projects/linux·blob master · refs

669 lines · 16.3 KiB

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 <[email protected]>
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 "dev.h"
55#include "fuse_i.h"
56#include "fuse_dev_i.h"
57
58#define CUSE_CONNTBL_LEN 64
59
60struct cuse_conn {
61 struct list_head list; /* linked on cuse_conntbl */
62 struct fuse_mount fm; /* Dummy mount referencing fc */
63 struct fuse_conn fc; /* fuse connection */
64 struct cdev *cdev; /* associated character device */
65 struct device *dev; /* device representing @cdev */
66
67 /* init parameters, set once during initialization */
68 bool unrestricted_ioctl;
69};
70
71static DEFINE_MUTEX(cuse_lock); /* protects registration */
72static struct list_head cuse_conntbl[CUSE_CONNTBL_LEN];
73static struct class *cuse_class;
74
75static struct cuse_conn *fc_to_cc(struct fuse_conn *fc)
76{
77 return container_of(fc, struct cuse_conn, fc);
78}
79
80static struct list_head *cuse_conntbl_head(dev_t devt)
81{
82 return &cuse_conntbl[(MAJOR(devt) + MINOR(devt)) % CUSE_CONNTBL_LEN];
83}
84
85
86/**************************************************************************
87 * CUSE frontend operations
88 *
89 * These are file operations for the character device.
90 *
91 * On open, CUSE opens a file from the FUSE mnt and stores it to
92 * private_data of the open file. All other ops call FUSE ops on the
93 * FUSE file.
94 */
95
96static ssize_t cuse_read_iter(struct kiocb *kiocb, struct iov_iter *to)
97{
98 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(kiocb);
99 loff_t pos = 0;
100
101 return fuse_direct_io(&io, to, &pos, FUSE_DIO_CUSE);
102}
103
104static ssize_t cuse_write_iter(struct kiocb *kiocb, struct iov_iter *from)
105{
106 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(kiocb);
107 loff_t pos = 0;
108 /*
109 * No locking or generic_write_checks(), the server is
110 * responsible for locking and sanity checks.
111 */
112 return fuse_direct_io(&io, from, &pos,
113 FUSE_DIO_WRITE | FUSE_DIO_CUSE);
114}
115
116static int cuse_open(struct inode *inode, struct file *file)
117{
118 dev_t devt = inode->i_cdev->dev;
119 struct cuse_conn *cc = NULL, *pos;
120 int rc;
121
122 /* look up and get the connection */
123 mutex_lock(&cuse_lock);
124 list_for_each_entry(pos, cuse_conntbl_head(devt), list)
125 if (pos->dev->devt == devt) {
126 fuse_conn_get(&pos->fc);
127 cc = pos;
128 break;
129 }
130 mutex_unlock(&cuse_lock);
131
132 /* dead? */
133 if (!cc)
134 return -ENODEV;
135
136 /*
137 * Generic permission check is already done against the chrdev
138 * file, proceed to open.
139 */
140 rc = fuse_do_open(&cc->fm, 0, file, 0);
141 if (rc)
142 fuse_conn_put(&cc->fc);
143 return rc;
144}
145
146static int cuse_release(struct inode *inode, struct file *file)
147{
148 struct fuse_file *ff = file->private_data;
149 struct fuse_mount *fm = ff->fm;
150
151 fuse_sync_release(NULL, ff, file->f_flags);
152 fuse_conn_put(fm->fc);
153
154 return 0;
155}
156
157static long cuse_file_ioctl(struct file *file, unsigned int cmd,
158 unsigned long arg)
159{
160 struct fuse_file *ff = file->private_data;
161 struct cuse_conn *cc = fc_to_cc(ff->fm->fc);
162 unsigned int flags = 0;
163
164 if (cc->unrestricted_ioctl)
165 flags |= FUSE_IOCTL_UNRESTRICTED;
166
167 return fuse_do_ioctl(file, cmd, arg, flags);
168}
169
170static long cuse_file_compat_ioctl(struct file *file, unsigned int cmd,
171 unsigned long arg)
172{
173 struct fuse_file *ff = file->private_data;
174 struct cuse_conn *cc = fc_to_cc(ff->fm->fc);
175 unsigned int flags = FUSE_IOCTL_COMPAT;
176
177 if (cc->unrestricted_ioctl)
178 flags |= FUSE_IOCTL_UNRESTRICTED;
179
180 return fuse_do_ioctl(file, cmd, arg, flags);
181}
182
183static const struct file_operations cuse_frontend_fops = {
184 .owner = THIS_MODULE,
185 .read_iter = cuse_read_iter,
186 .write_iter = cuse_write_iter,
187 .open = cuse_open,
188 .release = cuse_release,
189 .unlocked_ioctl = cuse_file_ioctl,
190 .compat_ioctl = cuse_file_compat_ioctl,
191 .poll = fuse_file_poll,
192 .llseek = noop_llseek,
193};
194
195
196/**************************************************************************
197 * CUSE channel initialization and destruction
198 */
199
200struct cuse_devinfo {
201 const char *name;
202};
203
204/**
205 * cuse_parse_one - parse one key=value pair
206 * @pp: i/o parameter for the current position
207 * @end: points to one past the end of the packed string
208 * @keyp: out parameter for key
209 * @valp: out parameter for value
210 *
211 * *@pp points to packed strings - "key0=val0\0key1=val1\0" which ends
212 * at @end - 1. This function parses one pair and set *@keyp to the
213 * start of the key and *@valp to the start of the value. Note that
214 * the original string is modified such that the key string is
215 * terminated with '\0'. *@pp is updated to point to the next string.
216 *
217 * RETURNS:
218 * 1 on successful parse, 0 on EOF, -errno on failure.
219 */
220static int cuse_parse_one(char **pp, char *end, char **keyp, char **valp)
221{
222 char *p = *pp;
223 char *key, *val;
224
225 while (p < end && *p == '\0')
226 p++;
227 if (p == end)
228 return 0;
229
230 if (end[-1] != '\0') {
231 pr_err("info not properly terminated\n");
232 return -EINVAL;
233 }
234
235 key = val = p;
236 p += strlen(p);
237
238 if (valp) {
239 strsep(&val, "=");
240 if (!val)
241 val = key + strlen(key);
242 key = strstrip(key);
243 val = strstrip(val);
244 } else
245 key = strstrip(key);
246
247 if (!strlen(key)) {
248 pr_err("zero length info key specified\n");
249 return -EINVAL;
250 }
251
252 *pp = p;
253 *keyp = key;
254 if (valp)
255 *valp = val;
256
257 return 1;
258}
259
260/**
261 * cuse_parse_devinfo - parse device info
262 * @p: device info string
263 * @len: length of device info string
264 * @devinfo: out parameter for parsed device info
265 *
266 * Parse @p to extract device info and store it into @devinfo. String
267 * pointed to by @p is modified by parsing and @devinfo points into
268 * them, so @p shouldn't be freed while @devinfo is in use.
269 *
270 * RETURNS:
271 * 0 on success, -errno on failure.
272 */
273static int cuse_parse_devinfo(char *p, size_t len, struct cuse_devinfo *devinfo)
274{
275 char *end = p + len;
276 char *key, *val;
277 int rc;
278
279 while (true) {
280 rc = cuse_parse_one(&p, end, &key, &val);
281 if (rc < 0)
282 return rc;
283 if (!rc)
284 break;
285 if (strcmp(key, "DEVNAME") == 0)
286 devinfo->name = val;
287 else
288 pr_warn("unknown device info \"%s\"\n", key);
289 }
290
291 if (!devinfo->name || !strlen(devinfo->name)) {
292 pr_err("DEVNAME unspecified\n");
293 return -EINVAL;
294 }
295
296 return 0;
297}
298
299static void cuse_gendev_release(struct device *dev)
300{
301 kfree(dev);
302}
303
304struct cuse_init_args {
305 struct fuse_args_pages ap;
306 struct cuse_init_in in;
307 struct cuse_init_out out;
308 struct folio *folio;
309 struct fuse_folio_desc desc;
310 struct fuse_conn *fc;
311};
312
313/**
314 * cuse_process_init_reply - finish initializing CUSE channel
315 *
316 * @fm: The fuse mount information containing the CUSE connection.
317 * @args: The arguments passed to the init reply.
318 * @error: The error code signifying if any error occurred during the process.
319 *
320 * This function creates the character device and sets up all the
321 * required data structures for it. Please read the comment at the
322 * top of this file for high level overview.
323 */
324static void cuse_process_init_reply(struct fuse_args *args, int error)
325{
326 struct cuse_init_args *ia = container_of(args, typeof(*ia), ap.args);
327 struct fuse_conn *fc = ia->fc;
328 struct fuse_args_pages *ap = &ia->ap;
329 struct cuse_conn *cc = fc_to_cc(fc), *pos;
330 struct cuse_init_out *arg = &ia->out;
331 struct folio *folio = ap->folios[0];
332 struct cuse_devinfo devinfo = { };
333 struct device *dev;
334 struct cdev *cdev;
335 dev_t devt;
336 int rc, i;
337
338 if (error || arg->major != FUSE_KERNEL_VERSION || arg->minor < 11)
339 goto err;
340
341 fc->minor = arg->minor;
342 fc->max_read = max_t(unsigned, arg->max_read, 4096);
343 fc->max_write = max_t(unsigned, arg->max_write, 4096);
344
345 /* parse init reply */
346 cc->unrestricted_ioctl = arg->flags & CUSE_UNRESTRICTED_IOCTL;
347
348 rc = cuse_parse_devinfo(folio_address(folio), ap->args.out_args[1].size,
349 &devinfo);
350 if (rc)
351 goto err;
352
353 /* determine and reserve devt */
354 devt = MKDEV(arg->dev_major, arg->dev_minor);
355 if (!MAJOR(devt))
356 rc = alloc_chrdev_region(&devt, MINOR(devt), 1, devinfo.name);
357 else
358 rc = register_chrdev_region(devt, 1, devinfo.name);
359 if (rc) {
360 pr_err("failed to register chrdev region\n");
361 goto err;
362 }
363
364 /* devt determined, create device */
365 rc = -ENOMEM;
366 dev = kzalloc_obj(*dev);
367 if (!dev)
368 goto err_region;
369
370 device_initialize(dev);
371 dev_set_uevent_suppress(dev, 1);
372 dev->class = cuse_class;
373 dev->devt = devt;
374 dev->release = cuse_gendev_release;
375 dev_set_drvdata(dev, cc);
376 dev_set_name(dev, "%s", devinfo.name);
377
378 mutex_lock(&cuse_lock);
379
380 /* make sure the device-name is unique */
381 for (i = 0; i < CUSE_CONNTBL_LEN; ++i) {
382 list_for_each_entry(pos, &cuse_conntbl[i], list)
383 if (!strcmp(dev_name(pos->dev), dev_name(dev)))
384 goto err_unlock;
385 }
386
387 rc = device_add(dev);
388 if (rc)
389 goto err_unlock;
390
391 /* register cdev */
392 rc = -ENOMEM;
393 cdev = cdev_alloc();
394 if (!cdev)
395 goto err_dev;
396
397 cdev->owner = THIS_MODULE;
398 cdev->ops = &cuse_frontend_fops;
399
400 rc = cdev_add(cdev, devt, 1);
401 if (rc)
402 goto err_cdev;
403
404 cc->dev = dev;
405 cc->cdev = cdev;
406
407 /* make the device available */
408 list_add(&cc->list, cuse_conntbl_head(devt));
409 mutex_unlock(&cuse_lock);
410
411 /* announce device availability */
412 dev_set_uevent_suppress(dev, 0);
413 kobject_uevent(&dev->kobj, KOBJ_ADD);
414out:
415 kfree(ia);
416 folio_put(folio);
417 return;
418
419err_cdev:
420 cdev_del(cdev);
421err_dev:
422 device_del(dev);
423err_unlock:
424 mutex_unlock(&cuse_lock);
425 put_device(dev);
426err_region:
427 unregister_chrdev_region(devt, 1);
428err:
429 fuse_chan_abort(fc->chan, false);
430 goto out;
431}
432
433static int cuse_send_init(struct cuse_conn *cc)
434{
435 int rc;
436 struct folio *folio;
437 struct fuse_mount *fm = &cc->fm;
438 struct cuse_init_args *ia;
439 struct fuse_args_pages *ap;
440
441 BUILD_BUG_ON(CUSE_INIT_INFO_MAX > PAGE_SIZE);
442
443 rc = -ENOMEM;
444
445 folio = folio_alloc(GFP_KERNEL | __GFP_ZERO, 0);
446 if (!folio)
447 goto err;
448
449 ia = kzalloc_obj(*ia);
450 if (!ia)
451 goto err_free_folio;
452
453 ap = &ia->ap;
454 ia->in.major = FUSE_KERNEL_VERSION;
455 ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
456 ia->in.flags |= CUSE_UNRESTRICTED_IOCTL;
457 ap->args.opcode = CUSE_INIT;
458 ap->args.in_numargs = 1;
459 ap->args.in_args[0].size = sizeof(ia->in);
460 ap->args.in_args[0].value = &ia->in;
461 ap->args.out_numargs = 2;
462 ap->args.out_args[0].size = sizeof(ia->out);
463 ap->args.out_args[0].value = &ia->out;
464 ap->args.out_args[1].size = CUSE_INIT_INFO_MAX;
465 ap->args.out_argvar = true;
466 ap->args.out_pages = true;
467 ap->num_folios = 1;
468 ap->folios = &ia->folio;
469 ap->descs = &ia->desc;
470 ia->folio = folio;
471 ia->desc.length = ap->args.out_args[1].size;
472 ia->fc = &cc->fc;
473 ap->args.end = cuse_process_init_reply;
474
475 rc = fuse_simple_background(fm, &ap->args, GFP_KERNEL);
476 if (rc) {
477 kfree(ia);
478err_free_folio:
479 folio_put(folio);
480 }
481err:
482 return rc;
483}
484
485static void cuse_fc_release(struct fuse_conn *fc)
486{
487 kfree(fc_to_cc(fc));
488}
489
490/**
491 * cuse_channel_open - open method for /dev/cuse
492 * @inode: inode for /dev/cuse
493 * @file: file struct being opened
494 *
495 * Userland CUSE server can create a CUSE device by opening /dev/cuse
496 * and replying to the initialization request kernel sends. This
497 * function is responsible for handling CUSE device initialization.
498 * Because the fd opened by this function is used during
499 * initialization, this function only creates cuse_conn and sends
500 * init. The rest is delegated to a kthread.
501 *
502 * RETURNS:
503 * 0 on success, -errno on failure.
504 */
505static int cuse_channel_open(struct inode *inode, struct file *file)
506{
507 struct fuse_dev *fud;
508 struct cuse_conn *cc;
509 struct fuse_chan *fch __free(fuse_chan_free) = fuse_dev_chan_new();
510 int rc;
511
512 if (!fch)
513 return -ENOMEM;
514
515 /* set up cuse_conn */
516 cc = kzalloc_obj(*cc);
517 if (!cc)
518 return -ENOMEM;
519
520 /*
521 * Limit the cuse channel to requests that can
522 * be represented in file->f_cred->user_ns.
523 */
524 fuse_conn_init(&cc->fc, &cc->fm, file->f_cred->user_ns, no_free_ptr(fch));
525 cc->fc.release = cuse_fc_release;
526 fud = fuse_dev_alloc_install(cc->fc.chan);
527 fuse_conn_put(&cc->fc);
528 if (!fud)
529 return -ENOMEM;
530
531 INIT_LIST_HEAD(&cc->list);
532
533 /* Pairs with smp_load_acquire() readers of fch->initialized */
534 smp_store_release(&cc->fc.chan->initialized, 1);
535 rc = cuse_send_init(cc);
536 if (rc) {
537 fuse_dev_put(fud);
538 return rc;
539 }
540 file->private_data = fud;
541
542 return 0;
543}
544
545/**
546 * cuse_channel_release - release method for /dev/cuse
547 * @inode: inode for /dev/cuse
548 * @file: file struct being closed
549 *
550 * Disconnect the channel, deregister CUSE device and initiate
551 * destruction by putting the default reference.
552 *
553 * RETURNS:
554 * 0 on success, -errno on failure.
555 */
556static int cuse_channel_release(struct inode *inode, struct file *file)
557{
558 struct fuse_dev *fud = __fuse_get_dev(file);
559 struct cuse_conn *cc = fc_to_cc(fud->chan->conn);
560
561 /* remove from the conntbl, no more access from this point on */
562 mutex_lock(&cuse_lock);
563 list_del_init(&cc->list);
564 mutex_unlock(&cuse_lock);
565
566 /* remove device */
567 if (cc->dev)
568 device_unregister(cc->dev);
569 if (cc->cdev) {
570 unregister_chrdev_region(cc->cdev->dev, 1);
571 cdev_del(cc->cdev);
572 }
573
574 return fuse_dev_release(inode, file);
575}
576
577static struct file_operations cuse_channel_fops; /* initialized during init */
578
579
580/**************************************************************************
581 * Misc stuff and module initializatiion
582 *
583 * CUSE exports the same set of attributes to sysfs as fusectl.
584 */
585
586static ssize_t cuse_class_waiting_show(struct device *dev,
587 struct device_attribute *attr, char *buf)
588{
589 struct cuse_conn *cc = dev_get_drvdata(dev);
590
591 return sprintf(buf, "%d\n", atomic_read(&cc->fc.chan->num_waiting));
592}
593static DEVICE_ATTR(waiting, 0400, cuse_class_waiting_show, NULL);
594
595static ssize_t cuse_class_abort_store(struct device *dev,
596 struct device_attribute *attr,
597 const char *buf, size_t count)
598{
599 struct cuse_conn *cc = dev_get_drvdata(dev);
600
601 fuse_chan_abort(cc->fc.chan, false);
602 return count;
603}
604static DEVICE_ATTR(abort, 0200, NULL, cuse_class_abort_store);
605
606static struct attribute *cuse_class_dev_attrs[] = {
607 &dev_attr_waiting.attr,
608 &dev_attr_abort.attr,
609 NULL,
610};
611ATTRIBUTE_GROUPS(cuse_class_dev);
612
613static struct miscdevice cuse_miscdev = {
614 .minor = CUSE_MINOR,
615 .name = "cuse",
616 .fops = &cuse_channel_fops,
617};
618
619MODULE_ALIAS_MISCDEV(CUSE_MINOR);
620MODULE_ALIAS("devname:cuse");
621
622static int __init cuse_init(void)
623{
624 int i, rc;
625
626 /* init conntbl */
627 for (i = 0; i < CUSE_CONNTBL_LEN; i++)
628 INIT_LIST_HEAD(&cuse_conntbl[i]);
629
630 /* inherit and extend fuse_dev_operations */
631 cuse_channel_fops = fuse_dev_operations;
632 cuse_channel_fops.owner = THIS_MODULE;
633 cuse_channel_fops.open = cuse_channel_open;
634 cuse_channel_fops.release = cuse_channel_release;
635 /* CUSE is not prepared for FUSE_DEV_IOC_CLONE */
636 cuse_channel_fops.unlocked_ioctl = NULL;
637
638 cuse_class = class_create("cuse");
639 if (IS_ERR(cuse_class))
640 return PTR_ERR(cuse_class);
641
642 cuse_class->dev_groups = cuse_class_dev_groups;
643
644 rc = misc_register(&cuse_miscdev);
645 if (rc) {
646 class_destroy(cuse_class);
647 return rc;
648 }
649
650 return 0;
651}
652
653static void __exit cuse_exit(void)
654{
655 misc_deregister(&cuse_miscdev);
656 class_destroy(cuse_class);
657 /*
658 * Wait for pending call_rcu() callbacks that call back into
659 * this module via fc->release (cuse_fc_release).
660 */
661 rcu_barrier();
662}
663
664module_init(cuse_init);
665module_exit(cuse_exit);
666
667MODULE_AUTHOR("Tejun Heo <[email protected]>");
668MODULE_DESCRIPTION("Character device in Userspace");
669MODULE_LICENSE("GPL");