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 struct timespec64 ctime = inode_get_ctime(inode);
236 struct timespec64 atime = inod 236 struct timespec64 atime = inode_get_atime(inode);
237 struct timespec64 mtime = inod 237 struct timespec64 mtime = inode_get_mtime(inode);
238 238
239 if ((timespec64_compare(&atime 239 if ((timespec64_compare(&atime, &mtime) <= 0) ||
240 (timespec64_compare(&atime 240 (timespec64_compare(&atime, &ctime) <= 0))
241 return 1; 241 return 1;
242 242
243 return 0; 243 return 0;
244 } 244 }
245 245
246 now = current_time(inode); 246 now = current_time(inode);
247 if ((now.tv_sec - inode_get_atime_sec( 247 if ((now.tv_sec - inode_get_atime_sec(inode) <= osb->s_atime_quantum))
248 return 0; 248 return 0;
249 else 249 else
250 return 1; 250 return 1;
251 } 251 }
252 252
253 int ocfs2_update_inode_atime(struct inode *ino 253 int ocfs2_update_inode_atime(struct inode *inode,
254 struct buffer_hea 254 struct buffer_head *bh)
255 { 255 {
256 int ret; 256 int ret;
257 struct ocfs2_super *osb = OCFS2_SB(ino 257 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
258 handle_t *handle; 258 handle_t *handle;
259 struct ocfs2_dinode *di = (struct ocfs 259 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
260 260
261 handle = ocfs2_start_trans(osb, OCFS2_ 261 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
262 if (IS_ERR(handle)) { 262 if (IS_ERR(handle)) {
263 ret = PTR_ERR(handle); 263 ret = PTR_ERR(handle);
264 mlog_errno(ret); 264 mlog_errno(ret);
265 goto out; 265 goto out;
266 } 266 }
267 267
268 ret = ocfs2_journal_access_di(handle, 268 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
269 OCFS2_JO 269 OCFS2_JOURNAL_ACCESS_WRITE);
270 if (ret) { 270 if (ret) {
271 mlog_errno(ret); 271 mlog_errno(ret);
272 goto out_commit; 272 goto out_commit;
273 } 273 }
274 274
275 /* 275 /*
276 * Don't use ocfs2_mark_inode_dirty() 276 * Don't use ocfs2_mark_inode_dirty() here as we don't always
277 * have i_rwsem to guard against concu 277 * have i_rwsem to guard against concurrent changes to other
278 * inode fields. 278 * inode fields.
279 */ 279 */
280 inode_set_atime_to_ts(inode, current_t 280 inode_set_atime_to_ts(inode, current_time(inode));
281 di->i_atime = cpu_to_le64(inode_get_at 281 di->i_atime = cpu_to_le64(inode_get_atime_sec(inode));
282 di->i_atime_nsec = cpu_to_le32(inode_g 282 di->i_atime_nsec = cpu_to_le32(inode_get_atime_nsec(inode));
283 ocfs2_update_inode_fsync_trans(handle, 283 ocfs2_update_inode_fsync_trans(handle, inode, 0);
284 ocfs2_journal_dirty(handle, bh); 284 ocfs2_journal_dirty(handle, bh);
285 285
286 out_commit: 286 out_commit:
287 ocfs2_commit_trans(osb, handle); 287 ocfs2_commit_trans(osb, handle);
288 out: 288 out:
289 return ret; 289 return ret;
290 } 290 }
291 291
292 int ocfs2_set_inode_size(handle_t *handle, 292 int ocfs2_set_inode_size(handle_t *handle,
293 struct inode * 293 struct inode *inode,
294 struct buffer_ 294 struct buffer_head *fe_bh,
295 u64 new_i_size 295 u64 new_i_size)
296 { 296 {
297 int status; 297 int status;
298 298
299 i_size_write(inode, new_i_size); 299 i_size_write(inode, new_i_size);
300 inode->i_blocks = ocfs2_inode_sector_c 300 inode->i_blocks = ocfs2_inode_sector_count(inode);
301 inode_set_mtime_to_ts(inode, inode_set 301 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
302 302
303 status = ocfs2_mark_inode_dirty(handle 303 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
304 if (status < 0) { 304 if (status < 0) {
305 mlog_errno(status); 305 mlog_errno(status);
306 goto bail; 306 goto bail;
307 } 307 }
308 308
309 bail: 309 bail:
310 return status; 310 return status;
311 } 311 }
312 312
313 int ocfs2_simple_size_update(struct inode *ino 313 int ocfs2_simple_size_update(struct inode *inode,
314 struct buffer_hea 314 struct buffer_head *di_bh,
315 u64 new_i_size) 315 u64 new_i_size)
316 { 316 {
317 int ret; 317 int ret;
318 struct ocfs2_super *osb = OCFS2_SB(ino 318 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
319 handle_t *handle = NULL; 319 handle_t *handle = NULL;
320 320
321 handle = ocfs2_start_trans(osb, OCFS2_ 321 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
322 if (IS_ERR(handle)) { 322 if (IS_ERR(handle)) {
323 ret = PTR_ERR(handle); 323 ret = PTR_ERR(handle);
324 mlog_errno(ret); 324 mlog_errno(ret);
325 goto out; 325 goto out;
326 } 326 }
327 327
328 ret = ocfs2_set_inode_size(handle, ino 328 ret = ocfs2_set_inode_size(handle, inode, di_bh,
329 new_i_size) 329 new_i_size);
330 if (ret < 0) 330 if (ret < 0)
331 mlog_errno(ret); 331 mlog_errno(ret);
332 332
333 ocfs2_update_inode_fsync_trans(handle, 333 ocfs2_update_inode_fsync_trans(handle, inode, 0);
334 ocfs2_commit_trans(osb, handle); 334 ocfs2_commit_trans(osb, handle);
335 out: 335 out:
336 return ret; 336 return ret;
337 } 337 }
338 338
339 static int ocfs2_cow_file_pos(struct inode *in 339 static int ocfs2_cow_file_pos(struct inode *inode,
340 struct buffer_he 340 struct buffer_head *fe_bh,
341 u64 offset) 341 u64 offset)
342 { 342 {
343 int status; 343 int status;
344 u32 phys, cpos = offset >> OCFS2_SB(in 344 u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
345 unsigned int num_clusters = 0; 345 unsigned int num_clusters = 0;
346 unsigned int ext_flags = 0; 346 unsigned int ext_flags = 0;
347 347
348 /* 348 /*
349 * If the new offset is aligned to the 349 * If the new offset is aligned to the range of the cluster, there is
350 * no space for ocfs2_zero_range_for_t 350 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
351 * CoW either. 351 * CoW either.
352 */ 352 */
353 if ((offset & (OCFS2_SB(inode->i_sb)-> 353 if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
354 return 0; 354 return 0;
355 355
356 status = ocfs2_get_clusters(inode, cpo 356 status = ocfs2_get_clusters(inode, cpos, &phys,
357 &num_clust 357 &num_clusters, &ext_flags);
358 if (status) { 358 if (status) {
359 mlog_errno(status); 359 mlog_errno(status);
360 goto out; 360 goto out;
361 } 361 }
362 362
363 if (!(ext_flags & OCFS2_EXT_REFCOUNTED 363 if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
364 goto out; 364 goto out;
365 365
366 return ocfs2_refcount_cow(inode, fe_bh 366 return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
367 367
368 out: 368 out:
369 return status; 369 return status;
370 } 370 }
371 371
372 static int ocfs2_orphan_for_truncate(struct oc 372 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
373 struct in 373 struct inode *inode,
374 struct bu 374 struct buffer_head *fe_bh,
375 u64 new_i 375 u64 new_i_size)
376 { 376 {
377 int status; 377 int status;
378 handle_t *handle; 378 handle_t *handle;
379 struct ocfs2_dinode *di; 379 struct ocfs2_dinode *di;
380 u64 cluster_bytes; 380 u64 cluster_bytes;
381 381
382 /* 382 /*
383 * We need to CoW the cluster contains 383 * We need to CoW the cluster contains the offset if it is reflinked
384 * since we will call ocfs2_zero_range 384 * since we will call ocfs2_zero_range_for_truncate later which will
385 * write "" from offset to the end of 385 * write "" from offset to the end of the cluster.
386 */ 386 */
387 status = ocfs2_cow_file_pos(inode, fe_ 387 status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
388 if (status) { 388 if (status) {
389 mlog_errno(status); 389 mlog_errno(status);
390 return status; 390 return status;
391 } 391 }
392 392
393 /* TODO: This needs to actually orphan 393 /* TODO: This needs to actually orphan the inode in this
394 * transaction. */ 394 * transaction. */
395 395
396 handle = ocfs2_start_trans(osb, OCFS2_ 396 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
397 if (IS_ERR(handle)) { 397 if (IS_ERR(handle)) {
398 status = PTR_ERR(handle); 398 status = PTR_ERR(handle);
399 mlog_errno(status); 399 mlog_errno(status);
400 goto out; 400 goto out;
401 } 401 }
402 402
403 status = ocfs2_journal_access_di(handl 403 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
404 OCFS2 404 OCFS2_JOURNAL_ACCESS_WRITE);
405 if (status < 0) { 405 if (status < 0) {
406 mlog_errno(status); 406 mlog_errno(status);
407 goto out_commit; 407 goto out_commit;
408 } 408 }
409 409
410 /* 410 /*
411 * Do this before setting i_size. 411 * Do this before setting i_size.
412 */ 412 */
413 cluster_bytes = ocfs2_align_bytes_to_c 413 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
414 status = ocfs2_zero_range_for_truncate 414 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
415 415 cluster_bytes);
416 if (status) { 416 if (status) {
417 mlog_errno(status); 417 mlog_errno(status);
418 goto out_commit; 418 goto out_commit;
419 } 419 }
420 420
421 i_size_write(inode, new_i_size); 421 i_size_write(inode, new_i_size);
422 inode_set_mtime_to_ts(inode, inode_set 422 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
423 423
424 di = (struct ocfs2_dinode *) fe_bh->b_ 424 di = (struct ocfs2_dinode *) fe_bh->b_data;
425 di->i_size = cpu_to_le64(new_i_size); 425 di->i_size = cpu_to_le64(new_i_size);
426 di->i_ctime = di->i_mtime = cpu_to_le6 426 di->i_ctime = di->i_mtime = cpu_to_le64(inode_get_ctime_sec(inode));
427 di->i_ctime_nsec = di->i_mtime_nsec = 427 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
428 ocfs2_update_inode_fsync_trans(handle, 428 ocfs2_update_inode_fsync_trans(handle, inode, 0);
429 429
430 ocfs2_journal_dirty(handle, fe_bh); 430 ocfs2_journal_dirty(handle, fe_bh);
431 431
432 out_commit: 432 out_commit:
433 ocfs2_commit_trans(osb, handle); 433 ocfs2_commit_trans(osb, handle);
434 out: 434 out:
435 return status; 435 return status;
436 } 436 }
437 437
438 int ocfs2_truncate_file(struct inode *inode, 438 int ocfs2_truncate_file(struct inode *inode,
439 struct buffer_h 439 struct buffer_head *di_bh,
440 u64 new_i_size) 440 u64 new_i_size)
441 { 441 {
442 int status = 0; 442 int status = 0;
443 struct ocfs2_dinode *fe = NULL; 443 struct ocfs2_dinode *fe = NULL;
444 struct ocfs2_super *osb = OCFS2_SB(ino 444 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
445 445
446 /* We trust di_bh because it comes fro 446 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
447 * already validated it */ 447 * already validated it */
448 fe = (struct ocfs2_dinode *) di_bh->b_ 448 fe = (struct ocfs2_dinode *) di_bh->b_data;
449 449
450 trace_ocfs2_truncate_file((unsigned lo 450 trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
451 (unsigned lo 451 (unsigned long long)le64_to_cpu(fe->i_size),
452 (unsigned lo 452 (unsigned long long)new_i_size);
453 453
454 mlog_bug_on_msg(le64_to_cpu(fe->i_size 454 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
455 "Inode %llu, inode i_s 455 "Inode %llu, inode i_size = %lld != di "
456 "i_size = %llu, i_flag 456 "i_size = %llu, i_flags = 0x%x\n",
457 (unsigned long long)OC 457 (unsigned long long)OCFS2_I(inode)->ip_blkno,
458 i_size_read(inode), 458 i_size_read(inode),
459 (unsigned long long)le 459 (unsigned long long)le64_to_cpu(fe->i_size),
460 le32_to_cpu(fe->i_flag 460 le32_to_cpu(fe->i_flags));
461 461
462 if (new_i_size > le64_to_cpu(fe->i_siz 462 if (new_i_size > le64_to_cpu(fe->i_size)) {
463 trace_ocfs2_truncate_file_erro 463 trace_ocfs2_truncate_file_error(
464 (unsigned long long)le 464 (unsigned long long)le64_to_cpu(fe->i_size),
465 (unsigned long long)ne 465 (unsigned long long)new_i_size);
466 status = -EINVAL; 466 status = -EINVAL;
467 mlog_errno(status); 467 mlog_errno(status);
468 goto bail; 468 goto bail;
469 } 469 }
470 470
471 down_write(&OCFS2_I(inode)->ip_alloc_s 471 down_write(&OCFS2_I(inode)->ip_alloc_sem);
472 472
473 ocfs2_resv_discard(&osb->osb_la_resmap 473 ocfs2_resv_discard(&osb->osb_la_resmap,
474 &OCFS2_I(inode)->ip 474 &OCFS2_I(inode)->ip_la_data_resv);
475 475
476 /* 476 /*
477 * The inode lock forced other nodes t 477 * The inode lock forced other nodes to sync and drop their
478 * pages, which (correctly) happens ev 478 * pages, which (correctly) happens even if we have a truncate
479 * without allocation change - ocfs2 c 479 * without allocation change - ocfs2 cluster sizes can be much
480 * greater than page size, so we have 480 * greater than page size, so we have to truncate them
481 * anyway. 481 * anyway.
482 */ 482 */
483 483
484 if (OCFS2_I(inode)->ip_dyn_features & 484 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
485 unmap_mapping_range(inode->i_m 485 unmap_mapping_range(inode->i_mapping,
486 new_i_size 486 new_i_size + PAGE_SIZE - 1, 0, 1);
487 truncate_inode_pages(inode->i_ 487 truncate_inode_pages(inode->i_mapping, new_i_size);
488 status = ocfs2_truncate_inline 488 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
489 489 i_size_read(inode), 1);
490 if (status) 490 if (status)
491 mlog_errno(status); 491 mlog_errno(status);
492 492
493 goto bail_unlock_sem; 493 goto bail_unlock_sem;
494 } 494 }
495 495
496 /* alright, we're going to need to do 496 /* alright, we're going to need to do a full blown alloc size
497 * change. Orphan the inode so that re 497 * change. Orphan the inode so that recovery can complete the
498 * truncate if necessary. This does th 498 * truncate if necessary. This does the task of marking
499 * i_size. */ 499 * i_size. */
500 status = ocfs2_orphan_for_truncate(osb 500 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
501 if (status < 0) { 501 if (status < 0) {
502 mlog_errno(status); 502 mlog_errno(status);
503 goto bail_unlock_sem; 503 goto bail_unlock_sem;
504 } 504 }
505 505
506 unmap_mapping_range(inode->i_mapping, 506 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
507 truncate_inode_pages(inode->i_mapping, 507 truncate_inode_pages(inode->i_mapping, new_i_size);
508 508
509 status = ocfs2_commit_truncate(osb, in 509 status = ocfs2_commit_truncate(osb, inode, di_bh);
510 if (status < 0) { 510 if (status < 0) {
511 mlog_errno(status); 511 mlog_errno(status);
512 goto bail_unlock_sem; 512 goto bail_unlock_sem;
513 } 513 }
514 514
515 /* TODO: orphan dir cleanup here. */ 515 /* TODO: orphan dir cleanup here. */
516 bail_unlock_sem: 516 bail_unlock_sem:
517 up_write(&OCFS2_I(inode)->ip_alloc_sem 517 up_write(&OCFS2_I(inode)->ip_alloc_sem);
518 518
519 bail: 519 bail:
520 if (!status && OCFS2_I(inode)->ip_clus 520 if (!status && OCFS2_I(inode)->ip_clusters == 0)
521 status = ocfs2_try_remove_refc 521 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
522 522
523 return status; 523 return status;
524 } 524 }
525 525
526 /* 526 /*
527 * extend file allocation only here. 527 * extend file allocation only here.
528 * we'll update all the disk stuff, and oip->a 528 * we'll update all the disk stuff, and oip->alloc_size
529 * 529 *
530 * expect stuff to be locked, a transaction st 530 * expect stuff to be locked, a transaction started and enough data /
531 * metadata reservations in the contexts. 531 * metadata reservations in the contexts.
532 * 532 *
533 * Will return -EAGAIN, and a reason if a rest 533 * Will return -EAGAIN, and a reason if a restart is needed.
534 * If passed in, *reason will always be set, e 534 * If passed in, *reason will always be set, even in error.
535 */ 535 */
536 int ocfs2_add_inode_data(struct ocfs2_super *o 536 int ocfs2_add_inode_data(struct ocfs2_super *osb,
537 struct inode *inode, 537 struct inode *inode,
538 u32 *logical_offset, 538 u32 *logical_offset,
539 u32 clusters_to_add, 539 u32 clusters_to_add,
540 int mark_unwritten, 540 int mark_unwritten,
541 struct buffer_head *f 541 struct buffer_head *fe_bh,
542 handle_t *handle, 542 handle_t *handle,
543 struct ocfs2_alloc_co 543 struct ocfs2_alloc_context *data_ac,
544 struct ocfs2_alloc_co 544 struct ocfs2_alloc_context *meta_ac,
545 enum ocfs2_alloc_rest 545 enum ocfs2_alloc_restarted *reason_ret)
546 { 546 {
547 struct ocfs2_extent_tree et; 547 struct ocfs2_extent_tree et;
548 548
549 ocfs2_init_dinode_extent_tree(&et, INO 549 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
550 return ocfs2_add_clusters_in_btree(han 550 return ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
551 clu 551 clusters_to_add, mark_unwritten,
552 dat 552 data_ac, meta_ac, reason_ret);
553 } 553 }
554 554
555 static int ocfs2_extend_allocation(struct inod 555 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
556 u32 cluster 556 u32 clusters_to_add, int mark_unwritten)
557 { 557 {
558 int status = 0; 558 int status = 0;
559 int restart_func = 0; 559 int restart_func = 0;
560 int credits; 560 int credits;
561 u32 prev_clusters; 561 u32 prev_clusters;
562 struct buffer_head *bh = NULL; 562 struct buffer_head *bh = NULL;
563 struct ocfs2_dinode *fe = NULL; 563 struct ocfs2_dinode *fe = NULL;
564 handle_t *handle = NULL; 564 handle_t *handle = NULL;
565 struct ocfs2_alloc_context *data_ac = 565 struct ocfs2_alloc_context *data_ac = NULL;
566 struct ocfs2_alloc_context *meta_ac = 566 struct ocfs2_alloc_context *meta_ac = NULL;
567 enum ocfs2_alloc_restarted why = RESTA 567 enum ocfs2_alloc_restarted why = RESTART_NONE;
568 struct ocfs2_super *osb = OCFS2_SB(ino 568 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
569 struct ocfs2_extent_tree et; 569 struct ocfs2_extent_tree et;
570 int did_quota = 0; 570 int did_quota = 0;
571 571
572 /* 572 /*
573 * Unwritten extent only exists for fi 573 * Unwritten extent only exists for file systems which
574 * support holes. 574 * support holes.
575 */ 575 */
576 BUG_ON(mark_unwritten && !ocfs2_sparse 576 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
577 577
578 status = ocfs2_read_inode_block(inode, 578 status = ocfs2_read_inode_block(inode, &bh);
579 if (status < 0) { 579 if (status < 0) {
580 mlog_errno(status); 580 mlog_errno(status);
581 goto leave; 581 goto leave;
582 } 582 }
583 fe = (struct ocfs2_dinode *) bh->b_dat 583 fe = (struct ocfs2_dinode *) bh->b_data;
584 584
585 restart_all: 585 restart_all:
586 BUG_ON(le32_to_cpu(fe->i_clusters) != 586 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
587 587
588 ocfs2_init_dinode_extent_tree(&et, INO 588 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
589 status = ocfs2_lock_allocators(inode, 589 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
590 &data_a 590 &data_ac, &meta_ac);
591 if (status) { 591 if (status) {
592 mlog_errno(status); 592 mlog_errno(status);
593 goto leave; 593 goto leave;
594 } 594 }
595 595
596 credits = ocfs2_calc_extend_credits(os 596 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
597 handle = ocfs2_start_trans(osb, credit 597 handle = ocfs2_start_trans(osb, credits);
598 if (IS_ERR(handle)) { 598 if (IS_ERR(handle)) {
599 status = PTR_ERR(handle); 599 status = PTR_ERR(handle);
600 handle = NULL; 600 handle = NULL;
601 mlog_errno(status); 601 mlog_errno(status);
602 goto leave; 602 goto leave;
603 } 603 }
604 604
605 restarted_transaction: 605 restarted_transaction:
606 trace_ocfs2_extend_allocation( 606 trace_ocfs2_extend_allocation(
607 (unsigned long long)OCFS2_I(in 607 (unsigned long long)OCFS2_I(inode)->ip_blkno,
608 (unsigned long long)i_size_rea 608 (unsigned long long)i_size_read(inode),
609 le32_to_cpu(fe->i_clusters), c 609 le32_to_cpu(fe->i_clusters), clusters_to_add,
610 why, restart_func); 610 why, restart_func);
611 611
612 status = dquot_alloc_space_nodirty(ino 612 status = dquot_alloc_space_nodirty(inode,
613 ocfs2_clusters_to_byte 613 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
614 if (status) 614 if (status)
615 goto leave; 615 goto leave;
616 did_quota = 1; 616 did_quota = 1;
617 617
618 /* reserve a write to the file entry e 618 /* reserve a write to the file entry early on - that we if we
619 * run out of credits in the allocatio 619 * run out of credits in the allocation path, we can still
620 * update i_size. */ 620 * update i_size. */
621 status = ocfs2_journal_access_di(handl 621 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
622 OCFS2 622 OCFS2_JOURNAL_ACCESS_WRITE);
623 if (status < 0) { 623 if (status < 0) {
624 mlog_errno(status); 624 mlog_errno(status);
625 goto leave; 625 goto leave;
626 } 626 }
627 627
628 prev_clusters = OCFS2_I(inode)->ip_clu 628 prev_clusters = OCFS2_I(inode)->ip_clusters;
629 629
630 status = ocfs2_add_inode_data(osb, 630 status = ocfs2_add_inode_data(osb,
631 inode, 631 inode,
632 &logical 632 &logical_start,
633 clusters 633 clusters_to_add,
634 mark_unw 634 mark_unwritten,
635 bh, 635 bh,
636 handle, 636 handle,
637 data_ac, 637 data_ac,
638 meta_ac, 638 meta_ac,
639 &why); 639 &why);
640 if ((status < 0) && (status != -EAGAIN 640 if ((status < 0) && (status != -EAGAIN)) {
641 if (status != -ENOSPC) 641 if (status != -ENOSPC)
642 mlog_errno(status); 642 mlog_errno(status);
643 goto leave; 643 goto leave;
644 } 644 }
645 ocfs2_update_inode_fsync_trans(handle, 645 ocfs2_update_inode_fsync_trans(handle, inode, 1);
646 ocfs2_journal_dirty(handle, bh); 646 ocfs2_journal_dirty(handle, bh);
647 647
648 spin_lock(&OCFS2_I(inode)->ip_lock); 648 spin_lock(&OCFS2_I(inode)->ip_lock);
649 clusters_to_add -= (OCFS2_I(inode)->ip 649 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
650 spin_unlock(&OCFS2_I(inode)->ip_lock); 650 spin_unlock(&OCFS2_I(inode)->ip_lock);
651 /* Release unused quota reservation */ 651 /* Release unused quota reservation */
652 dquot_free_space(inode, 652 dquot_free_space(inode,
653 ocfs2_clusters_to_byte 653 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
654 did_quota = 0; 654 did_quota = 0;
655 655
656 if (why != RESTART_NONE && clusters_to 656 if (why != RESTART_NONE && clusters_to_add) {
657 if (why == RESTART_META) { 657 if (why == RESTART_META) {
658 restart_func = 1; 658 restart_func = 1;
659 status = 0; 659 status = 0;
660 } else { 660 } else {
661 BUG_ON(why != RESTART_ 661 BUG_ON(why != RESTART_TRANS);
662 662
663 status = ocfs2_allocat 663 status = ocfs2_allocate_extend_trans(handle, 1);
664 if (status < 0) { 664 if (status < 0) {
665 /* handle stil 665 /* handle still has to be committed at
666 * this point. 666 * this point. */
667 status = -ENOM 667 status = -ENOMEM;
668 mlog_errno(sta 668 mlog_errno(status);
669 goto leave; 669 goto leave;
670 } 670 }
671 goto restarted_transac 671 goto restarted_transaction;
672 } 672 }
673 } 673 }
674 674
675 trace_ocfs2_extend_allocation_end(OCFS 675 trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
676 le32_to_cpu(fe->i_clusters), 676 le32_to_cpu(fe->i_clusters),
677 (unsigned long long)le64_to_cpu(f 677 (unsigned long long)le64_to_cpu(fe->i_size),
678 OCFS2_I(inode)->ip_clusters, 678 OCFS2_I(inode)->ip_clusters,
679 (unsigned long long)i_size_read(i 679 (unsigned long long)i_size_read(inode));
680 680
681 leave: 681 leave:
682 if (status < 0 && did_quota) 682 if (status < 0 && did_quota)
683 dquot_free_space(inode, 683 dquot_free_space(inode,
684 ocfs2_clusters_to_byte 684 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
685 if (handle) { 685 if (handle) {
686 ocfs2_commit_trans(osb, handle 686 ocfs2_commit_trans(osb, handle);
687 handle = NULL; 687 handle = NULL;
688 } 688 }
689 if (data_ac) { 689 if (data_ac) {
690 ocfs2_free_alloc_context(data_ 690 ocfs2_free_alloc_context(data_ac);
691 data_ac = NULL; 691 data_ac = NULL;
692 } 692 }
693 if (meta_ac) { 693 if (meta_ac) {
694 ocfs2_free_alloc_context(meta_ 694 ocfs2_free_alloc_context(meta_ac);
695 meta_ac = NULL; 695 meta_ac = NULL;
696 } 696 }
697 if ((!status) && restart_func) { 697 if ((!status) && restart_func) {
698 restart_func = 0; 698 restart_func = 0;
699 goto restart_all; 699 goto restart_all;
700 } 700 }
701 brelse(bh); 701 brelse(bh);
702 bh = NULL; 702 bh = NULL;
703 703
704 return status; 704 return status;
705 } 705 }
706 706
707 /* 707 /*
708 * While a write will already be ordering the 708 * While a write will already be ordering the data, a truncate will not.
709 * Thus, we need to explicitly order the zeroe 709 * Thus, we need to explicitly order the zeroed pages.
710 */ 710 */
711 static handle_t *ocfs2_zero_start_ordered_tran 711 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
712 712 struct buffer_head *di_bh,
713 713 loff_t start_byte,
714 714 loff_t length)
715 { 715 {
716 struct ocfs2_super *osb = OCFS2_SB(ino 716 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
717 handle_t *handle = NULL; 717 handle_t *handle = NULL;
718 int ret = 0; 718 int ret = 0;
719 719
720 if (!ocfs2_should_order_data(inode)) 720 if (!ocfs2_should_order_data(inode))
721 goto out; 721 goto out;
722 722
723 handle = ocfs2_start_trans(osb, OCFS2_ 723 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
724 if (IS_ERR(handle)) { 724 if (IS_ERR(handle)) {
725 ret = -ENOMEM; 725 ret = -ENOMEM;
726 mlog_errno(ret); 726 mlog_errno(ret);
727 goto out; 727 goto out;
728 } 728 }
729 729
730 ret = ocfs2_jbd2_inode_add_write(handl 730 ret = ocfs2_jbd2_inode_add_write(handle, inode, start_byte, length);
731 if (ret < 0) { 731 if (ret < 0) {
732 mlog_errno(ret); 732 mlog_errno(ret);
733 goto out; 733 goto out;
734 } 734 }
735 735
736 ret = ocfs2_journal_access_di(handle, 736 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
737 OCFS2_JO 737 OCFS2_JOURNAL_ACCESS_WRITE);
738 if (ret) 738 if (ret)
739 mlog_errno(ret); 739 mlog_errno(ret);
740 ocfs2_update_inode_fsync_trans(handle, 740 ocfs2_update_inode_fsync_trans(handle, inode, 1);
741 741
742 out: 742 out:
743 if (ret) { 743 if (ret) {
744 if (!IS_ERR(handle)) 744 if (!IS_ERR(handle))
745 ocfs2_commit_trans(osb 745 ocfs2_commit_trans(osb, handle);
746 handle = ERR_PTR(ret); 746 handle = ERR_PTR(ret);
747 } 747 }
748 return handle; 748 return handle;
749 } 749 }
750 750
751 /* Some parts of this taken from generic_cont_ 751 /* Some parts of this taken from generic_cont_expand, which turned out
752 * to be too fragile to do exactly what we nee 752 * to be too fragile to do exactly what we need without us having to
753 * worry about recursive locking in ->write_be 753 * worry about recursive locking in ->write_begin() and ->write_end(). */
754 static int ocfs2_write_zero_page(struct inode 754 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
755 u64 abs_to, s 755 u64 abs_to, struct buffer_head *di_bh)
756 { 756 {
757 struct address_space *mapping = inode- 757 struct address_space *mapping = inode->i_mapping;
758 struct folio *folio; !! 758 struct page *page;
759 unsigned long index = abs_from >> PAGE 759 unsigned long index = abs_from >> PAGE_SHIFT;
760 handle_t *handle; 760 handle_t *handle;
761 int ret = 0; 761 int ret = 0;
762 unsigned zero_from, zero_to, block_sta 762 unsigned zero_from, zero_to, block_start, block_end;
763 struct ocfs2_dinode *di = (struct ocfs 763 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
764 764
765 BUG_ON(abs_from >= abs_to); 765 BUG_ON(abs_from >= abs_to);
766 BUG_ON(abs_to > (((u64)index + 1) << P 766 BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
767 BUG_ON(abs_from & (inode->i_blkbits - 767 BUG_ON(abs_from & (inode->i_blkbits - 1));
768 768
769 handle = ocfs2_zero_start_ordered_tran 769 handle = ocfs2_zero_start_ordered_transaction(inode, di_bh,
770 770 abs_from,
771 771 abs_to - abs_from);
772 if (IS_ERR(handle)) { 772 if (IS_ERR(handle)) {
773 ret = PTR_ERR(handle); 773 ret = PTR_ERR(handle);
774 goto out; 774 goto out;
775 } 775 }
776 776
777 folio = __filemap_get_folio(mapping, i !! 777 page = find_or_create_page(mapping, index, GFP_NOFS);
778 FGP_LOCK | FGP_ACCESSE !! 778 if (!page) {
779 if (IS_ERR(folio)) { !! 779 ret = -ENOMEM;
780 ret = PTR_ERR(folio); <<
781 mlog_errno(ret); 780 mlog_errno(ret);
782 goto out_commit_trans; 781 goto out_commit_trans;
783 } 782 }
784 783
785 /* Get the offsets within the page tha 784 /* Get the offsets within the page that we want to zero */
786 zero_from = abs_from & (PAGE_SIZE - 1) 785 zero_from = abs_from & (PAGE_SIZE - 1);
787 zero_to = abs_to & (PAGE_SIZE - 1); 786 zero_to = abs_to & (PAGE_SIZE - 1);
788 if (!zero_to) 787 if (!zero_to)
789 zero_to = PAGE_SIZE; 788 zero_to = PAGE_SIZE;
790 789
791 trace_ocfs2_write_zero_page( 790 trace_ocfs2_write_zero_page(
792 (unsigned long long)OC 791 (unsigned long long)OCFS2_I(inode)->ip_blkno,
793 (unsigned long long)ab 792 (unsigned long long)abs_from,
794 (unsigned long long)ab 793 (unsigned long long)abs_to,
795 index, zero_from, zero 794 index, zero_from, zero_to);
796 795
797 /* We know that zero_from is block ali 796 /* We know that zero_from is block aligned */
798 for (block_start = zero_from; block_st 797 for (block_start = zero_from; block_start < zero_to;
799 block_start = block_end) { 798 block_start = block_end) {
800 block_end = block_start + i_bl 799 block_end = block_start + i_blocksize(inode);
801 800
802 /* 801 /*
803 * block_start is block-aligne 802 * block_start is block-aligned. Bump it by one to force
804 * __block_write_begin and blo 803 * __block_write_begin and block_commit_write to zero the
805 * whole block. 804 * whole block.
806 */ 805 */
807 ret = __block_write_begin(foli !! 806 ret = __block_write_begin(page, block_start + 1, 0,
808 ocfs 807 ocfs2_get_block);
809 if (ret < 0) { 808 if (ret < 0) {
810 mlog_errno(ret); 809 mlog_errno(ret);
811 goto out_unlock; 810 goto out_unlock;
812 } 811 }
813 812
814 813
815 /* must not update i_size! */ 814 /* must not update i_size! */
816 block_commit_write(&folio->pag !! 815 block_commit_write(page, block_start + 1, block_start + 1);
817 } 816 }
818 817
819 /* 818 /*
820 * fs-writeback will release the dirty 819 * fs-writeback will release the dirty pages without page lock
821 * whose offset are over inode size, t 820 * whose offset are over inode size, the release happens at
822 * block_write_full_folio(). 821 * block_write_full_folio().
823 */ 822 */
824 i_size_write(inode, abs_to); 823 i_size_write(inode, abs_to);
825 inode->i_blocks = ocfs2_inode_sector_c 824 inode->i_blocks = ocfs2_inode_sector_count(inode);
826 di->i_size = cpu_to_le64((u64)i_size_r 825 di->i_size = cpu_to_le64((u64)i_size_read(inode));
827 inode_set_mtime_to_ts(inode, inode_set 826 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
828 di->i_mtime = di->i_ctime = cpu_to_le6 827 di->i_mtime = di->i_ctime = cpu_to_le64(inode_get_mtime_sec(inode));
829 di->i_ctime_nsec = cpu_to_le32(inode_g 828 di->i_ctime_nsec = cpu_to_le32(inode_get_mtime_nsec(inode));
830 di->i_mtime_nsec = di->i_ctime_nsec; 829 di->i_mtime_nsec = di->i_ctime_nsec;
831 if (handle) { 830 if (handle) {
832 ocfs2_journal_dirty(handle, di 831 ocfs2_journal_dirty(handle, di_bh);
833 ocfs2_update_inode_fsync_trans 832 ocfs2_update_inode_fsync_trans(handle, inode, 1);
834 } 833 }
835 834
836 out_unlock: 835 out_unlock:
837 folio_unlock(folio); !! 836 unlock_page(page);
838 folio_put(folio); !! 837 put_page(page);
839 out_commit_trans: 838 out_commit_trans:
840 if (handle) 839 if (handle)
841 ocfs2_commit_trans(OCFS2_SB(in 840 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
842 out: 841 out:
843 return ret; 842 return ret;
844 } 843 }
845 844
846 /* 845 /*
847 * Find the next range to zero. We do this in 846 * Find the next range to zero. We do this in terms of bytes because
848 * that's what ocfs2_zero_extend() wants, and 847 * that's what ocfs2_zero_extend() wants, and it is dealing with the
849 * pagecache. We may return multiple extents. 848 * pagecache. We may return multiple extents.
850 * 849 *
851 * zero_start and zero_end are ocfs2_zero_exte 850 * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
852 * needs to be zeroed. range_start and range_ 851 * needs to be zeroed. range_start and range_end return the next zeroing
853 * range. A subsequent call should pass the p 852 * range. A subsequent call should pass the previous range_end as its
854 * zero_start. If range_end is 0, there's not 853 * zero_start. If range_end is 0, there's nothing to do.
855 * 854 *
856 * Unwritten extents are skipped over. Refcou 855 * Unwritten extents are skipped over. Refcounted extents are CoWd.
857 */ 856 */
858 static int ocfs2_zero_extend_get_range(struct 857 static int ocfs2_zero_extend_get_range(struct inode *inode,
859 struct 858 struct buffer_head *di_bh,
860 u64 zer 859 u64 zero_start, u64 zero_end,
861 u64 *ra 860 u64 *range_start, u64 *range_end)
862 { 861 {
863 int rc = 0, needs_cow = 0; 862 int rc = 0, needs_cow = 0;
864 u32 p_cpos, zero_clusters = 0; 863 u32 p_cpos, zero_clusters = 0;
865 u32 zero_cpos = 864 u32 zero_cpos =
866 zero_start >> OCFS2_SB(inode-> 865 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
867 u32 last_cpos = ocfs2_clusters_for_byt 866 u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
868 unsigned int num_clusters = 0; 867 unsigned int num_clusters = 0;
869 unsigned int ext_flags = 0; 868 unsigned int ext_flags = 0;
870 869
871 while (zero_cpos < last_cpos) { 870 while (zero_cpos < last_cpos) {
872 rc = ocfs2_get_clusters(inode, 871 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
873 &num_c 872 &num_clusters, &ext_flags);
874 if (rc) { 873 if (rc) {
875 mlog_errno(rc); 874 mlog_errno(rc);
876 goto out; 875 goto out;
877 } 876 }
878 877
879 if (p_cpos && !(ext_flags & OC 878 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
880 zero_clusters = num_cl 879 zero_clusters = num_clusters;
881 if (ext_flags & OCFS2_ 880 if (ext_flags & OCFS2_EXT_REFCOUNTED)
882 needs_cow = 1; 881 needs_cow = 1;
883 break; 882 break;
884 } 883 }
885 884
886 zero_cpos += num_clusters; 885 zero_cpos += num_clusters;
887 } 886 }
888 if (!zero_clusters) { 887 if (!zero_clusters) {
889 *range_end = 0; 888 *range_end = 0;
890 goto out; 889 goto out;
891 } 890 }
892 891
893 while ((zero_cpos + zero_clusters) < l 892 while ((zero_cpos + zero_clusters) < last_cpos) {
894 rc = ocfs2_get_clusters(inode, 893 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
895 &p_cpo 894 &p_cpos, &num_clusters,
896 &ext_f 895 &ext_flags);
897 if (rc) { 896 if (rc) {
898 mlog_errno(rc); 897 mlog_errno(rc);
899 goto out; 898 goto out;
900 } 899 }
901 900
902 if (!p_cpos || (ext_flags & OC 901 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
903 break; 902 break;
904 if (ext_flags & OCFS2_EXT_REFC 903 if (ext_flags & OCFS2_EXT_REFCOUNTED)
905 needs_cow = 1; 904 needs_cow = 1;
906 zero_clusters += num_clusters; 905 zero_clusters += num_clusters;
907 } 906 }
908 if ((zero_cpos + zero_clusters) > last 907 if ((zero_cpos + zero_clusters) > last_cpos)
909 zero_clusters = last_cpos - ze 908 zero_clusters = last_cpos - zero_cpos;
910 909
911 if (needs_cow) { 910 if (needs_cow) {
912 rc = ocfs2_refcount_cow(inode, 911 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
913 zero_c 912 zero_clusters, UINT_MAX);
914 if (rc) { 913 if (rc) {
915 mlog_errno(rc); 914 mlog_errno(rc);
916 goto out; 915 goto out;
917 } 916 }
918 } 917 }
919 918
920 *range_start = ocfs2_clusters_to_bytes 919 *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
921 *range_end = ocfs2_clusters_to_bytes(i 920 *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
922 z 921 zero_cpos + zero_clusters);
923 922
924 out: 923 out:
925 return rc; 924 return rc;
926 } 925 }
927 926
928 /* 927 /*
929 * Zero one range returned from ocfs2_zero_ext 928 * Zero one range returned from ocfs2_zero_extend_get_range(). The caller
930 * has made sure that the entire range needs z 929 * has made sure that the entire range needs zeroing.
931 */ 930 */
932 static int ocfs2_zero_extend_range(struct inod 931 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
933 u64 range_e 932 u64 range_end, struct buffer_head *di_bh)
934 { 933 {
935 int rc = 0; 934 int rc = 0;
936 u64 next_pos; 935 u64 next_pos;
937 u64 zero_pos = range_start; 936 u64 zero_pos = range_start;
938 937
939 trace_ocfs2_zero_extend_range( 938 trace_ocfs2_zero_extend_range(
940 (unsigned long long)OC 939 (unsigned long long)OCFS2_I(inode)->ip_blkno,
941 (unsigned long long)ra 940 (unsigned long long)range_start,
942 (unsigned long long)ra 941 (unsigned long long)range_end);
943 BUG_ON(range_start >= range_end); 942 BUG_ON(range_start >= range_end);
944 943
945 while (zero_pos < range_end) { 944 while (zero_pos < range_end) {
946 next_pos = (zero_pos & PAGE_MA 945 next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
947 if (next_pos > range_end) 946 if (next_pos > range_end)
948 next_pos = range_end; 947 next_pos = range_end;
949 rc = ocfs2_write_zero_page(ino 948 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
950 if (rc < 0) { 949 if (rc < 0) {
951 mlog_errno(rc); 950 mlog_errno(rc);
952 break; 951 break;
953 } 952 }
954 zero_pos = next_pos; 953 zero_pos = next_pos;
955 954
956 /* 955 /*
957 * Very large extends have the 956 * Very large extends have the potential to lock up
958 * the cpu for extended period 957 * the cpu for extended periods of time.
959 */ 958 */
960 cond_resched(); 959 cond_resched();
961 } 960 }
962 961
963 return rc; 962 return rc;
964 } 963 }
965 964
966 int ocfs2_zero_extend(struct inode *inode, str 965 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
967 loff_t zero_to_size) 966 loff_t zero_to_size)
968 { 967 {
969 int ret = 0; 968 int ret = 0;
970 u64 zero_start, range_start = 0, range 969 u64 zero_start, range_start = 0, range_end = 0;
971 struct super_block *sb = inode->i_sb; 970 struct super_block *sb = inode->i_sb;
972 971
973 zero_start = ocfs2_align_bytes_to_bloc 972 zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
974 trace_ocfs2_zero_extend((unsigned long 973 trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
975 (unsigned long 974 (unsigned long long)zero_start,
976 (unsigned long 975 (unsigned long long)i_size_read(inode));
977 while (zero_start < zero_to_size) { 976 while (zero_start < zero_to_size) {
978 ret = ocfs2_zero_extend_get_ra 977 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
979 978 zero_to_size,
980 979 &range_start,
981 980 &range_end);
982 if (ret) { 981 if (ret) {
983 mlog_errno(ret); 982 mlog_errno(ret);
984 break; 983 break;
985 } 984 }
986 if (!range_end) 985 if (!range_end)
987 break; 986 break;
988 /* Trim the ends */ 987 /* Trim the ends */
989 if (range_start < zero_start) 988 if (range_start < zero_start)
990 range_start = zero_sta 989 range_start = zero_start;
991 if (range_end > zero_to_size) 990 if (range_end > zero_to_size)
992 range_end = zero_to_si 991 range_end = zero_to_size;
993 992
994 ret = ocfs2_zero_extend_range( 993 ret = ocfs2_zero_extend_range(inode, range_start,
995 994 range_end, di_bh);
996 if (ret) { 995 if (ret) {
997 mlog_errno(ret); 996 mlog_errno(ret);
998 break; 997 break;
999 } 998 }
1000 zero_start = range_end; 999 zero_start = range_end;
1001 } 1000 }
1002 1001
1003 return ret; 1002 return ret;
1004 } 1003 }
1005 1004
1006 int ocfs2_extend_no_holes(struct inode *inode 1005 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1007 u64 new_i_size, u64 1006 u64 new_i_size, u64 zero_to)
1008 { 1007 {
1009 int ret; 1008 int ret;
1010 u32 clusters_to_add; 1009 u32 clusters_to_add;
1011 struct ocfs2_inode_info *oi = OCFS2_I 1010 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1012 1011
1013 /* 1012 /*
1014 * Only quota files call this without 1013 * Only quota files call this without a bh, and they can't be
1015 * refcounted. 1014 * refcounted.
1016 */ 1015 */
1017 BUG_ON(!di_bh && ocfs2_is_refcount_in 1016 BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1018 BUG_ON(!di_bh && !(oi->ip_flags & OCF 1017 BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1019 1018
1020 clusters_to_add = ocfs2_clusters_for_ 1019 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1021 if (clusters_to_add < oi->ip_clusters 1020 if (clusters_to_add < oi->ip_clusters)
1022 clusters_to_add = 0; 1021 clusters_to_add = 0;
1023 else 1022 else
1024 clusters_to_add -= oi->ip_clu 1023 clusters_to_add -= oi->ip_clusters;
1025 1024
1026 if (clusters_to_add) { 1025 if (clusters_to_add) {
1027 ret = ocfs2_extend_allocation 1026 ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1028 1027 clusters_to_add, 0);
1029 if (ret) { 1028 if (ret) {
1030 mlog_errno(ret); 1029 mlog_errno(ret);
1031 goto out; 1030 goto out;
1032 } 1031 }
1033 } 1032 }
1034 1033
1035 /* 1034 /*
1036 * Call this even if we don't add any 1035 * Call this even if we don't add any clusters to the tree. We
1037 * still need to zero the area betwee 1036 * still need to zero the area between the old i_size and the
1038 * new i_size. 1037 * new i_size.
1039 */ 1038 */
1040 ret = ocfs2_zero_extend(inode, di_bh, 1039 ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1041 if (ret < 0) 1040 if (ret < 0)
1042 mlog_errno(ret); 1041 mlog_errno(ret);
1043 1042
1044 out: 1043 out:
1045 return ret; 1044 return ret;
1046 } 1045 }
1047 1046
1048 static int ocfs2_extend_file(struct inode *in 1047 static int ocfs2_extend_file(struct inode *inode,
1049 struct buffer_he 1048 struct buffer_head *di_bh,
1050 u64 new_i_size) 1049 u64 new_i_size)
1051 { 1050 {
1052 int ret = 0; 1051 int ret = 0;
1053 struct ocfs2_inode_info *oi = OCFS2_I 1052 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1054 1053
1055 BUG_ON(!di_bh); 1054 BUG_ON(!di_bh);
1056 1055
1057 /* setattr sometimes calls us like th 1056 /* setattr sometimes calls us like this. */
1058 if (new_i_size == 0) 1057 if (new_i_size == 0)
1059 goto out; 1058 goto out;
1060 1059
1061 if (i_size_read(inode) == new_i_size) 1060 if (i_size_read(inode) == new_i_size)
1062 goto out; 1061 goto out;
1063 BUG_ON(new_i_size < i_size_read(inode 1062 BUG_ON(new_i_size < i_size_read(inode));
1064 1063
1065 /* 1064 /*
1066 * The alloc sem blocks people in rea 1065 * The alloc sem blocks people in read/write from reading our
1067 * allocation until we're done changi 1066 * allocation until we're done changing it. We depend on
1068 * i_rwsem to block other extend/trun 1067 * i_rwsem to block other extend/truncate calls while we're
1069 * here. We even have to hold it for 1068 * here. We even have to hold it for sparse files because there
1070 * might be some tail zeroing. 1069 * might be some tail zeroing.
1071 */ 1070 */
1072 down_write(&oi->ip_alloc_sem); 1071 down_write(&oi->ip_alloc_sem);
1073 1072
1074 if (oi->ip_dyn_features & OCFS2_INLIN 1073 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1075 /* 1074 /*
1076 * We can optimize small exte 1075 * We can optimize small extends by keeping the inodes
1077 * inline data. 1076 * inline data.
1078 */ 1077 */
1079 if (ocfs2_size_fits_inline_da 1078 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1080 up_write(&oi->ip_allo 1079 up_write(&oi->ip_alloc_sem);
1081 goto out_update_size; 1080 goto out_update_size;
1082 } 1081 }
1083 1082
1084 ret = ocfs2_convert_inline_da 1083 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1085 if (ret) { 1084 if (ret) {
1086 up_write(&oi->ip_allo 1085 up_write(&oi->ip_alloc_sem);
1087 mlog_errno(ret); 1086 mlog_errno(ret);
1088 goto out; 1087 goto out;
1089 } 1088 }
1090 } 1089 }
1091 1090
1092 if (ocfs2_sparse_alloc(OCFS2_SB(inode 1091 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1093 ret = ocfs2_zero_extend(inode 1092 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1094 else 1093 else
1095 ret = ocfs2_extend_no_holes(i 1094 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1096 n 1095 new_i_size);
1097 1096
1098 up_write(&oi->ip_alloc_sem); 1097 up_write(&oi->ip_alloc_sem);
1099 1098
1100 if (ret < 0) { 1099 if (ret < 0) {
1101 mlog_errno(ret); 1100 mlog_errno(ret);
1102 goto out; 1101 goto out;
1103 } 1102 }
1104 1103
1105 out_update_size: 1104 out_update_size:
1106 ret = ocfs2_simple_size_update(inode, 1105 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1107 if (ret < 0) 1106 if (ret < 0)
1108 mlog_errno(ret); 1107 mlog_errno(ret);
1109 1108
1110 out: 1109 out:
1111 return ret; 1110 return ret;
1112 } 1111 }
1113 1112
1114 int ocfs2_setattr(struct mnt_idmap *idmap, st 1113 int ocfs2_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
1115 struct iattr *attr) 1114 struct iattr *attr)
1116 { 1115 {
1117 int status = 0, size_change; 1116 int status = 0, size_change;
1118 int inode_locked = 0; 1117 int inode_locked = 0;
1119 struct inode *inode = d_inode(dentry) 1118 struct inode *inode = d_inode(dentry);
1120 struct super_block *sb = inode->i_sb; 1119 struct super_block *sb = inode->i_sb;
1121 struct ocfs2_super *osb = OCFS2_SB(sb 1120 struct ocfs2_super *osb = OCFS2_SB(sb);
1122 struct buffer_head *bh = NULL; 1121 struct buffer_head *bh = NULL;
1123 handle_t *handle = NULL; 1122 handle_t *handle = NULL;
1124 struct dquot *transfer_to[MAXQUOTAS] 1123 struct dquot *transfer_to[MAXQUOTAS] = { };
1125 int qtype; 1124 int qtype;
1126 int had_lock; 1125 int had_lock;
1127 struct ocfs2_lock_holder oh; 1126 struct ocfs2_lock_holder oh;
1128 1127
1129 trace_ocfs2_setattr(inode, dentry, 1128 trace_ocfs2_setattr(inode, dentry,
1130 (unsigned long lo 1129 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1131 dentry->d_name.le 1130 dentry->d_name.len, dentry->d_name.name,
1132 attr->ia_valid, !! 1131 attr->ia_valid, attr->ia_mode,
1133 attr->ia_vali !! 1132 from_kuid(&init_user_ns, attr->ia_uid),
1134 attr->ia_vali !! 1133 from_kgid(&init_user_ns, attr->ia_gid));
1135 from_ <<
1136 attr->ia_vali <<
1137 from_ <<
1138 1134
1139 /* ensuring we don't even attempt to 1135 /* ensuring we don't even attempt to truncate a symlink */
1140 if (S_ISLNK(inode->i_mode)) 1136 if (S_ISLNK(inode->i_mode))
1141 attr->ia_valid &= ~ATTR_SIZE; 1137 attr->ia_valid &= ~ATTR_SIZE;
1142 1138
1143 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_ 1139 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1144 | ATTR_GID | ATTR_ 1140 | ATTR_GID | ATTR_UID | ATTR_MODE)
1145 if (!(attr->ia_valid & OCFS2_VALID_AT 1141 if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1146 return 0; 1142 return 0;
1147 1143
1148 status = setattr_prepare(&nop_mnt_idm 1144 status = setattr_prepare(&nop_mnt_idmap, dentry, attr);
1149 if (status) 1145 if (status)
1150 return status; 1146 return status;
1151 1147
1152 if (is_quota_modification(&nop_mnt_id 1148 if (is_quota_modification(&nop_mnt_idmap, inode, attr)) {
1153 status = dquot_initialize(ino 1149 status = dquot_initialize(inode);
1154 if (status) 1150 if (status)
1155 return status; 1151 return status;
1156 } 1152 }
1157 size_change = S_ISREG(inode->i_mode) 1153 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1158 if (size_change) { 1154 if (size_change) {
1159 /* 1155 /*
1160 * Here we should wait dio to 1156 * Here we should wait dio to finish before inode lock
1161 * to avoid a deadlock betwee 1157 * to avoid a deadlock between ocfs2_setattr() and
1162 * ocfs2_dio_end_io_write() 1158 * ocfs2_dio_end_io_write()
1163 */ 1159 */
1164 inode_dio_wait(inode); 1160 inode_dio_wait(inode);
1165 1161
1166 status = ocfs2_rw_lock(inode, 1162 status = ocfs2_rw_lock(inode, 1);
1167 if (status < 0) { 1163 if (status < 0) {
1168 mlog_errno(status); 1164 mlog_errno(status);
1169 goto bail; 1165 goto bail;
1170 } 1166 }
1171 } 1167 }
1172 1168
1173 had_lock = ocfs2_inode_lock_tracker(i 1169 had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1174 if (had_lock < 0) { 1170 if (had_lock < 0) {
1175 status = had_lock; 1171 status = had_lock;
1176 goto bail_unlock_rw; 1172 goto bail_unlock_rw;
1177 } else if (had_lock) { 1173 } else if (had_lock) {
1178 /* 1174 /*
1179 * As far as we know, ocfs2_s 1175 * As far as we know, ocfs2_setattr() could only be the first
1180 * VFS entry point in the cal 1176 * VFS entry point in the call chain of recursive cluster
1181 * locking issue. 1177 * locking issue.
1182 * 1178 *
1183 * For instance: 1179 * For instance:
1184 * chmod_common() 1180 * chmod_common()
1185 * notify_change() 1181 * notify_change()
1186 * ocfs2_setattr() 1182 * ocfs2_setattr()
1187 * posix_acl_chmod() 1183 * posix_acl_chmod()
1188 * ocfs2_iop_get_acl() 1184 * ocfs2_iop_get_acl()
1189 * 1185 *
1190 * But, we're not 100% sure i 1186 * But, we're not 100% sure if it's always true, because the
1191 * ordering of the VFS entry 1187 * ordering of the VFS entry points in the call chain is out
1192 * of our control. So, we'd b 1188 * of our control. So, we'd better dump the stack here to
1193 * catch the other cases of r 1189 * catch the other cases of recursive locking.
1194 */ 1190 */
1195 mlog(ML_ERROR, "Another case 1191 mlog(ML_ERROR, "Another case of recursive locking:\n");
1196 dump_stack(); 1192 dump_stack();
1197 } 1193 }
1198 inode_locked = 1; 1194 inode_locked = 1;
1199 1195
1200 if (size_change) { 1196 if (size_change) {
1201 status = inode_newsize_ok(ino 1197 status = inode_newsize_ok(inode, attr->ia_size);
1202 if (status) 1198 if (status)
1203 goto bail_unlock; 1199 goto bail_unlock;
1204 1200
1205 if (i_size_read(inode) >= att 1201 if (i_size_read(inode) >= attr->ia_size) {
1206 if (ocfs2_should_orde 1202 if (ocfs2_should_order_data(inode)) {
1207 status = ocfs 1203 status = ocfs2_begin_ordered_truncate(inode,
1208 1204 attr->ia_size);
1209 if (status) 1205 if (status)
1210 goto 1206 goto bail_unlock;
1211 } 1207 }
1212 status = ocfs2_trunca 1208 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1213 } else 1209 } else
1214 status = ocfs2_extend 1210 status = ocfs2_extend_file(inode, bh, attr->ia_size);
1215 if (status < 0) { 1211 if (status < 0) {
1216 if (status != -ENOSPC 1212 if (status != -ENOSPC)
1217 mlog_errno(st 1213 mlog_errno(status);
1218 status = -ENOSPC; 1214 status = -ENOSPC;
1219 goto bail_unlock; 1215 goto bail_unlock;
1220 } 1216 }
1221 } 1217 }
1222 1218
1223 if ((attr->ia_valid & ATTR_UID && !ui 1219 if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1224 (attr->ia_valid & ATTR_GID && !gi 1220 (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1225 /* 1221 /*
1226 * Gather pointers to quota s 1222 * Gather pointers to quota structures so that allocation /
1227 * freeing of quota structure 1223 * freeing of quota structures happens here and not inside
1228 * dquot_transfer() where we 1224 * dquot_transfer() where we have problems with lock ordering
1229 */ 1225 */
1230 if (attr->ia_valid & ATTR_UID 1226 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1231 && OCFS2_HAS_RO_COMPAT_FE 1227 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1232 OCFS2_FEATURE_RO_COMPAT_U 1228 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1233 transfer_to[USRQUOTA] 1229 transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1234 if (IS_ERR(transfer_t 1230 if (IS_ERR(transfer_to[USRQUOTA])) {
1235 status = PTR_ 1231 status = PTR_ERR(transfer_to[USRQUOTA]);
1236 transfer_to[U 1232 transfer_to[USRQUOTA] = NULL;
1237 goto bail_unl 1233 goto bail_unlock;
1238 } 1234 }
1239 } 1235 }
1240 if (attr->ia_valid & ATTR_GID 1236 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1241 && OCFS2_HAS_RO_COMPAT_FE 1237 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1242 OCFS2_FEATURE_RO_COMPAT_G 1238 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1243 transfer_to[GRPQUOTA] 1239 transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1244 if (IS_ERR(transfer_t 1240 if (IS_ERR(transfer_to[GRPQUOTA])) {
1245 status = PTR_ 1241 status = PTR_ERR(transfer_to[GRPQUOTA]);
1246 transfer_to[G 1242 transfer_to[GRPQUOTA] = NULL;
1247 goto bail_unl 1243 goto bail_unlock;
1248 } 1244 }
1249 } 1245 }
1250 down_write(&OCFS2_I(inode)->i 1246 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1251 handle = ocfs2_start_trans(os 1247 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1252 2 1248 2 * ocfs2_quota_trans_credits(sb));
1253 if (IS_ERR(handle)) { 1249 if (IS_ERR(handle)) {
1254 status = PTR_ERR(hand 1250 status = PTR_ERR(handle);
1255 mlog_errno(status); 1251 mlog_errno(status);
1256 goto bail_unlock_allo 1252 goto bail_unlock_alloc;
1257 } 1253 }
1258 status = __dquot_transfer(ino 1254 status = __dquot_transfer(inode, transfer_to);
1259 if (status < 0) 1255 if (status < 0)
1260 goto bail_commit; 1256 goto bail_commit;
1261 } else { 1257 } else {
1262 down_write(&OCFS2_I(inode)->i 1258 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1263 handle = ocfs2_start_trans(os 1259 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1264 if (IS_ERR(handle)) { 1260 if (IS_ERR(handle)) {
1265 status = PTR_ERR(hand 1261 status = PTR_ERR(handle);
1266 mlog_errno(status); 1262 mlog_errno(status);
1267 goto bail_unlock_allo 1263 goto bail_unlock_alloc;
1268 } 1264 }
1269 } 1265 }
1270 1266
1271 setattr_copy(&nop_mnt_idmap, inode, a 1267 setattr_copy(&nop_mnt_idmap, inode, attr);
1272 mark_inode_dirty(inode); 1268 mark_inode_dirty(inode);
1273 1269
1274 status = ocfs2_mark_inode_dirty(handl 1270 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1275 if (status < 0) 1271 if (status < 0)
1276 mlog_errno(status); 1272 mlog_errno(status);
1277 1273
1278 bail_commit: 1274 bail_commit:
1279 ocfs2_commit_trans(osb, handle); 1275 ocfs2_commit_trans(osb, handle);
1280 bail_unlock_alloc: 1276 bail_unlock_alloc:
1281 up_write(&OCFS2_I(inode)->ip_alloc_se 1277 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1282 bail_unlock: 1278 bail_unlock:
1283 if (status && inode_locked) { 1279 if (status && inode_locked) {
1284 ocfs2_inode_unlock_tracker(in 1280 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1285 inode_locked = 0; 1281 inode_locked = 0;
1286 } 1282 }
1287 bail_unlock_rw: 1283 bail_unlock_rw:
1288 if (size_change) 1284 if (size_change)
1289 ocfs2_rw_unlock(inode, 1); 1285 ocfs2_rw_unlock(inode, 1);
1290 bail: 1286 bail:
1291 1287
1292 /* Release quota pointers in case we 1288 /* Release quota pointers in case we acquired them */
1293 for (qtype = 0; qtype < OCFS2_MAXQUOT 1289 for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1294 dqput(transfer_to[qtype]); 1290 dqput(transfer_to[qtype]);
1295 1291
1296 if (!status && attr->ia_valid & ATTR_ 1292 if (!status && attr->ia_valid & ATTR_MODE) {
1297 status = ocfs2_acl_chmod(inod 1293 status = ocfs2_acl_chmod(inode, bh);
1298 if (status < 0) 1294 if (status < 0)
1299 mlog_errno(status); 1295 mlog_errno(status);
1300 } 1296 }
1301 if (inode_locked) 1297 if (inode_locked)
1302 ocfs2_inode_unlock_tracker(in 1298 ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1303 1299
1304 brelse(bh); 1300 brelse(bh);
1305 return status; 1301 return status;
1306 } 1302 }
1307 1303
1308 int ocfs2_getattr(struct mnt_idmap *idmap, co 1304 int ocfs2_getattr(struct mnt_idmap *idmap, const struct path *path,
1309 struct kstat *stat, u32 req 1305 struct kstat *stat, u32 request_mask, unsigned int flags)
1310 { 1306 {
1311 struct inode *inode = d_inode(path->d 1307 struct inode *inode = d_inode(path->dentry);
1312 struct super_block *sb = path->dentry 1308 struct super_block *sb = path->dentry->d_sb;
1313 struct ocfs2_super *osb = sb->s_fs_in 1309 struct ocfs2_super *osb = sb->s_fs_info;
1314 int err; 1310 int err;
1315 1311
1316 err = ocfs2_inode_revalidate(path->de 1312 err = ocfs2_inode_revalidate(path->dentry);
1317 if (err) { 1313 if (err) {
1318 if (err != -ENOENT) 1314 if (err != -ENOENT)
1319 mlog_errno(err); 1315 mlog_errno(err);
1320 goto bail; 1316 goto bail;
1321 } 1317 }
1322 1318
1323 generic_fillattr(&nop_mnt_idmap, requ 1319 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1324 /* 1320 /*
1325 * If there is inline data in the ino 1321 * If there is inline data in the inode, the inode will normally not
1326 * have data blocks allocated (it may 1322 * have data blocks allocated (it may have an external xattr block).
1327 * Report at least one sector for suc 1323 * Report at least one sector for such files, so tools like tar, rsync,
1328 * others don't incorrectly think the 1324 * others don't incorrectly think the file is completely sparse.
1329 */ 1325 */
1330 if (unlikely(OCFS2_I(inode)->ip_dyn_f 1326 if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1331 stat->blocks += (stat->size + 1327 stat->blocks += (stat->size + 511)>>9;
1332 1328
1333 /* We set the blksize from the cluste 1329 /* We set the blksize from the cluster size for performance */
1334 stat->blksize = osb->s_clustersize; 1330 stat->blksize = osb->s_clustersize;
1335 1331
1336 bail: 1332 bail:
1337 return err; 1333 return err;
1338 } 1334 }
1339 1335
1340 int ocfs2_permission(struct mnt_idmap *idmap, 1336 int ocfs2_permission(struct mnt_idmap *idmap, struct inode *inode,
1341 int mask) 1337 int mask)
1342 { 1338 {
1343 int ret, had_lock; 1339 int ret, had_lock;
1344 struct ocfs2_lock_holder oh; 1340 struct ocfs2_lock_holder oh;
1345 1341
1346 if (mask & MAY_NOT_BLOCK) 1342 if (mask & MAY_NOT_BLOCK)
1347 return -ECHILD; 1343 return -ECHILD;
1348 1344
1349 had_lock = ocfs2_inode_lock_tracker(i 1345 had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1350 if (had_lock < 0) { 1346 if (had_lock < 0) {
1351 ret = had_lock; 1347 ret = had_lock;
1352 goto out; 1348 goto out;
1353 } else if (had_lock) { 1349 } else if (had_lock) {
1354 /* See comments in ocfs2_seta 1350 /* See comments in ocfs2_setattr() for details.
1355 * The call chain of this cas 1351 * The call chain of this case could be:
1356 * do_sys_open() 1352 * do_sys_open()
1357 * may_open() 1353 * may_open()
1358 * inode_permission() 1354 * inode_permission()
1359 * ocfs2_permission() 1355 * ocfs2_permission()
1360 * ocfs2_iop_get_acl() 1356 * ocfs2_iop_get_acl()
1361 */ 1357 */
1362 mlog(ML_ERROR, "Another case 1358 mlog(ML_ERROR, "Another case of recursive locking:\n");
1363 dump_stack(); 1359 dump_stack();
1364 } 1360 }
1365 1361
1366 ret = generic_permission(&nop_mnt_idm 1362 ret = generic_permission(&nop_mnt_idmap, inode, mask);
1367 1363
1368 ocfs2_inode_unlock_tracker(inode, 0, 1364 ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1369 out: 1365 out:
1370 return ret; 1366 return ret;
1371 } 1367 }
1372 1368
1373 static int __ocfs2_write_remove_suid(struct i 1369 static int __ocfs2_write_remove_suid(struct inode *inode,
1374 struct b 1370 struct buffer_head *bh)
1375 { 1371 {
1376 int ret; 1372 int ret;
1377 handle_t *handle; 1373 handle_t *handle;
1378 struct ocfs2_super *osb = OCFS2_SB(in 1374 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1379 struct ocfs2_dinode *di; 1375 struct ocfs2_dinode *di;
1380 1376
1381 trace_ocfs2_write_remove_suid( 1377 trace_ocfs2_write_remove_suid(
1382 (unsigned long long)O 1378 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1383 inode->i_mode); 1379 inode->i_mode);
1384 1380
1385 handle = ocfs2_start_trans(osb, OCFS2 1381 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1386 if (IS_ERR(handle)) { 1382 if (IS_ERR(handle)) {
1387 ret = PTR_ERR(handle); 1383 ret = PTR_ERR(handle);
1388 mlog_errno(ret); 1384 mlog_errno(ret);
1389 goto out; 1385 goto out;
1390 } 1386 }
1391 1387
1392 ret = ocfs2_journal_access_di(handle, 1388 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1393 OCFS2_J 1389 OCFS2_JOURNAL_ACCESS_WRITE);
1394 if (ret < 0) { 1390 if (ret < 0) {
1395 mlog_errno(ret); 1391 mlog_errno(ret);
1396 goto out_trans; 1392 goto out_trans;
1397 } 1393 }
1398 1394
1399 inode->i_mode &= ~S_ISUID; 1395 inode->i_mode &= ~S_ISUID;
1400 if ((inode->i_mode & S_ISGID) && (ino 1396 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1401 inode->i_mode &= ~S_ISGID; 1397 inode->i_mode &= ~S_ISGID;
1402 1398
1403 di = (struct ocfs2_dinode *) bh->b_da 1399 di = (struct ocfs2_dinode *) bh->b_data;
1404 di->i_mode = cpu_to_le16(inode->i_mod 1400 di->i_mode = cpu_to_le16(inode->i_mode);
1405 ocfs2_update_inode_fsync_trans(handle 1401 ocfs2_update_inode_fsync_trans(handle, inode, 0);
1406 1402
1407 ocfs2_journal_dirty(handle, bh); 1403 ocfs2_journal_dirty(handle, bh);
1408 1404
1409 out_trans: 1405 out_trans:
1410 ocfs2_commit_trans(osb, handle); 1406 ocfs2_commit_trans(osb, handle);
1411 out: 1407 out:
1412 return ret; 1408 return ret;
1413 } 1409 }
1414 1410
1415 static int ocfs2_write_remove_suid(struct ino 1411 static int ocfs2_write_remove_suid(struct inode *inode)
1416 { 1412 {
1417 int ret; 1413 int ret;
1418 struct buffer_head *bh = NULL; 1414 struct buffer_head *bh = NULL;
1419 1415
1420 ret = ocfs2_read_inode_block(inode, & 1416 ret = ocfs2_read_inode_block(inode, &bh);
1421 if (ret < 0) { 1417 if (ret < 0) {
1422 mlog_errno(ret); 1418 mlog_errno(ret);
1423 goto out; 1419 goto out;
1424 } 1420 }
1425 1421
1426 ret = __ocfs2_write_remove_suid(inod 1422 ret = __ocfs2_write_remove_suid(inode, bh);
1427 out: 1423 out:
1428 brelse(bh); 1424 brelse(bh);
1429 return ret; 1425 return ret;
1430 } 1426 }
1431 1427
1432 /* 1428 /*
1433 * Allocate enough extents to cover the regio 1429 * Allocate enough extents to cover the region starting at byte offset
1434 * start for len bytes. Existing extents are 1430 * start for len bytes. Existing extents are skipped, any extents
1435 * added are marked as "unwritten". 1431 * added are marked as "unwritten".
1436 */ 1432 */
1437 static int ocfs2_allocate_unwritten_extents(s 1433 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1438 u 1434 u64 start, u64 len)
1439 { 1435 {
1440 int ret; 1436 int ret;
1441 u32 cpos, phys_cpos, clusters, alloc_ 1437 u32 cpos, phys_cpos, clusters, alloc_size;
1442 u64 end = start + len; 1438 u64 end = start + len;
1443 struct buffer_head *di_bh = NULL; 1439 struct buffer_head *di_bh = NULL;
1444 1440
1445 if (OCFS2_I(inode)->ip_dyn_features & 1441 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1446 ret = ocfs2_read_inode_block( 1442 ret = ocfs2_read_inode_block(inode, &di_bh);
1447 if (ret) { 1443 if (ret) {
1448 mlog_errno(ret); 1444 mlog_errno(ret);
1449 goto out; 1445 goto out;
1450 } 1446 }
1451 1447
1452 /* 1448 /*
1453 * Nothing to do if the reque 1449 * Nothing to do if the requested reservation range
1454 * fits within the inode. 1450 * fits within the inode.
1455 */ 1451 */
1456 if (ocfs2_size_fits_inline_da 1452 if (ocfs2_size_fits_inline_data(di_bh, end))
1457 goto out; 1453 goto out;
1458 1454
1459 ret = ocfs2_convert_inline_da 1455 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1460 if (ret) { 1456 if (ret) {
1461 mlog_errno(ret); 1457 mlog_errno(ret);
1462 goto out; 1458 goto out;
1463 } 1459 }
1464 } 1460 }
1465 1461
1466 /* 1462 /*
1467 * We consider both start and len to 1463 * We consider both start and len to be inclusive.
1468 */ 1464 */
1469 cpos = start >> OCFS2_SB(inode->i_sb) 1465 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1470 clusters = ocfs2_clusters_for_bytes(i 1466 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1471 clusters -= cpos; 1467 clusters -= cpos;
1472 1468
1473 while (clusters) { 1469 while (clusters) {
1474 ret = ocfs2_get_clusters(inod 1470 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1475 &all 1471 &alloc_size, NULL);
1476 if (ret) { 1472 if (ret) {
1477 mlog_errno(ret); 1473 mlog_errno(ret);
1478 goto out; 1474 goto out;
1479 } 1475 }
1480 1476
1481 /* 1477 /*
1482 * Hole or existing extent le 1478 * Hole or existing extent len can be arbitrary, so
1483 * cap it to our own allocati 1479 * cap it to our own allocation request.
1484 */ 1480 */
1485 if (alloc_size > clusters) 1481 if (alloc_size > clusters)
1486 alloc_size = clusters 1482 alloc_size = clusters;
1487 1483
1488 if (phys_cpos) { 1484 if (phys_cpos) {
1489 /* 1485 /*
1490 * We already have an 1486 * We already have an allocation at this
1491 * region so we can s 1487 * region so we can safely skip it.
1492 */ 1488 */
1493 goto next; 1489 goto next;
1494 } 1490 }
1495 1491
1496 ret = ocfs2_extend_allocation 1492 ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1497 if (ret) { 1493 if (ret) {
1498 if (ret != -ENOSPC) 1494 if (ret != -ENOSPC)
1499 mlog_errno(re 1495 mlog_errno(ret);
1500 goto out; 1496 goto out;
1501 } 1497 }
1502 1498
1503 next: 1499 next:
1504 cpos += alloc_size; 1500 cpos += alloc_size;
1505 clusters -= alloc_size; 1501 clusters -= alloc_size;
1506 } 1502 }
1507 1503
1508 ret = 0; 1504 ret = 0;
1509 out: 1505 out:
1510 1506
1511 brelse(di_bh); 1507 brelse(di_bh);
1512 return ret; 1508 return ret;
1513 } 1509 }
1514 1510
1515 /* 1511 /*
1516 * Truncate a byte range, avoiding pages with 1512 * Truncate a byte range, avoiding pages within partial clusters. This
1517 * preserves those pages for the zeroing code 1513 * preserves those pages for the zeroing code to write to.
1518 */ 1514 */
1519 static void ocfs2_truncate_cluster_pages(stru 1515 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1520 u64 1516 u64 byte_len)
1521 { 1517 {
1522 struct ocfs2_super *osb = OCFS2_SB(in 1518 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1523 loff_t start, end; 1519 loff_t start, end;
1524 struct address_space *mapping = inode 1520 struct address_space *mapping = inode->i_mapping;
1525 1521
1526 start = (loff_t)ocfs2_align_bytes_to_ 1522 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1527 end = byte_start + byte_len; 1523 end = byte_start + byte_len;
1528 end = end & ~(osb->s_clustersize - 1) 1524 end = end & ~(osb->s_clustersize - 1);
1529 1525
1530 if (start < end) { 1526 if (start < end) {
1531 unmap_mapping_range(mapping, 1527 unmap_mapping_range(mapping, start, end - start, 0);
1532 truncate_inode_pages_range(ma 1528 truncate_inode_pages_range(mapping, start, end - 1);
1533 } 1529 }
1534 } 1530 }
1535 1531
1536 /* 1532 /*
1537 * zero out partial blocks of one cluster. 1533 * zero out partial blocks of one cluster.
1538 * 1534 *
1539 * start: file offset where zero starts, will 1535 * start: file offset where zero starts, will be made upper block aligned.
1540 * len: it will be trimmed to the end of curr 1536 * len: it will be trimmed to the end of current cluster if "start + len"
1541 * is bigger than it. 1537 * is bigger than it.
1542 */ 1538 */
1543 static int ocfs2_zeroout_partial_cluster(stru 1539 static int ocfs2_zeroout_partial_cluster(struct inode *inode,
1544 u64 s 1540 u64 start, u64 len)
1545 { 1541 {
1546 int ret; 1542 int ret;
1547 u64 start_block, end_block, nr_blocks 1543 u64 start_block, end_block, nr_blocks;
1548 u64 p_block, offset; 1544 u64 p_block, offset;
1549 u32 cluster, p_cluster, nr_clusters; 1545 u32 cluster, p_cluster, nr_clusters;
1550 struct super_block *sb = inode->i_sb; 1546 struct super_block *sb = inode->i_sb;
1551 u64 end = ocfs2_align_bytes_to_cluste 1547 u64 end = ocfs2_align_bytes_to_clusters(sb, start);
1552 1548
1553 if (start + len < end) 1549 if (start + len < end)
1554 end = start + len; 1550 end = start + len;
1555 1551
1556 start_block = ocfs2_blocks_for_bytes( 1552 start_block = ocfs2_blocks_for_bytes(sb, start);
1557 end_block = ocfs2_blocks_for_bytes(sb 1553 end_block = ocfs2_blocks_for_bytes(sb, end);
1558 nr_blocks = end_block - start_block; 1554 nr_blocks = end_block - start_block;
1559 if (!nr_blocks) 1555 if (!nr_blocks)
1560 return 0; 1556 return 0;
1561 1557
1562 cluster = ocfs2_bytes_to_clusters(sb, 1558 cluster = ocfs2_bytes_to_clusters(sb, start);
1563 ret = ocfs2_get_clusters(inode, clust 1559 ret = ocfs2_get_clusters(inode, cluster, &p_cluster,
1564 &nr_clusters, 1560 &nr_clusters, NULL);
1565 if (ret) 1561 if (ret)
1566 return ret; 1562 return ret;
1567 if (!p_cluster) 1563 if (!p_cluster)
1568 return 0; 1564 return 0;
1569 1565
1570 offset = start_block - ocfs2_clusters 1566 offset = start_block - ocfs2_clusters_to_blocks(sb, cluster);
1571 p_block = ocfs2_clusters_to_blocks(sb 1567 p_block = ocfs2_clusters_to_blocks(sb, p_cluster) + offset;
1572 return sb_issue_zeroout(sb, p_block, 1568 return sb_issue_zeroout(sb, p_block, nr_blocks, GFP_NOFS);
1573 } 1569 }
1574 1570
1575 static int ocfs2_zero_partial_clusters(struct 1571 static int ocfs2_zero_partial_clusters(struct inode *inode,
1576 u64 st 1572 u64 start, u64 len)
1577 { 1573 {
1578 int ret = 0; 1574 int ret = 0;
1579 u64 tmpend = 0; 1575 u64 tmpend = 0;
1580 u64 end = start + len; 1576 u64 end = start + len;
1581 struct ocfs2_super *osb = OCFS2_SB(in 1577 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1582 unsigned int csize = osb->s_clustersi 1578 unsigned int csize = osb->s_clustersize;
1583 handle_t *handle; 1579 handle_t *handle;
1584 loff_t isize = i_size_read(inode); 1580 loff_t isize = i_size_read(inode);
1585 1581
1586 /* 1582 /*
1587 * The "start" and "end" values are N 1583 * The "start" and "end" values are NOT necessarily part of
1588 * the range whose allocation is bein 1584 * the range whose allocation is being deleted. Rather, this
1589 * is what the user passed in with th 1585 * is what the user passed in with the request. We must zero
1590 * partial clusters here. There's no 1586 * partial clusters here. There's no need to worry about
1591 * physical allocation - the zeroing 1587 * physical allocation - the zeroing code knows to skip holes.
1592 */ 1588 */
1593 trace_ocfs2_zero_partial_clusters( 1589 trace_ocfs2_zero_partial_clusters(
1594 (unsigned long long)OCFS2_I(i 1590 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1595 (unsigned long long)start, (u 1591 (unsigned long long)start, (unsigned long long)end);
1596 1592
1597 /* 1593 /*
1598 * If both edges are on a cluster bou 1594 * If both edges are on a cluster boundary then there's no
1599 * zeroing required as the region is 1595 * zeroing required as the region is part of the allocation to
1600 * be truncated. 1596 * be truncated.
1601 */ 1597 */
1602 if ((start & (csize - 1)) == 0 && (en 1598 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1603 goto out; 1599 goto out;
1604 1600
1605 /* No page cache for EOF blocks, issu 1601 /* No page cache for EOF blocks, issue zero out to disk. */
1606 if (end > isize) { 1602 if (end > isize) {
1607 /* 1603 /*
1608 * zeroout eof blocks in last 1604 * zeroout eof blocks in last cluster starting from
1609 * "isize" even "start" > "is 1605 * "isize" even "start" > "isize" because it is
1610 * complicated to zeroout jus 1606 * complicated to zeroout just at "start" as "start"
1611 * may be not aligned with bl 1607 * may be not aligned with block size, buffer write
1612 * would be required to do th 1608 * would be required to do that, but out of eof buffer
1613 * write is not supported. 1609 * write is not supported.
1614 */ 1610 */
1615 ret = ocfs2_zeroout_partial_c 1611 ret = ocfs2_zeroout_partial_cluster(inode, isize,
1616 end - 1612 end - isize);
1617 if (ret) { 1613 if (ret) {
1618 mlog_errno(ret); 1614 mlog_errno(ret);
1619 goto out; 1615 goto out;
1620 } 1616 }
1621 if (start >= isize) 1617 if (start >= isize)
1622 goto out; 1618 goto out;
1623 end = isize; 1619 end = isize;
1624 } 1620 }
1625 handle = ocfs2_start_trans(osb, OCFS2 1621 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1626 if (IS_ERR(handle)) { 1622 if (IS_ERR(handle)) {
1627 ret = PTR_ERR(handle); 1623 ret = PTR_ERR(handle);
1628 mlog_errno(ret); 1624 mlog_errno(ret);
1629 goto out; 1625 goto out;
1630 } 1626 }
1631 1627
1632 /* 1628 /*
1633 * If start is on a cluster boundary 1629 * If start is on a cluster boundary and end is somewhere in another
1634 * cluster, we have not COWed the clu 1630 * cluster, we have not COWed the cluster starting at start, unless
1635 * end is also within the same cluste 1631 * end is also within the same cluster. So, in this case, we skip this
1636 * first call to ocfs2_zero_range_for 1632 * first call to ocfs2_zero_range_for_truncate() truncate and move on
1637 * to the next one. 1633 * to the next one.
1638 */ 1634 */
1639 if ((start & (csize - 1)) != 0) { 1635 if ((start & (csize - 1)) != 0) {
1640 /* 1636 /*
1641 * We want to get the byte of 1637 * We want to get the byte offset of the end of the 1st
1642 * cluster. 1638 * cluster.
1643 */ 1639 */
1644 tmpend = (u64)osb->s_clusters 1640 tmpend = (u64)osb->s_clustersize +
1645 (start & ~(osb->s_clu 1641 (start & ~(osb->s_clustersize - 1));
1646 if (tmpend > end) 1642 if (tmpend > end)
1647 tmpend = end; 1643 tmpend = end;
1648 1644
1649 trace_ocfs2_zero_partial_clus 1645 trace_ocfs2_zero_partial_clusters_range1(
1650 (unsigned long long)s 1646 (unsigned long long)start,
1651 (unsigned long long)t 1647 (unsigned long long)tmpend);
1652 1648
1653 ret = ocfs2_zero_range_for_tr 1649 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1654 1650 tmpend);
1655 if (ret) 1651 if (ret)
1656 mlog_errno(ret); 1652 mlog_errno(ret);
1657 } 1653 }
1658 1654
1659 if (tmpend < end) { 1655 if (tmpend < end) {
1660 /* 1656 /*
1661 * This may make start and en 1657 * This may make start and end equal, but the zeroing
1662 * code will skip any work in 1658 * code will skip any work in that case so there's no
1663 * need to catch it up here. 1659 * need to catch it up here.
1664 */ 1660 */
1665 start = end & ~(osb->s_cluste 1661 start = end & ~(osb->s_clustersize - 1);
1666 1662
1667 trace_ocfs2_zero_partial_clus 1663 trace_ocfs2_zero_partial_clusters_range2(
1668 (unsigned long long)s 1664 (unsigned long long)start, (unsigned long long)end);
1669 1665
1670 ret = ocfs2_zero_range_for_tr 1666 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1671 if (ret) 1667 if (ret)
1672 mlog_errno(ret); 1668 mlog_errno(ret);
1673 } 1669 }
1674 ocfs2_update_inode_fsync_trans(handle 1670 ocfs2_update_inode_fsync_trans(handle, inode, 1);
1675 1671
1676 ocfs2_commit_trans(osb, handle); 1672 ocfs2_commit_trans(osb, handle);
1677 out: 1673 out:
1678 return ret; 1674 return ret;
1679 } 1675 }
1680 1676
1681 static int ocfs2_find_rec(struct ocfs2_extent 1677 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1682 { 1678 {
1683 int i; 1679 int i;
1684 struct ocfs2_extent_rec *rec = NULL; 1680 struct ocfs2_extent_rec *rec = NULL;
1685 1681
1686 for (i = le16_to_cpu(el->l_next_free_ 1682 for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1687 1683
1688 rec = &el->l_recs[i]; 1684 rec = &el->l_recs[i];
1689 1685
1690 if (le32_to_cpu(rec->e_cpos) 1686 if (le32_to_cpu(rec->e_cpos) < pos)
1691 break; 1687 break;
1692 } 1688 }
1693 1689
1694 return i; 1690 return i;
1695 } 1691 }
1696 1692
1697 /* 1693 /*
1698 * Helper to calculate the punching pos and l 1694 * Helper to calculate the punching pos and length in one run, we handle the
1699 * following three cases in order: 1695 * following three cases in order:
1700 * 1696 *
1701 * - remove the entire record 1697 * - remove the entire record
1702 * - remove a partial record 1698 * - remove a partial record
1703 * - no record needs to be removed (hole-punc 1699 * - no record needs to be removed (hole-punching completed)
1704 */ 1700 */
1705 static void ocfs2_calc_trunc_pos(struct inode 1701 static void ocfs2_calc_trunc_pos(struct inode *inode,
1706 struct ocfs2 1702 struct ocfs2_extent_list *el,
1707 struct ocfs2 1703 struct ocfs2_extent_rec *rec,
1708 u32 trunc_st 1704 u32 trunc_start, u32 *trunc_cpos,
1709 u32 *trunc_l 1705 u32 *trunc_len, u32 *trunc_end,
1710 u64 *blkno, 1706 u64 *blkno, int *done)
1711 { 1707 {
1712 int ret = 0; 1708 int ret = 0;
1713 u32 coff, range; 1709 u32 coff, range;
1714 1710
1715 range = le32_to_cpu(rec->e_cpos) + oc 1711 range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1716 1712
1717 if (le32_to_cpu(rec->e_cpos) >= trunc 1713 if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1718 /* 1714 /*
1719 * remove an entire extent re 1715 * remove an entire extent record.
1720 */ 1716 */
1721 *trunc_cpos = le32_to_cpu(rec 1717 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1722 /* 1718 /*
1723 * Skip holes if any. 1719 * Skip holes if any.
1724 */ 1720 */
1725 if (range < *trunc_end) 1721 if (range < *trunc_end)
1726 *trunc_end = range; 1722 *trunc_end = range;
1727 *trunc_len = *trunc_end - le3 1723 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1728 *blkno = le64_to_cpu(rec->e_b 1724 *blkno = le64_to_cpu(rec->e_blkno);
1729 *trunc_end = le32_to_cpu(rec- 1725 *trunc_end = le32_to_cpu(rec->e_cpos);
1730 } else if (range > trunc_start) { 1726 } else if (range > trunc_start) {
1731 /* 1727 /*
1732 * remove a partial extent re 1728 * remove a partial extent record, which means we're
1733 * removing the last extent r 1729 * removing the last extent record.
1734 */ 1730 */
1735 *trunc_cpos = trunc_start; 1731 *trunc_cpos = trunc_start;
1736 /* 1732 /*
1737 * skip hole if any. 1733 * skip hole if any.
1738 */ 1734 */
1739 if (range < *trunc_end) 1735 if (range < *trunc_end)
1740 *trunc_end = range; 1736 *trunc_end = range;
1741 *trunc_len = *trunc_end - tru 1737 *trunc_len = *trunc_end - trunc_start;
1742 coff = trunc_start - le32_to_ 1738 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1743 *blkno = le64_to_cpu(rec->e_b 1739 *blkno = le64_to_cpu(rec->e_blkno) +
1744 ocfs2_cluster 1740 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1745 *trunc_end = trunc_start; 1741 *trunc_end = trunc_start;
1746 } else { 1742 } else {
1747 /* 1743 /*
1748 * It may have two following 1744 * It may have two following possibilities:
1749 * 1745 *
1750 * - last record has been rem 1746 * - last record has been removed
1751 * - trunc_start was within a 1747 * - trunc_start was within a hole
1752 * 1748 *
1753 * both two cases mean the co 1749 * both two cases mean the completion of hole punching.
1754 */ 1750 */
1755 ret = 1; 1751 ret = 1;
1756 } 1752 }
1757 1753
1758 *done = ret; 1754 *done = ret;
1759 } 1755 }
1760 1756
1761 int ocfs2_remove_inode_range(struct inode *in 1757 int ocfs2_remove_inode_range(struct inode *inode,
1762 struct buffer_he 1758 struct buffer_head *di_bh, u64 byte_start,
1763 u64 byte_len) 1759 u64 byte_len)
1764 { 1760 {
1765 int ret = 0, flags = 0, done = 0, i; 1761 int ret = 0, flags = 0, done = 0, i;
1766 u32 trunc_start, trunc_len, trunc_end 1762 u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1767 u32 cluster_in_el; 1763 u32 cluster_in_el;
1768 struct ocfs2_super *osb = OCFS2_SB(in 1764 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1769 struct ocfs2_cached_dealloc_ctxt deal 1765 struct ocfs2_cached_dealloc_ctxt dealloc;
1770 struct address_space *mapping = inode 1766 struct address_space *mapping = inode->i_mapping;
1771 struct ocfs2_extent_tree et; 1767 struct ocfs2_extent_tree et;
1772 struct ocfs2_path *path = NULL; 1768 struct ocfs2_path *path = NULL;
1773 struct ocfs2_extent_list *el = NULL; 1769 struct ocfs2_extent_list *el = NULL;
1774 struct ocfs2_extent_rec *rec = NULL; 1770 struct ocfs2_extent_rec *rec = NULL;
1775 struct ocfs2_dinode *di = (struct ocf 1771 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1776 u64 blkno, refcount_loc = le64_to_cpu 1772 u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1777 1773
1778 ocfs2_init_dinode_extent_tree(&et, IN 1774 ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1779 ocfs2_init_dealloc_ctxt(&dealloc); 1775 ocfs2_init_dealloc_ctxt(&dealloc);
1780 1776
1781 trace_ocfs2_remove_inode_range( 1777 trace_ocfs2_remove_inode_range(
1782 (unsigned long long)O 1778 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1783 (unsigned long long)b 1779 (unsigned long long)byte_start,
1784 (unsigned long long)b 1780 (unsigned long long)byte_len);
1785 1781
1786 if (byte_len == 0) 1782 if (byte_len == 0)
1787 return 0; 1783 return 0;
1788 1784
1789 if (OCFS2_I(inode)->ip_dyn_features & 1785 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 1786 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1799 b 1787 byte_start + byte_len, 0);
1800 if (ret) { 1788 if (ret) {
1801 mlog_errno(ret); 1789 mlog_errno(ret);
1802 goto out; 1790 goto out;
1803 } 1791 }
1804 /* 1792 /*
1805 * There's no need to get fan 1793 * There's no need to get fancy with the page cache
1806 * truncate of an inline-data 1794 * truncate of an inline-data inode. We're talking
1807 * about less than a page her 1795 * about less than a page here, which will be cached
1808 * in the dinode buffer anywa 1796 * in the dinode buffer anyway.
1809 */ 1797 */
1810 unmap_mapping_range(mapping, 1798 unmap_mapping_range(mapping, 0, 0, 0);
1811 truncate_inode_pages(mapping, 1799 truncate_inode_pages(mapping, 0);
1812 goto out; 1800 goto out;
1813 } 1801 }
1814 1802
1815 /* 1803 /*
1816 * For reflinks, we may need to CoW 2 1804 * For reflinks, we may need to CoW 2 clusters which might be
1817 * partially zero'd later, if hole's 1805 * partially zero'd later, if hole's start and end offset were
1818 * within one cluster(means is not ex 1806 * within one cluster(means is not exactly aligned to clustersize).
1819 */ 1807 */
1820 1808
1821 if (ocfs2_is_refcount_inode(inode)) { 1809 if (ocfs2_is_refcount_inode(inode)) {
1822 ret = ocfs2_cow_file_pos(inod 1810 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1823 if (ret) { 1811 if (ret) {
1824 mlog_errno(ret); 1812 mlog_errno(ret);
1825 goto out; 1813 goto out;
1826 } 1814 }
1827 1815
1828 ret = ocfs2_cow_file_pos(inod 1816 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1829 if (ret) { 1817 if (ret) {
1830 mlog_errno(ret); 1818 mlog_errno(ret);
1831 goto out; 1819 goto out;
1832 } 1820 }
1833 } 1821 }
1834 1822
1835 trunc_start = ocfs2_clusters_for_byte 1823 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1836 trunc_end = (byte_start + byte_len) > 1824 trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1837 cluster_in_el = trunc_end; 1825 cluster_in_el = trunc_end;
1838 1826
1839 ret = ocfs2_zero_partial_clusters(ino 1827 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1840 if (ret) { 1828 if (ret) {
1841 mlog_errno(ret); 1829 mlog_errno(ret);
1842 goto out; 1830 goto out;
1843 } 1831 }
1844 1832
1845 path = ocfs2_new_path_from_et(&et); 1833 path = ocfs2_new_path_from_et(&et);
1846 if (!path) { 1834 if (!path) {
1847 ret = -ENOMEM; 1835 ret = -ENOMEM;
1848 mlog_errno(ret); 1836 mlog_errno(ret);
1849 goto out; 1837 goto out;
1850 } 1838 }
1851 1839
1852 while (trunc_end > trunc_start) { 1840 while (trunc_end > trunc_start) {
1853 1841
1854 ret = ocfs2_find_path(INODE_C 1842 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1855 cluster 1843 cluster_in_el);
1856 if (ret) { 1844 if (ret) {
1857 mlog_errno(ret); 1845 mlog_errno(ret);
1858 goto out; 1846 goto out;
1859 } 1847 }
1860 1848
1861 el = path_leaf_el(path); 1849 el = path_leaf_el(path);
1862 1850
1863 i = ocfs2_find_rec(el, trunc_ 1851 i = ocfs2_find_rec(el, trunc_end);
1864 /* 1852 /*
1865 * Need to go to previous ext 1853 * Need to go to previous extent block.
1866 */ 1854 */
1867 if (i < 0) { 1855 if (i < 0) {
1868 if (path->p_tree_dept 1856 if (path->p_tree_depth == 0)
1869 break; 1857 break;
1870 1858
1871 ret = ocfs2_find_cpos 1859 ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1872 1860 path,
1873 1861 &cluster_in_el);
1874 if (ret) { 1862 if (ret) {
1875 mlog_errno(re 1863 mlog_errno(ret);
1876 goto out; 1864 goto out;
1877 } 1865 }
1878 1866
1879 /* 1867 /*
1880 * We've reached the 1868 * We've reached the leftmost extent block,
1881 * it's safe to leave 1869 * it's safe to leave.
1882 */ 1870 */
1883 if (cluster_in_el == 1871 if (cluster_in_el == 0)
1884 break; 1872 break;
1885 1873
1886 /* 1874 /*
1887 * The 'pos' searched 1875 * The 'pos' searched for previous extent block is
1888 * always one cluster 1876 * always one cluster less than actual trunc_end.
1889 */ 1877 */
1890 trunc_end = cluster_i 1878 trunc_end = cluster_in_el + 1;
1891 1879
1892 ocfs2_reinit_path(pat 1880 ocfs2_reinit_path(path, 1);
1893 1881
1894 continue; 1882 continue;
1895 1883
1896 } else 1884 } else
1897 rec = &el->l_recs[i]; 1885 rec = &el->l_recs[i];
1898 1886
1899 ocfs2_calc_trunc_pos(inode, e 1887 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1900 &trunc_l 1888 &trunc_len, &trunc_end, &blkno, &done);
1901 if (done) 1889 if (done)
1902 break; 1890 break;
1903 1891
1904 flags = rec->e_flags; 1892 flags = rec->e_flags;
1905 phys_cpos = ocfs2_blocks_to_c 1893 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1906 1894
1907 ret = ocfs2_remove_btree_rang 1895 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1908 1896 phys_cpos, trunc_len, flags,
1909 1897 &dealloc, refcount_loc, false);
1910 if (ret < 0) { 1898 if (ret < 0) {
1911 mlog_errno(ret); 1899 mlog_errno(ret);
1912 goto out; 1900 goto out;
1913 } 1901 }
1914 1902
1915 cluster_in_el = trunc_end; 1903 cluster_in_el = trunc_end;
1916 1904
1917 ocfs2_reinit_path(path, 1); 1905 ocfs2_reinit_path(path, 1);
1918 } 1906 }
1919 1907
1920 ocfs2_truncate_cluster_pages(inode, b 1908 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1921 1909
1922 out: 1910 out:
1923 ocfs2_free_path(path); 1911 ocfs2_free_path(path);
1924 ocfs2_schedule_truncate_log_flush(osb 1912 ocfs2_schedule_truncate_log_flush(osb, 1);
1925 ocfs2_run_deallocs(osb, &dealloc); 1913 ocfs2_run_deallocs(osb, &dealloc);
1926 1914
1927 return ret; 1915 return ret;
1928 } 1916 }
1929 1917
1930 /* 1918 /*
1931 * Parts of this function taken from xfs_chan 1919 * Parts of this function taken from xfs_change_file_space()
1932 */ 1920 */
1933 static int __ocfs2_change_file_space(struct f 1921 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1934 loff_t f 1922 loff_t f_pos, unsigned int cmd,
1935 struct o 1923 struct ocfs2_space_resv *sr,
1936 int chan 1924 int change_size)
1937 { 1925 {
1938 int ret; 1926 int ret;
1939 s64 llen; 1927 s64 llen;
1940 loff_t size, orig_isize; 1928 loff_t size, orig_isize;
1941 struct ocfs2_super *osb = OCFS2_SB(in 1929 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1942 struct buffer_head *di_bh = NULL; 1930 struct buffer_head *di_bh = NULL;
1943 handle_t *handle; 1931 handle_t *handle;
1944 unsigned long long max_off = inode->i 1932 unsigned long long max_off = inode->i_sb->s_maxbytes;
1945 1933
1946 if (ocfs2_is_hard_readonly(osb) || oc 1934 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1947 return -EROFS; 1935 return -EROFS;
1948 1936
1949 inode_lock(inode); 1937 inode_lock(inode);
1950 1938
1951 /* Wait all existing dio workers, new 1939 /* Wait all existing dio workers, newcomers will block on i_rwsem */
1952 inode_dio_wait(inode); 1940 inode_dio_wait(inode);
1953 /* 1941 /*
1954 * This prevents concurrent writes on 1942 * This prevents concurrent writes on other nodes
1955 */ 1943 */
1956 ret = ocfs2_rw_lock(inode, 1); 1944 ret = ocfs2_rw_lock(inode, 1);
1957 if (ret) { 1945 if (ret) {
1958 mlog_errno(ret); 1946 mlog_errno(ret);
1959 goto out; 1947 goto out;
1960 } 1948 }
1961 1949
1962 ret = ocfs2_inode_lock(inode, &di_bh, 1950 ret = ocfs2_inode_lock(inode, &di_bh, 1);
1963 if (ret) { 1951 if (ret) {
1964 mlog_errno(ret); 1952 mlog_errno(ret);
1965 goto out_rw_unlock; 1953 goto out_rw_unlock;
1966 } 1954 }
1967 1955
1968 if (inode->i_flags & (S_IMMUTABLE|S_A 1956 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1969 ret = -EPERM; 1957 ret = -EPERM;
1970 goto out_inode_unlock; 1958 goto out_inode_unlock;
1971 } 1959 }
1972 1960
1973 switch (sr->l_whence) { 1961 switch (sr->l_whence) {
1974 case 0: /*SEEK_SET*/ 1962 case 0: /*SEEK_SET*/
1975 break; 1963 break;
1976 case 1: /*SEEK_CUR*/ 1964 case 1: /*SEEK_CUR*/
1977 sr->l_start += f_pos; 1965 sr->l_start += f_pos;
1978 break; 1966 break;
1979 case 2: /*SEEK_END*/ 1967 case 2: /*SEEK_END*/
1980 sr->l_start += i_size_read(in 1968 sr->l_start += i_size_read(inode);
1981 break; 1969 break;
1982 default: 1970 default:
1983 ret = -EINVAL; 1971 ret = -EINVAL;
1984 goto out_inode_unlock; 1972 goto out_inode_unlock;
1985 } 1973 }
1986 sr->l_whence = 0; 1974 sr->l_whence = 0;
1987 1975
1988 llen = sr->l_len > 0 ? sr->l_len - 1 1976 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1989 1977
1990 if (sr->l_start < 0 1978 if (sr->l_start < 0
1991 || sr->l_start > max_off 1979 || sr->l_start > max_off
1992 || (sr->l_start + llen) < 0 1980 || (sr->l_start + llen) < 0
1993 || (sr->l_start + llen) > max_off 1981 || (sr->l_start + llen) > max_off) {
1994 ret = -EINVAL; 1982 ret = -EINVAL;
1995 goto out_inode_unlock; 1983 goto out_inode_unlock;
1996 } 1984 }
1997 size = sr->l_start + sr->l_len; 1985 size = sr->l_start + sr->l_len;
1998 1986
1999 if (cmd == OCFS2_IOC_RESVSP || cmd == 1987 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
2000 cmd == OCFS2_IOC_UNRESVSP || cmd 1988 cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
2001 if (sr->l_len <= 0) { 1989 if (sr->l_len <= 0) {
2002 ret = -EINVAL; 1990 ret = -EINVAL;
2003 goto out_inode_unlock 1991 goto out_inode_unlock;
2004 } 1992 }
2005 } 1993 }
2006 1994
2007 if (file && setattr_should_drop_suidg 1995 if (file && setattr_should_drop_suidgid(&nop_mnt_idmap, file_inode(file))) {
2008 ret = __ocfs2_write_remove_su 1996 ret = __ocfs2_write_remove_suid(inode, di_bh);
2009 if (ret) { 1997 if (ret) {
2010 mlog_errno(ret); 1998 mlog_errno(ret);
2011 goto out_inode_unlock 1999 goto out_inode_unlock;
2012 } 2000 }
2013 } 2001 }
2014 2002
2015 down_write(&OCFS2_I(inode)->ip_alloc_ 2003 down_write(&OCFS2_I(inode)->ip_alloc_sem);
2016 switch (cmd) { 2004 switch (cmd) {
2017 case OCFS2_IOC_RESVSP: 2005 case OCFS2_IOC_RESVSP:
2018 case OCFS2_IOC_RESVSP64: 2006 case OCFS2_IOC_RESVSP64:
2019 /* 2007 /*
2020 * This takes unsigned offset 2008 * This takes unsigned offsets, but the signed ones we
2021 * pass have been checked aga 2009 * pass have been checked against overflow above.
2022 */ 2010 */
2023 ret = ocfs2_allocate_unwritte 2011 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
2024 2012 sr->l_len);
2025 break; 2013 break;
2026 case OCFS2_IOC_UNRESVSP: 2014 case OCFS2_IOC_UNRESVSP:
2027 case OCFS2_IOC_UNRESVSP64: 2015 case OCFS2_IOC_UNRESVSP64:
2028 ret = ocfs2_remove_inode_rang 2016 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
2029 2017 sr->l_len);
2030 break; 2018 break;
2031 default: 2019 default:
2032 ret = -EINVAL; 2020 ret = -EINVAL;
2033 } 2021 }
2034 2022
2035 orig_isize = i_size_read(inode); 2023 orig_isize = i_size_read(inode);
2036 /* zeroout eof blocks in the cluster. 2024 /* zeroout eof blocks in the cluster. */
2037 if (!ret && change_size && orig_isize 2025 if (!ret && change_size && orig_isize < size) {
2038 ret = ocfs2_zeroout_partial_c 2026 ret = ocfs2_zeroout_partial_cluster(inode, orig_isize,
2039 size 2027 size - orig_isize);
2040 if (!ret) 2028 if (!ret)
2041 i_size_write(inode, s 2029 i_size_write(inode, size);
2042 } 2030 }
2043 up_write(&OCFS2_I(inode)->ip_alloc_se 2031 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2044 if (ret) { 2032 if (ret) {
2045 mlog_errno(ret); 2033 mlog_errno(ret);
2046 goto out_inode_unlock; 2034 goto out_inode_unlock;
2047 } 2035 }
2048 2036
2049 /* 2037 /*
2050 * We update c/mtime for these change 2038 * We update c/mtime for these changes
2051 */ 2039 */
2052 handle = ocfs2_start_trans(osb, OCFS2 2040 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2053 if (IS_ERR(handle)) { 2041 if (IS_ERR(handle)) {
2054 ret = PTR_ERR(handle); 2042 ret = PTR_ERR(handle);
2055 mlog_errno(ret); 2043 mlog_errno(ret);
2056 goto out_inode_unlock; 2044 goto out_inode_unlock;
2057 } 2045 }
2058 2046
2059 inode_set_mtime_to_ts(inode, inode_se 2047 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
2060 ret = ocfs2_mark_inode_dirty(handle, 2048 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2061 if (ret < 0) 2049 if (ret < 0)
2062 mlog_errno(ret); 2050 mlog_errno(ret);
2063 2051
2064 if (file && (file->f_flags & O_SYNC)) 2052 if (file && (file->f_flags & O_SYNC))
2065 handle->h_sync = 1; 2053 handle->h_sync = 1;
2066 2054
2067 ocfs2_commit_trans(osb, handle); 2055 ocfs2_commit_trans(osb, handle);
2068 2056
2069 out_inode_unlock: 2057 out_inode_unlock:
2070 brelse(di_bh); 2058 brelse(di_bh);
2071 ocfs2_inode_unlock(inode, 1); 2059 ocfs2_inode_unlock(inode, 1);
2072 out_rw_unlock: 2060 out_rw_unlock:
2073 ocfs2_rw_unlock(inode, 1); 2061 ocfs2_rw_unlock(inode, 1);
2074 2062
2075 out: 2063 out:
2076 inode_unlock(inode); 2064 inode_unlock(inode);
2077 return ret; 2065 return ret;
2078 } 2066 }
2079 2067
2080 int ocfs2_change_file_space(struct file *file 2068 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2081 struct ocfs2_spac 2069 struct ocfs2_space_resv *sr)
2082 { 2070 {
2083 struct inode *inode = file_inode(file 2071 struct inode *inode = file_inode(file);
2084 struct ocfs2_super *osb = OCFS2_SB(in 2072 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2085 int ret; 2073 int ret;
2086 2074
2087 if ((cmd == OCFS2_IOC_RESVSP || cmd = 2075 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2088 !ocfs2_writes_unwritten_extents(o 2076 !ocfs2_writes_unwritten_extents(osb))
2089 return -ENOTTY; 2077 return -ENOTTY;
2090 else if ((cmd == OCFS2_IOC_UNRESVSP | 2078 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2091 !ocfs2_sparse_alloc(osb)) 2079 !ocfs2_sparse_alloc(osb))
2092 return -ENOTTY; 2080 return -ENOTTY;
2093 2081
2094 if (!S_ISREG(inode->i_mode)) 2082 if (!S_ISREG(inode->i_mode))
2095 return -EINVAL; 2083 return -EINVAL;
2096 2084
2097 if (!(file->f_mode & FMODE_WRITE)) 2085 if (!(file->f_mode & FMODE_WRITE))
2098 return -EBADF; 2086 return -EBADF;
2099 2087
2100 ret = mnt_want_write_file(file); 2088 ret = mnt_want_write_file(file);
2101 if (ret) 2089 if (ret)
2102 return ret; 2090 return ret;
2103 ret = __ocfs2_change_file_space(file, 2091 ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2104 mnt_drop_write_file(file); 2092 mnt_drop_write_file(file);
2105 return ret; 2093 return ret;
2106 } 2094 }
2107 2095
2108 static long ocfs2_fallocate(struct file *file 2096 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2109 loff_t len) 2097 loff_t len)
2110 { 2098 {
2111 struct inode *inode = file_inode(file 2099 struct inode *inode = file_inode(file);
2112 struct ocfs2_super *osb = OCFS2_SB(in 2100 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2113 struct ocfs2_space_resv sr; 2101 struct ocfs2_space_resv sr;
2114 int change_size = 1; 2102 int change_size = 1;
2115 int cmd = OCFS2_IOC_RESVSP64; 2103 int cmd = OCFS2_IOC_RESVSP64;
2116 int ret = 0; 2104 int ret = 0;
2117 2105
2118 if (mode & ~(FALLOC_FL_KEEP_SIZE | FA 2106 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2119 return -EOPNOTSUPP; 2107 return -EOPNOTSUPP;
2120 if (!ocfs2_writes_unwritten_extents(o 2108 if (!ocfs2_writes_unwritten_extents(osb))
2121 return -EOPNOTSUPP; 2109 return -EOPNOTSUPP;
2122 2110
2123 if (mode & FALLOC_FL_KEEP_SIZE) { 2111 if (mode & FALLOC_FL_KEEP_SIZE) {
2124 change_size = 0; 2112 change_size = 0;
2125 } else { 2113 } else {
2126 ret = inode_newsize_ok(inode, 2114 ret = inode_newsize_ok(inode, offset + len);
2127 if (ret) 2115 if (ret)
2128 return ret; 2116 return ret;
2129 } 2117 }
2130 2118
2131 if (mode & FALLOC_FL_PUNCH_HOLE) 2119 if (mode & FALLOC_FL_PUNCH_HOLE)
2132 cmd = OCFS2_IOC_UNRESVSP64; 2120 cmd = OCFS2_IOC_UNRESVSP64;
2133 2121
2134 sr.l_whence = 0; 2122 sr.l_whence = 0;
2135 sr.l_start = (s64)offset; 2123 sr.l_start = (s64)offset;
2136 sr.l_len = (s64)len; 2124 sr.l_len = (s64)len;
2137 2125
2138 return __ocfs2_change_file_space(NULL 2126 return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2139 chan 2127 change_size);
2140 } 2128 }
2141 2129
2142 int ocfs2_check_range_for_refcount(struct ino 2130 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2143 size_t cou 2131 size_t count)
2144 { 2132 {
2145 int ret = 0; 2133 int ret = 0;
2146 unsigned int extent_flags; 2134 unsigned int extent_flags;
2147 u32 cpos, clusters, extent_len, phys_ 2135 u32 cpos, clusters, extent_len, phys_cpos;
2148 struct super_block *sb = inode->i_sb; 2136 struct super_block *sb = inode->i_sb;
2149 2137
2150 if (!ocfs2_refcount_tree(OCFS2_SB(ino 2138 if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2151 !ocfs2_is_refcount_inode(inode) | 2139 !ocfs2_is_refcount_inode(inode) ||
2152 OCFS2_I(inode)->ip_dyn_features & 2140 OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2153 return 0; 2141 return 0;
2154 2142
2155 cpos = pos >> OCFS2_SB(sb)->s_cluster 2143 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2156 clusters = ocfs2_clusters_for_bytes(s 2144 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2157 2145
2158 while (clusters) { 2146 while (clusters) {
2159 ret = ocfs2_get_clusters(inod 2147 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2160 &ext 2148 &extent_flags);
2161 if (ret < 0) { 2149 if (ret < 0) {
2162 mlog_errno(ret); 2150 mlog_errno(ret);
2163 goto out; 2151 goto out;
2164 } 2152 }
2165 2153
2166 if (phys_cpos && (extent_flag 2154 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2167 ret = 1; 2155 ret = 1;
2168 break; 2156 break;
2169 } 2157 }
2170 2158
2171 if (extent_len > clusters) 2159 if (extent_len > clusters)
2172 extent_len = clusters 2160 extent_len = clusters;
2173 2161
2174 clusters -= extent_len; 2162 clusters -= extent_len;
2175 cpos += extent_len; 2163 cpos += extent_len;
2176 } 2164 }
2177 out: 2165 out:
2178 return ret; 2166 return ret;
2179 } 2167 }
2180 2168
2181 static int ocfs2_is_io_unaligned(struct inode 2169 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2182 { 2170 {
2183 int blockmask = inode->i_sb->s_blocks 2171 int blockmask = inode->i_sb->s_blocksize - 1;
2184 loff_t final_size = pos + count; 2172 loff_t final_size = pos + count;
2185 2173
2186 if ((pos & blockmask) || (final_size 2174 if ((pos & blockmask) || (final_size & blockmask))
2187 return 1; 2175 return 1;
2188 return 0; 2176 return 0;
2189 } 2177 }
2190 2178
2191 static int ocfs2_inode_lock_for_extent_tree(s 2179 static int ocfs2_inode_lock_for_extent_tree(struct inode *inode,
2192 s 2180 struct buffer_head **di_bh,
2193 i 2181 int meta_level,
2194 i 2182 int write_sem,
2195 i 2183 int wait)
2196 { 2184 {
2197 int ret = 0; 2185 int ret = 0;
2198 2186
2199 if (wait) 2187 if (wait)
2200 ret = ocfs2_inode_lock(inode, 2188 ret = ocfs2_inode_lock(inode, di_bh, meta_level);
2201 else 2189 else
2202 ret = ocfs2_try_inode_lock(in 2190 ret = ocfs2_try_inode_lock(inode, di_bh, meta_level);
2203 if (ret < 0) 2191 if (ret < 0)
2204 goto out; 2192 goto out;
2205 2193
2206 if (wait) { 2194 if (wait) {
2207 if (write_sem) 2195 if (write_sem)
2208 down_write(&OCFS2_I(i 2196 down_write(&OCFS2_I(inode)->ip_alloc_sem);
2209 else 2197 else
2210 down_read(&OCFS2_I(in 2198 down_read(&OCFS2_I(inode)->ip_alloc_sem);
2211 } else { 2199 } else {
2212 if (write_sem) 2200 if (write_sem)
2213 ret = down_write_tryl 2201 ret = down_write_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2214 else 2202 else
2215 ret = down_read_trylo 2203 ret = down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2216 2204
2217 if (!ret) { 2205 if (!ret) {
2218 ret = -EAGAIN; 2206 ret = -EAGAIN;
2219 goto out_unlock; 2207 goto out_unlock;
2220 } 2208 }
2221 } 2209 }
2222 2210
2223 return ret; 2211 return ret;
2224 2212
2225 out_unlock: 2213 out_unlock:
2226 brelse(*di_bh); 2214 brelse(*di_bh);
2227 *di_bh = NULL; 2215 *di_bh = NULL;
2228 ocfs2_inode_unlock(inode, meta_level) 2216 ocfs2_inode_unlock(inode, meta_level);
2229 out: 2217 out:
2230 return ret; 2218 return ret;
2231 } 2219 }
2232 2220
2233 static void ocfs2_inode_unlock_for_extent_tre 2221 static void ocfs2_inode_unlock_for_extent_tree(struct inode *inode,
2234 2222 struct buffer_head **di_bh,
2235 2223 int meta_level,
2236 2224 int write_sem)
2237 { 2225 {
2238 if (write_sem) 2226 if (write_sem)
2239 up_write(&OCFS2_I(inode)->ip_ 2227 up_write(&OCFS2_I(inode)->ip_alloc_sem);
2240 else 2228 else
2241 up_read(&OCFS2_I(inode)->ip_a 2229 up_read(&OCFS2_I(inode)->ip_alloc_sem);
2242 2230
2243 brelse(*di_bh); 2231 brelse(*di_bh);
2244 *di_bh = NULL; 2232 *di_bh = NULL;
2245 2233
2246 if (meta_level >= 0) 2234 if (meta_level >= 0)
2247 ocfs2_inode_unlock(inode, met 2235 ocfs2_inode_unlock(inode, meta_level);
2248 } 2236 }
2249 2237
2250 static int ocfs2_prepare_inode_for_write(stru 2238 static int ocfs2_prepare_inode_for_write(struct file *file,
2251 loff 2239 loff_t pos, size_t count, int wait)
2252 { 2240 {
2253 int ret = 0, meta_level = 0, overwrit 2241 int ret = 0, meta_level = 0, overwrite_io = 0;
2254 int write_sem = 0; 2242 int write_sem = 0;
2255 struct dentry *dentry = file->f_path. 2243 struct dentry *dentry = file->f_path.dentry;
2256 struct inode *inode = d_inode(dentry) 2244 struct inode *inode = d_inode(dentry);
2257 struct buffer_head *di_bh = NULL; 2245 struct buffer_head *di_bh = NULL;
2258 u32 cpos; 2246 u32 cpos;
2259 u32 clusters; 2247 u32 clusters;
2260 2248
2261 /* 2249 /*
2262 * We start with a read level meta lo 2250 * We start with a read level meta lock and only jump to an ex
2263 * if we need to make modifications h 2251 * if we need to make modifications here.
2264 */ 2252 */
2265 for(;;) { 2253 for(;;) {
2266 ret = ocfs2_inode_lock_for_ex 2254 ret = ocfs2_inode_lock_for_extent_tree(inode,
2267 2255 &di_bh,
2268 2256 meta_level,
2269 2257 write_sem,
2270 2258 wait);
2271 if (ret < 0) { 2259 if (ret < 0) {
2272 if (ret != -EAGAIN) 2260 if (ret != -EAGAIN)
2273 mlog_errno(re 2261 mlog_errno(ret);
2274 goto out; 2262 goto out;
2275 } 2263 }
2276 2264
2277 /* 2265 /*
2278 * Check if IO will overwrite 2266 * Check if IO will overwrite allocated blocks in case
2279 * IOCB_NOWAIT flag is set. 2267 * IOCB_NOWAIT flag is set.
2280 */ 2268 */
2281 if (!wait && !overwrite_io) { 2269 if (!wait && !overwrite_io) {
2282 overwrite_io = 1; 2270 overwrite_io = 1;
2283 2271
2284 ret = ocfs2_overwrite 2272 ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
2285 if (ret < 0) { 2273 if (ret < 0) {
2286 if (ret != -E 2274 if (ret != -EAGAIN)
2287 mlog_ 2275 mlog_errno(ret);
2288 goto out_unlo 2276 goto out_unlock;
2289 } 2277 }
2290 } 2278 }
2291 2279
2292 /* Clear suid / sgid if neces 2280 /* Clear suid / sgid if necessary. We do this here
2293 * instead of later in the wr 2281 * instead of later in the write path because
2294 * remove_suid() calls ->seta 2282 * remove_suid() calls ->setattr without any hint that
2295 * we may have already done o 2283 * we may have already done our cluster locking. Since
2296 * ocfs2_setattr() *must* tak 2284 * ocfs2_setattr() *must* take cluster locks to
2297 * proceed, this will lead us 2285 * proceed, this will lead us to recursively lock the
2298 * inode. There's also the di 2286 * inode. There's also the dinode i_size state which
2299 * can be lost via setattr du 2287 * can be lost via setattr during extending writes (we
2300 * set inode->i_size at the e 2288 * set inode->i_size at the end of a write. */
2301 if (setattr_should_drop_suidg 2289 if (setattr_should_drop_suidgid(&nop_mnt_idmap, inode)) {
2302 if (meta_level == 0) 2290 if (meta_level == 0) {
2303 ocfs2_inode_u 2291 ocfs2_inode_unlock_for_extent_tree(inode,
2304 2292 &di_bh,
2305 2293 meta_level,
2306 2294 write_sem);
2307 meta_level = 2295 meta_level = 1;
2308 continue; 2296 continue;
2309 } 2297 }
2310 2298
2311 ret = ocfs2_write_rem 2299 ret = ocfs2_write_remove_suid(inode);
2312 if (ret < 0) { 2300 if (ret < 0) {
2313 mlog_errno(re 2301 mlog_errno(ret);
2314 goto out_unlo 2302 goto out_unlock;
2315 } 2303 }
2316 } 2304 }
2317 2305
2318 ret = ocfs2_check_range_for_r 2306 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2319 if (ret == 1) { 2307 if (ret == 1) {
2320 ocfs2_inode_unlock_fo 2308 ocfs2_inode_unlock_for_extent_tree(inode,
2321 2309 &di_bh,
2322 2310 meta_level,
2323 2311 write_sem);
2324 meta_level = 1; 2312 meta_level = 1;
2325 write_sem = 1; 2313 write_sem = 1;
2326 ret = ocfs2_inode_loc 2314 ret = ocfs2_inode_lock_for_extent_tree(inode,
2327 2315 &di_bh,
2328 2316 meta_level,
2329 2317 write_sem,
2330 2318 wait);
2331 if (ret < 0) { 2319 if (ret < 0) {
2332 if (ret != -E 2320 if (ret != -EAGAIN)
2333 mlog_ 2321 mlog_errno(ret);
2334 goto out; 2322 goto out;
2335 } 2323 }
2336 2324
2337 cpos = pos >> OCFS2_S 2325 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2338 clusters = 2326 clusters =
2339 ocfs2_cluster 2327 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2340 ret = ocfs2_refcount_ 2328 ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2341 } 2329 }
2342 2330
2343 if (ret < 0) { 2331 if (ret < 0) {
2344 if (ret != -EAGAIN) 2332 if (ret != -EAGAIN)
2345 mlog_errno(re 2333 mlog_errno(ret);
2346 goto out_unlock; 2334 goto out_unlock;
2347 } 2335 }
2348 2336
2349 break; 2337 break;
2350 } 2338 }
2351 2339
2352 out_unlock: 2340 out_unlock:
2353 trace_ocfs2_prepare_inode_for_write(O 2341 trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2354 p 2342 pos, count, wait);
2355 2343
2356 ocfs2_inode_unlock_for_extent_tree(in 2344 ocfs2_inode_unlock_for_extent_tree(inode,
2357 &d 2345 &di_bh,
2358 me 2346 meta_level,
2359 wr 2347 write_sem);
2360 2348
2361 out: 2349 out:
2362 return ret; 2350 return ret;
2363 } 2351 }
2364 2352
2365 static ssize_t ocfs2_file_write_iter(struct k 2353 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2366 struct io 2354 struct iov_iter *from)
2367 { 2355 {
2368 int rw_level; 2356 int rw_level;
2369 ssize_t written = 0; 2357 ssize_t written = 0;
2370 ssize_t ret; 2358 ssize_t ret;
2371 size_t count = iov_iter_count(from); 2359 size_t count = iov_iter_count(from);
2372 struct file *file = iocb->ki_filp; 2360 struct file *file = iocb->ki_filp;
2373 struct inode *inode = file_inode(file 2361 struct inode *inode = file_inode(file);
2374 struct ocfs2_super *osb = OCFS2_SB(in 2362 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2375 int full_coherency = !(osb->s_mount_o 2363 int full_coherency = !(osb->s_mount_opt &
2376 OCFS2_MOUNT_CO 2364 OCFS2_MOUNT_COHERENCY_BUFFERED);
2377 void *saved_ki_complete = NULL; 2365 void *saved_ki_complete = NULL;
2378 int append_write = ((iocb->ki_pos + c 2366 int append_write = ((iocb->ki_pos + count) >=
2379 i_size_read(inode) ? 2367 i_size_read(inode) ? 1 : 0);
2380 int direct_io = iocb->ki_flags & IOCB 2368 int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2381 int nowait = iocb->ki_flags & IOCB_NO 2369 int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2382 2370
2383 trace_ocfs2_file_write_iter(inode, fi 2371 trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2384 (unsigned long long)OCFS2_I(i 2372 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2385 file->f_path.dentry->d_name.l 2373 file->f_path.dentry->d_name.len,
2386 file->f_path.dentry->d_name.n 2374 file->f_path.dentry->d_name.name,
2387 (unsigned int)from->nr_segs); 2375 (unsigned int)from->nr_segs); /* GRRRRR */
2388 2376
2389 if (!direct_io && nowait) 2377 if (!direct_io && nowait)
2390 return -EOPNOTSUPP; 2378 return -EOPNOTSUPP;
2391 2379
2392 if (count == 0) 2380 if (count == 0)
2393 return 0; 2381 return 0;
2394 2382
2395 if (nowait) { 2383 if (nowait) {
2396 if (!inode_trylock(inode)) 2384 if (!inode_trylock(inode))
2397 return -EAGAIN; 2385 return -EAGAIN;
2398 } else 2386 } else
2399 inode_lock(inode); 2387 inode_lock(inode);
2400 2388
2401 /* 2389 /*
2402 * Concurrent O_DIRECT writes are all 2390 * Concurrent O_DIRECT writes are allowed with
2403 * mount_option "coherency=buffered". 2391 * mount_option "coherency=buffered".
2404 * For append write, we must take rw 2392 * For append write, we must take rw EX.
2405 */ 2393 */
2406 rw_level = (!direct_io || full_cohere 2394 rw_level = (!direct_io || full_coherency || append_write);
2407 2395
2408 if (nowait) 2396 if (nowait)
2409 ret = ocfs2_try_rw_lock(inode 2397 ret = ocfs2_try_rw_lock(inode, rw_level);
2410 else 2398 else
2411 ret = ocfs2_rw_lock(inode, rw 2399 ret = ocfs2_rw_lock(inode, rw_level);
2412 if (ret < 0) { 2400 if (ret < 0) {
2413 if (ret != -EAGAIN) 2401 if (ret != -EAGAIN)
2414 mlog_errno(ret); 2402 mlog_errno(ret);
2415 goto out_mutex; 2403 goto out_mutex;
2416 } 2404 }
2417 2405
2418 /* 2406 /*
2419 * O_DIRECT writes with "coherency=fu 2407 * O_DIRECT writes with "coherency=full" need to take EX cluster
2420 * inode_lock to guarantee coherency. 2408 * inode_lock to guarantee coherency.
2421 */ 2409 */
2422 if (direct_io && full_coherency) { 2410 if (direct_io && full_coherency) {
2423 /* 2411 /*
2424 * We need to take and drop t 2412 * We need to take and drop the inode lock to force
2425 * other nodes to drop their 2413 * other nodes to drop their caches. Buffered I/O
2426 * already does this in write 2414 * already does this in write_begin().
2427 */ 2415 */
2428 if (nowait) 2416 if (nowait)
2429 ret = ocfs2_try_inode 2417 ret = ocfs2_try_inode_lock(inode, NULL, 1);
2430 else 2418 else
2431 ret = ocfs2_inode_loc 2419 ret = ocfs2_inode_lock(inode, NULL, 1);
2432 if (ret < 0) { 2420 if (ret < 0) {
2433 if (ret != -EAGAIN) 2421 if (ret != -EAGAIN)
2434 mlog_errno(re 2422 mlog_errno(ret);
2435 goto out; 2423 goto out;
2436 } 2424 }
2437 2425
2438 ocfs2_inode_unlock(inode, 1); 2426 ocfs2_inode_unlock(inode, 1);
2439 } 2427 }
2440 2428
2441 ret = generic_write_checks(iocb, from 2429 ret = generic_write_checks(iocb, from);
2442 if (ret <= 0) { 2430 if (ret <= 0) {
2443 if (ret) 2431 if (ret)
2444 mlog_errno(ret); 2432 mlog_errno(ret);
2445 goto out; 2433 goto out;
2446 } 2434 }
2447 count = ret; 2435 count = ret;
2448 2436
2449 ret = ocfs2_prepare_inode_for_write(f 2437 ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2450 if (ret < 0) { 2438 if (ret < 0) {
2451 if (ret != -EAGAIN) 2439 if (ret != -EAGAIN)
2452 mlog_errno(ret); 2440 mlog_errno(ret);
2453 goto out; 2441 goto out;
2454 } 2442 }
2455 2443
2456 if (direct_io && !is_sync_kiocb(iocb) 2444 if (direct_io && !is_sync_kiocb(iocb) &&
2457 ocfs2_is_io_unaligned(inode, coun 2445 ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2458 /* 2446 /*
2459 * Make it a sync io if it's 2447 * Make it a sync io if it's an unaligned aio.
2460 */ 2448 */
2461 saved_ki_complete = xchg(&ioc 2449 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2462 } 2450 }
2463 2451
2464 /* communicate with ocfs2_dio_end_io 2452 /* communicate with ocfs2_dio_end_io */
2465 ocfs2_iocb_set_rw_locked(iocb, rw_lev 2453 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2466 2454
2467 written = __generic_file_write_iter(i 2455 written = __generic_file_write_iter(iocb, from);
2468 /* buffered aio wouldn't have proper 2456 /* buffered aio wouldn't have proper lock coverage today */
2469 BUG_ON(written == -EIOCBQUEUED && !di 2457 BUG_ON(written == -EIOCBQUEUED && !direct_io);
2470 2458
2471 /* 2459 /*
2472 * deep in g_f_a_w_n()->ocfs2_direct_ 2460 * 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 2461 * function pointer which is called when o_direct io completes so that
2474 * it can unlock our rw lock. 2462 * it can unlock our rw lock.
2475 * Unfortunately there are error case 2463 * Unfortunately there are error cases which call end_io and others
2476 * that don't. so we don't have to u 2464 * 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 2465 * async dio is going to do it in the future or an end_io after an
2478 * error has already done it. 2466 * error has already done it.
2479 */ 2467 */
2480 if ((written == -EIOCBQUEUED) || (!oc 2468 if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2481 rw_level = -1; 2469 rw_level = -1;
2482 } 2470 }
2483 2471
2484 if (unlikely(written <= 0)) 2472 if (unlikely(written <= 0))
2485 goto out; 2473 goto out;
2486 2474
2487 if (((file->f_flags & O_DSYNC) && !di 2475 if (((file->f_flags & O_DSYNC) && !direct_io) ||
2488 IS_SYNC(inode)) { 2476 IS_SYNC(inode)) {
2489 ret = filemap_fdatawrite_rang 2477 ret = filemap_fdatawrite_range(file->f_mapping,
2490 2478 iocb->ki_pos - written,
2491 2479 iocb->ki_pos - 1);
2492 if (ret < 0) 2480 if (ret < 0)
2493 written = ret; 2481 written = ret;
2494 2482
2495 if (!ret) { 2483 if (!ret) {
2496 ret = jbd2_journal_fo 2484 ret = jbd2_journal_force_commit(osb->journal->j_journal);
2497 if (ret < 0) 2485 if (ret < 0)
2498 written = ret 2486 written = ret;
2499 } 2487 }
2500 2488
2501 if (!ret) 2489 if (!ret)
2502 ret = filemap_fdatawa 2490 ret = filemap_fdatawait_range(file->f_mapping,
2503 2491 iocb->ki_pos - written,
2504 2492 iocb->ki_pos - 1);
2505 } 2493 }
2506 2494
2507 out: 2495 out:
2508 if (saved_ki_complete) 2496 if (saved_ki_complete)
2509 xchg(&iocb->ki_complete, save 2497 xchg(&iocb->ki_complete, saved_ki_complete);
2510 2498
2511 if (rw_level != -1) 2499 if (rw_level != -1)
2512 ocfs2_rw_unlock(inode, rw_lev 2500 ocfs2_rw_unlock(inode, rw_level);
2513 2501
2514 out_mutex: 2502 out_mutex:
2515 inode_unlock(inode); 2503 inode_unlock(inode);
2516 2504
2517 if (written) 2505 if (written)
2518 ret = written; 2506 ret = written;
2519 return ret; 2507 return ret;
2520 } 2508 }
2521 2509
2522 static ssize_t ocfs2_file_read_iter(struct ki 2510 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2523 struct iov 2511 struct iov_iter *to)
2524 { 2512 {
2525 int ret = 0, rw_level = -1, lock_leve 2513 int ret = 0, rw_level = -1, lock_level = 0;
2526 struct file *filp = iocb->ki_filp; 2514 struct file *filp = iocb->ki_filp;
2527 struct inode *inode = file_inode(filp 2515 struct inode *inode = file_inode(filp);
2528 int direct_io = iocb->ki_flags & IOCB 2516 int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2529 int nowait = iocb->ki_flags & IOCB_NO 2517 int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2530 2518
2531 trace_ocfs2_file_read_iter(inode, fil 2519 trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2532 (unsigned long long)O 2520 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2533 filp->f_path.dentry-> 2521 filp->f_path.dentry->d_name.len,
2534 filp->f_path.dentry-> 2522 filp->f_path.dentry->d_name.name,
2535 to->nr_segs); /* GR 2523 to->nr_segs); /* GRRRRR */
2536 2524
2537 2525
2538 if (!inode) { 2526 if (!inode) {
2539 ret = -EINVAL; 2527 ret = -EINVAL;
2540 mlog_errno(ret); 2528 mlog_errno(ret);
2541 goto bail; 2529 goto bail;
2542 } 2530 }
2543 2531
2544 if (!direct_io && nowait) 2532 if (!direct_io && nowait)
2545 return -EOPNOTSUPP; 2533 return -EOPNOTSUPP;
2546 2534
2547 /* 2535 /*
2548 * buffered reads protect themselves 2536 * buffered reads protect themselves in ->read_folio(). O_DIRECT reads
2549 * need locks to protect pending read 2537 * need locks to protect pending reads from racing with truncate.
2550 */ 2538 */
2551 if (direct_io) { 2539 if (direct_io) {
2552 if (nowait) 2540 if (nowait)
2553 ret = ocfs2_try_rw_lo 2541 ret = ocfs2_try_rw_lock(inode, 0);
2554 else 2542 else
2555 ret = ocfs2_rw_lock(i 2543 ret = ocfs2_rw_lock(inode, 0);
2556 2544
2557 if (ret < 0) { 2545 if (ret < 0) {
2558 if (ret != -EAGAIN) 2546 if (ret != -EAGAIN)
2559 mlog_errno(re 2547 mlog_errno(ret);
2560 goto bail; 2548 goto bail;
2561 } 2549 }
2562 rw_level = 0; 2550 rw_level = 0;
2563 /* communicate with ocfs2_dio 2551 /* communicate with ocfs2_dio_end_io */
2564 ocfs2_iocb_set_rw_locked(iocb 2552 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2565 } 2553 }
2566 2554
2567 /* 2555 /*
2568 * We're fine letting folks race trun 2556 * We're fine letting folks race truncates and extending
2569 * writes with read across the cluste 2557 * writes with read across the cluster, just like they can
2570 * locally. Hence no rw_lock during r 2558 * locally. Hence no rw_lock during read.
2571 * 2559 *
2572 * Take and drop the meta data lock t 2560 * Take and drop the meta data lock to update inode fields
2573 * like i_size. This allows the check 2561 * like i_size. This allows the checks down below
2574 * copy_splice_read() a chance of act 2562 * copy_splice_read() a chance of actually working.
2575 */ 2563 */
2576 ret = ocfs2_inode_lock_atime(inode, f 2564 ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2577 !nowait) 2565 !nowait);
2578 if (ret < 0) { 2566 if (ret < 0) {
2579 if (ret != -EAGAIN) 2567 if (ret != -EAGAIN)
2580 mlog_errno(ret); 2568 mlog_errno(ret);
2581 goto bail; 2569 goto bail;
2582 } 2570 }
2583 ocfs2_inode_unlock(inode, lock_level) 2571 ocfs2_inode_unlock(inode, lock_level);
2584 2572
2585 ret = generic_file_read_iter(iocb, to 2573 ret = generic_file_read_iter(iocb, to);
2586 trace_generic_file_read_iter_ret(ret) 2574 trace_generic_file_read_iter_ret(ret);
2587 2575
2588 /* buffered aio wouldn't have proper 2576 /* buffered aio wouldn't have proper lock coverage today */
2589 BUG_ON(ret == -EIOCBQUEUED && !direct 2577 BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2590 2578
2591 /* see ocfs2_file_write_iter */ 2579 /* see ocfs2_file_write_iter */
2592 if (ret == -EIOCBQUEUED || !ocfs2_ioc 2580 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2593 rw_level = -1; 2581 rw_level = -1;
2594 } 2582 }
2595 2583
2596 bail: 2584 bail:
2597 if (rw_level != -1) 2585 if (rw_level != -1)
2598 ocfs2_rw_unlock(inode, rw_lev 2586 ocfs2_rw_unlock(inode, rw_level);
2599 2587
2600 return ret; 2588 return ret;
2601 } 2589 }
2602 2590
2603 static ssize_t ocfs2_file_splice_read(struct 2591 static ssize_t ocfs2_file_splice_read(struct file *in, loff_t *ppos,
2604 struct 2592 struct pipe_inode_info *pipe,
2605 size_t 2593 size_t len, unsigned int flags)
2606 { 2594 {
2607 struct inode *inode = file_inode(in); 2595 struct inode *inode = file_inode(in);
2608 ssize_t ret = 0; 2596 ssize_t ret = 0;
2609 int lock_level = 0; 2597 int lock_level = 0;
2610 2598
2611 trace_ocfs2_file_splice_read(inode, i 2599 trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2612 (unsigne 2600 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2613 in->f_pa 2601 in->f_path.dentry->d_name.len,
2614 in->f_pa 2602 in->f_path.dentry->d_name.name,
2615 flags); 2603 flags);
2616 2604
2617 /* 2605 /*
2618 * We're fine letting folks race trun 2606 * We're fine letting folks race truncates and extending writes with
2619 * read across the cluster, just like 2607 * read across the cluster, just like they can locally. Hence no
2620 * rw_lock during read. 2608 * rw_lock during read.
2621 * 2609 *
2622 * Take and drop the meta data lock t 2610 * Take and drop the meta data lock to update inode fields like i_size.
2623 * This allows the checks down below 2611 * This allows the checks down below filemap_splice_read() a chance of
2624 * actually working. 2612 * actually working.
2625 */ 2613 */
2626 ret = ocfs2_inode_lock_atime(inode, i 2614 ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level, 1);
2627 if (ret < 0) { 2615 if (ret < 0) {
2628 if (ret != -EAGAIN) 2616 if (ret != -EAGAIN)
2629 mlog_errno(ret); 2617 mlog_errno(ret);
2630 goto bail; 2618 goto bail;
2631 } 2619 }
2632 ocfs2_inode_unlock(inode, lock_level) 2620 ocfs2_inode_unlock(inode, lock_level);
2633 2621
2634 ret = filemap_splice_read(in, ppos, p 2622 ret = filemap_splice_read(in, ppos, pipe, len, flags);
2635 trace_filemap_splice_read_ret(ret); 2623 trace_filemap_splice_read_ret(ret);
2636 bail: 2624 bail:
2637 return ret; 2625 return ret;
2638 } 2626 }
2639 2627
2640 /* Refer generic_file_llseek_unlocked() */ 2628 /* Refer generic_file_llseek_unlocked() */
2641 static loff_t ocfs2_file_llseek(struct file * 2629 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2642 { 2630 {
2643 struct inode *inode = file->f_mapping 2631 struct inode *inode = file->f_mapping->host;
2644 int ret = 0; 2632 int ret = 0;
2645 2633
2646 inode_lock(inode); 2634 inode_lock(inode);
2647 2635
2648 switch (whence) { 2636 switch (whence) {
2649 case SEEK_SET: 2637 case SEEK_SET:
2650 break; 2638 break;
2651 case SEEK_END: 2639 case SEEK_END:
2652 /* SEEK_END requires the OCFS 2640 /* SEEK_END requires the OCFS2 inode lock for the file
2653 * because it references the 2641 * because it references the file's size.
2654 */ 2642 */
2655 ret = ocfs2_inode_lock(inode, 2643 ret = ocfs2_inode_lock(inode, NULL, 0);
2656 if (ret < 0) { 2644 if (ret < 0) {
2657 mlog_errno(ret); 2645 mlog_errno(ret);
2658 goto out; 2646 goto out;
2659 } 2647 }
2660 offset += i_size_read(inode); 2648 offset += i_size_read(inode);
2661 ocfs2_inode_unlock(inode, 0); 2649 ocfs2_inode_unlock(inode, 0);
2662 break; 2650 break;
2663 case SEEK_CUR: 2651 case SEEK_CUR:
2664 if (offset == 0) { 2652 if (offset == 0) {
2665 offset = file->f_pos; 2653 offset = file->f_pos;
2666 goto out; 2654 goto out;
2667 } 2655 }
2668 offset += file->f_pos; 2656 offset += file->f_pos;
2669 break; 2657 break;
2670 case SEEK_DATA: 2658 case SEEK_DATA:
2671 case SEEK_HOLE: 2659 case SEEK_HOLE:
2672 ret = ocfs2_seek_data_hole_of 2660 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2673 if (ret) 2661 if (ret)
2674 goto out; 2662 goto out;
2675 break; 2663 break;
2676 default: 2664 default:
2677 ret = -EINVAL; 2665 ret = -EINVAL;
2678 goto out; 2666 goto out;
2679 } 2667 }
2680 2668
2681 offset = vfs_setpos(file, offset, ino 2669 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2682 2670
2683 out: 2671 out:
2684 inode_unlock(inode); 2672 inode_unlock(inode);
2685 if (ret) 2673 if (ret)
2686 return ret; 2674 return ret;
2687 return offset; 2675 return offset;
2688 } 2676 }
2689 2677
2690 static loff_t ocfs2_remap_file_range(struct f 2678 static loff_t ocfs2_remap_file_range(struct file *file_in, loff_t pos_in,
2691 struct f 2679 struct file *file_out, loff_t pos_out,
2692 loff_t l 2680 loff_t len, unsigned int remap_flags)
2693 { 2681 {
2694 struct inode *inode_in = file_inode(f 2682 struct inode *inode_in = file_inode(file_in);
2695 struct inode *inode_out = file_inode( 2683 struct inode *inode_out = file_inode(file_out);
2696 struct ocfs2_super *osb = OCFS2_SB(in 2684 struct ocfs2_super *osb = OCFS2_SB(inode_in->i_sb);
2697 struct buffer_head *in_bh = NULL, *ou 2685 struct buffer_head *in_bh = NULL, *out_bh = NULL;
2698 bool same_inode = (inode_in == inode_ 2686 bool same_inode = (inode_in == inode_out);
2699 loff_t remapped = 0; 2687 loff_t remapped = 0;
2700 ssize_t ret; 2688 ssize_t ret;
2701 2689
2702 if (remap_flags & ~(REMAP_FILE_DEDUP 2690 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
2703 return -EINVAL; 2691 return -EINVAL;
2704 if (!ocfs2_refcount_tree(osb)) 2692 if (!ocfs2_refcount_tree(osb))
2705 return -EOPNOTSUPP; 2693 return -EOPNOTSUPP;
2706 if (ocfs2_is_hard_readonly(osb) || oc 2694 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
2707 return -EROFS; 2695 return -EROFS;
2708 2696
2709 /* Lock both files against IO */ 2697 /* Lock both files against IO */
2710 ret = ocfs2_reflink_inodes_lock(inode 2698 ret = ocfs2_reflink_inodes_lock(inode_in, &in_bh, inode_out, &out_bh);
2711 if (ret) 2699 if (ret)
2712 return ret; 2700 return ret;
2713 2701
2714 /* Check file eligibility and prepare 2702 /* Check file eligibility and prepare for block sharing. */
2715 ret = -EINVAL; 2703 ret = -EINVAL;
2716 if ((OCFS2_I(inode_in)->ip_flags & OC 2704 if ((OCFS2_I(inode_in)->ip_flags & OCFS2_INODE_SYSTEM_FILE) ||
2717 (OCFS2_I(inode_out)->ip_flags & O 2705 (OCFS2_I(inode_out)->ip_flags & OCFS2_INODE_SYSTEM_FILE))
2718 goto out_unlock; 2706 goto out_unlock;
2719 2707
2720 ret = generic_remap_file_range_prep(f 2708 ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
2721 &len, remap_flags); 2709 &len, remap_flags);
2722 if (ret < 0 || len == 0) 2710 if (ret < 0 || len == 0)
2723 goto out_unlock; 2711 goto out_unlock;
2724 2712
2725 /* Lock out changes to the allocation 2713 /* Lock out changes to the allocation maps and remap. */
2726 down_write(&OCFS2_I(inode_in)->ip_all 2714 down_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2727 if (!same_inode) 2715 if (!same_inode)
2728 down_write_nested(&OCFS2_I(in 2716 down_write_nested(&OCFS2_I(inode_out)->ip_alloc_sem,
2729 SINGLE_DEPT 2717 SINGLE_DEPTH_NESTING);
2730 2718
2731 /* Zap any page cache for the destina 2719 /* Zap any page cache for the destination file's range. */
2732 truncate_inode_pages_range(&inode_out 2720 truncate_inode_pages_range(&inode_out->i_data,
2733 round_down 2721 round_down(pos_out, PAGE_SIZE),
2734 round_up(p 2722 round_up(pos_out + len, PAGE_SIZE) - 1);
2735 2723
2736 remapped = ocfs2_reflink_remap_blocks 2724 remapped = ocfs2_reflink_remap_blocks(inode_in, in_bh, pos_in,
2737 inode_out, out_bh, po 2725 inode_out, out_bh, pos_out, len);
2738 up_write(&OCFS2_I(inode_in)->ip_alloc 2726 up_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2739 if (!same_inode) 2727 if (!same_inode)
2740 up_write(&OCFS2_I(inode_out)- 2728 up_write(&OCFS2_I(inode_out)->ip_alloc_sem);
2741 if (remapped < 0) { 2729 if (remapped < 0) {
2742 ret = remapped; 2730 ret = remapped;
2743 mlog_errno(ret); 2731 mlog_errno(ret);
2744 goto out_unlock; 2732 goto out_unlock;
2745 } 2733 }
2746 2734
2747 /* 2735 /*
2748 * Empty the extent map so that we ma 2736 * Empty the extent map so that we may get the right extent
2749 * record from the disk. 2737 * record from the disk.
2750 */ 2738 */
2751 ocfs2_extent_map_trunc(inode_in, 0); 2739 ocfs2_extent_map_trunc(inode_in, 0);
2752 ocfs2_extent_map_trunc(inode_out, 0); 2740 ocfs2_extent_map_trunc(inode_out, 0);
2753 2741
2754 ret = ocfs2_reflink_update_dest(inode 2742 ret = ocfs2_reflink_update_dest(inode_out, out_bh, pos_out + len);
2755 if (ret) { 2743 if (ret) {
2756 mlog_errno(ret); 2744 mlog_errno(ret);
2757 goto out_unlock; 2745 goto out_unlock;
2758 } 2746 }
2759 2747
2760 out_unlock: 2748 out_unlock:
2761 ocfs2_reflink_inodes_unlock(inode_in, 2749 ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh);
2762 return remapped > 0 ? remapped : ret; 2750 return remapped > 0 ? remapped : ret;
2763 } 2751 }
2764 2752
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 2753 const struct inode_operations ocfs2_file_iops = {
2773 .setattr = ocfs2_setattr, 2754 .setattr = ocfs2_setattr,
2774 .getattr = ocfs2_getattr, 2755 .getattr = ocfs2_getattr,
2775 .permission = ocfs2_permission, 2756 .permission = ocfs2_permission,
2776 .listxattr = ocfs2_listxattr, 2757 .listxattr = ocfs2_listxattr,
2777 .fiemap = ocfs2_fiemap, 2758 .fiemap = ocfs2_fiemap,
2778 .get_inode_acl = ocfs2_iop_get_acl, 2759 .get_inode_acl = ocfs2_iop_get_acl,
2779 .set_acl = ocfs2_iop_set_acl, 2760 .set_acl = ocfs2_iop_set_acl,
2780 .fileattr_get = ocfs2_fileattr_get, 2761 .fileattr_get = ocfs2_fileattr_get,
2781 .fileattr_set = ocfs2_fileattr_set, 2762 .fileattr_set = ocfs2_fileattr_set,
2782 }; 2763 };
2783 2764
2784 const struct inode_operations ocfs2_special_f 2765 const struct inode_operations ocfs2_special_file_iops = {
2785 .setattr = ocfs2_setattr, 2766 .setattr = ocfs2_setattr,
2786 .getattr = ocfs2_getattr, 2767 .getattr = ocfs2_getattr,
2787 .listxattr = ocfs2_listxattr, 2768 .listxattr = ocfs2_listxattr,
2788 .permission = ocfs2_permission, 2769 .permission = ocfs2_permission,
2789 .get_inode_acl = ocfs2_iop_get_acl, 2770 .get_inode_acl = ocfs2_iop_get_acl,
2790 .set_acl = ocfs2_iop_set_acl, 2771 .set_acl = ocfs2_iop_set_acl,
2791 }; 2772 };
2792 2773
2793 /* 2774 /*
2794 * Other than ->lock, keep ocfs2_fops and ocf 2775 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2795 * ocfs2_fops_no_plocks and ocfs2_dops_no_plo 2776 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2796 */ 2777 */
2797 const struct file_operations ocfs2_fops = { 2778 const struct file_operations ocfs2_fops = {
2798 .llseek = ocfs2_file_llseek, 2779 .llseek = ocfs2_file_llseek,
2799 .mmap = ocfs2_mmap, 2780 .mmap = ocfs2_mmap,
2800 .fsync = ocfs2_sync_file, 2781 .fsync = ocfs2_sync_file,
2801 .release = ocfs2_file_release, 2782 .release = ocfs2_file_release,
2802 .open = ocfs2_file_open, 2783 .open = ocfs2_file_open,
2803 .read_iter = ocfs2_file_read_ite 2784 .read_iter = ocfs2_file_read_iter,
2804 .write_iter = ocfs2_file_write_it 2785 .write_iter = ocfs2_file_write_iter,
2805 .unlocked_ioctl = ocfs2_ioctl, 2786 .unlocked_ioctl = ocfs2_ioctl,
2806 #ifdef CONFIG_COMPAT 2787 #ifdef CONFIG_COMPAT
2807 .compat_ioctl = ocfs2_compat_ioctl, 2788 .compat_ioctl = ocfs2_compat_ioctl,
2808 #endif 2789 #endif
2809 .lock = ocfs2_lock, 2790 .lock = ocfs2_lock,
2810 .flock = ocfs2_flock, 2791 .flock = ocfs2_flock,
2811 .splice_read = ocfs2_file_splice_r 2792 .splice_read = ocfs2_file_splice_read,
2812 .splice_write = iter_file_splice_wr 2793 .splice_write = iter_file_splice_write,
2813 .fallocate = ocfs2_fallocate, 2794 .fallocate = ocfs2_fallocate,
2814 .remap_file_range = ocfs2_remap_file_ 2795 .remap_file_range = ocfs2_remap_file_range,
2815 }; 2796 };
2816 2797
2817 WRAP_DIR_ITER(ocfs2_readdir) // FIXME! 2798 WRAP_DIR_ITER(ocfs2_readdir) // FIXME!
2818 const struct file_operations ocfs2_dops = { 2799 const struct file_operations ocfs2_dops = {
2819 .llseek = ocfs2_dir_llseek, !! 2800 .llseek = generic_file_llseek,
2820 .read = generic_read_dir, 2801 .read = generic_read_dir,
2821 .iterate_shared = shared_ocfs2_readdi 2802 .iterate_shared = shared_ocfs2_readdir,
2822 .fsync = ocfs2_sync_file, 2803 .fsync = ocfs2_sync_file,
2823 .release = ocfs2_dir_release, 2804 .release = ocfs2_dir_release,
2824 .open = ocfs2_dir_open, 2805 .open = ocfs2_dir_open,
2825 .unlocked_ioctl = ocfs2_ioctl, 2806 .unlocked_ioctl = ocfs2_ioctl,
2826 #ifdef CONFIG_COMPAT 2807 #ifdef CONFIG_COMPAT
2827 .compat_ioctl = ocfs2_compat_ioctl, 2808 .compat_ioctl = ocfs2_compat_ioctl,
2828 #endif 2809 #endif
2829 .lock = ocfs2_lock, 2810 .lock = ocfs2_lock,
2830 .flock = ocfs2_flock, 2811 .flock = ocfs2_flock,
2831 }; 2812 };
2832 2813
2833 /* 2814 /*
2834 * POSIX-lockless variants of our file_operat 2815 * POSIX-lockless variants of our file_operations.
2835 * 2816 *
2836 * These will be used if the underlying clust 2817 * These will be used if the underlying cluster stack does not support
2837 * posix file locking, if the user passes the 2818 * posix file locking, if the user passes the "localflocks" mount
2838 * option, or if we have a local-only fs. 2819 * option, or if we have a local-only fs.
2839 * 2820 *
2840 * ocfs2_flock is in here because all stacks 2821 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2841 * so we still want it in the case of no stac 2822 * so we still want it in the case of no stack support for
2842 * plocks. Internally, it will do the right t 2823 * plocks. Internally, it will do the right thing when asked to ignore
2843 * the cluster. 2824 * the cluster.
2844 */ 2825 */
2845 const struct file_operations ocfs2_fops_no_pl 2826 const struct file_operations ocfs2_fops_no_plocks = {
2846 .llseek = ocfs2_file_llseek, 2827 .llseek = ocfs2_file_llseek,
2847 .mmap = ocfs2_mmap, 2828 .mmap = ocfs2_mmap,
2848 .fsync = ocfs2_sync_file, 2829 .fsync = ocfs2_sync_file,
2849 .release = ocfs2_file_release, 2830 .release = ocfs2_file_release,
2850 .open = ocfs2_file_open, 2831 .open = ocfs2_file_open,
2851 .read_iter = ocfs2_file_read_ite 2832 .read_iter = ocfs2_file_read_iter,
2852 .write_iter = ocfs2_file_write_it 2833 .write_iter = ocfs2_file_write_iter,
2853 .unlocked_ioctl = ocfs2_ioctl, 2834 .unlocked_ioctl = ocfs2_ioctl,
2854 #ifdef CONFIG_COMPAT 2835 #ifdef CONFIG_COMPAT
2855 .compat_ioctl = ocfs2_compat_ioctl, 2836 .compat_ioctl = ocfs2_compat_ioctl,
2856 #endif 2837 #endif
2857 .flock = ocfs2_flock, 2838 .flock = ocfs2_flock,
2858 .splice_read = filemap_splice_read 2839 .splice_read = filemap_splice_read,
2859 .splice_write = iter_file_splice_wr 2840 .splice_write = iter_file_splice_write,
2860 .fallocate = ocfs2_fallocate, 2841 .fallocate = ocfs2_fallocate,
2861 .remap_file_range = ocfs2_remap_file_ 2842 .remap_file_range = ocfs2_remap_file_range,
2862 }; 2843 };
2863 2844
2864 const struct file_operations ocfs2_dops_no_pl 2845 const struct file_operations ocfs2_dops_no_plocks = {
2865 .llseek = ocfs2_dir_llseek, !! 2846 .llseek = generic_file_llseek,
2866 .read = generic_read_dir, 2847 .read = generic_read_dir,
2867 .iterate_shared = shared_ocfs2_readdi 2848 .iterate_shared = shared_ocfs2_readdir,
2868 .fsync = ocfs2_sync_file, 2849 .fsync = ocfs2_sync_file,
2869 .release = ocfs2_dir_release, 2850 .release = ocfs2_dir_release,
2870 .open = ocfs2_dir_open, 2851 .open = ocfs2_dir_open,
2871 .unlocked_ioctl = ocfs2_ioctl, 2852 .unlocked_ioctl = ocfs2_ioctl,
2872 #ifdef CONFIG_COMPAT 2853 #ifdef CONFIG_COMPAT
2873 .compat_ioctl = ocfs2_compat_ioctl, 2854 .compat_ioctl = ocfs2_compat_ioctl,
2874 #endif 2855 #endif
2875 .flock = ocfs2_flock, 2856 .flock = ocfs2_flock,
2876 }; 2857 };
2877 2858
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.