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.
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>
19 static const struct file_operations fuse_direct_io_file_operations;
21 static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
22 int opcode, struct fuse_open_out *outargp)
24 struct fuse_open_in inarg;
28 req = fuse_get_req(fc);
32 memset(&inarg, 0, sizeof(inarg));
33 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY);
34 if (!fc->atomic_o_trunc)
35 inarg.flags &= ~O_TRUNC;
36 req->in.h.opcode = opcode;
37 req->in.h.nodeid = nodeid;
39 req->in.args[0].size = sizeof(inarg);
40 req->in.args[0].value = &inarg;
42 req->out.args[0].size = sizeof(*outargp);
43 req->out.args[0].value = outargp;
44 fuse_request_send(fc, req);
45 err = req->out.h.error;
46 fuse_put_request(fc, req);
51 struct fuse_file *fuse_file_alloc(struct fuse_conn *fc)
55 ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
60 ff->reserved_req = fuse_request_alloc();
61 if (unlikely(!ff->reserved_req)) {
66 INIT_LIST_HEAD(&ff->write_entry);
67 atomic_set(&ff->count, 0);
68 RB_CLEAR_NODE(&ff->polled_node);
69 init_waitqueue_head(&ff->poll_wait);
73 spin_unlock(&fc->lock);
78 void fuse_file_free(struct fuse_file *ff)
80 fuse_request_free(ff->reserved_req);
84 struct fuse_file *fuse_file_get(struct fuse_file *ff)
86 atomic_inc(&ff->count);
90 static void fuse_release_async(struct work_struct *work)
96 req = container_of(work, struct fuse_req, misc.release.work);
97 path = req->misc.release.path;
98 fc = get_fuse_conn(path.dentry->d_inode);
100 fuse_put_request(fc, req);
104 static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
106 if (fc->destroy_req) {
108 * If this is a fuseblk mount, then it's possible that
109 * releasing the path will result in releasing the
110 * super block and sending the DESTROY request. If
111 * the server is single threaded, this would hang.
112 * For this reason do the path_put() in a separate
115 atomic_inc(&req->count);
116 INIT_WORK(&req->misc.release.work, fuse_release_async);
117 schedule_work(&req->misc.release.work);
119 path_put(&req->misc.release.path);
123 static void fuse_file_put(struct fuse_file *ff, bool sync)
125 if (atomic_dec_and_test(&ff->count)) {
126 struct fuse_req *req = ff->reserved_req;
129 fuse_request_send(ff->fc, req);
130 path_put(&req->misc.release.path);
131 fuse_put_request(ff->fc, req);
133 req->end = fuse_release_end;
134 fuse_request_send_background(ff->fc, req);
140 int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
143 struct fuse_open_out outarg;
144 struct fuse_file *ff;
146 int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
148 ff = fuse_file_alloc(fc);
152 err = fuse_send_open(fc, nodeid, file, opcode, &outarg);
159 outarg.open_flags &= ~FOPEN_DIRECT_IO;
163 ff->open_flags = outarg.open_flags;
164 file->private_data = fuse_file_get(ff);
168 EXPORT_SYMBOL_GPL(fuse_do_open);
170 void fuse_finish_open(struct inode *inode, struct file *file)
172 struct fuse_file *ff = file->private_data;
173 struct fuse_conn *fc = get_fuse_conn(inode);
175 if (ff->open_flags & FOPEN_DIRECT_IO)
176 file->f_op = &fuse_direct_io_file_operations;
177 if (!(ff->open_flags & FOPEN_KEEP_CACHE))
178 invalidate_inode_pages2(inode->i_mapping);
179 if (ff->open_flags & FOPEN_NONSEEKABLE)
180 nonseekable_open(inode, file);
181 if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) {
182 struct fuse_inode *fi = get_fuse_inode(inode);
184 spin_lock(&fc->lock);
185 fi->attr_version = ++fc->attr_version;
186 i_size_write(inode, 0);
187 spin_unlock(&fc->lock);
188 fuse_invalidate_attr(inode);
192 int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
194 struct fuse_conn *fc = get_fuse_conn(inode);
197 /* VFS checks this, but only _after_ ->open() */
198 if (file->f_flags & O_DIRECT)
201 err = generic_file_open(inode, file);
205 err = fuse_do_open(fc, get_node_id(inode), file, isdir);
209 fuse_finish_open(inode, file);
214 static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode)
216 struct fuse_conn *fc = ff->fc;
217 struct fuse_req *req = ff->reserved_req;
218 struct fuse_release_in *inarg = &req->misc.release.in;
220 spin_lock(&fc->lock);
221 list_del(&ff->write_entry);
222 if (!RB_EMPTY_NODE(&ff->polled_node))
223 rb_erase(&ff->polled_node, &fc->polled_files);
224 spin_unlock(&fc->lock);
226 wake_up_interruptible_all(&ff->poll_wait);
229 inarg->flags = flags;
230 req->in.h.opcode = opcode;
231 req->in.h.nodeid = ff->nodeid;
233 req->in.args[0].size = sizeof(struct fuse_release_in);
234 req->in.args[0].value = inarg;
237 void fuse_release_common(struct file *file, int opcode)
239 struct fuse_file *ff;
240 struct fuse_req *req;
242 ff = file->private_data;
246 req = ff->reserved_req;
247 fuse_prepare_release(ff, file->f_flags, opcode);
250 struct fuse_release_in *inarg = &req->misc.release.in;
251 inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK;
252 inarg->lock_owner = fuse_lock_owner_id(ff->fc,
255 /* Hold vfsmount and dentry until release is finished */
256 path_get(&file->f_path);
257 req->misc.release.path = file->f_path;
260 * Normally this will send the RELEASE request, however if
261 * some asynchronous READ or WRITE requests are outstanding,
262 * the sending will be delayed.
264 * Make the release synchronous if this is a fuseblk mount,
265 * synchronous RELEASE is allowed (and desirable) in this case
266 * because the server can be trusted not to screw up.
268 fuse_file_put(ff, ff->fc->destroy_req != NULL);
271 static int fuse_open(struct inode *inode, struct file *file)
273 return fuse_open_common(inode, file, false);
276 static int fuse_release(struct inode *inode, struct file *file)
278 fuse_release_common(file, FUSE_RELEASE);
280 /* return value is ignored by VFS */
284 void fuse_sync_release(struct fuse_file *ff, int flags)
286 WARN_ON(atomic_read(&ff->count) > 1);
287 fuse_prepare_release(ff, flags, FUSE_RELEASE);
288 ff->reserved_req->force = 1;
289 fuse_request_send(ff->fc, ff->reserved_req);
290 fuse_put_request(ff->fc, ff->reserved_req);
293 EXPORT_SYMBOL_GPL(fuse_sync_release);
296 * Scramble the ID space with XTEA, so that the value of the files_struct
297 * pointer is not exposed to userspace.
299 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
301 u32 *k = fc->scramble_key;
302 u64 v = (unsigned long) id;
308 for (i = 0; i < 32; i++) {
309 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
311 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
314 return (u64) v0 + ((u64) v1 << 32);
318 * Check if page is under writeback
320 * This is currently done by walking the list of writepage requests
321 * for the inode, which can be pretty inefficient.
323 static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
325 struct fuse_conn *fc = get_fuse_conn(inode);
326 struct fuse_inode *fi = get_fuse_inode(inode);
327 struct fuse_req *req;
330 spin_lock(&fc->lock);
331 list_for_each_entry(req, &fi->writepages, writepages_entry) {
334 BUG_ON(req->inode != inode);
335 curr_index = req->misc.write.in.offset >> PAGE_CACHE_SHIFT;
336 if (curr_index == index) {
341 spin_unlock(&fc->lock);
347 * Wait for page writeback to be completed.
349 * Since fuse doesn't rely on the VM writeback tracking, this has to
350 * use some other means.
352 static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
354 struct fuse_inode *fi = get_fuse_inode(inode);
356 wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
360 static int fuse_flush(struct file *file, fl_owner_t id)
362 struct inode *inode = file->f_path.dentry->d_inode;
363 struct fuse_conn *fc = get_fuse_conn(inode);
364 struct fuse_file *ff = file->private_data;
365 struct fuse_req *req;
366 struct fuse_flush_in inarg;
369 if (is_bad_inode(inode))
375 req = fuse_get_req_nofail(fc, file);
376 memset(&inarg, 0, sizeof(inarg));
378 inarg.lock_owner = fuse_lock_owner_id(fc, id);
379 req->in.h.opcode = FUSE_FLUSH;
380 req->in.h.nodeid = get_node_id(inode);
382 req->in.args[0].size = sizeof(inarg);
383 req->in.args[0].value = &inarg;
385 fuse_request_send(fc, req);
386 err = req->out.h.error;
387 fuse_put_request(fc, req);
388 if (err == -ENOSYS) {
396 * Wait for all pending writepages on the inode to finish.
398 * This is currently done by blocking further writes with FUSE_NOWRITE
399 * and waiting for all sent writes to complete.
401 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
402 * could conflict with truncation.
404 static void fuse_sync_writes(struct inode *inode)
406 fuse_set_nowrite(inode);
407 fuse_release_nowrite(inode);
410 int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
411 int datasync, int isdir)
413 struct inode *inode = file->f_mapping->host;
414 struct fuse_conn *fc = get_fuse_conn(inode);
415 struct fuse_file *ff = file->private_data;
416 struct fuse_req *req;
417 struct fuse_fsync_in inarg;
420 if (is_bad_inode(inode))
423 err = filemap_write_and_wait_range(inode->i_mapping, start, end);
427 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
430 mutex_lock(&inode->i_mutex);
433 * Start writeback against all dirty pages of the inode, then
434 * wait for all outstanding writes, before sending the FSYNC
437 err = write_inode_now(inode, 0);
441 fuse_sync_writes(inode);
443 req = fuse_get_req(fc);
449 memset(&inarg, 0, sizeof(inarg));
451 inarg.fsync_flags = datasync ? 1 : 0;
452 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
453 req->in.h.nodeid = get_node_id(inode);
455 req->in.args[0].size = sizeof(inarg);
456 req->in.args[0].value = &inarg;
457 fuse_request_send(fc, req);
458 err = req->out.h.error;
459 fuse_put_request(fc, req);
460 if (err == -ENOSYS) {
468 mutex_unlock(&inode->i_mutex);
472 static int fuse_fsync(struct file *file, loff_t start, loff_t end,
475 return fuse_fsync_common(file, start, end, datasync, 0);
478 void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos,
479 size_t count, int opcode)
481 struct fuse_read_in *inarg = &req->misc.read.in;
482 struct fuse_file *ff = file->private_data;
487 inarg->flags = file->f_flags;
488 req->in.h.opcode = opcode;
489 req->in.h.nodeid = ff->nodeid;
491 req->in.args[0].size = sizeof(struct fuse_read_in);
492 req->in.args[0].value = inarg;
494 req->out.numargs = 1;
495 req->out.args[0].size = count;
498 static size_t fuse_send_read(struct fuse_req *req, struct file *file,
499 loff_t pos, size_t count, fl_owner_t owner)
501 struct fuse_file *ff = file->private_data;
502 struct fuse_conn *fc = ff->fc;
504 fuse_read_fill(req, file, pos, count, FUSE_READ);
506 struct fuse_read_in *inarg = &req->misc.read.in;
508 inarg->read_flags |= FUSE_READ_LOCKOWNER;
509 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
511 fuse_request_send(fc, req);
512 return req->out.args[0].size;
515 static void fuse_read_update_size(struct inode *inode, loff_t size,
518 struct fuse_conn *fc = get_fuse_conn(inode);
519 struct fuse_inode *fi = get_fuse_inode(inode);
521 spin_lock(&fc->lock);
522 if (attr_ver == fi->attr_version && size < inode->i_size &&
523 !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
524 fi->attr_version = ++fc->attr_version;
525 i_size_write(inode, size);
527 spin_unlock(&fc->lock);
530 static int fuse_readpage(struct file *file, struct page *page)
532 struct inode *inode = page->mapping->host;
533 struct fuse_conn *fc = get_fuse_conn(inode);
534 struct fuse_req *req;
536 loff_t pos = page_offset(page);
537 size_t count = PAGE_CACHE_SIZE;
542 if (is_bad_inode(inode))
546 * Page writeback can extend beyond the lifetime of the
547 * page-cache page, so make sure we read a properly synced
550 fuse_wait_on_page_writeback(inode, page->index);
552 req = fuse_get_req(fc);
557 attr_ver = fuse_get_attr_version(fc);
559 req->out.page_zeroing = 1;
560 req->out.argpages = 1;
562 req->pages[0] = page;
563 num_read = fuse_send_read(req, file, pos, count, NULL);
564 err = req->out.h.error;
565 fuse_put_request(fc, req);
569 * Short read means EOF. If file size is larger, truncate it
571 if (num_read < count)
572 fuse_read_update_size(inode, pos + num_read, attr_ver);
574 SetPageUptodate(page);
577 fuse_invalidate_attr(inode); /* atime changed */
583 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
586 size_t count = req->misc.read.in.size;
587 size_t num_read = req->out.args[0].size;
588 struct address_space *mapping = NULL;
590 for (i = 0; mapping == NULL && i < req->num_pages; i++)
591 mapping = req->pages[i]->mapping;
594 struct inode *inode = mapping->host;
597 * Short read means EOF. If file size is larger, truncate it
599 if (!req->out.h.error && num_read < count) {
602 pos = page_offset(req->pages[0]) + num_read;
603 fuse_read_update_size(inode, pos,
604 req->misc.read.attr_ver);
606 fuse_invalidate_attr(inode); /* atime changed */
609 for (i = 0; i < req->num_pages; i++) {
610 struct page *page = req->pages[i];
611 if (!req->out.h.error)
612 SetPageUptodate(page);
616 page_cache_release(page);
619 fuse_file_put(req->ff, false);
622 static void fuse_send_readpages(struct fuse_req *req, struct file *file)
624 struct fuse_file *ff = file->private_data;
625 struct fuse_conn *fc = ff->fc;
626 loff_t pos = page_offset(req->pages[0]);
627 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
629 req->out.argpages = 1;
630 req->out.page_zeroing = 1;
631 req->out.page_replace = 1;
632 fuse_read_fill(req, file, pos, count, FUSE_READ);
633 req->misc.read.attr_ver = fuse_get_attr_version(fc);
634 if (fc->async_read) {
635 req->ff = fuse_file_get(ff);
636 req->end = fuse_readpages_end;
637 fuse_request_send_background(fc, req);
639 fuse_request_send(fc, req);
640 fuse_readpages_end(fc, req);
641 fuse_put_request(fc, req);
645 struct fuse_fill_data {
646 struct fuse_req *req;
651 static int fuse_readpages_fill(void *_data, struct page *page)
653 struct fuse_fill_data *data = _data;
654 struct fuse_req *req = data->req;
655 struct inode *inode = data->inode;
656 struct fuse_conn *fc = get_fuse_conn(inode);
658 fuse_wait_on_page_writeback(inode, page->index);
660 if (req->num_pages &&
661 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
662 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
663 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
664 fuse_send_readpages(req, data->file);
665 data->req = req = fuse_get_req(fc);
671 page_cache_get(page);
672 req->pages[req->num_pages] = page;
677 static int fuse_readpages(struct file *file, struct address_space *mapping,
678 struct list_head *pages, unsigned nr_pages)
680 struct inode *inode = mapping->host;
681 struct fuse_conn *fc = get_fuse_conn(inode);
682 struct fuse_fill_data data;
686 if (is_bad_inode(inode))
691 data.req = fuse_get_req(fc);
692 err = PTR_ERR(data.req);
693 if (IS_ERR(data.req))
696 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
698 if (data.req->num_pages)
699 fuse_send_readpages(data.req, file);
701 fuse_put_request(fc, data.req);
707 static ssize_t fuse_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
708 unsigned long nr_segs, loff_t pos)
710 struct inode *inode = iocb->ki_filp->f_mapping->host;
712 if (pos + iov_length(iov, nr_segs) > i_size_read(inode)) {
715 * If trying to read past EOF, make sure the i_size
716 * attribute is up-to-date.
718 err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL);
723 return generic_file_aio_read(iocb, iov, nr_segs, pos);
726 static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff,
727 loff_t pos, size_t count)
729 struct fuse_write_in *inarg = &req->misc.write.in;
730 struct fuse_write_out *outarg = &req->misc.write.out;
735 req->in.h.opcode = FUSE_WRITE;
736 req->in.h.nodeid = ff->nodeid;
738 if (ff->fc->minor < 9)
739 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
741 req->in.args[0].size = sizeof(struct fuse_write_in);
742 req->in.args[0].value = inarg;
743 req->in.args[1].size = count;
744 req->out.numargs = 1;
745 req->out.args[0].size = sizeof(struct fuse_write_out);
746 req->out.args[0].value = outarg;
749 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
750 loff_t pos, size_t count, fl_owner_t owner)
752 struct fuse_file *ff = file->private_data;
753 struct fuse_conn *fc = ff->fc;
754 struct fuse_write_in *inarg = &req->misc.write.in;
756 fuse_write_fill(req, ff, pos, count);
757 inarg->flags = file->f_flags;
759 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
760 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
762 fuse_request_send(fc, req);
763 return req->misc.write.out.size;
766 void fuse_write_update_size(struct inode *inode, loff_t pos)
768 struct fuse_conn *fc = get_fuse_conn(inode);
769 struct fuse_inode *fi = get_fuse_inode(inode);
771 spin_lock(&fc->lock);
772 fi->attr_version = ++fc->attr_version;
773 if (pos > inode->i_size)
774 i_size_write(inode, pos);
775 spin_unlock(&fc->lock);
778 static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file,
779 struct inode *inode, loff_t pos,
786 for (i = 0; i < req->num_pages; i++)
787 fuse_wait_on_page_writeback(inode, req->pages[i]->index);
789 res = fuse_send_write(req, file, pos, count, NULL);
791 offset = req->page_offset;
793 for (i = 0; i < req->num_pages; i++) {
794 struct page *page = req->pages[i];
796 if (!req->out.h.error && !offset && count >= PAGE_CACHE_SIZE)
797 SetPageUptodate(page);
799 if (count > PAGE_CACHE_SIZE - offset)
800 count -= PAGE_CACHE_SIZE - offset;
806 page_cache_release(page);
812 static ssize_t fuse_fill_write_pages(struct fuse_req *req,
813 struct address_space *mapping,
814 struct iov_iter *ii, loff_t pos)
816 struct fuse_conn *fc = get_fuse_conn(mapping->host);
817 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
821 req->in.argpages = 1;
822 req->page_offset = offset;
827 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
828 size_t bytes = min_t(size_t, PAGE_CACHE_SIZE - offset,
831 bytes = min_t(size_t, bytes, fc->max_write - count);
835 if (iov_iter_fault_in_readable(ii, bytes))
839 page = grab_cache_page_write_begin(mapping, index, 0);
843 if (mapping_writably_mapped(mapping))
844 flush_dcache_page(page);
847 tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes);
849 flush_dcache_page(page);
851 mark_page_accessed(page);
853 iov_iter_advance(ii, tmp);
856 page_cache_release(page);
857 bytes = min(bytes, iov_iter_single_seg_count(ii));
862 req->pages[req->num_pages] = page;
868 if (offset == PAGE_CACHE_SIZE)
873 } while (iov_iter_count(ii) && count < fc->max_write &&
874 req->num_pages < FUSE_MAX_PAGES_PER_REQ && offset == 0);
876 return count > 0 ? count : err;
879 static ssize_t fuse_perform_write(struct file *file,
880 struct address_space *mapping,
881 struct iov_iter *ii, loff_t pos)
883 struct inode *inode = mapping->host;
884 struct fuse_conn *fc = get_fuse_conn(inode);
885 struct fuse_inode *fi = get_fuse_inode(inode);
889 if (is_bad_inode(inode))
892 if (inode->i_size < pos + iov_iter_count(ii))
893 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
896 struct fuse_req *req;
899 req = fuse_get_req(fc);
905 count = fuse_fill_write_pages(req, mapping, ii, pos);
911 num_written = fuse_send_write_pages(req, file, inode,
913 err = req->out.h.error;
918 /* break out of the loop on short write */
919 if (num_written != count)
923 fuse_put_request(fc, req);
924 } while (!err && iov_iter_count(ii));
927 fuse_write_update_size(inode, pos);
929 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
930 fuse_invalidate_attr(inode);
932 return res > 0 ? res : err;
935 static ssize_t fuse_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
936 unsigned long nr_segs, loff_t pos)
938 struct file *file = iocb->ki_filp;
939 struct address_space *mapping = file->f_mapping;
942 struct inode *inode = mapping->host;
946 WARN_ON(iocb->ki_pos != pos);
948 err = generic_segment_checks(iov, &nr_segs, &count, VERIFY_READ);
952 mutex_lock(&inode->i_mutex);
953 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
955 /* We can write back this queue in page reclaim */
956 current->backing_dev_info = mapping->backing_dev_info;
958 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
965 err = file_remove_suid(file);
969 file_update_time(file);
971 iov_iter_init(&i, iov, nr_segs, count, 0);
972 written = fuse_perform_write(file, mapping, &i, pos);
974 iocb->ki_pos = pos + written;
977 current->backing_dev_info = NULL;
978 mutex_unlock(&inode->i_mutex);
980 return written ? written : err;
983 static void fuse_release_user_pages(struct fuse_req *req, int write)
987 for (i = 0; i < req->num_pages; i++) {
988 struct page *page = req->pages[i];
990 set_page_dirty_lock(page);
995 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
996 size_t *nbytesp, int write)
998 size_t nbytes = *nbytesp;
999 unsigned long user_addr = (unsigned long) buf;
1000 unsigned offset = user_addr & ~PAGE_MASK;
1003 /* Special case for kernel I/O: can copy directly into the buffer */
1004 if (segment_eq(get_fs(), KERNEL_DS)) {
1006 req->in.args[1].value = (void *) user_addr;
1008 req->out.args[0].value = (void *) user_addr;
1013 nbytes = min_t(size_t, nbytes, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
1014 npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
1015 npages = clamp(npages, 1, FUSE_MAX_PAGES_PER_REQ);
1016 npages = get_user_pages_fast(user_addr, npages, !write, req->pages);
1020 req->num_pages = npages;
1021 req->page_offset = offset;
1024 req->in.argpages = 1;
1026 req->out.argpages = 1;
1028 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset;
1029 *nbytesp = min(*nbytesp, nbytes);
1034 ssize_t fuse_direct_io(struct file *file, const char __user *buf,
1035 size_t count, loff_t *ppos, int write)
1037 struct fuse_file *ff = file->private_data;
1038 struct fuse_conn *fc = ff->fc;
1039 size_t nmax = write ? fc->max_write : fc->max_read;
1042 struct fuse_req *req;
1044 req = fuse_get_req(fc);
1046 return PTR_ERR(req);
1050 fl_owner_t owner = current->files;
1051 size_t nbytes = min(count, nmax);
1052 int err = fuse_get_user_pages(req, buf, &nbytes, write);
1059 nres = fuse_send_write(req, file, pos, nbytes, owner);
1061 nres = fuse_send_read(req, file, pos, nbytes, owner);
1063 fuse_release_user_pages(req, !write);
1064 if (req->out.h.error) {
1066 res = req->out.h.error;
1068 } else if (nres > nbytes) {
1079 fuse_put_request(fc, req);
1080 req = fuse_get_req(fc);
1086 fuse_put_request(fc, req);
1092 EXPORT_SYMBOL_GPL(fuse_direct_io);
1094 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
1095 size_t count, loff_t *ppos)
1098 struct inode *inode = file->f_path.dentry->d_inode;
1100 if (is_bad_inode(inode))
1103 res = fuse_direct_io(file, buf, count, ppos, 0);
1105 fuse_invalidate_attr(inode);
1110 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
1111 size_t count, loff_t *ppos)
1113 struct inode *inode = file->f_path.dentry->d_inode;
1116 if (is_bad_inode(inode))
1119 /* Don't allow parallel writes to the same file */
1120 mutex_lock(&inode->i_mutex);
1121 res = generic_write_checks(file, ppos, &count, 0);
1123 res = fuse_direct_io(file, buf, count, ppos, 1);
1125 fuse_write_update_size(inode, *ppos);
1127 mutex_unlock(&inode->i_mutex);
1129 fuse_invalidate_attr(inode);
1134 static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
1136 __free_page(req->pages[0]);
1137 fuse_file_put(req->ff, false);
1140 static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
1142 struct inode *inode = req->inode;
1143 struct fuse_inode *fi = get_fuse_inode(inode);
1144 struct backing_dev_info *bdi = inode->i_mapping->backing_dev_info;
1146 list_del(&req->writepages_entry);
1147 dec_bdi_stat(bdi, BDI_WRITEBACK);
1148 dec_zone_page_state(req->pages[0], NR_WRITEBACK_TEMP);
1149 bdi_writeout_inc(bdi);
1150 wake_up(&fi->page_waitq);
1153 /* Called under fc->lock, may release and reacquire it */
1154 static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req)
1155 __releases(fc->lock)
1156 __acquires(fc->lock)
1158 struct fuse_inode *fi = get_fuse_inode(req->inode);
1159 loff_t size = i_size_read(req->inode);
1160 struct fuse_write_in *inarg = &req->misc.write.in;
1165 if (inarg->offset + PAGE_CACHE_SIZE <= size) {
1166 inarg->size = PAGE_CACHE_SIZE;
1167 } else if (inarg->offset < size) {
1168 inarg->size = size & (PAGE_CACHE_SIZE - 1);
1170 /* Got truncated off completely */
1174 req->in.args[1].size = inarg->size;
1176 fuse_request_send_background_locked(fc, req);
1180 fuse_writepage_finish(fc, req);
1181 spin_unlock(&fc->lock);
1182 fuse_writepage_free(fc, req);
1183 fuse_put_request(fc, req);
1184 spin_lock(&fc->lock);
1188 * If fi->writectr is positive (no truncate or fsync going on) send
1189 * all queued writepage requests.
1191 * Called with fc->lock
1193 void fuse_flush_writepages(struct inode *inode)
1194 __releases(fc->lock)
1195 __acquires(fc->lock)
1197 struct fuse_conn *fc = get_fuse_conn(inode);
1198 struct fuse_inode *fi = get_fuse_inode(inode);
1199 struct fuse_req *req;
1201 while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
1202 req = list_entry(fi->queued_writes.next, struct fuse_req, list);
1203 list_del_init(&req->list);
1204 fuse_send_writepage(fc, req);
1208 static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
1210 struct inode *inode = req->inode;
1211 struct fuse_inode *fi = get_fuse_inode(inode);
1213 mapping_set_error(inode->i_mapping, req->out.h.error);
1214 spin_lock(&fc->lock);
1216 fuse_writepage_finish(fc, req);
1217 spin_unlock(&fc->lock);
1218 fuse_writepage_free(fc, req);
1221 static int fuse_writepage_locked(struct page *page)
1223 struct address_space *mapping = page->mapping;
1224 struct inode *inode = mapping->host;
1225 struct fuse_conn *fc = get_fuse_conn(inode);
1226 struct fuse_inode *fi = get_fuse_inode(inode);
1227 struct fuse_req *req;
1228 struct fuse_file *ff;
1229 struct page *tmp_page;
1231 set_page_writeback(page);
1233 req = fuse_request_alloc_nofs();
1237 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1241 spin_lock(&fc->lock);
1242 BUG_ON(list_empty(&fi->write_files));
1243 ff = list_entry(fi->write_files.next, struct fuse_file, write_entry);
1244 req->ff = fuse_file_get(ff);
1245 spin_unlock(&fc->lock);
1247 fuse_write_fill(req, ff, page_offset(page), 0);
1249 copy_highpage(tmp_page, page);
1250 req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
1251 req->in.argpages = 1;
1253 req->pages[0] = tmp_page;
1254 req->page_offset = 0;
1255 req->end = fuse_writepage_end;
1258 inc_bdi_stat(mapping->backing_dev_info, BDI_WRITEBACK);
1259 inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP);
1261 spin_lock(&fc->lock);
1262 list_add(&req->writepages_entry, &fi->writepages);
1263 list_add_tail(&req->list, &fi->queued_writes);
1264 fuse_flush_writepages(inode);
1265 spin_unlock(&fc->lock);
1267 end_page_writeback(page);
1272 fuse_request_free(req);
1274 end_page_writeback(page);
1278 static int fuse_writepage(struct page *page, struct writeback_control *wbc)
1282 err = fuse_writepage_locked(page);
1288 static int fuse_launder_page(struct page *page)
1291 if (clear_page_dirty_for_io(page)) {
1292 struct inode *inode = page->mapping->host;
1293 err = fuse_writepage_locked(page);
1295 fuse_wait_on_page_writeback(inode, page->index);
1301 * Write back dirty pages now, because there may not be any suitable
1304 static void fuse_vma_close(struct vm_area_struct *vma)
1306 filemap_write_and_wait(vma->vm_file->f_mapping);
1310 * Wait for writeback against this page to complete before allowing it
1311 * to be marked dirty again, and hence written back again, possibly
1312 * before the previous writepage completed.
1314 * Block here, instead of in ->writepage(), so that the userspace fs
1315 * can only block processes actually operating on the filesystem.
1317 * Otherwise unprivileged userspace fs would be able to block
1322 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
1324 static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1326 struct page *page = vmf->page;
1328 * Don't use page->mapping as it may become NULL from a
1329 * concurrent truncate.
1331 struct inode *inode = vma->vm_file->f_mapping->host;
1333 fuse_wait_on_page_writeback(inode, page->index);
1337 static const struct vm_operations_struct fuse_file_vm_ops = {
1338 .close = fuse_vma_close,
1339 .fault = filemap_fault,
1340 .page_mkwrite = fuse_page_mkwrite,
1343 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
1345 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) {
1346 struct inode *inode = file->f_dentry->d_inode;
1347 struct fuse_conn *fc = get_fuse_conn(inode);
1348 struct fuse_inode *fi = get_fuse_inode(inode);
1349 struct fuse_file *ff = file->private_data;
1351 * file may be written through mmap, so chain it onto the
1352 * inodes's write_file list
1354 spin_lock(&fc->lock);
1355 if (list_empty(&ff->write_entry))
1356 list_add(&ff->write_entry, &fi->write_files);
1357 spin_unlock(&fc->lock);
1359 file_accessed(file);
1360 vma->vm_ops = &fuse_file_vm_ops;
1364 static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma)
1366 /* Can't provide the coherency needed for MAP_SHARED */
1367 if (vma->vm_flags & VM_MAYSHARE)
1370 invalidate_inode_pages2(file->f_mapping);
1372 return generic_file_mmap(file, vma);
1375 static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
1376 struct file_lock *fl)
1378 switch (ffl->type) {
1384 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
1385 ffl->end < ffl->start)
1388 fl->fl_start = ffl->start;
1389 fl->fl_end = ffl->end;
1390 fl->fl_pid = ffl->pid;
1396 fl->fl_type = ffl->type;
1400 static void fuse_lk_fill(struct fuse_req *req, struct file *file,
1401 const struct file_lock *fl, int opcode, pid_t pid,
1404 struct inode *inode = file->f_path.dentry->d_inode;
1405 struct fuse_conn *fc = get_fuse_conn(inode);
1406 struct fuse_file *ff = file->private_data;
1407 struct fuse_lk_in *arg = &req->misc.lk_in;
1410 arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
1411 arg->lk.start = fl->fl_start;
1412 arg->lk.end = fl->fl_end;
1413 arg->lk.type = fl->fl_type;
1416 arg->lk_flags |= FUSE_LK_FLOCK;
1417 req->in.h.opcode = opcode;
1418 req->in.h.nodeid = get_node_id(inode);
1419 req->in.numargs = 1;
1420 req->in.args[0].size = sizeof(*arg);
1421 req->in.args[0].value = arg;
1424 static int fuse_getlk(struct file *file, struct file_lock *fl)
1426 struct inode *inode = file->f_path.dentry->d_inode;
1427 struct fuse_conn *fc = get_fuse_conn(inode);
1428 struct fuse_req *req;
1429 struct fuse_lk_out outarg;
1432 req = fuse_get_req(fc);
1434 return PTR_ERR(req);
1436 fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0);
1437 req->out.numargs = 1;
1438 req->out.args[0].size = sizeof(outarg);
1439 req->out.args[0].value = &outarg;
1440 fuse_request_send(fc, req);
1441 err = req->out.h.error;
1442 fuse_put_request(fc, req);
1444 err = convert_fuse_file_lock(&outarg.lk, fl);
1449 static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
1451 struct inode *inode = file->f_path.dentry->d_inode;
1452 struct fuse_conn *fc = get_fuse_conn(inode);
1453 struct fuse_req *req;
1454 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
1455 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
1458 if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
1459 /* NLM needs asynchronous locks, which we don't support yet */
1463 /* Unlock on close is handled by the flush method */
1464 if (fl->fl_flags & FL_CLOSE)
1467 req = fuse_get_req(fc);
1469 return PTR_ERR(req);
1471 fuse_lk_fill(req, file, fl, opcode, pid, flock);
1472 fuse_request_send(fc, req);
1473 err = req->out.h.error;
1474 /* locking is restartable */
1477 fuse_put_request(fc, req);
1481 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
1483 struct inode *inode = file->f_path.dentry->d_inode;
1484 struct fuse_conn *fc = get_fuse_conn(inode);
1487 if (cmd == F_CANCELLK) {
1489 } else if (cmd == F_GETLK) {
1491 posix_test_lock(file, fl);
1494 err = fuse_getlk(file, fl);
1497 err = posix_lock_file(file, fl, NULL);
1499 err = fuse_setlk(file, fl, 0);
1504 static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
1506 struct inode *inode = file->f_path.dentry->d_inode;
1507 struct fuse_conn *fc = get_fuse_conn(inode);
1511 err = flock_lock_file_wait(file, fl);
1513 struct fuse_file *ff = file->private_data;
1515 /* emulate flock with POSIX locks */
1516 fl->fl_owner = (fl_owner_t) file;
1518 err = fuse_setlk(file, fl, 1);
1524 static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
1526 struct inode *inode = mapping->host;
1527 struct fuse_conn *fc = get_fuse_conn(inode);
1528 struct fuse_req *req;
1529 struct fuse_bmap_in inarg;
1530 struct fuse_bmap_out outarg;
1533 if (!inode->i_sb->s_bdev || fc->no_bmap)
1536 req = fuse_get_req(fc);
1540 memset(&inarg, 0, sizeof(inarg));
1541 inarg.block = block;
1542 inarg.blocksize = inode->i_sb->s_blocksize;
1543 req->in.h.opcode = FUSE_BMAP;
1544 req->in.h.nodeid = get_node_id(inode);
1545 req->in.numargs = 1;
1546 req->in.args[0].size = sizeof(inarg);
1547 req->in.args[0].value = &inarg;
1548 req->out.numargs = 1;
1549 req->out.args[0].size = sizeof(outarg);
1550 req->out.args[0].value = &outarg;
1551 fuse_request_send(fc, req);
1552 err = req->out.h.error;
1553 fuse_put_request(fc, req);
1557 return err ? 0 : outarg.block;
1560 static loff_t fuse_file_llseek(struct file *file, loff_t offset, int origin)
1563 struct inode *inode = file->f_path.dentry->d_inode;
1565 mutex_lock(&inode->i_mutex);
1566 if (origin != SEEK_CUR && origin != SEEK_SET) {
1567 retval = fuse_update_attributes(inode, NULL, file, NULL);
1574 offset += i_size_read(inode);
1578 retval = file->f_pos;
1581 offset += file->f_pos;
1584 if (offset >= i_size_read(inode)) {
1590 if (offset >= i_size_read(inode)) {
1594 offset = i_size_read(inode);
1598 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
1599 if (offset != file->f_pos) {
1600 file->f_pos = offset;
1601 file->f_version = 0;
1606 mutex_unlock(&inode->i_mutex);
1610 static int fuse_ioctl_copy_user(struct page **pages, struct iovec *iov,
1611 unsigned int nr_segs, size_t bytes, bool to_user)
1619 iov_iter_init(&ii, iov, nr_segs, bytes, 0);
1621 while (iov_iter_count(&ii)) {
1622 struct page *page = pages[page_idx++];
1623 size_t todo = min_t(size_t, PAGE_SIZE, iov_iter_count(&ii));
1629 char __user *uaddr = ii.iov->iov_base + ii.iov_offset;
1630 size_t iov_len = ii.iov->iov_len - ii.iov_offset;
1631 size_t copy = min(todo, iov_len);
1635 left = copy_from_user(kaddr, uaddr, copy);
1637 left = copy_to_user(uaddr, kaddr, copy);
1642 iov_iter_advance(&ii, copy);
1654 * CUSE servers compiled on 32bit broke on 64bit kernels because the
1655 * ABI was defined to be 'struct iovec' which is different on 32bit
1656 * and 64bit. Fortunately we can determine which structure the server
1657 * used from the size of the reply.
1659 static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
1660 size_t transferred, unsigned count,
1663 #ifdef CONFIG_COMPAT
1664 if (count * sizeof(struct compat_iovec) == transferred) {
1665 struct compat_iovec *ciov = src;
1669 * With this interface a 32bit server cannot support
1670 * non-compat (i.e. ones coming from 64bit apps) ioctl
1676 for (i = 0; i < count; i++) {
1677 dst[i].iov_base = compat_ptr(ciov[i].iov_base);
1678 dst[i].iov_len = ciov[i].iov_len;
1684 if (count * sizeof(struct iovec) != transferred)
1687 memcpy(dst, src, transferred);
1691 /* Make sure iov_length() won't overflow */
1692 static int fuse_verify_ioctl_iov(struct iovec *iov, size_t count)
1695 u32 max = FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT;
1697 for (n = 0; n < count; n++, iov++) {
1698 if (iov->iov_len > (size_t) max)
1700 max -= iov->iov_len;
1705 static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
1706 void *src, size_t transferred, unsigned count,
1710 struct fuse_ioctl_iovec *fiov = src;
1712 if (fc->minor < 16) {
1713 return fuse_copy_ioctl_iovec_old(dst, src, transferred,
1717 if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
1720 for (i = 0; i < count; i++) {
1721 /* Did the server supply an inappropriate value? */
1722 if (fiov[i].base != (unsigned long) fiov[i].base ||
1723 fiov[i].len != (unsigned long) fiov[i].len)
1726 dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
1727 dst[i].iov_len = (size_t) fiov[i].len;
1729 #ifdef CONFIG_COMPAT
1731 (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
1732 (compat_size_t) dst[i].iov_len != fiov[i].len))
1742 * For ioctls, there is no generic way to determine how much memory
1743 * needs to be read and/or written. Furthermore, ioctls are allowed
1744 * to dereference the passed pointer, so the parameter requires deep
1745 * copying but FUSE has no idea whatsoever about what to copy in or
1748 * This is solved by allowing FUSE server to retry ioctl with
1749 * necessary in/out iovecs. Let's assume the ioctl implementation
1750 * needs to read in the following structure.
1757 * On the first callout to FUSE server, inarg->in_size and
1758 * inarg->out_size will be NULL; then, the server completes the ioctl
1759 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
1760 * the actual iov array to
1762 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } }
1764 * which tells FUSE to copy in the requested area and retry the ioctl.
1765 * On the second round, the server has access to the structure and
1766 * from that it can tell what to look for next, so on the invocation,
1767 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
1769 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) },
1770 * { .iov_base = a.buf, .iov_len = a.buflen } }
1772 * FUSE will copy both struct a and the pointed buffer from the
1773 * process doing the ioctl and retry ioctl with both struct a and the
1776 * This time, FUSE server has everything it needs and completes ioctl
1777 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
1779 * Copying data out works the same way.
1781 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
1782 * automatically initializes in and out iovs by decoding @cmd with
1783 * _IOC_* macros and the server is not allowed to request RETRY. This
1784 * limits ioctl data transfers to well-formed ioctls and is the forced
1785 * behavior for all FUSE servers.
1787 long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1790 struct fuse_file *ff = file->private_data;
1791 struct fuse_conn *fc = ff->fc;
1792 struct fuse_ioctl_in inarg = {
1798 struct fuse_ioctl_out outarg;
1799 struct fuse_req *req = NULL;
1800 struct page **pages = NULL;
1801 struct iovec *iov_page = NULL;
1802 struct iovec *in_iov = NULL, *out_iov = NULL;
1803 unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages;
1804 size_t in_size, out_size, transferred;
1807 #if BITS_PER_LONG == 32
1808 inarg.flags |= FUSE_IOCTL_32BIT;
1810 if (flags & FUSE_IOCTL_COMPAT)
1811 inarg.flags |= FUSE_IOCTL_32BIT;
1814 /* assume all the iovs returned by client always fits in a page */
1815 BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
1818 pages = kzalloc(sizeof(pages[0]) * FUSE_MAX_PAGES_PER_REQ, GFP_KERNEL);
1819 iov_page = (struct iovec *) __get_free_page(GFP_KERNEL);
1820 if (!pages || !iov_page)
1824 * If restricted, initialize IO parameters as encoded in @cmd.
1825 * RETRY from server is not allowed.
1827 if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
1828 struct iovec *iov = iov_page;
1830 iov->iov_base = (void __user *)arg;
1831 iov->iov_len = _IOC_SIZE(cmd);
1833 if (_IOC_DIR(cmd) & _IOC_WRITE) {
1838 if (_IOC_DIR(cmd) & _IOC_READ) {
1845 inarg.in_size = in_size = iov_length(in_iov, in_iovs);
1846 inarg.out_size = out_size = iov_length(out_iov, out_iovs);
1849 * Out data can be used either for actual out data or iovs,
1850 * make sure there always is at least one page.
1852 out_size = max_t(size_t, out_size, PAGE_SIZE);
1853 max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
1855 /* make sure there are enough buffer pages and init request with them */
1857 if (max_pages > FUSE_MAX_PAGES_PER_REQ)
1859 while (num_pages < max_pages) {
1860 pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
1861 if (!pages[num_pages])
1866 req = fuse_get_req(fc);
1872 memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages);
1873 req->num_pages = num_pages;
1875 /* okay, let's send it to the client */
1876 req->in.h.opcode = FUSE_IOCTL;
1877 req->in.h.nodeid = ff->nodeid;
1878 req->in.numargs = 1;
1879 req->in.args[0].size = sizeof(inarg);
1880 req->in.args[0].value = &inarg;
1883 req->in.args[1].size = in_size;
1884 req->in.argpages = 1;
1886 err = fuse_ioctl_copy_user(pages, in_iov, in_iovs, in_size,
1892 req->out.numargs = 2;
1893 req->out.args[0].size = sizeof(outarg);
1894 req->out.args[0].value = &outarg;
1895 req->out.args[1].size = out_size;
1896 req->out.argpages = 1;
1897 req->out.argvar = 1;
1899 fuse_request_send(fc, req);
1900 err = req->out.h.error;
1901 transferred = req->out.args[1].size;
1902 fuse_put_request(fc, req);
1907 /* did it ask for retry? */
1908 if (outarg.flags & FUSE_IOCTL_RETRY) {
1911 /* no retry if in restricted mode */
1913 if (!(flags & FUSE_IOCTL_UNRESTRICTED))
1916 in_iovs = outarg.in_iovs;
1917 out_iovs = outarg.out_iovs;
1920 * Make sure things are in boundary, separate checks
1921 * are to protect against overflow.
1924 if (in_iovs > FUSE_IOCTL_MAX_IOV ||
1925 out_iovs > FUSE_IOCTL_MAX_IOV ||
1926 in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
1929 vaddr = kmap_atomic(pages[0], KM_USER0);
1930 err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr,
1931 transferred, in_iovs + out_iovs,
1932 (flags & FUSE_IOCTL_COMPAT) != 0);
1933 kunmap_atomic(vaddr, KM_USER0);
1938 out_iov = in_iov + in_iovs;
1940 err = fuse_verify_ioctl_iov(in_iov, in_iovs);
1944 err = fuse_verify_ioctl_iov(out_iov, out_iovs);
1952 if (transferred > inarg.out_size)
1955 err = fuse_ioctl_copy_user(pages, out_iov, out_iovs, transferred, true);
1958 fuse_put_request(fc, req);
1959 free_page((unsigned long) iov_page);
1961 __free_page(pages[--num_pages]);
1964 return err ? err : outarg.result;
1966 EXPORT_SYMBOL_GPL(fuse_do_ioctl);
1968 static long fuse_file_ioctl_common(struct file *file, unsigned int cmd,
1969 unsigned long arg, unsigned int flags)
1971 struct inode *inode = file->f_dentry->d_inode;
1972 struct fuse_conn *fc = get_fuse_conn(inode);
1974 if (!fuse_allow_task(fc, current))
1977 if (is_bad_inode(inode))
1980 return fuse_do_ioctl(file, cmd, arg, flags);
1983 static long fuse_file_ioctl(struct file *file, unsigned int cmd,
1986 return fuse_file_ioctl_common(file, cmd, arg, 0);
1989 static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
1992 return fuse_file_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
1996 * All files which have been polled are linked to RB tree
1997 * fuse_conn->polled_files which is indexed by kh. Walk the tree and
1998 * find the matching one.
2000 static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh,
2001 struct rb_node **parent_out)
2003 struct rb_node **link = &fc->polled_files.rb_node;
2004 struct rb_node *last = NULL;
2007 struct fuse_file *ff;
2010 ff = rb_entry(last, struct fuse_file, polled_node);
2013 link = &last->rb_left;
2014 else if (kh > ff->kh)
2015 link = &last->rb_right;
2026 * The file is about to be polled. Make sure it's on the polled_files
2027 * RB tree. Note that files once added to the polled_files tree are
2028 * not removed before the file is released. This is because a file
2029 * polled once is likely to be polled again.
2031 static void fuse_register_polled_file(struct fuse_conn *fc,
2032 struct fuse_file *ff)
2034 spin_lock(&fc->lock);
2035 if (RB_EMPTY_NODE(&ff->polled_node)) {
2036 struct rb_node **link, *parent;
2038 link = fuse_find_polled_node(fc, ff->kh, &parent);
2040 rb_link_node(&ff->polled_node, parent, link);
2041 rb_insert_color(&ff->polled_node, &fc->polled_files);
2043 spin_unlock(&fc->lock);
2046 unsigned fuse_file_poll(struct file *file, poll_table *wait)
2048 struct fuse_file *ff = file->private_data;
2049 struct fuse_conn *fc = ff->fc;
2050 struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh };
2051 struct fuse_poll_out outarg;
2052 struct fuse_req *req;
2056 return DEFAULT_POLLMASK;
2058 poll_wait(file, &ff->poll_wait, wait);
2061 * Ask for notification iff there's someone waiting for it.
2062 * The client may ignore the flag and always notify.
2064 if (waitqueue_active(&ff->poll_wait)) {
2065 inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY;
2066 fuse_register_polled_file(fc, ff);
2069 req = fuse_get_req(fc);
2073 req->in.h.opcode = FUSE_POLL;
2074 req->in.h.nodeid = ff->nodeid;
2075 req->in.numargs = 1;
2076 req->in.args[0].size = sizeof(inarg);
2077 req->in.args[0].value = &inarg;
2078 req->out.numargs = 1;
2079 req->out.args[0].size = sizeof(outarg);
2080 req->out.args[0].value = &outarg;
2081 fuse_request_send(fc, req);
2082 err = req->out.h.error;
2083 fuse_put_request(fc, req);
2086 return outarg.revents;
2087 if (err == -ENOSYS) {
2089 return DEFAULT_POLLMASK;
2093 EXPORT_SYMBOL_GPL(fuse_file_poll);
2096 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
2097 * wakes up the poll waiters.
2099 int fuse_notify_poll_wakeup(struct fuse_conn *fc,
2100 struct fuse_notify_poll_wakeup_out *outarg)
2102 u64 kh = outarg->kh;
2103 struct rb_node **link;
2105 spin_lock(&fc->lock);
2107 link = fuse_find_polled_node(fc, kh, NULL);
2109 struct fuse_file *ff;
2111 ff = rb_entry(*link, struct fuse_file, polled_node);
2112 wake_up_interruptible_sync(&ff->poll_wait);
2115 spin_unlock(&fc->lock);
2119 static const struct file_operations fuse_file_operations = {
2120 .llseek = fuse_file_llseek,
2121 .read = do_sync_read,
2122 .aio_read = fuse_file_aio_read,
2123 .write = do_sync_write,
2124 .aio_write = fuse_file_aio_write,
2125 .mmap = fuse_file_mmap,
2127 .flush = fuse_flush,
2128 .release = fuse_release,
2129 .fsync = fuse_fsync,
2130 .lock = fuse_file_lock,
2131 .flock = fuse_file_flock,
2132 .splice_read = generic_file_splice_read,
2133 .unlocked_ioctl = fuse_file_ioctl,
2134 .compat_ioctl = fuse_file_compat_ioctl,
2135 .poll = fuse_file_poll,
2138 static const struct file_operations fuse_direct_io_file_operations = {
2139 .llseek = fuse_file_llseek,
2140 .read = fuse_direct_read,
2141 .write = fuse_direct_write,
2142 .mmap = fuse_direct_mmap,
2144 .flush = fuse_flush,
2145 .release = fuse_release,
2146 .fsync = fuse_fsync,
2147 .lock = fuse_file_lock,
2148 .flock = fuse_file_flock,
2149 .unlocked_ioctl = fuse_file_ioctl,
2150 .compat_ioctl = fuse_file_compat_ioctl,
2151 .poll = fuse_file_poll,
2152 /* no splice_read */
2155 static const struct address_space_operations fuse_file_aops = {
2156 .readpage = fuse_readpage,
2157 .writepage = fuse_writepage,
2158 .launder_page = fuse_launder_page,
2159 .readpages = fuse_readpages,
2160 .set_page_dirty = __set_page_dirty_nobuffers,
2164 void fuse_init_file_inode(struct inode *inode)
2166 inode->i_fop = &fuse_file_operations;
2167 inode->i_data.a_ops = &fuse_file_aops;