[IA64] Minimize per_cpu reservations.
[linux-2.6] / fs / fuse / file.c
1 /*
2   FUSE: Filesystem in Userspace
3   Copyright (C) 2001-2006  Miklos Szeredi <miklos@szeredi.hu>
4
5   This program can be distributed under the terms of the GNU GPL.
6   See the file COPYING.
7 */
8
9 #include "fuse_i.h"
10
11 #include <linux/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/kernel.h>
14 #include <linux/sched.h>
15
16 static const struct file_operations fuse_direct_io_file_operations;
17
18 static int fuse_send_open(struct inode *inode, struct file *file, int isdir,
19                           struct fuse_open_out *outargp)
20 {
21         struct fuse_conn *fc = get_fuse_conn(inode);
22         struct fuse_open_in inarg;
23         struct fuse_req *req;
24         int err;
25
26         req = fuse_get_req(fc);
27         if (IS_ERR(req))
28                 return PTR_ERR(req);
29
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);
36         req->in.numargs = 1;
37         req->in.args[0].size = sizeof(inarg);
38         req->in.args[0].value = &inarg;
39         req->out.numargs = 1;
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);
45
46         return err;
47 }
48
49 struct fuse_file *fuse_file_alloc(void)
50 {
51         struct fuse_file *ff;
52         ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
53         if (ff) {
54                 ff->reserved_req = fuse_request_alloc();
55                 if (!ff->reserved_req) {
56                         kfree(ff);
57                         ff = NULL;
58                 } else {
59                         INIT_LIST_HEAD(&ff->write_entry);
60                         atomic_set(&ff->count, 0);
61                 }
62         }
63         return ff;
64 }
65
66 void fuse_file_free(struct fuse_file *ff)
67 {
68         fuse_request_free(ff->reserved_req);
69         kfree(ff);
70 }
71
72 static struct fuse_file *fuse_file_get(struct fuse_file *ff)
73 {
74         atomic_inc(&ff->count);
75         return ff;
76 }
77
78 static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
79 {
80         dput(req->misc.release.dentry);
81         mntput(req->misc.release.vfsmount);
82         fuse_put_request(fc, req);
83 }
84
85 static void fuse_file_put(struct fuse_file *ff)
86 {
87         if (atomic_dec_and_test(&ff->count)) {
88                 struct fuse_req *req = ff->reserved_req;
89                 struct inode *inode = req->misc.release.dentry->d_inode;
90                 struct fuse_conn *fc = get_fuse_conn(inode);
91                 req->end = fuse_release_end;
92                 request_send_background(fc, req);
93                 kfree(ff);
94         }
95 }
96
97 void fuse_finish_open(struct inode *inode, struct file *file,
98                       struct fuse_file *ff, struct fuse_open_out *outarg)
99 {
100         if (outarg->open_flags & FOPEN_DIRECT_IO)
101                 file->f_op = &fuse_direct_io_file_operations;
102         if (!(outarg->open_flags & FOPEN_KEEP_CACHE))
103                 invalidate_inode_pages2(inode->i_mapping);
104         ff->fh = outarg->fh;
105         file->private_data = fuse_file_get(ff);
106 }
107
108 int fuse_open_common(struct inode *inode, struct file *file, int isdir)
109 {
110         struct fuse_open_out outarg;
111         struct fuse_file *ff;
112         int err;
113
114         /* VFS checks this, but only _after_ ->open() */
115         if (file->f_flags & O_DIRECT)
116                 return -EINVAL;
117
118         err = generic_file_open(inode, file);
119         if (err)
120                 return err;
121
122         ff = fuse_file_alloc();
123         if (!ff)
124                 return -ENOMEM;
125
126         err = fuse_send_open(inode, file, isdir, &outarg);
127         if (err)
128                 fuse_file_free(ff);
129         else {
130                 if (isdir)
131                         outarg.open_flags &= ~FOPEN_DIRECT_IO;
132                 fuse_finish_open(inode, file, ff, &outarg);
133         }
134
135         return err;
136 }
137
138 void fuse_release_fill(struct fuse_file *ff, u64 nodeid, int flags, int opcode)
139 {
140         struct fuse_req *req = ff->reserved_req;
141         struct fuse_release_in *inarg = &req->misc.release.in;
142
143         inarg->fh = ff->fh;
144         inarg->flags = flags;
145         req->in.h.opcode = opcode;
146         req->in.h.nodeid = nodeid;
147         req->in.numargs = 1;
148         req->in.args[0].size = sizeof(struct fuse_release_in);
149         req->in.args[0].value = inarg;
150 }
151
152 int fuse_release_common(struct inode *inode, struct file *file, int isdir)
153 {
154         struct fuse_file *ff = file->private_data;
155         if (ff) {
156                 struct fuse_conn *fc = get_fuse_conn(inode);
157                 struct fuse_req *req = ff->reserved_req;
158
159                 fuse_release_fill(ff, get_node_id(inode), file->f_flags,
160                                   isdir ? FUSE_RELEASEDIR : FUSE_RELEASE);
161
162                 /* Hold vfsmount and dentry until release is finished */
163                 req->misc.release.vfsmount = mntget(file->f_path.mnt);
164                 req->misc.release.dentry = dget(file->f_path.dentry);
165
166                 spin_lock(&fc->lock);
167                 list_del(&ff->write_entry);
168                 spin_unlock(&fc->lock);
169                 /*
170                  * Normally this will send the RELEASE request,
171                  * however if some asynchronous READ or WRITE requests
172                  * are outstanding, the sending will be delayed
173                  */
174                 fuse_file_put(ff);
175         }
176
177         /* Return value is ignored by VFS */
178         return 0;
179 }
180
181 static int fuse_open(struct inode *inode, struct file *file)
182 {
183         return fuse_open_common(inode, file, 0);
184 }
185
186 static int fuse_release(struct inode *inode, struct file *file)
187 {
188         return fuse_release_common(inode, file, 0);
189 }
190
191 /*
192  * Scramble the ID space with XTEA, so that the value of the files_struct
193  * pointer is not exposed to userspace.
194  */
195 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
196 {
197         u32 *k = fc->scramble_key;
198         u64 v = (unsigned long) id;
199         u32 v0 = v;
200         u32 v1 = v >> 32;
201         u32 sum = 0;
202         int i;
203
204         for (i = 0; i < 32; i++) {
205                 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
206                 sum += 0x9E3779B9;
207                 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
208         }
209
210         return (u64) v0 + ((u64) v1 << 32);
211 }
212
213 static int fuse_flush(struct file *file, fl_owner_t id)
214 {
215         struct inode *inode = file->f_path.dentry->d_inode;
216         struct fuse_conn *fc = get_fuse_conn(inode);
217         struct fuse_file *ff = file->private_data;
218         struct fuse_req *req;
219         struct fuse_flush_in inarg;
220         int err;
221
222         if (is_bad_inode(inode))
223                 return -EIO;
224
225         if (fc->no_flush)
226                 return 0;
227
228         req = fuse_get_req_nofail(fc, file);
229         memset(&inarg, 0, sizeof(inarg));
230         inarg.fh = ff->fh;
231         inarg.lock_owner = fuse_lock_owner_id(fc, id);
232         req->in.h.opcode = FUSE_FLUSH;
233         req->in.h.nodeid = get_node_id(inode);
234         req->in.numargs = 1;
235         req->in.args[0].size = sizeof(inarg);
236         req->in.args[0].value = &inarg;
237         req->force = 1;
238         request_send(fc, req);
239         err = req->out.h.error;
240         fuse_put_request(fc, req);
241         if (err == -ENOSYS) {
242                 fc->no_flush = 1;
243                 err = 0;
244         }
245         return err;
246 }
247
248 int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
249                       int isdir)
250 {
251         struct inode *inode = de->d_inode;
252         struct fuse_conn *fc = get_fuse_conn(inode);
253         struct fuse_file *ff = file->private_data;
254         struct fuse_req *req;
255         struct fuse_fsync_in inarg;
256         int err;
257
258         if (is_bad_inode(inode))
259                 return -EIO;
260
261         if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
262                 return 0;
263
264         req = fuse_get_req(fc);
265         if (IS_ERR(req))
266                 return PTR_ERR(req);
267
268         memset(&inarg, 0, sizeof(inarg));
269         inarg.fh = ff->fh;
270         inarg.fsync_flags = datasync ? 1 : 0;
271         req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
272         req->in.h.nodeid = get_node_id(inode);
273         req->in.numargs = 1;
274         req->in.args[0].size = sizeof(inarg);
275         req->in.args[0].value = &inarg;
276         request_send(fc, req);
277         err = req->out.h.error;
278         fuse_put_request(fc, req);
279         if (err == -ENOSYS) {
280                 if (isdir)
281                         fc->no_fsyncdir = 1;
282                 else
283                         fc->no_fsync = 1;
284                 err = 0;
285         }
286         return err;
287 }
288
289 static int fuse_fsync(struct file *file, struct dentry *de, int datasync)
290 {
291         return fuse_fsync_common(file, de, datasync, 0);
292 }
293
294 void fuse_read_fill(struct fuse_req *req, struct file *file,
295                     struct inode *inode, loff_t pos, size_t count, int opcode)
296 {
297         struct fuse_read_in *inarg = &req->misc.read_in;
298         struct fuse_file *ff = file->private_data;
299
300         inarg->fh = ff->fh;
301         inarg->offset = pos;
302         inarg->size = count;
303         inarg->flags = file->f_flags;
304         req->in.h.opcode = opcode;
305         req->in.h.nodeid = get_node_id(inode);
306         req->in.numargs = 1;
307         req->in.args[0].size = sizeof(struct fuse_read_in);
308         req->in.args[0].value = inarg;
309         req->out.argpages = 1;
310         req->out.argvar = 1;
311         req->out.numargs = 1;
312         req->out.args[0].size = count;
313 }
314
315 static size_t fuse_send_read(struct fuse_req *req, struct file *file,
316                              struct inode *inode, loff_t pos, size_t count,
317                              fl_owner_t owner)
318 {
319         struct fuse_conn *fc = get_fuse_conn(inode);
320
321         fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
322         if (owner != NULL) {
323                 struct fuse_read_in *inarg = &req->misc.read_in;
324
325                 inarg->read_flags |= FUSE_READ_LOCKOWNER;
326                 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
327         }
328         request_send(fc, req);
329         return req->out.args[0].size;
330 }
331
332 static int fuse_readpage(struct file *file, struct page *page)
333 {
334         struct inode *inode = page->mapping->host;
335         struct fuse_conn *fc = get_fuse_conn(inode);
336         struct fuse_req *req;
337         int err;
338
339         err = -EIO;
340         if (is_bad_inode(inode))
341                 goto out;
342
343         req = fuse_get_req(fc);
344         err = PTR_ERR(req);
345         if (IS_ERR(req))
346                 goto out;
347
348         req->out.page_zeroing = 1;
349         req->num_pages = 1;
350         req->pages[0] = page;
351         fuse_send_read(req, file, inode, page_offset(page), PAGE_CACHE_SIZE,
352                        NULL);
353         err = req->out.h.error;
354         fuse_put_request(fc, req);
355         if (!err)
356                 SetPageUptodate(page);
357         fuse_invalidate_attr(inode); /* atime changed */
358  out:
359         unlock_page(page);
360         return err;
361 }
362
363 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
364 {
365         int i;
366
367         fuse_invalidate_attr(req->pages[0]->mapping->host); /* atime changed */
368
369         for (i = 0; i < req->num_pages; i++) {
370                 struct page *page = req->pages[i];
371                 if (!req->out.h.error)
372                         SetPageUptodate(page);
373                 else
374                         SetPageError(page);
375                 unlock_page(page);
376         }
377         if (req->ff)
378                 fuse_file_put(req->ff);
379         fuse_put_request(fc, req);
380 }
381
382 static void fuse_send_readpages(struct fuse_req *req, struct file *file,
383                                 struct inode *inode)
384 {
385         struct fuse_conn *fc = get_fuse_conn(inode);
386         loff_t pos = page_offset(req->pages[0]);
387         size_t count = req->num_pages << PAGE_CACHE_SHIFT;
388         req->out.page_zeroing = 1;
389         fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
390         if (fc->async_read) {
391                 struct fuse_file *ff = file->private_data;
392                 req->ff = fuse_file_get(ff);
393                 req->end = fuse_readpages_end;
394                 request_send_background(fc, req);
395         } else {
396                 request_send(fc, req);
397                 fuse_readpages_end(fc, req);
398         }
399 }
400
401 struct fuse_fill_data {
402         struct fuse_req *req;
403         struct file *file;
404         struct inode *inode;
405 };
406
407 static int fuse_readpages_fill(void *_data, struct page *page)
408 {
409         struct fuse_fill_data *data = _data;
410         struct fuse_req *req = data->req;
411         struct inode *inode = data->inode;
412         struct fuse_conn *fc = get_fuse_conn(inode);
413
414         if (req->num_pages &&
415             (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
416              (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
417              req->pages[req->num_pages - 1]->index + 1 != page->index)) {
418                 fuse_send_readpages(req, data->file, inode);
419                 data->req = req = fuse_get_req(fc);
420                 if (IS_ERR(req)) {
421                         unlock_page(page);
422                         return PTR_ERR(req);
423                 }
424         }
425         req->pages[req->num_pages] = page;
426         req->num_pages ++;
427         return 0;
428 }
429
430 static int fuse_readpages(struct file *file, struct address_space *mapping,
431                           struct list_head *pages, unsigned nr_pages)
432 {
433         struct inode *inode = mapping->host;
434         struct fuse_conn *fc = get_fuse_conn(inode);
435         struct fuse_fill_data data;
436         int err;
437
438         err = -EIO;
439         if (is_bad_inode(inode))
440                 goto out;
441
442         data.file = file;
443         data.inode = inode;
444         data.req = fuse_get_req(fc);
445         err = PTR_ERR(data.req);
446         if (IS_ERR(data.req))
447                 goto out;
448
449         err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
450         if (!err) {
451                 if (data.req->num_pages)
452                         fuse_send_readpages(data.req, file, inode);
453                 else
454                         fuse_put_request(fc, data.req);
455         }
456 out:
457         return err;
458 }
459
460 static ssize_t fuse_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
461                                   unsigned long nr_segs, loff_t pos)
462 {
463         struct inode *inode = iocb->ki_filp->f_mapping->host;
464
465         if (pos + iov_length(iov, nr_segs) > i_size_read(inode)) {
466                 int err;
467                 /*
468                  * If trying to read past EOF, make sure the i_size
469                  * attribute is up-to-date.
470                  */
471                 err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL);
472                 if (err)
473                         return err;
474         }
475
476         return generic_file_aio_read(iocb, iov, nr_segs, pos);
477 }
478
479 static void fuse_write_fill(struct fuse_req *req, struct file *file,
480                             struct inode *inode, loff_t pos, size_t count,
481                             int writepage)
482 {
483         struct fuse_conn *fc = get_fuse_conn(inode);
484         struct fuse_file *ff = file->private_data;
485         struct fuse_write_in *inarg = &req->misc.write.in;
486         struct fuse_write_out *outarg = &req->misc.write.out;
487
488         memset(inarg, 0, sizeof(struct fuse_write_in));
489         inarg->fh = ff->fh;
490         inarg->offset = pos;
491         inarg->size = count;
492         inarg->write_flags = writepage ? FUSE_WRITE_CACHE : 0;
493         inarg->flags = file->f_flags;
494         req->in.h.opcode = FUSE_WRITE;
495         req->in.h.nodeid = get_node_id(inode);
496         req->in.argpages = 1;
497         req->in.numargs = 2;
498         if (fc->minor < 9)
499                 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
500         else
501                 req->in.args[0].size = sizeof(struct fuse_write_in);
502         req->in.args[0].value = inarg;
503         req->in.args[1].size = count;
504         req->out.numargs = 1;
505         req->out.args[0].size = sizeof(struct fuse_write_out);
506         req->out.args[0].value = outarg;
507 }
508
509 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
510                               struct inode *inode, loff_t pos, size_t count,
511                               fl_owner_t owner)
512 {
513         struct fuse_conn *fc = get_fuse_conn(inode);
514         fuse_write_fill(req, file, inode, pos, count, 0);
515         if (owner != NULL) {
516                 struct fuse_write_in *inarg = &req->misc.write.in;
517                 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
518                 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
519         }
520         request_send(fc, req);
521         return req->misc.write.out.size;
522 }
523
524 static int fuse_write_begin(struct file *file, struct address_space *mapping,
525                         loff_t pos, unsigned len, unsigned flags,
526                         struct page **pagep, void **fsdata)
527 {
528         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
529
530         *pagep = __grab_cache_page(mapping, index);
531         if (!*pagep)
532                 return -ENOMEM;
533         return 0;
534 }
535
536 static int fuse_buffered_write(struct file *file, struct inode *inode,
537                                loff_t pos, unsigned count, struct page *page)
538 {
539         int err;
540         size_t nres;
541         struct fuse_conn *fc = get_fuse_conn(inode);
542         struct fuse_inode *fi = get_fuse_inode(inode);
543         unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
544         struct fuse_req *req;
545
546         if (is_bad_inode(inode))
547                 return -EIO;
548
549         req = fuse_get_req(fc);
550         if (IS_ERR(req))
551                 return PTR_ERR(req);
552
553         req->num_pages = 1;
554         req->pages[0] = page;
555         req->page_offset = offset;
556         nres = fuse_send_write(req, file, inode, pos, count, NULL);
557         err = req->out.h.error;
558         fuse_put_request(fc, req);
559         if (!err && !nres)
560                 err = -EIO;
561         if (!err) {
562                 pos += nres;
563                 spin_lock(&fc->lock);
564                 fi->attr_version = ++fc->attr_version;
565                 if (pos > inode->i_size)
566                         i_size_write(inode, pos);
567                 spin_unlock(&fc->lock);
568
569                 if (count == PAGE_CACHE_SIZE)
570                         SetPageUptodate(page);
571         }
572         fuse_invalidate_attr(inode);
573         return err ? err : nres;
574 }
575
576 static int fuse_write_end(struct file *file, struct address_space *mapping,
577                         loff_t pos, unsigned len, unsigned copied,
578                         struct page *page, void *fsdata)
579 {
580         struct inode *inode = mapping->host;
581         int res = 0;
582
583         if (copied)
584                 res = fuse_buffered_write(file, inode, pos, copied, page);
585
586         unlock_page(page);
587         page_cache_release(page);
588         return res;
589 }
590
591 static void fuse_release_user_pages(struct fuse_req *req, int write)
592 {
593         unsigned i;
594
595         for (i = 0; i < req->num_pages; i++) {
596                 struct page *page = req->pages[i];
597                 if (write)
598                         set_page_dirty_lock(page);
599                 put_page(page);
600         }
601 }
602
603 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
604                                unsigned nbytes, int write)
605 {
606         unsigned long user_addr = (unsigned long) buf;
607         unsigned offset = user_addr & ~PAGE_MASK;
608         int npages;
609
610         /* This doesn't work with nfsd */
611         if (!current->mm)
612                 return -EPERM;
613
614         nbytes = min(nbytes, (unsigned) FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
615         npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
616         npages = min(max(npages, 1), FUSE_MAX_PAGES_PER_REQ);
617         down_read(&current->mm->mmap_sem);
618         npages = get_user_pages(current, current->mm, user_addr, npages, write,
619                                 0, req->pages, NULL);
620         up_read(&current->mm->mmap_sem);
621         if (npages < 0)
622                 return npages;
623
624         req->num_pages = npages;
625         req->page_offset = offset;
626         return 0;
627 }
628
629 static ssize_t fuse_direct_io(struct file *file, const char __user *buf,
630                               size_t count, loff_t *ppos, int write)
631 {
632         struct inode *inode = file->f_path.dentry->d_inode;
633         struct fuse_conn *fc = get_fuse_conn(inode);
634         size_t nmax = write ? fc->max_write : fc->max_read;
635         loff_t pos = *ppos;
636         ssize_t res = 0;
637         struct fuse_req *req;
638
639         if (is_bad_inode(inode))
640                 return -EIO;
641
642         req = fuse_get_req(fc);
643         if (IS_ERR(req))
644                 return PTR_ERR(req);
645
646         while (count) {
647                 size_t nres;
648                 size_t nbytes = min(count, nmax);
649                 int err = fuse_get_user_pages(req, buf, nbytes, !write);
650                 if (err) {
651                         res = err;
652                         break;
653                 }
654                 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset;
655                 nbytes = min(count, nbytes);
656                 if (write)
657                         nres = fuse_send_write(req, file, inode, pos, nbytes,
658                                                current->files);
659                 else
660                         nres = fuse_send_read(req, file, inode, pos, nbytes,
661                                               current->files);
662                 fuse_release_user_pages(req, !write);
663                 if (req->out.h.error) {
664                         if (!res)
665                                 res = req->out.h.error;
666                         break;
667                 } else if (nres > nbytes) {
668                         res = -EIO;
669                         break;
670                 }
671                 count -= nres;
672                 res += nres;
673                 pos += nres;
674                 buf += nres;
675                 if (nres != nbytes)
676                         break;
677                 if (count) {
678                         fuse_put_request(fc, req);
679                         req = fuse_get_req(fc);
680                         if (IS_ERR(req))
681                                 break;
682                 }
683         }
684         fuse_put_request(fc, req);
685         if (res > 0) {
686                 if (write) {
687                         spin_lock(&fc->lock);
688                         if (pos > inode->i_size)
689                                 i_size_write(inode, pos);
690                         spin_unlock(&fc->lock);
691                 }
692                 *ppos = pos;
693         }
694         fuse_invalidate_attr(inode);
695
696         return res;
697 }
698
699 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
700                                      size_t count, loff_t *ppos)
701 {
702         return fuse_direct_io(file, buf, count, ppos, 0);
703 }
704
705 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
706                                  size_t count, loff_t *ppos)
707 {
708         struct inode *inode = file->f_path.dentry->d_inode;
709         ssize_t res;
710         /* Don't allow parallel writes to the same file */
711         mutex_lock(&inode->i_mutex);
712         res = generic_write_checks(file, ppos, &count, 0);
713         if (!res)
714                 res = fuse_direct_io(file, buf, count, ppos, 1);
715         mutex_unlock(&inode->i_mutex);
716         return res;
717 }
718
719 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
720 {
721         if ((vma->vm_flags & VM_SHARED)) {
722                 if ((vma->vm_flags & VM_WRITE))
723                         return -ENODEV;
724                 else
725                         vma->vm_flags &= ~VM_MAYWRITE;
726         }
727         return generic_file_mmap(file, vma);
728 }
729
730 static int fuse_set_page_dirty(struct page *page)
731 {
732         printk("fuse_set_page_dirty: should not happen\n");
733         dump_stack();
734         return 0;
735 }
736
737 static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
738                                   struct file_lock *fl)
739 {
740         switch (ffl->type) {
741         case F_UNLCK:
742                 break;
743
744         case F_RDLCK:
745         case F_WRLCK:
746                 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
747                     ffl->end < ffl->start)
748                         return -EIO;
749
750                 fl->fl_start = ffl->start;
751                 fl->fl_end = ffl->end;
752                 fl->fl_pid = ffl->pid;
753                 break;
754
755         default:
756                 return -EIO;
757         }
758         fl->fl_type = ffl->type;
759         return 0;
760 }
761
762 static void fuse_lk_fill(struct fuse_req *req, struct file *file,
763                          const struct file_lock *fl, int opcode, pid_t pid,
764                          int flock)
765 {
766         struct inode *inode = file->f_path.dentry->d_inode;
767         struct fuse_conn *fc = get_fuse_conn(inode);
768         struct fuse_file *ff = file->private_data;
769         struct fuse_lk_in *arg = &req->misc.lk_in;
770
771         arg->fh = ff->fh;
772         arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
773         arg->lk.start = fl->fl_start;
774         arg->lk.end = fl->fl_end;
775         arg->lk.type = fl->fl_type;
776         arg->lk.pid = pid;
777         if (flock)
778                 arg->lk_flags |= FUSE_LK_FLOCK;
779         req->in.h.opcode = opcode;
780         req->in.h.nodeid = get_node_id(inode);
781         req->in.numargs = 1;
782         req->in.args[0].size = sizeof(*arg);
783         req->in.args[0].value = arg;
784 }
785
786 static int fuse_getlk(struct file *file, struct file_lock *fl)
787 {
788         struct inode *inode = file->f_path.dentry->d_inode;
789         struct fuse_conn *fc = get_fuse_conn(inode);
790         struct fuse_req *req;
791         struct fuse_lk_out outarg;
792         int err;
793
794         req = fuse_get_req(fc);
795         if (IS_ERR(req))
796                 return PTR_ERR(req);
797
798         fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0);
799         req->out.numargs = 1;
800         req->out.args[0].size = sizeof(outarg);
801         req->out.args[0].value = &outarg;
802         request_send(fc, req);
803         err = req->out.h.error;
804         fuse_put_request(fc, req);
805         if (!err)
806                 err = convert_fuse_file_lock(&outarg.lk, fl);
807
808         return err;
809 }
810
811 static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
812 {
813         struct inode *inode = file->f_path.dentry->d_inode;
814         struct fuse_conn *fc = get_fuse_conn(inode);
815         struct fuse_req *req;
816         int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
817         pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
818         int err;
819
820         /* Unlock on close is handled by the flush method */
821         if (fl->fl_flags & FL_CLOSE)
822                 return 0;
823
824         req = fuse_get_req(fc);
825         if (IS_ERR(req))
826                 return PTR_ERR(req);
827
828         fuse_lk_fill(req, file, fl, opcode, pid, flock);
829         request_send(fc, req);
830         err = req->out.h.error;
831         /* locking is restartable */
832         if (err == -EINTR)
833                 err = -ERESTARTSYS;
834         fuse_put_request(fc, req);
835         return err;
836 }
837
838 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
839 {
840         struct inode *inode = file->f_path.dentry->d_inode;
841         struct fuse_conn *fc = get_fuse_conn(inode);
842         int err;
843
844         if (cmd == F_GETLK) {
845                 if (fc->no_lock) {
846                         posix_test_lock(file, fl);
847                         err = 0;
848                 } else
849                         err = fuse_getlk(file, fl);
850         } else {
851                 if (fc->no_lock)
852                         err = posix_lock_file_wait(file, fl);
853                 else
854                         err = fuse_setlk(file, fl, 0);
855         }
856         return err;
857 }
858
859 static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
860 {
861         struct inode *inode = file->f_path.dentry->d_inode;
862         struct fuse_conn *fc = get_fuse_conn(inode);
863         int err;
864
865         if (fc->no_lock) {
866                 err = flock_lock_file_wait(file, fl);
867         } else {
868                 /* emulate flock with POSIX locks */
869                 fl->fl_owner = (fl_owner_t) file;
870                 err = fuse_setlk(file, fl, 1);
871         }
872
873         return err;
874 }
875
876 static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
877 {
878         struct inode *inode = mapping->host;
879         struct fuse_conn *fc = get_fuse_conn(inode);
880         struct fuse_req *req;
881         struct fuse_bmap_in inarg;
882         struct fuse_bmap_out outarg;
883         int err;
884
885         if (!inode->i_sb->s_bdev || fc->no_bmap)
886                 return 0;
887
888         req = fuse_get_req(fc);
889         if (IS_ERR(req))
890                 return 0;
891
892         memset(&inarg, 0, sizeof(inarg));
893         inarg.block = block;
894         inarg.blocksize = inode->i_sb->s_blocksize;
895         req->in.h.opcode = FUSE_BMAP;
896         req->in.h.nodeid = get_node_id(inode);
897         req->in.numargs = 1;
898         req->in.args[0].size = sizeof(inarg);
899         req->in.args[0].value = &inarg;
900         req->out.numargs = 1;
901         req->out.args[0].size = sizeof(outarg);
902         req->out.args[0].value = &outarg;
903         request_send(fc, req);
904         err = req->out.h.error;
905         fuse_put_request(fc, req);
906         if (err == -ENOSYS)
907                 fc->no_bmap = 1;
908
909         return err ? 0 : outarg.block;
910 }
911
912 static const struct file_operations fuse_file_operations = {
913         .llseek         = generic_file_llseek,
914         .read           = do_sync_read,
915         .aio_read       = fuse_file_aio_read,
916         .write          = do_sync_write,
917         .aio_write      = generic_file_aio_write,
918         .mmap           = fuse_file_mmap,
919         .open           = fuse_open,
920         .flush          = fuse_flush,
921         .release        = fuse_release,
922         .fsync          = fuse_fsync,
923         .lock           = fuse_file_lock,
924         .flock          = fuse_file_flock,
925         .splice_read    = generic_file_splice_read,
926 };
927
928 static const struct file_operations fuse_direct_io_file_operations = {
929         .llseek         = generic_file_llseek,
930         .read           = fuse_direct_read,
931         .write          = fuse_direct_write,
932         .open           = fuse_open,
933         .flush          = fuse_flush,
934         .release        = fuse_release,
935         .fsync          = fuse_fsync,
936         .lock           = fuse_file_lock,
937         .flock          = fuse_file_flock,
938         /* no mmap and splice_read */
939 };
940
941 static const struct address_space_operations fuse_file_aops  = {
942         .readpage       = fuse_readpage,
943         .write_begin    = fuse_write_begin,
944         .write_end      = fuse_write_end,
945         .readpages      = fuse_readpages,
946         .set_page_dirty = fuse_set_page_dirty,
947         .bmap           = fuse_bmap,
948 };
949
950 void fuse_init_file_inode(struct inode *inode)
951 {
952         inode->i_fop = &fuse_file_operations;
953         inode->i_data.a_ops = &fuse_file_aops;
954 }