1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Copyright (C) 2017 Red Hat, Inc.
  4 */
  5
  6#include "fuse_i.h"
  7
  8#include <linux/uio.h>
  9#include <linux/compat.h>
 10#include <linux/fileattr.h>
 11
 12/*
 13 * CUSE servers compiled on 32bit broke on 64bit kernels because the
 14 * ABI was defined to be 'struct iovec' which is different on 32bit
 15 * and 64bit.  Fortunately we can determine which structure the server
 16 * used from the size of the reply.
 17 */
 18static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
 19				     size_t transferred, unsigned count,
 20				     bool is_compat)
 21{
 22#ifdef CONFIG_COMPAT
 23	if (count * sizeof(struct compat_iovec) == transferred) {
 24		struct compat_iovec *ciov = src;
 25		unsigned i;
 26
 27		/*
 28		 * With this interface a 32bit server cannot support
 29		 * non-compat (i.e. ones coming from 64bit apps) ioctl
 30		 * requests
 31		 */
 32		if (!is_compat)
 33			return -EINVAL;
 34
 35		for (i = 0; i < count; i++) {
 36			dst[i].iov_base = compat_ptr(ciov[i].iov_base);
 37			dst[i].iov_len = ciov[i].iov_len;
 38		}
 39		return 0;
 40	}
 41#endif
 42
 43	if (count * sizeof(struct iovec) != transferred)
 44		return -EIO;
 45
 46	memcpy(dst, src, transferred);
 47	return 0;
 48}
 49
 50/* Make sure iov_length() won't overflow */
 51static int fuse_verify_ioctl_iov(struct fuse_conn *fc, struct iovec *iov,
 52				 size_t count)
 53{
 54	size_t n;
 55	u32 max = fc->max_pages << PAGE_SHIFT;
 56
 57	for (n = 0; n < count; n++, iov++) {
 58		if (iov->iov_len > (size_t) max)
 59			return -ENOMEM;
 60		max -= iov->iov_len;
 61	}
 62	return 0;
 63}
 64
 65static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
 66				 void *src, size_t transferred, unsigned count,
 67				 bool is_compat)
 68{
 69	unsigned i;
 70	struct fuse_ioctl_iovec *fiov = src;
 71
 72	if (fc->minor < 16) {
 73		return fuse_copy_ioctl_iovec_old(dst, src, transferred,
 74						 count, is_compat);
 75	}
 76
 77	if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
 78		return -EIO;
 79
 80	for (i = 0; i < count; i++) {
 81		/* Did the server supply an inappropriate value? */
 82		if (fiov[i].base != (unsigned long) fiov[i].base ||
 83		    fiov[i].len != (unsigned long) fiov[i].len)
 84			return -EIO;
 85
 86		dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
 87		dst[i].iov_len = (size_t) fiov[i].len;
 88
 89#ifdef CONFIG_COMPAT
 90		if (is_compat &&
 91		    (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
 92		     (compat_size_t) dst[i].iov_len != fiov[i].len))
 93			return -EIO;
 94#endif
 95	}
 96
 97	return 0;
 98}
 99
100
101/*
102 * For ioctls, there is no generic way to determine how much memory
103 * needs to be read and/or written.  Furthermore, ioctls are allowed
104 * to dereference the passed pointer, so the parameter requires deep
105 * copying but FUSE has no idea whatsoever about what to copy in or
106 * out.
107 *
108 * This is solved by allowing FUSE server to retry ioctl with
109 * necessary in/out iovecs.  Let's assume the ioctl implementation
110 * needs to read in the following structure.
111 *
112 * struct a {
113 *	char	*buf;
114 *	size_t	buflen;
115 * }
116 *
117 * On the first callout to FUSE server, inarg->in_size and
118 * inarg->out_size will be NULL; then, the server completes the ioctl
119 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
120 * the actual iov array to
121 *
122 * { { .iov_base = inarg.arg,	.iov_len = sizeof(struct a) } }
123 *
124 * which tells FUSE to copy in the requested area and retry the ioctl.
125 * On the second round, the server has access to the structure and
126 * from that it can tell what to look for next, so on the invocation,
127 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
128 *
129 * { { .iov_base = inarg.arg,	.iov_len = sizeof(struct a)	},
130 *   { .iov_base = a.buf,	.iov_len = a.buflen		} }
131 *
132 * FUSE will copy both struct a and the pointed buffer from the
133 * process doing the ioctl and retry ioctl with both struct a and the
134 * buffer.
135 *
136 * This time, FUSE server has everything it needs and completes ioctl
137 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
138 *
139 * Copying data out works the same way.
140 *
141 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
142 * automatically initializes in and out iovs by decoding @cmd with
143 * _IOC_* macros and the server is not allowed to request RETRY.  This
144 * limits ioctl data transfers to well-formed ioctls and is the forced
145 * behavior for all FUSE servers.
146 */
147long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
148		   unsigned int flags)
149{
150	struct fuse_file *ff = file->private_data;
151	struct fuse_mount *fm = ff->fm;
152	struct fuse_ioctl_in inarg = {
153		.fh = ff->fh,
154		.cmd = cmd,
155		.arg = arg,
156		.flags = flags
157	};
158	struct fuse_ioctl_out outarg;
159	struct iovec *iov_page = NULL;
160	struct iovec *in_iov = NULL, *out_iov = NULL;
161	unsigned int in_iovs = 0, out_iovs = 0, max_pages;
162	size_t in_size, out_size, c;
163	ssize_t transferred;
164	int err, i;
165	struct iov_iter ii;
166	struct fuse_args_pages ap = {};
167
168#if BITS_PER_LONG == 32
169	inarg.flags |= FUSE_IOCTL_32BIT;
170#else
171	if (flags & FUSE_IOCTL_COMPAT) {
172		inarg.flags |= FUSE_IOCTL_32BIT;
173#ifdef CONFIG_X86_X32
174		if (in_x32_syscall())
175			inarg.flags |= FUSE_IOCTL_COMPAT_X32;
176#endif
177	}
178#endif
179
180	/* assume all the iovs returned by client always fits in a page */
181	BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
182
183	err = -ENOMEM;
184	ap.pages = fuse_pages_alloc(fm->fc->max_pages, GFP_KERNEL, &ap.descs);
185	iov_page = (struct iovec *) __get_free_page(GFP_KERNEL);
186	if (!ap.pages || !iov_page)
187		goto out;
188
189	fuse_page_descs_length_init(ap.descs, 0, fm->fc->max_pages);
190
191	/*
192	 * If restricted, initialize IO parameters as encoded in @cmd.
193	 * RETRY from server is not allowed.
194	 */
195	if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
196		struct iovec *iov = iov_page;
197
198		iov->iov_base = (void __user *)arg;
199		iov->iov_len = _IOC_SIZE(cmd);
200
201		if (_IOC_DIR(cmd) & _IOC_WRITE) {
202			in_iov = iov;
203			in_iovs = 1;
204		}
205
206		if (_IOC_DIR(cmd) & _IOC_READ) {
207			out_iov = iov;
208			out_iovs = 1;
209		}
210	}
211
212 retry:
213	inarg.in_size = in_size = iov_length(in_iov, in_iovs);
214	inarg.out_size = out_size = iov_length(out_iov, out_iovs);
215
216	/*
217	 * Out data can be used either for actual out data or iovs,
218	 * make sure there always is at least one page.
219	 */
220	out_size = max_t(size_t, out_size, PAGE_SIZE);
221	max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
222
223	/* make sure there are enough buffer pages and init request with them */
224	err = -ENOMEM;
225	if (max_pages > fm->fc->max_pages)
226		goto out;
227	while (ap.num_pages < max_pages) {
228		ap.pages[ap.num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
229		if (!ap.pages[ap.num_pages])
230			goto out;
231		ap.num_pages++;
232	}
233
234
235	/* okay, let's send it to the client */
236	ap.args.opcode = FUSE_IOCTL;
237	ap.args.nodeid = ff->nodeid;
238	ap.args.in_numargs = 1;
239	ap.args.in_args[0].size = sizeof(inarg);
240	ap.args.in_args[0].value = &inarg;
241	if (in_size) {
242		ap.args.in_numargs++;
243		ap.args.in_args[1].size = in_size;
244		ap.args.in_pages = true;
245
246		err = -EFAULT;
247		iov_iter_init(&ii, WRITE, in_iov, in_iovs, in_size);
248		for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= ap.num_pages); i++) {
249			c = copy_page_from_iter(ap.pages[i], 0, PAGE_SIZE, &ii);
250			if (c != PAGE_SIZE && iov_iter_count(&ii))
251				goto out;
252		}
253	}
254
255	ap.args.out_numargs = 2;
256	ap.args.out_args[0].size = sizeof(outarg);
257	ap.args.out_args[0].value = &outarg;
258	ap.args.out_args[1].size = out_size;
259	ap.args.out_pages = true;
260	ap.args.out_argvar = true;
261
262	transferred = fuse_simple_request(fm, &ap.args);
263	err = transferred;
264	if (transferred < 0)
265		goto out;
266
267	/* did it ask for retry? */
268	if (outarg.flags & FUSE_IOCTL_RETRY) {
269		void *vaddr;
270
271		/* no retry if in restricted mode */
272		err = -EIO;
273		if (!(flags & FUSE_IOCTL_UNRESTRICTED))
274			goto out;
275
276		in_iovs = outarg.in_iovs;
277		out_iovs = outarg.out_iovs;
278
279		/*
280		 * Make sure things are in boundary, separate checks
281		 * are to protect against overflow.
282		 */
283		err = -ENOMEM;
284		if (in_iovs > FUSE_IOCTL_MAX_IOV ||
285		    out_iovs > FUSE_IOCTL_MAX_IOV ||
286		    in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
287			goto out;
288
289		vaddr = kmap_atomic(ap.pages[0]);
290		err = fuse_copy_ioctl_iovec(fm->fc, iov_page, vaddr,
291					    transferred, in_iovs + out_iovs,
292					    (flags & FUSE_IOCTL_COMPAT) != 0);
293		kunmap_atomic(vaddr);
294		if (err)
295			goto out;
296
297		in_iov = iov_page;
298		out_iov = in_iov + in_iovs;
299
300		err = fuse_verify_ioctl_iov(fm->fc, in_iov, in_iovs);
301		if (err)
302			goto out;
303
304		err = fuse_verify_ioctl_iov(fm->fc, out_iov, out_iovs);
305		if (err)
306			goto out;
307
308		goto retry;
309	}
310
311	err = -EIO;
312	if (transferred > inarg.out_size)
313		goto out;
314
315	err = -EFAULT;
316	iov_iter_init(&ii, READ, out_iov, out_iovs, transferred);
317	for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= ap.num_pages); i++) {
318		c = copy_page_to_iter(ap.pages[i], 0, PAGE_SIZE, &ii);
319		if (c != PAGE_SIZE && iov_iter_count(&ii))
320			goto out;
321	}
322	err = 0;
323 out:
324	free_page((unsigned long) iov_page);
325	while (ap.num_pages)
326		__free_page(ap.pages[--ap.num_pages]);
327	kfree(ap.pages);
328
329	return err ? err : outarg.result;
330}
331EXPORT_SYMBOL_GPL(fuse_do_ioctl);
332
333long fuse_ioctl_common(struct file *file, unsigned int cmd,
334		       unsigned long arg, unsigned int flags)
335{
336	struct inode *inode = file_inode(file);
337	struct fuse_conn *fc = get_fuse_conn(inode);
338
339	if (!fuse_allow_current_process(fc))
340		return -EACCES;
341
342	if (fuse_is_bad(inode))
343		return -EIO;
344
345	return fuse_do_ioctl(file, cmd, arg, flags);
346}
347
348long fuse_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
349{
350	return fuse_ioctl_common(file, cmd, arg, 0);
351}
352
353long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
354			    unsigned long arg)
355{
356	return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
357}
358
359static int fuse_priv_ioctl(struct inode *inode, struct fuse_file *ff,
360			   unsigned int cmd, void *ptr, size_t size)
361{
362	struct fuse_mount *fm = ff->fm;
363	struct fuse_ioctl_in inarg;
364	struct fuse_ioctl_out outarg;
365	FUSE_ARGS(args);
366	int err;
367
368	memset(&inarg, 0, sizeof(inarg));
369	inarg.fh = ff->fh;
370	inarg.cmd = cmd;
371
372#if BITS_PER_LONG == 32
373	inarg.flags |= FUSE_IOCTL_32BIT;
374#endif
375	if (S_ISDIR(inode->i_mode))
376		inarg.flags |= FUSE_IOCTL_DIR;
377
378	if (_IOC_DIR(cmd) & _IOC_READ)
379		inarg.out_size = size;
380	if (_IOC_DIR(cmd) & _IOC_WRITE)
381		inarg.in_size = size;
382
383	args.opcode = FUSE_IOCTL;
384	args.nodeid = ff->nodeid;
385	args.in_numargs = 2;
386	args.in_args[0].size = sizeof(inarg);
387	args.in_args[0].value = &inarg;
388	args.in_args[1].size = inarg.in_size;
389	args.in_args[1].value = ptr;
390	args.out_numargs = 2;
391	args.out_args[0].size = sizeof(outarg);
392	args.out_args[0].value = &outarg;
393	args.out_args[1].size = inarg.out_size;
394	args.out_args[1].value = ptr;
395
396	err = fuse_simple_request(fm, &args);
397	if (!err && outarg.flags & FUSE_IOCTL_RETRY)
398		err = -EIO;
399
400	return err;
401}
402
403static struct fuse_file *fuse_priv_ioctl_prepare(struct inode *inode)
404{
405	struct fuse_mount *fm = get_fuse_mount(inode);
406	bool isdir = S_ISDIR(inode->i_mode);
407
408	if (!S_ISREG(inode->i_mode) && !isdir)
409		return ERR_PTR(-ENOTTY);
410
411	return fuse_file_open(fm, get_node_id(inode), O_RDONLY, isdir);
412}
413
414static void fuse_priv_ioctl_cleanup(struct inode *inode, struct fuse_file *ff)
415{
416	fuse_file_release(inode, ff, O_RDONLY, NULL, S_ISDIR(inode->i_mode));
417}
418
419int fuse_fileattr_get(struct dentry *dentry, struct fileattr *fa)
420{
421	struct inode *inode = d_inode(dentry);
422	struct fuse_file *ff;
423	unsigned int flags;
424	struct fsxattr xfa;
425	int err;
426
427	ff = fuse_priv_ioctl_prepare(inode);
428	if (IS_ERR(ff))
429		return PTR_ERR(ff);
430
431	if (fa->flags_valid) {
432		err = fuse_priv_ioctl(inode, ff, FS_IOC_GETFLAGS,
433				      &flags, sizeof(flags));
434		if (err)
435			goto cleanup;
436
437		fileattr_fill_flags(fa, flags);
438	} else {
439		err = fuse_priv_ioctl(inode, ff, FS_IOC_FSGETXATTR,
440				      &xfa, sizeof(xfa));
441		if (err)
442			goto cleanup;
443
444		fileattr_fill_xflags(fa, xfa.fsx_xflags);
445		fa->fsx_extsize = xfa.fsx_extsize;
446		fa->fsx_nextents = xfa.fsx_nextents;
447		fa->fsx_projid = xfa.fsx_projid;
448		fa->fsx_cowextsize = xfa.fsx_cowextsize;
449	}
450cleanup:
451	fuse_priv_ioctl_cleanup(inode, ff);
452
453	return err;
454}
455
456int fuse_fileattr_set(struct user_namespace *mnt_userns,
457		      struct dentry *dentry, struct fileattr *fa)
458{
459	struct inode *inode = d_inode(dentry);
460	struct fuse_file *ff;
461	unsigned int flags = fa->flags;
462	struct fsxattr xfa;
463	int err;
464
465	ff = fuse_priv_ioctl_prepare(inode);
466	if (IS_ERR(ff))
467		return PTR_ERR(ff);
468
469	if (fa->flags_valid) {
470		err = fuse_priv_ioctl(inode, ff, FS_IOC_SETFLAGS,
471				      &flags, sizeof(flags));
472		if (err)
473			goto cleanup;
474	} else {
475		memset(&xfa, 0, sizeof(xfa));
476		xfa.fsx_xflags = fa->fsx_xflags;
477		xfa.fsx_extsize = fa->fsx_extsize;
478		xfa.fsx_nextents = fa->fsx_nextents;
479		xfa.fsx_projid = fa->fsx_projid;
480		xfa.fsx_cowextsize = fa->fsx_cowextsize;
481
482		err = fuse_priv_ioctl(inode, ff, FS_IOC_FSSETXATTR,
483				      &xfa, sizeof(xfa));
484	}
485
486cleanup:
487	fuse_priv_ioctl_cleanup(inode, ff);
488
489	return err;
490}