xref: /openbmc/linux/fs/f2fs/file.c (revision 8b036556) 
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.h"
30 #include <trace/events/f2fs.h>
31 
32 static int f2fs_vm_page_mkwrite(struct vm_area_struct *vma,
33 						struct vm_fault *vmf)
34 {
35 	struct page *page = vmf->page;
36 	struct inode *inode = file_inode(vma->vm_file);
37 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
38 	struct dnode_of_data dn;
39 	int err;
40 
41 	f2fs_balance_fs(sbi);
42 
43 	sb_start_pagefault(inode->i_sb);
44 
45 	f2fs_bug_on(sbi, f2fs_has_inline_data(inode));
46 
47 	/* block allocation */
48 	f2fs_lock_op(sbi);
49 	set_new_dnode(&dn, inode, NULL, NULL, 0);
50 	err = f2fs_reserve_block(&dn, page->index);
51 	if (err) {
52 		f2fs_unlock_op(sbi);
53 		goto out;
54 	}
55 	f2fs_put_dnode(&dn);
56 	f2fs_unlock_op(sbi);
57 
58 	file_update_time(vma->vm_file);
59 	lock_page(page);
60 	if (unlikely(page->mapping != inode->i_mapping ||
61 			page_offset(page) > i_size_read(inode) ||
62 			!PageUptodate(page))) {
63 		unlock_page(page);
64 		err = -EFAULT;
65 		goto out;
66 	}
67 
68 	/*
69 	 * check to see if the page is mapped already (no holes)
70 	 */
71 	if (PageMappedToDisk(page))
72 		goto mapped;
73 
74 	/* page is wholly or partially inside EOF */
75 	if (((page->index + 1) << PAGE_CACHE_SHIFT) > i_size_read(inode)) {
76 		unsigned offset;
77 		offset = i_size_read(inode) & ~PAGE_CACHE_MASK;
78 		zero_user_segment(page, offset, PAGE_CACHE_SIZE);
79 	}
80 	set_page_dirty(page);
81 	SetPageUptodate(page);
82 
83 	trace_f2fs_vm_page_mkwrite(page, DATA);
84 mapped:
85 	/* fill the page */
86 	f2fs_wait_on_page_writeback(page, DATA);
87 out:
88 	sb_end_pagefault(inode->i_sb);
89 	return block_page_mkwrite_return(err);
90 }
91 
92 static const struct vm_operations_struct f2fs_file_vm_ops = {
93 	.fault		= filemap_fault,
94 	.map_pages	= filemap_map_pages,
95 	.page_mkwrite	= f2fs_vm_page_mkwrite,
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 cp_error was enabled, we should avoid infinite loop */
250 	if (unlikely(f2fs_cp_error(sbi)))
251 		goto out;
252 
253 	if (need_inode_block_update(sbi, ino)) {
254 		mark_inode_dirty_sync(inode);
255 		f2fs_write_inode(inode, NULL);
256 		goto sync_nodes;
257 	}
258 
259 	ret = wait_on_node_pages_writeback(sbi, ino);
260 	if (ret)
261 		goto out;
262 
263 	/* once recovery info is written, don't need to tack this */
264 	remove_dirty_inode(sbi, ino, APPEND_INO);
265 	clear_inode_flag(fi, FI_APPEND_WRITE);
266 flush_out:
267 	remove_dirty_inode(sbi, ino, UPDATE_INO);
268 	clear_inode_flag(fi, FI_UPDATE_WRITE);
269 	ret = f2fs_issue_flush(sbi);
270 out:
271 	trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret);
272 	f2fs_trace_ios(NULL, NULL, 1);
273 	return ret;
274 }
275 
276 static pgoff_t __get_first_dirty_index(struct address_space *mapping,
277 						pgoff_t pgofs, int whence)
278 {
279 	struct pagevec pvec;
280 	int nr_pages;
281 
282 	if (whence != SEEK_DATA)
283 		return 0;
284 
285 	/* find first dirty page index */
286 	pagevec_init(&pvec, 0);
287 	nr_pages = pagevec_lookup_tag(&pvec, mapping, &pgofs,
288 					PAGECACHE_TAG_DIRTY, 1);
289 	pgofs = nr_pages ? pvec.pages[0]->index : LONG_MAX;
290 	pagevec_release(&pvec);
291 	return pgofs;
292 }
293 
294 static bool __found_offset(block_t blkaddr, pgoff_t dirty, pgoff_t pgofs,
295 							int whence)
296 {
297 	switch (whence) {
298 	case SEEK_DATA:
299 		if ((blkaddr == NEW_ADDR && dirty == pgofs) ||
300 			(blkaddr != NEW_ADDR && blkaddr != NULL_ADDR))
301 			return true;
302 		break;
303 	case SEEK_HOLE:
304 		if (blkaddr == NULL_ADDR)
305 			return true;
306 		break;
307 	}
308 	return false;
309 }
310 
311 static loff_t f2fs_seek_block(struct file *file, loff_t offset, int whence)
312 {
313 	struct inode *inode = file->f_mapping->host;
314 	loff_t maxbytes = inode->i_sb->s_maxbytes;
315 	struct dnode_of_data dn;
316 	pgoff_t pgofs, end_offset, dirty;
317 	loff_t data_ofs = offset;
318 	loff_t isize;
319 	int err = 0;
320 
321 	mutex_lock(&inode->i_mutex);
322 
323 	isize = i_size_read(inode);
324 	if (offset >= isize)
325 		goto fail;
326 
327 	/* handle inline data case */
328 	if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
329 		if (whence == SEEK_HOLE)
330 			data_ofs = isize;
331 		goto found;
332 	}
333 
334 	pgofs = (pgoff_t)(offset >> PAGE_CACHE_SHIFT);
335 
336 	dirty = __get_first_dirty_index(inode->i_mapping, pgofs, whence);
337 
338 	for (; data_ofs < isize; data_ofs = pgofs << PAGE_CACHE_SHIFT) {
339 		set_new_dnode(&dn, inode, NULL, NULL, 0);
340 		err = get_dnode_of_data(&dn, pgofs, LOOKUP_NODE_RA);
341 		if (err && err != -ENOENT) {
342 			goto fail;
343 		} else if (err == -ENOENT) {
344 			/* direct node does not exists */
345 			if (whence == SEEK_DATA) {
346 				pgofs = PGOFS_OF_NEXT_DNODE(pgofs,
347 							F2FS_I(inode));
348 				continue;
349 			} else {
350 				goto found;
351 			}
352 		}
353 
354 		end_offset = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
355 
356 		/* find data/hole in dnode block */
357 		for (; dn.ofs_in_node < end_offset;
358 				dn.ofs_in_node++, pgofs++,
359 				data_ofs = (loff_t)pgofs << PAGE_CACHE_SHIFT) {
360 			block_t blkaddr;
361 			blkaddr = datablock_addr(dn.node_page, dn.ofs_in_node);
362 
363 			if (__found_offset(blkaddr, dirty, pgofs, whence)) {
364 				f2fs_put_dnode(&dn);
365 				goto found;
366 			}
367 		}
368 		f2fs_put_dnode(&dn);
369 	}
370 
371 	if (whence == SEEK_DATA)
372 		goto fail;
373 found:
374 	if (whence == SEEK_HOLE && data_ofs > isize)
375 		data_ofs = isize;
376 	mutex_unlock(&inode->i_mutex);
377 	return vfs_setpos(file, data_ofs, maxbytes);
378 fail:
379 	mutex_unlock(&inode->i_mutex);
380 	return -ENXIO;
381 }
382 
383 static loff_t f2fs_llseek(struct file *file, loff_t offset, int whence)
384 {
385 	struct inode *inode = file->f_mapping->host;
386 	loff_t maxbytes = inode->i_sb->s_maxbytes;
387 
388 	switch (whence) {
389 	case SEEK_SET:
390 	case SEEK_CUR:
391 	case SEEK_END:
392 		return generic_file_llseek_size(file, offset, whence,
393 						maxbytes, i_size_read(inode));
394 	case SEEK_DATA:
395 	case SEEK_HOLE:
396 		if (offset < 0)
397 			return -ENXIO;
398 		return f2fs_seek_block(file, offset, whence);
399 	}
400 
401 	return -EINVAL;
402 }
403 
404 static int f2fs_file_mmap(struct file *file, struct vm_area_struct *vma)
405 {
406 	struct inode *inode = file_inode(file);
407 
408 	/* we don't need to use inline_data strictly */
409 	if (f2fs_has_inline_data(inode)) {
410 		int err = f2fs_convert_inline_inode(inode);
411 		if (err)
412 			return err;
413 	}
414 
415 	file_accessed(file);
416 	vma->vm_ops = &f2fs_file_vm_ops;
417 	return 0;
418 }
419 
420 int truncate_data_blocks_range(struct dnode_of_data *dn, int count)
421 {
422 	int nr_free = 0, ofs = dn->ofs_in_node;
423 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
424 	struct f2fs_node *raw_node;
425 	__le32 *addr;
426 
427 	raw_node = F2FS_NODE(dn->node_page);
428 	addr = blkaddr_in_node(raw_node) + ofs;
429 
430 	for (; count > 0; count--, addr++, dn->ofs_in_node++) {
431 		block_t blkaddr = le32_to_cpu(*addr);
432 		if (blkaddr == NULL_ADDR)
433 			continue;
434 
435 		dn->data_blkaddr = NULL_ADDR;
436 		update_extent_cache(dn);
437 		invalidate_blocks(sbi, blkaddr);
438 		nr_free++;
439 	}
440 	if (nr_free) {
441 		dec_valid_block_count(sbi, dn->inode, nr_free);
442 		set_page_dirty(dn->node_page);
443 		sync_inode_page(dn);
444 	}
445 	dn->ofs_in_node = ofs;
446 
447 	trace_f2fs_truncate_data_blocks_range(dn->inode, dn->nid,
448 					 dn->ofs_in_node, nr_free);
449 	return nr_free;
450 }
451 
452 void truncate_data_blocks(struct dnode_of_data *dn)
453 {
454 	truncate_data_blocks_range(dn, ADDRS_PER_BLOCK);
455 }
456 
457 static int truncate_partial_data_page(struct inode *inode, u64 from)
458 {
459 	unsigned offset = from & (PAGE_CACHE_SIZE - 1);
460 	struct page *page;
461 
462 	if (!offset)
463 		return 0;
464 
465 	page = find_data_page(inode, from >> PAGE_CACHE_SHIFT, false);
466 	if (IS_ERR(page))
467 		return 0;
468 
469 	lock_page(page);
470 	if (unlikely(!PageUptodate(page) ||
471 			page->mapping != inode->i_mapping))
472 		goto out;
473 
474 	f2fs_wait_on_page_writeback(page, DATA);
475 	zero_user(page, offset, PAGE_CACHE_SIZE - offset);
476 	set_page_dirty(page);
477 out:
478 	f2fs_put_page(page, 1);
479 	return 0;
480 }
481 
482 int truncate_blocks(struct inode *inode, u64 from, bool lock)
483 {
484 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
485 	unsigned int blocksize = inode->i_sb->s_blocksize;
486 	struct dnode_of_data dn;
487 	pgoff_t free_from;
488 	int count = 0, err = 0;
489 	struct page *ipage;
490 
491 	trace_f2fs_truncate_blocks_enter(inode, from);
492 
493 	free_from = (pgoff_t)F2FS_BYTES_TO_BLK(from + blocksize - 1);
494 
495 	if (lock)
496 		f2fs_lock_op(sbi);
497 
498 	ipage = get_node_page(sbi, inode->i_ino);
499 	if (IS_ERR(ipage)) {
500 		err = PTR_ERR(ipage);
501 		goto out;
502 	}
503 
504 	if (f2fs_has_inline_data(inode)) {
505 		f2fs_put_page(ipage, 1);
506 		goto out;
507 	}
508 
509 	set_new_dnode(&dn, inode, ipage, NULL, 0);
510 	err = get_dnode_of_data(&dn, free_from, LOOKUP_NODE);
511 	if (err) {
512 		if (err == -ENOENT)
513 			goto free_next;
514 		goto out;
515 	}
516 
517 	count = ADDRS_PER_PAGE(dn.node_page, F2FS_I(inode));
518 
519 	count -= dn.ofs_in_node;
520 	f2fs_bug_on(sbi, count < 0);
521 
522 	if (dn.ofs_in_node || IS_INODE(dn.node_page)) {
523 		truncate_data_blocks_range(&dn, count);
524 		free_from += count;
525 	}
526 
527 	f2fs_put_dnode(&dn);
528 free_next:
529 	err = truncate_inode_blocks(inode, free_from);
530 out:
531 	if (lock)
532 		f2fs_unlock_op(sbi);
533 
534 	/* lastly zero out the first data page */
535 	if (!err)
536 		err = truncate_partial_data_page(inode, from);
537 
538 	trace_f2fs_truncate_blocks_exit(inode, err);
539 	return err;
540 }
541 
542 void f2fs_truncate(struct inode *inode)
543 {
544 	if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
545 				S_ISLNK(inode->i_mode)))
546 		return;
547 
548 	trace_f2fs_truncate(inode);
549 
550 	/* we should check inline_data size */
551 	if (f2fs_has_inline_data(inode) && !f2fs_may_inline(inode)) {
552 		if (f2fs_convert_inline_inode(inode))
553 			return;
554 	}
555 
556 	if (!truncate_blocks(inode, i_size_read(inode), true)) {
557 		inode->i_mtime = inode->i_ctime = CURRENT_TIME;
558 		mark_inode_dirty(inode);
559 	}
560 }
561 
562 int f2fs_getattr(struct vfsmount *mnt,
563 			 struct dentry *dentry, struct kstat *stat)
564 {
565 	struct inode *inode = dentry->d_inode;
566 	generic_fillattr(inode, stat);
567 	stat->blocks <<= 3;
568 	return 0;
569 }
570 
571 #ifdef CONFIG_F2FS_FS_POSIX_ACL
572 static void __setattr_copy(struct inode *inode, const struct iattr *attr)
573 {
574 	struct f2fs_inode_info *fi = F2FS_I(inode);
575 	unsigned int ia_valid = attr->ia_valid;
576 
577 	if (ia_valid & ATTR_UID)
578 		inode->i_uid = attr->ia_uid;
579 	if (ia_valid & ATTR_GID)
580 		inode->i_gid = attr->ia_gid;
581 	if (ia_valid & ATTR_ATIME)
582 		inode->i_atime = timespec_trunc(attr->ia_atime,
583 						inode->i_sb->s_time_gran);
584 	if (ia_valid & ATTR_MTIME)
585 		inode->i_mtime = timespec_trunc(attr->ia_mtime,
586 						inode->i_sb->s_time_gran);
587 	if (ia_valid & ATTR_CTIME)
588 		inode->i_ctime = timespec_trunc(attr->ia_ctime,
589 						inode->i_sb->s_time_gran);
590 	if (ia_valid & ATTR_MODE) {
591 		umode_t mode = attr->ia_mode;
592 
593 		if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
594 			mode &= ~S_ISGID;
595 		set_acl_inode(fi, mode);
596 	}
597 }
598 #else
599 #define __setattr_copy setattr_copy
600 #endif
601 
602 int f2fs_setattr(struct dentry *dentry, struct iattr *attr)
603 {
604 	struct inode *inode = dentry->d_inode;
605 	struct f2fs_inode_info *fi = F2FS_I(inode);
606 	int err;
607 
608 	err = inode_change_ok(inode, attr);
609 	if (err)
610 		return err;
611 
612 	if (attr->ia_valid & ATTR_SIZE) {
613 		if (attr->ia_size != i_size_read(inode)) {
614 			truncate_setsize(inode, attr->ia_size);
615 			f2fs_truncate(inode);
616 			f2fs_balance_fs(F2FS_I_SB(inode));
617 		} else {
618 			/*
619 			 * giving a chance to truncate blocks past EOF which
620 			 * are fallocated with FALLOC_FL_KEEP_SIZE.
621 			 */
622 			f2fs_truncate(inode);
623 		}
624 	}
625 
626 	__setattr_copy(inode, attr);
627 
628 	if (attr->ia_valid & ATTR_MODE) {
629 		err = posix_acl_chmod(inode, get_inode_mode(inode));
630 		if (err || is_inode_flag_set(fi, FI_ACL_MODE)) {
631 			inode->i_mode = fi->i_acl_mode;
632 			clear_inode_flag(fi, FI_ACL_MODE);
633 		}
634 	}
635 
636 	mark_inode_dirty(inode);
637 	return err;
638 }
639 
640 const struct inode_operations f2fs_file_inode_operations = {
641 	.getattr	= f2fs_getattr,
642 	.setattr	= f2fs_setattr,
643 	.get_acl	= f2fs_get_acl,
644 	.set_acl	= f2fs_set_acl,
645 #ifdef CONFIG_F2FS_FS_XATTR
646 	.setxattr	= generic_setxattr,
647 	.getxattr	= generic_getxattr,
648 	.listxattr	= f2fs_listxattr,
649 	.removexattr	= generic_removexattr,
650 #endif
651 	.fiemap		= f2fs_fiemap,
652 };
653 
654 static void fill_zero(struct inode *inode, pgoff_t index,
655 					loff_t start, loff_t len)
656 {
657 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
658 	struct page *page;
659 
660 	if (!len)
661 		return;
662 
663 	f2fs_balance_fs(sbi);
664 
665 	f2fs_lock_op(sbi);
666 	page = get_new_data_page(inode, NULL, index, false);
667 	f2fs_unlock_op(sbi);
668 
669 	if (!IS_ERR(page)) {
670 		f2fs_wait_on_page_writeback(page, DATA);
671 		zero_user(page, start, len);
672 		set_page_dirty(page);
673 		f2fs_put_page(page, 1);
674 	}
675 }
676 
677 int truncate_hole(struct inode *inode, pgoff_t pg_start, pgoff_t pg_end)
678 {
679 	pgoff_t index;
680 	int err;
681 
682 	for (index = pg_start; index < pg_end; index++) {
683 		struct dnode_of_data dn;
684 
685 		set_new_dnode(&dn, inode, NULL, NULL, 0);
686 		err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
687 		if (err) {
688 			if (err == -ENOENT)
689 				continue;
690 			return err;
691 		}
692 
693 		if (dn.data_blkaddr != NULL_ADDR)
694 			truncate_data_blocks_range(&dn, 1);
695 		f2fs_put_dnode(&dn);
696 	}
697 	return 0;
698 }
699 
700 static int punch_hole(struct inode *inode, loff_t offset, loff_t len)
701 {
702 	pgoff_t pg_start, pg_end;
703 	loff_t off_start, off_end;
704 	int ret = 0;
705 
706 	if (!S_ISREG(inode->i_mode))
707 		return -EOPNOTSUPP;
708 
709 	/* skip punching hole beyond i_size */
710 	if (offset >= inode->i_size)
711 		return ret;
712 
713 	if (f2fs_has_inline_data(inode)) {
714 		ret = f2fs_convert_inline_inode(inode);
715 		if (ret)
716 			return ret;
717 	}
718 
719 	pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT;
720 	pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT;
721 
722 	off_start = offset & (PAGE_CACHE_SIZE - 1);
723 	off_end = (offset + len) & (PAGE_CACHE_SIZE - 1);
724 
725 	if (pg_start == pg_end) {
726 		fill_zero(inode, pg_start, off_start,
727 						off_end - off_start);
728 	} else {
729 		if (off_start)
730 			fill_zero(inode, pg_start++, off_start,
731 					PAGE_CACHE_SIZE - off_start);
732 		if (off_end)
733 			fill_zero(inode, pg_end, 0, off_end);
734 
735 		if (pg_start < pg_end) {
736 			struct address_space *mapping = inode->i_mapping;
737 			loff_t blk_start, blk_end;
738 			struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
739 
740 			f2fs_balance_fs(sbi);
741 
742 			blk_start = pg_start << PAGE_CACHE_SHIFT;
743 			blk_end = pg_end << PAGE_CACHE_SHIFT;
744 			truncate_inode_pages_range(mapping, blk_start,
745 					blk_end - 1);
746 
747 			f2fs_lock_op(sbi);
748 			ret = truncate_hole(inode, pg_start, pg_end);
749 			f2fs_unlock_op(sbi);
750 		}
751 	}
752 
753 	return ret;
754 }
755 
756 static int expand_inode_data(struct inode *inode, loff_t offset,
757 					loff_t len, int mode)
758 {
759 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
760 	pgoff_t index, pg_start, pg_end;
761 	loff_t new_size = i_size_read(inode);
762 	loff_t off_start, off_end;
763 	int ret = 0;
764 
765 	f2fs_balance_fs(sbi);
766 
767 	ret = inode_newsize_ok(inode, (len + offset));
768 	if (ret)
769 		return ret;
770 
771 	if (f2fs_has_inline_data(inode)) {
772 		ret = f2fs_convert_inline_inode(inode);
773 		if (ret)
774 			return ret;
775 	}
776 
777 	pg_start = ((unsigned long long) offset) >> PAGE_CACHE_SHIFT;
778 	pg_end = ((unsigned long long) offset + len) >> PAGE_CACHE_SHIFT;
779 
780 	off_start = offset & (PAGE_CACHE_SIZE - 1);
781 	off_end = (offset + len) & (PAGE_CACHE_SIZE - 1);
782 
783 	f2fs_lock_op(sbi);
784 
785 	for (index = pg_start; index <= pg_end; index++) {
786 		struct dnode_of_data dn;
787 
788 		if (index == pg_end && !off_end)
789 			goto noalloc;
790 
791 		set_new_dnode(&dn, inode, NULL, NULL, 0);
792 		ret = f2fs_reserve_block(&dn, index);
793 		if (ret)
794 			break;
795 noalloc:
796 		if (pg_start == pg_end)
797 			new_size = offset + len;
798 		else if (index == pg_start && off_start)
799 			new_size = (index + 1) << PAGE_CACHE_SHIFT;
800 		else if (index == pg_end)
801 			new_size = (index << PAGE_CACHE_SHIFT) + off_end;
802 		else
803 			new_size += PAGE_CACHE_SIZE;
804 	}
805 
806 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
807 		i_size_read(inode) < new_size) {
808 		i_size_write(inode, new_size);
809 		mark_inode_dirty(inode);
810 		update_inode_page(inode);
811 	}
812 	f2fs_unlock_op(sbi);
813 
814 	return ret;
815 }
816 
817 static long f2fs_fallocate(struct file *file, int mode,
818 				loff_t offset, loff_t len)
819 {
820 	struct inode *inode = file_inode(file);
821 	long ret;
822 
823 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
824 		return -EOPNOTSUPP;
825 
826 	mutex_lock(&inode->i_mutex);
827 
828 	if (mode & FALLOC_FL_PUNCH_HOLE)
829 		ret = punch_hole(inode, offset, len);
830 	else
831 		ret = expand_inode_data(inode, offset, len, mode);
832 
833 	if (!ret) {
834 		inode->i_mtime = inode->i_ctime = CURRENT_TIME;
835 		mark_inode_dirty(inode);
836 	}
837 
838 	mutex_unlock(&inode->i_mutex);
839 
840 	trace_f2fs_fallocate(inode, mode, offset, len, ret);
841 	return ret;
842 }
843 
844 static int f2fs_release_file(struct inode *inode, struct file *filp)
845 {
846 	/* some remained atomic pages should discarded */
847 	if (f2fs_is_atomic_file(inode))
848 		commit_inmem_pages(inode, true);
849 	if (f2fs_is_volatile_file(inode)) {
850 		set_inode_flag(F2FS_I(inode), FI_DROP_CACHE);
851 		filemap_fdatawrite(inode->i_mapping);
852 		clear_inode_flag(F2FS_I(inode), FI_DROP_CACHE);
853 	}
854 	return 0;
855 }
856 
857 #define F2FS_REG_FLMASK		(~(FS_DIRSYNC_FL | FS_TOPDIR_FL))
858 #define F2FS_OTHER_FLMASK	(FS_NODUMP_FL | FS_NOATIME_FL)
859 
860 static inline __u32 f2fs_mask_flags(umode_t mode, __u32 flags)
861 {
862 	if (S_ISDIR(mode))
863 		return flags;
864 	else if (S_ISREG(mode))
865 		return flags & F2FS_REG_FLMASK;
866 	else
867 		return flags & F2FS_OTHER_FLMASK;
868 }
869 
870 static int f2fs_ioc_getflags(struct file *filp, unsigned long arg)
871 {
872 	struct inode *inode = file_inode(filp);
873 	struct f2fs_inode_info *fi = F2FS_I(inode);
874 	unsigned int flags = fi->i_flags & FS_FL_USER_VISIBLE;
875 	return put_user(flags, (int __user *)arg);
876 }
877 
878 static int f2fs_ioc_setflags(struct file *filp, unsigned long arg)
879 {
880 	struct inode *inode = file_inode(filp);
881 	struct f2fs_inode_info *fi = F2FS_I(inode);
882 	unsigned int flags = fi->i_flags & FS_FL_USER_VISIBLE;
883 	unsigned int oldflags;
884 	int ret;
885 
886 	ret = mnt_want_write_file(filp);
887 	if (ret)
888 		return ret;
889 
890 	if (!inode_owner_or_capable(inode)) {
891 		ret = -EACCES;
892 		goto out;
893 	}
894 
895 	if (get_user(flags, (int __user *)arg)) {
896 		ret = -EFAULT;
897 		goto out;
898 	}
899 
900 	flags = f2fs_mask_flags(inode->i_mode, flags);
901 
902 	mutex_lock(&inode->i_mutex);
903 
904 	oldflags = fi->i_flags;
905 
906 	if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) {
907 		if (!capable(CAP_LINUX_IMMUTABLE)) {
908 			mutex_unlock(&inode->i_mutex);
909 			ret = -EPERM;
910 			goto out;
911 		}
912 	}
913 
914 	flags = flags & FS_FL_USER_MODIFIABLE;
915 	flags |= oldflags & ~FS_FL_USER_MODIFIABLE;
916 	fi->i_flags = flags;
917 	mutex_unlock(&inode->i_mutex);
918 
919 	f2fs_set_inode_flags(inode);
920 	inode->i_ctime = CURRENT_TIME;
921 	mark_inode_dirty(inode);
922 out:
923 	mnt_drop_write_file(filp);
924 	return ret;
925 }
926 
927 static int f2fs_ioc_getversion(struct file *filp, unsigned long arg)
928 {
929 	struct inode *inode = file_inode(filp);
930 
931 	return put_user(inode->i_generation, (int __user *)arg);
932 }
933 
934 static int f2fs_ioc_start_atomic_write(struct file *filp)
935 {
936 	struct inode *inode = file_inode(filp);
937 
938 	if (!inode_owner_or_capable(inode))
939 		return -EACCES;
940 
941 	f2fs_balance_fs(F2FS_I_SB(inode));
942 
943 	if (f2fs_is_atomic_file(inode))
944 		return 0;
945 
946 	set_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE);
947 
948 	return f2fs_convert_inline_inode(inode);
949 }
950 
951 static int f2fs_ioc_commit_atomic_write(struct file *filp)
952 {
953 	struct inode *inode = file_inode(filp);
954 	int ret;
955 
956 	if (!inode_owner_or_capable(inode))
957 		return -EACCES;
958 
959 	if (f2fs_is_volatile_file(inode))
960 		return 0;
961 
962 	ret = mnt_want_write_file(filp);
963 	if (ret)
964 		return ret;
965 
966 	if (f2fs_is_atomic_file(inode))
967 		commit_inmem_pages(inode, false);
968 
969 	ret = f2fs_sync_file(filp, 0, LONG_MAX, 0);
970 	mnt_drop_write_file(filp);
971 	clear_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE);
972 	return ret;
973 }
974 
975 static int f2fs_ioc_start_volatile_write(struct file *filp)
976 {
977 	struct inode *inode = file_inode(filp);
978 
979 	if (!inode_owner_or_capable(inode))
980 		return -EACCES;
981 
982 	if (f2fs_is_volatile_file(inode))
983 		return 0;
984 
985 	set_inode_flag(F2FS_I(inode), FI_VOLATILE_FILE);
986 
987 	return f2fs_convert_inline_inode(inode);
988 }
989 
990 static int f2fs_ioc_release_volatile_write(struct file *filp)
991 {
992 	struct inode *inode = file_inode(filp);
993 
994 	if (!inode_owner_or_capable(inode))
995 		return -EACCES;
996 
997 	if (!f2fs_is_volatile_file(inode))
998 		return 0;
999 
1000 	punch_hole(inode, 0, F2FS_BLKSIZE);
1001 	return 0;
1002 }
1003 
1004 static int f2fs_ioc_abort_volatile_write(struct file *filp)
1005 {
1006 	struct inode *inode = file_inode(filp);
1007 	int ret;
1008 
1009 	if (!inode_owner_or_capable(inode))
1010 		return -EACCES;
1011 
1012 	ret = mnt_want_write_file(filp);
1013 	if (ret)
1014 		return ret;
1015 
1016 	f2fs_balance_fs(F2FS_I_SB(inode));
1017 
1018 	if (f2fs_is_atomic_file(inode)) {
1019 		commit_inmem_pages(inode, false);
1020 		clear_inode_flag(F2FS_I(inode), FI_ATOMIC_FILE);
1021 	}
1022 
1023 	if (f2fs_is_volatile_file(inode)) {
1024 		clear_inode_flag(F2FS_I(inode), FI_VOLATILE_FILE);
1025 		filemap_fdatawrite(inode->i_mapping);
1026 		set_inode_flag(F2FS_I(inode), FI_VOLATILE_FILE);
1027 	}
1028 	mnt_drop_write_file(filp);
1029 	return ret;
1030 }
1031 
1032 static int f2fs_ioc_fitrim(struct file *filp, unsigned long arg)
1033 {
1034 	struct inode *inode = file_inode(filp);
1035 	struct super_block *sb = inode->i_sb;
1036 	struct request_queue *q = bdev_get_queue(sb->s_bdev);
1037 	struct fstrim_range range;
1038 	int ret;
1039 
1040 	if (!capable(CAP_SYS_ADMIN))
1041 		return -EPERM;
1042 
1043 	if (!blk_queue_discard(q))
1044 		return -EOPNOTSUPP;
1045 
1046 	if (copy_from_user(&range, (struct fstrim_range __user *)arg,
1047 				sizeof(range)))
1048 		return -EFAULT;
1049 
1050 	range.minlen = max((unsigned int)range.minlen,
1051 				q->limits.discard_granularity);
1052 	ret = f2fs_trim_fs(F2FS_SB(sb), &range);
1053 	if (ret < 0)
1054 		return ret;
1055 
1056 	if (copy_to_user((struct fstrim_range __user *)arg, &range,
1057 				sizeof(range)))
1058 		return -EFAULT;
1059 	return 0;
1060 }
1061 
1062 long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
1063 {
1064 	switch (cmd) {
1065 	case F2FS_IOC_GETFLAGS:
1066 		return f2fs_ioc_getflags(filp, arg);
1067 	case F2FS_IOC_SETFLAGS:
1068 		return f2fs_ioc_setflags(filp, arg);
1069 	case F2FS_IOC_GETVERSION:
1070 		return f2fs_ioc_getversion(filp, arg);
1071 	case F2FS_IOC_START_ATOMIC_WRITE:
1072 		return f2fs_ioc_start_atomic_write(filp);
1073 	case F2FS_IOC_COMMIT_ATOMIC_WRITE:
1074 		return f2fs_ioc_commit_atomic_write(filp);
1075 	case F2FS_IOC_START_VOLATILE_WRITE:
1076 		return f2fs_ioc_start_volatile_write(filp);
1077 	case F2FS_IOC_RELEASE_VOLATILE_WRITE:
1078 		return f2fs_ioc_release_volatile_write(filp);
1079 	case F2FS_IOC_ABORT_VOLATILE_WRITE:
1080 		return f2fs_ioc_abort_volatile_write(filp);
1081 	case FITRIM:
1082 		return f2fs_ioc_fitrim(filp, arg);
1083 	default:
1084 		return -ENOTTY;
1085 	}
1086 }
1087 
1088 #ifdef CONFIG_COMPAT
1089 long f2fs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1090 {
1091 	switch (cmd) {
1092 	case F2FS_IOC32_GETFLAGS:
1093 		cmd = F2FS_IOC_GETFLAGS;
1094 		break;
1095 	case F2FS_IOC32_SETFLAGS:
1096 		cmd = F2FS_IOC_SETFLAGS;
1097 		break;
1098 	default:
1099 		return -ENOIOCTLCMD;
1100 	}
1101 	return f2fs_ioctl(file, cmd, (unsigned long) compat_ptr(arg));
1102 }
1103 #endif
1104 
1105 const struct file_operations f2fs_file_operations = {
1106 	.llseek		= f2fs_llseek,
1107 	.read		= new_sync_read,
1108 	.write		= new_sync_write,
1109 	.read_iter	= generic_file_read_iter,
1110 	.write_iter	= generic_file_write_iter,
1111 	.open		= generic_file_open,
1112 	.release	= f2fs_release_file,
1113 	.mmap		= f2fs_file_mmap,
1114 	.fsync		= f2fs_sync_file,
1115 	.fallocate	= f2fs_fallocate,
1116 	.unlocked_ioctl	= f2fs_ioctl,
1117 #ifdef CONFIG_COMPAT
1118 	.compat_ioctl	= f2fs_compat_ioctl,
1119 #endif
1120 	.splice_read	= generic_file_splice_read,
1121 	.splice_write	= iter_file_splice_write,
1122 };
1123