xref: /openbmc/linux/fs/ocfs2/file.c (revision 8e0de82e)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * file.c
4  *
5  * File open, close, extend, truncate
6  *
7  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
8  */
9 
10 #include <linux/capability.h>
11 #include <linux/fs.h>
12 #include <linux/types.h>
13 #include <linux/slab.h>
14 #include <linux/highmem.h>
15 #include <linux/pagemap.h>
16 #include <linux/uio.h>
17 #include <linux/sched.h>
18 #include <linux/splice.h>
19 #include <linux/mount.h>
20 #include <linux/writeback.h>
21 #include <linux/falloc.h>
22 #include <linux/quotaops.h>
23 #include <linux/blkdev.h>
24 #include <linux/backing-dev.h>
25 
26 #include <cluster/masklog.h>
27 
28 #include "ocfs2.h"
29 
30 #include "alloc.h"
31 #include "aops.h"
32 #include "dir.h"
33 #include "dlmglue.h"
34 #include "extent_map.h"
35 #include "file.h"
36 #include "sysfile.h"
37 #include "inode.h"
38 #include "ioctl.h"
39 #include "journal.h"
40 #include "locks.h"
41 #include "mmap.h"
42 #include "suballoc.h"
43 #include "super.h"
44 #include "xattr.h"
45 #include "acl.h"
46 #include "quota.h"
47 #include "refcounttree.h"
48 #include "ocfs2_trace.h"
49 
50 #include "buffer_head_io.h"
51 
ocfs2_init_file_private(struct inode * inode,struct file * file)52 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
53 {
54 	struct ocfs2_file_private *fp;
55 
56 	fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
57 	if (!fp)
58 		return -ENOMEM;
59 
60 	fp->fp_file = file;
61 	mutex_init(&fp->fp_mutex);
62 	ocfs2_file_lock_res_init(&fp->fp_flock, fp);
63 	file->private_data = fp;
64 
65 	return 0;
66 }
67 
ocfs2_free_file_private(struct inode * inode,struct file * file)68 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
69 {
70 	struct ocfs2_file_private *fp = file->private_data;
71 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
72 
73 	if (fp) {
74 		ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
75 		ocfs2_lock_res_free(&fp->fp_flock);
76 		kfree(fp);
77 		file->private_data = NULL;
78 	}
79 }
80 
ocfs2_file_open(struct inode * inode,struct file * file)81 static int ocfs2_file_open(struct inode *inode, struct file *file)
82 {
83 	int status;
84 	int mode = file->f_flags;
85 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
86 
87 	trace_ocfs2_file_open(inode, file, file->f_path.dentry,
88 			      (unsigned long long)oi->ip_blkno,
89 			      file->f_path.dentry->d_name.len,
90 			      file->f_path.dentry->d_name.name, mode);
91 
92 	if (file->f_mode & FMODE_WRITE) {
93 		status = dquot_initialize(inode);
94 		if (status)
95 			goto leave;
96 	}
97 
98 	spin_lock(&oi->ip_lock);
99 
100 	/* Check that the inode hasn't been wiped from disk by another
101 	 * node. If it hasn't then we're safe as long as we hold the
102 	 * spin lock until our increment of open count. */
103 	if (oi->ip_flags & OCFS2_INODE_DELETED) {
104 		spin_unlock(&oi->ip_lock);
105 
106 		status = -ENOENT;
107 		goto leave;
108 	}
109 
110 	if (mode & O_DIRECT)
111 		oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
112 
113 	oi->ip_open_count++;
114 	spin_unlock(&oi->ip_lock);
115 
116 	status = ocfs2_init_file_private(inode, file);
117 	if (status) {
118 		/*
119 		 * We want to set open count back if we're failing the
120 		 * open.
121 		 */
122 		spin_lock(&oi->ip_lock);
123 		oi->ip_open_count--;
124 		spin_unlock(&oi->ip_lock);
125 	}
126 
127 	file->f_mode |= FMODE_NOWAIT;
128 
129 leave:
130 	return status;
131 }
132 
ocfs2_file_release(struct inode * inode,struct file * file)133 static int ocfs2_file_release(struct inode *inode, struct file *file)
134 {
135 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
136 
137 	spin_lock(&oi->ip_lock);
138 	if (!--oi->ip_open_count)
139 		oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
140 
141 	trace_ocfs2_file_release(inode, file, file->f_path.dentry,
142 				 oi->ip_blkno,
143 				 file->f_path.dentry->d_name.len,
144 				 file->f_path.dentry->d_name.name,
145 				 oi->ip_open_count);
146 	spin_unlock(&oi->ip_lock);
147 
148 	ocfs2_free_file_private(inode, file);
149 
150 	return 0;
151 }
152 
ocfs2_dir_open(struct inode * inode,struct file * file)153 static int ocfs2_dir_open(struct inode *inode, struct file *file)
154 {
155 	return ocfs2_init_file_private(inode, file);
156 }
157 
ocfs2_dir_release(struct inode * inode,struct file * file)158 static int ocfs2_dir_release(struct inode *inode, struct file *file)
159 {
160 	ocfs2_free_file_private(inode, file);
161 	return 0;
162 }
163 
ocfs2_sync_file(struct file * file,loff_t start,loff_t end,int datasync)164 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
165 			   int datasync)
166 {
167 	int err = 0;
168 	struct inode *inode = file->f_mapping->host;
169 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
170 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
171 	journal_t *journal = osb->journal->j_journal;
172 	int ret;
173 	tid_t commit_tid;
174 	bool needs_barrier = false;
175 
176 	trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
177 			      oi->ip_blkno,
178 			      file->f_path.dentry->d_name.len,
179 			      file->f_path.dentry->d_name.name,
180 			      (unsigned long long)datasync);
181 
182 	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
183 		return -EROFS;
184 
185 	err = file_write_and_wait_range(file, start, end);
186 	if (err)
187 		return err;
188 
189 	commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
190 	if (journal->j_flags & JBD2_BARRIER &&
191 	    !jbd2_trans_will_send_data_barrier(journal, commit_tid))
192 		needs_barrier = true;
193 	err = jbd2_complete_transaction(journal, commit_tid);
194 	if (needs_barrier) {
195 		ret = blkdev_issue_flush(inode->i_sb->s_bdev);
196 		if (!err)
197 			err = ret;
198 	}
199 
200 	if (err)
201 		mlog_errno(err);
202 
203 	return (err < 0) ? -EIO : 0;
204 }
205 
ocfs2_should_update_atime(struct inode * inode,struct vfsmount * vfsmnt)206 int ocfs2_should_update_atime(struct inode *inode,
207 			      struct vfsmount *vfsmnt)
208 {
209 	struct timespec64 now;
210 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
211 
212 	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
213 		return 0;
214 
215 	if ((inode->i_flags & S_NOATIME) ||
216 	    ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode)))
217 		return 0;
218 
219 	/*
220 	 * We can be called with no vfsmnt structure - NFSD will
221 	 * sometimes do this.
222 	 *
223 	 * Note that our action here is different than touch_atime() -
224 	 * if we can't tell whether this is a noatime mount, then we
225 	 * don't know whether to trust the value of s_atime_quantum.
226 	 */
227 	if (vfsmnt == NULL)
228 		return 0;
229 
230 	if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
231 	    ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
232 		return 0;
233 
234 	if (vfsmnt->mnt_flags & MNT_RELATIME) {
235 		struct timespec64 ctime = inode_get_ctime(inode);
236 		struct timespec64 atime = inode_get_atime(inode);
237 		struct timespec64 mtime = inode_get_mtime(inode);
238 
239 		if ((timespec64_compare(&atime, &mtime) <= 0) ||
240 		    (timespec64_compare(&atime, &ctime) <= 0))
241 			return 1;
242 
243 		return 0;
244 	}
245 
246 	now = current_time(inode);
247 	if ((now.tv_sec - inode_get_atime_sec(inode) <= osb->s_atime_quantum))
248 		return 0;
249 	else
250 		return 1;
251 }
252 
ocfs2_update_inode_atime(struct inode * inode,struct buffer_head * bh)253 int ocfs2_update_inode_atime(struct inode *inode,
254 			     struct buffer_head *bh)
255 {
256 	int ret;
257 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
258 	handle_t *handle;
259 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
260 
261 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
262 	if (IS_ERR(handle)) {
263 		ret = PTR_ERR(handle);
264 		mlog_errno(ret);
265 		goto out;
266 	}
267 
268 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
269 				      OCFS2_JOURNAL_ACCESS_WRITE);
270 	if (ret) {
271 		mlog_errno(ret);
272 		goto out_commit;
273 	}
274 
275 	/*
276 	 * Don't use ocfs2_mark_inode_dirty() here as we don't always
277 	 * have i_rwsem to guard against concurrent changes to other
278 	 * inode fields.
279 	 */
280 	inode_set_atime_to_ts(inode, current_time(inode));
281 	di->i_atime = cpu_to_le64(inode_get_atime_sec(inode));
282 	di->i_atime_nsec = cpu_to_le32(inode_get_atime_nsec(inode));
283 	ocfs2_update_inode_fsync_trans(handle, inode, 0);
284 	ocfs2_journal_dirty(handle, bh);
285 
286 out_commit:
287 	ocfs2_commit_trans(osb, handle);
288 out:
289 	return ret;
290 }
291 
ocfs2_set_inode_size(handle_t * handle,struct inode * inode,struct buffer_head * fe_bh,u64 new_i_size)292 int ocfs2_set_inode_size(handle_t *handle,
293 				struct inode *inode,
294 				struct buffer_head *fe_bh,
295 				u64 new_i_size)
296 {
297 	int status;
298 
299 	i_size_write(inode, new_i_size);
300 	inode->i_blocks = ocfs2_inode_sector_count(inode);
301 	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
302 
303 	status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
304 	if (status < 0) {
305 		mlog_errno(status);
306 		goto bail;
307 	}
308 
309 bail:
310 	return status;
311 }
312 
ocfs2_simple_size_update(struct inode * inode,struct buffer_head * di_bh,u64 new_i_size)313 int ocfs2_simple_size_update(struct inode *inode,
314 			     struct buffer_head *di_bh,
315 			     u64 new_i_size)
316 {
317 	int ret;
318 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
319 	handle_t *handle = NULL;
320 
321 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
322 	if (IS_ERR(handle)) {
323 		ret = PTR_ERR(handle);
324 		mlog_errno(ret);
325 		goto out;
326 	}
327 
328 	ret = ocfs2_set_inode_size(handle, inode, di_bh,
329 				   new_i_size);
330 	if (ret < 0)
331 		mlog_errno(ret);
332 
333 	ocfs2_update_inode_fsync_trans(handle, inode, 0);
334 	ocfs2_commit_trans(osb, handle);
335 out:
336 	return ret;
337 }
338 
ocfs2_cow_file_pos(struct inode * inode,struct buffer_head * fe_bh,u64 offset)339 static int ocfs2_cow_file_pos(struct inode *inode,
340 			      struct buffer_head *fe_bh,
341 			      u64 offset)
342 {
343 	int status;
344 	u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
345 	unsigned int num_clusters = 0;
346 	unsigned int ext_flags = 0;
347 
348 	/*
349 	 * If the new offset is aligned to the range of the cluster, there is
350 	 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
351 	 * CoW either.
352 	 */
353 	if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
354 		return 0;
355 
356 	status = ocfs2_get_clusters(inode, cpos, &phys,
357 				    &num_clusters, &ext_flags);
358 	if (status) {
359 		mlog_errno(status);
360 		goto out;
361 	}
362 
363 	if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
364 		goto out;
365 
366 	return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
367 
368 out:
369 	return status;
370 }
371 
ocfs2_orphan_for_truncate(struct ocfs2_super * osb,struct inode * inode,struct buffer_head * fe_bh,u64 new_i_size)372 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
373 				     struct inode *inode,
374 				     struct buffer_head *fe_bh,
375 				     u64 new_i_size)
376 {
377 	int status;
378 	handle_t *handle;
379 	struct ocfs2_dinode *di;
380 	u64 cluster_bytes;
381 
382 	/*
383 	 * We need to CoW the cluster contains the offset if it is reflinked
384 	 * since we will call ocfs2_zero_range_for_truncate later which will
385 	 * write "0" from offset to the end of the cluster.
386 	 */
387 	status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
388 	if (status) {
389 		mlog_errno(status);
390 		return status;
391 	}
392 
393 	/* TODO: This needs to actually orphan the inode in this
394 	 * transaction. */
395 
396 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
397 	if (IS_ERR(handle)) {
398 		status = PTR_ERR(handle);
399 		mlog_errno(status);
400 		goto out;
401 	}
402 
403 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
404 					 OCFS2_JOURNAL_ACCESS_WRITE);
405 	if (status < 0) {
406 		mlog_errno(status);
407 		goto out_commit;
408 	}
409 
410 	/*
411 	 * Do this before setting i_size.
412 	 */
413 	cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
414 	status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
415 					       cluster_bytes);
416 	if (status) {
417 		mlog_errno(status);
418 		goto out_commit;
419 	}
420 
421 	i_size_write(inode, new_i_size);
422 	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
423 
424 	di = (struct ocfs2_dinode *) fe_bh->b_data;
425 	di->i_size = cpu_to_le64(new_i_size);
426 	di->i_ctime = di->i_mtime = cpu_to_le64(inode_get_ctime_sec(inode));
427 	di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
428 	ocfs2_update_inode_fsync_trans(handle, inode, 0);
429 
430 	ocfs2_journal_dirty(handle, fe_bh);
431 
432 out_commit:
433 	ocfs2_commit_trans(osb, handle);
434 out:
435 	return status;
436 }
437 
ocfs2_truncate_file(struct inode * inode,struct buffer_head * di_bh,u64 new_i_size)438 int ocfs2_truncate_file(struct inode *inode,
439 			       struct buffer_head *di_bh,
440 			       u64 new_i_size)
441 {
442 	int status = 0;
443 	struct ocfs2_dinode *fe = NULL;
444 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
445 
446 	/* We trust di_bh because it comes from ocfs2_inode_lock(), which
447 	 * already validated it */
448 	fe = (struct ocfs2_dinode *) di_bh->b_data;
449 
450 	trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
451 				  (unsigned long long)le64_to_cpu(fe->i_size),
452 				  (unsigned long long)new_i_size);
453 
454 	mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
455 			"Inode %llu, inode i_size = %lld != di "
456 			"i_size = %llu, i_flags = 0x%x\n",
457 			(unsigned long long)OCFS2_I(inode)->ip_blkno,
458 			i_size_read(inode),
459 			(unsigned long long)le64_to_cpu(fe->i_size),
460 			le32_to_cpu(fe->i_flags));
461 
462 	if (new_i_size > le64_to_cpu(fe->i_size)) {
463 		trace_ocfs2_truncate_file_error(
464 			(unsigned long long)le64_to_cpu(fe->i_size),
465 			(unsigned long long)new_i_size);
466 		status = -EINVAL;
467 		mlog_errno(status);
468 		goto bail;
469 	}
470 
471 	down_write(&OCFS2_I(inode)->ip_alloc_sem);
472 
473 	ocfs2_resv_discard(&osb->osb_la_resmap,
474 			   &OCFS2_I(inode)->ip_la_data_resv);
475 
476 	/*
477 	 * The inode lock forced other nodes to sync and drop their
478 	 * pages, which (correctly) happens even if we have a truncate
479 	 * without allocation change - ocfs2 cluster sizes can be much
480 	 * greater than page size, so we have to truncate them
481 	 * anyway.
482 	 */
483 
484 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
485 		unmap_mapping_range(inode->i_mapping,
486 				    new_i_size + PAGE_SIZE - 1, 0, 1);
487 		truncate_inode_pages(inode->i_mapping, new_i_size);
488 		status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
489 					       i_size_read(inode), 1);
490 		if (status)
491 			mlog_errno(status);
492 
493 		goto bail_unlock_sem;
494 	}
495 
496 	/* alright, we're going to need to do a full blown alloc size
497 	 * change. Orphan the inode so that recovery can complete the
498 	 * truncate if necessary. This does the task of marking
499 	 * i_size. */
500 	status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
501 	if (status < 0) {
502 		mlog_errno(status);
503 		goto bail_unlock_sem;
504 	}
505 
506 	unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
507 	truncate_inode_pages(inode->i_mapping, new_i_size);
508 
509 	status = ocfs2_commit_truncate(osb, inode, di_bh);
510 	if (status < 0) {
511 		mlog_errno(status);
512 		goto bail_unlock_sem;
513 	}
514 
515 	/* TODO: orphan dir cleanup here. */
516 bail_unlock_sem:
517 	up_write(&OCFS2_I(inode)->ip_alloc_sem);
518 
519 bail:
520 	if (!status && OCFS2_I(inode)->ip_clusters == 0)
521 		status = ocfs2_try_remove_refcount_tree(inode, di_bh);
522 
523 	return status;
524 }
525 
526 /*
527  * extend file allocation only here.
528  * we'll update all the disk stuff, and oip->alloc_size
529  *
530  * expect stuff to be locked, a transaction started and enough data /
531  * metadata reservations in the contexts.
532  *
533  * Will return -EAGAIN, and a reason if a restart is needed.
534  * If passed in, *reason will always be set, even in error.
535  */
ocfs2_add_inode_data(struct ocfs2_super * osb,struct inode * inode,u32 * logical_offset,u32 clusters_to_add,int mark_unwritten,struct buffer_head * fe_bh,handle_t * handle,struct ocfs2_alloc_context * data_ac,struct ocfs2_alloc_context * meta_ac,enum ocfs2_alloc_restarted * reason_ret)536 int ocfs2_add_inode_data(struct ocfs2_super *osb,
537 			 struct inode *inode,
538 			 u32 *logical_offset,
539 			 u32 clusters_to_add,
540 			 int mark_unwritten,
541 			 struct buffer_head *fe_bh,
542 			 handle_t *handle,
543 			 struct ocfs2_alloc_context *data_ac,
544 			 struct ocfs2_alloc_context *meta_ac,
545 			 enum ocfs2_alloc_restarted *reason_ret)
546 {
547 	struct ocfs2_extent_tree et;
548 
549 	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
550 	return ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
551 					   clusters_to_add, mark_unwritten,
552 					   data_ac, meta_ac, reason_ret);
553 }
554 
ocfs2_extend_allocation(struct inode * inode,u32 logical_start,u32 clusters_to_add,int mark_unwritten)555 static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
556 				   u32 clusters_to_add, int mark_unwritten)
557 {
558 	int status = 0;
559 	int restart_func = 0;
560 	int credits;
561 	u32 prev_clusters;
562 	struct buffer_head *bh = NULL;
563 	struct ocfs2_dinode *fe = NULL;
564 	handle_t *handle = NULL;
565 	struct ocfs2_alloc_context *data_ac = NULL;
566 	struct ocfs2_alloc_context *meta_ac = NULL;
567 	enum ocfs2_alloc_restarted why = RESTART_NONE;
568 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
569 	struct ocfs2_extent_tree et;
570 	int did_quota = 0;
571 
572 	/*
573 	 * Unwritten extent only exists for file systems which
574 	 * support holes.
575 	 */
576 	BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
577 
578 	status = ocfs2_read_inode_block(inode, &bh);
579 	if (status < 0) {
580 		mlog_errno(status);
581 		goto leave;
582 	}
583 	fe = (struct ocfs2_dinode *) bh->b_data;
584 
585 restart_all:
586 	BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
587 
588 	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
589 	status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
590 				       &data_ac, &meta_ac);
591 	if (status) {
592 		mlog_errno(status);
593 		goto leave;
594 	}
595 
596 	credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
597 	handle = ocfs2_start_trans(osb, credits);
598 	if (IS_ERR(handle)) {
599 		status = PTR_ERR(handle);
600 		handle = NULL;
601 		mlog_errno(status);
602 		goto leave;
603 	}
604 
605 restarted_transaction:
606 	trace_ocfs2_extend_allocation(
607 		(unsigned long long)OCFS2_I(inode)->ip_blkno,
608 		(unsigned long long)i_size_read(inode),
609 		le32_to_cpu(fe->i_clusters), clusters_to_add,
610 		why, restart_func);
611 
612 	status = dquot_alloc_space_nodirty(inode,
613 			ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
614 	if (status)
615 		goto leave;
616 	did_quota = 1;
617 
618 	/* reserve a write to the file entry early on - that we if we
619 	 * run out of credits in the allocation path, we can still
620 	 * update i_size. */
621 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
622 					 OCFS2_JOURNAL_ACCESS_WRITE);
623 	if (status < 0) {
624 		mlog_errno(status);
625 		goto leave;
626 	}
627 
628 	prev_clusters = OCFS2_I(inode)->ip_clusters;
629 
630 	status = ocfs2_add_inode_data(osb,
631 				      inode,
632 				      &logical_start,
633 				      clusters_to_add,
634 				      mark_unwritten,
635 				      bh,
636 				      handle,
637 				      data_ac,
638 				      meta_ac,
639 				      &why);
640 	if ((status < 0) && (status != -EAGAIN)) {
641 		if (status != -ENOSPC)
642 			mlog_errno(status);
643 		goto leave;
644 	}
645 	ocfs2_update_inode_fsync_trans(handle, inode, 1);
646 	ocfs2_journal_dirty(handle, bh);
647 
648 	spin_lock(&OCFS2_I(inode)->ip_lock);
649 	clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
650 	spin_unlock(&OCFS2_I(inode)->ip_lock);
651 	/* Release unused quota reservation */
652 	dquot_free_space(inode,
653 			ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
654 	did_quota = 0;
655 
656 	if (why != RESTART_NONE && clusters_to_add) {
657 		if (why == RESTART_META) {
658 			restart_func = 1;
659 			status = 0;
660 		} else {
661 			BUG_ON(why != RESTART_TRANS);
662 
663 			status = ocfs2_allocate_extend_trans(handle, 1);
664 			if (status < 0) {
665 				/* handle still has to be committed at
666 				 * this point. */
667 				status = -ENOMEM;
668 				mlog_errno(status);
669 				goto leave;
670 			}
671 			goto restarted_transaction;
672 		}
673 	}
674 
675 	trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
676 	     le32_to_cpu(fe->i_clusters),
677 	     (unsigned long long)le64_to_cpu(fe->i_size),
678 	     OCFS2_I(inode)->ip_clusters,
679 	     (unsigned long long)i_size_read(inode));
680 
681 leave:
682 	if (status < 0 && did_quota)
683 		dquot_free_space(inode,
684 			ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
685 	if (handle) {
686 		ocfs2_commit_trans(osb, handle);
687 		handle = NULL;
688 	}
689 	if (data_ac) {
690 		ocfs2_free_alloc_context(data_ac);
691 		data_ac = NULL;
692 	}
693 	if (meta_ac) {
694 		ocfs2_free_alloc_context(meta_ac);
695 		meta_ac = NULL;
696 	}
697 	if ((!status) && restart_func) {
698 		restart_func = 0;
699 		goto restart_all;
700 	}
701 	brelse(bh);
702 	bh = NULL;
703 
704 	return status;
705 }
706 
707 /*
708  * While a write will already be ordering the data, a truncate will not.
709  * Thus, we need to explicitly order the zeroed pages.
710  */
ocfs2_zero_start_ordered_transaction(struct inode * inode,struct buffer_head * di_bh,loff_t start_byte,loff_t length)711 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
712 						      struct buffer_head *di_bh,
713 						      loff_t start_byte,
714 						      loff_t length)
715 {
716 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
717 	handle_t *handle = NULL;
718 	int ret = 0;
719 
720 	if (!ocfs2_should_order_data(inode))
721 		goto out;
722 
723 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
724 	if (IS_ERR(handle)) {
725 		ret = -ENOMEM;
726 		mlog_errno(ret);
727 		goto out;
728 	}
729 
730 	ret = ocfs2_jbd2_inode_add_write(handle, inode, start_byte, length);
731 	if (ret < 0) {
732 		mlog_errno(ret);
733 		goto out;
734 	}
735 
736 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
737 				      OCFS2_JOURNAL_ACCESS_WRITE);
738 	if (ret)
739 		mlog_errno(ret);
740 	ocfs2_update_inode_fsync_trans(handle, inode, 1);
741 
742 out:
743 	if (ret) {
744 		if (!IS_ERR(handle))
745 			ocfs2_commit_trans(osb, handle);
746 		handle = ERR_PTR(ret);
747 	}
748 	return handle;
749 }
750 
751 /* Some parts of this taken from generic_cont_expand, which turned out
752  * to be too fragile to do exactly what we need without us having to
753  * worry about recursive locking in ->write_begin() and ->write_end(). */
ocfs2_write_zero_page(struct inode * inode,u64 abs_from,u64 abs_to,struct buffer_head * di_bh)754 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
755 				 u64 abs_to, struct buffer_head *di_bh)
756 {
757 	struct address_space *mapping = inode->i_mapping;
758 	struct page *page;
759 	unsigned long index = abs_from >> PAGE_SHIFT;
760 	handle_t *handle;
761 	int ret = 0;
762 	unsigned zero_from, zero_to, block_start, block_end;
763 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
764 
765 	BUG_ON(abs_from >= abs_to);
766 	BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
767 	BUG_ON(abs_from & (inode->i_blkbits - 1));
768 
769 	handle = ocfs2_zero_start_ordered_transaction(inode, di_bh,
770 						      abs_from,
771 						      abs_to - abs_from);
772 	if (IS_ERR(handle)) {
773 		ret = PTR_ERR(handle);
774 		goto out;
775 	}
776 
777 	page = find_or_create_page(mapping, index, GFP_NOFS);
778 	if (!page) {
779 		ret = -ENOMEM;
780 		mlog_errno(ret);
781 		goto out_commit_trans;
782 	}
783 
784 	/* Get the offsets within the page that we want to zero */
785 	zero_from = abs_from & (PAGE_SIZE - 1);
786 	zero_to = abs_to & (PAGE_SIZE - 1);
787 	if (!zero_to)
788 		zero_to = PAGE_SIZE;
789 
790 	trace_ocfs2_write_zero_page(
791 			(unsigned long long)OCFS2_I(inode)->ip_blkno,
792 			(unsigned long long)abs_from,
793 			(unsigned long long)abs_to,
794 			index, zero_from, zero_to);
795 
796 	/* We know that zero_from is block aligned */
797 	for (block_start = zero_from; block_start < zero_to;
798 	     block_start = block_end) {
799 		block_end = block_start + i_blocksize(inode);
800 
801 		/*
802 		 * block_start is block-aligned.  Bump it by one to force
803 		 * __block_write_begin and block_commit_write to zero the
804 		 * whole block.
805 		 */
806 		ret = __block_write_begin(page, block_start + 1, 0,
807 					  ocfs2_get_block);
808 		if (ret < 0) {
809 			mlog_errno(ret);
810 			goto out_unlock;
811 		}
812 
813 
814 		/* must not update i_size! */
815 		block_commit_write(page, block_start + 1, block_start + 1);
816 	}
817 
818 	/*
819 	 * fs-writeback will release the dirty pages without page lock
820 	 * whose offset are over inode size, the release happens at
821 	 * block_write_full_page().
822 	 */
823 	i_size_write(inode, abs_to);
824 	inode->i_blocks = ocfs2_inode_sector_count(inode);
825 	di->i_size = cpu_to_le64((u64)i_size_read(inode));
826 	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
827 	di->i_mtime = di->i_ctime = cpu_to_le64(inode_get_mtime_sec(inode));
828 	di->i_ctime_nsec = cpu_to_le32(inode_get_mtime_nsec(inode));
829 	di->i_mtime_nsec = di->i_ctime_nsec;
830 	if (handle) {
831 		ocfs2_journal_dirty(handle, di_bh);
832 		ocfs2_update_inode_fsync_trans(handle, inode, 1);
833 	}
834 
835 out_unlock:
836 	unlock_page(page);
837 	put_page(page);
838 out_commit_trans:
839 	if (handle)
840 		ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
841 out:
842 	return ret;
843 }
844 
845 /*
846  * Find the next range to zero.  We do this in terms of bytes because
847  * that's what ocfs2_zero_extend() wants, and it is dealing with the
848  * pagecache.  We may return multiple extents.
849  *
850  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
851  * needs to be zeroed.  range_start and range_end return the next zeroing
852  * range.  A subsequent call should pass the previous range_end as its
853  * zero_start.  If range_end is 0, there's nothing to do.
854  *
855  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
856  */
ocfs2_zero_extend_get_range(struct inode * inode,struct buffer_head * di_bh,u64 zero_start,u64 zero_end,u64 * range_start,u64 * range_end)857 static int ocfs2_zero_extend_get_range(struct inode *inode,
858 				       struct buffer_head *di_bh,
859 				       u64 zero_start, u64 zero_end,
860 				       u64 *range_start, u64 *range_end)
861 {
862 	int rc = 0, needs_cow = 0;
863 	u32 p_cpos, zero_clusters = 0;
864 	u32 zero_cpos =
865 		zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
866 	u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
867 	unsigned int num_clusters = 0;
868 	unsigned int ext_flags = 0;
869 
870 	while (zero_cpos < last_cpos) {
871 		rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
872 					&num_clusters, &ext_flags);
873 		if (rc) {
874 			mlog_errno(rc);
875 			goto out;
876 		}
877 
878 		if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
879 			zero_clusters = num_clusters;
880 			if (ext_flags & OCFS2_EXT_REFCOUNTED)
881 				needs_cow = 1;
882 			break;
883 		}
884 
885 		zero_cpos += num_clusters;
886 	}
887 	if (!zero_clusters) {
888 		*range_end = 0;
889 		goto out;
890 	}
891 
892 	while ((zero_cpos + zero_clusters) < last_cpos) {
893 		rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
894 					&p_cpos, &num_clusters,
895 					&ext_flags);
896 		if (rc) {
897 			mlog_errno(rc);
898 			goto out;
899 		}
900 
901 		if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
902 			break;
903 		if (ext_flags & OCFS2_EXT_REFCOUNTED)
904 			needs_cow = 1;
905 		zero_clusters += num_clusters;
906 	}
907 	if ((zero_cpos + zero_clusters) > last_cpos)
908 		zero_clusters = last_cpos - zero_cpos;
909 
910 	if (needs_cow) {
911 		rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
912 					zero_clusters, UINT_MAX);
913 		if (rc) {
914 			mlog_errno(rc);
915 			goto out;
916 		}
917 	}
918 
919 	*range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
920 	*range_end = ocfs2_clusters_to_bytes(inode->i_sb,
921 					     zero_cpos + zero_clusters);
922 
923 out:
924 	return rc;
925 }
926 
927 /*
928  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
929  * has made sure that the entire range needs zeroing.
930  */
ocfs2_zero_extend_range(struct inode * inode,u64 range_start,u64 range_end,struct buffer_head * di_bh)931 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
932 				   u64 range_end, struct buffer_head *di_bh)
933 {
934 	int rc = 0;
935 	u64 next_pos;
936 	u64 zero_pos = range_start;
937 
938 	trace_ocfs2_zero_extend_range(
939 			(unsigned long long)OCFS2_I(inode)->ip_blkno,
940 			(unsigned long long)range_start,
941 			(unsigned long long)range_end);
942 	BUG_ON(range_start >= range_end);
943 
944 	while (zero_pos < range_end) {
945 		next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
946 		if (next_pos > range_end)
947 			next_pos = range_end;
948 		rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
949 		if (rc < 0) {
950 			mlog_errno(rc);
951 			break;
952 		}
953 		zero_pos = next_pos;
954 
955 		/*
956 		 * Very large extends have the potential to lock up
957 		 * the cpu for extended periods of time.
958 		 */
959 		cond_resched();
960 	}
961 
962 	return rc;
963 }
964 
ocfs2_zero_extend(struct inode * inode,struct buffer_head * di_bh,loff_t zero_to_size)965 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
966 		      loff_t zero_to_size)
967 {
968 	int ret = 0;
969 	u64 zero_start, range_start = 0, range_end = 0;
970 	struct super_block *sb = inode->i_sb;
971 
972 	zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
973 	trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
974 				(unsigned long long)zero_start,
975 				(unsigned long long)i_size_read(inode));
976 	while (zero_start < zero_to_size) {
977 		ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
978 						  zero_to_size,
979 						  &range_start,
980 						  &range_end);
981 		if (ret) {
982 			mlog_errno(ret);
983 			break;
984 		}
985 		if (!range_end)
986 			break;
987 		/* Trim the ends */
988 		if (range_start < zero_start)
989 			range_start = zero_start;
990 		if (range_end > zero_to_size)
991 			range_end = zero_to_size;
992 
993 		ret = ocfs2_zero_extend_range(inode, range_start,
994 					      range_end, di_bh);
995 		if (ret) {
996 			mlog_errno(ret);
997 			break;
998 		}
999 		zero_start = range_end;
1000 	}
1001 
1002 	return ret;
1003 }
1004 
ocfs2_extend_no_holes(struct inode * inode,struct buffer_head * di_bh,u64 new_i_size,u64 zero_to)1005 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1006 			  u64 new_i_size, u64 zero_to)
1007 {
1008 	int ret;
1009 	u32 clusters_to_add;
1010 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1011 
1012 	/*
1013 	 * Only quota files call this without a bh, and they can't be
1014 	 * refcounted.
1015 	 */
1016 	BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1017 	BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1018 
1019 	clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1020 	if (clusters_to_add < oi->ip_clusters)
1021 		clusters_to_add = 0;
1022 	else
1023 		clusters_to_add -= oi->ip_clusters;
1024 
1025 	if (clusters_to_add) {
1026 		ret = ocfs2_extend_allocation(inode, oi->ip_clusters,
1027 					      clusters_to_add, 0);
1028 		if (ret) {
1029 			mlog_errno(ret);
1030 			goto out;
1031 		}
1032 	}
1033 
1034 	/*
1035 	 * Call this even if we don't add any clusters to the tree. We
1036 	 * still need to zero the area between the old i_size and the
1037 	 * new i_size.
1038 	 */
1039 	ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1040 	if (ret < 0)
1041 		mlog_errno(ret);
1042 
1043 out:
1044 	return ret;
1045 }
1046 
ocfs2_extend_file(struct inode * inode,struct buffer_head * di_bh,u64 new_i_size)1047 static int ocfs2_extend_file(struct inode *inode,
1048 			     struct buffer_head *di_bh,
1049 			     u64 new_i_size)
1050 {
1051 	int ret = 0;
1052 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1053 
1054 	BUG_ON(!di_bh);
1055 
1056 	/* setattr sometimes calls us like this. */
1057 	if (new_i_size == 0)
1058 		goto out;
1059 
1060 	if (i_size_read(inode) == new_i_size)
1061 		goto out;
1062 	BUG_ON(new_i_size < i_size_read(inode));
1063 
1064 	/*
1065 	 * The alloc sem blocks people in read/write from reading our
1066 	 * allocation until we're done changing it. We depend on
1067 	 * i_rwsem to block other extend/truncate calls while we're
1068 	 * here.  We even have to hold it for sparse files because there
1069 	 * might be some tail zeroing.
1070 	 */
1071 	down_write(&oi->ip_alloc_sem);
1072 
1073 	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1074 		/*
1075 		 * We can optimize small extends by keeping the inodes
1076 		 * inline data.
1077 		 */
1078 		if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1079 			up_write(&oi->ip_alloc_sem);
1080 			goto out_update_size;
1081 		}
1082 
1083 		ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1084 		if (ret) {
1085 			up_write(&oi->ip_alloc_sem);
1086 			mlog_errno(ret);
1087 			goto out;
1088 		}
1089 	}
1090 
1091 	if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1092 		ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1093 	else
1094 		ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1095 					    new_i_size);
1096 
1097 	up_write(&oi->ip_alloc_sem);
1098 
1099 	if (ret < 0) {
1100 		mlog_errno(ret);
1101 		goto out;
1102 	}
1103 
1104 out_update_size:
1105 	ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1106 	if (ret < 0)
1107 		mlog_errno(ret);
1108 
1109 out:
1110 	return ret;
1111 }
1112 
ocfs2_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * attr)1113 int ocfs2_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
1114 		  struct iattr *attr)
1115 {
1116 	int status = 0, size_change;
1117 	int inode_locked = 0;
1118 	struct inode *inode = d_inode(dentry);
1119 	struct super_block *sb = inode->i_sb;
1120 	struct ocfs2_super *osb = OCFS2_SB(sb);
1121 	struct buffer_head *bh = NULL;
1122 	handle_t *handle = NULL;
1123 	struct dquot *transfer_to[MAXQUOTAS] = { };
1124 	int qtype;
1125 	int had_lock;
1126 	struct ocfs2_lock_holder oh;
1127 
1128 	trace_ocfs2_setattr(inode, dentry,
1129 			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
1130 			    dentry->d_name.len, dentry->d_name.name,
1131 			    attr->ia_valid,
1132 				attr->ia_valid & ATTR_MODE ? attr->ia_mode : 0,
1133 				attr->ia_valid & ATTR_UID ?
1134 					from_kuid(&init_user_ns, attr->ia_uid) : 0,
1135 				attr->ia_valid & ATTR_GID ?
1136 					from_kgid(&init_user_ns, attr->ia_gid) : 0);
1137 
1138 	/* ensuring we don't even attempt to truncate a symlink */
1139 	if (S_ISLNK(inode->i_mode))
1140 		attr->ia_valid &= ~ATTR_SIZE;
1141 
1142 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1143 			   | ATTR_GID | ATTR_UID | ATTR_MODE)
1144 	if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1145 		return 0;
1146 
1147 	status = setattr_prepare(&nop_mnt_idmap, dentry, attr);
1148 	if (status)
1149 		return status;
1150 
1151 	if (is_quota_modification(&nop_mnt_idmap, inode, attr)) {
1152 		status = dquot_initialize(inode);
1153 		if (status)
1154 			return status;
1155 	}
1156 	size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1157 	if (size_change) {
1158 		/*
1159 		 * Here we should wait dio to finish before inode lock
1160 		 * to avoid a deadlock between ocfs2_setattr() and
1161 		 * ocfs2_dio_end_io_write()
1162 		 */
1163 		inode_dio_wait(inode);
1164 
1165 		status = ocfs2_rw_lock(inode, 1);
1166 		if (status < 0) {
1167 			mlog_errno(status);
1168 			goto bail;
1169 		}
1170 	}
1171 
1172 	had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1173 	if (had_lock < 0) {
1174 		status = had_lock;
1175 		goto bail_unlock_rw;
1176 	} else if (had_lock) {
1177 		/*
1178 		 * As far as we know, ocfs2_setattr() could only be the first
1179 		 * VFS entry point in the call chain of recursive cluster
1180 		 * locking issue.
1181 		 *
1182 		 * For instance:
1183 		 * chmod_common()
1184 		 *  notify_change()
1185 		 *   ocfs2_setattr()
1186 		 *    posix_acl_chmod()
1187 		 *     ocfs2_iop_get_acl()
1188 		 *
1189 		 * But, we're not 100% sure if it's always true, because the
1190 		 * ordering of the VFS entry points in the call chain is out
1191 		 * of our control. So, we'd better dump the stack here to
1192 		 * catch the other cases of recursive locking.
1193 		 */
1194 		mlog(ML_ERROR, "Another case of recursive locking:\n");
1195 		dump_stack();
1196 	}
1197 	inode_locked = 1;
1198 
1199 	if (size_change) {
1200 		status = inode_newsize_ok(inode, attr->ia_size);
1201 		if (status)
1202 			goto bail_unlock;
1203 
1204 		if (i_size_read(inode) >= attr->ia_size) {
1205 			if (ocfs2_should_order_data(inode)) {
1206 				status = ocfs2_begin_ordered_truncate(inode,
1207 								      attr->ia_size);
1208 				if (status)
1209 					goto bail_unlock;
1210 			}
1211 			status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1212 		} else
1213 			status = ocfs2_extend_file(inode, bh, attr->ia_size);
1214 		if (status < 0) {
1215 			if (status != -ENOSPC)
1216 				mlog_errno(status);
1217 			status = -ENOSPC;
1218 			goto bail_unlock;
1219 		}
1220 	}
1221 
1222 	if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1223 	    (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1224 		/*
1225 		 * Gather pointers to quota structures so that allocation /
1226 		 * freeing of quota structures happens here and not inside
1227 		 * dquot_transfer() where we have problems with lock ordering
1228 		 */
1229 		if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1230 		    && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1231 		    OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1232 			transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1233 			if (IS_ERR(transfer_to[USRQUOTA])) {
1234 				status = PTR_ERR(transfer_to[USRQUOTA]);
1235 				transfer_to[USRQUOTA] = NULL;
1236 				goto bail_unlock;
1237 			}
1238 		}
1239 		if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1240 		    && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1241 		    OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1242 			transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1243 			if (IS_ERR(transfer_to[GRPQUOTA])) {
1244 				status = PTR_ERR(transfer_to[GRPQUOTA]);
1245 				transfer_to[GRPQUOTA] = NULL;
1246 				goto bail_unlock;
1247 			}
1248 		}
1249 		down_write(&OCFS2_I(inode)->ip_alloc_sem);
1250 		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1251 					   2 * ocfs2_quota_trans_credits(sb));
1252 		if (IS_ERR(handle)) {
1253 			status = PTR_ERR(handle);
1254 			mlog_errno(status);
1255 			goto bail_unlock_alloc;
1256 		}
1257 		status = __dquot_transfer(inode, transfer_to);
1258 		if (status < 0)
1259 			goto bail_commit;
1260 	} else {
1261 		down_write(&OCFS2_I(inode)->ip_alloc_sem);
1262 		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1263 		if (IS_ERR(handle)) {
1264 			status = PTR_ERR(handle);
1265 			mlog_errno(status);
1266 			goto bail_unlock_alloc;
1267 		}
1268 	}
1269 
1270 	setattr_copy(&nop_mnt_idmap, inode, attr);
1271 	mark_inode_dirty(inode);
1272 
1273 	status = ocfs2_mark_inode_dirty(handle, inode, bh);
1274 	if (status < 0)
1275 		mlog_errno(status);
1276 
1277 bail_commit:
1278 	ocfs2_commit_trans(osb, handle);
1279 bail_unlock_alloc:
1280 	up_write(&OCFS2_I(inode)->ip_alloc_sem);
1281 bail_unlock:
1282 	if (status && inode_locked) {
1283 		ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1284 		inode_locked = 0;
1285 	}
1286 bail_unlock_rw:
1287 	if (size_change)
1288 		ocfs2_rw_unlock(inode, 1);
1289 bail:
1290 
1291 	/* Release quota pointers in case we acquired them */
1292 	for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1293 		dqput(transfer_to[qtype]);
1294 
1295 	if (!status && attr->ia_valid & ATTR_MODE) {
1296 		status = ocfs2_acl_chmod(inode, bh);
1297 		if (status < 0)
1298 			mlog_errno(status);
1299 	}
1300 	if (inode_locked)
1301 		ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1302 
1303 	brelse(bh);
1304 	return status;
1305 }
1306 
ocfs2_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int flags)1307 int ocfs2_getattr(struct mnt_idmap *idmap, const struct path *path,
1308 		  struct kstat *stat, u32 request_mask, unsigned int flags)
1309 {
1310 	struct inode *inode = d_inode(path->dentry);
1311 	struct super_block *sb = path->dentry->d_sb;
1312 	struct ocfs2_super *osb = sb->s_fs_info;
1313 	int err;
1314 
1315 	err = ocfs2_inode_revalidate(path->dentry);
1316 	if (err) {
1317 		if (err != -ENOENT)
1318 			mlog_errno(err);
1319 		goto bail;
1320 	}
1321 
1322 	generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1323 	/*
1324 	 * If there is inline data in the inode, the inode will normally not
1325 	 * have data blocks allocated (it may have an external xattr block).
1326 	 * Report at least one sector for such files, so tools like tar, rsync,
1327 	 * others don't incorrectly think the file is completely sparse.
1328 	 */
1329 	if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1330 		stat->blocks += (stat->size + 511)>>9;
1331 
1332 	/* We set the blksize from the cluster size for performance */
1333 	stat->blksize = osb->s_clustersize;
1334 
1335 bail:
1336 	return err;
1337 }
1338 
ocfs2_permission(struct mnt_idmap * idmap,struct inode * inode,int mask)1339 int ocfs2_permission(struct mnt_idmap *idmap, struct inode *inode,
1340 		     int mask)
1341 {
1342 	int ret, had_lock;
1343 	struct ocfs2_lock_holder oh;
1344 
1345 	if (mask & MAY_NOT_BLOCK)
1346 		return -ECHILD;
1347 
1348 	had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1349 	if (had_lock < 0) {
1350 		ret = had_lock;
1351 		goto out;
1352 	} else if (had_lock) {
1353 		/* See comments in ocfs2_setattr() for details.
1354 		 * The call chain of this case could be:
1355 		 * do_sys_open()
1356 		 *  may_open()
1357 		 *   inode_permission()
1358 		 *    ocfs2_permission()
1359 		 *     ocfs2_iop_get_acl()
1360 		 */
1361 		mlog(ML_ERROR, "Another case of recursive locking:\n");
1362 		dump_stack();
1363 	}
1364 
1365 	ret = generic_permission(&nop_mnt_idmap, inode, mask);
1366 
1367 	ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1368 out:
1369 	return ret;
1370 }
1371 
__ocfs2_write_remove_suid(struct inode * inode,struct buffer_head * bh)1372 static int __ocfs2_write_remove_suid(struct inode *inode,
1373 				     struct buffer_head *bh)
1374 {
1375 	int ret;
1376 	handle_t *handle;
1377 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1378 	struct ocfs2_dinode *di;
1379 
1380 	trace_ocfs2_write_remove_suid(
1381 			(unsigned long long)OCFS2_I(inode)->ip_blkno,
1382 			inode->i_mode);
1383 
1384 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1385 	if (IS_ERR(handle)) {
1386 		ret = PTR_ERR(handle);
1387 		mlog_errno(ret);
1388 		goto out;
1389 	}
1390 
1391 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1392 				      OCFS2_JOURNAL_ACCESS_WRITE);
1393 	if (ret < 0) {
1394 		mlog_errno(ret);
1395 		goto out_trans;
1396 	}
1397 
1398 	inode->i_mode &= ~S_ISUID;
1399 	if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1400 		inode->i_mode &= ~S_ISGID;
1401 
1402 	di = (struct ocfs2_dinode *) bh->b_data;
1403 	di->i_mode = cpu_to_le16(inode->i_mode);
1404 	ocfs2_update_inode_fsync_trans(handle, inode, 0);
1405 
1406 	ocfs2_journal_dirty(handle, bh);
1407 
1408 out_trans:
1409 	ocfs2_commit_trans(osb, handle);
1410 out:
1411 	return ret;
1412 }
1413 
ocfs2_write_remove_suid(struct inode * inode)1414 static int ocfs2_write_remove_suid(struct inode *inode)
1415 {
1416 	int ret;
1417 	struct buffer_head *bh = NULL;
1418 
1419 	ret = ocfs2_read_inode_block(inode, &bh);
1420 	if (ret < 0) {
1421 		mlog_errno(ret);
1422 		goto out;
1423 	}
1424 
1425 	ret =  __ocfs2_write_remove_suid(inode, bh);
1426 out:
1427 	brelse(bh);
1428 	return ret;
1429 }
1430 
1431 /*
1432  * Allocate enough extents to cover the region starting at byte offset
1433  * start for len bytes. Existing extents are skipped, any extents
1434  * added are marked as "unwritten".
1435  */
ocfs2_allocate_unwritten_extents(struct inode * inode,u64 start,u64 len)1436 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1437 					    u64 start, u64 len)
1438 {
1439 	int ret;
1440 	u32 cpos, phys_cpos, clusters, alloc_size;
1441 	u64 end = start + len;
1442 	struct buffer_head *di_bh = NULL;
1443 
1444 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1445 		ret = ocfs2_read_inode_block(inode, &di_bh);
1446 		if (ret) {
1447 			mlog_errno(ret);
1448 			goto out;
1449 		}
1450 
1451 		/*
1452 		 * Nothing to do if the requested reservation range
1453 		 * fits within the inode.
1454 		 */
1455 		if (ocfs2_size_fits_inline_data(di_bh, end))
1456 			goto out;
1457 
1458 		ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1459 		if (ret) {
1460 			mlog_errno(ret);
1461 			goto out;
1462 		}
1463 	}
1464 
1465 	/*
1466 	 * We consider both start and len to be inclusive.
1467 	 */
1468 	cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1469 	clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1470 	clusters -= cpos;
1471 
1472 	while (clusters) {
1473 		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1474 					 &alloc_size, NULL);
1475 		if (ret) {
1476 			mlog_errno(ret);
1477 			goto out;
1478 		}
1479 
1480 		/*
1481 		 * Hole or existing extent len can be arbitrary, so
1482 		 * cap it to our own allocation request.
1483 		 */
1484 		if (alloc_size > clusters)
1485 			alloc_size = clusters;
1486 
1487 		if (phys_cpos) {
1488 			/*
1489 			 * We already have an allocation at this
1490 			 * region so we can safely skip it.
1491 			 */
1492 			goto next;
1493 		}
1494 
1495 		ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1496 		if (ret) {
1497 			if (ret != -ENOSPC)
1498 				mlog_errno(ret);
1499 			goto out;
1500 		}
1501 
1502 next:
1503 		cpos += alloc_size;
1504 		clusters -= alloc_size;
1505 	}
1506 
1507 	ret = 0;
1508 out:
1509 
1510 	brelse(di_bh);
1511 	return ret;
1512 }
1513 
1514 /*
1515  * Truncate a byte range, avoiding pages within partial clusters. This
1516  * preserves those pages for the zeroing code to write to.
1517  */
ocfs2_truncate_cluster_pages(struct inode * inode,u64 byte_start,u64 byte_len)1518 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1519 					 u64 byte_len)
1520 {
1521 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1522 	loff_t start, end;
1523 	struct address_space *mapping = inode->i_mapping;
1524 
1525 	start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1526 	end = byte_start + byte_len;
1527 	end = end & ~(osb->s_clustersize - 1);
1528 
1529 	if (start < end) {
1530 		unmap_mapping_range(mapping, start, end - start, 0);
1531 		truncate_inode_pages_range(mapping, start, end - 1);
1532 	}
1533 }
1534 
1535 /*
1536  * zero out partial blocks of one cluster.
1537  *
1538  * start: file offset where zero starts, will be made upper block aligned.
1539  * len: it will be trimmed to the end of current cluster if "start + len"
1540  *      is bigger than it.
1541  */
ocfs2_zeroout_partial_cluster(struct inode * inode,u64 start,u64 len)1542 static int ocfs2_zeroout_partial_cluster(struct inode *inode,
1543 					u64 start, u64 len)
1544 {
1545 	int ret;
1546 	u64 start_block, end_block, nr_blocks;
1547 	u64 p_block, offset;
1548 	u32 cluster, p_cluster, nr_clusters;
1549 	struct super_block *sb = inode->i_sb;
1550 	u64 end = ocfs2_align_bytes_to_clusters(sb, start);
1551 
1552 	if (start + len < end)
1553 		end = start + len;
1554 
1555 	start_block = ocfs2_blocks_for_bytes(sb, start);
1556 	end_block = ocfs2_blocks_for_bytes(sb, end);
1557 	nr_blocks = end_block - start_block;
1558 	if (!nr_blocks)
1559 		return 0;
1560 
1561 	cluster = ocfs2_bytes_to_clusters(sb, start);
1562 	ret = ocfs2_get_clusters(inode, cluster, &p_cluster,
1563 				&nr_clusters, NULL);
1564 	if (ret)
1565 		return ret;
1566 	if (!p_cluster)
1567 		return 0;
1568 
1569 	offset = start_block - ocfs2_clusters_to_blocks(sb, cluster);
1570 	p_block = ocfs2_clusters_to_blocks(sb, p_cluster) + offset;
1571 	return sb_issue_zeroout(sb, p_block, nr_blocks, GFP_NOFS);
1572 }
1573 
ocfs2_zero_partial_clusters(struct inode * inode,u64 start,u64 len)1574 static int ocfs2_zero_partial_clusters(struct inode *inode,
1575 				       u64 start, u64 len)
1576 {
1577 	int ret = 0;
1578 	u64 tmpend = 0;
1579 	u64 end = start + len;
1580 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1581 	unsigned int csize = osb->s_clustersize;
1582 	handle_t *handle;
1583 	loff_t isize = i_size_read(inode);
1584 
1585 	/*
1586 	 * The "start" and "end" values are NOT necessarily part of
1587 	 * the range whose allocation is being deleted. Rather, this
1588 	 * is what the user passed in with the request. We must zero
1589 	 * partial clusters here. There's no need to worry about
1590 	 * physical allocation - the zeroing code knows to skip holes.
1591 	 */
1592 	trace_ocfs2_zero_partial_clusters(
1593 		(unsigned long long)OCFS2_I(inode)->ip_blkno,
1594 		(unsigned long long)start, (unsigned long long)end);
1595 
1596 	/*
1597 	 * If both edges are on a cluster boundary then there's no
1598 	 * zeroing required as the region is part of the allocation to
1599 	 * be truncated.
1600 	 */
1601 	if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1602 		goto out;
1603 
1604 	/* No page cache for EOF blocks, issue zero out to disk. */
1605 	if (end > isize) {
1606 		/*
1607 		 * zeroout eof blocks in last cluster starting from
1608 		 * "isize" even "start" > "isize" because it is
1609 		 * complicated to zeroout just at "start" as "start"
1610 		 * may be not aligned with block size, buffer write
1611 		 * would be required to do that, but out of eof buffer
1612 		 * write is not supported.
1613 		 */
1614 		ret = ocfs2_zeroout_partial_cluster(inode, isize,
1615 					end - isize);
1616 		if (ret) {
1617 			mlog_errno(ret);
1618 			goto out;
1619 		}
1620 		if (start >= isize)
1621 			goto out;
1622 		end = isize;
1623 	}
1624 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1625 	if (IS_ERR(handle)) {
1626 		ret = PTR_ERR(handle);
1627 		mlog_errno(ret);
1628 		goto out;
1629 	}
1630 
1631 	/*
1632 	 * If start is on a cluster boundary and end is somewhere in another
1633 	 * cluster, we have not COWed the cluster starting at start, unless
1634 	 * end is also within the same cluster. So, in this case, we skip this
1635 	 * first call to ocfs2_zero_range_for_truncate() truncate and move on
1636 	 * to the next one.
1637 	 */
1638 	if ((start & (csize - 1)) != 0) {
1639 		/*
1640 		 * We want to get the byte offset of the end of the 1st
1641 		 * cluster.
1642 		 */
1643 		tmpend = (u64)osb->s_clustersize +
1644 			(start & ~(osb->s_clustersize - 1));
1645 		if (tmpend > end)
1646 			tmpend = end;
1647 
1648 		trace_ocfs2_zero_partial_clusters_range1(
1649 			(unsigned long long)start,
1650 			(unsigned long long)tmpend);
1651 
1652 		ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1653 						    tmpend);
1654 		if (ret)
1655 			mlog_errno(ret);
1656 	}
1657 
1658 	if (tmpend < end) {
1659 		/*
1660 		 * This may make start and end equal, but the zeroing
1661 		 * code will skip any work in that case so there's no
1662 		 * need to catch it up here.
1663 		 */
1664 		start = end & ~(osb->s_clustersize - 1);
1665 
1666 		trace_ocfs2_zero_partial_clusters_range2(
1667 			(unsigned long long)start, (unsigned long long)end);
1668 
1669 		ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1670 		if (ret)
1671 			mlog_errno(ret);
1672 	}
1673 	ocfs2_update_inode_fsync_trans(handle, inode, 1);
1674 
1675 	ocfs2_commit_trans(osb, handle);
1676 out:
1677 	return ret;
1678 }
1679 
ocfs2_find_rec(struct ocfs2_extent_list * el,u32 pos)1680 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1681 {
1682 	int i;
1683 	struct ocfs2_extent_rec *rec = NULL;
1684 
1685 	for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1686 
1687 		rec = &el->l_recs[i];
1688 
1689 		if (le32_to_cpu(rec->e_cpos) < pos)
1690 			break;
1691 	}
1692 
1693 	return i;
1694 }
1695 
1696 /*
1697  * Helper to calculate the punching pos and length in one run, we handle the
1698  * following three cases in order:
1699  *
1700  * - remove the entire record
1701  * - remove a partial record
1702  * - no record needs to be removed (hole-punching completed)
1703 */
ocfs2_calc_trunc_pos(struct inode * inode,struct ocfs2_extent_list * el,struct ocfs2_extent_rec * rec,u32 trunc_start,u32 * trunc_cpos,u32 * trunc_len,u32 * trunc_end,u64 * blkno,int * done)1704 static void ocfs2_calc_trunc_pos(struct inode *inode,
1705 				 struct ocfs2_extent_list *el,
1706 				 struct ocfs2_extent_rec *rec,
1707 				 u32 trunc_start, u32 *trunc_cpos,
1708 				 u32 *trunc_len, u32 *trunc_end,
1709 				 u64 *blkno, int *done)
1710 {
1711 	int ret = 0;
1712 	u32 coff, range;
1713 
1714 	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1715 
1716 	if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1717 		/*
1718 		 * remove an entire extent record.
1719 		 */
1720 		*trunc_cpos = le32_to_cpu(rec->e_cpos);
1721 		/*
1722 		 * Skip holes if any.
1723 		 */
1724 		if (range < *trunc_end)
1725 			*trunc_end = range;
1726 		*trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1727 		*blkno = le64_to_cpu(rec->e_blkno);
1728 		*trunc_end = le32_to_cpu(rec->e_cpos);
1729 	} else if (range > trunc_start) {
1730 		/*
1731 		 * remove a partial extent record, which means we're
1732 		 * removing the last extent record.
1733 		 */
1734 		*trunc_cpos = trunc_start;
1735 		/*
1736 		 * skip hole if any.
1737 		 */
1738 		if (range < *trunc_end)
1739 			*trunc_end = range;
1740 		*trunc_len = *trunc_end - trunc_start;
1741 		coff = trunc_start - le32_to_cpu(rec->e_cpos);
1742 		*blkno = le64_to_cpu(rec->e_blkno) +
1743 				ocfs2_clusters_to_blocks(inode->i_sb, coff);
1744 		*trunc_end = trunc_start;
1745 	} else {
1746 		/*
1747 		 * It may have two following possibilities:
1748 		 *
1749 		 * - last record has been removed
1750 		 * - trunc_start was within a hole
1751 		 *
1752 		 * both two cases mean the completion of hole punching.
1753 		 */
1754 		ret = 1;
1755 	}
1756 
1757 	*done = ret;
1758 }
1759 
ocfs2_remove_inode_range(struct inode * inode,struct buffer_head * di_bh,u64 byte_start,u64 byte_len)1760 int ocfs2_remove_inode_range(struct inode *inode,
1761 			     struct buffer_head *di_bh, u64 byte_start,
1762 			     u64 byte_len)
1763 {
1764 	int ret = 0, flags = 0, done = 0, i;
1765 	u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1766 	u32 cluster_in_el;
1767 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1768 	struct ocfs2_cached_dealloc_ctxt dealloc;
1769 	struct address_space *mapping = inode->i_mapping;
1770 	struct ocfs2_extent_tree et;
1771 	struct ocfs2_path *path = NULL;
1772 	struct ocfs2_extent_list *el = NULL;
1773 	struct ocfs2_extent_rec *rec = NULL;
1774 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1775 	u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1776 
1777 	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1778 	ocfs2_init_dealloc_ctxt(&dealloc);
1779 
1780 	trace_ocfs2_remove_inode_range(
1781 			(unsigned long long)OCFS2_I(inode)->ip_blkno,
1782 			(unsigned long long)byte_start,
1783 			(unsigned long long)byte_len);
1784 
1785 	if (byte_len == 0)
1786 		return 0;
1787 
1788 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1789 		int id_count = ocfs2_max_inline_data_with_xattr(inode->i_sb, di);
1790 
1791 		if (byte_start > id_count || byte_start + byte_len > id_count) {
1792 			ret = -EINVAL;
1793 			mlog_errno(ret);
1794 			goto out;
1795 		}
1796 
1797 		ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1798 					    byte_start + byte_len, 0);
1799 		if (ret) {
1800 			mlog_errno(ret);
1801 			goto out;
1802 		}
1803 		/*
1804 		 * There's no need to get fancy with the page cache
1805 		 * truncate of an inline-data inode. We're talking
1806 		 * about less than a page here, which will be cached
1807 		 * in the dinode buffer anyway.
1808 		 */
1809 		unmap_mapping_range(mapping, 0, 0, 0);
1810 		truncate_inode_pages(mapping, 0);
1811 		goto out;
1812 	}
1813 
1814 	/*
1815 	 * For reflinks, we may need to CoW 2 clusters which might be
1816 	 * partially zero'd later, if hole's start and end offset were
1817 	 * within one cluster(means is not exactly aligned to clustersize).
1818 	 */
1819 
1820 	if (ocfs2_is_refcount_inode(inode)) {
1821 		ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1822 		if (ret) {
1823 			mlog_errno(ret);
1824 			goto out;
1825 		}
1826 
1827 		ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1828 		if (ret) {
1829 			mlog_errno(ret);
1830 			goto out;
1831 		}
1832 	}
1833 
1834 	trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1835 	trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1836 	cluster_in_el = trunc_end;
1837 
1838 	ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1839 	if (ret) {
1840 		mlog_errno(ret);
1841 		goto out;
1842 	}
1843 
1844 	path = ocfs2_new_path_from_et(&et);
1845 	if (!path) {
1846 		ret = -ENOMEM;
1847 		mlog_errno(ret);
1848 		goto out;
1849 	}
1850 
1851 	while (trunc_end > trunc_start) {
1852 
1853 		ret = ocfs2_find_path(INODE_CACHE(inode), path,
1854 				      cluster_in_el);
1855 		if (ret) {
1856 			mlog_errno(ret);
1857 			goto out;
1858 		}
1859 
1860 		el = path_leaf_el(path);
1861 
1862 		i = ocfs2_find_rec(el, trunc_end);
1863 		/*
1864 		 * Need to go to previous extent block.
1865 		 */
1866 		if (i < 0) {
1867 			if (path->p_tree_depth == 0)
1868 				break;
1869 
1870 			ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1871 							    path,
1872 							    &cluster_in_el);
1873 			if (ret) {
1874 				mlog_errno(ret);
1875 				goto out;
1876 			}
1877 
1878 			/*
1879 			 * We've reached the leftmost extent block,
1880 			 * it's safe to leave.
1881 			 */
1882 			if (cluster_in_el == 0)
1883 				break;
1884 
1885 			/*
1886 			 * The 'pos' searched for previous extent block is
1887 			 * always one cluster less than actual trunc_end.
1888 			 */
1889 			trunc_end = cluster_in_el + 1;
1890 
1891 			ocfs2_reinit_path(path, 1);
1892 
1893 			continue;
1894 
1895 		} else
1896 			rec = &el->l_recs[i];
1897 
1898 		ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1899 				     &trunc_len, &trunc_end, &blkno, &done);
1900 		if (done)
1901 			break;
1902 
1903 		flags = rec->e_flags;
1904 		phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1905 
1906 		ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1907 					       phys_cpos, trunc_len, flags,
1908 					       &dealloc, refcount_loc, false);
1909 		if (ret < 0) {
1910 			mlog_errno(ret);
1911 			goto out;
1912 		}
1913 
1914 		cluster_in_el = trunc_end;
1915 
1916 		ocfs2_reinit_path(path, 1);
1917 	}
1918 
1919 	ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1920 
1921 out:
1922 	ocfs2_free_path(path);
1923 	ocfs2_schedule_truncate_log_flush(osb, 1);
1924 	ocfs2_run_deallocs(osb, &dealloc);
1925 
1926 	return ret;
1927 }
1928 
1929 /*
1930  * Parts of this function taken from xfs_change_file_space()
1931  */
__ocfs2_change_file_space(struct file * file,struct inode * inode,loff_t f_pos,unsigned int cmd,struct ocfs2_space_resv * sr,int change_size)1932 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1933 				     loff_t f_pos, unsigned int cmd,
1934 				     struct ocfs2_space_resv *sr,
1935 				     int change_size)
1936 {
1937 	int ret;
1938 	s64 llen;
1939 	loff_t size, orig_isize;
1940 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1941 	struct buffer_head *di_bh = NULL;
1942 	handle_t *handle;
1943 	unsigned long long max_off = inode->i_sb->s_maxbytes;
1944 
1945 	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1946 		return -EROFS;
1947 
1948 	inode_lock(inode);
1949 
1950 	/* Wait all existing dio workers, newcomers will block on i_rwsem */
1951 	inode_dio_wait(inode);
1952 	/*
1953 	 * This prevents concurrent writes on other nodes
1954 	 */
1955 	ret = ocfs2_rw_lock(inode, 1);
1956 	if (ret) {
1957 		mlog_errno(ret);
1958 		goto out;
1959 	}
1960 
1961 	ret = ocfs2_inode_lock(inode, &di_bh, 1);
1962 	if (ret) {
1963 		mlog_errno(ret);
1964 		goto out_rw_unlock;
1965 	}
1966 
1967 	if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1968 		ret = -EPERM;
1969 		goto out_inode_unlock;
1970 	}
1971 
1972 	switch (sr->l_whence) {
1973 	case 0: /*SEEK_SET*/
1974 		break;
1975 	case 1: /*SEEK_CUR*/
1976 		sr->l_start += f_pos;
1977 		break;
1978 	case 2: /*SEEK_END*/
1979 		sr->l_start += i_size_read(inode);
1980 		break;
1981 	default:
1982 		ret = -EINVAL;
1983 		goto out_inode_unlock;
1984 	}
1985 	sr->l_whence = 0;
1986 
1987 	llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1988 
1989 	if (sr->l_start < 0
1990 	    || sr->l_start > max_off
1991 	    || (sr->l_start + llen) < 0
1992 	    || (sr->l_start + llen) > max_off) {
1993 		ret = -EINVAL;
1994 		goto out_inode_unlock;
1995 	}
1996 	size = sr->l_start + sr->l_len;
1997 
1998 	if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1999 	    cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
2000 		if (sr->l_len <= 0) {
2001 			ret = -EINVAL;
2002 			goto out_inode_unlock;
2003 		}
2004 	}
2005 
2006 	if (file && setattr_should_drop_suidgid(&nop_mnt_idmap, file_inode(file))) {
2007 		ret = __ocfs2_write_remove_suid(inode, di_bh);
2008 		if (ret) {
2009 			mlog_errno(ret);
2010 			goto out_inode_unlock;
2011 		}
2012 	}
2013 
2014 	down_write(&OCFS2_I(inode)->ip_alloc_sem);
2015 	switch (cmd) {
2016 	case OCFS2_IOC_RESVSP:
2017 	case OCFS2_IOC_RESVSP64:
2018 		/*
2019 		 * This takes unsigned offsets, but the signed ones we
2020 		 * pass have been checked against overflow above.
2021 		 */
2022 		ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
2023 						       sr->l_len);
2024 		break;
2025 	case OCFS2_IOC_UNRESVSP:
2026 	case OCFS2_IOC_UNRESVSP64:
2027 		ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
2028 					       sr->l_len);
2029 		break;
2030 	default:
2031 		ret = -EINVAL;
2032 	}
2033 
2034 	orig_isize = i_size_read(inode);
2035 	/* zeroout eof blocks in the cluster. */
2036 	if (!ret && change_size && orig_isize < size) {
2037 		ret = ocfs2_zeroout_partial_cluster(inode, orig_isize,
2038 					size - orig_isize);
2039 		if (!ret)
2040 			i_size_write(inode, size);
2041 	}
2042 	up_write(&OCFS2_I(inode)->ip_alloc_sem);
2043 	if (ret) {
2044 		mlog_errno(ret);
2045 		goto out_inode_unlock;
2046 	}
2047 
2048 	/*
2049 	 * We update c/mtime for these changes
2050 	 */
2051 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2052 	if (IS_ERR(handle)) {
2053 		ret = PTR_ERR(handle);
2054 		mlog_errno(ret);
2055 		goto out_inode_unlock;
2056 	}
2057 
2058 	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
2059 	ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2060 	if (ret < 0)
2061 		mlog_errno(ret);
2062 
2063 	if (file && (file->f_flags & O_SYNC))
2064 		handle->h_sync = 1;
2065 
2066 	ocfs2_commit_trans(osb, handle);
2067 
2068 out_inode_unlock:
2069 	brelse(di_bh);
2070 	ocfs2_inode_unlock(inode, 1);
2071 out_rw_unlock:
2072 	ocfs2_rw_unlock(inode, 1);
2073 
2074 out:
2075 	inode_unlock(inode);
2076 	return ret;
2077 }
2078 
ocfs2_change_file_space(struct file * file,unsigned int cmd,struct ocfs2_space_resv * sr)2079 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2080 			    struct ocfs2_space_resv *sr)
2081 {
2082 	struct inode *inode = file_inode(file);
2083 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2084 	int ret;
2085 
2086 	if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2087 	    !ocfs2_writes_unwritten_extents(osb))
2088 		return -ENOTTY;
2089 	else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2090 		 !ocfs2_sparse_alloc(osb))
2091 		return -ENOTTY;
2092 
2093 	if (!S_ISREG(inode->i_mode))
2094 		return -EINVAL;
2095 
2096 	if (!(file->f_mode & FMODE_WRITE))
2097 		return -EBADF;
2098 
2099 	ret = mnt_want_write_file(file);
2100 	if (ret)
2101 		return ret;
2102 	ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2103 	mnt_drop_write_file(file);
2104 	return ret;
2105 }
2106 
ocfs2_fallocate(struct file * file,int mode,loff_t offset,loff_t len)2107 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2108 			    loff_t len)
2109 {
2110 	struct inode *inode = file_inode(file);
2111 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2112 	struct ocfs2_space_resv sr;
2113 	int change_size = 1;
2114 	int cmd = OCFS2_IOC_RESVSP64;
2115 	int ret = 0;
2116 
2117 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2118 		return -EOPNOTSUPP;
2119 	if (!ocfs2_writes_unwritten_extents(osb))
2120 		return -EOPNOTSUPP;
2121 
2122 	if (mode & FALLOC_FL_KEEP_SIZE) {
2123 		change_size = 0;
2124 	} else {
2125 		ret = inode_newsize_ok(inode, offset + len);
2126 		if (ret)
2127 			return ret;
2128 	}
2129 
2130 	if (mode & FALLOC_FL_PUNCH_HOLE)
2131 		cmd = OCFS2_IOC_UNRESVSP64;
2132 
2133 	sr.l_whence = 0;
2134 	sr.l_start = (s64)offset;
2135 	sr.l_len = (s64)len;
2136 
2137 	return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2138 					 change_size);
2139 }
2140 
ocfs2_check_range_for_refcount(struct inode * inode,loff_t pos,size_t count)2141 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2142 				   size_t count)
2143 {
2144 	int ret = 0;
2145 	unsigned int extent_flags;
2146 	u32 cpos, clusters, extent_len, phys_cpos;
2147 	struct super_block *sb = inode->i_sb;
2148 
2149 	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2150 	    !ocfs2_is_refcount_inode(inode) ||
2151 	    OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2152 		return 0;
2153 
2154 	cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2155 	clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2156 
2157 	while (clusters) {
2158 		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2159 					 &extent_flags);
2160 		if (ret < 0) {
2161 			mlog_errno(ret);
2162 			goto out;
2163 		}
2164 
2165 		if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2166 			ret = 1;
2167 			break;
2168 		}
2169 
2170 		if (extent_len > clusters)
2171 			extent_len = clusters;
2172 
2173 		clusters -= extent_len;
2174 		cpos += extent_len;
2175 	}
2176 out:
2177 	return ret;
2178 }
2179 
ocfs2_is_io_unaligned(struct inode * inode,size_t count,loff_t pos)2180 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2181 {
2182 	int blockmask = inode->i_sb->s_blocksize - 1;
2183 	loff_t final_size = pos + count;
2184 
2185 	if ((pos & blockmask) || (final_size & blockmask))
2186 		return 1;
2187 	return 0;
2188 }
2189 
ocfs2_inode_lock_for_extent_tree(struct inode * inode,struct buffer_head ** di_bh,int meta_level,int write_sem,int wait)2190 static int ocfs2_inode_lock_for_extent_tree(struct inode *inode,
2191 					    struct buffer_head **di_bh,
2192 					    int meta_level,
2193 					    int write_sem,
2194 					    int wait)
2195 {
2196 	int ret = 0;
2197 
2198 	if (wait)
2199 		ret = ocfs2_inode_lock(inode, di_bh, meta_level);
2200 	else
2201 		ret = ocfs2_try_inode_lock(inode, di_bh, meta_level);
2202 	if (ret < 0)
2203 		goto out;
2204 
2205 	if (wait) {
2206 		if (write_sem)
2207 			down_write(&OCFS2_I(inode)->ip_alloc_sem);
2208 		else
2209 			down_read(&OCFS2_I(inode)->ip_alloc_sem);
2210 	} else {
2211 		if (write_sem)
2212 			ret = down_write_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2213 		else
2214 			ret = down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem);
2215 
2216 		if (!ret) {
2217 			ret = -EAGAIN;
2218 			goto out_unlock;
2219 		}
2220 	}
2221 
2222 	return ret;
2223 
2224 out_unlock:
2225 	brelse(*di_bh);
2226 	*di_bh = NULL;
2227 	ocfs2_inode_unlock(inode, meta_level);
2228 out:
2229 	return ret;
2230 }
2231 
ocfs2_inode_unlock_for_extent_tree(struct inode * inode,struct buffer_head ** di_bh,int meta_level,int write_sem)2232 static void ocfs2_inode_unlock_for_extent_tree(struct inode *inode,
2233 					       struct buffer_head **di_bh,
2234 					       int meta_level,
2235 					       int write_sem)
2236 {
2237 	if (write_sem)
2238 		up_write(&OCFS2_I(inode)->ip_alloc_sem);
2239 	else
2240 		up_read(&OCFS2_I(inode)->ip_alloc_sem);
2241 
2242 	brelse(*di_bh);
2243 	*di_bh = NULL;
2244 
2245 	if (meta_level >= 0)
2246 		ocfs2_inode_unlock(inode, meta_level);
2247 }
2248 
ocfs2_prepare_inode_for_write(struct file * file,loff_t pos,size_t count,int wait)2249 static int ocfs2_prepare_inode_for_write(struct file *file,
2250 					 loff_t pos, size_t count, int wait)
2251 {
2252 	int ret = 0, meta_level = 0, overwrite_io = 0;
2253 	int write_sem = 0;
2254 	struct dentry *dentry = file->f_path.dentry;
2255 	struct inode *inode = d_inode(dentry);
2256 	struct buffer_head *di_bh = NULL;
2257 	u32 cpos;
2258 	u32 clusters;
2259 
2260 	/*
2261 	 * We start with a read level meta lock and only jump to an ex
2262 	 * if we need to make modifications here.
2263 	 */
2264 	for(;;) {
2265 		ret = ocfs2_inode_lock_for_extent_tree(inode,
2266 						       &di_bh,
2267 						       meta_level,
2268 						       write_sem,
2269 						       wait);
2270 		if (ret < 0) {
2271 			if (ret != -EAGAIN)
2272 				mlog_errno(ret);
2273 			goto out;
2274 		}
2275 
2276 		/*
2277 		 * Check if IO will overwrite allocated blocks in case
2278 		 * IOCB_NOWAIT flag is set.
2279 		 */
2280 		if (!wait && !overwrite_io) {
2281 			overwrite_io = 1;
2282 
2283 			ret = ocfs2_overwrite_io(inode, di_bh, pos, count);
2284 			if (ret < 0) {
2285 				if (ret != -EAGAIN)
2286 					mlog_errno(ret);
2287 				goto out_unlock;
2288 			}
2289 		}
2290 
2291 		/* Clear suid / sgid if necessary. We do this here
2292 		 * instead of later in the write path because
2293 		 * remove_suid() calls ->setattr without any hint that
2294 		 * we may have already done our cluster locking. Since
2295 		 * ocfs2_setattr() *must* take cluster locks to
2296 		 * proceed, this will lead us to recursively lock the
2297 		 * inode. There's also the dinode i_size state which
2298 		 * can be lost via setattr during extending writes (we
2299 		 * set inode->i_size at the end of a write. */
2300 		if (setattr_should_drop_suidgid(&nop_mnt_idmap, inode)) {
2301 			if (meta_level == 0) {
2302 				ocfs2_inode_unlock_for_extent_tree(inode,
2303 								   &di_bh,
2304 								   meta_level,
2305 								   write_sem);
2306 				meta_level = 1;
2307 				continue;
2308 			}
2309 
2310 			ret = ocfs2_write_remove_suid(inode);
2311 			if (ret < 0) {
2312 				mlog_errno(ret);
2313 				goto out_unlock;
2314 			}
2315 		}
2316 
2317 		ret = ocfs2_check_range_for_refcount(inode, pos, count);
2318 		if (ret == 1) {
2319 			ocfs2_inode_unlock_for_extent_tree(inode,
2320 							   &di_bh,
2321 							   meta_level,
2322 							   write_sem);
2323 			meta_level = 1;
2324 			write_sem = 1;
2325 			ret = ocfs2_inode_lock_for_extent_tree(inode,
2326 							       &di_bh,
2327 							       meta_level,
2328 							       write_sem,
2329 							       wait);
2330 			if (ret < 0) {
2331 				if (ret != -EAGAIN)
2332 					mlog_errno(ret);
2333 				goto out;
2334 			}
2335 
2336 			cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2337 			clusters =
2338 				ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2339 			ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2340 		}
2341 
2342 		if (ret < 0) {
2343 			if (ret != -EAGAIN)
2344 				mlog_errno(ret);
2345 			goto out_unlock;
2346 		}
2347 
2348 		break;
2349 	}
2350 
2351 out_unlock:
2352 	trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2353 					    pos, count, wait);
2354 
2355 	ocfs2_inode_unlock_for_extent_tree(inode,
2356 					   &di_bh,
2357 					   meta_level,
2358 					   write_sem);
2359 
2360 out:
2361 	return ret;
2362 }
2363 
ocfs2_file_write_iter(struct kiocb * iocb,struct iov_iter * from)2364 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2365 				    struct iov_iter *from)
2366 {
2367 	int rw_level;
2368 	ssize_t written = 0;
2369 	ssize_t ret;
2370 	size_t count = iov_iter_count(from);
2371 	struct file *file = iocb->ki_filp;
2372 	struct inode *inode = file_inode(file);
2373 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2374 	int full_coherency = !(osb->s_mount_opt &
2375 			       OCFS2_MOUNT_COHERENCY_BUFFERED);
2376 	void *saved_ki_complete = NULL;
2377 	int append_write = ((iocb->ki_pos + count) >=
2378 			i_size_read(inode) ? 1 : 0);
2379 	int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2380 	int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2381 
2382 	trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry,
2383 		(unsigned long long)OCFS2_I(inode)->ip_blkno,
2384 		file->f_path.dentry->d_name.len,
2385 		file->f_path.dentry->d_name.name,
2386 		(unsigned int)from->nr_segs);	/* GRRRRR */
2387 
2388 	if (!direct_io && nowait)
2389 		return -EOPNOTSUPP;
2390 
2391 	if (count == 0)
2392 		return 0;
2393 
2394 	if (nowait) {
2395 		if (!inode_trylock(inode))
2396 			return -EAGAIN;
2397 	} else
2398 		inode_lock(inode);
2399 
2400 	ocfs2_iocb_init_rw_locked(iocb);
2401 
2402 	/*
2403 	 * Concurrent O_DIRECT writes are allowed with
2404 	 * mount_option "coherency=buffered".
2405 	 * For append write, we must take rw EX.
2406 	 */
2407 	rw_level = (!direct_io || full_coherency || append_write);
2408 
2409 	if (nowait)
2410 		ret = ocfs2_try_rw_lock(inode, rw_level);
2411 	else
2412 		ret = ocfs2_rw_lock(inode, rw_level);
2413 	if (ret < 0) {
2414 		if (ret != -EAGAIN)
2415 			mlog_errno(ret);
2416 		goto out_mutex;
2417 	}
2418 
2419 	/*
2420 	 * O_DIRECT writes with "coherency=full" need to take EX cluster
2421 	 * inode_lock to guarantee coherency.
2422 	 */
2423 	if (direct_io && full_coherency) {
2424 		/*
2425 		 * We need to take and drop the inode lock to force
2426 		 * other nodes to drop their caches.  Buffered I/O
2427 		 * already does this in write_begin().
2428 		 */
2429 		if (nowait)
2430 			ret = ocfs2_try_inode_lock(inode, NULL, 1);
2431 		else
2432 			ret = ocfs2_inode_lock(inode, NULL, 1);
2433 		if (ret < 0) {
2434 			if (ret != -EAGAIN)
2435 				mlog_errno(ret);
2436 			goto out;
2437 		}
2438 
2439 		ocfs2_inode_unlock(inode, 1);
2440 	}
2441 
2442 	ret = generic_write_checks(iocb, from);
2443 	if (ret <= 0) {
2444 		if (ret)
2445 			mlog_errno(ret);
2446 		goto out;
2447 	}
2448 	count = ret;
2449 
2450 	ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait);
2451 	if (ret < 0) {
2452 		if (ret != -EAGAIN)
2453 			mlog_errno(ret);
2454 		goto out;
2455 	}
2456 
2457 	if (direct_io && !is_sync_kiocb(iocb) &&
2458 	    ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2459 		/*
2460 		 * Make it a sync io if it's an unaligned aio.
2461 		 */
2462 		saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2463 	}
2464 
2465 	/* communicate with ocfs2_dio_end_io */
2466 	ocfs2_iocb_set_rw_locked(iocb, rw_level);
2467 
2468 	written = __generic_file_write_iter(iocb, from);
2469 	/* buffered aio wouldn't have proper lock coverage today */
2470 	BUG_ON(written == -EIOCBQUEUED && !direct_io);
2471 
2472 	/*
2473 	 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2474 	 * function pointer which is called when o_direct io completes so that
2475 	 * it can unlock our rw lock.
2476 	 * Unfortunately there are error cases which call end_io and others
2477 	 * that don't.  so we don't have to unlock the rw_lock if either an
2478 	 * async dio is going to do it in the future or an end_io after an
2479 	 * error has already done it.
2480 	 */
2481 	if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2482 		rw_level = -1;
2483 	}
2484 
2485 	if (unlikely(written <= 0))
2486 		goto out;
2487 
2488 	if (((file->f_flags & O_DSYNC) && !direct_io) ||
2489 	    IS_SYNC(inode)) {
2490 		ret = filemap_fdatawrite_range(file->f_mapping,
2491 					       iocb->ki_pos - written,
2492 					       iocb->ki_pos - 1);
2493 		if (ret < 0)
2494 			written = ret;
2495 
2496 		if (!ret) {
2497 			ret = jbd2_journal_force_commit(osb->journal->j_journal);
2498 			if (ret < 0)
2499 				written = ret;
2500 		}
2501 
2502 		if (!ret)
2503 			ret = filemap_fdatawait_range(file->f_mapping,
2504 						      iocb->ki_pos - written,
2505 						      iocb->ki_pos - 1);
2506 	}
2507 
2508 out:
2509 	if (saved_ki_complete)
2510 		xchg(&iocb->ki_complete, saved_ki_complete);
2511 
2512 	if (rw_level != -1)
2513 		ocfs2_rw_unlock(inode, rw_level);
2514 
2515 out_mutex:
2516 	inode_unlock(inode);
2517 
2518 	if (written)
2519 		ret = written;
2520 	return ret;
2521 }
2522 
ocfs2_file_read_iter(struct kiocb * iocb,struct iov_iter * to)2523 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2524 				   struct iov_iter *to)
2525 {
2526 	int ret = 0, rw_level = -1, lock_level = 0;
2527 	struct file *filp = iocb->ki_filp;
2528 	struct inode *inode = file_inode(filp);
2529 	int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2530 	int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0;
2531 
2532 	trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry,
2533 			(unsigned long long)OCFS2_I(inode)->ip_blkno,
2534 			filp->f_path.dentry->d_name.len,
2535 			filp->f_path.dentry->d_name.name,
2536 			to->nr_segs);	/* GRRRRR */
2537 
2538 
2539 	if (!inode) {
2540 		ret = -EINVAL;
2541 		mlog_errno(ret);
2542 		goto bail;
2543 	}
2544 
2545 	if (!direct_io && nowait)
2546 		return -EOPNOTSUPP;
2547 
2548 	ocfs2_iocb_init_rw_locked(iocb);
2549 
2550 	/*
2551 	 * buffered reads protect themselves in ->read_folio().  O_DIRECT reads
2552 	 * need locks to protect pending reads from racing with truncate.
2553 	 */
2554 	if (direct_io) {
2555 		if (nowait)
2556 			ret = ocfs2_try_rw_lock(inode, 0);
2557 		else
2558 			ret = ocfs2_rw_lock(inode, 0);
2559 
2560 		if (ret < 0) {
2561 			if (ret != -EAGAIN)
2562 				mlog_errno(ret);
2563 			goto bail;
2564 		}
2565 		rw_level = 0;
2566 		/* communicate with ocfs2_dio_end_io */
2567 		ocfs2_iocb_set_rw_locked(iocb, rw_level);
2568 	}
2569 
2570 	/*
2571 	 * We're fine letting folks race truncates and extending
2572 	 * writes with read across the cluster, just like they can
2573 	 * locally. Hence no rw_lock during read.
2574 	 *
2575 	 * Take and drop the meta data lock to update inode fields
2576 	 * like i_size. This allows the checks down below
2577 	 * copy_splice_read() a chance of actually working.
2578 	 */
2579 	ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level,
2580 				     !nowait);
2581 	if (ret < 0) {
2582 		if (ret != -EAGAIN)
2583 			mlog_errno(ret);
2584 		goto bail;
2585 	}
2586 	ocfs2_inode_unlock(inode, lock_level);
2587 
2588 	ret = generic_file_read_iter(iocb, to);
2589 	trace_generic_file_read_iter_ret(ret);
2590 
2591 	/* buffered aio wouldn't have proper lock coverage today */
2592 	BUG_ON(ret == -EIOCBQUEUED && !direct_io);
2593 
2594 	/* see ocfs2_file_write_iter */
2595 	if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2596 		rw_level = -1;
2597 	}
2598 
2599 bail:
2600 	if (rw_level != -1)
2601 		ocfs2_rw_unlock(inode, rw_level);
2602 
2603 	return ret;
2604 }
2605 
ocfs2_file_splice_read(struct file * in,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)2606 static ssize_t ocfs2_file_splice_read(struct file *in, loff_t *ppos,
2607 				      struct pipe_inode_info *pipe,
2608 				      size_t len, unsigned int flags)
2609 {
2610 	struct inode *inode = file_inode(in);
2611 	ssize_t ret = 0;
2612 	int lock_level = 0;
2613 
2614 	trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2615 				     (unsigned long long)OCFS2_I(inode)->ip_blkno,
2616 				     in->f_path.dentry->d_name.len,
2617 				     in->f_path.dentry->d_name.name,
2618 				     flags);
2619 
2620 	/*
2621 	 * We're fine letting folks race truncates and extending writes with
2622 	 * read across the cluster, just like they can locally.  Hence no
2623 	 * rw_lock during read.
2624 	 *
2625 	 * Take and drop the meta data lock to update inode fields like i_size.
2626 	 * This allows the checks down below filemap_splice_read() a chance of
2627 	 * actually working.
2628 	 */
2629 	ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level, 1);
2630 	if (ret < 0) {
2631 		if (ret != -EAGAIN)
2632 			mlog_errno(ret);
2633 		goto bail;
2634 	}
2635 	ocfs2_inode_unlock(inode, lock_level);
2636 
2637 	ret = filemap_splice_read(in, ppos, pipe, len, flags);
2638 	trace_filemap_splice_read_ret(ret);
2639 bail:
2640 	return ret;
2641 }
2642 
2643 /* Refer generic_file_llseek_unlocked() */
ocfs2_file_llseek(struct file * file,loff_t offset,int whence)2644 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2645 {
2646 	struct inode *inode = file->f_mapping->host;
2647 	int ret = 0;
2648 
2649 	inode_lock(inode);
2650 
2651 	switch (whence) {
2652 	case SEEK_SET:
2653 		break;
2654 	case SEEK_END:
2655 		/* SEEK_END requires the OCFS2 inode lock for the file
2656 		 * because it references the file's size.
2657 		 */
2658 		ret = ocfs2_inode_lock(inode, NULL, 0);
2659 		if (ret < 0) {
2660 			mlog_errno(ret);
2661 			goto out;
2662 		}
2663 		offset += i_size_read(inode);
2664 		ocfs2_inode_unlock(inode, 0);
2665 		break;
2666 	case SEEK_CUR:
2667 		if (offset == 0) {
2668 			offset = file->f_pos;
2669 			goto out;
2670 		}
2671 		offset += file->f_pos;
2672 		break;
2673 	case SEEK_DATA:
2674 	case SEEK_HOLE:
2675 		ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2676 		if (ret)
2677 			goto out;
2678 		break;
2679 	default:
2680 		ret = -EINVAL;
2681 		goto out;
2682 	}
2683 
2684 	offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2685 
2686 out:
2687 	inode_unlock(inode);
2688 	if (ret)
2689 		return ret;
2690 	return offset;
2691 }
2692 
ocfs2_remap_file_range(struct file * file_in,loff_t pos_in,struct file * file_out,loff_t pos_out,loff_t len,unsigned int remap_flags)2693 static loff_t ocfs2_remap_file_range(struct file *file_in, loff_t pos_in,
2694 				     struct file *file_out, loff_t pos_out,
2695 				     loff_t len, unsigned int remap_flags)
2696 {
2697 	struct inode *inode_in = file_inode(file_in);
2698 	struct inode *inode_out = file_inode(file_out);
2699 	struct ocfs2_super *osb = OCFS2_SB(inode_in->i_sb);
2700 	struct buffer_head *in_bh = NULL, *out_bh = NULL;
2701 	bool same_inode = (inode_in == inode_out);
2702 	loff_t remapped = 0;
2703 	ssize_t ret;
2704 
2705 	if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
2706 		return -EINVAL;
2707 	if (!ocfs2_refcount_tree(osb))
2708 		return -EOPNOTSUPP;
2709 	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
2710 		return -EROFS;
2711 
2712 	/* Lock both files against IO */
2713 	ret = ocfs2_reflink_inodes_lock(inode_in, &in_bh, inode_out, &out_bh);
2714 	if (ret)
2715 		return ret;
2716 
2717 	/* Check file eligibility and prepare for block sharing. */
2718 	ret = -EINVAL;
2719 	if ((OCFS2_I(inode_in)->ip_flags & OCFS2_INODE_SYSTEM_FILE) ||
2720 	    (OCFS2_I(inode_out)->ip_flags & OCFS2_INODE_SYSTEM_FILE))
2721 		goto out_unlock;
2722 
2723 	ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
2724 			&len, remap_flags);
2725 	if (ret < 0 || len == 0)
2726 		goto out_unlock;
2727 
2728 	/* Lock out changes to the allocation maps and remap. */
2729 	down_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2730 	if (!same_inode)
2731 		down_write_nested(&OCFS2_I(inode_out)->ip_alloc_sem,
2732 				  SINGLE_DEPTH_NESTING);
2733 
2734 	/* Zap any page cache for the destination file's range. */
2735 	truncate_inode_pages_range(&inode_out->i_data,
2736 				   round_down(pos_out, PAGE_SIZE),
2737 				   round_up(pos_out + len, PAGE_SIZE) - 1);
2738 
2739 	remapped = ocfs2_reflink_remap_blocks(inode_in, in_bh, pos_in,
2740 			inode_out, out_bh, pos_out, len);
2741 	up_write(&OCFS2_I(inode_in)->ip_alloc_sem);
2742 	if (!same_inode)
2743 		up_write(&OCFS2_I(inode_out)->ip_alloc_sem);
2744 	if (remapped < 0) {
2745 		ret = remapped;
2746 		mlog_errno(ret);
2747 		goto out_unlock;
2748 	}
2749 
2750 	/*
2751 	 * Empty the extent map so that we may get the right extent
2752 	 * record from the disk.
2753 	 */
2754 	ocfs2_extent_map_trunc(inode_in, 0);
2755 	ocfs2_extent_map_trunc(inode_out, 0);
2756 
2757 	ret = ocfs2_reflink_update_dest(inode_out, out_bh, pos_out + len);
2758 	if (ret) {
2759 		mlog_errno(ret);
2760 		goto out_unlock;
2761 	}
2762 
2763 out_unlock:
2764 	ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh);
2765 	return remapped > 0 ? remapped : ret;
2766 }
2767 
2768 const struct inode_operations ocfs2_file_iops = {
2769 	.setattr	= ocfs2_setattr,
2770 	.getattr	= ocfs2_getattr,
2771 	.permission	= ocfs2_permission,
2772 	.listxattr	= ocfs2_listxattr,
2773 	.fiemap		= ocfs2_fiemap,
2774 	.get_inode_acl	= ocfs2_iop_get_acl,
2775 	.set_acl	= ocfs2_iop_set_acl,
2776 	.fileattr_get	= ocfs2_fileattr_get,
2777 	.fileattr_set	= ocfs2_fileattr_set,
2778 };
2779 
2780 const struct inode_operations ocfs2_special_file_iops = {
2781 	.setattr	= ocfs2_setattr,
2782 	.getattr	= ocfs2_getattr,
2783 	.permission	= ocfs2_permission,
2784 	.get_inode_acl	= ocfs2_iop_get_acl,
2785 	.set_acl	= ocfs2_iop_set_acl,
2786 };
2787 
2788 /*
2789  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2790  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2791  */
2792 const struct file_operations ocfs2_fops = {
2793 	.llseek		= ocfs2_file_llseek,
2794 	.mmap		= ocfs2_mmap,
2795 	.fsync		= ocfs2_sync_file,
2796 	.release	= ocfs2_file_release,
2797 	.open		= ocfs2_file_open,
2798 	.read_iter	= ocfs2_file_read_iter,
2799 	.write_iter	= ocfs2_file_write_iter,
2800 	.unlocked_ioctl	= ocfs2_ioctl,
2801 #ifdef CONFIG_COMPAT
2802 	.compat_ioctl   = ocfs2_compat_ioctl,
2803 #endif
2804 	.lock		= ocfs2_lock,
2805 	.flock		= ocfs2_flock,
2806 	.splice_read	= ocfs2_file_splice_read,
2807 	.splice_write	= iter_file_splice_write,
2808 	.fallocate	= ocfs2_fallocate,
2809 	.remap_file_range = ocfs2_remap_file_range,
2810 };
2811 
2812 WRAP_DIR_ITER(ocfs2_readdir) // FIXME!
2813 const struct file_operations ocfs2_dops = {
2814 	.llseek		= generic_file_llseek,
2815 	.read		= generic_read_dir,
2816 	.iterate_shared	= shared_ocfs2_readdir,
2817 	.fsync		= ocfs2_sync_file,
2818 	.release	= ocfs2_dir_release,
2819 	.open		= ocfs2_dir_open,
2820 	.unlocked_ioctl	= ocfs2_ioctl,
2821 #ifdef CONFIG_COMPAT
2822 	.compat_ioctl   = ocfs2_compat_ioctl,
2823 #endif
2824 	.lock		= ocfs2_lock,
2825 	.flock		= ocfs2_flock,
2826 };
2827 
2828 /*
2829  * POSIX-lockless variants of our file_operations.
2830  *
2831  * These will be used if the underlying cluster stack does not support
2832  * posix file locking, if the user passes the "localflocks" mount
2833  * option, or if we have a local-only fs.
2834  *
2835  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2836  * so we still want it in the case of no stack support for
2837  * plocks. Internally, it will do the right thing when asked to ignore
2838  * the cluster.
2839  */
2840 const struct file_operations ocfs2_fops_no_plocks = {
2841 	.llseek		= ocfs2_file_llseek,
2842 	.mmap		= ocfs2_mmap,
2843 	.fsync		= ocfs2_sync_file,
2844 	.release	= ocfs2_file_release,
2845 	.open		= ocfs2_file_open,
2846 	.read_iter	= ocfs2_file_read_iter,
2847 	.write_iter	= ocfs2_file_write_iter,
2848 	.unlocked_ioctl	= ocfs2_ioctl,
2849 #ifdef CONFIG_COMPAT
2850 	.compat_ioctl   = ocfs2_compat_ioctl,
2851 #endif
2852 	.flock		= ocfs2_flock,
2853 	.splice_read	= filemap_splice_read,
2854 	.splice_write	= iter_file_splice_write,
2855 	.fallocate	= ocfs2_fallocate,
2856 	.remap_file_range = ocfs2_remap_file_range,
2857 };
2858 
2859 const struct file_operations ocfs2_dops_no_plocks = {
2860 	.llseek		= generic_file_llseek,
2861 	.read		= generic_read_dir,
2862 	.iterate_shared	= shared_ocfs2_readdir,
2863 	.fsync		= ocfs2_sync_file,
2864 	.release	= ocfs2_dir_release,
2865 	.open		= ocfs2_dir_open,
2866 	.unlocked_ioctl	= ocfs2_ioctl,
2867 #ifdef CONFIG_COMPAT
2868 	.compat_ioctl   = ocfs2_compat_ioctl,
2869 #endif
2870 	.flock		= ocfs2_flock,
2871 };
2872