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);
66 void fuse_file_free(struct fuse_file *ff)
68 fuse_request_free(ff->reserved_req);
72 static struct fuse_file *fuse_file_get(struct fuse_file *ff)
74 atomic_inc(&ff->count);
78 static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
81 mntput(req->vfsmount);
82 fuse_put_request(fc, req);
85 static void fuse_file_put(struct fuse_file *ff)
87 if (atomic_dec_and_test(&ff->count)) {
88 struct fuse_req *req = ff->reserved_req;
89 struct fuse_conn *fc = get_fuse_conn(req->dentry->d_inode);
90 req->end = fuse_release_end;
91 request_send_background(fc, req);
96 void fuse_finish_open(struct inode *inode, struct file *file,
97 struct fuse_file *ff, struct fuse_open_out *outarg)
99 if (outarg->open_flags & FOPEN_DIRECT_IO)
100 file->f_op = &fuse_direct_io_file_operations;
101 if (!(outarg->open_flags & FOPEN_KEEP_CACHE))
102 invalidate_inode_pages2(inode->i_mapping);
104 file->private_data = fuse_file_get(ff);
107 int fuse_open_common(struct inode *inode, struct file *file, int isdir)
109 struct fuse_open_out outarg;
110 struct fuse_file *ff;
113 /* VFS checks this, but only _after_ ->open() */
114 if (file->f_flags & O_DIRECT)
117 err = generic_file_open(inode, file);
121 ff = fuse_file_alloc();
125 err = fuse_send_open(inode, file, isdir, &outarg);
130 outarg.open_flags &= ~FOPEN_DIRECT_IO;
131 fuse_finish_open(inode, file, ff, &outarg);
137 void fuse_release_fill(struct fuse_file *ff, u64 nodeid, int flags, int opcode)
139 struct fuse_req *req = ff->reserved_req;
140 struct fuse_release_in *inarg = &req->misc.release_in;
143 inarg->flags = flags;
144 req->in.h.opcode = opcode;
145 req->in.h.nodeid = nodeid;
147 req->in.args[0].size = sizeof(struct fuse_release_in);
148 req->in.args[0].value = inarg;
151 int fuse_release_common(struct inode *inode, struct file *file, int isdir)
153 struct fuse_file *ff = file->private_data;
155 struct fuse_conn *fc = get_fuse_conn(inode);
157 fuse_release_fill(ff, get_node_id(inode), file->f_flags,
158 isdir ? FUSE_RELEASEDIR : FUSE_RELEASE);
160 /* Hold vfsmount and dentry until release is finished */
161 ff->reserved_req->vfsmount = mntget(file->f_path.mnt);
162 ff->reserved_req->dentry = dget(file->f_path.dentry);
164 spin_lock(&fc->lock);
165 list_del(&ff->write_entry);
166 spin_unlock(&fc->lock);
168 * Normally this will send the RELEASE request,
169 * however if some asynchronous READ or WRITE requests
170 * are outstanding, the sending will be delayed
175 /* Return value is ignored by VFS */
179 static int fuse_open(struct inode *inode, struct file *file)
181 return fuse_open_common(inode, file, 0);
184 static int fuse_release(struct inode *inode, struct file *file)
186 return fuse_release_common(inode, file, 0);
190 * Scramble the ID space with XTEA, so that the value of the files_struct
191 * pointer is not exposed to userspace.
193 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
195 u32 *k = fc->scramble_key;
196 u64 v = (unsigned long) id;
202 for (i = 0; i < 32; i++) {
203 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
205 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
208 return (u64) v0 + ((u64) v1 << 32);
211 static int fuse_flush(struct file *file, fl_owner_t id)
213 struct inode *inode = file->f_path.dentry->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_flush_in inarg;
220 if (is_bad_inode(inode))
226 req = fuse_get_req_nofail(fc, file);
227 memset(&inarg, 0, sizeof(inarg));
229 inarg.lock_owner = fuse_lock_owner_id(fc, id);
230 req->in.h.opcode = FUSE_FLUSH;
231 req->in.h.nodeid = get_node_id(inode);
233 req->in.args[0].size = sizeof(inarg);
234 req->in.args[0].value = &inarg;
236 request_send(fc, req);
237 err = req->out.h.error;
238 fuse_put_request(fc, req);
239 if (err == -ENOSYS) {
246 int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
249 struct inode *inode = de->d_inode;
250 struct fuse_conn *fc = get_fuse_conn(inode);
251 struct fuse_file *ff = file->private_data;
252 struct fuse_req *req;
253 struct fuse_fsync_in inarg;
256 if (is_bad_inode(inode))
259 if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
262 req = fuse_get_req(fc);
266 memset(&inarg, 0, sizeof(inarg));
268 inarg.fsync_flags = datasync ? 1 : 0;
269 req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
270 req->in.h.nodeid = get_node_id(inode);
272 req->in.args[0].size = sizeof(inarg);
273 req->in.args[0].value = &inarg;
274 request_send(fc, req);
275 err = req->out.h.error;
276 fuse_put_request(fc, req);
277 if (err == -ENOSYS) {
287 static int fuse_fsync(struct file *file, struct dentry *de, int datasync)
289 return fuse_fsync_common(file, de, datasync, 0);
292 void fuse_read_fill(struct fuse_req *req, struct file *file,
293 struct inode *inode, loff_t pos, size_t count, int opcode)
295 struct fuse_read_in *inarg = &req->misc.read_in;
296 struct fuse_file *ff = file->private_data;
301 inarg->flags = file->f_flags;
302 req->in.h.opcode = opcode;
303 req->in.h.nodeid = get_node_id(inode);
305 req->in.args[0].size = sizeof(struct fuse_read_in);
306 req->in.args[0].value = inarg;
307 req->out.argpages = 1;
309 req->out.numargs = 1;
310 req->out.args[0].size = count;
313 static size_t fuse_send_read(struct fuse_req *req, struct file *file,
314 struct inode *inode, loff_t pos, size_t count,
317 struct fuse_conn *fc = get_fuse_conn(inode);
319 fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
321 struct fuse_read_in *inarg = &req->misc.read_in;
323 inarg->read_flags |= FUSE_READ_LOCKOWNER;
324 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
326 request_send(fc, req);
327 return req->out.args[0].size;
330 static int fuse_readpage(struct file *file, struct page *page)
332 struct inode *inode = page->mapping->host;
333 struct fuse_conn *fc = get_fuse_conn(inode);
334 struct fuse_req *req;
338 if (is_bad_inode(inode))
341 req = fuse_get_req(fc);
346 req->out.page_zeroing = 1;
348 req->pages[0] = page;
349 fuse_send_read(req, file, inode, page_offset(page), PAGE_CACHE_SIZE,
351 err = req->out.h.error;
352 fuse_put_request(fc, req);
354 SetPageUptodate(page);
355 fuse_invalidate_attr(inode); /* atime changed */
361 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
365 fuse_invalidate_attr(req->pages[0]->mapping->host); /* atime changed */
367 for (i = 0; i < req->num_pages; i++) {
368 struct page *page = req->pages[i];
369 if (!req->out.h.error)
370 SetPageUptodate(page);
376 fuse_file_put(req->ff);
377 fuse_put_request(fc, req);
380 static void fuse_send_readpages(struct fuse_req *req, struct file *file,
383 struct fuse_conn *fc = get_fuse_conn(inode);
384 loff_t pos = page_offset(req->pages[0]);
385 size_t count = req->num_pages << PAGE_CACHE_SHIFT;
386 req->out.page_zeroing = 1;
387 fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
388 if (fc->async_read) {
389 struct fuse_file *ff = file->private_data;
390 req->ff = fuse_file_get(ff);
391 req->end = fuse_readpages_end;
392 request_send_background(fc, req);
394 request_send(fc, req);
395 fuse_readpages_end(fc, req);
399 struct fuse_fill_data {
400 struct fuse_req *req;
405 static int fuse_readpages_fill(void *_data, struct page *page)
407 struct fuse_fill_data *data = _data;
408 struct fuse_req *req = data->req;
409 struct inode *inode = data->inode;
410 struct fuse_conn *fc = get_fuse_conn(inode);
412 if (req->num_pages &&
413 (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
414 (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
415 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
416 fuse_send_readpages(req, data->file, inode);
417 data->req = req = fuse_get_req(fc);
423 req->pages[req->num_pages] = page;
428 static int fuse_readpages(struct file *file, struct address_space *mapping,
429 struct list_head *pages, unsigned nr_pages)
431 struct inode *inode = mapping->host;
432 struct fuse_conn *fc = get_fuse_conn(inode);
433 struct fuse_fill_data data;
437 if (is_bad_inode(inode))
442 data.req = fuse_get_req(fc);
443 err = PTR_ERR(data.req);
444 if (IS_ERR(data.req))
447 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
449 if (data.req->num_pages)
450 fuse_send_readpages(data.req, file, inode);
452 fuse_put_request(fc, data.req);
458 static ssize_t fuse_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
459 unsigned long nr_segs, loff_t pos)
461 struct inode *inode = iocb->ki_filp->f_mapping->host;
463 if (pos + iov_length(iov, nr_segs) > i_size_read(inode)) {
466 * If trying to read past EOF, make sure the i_size
467 * attribute is up-to-date.
469 err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL);
474 return generic_file_aio_read(iocb, iov, nr_segs, pos);
477 static void fuse_write_fill(struct fuse_req *req, struct file *file,
478 struct inode *inode, loff_t pos, size_t count,
481 struct fuse_conn *fc = get_fuse_conn(inode);
482 struct fuse_file *ff = file->private_data;
483 struct fuse_write_in *inarg = &req->misc.write.in;
484 struct fuse_write_out *outarg = &req->misc.write.out;
486 memset(inarg, 0, sizeof(struct fuse_write_in));
490 inarg->write_flags = writepage ? FUSE_WRITE_CACHE : 0;
491 inarg->flags = file->f_flags;
492 req->in.h.opcode = FUSE_WRITE;
493 req->in.h.nodeid = get_node_id(inode);
494 req->in.argpages = 1;
497 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
499 req->in.args[0].size = sizeof(struct fuse_write_in);
500 req->in.args[0].value = inarg;
501 req->in.args[1].size = count;
502 req->out.numargs = 1;
503 req->out.args[0].size = sizeof(struct fuse_write_out);
504 req->out.args[0].value = outarg;
507 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
508 struct inode *inode, loff_t pos, size_t count,
511 struct fuse_conn *fc = get_fuse_conn(inode);
512 fuse_write_fill(req, file, inode, pos, count, 0);
514 struct fuse_write_in *inarg = &req->misc.write.in;
515 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
516 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
518 request_send(fc, req);
519 return req->misc.write.out.size;
522 static int fuse_write_begin(struct file *file, struct address_space *mapping,
523 loff_t pos, unsigned len, unsigned flags,
524 struct page **pagep, void **fsdata)
526 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
528 *pagep = __grab_cache_page(mapping, index);
534 static int fuse_buffered_write(struct file *file, struct inode *inode,
535 loff_t pos, unsigned count, struct page *page)
539 struct fuse_conn *fc = get_fuse_conn(inode);
540 struct fuse_inode *fi = get_fuse_inode(inode);
541 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
542 struct fuse_req *req;
544 if (is_bad_inode(inode))
547 req = fuse_get_req(fc);
552 req->pages[0] = page;
553 req->page_offset = offset;
554 nres = fuse_send_write(req, file, inode, pos, count, NULL);
555 err = req->out.h.error;
556 fuse_put_request(fc, req);
561 spin_lock(&fc->lock);
562 fi->attr_version = ++fc->attr_version;
563 if (pos > inode->i_size)
564 i_size_write(inode, pos);
565 spin_unlock(&fc->lock);
567 if (count == PAGE_CACHE_SIZE)
568 SetPageUptodate(page);
570 fuse_invalidate_attr(inode);
571 return err ? err : nres;
574 static int fuse_write_end(struct file *file, struct address_space *mapping,
575 loff_t pos, unsigned len, unsigned copied,
576 struct page *page, void *fsdata)
578 struct inode *inode = mapping->host;
582 res = fuse_buffered_write(file, inode, pos, copied, page);
585 page_cache_release(page);
589 static void fuse_release_user_pages(struct fuse_req *req, int write)
593 for (i = 0; i < req->num_pages; i++) {
594 struct page *page = req->pages[i];
596 set_page_dirty_lock(page);
601 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
602 unsigned nbytes, int write)
604 unsigned long user_addr = (unsigned long) buf;
605 unsigned offset = user_addr & ~PAGE_MASK;
608 /* This doesn't work with nfsd */
612 nbytes = min(nbytes, (unsigned) FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
613 npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
614 npages = min(max(npages, 1), FUSE_MAX_PAGES_PER_REQ);
615 down_read(¤t->mm->mmap_sem);
616 npages = get_user_pages(current, current->mm, user_addr, npages, write,
617 0, req->pages, NULL);
618 up_read(¤t->mm->mmap_sem);
622 req->num_pages = npages;
623 req->page_offset = offset;
627 static ssize_t fuse_direct_io(struct file *file, const char __user *buf,
628 size_t count, loff_t *ppos, int write)
630 struct inode *inode = file->f_path.dentry->d_inode;
631 struct fuse_conn *fc = get_fuse_conn(inode);
632 size_t nmax = write ? fc->max_write : fc->max_read;
635 struct fuse_req *req;
637 if (is_bad_inode(inode))
640 req = fuse_get_req(fc);
646 size_t nbytes = min(count, nmax);
647 int err = fuse_get_user_pages(req, buf, nbytes, !write);
652 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset;
653 nbytes = min(count, nbytes);
655 nres = fuse_send_write(req, file, inode, pos, nbytes,
658 nres = fuse_send_read(req, file, inode, pos, nbytes,
660 fuse_release_user_pages(req, !write);
661 if (req->out.h.error) {
663 res = req->out.h.error;
665 } else if (nres > nbytes) {
676 fuse_put_request(fc, req);
677 req = fuse_get_req(fc);
682 fuse_put_request(fc, req);
685 spin_lock(&fc->lock);
686 if (pos > inode->i_size)
687 i_size_write(inode, pos);
688 spin_unlock(&fc->lock);
692 fuse_invalidate_attr(inode);
697 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
698 size_t count, loff_t *ppos)
700 return fuse_direct_io(file, buf, count, ppos, 0);
703 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
704 size_t count, loff_t *ppos)
706 struct inode *inode = file->f_path.dentry->d_inode;
708 /* Don't allow parallel writes to the same file */
709 mutex_lock(&inode->i_mutex);
710 res = generic_write_checks(file, ppos, &count, 0);
712 res = fuse_direct_io(file, buf, count, ppos, 1);
713 mutex_unlock(&inode->i_mutex);
717 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
719 if ((vma->vm_flags & VM_SHARED)) {
720 if ((vma->vm_flags & VM_WRITE))
723 vma->vm_flags &= ~VM_MAYWRITE;
725 return generic_file_mmap(file, vma);
728 static int fuse_set_page_dirty(struct page *page)
730 printk("fuse_set_page_dirty: should not happen\n");
735 static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
736 struct file_lock *fl)
744 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
745 ffl->end < ffl->start)
748 fl->fl_start = ffl->start;
749 fl->fl_end = ffl->end;
750 fl->fl_pid = ffl->pid;
756 fl->fl_type = ffl->type;
760 static void fuse_lk_fill(struct fuse_req *req, struct file *file,
761 const struct file_lock *fl, int opcode, pid_t pid,
764 struct inode *inode = file->f_path.dentry->d_inode;
765 struct fuse_conn *fc = get_fuse_conn(inode);
766 struct fuse_file *ff = file->private_data;
767 struct fuse_lk_in *arg = &req->misc.lk_in;
770 arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
771 arg->lk.start = fl->fl_start;
772 arg->lk.end = fl->fl_end;
773 arg->lk.type = fl->fl_type;
776 arg->lk_flags |= FUSE_LK_FLOCK;
777 req->in.h.opcode = opcode;
778 req->in.h.nodeid = get_node_id(inode);
780 req->in.args[0].size = sizeof(*arg);
781 req->in.args[0].value = arg;
784 static int fuse_getlk(struct file *file, struct file_lock *fl)
786 struct inode *inode = file->f_path.dentry->d_inode;
787 struct fuse_conn *fc = get_fuse_conn(inode);
788 struct fuse_req *req;
789 struct fuse_lk_out outarg;
792 req = fuse_get_req(fc);
796 fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0);
797 req->out.numargs = 1;
798 req->out.args[0].size = sizeof(outarg);
799 req->out.args[0].value = &outarg;
800 request_send(fc, req);
801 err = req->out.h.error;
802 fuse_put_request(fc, req);
804 err = convert_fuse_file_lock(&outarg.lk, fl);
809 static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
811 struct inode *inode = file->f_path.dentry->d_inode;
812 struct fuse_conn *fc = get_fuse_conn(inode);
813 struct fuse_req *req;
814 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
815 pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
818 /* Unlock on close is handled by the flush method */
819 if (fl->fl_flags & FL_CLOSE)
822 req = fuse_get_req(fc);
826 fuse_lk_fill(req, file, fl, opcode, pid, flock);
827 request_send(fc, req);
828 err = req->out.h.error;
829 /* locking is restartable */
832 fuse_put_request(fc, req);
836 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
838 struct inode *inode = file->f_path.dentry->d_inode;
839 struct fuse_conn *fc = get_fuse_conn(inode);
842 if (cmd == F_GETLK) {
844 posix_test_lock(file, fl);
847 err = fuse_getlk(file, fl);
850 err = posix_lock_file_wait(file, fl);
852 err = fuse_setlk(file, fl, 0);
857 static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
859 struct inode *inode = file->f_path.dentry->d_inode;
860 struct fuse_conn *fc = get_fuse_conn(inode);
864 err = flock_lock_file_wait(file, fl);
866 /* emulate flock with POSIX locks */
867 fl->fl_owner = (fl_owner_t) file;
868 err = fuse_setlk(file, fl, 1);
874 static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
876 struct inode *inode = mapping->host;
877 struct fuse_conn *fc = get_fuse_conn(inode);
878 struct fuse_req *req;
879 struct fuse_bmap_in inarg;
880 struct fuse_bmap_out outarg;
883 if (!inode->i_sb->s_bdev || fc->no_bmap)
886 req = fuse_get_req(fc);
890 memset(&inarg, 0, sizeof(inarg));
892 inarg.blocksize = inode->i_sb->s_blocksize;
893 req->in.h.opcode = FUSE_BMAP;
894 req->in.h.nodeid = get_node_id(inode);
896 req->in.args[0].size = sizeof(inarg);
897 req->in.args[0].value = &inarg;
898 req->out.numargs = 1;
899 req->out.args[0].size = sizeof(outarg);
900 req->out.args[0].value = &outarg;
901 request_send(fc, req);
902 err = req->out.h.error;
903 fuse_put_request(fc, req);
907 return err ? 0 : outarg.block;
910 static const struct file_operations fuse_file_operations = {
911 .llseek = generic_file_llseek,
912 .read = do_sync_read,
913 .aio_read = fuse_file_aio_read,
914 .write = do_sync_write,
915 .aio_write = generic_file_aio_write,
916 .mmap = fuse_file_mmap,
919 .release = fuse_release,
921 .lock = fuse_file_lock,
922 .flock = fuse_file_flock,
923 .splice_read = generic_file_splice_read,
926 static const struct file_operations fuse_direct_io_file_operations = {
927 .llseek = generic_file_llseek,
928 .read = fuse_direct_read,
929 .write = fuse_direct_write,
932 .release = fuse_release,
934 .lock = fuse_file_lock,
935 .flock = fuse_file_flock,
936 /* no mmap and splice_read */
939 static const struct address_space_operations fuse_file_aops = {
940 .readpage = fuse_readpage,
941 .write_begin = fuse_write_begin,
942 .write_end = fuse_write_end,
943 .readpages = fuse_readpages,
944 .set_page_dirty = fuse_set_page_dirty,
948 void fuse_init_file_inode(struct inode *inode)
950 inode->i_fop = &fuse_file_operations;
951 inode->i_data.a_ops = &fuse_file_aops;