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 static 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)
313 struct fuse_conn *fc = get_fuse_conn(inode);
314 struct fuse_file *ff = file->private_data;
315 fuse_read_fill(req, ff, inode, pos, count, FUSE_READ);
316 request_send(fc, req);
317 return req->out.args[0].size;
320 static int fuse_readpage(struct file *file, struct page *page)
322 struct inode *inode = page->mapping->host;
323 struct fuse_conn *fc = get_fuse_conn(inode);
324 struct fuse_req *req;
328 if (is_bad_inode(inode))
331 req = fuse_get_req(fc);
336 req->out.page_zeroing = 1;
338 req->pages[0] = page;
339 fuse_send_read(req, file, inode, page_offset(page), PAGE_CACHE_SIZE);
340 err = req->out.h.error;
341 fuse_put_request(fc, req);
343 SetPageUptodate(page);
344 fuse_invalidate_attr(inode); /* atime changed */
350 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
354 fuse_invalidate_attr(req->pages[0]->mapping->host); /* atime changed */
356 for (i = 0; i < req->num_pages; i++) {
357 struct page *page = req->pages[i];
358 if (!req->out.h.error)
359 SetPageUptodate(page);
365 fuse_file_put(req->ff);
366 fuse_put_request(fc, req);
369 static void fuse_send_readpages(struct fuse_req *req, struct fuse_file *ff,
372 struct fuse_conn *fc = get_fuse_conn(inode);
373 loff_t pos = page_offset(req->pages[0]);
374 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
375 req->out.page_zeroing = 1;
376 fuse_read_fill(req, ff, inode, pos, count, FUSE_READ);
377 if (fc->async_read) {
378 req->ff = fuse_file_get(ff);
379 req->end = fuse_readpages_end;
380 request_send_background(fc, req);
382 request_send(fc, req);
383 fuse_readpages_end(fc, req);
387 struct fuse_fill_data {
388 struct fuse_req *req;
389 struct fuse_file *ff;
393 static int fuse_readpages_fill(void *_data, struct page *page)
395 struct fuse_fill_data *data = _data;
396 struct fuse_req *req = data->req;
397 struct inode *inode = data->inode;
398 struct fuse_conn *fc = get_fuse_conn(inode);
400 if (req->num_pages &&
401 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
402 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
403 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
404 fuse_send_readpages(req, data->ff, inode);
405 data->req = req = fuse_get_req(fc);
411 req->pages[req->num_pages] = page;
416 static int fuse_readpages(struct file *file, struct address_space *mapping,
417 struct list_head *pages, unsigned nr_pages)
419 struct inode *inode = mapping->host;
420 struct fuse_conn *fc = get_fuse_conn(inode);
421 struct fuse_fill_data data;
425 if (is_bad_inode(inode))
428 data.ff = file->private_data;
430 data.req = fuse_get_req(fc);
431 err = PTR_ERR(data.req);
432 if (IS_ERR(data.req))
435 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
437 if (data.req->num_pages)
438 fuse_send_readpages(data.req, data.ff, inode);
440 fuse_put_request(fc, data.req);
446 static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff,
447 struct inode *inode, loff_t pos, size_t count,
450 struct fuse_write_in *inarg = &req->misc.write.in;
451 struct fuse_write_out *outarg = &req->misc.write.out;
453 memset(inarg, 0, sizeof(struct fuse_write_in));
457 inarg->write_flags = writepage ? FUSE_WRITE_CACHE : 0;
458 req->in.h.opcode = FUSE_WRITE;
459 req->in.h.nodeid = get_node_id(inode);
460 req->in.argpages = 1;
462 req->in.args[0].size = sizeof(struct fuse_write_in);
463 req->in.args[0].value = inarg;
464 req->in.args[1].size = count;
465 req->out.numargs = 1;
466 req->out.args[0].size = sizeof(struct fuse_write_out);
467 req->out.args[0].value = outarg;
470 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
471 struct inode *inode, loff_t pos, size_t count)
473 struct fuse_conn *fc = get_fuse_conn(inode);
474 fuse_write_fill(req, file->private_data, inode, pos, count, 0);
475 request_send(fc, req);
476 return req->misc.write.out.size;
479 static int fuse_write_begin(struct file *file, struct address_space *mapping,
480 loff_t pos, unsigned len, unsigned flags,
481 struct page **pagep, void **fsdata)
483 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
485 *pagep = __grab_cache_page(mapping, index);
491 static int fuse_buffered_write(struct file *file, struct inode *inode,
492 loff_t pos, unsigned count, struct page *page)
496 struct fuse_conn *fc = get_fuse_conn(inode);
497 struct fuse_inode *fi = get_fuse_inode(inode);
498 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
499 struct fuse_req *req;
501 if (is_bad_inode(inode))
504 req = fuse_get_req(fc);
509 req->pages[0] = page;
510 req->page_offset = offset;
511 nres = fuse_send_write(req, file, inode, pos, count);
512 err = req->out.h.error;
513 fuse_put_request(fc, req);
518 spin_lock(&fc->lock);
519 fi->attr_version = ++fc->attr_version;
520 if (pos > inode->i_size)
521 i_size_write(inode, pos);
522 spin_unlock(&fc->lock);
524 if (count == PAGE_CACHE_SIZE)
525 SetPageUptodate(page);
527 fuse_invalidate_attr(inode);
528 return err ? err : nres;
531 static int fuse_write_end(struct file *file, struct address_space *mapping,
532 loff_t pos, unsigned len, unsigned copied,
533 struct page *page, void *fsdata)
535 struct inode *inode = mapping->host;
539 res = fuse_buffered_write(file, inode, pos, copied, page);
542 page_cache_release(page);
546 static void fuse_release_user_pages(struct fuse_req *req, int write)
550 for (i = 0; i < req->num_pages; i++) {
551 struct page *page = req->pages[i];
553 set_page_dirty_lock(page);
558 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
559 unsigned nbytes, int write)
561 unsigned long user_addr = (unsigned long) buf;
562 unsigned offset = user_addr & ~PAGE_MASK;
565 /* This doesn't work with nfsd */
569 nbytes = min(nbytes, (unsigned) FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
570 npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
571 npages = min(max(npages, 1), FUSE_MAX_PAGES_PER_REQ);
572 down_read(¤t->mm->mmap_sem);
573 npages = get_user_pages(current, current->mm, user_addr, npages, write,
574 0, req->pages, NULL);
575 up_read(¤t->mm->mmap_sem);
579 req->num_pages = npages;
580 req->page_offset = offset;
584 static ssize_t fuse_direct_io(struct file *file, const char __user *buf,
585 size_t count, loff_t *ppos, int write)
587 struct inode *inode = file->f_path.dentry->d_inode;
588 struct fuse_conn *fc = get_fuse_conn(inode);
589 size_t nmax = write ? fc->max_write : fc->max_read;
592 struct fuse_req *req;
594 if (is_bad_inode(inode))
597 req = fuse_get_req(fc);
603 size_t nbytes = min(count, nmax);
604 int err = fuse_get_user_pages(req, buf, nbytes, !write);
609 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset;
610 nbytes = min(count, nbytes);
612 nres = fuse_send_write(req, file, inode, pos, nbytes);
614 nres = fuse_send_read(req, file, inode, pos, nbytes);
615 fuse_release_user_pages(req, !write);
616 if (req->out.h.error) {
618 res = req->out.h.error;
620 } else if (nres > nbytes) {
631 fuse_put_request(fc, req);
632 req = fuse_get_req(fc);
637 fuse_put_request(fc, req);
640 spin_lock(&fc->lock);
641 if (pos > inode->i_size)
642 i_size_write(inode, pos);
643 spin_unlock(&fc->lock);
647 fuse_invalidate_attr(inode);
652 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
653 size_t count, loff_t *ppos)
655 return fuse_direct_io(file, buf, count, ppos, 0);
658 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
659 size_t count, loff_t *ppos)
661 struct inode *inode = file->f_path.dentry->d_inode;
663 /* Don't allow parallel writes to the same file */
664 mutex_lock(&inode->i_mutex);
665 res = generic_write_checks(file, ppos, &count, 0);
667 res = fuse_direct_io(file, buf, count, ppos, 1);
668 mutex_unlock(&inode->i_mutex);
672 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
674 if ((vma->vm_flags & VM_SHARED)) {
675 if ((vma->vm_flags & VM_WRITE))
678 vma->vm_flags &= ~VM_MAYWRITE;
680 return generic_file_mmap(file, vma);
683 static int fuse_set_page_dirty(struct page *page)
685 printk("fuse_set_page_dirty: should not happen\n");
690 static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
691 struct file_lock *fl)
699 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
700 ffl->end < ffl->start)
703 fl->fl_start = ffl->start;
704 fl->fl_end = ffl->end;
705 fl->fl_pid = ffl->pid;
711 fl->fl_type = ffl->type;
715 static void fuse_lk_fill(struct fuse_req *req, struct file *file,
716 const struct file_lock *fl, int opcode, pid_t pid,
719 struct inode *inode = file->f_path.dentry->d_inode;
720 struct fuse_conn *fc = get_fuse_conn(inode);
721 struct fuse_file *ff = file->private_data;
722 struct fuse_lk_in *arg = &req->misc.lk_in;
725 arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
726 arg->lk.start = fl->fl_start;
727 arg->lk.end = fl->fl_end;
728 arg->lk.type = fl->fl_type;
731 arg->lk_flags |= FUSE_LK_FLOCK;
732 req->in.h.opcode = opcode;
733 req->in.h.nodeid = get_node_id(inode);
735 req->in.args[0].size = sizeof(*arg);
736 req->in.args[0].value = arg;
739 static int fuse_getlk(struct file *file, struct file_lock *fl)
741 struct inode *inode = file->f_path.dentry->d_inode;
742 struct fuse_conn *fc = get_fuse_conn(inode);
743 struct fuse_req *req;
744 struct fuse_lk_out outarg;
747 req = fuse_get_req(fc);
751 fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0);
752 req->out.numargs = 1;
753 req->out.args[0].size = sizeof(outarg);
754 req->out.args[0].value = &outarg;
755 request_send(fc, req);
756 err = req->out.h.error;
757 fuse_put_request(fc, req);
759 err = convert_fuse_file_lock(&outarg.lk, fl);
764 static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
766 struct inode *inode = file->f_path.dentry->d_inode;
767 struct fuse_conn *fc = get_fuse_conn(inode);
768 struct fuse_req *req;
769 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
770 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
773 /* Unlock on close is handled by the flush method */
774 if (fl->fl_flags & FL_CLOSE)
777 req = fuse_get_req(fc);
781 fuse_lk_fill(req, file, fl, opcode, pid, flock);
782 request_send(fc, req);
783 err = req->out.h.error;
784 /* locking is restartable */
787 fuse_put_request(fc, req);
791 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
793 struct inode *inode = file->f_path.dentry->d_inode;
794 struct fuse_conn *fc = get_fuse_conn(inode);
797 if (cmd == F_GETLK) {
799 posix_test_lock(file, fl);
802 err = fuse_getlk(file, fl);
805 err = posix_lock_file_wait(file, fl);
807 err = fuse_setlk(file, fl, 0);
812 static int fuse_file_flock(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);
819 err = flock_lock_file_wait(file, fl);
821 /* emulate flock with POSIX locks */
822 fl->fl_owner = (fl_owner_t) file;
823 err = fuse_setlk(file, fl, 1);
829 static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
831 struct inode *inode = mapping->host;
832 struct fuse_conn *fc = get_fuse_conn(inode);
833 struct fuse_req *req;
834 struct fuse_bmap_in inarg;
835 struct fuse_bmap_out outarg;
838 if (!inode->i_sb->s_bdev || fc->no_bmap)
841 req = fuse_get_req(fc);
845 memset(&inarg, 0, sizeof(inarg));
847 inarg.blocksize = inode->i_sb->s_blocksize;
848 req->in.h.opcode = FUSE_BMAP;
849 req->in.h.nodeid = get_node_id(inode);
851 req->in.args[0].size = sizeof(inarg);
852 req->in.args[0].value = &inarg;
853 req->out.numargs = 1;
854 req->out.args[0].size = sizeof(outarg);
855 req->out.args[0].value = &outarg;
856 request_send(fc, req);
857 err = req->out.h.error;
858 fuse_put_request(fc, req);
862 return err ? 0 : outarg.block;
865 static const struct file_operations fuse_file_operations = {
866 .llseek = generic_file_llseek,
867 .read = do_sync_read,
868 .aio_read = generic_file_aio_read,
869 .write = do_sync_write,
870 .aio_write = generic_file_aio_write,
871 .mmap = fuse_file_mmap,
874 .release = fuse_release,
876 .lock = fuse_file_lock,
877 .flock = fuse_file_flock,
878 .splice_read = generic_file_splice_read,
881 static const struct file_operations fuse_direct_io_file_operations = {
882 .llseek = generic_file_llseek,
883 .read = fuse_direct_read,
884 .write = fuse_direct_write,
887 .release = fuse_release,
889 .lock = fuse_file_lock,
890 .flock = fuse_file_flock,
891 /* no mmap and splice_read */
894 static const struct address_space_operations fuse_file_aops = {
895 .readpage = fuse_readpage,
896 .write_begin = fuse_write_begin,
897 .write_end = fuse_write_end,
898 .readpages = fuse_readpages,
899 .set_page_dirty = fuse_set_page_dirty,
903 void fuse_init_file_inode(struct inode *inode)
905 inode->i_fop = &fuse_file_operations;
906 inode->i_data.a_ops = &fuse_file_aops;