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