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