blob: 6e07696308dc17a2c8ce26f852b04550d070175f [file] [log] [blame]
Miklos Szeredi334f4852005-09-09 13:10:27 -07001/*
2 FUSE: Filesystem in Userspace
Miklos Szeredi1729a162008-11-26 12:03:54 +01003 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
Miklos Szeredi334f4852005-09-09 13:10:27 -07004
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7*/
8
9#include "fuse_i.h"
10
11#include <linux/init.h>
12#include <linux/module.h>
13#include <linux/poll.h>
14#include <linux/uio.h>
15#include <linux/miscdevice.h>
16#include <linux/pagemap.h>
17#include <linux/file.h>
18#include <linux/slab.h>
Miklos Szeredidd3bb142010-05-25 15:06:06 +020019#include <linux/pipe_fs_i.h>
Miklos Szeredice534fb2010-05-25 15:06:07 +020020#include <linux/swap.h>
21#include <linux/splice.h>
Miklos Szeredi334f4852005-09-09 13:10:27 -070022
23MODULE_ALIAS_MISCDEV(FUSE_MINOR);
Kay Sievers578454f2010-05-20 18:07:20 +020024MODULE_ALIAS("devname:fuse");
Miklos Szeredi334f4852005-09-09 13:10:27 -070025
Christoph Lametere18b8902006-12-06 20:33:20 -080026static struct kmem_cache *fuse_req_cachep;
Miklos Szeredi334f4852005-09-09 13:10:27 -070027
Miklos Szeredi8bfc0162006-01-16 22:14:28 -080028static struct fuse_conn *fuse_get_conn(struct file *file)
Miklos Szeredi334f4852005-09-09 13:10:27 -070029{
Miklos Szeredi0720b312006-04-10 22:54:55 -070030 /*
31 * Lockless access is OK, because file->private data is set
32 * once during mount and is valid until the file is released.
33 */
34 return file->private_data;
Miklos Szeredi334f4852005-09-09 13:10:27 -070035}
36
Miklos Szeredi8bfc0162006-01-16 22:14:28 -080037static void fuse_request_init(struct fuse_req *req)
Miklos Szeredi334f4852005-09-09 13:10:27 -070038{
39 memset(req, 0, sizeof(*req));
40 INIT_LIST_HEAD(&req->list);
Miklos Szeredia4d27e72006-06-25 05:48:54 -070041 INIT_LIST_HEAD(&req->intr_entry);
Miklos Szeredi334f4852005-09-09 13:10:27 -070042 init_waitqueue_head(&req->waitq);
43 atomic_set(&req->count, 1);
44}
45
46struct fuse_req *fuse_request_alloc(void)
47{
Christoph Lametere94b1762006-12-06 20:33:17 -080048 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_KERNEL);
Miklos Szeredi334f4852005-09-09 13:10:27 -070049 if (req)
50 fuse_request_init(req);
51 return req;
52}
Tejun Heo08cbf542009-04-14 10:54:53 +090053EXPORT_SYMBOL_GPL(fuse_request_alloc);
Miklos Szeredi334f4852005-09-09 13:10:27 -070054
Miklos Szeredi3be5a522008-04-30 00:54:41 -070055struct fuse_req *fuse_request_alloc_nofs(void)
56{
57 struct fuse_req *req = kmem_cache_alloc(fuse_req_cachep, GFP_NOFS);
58 if (req)
59 fuse_request_init(req);
60 return req;
61}
62
Miklos Szeredi334f4852005-09-09 13:10:27 -070063void fuse_request_free(struct fuse_req *req)
64{
65 kmem_cache_free(fuse_req_cachep, req);
66}
67
Miklos Szeredi8bfc0162006-01-16 22:14:28 -080068static void block_sigs(sigset_t *oldset)
Miklos Szeredi334f4852005-09-09 13:10:27 -070069{
70 sigset_t mask;
71
72 siginitsetinv(&mask, sigmask(SIGKILL));
73 sigprocmask(SIG_BLOCK, &mask, oldset);
74}
75
Miklos Szeredi8bfc0162006-01-16 22:14:28 -080076static void restore_sigs(sigset_t *oldset)
Miklos Szeredi334f4852005-09-09 13:10:27 -070077{
78 sigprocmask(SIG_SETMASK, oldset, NULL);
79}
80
Miklos Szeredi334f4852005-09-09 13:10:27 -070081static void __fuse_get_request(struct fuse_req *req)
82{
83 atomic_inc(&req->count);
84}
85
86/* Must be called with > 1 refcount */
87static void __fuse_put_request(struct fuse_req *req)
88{
89 BUG_ON(atomic_read(&req->count) < 2);
90 atomic_dec(&req->count);
91}
92
Miklos Szeredi33649c92006-06-25 05:48:52 -070093static void fuse_req_init_context(struct fuse_req *req)
94{
David Howells2186a712008-11-14 10:38:53 +110095 req->in.h.uid = current_fsuid();
96 req->in.h.gid = current_fsgid();
Miklos Szeredi33649c92006-06-25 05:48:52 -070097 req->in.h.pid = current->pid;
98}
99
Miklos Szeredice1d5a42006-04-10 22:54:58 -0700100struct fuse_req *fuse_get_req(struct fuse_conn *fc)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700101{
Miklos Szeredi08a53cd2006-04-10 22:54:59 -0700102 struct fuse_req *req;
103 sigset_t oldset;
Miklos Szeredi9bc5ddd2006-04-11 21:16:09 +0200104 int intr;
Miklos Szeredi08a53cd2006-04-10 22:54:59 -0700105 int err;
106
Miklos Szeredi9bc5ddd2006-04-11 21:16:09 +0200107 atomic_inc(&fc->num_waiting);
Miklos Szeredi08a53cd2006-04-10 22:54:59 -0700108 block_sigs(&oldset);
Miklos Szeredi9bc5ddd2006-04-11 21:16:09 +0200109 intr = wait_event_interruptible(fc->blocked_waitq, !fc->blocked);
Miklos Szeredi08a53cd2006-04-10 22:54:59 -0700110 restore_sigs(&oldset);
Miklos Szeredi9bc5ddd2006-04-11 21:16:09 +0200111 err = -EINTR;
112 if (intr)
113 goto out;
Miklos Szeredi08a53cd2006-04-10 22:54:59 -0700114
Miklos Szeredi51eb01e2006-06-25 05:48:50 -0700115 err = -ENOTCONN;
116 if (!fc->connected)
117 goto out;
118
Miklos Szeredi08a53cd2006-04-10 22:54:59 -0700119 req = fuse_request_alloc();
Miklos Szeredi9bc5ddd2006-04-11 21:16:09 +0200120 err = -ENOMEM;
Miklos Szeredice1d5a42006-04-10 22:54:58 -0700121 if (!req)
Miklos Szeredi9bc5ddd2006-04-11 21:16:09 +0200122 goto out;
Miklos Szeredi334f4852005-09-09 13:10:27 -0700123
Miklos Szeredi33649c92006-06-25 05:48:52 -0700124 fuse_req_init_context(req);
Miklos Szeredi9bc5ddd2006-04-11 21:16:09 +0200125 req->waiting = 1;
Miklos Szeredi334f4852005-09-09 13:10:27 -0700126 return req;
Miklos Szeredi9bc5ddd2006-04-11 21:16:09 +0200127
128 out:
129 atomic_dec(&fc->num_waiting);
130 return ERR_PTR(err);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700131}
Tejun Heo08cbf542009-04-14 10:54:53 +0900132EXPORT_SYMBOL_GPL(fuse_get_req);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700133
Miklos Szeredi33649c92006-06-25 05:48:52 -0700134/*
135 * Return request in fuse_file->reserved_req. However that may
136 * currently be in use. If that is the case, wait for it to become
137 * available.
138 */
139static struct fuse_req *get_reserved_req(struct fuse_conn *fc,
140 struct file *file)
141{
142 struct fuse_req *req = NULL;
143 struct fuse_file *ff = file->private_data;
144
145 do {
Miklos Szeredide5e3de2007-10-16 23:31:00 -0700146 wait_event(fc->reserved_req_waitq, ff->reserved_req);
Miklos Szeredi33649c92006-06-25 05:48:52 -0700147 spin_lock(&fc->lock);
148 if (ff->reserved_req) {
149 req = ff->reserved_req;
150 ff->reserved_req = NULL;
151 get_file(file);
152 req->stolen_file = file;
153 }
154 spin_unlock(&fc->lock);
155 } while (!req);
156
157 return req;
158}
159
160/*
161 * Put stolen request back into fuse_file->reserved_req
162 */
163static void put_reserved_req(struct fuse_conn *fc, struct fuse_req *req)
164{
165 struct file *file = req->stolen_file;
166 struct fuse_file *ff = file->private_data;
167
168 spin_lock(&fc->lock);
169 fuse_request_init(req);
170 BUG_ON(ff->reserved_req);
171 ff->reserved_req = req;
Miklos Szeredide5e3de2007-10-16 23:31:00 -0700172 wake_up_all(&fc->reserved_req_waitq);
Miklos Szeredi33649c92006-06-25 05:48:52 -0700173 spin_unlock(&fc->lock);
174 fput(file);
175}
176
177/*
178 * Gets a requests for a file operation, always succeeds
179 *
180 * This is used for sending the FLUSH request, which must get to
181 * userspace, due to POSIX locks which may need to be unlocked.
182 *
183 * If allocation fails due to OOM, use the reserved request in
184 * fuse_file.
185 *
186 * This is very unlikely to deadlock accidentally, since the
187 * filesystem should not have it's own file open. If deadlock is
188 * intentional, it can still be broken by "aborting" the filesystem.
189 */
190struct fuse_req *fuse_get_req_nofail(struct fuse_conn *fc, struct file *file)
191{
192 struct fuse_req *req;
193
194 atomic_inc(&fc->num_waiting);
195 wait_event(fc->blocked_waitq, !fc->blocked);
196 req = fuse_request_alloc();
197 if (!req)
198 req = get_reserved_req(fc, file);
199
200 fuse_req_init_context(req);
201 req->waiting = 1;
202 return req;
203}
204
Miklos Szeredi334f4852005-09-09 13:10:27 -0700205void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
206{
Miklos Szeredi7128ec22006-02-04 23:27:40 -0800207 if (atomic_dec_and_test(&req->count)) {
Miklos Szeredi9bc5ddd2006-04-11 21:16:09 +0200208 if (req->waiting)
209 atomic_dec(&fc->num_waiting);
Miklos Szeredi33649c92006-06-25 05:48:52 -0700210
211 if (req->stolen_file)
212 put_reserved_req(fc, req);
213 else
214 fuse_request_free(req);
Miklos Szeredi7128ec22006-02-04 23:27:40 -0800215 }
216}
Tejun Heo08cbf542009-04-14 10:54:53 +0900217EXPORT_SYMBOL_GPL(fuse_put_request);
Miklos Szeredi7128ec22006-02-04 23:27:40 -0800218
Miklos Szeredid12def12008-02-06 01:38:39 -0800219static unsigned len_args(unsigned numargs, struct fuse_arg *args)
220{
221 unsigned nbytes = 0;
222 unsigned i;
223
224 for (i = 0; i < numargs; i++)
225 nbytes += args[i].size;
226
227 return nbytes;
228}
229
230static u64 fuse_get_unique(struct fuse_conn *fc)
231{
232 fc->reqctr++;
233 /* zero is special */
234 if (fc->reqctr == 0)
235 fc->reqctr = 1;
236
237 return fc->reqctr;
238}
239
240static void queue_request(struct fuse_conn *fc, struct fuse_req *req)
241{
Miklos Szeredid12def12008-02-06 01:38:39 -0800242 req->in.h.len = sizeof(struct fuse_in_header) +
243 len_args(req->in.numargs, (struct fuse_arg *) req->in.args);
244 list_add_tail(&req->list, &fc->pending);
245 req->state = FUSE_REQ_PENDING;
246 if (!req->waiting) {
247 req->waiting = 1;
248 atomic_inc(&fc->num_waiting);
249 }
250 wake_up(&fc->waitq);
251 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
252}
253
254static void flush_bg_queue(struct fuse_conn *fc)
255{
Csaba Henk7a6d3c82009-07-01 17:28:41 -0700256 while (fc->active_background < fc->max_background &&
Miklos Szeredid12def12008-02-06 01:38:39 -0800257 !list_empty(&fc->bg_queue)) {
258 struct fuse_req *req;
259
260 req = list_entry(fc->bg_queue.next, struct fuse_req, list);
261 list_del(&req->list);
262 fc->active_background++;
Miklos Szeredi2d45ba32010-07-12 14:41:40 +0200263 req->in.h.unique = fuse_get_unique(fc);
Miklos Szeredid12def12008-02-06 01:38:39 -0800264 queue_request(fc, req);
265 }
266}
267
Miklos Szeredi6dbbcb12006-04-26 10:49:06 +0200268/*
Miklos Szeredi334f4852005-09-09 13:10:27 -0700269 * This function is called when a request is finished. Either a reply
Miklos Szeredif9a28422006-06-25 05:48:53 -0700270 * has arrived or it was aborted (and not yet sent) or some error
Miklos Szeredif43b1552006-01-16 22:14:26 -0800271 * occurred during communication with userspace, or the device file
Miklos Szeredi51eb01e2006-06-25 05:48:50 -0700272 * was closed. The requester thread is woken up (if still waiting),
273 * the 'end' callback is called if given, else the reference to the
274 * request is released
Miklos Szeredi7128ec22006-02-04 23:27:40 -0800275 *
Miklos Szeredid7133112006-04-10 22:54:55 -0700276 * Called with fc->lock, unlocks it
Miklos Szeredi334f4852005-09-09 13:10:27 -0700277 */
278static void request_end(struct fuse_conn *fc, struct fuse_req *req)
Miklos Szeredib9ca67b2010-09-07 13:42:41 +0200279__releases(fc->lock)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700280{
Miklos Szeredi51eb01e2006-06-25 05:48:50 -0700281 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
282 req->end = NULL;
Miklos Szeredid77a1d52006-01-16 22:14:31 -0800283 list_del(&req->list);
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700284 list_del(&req->intr_entry);
Miklos Szeredi83cfd492006-01-16 22:14:31 -0800285 req->state = FUSE_REQ_FINISHED;
Miklos Szeredi51eb01e2006-06-25 05:48:50 -0700286 if (req->background) {
Csaba Henk7a6d3c82009-07-01 17:28:41 -0700287 if (fc->num_background == fc->max_background) {
Miklos Szeredi51eb01e2006-06-25 05:48:50 -0700288 fc->blocked = 0;
289 wake_up_all(&fc->blocked_waitq);
290 }
Csaba Henk7a6d3c82009-07-01 17:28:41 -0700291 if (fc->num_background == fc->congestion_threshold &&
Tejun Heoa325f9b2009-04-14 10:54:52 +0900292 fc->connected && fc->bdi_initialized) {
Jens Axboe8aa7e842009-07-09 14:52:32 +0200293 clear_bdi_congested(&fc->bdi, BLK_RW_SYNC);
294 clear_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
Miklos Szeredif92b99b2007-10-16 23:30:59 -0700295 }
Miklos Szeredi51eb01e2006-06-25 05:48:50 -0700296 fc->num_background--;
Miklos Szeredid12def12008-02-06 01:38:39 -0800297 fc->active_background--;
298 flush_bg_queue(fc);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700299 }
Miklos Szeredi51eb01e2006-06-25 05:48:50 -0700300 spin_unlock(&fc->lock);
Miklos Szeredi51eb01e2006-06-25 05:48:50 -0700301 wake_up(&req->waitq);
302 if (end)
303 end(fc, req);
Tejun Heoe9bb09d2008-11-26 12:03:54 +0100304 fuse_put_request(fc, req);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700305}
306
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700307static void wait_answer_interruptible(struct fuse_conn *fc,
308 struct fuse_req *req)
Miklos Szeredib9ca67b2010-09-07 13:42:41 +0200309__releases(fc->lock)
310__acquires(fc->lock)
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700311{
312 if (signal_pending(current))
313 return;
314
315 spin_unlock(&fc->lock);
316 wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED);
317 spin_lock(&fc->lock);
318}
319
320static void queue_interrupt(struct fuse_conn *fc, struct fuse_req *req)
321{
322 list_add_tail(&req->intr_entry, &fc->interrupts);
323 wake_up(&fc->waitq);
324 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
325}
326
Miklos Szeredi7c352bd2005-09-09 13:10:39 -0700327static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
Miklos Szeredib9ca67b2010-09-07 13:42:41 +0200328__releases(fc->lock)
329__acquires(fc->lock)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700330{
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700331 if (!fc->no_interrupt) {
332 /* Any signal may interrupt this */
333 wait_answer_interruptible(fc, req);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700334
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700335 if (req->aborted)
336 goto aborted;
337 if (req->state == FUSE_REQ_FINISHED)
338 return;
339
340 req->interrupted = 1;
341 if (req->state == FUSE_REQ_SENT)
342 queue_interrupt(fc, req);
343 }
344
Miklos Szeredia131de02007-10-16 23:31:04 -0700345 if (!req->force) {
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700346 sigset_t oldset;
347
348 /* Only fatal signals may interrupt this */
Miklos Szeredi51eb01e2006-06-25 05:48:50 -0700349 block_sigs(&oldset);
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700350 wait_answer_interruptible(fc, req);
Miklos Szeredi51eb01e2006-06-25 05:48:50 -0700351 restore_sigs(&oldset);
Miklos Szeredia131de02007-10-16 23:31:04 -0700352
353 if (req->aborted)
354 goto aborted;
355 if (req->state == FUSE_REQ_FINISHED)
356 return;
357
358 /* Request is not yet in userspace, bail out */
359 if (req->state == FUSE_REQ_PENDING) {
360 list_del(&req->list);
361 __fuse_put_request(req);
362 req->out.h.error = -EINTR;
363 return;
364 }
Miklos Szeredi51eb01e2006-06-25 05:48:50 -0700365 }
Miklos Szeredi334f4852005-09-09 13:10:27 -0700366
Miklos Szeredia131de02007-10-16 23:31:04 -0700367 /*
368 * Either request is already in userspace, or it was forced.
369 * Wait it out.
370 */
371 spin_unlock(&fc->lock);
372 wait_event(req->waitq, req->state == FUSE_REQ_FINISHED);
373 spin_lock(&fc->lock);
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700374
Miklos Szeredia131de02007-10-16 23:31:04 -0700375 if (!req->aborted)
376 return;
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700377
378 aborted:
Miklos Szeredia131de02007-10-16 23:31:04 -0700379 BUG_ON(req->state != FUSE_REQ_FINISHED);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700380 if (req->locked) {
381 /* This is uninterruptible sleep, because data is
382 being copied to/from the buffers of req. During
383 locked state, there mustn't be any filesystem
384 operation (e.g. page fault), since that could lead
385 to deadlock */
Miklos Szeredid7133112006-04-10 22:54:55 -0700386 spin_unlock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700387 wait_event(req->waitq, !req->locked);
Miklos Szeredid7133112006-04-10 22:54:55 -0700388 spin_lock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700389 }
Miklos Szeredi334f4852005-09-09 13:10:27 -0700390}
391
Tejun Heob93f8582008-11-26 12:03:55 +0100392void fuse_request_send(struct fuse_conn *fc, struct fuse_req *req)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700393{
394 req->isreply = 1;
Miklos Szeredid7133112006-04-10 22:54:55 -0700395 spin_lock(&fc->lock);
Miklos Szeredi1e9a4ed2005-09-09 13:10:31 -0700396 if (!fc->connected)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700397 req->out.h.error = -ENOTCONN;
398 else if (fc->conn_error)
399 req->out.h.error = -ECONNREFUSED;
400 else {
Miklos Szeredi2d45ba32010-07-12 14:41:40 +0200401 req->in.h.unique = fuse_get_unique(fc);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700402 queue_request(fc, req);
403 /* acquire extra reference, since request is still needed
404 after request_end() */
405 __fuse_get_request(req);
406
Miklos Szeredi7c352bd2005-09-09 13:10:39 -0700407 request_wait_answer(fc, req);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700408 }
Miklos Szeredid7133112006-04-10 22:54:55 -0700409 spin_unlock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700410}
Tejun Heo08cbf542009-04-14 10:54:53 +0900411EXPORT_SYMBOL_GPL(fuse_request_send);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700412
Tejun Heob93f8582008-11-26 12:03:55 +0100413static void fuse_request_send_nowait_locked(struct fuse_conn *fc,
414 struct fuse_req *req)
Miklos Szeredid12def12008-02-06 01:38:39 -0800415{
416 req->background = 1;
417 fc->num_background++;
Csaba Henk7a6d3c82009-07-01 17:28:41 -0700418 if (fc->num_background == fc->max_background)
Miklos Szeredid12def12008-02-06 01:38:39 -0800419 fc->blocked = 1;
Csaba Henk7a6d3c82009-07-01 17:28:41 -0700420 if (fc->num_background == fc->congestion_threshold &&
Tejun Heoa325f9b2009-04-14 10:54:52 +0900421 fc->bdi_initialized) {
Jens Axboe8aa7e842009-07-09 14:52:32 +0200422 set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
423 set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
Miklos Szeredid12def12008-02-06 01:38:39 -0800424 }
425 list_add_tail(&req->list, &fc->bg_queue);
426 flush_bg_queue(fc);
427}
428
Tejun Heob93f8582008-11-26 12:03:55 +0100429static void fuse_request_send_nowait(struct fuse_conn *fc, struct fuse_req *req)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700430{
Miklos Szeredid7133112006-04-10 22:54:55 -0700431 spin_lock(&fc->lock);
Miklos Szeredi1e9a4ed2005-09-09 13:10:31 -0700432 if (fc->connected) {
Tejun Heob93f8582008-11-26 12:03:55 +0100433 fuse_request_send_nowait_locked(fc, req);
Miklos Szeredid7133112006-04-10 22:54:55 -0700434 spin_unlock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700435 } else {
436 req->out.h.error = -ENOTCONN;
437 request_end(fc, req);
438 }
439}
440
Tejun Heob93f8582008-11-26 12:03:55 +0100441void fuse_request_send_noreply(struct fuse_conn *fc, struct fuse_req *req)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700442{
443 req->isreply = 0;
Tejun Heob93f8582008-11-26 12:03:55 +0100444 fuse_request_send_nowait(fc, req);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700445}
446
Tejun Heob93f8582008-11-26 12:03:55 +0100447void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700448{
449 req->isreply = 1;
Tejun Heob93f8582008-11-26 12:03:55 +0100450 fuse_request_send_nowait(fc, req);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700451}
Tejun Heo08cbf542009-04-14 10:54:53 +0900452EXPORT_SYMBOL_GPL(fuse_request_send_background);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700453
Miklos Szeredi2d45ba32010-07-12 14:41:40 +0200454static int fuse_request_send_notify_reply(struct fuse_conn *fc,
455 struct fuse_req *req, u64 unique)
456{
457 int err = -ENODEV;
458
459 req->isreply = 0;
460 req->in.h.unique = unique;
461 spin_lock(&fc->lock);
462 if (fc->connected) {
463 queue_request(fc, req);
464 err = 0;
465 }
466 spin_unlock(&fc->lock);
467
468 return err;
469}
470
Miklos Szeredi334f4852005-09-09 13:10:27 -0700471/*
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700472 * Called under fc->lock
473 *
474 * fc->connected must have been checked previously
475 */
Tejun Heob93f8582008-11-26 12:03:55 +0100476void fuse_request_send_background_locked(struct fuse_conn *fc,
477 struct fuse_req *req)
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700478{
479 req->isreply = 1;
Tejun Heob93f8582008-11-26 12:03:55 +0100480 fuse_request_send_nowait_locked(fc, req);
Miklos Szeredi3be5a522008-04-30 00:54:41 -0700481}
482
483/*
Miklos Szeredi334f4852005-09-09 13:10:27 -0700484 * Lock the request. Up to the next unlock_request() there mustn't be
485 * anything that could cause a page-fault. If the request was already
Miklos Szeredif9a28422006-06-25 05:48:53 -0700486 * aborted bail out.
Miklos Szeredi334f4852005-09-09 13:10:27 -0700487 */
Miklos Szeredid7133112006-04-10 22:54:55 -0700488static int lock_request(struct fuse_conn *fc, struct fuse_req *req)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700489{
490 int err = 0;
491 if (req) {
Miklos Szeredid7133112006-04-10 22:54:55 -0700492 spin_lock(&fc->lock);
Miklos Szeredif9a28422006-06-25 05:48:53 -0700493 if (req->aborted)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700494 err = -ENOENT;
495 else
496 req->locked = 1;
Miklos Szeredid7133112006-04-10 22:54:55 -0700497 spin_unlock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700498 }
499 return err;
500}
501
502/*
Miklos Szeredif9a28422006-06-25 05:48:53 -0700503 * Unlock request. If it was aborted during being locked, the
Miklos Szeredi334f4852005-09-09 13:10:27 -0700504 * requester thread is currently waiting for it to be unlocked, so
505 * wake it up.
506 */
Miklos Szeredid7133112006-04-10 22:54:55 -0700507static void unlock_request(struct fuse_conn *fc, struct fuse_req *req)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700508{
509 if (req) {
Miklos Szeredid7133112006-04-10 22:54:55 -0700510 spin_lock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700511 req->locked = 0;
Miklos Szeredif9a28422006-06-25 05:48:53 -0700512 if (req->aborted)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700513 wake_up(&req->waitq);
Miklos Szeredid7133112006-04-10 22:54:55 -0700514 spin_unlock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700515 }
516}
517
518struct fuse_copy_state {
Miklos Szeredid7133112006-04-10 22:54:55 -0700519 struct fuse_conn *fc;
Miklos Szeredi334f4852005-09-09 13:10:27 -0700520 int write;
521 struct fuse_req *req;
522 const struct iovec *iov;
Miklos Szeredidd3bb142010-05-25 15:06:06 +0200523 struct pipe_buffer *pipebufs;
524 struct pipe_buffer *currbuf;
525 struct pipe_inode_info *pipe;
Miklos Szeredi334f4852005-09-09 13:10:27 -0700526 unsigned long nr_segs;
527 unsigned long seglen;
528 unsigned long addr;
529 struct page *pg;
530 void *mapaddr;
531 void *buf;
532 unsigned len;
Miklos Szeredice534fb2010-05-25 15:06:07 +0200533 unsigned move_pages:1;
Miklos Szeredi334f4852005-09-09 13:10:27 -0700534};
535
Miklos Szeredid7133112006-04-10 22:54:55 -0700536static void fuse_copy_init(struct fuse_copy_state *cs, struct fuse_conn *fc,
Miklos Szeredic3021622010-05-25 15:06:07 +0200537 int write,
Miklos Szeredid7133112006-04-10 22:54:55 -0700538 const struct iovec *iov, unsigned long nr_segs)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700539{
540 memset(cs, 0, sizeof(*cs));
Miklos Szeredid7133112006-04-10 22:54:55 -0700541 cs->fc = fc;
Miklos Szeredi334f4852005-09-09 13:10:27 -0700542 cs->write = write;
Miklos Szeredi334f4852005-09-09 13:10:27 -0700543 cs->iov = iov;
544 cs->nr_segs = nr_segs;
545}
546
547/* Unmap and put previous page of userspace buffer */
Miklos Szeredi8bfc0162006-01-16 22:14:28 -0800548static void fuse_copy_finish(struct fuse_copy_state *cs)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700549{
Miklos Szeredidd3bb142010-05-25 15:06:06 +0200550 if (cs->currbuf) {
551 struct pipe_buffer *buf = cs->currbuf;
552
Miklos Szeredic3021622010-05-25 15:06:07 +0200553 if (!cs->write) {
554 buf->ops->unmap(cs->pipe, buf, cs->mapaddr);
555 } else {
Miklos Szeredi7909b1c2010-07-12 14:41:40 +0200556 kunmap(buf->page);
Miklos Szeredic3021622010-05-25 15:06:07 +0200557 buf->len = PAGE_SIZE - cs->len;
558 }
Miklos Szeredidd3bb142010-05-25 15:06:06 +0200559 cs->currbuf = NULL;
560 cs->mapaddr = NULL;
561 } else if (cs->mapaddr) {
Miklos Szeredi7909b1c2010-07-12 14:41:40 +0200562 kunmap(cs->pg);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700563 if (cs->write) {
564 flush_dcache_page(cs->pg);
565 set_page_dirty_lock(cs->pg);
566 }
567 put_page(cs->pg);
568 cs->mapaddr = NULL;
569 }
570}
571
572/*
573 * Get another pagefull of userspace buffer, and map it to kernel
574 * address space, and lock request
575 */
576static int fuse_copy_fill(struct fuse_copy_state *cs)
577{
578 unsigned long offset;
579 int err;
580
Miklos Szeredid7133112006-04-10 22:54:55 -0700581 unlock_request(cs->fc, cs->req);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700582 fuse_copy_finish(cs);
Miklos Szeredidd3bb142010-05-25 15:06:06 +0200583 if (cs->pipebufs) {
584 struct pipe_buffer *buf = cs->pipebufs;
585
Miklos Szeredic3021622010-05-25 15:06:07 +0200586 if (!cs->write) {
587 err = buf->ops->confirm(cs->pipe, buf);
588 if (err)
589 return err;
Miklos Szeredidd3bb142010-05-25 15:06:06 +0200590
Miklos Szeredic3021622010-05-25 15:06:07 +0200591 BUG_ON(!cs->nr_segs);
592 cs->currbuf = buf;
Miklos Szeredi7909b1c2010-07-12 14:41:40 +0200593 cs->mapaddr = buf->ops->map(cs->pipe, buf, 0);
Miklos Szeredic3021622010-05-25 15:06:07 +0200594 cs->len = buf->len;
595 cs->buf = cs->mapaddr + buf->offset;
596 cs->pipebufs++;
597 cs->nr_segs--;
598 } else {
599 struct page *page;
600
601 if (cs->nr_segs == cs->pipe->buffers)
602 return -EIO;
603
604 page = alloc_page(GFP_HIGHUSER);
605 if (!page)
606 return -ENOMEM;
607
608 buf->page = page;
609 buf->offset = 0;
610 buf->len = 0;
611
612 cs->currbuf = buf;
Miklos Szeredi7909b1c2010-07-12 14:41:40 +0200613 cs->mapaddr = kmap(page);
Miklos Szeredic3021622010-05-25 15:06:07 +0200614 cs->buf = cs->mapaddr;
615 cs->len = PAGE_SIZE;
616 cs->pipebufs++;
617 cs->nr_segs++;
618 }
Miklos Szeredidd3bb142010-05-25 15:06:06 +0200619 } else {
620 if (!cs->seglen) {
621 BUG_ON(!cs->nr_segs);
622 cs->seglen = cs->iov[0].iov_len;
623 cs->addr = (unsigned long) cs->iov[0].iov_base;
624 cs->iov++;
625 cs->nr_segs--;
626 }
627 err = get_user_pages_fast(cs->addr, 1, cs->write, &cs->pg);
628 if (err < 0)
629 return err;
630 BUG_ON(err != 1);
631 offset = cs->addr % PAGE_SIZE;
Miklos Szeredi7909b1c2010-07-12 14:41:40 +0200632 cs->mapaddr = kmap(cs->pg);
Miklos Szeredidd3bb142010-05-25 15:06:06 +0200633 cs->buf = cs->mapaddr + offset;
634 cs->len = min(PAGE_SIZE - offset, cs->seglen);
635 cs->seglen -= cs->len;
636 cs->addr += cs->len;
Miklos Szeredi334f4852005-09-09 13:10:27 -0700637 }
Miklos Szeredi334f4852005-09-09 13:10:27 -0700638
Miklos Szeredid7133112006-04-10 22:54:55 -0700639 return lock_request(cs->fc, cs->req);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700640}
641
642/* Do as much copy to/from userspace buffer as we can */
Miklos Szeredi8bfc0162006-01-16 22:14:28 -0800643static int fuse_copy_do(struct fuse_copy_state *cs, void **val, unsigned *size)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700644{
645 unsigned ncpy = min(*size, cs->len);
646 if (val) {
647 if (cs->write)
648 memcpy(cs->buf, *val, ncpy);
649 else
650 memcpy(*val, cs->buf, ncpy);
651 *val += ncpy;
652 }
653 *size -= ncpy;
654 cs->len -= ncpy;
655 cs->buf += ncpy;
656 return ncpy;
657}
658
Miklos Szeredice534fb2010-05-25 15:06:07 +0200659static int fuse_check_page(struct page *page)
660{
661 if (page_mapcount(page) ||
662 page->mapping != NULL ||
663 page_count(page) != 1 ||
664 (page->flags & PAGE_FLAGS_CHECK_AT_PREP &
665 ~(1 << PG_locked |
666 1 << PG_referenced |
667 1 << PG_uptodate |
668 1 << PG_lru |
669 1 << PG_active |
670 1 << PG_reclaim))) {
671 printk(KERN_WARNING "fuse: trying to steal weird page\n");
672 printk(KERN_WARNING " page=%p index=%li flags=%08lx, count=%i, mapcount=%i, mapping=%p\n", page, page->index, page->flags, page_count(page), page_mapcount(page), page->mapping);
673 return 1;
674 }
675 return 0;
676}
677
678static int fuse_try_move_page(struct fuse_copy_state *cs, struct page **pagep)
679{
680 int err;
681 struct page *oldpage = *pagep;
682 struct page *newpage;
683 struct pipe_buffer *buf = cs->pipebufs;
684 struct address_space *mapping;
685 pgoff_t index;
686
687 unlock_request(cs->fc, cs->req);
688 fuse_copy_finish(cs);
689
690 err = buf->ops->confirm(cs->pipe, buf);
691 if (err)
692 return err;
693
694 BUG_ON(!cs->nr_segs);
695 cs->currbuf = buf;
696 cs->len = buf->len;
697 cs->pipebufs++;
698 cs->nr_segs--;
699
700 if (cs->len != PAGE_SIZE)
701 goto out_fallback;
702
703 if (buf->ops->steal(cs->pipe, buf) != 0)
704 goto out_fallback;
705
706 newpage = buf->page;
707
708 if (WARN_ON(!PageUptodate(newpage)))
709 return -EIO;
710
711 ClearPageMappedToDisk(newpage);
712
713 if (fuse_check_page(newpage) != 0)
714 goto out_fallback_unlock;
715
716 mapping = oldpage->mapping;
717 index = oldpage->index;
718
719 /*
720 * This is a new and locked page, it shouldn't be mapped or
721 * have any special flags on it
722 */
723 if (WARN_ON(page_mapped(oldpage)))
724 goto out_fallback_unlock;
725 if (WARN_ON(page_has_private(oldpage)))
726 goto out_fallback_unlock;
727 if (WARN_ON(PageDirty(oldpage) || PageWriteback(oldpage)))
728 goto out_fallback_unlock;
729 if (WARN_ON(PageMlocked(oldpage)))
730 goto out_fallback_unlock;
731
732 remove_from_page_cache(oldpage);
733 page_cache_release(oldpage);
734
735 err = add_to_page_cache_locked(newpage, mapping, index, GFP_KERNEL);
736 if (err) {
737 printk(KERN_WARNING "fuse_try_move_page: failed to add page");
738 goto out_fallback_unlock;
739 }
740 page_cache_get(newpage);
741
742 if (!(buf->flags & PIPE_BUF_FLAG_LRU))
743 lru_cache_add_file(newpage);
744
745 err = 0;
746 spin_lock(&cs->fc->lock);
747 if (cs->req->aborted)
748 err = -ENOENT;
749 else
750 *pagep = newpage;
751 spin_unlock(&cs->fc->lock);
752
753 if (err) {
754 unlock_page(newpage);
755 page_cache_release(newpage);
756 return err;
757 }
758
759 unlock_page(oldpage);
760 page_cache_release(oldpage);
761 cs->len = 0;
762
763 return 0;
764
765out_fallback_unlock:
766 unlock_page(newpage);
767out_fallback:
768 cs->mapaddr = buf->ops->map(cs->pipe, buf, 1);
769 cs->buf = cs->mapaddr + buf->offset;
770
771 err = lock_request(cs->fc, cs->req);
772 if (err)
773 return err;
774
775 return 1;
776}
777
Miklos Szeredic3021622010-05-25 15:06:07 +0200778static int fuse_ref_page(struct fuse_copy_state *cs, struct page *page,
779 unsigned offset, unsigned count)
780{
781 struct pipe_buffer *buf;
782
783 if (cs->nr_segs == cs->pipe->buffers)
784 return -EIO;
785
786 unlock_request(cs->fc, cs->req);
787 fuse_copy_finish(cs);
788
789 buf = cs->pipebufs;
790 page_cache_get(page);
791 buf->page = page;
792 buf->offset = offset;
793 buf->len = count;
794
795 cs->pipebufs++;
796 cs->nr_segs++;
797 cs->len = 0;
798
799 return 0;
800}
801
Miklos Szeredi334f4852005-09-09 13:10:27 -0700802/*
803 * Copy a page in the request to/from the userspace buffer. Must be
804 * done atomically
805 */
Miklos Szeredice534fb2010-05-25 15:06:07 +0200806static int fuse_copy_page(struct fuse_copy_state *cs, struct page **pagep,
Miklos Szeredi8bfc0162006-01-16 22:14:28 -0800807 unsigned offset, unsigned count, int zeroing)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700808{
Miklos Szeredice534fb2010-05-25 15:06:07 +0200809 int err;
810 struct page *page = *pagep;
811
Miklos Szeredib6777c42010-10-26 14:22:27 -0700812 if (page && zeroing && count < PAGE_SIZE)
813 clear_highpage(page);
814
Miklos Szeredi334f4852005-09-09 13:10:27 -0700815 while (count) {
Miklos Szeredic3021622010-05-25 15:06:07 +0200816 if (cs->write && cs->pipebufs && page) {
817 return fuse_ref_page(cs, page, offset, count);
818 } else if (!cs->len) {
Miklos Szeredice534fb2010-05-25 15:06:07 +0200819 if (cs->move_pages && page &&
820 offset == 0 && count == PAGE_SIZE) {
821 err = fuse_try_move_page(cs, pagep);
822 if (err <= 0)
823 return err;
824 } else {
825 err = fuse_copy_fill(cs);
826 if (err)
827 return err;
828 }
Miklos Szeredi1729a162008-11-26 12:03:54 +0100829 }
Miklos Szeredi334f4852005-09-09 13:10:27 -0700830 if (page) {
Miklos Szeredib6777c42010-10-26 14:22:27 -0700831 void *mapaddr = kmap_atomic(page, KM_USER0);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700832 void *buf = mapaddr + offset;
833 offset += fuse_copy_do(cs, &buf, &count);
Miklos Szeredib6777c42010-10-26 14:22:27 -0700834 kunmap_atomic(mapaddr, KM_USER0);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700835 } else
836 offset += fuse_copy_do(cs, NULL, &count);
837 }
838 if (page && !cs->write)
839 flush_dcache_page(page);
840 return 0;
841}
842
843/* Copy pages in the request to/from userspace buffer */
844static int fuse_copy_pages(struct fuse_copy_state *cs, unsigned nbytes,
845 int zeroing)
846{
847 unsigned i;
848 struct fuse_req *req = cs->req;
849 unsigned offset = req->page_offset;
850 unsigned count = min(nbytes, (unsigned) PAGE_SIZE - offset);
851
852 for (i = 0; i < req->num_pages && (nbytes || zeroing); i++) {
Miklos Szeredice534fb2010-05-25 15:06:07 +0200853 int err;
854
855 err = fuse_copy_page(cs, &req->pages[i], offset, count,
856 zeroing);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700857 if (err)
858 return err;
859
860 nbytes -= count;
861 count = min(nbytes, (unsigned) PAGE_SIZE);
862 offset = 0;
863 }
864 return 0;
865}
866
867/* Copy a single argument in the request to/from userspace buffer */
868static int fuse_copy_one(struct fuse_copy_state *cs, void *val, unsigned size)
869{
870 while (size) {
Miklos Szeredi1729a162008-11-26 12:03:54 +0100871 if (!cs->len) {
872 int err = fuse_copy_fill(cs);
873 if (err)
874 return err;
875 }
Miklos Szeredi334f4852005-09-09 13:10:27 -0700876 fuse_copy_do(cs, &val, &size);
877 }
878 return 0;
879}
880
881/* Copy request arguments to/from userspace buffer */
882static int fuse_copy_args(struct fuse_copy_state *cs, unsigned numargs,
883 unsigned argpages, struct fuse_arg *args,
884 int zeroing)
885{
886 int err = 0;
887 unsigned i;
888
889 for (i = 0; !err && i < numargs; i++) {
890 struct fuse_arg *arg = &args[i];
891 if (i == numargs - 1 && argpages)
892 err = fuse_copy_pages(cs, arg->size, zeroing);
893 else
894 err = fuse_copy_one(cs, arg->value, arg->size);
895 }
896 return err;
897}
898
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700899static int request_pending(struct fuse_conn *fc)
900{
901 return !list_empty(&fc->pending) || !list_empty(&fc->interrupts);
902}
903
Miklos Szeredi334f4852005-09-09 13:10:27 -0700904/* Wait until a request is available on the pending list */
905static void request_wait(struct fuse_conn *fc)
Miklos Szeredib9ca67b2010-09-07 13:42:41 +0200906__releases(fc->lock)
907__acquires(fc->lock)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700908{
909 DECLARE_WAITQUEUE(wait, current);
910
911 add_wait_queue_exclusive(&fc->waitq, &wait);
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700912 while (fc->connected && !request_pending(fc)) {
Miklos Szeredi334f4852005-09-09 13:10:27 -0700913 set_current_state(TASK_INTERRUPTIBLE);
914 if (signal_pending(current))
915 break;
916
Miklos Szeredid7133112006-04-10 22:54:55 -0700917 spin_unlock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700918 schedule();
Miklos Szeredid7133112006-04-10 22:54:55 -0700919 spin_lock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -0700920 }
921 set_current_state(TASK_RUNNING);
922 remove_wait_queue(&fc->waitq, &wait);
923}
924
925/*
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700926 * Transfer an interrupt request to userspace
927 *
928 * Unlike other requests this is assembled on demand, without a need
929 * to allocate a separate fuse_req structure.
930 *
931 * Called with fc->lock held, releases it
932 */
Miklos Szeredic3021622010-05-25 15:06:07 +0200933static int fuse_read_interrupt(struct fuse_conn *fc, struct fuse_copy_state *cs,
934 size_t nbytes, struct fuse_req *req)
Miklos Szeredib9ca67b2010-09-07 13:42:41 +0200935__releases(fc->lock)
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700936{
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700937 struct fuse_in_header ih;
938 struct fuse_interrupt_in arg;
939 unsigned reqsize = sizeof(ih) + sizeof(arg);
940 int err;
941
942 list_del_init(&req->intr_entry);
943 req->intr_unique = fuse_get_unique(fc);
944 memset(&ih, 0, sizeof(ih));
945 memset(&arg, 0, sizeof(arg));
946 ih.len = reqsize;
947 ih.opcode = FUSE_INTERRUPT;
948 ih.unique = req->intr_unique;
949 arg.unique = req->in.h.unique;
950
951 spin_unlock(&fc->lock);
Miklos Szeredic3021622010-05-25 15:06:07 +0200952 if (nbytes < reqsize)
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700953 return -EINVAL;
954
Miklos Szeredic3021622010-05-25 15:06:07 +0200955 err = fuse_copy_one(cs, &ih, sizeof(ih));
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700956 if (!err)
Miklos Szeredic3021622010-05-25 15:06:07 +0200957 err = fuse_copy_one(cs, &arg, sizeof(arg));
958 fuse_copy_finish(cs);
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700959
960 return err ? err : reqsize;
961}
962
963/*
Miklos Szeredi334f4852005-09-09 13:10:27 -0700964 * Read a single request into the userspace filesystem's buffer. This
965 * function waits until a request is available, then removes it from
966 * the pending list and copies request data to userspace buffer. If
Miklos Szeredif9a28422006-06-25 05:48:53 -0700967 * no reply is needed (FORGET) or request has been aborted or there
968 * was an error during the copying then it's finished by calling
Miklos Szeredi334f4852005-09-09 13:10:27 -0700969 * request_end(). Otherwise add it to the processing list, and set
970 * the 'sent' flag.
971 */
Miklos Szeredic3021622010-05-25 15:06:07 +0200972static ssize_t fuse_dev_do_read(struct fuse_conn *fc, struct file *file,
973 struct fuse_copy_state *cs, size_t nbytes)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700974{
975 int err;
Miklos Szeredi334f4852005-09-09 13:10:27 -0700976 struct fuse_req *req;
977 struct fuse_in *in;
Miklos Szeredi334f4852005-09-09 13:10:27 -0700978 unsigned reqsize;
979
Miklos Szeredi1d3d7522006-01-06 00:19:40 -0800980 restart:
Miklos Szeredid7133112006-04-10 22:54:55 -0700981 spin_lock(&fc->lock);
Jeff Dikee5ac1d12006-04-10 22:54:53 -0700982 err = -EAGAIN;
983 if ((file->f_flags & O_NONBLOCK) && fc->connected &&
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700984 !request_pending(fc))
Jeff Dikee5ac1d12006-04-10 22:54:53 -0700985 goto err_unlock;
986
Miklos Szeredi334f4852005-09-09 13:10:27 -0700987 request_wait(fc);
988 err = -ENODEV;
Miklos Szeredi9ba7cbb2006-01-16 22:14:34 -0800989 if (!fc->connected)
Miklos Szeredi334f4852005-09-09 13:10:27 -0700990 goto err_unlock;
991 err = -ERESTARTSYS;
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700992 if (!request_pending(fc))
Miklos Szeredi334f4852005-09-09 13:10:27 -0700993 goto err_unlock;
994
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700995 if (!list_empty(&fc->interrupts)) {
996 req = list_entry(fc->interrupts.next, struct fuse_req,
997 intr_entry);
Miklos Szeredic3021622010-05-25 15:06:07 +0200998 return fuse_read_interrupt(fc, cs, nbytes, req);
Miklos Szeredia4d27e72006-06-25 05:48:54 -0700999 }
1000
Miklos Szeredi334f4852005-09-09 13:10:27 -07001001 req = list_entry(fc->pending.next, struct fuse_req, list);
Miklos Szeredi83cfd492006-01-16 22:14:31 -08001002 req->state = FUSE_REQ_READING;
Miklos Szeredid77a1d52006-01-16 22:14:31 -08001003 list_move(&req->list, &fc->io);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001004
1005 in = &req->in;
Miklos Szeredi1d3d7522006-01-06 00:19:40 -08001006 reqsize = in->h.len;
1007 /* If request is too large, reply with an error and restart the read */
Miklos Szeredic3021622010-05-25 15:06:07 +02001008 if (nbytes < reqsize) {
Miklos Szeredi1d3d7522006-01-06 00:19:40 -08001009 req->out.h.error = -EIO;
1010 /* SETXATTR is special, since it may contain too large data */
1011 if (in->h.opcode == FUSE_SETXATTR)
1012 req->out.h.error = -E2BIG;
1013 request_end(fc, req);
1014 goto restart;
Miklos Szeredi334f4852005-09-09 13:10:27 -07001015 }
Miklos Szeredid7133112006-04-10 22:54:55 -07001016 spin_unlock(&fc->lock);
Miklos Szeredic3021622010-05-25 15:06:07 +02001017 cs->req = req;
1018 err = fuse_copy_one(cs, &in->h, sizeof(in->h));
Miklos Szeredi1d3d7522006-01-06 00:19:40 -08001019 if (!err)
Miklos Szeredic3021622010-05-25 15:06:07 +02001020 err = fuse_copy_args(cs, in->numargs, in->argpages,
Miklos Szeredi1d3d7522006-01-06 00:19:40 -08001021 (struct fuse_arg *) in->args, 0);
Miklos Szeredic3021622010-05-25 15:06:07 +02001022 fuse_copy_finish(cs);
Miklos Szeredid7133112006-04-10 22:54:55 -07001023 spin_lock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001024 req->locked = 0;
Miklos Szeredic9c9d7d2007-10-16 23:31:05 -07001025 if (req->aborted) {
1026 request_end(fc, req);
1027 return -ENODEV;
1028 }
Miklos Szeredi334f4852005-09-09 13:10:27 -07001029 if (err) {
Miklos Szeredic9c9d7d2007-10-16 23:31:05 -07001030 req->out.h.error = -EIO;
Miklos Szeredi334f4852005-09-09 13:10:27 -07001031 request_end(fc, req);
1032 return err;
1033 }
1034 if (!req->isreply)
1035 request_end(fc, req);
1036 else {
Miklos Szeredi83cfd492006-01-16 22:14:31 -08001037 req->state = FUSE_REQ_SENT;
Miklos Szeredid77a1d52006-01-16 22:14:31 -08001038 list_move_tail(&req->list, &fc->processing);
Miklos Szeredia4d27e72006-06-25 05:48:54 -07001039 if (req->interrupted)
1040 queue_interrupt(fc, req);
Miklos Szeredid7133112006-04-10 22:54:55 -07001041 spin_unlock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001042 }
1043 return reqsize;
1044
1045 err_unlock:
Miklos Szeredid7133112006-04-10 22:54:55 -07001046 spin_unlock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001047 return err;
1048}
1049
Miklos Szeredic3021622010-05-25 15:06:07 +02001050static ssize_t fuse_dev_read(struct kiocb *iocb, const struct iovec *iov,
1051 unsigned long nr_segs, loff_t pos)
1052{
1053 struct fuse_copy_state cs;
1054 struct file *file = iocb->ki_filp;
1055 struct fuse_conn *fc = fuse_get_conn(file);
1056 if (!fc)
1057 return -EPERM;
1058
1059 fuse_copy_init(&cs, fc, 1, iov, nr_segs);
1060
1061 return fuse_dev_do_read(fc, file, &cs, iov_length(iov, nr_segs));
1062}
1063
1064static int fuse_dev_pipe_buf_steal(struct pipe_inode_info *pipe,
1065 struct pipe_buffer *buf)
1066{
1067 return 1;
1068}
1069
1070static const struct pipe_buf_operations fuse_dev_pipe_buf_ops = {
1071 .can_merge = 0,
1072 .map = generic_pipe_buf_map,
1073 .unmap = generic_pipe_buf_unmap,
1074 .confirm = generic_pipe_buf_confirm,
1075 .release = generic_pipe_buf_release,
1076 .steal = fuse_dev_pipe_buf_steal,
1077 .get = generic_pipe_buf_get,
1078};
1079
1080static ssize_t fuse_dev_splice_read(struct file *in, loff_t *ppos,
1081 struct pipe_inode_info *pipe,
1082 size_t len, unsigned int flags)
1083{
1084 int ret;
1085 int page_nr = 0;
1086 int do_wakeup = 0;
1087 struct pipe_buffer *bufs;
1088 struct fuse_copy_state cs;
1089 struct fuse_conn *fc = fuse_get_conn(in);
1090 if (!fc)
1091 return -EPERM;
1092
1093 bufs = kmalloc(pipe->buffers * sizeof (struct pipe_buffer), GFP_KERNEL);
1094 if (!bufs)
1095 return -ENOMEM;
1096
1097 fuse_copy_init(&cs, fc, 1, NULL, 0);
1098 cs.pipebufs = bufs;
1099 cs.pipe = pipe;
1100 ret = fuse_dev_do_read(fc, in, &cs, len);
1101 if (ret < 0)
1102 goto out;
1103
1104 ret = 0;
1105 pipe_lock(pipe);
1106
1107 if (!pipe->readers) {
1108 send_sig(SIGPIPE, current, 0);
1109 if (!ret)
1110 ret = -EPIPE;
1111 goto out_unlock;
1112 }
1113
1114 if (pipe->nrbufs + cs.nr_segs > pipe->buffers) {
1115 ret = -EIO;
1116 goto out_unlock;
1117 }
1118
1119 while (page_nr < cs.nr_segs) {
1120 int newbuf = (pipe->curbuf + pipe->nrbufs) & (pipe->buffers - 1);
1121 struct pipe_buffer *buf = pipe->bufs + newbuf;
1122
1123 buf->page = bufs[page_nr].page;
1124 buf->offset = bufs[page_nr].offset;
1125 buf->len = bufs[page_nr].len;
1126 buf->ops = &fuse_dev_pipe_buf_ops;
1127
1128 pipe->nrbufs++;
1129 page_nr++;
1130 ret += buf->len;
1131
1132 if (pipe->inode)
1133 do_wakeup = 1;
1134 }
1135
1136out_unlock:
1137 pipe_unlock(pipe);
1138
1139 if (do_wakeup) {
1140 smp_mb();
1141 if (waitqueue_active(&pipe->wait))
1142 wake_up_interruptible(&pipe->wait);
1143 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1144 }
1145
1146out:
1147 for (; page_nr < cs.nr_segs; page_nr++)
1148 page_cache_release(bufs[page_nr].page);
1149
1150 kfree(bufs);
1151 return ret;
1152}
1153
Tejun Heo95668a62008-11-26 12:03:55 +01001154static int fuse_notify_poll(struct fuse_conn *fc, unsigned int size,
1155 struct fuse_copy_state *cs)
1156{
1157 struct fuse_notify_poll_wakeup_out outarg;
Miklos Szeredif6d47a12009-01-26 15:00:59 +01001158 int err = -EINVAL;
Tejun Heo95668a62008-11-26 12:03:55 +01001159
1160 if (size != sizeof(outarg))
Miklos Szeredif6d47a12009-01-26 15:00:59 +01001161 goto err;
Tejun Heo95668a62008-11-26 12:03:55 +01001162
1163 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1164 if (err)
Miklos Szeredif6d47a12009-01-26 15:00:59 +01001165 goto err;
Tejun Heo95668a62008-11-26 12:03:55 +01001166
Miklos Szeredif6d47a12009-01-26 15:00:59 +01001167 fuse_copy_finish(cs);
Tejun Heo95668a62008-11-26 12:03:55 +01001168 return fuse_notify_poll_wakeup(fc, &outarg);
Miklos Szeredif6d47a12009-01-26 15:00:59 +01001169
1170err:
1171 fuse_copy_finish(cs);
1172 return err;
Tejun Heo95668a62008-11-26 12:03:55 +01001173}
1174
John Muir3b463ae2009-05-31 11:13:57 -04001175static int fuse_notify_inval_inode(struct fuse_conn *fc, unsigned int size,
1176 struct fuse_copy_state *cs)
1177{
1178 struct fuse_notify_inval_inode_out outarg;
1179 int err = -EINVAL;
1180
1181 if (size != sizeof(outarg))
1182 goto err;
1183
1184 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1185 if (err)
1186 goto err;
1187 fuse_copy_finish(cs);
1188
1189 down_read(&fc->killsb);
1190 err = -ENOENT;
Miklos Szeredib21dda42010-02-05 12:08:31 +01001191 if (fc->sb) {
1192 err = fuse_reverse_inval_inode(fc->sb, outarg.ino,
1193 outarg.off, outarg.len);
1194 }
John Muir3b463ae2009-05-31 11:13:57 -04001195 up_read(&fc->killsb);
1196 return err;
1197
1198err:
1199 fuse_copy_finish(cs);
1200 return err;
1201}
1202
1203static int fuse_notify_inval_entry(struct fuse_conn *fc, unsigned int size,
1204 struct fuse_copy_state *cs)
1205{
1206 struct fuse_notify_inval_entry_out outarg;
Fang Wenqib2d82ee2009-12-30 18:37:13 +08001207 int err = -ENOMEM;
1208 char *buf;
John Muir3b463ae2009-05-31 11:13:57 -04001209 struct qstr name;
1210
Fang Wenqib2d82ee2009-12-30 18:37:13 +08001211 buf = kzalloc(FUSE_NAME_MAX + 1, GFP_KERNEL);
1212 if (!buf)
1213 goto err;
1214
1215 err = -EINVAL;
John Muir3b463ae2009-05-31 11:13:57 -04001216 if (size < sizeof(outarg))
1217 goto err;
1218
1219 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1220 if (err)
1221 goto err;
1222
1223 err = -ENAMETOOLONG;
1224 if (outarg.namelen > FUSE_NAME_MAX)
1225 goto err;
1226
1227 name.name = buf;
1228 name.len = outarg.namelen;
1229 err = fuse_copy_one(cs, buf, outarg.namelen + 1);
1230 if (err)
1231 goto err;
1232 fuse_copy_finish(cs);
1233 buf[outarg.namelen] = 0;
1234 name.hash = full_name_hash(name.name, name.len);
1235
1236 down_read(&fc->killsb);
1237 err = -ENOENT;
Miklos Szeredib21dda42010-02-05 12:08:31 +01001238 if (fc->sb)
1239 err = fuse_reverse_inval_entry(fc->sb, outarg.parent, &name);
John Muir3b463ae2009-05-31 11:13:57 -04001240 up_read(&fc->killsb);
Fang Wenqib2d82ee2009-12-30 18:37:13 +08001241 kfree(buf);
John Muir3b463ae2009-05-31 11:13:57 -04001242 return err;
1243
1244err:
Fang Wenqib2d82ee2009-12-30 18:37:13 +08001245 kfree(buf);
John Muir3b463ae2009-05-31 11:13:57 -04001246 fuse_copy_finish(cs);
1247 return err;
1248}
1249
Miklos Szeredia1d75f22010-07-12 14:41:40 +02001250static int fuse_notify_store(struct fuse_conn *fc, unsigned int size,
1251 struct fuse_copy_state *cs)
1252{
1253 struct fuse_notify_store_out outarg;
1254 struct inode *inode;
1255 struct address_space *mapping;
1256 u64 nodeid;
1257 int err;
1258 pgoff_t index;
1259 unsigned int offset;
1260 unsigned int num;
1261 loff_t file_size;
1262 loff_t end;
1263
1264 err = -EINVAL;
1265 if (size < sizeof(outarg))
1266 goto out_finish;
1267
1268 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1269 if (err)
1270 goto out_finish;
1271
1272 err = -EINVAL;
1273 if (size - sizeof(outarg) != outarg.size)
1274 goto out_finish;
1275
1276 nodeid = outarg.nodeid;
1277
1278 down_read(&fc->killsb);
1279
1280 err = -ENOENT;
1281 if (!fc->sb)
1282 goto out_up_killsb;
1283
1284 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1285 if (!inode)
1286 goto out_up_killsb;
1287
1288 mapping = inode->i_mapping;
1289 index = outarg.offset >> PAGE_CACHE_SHIFT;
1290 offset = outarg.offset & ~PAGE_CACHE_MASK;
1291 file_size = i_size_read(inode);
1292 end = outarg.offset + outarg.size;
1293 if (end > file_size) {
1294 file_size = end;
1295 fuse_write_update_size(inode, file_size);
1296 }
1297
1298 num = outarg.size;
1299 while (num) {
1300 struct page *page;
1301 unsigned int this_num;
1302
1303 err = -ENOMEM;
1304 page = find_or_create_page(mapping, index,
1305 mapping_gfp_mask(mapping));
1306 if (!page)
1307 goto out_iput;
1308
1309 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1310 err = fuse_copy_page(cs, &page, offset, this_num, 0);
1311 if (!err && offset == 0 && (num != 0 || file_size == end))
1312 SetPageUptodate(page);
1313 unlock_page(page);
1314 page_cache_release(page);
1315
1316 if (err)
1317 goto out_iput;
1318
1319 num -= this_num;
1320 offset = 0;
1321 index++;
1322 }
1323
1324 err = 0;
1325
1326out_iput:
1327 iput(inode);
1328out_up_killsb:
1329 up_read(&fc->killsb);
1330out_finish:
1331 fuse_copy_finish(cs);
1332 return err;
1333}
1334
Miklos Szeredi2d45ba32010-07-12 14:41:40 +02001335static void fuse_retrieve_end(struct fuse_conn *fc, struct fuse_req *req)
1336{
Miklos Szeredi0be85572010-10-27 15:34:46 -07001337 release_pages(req->pages, req->num_pages, 0);
Miklos Szeredi2d45ba32010-07-12 14:41:40 +02001338}
1339
1340static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode,
1341 struct fuse_notify_retrieve_out *outarg)
1342{
1343 int err;
1344 struct address_space *mapping = inode->i_mapping;
1345 struct fuse_req *req;
1346 pgoff_t index;
1347 loff_t file_size;
1348 unsigned int num;
1349 unsigned int offset;
Geert Uytterhoeven01574432010-09-30 22:06:21 +02001350 size_t total_len = 0;
Miklos Szeredi2d45ba32010-07-12 14:41:40 +02001351
1352 req = fuse_get_req(fc);
1353 if (IS_ERR(req))
1354 return PTR_ERR(req);
1355
1356 offset = outarg->offset & ~PAGE_CACHE_MASK;
1357
1358 req->in.h.opcode = FUSE_NOTIFY_REPLY;
1359 req->in.h.nodeid = outarg->nodeid;
1360 req->in.numargs = 2;
1361 req->in.argpages = 1;
1362 req->page_offset = offset;
1363 req->end = fuse_retrieve_end;
1364
1365 index = outarg->offset >> PAGE_CACHE_SHIFT;
1366 file_size = i_size_read(inode);
1367 num = outarg->size;
1368 if (outarg->offset > file_size)
1369 num = 0;
1370 else if (outarg->offset + num > file_size)
1371 num = file_size - outarg->offset;
1372
1373 while (num) {
1374 struct page *page;
1375 unsigned int this_num;
1376
1377 page = find_get_page(mapping, index);
1378 if (!page)
1379 break;
1380
1381 this_num = min_t(unsigned, num, PAGE_CACHE_SIZE - offset);
1382 req->pages[req->num_pages] = page;
1383 req->num_pages++;
1384
1385 num -= this_num;
1386 total_len += this_num;
1387 }
1388 req->misc.retrieve_in.offset = outarg->offset;
1389 req->misc.retrieve_in.size = total_len;
1390 req->in.args[0].size = sizeof(req->misc.retrieve_in);
1391 req->in.args[0].value = &req->misc.retrieve_in;
1392 req->in.args[1].size = total_len;
1393
1394 err = fuse_request_send_notify_reply(fc, req, outarg->notify_unique);
1395 if (err)
1396 fuse_retrieve_end(fc, req);
1397
1398 return err;
1399}
1400
1401static int fuse_notify_retrieve(struct fuse_conn *fc, unsigned int size,
1402 struct fuse_copy_state *cs)
1403{
1404 struct fuse_notify_retrieve_out outarg;
1405 struct inode *inode;
1406 int err;
1407
1408 err = -EINVAL;
1409 if (size != sizeof(outarg))
1410 goto copy_finish;
1411
1412 err = fuse_copy_one(cs, &outarg, sizeof(outarg));
1413 if (err)
1414 goto copy_finish;
1415
1416 fuse_copy_finish(cs);
1417
1418 down_read(&fc->killsb);
1419 err = -ENOENT;
1420 if (fc->sb) {
1421 u64 nodeid = outarg.nodeid;
1422
1423 inode = ilookup5(fc->sb, nodeid, fuse_inode_eq, &nodeid);
1424 if (inode) {
1425 err = fuse_retrieve(fc, inode, &outarg);
1426 iput(inode);
1427 }
1428 }
1429 up_read(&fc->killsb);
1430
1431 return err;
1432
1433copy_finish:
1434 fuse_copy_finish(cs);
1435 return err;
1436}
1437
Tejun Heo85993962008-11-26 12:03:55 +01001438static int fuse_notify(struct fuse_conn *fc, enum fuse_notify_code code,
1439 unsigned int size, struct fuse_copy_state *cs)
1440{
1441 switch (code) {
Tejun Heo95668a62008-11-26 12:03:55 +01001442 case FUSE_NOTIFY_POLL:
1443 return fuse_notify_poll(fc, size, cs);
1444
John Muir3b463ae2009-05-31 11:13:57 -04001445 case FUSE_NOTIFY_INVAL_INODE:
1446 return fuse_notify_inval_inode(fc, size, cs);
1447
1448 case FUSE_NOTIFY_INVAL_ENTRY:
1449 return fuse_notify_inval_entry(fc, size, cs);
1450
Miklos Szeredia1d75f22010-07-12 14:41:40 +02001451 case FUSE_NOTIFY_STORE:
1452 return fuse_notify_store(fc, size, cs);
1453
Miklos Szeredi2d45ba32010-07-12 14:41:40 +02001454 case FUSE_NOTIFY_RETRIEVE:
1455 return fuse_notify_retrieve(fc, size, cs);
1456
Tejun Heo85993962008-11-26 12:03:55 +01001457 default:
Miklos Szeredif6d47a12009-01-26 15:00:59 +01001458 fuse_copy_finish(cs);
Tejun Heo85993962008-11-26 12:03:55 +01001459 return -EINVAL;
1460 }
1461}
1462
Miklos Szeredi334f4852005-09-09 13:10:27 -07001463/* Look up request on processing list by unique ID */
1464static struct fuse_req *request_find(struct fuse_conn *fc, u64 unique)
1465{
1466 struct list_head *entry;
1467
1468 list_for_each(entry, &fc->processing) {
1469 struct fuse_req *req;
1470 req = list_entry(entry, struct fuse_req, list);
Miklos Szeredia4d27e72006-06-25 05:48:54 -07001471 if (req->in.h.unique == unique || req->intr_unique == unique)
Miklos Szeredi334f4852005-09-09 13:10:27 -07001472 return req;
1473 }
1474 return NULL;
1475}
1476
1477static int copy_out_args(struct fuse_copy_state *cs, struct fuse_out *out,
1478 unsigned nbytes)
1479{
1480 unsigned reqsize = sizeof(struct fuse_out_header);
1481
1482 if (out->h.error)
1483 return nbytes != reqsize ? -EINVAL : 0;
1484
1485 reqsize += len_args(out->numargs, out->args);
1486
1487 if (reqsize < nbytes || (reqsize > nbytes && !out->argvar))
1488 return -EINVAL;
1489 else if (reqsize > nbytes) {
1490 struct fuse_arg *lastarg = &out->args[out->numargs-1];
1491 unsigned diffsize = reqsize - nbytes;
1492 if (diffsize > lastarg->size)
1493 return -EINVAL;
1494 lastarg->size -= diffsize;
1495 }
1496 return fuse_copy_args(cs, out->numargs, out->argpages, out->args,
1497 out->page_zeroing);
1498}
1499
1500/*
1501 * Write a single reply to a request. First the header is copied from
1502 * the write buffer. The request is then searched on the processing
1503 * list by the unique ID found in the header. If found, then remove
1504 * it from the list and copy the rest of the buffer to the request.
1505 * The request is finished by calling request_end()
1506 */
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001507static ssize_t fuse_dev_do_write(struct fuse_conn *fc,
1508 struct fuse_copy_state *cs, size_t nbytes)
Miklos Szeredi334f4852005-09-09 13:10:27 -07001509{
1510 int err;
Miklos Szeredi334f4852005-09-09 13:10:27 -07001511 struct fuse_req *req;
1512 struct fuse_out_header oh;
Miklos Szeredi334f4852005-09-09 13:10:27 -07001513
Miklos Szeredi334f4852005-09-09 13:10:27 -07001514 if (nbytes < sizeof(struct fuse_out_header))
1515 return -EINVAL;
1516
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001517 err = fuse_copy_one(cs, &oh, sizeof(oh));
Miklos Szeredi334f4852005-09-09 13:10:27 -07001518 if (err)
1519 goto err_finish;
Tejun Heo85993962008-11-26 12:03:55 +01001520
Miklos Szeredi334f4852005-09-09 13:10:27 -07001521 err = -EINVAL;
Tejun Heo85993962008-11-26 12:03:55 +01001522 if (oh.len != nbytes)
1523 goto err_finish;
1524
1525 /*
1526 * Zero oh.unique indicates unsolicited notification message
1527 * and error contains notification code.
1528 */
1529 if (!oh.unique) {
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001530 err = fuse_notify(fc, oh.error, nbytes - sizeof(oh), cs);
Tejun Heo85993962008-11-26 12:03:55 +01001531 return err ? err : nbytes;
1532 }
1533
1534 err = -EINVAL;
1535 if (oh.error <= -1000 || oh.error > 0)
Miklos Szeredi334f4852005-09-09 13:10:27 -07001536 goto err_finish;
1537
Miklos Szeredid7133112006-04-10 22:54:55 -07001538 spin_lock(&fc->lock);
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001539 err = -ENOENT;
1540 if (!fc->connected)
1541 goto err_unlock;
1542
Miklos Szeredi334f4852005-09-09 13:10:27 -07001543 req = request_find(fc, oh.unique);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001544 if (!req)
1545 goto err_unlock;
1546
Miklos Szeredif9a28422006-06-25 05:48:53 -07001547 if (req->aborted) {
Miklos Szeredid7133112006-04-10 22:54:55 -07001548 spin_unlock(&fc->lock);
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001549 fuse_copy_finish(cs);
Miklos Szeredid7133112006-04-10 22:54:55 -07001550 spin_lock(&fc->lock);
Miklos Szeredi222f1d62006-01-16 22:14:25 -08001551 request_end(fc, req);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001552 return -ENOENT;
1553 }
Miklos Szeredia4d27e72006-06-25 05:48:54 -07001554 /* Is it an interrupt reply? */
1555 if (req->intr_unique == oh.unique) {
1556 err = -EINVAL;
1557 if (nbytes != sizeof(struct fuse_out_header))
1558 goto err_unlock;
1559
1560 if (oh.error == -ENOSYS)
1561 fc->no_interrupt = 1;
1562 else if (oh.error == -EAGAIN)
1563 queue_interrupt(fc, req);
1564
1565 spin_unlock(&fc->lock);
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001566 fuse_copy_finish(cs);
Miklos Szeredia4d27e72006-06-25 05:48:54 -07001567 return nbytes;
1568 }
1569
1570 req->state = FUSE_REQ_WRITING;
Miklos Szeredid77a1d52006-01-16 22:14:31 -08001571 list_move(&req->list, &fc->io);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001572 req->out.h = oh;
1573 req->locked = 1;
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001574 cs->req = req;
Miklos Szeredice534fb2010-05-25 15:06:07 +02001575 if (!req->out.page_replace)
1576 cs->move_pages = 0;
Miklos Szeredid7133112006-04-10 22:54:55 -07001577 spin_unlock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001578
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001579 err = copy_out_args(cs, &req->out, nbytes);
1580 fuse_copy_finish(cs);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001581
Miklos Szeredid7133112006-04-10 22:54:55 -07001582 spin_lock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001583 req->locked = 0;
1584 if (!err) {
Miklos Szeredif9a28422006-06-25 05:48:53 -07001585 if (req->aborted)
Miklos Szeredi334f4852005-09-09 13:10:27 -07001586 err = -ENOENT;
Miklos Szeredif9a28422006-06-25 05:48:53 -07001587 } else if (!req->aborted)
Miklos Szeredi334f4852005-09-09 13:10:27 -07001588 req->out.h.error = -EIO;
1589 request_end(fc, req);
1590
1591 return err ? err : nbytes;
1592
1593 err_unlock:
Miklos Szeredid7133112006-04-10 22:54:55 -07001594 spin_unlock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001595 err_finish:
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001596 fuse_copy_finish(cs);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001597 return err;
1598}
1599
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001600static ssize_t fuse_dev_write(struct kiocb *iocb, const struct iovec *iov,
1601 unsigned long nr_segs, loff_t pos)
1602{
1603 struct fuse_copy_state cs;
1604 struct fuse_conn *fc = fuse_get_conn(iocb->ki_filp);
1605 if (!fc)
1606 return -EPERM;
1607
Miklos Szeredic3021622010-05-25 15:06:07 +02001608 fuse_copy_init(&cs, fc, 0, iov, nr_segs);
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001609
1610 return fuse_dev_do_write(fc, &cs, iov_length(iov, nr_segs));
1611}
1612
1613static ssize_t fuse_dev_splice_write(struct pipe_inode_info *pipe,
1614 struct file *out, loff_t *ppos,
1615 size_t len, unsigned int flags)
1616{
1617 unsigned nbuf;
1618 unsigned idx;
1619 struct pipe_buffer *bufs;
1620 struct fuse_copy_state cs;
1621 struct fuse_conn *fc;
1622 size_t rem;
1623 ssize_t ret;
1624
1625 fc = fuse_get_conn(out);
1626 if (!fc)
1627 return -EPERM;
1628
1629 bufs = kmalloc(pipe->buffers * sizeof (struct pipe_buffer), GFP_KERNEL);
1630 if (!bufs)
1631 return -ENOMEM;
1632
1633 pipe_lock(pipe);
1634 nbuf = 0;
1635 rem = 0;
1636 for (idx = 0; idx < pipe->nrbufs && rem < len; idx++)
1637 rem += pipe->bufs[(pipe->curbuf + idx) & (pipe->buffers - 1)].len;
1638
1639 ret = -EINVAL;
1640 if (rem < len) {
1641 pipe_unlock(pipe);
1642 goto out;
1643 }
1644
1645 rem = len;
1646 while (rem) {
1647 struct pipe_buffer *ibuf;
1648 struct pipe_buffer *obuf;
1649
1650 BUG_ON(nbuf >= pipe->buffers);
1651 BUG_ON(!pipe->nrbufs);
1652 ibuf = &pipe->bufs[pipe->curbuf];
1653 obuf = &bufs[nbuf];
1654
1655 if (rem >= ibuf->len) {
1656 *obuf = *ibuf;
1657 ibuf->ops = NULL;
1658 pipe->curbuf = (pipe->curbuf + 1) & (pipe->buffers - 1);
1659 pipe->nrbufs--;
1660 } else {
1661 ibuf->ops->get(pipe, ibuf);
1662 *obuf = *ibuf;
1663 obuf->flags &= ~PIPE_BUF_FLAG_GIFT;
1664 obuf->len = rem;
1665 ibuf->offset += obuf->len;
1666 ibuf->len -= obuf->len;
1667 }
1668 nbuf++;
1669 rem -= obuf->len;
1670 }
1671 pipe_unlock(pipe);
1672
Miklos Szeredic3021622010-05-25 15:06:07 +02001673 fuse_copy_init(&cs, fc, 0, NULL, nbuf);
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001674 cs.pipebufs = bufs;
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001675 cs.pipe = pipe;
1676
Miklos Szeredice534fb2010-05-25 15:06:07 +02001677 if (flags & SPLICE_F_MOVE)
1678 cs.move_pages = 1;
1679
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001680 ret = fuse_dev_do_write(fc, &cs, len);
1681
1682 for (idx = 0; idx < nbuf; idx++) {
1683 struct pipe_buffer *buf = &bufs[idx];
1684 buf->ops->release(pipe, buf);
1685 }
1686out:
1687 kfree(bufs);
1688 return ret;
1689}
1690
Miklos Szeredi334f4852005-09-09 13:10:27 -07001691static unsigned fuse_dev_poll(struct file *file, poll_table *wait)
1692{
Miklos Szeredi334f4852005-09-09 13:10:27 -07001693 unsigned mask = POLLOUT | POLLWRNORM;
Miklos Szeredi7025d9a2006-04-10 22:54:50 -07001694 struct fuse_conn *fc = fuse_get_conn(file);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001695 if (!fc)
Miklos Szeredi7025d9a2006-04-10 22:54:50 -07001696 return POLLERR;
Miklos Szeredi334f4852005-09-09 13:10:27 -07001697
1698 poll_wait(file, &fc->waitq, wait);
1699
Miklos Szeredid7133112006-04-10 22:54:55 -07001700 spin_lock(&fc->lock);
Miklos Szeredi7025d9a2006-04-10 22:54:50 -07001701 if (!fc->connected)
1702 mask = POLLERR;
Miklos Szeredia4d27e72006-06-25 05:48:54 -07001703 else if (request_pending(fc))
Miklos Szeredi7025d9a2006-04-10 22:54:50 -07001704 mask |= POLLIN | POLLRDNORM;
Miklos Szeredid7133112006-04-10 22:54:55 -07001705 spin_unlock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001706
1707 return mask;
1708}
1709
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001710/*
1711 * Abort all requests on the given list (pending or processing)
1712 *
Miklos Szeredid7133112006-04-10 22:54:55 -07001713 * This function releases and reacquires fc->lock
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001714 */
Miklos Szeredi334f4852005-09-09 13:10:27 -07001715static void end_requests(struct fuse_conn *fc, struct list_head *head)
Miklos Szeredib9ca67b2010-09-07 13:42:41 +02001716__releases(fc->lock)
1717__acquires(fc->lock)
Miklos Szeredi334f4852005-09-09 13:10:27 -07001718{
1719 while (!list_empty(head)) {
1720 struct fuse_req *req;
1721 req = list_entry(head->next, struct fuse_req, list);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001722 req->out.h.error = -ECONNABORTED;
1723 request_end(fc, req);
Miklos Szeredid7133112006-04-10 22:54:55 -07001724 spin_lock(&fc->lock);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001725 }
1726}
1727
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001728/*
1729 * Abort requests under I/O
1730 *
Miklos Szeredif9a28422006-06-25 05:48:53 -07001731 * The requests are set to aborted and finished, and the request
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001732 * waiter is woken up. This will make request_wait_answer() wait
1733 * until the request is unlocked and then return.
Miklos Szeredi64c6d8e2006-01-16 22:14:42 -08001734 *
1735 * If the request is asynchronous, then the end function needs to be
1736 * called after waiting for the request to be unlocked (if it was
1737 * locked).
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001738 */
1739static void end_io_requests(struct fuse_conn *fc)
Miklos Szeredib9ca67b2010-09-07 13:42:41 +02001740__releases(fc->lock)
1741__acquires(fc->lock)
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001742{
1743 while (!list_empty(&fc->io)) {
Miklos Szeredi64c6d8e2006-01-16 22:14:42 -08001744 struct fuse_req *req =
1745 list_entry(fc->io.next, struct fuse_req, list);
1746 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
1747
Miklos Szeredif9a28422006-06-25 05:48:53 -07001748 req->aborted = 1;
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001749 req->out.h.error = -ECONNABORTED;
1750 req->state = FUSE_REQ_FINISHED;
1751 list_del_init(&req->list);
1752 wake_up(&req->waitq);
Miklos Szeredi64c6d8e2006-01-16 22:14:42 -08001753 if (end) {
1754 req->end = NULL;
Miklos Szeredi64c6d8e2006-01-16 22:14:42 -08001755 __fuse_get_request(req);
Miklos Szeredid7133112006-04-10 22:54:55 -07001756 spin_unlock(&fc->lock);
Miklos Szeredi64c6d8e2006-01-16 22:14:42 -08001757 wait_event(req->waitq, !req->locked);
1758 end(fc, req);
Tejun Heoe9bb09d2008-11-26 12:03:54 +01001759 fuse_put_request(fc, req);
Miklos Szeredid7133112006-04-10 22:54:55 -07001760 spin_lock(&fc->lock);
Miklos Szeredi64c6d8e2006-01-16 22:14:42 -08001761 }
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001762 }
1763}
1764
Miklos Szeredi595afaf2010-09-07 13:42:41 +02001765static void end_queued_requests(struct fuse_conn *fc)
Miklos Szeredib9ca67b2010-09-07 13:42:41 +02001766__releases(fc->lock)
1767__acquires(fc->lock)
Miklos Szeredi595afaf2010-09-07 13:42:41 +02001768{
1769 fc->max_background = UINT_MAX;
1770 flush_bg_queue(fc);
1771 end_requests(fc, &fc->pending);
1772 end_requests(fc, &fc->processing);
1773}
1774
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001775/*
1776 * Abort all requests.
1777 *
1778 * Emergency exit in case of a malicious or accidental deadlock, or
1779 * just a hung filesystem.
1780 *
1781 * The same effect is usually achievable through killing the
1782 * filesystem daemon and all users of the filesystem. The exception
1783 * is the combination of an asynchronous request and the tricky
1784 * deadlock (see Documentation/filesystems/fuse.txt).
1785 *
1786 * During the aborting, progression of requests from the pending and
1787 * processing lists onto the io list, and progression of new requests
1788 * onto the pending list is prevented by req->connected being false.
1789 *
1790 * Progression of requests under I/O to the processing list is
Miklos Szeredif9a28422006-06-25 05:48:53 -07001791 * prevented by the req->aborted flag being true for these requests.
1792 * For this reason requests on the io list must be aborted first.
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001793 */
1794void fuse_abort_conn(struct fuse_conn *fc)
1795{
Miklos Szeredid7133112006-04-10 22:54:55 -07001796 spin_lock(&fc->lock);
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001797 if (fc->connected) {
1798 fc->connected = 0;
Miklos Szeredi51eb01e2006-06-25 05:48:50 -07001799 fc->blocked = 0;
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001800 end_io_requests(fc);
Miklos Szeredi595afaf2010-09-07 13:42:41 +02001801 end_queued_requests(fc);
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001802 wake_up_all(&fc->waitq);
Miklos Szeredi51eb01e2006-06-25 05:48:50 -07001803 wake_up_all(&fc->blocked_waitq);
Jeff Dike385a17b2006-04-10 22:54:52 -07001804 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001805 }
Miklos Szeredid7133112006-04-10 22:54:55 -07001806 spin_unlock(&fc->lock);
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001807}
Tejun Heo08cbf542009-04-14 10:54:53 +09001808EXPORT_SYMBOL_GPL(fuse_abort_conn);
Miklos Szeredi69a53bf2006-01-16 22:14:41 -08001809
Tejun Heo08cbf542009-04-14 10:54:53 +09001810int fuse_dev_release(struct inode *inode, struct file *file)
Miklos Szeredi334f4852005-09-09 13:10:27 -07001811{
Miklos Szeredi0720b312006-04-10 22:54:55 -07001812 struct fuse_conn *fc = fuse_get_conn(file);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001813 if (fc) {
Miklos Szeredid7133112006-04-10 22:54:55 -07001814 spin_lock(&fc->lock);
Miklos Szeredi1e9a4ed2005-09-09 13:10:31 -07001815 fc->connected = 0;
Miklos Szeredi595afaf2010-09-07 13:42:41 +02001816 fc->blocked = 0;
1817 end_queued_requests(fc);
1818 wake_up_all(&fc->blocked_waitq);
Miklos Szeredid7133112006-04-10 22:54:55 -07001819 spin_unlock(&fc->lock);
Miklos Szeredibafa9652006-06-25 05:48:51 -07001820 fuse_conn_put(fc);
Jeff Dike385a17b2006-04-10 22:54:52 -07001821 }
Miklos Szeredif543f252006-01-16 22:14:35 -08001822
Miklos Szeredi334f4852005-09-09 13:10:27 -07001823 return 0;
1824}
Tejun Heo08cbf542009-04-14 10:54:53 +09001825EXPORT_SYMBOL_GPL(fuse_dev_release);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001826
Jeff Dike385a17b2006-04-10 22:54:52 -07001827static int fuse_dev_fasync(int fd, struct file *file, int on)
1828{
1829 struct fuse_conn *fc = fuse_get_conn(file);
1830 if (!fc)
Miklos Szeredia87046d2006-04-10 22:54:56 -07001831 return -EPERM;
Jeff Dike385a17b2006-04-10 22:54:52 -07001832
1833 /* No locking - fasync_helper does its own locking */
1834 return fasync_helper(fd, file, on, &fc->fasync);
1835}
1836
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001837const struct file_operations fuse_dev_operations = {
Miklos Szeredi334f4852005-09-09 13:10:27 -07001838 .owner = THIS_MODULE,
1839 .llseek = no_llseek,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -07001840 .read = do_sync_read,
1841 .aio_read = fuse_dev_read,
Miklos Szeredic3021622010-05-25 15:06:07 +02001842 .splice_read = fuse_dev_splice_read,
Badari Pulavartyee0b3e62006-09-30 23:28:47 -07001843 .write = do_sync_write,
1844 .aio_write = fuse_dev_write,
Miklos Szeredidd3bb142010-05-25 15:06:06 +02001845 .splice_write = fuse_dev_splice_write,
Miklos Szeredi334f4852005-09-09 13:10:27 -07001846 .poll = fuse_dev_poll,
1847 .release = fuse_dev_release,
Jeff Dike385a17b2006-04-10 22:54:52 -07001848 .fasync = fuse_dev_fasync,
Miklos Szeredi334f4852005-09-09 13:10:27 -07001849};
Tejun Heo08cbf542009-04-14 10:54:53 +09001850EXPORT_SYMBOL_GPL(fuse_dev_operations);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001851
1852static struct miscdevice fuse_miscdevice = {
1853 .minor = FUSE_MINOR,
1854 .name = "fuse",
1855 .fops = &fuse_dev_operations,
1856};
1857
1858int __init fuse_dev_init(void)
1859{
1860 int err = -ENOMEM;
1861 fuse_req_cachep = kmem_cache_create("fuse_request",
1862 sizeof(struct fuse_req),
Paul Mundt20c2df82007-07-20 10:11:58 +09001863 0, 0, NULL);
Miklos Szeredi334f4852005-09-09 13:10:27 -07001864 if (!fuse_req_cachep)
1865 goto out;
1866
1867 err = misc_register(&fuse_miscdevice);
1868 if (err)
1869 goto out_cache_clean;
1870
1871 return 0;
1872
1873 out_cache_clean:
1874 kmem_cache_destroy(fuse_req_cachep);
1875 out:
1876 return err;
1877}
1878
1879void fuse_dev_cleanup(void)
1880{
1881 misc_deregister(&fuse_miscdevice);
1882 kmem_cache_destroy(fuse_req_cachep);
1883}