xref: /openbmc/linux/fs/fuse/ioctl.c (revision 47aab533)
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  static ssize_t fuse_send_ioctl(struct fuse_mount *fm, struct fuse_args *args)
13  {
14  	ssize_t ret = fuse_simple_request(fm, args);
15  
16  	/* Translate ENOSYS, which shouldn't be returned from fs */
17  	if (ret == -ENOSYS)
18  		ret = -ENOTTY;
19  
20  	return ret;
21  }
22  
23  /*
24   * CUSE servers compiled on 32bit broke on 64bit kernels because the
25   * ABI was defined to be 'struct iovec' which is different on 32bit
26   * and 64bit.  Fortunately we can determine which structure the server
27   * used from the size of the reply.
28   */
29  static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
30  				     size_t transferred, unsigned count,
31  				     bool is_compat)
32  {
33  #ifdef CONFIG_COMPAT
34  	if (count * sizeof(struct compat_iovec) == transferred) {
35  		struct compat_iovec *ciov = src;
36  		unsigned i;
37  
38  		/*
39  		 * With this interface a 32bit server cannot support
40  		 * non-compat (i.e. ones coming from 64bit apps) ioctl
41  		 * requests
42  		 */
43  		if (!is_compat)
44  			return -EINVAL;
45  
46  		for (i = 0; i < count; i++) {
47  			dst[i].iov_base = compat_ptr(ciov[i].iov_base);
48  			dst[i].iov_len = ciov[i].iov_len;
49  		}
50  		return 0;
51  	}
52  #endif
53  
54  	if (count * sizeof(struct iovec) != transferred)
55  		return -EIO;
56  
57  	memcpy(dst, src, transferred);
58  	return 0;
59  }
60  
61  /* Make sure iov_length() won't overflow */
62  static int fuse_verify_ioctl_iov(struct fuse_conn *fc, struct iovec *iov,
63  				 size_t count)
64  {
65  	size_t n;
66  	u32 max = fc->max_pages << PAGE_SHIFT;
67  
68  	for (n = 0; n < count; n++, iov++) {
69  		if (iov->iov_len > (size_t) max)
70  			return -ENOMEM;
71  		max -= iov->iov_len;
72  	}
73  	return 0;
74  }
75  
76  static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
77  				 void *src, size_t transferred, unsigned count,
78  				 bool is_compat)
79  {
80  	unsigned i;
81  	struct fuse_ioctl_iovec *fiov = src;
82  
83  	if (fc->minor < 16) {
84  		return fuse_copy_ioctl_iovec_old(dst, src, transferred,
85  						 count, is_compat);
86  	}
87  
88  	if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
89  		return -EIO;
90  
91  	for (i = 0; i < count; i++) {
92  		/* Did the server supply an inappropriate value? */
93  		if (fiov[i].base != (unsigned long) fiov[i].base ||
94  		    fiov[i].len != (unsigned long) fiov[i].len)
95  			return -EIO;
96  
97  		dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
98  		dst[i].iov_len = (size_t) fiov[i].len;
99  
100  #ifdef CONFIG_COMPAT
101  		if (is_compat &&
102  		    (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
103  		     (compat_size_t) dst[i].iov_len != fiov[i].len))
104  			return -EIO;
105  #endif
106  	}
107  
108  	return 0;
109  }
110  
111  
112  /*
113   * For ioctls, there is no generic way to determine how much memory
114   * needs to be read and/or written.  Furthermore, ioctls are allowed
115   * to dereference the passed pointer, so the parameter requires deep
116   * copying but FUSE has no idea whatsoever about what to copy in or
117   * out.
118   *
119   * This is solved by allowing FUSE server to retry ioctl with
120   * necessary in/out iovecs.  Let's assume the ioctl implementation
121   * needs to read in the following structure.
122   *
123   * struct a {
124   *	char	*buf;
125   *	size_t	buflen;
126   * }
127   *
128   * On the first callout to FUSE server, inarg->in_size and
129   * inarg->out_size will be NULL; then, the server completes the ioctl
130   * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
131   * the actual iov array to
132   *
133   * { { .iov_base = inarg.arg,	.iov_len = sizeof(struct a) } }
134   *
135   * which tells FUSE to copy in the requested area and retry the ioctl.
136   * On the second round, the server has access to the structure and
137   * from that it can tell what to look for next, so on the invocation,
138   * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
139   *
140   * { { .iov_base = inarg.arg,	.iov_len = sizeof(struct a)	},
141   *   { .iov_base = a.buf,	.iov_len = a.buflen		} }
142   *
143   * FUSE will copy both struct a and the pointed buffer from the
144   * process doing the ioctl and retry ioctl with both struct a and the
145   * buffer.
146   *
147   * This time, FUSE server has everything it needs and completes ioctl
148   * without FUSE_IOCTL_RETRY which finishes the ioctl call.
149   *
150   * Copying data out works the same way.
151   *
152   * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
153   * automatically initializes in and out iovs by decoding @cmd with
154   * _IOC_* macros and the server is not allowed to request RETRY.  This
155   * limits ioctl data transfers to well-formed ioctls and is the forced
156   * behavior for all FUSE servers.
157   */
158  long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
159  		   unsigned int flags)
160  {
161  	struct fuse_file *ff = file->private_data;
162  	struct fuse_mount *fm = ff->fm;
163  	struct fuse_ioctl_in inarg = {
164  		.fh = ff->fh,
165  		.cmd = cmd,
166  		.arg = arg,
167  		.flags = flags
168  	};
169  	struct fuse_ioctl_out outarg;
170  	struct iovec *iov_page = NULL;
171  	struct iovec *in_iov = NULL, *out_iov = NULL;
172  	unsigned int in_iovs = 0, out_iovs = 0, max_pages;
173  	size_t in_size, out_size, c;
174  	ssize_t transferred;
175  	int err, i;
176  	struct iov_iter ii;
177  	struct fuse_args_pages ap = {};
178  
179  #if BITS_PER_LONG == 32
180  	inarg.flags |= FUSE_IOCTL_32BIT;
181  #else
182  	if (flags & FUSE_IOCTL_COMPAT) {
183  		inarg.flags |= FUSE_IOCTL_32BIT;
184  #ifdef CONFIG_X86_X32_ABI
185  		if (in_x32_syscall())
186  			inarg.flags |= FUSE_IOCTL_COMPAT_X32;
187  #endif
188  	}
189  #endif
190  
191  	/* assume all the iovs returned by client always fits in a page */
192  	BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
193  
194  	err = -ENOMEM;
195  	ap.pages = fuse_pages_alloc(fm->fc->max_pages, GFP_KERNEL, &ap.descs);
196  	iov_page = (struct iovec *) __get_free_page(GFP_KERNEL);
197  	if (!ap.pages || !iov_page)
198  		goto out;
199  
200  	fuse_page_descs_length_init(ap.descs, 0, fm->fc->max_pages);
201  
202  	/*
203  	 * If restricted, initialize IO parameters as encoded in @cmd.
204  	 * RETRY from server is not allowed.
205  	 */
206  	if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
207  		struct iovec *iov = iov_page;
208  
209  		iov->iov_base = (void __user *)arg;
210  		iov->iov_len = _IOC_SIZE(cmd);
211  
212  		if (_IOC_DIR(cmd) & _IOC_WRITE) {
213  			in_iov = iov;
214  			in_iovs = 1;
215  		}
216  
217  		if (_IOC_DIR(cmd) & _IOC_READ) {
218  			out_iov = iov;
219  			out_iovs = 1;
220  		}
221  	}
222  
223   retry:
224  	inarg.in_size = in_size = iov_length(in_iov, in_iovs);
225  	inarg.out_size = out_size = iov_length(out_iov, out_iovs);
226  
227  	/*
228  	 * Out data can be used either for actual out data or iovs,
229  	 * make sure there always is at least one page.
230  	 */
231  	out_size = max_t(size_t, out_size, PAGE_SIZE);
232  	max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
233  
234  	/* make sure there are enough buffer pages and init request with them */
235  	err = -ENOMEM;
236  	if (max_pages > fm->fc->max_pages)
237  		goto out;
238  	while (ap.num_pages < max_pages) {
239  		ap.pages[ap.num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
240  		if (!ap.pages[ap.num_pages])
241  			goto out;
242  		ap.num_pages++;
243  	}
244  
245  
246  	/* okay, let's send it to the client */
247  	ap.args.opcode = FUSE_IOCTL;
248  	ap.args.nodeid = ff->nodeid;
249  	ap.args.in_numargs = 1;
250  	ap.args.in_args[0].size = sizeof(inarg);
251  	ap.args.in_args[0].value = &inarg;
252  	if (in_size) {
253  		ap.args.in_numargs++;
254  		ap.args.in_args[1].size = in_size;
255  		ap.args.in_pages = true;
256  
257  		err = -EFAULT;
258  		iov_iter_init(&ii, ITER_SOURCE, in_iov, in_iovs, in_size);
259  		for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= ap.num_pages); i++) {
260  			c = copy_page_from_iter(ap.pages[i], 0, PAGE_SIZE, &ii);
261  			if (c != PAGE_SIZE && iov_iter_count(&ii))
262  				goto out;
263  		}
264  	}
265  
266  	ap.args.out_numargs = 2;
267  	ap.args.out_args[0].size = sizeof(outarg);
268  	ap.args.out_args[0].value = &outarg;
269  	ap.args.out_args[1].size = out_size;
270  	ap.args.out_pages = true;
271  	ap.args.out_argvar = true;
272  
273  	transferred = fuse_send_ioctl(fm, &ap.args);
274  	err = transferred;
275  	if (transferred < 0)
276  		goto out;
277  
278  	/* did it ask for retry? */
279  	if (outarg.flags & FUSE_IOCTL_RETRY) {
280  		void *vaddr;
281  
282  		/* no retry if in restricted mode */
283  		err = -EIO;
284  		if (!(flags & FUSE_IOCTL_UNRESTRICTED))
285  			goto out;
286  
287  		in_iovs = outarg.in_iovs;
288  		out_iovs = outarg.out_iovs;
289  
290  		/*
291  		 * Make sure things are in boundary, separate checks
292  		 * are to protect against overflow.
293  		 */
294  		err = -ENOMEM;
295  		if (in_iovs > FUSE_IOCTL_MAX_IOV ||
296  		    out_iovs > FUSE_IOCTL_MAX_IOV ||
297  		    in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
298  			goto out;
299  
300  		vaddr = kmap_local_page(ap.pages[0]);
301  		err = fuse_copy_ioctl_iovec(fm->fc, iov_page, vaddr,
302  					    transferred, in_iovs + out_iovs,
303  					    (flags & FUSE_IOCTL_COMPAT) != 0);
304  		kunmap_local(vaddr);
305  		if (err)
306  			goto out;
307  
308  		in_iov = iov_page;
309  		out_iov = in_iov + in_iovs;
310  
311  		err = fuse_verify_ioctl_iov(fm->fc, in_iov, in_iovs);
312  		if (err)
313  			goto out;
314  
315  		err = fuse_verify_ioctl_iov(fm->fc, out_iov, out_iovs);
316  		if (err)
317  			goto out;
318  
319  		goto retry;
320  	}
321  
322  	err = -EIO;
323  	if (transferred > inarg.out_size)
324  		goto out;
325  
326  	err = -EFAULT;
327  	iov_iter_init(&ii, ITER_DEST, out_iov, out_iovs, transferred);
328  	for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= ap.num_pages); i++) {
329  		c = copy_page_to_iter(ap.pages[i], 0, PAGE_SIZE, &ii);
330  		if (c != PAGE_SIZE && iov_iter_count(&ii))
331  			goto out;
332  	}
333  	err = 0;
334   out:
335  	free_page((unsigned long) iov_page);
336  	while (ap.num_pages)
337  		__free_page(ap.pages[--ap.num_pages]);
338  	kfree(ap.pages);
339  
340  	return err ? err : outarg.result;
341  }
342  EXPORT_SYMBOL_GPL(fuse_do_ioctl);
343  
344  long fuse_ioctl_common(struct file *file, unsigned int cmd,
345  		       unsigned long arg, unsigned int flags)
346  {
347  	struct inode *inode = file_inode(file);
348  	struct fuse_conn *fc = get_fuse_conn(inode);
349  
350  	if (!fuse_allow_current_process(fc))
351  		return -EACCES;
352  
353  	if (fuse_is_bad(inode))
354  		return -EIO;
355  
356  	return fuse_do_ioctl(file, cmd, arg, flags);
357  }
358  
359  long fuse_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
360  {
361  	return fuse_ioctl_common(file, cmd, arg, 0);
362  }
363  
364  long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
365  			    unsigned long arg)
366  {
367  	return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
368  }
369  
370  static int fuse_priv_ioctl(struct inode *inode, struct fuse_file *ff,
371  			   unsigned int cmd, void *ptr, size_t size)
372  {
373  	struct fuse_mount *fm = ff->fm;
374  	struct fuse_ioctl_in inarg;
375  	struct fuse_ioctl_out outarg;
376  	FUSE_ARGS(args);
377  	int err;
378  
379  	memset(&inarg, 0, sizeof(inarg));
380  	inarg.fh = ff->fh;
381  	inarg.cmd = cmd;
382  
383  #if BITS_PER_LONG == 32
384  	inarg.flags |= FUSE_IOCTL_32BIT;
385  #endif
386  	if (S_ISDIR(inode->i_mode))
387  		inarg.flags |= FUSE_IOCTL_DIR;
388  
389  	if (_IOC_DIR(cmd) & _IOC_READ)
390  		inarg.out_size = size;
391  	if (_IOC_DIR(cmd) & _IOC_WRITE)
392  		inarg.in_size = size;
393  
394  	args.opcode = FUSE_IOCTL;
395  	args.nodeid = ff->nodeid;
396  	args.in_numargs = 2;
397  	args.in_args[0].size = sizeof(inarg);
398  	args.in_args[0].value = &inarg;
399  	args.in_args[1].size = inarg.in_size;
400  	args.in_args[1].value = ptr;
401  	args.out_numargs = 2;
402  	args.out_args[0].size = sizeof(outarg);
403  	args.out_args[0].value = &outarg;
404  	args.out_args[1].size = inarg.out_size;
405  	args.out_args[1].value = ptr;
406  
407  	err = fuse_send_ioctl(fm, &args);
408  	if (!err) {
409  		if (outarg.result < 0)
410  			err = outarg.result;
411  		else if (outarg.flags & FUSE_IOCTL_RETRY)
412  			err = -EIO;
413  	}
414  	return err;
415  }
416  
417  static struct fuse_file *fuse_priv_ioctl_prepare(struct inode *inode)
418  {
419  	struct fuse_mount *fm = get_fuse_mount(inode);
420  	bool isdir = S_ISDIR(inode->i_mode);
421  
422  	if (!fuse_allow_current_process(fm->fc))
423  		return ERR_PTR(-EACCES);
424  
425  	if (fuse_is_bad(inode))
426  		return ERR_PTR(-EIO);
427  
428  	if (!S_ISREG(inode->i_mode) && !isdir)
429  		return ERR_PTR(-ENOTTY);
430  
431  	return fuse_file_open(fm, get_node_id(inode), O_RDONLY, isdir);
432  }
433  
434  static void fuse_priv_ioctl_cleanup(struct inode *inode, struct fuse_file *ff)
435  {
436  	fuse_file_release(inode, ff, O_RDONLY, NULL, S_ISDIR(inode->i_mode));
437  }
438  
439  int fuse_fileattr_get(struct dentry *dentry, struct fileattr *fa)
440  {
441  	struct inode *inode = d_inode(dentry);
442  	struct fuse_file *ff;
443  	unsigned int flags;
444  	struct fsxattr xfa;
445  	int err;
446  
447  	ff = fuse_priv_ioctl_prepare(inode);
448  	if (IS_ERR(ff))
449  		return PTR_ERR(ff);
450  
451  	if (fa->flags_valid) {
452  		err = fuse_priv_ioctl(inode, ff, FS_IOC_GETFLAGS,
453  				      &flags, sizeof(flags));
454  		if (err)
455  			goto cleanup;
456  
457  		fileattr_fill_flags(fa, flags);
458  	} else {
459  		err = fuse_priv_ioctl(inode, ff, FS_IOC_FSGETXATTR,
460  				      &xfa, sizeof(xfa));
461  		if (err)
462  			goto cleanup;
463  
464  		fileattr_fill_xflags(fa, xfa.fsx_xflags);
465  		fa->fsx_extsize = xfa.fsx_extsize;
466  		fa->fsx_nextents = xfa.fsx_nextents;
467  		fa->fsx_projid = xfa.fsx_projid;
468  		fa->fsx_cowextsize = xfa.fsx_cowextsize;
469  	}
470  cleanup:
471  	fuse_priv_ioctl_cleanup(inode, ff);
472  
473  	return err;
474  }
475  
476  int fuse_fileattr_set(struct mnt_idmap *idmap,
477  		      struct dentry *dentry, struct fileattr *fa)
478  {
479  	struct inode *inode = d_inode(dentry);
480  	struct fuse_file *ff;
481  	unsigned int flags = fa->flags;
482  	struct fsxattr xfa;
483  	int err;
484  
485  	ff = fuse_priv_ioctl_prepare(inode);
486  	if (IS_ERR(ff))
487  		return PTR_ERR(ff);
488  
489  	if (fa->flags_valid) {
490  		err = fuse_priv_ioctl(inode, ff, FS_IOC_SETFLAGS,
491  				      &flags, sizeof(flags));
492  		if (err)
493  			goto cleanup;
494  	} else {
495  		memset(&xfa, 0, sizeof(xfa));
496  		xfa.fsx_xflags = fa->fsx_xflags;
497  		xfa.fsx_extsize = fa->fsx_extsize;
498  		xfa.fsx_nextents = fa->fsx_nextents;
499  		xfa.fsx_projid = fa->fsx_projid;
500  		xfa.fsx_cowextsize = fa->fsx_cowextsize;
501  
502  		err = fuse_priv_ioctl(inode, ff, FS_IOC_FSSETXATTR,
503  				      &xfa, sizeof(xfa));
504  	}
505  
506  cleanup:
507  	fuse_priv_ioctl_cleanup(inode, ff);
508  
509  	return err;
510  }
511