projects/linux·blob master · refs

577 lines · 14.4 KiB

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#include <linux/fsverity.h>
12
13#include <linux/slab.h>
14#define FUSE_VERITY_ENABLE_ARG_MAX_PAGES 256
15
16static ssize_t fuse_send_ioctl(struct fuse_mount *fm, struct fuse_args *args,
17 struct fuse_ioctl_out *outarg)
18{
19 ssize_t ret;
20
21 args->out_args[0].size = sizeof(*outarg);
22 args->out_args[0].value = outarg;
23
24 ret = fuse_simple_request(fm, args);
25
26 /* Translate ENOSYS, which shouldn't be returned from fs */
27 if (ret == -ENOSYS)
28 ret = -ENOTTY;
29
30 if (ret >= 0 && outarg->result == -ENOSYS)
31 outarg->result = -ENOTTY;
32
33 return ret;
34}
35
36/*
37 * CUSE servers compiled on 32bit broke on 64bit kernels because the
38 * ABI was defined to be 'struct iovec' which is different on 32bit
39 * and 64bit. Fortunately we can determine which structure the server
40 * used from the size of the reply.
41 */
42static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
43 size_t transferred, unsigned count,
44 bool is_compat)
45{
46#ifdef CONFIG_COMPAT
47 if (count * sizeof(struct compat_iovec) == transferred) {
48 struct compat_iovec *ciov = src;
49 unsigned i;
50
51 /*
52 * With this interface a 32bit server cannot support
53 * non-compat (i.e. ones coming from 64bit apps) ioctl
54 * requests
55 */
56 if (!is_compat)
57 return -EINVAL;
58
59 for (i = 0; i < count; i++) {
60 dst[i].iov_base = compat_ptr(ciov[i].iov_base);
61 dst[i].iov_len = ciov[i].iov_len;
62 }
63 return 0;
64 }
65#endif
66
67 if (count * sizeof(struct iovec) != transferred)
68 return -EIO;
69
70 memcpy(dst, src, transferred);
71 return 0;
72}
73
74/* Make sure iov_length() won't overflow */
75static int fuse_verify_ioctl_iov(struct fuse_conn *fc, struct iovec *iov,
76 size_t count)
77{
78 size_t n;
79 u32 max = fc->max_pages << PAGE_SHIFT;
80
81 for (n = 0; n < count; n++, iov++) {
82 if (iov->iov_len > (size_t) max)
83 return -ENOMEM;
84 max -= iov->iov_len;
85 }
86 return 0;
87}
88
89static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
90 void *src, size_t transferred, unsigned count,
91 bool is_compat)
92{
93 unsigned i;
94 struct fuse_ioctl_iovec *fiov = src;
95
96 if (fc->minor < 16) {
97 return fuse_copy_ioctl_iovec_old(dst, src, transferred,
98 count, is_compat);
99 }
100
101 if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
102 return -EIO;
103
104 for (i = 0; i < count; i++) {
105 /* Did the server supply an inappropriate value? */
106 if (fiov[i].base != (unsigned long) fiov[i].base ||
107 fiov[i].len != (unsigned long) fiov[i].len)
108 return -EIO;
109
110 dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
111 dst[i].iov_len = (size_t) fiov[i].len;
112
113#ifdef CONFIG_COMPAT
114 if (is_compat &&
115 (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
116 (compat_size_t) dst[i].iov_len != fiov[i].len))
117 return -EIO;
118#endif
119 }
120
121 return 0;
122}
123
124/* For fs-verity, determine iov lengths from input */
125static int fuse_setup_measure_verity(unsigned long arg, struct iovec *iov)
126{
127 __u16 digest_size;
128 struct fsverity_digest __user *uarg = (void __user *)arg;
129
130 if (copy_from_user(&digest_size, &uarg->digest_size, sizeof(digest_size)))
131 return -EFAULT;
132
133 if (digest_size > SIZE_MAX - sizeof(struct fsverity_digest))
134 return -EINVAL;
135
136 iov->iov_len = sizeof(struct fsverity_digest) + digest_size;
137
138 return 0;
139}
140
141static int fuse_setup_enable_verity(unsigned long arg, struct iovec *iov,
142 unsigned int *in_iovs)
143{
144 struct fsverity_enable_arg enable;
145 struct fsverity_enable_arg __user *uarg = (void __user *)arg;
146 const __u32 max_buffer_len = FUSE_VERITY_ENABLE_ARG_MAX_PAGES * PAGE_SIZE;
147
148 if (copy_from_user(&enable, uarg, sizeof(enable)))
149 return -EFAULT;
150
151 if (enable.salt_size > max_buffer_len || enable.sig_size > max_buffer_len)
152 return -ENOMEM;
153
154 if (enable.salt_size > 0) {
155 iov++;
156 (*in_iovs)++;
157
158 iov->iov_base = u64_to_user_ptr(enable.salt_ptr);
159 iov->iov_len = enable.salt_size;
160 }
161
162 if (enable.sig_size > 0) {
163 iov++;
164 (*in_iovs)++;
165
166 iov->iov_base = u64_to_user_ptr(enable.sig_ptr);
167 iov->iov_len = enable.sig_size;
168 }
169 return 0;
170}
171
172/*
173 * For ioctls, there is no generic way to determine how much memory
174 * needs to be read and/or written. Furthermore, ioctls are allowed
175 * to dereference the passed pointer, so the parameter requires deep
176 * copying but FUSE has no idea whatsoever about what to copy in or
177 * out.
178 *
179 * This is solved by allowing FUSE server to retry ioctl with
180 * necessary in/out iovecs. Let's assume the ioctl implementation
181 * needs to read in the following structure.
182 *
183 * struct a {
184 * char *buf;
185 * size_t buflen;
186 * }
187 *
188 * On the first callout to FUSE server, inarg->in_size and
189 * inarg->out_size will be NULL; then, the server completes the ioctl
190 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
191 * the actual iov array to
192 *
193 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } }
194 *
195 * which tells FUSE to copy in the requested area and retry the ioctl.
196 * On the second round, the server has access to the structure and
197 * from that it can tell what to look for next, so on the invocation,
198 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
199 *
200 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) },
201 * { .iov_base = a.buf, .iov_len = a.buflen } }
202 *
203 * FUSE will copy both struct a and the pointed buffer from the
204 * process doing the ioctl and retry ioctl with both struct a and the
205 * buffer.
206 *
207 * This time, FUSE server has everything it needs and completes ioctl
208 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
209 *
210 * Copying data out works the same way.
211 *
212 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
213 * automatically initializes in and out iovs by decoding @cmd with
214 * _IOC_* macros and the server is not allowed to request RETRY. This
215 * limits ioctl data transfers to well-formed ioctls and is the forced
216 * behavior for all FUSE servers.
217 */
218long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
219 unsigned int flags)
220{
221 struct fuse_file *ff = file->private_data;
222 struct fuse_mount *fm = ff->fm;
223 struct fuse_ioctl_in inarg = {
224 .fh = ff->fh,
225 .cmd = cmd,
226 .arg = arg,
227 .flags = flags
228 };
229 struct fuse_ioctl_out outarg;
230 struct iovec *iov_page = NULL;
231 struct iovec *in_iov = NULL, *out_iov = NULL;
232 unsigned int in_iovs = 0, out_iovs = 0, max_pages;
233 size_t in_size, out_size, c;
234 ssize_t transferred;
235 int err, i;
236 struct iov_iter ii;
237 struct fuse_args_pages ap = {};
238
239#if BITS_PER_LONG == 32
240 inarg.flags |= FUSE_IOCTL_32BIT;
241#else
242 if (flags & FUSE_IOCTL_COMPAT) {
243 inarg.flags |= FUSE_IOCTL_32BIT;
244#ifdef CONFIG_X86_X32_ABI
245 if (in_x32_syscall())
246 inarg.flags |= FUSE_IOCTL_COMPAT_X32;
247#endif
248 }
249#endif
250
251 /* assume all the iovs returned by client always fits in a page */
252 BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
253
254 err = -ENOMEM;
255 ap.folios = fuse_folios_alloc(fm->fc->max_pages, GFP_KERNEL, &ap.descs);
256 iov_page = kmalloc(PAGE_SIZE, GFP_KERNEL);
257 if (!ap.folios || !iov_page)
258 goto out;
259
260 fuse_folio_descs_length_init(ap.descs, 0, fm->fc->max_pages);
261
262 /*
263 * If restricted, initialize IO parameters as encoded in @cmd.
264 * RETRY from server is not allowed.
265 */
266 if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
267 struct iovec *iov = iov_page;
268
269 iov->iov_base = (void __user *)arg;
270 iov->iov_len = _IOC_SIZE(cmd);
271
272 if (_IOC_DIR(cmd) & _IOC_WRITE) {
273 in_iov = iov;
274 in_iovs = 1;
275 }
276
277 if (_IOC_DIR(cmd) & _IOC_READ) {
278 out_iov = iov;
279 out_iovs = 1;
280 }
281
282 err = 0;
283 switch (cmd) {
284 case FS_IOC_MEASURE_VERITY:
285 err = fuse_setup_measure_verity(arg, iov);
286 break;
287 case FS_IOC_ENABLE_VERITY:
288 err = fuse_setup_enable_verity(arg, iov, &in_iovs);
289 break;
290 }
291 if (err)
292 goto out;
293 }
294
295 retry:
296 inarg.in_size = in_size = iov_length(in_iov, in_iovs);
297 inarg.out_size = out_size = iov_length(out_iov, out_iovs);
298
299 /*
300 * Out data can be used either for actual out data or iovs,
301 * make sure there always is at least one page.
302 */
303 out_size = max_t(size_t, out_size, PAGE_SIZE);
304 max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
305
306 /* make sure there are enough buffer pages and init request with them */
307 err = -ENOMEM;
308 if (max_pages > fm->fc->max_pages)
309 goto out;
310 while (ap.num_folios < max_pages) {
311 ap.folios[ap.num_folios] = folio_alloc(GFP_KERNEL | __GFP_HIGHMEM, 0);
312 if (!ap.folios[ap.num_folios])
313 goto out;
314 ap.num_folios++;
315 }
316
317 /* okay, let's send it to the client */
318 ap.args.opcode = FUSE_IOCTL;
319 ap.args.nodeid = ff->nodeid;
320 ap.args.in_numargs = 1;
321 ap.args.in_args[0].size = sizeof(inarg);
322 ap.args.in_args[0].value = &inarg;
323 if (in_size) {
324 ap.args.in_numargs++;
325 ap.args.in_args[1].size = in_size;
326 ap.args.in_pages = true;
327
328 err = -EFAULT;
329 iov_iter_init(&ii, ITER_SOURCE, in_iov, in_iovs, in_size);
330 for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= ap.num_folios); i++) {
331 c = copy_folio_from_iter(ap.folios[i], 0, PAGE_SIZE, &ii);
332 if (c != PAGE_SIZE && iov_iter_count(&ii))
333 goto out;
334 }
335 }
336
337 ap.args.out_numargs = 2;
338 ap.args.out_args[1].size = out_size;
339 ap.args.out_pages = true;
340 ap.args.out_argvar = true;
341
342 transferred = fuse_send_ioctl(fm, &ap.args, &outarg);
343 err = transferred;
344 if (transferred < 0)
345 goto out;
346
347 /* did it ask for retry? */
348 if (outarg.flags & FUSE_IOCTL_RETRY) {
349 void *vaddr;
350
351 /* no retry if in restricted mode */
352 err = -EIO;
353 if (!(flags & FUSE_IOCTL_UNRESTRICTED))
354 goto out;
355
356 in_iovs = outarg.in_iovs;
357 out_iovs = outarg.out_iovs;
358
359 /*
360 * Make sure things are in boundary, separate checks
361 * are to protect against overflow.
362 */
363 err = -ENOMEM;
364 if (in_iovs > FUSE_IOCTL_MAX_IOV ||
365 out_iovs > FUSE_IOCTL_MAX_IOV ||
366 in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
367 goto out;
368
369 vaddr = kmap_local_folio(ap.folios[0], 0);
370 err = fuse_copy_ioctl_iovec(fm->fc, iov_page, vaddr,
371 transferred, in_iovs + out_iovs,
372 (flags & FUSE_IOCTL_COMPAT) != 0);
373 kunmap_local(vaddr);
374 if (err)
375 goto out;
376
377 in_iov = iov_page;
378 out_iov = in_iov + in_iovs;
379
380 err = fuse_verify_ioctl_iov(fm->fc, in_iov, in_iovs);
381 if (err)
382 goto out;
383
384 err = fuse_verify_ioctl_iov(fm->fc, out_iov, out_iovs);
385 if (err)
386 goto out;
387
388 goto retry;
389 }
390
391 err = -EIO;
392 if (transferred > inarg.out_size)
393 goto out;
394
395 err = -EFAULT;
396 iov_iter_init(&ii, ITER_DEST, out_iov, out_iovs, transferred);
397 for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= ap.num_folios); i++) {
398 c = copy_folio_to_iter(ap.folios[i], 0, PAGE_SIZE, &ii);
399 if (c != PAGE_SIZE && iov_iter_count(&ii))
400 goto out;
401 }
402 err = 0;
403 out:
404 kfree(iov_page);
405 while (ap.num_folios)
406 folio_put(ap.folios[--ap.num_folios]);
407 kfree(ap.folios);
408
409 return err ? err : outarg.result;
410}
411EXPORT_SYMBOL_GPL(fuse_do_ioctl);
412
413long fuse_ioctl_common(struct file *file, unsigned int cmd,
414 unsigned long arg, unsigned int flags)
415{
416 struct inode *inode = file_inode(file);
417 struct fuse_conn *fc = get_fuse_conn(inode);
418
419 if (!fuse_allow_current_process(fc))
420 return -EACCES;
421
422 if (fuse_is_bad(inode))
423 return -EIO;
424
425 return fuse_do_ioctl(file, cmd, arg, flags);
426}
427
428long fuse_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
429{
430 return fuse_ioctl_common(file, cmd, arg, 0);
431}
432
433long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
434 unsigned long arg)
435{
436 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
437}
438
439static int fuse_priv_ioctl(struct inode *inode, struct fuse_file *ff,
440 unsigned int cmd, void *ptr, size_t size)
441{
442 struct fuse_mount *fm = ff->fm;
443 struct fuse_ioctl_in inarg;
444 struct fuse_ioctl_out outarg;
445 FUSE_ARGS(args);
446 int err;
447
448 memset(&inarg, 0, sizeof(inarg));
449 inarg.fh = ff->fh;
450 inarg.cmd = cmd;
451
452#if BITS_PER_LONG == 32
453 inarg.flags |= FUSE_IOCTL_32BIT;
454#endif
455 if (S_ISDIR(inode->i_mode))
456 inarg.flags |= FUSE_IOCTL_DIR;
457
458 if (_IOC_DIR(cmd) & _IOC_READ)
459 inarg.out_size = size;
460 if (_IOC_DIR(cmd) & _IOC_WRITE)
461 inarg.in_size = size;
462
463 args.opcode = FUSE_IOCTL;
464 args.nodeid = ff->nodeid;
465 args.in_numargs = 2;
466 args.in_args[0].size = sizeof(inarg);
467 args.in_args[0].value = &inarg;
468 args.in_args[1].size = inarg.in_size;
469 args.in_args[1].value = ptr;
470 args.out_numargs = 2;
471 args.out_args[1].size = inarg.out_size;
472 args.out_args[1].value = ptr;
473
474 err = fuse_send_ioctl(fm, &args, &outarg);
475 if (!err) {
476 if (outarg.result < 0)
477 err = outarg.result;
478 else if (outarg.flags & FUSE_IOCTL_RETRY)
479 err = -EIO;
480 }
481 return err;
482}
483
484static struct fuse_file *fuse_priv_ioctl_prepare(struct inode *inode)
485{
486 struct fuse_mount *fm = get_fuse_mount(inode);
487 bool isdir = S_ISDIR(inode->i_mode);
488
489 if (!fuse_allow_current_process(fm->fc))
490 return ERR_PTR(-EACCES);
491
492 if (fuse_is_bad(inode))
493 return ERR_PTR(-EIO);
494
495 if (!S_ISREG(inode->i_mode) && !isdir)
496 return ERR_PTR(-ENOTTY);
497
498 return fuse_file_open(fm, get_node_id(inode), O_RDONLY, isdir);
499}
500
501static void fuse_priv_ioctl_cleanup(struct inode *inode, struct fuse_file *ff)
502{
503 fuse_file_release(inode, ff, O_RDONLY, NULL, S_ISDIR(inode->i_mode));
504}
505
506int fuse_fileattr_get(struct dentry *dentry, struct file_kattr *fa)
507{
508 struct inode *inode = d_inode(dentry);
509 struct fuse_file *ff;
510 unsigned int flags;
511 struct fsxattr xfa;
512 int err;
513
514 ff = fuse_priv_ioctl_prepare(inode);
515 if (IS_ERR(ff))
516 return PTR_ERR(ff);
517
518 if (fa->flags_valid) {
519 err = fuse_priv_ioctl(inode, ff, FS_IOC_GETFLAGS,
520 &flags, sizeof(flags));
521 if (err)
522 goto cleanup;
523
524 fileattr_fill_flags(fa, flags);
525 } else {
526 err = fuse_priv_ioctl(inode, ff, FS_IOC_FSGETXATTR,
527 &xfa, sizeof(xfa));
528 if (err)
529 goto cleanup;
530
531 fileattr_fill_xflags(fa, xfa.fsx_xflags);
532 fa->fsx_extsize = xfa.fsx_extsize;
533 fa->fsx_nextents = xfa.fsx_nextents;
534 fa->fsx_projid = xfa.fsx_projid;
535 fa->fsx_cowextsize = xfa.fsx_cowextsize;
536 }
537cleanup:
538 fuse_priv_ioctl_cleanup(inode, ff);
539
540 return err;
541}
542
543int fuse_fileattr_set(const struct mnt_idmap *idmap,
544 struct dentry *dentry, struct file_kattr *fa)
545{
546 struct inode *inode = d_inode(dentry);
547 struct fuse_file *ff;
548 unsigned int flags = fa->flags;
549 struct fsxattr xfa;
550 int err;
551
552 ff = fuse_priv_ioctl_prepare(inode);
553 if (IS_ERR(ff))
554 return PTR_ERR(ff);
555
556 if (fa->flags_valid) {
557 err = fuse_priv_ioctl(inode, ff, FS_IOC_SETFLAGS,
558 &flags, sizeof(flags));
559 if (err)
560 goto cleanup;
561 } else {
562 memset(&xfa, 0, sizeof(xfa));
563 xfa.fsx_xflags = fa->fsx_xflags;
564 xfa.fsx_extsize = fa->fsx_extsize;
565 xfa.fsx_nextents = fa->fsx_nextents;
566 xfa.fsx_projid = fa->fsx_projid;
567 xfa.fsx_cowextsize = fa->fsx_cowextsize;
568
569 err = fuse_priv_ioctl(inode, ff, FS_IOC_FSSETXATTR,
570 &xfa, sizeof(xfa));
571 }
572
573cleanup:
574 fuse_priv_ioctl_cleanup(inode, ff);
575
576 return err;
577}