]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/fuse/file.c
mm: move most file-based accounting to the node
[mirror_ubuntu-bionic-kernel.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>
3634a632 18#include <linux/falloc.h>
e2e40f2c 19#include <linux/uio.h>
b6aeaded 20
4b6f5d20 21static const struct file_operations fuse_direct_io_file_operations;
45323fb7 22
91fe96b4
MS
23static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
24 int opcode, struct fuse_open_out *outargp)
b6aeaded 25{
b6aeaded 26 struct fuse_open_in inarg;
7078187a 27 FUSE_ARGS(args);
fd72faac
MS
28
29 memset(&inarg, 0, sizeof(inarg));
6ff958ed
MS
30 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY);
31 if (!fc->atomic_o_trunc)
32 inarg.flags &= ~O_TRUNC;
7078187a
MS
33 args.in.h.opcode = opcode;
34 args.in.h.nodeid = nodeid;
35 args.in.numargs = 1;
36 args.in.args[0].size = sizeof(inarg);
37 args.in.args[0].value = &inarg;
38 args.out.numargs = 1;
39 args.out.args[0].size = sizeof(*outargp);
40 args.out.args[0].value = outargp;
fd72faac 41
7078187a 42 return fuse_simple_request(fc, &args);
fd72faac
MS
43}
44
acf99433 45struct fuse_file *fuse_file_alloc(struct fuse_conn *fc)
fd72faac
MS
46{
47 struct fuse_file *ff;
6b2db28a 48
fd72faac 49 ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
6b2db28a
TH
50 if (unlikely(!ff))
51 return NULL;
52
da5e4714 53 ff->fc = fc;
4250c066 54 ff->reserved_req = fuse_request_alloc(0);
6b2db28a
TH
55 if (unlikely(!ff->reserved_req)) {
56 kfree(ff);
57 return NULL;
fd72faac 58 }
6b2db28a
TH
59
60 INIT_LIST_HEAD(&ff->write_entry);
61 atomic_set(&ff->count, 0);
62 RB_CLEAR_NODE(&ff->polled_node);
63 init_waitqueue_head(&ff->poll_wait);
64
65 spin_lock(&fc->lock);
66 ff->kh = ++fc->khctr;
67 spin_unlock(&fc->lock);
68
fd72faac
MS
69 return ff;
70}
71
72void fuse_file_free(struct fuse_file *ff)
73{
33649c91 74 fuse_request_free(ff->reserved_req);
fd72faac
MS
75 kfree(ff);
76}
77
c7b7143c 78struct fuse_file *fuse_file_get(struct fuse_file *ff)
c756e0a4
MS
79{
80 atomic_inc(&ff->count);
81 return ff;
82}
83
819c4b3b
MS
84static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
85{
baebccbe 86 iput(req->misc.release.inode);
819c4b3b
MS
87}
88
5a18ec17 89static void fuse_file_put(struct fuse_file *ff, bool sync)
c756e0a4
MS
90{
91 if (atomic_dec_and_test(&ff->count)) {
92 struct fuse_req *req = ff->reserved_req;
8b0797a4 93
7678ac50
AG
94 if (ff->fc->no_open) {
95 /*
96 * Drop the release request when client does not
97 * implement 'open'
98 */
825d6d33 99 __clear_bit(FR_BACKGROUND, &req->flags);
baebccbe 100 iput(req->misc.release.inode);
7678ac50
AG
101 fuse_put_request(ff->fc, req);
102 } else if (sync) {
825d6d33 103 __clear_bit(FR_BACKGROUND, &req->flags);
5a18ec17 104 fuse_request_send(ff->fc, req);
baebccbe 105 iput(req->misc.release.inode);
5a18ec17
MS
106 fuse_put_request(ff->fc, req);
107 } else {
108 req->end = fuse_release_end;
825d6d33 109 __set_bit(FR_BACKGROUND, &req->flags);
5a18ec17
MS
110 fuse_request_send_background(ff->fc, req);
111 }
c756e0a4
MS
112 kfree(ff);
113 }
114}
115
08cbf542
TH
116int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
117 bool isdir)
91fe96b4 118{
91fe96b4 119 struct fuse_file *ff;
91fe96b4
MS
120 int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
121
122 ff = fuse_file_alloc(fc);
123 if (!ff)
124 return -ENOMEM;
125
7678ac50
AG
126 ff->fh = 0;
127 ff->open_flags = FOPEN_KEEP_CACHE; /* Default for no-open */
128 if (!fc->no_open || isdir) {
129 struct fuse_open_out outarg;
130 int err;
131
132 err = fuse_send_open(fc, nodeid, file, opcode, &outarg);
133 if (!err) {
134 ff->fh = outarg.fh;
135 ff->open_flags = outarg.open_flags;
136
137 } else if (err != -ENOSYS || isdir) {
138 fuse_file_free(ff);
139 return err;
140 } else {
141 fc->no_open = 1;
142 }
91fe96b4
MS
143 }
144
145 if (isdir)
7678ac50 146 ff->open_flags &= ~FOPEN_DIRECT_IO;
91fe96b4 147
91fe96b4 148 ff->nodeid = nodeid;
91fe96b4
MS
149 file->private_data = fuse_file_get(ff);
150
151 return 0;
152}
08cbf542 153EXPORT_SYMBOL_GPL(fuse_do_open);
91fe96b4 154
650b22b9
PE
155static void fuse_link_write_file(struct file *file)
156{
157 struct inode *inode = file_inode(file);
158 struct fuse_conn *fc = get_fuse_conn(inode);
159 struct fuse_inode *fi = get_fuse_inode(inode);
160 struct fuse_file *ff = file->private_data;
161 /*
162 * file may be written through mmap, so chain it onto the
163 * inodes's write_file list
164 */
165 spin_lock(&fc->lock);
166 if (list_empty(&ff->write_entry))
167 list_add(&ff->write_entry, &fi->write_files);
168 spin_unlock(&fc->lock);
169}
170
c7b7143c 171void fuse_finish_open(struct inode *inode, struct file *file)
fd72faac 172{
c7b7143c 173 struct fuse_file *ff = file->private_data;
a0822c55 174 struct fuse_conn *fc = get_fuse_conn(inode);
c7b7143c
MS
175
176 if (ff->open_flags & FOPEN_DIRECT_IO)
fd72faac 177 file->f_op = &fuse_direct_io_file_operations;
c7b7143c 178 if (!(ff->open_flags & FOPEN_KEEP_CACHE))
b1009979 179 invalidate_inode_pages2(inode->i_mapping);
c7b7143c 180 if (ff->open_flags & FOPEN_NONSEEKABLE)
a7c1b990 181 nonseekable_open(inode, file);
a0822c55
KS
182 if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) {
183 struct fuse_inode *fi = get_fuse_inode(inode);
184
185 spin_lock(&fc->lock);
186 fi->attr_version = ++fc->attr_version;
187 i_size_write(inode, 0);
188 spin_unlock(&fc->lock);
189 fuse_invalidate_attr(inode);
75caeecd
MP
190 if (fc->writeback_cache)
191 file_update_time(file);
a0822c55 192 }
4d99ff8f
PE
193 if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache)
194 fuse_link_write_file(file);
fd72faac
MS
195}
196
91fe96b4 197int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
fd72faac 198{
acf99433 199 struct fuse_conn *fc = get_fuse_conn(inode);
b6aeaded 200 int err;
75caeecd
MP
201 bool lock_inode = (file->f_flags & O_TRUNC) &&
202 fc->atomic_o_trunc &&
203 fc->writeback_cache;
b6aeaded
MS
204
205 err = generic_file_open(inode, file);
206 if (err)
207 return err;
208
75caeecd 209 if (lock_inode)
5955102c 210 inode_lock(inode);
75caeecd 211
91fe96b4 212 err = fuse_do_open(fc, get_node_id(inode), file, isdir);
b6aeaded 213
75caeecd
MP
214 if (!err)
215 fuse_finish_open(inode, file);
91fe96b4 216
75caeecd 217 if (lock_inode)
5955102c 218 inode_unlock(inode);
75caeecd
MP
219
220 return err;
b6aeaded
MS
221}
222
8b0797a4 223static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode)
64c6d8ed 224{
8b0797a4 225 struct fuse_conn *fc = ff->fc;
33649c91 226 struct fuse_req *req = ff->reserved_req;
b57d4264 227 struct fuse_release_in *inarg = &req->misc.release.in;
b6aeaded 228
8b0797a4
MS
229 spin_lock(&fc->lock);
230 list_del(&ff->write_entry);
231 if (!RB_EMPTY_NODE(&ff->polled_node))
232 rb_erase(&ff->polled_node, &fc->polled_files);
233 spin_unlock(&fc->lock);
234
357ccf2b 235 wake_up_interruptible_all(&ff->poll_wait);
8b0797a4 236
b6aeaded 237 inarg->fh = ff->fh;
fd72faac 238 inarg->flags = flags;
51eb01e7 239 req->in.h.opcode = opcode;
c7b7143c 240 req->in.h.nodeid = ff->nodeid;
b6aeaded
MS
241 req->in.numargs = 1;
242 req->in.args[0].size = sizeof(struct fuse_release_in);
243 req->in.args[0].value = inarg;
fd72faac
MS
244}
245
8b0797a4 246void fuse_release_common(struct file *file, int opcode)
fd72faac 247{
6b2db28a
TH
248 struct fuse_file *ff;
249 struct fuse_req *req;
b6aeaded 250
6b2db28a
TH
251 ff = file->private_data;
252 if (unlikely(!ff))
8b0797a4 253 return;
6b2db28a 254
6b2db28a 255 req = ff->reserved_req;
8b0797a4 256 fuse_prepare_release(ff, file->f_flags, opcode);
6b2db28a 257
37fb3a30
MS
258 if (ff->flock) {
259 struct fuse_release_in *inarg = &req->misc.release.in;
260 inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK;
261 inarg->lock_owner = fuse_lock_owner_id(ff->fc,
262 (fl_owner_t) file);
263 }
baebccbe
MS
264 /* Hold inode until release is finished */
265 req->misc.release.inode = igrab(file_inode(file));
6b2db28a 266
6b2db28a
TH
267 /*
268 * Normally this will send the RELEASE request, however if
269 * some asynchronous READ or WRITE requests are outstanding,
270 * the sending will be delayed.
5a18ec17
MS
271 *
272 * Make the release synchronous if this is a fuseblk mount,
273 * synchronous RELEASE is allowed (and desirable) in this case
274 * because the server can be trusted not to screw up.
6b2db28a 275 */
5a18ec17 276 fuse_file_put(ff, ff->fc->destroy_req != NULL);
b6aeaded
MS
277}
278
04730fef
MS
279static int fuse_open(struct inode *inode, struct file *file)
280{
91fe96b4 281 return fuse_open_common(inode, file, false);
04730fef
MS
282}
283
284static int fuse_release(struct inode *inode, struct file *file)
285{
e7cc133c
PE
286 struct fuse_conn *fc = get_fuse_conn(inode);
287
288 /* see fuse_vma_close() for !writeback_cache case */
289 if (fc->writeback_cache)
1e18bda8 290 write_inode_now(inode, 1);
b0aa7606 291
8b0797a4
MS
292 fuse_release_common(file, FUSE_RELEASE);
293
294 /* return value is ignored by VFS */
295 return 0;
296}
297
298void fuse_sync_release(struct fuse_file *ff, int flags)
299{
300 WARN_ON(atomic_read(&ff->count) > 1);
301 fuse_prepare_release(ff, flags, FUSE_RELEASE);
825d6d33
MS
302 __set_bit(FR_FORCE, &ff->reserved_req->flags);
303 __clear_bit(FR_BACKGROUND, &ff->reserved_req->flags);
8b0797a4
MS
304 fuse_request_send(ff->fc, ff->reserved_req);
305 fuse_put_request(ff->fc, ff->reserved_req);
306 kfree(ff);
04730fef 307}
08cbf542 308EXPORT_SYMBOL_GPL(fuse_sync_release);
04730fef 309
71421259 310/*
9c8ef561
MS
311 * Scramble the ID space with XTEA, so that the value of the files_struct
312 * pointer is not exposed to userspace.
71421259 313 */
f3332114 314u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
71421259 315{
9c8ef561
MS
316 u32 *k = fc->scramble_key;
317 u64 v = (unsigned long) id;
318 u32 v0 = v;
319 u32 v1 = v >> 32;
320 u32 sum = 0;
321 int i;
322
323 for (i = 0; i < 32; i++) {
324 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
325 sum += 0x9E3779B9;
326 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
327 }
328
329 return (u64) v0 + ((u64) v1 << 32);
71421259
MS
330}
331
3be5a52b 332/*
ea8cd333 333 * Check if any page in a range is under writeback
3be5a52b
MS
334 *
335 * This is currently done by walking the list of writepage requests
336 * for the inode, which can be pretty inefficient.
337 */
ea8cd333
PE
338static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from,
339 pgoff_t idx_to)
3be5a52b
MS
340{
341 struct fuse_conn *fc = get_fuse_conn(inode);
342 struct fuse_inode *fi = get_fuse_inode(inode);
343 struct fuse_req *req;
344 bool found = false;
345
346 spin_lock(&fc->lock);
347 list_for_each_entry(req, &fi->writepages, writepages_entry) {
348 pgoff_t curr_index;
349
350 BUG_ON(req->inode != inode);
09cbfeaf 351 curr_index = req->misc.write.in.offset >> PAGE_SHIFT;
ea8cd333
PE
352 if (idx_from < curr_index + req->num_pages &&
353 curr_index <= idx_to) {
3be5a52b
MS
354 found = true;
355 break;
356 }
357 }
358 spin_unlock(&fc->lock);
359
360 return found;
361}
362
ea8cd333
PE
363static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
364{
365 return fuse_range_is_writeback(inode, index, index);
366}
367
3be5a52b
MS
368/*
369 * Wait for page writeback to be completed.
370 *
371 * Since fuse doesn't rely on the VM writeback tracking, this has to
372 * use some other means.
373 */
374static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
375{
376 struct fuse_inode *fi = get_fuse_inode(inode);
377
378 wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
379 return 0;
380}
381
fe38d7df
MP
382/*
383 * Wait for all pending writepages on the inode to finish.
384 *
385 * This is currently done by blocking further writes with FUSE_NOWRITE
386 * and waiting for all sent writes to complete.
387 *
388 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
389 * could conflict with truncation.
390 */
391static void fuse_sync_writes(struct inode *inode)
392{
393 fuse_set_nowrite(inode);
394 fuse_release_nowrite(inode);
395}
396
75e1fcc0 397static int fuse_flush(struct file *file, fl_owner_t id)
b6aeaded 398{
6131ffaa 399 struct inode *inode = file_inode(file);
b6aeaded
MS
400 struct fuse_conn *fc = get_fuse_conn(inode);
401 struct fuse_file *ff = file->private_data;
402 struct fuse_req *req;
403 struct fuse_flush_in inarg;
404 int err;
405
248d86e8
MS
406 if (is_bad_inode(inode))
407 return -EIO;
408
b6aeaded
MS
409 if (fc->no_flush)
410 return 0;
411
1e18bda8 412 err = write_inode_now(inode, 1);
fe38d7df
MP
413 if (err)
414 return err;
415
5955102c 416 inode_lock(inode);
fe38d7df 417 fuse_sync_writes(inode);
5955102c 418 inode_unlock(inode);
fe38d7df 419
b111c8c0 420 req = fuse_get_req_nofail_nopages(fc, file);
b6aeaded
MS
421 memset(&inarg, 0, sizeof(inarg));
422 inarg.fh = ff->fh;
9c8ef561 423 inarg.lock_owner = fuse_lock_owner_id(fc, id);
b6aeaded
MS
424 req->in.h.opcode = FUSE_FLUSH;
425 req->in.h.nodeid = get_node_id(inode);
b6aeaded
MS
426 req->in.numargs = 1;
427 req->in.args[0].size = sizeof(inarg);
428 req->in.args[0].value = &inarg;
825d6d33 429 __set_bit(FR_FORCE, &req->flags);
b93f858a 430 fuse_request_send(fc, req);
b6aeaded
MS
431 err = req->out.h.error;
432 fuse_put_request(fc, req);
433 if (err == -ENOSYS) {
434 fc->no_flush = 1;
435 err = 0;
436 }
437 return err;
438}
439
02c24a82
JB
440int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
441 int datasync, int isdir)
b6aeaded 442{
7ea80859 443 struct inode *inode = file->f_mapping->host;
b6aeaded
MS
444 struct fuse_conn *fc = get_fuse_conn(inode);
445 struct fuse_file *ff = file->private_data;
7078187a 446 FUSE_ARGS(args);
b6aeaded
MS
447 struct fuse_fsync_in inarg;
448 int err;
449
248d86e8
MS
450 if (is_bad_inode(inode))
451 return -EIO;
452
5955102c 453 inode_lock(inode);
02c24a82 454
3be5a52b
MS
455 /*
456 * Start writeback against all dirty pages of the inode, then
457 * wait for all outstanding writes, before sending the FSYNC
458 * request.
459 */
22401e7b 460 err = filemap_write_and_wait_range(inode->i_mapping, start, end);
3be5a52b 461 if (err)
02c24a82 462 goto out;
3be5a52b
MS
463
464 fuse_sync_writes(inode);
1e18bda8
MS
465 err = sync_inode_metadata(inode, 1);
466 if (err)
467 goto out;
3be5a52b 468
22401e7b
MS
469 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
470 goto out;
b0aa7606 471
b6aeaded
MS
472 memset(&inarg, 0, sizeof(inarg));
473 inarg.fh = ff->fh;
474 inarg.fsync_flags = datasync ? 1 : 0;
7078187a
MS
475 args.in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
476 args.in.h.nodeid = get_node_id(inode);
477 args.in.numargs = 1;
478 args.in.args[0].size = sizeof(inarg);
479 args.in.args[0].value = &inarg;
480 err = fuse_simple_request(fc, &args);
b6aeaded 481 if (err == -ENOSYS) {
82547981
MS
482 if (isdir)
483 fc->no_fsyncdir = 1;
484 else
485 fc->no_fsync = 1;
b6aeaded
MS
486 err = 0;
487 }
02c24a82 488out:
5955102c 489 inode_unlock(inode);
b6aeaded
MS
490 return err;
491}
492
02c24a82
JB
493static int fuse_fsync(struct file *file, loff_t start, loff_t end,
494 int datasync)
82547981 495{
02c24a82 496 return fuse_fsync_common(file, start, end, datasync, 0);
82547981
MS
497}
498
2106cb18
MS
499void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos,
500 size_t count, int opcode)
b6aeaded 501{
5c5c5e51 502 struct fuse_read_in *inarg = &req->misc.read.in;
a6643094 503 struct fuse_file *ff = file->private_data;
b6aeaded 504
361b1eb5
MS
505 inarg->fh = ff->fh;
506 inarg->offset = pos;
507 inarg->size = count;
a6643094 508 inarg->flags = file->f_flags;
361b1eb5 509 req->in.h.opcode = opcode;
2106cb18 510 req->in.h.nodeid = ff->nodeid;
b6aeaded
MS
511 req->in.numargs = 1;
512 req->in.args[0].size = sizeof(struct fuse_read_in);
c1aa96a5 513 req->in.args[0].value = inarg;
b6aeaded
MS
514 req->out.argvar = 1;
515 req->out.numargs = 1;
516 req->out.args[0].size = count;
b6aeaded
MS
517}
518
187c5c36
MP
519static void fuse_release_user_pages(struct fuse_req *req, int write)
520{
521 unsigned i;
522
523 for (i = 0; i < req->num_pages; i++) {
524 struct page *page = req->pages[i];
525 if (write)
526 set_page_dirty_lock(page);
527 put_page(page);
528 }
529}
530
744742d6
SF
531static void fuse_io_release(struct kref *kref)
532{
533 kfree(container_of(kref, struct fuse_io_priv, refcnt));
534}
535
9d5722b7
CH
536static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io)
537{
538 if (io->err)
539 return io->err;
540
541 if (io->bytes >= 0 && io->write)
542 return -EIO;
543
544 return io->bytes < 0 ? io->size : io->bytes;
545}
546
01e9d11a
MP
547/**
548 * In case of short read, the caller sets 'pos' to the position of
549 * actual end of fuse request in IO request. Otherwise, if bytes_requested
550 * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1.
551 *
552 * An example:
553 * User requested DIO read of 64K. It was splitted into two 32K fuse requests,
554 * both submitted asynchronously. The first of them was ACKed by userspace as
555 * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The
556 * second request was ACKed as short, e.g. only 1K was read, resulting in
557 * pos == 33K.
558 *
559 * Thus, when all fuse requests are completed, the minimal non-negative 'pos'
560 * will be equal to the length of the longest contiguous fragment of
561 * transferred data starting from the beginning of IO request.
562 */
563static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
564{
9d5722b7 565 bool is_sync = is_sync_kiocb(io->iocb);
01e9d11a
MP
566 int left;
567
568 spin_lock(&io->lock);
569 if (err)
570 io->err = io->err ? : err;
571 else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes))
572 io->bytes = pos;
573
574 left = --io->reqs;
9d5722b7
CH
575 if (!left && is_sync)
576 complete(io->done);
01e9d11a
MP
577 spin_unlock(&io->lock);
578
9d5722b7
CH
579 if (!left && !is_sync) {
580 ssize_t res = fuse_get_res_by_io(io);
01e9d11a 581
9d5722b7
CH
582 if (res >= 0) {
583 struct inode *inode = file_inode(io->iocb->ki_filp);
584 struct fuse_conn *fc = get_fuse_conn(inode);
585 struct fuse_inode *fi = get_fuse_inode(inode);
01e9d11a 586
9d5722b7
CH
587 spin_lock(&fc->lock);
588 fi->attr_version = ++fc->attr_version;
589 spin_unlock(&fc->lock);
01e9d11a
MP
590 }
591
04b2fa9f 592 io->iocb->ki_complete(io->iocb, res, 0);
01e9d11a 593 }
744742d6
SF
594
595 kref_put(&io->refcnt, fuse_io_release);
01e9d11a
MP
596}
597
598static void fuse_aio_complete_req(struct fuse_conn *fc, struct fuse_req *req)
599{
600 struct fuse_io_priv *io = req->io;
601 ssize_t pos = -1;
602
603 fuse_release_user_pages(req, !io->write);
604
605 if (io->write) {
606 if (req->misc.write.in.size != req->misc.write.out.size)
607 pos = req->misc.write.in.offset - io->offset +
608 req->misc.write.out.size;
609 } else {
610 if (req->misc.read.in.size != req->out.args[0].size)
611 pos = req->misc.read.in.offset - io->offset +
612 req->out.args[0].size;
613 }
614
615 fuse_aio_complete(io, req->out.h.error, pos);
616}
617
618static size_t fuse_async_req_send(struct fuse_conn *fc, struct fuse_req *req,
619 size_t num_bytes, struct fuse_io_priv *io)
620{
621 spin_lock(&io->lock);
744742d6 622 kref_get(&io->refcnt);
01e9d11a
MP
623 io->size += num_bytes;
624 io->reqs++;
625 spin_unlock(&io->lock);
626
627 req->io = io;
628 req->end = fuse_aio_complete_req;
629
36cf66ed 630 __fuse_get_request(req);
01e9d11a
MP
631 fuse_request_send_background(fc, req);
632
633 return num_bytes;
634}
635
36cf66ed 636static size_t fuse_send_read(struct fuse_req *req, struct fuse_io_priv *io,
2106cb18 637 loff_t pos, size_t count, fl_owner_t owner)
04730fef 638{
36cf66ed 639 struct file *file = io->file;
2106cb18
MS
640 struct fuse_file *ff = file->private_data;
641 struct fuse_conn *fc = ff->fc;
f3332114 642
2106cb18 643 fuse_read_fill(req, file, pos, count, FUSE_READ);
f3332114 644 if (owner != NULL) {
5c5c5e51 645 struct fuse_read_in *inarg = &req->misc.read.in;
f3332114
MS
646
647 inarg->read_flags |= FUSE_READ_LOCKOWNER;
648 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
649 }
36cf66ed
MP
650
651 if (io->async)
652 return fuse_async_req_send(fc, req, count, io);
653
b93f858a 654 fuse_request_send(fc, req);
361b1eb5 655 return req->out.args[0].size;
04730fef
MS
656}
657
5c5c5e51
MS
658static void fuse_read_update_size(struct inode *inode, loff_t size,
659 u64 attr_ver)
660{
661 struct fuse_conn *fc = get_fuse_conn(inode);
662 struct fuse_inode *fi = get_fuse_inode(inode);
663
664 spin_lock(&fc->lock);
06a7c3c2
MP
665 if (attr_ver == fi->attr_version && size < inode->i_size &&
666 !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
5c5c5e51
MS
667 fi->attr_version = ++fc->attr_version;
668 i_size_write(inode, size);
669 }
670 spin_unlock(&fc->lock);
671}
672
a92adc82
PE
673static void fuse_short_read(struct fuse_req *req, struct inode *inode,
674 u64 attr_ver)
675{
676 size_t num_read = req->out.args[0].size;
8373200b
PE
677 struct fuse_conn *fc = get_fuse_conn(inode);
678
679 if (fc->writeback_cache) {
680 /*
681 * A hole in a file. Some data after the hole are in page cache,
682 * but have not reached the client fs yet. So, the hole is not
683 * present there.
684 */
685 int i;
09cbfeaf
KS
686 int start_idx = num_read >> PAGE_SHIFT;
687 size_t off = num_read & (PAGE_SIZE - 1);
a92adc82 688
8373200b 689 for (i = start_idx; i < req->num_pages; i++) {
09cbfeaf 690 zero_user_segment(req->pages[i], off, PAGE_SIZE);
8373200b
PE
691 off = 0;
692 }
693 } else {
694 loff_t pos = page_offset(req->pages[0]) + num_read;
695 fuse_read_update_size(inode, pos, attr_ver);
696 }
a92adc82
PE
697}
698
482fce55 699static int fuse_do_readpage(struct file *file, struct page *page)
b6aeaded 700{
744742d6 701 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(file);
b6aeaded
MS
702 struct inode *inode = page->mapping->host;
703 struct fuse_conn *fc = get_fuse_conn(inode);
248d86e8 704 struct fuse_req *req;
5c5c5e51
MS
705 size_t num_read;
706 loff_t pos = page_offset(page);
09cbfeaf 707 size_t count = PAGE_SIZE;
5c5c5e51 708 u64 attr_ver;
248d86e8
MS
709 int err;
710
3be5a52b 711 /*
25985edc 712 * Page writeback can extend beyond the lifetime of the
3be5a52b
MS
713 * page-cache page, so make sure we read a properly synced
714 * page.
715 */
716 fuse_wait_on_page_writeback(inode, page->index);
717
b111c8c0 718 req = fuse_get_req(fc, 1);
ce1d5a49 719 if (IS_ERR(req))
482fce55 720 return PTR_ERR(req);
b6aeaded 721
5c5c5e51
MS
722 attr_ver = fuse_get_attr_version(fc);
723
b6aeaded 724 req->out.page_zeroing = 1;
f4975c67 725 req->out.argpages = 1;
b6aeaded
MS
726 req->num_pages = 1;
727 req->pages[0] = page;
85f40aec 728 req->page_descs[0].length = count;
36cf66ed 729 num_read = fuse_send_read(req, &io, pos, count, NULL);
b6aeaded 730 err = req->out.h.error;
5c5c5e51
MS
731
732 if (!err) {
733 /*
734 * Short read means EOF. If file size is larger, truncate it
735 */
736 if (num_read < count)
a92adc82 737 fuse_short_read(req, inode, attr_ver);
5c5c5e51 738
b6aeaded 739 SetPageUptodate(page);
5c5c5e51
MS
740 }
741
a92adc82 742 fuse_put_request(fc, req);
482fce55
MP
743
744 return err;
745}
746
747static int fuse_readpage(struct file *file, struct page *page)
748{
749 struct inode *inode = page->mapping->host;
750 int err;
751
752 err = -EIO;
753 if (is_bad_inode(inode))
754 goto out;
755
756 err = fuse_do_readpage(file, page);
451418fc 757 fuse_invalidate_atime(inode);
b6aeaded
MS
758 out:
759 unlock_page(page);
760 return err;
761}
762
c1aa96a5 763static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
db50b96c 764{
c1aa96a5 765 int i;
5c5c5e51
MS
766 size_t count = req->misc.read.in.size;
767 size_t num_read = req->out.args[0].size;
ce534fb0 768 struct address_space *mapping = NULL;
c1aa96a5 769
ce534fb0
MS
770 for (i = 0; mapping == NULL && i < req->num_pages; i++)
771 mapping = req->pages[i]->mapping;
5c5c5e51 772
ce534fb0
MS
773 if (mapping) {
774 struct inode *inode = mapping->host;
775
776 /*
777 * Short read means EOF. If file size is larger, truncate it
778 */
a92adc82
PE
779 if (!req->out.h.error && num_read < count)
780 fuse_short_read(req, inode, req->misc.read.attr_ver);
ce534fb0 781
451418fc 782 fuse_invalidate_atime(inode);
ce534fb0 783 }
c1aa96a5 784
db50b96c
MS
785 for (i = 0; i < req->num_pages; i++) {
786 struct page *page = req->pages[i];
787 if (!req->out.h.error)
788 SetPageUptodate(page);
c1aa96a5
MS
789 else
790 SetPageError(page);
db50b96c 791 unlock_page(page);
09cbfeaf 792 put_page(page);
db50b96c 793 }
c756e0a4 794 if (req->ff)
5a18ec17 795 fuse_file_put(req->ff, false);
c1aa96a5
MS
796}
797
2106cb18 798static void fuse_send_readpages(struct fuse_req *req, struct file *file)
c1aa96a5 799{
2106cb18
MS
800 struct fuse_file *ff = file->private_data;
801 struct fuse_conn *fc = ff->fc;
c1aa96a5 802 loff_t pos = page_offset(req->pages[0]);
09cbfeaf 803 size_t count = req->num_pages << PAGE_SHIFT;
f4975c67
MS
804
805 req->out.argpages = 1;
c1aa96a5 806 req->out.page_zeroing = 1;
ce534fb0 807 req->out.page_replace = 1;
2106cb18 808 fuse_read_fill(req, file, pos, count, FUSE_READ);
5c5c5e51 809 req->misc.read.attr_ver = fuse_get_attr_version(fc);
9cd68455 810 if (fc->async_read) {
c756e0a4 811 req->ff = fuse_file_get(ff);
9cd68455 812 req->end = fuse_readpages_end;
b93f858a 813 fuse_request_send_background(fc, req);
9cd68455 814 } else {
b93f858a 815 fuse_request_send(fc, req);
9cd68455 816 fuse_readpages_end(fc, req);
e9bb09dd 817 fuse_put_request(fc, req);
9cd68455 818 }
db50b96c
MS
819}
820
c756e0a4 821struct fuse_fill_data {
db50b96c 822 struct fuse_req *req;
a6643094 823 struct file *file;
db50b96c 824 struct inode *inode;
f8dbdf81 825 unsigned nr_pages;
db50b96c
MS
826};
827
828static int fuse_readpages_fill(void *_data, struct page *page)
829{
c756e0a4 830 struct fuse_fill_data *data = _data;
db50b96c
MS
831 struct fuse_req *req = data->req;
832 struct inode *inode = data->inode;
833 struct fuse_conn *fc = get_fuse_conn(inode);
834
3be5a52b
MS
835 fuse_wait_on_page_writeback(inode, page->index);
836
db50b96c
MS
837 if (req->num_pages &&
838 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
09cbfeaf 839 (req->num_pages + 1) * PAGE_SIZE > fc->max_read ||
db50b96c 840 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
f8dbdf81
MP
841 int nr_alloc = min_t(unsigned, data->nr_pages,
842 FUSE_MAX_PAGES_PER_REQ);
2106cb18 843 fuse_send_readpages(req, data->file);
8b41e671
MP
844 if (fc->async_read)
845 req = fuse_get_req_for_background(fc, nr_alloc);
846 else
847 req = fuse_get_req(fc, nr_alloc);
848
849 data->req = req;
ce1d5a49 850 if (IS_ERR(req)) {
db50b96c 851 unlock_page(page);
ce1d5a49 852 return PTR_ERR(req);
db50b96c 853 }
db50b96c 854 }
f8dbdf81
MP
855
856 if (WARN_ON(req->num_pages >= req->max_pages)) {
857 fuse_put_request(fc, req);
858 return -EIO;
859 }
860
09cbfeaf 861 get_page(page);
db50b96c 862 req->pages[req->num_pages] = page;
85f40aec 863 req->page_descs[req->num_pages].length = PAGE_SIZE;
1729a16c 864 req->num_pages++;
f8dbdf81 865 data->nr_pages--;
db50b96c
MS
866 return 0;
867}
868
869static int fuse_readpages(struct file *file, struct address_space *mapping,
870 struct list_head *pages, unsigned nr_pages)
871{
872 struct inode *inode = mapping->host;
873 struct fuse_conn *fc = get_fuse_conn(inode);
c756e0a4 874 struct fuse_fill_data data;
db50b96c 875 int err;
f8dbdf81 876 int nr_alloc = min_t(unsigned, nr_pages, FUSE_MAX_PAGES_PER_REQ);
248d86e8 877
1d7ea732 878 err = -EIO;
248d86e8 879 if (is_bad_inode(inode))
2e990021 880 goto out;
248d86e8 881
a6643094 882 data.file = file;
db50b96c 883 data.inode = inode;
8b41e671
MP
884 if (fc->async_read)
885 data.req = fuse_get_req_for_background(fc, nr_alloc);
886 else
887 data.req = fuse_get_req(fc, nr_alloc);
f8dbdf81 888 data.nr_pages = nr_pages;
1d7ea732 889 err = PTR_ERR(data.req);
ce1d5a49 890 if (IS_ERR(data.req))
2e990021 891 goto out;
db50b96c
MS
892
893 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
d3406ffa
MS
894 if (!err) {
895 if (data.req->num_pages)
2106cb18 896 fuse_send_readpages(data.req, file);
d3406ffa
MS
897 else
898 fuse_put_request(fc, data.req);
899 }
2e990021 900out:
1d7ea732 901 return err;
db50b96c
MS
902}
903
37c20f16 904static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
bcb4be80
MS
905{
906 struct inode *inode = iocb->ki_filp->f_mapping->host;
a8894274 907 struct fuse_conn *fc = get_fuse_conn(inode);
bcb4be80 908
a8894274
BF
909 /*
910 * In auto invalidate mode, always update attributes on read.
911 * Otherwise, only update if we attempt to read past EOF (to ensure
912 * i_size is up to date).
913 */
914 if (fc->auto_inval_data ||
37c20f16 915 (iocb->ki_pos + iov_iter_count(to) > i_size_read(inode))) {
bcb4be80 916 int err;
bcb4be80
MS
917 err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL);
918 if (err)
919 return err;
920 }
921
37c20f16 922 return generic_file_read_iter(iocb, to);
bcb4be80
MS
923}
924
2d698b07 925static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff,
2106cb18 926 loff_t pos, size_t count)
b6aeaded 927{
b25e82e5
MS
928 struct fuse_write_in *inarg = &req->misc.write.in;
929 struct fuse_write_out *outarg = &req->misc.write.out;
b6aeaded 930
b25e82e5
MS
931 inarg->fh = ff->fh;
932 inarg->offset = pos;
933 inarg->size = count;
b6aeaded 934 req->in.h.opcode = FUSE_WRITE;
2106cb18 935 req->in.h.nodeid = ff->nodeid;
b6aeaded 936 req->in.numargs = 2;
2106cb18 937 if (ff->fc->minor < 9)
f3332114
MS
938 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
939 else
940 req->in.args[0].size = sizeof(struct fuse_write_in);
b25e82e5 941 req->in.args[0].value = inarg;
b6aeaded
MS
942 req->in.args[1].size = count;
943 req->out.numargs = 1;
944 req->out.args[0].size = sizeof(struct fuse_write_out);
b25e82e5
MS
945 req->out.args[0].value = outarg;
946}
947
36cf66ed 948static size_t fuse_send_write(struct fuse_req *req, struct fuse_io_priv *io,
2106cb18 949 loff_t pos, size_t count, fl_owner_t owner)
b25e82e5 950{
36cf66ed 951 struct file *file = io->file;
2106cb18
MS
952 struct fuse_file *ff = file->private_data;
953 struct fuse_conn *fc = ff->fc;
2d698b07
MS
954 struct fuse_write_in *inarg = &req->misc.write.in;
955
2106cb18 956 fuse_write_fill(req, ff, pos, count);
2d698b07 957 inarg->flags = file->f_flags;
f3332114 958 if (owner != NULL) {
f3332114
MS
959 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
960 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
961 }
36cf66ed
MP
962
963 if (io->async)
964 return fuse_async_req_send(fc, req, count, io);
965
b93f858a 966 fuse_request_send(fc, req);
b25e82e5 967 return req->misc.write.out.size;
b6aeaded
MS
968}
969
b0aa7606 970bool fuse_write_update_size(struct inode *inode, loff_t pos)
854512ec
MS
971{
972 struct fuse_conn *fc = get_fuse_conn(inode);
973 struct fuse_inode *fi = get_fuse_inode(inode);
b0aa7606 974 bool ret = false;
854512ec
MS
975
976 spin_lock(&fc->lock);
977 fi->attr_version = ++fc->attr_version;
b0aa7606 978 if (pos > inode->i_size) {
854512ec 979 i_size_write(inode, pos);
b0aa7606
MP
980 ret = true;
981 }
854512ec 982 spin_unlock(&fc->lock);
b0aa7606
MP
983
984 return ret;
854512ec
MS
985}
986
ea9b9907
NP
987static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file,
988 struct inode *inode, loff_t pos,
989 size_t count)
990{
991 size_t res;
992 unsigned offset;
993 unsigned i;
744742d6 994 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(file);
ea9b9907
NP
995
996 for (i = 0; i < req->num_pages; i++)
997 fuse_wait_on_page_writeback(inode, req->pages[i]->index);
998
36cf66ed 999 res = fuse_send_write(req, &io, pos, count, NULL);
ea9b9907 1000
b2430d75 1001 offset = req->page_descs[0].offset;
ea9b9907
NP
1002 count = res;
1003 for (i = 0; i < req->num_pages; i++) {
1004 struct page *page = req->pages[i];
1005
09cbfeaf 1006 if (!req->out.h.error && !offset && count >= PAGE_SIZE)
ea9b9907
NP
1007 SetPageUptodate(page);
1008
09cbfeaf
KS
1009 if (count > PAGE_SIZE - offset)
1010 count -= PAGE_SIZE - offset;
ea9b9907
NP
1011 else
1012 count = 0;
1013 offset = 0;
1014
1015 unlock_page(page);
09cbfeaf 1016 put_page(page);
ea9b9907
NP
1017 }
1018
1019 return res;
1020}
1021
1022static ssize_t fuse_fill_write_pages(struct fuse_req *req,
1023 struct address_space *mapping,
1024 struct iov_iter *ii, loff_t pos)
1025{
1026 struct fuse_conn *fc = get_fuse_conn(mapping->host);
09cbfeaf 1027 unsigned offset = pos & (PAGE_SIZE - 1);
ea9b9907
NP
1028 size_t count = 0;
1029 int err;
1030
f4975c67 1031 req->in.argpages = 1;
b2430d75 1032 req->page_descs[0].offset = offset;
ea9b9907
NP
1033
1034 do {
1035 size_t tmp;
1036 struct page *page;
09cbfeaf
KS
1037 pgoff_t index = pos >> PAGE_SHIFT;
1038 size_t bytes = min_t(size_t, PAGE_SIZE - offset,
ea9b9907
NP
1039 iov_iter_count(ii));
1040
1041 bytes = min_t(size_t, bytes, fc->max_write - count);
1042
1043 again:
1044 err = -EFAULT;
1045 if (iov_iter_fault_in_readable(ii, bytes))
1046 break;
1047
1048 err = -ENOMEM;
54566b2c 1049 page = grab_cache_page_write_begin(mapping, index, 0);
ea9b9907
NP
1050 if (!page)
1051 break;
1052
931e80e4 1053 if (mapping_writably_mapped(mapping))
1054 flush_dcache_page(page);
1055
ea9b9907 1056 tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes);
ea9b9907
NP
1057 flush_dcache_page(page);
1058
3ca8138f 1059 iov_iter_advance(ii, tmp);
ea9b9907
NP
1060 if (!tmp) {
1061 unlock_page(page);
09cbfeaf 1062 put_page(page);
ea9b9907
NP
1063 bytes = min(bytes, iov_iter_single_seg_count(ii));
1064 goto again;
1065 }
1066
1067 err = 0;
1068 req->pages[req->num_pages] = page;
85f40aec 1069 req->page_descs[req->num_pages].length = tmp;
ea9b9907
NP
1070 req->num_pages++;
1071
ea9b9907
NP
1072 count += tmp;
1073 pos += tmp;
1074 offset += tmp;
09cbfeaf 1075 if (offset == PAGE_SIZE)
ea9b9907
NP
1076 offset = 0;
1077
78bb6cb9
MS
1078 if (!fc->big_writes)
1079 break;
ea9b9907 1080 } while (iov_iter_count(ii) && count < fc->max_write &&
d07f09f5 1081 req->num_pages < req->max_pages && offset == 0);
ea9b9907
NP
1082
1083 return count > 0 ? count : err;
1084}
1085
d07f09f5
MP
1086static inline unsigned fuse_wr_pages(loff_t pos, size_t len)
1087{
1088 return min_t(unsigned,
09cbfeaf
KS
1089 ((pos + len - 1) >> PAGE_SHIFT) -
1090 (pos >> PAGE_SHIFT) + 1,
d07f09f5
MP
1091 FUSE_MAX_PAGES_PER_REQ);
1092}
1093
ea9b9907
NP
1094static ssize_t fuse_perform_write(struct file *file,
1095 struct address_space *mapping,
1096 struct iov_iter *ii, loff_t pos)
1097{
1098 struct inode *inode = mapping->host;
1099 struct fuse_conn *fc = get_fuse_conn(inode);
06a7c3c2 1100 struct fuse_inode *fi = get_fuse_inode(inode);
ea9b9907
NP
1101 int err = 0;
1102 ssize_t res = 0;
1103
1104 if (is_bad_inode(inode))
1105 return -EIO;
1106
06a7c3c2
MP
1107 if (inode->i_size < pos + iov_iter_count(ii))
1108 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1109
ea9b9907
NP
1110 do {
1111 struct fuse_req *req;
1112 ssize_t count;
d07f09f5 1113 unsigned nr_pages = fuse_wr_pages(pos, iov_iter_count(ii));
ea9b9907 1114
d07f09f5 1115 req = fuse_get_req(fc, nr_pages);
ea9b9907
NP
1116 if (IS_ERR(req)) {
1117 err = PTR_ERR(req);
1118 break;
1119 }
1120
1121 count = fuse_fill_write_pages(req, mapping, ii, pos);
1122 if (count <= 0) {
1123 err = count;
1124 } else {
1125 size_t num_written;
1126
1127 num_written = fuse_send_write_pages(req, file, inode,
1128 pos, count);
1129 err = req->out.h.error;
1130 if (!err) {
1131 res += num_written;
1132 pos += num_written;
1133
1134 /* break out of the loop on short write */
1135 if (num_written != count)
1136 err = -EIO;
1137 }
1138 }
1139 fuse_put_request(fc, req);
1140 } while (!err && iov_iter_count(ii));
1141
1142 if (res > 0)
1143 fuse_write_update_size(inode, pos);
1144
06a7c3c2 1145 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
ea9b9907
NP
1146 fuse_invalidate_attr(inode);
1147
1148 return res > 0 ? res : err;
1149}
1150
84c3d55c 1151static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
ea9b9907
NP
1152{
1153 struct file *file = iocb->ki_filp;
1154 struct address_space *mapping = file->f_mapping;
ea9b9907 1155 ssize_t written = 0;
4273b793 1156 ssize_t written_buffered = 0;
ea9b9907
NP
1157 struct inode *inode = mapping->host;
1158 ssize_t err;
4273b793 1159 loff_t endbyte = 0;
ea9b9907 1160
4d99ff8f
PE
1161 if (get_fuse_conn(inode)->writeback_cache) {
1162 /* Update size (EOF optimization) and mode (SUID clearing) */
1163 err = fuse_update_attributes(mapping->host, NULL, file, NULL);
1164 if (err)
1165 return err;
1166
84c3d55c 1167 return generic_file_write_iter(iocb, from);
4d99ff8f
PE
1168 }
1169
5955102c 1170 inode_lock(inode);
ea9b9907
NP
1171
1172 /* We can write back this queue in page reclaim */
de1414a6 1173 current->backing_dev_info = inode_to_bdi(inode);
ea9b9907 1174
3309dd04
AV
1175 err = generic_write_checks(iocb, from);
1176 if (err <= 0)
ea9b9907
NP
1177 goto out;
1178
5fa8e0a1 1179 err = file_remove_privs(file);
ea9b9907
NP
1180 if (err)
1181 goto out;
1182
c3b2da31
JB
1183 err = file_update_time(file);
1184 if (err)
1185 goto out;
ea9b9907 1186
2ba48ce5 1187 if (iocb->ki_flags & IOCB_DIRECT) {
3309dd04 1188 loff_t pos = iocb->ki_pos;
1af5bb49 1189 written = generic_file_direct_write(iocb, from);
84c3d55c 1190 if (written < 0 || !iov_iter_count(from))
4273b793
AA
1191 goto out;
1192
1193 pos += written;
ea9b9907 1194
84c3d55c 1195 written_buffered = fuse_perform_write(file, mapping, from, pos);
4273b793
AA
1196 if (written_buffered < 0) {
1197 err = written_buffered;
1198 goto out;
1199 }
1200 endbyte = pos + written_buffered - 1;
1201
1202 err = filemap_write_and_wait_range(file->f_mapping, pos,
1203 endbyte);
1204 if (err)
1205 goto out;
1206
1207 invalidate_mapping_pages(file->f_mapping,
09cbfeaf
KS
1208 pos >> PAGE_SHIFT,
1209 endbyte >> PAGE_SHIFT);
4273b793
AA
1210
1211 written += written_buffered;
1212 iocb->ki_pos = pos + written_buffered;
1213 } else {
3309dd04 1214 written = fuse_perform_write(file, mapping, from, iocb->ki_pos);
4273b793 1215 if (written >= 0)
3309dd04 1216 iocb->ki_pos += written;
4273b793 1217 }
ea9b9907
NP
1218out:
1219 current->backing_dev_info = NULL;
5955102c 1220 inode_unlock(inode);
ea9b9907
NP
1221
1222 return written ? written : err;
1223}
1224
7c190c8b
MP
1225static inline void fuse_page_descs_length_init(struct fuse_req *req,
1226 unsigned index, unsigned nr_pages)
85f40aec
MP
1227{
1228 int i;
1229
7c190c8b 1230 for (i = index; i < index + nr_pages; i++)
85f40aec
MP
1231 req->page_descs[i].length = PAGE_SIZE -
1232 req->page_descs[i].offset;
1233}
1234
7c190c8b
MP
1235static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii)
1236{
1237 return (unsigned long)ii->iov->iov_base + ii->iov_offset;
1238}
1239
1240static inline size_t fuse_get_frag_size(const struct iov_iter *ii,
1241 size_t max_size)
1242{
1243 return min(iov_iter_single_seg_count(ii), max_size);
1244}
1245
b98d023a 1246static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii,
ce60a2f1 1247 size_t *nbytesp, int write)
413ef8cb 1248{
7c190c8b 1249 size_t nbytes = 0; /* # bytes already packed in req */
742f9927 1250 ssize_t ret = 0;
b98d023a 1251
f4975c67 1252 /* Special case for kernel I/O: can copy directly into the buffer */
62a8067a 1253 if (ii->type & ITER_KVEC) {
7c190c8b
MP
1254 unsigned long user_addr = fuse_get_user_addr(ii);
1255 size_t frag_size = fuse_get_frag_size(ii, *nbytesp);
1256
f4975c67
MS
1257 if (write)
1258 req->in.args[1].value = (void *) user_addr;
1259 else
1260 req->out.args[0].value = (void *) user_addr;
1261
b98d023a
MP
1262 iov_iter_advance(ii, frag_size);
1263 *nbytesp = frag_size;
f4975c67
MS
1264 return 0;
1265 }
413ef8cb 1266
5565a9d8 1267 while (nbytes < *nbytesp && req->num_pages < req->max_pages) {
7c190c8b 1268 unsigned npages;
f67da30c 1269 size_t start;
742f9927 1270 ret = iov_iter_get_pages(ii, &req->pages[req->num_pages],
2c80929c 1271 *nbytesp - nbytes,
c7f3888a
AV
1272 req->max_pages - req->num_pages,
1273 &start);
7c190c8b 1274 if (ret < 0)
742f9927 1275 break;
7c190c8b 1276
c9c37e2e
AV
1277 iov_iter_advance(ii, ret);
1278 nbytes += ret;
7c190c8b 1279
c9c37e2e
AV
1280 ret += start;
1281 npages = (ret + PAGE_SIZE - 1) / PAGE_SIZE;
7c190c8b 1282
c9c37e2e 1283 req->page_descs[req->num_pages].offset = start;
7c190c8b
MP
1284 fuse_page_descs_length_init(req, req->num_pages, npages);
1285
1286 req->num_pages += npages;
1287 req->page_descs[req->num_pages - 1].length -=
c9c37e2e 1288 (PAGE_SIZE - ret) & (PAGE_SIZE - 1);
7c190c8b 1289 }
f4975c67
MS
1290
1291 if (write)
1292 req->in.argpages = 1;
1293 else
1294 req->out.argpages = 1;
1295
7c190c8b 1296 *nbytesp = nbytes;
f4975c67 1297
2c932d4c 1298 return ret < 0 ? ret : 0;
413ef8cb
MS
1299}
1300
5565a9d8
MP
1301static inline int fuse_iter_npages(const struct iov_iter *ii_p)
1302{
f67da30c 1303 return iov_iter_npages(ii_p, FUSE_MAX_PAGES_PER_REQ);
5565a9d8
MP
1304}
1305
d22a943f
AV
1306ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
1307 loff_t *ppos, int flags)
413ef8cb 1308{
ea8cd333
PE
1309 int write = flags & FUSE_DIO_WRITE;
1310 int cuse = flags & FUSE_DIO_CUSE;
36cf66ed 1311 struct file *file = io->file;
ea8cd333 1312 struct inode *inode = file->f_mapping->host;
2106cb18
MS
1313 struct fuse_file *ff = file->private_data;
1314 struct fuse_conn *fc = ff->fc;
413ef8cb
MS
1315 size_t nmax = write ? fc->max_write : fc->max_read;
1316 loff_t pos = *ppos;
d22a943f 1317 size_t count = iov_iter_count(iter);
09cbfeaf
KS
1318 pgoff_t idx_from = pos >> PAGE_SHIFT;
1319 pgoff_t idx_to = (pos + count - 1) >> PAGE_SHIFT;
413ef8cb 1320 ssize_t res = 0;
248d86e8 1321 struct fuse_req *req;
742f9927 1322 int err = 0;
248d86e8 1323
de82b923 1324 if (io->async)
d22a943f 1325 req = fuse_get_req_for_background(fc, fuse_iter_npages(iter));
de82b923 1326 else
d22a943f 1327 req = fuse_get_req(fc, fuse_iter_npages(iter));
ce1d5a49
MS
1328 if (IS_ERR(req))
1329 return PTR_ERR(req);
413ef8cb 1330
ea8cd333
PE
1331 if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) {
1332 if (!write)
5955102c 1333 inode_lock(inode);
ea8cd333
PE
1334 fuse_sync_writes(inode);
1335 if (!write)
5955102c 1336 inode_unlock(inode);
ea8cd333
PE
1337 }
1338
413ef8cb 1339 while (count) {
413ef8cb 1340 size_t nres;
2106cb18 1341 fl_owner_t owner = current->files;
f4975c67 1342 size_t nbytes = min(count, nmax);
742f9927
AS
1343 err = fuse_get_user_pages(req, iter, &nbytes, write);
1344 if (err && !nbytes)
413ef8cb 1345 break;
f4975c67 1346
413ef8cb 1347 if (write)
36cf66ed 1348 nres = fuse_send_write(req, io, pos, nbytes, owner);
413ef8cb 1349 else
36cf66ed 1350 nres = fuse_send_read(req, io, pos, nbytes, owner);
2106cb18 1351
36cf66ed
MP
1352 if (!io->async)
1353 fuse_release_user_pages(req, !write);
413ef8cb 1354 if (req->out.h.error) {
742f9927 1355 err = req->out.h.error;
413ef8cb
MS
1356 break;
1357 } else if (nres > nbytes) {
742f9927
AS
1358 res = 0;
1359 err = -EIO;
413ef8cb
MS
1360 break;
1361 }
1362 count -= nres;
1363 res += nres;
1364 pos += nres;
413ef8cb
MS
1365 if (nres != nbytes)
1366 break;
56cf34ff
MS
1367 if (count) {
1368 fuse_put_request(fc, req);
de82b923
BF
1369 if (io->async)
1370 req = fuse_get_req_for_background(fc,
d22a943f 1371 fuse_iter_npages(iter));
de82b923 1372 else
d22a943f 1373 req = fuse_get_req(fc, fuse_iter_npages(iter));
56cf34ff
MS
1374 if (IS_ERR(req))
1375 break;
1376 }
413ef8cb 1377 }
f60311d5
AA
1378 if (!IS_ERR(req))
1379 fuse_put_request(fc, req);
d09cb9d7 1380 if (res > 0)
413ef8cb 1381 *ppos = pos;
413ef8cb 1382
742f9927 1383 return res > 0 ? res : err;
413ef8cb 1384}
08cbf542 1385EXPORT_SYMBOL_GPL(fuse_direct_io);
413ef8cb 1386
36cf66ed 1387static ssize_t __fuse_direct_read(struct fuse_io_priv *io,
d22a943f
AV
1388 struct iov_iter *iter,
1389 loff_t *ppos)
413ef8cb 1390{
d09cb9d7 1391 ssize_t res;
36cf66ed 1392 struct file *file = io->file;
6131ffaa 1393 struct inode *inode = file_inode(file);
d09cb9d7
MS
1394
1395 if (is_bad_inode(inode))
1396 return -EIO;
1397
d22a943f 1398 res = fuse_direct_io(io, iter, ppos, 0);
d09cb9d7
MS
1399
1400 fuse_invalidate_attr(inode);
1401
1402 return res;
413ef8cb
MS
1403}
1404
15316263 1405static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to)
b98d023a 1406{
744742d6 1407 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb->ki_filp);
15316263 1408 return __fuse_direct_read(&io, to, &iocb->ki_pos);
b98d023a
MP
1409}
1410
15316263 1411static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
4273b793 1412{
15316263 1413 struct file *file = iocb->ki_filp;
6131ffaa 1414 struct inode *inode = file_inode(file);
744742d6 1415 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(file);
15316263 1416 ssize_t res;
4273b793
AA
1417
1418 if (is_bad_inode(inode))
1419 return -EIO;
1420
1421 /* Don't allow parallel writes to the same file */
5955102c 1422 inode_lock(inode);
3309dd04
AV
1423 res = generic_write_checks(iocb, from);
1424 if (res > 0)
812408fb 1425 res = fuse_direct_io(&io, from, &iocb->ki_pos, FUSE_DIO_WRITE);
812408fb 1426 fuse_invalidate_attr(inode);
bcba24cc 1427 if (res > 0)
15316263 1428 fuse_write_update_size(inode, iocb->ki_pos);
5955102c 1429 inode_unlock(inode);
4273b793
AA
1430
1431 return res;
1432}
1433
3be5a52b 1434static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
b6aeaded 1435{
385b1268
PE
1436 int i;
1437
1438 for (i = 0; i < req->num_pages; i++)
1439 __free_page(req->pages[i]);
8b284dc4
MS
1440
1441 if (req->ff)
1442 fuse_file_put(req->ff, false);
3be5a52b
MS
1443}
1444
1445static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
1446{
1447 struct inode *inode = req->inode;
1448 struct fuse_inode *fi = get_fuse_inode(inode);
de1414a6 1449 struct backing_dev_info *bdi = inode_to_bdi(inode);
385b1268 1450 int i;
3be5a52b
MS
1451
1452 list_del(&req->writepages_entry);
385b1268 1453 for (i = 0; i < req->num_pages; i++) {
93f78d88 1454 dec_wb_stat(&bdi->wb, WB_WRITEBACK);
11fb9989 1455 dec_node_page_state(req->pages[i], NR_WRITEBACK_TEMP);
93f78d88 1456 wb_writeout_inc(&bdi->wb);
385b1268 1457 }
3be5a52b
MS
1458 wake_up(&fi->page_waitq);
1459}
1460
1461/* Called under fc->lock, may release and reacquire it */
6eaf4782
MP
1462static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req,
1463 loff_t size)
b9ca67b2
MS
1464__releases(fc->lock)
1465__acquires(fc->lock)
3be5a52b
MS
1466{
1467 struct fuse_inode *fi = get_fuse_inode(req->inode);
3be5a52b 1468 struct fuse_write_in *inarg = &req->misc.write.in;
09cbfeaf 1469 __u64 data_size = req->num_pages * PAGE_SIZE;
3be5a52b
MS
1470
1471 if (!fc->connected)
1472 goto out_free;
1473
385b1268
PE
1474 if (inarg->offset + data_size <= size) {
1475 inarg->size = data_size;
3be5a52b 1476 } else if (inarg->offset < size) {
385b1268 1477 inarg->size = size - inarg->offset;
3be5a52b
MS
1478 } else {
1479 /* Got truncated off completely */
1480 goto out_free;
b6aeaded 1481 }
3be5a52b
MS
1482
1483 req->in.args[1].size = inarg->size;
1484 fi->writectr++;
b93f858a 1485 fuse_request_send_background_locked(fc, req);
3be5a52b
MS
1486 return;
1487
1488 out_free:
1489 fuse_writepage_finish(fc, req);
1490 spin_unlock(&fc->lock);
1491 fuse_writepage_free(fc, req);
e9bb09dd 1492 fuse_put_request(fc, req);
3be5a52b 1493 spin_lock(&fc->lock);
b6aeaded
MS
1494}
1495
3be5a52b
MS
1496/*
1497 * If fi->writectr is positive (no truncate or fsync going on) send
1498 * all queued writepage requests.
1499 *
1500 * Called with fc->lock
1501 */
1502void fuse_flush_writepages(struct inode *inode)
b9ca67b2
MS
1503__releases(fc->lock)
1504__acquires(fc->lock)
b6aeaded 1505{
3be5a52b
MS
1506 struct fuse_conn *fc = get_fuse_conn(inode);
1507 struct fuse_inode *fi = get_fuse_inode(inode);
6eaf4782 1508 size_t crop = i_size_read(inode);
3be5a52b
MS
1509 struct fuse_req *req;
1510
1511 while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
1512 req = list_entry(fi->queued_writes.next, struct fuse_req, list);
1513 list_del_init(&req->list);
6eaf4782 1514 fuse_send_writepage(fc, req, crop);
3be5a52b
MS
1515 }
1516}
1517
1518static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
1519{
1520 struct inode *inode = req->inode;
1521 struct fuse_inode *fi = get_fuse_inode(inode);
1522
1523 mapping_set_error(inode->i_mapping, req->out.h.error);
1524 spin_lock(&fc->lock);
8b284dc4 1525 while (req->misc.write.next) {
6eaf4782
MP
1526 struct fuse_conn *fc = get_fuse_conn(inode);
1527 struct fuse_write_in *inarg = &req->misc.write.in;
8b284dc4
MS
1528 struct fuse_req *next = req->misc.write.next;
1529 req->misc.write.next = next->misc.write.next;
1530 next->misc.write.next = NULL;
ce128de6 1531 next->ff = fuse_file_get(req->ff);
8b284dc4 1532 list_add(&next->writepages_entry, &fi->writepages);
6eaf4782
MP
1533
1534 /*
1535 * Skip fuse_flush_writepages() to make it easy to crop requests
1536 * based on primary request size.
1537 *
1538 * 1st case (trivial): there are no concurrent activities using
1539 * fuse_set/release_nowrite. Then we're on safe side because
1540 * fuse_flush_writepages() would call fuse_send_writepage()
1541 * anyway.
1542 *
1543 * 2nd case: someone called fuse_set_nowrite and it is waiting
1544 * now for completion of all in-flight requests. This happens
1545 * rarely and no more than once per page, so this should be
1546 * okay.
1547 *
1548 * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle
1549 * of fuse_set_nowrite..fuse_release_nowrite section. The fact
1550 * that fuse_set_nowrite returned implies that all in-flight
1551 * requests were completed along with all of their secondary
1552 * requests. Further primary requests are blocked by negative
1553 * writectr. Hence there cannot be any in-flight requests and
1554 * no invocations of fuse_writepage_end() while we're in
1555 * fuse_set_nowrite..fuse_release_nowrite section.
1556 */
1557 fuse_send_writepage(fc, next, inarg->offset + inarg->size);
8b284dc4 1558 }
3be5a52b
MS
1559 fi->writectr--;
1560 fuse_writepage_finish(fc, req);
1561 spin_unlock(&fc->lock);
1562 fuse_writepage_free(fc, req);
1563}
1564
1e18bda8
MS
1565static struct fuse_file *__fuse_write_file_get(struct fuse_conn *fc,
1566 struct fuse_inode *fi)
adcadfa8 1567{
72523425 1568 struct fuse_file *ff = NULL;
adcadfa8
PE
1569
1570 spin_lock(&fc->lock);
1e18bda8 1571 if (!list_empty(&fi->write_files)) {
72523425
MS
1572 ff = list_entry(fi->write_files.next, struct fuse_file,
1573 write_entry);
1574 fuse_file_get(ff);
1575 }
adcadfa8
PE
1576 spin_unlock(&fc->lock);
1577
1578 return ff;
1579}
1580
1e18bda8
MS
1581static struct fuse_file *fuse_write_file_get(struct fuse_conn *fc,
1582 struct fuse_inode *fi)
1583{
1584 struct fuse_file *ff = __fuse_write_file_get(fc, fi);
1585 WARN_ON(!ff);
1586 return ff;
1587}
1588
1589int fuse_write_inode(struct inode *inode, struct writeback_control *wbc)
1590{
1591 struct fuse_conn *fc = get_fuse_conn(inode);
1592 struct fuse_inode *fi = get_fuse_inode(inode);
1593 struct fuse_file *ff;
1594 int err;
1595
1596 ff = __fuse_write_file_get(fc, fi);
ab9e13f7 1597 err = fuse_flush_times(inode, ff);
1e18bda8
MS
1598 if (ff)
1599 fuse_file_put(ff, 0);
1600
1601 return err;
1602}
1603
3be5a52b
MS
1604static int fuse_writepage_locked(struct page *page)
1605{
1606 struct address_space *mapping = page->mapping;
1607 struct inode *inode = mapping->host;
1608 struct fuse_conn *fc = get_fuse_conn(inode);
1609 struct fuse_inode *fi = get_fuse_inode(inode);
1610 struct fuse_req *req;
3be5a52b 1611 struct page *tmp_page;
72523425 1612 int error = -ENOMEM;
3be5a52b
MS
1613
1614 set_page_writeback(page);
1615
4250c066 1616 req = fuse_request_alloc_nofs(1);
3be5a52b
MS
1617 if (!req)
1618 goto err;
1619
825d6d33
MS
1620 /* writeback always goes to bg_queue */
1621 __set_bit(FR_BACKGROUND, &req->flags);
3be5a52b
MS
1622 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1623 if (!tmp_page)
1624 goto err_free;
1625
72523425 1626 error = -EIO;
26d614df 1627 req->ff = fuse_write_file_get(fc, fi);
72523425 1628 if (!req->ff)
27f1b363 1629 goto err_nofile;
72523425 1630
adcadfa8 1631 fuse_write_fill(req, req->ff, page_offset(page), 0);
3be5a52b
MS
1632
1633 copy_highpage(tmp_page, page);
2d698b07 1634 req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
8b284dc4 1635 req->misc.write.next = NULL;
f4975c67 1636 req->in.argpages = 1;
3be5a52b
MS
1637 req->num_pages = 1;
1638 req->pages[0] = tmp_page;
b2430d75 1639 req->page_descs[0].offset = 0;
85f40aec 1640 req->page_descs[0].length = PAGE_SIZE;
3be5a52b
MS
1641 req->end = fuse_writepage_end;
1642 req->inode = inode;
1643
93f78d88 1644 inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
11fb9989 1645 inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP);
3be5a52b
MS
1646
1647 spin_lock(&fc->lock);
1648 list_add(&req->writepages_entry, &fi->writepages);
1649 list_add_tail(&req->list, &fi->queued_writes);
1650 fuse_flush_writepages(inode);
1651 spin_unlock(&fc->lock);
1652
4a4ac4eb
MP
1653 end_page_writeback(page);
1654
3be5a52b
MS
1655 return 0;
1656
27f1b363
MP
1657err_nofile:
1658 __free_page(tmp_page);
3be5a52b
MS
1659err_free:
1660 fuse_request_free(req);
1661err:
1662 end_page_writeback(page);
72523425 1663 return error;
3be5a52b
MS
1664}
1665
1666static int fuse_writepage(struct page *page, struct writeback_control *wbc)
1667{
1668 int err;
1669
ff17be08
MS
1670 if (fuse_page_is_writeback(page->mapping->host, page->index)) {
1671 /*
1672 * ->writepages() should be called for sync() and friends. We
1673 * should only get here on direct reclaim and then we are
1674 * allowed to skip a page which is already in flight
1675 */
1676 WARN_ON(wbc->sync_mode == WB_SYNC_ALL);
1677
1678 redirty_page_for_writepage(wbc, page);
1679 return 0;
1680 }
1681
3be5a52b
MS
1682 err = fuse_writepage_locked(page);
1683 unlock_page(page);
1684
1685 return err;
1686}
1687
26d614df
PE
1688struct fuse_fill_wb_data {
1689 struct fuse_req *req;
1690 struct fuse_file *ff;
1691 struct inode *inode;
2d033eaa 1692 struct page **orig_pages;
26d614df
PE
1693};
1694
1695static void fuse_writepages_send(struct fuse_fill_wb_data *data)
1696{
1697 struct fuse_req *req = data->req;
1698 struct inode *inode = data->inode;
1699 struct fuse_conn *fc = get_fuse_conn(inode);
1700 struct fuse_inode *fi = get_fuse_inode(inode);
2d033eaa
MP
1701 int num_pages = req->num_pages;
1702 int i;
26d614df
PE
1703
1704 req->ff = fuse_file_get(data->ff);
1705 spin_lock(&fc->lock);
1706 list_add_tail(&req->list, &fi->queued_writes);
1707 fuse_flush_writepages(inode);
1708 spin_unlock(&fc->lock);
2d033eaa
MP
1709
1710 for (i = 0; i < num_pages; i++)
1711 end_page_writeback(data->orig_pages[i]);
26d614df
PE
1712}
1713
8b284dc4
MS
1714static bool fuse_writepage_in_flight(struct fuse_req *new_req,
1715 struct page *page)
1716{
1717 struct fuse_conn *fc = get_fuse_conn(new_req->inode);
1718 struct fuse_inode *fi = get_fuse_inode(new_req->inode);
1719 struct fuse_req *tmp;
1720 struct fuse_req *old_req;
1721 bool found = false;
1722 pgoff_t curr_index;
1723
1724 BUG_ON(new_req->num_pages != 0);
1725
1726 spin_lock(&fc->lock);
1727 list_del(&new_req->writepages_entry);
8b284dc4
MS
1728 list_for_each_entry(old_req, &fi->writepages, writepages_entry) {
1729 BUG_ON(old_req->inode != new_req->inode);
09cbfeaf 1730 curr_index = old_req->misc.write.in.offset >> PAGE_SHIFT;
8b284dc4
MS
1731 if (curr_index <= page->index &&
1732 page->index < curr_index + old_req->num_pages) {
1733 found = true;
1734 break;
1735 }
1736 }
f6011081
MP
1737 if (!found) {
1738 list_add(&new_req->writepages_entry, &fi->writepages);
8b284dc4 1739 goto out_unlock;
f6011081 1740 }
8b284dc4 1741
f6011081 1742 new_req->num_pages = 1;
8b284dc4
MS
1743 for (tmp = old_req; tmp != NULL; tmp = tmp->misc.write.next) {
1744 BUG_ON(tmp->inode != new_req->inode);
09cbfeaf 1745 curr_index = tmp->misc.write.in.offset >> PAGE_SHIFT;
8b284dc4
MS
1746 if (tmp->num_pages == 1 &&
1747 curr_index == page->index) {
1748 old_req = tmp;
1749 }
1750 }
1751
33e14b4d 1752 if (old_req->num_pages == 1 && test_bit(FR_PENDING, &old_req->flags)) {
de1414a6 1753 struct backing_dev_info *bdi = inode_to_bdi(page->mapping->host);
41b6e41f 1754
8b284dc4
MS
1755 copy_highpage(old_req->pages[0], page);
1756 spin_unlock(&fc->lock);
1757
93f78d88 1758 dec_wb_stat(&bdi->wb, WB_WRITEBACK);
11fb9989 1759 dec_node_page_state(page, NR_WRITEBACK_TEMP);
93f78d88 1760 wb_writeout_inc(&bdi->wb);
8b284dc4
MS
1761 fuse_writepage_free(fc, new_req);
1762 fuse_request_free(new_req);
1763 goto out;
1764 } else {
1765 new_req->misc.write.next = old_req->misc.write.next;
1766 old_req->misc.write.next = new_req;
1767 }
1768out_unlock:
1769 spin_unlock(&fc->lock);
1770out:
1771 return found;
1772}
1773
26d614df
PE
1774static int fuse_writepages_fill(struct page *page,
1775 struct writeback_control *wbc, void *_data)
1776{
1777 struct fuse_fill_wb_data *data = _data;
1778 struct fuse_req *req = data->req;
1779 struct inode *inode = data->inode;
1780 struct fuse_conn *fc = get_fuse_conn(inode);
1781 struct page *tmp_page;
8b284dc4 1782 bool is_writeback;
26d614df
PE
1783 int err;
1784
1785 if (!data->ff) {
1786 err = -EIO;
1787 data->ff = fuse_write_file_get(fc, get_fuse_inode(inode));
1788 if (!data->ff)
1789 goto out_unlock;
1790 }
1791
8b284dc4
MS
1792 /*
1793 * Being under writeback is unlikely but possible. For example direct
1794 * read to an mmaped fuse file will set the page dirty twice; once when
1795 * the pages are faulted with get_user_pages(), and then after the read
1796 * completed.
1797 */
1798 is_writeback = fuse_page_is_writeback(inode, page->index);
1799
1800 if (req && req->num_pages &&
1801 (is_writeback || req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
09cbfeaf 1802 (req->num_pages + 1) * PAGE_SIZE > fc->max_write ||
8b284dc4
MS
1803 data->orig_pages[req->num_pages - 1]->index + 1 != page->index)) {
1804 fuse_writepages_send(data);
1805 data->req = NULL;
26d614df
PE
1806 }
1807 err = -ENOMEM;
1808 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1809 if (!tmp_page)
1810 goto out_unlock;
1811
1812 /*
1813 * The page must not be redirtied until the writeout is completed
1814 * (i.e. userspace has sent a reply to the write request). Otherwise
1815 * there could be more than one temporary page instance for each real
1816 * page.
1817 *
1818 * This is ensured by holding the page lock in page_mkwrite() while
1819 * checking fuse_page_is_writeback(). We already hold the page lock
1820 * since clear_page_dirty_for_io() and keep it held until we add the
1821 * request to the fi->writepages list and increment req->num_pages.
1822 * After this fuse_page_is_writeback() will indicate that the page is
1823 * under writeback, so we can release the page lock.
1824 */
1825 if (data->req == NULL) {
1826 struct fuse_inode *fi = get_fuse_inode(inode);
1827
1828 err = -ENOMEM;
1829 req = fuse_request_alloc_nofs(FUSE_MAX_PAGES_PER_REQ);
1830 if (!req) {
1831 __free_page(tmp_page);
1832 goto out_unlock;
1833 }
1834
1835 fuse_write_fill(req, data->ff, page_offset(page), 0);
1836 req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
8b284dc4 1837 req->misc.write.next = NULL;
26d614df 1838 req->in.argpages = 1;
825d6d33 1839 __set_bit(FR_BACKGROUND, &req->flags);
26d614df
PE
1840 req->num_pages = 0;
1841 req->end = fuse_writepage_end;
1842 req->inode = inode;
1843
1844 spin_lock(&fc->lock);
1845 list_add(&req->writepages_entry, &fi->writepages);
1846 spin_unlock(&fc->lock);
1847
1848 data->req = req;
1849 }
1850 set_page_writeback(page);
1851
1852 copy_highpage(tmp_page, page);
1853 req->pages[req->num_pages] = tmp_page;
1854 req->page_descs[req->num_pages].offset = 0;
1855 req->page_descs[req->num_pages].length = PAGE_SIZE;
1856
93f78d88 1857 inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
11fb9989 1858 inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP);
8b284dc4
MS
1859
1860 err = 0;
1861 if (is_writeback && fuse_writepage_in_flight(req, page)) {
1862 end_page_writeback(page);
1863 data->req = NULL;
1864 goto out_unlock;
1865 }
2d033eaa 1866 data->orig_pages[req->num_pages] = page;
26d614df
PE
1867
1868 /*
1869 * Protected by fc->lock against concurrent access by
1870 * fuse_page_is_writeback().
1871 */
1872 spin_lock(&fc->lock);
1873 req->num_pages++;
1874 spin_unlock(&fc->lock);
1875
26d614df
PE
1876out_unlock:
1877 unlock_page(page);
1878
1879 return err;
1880}
1881
1882static int fuse_writepages(struct address_space *mapping,
1883 struct writeback_control *wbc)
1884{
1885 struct inode *inode = mapping->host;
1886 struct fuse_fill_wb_data data;
1887 int err;
1888
1889 err = -EIO;
1890 if (is_bad_inode(inode))
1891 goto out;
1892
1893 data.inode = inode;
1894 data.req = NULL;
1895 data.ff = NULL;
1896
2d033eaa 1897 err = -ENOMEM;
f2b3455e
FF
1898 data.orig_pages = kcalloc(FUSE_MAX_PAGES_PER_REQ,
1899 sizeof(struct page *),
2d033eaa
MP
1900 GFP_NOFS);
1901 if (!data.orig_pages)
1902 goto out;
1903
26d614df
PE
1904 err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data);
1905 if (data.req) {
1906 /* Ignore errors if we can write at least one page */
1907 BUG_ON(!data.req->num_pages);
1908 fuse_writepages_send(&data);
1909 err = 0;
1910 }
1911 if (data.ff)
1912 fuse_file_put(data.ff, false);
2d033eaa
MP
1913
1914 kfree(data.orig_pages);
26d614df
PE
1915out:
1916 return err;
1917}
1918
6b12c1b3
PE
1919/*
1920 * It's worthy to make sure that space is reserved on disk for the write,
1921 * but how to implement it without killing performance need more thinking.
1922 */
1923static int fuse_write_begin(struct file *file, struct address_space *mapping,
1924 loff_t pos, unsigned len, unsigned flags,
1925 struct page **pagep, void **fsdata)
1926{
09cbfeaf 1927 pgoff_t index = pos >> PAGE_SHIFT;
a455589f 1928 struct fuse_conn *fc = get_fuse_conn(file_inode(file));
6b12c1b3
PE
1929 struct page *page;
1930 loff_t fsize;
1931 int err = -ENOMEM;
1932
1933 WARN_ON(!fc->writeback_cache);
1934
1935 page = grab_cache_page_write_begin(mapping, index, flags);
1936 if (!page)
1937 goto error;
1938
1939 fuse_wait_on_page_writeback(mapping->host, page->index);
1940
09cbfeaf 1941 if (PageUptodate(page) || len == PAGE_SIZE)
6b12c1b3
PE
1942 goto success;
1943 /*
1944 * Check if the start this page comes after the end of file, in which
1945 * case the readpage can be optimized away.
1946 */
1947 fsize = i_size_read(mapping->host);
09cbfeaf
KS
1948 if (fsize <= (pos & PAGE_MASK)) {
1949 size_t off = pos & ~PAGE_MASK;
6b12c1b3
PE
1950 if (off)
1951 zero_user_segment(page, 0, off);
1952 goto success;
1953 }
1954 err = fuse_do_readpage(file, page);
1955 if (err)
1956 goto cleanup;
1957success:
1958 *pagep = page;
1959 return 0;
1960
1961cleanup:
1962 unlock_page(page);
09cbfeaf 1963 put_page(page);
6b12c1b3
PE
1964error:
1965 return err;
1966}
1967
1968static int fuse_write_end(struct file *file, struct address_space *mapping,
1969 loff_t pos, unsigned len, unsigned copied,
1970 struct page *page, void *fsdata)
1971{
1972 struct inode *inode = page->mapping->host;
1973
1974 if (!PageUptodate(page)) {
1975 /* Zero any unwritten bytes at the end of the page */
09cbfeaf 1976 size_t endoff = (pos + copied) & ~PAGE_MASK;
6b12c1b3 1977 if (endoff)
09cbfeaf 1978 zero_user_segment(page, endoff, PAGE_SIZE);
6b12c1b3
PE
1979 SetPageUptodate(page);
1980 }
1981
1982 fuse_write_update_size(inode, pos + copied);
1983 set_page_dirty(page);
1984 unlock_page(page);
09cbfeaf 1985 put_page(page);
6b12c1b3
PE
1986
1987 return copied;
1988}
1989
3be5a52b
MS
1990static int fuse_launder_page(struct page *page)
1991{
1992 int err = 0;
1993 if (clear_page_dirty_for_io(page)) {
1994 struct inode *inode = page->mapping->host;
1995 err = fuse_writepage_locked(page);
1996 if (!err)
1997 fuse_wait_on_page_writeback(inode, page->index);
1998 }
1999 return err;
2000}
2001
2002/*
2003 * Write back dirty pages now, because there may not be any suitable
2004 * open files later
2005 */
2006static void fuse_vma_close(struct vm_area_struct *vma)
2007{
2008 filemap_write_and_wait(vma->vm_file->f_mapping);
2009}
2010
2011/*
2012 * Wait for writeback against this page to complete before allowing it
2013 * to be marked dirty again, and hence written back again, possibly
2014 * before the previous writepage completed.
2015 *
2016 * Block here, instead of in ->writepage(), so that the userspace fs
2017 * can only block processes actually operating on the filesystem.
2018 *
2019 * Otherwise unprivileged userspace fs would be able to block
2020 * unrelated:
2021 *
2022 * - page migration
2023 * - sync(2)
2024 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
2025 */
c2ec175c 2026static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
3be5a52b 2027{
c2ec175c 2028 struct page *page = vmf->page;
cca24370
MS
2029 struct inode *inode = file_inode(vma->vm_file);
2030
2031 file_update_time(vma->vm_file);
2032 lock_page(page);
2033 if (page->mapping != inode->i_mapping) {
2034 unlock_page(page);
2035 return VM_FAULT_NOPAGE;
2036 }
3be5a52b
MS
2037
2038 fuse_wait_on_page_writeback(inode, page->index);
cca24370 2039 return VM_FAULT_LOCKED;
3be5a52b
MS
2040}
2041
f0f37e2f 2042static const struct vm_operations_struct fuse_file_vm_ops = {
3be5a52b
MS
2043 .close = fuse_vma_close,
2044 .fault = filemap_fault,
f1820361 2045 .map_pages = filemap_map_pages,
3be5a52b
MS
2046 .page_mkwrite = fuse_page_mkwrite,
2047};
2048
2049static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
2050{
650b22b9
PE
2051 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2052 fuse_link_write_file(file);
2053
3be5a52b
MS
2054 file_accessed(file);
2055 vma->vm_ops = &fuse_file_vm_ops;
b6aeaded
MS
2056 return 0;
2057}
2058
fc280c96
MS
2059static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma)
2060{
2061 /* Can't provide the coherency needed for MAP_SHARED */
2062 if (vma->vm_flags & VM_MAYSHARE)
2063 return -ENODEV;
2064
3121bfe7
MS
2065 invalidate_inode_pages2(file->f_mapping);
2066
fc280c96
MS
2067 return generic_file_mmap(file, vma);
2068}
2069
71421259
MS
2070static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
2071 struct file_lock *fl)
2072{
2073 switch (ffl->type) {
2074 case F_UNLCK:
2075 break;
2076
2077 case F_RDLCK:
2078 case F_WRLCK:
2079 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
2080 ffl->end < ffl->start)
2081 return -EIO;
2082
2083 fl->fl_start = ffl->start;
2084 fl->fl_end = ffl->end;
2085 fl->fl_pid = ffl->pid;
2086 break;
2087
2088 default:
2089 return -EIO;
2090 }
2091 fl->fl_type = ffl->type;
2092 return 0;
2093}
2094
7078187a 2095static void fuse_lk_fill(struct fuse_args *args, struct file *file,
a9ff4f87 2096 const struct file_lock *fl, int opcode, pid_t pid,
7078187a 2097 int flock, struct fuse_lk_in *inarg)
71421259 2098{
6131ffaa 2099 struct inode *inode = file_inode(file);
9c8ef561 2100 struct fuse_conn *fc = get_fuse_conn(inode);
71421259 2101 struct fuse_file *ff = file->private_data;
7078187a
MS
2102
2103 memset(inarg, 0, sizeof(*inarg));
2104 inarg->fh = ff->fh;
2105 inarg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
2106 inarg->lk.start = fl->fl_start;
2107 inarg->lk.end = fl->fl_end;
2108 inarg->lk.type = fl->fl_type;
2109 inarg->lk.pid = pid;
a9ff4f87 2110 if (flock)
7078187a
MS
2111 inarg->lk_flags |= FUSE_LK_FLOCK;
2112 args->in.h.opcode = opcode;
2113 args->in.h.nodeid = get_node_id(inode);
2114 args->in.numargs = 1;
2115 args->in.args[0].size = sizeof(*inarg);
2116 args->in.args[0].value = inarg;
71421259
MS
2117}
2118
2119static int fuse_getlk(struct file *file, struct file_lock *fl)
2120{
6131ffaa 2121 struct inode *inode = file_inode(file);
71421259 2122 struct fuse_conn *fc = get_fuse_conn(inode);
7078187a
MS
2123 FUSE_ARGS(args);
2124 struct fuse_lk_in inarg;
71421259
MS
2125 struct fuse_lk_out outarg;
2126 int err;
2127
7078187a
MS
2128 fuse_lk_fill(&args, file, fl, FUSE_GETLK, 0, 0, &inarg);
2129 args.out.numargs = 1;
2130 args.out.args[0].size = sizeof(outarg);
2131 args.out.args[0].value = &outarg;
2132 err = fuse_simple_request(fc, &args);
71421259
MS
2133 if (!err)
2134 err = convert_fuse_file_lock(&outarg.lk, fl);
2135
2136 return err;
2137}
2138
a9ff4f87 2139static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
71421259 2140{
6131ffaa 2141 struct inode *inode = file_inode(file);
71421259 2142 struct fuse_conn *fc = get_fuse_conn(inode);
7078187a
MS
2143 FUSE_ARGS(args);
2144 struct fuse_lk_in inarg;
71421259
MS
2145 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
2146 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
2147 int err;
2148
8fb47a4f 2149 if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
48e90761
MS
2150 /* NLM needs asynchronous locks, which we don't support yet */
2151 return -ENOLCK;
2152 }
2153
71421259
MS
2154 /* Unlock on close is handled by the flush method */
2155 if (fl->fl_flags & FL_CLOSE)
2156 return 0;
2157
7078187a
MS
2158 fuse_lk_fill(&args, file, fl, opcode, pid, flock, &inarg);
2159 err = fuse_simple_request(fc, &args);
71421259 2160
a4d27e75
MS
2161 /* locking is restartable */
2162 if (err == -EINTR)
2163 err = -ERESTARTSYS;
7078187a 2164
71421259
MS
2165 return err;
2166}
2167
2168static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
2169{
6131ffaa 2170 struct inode *inode = file_inode(file);
71421259
MS
2171 struct fuse_conn *fc = get_fuse_conn(inode);
2172 int err;
2173
48e90761
MS
2174 if (cmd == F_CANCELLK) {
2175 err = 0;
2176 } else if (cmd == F_GETLK) {
71421259 2177 if (fc->no_lock) {
9d6a8c5c 2178 posix_test_lock(file, fl);
71421259
MS
2179 err = 0;
2180 } else
2181 err = fuse_getlk(file, fl);
2182 } else {
2183 if (fc->no_lock)
48e90761 2184 err = posix_lock_file(file, fl, NULL);
71421259 2185 else
a9ff4f87 2186 err = fuse_setlk(file, fl, 0);
71421259
MS
2187 }
2188 return err;
2189}
2190
a9ff4f87
MS
2191static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
2192{
6131ffaa 2193 struct inode *inode = file_inode(file);
a9ff4f87
MS
2194 struct fuse_conn *fc = get_fuse_conn(inode);
2195 int err;
2196
37fb3a30 2197 if (fc->no_flock) {
4f656367 2198 err = locks_lock_file_wait(file, fl);
a9ff4f87 2199 } else {
37fb3a30
MS
2200 struct fuse_file *ff = file->private_data;
2201
a9ff4f87 2202 /* emulate flock with POSIX locks */
37fb3a30 2203 ff->flock = true;
a9ff4f87
MS
2204 err = fuse_setlk(file, fl, 1);
2205 }
2206
2207 return err;
2208}
2209
b2d2272f
MS
2210static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
2211{
2212 struct inode *inode = mapping->host;
2213 struct fuse_conn *fc = get_fuse_conn(inode);
7078187a 2214 FUSE_ARGS(args);
b2d2272f
MS
2215 struct fuse_bmap_in inarg;
2216 struct fuse_bmap_out outarg;
2217 int err;
2218
2219 if (!inode->i_sb->s_bdev || fc->no_bmap)
2220 return 0;
2221
b2d2272f
MS
2222 memset(&inarg, 0, sizeof(inarg));
2223 inarg.block = block;
2224 inarg.blocksize = inode->i_sb->s_blocksize;
7078187a
MS
2225 args.in.h.opcode = FUSE_BMAP;
2226 args.in.h.nodeid = get_node_id(inode);
2227 args.in.numargs = 1;
2228 args.in.args[0].size = sizeof(inarg);
2229 args.in.args[0].value = &inarg;
2230 args.out.numargs = 1;
2231 args.out.args[0].size = sizeof(outarg);
2232 args.out.args[0].value = &outarg;
2233 err = fuse_simple_request(fc, &args);
b2d2272f
MS
2234 if (err == -ENOSYS)
2235 fc->no_bmap = 1;
2236
2237 return err ? 0 : outarg.block;
2238}
2239
0b5da8db
R
2240static loff_t fuse_lseek(struct file *file, loff_t offset, int whence)
2241{
2242 struct inode *inode = file->f_mapping->host;
2243 struct fuse_conn *fc = get_fuse_conn(inode);
2244 struct fuse_file *ff = file->private_data;
2245 FUSE_ARGS(args);
2246 struct fuse_lseek_in inarg = {
2247 .fh = ff->fh,
2248 .offset = offset,
2249 .whence = whence
2250 };
2251 struct fuse_lseek_out outarg;
2252 int err;
2253
2254 if (fc->no_lseek)
2255 goto fallback;
2256
2257 args.in.h.opcode = FUSE_LSEEK;
2258 args.in.h.nodeid = ff->nodeid;
2259 args.in.numargs = 1;
2260 args.in.args[0].size = sizeof(inarg);
2261 args.in.args[0].value = &inarg;
2262 args.out.numargs = 1;
2263 args.out.args[0].size = sizeof(outarg);
2264 args.out.args[0].value = &outarg;
2265 err = fuse_simple_request(fc, &args);
2266 if (err) {
2267 if (err == -ENOSYS) {
2268 fc->no_lseek = 1;
2269 goto fallback;
2270 }
2271 return err;
2272 }
2273
2274 return vfs_setpos(file, outarg.offset, inode->i_sb->s_maxbytes);
2275
2276fallback:
2277 err = fuse_update_attributes(inode, NULL, file, NULL);
2278 if (!err)
2279 return generic_file_llseek(file, offset, whence);
2280 else
2281 return err;
2282}
2283
965c8e59 2284static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence)
5559b8f4
MS
2285{
2286 loff_t retval;
6131ffaa 2287 struct inode *inode = file_inode(file);
5559b8f4 2288
0b5da8db
R
2289 switch (whence) {
2290 case SEEK_SET:
2291 case SEEK_CUR:
2292 /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */
965c8e59 2293 retval = generic_file_llseek(file, offset, whence);
0b5da8db
R
2294 break;
2295 case SEEK_END:
5955102c 2296 inode_lock(inode);
0b5da8db
R
2297 retval = fuse_update_attributes(inode, NULL, file, NULL);
2298 if (!retval)
2299 retval = generic_file_llseek(file, offset, whence);
5955102c 2300 inode_unlock(inode);
0b5da8db
R
2301 break;
2302 case SEEK_HOLE:
2303 case SEEK_DATA:
5955102c 2304 inode_lock(inode);
0b5da8db 2305 retval = fuse_lseek(file, offset, whence);
5955102c 2306 inode_unlock(inode);
0b5da8db
R
2307 break;
2308 default:
2309 retval = -EINVAL;
2310 }
c07c3d19 2311
5559b8f4
MS
2312 return retval;
2313}
2314
59efec7b
TH
2315static int fuse_ioctl_copy_user(struct page **pages, struct iovec *iov,
2316 unsigned int nr_segs, size_t bytes, bool to_user)
2317{
2318 struct iov_iter ii;
2319 int page_idx = 0;
2320
2321 if (!bytes)
2322 return 0;
2323
71d8e532 2324 iov_iter_init(&ii, to_user ? READ : WRITE, iov, nr_segs, bytes);
59efec7b
TH
2325
2326 while (iov_iter_count(&ii)) {
2327 struct page *page = pages[page_idx++];
2328 size_t todo = min_t(size_t, PAGE_SIZE, iov_iter_count(&ii));
4aa0edd2 2329 void *kaddr;
59efec7b 2330
4aa0edd2 2331 kaddr = kmap(page);
59efec7b
TH
2332
2333 while (todo) {
2334 char __user *uaddr = ii.iov->iov_base + ii.iov_offset;
2335 size_t iov_len = ii.iov->iov_len - ii.iov_offset;
2336 size_t copy = min(todo, iov_len);
2337 size_t left;
2338
2339 if (!to_user)
2340 left = copy_from_user(kaddr, uaddr, copy);
2341 else
2342 left = copy_to_user(uaddr, kaddr, copy);
2343
2344 if (unlikely(left))
2345 return -EFAULT;
2346
2347 iov_iter_advance(&ii, copy);
2348 todo -= copy;
2349 kaddr += copy;
2350 }
2351
0bd87182 2352 kunmap(page);
59efec7b
TH
2353 }
2354
2355 return 0;
2356}
2357
d9d318d3
MS
2358/*
2359 * CUSE servers compiled on 32bit broke on 64bit kernels because the
2360 * ABI was defined to be 'struct iovec' which is different on 32bit
2361 * and 64bit. Fortunately we can determine which structure the server
2362 * used from the size of the reply.
2363 */
1baa26b2
MS
2364static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
2365 size_t transferred, unsigned count,
2366 bool is_compat)
d9d318d3
MS
2367{
2368#ifdef CONFIG_COMPAT
2369 if (count * sizeof(struct compat_iovec) == transferred) {
2370 struct compat_iovec *ciov = src;
2371 unsigned i;
2372
2373 /*
2374 * With this interface a 32bit server cannot support
2375 * non-compat (i.e. ones coming from 64bit apps) ioctl
2376 * requests
2377 */
2378 if (!is_compat)
2379 return -EINVAL;
2380
2381 for (i = 0; i < count; i++) {
2382 dst[i].iov_base = compat_ptr(ciov[i].iov_base);
2383 dst[i].iov_len = ciov[i].iov_len;
2384 }
2385 return 0;
2386 }
2387#endif
2388
2389 if (count * sizeof(struct iovec) != transferred)
2390 return -EIO;
2391
2392 memcpy(dst, src, transferred);
2393 return 0;
2394}
2395
7572777e
MS
2396/* Make sure iov_length() won't overflow */
2397static int fuse_verify_ioctl_iov(struct iovec *iov, size_t count)
2398{
2399 size_t n;
2400 u32 max = FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT;
2401
fb6ccff6 2402 for (n = 0; n < count; n++, iov++) {
7572777e
MS
2403 if (iov->iov_len > (size_t) max)
2404 return -ENOMEM;
2405 max -= iov->iov_len;
2406 }
2407 return 0;
2408}
2409
1baa26b2
MS
2410static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
2411 void *src, size_t transferred, unsigned count,
2412 bool is_compat)
2413{
2414 unsigned i;
2415 struct fuse_ioctl_iovec *fiov = src;
2416
2417 if (fc->minor < 16) {
2418 return fuse_copy_ioctl_iovec_old(dst, src, transferred,
2419 count, is_compat);
2420 }
2421
2422 if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
2423 return -EIO;
2424
2425 for (i = 0; i < count; i++) {
2426 /* Did the server supply an inappropriate value? */
2427 if (fiov[i].base != (unsigned long) fiov[i].base ||
2428 fiov[i].len != (unsigned long) fiov[i].len)
2429 return -EIO;
2430
2431 dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
2432 dst[i].iov_len = (size_t) fiov[i].len;
2433
2434#ifdef CONFIG_COMPAT
2435 if (is_compat &&
2436 (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
2437 (compat_size_t) dst[i].iov_len != fiov[i].len))
2438 return -EIO;
2439#endif
2440 }
2441
2442 return 0;
2443}
2444
2445
59efec7b
TH
2446/*
2447 * For ioctls, there is no generic way to determine how much memory
2448 * needs to be read and/or written. Furthermore, ioctls are allowed
2449 * to dereference the passed pointer, so the parameter requires deep
2450 * copying but FUSE has no idea whatsoever about what to copy in or
2451 * out.
2452 *
2453 * This is solved by allowing FUSE server to retry ioctl with
2454 * necessary in/out iovecs. Let's assume the ioctl implementation
2455 * needs to read in the following structure.
2456 *
2457 * struct a {
2458 * char *buf;
2459 * size_t buflen;
2460 * }
2461 *
2462 * On the first callout to FUSE server, inarg->in_size and
2463 * inarg->out_size will be NULL; then, the server completes the ioctl
2464 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
2465 * the actual iov array to
2466 *
2467 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } }
2468 *
2469 * which tells FUSE to copy in the requested area and retry the ioctl.
2470 * On the second round, the server has access to the structure and
2471 * from that it can tell what to look for next, so on the invocation,
2472 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
2473 *
2474 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) },
2475 * { .iov_base = a.buf, .iov_len = a.buflen } }
2476 *
2477 * FUSE will copy both struct a and the pointed buffer from the
2478 * process doing the ioctl and retry ioctl with both struct a and the
2479 * buffer.
2480 *
2481 * This time, FUSE server has everything it needs and completes ioctl
2482 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
2483 *
2484 * Copying data out works the same way.
2485 *
2486 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
2487 * automatically initializes in and out iovs by decoding @cmd with
2488 * _IOC_* macros and the server is not allowed to request RETRY. This
2489 * limits ioctl data transfers to well-formed ioctls and is the forced
2490 * behavior for all FUSE servers.
2491 */
08cbf542
TH
2492long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
2493 unsigned int flags)
59efec7b 2494{
59efec7b 2495 struct fuse_file *ff = file->private_data;
d36f2487 2496 struct fuse_conn *fc = ff->fc;
59efec7b
TH
2497 struct fuse_ioctl_in inarg = {
2498 .fh = ff->fh,
2499 .cmd = cmd,
2500 .arg = arg,
2501 .flags = flags
2502 };
2503 struct fuse_ioctl_out outarg;
2504 struct fuse_req *req = NULL;
2505 struct page **pages = NULL;
8ac83505 2506 struct iovec *iov_page = NULL;
59efec7b
TH
2507 struct iovec *in_iov = NULL, *out_iov = NULL;
2508 unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages;
2509 size_t in_size, out_size, transferred;
2510 int err;
2511
1baa26b2
MS
2512#if BITS_PER_LONG == 32
2513 inarg.flags |= FUSE_IOCTL_32BIT;
2514#else
2515 if (flags & FUSE_IOCTL_COMPAT)
2516 inarg.flags |= FUSE_IOCTL_32BIT;
2517#endif
2518
59efec7b 2519 /* assume all the iovs returned by client always fits in a page */
1baa26b2 2520 BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
59efec7b 2521
59efec7b 2522 err = -ENOMEM;
c411cc88 2523 pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, sizeof(pages[0]), GFP_KERNEL);
8ac83505 2524 iov_page = (struct iovec *) __get_free_page(GFP_KERNEL);
59efec7b
TH
2525 if (!pages || !iov_page)
2526 goto out;
2527
2528 /*
2529 * If restricted, initialize IO parameters as encoded in @cmd.
2530 * RETRY from server is not allowed.
2531 */
2532 if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
8ac83505 2533 struct iovec *iov = iov_page;
59efec7b 2534
c9f0523d 2535 iov->iov_base = (void __user *)arg;
59efec7b
TH
2536 iov->iov_len = _IOC_SIZE(cmd);
2537
2538 if (_IOC_DIR(cmd) & _IOC_WRITE) {
2539 in_iov = iov;
2540 in_iovs = 1;
2541 }
2542
2543 if (_IOC_DIR(cmd) & _IOC_READ) {
2544 out_iov = iov;
2545 out_iovs = 1;
2546 }
2547 }
2548
2549 retry:
2550 inarg.in_size = in_size = iov_length(in_iov, in_iovs);
2551 inarg.out_size = out_size = iov_length(out_iov, out_iovs);
2552
2553 /*
2554 * Out data can be used either for actual out data or iovs,
2555 * make sure there always is at least one page.
2556 */
2557 out_size = max_t(size_t, out_size, PAGE_SIZE);
2558 max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
2559
2560 /* make sure there are enough buffer pages and init request with them */
2561 err = -ENOMEM;
2562 if (max_pages > FUSE_MAX_PAGES_PER_REQ)
2563 goto out;
2564 while (num_pages < max_pages) {
2565 pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2566 if (!pages[num_pages])
2567 goto out;
2568 num_pages++;
2569 }
2570
54b96670 2571 req = fuse_get_req(fc, num_pages);
59efec7b
TH
2572 if (IS_ERR(req)) {
2573 err = PTR_ERR(req);
2574 req = NULL;
2575 goto out;
2576 }
2577 memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages);
2578 req->num_pages = num_pages;
7c190c8b 2579 fuse_page_descs_length_init(req, 0, req->num_pages);
59efec7b
TH
2580
2581 /* okay, let's send it to the client */
2582 req->in.h.opcode = FUSE_IOCTL;
d36f2487 2583 req->in.h.nodeid = ff->nodeid;
59efec7b
TH
2584 req->in.numargs = 1;
2585 req->in.args[0].size = sizeof(inarg);
2586 req->in.args[0].value = &inarg;
2587 if (in_size) {
2588 req->in.numargs++;
2589 req->in.args[1].size = in_size;
2590 req->in.argpages = 1;
2591
2592 err = fuse_ioctl_copy_user(pages, in_iov, in_iovs, in_size,
2593 false);
2594 if (err)
2595 goto out;
2596 }
2597
2598 req->out.numargs = 2;
2599 req->out.args[0].size = sizeof(outarg);
2600 req->out.args[0].value = &outarg;
2601 req->out.args[1].size = out_size;
2602 req->out.argpages = 1;
2603 req->out.argvar = 1;
2604
b93f858a 2605 fuse_request_send(fc, req);
59efec7b
TH
2606 err = req->out.h.error;
2607 transferred = req->out.args[1].size;
2608 fuse_put_request(fc, req);
2609 req = NULL;
2610 if (err)
2611 goto out;
2612
2613 /* did it ask for retry? */
2614 if (outarg.flags & FUSE_IOCTL_RETRY) {
8ac83505 2615 void *vaddr;
59efec7b
TH
2616
2617 /* no retry if in restricted mode */
2618 err = -EIO;
2619 if (!(flags & FUSE_IOCTL_UNRESTRICTED))
2620 goto out;
2621
2622 in_iovs = outarg.in_iovs;
2623 out_iovs = outarg.out_iovs;
2624
2625 /*
2626 * Make sure things are in boundary, separate checks
2627 * are to protect against overflow.
2628 */
2629 err = -ENOMEM;
2630 if (in_iovs > FUSE_IOCTL_MAX_IOV ||
2631 out_iovs > FUSE_IOCTL_MAX_IOV ||
2632 in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
2633 goto out;
2634
2408f6ef 2635 vaddr = kmap_atomic(pages[0]);
1baa26b2 2636 err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr,
d9d318d3
MS
2637 transferred, in_iovs + out_iovs,
2638 (flags & FUSE_IOCTL_COMPAT) != 0);
2408f6ef 2639 kunmap_atomic(vaddr);
d9d318d3
MS
2640 if (err)
2641 goto out;
59efec7b 2642
8ac83505 2643 in_iov = iov_page;
59efec7b
TH
2644 out_iov = in_iov + in_iovs;
2645
7572777e
MS
2646 err = fuse_verify_ioctl_iov(in_iov, in_iovs);
2647 if (err)
2648 goto out;
2649
2650 err = fuse_verify_ioctl_iov(out_iov, out_iovs);
2651 if (err)
2652 goto out;
2653
59efec7b
TH
2654 goto retry;
2655 }
2656
2657 err = -EIO;
2658 if (transferred > inarg.out_size)
2659 goto out;
2660
2661 err = fuse_ioctl_copy_user(pages, out_iov, out_iovs, transferred, true);
2662 out:
2663 if (req)
2664 fuse_put_request(fc, req);
8ac83505 2665 free_page((unsigned long) iov_page);
59efec7b
TH
2666 while (num_pages)
2667 __free_page(pages[--num_pages]);
2668 kfree(pages);
2669
2670 return err ? err : outarg.result;
2671}
08cbf542 2672EXPORT_SYMBOL_GPL(fuse_do_ioctl);
59efec7b 2673
b18da0c5
MS
2674long fuse_ioctl_common(struct file *file, unsigned int cmd,
2675 unsigned long arg, unsigned int flags)
d36f2487 2676{
6131ffaa 2677 struct inode *inode = file_inode(file);
d36f2487
MS
2678 struct fuse_conn *fc = get_fuse_conn(inode);
2679
c2132c1b 2680 if (!fuse_allow_current_process(fc))
d36f2487
MS
2681 return -EACCES;
2682
2683 if (is_bad_inode(inode))
2684 return -EIO;
2685
2686 return fuse_do_ioctl(file, cmd, arg, flags);
2687}
2688
59efec7b
TH
2689static long fuse_file_ioctl(struct file *file, unsigned int cmd,
2690 unsigned long arg)
2691{
b18da0c5 2692 return fuse_ioctl_common(file, cmd, arg, 0);
59efec7b
TH
2693}
2694
2695static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
2696 unsigned long arg)
2697{
b18da0c5 2698 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
59efec7b
TH
2699}
2700
95668a69
TH
2701/*
2702 * All files which have been polled are linked to RB tree
2703 * fuse_conn->polled_files which is indexed by kh. Walk the tree and
2704 * find the matching one.
2705 */
2706static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh,
2707 struct rb_node **parent_out)
2708{
2709 struct rb_node **link = &fc->polled_files.rb_node;
2710 struct rb_node *last = NULL;
2711
2712 while (*link) {
2713 struct fuse_file *ff;
2714
2715 last = *link;
2716 ff = rb_entry(last, struct fuse_file, polled_node);
2717
2718 if (kh < ff->kh)
2719 link = &last->rb_left;
2720 else if (kh > ff->kh)
2721 link = &last->rb_right;
2722 else
2723 return link;
2724 }
2725
2726 if (parent_out)
2727 *parent_out = last;
2728 return link;
2729}
2730
2731/*
2732 * The file is about to be polled. Make sure it's on the polled_files
2733 * RB tree. Note that files once added to the polled_files tree are
2734 * not removed before the file is released. This is because a file
2735 * polled once is likely to be polled again.
2736 */
2737static void fuse_register_polled_file(struct fuse_conn *fc,
2738 struct fuse_file *ff)
2739{
2740 spin_lock(&fc->lock);
2741 if (RB_EMPTY_NODE(&ff->polled_node)) {
f3846266 2742 struct rb_node **link, *uninitialized_var(parent);
95668a69
TH
2743
2744 link = fuse_find_polled_node(fc, ff->kh, &parent);
2745 BUG_ON(*link);
2746 rb_link_node(&ff->polled_node, parent, link);
2747 rb_insert_color(&ff->polled_node, &fc->polled_files);
2748 }
2749 spin_unlock(&fc->lock);
2750}
2751
08cbf542 2752unsigned fuse_file_poll(struct file *file, poll_table *wait)
95668a69 2753{
95668a69 2754 struct fuse_file *ff = file->private_data;
797759aa 2755 struct fuse_conn *fc = ff->fc;
95668a69
TH
2756 struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh };
2757 struct fuse_poll_out outarg;
7078187a 2758 FUSE_ARGS(args);
95668a69
TH
2759 int err;
2760
2761 if (fc->no_poll)
2762 return DEFAULT_POLLMASK;
2763
2764 poll_wait(file, &ff->poll_wait, wait);
0415d291 2765 inarg.events = (__u32)poll_requested_events(wait);
95668a69
TH
2766
2767 /*
2768 * Ask for notification iff there's someone waiting for it.
2769 * The client may ignore the flag and always notify.
2770 */
2771 if (waitqueue_active(&ff->poll_wait)) {
2772 inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY;
2773 fuse_register_polled_file(fc, ff);
2774 }
2775
7078187a
MS
2776 args.in.h.opcode = FUSE_POLL;
2777 args.in.h.nodeid = ff->nodeid;
2778 args.in.numargs = 1;
2779 args.in.args[0].size = sizeof(inarg);
2780 args.in.args[0].value = &inarg;
2781 args.out.numargs = 1;
2782 args.out.args[0].size = sizeof(outarg);
2783 args.out.args[0].value = &outarg;
2784 err = fuse_simple_request(fc, &args);
95668a69
TH
2785
2786 if (!err)
2787 return outarg.revents;
2788 if (err == -ENOSYS) {
2789 fc->no_poll = 1;
2790 return DEFAULT_POLLMASK;
2791 }
2792 return POLLERR;
2793}
08cbf542 2794EXPORT_SYMBOL_GPL(fuse_file_poll);
95668a69
TH
2795
2796/*
2797 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
2798 * wakes up the poll waiters.
2799 */
2800int fuse_notify_poll_wakeup(struct fuse_conn *fc,
2801 struct fuse_notify_poll_wakeup_out *outarg)
2802{
2803 u64 kh = outarg->kh;
2804 struct rb_node **link;
2805
2806 spin_lock(&fc->lock);
2807
2808 link = fuse_find_polled_node(fc, kh, NULL);
2809 if (*link) {
2810 struct fuse_file *ff;
2811
2812 ff = rb_entry(*link, struct fuse_file, polled_node);
2813 wake_up_interruptible_sync(&ff->poll_wait);
2814 }
2815
2816 spin_unlock(&fc->lock);
2817 return 0;
2818}
2819
efb9fa9e
MP
2820static void fuse_do_truncate(struct file *file)
2821{
2822 struct inode *inode = file->f_mapping->host;
2823 struct iattr attr;
2824
2825 attr.ia_valid = ATTR_SIZE;
2826 attr.ia_size = i_size_read(inode);
2827
2828 attr.ia_file = file;
2829 attr.ia_valid |= ATTR_FILE;
2830
2831 fuse_do_setattr(inode, &attr, file);
2832}
2833
e5c5f05d
MP
2834static inline loff_t fuse_round_up(loff_t off)
2835{
2836 return round_up(off, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
2837}
2838
4273b793 2839static ssize_t
c8b8e32d 2840fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
4273b793 2841{
9d5722b7 2842 DECLARE_COMPLETION_ONSTACK(wait);
4273b793 2843 ssize_t ret = 0;
60b9df7a
MS
2844 struct file *file = iocb->ki_filp;
2845 struct fuse_file *ff = file->private_data;
e5c5f05d 2846 bool async_dio = ff->fc->async_dio;
4273b793 2847 loff_t pos = 0;
bcba24cc
MP
2848 struct inode *inode;
2849 loff_t i_size;
a6cbcd4a 2850 size_t count = iov_iter_count(iter);
c8b8e32d 2851 loff_t offset = iocb->ki_pos;
36cf66ed 2852 struct fuse_io_priv *io;
7cabc61e 2853 bool is_sync = is_sync_kiocb(iocb);
4273b793 2854
4273b793 2855 pos = offset;
bcba24cc
MP
2856 inode = file->f_mapping->host;
2857 i_size = i_size_read(inode);
4273b793 2858
6f673763 2859 if ((iov_iter_rw(iter) == READ) && (offset > i_size))
9fe55eea
SW
2860 return 0;
2861
439ee5f0 2862 /* optimization for short read */
6f673763 2863 if (async_dio && iov_iter_rw(iter) != WRITE && offset + count > i_size) {
439ee5f0
MP
2864 if (offset >= i_size)
2865 return 0;
6b775b18
AV
2866 iov_iter_truncate(iter, fuse_round_up(i_size - offset));
2867 count = iov_iter_count(iter);
439ee5f0
MP
2868 }
2869
bcba24cc 2870 io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL);
36cf66ed
MP
2871 if (!io)
2872 return -ENOMEM;
bcba24cc 2873 spin_lock_init(&io->lock);
744742d6 2874 kref_init(&io->refcnt);
bcba24cc
MP
2875 io->reqs = 1;
2876 io->bytes = -1;
2877 io->size = 0;
2878 io->offset = offset;
6f673763 2879 io->write = (iov_iter_rw(iter) == WRITE);
bcba24cc 2880 io->err = 0;
36cf66ed 2881 io->file = file;
bcba24cc
MP
2882 /*
2883 * By default, we want to optimize all I/Os with async request
60b9df7a 2884 * submission to the client filesystem if supported.
bcba24cc 2885 */
e5c5f05d 2886 io->async = async_dio;
bcba24cc
MP
2887 io->iocb = iocb;
2888
2889 /*
2890 * We cannot asynchronously extend the size of a file. We have no method
2891 * to wait on real async I/O requests, so we must submit this request
2892 * synchronously.
2893 */
7cabc61e 2894 if (!is_sync && (offset + count > i_size) &&
6f673763 2895 iov_iter_rw(iter) == WRITE)
bcba24cc 2896 io->async = false;
4273b793 2897
744742d6
SF
2898 if (io->async && is_sync) {
2899 /*
2900 * Additional reference to keep io around after
2901 * calling fuse_aio_complete()
2902 */
2903 kref_get(&io->refcnt);
9d5722b7 2904 io->done = &wait;
744742d6 2905 }
9d5722b7 2906
6f673763 2907 if (iov_iter_rw(iter) == WRITE) {
6b775b18 2908 ret = fuse_direct_io(io, iter, &pos, FUSE_DIO_WRITE);
812408fb
AV
2909 fuse_invalidate_attr(inode);
2910 } else {
d22a943f 2911 ret = __fuse_direct_read(io, iter, &pos);
812408fb 2912 }
36cf66ed 2913
bcba24cc
MP
2914 if (io->async) {
2915 fuse_aio_complete(io, ret < 0 ? ret : 0, -1);
2916
2917 /* we have a non-extending, async request, so return */
7cabc61e 2918 if (!is_sync)
bcba24cc
MP
2919 return -EIOCBQUEUED;
2920
9d5722b7
CH
2921 wait_for_completion(&wait);
2922 ret = fuse_get_res_by_io(io);
bcba24cc
MP
2923 }
2924
744742d6 2925 kref_put(&io->refcnt, fuse_io_release);
9d5722b7 2926
6f673763 2927 if (iov_iter_rw(iter) == WRITE) {
efb9fa9e
MP
2928 if (ret > 0)
2929 fuse_write_update_size(inode, pos);
2930 else if (ret < 0 && offset + count > i_size)
2931 fuse_do_truncate(file);
2932 }
4273b793
AA
2933
2934 return ret;
2935}
2936
cdadb11c
MS
2937static long fuse_file_fallocate(struct file *file, int mode, loff_t offset,
2938 loff_t length)
05ba1f08
AP
2939{
2940 struct fuse_file *ff = file->private_data;
1c68271c 2941 struct inode *inode = file_inode(file);
0ab08f57 2942 struct fuse_inode *fi = get_fuse_inode(inode);
05ba1f08 2943 struct fuse_conn *fc = ff->fc;
7078187a 2944 FUSE_ARGS(args);
05ba1f08
AP
2945 struct fuse_fallocate_in inarg = {
2946 .fh = ff->fh,
2947 .offset = offset,
2948 .length = length,
2949 .mode = mode
2950 };
2951 int err;
14c14414
MP
2952 bool lock_inode = !(mode & FALLOC_FL_KEEP_SIZE) ||
2953 (mode & FALLOC_FL_PUNCH_HOLE);
05ba1f08 2954
4adb8302
MS
2955 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2956 return -EOPNOTSUPP;
2957
519c6040
MS
2958 if (fc->no_fallocate)
2959 return -EOPNOTSUPP;
2960
14c14414 2961 if (lock_inode) {
5955102c 2962 inode_lock(inode);
bde52788
MP
2963 if (mode & FALLOC_FL_PUNCH_HOLE) {
2964 loff_t endbyte = offset + length - 1;
2965 err = filemap_write_and_wait_range(inode->i_mapping,
2966 offset, endbyte);
2967 if (err)
2968 goto out;
2969
2970 fuse_sync_writes(inode);
2971 }
3634a632
BF
2972 }
2973
0ab08f57
MP
2974 if (!(mode & FALLOC_FL_KEEP_SIZE))
2975 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
2976
7078187a
MS
2977 args.in.h.opcode = FUSE_FALLOCATE;
2978 args.in.h.nodeid = ff->nodeid;
2979 args.in.numargs = 1;
2980 args.in.args[0].size = sizeof(inarg);
2981 args.in.args[0].value = &inarg;
2982 err = fuse_simple_request(fc, &args);
519c6040
MS
2983 if (err == -ENOSYS) {
2984 fc->no_fallocate = 1;
2985 err = -EOPNOTSUPP;
2986 }
bee6c307
BF
2987 if (err)
2988 goto out;
2989
2990 /* we could have extended the file */
b0aa7606
MP
2991 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
2992 bool changed = fuse_write_update_size(inode, offset + length);
2993
93d2269d
MS
2994 if (changed && fc->writeback_cache)
2995 file_update_time(file);
b0aa7606 2996 }
bee6c307
BF
2997
2998 if (mode & FALLOC_FL_PUNCH_HOLE)
2999 truncate_pagecache_range(inode, offset, offset + length - 1);
3000
3001 fuse_invalidate_attr(inode);
3002
3634a632 3003out:
0ab08f57
MP
3004 if (!(mode & FALLOC_FL_KEEP_SIZE))
3005 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
3006
bde52788 3007 if (lock_inode)
5955102c 3008 inode_unlock(inode);
3634a632 3009
05ba1f08
AP
3010 return err;
3011}
05ba1f08 3012
4b6f5d20 3013static const struct file_operations fuse_file_operations = {
5559b8f4 3014 .llseek = fuse_file_llseek,
37c20f16 3015 .read_iter = fuse_file_read_iter,
84c3d55c 3016 .write_iter = fuse_file_write_iter,
b6aeaded
MS
3017 .mmap = fuse_file_mmap,
3018 .open = fuse_open,
3019 .flush = fuse_flush,
3020 .release = fuse_release,
3021 .fsync = fuse_fsync,
71421259 3022 .lock = fuse_file_lock,
a9ff4f87 3023 .flock = fuse_file_flock,
5ffc4ef4 3024 .splice_read = generic_file_splice_read,
59efec7b
TH
3025 .unlocked_ioctl = fuse_file_ioctl,
3026 .compat_ioctl = fuse_file_compat_ioctl,
95668a69 3027 .poll = fuse_file_poll,
05ba1f08 3028 .fallocate = fuse_file_fallocate,
b6aeaded
MS
3029};
3030
4b6f5d20 3031static const struct file_operations fuse_direct_io_file_operations = {
5559b8f4 3032 .llseek = fuse_file_llseek,
15316263 3033 .read_iter = fuse_direct_read_iter,
15316263 3034 .write_iter = fuse_direct_write_iter,
fc280c96 3035 .mmap = fuse_direct_mmap,
413ef8cb
MS
3036 .open = fuse_open,
3037 .flush = fuse_flush,
3038 .release = fuse_release,
3039 .fsync = fuse_fsync,
71421259 3040 .lock = fuse_file_lock,
a9ff4f87 3041 .flock = fuse_file_flock,
59efec7b
TH
3042 .unlocked_ioctl = fuse_file_ioctl,
3043 .compat_ioctl = fuse_file_compat_ioctl,
95668a69 3044 .poll = fuse_file_poll,
05ba1f08 3045 .fallocate = fuse_file_fallocate,
fc280c96 3046 /* no splice_read */
413ef8cb
MS
3047};
3048
f5e54d6e 3049static const struct address_space_operations fuse_file_aops = {
b6aeaded 3050 .readpage = fuse_readpage,
3be5a52b 3051 .writepage = fuse_writepage,
26d614df 3052 .writepages = fuse_writepages,
3be5a52b 3053 .launder_page = fuse_launder_page,
db50b96c 3054 .readpages = fuse_readpages,
3be5a52b 3055 .set_page_dirty = __set_page_dirty_nobuffers,
b2d2272f 3056 .bmap = fuse_bmap,
4273b793 3057 .direct_IO = fuse_direct_IO,
6b12c1b3
PE
3058 .write_begin = fuse_write_begin,
3059 .write_end = fuse_write_end,
b6aeaded
MS
3060};
3061
3062void fuse_init_file_inode(struct inode *inode)
3063{
45323fb7
MS
3064 inode->i_fop = &fuse_file_operations;
3065 inode->i_data.a_ops = &fuse_file_aops;
b6aeaded 3066}