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>
15 static const struct file_operations fuse_direct_io_file_operations;
17 static int fuse_send_open(struct inode *inode, struct file *file, int isdir,
18 struct fuse_open_out *outargp)
20 struct fuse_conn *fc = get_fuse_conn(inode);
21 struct fuse_open_in inarg;
25 req = fuse_get_req(fc);
29 memset(&inarg, 0, sizeof(inarg));
30 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
31 req->in.h.opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
32 req->in.h.nodeid = get_node_id(inode);
34 req->in.args[0].size = sizeof(inarg);
35 req->in.args[0].value = &inarg;
37 req->out.args[0].size = sizeof(*outargp);
38 req->out.args[0].value = outargp;
39 request_send(fc, req);
40 err = req->out.h.error;
41 fuse_put_request(fc, req);
46 struct fuse_file *fuse_file_alloc(void)
49 ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
51 ff->reserved_req = fuse_request_alloc();
52 if (!ff->reserved_req) {
60 void fuse_file_free(struct fuse_file *ff)
62 fuse_request_free(ff->reserved_req);
66 void fuse_finish_open(struct inode *inode, struct file *file,
67 struct fuse_file *ff, struct fuse_open_out *outarg)
69 if (outarg->open_flags & FOPEN_DIRECT_IO)
70 file->f_op = &fuse_direct_io_file_operations;
71 if (!(outarg->open_flags & FOPEN_KEEP_CACHE))
72 invalidate_inode_pages(inode->i_mapping);
74 file->private_data = ff;
77 int fuse_open_common(struct inode *inode, struct file *file, int isdir)
79 struct fuse_open_out outarg;
83 /* VFS checks this, but only _after_ ->open() */
84 if (file->f_flags & O_DIRECT)
87 err = generic_file_open(inode, file);
91 /* If opening the root node, no lookup has been performed on
92 it, so the attributes must be refreshed */
93 if (get_node_id(inode) == FUSE_ROOT_ID) {
94 err = fuse_do_getattr(inode);
99 ff = fuse_file_alloc();
103 err = fuse_send_open(inode, file, isdir, &outarg);
108 outarg.open_flags &= ~FOPEN_DIRECT_IO;
109 fuse_finish_open(inode, file, ff, &outarg);
115 struct fuse_req *fuse_release_fill(struct fuse_file *ff, u64 nodeid, int flags,
118 struct fuse_req *req = ff->reserved_req;
119 struct fuse_release_in *inarg = &req->misc.release_in;
122 inarg->flags = flags;
123 req->in.h.opcode = opcode;
124 req->in.h.nodeid = nodeid;
126 req->in.args[0].size = sizeof(struct fuse_release_in);
127 req->in.args[0].value = inarg;
133 int fuse_release_common(struct inode *inode, struct file *file, int isdir)
135 struct fuse_file *ff = file->private_data;
137 struct fuse_conn *fc = get_fuse_conn(inode);
138 struct fuse_req *req;
140 req = fuse_release_fill(ff, get_node_id(inode), file->f_flags,
141 isdir ? FUSE_RELEASEDIR : FUSE_RELEASE);
143 /* Hold vfsmount and dentry until release is finished */
144 req->vfsmount = mntget(file->f_vfsmnt);
145 req->dentry = dget(file->f_dentry);
146 request_send_background(fc, req);
149 /* Return value is ignored by VFS */
153 static int fuse_open(struct inode *inode, struct file *file)
155 return fuse_open_common(inode, file, 0);
158 static int fuse_release(struct inode *inode, struct file *file)
160 return fuse_release_common(inode, file, 0);
164 * Scramble the ID space with XTEA, so that the value of the files_struct
165 * pointer is not exposed to userspace.
167 static u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
169 u32 *k = fc->scramble_key;
170 u64 v = (unsigned long) id;
176 for (i = 0; i < 32; i++) {
177 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
179 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
182 return (u64) v0 + ((u64) v1 << 32);
185 static int fuse_flush(struct file *file, fl_owner_t id)
187 struct inode *inode = file->f_dentry->d_inode;
188 struct fuse_conn *fc = get_fuse_conn(inode);
189 struct fuse_file *ff = file->private_data;
190 struct fuse_req *req;
191 struct fuse_flush_in inarg;
194 if (is_bad_inode(inode))
200 req = fuse_get_req_nofail(fc, file);
201 memset(&inarg, 0, sizeof(inarg));
203 inarg.lock_owner = fuse_lock_owner_id(fc, id);
204 req->in.h.opcode = FUSE_FLUSH;
205 req->in.h.nodeid = get_node_id(inode);
207 req->in.args[0].size = sizeof(inarg);
208 req->in.args[0].value = &inarg;
210 request_send(fc, req);
211 err = req->out.h.error;
212 fuse_put_request(fc, req);
213 if (err == -ENOSYS) {
220 int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
223 struct inode *inode = de->d_inode;
224 struct fuse_conn *fc = get_fuse_conn(inode);
225 struct fuse_file *ff = file->private_data;
226 struct fuse_req *req;
227 struct fuse_fsync_in inarg;
230 if (is_bad_inode(inode))
233 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
236 req = fuse_get_req(fc);
240 memset(&inarg, 0, sizeof(inarg));
242 inarg.fsync_flags = datasync ? 1 : 0;
243 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
244 req->in.h.nodeid = get_node_id(inode);
246 req->in.args[0].size = sizeof(inarg);
247 req->in.args[0].value = &inarg;
248 request_send(fc, req);
249 err = req->out.h.error;
250 fuse_put_request(fc, req);
251 if (err == -ENOSYS) {
261 static int fuse_fsync(struct file *file, struct dentry *de, int datasync)
263 return fuse_fsync_common(file, de, datasync, 0);
266 void fuse_read_fill(struct fuse_req *req, struct file *file,
267 struct inode *inode, loff_t pos, size_t count, int opcode)
269 struct fuse_file *ff = file->private_data;
270 struct fuse_read_in *inarg = &req->misc.read_in;
275 req->in.h.opcode = opcode;
276 req->in.h.nodeid = get_node_id(inode);
278 req->in.args[0].size = sizeof(struct fuse_read_in);
279 req->in.args[0].value = inarg;
280 req->out.argpages = 1;
282 req->out.numargs = 1;
283 req->out.args[0].size = count;
286 static size_t fuse_send_read(struct fuse_req *req, struct file *file,
287 struct inode *inode, loff_t pos, size_t count)
289 struct fuse_conn *fc = get_fuse_conn(inode);
290 fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
291 request_send(fc, req);
292 return req->out.args[0].size;
295 static int fuse_readpage(struct file *file, struct page *page)
297 struct inode *inode = page->mapping->host;
298 struct fuse_conn *fc = get_fuse_conn(inode);
299 struct fuse_req *req;
303 if (is_bad_inode(inode))
306 req = fuse_get_req(fc);
311 req->out.page_zeroing = 1;
313 req->pages[0] = page;
314 fuse_send_read(req, file, inode, page_offset(page), PAGE_CACHE_SIZE);
315 err = req->out.h.error;
316 fuse_put_request(fc, req);
318 SetPageUptodate(page);
319 fuse_invalidate_attr(inode); /* atime changed */
325 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
329 fuse_invalidate_attr(req->pages[0]->mapping->host); /* atime changed */
331 for (i = 0; i < req->num_pages; i++) {
332 struct page *page = req->pages[i];
333 if (!req->out.h.error)
334 SetPageUptodate(page);
339 fuse_put_request(fc, req);
342 static void fuse_send_readpages(struct fuse_req *req, struct file *file,
345 struct fuse_conn *fc = get_fuse_conn(inode);
346 loff_t pos = page_offset(req->pages[0]);
347 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
348 req->out.page_zeroing = 1;
349 fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
350 if (fc->async_read) {
353 req->end = fuse_readpages_end;
354 request_send_background(fc, req);
356 request_send(fc, req);
357 fuse_readpages_end(fc, req);
361 struct fuse_readpages_data {
362 struct fuse_req *req;
367 static int fuse_readpages_fill(void *_data, struct page *page)
369 struct fuse_readpages_data *data = _data;
370 struct fuse_req *req = data->req;
371 struct inode *inode = data->inode;
372 struct fuse_conn *fc = get_fuse_conn(inode);
374 if (req->num_pages &&
375 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
376 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
377 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
378 fuse_send_readpages(req, data->file, inode);
379 data->req = req = fuse_get_req(fc);
385 req->pages[req->num_pages] = page;
390 static int fuse_readpages(struct file *file, struct address_space *mapping,
391 struct list_head *pages, unsigned nr_pages)
393 struct inode *inode = mapping->host;
394 struct fuse_conn *fc = get_fuse_conn(inode);
395 struct fuse_readpages_data data;
399 if (is_bad_inode(inode))
404 data.req = fuse_get_req(fc);
405 err = PTR_ERR(data.req);
406 if (IS_ERR(data.req))
409 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
411 if (data.req->num_pages)
412 fuse_send_readpages(data.req, file, inode);
414 fuse_put_request(fc, data.req);
419 put_pages_list(pages);
423 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
424 struct inode *inode, loff_t pos, size_t count)
426 struct fuse_conn *fc = get_fuse_conn(inode);
427 struct fuse_file *ff = file->private_data;
428 struct fuse_write_in inarg;
429 struct fuse_write_out outarg;
431 memset(&inarg, 0, sizeof(struct fuse_write_in));
435 req->in.h.opcode = FUSE_WRITE;
436 req->in.h.nodeid = get_node_id(inode);
437 req->in.argpages = 1;
439 req->in.args[0].size = sizeof(struct fuse_write_in);
440 req->in.args[0].value = &inarg;
441 req->in.args[1].size = count;
442 req->out.numargs = 1;
443 req->out.args[0].size = sizeof(struct fuse_write_out);
444 req->out.args[0].value = &outarg;
445 request_send(fc, req);
449 static int fuse_prepare_write(struct file *file, struct page *page,
450 unsigned offset, unsigned to)
456 static int fuse_commit_write(struct file *file, struct page *page,
457 unsigned offset, unsigned to)
461 unsigned count = to - offset;
462 struct inode *inode = page->mapping->host;
463 struct fuse_conn *fc = get_fuse_conn(inode);
464 loff_t pos = page_offset(page) + offset;
465 struct fuse_req *req;
467 if (is_bad_inode(inode))
470 req = fuse_get_req(fc);
475 req->pages[0] = page;
476 req->page_offset = offset;
477 nres = fuse_send_write(req, file, inode, pos, count);
478 err = req->out.h.error;
479 fuse_put_request(fc, req);
480 if (!err && nres != count)
484 if (pos > i_size_read(inode))
485 i_size_write(inode, pos);
487 if (offset == 0 && to == PAGE_CACHE_SIZE) {
488 clear_page_dirty(page);
489 SetPageUptodate(page);
492 fuse_invalidate_attr(inode);
496 static void fuse_release_user_pages(struct fuse_req *req, int write)
500 for (i = 0; i < req->num_pages; i++) {
501 struct page *page = req->pages[i];
503 set_page_dirty_lock(page);
508 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
509 unsigned nbytes, int write)
511 unsigned long user_addr = (unsigned long) buf;
512 unsigned offset = user_addr & ~PAGE_MASK;
515 /* This doesn't work with nfsd */
519 nbytes = min(nbytes, (unsigned) FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
520 npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
521 npages = min(max(npages, 1), FUSE_MAX_PAGES_PER_REQ);
522 down_read(¤t->mm->mmap_sem);
523 npages = get_user_pages(current, current->mm, user_addr, npages, write,
524 0, req->pages, NULL);
525 up_read(¤t->mm->mmap_sem);
529 req->num_pages = npages;
530 req->page_offset = offset;
534 static ssize_t fuse_direct_io(struct file *file, const char __user *buf,
535 size_t count, loff_t *ppos, int write)
537 struct inode *inode = file->f_dentry->d_inode;
538 struct fuse_conn *fc = get_fuse_conn(inode);
539 size_t nmax = write ? fc->max_write : fc->max_read;
542 struct fuse_req *req;
544 if (is_bad_inode(inode))
547 req = fuse_get_req(fc);
553 size_t nbytes = min(count, nmax);
554 int err = fuse_get_user_pages(req, buf, nbytes, !write);
559 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset;
560 nbytes = min(count, nbytes);
562 nres = fuse_send_write(req, file, inode, pos, nbytes);
564 nres = fuse_send_read(req, file, inode, pos, nbytes);
565 fuse_release_user_pages(req, !write);
566 if (req->out.h.error) {
568 res = req->out.h.error;
570 } else if (nres > nbytes) {
581 fuse_put_request(fc, req);
582 req = fuse_get_req(fc);
587 fuse_put_request(fc, req);
589 if (write && pos > i_size_read(inode))
590 i_size_write(inode, pos);
593 fuse_invalidate_attr(inode);
598 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
599 size_t count, loff_t *ppos)
601 return fuse_direct_io(file, buf, count, ppos, 0);
604 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
605 size_t count, loff_t *ppos)
607 struct inode *inode = file->f_dentry->d_inode;
609 /* Don't allow parallel writes to the same file */
610 mutex_lock(&inode->i_mutex);
611 res = fuse_direct_io(file, buf, count, ppos, 1);
612 mutex_unlock(&inode->i_mutex);
616 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
618 if ((vma->vm_flags & VM_SHARED)) {
619 if ((vma->vm_flags & VM_WRITE))
622 vma->vm_flags &= ~VM_MAYWRITE;
624 return generic_file_mmap(file, vma);
627 static int fuse_set_page_dirty(struct page *page)
629 printk("fuse_set_page_dirty: should not happen\n");
634 static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
635 struct file_lock *fl)
643 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
644 ffl->end < ffl->start)
647 fl->fl_start = ffl->start;
648 fl->fl_end = ffl->end;
649 fl->fl_pid = ffl->pid;
655 fl->fl_type = ffl->type;
659 static void fuse_lk_fill(struct fuse_req *req, struct file *file,
660 const struct file_lock *fl, int opcode, pid_t pid)
662 struct inode *inode = file->f_dentry->d_inode;
663 struct fuse_conn *fc = get_fuse_conn(inode);
664 struct fuse_file *ff = file->private_data;
665 struct fuse_lk_in *arg = &req->misc.lk_in;
668 arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
669 arg->lk.start = fl->fl_start;
670 arg->lk.end = fl->fl_end;
671 arg->lk.type = fl->fl_type;
673 req->in.h.opcode = opcode;
674 req->in.h.nodeid = get_node_id(inode);
676 req->in.args[0].size = sizeof(*arg);
677 req->in.args[0].value = arg;
680 static int fuse_getlk(struct file *file, struct file_lock *fl)
682 struct inode *inode = file->f_dentry->d_inode;
683 struct fuse_conn *fc = get_fuse_conn(inode);
684 struct fuse_req *req;
685 struct fuse_lk_out outarg;
688 req = fuse_get_req(fc);
692 fuse_lk_fill(req, file, fl, FUSE_GETLK, 0);
693 req->out.numargs = 1;
694 req->out.args[0].size = sizeof(outarg);
695 req->out.args[0].value = &outarg;
696 request_send(fc, req);
697 err = req->out.h.error;
698 fuse_put_request(fc, req);
700 err = convert_fuse_file_lock(&outarg.lk, fl);
705 static int fuse_setlk(struct file *file, struct file_lock *fl)
707 struct inode *inode = file->f_dentry->d_inode;
708 struct fuse_conn *fc = get_fuse_conn(inode);
709 struct fuse_req *req;
710 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
711 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
714 /* Unlock on close is handled by the flush method */
715 if (fl->fl_flags & FL_CLOSE)
718 req = fuse_get_req(fc);
722 fuse_lk_fill(req, file, fl, opcode, pid);
723 request_send(fc, req);
724 err = req->out.h.error;
725 /* locking is restartable */
728 fuse_put_request(fc, req);
732 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
734 struct inode *inode = file->f_dentry->d_inode;
735 struct fuse_conn *fc = get_fuse_conn(inode);
738 if (cmd == F_GETLK) {
740 if (!posix_test_lock(file, fl, fl))
741 fl->fl_type = F_UNLCK;
744 err = fuse_getlk(file, fl);
747 err = posix_lock_file_wait(file, fl);
749 err = fuse_setlk(file, fl);
754 static const struct file_operations fuse_file_operations = {
755 .llseek = generic_file_llseek,
756 .read = do_sync_read,
757 .aio_read = generic_file_aio_read,
758 .write = do_sync_write,
759 .aio_write = generic_file_aio_write,
760 .mmap = fuse_file_mmap,
763 .release = fuse_release,
765 .lock = fuse_file_lock,
766 .sendfile = generic_file_sendfile,
769 static const struct file_operations fuse_direct_io_file_operations = {
770 .llseek = generic_file_llseek,
771 .read = fuse_direct_read,
772 .write = fuse_direct_write,
775 .release = fuse_release,
777 .lock = fuse_file_lock,
778 /* no mmap and sendfile */
781 static const struct address_space_operations fuse_file_aops = {
782 .readpage = fuse_readpage,
783 .prepare_write = fuse_prepare_write,
784 .commit_write = fuse_commit_write,
785 .readpages = fuse_readpages,
786 .set_page_dirty = fuse_set_page_dirty,
789 void fuse_init_file_inode(struct inode *inode)
791 inode->i_fop = &fuse_file_operations;
792 inode->i_data.a_ops = &fuse_file_aops;