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 * It would be nice to scramble the ID space, so that the value of the
165 * files_struct pointer is not exposed to userspace. Symmetric crypto
166 * functions are overkill, since the inverse function doesn't need to
167 * be implemented (though it does have to exist). Is there something
170 static inline u64 fuse_lock_owner_id(fl_owner_t id)
172 return (unsigned long) id;
175 static int fuse_flush(struct file *file, fl_owner_t id)
177 struct inode *inode = file->f_dentry->d_inode;
178 struct fuse_conn *fc = get_fuse_conn(inode);
179 struct fuse_file *ff = file->private_data;
180 struct fuse_req *req;
181 struct fuse_flush_in inarg;
184 if (is_bad_inode(inode))
190 req = fuse_get_req_nofail(fc, file);
191 memset(&inarg, 0, sizeof(inarg));
193 inarg.lock_owner = fuse_lock_owner_id(id);
194 req->in.h.opcode = FUSE_FLUSH;
195 req->in.h.nodeid = get_node_id(inode);
197 req->in.args[0].size = sizeof(inarg);
198 req->in.args[0].value = &inarg;
200 request_send(fc, req);
201 err = req->out.h.error;
202 fuse_put_request(fc, req);
203 if (err == -ENOSYS) {
210 int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
213 struct inode *inode = de->d_inode;
214 struct fuse_conn *fc = get_fuse_conn(inode);
215 struct fuse_file *ff = file->private_data;
216 struct fuse_req *req;
217 struct fuse_fsync_in inarg;
220 if (is_bad_inode(inode))
223 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
226 req = fuse_get_req(fc);
230 memset(&inarg, 0, sizeof(inarg));
232 inarg.fsync_flags = datasync ? 1 : 0;
233 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
234 req->in.h.nodeid = get_node_id(inode);
236 req->in.args[0].size = sizeof(inarg);
237 req->in.args[0].value = &inarg;
238 request_send(fc, req);
239 err = req->out.h.error;
240 fuse_put_request(fc, req);
241 if (err == -ENOSYS) {
251 static int fuse_fsync(struct file *file, struct dentry *de, int datasync)
253 return fuse_fsync_common(file, de, datasync, 0);
256 void fuse_read_fill(struct fuse_req *req, struct file *file,
257 struct inode *inode, loff_t pos, size_t count, int opcode)
259 struct fuse_file *ff = file->private_data;
260 struct fuse_read_in *inarg = &req->misc.read_in;
265 req->in.h.opcode = opcode;
266 req->in.h.nodeid = get_node_id(inode);
268 req->in.args[0].size = sizeof(struct fuse_read_in);
269 req->in.args[0].value = inarg;
270 req->out.argpages = 1;
272 req->out.numargs = 1;
273 req->out.args[0].size = count;
276 static size_t fuse_send_read(struct fuse_req *req, struct file *file,
277 struct inode *inode, loff_t pos, size_t count)
279 struct fuse_conn *fc = get_fuse_conn(inode);
280 fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
281 request_send(fc, req);
282 return req->out.args[0].size;
285 static int fuse_readpage(struct file *file, struct page *page)
287 struct inode *inode = page->mapping->host;
288 struct fuse_conn *fc = get_fuse_conn(inode);
289 struct fuse_req *req;
293 if (is_bad_inode(inode))
296 req = fuse_get_req(fc);
301 req->out.page_zeroing = 1;
303 req->pages[0] = page;
304 fuse_send_read(req, file, inode, page_offset(page), PAGE_CACHE_SIZE);
305 err = req->out.h.error;
306 fuse_put_request(fc, req);
308 SetPageUptodate(page);
309 fuse_invalidate_attr(inode); /* atime changed */
315 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
319 fuse_invalidate_attr(req->pages[0]->mapping->host); /* atime changed */
321 for (i = 0; i < req->num_pages; i++) {
322 struct page *page = req->pages[i];
323 if (!req->out.h.error)
324 SetPageUptodate(page);
329 fuse_put_request(fc, req);
332 static void fuse_send_readpages(struct fuse_req *req, struct file *file,
335 struct fuse_conn *fc = get_fuse_conn(inode);
336 loff_t pos = page_offset(req->pages[0]);
337 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
338 req->out.page_zeroing = 1;
339 fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
340 if (fc->async_read) {
343 req->end = fuse_readpages_end;
344 request_send_background(fc, req);
346 request_send(fc, req);
347 fuse_readpages_end(fc, req);
351 struct fuse_readpages_data {
352 struct fuse_req *req;
357 static int fuse_readpages_fill(void *_data, struct page *page)
359 struct fuse_readpages_data *data = _data;
360 struct fuse_req *req = data->req;
361 struct inode *inode = data->inode;
362 struct fuse_conn *fc = get_fuse_conn(inode);
364 if (req->num_pages &&
365 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
366 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
367 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
368 fuse_send_readpages(req, data->file, inode);
369 data->req = req = fuse_get_req(fc);
375 req->pages[req->num_pages] = page;
380 static int fuse_readpages(struct file *file, struct address_space *mapping,
381 struct list_head *pages, unsigned nr_pages)
383 struct inode *inode = mapping->host;
384 struct fuse_conn *fc = get_fuse_conn(inode);
385 struct fuse_readpages_data data;
388 if (is_bad_inode(inode))
393 data.req = fuse_get_req(fc);
394 if (IS_ERR(data.req))
395 return PTR_ERR(data.req);
397 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
399 if (data.req->num_pages)
400 fuse_send_readpages(data.req, file, inode);
402 fuse_put_request(fc, data.req);
407 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
408 struct inode *inode, loff_t pos, size_t count)
410 struct fuse_conn *fc = get_fuse_conn(inode);
411 struct fuse_file *ff = file->private_data;
412 struct fuse_write_in inarg;
413 struct fuse_write_out outarg;
415 memset(&inarg, 0, sizeof(struct fuse_write_in));
419 req->in.h.opcode = FUSE_WRITE;
420 req->in.h.nodeid = get_node_id(inode);
421 req->in.argpages = 1;
423 req->in.args[0].size = sizeof(struct fuse_write_in);
424 req->in.args[0].value = &inarg;
425 req->in.args[1].size = count;
426 req->out.numargs = 1;
427 req->out.args[0].size = sizeof(struct fuse_write_out);
428 req->out.args[0].value = &outarg;
429 request_send(fc, req);
433 static int fuse_prepare_write(struct file *file, struct page *page,
434 unsigned offset, unsigned to)
440 static int fuse_commit_write(struct file *file, struct page *page,
441 unsigned offset, unsigned to)
445 unsigned count = to - offset;
446 struct inode *inode = page->mapping->host;
447 struct fuse_conn *fc = get_fuse_conn(inode);
448 loff_t pos = page_offset(page) + offset;
449 struct fuse_req *req;
451 if (is_bad_inode(inode))
454 req = fuse_get_req(fc);
459 req->pages[0] = page;
460 req->page_offset = offset;
461 nres = fuse_send_write(req, file, inode, pos, count);
462 err = req->out.h.error;
463 fuse_put_request(fc, req);
464 if (!err && nres != count)
468 if (pos > i_size_read(inode))
469 i_size_write(inode, pos);
471 if (offset == 0 && to == PAGE_CACHE_SIZE) {
472 clear_page_dirty(page);
473 SetPageUptodate(page);
476 fuse_invalidate_attr(inode);
480 static void fuse_release_user_pages(struct fuse_req *req, int write)
484 for (i = 0; i < req->num_pages; i++) {
485 struct page *page = req->pages[i];
487 set_page_dirty_lock(page);
492 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
493 unsigned nbytes, int write)
495 unsigned long user_addr = (unsigned long) buf;
496 unsigned offset = user_addr & ~PAGE_MASK;
499 /* This doesn't work with nfsd */
503 nbytes = min(nbytes, (unsigned) FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
504 npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
505 npages = min(max(npages, 1), FUSE_MAX_PAGES_PER_REQ);
506 down_read(¤t->mm->mmap_sem);
507 npages = get_user_pages(current, current->mm, user_addr, npages, write,
508 0, req->pages, NULL);
509 up_read(¤t->mm->mmap_sem);
513 req->num_pages = npages;
514 req->page_offset = offset;
518 static ssize_t fuse_direct_io(struct file *file, const char __user *buf,
519 size_t count, loff_t *ppos, int write)
521 struct inode *inode = file->f_dentry->d_inode;
522 struct fuse_conn *fc = get_fuse_conn(inode);
523 size_t nmax = write ? fc->max_write : fc->max_read;
526 struct fuse_req *req;
528 if (is_bad_inode(inode))
531 req = fuse_get_req(fc);
537 size_t nbytes = min(count, nmax);
538 int err = fuse_get_user_pages(req, buf, nbytes, !write);
543 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset;
544 nbytes = min(count, nbytes);
546 nres = fuse_send_write(req, file, inode, pos, nbytes);
548 nres = fuse_send_read(req, file, inode, pos, nbytes);
549 fuse_release_user_pages(req, !write);
550 if (req->out.h.error) {
552 res = req->out.h.error;
554 } else if (nres > nbytes) {
565 fuse_put_request(fc, req);
566 req = fuse_get_req(fc);
571 fuse_put_request(fc, req);
573 if (write && pos > i_size_read(inode))
574 i_size_write(inode, pos);
577 fuse_invalidate_attr(inode);
582 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
583 size_t count, loff_t *ppos)
585 return fuse_direct_io(file, buf, count, ppos, 0);
588 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
589 size_t count, loff_t *ppos)
591 struct inode *inode = file->f_dentry->d_inode;
593 /* Don't allow parallel writes to the same file */
594 mutex_lock(&inode->i_mutex);
595 res = fuse_direct_io(file, buf, count, ppos, 1);
596 mutex_unlock(&inode->i_mutex);
600 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
602 if ((vma->vm_flags & VM_SHARED)) {
603 if ((vma->vm_flags & VM_WRITE))
606 vma->vm_flags &= ~VM_MAYWRITE;
608 return generic_file_mmap(file, vma);
611 static int fuse_set_page_dirty(struct page *page)
613 printk("fuse_set_page_dirty: should not happen\n");
618 static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
619 struct file_lock *fl)
627 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
628 ffl->end < ffl->start)
631 fl->fl_start = ffl->start;
632 fl->fl_end = ffl->end;
633 fl->fl_pid = ffl->pid;
639 fl->fl_type = ffl->type;
643 static void fuse_lk_fill(struct fuse_req *req, struct file *file,
644 const struct file_lock *fl, int opcode, pid_t pid)
646 struct inode *inode = file->f_dentry->d_inode;
647 struct fuse_file *ff = file->private_data;
648 struct fuse_lk_in *arg = &req->misc.lk_in;
651 arg->owner = fuse_lock_owner_id(fl->fl_owner);
652 arg->lk.start = fl->fl_start;
653 arg->lk.end = fl->fl_end;
654 arg->lk.type = fl->fl_type;
656 req->in.h.opcode = opcode;
657 req->in.h.nodeid = get_node_id(inode);
659 req->in.args[0].size = sizeof(*arg);
660 req->in.args[0].value = arg;
663 static int fuse_getlk(struct file *file, struct file_lock *fl)
665 struct inode *inode = file->f_dentry->d_inode;
666 struct fuse_conn *fc = get_fuse_conn(inode);
667 struct fuse_req *req;
668 struct fuse_lk_out outarg;
671 req = fuse_get_req(fc);
675 fuse_lk_fill(req, file, fl, FUSE_GETLK, 0);
676 req->out.numargs = 1;
677 req->out.args[0].size = sizeof(outarg);
678 req->out.args[0].value = &outarg;
679 request_send(fc, req);
680 err = req->out.h.error;
681 fuse_put_request(fc, req);
683 err = convert_fuse_file_lock(&outarg.lk, fl);
688 static int fuse_setlk(struct file *file, struct file_lock *fl)
690 struct inode *inode = file->f_dentry->d_inode;
691 struct fuse_conn *fc = get_fuse_conn(inode);
692 struct fuse_req *req;
693 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
694 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
697 /* Unlock on close is handled by the flush method */
698 if (fl->fl_flags & FL_CLOSE)
701 req = fuse_get_req(fc);
705 fuse_lk_fill(req, file, fl, opcode, pid);
706 request_send(fc, req);
707 err = req->out.h.error;
708 fuse_put_request(fc, req);
712 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
714 struct inode *inode = file->f_dentry->d_inode;
715 struct fuse_conn *fc = get_fuse_conn(inode);
718 if (cmd == F_GETLK) {
720 if (!posix_test_lock(file, fl, fl))
721 fl->fl_type = F_UNLCK;
724 err = fuse_getlk(file, fl);
727 err = posix_lock_file_wait(file, fl);
729 err = fuse_setlk(file, fl);
734 static const struct file_operations fuse_file_operations = {
735 .llseek = generic_file_llseek,
736 .read = generic_file_read,
737 .write = generic_file_write,
738 .mmap = fuse_file_mmap,
741 .release = fuse_release,
743 .lock = fuse_file_lock,
744 .sendfile = generic_file_sendfile,
747 static const struct file_operations fuse_direct_io_file_operations = {
748 .llseek = generic_file_llseek,
749 .read = fuse_direct_read,
750 .write = fuse_direct_write,
753 .release = fuse_release,
755 .lock = fuse_file_lock,
756 /* no mmap and sendfile */
759 static struct address_space_operations fuse_file_aops = {
760 .readpage = fuse_readpage,
761 .prepare_write = fuse_prepare_write,
762 .commit_write = fuse_commit_write,
763 .readpages = fuse_readpages,
764 .set_page_dirty = fuse_set_page_dirty,
767 void fuse_init_file_inode(struct inode *inode)
769 inode->i_fop = &fuse_file_operations;
770 inode->i_data.a_ops = &fuse_file_aops;