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) {
57 atomic_set(&ff->count, 0);
62 void fuse_file_free(struct fuse_file *ff)
64 fuse_request_free(ff->reserved_req);
68 static struct fuse_file *fuse_file_get(struct fuse_file *ff)
70 atomic_inc(&ff->count);
74 static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
77 mntput(req->vfsmount);
78 fuse_put_request(fc, req);
81 static void fuse_file_put(struct fuse_file *ff)
83 if (atomic_dec_and_test(&ff->count)) {
84 struct fuse_req *req = ff->reserved_req;
85 struct fuse_conn *fc = get_fuse_conn(req->dentry->d_inode);
86 req->end = fuse_release_end;
87 request_send_background(fc, req);
92 void fuse_finish_open(struct inode *inode, struct file *file,
93 struct fuse_file *ff, struct fuse_open_out *outarg)
95 if (outarg->open_flags & FOPEN_DIRECT_IO)
96 file->f_op = &fuse_direct_io_file_operations;
97 if (!(outarg->open_flags & FOPEN_KEEP_CACHE))
98 invalidate_inode_pages2(inode->i_mapping);
100 file->private_data = fuse_file_get(ff);
103 int fuse_open_common(struct inode *inode, struct file *file, int isdir)
105 struct fuse_open_out outarg;
106 struct fuse_file *ff;
109 /* VFS checks this, but only _after_ ->open() */
110 if (file->f_flags & O_DIRECT)
113 err = generic_file_open(inode, file);
117 ff = fuse_file_alloc();
121 err = fuse_send_open(inode, file, isdir, &outarg);
126 outarg.open_flags &= ~FOPEN_DIRECT_IO;
127 fuse_finish_open(inode, file, ff, &outarg);
133 void fuse_release_fill(struct fuse_file *ff, u64 nodeid, int flags, int opcode)
135 struct fuse_req *req = ff->reserved_req;
136 struct fuse_release_in *inarg = &req->misc.release_in;
139 inarg->flags = flags;
140 req->in.h.opcode = opcode;
141 req->in.h.nodeid = nodeid;
143 req->in.args[0].size = sizeof(struct fuse_release_in);
144 req->in.args[0].value = inarg;
147 int fuse_release_common(struct inode *inode, struct file *file, int isdir)
149 struct fuse_file *ff = file->private_data;
151 fuse_release_fill(ff, get_node_id(inode), file->f_flags,
152 isdir ? FUSE_RELEASEDIR : FUSE_RELEASE);
154 /* Hold vfsmount and dentry until release is finished */
155 ff->reserved_req->vfsmount = mntget(file->f_path.mnt);
156 ff->reserved_req->dentry = dget(file->f_path.dentry);
158 * Normally this will send the RELEASE request,
159 * however if some asynchronous READ or WRITE requests
160 * are outstanding, the sending will be delayed
165 /* Return value is ignored by VFS */
169 static int fuse_open(struct inode *inode, struct file *file)
171 return fuse_open_common(inode, file, 0);
174 static int fuse_release(struct inode *inode, struct file *file)
176 return fuse_release_common(inode, file, 0);
180 * Scramble the ID space with XTEA, so that the value of the files_struct
181 * pointer is not exposed to userspace.
183 static u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
185 u32 *k = fc->scramble_key;
186 u64 v = (unsigned long) id;
192 for (i = 0; i < 32; i++) {
193 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
195 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
198 return (u64) v0 + ((u64) v1 << 32);
201 static int fuse_flush(struct file *file, fl_owner_t id)
203 struct inode *inode = file->f_path.dentry->d_inode;
204 struct fuse_conn *fc = get_fuse_conn(inode);
205 struct fuse_file *ff = file->private_data;
206 struct fuse_req *req;
207 struct fuse_flush_in inarg;
210 if (is_bad_inode(inode))
216 req = fuse_get_req_nofail(fc, file);
217 memset(&inarg, 0, sizeof(inarg));
219 inarg.lock_owner = fuse_lock_owner_id(fc, id);
220 req->in.h.opcode = FUSE_FLUSH;
221 req->in.h.nodeid = get_node_id(inode);
223 req->in.args[0].size = sizeof(inarg);
224 req->in.args[0].value = &inarg;
226 request_send(fc, req);
227 err = req->out.h.error;
228 fuse_put_request(fc, req);
229 if (err == -ENOSYS) {
236 int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
239 struct inode *inode = de->d_inode;
240 struct fuse_conn *fc = get_fuse_conn(inode);
241 struct fuse_file *ff = file->private_data;
242 struct fuse_req *req;
243 struct fuse_fsync_in inarg;
246 if (is_bad_inode(inode))
249 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
252 req = fuse_get_req(fc);
256 memset(&inarg, 0, sizeof(inarg));
258 inarg.fsync_flags = datasync ? 1 : 0;
259 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
260 req->in.h.nodeid = get_node_id(inode);
262 req->in.args[0].size = sizeof(inarg);
263 req->in.args[0].value = &inarg;
264 request_send(fc, req);
265 err = req->out.h.error;
266 fuse_put_request(fc, req);
267 if (err == -ENOSYS) {
277 static int fuse_fsync(struct file *file, struct dentry *de, int datasync)
279 return fuse_fsync_common(file, de, datasync, 0);
282 void fuse_read_fill(struct fuse_req *req, struct fuse_file *ff,
283 struct inode *inode, loff_t pos, size_t count, int opcode)
285 struct fuse_read_in *inarg = &req->misc.read_in;
290 req->in.h.opcode = opcode;
291 req->in.h.nodeid = get_node_id(inode);
293 req->in.args[0].size = sizeof(struct fuse_read_in);
294 req->in.args[0].value = inarg;
295 req->out.argpages = 1;
297 req->out.numargs = 1;
298 req->out.args[0].size = count;
301 static size_t fuse_send_read(struct fuse_req *req, struct file *file,
302 struct inode *inode, loff_t pos, size_t count)
304 struct fuse_conn *fc = get_fuse_conn(inode);
305 struct fuse_file *ff = file->private_data;
306 fuse_read_fill(req, ff, inode, pos, count, FUSE_READ);
307 request_send(fc, req);
308 return req->out.args[0].size;
311 static int fuse_readpage(struct file *file, struct page *page)
313 struct inode *inode = page->mapping->host;
314 struct fuse_conn *fc = get_fuse_conn(inode);
315 struct fuse_req *req;
319 if (is_bad_inode(inode))
322 req = fuse_get_req(fc);
327 req->out.page_zeroing = 1;
329 req->pages[0] = page;
330 fuse_send_read(req, file, inode, page_offset(page), PAGE_CACHE_SIZE);
331 err = req->out.h.error;
332 fuse_put_request(fc, req);
334 SetPageUptodate(page);
335 fuse_invalidate_attr(inode); /* atime changed */
341 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
345 fuse_invalidate_attr(req->pages[0]->mapping->host); /* atime changed */
347 for (i = 0; i < req->num_pages; i++) {
348 struct page *page = req->pages[i];
349 if (!req->out.h.error)
350 SetPageUptodate(page);
356 fuse_file_put(req->ff);
357 fuse_put_request(fc, req);
360 static void fuse_send_readpages(struct fuse_req *req, struct fuse_file *ff,
363 struct fuse_conn *fc = get_fuse_conn(inode);
364 loff_t pos = page_offset(req->pages[0]);
365 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
366 req->out.page_zeroing = 1;
367 fuse_read_fill(req, ff, inode, pos, count, FUSE_READ);
368 if (fc->async_read) {
369 req->ff = fuse_file_get(ff);
370 req->end = fuse_readpages_end;
371 request_send_background(fc, req);
373 request_send(fc, req);
374 fuse_readpages_end(fc, req);
378 struct fuse_fill_data {
379 struct fuse_req *req;
380 struct fuse_file *ff;
384 static int fuse_readpages_fill(void *_data, struct page *page)
386 struct fuse_fill_data *data = _data;
387 struct fuse_req *req = data->req;
388 struct inode *inode = data->inode;
389 struct fuse_conn *fc = get_fuse_conn(inode);
391 if (req->num_pages &&
392 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
393 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
394 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
395 fuse_send_readpages(req, data->ff, inode);
396 data->req = req = fuse_get_req(fc);
402 req->pages[req->num_pages] = page;
407 static int fuse_readpages(struct file *file, struct address_space *mapping,
408 struct list_head *pages, unsigned nr_pages)
410 struct inode *inode = mapping->host;
411 struct fuse_conn *fc = get_fuse_conn(inode);
412 struct fuse_fill_data data;
416 if (is_bad_inode(inode))
419 data.ff = file->private_data;
421 data.req = fuse_get_req(fc);
422 err = PTR_ERR(data.req);
423 if (IS_ERR(data.req))
426 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
428 if (data.req->num_pages)
429 fuse_send_readpages(data.req, data.ff, inode);
431 fuse_put_request(fc, data.req);
437 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
438 struct inode *inode, loff_t pos, size_t count)
440 struct fuse_conn *fc = get_fuse_conn(inode);
441 struct fuse_file *ff = file->private_data;
442 struct fuse_write_in inarg;
443 struct fuse_write_out outarg;
445 memset(&inarg, 0, sizeof(struct fuse_write_in));
449 req->in.h.opcode = FUSE_WRITE;
450 req->in.h.nodeid = get_node_id(inode);
451 req->in.argpages = 1;
453 req->in.args[0].size = sizeof(struct fuse_write_in);
454 req->in.args[0].value = &inarg;
455 req->in.args[1].size = count;
456 req->out.numargs = 1;
457 req->out.args[0].size = sizeof(struct fuse_write_out);
458 req->out.args[0].value = &outarg;
459 request_send(fc, req);
463 static int fuse_write_begin(struct file *file, struct address_space *mapping,
464 loff_t pos, unsigned len, unsigned flags,
465 struct page **pagep, void **fsdata)
467 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
469 *pagep = __grab_cache_page(mapping, index);
475 static int fuse_buffered_write(struct file *file, struct inode *inode,
476 loff_t pos, unsigned count, struct page *page)
480 struct fuse_conn *fc = get_fuse_conn(inode);
481 struct fuse_inode *fi = get_fuse_inode(inode);
482 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
483 struct fuse_req *req;
485 if (is_bad_inode(inode))
488 req = fuse_get_req(fc);
493 req->pages[0] = page;
494 req->page_offset = offset;
495 nres = fuse_send_write(req, file, inode, pos, count);
496 err = req->out.h.error;
497 fuse_put_request(fc, req);
502 spin_lock(&fc->lock);
503 fi->attr_version = ++fc->attr_version;
504 if (pos > inode->i_size)
505 i_size_write(inode, pos);
506 spin_unlock(&fc->lock);
508 if (count == PAGE_CACHE_SIZE)
509 SetPageUptodate(page);
511 fuse_invalidate_attr(inode);
512 return err ? err : nres;
515 static int fuse_write_end(struct file *file, struct address_space *mapping,
516 loff_t pos, unsigned len, unsigned copied,
517 struct page *page, void *fsdata)
519 struct inode *inode = mapping->host;
523 res = fuse_buffered_write(file, inode, pos, copied, page);
526 page_cache_release(page);
530 static void fuse_release_user_pages(struct fuse_req *req, int write)
534 for (i = 0; i < req->num_pages; i++) {
535 struct page *page = req->pages[i];
537 set_page_dirty_lock(page);
542 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
543 unsigned nbytes, int write)
545 unsigned long user_addr = (unsigned long) buf;
546 unsigned offset = user_addr & ~PAGE_MASK;
549 /* This doesn't work with nfsd */
553 nbytes = min(nbytes, (unsigned) FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
554 npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
555 npages = min(max(npages, 1), FUSE_MAX_PAGES_PER_REQ);
556 down_read(¤t->mm->mmap_sem);
557 npages = get_user_pages(current, current->mm, user_addr, npages, write,
558 0, req->pages, NULL);
559 up_read(¤t->mm->mmap_sem);
563 req->num_pages = npages;
564 req->page_offset = offset;
568 static ssize_t fuse_direct_io(struct file *file, const char __user *buf,
569 size_t count, loff_t *ppos, int write)
571 struct inode *inode = file->f_path.dentry->d_inode;
572 struct fuse_conn *fc = get_fuse_conn(inode);
573 size_t nmax = write ? fc->max_write : fc->max_read;
576 struct fuse_req *req;
578 if (is_bad_inode(inode))
581 req = fuse_get_req(fc);
587 size_t nbytes = min(count, nmax);
588 int err = fuse_get_user_pages(req, buf, nbytes, !write);
593 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset;
594 nbytes = min(count, nbytes);
596 nres = fuse_send_write(req, file, inode, pos, nbytes);
598 nres = fuse_send_read(req, file, inode, pos, nbytes);
599 fuse_release_user_pages(req, !write);
600 if (req->out.h.error) {
602 res = req->out.h.error;
604 } else if (nres > nbytes) {
615 fuse_put_request(fc, req);
616 req = fuse_get_req(fc);
621 fuse_put_request(fc, req);
624 spin_lock(&fc->lock);
625 if (pos > inode->i_size)
626 i_size_write(inode, pos);
627 spin_unlock(&fc->lock);
631 fuse_invalidate_attr(inode);
636 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
637 size_t count, loff_t *ppos)
639 return fuse_direct_io(file, buf, count, ppos, 0);
642 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
643 size_t count, loff_t *ppos)
645 struct inode *inode = file->f_path.dentry->d_inode;
647 /* Don't allow parallel writes to the same file */
648 mutex_lock(&inode->i_mutex);
649 res = generic_write_checks(file, ppos, &count, 0);
651 res = fuse_direct_io(file, buf, count, ppos, 1);
652 mutex_unlock(&inode->i_mutex);
656 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
658 if ((vma->vm_flags & VM_SHARED)) {
659 if ((vma->vm_flags & VM_WRITE))
662 vma->vm_flags &= ~VM_MAYWRITE;
664 return generic_file_mmap(file, vma);
667 static int fuse_set_page_dirty(struct page *page)
669 printk("fuse_set_page_dirty: should not happen\n");
674 static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
675 struct file_lock *fl)
683 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
684 ffl->end < ffl->start)
687 fl->fl_start = ffl->start;
688 fl->fl_end = ffl->end;
689 fl->fl_pid = ffl->pid;
695 fl->fl_type = ffl->type;
699 static void fuse_lk_fill(struct fuse_req *req, struct file *file,
700 const struct file_lock *fl, int opcode, pid_t pid)
702 struct inode *inode = file->f_path.dentry->d_inode;
703 struct fuse_conn *fc = get_fuse_conn(inode);
704 struct fuse_file *ff = file->private_data;
705 struct fuse_lk_in *arg = &req->misc.lk_in;
708 arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
709 arg->lk.start = fl->fl_start;
710 arg->lk.end = fl->fl_end;
711 arg->lk.type = fl->fl_type;
713 req->in.h.opcode = opcode;
714 req->in.h.nodeid = get_node_id(inode);
716 req->in.args[0].size = sizeof(*arg);
717 req->in.args[0].value = arg;
720 static int fuse_getlk(struct file *file, struct file_lock *fl)
722 struct inode *inode = file->f_path.dentry->d_inode;
723 struct fuse_conn *fc = get_fuse_conn(inode);
724 struct fuse_req *req;
725 struct fuse_lk_out outarg;
728 req = fuse_get_req(fc);
732 fuse_lk_fill(req, file, fl, FUSE_GETLK, 0);
733 req->out.numargs = 1;
734 req->out.args[0].size = sizeof(outarg);
735 req->out.args[0].value = &outarg;
736 request_send(fc, req);
737 err = req->out.h.error;
738 fuse_put_request(fc, req);
740 err = convert_fuse_file_lock(&outarg.lk, fl);
745 static int fuse_setlk(struct file *file, struct file_lock *fl)
747 struct inode *inode = file->f_path.dentry->d_inode;
748 struct fuse_conn *fc = get_fuse_conn(inode);
749 struct fuse_req *req;
750 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
751 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
754 /* Unlock on close is handled by the flush method */
755 if (fl->fl_flags & FL_CLOSE)
758 req = fuse_get_req(fc);
762 fuse_lk_fill(req, file, fl, opcode, pid);
763 request_send(fc, req);
764 err = req->out.h.error;
765 /* locking is restartable */
768 fuse_put_request(fc, req);
772 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
774 struct inode *inode = file->f_path.dentry->d_inode;
775 struct fuse_conn *fc = get_fuse_conn(inode);
778 if (cmd == F_GETLK) {
780 posix_test_lock(file, fl);
783 err = fuse_getlk(file, fl);
786 err = posix_lock_file_wait(file, fl);
788 err = fuse_setlk(file, fl);
793 static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
795 struct inode *inode = mapping->host;
796 struct fuse_conn *fc = get_fuse_conn(inode);
797 struct fuse_req *req;
798 struct fuse_bmap_in inarg;
799 struct fuse_bmap_out outarg;
802 if (!inode->i_sb->s_bdev || fc->no_bmap)
805 req = fuse_get_req(fc);
809 memset(&inarg, 0, sizeof(inarg));
811 inarg.blocksize = inode->i_sb->s_blocksize;
812 req->in.h.opcode = FUSE_BMAP;
813 req->in.h.nodeid = get_node_id(inode);
815 req->in.args[0].size = sizeof(inarg);
816 req->in.args[0].value = &inarg;
817 req->out.numargs = 1;
818 req->out.args[0].size = sizeof(outarg);
819 req->out.args[0].value = &outarg;
820 request_send(fc, req);
821 err = req->out.h.error;
822 fuse_put_request(fc, req);
826 return err ? 0 : outarg.block;
829 static const struct file_operations fuse_file_operations = {
830 .llseek = generic_file_llseek,
831 .read = do_sync_read,
832 .aio_read = generic_file_aio_read,
833 .write = do_sync_write,
834 .aio_write = generic_file_aio_write,
835 .mmap = fuse_file_mmap,
838 .release = fuse_release,
840 .lock = fuse_file_lock,
841 .splice_read = generic_file_splice_read,
844 static const struct file_operations fuse_direct_io_file_operations = {
845 .llseek = generic_file_llseek,
846 .read = fuse_direct_read,
847 .write = fuse_direct_write,
850 .release = fuse_release,
852 .lock = fuse_file_lock,
853 /* no mmap and splice_read */
856 static const struct address_space_operations fuse_file_aops = {
857 .readpage = fuse_readpage,
858 .write_begin = fuse_write_begin,
859 .write_end = fuse_write_end,
860 .readpages = fuse_readpages,
861 .set_page_dirty = fuse_set_page_dirty,
865 void fuse_init_file_inode(struct inode *inode)
867 inode->i_fop = &fuse_file_operations;
868 inode->i_data.a_ops = &fuse_file_aops;