2 FUSE: Filesystem in Userspace
3 Copyright (C) 2001-2006 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>
16 static const struct file_operations fuse_direct_io_file_operations;
18 static int fuse_send_open(struct inode *inode, struct file *file, int isdir,
19 struct fuse_open_out *outargp)
21 struct fuse_conn *fc = get_fuse_conn(inode);
22 struct fuse_open_in inarg;
26 req = fuse_get_req(fc);
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 = isdir ? FUSE_OPENDIR : FUSE_OPEN;
35 req->in.h.nodeid = get_node_id(inode);
37 req->in.args[0].size = sizeof(inarg);
38 req->in.args[0].value = &inarg;
40 req->out.args[0].size = sizeof(*outargp);
41 req->out.args[0].value = outargp;
42 request_send(fc, req);
43 err = req->out.h.error;
44 fuse_put_request(fc, req);
49 struct fuse_file *fuse_file_alloc(void)
52 ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
54 ff->reserved_req = fuse_request_alloc();
55 if (!ff->reserved_req) {
59 INIT_LIST_HEAD(&ff->write_entry);
60 atomic_set(&ff->count, 0);
66 void fuse_file_free(struct fuse_file *ff)
68 fuse_request_free(ff->reserved_req);
72 static struct fuse_file *fuse_file_get(struct fuse_file *ff)
74 atomic_inc(&ff->count);
78 static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
80 dput(req->misc.release.dentry);
81 mntput(req->misc.release.vfsmount);
82 fuse_put_request(fc, req);
85 static void fuse_file_put(struct fuse_file *ff)
87 if (atomic_dec_and_test(&ff->count)) {
88 struct fuse_req *req = ff->reserved_req;
89 struct inode *inode = req->misc.release.dentry->d_inode;
90 struct fuse_conn *fc = get_fuse_conn(inode);
91 req->end = fuse_release_end;
92 request_send_background(fc, req);
97 void fuse_finish_open(struct inode *inode, struct file *file,
98 struct fuse_file *ff, struct fuse_open_out *outarg)
100 if (outarg->open_flags & FOPEN_DIRECT_IO)
101 file->f_op = &fuse_direct_io_file_operations;
102 if (!(outarg->open_flags & FOPEN_KEEP_CACHE))
103 invalidate_inode_pages2(inode->i_mapping);
105 file->private_data = fuse_file_get(ff);
108 int fuse_open_common(struct inode *inode, struct file *file, int isdir)
110 struct fuse_open_out outarg;
111 struct fuse_file *ff;
114 /* VFS checks this, but only _after_ ->open() */
115 if (file->f_flags & O_DIRECT)
118 err = generic_file_open(inode, file);
122 ff = fuse_file_alloc();
126 err = fuse_send_open(inode, file, isdir, &outarg);
131 outarg.open_flags &= ~FOPEN_DIRECT_IO;
132 fuse_finish_open(inode, file, ff, &outarg);
138 void fuse_release_fill(struct fuse_file *ff, u64 nodeid, int flags, int opcode)
140 struct fuse_req *req = ff->reserved_req;
141 struct fuse_release_in *inarg = &req->misc.release.in;
144 inarg->flags = flags;
145 req->in.h.opcode = opcode;
146 req->in.h.nodeid = nodeid;
148 req->in.args[0].size = sizeof(struct fuse_release_in);
149 req->in.args[0].value = inarg;
152 int fuse_release_common(struct inode *inode, struct file *file, int isdir)
154 struct fuse_file *ff = file->private_data;
156 struct fuse_conn *fc = get_fuse_conn(inode);
157 struct fuse_req *req = ff->reserved_req;
159 fuse_release_fill(ff, get_node_id(inode), file->f_flags,
160 isdir ? FUSE_RELEASEDIR : FUSE_RELEASE);
162 /* Hold vfsmount and dentry until release is finished */
163 req->misc.release.vfsmount = mntget(file->f_path.mnt);
164 req->misc.release.dentry = dget(file->f_path.dentry);
166 spin_lock(&fc->lock);
167 list_del(&ff->write_entry);
168 spin_unlock(&fc->lock);
170 * Normally this will send the RELEASE request,
171 * however if some asynchronous READ or WRITE requests
172 * are outstanding, the sending will be delayed
177 /* Return value is ignored by VFS */
181 static int fuse_open(struct inode *inode, struct file *file)
183 return fuse_open_common(inode, file, 0);
186 static int fuse_release(struct inode *inode, struct file *file)
188 return fuse_release_common(inode, file, 0);
192 * Scramble the ID space with XTEA, so that the value of the files_struct
193 * pointer is not exposed to userspace.
195 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
197 u32 *k = fc->scramble_key;
198 u64 v = (unsigned long) id;
204 for (i = 0; i < 32; i++) {
205 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
207 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
210 return (u64) v0 + ((u64) v1 << 32);
214 * Check if page is under writeback
216 * This is currently done by walking the list of writepage requests
217 * for the inode, which can be pretty inefficient.
219 static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
221 struct fuse_conn *fc = get_fuse_conn(inode);
222 struct fuse_inode *fi = get_fuse_inode(inode);
223 struct fuse_req *req;
226 spin_lock(&fc->lock);
227 list_for_each_entry(req, &fi->writepages, writepages_entry) {
230 BUG_ON(req->inode != inode);
231 curr_index = req->misc.write.in.offset >> PAGE_CACHE_SHIFT;
232 if (curr_index == index) {
237 spin_unlock(&fc->lock);
243 * Wait for page writeback to be completed.
245 * Since fuse doesn't rely on the VM writeback tracking, this has to
246 * use some other means.
248 static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
250 struct fuse_inode *fi = get_fuse_inode(inode);
252 wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
256 static int fuse_flush(struct file *file, fl_owner_t id)
258 struct inode *inode = file->f_path.dentry->d_inode;
259 struct fuse_conn *fc = get_fuse_conn(inode);
260 struct fuse_file *ff = file->private_data;
261 struct fuse_req *req;
262 struct fuse_flush_in inarg;
265 if (is_bad_inode(inode))
271 req = fuse_get_req_nofail(fc, file);
272 memset(&inarg, 0, sizeof(inarg));
274 inarg.lock_owner = fuse_lock_owner_id(fc, id);
275 req->in.h.opcode = FUSE_FLUSH;
276 req->in.h.nodeid = get_node_id(inode);
278 req->in.args[0].size = sizeof(inarg);
279 req->in.args[0].value = &inarg;
281 request_send(fc, req);
282 err = req->out.h.error;
283 fuse_put_request(fc, req);
284 if (err == -ENOSYS) {
292 * Wait for all pending writepages on the inode to finish.
294 * This is currently done by blocking further writes with FUSE_NOWRITE
295 * and waiting for all sent writes to complete.
297 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
298 * could conflict with truncation.
300 static void fuse_sync_writes(struct inode *inode)
302 fuse_set_nowrite(inode);
303 fuse_release_nowrite(inode);
306 int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
309 struct inode *inode = de->d_inode;
310 struct fuse_conn *fc = get_fuse_conn(inode);
311 struct fuse_file *ff = file->private_data;
312 struct fuse_req *req;
313 struct fuse_fsync_in inarg;
316 if (is_bad_inode(inode))
319 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
323 * Start writeback against all dirty pages of the inode, then
324 * wait for all outstanding writes, before sending the FSYNC
327 err = write_inode_now(inode, 0);
331 fuse_sync_writes(inode);
333 req = fuse_get_req(fc);
337 memset(&inarg, 0, sizeof(inarg));
339 inarg.fsync_flags = datasync ? 1 : 0;
340 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
341 req->in.h.nodeid = get_node_id(inode);
343 req->in.args[0].size = sizeof(inarg);
344 req->in.args[0].value = &inarg;
345 request_send(fc, req);
346 err = req->out.h.error;
347 fuse_put_request(fc, req);
348 if (err == -ENOSYS) {
358 static int fuse_fsync(struct file *file, struct dentry *de, int datasync)
360 return fuse_fsync_common(file, de, datasync, 0);
363 void fuse_read_fill(struct fuse_req *req, struct file *file,
364 struct inode *inode, loff_t pos, size_t count, int opcode)
366 struct fuse_read_in *inarg = &req->misc.read_in;
367 struct fuse_file *ff = file->private_data;
372 inarg->flags = file->f_flags;
373 req->in.h.opcode = opcode;
374 req->in.h.nodeid = get_node_id(inode);
376 req->in.args[0].size = sizeof(struct fuse_read_in);
377 req->in.args[0].value = inarg;
378 req->out.argpages = 1;
380 req->out.numargs = 1;
381 req->out.args[0].size = count;
384 static size_t fuse_send_read(struct fuse_req *req, struct file *file,
385 struct inode *inode, loff_t pos, size_t count,
388 struct fuse_conn *fc = get_fuse_conn(inode);
390 fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
392 struct fuse_read_in *inarg = &req->misc.read_in;
394 inarg->read_flags |= FUSE_READ_LOCKOWNER;
395 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
397 request_send(fc, req);
398 return req->out.args[0].size;
401 static int fuse_readpage(struct file *file, struct page *page)
403 struct inode *inode = page->mapping->host;
404 struct fuse_conn *fc = get_fuse_conn(inode);
405 struct fuse_req *req;
409 if (is_bad_inode(inode))
413 * Page writeback can extend beyond the liftime of the
414 * page-cache page, so make sure we read a properly synced
417 fuse_wait_on_page_writeback(inode, page->index);
419 req = fuse_get_req(fc);
424 req->out.page_zeroing = 1;
426 req->pages[0] = page;
427 fuse_send_read(req, file, inode, page_offset(page), PAGE_CACHE_SIZE,
429 err = req->out.h.error;
430 fuse_put_request(fc, req);
432 SetPageUptodate(page);
433 fuse_invalidate_attr(inode); /* atime changed */
439 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
443 fuse_invalidate_attr(req->pages[0]->mapping->host); /* atime changed */
445 for (i = 0; i < req->num_pages; i++) {
446 struct page *page = req->pages[i];
447 if (!req->out.h.error)
448 SetPageUptodate(page);
454 fuse_file_put(req->ff);
455 fuse_put_request(fc, req);
458 static void fuse_send_readpages(struct fuse_req *req, struct file *file,
461 struct fuse_conn *fc = get_fuse_conn(inode);
462 loff_t pos = page_offset(req->pages[0]);
463 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
464 req->out.page_zeroing = 1;
465 fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
466 if (fc->async_read) {
467 struct fuse_file *ff = file->private_data;
468 req->ff = fuse_file_get(ff);
469 req->end = fuse_readpages_end;
470 request_send_background(fc, req);
472 request_send(fc, req);
473 fuse_readpages_end(fc, req);
477 struct fuse_fill_data {
478 struct fuse_req *req;
483 static int fuse_readpages_fill(void *_data, struct page *page)
485 struct fuse_fill_data *data = _data;
486 struct fuse_req *req = data->req;
487 struct inode *inode = data->inode;
488 struct fuse_conn *fc = get_fuse_conn(inode);
490 fuse_wait_on_page_writeback(inode, page->index);
492 if (req->num_pages &&
493 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
494 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
495 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
496 fuse_send_readpages(req, data->file, inode);
497 data->req = req = fuse_get_req(fc);
503 req->pages[req->num_pages] = page;
508 static int fuse_readpages(struct file *file, struct address_space *mapping,
509 struct list_head *pages, unsigned nr_pages)
511 struct inode *inode = mapping->host;
512 struct fuse_conn *fc = get_fuse_conn(inode);
513 struct fuse_fill_data data;
517 if (is_bad_inode(inode))
522 data.req = fuse_get_req(fc);
523 err = PTR_ERR(data.req);
524 if (IS_ERR(data.req))
527 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
529 if (data.req->num_pages)
530 fuse_send_readpages(data.req, file, inode);
532 fuse_put_request(fc, data.req);
538 static ssize_t fuse_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
539 unsigned long nr_segs, loff_t pos)
541 struct inode *inode = iocb->ki_filp->f_mapping->host;
543 if (pos + iov_length(iov, nr_segs) > i_size_read(inode)) {
546 * If trying to read past EOF, make sure the i_size
547 * attribute is up-to-date.
549 err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL);
554 return generic_file_aio_read(iocb, iov, nr_segs, pos);
557 static void fuse_write_fill(struct fuse_req *req, struct file *file,
558 struct fuse_file *ff, struct inode *inode,
559 loff_t pos, size_t count, int writepage)
561 struct fuse_conn *fc = get_fuse_conn(inode);
562 struct fuse_write_in *inarg = &req->misc.write.in;
563 struct fuse_write_out *outarg = &req->misc.write.out;
565 memset(inarg, 0, sizeof(struct fuse_write_in));
569 inarg->write_flags = writepage ? FUSE_WRITE_CACHE : 0;
570 inarg->flags = file ? file->f_flags : 0;
571 req->in.h.opcode = FUSE_WRITE;
572 req->in.h.nodeid = get_node_id(inode);
573 req->in.argpages = 1;
576 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
578 req->in.args[0].size = sizeof(struct fuse_write_in);
579 req->in.args[0].value = inarg;
580 req->in.args[1].size = count;
581 req->out.numargs = 1;
582 req->out.args[0].size = sizeof(struct fuse_write_out);
583 req->out.args[0].value = outarg;
586 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
587 struct inode *inode, loff_t pos, size_t count,
590 struct fuse_conn *fc = get_fuse_conn(inode);
591 fuse_write_fill(req, file, file->private_data, inode, pos, count, 0);
593 struct fuse_write_in *inarg = &req->misc.write.in;
594 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
595 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
597 request_send(fc, req);
598 return req->misc.write.out.size;
601 static int fuse_write_begin(struct file *file, struct address_space *mapping,
602 loff_t pos, unsigned len, unsigned flags,
603 struct page **pagep, void **fsdata)
605 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
607 *pagep = __grab_cache_page(mapping, index);
613 static int fuse_buffered_write(struct file *file, struct inode *inode,
614 loff_t pos, unsigned count, struct page *page)
618 struct fuse_conn *fc = get_fuse_conn(inode);
619 struct fuse_inode *fi = get_fuse_inode(inode);
620 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
621 struct fuse_req *req;
623 if (is_bad_inode(inode))
627 * Make sure writepages on the same page are not mixed up with
630 fuse_wait_on_page_writeback(inode, page->index);
632 req = fuse_get_req(fc);
637 req->pages[0] = page;
638 req->page_offset = offset;
639 nres = fuse_send_write(req, file, inode, pos, count, NULL);
640 err = req->out.h.error;
641 fuse_put_request(fc, req);
646 spin_lock(&fc->lock);
647 fi->attr_version = ++fc->attr_version;
648 if (pos > inode->i_size)
649 i_size_write(inode, pos);
650 spin_unlock(&fc->lock);
652 if (count == PAGE_CACHE_SIZE)
653 SetPageUptodate(page);
655 fuse_invalidate_attr(inode);
656 return err ? err : nres;
659 static int fuse_write_end(struct file *file, struct address_space *mapping,
660 loff_t pos, unsigned len, unsigned copied,
661 struct page *page, void *fsdata)
663 struct inode *inode = mapping->host;
667 res = fuse_buffered_write(file, inode, pos, copied, page);
670 page_cache_release(page);
674 static void fuse_release_user_pages(struct fuse_req *req, int write)
678 for (i = 0; i < req->num_pages; i++) {
679 struct page *page = req->pages[i];
681 set_page_dirty_lock(page);
686 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
687 unsigned nbytes, int write)
689 unsigned long user_addr = (unsigned long) buf;
690 unsigned offset = user_addr & ~PAGE_MASK;
693 /* This doesn't work with nfsd */
697 nbytes = min(nbytes, (unsigned) FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
698 npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
699 npages = min(max(npages, 1), FUSE_MAX_PAGES_PER_REQ);
700 down_read(¤t->mm->mmap_sem);
701 npages = get_user_pages(current, current->mm, user_addr, npages, write,
702 0, req->pages, NULL);
703 up_read(¤t->mm->mmap_sem);
707 req->num_pages = npages;
708 req->page_offset = offset;
712 static ssize_t fuse_direct_io(struct file *file, const char __user *buf,
713 size_t count, loff_t *ppos, int write)
715 struct inode *inode = file->f_path.dentry->d_inode;
716 struct fuse_conn *fc = get_fuse_conn(inode);
717 size_t nmax = write ? fc->max_write : fc->max_read;
720 struct fuse_req *req;
722 if (is_bad_inode(inode))
725 req = fuse_get_req(fc);
731 size_t nbytes = min(count, nmax);
732 int err = fuse_get_user_pages(req, buf, nbytes, !write);
737 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset;
738 nbytes = min(count, nbytes);
740 nres = fuse_send_write(req, file, inode, pos, nbytes,
743 nres = fuse_send_read(req, file, inode, pos, nbytes,
745 fuse_release_user_pages(req, !write);
746 if (req->out.h.error) {
748 res = req->out.h.error;
750 } else if (nres > nbytes) {
761 fuse_put_request(fc, req);
762 req = fuse_get_req(fc);
767 fuse_put_request(fc, req);
770 spin_lock(&fc->lock);
771 if (pos > inode->i_size)
772 i_size_write(inode, pos);
773 spin_unlock(&fc->lock);
777 fuse_invalidate_attr(inode);
782 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
783 size_t count, loff_t *ppos)
785 return fuse_direct_io(file, buf, count, ppos, 0);
788 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
789 size_t count, loff_t *ppos)
791 struct inode *inode = file->f_path.dentry->d_inode;
793 /* Don't allow parallel writes to the same file */
794 mutex_lock(&inode->i_mutex);
795 res = generic_write_checks(file, ppos, &count, 0);
797 res = fuse_direct_io(file, buf, count, ppos, 1);
798 mutex_unlock(&inode->i_mutex);
802 static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
804 __free_page(req->pages[0]);
805 fuse_file_put(req->ff);
806 fuse_put_request(fc, req);
809 static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
811 struct inode *inode = req->inode;
812 struct fuse_inode *fi = get_fuse_inode(inode);
813 struct backing_dev_info *bdi = inode->i_mapping->backing_dev_info;
815 list_del(&req->writepages_entry);
816 dec_bdi_stat(bdi, BDI_WRITEBACK);
817 dec_zone_page_state(req->pages[0], NR_WRITEBACK_TEMP);
818 bdi_writeout_inc(bdi);
819 wake_up(&fi->page_waitq);
822 /* Called under fc->lock, may release and reacquire it */
823 static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req)
825 struct fuse_inode *fi = get_fuse_inode(req->inode);
826 loff_t size = i_size_read(req->inode);
827 struct fuse_write_in *inarg = &req->misc.write.in;
832 if (inarg->offset + PAGE_CACHE_SIZE <= size) {
833 inarg->size = PAGE_CACHE_SIZE;
834 } else if (inarg->offset < size) {
835 inarg->size = size & (PAGE_CACHE_SIZE - 1);
837 /* Got truncated off completely */
841 req->in.args[1].size = inarg->size;
843 request_send_background_locked(fc, req);
847 fuse_writepage_finish(fc, req);
848 spin_unlock(&fc->lock);
849 fuse_writepage_free(fc, req);
850 spin_lock(&fc->lock);
854 * If fi->writectr is positive (no truncate or fsync going on) send
855 * all queued writepage requests.
857 * Called with fc->lock
859 void fuse_flush_writepages(struct inode *inode)
861 struct fuse_conn *fc = get_fuse_conn(inode);
862 struct fuse_inode *fi = get_fuse_inode(inode);
863 struct fuse_req *req;
865 while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
866 req = list_entry(fi->queued_writes.next, struct fuse_req, list);
867 list_del_init(&req->list);
868 fuse_send_writepage(fc, req);
872 static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
874 struct inode *inode = req->inode;
875 struct fuse_inode *fi = get_fuse_inode(inode);
877 mapping_set_error(inode->i_mapping, req->out.h.error);
878 spin_lock(&fc->lock);
880 fuse_writepage_finish(fc, req);
881 spin_unlock(&fc->lock);
882 fuse_writepage_free(fc, req);
885 static int fuse_writepage_locked(struct page *page)
887 struct address_space *mapping = page->mapping;
888 struct inode *inode = mapping->host;
889 struct fuse_conn *fc = get_fuse_conn(inode);
890 struct fuse_inode *fi = get_fuse_inode(inode);
891 struct fuse_req *req;
892 struct fuse_file *ff;
893 struct page *tmp_page;
895 set_page_writeback(page);
897 req = fuse_request_alloc_nofs();
901 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
905 spin_lock(&fc->lock);
906 BUG_ON(list_empty(&fi->write_files));
907 ff = list_entry(fi->write_files.next, struct fuse_file, write_entry);
908 req->ff = fuse_file_get(ff);
909 spin_unlock(&fc->lock);
911 fuse_write_fill(req, NULL, ff, inode, page_offset(page), 0, 1);
913 copy_highpage(tmp_page, page);
915 req->pages[0] = tmp_page;
916 req->page_offset = 0;
917 req->end = fuse_writepage_end;
920 inc_bdi_stat(mapping->backing_dev_info, BDI_WRITEBACK);
921 inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP);
922 end_page_writeback(page);
924 spin_lock(&fc->lock);
925 list_add(&req->writepages_entry, &fi->writepages);
926 list_add_tail(&req->list, &fi->queued_writes);
927 fuse_flush_writepages(inode);
928 spin_unlock(&fc->lock);
933 fuse_request_free(req);
935 end_page_writeback(page);
939 static int fuse_writepage(struct page *page, struct writeback_control *wbc)
943 err = fuse_writepage_locked(page);
949 static int fuse_launder_page(struct page *page)
952 if (clear_page_dirty_for_io(page)) {
953 struct inode *inode = page->mapping->host;
954 err = fuse_writepage_locked(page);
956 fuse_wait_on_page_writeback(inode, page->index);
962 * Write back dirty pages now, because there may not be any suitable
965 static void fuse_vma_close(struct vm_area_struct *vma)
967 filemap_write_and_wait(vma->vm_file->f_mapping);
971 * Wait for writeback against this page to complete before allowing it
972 * to be marked dirty again, and hence written back again, possibly
973 * before the previous writepage completed.
975 * Block here, instead of in ->writepage(), so that the userspace fs
976 * can only block processes actually operating on the filesystem.
978 * Otherwise unprivileged userspace fs would be able to block
983 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
985 static int fuse_page_mkwrite(struct vm_area_struct *vma, struct page *page)
988 * Don't use page->mapping as it may become NULL from a
989 * concurrent truncate.
991 struct inode *inode = vma->vm_file->f_mapping->host;
993 fuse_wait_on_page_writeback(inode, page->index);
997 static struct vm_operations_struct fuse_file_vm_ops = {
998 .close = fuse_vma_close,
999 .fault = filemap_fault,
1000 .page_mkwrite = fuse_page_mkwrite,
1003 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
1005 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) {
1006 struct inode *inode = file->f_dentry->d_inode;
1007 struct fuse_conn *fc = get_fuse_conn(inode);
1008 struct fuse_inode *fi = get_fuse_inode(inode);
1009 struct fuse_file *ff = file->private_data;
1011 * file may be written through mmap, so chain it onto the
1012 * inodes's write_file list
1014 spin_lock(&fc->lock);
1015 if (list_empty(&ff->write_entry))
1016 list_add(&ff->write_entry, &fi->write_files);
1017 spin_unlock(&fc->lock);
1019 file_accessed(file);
1020 vma->vm_ops = &fuse_file_vm_ops;
1024 static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
1025 struct file_lock *fl)
1027 switch (ffl->type) {
1033 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
1034 ffl->end < ffl->start)
1037 fl->fl_start = ffl->start;
1038 fl->fl_end = ffl->end;
1039 fl->fl_pid = ffl->pid;
1045 fl->fl_type = ffl->type;
1049 static void fuse_lk_fill(struct fuse_req *req, struct file *file,
1050 const struct file_lock *fl, int opcode, pid_t pid,
1053 struct inode *inode = file->f_path.dentry->d_inode;
1054 struct fuse_conn *fc = get_fuse_conn(inode);
1055 struct fuse_file *ff = file->private_data;
1056 struct fuse_lk_in *arg = &req->misc.lk_in;
1059 arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
1060 arg->lk.start = fl->fl_start;
1061 arg->lk.end = fl->fl_end;
1062 arg->lk.type = fl->fl_type;
1065 arg->lk_flags |= FUSE_LK_FLOCK;
1066 req->in.h.opcode = opcode;
1067 req->in.h.nodeid = get_node_id(inode);
1068 req->in.numargs = 1;
1069 req->in.args[0].size = sizeof(*arg);
1070 req->in.args[0].value = arg;
1073 static int fuse_getlk(struct file *file, struct file_lock *fl)
1075 struct inode *inode = file->f_path.dentry->d_inode;
1076 struct fuse_conn *fc = get_fuse_conn(inode);
1077 struct fuse_req *req;
1078 struct fuse_lk_out outarg;
1081 req = fuse_get_req(fc);
1083 return PTR_ERR(req);
1085 fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0);
1086 req->out.numargs = 1;
1087 req->out.args[0].size = sizeof(outarg);
1088 req->out.args[0].value = &outarg;
1089 request_send(fc, req);
1090 err = req->out.h.error;
1091 fuse_put_request(fc, req);
1093 err = convert_fuse_file_lock(&outarg.lk, fl);
1098 static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
1100 struct inode *inode = file->f_path.dentry->d_inode;
1101 struct fuse_conn *fc = get_fuse_conn(inode);
1102 struct fuse_req *req;
1103 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
1104 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
1107 /* Unlock on close is handled by the flush method */
1108 if (fl->fl_flags & FL_CLOSE)
1111 req = fuse_get_req(fc);
1113 return PTR_ERR(req);
1115 fuse_lk_fill(req, file, fl, opcode, pid, flock);
1116 request_send(fc, req);
1117 err = req->out.h.error;
1118 /* locking is restartable */
1121 fuse_put_request(fc, req);
1125 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
1127 struct inode *inode = file->f_path.dentry->d_inode;
1128 struct fuse_conn *fc = get_fuse_conn(inode);
1131 if (cmd == F_GETLK) {
1133 posix_test_lock(file, fl);
1136 err = fuse_getlk(file, fl);
1139 err = posix_lock_file_wait(file, fl);
1141 err = fuse_setlk(file, fl, 0);
1146 static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
1148 struct inode *inode = file->f_path.dentry->d_inode;
1149 struct fuse_conn *fc = get_fuse_conn(inode);
1153 err = flock_lock_file_wait(file, fl);
1155 /* emulate flock with POSIX locks */
1156 fl->fl_owner = (fl_owner_t) file;
1157 err = fuse_setlk(file, fl, 1);
1163 static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
1165 struct inode *inode = mapping->host;
1166 struct fuse_conn *fc = get_fuse_conn(inode);
1167 struct fuse_req *req;
1168 struct fuse_bmap_in inarg;
1169 struct fuse_bmap_out outarg;
1172 if (!inode->i_sb->s_bdev || fc->no_bmap)
1175 req = fuse_get_req(fc);
1179 memset(&inarg, 0, sizeof(inarg));
1180 inarg.block = block;
1181 inarg.blocksize = inode->i_sb->s_blocksize;
1182 req->in.h.opcode = FUSE_BMAP;
1183 req->in.h.nodeid = get_node_id(inode);
1184 req->in.numargs = 1;
1185 req->in.args[0].size = sizeof(inarg);
1186 req->in.args[0].value = &inarg;
1187 req->out.numargs = 1;
1188 req->out.args[0].size = sizeof(outarg);
1189 req->out.args[0].value = &outarg;
1190 request_send(fc, req);
1191 err = req->out.h.error;
1192 fuse_put_request(fc, req);
1196 return err ? 0 : outarg.block;
1199 static const struct file_operations fuse_file_operations = {
1200 .llseek = generic_file_llseek,
1201 .read = do_sync_read,
1202 .aio_read = fuse_file_aio_read,
1203 .write = do_sync_write,
1204 .aio_write = generic_file_aio_write,
1205 .mmap = fuse_file_mmap,
1207 .flush = fuse_flush,
1208 .release = fuse_release,
1209 .fsync = fuse_fsync,
1210 .lock = fuse_file_lock,
1211 .flock = fuse_file_flock,
1212 .splice_read = generic_file_splice_read,
1215 static const struct file_operations fuse_direct_io_file_operations = {
1216 .llseek = generic_file_llseek,
1217 .read = fuse_direct_read,
1218 .write = fuse_direct_write,
1220 .flush = fuse_flush,
1221 .release = fuse_release,
1222 .fsync = fuse_fsync,
1223 .lock = fuse_file_lock,
1224 .flock = fuse_file_flock,
1225 /* no mmap and splice_read */
1228 static const struct address_space_operations fuse_file_aops = {
1229 .readpage = fuse_readpage,
1230 .writepage = fuse_writepage,
1231 .launder_page = fuse_launder_page,
1232 .write_begin = fuse_write_begin,
1233 .write_end = fuse_write_end,
1234 .readpages = fuse_readpages,
1235 .set_page_dirty = __set_page_dirty_nobuffers,
1239 void fuse_init_file_inode(struct inode *inode)
1241 inode->i_fop = &fuse_file_operations;
1242 inode->i_data.a_ops = &fuse_file_aops;