aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorMiklos Szeredi <miklos@szeredi.hu>2006-06-25 08:48:50 -0400
committerLinus Torvalds <torvalds@g5.osdl.org>2006-06-25 13:01:19 -0400
commit51eb01e73599efb88c6c20b1c226d20309a75450 (patch)
treec37e7a61ce7c049a4615dfb2ee994866d3c85bd6
parent3e8c54fad89144b8d63cc41619f363df1ec7cc42 (diff)
[PATCH] fuse: no backgrounding on interrupt
Don't put requests into the background when a fatal interrupt occurs while the request is in userspace. This removes a major wart from the implementation. Backgrounding of requests was introduced to allow breaking of deadlocks. However now the same can be achieved by aborting the filesystem through the 'abort' sysfs attribute. This is a change in the interface, but should not cause problems, since these kinds of deadlocks never happen during normal operation. Signed-off-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
-rw-r--r--Documentation/filesystems/fuse.txt40
-rw-r--r--fs/fuse/dev.c157
-rw-r--r--fs/fuse/dir.c56
-rw-r--r--fs/fuse/file.c49
-rw-r--r--fs/fuse/fuse_i.h49
-rw-r--r--fs/fuse/inode.c14
6 files changed, 112 insertions, 253 deletions
diff --git a/Documentation/filesystems/fuse.txt b/Documentation/filesystems/fuse.txt
index 33f74310d161..e7747774ceb9 100644
--- a/Documentation/filesystems/fuse.txt
+++ b/Documentation/filesystems/fuse.txt
@@ -304,25 +304,7 @@ Scenario 1 - Simple deadlock
304 | | for "file"] 304 | | for "file"]
305 | | *DEADLOCK* 305 | | *DEADLOCK*
306 306
307The solution for this is to allow requests to be interrupted while 307The solution for this is to allow the filesystem to be aborted.
308they are in userspace:
309
310 | [interrupted by signal] |
311 | <fuse_unlink() |
312 | [release semaphore] | [semaphore acquired]
313 | <sys_unlink() |
314 | | >fuse_unlink()
315 | | [queue req on fc->pending]
316 | | [wake up fc->waitq]
317 | | [sleep on req->waitq]
318
319If the filesystem daemon was single threaded, this will stop here,
320since there's no other thread to dequeue and execute the request.
321In this case the solution is to kill the FUSE daemon as well. If
322there are multiple serving threads, you just have to kill them as
323long as any remain.
324
325Moral: a filesystem which deadlocks, can soon find itself dead.
326 308
327Scenario 2 - Tricky deadlock 309Scenario 2 - Tricky deadlock
328---------------------------- 310----------------------------
@@ -355,24 +337,14 @@ but is caused by a pagefault.
355 | | [lock page] 337 | | [lock page]
356 | | * DEADLOCK * 338 | | * DEADLOCK *
357 339
358Solution is again to let the the request be interrupted (not 340Solution is basically the same as above.
359elaborated further).
360 341
361An additional problem is that while the write buffer is being 342An additional problem is that while the write buffer is being
362copied to the request, the request must not be interrupted. This 343copied to the request, the request must not be interrupted. This
363is because the destination address of the copy may not be valid 344is because the destination address of the copy may not be valid
364after the request is interrupted. 345after the request is interrupted.
365 346
366This is solved with doing the copy atomically, and allowing 347This is solved with doing the copy atomically, and allowing abort
367interruption while the page(s) belonging to the write buffer are 348while the page(s) belonging to the write buffer are faulted with
368faulted with get_user_pages(). The 'req->locked' flag indicates 349get_user_pages(). The 'req->locked' flag indicates when the copy is
369when the copy is taking place, and interruption is delayed until 350taking place, and abort is delayed until this flag is unset.
370this flag is unset.
371
372Scenario 3 - Tricky deadlock with asynchronous read
373---------------------------------------------------
374
375The same situation as above, except thread-1 will wait on page lock
376and hence it will be uninterruptible as well. The solution is to
377abort the connection with forced umount (if mount is attached) or
378through the abort attribute in sysfs.
diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
index 104a62dadb94..fec4779e2b55 100644
--- a/fs/fuse/dev.c
+++ b/fs/fuse/dev.c
@@ -64,18 +64,6 @@ static void restore_sigs(sigset_t *oldset)
64 sigprocmask(SIG_SETMASK, oldset, NULL); 64 sigprocmask(SIG_SETMASK, oldset, NULL);
65} 65}
66 66
67/*
68 * Reset request, so that it can be reused
69 *
70 * The caller must be _very_ careful to make sure, that it is holding
71 * the only reference to req
72 */
73void fuse_reset_request(struct fuse_req *req)
74{
75 BUG_ON(atomic_read(&req->count) != 1);
76 fuse_request_init(req);
77}
78
79static void __fuse_get_request(struct fuse_req *req) 67static void __fuse_get_request(struct fuse_req *req)
80{ 68{
81 atomic_inc(&req->count); 69 atomic_inc(&req->count);
@@ -103,6 +91,10 @@ struct fuse_req *fuse_get_req(struct fuse_conn *fc)
103 if (intr) 91 if (intr)
104 goto out; 92 goto out;
105 93
94 err = -ENOTCONN;
95 if (!fc->connected)
96 goto out;
97
106 req = fuse_request_alloc(); 98 req = fuse_request_alloc();
107 err = -ENOMEM; 99 err = -ENOMEM;
108 if (!req) 100 if (!req)
@@ -129,113 +121,38 @@ void fuse_put_request(struct fuse_conn *fc, struct fuse_req *req)
129} 121}
130 122
131/* 123/*
132 * Called with sbput_sem held for read (request_end) or write
133 * (fuse_put_super). By the time fuse_put_super() is finished, all
134 * inodes belonging to background requests must be released, so the
135 * iputs have to be done within the locked region.
136 */
137void fuse_release_background(struct fuse_conn *fc, struct fuse_req *req)
138{
139 iput(req->inode);
140 iput(req->inode2);
141 spin_lock(&fc->lock);
142 list_del(&req->bg_entry);
143 if (fc->num_background == FUSE_MAX_BACKGROUND) {
144 fc->blocked = 0;
145 wake_up_all(&fc->blocked_waitq);
146 }
147 fc->num_background--;
148 spin_unlock(&fc->lock);
149}
150
151/*
152 * This function is called when a request is finished. Either a reply 124 * This function is called when a request is finished. Either a reply
153 * has arrived or it was interrupted (and not yet sent) or some error 125 * has arrived or it was interrupted (and not yet sent) or some error
154 * occurred during communication with userspace, or the device file 126 * occurred during communication with userspace, or the device file
155 * was closed. In case of a background request the reference to the 127 * was closed. The requester thread is woken up (if still waiting),
156 * stored objects are released. The requester thread is woken up (if 128 * the 'end' callback is called if given, else the reference to the
157 * still waiting), the 'end' callback is called if given, else the 129 * request is released
158 * reference to the request is released
159 *
160 * Releasing extra reference for foreground requests must be done
161 * within the same locked region as setting state to finished. This
162 * is because fuse_reset_request() may be called after request is
163 * finished and it must be the sole possessor. If request is
164 * interrupted and put in the background, it will return with an error
165 * and hence never be reset and reused.
166 * 130 *
167 * Called with fc->lock, unlocks it 131 * Called with fc->lock, unlocks it
168 */ 132 */
169static void request_end(struct fuse_conn *fc, struct fuse_req *req) 133static void request_end(struct fuse_conn *fc, struct fuse_req *req)
170{ 134{
135 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end;
136 req->end = NULL;
171 list_del(&req->list); 137 list_del(&req->list);
172 req->state = FUSE_REQ_FINISHED; 138 req->state = FUSE_REQ_FINISHED;
173 if (!req->background) { 139 if (req->background) {
174 spin_unlock(&fc->lock); 140 if (fc->num_background == FUSE_MAX_BACKGROUND) {
175 wake_up(&req->waitq); 141 fc->blocked = 0;
176 fuse_put_request(fc, req); 142 wake_up_all(&fc->blocked_waitq);
177 } else { 143 }
178 void (*end) (struct fuse_conn *, struct fuse_req *) = req->end; 144 fc->num_background--;
179 req->end = NULL;
180 spin_unlock(&fc->lock);
181 down_read(&fc->sbput_sem);
182 if (fc->mounted)
183 fuse_release_background(fc, req);
184 up_read(&fc->sbput_sem);
185
186 /* fput must go outside sbput_sem, otherwise it can deadlock */
187 if (req->file)
188 fput(req->file);
189
190 if (end)
191 end(fc, req);
192 else
193 fuse_put_request(fc, req);
194 } 145 }
195} 146 spin_unlock(&fc->lock);
196 147 dput(req->dentry);
197/* 148 mntput(req->vfsmount);
198 * Unfortunately request interruption not just solves the deadlock
199 * problem, it causes problems too. These stem from the fact, that an
200 * interrupted request is continued to be processed in userspace,
201 * while all the locks and object references (inode and file) held
202 * during the operation are released.
203 *
204 * To release the locks is exactly why there's a need to interrupt the
205 * request, so there's not a lot that can be done about this, except
206 * introduce additional locking in userspace.
207 *
208 * More important is to keep inode and file references until userspace
209 * has replied, otherwise FORGET and RELEASE could be sent while the
210 * inode/file is still used by the filesystem.
211 *
212 * For this reason the concept of "background" request is introduced.
213 * An interrupted request is backgrounded if it has been already sent
214 * to userspace. Backgrounding involves getting an extra reference to
215 * inode(s) or file used in the request, and adding the request to
216 * fc->background list. When a reply is received for a background
217 * request, the object references are released, and the request is
218 * removed from the list. If the filesystem is unmounted while there
219 * are still background requests, the list is walked and references
220 * are released as if a reply was received.
221 *
222 * There's one more use for a background request. The RELEASE message is
223 * always sent as background, since it doesn't return an error or
224 * data.
225 */
226static void background_request(struct fuse_conn *fc, struct fuse_req *req)
227{
228 req->background = 1;
229 list_add(&req->bg_entry, &fc->background);
230 fc->num_background++;
231 if (fc->num_background == FUSE_MAX_BACKGROUND)
232 fc->blocked = 1;
233 if (req->inode)
234 req->inode = igrab(req->inode);
235 if (req->inode2)
236 req->inode2 = igrab(req->inode2);
237 if (req->file) 149 if (req->file)
238 get_file(req->file); 150 fput(req->file);
151 wake_up(&req->waitq);
152 if (end)
153 end(fc, req);
154 else
155 fuse_put_request(fc, req);
239} 156}
240 157
241/* Called with fc->lock held. Releases, and then reacquires it. */ 158/* Called with fc->lock held. Releases, and then reacquires it. */
@@ -244,9 +161,14 @@ static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
244 sigset_t oldset; 161 sigset_t oldset;
245 162
246 spin_unlock(&fc->lock); 163 spin_unlock(&fc->lock);
247 block_sigs(&oldset); 164 if (req->force)
248 wait_event_interruptible(req->waitq, req->state == FUSE_REQ_FINISHED); 165 wait_event(req->waitq, req->state == FUSE_REQ_FINISHED);
249 restore_sigs(&oldset); 166 else {
167 block_sigs(&oldset);
168 wait_event_interruptible(req->waitq,
169 req->state == FUSE_REQ_FINISHED);
170 restore_sigs(&oldset);
171 }
250 spin_lock(&fc->lock); 172 spin_lock(&fc->lock);
251 if (req->state == FUSE_REQ_FINISHED && !req->interrupted) 173 if (req->state == FUSE_REQ_FINISHED && !req->interrupted)
252 return; 174 return;
@@ -268,8 +190,11 @@ static void request_wait_answer(struct fuse_conn *fc, struct fuse_req *req)
268 if (req->state == FUSE_REQ_PENDING) { 190 if (req->state == FUSE_REQ_PENDING) {
269 list_del(&req->list); 191 list_del(&req->list);
270 __fuse_put_request(req); 192 __fuse_put_request(req);
271 } else if (req->state == FUSE_REQ_SENT) 193 } else if (req->state == FUSE_REQ_SENT) {
272 background_request(fc, req); 194 spin_unlock(&fc->lock);
195 wait_event(req->waitq, req->state == FUSE_REQ_FINISHED);
196 spin_lock(&fc->lock);
197 }
273} 198}
274 199
275static unsigned len_args(unsigned numargs, struct fuse_arg *args) 200static unsigned len_args(unsigned numargs, struct fuse_arg *args)
@@ -327,8 +252,12 @@ void request_send(struct fuse_conn *fc, struct fuse_req *req)
327static void request_send_nowait(struct fuse_conn *fc, struct fuse_req *req) 252static void request_send_nowait(struct fuse_conn *fc, struct fuse_req *req)
328{ 253{
329 spin_lock(&fc->lock); 254 spin_lock(&fc->lock);
330 background_request(fc, req);
331 if (fc->connected) { 255 if (fc->connected) {
256 req->background = 1;
257 fc->num_background++;
258 if (fc->num_background == FUSE_MAX_BACKGROUND)
259 fc->blocked = 1;
260
332 queue_request(fc, req); 261 queue_request(fc, req);
333 spin_unlock(&fc->lock); 262 spin_unlock(&fc->lock);
334 } else { 263 } else {
@@ -883,10 +812,12 @@ void fuse_abort_conn(struct fuse_conn *fc)
883 spin_lock(&fc->lock); 812 spin_lock(&fc->lock);
884 if (fc->connected) { 813 if (fc->connected) {
885 fc->connected = 0; 814 fc->connected = 0;
815 fc->blocked = 0;
886 end_io_requests(fc); 816 end_io_requests(fc);
887 end_requests(fc, &fc->pending); 817 end_requests(fc, &fc->pending);
888 end_requests(fc, &fc->processing); 818 end_requests(fc, &fc->processing);
889 wake_up_all(&fc->waitq); 819 wake_up_all(&fc->waitq);
820 wake_up_all(&fc->blocked_waitq);
890 kill_fasync(&fc->fasync, SIGIO, POLL_IN); 821 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
891 } 822 }
892 spin_unlock(&fc->lock); 823 spin_unlock(&fc->lock);
diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index 8d7546e832e8..72a74cde6de8 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -1,6 +1,6 @@
1/* 1/*
2 FUSE: Filesystem in Userspace 2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2005 Miklos Szeredi <miklos@szeredi.hu> 3 Copyright (C) 2001-2006 Miklos Szeredi <miklos@szeredi.hu>
4 4
5 This program can be distributed under the terms of the GNU GPL. 5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING. 6 See the file COPYING.
@@ -79,7 +79,6 @@ static void fuse_lookup_init(struct fuse_req *req, struct inode *dir,
79{ 79{
80 req->in.h.opcode = FUSE_LOOKUP; 80 req->in.h.opcode = FUSE_LOOKUP;
81 req->in.h.nodeid = get_node_id(dir); 81 req->in.h.nodeid = get_node_id(dir);
82 req->inode = dir;
83 req->in.numargs = 1; 82 req->in.numargs = 1;
84 req->in.args[0].size = entry->d_name.len + 1; 83 req->in.args[0].size = entry->d_name.len + 1;
85 req->in.args[0].value = entry->d_name.name; 84 req->in.args[0].value = entry->d_name.name;
@@ -225,6 +224,20 @@ static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
225} 224}
226 225
227/* 226/*
227 * Synchronous release for the case when something goes wrong in CREATE_OPEN
228 */
229static void fuse_sync_release(struct fuse_conn *fc, struct fuse_file *ff,
230 u64 nodeid, int flags)
231{
232 struct fuse_req *req;
233
234 req = fuse_release_fill(ff, nodeid, flags, FUSE_RELEASE);
235 req->force = 1;
236 request_send(fc, req);
237 fuse_put_request(fc, req);
238}
239
240/*
228 * Atomic create+open operation 241 * Atomic create+open operation
229 * 242 *
230 * If the filesystem doesn't support this, then fall back to separate 243 * If the filesystem doesn't support this, then fall back to separate
@@ -237,6 +250,7 @@ static int fuse_create_open(struct inode *dir, struct dentry *entry, int mode,
237 struct inode *inode; 250 struct inode *inode;
238 struct fuse_conn *fc = get_fuse_conn(dir); 251 struct fuse_conn *fc = get_fuse_conn(dir);
239 struct fuse_req *req; 252 struct fuse_req *req;
253 struct fuse_req *forget_req;
240 struct fuse_open_in inarg; 254 struct fuse_open_in inarg;
241 struct fuse_open_out outopen; 255 struct fuse_open_out outopen;
242 struct fuse_entry_out outentry; 256 struct fuse_entry_out outentry;
@@ -247,9 +261,14 @@ static int fuse_create_open(struct inode *dir, struct dentry *entry, int mode,
247 if (fc->no_create) 261 if (fc->no_create)
248 return -ENOSYS; 262 return -ENOSYS;
249 263
264 forget_req = fuse_get_req(fc);
265 if (IS_ERR(forget_req))
266 return PTR_ERR(forget_req);
267
250 req = fuse_get_req(fc); 268 req = fuse_get_req(fc);
269 err = PTR_ERR(req);
251 if (IS_ERR(req)) 270 if (IS_ERR(req))
252 return PTR_ERR(req); 271 goto out_put_forget_req;
253 272
254 err = -ENOMEM; 273 err = -ENOMEM;
255 ff = fuse_file_alloc(); 274 ff = fuse_file_alloc();
@@ -262,7 +281,6 @@ static int fuse_create_open(struct inode *dir, struct dentry *entry, int mode,
262 inarg.mode = mode; 281 inarg.mode = mode;
263 req->in.h.opcode = FUSE_CREATE; 282 req->in.h.opcode = FUSE_CREATE;
264 req->in.h.nodeid = get_node_id(dir); 283 req->in.h.nodeid = get_node_id(dir);
265 req->inode = dir;
266 req->in.numargs = 2; 284 req->in.numargs = 2;
267 req->in.args[0].size = sizeof(inarg); 285 req->in.args[0].size = sizeof(inarg);
268 req->in.args[0].value = &inarg; 286 req->in.args[0].value = &inarg;
@@ -285,25 +303,23 @@ static int fuse_create_open(struct inode *dir, struct dentry *entry, int mode,
285 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid)) 303 if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
286 goto out_free_ff; 304 goto out_free_ff;
287 305
306 fuse_put_request(fc, req);
288 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation, 307 inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
289 &outentry.attr); 308 &outentry.attr);
290 err = -ENOMEM;
291 if (!inode) { 309 if (!inode) {
292 flags &= ~(O_CREAT | O_EXCL | O_TRUNC); 310 flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
293 ff->fh = outopen.fh; 311 ff->fh = outopen.fh;
294 /* Special release, with inode = NULL, this will 312 fuse_sync_release(fc, ff, outentry.nodeid, flags);
295 trigger a 'forget' request when the release is 313 fuse_send_forget(fc, forget_req, outentry.nodeid, 1);
296 complete */ 314 return -ENOMEM;
297 fuse_send_release(fc, ff, outentry.nodeid, NULL, flags, 0);
298 goto out_put_request;
299 } 315 }
300 fuse_put_request(fc, req); 316 fuse_put_request(fc, forget_req);
301 d_instantiate(entry, inode); 317 d_instantiate(entry, inode);
302 fuse_change_timeout(entry, &outentry); 318 fuse_change_timeout(entry, &outentry);
303 file = lookup_instantiate_filp(nd, entry, generic_file_open); 319 file = lookup_instantiate_filp(nd, entry, generic_file_open);
304 if (IS_ERR(file)) { 320 if (IS_ERR(file)) {
305 ff->fh = outopen.fh; 321 ff->fh = outopen.fh;
306 fuse_send_release(fc, ff, outentry.nodeid, inode, flags, 0); 322 fuse_sync_release(fc, ff, outentry.nodeid, flags);
307 return PTR_ERR(file); 323 return PTR_ERR(file);
308 } 324 }
309 fuse_finish_open(inode, file, ff, &outopen); 325 fuse_finish_open(inode, file, ff, &outopen);
@@ -313,6 +329,8 @@ static int fuse_create_open(struct inode *dir, struct dentry *entry, int mode,
313 fuse_file_free(ff); 329 fuse_file_free(ff);
314 out_put_request: 330 out_put_request:
315 fuse_put_request(fc, req); 331 fuse_put_request(fc, req);
332 out_put_forget_req:
333 fuse_put_request(fc, forget_req);
316 return err; 334 return err;
317} 335}
318 336
@@ -328,7 +346,6 @@ static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
328 int err; 346 int err;
329 347
330 req->in.h.nodeid = get_node_id(dir); 348 req->in.h.nodeid = get_node_id(dir);
331 req->inode = dir;
332 req->out.numargs = 1; 349 req->out.numargs = 1;
333 req->out.args[0].size = sizeof(outarg); 350 req->out.args[0].size = sizeof(outarg);
334 req->out.args[0].value = &outarg; 351 req->out.args[0].value = &outarg;
@@ -448,7 +465,6 @@ static int fuse_unlink(struct inode *dir, struct dentry *entry)
448 465
449 req->in.h.opcode = FUSE_UNLINK; 466 req->in.h.opcode = FUSE_UNLINK;
450 req->in.h.nodeid = get_node_id(dir); 467 req->in.h.nodeid = get_node_id(dir);
451 req->inode = dir;
452 req->in.numargs = 1; 468 req->in.numargs = 1;
453 req->in.args[0].size = entry->d_name.len + 1; 469 req->in.args[0].size = entry->d_name.len + 1;
454 req->in.args[0].value = entry->d_name.name; 470 req->in.args[0].value = entry->d_name.name;
@@ -480,7 +496,6 @@ static int fuse_rmdir(struct inode *dir, struct dentry *entry)
480 496
481 req->in.h.opcode = FUSE_RMDIR; 497 req->in.h.opcode = FUSE_RMDIR;
482 req->in.h.nodeid = get_node_id(dir); 498 req->in.h.nodeid = get_node_id(dir);
483 req->inode = dir;
484 req->in.numargs = 1; 499 req->in.numargs = 1;
485 req->in.args[0].size = entry->d_name.len + 1; 500 req->in.args[0].size = entry->d_name.len + 1;
486 req->in.args[0].value = entry->d_name.name; 501 req->in.args[0].value = entry->d_name.name;
@@ -510,8 +525,6 @@ static int fuse_rename(struct inode *olddir, struct dentry *oldent,
510 inarg.newdir = get_node_id(newdir); 525 inarg.newdir = get_node_id(newdir);
511 req->in.h.opcode = FUSE_RENAME; 526 req->in.h.opcode = FUSE_RENAME;
512 req->in.h.nodeid = get_node_id(olddir); 527 req->in.h.nodeid = get_node_id(olddir);
513 req->inode = olddir;
514 req->inode2 = newdir;
515 req->in.numargs = 3; 528 req->in.numargs = 3;
516 req->in.args[0].size = sizeof(inarg); 529 req->in.args[0].size = sizeof(inarg);
517 req->in.args[0].value = &inarg; 530 req->in.args[0].value = &inarg;
@@ -558,7 +571,6 @@ static int fuse_link(struct dentry *entry, struct inode *newdir,
558 memset(&inarg, 0, sizeof(inarg)); 571 memset(&inarg, 0, sizeof(inarg));
559 inarg.oldnodeid = get_node_id(inode); 572 inarg.oldnodeid = get_node_id(inode);
560 req->in.h.opcode = FUSE_LINK; 573 req->in.h.opcode = FUSE_LINK;
561 req->inode2 = inode;
562 req->in.numargs = 2; 574 req->in.numargs = 2;
563 req->in.args[0].size = sizeof(inarg); 575 req->in.args[0].size = sizeof(inarg);
564 req->in.args[0].value = &inarg; 576 req->in.args[0].value = &inarg;
@@ -587,7 +599,6 @@ int fuse_do_getattr(struct inode *inode)
587 599
588 req->in.h.opcode = FUSE_GETATTR; 600 req->in.h.opcode = FUSE_GETATTR;
589 req->in.h.nodeid = get_node_id(inode); 601 req->in.h.nodeid = get_node_id(inode);
590 req->inode = inode;
591 req->out.numargs = 1; 602 req->out.numargs = 1;
592 req->out.args[0].size = sizeof(arg); 603 req->out.args[0].size = sizeof(arg);
593 req->out.args[0].value = &arg; 604 req->out.args[0].value = &arg;
@@ -679,7 +690,6 @@ static int fuse_access(struct inode *inode, int mask)
679 inarg.mask = mask; 690 inarg.mask = mask;
680 req->in.h.opcode = FUSE_ACCESS; 691 req->in.h.opcode = FUSE_ACCESS;
681 req->in.h.nodeid = get_node_id(inode); 692 req->in.h.nodeid = get_node_id(inode);
682 req->inode = inode;
683 req->in.numargs = 1; 693 req->in.numargs = 1;
684 req->in.args[0].size = sizeof(inarg); 694 req->in.args[0].size = sizeof(inarg);
685 req->in.args[0].value = &inarg; 695 req->in.args[0].value = &inarg;
@@ -820,7 +830,6 @@ static char *read_link(struct dentry *dentry)
820 } 830 }
821 req->in.h.opcode = FUSE_READLINK; 831 req->in.h.opcode = FUSE_READLINK;
822 req->in.h.nodeid = get_node_id(inode); 832 req->in.h.nodeid = get_node_id(inode);
823 req->inode = inode;
824 req->out.argvar = 1; 833 req->out.argvar = 1;
825 req->out.numargs = 1; 834 req->out.numargs = 1;
826 req->out.args[0].size = PAGE_SIZE - 1; 835 req->out.args[0].size = PAGE_SIZE - 1;
@@ -939,7 +948,6 @@ static int fuse_setattr(struct dentry *entry, struct iattr *attr)
939 iattr_to_fattr(attr, &inarg); 948 iattr_to_fattr(attr, &inarg);
940 req->in.h.opcode = FUSE_SETATTR; 949 req->in.h.opcode = FUSE_SETATTR;
941 req->in.h.nodeid = get_node_id(inode); 950 req->in.h.nodeid = get_node_id(inode);
942 req->inode = inode;
943 req->in.numargs = 1; 951 req->in.numargs = 1;
944 req->in.args[0].size = sizeof(inarg); 952 req->in.args[0].size = sizeof(inarg);
945 req->in.args[0].value = &inarg; 953 req->in.args[0].value = &inarg;
@@ -1002,7 +1010,6 @@ static int fuse_setxattr(struct dentry *entry, const char *name,
1002 inarg.flags = flags; 1010 inarg.flags = flags;
1003 req->in.h.opcode = FUSE_SETXATTR; 1011 req->in.h.opcode = FUSE_SETXATTR;
1004 req->in.h.nodeid = get_node_id(inode); 1012 req->in.h.nodeid = get_node_id(inode);
1005 req->inode = inode;
1006 req->in.numargs = 3; 1013 req->in.numargs = 3;
1007 req->in.args[0].size = sizeof(inarg); 1014 req->in.args[0].size = sizeof(inarg);
1008 req->in.args[0].value = &inarg; 1015 req->in.args[0].value = &inarg;
@@ -1041,7 +1048,6 @@ static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
1041 inarg.size = size; 1048 inarg.size = size;
1042 req->in.h.opcode = FUSE_GETXATTR; 1049 req->in.h.opcode = FUSE_GETXATTR;
1043 req->in.h.nodeid = get_node_id(inode); 1050 req->in.h.nodeid = get_node_id(inode);
1044 req->inode = inode;
1045 req->in.numargs = 2; 1051 req->in.numargs = 2;
1046 req->in.args[0].size = sizeof(inarg); 1052 req->in.args[0].size = sizeof(inarg);
1047 req->in.args[0].value = &inarg; 1053 req->in.args[0].value = &inarg;
@@ -1091,7 +1097,6 @@ static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
1091 inarg.size = size; 1097 inarg.size = size;
1092 req->in.h.opcode = FUSE_LISTXATTR; 1098 req->in.h.opcode = FUSE_LISTXATTR;
1093 req->in.h.nodeid = get_node_id(inode); 1099 req->in.h.nodeid = get_node_id(inode);
1094 req->inode = inode;
1095 req->in.numargs = 1; 1100 req->in.numargs = 1;
1096 req->in.args[0].size = sizeof(inarg); 1101 req->in.args[0].size = sizeof(inarg);
1097 req->in.args[0].value = &inarg; 1102 req->in.args[0].value = &inarg;
@@ -1135,7 +1140,6 @@ static int fuse_removexattr(struct dentry *entry, const char *name)
1135 1140
1136 req->in.h.opcode = FUSE_REMOVEXATTR; 1141 req->in.h.opcode = FUSE_REMOVEXATTR;
1137 req->in.h.nodeid = get_node_id(inode); 1142 req->in.h.nodeid = get_node_id(inode);
1138 req->inode = inode;
1139 req->in.numargs = 1; 1143 req->in.numargs = 1;
1140 req->in.args[0].size = strlen(name) + 1; 1144 req->in.args[0].size = strlen(name) + 1;
1141 req->in.args[0].value = name; 1145 req->in.args[0].value = name;
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index 087f3b734f40..1d59af306b28 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -30,7 +30,6 @@ static int fuse_send_open(struct inode *inode, struct file *file, int isdir,
30 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC); 30 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
31 req->in.h.opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN; 31 req->in.h.opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
32 req->in.h.nodeid = get_node_id(inode); 32 req->in.h.nodeid = get_node_id(inode);
33 req->inode = inode;
34 req->in.numargs = 1; 33 req->in.numargs = 1;
35 req->in.args[0].size = sizeof(inarg); 34 req->in.args[0].size = sizeof(inarg);
36 req->in.args[0].value = &inarg; 35 req->in.args[0].value = &inarg;
@@ -113,37 +112,22 @@ int fuse_open_common(struct inode *inode, struct file *file, int isdir)
113 return err; 112 return err;
114} 113}
115 114
116/* Special case for failed iget in CREATE */ 115struct fuse_req *fuse_release_fill(struct fuse_file *ff, u64 nodeid, int flags,
117static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req) 116 int opcode)
118{ 117{
119 /* If called from end_io_requests(), req has more than one 118 struct fuse_req *req = ff->release_req;
120 reference and fuse_reset_request() cannot work */
121 if (fc->connected) {
122 u64 nodeid = req->in.h.nodeid;
123 fuse_reset_request(req);
124 fuse_send_forget(fc, req, nodeid, 1);
125 } else
126 fuse_put_request(fc, req);
127}
128
129void fuse_send_release(struct fuse_conn *fc, struct fuse_file *ff,
130 u64 nodeid, struct inode *inode, int flags, int isdir)
131{
132 struct fuse_req * req = ff->release_req;
133 struct fuse_release_in *inarg = &req->misc.release_in; 119 struct fuse_release_in *inarg = &req->misc.release_in;
134 120
135 inarg->fh = ff->fh; 121 inarg->fh = ff->fh;
136 inarg->flags = flags; 122 inarg->flags = flags;
137 req->in.h.opcode = isdir ? FUSE_RELEASEDIR : FUSE_RELEASE; 123 req->in.h.opcode = opcode;
138 req->in.h.nodeid = nodeid; 124 req->in.h.nodeid = nodeid;
139 req->inode = inode;
140 req->in.numargs = 1; 125 req->in.numargs = 1;
141 req->in.args[0].size = sizeof(struct fuse_release_in); 126 req->in.args[0].size = sizeof(struct fuse_release_in);
142 req->in.args[0].value = inarg; 127 req->in.args[0].value = inarg;
143 request_send_background(fc, req);
144 if (!inode)
145 req->end = fuse_release_end;
146 kfree(ff); 128 kfree(ff);
129
130 return req;
147} 131}
148 132
149int fuse_release_common(struct inode *inode, struct file *file, int isdir) 133int fuse_release_common(struct inode *inode, struct file *file, int isdir)
@@ -151,8 +135,15 @@ int fuse_release_common(struct inode *inode, struct file *file, int isdir)
151 struct fuse_file *ff = file->private_data; 135 struct fuse_file *ff = file->private_data;
152 if (ff) { 136 if (ff) {
153 struct fuse_conn *fc = get_fuse_conn(inode); 137 struct fuse_conn *fc = get_fuse_conn(inode);
154 u64 nodeid = get_node_id(inode); 138 struct fuse_req *req;
155 fuse_send_release(fc, ff, nodeid, inode, file->f_flags, isdir); 139
140 req = fuse_release_fill(ff, get_node_id(inode), file->f_flags,
141 isdir ? FUSE_RELEASEDIR : FUSE_RELEASE);
142
143 /* Hold vfsmount and dentry until release is finished */
144 req->vfsmount = mntget(file->f_vfsmnt);
145 req->dentry = dget(file->f_dentry);
146 request_send_background(fc, req);
156 } 147 }
157 148
158 /* Return value is ignored by VFS */ 149 /* Return value is ignored by VFS */
@@ -192,8 +183,6 @@ static int fuse_flush(struct file *file, fl_owner_t id)
192 inarg.fh = ff->fh; 183 inarg.fh = ff->fh;
193 req->in.h.opcode = FUSE_FLUSH; 184 req->in.h.opcode = FUSE_FLUSH;
194 req->in.h.nodeid = get_node_id(inode); 185 req->in.h.nodeid = get_node_id(inode);
195 req->inode = inode;
196 req->file = file;
197 req->in.numargs = 1; 186 req->in.numargs = 1;
198 req->in.args[0].size = sizeof(inarg); 187 req->in.args[0].size = sizeof(inarg);
199 req->in.args[0].value = &inarg; 188 req->in.args[0].value = &inarg;
@@ -232,8 +221,6 @@ int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
232 inarg.fsync_flags = datasync ? 1 : 0; 221 inarg.fsync_flags = datasync ? 1 : 0;
233 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC; 222 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
234 req->in.h.nodeid = get_node_id(inode); 223 req->in.h.nodeid = get_node_id(inode);
235 req->inode = inode;
236 req->file = file;
237 req->in.numargs = 1; 224 req->in.numargs = 1;
238 req->in.args[0].size = sizeof(inarg); 225 req->in.args[0].size = sizeof(inarg);
239 req->in.args[0].value = &inarg; 226 req->in.args[0].value = &inarg;
@@ -266,8 +253,6 @@ void fuse_read_fill(struct fuse_req *req, struct file *file,
266 inarg->size = count; 253 inarg->size = count;
267 req->in.h.opcode = opcode; 254 req->in.h.opcode = opcode;
268 req->in.h.nodeid = get_node_id(inode); 255 req->in.h.nodeid = get_node_id(inode);
269 req->inode = inode;
270 req->file = file;
271 req->in.numargs = 1; 256 req->in.numargs = 1;
272 req->in.args[0].size = sizeof(struct fuse_read_in); 257 req->in.args[0].size = sizeof(struct fuse_read_in);
273 req->in.args[0].value = inarg; 258 req->in.args[0].value = inarg;
@@ -342,6 +327,8 @@ static void fuse_send_readpages(struct fuse_req *req, struct file *file,
342 req->out.page_zeroing = 1; 327 req->out.page_zeroing = 1;
343 fuse_read_fill(req, file, inode, pos, count, FUSE_READ); 328 fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
344 if (fc->async_read) { 329 if (fc->async_read) {
330 get_file(file);
331 req->file = file;
345 req->end = fuse_readpages_end; 332 req->end = fuse_readpages_end;
346 request_send_background(fc, req); 333 request_send_background(fc, req);
347 } else { 334 } else {
@@ -420,8 +407,6 @@ static size_t fuse_send_write(struct fuse_req *req, struct file *file,
420 inarg.size = count; 407 inarg.size = count;
421 req->in.h.opcode = FUSE_WRITE; 408 req->in.h.opcode = FUSE_WRITE;
422 req->in.h.nodeid = get_node_id(inode); 409 req->in.h.nodeid = get_node_id(inode);
423 req->inode = inode;
424 req->file = file;
425 req->in.argpages = 1; 410 req->in.argpages = 1;
426 req->in.numargs = 2; 411 req->in.numargs = 2;
427 req->in.args[0].size = sizeof(struct fuse_write_in); 412 req->in.args[0].size = sizeof(struct fuse_write_in);
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index 0474202cb5dc..25f8581a770c 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -8,12 +8,12 @@
8 8
9#include <linux/fuse.h> 9#include <linux/fuse.h>
10#include <linux/fs.h> 10#include <linux/fs.h>
11#include <linux/mount.h>
11#include <linux/wait.h> 12#include <linux/wait.h>
12#include <linux/list.h> 13#include <linux/list.h>
13#include <linux/spinlock.h> 14#include <linux/spinlock.h>
14#include <linux/mm.h> 15#include <linux/mm.h>
15#include <linux/backing-dev.h> 16#include <linux/backing-dev.h>
16#include <asm/semaphore.h>
17 17
18/** Max number of pages that can be used in a single read request */ 18/** Max number of pages that can be used in a single read request */
19#define FUSE_MAX_PAGES_PER_REQ 32 19#define FUSE_MAX_PAGES_PER_REQ 32
@@ -135,9 +135,6 @@ struct fuse_req {
135 fuse_conn */ 135 fuse_conn */
136 struct list_head list; 136 struct list_head list;
137 137
138 /** Entry on the background list */
139 struct list_head bg_entry;
140
141 /** refcount */ 138 /** refcount */
142 atomic_t count; 139 atomic_t count;
143 140
@@ -150,6 +147,9 @@ struct fuse_req {
150 /** True if the request has reply */ 147 /** True if the request has reply */
151 unsigned isreply:1; 148 unsigned isreply:1;
152 149
150 /** Force sending of the request even if interrupted */
151 unsigned force:1;
152
153 /** The request was interrupted */ 153 /** The request was interrupted */
154 unsigned interrupted:1; 154 unsigned interrupted:1;
155 155
@@ -192,15 +192,15 @@ struct fuse_req {
192 /** offset of data on first page */ 192 /** offset of data on first page */
193 unsigned page_offset; 193 unsigned page_offset;
194 194
195 /** Inode used in the request */
196 struct inode *inode;
197
198 /** Second inode used in the request (or NULL) */
199 struct inode *inode2;
200
201 /** File used in the request (or NULL) */ 195 /** File used in the request (or NULL) */
202 struct file *file; 196 struct file *file;
203 197
198 /** vfsmount used in release */
199 struct vfsmount *vfsmount;
200
201 /** dentry used in release */
202 struct dentry *dentry;
203
204 /** Request completion callback */ 204 /** Request completion callback */
205 void (*end)(struct fuse_conn *, struct fuse_req *); 205 void (*end)(struct fuse_conn *, struct fuse_req *);
206}; 206};
@@ -243,10 +243,6 @@ struct fuse_conn {
243 /** The list of requests under I/O */ 243 /** The list of requests under I/O */
244 struct list_head io; 244 struct list_head io;
245 245
246 /** Requests put in the background (RELEASE or any other
247 interrupted request) */
248 struct list_head background;
249
250 /** Number of requests currently in the background */ 246 /** Number of requests currently in the background */
251 unsigned num_background; 247 unsigned num_background;
252 248
@@ -258,15 +254,9 @@ struct fuse_conn {
258 /** waitq for blocked connection */ 254 /** waitq for blocked connection */
259 wait_queue_head_t blocked_waitq; 255 wait_queue_head_t blocked_waitq;
260 256
261 /** RW semaphore for exclusion with fuse_put_super() */
262 struct rw_semaphore sbput_sem;
263
264 /** The next unique request id */ 257 /** The next unique request id */
265 u64 reqctr; 258 u64 reqctr;
266 259
267 /** Mount is active */
268 unsigned mounted;
269
270 /** Connection established, cleared on umount, connection 260 /** Connection established, cleared on umount, connection
271 abort and device release */ 261 abort and device release */
272 unsigned connected; 262 unsigned connected;
@@ -383,12 +373,9 @@ void fuse_file_free(struct fuse_file *ff);
383void fuse_finish_open(struct inode *inode, struct file *file, 373void fuse_finish_open(struct inode *inode, struct file *file,
384 struct fuse_file *ff, struct fuse_open_out *outarg); 374 struct fuse_file *ff, struct fuse_open_out *outarg);
385 375
386/** 376/** */
387 * Send a RELEASE request 377struct fuse_req *fuse_release_fill(struct fuse_file *ff, u64 nodeid, int flags,
388 */ 378 int opcode);
389void fuse_send_release(struct fuse_conn *fc, struct fuse_file *ff,
390 u64 nodeid, struct inode *inode, int flags, int isdir);
391
392/** 379/**
393 * Send RELEASE or RELEASEDIR request 380 * Send RELEASE or RELEASEDIR request
394 */ 381 */
@@ -446,11 +433,6 @@ struct fuse_req *fuse_request_alloc(void);
446void fuse_request_free(struct fuse_req *req); 433void fuse_request_free(struct fuse_req *req);
447 434
448/** 435/**
449 * Reinitialize a request, the preallocated flag is left unmodified
450 */
451void fuse_reset_request(struct fuse_req *req);
452
453/**
454 * Reserve a preallocated request 436 * Reserve a preallocated request
455 */ 437 */
456struct fuse_req *fuse_get_req(struct fuse_conn *fc); 438struct fuse_req *fuse_get_req(struct fuse_conn *fc);
@@ -476,11 +458,6 @@ void request_send_noreply(struct fuse_conn *fc, struct fuse_req *req);
476 */ 458 */
477void request_send_background(struct fuse_conn *fc, struct fuse_req *req); 459void request_send_background(struct fuse_conn *fc, struct fuse_req *req);
478 460
479/**
480 * Release inodes and file associated with background request
481 */
482void fuse_release_background(struct fuse_conn *fc, struct fuse_req *req);
483
484/* Abort all requests */ 461/* Abort all requests */
485void fuse_abort_conn(struct fuse_conn *fc); 462void fuse_abort_conn(struct fuse_conn *fc);
486 463
diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
index a13c0f529058..0225729977c4 100644
--- a/fs/fuse/inode.c
+++ b/fs/fuse/inode.c
@@ -11,7 +11,6 @@
11#include <linux/pagemap.h> 11#include <linux/pagemap.h>
12#include <linux/slab.h> 12#include <linux/slab.h>
13#include <linux/file.h> 13#include <linux/file.h>
14#include <linux/mount.h>
15#include <linux/seq_file.h> 14#include <linux/seq_file.h>
16#include <linux/init.h> 15#include <linux/init.h>
17#include <linux/module.h> 16#include <linux/module.h>
@@ -204,20 +203,14 @@ static void fuse_put_super(struct super_block *sb)
204{ 203{
205 struct fuse_conn *fc = get_fuse_conn_super(sb); 204 struct fuse_conn *fc = get_fuse_conn_super(sb);
206 205
207 down_write(&fc->sbput_sem);
208 while (!list_empty(&fc->background))
209 fuse_release_background(fc,
210 list_entry(fc->background.next,
211 struct fuse_req, bg_entry));
212
213 spin_lock(&fc->lock); 206 spin_lock(&fc->lock);
214 fc->mounted = 0;
215 fc->connected = 0; 207 fc->connected = 0;
208 fc->blocked = 0;
216 spin_unlock(&fc->lock); 209 spin_unlock(&fc->lock);
217 up_write(&fc->sbput_sem);
218 /* Flush all readers on this fs */ 210 /* Flush all readers on this fs */
219 kill_fasync(&fc->fasync, SIGIO, POLL_IN); 211 kill_fasync(&fc->fasync, SIGIO, POLL_IN);
220 wake_up_all(&fc->waitq); 212 wake_up_all(&fc->waitq);
213 wake_up_all(&fc->blocked_waitq);
221 kobject_del(&fc->kobj); 214 kobject_del(&fc->kobj);
222 kobject_put(&fc->kobj); 215 kobject_put(&fc->kobj);
223} 216}
@@ -386,8 +379,6 @@ static struct fuse_conn *new_conn(void)
386 INIT_LIST_HEAD(&fc->pending); 379 INIT_LIST_HEAD(&fc->pending);
387 INIT_LIST_HEAD(&fc->processing); 380 INIT_LIST_HEAD(&fc->processing);
388 INIT_LIST_HEAD(&fc->io); 381 INIT_LIST_HEAD(&fc->io);
389 INIT_LIST_HEAD(&fc->background);
390 init_rwsem(&fc->sbput_sem);
391 kobj_set_kset_s(fc, connections_subsys); 382 kobj_set_kset_s(fc, connections_subsys);
392 kobject_init(&fc->kobj); 383 kobject_init(&fc->kobj);
393 atomic_set(&fc->num_waiting, 0); 384 atomic_set(&fc->num_waiting, 0);
@@ -543,7 +534,6 @@ static int fuse_fill_super(struct super_block *sb, void *data, int silent)
543 goto err_kobject_del; 534 goto err_kobject_del;
544 535
545 sb->s_root = root_dentry; 536 sb->s_root = root_dentry;
546 fc->mounted = 1;
547 fc->connected = 1; 537 fc->connected = 1;
548 kobject_get(&fc->kobj); 538 kobject_get(&fc->kobj);
549 file->private_data = fc; 539 file->private_data = fc;