xref: /openbmc/linux/fs/f2fs/file.c (revision c819e2cf)
1 /*
2  * fs/f2fs/file.c
3  *
4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5  *             http://www.samsung.com/
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/stat.h>
14 #include <linux/buffer_head.h>
15 #include <linux/writeback.h>
16 #include <linux/blkdev.h>
17 #include <linux/falloc.h>
18 #include <linux/types.h>
19 #include <linux/compat.h>
20 #include <linux/uaccess.h>
21 #include <linux/mount.h>
22 #include <linux/pagevec.h>
23 
24 #include "f2fs.h"
25 #include "node.h"
26 #include "segment.h"
27 #include "xattr.h"
28 #include "acl.h"
29 #include <trace/events/f2fs.h>
30 
31 static int f2fs_vm_page_mkwrite(struct vm_area_struct *vma,
32 						struct vm_fault *vmf)
33 {
34 	struct page *page = vmf->page;
35 	struct inode *inode = file_inode(vma->vm_file);
36 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
37 	struct dnode_of_data dn;
38 	int err;
39 
40 	f2fs_balance_fs(sbi);
41 
42 	sb_start_pagefault(inode->i_sb);
43 
44 	f2fs_bug_on(sbi, f2fs_has_inline_data(inode));
45 
46 	/* block allocation */
47 	f2fs_lock_op(sbi);
48 	set_new_dnode(&dn, inode, NULL, NULL, 0);
49 	err = f2fs_reserve_block(&dn, page->index);
50 	if (err) {
51 		f2fs_unlock_op(sbi);
52 		goto out;
53 	}
54 	f2fs_put_dnode(&dn);
55 	f2fs_unlock_op(sbi);
56 
57 	file_update_time(vma->vm_file);
58 	lock_page(page);
59 	if (unlikely(page->mapping != inode->i_mapping ||
60 			page_offset(page) > i_size_read(inode) ||
61 			!PageUptodate(page))) {
62 		unlock_page(page);
63 		err = -EFAULT;
64 		goto out;
65 	}
66 
67 	/*
68 	 * check to see if the page is mapped already (no holes)
69 	 */
70 	if (PageMappedToDisk(page))
71 		goto mapped;
72 
73 	/* page is wholly or partially inside EOF */
74 	if (((page->index + 1) << PAGE_CACHE_SHIFT) > i_size_read(inode)) {
75 		unsigned offset;
76 		offset = i_size_read(inode) & ~PAGE_CACHE_MASK;
77 		zero_user_segment(page, offset, PAGE_CACHE_SIZE);
78 	}
79 	set_page_dirty(page);
80 	SetPageUptodate(page);
81 
82 	trace_f2fs_vm_page_mkwrite(page, DATA);
83 mapped:
84 	/* fill the page */
85 	f2fs_wait_on_page_writeback(page, DATA);
86 out:
87 	sb_end_pagefault(inode->i_sb);
88 	return block_page_mkwrite_return(err);
89 }
90 
91 static const struct vm_operations_struct f2fs_file_vm_ops = {
92 	.fault		= filemap_fault,
93 	.map_pages	= filemap_map_pages,
94 	.page_mkwrite	= f2fs_vm_page_mkwrite,
95 	.remap_pages	= generic_file_remap_pages,
96 };
97 
98 static int get_parent_ino(struct inode *inode, nid_t *pino)
99 {
100 	struct dentry *dentry;
101 
102 	inode = igrab(inode);
103 	dentry = d_find_any_alias(inode);
104 	iput(inode);
105 	if (!dentry)
106 		return 0;
107 
108 	if (update_dent_inode(inode, &dentry->d_name)) {
109 		dput(dentry);
110 		return 0;
111 	}
112 
113 	*pino = parent_ino(dentry);
114 	dput(dentry);
115 	return 1;
116 }
117 
118 static inline bool need_do_checkpoint(struct inode *inode)
119 {
120 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
121 	bool need_cp = false;
122 
123 	if (!S_ISREG(inode->i_mode) || inode->i_nlink != 1)
124 		need_cp = true;
125 	else if (file_wrong_pino(inode))
126 		need_cp = true;
127 	else if (!space_for_roll_forward(sbi))
128 		need_cp = true;
129 	else if (!is_checkpointed_node(sbi, F2FS_I(inode)->i_pino))
130 		need_cp = true;
131 	else if (F2FS_I(inode)->xattr_ver == cur_cp_version(F2FS_CKPT(sbi)))
132 		need_cp = true;
133 	else if (test_opt(sbi, FASTBOOT))
134 		need_cp = true;
135 	else if (sbi->active_logs == 2)
136 		need_cp = true;
137 
138 	return need_cp;
139 }
140 
141 static bool need_inode_page_update(struct f2fs_sb_info *sbi, nid_t ino)
142 {
143 	struct page *i = find_get_page(NODE_MAPPING(sbi), ino);
144 	bool ret = false;
145 	/* But we need to avoid that there are some inode updates */
146 	if ((i && PageDirty(i)) || need_inode_block_update(sbi, ino))
147 		ret = true;
148 	f2fs_put_page(i, 0);
149 	return ret;
150 }
151 
152 static void try_to_fix_pino(struct inode *inode)
153 {
154 	struct f2fs_inode_info *fi = F2FS_I(inode);
155 	nid_t pino;
156 
157 	down_write(&fi->i_sem);
158 	fi->xattr_ver = 0;
159 	if (file_wrong_pino(inode) && inode->i_nlink == 1 &&
160 			get_parent_ino(inode, &pino)) {
161 		fi->i_pino = pino;
162 		file_got_pino(inode);
163 		up_write(&fi->i_sem);
164 
165 		mark_inode_dirty_sync(inode);
166 		f2fs_write_inode(inode, NULL);
167 	} else {
168 		up_write(&fi->i_sem);
169 	}
170 }
171 
172 int f2fs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
173 {
174 	struct inode *inode = file->f_mapping->host;
175 	struct f2fs_inode_info *fi = F2FS_I(inode);
176 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
177 	nid_t ino = inode->i_ino;
178 	int ret = 0;
179 	bool need_cp = false;
180 	struct writeback_control wbc = {
181 		.sync_mode = WB_SYNC_ALL,
182 		.nr_to_write = LONG_MAX,
183 		.for_reclaim = 0,
184 	};
185 
186 	if (unlikely(f2fs_readonly(inode->i_sb)))
187 		return 0;
188 
189 	trace_f2fs_sync_file_enter(inode);
190 
191 	/* if fdatasync is triggered, let's do in-place-update */
192 	if (get_dirty_pages(inode) <= SM_I(sbi)->min_fsync_blocks)
193 		set_inode_flag(fi, FI_NEED_IPU);
194 	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
195 	clear_inode_flag(fi, FI_NEED_IPU);
196 
197 	if (ret) {
198 		trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret);
199 		return ret;
200 	}
201 
202 	/* if the inode is dirty, let's recover all the time */
203 	if (!datasync && is_inode_flag_set(fi, FI_DIRTY_INODE)) {
204 		update_inode_page(inode);
205 		goto go_write;
206 	}
207 
208 	/*
209 	 * if there is no written data, don't waste time to write recovery info.
210 	 */
211 	if (!is_inode_flag_set(fi, FI_APPEND_WRITE) &&
212 			!exist_written_data(sbi, ino, APPEND_INO)) {
213 
214 		/* it may call write_inode just prior to fsync */
215 		if (need_inode_page_update(sbi, ino))
216 			goto go_write;
217 
218 		if (is_inode_flag_set(fi, FI_UPDATE_WRITE) ||
219 				exist_written_data(sbi, ino, UPDATE_INO))
220 			goto flush_out;
221 		goto out;
222 	}
223 go_write:
224 	/* guarantee free sections for fsync */
225 	f2fs_balance_fs(sbi);
226 
227 	/*
228 	 * Both of fdatasync() and fsync() are able to be recovered from
229 	 * sudden-power-off.
230 	 */
231 	down_read(&fi->i_sem);
232 	need_cp = need_do_checkpoint(inode);
233 	up_read(&fi->i_sem);
234 
235 	if (need_cp) {
236 		/* all the dirty node pages should be flushed for POR */
237 		ret = f2fs_sync_fs(inode->i_sb, 1);
238 
239 		/*
240 		 * We've secured consistency through sync_fs. Following pino
241 		 * will be used only for fsynced inodes after checkpoint.
242 		 */
243 		try_to_fix_pino(inode);
244 		goto out;
245 	}
246 sync_nodes:
247 	sync_node_pages(sbi, ino, &wbc);
248 
249 	if (need_inode_block_update(sbi, ino)) {
250 		mark_inode_dirty_sync(inode);
251 		f2fs_write_inode(inode, NULL);
252 		goto sync_nodes;
253 	}
254 
255 	ret = wait_on_node_pages_writeback(sbi, ino);
256 	if (ret)
257 		goto out;
258 
259 	/* once recovery info is written, don't need to tack this */
260 	remove_dirty_inode(sbi, ino, APPEND_INO);
261 	clear_inode_flag(fi, FI_APPEND_WRITE);
262 flush_out:
263 	remove_dirty_inode(sbi, ino, UPDATE_INO);
264 	clear_inode_flag(fi, FI_UPDATE_WRITE);
265 	ret = f2fs_issue_flush(sbi);
266 out:
267 	trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret);
268 	return ret;
269 }
270 
271 static pgoff_t __get_first_dirty_index(struct address_space *mapping,
272 						pgoff_t pgofs, int whence)
273 {
274 	struct pagevec pvec;
275 	int nr_pages;
276 
277 	if (whence != SEEK_DATA)
278 		return 0;
279 
280 	/* find first dirty page index */
281 	pagevec_init(&pvec, 0);
282 	nr_pages = pagevec_lookup_tag(&pvec, mapping, &pgofs,
283 					PAGECACHE_TAG_DIRTY, 1);
284 	pgofs = nr_pages ? pvec.pages[0]->index : LONG_MAX;
285 	pagevec_release(&pvec);
286 	return pgofs;
287 }
288 
289 static bool __found_offset(block_t blkaddr, pgoff_t dirty, pgoff_t pgofs,
290 							int whence)
291 {
292 	switch (whence) {
293 	case SEEK_DATA:
294 		if ((blkaddr == NEW_ADDR && dirty == pgofs) ||
295 			(blkaddr != NEW_ADDR && blkaddr != NULL_ADDR))
296 			return true;
297 		break;
298 	case SEEK_HOLE:
299 		if (blkaddr == NULL_ADDR)
300 			return true;
301 		break;
302 	}
303 	return false;
304 }
305 
306 static loff_t f2fs_seek_block(struct file *file, loff_t offset, int whence)
307 {
308 	struct inode *inode = file->f_mapping->host;
309 	loff_t maxbytes = inode->i_sb->s_maxbytes;
310 	struct dnode_of_data dn;
311 	pgoff_t pgofs, end_offset, dirty;
312 	loff_t data_ofs = offset;
313 	loff_t isize;
314 	int err = 0;
315 
316 	mutex_lock(&inode->i_mutex);
317 
318 	isize = i_size_read(inode);
319 	if (offset >= isize)
320 		goto fail;
321 
322 	/* handle inline data case */
323 	if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
324 		if (whence == SEEK_HOLE)
325 			data_ofs = isize;
326 		goto found;
327 	}
328 
329 	pgofs = (pgoff_t)(offset >> PAGE_CACHE_SHIFT);
330 
331 	dirty = __get_first_dirty_index(inode->i_mapping, pgofs, whence);
332 
333 	for (; data_ofs < isize; data_ofs = pgofs << PAGE_CACHE_SHIFT) {
334 		set_new_dnode(&dn, inode, NULL, NULL, 0);
335 		err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
336 		if (err && err != -ENOENT) {
337 			goto fail;
338 		} else if (err == -ENOENT) {
339 			/* direct node does not exists */
340 			if (whence == SEEK_DATA) {
341 				pgofs = PGOFS_OF_NEXT_DNODE(pgofs,
342 							F2FS_I(inode));
343 				continue;
344 			} else {
345 				goto found;
346 			}
347 		}
348 
349 		end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
350 
351 		/* find data/hole in dnode block */
352 		for (; dn.ofs_in_node < end_offset;
353 				dn.ofs_in_node++, pgofs++,
354 				data_ofs = pgofs << PAGE_CACHE_SHIFT) {
355 			block_t blkaddr;
356 			blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
357 
358 			if (__found_offset(blkaddr, dirty, pgofs, whence)) {
359 				f2fs_put_dnode(&dn);
360 				goto found;
361 			}
362 		}
363 		f2fs_put_dnode(&dn);
364 	}
365 
366 	if (whence == SEEK_DATA)
367 		goto fail;
368 found:
369 	if (whence == SEEK_HOLE && data_ofs > isize)
370 		data_ofs = isize;
371 	mutex_unlock(&inode->i_mutex);
372 	return vfs_setpos(file, data_ofs, maxbytes);
373 fail:
374 	mutex_unlock(&inode->i_mutex);
375 	return -ENXIO;
376 }
377 
378 static loff_t f2fs_llseek(struct file *file, loff_t offset, int whence)
379 {
380 	struct inode *inode = file->f_mapping->host;
381 	loff_t maxbytes = inode->i_sb->s_maxbytes;
382 
383 	switch (whence) {
384 	case SEEK_SET:
385 	case SEEK_CUR:
386 	case SEEK_END:
387 		return generic_file_llseek_size(file, offset, whence,
388 						maxbytes, i_size_read(inode));
389 	case SEEK_DATA:
390 	case SEEK_HOLE:
391 		if (offset < 0)
392 			return -ENXIO;
393 		return f2fs_seek_block(file, offset, whence);
394 	}
395 
396 	return -EINVAL;
397 }
398 
399 static int f2fs_file_mmap(struct file *file, struct vm_area_struct *vma)
400 {
401 	struct inode *inode = file_inode(file);
402 
403 	/* we don't need to use inline_data strictly */
404 	if (f2fs_has_inline_data(inode)) {
405 		int err = f2fs_convert_inline_inode(inode);
406 		if (err)
407 			return err;
408 	}
409 
410 	file_accessed(file);
411 	vma->vm_ops = &f2fs_file_vm_ops;
412 	return 0;
413 }
414 
415 int truncate_data_blocks_range(struct dnode_of_data *dn, int count)
416 {
417 	int nr_free = 0, ofs = dn->ofs_in_node;
418 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
419 	struct f2fs_node *raw_node;
420 	__le32 *addr;
421 
422 	raw_node = F2FS_NODE(dn->node_page);
423 	addr = blkaddr_in_node(raw_node) + ofs;
424 
425 	for (; count > 0; count--, addr++, dn->ofs_in_node++) {
426 		block_t blkaddr = le32_to_cpu(*addr);
427 		if (blkaddr == NULL_ADDR)
428 			continue;
429 
430 		update_extent_cache(NULL_ADDR, dn);
431 		invalidate_blocks(sbi, blkaddr);
432 		nr_free++;
433 	}
434 	if (nr_free) {
435 		dec_valid_block_count(sbi, dn->inode, nr_free);
436 		set_page_dirty(dn->node_page);
437 		sync_inode_page(dn);
438 	}
439 	dn->ofs_in_node = ofs;
440 
441 	trace_f2fs_truncate_data_blocks_range(dn->inode, dn->nid,
442 					 dn->ofs_in_node, nr_free);
443 	return nr_free;
444 }
445 
446 void truncate_data_blocks(struct dnode_of_data *dn)
447 {
448 	truncate_data_blocks_range(dn, ADDRS_PER_BLOCK);
449 }
450 
451 static int truncate_partial_data_page(struct inode *inode, u64 from)
452 {
453 	unsigned offset = from & (PAGE_CACHE_SIZE - 1);
454 	struct page *page;
455 
456 	if (!offset)
457 		return 0;
458 
459 	page = find_data_page(inode, from >> PAGE_CACHE_SHIFT, false);
460 	if (IS_ERR(page))
461 		return 0;
462 
463 	lock_page(page);
464 	if (unlikely(!PageUptodate(page) ||
465 			page->mapping != inode->i_mapping))
466 		goto out;
467 
468 	f2fs_wait_on_page_writeback(page, DATA);
469 	zero_user(page, offset, PAGE_CACHE_SIZE - offset);
470 	set_page_dirty(page);
471 out:
472 	f2fs_put_page(page, 1);
473 	return 0;
474 }
475 
476 int truncate_blocks(struct inode *inode, u64 from, bool lock)
477 {
478 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
479 	unsigned int blocksize = inode->i_sb->s_blocksize;
480 	struct dnode_of_data dn;
481 	pgoff_t free_from;
482 	int count = 0, err = 0;
483 	struct page *ipage;
484 
485 	trace_f2fs_truncate_blocks_enter(inode, from);
486 
487 	free_from = (pgoff_t)
488 		((from + blocksize - 1) >> (sbi->log_blocksize));
489 
490 	if (lock)
491 		f2fs_lock_op(sbi);
492 
493 	ipage = get_node_page(sbi, inode->i_ino);
494 	if (IS_ERR(ipage)) {
495 		err = PTR_ERR(ipage);
496 		goto out;
497 	}
498 
499 	if (f2fs_has_inline_data(inode)) {
500 		f2fs_put_page(ipage, 1);
501 		goto out;
502 	}
503 
504 	set_new_dnode(&dn, inode, ipage, NULL, 0);
505 	err = get_dnode_of_data(&dn, free_from, LOOKUP_NODE);
506 	if (err) {
507 		if (err == -ENOENT)
508 			goto free_next;
509 		goto out;
510 	}
511 
512 	count = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
513 
514 	count -= dn.ofs_in_node;
515 	f2fs_bug_on(sbi, count < 0);
516 
517 	if (dn.ofs_in_node || IS_INODE(dn.node_page)) {
518 		truncate_data_blocks_range(&dn, count);
519 		free_from += count;
520 	}
521 
522 	f2fs_put_dnode(&dn);
523 free_next:
524 	err = truncate_inode_blocks(inode, free_from);
525 out:
526 	if (lock)
527 		f2fs_unlock_op(sbi);
528 
529 	/* lastly zero out the first data page */
530 	if (!err)
531 		err = truncate_partial_data_page(inode, from);
532 
533 	trace_f2fs_truncate_blocks_exit(inode, err);
534 	return err;
535 }
536 
537 void f2fs_truncate(struct inode *inode)
538 {
539 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
540 				S_ISLNK(inode->i_mode)))
541 		return;
542 
543 	trace_f2fs_truncate(inode);
544 
545 	/* we should check inline_data size */
546 	if (f2fs_has_inline_data(inode) && !f2fs_may_inline(inode)) {
547 		if (f2fs_convert_inline_inode(inode))
548 			return;
549 	}
550 
551 	if (!truncate_blocks(inode, i_size_read(inode), true)) {
552 		inode->i_mtime = inode->i_ctime = CURRENT_TIME;
553 		mark_inode_dirty(inode);
554 	}
555 }
556 
557 int f2fs_getattr(struct vfsmount *mnt,
558 			 struct dentry *dentry, struct kstat *stat)
559 {
560 	struct inode *inode = dentry->d_inode;
561 	generic_fillattr(inode, stat);
562 	stat->blocks <<= 3;
563 	return 0;
564 }
565 
566 #ifdef CONFIG_F2FS_FS_POSIX_ACL
567 static void __setattr_copy(struct inode *inode, const struct iattr *attr)
568 {
569 	struct f2fs_inode_info *fi = F2FS_I(inode);
570 	unsigned int ia_valid = attr->ia_valid;
571 
572 	if (ia_valid & ATTR_UID)
573 		inode->i_uid = attr->ia_uid;
574 	if (ia_valid & ATTR_GID)
575 		inode->i_gid = attr->ia_gid;
576 	if (ia_valid & ATTR_ATIME)
577 		inode->i_atime = timespec_trunc(attr->ia_atime,
578 						inode->i_sb->s_time_gran);
579 	if (ia_valid & ATTR_MTIME)
580 		inode->i_mtime = timespec_trunc(attr->ia_mtime,
581 						inode->i_sb->s_time_gran);
582 	if (ia_valid & ATTR_CTIME)
583 		inode->i_ctime = timespec_trunc(attr->ia_ctime,
584 						inode->i_sb->s_time_gran);
585 	if (ia_valid & ATTR_MODE) {
586 		umode_t mode = attr->ia_mode;
587 
588 		if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
589 			mode &= ~S_ISGID;
590 		set_acl_inode(fi, mode);
591 	}
592 }
593 #else
594 #define __setattr_copy setattr_copy
595 #endif
596 
597 int f2fs_setattr(struct dentry *dentry, struct iattr *attr)
598 {
599 	struct inode *inode = dentry->d_inode;
600 	struct f2fs_inode_info *fi = F2FS_I(inode);
601 	int err;
602 
603 	err = inode_change_ok(inode, attr);
604 	if (err)
605 		return err;
606 
607 	if (attr->ia_valid & ATTR_SIZE) {
608 		if (attr->ia_size != i_size_read(inode)) {
609 			truncate_setsize(inode, attr->ia_size);
610 			f2fs_truncate(inode);
611 			f2fs_balance_fs(F2FS_I_SB(inode));
612 		} else {
613 			/*
614 			 * giving a chance to truncate blocks past EOF which
615 			 * are fallocated with FALLOC_FL_KEEP_SIZE.
616 			 */
617 			f2fs_truncate(inode);
618 		}
619 	}
620 
621 	__setattr_copy(inode, attr);
622 
623 	if (attr->ia_valid & ATTR_MODE) {
624 		err = posix_acl_chmod(inode, get_inode_mode(inode));
625 		if (err || is_inode_flag_set(fi, FI_ACL_MODE)) {
626 			inode->i_mode = fi->i_acl_mode;
627 			clear_inode_flag(fi, FI_ACL_MODE);
628 		}
629 	}
630 
631 	mark_inode_dirty(inode);
632 	return err;
633 }
634 
635 const struct inode_operations f2fs_file_inode_operations = {
636 	.getattr	= f2fs_getattr,
637 	.setattr	= f2fs_setattr,
638 	.get_acl	= f2fs_get_acl,
639 	.set_acl	= f2fs_set_acl,
640 #ifdef CONFIG_F2FS_FS_XATTR
641 	.setxattr	= generic_setxattr,
642 	.getxattr	= generic_getxattr,
643 	.listxattr	= f2fs_listxattr,
644 	.removexattr	= generic_removexattr,
645 #endif
646 	.fiemap		= f2fs_fiemap,
647 };
648 
649 static void fill_zero(struct inode *inode, pgoff_t index,
650 					loff_t start, loff_t len)
651 {
652 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
653 	struct page *page;
654 
655 	if (!len)
656 		return;
657 
658 	f2fs_balance_fs(sbi);
659 
660 	f2fs_lock_op(sbi);
661 	page = get_new_data_page(inode, NULL, index, false);
662 	f2fs_unlock_op(sbi);
663 
664 	if (!IS_ERR(page)) {
665 		f2fs_wait_on_page_writeback(page, DATA);
666 		zero_user(page, start, len);
667 		set_page_dirty(page);
668 		f2fs_put_page(page, 1);
669 	}
670 }
671 
672 int truncate_hole(struct inode *inode, pgoff_t pg_start, pgoff_t pg_end)
673 {
674 	pgoff_t index;
675 	int err;
676 
677 	for (index = pg_start; index < pg_end; index++) {
678 		struct dnode_of_data dn;
679 
680 		set_new_dnode(&dn, inode, NULL, NULL, 0);
681 		err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
682 		if (err) {
683 			if (err == -ENOENT)
684 				continue;
685 			return err;
686 		}
687 
688 		if (dn.data_blkaddr != NULL_ADDR)
689 			truncate_data_blocks_range(&dn, 1);
690 		f2fs_put_dnode(&dn);
691 	}
692 	return 0;
693 }
694 
695 static int punch_hole(struct inode *inode, loff_t offset, loff_t len)
696 {
697 	pgoff_t pg_start, pg_end;
698 	loff_t off_start, off_end;
699 	int ret = 0;
700 
701 	if (!S_ISREG(inode->i_mode))
702 		return -EOPNOTSUPP;
703 
704 	/* skip punching hole beyond i_size */
705 	if (offset >= inode->i_size)
706 		return ret;
707 
708 	if (f2fs_has_inline_data(inode)) {
709 		ret = f2fs_convert_inline_inode(inode);
710 		if (ret)
711 			return ret;
712 	}
713 
714 	pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT;
715 	pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT;
716 
717 	off_start = offset & (PAGE_CACHE_SIZE - 1);
718 	off_end = (offset + len) & (PAGE_CACHE_SIZE - 1);
719 
720 	if (pg_start == pg_end) {
721 		fill_zero(inode, pg_start, off_start,
722 						off_end - off_start);
723 	} else {
724 		if (off_start)
725 			fill_zero(inode, pg_start++, off_start,
726 					PAGE_CACHE_SIZE - off_start);
727 		if (off_end)
728 			fill_zero(inode, pg_end, 0, off_end);
729 
730 		if (pg_start < pg_end) {
731 			struct address_space *mapping = inode->i_mapping;
732 			loff_t blk_start, blk_end;
733 			struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
734 
735 			f2fs_balance_fs(sbi);
736 
737 			blk_start = pg_start << PAGE_CACHE_SHIFT;
738 			blk_end = pg_end << PAGE_CACHE_SHIFT;
739 			truncate_inode_pages_range(mapping, blk_start,
740 					blk_end - 1);
741 
742 			f2fs_lock_op(sbi);
743 			ret = truncate_hole(inode, pg_start, pg_end);
744 			f2fs_unlock_op(sbi);
745 		}
746 	}
747 
748 	return ret;
749 }
750 
751 static int expand_inode_data(struct inode *inode, loff_t offset,
752 					loff_t len, int mode)
753 {
754 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
755 	pgoff_t index, pg_start, pg_end;
756 	loff_t new_size = i_size_read(inode);
757 	loff_t off_start, off_end;
758 	int ret = 0;
759 
760 	f2fs_balance_fs(sbi);
761 
762 	ret = inode_newsize_ok(inode, (len + offset));
763 	if (ret)
764 		return ret;
765 
766 	if (f2fs_has_inline_data(inode)) {
767 		ret = f2fs_convert_inline_inode(inode);
768 		if (ret)
769 			return ret;
770 	}
771 
772 	pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT;
773 	pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT;
774 
775 	off_start = offset & (PAGE_CACHE_SIZE - 1);
776 	off_end = (offset + len) & (PAGE_CACHE_SIZE - 1);
777 
778 	f2fs_lock_op(sbi);
779 
780 	for (index = pg_start; index <= pg_end; index++) {
781 		struct dnode_of_data dn;
782 
783 		if (index == pg_end && !off_end)
784 			goto noalloc;
785 
786 		set_new_dnode(&dn, inode, NULL, NULL, 0);
787 		ret = f2fs_reserve_block(&dn, index);
788 		if (ret)
789 			break;
790 noalloc:
791 		if (pg_start == pg_end)
792 			new_size = offset + len;
793 		else if (index == pg_start && off_start)
794 			new_size = (index + 1) << PAGE_CACHE_SHIFT;
795 		else if (index == pg_end)
796 			new_size = (index << PAGE_CACHE_SHIFT) + off_end;
797 		else
798 			new_size += PAGE_CACHE_SIZE;
799 	}
800 
801 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
802 		i_size_read(inode) < new_size) {
803 		i_size_write(inode, new_size);
804 		mark_inode_dirty(inode);
805 		update_inode_page(inode);
806 	}
807 	f2fs_unlock_op(sbi);
808 
809 	return ret;
810 }
811 
812 static long f2fs_fallocate(struct file *file, int mode,
813 				loff_t offset, loff_t len)
814 {
815 	struct inode *inode = file_inode(file);
816 	long ret;
817 
818 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
819 		return -EOPNOTSUPP;
820 
821 	mutex_lock(&inode->i_mutex);
822 
823 	if (mode & FALLOC_FL_PUNCH_HOLE)
824 		ret = punch_hole(inode, offset, len);
825 	else
826 		ret = expand_inode_data(inode, offset, len, mode);
827 
828 	if (!ret) {
829 		inode->i_mtime = inode->i_ctime = CURRENT_TIME;
830 		mark_inode_dirty(inode);
831 	}
832 
833 	mutex_unlock(&inode->i_mutex);
834 
835 	trace_f2fs_fallocate(inode, mode, offset, len, ret);
836 	return ret;
837 }
838 
839 #define F2FS_REG_FLMASK		(~(FS_DIRSYNC_FL | FS_TOPDIR_FL))
840 #define F2FS_OTHER_FLMASK	(FS_NODUMP_FL | FS_NOATIME_FL)
841 
842 static inline __u32 f2fs_mask_flags(umode_t mode, __u32 flags)
843 {
844 	if (S_ISDIR(mode))
845 		return flags;
846 	else if (S_ISREG(mode))
847 		return flags & F2FS_REG_FLMASK;
848 	else
849 		return flags & F2FS_OTHER_FLMASK;
850 }
851 
852 static int f2fs_ioc_getflags(struct file *filp, unsigned long arg)
853 {
854 	struct inode *inode = file_inode(filp);
855 	struct f2fs_inode_info *fi = F2FS_I(inode);
856 	unsigned int flags = fi->i_flags & FS_FL_USER_VISIBLE;
857 	return put_user(flags, (int __user *)arg);
858 }
859 
860 static int f2fs_ioc_setflags(struct file *filp, unsigned long arg)
861 {
862 	struct inode *inode = file_inode(filp);
863 	struct f2fs_inode_info *fi = F2FS_I(inode);
864 	unsigned int flags = fi->i_flags & FS_FL_USER_VISIBLE;
865 	unsigned int oldflags;
866 	int ret;
867 
868 	ret = mnt_want_write_file(filp);
869 	if (ret)
870 		return ret;
871 
872 	if (!inode_owner_or_capable(inode)) {
873 		ret = -EACCES;
874 		goto out;
875 	}
876 
877 	if (get_user(flags, (int __user *)arg)) {
878 		ret = -EFAULT;
879 		goto out;
880 	}
881 
882 	flags = f2fs_mask_flags(inode->i_mode, flags);
883 
884 	mutex_lock(&inode->i_mutex);
885 
886 	oldflags = fi->i_flags;
887 
888 	if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
889 		if (!capable(CAP_LINUX_IMMUTABLE)) {
890 			mutex_unlock(&inode->i_mutex);
891 			ret = -EPERM;
892 			goto out;
893 		}
894 	}
895 
896 	flags = flags & FS_FL_USER_MODIFIABLE;
897 	flags |= oldflags & ~FS_FL_USER_MODIFIABLE;
898 	fi->i_flags = flags;
899 	mutex_unlock(&inode->i_mutex);
900 
901 	f2fs_set_inode_flags(inode);
902 	inode->i_ctime = CURRENT_TIME;
903 	mark_inode_dirty(inode);
904 out:
905 	mnt_drop_write_file(filp);
906 	return ret;
907 }
908 
909 static int f2fs_ioc_start_atomic_write(struct file *filp)
910 {
911 	struct inode *inode = file_inode(filp);
912 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
913 
914 	if (!inode_owner_or_capable(inode))
915 		return -EACCES;
916 
917 	f2fs_balance_fs(sbi);
918 
919 	set_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE);
920 
921 	return f2fs_convert_inline_inode(inode);
922 }
923 
924 static int f2fs_release_file(struct inode *inode, struct file *filp)
925 {
926 	/* some remained atomic pages should discarded */
927 	if (f2fs_is_atomic_file(inode) || f2fs_is_volatile_file(inode))
928 		commit_inmem_pages(inode, true);
929 	return 0;
930 }
931 
932 static int f2fs_ioc_commit_atomic_write(struct file *filp)
933 {
934 	struct inode *inode = file_inode(filp);
935 	int ret;
936 
937 	if (!inode_owner_or_capable(inode))
938 		return -EACCES;
939 
940 	if (f2fs_is_volatile_file(inode))
941 		return 0;
942 
943 	ret = mnt_want_write_file(filp);
944 	if (ret)
945 		return ret;
946 
947 	if (f2fs_is_atomic_file(inode))
948 		commit_inmem_pages(inode, false);
949 
950 	ret = f2fs_sync_file(filp, 0, LONG_MAX, 0);
951 	mnt_drop_write_file(filp);
952 	return ret;
953 }
954 
955 static int f2fs_ioc_start_volatile_write(struct file *filp)
956 {
957 	struct inode *inode = file_inode(filp);
958 
959 	if (!inode_owner_or_capable(inode))
960 		return -EACCES;
961 
962 	set_inode_flag(F2FS_I(inode), FI_VOLATILE_FILE);
963 
964 	return f2fs_convert_inline_inode(inode);
965 }
966 
967 static int f2fs_ioc_fitrim(struct file *filp, unsigned long arg)
968 {
969 	struct inode *inode = file_inode(filp);
970 	struct super_block *sb = inode->i_sb;
971 	struct request_queue *q = bdev_get_queue(sb->s_bdev);
972 	struct fstrim_range range;
973 	int ret;
974 
975 	if (!capable(CAP_SYS_ADMIN))
976 		return -EPERM;
977 
978 	if (!blk_queue_discard(q))
979 		return -EOPNOTSUPP;
980 
981 	if (copy_from_user(&range, (struct fstrim_range __user *)arg,
982 				sizeof(range)))
983 		return -EFAULT;
984 
985 	range.minlen = max((unsigned int)range.minlen,
986 				q->limits.discard_granularity);
987 	ret = f2fs_trim_fs(F2FS_SB(sb), &range);
988 	if (ret < 0)
989 		return ret;
990 
991 	if (copy_to_user((struct fstrim_range __user *)arg, &range,
992 				sizeof(range)))
993 		return -EFAULT;
994 	return 0;
995 }
996 
997 long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
998 {
999 	switch (cmd) {
1000 	case F2FS_IOC_GETFLAGS:
1001 		return f2fs_ioc_getflags(filp, arg);
1002 	case F2FS_IOC_SETFLAGS:
1003 		return f2fs_ioc_setflags(filp, arg);
1004 	case F2FS_IOC_START_ATOMIC_WRITE:
1005 		return f2fs_ioc_start_atomic_write(filp);
1006 	case F2FS_IOC_COMMIT_ATOMIC_WRITE:
1007 		return f2fs_ioc_commit_atomic_write(filp);
1008 	case F2FS_IOC_START_VOLATILE_WRITE:
1009 		return f2fs_ioc_start_volatile_write(filp);
1010 	case FITRIM:
1011 		return f2fs_ioc_fitrim(filp, arg);
1012 	default:
1013 		return -ENOTTY;
1014 	}
1015 }
1016 
1017 #ifdef CONFIG_COMPAT
1018 long f2fs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1019 {
1020 	switch (cmd) {
1021 	case F2FS_IOC32_GETFLAGS:
1022 		cmd = F2FS_IOC_GETFLAGS;
1023 		break;
1024 	case F2FS_IOC32_SETFLAGS:
1025 		cmd = F2FS_IOC_SETFLAGS;
1026 		break;
1027 	default:
1028 		return -ENOIOCTLCMD;
1029 	}
1030 	return f2fs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
1031 }
1032 #endif
1033 
1034 const struct file_operations f2fs_file_operations = {
1035 	.llseek		= f2fs_llseek,
1036 	.read		= new_sync_read,
1037 	.write		= new_sync_write,
1038 	.read_iter	= generic_file_read_iter,
1039 	.write_iter	= generic_file_write_iter,
1040 	.open		= generic_file_open,
1041 	.release	= f2fs_release_file,
1042 	.mmap		= f2fs_file_mmap,
1043 	.fsync		= f2fs_sync_file,
1044 	.fallocate	= f2fs_fallocate,
1045 	.unlocked_ioctl	= f2fs_ioctl,
1046 #ifdef CONFIG_COMPAT
1047 	.compat_ioctl	= f2fs_compat_ioctl,
1048 #endif
1049 	.splice_read	= generic_file_splice_read,
1050 	.splice_write	= iter_file_splice_write,
1051 };
1052