cuse.c

  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#include "fuse_dev_i.h"
 56
 57#define CUSE_CONNTBL_LEN	64
 58
 59struct cuse_conn {
 60	struct list_head	list;	/* linked on cuse_conntbl */
 61	struct fuse_mount	fm;	/* Dummy mount referencing fc */
 62	struct fuse_conn	fc;	/* fuse connection */
 63	struct cdev		*cdev;	/* associated character device */
 64	struct device		*dev;	/* device representing @cdev */
 65
 66	/* init parameters, set once during initialization */
 67	bool			unrestricted_ioctl;
 68};
 69
 70static DEFINE_MUTEX(cuse_lock);		/* protects registration */
 71static struct list_head cuse_conntbl[CUSE_CONNTBL_LEN];
 72static struct class *cuse_class;
 73
 74static struct cuse_conn *fc_to_cc(struct fuse_conn *fc)
 75{
 76	return container_of(fc, struct cuse_conn, fc);
 77}
 78
 79static struct list_head *cuse_conntbl_head(dev_t devt)
 80{
 81	return &cuse_conntbl[(MAJOR(devt) + MINOR(devt)) % CUSE_CONNTBL_LEN];
 82}
 83
 84
 85/**************************************************************************
 86 * CUSE frontend operations
 87 *
 88 * These are file operations for the character device.
 89 *
 90 * On open, CUSE opens a file from the FUSE mnt and stores it to
 91 * private_data of the open file.  All other ops call FUSE ops on the
 92 * FUSE file.
 93 */
 94
 95static ssize_t cuse_read_iter(struct kiocb *kiocb, struct iov_iter *to)
 96{
 97	struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(kiocb);
 98	loff_t pos = 0;
 99
100	return fuse_direct_io(&io, to, &pos, FUSE_DIO_CUSE);
101}
102
103static ssize_t cuse_write_iter(struct kiocb *kiocb, struct iov_iter *from)
104{
105	struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(kiocb);
106	loff_t pos = 0;
107	/*
108	 * No locking or generic_write_checks(), the server is
109	 * responsible for locking and sanity checks.
110	 */
111	return fuse_direct_io(&io, from, &pos,
112			      FUSE_DIO_WRITE | FUSE_DIO_CUSE);
113}
114
115static int cuse_open(struct inode *inode, struct file *file)
116{
117	dev_t devt = inode->i_cdev->dev;
118	struct cuse_conn *cc = NULL, *pos;
119	int rc;
120
121	/* look up and get the connection */
122	mutex_lock(&cuse_lock);
123	list_for_each_entry(pos, cuse_conntbl_head(devt), list)
124		if (pos->dev->devt == devt) {
125			fuse_conn_get(&pos->fc);
126			cc = pos;
127			break;
128		}
129	mutex_unlock(&cuse_lock);
130
131	/* dead? */
132	if (!cc)
133		return -ENODEV;
134
135	/*
136	 * Generic permission check is already done against the chrdev
137	 * file, proceed to open.
138	 */
139	rc = fuse_do_open(&cc->fm, 0, file, 0);
140	if (rc)
141		fuse_conn_put(&cc->fc);
142	return rc;
143}
144
145static int cuse_release(struct inode *inode, struct file *file)
146{
147	struct fuse_file *ff = file->private_data;
148	struct fuse_mount *fm = ff->fm;
149
150	fuse_sync_release(NULL, ff, file->f_flags);
151	fuse_conn_put(fm->fc);
152
153	return 0;
154}
155
156static long cuse_file_ioctl(struct file *file, unsigned int cmd,
157			    unsigned long arg)
158{
159	struct fuse_file *ff = file->private_data;
160	struct cuse_conn *cc = fc_to_cc(ff->fm->fc);
161	unsigned int flags = 0;
162
163	if (cc->unrestricted_ioctl)
164		flags |= FUSE_IOCTL_UNRESTRICTED;
165
166	return fuse_do_ioctl(file, cmd, arg, flags);
167}
168
169static long cuse_file_compat_ioctl(struct file *file, unsigned int cmd,
170				   unsigned long arg)
171{
172	struct fuse_file *ff = file->private_data;
173	struct cuse_conn *cc = fc_to_cc(ff->fm->fc);
174	unsigned int flags = FUSE_IOCTL_COMPAT;
175
176	if (cc->unrestricted_ioctl)
177		flags |= FUSE_IOCTL_UNRESTRICTED;
178
179	return fuse_do_ioctl(file, cmd, arg, flags);
180}
181
182static const struct file_operations cuse_frontend_fops = {
183	.owner			= THIS_MODULE,
184	.read_iter		= cuse_read_iter,
185	.write_iter		= cuse_write_iter,
186	.open			= cuse_open,
187	.release		= cuse_release,
188	.unlocked_ioctl		= cuse_file_ioctl,
189	.compat_ioctl		= cuse_file_compat_ioctl,
190	.poll			= fuse_file_poll,
191	.llseek		= noop_llseek,
192};
193
194
195/**************************************************************************
196 * CUSE channel initialization and destruction
197 */
198
199struct cuse_devinfo {
200	const char		*name;
201};
202
203/**
204 * cuse_parse_one - parse one key=value pair
205 * @pp: i/o parameter for the current position
206 * @end: points to one past the end of the packed string
207 * @keyp: out parameter for key
208 * @valp: out parameter for value
209 *
210 * *@pp points to packed strings - "key0=val0\0key1=val1\0" which ends
211 * at @end - 1.  This function parses one pair and set *@keyp to the
212 * start of the key and *@valp to the start of the value.  Note that
213 * the original string is modified such that the key string is
214 * terminated with '\0'.  *@pp is updated to point to the next string.
215 *
216 * RETURNS:
217 * 1 on successful parse, 0 on EOF, -errno on failure.
218 */
219static int cuse_parse_one(char **pp, char *end, char **keyp, char **valp)
220{
221	char *p = *pp;
222	char *key, *val;
223
224	while (p < end && *p == '\0')
225		p++;
226	if (p == end)
227		return 0;
228
229	if (end[-1] != '\0') {
230		pr_err("info not properly terminated\n");
231		return -EINVAL;
232	}
233
234	key = val = p;
235	p += strlen(p);
236
237	if (valp) {
238		strsep(&val, "=");
239		if (!val)
240			val = key + strlen(key);
241		key = strstrip(key);
242		val = strstrip(val);
243	} else
244		key = strstrip(key);
245
246	if (!strlen(key)) {
247		pr_err("zero length info key specified\n");
248		return -EINVAL;
249	}
250
251	*pp = p;
252	*keyp = key;
253	if (valp)
254		*valp = val;
255
256	return 1;
257}
258
259/**
260 * cuse_parse_devinfo - parse device info
261 * @p: device info string
262 * @len: length of device info string
263 * @devinfo: out parameter for parsed device info
264 *
265 * Parse @p to extract device info and store it into @devinfo.  String
266 * pointed to by @p is modified by parsing and @devinfo points into
267 * them, so @p shouldn't be freed while @devinfo is in use.
268 *
269 * RETURNS:
270 * 0 on success, -errno on failure.
271 */
272static int cuse_parse_devinfo(char *p, size_t len, struct cuse_devinfo *devinfo)
273{
274	char *end = p + len;
275	char *key, *val;
276	int rc;
277
278	while (true) {
279		rc = cuse_parse_one(&p, end, &key, &val);
280		if (rc < 0)
281			return rc;
282		if (!rc)
283			break;
284		if (strcmp(key, "DEVNAME") == 0)
285			devinfo->name = val;
286		else
287			pr_warn("unknown device info \"%s\"\n", key);
288	}
289
290	if (!devinfo->name || !strlen(devinfo->name)) {
291		pr_err("DEVNAME unspecified\n");
292		return -EINVAL;
293	}
294
295	return 0;
296}
297
298static void cuse_gendev_release(struct device *dev)
299{
300	kfree(dev);
301}
302
303struct cuse_init_args {
304	struct fuse_args_pages ap;
305	struct cuse_init_in in;
306	struct cuse_init_out out;
307	struct folio *folio;
308	struct fuse_folio_desc desc;
309};
310
311/**
312 * cuse_process_init_reply - finish initializing CUSE channel
313 *
314 * @fm: The fuse mount information containing the CUSE connection.
315 * @args: The arguments passed to the init reply.
316 * @error: The error code signifying if any error occurred during the process.
317 *
318 * This function creates the character device and sets up all the
319 * required data structures for it.  Please read the comment at the
320 * top of this file for high level overview.
321 */
322static void cuse_process_init_reply(struct fuse_mount *fm,
323				    struct fuse_args *args, int error)
324{
325	struct fuse_conn *fc = fm->fc;
326	struct cuse_init_args *ia = container_of(args, typeof(*ia), ap.args);
327	struct fuse_args_pages *ap = &ia->ap;
328	struct cuse_conn *cc = fc_to_cc(fc), *pos;
329	struct cuse_init_out *arg = &ia->out;
330	struct folio *folio = ap->folios[0];
331	struct cuse_devinfo devinfo = { };
332	struct device *dev;
333	struct cdev *cdev;
334	dev_t devt;
335	int rc, i;
336
337	if (error || arg->major != FUSE_KERNEL_VERSION || arg->minor < 11)
338		goto err;
339
340	fc->minor = arg->minor;
341	fc->max_read = max_t(unsigned, arg->max_read, 4096);
342	fc->max_write = max_t(unsigned, arg->max_write, 4096);
343
344	/* parse init reply */
345	cc->unrestricted_ioctl = arg->flags & CUSE_UNRESTRICTED_IOCTL;
346
347	rc = cuse_parse_devinfo(folio_address(folio), ap->args.out_args[1].size,
348				&devinfo);
349	if (rc)
350		goto err;
351
352	/* determine and reserve devt */
353	devt = MKDEV(arg->dev_major, arg->dev_minor);
354	if (!MAJOR(devt))
355		rc = alloc_chrdev_region(&devt, MINOR(devt), 1, devinfo.name);
356	else
357		rc = register_chrdev_region(devt, 1, devinfo.name);
358	if (rc) {
359		pr_err("failed to register chrdev region\n");
360		goto err;
361	}
362
363	/* devt determined, create device */
364	rc = -ENOMEM;
365	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
366	if (!dev)
367		goto err_region;
368
369	device_initialize(dev);
370	dev_set_uevent_suppress(dev, 1);
371	dev->class = cuse_class;
372	dev->devt = devt;
373	dev->release = cuse_gendev_release;
374	dev_set_drvdata(dev, cc);
375	dev_set_name(dev, "%s", devinfo.name);
376
377	mutex_lock(&cuse_lock);
378
379	/* make sure the device-name is unique */
380	for (i = 0; i < CUSE_CONNTBL_LEN; ++i) {
381		list_for_each_entry(pos, &cuse_conntbl[i], list)
382			if (!strcmp(dev_name(pos->dev), dev_name(dev)))
383				goto err_unlock;
384	}
385
386	rc = device_add(dev);
387	if (rc)
388		goto err_unlock;
389
390	/* register cdev */
391	rc = -ENOMEM;
392	cdev = cdev_alloc();
393	if (!cdev)
394		goto err_unlock;
395
396	cdev->owner = THIS_MODULE;
397	cdev->ops = &cuse_frontend_fops;
398
399	rc = cdev_add(cdev, devt, 1);
400	if (rc)
401		goto err_cdev;
402
403	cc->dev = dev;
404	cc->cdev = cdev;
405
406	/* make the device available */
407	list_add(&cc->list, cuse_conntbl_head(devt));
408	mutex_unlock(&cuse_lock);
409
410	/* announce device availability */
411	dev_set_uevent_suppress(dev, 0);
412	kobject_uevent(&dev->kobj, KOBJ_ADD);
413out:
414	kfree(ia);
415	folio_put(folio);
416	return;
417
418err_cdev:
419	cdev_del(cdev);
420err_unlock:
421	mutex_unlock(&cuse_lock);
422	put_device(dev);
423err_region:
424	unregister_chrdev_region(devt, 1);
425err:
426	fuse_abort_conn(fc);
427	goto out;
428}
429
430static int cuse_send_init(struct cuse_conn *cc)
431{
432	int rc;
433	struct folio *folio;
434	struct fuse_mount *fm = &cc->fm;
435	struct cuse_init_args *ia;
436	struct fuse_args_pages *ap;
437
438	BUILD_BUG_ON(CUSE_INIT_INFO_MAX > PAGE_SIZE);
439
440	rc = -ENOMEM;
441
442	folio = folio_alloc(GFP_KERNEL | __GFP_ZERO, 0);
443	if (!folio)
444		goto err;
445
446	ia = kzalloc(sizeof(*ia), GFP_KERNEL);
447	if (!ia)
448		goto err_free_folio;
449
450	ap = &ia->ap;
451	ia->in.major = FUSE_KERNEL_VERSION;
452	ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
453	ia->in.flags |= CUSE_UNRESTRICTED_IOCTL;
454	ap->args.opcode = CUSE_INIT;
455	ap->args.in_numargs = 1;
456	ap->args.in_args[0].size = sizeof(ia->in);
457	ap->args.in_args[0].value = &ia->in;
458	ap->args.out_numargs = 2;
459	ap->args.out_args[0].size = sizeof(ia->out);
460	ap->args.out_args[0].value = &ia->out;
461	ap->args.out_args[1].size = CUSE_INIT_INFO_MAX;
462	ap->args.out_argvar = true;
463	ap->args.out_pages = true;
464	ap->num_folios = 1;
465	ap->folios = &ia->folio;
466	ap->descs = &ia->desc;
467	ia->folio = folio;
468	ia->desc.length = ap->args.out_args[1].size;
469	ap->args.end = cuse_process_init_reply;
470
471	rc = fuse_simple_background(fm, &ap->args, GFP_KERNEL);
472	if (rc) {
473		kfree(ia);
474err_free_folio:
475		folio_put(folio);
476	}
477err:
478	return rc;
479}
480
481static void cuse_fc_release(struct fuse_conn *fc)
482{
483	kfree(fc_to_cc(fc));
484}
485
486/**
487 * cuse_channel_open - open method for /dev/cuse
488 * @inode: inode for /dev/cuse
489 * @file: file struct being opened
490 *
491 * Userland CUSE server can create a CUSE device by opening /dev/cuse
492 * and replying to the initialization request kernel sends.  This
493 * function is responsible for handling CUSE device initialization.
494 * Because the fd opened by this function is used during
495 * initialization, this function only creates cuse_conn and sends
496 * init.  The rest is delegated to a kthread.
497 *
498 * RETURNS:
499 * 0 on success, -errno on failure.
500 */
501static int cuse_channel_open(struct inode *inode, struct file *file)
502{
503	struct fuse_dev *fud;
504	struct cuse_conn *cc;
505	int rc;
506
507	/* set up cuse_conn */
508	cc = kzalloc(sizeof(*cc), GFP_KERNEL);
509	if (!cc)
510		return -ENOMEM;
511
512	/*
513	 * Limit the cuse channel to requests that can
514	 * be represented in file->f_cred->user_ns.
515	 */
516	fuse_conn_init(&cc->fc, &cc->fm, file->f_cred->user_ns,
517		       &fuse_dev_fiq_ops, NULL);
518
519	cc->fc.release = cuse_fc_release;
520	fud = fuse_dev_alloc_install(&cc->fc);
521	fuse_conn_put(&cc->fc);
522	if (!fud)
523		return -ENOMEM;
524
525	INIT_LIST_HEAD(&cc->list);
526
527	cc->fc.initialized = 1;
528	rc = cuse_send_init(cc);
529	if (rc) {
530		fuse_dev_free(fud);
531		return rc;
532	}
533	file->private_data = fud;
534
535	return 0;
536}
537
538/**
539 * cuse_channel_release - release method for /dev/cuse
540 * @inode: inode for /dev/cuse
541 * @file: file struct being closed
542 *
543 * Disconnect the channel, deregister CUSE device and initiate
544 * destruction by putting the default reference.
545 *
546 * RETURNS:
547 * 0 on success, -errno on failure.
548 */
549static int cuse_channel_release(struct inode *inode, struct file *file)
550{
551	struct fuse_dev *fud = __fuse_get_dev(file);
552	struct cuse_conn *cc = fc_to_cc(fud->fc);
553
554	/* remove from the conntbl, no more access from this point on */
555	mutex_lock(&cuse_lock);
556	list_del_init(&cc->list);
557	mutex_unlock(&cuse_lock);
558
559	/* remove device */
560	if (cc->dev)
561		device_unregister(cc->dev);
562	if (cc->cdev) {
563		unregister_chrdev_region(cc->cdev->dev, 1);
564		cdev_del(cc->cdev);
565	}
566
567	return fuse_dev_release(inode, file);
568}
569
570static struct file_operations cuse_channel_fops; /* initialized during init */
571
572
573/**************************************************************************
574 * Misc stuff and module initializatiion
575 *
576 * CUSE exports the same set of attributes to sysfs as fusectl.
577 */
578
579static ssize_t cuse_class_waiting_show(struct device *dev,
580				       struct device_attribute *attr, char *buf)
581{
582	struct cuse_conn *cc = dev_get_drvdata(dev);
583
584	return sprintf(buf, "%d\n", atomic_read(&cc->fc.num_waiting));
585}
586static DEVICE_ATTR(waiting, 0400, cuse_class_waiting_show, NULL);
587
588static ssize_t cuse_class_abort_store(struct device *dev,
589				      struct device_attribute *attr,
590				      const char *buf, size_t count)
591{
592	struct cuse_conn *cc = dev_get_drvdata(dev);
593
594	fuse_abort_conn(&cc->fc);
595	return count;
596}
597static DEVICE_ATTR(abort, 0200, NULL, cuse_class_abort_store);
598
599static struct attribute *cuse_class_dev_attrs[] = {
600	&dev_attr_waiting.attr,
601	&dev_attr_abort.attr,
602	NULL,
603};
604ATTRIBUTE_GROUPS(cuse_class_dev);
605
606static struct miscdevice cuse_miscdev = {
607	.minor		= CUSE_MINOR,
608	.name		= "cuse",
609	.fops		= &cuse_channel_fops,
610};
611
612MODULE_ALIAS_MISCDEV(CUSE_MINOR);
613MODULE_ALIAS("devname:cuse");
614
615static int __init cuse_init(void)
616{
617	int i, rc;
618
619	/* init conntbl */
620	for (i = 0; i < CUSE_CONNTBL_LEN; i++)
621		INIT_LIST_HEAD(&cuse_conntbl[i]);
622
623	/* inherit and extend fuse_dev_operations */
624	cuse_channel_fops		= fuse_dev_operations;
625	cuse_channel_fops.owner		= THIS_MODULE;
626	cuse_channel_fops.open		= cuse_channel_open;
627	cuse_channel_fops.release	= cuse_channel_release;
628	/* CUSE is not prepared for FUSE_DEV_IOC_CLONE */
629	cuse_channel_fops.unlocked_ioctl	= NULL;
630
631	cuse_class = class_create("cuse");
632	if (IS_ERR(cuse_class))
633		return PTR_ERR(cuse_class);
634
635	cuse_class->dev_groups = cuse_class_dev_groups;
636
637	rc = misc_register(&cuse_miscdev);
638	if (rc) {
639		class_destroy(cuse_class);
640		return rc;
641	}
642
643	return 0;
644}
645
646static void __exit cuse_exit(void)
647{
648	misc_deregister(&cuse_miscdev);
649	class_destroy(cuse_class);
650}
651
652module_init(cuse_init);
653module_exit(cuse_exit);
654
655MODULE_AUTHOR("Tejun Heo <tj@kernel.org>");
656MODULE_DESCRIPTION("Character device in Userspace");
657MODULE_LICENSE("GPL");