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/module.h>
16 #include <linux/compat.h>
17 #include <linux/swap.h>
18 #include <linux/falloc.h>
19 #include <linux/uio.h>
20
21 static const struct file_operations fuse_direct_io_file_operations;
22
fuse_send_open(struct fuse_conn * fc,u64 nodeid,struct file * file,int opcode,struct fuse_open_out * outargp)23 static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
24 int opcode, struct fuse_open_out *outargp)
25 {
26 struct fuse_open_in inarg;
27 FUSE_ARGS(args);
28
29 memset(&inarg, 0, sizeof(inarg));
30 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY);
31 if (!fc->atomic_o_trunc)
32 inarg.flags &= ~O_TRUNC;
33 args.in.h.opcode = opcode;
34 args.in.h.nodeid = nodeid;
35 args.in.numargs = 1;
36 args.in.args[0].size = sizeof(inarg);
37 args.in.args[0].value = &inarg;
38 args.out.numargs = 1;
39 args.out.args[0].size = sizeof(*outargp);
40 args.out.args[0].value = outargp;
41
42 return fuse_simple_request(fc, &args);
43 }
44
fuse_file_alloc(struct fuse_conn * fc)45 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc)
46 {
47 struct fuse_file *ff;
48
49 ff = kzalloc(sizeof(struct fuse_file), GFP_KERNEL);
50 if (unlikely(!ff))
51 return NULL;
52
53 ff->fc = fc;
54 ff->reserved_req = fuse_request_alloc(0);
55 if (unlikely(!ff->reserved_req)) {
56 kfree(ff);
57 return NULL;
58 }
59
60 INIT_LIST_HEAD(&ff->write_entry);
61 refcount_set(&ff->count, 1);
62 RB_CLEAR_NODE(&ff->polled_node);
63 init_waitqueue_head(&ff->poll_wait);
64
65 spin_lock(&fc->lock);
66 ff->kh = ++fc->khctr;
67 spin_unlock(&fc->lock);
68
69 return ff;
70 }
71
fuse_file_free(struct fuse_file * ff)72 void fuse_file_free(struct fuse_file *ff)
73 {
74 fuse_request_free(ff->reserved_req);
75 kfree(ff);
76 }
77
fuse_file_get(struct fuse_file * ff)78 static struct fuse_file *fuse_file_get(struct fuse_file *ff)
79 {
80 refcount_inc(&ff->count);
81 return ff;
82 }
83
fuse_release_end(struct fuse_conn * fc,struct fuse_req * req)84 static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
85 {
86 iput(req->misc.release.inode);
87 }
88
fuse_file_put(struct fuse_file * ff,bool sync,bool isdir)89 static void fuse_file_put(struct fuse_file *ff, bool sync, bool isdir)
90 {
91 if (refcount_dec_and_test(&ff->count)) {
92 struct fuse_req *req = ff->reserved_req;
93
94 if (ff->fc->no_open && !isdir) {
95 /*
96 * Drop the release request when client does not
97 * implement 'open'
98 */
99 __clear_bit(FR_BACKGROUND, &req->flags);
100 iput(req->misc.release.inode);
101 fuse_put_request(ff->fc, req);
102 } else if (sync) {
103 __set_bit(FR_FORCE, &req->flags);
104 __clear_bit(FR_BACKGROUND, &req->flags);
105 fuse_request_send(ff->fc, req);
106 iput(req->misc.release.inode);
107 fuse_put_request(ff->fc, req);
108 } else {
109 req->end = fuse_release_end;
110 __set_bit(FR_BACKGROUND, &req->flags);
111 fuse_request_send_background(ff->fc, req);
112 }
113 kfree(ff);
114 }
115 }
116
fuse_do_open(struct fuse_conn * fc,u64 nodeid,struct file * file,bool isdir)117 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
118 bool isdir)
119 {
120 struct fuse_file *ff;
121 int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
122
123 ff = fuse_file_alloc(fc);
124 if (!ff)
125 return -ENOMEM;
126
127 ff->fh = 0;
128 ff->open_flags = FOPEN_KEEP_CACHE; /* Default for no-open */
129 if (!fc->no_open || isdir) {
130 struct fuse_open_out outarg;
131 int err;
132
133 err = fuse_send_open(fc, nodeid, file, opcode, &outarg);
134 if (!err) {
135 ff->fh = outarg.fh;
136 ff->open_flags = outarg.open_flags;
137
138 } else if (err != -ENOSYS || isdir) {
139 fuse_file_free(ff);
140 return err;
141 } else {
142 fc->no_open = 1;
143 }
144 }
145
146 if (isdir)
147 ff->open_flags &= ~FOPEN_DIRECT_IO;
148
149 ff->nodeid = nodeid;
150 file->private_data = ff;
151
152 return 0;
153 }
154 EXPORT_SYMBOL_GPL(fuse_do_open);
155
fuse_link_write_file(struct file * file)156 static void fuse_link_write_file(struct file *file)
157 {
158 struct inode *inode = file_inode(file);
159 struct fuse_conn *fc = get_fuse_conn(inode);
160 struct fuse_inode *fi = get_fuse_inode(inode);
161 struct fuse_file *ff = file->private_data;
162 /*
163 * file may be written through mmap, so chain it onto the
164 * inodes's write_file list
165 */
166 spin_lock(&fc->lock);
167 if (list_empty(&ff->write_entry))
168 list_add(&ff->write_entry, &fi->write_files);
169 spin_unlock(&fc->lock);
170 }
171
fuse_finish_open(struct inode * inode,struct file * file)172 void fuse_finish_open(struct inode *inode, struct file *file)
173 {
174 struct fuse_file *ff = file->private_data;
175 struct fuse_conn *fc = get_fuse_conn(inode);
176
177 if (ff->open_flags & FOPEN_DIRECT_IO)
178 file->f_op = &fuse_direct_io_file_operations;
179 if (!(ff->open_flags & FOPEN_KEEP_CACHE))
180 invalidate_inode_pages2(inode->i_mapping);
181 if (ff->open_flags & FOPEN_STREAM)
182 stream_open(inode, file);
183 else if (ff->open_flags & FOPEN_NONSEEKABLE)
184 nonseekable_open(inode, file);
185 if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) {
186 struct fuse_inode *fi = get_fuse_inode(inode);
187
188 spin_lock(&fc->lock);
189 fi->attr_version = ++fc->attr_version;
190 i_size_write(inode, 0);
191 spin_unlock(&fc->lock);
192 fuse_invalidate_attr(inode);
193 if (fc->writeback_cache)
194 file_update_time(file);
195 }
196 if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache)
197 fuse_link_write_file(file);
198 }
199
fuse_open_common(struct inode * inode,struct file * file,bool isdir)200 int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
201 {
202 struct fuse_conn *fc = get_fuse_conn(inode);
203 int err;
204 bool is_wb_truncate = (file->f_flags & O_TRUNC) &&
205 fc->atomic_o_trunc &&
206 fc->writeback_cache;
207
208 err = generic_file_open(inode, file);
209 if (err)
210 return err;
211
212 if (is_wb_truncate) {
213 inode_lock(inode);
214 fuse_set_nowrite(inode);
215 }
216
217 err = fuse_do_open(fc, get_node_id(inode), file, isdir);
218
219 if (!err)
220 fuse_finish_open(inode, file);
221
222 if (is_wb_truncate) {
223 fuse_release_nowrite(inode);
224 inode_unlock(inode);
225 }
226
227 return err;
228 }
229
fuse_prepare_release(struct fuse_file * ff,int flags,int opcode)230 static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode)
231 {
232 struct fuse_conn *fc = ff->fc;
233 struct fuse_req *req = ff->reserved_req;
234 struct fuse_release_in *inarg = &req->misc.release.in;
235
236 spin_lock(&fc->lock);
237 list_del(&ff->write_entry);
238 if (!RB_EMPTY_NODE(&ff->polled_node))
239 rb_erase(&ff->polled_node, &fc->polled_files);
240 spin_unlock(&fc->lock);
241
242 wake_up_interruptible_all(&ff->poll_wait);
243
244 inarg->fh = ff->fh;
245 inarg->flags = flags;
246 req->in.h.opcode = opcode;
247 req->in.h.nodeid = ff->nodeid;
248 req->in.numargs = 1;
249 req->in.args[0].size = sizeof(struct fuse_release_in);
250 req->in.args[0].value = inarg;
251 }
252
fuse_release_common(struct file * file,bool isdir)253 void fuse_release_common(struct file *file, bool isdir)
254 {
255 struct fuse_file *ff = file->private_data;
256 struct fuse_req *req = ff->reserved_req;
257 int opcode = isdir ? FUSE_RELEASEDIR : FUSE_RELEASE;
258
259 fuse_prepare_release(ff, file->f_flags, opcode);
260
261 if (ff->flock) {
262 struct fuse_release_in *inarg = &req->misc.release.in;
263 inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK;
264 inarg->lock_owner = fuse_lock_owner_id(ff->fc,
265 (fl_owner_t) file);
266 }
267 /* Hold inode until release is finished */
268 req->misc.release.inode = igrab(file_inode(file));
269
270 /*
271 * Normally this will send the RELEASE request, however if
272 * some asynchronous READ or WRITE requests are outstanding,
273 * the sending will be delayed.
274 *
275 * Make the release synchronous if this is a fuseblk mount,
276 * synchronous RELEASE is allowed (and desirable) in this case
277 * because the server can be trusted not to screw up.
278 */
279 fuse_file_put(ff, ff->fc->destroy_req != NULL, isdir);
280 }
281
fuse_open(struct inode * inode,struct file * file)282 static int fuse_open(struct inode *inode, struct file *file)
283 {
284 return fuse_open_common(inode, file, false);
285 }
286
fuse_release(struct inode * inode,struct file * file)287 static int fuse_release(struct inode *inode, struct file *file)
288 {
289 struct fuse_conn *fc = get_fuse_conn(inode);
290
291 /* see fuse_vma_close() for !writeback_cache case */
292 if (fc->writeback_cache)
293 write_inode_now(inode, 1);
294
295 fuse_release_common(file, false);
296
297 /* return value is ignored by VFS */
298 return 0;
299 }
300
fuse_sync_release(struct fuse_file * ff,int flags)301 void fuse_sync_release(struct fuse_file *ff, int flags)
302 {
303 WARN_ON(refcount_read(&ff->count) > 1);
304 fuse_prepare_release(ff, flags, FUSE_RELEASE);
305 /*
306 * iput(NULL) is a no-op and since the refcount is 1 and everything's
307 * synchronous, we are fine with not doing igrab() here"
308 */
309 fuse_file_put(ff, true, false);
310 }
311 EXPORT_SYMBOL_GPL(fuse_sync_release);
312
313 /*
314 * Scramble the ID space with XTEA, so that the value of the files_struct
315 * pointer is not exposed to userspace.
316 */
fuse_lock_owner_id(struct fuse_conn * fc,fl_owner_t id)317 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
318 {
319 u32 *k = fc->scramble_key;
320 u64 v = (unsigned long) id;
321 u32 v0 = v;
322 u32 v1 = v >> 32;
323 u32 sum = 0;
324 int i;
325
326 for (i = 0; i < 32; i++) {
327 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
328 sum += 0x9E3779B9;
329 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
330 }
331
332 return (u64) v0 + ((u64) v1 << 32);
333 }
334
335 /*
336 * Check if any page in a range is under writeback
337 *
338 * This is currently done by walking the list of writepage requests
339 * for the inode, which can be pretty inefficient.
340 */
fuse_range_is_writeback(struct inode * inode,pgoff_t idx_from,pgoff_t idx_to)341 static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from,
342 pgoff_t idx_to)
343 {
344 struct fuse_conn *fc = get_fuse_conn(inode);
345 struct fuse_inode *fi = get_fuse_inode(inode);
346 struct fuse_req *req;
347 bool found = false;
348
349 spin_lock(&fc->lock);
350 list_for_each_entry(req, &fi->writepages, writepages_entry) {
351 pgoff_t curr_index;
352
353 BUG_ON(req->inode != inode);
354 curr_index = req->misc.write.in.offset >> PAGE_SHIFT;
355 if (idx_from < curr_index + req->num_pages &&
356 curr_index <= idx_to) {
357 found = true;
358 break;
359 }
360 }
361 spin_unlock(&fc->lock);
362
363 return found;
364 }
365
fuse_page_is_writeback(struct inode * inode,pgoff_t index)366 static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
367 {
368 return fuse_range_is_writeback(inode, index, index);
369 }
370
371 /*
372 * Wait for page writeback to be completed.
373 *
374 * Since fuse doesn't rely on the VM writeback tracking, this has to
375 * use some other means.
376 */
fuse_wait_on_page_writeback(struct inode * inode,pgoff_t index)377 static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
378 {
379 struct fuse_inode *fi = get_fuse_inode(inode);
380
381 wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
382 return 0;
383 }
384
385 /*
386 * Wait for all pending writepages on the inode to finish.
387 *
388 * This is currently done by blocking further writes with FUSE_NOWRITE
389 * and waiting for all sent writes to complete.
390 *
391 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
392 * could conflict with truncation.
393 */
fuse_sync_writes(struct inode * inode)394 static void fuse_sync_writes(struct inode *inode)
395 {
396 fuse_set_nowrite(inode);
397 fuse_release_nowrite(inode);
398 }
399
fuse_flush(struct file * file,fl_owner_t id)400 static int fuse_flush(struct file *file, fl_owner_t id)
401 {
402 struct inode *inode = file_inode(file);
403 struct fuse_conn *fc = get_fuse_conn(inode);
404 struct fuse_file *ff = file->private_data;
405 struct fuse_req *req;
406 struct fuse_flush_in inarg;
407 int err;
408
409 if (is_bad_inode(inode))
410 return -EIO;
411
412 if (fc->no_flush)
413 return 0;
414
415 err = write_inode_now(inode, 1);
416 if (err)
417 return err;
418
419 inode_lock(inode);
420 fuse_sync_writes(inode);
421 inode_unlock(inode);
422
423 err = filemap_check_errors(file->f_mapping);
424 if (err)
425 return err;
426
427 req = fuse_get_req_nofail_nopages(fc, file);
428 memset(&inarg, 0, sizeof(inarg));
429 inarg.fh = ff->fh;
430 inarg.lock_owner = fuse_lock_owner_id(fc, id);
431 req->in.h.opcode = FUSE_FLUSH;
432 req->in.h.nodeid = get_node_id(inode);
433 req->in.numargs = 1;
434 req->in.args[0].size = sizeof(inarg);
435 req->in.args[0].value = &inarg;
436 __set_bit(FR_FORCE, &req->flags);
437 fuse_request_send(fc, req);
438 err = req->out.h.error;
439 fuse_put_request(fc, req);
440 if (err == -ENOSYS) {
441 fc->no_flush = 1;
442 err = 0;
443 }
444 return err;
445 }
446
fuse_fsync_common(struct file * file,loff_t start,loff_t end,int datasync,int isdir)447 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
448 int datasync, int isdir)
449 {
450 struct inode *inode = file->f_mapping->host;
451 struct fuse_conn *fc = get_fuse_conn(inode);
452 struct fuse_file *ff = file->private_data;
453 FUSE_ARGS(args);
454 struct fuse_fsync_in inarg;
455 int err;
456
457 if (is_bad_inode(inode))
458 return -EIO;
459
460 inode_lock(inode);
461
462 /*
463 * Start writeback against all dirty pages of the inode, then
464 * wait for all outstanding writes, before sending the FSYNC
465 * request.
466 */
467 err = file_write_and_wait_range(file, start, end);
468 if (err)
469 goto out;
470
471 fuse_sync_writes(inode);
472
473 /*
474 * Due to implementation of fuse writeback
475 * file_write_and_wait_range() does not catch errors.
476 * We have to do this directly after fuse_sync_writes()
477 */
478 err = file_check_and_advance_wb_err(file);
479 if (err)
480 goto out;
481
482 err = sync_inode_metadata(inode, 1);
483 if (err)
484 goto out;
485
486 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
487 goto out;
488
489 memset(&inarg, 0, sizeof(inarg));
490 inarg.fh = ff->fh;
491 inarg.fsync_flags = datasync ? 1 : 0;
492 args.in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
493 args.in.h.nodeid = get_node_id(inode);
494 args.in.numargs = 1;
495 args.in.args[0].size = sizeof(inarg);
496 args.in.args[0].value = &inarg;
497 err = fuse_simple_request(fc, &args);
498 if (err == -ENOSYS) {
499 if (isdir)
500 fc->no_fsyncdir = 1;
501 else
502 fc->no_fsync = 1;
503 err = 0;
504 }
505 out:
506 inode_unlock(inode);
507 return err;
508 }
509
fuse_fsync(struct file * file,loff_t start,loff_t end,int datasync)510 static int fuse_fsync(struct file *file, loff_t start, loff_t end,
511 int datasync)
512 {
513 return fuse_fsync_common(file, start, end, datasync, 0);
514 }
515
fuse_read_fill(struct fuse_req * req,struct file * file,loff_t pos,size_t count,int opcode)516 void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos,
517 size_t count, int opcode)
518 {
519 struct fuse_read_in *inarg = &req->misc.read.in;
520 struct fuse_file *ff = file->private_data;
521
522 inarg->fh = ff->fh;
523 inarg->offset = pos;
524 inarg->size = count;
525 inarg->flags = file->f_flags;
526 req->in.h.opcode = opcode;
527 req->in.h.nodeid = ff->nodeid;
528 req->in.numargs = 1;
529 req->in.args[0].size = sizeof(struct fuse_read_in);
530 req->in.args[0].value = inarg;
531 req->out.argvar = 1;
532 req->out.numargs = 1;
533 req->out.args[0].size = count;
534 }
535
fuse_release_user_pages(struct fuse_req * req,bool should_dirty)536 static void fuse_release_user_pages(struct fuse_req *req, bool should_dirty)
537 {
538 unsigned i;
539
540 for (i = 0; i < req->num_pages; i++) {
541 struct page *page = req->pages[i];
542 if (should_dirty)
543 set_page_dirty_lock(page);
544 put_page(page);
545 }
546 }
547
fuse_io_release(struct kref * kref)548 static void fuse_io_release(struct kref *kref)
549 {
550 kfree(container_of(kref, struct fuse_io_priv, refcnt));
551 }
552
fuse_get_res_by_io(struct fuse_io_priv * io)553 static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io)
554 {
555 if (io->err)
556 return io->err;
557
558 if (io->bytes >= 0 && io->write)
559 return -EIO;
560
561 return io->bytes < 0 ? io->size : io->bytes;
562 }
563
564 /**
565 * In case of short read, the caller sets 'pos' to the position of
566 * actual end of fuse request in IO request. Otherwise, if bytes_requested
567 * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1.
568 *
569 * An example:
570 * User requested DIO read of 64K. It was splitted into two 32K fuse requests,
571 * both submitted asynchronously. The first of them was ACKed by userspace as
572 * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The
573 * second request was ACKed as short, e.g. only 1K was read, resulting in
574 * pos == 33K.
575 *
576 * Thus, when all fuse requests are completed, the minimal non-negative 'pos'
577 * will be equal to the length of the longest contiguous fragment of
578 * transferred data starting from the beginning of IO request.
579 */
fuse_aio_complete(struct fuse_io_priv * io,int err,ssize_t pos)580 static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
581 {
582 int left;
583
584 spin_lock(&io->lock);
585 if (err)
586 io->err = io->err ? : err;
587 else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes))
588 io->bytes = pos;
589
590 left = --io->reqs;
591 if (!left && io->blocking)
592 complete(io->done);
593 spin_unlock(&io->lock);
594
595 if (!left && !io->blocking) {
596 ssize_t res = fuse_get_res_by_io(io);
597
598 if (res >= 0) {
599 struct inode *inode = file_inode(io->iocb->ki_filp);
600 struct fuse_conn *fc = get_fuse_conn(inode);
601 struct fuse_inode *fi = get_fuse_inode(inode);
602
603 spin_lock(&fc->lock);
604 fi->attr_version = ++fc->attr_version;
605 spin_unlock(&fc->lock);
606 }
607
608 io->iocb->ki_complete(io->iocb, res, 0);
609 }
610
611 kref_put(&io->refcnt, fuse_io_release);
612 }
613
fuse_aio_complete_req(struct fuse_conn * fc,struct fuse_req * req)614 static void fuse_aio_complete_req(struct fuse_conn *fc, struct fuse_req *req)
615 {
616 struct fuse_io_priv *io = req->io;
617 ssize_t pos = -1;
618
619 fuse_release_user_pages(req, io->should_dirty);
620
621 if (io->write) {
622 if (req->misc.write.in.size != req->misc.write.out.size)
623 pos = req->misc.write.in.offset - io->offset +
624 req->misc.write.out.size;
625 } else {
626 if (req->misc.read.in.size != req->out.args[0].size)
627 pos = req->misc.read.in.offset - io->offset +
628 req->out.args[0].size;
629 }
630
631 fuse_aio_complete(io, req->out.h.error, pos);
632 }
633
fuse_async_req_send(struct fuse_conn * fc,struct fuse_req * req,size_t num_bytes,struct fuse_io_priv * io)634 static size_t fuse_async_req_send(struct fuse_conn *fc, struct fuse_req *req,
635 size_t num_bytes, struct fuse_io_priv *io)
636 {
637 spin_lock(&io->lock);
638 kref_get(&io->refcnt);
639 io->size += num_bytes;
640 io->reqs++;
641 spin_unlock(&io->lock);
642
643 req->io = io;
644 req->end = fuse_aio_complete_req;
645
646 __fuse_get_request(req);
647 fuse_request_send_background(fc, req);
648
649 return num_bytes;
650 }
651
fuse_send_read(struct fuse_req * req,struct fuse_io_priv * io,loff_t pos,size_t count,fl_owner_t owner)652 static size_t fuse_send_read(struct fuse_req *req, struct fuse_io_priv *io,
653 loff_t pos, size_t count, fl_owner_t owner)
654 {
655 struct file *file = io->iocb->ki_filp;
656 struct fuse_file *ff = file->private_data;
657 struct fuse_conn *fc = ff->fc;
658
659 fuse_read_fill(req, file, pos, count, FUSE_READ);
660 if (owner != NULL) {
661 struct fuse_read_in *inarg = &req->misc.read.in;
662
663 inarg->read_flags |= FUSE_READ_LOCKOWNER;
664 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
665 }
666
667 if (io->async)
668 return fuse_async_req_send(fc, req, count, io);
669
670 fuse_request_send(fc, req);
671 return req->out.args[0].size;
672 }
673
fuse_read_update_size(struct inode * inode,loff_t size,u64 attr_ver)674 static void fuse_read_update_size(struct inode *inode, loff_t size,
675 u64 attr_ver)
676 {
677 struct fuse_conn *fc = get_fuse_conn(inode);
678 struct fuse_inode *fi = get_fuse_inode(inode);
679
680 spin_lock(&fc->lock);
681 if (attr_ver == fi->attr_version && size < inode->i_size &&
682 !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
683 fi->attr_version = ++fc->attr_version;
684 i_size_write(inode, size);
685 }
686 spin_unlock(&fc->lock);
687 }
688
fuse_short_read(struct fuse_req * req,struct inode * inode,u64 attr_ver)689 static void fuse_short_read(struct fuse_req *req, struct inode *inode,
690 u64 attr_ver)
691 {
692 size_t num_read = req->out.args[0].size;
693 struct fuse_conn *fc = get_fuse_conn(inode);
694
695 if (fc->writeback_cache) {
696 /*
697 * A hole in a file. Some data after the hole are in page cache,
698 * but have not reached the client fs yet. So, the hole is not
699 * present there.
700 */
701 int i;
702 int start_idx = num_read >> PAGE_SHIFT;
703 size_t off = num_read & (PAGE_SIZE - 1);
704
705 for (i = start_idx; i < req->num_pages; i++) {
706 zero_user_segment(req->pages[i], off, PAGE_SIZE);
707 off = 0;
708 }
709 } else {
710 loff_t pos = page_offset(req->pages[0]) + num_read;
711 fuse_read_update_size(inode, pos, attr_ver);
712 }
713 }
714
fuse_do_readpage(struct file * file,struct page * page)715 static int fuse_do_readpage(struct file *file, struct page *page)
716 {
717 struct kiocb iocb;
718 struct fuse_io_priv io;
719 struct inode *inode = page->mapping->host;
720 struct fuse_conn *fc = get_fuse_conn(inode);
721 struct fuse_req *req;
722 size_t num_read;
723 loff_t pos = page_offset(page);
724 size_t count = PAGE_SIZE;
725 u64 attr_ver;
726 int err;
727
728 /*
729 * Page writeback can extend beyond the lifetime of the
730 * page-cache page, so make sure we read a properly synced
731 * page.
732 */
733 fuse_wait_on_page_writeback(inode, page->index);
734
735 req = fuse_get_req(fc, 1);
736 if (IS_ERR(req))
737 return PTR_ERR(req);
738
739 attr_ver = fuse_get_attr_version(fc);
740
741 req->out.page_zeroing = 1;
742 req->out.argpages = 1;
743 req->num_pages = 1;
744 req->pages[0] = page;
745 req->page_descs[0].length = count;
746 init_sync_kiocb(&iocb, file);
747 io = (struct fuse_io_priv) FUSE_IO_PRIV_SYNC(&iocb);
748 num_read = fuse_send_read(req, &io, pos, count, NULL);
749 err = req->out.h.error;
750
751 if (!err) {
752 /*
753 * Short read means EOF. If file size is larger, truncate it
754 */
755 if (num_read < count)
756 fuse_short_read(req, inode, attr_ver);
757
758 SetPageUptodate(page);
759 }
760
761 fuse_put_request(fc, req);
762
763 return err;
764 }
765
fuse_readpage(struct file * file,struct page * page)766 static int fuse_readpage(struct file *file, struct page *page)
767 {
768 struct inode *inode = page->mapping->host;
769 int err;
770
771 err = -EIO;
772 if (is_bad_inode(inode))
773 goto out;
774
775 err = fuse_do_readpage(file, page);
776 fuse_invalidate_atime(inode);
777 out:
778 unlock_page(page);
779 return err;
780 }
781
fuse_readpages_end(struct fuse_conn * fc,struct fuse_req * req)782 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
783 {
784 int i;
785 size_t count = req->misc.read.in.size;
786 size_t num_read = req->out.args[0].size;
787 struct address_space *mapping = NULL;
788
789 for (i = 0; mapping == NULL && i < req->num_pages; i++)
790 mapping = req->pages[i]->mapping;
791
792 if (mapping) {
793 struct inode *inode = mapping->host;
794
795 /*
796 * Short read means EOF. If file size is larger, truncate it
797 */
798 if (!req->out.h.error && num_read < count)
799 fuse_short_read(req, inode, req->misc.read.attr_ver);
800
801 fuse_invalidate_atime(inode);
802 }
803
804 for (i = 0; i < req->num_pages; i++) {
805 struct page *page = req->pages[i];
806 if (!req->out.h.error)
807 SetPageUptodate(page);
808 else
809 SetPageError(page);
810 unlock_page(page);
811 put_page(page);
812 }
813 if (req->ff)
814 fuse_file_put(req->ff, false, false);
815 }
816
fuse_send_readpages(struct fuse_req * req,struct file * file)817 static void fuse_send_readpages(struct fuse_req *req, struct file *file)
818 {
819 struct fuse_file *ff = file->private_data;
820 struct fuse_conn *fc = ff->fc;
821 loff_t pos = page_offset(req->pages[0]);
822 size_t count = req->num_pages << PAGE_SHIFT;
823
824 req->out.argpages = 1;
825 req->out.page_zeroing = 1;
826 req->out.page_replace = 1;
827 fuse_read_fill(req, file, pos, count, FUSE_READ);
828 req->misc.read.attr_ver = fuse_get_attr_version(fc);
829 if (fc->async_read) {
830 req->ff = fuse_file_get(ff);
831 req->end = fuse_readpages_end;
832 fuse_request_send_background(fc, req);
833 } else {
834 fuse_request_send(fc, req);
835 fuse_readpages_end(fc, req);
836 fuse_put_request(fc, req);
837 }
838 }
839
840 struct fuse_fill_data {
841 struct fuse_req *req;
842 struct file *file;
843 struct inode *inode;
844 unsigned nr_pages;
845 };
846
fuse_readpages_fill(struct file * _data,struct page * page)847 static int fuse_readpages_fill(struct file *_data, struct page *page)
848 {
849 struct fuse_fill_data *data = (struct fuse_fill_data *)_data;
850 struct fuse_req *req = data->req;
851 struct inode *inode = data->inode;
852 struct fuse_conn *fc = get_fuse_conn(inode);
853
854 fuse_wait_on_page_writeback(inode, page->index);
855
856 if (req->num_pages &&
857 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
858 (req->num_pages + 1) * PAGE_SIZE > fc->max_read ||
859 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
860 int nr_alloc = min_t(unsigned, data->nr_pages,
861 FUSE_MAX_PAGES_PER_REQ);
862 fuse_send_readpages(req, data->file);
863 if (fc->async_read)
864 req = fuse_get_req_for_background(fc, nr_alloc);
865 else
866 req = fuse_get_req(fc, nr_alloc);
867
868 data->req = req;
869 if (IS_ERR(req)) {
870 unlock_page(page);
871 return PTR_ERR(req);
872 }
873 }
874
875 if (WARN_ON(req->num_pages >= req->max_pages)) {
876 unlock_page(page);
877 fuse_put_request(fc, req);
878 return -EIO;
879 }
880
881 get_page(page);
882 req->pages[req->num_pages] = page;
883 req->page_descs[req->num_pages].length = PAGE_SIZE;
884 req->num_pages++;
885 data->nr_pages--;
886 return 0;
887 }
888
fuse_readpages(struct file * file,struct address_space * mapping,struct list_head * pages,unsigned nr_pages)889 static int fuse_readpages(struct file *file, struct address_space *mapping,
890 struct list_head *pages, unsigned nr_pages)
891 {
892 struct inode *inode = mapping->host;
893 struct fuse_conn *fc = get_fuse_conn(inode);
894 struct fuse_fill_data data;
895 int err;
896 int nr_alloc = min_t(unsigned, nr_pages, FUSE_MAX_PAGES_PER_REQ);
897
898 err = -EIO;
899 if (is_bad_inode(inode))
900 goto out;
901
902 data.file = file;
903 data.inode = inode;
904 if (fc->async_read)
905 data.req = fuse_get_req_for_background(fc, nr_alloc);
906 else
907 data.req = fuse_get_req(fc, nr_alloc);
908 data.nr_pages = nr_pages;
909 err = PTR_ERR(data.req);
910 if (IS_ERR(data.req))
911 goto out;
912
913 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
914 if (!err) {
915 if (data.req->num_pages)
916 fuse_send_readpages(data.req, file);
917 else
918 fuse_put_request(fc, data.req);
919 }
920 out:
921 return err;
922 }
923
fuse_file_read_iter(struct kiocb * iocb,struct iov_iter * to)924 static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
925 {
926 struct inode *inode = iocb->ki_filp->f_mapping->host;
927 struct fuse_conn *fc = get_fuse_conn(inode);
928
929 /*
930 * In auto invalidate mode, always update attributes on read.
931 * Otherwise, only update if we attempt to read past EOF (to ensure
932 * i_size is up to date).
933 */
934 if (fc->auto_inval_data ||
935 (iocb->ki_pos + iov_iter_count(to) > i_size_read(inode))) {
936 int err;
937 err = fuse_update_attributes(inode, iocb->ki_filp);
938 if (err)
939 return err;
940 }
941
942 return generic_file_read_iter(iocb, to);
943 }
944
fuse_write_fill(struct fuse_req * req,struct fuse_file * ff,loff_t pos,size_t count)945 static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff,
946 loff_t pos, size_t count)
947 {
948 struct fuse_write_in *inarg = &req->misc.write.in;
949 struct fuse_write_out *outarg = &req->misc.write.out;
950
951 inarg->fh = ff->fh;
952 inarg->offset = pos;
953 inarg->size = count;
954 req->in.h.opcode = FUSE_WRITE;
955 req->in.h.nodeid = ff->nodeid;
956 req->in.numargs = 2;
957 if (ff->fc->minor < 9)
958 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
959 else
960 req->in.args[0].size = sizeof(struct fuse_write_in);
961 req->in.args[0].value = inarg;
962 req->in.args[1].size = count;
963 req->out.numargs = 1;
964 req->out.args[0].size = sizeof(struct fuse_write_out);
965 req->out.args[0].value = outarg;
966 }
967
fuse_send_write(struct fuse_req * req,struct fuse_io_priv * io,loff_t pos,size_t count,fl_owner_t owner)968 static size_t fuse_send_write(struct fuse_req *req, struct fuse_io_priv *io,
969 loff_t pos, size_t count, fl_owner_t owner)
970 {
971 struct kiocb *iocb = io->iocb;
972 struct file *file = iocb->ki_filp;
973 struct fuse_file *ff = file->private_data;
974 struct fuse_conn *fc = ff->fc;
975 struct fuse_write_in *inarg = &req->misc.write.in;
976
977 fuse_write_fill(req, ff, pos, count);
978 inarg->flags = file->f_flags;
979 if (iocb->ki_flags & IOCB_DSYNC)
980 inarg->flags |= O_DSYNC;
981 if (iocb->ki_flags & IOCB_SYNC)
982 inarg->flags |= O_SYNC;
983 if (owner != NULL) {
984 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
985 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
986 }
987
988 if (io->async)
989 return fuse_async_req_send(fc, req, count, io);
990
991 fuse_request_send(fc, req);
992 return req->misc.write.out.size;
993 }
994
fuse_write_update_size(struct inode * inode,loff_t pos)995 bool fuse_write_update_size(struct inode *inode, loff_t pos)
996 {
997 struct fuse_conn *fc = get_fuse_conn(inode);
998 struct fuse_inode *fi = get_fuse_inode(inode);
999 bool ret = false;
1000
1001 spin_lock(&fc->lock);
1002 fi->attr_version = ++fc->attr_version;
1003 if (pos > inode->i_size) {
1004 i_size_write(inode, pos);
1005 ret = true;
1006 }
1007 spin_unlock(&fc->lock);
1008
1009 return ret;
1010 }
1011
fuse_send_write_pages(struct fuse_req * req,struct kiocb * iocb,struct inode * inode,loff_t pos,size_t count)1012 static size_t fuse_send_write_pages(struct fuse_req *req, struct kiocb *iocb,
1013 struct inode *inode, loff_t pos,
1014 size_t count)
1015 {
1016 size_t res;
1017 unsigned offset;
1018 unsigned i;
1019 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
1020
1021 for (i = 0; i < req->num_pages; i++)
1022 fuse_wait_on_page_writeback(inode, req->pages[i]->index);
1023
1024 res = fuse_send_write(req, &io, pos, count, NULL);
1025
1026 offset = req->page_descs[0].offset;
1027 count = res;
1028 for (i = 0; i < req->num_pages; i++) {
1029 struct page *page = req->pages[i];
1030
1031 if (!req->out.h.error && !offset && count >= PAGE_SIZE)
1032 SetPageUptodate(page);
1033
1034 if (count > PAGE_SIZE - offset)
1035 count -= PAGE_SIZE - offset;
1036 else
1037 count = 0;
1038 offset = 0;
1039
1040 unlock_page(page);
1041 put_page(page);
1042 }
1043
1044 return res;
1045 }
1046
fuse_fill_write_pages(struct fuse_req * req,struct address_space * mapping,struct iov_iter * ii,loff_t pos)1047 static ssize_t fuse_fill_write_pages(struct fuse_req *req,
1048 struct address_space *mapping,
1049 struct iov_iter *ii, loff_t pos)
1050 {
1051 struct fuse_conn *fc = get_fuse_conn(mapping->host);
1052 unsigned offset = pos & (PAGE_SIZE - 1);
1053 size_t count = 0;
1054 int err;
1055
1056 req->in.argpages = 1;
1057 req->page_descs[0].offset = offset;
1058
1059 do {
1060 size_t tmp;
1061 struct page *page;
1062 pgoff_t index = pos >> PAGE_SHIFT;
1063 size_t bytes = min_t(size_t, PAGE_SIZE - offset,
1064 iov_iter_count(ii));
1065
1066 bytes = min_t(size_t, bytes, fc->max_write - count);
1067
1068 again:
1069 err = -EFAULT;
1070 if (iov_iter_fault_in_readable(ii, bytes))
1071 break;
1072
1073 err = -ENOMEM;
1074 page = grab_cache_page_write_begin(mapping, index, 0);
1075 if (!page)
1076 break;
1077
1078 if (mapping_writably_mapped(mapping))
1079 flush_dcache_page(page);
1080
1081 tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes);
1082 flush_dcache_page(page);
1083
1084 iov_iter_advance(ii, tmp);
1085 if (!tmp) {
1086 unlock_page(page);
1087 put_page(page);
1088 bytes = min(bytes, iov_iter_single_seg_count(ii));
1089 goto again;
1090 }
1091
1092 err = 0;
1093 req->pages[req->num_pages] = page;
1094 req->page_descs[req->num_pages].length = tmp;
1095 req->num_pages++;
1096
1097 count += tmp;
1098 pos += tmp;
1099 offset += tmp;
1100 if (offset == PAGE_SIZE)
1101 offset = 0;
1102
1103 if (!fc->big_writes)
1104 break;
1105 } while (iov_iter_count(ii) && count < fc->max_write &&
1106 req->num_pages < req->max_pages && offset == 0);
1107
1108 return count > 0 ? count : err;
1109 }
1110
fuse_wr_pages(loff_t pos,size_t len)1111 static inline unsigned fuse_wr_pages(loff_t pos, size_t len)
1112 {
1113 return min_t(unsigned,
1114 ((pos + len - 1) >> PAGE_SHIFT) -
1115 (pos >> PAGE_SHIFT) + 1,
1116 FUSE_MAX_PAGES_PER_REQ);
1117 }
1118
fuse_perform_write(struct kiocb * iocb,struct address_space * mapping,struct iov_iter * ii,loff_t pos)1119 static ssize_t fuse_perform_write(struct kiocb *iocb,
1120 struct address_space *mapping,
1121 struct iov_iter *ii, loff_t pos)
1122 {
1123 struct inode *inode = mapping->host;
1124 struct fuse_conn *fc = get_fuse_conn(inode);
1125 struct fuse_inode *fi = get_fuse_inode(inode);
1126 int err = 0;
1127 ssize_t res = 0;
1128
1129 if (is_bad_inode(inode))
1130 return -EIO;
1131
1132 if (inode->i_size < pos + iov_iter_count(ii))
1133 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1134
1135 do {
1136 struct fuse_req *req;
1137 ssize_t count;
1138 unsigned nr_pages = fuse_wr_pages(pos, iov_iter_count(ii));
1139
1140 req = fuse_get_req(fc, nr_pages);
1141 if (IS_ERR(req)) {
1142 err = PTR_ERR(req);
1143 break;
1144 }
1145
1146 count = fuse_fill_write_pages(req, mapping, ii, pos);
1147 if (count <= 0) {
1148 err = count;
1149 } else {
1150 size_t num_written;
1151
1152 num_written = fuse_send_write_pages(req, iocb, inode,
1153 pos, count);
1154 err = req->out.h.error;
1155 if (!err) {
1156 res += num_written;
1157 pos += num_written;
1158
1159 /* break out of the loop on short write */
1160 if (num_written != count)
1161 err = -EIO;
1162 }
1163 }
1164 fuse_put_request(fc, req);
1165 } while (!err && iov_iter_count(ii));
1166
1167 if (res > 0)
1168 fuse_write_update_size(inode, pos);
1169
1170 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1171 fuse_invalidate_attr(inode);
1172
1173 return res > 0 ? res : err;
1174 }
1175
fuse_file_write_iter(struct kiocb * iocb,struct iov_iter * from)1176 static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1177 {
1178 struct file *file = iocb->ki_filp;
1179 struct address_space *mapping = file->f_mapping;
1180 ssize_t written = 0;
1181 ssize_t written_buffered = 0;
1182 struct inode *inode = mapping->host;
1183 ssize_t err;
1184 loff_t endbyte = 0;
1185
1186 if (get_fuse_conn(inode)->writeback_cache) {
1187 /* Update size (EOF optimization) and mode (SUID clearing) */
1188 err = fuse_update_attributes(mapping->host, file);
1189 if (err)
1190 return err;
1191
1192 return generic_file_write_iter(iocb, from);
1193 }
1194
1195 inode_lock(inode);
1196
1197 /* We can write back this queue in page reclaim */
1198 current->backing_dev_info = inode_to_bdi(inode);
1199
1200 err = generic_write_checks(iocb, from);
1201 if (err <= 0)
1202 goto out;
1203
1204 err = file_remove_privs(file);
1205 if (err)
1206 goto out;
1207
1208 err = file_update_time(file);
1209 if (err)
1210 goto out;
1211
1212 if (iocb->ki_flags & IOCB_DIRECT) {
1213 loff_t pos = iocb->ki_pos;
1214 written = generic_file_direct_write(iocb, from);
1215 if (written < 0 || !iov_iter_count(from))
1216 goto out;
1217
1218 pos += written;
1219
1220 written_buffered = fuse_perform_write(iocb, mapping, from, pos);
1221 if (written_buffered < 0) {
1222 err = written_buffered;
1223 goto out;
1224 }
1225 endbyte = pos + written_buffered - 1;
1226
1227 err = filemap_write_and_wait_range(file->f_mapping, pos,
1228 endbyte);
1229 if (err)
1230 goto out;
1231
1232 invalidate_mapping_pages(file->f_mapping,
1233 pos >> PAGE_SHIFT,
1234 endbyte >> PAGE_SHIFT);
1235
1236 written += written_buffered;
1237 iocb->ki_pos = pos + written_buffered;
1238 } else {
1239 written = fuse_perform_write(iocb, mapping, from, iocb->ki_pos);
1240 if (written >= 0)
1241 iocb->ki_pos += written;
1242 }
1243 out:
1244 current->backing_dev_info = NULL;
1245 inode_unlock(inode);
1246 if (written > 0)
1247 written = generic_write_sync(iocb, written);
1248
1249 return written ? written : err;
1250 }
1251
fuse_page_descs_length_init(struct fuse_req * req,unsigned index,unsigned nr_pages)1252 static inline void fuse_page_descs_length_init(struct fuse_req *req,
1253 unsigned index, unsigned nr_pages)
1254 {
1255 int i;
1256
1257 for (i = index; i < index + nr_pages; i++)
1258 req->page_descs[i].length = PAGE_SIZE -
1259 req->page_descs[i].offset;
1260 }
1261
fuse_get_user_addr(const struct iov_iter * ii)1262 static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii)
1263 {
1264 return (unsigned long)ii->iov->iov_base + ii->iov_offset;
1265 }
1266
fuse_get_frag_size(const struct iov_iter * ii,size_t max_size)1267 static inline size_t fuse_get_frag_size(const struct iov_iter *ii,
1268 size_t max_size)
1269 {
1270 return min(iov_iter_single_seg_count(ii), max_size);
1271 }
1272
fuse_get_user_pages(struct fuse_req * req,struct iov_iter * ii,size_t * nbytesp,int write)1273 static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii,
1274 size_t *nbytesp, int write)
1275 {
1276 size_t nbytes = 0; /* # bytes already packed in req */
1277 ssize_t ret = 0;
1278
1279 /* Special case for kernel I/O: can copy directly into the buffer */
1280 if (ii->type & ITER_KVEC) {
1281 unsigned long user_addr = fuse_get_user_addr(ii);
1282 size_t frag_size = fuse_get_frag_size(ii, *nbytesp);
1283
1284 if (write)
1285 req->in.args[1].value = (void *) user_addr;
1286 else
1287 req->out.args[0].value = (void *) user_addr;
1288
1289 iov_iter_advance(ii, frag_size);
1290 *nbytesp = frag_size;
1291 return 0;
1292 }
1293
1294 while (nbytes < *nbytesp && req->num_pages < req->max_pages) {
1295 unsigned npages;
1296 size_t start;
1297 ret = iov_iter_get_pages(ii, &req->pages[req->num_pages],
1298 *nbytesp - nbytes,
1299 req->max_pages - req->num_pages,
1300 &start);
1301 if (ret < 0)
1302 break;
1303
1304 iov_iter_advance(ii, ret);
1305 nbytes += ret;
1306
1307 ret += start;
1308 npages = (ret + PAGE_SIZE - 1) / PAGE_SIZE;
1309
1310 req->page_descs[req->num_pages].offset = start;
1311 fuse_page_descs_length_init(req, req->num_pages, npages);
1312
1313 req->num_pages += npages;
1314 req->page_descs[req->num_pages - 1].length -=
1315 (PAGE_SIZE - ret) & (PAGE_SIZE - 1);
1316 }
1317
1318 if (write)
1319 req->in.argpages = 1;
1320 else
1321 req->out.argpages = 1;
1322
1323 *nbytesp = nbytes;
1324
1325 return ret < 0 ? ret : 0;
1326 }
1327
fuse_iter_npages(const struct iov_iter * ii_p)1328 static inline int fuse_iter_npages(const struct iov_iter *ii_p)
1329 {
1330 return iov_iter_npages(ii_p, FUSE_MAX_PAGES_PER_REQ);
1331 }
1332
fuse_direct_io(struct fuse_io_priv * io,struct iov_iter * iter,loff_t * ppos,int flags)1333 ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
1334 loff_t *ppos, int flags)
1335 {
1336 int write = flags & FUSE_DIO_WRITE;
1337 int cuse = flags & FUSE_DIO_CUSE;
1338 struct file *file = io->iocb->ki_filp;
1339 struct inode *inode = file->f_mapping->host;
1340 struct fuse_file *ff = file->private_data;
1341 struct fuse_conn *fc = ff->fc;
1342 size_t nmax = write ? fc->max_write : fc->max_read;
1343 loff_t pos = *ppos;
1344 size_t count = iov_iter_count(iter);
1345 pgoff_t idx_from = pos >> PAGE_SHIFT;
1346 pgoff_t idx_to = (pos + count - 1) >> PAGE_SHIFT;
1347 ssize_t res = 0;
1348 struct fuse_req *req;
1349 int err = 0;
1350
1351 if (io->async)
1352 req = fuse_get_req_for_background(fc, fuse_iter_npages(iter));
1353 else
1354 req = fuse_get_req(fc, fuse_iter_npages(iter));
1355 if (IS_ERR(req))
1356 return PTR_ERR(req);
1357
1358 if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) {
1359 if (!write)
1360 inode_lock(inode);
1361 fuse_sync_writes(inode);
1362 if (!write)
1363 inode_unlock(inode);
1364 }
1365
1366 io->should_dirty = !write && iter_is_iovec(iter);
1367 while (count) {
1368 size_t nres;
1369 fl_owner_t owner = current->files;
1370 size_t nbytes = min(count, nmax);
1371 err = fuse_get_user_pages(req, iter, &nbytes, write);
1372 if (err && !nbytes)
1373 break;
1374
1375 if (write)
1376 nres = fuse_send_write(req, io, pos, nbytes, owner);
1377 else
1378 nres = fuse_send_read(req, io, pos, nbytes, owner);
1379
1380 if (!io->async)
1381 fuse_release_user_pages(req, io->should_dirty);
1382 if (req->out.h.error) {
1383 err = req->out.h.error;
1384 break;
1385 } else if (nres > nbytes) {
1386 res = 0;
1387 err = -EIO;
1388 break;
1389 }
1390 count -= nres;
1391 res += nres;
1392 pos += nres;
1393 if (nres != nbytes)
1394 break;
1395 if (count) {
1396 fuse_put_request(fc, req);
1397 if (io->async)
1398 req = fuse_get_req_for_background(fc,
1399 fuse_iter_npages(iter));
1400 else
1401 req = fuse_get_req(fc, fuse_iter_npages(iter));
1402 if (IS_ERR(req))
1403 break;
1404 }
1405 }
1406 if (!IS_ERR(req))
1407 fuse_put_request(fc, req);
1408 if (res > 0)
1409 *ppos = pos;
1410
1411 return res > 0 ? res : err;
1412 }
1413 EXPORT_SYMBOL_GPL(fuse_direct_io);
1414
__fuse_direct_read(struct fuse_io_priv * io,struct iov_iter * iter,loff_t * ppos)1415 static ssize_t __fuse_direct_read(struct fuse_io_priv *io,
1416 struct iov_iter *iter,
1417 loff_t *ppos)
1418 {
1419 ssize_t res;
1420 struct inode *inode = file_inode(io->iocb->ki_filp);
1421
1422 if (is_bad_inode(inode))
1423 return -EIO;
1424
1425 res = fuse_direct_io(io, iter, ppos, 0);
1426
1427 fuse_invalidate_attr(inode);
1428
1429 return res;
1430 }
1431
fuse_direct_read_iter(struct kiocb * iocb,struct iov_iter * to)1432 static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to)
1433 {
1434 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
1435 return __fuse_direct_read(&io, to, &iocb->ki_pos);
1436 }
1437
fuse_direct_write_iter(struct kiocb * iocb,struct iov_iter * from)1438 static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
1439 {
1440 struct inode *inode = file_inode(iocb->ki_filp);
1441 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
1442 ssize_t res;
1443
1444 if (is_bad_inode(inode))
1445 return -EIO;
1446
1447 /* Don't allow parallel writes to the same file */
1448 inode_lock(inode);
1449 res = generic_write_checks(iocb, from);
1450 if (res > 0)
1451 res = fuse_direct_io(&io, from, &iocb->ki_pos, FUSE_DIO_WRITE);
1452 fuse_invalidate_attr(inode);
1453 if (res > 0)
1454 fuse_write_update_size(inode, iocb->ki_pos);
1455 inode_unlock(inode);
1456
1457 return res;
1458 }
1459
fuse_writepage_free(struct fuse_conn * fc,struct fuse_req * req)1460 static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
1461 {
1462 int i;
1463
1464 for (i = 0; i < req->num_pages; i++)
1465 __free_page(req->pages[i]);
1466
1467 if (req->ff)
1468 fuse_file_put(req->ff, false, false);
1469 }
1470
fuse_writepage_finish(struct fuse_conn * fc,struct fuse_req * req)1471 static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
1472 {
1473 struct inode *inode = req->inode;
1474 struct fuse_inode *fi = get_fuse_inode(inode);
1475 struct backing_dev_info *bdi = inode_to_bdi(inode);
1476 int i;
1477
1478 list_del(&req->writepages_entry);
1479 for (i = 0; i < req->num_pages; i++) {
1480 dec_wb_stat(&bdi->wb, WB_WRITEBACK);
1481 dec_node_page_state(req->pages[i], NR_WRITEBACK_TEMP);
1482 wb_writeout_inc(&bdi->wb);
1483 }
1484 wake_up(&fi->page_waitq);
1485 }
1486
1487 /* Called under fc->lock, may release and reacquire it */
fuse_send_writepage(struct fuse_conn * fc,struct fuse_req * req,loff_t size)1488 static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req,
1489 loff_t size)
1490 __releases(fc->lock)
1491 __acquires(fc->lock)
1492 {
1493 struct fuse_inode *fi = get_fuse_inode(req->inode);
1494 struct fuse_write_in *inarg = &req->misc.write.in;
1495 __u64 data_size = req->num_pages * PAGE_SIZE;
1496
1497 if (!fc->connected)
1498 goto out_free;
1499
1500 if (inarg->offset + data_size <= size) {
1501 inarg->size = data_size;
1502 } else if (inarg->offset < size) {
1503 inarg->size = size - inarg->offset;
1504 } else {
1505 /* Got truncated off completely */
1506 goto out_free;
1507 }
1508
1509 req->in.args[1].size = inarg->size;
1510 fi->writectr++;
1511 fuse_request_send_background_locked(fc, req);
1512 return;
1513
1514 out_free:
1515 fuse_writepage_finish(fc, req);
1516 spin_unlock(&fc->lock);
1517 fuse_writepage_free(fc, req);
1518 fuse_put_request(fc, req);
1519 spin_lock(&fc->lock);
1520 }
1521
1522 /*
1523 * If fi->writectr is positive (no truncate or fsync going on) send
1524 * all queued writepage requests.
1525 *
1526 * Called with fc->lock
1527 */
fuse_flush_writepages(struct inode * inode)1528 void fuse_flush_writepages(struct inode *inode)
1529 __releases(fc->lock)
1530 __acquires(fc->lock)
1531 {
1532 struct fuse_conn *fc = get_fuse_conn(inode);
1533 struct fuse_inode *fi = get_fuse_inode(inode);
1534 loff_t crop = i_size_read(inode);
1535 struct fuse_req *req;
1536
1537 while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
1538 req = list_entry(fi->queued_writes.next, struct fuse_req, list);
1539 list_del_init(&req->list);
1540 fuse_send_writepage(fc, req, crop);
1541 }
1542 }
1543
fuse_writepage_end(struct fuse_conn * fc,struct fuse_req * req)1544 static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
1545 {
1546 struct inode *inode = req->inode;
1547 struct fuse_inode *fi = get_fuse_inode(inode);
1548
1549 mapping_set_error(inode->i_mapping, req->out.h.error);
1550 spin_lock(&fc->lock);
1551 while (req->misc.write.next) {
1552 struct fuse_conn *fc = get_fuse_conn(inode);
1553 struct fuse_write_in *inarg = &req->misc.write.in;
1554 struct fuse_req *next = req->misc.write.next;
1555 req->misc.write.next = next->misc.write.next;
1556 next->misc.write.next = NULL;
1557 next->ff = fuse_file_get(req->ff);
1558 list_add(&next->writepages_entry, &fi->writepages);
1559
1560 /*
1561 * Skip fuse_flush_writepages() to make it easy to crop requests
1562 * based on primary request size.
1563 *
1564 * 1st case (trivial): there are no concurrent activities using
1565 * fuse_set/release_nowrite. Then we're on safe side because
1566 * fuse_flush_writepages() would call fuse_send_writepage()
1567 * anyway.
1568 *
1569 * 2nd case: someone called fuse_set_nowrite and it is waiting
1570 * now for completion of all in-flight requests. This happens
1571 * rarely and no more than once per page, so this should be
1572 * okay.
1573 *
1574 * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle
1575 * of fuse_set_nowrite..fuse_release_nowrite section. The fact
1576 * that fuse_set_nowrite returned implies that all in-flight
1577 * requests were completed along with all of their secondary
1578 * requests. Further primary requests are blocked by negative
1579 * writectr. Hence there cannot be any in-flight requests and
1580 * no invocations of fuse_writepage_end() while we're in
1581 * fuse_set_nowrite..fuse_release_nowrite section.
1582 */
1583 fuse_send_writepage(fc, next, inarg->offset + inarg->size);
1584 }
1585 fi->writectr--;
1586 fuse_writepage_finish(fc, req);
1587 spin_unlock(&fc->lock);
1588 fuse_writepage_free(fc, req);
1589 }
1590
__fuse_write_file_get(struct fuse_conn * fc,struct fuse_inode * fi)1591 static struct fuse_file *__fuse_write_file_get(struct fuse_conn *fc,
1592 struct fuse_inode *fi)
1593 {
1594 struct fuse_file *ff = NULL;
1595
1596 spin_lock(&fc->lock);
1597 if (!list_empty(&fi->write_files)) {
1598 ff = list_entry(fi->write_files.next, struct fuse_file,
1599 write_entry);
1600 fuse_file_get(ff);
1601 }
1602 spin_unlock(&fc->lock);
1603
1604 return ff;
1605 }
1606
fuse_write_file_get(struct fuse_conn * fc,struct fuse_inode * fi)1607 static struct fuse_file *fuse_write_file_get(struct fuse_conn *fc,
1608 struct fuse_inode *fi)
1609 {
1610 struct fuse_file *ff = __fuse_write_file_get(fc, fi);
1611 WARN_ON(!ff);
1612 return ff;
1613 }
1614
fuse_write_inode(struct inode * inode,struct writeback_control * wbc)1615 int fuse_write_inode(struct inode *inode, struct writeback_control *wbc)
1616 {
1617 struct fuse_conn *fc = get_fuse_conn(inode);
1618 struct fuse_inode *fi = get_fuse_inode(inode);
1619 struct fuse_file *ff;
1620 int err;
1621
1622 ff = __fuse_write_file_get(fc, fi);
1623 err = fuse_flush_times(inode, ff);
1624 if (ff)
1625 fuse_file_put(ff, false, false);
1626
1627 return err;
1628 }
1629
fuse_writepage_locked(struct page * page)1630 static int fuse_writepage_locked(struct page *page)
1631 {
1632 struct address_space *mapping = page->mapping;
1633 struct inode *inode = mapping->host;
1634 struct fuse_conn *fc = get_fuse_conn(inode);
1635 struct fuse_inode *fi = get_fuse_inode(inode);
1636 struct fuse_req *req;
1637 struct page *tmp_page;
1638 int error = -ENOMEM;
1639
1640 set_page_writeback(page);
1641
1642 req = fuse_request_alloc_nofs(1);
1643 if (!req)
1644 goto err;
1645
1646 /* writeback always goes to bg_queue */
1647 __set_bit(FR_BACKGROUND, &req->flags);
1648 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1649 if (!tmp_page)
1650 goto err_free;
1651
1652 error = -EIO;
1653 req->ff = fuse_write_file_get(fc, fi);
1654 if (!req->ff)
1655 goto err_nofile;
1656
1657 fuse_write_fill(req, req->ff, page_offset(page), 0);
1658
1659 copy_highpage(tmp_page, page);
1660 req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
1661 req->misc.write.next = NULL;
1662 req->in.argpages = 1;
1663 req->num_pages = 1;
1664 req->pages[0] = tmp_page;
1665 req->page_descs[0].offset = 0;
1666 req->page_descs[0].length = PAGE_SIZE;
1667 req->end = fuse_writepage_end;
1668 req->inode = inode;
1669
1670 inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
1671 inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP);
1672
1673 spin_lock(&fc->lock);
1674 list_add(&req->writepages_entry, &fi->writepages);
1675 list_add_tail(&req->list, &fi->queued_writes);
1676 fuse_flush_writepages(inode);
1677 spin_unlock(&fc->lock);
1678
1679 end_page_writeback(page);
1680
1681 return 0;
1682
1683 err_nofile:
1684 __free_page(tmp_page);
1685 err_free:
1686 fuse_request_free(req);
1687 err:
1688 mapping_set_error(page->mapping, error);
1689 end_page_writeback(page);
1690 return error;
1691 }
1692
fuse_writepage(struct page * page,struct writeback_control * wbc)1693 static int fuse_writepage(struct page *page, struct writeback_control *wbc)
1694 {
1695 int err;
1696
1697 if (fuse_page_is_writeback(page->mapping->host, page->index)) {
1698 /*
1699 * ->writepages() should be called for sync() and friends. We
1700 * should only get here on direct reclaim and then we are
1701 * allowed to skip a page which is already in flight
1702 */
1703 WARN_ON(wbc->sync_mode == WB_SYNC_ALL);
1704
1705 redirty_page_for_writepage(wbc, page);
1706 unlock_page(page);
1707 return 0;
1708 }
1709
1710 err = fuse_writepage_locked(page);
1711 unlock_page(page);
1712
1713 return err;
1714 }
1715
1716 struct fuse_fill_wb_data {
1717 struct fuse_req *req;
1718 struct fuse_file *ff;
1719 struct inode *inode;
1720 struct page **orig_pages;
1721 };
1722
fuse_writepages_send(struct fuse_fill_wb_data * data)1723 static void fuse_writepages_send(struct fuse_fill_wb_data *data)
1724 {
1725 struct fuse_req *req = data->req;
1726 struct inode *inode = data->inode;
1727 struct fuse_conn *fc = get_fuse_conn(inode);
1728 struct fuse_inode *fi = get_fuse_inode(inode);
1729 int num_pages = req->num_pages;
1730 int i;
1731
1732 req->ff = fuse_file_get(data->ff);
1733 spin_lock(&fc->lock);
1734 list_add_tail(&req->list, &fi->queued_writes);
1735 fuse_flush_writepages(inode);
1736 spin_unlock(&fc->lock);
1737
1738 for (i = 0; i < num_pages; i++)
1739 end_page_writeback(data->orig_pages[i]);
1740 }
1741
fuse_writepage_in_flight(struct fuse_req * new_req,struct page * page)1742 static bool fuse_writepage_in_flight(struct fuse_req *new_req,
1743 struct page *page)
1744 {
1745 struct fuse_conn *fc = get_fuse_conn(new_req->inode);
1746 struct fuse_inode *fi = get_fuse_inode(new_req->inode);
1747 struct fuse_req *tmp;
1748 struct fuse_req *old_req;
1749 bool found = false;
1750 pgoff_t curr_index;
1751
1752 BUG_ON(new_req->num_pages != 0);
1753
1754 spin_lock(&fc->lock);
1755 list_del(&new_req->writepages_entry);
1756 list_for_each_entry(old_req, &fi->writepages, writepages_entry) {
1757 BUG_ON(old_req->inode != new_req->inode);
1758 curr_index = old_req->misc.write.in.offset >> PAGE_SHIFT;
1759 if (curr_index <= page->index &&
1760 page->index < curr_index + old_req->num_pages) {
1761 found = true;
1762 break;
1763 }
1764 }
1765 if (!found) {
1766 list_add(&new_req->writepages_entry, &fi->writepages);
1767 goto out_unlock;
1768 }
1769
1770 new_req->num_pages = 1;
1771 for (tmp = old_req; tmp != NULL; tmp = tmp->misc.write.next) {
1772 BUG_ON(tmp->inode != new_req->inode);
1773 curr_index = tmp->misc.write.in.offset >> PAGE_SHIFT;
1774 if (tmp->num_pages == 1 &&
1775 curr_index == page->index) {
1776 old_req = tmp;
1777 }
1778 }
1779
1780 if (old_req->num_pages == 1 && test_bit(FR_PENDING, &old_req->flags)) {
1781 struct backing_dev_info *bdi = inode_to_bdi(page->mapping->host);
1782
1783 copy_highpage(old_req->pages[0], page);
1784 spin_unlock(&fc->lock);
1785
1786 dec_wb_stat(&bdi->wb, WB_WRITEBACK);
1787 dec_node_page_state(new_req->pages[0], NR_WRITEBACK_TEMP);
1788 wb_writeout_inc(&bdi->wb);
1789 fuse_writepage_free(fc, new_req);
1790 fuse_request_free(new_req);
1791 goto out;
1792 } else {
1793 new_req->misc.write.next = old_req->misc.write.next;
1794 old_req->misc.write.next = new_req;
1795 }
1796 out_unlock:
1797 spin_unlock(&fc->lock);
1798 out:
1799 return found;
1800 }
1801
fuse_writepages_fill(struct page * page,struct writeback_control * wbc,void * _data)1802 static int fuse_writepages_fill(struct page *page,
1803 struct writeback_control *wbc, void *_data)
1804 {
1805 struct fuse_fill_wb_data *data = _data;
1806 struct fuse_req *req = data->req;
1807 struct inode *inode = data->inode;
1808 struct fuse_conn *fc = get_fuse_conn(inode);
1809 struct page *tmp_page;
1810 bool is_writeback;
1811 int err;
1812
1813 if (!data->ff) {
1814 err = -EIO;
1815 data->ff = fuse_write_file_get(fc, get_fuse_inode(inode));
1816 if (!data->ff)
1817 goto out_unlock;
1818 }
1819
1820 /*
1821 * Being under writeback is unlikely but possible. For example direct
1822 * read to an mmaped fuse file will set the page dirty twice; once when
1823 * the pages are faulted with get_user_pages(), and then after the read
1824 * completed.
1825 */
1826 is_writeback = fuse_page_is_writeback(inode, page->index);
1827
1828 if (req && req->num_pages &&
1829 (is_writeback || req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
1830 (req->num_pages + 1) * PAGE_SIZE > fc->max_write ||
1831 data->orig_pages[req->num_pages - 1]->index + 1 != page->index)) {
1832 fuse_writepages_send(data);
1833 data->req = NULL;
1834 }
1835 err = -ENOMEM;
1836 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1837 if (!tmp_page)
1838 goto out_unlock;
1839
1840 /*
1841 * The page must not be redirtied until the writeout is completed
1842 * (i.e. userspace has sent a reply to the write request). Otherwise
1843 * there could be more than one temporary page instance for each real
1844 * page.
1845 *
1846 * This is ensured by holding the page lock in page_mkwrite() while
1847 * checking fuse_page_is_writeback(). We already hold the page lock
1848 * since clear_page_dirty_for_io() and keep it held until we add the
1849 * request to the fi->writepages list and increment req->num_pages.
1850 * After this fuse_page_is_writeback() will indicate that the page is
1851 * under writeback, so we can release the page lock.
1852 */
1853 if (data->req == NULL) {
1854 struct fuse_inode *fi = get_fuse_inode(inode);
1855
1856 err = -ENOMEM;
1857 req = fuse_request_alloc_nofs(FUSE_MAX_PAGES_PER_REQ);
1858 if (!req) {
1859 __free_page(tmp_page);
1860 goto out_unlock;
1861 }
1862
1863 fuse_write_fill(req, data->ff, page_offset(page), 0);
1864 req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
1865 req->misc.write.next = NULL;
1866 req->in.argpages = 1;
1867 __set_bit(FR_BACKGROUND, &req->flags);
1868 req->num_pages = 0;
1869 req->end = fuse_writepage_end;
1870 req->inode = inode;
1871
1872 spin_lock(&fc->lock);
1873 list_add(&req->writepages_entry, &fi->writepages);
1874 spin_unlock(&fc->lock);
1875
1876 data->req = req;
1877 }
1878 set_page_writeback(page);
1879
1880 copy_highpage(tmp_page, page);
1881 req->pages[req->num_pages] = tmp_page;
1882 req->page_descs[req->num_pages].offset = 0;
1883 req->page_descs[req->num_pages].length = PAGE_SIZE;
1884
1885 inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
1886 inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP);
1887
1888 err = 0;
1889 if (is_writeback && fuse_writepage_in_flight(req, page)) {
1890 end_page_writeback(page);
1891 data->req = NULL;
1892 goto out_unlock;
1893 }
1894 data->orig_pages[req->num_pages] = page;
1895
1896 /*
1897 * Protected by fc->lock against concurrent access by
1898 * fuse_page_is_writeback().
1899 */
1900 spin_lock(&fc->lock);
1901 req->num_pages++;
1902 spin_unlock(&fc->lock);
1903
1904 out_unlock:
1905 unlock_page(page);
1906
1907 return err;
1908 }
1909
fuse_writepages(struct address_space * mapping,struct writeback_control * wbc)1910 static int fuse_writepages(struct address_space *mapping,
1911 struct writeback_control *wbc)
1912 {
1913 struct inode *inode = mapping->host;
1914 struct fuse_fill_wb_data data;
1915 int err;
1916
1917 err = -EIO;
1918 if (is_bad_inode(inode))
1919 goto out;
1920
1921 data.inode = inode;
1922 data.req = NULL;
1923 data.ff = NULL;
1924
1925 err = -ENOMEM;
1926 data.orig_pages = kcalloc(FUSE_MAX_PAGES_PER_REQ,
1927 sizeof(struct page *),
1928 GFP_NOFS);
1929 if (!data.orig_pages)
1930 goto out;
1931
1932 err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data);
1933 if (data.req) {
1934 /* Ignore errors if we can write at least one page */
1935 BUG_ON(!data.req->num_pages);
1936 fuse_writepages_send(&data);
1937 err = 0;
1938 }
1939 if (data.ff)
1940 fuse_file_put(data.ff, false, false);
1941
1942 kfree(data.orig_pages);
1943 out:
1944 return err;
1945 }
1946
1947 /*
1948 * It's worthy to make sure that space is reserved on disk for the write,
1949 * but how to implement it without killing performance need more thinking.
1950 */
fuse_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned flags,struct page ** pagep,void ** fsdata)1951 static int fuse_write_begin(struct file *file, struct address_space *mapping,
1952 loff_t pos, unsigned len, unsigned flags,
1953 struct page **pagep, void **fsdata)
1954 {
1955 pgoff_t index = pos >> PAGE_SHIFT;
1956 struct fuse_conn *fc = get_fuse_conn(file_inode(file));
1957 struct page *page;
1958 loff_t fsize;
1959 int err = -ENOMEM;
1960
1961 WARN_ON(!fc->writeback_cache);
1962
1963 page = grab_cache_page_write_begin(mapping, index, flags);
1964 if (!page)
1965 goto error;
1966
1967 fuse_wait_on_page_writeback(mapping->host, page->index);
1968
1969 if (PageUptodate(page) || len == PAGE_SIZE)
1970 goto success;
1971 /*
1972 * Check if the start this page comes after the end of file, in which
1973 * case the readpage can be optimized away.
1974 */
1975 fsize = i_size_read(mapping->host);
1976 if (fsize <= (pos & PAGE_MASK)) {
1977 size_t off = pos & ~PAGE_MASK;
1978 if (off)
1979 zero_user_segment(page, 0, off);
1980 goto success;
1981 }
1982 err = fuse_do_readpage(file, page);
1983 if (err)
1984 goto cleanup;
1985 success:
1986 *pagep = page;
1987 return 0;
1988
1989 cleanup:
1990 unlock_page(page);
1991 put_page(page);
1992 error:
1993 return err;
1994 }
1995
fuse_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * page,void * fsdata)1996 static int fuse_write_end(struct file *file, struct address_space *mapping,
1997 loff_t pos, unsigned len, unsigned copied,
1998 struct page *page, void *fsdata)
1999 {
2000 struct inode *inode = page->mapping->host;
2001
2002 /* Haven't copied anything? Skip zeroing, size extending, dirtying. */
2003 if (!copied)
2004 goto unlock;
2005
2006 if (!PageUptodate(page)) {
2007 /* Zero any unwritten bytes at the end of the page */
2008 size_t endoff = (pos + copied) & ~PAGE_MASK;
2009 if (endoff)
2010 zero_user_segment(page, endoff, PAGE_SIZE);
2011 SetPageUptodate(page);
2012 }
2013
2014 fuse_write_update_size(inode, pos + copied);
2015 set_page_dirty(page);
2016
2017 unlock:
2018 unlock_page(page);
2019 put_page(page);
2020
2021 return copied;
2022 }
2023
fuse_launder_page(struct page * page)2024 static int fuse_launder_page(struct page *page)
2025 {
2026 int err = 0;
2027 if (clear_page_dirty_for_io(page)) {
2028 struct inode *inode = page->mapping->host;
2029 err = fuse_writepage_locked(page);
2030 if (!err)
2031 fuse_wait_on_page_writeback(inode, page->index);
2032 }
2033 return err;
2034 }
2035
2036 /*
2037 * Write back dirty pages now, because there may not be any suitable
2038 * open files later
2039 */
fuse_vma_close(struct vm_area_struct * vma)2040 static void fuse_vma_close(struct vm_area_struct *vma)
2041 {
2042 filemap_write_and_wait(vma->vm_file->f_mapping);
2043 }
2044
2045 /*
2046 * Wait for writeback against this page to complete before allowing it
2047 * to be marked dirty again, and hence written back again, possibly
2048 * before the previous writepage completed.
2049 *
2050 * Block here, instead of in ->writepage(), so that the userspace fs
2051 * can only block processes actually operating on the filesystem.
2052 *
2053 * Otherwise unprivileged userspace fs would be able to block
2054 * unrelated:
2055 *
2056 * - page migration
2057 * - sync(2)
2058 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
2059 */
fuse_page_mkwrite(struct vm_fault * vmf)2060 static int fuse_page_mkwrite(struct vm_fault *vmf)
2061 {
2062 struct page *page = vmf->page;
2063 struct inode *inode = file_inode(vmf->vma->vm_file);
2064
2065 file_update_time(vmf->vma->vm_file);
2066 lock_page(page);
2067 if (page->mapping != inode->i_mapping) {
2068 unlock_page(page);
2069 return VM_FAULT_NOPAGE;
2070 }
2071
2072 fuse_wait_on_page_writeback(inode, page->index);
2073 return VM_FAULT_LOCKED;
2074 }
2075
2076 static const struct vm_operations_struct fuse_file_vm_ops = {
2077 .close = fuse_vma_close,
2078 .fault = filemap_fault,
2079 .map_pages = filemap_map_pages,
2080 .page_mkwrite = fuse_page_mkwrite,
2081 };
2082
fuse_file_mmap(struct file * file,struct vm_area_struct * vma)2083 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
2084 {
2085 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2086 fuse_link_write_file(file);
2087
2088 file_accessed(file);
2089 vma->vm_ops = &fuse_file_vm_ops;
2090 return 0;
2091 }
2092
fuse_direct_mmap(struct file * file,struct vm_area_struct * vma)2093 static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma)
2094 {
2095 /* Can't provide the coherency needed for MAP_SHARED */
2096 if (vma->vm_flags & VM_MAYSHARE)
2097 return -ENODEV;
2098
2099 invalidate_inode_pages2(file->f_mapping);
2100
2101 return generic_file_mmap(file, vma);
2102 }
2103
convert_fuse_file_lock(struct fuse_conn * fc,const struct fuse_file_lock * ffl,struct file_lock * fl)2104 static int convert_fuse_file_lock(struct fuse_conn *fc,
2105 const struct fuse_file_lock *ffl,
2106 struct file_lock *fl)
2107 {
2108 switch (ffl->type) {
2109 case F_UNLCK:
2110 break;
2111
2112 case F_RDLCK:
2113 case F_WRLCK:
2114 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
2115 ffl->end < ffl->start)
2116 return -EIO;
2117
2118 fl->fl_start = ffl->start;
2119 fl->fl_end = ffl->end;
2120
2121 /*
2122 * Convert pid into init's pid namespace. The locks API will
2123 * translate it into the caller's pid namespace.
2124 */
2125 rcu_read_lock();
2126 fl->fl_pid = pid_nr_ns(find_pid_ns(ffl->pid, fc->pid_ns), &init_pid_ns);
2127 rcu_read_unlock();
2128 break;
2129
2130 default:
2131 return -EIO;
2132 }
2133 fl->fl_type = ffl->type;
2134 return 0;
2135 }
2136
fuse_lk_fill(struct fuse_args * args,struct file * file,const struct file_lock * fl,int opcode,pid_t pid,int flock,struct fuse_lk_in * inarg)2137 static void fuse_lk_fill(struct fuse_args *args, struct file *file,
2138 const struct file_lock *fl, int opcode, pid_t pid,
2139 int flock, struct fuse_lk_in *inarg)
2140 {
2141 struct inode *inode = file_inode(file);
2142 struct fuse_conn *fc = get_fuse_conn(inode);
2143 struct fuse_file *ff = file->private_data;
2144
2145 memset(inarg, 0, sizeof(*inarg));
2146 inarg->fh = ff->fh;
2147 inarg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
2148 inarg->lk.start = fl->fl_start;
2149 inarg->lk.end = fl->fl_end;
2150 inarg->lk.type = fl->fl_type;
2151 inarg->lk.pid = pid;
2152 if (flock)
2153 inarg->lk_flags |= FUSE_LK_FLOCK;
2154 args->in.h.opcode = opcode;
2155 args->in.h.nodeid = get_node_id(inode);
2156 args->in.numargs = 1;
2157 args->in.args[0].size = sizeof(*inarg);
2158 args->in.args[0].value = inarg;
2159 }
2160
fuse_getlk(struct file * file,struct file_lock * fl)2161 static int fuse_getlk(struct file *file, struct file_lock *fl)
2162 {
2163 struct inode *inode = file_inode(file);
2164 struct fuse_conn *fc = get_fuse_conn(inode);
2165 FUSE_ARGS(args);
2166 struct fuse_lk_in inarg;
2167 struct fuse_lk_out outarg;
2168 int err;
2169
2170 fuse_lk_fill(&args, file, fl, FUSE_GETLK, 0, 0, &inarg);
2171 args.out.numargs = 1;
2172 args.out.args[0].size = sizeof(outarg);
2173 args.out.args[0].value = &outarg;
2174 err = fuse_simple_request(fc, &args);
2175 if (!err)
2176 err = convert_fuse_file_lock(fc, &outarg.lk, fl);
2177
2178 return err;
2179 }
2180
fuse_setlk(struct file * file,struct file_lock * fl,int flock)2181 static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
2182 {
2183 struct inode *inode = file_inode(file);
2184 struct fuse_conn *fc = get_fuse_conn(inode);
2185 FUSE_ARGS(args);
2186 struct fuse_lk_in inarg;
2187 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
2188 struct pid *pid = fl->fl_type != F_UNLCK ? task_tgid(current) : NULL;
2189 pid_t pid_nr = pid_nr_ns(pid, fc->pid_ns);
2190 int err;
2191
2192 if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
2193 /* NLM needs asynchronous locks, which we don't support yet */
2194 return -ENOLCK;
2195 }
2196
2197 /* Unlock on close is handled by the flush method */
2198 if ((fl->fl_flags & FL_CLOSE_POSIX) == FL_CLOSE_POSIX)
2199 return 0;
2200
2201 fuse_lk_fill(&args, file, fl, opcode, pid_nr, flock, &inarg);
2202 err = fuse_simple_request(fc, &args);
2203
2204 /* locking is restartable */
2205 if (err == -EINTR)
2206 err = -ERESTARTSYS;
2207
2208 return err;
2209 }
2210
fuse_file_lock(struct file * file,int cmd,struct file_lock * fl)2211 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
2212 {
2213 struct inode *inode = file_inode(file);
2214 struct fuse_conn *fc = get_fuse_conn(inode);
2215 int err;
2216
2217 if (cmd == F_CANCELLK) {
2218 err = 0;
2219 } else if (cmd == F_GETLK) {
2220 if (fc->no_lock) {
2221 posix_test_lock(file, fl);
2222 err = 0;
2223 } else
2224 err = fuse_getlk(file, fl);
2225 } else {
2226 if (fc->no_lock)
2227 err = posix_lock_file(file, fl, NULL);
2228 else
2229 err = fuse_setlk(file, fl, 0);
2230 }
2231 return err;
2232 }
2233
fuse_file_flock(struct file * file,int cmd,struct file_lock * fl)2234 static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
2235 {
2236 struct inode *inode = file_inode(file);
2237 struct fuse_conn *fc = get_fuse_conn(inode);
2238 int err;
2239
2240 if (fc->no_flock) {
2241 err = locks_lock_file_wait(file, fl);
2242 } else {
2243 struct fuse_file *ff = file->private_data;
2244
2245 /* emulate flock with POSIX locks */
2246 ff->flock = true;
2247 err = fuse_setlk(file, fl, 1);
2248 }
2249
2250 return err;
2251 }
2252
fuse_bmap(struct address_space * mapping,sector_t block)2253 static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
2254 {
2255 struct inode *inode = mapping->host;
2256 struct fuse_conn *fc = get_fuse_conn(inode);
2257 FUSE_ARGS(args);
2258 struct fuse_bmap_in inarg;
2259 struct fuse_bmap_out outarg;
2260 int err;
2261
2262 if (!inode->i_sb->s_bdev || fc->no_bmap)
2263 return 0;
2264
2265 memset(&inarg, 0, sizeof(inarg));
2266 inarg.block = block;
2267 inarg.blocksize = inode->i_sb->s_blocksize;
2268 args.in.h.opcode = FUSE_BMAP;
2269 args.in.h.nodeid = get_node_id(inode);
2270 args.in.numargs = 1;
2271 args.in.args[0].size = sizeof(inarg);
2272 args.in.args[0].value = &inarg;
2273 args.out.numargs = 1;
2274 args.out.args[0].size = sizeof(outarg);
2275 args.out.args[0].value = &outarg;
2276 err = fuse_simple_request(fc, &args);
2277 if (err == -ENOSYS)
2278 fc->no_bmap = 1;
2279
2280 return err ? 0 : outarg.block;
2281 }
2282
fuse_lseek(struct file * file,loff_t offset,int whence)2283 static loff_t fuse_lseek(struct file *file, loff_t offset, int whence)
2284 {
2285 struct inode *inode = file->f_mapping->host;
2286 struct fuse_conn *fc = get_fuse_conn(inode);
2287 struct fuse_file *ff = file->private_data;
2288 FUSE_ARGS(args);
2289 struct fuse_lseek_in inarg = {
2290 .fh = ff->fh,
2291 .offset = offset,
2292 .whence = whence
2293 };
2294 struct fuse_lseek_out outarg;
2295 int err;
2296
2297 if (fc->no_lseek)
2298 goto fallback;
2299
2300 args.in.h.opcode = FUSE_LSEEK;
2301 args.in.h.nodeid = ff->nodeid;
2302 args.in.numargs = 1;
2303 args.in.args[0].size = sizeof(inarg);
2304 args.in.args[0].value = &inarg;
2305 args.out.numargs = 1;
2306 args.out.args[0].size = sizeof(outarg);
2307 args.out.args[0].value = &outarg;
2308 err = fuse_simple_request(fc, &args);
2309 if (err) {
2310 if (err == -ENOSYS) {
2311 fc->no_lseek = 1;
2312 goto fallback;
2313 }
2314 return err;
2315 }
2316
2317 return vfs_setpos(file, outarg.offset, inode->i_sb->s_maxbytes);
2318
2319 fallback:
2320 err = fuse_update_attributes(inode, file);
2321 if (!err)
2322 return generic_file_llseek(file, offset, whence);
2323 else
2324 return err;
2325 }
2326
fuse_file_llseek(struct file * file,loff_t offset,int whence)2327 static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence)
2328 {
2329 loff_t retval;
2330 struct inode *inode = file_inode(file);
2331
2332 switch (whence) {
2333 case SEEK_SET:
2334 case SEEK_CUR:
2335 /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */
2336 retval = generic_file_llseek(file, offset, whence);
2337 break;
2338 case SEEK_END:
2339 inode_lock(inode);
2340 retval = fuse_update_attributes(inode, file);
2341 if (!retval)
2342 retval = generic_file_llseek(file, offset, whence);
2343 inode_unlock(inode);
2344 break;
2345 case SEEK_HOLE:
2346 case SEEK_DATA:
2347 inode_lock(inode);
2348 retval = fuse_lseek(file, offset, whence);
2349 inode_unlock(inode);
2350 break;
2351 default:
2352 retval = -EINVAL;
2353 }
2354
2355 return retval;
2356 }
2357
2358 /*
2359 * CUSE servers compiled on 32bit broke on 64bit kernels because the
2360 * ABI was defined to be 'struct iovec' which is different on 32bit
2361 * and 64bit. Fortunately we can determine which structure the server
2362 * used from the size of the reply.
2363 */
fuse_copy_ioctl_iovec_old(struct iovec * dst,void * src,size_t transferred,unsigned count,bool is_compat)2364 static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
2365 size_t transferred, unsigned count,
2366 bool is_compat)
2367 {
2368 #ifdef CONFIG_COMPAT
2369 if (count * sizeof(struct compat_iovec) == transferred) {
2370 struct compat_iovec *ciov = src;
2371 unsigned i;
2372
2373 /*
2374 * With this interface a 32bit server cannot support
2375 * non-compat (i.e. ones coming from 64bit apps) ioctl
2376 * requests
2377 */
2378 if (!is_compat)
2379 return -EINVAL;
2380
2381 for (i = 0; i < count; i++) {
2382 dst[i].iov_base = compat_ptr(ciov[i].iov_base);
2383 dst[i].iov_len = ciov[i].iov_len;
2384 }
2385 return 0;
2386 }
2387 #endif
2388
2389 if (count * sizeof(struct iovec) != transferred)
2390 return -EIO;
2391
2392 memcpy(dst, src, transferred);
2393 return 0;
2394 }
2395
2396 /* Make sure iov_length() won't overflow */
fuse_verify_ioctl_iov(struct iovec * iov,size_t count)2397 static int fuse_verify_ioctl_iov(struct iovec *iov, size_t count)
2398 {
2399 size_t n;
2400 u32 max = FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT;
2401
2402 for (n = 0; n < count; n++, iov++) {
2403 if (iov->iov_len > (size_t) max)
2404 return -ENOMEM;
2405 max -= iov->iov_len;
2406 }
2407 return 0;
2408 }
2409
fuse_copy_ioctl_iovec(struct fuse_conn * fc,struct iovec * dst,void * src,size_t transferred,unsigned count,bool is_compat)2410 static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
2411 void *src, size_t transferred, unsigned count,
2412 bool is_compat)
2413 {
2414 unsigned i;
2415 struct fuse_ioctl_iovec *fiov = src;
2416
2417 if (fc->minor < 16) {
2418 return fuse_copy_ioctl_iovec_old(dst, src, transferred,
2419 count, is_compat);
2420 }
2421
2422 if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
2423 return -EIO;
2424
2425 for (i = 0; i < count; i++) {
2426 /* Did the server supply an inappropriate value? */
2427 if (fiov[i].base != (unsigned long) fiov[i].base ||
2428 fiov[i].len != (unsigned long) fiov[i].len)
2429 return -EIO;
2430
2431 dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
2432 dst[i].iov_len = (size_t) fiov[i].len;
2433
2434 #ifdef CONFIG_COMPAT
2435 if (is_compat &&
2436 (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
2437 (compat_size_t) dst[i].iov_len != fiov[i].len))
2438 return -EIO;
2439 #endif
2440 }
2441
2442 return 0;
2443 }
2444
2445
2446 /*
2447 * For ioctls, there is no generic way to determine how much memory
2448 * needs to be read and/or written. Furthermore, ioctls are allowed
2449 * to dereference the passed pointer, so the parameter requires deep
2450 * copying but FUSE has no idea whatsoever about what to copy in or
2451 * out.
2452 *
2453 * This is solved by allowing FUSE server to retry ioctl with
2454 * necessary in/out iovecs. Let's assume the ioctl implementation
2455 * needs to read in the following structure.
2456 *
2457 * struct a {
2458 * char *buf;
2459 * size_t buflen;
2460 * }
2461 *
2462 * On the first callout to FUSE server, inarg->in_size and
2463 * inarg->out_size will be NULL; then, the server completes the ioctl
2464 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
2465 * the actual iov array to
2466 *
2467 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } }
2468 *
2469 * which tells FUSE to copy in the requested area and retry the ioctl.
2470 * On the second round, the server has access to the structure and
2471 * from that it can tell what to look for next, so on the invocation,
2472 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
2473 *
2474 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) },
2475 * { .iov_base = a.buf, .iov_len = a.buflen } }
2476 *
2477 * FUSE will copy both struct a and the pointed buffer from the
2478 * process doing the ioctl and retry ioctl with both struct a and the
2479 * buffer.
2480 *
2481 * This time, FUSE server has everything it needs and completes ioctl
2482 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
2483 *
2484 * Copying data out works the same way.
2485 *
2486 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
2487 * automatically initializes in and out iovs by decoding @cmd with
2488 * _IOC_* macros and the server is not allowed to request RETRY. This
2489 * limits ioctl data transfers to well-formed ioctls and is the forced
2490 * behavior for all FUSE servers.
2491 */
fuse_do_ioctl(struct file * file,unsigned int cmd,unsigned long arg,unsigned int flags)2492 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
2493 unsigned int flags)
2494 {
2495 struct fuse_file *ff = file->private_data;
2496 struct fuse_conn *fc = ff->fc;
2497 struct fuse_ioctl_in inarg = {
2498 .fh = ff->fh,
2499 .cmd = cmd,
2500 .arg = arg,
2501 .flags = flags
2502 };
2503 struct fuse_ioctl_out outarg;
2504 struct fuse_req *req = NULL;
2505 struct page **pages = NULL;
2506 struct iovec *iov_page = NULL;
2507 struct iovec *in_iov = NULL, *out_iov = NULL;
2508 unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages;
2509 size_t in_size, out_size, transferred, c;
2510 int err, i;
2511 struct iov_iter ii;
2512
2513 #if BITS_PER_LONG == 32
2514 inarg.flags |= FUSE_IOCTL_32BIT;
2515 #else
2516 if (flags & FUSE_IOCTL_COMPAT)
2517 inarg.flags |= FUSE_IOCTL_32BIT;
2518 #endif
2519
2520 /* assume all the iovs returned by client always fits in a page */
2521 BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
2522
2523 err = -ENOMEM;
2524 pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, sizeof(pages[0]), GFP_KERNEL);
2525 iov_page = (struct iovec *) __get_free_page(GFP_KERNEL);
2526 if (!pages || !iov_page)
2527 goto out;
2528
2529 /*
2530 * If restricted, initialize IO parameters as encoded in @cmd.
2531 * RETRY from server is not allowed.
2532 */
2533 if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
2534 struct iovec *iov = iov_page;
2535
2536 iov->iov_base = (void __user *)arg;
2537 iov->iov_len = _IOC_SIZE(cmd);
2538
2539 if (_IOC_DIR(cmd) & _IOC_WRITE) {
2540 in_iov = iov;
2541 in_iovs = 1;
2542 }
2543
2544 if (_IOC_DIR(cmd) & _IOC_READ) {
2545 out_iov = iov;
2546 out_iovs = 1;
2547 }
2548 }
2549
2550 retry:
2551 inarg.in_size = in_size = iov_length(in_iov, in_iovs);
2552 inarg.out_size = out_size = iov_length(out_iov, out_iovs);
2553
2554 /*
2555 * Out data can be used either for actual out data or iovs,
2556 * make sure there always is at least one page.
2557 */
2558 out_size = max_t(size_t, out_size, PAGE_SIZE);
2559 max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
2560
2561 /* make sure there are enough buffer pages and init request with them */
2562 err = -ENOMEM;
2563 if (max_pages > FUSE_MAX_PAGES_PER_REQ)
2564 goto out;
2565 while (num_pages < max_pages) {
2566 pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2567 if (!pages[num_pages])
2568 goto out;
2569 num_pages++;
2570 }
2571
2572 req = fuse_get_req(fc, num_pages);
2573 if (IS_ERR(req)) {
2574 err = PTR_ERR(req);
2575 req = NULL;
2576 goto out;
2577 }
2578 memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages);
2579 req->num_pages = num_pages;
2580 fuse_page_descs_length_init(req, 0, req->num_pages);
2581
2582 /* okay, let's send it to the client */
2583 req->in.h.opcode = FUSE_IOCTL;
2584 req->in.h.nodeid = ff->nodeid;
2585 req->in.numargs = 1;
2586 req->in.args[0].size = sizeof(inarg);
2587 req->in.args[0].value = &inarg;
2588 if (in_size) {
2589 req->in.numargs++;
2590 req->in.args[1].size = in_size;
2591 req->in.argpages = 1;
2592
2593 err = -EFAULT;
2594 iov_iter_init(&ii, WRITE, in_iov, in_iovs, in_size);
2595 for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= num_pages); i++) {
2596 c = copy_page_from_iter(pages[i], 0, PAGE_SIZE, &ii);
2597 if (c != PAGE_SIZE && iov_iter_count(&ii))
2598 goto out;
2599 }
2600 }
2601
2602 req->out.numargs = 2;
2603 req->out.args[0].size = sizeof(outarg);
2604 req->out.args[0].value = &outarg;
2605 req->out.args[1].size = out_size;
2606 req->out.argpages = 1;
2607 req->out.argvar = 1;
2608
2609 fuse_request_send(fc, req);
2610 err = req->out.h.error;
2611 transferred = req->out.args[1].size;
2612 fuse_put_request(fc, req);
2613 req = NULL;
2614 if (err)
2615 goto out;
2616
2617 /* did it ask for retry? */
2618 if (outarg.flags & FUSE_IOCTL_RETRY) {
2619 void *vaddr;
2620
2621 /* no retry if in restricted mode */
2622 err = -EIO;
2623 if (!(flags & FUSE_IOCTL_UNRESTRICTED))
2624 goto out;
2625
2626 in_iovs = outarg.in_iovs;
2627 out_iovs = outarg.out_iovs;
2628
2629 /*
2630 * Make sure things are in boundary, separate checks
2631 * are to protect against overflow.
2632 */
2633 err = -ENOMEM;
2634 if (in_iovs > FUSE_IOCTL_MAX_IOV ||
2635 out_iovs > FUSE_IOCTL_MAX_IOV ||
2636 in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
2637 goto out;
2638
2639 vaddr = kmap_atomic(pages[0]);
2640 err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr,
2641 transferred, in_iovs + out_iovs,
2642 (flags & FUSE_IOCTL_COMPAT) != 0);
2643 kunmap_atomic(vaddr);
2644 if (err)
2645 goto out;
2646
2647 in_iov = iov_page;
2648 out_iov = in_iov + in_iovs;
2649
2650 err = fuse_verify_ioctl_iov(in_iov, in_iovs);
2651 if (err)
2652 goto out;
2653
2654 err = fuse_verify_ioctl_iov(out_iov, out_iovs);
2655 if (err)
2656 goto out;
2657
2658 goto retry;
2659 }
2660
2661 err = -EIO;
2662 if (transferred > inarg.out_size)
2663 goto out;
2664
2665 err = -EFAULT;
2666 iov_iter_init(&ii, READ, out_iov, out_iovs, transferred);
2667 for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= num_pages); i++) {
2668 c = copy_page_to_iter(pages[i], 0, PAGE_SIZE, &ii);
2669 if (c != PAGE_SIZE && iov_iter_count(&ii))
2670 goto out;
2671 }
2672 err = 0;
2673 out:
2674 if (req)
2675 fuse_put_request(fc, req);
2676 free_page((unsigned long) iov_page);
2677 while (num_pages)
2678 __free_page(pages[--num_pages]);
2679 kfree(pages);
2680
2681 return err ? err : outarg.result;
2682 }
2683 EXPORT_SYMBOL_GPL(fuse_do_ioctl);
2684
fuse_ioctl_common(struct file * file,unsigned int cmd,unsigned long arg,unsigned int flags)2685 long fuse_ioctl_common(struct file *file, unsigned int cmd,
2686 unsigned long arg, unsigned int flags)
2687 {
2688 struct inode *inode = file_inode(file);
2689 struct fuse_conn *fc = get_fuse_conn(inode);
2690
2691 if (!fuse_allow_current_process(fc))
2692 return -EACCES;
2693
2694 if (is_bad_inode(inode))
2695 return -EIO;
2696
2697 return fuse_do_ioctl(file, cmd, arg, flags);
2698 }
2699
fuse_file_ioctl(struct file * file,unsigned int cmd,unsigned long arg)2700 static long fuse_file_ioctl(struct file *file, unsigned int cmd,
2701 unsigned long arg)
2702 {
2703 return fuse_ioctl_common(file, cmd, arg, 0);
2704 }
2705
fuse_file_compat_ioctl(struct file * file,unsigned int cmd,unsigned long arg)2706 static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
2707 unsigned long arg)
2708 {
2709 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
2710 }
2711
2712 /*
2713 * All files which have been polled are linked to RB tree
2714 * fuse_conn->polled_files which is indexed by kh. Walk the tree and
2715 * find the matching one.
2716 */
fuse_find_polled_node(struct fuse_conn * fc,u64 kh,struct rb_node ** parent_out)2717 static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh,
2718 struct rb_node **parent_out)
2719 {
2720 struct rb_node **link = &fc->polled_files.rb_node;
2721 struct rb_node *last = NULL;
2722
2723 while (*link) {
2724 struct fuse_file *ff;
2725
2726 last = *link;
2727 ff = rb_entry(last, struct fuse_file, polled_node);
2728
2729 if (kh < ff->kh)
2730 link = &last->rb_left;
2731 else if (kh > ff->kh)
2732 link = &last->rb_right;
2733 else
2734 return link;
2735 }
2736
2737 if (parent_out)
2738 *parent_out = last;
2739 return link;
2740 }
2741
2742 /*
2743 * The file is about to be polled. Make sure it's on the polled_files
2744 * RB tree. Note that files once added to the polled_files tree are
2745 * not removed before the file is released. This is because a file
2746 * polled once is likely to be polled again.
2747 */
fuse_register_polled_file(struct fuse_conn * fc,struct fuse_file * ff)2748 static void fuse_register_polled_file(struct fuse_conn *fc,
2749 struct fuse_file *ff)
2750 {
2751 spin_lock(&fc->lock);
2752 if (RB_EMPTY_NODE(&ff->polled_node)) {
2753 struct rb_node **link, *uninitialized_var(parent);
2754
2755 link = fuse_find_polled_node(fc, ff->kh, &parent);
2756 BUG_ON(*link);
2757 rb_link_node(&ff->polled_node, parent, link);
2758 rb_insert_color(&ff->polled_node, &fc->polled_files);
2759 }
2760 spin_unlock(&fc->lock);
2761 }
2762
fuse_file_poll(struct file * file,poll_table * wait)2763 unsigned fuse_file_poll(struct file *file, poll_table *wait)
2764 {
2765 struct fuse_file *ff = file->private_data;
2766 struct fuse_conn *fc = ff->fc;
2767 struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh };
2768 struct fuse_poll_out outarg;
2769 FUSE_ARGS(args);
2770 int err;
2771
2772 if (fc->no_poll)
2773 return DEFAULT_POLLMASK;
2774
2775 poll_wait(file, &ff->poll_wait, wait);
2776 inarg.events = (__u32)poll_requested_events(wait);
2777
2778 /*
2779 * Ask for notification iff there's someone waiting for it.
2780 * The client may ignore the flag and always notify.
2781 */
2782 if (waitqueue_active(&ff->poll_wait)) {
2783 inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY;
2784 fuse_register_polled_file(fc, ff);
2785 }
2786
2787 args.in.h.opcode = FUSE_POLL;
2788 args.in.h.nodeid = ff->nodeid;
2789 args.in.numargs = 1;
2790 args.in.args[0].size = sizeof(inarg);
2791 args.in.args[0].value = &inarg;
2792 args.out.numargs = 1;
2793 args.out.args[0].size = sizeof(outarg);
2794 args.out.args[0].value = &outarg;
2795 err = fuse_simple_request(fc, &args);
2796
2797 if (!err)
2798 return outarg.revents;
2799 if (err == -ENOSYS) {
2800 fc->no_poll = 1;
2801 return DEFAULT_POLLMASK;
2802 }
2803 return POLLERR;
2804 }
2805 EXPORT_SYMBOL_GPL(fuse_file_poll);
2806
2807 /*
2808 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
2809 * wakes up the poll waiters.
2810 */
fuse_notify_poll_wakeup(struct fuse_conn * fc,struct fuse_notify_poll_wakeup_out * outarg)2811 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
2812 struct fuse_notify_poll_wakeup_out *outarg)
2813 {
2814 u64 kh = outarg->kh;
2815 struct rb_node **link;
2816
2817 spin_lock(&fc->lock);
2818
2819 link = fuse_find_polled_node(fc, kh, NULL);
2820 if (*link) {
2821 struct fuse_file *ff;
2822
2823 ff = rb_entry(*link, struct fuse_file, polled_node);
2824 wake_up_interruptible_sync(&ff->poll_wait);
2825 }
2826
2827 spin_unlock(&fc->lock);
2828 return 0;
2829 }
2830
fuse_do_truncate(struct file * file)2831 static void fuse_do_truncate(struct file *file)
2832 {
2833 struct inode *inode = file->f_mapping->host;
2834 struct iattr attr;
2835
2836 attr.ia_valid = ATTR_SIZE;
2837 attr.ia_size = i_size_read(inode);
2838
2839 attr.ia_file = file;
2840 attr.ia_valid |= ATTR_FILE;
2841
2842 fuse_do_setattr(file_dentry(file), &attr, file);
2843 }
2844
fuse_round_up(loff_t off)2845 static inline loff_t fuse_round_up(loff_t off)
2846 {
2847 return round_up(off, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
2848 }
2849
2850 static ssize_t
fuse_direct_IO(struct kiocb * iocb,struct iov_iter * iter)2851 fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2852 {
2853 DECLARE_COMPLETION_ONSTACK(wait);
2854 ssize_t ret = 0;
2855 struct file *file = iocb->ki_filp;
2856 struct fuse_file *ff = file->private_data;
2857 bool async_dio = ff->fc->async_dio;
2858 loff_t pos = 0;
2859 struct inode *inode;
2860 loff_t i_size;
2861 size_t count = iov_iter_count(iter);
2862 loff_t offset = iocb->ki_pos;
2863 struct fuse_io_priv *io;
2864
2865 pos = offset;
2866 inode = file->f_mapping->host;
2867 i_size = i_size_read(inode);
2868
2869 if ((iov_iter_rw(iter) == READ) && (offset > i_size))
2870 return 0;
2871
2872 /* optimization for short read */
2873 if (async_dio && iov_iter_rw(iter) != WRITE && offset + count > i_size) {
2874 if (offset >= i_size)
2875 return 0;
2876 iov_iter_truncate(iter, fuse_round_up(i_size - offset));
2877 count = iov_iter_count(iter);
2878 }
2879
2880 io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL);
2881 if (!io)
2882 return -ENOMEM;
2883 spin_lock_init(&io->lock);
2884 kref_init(&io->refcnt);
2885 io->reqs = 1;
2886 io->bytes = -1;
2887 io->size = 0;
2888 io->offset = offset;
2889 io->write = (iov_iter_rw(iter) == WRITE);
2890 io->err = 0;
2891 /*
2892 * By default, we want to optimize all I/Os with async request
2893 * submission to the client filesystem if supported.
2894 */
2895 io->async = async_dio;
2896 io->iocb = iocb;
2897 io->blocking = is_sync_kiocb(iocb);
2898
2899 /*
2900 * We cannot asynchronously extend the size of a file.
2901 * In such case the aio will behave exactly like sync io.
2902 */
2903 if ((offset + count > i_size) && iov_iter_rw(iter) == WRITE)
2904 io->blocking = true;
2905
2906 if (io->async && io->blocking) {
2907 /*
2908 * Additional reference to keep io around after
2909 * calling fuse_aio_complete()
2910 */
2911 kref_get(&io->refcnt);
2912 io->done = &wait;
2913 }
2914
2915 if (iov_iter_rw(iter) == WRITE) {
2916 ret = fuse_direct_io(io, iter, &pos, FUSE_DIO_WRITE);
2917 fuse_invalidate_attr(inode);
2918 } else {
2919 ret = __fuse_direct_read(io, iter, &pos);
2920 }
2921
2922 if (io->async) {
2923 bool blocking = io->blocking;
2924
2925 fuse_aio_complete(io, ret < 0 ? ret : 0, -1);
2926
2927 /* we have a non-extending, async request, so return */
2928 if (!blocking)
2929 return -EIOCBQUEUED;
2930
2931 wait_for_completion(&wait);
2932 ret = fuse_get_res_by_io(io);
2933 }
2934
2935 kref_put(&io->refcnt, fuse_io_release);
2936
2937 if (iov_iter_rw(iter) == WRITE) {
2938 if (ret > 0)
2939 fuse_write_update_size(inode, pos);
2940 else if (ret < 0 && offset + count > i_size)
2941 fuse_do_truncate(file);
2942 }
2943
2944 return ret;
2945 }
2946
fuse_file_fallocate(struct file * file,int mode,loff_t offset,loff_t length)2947 static long fuse_file_fallocate(struct file *file, int mode, loff_t offset,
2948 loff_t length)
2949 {
2950 struct fuse_file *ff = file->private_data;
2951 struct inode *inode = file_inode(file);
2952 struct fuse_inode *fi = get_fuse_inode(inode);
2953 struct fuse_conn *fc = ff->fc;
2954 FUSE_ARGS(args);
2955 struct fuse_fallocate_in inarg = {
2956 .fh = ff->fh,
2957 .offset = offset,
2958 .length = length,
2959 .mode = mode
2960 };
2961 int err;
2962 bool lock_inode = !(mode & FALLOC_FL_KEEP_SIZE) ||
2963 (mode & FALLOC_FL_PUNCH_HOLE);
2964
2965 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2966 return -EOPNOTSUPP;
2967
2968 if (fc->no_fallocate)
2969 return -EOPNOTSUPP;
2970
2971 if (lock_inode) {
2972 inode_lock(inode);
2973 if (mode & FALLOC_FL_PUNCH_HOLE) {
2974 loff_t endbyte = offset + length - 1;
2975 err = filemap_write_and_wait_range(inode->i_mapping,
2976 offset, endbyte);
2977 if (err)
2978 goto out;
2979
2980 fuse_sync_writes(inode);
2981 }
2982 }
2983
2984 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
2985 offset + length > i_size_read(inode)) {
2986 err = inode_newsize_ok(inode, offset + length);
2987 if (err)
2988 goto out;
2989 }
2990
2991 if (!(mode & FALLOC_FL_KEEP_SIZE))
2992 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
2993
2994 args.in.h.opcode = FUSE_FALLOCATE;
2995 args.in.h.nodeid = ff->nodeid;
2996 args.in.numargs = 1;
2997 args.in.args[0].size = sizeof(inarg);
2998 args.in.args[0].value = &inarg;
2999 err = fuse_simple_request(fc, &args);
3000 if (err == -ENOSYS) {
3001 fc->no_fallocate = 1;
3002 err = -EOPNOTSUPP;
3003 }
3004 if (err)
3005 goto out;
3006
3007 /* we could have extended the file */
3008 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3009 bool changed = fuse_write_update_size(inode, offset + length);
3010
3011 if (changed && fc->writeback_cache)
3012 file_update_time(file);
3013 }
3014
3015 if (mode & FALLOC_FL_PUNCH_HOLE)
3016 truncate_pagecache_range(inode, offset, offset + length - 1);
3017
3018 fuse_invalidate_attr(inode);
3019
3020 out:
3021 if (!(mode & FALLOC_FL_KEEP_SIZE))
3022 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
3023
3024 if (lock_inode)
3025 inode_unlock(inode);
3026
3027 return err;
3028 }
3029
3030 static const struct file_operations fuse_file_operations = {
3031 .llseek = fuse_file_llseek,
3032 .read_iter = fuse_file_read_iter,
3033 .write_iter = fuse_file_write_iter,
3034 .mmap = fuse_file_mmap,
3035 .open = fuse_open,
3036 .flush = fuse_flush,
3037 .release = fuse_release,
3038 .fsync = fuse_fsync,
3039 .lock = fuse_file_lock,
3040 .flock = fuse_file_flock,
3041 .splice_read = generic_file_splice_read,
3042 .unlocked_ioctl = fuse_file_ioctl,
3043 .compat_ioctl = fuse_file_compat_ioctl,
3044 .poll = fuse_file_poll,
3045 .fallocate = fuse_file_fallocate,
3046 };
3047
3048 static const struct file_operations fuse_direct_io_file_operations = {
3049 .llseek = fuse_file_llseek,
3050 .read_iter = fuse_direct_read_iter,
3051 .write_iter = fuse_direct_write_iter,
3052 .mmap = fuse_direct_mmap,
3053 .open = fuse_open,
3054 .flush = fuse_flush,
3055 .release = fuse_release,
3056 .fsync = fuse_fsync,
3057 .lock = fuse_file_lock,
3058 .flock = fuse_file_flock,
3059 .unlocked_ioctl = fuse_file_ioctl,
3060 .compat_ioctl = fuse_file_compat_ioctl,
3061 .poll = fuse_file_poll,
3062 .fallocate = fuse_file_fallocate,
3063 /* no splice_read */
3064 };
3065
3066 static const struct address_space_operations fuse_file_aops = {
3067 .readpage = fuse_readpage,
3068 .writepage = fuse_writepage,
3069 .writepages = fuse_writepages,
3070 .launder_page = fuse_launder_page,
3071 .readpages = fuse_readpages,
3072 .set_page_dirty = __set_page_dirty_nobuffers,
3073 .bmap = fuse_bmap,
3074 .direct_IO = fuse_direct_IO,
3075 .write_begin = fuse_write_begin,
3076 .write_end = fuse_write_end,
3077 };
3078
fuse_init_file_inode(struct inode * inode)3079 void fuse_init_file_inode(struct inode *inode)
3080 {
3081 inode->i_fop = &fuse_file_operations;
3082 inode->i_data.a_ops = &fuse_file_aops;
3083 }
3084