file.c

  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 *  linux/fs/ext2/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 *  ext2 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/pagemap.h>
 24#include <linux/dax.h>
 25#include <linux/quotaops.h>
 26#include <linux/iomap.h>
 27#include <linux/uio.h>
 28#include <linux/buffer_head.h>
 29#include "ext2.h"
 30#include "xattr.h"
 31#include "acl.h"
 32#include "trace.h"
 33
 34#ifdef CONFIG_FS_DAX
 35static ssize_t ext2_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
 36{
 37	struct inode *inode = iocb->ki_filp->f_mapping->host;
 38	ssize_t ret;
 39
 40	if (!iov_iter_count(to))
 41		return 0; /* skip atime */
 42
 43	inode_lock_shared(inode);
 44	ret = dax_iomap_rw(iocb, to, &ext2_iomap_ops);
 45	inode_unlock_shared(inode);
 46
 47	file_accessed(iocb->ki_filp);
 48	return ret;
 49}
 50
 51static ssize_t ext2_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
 52{
 53	struct file *file = iocb->ki_filp;
 54	struct inode *inode = file->f_mapping->host;
 55	ssize_t ret;
 56
 57	inode_lock(inode);
 58	ret = generic_write_checks(iocb, from);
 59	if (ret <= 0)
 60		goto out_unlock;
 61	ret = file_remove_privs(file);
 62	if (ret)
 63		goto out_unlock;
 64	ret = file_update_time(file);
 65	if (ret)
 66		goto out_unlock;
 67
 68	ret = dax_iomap_rw(iocb, from, &ext2_iomap_ops);
 69	if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
 70		i_size_write(inode, iocb->ki_pos);
 71		mark_inode_dirty(inode);
 72	}
 73
 74out_unlock:
 75	inode_unlock(inode);
 76	if (ret > 0)
 77		ret = generic_write_sync(iocb, ret);
 78	return ret;
 79}
 80
 81/*
 82 * The lock ordering for ext2 DAX fault paths is:
 83 *
 84 * mmap_lock (MM)
 85 *   sb_start_pagefault (vfs, freeze)
 86 *     address_space->invalidate_lock
 87 *       address_space->i_mmap_rwsem or page_lock (mutually exclusive in DAX)
 88 *         ext2_inode_info->truncate_mutex
 89 *
 90 * The default page_lock and i_size verification done by non-DAX fault paths
 91 * is sufficient because ext2 doesn't support hole punching.
 92 */
 93static vm_fault_t ext2_dax_fault(struct vm_fault *vmf)
 94{
 95	struct inode *inode = file_inode(vmf->vma->vm_file);
 96	vm_fault_t ret;
 97	bool write = (vmf->flags & FAULT_FLAG_WRITE) &&
 98		(vmf->vma->vm_flags & VM_SHARED);
 99
100	if (write) {
101		sb_start_pagefault(inode->i_sb);
102		file_update_time(vmf->vma->vm_file);
103	}
104	filemap_invalidate_lock_shared(inode->i_mapping);
105
106	ret = dax_iomap_fault(vmf, 0, NULL, NULL, &ext2_iomap_ops);
107
108	filemap_invalidate_unlock_shared(inode->i_mapping);
109	if (write)
110		sb_end_pagefault(inode->i_sb);
111	return ret;
112}
113
114static const struct vm_operations_struct ext2_dax_vm_ops = {
115	.fault		= ext2_dax_fault,
116	/*
117	 * .huge_fault is not supported for DAX because allocation in ext2
118	 * cannot be reliably aligned to huge page sizes and so pmd faults
119	 * will always fail and fail back to regular faults.
120	 */
121	.page_mkwrite	= ext2_dax_fault,
122	.pfn_mkwrite	= ext2_dax_fault,
123};
124
125static int ext2_file_mmap_prepare(struct vm_area_desc *desc)
126{
127	struct file *file = desc->file;
128
129	if (!IS_DAX(file_inode(file)))
130		return generic_file_mmap_prepare(desc);
131
132	file_accessed(file);
133	desc->vm_ops = &ext2_dax_vm_ops;
134	return 0;
135}
136#else
137#define ext2_file_mmap_prepare	generic_file_mmap_prepare
138#endif
139
140/*
141 * Called when filp is released. This happens when all file descriptors
142 * for a single struct file are closed. Note that different open() calls
143 * for the same file yield different struct file structures.
144 */
145static int ext2_release_file (struct inode * inode, struct file * filp)
146{
147	if (filp->f_mode & FMODE_WRITE) {
148		mutex_lock(&EXT2_I(inode)->truncate_mutex);
149		ext2_discard_reservation(inode);
150		mutex_unlock(&EXT2_I(inode)->truncate_mutex);
151	}
152	return 0;
153}
154
155int ext2_fsync(struct file *file, loff_t start, loff_t end, int datasync)
156{
157	int ret;
158	struct super_block *sb = file->f_mapping->host->i_sb;
159
160	ret = generic_buffers_fsync(file, start, end, datasync);
161	if (ret == -EIO)
162		/* We don't really know where the IO error happened... */
163		ext2_error(sb, __func__,
164			   "detected IO error when writing metadata buffers");
165	return ret;
166}
167
168static ssize_t ext2_dio_read_iter(struct kiocb *iocb, struct iov_iter *to)
169{
170	struct file *file = iocb->ki_filp;
171	struct inode *inode = file->f_mapping->host;
172	ssize_t ret;
173
174	trace_ext2_dio_read_begin(iocb, to, 0);
175	inode_lock_shared(inode);
176	ret = iomap_dio_rw(iocb, to, &ext2_iomap_ops, NULL, 0, NULL, 0);
177	inode_unlock_shared(inode);
178	trace_ext2_dio_read_end(iocb, to, ret);
179
180	return ret;
181}
182
183static int ext2_dio_write_end_io(struct kiocb *iocb, ssize_t size,
184				 int error, unsigned int flags)
185{
186	loff_t pos = iocb->ki_pos;
187	struct inode *inode = file_inode(iocb->ki_filp);
188
189	if (error)
190		goto out;
191
192	/*
193	 * If we are extending the file, we have to update i_size here before
194	 * page cache gets invalidated in iomap_dio_rw(). This prevents racing
195	 * buffered reads from zeroing out too much from page cache pages.
196	 * Note that all extending writes always happens synchronously with
197	 * inode lock held by ext2_dio_write_iter(). So it is safe to update
198	 * inode size here for extending file writes.
199	 */
200	pos += size;
201	if (pos > i_size_read(inode)) {
202		i_size_write(inode, pos);
203		mark_inode_dirty(inode);
204	}
205out:
206	trace_ext2_dio_write_endio(iocb, size, error);
207	return error;
208}
209
210static const struct iomap_dio_ops ext2_dio_write_ops = {
211	.end_io = ext2_dio_write_end_io,
212};
213
214static ssize_t ext2_dio_write_iter(struct kiocb *iocb, struct iov_iter *from)
215{
216	struct file *file = iocb->ki_filp;
217	struct inode *inode = file->f_mapping->host;
218	ssize_t ret;
219	unsigned int flags = 0;
220	unsigned long blocksize = inode->i_sb->s_blocksize;
221	loff_t offset = iocb->ki_pos;
222	loff_t count = iov_iter_count(from);
223	ssize_t status = 0;
224
225	trace_ext2_dio_write_begin(iocb, from, 0);
226	inode_lock(inode);
227	ret = generic_write_checks(iocb, from);
228	if (ret <= 0)
229		goto out_unlock;
230
231	ret = kiocb_modified(iocb);
232	if (ret)
233		goto out_unlock;
234
235	/* use IOMAP_DIO_FORCE_WAIT for unaligned or extending writes */
236	if (iocb->ki_pos + iov_iter_count(from) > i_size_read(inode) ||
237	   (!IS_ALIGNED(iocb->ki_pos | iov_iter_alignment(from), blocksize)))
238		flags |= IOMAP_DIO_FORCE_WAIT;
239
240	ret = iomap_dio_rw(iocb, from, &ext2_iomap_ops, &ext2_dio_write_ops,
241			   flags, NULL, 0);
242
243	/* ENOTBLK is magic return value for fallback to buffered-io */
244	if (ret == -ENOTBLK)
245		ret = 0;
246
247	if (ret < 0 && ret != -EIOCBQUEUED)
248		ext2_write_failed(inode->i_mapping, offset + count);
249
250	/* handle case for partial write and for fallback to buffered write */
251	if (ret >= 0 && iov_iter_count(from)) {
252		loff_t pos, endbyte;
253		int ret2;
254
255		iocb->ki_flags &= ~IOCB_DIRECT;
256		pos = iocb->ki_pos;
257		status = generic_perform_write(iocb, from);
258		if (unlikely(status < 0)) {
259			ret = status;
260			goto out_unlock;
261		}
262
263		ret += status;
264		endbyte = pos + status - 1;
265		ret2 = filemap_write_and_wait_range(inode->i_mapping, pos,
266						    endbyte);
267		if (!ret2)
268			invalidate_mapping_pages(inode->i_mapping,
269						 pos >> PAGE_SHIFT,
270						 endbyte >> PAGE_SHIFT);
271		if (ret > 0)
272			generic_write_sync(iocb, ret);
273	}
274
275out_unlock:
276	inode_unlock(inode);
277	if (status)
278		trace_ext2_dio_write_buff_end(iocb, from, status);
279	trace_ext2_dio_write_end(iocb, from, ret);
280	return ret;
281}
282
283static ssize_t ext2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
284{
285#ifdef CONFIG_FS_DAX
286	if (IS_DAX(iocb->ki_filp->f_mapping->host))
287		return ext2_dax_read_iter(iocb, to);
288#endif
289	if (iocb->ki_flags & IOCB_DIRECT)
290		return ext2_dio_read_iter(iocb, to);
291
292	return generic_file_read_iter(iocb, to);
293}
294
295static ssize_t ext2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
296{
297#ifdef CONFIG_FS_DAX
298	if (IS_DAX(iocb->ki_filp->f_mapping->host))
299		return ext2_dax_write_iter(iocb, from);
300#endif
301	if (iocb->ki_flags & IOCB_DIRECT)
302		return ext2_dio_write_iter(iocb, from);
303
304	return generic_file_write_iter(iocb, from);
305}
306
307static int ext2_file_open(struct inode *inode, struct file *filp)
308{
309	filp->f_mode |= FMODE_CAN_ODIRECT;
310	return dquot_file_open(inode, filp);
311}
312
313const struct file_operations ext2_file_operations = {
314	.llseek		= generic_file_llseek,
315	.read_iter	= ext2_file_read_iter,
316	.write_iter	= ext2_file_write_iter,
317	.unlocked_ioctl = ext2_ioctl,
318#ifdef CONFIG_COMPAT
319	.compat_ioctl	= ext2_compat_ioctl,
320#endif
321	.mmap_prepare	= ext2_file_mmap_prepare,
322	.open		= ext2_file_open,
323	.release	= ext2_release_file,
324	.fsync		= ext2_fsync,
325	.get_unmapped_area = thp_get_unmapped_area,
326	.splice_read	= filemap_splice_read,
327	.splice_write	= iter_file_splice_write,
328};
329
330const struct inode_operations ext2_file_inode_operations = {
331	.listxattr	= ext2_listxattr,
332	.getattr	= ext2_getattr,
333	.setattr	= ext2_setattr,
334	.get_inode_acl	= ext2_get_acl,
335	.set_acl	= ext2_set_acl,
336	.fiemap		= ext2_fiemap,
337	.fileattr_get	= ext2_fileattr_get,
338	.fileattr_set	= ext2_fileattr_set,
339};