1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * CUSE: Character device in Userspace
4 *
5 * Copyright (C) 2008-2009 SUSE Linux Products GmbH
6 * Copyright (C) 2008-2009 Tejun Heo <tj@kernel.org>
7 *
8 * CUSE enables character devices to be implemented from userland much
9 * like FUSE allows filesystems. On initialization /dev/cuse is
10 * created. By opening the file and replying to the CUSE_INIT request
11 * userland CUSE server can create a character device. After that the
12 * operation is very similar to FUSE.
13 *
14 * A CUSE instance involves the following objects.
15 *
16 * cuse_conn : contains fuse_conn and serves as bonding structure
17 * channel : file handle connected to the userland CUSE server
18 * cdev : the implemented character device
19 * dev : generic device for cdev
20 *
21 * Note that 'channel' is what 'dev' is in FUSE. As CUSE deals with
22 * devices, it's called 'channel' to reduce confusion.
23 *
24 * channel determines when the character device dies. When channel is
25 * closed, everything begins to destruct. The cuse_conn is taken off
26 * the lookup table preventing further access from cdev, cdev and
27 * generic device are removed and the base reference of cuse_conn is
28 * put.
29 *
30 * On each open, the matching cuse_conn is looked up and if found an
31 * additional reference is taken which is released when the file is
32 * closed.
33 */
34
35 #define pr_fmt(fmt) "CUSE: " fmt
36
37 #include <linux/fuse.h>
38 #include <linux/cdev.h>
39 #include <linux/device.h>
40 #include <linux/file.h>
41 #include <linux/fs.h>
42 #include <linux/kdev_t.h>
43 #include <linux/kthread.h>
44 #include <linux/list.h>
45 #include <linux/magic.h>
46 #include <linux/miscdevice.h>
47 #include <linux/mutex.h>
48 #include <linux/slab.h>
49 #include <linux/stat.h>
50 #include <linux/module.h>
51 #include <linux/uio.h>
52 #include <linux/user_namespace.h>
53
54 #include "fuse_i.h"
55
56 #define CUSE_CONNTBL_LEN 64
57
58 struct cuse_conn {
59 struct list_head list; /* linked on cuse_conntbl */
60 struct fuse_mount fm; /* Dummy mount referencing fc */
61 struct fuse_conn fc; /* fuse connection */
62 struct cdev *cdev; /* associated character device */
63 struct device *dev; /* device representing @cdev */
64
65 /* init parameters, set once during initialization */
66 bool unrestricted_ioctl;
67 };
68
69 static DEFINE_MUTEX(cuse_lock); /* protects registration */
70 static struct list_head cuse_conntbl[CUSE_CONNTBL_LEN];
71 static struct class *cuse_class;
72
73 static struct cuse_conn *fc_to_cc(struct fuse_conn *fc)
74 {
75 return container_of(fc, struct cuse_conn, fc);
76 }
77
78 static struct list_head *cuse_conntbl_head(dev_t devt)
79 {
80 return &cuse_conntbl[(MAJOR(devt) + MINOR(devt)) % CUSE_CONNTBL_LEN];
81 }
82
83
84 /**************************************************************************
85 * CUSE frontend operations
86 *
87 * These are file operations for the character device.
88 *
89 * On open, CUSE opens a file from the FUSE mnt and stores it to
90 * private_data of the open file. All other ops call FUSE ops on the
91 * FUSE file.
92 */
93
94 static ssize_t cuse_read_iter(struct kiocb *kiocb, struct iov_iter *to)
95 {
96 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(kiocb);
97 loff_t pos = 0;
98
99 return fuse_direct_io(&io, to, &pos, FUSE_DIO_CUSE);
100 }
101
102 static ssize_t cuse_write_iter(struct kiocb *kiocb, struct iov_iter *from)
103 {
104 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(kiocb);
105 loff_t pos = 0;
106 /*
107 * No locking or generic_write_checks(), the server is
108 * responsible for locking and sanity checks.
109 */
110 return fuse_direct_io(&io, from, &pos,
111 FUSE_DIO_WRITE | FUSE_DIO_CUSE);
112 }
113
114 static int cuse_open(struct inode *inode, struct file *file)
115 {
116 dev_t devt = inode->i_cdev->dev;
117 struct cuse_conn *cc = NULL, *pos;
118 int rc;
119
120 /* look up and get the connection */
121 mutex_lock(&cuse_lock);
122 list_for_each_entry(pos, cuse_conntbl_head(devt), list)
123 if (pos->dev->devt == devt) {
124 fuse_conn_get(&pos->fc);
125 cc = pos;
126 break;
127 }
128 mutex_unlock(&cuse_lock);
129
130 /* dead? */
131 if (!cc)
132 return -ENODEV;
133
134 /*
135 * Generic permission check is already done against the chrdev
136 * file, proceed to open.
137 */
138 rc = fuse_do_open(&cc->fm, 0, file, 0);
139 if (rc)
140 fuse_conn_put(&cc->fc);
141 return rc;
142 }
143
144 static int cuse_release(struct inode *inode, struct file *file)
145 {
146 struct fuse_file *ff = file->private_data;
147 struct fuse_mount *fm = ff->fm;
148
149 fuse_sync_release(NULL, ff, file->f_flags);
150 fuse_conn_put(fm->fc);
151
152 return 0;
153 }
154
155 static long cuse_file_ioctl(struct file *file, unsigned int cmd,
156 unsigned long arg)
157 {
158 struct fuse_file *ff = file->private_data;
159 struct cuse_conn *cc = fc_to_cc(ff->fm->fc);
160 unsigned int flags = 0;
161
162 if (cc->unrestricted_ioctl)
163 flags |= FUSE_IOCTL_UNRESTRICTED;
164
165 return fuse_do_ioctl(file, cmd, arg, flags);
166 }
167
168 static long cuse_file_compat_ioctl(struct file *file, unsigned int cmd,
169 unsigned long arg)
170 {
171 struct fuse_file *ff = file->private_data;
172 struct cuse_conn *cc = fc_to_cc(ff->fm->fc);
173 unsigned int flags = FUSE_IOCTL_COMPAT;
174
175 if (cc->unrestricted_ioctl)
176 flags |= FUSE_IOCTL_UNRESTRICTED;
177
178 return fuse_do_ioctl(file, cmd, arg, flags);
179 }
180
181 static const struct file_operations cuse_frontend_fops = {
182 .owner = THIS_MODULE,
183 .read_iter = cuse_read_iter,
184 .write_iter = cuse_write_iter,
185 .open = cuse_open,
186 .release = cuse_release,
187 .unlocked_ioctl = cuse_file_ioctl,
188 .compat_ioctl = cuse_file_compat_ioctl,
189 .poll = fuse_file_poll,
190 .llseek = noop_llseek,
191 };
192
193
194 /**************************************************************************
195 * CUSE channel initialization and destruction
196 */
197
198 struct cuse_devinfo {
199 const char *name;
200 };
201
202 /**
203 * cuse_parse_one - parse one key=value pair
204 * @pp: i/o parameter for the current position
205 * @end: points to one past the end of the packed string
206 * @keyp: out parameter for key
207 * @valp: out parameter for value
208 *
209 * *@pp points to packed strings - "key0=val0\0key1=val1\0" which ends
210 * at @end - 1. This function parses one pair and set *@keyp to the
211 * start of the key and *@valp to the start of the value. Note that
212 * the original string is modified such that the key string is
213 * terminated with '\0'. *@pp is updated to point to the next string.
214 *
215 * RETURNS:
216 * 1 on successful parse, 0 on EOF, -errno on failure.
217 */
218 static int cuse_parse_one(char **pp, char *end, char **keyp, char **valp)
219 {
220 char *p = *pp;
221 char *key, *val;
222
223 while (p < end && *p == '\0')
224 p++;
225 if (p == end)
226 return 0;
227
228 if (end[-1] != '\0') {
229 pr_err("info not properly terminated\n");
230 return -EINVAL;
231 }
232
233 key = val = p;
234 p += strlen(p);
235
236 if (valp) {
237 strsep(&val, "=");
238 if (!val)
239 val = key + strlen(key);
240 key = strstrip(key);
241 val = strstrip(val);
242 } else
243 key = strstrip(key);
244
245 if (!strlen(key)) {
246 pr_err("zero length info key specified\n");
247 return -EINVAL;
248 }
249
250 *pp = p;
251 *keyp = key;
252 if (valp)
253 *valp = val;
254
255 return 1;
256 }
257
258 /**
259 * cuse_parse_devinfo - parse device info
260 * @p: device info string
261 * @len: length of device info string
262 * @devinfo: out parameter for parsed device info
263 *
264 * Parse @p to extract device info and store it into @devinfo. String
265 * pointed to by @p is modified by parsing and @devinfo points into
266 * them, so @p shouldn't be freed while @devinfo is in use.
267 *
268 * RETURNS:
269 * 0 on success, -errno on failure.
270 */
271 static int cuse_parse_devinfo(char *p, size_t len, struct cuse_devinfo *devinfo)
272 {
273 char *end = p + len;
274 char *key, *val;
275 int rc;
276
277 while (true) {
278 rc = cuse_parse_one(&p, end, &key, &val);
279 if (rc < 0)
280 return rc;
281 if (!rc)
282 break;
283 if (strcmp(key, "DEVNAME") == 0)
284 devinfo->name = val;
285 else
286 pr_warn("unknown device info \"%s\"\n", key);
287 }
288
289 if (!devinfo->name || !strlen(devinfo->name)) {
290 pr_err("DEVNAME unspecified\n");
291 return -EINVAL;
292 }
293
294 return 0;
295 }
296
297 static void cuse_gendev_release(struct device *dev)
298 {
299 kfree(dev);
300 }
301
302 struct cuse_init_args {
303 struct fuse_args_pages ap;
304 struct cuse_init_in in;
305 struct cuse_init_out out;
306 struct page *page;
307 struct fuse_page_desc desc;
308 };
309
310 /**
311 * cuse_process_init_reply - finish initializing CUSE channel
312 *
313 * @fm: The fuse mount information containing the CUSE connection.
314 * @args: The arguments passed to the init reply.
315 * @error: The error code signifying if any error occurred during the process.
316 *
317 * This function creates the character device and sets up all the
318 * required data structures for it. Please read the comment at the
319 * top of this file for high level overview.
320 */
321 static void cuse_process_init_reply(struct fuse_mount *fm,
322 struct fuse_args *args, int error)
323 {
324 struct fuse_conn *fc = fm->fc;
325 struct cuse_init_args *ia = container_of(args, typeof(*ia), ap.args);
326 struct fuse_args_pages *ap = &ia->ap;
327 struct cuse_conn *cc = fc_to_cc(fc), *pos;
328 struct cuse_init_out *arg = &ia->out;
329 struct page *page = ap->pages[0];
330 struct cuse_devinfo devinfo = { };
331 struct device *dev;
332 struct cdev *cdev;
333 dev_t devt;
334 int rc, i;
335
336 if (error || arg->major != FUSE_KERNEL_VERSION || arg->minor < 11)
337 goto err;
338
339 fc->minor = arg->minor;
340 fc->max_read = max_t(unsigned, arg->max_read, 4096);
341 fc->max_write = max_t(unsigned, arg->max_write, 4096);
342
343 /* parse init reply */
344 cc->unrestricted_ioctl = arg->flags & CUSE_UNRESTRICTED_IOCTL;
345
346 rc = cuse_parse_devinfo(page_address(page), ap->args.out_args[1].size,
347 &devinfo);
348 if (rc)
349 goto err;
350
351 /* determine and reserve devt */
352 devt = MKDEV(arg->dev_major, arg->dev_minor);
353 if (!MAJOR(devt))
354 rc = alloc_chrdev_region(&devt, MINOR(devt), 1, devinfo.name);
355 else
356 rc = register_chrdev_region(devt, 1, devinfo.name);
357 if (rc) {
358 pr_err("failed to register chrdev region\n");
359 goto err;
360 }
361
362 /* devt determined, create device */
363 rc = -ENOMEM;
364 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
365 if (!dev)
366 goto err_region;
367
368 device_initialize(dev);
369 dev_set_uevent_suppress(dev, 1);
370 dev->class = cuse_class;
371 dev->devt = devt;
372 dev->release = cuse_gendev_release;
373 dev_set_drvdata(dev, cc);
374 dev_set_name(dev, "%s", devinfo.name);
375
376 mutex_lock(&cuse_lock);
377
378 /* make sure the device-name is unique */
379 for (i = 0; i < CUSE_CONNTBL_LEN; ++i) {
380 list_for_each_entry(pos, &cuse_conntbl[i], list)
381 if (!strcmp(dev_name(pos->dev), dev_name(dev)))
382 goto err_unlock;
383 }
384
385 rc = device_add(dev);
386 if (rc)
387 goto err_unlock;
388
389 /* register cdev */
390 rc = -ENOMEM;
391 cdev = cdev_alloc();
392 if (!cdev)
393 goto err_unlock;
394
395 cdev->owner = THIS_MODULE;
396 cdev->ops = &cuse_frontend_fops;
397
398 rc = cdev_add(cdev, devt, 1);
399 if (rc)
400 goto err_cdev;
401
402 cc->dev = dev;
403 cc->cdev = cdev;
404
405 /* make the device available */
406 list_add(&cc->list, cuse_conntbl_head(devt));
407 mutex_unlock(&cuse_lock);
408
409 /* announce device availability */
410 dev_set_uevent_suppress(dev, 0);
411 kobject_uevent(&dev->kobj, KOBJ_ADD);
412 out:
413 kfree(ia);
414 __free_page(page);
415 return;
416
417 err_cdev:
418 cdev_del(cdev);
419 err_unlock:
420 mutex_unlock(&cuse_lock);
421 put_device(dev);
422 err_region:
423 unregister_chrdev_region(devt, 1);
424 err:
425 fuse_abort_conn(fc);
426 goto out;
427 }
428
429 static int cuse_send_init(struct cuse_conn *cc)
430 {
431 int rc;
432 struct page *page;
433 struct fuse_mount *fm = &cc->fm;
434 struct cuse_init_args *ia;
435 struct fuse_args_pages *ap;
436
437 BUILD_BUG_ON(CUSE_INIT_INFO_MAX > PAGE_SIZE);
438
439 rc = -ENOMEM;
440 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
441 if (!page)
442 goto err;
443
444 ia = kzalloc(sizeof(*ia), GFP_KERNEL);
445 if (!ia)
446 goto err_free_page;
447
448 ap = &ia->ap;
449 ia->in.major = FUSE_KERNEL_VERSION;
450 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
451 ia->in.flags |= CUSE_UNRESTRICTED_IOCTL;
452 ap->args.opcode = CUSE_INIT;
453 ap->args.in_numargs = 1;
454 ap->args.in_args[0].size = sizeof(ia->in);
455 ap->args.in_args[0].value = &ia->in;
456 ap->args.out_numargs = 2;
457 ap->args.out_args[0].size = sizeof(ia->out);
458 ap->args.out_args[0].value = &ia->out;
459 ap->args.out_args[1].size = CUSE_INIT_INFO_MAX;
460 ap->args.out_argvar = true;
461 ap->args.out_pages = true;
462 ap->num_pages = 1;
463 ap->pages = &ia->page;
464 ap->descs = &ia->desc;
465 ia->page = page;
466 ia->desc.length = ap->args.out_args[1].size;
467 ap->args.end = cuse_process_init_reply;
468
469 rc = fuse_simple_background(fm, &ap->args, GFP_KERNEL);
470 if (rc) {
471 kfree(ia);
472 err_free_page:
473 __free_page(page);
474 }
475 err:
476 return rc;
477 }
478
479 static void cuse_fc_release(struct fuse_conn *fc)
480 {
481 kfree(fc_to_cc(fc));
482 }
483
484 /**
485 * cuse_channel_open - open method for /dev/cuse
486 * @inode: inode for /dev/cuse
487 * @file: file struct being opened
488 *
489 * Userland CUSE server can create a CUSE device by opening /dev/cuse
490 * and replying to the initialization request kernel sends. This
491 * function is responsible for handling CUSE device initialization.
492 * Because the fd opened by this function is used during
493 * initialization, this function only creates cuse_conn and sends
494 * init. The rest is delegated to a kthread.
495 *
496 * RETURNS:
497 * 0 on success, -errno on failure.
498 */
499 static int cuse_channel_open(struct inode *inode, struct file *file)
500 {
501 struct fuse_dev *fud;
502 struct cuse_conn *cc;
503 int rc;
504
505 /* set up cuse_conn */
506 cc = kzalloc(sizeof(*cc), GFP_KERNEL);
507 if (!cc)
508 return -ENOMEM;
509
510 /*
511 * Limit the cuse channel to requests that can
512 * be represented in file->f_cred->user_ns.
513 */
514 fuse_conn_init(&cc->fc, &cc->fm, file->f_cred->user_ns,
515 &fuse_dev_fiq_ops, NULL);
516
517 cc->fc.release = cuse_fc_release;
518 fud = fuse_dev_alloc_install(&cc->fc);
519 fuse_conn_put(&cc->fc);
520 if (!fud)
521 return -ENOMEM;
522
523 INIT_LIST_HEAD(&cc->list);
524
525 cc->fc.initialized = 1;
526 rc = cuse_send_init(cc);
527 if (rc) {
528 fuse_dev_free(fud);
529 return rc;
530 }
531 file->private_data = fud;
532
533 return 0;
534 }
535
536 /**
537 * cuse_channel_release - release method for /dev/cuse
538 * @inode: inode for /dev/cuse
539 * @file: file struct being closed
540 *
541 * Disconnect the channel, deregister CUSE device and initiate
542 * destruction by putting the default reference.
543 *
544 * RETURNS:
545 * 0 on success, -errno on failure.
546 */
547 static int cuse_channel_release(struct inode *inode, struct file *file)
548 {
549 struct fuse_dev *fud = file->private_data;
550 struct cuse_conn *cc = fc_to_cc(fud->fc);
551
552 /* remove from the conntbl, no more access from this point on */
553 mutex_lock(&cuse_lock);
554 list_del_init(&cc->list);
555 mutex_unlock(&cuse_lock);
556
557 /* remove device */
558 if (cc->dev)
559 device_unregister(cc->dev);
560 if (cc->cdev) {
561 unregister_chrdev_region(cc->cdev->dev, 1);
562 cdev_del(cc->cdev);
563 }
564
565 return fuse_dev_release(inode, file);
566 }
567
568 static struct file_operations cuse_channel_fops; /* initialized during init */
569
570
571 /**************************************************************************
572 * Misc stuff and module initializatiion
573 *
574 * CUSE exports the same set of attributes to sysfs as fusectl.
575 */
576
577 static ssize_t cuse_class_waiting_show(struct device *dev,
578 struct device_attribute *attr, char *buf)
579 {
580 struct cuse_conn *cc = dev_get_drvdata(dev);
581
582 return sprintf(buf, "%d\n", atomic_read(&cc->fc.num_waiting));
583 }
584 static DEVICE_ATTR(waiting, 0400, cuse_class_waiting_show, NULL);
585
586 static ssize_t cuse_class_abort_store(struct device *dev,
587 struct device_attribute *attr,
588 const char *buf, size_t count)
589 {
590 struct cuse_conn *cc = dev_get_drvdata(dev);
591
592 fuse_abort_conn(&cc->fc);
593 return count;
594 }
595 static DEVICE_ATTR(abort, 0200, NULL, cuse_class_abort_store);
596
597 static struct attribute *cuse_class_dev_attrs[] = {
598 &dev_attr_waiting.attr,
599 &dev_attr_abort.attr,
600 NULL,
601 };
602 ATTRIBUTE_GROUPS(cuse_class_dev);
603
604 static struct miscdevice cuse_miscdev = {
605 .minor = CUSE_MINOR,
606 .name = "cuse",
607 .fops = &cuse_channel_fops,
608 };
609
610 MODULE_ALIAS_MISCDEV(CUSE_MINOR);
611 MODULE_ALIAS("devname:cuse");
612
613 static int __init cuse_init(void)
614 {
615 int i, rc;
616
617 /* init conntbl */
618 for (i = 0; i < CUSE_CONNTBL_LEN; i++)
619 INIT_LIST_HEAD(&cuse_conntbl[i]);
620
621 /* inherit and extend fuse_dev_operations */
622 cuse_channel_fops = fuse_dev_operations;
623 cuse_channel_fops.owner = THIS_MODULE;
624 cuse_channel_fops.open = cuse_channel_open;
625 cuse_channel_fops.release = cuse_channel_release;
626 /* CUSE is not prepared for FUSE_DEV_IOC_CLONE */
627 cuse_channel_fops.unlocked_ioctl = NULL;
628
629 cuse_class = class_create("cuse");
630 if (IS_ERR(cuse_class))
631 return PTR_ERR(cuse_class);
632
633 cuse_class->dev_groups = cuse_class_dev_groups;
634
635 rc = misc_register(&cuse_miscdev);
636 if (rc) {
637 class_destroy(cuse_class);
638 return rc;
639 }
640
641 return 0;
642 }
643
644 static void __exit cuse_exit(void)
645 {
646 misc_deregister(&cuse_miscdev);
647 class_destroy(cuse_class);
648 }
649
650 module_init(cuse_init);
651 module_exit(cuse_exit);
652
653 MODULE_AUTHOR("Tejun Heo <tj@kernel.org>");
654 MODULE_DESCRIPTION("Character device in Userspace");
655 MODULE_LICENSE("GPL");
656
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.