xref: /openbmc/linux/fs/ext4/file.c (revision 9726bfcd)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/file.c
4  *
5  * Copyright (C) 1992, 1993, 1994, 1995
6  * Remy Card (card@masi.ibp.fr)
7  * Laboratoire MASI - Institut Blaise Pascal
8  * Universite Pierre et Marie Curie (Paris VI)
9  *
10  *  from
11  *
12  *  linux/fs/minix/file.c
13  *
14  *  Copyright (C) 1991, 1992  Linus Torvalds
15  *
16  *  ext4 fs regular file handling primitives
17  *
18  *  64-bit file support on 64-bit platforms by Jakub Jelinek
19  *	(jj@sunsite.ms.mff.cuni.cz)
20  */
21 
22 #include <linux/time.h>
23 #include <linux/fs.h>
24 #include <linux/iomap.h>
25 #include <linux/mount.h>
26 #include <linux/path.h>
27 #include <linux/dax.h>
28 #include <linux/quotaops.h>
29 #include <linux/pagevec.h>
30 #include <linux/uio.h>
31 #include <linux/mman.h>
32 #include "ext4.h"
33 #include "ext4_jbd2.h"
34 #include "xattr.h"
35 #include "acl.h"
36 
37 #ifdef CONFIG_FS_DAX
38 static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
39 {
40 	struct inode *inode = file_inode(iocb->ki_filp);
41 	ssize_t ret;
42 
43 	if (!inode_trylock_shared(inode)) {
44 		if (iocb->ki_flags & IOCB_NOWAIT)
45 			return -EAGAIN;
46 		inode_lock_shared(inode);
47 	}
48 	/*
49 	 * Recheck under inode lock - at this point we are sure it cannot
50 	 * change anymore
51 	 */
52 	if (!IS_DAX(inode)) {
53 		inode_unlock_shared(inode);
54 		/* Fallback to buffered IO in case we cannot support DAX */
55 		return generic_file_read_iter(iocb, to);
56 	}
57 	ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
58 	inode_unlock_shared(inode);
59 
60 	file_accessed(iocb->ki_filp);
61 	return ret;
62 }
63 #endif
64 
65 static ssize_t ext4_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
66 {
67 	if (unlikely(ext4_forced_shutdown(EXT4_SB(file_inode(iocb->ki_filp)->i_sb))))
68 		return -EIO;
69 
70 	if (!iov_iter_count(to))
71 		return 0; /* skip atime */
72 
73 #ifdef CONFIG_FS_DAX
74 	if (IS_DAX(file_inode(iocb->ki_filp)))
75 		return ext4_dax_read_iter(iocb, to);
76 #endif
77 	return generic_file_read_iter(iocb, to);
78 }
79 
80 /*
81  * Called when an inode is released. Note that this is different
82  * from ext4_file_open: open gets called at every open, but release
83  * gets called only when /all/ the files are closed.
84  */
85 static int ext4_release_file(struct inode *inode, struct file *filp)
86 {
87 	if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
88 		ext4_alloc_da_blocks(inode);
89 		ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
90 	}
91 	/* if we are the last writer on the inode, drop the block reservation */
92 	if ((filp->f_mode & FMODE_WRITE) &&
93 			(atomic_read(&inode->i_writecount) == 1) &&
94 		        !EXT4_I(inode)->i_reserved_data_blocks)
95 	{
96 		down_write(&EXT4_I(inode)->i_data_sem);
97 		ext4_discard_preallocations(inode);
98 		up_write(&EXT4_I(inode)->i_data_sem);
99 	}
100 	if (is_dx(inode) && filp->private_data)
101 		ext4_htree_free_dir_info(filp->private_data);
102 
103 	return 0;
104 }
105 
106 static void ext4_unwritten_wait(struct inode *inode)
107 {
108 	wait_queue_head_t *wq = ext4_ioend_wq(inode);
109 
110 	wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_unwritten) == 0));
111 }
112 
113 /*
114  * This tests whether the IO in question is block-aligned or not.
115  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
116  * are converted to written only after the IO is complete.  Until they are
117  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
118  * it needs to zero out portions of the start and/or end block.  If 2 AIO
119  * threads are at work on the same unwritten block, they must be synchronized
120  * or one thread will zero the other's data, causing corruption.
121  */
122 static int
123 ext4_unaligned_aio(struct inode *inode, struct iov_iter *from, loff_t pos)
124 {
125 	struct super_block *sb = inode->i_sb;
126 	int blockmask = sb->s_blocksize - 1;
127 
128 	if (pos >= ALIGN(i_size_read(inode), sb->s_blocksize))
129 		return 0;
130 
131 	if ((pos | iov_iter_alignment(from)) & blockmask)
132 		return 1;
133 
134 	return 0;
135 }
136 
137 /* Is IO overwriting allocated and initialized blocks? */
138 static bool ext4_overwrite_io(struct inode *inode, loff_t pos, loff_t len)
139 {
140 	struct ext4_map_blocks map;
141 	unsigned int blkbits = inode->i_blkbits;
142 	int err, blklen;
143 
144 	if (pos + len > i_size_read(inode))
145 		return false;
146 
147 	map.m_lblk = pos >> blkbits;
148 	map.m_len = EXT4_MAX_BLOCKS(len, pos, blkbits);
149 	blklen = map.m_len;
150 
151 	err = ext4_map_blocks(NULL, inode, &map, 0);
152 	/*
153 	 * 'err==len' means that all of the blocks have been preallocated,
154 	 * regardless of whether they have been initialized or not. To exclude
155 	 * unwritten extents, we need to check m_flags.
156 	 */
157 	return err == blklen && (map.m_flags & EXT4_MAP_MAPPED);
158 }
159 
160 static ssize_t ext4_write_checks(struct kiocb *iocb, struct iov_iter *from)
161 {
162 	struct inode *inode = file_inode(iocb->ki_filp);
163 	ssize_t ret;
164 
165 	ret = generic_write_checks(iocb, from);
166 	if (ret <= 0)
167 		return ret;
168 
169 	if (unlikely(IS_IMMUTABLE(inode)))
170 		return -EPERM;
171 
172 	/*
173 	 * If we have encountered a bitmap-format file, the size limit
174 	 * is smaller than s_maxbytes, which is for extent-mapped files.
175 	 */
176 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
177 		struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
178 
179 		if (iocb->ki_pos >= sbi->s_bitmap_maxbytes)
180 			return -EFBIG;
181 		iov_iter_truncate(from, sbi->s_bitmap_maxbytes - iocb->ki_pos);
182 	}
183 	return iov_iter_count(from);
184 }
185 
186 #ifdef CONFIG_FS_DAX
187 static ssize_t
188 ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
189 {
190 	struct inode *inode = file_inode(iocb->ki_filp);
191 	ssize_t ret;
192 
193 	if (!inode_trylock(inode)) {
194 		if (iocb->ki_flags & IOCB_NOWAIT)
195 			return -EAGAIN;
196 		inode_lock(inode);
197 	}
198 	ret = ext4_write_checks(iocb, from);
199 	if (ret <= 0)
200 		goto out;
201 	ret = file_remove_privs(iocb->ki_filp);
202 	if (ret)
203 		goto out;
204 	ret = file_update_time(iocb->ki_filp);
205 	if (ret)
206 		goto out;
207 
208 	ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
209 out:
210 	inode_unlock(inode);
211 	if (ret > 0)
212 		ret = generic_write_sync(iocb, ret);
213 	return ret;
214 }
215 #endif
216 
217 static ssize_t
218 ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
219 {
220 	struct inode *inode = file_inode(iocb->ki_filp);
221 	int o_direct = iocb->ki_flags & IOCB_DIRECT;
222 	int unaligned_aio = 0;
223 	int overwrite = 0;
224 	ssize_t ret;
225 
226 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
227 		return -EIO;
228 
229 #ifdef CONFIG_FS_DAX
230 	if (IS_DAX(inode))
231 		return ext4_dax_write_iter(iocb, from);
232 #endif
233 	if (!o_direct && (iocb->ki_flags & IOCB_NOWAIT))
234 		return -EOPNOTSUPP;
235 
236 	if (!inode_trylock(inode)) {
237 		if (iocb->ki_flags & IOCB_NOWAIT)
238 			return -EAGAIN;
239 		inode_lock(inode);
240 	}
241 
242 	ret = ext4_write_checks(iocb, from);
243 	if (ret <= 0)
244 		goto out;
245 
246 	/*
247 	 * Unaligned direct AIO must be serialized among each other as zeroing
248 	 * of partial blocks of two competing unaligned AIOs can result in data
249 	 * corruption.
250 	 */
251 	if (o_direct && ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) &&
252 	    !is_sync_kiocb(iocb) &&
253 	    ext4_unaligned_aio(inode, from, iocb->ki_pos)) {
254 		unaligned_aio = 1;
255 		ext4_unwritten_wait(inode);
256 	}
257 
258 	iocb->private = &overwrite;
259 	/* Check whether we do a DIO overwrite or not */
260 	if (o_direct && !unaligned_aio) {
261 		if (ext4_overwrite_io(inode, iocb->ki_pos, iov_iter_count(from))) {
262 			if (ext4_should_dioread_nolock(inode))
263 				overwrite = 1;
264 		} else if (iocb->ki_flags & IOCB_NOWAIT) {
265 			ret = -EAGAIN;
266 			goto out;
267 		}
268 	}
269 
270 	ret = __generic_file_write_iter(iocb, from);
271 	/*
272 	 * Unaligned direct AIO must be the only IO in flight. Otherwise
273 	 * overlapping aligned IO after unaligned might result in data
274 	 * corruption.
275 	 */
276 	if (ret == -EIOCBQUEUED && unaligned_aio)
277 		ext4_unwritten_wait(inode);
278 	inode_unlock(inode);
279 
280 	if (ret > 0)
281 		ret = generic_write_sync(iocb, ret);
282 
283 	return ret;
284 
285 out:
286 	inode_unlock(inode);
287 	return ret;
288 }
289 
290 #ifdef CONFIG_FS_DAX
291 static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf,
292 		enum page_entry_size pe_size)
293 {
294 	int error = 0;
295 	vm_fault_t result;
296 	int retries = 0;
297 	handle_t *handle = NULL;
298 	struct inode *inode = file_inode(vmf->vma->vm_file);
299 	struct super_block *sb = inode->i_sb;
300 
301 	/*
302 	 * We have to distinguish real writes from writes which will result in a
303 	 * COW page; COW writes should *not* poke the journal (the file will not
304 	 * be changed). Doing so would cause unintended failures when mounted
305 	 * read-only.
306 	 *
307 	 * We check for VM_SHARED rather than vmf->cow_page since the latter is
308 	 * unset for pe_size != PE_SIZE_PTE (i.e. only in do_cow_fault); for
309 	 * other sizes, dax_iomap_fault will handle splitting / fallback so that
310 	 * we eventually come back with a COW page.
311 	 */
312 	bool write = (vmf->flags & FAULT_FLAG_WRITE) &&
313 		(vmf->vma->vm_flags & VM_SHARED);
314 	pfn_t pfn;
315 
316 	if (write) {
317 		sb_start_pagefault(sb);
318 		file_update_time(vmf->vma->vm_file);
319 		down_read(&EXT4_I(inode)->i_mmap_sem);
320 retry:
321 		handle = ext4_journal_start_sb(sb, EXT4_HT_WRITE_PAGE,
322 					       EXT4_DATA_TRANS_BLOCKS(sb));
323 		if (IS_ERR(handle)) {
324 			up_read(&EXT4_I(inode)->i_mmap_sem);
325 			sb_end_pagefault(sb);
326 			return VM_FAULT_SIGBUS;
327 		}
328 	} else {
329 		down_read(&EXT4_I(inode)->i_mmap_sem);
330 	}
331 	result = dax_iomap_fault(vmf, pe_size, &pfn, &error, &ext4_iomap_ops);
332 	if (write) {
333 		ext4_journal_stop(handle);
334 
335 		if ((result & VM_FAULT_ERROR) && error == -ENOSPC &&
336 		    ext4_should_retry_alloc(sb, &retries))
337 			goto retry;
338 		/* Handling synchronous page fault? */
339 		if (result & VM_FAULT_NEEDDSYNC)
340 			result = dax_finish_sync_fault(vmf, pe_size, pfn);
341 		up_read(&EXT4_I(inode)->i_mmap_sem);
342 		sb_end_pagefault(sb);
343 	} else {
344 		up_read(&EXT4_I(inode)->i_mmap_sem);
345 	}
346 
347 	return result;
348 }
349 
350 static vm_fault_t ext4_dax_fault(struct vm_fault *vmf)
351 {
352 	return ext4_dax_huge_fault(vmf, PE_SIZE_PTE);
353 }
354 
355 static const struct vm_operations_struct ext4_dax_vm_ops = {
356 	.fault		= ext4_dax_fault,
357 	.huge_fault	= ext4_dax_huge_fault,
358 	.page_mkwrite	= ext4_dax_fault,
359 	.pfn_mkwrite	= ext4_dax_fault,
360 };
361 #else
362 #define ext4_dax_vm_ops	ext4_file_vm_ops
363 #endif
364 
365 static const struct vm_operations_struct ext4_file_vm_ops = {
366 	.fault		= ext4_filemap_fault,
367 	.map_pages	= filemap_map_pages,
368 	.page_mkwrite   = ext4_page_mkwrite,
369 };
370 
371 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
372 {
373 	struct inode *inode = file->f_mapping->host;
374 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
375 	struct dax_device *dax_dev = sbi->s_daxdev;
376 
377 	if (unlikely(ext4_forced_shutdown(sbi)))
378 		return -EIO;
379 
380 	/*
381 	 * We don't support synchronous mappings for non-DAX files and
382 	 * for DAX files if underneath dax_device is not synchronous.
383 	 */
384 	if (!daxdev_mapping_supported(vma, dax_dev))
385 		return -EOPNOTSUPP;
386 
387 	file_accessed(file);
388 	if (IS_DAX(file_inode(file))) {
389 		vma->vm_ops = &ext4_dax_vm_ops;
390 		vma->vm_flags |= VM_HUGEPAGE;
391 	} else {
392 		vma->vm_ops = &ext4_file_vm_ops;
393 	}
394 	return 0;
395 }
396 
397 static int ext4_sample_last_mounted(struct super_block *sb,
398 				    struct vfsmount *mnt)
399 {
400 	struct ext4_sb_info *sbi = EXT4_SB(sb);
401 	struct path path;
402 	char buf[64], *cp;
403 	handle_t *handle;
404 	int err;
405 
406 	if (likely(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED))
407 		return 0;
408 
409 	if (sb_rdonly(sb) || !sb_start_intwrite_trylock(sb))
410 		return 0;
411 
412 	sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
413 	/*
414 	 * Sample where the filesystem has been mounted and
415 	 * store it in the superblock for sysadmin convenience
416 	 * when trying to sort through large numbers of block
417 	 * devices or filesystem images.
418 	 */
419 	memset(buf, 0, sizeof(buf));
420 	path.mnt = mnt;
421 	path.dentry = mnt->mnt_root;
422 	cp = d_path(&path, buf, sizeof(buf));
423 	err = 0;
424 	if (IS_ERR(cp))
425 		goto out;
426 
427 	handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
428 	err = PTR_ERR(handle);
429 	if (IS_ERR(handle))
430 		goto out;
431 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
432 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
433 	if (err)
434 		goto out_journal;
435 	strlcpy(sbi->s_es->s_last_mounted, cp,
436 		sizeof(sbi->s_es->s_last_mounted));
437 	ext4_handle_dirty_super(handle, sb);
438 out_journal:
439 	ext4_journal_stop(handle);
440 out:
441 	sb_end_intwrite(sb);
442 	return err;
443 }
444 
445 static int ext4_file_open(struct inode * inode, struct file * filp)
446 {
447 	int ret;
448 
449 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
450 		return -EIO;
451 
452 	ret = ext4_sample_last_mounted(inode->i_sb, filp->f_path.mnt);
453 	if (ret)
454 		return ret;
455 
456 	ret = fscrypt_file_open(inode, filp);
457 	if (ret)
458 		return ret;
459 
460 	/*
461 	 * Set up the jbd2_inode if we are opening the inode for
462 	 * writing and the journal is present
463 	 */
464 	if (filp->f_mode & FMODE_WRITE) {
465 		ret = ext4_inode_attach_jinode(inode);
466 		if (ret < 0)
467 			return ret;
468 	}
469 
470 	filp->f_mode |= FMODE_NOWAIT;
471 	return dquot_file_open(inode, filp);
472 }
473 
474 /*
475  * ext4_llseek() handles both block-mapped and extent-mapped maxbytes values
476  * by calling generic_file_llseek_size() with the appropriate maxbytes
477  * value for each.
478  */
479 loff_t ext4_llseek(struct file *file, loff_t offset, int whence)
480 {
481 	struct inode *inode = file->f_mapping->host;
482 	loff_t maxbytes;
483 
484 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
485 		maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
486 	else
487 		maxbytes = inode->i_sb->s_maxbytes;
488 
489 	switch (whence) {
490 	default:
491 		return generic_file_llseek_size(file, offset, whence,
492 						maxbytes, i_size_read(inode));
493 	case SEEK_HOLE:
494 		inode_lock_shared(inode);
495 		offset = iomap_seek_hole(inode, offset, &ext4_iomap_ops);
496 		inode_unlock_shared(inode);
497 		break;
498 	case SEEK_DATA:
499 		inode_lock_shared(inode);
500 		offset = iomap_seek_data(inode, offset, &ext4_iomap_ops);
501 		inode_unlock_shared(inode);
502 		break;
503 	}
504 
505 	if (offset < 0)
506 		return offset;
507 	return vfs_setpos(file, offset, maxbytes);
508 }
509 
510 const struct file_operations ext4_file_operations = {
511 	.llseek		= ext4_llseek,
512 	.read_iter	= ext4_file_read_iter,
513 	.write_iter	= ext4_file_write_iter,
514 	.unlocked_ioctl = ext4_ioctl,
515 #ifdef CONFIG_COMPAT
516 	.compat_ioctl	= ext4_compat_ioctl,
517 #endif
518 	.mmap		= ext4_file_mmap,
519 	.mmap_supported_flags = MAP_SYNC,
520 	.open		= ext4_file_open,
521 	.release	= ext4_release_file,
522 	.fsync		= ext4_sync_file,
523 	.get_unmapped_area = thp_get_unmapped_area,
524 	.splice_read	= generic_file_splice_read,
525 	.splice_write	= iter_file_splice_write,
526 	.fallocate	= ext4_fallocate,
527 };
528 
529 const struct inode_operations ext4_file_inode_operations = {
530 	.setattr	= ext4_setattr,
531 	.getattr	= ext4_file_getattr,
532 	.listxattr	= ext4_listxattr,
533 	.get_acl	= ext4_get_acl,
534 	.set_acl	= ext4_set_acl,
535 	.fiemap		= ext4_fiemap,
536 };
537 
538