projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'v3.10.108' into update
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / fuse / file.c
CommitLineData
b6aeaded
MS		 1/*
2 FUSE: Filesystem in Userspace
1729a16c 3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
b6aeaded
MS		 4
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"
6fa3eb70 10#include "fuse.h"
b6aeaded
MS		 11
12#include <linux/pagemap.h>
13#include <linux/slab.h>
14#include <linux/kernel.h>
e8edc6e0 15#include <linux/sched.h>
08cbf542 16#include <linux/module.h>
d9d318d3 17#include <linux/compat.h>
478e0841 18#include <linux/swap.h>
a27bb332 19#include <linux/aio.h>
3634a632 20#include <linux/falloc.h>
6fa3eb70
S		 21#include <asm/div64.h>
22
23#ifdef FUSEIO_TRACE
24struct mutex fuse_iolog_lock;
25
26static struct fuse_proc_info fuse_iolog[FUSE_IOLOG_MAX];
27static struct timespec fuse_iolog_time;
28static struct task_struct *fuse_iolog_thread=NULL;
29
30void fuse_time_diff(
31 struct timespec *start,
32 struct timespec *end,
33 struct timespec *diff)
34{
35 if ((end->tv_nsec-start->tv_nsec)<0) {
36 diff->tv_sec = end->tv_sec-start->tv_sec-1;
37 diff->tv_nsec = 1000000000+end->tv_nsec-start->tv_nsec;
38 } else {
39 diff->tv_sec = end->tv_sec-start->tv_sec;
40 diff->tv_nsec = end->tv_nsec-start->tv_nsec;
41 }
42 return;
43}
44
45struct fuse_ioiog_type_map
46{
47 int type;
48 const char *str;
49};
50
51#define FUSE_IOLOG_TYPE_MAX 46
52
53static const char *fuse_iolog_type[FUSE_IOLOG_TYPE_MAX]={
54 "unknown", /*0*/
55 "lookup", /*1*/
56 "forget", /*2*/
57 "getattr", /*3*/
58 "setattr", /*4*/
59 "readlink", /*5*/
60 "symlink", /*6*/
61 "", /*7*/
62 "mknod", /*8*/
63 "mkdir", /*9*/
64 "unlink", /*10*/
65 "rmdir", /*11*/
66 "rename", /*12*/
67 "link", /*13*/
68 "open", /*14*/
69 "read", /*15*/
70 "write", /*16*/
71 "statfs", /*17*/
72 "release", /*18*/
73 "", /*19*/
74 "fsync", /*20*/
75 "setxattr", /*21*/
76 "getxattr", /*22*/
77 "listxattr", /*23*/
78 "removexattr", /*24*/
79 "flush", /*25*/
80 "init", /*26*/
81 "opendir", /*27*/
82 "readdir", /*28*/
83 "releasedir", /*29*/
84 "fsyncdir", /*30*/
85 "getlk", /*31*/
86 "setlk", /*32*/
87 "setlkw", /*33*/
88 "access", /*34*/
89 "create", /*35*/
90 "interrupt", /*36*/
91 "bmap", /*37*/
92 "destroy", /*38*/
93 "ioctl", /*39*/
94 "poll", /*40*/
95 "notify_reply", /*41*/
96 "batch_forget", /*42*/
97 "fallocate", /*43*/
98 "readdirplus", /*44*/
99 "cuse"
100};
101
102static const char *fuse_iolog_type2str(int type)
103{
104 if (type>=0 && type<FUSE_IOLOG_TYPE_MAX)
105 goto out;
106 else if (type==CUSE_INIT)
107 type=FUSE_IOLOG_TYPE_MAX-1;
108 else
109 type=0;
110out:
111 return fuse_iolog_type[type];
112}
113
114int fuse_iolog_print(void)
115{
116 int i, len, n;
117 char buf[FUSE_IOLOG_BUFLEN], *ptr;
118
119 len=FUSE_IOLOG_BUFLEN-1;
120 ptr=&buf[0];
121
122 for (i=0;i<FUSE_IOLOG_MAX && fuse_iolog[i].valid;i++) {
123
124 if (fuse_iolog[i].read.count || fuse_iolog[i].write.count) {
125 n=snprintf(ptr, len, "{%d:R(%d,%d,%d),W(%d,%d,%d)}",
126 fuse_iolog[i].pid,
127 fuse_iolog[i].read.bytes,
128 fuse_iolog[i].read.count,
129 fuse_iolog[i].read.us,
130 fuse_iolog[i].write.bytes,
131 fuse_iolog[i].write.count,
132 fuse_iolog[i].write.us);
133
134 len -=n;
135 ptr +=n;
136
137 if (len<0)
138 goto overflow;
139 }
140
141 if (fuse_iolog[i].misc_type) {
142 n=snprintf(ptr, len, "{%d:%s(%d,%d,%d)}",
143 fuse_iolog[i].pid,
144 fuse_iolog_type2str(fuse_iolog[i].misc_type),
145 fuse_iolog[i].misc.bytes,
146 fuse_iolog[i].misc.count,
147 fuse_iolog[i].misc.us);
148 len -=n;
149 ptr +=n;
150
151 if (len<0)
152 goto overflow;
153 }
154 }
155
156 if (i>0)
157 xlog_printk(ANDROID_LOG_DEBUG, "BLOCK_TAG", "FUSEIO %s\n", buf);
158
159 return ptr - &buf[0];
160
161overflow:
162 xlog_printk(ANDROID_LOG_DEBUG, "BLOCK_TAG",
163 "FUSEIO log buffer overflow \n");
164
165 return -1;
166}
167
168void fuse_iolog_proc_clear(void)
169{
170 memset(&fuse_iolog[0], 0, sizeof(struct fuse_proc_info)*FUSE_IOLOG_MAX);
171 get_monotonic_boottime(&fuse_iolog_time);
172}
173
174inline __u32 fuse_iolog_timeus(struct timespec *t)
175{
176 __u32 _t;
177 long us;
178
179 us = t->tv_nsec;
180 do_div(us, 1000);
181
182 if (t->tv_sec > 3600)
183 return 0xD693A400; /* 3600000000 */
184 else
185 _t = t->tv_sec * 1000000 + us;
186
187 if (_t)
188 return _t;
189 else
190 return 1;
191}
192
193__u32 fuse_iolog_timeus_diff(struct timespec *start, struct timespec *end)
194{
195 struct timespec diff;
196 fuse_time_diff(start, end, &diff);
197 return fuse_iolog_timeus(&diff);
198}
199
200
201inline int fuse_iolog_proc_update(struct fuse_proc_info *info,
202 __u32 io_bytes, int type, struct timespec *diff)
203{
204 struct fuse_rw_info *rwi;
205 __u32 _t;
206
207 _t = fuse_iolog_timeus(diff);
208
209 if (type==FUSE_READ)
210 rwi = &info->read;
211 else if (type == FUSE_WRITE)
212 rwi = &info->write;
213 else {
214 if (info->misc_type==0)
215 info->misc_type=type;
216 else if (info->misc_type!=type) /* misc type mismatch => continue */
217 return -1;
218 rwi = &info->misc;
219 }
220
221 rwi->bytes += io_bytes;
222 rwi->us += _t;
223 rwi->count ++;
224
225 return 0;
226}
227
228static int fuse_iolog_watch(void *arg)
229{
230 unsigned int timeout;
231 int n;
232 struct timespec curr, diff;
233
234 while (1) {
235 if (kthread_should_stop()) break;
236
237 get_monotonic_boottime(&curr);
238
239 mutex_lock(&fuse_iolog_lock);
240 fuse_time_diff(&fuse_iolog_time, &curr, &diff);
241
242 n=fuse_iolog_print();
243
244 if (n>0)
245 fuse_iolog_proc_clear();
246
247 mutex_unlock(&fuse_iolog_lock);
248
249 do {
250 set_current_state(TASK_INTERRUPTIBLE);
251 timeout = schedule_timeout(FUSE_IOLOG_LATENCY*HZ);
252 } while(timeout);
253 }
254
255 return 0;
256}
257
258void fuse_iolog_init(void)
259{
260 int ret;
261
262 mutex_init(&fuse_iolog_lock);
263 mutex_lock(&fuse_iolog_lock);
264 fuse_iolog_proc_clear();
265 mutex_unlock(&fuse_iolog_lock);
266
267 fuse_iolog_thread=kthread_create(fuse_iolog_watch, NULL, "fuse_log");
268 if (IS_ERR(fuse_iolog_thread)) {
269 ret = PTR_ERR(fuse_iolog_thread);
270 xlog_printk(ANDROID_LOG_DEBUG, "BLOCK_TAG",
271 "Fail to create fuse_log thread %d\n", ret);
272 fuse_iolog_thread = NULL;
273 goto out;
274 }
275 wake_up_process(fuse_iolog_thread);
276out:
277 return;
278}
279
280void fuse_iolog_exit(void)
281{
282 kthread_stop(fuse_iolog_thread);
283}
284void fuse_iolog_add(__u32 io_bytes, int type,
285 struct timespec *start,
286 struct timespec *end)
287{
288 struct fuse_proc_info *info;
289 struct timespec diff;
290 pid_t pid;
291 int i;
292 pid = task_pid_nr(current);
293 fuse_time_diff(start, end, &diff);
294
295 mutex_lock(&fuse_iolog_lock);
296
297 for (i=0;i<FUSE_IOLOG_MAX;i++) {
298 info=&fuse_iolog[i];
299
300 if (info->valid) {
301 if (info->pid == pid) {
302 if (fuse_iolog_proc_update(info, io_bytes, type, &diff)) {
303 continue; // ops mismatch
304 }
305 else
306 goto out;
307 }
308 else {
309 continue;
310 }
311 }
312 else {
313 info->valid=1;
314 info->pid=pid;
315 fuse_iolog_proc_update(info, io_bytes, type, &diff);
316 if (i==0)
317 get_monotonic_boottime(&fuse_iolog_time);
318 goto out;
319 }
320 }
321
322 if (i==FUSE_IOLOG_MAX) {
323 fuse_iolog_print();
324 fuse_iolog_proc_clear();
325 info=&fuse_iolog[0];
326 info->valid=1;
327 info->pid=pid;
328 fuse_iolog_proc_update(info, io_bytes, type, &diff);
329 }
330out:
331 mutex_unlock(&fuse_iolog_lock);
332}
333
334#endif
b6aeaded 335
4b6f5d20 336static const struct file_operations fuse_direct_io_file_operations;
45323fb7 337
91fe96b4
MS		 338static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
339 int opcode, struct fuse_open_out *outargp)
b6aeaded 340{
b6aeaded 341 struct fuse_open_in inarg;
fd72faac
MS		 342 struct fuse_req *req;
343 int err;
344
b111c8c0 345 req = fuse_get_req_nopages(fc);
ce1d5a49
MS		 346 if (IS_ERR(req))
347 return PTR_ERR(req);
fd72faac
MS		 348
349 memset(&inarg, 0, sizeof(inarg));
6ff958ed
MS		 350 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY);
351 if (!fc->atomic_o_trunc)
352 inarg.flags &= ~O_TRUNC;
91fe96b4
MS		 353 req->in.h.opcode = opcode;
354 req->in.h.nodeid = nodeid;
fd72faac
MS		 355 req->in.numargs = 1;
356 req->in.args[0].size = sizeof(inarg);
357 req->in.args[0].value = &inarg;
358 req->out.numargs = 1;
359 req->out.args[0].size = sizeof(*outargp);
360 req->out.args[0].value = outargp;
b93f858a 361 fuse_request_send(fc, req);
fd72faac
MS		 362 err = req->out.h.error;
363 fuse_put_request(fc, req);
364
365 return err;
366}
367
acf99433 368struct fuse_file *fuse_file_alloc(struct fuse_conn *fc)
fd72faac
MS		 369{
370 struct fuse_file *ff;
6b2db28a 371
c2ace304 372 ff = kzalloc(sizeof(struct fuse_file), GFP_KERNEL);
6b2db28a
TH		 373 if (unlikely(!ff))
374 return NULL;
375
da5e4714 376 ff->fc = fc;
4250c066 377 ff->reserved_req = fuse_request_alloc(0);
6b2db28a
TH		 378 if (unlikely(!ff->reserved_req)) {
379 kfree(ff);
380 return NULL;
fd72faac 381 }
6b2db28a
TH		 382
383 INIT_LIST_HEAD(&ff->write_entry);
384 atomic_set(&ff->count, 0);
385 RB_CLEAR_NODE(&ff->polled_node);
386 init_waitqueue_head(&ff->poll_wait);
387
388 spin_lock(&fc->lock);
389 ff->kh = ++fc->khctr;
390 spin_unlock(&fc->lock);
391
fd72faac
MS		 392 return ff;
393}
394
395void fuse_file_free(struct fuse_file *ff)
396{
33649c91 397 fuse_request_free(ff->reserved_req);
fd72faac
MS		 398 kfree(ff);
399}
400
c7b7143c 401struct fuse_file *fuse_file_get(struct fuse_file *ff)
c756e0a4
MS		 402{
403 atomic_inc(&ff->count);
404 return ff;
405}
406
5a18ec17
MS		 407static void fuse_release_async(struct work_struct *work)
408{
409 struct fuse_req *req;
410 struct fuse_conn *fc;
411 struct path path;
412
413 req = container_of(work, struct fuse_req, misc.release.work);
414 path = req->misc.release.path;
415 fc = get_fuse_conn(path.dentry->d_inode);
416
417 fuse_put_request(fc, req);
418 path_put(&path);
419}
420
819c4b3b
MS		 421static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
422{
5a18ec17
MS		 423 if (fc->destroy_req) {
424 /*
425 * If this is a fuseblk mount, then it's possible that
426 * releasing the path will result in releasing the
427 * super block and sending the DESTROY request. If
428 * the server is single threaded, this would hang.
429 * For this reason do the path_put() in a separate
430 * thread.
431 */
432 atomic_inc(&req->count);
433 INIT_WORK(&req->misc.release.work, fuse_release_async);
434 schedule_work(&req->misc.release.work);
435 } else {
436 path_put(&req->misc.release.path);
437 }
819c4b3b
MS		 438}
439
5a18ec17 440static void fuse_file_put(struct fuse_file *ff, bool sync)
c756e0a4
MS		 441{
442 if (atomic_dec_and_test(&ff->count)) {
443 struct fuse_req *req = ff->reserved_req;
8b0797a4 444
5a18ec17 445 if (sync) {
5beea857 446 req->force = 1;
8b41e671 447 req->background = 0;
5a18ec17
MS		 448 fuse_request_send(ff->fc, req);
449 path_put(&req->misc.release.path);
450 fuse_put_request(ff->fc, req);
451 } else {
452 req->end = fuse_release_end;
8b41e671 453 req->background = 1;
5a18ec17
MS		 454 fuse_request_send_background(ff->fc, req);
455 }
c756e0a4
MS		 456 kfree(ff);
457 }
458}
459
08cbf542
TH		 460int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
461 bool isdir)
91fe96b4
MS		 462{
463 struct fuse_open_out outarg;
464 struct fuse_file *ff;
465 int err;
466 int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
467
468 ff = fuse_file_alloc(fc);
469 if (!ff)
470 return -ENOMEM;
471
472 err = fuse_send_open(fc, nodeid, file, opcode, &outarg);
473 if (err) {
474 fuse_file_free(ff);
475 return err;
476 }
477
478 if (isdir)
479 outarg.open_flags &= ~FOPEN_DIRECT_IO;
480
481 ff->fh = outarg.fh;
482 ff->nodeid = nodeid;
483 ff->open_flags = outarg.open_flags;
484 file->private_data = fuse_file_get(ff);
485
486 return 0;
487}
08cbf542 488EXPORT_SYMBOL_GPL(fuse_do_open);
91fe96b4 489
c7b7143c 490void fuse_finish_open(struct inode *inode, struct file *file)
fd72faac 491{
c7b7143c 492 struct fuse_file *ff = file->private_data;
a0822c55 493 struct fuse_conn *fc = get_fuse_conn(inode);
c7b7143c
MS		 494
495 if (ff->open_flags & FOPEN_DIRECT_IO)
fd72faac 496 file->f_op = &fuse_direct_io_file_operations;
c7b7143c 497 if (!(ff->open_flags & FOPEN_KEEP_CACHE))
b1009979 498 invalidate_inode_pages2(inode->i_mapping);
c7b7143c 499 if (ff->open_flags & FOPEN_NONSEEKABLE)
a7c1b990 500 nonseekable_open(inode, file);
a0822c55
KS		 501 if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) {
502 struct fuse_inode *fi = get_fuse_inode(inode);
503
504 spin_lock(&fc->lock);
505 fi->attr_version = ++fc->attr_version;
506 i_size_write(inode, 0);
507 spin_unlock(&fc->lock);
508 fuse_invalidate_attr(inode);
509 }
fd72faac
MS		 510}
511
91fe96b4 512int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
fd72faac 513{
acf99433 514 struct fuse_conn *fc = get_fuse_conn(inode);
b6aeaded 515 int err;
b6aeaded
MS		 516
517 err = generic_file_open(inode, file);
518 if (err)
519 return err;
520
91fe96b4 521 err = fuse_do_open(fc, get_node_id(inode), file, isdir);
fd72faac 522 if (err)
91fe96b4 523 return err;
b6aeaded 524
91fe96b4
MS		 525 fuse_finish_open(inode, file);
526
527 return 0;
b6aeaded
MS		 528}
529
8b0797a4 530static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode)
64c6d8ed 531{
8b0797a4 532 struct fuse_conn *fc = ff->fc;
33649c91 533 struct fuse_req *req = ff->reserved_req;
b57d4264 534 struct fuse_release_in *inarg = &req->misc.release.in;
b6aeaded 535
8b0797a4
MS		 536 spin_lock(&fc->lock);
537 list_del(&ff->write_entry);
538 if (!RB_EMPTY_NODE(&ff->polled_node))
539 rb_erase(&ff->polled_node, &fc->polled_files);
540 spin_unlock(&fc->lock);
541
357ccf2b 542 wake_up_interruptible_all(&ff->poll_wait);
8b0797a4 543
b6aeaded 544 inarg->fh = ff->fh;
fd72faac 545 inarg->flags = flags;
51eb01e7 546 req->in.h.opcode = opcode;
c7b7143c 547 req->in.h.nodeid = ff->nodeid;
b6aeaded
MS		 548 req->in.numargs = 1;
549 req->in.args[0].size = sizeof(struct fuse_release_in);
550 req->in.args[0].value = inarg;
fd72faac
MS		 551}
552
8b0797a4 553void fuse_release_common(struct file *file, int opcode)
fd72faac 554{
6b2db28a
TH		 555 struct fuse_file *ff;
556 struct fuse_req *req;
b6aeaded 557
6b2db28a
TH		 558 ff = file->private_data;
559 if (unlikely(!ff))
8b0797a4 560 return;
6b2db28a 561
6b2db28a 562 req = ff->reserved_req;
8b0797a4 563 fuse_prepare_release(ff, file->f_flags, opcode);
6b2db28a 564
37fb3a30
MS		 565 if (ff->flock) {
566 struct fuse_release_in *inarg = &req->misc.release.in;
567 inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK;
568 inarg->lock_owner = fuse_lock_owner_id(ff->fc,
569 (fl_owner_t) file);
570 }
6b2db28a 571 /* Hold vfsmount and dentry until release is finished */
b0be46eb
MS		 572 path_get(&file->f_path);
573 req->misc.release.path = file->f_path;
6b2db28a 574
6b2db28a
TH		 575 /*
576 * Normally this will send the RELEASE request, however if
577 * some asynchronous READ or WRITE requests are outstanding,
578 * the sending will be delayed.
5a18ec17
MS		 579 *
580 * Make the release synchronous if this is a fuseblk mount,
581 * synchronous RELEASE is allowed (and desirable) in this case
582 * because the server can be trusted not to screw up.
6b2db28a 583 */
5a18ec17 584 fuse_file_put(ff, ff->fc->destroy_req != NULL);
b6aeaded
MS		 585}
586
04730fef
MS		 587static int fuse_open(struct inode *inode, struct file *file)
588{
91fe96b4 589 return fuse_open_common(inode, file, false);
04730fef
MS		 590}
591
592static int fuse_release(struct inode *inode, struct file *file)
593{
8b0797a4
MS		 594 fuse_release_common(file, FUSE_RELEASE);
595
596 /* return value is ignored by VFS */
597 return 0;
598}
599
600void fuse_sync_release(struct fuse_file *ff, int flags)
601{
602 WARN_ON(atomic_read(&ff->count) > 1);
603 fuse_prepare_release(ff, flags, FUSE_RELEASE);
604 ff->reserved_req->force = 1;
8b41e671 605 ff->reserved_req->background = 0;
8b0797a4
MS		 606 fuse_request_send(ff->fc, ff->reserved_req);
607 fuse_put_request(ff->fc, ff->reserved_req);
608 kfree(ff);
04730fef 609}
08cbf542 610EXPORT_SYMBOL_GPL(fuse_sync_release);
04730fef 611
71421259 612/*
9c8ef561
MS		 613 * Scramble the ID space with XTEA, so that the value of the files_struct
614 * pointer is not exposed to userspace.
71421259 615 */
f3332114 616u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
71421259 617{
9c8ef561
MS		 618 u32 *k = fc->scramble_key;
619 u64 v = (unsigned long) id;
620 u32 v0 = v;
621 u32 v1 = v >> 32;
622 u32 sum = 0;
623 int i;
624
625 for (i = 0; i < 32; i++) {
626 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
627 sum += 0x9E3779B9;
628 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
629 }
630
631 return (u64) v0 + ((u64) v1 << 32);
71421259
MS		 632}
633
3be5a52b
MS		 634/*
635 * Check if page is under writeback
636 *
637 * This is currently done by walking the list of writepage requests
638 * for the inode, which can be pretty inefficient.
639 */
640static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
641{
642 struct fuse_conn *fc = get_fuse_conn(inode);
643 struct fuse_inode *fi = get_fuse_inode(inode);
644 struct fuse_req *req;
645 bool found = false;
646
647 spin_lock(&fc->lock);
648 list_for_each_entry(req, &fi->writepages, writepages_entry) {
649 pgoff_t curr_index;
650
651 BUG_ON(req->inode != inode);
652 curr_index = req->misc.write.in.offset >> PAGE_CACHE_SHIFT;
653 if (curr_index == index) {
654 found = true;
655 break;
656 }
657 }
658 spin_unlock(&fc->lock);
659
660 return found;
661}
662
663/*
664 * Wait for page writeback to be completed.
665 *
666 * Since fuse doesn't rely on the VM writeback tracking, this has to
667 * use some other means.
668 */
669static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
670{
671 struct fuse_inode *fi = get_fuse_inode(inode);
672
673 wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
674 return 0;
675}
676
75e1fcc0 677static int fuse_flush(struct file *file, fl_owner_t id)
b6aeaded 678{
6131ffaa 679 struct inode *inode = file_inode(file);
b6aeaded
MS		 680 struct fuse_conn *fc = get_fuse_conn(inode);
681 struct fuse_file *ff = file->private_data;
682 struct fuse_req *req;
683 struct fuse_flush_in inarg;
684 int err;
685
248d86e8
MS		 686 if (is_bad_inode(inode))
687 return -EIO;
688
b6aeaded
MS		 689 if (fc->no_flush)
690 return 0;
691
b111c8c0 692 req = fuse_get_req_nofail_nopages(fc, file);
b6aeaded
MS		 693 memset(&inarg, 0, sizeof(inarg));
694 inarg.fh = ff->fh;
9c8ef561 695 inarg.lock_owner = fuse_lock_owner_id(fc, id);
b6aeaded
MS		 696 req->in.h.opcode = FUSE_FLUSH;
697 req->in.h.nodeid = get_node_id(inode);
b6aeaded
MS		 698 req->in.numargs = 1;
699 req->in.args[0].size = sizeof(inarg);
700 req->in.args[0].value = &inarg;
71421259 701 req->force = 1;
b93f858a 702 fuse_request_send(fc, req);
b6aeaded
MS		 703 err = req->out.h.error;
704 fuse_put_request(fc, req);
705 if (err == -ENOSYS) {
706 fc->no_flush = 1;
707 err = 0;
708 }
709 return err;
710}
711
3be5a52b
MS		 712/*
713 * Wait for all pending writepages on the inode to finish.
714 *
715 * This is currently done by blocking further writes with FUSE_NOWRITE
716 * and waiting for all sent writes to complete.
717 *
718 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
719 * could conflict with truncation.
720 */
721static void fuse_sync_writes(struct inode *inode)
722{
723 fuse_set_nowrite(inode);
724 fuse_release_nowrite(inode);
725}
726
02c24a82
JB		 727int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
728 int datasync, int isdir)
b6aeaded 729{
7ea80859 730 struct inode *inode = file->f_mapping->host;
b6aeaded
MS		 731 struct fuse_conn *fc = get_fuse_conn(inode);
732 struct fuse_file *ff = file->private_data;
733 struct fuse_req *req;
734 struct fuse_fsync_in inarg;
735 int err;
736
248d86e8
MS		 737 if (is_bad_inode(inode))
738 return -EIO;
739
02c24a82
JB		 740 err = filemap_write_and_wait_range(inode->i_mapping, start, end);
741 if (err)
742 return err;
743
82547981 744 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
b6aeaded
MS		 745 return 0;
746
02c24a82
JB		 747 mutex_lock(&inode->i_mutex);
748
3be5a52b
MS		 749 /*
750 * Start writeback against all dirty pages of the inode, then
751 * wait for all outstanding writes, before sending the FSYNC
752 * request.
753 */
754 err = write_inode_now(inode, 0);
755 if (err)
02c24a82 756 goto out;
3be5a52b
MS		 757
758 fuse_sync_writes(inode);
759
b111c8c0 760 req = fuse_get_req_nopages(fc);
02c24a82
JB		 761 if (IS_ERR(req)) {
762 err = PTR_ERR(req);
763 goto out;
764 }
b6aeaded
MS		 765
766 memset(&inarg, 0, sizeof(inarg));
767 inarg.fh = ff->fh;
768 inarg.fsync_flags = datasync ? 1 : 0;
82547981 769 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
b6aeaded 770 req->in.h.nodeid = get_node_id(inode);
b6aeaded
MS		 771 req->in.numargs = 1;
772 req->in.args[0].size = sizeof(inarg);
773 req->in.args[0].value = &inarg;
b93f858a 774 fuse_request_send(fc, req);
b6aeaded
MS		 775 err = req->out.h.error;
776 fuse_put_request(fc, req);
777 if (err == -ENOSYS) {
82547981
MS		 778 if (isdir)
779 fc->no_fsyncdir = 1;
780 else
781 fc->no_fsync = 1;
b6aeaded
MS		 782 err = 0;
783 }
02c24a82
JB		 784out:
785 mutex_unlock(&inode->i_mutex);
b6aeaded
MS		 786 return err;
787}
788
02c24a82
JB		 789static int fuse_fsync(struct file *file, loff_t start, loff_t end,
790 int datasync)
82547981 791{
02c24a82 792 return fuse_fsync_common(file, start, end, datasync, 0);
82547981
MS		 793}
794
2106cb18
MS		 795void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos,
796 size_t count, int opcode)
b6aeaded 797{
5c5c5e51 798 struct fuse_read_in *inarg = &req->misc.read.in;
a6643094 799 struct fuse_file *ff = file->private_data;
b6aeaded 800
361b1eb5
MS		 801 inarg->fh = ff->fh;
802 inarg->offset = pos;
803 inarg->size = count;
a6643094 804 inarg->flags = file->f_flags;
361b1eb5 805 req->in.h.opcode = opcode;
2106cb18 806 req->in.h.nodeid = ff->nodeid;
b6aeaded
MS		 807 req->in.numargs = 1;
808 req->in.args[0].size = sizeof(struct fuse_read_in);
c1aa96a5 809 req->in.args[0].value = inarg;
b6aeaded
MS		 810 req->out.argvar = 1;
811 req->out.numargs = 1;
812 req->out.args[0].size = count;
b6aeaded
MS		 813}
814
187c5c36
MP		 815static void fuse_release_user_pages(struct fuse_req *req, int write)
816{
817 unsigned i;
818
819 for (i = 0; i < req->num_pages; i++) {
820 struct page *page = req->pages[i];
821 if (write)
822 set_page_dirty_lock(page);
823 put_page(page);
824 }
825}
826
01e9d11a
MP		 827/**
828 * In case of short read, the caller sets 'pos' to the position of
829 * actual end of fuse request in IO request. Otherwise, if bytes_requested
830 * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1.
831 *
832 * An example:
833 * User requested DIO read of 64K. It was splitted into two 32K fuse requests,
834 * both submitted asynchronously. The first of them was ACKed by userspace as
835 * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The
836 * second request was ACKed as short, e.g. only 1K was read, resulting in
837 * pos == 33K.
838 *
839 * Thus, when all fuse requests are completed, the minimal non-negative 'pos'
840 * will be equal to the length of the longest contiguous fragment of
841 * transferred data starting from the beginning of IO request.
842 */
843static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
844{
845 int left;
846
847 spin_lock(&io->lock);
848 if (err)
849 io->err = io->err ? : err;
850 else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes))
851 io->bytes = pos;
852
853 left = --io->reqs;
854 spin_unlock(&io->lock);
855
856 if (!left) {
857 long res;
858
859 if (io->err)
860 res = io->err;
861 else if (io->bytes >= 0 && io->write)
862 res = -EIO;
863 else {
864 res = io->bytes < 0 ? io->size : io->bytes;
865
866 if (!is_sync_kiocb(io->iocb)) {
867 struct path *path = &io->iocb->ki_filp->f_path;
868 struct inode *inode = path->dentry->d_inode;
869 struct fuse_conn *fc = get_fuse_conn(inode);
870 struct fuse_inode *fi = get_fuse_inode(inode);
871
872 spin_lock(&fc->lock);
873 fi->attr_version = ++fc->attr_version;
874 spin_unlock(&fc->lock);
875 }
876 }
877
878 aio_complete(io->iocb, res, 0);
879 kfree(io);
880 }
881}
882
883static void fuse_aio_complete_req(struct fuse_conn *fc, struct fuse_req *req)
884{
885 struct fuse_io_priv *io = req->io;
886 ssize_t pos = -1;
887
888 fuse_release_user_pages(req, !io->write);
889
890 if (io->write) {
891 if (req->misc.write.in.size != req->misc.write.out.size)
892 pos = req->misc.write.in.offset - io->offset +
893 req->misc.write.out.size;
894 } else {
895 if (req->misc.read.in.size != req->out.args[0].size)
896 pos = req->misc.read.in.offset - io->offset +
897 req->out.args[0].size;
898 }
899
900 fuse_aio_complete(io, req->out.h.error, pos);
901}
902
903static size_t fuse_async_req_send(struct fuse_conn *fc, struct fuse_req *req,
904 size_t num_bytes, struct fuse_io_priv *io)
905{
906 spin_lock(&io->lock);
907 io->size += num_bytes;
908 io->reqs++;
909 spin_unlock(&io->lock);
910
911 req->io = io;
912 req->end = fuse_aio_complete_req;
913
36cf66ed 914 __fuse_get_request(req);
01e9d11a
MP		 915 fuse_request_send_background(fc, req);
916
917 return num_bytes;
918}
919
36cf66ed 920static size_t fuse_send_read(struct fuse_req *req, struct fuse_io_priv *io,
2106cb18 921 loff_t pos, size_t count, fl_owner_t owner)
04730fef 922{
36cf66ed 923 struct file *file = io->file;
2106cb18
MS		 924 struct fuse_file *ff = file->private_data;
925 struct fuse_conn *fc = ff->fc;
f3332114 926
2106cb18 927 fuse_read_fill(req, file, pos, count, FUSE_READ);
f3332114 928 if (owner != NULL) {
5c5c5e51 929 struct fuse_read_in *inarg = &req->misc.read.in;
f3332114
MS		 930
931 inarg->read_flags |= FUSE_READ_LOCKOWNER;
932 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
933 }
36cf66ed
MP		 934
935 if (io->async)
936 return fuse_async_req_send(fc, req, count, io);
937
b93f858a 938 fuse_request_send(fc, req);
361b1eb5 939 return req->out.args[0].size;
04730fef
MS		 940}
941
5c5c5e51
MS		 942static void fuse_read_update_size(struct inode *inode, loff_t size,
943 u64 attr_ver)
944{
945 struct fuse_conn *fc = get_fuse_conn(inode);
946 struct fuse_inode *fi = get_fuse_inode(inode);
947
948 spin_lock(&fc->lock);
4e208303
MP		 949 if (attr_ver == fi->attr_version && size < inode->i_size &&
950 !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
5c5c5e51
MS		 951 fi->attr_version = ++fc->attr_version;
952 i_size_write(inode, size);
953 }
954 spin_unlock(&fc->lock);
955}
956
b6aeaded
MS		 957static int fuse_readpage(struct file *file, struct page *page)
958{
36cf66ed 959 struct fuse_io_priv io = { .async = 0, .file = file };
b6aeaded
MS		 960 struct inode *inode = page->mapping->host;
961 struct fuse_conn *fc = get_fuse_conn(inode);
248d86e8 962 struct fuse_req *req;
5c5c5e51
MS		 963 size_t num_read;
964 loff_t pos = page_offset(page);
965 size_t count = PAGE_CACHE_SIZE;
966 u64 attr_ver;
248d86e8
MS		 967 int err;
968
969 err = -EIO;
970 if (is_bad_inode(inode))
971 goto out;
972
3be5a52b 973 /*
25985edc 974 * Page writeback can extend beyond the lifetime of the
3be5a52b
MS		 975 * page-cache page, so make sure we read a properly synced
976 * page.
977 */
978 fuse_wait_on_page_writeback(inode, page->index);
979
b111c8c0 980 req = fuse_get_req(fc, 1);
ce1d5a49
MS		 981 err = PTR_ERR(req);
982 if (IS_ERR(req))
b6aeaded
MS		 983 goto out;
984
5c5c5e51
MS		 985 attr_ver = fuse_get_attr_version(fc);
986
b6aeaded 987 req->out.page_zeroing = 1;
f4975c67 988 req->out.argpages = 1;
b6aeaded
MS		 989 req->num_pages = 1;
990 req->pages[0] = page;
85f40aec 991 req->page_descs[0].length = count;
36cf66ed 992 num_read = fuse_send_read(req, &io, pos, count, NULL);
b6aeaded
MS		 993 err = req->out.h.error;
994 fuse_put_request(fc, req);
5c5c5e51
MS		 995
996 if (!err) {
997 /*
998 * Short read means EOF. If file size is larger, truncate it
999 */
1000 if (num_read < count)
1001 fuse_read_update_size(inode, pos + num_read, attr_ver);
1002
b6aeaded 1003 SetPageUptodate(page);
5c5c5e51
MS		 1004 }
1005
b36c31ba 1006 fuse_invalidate_attr(inode); /* atime changed */
b6aeaded
MS		 1007 out:
1008 unlock_page(page);
1009 return err;
1010}
1011
c1aa96a5 1012static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
db50b96c 1013{
c1aa96a5 1014 int i;
5c5c5e51
MS		 1015 size_t count = req->misc.read.in.size;
1016 size_t num_read = req->out.args[0].size;
ce534fb0 1017 struct address_space *mapping = NULL;
c1aa96a5 1018
ce534fb0
MS		 1019 for (i = 0; mapping == NULL && i < req->num_pages; i++)
1020 mapping = req->pages[i]->mapping;
5c5c5e51 1021
ce534fb0
MS		 1022 if (mapping) {
1023 struct inode *inode = mapping->host;
1024
1025 /*
1026 * Short read means EOF. If file size is larger, truncate it
1027 */
1028 if (!req->out.h.error && num_read < count) {
1029 loff_t pos;
1030
1031 pos = page_offset(req->pages[0]) + num_read;
1032 fuse_read_update_size(inode, pos,
1033 req->misc.read.attr_ver);
1034 }
1035 fuse_invalidate_attr(inode); /* atime changed */
1036 }
c1aa96a5 1037
db50b96c
MS		 1038 for (i = 0; i < req->num_pages; i++) {
1039 struct page *page = req->pages[i];
1040 if (!req->out.h.error)
1041 SetPageUptodate(page);
c1aa96a5
MS		 1042 else
1043 SetPageError(page);
db50b96c 1044 unlock_page(page);
b5dd3285 1045 page_cache_release(page);
db50b96c 1046 }
c756e0a4 1047 if (req->ff)
5a18ec17 1048 fuse_file_put(req->ff, false);
c1aa96a5
MS		 1049}
1050
2106cb18 1051static void fuse_send_readpages(struct fuse_req *req, struct file *file)
c1aa96a5 1052{
2106cb18
MS		 1053 struct fuse_file *ff = file->private_data;
1054 struct fuse_conn *fc = ff->fc;
c1aa96a5
MS		 1055 loff_t pos = page_offset(req->pages[0]);
1056 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
f4975c67
MS		 1057
1058 req->out.argpages = 1;
c1aa96a5 1059 req->out.page_zeroing = 1;
ce534fb0 1060 req->out.page_replace = 1;
2106cb18 1061 fuse_read_fill(req, file, pos, count, FUSE_READ);
5c5c5e51 1062 req->misc.read.attr_ver = fuse_get_attr_version(fc);
9cd68455 1063 if (fc->async_read) {
c756e0a4 1064 req->ff = fuse_file_get(ff);
9cd68455 1065 req->end = fuse_readpages_end;
6fa3eb70 1066 fuse_request_send_background_ex(fc, req, count);
9cd68455 1067 } else {
6fa3eb70 1068 fuse_request_send_ex(fc, req, count);
9cd68455 1069 fuse_readpages_end(fc, req);
e9bb09dd 1070 fuse_put_request(fc, req);
9cd68455 1071 }
db50b96c
MS		 1072}
1073
c756e0a4 1074struct fuse_fill_data {
db50b96c 1075 struct fuse_req *req;
a6643094 1076 struct file *file;
db50b96c 1077 struct inode *inode;
f8dbdf81 1078 unsigned nr_pages;
db50b96c
MS		 1079};
1080
1081static int fuse_readpages_fill(void *_data, struct page *page)
1082{
c756e0a4 1083 struct fuse_fill_data *data = _data;
db50b96c
MS		 1084 struct fuse_req *req = data->req;
1085 struct inode *inode = data->inode;
1086 struct fuse_conn *fc = get_fuse_conn(inode);
1087
3be5a52b
MS		 1088 fuse_wait_on_page_writeback(inode, page->index);
1089
db50b96c
MS		 1090 if (req->num_pages &&
1091 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
1092 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
1093 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
f8dbdf81
MP		 1094 int nr_alloc = min_t(unsigned, data->nr_pages,
1095 FUSE_MAX_PAGES_PER_REQ);
2106cb18 1096 fuse_send_readpages(req, data->file);
8b41e671
MP		 1097 if (fc->async_read)
1098 req = fuse_get_req_for_background(fc, nr_alloc);
1099 else
1100 req = fuse_get_req(fc, nr_alloc);
1101
1102 data->req = req;
ce1d5a49 1103 if (IS_ERR(req)) {
db50b96c 1104 unlock_page(page);
ce1d5a49 1105 return PTR_ERR(req);
db50b96c 1106 }
db50b96c 1107 }
f8dbdf81
MP		 1108
1109 if (WARN_ON(req->num_pages >= req->max_pages)) {
1110 fuse_put_request(fc, req);
1111 return -EIO;
1112 }
1113
b5dd3285 1114 page_cache_get(page);
db50b96c 1115 req->pages[req->num_pages] = page;
85f40aec 1116 req->page_descs[req->num_pages].length = PAGE_SIZE;
1729a16c 1117 req->num_pages++;
f8dbdf81 1118 data->nr_pages--;
db50b96c
MS		 1119 return 0;
1120}
1121
1122static int fuse_readpages(struct file *file, struct address_space *mapping,
1123 struct list_head *pages, unsigned nr_pages)
1124{
1125 struct inode *inode = mapping->host;
1126 struct fuse_conn *fc = get_fuse_conn(inode);
c756e0a4 1127 struct fuse_fill_data data;
db50b96c 1128 int err;
f8dbdf81 1129 int nr_alloc = min_t(unsigned, nr_pages, FUSE_MAX_PAGES_PER_REQ);
248d86e8 1130
1d7ea732 1131 err = -EIO;
248d86e8 1132 if (is_bad_inode(inode))
2e990021 1133 goto out;
248d86e8 1134
a6643094 1135 data.file = file;
db50b96c 1136 data.inode = inode;
8b41e671
MP		 1137 if (fc->async_read)
1138 data.req = fuse_get_req_for_background(fc, nr_alloc);
1139 else
1140 data.req = fuse_get_req(fc, nr_alloc);
f8dbdf81 1141 data.nr_pages = nr_pages;
1d7ea732 1142 err = PTR_ERR(data.req);
ce1d5a49 1143 if (IS_ERR(data.req))
2e990021 1144 goto out;
db50b96c
MS		 1145
1146 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
d3406ffa
MS		 1147 if (!err) {
1148 if (data.req->num_pages)
2106cb18 1149 fuse_send_readpages(data.req, file);
d3406ffa
MS		 1150 else
1151 fuse_put_request(fc, data.req);
1152 }
2e990021 1153out:
1d7ea732 1154 return err;
db50b96c
MS		 1155}
1156
bcb4be80
MS		 1157static ssize_t fuse_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1158 unsigned long nr_segs, loff_t pos)
1159{
1160 struct inode *inode = iocb->ki_filp->f_mapping->host;
a8894274 1161 struct fuse_conn *fc = get_fuse_conn(inode);
bcb4be80 1162
a8894274
BF		 1163 /*
1164 * In auto invalidate mode, always update attributes on read.
1165 * Otherwise, only update if we attempt to read past EOF (to ensure
1166 * i_size is up to date).
1167 */
1168 if (fc->auto_inval_data ||
1169 (pos + iov_length(iov, nr_segs) > i_size_read(inode))) {
bcb4be80 1170 int err;
bcb4be80
MS		 1171 err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL);
1172 if (err)
1173 return err;
1174 }
1175
1176 return generic_file_aio_read(iocb, iov, nr_segs, pos);
1177}
1178
2d698b07 1179static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff,
2106cb18 1180 loff_t pos, size_t count)
b6aeaded 1181{
b25e82e5
MS		 1182 struct fuse_write_in *inarg = &req->misc.write.in;
1183 struct fuse_write_out *outarg = &req->misc.write.out;
b6aeaded 1184
b25e82e5
MS		 1185 inarg->fh = ff->fh;
1186 inarg->offset = pos;
1187 inarg->size = count;
b6aeaded 1188 req->in.h.opcode = FUSE_WRITE;
2106cb18 1189 req->in.h.nodeid = ff->nodeid;
b6aeaded 1190 req->in.numargs = 2;
2106cb18 1191 if (ff->fc->minor < 9)
f3332114
MS		 1192 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
1193 else
1194 req->in.args[0].size = sizeof(struct fuse_write_in);
b25e82e5 1195 req->in.args[0].value = inarg;
b6aeaded
MS		 1196 req->in.args[1].size = count;
1197 req->out.numargs = 1;
1198 req->out.args[0].size = sizeof(struct fuse_write_out);
b25e82e5
MS		 1199 req->out.args[0].value = outarg;
1200}
1201
36cf66ed 1202static size_t fuse_send_write(struct fuse_req *req, struct fuse_io_priv *io,
2106cb18 1203 loff_t pos, size_t count, fl_owner_t owner)
b25e82e5 1204{
36cf66ed 1205 struct file *file = io->file;
2106cb18
MS		 1206 struct fuse_file *ff = file->private_data;
1207 struct fuse_conn *fc = ff->fc;
2d698b07
MS		 1208 struct fuse_write_in *inarg = &req->misc.write.in;
1209
2106cb18 1210 fuse_write_fill(req, ff, pos, count);
2d698b07 1211 inarg->flags = file->f_flags;
f3332114 1212 if (owner != NULL) {
f3332114
MS		 1213 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
1214 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
1215 }
36cf66ed
MP		 1216
1217 if (io->async)
1218 return fuse_async_req_send(fc, req, count, io);
1219
6fa3eb70 1220 fuse_request_send_ex(fc, req, count);
b25e82e5 1221 return req->misc.write.out.size;
b6aeaded
MS		 1222}
1223
a1d75f25 1224void fuse_write_update_size(struct inode *inode, loff_t pos)
854512ec
MS		 1225{
1226 struct fuse_conn *fc = get_fuse_conn(inode);
1227 struct fuse_inode *fi = get_fuse_inode(inode);
1228
1229 spin_lock(&fc->lock);
1230 fi->attr_version = ++fc->attr_version;
1231 if (pos > inode->i_size)
1232 i_size_write(inode, pos);
1233 spin_unlock(&fc->lock);
1234}
1235
ea9b9907
NP		 1236static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file,
1237 struct inode *inode, loff_t pos,
1238 size_t count)
1239{
1240 size_t res;
1241 unsigned offset;
1242 unsigned i;
36cf66ed 1243 struct fuse_io_priv io = { .async = 0, .file = file };
ea9b9907
NP		 1244
1245 for (i = 0; i < req->num_pages; i++)
1246 fuse_wait_on_page_writeback(inode, req->pages[i]->index);
1247
36cf66ed 1248 res = fuse_send_write(req, &io, pos, count, NULL);
ea9b9907 1249
b2430d75 1250 offset = req->page_descs[0].offset;
ea9b9907
NP		 1251 count = res;
1252 for (i = 0; i < req->num_pages; i++) {
1253 struct page *page = req->pages[i];
1254
1255 if (!req->out.h.error && !offset && count >= PAGE_CACHE_SIZE)
1256 SetPageUptodate(page);
1257
1258 if (count > PAGE_CACHE_SIZE - offset)
1259 count -= PAGE_CACHE_SIZE - offset;
1260 else
1261 count = 0;
1262 offset = 0;
1263
1264 unlock_page(page);
1265 page_cache_release(page);
1266 }
1267
1268 return res;
1269}
1270
1271static ssize_t fuse_fill_write_pages(struct fuse_req *req,
1272 struct address_space *mapping,
1273 struct iov_iter *ii, loff_t pos)
1274{
1275 struct fuse_conn *fc = get_fuse_conn(mapping->host);
1276 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1277 size_t count = 0;
1278 int err;
1279
f4975c67 1280 req->in.argpages = 1;
b2430d75 1281 req->page_descs[0].offset = offset;
ea9b9907
NP		 1282
1283 do {
1284 size_t tmp;
1285 struct page *page;
1286 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1287 size_t bytes = min_t(size_t, PAGE_CACHE_SIZE - offset,
1288 iov_iter_count(ii));
1289
1290 bytes = min_t(size_t, bytes, fc->max_write - count);
1291
1292 again:
1293 err = -EFAULT;
1294 if (iov_iter_fault_in_readable(ii, bytes))
1295 break;
1296
1297 err = -ENOMEM;
54566b2c 1298 page = grab_cache_page_write_begin(mapping, index, 0);
ea9b9907
NP		 1299 if (!page)
1300 break;
1301
931e80e4 1302 if (mapping_writably_mapped(mapping))
1303 flush_dcache_page(page);
1304
ea9b9907
NP		 1305 pagefault_disable();
1306 tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes);
1307 pagefault_enable();
1308 flush_dcache_page(page);
1309
478e0841
JW		 1310 mark_page_accessed(page);
1311
020ef191 1312 iov_iter_advance(ii, tmp);
ea9b9907
NP		 1313 if (!tmp) {
1314 unlock_page(page);
1315 page_cache_release(page);
1316 bytes = min(bytes, iov_iter_single_seg_count(ii));
1317 goto again;
1318 }
1319
1320 err = 0;
1321 req->pages[req->num_pages] = page;
85f40aec 1322 req->page_descs[req->num_pages].length = tmp;
ea9b9907
NP		 1323 req->num_pages++;
1324
ea9b9907
NP		 1325 count += tmp;
1326 pos += tmp;
1327 offset += tmp;
1328 if (offset == PAGE_CACHE_SIZE)
1329 offset = 0;
1330
78bb6cb9
MS		 1331 if (!fc->big_writes)
1332 break;
ea9b9907 1333 } while (iov_iter_count(ii) && count < fc->max_write &&
d07f09f5 1334 req->num_pages < req->max_pages && offset == 0);
ea9b9907
NP		 1335
1336 return count > 0 ? count : err;
1337}
1338
d07f09f5
MP		 1339static inline unsigned fuse_wr_pages(loff_t pos, size_t len)
1340{
1341 return min_t(unsigned,
1342 ((pos + len - 1) >> PAGE_CACHE_SHIFT) -
1343 (pos >> PAGE_CACHE_SHIFT) + 1,
1344 FUSE_MAX_PAGES_PER_REQ);
1345}
1346
ea9b9907
NP		 1347static ssize_t fuse_perform_write(struct file *file,
1348 struct address_space *mapping,
1349 struct iov_iter *ii, loff_t pos)
1350{
1351 struct inode *inode = mapping->host;
1352 struct fuse_conn *fc = get_fuse_conn(inode);
4e208303 1353 struct fuse_inode *fi = get_fuse_inode(inode);
ea9b9907
NP		 1354 int err = 0;
1355 ssize_t res = 0;
1356
1357 if (is_bad_inode(inode))
1358 return -EIO;
1359
4e208303
MP		 1360 if (inode->i_size < pos + iov_iter_count(ii))
1361 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1362
ea9b9907
NP		 1363 do {
1364 struct fuse_req *req;
1365 ssize_t count;
d07f09f5 1366 unsigned nr_pages = fuse_wr_pages(pos, iov_iter_count(ii));
ea9b9907 1367
d07f09f5 1368 req = fuse_get_req(fc, nr_pages);
ea9b9907
NP		 1369 if (IS_ERR(req)) {
1370 err = PTR_ERR(req);
1371 break;
1372 }
1373
1374 count = fuse_fill_write_pages(req, mapping, ii, pos);
1375 if (count <= 0) {
1376 err = count;
1377 } else {
1378 size_t num_written;
1379
1380 num_written = fuse_send_write_pages(req, file, inode,
1381 pos, count);
1382 err = req->out.h.error;
1383 if (!err) {
1384 res += num_written;
1385 pos += num_written;
1386
1387 /* break out of the loop on short write */
1388 if (num_written != count)
1389 err = -EIO;
1390 }
1391 }
1392 fuse_put_request(fc, req);
1393 } while (!err && iov_iter_count(ii));
1394
1395 if (res > 0)
1396 fuse_write_update_size(inode, pos);
1397
4e208303 1398 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
ea9b9907
NP		 1399 fuse_invalidate_attr(inode);
1400
1401 return res > 0 ? res : err;
1402}
1403
1404static ssize_t fuse_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
1405 unsigned long nr_segs, loff_t pos)
1406{
1407 struct file *file = iocb->ki_filp;
1408 struct address_space *mapping = file->f_mapping;
1409 size_t count = 0;
4273b793 1410 size_t ocount = 0;
ea9b9907 1411 ssize_t written = 0;
4273b793 1412 ssize_t written_buffered = 0;
ea9b9907
NP		 1413 struct inode *inode = mapping->host;
1414 ssize_t err;
1415 struct iov_iter i;
4273b793 1416 loff_t endbyte = 0;
ea9b9907
NP		 1417
1418 WARN_ON(iocb->ki_pos != pos);
1419
4273b793
AA		 1420 ocount = 0;
1421 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
ea9b9907
NP		 1422 if (err)
1423 return err;
1424
4273b793 1425 count = ocount;
ea9b9907 1426 mutex_lock(&inode->i_mutex);
ea9b9907
NP		 1427
1428 /* We can write back this queue in page reclaim */
1429 current->backing_dev_info = mapping->backing_dev_info;
1430
1431 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1432 if (err)
1433 goto out;
1434
1435 if (count == 0)
1436 goto out;
1437
2f1936b8 1438 err = file_remove_suid(file);
ea9b9907
NP		 1439 if (err)
1440 goto out;
1441
c3b2da31
JB		 1442 err = file_update_time(file);
1443 if (err)
1444 goto out;
ea9b9907 1445
4273b793
AA		 1446 if (file->f_flags & O_DIRECT) {
1447 written = generic_file_direct_write(iocb, iov, &nr_segs,
1448 pos, &iocb->ki_pos,
1449 count, ocount);
1450 if (written < 0 || written == count)
1451 goto out;
1452
1453 pos += written;
1454 count -= written;
ea9b9907 1455
4273b793
AA		 1456 iov_iter_init(&i, iov, nr_segs, count, written);
1457 written_buffered = fuse_perform_write(file, mapping, &i, pos);
1458 if (written_buffered < 0) {
1459 err = written_buffered;
1460 goto out;
1461 }
1462 endbyte = pos + written_buffered - 1;
1463
1464 err = filemap_write_and_wait_range(file->f_mapping, pos,
1465 endbyte);
1466 if (err)
1467 goto out;
1468
1469 invalidate_mapping_pages(file->f_mapping,
1470 pos >> PAGE_CACHE_SHIFT,
1471 endbyte >> PAGE_CACHE_SHIFT);
1472
1473 written += written_buffered;
1474 iocb->ki_pos = pos + written_buffered;
1475 } else {
1476 iov_iter_init(&i, iov, nr_segs, count, 0);
1477 written = fuse_perform_write(file, mapping, &i, pos);
1478 if (written >= 0)
1479 iocb->ki_pos = pos + written;
1480 }
ea9b9907
NP		 1481out:
1482 current->backing_dev_info = NULL;
1483 mutex_unlock(&inode->i_mutex);
1484
1485 return written ? written : err;
1486}
1487
7c190c8b
MP		 1488static inline void fuse_page_descs_length_init(struct fuse_req *req,
1489 unsigned index, unsigned nr_pages)
85f40aec
MP		 1490{
1491 int i;
1492
7c190c8b 1493 for (i = index; i < index + nr_pages; i++)
85f40aec
MP		 1494 req->page_descs[i].length = PAGE_SIZE -
1495 req->page_descs[i].offset;
1496}
1497
7c190c8b
MP		 1498static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii)
1499{
1500 return (unsigned long)ii->iov->iov_base + ii->iov_offset;
1501}
1502
1503static inline size_t fuse_get_frag_size(const struct iov_iter *ii,
1504 size_t max_size)
1505{
1506 return min(iov_iter_single_seg_count(ii), max_size);
1507}
1508
b98d023a 1509static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii,
ce60a2f1 1510 size_t *nbytesp, int write)
413ef8cb 1511{
7c190c8b 1512 size_t nbytes = 0; /* # bytes already packed in req */
b98d023a 1513
f4975c67
MS		 1514 /* Special case for kernel I/O: can copy directly into the buffer */
1515 if (segment_eq(get_fs(), KERNEL_DS)) {
7c190c8b
MP		 1516 unsigned long user_addr = fuse_get_user_addr(ii);
1517 size_t frag_size = fuse_get_frag_size(ii, *nbytesp);
1518
f4975c67
MS		 1519 if (write)
1520 req->in.args[1].value = (void *) user_addr;
1521 else
1522 req->out.args[0].value = (void *) user_addr;
1523
b98d023a
MP		 1524 iov_iter_advance(ii, frag_size);
1525 *nbytesp = frag_size;
f4975c67
MS		 1526 return 0;
1527 }
413ef8cb 1528
5565a9d8 1529 while (nbytes < *nbytesp && req->num_pages < req->max_pages) {
7c190c8b
MP		 1530 unsigned npages;
1531 unsigned long user_addr = fuse_get_user_addr(ii);
1532 unsigned offset = user_addr & ~PAGE_MASK;
1533 size_t frag_size = fuse_get_frag_size(ii, *nbytesp - nbytes);
1534 int ret;
413ef8cb 1535
5565a9d8 1536 unsigned n = req->max_pages - req->num_pages;
7c190c8b
MP		 1537 frag_size = min_t(size_t, frag_size, n << PAGE_SHIFT);
1538
1539 npages = (frag_size + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1540 npages = clamp(npages, 1U, n);
1541
1542 ret = get_user_pages_fast(user_addr, npages, !write,
1543 &req->pages[req->num_pages]);
1544 if (ret < 0)
1545 return ret;
1546
1547 npages = ret;
1548 frag_size = min_t(size_t, frag_size,
1549 (npages << PAGE_SHIFT) - offset);
1550 iov_iter_advance(ii, frag_size);
1551
1552 req->page_descs[req->num_pages].offset = offset;
1553 fuse_page_descs_length_init(req, req->num_pages, npages);
1554
1555 req->num_pages += npages;
1556 req->page_descs[req->num_pages - 1].length -=
1557 (npages << PAGE_SHIFT) - offset - frag_size;
1558
1559 nbytes += frag_size;
1560 }
f4975c67
MS		 1561
1562 if (write)
1563 req->in.argpages = 1;
1564 else
1565 req->out.argpages = 1;
1566
7c190c8b 1567 *nbytesp = nbytes;
f4975c67 1568
413ef8cb
MS		 1569 return 0;
1570}
1571
5565a9d8
MP		 1572static inline int fuse_iter_npages(const struct iov_iter *ii_p)
1573{
1574 struct iov_iter ii = *ii_p;
1575 int npages = 0;
1576
1577 while (iov_iter_count(&ii) && npages < FUSE_MAX_PAGES_PER_REQ) {
1578 unsigned long user_addr = fuse_get_user_addr(&ii);
1579 unsigned offset = user_addr & ~PAGE_MASK;
1580 size_t frag_size = iov_iter_single_seg_count(&ii);
1581
1582 npages += (frag_size + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1583 iov_iter_advance(&ii, frag_size);
1584 }
1585
1586 return min(npages, FUSE_MAX_PAGES_PER_REQ);
1587}
1588
36cf66ed 1589ssize_t fuse_direct_io(struct fuse_io_priv *io, const struct iovec *iov,
fb05f41f
MS		 1590 unsigned long nr_segs, size_t count, loff_t *ppos,
1591 int write)
413ef8cb 1592{
36cf66ed 1593 struct file *file = io->file;
2106cb18
MS		 1594 struct fuse_file *ff = file->private_data;
1595 struct fuse_conn *fc = ff->fc;
413ef8cb
MS		 1596 size_t nmax = write ? fc->max_write : fc->max_read;
1597 loff_t pos = *ppos;
1598 ssize_t res = 0;
248d86e8 1599 struct fuse_req *req;
b98d023a
MP		 1600 struct iov_iter ii;
1601
1602 iov_iter_init(&ii, iov, nr_segs, count, 0);
248d86e8 1603
de82b923
BF		 1604 if (io->async)
1605 req = fuse_get_req_for_background(fc, fuse_iter_npages(&ii));
1606 else
1607 req = fuse_get_req(fc, fuse_iter_npages(&ii));
ce1d5a49
MS		 1608 if (IS_ERR(req))
1609 return PTR_ERR(req);
413ef8cb
MS		 1610
1611 while (count) {
413ef8cb 1612 size_t nres;
2106cb18 1613 fl_owner_t owner = current->files;
f4975c67 1614 size_t nbytes = min(count, nmax);
b98d023a 1615 int err = fuse_get_user_pages(req, &ii, &nbytes, write);
413ef8cb
MS		 1616 if (err) {
1617 res = err;
1618 break;
1619 }
f4975c67 1620
413ef8cb 1621 if (write)
36cf66ed 1622 nres = fuse_send_write(req, io, pos, nbytes, owner);
413ef8cb 1623 else
36cf66ed 1624 nres = fuse_send_read(req, io, pos, nbytes, owner);
2106cb18 1625
36cf66ed
MP		 1626 if (!io->async)
1627 fuse_release_user_pages(req, !write);
413ef8cb
MS		 1628 if (req->out.h.error) {
1629 if (!res)
1630 res = req->out.h.error;
1631 break;
1632 } else if (nres > nbytes) {
1633 res = -EIO;
1634 break;
1635 }
1636 count -= nres;
1637 res += nres;
1638 pos += nres;
413ef8cb
MS		 1639 if (nres != nbytes)
1640 break;
56cf34ff
MS		 1641 if (count) {
1642 fuse_put_request(fc, req);
de82b923
BF		 1643 if (io->async)
1644 req = fuse_get_req_for_background(fc,
1645 fuse_iter_npages(&ii));
1646 else
1647 req = fuse_get_req(fc, fuse_iter_npages(&ii));
56cf34ff
MS		 1648 if (IS_ERR(req))
1649 break;
1650 }
413ef8cb 1651 }
f60311d5
AA		 1652 if (!IS_ERR(req))
1653 fuse_put_request(fc, req);
d09cb9d7 1654 if (res > 0)
413ef8cb 1655 *ppos = pos;
413ef8cb
MS		 1656
1657 return res;
1658}
08cbf542 1659EXPORT_SYMBOL_GPL(fuse_direct_io);
413ef8cb 1660
36cf66ed
MP		 1661static ssize_t __fuse_direct_read(struct fuse_io_priv *io,
1662 const struct iovec *iov,
439ee5f0
MP		 1663 unsigned long nr_segs, loff_t *ppos,
1664 size_t count)
413ef8cb 1665{
d09cb9d7 1666 ssize_t res;
36cf66ed 1667 struct file *file = io->file;
6131ffaa 1668 struct inode *inode = file_inode(file);
d09cb9d7
MS		 1669
1670 if (is_bad_inode(inode))
1671 return -EIO;
1672
439ee5f0 1673 res = fuse_direct_io(io, iov, nr_segs, count, ppos, 0);
d09cb9d7
MS		 1674
1675 fuse_invalidate_attr(inode);
1676
1677 return res;
413ef8cb
MS		 1678}
1679
b98d023a
MP		 1680static ssize_t fuse_direct_read(struct file *file, char __user *buf,
1681 size_t count, loff_t *ppos)
1682{
36cf66ed 1683 struct fuse_io_priv io = { .async = 0, .file = file };
fb05f41f 1684 struct iovec iov = { .iov_base = buf, .iov_len = count };
439ee5f0 1685 return __fuse_direct_read(&io, &iov, 1, ppos, count);
b98d023a
MP		 1686}
1687
36cf66ed
MP		 1688static ssize_t __fuse_direct_write(struct fuse_io_priv *io,
1689 const struct iovec *iov,
b98d023a 1690 unsigned long nr_segs, loff_t *ppos)
413ef8cb 1691{
36cf66ed 1692 struct file *file = io->file;
6131ffaa 1693 struct inode *inode = file_inode(file);
b98d023a 1694 size_t count = iov_length(iov, nr_segs);
413ef8cb 1695 ssize_t res;
d09cb9d7 1696
889f7848 1697 res = generic_write_checks(file, ppos, &count, 0);
bcba24cc 1698 if (!res)
36cf66ed 1699 res = fuse_direct_io(io, iov, nr_segs, count, ppos, 1);
d09cb9d7
MS		 1700
1701 fuse_invalidate_attr(inode);
1702
413ef8cb
MS		 1703 return res;
1704}
1705
4273b793
AA		 1706static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
1707 size_t count, loff_t *ppos)
1708{
fb05f41f 1709 struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = count };
6131ffaa 1710 struct inode *inode = file_inode(file);
4273b793 1711 ssize_t res;
36cf66ed 1712 struct fuse_io_priv io = { .async = 0, .file = file };
4273b793
AA		 1713
1714 if (is_bad_inode(inode))
1715 return -EIO;
1716
1717 /* Don't allow parallel writes to the same file */
1718 mutex_lock(&inode->i_mutex);
36cf66ed 1719 res = __fuse_direct_write(&io, &iov, 1, ppos);
bcba24cc
MP		 1720 if (res > 0)
1721 fuse_write_update_size(inode, *ppos);
4273b793
AA		 1722 mutex_unlock(&inode->i_mutex);
1723
1724 return res;
1725}
1726
3be5a52b 1727static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
b6aeaded 1728{
3be5a52b 1729 __free_page(req->pages[0]);
5a18ec17 1730 fuse_file_put(req->ff, false);
3be5a52b
MS		 1731}
1732
1733static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
1734{
1735 struct inode *inode = req->inode;
1736 struct fuse_inode *fi = get_fuse_inode(inode);
1737 struct backing_dev_info *bdi = inode->i_mapping->backing_dev_info;
1738
1739 list_del(&req->writepages_entry);
1740 dec_bdi_stat(bdi, BDI_WRITEBACK);
1741 dec_zone_page_state(req->pages[0], NR_WRITEBACK_TEMP);
1742 bdi_writeout_inc(bdi);
1743 wake_up(&fi->page_waitq);
1744}
1745
1746/* Called under fc->lock, may release and reacquire it */
1747static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req)
b9ca67b2
MS		 1748__releases(fc->lock)
1749__acquires(fc->lock)
3be5a52b
MS		 1750{
1751 struct fuse_inode *fi = get_fuse_inode(req->inode);
1752 loff_t size = i_size_read(req->inode);
1753 struct fuse_write_in *inarg = &req->misc.write.in;
1754
1755 if (!fc->connected)
1756 goto out_free;
1757
1758 if (inarg->offset + PAGE_CACHE_SIZE <= size) {
1759 inarg->size = PAGE_CACHE_SIZE;
1760 } else if (inarg->offset < size) {
1761 inarg->size = size & (PAGE_CACHE_SIZE - 1);
1762 } else {
1763 /* Got truncated off completely */
1764 goto out_free;
b6aeaded 1765 }
3be5a52b
MS		 1766
1767 req->in.args[1].size = inarg->size;
1768 fi->writectr++;
b93f858a 1769 fuse_request_send_background_locked(fc, req);
3be5a52b
MS		 1770 return;
1771
1772 out_free:
1773 fuse_writepage_finish(fc, req);
1774 spin_unlock(&fc->lock);
1775 fuse_writepage_free(fc, req);
e9bb09dd 1776 fuse_put_request(fc, req);
3be5a52b 1777 spin_lock(&fc->lock);
b6aeaded
MS		 1778}
1779
3be5a52b
MS		 1780/*
1781 * If fi->writectr is positive (no truncate or fsync going on) send
1782 * all queued writepage requests.
1783 *
1784 * Called with fc->lock
1785 */
1786void fuse_flush_writepages(struct inode *inode)
b9ca67b2
MS		 1787__releases(fc->lock)
1788__acquires(fc->lock)
b6aeaded 1789{
3be5a52b
MS		 1790 struct fuse_conn *fc = get_fuse_conn(inode);
1791 struct fuse_inode *fi = get_fuse_inode(inode);
1792 struct fuse_req *req;
1793
1794 while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
1795 req = list_entry(fi->queued_writes.next, struct fuse_req, list);
1796 list_del_init(&req->list);
1797 fuse_send_writepage(fc, req);
1798 }
1799}
1800
1801static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
1802{
1803 struct inode *inode = req->inode;
1804 struct fuse_inode *fi = get_fuse_inode(inode);
1805
1806 mapping_set_error(inode->i_mapping, req->out.h.error);
1807 spin_lock(&fc->lock);
1808 fi->writectr--;
1809 fuse_writepage_finish(fc, req);
1810 spin_unlock(&fc->lock);
1811 fuse_writepage_free(fc, req);
1812}
1813
1814static int fuse_writepage_locked(struct page *page)
1815{
1816 struct address_space *mapping = page->mapping;
1817 struct inode *inode = mapping->host;
1818 struct fuse_conn *fc = get_fuse_conn(inode);
1819 struct fuse_inode *fi = get_fuse_inode(inode);
1820 struct fuse_req *req;
1821 struct fuse_file *ff;
1822 struct page *tmp_page;
1823
1824 set_page_writeback(page);
1825
4250c066 1826 req = fuse_request_alloc_nofs(1);
3be5a52b
MS		 1827 if (!req)
1828 goto err;
1829
8b41e671 1830 req->background = 1; /* writeback always goes to bg_queue */
3be5a52b
MS		 1831 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1832 if (!tmp_page)
1833 goto err_free;
1834
1835 spin_lock(&fc->lock);
1836 BUG_ON(list_empty(&fi->write_files));
1837 ff = list_entry(fi->write_files.next, struct fuse_file, write_entry);
1838 req->ff = fuse_file_get(ff);
1839 spin_unlock(&fc->lock);
1840
2106cb18 1841 fuse_write_fill(req, ff, page_offset(page), 0);
3be5a52b
MS		 1842
1843 copy_highpage(tmp_page, page);
2d698b07 1844 req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
f4975c67 1845 req->in.argpages = 1;
3be5a52b
MS		 1846 req->num_pages = 1;
1847 req->pages[0] = tmp_page;
b2430d75 1848 req->page_descs[0].offset = 0;
85f40aec 1849 req->page_descs[0].length = PAGE_SIZE;
3be5a52b
MS		 1850 req->end = fuse_writepage_end;
1851 req->inode = inode;
1852
1853 inc_bdi_stat(mapping->backing_dev_info, BDI_WRITEBACK);
1854 inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP);
3be5a52b
MS		 1855
1856 spin_lock(&fc->lock);
1857 list_add(&req->writepages_entry, &fi->writepages);
1858 list_add_tail(&req->list, &fi->queued_writes);
1859 fuse_flush_writepages(inode);
1860 spin_unlock(&fc->lock);
1861
3002e63b
MP		 1862 end_page_writeback(page);
1863
3be5a52b
MS		 1864 return 0;
1865
1866err_free:
1867 fuse_request_free(req);
1868err:
1869 end_page_writeback(page);
1870 return -ENOMEM;
1871}
1872
1873static int fuse_writepage(struct page *page, struct writeback_control *wbc)
1874{
1875 int err;
1876
1877 err = fuse_writepage_locked(page);
1878 unlock_page(page);
1879
1880 return err;
1881}
1882
1883static int fuse_launder_page(struct page *page)
1884{
1885 int err = 0;
1886 if (clear_page_dirty_for_io(page)) {
1887 struct inode *inode = page->mapping->host;
1888 err = fuse_writepage_locked(page);
1889 if (!err)
1890 fuse_wait_on_page_writeback(inode, page->index);
1891 }
1892 return err;
1893}
1894
1895/*
1896 * Write back dirty pages now, because there may not be any suitable
1897 * open files later
1898 */
1899static void fuse_vma_close(struct vm_area_struct *vma)
1900{
1901 filemap_write_and_wait(vma->vm_file->f_mapping);
1902}
1903
1904/*
1905 * Wait for writeback against this page to complete before allowing it
1906 * to be marked dirty again, and hence written back again, possibly
1907 * before the previous writepage completed.
1908 *
1909 * Block here, instead of in ->writepage(), so that the userspace fs
1910 * can only block processes actually operating on the filesystem.
1911 *
1912 * Otherwise unprivileged userspace fs would be able to block
1913 * unrelated:
1914 *
1915 * - page migration
1916 * - sync(2)
1917 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
1918 */
c2ec175c 1919static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
3be5a52b 1920{
c2ec175c 1921 struct page *page = vmf->page;
3be5a52b
MS		 1922 /*
1923 * Don't use page->mapping as it may become NULL from a
1924 * concurrent truncate.
1925 */
1926 struct inode *inode = vma->vm_file->f_mapping->host;
1927
1928 fuse_wait_on_page_writeback(inode, page->index);
1929 return 0;
1930}
1931
f0f37e2f 1932static const struct vm_operations_struct fuse_file_vm_ops = {
3be5a52b
MS		 1933 .close = fuse_vma_close,
1934 .fault = filemap_fault,
1935 .page_mkwrite = fuse_page_mkwrite,
0b173bc4 1936 .remap_pages = generic_file_remap_pages,
3be5a52b
MS		 1937};
1938
1939static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
1940{
1941 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) {
6131ffaa 1942 struct inode *inode = file_inode(file);
3be5a52b
MS		 1943 struct fuse_conn *fc = get_fuse_conn(inode);
1944 struct fuse_inode *fi = get_fuse_inode(inode);
1945 struct fuse_file *ff = file->private_data;
1946 /*
1947 * file may be written through mmap, so chain it onto the
1948 * inodes's write_file list
1949 */
1950 spin_lock(&fc->lock);
1951 if (list_empty(&ff->write_entry))
1952 list_add(&ff->write_entry, &fi->write_files);
1953 spin_unlock(&fc->lock);
1954 }
1955 file_accessed(file);
1956 vma->vm_ops = &fuse_file_vm_ops;
b6aeaded
MS		 1957 return 0;
1958}
1959
fc280c96
MS		 1960static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma)
1961{
1962 /* Can't provide the coherency needed for MAP_SHARED */
1963 if (vma->vm_flags & VM_MAYSHARE)
1964 return -ENODEV;
1965
3121bfe7
MS		 1966 invalidate_inode_pages2(file->f_mapping);
1967
fc280c96
MS		 1968 return generic_file_mmap(file, vma);
1969}
1970
71421259
MS		 1971static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
1972 struct file_lock *fl)
1973{
1974 switch (ffl->type) {
1975 case F_UNLCK:
1976 break;
1977
1978 case F_RDLCK:
1979 case F_WRLCK:
1980 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
1981 ffl->end < ffl->start)
1982 return -EIO;
1983
1984 fl->fl_start = ffl->start;
1985 fl->fl_end = ffl->end;
1986 fl->fl_pid = ffl->pid;
1987 break;
1988
1989 default:
1990 return -EIO;
1991 }
1992 fl->fl_type = ffl->type;
1993 return 0;
1994}
1995
1996static void fuse_lk_fill(struct fuse_req *req, struct file *file,
a9ff4f87
MS		 1997 const struct file_lock *fl, int opcode, pid_t pid,
1998 int flock)
71421259 1999{
6131ffaa 2000 struct inode *inode = file_inode(file);
9c8ef561 2001 struct fuse_conn *fc = get_fuse_conn(inode);
71421259
MS		 2002 struct fuse_file *ff = file->private_data;
2003 struct fuse_lk_in *arg = &req->misc.lk_in;
2004
2005 arg->fh = ff->fh;
9c8ef561 2006 arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
71421259
MS		 2007 arg->lk.start = fl->fl_start;
2008 arg->lk.end = fl->fl_end;
2009 arg->lk.type = fl->fl_type;
2010 arg->lk.pid = pid;
a9ff4f87
MS		 2011 if (flock)
2012 arg->lk_flags |= FUSE_LK_FLOCK;
71421259
MS		 2013 req->in.h.opcode = opcode;
2014 req->in.h.nodeid = get_node_id(inode);
2015 req->in.numargs = 1;
2016 req->in.args[0].size = sizeof(*arg);
2017 req->in.args[0].value = arg;
2018}
2019
2020static int fuse_getlk(struct file *file, struct file_lock *fl)
2021{
6131ffaa 2022 struct inode *inode = file_inode(file);
71421259
MS		 2023 struct fuse_conn *fc = get_fuse_conn(inode);
2024 struct fuse_req *req;
2025 struct fuse_lk_out outarg;
2026 int err;
2027
b111c8c0 2028 req = fuse_get_req_nopages(fc);
71421259
MS		 2029 if (IS_ERR(req))
2030 return PTR_ERR(req);
2031
a9ff4f87 2032 fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0);
71421259
MS		 2033 req->out.numargs = 1;
2034 req->out.args[0].size = sizeof(outarg);
2035 req->out.args[0].value = &outarg;
b93f858a 2036 fuse_request_send(fc, req);
71421259
MS		 2037 err = req->out.h.error;
2038 fuse_put_request(fc, req);
2039 if (!err)
2040 err = convert_fuse_file_lock(&outarg.lk, fl);
2041
2042 return err;
2043}
2044
a9ff4f87 2045static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
71421259 2046{
6131ffaa 2047 struct inode *inode = file_inode(file);
71421259
MS		 2048 struct fuse_conn *fc = get_fuse_conn(inode);
2049 struct fuse_req *req;
2050 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
2051 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
2052 int err;
2053
8fb47a4f 2054 if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
48e90761
MS		 2055 /* NLM needs asynchronous locks, which we don't support yet */
2056 return -ENOLCK;
2057 }
2058
71421259
MS		 2059 /* Unlock on close is handled by the flush method */
2060 if (fl->fl_flags & FL_CLOSE)
2061 return 0;
2062
b111c8c0 2063 req = fuse_get_req_nopages(fc);
71421259
MS		 2064 if (IS_ERR(req))
2065 return PTR_ERR(req);
2066
a9ff4f87 2067 fuse_lk_fill(req, file, fl, opcode, pid, flock);
b93f858a 2068 fuse_request_send(fc, req);
71421259 2069 err = req->out.h.error;
a4d27e75
MS		 2070 /* locking is restartable */
2071 if (err == -EINTR)
2072 err = -ERESTARTSYS;
71421259
MS		 2073 fuse_put_request(fc, req);
2074 return err;
2075}
2076
2077static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
2078{
6131ffaa 2079 struct inode *inode = file_inode(file);
71421259
MS		 2080 struct fuse_conn *fc = get_fuse_conn(inode);
2081 int err;
2082
48e90761
MS		 2083 if (cmd == F_CANCELLK) {
2084 err = 0;
2085 } else if (cmd == F_GETLK) {
71421259 2086 if (fc->no_lock) {
9d6a8c5c 2087 posix_test_lock(file, fl);
71421259
MS		 2088 err = 0;
2089 } else
2090 err = fuse_getlk(file, fl);
2091 } else {
2092 if (fc->no_lock)
48e90761 2093 err = posix_lock_file(file, fl, NULL);
71421259 2094 else
a9ff4f87 2095 err = fuse_setlk(file, fl, 0);
71421259
MS		 2096 }
2097 return err;
2098}
2099
a9ff4f87
MS		 2100static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
2101{
6131ffaa 2102 struct inode *inode = file_inode(file);
a9ff4f87
MS		 2103 struct fuse_conn *fc = get_fuse_conn(inode);
2104 int err;
2105
37fb3a30 2106 if (fc->no_flock) {
a9ff4f87
MS		 2107 err = flock_lock_file_wait(file, fl);
2108 } else {
37fb3a30
MS		 2109 struct fuse_file *ff = file->private_data;
2110
a9ff4f87
MS		 2111 /* emulate flock with POSIX locks */
2112 fl->fl_owner = (fl_owner_t) file;
37fb3a30 2113 ff->flock = true;
a9ff4f87
MS		 2114 err = fuse_setlk(file, fl, 1);
2115 }
2116
2117 return err;
2118}
2119
b2d2272f
MS		 2120static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
2121{
2122 struct inode *inode = mapping->host;
2123 struct fuse_conn *fc = get_fuse_conn(inode);
2124 struct fuse_req *req;
2125 struct fuse_bmap_in inarg;
2126 struct fuse_bmap_out outarg;
2127 int err;
2128
2129 if (!inode->i_sb->s_bdev || fc->no_bmap)
2130 return 0;
2131
b111c8c0 2132 req = fuse_get_req_nopages(fc);
b2d2272f
MS		 2133 if (IS_ERR(req))
2134 return 0;
2135
2136 memset(&inarg, 0, sizeof(inarg));
2137 inarg.block = block;
2138 inarg.blocksize = inode->i_sb->s_blocksize;
2139 req->in.h.opcode = FUSE_BMAP;
2140 req->in.h.nodeid = get_node_id(inode);
2141 req->in.numargs = 1;
2142 req->in.args[0].size = sizeof(inarg);
2143 req->in.args[0].value = &inarg;
2144 req->out.numargs = 1;
2145 req->out.args[0].size = sizeof(outarg);
2146 req->out.args[0].value = &outarg;
b93f858a 2147 fuse_request_send(fc, req);
b2d2272f
MS		 2148 err = req->out.h.error;
2149 fuse_put_request(fc, req);
2150 if (err == -ENOSYS)
2151 fc->no_bmap = 1;
2152
2153 return err ? 0 : outarg.block;
2154}
2155
965c8e59 2156static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence)
5559b8f4
MS		 2157{
2158 loff_t retval;
6131ffaa 2159 struct inode *inode = file_inode(file);
5559b8f4 2160
c07c3d19 2161 /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */
965c8e59
AM		 2162 if (whence == SEEK_CUR || whence == SEEK_SET)
2163 return generic_file_llseek(file, offset, whence);
06222e49 2164
c07c3d19
MS		 2165 mutex_lock(&inode->i_mutex);
2166 retval = fuse_update_attributes(inode, NULL, file, NULL);
2167 if (!retval)
965c8e59 2168 retval = generic_file_llseek(file, offset, whence);
5559b8f4 2169 mutex_unlock(&inode->i_mutex);
c07c3d19 2170
5559b8f4
MS		 2171 return retval;
2172}
2173
59efec7b
TH		 2174static int fuse_ioctl_copy_user(struct page **pages, struct iovec *iov,
2175 unsigned int nr_segs, size_t bytes, bool to_user)
2176{
2177 struct iov_iter ii;
2178 int page_idx = 0;
2179
2180 if (!bytes)
2181 return 0;
2182
2183 iov_iter_init(&ii, iov, nr_segs, bytes, 0);
2184
2185 while (iov_iter_count(&ii)) {
2186 struct page *page = pages[page_idx++];
2187 size_t todo = min_t(size_t, PAGE_SIZE, iov_iter_count(&ii));
4aa0edd2 2188 void *kaddr;
59efec7b 2189
4aa0edd2 2190 kaddr = kmap(page);
59efec7b
TH		 2191
2192 while (todo) {
2193 char __user *uaddr = ii.iov->iov_base + ii.iov_offset;
2194 size_t iov_len = ii.iov->iov_len - ii.iov_offset;
2195 size_t copy = min(todo, iov_len);
2196 size_t left;
2197
2198 if (!to_user)
2199 left = copy_from_user(kaddr, uaddr, copy);
2200 else
2201 left = copy_to_user(uaddr, kaddr, copy);
2202
2203 if (unlikely(left))
2204 return -EFAULT;
2205
2206 iov_iter_advance(&ii, copy);
2207 todo -= copy;
2208 kaddr += copy;
2209 }
2210
0bd87182 2211 kunmap(page);
59efec7b
TH		 2212 }
2213
2214 return 0;
2215}
2216
d9d318d3
MS		 2217/*
2218 * CUSE servers compiled on 32bit broke on 64bit kernels because the
2219 * ABI was defined to be 'struct iovec' which is different on 32bit
2220 * and 64bit. Fortunately we can determine which structure the server
2221 * used from the size of the reply.
2222 */
1baa26b2
MS		 2223static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
2224 size_t transferred, unsigned count,
2225 bool is_compat)
d9d318d3
MS		 2226{
2227#ifdef CONFIG_COMPAT
2228 if (count * sizeof(struct compat_iovec) == transferred) {
2229 struct compat_iovec *ciov = src;
2230 unsigned i;
2231
2232 /*
2233 * With this interface a 32bit server cannot support
2234 * non-compat (i.e. ones coming from 64bit apps) ioctl
2235 * requests
2236 */
2237 if (!is_compat)
2238 return -EINVAL;
2239
2240 for (i = 0; i < count; i++) {
2241 dst[i].iov_base = compat_ptr(ciov[i].iov_base);
2242 dst[i].iov_len = ciov[i].iov_len;
2243 }
2244 return 0;
2245 }
2246#endif
2247
2248 if (count * sizeof(struct iovec) != transferred)
2249 return -EIO;
2250
2251 memcpy(dst, src, transferred);
2252 return 0;
2253}
2254
7572777e
MS		 2255/* Make sure iov_length() won't overflow */
2256static int fuse_verify_ioctl_iov(struct iovec *iov, size_t count)
2257{
2258 size_t n;
2259 u32 max = FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT;
2260
fb6ccff6 2261 for (n = 0; n < count; n++, iov++) {
7572777e
MS		 2262 if (iov->iov_len > (size_t) max)
2263 return -ENOMEM;
2264 max -= iov->iov_len;
2265 }
2266 return 0;
2267}
2268
1baa26b2
MS		 2269static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
2270 void *src, size_t transferred, unsigned count,
2271 bool is_compat)
2272{
2273 unsigned i;
2274 struct fuse_ioctl_iovec *fiov = src;
2275
2276 if (fc->minor < 16) {
2277 return fuse_copy_ioctl_iovec_old(dst, src, transferred,
2278 count, is_compat);
2279 }
2280
2281 if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
2282 return -EIO;
2283
2284 for (i = 0; i < count; i++) {
2285 /* Did the server supply an inappropriate value? */
2286 if (fiov[i].base != (unsigned long) fiov[i].base ||
2287 fiov[i].len != (unsigned long) fiov[i].len)
2288 return -EIO;
2289
2290 dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
2291 dst[i].iov_len = (size_t) fiov[i].len;
2292
2293#ifdef CONFIG_COMPAT
2294 if (is_compat &&
2295 (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
2296 (compat_size_t) dst[i].iov_len != fiov[i].len))
2297 return -EIO;
2298#endif
2299 }
2300
2301 return 0;
2302}
2303
2304
59efec7b
TH		 2305/*
2306 * For ioctls, there is no generic way to determine how much memory
2307 * needs to be read and/or written. Furthermore, ioctls are allowed
2308 * to dereference the passed pointer, so the parameter requires deep
2309 * copying but FUSE has no idea whatsoever about what to copy in or
2310 * out.
2311 *
2312 * This is solved by allowing FUSE server to retry ioctl with
2313 * necessary in/out iovecs. Let's assume the ioctl implementation
2314 * needs to read in the following structure.
2315 *
2316 * struct a {
2317 * char *buf;
2318 * size_t buflen;
2319 * }
2320 *
2321 * On the first callout to FUSE server, inarg->in_size and
2322 * inarg->out_size will be NULL; then, the server completes the ioctl
2323 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
2324 * the actual iov array to
2325 *
2326 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } }
2327 *
2328 * which tells FUSE to copy in the requested area and retry the ioctl.
2329 * On the second round, the server has access to the structure and
2330 * from that it can tell what to look for next, so on the invocation,
2331 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
2332 *
2333 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) },
2334 * { .iov_base = a.buf, .iov_len = a.buflen } }
2335 *
2336 * FUSE will copy both struct a and the pointed buffer from the
2337 * process doing the ioctl and retry ioctl with both struct a and the
2338 * buffer.
2339 *
2340 * This time, FUSE server has everything it needs and completes ioctl
2341 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
2342 *
2343 * Copying data out works the same way.
2344 *
2345 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
2346 * automatically initializes in and out iovs by decoding @cmd with
2347 * _IOC_* macros and the server is not allowed to request RETRY. This
2348 * limits ioctl data transfers to well-formed ioctls and is the forced
2349 * behavior for all FUSE servers.
2350 */
08cbf542
TH		 2351long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
2352 unsigned int flags)
59efec7b 2353{
59efec7b 2354 struct fuse_file *ff = file->private_data;
d36f2487 2355 struct fuse_conn *fc = ff->fc;
59efec7b
TH		 2356 struct fuse_ioctl_in inarg = {
2357 .fh = ff->fh,
2358 .cmd = cmd,
2359 .arg = arg,
2360 .flags = flags
2361 };
2362 struct fuse_ioctl_out outarg;
2363 struct fuse_req *req = NULL;
2364 struct page **pages = NULL;
8ac83505 2365 struct iovec *iov_page = NULL;
59efec7b
TH		 2366 struct iovec *in_iov = NULL, *out_iov = NULL;
2367 unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages;
2368 size_t in_size, out_size, transferred;
2369 int err;
2370
1baa26b2
MS		 2371#if BITS_PER_LONG == 32
2372 inarg.flags |= FUSE_IOCTL_32BIT;
2373#else
2374 if (flags & FUSE_IOCTL_COMPAT)
2375 inarg.flags |= FUSE_IOCTL_32BIT;
2376#endif
2377
59efec7b 2378 /* assume all the iovs returned by client always fits in a page */
1baa26b2 2379 BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
59efec7b 2380
59efec7b 2381 err = -ENOMEM;
c411cc88 2382 pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, sizeof(pages[0]), GFP_KERNEL);
8ac83505 2383 iov_page = (struct iovec *) __get_free_page(GFP_KERNEL);
59efec7b
TH		 2384 if (!pages || !iov_page)
2385 goto out;
2386
2387 /*
2388 * If restricted, initialize IO parameters as encoded in @cmd.
2389 * RETRY from server is not allowed.
2390 */
2391 if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
8ac83505 2392 struct iovec *iov = iov_page;
59efec7b 2393
c9f0523d 2394 iov->iov_base = (void __user *)arg;
59efec7b
TH		 2395 iov->iov_len = _IOC_SIZE(cmd);
2396
2397 if (_IOC_DIR(cmd) & _IOC_WRITE) {
2398 in_iov = iov;
2399 in_iovs = 1;
2400 }
2401
2402 if (_IOC_DIR(cmd) & _IOC_READ) {
2403 out_iov = iov;
2404 out_iovs = 1;
2405 }
2406 }
2407
2408 retry:
2409 inarg.in_size = in_size = iov_length(in_iov, in_iovs);
2410 inarg.out_size = out_size = iov_length(out_iov, out_iovs);
2411
2412 /*
2413 * Out data can be used either for actual out data or iovs,
2414 * make sure there always is at least one page.
2415 */
2416 out_size = max_t(size_t, out_size, PAGE_SIZE);
2417 max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
2418
2419 /* make sure there are enough buffer pages and init request with them */
2420 err = -ENOMEM;
2421 if (max_pages > FUSE_MAX_PAGES_PER_REQ)
2422 goto out;
2423 while (num_pages < max_pages) {
2424 pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2425 if (!pages[num_pages])
2426 goto out;
2427 num_pages++;
2428 }
2429
54b96670 2430 req = fuse_get_req(fc, num_pages);
59efec7b
TH		 2431 if (IS_ERR(req)) {
2432 err = PTR_ERR(req);
2433 req = NULL;
2434 goto out;
2435 }
2436 memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages);
2437 req->num_pages = num_pages;
7c190c8b 2438 fuse_page_descs_length_init(req, 0, req->num_pages);
59efec7b
TH		 2439
2440 /* okay, let's send it to the client */
2441 req->in.h.opcode = FUSE_IOCTL;
d36f2487 2442 req->in.h.nodeid = ff->nodeid;
59efec7b
TH		 2443 req->in.numargs = 1;
2444 req->in.args[0].size = sizeof(inarg);
2445 req->in.args[0].value = &inarg;
2446 if (in_size) {
2447 req->in.numargs++;
2448 req->in.args[1].size = in_size;
2449 req->in.argpages = 1;
2450
2451 err = fuse_ioctl_copy_user(pages, in_iov, in_iovs, in_size,
2452 false);
2453 if (err)
2454 goto out;
2455 }
2456
2457 req->out.numargs = 2;
2458 req->out.args[0].size = sizeof(outarg);
2459 req->out.args[0].value = &outarg;
2460 req->out.args[1].size = out_size;
2461 req->out.argpages = 1;
2462 req->out.argvar = 1;
2463
b93f858a 2464 fuse_request_send(fc, req);
59efec7b
TH		 2465 err = req->out.h.error;
2466 transferred = req->out.args[1].size;
2467 fuse_put_request(fc, req);
2468 req = NULL;
2469 if (err)
2470 goto out;
2471
2472 /* did it ask for retry? */
2473 if (outarg.flags & FUSE_IOCTL_RETRY) {
8ac83505 2474 void *vaddr;
59efec7b
TH		 2475
2476 /* no retry if in restricted mode */
2477 err = -EIO;
2478 if (!(flags & FUSE_IOCTL_UNRESTRICTED))
2479 goto out;
2480
2481 in_iovs = outarg.in_iovs;
2482 out_iovs = outarg.out_iovs;
2483
2484 /*
2485 * Make sure things are in boundary, separate checks
2486 * are to protect against overflow.
2487 */
2488 err = -ENOMEM;
2489 if (in_iovs > FUSE_IOCTL_MAX_IOV ||
2490 out_iovs > FUSE_IOCTL_MAX_IOV ||
2491 in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
2492 goto out;
2493
2408f6ef 2494 vaddr = kmap_atomic(pages[0]);
1baa26b2 2495 err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr,
d9d318d3
MS		 2496 transferred, in_iovs + out_iovs,
2497 (flags & FUSE_IOCTL_COMPAT) != 0);
2408f6ef 2498 kunmap_atomic(vaddr);
d9d318d3
MS		 2499 if (err)
2500 goto out;
59efec7b 2501
8ac83505 2502 in_iov = iov_page;
59efec7b
TH		 2503 out_iov = in_iov + in_iovs;
2504
7572777e
MS		 2505 err = fuse_verify_ioctl_iov(in_iov, in_iovs);
2506 if (err)
2507 goto out;
2508
2509 err = fuse_verify_ioctl_iov(out_iov, out_iovs);
2510 if (err)
2511 goto out;
2512
59efec7b
TH		 2513 goto retry;
2514 }
2515
2516 err = -EIO;
2517 if (transferred > inarg.out_size)
2518 goto out;
2519
2520 err = fuse_ioctl_copy_user(pages, out_iov, out_iovs, transferred, true);
2521 out:
2522 if (req)
2523 fuse_put_request(fc, req);
8ac83505 2524 free_page((unsigned long) iov_page);
59efec7b
TH		 2525 while (num_pages)
2526 __free_page(pages[--num_pages]);
2527 kfree(pages);
2528
2529 return err ? err : outarg.result;
2530}
08cbf542 2531EXPORT_SYMBOL_GPL(fuse_do_ioctl);
59efec7b 2532
b18da0c5
MS		 2533long fuse_ioctl_common(struct file *file, unsigned int cmd,
2534 unsigned long arg, unsigned int flags)
d36f2487 2535{
6131ffaa 2536 struct inode *inode = file_inode(file);
d36f2487
MS		 2537 struct fuse_conn *fc = get_fuse_conn(inode);
2538
c2132c1b 2539 if (!fuse_allow_current_process(fc))
d36f2487
MS		 2540 return -EACCES;
2541
2542 if (is_bad_inode(inode))
2543 return -EIO;
2544
2545 return fuse_do_ioctl(file, cmd, arg, flags);
2546}
2547
59efec7b
TH		 2548static long fuse_file_ioctl(struct file *file, unsigned int cmd,
2549 unsigned long arg)
2550{
b18da0c5 2551 return fuse_ioctl_common(file, cmd, arg, 0);
59efec7b
TH		 2552}
2553
2554static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
2555 unsigned long arg)
2556{
b18da0c5 2557 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
59efec7b
TH		 2558}
2559
95668a69
TH		 2560/*
2561 * All files which have been polled are linked to RB tree
2562 * fuse_conn->polled_files which is indexed by kh. Walk the tree and
2563 * find the matching one.
2564 */
2565static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh,
2566 struct rb_node **parent_out)
2567{
2568 struct rb_node **link = &fc->polled_files.rb_node;
2569 struct rb_node *last = NULL;
2570
2571 while (*link) {
2572 struct fuse_file *ff;
2573
2574 last = *link;
2575 ff = rb_entry(last, struct fuse_file, polled_node);
2576
2577 if (kh < ff->kh)
2578 link = &last->rb_left;
2579 else if (kh > ff->kh)
2580 link = &last->rb_right;
2581 else
2582 return link;
2583 }
2584
2585 if (parent_out)
2586 *parent_out = last;
2587 return link;
2588}
2589
2590/*
2591 * The file is about to be polled. Make sure it's on the polled_files
2592 * RB tree. Note that files once added to the polled_files tree are
2593 * not removed before the file is released. This is because a file
2594 * polled once is likely to be polled again.
2595 */
2596static void fuse_register_polled_file(struct fuse_conn *fc,
2597 struct fuse_file *ff)
2598{
2599 spin_lock(&fc->lock);
2600 if (RB_EMPTY_NODE(&ff->polled_node)) {
2601 struct rb_node **link, *parent;
2602
2603 link = fuse_find_polled_node(fc, ff->kh, &parent);
2604 BUG_ON(*link);
2605 rb_link_node(&ff->polled_node, parent, link);
2606 rb_insert_color(&ff->polled_node, &fc->polled_files);
2607 }
2608 spin_unlock(&fc->lock);
2609}
2610
08cbf542 2611unsigned fuse_file_poll(struct file *file, poll_table *wait)
95668a69 2612{
95668a69 2613 struct fuse_file *ff = file->private_data;
797759aa 2614 struct fuse_conn *fc = ff->fc;
95668a69
TH		 2615 struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh };
2616 struct fuse_poll_out outarg;
2617 struct fuse_req *req;
2618 int err;
2619
2620 if (fc->no_poll)
2621 return DEFAULT_POLLMASK;
2622
2623 poll_wait(file, &ff->poll_wait, wait);
0415d291 2624 inarg.events = (__u32)poll_requested_events(wait);
95668a69
TH		 2625
2626 /*
2627 * Ask for notification iff there's someone waiting for it.
2628 * The client may ignore the flag and always notify.
2629 */
2630 if (waitqueue_active(&ff->poll_wait)) {
2631 inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY;
2632 fuse_register_polled_file(fc, ff);
2633 }
2634
b111c8c0 2635 req = fuse_get_req_nopages(fc);
95668a69 2636 if (IS_ERR(req))
201fa69a 2637 return POLLERR;
95668a69
TH		 2638
2639 req->in.h.opcode = FUSE_POLL;
797759aa 2640 req->in.h.nodeid = ff->nodeid;
95668a69
TH		 2641 req->in.numargs = 1;
2642 req->in.args[0].size = sizeof(inarg);
2643 req->in.args[0].value = &inarg;
2644 req->out.numargs = 1;
2645 req->out.args[0].size = sizeof(outarg);
2646 req->out.args[0].value = &outarg;
b93f858a 2647 fuse_request_send(fc, req);
95668a69
TH		 2648 err = req->out.h.error;
2649 fuse_put_request(fc, req);
2650
2651 if (!err)
2652 return outarg.revents;
2653 if (err == -ENOSYS) {
2654 fc->no_poll = 1;
2655 return DEFAULT_POLLMASK;
2656 }
2657 return POLLERR;
2658}
08cbf542 2659EXPORT_SYMBOL_GPL(fuse_file_poll);
95668a69
TH		 2660
2661/*
2662 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
2663 * wakes up the poll waiters.
2664 */
2665int fuse_notify_poll_wakeup(struct fuse_conn *fc,
2666 struct fuse_notify_poll_wakeup_out *outarg)
2667{
2668 u64 kh = outarg->kh;
2669 struct rb_node **link;
2670
2671 spin_lock(&fc->lock);
2672
2673 link = fuse_find_polled_node(fc, kh, NULL);
2674 if (*link) {
2675 struct fuse_file *ff;
2676
2677 ff = rb_entry(*link, struct fuse_file, polled_node);
2678 wake_up_interruptible_sync(&ff->poll_wait);
2679 }
2680
2681 spin_unlock(&fc->lock);
2682 return 0;
2683}
2684
efb9fa9e
MP		 2685static void fuse_do_truncate(struct file *file)
2686{
2687 struct inode *inode = file->f_mapping->host;
2688 struct iattr attr;
2689
2690 attr.ia_valid = ATTR_SIZE;
2691 attr.ia_size = i_size_read(inode);
2692
2693 attr.ia_file = file;
2694 attr.ia_valid |= ATTR_FILE;
2695
2696 fuse_do_setattr(inode, &attr, file);
2697}
2698
e5c5f05d
MP		 2699static inline loff_t fuse_round_up(loff_t off)
2700{
2701 return round_up(off, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
2702}
2703
4273b793
AA		 2704static ssize_t
2705fuse_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2706 loff_t offset, unsigned long nr_segs)
2707{
2708 ssize_t ret = 0;
60b9df7a
MS		 2709 struct file *file = iocb->ki_filp;
2710 struct fuse_file *ff = file->private_data;
e5c5f05d 2711 bool async_dio = ff->fc->async_dio;
4273b793 2712 loff_t pos = 0;
bcba24cc
MP		 2713 struct inode *inode;
2714 loff_t i_size;
2715 size_t count = iov_length(iov, nr_segs);
36cf66ed 2716 struct fuse_io_priv *io;
d3005b09 2717 bool is_sync = is_sync_kiocb(iocb);
4273b793 2718
4273b793 2719 pos = offset;
bcba24cc
MP		 2720 inode = file->f_mapping->host;
2721 i_size = i_size_read(inode);
4273b793 2722
439ee5f0 2723 /* optimization for short read */
e5c5f05d 2724 if (async_dio && rw != WRITE && offset + count > i_size) {
439ee5f0
MP		 2725 if (offset >= i_size)
2726 return 0;
e5c5f05d 2727 count = min_t(loff_t, count, fuse_round_up(i_size - offset));
439ee5f0
MP		 2728 }
2729
bcba24cc 2730 io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL);
36cf66ed
MP		 2731 if (!io)
2732 return -ENOMEM;
bcba24cc
MP		 2733 spin_lock_init(&io->lock);
2734 io->reqs = 1;
2735 io->bytes = -1;
2736 io->size = 0;
2737 io->offset = offset;
2738 io->write = (rw == WRITE);
2739 io->err = 0;
36cf66ed 2740 io->file = file;
bcba24cc
MP		 2741 /*
2742 * By default, we want to optimize all I/Os with async request
60b9df7a 2743 * submission to the client filesystem if supported.
bcba24cc 2744 */
e5c5f05d 2745 io->async = async_dio;
bcba24cc
MP		 2746 io->iocb = iocb;
2747
2748 /*
2749 * We cannot asynchronously extend the size of a file. We have no method
2750 * to wait on real async I/O requests, so we must submit this request
2751 * synchronously.
2752 */
d3005b09 2753 if (!is_sync && (offset + count > i_size) && rw == WRITE)
bcba24cc 2754 io->async = false;
4273b793 2755
b98d023a 2756 if (rw == WRITE)
36cf66ed 2757 ret = __fuse_direct_write(io, iov, nr_segs, &pos);
b98d023a 2758 else
439ee5f0 2759 ret = __fuse_direct_read(io, iov, nr_segs, &pos, count);
36cf66ed 2760
bcba24cc
MP		 2761 if (io->async) {
2762 fuse_aio_complete(io, ret < 0 ? ret : 0, -1);
2763
2764 /* we have a non-extending, async request, so return */
d3005b09 2765 if (!is_sync)
bcba24cc
MP		 2766 return -EIOCBQUEUED;
2767
2768 ret = wait_on_sync_kiocb(iocb);
2769 } else {
2770 kfree(io);
2771 }
2772
efb9fa9e
MP		 2773 if (rw == WRITE) {
2774 if (ret > 0)
2775 fuse_write_update_size(inode, pos);
2776 else if (ret < 0 && offset + count > i_size)
2777 fuse_do_truncate(file);
2778 }
4273b793
AA		 2779
2780 return ret;
2781}
2782
cdadb11c
MS		 2783static long fuse_file_fallocate(struct file *file, int mode, loff_t offset,
2784 loff_t length)
05ba1f08
AP		 2785{
2786 struct fuse_file *ff = file->private_data;
3634a632 2787 struct inode *inode = file->f_inode;
9abd30b4 2788 struct fuse_inode *fi = get_fuse_inode(inode);
05ba1f08
AP		 2789 struct fuse_conn *fc = ff->fc;
2790 struct fuse_req *req;
2791 struct fuse_fallocate_in inarg = {
2792 .fh = ff->fh,
2793 .offset = offset,
2794 .length = length,
2795 .mode = mode
2796 };
2797 int err;
14c14414
MP		 2798 bool lock_inode = !(mode & FALLOC_FL_KEEP_SIZE) ||
2799 (mode & FALLOC_FL_PUNCH_HOLE);
05ba1f08 2800
519c6040
MS		 2801 if (fc->no_fallocate)
2802 return -EOPNOTSUPP;
2803
14c14414 2804 if (lock_inode) {
3634a632 2805 mutex_lock(&inode->i_mutex);
eb12ca30
MP		 2806 if (mode & FALLOC_FL_PUNCH_HOLE) {
2807 loff_t endbyte = offset + length - 1;
2808 err = filemap_write_and_wait_range(inode->i_mapping,
2809 offset, endbyte);
2810 if (err)
2811 goto out;
2812
2813 fuse_sync_writes(inode);
2814 }
3634a632
BF		 2815 }
2816
9abd30b4
MP		 2817 if (!(mode & FALLOC_FL_KEEP_SIZE))
2818 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
2819
b111c8c0 2820 req = fuse_get_req_nopages(fc);
3634a632
BF		 2821 if (IS_ERR(req)) {
2822 err = PTR_ERR(req);
2823 goto out;
2824 }
05ba1f08
AP		 2825
2826 req->in.h.opcode = FUSE_FALLOCATE;
2827 req->in.h.nodeid = ff->nodeid;
2828 req->in.numargs = 1;
2829 req->in.args[0].size = sizeof(inarg);
2830 req->in.args[0].value = &inarg;
2831 fuse_request_send(fc, req);
2832 err = req->out.h.error;
519c6040
MS		 2833 if (err == -ENOSYS) {
2834 fc->no_fallocate = 1;
2835 err = -EOPNOTSUPP;
2836 }
05ba1f08
AP		 2837 fuse_put_request(fc, req);
2838
bee6c307
BF		 2839 if (err)
2840 goto out;
2841
2842 /* we could have extended the file */
2843 if (!(mode & FALLOC_FL_KEEP_SIZE))
2844 fuse_write_update_size(inode, offset + length);
2845
2846 if (mode & FALLOC_FL_PUNCH_HOLE)
2847 truncate_pagecache_range(inode, offset, offset + length - 1);
2848
2849 fuse_invalidate_attr(inode);
2850
3634a632 2851out:
9abd30b4
MP		 2852 if (!(mode & FALLOC_FL_KEEP_SIZE))
2853 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
2854
eb12ca30 2855 if (lock_inode)
3634a632 2856 mutex_unlock(&inode->i_mutex);
3634a632 2857
05ba1f08
AP		 2858 return err;
2859}
05ba1f08 2860
4b6f5d20 2861static const struct file_operations fuse_file_operations = {
5559b8f4 2862 .llseek = fuse_file_llseek,
543ade1f 2863 .read = do_sync_read,
bcb4be80 2864 .aio_read = fuse_file_aio_read,
543ade1f 2865 .write = do_sync_write,
ea9b9907 2866 .aio_write = fuse_file_aio_write,
b6aeaded
MS		 2867 .mmap = fuse_file_mmap,
2868 .open = fuse_open,
2869 .flush = fuse_flush,
2870 .release = fuse_release,
2871 .fsync = fuse_fsync,
71421259 2872 .lock = fuse_file_lock,
a9ff4f87 2873 .flock = fuse_file_flock,
5ffc4ef4 2874 .splice_read = generic_file_splice_read,
59efec7b
TH		 2875 .unlocked_ioctl = fuse_file_ioctl,
2876 .compat_ioctl = fuse_file_compat_ioctl,
95668a69 2877 .poll = fuse_file_poll,
05ba1f08 2878 .fallocate = fuse_file_fallocate,
b6aeaded
MS		 2879};
2880
4b6f5d20 2881static const struct file_operations fuse_direct_io_file_operations = {
5559b8f4 2882 .llseek = fuse_file_llseek,
413ef8cb
MS		 2883 .read = fuse_direct_read,
2884 .write = fuse_direct_write,
fc280c96 2885 .mmap = fuse_direct_mmap,
413ef8cb
MS		 2886 .open = fuse_open,
2887 .flush = fuse_flush,
2888 .release = fuse_release,
2889 .fsync = fuse_fsync,
71421259 2890 .lock = fuse_file_lock,
a9ff4f87 2891 .flock = fuse_file_flock,
59efec7b
TH		 2892 .unlocked_ioctl = fuse_file_ioctl,
2893 .compat_ioctl = fuse_file_compat_ioctl,
95668a69 2894 .poll = fuse_file_poll,
05ba1f08 2895 .fallocate = fuse_file_fallocate,
fc280c96 2896 /* no splice_read */
413ef8cb
MS		 2897};
2898
f5e54d6e 2899static const struct address_space_operations fuse_file_aops = {
b6aeaded 2900 .readpage = fuse_readpage,
3be5a52b
MS		 2901 .writepage = fuse_writepage,
2902 .launder_page = fuse_launder_page,
db50b96c 2903 .readpages = fuse_readpages,
3be5a52b 2904 .set_page_dirty = __set_page_dirty_nobuffers,
b2d2272f 2905 .bmap = fuse_bmap,
4273b793 2906 .direct_IO = fuse_direct_IO,
b6aeaded
MS		 2907};
2908
2909void fuse_init_file_inode(struct inode *inode)
2910{
45323fb7
MS		 2911 inode->i_fop = &fuse_file_operations;
2912 inode->i_data.a_ops = &fuse_file_aops;
b6aeaded 2913}
kernel source for telekom puls (alcatel/mediatek)