?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 5.10.7
[linux/fpc-iii.git] / fs / fuse / file.c
blob8b306005453ccff01cd545917391d93a16d1ea60
1 /*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
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>
14 #include <linux/sched.h>
15 #include <linux/sched/signal.h>
16 #include <linux/module.h>
17 #include <linux/compat.h>
18 #include <linux/swap.h>
19 #include <linux/falloc.h>
20 #include <linux/uio.h>
21 #include <linux/fs.h>
22
23 static struct page **fuse_pages_alloc(unsigned int npages, gfp_t flags,
24 struct fuse_page_desc **desc)
25 {
26 struct page **pages;
27
28 pages = kzalloc(npages * (sizeof(struct page *) +
29 sizeof(struct fuse_page_desc)), flags);
30 *desc = (void *) (pages + npages);
31
32 return pages;
33 }
34
35 static int fuse_send_open(struct fuse_mount *fm, u64 nodeid, struct file *file,
36 int opcode, struct fuse_open_out *outargp)
37 {
38 struct fuse_open_in inarg;
39 FUSE_ARGS(args);
40
41 memset(&inarg, 0, sizeof(inarg));
42 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY);
43 if (!fm->fc->atomic_o_trunc)
44 inarg.flags &= ~O_TRUNC;
45 args.opcode = opcode;
46 args.nodeid = nodeid;
47 args.in_numargs = 1;
48 args.in_args[0].size = sizeof(inarg);
49 args.in_args[0].value = &inarg;
50 args.out_numargs = 1;
51 args.out_args[0].size = sizeof(*outargp);
52 args.out_args[0].value = outargp;
53
54 return fuse_simple_request(fm, &args);
55 }
56
57 struct fuse_release_args {
58 struct fuse_args args;
59 struct fuse_release_in inarg;
60 struct inode *inode;
61 };
62
63 struct fuse_file *fuse_file_alloc(struct fuse_mount *fm)
64 {
65 struct fuse_file *ff;
66
67 ff = kzalloc(sizeof(struct fuse_file), GFP_KERNEL_ACCOUNT);
68 if (unlikely(!ff))
69 return NULL;
70
71 ff->fm = fm;
72 ff->release_args = kzalloc(sizeof(*ff->release_args),
73 GFP_KERNEL_ACCOUNT);
74 if (!ff->release_args) {
75 kfree(ff);
76 return NULL;
77 }
78
79 INIT_LIST_HEAD(&ff->write_entry);
80 mutex_init(&ff->readdir.lock);
81 refcount_set(&ff->count, 1);
82 RB_CLEAR_NODE(&ff->polled_node);
83 init_waitqueue_head(&ff->poll_wait);
84
85 ff->kh = atomic64_inc_return(&fm->fc->khctr);
86
87 return ff;
88 }
89
90 void fuse_file_free(struct fuse_file *ff)
91 {
92 kfree(ff->release_args);
93 mutex_destroy(&ff->readdir.lock);
94 kfree(ff);
95 }
96
97 static struct fuse_file *fuse_file_get(struct fuse_file *ff)
98 {
99 refcount_inc(&ff->count);
100 return ff;
101 }
102
103 static void fuse_release_end(struct fuse_mount *fm, struct fuse_args *args,
104 int error)
105 {
106 struct fuse_release_args *ra = container_of(args, typeof(*ra), args);
107
108 iput(ra->inode);
109 kfree(ra);
110 }
111
112 static void fuse_file_put(struct fuse_file *ff, bool sync, bool isdir)
113 {
114 if (refcount_dec_and_test(&ff->count)) {
115 struct fuse_args *args = &ff->release_args->args;
116
117 if (isdir ? ff->fm->fc->no_opendir : ff->fm->fc->no_open) {
118 /* Do nothing when client does not implement 'open' */
119 fuse_release_end(ff->fm, args, 0);
120 } else if (sync) {
121 fuse_simple_request(ff->fm, args);
122 fuse_release_end(ff->fm, args, 0);
123 } else {
124 args->end = fuse_release_end;
125 if (fuse_simple_background(ff->fm, args,
126 GFP_KERNEL | __GFP_NOFAIL))
127 fuse_release_end(ff->fm, args, -ENOTCONN);
128 }
129 kfree(ff);
130 }
131 }
132
133 int fuse_do_open(struct fuse_mount *fm, u64 nodeid, struct file *file,
134 bool isdir)
135 {
136 struct fuse_conn *fc = fm->fc;
137 struct fuse_file *ff;
138 int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
139
140 ff = fuse_file_alloc(fm);
141 if (!ff)
142 return -ENOMEM;
143
144 ff->fh = 0;
145 /* Default for no-open */
146 ff->open_flags = FOPEN_KEEP_CACHE | (isdir ? FOPEN_CACHE_DIR : 0);
147 if (isdir ? !fc->no_opendir : !fc->no_open) {
148 struct fuse_open_out outarg;
149 int err;
150
151 err = fuse_send_open(fm, nodeid, file, opcode, &outarg);
152 if (!err) {
153 ff->fh = outarg.fh;
154 ff->open_flags = outarg.open_flags;
155
156 } else if (err != -ENOSYS) {
157 fuse_file_free(ff);
158 return err;
159 } else {
160 if (isdir)
161 fc->no_opendir = 1;
162 else
163 fc->no_open = 1;
164 }
165 }
166
167 if (isdir)
168 ff->open_flags &= ~FOPEN_DIRECT_IO;
169
170 ff->nodeid = nodeid;
171 file->private_data = ff;
172
173 return 0;
174 }
175 EXPORT_SYMBOL_GPL(fuse_do_open);
176
177 static void fuse_link_write_file(struct file *file)
178 {
179 struct inode *inode = file_inode(file);
180 struct fuse_inode *fi = get_fuse_inode(inode);
181 struct fuse_file *ff = file->private_data;
182 /*
183 * file may be written through mmap, so chain it onto the
184 * inodes's write_file list
185 */
186 spin_lock(&fi->lock);
187 if (list_empty(&ff->write_entry))
188 list_add(&ff->write_entry, &fi->write_files);
189 spin_unlock(&fi->lock);
190 }
191
192 void fuse_finish_open(struct inode *inode, struct file *file)
193 {
194 struct fuse_file *ff = file->private_data;
195 struct fuse_conn *fc = get_fuse_conn(inode);
196
197 if (!(ff->open_flags & FOPEN_KEEP_CACHE))
198 invalidate_inode_pages2(inode->i_mapping);
199 if (ff->open_flags & FOPEN_STREAM)
200 stream_open(inode, file);
201 else if (ff->open_flags & FOPEN_NONSEEKABLE)
202 nonseekable_open(inode, file);
203 if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) {
204 struct fuse_inode *fi = get_fuse_inode(inode);
205
206 spin_lock(&fi->lock);
207 fi->attr_version = atomic64_inc_return(&fc->attr_version);
208 i_size_write(inode, 0);
209 spin_unlock(&fi->lock);
210 fuse_invalidate_attr(inode);
211 if (fc->writeback_cache)
212 file_update_time(file);
213 }
214 if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache)
215 fuse_link_write_file(file);
216 }
217
218 int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
219 {
220 struct fuse_mount *fm = get_fuse_mount(inode);
221 struct fuse_conn *fc = fm->fc;
222 int err;
223 bool is_wb_truncate = (file->f_flags & O_TRUNC) &&
224 fc->atomic_o_trunc &&
225 fc->writeback_cache;
226 bool dax_truncate = (file->f_flags & O_TRUNC) &&
227 fc->atomic_o_trunc && FUSE_IS_DAX(inode);
228
229 if (fuse_is_bad(inode))
230 return -EIO;
231
232 err = generic_file_open(inode, file);
233 if (err)
234 return err;
235
236 if (is_wb_truncate || dax_truncate) {
237 inode_lock(inode);
238 fuse_set_nowrite(inode);
239 }
240
241 if (dax_truncate) {
242 down_write(&get_fuse_inode(inode)->i_mmap_sem);
243 err = fuse_dax_break_layouts(inode, 0, 0);
244 if (err)
245 goto out;
246 }
247
248 err = fuse_do_open(fm, get_node_id(inode), file, isdir);
249 if (!err)
250 fuse_finish_open(inode, file);
251
252 out:
253 if (dax_truncate)
254 up_write(&get_fuse_inode(inode)->i_mmap_sem);
255
256 if (is_wb_truncate | dax_truncate) {
257 fuse_release_nowrite(inode);
258 inode_unlock(inode);
259 }
260
261 return err;
262 }
263
264 static void fuse_prepare_release(struct fuse_inode *fi, struct fuse_file *ff,
265 int flags, int opcode)
266 {
267 struct fuse_conn *fc = ff->fm->fc;
268 struct fuse_release_args *ra = ff->release_args;
269
270 /* Inode is NULL on error path of fuse_create_open() */
271 if (likely(fi)) {
272 spin_lock(&fi->lock);
273 list_del(&ff->write_entry);
274 spin_unlock(&fi->lock);
275 }
276 spin_lock(&fc->lock);
277 if (!RB_EMPTY_NODE(&ff->polled_node))
278 rb_erase(&ff->polled_node, &fc->polled_files);
279 spin_unlock(&fc->lock);
280
281 wake_up_interruptible_all(&ff->poll_wait);
282
283 ra->inarg.fh = ff->fh;
284 ra->inarg.flags = flags;
285 ra->args.in_numargs = 1;
286 ra->args.in_args[0].size = sizeof(struct fuse_release_in);
287 ra->args.in_args[0].value = &ra->inarg;
288 ra->args.opcode = opcode;
289 ra->args.nodeid = ff->nodeid;
290 ra->args.force = true;
291 ra->args.nocreds = true;
292 }
293
294 void fuse_release_common(struct file *file, bool isdir)
295 {
296 struct fuse_inode *fi = get_fuse_inode(file_inode(file));
297 struct fuse_file *ff = file->private_data;
298 struct fuse_release_args *ra = ff->release_args;
299 int opcode = isdir ? FUSE_RELEASEDIR : FUSE_RELEASE;
300
301 fuse_prepare_release(fi, ff, file->f_flags, opcode);
302
303 if (ff->flock) {
304 ra->inarg.release_flags |= FUSE_RELEASE_FLOCK_UNLOCK;
305 ra->inarg.lock_owner = fuse_lock_owner_id(ff->fm->fc,
306 (fl_owner_t) file);
307 }
308 /* Hold inode until release is finished */
309 ra->inode = igrab(file_inode(file));
310
311 /*
312 * Normally this will send the RELEASE request, however if
313 * some asynchronous READ or WRITE requests are outstanding,
314 * the sending will be delayed.
315 *
316 * Make the release synchronous if this is a fuseblk mount,
317 * synchronous RELEASE is allowed (and desirable) in this case
318 * because the server can be trusted not to screw up.
319 */
320 fuse_file_put(ff, ff->fm->fc->destroy, isdir);
321 }
322
323 static int fuse_open(struct inode *inode, struct file *file)
324 {
325 return fuse_open_common(inode, file, false);
326 }
327
328 static int fuse_release(struct inode *inode, struct file *file)
329 {
330 struct fuse_conn *fc = get_fuse_conn(inode);
331
332 /* see fuse_vma_close() for !writeback_cache case */
333 if (fc->writeback_cache)
334 write_inode_now(inode, 1);
335
336 fuse_release_common(file, false);
337
338 /* return value is ignored by VFS */
339 return 0;
340 }
341
342 void fuse_sync_release(struct fuse_inode *fi, struct fuse_file *ff, int flags)
343 {
344 WARN_ON(refcount_read(&ff->count) > 1);
345 fuse_prepare_release(fi, ff, flags, FUSE_RELEASE);
346 /*
347 * iput(NULL) is a no-op and since the refcount is 1 and everything's
348 * synchronous, we are fine with not doing igrab() here"
349 */
350 fuse_file_put(ff, true, false);
351 }
352 EXPORT_SYMBOL_GPL(fuse_sync_release);
353
354 /*
355 * Scramble the ID space with XTEA, so that the value of the files_struct
356 * pointer is not exposed to userspace.
357 */
358 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
359 {
360 u32 *k = fc->scramble_key;
361 u64 v = (unsigned long) id;
362 u32 v0 = v;
363 u32 v1 = v >> 32;
364 u32 sum = 0;
365 int i;
366
367 for (i = 0; i < 32; i++) {
368 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
369 sum += 0x9E3779B9;
370 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
371 }
372
373 return (u64) v0 + ((u64) v1 << 32);
374 }
375
376 struct fuse_writepage_args {
377 struct fuse_io_args ia;
378 struct rb_node writepages_entry;
379 struct list_head queue_entry;
380 struct fuse_writepage_args *next;
381 struct inode *inode;
382 };
383
384 static struct fuse_writepage_args *fuse_find_writeback(struct fuse_inode *fi,
385 pgoff_t idx_from, pgoff_t idx_to)
386 {
387 struct rb_node *n;
388
389 n = fi->writepages.rb_node;
390
391 while (n) {
392 struct fuse_writepage_args *wpa;
393 pgoff_t curr_index;
394
395 wpa = rb_entry(n, struct fuse_writepage_args, writepages_entry);
396 WARN_ON(get_fuse_inode(wpa->inode) != fi);
397 curr_index = wpa->ia.write.in.offset >> PAGE_SHIFT;
398 if (idx_from >= curr_index + wpa->ia.ap.num_pages)
399 n = n->rb_right;
400 else if (idx_to < curr_index)
401 n = n->rb_left;
402 else
403 return wpa;
404 }
405 return NULL;
406 }
407
408 /*
409 * Check if any page in a range is under writeback
410 *
411 * This is currently done by walking the list of writepage requests
412 * for the inode, which can be pretty inefficient.
413 */
414 static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from,
415 pgoff_t idx_to)
416 {
417 struct fuse_inode *fi = get_fuse_inode(inode);
418 bool found;
419
420 spin_lock(&fi->lock);
421 found = fuse_find_writeback(fi, idx_from, idx_to);
422 spin_unlock(&fi->lock);
423
424 return found;
425 }
426
427 static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
428 {
429 return fuse_range_is_writeback(inode, index, index);
430 }
431
432 /*
433 * Wait for page writeback to be completed.
434 *
435 * Since fuse doesn't rely on the VM writeback tracking, this has to
436 * use some other means.
437 */
438 static void fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
439 {
440 struct fuse_inode *fi = get_fuse_inode(inode);
441
442 wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
443 }
444
445 /*
446 * Wait for all pending writepages on the inode to finish.
447 *
448 * This is currently done by blocking further writes with FUSE_NOWRITE
449 * and waiting for all sent writes to complete.
450 *
451 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
452 * could conflict with truncation.
453 */
454 static void fuse_sync_writes(struct inode *inode)
455 {
456 fuse_set_nowrite(inode);
457 fuse_release_nowrite(inode);
458 }
459
460 static int fuse_flush(struct file *file, fl_owner_t id)
461 {
462 struct inode *inode = file_inode(file);
463 struct fuse_mount *fm = get_fuse_mount(inode);
464 struct fuse_file *ff = file->private_data;
465 struct fuse_flush_in inarg;
466 FUSE_ARGS(args);
467 int err;
468
469 if (fuse_is_bad(inode))
470 return -EIO;
471
472 err = write_inode_now(inode, 1);
473 if (err)
474 return err;
475
476 inode_lock(inode);
477 fuse_sync_writes(inode);
478 inode_unlock(inode);
479
480 err = filemap_check_errors(file->f_mapping);
481 if (err)
482 return err;
483
484 err = 0;
485 if (fm->fc->no_flush)
486 goto inval_attr_out;
487
488 memset(&inarg, 0, sizeof(inarg));
489 inarg.fh = ff->fh;
490 inarg.lock_owner = fuse_lock_owner_id(fm->fc, id);
491 args.opcode = FUSE_FLUSH;
492 args.nodeid = get_node_id(inode);
493 args.in_numargs = 1;
494 args.in_args[0].size = sizeof(inarg);
495 args.in_args[0].value = &inarg;
496 args.force = true;
497
498 err = fuse_simple_request(fm, &args);
499 if (err == -ENOSYS) {
500 fm->fc->no_flush = 1;
501 err = 0;
502 }
503
504 inval_attr_out:
505 /*
506 * In memory i_blocks is not maintained by fuse, if writeback cache is
507 * enabled, i_blocks from cached attr may not be accurate.
508 */
509 if (!err && fm->fc->writeback_cache)
510 fuse_invalidate_attr(inode);
511 return err;
512 }
513
514 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
515 int datasync, int opcode)
516 {
517 struct inode *inode = file->f_mapping->host;
518 struct fuse_mount *fm = get_fuse_mount(inode);
519 struct fuse_file *ff = file->private_data;
520 FUSE_ARGS(args);
521 struct fuse_fsync_in inarg;
522
523 memset(&inarg, 0, sizeof(inarg));
524 inarg.fh = ff->fh;
525 inarg.fsync_flags = datasync ? FUSE_FSYNC_FDATASYNC : 0;
526 args.opcode = opcode;
527 args.nodeid = get_node_id(inode);
528 args.in_numargs = 1;
529 args.in_args[0].size = sizeof(inarg);
530 args.in_args[0].value = &inarg;
531 return fuse_simple_request(fm, &args);
532 }
533
534 static int fuse_fsync(struct file *file, loff_t start, loff_t end,
535 int datasync)
536 {
537 struct inode *inode = file->f_mapping->host;
538 struct fuse_conn *fc = get_fuse_conn(inode);
539 int err;
540
541 if (fuse_is_bad(inode))
542 return -EIO;
543
544 inode_lock(inode);
545
546 /*
547 * Start writeback against all dirty pages of the inode, then
548 * wait for all outstanding writes, before sending the FSYNC
549 * request.
550 */
551 err = file_write_and_wait_range(file, start, end);
552 if (err)
553 goto out;
554
555 fuse_sync_writes(inode);
556
557 /*
558 * Due to implementation of fuse writeback
559 * file_write_and_wait_range() does not catch errors.
560 * We have to do this directly after fuse_sync_writes()
561 */
562 err = file_check_and_advance_wb_err(file);
563 if (err)
564 goto out;
565
566 err = sync_inode_metadata(inode, 1);
567 if (err)
568 goto out;
569
570 if (fc->no_fsync)
571 goto out;
572
573 err = fuse_fsync_common(file, start, end, datasync, FUSE_FSYNC);
574 if (err == -ENOSYS) {
575 fc->no_fsync = 1;
576 err = 0;
577 }
578 out:
579 inode_unlock(inode);
580
581 return err;
582 }
583
584 void fuse_read_args_fill(struct fuse_io_args *ia, struct file *file, loff_t pos,
585 size_t count, int opcode)
586 {
587 struct fuse_file *ff = file->private_data;
588 struct fuse_args *args = &ia->ap.args;
589
590 ia->read.in.fh = ff->fh;
591 ia->read.in.offset = pos;
592 ia->read.in.size = count;
593 ia->read.in.flags = file->f_flags;
594 args->opcode = opcode;
595 args->nodeid = ff->nodeid;
596 args->in_numargs = 1;
597 args->in_args[0].size = sizeof(ia->read.in);
598 args->in_args[0].value = &ia->read.in;
599 args->out_argvar = true;
600 args->out_numargs = 1;
601 args->out_args[0].size = count;
602 }
603
604 static void fuse_release_user_pages(struct fuse_args_pages *ap,
605 bool should_dirty)
606 {
607 unsigned int i;
608
609 for (i = 0; i < ap->num_pages; i++) {
610 if (should_dirty)
611 set_page_dirty_lock(ap->pages[i]);
612 put_page(ap->pages[i]);
613 }
614 }
615
616 static void fuse_io_release(struct kref *kref)
617 {
618 kfree(container_of(kref, struct fuse_io_priv, refcnt));
619 }
620
621 static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io)
622 {
623 if (io->err)
624 return io->err;
625
626 if (io->bytes >= 0 && io->write)
627 return -EIO;
628
629 return io->bytes < 0 ? io->size : io->bytes;
630 }
631
632 /**
633 * In case of short read, the caller sets 'pos' to the position of
634 * actual end of fuse request in IO request. Otherwise, if bytes_requested
635 * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1.
636 *
637 * An example:
638 * User requested DIO read of 64K. It was splitted into two 32K fuse requests,
639 * both submitted asynchronously. The first of them was ACKed by userspace as
640 * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The
641 * second request was ACKed as short, e.g. only 1K was read, resulting in
642 * pos == 33K.
643 *
644 * Thus, when all fuse requests are completed, the minimal non-negative 'pos'
645 * will be equal to the length of the longest contiguous fragment of
646 * transferred data starting from the beginning of IO request.
647 */
648 static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
649 {
650 int left;
651
652 spin_lock(&io->lock);
653 if (err)
654 io->err = io->err ? : err;
655 else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes))
656 io->bytes = pos;
657
658 left = --io->reqs;
659 if (!left && io->blocking)
660 complete(io->done);
661 spin_unlock(&io->lock);
662
663 if (!left && !io->blocking) {
664 ssize_t res = fuse_get_res_by_io(io);
665
666 if (res >= 0) {
667 struct inode *inode = file_inode(io->iocb->ki_filp);
668 struct fuse_conn *fc = get_fuse_conn(inode);
669 struct fuse_inode *fi = get_fuse_inode(inode);
670
671 spin_lock(&fi->lock);
672 fi->attr_version = atomic64_inc_return(&fc->attr_version);
673 spin_unlock(&fi->lock);
674 }
675
676 io->iocb->ki_complete(io->iocb, res, 0);
677 }
678
679 kref_put(&io->refcnt, fuse_io_release);
680 }
681
682 static struct fuse_io_args *fuse_io_alloc(struct fuse_io_priv *io,
683 unsigned int npages)
684 {
685 struct fuse_io_args *ia;
686
687 ia = kzalloc(sizeof(*ia), GFP_KERNEL);
688 if (ia) {
689 ia->io = io;
690 ia->ap.pages = fuse_pages_alloc(npages, GFP_KERNEL,
691 &ia->ap.descs);
692 if (!ia->ap.pages) {
693 kfree(ia);
694 ia = NULL;
695 }
696 }
697 return ia;
698 }
699
700 static void fuse_io_free(struct fuse_io_args *ia)
701 {
702 kfree(ia->ap.pages);
703 kfree(ia);
704 }
705
706 static void fuse_aio_complete_req(struct fuse_mount *fm, struct fuse_args *args,
707 int err)
708 {
709 struct fuse_io_args *ia = container_of(args, typeof(*ia), ap.args);
710 struct fuse_io_priv *io = ia->io;
711 ssize_t pos = -1;
712
713 fuse_release_user_pages(&ia->ap, io->should_dirty);
714
715 if (err) {
716 /* Nothing */
717 } else if (io->write) {
718 if (ia->write.out.size > ia->write.in.size) {
719 err = -EIO;
720 } else if (ia->write.in.size != ia->write.out.size) {
721 pos = ia->write.in.offset - io->offset +
722 ia->write.out.size;
723 }
724 } else {
725 u32 outsize = args->out_args[0].size;
726
727 if (ia->read.in.size != outsize)
728 pos = ia->read.in.offset - io->offset + outsize;
729 }
730
731 fuse_aio_complete(io, err, pos);
732 fuse_io_free(ia);
733 }
734
735 static ssize_t fuse_async_req_send(struct fuse_mount *fm,
736 struct fuse_io_args *ia, size_t num_bytes)
737 {
738 ssize_t err;
739 struct fuse_io_priv *io = ia->io;
740
741 spin_lock(&io->lock);
742 kref_get(&io->refcnt);
743 io->size += num_bytes;
744 io->reqs++;
745 spin_unlock(&io->lock);
746
747 ia->ap.args.end = fuse_aio_complete_req;
748 ia->ap.args.may_block = io->should_dirty;
749 err = fuse_simple_background(fm, &ia->ap.args, GFP_KERNEL);
750 if (err)
751 fuse_aio_complete_req(fm, &ia->ap.args, err);
752
753 return num_bytes;
754 }
755
756 static ssize_t fuse_send_read(struct fuse_io_args *ia, loff_t pos, size_t count,
757 fl_owner_t owner)
758 {
759 struct file *file = ia->io->iocb->ki_filp;
760 struct fuse_file *ff = file->private_data;
761 struct fuse_mount *fm = ff->fm;
762
763 fuse_read_args_fill(ia, file, pos, count, FUSE_READ);
764 if (owner != NULL) {
765 ia->read.in.read_flags |= FUSE_READ_LOCKOWNER;
766 ia->read.in.lock_owner = fuse_lock_owner_id(fm->fc, owner);
767 }
768
769 if (ia->io->async)
770 return fuse_async_req_send(fm, ia, count);
771
772 return fuse_simple_request(fm, &ia->ap.args);
773 }
774
775 static void fuse_read_update_size(struct inode *inode, loff_t size,
776 u64 attr_ver)
777 {
778 struct fuse_conn *fc = get_fuse_conn(inode);
779 struct fuse_inode *fi = get_fuse_inode(inode);
780
781 spin_lock(&fi->lock);
782 if (attr_ver == fi->attr_version && size < inode->i_size &&
783 !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
784 fi->attr_version = atomic64_inc_return(&fc->attr_version);
785 i_size_write(inode, size);
786 }
787 spin_unlock(&fi->lock);
788 }
789
790 static void fuse_short_read(struct inode *inode, u64 attr_ver, size_t num_read,
791 struct fuse_args_pages *ap)
792 {
793 struct fuse_conn *fc = get_fuse_conn(inode);
794
795 if (fc->writeback_cache) {
796 /*
797 * A hole in a file. Some data after the hole are in page cache,
798 * but have not reached the client fs yet. So, the hole is not
799 * present there.
800 */
801 int i;
802 int start_idx = num_read >> PAGE_SHIFT;
803 size_t off = num_read & (PAGE_SIZE - 1);
804
805 for (i = start_idx; i < ap->num_pages; i++) {
806 zero_user_segment(ap->pages[i], off, PAGE_SIZE);
807 off = 0;
808 }
809 } else {
810 loff_t pos = page_offset(ap->pages[0]) + num_read;
811 fuse_read_update_size(inode, pos, attr_ver);
812 }
813 }
814
815 static int fuse_do_readpage(struct file *file, struct page *page)
816 {
817 struct inode *inode = page->mapping->host;
818 struct fuse_mount *fm = get_fuse_mount(inode);
819 loff_t pos = page_offset(page);
820 struct fuse_page_desc desc = { .length = PAGE_SIZE };
821 struct fuse_io_args ia = {
822 .ap.args.page_zeroing = true,
823 .ap.args.out_pages = true,
824 .ap.num_pages = 1,
825 .ap.pages = &page,
826 .ap.descs = &desc,
827 };
828 ssize_t res;
829 u64 attr_ver;
830
831 /*
832 * Page writeback can extend beyond the lifetime of the
833 * page-cache page, so make sure we read a properly synced
834 * page.
835 */
836 fuse_wait_on_page_writeback(inode, page->index);
837
838 attr_ver = fuse_get_attr_version(fm->fc);
839
840 /* Don't overflow end offset */
841 if (pos + (desc.length - 1) == LLONG_MAX)
842 desc.length--;
843
844 fuse_read_args_fill(&ia, file, pos, desc.length, FUSE_READ);
845 res = fuse_simple_request(fm, &ia.ap.args);
846 if (res < 0)
847 return res;
848 /*
849 * Short read means EOF. If file size is larger, truncate it
850 */
851 if (res < desc.length)
852 fuse_short_read(inode, attr_ver, res, &ia.ap);
853
854 SetPageUptodate(page);
855
856 return 0;
857 }
858
859 static int fuse_readpage(struct file *file, struct page *page)
860 {
861 struct inode *inode = page->mapping->host;
862 int err;
863
864 err = -EIO;
865 if (fuse_is_bad(inode))
866 goto out;
867
868 err = fuse_do_readpage(file, page);
869 fuse_invalidate_atime(inode);
870 out:
871 unlock_page(page);
872 return err;
873 }
874
875 static void fuse_readpages_end(struct fuse_mount *fm, struct fuse_args *args,
876 int err)
877 {
878 int i;
879 struct fuse_io_args *ia = container_of(args, typeof(*ia), ap.args);
880 struct fuse_args_pages *ap = &ia->ap;
881 size_t count = ia->read.in.size;
882 size_t num_read = args->out_args[0].size;
883 struct address_space *mapping = NULL;
884
885 for (i = 0; mapping == NULL && i < ap->num_pages; i++)
886 mapping = ap->pages[i]->mapping;
887
888 if (mapping) {
889 struct inode *inode = mapping->host;
890
891 /*
892 * Short read means EOF. If file size is larger, truncate it
893 */
894 if (!err && num_read < count)
895 fuse_short_read(inode, ia->read.attr_ver, num_read, ap);
896
897 fuse_invalidate_atime(inode);
898 }
899
900 for (i = 0; i < ap->num_pages; i++) {
901 struct page *page = ap->pages[i];
902
903 if (!err)
904 SetPageUptodate(page);
905 else
906 SetPageError(page);
907 unlock_page(page);
908 put_page(page);
909 }
910 if (ia->ff)
911 fuse_file_put(ia->ff, false, false);
912
913 fuse_io_free(ia);
914 }
915
916 static void fuse_send_readpages(struct fuse_io_args *ia, struct file *file)
917 {
918 struct fuse_file *ff = file->private_data;
919 struct fuse_mount *fm = ff->fm;
920 struct fuse_args_pages *ap = &ia->ap;
921 loff_t pos = page_offset(ap->pages[0]);
922 size_t count = ap->num_pages << PAGE_SHIFT;
923 ssize_t res;
924 int err;
925
926 ap->args.out_pages = true;
927 ap->args.page_zeroing = true;
928 ap->args.page_replace = true;
929
930 /* Don't overflow end offset */
931 if (pos + (count - 1) == LLONG_MAX) {
932 count--;
933 ap->descs[ap->num_pages - 1].length--;
934 }
935 WARN_ON((loff_t) (pos + count) < 0);
936
937 fuse_read_args_fill(ia, file, pos, count, FUSE_READ);
938 ia->read.attr_ver = fuse_get_attr_version(fm->fc);
939 if (fm->fc->async_read) {
940 ia->ff = fuse_file_get(ff);
941 ap->args.end = fuse_readpages_end;
942 err = fuse_simple_background(fm, &ap->args, GFP_KERNEL);
943 if (!err)
944 return;
945 } else {
946 res = fuse_simple_request(fm, &ap->args);
947 err = res < 0 ? res : 0;
948 }
949 fuse_readpages_end(fm, &ap->args, err);
950 }
951
952 static void fuse_readahead(struct readahead_control *rac)
953 {
954 struct inode *inode = rac->mapping->host;
955 struct fuse_conn *fc = get_fuse_conn(inode);
956 unsigned int i, max_pages, nr_pages = 0;
957
958 if (fuse_is_bad(inode))
959 return;
960
961 max_pages = min_t(unsigned int, fc->max_pages,
962 fc->max_read / PAGE_SIZE);
963
964 for (;;) {
965 struct fuse_io_args *ia;
966 struct fuse_args_pages *ap;
967
968 nr_pages = readahead_count(rac) - nr_pages;
969 if (nr_pages > max_pages)
970 nr_pages = max_pages;
971 if (nr_pages == 0)
972 break;
973 ia = fuse_io_alloc(NULL, nr_pages);
974 if (!ia)
975 return;
976 ap = &ia->ap;
977 nr_pages = __readahead_batch(rac, ap->pages, nr_pages);
978 for (i = 0; i < nr_pages; i++) {
979 fuse_wait_on_page_writeback(inode,
980 readahead_index(rac) + i);
981 ap->descs[i].length = PAGE_SIZE;
982 }
983 ap->num_pages = nr_pages;
984 fuse_send_readpages(ia, rac->file);
985 }
986 }
987
988 static ssize_t fuse_cache_read_iter(struct kiocb *iocb, struct iov_iter *to)
989 {
990 struct inode *inode = iocb->ki_filp->f_mapping->host;
991 struct fuse_conn *fc = get_fuse_conn(inode);
992
993 /*
994 * In auto invalidate mode, always update attributes on read.
995 * Otherwise, only update if we attempt to read past EOF (to ensure
996 * i_size is up to date).
997 */
998 if (fc->auto_inval_data ||
999 (iocb->ki_pos + iov_iter_count(to) > i_size_read(inode))) {
1000 int err;
1001 err = fuse_update_attributes(inode, iocb->ki_filp);
1002 if (err)
1003 return err;
1004 }
1005
1006 return generic_file_read_iter(iocb, to);
1007 }
1008
1009 static void fuse_write_args_fill(struct fuse_io_args *ia, struct fuse_file *ff,
1010 loff_t pos, size_t count)
1011 {
1012 struct fuse_args *args = &ia->ap.args;
1013
1014 ia->write.in.fh = ff->fh;
1015 ia->write.in.offset = pos;
1016 ia->write.in.size = count;
1017 args->opcode = FUSE_WRITE;
1018 args->nodeid = ff->nodeid;
1019 args->in_numargs = 2;
1020 if (ff->fm->fc->minor < 9)
1021 args->in_args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
1022 else
1023 args->in_args[0].size = sizeof(ia->write.in);
1024 args->in_args[0].value = &ia->write.in;
1025 args->in_args[1].size = count;
1026 args->out_numargs = 1;
1027 args->out_args[0].size = sizeof(ia->write.out);
1028 args->out_args[0].value = &ia->write.out;
1029 }
1030
1031 static unsigned int fuse_write_flags(struct kiocb *iocb)
1032 {
1033 unsigned int flags = iocb->ki_filp->f_flags;
1034
1035 if (iocb->ki_flags & IOCB_DSYNC)
1036 flags |= O_DSYNC;
1037 if (iocb->ki_flags & IOCB_SYNC)
1038 flags |= O_SYNC;
1039
1040 return flags;
1041 }
1042
1043 static ssize_t fuse_send_write(struct fuse_io_args *ia, loff_t pos,
1044 size_t count, fl_owner_t owner)
1045 {
1046 struct kiocb *iocb = ia->io->iocb;
1047 struct file *file = iocb->ki_filp;
1048 struct fuse_file *ff = file->private_data;
1049 struct fuse_mount *fm = ff->fm;
1050 struct fuse_write_in *inarg = &ia->write.in;
1051 ssize_t err;
1052
1053 fuse_write_args_fill(ia, ff, pos, count);
1054 inarg->flags = fuse_write_flags(iocb);
1055 if (owner != NULL) {
1056 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
1057 inarg->lock_owner = fuse_lock_owner_id(fm->fc, owner);
1058 }
1059
1060 if (ia->io->async)
1061 return fuse_async_req_send(fm, ia, count);
1062
1063 err = fuse_simple_request(fm, &ia->ap.args);
1064 if (!err && ia->write.out.size > count)
1065 err = -EIO;
1066
1067 return err ?: ia->write.out.size;
1068 }
1069
1070 bool fuse_write_update_size(struct inode *inode, loff_t pos)
1071 {
1072 struct fuse_conn *fc = get_fuse_conn(inode);
1073 struct fuse_inode *fi = get_fuse_inode(inode);
1074 bool ret = false;
1075
1076 spin_lock(&fi->lock);
1077 fi->attr_version = atomic64_inc_return(&fc->attr_version);
1078 if (pos > inode->i_size) {
1079 i_size_write(inode, pos);
1080 ret = true;
1081 }
1082 spin_unlock(&fi->lock);
1083
1084 return ret;
1085 }
1086
1087 static ssize_t fuse_send_write_pages(struct fuse_io_args *ia,
1088 struct kiocb *iocb, struct inode *inode,
1089 loff_t pos, size_t count)
1090 {
1091 struct fuse_args_pages *ap = &ia->ap;
1092 struct file *file = iocb->ki_filp;
1093 struct fuse_file *ff = file->private_data;
1094 struct fuse_mount *fm = ff->fm;
1095 unsigned int offset, i;
1096 int err;
1097
1098 for (i = 0; i < ap->num_pages; i++)
1099 fuse_wait_on_page_writeback(inode, ap->pages[i]->index);
1100
1101 fuse_write_args_fill(ia, ff, pos, count);
1102 ia->write.in.flags = fuse_write_flags(iocb);
1103
1104 err = fuse_simple_request(fm, &ap->args);
1105 if (!err && ia->write.out.size > count)
1106 err = -EIO;
1107
1108 offset = ap->descs[0].offset;
1109 count = ia->write.out.size;
1110 for (i = 0; i < ap->num_pages; i++) {
1111 struct page *page = ap->pages[i];
1112
1113 if (!err && !offset && count >= PAGE_SIZE)
1114 SetPageUptodate(page);
1115
1116 if (count > PAGE_SIZE - offset)
1117 count -= PAGE_SIZE - offset;
1118 else
1119 count = 0;
1120 offset = 0;
1121
1122 unlock_page(page);
1123 put_page(page);
1124 }
1125
1126 return err;
1127 }
1128
1129 static ssize_t fuse_fill_write_pages(struct fuse_args_pages *ap,
1130 struct address_space *mapping,
1131 struct iov_iter *ii, loff_t pos,
1132 unsigned int max_pages)
1133 {
1134 struct fuse_conn *fc = get_fuse_conn(mapping->host);
1135 unsigned offset = pos & (PAGE_SIZE - 1);
1136 size_t count = 0;
1137 int err;
1138
1139 ap->args.in_pages = true;
1140 ap->descs[0].offset = offset;
1141
1142 do {
1143 size_t tmp;
1144 struct page *page;
1145 pgoff_t index = pos >> PAGE_SHIFT;
1146 size_t bytes = min_t(size_t, PAGE_SIZE - offset,
1147 iov_iter_count(ii));
1148
1149 bytes = min_t(size_t, bytes, fc->max_write - count);
1150
1151 again:
1152 err = -EFAULT;
1153 if (iov_iter_fault_in_readable(ii, bytes))
1154 break;
1155
1156 err = -ENOMEM;
1157 page = grab_cache_page_write_begin(mapping, index, 0);
1158 if (!page)
1159 break;
1160
1161 if (mapping_writably_mapped(mapping))
1162 flush_dcache_page(page);
1163
1164 tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes);
1165 flush_dcache_page(page);
1166
1167 iov_iter_advance(ii, tmp);
1168 if (!tmp) {
1169 unlock_page(page);
1170 put_page(page);
1171 bytes = min(bytes, iov_iter_single_seg_count(ii));
1172 goto again;
1173 }
1174
1175 err = 0;
1176 ap->pages[ap->num_pages] = page;
1177 ap->descs[ap->num_pages].length = tmp;
1178 ap->num_pages++;
1179
1180 count += tmp;
1181 pos += tmp;
1182 offset += tmp;
1183 if (offset == PAGE_SIZE)
1184 offset = 0;
1185
1186 if (!fc->big_writes)
1187 break;
1188 } while (iov_iter_count(ii) && count < fc->max_write &&
1189 ap->num_pages < max_pages && offset == 0);
1190
1191 return count > 0 ? count : err;
1192 }
1193
1194 static inline unsigned int fuse_wr_pages(loff_t pos, size_t len,
1195 unsigned int max_pages)
1196 {
1197 return min_t(unsigned int,
1198 ((pos + len - 1) >> PAGE_SHIFT) -
1199 (pos >> PAGE_SHIFT) + 1,
1200 max_pages);
1201 }
1202
1203 static ssize_t fuse_perform_write(struct kiocb *iocb,
1204 struct address_space *mapping,
1205 struct iov_iter *ii, loff_t pos)
1206 {
1207 struct inode *inode = mapping->host;
1208 struct fuse_conn *fc = get_fuse_conn(inode);
1209 struct fuse_inode *fi = get_fuse_inode(inode);
1210 int err = 0;
1211 ssize_t res = 0;
1212
1213 if (inode->i_size < pos + iov_iter_count(ii))
1214 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1215
1216 do {
1217 ssize_t count;
1218 struct fuse_io_args ia = {};
1219 struct fuse_args_pages *ap = &ia.ap;
1220 unsigned int nr_pages = fuse_wr_pages(pos, iov_iter_count(ii),
1221 fc->max_pages);
1222
1223 ap->pages = fuse_pages_alloc(nr_pages, GFP_KERNEL, &ap->descs);
1224 if (!ap->pages) {
1225 err = -ENOMEM;
1226 break;
1227 }
1228
1229 count = fuse_fill_write_pages(ap, mapping, ii, pos, nr_pages);
1230 if (count <= 0) {
1231 err = count;
1232 } else {
1233 err = fuse_send_write_pages(&ia, iocb, inode,
1234 pos, count);
1235 if (!err) {
1236 size_t num_written = ia.write.out.size;
1237
1238 res += num_written;
1239 pos += num_written;
1240
1241 /* break out of the loop on short write */
1242 if (num_written != count)
1243 err = -EIO;
1244 }
1245 }
1246 kfree(ap->pages);
1247 } while (!err && iov_iter_count(ii));
1248
1249 if (res > 0)
1250 fuse_write_update_size(inode, pos);
1251
1252 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1253 fuse_invalidate_attr(inode);
1254
1255 return res > 0 ? res : err;
1256 }
1257
1258 static ssize_t fuse_cache_write_iter(struct kiocb *iocb, struct iov_iter *from)
1259 {
1260 struct file *file = iocb->ki_filp;
1261 struct address_space *mapping = file->f_mapping;
1262 ssize_t written = 0;
1263 ssize_t written_buffered = 0;
1264 struct inode *inode = mapping->host;
1265 ssize_t err;
1266 loff_t endbyte = 0;
1267
1268 if (get_fuse_conn(inode)->writeback_cache) {
1269 /* Update size (EOF optimization) and mode (SUID clearing) */
1270 err = fuse_update_attributes(mapping->host, file);
1271 if (err)
1272 return err;
1273
1274 return generic_file_write_iter(iocb, from);
1275 }
1276
1277 inode_lock(inode);
1278
1279 /* We can write back this queue in page reclaim */
1280 current->backing_dev_info = inode_to_bdi(inode);
1281
1282 err = generic_write_checks(iocb, from);
1283 if (err <= 0)
1284 goto out;
1285
1286 err = file_remove_privs(file);
1287 if (err)
1288 goto out;
1289
1290 err = file_update_time(file);
1291 if (err)
1292 goto out;
1293
1294 if (iocb->ki_flags & IOCB_DIRECT) {
1295 loff_t pos = iocb->ki_pos;
1296 written = generic_file_direct_write(iocb, from);
1297 if (written < 0 || !iov_iter_count(from))
1298 goto out;
1299
1300 pos += written;
1301
1302 written_buffered = fuse_perform_write(iocb, mapping, from, pos);
1303 if (written_buffered < 0) {
1304 err = written_buffered;
1305 goto out;
1306 }
1307 endbyte = pos + written_buffered - 1;
1308
1309 err = filemap_write_and_wait_range(file->f_mapping, pos,
1310 endbyte);
1311 if (err)
1312 goto out;
1313
1314 invalidate_mapping_pages(file->f_mapping,
1315 pos >> PAGE_SHIFT,
1316 endbyte >> PAGE_SHIFT);
1317
1318 written += written_buffered;
1319 iocb->ki_pos = pos + written_buffered;
1320 } else {
1321 written = fuse_perform_write(iocb, mapping, from, iocb->ki_pos);
1322 if (written >= 0)
1323 iocb->ki_pos += written;
1324 }
1325 out:
1326 current->backing_dev_info = NULL;
1327 inode_unlock(inode);
1328 if (written > 0)
1329 written = generic_write_sync(iocb, written);
1330
1331 return written ? written : err;
1332 }
1333
1334 static inline void fuse_page_descs_length_init(struct fuse_page_desc *descs,
1335 unsigned int index,
1336 unsigned int nr_pages)
1337 {
1338 int i;
1339
1340 for (i = index; i < index + nr_pages; i++)
1341 descs[i].length = PAGE_SIZE - descs[i].offset;
1342 }
1343
1344 static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii)
1345 {
1346 return (unsigned long)ii->iov->iov_base + ii->iov_offset;
1347 }
1348
1349 static inline size_t fuse_get_frag_size(const struct iov_iter *ii,
1350 size_t max_size)
1351 {
1352 return min(iov_iter_single_seg_count(ii), max_size);
1353 }
1354
1355 static int fuse_get_user_pages(struct fuse_args_pages *ap, struct iov_iter *ii,
1356 size_t *nbytesp, int write,
1357 unsigned int max_pages)
1358 {
1359 size_t nbytes = 0; /* # bytes already packed in req */
1360 ssize_t ret = 0;
1361
1362 /* Special case for kernel I/O: can copy directly into the buffer */
1363 if (iov_iter_is_kvec(ii)) {
1364 unsigned long user_addr = fuse_get_user_addr(ii);
1365 size_t frag_size = fuse_get_frag_size(ii, *nbytesp);
1366
1367 if (write)
1368 ap->args.in_args[1].value = (void *) user_addr;
1369 else
1370 ap->args.out_args[0].value = (void *) user_addr;
1371
1372 iov_iter_advance(ii, frag_size);
1373 *nbytesp = frag_size;
1374 return 0;
1375 }
1376
1377 while (nbytes < *nbytesp && ap->num_pages < max_pages) {
1378 unsigned npages;
1379 size_t start;
1380 ret = iov_iter_get_pages(ii, &ap->pages[ap->num_pages],
1381 *nbytesp - nbytes,
1382 max_pages - ap->num_pages,
1383 &start);
1384 if (ret < 0)
1385 break;
1386
1387 iov_iter_advance(ii, ret);
1388 nbytes += ret;
1389
1390 ret += start;
1391 npages = (ret + PAGE_SIZE - 1) / PAGE_SIZE;
1392
1393 ap->descs[ap->num_pages].offset = start;
1394 fuse_page_descs_length_init(ap->descs, ap->num_pages, npages);
1395
1396 ap->num_pages += npages;
1397 ap->descs[ap->num_pages - 1].length -=
1398 (PAGE_SIZE - ret) & (PAGE_SIZE - 1);
1399 }
1400
1401 if (write)
1402 ap->args.in_pages = true;
1403 else
1404 ap->args.out_pages = true;
1405
1406 *nbytesp = nbytes;
1407
1408 return ret < 0 ? ret : 0;
1409 }
1410
1411 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
1412 loff_t *ppos, int flags)
1413 {
1414 int write = flags & FUSE_DIO_WRITE;
1415 int cuse = flags & FUSE_DIO_CUSE;
1416 struct file *file = io->iocb->ki_filp;
1417 struct inode *inode = file->f_mapping->host;
1418 struct fuse_file *ff = file->private_data;
1419 struct fuse_conn *fc = ff->fm->fc;
1420 size_t nmax = write ? fc->max_write : fc->max_read;
1421 loff_t pos = *ppos;
1422 size_t count = iov_iter_count(iter);
1423 pgoff_t idx_from = pos >> PAGE_SHIFT;
1424 pgoff_t idx_to = (pos + count - 1) >> PAGE_SHIFT;
1425 ssize_t res = 0;
1426 int err = 0;
1427 struct fuse_io_args *ia;
1428 unsigned int max_pages;
1429
1430 max_pages = iov_iter_npages(iter, fc->max_pages);
1431 ia = fuse_io_alloc(io, max_pages);
1432 if (!ia)
1433 return -ENOMEM;
1434
1435 ia->io = io;
1436 if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) {
1437 if (!write)
1438 inode_lock(inode);
1439 fuse_sync_writes(inode);
1440 if (!write)
1441 inode_unlock(inode);
1442 }
1443
1444 io->should_dirty = !write && iter_is_iovec(iter);
1445 while (count) {
1446 ssize_t nres;
1447 fl_owner_t owner = current->files;
1448 size_t nbytes = min(count, nmax);
1449
1450 err = fuse_get_user_pages(&ia->ap, iter, &nbytes, write,
1451 max_pages);
1452 if (err && !nbytes)
1453 break;
1454
1455 if (write) {
1456 if (!capable(CAP_FSETID))
1457 ia->write.in.write_flags |= FUSE_WRITE_KILL_PRIV;
1458
1459 nres = fuse_send_write(ia, pos, nbytes, owner);
1460 } else {
1461 nres = fuse_send_read(ia, pos, nbytes, owner);
1462 }
1463
1464 if (!io->async || nres < 0) {
1465 fuse_release_user_pages(&ia->ap, io->should_dirty);
1466 fuse_io_free(ia);
1467 }
1468 ia = NULL;
1469 if (nres < 0) {
1470 iov_iter_revert(iter, nbytes);
1471 err = nres;
1472 break;
1473 }
1474 WARN_ON(nres > nbytes);
1475
1476 count -= nres;
1477 res += nres;
1478 pos += nres;
1479 if (nres != nbytes) {
1480 iov_iter_revert(iter, nbytes - nres);
1481 break;
1482 }
1483 if (count) {
1484 max_pages = iov_iter_npages(iter, fc->max_pages);
1485 ia = fuse_io_alloc(io, max_pages);
1486 if (!ia)
1487 break;
1488 }
1489 }
1490 if (ia)
1491 fuse_io_free(ia);
1492 if (res > 0)
1493 *ppos = pos;
1494
1495 return res > 0 ? res : err;
1496 }
1497 EXPORT_SYMBOL_GPL(fuse_direct_io);
1498
1499 static ssize_t __fuse_direct_read(struct fuse_io_priv *io,
1500 struct iov_iter *iter,
1501 loff_t *ppos)
1502 {
1503 ssize_t res;
1504 struct inode *inode = file_inode(io->iocb->ki_filp);
1505
1506 res = fuse_direct_io(io, iter, ppos, 0);
1507
1508 fuse_invalidate_atime(inode);
1509
1510 return res;
1511 }
1512
1513 static ssize_t fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter);
1514
1515 static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to)
1516 {
1517 ssize_t res;
1518
1519 if (!is_sync_kiocb(iocb) && iocb->ki_flags & IOCB_DIRECT) {
1520 res = fuse_direct_IO(iocb, to);
1521 } else {
1522 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
1523
1524 res = __fuse_direct_read(&io, to, &iocb->ki_pos);
1525 }
1526
1527 return res;
1528 }
1529
1530 static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
1531 {
1532 struct inode *inode = file_inode(iocb->ki_filp);
1533 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
1534 ssize_t res;
1535
1536 /* Don't allow parallel writes to the same file */
1537 inode_lock(inode);
1538 res = generic_write_checks(iocb, from);
1539 if (res > 0) {
1540 if (!is_sync_kiocb(iocb) && iocb->ki_flags & IOCB_DIRECT) {
1541 res = fuse_direct_IO(iocb, from);
1542 } else {
1543 res = fuse_direct_io(&io, from, &iocb->ki_pos,
1544 FUSE_DIO_WRITE);
1545 }
1546 }
1547 fuse_invalidate_attr(inode);
1548 if (res > 0)
1549 fuse_write_update_size(inode, iocb->ki_pos);
1550 inode_unlock(inode);
1551
1552 return res;
1553 }
1554
1555 static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1556 {
1557 struct file *file = iocb->ki_filp;
1558 struct fuse_file *ff = file->private_data;
1559 struct inode *inode = file_inode(file);
1560
1561 if (fuse_is_bad(inode))
1562 return -EIO;
1563
1564 if (FUSE_IS_DAX(inode))
1565 return fuse_dax_read_iter(iocb, to);
1566
1567 if (!(ff->open_flags & FOPEN_DIRECT_IO))
1568 return fuse_cache_read_iter(iocb, to);
1569 else
1570 return fuse_direct_read_iter(iocb, to);
1571 }
1572
1573 static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1574 {
1575 struct file *file = iocb->ki_filp;
1576 struct fuse_file *ff = file->private_data;
1577 struct inode *inode = file_inode(file);
1578
1579 if (fuse_is_bad(inode))
1580 return -EIO;
1581
1582 if (FUSE_IS_DAX(inode))
1583 return fuse_dax_write_iter(iocb, from);
1584
1585 if (!(ff->open_flags & FOPEN_DIRECT_IO))
1586 return fuse_cache_write_iter(iocb, from);
1587 else
1588 return fuse_direct_write_iter(iocb, from);
1589 }
1590
1591 static void fuse_writepage_free(struct fuse_writepage_args *wpa)
1592 {
1593 struct fuse_args_pages *ap = &wpa->ia.ap;
1594 int i;
1595
1596 for (i = 0; i < ap->num_pages; i++)
1597 __free_page(ap->pages[i]);
1598
1599 if (wpa->ia.ff)
1600 fuse_file_put(wpa->ia.ff, false, false);
1601
1602 kfree(ap->pages);
1603 kfree(wpa);
1604 }
1605
1606 static void fuse_writepage_finish(struct fuse_mount *fm,
1607 struct fuse_writepage_args *wpa)
1608 {
1609 struct fuse_args_pages *ap = &wpa->ia.ap;
1610 struct inode *inode = wpa->inode;
1611 struct fuse_inode *fi = get_fuse_inode(inode);
1612 struct backing_dev_info *bdi = inode_to_bdi(inode);
1613 int i;
1614
1615 for (i = 0; i < ap->num_pages; i++) {
1616 dec_wb_stat(&bdi->wb, WB_WRITEBACK);
1617 dec_node_page_state(ap->pages[i], NR_WRITEBACK_TEMP);
1618 wb_writeout_inc(&bdi->wb);
1619 }
1620 wake_up(&fi->page_waitq);
1621 }
1622
1623 /* Called under fi->lock, may release and reacquire it */
1624 static void fuse_send_writepage(struct fuse_mount *fm,
1625 struct fuse_writepage_args *wpa, loff_t size)
1626 __releases(fi->lock)
1627 __acquires(fi->lock)
1628 {
1629 struct fuse_writepage_args *aux, *next;
1630 struct fuse_inode *fi = get_fuse_inode(wpa->inode);
1631 struct fuse_write_in *inarg = &wpa->ia.write.in;
1632 struct fuse_args *args = &wpa->ia.ap.args;
1633 __u64 data_size = wpa->ia.ap.num_pages * PAGE_SIZE;
1634 int err;
1635
1636 fi->writectr++;
1637 if (inarg->offset + data_size <= size) {
1638 inarg->size = data_size;
1639 } else if (inarg->offset < size) {
1640 inarg->size = size - inarg->offset;
1641 } else {
1642 /* Got truncated off completely */
1643 goto out_free;
1644 }
1645
1646 args->in_args[1].size = inarg->size;
1647 args->force = true;
1648 args->nocreds = true;
1649
1650 err = fuse_simple_background(fm, args, GFP_ATOMIC);
1651 if (err == -ENOMEM) {
1652 spin_unlock(&fi->lock);
1653 err = fuse_simple_background(fm, args, GFP_NOFS | __GFP_NOFAIL);
1654 spin_lock(&fi->lock);
1655 }
1656
1657 /* Fails on broken connection only */
1658 if (unlikely(err))
1659 goto out_free;
1660
1661 return;
1662
1663 out_free:
1664 fi->writectr--;
1665 rb_erase(&wpa->writepages_entry, &fi->writepages);
1666 fuse_writepage_finish(fm, wpa);
1667 spin_unlock(&fi->lock);
1668
1669 /* After fuse_writepage_finish() aux request list is private */
1670 for (aux = wpa->next; aux; aux = next) {
1671 next = aux->next;
1672 aux->next = NULL;
1673 fuse_writepage_free(aux);
1674 }
1675
1676 fuse_writepage_free(wpa);
1677 spin_lock(&fi->lock);
1678 }
1679
1680 /*
1681 * If fi->writectr is positive (no truncate or fsync going on) send
1682 * all queued writepage requests.
1683 *
1684 * Called with fi->lock
1685 */
1686 void fuse_flush_writepages(struct inode *inode)
1687 __releases(fi->lock)
1688 __acquires(fi->lock)
1689 {
1690 struct fuse_mount *fm = get_fuse_mount(inode);
1691 struct fuse_inode *fi = get_fuse_inode(inode);
1692 loff_t crop = i_size_read(inode);
1693 struct fuse_writepage_args *wpa;
1694
1695 while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
1696 wpa = list_entry(fi->queued_writes.next,
1697 struct fuse_writepage_args, queue_entry);
1698 list_del_init(&wpa->queue_entry);
1699 fuse_send_writepage(fm, wpa, crop);
1700 }
1701 }
1702
1703 static struct fuse_writepage_args *fuse_insert_writeback(struct rb_root *root,
1704 struct fuse_writepage_args *wpa)
1705 {
1706 pgoff_t idx_from = wpa->ia.write.in.offset >> PAGE_SHIFT;
1707 pgoff_t idx_to = idx_from + wpa->ia.ap.num_pages - 1;
1708 struct rb_node **p = &root->rb_node;
1709 struct rb_node *parent = NULL;
1710
1711 WARN_ON(!wpa->ia.ap.num_pages);
1712 while (*p) {
1713 struct fuse_writepage_args *curr;
1714 pgoff_t curr_index;
1715
1716 parent = *p;
1717 curr = rb_entry(parent, struct fuse_writepage_args,
1718 writepages_entry);
1719 WARN_ON(curr->inode != wpa->inode);
1720 curr_index = curr->ia.write.in.offset >> PAGE_SHIFT;
1721
1722 if (idx_from >= curr_index + curr->ia.ap.num_pages)
1723 p = &(*p)->rb_right;
1724 else if (idx_to < curr_index)
1725 p = &(*p)->rb_left;
1726 else
1727 return curr;
1728 }
1729
1730 rb_link_node(&wpa->writepages_entry, parent, p);
1731 rb_insert_color(&wpa->writepages_entry, root);
1732 return NULL;
1733 }
1734
1735 static void tree_insert(struct rb_root *root, struct fuse_writepage_args *wpa)
1736 {
1737 WARN_ON(fuse_insert_writeback(root, wpa));
1738 }
1739
1740 static void fuse_writepage_end(struct fuse_mount *fm, struct fuse_args *args,
1741 int error)
1742 {
1743 struct fuse_writepage_args *wpa =
1744 container_of(args, typeof(*wpa), ia.ap.args);
1745 struct inode *inode = wpa->inode;
1746 struct fuse_inode *fi = get_fuse_inode(inode);
1747
1748 mapping_set_error(inode->i_mapping, error);
1749 spin_lock(&fi->lock);
1750 rb_erase(&wpa->writepages_entry, &fi->writepages);
1751 while (wpa->next) {
1752 struct fuse_mount *fm = get_fuse_mount(inode);
1753 struct fuse_write_in *inarg = &wpa->ia.write.in;
1754 struct fuse_writepage_args *next = wpa->next;
1755
1756 wpa->next = next->next;
1757 next->next = NULL;
1758 next->ia.ff = fuse_file_get(wpa->ia.ff);
1759 tree_insert(&fi->writepages, next);
1760
1761 /*
1762 * Skip fuse_flush_writepages() to make it easy to crop requests
1763 * based on primary request size.
1764 *
1765 * 1st case (trivial): there are no concurrent activities using
1766 * fuse_set/release_nowrite. Then we're on safe side because
1767 * fuse_flush_writepages() would call fuse_send_writepage()
1768 * anyway.
1769 *
1770 * 2nd case: someone called fuse_set_nowrite and it is waiting
1771 * now for completion of all in-flight requests. This happens
1772 * rarely and no more than once per page, so this should be
1773 * okay.
1774 *
1775 * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle
1776 * of fuse_set_nowrite..fuse_release_nowrite section. The fact
1777 * that fuse_set_nowrite returned implies that all in-flight
1778 * requests were completed along with all of their secondary
1779 * requests. Further primary requests are blocked by negative
1780 * writectr. Hence there cannot be any in-flight requests and
1781 * no invocations of fuse_writepage_end() while we're in
1782 * fuse_set_nowrite..fuse_release_nowrite section.
1783 */
1784 fuse_send_writepage(fm, next, inarg->offset + inarg->size);
1785 }
1786 fi->writectr--;
1787 fuse_writepage_finish(fm, wpa);
1788 spin_unlock(&fi->lock);
1789 fuse_writepage_free(wpa);
1790 }
1791
1792 static struct fuse_file *__fuse_write_file_get(struct fuse_conn *fc,
1793 struct fuse_inode *fi)
1794 {
1795 struct fuse_file *ff = NULL;
1796
1797 spin_lock(&fi->lock);
1798 if (!list_empty(&fi->write_files)) {
1799 ff = list_entry(fi->write_files.next, struct fuse_file,
1800 write_entry);
1801 fuse_file_get(ff);
1802 }
1803 spin_unlock(&fi->lock);
1804
1805 return ff;
1806 }
1807
1808 static struct fuse_file *fuse_write_file_get(struct fuse_conn *fc,
1809 struct fuse_inode *fi)
1810 {
1811 struct fuse_file *ff = __fuse_write_file_get(fc, fi);
1812 WARN_ON(!ff);
1813 return ff;
1814 }
1815
1816 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc)
1817 {
1818 struct fuse_conn *fc = get_fuse_conn(inode);
1819 struct fuse_inode *fi = get_fuse_inode(inode);
1820 struct fuse_file *ff;
1821 int err;
1822
1823 ff = __fuse_write_file_get(fc, fi);
1824 err = fuse_flush_times(inode, ff);
1825 if (ff)
1826 fuse_file_put(ff, false, false);
1827
1828 return err;
1829 }
1830
1831 static struct fuse_writepage_args *fuse_writepage_args_alloc(void)
1832 {
1833 struct fuse_writepage_args *wpa;
1834 struct fuse_args_pages *ap;
1835
1836 wpa = kzalloc(sizeof(*wpa), GFP_NOFS);
1837 if (wpa) {
1838 ap = &wpa->ia.ap;
1839 ap->num_pages = 0;
1840 ap->pages = fuse_pages_alloc(1, GFP_NOFS, &ap->descs);
1841 if (!ap->pages) {
1842 kfree(wpa);
1843 wpa = NULL;
1844 }
1845 }
1846 return wpa;
1847
1848 }
1849
1850 static int fuse_writepage_locked(struct page *page)
1851 {
1852 struct address_space *mapping = page->mapping;
1853 struct inode *inode = mapping->host;
1854 struct fuse_conn *fc = get_fuse_conn(inode);
1855 struct fuse_inode *fi = get_fuse_inode(inode);
1856 struct fuse_writepage_args *wpa;
1857 struct fuse_args_pages *ap;
1858 struct page *tmp_page;
1859 int error = -ENOMEM;
1860
1861 set_page_writeback(page);
1862
1863 wpa = fuse_writepage_args_alloc();
1864 if (!wpa)
1865 goto err;
1866 ap = &wpa->ia.ap;
1867
1868 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1869 if (!tmp_page)
1870 goto err_free;
1871
1872 error = -EIO;
1873 wpa->ia.ff = fuse_write_file_get(fc, fi);
1874 if (!wpa->ia.ff)
1875 goto err_nofile;
1876
1877 fuse_write_args_fill(&wpa->ia, wpa->ia.ff, page_offset(page), 0);
1878
1879 copy_highpage(tmp_page, page);
1880 wpa->ia.write.in.write_flags |= FUSE_WRITE_CACHE;
1881 wpa->next = NULL;
1882 ap->args.in_pages = true;
1883 ap->num_pages = 1;
1884 ap->pages[0] = tmp_page;
1885 ap->descs[0].offset = 0;
1886 ap->descs[0].length = PAGE_SIZE;
1887 ap->args.end = fuse_writepage_end;
1888 wpa->inode = inode;
1889
1890 inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
1891 inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP);
1892
1893 spin_lock(&fi->lock);
1894 tree_insert(&fi->writepages, wpa);
1895 list_add_tail(&wpa->queue_entry, &fi->queued_writes);
1896 fuse_flush_writepages(inode);
1897 spin_unlock(&fi->lock);
1898
1899 end_page_writeback(page);
1900
1901 return 0;
1902
1903 err_nofile:
1904 __free_page(tmp_page);
1905 err_free:
1906 kfree(wpa);
1907 err:
1908 mapping_set_error(page->mapping, error);
1909 end_page_writeback(page);
1910 return error;
1911 }
1912
1913 static int fuse_writepage(struct page *page, struct writeback_control *wbc)
1914 {
1915 int err;
1916
1917 if (fuse_page_is_writeback(page->mapping->host, page->index)) {
1918 /*
1919 * ->writepages() should be called for sync() and friends. We
1920 * should only get here on direct reclaim and then we are
1921 * allowed to skip a page which is already in flight
1922 */
1923 WARN_ON(wbc->sync_mode == WB_SYNC_ALL);
1924
1925 redirty_page_for_writepage(wbc, page);
1926 unlock_page(page);
1927 return 0;
1928 }
1929
1930 err = fuse_writepage_locked(page);
1931 unlock_page(page);
1932
1933 return err;
1934 }
1935
1936 struct fuse_fill_wb_data {
1937 struct fuse_writepage_args *wpa;
1938 struct fuse_file *ff;
1939 struct inode *inode;
1940 struct page **orig_pages;
1941 unsigned int max_pages;
1942 };
1943
1944 static bool fuse_pages_realloc(struct fuse_fill_wb_data *data)
1945 {
1946 struct fuse_args_pages *ap = &data->wpa->ia.ap;
1947 struct fuse_conn *fc = get_fuse_conn(data->inode);
1948 struct page **pages;
1949 struct fuse_page_desc *descs;
1950 unsigned int npages = min_t(unsigned int,
1951 max_t(unsigned int, data->max_pages * 2,
1952 FUSE_DEFAULT_MAX_PAGES_PER_REQ),
1953 fc->max_pages);
1954 WARN_ON(npages <= data->max_pages);
1955
1956 pages = fuse_pages_alloc(npages, GFP_NOFS, &descs);
1957 if (!pages)
1958 return false;
1959
1960 memcpy(pages, ap->pages, sizeof(struct page *) * ap->num_pages);
1961 memcpy(descs, ap->descs, sizeof(struct fuse_page_desc) * ap->num_pages);
1962 kfree(ap->pages);
1963 ap->pages = pages;
1964 ap->descs = descs;
1965 data->max_pages = npages;
1966
1967 return true;
1968 }
1969
1970 static void fuse_writepages_send(struct fuse_fill_wb_data *data)
1971 {
1972 struct fuse_writepage_args *wpa = data->wpa;
1973 struct inode *inode = data->inode;
1974 struct fuse_inode *fi = get_fuse_inode(inode);
1975 int num_pages = wpa->ia.ap.num_pages;
1976 int i;
1977
1978 wpa->ia.ff = fuse_file_get(data->ff);
1979 spin_lock(&fi->lock);
1980 list_add_tail(&wpa->queue_entry, &fi->queued_writes);
1981 fuse_flush_writepages(inode);
1982 spin_unlock(&fi->lock);
1983
1984 for (i = 0; i < num_pages; i++)
1985 end_page_writeback(data->orig_pages[i]);
1986 }
1987
1988 /*
1989 * Check under fi->lock if the page is under writeback, and insert it onto the
1990 * rb_tree if not. Otherwise iterate auxiliary write requests, to see if there's
1991 * one already added for a page at this offset. If there's none, then insert
1992 * this new request onto the auxiliary list, otherwise reuse the existing one by
1993 * swapping the new temp page with the old one.
1994 */
1995 static bool fuse_writepage_add(struct fuse_writepage_args *new_wpa,
1996 struct page *page)
1997 {
1998 struct fuse_inode *fi = get_fuse_inode(new_wpa->inode);
1999 struct fuse_writepage_args *tmp;
2000 struct fuse_writepage_args *old_wpa;
2001 struct fuse_args_pages *new_ap = &new_wpa->ia.ap;
2002
2003 WARN_ON(new_ap->num_pages != 0);
2004 new_ap->num_pages = 1;
2005
2006 spin_lock(&fi->lock);
2007 old_wpa = fuse_insert_writeback(&fi->writepages, new_wpa);
2008 if (!old_wpa) {
2009 spin_unlock(&fi->lock);
2010 return true;
2011 }
2012
2013 for (tmp = old_wpa->next; tmp; tmp = tmp->next) {
2014 pgoff_t curr_index;
2015
2016 WARN_ON(tmp->inode != new_wpa->inode);
2017 curr_index = tmp->ia.write.in.offset >> PAGE_SHIFT;
2018 if (curr_index == page->index) {
2019 WARN_ON(tmp->ia.ap.num_pages != 1);
2020 swap(tmp->ia.ap.pages[0], new_ap->pages[0]);
2021 break;
2022 }
2023 }
2024
2025 if (!tmp) {
2026 new_wpa->next = old_wpa->next;
2027 old_wpa->next = new_wpa;
2028 }
2029
2030 spin_unlock(&fi->lock);
2031
2032 if (tmp) {
2033 struct backing_dev_info *bdi = inode_to_bdi(new_wpa->inode);
2034
2035 dec_wb_stat(&bdi->wb, WB_WRITEBACK);
2036 dec_node_page_state(new_ap->pages[0], NR_WRITEBACK_TEMP);
2037 wb_writeout_inc(&bdi->wb);
2038 fuse_writepage_free(new_wpa);
2039 }
2040
2041 return false;
2042 }
2043
2044 static bool fuse_writepage_need_send(struct fuse_conn *fc, struct page *page,
2045 struct fuse_args_pages *ap,
2046 struct fuse_fill_wb_data *data)
2047 {
2048 WARN_ON(!ap->num_pages);
2049
2050 /*
2051 * Being under writeback is unlikely but possible. For example direct
2052 * read to an mmaped fuse file will set the page dirty twice; once when
2053 * the pages are faulted with get_user_pages(), and then after the read
2054 * completed.
2055 */
2056 if (fuse_page_is_writeback(data->inode, page->index))
2057 return true;
2058
2059 /* Reached max pages */
2060 if (ap->num_pages == fc->max_pages)
2061 return true;
2062
2063 /* Reached max write bytes */
2064 if ((ap->num_pages + 1) * PAGE_SIZE > fc->max_write)
2065 return true;
2066
2067 /* Discontinuity */
2068 if (data->orig_pages[ap->num_pages - 1]->index + 1 != page->index)
2069 return true;
2070
2071 /* Need to grow the pages array? If so, did the expansion fail? */
2072 if (ap->num_pages == data->max_pages && !fuse_pages_realloc(data))
2073 return true;
2074
2075 return false;
2076 }
2077
2078 static int fuse_writepages_fill(struct page *page,
2079 struct writeback_control *wbc, void *_data)
2080 {
2081 struct fuse_fill_wb_data *data = _data;
2082 struct fuse_writepage_args *wpa = data->wpa;
2083 struct fuse_args_pages *ap = &wpa->ia.ap;
2084 struct inode *inode = data->inode;
2085 struct fuse_inode *fi = get_fuse_inode(inode);
2086 struct fuse_conn *fc = get_fuse_conn(inode);
2087 struct page *tmp_page;
2088 int err;
2089
2090 if (!data->ff) {
2091 err = -EIO;
2092 data->ff = fuse_write_file_get(fc, fi);
2093 if (!data->ff)
2094 goto out_unlock;
2095 }
2096
2097 if (wpa && fuse_writepage_need_send(fc, page, ap, data)) {
2098 fuse_writepages_send(data);
2099 data->wpa = NULL;
2100 }
2101
2102 err = -ENOMEM;
2103 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
2104 if (!tmp_page)
2105 goto out_unlock;
2106
2107 /*
2108 * The page must not be redirtied until the writeout is completed
2109 * (i.e. userspace has sent a reply to the write request). Otherwise
2110 * there could be more than one temporary page instance for each real
2111 * page.
2112 *
2113 * This is ensured by holding the page lock in page_mkwrite() while
2114 * checking fuse_page_is_writeback(). We already hold the page lock
2115 * since clear_page_dirty_for_io() and keep it held until we add the
2116 * request to the fi->writepages list and increment ap->num_pages.
2117 * After this fuse_page_is_writeback() will indicate that the page is
2118 * under writeback, so we can release the page lock.
2119 */
2120 if (data->wpa == NULL) {
2121 err = -ENOMEM;
2122 wpa = fuse_writepage_args_alloc();
2123 if (!wpa) {
2124 __free_page(tmp_page);
2125 goto out_unlock;
2126 }
2127 data->max_pages = 1;
2128
2129 ap = &wpa->ia.ap;
2130 fuse_write_args_fill(&wpa->ia, data->ff, page_offset(page), 0);
2131 wpa->ia.write.in.write_flags |= FUSE_WRITE_CACHE;
2132 wpa->next = NULL;
2133 ap->args.in_pages = true;
2134 ap->args.end = fuse_writepage_end;
2135 ap->num_pages = 0;
2136 wpa->inode = inode;
2137 }
2138 set_page_writeback(page);
2139
2140 copy_highpage(tmp_page, page);
2141 ap->pages[ap->num_pages] = tmp_page;
2142 ap->descs[ap->num_pages].offset = 0;
2143 ap->descs[ap->num_pages].length = PAGE_SIZE;
2144 data->orig_pages[ap->num_pages] = page;
2145
2146 inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
2147 inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP);
2148
2149 err = 0;
2150 if (data->wpa) {
2151 /*
2152 * Protected by fi->lock against concurrent access by
2153 * fuse_page_is_writeback().
2154 */
2155 spin_lock(&fi->lock);
2156 ap->num_pages++;
2157 spin_unlock(&fi->lock);
2158 } else if (fuse_writepage_add(wpa, page)) {
2159 data->wpa = wpa;
2160 } else {
2161 end_page_writeback(page);
2162 }
2163 out_unlock:
2164 unlock_page(page);
2165
2166 return err;
2167 }
2168
2169 static int fuse_writepages(struct address_space *mapping,
2170 struct writeback_control *wbc)
2171 {
2172 struct inode *inode = mapping->host;
2173 struct fuse_conn *fc = get_fuse_conn(inode);
2174 struct fuse_fill_wb_data data;
2175 int err;
2176
2177 err = -EIO;
2178 if (fuse_is_bad(inode))
2179 goto out;
2180
2181 data.inode = inode;
2182 data.wpa = NULL;
2183 data.ff = NULL;
2184
2185 err = -ENOMEM;
2186 data.orig_pages = kcalloc(fc->max_pages,
2187 sizeof(struct page *),
2188 GFP_NOFS);
2189 if (!data.orig_pages)
2190 goto out;
2191
2192 err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data);
2193 if (data.wpa) {
2194 WARN_ON(!data.wpa->ia.ap.num_pages);
2195 fuse_writepages_send(&data);
2196 }
2197 if (data.ff)
2198 fuse_file_put(data.ff, false, false);
2199
2200 kfree(data.orig_pages);
2201 out:
2202 return err;
2203 }
2204
2205 /*
2206 * It's worthy to make sure that space is reserved on disk for the write,
2207 * but how to implement it without killing performance need more thinking.
2208 */
2209 static int fuse_write_begin(struct file *file, struct address_space *mapping,
2210 loff_t pos, unsigned len, unsigned flags,
2211 struct page **pagep, void **fsdata)
2212 {
2213 pgoff_t index = pos >> PAGE_SHIFT;
2214 struct fuse_conn *fc = get_fuse_conn(file_inode(file));
2215 struct page *page;
2216 loff_t fsize;
2217 int err = -ENOMEM;
2218
2219 WARN_ON(!fc->writeback_cache);
2220
2221 page = grab_cache_page_write_begin(mapping, index, flags);
2222 if (!page)
2223 goto error;
2224
2225 fuse_wait_on_page_writeback(mapping->host, page->index);
2226
2227 if (PageUptodate(page) || len == PAGE_SIZE)
2228 goto success;
2229 /*
2230 * Check if the start this page comes after the end of file, in which
2231 * case the readpage can be optimized away.
2232 */
2233 fsize = i_size_read(mapping->host);
2234 if (fsize <= (pos & PAGE_MASK)) {
2235 size_t off = pos & ~PAGE_MASK;
2236 if (off)
2237 zero_user_segment(page, 0, off);
2238 goto success;
2239 }
2240 err = fuse_do_readpage(file, page);
2241 if (err)
2242 goto cleanup;
2243 success:
2244 *pagep = page;
2245 return 0;
2246
2247 cleanup:
2248 unlock_page(page);
2249 put_page(page);
2250 error:
2251 return err;
2252 }
2253
2254 static int fuse_write_end(struct file *file, struct address_space *mapping,
2255 loff_t pos, unsigned len, unsigned copied,
2256 struct page *page, void *fsdata)
2257 {
2258 struct inode *inode = page->mapping->host;
2259
2260 /* Haven't copied anything? Skip zeroing, size extending, dirtying. */
2261 if (!copied)
2262 goto unlock;
2263
2264 if (!PageUptodate(page)) {
2265 /* Zero any unwritten bytes at the end of the page */
2266 size_t endoff = (pos + copied) & ~PAGE_MASK;
2267 if (endoff)
2268 zero_user_segment(page, endoff, PAGE_SIZE);
2269 SetPageUptodate(page);
2270 }
2271
2272 fuse_write_update_size(inode, pos + copied);
2273 set_page_dirty(page);
2274
2275 unlock:
2276 unlock_page(page);
2277 put_page(page);
2278
2279 return copied;
2280 }
2281
2282 static int fuse_launder_page(struct page *page)
2283 {
2284 int err = 0;
2285 if (clear_page_dirty_for_io(page)) {
2286 struct inode *inode = page->mapping->host;
2287 err = fuse_writepage_locked(page);
2288 if (!err)
2289 fuse_wait_on_page_writeback(inode, page->index);
2290 }
2291 return err;
2292 }
2293
2294 /*
2295 * Write back dirty pages now, because there may not be any suitable
2296 * open files later
2297 */
2298 static void fuse_vma_close(struct vm_area_struct *vma)
2299 {
2300 filemap_write_and_wait(vma->vm_file->f_mapping);
2301 }
2302
2303 /*
2304 * Wait for writeback against this page to complete before allowing it
2305 * to be marked dirty again, and hence written back again, possibly
2306 * before the previous writepage completed.
2307 *
2308 * Block here, instead of in ->writepage(), so that the userspace fs
2309 * can only block processes actually operating on the filesystem.
2310 *
2311 * Otherwise unprivileged userspace fs would be able to block
2312 * unrelated:
2313 *
2314 * - page migration
2315 * - sync(2)
2316 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
2317 */
2318 static vm_fault_t fuse_page_mkwrite(struct vm_fault *vmf)
2319 {
2320 struct page *page = vmf->page;
2321 struct inode *inode = file_inode(vmf->vma->vm_file);
2322
2323 file_update_time(vmf->vma->vm_file);
2324 lock_page(page);
2325 if (page->mapping != inode->i_mapping) {
2326 unlock_page(page);
2327 return VM_FAULT_NOPAGE;
2328 }
2329
2330 fuse_wait_on_page_writeback(inode, page->index);
2331 return VM_FAULT_LOCKED;
2332 }
2333
2334 static const struct vm_operations_struct fuse_file_vm_ops = {
2335 .close = fuse_vma_close,
2336 .fault = filemap_fault,
2337 .map_pages = filemap_map_pages,
2338 .page_mkwrite = fuse_page_mkwrite,
2339 };
2340
2341 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
2342 {
2343 struct fuse_file *ff = file->private_data;
2344
2345 /* DAX mmap is superior to direct_io mmap */
2346 if (FUSE_IS_DAX(file_inode(file)))
2347 return fuse_dax_mmap(file, vma);
2348
2349 if (ff->open_flags & FOPEN_DIRECT_IO) {
2350 /* Can't provide the coherency needed for MAP_SHARED */
2351 if (vma->vm_flags & VM_MAYSHARE)
2352 return -ENODEV;
2353
2354 invalidate_inode_pages2(file->f_mapping);
2355
2356 return generic_file_mmap(file, vma);
2357 }
2358
2359 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2360 fuse_link_write_file(file);
2361
2362 file_accessed(file);
2363 vma->vm_ops = &fuse_file_vm_ops;
2364 return 0;
2365 }
2366
2367 static int convert_fuse_file_lock(struct fuse_conn *fc,
2368 const struct fuse_file_lock *ffl,
2369 struct file_lock *fl)
2370 {
2371 switch (ffl->type) {
2372 case F_UNLCK:
2373 break;
2374
2375 case F_RDLCK:
2376 case F_WRLCK:
2377 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
2378 ffl->end < ffl->start)
2379 return -EIO;
2380
2381 fl->fl_start = ffl->start;
2382 fl->fl_end = ffl->end;
2383
2384 /*
2385 * Convert pid into init's pid namespace. The locks API will
2386 * translate it into the caller's pid namespace.
2387 */
2388 rcu_read_lock();
2389 fl->fl_pid = pid_nr_ns(find_pid_ns(ffl->pid, fc->pid_ns), &init_pid_ns);
2390 rcu_read_unlock();
2391 break;
2392
2393 default:
2394 return -EIO;
2395 }
2396 fl->fl_type = ffl->type;
2397 return 0;
2398 }
2399
2400 static void fuse_lk_fill(struct fuse_args *args, struct file *file,
2401 const struct file_lock *fl, int opcode, pid_t pid,
2402 int flock, struct fuse_lk_in *inarg)
2403 {
2404 struct inode *inode = file_inode(file);
2405 struct fuse_conn *fc = get_fuse_conn(inode);
2406 struct fuse_file *ff = file->private_data;
2407
2408 memset(inarg, 0, sizeof(*inarg));
2409 inarg->fh = ff->fh;
2410 inarg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
2411 inarg->lk.start = fl->fl_start;
2412 inarg->lk.end = fl->fl_end;
2413 inarg->lk.type = fl->fl_type;
2414 inarg->lk.pid = pid;
2415 if (flock)
2416 inarg->lk_flags |= FUSE_LK_FLOCK;
2417 args->opcode = opcode;
2418 args->nodeid = get_node_id(inode);
2419 args->in_numargs = 1;
2420 args->in_args[0].size = sizeof(*inarg);
2421 args->in_args[0].value = inarg;
2422 }
2423
2424 static int fuse_getlk(struct file *file, struct file_lock *fl)
2425 {
2426 struct inode *inode = file_inode(file);
2427 struct fuse_mount *fm = get_fuse_mount(inode);
2428 FUSE_ARGS(args);
2429 struct fuse_lk_in inarg;
2430 struct fuse_lk_out outarg;
2431 int err;
2432
2433 fuse_lk_fill(&args, file, fl, FUSE_GETLK, 0, 0, &inarg);
2434 args.out_numargs = 1;
2435 args.out_args[0].size = sizeof(outarg);
2436 args.out_args[0].value = &outarg;
2437 err = fuse_simple_request(fm, &args);
2438 if (!err)
2439 err = convert_fuse_file_lock(fm->fc, &outarg.lk, fl);
2440
2441 return err;
2442 }
2443
2444 static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
2445 {
2446 struct inode *inode = file_inode(file);
2447 struct fuse_mount *fm = get_fuse_mount(inode);
2448 FUSE_ARGS(args);
2449 struct fuse_lk_in inarg;
2450 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
2451 struct pid *pid = fl->fl_type != F_UNLCK ? task_tgid(current) : NULL;
2452 pid_t pid_nr = pid_nr_ns(pid, fm->fc->pid_ns);
2453 int err;
2454
2455 if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
2456 /* NLM needs asynchronous locks, which we don't support yet */
2457 return -ENOLCK;
2458 }
2459
2460 /* Unlock on close is handled by the flush method */
2461 if ((fl->fl_flags & FL_CLOSE_POSIX) == FL_CLOSE_POSIX)
2462 return 0;
2463
2464 fuse_lk_fill(&args, file, fl, opcode, pid_nr, flock, &inarg);
2465 err = fuse_simple_request(fm, &args);
2466
2467 /* locking is restartable */
2468 if (err == -EINTR)
2469 err = -ERESTARTSYS;
2470
2471 return err;
2472 }
2473
2474 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
2475 {
2476 struct inode *inode = file_inode(file);
2477 struct fuse_conn *fc = get_fuse_conn(inode);
2478 int err;
2479
2480 if (cmd == F_CANCELLK) {
2481 err = 0;
2482 } else if (cmd == F_GETLK) {
2483 if (fc->no_lock) {
2484 posix_test_lock(file, fl);
2485 err = 0;
2486 } else
2487 err = fuse_getlk(file, fl);
2488 } else {
2489 if (fc->no_lock)
2490 err = posix_lock_file(file, fl, NULL);
2491 else
2492 err = fuse_setlk(file, fl, 0);
2493 }
2494 return err;
2495 }
2496
2497 static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
2498 {
2499 struct inode *inode = file_inode(file);
2500 struct fuse_conn *fc = get_fuse_conn(inode);
2501 int err;
2502
2503 if (fc->no_flock) {
2504 err = locks_lock_file_wait(file, fl);
2505 } else {
2506 struct fuse_file *ff = file->private_data;
2507
2508 /* emulate flock with POSIX locks */
2509 ff->flock = true;
2510 err = fuse_setlk(file, fl, 1);
2511 }
2512
2513 return err;
2514 }
2515
2516 static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
2517 {
2518 struct inode *inode = mapping->host;
2519 struct fuse_mount *fm = get_fuse_mount(inode);
2520 FUSE_ARGS(args);
2521 struct fuse_bmap_in inarg;
2522 struct fuse_bmap_out outarg;
2523 int err;
2524
2525 if (!inode->i_sb->s_bdev || fm->fc->no_bmap)
2526 return 0;
2527
2528 memset(&inarg, 0, sizeof(inarg));
2529 inarg.block = block;
2530 inarg.blocksize = inode->i_sb->s_blocksize;
2531 args.opcode = FUSE_BMAP;
2532 args.nodeid = get_node_id(inode);
2533 args.in_numargs = 1;
2534 args.in_args[0].size = sizeof(inarg);
2535 args.in_args[0].value = &inarg;
2536 args.out_numargs = 1;
2537 args.out_args[0].size = sizeof(outarg);
2538 args.out_args[0].value = &outarg;
2539 err = fuse_simple_request(fm, &args);
2540 if (err == -ENOSYS)
2541 fm->fc->no_bmap = 1;
2542
2543 return err ? 0 : outarg.block;
2544 }
2545
2546 static loff_t fuse_lseek(struct file *file, loff_t offset, int whence)
2547 {
2548 struct inode *inode = file->f_mapping->host;
2549 struct fuse_mount *fm = get_fuse_mount(inode);
2550 struct fuse_file *ff = file->private_data;
2551 FUSE_ARGS(args);
2552 struct fuse_lseek_in inarg = {
2553 .fh = ff->fh,
2554 .offset = offset,
2555 .whence = whence
2556 };
2557 struct fuse_lseek_out outarg;
2558 int err;
2559
2560 if (fm->fc->no_lseek)
2561 goto fallback;
2562
2563 args.opcode = FUSE_LSEEK;
2564 args.nodeid = ff->nodeid;
2565 args.in_numargs = 1;
2566 args.in_args[0].size = sizeof(inarg);
2567 args.in_args[0].value = &inarg;
2568 args.out_numargs = 1;
2569 args.out_args[0].size = sizeof(outarg);
2570 args.out_args[0].value = &outarg;
2571 err = fuse_simple_request(fm, &args);
2572 if (err) {
2573 if (err == -ENOSYS) {
2574 fm->fc->no_lseek = 1;
2575 goto fallback;
2576 }
2577 return err;
2578 }
2579
2580 return vfs_setpos(file, outarg.offset, inode->i_sb->s_maxbytes);
2581
2582 fallback:
2583 err = fuse_update_attributes(inode, file);
2584 if (!err)
2585 return generic_file_llseek(file, offset, whence);
2586 else
2587 return err;
2588 }
2589
2590 static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence)
2591 {
2592 loff_t retval;
2593 struct inode *inode = file_inode(file);
2594
2595 switch (whence) {
2596 case SEEK_SET:
2597 case SEEK_CUR:
2598 /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */
2599 retval = generic_file_llseek(file, offset, whence);
2600 break;
2601 case SEEK_END:
2602 inode_lock(inode);
2603 retval = fuse_update_attributes(inode, file);
2604 if (!retval)
2605 retval = generic_file_llseek(file, offset, whence);
2606 inode_unlock(inode);
2607 break;
2608 case SEEK_HOLE:
2609 case SEEK_DATA:
2610 inode_lock(inode);
2611 retval = fuse_lseek(file, offset, whence);
2612 inode_unlock(inode);
2613 break;
2614 default:
2615 retval = -EINVAL;
2616 }
2617
2618 return retval;
2619 }
2620
2621 /*
2622 * CUSE servers compiled on 32bit broke on 64bit kernels because the
2623 * ABI was defined to be 'struct iovec' which is different on 32bit
2624 * and 64bit. Fortunately we can determine which structure the server
2625 * used from the size of the reply.
2626 */
2627 static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
2628 size_t transferred, unsigned count,
2629 bool is_compat)
2630 {
2631 #ifdef CONFIG_COMPAT
2632 if (count * sizeof(struct compat_iovec) == transferred) {
2633 struct compat_iovec *ciov = src;
2634 unsigned i;
2635
2636 /*
2637 * With this interface a 32bit server cannot support
2638 * non-compat (i.e. ones coming from 64bit apps) ioctl
2639 * requests
2640 */
2641 if (!is_compat)
2642 return -EINVAL;
2643
2644 for (i = 0; i < count; i++) {
2645 dst[i].iov_base = compat_ptr(ciov[i].iov_base);
2646 dst[i].iov_len = ciov[i].iov_len;
2647 }
2648 return 0;
2649 }
2650 #endif
2651
2652 if (count * sizeof(struct iovec) != transferred)
2653 return -EIO;
2654
2655 memcpy(dst, src, transferred);
2656 return 0;
2657 }
2658
2659 /* Make sure iov_length() won't overflow */
2660 static int fuse_verify_ioctl_iov(struct fuse_conn *fc, struct iovec *iov,
2661 size_t count)
2662 {
2663 size_t n;
2664 u32 max = fc->max_pages << PAGE_SHIFT;
2665
2666 for (n = 0; n < count; n++, iov++) {
2667 if (iov->iov_len > (size_t) max)
2668 return -ENOMEM;
2669 max -= iov->iov_len;
2670 }
2671 return 0;
2672 }
2673
2674 static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
2675 void *src, size_t transferred, unsigned count,
2676 bool is_compat)
2677 {
2678 unsigned i;
2679 struct fuse_ioctl_iovec *fiov = src;
2680
2681 if (fc->minor < 16) {
2682 return fuse_copy_ioctl_iovec_old(dst, src, transferred,
2683 count, is_compat);
2684 }
2685
2686 if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
2687 return -EIO;
2688
2689 for (i = 0; i < count; i++) {
2690 /* Did the server supply an inappropriate value? */
2691 if (fiov[i].base != (unsigned long) fiov[i].base ||
2692 fiov[i].len != (unsigned long) fiov[i].len)
2693 return -EIO;
2694
2695 dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
2696 dst[i].iov_len = (size_t) fiov[i].len;
2697
2698 #ifdef CONFIG_COMPAT
2699 if (is_compat &&
2700 (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
2701 (compat_size_t) dst[i].iov_len != fiov[i].len))
2702 return -EIO;
2703 #endif
2704 }
2705
2706 return 0;
2707 }
2708
2709
2710 /*
2711 * For ioctls, there is no generic way to determine how much memory
2712 * needs to be read and/or written. Furthermore, ioctls are allowed
2713 * to dereference the passed pointer, so the parameter requires deep
2714 * copying but FUSE has no idea whatsoever about what to copy in or
2715 * out.
2716 *
2717 * This is solved by allowing FUSE server to retry ioctl with
2718 * necessary in/out iovecs. Let's assume the ioctl implementation
2719 * needs to read in the following structure.
2720 *
2721 * struct a {
2722 * char *buf;
2723 * size_t buflen;
2724 * }
2725 *
2726 * On the first callout to FUSE server, inarg->in_size and
2727 * inarg->out_size will be NULL; then, the server completes the ioctl
2728 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
2729 * the actual iov array to
2730 *
2731 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } }
2732 *
2733 * which tells FUSE to copy in the requested area and retry the ioctl.
2734 * On the second round, the server has access to the structure and
2735 * from that it can tell what to look for next, so on the invocation,
2736 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
2737 *
2738 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) },
2739 * { .iov_base = a.buf, .iov_len = a.buflen } }
2740 *
2741 * FUSE will copy both struct a and the pointed buffer from the
2742 * process doing the ioctl and retry ioctl with both struct a and the
2743 * buffer.
2744 *
2745 * This time, FUSE server has everything it needs and completes ioctl
2746 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
2747 *
2748 * Copying data out works the same way.
2749 *
2750 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
2751 * automatically initializes in and out iovs by decoding @cmd with
2752 * _IOC_* macros and the server is not allowed to request RETRY. This
2753 * limits ioctl data transfers to well-formed ioctls and is the forced
2754 * behavior for all FUSE servers.
2755 */
2756 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
2757 unsigned int flags)
2758 {
2759 struct fuse_file *ff = file->private_data;
2760 struct fuse_mount *fm = ff->fm;
2761 struct fuse_ioctl_in inarg = {
2762 .fh = ff->fh,
2763 .cmd = cmd,
2764 .arg = arg,
2765 .flags = flags
2766 };
2767 struct fuse_ioctl_out outarg;
2768 struct iovec *iov_page = NULL;
2769 struct iovec *in_iov = NULL, *out_iov = NULL;
2770 unsigned int in_iovs = 0, out_iovs = 0, max_pages;
2771 size_t in_size, out_size, c;
2772 ssize_t transferred;
2773 int err, i;
2774 struct iov_iter ii;
2775 struct fuse_args_pages ap = {};
2776
2777 #if BITS_PER_LONG == 32
2778 inarg.flags |= FUSE_IOCTL_32BIT;
2779 #else
2780 if (flags & FUSE_IOCTL_COMPAT) {
2781 inarg.flags |= FUSE_IOCTL_32BIT;
2782 #ifdef CONFIG_X86_X32
2783 if (in_x32_syscall())
2784 inarg.flags |= FUSE_IOCTL_COMPAT_X32;
2785 #endif
2786 }
2787 #endif
2788
2789 /* assume all the iovs returned by client always fits in a page */
2790 BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
2791
2792 err = -ENOMEM;
2793 ap.pages = fuse_pages_alloc(fm->fc->max_pages, GFP_KERNEL, &ap.descs);
2794 iov_page = (struct iovec *) __get_free_page(GFP_KERNEL);
2795 if (!ap.pages || !iov_page)
2796 goto out;
2797
2798 fuse_page_descs_length_init(ap.descs, 0, fm->fc->max_pages);
2799
2800 /*
2801 * If restricted, initialize IO parameters as encoded in @cmd.
2802 * RETRY from server is not allowed.
2803 */
2804 if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
2805 struct iovec *iov = iov_page;
2806
2807 iov->iov_base = (void __user *)arg;
2808
2809 switch (cmd) {
2810 case FS_IOC_GETFLAGS:
2811 case FS_IOC_SETFLAGS:
2812 iov->iov_len = sizeof(int);
2813 break;
2814 default:
2815 iov->iov_len = _IOC_SIZE(cmd);
2816 break;
2817 }
2818
2819 if (_IOC_DIR(cmd) & _IOC_WRITE) {
2820 in_iov = iov;
2821 in_iovs = 1;
2822 }
2823
2824 if (_IOC_DIR(cmd) & _IOC_READ) {
2825 out_iov = iov;
2826 out_iovs = 1;
2827 }
2828 }
2829
2830 retry:
2831 inarg.in_size = in_size = iov_length(in_iov, in_iovs);
2832 inarg.out_size = out_size = iov_length(out_iov, out_iovs);
2833
2834 /*
2835 * Out data can be used either for actual out data or iovs,
2836 * make sure there always is at least one page.
2837 */
2838 out_size = max_t(size_t, out_size, PAGE_SIZE);
2839 max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
2840
2841 /* make sure there are enough buffer pages and init request with them */
2842 err = -ENOMEM;
2843 if (max_pages > fm->fc->max_pages)
2844 goto out;
2845 while (ap.num_pages < max_pages) {
2846 ap.pages[ap.num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2847 if (!ap.pages[ap.num_pages])
2848 goto out;
2849 ap.num_pages++;
2850 }
2851
2852
2853 /* okay, let's send it to the client */
2854 ap.args.opcode = FUSE_IOCTL;
2855 ap.args.nodeid = ff->nodeid;
2856 ap.args.in_numargs = 1;
2857 ap.args.in_args[0].size = sizeof(inarg);
2858 ap.args.in_args[0].value = &inarg;
2859 if (in_size) {
2860 ap.args.in_numargs++;
2861 ap.args.in_args[1].size = in_size;
2862 ap.args.in_pages = true;
2863
2864 err = -EFAULT;
2865 iov_iter_init(&ii, WRITE, in_iov, in_iovs, in_size);
2866 for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= ap.num_pages); i++) {
2867 c = copy_page_from_iter(ap.pages[i], 0, PAGE_SIZE, &ii);
2868 if (c != PAGE_SIZE && iov_iter_count(&ii))
2869 goto out;
2870 }
2871 }
2872
2873 ap.args.out_numargs = 2;
2874 ap.args.out_args[0].size = sizeof(outarg);
2875 ap.args.out_args[0].value = &outarg;
2876 ap.args.out_args[1].size = out_size;
2877 ap.args.out_pages = true;
2878 ap.args.out_argvar = true;
2879
2880 transferred = fuse_simple_request(fm, &ap.args);
2881 err = transferred;
2882 if (transferred < 0)
2883 goto out;
2884
2885 /* did it ask for retry? */
2886 if (outarg.flags & FUSE_IOCTL_RETRY) {
2887 void *vaddr;
2888
2889 /* no retry if in restricted mode */
2890 err = -EIO;
2891 if (!(flags & FUSE_IOCTL_UNRESTRICTED))
2892 goto out;
2893
2894 in_iovs = outarg.in_iovs;
2895 out_iovs = outarg.out_iovs;
2896
2897 /*
2898 * Make sure things are in boundary, separate checks
2899 * are to protect against overflow.
2900 */
2901 err = -ENOMEM;
2902 if (in_iovs > FUSE_IOCTL_MAX_IOV ||
2903 out_iovs > FUSE_IOCTL_MAX_IOV ||
2904 in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
2905 goto out;
2906
2907 vaddr = kmap_atomic(ap.pages[0]);
2908 err = fuse_copy_ioctl_iovec(fm->fc, iov_page, vaddr,
2909 transferred, in_iovs + out_iovs,
2910 (flags & FUSE_IOCTL_COMPAT) != 0);
2911 kunmap_atomic(vaddr);
2912 if (err)
2913 goto out;
2914
2915 in_iov = iov_page;
2916 out_iov = in_iov + in_iovs;
2917
2918 err = fuse_verify_ioctl_iov(fm->fc, in_iov, in_iovs);
2919 if (err)
2920 goto out;
2921
2922 err = fuse_verify_ioctl_iov(fm->fc, out_iov, out_iovs);
2923 if (err)
2924 goto out;
2925
2926 goto retry;
2927 }
2928
2929 err = -EIO;
2930 if (transferred > inarg.out_size)
2931 goto out;
2932
2933 err = -EFAULT;
2934 iov_iter_init(&ii, READ, out_iov, out_iovs, transferred);
2935 for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= ap.num_pages); i++) {
2936 c = copy_page_to_iter(ap.pages[i], 0, PAGE_SIZE, &ii);
2937 if (c != PAGE_SIZE && iov_iter_count(&ii))
2938 goto out;
2939 }
2940 err = 0;
2941 out:
2942 free_page((unsigned long) iov_page);
2943 while (ap.num_pages)
2944 __free_page(ap.pages[--ap.num_pages]);
2945 kfree(ap.pages);
2946
2947 return err ? err : outarg.result;
2948 }
2949 EXPORT_SYMBOL_GPL(fuse_do_ioctl);
2950
2951 long fuse_ioctl_common(struct file *file, unsigned int cmd,
2952 unsigned long arg, unsigned int flags)
2953 {
2954 struct inode *inode = file_inode(file);
2955 struct fuse_conn *fc = get_fuse_conn(inode);
2956
2957 if (!fuse_allow_current_process(fc))
2958 return -EACCES;
2959
2960 if (fuse_is_bad(inode))
2961 return -EIO;
2962
2963 return fuse_do_ioctl(file, cmd, arg, flags);
2964 }
2965
2966 static long fuse_file_ioctl(struct file *file, unsigned int cmd,
2967 unsigned long arg)
2968 {
2969 return fuse_ioctl_common(file, cmd, arg, 0);
2970 }
2971
2972 static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
2973 unsigned long arg)
2974 {
2975 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
2976 }
2977
2978 /*
2979 * All files which have been polled are linked to RB tree
2980 * fuse_conn->polled_files which is indexed by kh. Walk the tree and
2981 * find the matching one.
2982 */
2983 static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh,
2984 struct rb_node **parent_out)
2985 {
2986 struct rb_node **link = &fc->polled_files.rb_node;
2987 struct rb_node *last = NULL;
2988
2989 while (*link) {
2990 struct fuse_file *ff;
2991
2992 last = *link;
2993 ff = rb_entry(last, struct fuse_file, polled_node);
2994
2995 if (kh < ff->kh)
2996 link = &last->rb_left;
2997 else if (kh > ff->kh)
2998 link = &last->rb_right;
2999 else
3000 return link;
3001 }
3002
3003 if (parent_out)
3004 *parent_out = last;
3005 return link;
3006 }
3007
3008 /*
3009 * The file is about to be polled. Make sure it's on the polled_files
3010 * RB tree. Note that files once added to the polled_files tree are
3011 * not removed before the file is released. This is because a file
3012 * polled once is likely to be polled again.
3013 */
3014 static void fuse_register_polled_file(struct fuse_conn *fc,
3015 struct fuse_file *ff)
3016 {
3017 spin_lock(&fc->lock);
3018 if (RB_EMPTY_NODE(&ff->polled_node)) {
3019 struct rb_node **link, *parent;
3020
3021 link = fuse_find_polled_node(fc, ff->kh, &parent);
3022 BUG_ON(*link);
3023 rb_link_node(&ff->polled_node, parent, link);
3024 rb_insert_color(&ff->polled_node, &fc->polled_files);
3025 }
3026 spin_unlock(&fc->lock);
3027 }
3028
3029 __poll_t fuse_file_poll(struct file *file, poll_table *wait)
3030 {
3031 struct fuse_file *ff = file->private_data;
3032 struct fuse_mount *fm = ff->fm;
3033 struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh };
3034 struct fuse_poll_out outarg;
3035 FUSE_ARGS(args);
3036 int err;
3037
3038 if (fm->fc->no_poll)
3039 return DEFAULT_POLLMASK;
3040
3041 poll_wait(file, &ff->poll_wait, wait);
3042 inarg.events = mangle_poll(poll_requested_events(wait));
3043
3044 /*
3045 * Ask for notification iff there's someone waiting for it.
3046 * The client may ignore the flag and always notify.
3047 */
3048 if (waitqueue_active(&ff->poll_wait)) {
3049 inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY;
3050 fuse_register_polled_file(fm->fc, ff);
3051 }
3052
3053 args.opcode = FUSE_POLL;
3054 args.nodeid = ff->nodeid;
3055 args.in_numargs = 1;
3056 args.in_args[0].size = sizeof(inarg);
3057 args.in_args[0].value = &inarg;
3058 args.out_numargs = 1;
3059 args.out_args[0].size = sizeof(outarg);
3060 args.out_args[0].value = &outarg;
3061 err = fuse_simple_request(fm, &args);
3062
3063 if (!err)
3064 return demangle_poll(outarg.revents);
3065 if (err == -ENOSYS) {
3066 fm->fc->no_poll = 1;
3067 return DEFAULT_POLLMASK;
3068 }
3069 return EPOLLERR;
3070 }
3071 EXPORT_SYMBOL_GPL(fuse_file_poll);
3072
3073 /*
3074 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
3075 * wakes up the poll waiters.
3076 */
3077 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
3078 struct fuse_notify_poll_wakeup_out *outarg)
3079 {
3080 u64 kh = outarg->kh;
3081 struct rb_node **link;
3082
3083 spin_lock(&fc->lock);
3084
3085 link = fuse_find_polled_node(fc, kh, NULL);
3086 if (*link) {
3087 struct fuse_file *ff;
3088
3089 ff = rb_entry(*link, struct fuse_file, polled_node);
3090 wake_up_interruptible_sync(&ff->poll_wait);
3091 }
3092
3093 spin_unlock(&fc->lock);
3094 return 0;
3095 }
3096
3097 static void fuse_do_truncate(struct file *file)
3098 {
3099 struct inode *inode = file->f_mapping->host;
3100 struct iattr attr;
3101
3102 attr.ia_valid = ATTR_SIZE;
3103 attr.ia_size = i_size_read(inode);
3104
3105 attr.ia_file = file;
3106 attr.ia_valid |= ATTR_FILE;
3107
3108 fuse_do_setattr(file_dentry(file), &attr, file);
3109 }
3110
3111 static inline loff_t fuse_round_up(struct fuse_conn *fc, loff_t off)
3112 {
3113 return round_up(off, fc->max_pages << PAGE_SHIFT);
3114 }
3115
3116 static ssize_t
3117 fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3118 {
3119 DECLARE_COMPLETION_ONSTACK(wait);
3120 ssize_t ret = 0;
3121 struct file *file = iocb->ki_filp;
3122 struct fuse_file *ff = file->private_data;
3123 loff_t pos = 0;
3124 struct inode *inode;
3125 loff_t i_size;
3126 size_t count = iov_iter_count(iter), shortened = 0;
3127 loff_t offset = iocb->ki_pos;
3128 struct fuse_io_priv *io;
3129
3130 pos = offset;
3131 inode = file->f_mapping->host;
3132 i_size = i_size_read(inode);
3133
3134 if ((iov_iter_rw(iter) == READ) && (offset >= i_size))
3135 return 0;
3136
3137 io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL);
3138 if (!io)
3139 return -ENOMEM;
3140 spin_lock_init(&io->lock);
3141 kref_init(&io->refcnt);
3142 io->reqs = 1;
3143 io->bytes = -1;
3144 io->size = 0;
3145 io->offset = offset;
3146 io->write = (iov_iter_rw(iter) == WRITE);
3147 io->err = 0;
3148 /*
3149 * By default, we want to optimize all I/Os with async request
3150 * submission to the client filesystem if supported.
3151 */
3152 io->async = ff->fm->fc->async_dio;
3153 io->iocb = iocb;
3154 io->blocking = is_sync_kiocb(iocb);
3155
3156 /* optimization for short read */
3157 if (io->async && !io->write && offset + count > i_size) {
3158 iov_iter_truncate(iter, fuse_round_up(ff->fm->fc, i_size - offset));
3159 shortened = count - iov_iter_count(iter);
3160 count -= shortened;
3161 }
3162
3163 /*
3164 * We cannot asynchronously extend the size of a file.
3165 * In such case the aio will behave exactly like sync io.
3166 */
3167 if ((offset + count > i_size) && io->write)
3168 io->blocking = true;
3169
3170 if (io->async && io->blocking) {
3171 /*
3172 * Additional reference to keep io around after
3173 * calling fuse_aio_complete()
3174 */
3175 kref_get(&io->refcnt);
3176 io->done = &wait;
3177 }
3178
3179 if (iov_iter_rw(iter) == WRITE) {
3180 ret = fuse_direct_io(io, iter, &pos, FUSE_DIO_WRITE);
3181 fuse_invalidate_attr(inode);
3182 } else {
3183 ret = __fuse_direct_read(io, iter, &pos);
3184 }
3185 iov_iter_reexpand(iter, iov_iter_count(iter) + shortened);
3186
3187 if (io->async) {
3188 bool blocking = io->blocking;
3189
3190 fuse_aio_complete(io, ret < 0 ? ret : 0, -1);
3191
3192 /* we have a non-extending, async request, so return */
3193 if (!blocking)
3194 return -EIOCBQUEUED;
3195
3196 wait_for_completion(&wait);
3197 ret = fuse_get_res_by_io(io);
3198 }
3199
3200 kref_put(&io->refcnt, fuse_io_release);
3201
3202 if (iov_iter_rw(iter) == WRITE) {
3203 if (ret > 0)
3204 fuse_write_update_size(inode, pos);
3205 else if (ret < 0 && offset + count > i_size)
3206 fuse_do_truncate(file);
3207 }
3208
3209 return ret;
3210 }
3211
3212 static int fuse_writeback_range(struct inode *inode, loff_t start, loff_t end)
3213 {
3214 int err = filemap_write_and_wait_range(inode->i_mapping, start, end);
3215
3216 if (!err)
3217 fuse_sync_writes(inode);
3218
3219 return err;
3220 }
3221
3222 static long fuse_file_fallocate(struct file *file, int mode, loff_t offset,
3223 loff_t length)
3224 {
3225 struct fuse_file *ff = file->private_data;
3226 struct inode *inode = file_inode(file);
3227 struct fuse_inode *fi = get_fuse_inode(inode);
3228 struct fuse_mount *fm = ff->fm;
3229 FUSE_ARGS(args);
3230 struct fuse_fallocate_in inarg = {
3231 .fh = ff->fh,
3232 .offset = offset,
3233 .length = length,
3234 .mode = mode
3235 };
3236 int err;
3237 bool lock_inode = !(mode & FALLOC_FL_KEEP_SIZE) ||
3238 (mode & FALLOC_FL_PUNCH_HOLE);
3239
3240 bool block_faults = FUSE_IS_DAX(inode) && lock_inode;
3241
3242 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3243 return -EOPNOTSUPP;
3244
3245 if (fm->fc->no_fallocate)
3246 return -EOPNOTSUPP;
3247
3248 if (lock_inode) {
3249 inode_lock(inode);
3250 if (block_faults) {
3251 down_write(&fi->i_mmap_sem);
3252 err = fuse_dax_break_layouts(inode, 0, 0);
3253 if (err)
3254 goto out;
3255 }
3256
3257 if (mode & FALLOC_FL_PUNCH_HOLE) {
3258 loff_t endbyte = offset + length - 1;
3259
3260 err = fuse_writeback_range(inode, offset, endbyte);
3261 if (err)
3262 goto out;
3263 }
3264 }
3265
3266 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
3267 offset + length > i_size_read(inode)) {
3268 err = inode_newsize_ok(inode, offset + length);
3269 if (err)
3270 goto out;
3271 }
3272
3273 if (!(mode & FALLOC_FL_KEEP_SIZE))
3274 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
3275
3276 args.opcode = FUSE_FALLOCATE;
3277 args.nodeid = ff->nodeid;
3278 args.in_numargs = 1;
3279 args.in_args[0].size = sizeof(inarg);
3280 args.in_args[0].value = &inarg;
3281 err = fuse_simple_request(fm, &args);
3282 if (err == -ENOSYS) {
3283 fm->fc->no_fallocate = 1;
3284 err = -EOPNOTSUPP;
3285 }
3286 if (err)
3287 goto out;
3288
3289 /* we could have extended the file */
3290 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3291 bool changed = fuse_write_update_size(inode, offset + length);
3292
3293 if (changed && fm->fc->writeback_cache)
3294 file_update_time(file);
3295 }
3296
3297 if (mode & FALLOC_FL_PUNCH_HOLE)
3298 truncate_pagecache_range(inode, offset, offset + length - 1);
3299
3300 fuse_invalidate_attr(inode);
3301
3302 out:
3303 if (!(mode & FALLOC_FL_KEEP_SIZE))
3304 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
3305
3306 if (block_faults)
3307 up_write(&fi->i_mmap_sem);
3308
3309 if (lock_inode)
3310 inode_unlock(inode);
3311
3312 return err;
3313 }
3314
3315 static ssize_t __fuse_copy_file_range(struct file *file_in, loff_t pos_in,
3316 struct file *file_out, loff_t pos_out,
3317 size_t len, unsigned int flags)
3318 {
3319 struct fuse_file *ff_in = file_in->private_data;
3320 struct fuse_file *ff_out = file_out->private_data;
3321 struct inode *inode_in = file_inode(file_in);
3322 struct inode *inode_out = file_inode(file_out);
3323 struct fuse_inode *fi_out = get_fuse_inode(inode_out);
3324 struct fuse_mount *fm = ff_in->fm;
3325 struct fuse_conn *fc = fm->fc;
3326 FUSE_ARGS(args);
3327 struct fuse_copy_file_range_in inarg = {
3328 .fh_in = ff_in->fh,
3329 .off_in = pos_in,
3330 .nodeid_out = ff_out->nodeid,
3331 .fh_out = ff_out->fh,
3332 .off_out = pos_out,
3333 .len = len,
3334 .flags = flags
3335 };
3336 struct fuse_write_out outarg;
3337 ssize_t err;
3338 /* mark unstable when write-back is not used, and file_out gets
3339 * extended */
3340 bool is_unstable = (!fc->writeback_cache) &&
3341 ((pos_out + len) > inode_out->i_size);
3342
3343 if (fc->no_copy_file_range)
3344 return -EOPNOTSUPP;
3345
3346 if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
3347 return -EXDEV;
3348
3349 inode_lock(inode_in);
3350 err = fuse_writeback_range(inode_in, pos_in, pos_in + len - 1);
3351 inode_unlock(inode_in);
3352 if (err)
3353 return err;
3354
3355 inode_lock(inode_out);
3356
3357 err = file_modified(file_out);
3358 if (err)
3359 goto out;
3360
3361 /*
3362 * Write out dirty pages in the destination file before sending the COPY
3363 * request to userspace. After the request is completed, truncate off
3364 * pages (including partial ones) from the cache that have been copied,
3365 * since these contain stale data at that point.
3366 *
3367 * This should be mostly correct, but if the COPY writes to partial
3368 * pages (at the start or end) and the parts not covered by the COPY are
3369 * written through a memory map after calling fuse_writeback_range(),
3370 * then these partial page modifications will be lost on truncation.
3371 *
3372 * It is unlikely that someone would rely on such mixed style
3373 * modifications. Yet this does give less guarantees than if the
3374 * copying was performed with write(2).
3375 *
3376 * To fix this a i_mmap_sem style lock could be used to prevent new
3377 * faults while the copy is ongoing.
3378 */
3379 err = fuse_writeback_range(inode_out, pos_out, pos_out + len - 1);
3380 if (err)
3381 goto out;
3382
3383 if (is_unstable)
3384 set_bit(FUSE_I_SIZE_UNSTABLE, &fi_out->state);
3385
3386 args.opcode = FUSE_COPY_FILE_RANGE;
3387 args.nodeid = ff_in->nodeid;
3388 args.in_numargs = 1;
3389 args.in_args[0].size = sizeof(inarg);
3390 args.in_args[0].value = &inarg;
3391 args.out_numargs = 1;
3392 args.out_args[0].size = sizeof(outarg);
3393 args.out_args[0].value = &outarg;
3394 err = fuse_simple_request(fm, &args);
3395 if (err == -ENOSYS) {
3396 fc->no_copy_file_range = 1;
3397 err = -EOPNOTSUPP;
3398 }
3399 if (err)
3400 goto out;
3401
3402 truncate_inode_pages_range(inode_out->i_mapping,
3403 ALIGN_DOWN(pos_out, PAGE_SIZE),
3404 ALIGN(pos_out + outarg.size, PAGE_SIZE) - 1);
3405
3406 if (fc->writeback_cache) {
3407 fuse_write_update_size(inode_out, pos_out + outarg.size);
3408 file_update_time(file_out);
3409 }
3410
3411 fuse_invalidate_attr(inode_out);
3412
3413 err = outarg.size;
3414 out:
3415 if (is_unstable)
3416 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi_out->state);
3417
3418 inode_unlock(inode_out);
3419 file_accessed(file_in);
3420
3421 return err;
3422 }
3423
3424 static ssize_t fuse_copy_file_range(struct file *src_file, loff_t src_off,
3425 struct file *dst_file, loff_t dst_off,
3426 size_t len, unsigned int flags)
3427 {
3428 ssize_t ret;
3429
3430 ret = __fuse_copy_file_range(src_file, src_off, dst_file, dst_off,
3431 len, flags);
3432
3433 if (ret == -EOPNOTSUPP || ret == -EXDEV)
3434 ret = generic_copy_file_range(src_file, src_off, dst_file,
3435 dst_off, len, flags);
3436 return ret;
3437 }
3438
3439 static const struct file_operations fuse_file_operations = {
3440 .llseek = fuse_file_llseek,
3441 .read_iter = fuse_file_read_iter,
3442 .write_iter = fuse_file_write_iter,
3443 .mmap = fuse_file_mmap,
3444 .open = fuse_open,
3445 .flush = fuse_flush,
3446 .release = fuse_release,
3447 .fsync = fuse_fsync,
3448 .lock = fuse_file_lock,
3449 .get_unmapped_area = thp_get_unmapped_area,
3450 .flock = fuse_file_flock,
3451 .splice_read = generic_file_splice_read,
3452 .splice_write = iter_file_splice_write,
3453 .unlocked_ioctl = fuse_file_ioctl,
3454 .compat_ioctl = fuse_file_compat_ioctl,
3455 .poll = fuse_file_poll,
3456 .fallocate = fuse_file_fallocate,
3457 .copy_file_range = fuse_copy_file_range,
3458 };
3459
3460 static const struct address_space_operations fuse_file_aops = {
3461 .readpage = fuse_readpage,
3462 .readahead = fuse_readahead,
3463 .writepage = fuse_writepage,
3464 .writepages = fuse_writepages,
3465 .launder_page = fuse_launder_page,
3466 .set_page_dirty = __set_page_dirty_nobuffers,
3467 .bmap = fuse_bmap,
3468 .direct_IO = fuse_direct_IO,
3469 .write_begin = fuse_write_begin,
3470 .write_end = fuse_write_end,
3471 };
3472
3473 void fuse_init_file_inode(struct inode *inode)
3474 {
3475 struct fuse_inode *fi = get_fuse_inode(inode);
3476
3477 inode->i_fop = &fuse_file_operations;
3478 inode->i_data.a_ops = &fuse_file_aops;
3479
3480 INIT_LIST_HEAD(&fi->write_files);
3481 INIT_LIST_HEAD(&fi->queued_writes);
3482 fi->writectr = 0;
3483 init_waitqueue_head(&fi->page_waitq);
3484 fi->writepages = RB_ROOT;
3485
3486 if (IS_ENABLED(CONFIG_FUSE_DAX))
3487 fuse_dax_inode_init(inode);
3488 }
Linux with added FPC-III support