Linux 2.6.31.6
[linux/fpc-iii.git] / fs / fuse / file.c
blob899870ada1383672e1e58b770a966699e5f14fc2
1 /*
2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
5 This program can be distributed under the terms of the GNU GPL.
6 See the file COPYING.
7 */
9 #include "fuse_i.h"
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>
17 static const struct file_operations fuse_direct_io_file_operations;
19 static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
20 int opcode, struct fuse_open_out *outargp)
22 struct fuse_open_in inarg;
23 struct fuse_req *req;
24 int err;
26 req = fuse_get_req(fc);
27 if (IS_ERR(req))
28 return PTR_ERR(req);
30 memset(&inarg, 0, sizeof(inarg));
31 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY);
32 if (!fc->atomic_o_trunc)
33 inarg.flags &= ~O_TRUNC;
34 req->in.h.opcode = opcode;
35 req->in.h.nodeid = nodeid;
36 req->in.numargs = 1;
37 req->in.args[0].size = sizeof(inarg);
38 req->in.args[0].value = &inarg;
39 req->out.numargs = 1;
40 req->out.args[0].size = sizeof(*outargp);
41 req->out.args[0].value = outargp;
42 fuse_request_send(fc, req);
43 err = req->out.h.error;
44 fuse_put_request(fc, req);
46 return err;
49 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc)
51 struct fuse_file *ff;
53 ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
54 if (unlikely(!ff))
55 return NULL;
57 ff->fc = fc;
58 ff->reserved_req = fuse_request_alloc();
59 if (unlikely(!ff->reserved_req)) {
60 kfree(ff);
61 return NULL;
64 INIT_LIST_HEAD(&ff->write_entry);
65 atomic_set(&ff->count, 0);
66 RB_CLEAR_NODE(&ff->polled_node);
67 init_waitqueue_head(&ff->poll_wait);
69 spin_lock(&fc->lock);
70 ff->kh = ++fc->khctr;
71 spin_unlock(&fc->lock);
73 return ff;
76 void fuse_file_free(struct fuse_file *ff)
78 fuse_request_free(ff->reserved_req);
79 kfree(ff);
82 struct fuse_file *fuse_file_get(struct fuse_file *ff)
84 atomic_inc(&ff->count);
85 return ff;
88 static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
90 path_put(&req->misc.release.path);
93 static void fuse_file_put(struct fuse_file *ff)
95 if (atomic_dec_and_test(&ff->count)) {
96 struct fuse_req *req = ff->reserved_req;
98 req->end = fuse_release_end;
99 fuse_request_send_background(ff->fc, req);
100 kfree(ff);
104 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
105 bool isdir)
107 struct fuse_open_out outarg;
108 struct fuse_file *ff;
109 int err;
110 int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
112 ff = fuse_file_alloc(fc);
113 if (!ff)
114 return -ENOMEM;
116 err = fuse_send_open(fc, nodeid, file, opcode, &outarg);
117 if (err) {
118 fuse_file_free(ff);
119 return err;
122 if (isdir)
123 outarg.open_flags &= ~FOPEN_DIRECT_IO;
125 ff->fh = outarg.fh;
126 ff->nodeid = nodeid;
127 ff->open_flags = outarg.open_flags;
128 file->private_data = fuse_file_get(ff);
130 return 0;
132 EXPORT_SYMBOL_GPL(fuse_do_open);
134 void fuse_finish_open(struct inode *inode, struct file *file)
136 struct fuse_file *ff = file->private_data;
138 if (ff->open_flags & FOPEN_DIRECT_IO)
139 file->f_op = &fuse_direct_io_file_operations;
140 if (!(ff->open_flags & FOPEN_KEEP_CACHE))
141 invalidate_inode_pages2(inode->i_mapping);
142 if (ff->open_flags & FOPEN_NONSEEKABLE)
143 nonseekable_open(inode, file);
146 int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
148 struct fuse_conn *fc = get_fuse_conn(inode);
149 int err;
151 /* VFS checks this, but only _after_ ->open() */
152 if (file->f_flags & O_DIRECT)
153 return -EINVAL;
155 err = generic_file_open(inode, file);
156 if (err)
157 return err;
159 err = fuse_do_open(fc, get_node_id(inode), file, isdir);
160 if (err)
161 return err;
163 fuse_finish_open(inode, file);
165 return 0;
168 static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode)
170 struct fuse_conn *fc = ff->fc;
171 struct fuse_req *req = ff->reserved_req;
172 struct fuse_release_in *inarg = &req->misc.release.in;
174 spin_lock(&fc->lock);
175 list_del(&ff->write_entry);
176 if (!RB_EMPTY_NODE(&ff->polled_node))
177 rb_erase(&ff->polled_node, &fc->polled_files);
178 spin_unlock(&fc->lock);
180 wake_up_interruptible_sync(&ff->poll_wait);
182 inarg->fh = ff->fh;
183 inarg->flags = flags;
184 req->in.h.opcode = opcode;
185 req->in.h.nodeid = ff->nodeid;
186 req->in.numargs = 1;
187 req->in.args[0].size = sizeof(struct fuse_release_in);
188 req->in.args[0].value = inarg;
191 void fuse_release_common(struct file *file, int opcode)
193 struct fuse_file *ff;
194 struct fuse_req *req;
196 ff = file->private_data;
197 if (unlikely(!ff))
198 return;
200 req = ff->reserved_req;
201 fuse_prepare_release(ff, file->f_flags, opcode);
203 /* Hold vfsmount and dentry until release is finished */
204 path_get(&file->f_path);
205 req->misc.release.path = file->f_path;
208 * Normally this will send the RELEASE request, however if
209 * some asynchronous READ or WRITE requests are outstanding,
210 * the sending will be delayed.
212 fuse_file_put(ff);
215 static int fuse_open(struct inode *inode, struct file *file)
217 return fuse_open_common(inode, file, false);
220 static int fuse_release(struct inode *inode, struct file *file)
222 fuse_release_common(file, FUSE_RELEASE);
224 /* return value is ignored by VFS */
225 return 0;
228 void fuse_sync_release(struct fuse_file *ff, int flags)
230 WARN_ON(atomic_read(&ff->count) > 1);
231 fuse_prepare_release(ff, flags, FUSE_RELEASE);
232 ff->reserved_req->force = 1;
233 fuse_request_send(ff->fc, ff->reserved_req);
234 fuse_put_request(ff->fc, ff->reserved_req);
235 kfree(ff);
237 EXPORT_SYMBOL_GPL(fuse_sync_release);
240 * Scramble the ID space with XTEA, so that the value of the files_struct
241 * pointer is not exposed to userspace.
243 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
245 u32 *k = fc->scramble_key;
246 u64 v = (unsigned long) id;
247 u32 v0 = v;
248 u32 v1 = v >> 32;
249 u32 sum = 0;
250 int i;
252 for (i = 0; i < 32; i++) {
253 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
254 sum += 0x9E3779B9;
255 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
258 return (u64) v0 + ((u64) v1 << 32);
262 * Check if page is under writeback
264 * This is currently done by walking the list of writepage requests
265 * for the inode, which can be pretty inefficient.
267 static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
269 struct fuse_conn *fc = get_fuse_conn(inode);
270 struct fuse_inode *fi = get_fuse_inode(inode);
271 struct fuse_req *req;
272 bool found = false;
274 spin_lock(&fc->lock);
275 list_for_each_entry(req, &fi->writepages, writepages_entry) {
276 pgoff_t curr_index;
278 BUG_ON(req->inode != inode);
279 curr_index = req->misc.write.in.offset >> PAGE_CACHE_SHIFT;
280 if (curr_index == index) {
281 found = true;
282 break;
285 spin_unlock(&fc->lock);
287 return found;
291 * Wait for page writeback to be completed.
293 * Since fuse doesn't rely on the VM writeback tracking, this has to
294 * use some other means.
296 static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
298 struct fuse_inode *fi = get_fuse_inode(inode);
300 wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
301 return 0;
304 static int fuse_flush(struct file *file, fl_owner_t id)
306 struct inode *inode = file->f_path.dentry->d_inode;
307 struct fuse_conn *fc = get_fuse_conn(inode);
308 struct fuse_file *ff = file->private_data;
309 struct fuse_req *req;
310 struct fuse_flush_in inarg;
311 int err;
313 if (is_bad_inode(inode))
314 return -EIO;
316 if (fc->no_flush)
317 return 0;
319 req = fuse_get_req_nofail(fc, file);
320 memset(&inarg, 0, sizeof(inarg));
321 inarg.fh = ff->fh;
322 inarg.lock_owner = fuse_lock_owner_id(fc, id);
323 req->in.h.opcode = FUSE_FLUSH;
324 req->in.h.nodeid = get_node_id(inode);
325 req->in.numargs = 1;
326 req->in.args[0].size = sizeof(inarg);
327 req->in.args[0].value = &inarg;
328 req->force = 1;
329 fuse_request_send(fc, req);
330 err = req->out.h.error;
331 fuse_put_request(fc, req);
332 if (err == -ENOSYS) {
333 fc->no_flush = 1;
334 err = 0;
336 return err;
340 * Wait for all pending writepages on the inode to finish.
342 * This is currently done by blocking further writes with FUSE_NOWRITE
343 * and waiting for all sent writes to complete.
345 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
346 * could conflict with truncation.
348 static void fuse_sync_writes(struct inode *inode)
350 fuse_set_nowrite(inode);
351 fuse_release_nowrite(inode);
354 int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
355 int isdir)
357 struct inode *inode = de->d_inode;
358 struct fuse_conn *fc = get_fuse_conn(inode);
359 struct fuse_file *ff = file->private_data;
360 struct fuse_req *req;
361 struct fuse_fsync_in inarg;
362 int err;
364 if (is_bad_inode(inode))
365 return -EIO;
367 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
368 return 0;
371 * Start writeback against all dirty pages of the inode, then
372 * wait for all outstanding writes, before sending the FSYNC
373 * request.
375 err = write_inode_now(inode, 0);
376 if (err)
377 return err;
379 fuse_sync_writes(inode);
381 req = fuse_get_req(fc);
382 if (IS_ERR(req))
383 return PTR_ERR(req);
385 memset(&inarg, 0, sizeof(inarg));
386 inarg.fh = ff->fh;
387 inarg.fsync_flags = datasync ? 1 : 0;
388 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
389 req->in.h.nodeid = get_node_id(inode);
390 req->in.numargs = 1;
391 req->in.args[0].size = sizeof(inarg);
392 req->in.args[0].value = &inarg;
393 fuse_request_send(fc, req);
394 err = req->out.h.error;
395 fuse_put_request(fc, req);
396 if (err == -ENOSYS) {
397 if (isdir)
398 fc->no_fsyncdir = 1;
399 else
400 fc->no_fsync = 1;
401 err = 0;
403 return err;
406 static int fuse_fsync(struct file *file, struct dentry *de, int datasync)
408 return fuse_fsync_common(file, de, datasync, 0);
411 void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos,
412 size_t count, int opcode)
414 struct fuse_read_in *inarg = &req->misc.read.in;
415 struct fuse_file *ff = file->private_data;
417 inarg->fh = ff->fh;
418 inarg->offset = pos;
419 inarg->size = count;
420 inarg->flags = file->f_flags;
421 req->in.h.opcode = opcode;
422 req->in.h.nodeid = ff->nodeid;
423 req->in.numargs = 1;
424 req->in.args[0].size = sizeof(struct fuse_read_in);
425 req->in.args[0].value = inarg;
426 req->out.argvar = 1;
427 req->out.numargs = 1;
428 req->out.args[0].size = count;
431 static size_t fuse_send_read(struct fuse_req *req, struct file *file,
432 loff_t pos, size_t count, fl_owner_t owner)
434 struct fuse_file *ff = file->private_data;
435 struct fuse_conn *fc = ff->fc;
437 fuse_read_fill(req, file, pos, count, FUSE_READ);
438 if (owner != NULL) {
439 struct fuse_read_in *inarg = &req->misc.read.in;
441 inarg->read_flags |= FUSE_READ_LOCKOWNER;
442 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
444 fuse_request_send(fc, req);
445 return req->out.args[0].size;
448 static void fuse_read_update_size(struct inode *inode, loff_t size,
449 u64 attr_ver)
451 struct fuse_conn *fc = get_fuse_conn(inode);
452 struct fuse_inode *fi = get_fuse_inode(inode);
454 spin_lock(&fc->lock);
455 if (attr_ver == fi->attr_version && size < inode->i_size) {
456 fi->attr_version = ++fc->attr_version;
457 i_size_write(inode, size);
459 spin_unlock(&fc->lock);
462 static int fuse_readpage(struct file *file, struct page *page)
464 struct inode *inode = page->mapping->host;
465 struct fuse_conn *fc = get_fuse_conn(inode);
466 struct fuse_req *req;
467 size_t num_read;
468 loff_t pos = page_offset(page);
469 size_t count = PAGE_CACHE_SIZE;
470 u64 attr_ver;
471 int err;
473 err = -EIO;
474 if (is_bad_inode(inode))
475 goto out;
478 * Page writeback can extend beyond the liftime of the
479 * page-cache page, so make sure we read a properly synced
480 * page.
482 fuse_wait_on_page_writeback(inode, page->index);
484 req = fuse_get_req(fc);
485 err = PTR_ERR(req);
486 if (IS_ERR(req))
487 goto out;
489 attr_ver = fuse_get_attr_version(fc);
491 req->out.page_zeroing = 1;
492 req->out.argpages = 1;
493 req->num_pages = 1;
494 req->pages[0] = page;
495 num_read = fuse_send_read(req, file, pos, count, NULL);
496 err = req->out.h.error;
497 fuse_put_request(fc, req);
499 if (!err) {
501 * Short read means EOF. If file size is larger, truncate it
503 if (num_read < count)
504 fuse_read_update_size(inode, pos + num_read, attr_ver);
506 SetPageUptodate(page);
509 fuse_invalidate_attr(inode); /* atime changed */
510 out:
511 unlock_page(page);
512 return err;
515 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
517 int i;
518 size_t count = req->misc.read.in.size;
519 size_t num_read = req->out.args[0].size;
520 struct inode *inode = req->pages[0]->mapping->host;
523 * Short read means EOF. If file size is larger, truncate it
525 if (!req->out.h.error && num_read < count) {
526 loff_t pos = page_offset(req->pages[0]) + num_read;
527 fuse_read_update_size(inode, pos, req->misc.read.attr_ver);
530 fuse_invalidate_attr(inode); /* atime changed */
532 for (i = 0; i < req->num_pages; i++) {
533 struct page *page = req->pages[i];
534 if (!req->out.h.error)
535 SetPageUptodate(page);
536 else
537 SetPageError(page);
538 unlock_page(page);
540 if (req->ff)
541 fuse_file_put(req->ff);
544 static void fuse_send_readpages(struct fuse_req *req, struct file *file)
546 struct fuse_file *ff = file->private_data;
547 struct fuse_conn *fc = ff->fc;
548 loff_t pos = page_offset(req->pages[0]);
549 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
551 req->out.argpages = 1;
552 req->out.page_zeroing = 1;
553 fuse_read_fill(req, file, pos, count, FUSE_READ);
554 req->misc.read.attr_ver = fuse_get_attr_version(fc);
555 if (fc->async_read) {
556 req->ff = fuse_file_get(ff);
557 req->end = fuse_readpages_end;
558 fuse_request_send_background(fc, req);
559 } else {
560 fuse_request_send(fc, req);
561 fuse_readpages_end(fc, req);
562 fuse_put_request(fc, req);
566 struct fuse_fill_data {
567 struct fuse_req *req;
568 struct file *file;
569 struct inode *inode;
572 static int fuse_readpages_fill(void *_data, struct page *page)
574 struct fuse_fill_data *data = _data;
575 struct fuse_req *req = data->req;
576 struct inode *inode = data->inode;
577 struct fuse_conn *fc = get_fuse_conn(inode);
579 fuse_wait_on_page_writeback(inode, page->index);
581 if (req->num_pages &&
582 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
583 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
584 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
585 fuse_send_readpages(req, data->file);
586 data->req = req = fuse_get_req(fc);
587 if (IS_ERR(req)) {
588 unlock_page(page);
589 return PTR_ERR(req);
592 req->pages[req->num_pages] = page;
593 req->num_pages++;
594 return 0;
597 static int fuse_readpages(struct file *file, struct address_space *mapping,
598 struct list_head *pages, unsigned nr_pages)
600 struct inode *inode = mapping->host;
601 struct fuse_conn *fc = get_fuse_conn(inode);
602 struct fuse_fill_data data;
603 int err;
605 err = -EIO;
606 if (is_bad_inode(inode))
607 goto out;
609 data.file = file;
610 data.inode = inode;
611 data.req = fuse_get_req(fc);
612 err = PTR_ERR(data.req);
613 if (IS_ERR(data.req))
614 goto out;
616 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
617 if (!err) {
618 if (data.req->num_pages)
619 fuse_send_readpages(data.req, file);
620 else
621 fuse_put_request(fc, data.req);
623 out:
624 return err;
627 static ssize_t fuse_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
628 unsigned long nr_segs, loff_t pos)
630 struct inode *inode = iocb->ki_filp->f_mapping->host;
632 if (pos + iov_length(iov, nr_segs) > i_size_read(inode)) {
633 int err;
635 * If trying to read past EOF, make sure the i_size
636 * attribute is up-to-date.
638 err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL);
639 if (err)
640 return err;
643 return generic_file_aio_read(iocb, iov, nr_segs, pos);
646 static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff,
647 loff_t pos, size_t count)
649 struct fuse_write_in *inarg = &req->misc.write.in;
650 struct fuse_write_out *outarg = &req->misc.write.out;
652 inarg->fh = ff->fh;
653 inarg->offset = pos;
654 inarg->size = count;
655 req->in.h.opcode = FUSE_WRITE;
656 req->in.h.nodeid = ff->nodeid;
657 req->in.numargs = 2;
658 if (ff->fc->minor < 9)
659 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
660 else
661 req->in.args[0].size = sizeof(struct fuse_write_in);
662 req->in.args[0].value = inarg;
663 req->in.args[1].size = count;
664 req->out.numargs = 1;
665 req->out.args[0].size = sizeof(struct fuse_write_out);
666 req->out.args[0].value = outarg;
669 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
670 loff_t pos, size_t count, fl_owner_t owner)
672 struct fuse_file *ff = file->private_data;
673 struct fuse_conn *fc = ff->fc;
674 struct fuse_write_in *inarg = &req->misc.write.in;
676 fuse_write_fill(req, ff, pos, count);
677 inarg->flags = file->f_flags;
678 if (owner != NULL) {
679 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
680 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
682 fuse_request_send(fc, req);
683 return req->misc.write.out.size;
686 static int fuse_write_begin(struct file *file, struct address_space *mapping,
687 loff_t pos, unsigned len, unsigned flags,
688 struct page **pagep, void **fsdata)
690 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
692 *pagep = grab_cache_page_write_begin(mapping, index, flags);
693 if (!*pagep)
694 return -ENOMEM;
695 return 0;
698 static void fuse_write_update_size(struct inode *inode, loff_t pos)
700 struct fuse_conn *fc = get_fuse_conn(inode);
701 struct fuse_inode *fi = get_fuse_inode(inode);
703 spin_lock(&fc->lock);
704 fi->attr_version = ++fc->attr_version;
705 if (pos > inode->i_size)
706 i_size_write(inode, pos);
707 spin_unlock(&fc->lock);
710 static int fuse_buffered_write(struct file *file, struct inode *inode,
711 loff_t pos, unsigned count, struct page *page)
713 int err;
714 size_t nres;
715 struct fuse_conn *fc = get_fuse_conn(inode);
716 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
717 struct fuse_req *req;
719 if (is_bad_inode(inode))
720 return -EIO;
723 * Make sure writepages on the same page are not mixed up with
724 * plain writes.
726 fuse_wait_on_page_writeback(inode, page->index);
728 req = fuse_get_req(fc);
729 if (IS_ERR(req))
730 return PTR_ERR(req);
732 req->in.argpages = 1;
733 req->num_pages = 1;
734 req->pages[0] = page;
735 req->page_offset = offset;
736 nres = fuse_send_write(req, file, pos, count, NULL);
737 err = req->out.h.error;
738 fuse_put_request(fc, req);
739 if (!err && !nres)
740 err = -EIO;
741 if (!err) {
742 pos += nres;
743 fuse_write_update_size(inode, pos);
744 if (count == PAGE_CACHE_SIZE)
745 SetPageUptodate(page);
747 fuse_invalidate_attr(inode);
748 return err ? err : nres;
751 static int fuse_write_end(struct file *file, struct address_space *mapping,
752 loff_t pos, unsigned len, unsigned copied,
753 struct page *page, void *fsdata)
755 struct inode *inode = mapping->host;
756 int res = 0;
758 if (copied)
759 res = fuse_buffered_write(file, inode, pos, copied, page);
761 unlock_page(page);
762 page_cache_release(page);
763 return res;
766 static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file,
767 struct inode *inode, loff_t pos,
768 size_t count)
770 size_t res;
771 unsigned offset;
772 unsigned i;
774 for (i = 0; i < req->num_pages; i++)
775 fuse_wait_on_page_writeback(inode, req->pages[i]->index);
777 res = fuse_send_write(req, file, pos, count, NULL);
779 offset = req->page_offset;
780 count = res;
781 for (i = 0; i < req->num_pages; i++) {
782 struct page *page = req->pages[i];
784 if (!req->out.h.error && !offset && count >= PAGE_CACHE_SIZE)
785 SetPageUptodate(page);
787 if (count > PAGE_CACHE_SIZE - offset)
788 count -= PAGE_CACHE_SIZE - offset;
789 else
790 count = 0;
791 offset = 0;
793 unlock_page(page);
794 page_cache_release(page);
797 return res;
800 static ssize_t fuse_fill_write_pages(struct fuse_req *req,
801 struct address_space *mapping,
802 struct iov_iter *ii, loff_t pos)
804 struct fuse_conn *fc = get_fuse_conn(mapping->host);
805 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
806 size_t count = 0;
807 int err;
809 req->in.argpages = 1;
810 req->page_offset = offset;
812 do {
813 size_t tmp;
814 struct page *page;
815 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
816 size_t bytes = min_t(size_t, PAGE_CACHE_SIZE - offset,
817 iov_iter_count(ii));
819 bytes = min_t(size_t, bytes, fc->max_write - count);
821 again:
822 err = -EFAULT;
823 if (iov_iter_fault_in_readable(ii, bytes))
824 break;
826 err = -ENOMEM;
827 page = grab_cache_page_write_begin(mapping, index, 0);
828 if (!page)
829 break;
831 pagefault_disable();
832 tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes);
833 pagefault_enable();
834 flush_dcache_page(page);
836 if (!tmp) {
837 unlock_page(page);
838 page_cache_release(page);
839 bytes = min(bytes, iov_iter_single_seg_count(ii));
840 goto again;
843 err = 0;
844 req->pages[req->num_pages] = page;
845 req->num_pages++;
847 iov_iter_advance(ii, tmp);
848 count += tmp;
849 pos += tmp;
850 offset += tmp;
851 if (offset == PAGE_CACHE_SIZE)
852 offset = 0;
854 if (!fc->big_writes)
855 break;
856 } while (iov_iter_count(ii) && count < fc->max_write &&
857 req->num_pages < FUSE_MAX_PAGES_PER_REQ && offset == 0);
859 return count > 0 ? count : err;
862 static ssize_t fuse_perform_write(struct file *file,
863 struct address_space *mapping,
864 struct iov_iter *ii, loff_t pos)
866 struct inode *inode = mapping->host;
867 struct fuse_conn *fc = get_fuse_conn(inode);
868 int err = 0;
869 ssize_t res = 0;
871 if (is_bad_inode(inode))
872 return -EIO;
874 do {
875 struct fuse_req *req;
876 ssize_t count;
878 req = fuse_get_req(fc);
879 if (IS_ERR(req)) {
880 err = PTR_ERR(req);
881 break;
884 count = fuse_fill_write_pages(req, mapping, ii, pos);
885 if (count <= 0) {
886 err = count;
887 } else {
888 size_t num_written;
890 num_written = fuse_send_write_pages(req, file, inode,
891 pos, count);
892 err = req->out.h.error;
893 if (!err) {
894 res += num_written;
895 pos += num_written;
897 /* break out of the loop on short write */
898 if (num_written != count)
899 err = -EIO;
902 fuse_put_request(fc, req);
903 } while (!err && iov_iter_count(ii));
905 if (res > 0)
906 fuse_write_update_size(inode, pos);
908 fuse_invalidate_attr(inode);
910 return res > 0 ? res : err;
913 static ssize_t fuse_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
914 unsigned long nr_segs, loff_t pos)
916 struct file *file = iocb->ki_filp;
917 struct address_space *mapping = file->f_mapping;
918 size_t count = 0;
919 ssize_t written = 0;
920 struct inode *inode = mapping->host;
921 ssize_t err;
922 struct iov_iter i;
924 WARN_ON(iocb->ki_pos != pos);
926 err = generic_segment_checks(iov, &nr_segs, &count, VERIFY_READ);
927 if (err)
928 return err;
930 mutex_lock(&inode->i_mutex);
931 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
933 /* We can write back this queue in page reclaim */
934 current->backing_dev_info = mapping->backing_dev_info;
936 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
937 if (err)
938 goto out;
940 if (count == 0)
941 goto out;
943 err = file_remove_suid(file);
944 if (err)
945 goto out;
947 file_update_time(file);
949 iov_iter_init(&i, iov, nr_segs, count, 0);
950 written = fuse_perform_write(file, mapping, &i, pos);
951 if (written >= 0)
952 iocb->ki_pos = pos + written;
954 out:
955 current->backing_dev_info = NULL;
956 mutex_unlock(&inode->i_mutex);
958 return written ? written : err;
961 static void fuse_release_user_pages(struct fuse_req *req, int write)
963 unsigned i;
965 for (i = 0; i < req->num_pages; i++) {
966 struct page *page = req->pages[i];
967 if (write)
968 set_page_dirty_lock(page);
969 put_page(page);
973 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
974 size_t *nbytesp, int write)
976 size_t nbytes = *nbytesp;
977 unsigned long user_addr = (unsigned long) buf;
978 unsigned offset = user_addr & ~PAGE_MASK;
979 int npages;
981 /* Special case for kernel I/O: can copy directly into the buffer */
982 if (segment_eq(get_fs(), KERNEL_DS)) {
983 if (write)
984 req->in.args[1].value = (void *) user_addr;
985 else
986 req->out.args[0].value = (void *) user_addr;
988 return 0;
991 nbytes = min_t(size_t, nbytes, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
992 npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
993 npages = clamp(npages, 1, FUSE_MAX_PAGES_PER_REQ);
994 down_read(&current->mm->mmap_sem);
995 npages = get_user_pages(current, current->mm, user_addr, npages, !write,
996 0, req->pages, NULL);
997 up_read(&current->mm->mmap_sem);
998 if (npages < 0)
999 return npages;
1001 req->num_pages = npages;
1002 req->page_offset = offset;
1004 if (write)
1005 req->in.argpages = 1;
1006 else
1007 req->out.argpages = 1;
1009 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset;
1010 *nbytesp = min(*nbytesp, nbytes);
1012 return 0;
1015 ssize_t fuse_direct_io(struct file *file, const char __user *buf,
1016 size_t count, loff_t *ppos, int write)
1018 struct fuse_file *ff = file->private_data;
1019 struct fuse_conn *fc = ff->fc;
1020 size_t nmax = write ? fc->max_write : fc->max_read;
1021 loff_t pos = *ppos;
1022 ssize_t res = 0;
1023 struct fuse_req *req;
1025 req = fuse_get_req(fc);
1026 if (IS_ERR(req))
1027 return PTR_ERR(req);
1029 while (count) {
1030 size_t nres;
1031 fl_owner_t owner = current->files;
1032 size_t nbytes = min(count, nmax);
1033 int err = fuse_get_user_pages(req, buf, &nbytes, write);
1034 if (err) {
1035 res = err;
1036 break;
1039 if (write)
1040 nres = fuse_send_write(req, file, pos, nbytes, owner);
1041 else
1042 nres = fuse_send_read(req, file, pos, nbytes, owner);
1044 fuse_release_user_pages(req, !write);
1045 if (req->out.h.error) {
1046 if (!res)
1047 res = req->out.h.error;
1048 break;
1049 } else if (nres > nbytes) {
1050 res = -EIO;
1051 break;
1053 count -= nres;
1054 res += nres;
1055 pos += nres;
1056 buf += nres;
1057 if (nres != nbytes)
1058 break;
1059 if (count) {
1060 fuse_put_request(fc, req);
1061 req = fuse_get_req(fc);
1062 if (IS_ERR(req))
1063 break;
1066 if (!IS_ERR(req))
1067 fuse_put_request(fc, req);
1068 if (res > 0)
1069 *ppos = pos;
1071 return res;
1073 EXPORT_SYMBOL_GPL(fuse_direct_io);
1075 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
1076 size_t count, loff_t *ppos)
1078 ssize_t res;
1079 struct inode *inode = file->f_path.dentry->d_inode;
1081 if (is_bad_inode(inode))
1082 return -EIO;
1084 res = fuse_direct_io(file, buf, count, ppos, 0);
1086 fuse_invalidate_attr(inode);
1088 return res;
1091 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
1092 size_t count, loff_t *ppos)
1094 struct inode *inode = file->f_path.dentry->d_inode;
1095 ssize_t res;
1097 if (is_bad_inode(inode))
1098 return -EIO;
1100 /* Don't allow parallel writes to the same file */
1101 mutex_lock(&inode->i_mutex);
1102 res = generic_write_checks(file, ppos, &count, 0);
1103 if (!res) {
1104 res = fuse_direct_io(file, buf, count, ppos, 1);
1105 if (res > 0)
1106 fuse_write_update_size(inode, *ppos);
1108 mutex_unlock(&inode->i_mutex);
1110 fuse_invalidate_attr(inode);
1112 return res;
1115 static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
1117 __free_page(req->pages[0]);
1118 fuse_file_put(req->ff);
1121 static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
1123 struct inode *inode = req->inode;
1124 struct fuse_inode *fi = get_fuse_inode(inode);
1125 struct backing_dev_info *bdi = inode->i_mapping->backing_dev_info;
1127 list_del(&req->writepages_entry);
1128 dec_bdi_stat(bdi, BDI_WRITEBACK);
1129 dec_zone_page_state(req->pages[0], NR_WRITEBACK_TEMP);
1130 bdi_writeout_inc(bdi);
1131 wake_up(&fi->page_waitq);
1134 /* Called under fc->lock, may release and reacquire it */
1135 static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req)
1136 __releases(&fc->lock)
1137 __acquires(&fc->lock)
1139 struct fuse_inode *fi = get_fuse_inode(req->inode);
1140 loff_t size = i_size_read(req->inode);
1141 struct fuse_write_in *inarg = &req->misc.write.in;
1143 if (!fc->connected)
1144 goto out_free;
1146 if (inarg->offset + PAGE_CACHE_SIZE <= size) {
1147 inarg->size = PAGE_CACHE_SIZE;
1148 } else if (inarg->offset < size) {
1149 inarg->size = size & (PAGE_CACHE_SIZE - 1);
1150 } else {
1151 /* Got truncated off completely */
1152 goto out_free;
1155 req->in.args[1].size = inarg->size;
1156 fi->writectr++;
1157 fuse_request_send_background_locked(fc, req);
1158 return;
1160 out_free:
1161 fuse_writepage_finish(fc, req);
1162 spin_unlock(&fc->lock);
1163 fuse_writepage_free(fc, req);
1164 fuse_put_request(fc, req);
1165 spin_lock(&fc->lock);
1169 * If fi->writectr is positive (no truncate or fsync going on) send
1170 * all queued writepage requests.
1172 * Called with fc->lock
1174 void fuse_flush_writepages(struct inode *inode)
1175 __releases(&fc->lock)
1176 __acquires(&fc->lock)
1178 struct fuse_conn *fc = get_fuse_conn(inode);
1179 struct fuse_inode *fi = get_fuse_inode(inode);
1180 struct fuse_req *req;
1182 while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
1183 req = list_entry(fi->queued_writes.next, struct fuse_req, list);
1184 list_del_init(&req->list);
1185 fuse_send_writepage(fc, req);
1189 static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
1191 struct inode *inode = req->inode;
1192 struct fuse_inode *fi = get_fuse_inode(inode);
1194 mapping_set_error(inode->i_mapping, req->out.h.error);
1195 spin_lock(&fc->lock);
1196 fi->writectr--;
1197 fuse_writepage_finish(fc, req);
1198 spin_unlock(&fc->lock);
1199 fuse_writepage_free(fc, req);
1202 static int fuse_writepage_locked(struct page *page)
1204 struct address_space *mapping = page->mapping;
1205 struct inode *inode = mapping->host;
1206 struct fuse_conn *fc = get_fuse_conn(inode);
1207 struct fuse_inode *fi = get_fuse_inode(inode);
1208 struct fuse_req *req;
1209 struct fuse_file *ff;
1210 struct page *tmp_page;
1212 set_page_writeback(page);
1214 req = fuse_request_alloc_nofs();
1215 if (!req)
1216 goto err;
1218 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1219 if (!tmp_page)
1220 goto err_free;
1222 spin_lock(&fc->lock);
1223 BUG_ON(list_empty(&fi->write_files));
1224 ff = list_entry(fi->write_files.next, struct fuse_file, write_entry);
1225 req->ff = fuse_file_get(ff);
1226 spin_unlock(&fc->lock);
1228 fuse_write_fill(req, ff, page_offset(page), 0);
1230 copy_highpage(tmp_page, page);
1231 req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
1232 req->in.argpages = 1;
1233 req->num_pages = 1;
1234 req->pages[0] = tmp_page;
1235 req->page_offset = 0;
1236 req->end = fuse_writepage_end;
1237 req->inode = inode;
1239 inc_bdi_stat(mapping->backing_dev_info, BDI_WRITEBACK);
1240 inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP);
1241 end_page_writeback(page);
1243 spin_lock(&fc->lock);
1244 list_add(&req->writepages_entry, &fi->writepages);
1245 list_add_tail(&req->list, &fi->queued_writes);
1246 fuse_flush_writepages(inode);
1247 spin_unlock(&fc->lock);
1249 return 0;
1251 err_free:
1252 fuse_request_free(req);
1253 err:
1254 end_page_writeback(page);
1255 return -ENOMEM;
1258 static int fuse_writepage(struct page *page, struct writeback_control *wbc)
1260 int err;
1262 err = fuse_writepage_locked(page);
1263 unlock_page(page);
1265 return err;
1268 static int fuse_launder_page(struct page *page)
1270 int err = 0;
1271 if (clear_page_dirty_for_io(page)) {
1272 struct inode *inode = page->mapping->host;
1273 err = fuse_writepage_locked(page);
1274 if (!err)
1275 fuse_wait_on_page_writeback(inode, page->index);
1277 return err;
1281 * Write back dirty pages now, because there may not be any suitable
1282 * open files later
1284 static void fuse_vma_close(struct vm_area_struct *vma)
1286 filemap_write_and_wait(vma->vm_file->f_mapping);
1290 * Wait for writeback against this page to complete before allowing it
1291 * to be marked dirty again, and hence written back again, possibly
1292 * before the previous writepage completed.
1294 * Block here, instead of in ->writepage(), so that the userspace fs
1295 * can only block processes actually operating on the filesystem.
1297 * Otherwise unprivileged userspace fs would be able to block
1298 * unrelated:
1300 * - page migration
1301 * - sync(2)
1302 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
1304 static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1306 struct page *page = vmf->page;
1308 * Don't use page->mapping as it may become NULL from a
1309 * concurrent truncate.
1311 struct inode *inode = vma->vm_file->f_mapping->host;
1313 fuse_wait_on_page_writeback(inode, page->index);
1314 return 0;
1317 static struct vm_operations_struct fuse_file_vm_ops = {
1318 .close = fuse_vma_close,
1319 .fault = filemap_fault,
1320 .page_mkwrite = fuse_page_mkwrite,
1323 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
1325 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) {
1326 struct inode *inode = file->f_dentry->d_inode;
1327 struct fuse_conn *fc = get_fuse_conn(inode);
1328 struct fuse_inode *fi = get_fuse_inode(inode);
1329 struct fuse_file *ff = file->private_data;
1331 * file may be written through mmap, so chain it onto the
1332 * inodes's write_file list
1334 spin_lock(&fc->lock);
1335 if (list_empty(&ff->write_entry))
1336 list_add(&ff->write_entry, &fi->write_files);
1337 spin_unlock(&fc->lock);
1339 file_accessed(file);
1340 vma->vm_ops = &fuse_file_vm_ops;
1341 return 0;
1344 static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma)
1346 /* Can't provide the coherency needed for MAP_SHARED */
1347 if (vma->vm_flags & VM_MAYSHARE)
1348 return -ENODEV;
1350 invalidate_inode_pages2(file->f_mapping);
1352 return generic_file_mmap(file, vma);
1355 static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
1356 struct file_lock *fl)
1358 switch (ffl->type) {
1359 case F_UNLCK:
1360 break;
1362 case F_RDLCK:
1363 case F_WRLCK:
1364 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
1365 ffl->end < ffl->start)
1366 return -EIO;
1368 fl->fl_start = ffl->start;
1369 fl->fl_end = ffl->end;
1370 fl->fl_pid = ffl->pid;
1371 break;
1373 default:
1374 return -EIO;
1376 fl->fl_type = ffl->type;
1377 return 0;
1380 static void fuse_lk_fill(struct fuse_req *req, struct file *file,
1381 const struct file_lock *fl, int opcode, pid_t pid,
1382 int flock)
1384 struct inode *inode = file->f_path.dentry->d_inode;
1385 struct fuse_conn *fc = get_fuse_conn(inode);
1386 struct fuse_file *ff = file->private_data;
1387 struct fuse_lk_in *arg = &req->misc.lk_in;
1389 arg->fh = ff->fh;
1390 arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
1391 arg->lk.start = fl->fl_start;
1392 arg->lk.end = fl->fl_end;
1393 arg->lk.type = fl->fl_type;
1394 arg->lk.pid = pid;
1395 if (flock)
1396 arg->lk_flags |= FUSE_LK_FLOCK;
1397 req->in.h.opcode = opcode;
1398 req->in.h.nodeid = get_node_id(inode);
1399 req->in.numargs = 1;
1400 req->in.args[0].size = sizeof(*arg);
1401 req->in.args[0].value = arg;
1404 static int fuse_getlk(struct file *file, struct file_lock *fl)
1406 struct inode *inode = file->f_path.dentry->d_inode;
1407 struct fuse_conn *fc = get_fuse_conn(inode);
1408 struct fuse_req *req;
1409 struct fuse_lk_out outarg;
1410 int err;
1412 req = fuse_get_req(fc);
1413 if (IS_ERR(req))
1414 return PTR_ERR(req);
1416 fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0);
1417 req->out.numargs = 1;
1418 req->out.args[0].size = sizeof(outarg);
1419 req->out.args[0].value = &outarg;
1420 fuse_request_send(fc, req);
1421 err = req->out.h.error;
1422 fuse_put_request(fc, req);
1423 if (!err)
1424 err = convert_fuse_file_lock(&outarg.lk, fl);
1426 return err;
1429 static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
1431 struct inode *inode = file->f_path.dentry->d_inode;
1432 struct fuse_conn *fc = get_fuse_conn(inode);
1433 struct fuse_req *req;
1434 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
1435 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
1436 int err;
1438 if (fl->fl_lmops && fl->fl_lmops->fl_grant) {
1439 /* NLM needs asynchronous locks, which we don't support yet */
1440 return -ENOLCK;
1443 /* Unlock on close is handled by the flush method */
1444 if (fl->fl_flags & FL_CLOSE)
1445 return 0;
1447 req = fuse_get_req(fc);
1448 if (IS_ERR(req))
1449 return PTR_ERR(req);
1451 fuse_lk_fill(req, file, fl, opcode, pid, flock);
1452 fuse_request_send(fc, req);
1453 err = req->out.h.error;
1454 /* locking is restartable */
1455 if (err == -EINTR)
1456 err = -ERESTARTSYS;
1457 fuse_put_request(fc, req);
1458 return err;
1461 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
1463 struct inode *inode = file->f_path.dentry->d_inode;
1464 struct fuse_conn *fc = get_fuse_conn(inode);
1465 int err;
1467 if (cmd == F_CANCELLK) {
1468 err = 0;
1469 } else if (cmd == F_GETLK) {
1470 if (fc->no_lock) {
1471 posix_test_lock(file, fl);
1472 err = 0;
1473 } else
1474 err = fuse_getlk(file, fl);
1475 } else {
1476 if (fc->no_lock)
1477 err = posix_lock_file(file, fl, NULL);
1478 else
1479 err = fuse_setlk(file, fl, 0);
1481 return err;
1484 static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
1486 struct inode *inode = file->f_path.dentry->d_inode;
1487 struct fuse_conn *fc = get_fuse_conn(inode);
1488 int err;
1490 if (fc->no_lock) {
1491 err = flock_lock_file_wait(file, fl);
1492 } else {
1493 /* emulate flock with POSIX locks */
1494 fl->fl_owner = (fl_owner_t) file;
1495 err = fuse_setlk(file, fl, 1);
1498 return err;
1501 static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
1503 struct inode *inode = mapping->host;
1504 struct fuse_conn *fc = get_fuse_conn(inode);
1505 struct fuse_req *req;
1506 struct fuse_bmap_in inarg;
1507 struct fuse_bmap_out outarg;
1508 int err;
1510 if (!inode->i_sb->s_bdev || fc->no_bmap)
1511 return 0;
1513 req = fuse_get_req(fc);
1514 if (IS_ERR(req))
1515 return 0;
1517 memset(&inarg, 0, sizeof(inarg));
1518 inarg.block = block;
1519 inarg.blocksize = inode->i_sb->s_blocksize;
1520 req->in.h.opcode = FUSE_BMAP;
1521 req->in.h.nodeid = get_node_id(inode);
1522 req->in.numargs = 1;
1523 req->in.args[0].size = sizeof(inarg);
1524 req->in.args[0].value = &inarg;
1525 req->out.numargs = 1;
1526 req->out.args[0].size = sizeof(outarg);
1527 req->out.args[0].value = &outarg;
1528 fuse_request_send(fc, req);
1529 err = req->out.h.error;
1530 fuse_put_request(fc, req);
1531 if (err == -ENOSYS)
1532 fc->no_bmap = 1;
1534 return err ? 0 : outarg.block;
1537 static loff_t fuse_file_llseek(struct file *file, loff_t offset, int origin)
1539 loff_t retval;
1540 struct inode *inode = file->f_path.dentry->d_inode;
1542 mutex_lock(&inode->i_mutex);
1543 switch (origin) {
1544 case SEEK_END:
1545 retval = fuse_update_attributes(inode, NULL, file, NULL);
1546 if (retval)
1547 goto exit;
1548 offset += i_size_read(inode);
1549 break;
1550 case SEEK_CUR:
1551 offset += file->f_pos;
1553 retval = -EINVAL;
1554 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
1555 if (offset != file->f_pos) {
1556 file->f_pos = offset;
1557 file->f_version = 0;
1559 retval = offset;
1561 exit:
1562 mutex_unlock(&inode->i_mutex);
1563 return retval;
1566 static int fuse_ioctl_copy_user(struct page **pages, struct iovec *iov,
1567 unsigned int nr_segs, size_t bytes, bool to_user)
1569 struct iov_iter ii;
1570 int page_idx = 0;
1572 if (!bytes)
1573 return 0;
1575 iov_iter_init(&ii, iov, nr_segs, bytes, 0);
1577 while (iov_iter_count(&ii)) {
1578 struct page *page = pages[page_idx++];
1579 size_t todo = min_t(size_t, PAGE_SIZE, iov_iter_count(&ii));
1580 void *kaddr, *map;
1582 kaddr = map = kmap(page);
1584 while (todo) {
1585 char __user *uaddr = ii.iov->iov_base + ii.iov_offset;
1586 size_t iov_len = ii.iov->iov_len - ii.iov_offset;
1587 size_t copy = min(todo, iov_len);
1588 size_t left;
1590 if (!to_user)
1591 left = copy_from_user(kaddr, uaddr, copy);
1592 else
1593 left = copy_to_user(uaddr, kaddr, copy);
1595 if (unlikely(left))
1596 return -EFAULT;
1598 iov_iter_advance(&ii, copy);
1599 todo -= copy;
1600 kaddr += copy;
1603 kunmap(page);
1606 return 0;
1610 * For ioctls, there is no generic way to determine how much memory
1611 * needs to be read and/or written. Furthermore, ioctls are allowed
1612 * to dereference the passed pointer, so the parameter requires deep
1613 * copying but FUSE has no idea whatsoever about what to copy in or
1614 * out.
1616 * This is solved by allowing FUSE server to retry ioctl with
1617 * necessary in/out iovecs. Let's assume the ioctl implementation
1618 * needs to read in the following structure.
1620 * struct a {
1621 * char *buf;
1622 * size_t buflen;
1625 * On the first callout to FUSE server, inarg->in_size and
1626 * inarg->out_size will be NULL; then, the server completes the ioctl
1627 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
1628 * the actual iov array to
1630 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } }
1632 * which tells FUSE to copy in the requested area and retry the ioctl.
1633 * On the second round, the server has access to the structure and
1634 * from that it can tell what to look for next, so on the invocation,
1635 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
1637 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) },
1638 * { .iov_base = a.buf, .iov_len = a.buflen } }
1640 * FUSE will copy both struct a and the pointed buffer from the
1641 * process doing the ioctl and retry ioctl with both struct a and the
1642 * buffer.
1644 * This time, FUSE server has everything it needs and completes ioctl
1645 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
1647 * Copying data out works the same way.
1649 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
1650 * automatically initializes in and out iovs by decoding @cmd with
1651 * _IOC_* macros and the server is not allowed to request RETRY. This
1652 * limits ioctl data transfers to well-formed ioctls and is the forced
1653 * behavior for all FUSE servers.
1655 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1656 unsigned int flags)
1658 struct fuse_file *ff = file->private_data;
1659 struct fuse_conn *fc = ff->fc;
1660 struct fuse_ioctl_in inarg = {
1661 .fh = ff->fh,
1662 .cmd = cmd,
1663 .arg = arg,
1664 .flags = flags
1666 struct fuse_ioctl_out outarg;
1667 struct fuse_req *req = NULL;
1668 struct page **pages = NULL;
1669 struct page *iov_page = NULL;
1670 struct iovec *in_iov = NULL, *out_iov = NULL;
1671 unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages;
1672 size_t in_size, out_size, transferred;
1673 int err;
1675 /* assume all the iovs returned by client always fits in a page */
1676 BUILD_BUG_ON(sizeof(struct iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
1678 err = -ENOMEM;
1679 pages = kzalloc(sizeof(pages[0]) * FUSE_MAX_PAGES_PER_REQ, GFP_KERNEL);
1680 iov_page = alloc_page(GFP_KERNEL);
1681 if (!pages || !iov_page)
1682 goto out;
1685 * If restricted, initialize IO parameters as encoded in @cmd.
1686 * RETRY from server is not allowed.
1688 if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
1689 struct iovec *iov = page_address(iov_page);
1691 iov->iov_base = (void __user *)arg;
1692 iov->iov_len = _IOC_SIZE(cmd);
1694 if (_IOC_DIR(cmd) & _IOC_WRITE) {
1695 in_iov = iov;
1696 in_iovs = 1;
1699 if (_IOC_DIR(cmd) & _IOC_READ) {
1700 out_iov = iov;
1701 out_iovs = 1;
1705 retry:
1706 inarg.in_size = in_size = iov_length(in_iov, in_iovs);
1707 inarg.out_size = out_size = iov_length(out_iov, out_iovs);
1710 * Out data can be used either for actual out data or iovs,
1711 * make sure there always is at least one page.
1713 out_size = max_t(size_t, out_size, PAGE_SIZE);
1714 max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
1716 /* make sure there are enough buffer pages and init request with them */
1717 err = -ENOMEM;
1718 if (max_pages > FUSE_MAX_PAGES_PER_REQ)
1719 goto out;
1720 while (num_pages < max_pages) {
1721 pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
1722 if (!pages[num_pages])
1723 goto out;
1724 num_pages++;
1727 req = fuse_get_req(fc);
1728 if (IS_ERR(req)) {
1729 err = PTR_ERR(req);
1730 req = NULL;
1731 goto out;
1733 memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages);
1734 req->num_pages = num_pages;
1736 /* okay, let's send it to the client */
1737 req->in.h.opcode = FUSE_IOCTL;
1738 req->in.h.nodeid = ff->nodeid;
1739 req->in.numargs = 1;
1740 req->in.args[0].size = sizeof(inarg);
1741 req->in.args[0].value = &inarg;
1742 if (in_size) {
1743 req->in.numargs++;
1744 req->in.args[1].size = in_size;
1745 req->in.argpages = 1;
1747 err = fuse_ioctl_copy_user(pages, in_iov, in_iovs, in_size,
1748 false);
1749 if (err)
1750 goto out;
1753 req->out.numargs = 2;
1754 req->out.args[0].size = sizeof(outarg);
1755 req->out.args[0].value = &outarg;
1756 req->out.args[1].size = out_size;
1757 req->out.argpages = 1;
1758 req->out.argvar = 1;
1760 fuse_request_send(fc, req);
1761 err = req->out.h.error;
1762 transferred = req->out.args[1].size;
1763 fuse_put_request(fc, req);
1764 req = NULL;
1765 if (err)
1766 goto out;
1768 /* did it ask for retry? */
1769 if (outarg.flags & FUSE_IOCTL_RETRY) {
1770 char *vaddr;
1772 /* no retry if in restricted mode */
1773 err = -EIO;
1774 if (!(flags & FUSE_IOCTL_UNRESTRICTED))
1775 goto out;
1777 in_iovs = outarg.in_iovs;
1778 out_iovs = outarg.out_iovs;
1781 * Make sure things are in boundary, separate checks
1782 * are to protect against overflow.
1784 err = -ENOMEM;
1785 if (in_iovs > FUSE_IOCTL_MAX_IOV ||
1786 out_iovs > FUSE_IOCTL_MAX_IOV ||
1787 in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
1788 goto out;
1790 err = -EIO;
1791 if ((in_iovs + out_iovs) * sizeof(struct iovec) != transferred)
1792 goto out;
1794 /* okay, copy in iovs and retry */
1795 vaddr = kmap_atomic(pages[0], KM_USER0);
1796 memcpy(page_address(iov_page), vaddr, transferred);
1797 kunmap_atomic(vaddr, KM_USER0);
1799 in_iov = page_address(iov_page);
1800 out_iov = in_iov + in_iovs;
1802 goto retry;
1805 err = -EIO;
1806 if (transferred > inarg.out_size)
1807 goto out;
1809 err = fuse_ioctl_copy_user(pages, out_iov, out_iovs, transferred, true);
1810 out:
1811 if (req)
1812 fuse_put_request(fc, req);
1813 if (iov_page)
1814 __free_page(iov_page);
1815 while (num_pages)
1816 __free_page(pages[--num_pages]);
1817 kfree(pages);
1819 return err ? err : outarg.result;
1821 EXPORT_SYMBOL_GPL(fuse_do_ioctl);
1823 static long fuse_file_ioctl_common(struct file *file, unsigned int cmd,
1824 unsigned long arg, unsigned int flags)
1826 struct inode *inode = file->f_dentry->d_inode;
1827 struct fuse_conn *fc = get_fuse_conn(inode);
1829 if (!fuse_allow_task(fc, current))
1830 return -EACCES;
1832 if (is_bad_inode(inode))
1833 return -EIO;
1835 return fuse_do_ioctl(file, cmd, arg, flags);
1838 static long fuse_file_ioctl(struct file *file, unsigned int cmd,
1839 unsigned long arg)
1841 return fuse_file_ioctl_common(file, cmd, arg, 0);
1844 static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
1845 unsigned long arg)
1847 return fuse_file_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
1851 * All files which have been polled are linked to RB tree
1852 * fuse_conn->polled_files which is indexed by kh. Walk the tree and
1853 * find the matching one.
1855 static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh,
1856 struct rb_node **parent_out)
1858 struct rb_node **link = &fc->polled_files.rb_node;
1859 struct rb_node *last = NULL;
1861 while (*link) {
1862 struct fuse_file *ff;
1864 last = *link;
1865 ff = rb_entry(last, struct fuse_file, polled_node);
1867 if (kh < ff->kh)
1868 link = &last->rb_left;
1869 else if (kh > ff->kh)
1870 link = &last->rb_right;
1871 else
1872 return link;
1875 if (parent_out)
1876 *parent_out = last;
1877 return link;
1881 * The file is about to be polled. Make sure it's on the polled_files
1882 * RB tree. Note that files once added to the polled_files tree are
1883 * not removed before the file is released. This is because a file
1884 * polled once is likely to be polled again.
1886 static void fuse_register_polled_file(struct fuse_conn *fc,
1887 struct fuse_file *ff)
1889 spin_lock(&fc->lock);
1890 if (RB_EMPTY_NODE(&ff->polled_node)) {
1891 struct rb_node **link, *parent;
1893 link = fuse_find_polled_node(fc, ff->kh, &parent);
1894 BUG_ON(*link);
1895 rb_link_node(&ff->polled_node, parent, link);
1896 rb_insert_color(&ff->polled_node, &fc->polled_files);
1898 spin_unlock(&fc->lock);
1901 unsigned fuse_file_poll(struct file *file, poll_table *wait)
1903 struct fuse_file *ff = file->private_data;
1904 struct fuse_conn *fc = ff->fc;
1905 struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh };
1906 struct fuse_poll_out outarg;
1907 struct fuse_req *req;
1908 int err;
1910 if (fc->no_poll)
1911 return DEFAULT_POLLMASK;
1913 poll_wait(file, &ff->poll_wait, wait);
1916 * Ask for notification iff there's someone waiting for it.
1917 * The client may ignore the flag and always notify.
1919 if (waitqueue_active(&ff->poll_wait)) {
1920 inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY;
1921 fuse_register_polled_file(fc, ff);
1924 req = fuse_get_req(fc);
1925 if (IS_ERR(req))
1926 return POLLERR;
1928 req->in.h.opcode = FUSE_POLL;
1929 req->in.h.nodeid = ff->nodeid;
1930 req->in.numargs = 1;
1931 req->in.args[0].size = sizeof(inarg);
1932 req->in.args[0].value = &inarg;
1933 req->out.numargs = 1;
1934 req->out.args[0].size = sizeof(outarg);
1935 req->out.args[0].value = &outarg;
1936 fuse_request_send(fc, req);
1937 err = req->out.h.error;
1938 fuse_put_request(fc, req);
1940 if (!err)
1941 return outarg.revents;
1942 if (err == -ENOSYS) {
1943 fc->no_poll = 1;
1944 return DEFAULT_POLLMASK;
1946 return POLLERR;
1948 EXPORT_SYMBOL_GPL(fuse_file_poll);
1951 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
1952 * wakes up the poll waiters.
1954 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
1955 struct fuse_notify_poll_wakeup_out *outarg)
1957 u64 kh = outarg->kh;
1958 struct rb_node **link;
1960 spin_lock(&fc->lock);
1962 link = fuse_find_polled_node(fc, kh, NULL);
1963 if (*link) {
1964 struct fuse_file *ff;
1966 ff = rb_entry(*link, struct fuse_file, polled_node);
1967 wake_up_interruptible_sync(&ff->poll_wait);
1970 spin_unlock(&fc->lock);
1971 return 0;
1974 static const struct file_operations fuse_file_operations = {
1975 .llseek = fuse_file_llseek,
1976 .read = do_sync_read,
1977 .aio_read = fuse_file_aio_read,
1978 .write = do_sync_write,
1979 .aio_write = fuse_file_aio_write,
1980 .mmap = fuse_file_mmap,
1981 .open = fuse_open,
1982 .flush = fuse_flush,
1983 .release = fuse_release,
1984 .fsync = fuse_fsync,
1985 .lock = fuse_file_lock,
1986 .flock = fuse_file_flock,
1987 .splice_read = generic_file_splice_read,
1988 .unlocked_ioctl = fuse_file_ioctl,
1989 .compat_ioctl = fuse_file_compat_ioctl,
1990 .poll = fuse_file_poll,
1993 static const struct file_operations fuse_direct_io_file_operations = {
1994 .llseek = fuse_file_llseek,
1995 .read = fuse_direct_read,
1996 .write = fuse_direct_write,
1997 .mmap = fuse_direct_mmap,
1998 .open = fuse_open,
1999 .flush = fuse_flush,
2000 .release = fuse_release,
2001 .fsync = fuse_fsync,
2002 .lock = fuse_file_lock,
2003 .flock = fuse_file_flock,
2004 .unlocked_ioctl = fuse_file_ioctl,
2005 .compat_ioctl = fuse_file_compat_ioctl,
2006 .poll = fuse_file_poll,
2007 /* no splice_read */
2010 static const struct address_space_operations fuse_file_aops = {
2011 .readpage = fuse_readpage,
2012 .writepage = fuse_writepage,
2013 .launder_page = fuse_launder_page,
2014 .write_begin = fuse_write_begin,
2015 .write_end = fuse_write_end,
2016 .readpages = fuse_readpages,
2017 .set_page_dirty = __set_page_dirty_nobuffers,
2018 .bmap = fuse_bmap,
2021 void fuse_init_file_inode(struct inode *inode)
2023 inode->i_fop = &fuse_file_operations;
2024 inode->i_data.a_ops = &fuse_file_aops;