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);
65 void fuse_file_free(struct fuse_file *ff)
67 fuse_request_free(ff->reserved_req);
71 static struct fuse_file *fuse_file_get(struct fuse_file *ff)
73 atomic_inc(&ff->count);
77 static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
80 mntput(req->vfsmount);
81 fuse_put_request(fc, req);
84 static void fuse_file_put(struct fuse_file *ff)
86 if (atomic_dec_and_test(&ff->count)) {
87 struct fuse_req *req = ff->reserved_req;
88 struct fuse_conn *fc = get_fuse_conn(req->dentry->d_inode);
89 req->end = fuse_release_end;
90 request_send_background(fc, req);
95 void fuse_finish_open(struct inode *inode, struct file *file,
96 struct fuse_file *ff, struct fuse_open_out *outarg)
98 if (outarg->open_flags & FOPEN_DIRECT_IO)
99 file->f_op = &fuse_direct_io_file_operations;
100 if (!(outarg->open_flags & FOPEN_KEEP_CACHE))
101 invalidate_inode_pages2(inode->i_mapping);
103 file->private_data = fuse_file_get(ff);
106 int fuse_open_common(struct inode *inode, struct file *file, int isdir)
108 struct fuse_open_out outarg;
109 struct fuse_file *ff;
112 /* VFS checks this, but only _after_ ->open() */
113 if (file->f_flags & O_DIRECT)
116 err = generic_file_open(inode, file);
120 ff = fuse_file_alloc();
124 err = fuse_send_open(inode, file, isdir, &outarg);
129 outarg.open_flags &= ~FOPEN_DIRECT_IO;
130 fuse_finish_open(inode, file, ff, &outarg);
136 void fuse_release_fill(struct fuse_file *ff, u64 nodeid, int flags, int opcode)
138 struct fuse_req *req = ff->reserved_req;
139 struct fuse_release_in *inarg = &req->misc.release_in;
142 inarg->flags = flags;
143 req->in.h.opcode = opcode;
144 req->in.h.nodeid = nodeid;
146 req->in.args[0].size = sizeof(struct fuse_release_in);
147 req->in.args[0].value = inarg;
150 int fuse_release_common(struct inode *inode, struct file *file, int isdir)
152 struct fuse_file *ff = file->private_data;
154 struct fuse_conn *fc = get_fuse_conn(inode);
156 fuse_release_fill(ff, get_node_id(inode), file->f_flags,
157 isdir ? FUSE_RELEASEDIR : FUSE_RELEASE);
159 /* Hold vfsmount and dentry until release is finished */
160 ff->reserved_req->vfsmount = mntget(file->f_path.mnt);
161 ff->reserved_req->dentry = dget(file->f_path.dentry);
163 spin_lock(&fc->lock);
164 list_del(&ff->write_entry);
165 spin_unlock(&fc->lock);
167 * Normally this will send the RELEASE request,
168 * however if some asynchronous READ or WRITE requests
169 * are outstanding, the sending will be delayed
174 /* Return value is ignored by VFS */
178 static int fuse_open(struct inode *inode, struct file *file)
180 return fuse_open_common(inode, file, 0);
183 static int fuse_release(struct inode *inode, struct file *file)
185 return fuse_release_common(inode, file, 0);
189 * Scramble the ID space with XTEA, so that the value of the files_struct
190 * pointer is not exposed to userspace.
192 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
194 u32 *k = fc->scramble_key;
195 u64 v = (unsigned long) id;
201 for (i = 0; i < 32; i++) {
202 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
204 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
207 return (u64) v0 + ((u64) v1 << 32);
210 static int fuse_flush(struct file *file, fl_owner_t id)
212 struct inode *inode = file->f_path.dentry->d_inode;
213 struct fuse_conn *fc = get_fuse_conn(inode);
214 struct fuse_file *ff = file->private_data;
215 struct fuse_req *req;
216 struct fuse_flush_in inarg;
219 if (is_bad_inode(inode))
225 req = fuse_get_req_nofail(fc, file);
226 memset(&inarg, 0, sizeof(inarg));
228 inarg.lock_owner = fuse_lock_owner_id(fc, id);
229 req->in.h.opcode = FUSE_FLUSH;
230 req->in.h.nodeid = get_node_id(inode);
232 req->in.args[0].size = sizeof(inarg);
233 req->in.args[0].value = &inarg;
235 request_send(fc, req);
236 err = req->out.h.error;
237 fuse_put_request(fc, req);
238 if (err == -ENOSYS) {
245 int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
248 struct inode *inode = de->d_inode;
249 struct fuse_conn *fc = get_fuse_conn(inode);
250 struct fuse_file *ff = file->private_data;
251 struct fuse_req *req;
252 struct fuse_fsync_in inarg;
255 if (is_bad_inode(inode))
258 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
261 req = fuse_get_req(fc);
265 memset(&inarg, 0, sizeof(inarg));
267 inarg.fsync_flags = datasync ? 1 : 0;
268 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
269 req->in.h.nodeid = get_node_id(inode);
271 req->in.args[0].size = sizeof(inarg);
272 req->in.args[0].value = &inarg;
273 request_send(fc, req);
274 err = req->out.h.error;
275 fuse_put_request(fc, req);
276 if (err == -ENOSYS) {
286 static int fuse_fsync(struct file *file, struct dentry *de, int datasync)
288 return fuse_fsync_common(file, de, datasync, 0);
291 void fuse_read_fill(struct fuse_req *req, struct fuse_file *ff,
292 struct inode *inode, loff_t pos, size_t count, int opcode)
294 struct fuse_read_in *inarg = &req->misc.read_in;
299 req->in.h.opcode = opcode;
300 req->in.h.nodeid = get_node_id(inode);
302 req->in.args[0].size = sizeof(struct fuse_read_in);
303 req->in.args[0].value = inarg;
304 req->out.argpages = 1;
306 req->out.numargs = 1;
307 req->out.args[0].size = count;
310 static size_t fuse_send_read(struct fuse_req *req, struct file *file,
311 struct inode *inode, loff_t pos, size_t count,
314 struct fuse_conn *fc = get_fuse_conn(inode);
315 struct fuse_file *ff = file->private_data;
317 fuse_read_fill(req, ff, inode, pos, count, FUSE_READ);
319 struct fuse_read_in *inarg = &req->misc.read_in;
321 inarg->read_flags |= FUSE_READ_LOCKOWNER;
322 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
324 request_send(fc, req);
325 return req->out.args[0].size;
328 static int fuse_readpage(struct file *file, struct page *page)
330 struct inode *inode = page->mapping->host;
331 struct fuse_conn *fc = get_fuse_conn(inode);
332 struct fuse_req *req;
336 if (is_bad_inode(inode))
339 req = fuse_get_req(fc);
344 req->out.page_zeroing = 1;
346 req->pages[0] = page;
347 fuse_send_read(req, file, inode, page_offset(page), PAGE_CACHE_SIZE,
349 err = req->out.h.error;
350 fuse_put_request(fc, req);
352 SetPageUptodate(page);
353 fuse_invalidate_attr(inode); /* atime changed */
359 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
363 fuse_invalidate_attr(req->pages[0]->mapping->host); /* atime changed */
365 for (i = 0; i < req->num_pages; i++) {
366 struct page *page = req->pages[i];
367 if (!req->out.h.error)
368 SetPageUptodate(page);
374 fuse_file_put(req->ff);
375 fuse_put_request(fc, req);
378 static void fuse_send_readpages(struct fuse_req *req, struct fuse_file *ff,
381 struct fuse_conn *fc = get_fuse_conn(inode);
382 loff_t pos = page_offset(req->pages[0]);
383 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
384 req->out.page_zeroing = 1;
385 fuse_read_fill(req, ff, inode, pos, count, FUSE_READ);
386 if (fc->async_read) {
387 req->ff = fuse_file_get(ff);
388 req->end = fuse_readpages_end;
389 request_send_background(fc, req);
391 request_send(fc, req);
392 fuse_readpages_end(fc, req);
396 struct fuse_fill_data {
397 struct fuse_req *req;
398 struct fuse_file *ff;
402 static int fuse_readpages_fill(void *_data, struct page *page)
404 struct fuse_fill_data *data = _data;
405 struct fuse_req *req = data->req;
406 struct inode *inode = data->inode;
407 struct fuse_conn *fc = get_fuse_conn(inode);
409 if (req->num_pages &&
410 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
411 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
412 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
413 fuse_send_readpages(req, data->ff, inode);
414 data->req = req = fuse_get_req(fc);
420 req->pages[req->num_pages] = page;
425 static int fuse_readpages(struct file *file, struct address_space *mapping,
426 struct list_head *pages, unsigned nr_pages)
428 struct inode *inode = mapping->host;
429 struct fuse_conn *fc = get_fuse_conn(inode);
430 struct fuse_fill_data data;
434 if (is_bad_inode(inode))
437 data.ff = file->private_data;
439 data.req = fuse_get_req(fc);
440 err = PTR_ERR(data.req);
441 if (IS_ERR(data.req))
444 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
446 if (data.req->num_pages)
447 fuse_send_readpages(data.req, data.ff, inode);
449 fuse_put_request(fc, data.req);
455 static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff,
456 struct inode *inode, loff_t pos, size_t count,
459 struct fuse_conn *fc = get_fuse_conn(inode);
460 struct fuse_write_in *inarg = &req->misc.write.in;
461 struct fuse_write_out *outarg = &req->misc.write.out;
463 memset(inarg, 0, sizeof(struct fuse_write_in));
467 inarg->write_flags = writepage ? FUSE_WRITE_CACHE : 0;
468 req->in.h.opcode = FUSE_WRITE;
469 req->in.h.nodeid = get_node_id(inode);
470 req->in.argpages = 1;
473 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
475 req->in.args[0].size = sizeof(struct fuse_write_in);
476 req->in.args[0].value = inarg;
477 req->in.args[1].size = count;
478 req->out.numargs = 1;
479 req->out.args[0].size = sizeof(struct fuse_write_out);
480 req->out.args[0].value = outarg;
483 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
484 struct inode *inode, loff_t pos, size_t count,
487 struct fuse_conn *fc = get_fuse_conn(inode);
488 fuse_write_fill(req, file->private_data, inode, pos, count, 0);
490 struct fuse_write_in *inarg = &req->misc.write.in;
491 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
492 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
494 request_send(fc, req);
495 return req->misc.write.out.size;
498 static int fuse_write_begin(struct file *file, struct address_space *mapping,
499 loff_t pos, unsigned len, unsigned flags,
500 struct page **pagep, void **fsdata)
502 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
504 *pagep = __grab_cache_page(mapping, index);
510 static int fuse_buffered_write(struct file *file, struct inode *inode,
511 loff_t pos, unsigned count, struct page *page)
515 struct fuse_conn *fc = get_fuse_conn(inode);
516 struct fuse_inode *fi = get_fuse_inode(inode);
517 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
518 struct fuse_req *req;
520 if (is_bad_inode(inode))
523 req = fuse_get_req(fc);
528 req->pages[0] = page;
529 req->page_offset = offset;
530 nres = fuse_send_write(req, file, inode, pos, count, NULL);
531 err = req->out.h.error;
532 fuse_put_request(fc, req);
537 spin_lock(&fc->lock);
538 fi->attr_version = ++fc->attr_version;
539 if (pos > inode->i_size)
540 i_size_write(inode, pos);
541 spin_unlock(&fc->lock);
543 if (count == PAGE_CACHE_SIZE)
544 SetPageUptodate(page);
546 fuse_invalidate_attr(inode);
547 return err ? err : nres;
550 static int fuse_write_end(struct file *file, struct address_space *mapping,
551 loff_t pos, unsigned len, unsigned copied,
552 struct page *page, void *fsdata)
554 struct inode *inode = mapping->host;
558 res = fuse_buffered_write(file, inode, pos, copied, page);
561 page_cache_release(page);
565 static void fuse_release_user_pages(struct fuse_req *req, int write)
569 for (i = 0; i < req->num_pages; i++) {
570 struct page *page = req->pages[i];
572 set_page_dirty_lock(page);
577 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
578 unsigned nbytes, int write)
580 unsigned long user_addr = (unsigned long) buf;
581 unsigned offset = user_addr & ~PAGE_MASK;
584 /* This doesn't work with nfsd */
588 nbytes = min(nbytes, (unsigned) FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
589 npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
590 npages = min(max(npages, 1), FUSE_MAX_PAGES_PER_REQ);
591 down_read(¤t->mm->mmap_sem);
592 npages = get_user_pages(current, current->mm, user_addr, npages, write,
593 0, req->pages, NULL);
594 up_read(¤t->mm->mmap_sem);
598 req->num_pages = npages;
599 req->page_offset = offset;
603 static ssize_t fuse_direct_io(struct file *file, const char __user *buf,
604 size_t count, loff_t *ppos, int write)
606 struct inode *inode = file->f_path.dentry->d_inode;
607 struct fuse_conn *fc = get_fuse_conn(inode);
608 size_t nmax = write ? fc->max_write : fc->max_read;
611 struct fuse_req *req;
613 if (is_bad_inode(inode))
616 req = fuse_get_req(fc);
622 size_t nbytes = min(count, nmax);
623 int err = fuse_get_user_pages(req, buf, nbytes, !write);
628 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset;
629 nbytes = min(count, nbytes);
631 nres = fuse_send_write(req, file, inode, pos, nbytes,
634 nres = fuse_send_read(req, file, inode, pos, nbytes,
636 fuse_release_user_pages(req, !write);
637 if (req->out.h.error) {
639 res = req->out.h.error;
641 } else if (nres > nbytes) {
652 fuse_put_request(fc, req);
653 req = fuse_get_req(fc);
658 fuse_put_request(fc, req);
661 spin_lock(&fc->lock);
662 if (pos > inode->i_size)
663 i_size_write(inode, pos);
664 spin_unlock(&fc->lock);
668 fuse_invalidate_attr(inode);
673 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
674 size_t count, loff_t *ppos)
676 return fuse_direct_io(file, buf, count, ppos, 0);
679 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
680 size_t count, loff_t *ppos)
682 struct inode *inode = file->f_path.dentry->d_inode;
684 /* Don't allow parallel writes to the same file */
685 mutex_lock(&inode->i_mutex);
686 res = generic_write_checks(file, ppos, &count, 0);
688 res = fuse_direct_io(file, buf, count, ppos, 1);
689 mutex_unlock(&inode->i_mutex);
693 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
695 if ((vma->vm_flags & VM_SHARED)) {
696 if ((vma->vm_flags & VM_WRITE))
699 vma->vm_flags &= ~VM_MAYWRITE;
701 return generic_file_mmap(file, vma);
704 static int fuse_set_page_dirty(struct page *page)
706 printk("fuse_set_page_dirty: should not happen\n");
711 static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
712 struct file_lock *fl)
720 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
721 ffl->end < ffl->start)
724 fl->fl_start = ffl->start;
725 fl->fl_end = ffl->end;
726 fl->fl_pid = ffl->pid;
732 fl->fl_type = ffl->type;
736 static void fuse_lk_fill(struct fuse_req *req, struct file *file,
737 const struct file_lock *fl, int opcode, pid_t pid,
740 struct inode *inode = file->f_path.dentry->d_inode;
741 struct fuse_conn *fc = get_fuse_conn(inode);
742 struct fuse_file *ff = file->private_data;
743 struct fuse_lk_in *arg = &req->misc.lk_in;
746 arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
747 arg->lk.start = fl->fl_start;
748 arg->lk.end = fl->fl_end;
749 arg->lk.type = fl->fl_type;
752 arg->lk_flags |= FUSE_LK_FLOCK;
753 req->in.h.opcode = opcode;
754 req->in.h.nodeid = get_node_id(inode);
756 req->in.args[0].size = sizeof(*arg);
757 req->in.args[0].value = arg;
760 static int fuse_getlk(struct file *file, struct file_lock *fl)
762 struct inode *inode = file->f_path.dentry->d_inode;
763 struct fuse_conn *fc = get_fuse_conn(inode);
764 struct fuse_req *req;
765 struct fuse_lk_out outarg;
768 req = fuse_get_req(fc);
772 fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0);
773 req->out.numargs = 1;
774 req->out.args[0].size = sizeof(outarg);
775 req->out.args[0].value = &outarg;
776 request_send(fc, req);
777 err = req->out.h.error;
778 fuse_put_request(fc, req);
780 err = convert_fuse_file_lock(&outarg.lk, fl);
785 static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
787 struct inode *inode = file->f_path.dentry->d_inode;
788 struct fuse_conn *fc = get_fuse_conn(inode);
789 struct fuse_req *req;
790 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
791 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
794 /* Unlock on close is handled by the flush method */
795 if (fl->fl_flags & FL_CLOSE)
798 req = fuse_get_req(fc);
802 fuse_lk_fill(req, file, fl, opcode, pid, flock);
803 request_send(fc, req);
804 err = req->out.h.error;
805 /* locking is restartable */
808 fuse_put_request(fc, req);
812 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
814 struct inode *inode = file->f_path.dentry->d_inode;
815 struct fuse_conn *fc = get_fuse_conn(inode);
818 if (cmd == F_GETLK) {
820 posix_test_lock(file, fl);
823 err = fuse_getlk(file, fl);
826 err = posix_lock_file_wait(file, fl);
828 err = fuse_setlk(file, fl, 0);
833 static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
835 struct inode *inode = file->f_path.dentry->d_inode;
836 struct fuse_conn *fc = get_fuse_conn(inode);
840 err = flock_lock_file_wait(file, fl);
842 /* emulate flock with POSIX locks */
843 fl->fl_owner = (fl_owner_t) file;
844 err = fuse_setlk(file, fl, 1);
850 static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
852 struct inode *inode = mapping->host;
853 struct fuse_conn *fc = get_fuse_conn(inode);
854 struct fuse_req *req;
855 struct fuse_bmap_in inarg;
856 struct fuse_bmap_out outarg;
859 if (!inode->i_sb->s_bdev || fc->no_bmap)
862 req = fuse_get_req(fc);
866 memset(&inarg, 0, sizeof(inarg));
868 inarg.blocksize = inode->i_sb->s_blocksize;
869 req->in.h.opcode = FUSE_BMAP;
870 req->in.h.nodeid = get_node_id(inode);
872 req->in.args[0].size = sizeof(inarg);
873 req->in.args[0].value = &inarg;
874 req->out.numargs = 1;
875 req->out.args[0].size = sizeof(outarg);
876 req->out.args[0].value = &outarg;
877 request_send(fc, req);
878 err = req->out.h.error;
879 fuse_put_request(fc, req);
883 return err ? 0 : outarg.block;
886 static const struct file_operations fuse_file_operations = {
887 .llseek = generic_file_llseek,
888 .read = do_sync_read,
889 .aio_read = generic_file_aio_read,
890 .write = do_sync_write,
891 .aio_write = generic_file_aio_write,
892 .mmap = fuse_file_mmap,
895 .release = fuse_release,
897 .lock = fuse_file_lock,
898 .flock = fuse_file_flock,
899 .splice_read = generic_file_splice_read,
902 static const struct file_operations fuse_direct_io_file_operations = {
903 .llseek = generic_file_llseek,
904 .read = fuse_direct_read,
905 .write = fuse_direct_write,
908 .release = fuse_release,
910 .lock = fuse_file_lock,
911 .flock = fuse_file_flock,
912 /* no mmap and splice_read */
915 static const struct address_space_operations fuse_file_aops = {
916 .readpage = fuse_readpage,
917 .write_begin = fuse_write_begin,
918 .write_end = fuse_write_end,
919 .readpages = fuse_readpages,
920 .set_page_dirty = fuse_set_page_dirty,
924 void fuse_init_file_inode(struct inode *inode)
926 inode->i_fop = &fuse_file_operations;
927 inode->i_data.a_ops = &fuse_file_aops;