1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Common helpers for stackable filesystems and backing files.
  4 *
  5 * Forked from fs/overlayfs/file.c.
  6 *
  7 * Copyright (C) 2017 Red Hat, Inc.
  8 * Copyright (C) 2023 CTERA Networks.
  9 */
 10
 11#include <linux/fs.h>
 12#include <linux/backing-file.h>
 13#include <linux/splice.h>
 14#include <linux/mm.h>
 15
 16#include "internal.h"
 17
 18/**
 19 * backing_file_open - open a backing file for kernel internal use
 20 * @user_path:	path that the user reuqested to open
 21 * @flags:	open flags
 22 * @real_path:	path of the backing file
 23 * @cred:	credentials for open
 24 *
 25 * Open a backing file for a stackable filesystem (e.g., overlayfs).
 26 * @user_path may be on the stackable filesystem and @real_path on the
 27 * underlying filesystem.  In this case, we want to be able to return the
 28 * @user_path of the stackable filesystem. This is done by embedding the
 29 * returned file into a container structure that also stores the stacked
 30 * file's path, which can be retrieved using backing_file_user_path().
 31 */
 32struct file *backing_file_open(const struct path *user_path, int flags,
 33			       const struct path *real_path,
 34			       const struct cred *cred)
 35{
 36	struct file *f;
 37	int error;
 38
 39	f = alloc_empty_backing_file(flags, cred);
 40	if (IS_ERR(f))
 41		return f;
 42
 43	path_get(user_path);
 44	*backing_file_user_path(f) = *user_path;
 45	error = vfs_open(real_path, f);
 46	if (error) {
 47		fput(f);
 48		f = ERR_PTR(error);
 49	}
 50
 51	return f;
 52}
 53EXPORT_SYMBOL_GPL(backing_file_open);
 54
 55struct file *backing_tmpfile_open(const struct path *user_path, int flags,
 56				  const struct path *real_parentpath,
 57				  umode_t mode, const struct cred *cred)
 58{
 59	struct mnt_idmap *real_idmap = mnt_idmap(real_parentpath->mnt);
 60	struct file *f;
 61	int error;
 62
 63	f = alloc_empty_backing_file(flags, cred);
 64	if (IS_ERR(f))
 65		return f;
 66
 67	path_get(user_path);
 68	*backing_file_user_path(f) = *user_path;
 69	error = vfs_tmpfile(real_idmap, real_parentpath, f, mode);
 70	if (error) {
 71		fput(f);
 72		f = ERR_PTR(error);
 73	}
 74	return f;
 75}
 76EXPORT_SYMBOL(backing_tmpfile_open);
 77
 78struct backing_aio {
 79	struct kiocb iocb;
 80	refcount_t ref;
 81	struct kiocb *orig_iocb;
 82	/* used for aio completion */
 83	void (*end_write)(struct kiocb *iocb, ssize_t);
 84	struct work_struct work;
 85	long res;
 86};
 87
 88static struct kmem_cache *backing_aio_cachep;
 89
 90#define BACKING_IOCB_MASK \
 91	(IOCB_NOWAIT | IOCB_HIPRI | IOCB_DSYNC | IOCB_SYNC | IOCB_APPEND)
 92
 93static rwf_t iocb_to_rw_flags(int flags)
 94{
 95	return (__force rwf_t)(flags & BACKING_IOCB_MASK);
 96}
 97
 98static void backing_aio_put(struct backing_aio *aio)
 99{
100	if (refcount_dec_and_test(&aio->ref)) {
101		fput(aio->iocb.ki_filp);
102		kmem_cache_free(backing_aio_cachep, aio);
103	}
104}
105
106static void backing_aio_cleanup(struct backing_aio *aio, long res)
107{
108	struct kiocb *iocb = &aio->iocb;
109	struct kiocb *orig_iocb = aio->orig_iocb;
110
111	orig_iocb->ki_pos = iocb->ki_pos;
112	if (aio->end_write)
113		aio->end_write(orig_iocb, res);
114
115	backing_aio_put(aio);
116}
117
118static void backing_aio_rw_complete(struct kiocb *iocb, long res)
119{
120	struct backing_aio *aio = container_of(iocb, struct backing_aio, iocb);
121	struct kiocb *orig_iocb = aio->orig_iocb;
122
123	if (iocb->ki_flags & IOCB_WRITE)
124		kiocb_end_write(iocb);
125
126	backing_aio_cleanup(aio, res);
127	orig_iocb->ki_complete(orig_iocb, res);
128}
129
130static void backing_aio_complete_work(struct work_struct *work)
131{
132	struct backing_aio *aio = container_of(work, struct backing_aio, work);
133
134	backing_aio_rw_complete(&aio->iocb, aio->res);
135}
136
137static void backing_aio_queue_completion(struct kiocb *iocb, long res)
138{
139	struct backing_aio *aio = container_of(iocb, struct backing_aio, iocb);
140
141	/*
142	 * Punt to a work queue to serialize updates of mtime/size.
143	 */
144	aio->res = res;
145	INIT_WORK(&aio->work, backing_aio_complete_work);
146	queue_work(file_inode(aio->orig_iocb->ki_filp)->i_sb->s_dio_done_wq,
147		   &aio->work);
148}
149
150static int backing_aio_init_wq(struct kiocb *iocb)
151{
152	struct super_block *sb = file_inode(iocb->ki_filp)->i_sb;
153
154	if (sb->s_dio_done_wq)
155		return 0;
156
157	return sb_init_dio_done_wq(sb);
158}
159
160
161ssize_t backing_file_read_iter(struct file *file, struct iov_iter *iter,
162			       struct kiocb *iocb, int flags,
163			       struct backing_file_ctx *ctx)
164{
165	struct backing_aio *aio = NULL;
166	const struct cred *old_cred;
167	ssize_t ret;
168
169	if (WARN_ON_ONCE(!(file->f_mode & FMODE_BACKING)))
170		return -EIO;
171
172	if (!iov_iter_count(iter))
173		return 0;
174
175	if (iocb->ki_flags & IOCB_DIRECT &&
176	    !(file->f_mode & FMODE_CAN_ODIRECT))
177		return -EINVAL;
178
179	old_cred = override_creds_light(ctx->cred);
180	if (is_sync_kiocb(iocb)) {
181		rwf_t rwf = iocb_to_rw_flags(flags);
182
183		ret = vfs_iter_read(file, iter, &iocb->ki_pos, rwf);
184	} else {
185		ret = -ENOMEM;
186		aio = kmem_cache_zalloc(backing_aio_cachep, GFP_KERNEL);
187		if (!aio)
188			goto out;
189
190		aio->orig_iocb = iocb;
191		kiocb_clone(&aio->iocb, iocb, get_file(file));
192		aio->iocb.ki_complete = backing_aio_rw_complete;
193		refcount_set(&aio->ref, 2);
194		ret = vfs_iocb_iter_read(file, &aio->iocb, iter);
195		backing_aio_put(aio);
196		if (ret != -EIOCBQUEUED)
197			backing_aio_cleanup(aio, ret);
198	}
199out:
200	revert_creds_light(old_cred);
201
202	if (ctx->accessed)
203		ctx->accessed(iocb->ki_filp);
204
205	return ret;
206}
207EXPORT_SYMBOL_GPL(backing_file_read_iter);
208
209ssize_t backing_file_write_iter(struct file *file, struct iov_iter *iter,
210				struct kiocb *iocb, int flags,
211				struct backing_file_ctx *ctx)
212{
213	const struct cred *old_cred;
214	ssize_t ret;
215
216	if (WARN_ON_ONCE(!(file->f_mode & FMODE_BACKING)))
217		return -EIO;
218
219	if (!iov_iter_count(iter))
220		return 0;
221
222	ret = file_remove_privs(iocb->ki_filp);
223	if (ret)
224		return ret;
225
226	if (iocb->ki_flags & IOCB_DIRECT &&
227	    !(file->f_mode & FMODE_CAN_ODIRECT))
228		return -EINVAL;
229
230	/*
231	 * Stacked filesystems don't support deferred completions, don't copy
232	 * this property in case it is set by the issuer.
233	 */
234	flags &= ~IOCB_DIO_CALLER_COMP;
235
236	old_cred = override_creds_light(ctx->cred);
237	if (is_sync_kiocb(iocb)) {
238		rwf_t rwf = iocb_to_rw_flags(flags);
239
240		ret = vfs_iter_write(file, iter, &iocb->ki_pos, rwf);
241		if (ctx->end_write)
242			ctx->end_write(iocb, ret);
243	} else {
244		struct backing_aio *aio;
245
246		ret = backing_aio_init_wq(iocb);
247		if (ret)
248			goto out;
249
250		ret = -ENOMEM;
251		aio = kmem_cache_zalloc(backing_aio_cachep, GFP_KERNEL);
252		if (!aio)
253			goto out;
254
255		aio->orig_iocb = iocb;
256		aio->end_write = ctx->end_write;
257		kiocb_clone(&aio->iocb, iocb, get_file(file));
258		aio->iocb.ki_flags = flags;
259		aio->iocb.ki_complete = backing_aio_queue_completion;
260		refcount_set(&aio->ref, 2);
261		ret = vfs_iocb_iter_write(file, &aio->iocb, iter);
262		backing_aio_put(aio);
263		if (ret != -EIOCBQUEUED)
264			backing_aio_cleanup(aio, ret);
265	}
266out:
267	revert_creds_light(old_cred);
268
269	return ret;
270}
271EXPORT_SYMBOL_GPL(backing_file_write_iter);
272
273ssize_t backing_file_splice_read(struct file *in, struct kiocb *iocb,
274				 struct pipe_inode_info *pipe, size_t len,
275				 unsigned int flags,
276				 struct backing_file_ctx *ctx)
277{
278	const struct cred *old_cred;
279	ssize_t ret;
280
281	if (WARN_ON_ONCE(!(in->f_mode & FMODE_BACKING)))
282		return -EIO;
283
284	old_cred = override_creds_light(ctx->cred);
285	ret = vfs_splice_read(in, &iocb->ki_pos, pipe, len, flags);
286	revert_creds_light(old_cred);
287
288	if (ctx->accessed)
289		ctx->accessed(iocb->ki_filp);
290
291	return ret;
292}
293EXPORT_SYMBOL_GPL(backing_file_splice_read);
294
295ssize_t backing_file_splice_write(struct pipe_inode_info *pipe,
296				  struct file *out, struct kiocb *iocb,
297				  size_t len, unsigned int flags,
298				  struct backing_file_ctx *ctx)
299{
300	const struct cred *old_cred;
301	ssize_t ret;
302
303	if (WARN_ON_ONCE(!(out->f_mode & FMODE_BACKING)))
304		return -EIO;
305
306	if (!out->f_op->splice_write)
307		return -EINVAL;
308
309	ret = file_remove_privs(iocb->ki_filp);
310	if (ret)
311		return ret;
312
313	old_cred = override_creds_light(ctx->cred);
314	file_start_write(out);
315	ret = out->f_op->splice_write(pipe, out, &iocb->ki_pos, len, flags);
316	file_end_write(out);
317	revert_creds_light(old_cred);
318
319	if (ctx->end_write)
320		ctx->end_write(iocb, ret);
321
322	return ret;
323}
324EXPORT_SYMBOL_GPL(backing_file_splice_write);
325
326int backing_file_mmap(struct file *file, struct vm_area_struct *vma,
327		      struct backing_file_ctx *ctx)
328{
329	const struct cred *old_cred;
330	struct file *user_file = vma->vm_file;
331	int ret;
332
333	if (WARN_ON_ONCE(!(file->f_mode & FMODE_BACKING)))
334		return -EIO;
335
336	if (!file->f_op->mmap)
337		return -ENODEV;
338
339	vma_set_file(vma, file);
340
341	old_cred = override_creds_light(ctx->cred);
342	ret = call_mmap(vma->vm_file, vma);
343	revert_creds_light(old_cred);
344
345	if (ctx->accessed)
346		ctx->accessed(user_file);
347
348	return ret;
349}
350EXPORT_SYMBOL_GPL(backing_file_mmap);
351
352static int __init backing_aio_init(void)
353{
354	backing_aio_cachep = KMEM_CACHE(backing_aio, SLAB_HWCACHE_ALIGN);
355	if (!backing_aio_cachep)
356		return -ENOMEM;
357
358	return 0;
359}
360fs_initcall(backing_aio_init);