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 | O_TRUNC);
32 req->in.h.opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
33 req->in.h.nodeid = get_node_id(inode);
35 req->in.args[0].size = sizeof(inarg);
36 req->in.args[0].value = &inarg;
38 req->out.args[0].size = sizeof(*outargp);
39 req->out.args[0].value = outargp;
40 request_send(fc, req);
41 err = req->out.h.error;
42 fuse_put_request(fc, req);
47 struct fuse_file *fuse_file_alloc(void)
50 ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
52 ff->reserved_req = fuse_request_alloc();
53 if (!ff->reserved_req) {
61 void fuse_file_free(struct fuse_file *ff)
63 fuse_request_free(ff->reserved_req);
67 void fuse_finish_open(struct inode *inode, struct file *file,
68 struct fuse_file *ff, struct fuse_open_out *outarg)
70 if (outarg->open_flags & FOPEN_DIRECT_IO)
71 file->f_op = &fuse_direct_io_file_operations;
72 if (!(outarg->open_flags & FOPEN_KEEP_CACHE))
73 invalidate_mapping_pages(inode->i_mapping, 0, -1);
75 file->private_data = ff;
78 int fuse_open_common(struct inode *inode, struct file *file, int isdir)
80 struct fuse_open_out outarg;
84 /* VFS checks this, but only _after_ ->open() */
85 if (file->f_flags & O_DIRECT)
88 err = generic_file_open(inode, file);
92 /* If opening the root node, no lookup has been performed on
93 it, so the attributes must be refreshed */
94 if (get_node_id(inode) == FUSE_ROOT_ID) {
95 err = fuse_do_getattr(inode);
100 ff = fuse_file_alloc();
104 err = fuse_send_open(inode, file, isdir, &outarg);
109 outarg.open_flags &= ~FOPEN_DIRECT_IO;
110 fuse_finish_open(inode, file, ff, &outarg);
116 struct fuse_req *fuse_release_fill(struct fuse_file *ff, u64 nodeid, int flags,
119 struct fuse_req *req = ff->reserved_req;
120 struct fuse_release_in *inarg = &req->misc.release_in;
123 inarg->flags = flags;
124 req->in.h.opcode = opcode;
125 req->in.h.nodeid = nodeid;
127 req->in.args[0].size = sizeof(struct fuse_release_in);
128 req->in.args[0].value = inarg;
134 int fuse_release_common(struct inode *inode, struct file *file, int isdir)
136 struct fuse_file *ff = file->private_data;
138 struct fuse_conn *fc = get_fuse_conn(inode);
139 struct fuse_req *req;
141 req = fuse_release_fill(ff, get_node_id(inode), file->f_flags,
142 isdir ? FUSE_RELEASEDIR : FUSE_RELEASE);
144 /* Hold vfsmount and dentry until release is finished */
145 req->vfsmount = mntget(file->f_path.mnt);
146 req->dentry = dget(file->f_path.dentry);
147 request_send_background(fc, req);
150 /* Return value is ignored by VFS */
154 static int fuse_open(struct inode *inode, struct file *file)
156 return fuse_open_common(inode, file, 0);
159 static int fuse_release(struct inode *inode, struct file *file)
161 return fuse_release_common(inode, file, 0);
165 * Scramble the ID space with XTEA, so that the value of the files_struct
166 * pointer is not exposed to userspace.
168 static u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
170 u32 *k = fc->scramble_key;
171 u64 v = (unsigned long) id;
177 for (i = 0; i < 32; i++) {
178 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
180 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
183 return (u64) v0 + ((u64) v1 << 32);
186 static int fuse_flush(struct file *file, fl_owner_t id)
188 struct inode *inode = file->f_path.dentry->d_inode;
189 struct fuse_conn *fc = get_fuse_conn(inode);
190 struct fuse_file *ff = file->private_data;
191 struct fuse_req *req;
192 struct fuse_flush_in inarg;
195 if (is_bad_inode(inode))
201 req = fuse_get_req_nofail(fc, file);
202 memset(&inarg, 0, sizeof(inarg));
204 inarg.lock_owner = fuse_lock_owner_id(fc, id);
205 req->in.h.opcode = FUSE_FLUSH;
206 req->in.h.nodeid = get_node_id(inode);
208 req->in.args[0].size = sizeof(inarg);
209 req->in.args[0].value = &inarg;
211 request_send(fc, req);
212 err = req->out.h.error;
213 fuse_put_request(fc, req);
214 if (err == -ENOSYS) {
221 int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
224 struct inode *inode = de->d_inode;
225 struct fuse_conn *fc = get_fuse_conn(inode);
226 struct fuse_file *ff = file->private_data;
227 struct fuse_req *req;
228 struct fuse_fsync_in inarg;
231 if (is_bad_inode(inode))
234 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
237 req = fuse_get_req(fc);
241 memset(&inarg, 0, sizeof(inarg));
243 inarg.fsync_flags = datasync ? 1 : 0;
244 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
245 req->in.h.nodeid = get_node_id(inode);
247 req->in.args[0].size = sizeof(inarg);
248 req->in.args[0].value = &inarg;
249 request_send(fc, req);
250 err = req->out.h.error;
251 fuse_put_request(fc, req);
252 if (err == -ENOSYS) {
262 static int fuse_fsync(struct file *file, struct dentry *de, int datasync)
264 return fuse_fsync_common(file, de, datasync, 0);
267 void fuse_read_fill(struct fuse_req *req, struct file *file,
268 struct inode *inode, loff_t pos, size_t count, int opcode)
270 struct fuse_file *ff = file->private_data;
271 struct fuse_read_in *inarg = &req->misc.read_in;
276 req->in.h.opcode = opcode;
277 req->in.h.nodeid = get_node_id(inode);
279 req->in.args[0].size = sizeof(struct fuse_read_in);
280 req->in.args[0].value = inarg;
281 req->out.argpages = 1;
283 req->out.numargs = 1;
284 req->out.args[0].size = count;
287 static size_t fuse_send_read(struct fuse_req *req, struct file *file,
288 struct inode *inode, loff_t pos, size_t count)
290 struct fuse_conn *fc = get_fuse_conn(inode);
291 fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
292 request_send(fc, req);
293 return req->out.args[0].size;
296 static int fuse_readpage(struct file *file, struct page *page)
298 struct inode *inode = page->mapping->host;
299 struct fuse_conn *fc = get_fuse_conn(inode);
300 struct fuse_req *req;
304 if (is_bad_inode(inode))
307 req = fuse_get_req(fc);
312 req->out.page_zeroing = 1;
314 req->pages[0] = page;
315 fuse_send_read(req, file, inode, page_offset(page), PAGE_CACHE_SIZE);
316 err = req->out.h.error;
317 fuse_put_request(fc, req);
319 SetPageUptodate(page);
320 fuse_invalidate_attr(inode); /* atime changed */
326 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
330 fuse_invalidate_attr(req->pages[0]->mapping->host); /* atime changed */
332 for (i = 0; i < req->num_pages; i++) {
333 struct page *page = req->pages[i];
334 if (!req->out.h.error)
335 SetPageUptodate(page);
340 fuse_put_request(fc, req);
343 static void fuse_send_readpages(struct fuse_req *req, struct file *file,
346 struct fuse_conn *fc = get_fuse_conn(inode);
347 loff_t pos = page_offset(req->pages[0]);
348 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
349 req->out.page_zeroing = 1;
350 fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
351 if (fc->async_read) {
354 req->end = fuse_readpages_end;
355 request_send_background(fc, req);
357 request_send(fc, req);
358 fuse_readpages_end(fc, req);
362 struct fuse_readpages_data {
363 struct fuse_req *req;
368 static int fuse_readpages_fill(void *_data, struct page *page)
370 struct fuse_readpages_data *data = _data;
371 struct fuse_req *req = data->req;
372 struct inode *inode = data->inode;
373 struct fuse_conn *fc = get_fuse_conn(inode);
375 if (req->num_pages &&
376 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
377 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
378 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
379 fuse_send_readpages(req, data->file, inode);
380 data->req = req = fuse_get_req(fc);
386 req->pages[req->num_pages] = page;
391 static int fuse_readpages(struct file *file, struct address_space *mapping,
392 struct list_head *pages, unsigned nr_pages)
394 struct inode *inode = mapping->host;
395 struct fuse_conn *fc = get_fuse_conn(inode);
396 struct fuse_readpages_data data;
400 if (is_bad_inode(inode))
405 data.req = fuse_get_req(fc);
406 err = PTR_ERR(data.req);
407 if (IS_ERR(data.req))
410 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
412 if (data.req->num_pages)
413 fuse_send_readpages(data.req, file, inode);
415 fuse_put_request(fc, data.req);
421 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
422 struct inode *inode, loff_t pos, size_t count)
424 struct fuse_conn *fc = get_fuse_conn(inode);
425 struct fuse_file *ff = file->private_data;
426 struct fuse_write_in inarg;
427 struct fuse_write_out outarg;
429 memset(&inarg, 0, sizeof(struct fuse_write_in));
433 req->in.h.opcode = FUSE_WRITE;
434 req->in.h.nodeid = get_node_id(inode);
435 req->in.argpages = 1;
437 req->in.args[0].size = sizeof(struct fuse_write_in);
438 req->in.args[0].value = &inarg;
439 req->in.args[1].size = count;
440 req->out.numargs = 1;
441 req->out.args[0].size = sizeof(struct fuse_write_out);
442 req->out.args[0].value = &outarg;
443 request_send(fc, req);
447 static int fuse_prepare_write(struct file *file, struct page *page,
448 unsigned offset, unsigned to)
454 static int fuse_commit_write(struct file *file, struct page *page,
455 unsigned offset, unsigned to)
459 unsigned count = to - offset;
460 struct inode *inode = page->mapping->host;
461 struct fuse_conn *fc = get_fuse_conn(inode);
462 loff_t pos = page_offset(page) + offset;
463 struct fuse_req *req;
465 if (is_bad_inode(inode))
468 req = fuse_get_req(fc);
473 req->pages[0] = page;
474 req->page_offset = offset;
475 nres = fuse_send_write(req, file, inode, pos, count);
476 err = req->out.h.error;
477 fuse_put_request(fc, req);
478 if (!err && nres != count)
482 spin_lock(&fc->lock);
483 if (pos > inode->i_size)
484 i_size_write(inode, pos);
485 spin_unlock(&fc->lock);
487 if (offset == 0 && to == PAGE_CACHE_SIZE)
488 SetPageUptodate(page);
490 fuse_invalidate_attr(inode);
494 static void fuse_release_user_pages(struct fuse_req *req, int write)
498 for (i = 0; i < req->num_pages; i++) {
499 struct page *page = req->pages[i];
501 set_page_dirty_lock(page);
506 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
507 unsigned nbytes, int write)
509 unsigned long user_addr = (unsigned long) buf;
510 unsigned offset = user_addr & ~PAGE_MASK;
513 /* This doesn't work with nfsd */
517 nbytes = min(nbytes, (unsigned) FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
518 npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
519 npages = min(max(npages, 1), FUSE_MAX_PAGES_PER_REQ);
520 down_read(¤t->mm->mmap_sem);
521 npages = get_user_pages(current, current->mm, user_addr, npages, write,
522 0, req->pages, NULL);
523 up_read(¤t->mm->mmap_sem);
527 req->num_pages = npages;
528 req->page_offset = offset;
532 static ssize_t fuse_direct_io(struct file *file, const char __user *buf,
533 size_t count, loff_t *ppos, int write)
535 struct inode *inode = file->f_path.dentry->d_inode;
536 struct fuse_conn *fc = get_fuse_conn(inode);
537 size_t nmax = write ? fc->max_write : fc->max_read;
540 struct fuse_req *req;
542 if (is_bad_inode(inode))
545 req = fuse_get_req(fc);
551 size_t nbytes = min(count, nmax);
552 int err = fuse_get_user_pages(req, buf, nbytes, !write);
557 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset;
558 nbytes = min(count, nbytes);
560 nres = fuse_send_write(req, file, inode, pos, nbytes);
562 nres = fuse_send_read(req, file, inode, pos, nbytes);
563 fuse_release_user_pages(req, !write);
564 if (req->out.h.error) {
566 res = req->out.h.error;
568 } else if (nres > nbytes) {
579 fuse_put_request(fc, req);
580 req = fuse_get_req(fc);
585 fuse_put_request(fc, req);
588 spin_lock(&fc->lock);
589 if (pos > inode->i_size)
590 i_size_write(inode, pos);
591 spin_unlock(&fc->lock);
595 fuse_invalidate_attr(inode);
600 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
601 size_t count, loff_t *ppos)
603 return fuse_direct_io(file, buf, count, ppos, 0);
606 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
607 size_t count, loff_t *ppos)
609 struct inode *inode = file->f_path.dentry->d_inode;
611 /* Don't allow parallel writes to the same file */
612 mutex_lock(&inode->i_mutex);
613 res = fuse_direct_io(file, buf, count, ppos, 1);
614 mutex_unlock(&inode->i_mutex);
618 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
620 if ((vma->vm_flags & VM_SHARED)) {
621 if ((vma->vm_flags & VM_WRITE))
624 vma->vm_flags &= ~VM_MAYWRITE;
626 return generic_file_mmap(file, vma);
629 static int fuse_set_page_dirty(struct page *page)
631 printk("fuse_set_page_dirty: should not happen\n");
636 static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
637 struct file_lock *fl)
645 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
646 ffl->end < ffl->start)
649 fl->fl_start = ffl->start;
650 fl->fl_end = ffl->end;
651 fl->fl_pid = ffl->pid;
657 fl->fl_type = ffl->type;
661 static void fuse_lk_fill(struct fuse_req *req, struct file *file,
662 const struct file_lock *fl, int opcode, pid_t pid)
664 struct inode *inode = file->f_path.dentry->d_inode;
665 struct fuse_conn *fc = get_fuse_conn(inode);
666 struct fuse_file *ff = file->private_data;
667 struct fuse_lk_in *arg = &req->misc.lk_in;
670 arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
671 arg->lk.start = fl->fl_start;
672 arg->lk.end = fl->fl_end;
673 arg->lk.type = fl->fl_type;
675 req->in.h.opcode = opcode;
676 req->in.h.nodeid = get_node_id(inode);
678 req->in.args[0].size = sizeof(*arg);
679 req->in.args[0].value = arg;
682 static int fuse_getlk(struct file *file, struct file_lock *fl)
684 struct inode *inode = file->f_path.dentry->d_inode;
685 struct fuse_conn *fc = get_fuse_conn(inode);
686 struct fuse_req *req;
687 struct fuse_lk_out outarg;
690 req = fuse_get_req(fc);
694 fuse_lk_fill(req, file, fl, FUSE_GETLK, 0);
695 req->out.numargs = 1;
696 req->out.args[0].size = sizeof(outarg);
697 req->out.args[0].value = &outarg;
698 request_send(fc, req);
699 err = req->out.h.error;
700 fuse_put_request(fc, req);
702 err = convert_fuse_file_lock(&outarg.lk, fl);
707 static int fuse_setlk(struct file *file, struct file_lock *fl)
709 struct inode *inode = file->f_path.dentry->d_inode;
710 struct fuse_conn *fc = get_fuse_conn(inode);
711 struct fuse_req *req;
712 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
713 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
716 /* Unlock on close is handled by the flush method */
717 if (fl->fl_flags & FL_CLOSE)
720 req = fuse_get_req(fc);
724 fuse_lk_fill(req, file, fl, opcode, pid);
725 request_send(fc, req);
726 err = req->out.h.error;
727 /* locking is restartable */
730 fuse_put_request(fc, req);
734 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
736 struct inode *inode = file->f_path.dentry->d_inode;
737 struct fuse_conn *fc = get_fuse_conn(inode);
740 if (cmd == F_GETLK) {
742 posix_test_lock(file, fl);
745 err = fuse_getlk(file, fl);
748 err = posix_lock_file_wait(file, fl);
750 err = fuse_setlk(file, fl);
755 static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
757 struct inode *inode = mapping->host;
758 struct fuse_conn *fc = get_fuse_conn(inode);
759 struct fuse_req *req;
760 struct fuse_bmap_in inarg;
761 struct fuse_bmap_out outarg;
764 if (!inode->i_sb->s_bdev || fc->no_bmap)
767 req = fuse_get_req(fc);
771 memset(&inarg, 0, sizeof(inarg));
773 inarg.blocksize = inode->i_sb->s_blocksize;
774 req->in.h.opcode = FUSE_BMAP;
775 req->in.h.nodeid = get_node_id(inode);
777 req->in.args[0].size = sizeof(inarg);
778 req->in.args[0].value = &inarg;
779 req->out.numargs = 1;
780 req->out.args[0].size = sizeof(outarg);
781 req->out.args[0].value = &outarg;
782 request_send(fc, req);
783 err = req->out.h.error;
784 fuse_put_request(fc, req);
788 return err ? 0 : outarg.block;
791 static const struct file_operations fuse_file_operations = {
792 .llseek = generic_file_llseek,
793 .read = do_sync_read,
794 .aio_read = generic_file_aio_read,
795 .write = do_sync_write,
796 .aio_write = generic_file_aio_write,
797 .mmap = fuse_file_mmap,
800 .release = fuse_release,
802 .lock = fuse_file_lock,
803 .sendfile = generic_file_sendfile,
806 static const struct file_operations fuse_direct_io_file_operations = {
807 .llseek = generic_file_llseek,
808 .read = fuse_direct_read,
809 .write = fuse_direct_write,
812 .release = fuse_release,
814 .lock = fuse_file_lock,
815 /* no mmap and sendfile */
818 static const struct address_space_operations fuse_file_aops = {
819 .readpage = fuse_readpage,
820 .prepare_write = fuse_prepare_write,
821 .commit_write = fuse_commit_write,
822 .readpages = fuse_readpages,
823 .set_page_dirty = fuse_set_page_dirty,
827 void fuse_init_file_inode(struct inode *inode)
829 inode->i_fop = &fuse_file_operations;
830 inode->i_data.a_ops = &fuse_file_aops;