fuse: optimize fuse_get_user_pages()
[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"
10
11#include <linux/pagemap.h>
12#include <linux/slab.h>
13#include <linux/kernel.h>
e8edc6e0 14#include <linux/sched.h>
08cbf542 15#include <linux/module.h>
d9d318d3 16#include <linux/compat.h>
478e0841 17#include <linux/swap.h>
b6aeaded 18
4b6f5d20 19static const struct file_operations fuse_direct_io_file_operations;
45323fb7 20
91fe96b4
MS
21static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
22 int opcode, struct fuse_open_out *outargp)
b6aeaded 23{
b6aeaded 24 struct fuse_open_in inarg;
fd72faac
MS
25 struct fuse_req *req;
26 int err;
27
b111c8c0 28 req = fuse_get_req_nopages(fc);
ce1d5a49
MS
29 if (IS_ERR(req))
30 return PTR_ERR(req);
fd72faac
MS
31
32 memset(&inarg, 0, sizeof(inarg));
6ff958ed
MS
33 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY);
34 if (!fc->atomic_o_trunc)
35 inarg.flags &= ~O_TRUNC;
91fe96b4
MS
36 req->in.h.opcode = opcode;
37 req->in.h.nodeid = nodeid;
fd72faac
MS
38 req->in.numargs = 1;
39 req->in.args[0].size = sizeof(inarg);
40 req->in.args[0].value = &inarg;
41 req->out.numargs = 1;
42 req->out.args[0].size = sizeof(*outargp);
43 req->out.args[0].value = outargp;
b93f858a 44 fuse_request_send(fc, req);
fd72faac
MS
45 err = req->out.h.error;
46 fuse_put_request(fc, req);
47
48 return err;
49}
50
acf99433 51struct fuse_file *fuse_file_alloc(struct fuse_conn *fc)
fd72faac
MS
52{
53 struct fuse_file *ff;
6b2db28a 54
fd72faac 55 ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
6b2db28a
TH
56 if (unlikely(!ff))
57 return NULL;
58
da5e4714 59 ff->fc = fc;
4250c066 60 ff->reserved_req = fuse_request_alloc(0);
6b2db28a
TH
61 if (unlikely(!ff->reserved_req)) {
62 kfree(ff);
63 return NULL;
fd72faac 64 }
6b2db28a
TH
65
66 INIT_LIST_HEAD(&ff->write_entry);
67 atomic_set(&ff->count, 0);
68 RB_CLEAR_NODE(&ff->polled_node);
69 init_waitqueue_head(&ff->poll_wait);
70
71 spin_lock(&fc->lock);
72 ff->kh = ++fc->khctr;
73 spin_unlock(&fc->lock);
74
fd72faac
MS
75 return ff;
76}
77
78void fuse_file_free(struct fuse_file *ff)
79{
33649c91 80 fuse_request_free(ff->reserved_req);
fd72faac
MS
81 kfree(ff);
82}
83
c7b7143c 84struct fuse_file *fuse_file_get(struct fuse_file *ff)
c756e0a4
MS
85{
86 atomic_inc(&ff->count);
87 return ff;
88}
89
5a18ec17
MS
90static void fuse_release_async(struct work_struct *work)
91{
92 struct fuse_req *req;
93 struct fuse_conn *fc;
94 struct path path;
95
96 req = container_of(work, struct fuse_req, misc.release.work);
97 path = req->misc.release.path;
98 fc = get_fuse_conn(path.dentry->d_inode);
99
100 fuse_put_request(fc, req);
101 path_put(&path);
102}
103
819c4b3b
MS
104static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
105{
5a18ec17
MS
106 if (fc->destroy_req) {
107 /*
108 * If this is a fuseblk mount, then it's possible that
109 * releasing the path will result in releasing the
110 * super block and sending the DESTROY request. If
111 * the server is single threaded, this would hang.
112 * For this reason do the path_put() in a separate
113 * thread.
114 */
115 atomic_inc(&req->count);
116 INIT_WORK(&req->misc.release.work, fuse_release_async);
117 schedule_work(&req->misc.release.work);
118 } else {
119 path_put(&req->misc.release.path);
120 }
819c4b3b
MS
121}
122
5a18ec17 123static void fuse_file_put(struct fuse_file *ff, bool sync)
c756e0a4
MS
124{
125 if (atomic_dec_and_test(&ff->count)) {
126 struct fuse_req *req = ff->reserved_req;
8b0797a4 127
5a18ec17
MS
128 if (sync) {
129 fuse_request_send(ff->fc, req);
130 path_put(&req->misc.release.path);
131 fuse_put_request(ff->fc, req);
132 } else {
133 req->end = fuse_release_end;
134 fuse_request_send_background(ff->fc, req);
135 }
c756e0a4
MS
136 kfree(ff);
137 }
138}
139
08cbf542
TH
140int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
141 bool isdir)
91fe96b4
MS
142{
143 struct fuse_open_out outarg;
144 struct fuse_file *ff;
145 int err;
146 int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
147
148 ff = fuse_file_alloc(fc);
149 if (!ff)
150 return -ENOMEM;
151
152 err = fuse_send_open(fc, nodeid, file, opcode, &outarg);
153 if (err) {
154 fuse_file_free(ff);
155 return err;
156 }
157
158 if (isdir)
159 outarg.open_flags &= ~FOPEN_DIRECT_IO;
160
161 ff->fh = outarg.fh;
162 ff->nodeid = nodeid;
163 ff->open_flags = outarg.open_flags;
164 file->private_data = fuse_file_get(ff);
165
166 return 0;
167}
08cbf542 168EXPORT_SYMBOL_GPL(fuse_do_open);
91fe96b4 169
c7b7143c 170void fuse_finish_open(struct inode *inode, struct file *file)
fd72faac 171{
c7b7143c 172 struct fuse_file *ff = file->private_data;
a0822c55 173 struct fuse_conn *fc = get_fuse_conn(inode);
c7b7143c
MS
174
175 if (ff->open_flags & FOPEN_DIRECT_IO)
fd72faac 176 file->f_op = &fuse_direct_io_file_operations;
c7b7143c 177 if (!(ff->open_flags & FOPEN_KEEP_CACHE))
b1009979 178 invalidate_inode_pages2(inode->i_mapping);
c7b7143c 179 if (ff->open_flags & FOPEN_NONSEEKABLE)
a7c1b990 180 nonseekable_open(inode, file);
a0822c55
KS
181 if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) {
182 struct fuse_inode *fi = get_fuse_inode(inode);
183
184 spin_lock(&fc->lock);
185 fi->attr_version = ++fc->attr_version;
186 i_size_write(inode, 0);
187 spin_unlock(&fc->lock);
188 fuse_invalidate_attr(inode);
189 }
fd72faac
MS
190}
191
91fe96b4 192int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
fd72faac 193{
acf99433 194 struct fuse_conn *fc = get_fuse_conn(inode);
b6aeaded 195 int err;
b6aeaded
MS
196
197 err = generic_file_open(inode, file);
198 if (err)
199 return err;
200
91fe96b4 201 err = fuse_do_open(fc, get_node_id(inode), file, isdir);
fd72faac 202 if (err)
91fe96b4 203 return err;
b6aeaded 204
91fe96b4
MS
205 fuse_finish_open(inode, file);
206
207 return 0;
b6aeaded
MS
208}
209
8b0797a4 210static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode)
64c6d8ed 211{
8b0797a4 212 struct fuse_conn *fc = ff->fc;
33649c91 213 struct fuse_req *req = ff->reserved_req;
b57d4264 214 struct fuse_release_in *inarg = &req->misc.release.in;
b6aeaded 215
8b0797a4
MS
216 spin_lock(&fc->lock);
217 list_del(&ff->write_entry);
218 if (!RB_EMPTY_NODE(&ff->polled_node))
219 rb_erase(&ff->polled_node, &fc->polled_files);
220 spin_unlock(&fc->lock);
221
357ccf2b 222 wake_up_interruptible_all(&ff->poll_wait);
8b0797a4 223
b6aeaded 224 inarg->fh = ff->fh;
fd72faac 225 inarg->flags = flags;
51eb01e7 226 req->in.h.opcode = opcode;
c7b7143c 227 req->in.h.nodeid = ff->nodeid;
b6aeaded
MS
228 req->in.numargs = 1;
229 req->in.args[0].size = sizeof(struct fuse_release_in);
230 req->in.args[0].value = inarg;
fd72faac
MS
231}
232
8b0797a4 233void fuse_release_common(struct file *file, int opcode)
fd72faac 234{
6b2db28a
TH
235 struct fuse_file *ff;
236 struct fuse_req *req;
b6aeaded 237
6b2db28a
TH
238 ff = file->private_data;
239 if (unlikely(!ff))
8b0797a4 240 return;
6b2db28a 241
6b2db28a 242 req = ff->reserved_req;
8b0797a4 243 fuse_prepare_release(ff, file->f_flags, opcode);
6b2db28a 244
37fb3a30
MS
245 if (ff->flock) {
246 struct fuse_release_in *inarg = &req->misc.release.in;
247 inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK;
248 inarg->lock_owner = fuse_lock_owner_id(ff->fc,
249 (fl_owner_t) file);
250 }
6b2db28a 251 /* Hold vfsmount and dentry until release is finished */
b0be46eb
MS
252 path_get(&file->f_path);
253 req->misc.release.path = file->f_path;
6b2db28a 254
6b2db28a
TH
255 /*
256 * Normally this will send the RELEASE request, however if
257 * some asynchronous READ or WRITE requests are outstanding,
258 * the sending will be delayed.
5a18ec17
MS
259 *
260 * Make the release synchronous if this is a fuseblk mount,
261 * synchronous RELEASE is allowed (and desirable) in this case
262 * because the server can be trusted not to screw up.
6b2db28a 263 */
5a18ec17 264 fuse_file_put(ff, ff->fc->destroy_req != NULL);
b6aeaded
MS
265}
266
04730fef
MS
267static int fuse_open(struct inode *inode, struct file *file)
268{
91fe96b4 269 return fuse_open_common(inode, file, false);
04730fef
MS
270}
271
272static int fuse_release(struct inode *inode, struct file *file)
273{
8b0797a4
MS
274 fuse_release_common(file, FUSE_RELEASE);
275
276 /* return value is ignored by VFS */
277 return 0;
278}
279
280void fuse_sync_release(struct fuse_file *ff, int flags)
281{
282 WARN_ON(atomic_read(&ff->count) > 1);
283 fuse_prepare_release(ff, flags, FUSE_RELEASE);
284 ff->reserved_req->force = 1;
285 fuse_request_send(ff->fc, ff->reserved_req);
286 fuse_put_request(ff->fc, ff->reserved_req);
287 kfree(ff);
04730fef 288}
08cbf542 289EXPORT_SYMBOL_GPL(fuse_sync_release);
04730fef 290
71421259 291/*
9c8ef561
MS
292 * Scramble the ID space with XTEA, so that the value of the files_struct
293 * pointer is not exposed to userspace.
71421259 294 */
f3332114 295u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
71421259 296{
9c8ef561
MS
297 u32 *k = fc->scramble_key;
298 u64 v = (unsigned long) id;
299 u32 v0 = v;
300 u32 v1 = v >> 32;
301 u32 sum = 0;
302 int i;
303
304 for (i = 0; i < 32; i++) {
305 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
306 sum += 0x9E3779B9;
307 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
308 }
309
310 return (u64) v0 + ((u64) v1 << 32);
71421259
MS
311}
312
3be5a52b
MS
313/*
314 * Check if page is under writeback
315 *
316 * This is currently done by walking the list of writepage requests
317 * for the inode, which can be pretty inefficient.
318 */
319static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
320{
321 struct fuse_conn *fc = get_fuse_conn(inode);
322 struct fuse_inode *fi = get_fuse_inode(inode);
323 struct fuse_req *req;
324 bool found = false;
325
326 spin_lock(&fc->lock);
327 list_for_each_entry(req, &fi->writepages, writepages_entry) {
328 pgoff_t curr_index;
329
330 BUG_ON(req->inode != inode);
331 curr_index = req->misc.write.in.offset >> PAGE_CACHE_SHIFT;
332 if (curr_index == index) {
333 found = true;
334 break;
335 }
336 }
337 spin_unlock(&fc->lock);
338
339 return found;
340}
341
342/*
343 * Wait for page writeback to be completed.
344 *
345 * Since fuse doesn't rely on the VM writeback tracking, this has to
346 * use some other means.
347 */
348static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
349{
350 struct fuse_inode *fi = get_fuse_inode(inode);
351
352 wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
353 return 0;
354}
355
75e1fcc0 356static int fuse_flush(struct file *file, fl_owner_t id)
b6aeaded 357{
7706a9d6 358 struct inode *inode = file->f_path.dentry->d_inode;
b6aeaded
MS
359 struct fuse_conn *fc = get_fuse_conn(inode);
360 struct fuse_file *ff = file->private_data;
361 struct fuse_req *req;
362 struct fuse_flush_in inarg;
363 int err;
364
248d86e8
MS
365 if (is_bad_inode(inode))
366 return -EIO;
367
b6aeaded
MS
368 if (fc->no_flush)
369 return 0;
370
b111c8c0 371 req = fuse_get_req_nofail_nopages(fc, file);
b6aeaded
MS
372 memset(&inarg, 0, sizeof(inarg));
373 inarg.fh = ff->fh;
9c8ef561 374 inarg.lock_owner = fuse_lock_owner_id(fc, id);
b6aeaded
MS
375 req->in.h.opcode = FUSE_FLUSH;
376 req->in.h.nodeid = get_node_id(inode);
b6aeaded
MS
377 req->in.numargs = 1;
378 req->in.args[0].size = sizeof(inarg);
379 req->in.args[0].value = &inarg;
71421259 380 req->force = 1;
b93f858a 381 fuse_request_send(fc, req);
b6aeaded
MS
382 err = req->out.h.error;
383 fuse_put_request(fc, req);
384 if (err == -ENOSYS) {
385 fc->no_flush = 1;
386 err = 0;
387 }
388 return err;
389}
390
3be5a52b
MS
391/*
392 * Wait for all pending writepages on the inode to finish.
393 *
394 * This is currently done by blocking further writes with FUSE_NOWRITE
395 * and waiting for all sent writes to complete.
396 *
397 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
398 * could conflict with truncation.
399 */
400static void fuse_sync_writes(struct inode *inode)
401{
402 fuse_set_nowrite(inode);
403 fuse_release_nowrite(inode);
404}
405
02c24a82
JB
406int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
407 int datasync, int isdir)
b6aeaded 408{
7ea80859 409 struct inode *inode = file->f_mapping->host;
b6aeaded
MS
410 struct fuse_conn *fc = get_fuse_conn(inode);
411 struct fuse_file *ff = file->private_data;
412 struct fuse_req *req;
413 struct fuse_fsync_in inarg;
414 int err;
415
248d86e8
MS
416 if (is_bad_inode(inode))
417 return -EIO;
418
02c24a82
JB
419 err = filemap_write_and_wait_range(inode->i_mapping, start, end);
420 if (err)
421 return err;
422
82547981 423 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
b6aeaded
MS
424 return 0;
425
02c24a82
JB
426 mutex_lock(&inode->i_mutex);
427
3be5a52b
MS
428 /*
429 * Start writeback against all dirty pages of the inode, then
430 * wait for all outstanding writes, before sending the FSYNC
431 * request.
432 */
433 err = write_inode_now(inode, 0);
434 if (err)
02c24a82 435 goto out;
3be5a52b
MS
436
437 fuse_sync_writes(inode);
438
b111c8c0 439 req = fuse_get_req_nopages(fc);
02c24a82
JB
440 if (IS_ERR(req)) {
441 err = PTR_ERR(req);
442 goto out;
443 }
b6aeaded
MS
444
445 memset(&inarg, 0, sizeof(inarg));
446 inarg.fh = ff->fh;
447 inarg.fsync_flags = datasync ? 1 : 0;
82547981 448 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
b6aeaded 449 req->in.h.nodeid = get_node_id(inode);
b6aeaded
MS
450 req->in.numargs = 1;
451 req->in.args[0].size = sizeof(inarg);
452 req->in.args[0].value = &inarg;
b93f858a 453 fuse_request_send(fc, req);
b6aeaded
MS
454 err = req->out.h.error;
455 fuse_put_request(fc, req);
456 if (err == -ENOSYS) {
82547981
MS
457 if (isdir)
458 fc->no_fsyncdir = 1;
459 else
460 fc->no_fsync = 1;
b6aeaded
MS
461 err = 0;
462 }
02c24a82
JB
463out:
464 mutex_unlock(&inode->i_mutex);
b6aeaded
MS
465 return err;
466}
467
02c24a82
JB
468static int fuse_fsync(struct file *file, loff_t start, loff_t end,
469 int datasync)
82547981 470{
02c24a82 471 return fuse_fsync_common(file, start, end, datasync, 0);
82547981
MS
472}
473
2106cb18
MS
474void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos,
475 size_t count, int opcode)
b6aeaded 476{
5c5c5e51 477 struct fuse_read_in *inarg = &req->misc.read.in;
a6643094 478 struct fuse_file *ff = file->private_data;
b6aeaded 479
361b1eb5
MS
480 inarg->fh = ff->fh;
481 inarg->offset = pos;
482 inarg->size = count;
a6643094 483 inarg->flags = file->f_flags;
361b1eb5 484 req->in.h.opcode = opcode;
2106cb18 485 req->in.h.nodeid = ff->nodeid;
b6aeaded
MS
486 req->in.numargs = 1;
487 req->in.args[0].size = sizeof(struct fuse_read_in);
c1aa96a5 488 req->in.args[0].value = inarg;
b6aeaded
MS
489 req->out.argvar = 1;
490 req->out.numargs = 1;
491 req->out.args[0].size = count;
b6aeaded
MS
492}
493
8bfc016d 494static size_t fuse_send_read(struct fuse_req *req, struct file *file,
2106cb18 495 loff_t pos, size_t count, fl_owner_t owner)
04730fef 496{
2106cb18
MS
497 struct fuse_file *ff = file->private_data;
498 struct fuse_conn *fc = ff->fc;
f3332114 499
2106cb18 500 fuse_read_fill(req, file, pos, count, FUSE_READ);
f3332114 501 if (owner != NULL) {
5c5c5e51 502 struct fuse_read_in *inarg = &req->misc.read.in;
f3332114
MS
503
504 inarg->read_flags |= FUSE_READ_LOCKOWNER;
505 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
506 }
b93f858a 507 fuse_request_send(fc, req);
361b1eb5 508 return req->out.args[0].size;
04730fef
MS
509}
510
5c5c5e51
MS
511static void fuse_read_update_size(struct inode *inode, loff_t size,
512 u64 attr_ver)
513{
514 struct fuse_conn *fc = get_fuse_conn(inode);
515 struct fuse_inode *fi = get_fuse_inode(inode);
516
517 spin_lock(&fc->lock);
518 if (attr_ver == fi->attr_version && size < inode->i_size) {
519 fi->attr_version = ++fc->attr_version;
520 i_size_write(inode, size);
521 }
522 spin_unlock(&fc->lock);
523}
524
b6aeaded
MS
525static int fuse_readpage(struct file *file, struct page *page)
526{
527 struct inode *inode = page->mapping->host;
528 struct fuse_conn *fc = get_fuse_conn(inode);
248d86e8 529 struct fuse_req *req;
5c5c5e51
MS
530 size_t num_read;
531 loff_t pos = page_offset(page);
532 size_t count = PAGE_CACHE_SIZE;
533 u64 attr_ver;
248d86e8
MS
534 int err;
535
536 err = -EIO;
537 if (is_bad_inode(inode))
538 goto out;
539
3be5a52b 540 /*
25985edc 541 * Page writeback can extend beyond the lifetime of the
3be5a52b
MS
542 * page-cache page, so make sure we read a properly synced
543 * page.
544 */
545 fuse_wait_on_page_writeback(inode, page->index);
546
b111c8c0 547 req = fuse_get_req(fc, 1);
ce1d5a49
MS
548 err = PTR_ERR(req);
549 if (IS_ERR(req))
b6aeaded
MS
550 goto out;
551
5c5c5e51
MS
552 attr_ver = fuse_get_attr_version(fc);
553
b6aeaded 554 req->out.page_zeroing = 1;
f4975c67 555 req->out.argpages = 1;
b6aeaded
MS
556 req->num_pages = 1;
557 req->pages[0] = page;
85f40aec 558 req->page_descs[0].length = count;
2106cb18 559 num_read = fuse_send_read(req, file, pos, count, NULL);
b6aeaded
MS
560 err = req->out.h.error;
561 fuse_put_request(fc, req);
5c5c5e51
MS
562
563 if (!err) {
564 /*
565 * Short read means EOF. If file size is larger, truncate it
566 */
567 if (num_read < count)
568 fuse_read_update_size(inode, pos + num_read, attr_ver);
569
b6aeaded 570 SetPageUptodate(page);
5c5c5e51
MS
571 }
572
b36c31ba 573 fuse_invalidate_attr(inode); /* atime changed */
b6aeaded
MS
574 out:
575 unlock_page(page);
576 return err;
577}
578
c1aa96a5 579static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
db50b96c 580{
c1aa96a5 581 int i;
5c5c5e51
MS
582 size_t count = req->misc.read.in.size;
583 size_t num_read = req->out.args[0].size;
ce534fb0 584 struct address_space *mapping = NULL;
c1aa96a5 585
ce534fb0
MS
586 for (i = 0; mapping == NULL && i < req->num_pages; i++)
587 mapping = req->pages[i]->mapping;
5c5c5e51 588
ce534fb0
MS
589 if (mapping) {
590 struct inode *inode = mapping->host;
591
592 /*
593 * Short read means EOF. If file size is larger, truncate it
594 */
595 if (!req->out.h.error && num_read < count) {
596 loff_t pos;
597
598 pos = page_offset(req->pages[0]) + num_read;
599 fuse_read_update_size(inode, pos,
600 req->misc.read.attr_ver);
601 }
602 fuse_invalidate_attr(inode); /* atime changed */
603 }
c1aa96a5 604
db50b96c
MS
605 for (i = 0; i < req->num_pages; i++) {
606 struct page *page = req->pages[i];
607 if (!req->out.h.error)
608 SetPageUptodate(page);
c1aa96a5
MS
609 else
610 SetPageError(page);
db50b96c 611 unlock_page(page);
b5dd3285 612 page_cache_release(page);
db50b96c 613 }
c756e0a4 614 if (req->ff)
5a18ec17 615 fuse_file_put(req->ff, false);
c1aa96a5
MS
616}
617
2106cb18 618static void fuse_send_readpages(struct fuse_req *req, struct file *file)
c1aa96a5 619{
2106cb18
MS
620 struct fuse_file *ff = file->private_data;
621 struct fuse_conn *fc = ff->fc;
c1aa96a5
MS
622 loff_t pos = page_offset(req->pages[0]);
623 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
f4975c67
MS
624
625 req->out.argpages = 1;
c1aa96a5 626 req->out.page_zeroing = 1;
ce534fb0 627 req->out.page_replace = 1;
2106cb18 628 fuse_read_fill(req, file, pos, count, FUSE_READ);
5c5c5e51 629 req->misc.read.attr_ver = fuse_get_attr_version(fc);
9cd68455 630 if (fc->async_read) {
c756e0a4 631 req->ff = fuse_file_get(ff);
9cd68455 632 req->end = fuse_readpages_end;
b93f858a 633 fuse_request_send_background(fc, req);
9cd68455 634 } else {
b93f858a 635 fuse_request_send(fc, req);
9cd68455 636 fuse_readpages_end(fc, req);
e9bb09dd 637 fuse_put_request(fc, req);
9cd68455 638 }
db50b96c
MS
639}
640
c756e0a4 641struct fuse_fill_data {
db50b96c 642 struct fuse_req *req;
a6643094 643 struct file *file;
db50b96c 644 struct inode *inode;
f8dbdf81 645 unsigned nr_pages;
db50b96c
MS
646};
647
648static int fuse_readpages_fill(void *_data, struct page *page)
649{
c756e0a4 650 struct fuse_fill_data *data = _data;
db50b96c
MS
651 struct fuse_req *req = data->req;
652 struct inode *inode = data->inode;
653 struct fuse_conn *fc = get_fuse_conn(inode);
654
3be5a52b
MS
655 fuse_wait_on_page_writeback(inode, page->index);
656
db50b96c
MS
657 if (req->num_pages &&
658 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
659 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
660 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
f8dbdf81
MP
661 int nr_alloc = min_t(unsigned, data->nr_pages,
662 FUSE_MAX_PAGES_PER_REQ);
2106cb18 663 fuse_send_readpages(req, data->file);
f8dbdf81 664 data->req = req = fuse_get_req(fc, nr_alloc);
ce1d5a49 665 if (IS_ERR(req)) {
db50b96c 666 unlock_page(page);
ce1d5a49 667 return PTR_ERR(req);
db50b96c 668 }
db50b96c 669 }
f8dbdf81
MP
670
671 if (WARN_ON(req->num_pages >= req->max_pages)) {
672 fuse_put_request(fc, req);
673 return -EIO;
674 }
675
b5dd3285 676 page_cache_get(page);
db50b96c 677 req->pages[req->num_pages] = page;
85f40aec 678 req->page_descs[req->num_pages].length = PAGE_SIZE;
1729a16c 679 req->num_pages++;
f8dbdf81 680 data->nr_pages--;
db50b96c
MS
681 return 0;
682}
683
684static int fuse_readpages(struct file *file, struct address_space *mapping,
685 struct list_head *pages, unsigned nr_pages)
686{
687 struct inode *inode = mapping->host;
688 struct fuse_conn *fc = get_fuse_conn(inode);
c756e0a4 689 struct fuse_fill_data data;
db50b96c 690 int err;
f8dbdf81 691 int nr_alloc = min_t(unsigned, nr_pages, FUSE_MAX_PAGES_PER_REQ);
248d86e8 692
1d7ea732 693 err = -EIO;
248d86e8 694 if (is_bad_inode(inode))
2e990021 695 goto out;
248d86e8 696
a6643094 697 data.file = file;
db50b96c 698 data.inode = inode;
f8dbdf81
MP
699 data.req = fuse_get_req(fc, nr_alloc);
700 data.nr_pages = nr_pages;
1d7ea732 701 err = PTR_ERR(data.req);
ce1d5a49 702 if (IS_ERR(data.req))
2e990021 703 goto out;
db50b96c
MS
704
705 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
d3406ffa
MS
706 if (!err) {
707 if (data.req->num_pages)
2106cb18 708 fuse_send_readpages(data.req, file);
d3406ffa
MS
709 else
710 fuse_put_request(fc, data.req);
711 }
2e990021 712out:
1d7ea732 713 return err;
db50b96c
MS
714}
715
bcb4be80
MS
716static ssize_t fuse_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
717 unsigned long nr_segs, loff_t pos)
718{
719 struct inode *inode = iocb->ki_filp->f_mapping->host;
a8894274 720 struct fuse_conn *fc = get_fuse_conn(inode);
bcb4be80 721
a8894274
BF
722 /*
723 * In auto invalidate mode, always update attributes on read.
724 * Otherwise, only update if we attempt to read past EOF (to ensure
725 * i_size is up to date).
726 */
727 if (fc->auto_inval_data ||
728 (pos + iov_length(iov, nr_segs) > i_size_read(inode))) {
bcb4be80 729 int err;
bcb4be80
MS
730 err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL);
731 if (err)
732 return err;
733 }
734
735 return generic_file_aio_read(iocb, iov, nr_segs, pos);
736}
737
2d698b07 738static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff,
2106cb18 739 loff_t pos, size_t count)
b6aeaded 740{
b25e82e5
MS
741 struct fuse_write_in *inarg = &req->misc.write.in;
742 struct fuse_write_out *outarg = &req->misc.write.out;
b6aeaded 743
b25e82e5
MS
744 inarg->fh = ff->fh;
745 inarg->offset = pos;
746 inarg->size = count;
b6aeaded 747 req->in.h.opcode = FUSE_WRITE;
2106cb18 748 req->in.h.nodeid = ff->nodeid;
b6aeaded 749 req->in.numargs = 2;
2106cb18 750 if (ff->fc->minor < 9)
f3332114
MS
751 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
752 else
753 req->in.args[0].size = sizeof(struct fuse_write_in);
b25e82e5 754 req->in.args[0].value = inarg;
b6aeaded
MS
755 req->in.args[1].size = count;
756 req->out.numargs = 1;
757 req->out.args[0].size = sizeof(struct fuse_write_out);
b25e82e5
MS
758 req->out.args[0].value = outarg;
759}
760
761static size_t fuse_send_write(struct fuse_req *req, struct file *file,
2106cb18 762 loff_t pos, size_t count, fl_owner_t owner)
b25e82e5 763{
2106cb18
MS
764 struct fuse_file *ff = file->private_data;
765 struct fuse_conn *fc = ff->fc;
2d698b07
MS
766 struct fuse_write_in *inarg = &req->misc.write.in;
767
2106cb18 768 fuse_write_fill(req, ff, pos, count);
2d698b07 769 inarg->flags = file->f_flags;
f3332114 770 if (owner != NULL) {
f3332114
MS
771 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
772 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
773 }
b93f858a 774 fuse_request_send(fc, req);
b25e82e5 775 return req->misc.write.out.size;
b6aeaded
MS
776}
777
a1d75f25 778void fuse_write_update_size(struct inode *inode, loff_t pos)
854512ec
MS
779{
780 struct fuse_conn *fc = get_fuse_conn(inode);
781 struct fuse_inode *fi = get_fuse_inode(inode);
782
783 spin_lock(&fc->lock);
784 fi->attr_version = ++fc->attr_version;
785 if (pos > inode->i_size)
786 i_size_write(inode, pos);
787 spin_unlock(&fc->lock);
788}
789
ea9b9907
NP
790static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file,
791 struct inode *inode, loff_t pos,
792 size_t count)
793{
794 size_t res;
795 unsigned offset;
796 unsigned i;
797
798 for (i = 0; i < req->num_pages; i++)
799 fuse_wait_on_page_writeback(inode, req->pages[i]->index);
800
2106cb18 801 res = fuse_send_write(req, file, pos, count, NULL);
ea9b9907 802
b2430d75 803 offset = req->page_descs[0].offset;
ea9b9907
NP
804 count = res;
805 for (i = 0; i < req->num_pages; i++) {
806 struct page *page = req->pages[i];
807
808 if (!req->out.h.error && !offset && count >= PAGE_CACHE_SIZE)
809 SetPageUptodate(page);
810
811 if (count > PAGE_CACHE_SIZE - offset)
812 count -= PAGE_CACHE_SIZE - offset;
813 else
814 count = 0;
815 offset = 0;
816
817 unlock_page(page);
818 page_cache_release(page);
819 }
820
821 return res;
822}
823
824static ssize_t fuse_fill_write_pages(struct fuse_req *req,
825 struct address_space *mapping,
826 struct iov_iter *ii, loff_t pos)
827{
828 struct fuse_conn *fc = get_fuse_conn(mapping->host);
829 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
830 size_t count = 0;
831 int err;
832
f4975c67 833 req->in.argpages = 1;
b2430d75 834 req->page_descs[0].offset = offset;
ea9b9907
NP
835
836 do {
837 size_t tmp;
838 struct page *page;
839 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
840 size_t bytes = min_t(size_t, PAGE_CACHE_SIZE - offset,
841 iov_iter_count(ii));
842
843 bytes = min_t(size_t, bytes, fc->max_write - count);
844
845 again:
846 err = -EFAULT;
847 if (iov_iter_fault_in_readable(ii, bytes))
848 break;
849
850 err = -ENOMEM;
54566b2c 851 page = grab_cache_page_write_begin(mapping, index, 0);
ea9b9907
NP
852 if (!page)
853 break;
854
931e80e4 855 if (mapping_writably_mapped(mapping))
856 flush_dcache_page(page);
857
ea9b9907
NP
858 pagefault_disable();
859 tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes);
860 pagefault_enable();
861 flush_dcache_page(page);
862
478e0841
JW
863 mark_page_accessed(page);
864
ea9b9907
NP
865 if (!tmp) {
866 unlock_page(page);
867 page_cache_release(page);
868 bytes = min(bytes, iov_iter_single_seg_count(ii));
869 goto again;
870 }
871
872 err = 0;
873 req->pages[req->num_pages] = page;
85f40aec 874 req->page_descs[req->num_pages].length = tmp;
ea9b9907
NP
875 req->num_pages++;
876
877 iov_iter_advance(ii, tmp);
878 count += tmp;
879 pos += tmp;
880 offset += tmp;
881 if (offset == PAGE_CACHE_SIZE)
882 offset = 0;
883
78bb6cb9
MS
884 if (!fc->big_writes)
885 break;
ea9b9907 886 } while (iov_iter_count(ii) && count < fc->max_write &&
d07f09f5 887 req->num_pages < req->max_pages && offset == 0);
ea9b9907
NP
888
889 return count > 0 ? count : err;
890}
891
d07f09f5
MP
892static inline unsigned fuse_wr_pages(loff_t pos, size_t len)
893{
894 return min_t(unsigned,
895 ((pos + len - 1) >> PAGE_CACHE_SHIFT) -
896 (pos >> PAGE_CACHE_SHIFT) + 1,
897 FUSE_MAX_PAGES_PER_REQ);
898}
899
ea9b9907
NP
900static ssize_t fuse_perform_write(struct file *file,
901 struct address_space *mapping,
902 struct iov_iter *ii, loff_t pos)
903{
904 struct inode *inode = mapping->host;
905 struct fuse_conn *fc = get_fuse_conn(inode);
906 int err = 0;
907 ssize_t res = 0;
908
909 if (is_bad_inode(inode))
910 return -EIO;
911
912 do {
913 struct fuse_req *req;
914 ssize_t count;
d07f09f5 915 unsigned nr_pages = fuse_wr_pages(pos, iov_iter_count(ii));
ea9b9907 916
d07f09f5 917 req = fuse_get_req(fc, nr_pages);
ea9b9907
NP
918 if (IS_ERR(req)) {
919 err = PTR_ERR(req);
920 break;
921 }
922
923 count = fuse_fill_write_pages(req, mapping, ii, pos);
924 if (count <= 0) {
925 err = count;
926 } else {
927 size_t num_written;
928
929 num_written = fuse_send_write_pages(req, file, inode,
930 pos, count);
931 err = req->out.h.error;
932 if (!err) {
933 res += num_written;
934 pos += num_written;
935
936 /* break out of the loop on short write */
937 if (num_written != count)
938 err = -EIO;
939 }
940 }
941 fuse_put_request(fc, req);
942 } while (!err && iov_iter_count(ii));
943
944 if (res > 0)
945 fuse_write_update_size(inode, pos);
946
947 fuse_invalidate_attr(inode);
948
949 return res > 0 ? res : err;
950}
951
952static ssize_t fuse_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
953 unsigned long nr_segs, loff_t pos)
954{
955 struct file *file = iocb->ki_filp;
956 struct address_space *mapping = file->f_mapping;
957 size_t count = 0;
4273b793 958 size_t ocount = 0;
ea9b9907 959 ssize_t written = 0;
4273b793 960 ssize_t written_buffered = 0;
ea9b9907
NP
961 struct inode *inode = mapping->host;
962 ssize_t err;
963 struct iov_iter i;
4273b793 964 loff_t endbyte = 0;
ea9b9907
NP
965
966 WARN_ON(iocb->ki_pos != pos);
967
4273b793
AA
968 ocount = 0;
969 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
ea9b9907
NP
970 if (err)
971 return err;
972
4273b793 973 count = ocount;
58ef6a75 974 sb_start_write(inode->i_sb);
ea9b9907 975 mutex_lock(&inode->i_mutex);
ea9b9907
NP
976
977 /* We can write back this queue in page reclaim */
978 current->backing_dev_info = mapping->backing_dev_info;
979
980 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
981 if (err)
982 goto out;
983
984 if (count == 0)
985 goto out;
986
2f1936b8 987 err = file_remove_suid(file);
ea9b9907
NP
988 if (err)
989 goto out;
990
c3b2da31
JB
991 err = file_update_time(file);
992 if (err)
993 goto out;
ea9b9907 994
4273b793
AA
995 if (file->f_flags & O_DIRECT) {
996 written = generic_file_direct_write(iocb, iov, &nr_segs,
997 pos, &iocb->ki_pos,
998 count, ocount);
999 if (written < 0 || written == count)
1000 goto out;
1001
1002 pos += written;
1003 count -= written;
ea9b9907 1004
4273b793
AA
1005 iov_iter_init(&i, iov, nr_segs, count, written);
1006 written_buffered = fuse_perform_write(file, mapping, &i, pos);
1007 if (written_buffered < 0) {
1008 err = written_buffered;
1009 goto out;
1010 }
1011 endbyte = pos + written_buffered - 1;
1012
1013 err = filemap_write_and_wait_range(file->f_mapping, pos,
1014 endbyte);
1015 if (err)
1016 goto out;
1017
1018 invalidate_mapping_pages(file->f_mapping,
1019 pos >> PAGE_CACHE_SHIFT,
1020 endbyte >> PAGE_CACHE_SHIFT);
1021
1022 written += written_buffered;
1023 iocb->ki_pos = pos + written_buffered;
1024 } else {
1025 iov_iter_init(&i, iov, nr_segs, count, 0);
1026 written = fuse_perform_write(file, mapping, &i, pos);
1027 if (written >= 0)
1028 iocb->ki_pos = pos + written;
1029 }
ea9b9907
NP
1030out:
1031 current->backing_dev_info = NULL;
1032 mutex_unlock(&inode->i_mutex);
58ef6a75 1033 sb_end_write(inode->i_sb);
ea9b9907
NP
1034
1035 return written ? written : err;
1036}
1037
413ef8cb
MS
1038static void fuse_release_user_pages(struct fuse_req *req, int write)
1039{
1040 unsigned i;
1041
1042 for (i = 0; i < req->num_pages; i++) {
1043 struct page *page = req->pages[i];
1044 if (write)
1045 set_page_dirty_lock(page);
1046 put_page(page);
1047 }
1048}
1049
7c190c8b
MP
1050static inline void fuse_page_descs_length_init(struct fuse_req *req,
1051 unsigned index, unsigned nr_pages)
85f40aec
MP
1052{
1053 int i;
1054
7c190c8b 1055 for (i = index; i < index + nr_pages; i++)
85f40aec
MP
1056 req->page_descs[i].length = PAGE_SIZE -
1057 req->page_descs[i].offset;
1058}
1059
7c190c8b
MP
1060static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii)
1061{
1062 return (unsigned long)ii->iov->iov_base + ii->iov_offset;
1063}
1064
1065static inline size_t fuse_get_frag_size(const struct iov_iter *ii,
1066 size_t max_size)
1067{
1068 return min(iov_iter_single_seg_count(ii), max_size);
1069}
1070
b98d023a 1071static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii,
ce60a2f1 1072 size_t *nbytesp, int write)
413ef8cb 1073{
7c190c8b 1074 size_t nbytes = 0; /* # bytes already packed in req */
b98d023a 1075
f4975c67
MS
1076 /* Special case for kernel I/O: can copy directly into the buffer */
1077 if (segment_eq(get_fs(), KERNEL_DS)) {
7c190c8b
MP
1078 unsigned long user_addr = fuse_get_user_addr(ii);
1079 size_t frag_size = fuse_get_frag_size(ii, *nbytesp);
1080
f4975c67
MS
1081 if (write)
1082 req->in.args[1].value = (void *) user_addr;
1083 else
1084 req->out.args[0].value = (void *) user_addr;
1085
b98d023a
MP
1086 iov_iter_advance(ii, frag_size);
1087 *nbytesp = frag_size;
f4975c67
MS
1088 return 0;
1089 }
413ef8cb 1090
7c190c8b
MP
1091 while (nbytes < *nbytesp && req->num_pages < FUSE_MAX_PAGES_PER_REQ) {
1092 unsigned npages;
1093 unsigned long user_addr = fuse_get_user_addr(ii);
1094 unsigned offset = user_addr & ~PAGE_MASK;
1095 size_t frag_size = fuse_get_frag_size(ii, *nbytesp - nbytes);
1096 int ret;
413ef8cb 1097
7c190c8b
MP
1098 unsigned n = FUSE_MAX_PAGES_PER_REQ - req->num_pages;
1099 frag_size = min_t(size_t, frag_size, n << PAGE_SHIFT);
1100
1101 npages = (frag_size + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1102 npages = clamp(npages, 1U, n);
1103
1104 ret = get_user_pages_fast(user_addr, npages, !write,
1105 &req->pages[req->num_pages]);
1106 if (ret < 0)
1107 return ret;
1108
1109 npages = ret;
1110 frag_size = min_t(size_t, frag_size,
1111 (npages << PAGE_SHIFT) - offset);
1112 iov_iter_advance(ii, frag_size);
1113
1114 req->page_descs[req->num_pages].offset = offset;
1115 fuse_page_descs_length_init(req, req->num_pages, npages);
1116
1117 req->num_pages += npages;
1118 req->page_descs[req->num_pages - 1].length -=
1119 (npages << PAGE_SHIFT) - offset - frag_size;
1120
1121 nbytes += frag_size;
1122 }
f4975c67
MS
1123
1124 if (write)
1125 req->in.argpages = 1;
1126 else
1127 req->out.argpages = 1;
1128
7c190c8b 1129 *nbytesp = nbytes;
f4975c67 1130
413ef8cb
MS
1131 return 0;
1132}
1133
b98d023a
MP
1134static ssize_t __fuse_direct_io(struct file *file, const struct iovec *iov,
1135 unsigned long nr_segs, size_t count,
1136 loff_t *ppos, int write)
413ef8cb 1137{
2106cb18
MS
1138 struct fuse_file *ff = file->private_data;
1139 struct fuse_conn *fc = ff->fc;
413ef8cb
MS
1140 size_t nmax = write ? fc->max_write : fc->max_read;
1141 loff_t pos = *ppos;
1142 ssize_t res = 0;
248d86e8 1143 struct fuse_req *req;
b98d023a
MP
1144 struct iov_iter ii;
1145
1146 iov_iter_init(&ii, iov, nr_segs, count, 0);
248d86e8 1147
b111c8c0 1148 req = fuse_get_req(fc, FUSE_MAX_PAGES_PER_REQ);
ce1d5a49
MS
1149 if (IS_ERR(req))
1150 return PTR_ERR(req);
413ef8cb
MS
1151
1152 while (count) {
413ef8cb 1153 size_t nres;
2106cb18 1154 fl_owner_t owner = current->files;
f4975c67 1155 size_t nbytes = min(count, nmax);
b98d023a 1156 int err = fuse_get_user_pages(req, &ii, &nbytes, write);
413ef8cb
MS
1157 if (err) {
1158 res = err;
1159 break;
1160 }
f4975c67 1161
413ef8cb 1162 if (write)
2106cb18 1163 nres = fuse_send_write(req, file, pos, nbytes, owner);
413ef8cb 1164 else
2106cb18
MS
1165 nres = fuse_send_read(req, file, pos, nbytes, owner);
1166
413ef8cb
MS
1167 fuse_release_user_pages(req, !write);
1168 if (req->out.h.error) {
1169 if (!res)
1170 res = req->out.h.error;
1171 break;
1172 } else if (nres > nbytes) {
1173 res = -EIO;
1174 break;
1175 }
1176 count -= nres;
1177 res += nres;
1178 pos += nres;
413ef8cb
MS
1179 if (nres != nbytes)
1180 break;
56cf34ff
MS
1181 if (count) {
1182 fuse_put_request(fc, req);
b111c8c0 1183 req = fuse_get_req(fc, FUSE_MAX_PAGES_PER_REQ);
56cf34ff
MS
1184 if (IS_ERR(req))
1185 break;
1186 }
413ef8cb 1187 }
f60311d5
AA
1188 if (!IS_ERR(req))
1189 fuse_put_request(fc, req);
d09cb9d7 1190 if (res > 0)
413ef8cb 1191 *ppos = pos;
413ef8cb
MS
1192
1193 return res;
1194}
b98d023a
MP
1195
1196ssize_t fuse_direct_io(struct file *file, const char __user *buf,
1197 size_t count, loff_t *ppos, int write)
1198{
1199 struct iovec iov = { .iov_base = (void *)buf, .iov_len = count };
1200 return __fuse_direct_io(file, &iov, 1, count, ppos, write);
1201}
08cbf542 1202EXPORT_SYMBOL_GPL(fuse_direct_io);
413ef8cb 1203
b98d023a
MP
1204static ssize_t __fuse_direct_read(struct file *file, const struct iovec *iov,
1205 unsigned long nr_segs, loff_t *ppos)
413ef8cb 1206{
d09cb9d7
MS
1207 ssize_t res;
1208 struct inode *inode = file->f_path.dentry->d_inode;
1209
1210 if (is_bad_inode(inode))
1211 return -EIO;
1212
b98d023a
MP
1213 res = __fuse_direct_io(file, iov, nr_segs, iov_length(iov, nr_segs),
1214 ppos, 0);
d09cb9d7
MS
1215
1216 fuse_invalidate_attr(inode);
1217
1218 return res;
413ef8cb
MS
1219}
1220
b98d023a
MP
1221static ssize_t fuse_direct_read(struct file *file, char __user *buf,
1222 size_t count, loff_t *ppos)
1223{
1224 struct iovec iov = { .iov_base = (void *)buf, .iov_len = count };
1225 return __fuse_direct_read(file, &iov, 1, ppos);
1226}
1227
1228static ssize_t __fuse_direct_write(struct file *file, const struct iovec *iov,
1229 unsigned long nr_segs, loff_t *ppos)
413ef8cb 1230{
7706a9d6 1231 struct inode *inode = file->f_path.dentry->d_inode;
b98d023a 1232 size_t count = iov_length(iov, nr_segs);
413ef8cb 1233 ssize_t res;
d09cb9d7 1234
889f7848 1235 res = generic_write_checks(file, ppos, &count, 0);
d09cb9d7 1236 if (!res) {
b98d023a 1237 res = __fuse_direct_io(file, iov, nr_segs, count, ppos, 1);
d09cb9d7
MS
1238 if (res > 0)
1239 fuse_write_update_size(inode, *ppos);
1240 }
d09cb9d7
MS
1241
1242 fuse_invalidate_attr(inode);
1243
413ef8cb
MS
1244 return res;
1245}
1246
4273b793
AA
1247static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
1248 size_t count, loff_t *ppos)
1249{
b98d023a 1250 struct iovec iov = { .iov_base = (void *)buf, .iov_len = count };
4273b793
AA
1251 struct inode *inode = file->f_path.dentry->d_inode;
1252 ssize_t res;
1253
1254 if (is_bad_inode(inode))
1255 return -EIO;
1256
1257 /* Don't allow parallel writes to the same file */
1258 mutex_lock(&inode->i_mutex);
b98d023a 1259 res = __fuse_direct_write(file, &iov, 1, ppos);
4273b793
AA
1260 mutex_unlock(&inode->i_mutex);
1261
1262 return res;
1263}
1264
3be5a52b 1265static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
b6aeaded 1266{
3be5a52b 1267 __free_page(req->pages[0]);
5a18ec17 1268 fuse_file_put(req->ff, false);
3be5a52b
MS
1269}
1270
1271static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
1272{
1273 struct inode *inode = req->inode;
1274 struct fuse_inode *fi = get_fuse_inode(inode);
1275 struct backing_dev_info *bdi = inode->i_mapping->backing_dev_info;
1276
1277 list_del(&req->writepages_entry);
1278 dec_bdi_stat(bdi, BDI_WRITEBACK);
1279 dec_zone_page_state(req->pages[0], NR_WRITEBACK_TEMP);
1280 bdi_writeout_inc(bdi);
1281 wake_up(&fi->page_waitq);
1282}
1283
1284/* Called under fc->lock, may release and reacquire it */
1285static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req)
b9ca67b2
MS
1286__releases(fc->lock)
1287__acquires(fc->lock)
3be5a52b
MS
1288{
1289 struct fuse_inode *fi = get_fuse_inode(req->inode);
1290 loff_t size = i_size_read(req->inode);
1291 struct fuse_write_in *inarg = &req->misc.write.in;
1292
1293 if (!fc->connected)
1294 goto out_free;
1295
1296 if (inarg->offset + PAGE_CACHE_SIZE <= size) {
1297 inarg->size = PAGE_CACHE_SIZE;
1298 } else if (inarg->offset < size) {
1299 inarg->size = size & (PAGE_CACHE_SIZE - 1);
1300 } else {
1301 /* Got truncated off completely */
1302 goto out_free;
b6aeaded 1303 }
3be5a52b
MS
1304
1305 req->in.args[1].size = inarg->size;
1306 fi->writectr++;
b93f858a 1307 fuse_request_send_background_locked(fc, req);
3be5a52b
MS
1308 return;
1309
1310 out_free:
1311 fuse_writepage_finish(fc, req);
1312 spin_unlock(&fc->lock);
1313 fuse_writepage_free(fc, req);
e9bb09dd 1314 fuse_put_request(fc, req);
3be5a52b 1315 spin_lock(&fc->lock);
b6aeaded
MS
1316}
1317
3be5a52b
MS
1318/*
1319 * If fi->writectr is positive (no truncate or fsync going on) send
1320 * all queued writepage requests.
1321 *
1322 * Called with fc->lock
1323 */
1324void fuse_flush_writepages(struct inode *inode)
b9ca67b2
MS
1325__releases(fc->lock)
1326__acquires(fc->lock)
b6aeaded 1327{
3be5a52b
MS
1328 struct fuse_conn *fc = get_fuse_conn(inode);
1329 struct fuse_inode *fi = get_fuse_inode(inode);
1330 struct fuse_req *req;
1331
1332 while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
1333 req = list_entry(fi->queued_writes.next, struct fuse_req, list);
1334 list_del_init(&req->list);
1335 fuse_send_writepage(fc, req);
1336 }
1337}
1338
1339static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
1340{
1341 struct inode *inode = req->inode;
1342 struct fuse_inode *fi = get_fuse_inode(inode);
1343
1344 mapping_set_error(inode->i_mapping, req->out.h.error);
1345 spin_lock(&fc->lock);
1346 fi->writectr--;
1347 fuse_writepage_finish(fc, req);
1348 spin_unlock(&fc->lock);
1349 fuse_writepage_free(fc, req);
1350}
1351
1352static int fuse_writepage_locked(struct page *page)
1353{
1354 struct address_space *mapping = page->mapping;
1355 struct inode *inode = mapping->host;
1356 struct fuse_conn *fc = get_fuse_conn(inode);
1357 struct fuse_inode *fi = get_fuse_inode(inode);
1358 struct fuse_req *req;
1359 struct fuse_file *ff;
1360 struct page *tmp_page;
1361
1362 set_page_writeback(page);
1363
4250c066 1364 req = fuse_request_alloc_nofs(1);
3be5a52b
MS
1365 if (!req)
1366 goto err;
1367
1368 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1369 if (!tmp_page)
1370 goto err_free;
1371
1372 spin_lock(&fc->lock);
1373 BUG_ON(list_empty(&fi->write_files));
1374 ff = list_entry(fi->write_files.next, struct fuse_file, write_entry);
1375 req->ff = fuse_file_get(ff);
1376 spin_unlock(&fc->lock);
1377
2106cb18 1378 fuse_write_fill(req, ff, page_offset(page), 0);
3be5a52b
MS
1379
1380 copy_highpage(tmp_page, page);
2d698b07 1381 req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
f4975c67 1382 req->in.argpages = 1;
3be5a52b
MS
1383 req->num_pages = 1;
1384 req->pages[0] = tmp_page;
b2430d75 1385 req->page_descs[0].offset = 0;
85f40aec 1386 req->page_descs[0].length = PAGE_SIZE;
3be5a52b
MS
1387 req->end = fuse_writepage_end;
1388 req->inode = inode;
1389
1390 inc_bdi_stat(mapping->backing_dev_info, BDI_WRITEBACK);
1391 inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP);
1392 end_page_writeback(page);
1393
1394 spin_lock(&fc->lock);
1395 list_add(&req->writepages_entry, &fi->writepages);
1396 list_add_tail(&req->list, &fi->queued_writes);
1397 fuse_flush_writepages(inode);
1398 spin_unlock(&fc->lock);
1399
1400 return 0;
1401
1402err_free:
1403 fuse_request_free(req);
1404err:
1405 end_page_writeback(page);
1406 return -ENOMEM;
1407}
1408
1409static int fuse_writepage(struct page *page, struct writeback_control *wbc)
1410{
1411 int err;
1412
1413 err = fuse_writepage_locked(page);
1414 unlock_page(page);
1415
1416 return err;
1417}
1418
1419static int fuse_launder_page(struct page *page)
1420{
1421 int err = 0;
1422 if (clear_page_dirty_for_io(page)) {
1423 struct inode *inode = page->mapping->host;
1424 err = fuse_writepage_locked(page);
1425 if (!err)
1426 fuse_wait_on_page_writeback(inode, page->index);
1427 }
1428 return err;
1429}
1430
1431/*
1432 * Write back dirty pages now, because there may not be any suitable
1433 * open files later
1434 */
1435static void fuse_vma_close(struct vm_area_struct *vma)
1436{
1437 filemap_write_and_wait(vma->vm_file->f_mapping);
1438}
1439
1440/*
1441 * Wait for writeback against this page to complete before allowing it
1442 * to be marked dirty again, and hence written back again, possibly
1443 * before the previous writepage completed.
1444 *
1445 * Block here, instead of in ->writepage(), so that the userspace fs
1446 * can only block processes actually operating on the filesystem.
1447 *
1448 * Otherwise unprivileged userspace fs would be able to block
1449 * unrelated:
1450 *
1451 * - page migration
1452 * - sync(2)
1453 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
1454 */
c2ec175c 1455static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
3be5a52b 1456{
c2ec175c 1457 struct page *page = vmf->page;
3be5a52b
MS
1458 /*
1459 * Don't use page->mapping as it may become NULL from a
1460 * concurrent truncate.
1461 */
1462 struct inode *inode = vma->vm_file->f_mapping->host;
1463
1464 fuse_wait_on_page_writeback(inode, page->index);
1465 return 0;
1466}
1467
f0f37e2f 1468static const struct vm_operations_struct fuse_file_vm_ops = {
3be5a52b
MS
1469 .close = fuse_vma_close,
1470 .fault = filemap_fault,
1471 .page_mkwrite = fuse_page_mkwrite,
0b173bc4 1472 .remap_pages = generic_file_remap_pages,
3be5a52b
MS
1473};
1474
1475static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
1476{
1477 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) {
1478 struct inode *inode = file->f_dentry->d_inode;
1479 struct fuse_conn *fc = get_fuse_conn(inode);
1480 struct fuse_inode *fi = get_fuse_inode(inode);
1481 struct fuse_file *ff = file->private_data;
1482 /*
1483 * file may be written through mmap, so chain it onto the
1484 * inodes's write_file list
1485 */
1486 spin_lock(&fc->lock);
1487 if (list_empty(&ff->write_entry))
1488 list_add(&ff->write_entry, &fi->write_files);
1489 spin_unlock(&fc->lock);
1490 }
1491 file_accessed(file);
1492 vma->vm_ops = &fuse_file_vm_ops;
b6aeaded
MS
1493 return 0;
1494}
1495
fc280c96
MS
1496static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma)
1497{
1498 /* Can't provide the coherency needed for MAP_SHARED */
1499 if (vma->vm_flags & VM_MAYSHARE)
1500 return -ENODEV;
1501
3121bfe7
MS
1502 invalidate_inode_pages2(file->f_mapping);
1503
fc280c96
MS
1504 return generic_file_mmap(file, vma);
1505}
1506
71421259
MS
1507static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
1508 struct file_lock *fl)
1509{
1510 switch (ffl->type) {
1511 case F_UNLCK:
1512 break;
1513
1514 case F_RDLCK:
1515 case F_WRLCK:
1516 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
1517 ffl->end < ffl->start)
1518 return -EIO;
1519
1520 fl->fl_start = ffl->start;
1521 fl->fl_end = ffl->end;
1522 fl->fl_pid = ffl->pid;
1523 break;
1524
1525 default:
1526 return -EIO;
1527 }
1528 fl->fl_type = ffl->type;
1529 return 0;
1530}
1531
1532static void fuse_lk_fill(struct fuse_req *req, struct file *file,
a9ff4f87
MS
1533 const struct file_lock *fl, int opcode, pid_t pid,
1534 int flock)
71421259 1535{
7706a9d6 1536 struct inode *inode = file->f_path.dentry->d_inode;
9c8ef561 1537 struct fuse_conn *fc = get_fuse_conn(inode);
71421259
MS
1538 struct fuse_file *ff = file->private_data;
1539 struct fuse_lk_in *arg = &req->misc.lk_in;
1540
1541 arg->fh = ff->fh;
9c8ef561 1542 arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
71421259
MS
1543 arg->lk.start = fl->fl_start;
1544 arg->lk.end = fl->fl_end;
1545 arg->lk.type = fl->fl_type;
1546 arg->lk.pid = pid;
a9ff4f87
MS
1547 if (flock)
1548 arg->lk_flags |= FUSE_LK_FLOCK;
71421259
MS
1549 req->in.h.opcode = opcode;
1550 req->in.h.nodeid = get_node_id(inode);
1551 req->in.numargs = 1;
1552 req->in.args[0].size = sizeof(*arg);
1553 req->in.args[0].value = arg;
1554}
1555
1556static int fuse_getlk(struct file *file, struct file_lock *fl)
1557{
7706a9d6 1558 struct inode *inode = file->f_path.dentry->d_inode;
71421259
MS
1559 struct fuse_conn *fc = get_fuse_conn(inode);
1560 struct fuse_req *req;
1561 struct fuse_lk_out outarg;
1562 int err;
1563
b111c8c0 1564 req = fuse_get_req_nopages(fc);
71421259
MS
1565 if (IS_ERR(req))
1566 return PTR_ERR(req);
1567
a9ff4f87 1568 fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0);
71421259
MS
1569 req->out.numargs = 1;
1570 req->out.args[0].size = sizeof(outarg);
1571 req->out.args[0].value = &outarg;
b93f858a 1572 fuse_request_send(fc, req);
71421259
MS
1573 err = req->out.h.error;
1574 fuse_put_request(fc, req);
1575 if (!err)
1576 err = convert_fuse_file_lock(&outarg.lk, fl);
1577
1578 return err;
1579}
1580
a9ff4f87 1581static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
71421259 1582{
7706a9d6 1583 struct inode *inode = file->f_path.dentry->d_inode;
71421259
MS
1584 struct fuse_conn *fc = get_fuse_conn(inode);
1585 struct fuse_req *req;
1586 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
1587 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
1588 int err;
1589
8fb47a4f 1590 if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
48e90761
MS
1591 /* NLM needs asynchronous locks, which we don't support yet */
1592 return -ENOLCK;
1593 }
1594
71421259
MS
1595 /* Unlock on close is handled by the flush method */
1596 if (fl->fl_flags & FL_CLOSE)
1597 return 0;
1598
b111c8c0 1599 req = fuse_get_req_nopages(fc);
71421259
MS
1600 if (IS_ERR(req))
1601 return PTR_ERR(req);
1602
a9ff4f87 1603 fuse_lk_fill(req, file, fl, opcode, pid, flock);
b93f858a 1604 fuse_request_send(fc, req);
71421259 1605 err = req->out.h.error;
a4d27e75
MS
1606 /* locking is restartable */
1607 if (err == -EINTR)
1608 err = -ERESTARTSYS;
71421259
MS
1609 fuse_put_request(fc, req);
1610 return err;
1611}
1612
1613static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
1614{
7706a9d6 1615 struct inode *inode = file->f_path.dentry->d_inode;
71421259
MS
1616 struct fuse_conn *fc = get_fuse_conn(inode);
1617 int err;
1618
48e90761
MS
1619 if (cmd == F_CANCELLK) {
1620 err = 0;
1621 } else if (cmd == F_GETLK) {
71421259 1622 if (fc->no_lock) {
9d6a8c5c 1623 posix_test_lock(file, fl);
71421259
MS
1624 err = 0;
1625 } else
1626 err = fuse_getlk(file, fl);
1627 } else {
1628 if (fc->no_lock)
48e90761 1629 err = posix_lock_file(file, fl, NULL);
71421259 1630 else
a9ff4f87 1631 err = fuse_setlk(file, fl, 0);
71421259
MS
1632 }
1633 return err;
1634}
1635
a9ff4f87
MS
1636static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
1637{
1638 struct inode *inode = file->f_path.dentry->d_inode;
1639 struct fuse_conn *fc = get_fuse_conn(inode);
1640 int err;
1641
37fb3a30 1642 if (fc->no_flock) {
a9ff4f87
MS
1643 err = flock_lock_file_wait(file, fl);
1644 } else {
37fb3a30
MS
1645 struct fuse_file *ff = file->private_data;
1646
a9ff4f87
MS
1647 /* emulate flock with POSIX locks */
1648 fl->fl_owner = (fl_owner_t) file;
37fb3a30 1649 ff->flock = true;
a9ff4f87
MS
1650 err = fuse_setlk(file, fl, 1);
1651 }
1652
1653 return err;
1654}
1655
b2d2272f
MS
1656static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
1657{
1658 struct inode *inode = mapping->host;
1659 struct fuse_conn *fc = get_fuse_conn(inode);
1660 struct fuse_req *req;
1661 struct fuse_bmap_in inarg;
1662 struct fuse_bmap_out outarg;
1663 int err;
1664
1665 if (!inode->i_sb->s_bdev || fc->no_bmap)
1666 return 0;
1667
b111c8c0 1668 req = fuse_get_req_nopages(fc);
b2d2272f
MS
1669 if (IS_ERR(req))
1670 return 0;
1671
1672 memset(&inarg, 0, sizeof(inarg));
1673 inarg.block = block;
1674 inarg.blocksize = inode->i_sb->s_blocksize;
1675 req->in.h.opcode = FUSE_BMAP;
1676 req->in.h.nodeid = get_node_id(inode);
1677 req->in.numargs = 1;
1678 req->in.args[0].size = sizeof(inarg);
1679 req->in.args[0].value = &inarg;
1680 req->out.numargs = 1;
1681 req->out.args[0].size = sizeof(outarg);
1682 req->out.args[0].value = &outarg;
b93f858a 1683 fuse_request_send(fc, req);
b2d2272f
MS
1684 err = req->out.h.error;
1685 fuse_put_request(fc, req);
1686 if (err == -ENOSYS)
1687 fc->no_bmap = 1;
1688
1689 return err ? 0 : outarg.block;
1690}
1691
965c8e59 1692static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence)
5559b8f4
MS
1693{
1694 loff_t retval;
1695 struct inode *inode = file->f_path.dentry->d_inode;
1696
c07c3d19 1697 /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */
965c8e59
AM
1698 if (whence == SEEK_CUR || whence == SEEK_SET)
1699 return generic_file_llseek(file, offset, whence);
06222e49 1700
c07c3d19
MS
1701 mutex_lock(&inode->i_mutex);
1702 retval = fuse_update_attributes(inode, NULL, file, NULL);
1703 if (!retval)
965c8e59 1704 retval = generic_file_llseek(file, offset, whence);
5559b8f4 1705 mutex_unlock(&inode->i_mutex);
c07c3d19 1706
5559b8f4
MS
1707 return retval;
1708}
1709
59efec7b
TH
1710static int fuse_ioctl_copy_user(struct page **pages, struct iovec *iov,
1711 unsigned int nr_segs, size_t bytes, bool to_user)
1712{
1713 struct iov_iter ii;
1714 int page_idx = 0;
1715
1716 if (!bytes)
1717 return 0;
1718
1719 iov_iter_init(&ii, iov, nr_segs, bytes, 0);
1720
1721 while (iov_iter_count(&ii)) {
1722 struct page *page = pages[page_idx++];
1723 size_t todo = min_t(size_t, PAGE_SIZE, iov_iter_count(&ii));
4aa0edd2 1724 void *kaddr;
59efec7b 1725
4aa0edd2 1726 kaddr = kmap(page);
59efec7b
TH
1727
1728 while (todo) {
1729 char __user *uaddr = ii.iov->iov_base + ii.iov_offset;
1730 size_t iov_len = ii.iov->iov_len - ii.iov_offset;
1731 size_t copy = min(todo, iov_len);
1732 size_t left;
1733
1734 if (!to_user)
1735 left = copy_from_user(kaddr, uaddr, copy);
1736 else
1737 left = copy_to_user(uaddr, kaddr, copy);
1738
1739 if (unlikely(left))
1740 return -EFAULT;
1741
1742 iov_iter_advance(&ii, copy);
1743 todo -= copy;
1744 kaddr += copy;
1745 }
1746
0bd87182 1747 kunmap(page);
59efec7b
TH
1748 }
1749
1750 return 0;
1751}
1752
d9d318d3
MS
1753/*
1754 * CUSE servers compiled on 32bit broke on 64bit kernels because the
1755 * ABI was defined to be 'struct iovec' which is different on 32bit
1756 * and 64bit. Fortunately we can determine which structure the server
1757 * used from the size of the reply.
1758 */
1baa26b2
MS
1759static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
1760 size_t transferred, unsigned count,
1761 bool is_compat)
d9d318d3
MS
1762{
1763#ifdef CONFIG_COMPAT
1764 if (count * sizeof(struct compat_iovec) == transferred) {
1765 struct compat_iovec *ciov = src;
1766 unsigned i;
1767
1768 /*
1769 * With this interface a 32bit server cannot support
1770 * non-compat (i.e. ones coming from 64bit apps) ioctl
1771 * requests
1772 */
1773 if (!is_compat)
1774 return -EINVAL;
1775
1776 for (i = 0; i < count; i++) {
1777 dst[i].iov_base = compat_ptr(ciov[i].iov_base);
1778 dst[i].iov_len = ciov[i].iov_len;
1779 }
1780 return 0;
1781 }
1782#endif
1783
1784 if (count * sizeof(struct iovec) != transferred)
1785 return -EIO;
1786
1787 memcpy(dst, src, transferred);
1788 return 0;
1789}
1790
7572777e
MS
1791/* Make sure iov_length() won't overflow */
1792static int fuse_verify_ioctl_iov(struct iovec *iov, size_t count)
1793{
1794 size_t n;
1795 u32 max = FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT;
1796
fb6ccff6 1797 for (n = 0; n < count; n++, iov++) {
7572777e
MS
1798 if (iov->iov_len > (size_t) max)
1799 return -ENOMEM;
1800 max -= iov->iov_len;
1801 }
1802 return 0;
1803}
1804
1baa26b2
MS
1805static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
1806 void *src, size_t transferred, unsigned count,
1807 bool is_compat)
1808{
1809 unsigned i;
1810 struct fuse_ioctl_iovec *fiov = src;
1811
1812 if (fc->minor < 16) {
1813 return fuse_copy_ioctl_iovec_old(dst, src, transferred,
1814 count, is_compat);
1815 }
1816
1817 if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
1818 return -EIO;
1819
1820 for (i = 0; i < count; i++) {
1821 /* Did the server supply an inappropriate value? */
1822 if (fiov[i].base != (unsigned long) fiov[i].base ||
1823 fiov[i].len != (unsigned long) fiov[i].len)
1824 return -EIO;
1825
1826 dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
1827 dst[i].iov_len = (size_t) fiov[i].len;
1828
1829#ifdef CONFIG_COMPAT
1830 if (is_compat &&
1831 (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
1832 (compat_size_t) dst[i].iov_len != fiov[i].len))
1833 return -EIO;
1834#endif
1835 }
1836
1837 return 0;
1838}
1839
1840
59efec7b
TH
1841/*
1842 * For ioctls, there is no generic way to determine how much memory
1843 * needs to be read and/or written. Furthermore, ioctls are allowed
1844 * to dereference the passed pointer, so the parameter requires deep
1845 * copying but FUSE has no idea whatsoever about what to copy in or
1846 * out.
1847 *
1848 * This is solved by allowing FUSE server to retry ioctl with
1849 * necessary in/out iovecs. Let's assume the ioctl implementation
1850 * needs to read in the following structure.
1851 *
1852 * struct a {
1853 * char *buf;
1854 * size_t buflen;
1855 * }
1856 *
1857 * On the first callout to FUSE server, inarg->in_size and
1858 * inarg->out_size will be NULL; then, the server completes the ioctl
1859 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
1860 * the actual iov array to
1861 *
1862 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } }
1863 *
1864 * which tells FUSE to copy in the requested area and retry the ioctl.
1865 * On the second round, the server has access to the structure and
1866 * from that it can tell what to look for next, so on the invocation,
1867 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
1868 *
1869 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) },
1870 * { .iov_base = a.buf, .iov_len = a.buflen } }
1871 *
1872 * FUSE will copy both struct a and the pointed buffer from the
1873 * process doing the ioctl and retry ioctl with both struct a and the
1874 * buffer.
1875 *
1876 * This time, FUSE server has everything it needs and completes ioctl
1877 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
1878 *
1879 * Copying data out works the same way.
1880 *
1881 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
1882 * automatically initializes in and out iovs by decoding @cmd with
1883 * _IOC_* macros and the server is not allowed to request RETRY. This
1884 * limits ioctl data transfers to well-formed ioctls and is the forced
1885 * behavior for all FUSE servers.
1886 */
08cbf542
TH
1887long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1888 unsigned int flags)
59efec7b 1889{
59efec7b 1890 struct fuse_file *ff = file->private_data;
d36f2487 1891 struct fuse_conn *fc = ff->fc;
59efec7b
TH
1892 struct fuse_ioctl_in inarg = {
1893 .fh = ff->fh,
1894 .cmd = cmd,
1895 .arg = arg,
1896 .flags = flags
1897 };
1898 struct fuse_ioctl_out outarg;
1899 struct fuse_req *req = NULL;
1900 struct page **pages = NULL;
8ac83505 1901 struct iovec *iov_page = NULL;
59efec7b
TH
1902 struct iovec *in_iov = NULL, *out_iov = NULL;
1903 unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages;
1904 size_t in_size, out_size, transferred;
1905 int err;
1906
1baa26b2
MS
1907#if BITS_PER_LONG == 32
1908 inarg.flags |= FUSE_IOCTL_32BIT;
1909#else
1910 if (flags & FUSE_IOCTL_COMPAT)
1911 inarg.flags |= FUSE_IOCTL_32BIT;
1912#endif
1913
59efec7b 1914 /* assume all the iovs returned by client always fits in a page */
1baa26b2 1915 BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
59efec7b 1916
59efec7b 1917 err = -ENOMEM;
c411cc88 1918 pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, sizeof(pages[0]), GFP_KERNEL);
8ac83505 1919 iov_page = (struct iovec *) __get_free_page(GFP_KERNEL);
59efec7b
TH
1920 if (!pages || !iov_page)
1921 goto out;
1922
1923 /*
1924 * If restricted, initialize IO parameters as encoded in @cmd.
1925 * RETRY from server is not allowed.
1926 */
1927 if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
8ac83505 1928 struct iovec *iov = iov_page;
59efec7b 1929
c9f0523d 1930 iov->iov_base = (void __user *)arg;
59efec7b
TH
1931 iov->iov_len = _IOC_SIZE(cmd);
1932
1933 if (_IOC_DIR(cmd) & _IOC_WRITE) {
1934 in_iov = iov;
1935 in_iovs = 1;
1936 }
1937
1938 if (_IOC_DIR(cmd) & _IOC_READ) {
1939 out_iov = iov;
1940 out_iovs = 1;
1941 }
1942 }
1943
1944 retry:
1945 inarg.in_size = in_size = iov_length(in_iov, in_iovs);
1946 inarg.out_size = out_size = iov_length(out_iov, out_iovs);
1947
1948 /*
1949 * Out data can be used either for actual out data or iovs,
1950 * make sure there always is at least one page.
1951 */
1952 out_size = max_t(size_t, out_size, PAGE_SIZE);
1953 max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
1954
1955 /* make sure there are enough buffer pages and init request with them */
1956 err = -ENOMEM;
1957 if (max_pages > FUSE_MAX_PAGES_PER_REQ)
1958 goto out;
1959 while (num_pages < max_pages) {
1960 pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
1961 if (!pages[num_pages])
1962 goto out;
1963 num_pages++;
1964 }
1965
54b96670 1966 req = fuse_get_req(fc, num_pages);
59efec7b
TH
1967 if (IS_ERR(req)) {
1968 err = PTR_ERR(req);
1969 req = NULL;
1970 goto out;
1971 }
1972 memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages);
1973 req->num_pages = num_pages;
7c190c8b 1974 fuse_page_descs_length_init(req, 0, req->num_pages);
59efec7b
TH
1975
1976 /* okay, let's send it to the client */
1977 req->in.h.opcode = FUSE_IOCTL;
d36f2487 1978 req->in.h.nodeid = ff->nodeid;
59efec7b
TH
1979 req->in.numargs = 1;
1980 req->in.args[0].size = sizeof(inarg);
1981 req->in.args[0].value = &inarg;
1982 if (in_size) {
1983 req->in.numargs++;
1984 req->in.args[1].size = in_size;
1985 req->in.argpages = 1;
1986
1987 err = fuse_ioctl_copy_user(pages, in_iov, in_iovs, in_size,
1988 false);
1989 if (err)
1990 goto out;
1991 }
1992
1993 req->out.numargs = 2;
1994 req->out.args[0].size = sizeof(outarg);
1995 req->out.args[0].value = &outarg;
1996 req->out.args[1].size = out_size;
1997 req->out.argpages = 1;
1998 req->out.argvar = 1;
1999
b93f858a 2000 fuse_request_send(fc, req);
59efec7b
TH
2001 err = req->out.h.error;
2002 transferred = req->out.args[1].size;
2003 fuse_put_request(fc, req);
2004 req = NULL;
2005 if (err)
2006 goto out;
2007
2008 /* did it ask for retry? */
2009 if (outarg.flags & FUSE_IOCTL_RETRY) {
8ac83505 2010 void *vaddr;
59efec7b
TH
2011
2012 /* no retry if in restricted mode */
2013 err = -EIO;
2014 if (!(flags & FUSE_IOCTL_UNRESTRICTED))
2015 goto out;
2016
2017 in_iovs = outarg.in_iovs;
2018 out_iovs = outarg.out_iovs;
2019
2020 /*
2021 * Make sure things are in boundary, separate checks
2022 * are to protect against overflow.
2023 */
2024 err = -ENOMEM;
2025 if (in_iovs > FUSE_IOCTL_MAX_IOV ||
2026 out_iovs > FUSE_IOCTL_MAX_IOV ||
2027 in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
2028 goto out;
2029
2408f6ef 2030 vaddr = kmap_atomic(pages[0]);
1baa26b2 2031 err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr,
d9d318d3
MS
2032 transferred, in_iovs + out_iovs,
2033 (flags & FUSE_IOCTL_COMPAT) != 0);
2408f6ef 2034 kunmap_atomic(vaddr);
d9d318d3
MS
2035 if (err)
2036 goto out;
59efec7b 2037
8ac83505 2038 in_iov = iov_page;
59efec7b
TH
2039 out_iov = in_iov + in_iovs;
2040
7572777e
MS
2041 err = fuse_verify_ioctl_iov(in_iov, in_iovs);
2042 if (err)
2043 goto out;
2044
2045 err = fuse_verify_ioctl_iov(out_iov, out_iovs);
2046 if (err)
2047 goto out;
2048
59efec7b
TH
2049 goto retry;
2050 }
2051
2052 err = -EIO;
2053 if (transferred > inarg.out_size)
2054 goto out;
2055
2056 err = fuse_ioctl_copy_user(pages, out_iov, out_iovs, transferred, true);
2057 out:
2058 if (req)
2059 fuse_put_request(fc, req);
8ac83505 2060 free_page((unsigned long) iov_page);
59efec7b
TH
2061 while (num_pages)
2062 __free_page(pages[--num_pages]);
2063 kfree(pages);
2064
2065 return err ? err : outarg.result;
2066}
08cbf542 2067EXPORT_SYMBOL_GPL(fuse_do_ioctl);
59efec7b 2068
b18da0c5
MS
2069long fuse_ioctl_common(struct file *file, unsigned int cmd,
2070 unsigned long arg, unsigned int flags)
d36f2487
MS
2071{
2072 struct inode *inode = file->f_dentry->d_inode;
2073 struct fuse_conn *fc = get_fuse_conn(inode);
2074
2075 if (!fuse_allow_task(fc, current))
2076 return -EACCES;
2077
2078 if (is_bad_inode(inode))
2079 return -EIO;
2080
2081 return fuse_do_ioctl(file, cmd, arg, flags);
2082}
2083
59efec7b
TH
2084static long fuse_file_ioctl(struct file *file, unsigned int cmd,
2085 unsigned long arg)
2086{
b18da0c5 2087 return fuse_ioctl_common(file, cmd, arg, 0);
59efec7b
TH
2088}
2089
2090static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
2091 unsigned long arg)
2092{
b18da0c5 2093 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
59efec7b
TH
2094}
2095
95668a69
TH
2096/*
2097 * All files which have been polled are linked to RB tree
2098 * fuse_conn->polled_files which is indexed by kh. Walk the tree and
2099 * find the matching one.
2100 */
2101static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh,
2102 struct rb_node **parent_out)
2103{
2104 struct rb_node **link = &fc->polled_files.rb_node;
2105 struct rb_node *last = NULL;
2106
2107 while (*link) {
2108 struct fuse_file *ff;
2109
2110 last = *link;
2111 ff = rb_entry(last, struct fuse_file, polled_node);
2112
2113 if (kh < ff->kh)
2114 link = &last->rb_left;
2115 else if (kh > ff->kh)
2116 link = &last->rb_right;
2117 else
2118 return link;
2119 }
2120
2121 if (parent_out)
2122 *parent_out = last;
2123 return link;
2124}
2125
2126/*
2127 * The file is about to be polled. Make sure it's on the polled_files
2128 * RB tree. Note that files once added to the polled_files tree are
2129 * not removed before the file is released. This is because a file
2130 * polled once is likely to be polled again.
2131 */
2132static void fuse_register_polled_file(struct fuse_conn *fc,
2133 struct fuse_file *ff)
2134{
2135 spin_lock(&fc->lock);
2136 if (RB_EMPTY_NODE(&ff->polled_node)) {
2137 struct rb_node **link, *parent;
2138
2139 link = fuse_find_polled_node(fc, ff->kh, &parent);
2140 BUG_ON(*link);
2141 rb_link_node(&ff->polled_node, parent, link);
2142 rb_insert_color(&ff->polled_node, &fc->polled_files);
2143 }
2144 spin_unlock(&fc->lock);
2145}
2146
08cbf542 2147unsigned fuse_file_poll(struct file *file, poll_table *wait)
95668a69 2148{
95668a69 2149 struct fuse_file *ff = file->private_data;
797759aa 2150 struct fuse_conn *fc = ff->fc;
95668a69
TH
2151 struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh };
2152 struct fuse_poll_out outarg;
2153 struct fuse_req *req;
2154 int err;
2155
2156 if (fc->no_poll)
2157 return DEFAULT_POLLMASK;
2158
2159 poll_wait(file, &ff->poll_wait, wait);
2160
2161 /*
2162 * Ask for notification iff there's someone waiting for it.
2163 * The client may ignore the flag and always notify.
2164 */
2165 if (waitqueue_active(&ff->poll_wait)) {
2166 inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY;
2167 fuse_register_polled_file(fc, ff);
2168 }
2169
b111c8c0 2170 req = fuse_get_req_nopages(fc);
95668a69 2171 if (IS_ERR(req))
201fa69a 2172 return POLLERR;
95668a69
TH
2173
2174 req->in.h.opcode = FUSE_POLL;
797759aa 2175 req->in.h.nodeid = ff->nodeid;
95668a69
TH
2176 req->in.numargs = 1;
2177 req->in.args[0].size = sizeof(inarg);
2178 req->in.args[0].value = &inarg;
2179 req->out.numargs = 1;
2180 req->out.args[0].size = sizeof(outarg);
2181 req->out.args[0].value = &outarg;
b93f858a 2182 fuse_request_send(fc, req);
95668a69
TH
2183 err = req->out.h.error;
2184 fuse_put_request(fc, req);
2185
2186 if (!err)
2187 return outarg.revents;
2188 if (err == -ENOSYS) {
2189 fc->no_poll = 1;
2190 return DEFAULT_POLLMASK;
2191 }
2192 return POLLERR;
2193}
08cbf542 2194EXPORT_SYMBOL_GPL(fuse_file_poll);
95668a69
TH
2195
2196/*
2197 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
2198 * wakes up the poll waiters.
2199 */
2200int fuse_notify_poll_wakeup(struct fuse_conn *fc,
2201 struct fuse_notify_poll_wakeup_out *outarg)
2202{
2203 u64 kh = outarg->kh;
2204 struct rb_node **link;
2205
2206 spin_lock(&fc->lock);
2207
2208 link = fuse_find_polled_node(fc, kh, NULL);
2209 if (*link) {
2210 struct fuse_file *ff;
2211
2212 ff = rb_entry(*link, struct fuse_file, polled_node);
2213 wake_up_interruptible_sync(&ff->poll_wait);
2214 }
2215
2216 spin_unlock(&fc->lock);
2217 return 0;
2218}
2219
4273b793
AA
2220static ssize_t
2221fuse_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
2222 loff_t offset, unsigned long nr_segs)
2223{
2224 ssize_t ret = 0;
2225 struct file *file = NULL;
2226 loff_t pos = 0;
2227
2228 file = iocb->ki_filp;
2229 pos = offset;
2230
b98d023a
MP
2231 if (rw == WRITE)
2232 ret = __fuse_direct_write(file, iov, nr_segs, &pos);
2233 else
2234 ret = __fuse_direct_read(file, iov, nr_segs, &pos);
4273b793
AA
2235
2236 return ret;
2237}
2238
cdadb11c
MS
2239static long fuse_file_fallocate(struct file *file, int mode, loff_t offset,
2240 loff_t length)
05ba1f08
AP
2241{
2242 struct fuse_file *ff = file->private_data;
2243 struct fuse_conn *fc = ff->fc;
2244 struct fuse_req *req;
2245 struct fuse_fallocate_in inarg = {
2246 .fh = ff->fh,
2247 .offset = offset,
2248 .length = length,
2249 .mode = mode
2250 };
2251 int err;
2252
519c6040
MS
2253 if (fc->no_fallocate)
2254 return -EOPNOTSUPP;
2255
b111c8c0 2256 req = fuse_get_req_nopages(fc);
05ba1f08
AP
2257 if (IS_ERR(req))
2258 return PTR_ERR(req);
2259
2260 req->in.h.opcode = FUSE_FALLOCATE;
2261 req->in.h.nodeid = ff->nodeid;
2262 req->in.numargs = 1;
2263 req->in.args[0].size = sizeof(inarg);
2264 req->in.args[0].value = &inarg;
2265 fuse_request_send(fc, req);
2266 err = req->out.h.error;
519c6040
MS
2267 if (err == -ENOSYS) {
2268 fc->no_fallocate = 1;
2269 err = -EOPNOTSUPP;
2270 }
05ba1f08
AP
2271 fuse_put_request(fc, req);
2272
2273 return err;
2274}
05ba1f08 2275
4b6f5d20 2276static const struct file_operations fuse_file_operations = {
5559b8f4 2277 .llseek = fuse_file_llseek,
543ade1f 2278 .read = do_sync_read,
bcb4be80 2279 .aio_read = fuse_file_aio_read,
543ade1f 2280 .write = do_sync_write,
ea9b9907 2281 .aio_write = fuse_file_aio_write,
b6aeaded
MS
2282 .mmap = fuse_file_mmap,
2283 .open = fuse_open,
2284 .flush = fuse_flush,
2285 .release = fuse_release,
2286 .fsync = fuse_fsync,
71421259 2287 .lock = fuse_file_lock,
a9ff4f87 2288 .flock = fuse_file_flock,
5ffc4ef4 2289 .splice_read = generic_file_splice_read,
59efec7b
TH
2290 .unlocked_ioctl = fuse_file_ioctl,
2291 .compat_ioctl = fuse_file_compat_ioctl,
95668a69 2292 .poll = fuse_file_poll,
05ba1f08 2293 .fallocate = fuse_file_fallocate,
b6aeaded
MS
2294};
2295
4b6f5d20 2296static const struct file_operations fuse_direct_io_file_operations = {
5559b8f4 2297 .llseek = fuse_file_llseek,
413ef8cb
MS
2298 .read = fuse_direct_read,
2299 .write = fuse_direct_write,
fc280c96 2300 .mmap = fuse_direct_mmap,
413ef8cb
MS
2301 .open = fuse_open,
2302 .flush = fuse_flush,
2303 .release = fuse_release,
2304 .fsync = fuse_fsync,
71421259 2305 .lock = fuse_file_lock,
a9ff4f87 2306 .flock = fuse_file_flock,
59efec7b
TH
2307 .unlocked_ioctl = fuse_file_ioctl,
2308 .compat_ioctl = fuse_file_compat_ioctl,
95668a69 2309 .poll = fuse_file_poll,
05ba1f08 2310 .fallocate = fuse_file_fallocate,
fc280c96 2311 /* no splice_read */
413ef8cb
MS
2312};
2313
f5e54d6e 2314static const struct address_space_operations fuse_file_aops = {
b6aeaded 2315 .readpage = fuse_readpage,
3be5a52b
MS
2316 .writepage = fuse_writepage,
2317 .launder_page = fuse_launder_page,
db50b96c 2318 .readpages = fuse_readpages,
3be5a52b 2319 .set_page_dirty = __set_page_dirty_nobuffers,
b2d2272f 2320 .bmap = fuse_bmap,
4273b793 2321 .direct_IO = fuse_direct_IO,
b6aeaded
MS
2322};
2323
2324void fuse_init_file_inode(struct inode *inode)
2325{
45323fb7
MS
2326 inode->i_fop = &fuse_file_operations;
2327 inode->i_data.a_ops = &fuse_file_aops;
b6aeaded 2328}