projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
fuse: multiplex cached/direct_io file operations
[linux-2.6-block.git] / fs / fuse / file.c
CommitLineData
b6aeaded
MS		 1/*
2 FUSE: Filesystem in Userspace
1729a16c 3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu>
b6aeaded
MS		 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>
e8edc6e0 14#include <linux/sched.h>
7a36094d 15#include <linux/sched/signal.h>
08cbf542 16#include <linux/module.h>
d9d318d3 17#include <linux/compat.h>
478e0841 18#include <linux/swap.h>
3634a632 19#include <linux/falloc.h>
e2e40f2c 20#include <linux/uio.h>
b6aeaded 21
91fe96b4
MS		 22static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
23 int opcode, struct fuse_open_out *outargp)
b6aeaded 24{
b6aeaded 25 struct fuse_open_in inarg;
7078187a 26 FUSE_ARGS(args);
fd72faac
MS		 27
28 memset(&inarg, 0, sizeof(inarg));
6ff958ed
MS		 29 inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY);
30 if (!fc->atomic_o_trunc)
31 inarg.flags &= ~O_TRUNC;
7078187a
MS		 32 args.in.h.opcode = opcode;
33 args.in.h.nodeid = nodeid;
34 args.in.numargs = 1;
35 args.in.args[0].size = sizeof(inarg);
36 args.in.args[0].value = &inarg;
37 args.out.numargs = 1;
38 args.out.args[0].size = sizeof(*outargp);
39 args.out.args[0].value = outargp;
fd72faac 40
7078187a 41 return fuse_simple_request(fc, &args);
fd72faac
MS		 42}
43
acf99433 44struct fuse_file *fuse_file_alloc(struct fuse_conn *fc)
fd72faac
MS		 45{
46 struct fuse_file *ff;
6b2db28a 47
68227c03 48 ff = kzalloc(sizeof(struct fuse_file), GFP_KERNEL);
6b2db28a
TH		 49 if (unlikely(!ff))
50 return NULL;
51
da5e4714 52 ff->fc = fc;
4250c066 53 ff->reserved_req = fuse_request_alloc(0);
6b2db28a
TH		 54 if (unlikely(!ff->reserved_req)) {
55 kfree(ff);
56 return NULL;
fd72faac 57 }
6b2db28a
TH		 58
59 INIT_LIST_HEAD(&ff->write_entry);
5d7bc7e8 60 mutex_init(&ff->readdir.lock);
4e8c2eb5 61 refcount_set(&ff->count, 1);
6b2db28a
TH		 62 RB_CLEAR_NODE(&ff->polled_node);
63 init_waitqueue_head(&ff->poll_wait);
64
75126f55 65 ff->kh = atomic64_inc_return(&fc->khctr);
6b2db28a 66
fd72faac
MS		 67 return ff;
68}
69
70void fuse_file_free(struct fuse_file *ff)
71{
33649c91 72 fuse_request_free(ff->reserved_req);
5d7bc7e8 73 mutex_destroy(&ff->readdir.lock);
fd72faac
MS		 74 kfree(ff);
75}
76
267d8444 77static struct fuse_file *fuse_file_get(struct fuse_file *ff)
c756e0a4 78{
4e8c2eb5 79 refcount_inc(&ff->count);
c756e0a4
MS		 80 return ff;
81}
82
819c4b3b
MS		 83static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
84{
baebccbe 85 iput(req->misc.release.inode);
819c4b3b
MS		 86}
87
2e64ff15 88static void fuse_file_put(struct fuse_file *ff, bool sync, bool isdir)
c756e0a4 89{
4e8c2eb5 90 if (refcount_dec_and_test(&ff->count)) {
c756e0a4 91 struct fuse_req *req = ff->reserved_req;
8b0797a4 92
2e64ff15 93 if (ff->fc->no_open && !isdir) {
7678ac50
AG		 94 /*
95 * Drop the release request when client does not
96 * implement 'open'
97 */
825d6d33 98 __clear_bit(FR_BACKGROUND, &req->flags);
baebccbe 99 iput(req->misc.release.inode);
7678ac50
AG		 100 fuse_put_request(ff->fc, req);
101 } else if (sync) {
2e38bea9 102 __set_bit(FR_FORCE, &req->flags);
825d6d33 103 __clear_bit(FR_BACKGROUND, &req->flags);
5a18ec17 104 fuse_request_send(ff->fc, req);
baebccbe 105 iput(req->misc.release.inode);
5a18ec17
MS		 106 fuse_put_request(ff->fc, req);
107 } else {
108 req->end = fuse_release_end;
825d6d33 109 __set_bit(FR_BACKGROUND, &req->flags);
5a18ec17
MS		 110 fuse_request_send_background(ff->fc, req);
111 }
c756e0a4
MS		 112 kfree(ff);
113 }
114}
115
08cbf542
TH		 116int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
117 bool isdir)
91fe96b4 118{
91fe96b4 119 struct fuse_file *ff;
91fe96b4
MS		 120 int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
121
122 ff = fuse_file_alloc(fc);
123 if (!ff)
124 return -ENOMEM;
125
7678ac50
AG		 126 ff->fh = 0;
127 ff->open_flags = FOPEN_KEEP_CACHE; /* Default for no-open */
128 if (!fc->no_open || isdir) {
129 struct fuse_open_out outarg;
130 int err;
131
132 err = fuse_send_open(fc, nodeid, file, opcode, &outarg);
133 if (!err) {
134 ff->fh = outarg.fh;
135 ff->open_flags = outarg.open_flags;
136
137 } else if (err != -ENOSYS || isdir) {
138 fuse_file_free(ff);
139 return err;
140 } else {
141 fc->no_open = 1;
142 }
91fe96b4
MS		 143 }
144
145 if (isdir)
7678ac50 146 ff->open_flags &= ~FOPEN_DIRECT_IO;
91fe96b4 147
91fe96b4 148 ff->nodeid = nodeid;
267d8444 149 file->private_data = ff;
91fe96b4
MS		 150
151 return 0;
152}
08cbf542 153EXPORT_SYMBOL_GPL(fuse_do_open);
91fe96b4 154
650b22b9
PE		 155static void fuse_link_write_file(struct file *file)
156{
157 struct inode *inode = file_inode(file);
650b22b9
PE		 158 struct fuse_inode *fi = get_fuse_inode(inode);
159 struct fuse_file *ff = file->private_data;
160 /*
161 * file may be written through mmap, so chain it onto the
162 * inodes's write_file list
163 */
f15ecfef 164 spin_lock(&fi->lock);
650b22b9
PE		 165 if (list_empty(&ff->write_entry))
166 list_add(&ff->write_entry, &fi->write_files);
f15ecfef 167 spin_unlock(&fi->lock);
650b22b9
PE		 168}
169
c7b7143c 170void fuse_finish_open(struct inode *inode, struct file *file)
fd72faac 171{
c7b7143c 172 struct fuse_file *ff = file->private_data;
a0822c55 173 struct fuse_conn *fc = get_fuse_conn(inode);
c7b7143c 174
c7b7143c 175 if (!(ff->open_flags & FOPEN_KEEP_CACHE))
b1009979 176 invalidate_inode_pages2(inode->i_mapping);
c7b7143c 177 if (ff->open_flags & FOPEN_NONSEEKABLE)
a7c1b990 178 nonseekable_open(inode, file);
a0822c55
KS		 179 if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) {
180 struct fuse_inode *fi = get_fuse_inode(inode);
181
f15ecfef 182 spin_lock(&fi->lock);
4510d86f 183 fi->attr_version = atomic64_inc_return(&fc->attr_version);
a0822c55 184 i_size_write(inode, 0);
f15ecfef 185 spin_unlock(&fi->lock);
a0822c55 186 fuse_invalidate_attr(inode);
75caeecd
MP		 187 if (fc->writeback_cache)
188 file_update_time(file);
a0822c55 189 }
4d99ff8f
PE		 190 if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache)
191 fuse_link_write_file(file);
fd72faac
MS		 192}
193
91fe96b4 194int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
fd72faac 195{
acf99433 196 struct fuse_conn *fc = get_fuse_conn(inode);
b6aeaded 197 int err;
75caeecd
MP		 198 bool lock_inode = (file->f_flags & O_TRUNC) &&
199 fc->atomic_o_trunc &&
200 fc->writeback_cache;
b6aeaded
MS		 201
202 err = generic_file_open(inode, file);
203 if (err)
204 return err;
205
75caeecd 206 if (lock_inode)
5955102c 207 inode_lock(inode);
75caeecd 208
91fe96b4 209 err = fuse_do_open(fc, get_node_id(inode), file, isdir);
b6aeaded 210
75caeecd
MP		 211 if (!err)
212 fuse_finish_open(inode, file);
91fe96b4 213
75caeecd 214 if (lock_inode)
5955102c 215 inode_unlock(inode);
75caeecd
MP		 216
217 return err;
b6aeaded
MS		 218}
219
ebf84d0c
KT		 220static void fuse_prepare_release(struct fuse_inode *fi, struct fuse_file *ff,
221 int flags, int opcode)
64c6d8ed 222{
8b0797a4 223 struct fuse_conn *fc = ff->fc;
33649c91 224 struct fuse_req *req = ff->reserved_req;
b57d4264 225 struct fuse_release_in *inarg = &req->misc.release.in;
b6aeaded 226
f15ecfef
KT		 227 /* Inode is NULL on error path of fuse_create_open() */
228 if (likely(fi)) {
229 spin_lock(&fi->lock);
230 list_del(&ff->write_entry);
231 spin_unlock(&fi->lock);
232 }
8b0797a4 233 spin_lock(&fc->lock);
8b0797a4
MS		 234 if (!RB_EMPTY_NODE(&ff->polled_node))
235 rb_erase(&ff->polled_node, &fc->polled_files);
236 spin_unlock(&fc->lock);
237
357ccf2b 238 wake_up_interruptible_all(&ff->poll_wait);
8b0797a4 239
b6aeaded 240 inarg->fh = ff->fh;
fd72faac 241 inarg->flags = flags;
51eb01e7 242 req->in.h.opcode = opcode;
c7b7143c 243 req->in.h.nodeid = ff->nodeid;
b6aeaded
MS		 244 req->in.numargs = 1;
245 req->in.args[0].size = sizeof(struct fuse_release_in);
246 req->in.args[0].value = inarg;
fd72faac
MS		 247}
248
2e64ff15 249void fuse_release_common(struct file *file, bool isdir)
fd72faac 250{
ebf84d0c 251 struct fuse_inode *fi = get_fuse_inode(file_inode(file));
9a87ad3d
MS		 252 struct fuse_file *ff = file->private_data;
253 struct fuse_req *req = ff->reserved_req;
2e64ff15 254 int opcode = isdir ? FUSE_RELEASEDIR : FUSE_RELEASE;
6b2db28a 255
ebf84d0c 256 fuse_prepare_release(fi, ff, file->f_flags, opcode);
6b2db28a 257
37fb3a30
MS		 258 if (ff->flock) {
259 struct fuse_release_in *inarg = &req->misc.release.in;
260 inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK;
261 inarg->lock_owner = fuse_lock_owner_id(ff->fc,
262 (fl_owner_t) file);
263 }
baebccbe
MS		 264 /* Hold inode until release is finished */
265 req->misc.release.inode = igrab(file_inode(file));
6b2db28a 266
6b2db28a
TH		 267 /*
268 * Normally this will send the RELEASE request, however if
269 * some asynchronous READ or WRITE requests are outstanding,
270 * the sending will be delayed.
5a18ec17
MS		 271 *
272 * Make the release synchronous if this is a fuseblk mount,
273 * synchronous RELEASE is allowed (and desirable) in this case
274 * because the server can be trusted not to screw up.
6b2db28a 275 */
2e64ff15 276 fuse_file_put(ff, ff->fc->destroy_req != NULL, isdir);
b6aeaded
MS		 277}
278
04730fef
MS		 279static int fuse_open(struct inode *inode, struct file *file)
280{
91fe96b4 281 return fuse_open_common(inode, file, false);
04730fef
MS		 282}
283
284static int fuse_release(struct inode *inode, struct file *file)
285{
e7cc133c
PE		 286 struct fuse_conn *fc = get_fuse_conn(inode);
287
288 /* see fuse_vma_close() for !writeback_cache case */
289 if (fc->writeback_cache)
1e18bda8 290 write_inode_now(inode, 1);
b0aa7606 291
2e64ff15 292 fuse_release_common(file, false);
8b0797a4
MS		 293
294 /* return value is ignored by VFS */
295 return 0;
296}
297
ebf84d0c 298void fuse_sync_release(struct fuse_inode *fi, struct fuse_file *ff, int flags)
8b0797a4 299{
4e8c2eb5 300 WARN_ON(refcount_read(&ff->count) > 1);
ebf84d0c 301 fuse_prepare_release(fi, ff, flags, FUSE_RELEASE);
267d8444
MS		 302 /*
303 * iput(NULL) is a no-op and since the refcount is 1 and everything's
304 * synchronous, we are fine with not doing igrab() here"
305 */
2e64ff15 306 fuse_file_put(ff, true, false);
04730fef 307}
08cbf542 308EXPORT_SYMBOL_GPL(fuse_sync_release);
04730fef 309
71421259 310/*
9c8ef561
MS		 311 * Scramble the ID space with XTEA, so that the value of the files_struct
312 * pointer is not exposed to userspace.
71421259 313 */
f3332114 314u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
71421259 315{
9c8ef561
MS		 316 u32 *k = fc->scramble_key;
317 u64 v = (unsigned long) id;
318 u32 v0 = v;
319 u32 v1 = v >> 32;
320 u32 sum = 0;
321 int i;
322
323 for (i = 0; i < 32; i++) {
324 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
325 sum += 0x9E3779B9;
326 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
327 }
328
329 return (u64) v0 + ((u64) v1 << 32);
71421259
MS		 330}
331
2fe93bd4
MS		 332static struct fuse_req *fuse_find_writeback(struct fuse_inode *fi,
333 pgoff_t idx_from, pgoff_t idx_to)
334{
335 struct fuse_req *req;
336
337 list_for_each_entry(req, &fi->writepages, writepages_entry) {
338 pgoff_t curr_index;
339
340 WARN_ON(get_fuse_inode(req->inode) != fi);
341 curr_index = req->misc.write.in.offset >> PAGE_SHIFT;
342 if (idx_from < curr_index + req->num_pages &&
343 curr_index <= idx_to) {
344 return req;
345 }
346 }
347 return NULL;
348}
349
3be5a52b 350/*
ea8cd333 351 * Check if any page in a range is under writeback
3be5a52b
MS		 352 *
353 * This is currently done by walking the list of writepage requests
354 * for the inode, which can be pretty inefficient.
355 */
ea8cd333
PE		 356static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from,
357 pgoff_t idx_to)
3be5a52b 358{
3be5a52b 359 struct fuse_inode *fi = get_fuse_inode(inode);
2fe93bd4 360 bool found;
3be5a52b 361
f15ecfef 362 spin_lock(&fi->lock);
2fe93bd4 363 found = fuse_find_writeback(fi, idx_from, idx_to);
f15ecfef 364 spin_unlock(&fi->lock);
3be5a52b
MS		 365
366 return found;
367}
368
ea8cd333
PE		 369static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
370{
371 return fuse_range_is_writeback(inode, index, index);
372}
373
3be5a52b
MS		 374/*
375 * Wait for page writeback to be completed.
376 *
377 * Since fuse doesn't rely on the VM writeback tracking, this has to
378 * use some other means.
379 */
380static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
381{
382 struct fuse_inode *fi = get_fuse_inode(inode);
383
384 wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
385 return 0;
386}
387
fe38d7df
MP		 388/*
389 * Wait for all pending writepages on the inode to finish.
390 *
391 * This is currently done by blocking further writes with FUSE_NOWRITE
392 * and waiting for all sent writes to complete.
393 *
394 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
395 * could conflict with truncation.
396 */
397static void fuse_sync_writes(struct inode *inode)
398{
399 fuse_set_nowrite(inode);
400 fuse_release_nowrite(inode);
401}
402
75e1fcc0 403static int fuse_flush(struct file *file, fl_owner_t id)
b6aeaded 404{
6131ffaa 405 struct inode *inode = file_inode(file);
b6aeaded
MS		 406 struct fuse_conn *fc = get_fuse_conn(inode);
407 struct fuse_file *ff = file->private_data;
408 struct fuse_req *req;
409 struct fuse_flush_in inarg;
410 int err;
411
248d86e8
MS		 412 if (is_bad_inode(inode))
413 return -EIO;
414
b6aeaded
MS		 415 if (fc->no_flush)
416 return 0;
417
1e18bda8 418 err = write_inode_now(inode, 1);
fe38d7df
MP		 419 if (err)
420 return err;
421
5955102c 422 inode_lock(inode);
fe38d7df 423 fuse_sync_writes(inode);
5955102c 424 inode_unlock(inode);
fe38d7df 425
4a7f4e88 426 err = filemap_check_errors(file->f_mapping);
9ebce595
MP		 427 if (err)
428 return err;
429
b111c8c0 430 req = fuse_get_req_nofail_nopages(fc, file);
b6aeaded
MS		 431 memset(&inarg, 0, sizeof(inarg));
432 inarg.fh = ff->fh;
9c8ef561 433 inarg.lock_owner = fuse_lock_owner_id(fc, id);
b6aeaded
MS		 434 req->in.h.opcode = FUSE_FLUSH;
435 req->in.h.nodeid = get_node_id(inode);
b6aeaded
MS		 436 req->in.numargs = 1;
437 req->in.args[0].size = sizeof(inarg);
438 req->in.args[0].value = &inarg;
825d6d33 439 __set_bit(FR_FORCE, &req->flags);
b93f858a 440 fuse_request_send(fc, req);
b6aeaded
MS		 441 err = req->out.h.error;
442 fuse_put_request(fc, req);
443 if (err == -ENOSYS) {
444 fc->no_flush = 1;
445 err = 0;
446 }
447 return err;
448}
449
02c24a82 450int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
a9c2d1e8 451 int datasync, int opcode)
b6aeaded 452{
7ea80859 453 struct inode *inode = file->f_mapping->host;
b6aeaded
MS		 454 struct fuse_conn *fc = get_fuse_conn(inode);
455 struct fuse_file *ff = file->private_data;
7078187a 456 FUSE_ARGS(args);
b6aeaded 457 struct fuse_fsync_in inarg;
a9c2d1e8
MS		 458
459 memset(&inarg, 0, sizeof(inarg));
460 inarg.fh = ff->fh;
461 inarg.fsync_flags = datasync ? 1 : 0;
462 args.in.h.opcode = opcode;
463 args.in.h.nodeid = get_node_id(inode);
464 args.in.numargs = 1;
465 args.in.args[0].size = sizeof(inarg);
466 args.in.args[0].value = &inarg;
467 return fuse_simple_request(fc, &args);
468}
469
470static int fuse_fsync(struct file *file, loff_t start, loff_t end,
471 int datasync)
472{
473 struct inode *inode = file->f_mapping->host;
474 struct fuse_conn *fc = get_fuse_conn(inode);
b6aeaded
MS		 475 int err;
476
248d86e8
MS		 477 if (is_bad_inode(inode))
478 return -EIO;
479
5955102c 480 inode_lock(inode);
02c24a82 481
3be5a52b
MS		 482 /*
483 * Start writeback against all dirty pages of the inode, then
484 * wait for all outstanding writes, before sending the FSYNC
485 * request.
486 */
7e51fe1d 487 err = file_write_and_wait_range(file, start, end);
3be5a52b 488 if (err)
02c24a82 489 goto out;
3be5a52b
MS		 490
491 fuse_sync_writes(inode);
ac7f052b
AK		 492
493 /*
494 * Due to implementation of fuse writeback
7e51fe1d 495 * file_write_and_wait_range() does not catch errors.
ac7f052b
AK		 496 * We have to do this directly after fuse_sync_writes()
497 */
7e51fe1d 498 err = file_check_and_advance_wb_err(file);
ac7f052b
AK		 499 if (err)
500 goto out;
501
1e18bda8
MS		 502 err = sync_inode_metadata(inode, 1);
503 if (err)
504 goto out;
3be5a52b 505
a9c2d1e8 506 if (fc->no_fsync)
22401e7b 507 goto out;
b0aa7606 508
a9c2d1e8 509 err = fuse_fsync_common(file, start, end, datasync, FUSE_FSYNC);
b6aeaded 510 if (err == -ENOSYS) {
a9c2d1e8 511 fc->no_fsync = 1;
b6aeaded
MS		 512 err = 0;
513 }
02c24a82 514out:
5955102c 515 inode_unlock(inode);
b6aeaded 516
a9c2d1e8 517 return err;
82547981
MS		 518}
519
2106cb18
MS		 520void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos,
521 size_t count, int opcode)
b6aeaded 522{
5c5c5e51 523 struct fuse_read_in *inarg = &req->misc.read.in;
a6643094 524 struct fuse_file *ff = file->private_data;
b6aeaded 525
361b1eb5
MS		 526 inarg->fh = ff->fh;
527 inarg->offset = pos;
528 inarg->size = count;
a6643094 529 inarg->flags = file->f_flags;
361b1eb5 530 req->in.h.opcode = opcode;
2106cb18 531 req->in.h.nodeid = ff->nodeid;
b6aeaded
MS		 532 req->in.numargs = 1;
533 req->in.args[0].size = sizeof(struct fuse_read_in);
c1aa96a5 534 req->in.args[0].value = inarg;
b6aeaded
MS		 535 req->out.argvar = 1;
536 req->out.numargs = 1;
537 req->out.args[0].size = count;
b6aeaded
MS		 538}
539
8fba54ae 540static void fuse_release_user_pages(struct fuse_req *req, bool should_dirty)
187c5c36
MP		 541{
542 unsigned i;
543
544 for (i = 0; i < req->num_pages; i++) {
545 struct page *page = req->pages[i];
8fba54ae 546 if (should_dirty)
187c5c36
MP		 547 set_page_dirty_lock(page);
548 put_page(page);
549 }
550}
551
744742d6
SF		 552static void fuse_io_release(struct kref *kref)
553{
554 kfree(container_of(kref, struct fuse_io_priv, refcnt));
555}
556
9d5722b7
CH		 557static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io)
558{
559 if (io->err)
560 return io->err;
561
562 if (io->bytes >= 0 && io->write)
563 return -EIO;
564
565 return io->bytes < 0 ? io->size : io->bytes;
566}
567
01e9d11a
MP		 568/**
569 * In case of short read, the caller sets 'pos' to the position of
570 * actual end of fuse request in IO request. Otherwise, if bytes_requested
571 * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1.
572 *
573 * An example:
574 * User requested DIO read of 64K. It was splitted into two 32K fuse requests,
575 * both submitted asynchronously. The first of them was ACKed by userspace as
576 * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The
577 * second request was ACKed as short, e.g. only 1K was read, resulting in
578 * pos == 33K.
579 *
580 * Thus, when all fuse requests are completed, the minimal non-negative 'pos'
581 * will be equal to the length of the longest contiguous fragment of
582 * transferred data starting from the beginning of IO request.
583 */
584static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
585{
586 int left;
587
588 spin_lock(&io->lock);
589 if (err)
590 io->err = io->err ? : err;
591 else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes))
592 io->bytes = pos;
593
594 left = --io->reqs;
7879c4e5 595 if (!left && io->blocking)
9d5722b7 596 complete(io->done);
01e9d11a
MP		 597 spin_unlock(&io->lock);
598
7879c4e5 599 if (!left && !io->blocking) {
9d5722b7 600 ssize_t res = fuse_get_res_by_io(io);
01e9d11a 601
9d5722b7
CH		 602 if (res >= 0) {
603 struct inode *inode = file_inode(io->iocb->ki_filp);
604 struct fuse_conn *fc = get_fuse_conn(inode);
605 struct fuse_inode *fi = get_fuse_inode(inode);
01e9d11a 606
f15ecfef 607 spin_lock(&fi->lock);
4510d86f 608 fi->attr_version = atomic64_inc_return(&fc->attr_version);
f15ecfef 609 spin_unlock(&fi->lock);
01e9d11a
MP		 610 }
611
04b2fa9f 612 io->iocb->ki_complete(io->iocb, res, 0);
01e9d11a 613 }
744742d6
SF		 614
615 kref_put(&io->refcnt, fuse_io_release);
01e9d11a
MP		 616}
617
618static void fuse_aio_complete_req(struct fuse_conn *fc, struct fuse_req *req)
619{
620 struct fuse_io_priv *io = req->io;
621 ssize_t pos = -1;
622
61c12b49 623 fuse_release_user_pages(req, io->should_dirty);
01e9d11a
MP		 624
625 if (io->write) {
626 if (req->misc.write.in.size != req->misc.write.out.size)
627 pos = req->misc.write.in.offset - io->offset +
628 req->misc.write.out.size;
629 } else {
630 if (req->misc.read.in.size != req->out.args[0].size)
631 pos = req->misc.read.in.offset - io->offset +
632 req->out.args[0].size;
633 }
634
635 fuse_aio_complete(io, req->out.h.error, pos);
636}
637
638static size_t fuse_async_req_send(struct fuse_conn *fc, struct fuse_req *req,
639 size_t num_bytes, struct fuse_io_priv *io)
640{
641 spin_lock(&io->lock);
744742d6 642 kref_get(&io->refcnt);
01e9d11a
MP		 643 io->size += num_bytes;
644 io->reqs++;
645 spin_unlock(&io->lock);
646
647 req->io = io;
648 req->end = fuse_aio_complete_req;
649
36cf66ed 650 __fuse_get_request(req);
01e9d11a
MP		 651 fuse_request_send_background(fc, req);
652
653 return num_bytes;
654}
655
36cf66ed 656static size_t fuse_send_read(struct fuse_req *req, struct fuse_io_priv *io,
2106cb18 657 loff_t pos, size_t count, fl_owner_t owner)
04730fef 658{
e1c0eecb 659 struct file *file = io->iocb->ki_filp;
2106cb18
MS		 660 struct fuse_file *ff = file->private_data;
661 struct fuse_conn *fc = ff->fc;
f3332114 662
2106cb18 663 fuse_read_fill(req, file, pos, count, FUSE_READ);
f3332114 664 if (owner != NULL) {
5c5c5e51 665 struct fuse_read_in *inarg = &req->misc.read.in;
f3332114
MS		 666
667 inarg->read_flags |= FUSE_READ_LOCKOWNER;
668 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
669 }
36cf66ed
MP		 670
671 if (io->async)
672 return fuse_async_req_send(fc, req, count, io);
673
b93f858a 674 fuse_request_send(fc, req);
361b1eb5 675 return req->out.args[0].size;
04730fef
MS		 676}
677
5c5c5e51
MS		 678static void fuse_read_update_size(struct inode *inode, loff_t size,
679 u64 attr_ver)
680{
681 struct fuse_conn *fc = get_fuse_conn(inode);
682 struct fuse_inode *fi = get_fuse_inode(inode);
683
f15ecfef 684 spin_lock(&fi->lock);
06a7c3c2
MP		 685 if (attr_ver == fi->attr_version && size < inode->i_size &&
686 !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
4510d86f 687 fi->attr_version = atomic64_inc_return(&fc->attr_version);
5c5c5e51
MS		 688 i_size_write(inode, size);
689 }
f15ecfef 690 spin_unlock(&fi->lock);
5c5c5e51
MS		 691}
692
a92adc82
PE		 693static void fuse_short_read(struct fuse_req *req, struct inode *inode,
694 u64 attr_ver)
695{
696 size_t num_read = req->out.args[0].size;
8373200b
PE		 697 struct fuse_conn *fc = get_fuse_conn(inode);
698
699 if (fc->writeback_cache) {
700 /*
701 * A hole in a file. Some data after the hole are in page cache,
702 * but have not reached the client fs yet. So, the hole is not
703 * present there.
704 */
705 int i;
09cbfeaf
KS		 706 int start_idx = num_read >> PAGE_SHIFT;
707 size_t off = num_read & (PAGE_SIZE - 1);
a92adc82 708
8373200b 709 for (i = start_idx; i < req->num_pages; i++) {
09cbfeaf 710 zero_user_segment(req->pages[i], off, PAGE_SIZE);
8373200b
PE		 711 off = 0;
712 }
713 } else {
714 loff_t pos = page_offset(req->pages[0]) + num_read;
715 fuse_read_update_size(inode, pos, attr_ver);
716 }
a92adc82
PE		 717}
718
482fce55 719static int fuse_do_readpage(struct file *file, struct page *page)
b6aeaded 720{
e1c0eecb
MS		 721 struct kiocb iocb;
722 struct fuse_io_priv io;
b6aeaded
MS		 723 struct inode *inode = page->mapping->host;
724 struct fuse_conn *fc = get_fuse_conn(inode);
248d86e8 725 struct fuse_req *req;
5c5c5e51
MS		 726 size_t num_read;
727 loff_t pos = page_offset(page);
09cbfeaf 728 size_t count = PAGE_SIZE;
5c5c5e51 729 u64 attr_ver;
248d86e8
MS		 730 int err;
731
3be5a52b 732 /*
25985edc 733 * Page writeback can extend beyond the lifetime of the
3be5a52b
MS		 734 * page-cache page, so make sure we read a properly synced
735 * page.
736 */
737 fuse_wait_on_page_writeback(inode, page->index);
738
b111c8c0 739 req = fuse_get_req(fc, 1);
ce1d5a49 740 if (IS_ERR(req))
482fce55 741 return PTR_ERR(req);
b6aeaded 742
5c5c5e51
MS		 743 attr_ver = fuse_get_attr_version(fc);
744
b6aeaded 745 req->out.page_zeroing = 1;
f4975c67 746 req->out.argpages = 1;
b6aeaded
MS		 747 req->num_pages = 1;
748 req->pages[0] = page;
85f40aec 749 req->page_descs[0].length = count;
e1c0eecb
MS		 750 init_sync_kiocb(&iocb, file);
751 io = (struct fuse_io_priv) FUSE_IO_PRIV_SYNC(&iocb);
36cf66ed 752 num_read = fuse_send_read(req, &io, pos, count, NULL);
b6aeaded 753 err = req->out.h.error;
5c5c5e51
MS		 754
755 if (!err) {
756 /*
757 * Short read means EOF. If file size is larger, truncate it
758 */
759 if (num_read < count)
a92adc82 760 fuse_short_read(req, inode, attr_ver);
5c5c5e51 761
b6aeaded 762 SetPageUptodate(page);
5c5c5e51
MS		 763 }
764
a92adc82 765 fuse_put_request(fc, req);
482fce55
MP		 766
767 return err;
768}
769
770static int fuse_readpage(struct file *file, struct page *page)
771{
772 struct inode *inode = page->mapping->host;
773 int err;
774
775 err = -EIO;
776 if (is_bad_inode(inode))
777 goto out;
778
779 err = fuse_do_readpage(file, page);
451418fc 780 fuse_invalidate_atime(inode);
b6aeaded
MS		 781 out:
782 unlock_page(page);
783 return err;
784}
785
c1aa96a5 786static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
db50b96c 787{
c1aa96a5 788 int i;
5c5c5e51
MS		 789 size_t count = req->misc.read.in.size;
790 size_t num_read = req->out.args[0].size;
ce534fb0 791 struct address_space *mapping = NULL;
c1aa96a5 792
ce534fb0
MS		 793 for (i = 0; mapping == NULL && i < req->num_pages; i++)
794 mapping = req->pages[i]->mapping;
5c5c5e51 795
ce534fb0
MS		 796 if (mapping) {
797 struct inode *inode = mapping->host;
798
799 /*
800 * Short read means EOF. If file size is larger, truncate it
801 */
a92adc82
PE		 802 if (!req->out.h.error && num_read < count)
803 fuse_short_read(req, inode, req->misc.read.attr_ver);
ce534fb0 804
451418fc 805 fuse_invalidate_atime(inode);
ce534fb0 806 }
c1aa96a5 807
db50b96c
MS		 808 for (i = 0; i < req->num_pages; i++) {
809 struct page *page = req->pages[i];
810 if (!req->out.h.error)
811 SetPageUptodate(page);
c1aa96a5
MS		 812 else
813 SetPageError(page);
db50b96c 814 unlock_page(page);
09cbfeaf 815 put_page(page);
db50b96c 816 }
c756e0a4 817 if (req->ff)
2e64ff15 818 fuse_file_put(req->ff, false, false);
c1aa96a5
MS		 819}
820
2106cb18 821static void fuse_send_readpages(struct fuse_req *req, struct file *file)
c1aa96a5 822{
2106cb18
MS		 823 struct fuse_file *ff = file->private_data;
824 struct fuse_conn *fc = ff->fc;
c1aa96a5 825 loff_t pos = page_offset(req->pages[0]);
09cbfeaf 826 size_t count = req->num_pages << PAGE_SHIFT;
f4975c67
MS		 827
828 req->out.argpages = 1;
c1aa96a5 829 req->out.page_zeroing = 1;
ce534fb0 830 req->out.page_replace = 1;
2106cb18 831 fuse_read_fill(req, file, pos, count, FUSE_READ);
5c5c5e51 832 req->misc.read.attr_ver = fuse_get_attr_version(fc);
9cd68455 833 if (fc->async_read) {
c756e0a4 834 req->ff = fuse_file_get(ff);
9cd68455 835 req->end = fuse_readpages_end;
b93f858a 836 fuse_request_send_background(fc, req);
9cd68455 837 } else {
b93f858a 838 fuse_request_send(fc, req);
9cd68455 839 fuse_readpages_end(fc, req);
e9bb09dd 840 fuse_put_request(fc, req);
9cd68455 841 }
db50b96c
MS		 842}
843
c756e0a4 844struct fuse_fill_data {
db50b96c 845 struct fuse_req *req;
a6643094 846 struct file *file;
db50b96c 847 struct inode *inode;
f8dbdf81 848 unsigned nr_pages;
db50b96c
MS		 849};
850
851static int fuse_readpages_fill(void *_data, struct page *page)
852{
c756e0a4 853 struct fuse_fill_data *data = _data;
db50b96c
MS		 854 struct fuse_req *req = data->req;
855 struct inode *inode = data->inode;
856 struct fuse_conn *fc = get_fuse_conn(inode);
857
3be5a52b
MS		 858 fuse_wait_on_page_writeback(inode, page->index);
859
db50b96c 860 if (req->num_pages &&
5da784cc 861 (req->num_pages == fc->max_pages ||
09cbfeaf 862 (req->num_pages + 1) * PAGE_SIZE > fc->max_read ||
db50b96c 863 req->pages[req->num_pages - 1]->index + 1 != page->index)) {
5da784cc
CS		 864 unsigned int nr_alloc = min_t(unsigned int, data->nr_pages,
865 fc->max_pages);
2106cb18 866 fuse_send_readpages(req, data->file);
8b41e671
MP		 867 if (fc->async_read)
868 req = fuse_get_req_for_background(fc, nr_alloc);
869 else
870 req = fuse_get_req(fc, nr_alloc);
871
872 data->req = req;
ce1d5a49 873 if (IS_ERR(req)) {
db50b96c 874 unlock_page(page);
ce1d5a49 875 return PTR_ERR(req);
db50b96c 876 }
db50b96c 877 }
f8dbdf81
MP		 878
879 if (WARN_ON(req->num_pages >= req->max_pages)) {
109728cc 880 unlock_page(page);
f8dbdf81
MP		 881 fuse_put_request(fc, req);
882 return -EIO;
883 }
884
09cbfeaf 885 get_page(page);
db50b96c 886 req->pages[req->num_pages] = page;
85f40aec 887 req->page_descs[req->num_pages].length = PAGE_SIZE;
1729a16c 888 req->num_pages++;
f8dbdf81 889 data->nr_pages--;
db50b96c
MS		 890 return 0;
891}
892
893static int fuse_readpages(struct file *file, struct address_space *mapping,
894 struct list_head *pages, unsigned nr_pages)
895{
896 struct inode *inode = mapping->host;
897 struct fuse_conn *fc = get_fuse_conn(inode);
c756e0a4 898 struct fuse_fill_data data;
db50b96c 899 int err;
5da784cc 900 unsigned int nr_alloc = min_t(unsigned int, nr_pages, fc->max_pages);
248d86e8 901
1d7ea732 902 err = -EIO;
248d86e8 903 if (is_bad_inode(inode))
2e990021 904 goto out;
248d86e8 905
a6643094 906 data.file = file;
db50b96c 907 data.inode = inode;
8b41e671
MP		 908 if (fc->async_read)
909 data.req = fuse_get_req_for_background(fc, nr_alloc);
910 else
911 data.req = fuse_get_req(fc, nr_alloc);
f8dbdf81 912 data.nr_pages = nr_pages;
1d7ea732 913 err = PTR_ERR(data.req);
ce1d5a49 914 if (IS_ERR(data.req))
2e990021 915 goto out;
db50b96c
MS		 916
917 err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
d3406ffa
MS		 918 if (!err) {
919 if (data.req->num_pages)
2106cb18 920 fuse_send_readpages(data.req, file);
d3406ffa
MS		 921 else
922 fuse_put_request(fc, data.req);
923 }
2e990021 924out:
1d7ea732 925 return err;
db50b96c
MS		 926}
927
55752a3a 928static ssize_t fuse_cache_read_iter(struct kiocb *iocb, struct iov_iter *to)
bcb4be80
MS		 929{
930 struct inode *inode = iocb->ki_filp->f_mapping->host;
a8894274 931 struct fuse_conn *fc = get_fuse_conn(inode);
bcb4be80 932
a8894274
BF		 933 /*
934 * In auto invalidate mode, always update attributes on read.
935 * Otherwise, only update if we attempt to read past EOF (to ensure
936 * i_size is up to date).
937 */
938 if (fc->auto_inval_data ||
37c20f16 939 (iocb->ki_pos + iov_iter_count(to) > i_size_read(inode))) {
bcb4be80 940 int err;
5b97eeac 941 err = fuse_update_attributes(inode, iocb->ki_filp);
bcb4be80
MS		 942 if (err)
943 return err;
944 }
945
37c20f16 946 return generic_file_read_iter(iocb, to);
bcb4be80
MS		 947}
948
2d698b07 949static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff,
2106cb18 950 loff_t pos, size_t count)
b6aeaded 951{
b25e82e5
MS		 952 struct fuse_write_in *inarg = &req->misc.write.in;
953 struct fuse_write_out *outarg = &req->misc.write.out;
b6aeaded 954
b25e82e5
MS		 955 inarg->fh = ff->fh;
956 inarg->offset = pos;
957 inarg->size = count;
b6aeaded 958 req->in.h.opcode = FUSE_WRITE;
2106cb18 959 req->in.h.nodeid = ff->nodeid;
b6aeaded 960 req->in.numargs = 2;
2106cb18 961 if (ff->fc->minor < 9)
f3332114
MS		 962 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
963 else
964 req->in.args[0].size = sizeof(struct fuse_write_in);
b25e82e5 965 req->in.args[0].value = inarg;
b6aeaded
MS		 966 req->in.args[1].size = count;
967 req->out.numargs = 1;
968 req->out.args[0].size = sizeof(struct fuse_write_out);
b25e82e5
MS		 969 req->out.args[0].value = outarg;
970}
971
36cf66ed 972static size_t fuse_send_write(struct fuse_req *req, struct fuse_io_priv *io,
2106cb18 973 loff_t pos, size_t count, fl_owner_t owner)
b25e82e5 974{
e1c0eecb
MS		 975 struct kiocb *iocb = io->iocb;
976 struct file *file = iocb->ki_filp;
2106cb18
MS		 977 struct fuse_file *ff = file->private_data;
978 struct fuse_conn *fc = ff->fc;
2d698b07
MS		 979 struct fuse_write_in *inarg = &req->misc.write.in;
980
2106cb18 981 fuse_write_fill(req, ff, pos, count);
2d698b07 982 inarg->flags = file->f_flags;
e1c0eecb
MS		 983 if (iocb->ki_flags & IOCB_DSYNC)
984 inarg->flags |= O_DSYNC;
985 if (iocb->ki_flags & IOCB_SYNC)
986 inarg->flags |= O_SYNC;
f3332114 987 if (owner != NULL) {
f3332114
MS		 988 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
989 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
990 }
36cf66ed
MP		 991
992 if (io->async)
993 return fuse_async_req_send(fc, req, count, io);
994
b93f858a 995 fuse_request_send(fc, req);
b25e82e5 996 return req->misc.write.out.size;
b6aeaded
MS		 997}
998
b0aa7606 999bool fuse_write_update_size(struct inode *inode, loff_t pos)
854512ec
MS		 1000{
1001 struct fuse_conn *fc = get_fuse_conn(inode);
1002 struct fuse_inode *fi = get_fuse_inode(inode);
b0aa7606 1003 bool ret = false;
854512ec 1004
f15ecfef 1005 spin_lock(&fi->lock);
4510d86f 1006 fi->attr_version = atomic64_inc_return(&fc->attr_version);
b0aa7606 1007 if (pos > inode->i_size) {
854512ec 1008 i_size_write(inode, pos);
b0aa7606
MP		 1009 ret = true;
1010 }
f15ecfef 1011 spin_unlock(&fi->lock);
b0aa7606
MP		 1012
1013 return ret;
854512ec
MS		 1014}
1015
e1c0eecb 1016static size_t fuse_send_write_pages(struct fuse_req *req, struct kiocb *iocb,
ea9b9907
NP		 1017 struct inode *inode, loff_t pos,
1018 size_t count)
1019{
1020 size_t res;
1021 unsigned offset;
1022 unsigned i;
e1c0eecb 1023 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
ea9b9907
NP		 1024
1025 for (i = 0; i < req->num_pages; i++)
1026 fuse_wait_on_page_writeback(inode, req->pages[i]->index);
1027
36cf66ed 1028 res = fuse_send_write(req, &io, pos, count, NULL);
ea9b9907 1029
b2430d75 1030 offset = req->page_descs[0].offset;
ea9b9907
NP		 1031 count = res;
1032 for (i = 0; i < req->num_pages; i++) {
1033 struct page *page = req->pages[i];
1034
09cbfeaf 1035 if (!req->out.h.error && !offset && count >= PAGE_SIZE)
ea9b9907
NP		 1036 SetPageUptodate(page);
1037
09cbfeaf
KS		 1038 if (count > PAGE_SIZE - offset)
1039 count -= PAGE_SIZE - offset;
ea9b9907
NP		 1040 else
1041 count = 0;
1042 offset = 0;
1043
1044 unlock_page(page);
09cbfeaf 1045 put_page(page);
ea9b9907
NP		 1046 }
1047
1048 return res;
1049}
1050
1051static ssize_t fuse_fill_write_pages(struct fuse_req *req,
1052 struct address_space *mapping,
1053 struct iov_iter *ii, loff_t pos)
1054{
1055 struct fuse_conn *fc = get_fuse_conn(mapping->host);
09cbfeaf 1056 unsigned offset = pos & (PAGE_SIZE - 1);
ea9b9907
NP		 1057 size_t count = 0;
1058 int err;
1059
f4975c67 1060 req->in.argpages = 1;
b2430d75 1061 req->page_descs[0].offset = offset;
ea9b9907
NP		 1062
1063 do {
1064 size_t tmp;
1065 struct page *page;
09cbfeaf
KS		 1066 pgoff_t index = pos >> PAGE_SHIFT;
1067 size_t bytes = min_t(size_t, PAGE_SIZE - offset,
ea9b9907
NP		 1068 iov_iter_count(ii));
1069
1070 bytes = min_t(size_t, bytes, fc->max_write - count);
1071
1072 again:
1073 err = -EFAULT;
1074 if (iov_iter_fault_in_readable(ii, bytes))
1075 break;
1076
1077 err = -ENOMEM;
54566b2c 1078 page = grab_cache_page_write_begin(mapping, index, 0);
ea9b9907
NP		 1079 if (!page)
1080 break;
1081
931e80e4 1082 if (mapping_writably_mapped(mapping))
1083 flush_dcache_page(page);
1084
ea9b9907 1085 tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes);
ea9b9907
NP		 1086 flush_dcache_page(page);
1087
3ca8138f 1088 iov_iter_advance(ii, tmp);
ea9b9907
NP		 1089 if (!tmp) {
1090 unlock_page(page);
09cbfeaf 1091 put_page(page);
ea9b9907
NP		 1092 bytes = min(bytes, iov_iter_single_seg_count(ii));
1093 goto again;
1094 }
1095
1096 err = 0;
1097 req->pages[req->num_pages] = page;
85f40aec 1098 req->page_descs[req->num_pages].length = tmp;
ea9b9907
NP		 1099 req->num_pages++;
1100
ea9b9907
NP		 1101 count += tmp;
1102 pos += tmp;
1103 offset += tmp;
09cbfeaf 1104 if (offset == PAGE_SIZE)
ea9b9907
NP		 1105 offset = 0;
1106
78bb6cb9
MS		 1107 if (!fc->big_writes)
1108 break;
ea9b9907 1109 } while (iov_iter_count(ii) && count < fc->max_write &&
d07f09f5 1110 req->num_pages < req->max_pages && offset == 0);
ea9b9907
NP		 1111
1112 return count > 0 ? count : err;
1113}
1114
5da784cc
CS		 1115static inline unsigned int fuse_wr_pages(loff_t pos, size_t len,
1116 unsigned int max_pages)
d07f09f5 1117{
5da784cc 1118 return min_t(unsigned int,
09cbfeaf
KS		 1119 ((pos + len - 1) >> PAGE_SHIFT) -
1120 (pos >> PAGE_SHIFT) + 1,
5da784cc 1121 max_pages);
d07f09f5
MP		 1122}
1123
e1c0eecb 1124static ssize_t fuse_perform_write(struct kiocb *iocb,
ea9b9907
NP		 1125 struct address_space *mapping,
1126 struct iov_iter *ii, loff_t pos)
1127{
1128 struct inode *inode = mapping->host;
1129 struct fuse_conn *fc = get_fuse_conn(inode);
06a7c3c2 1130 struct fuse_inode *fi = get_fuse_inode(inode);
ea9b9907
NP		 1131 int err = 0;
1132 ssize_t res = 0;
1133
1134 if (is_bad_inode(inode))
1135 return -EIO;
1136
06a7c3c2
MP		 1137 if (inode->i_size < pos + iov_iter_count(ii))
1138 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1139
ea9b9907
NP		 1140 do {
1141 struct fuse_req *req;
1142 ssize_t count;
5da784cc
CS		 1143 unsigned int nr_pages = fuse_wr_pages(pos, iov_iter_count(ii),
1144 fc->max_pages);
ea9b9907 1145
d07f09f5 1146 req = fuse_get_req(fc, nr_pages);
ea9b9907
NP		 1147 if (IS_ERR(req)) {
1148 err = PTR_ERR(req);
1149 break;
1150 }
1151
1152 count = fuse_fill_write_pages(req, mapping, ii, pos);
1153 if (count <= 0) {
1154 err = count;
1155 } else {
1156 size_t num_written;
1157
e1c0eecb 1158 num_written = fuse_send_write_pages(req, iocb, inode,
ea9b9907
NP		 1159 pos, count);
1160 err = req->out.h.error;
1161 if (!err) {
1162 res += num_written;
1163 pos += num_written;
1164
1165 /* break out of the loop on short write */
1166 if (num_written != count)
1167 err = -EIO;
1168 }
1169 }
1170 fuse_put_request(fc, req);
1171 } while (!err && iov_iter_count(ii));
1172
1173 if (res > 0)
1174 fuse_write_update_size(inode, pos);
1175
06a7c3c2 1176 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
ea9b9907
NP		 1177 fuse_invalidate_attr(inode);
1178
1179 return res > 0 ? res : err;
1180}
1181
55752a3a 1182static ssize_t fuse_cache_write_iter(struct kiocb *iocb, struct iov_iter *from)
ea9b9907
NP		 1183{
1184 struct file *file = iocb->ki_filp;
1185 struct address_space *mapping = file->f_mapping;
ea9b9907 1186 ssize_t written = 0;
4273b793 1187 ssize_t written_buffered = 0;
ea9b9907
NP		 1188 struct inode *inode = mapping->host;
1189 ssize_t err;
4273b793 1190 loff_t endbyte = 0;
ea9b9907 1191
4d99ff8f
PE		 1192 if (get_fuse_conn(inode)->writeback_cache) {
1193 /* Update size (EOF optimization) and mode (SUID clearing) */
5b97eeac 1194 err = fuse_update_attributes(mapping->host, file);
4d99ff8f
PE		 1195 if (err)
1196 return err;
1197
84c3d55c 1198 return generic_file_write_iter(iocb, from);
4d99ff8f
PE		 1199 }
1200
5955102c 1201 inode_lock(inode);
ea9b9907
NP		 1202
1203 /* We can write back this queue in page reclaim */
de1414a6 1204 current->backing_dev_info = inode_to_bdi(inode);
ea9b9907 1205
3309dd04
AV		 1206 err = generic_write_checks(iocb, from);
1207 if (err <= 0)
ea9b9907
NP		 1208 goto out;
1209
5fa8e0a1 1210 err = file_remove_privs(file);
ea9b9907
NP		 1211 if (err)
1212 goto out;
1213
c3b2da31
JB		 1214 err = file_update_time(file);
1215 if (err)
1216 goto out;
ea9b9907 1217
2ba48ce5 1218 if (iocb->ki_flags & IOCB_DIRECT) {
3309dd04 1219 loff_t pos = iocb->ki_pos;
1af5bb49 1220 written = generic_file_direct_write(iocb, from);
84c3d55c 1221 if (written < 0 || !iov_iter_count(from))
4273b793
AA		 1222 goto out;
1223
1224 pos += written;
ea9b9907 1225
e1c0eecb 1226 written_buffered = fuse_perform_write(iocb, mapping, from, pos);
4273b793
AA		 1227 if (written_buffered < 0) {
1228 err = written_buffered;
1229 goto out;
1230 }
1231 endbyte = pos + written_buffered - 1;
1232
1233 err = filemap_write_and_wait_range(file->f_mapping, pos,
1234 endbyte);
1235 if (err)
1236 goto out;
1237
1238 invalidate_mapping_pages(file->f_mapping,
09cbfeaf
KS		 1239 pos >> PAGE_SHIFT,
1240 endbyte >> PAGE_SHIFT);
4273b793
AA		 1241
1242 written += written_buffered;
1243 iocb->ki_pos = pos + written_buffered;
1244 } else {
e1c0eecb 1245 written = fuse_perform_write(iocb, mapping, from, iocb->ki_pos);
4273b793 1246 if (written >= 0)
3309dd04 1247 iocb->ki_pos += written;
4273b793 1248 }
ea9b9907
NP		 1249out:
1250 current->backing_dev_info = NULL;
5955102c 1251 inode_unlock(inode);
e1c0eecb
MS		 1252 if (written > 0)
1253 written = generic_write_sync(iocb, written);
ea9b9907
NP		 1254
1255 return written ? written : err;
1256}
1257
7c190c8b
MP		 1258static inline void fuse_page_descs_length_init(struct fuse_req *req,
1259 unsigned index, unsigned nr_pages)
85f40aec
MP		 1260{
1261 int i;
1262
7c190c8b 1263 for (i = index; i < index + nr_pages; i++)
85f40aec
MP		 1264 req->page_descs[i].length = PAGE_SIZE -
1265 req->page_descs[i].offset;
1266}
1267
7c190c8b
MP		 1268static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii)
1269{
1270 return (unsigned long)ii->iov->iov_base + ii->iov_offset;
1271}
1272
1273static inline size_t fuse_get_frag_size(const struct iov_iter *ii,
1274 size_t max_size)
1275{
1276 return min(iov_iter_single_seg_count(ii), max_size);
1277}
1278
b98d023a 1279static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii,
ce60a2f1 1280 size_t *nbytesp, int write)
413ef8cb 1281{
7c190c8b 1282 size_t nbytes = 0; /* # bytes already packed in req */
742f9927 1283 ssize_t ret = 0;
b98d023a 1284
f4975c67 1285 /* Special case for kernel I/O: can copy directly into the buffer */
00e23707 1286 if (iov_iter_is_kvec(ii)) {
7c190c8b
MP		 1287 unsigned long user_addr = fuse_get_user_addr(ii);
1288 size_t frag_size = fuse_get_frag_size(ii, *nbytesp);
1289
f4975c67
MS		 1290 if (write)
1291 req->in.args[1].value = (void *) user_addr;
1292 else
1293 req->out.args[0].value = (void *) user_addr;
1294
b98d023a
MP		 1295 iov_iter_advance(ii, frag_size);
1296 *nbytesp = frag_size;
f4975c67
MS		 1297 return 0;
1298 }
413ef8cb 1299
5565a9d8 1300 while (nbytes < *nbytesp && req->num_pages < req->max_pages) {
7c190c8b 1301 unsigned npages;
f67da30c 1302 size_t start;
742f9927 1303 ret = iov_iter_get_pages(ii, &req->pages[req->num_pages],
2c80929c 1304 *nbytesp - nbytes,
c7f3888a
AV		 1305 req->max_pages - req->num_pages,
1306 &start);
7c190c8b 1307 if (ret < 0)
742f9927 1308 break;
7c190c8b 1309
c9c37e2e
AV		 1310 iov_iter_advance(ii, ret);
1311 nbytes += ret;
7c190c8b 1312
c9c37e2e
AV		 1313 ret += start;
1314 npages = (ret + PAGE_SIZE - 1) / PAGE_SIZE;
7c190c8b 1315
c9c37e2e 1316 req->page_descs[req->num_pages].offset = start;
7c190c8b
MP		 1317 fuse_page_descs_length_init(req, req->num_pages, npages);
1318
1319 req->num_pages += npages;
1320 req->page_descs[req->num_pages - 1].length -=
c9c37e2e 1321 (PAGE_SIZE - ret) & (PAGE_SIZE - 1);
7c190c8b 1322 }
f4975c67
MS		 1323
1324 if (write)
1325 req->in.argpages = 1;
1326 else
1327 req->out.argpages = 1;
1328
7c190c8b 1329 *nbytesp = nbytes;
f4975c67 1330
2c932d4c 1331 return ret < 0 ? ret : 0;
413ef8cb
MS		 1332}
1333
d22a943f
AV		 1334ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
1335 loff_t *ppos, int flags)
413ef8cb 1336{
ea8cd333
PE		 1337 int write = flags & FUSE_DIO_WRITE;
1338 int cuse = flags & FUSE_DIO_CUSE;
e1c0eecb 1339 struct file *file = io->iocb->ki_filp;
ea8cd333 1340 struct inode *inode = file->f_mapping->host;
2106cb18
MS		 1341 struct fuse_file *ff = file->private_data;
1342 struct fuse_conn *fc = ff->fc;
413ef8cb
MS		 1343 size_t nmax = write ? fc->max_write : fc->max_read;
1344 loff_t pos = *ppos;
d22a943f 1345 size_t count = iov_iter_count(iter);
09cbfeaf
KS		 1346 pgoff_t idx_from = pos >> PAGE_SHIFT;
1347 pgoff_t idx_to = (pos + count - 1) >> PAGE_SHIFT;
413ef8cb 1348 ssize_t res = 0;
248d86e8 1349 struct fuse_req *req;
742f9927 1350 int err = 0;
248d86e8 1351
de82b923 1352 if (io->async)
5da784cc
CS		 1353 req = fuse_get_req_for_background(fc, iov_iter_npages(iter,
1354 fc->max_pages));
de82b923 1355 else
5da784cc 1356 req = fuse_get_req(fc, iov_iter_npages(iter, fc->max_pages));
ce1d5a49
MS		 1357 if (IS_ERR(req))
1358 return PTR_ERR(req);
413ef8cb 1359
ea8cd333
PE		 1360 if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) {
1361 if (!write)
5955102c 1362 inode_lock(inode);
ea8cd333
PE		 1363 fuse_sync_writes(inode);
1364 if (!write)
5955102c 1365 inode_unlock(inode);
ea8cd333
PE		 1366 }
1367
61c12b49 1368 io->should_dirty = !write && iter_is_iovec(iter);
413ef8cb 1369 while (count) {
413ef8cb 1370 size_t nres;
2106cb18 1371 fl_owner_t owner = current->files;
f4975c67 1372 size_t nbytes = min(count, nmax);
742f9927
AS		 1373 err = fuse_get_user_pages(req, iter, &nbytes, write);
1374 if (err && !nbytes)
413ef8cb 1375 break;
f4975c67 1376
413ef8cb 1377 if (write)
36cf66ed 1378 nres = fuse_send_write(req, io, pos, nbytes, owner);
413ef8cb 1379 else
36cf66ed 1380 nres = fuse_send_read(req, io, pos, nbytes, owner);
2106cb18 1381
36cf66ed 1382 if (!io->async)
61c12b49 1383 fuse_release_user_pages(req, io->should_dirty);
413ef8cb 1384 if (req->out.h.error) {
742f9927 1385 err = req->out.h.error;
413ef8cb
MS		 1386 break;
1387 } else if (nres > nbytes) {
742f9927
AS		 1388 res = 0;
1389 err = -EIO;
413ef8cb
MS		 1390 break;
1391 }
1392 count -= nres;
1393 res += nres;
1394 pos += nres;
413ef8cb
MS		 1395 if (nres != nbytes)
1396 break;
56cf34ff
MS		 1397 if (count) {
1398 fuse_put_request(fc, req);
de82b923
BF		 1399 if (io->async)
1400 req = fuse_get_req_for_background(fc,
5da784cc 1401 iov_iter_npages(iter, fc->max_pages));
de82b923 1402 else
5da784cc
CS		 1403 req = fuse_get_req(fc, iov_iter_npages(iter,
1404 fc->max_pages));
56cf34ff
MS		 1405 if (IS_ERR(req))
1406 break;
1407 }
413ef8cb 1408 }
f60311d5
AA		 1409 if (!IS_ERR(req))
1410 fuse_put_request(fc, req);
d09cb9d7 1411 if (res > 0)
413ef8cb 1412 *ppos = pos;
413ef8cb 1413
742f9927 1414 return res > 0 ? res : err;
413ef8cb 1415}
08cbf542 1416EXPORT_SYMBOL_GPL(fuse_direct_io);
413ef8cb 1417
36cf66ed 1418static ssize_t __fuse_direct_read(struct fuse_io_priv *io,
d22a943f
AV		 1419 struct iov_iter *iter,
1420 loff_t *ppos)
413ef8cb 1421{
d09cb9d7 1422 ssize_t res;
e1c0eecb 1423 struct inode *inode = file_inode(io->iocb->ki_filp);
d09cb9d7
MS		 1424
1425 if (is_bad_inode(inode))
1426 return -EIO;
1427
d22a943f 1428 res = fuse_direct_io(io, iter, ppos, 0);
d09cb9d7 1429
9a2eb24d 1430 fuse_invalidate_atime(inode);
d09cb9d7
MS		 1431
1432 return res;
413ef8cb
MS		 1433}
1434
23c94e1c
MR		 1435static ssize_t fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter);
1436
15316263 1437static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to)
b98d023a 1438{
23c94e1c
MR		 1439 ssize_t res;
1440
1441 if (!is_sync_kiocb(iocb) && iocb->ki_flags & IOCB_DIRECT) {
1442 struct file *file = iocb->ki_filp;
1443
1444 if (is_bad_inode(file_inode(file)))
1445 return -EIO;
1446
1447 res = fuse_direct_IO(iocb, to);
1448 } else {
1449 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
1450
1451 res = __fuse_direct_read(&io, to, &iocb->ki_pos);
1452 }
1453
1454 return res;
b98d023a
MP		 1455}
1456
15316263 1457static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
4273b793 1458{
e1c0eecb
MS		 1459 struct inode *inode = file_inode(iocb->ki_filp);
1460 struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
15316263 1461 ssize_t res;
4273b793
AA		 1462
1463 if (is_bad_inode(inode))
1464 return -EIO;
1465
1466 /* Don't allow parallel writes to the same file */
5955102c 1467 inode_lock(inode);
3309dd04 1468 res = generic_write_checks(iocb, from);
23c94e1c
MR		 1469 if (res > 0) {
1470 if (!is_sync_kiocb(iocb) && iocb->ki_flags & IOCB_DIRECT) {
1471 res = fuse_direct_IO(iocb, from);
1472 } else {
1473 res = fuse_direct_io(&io, from, &iocb->ki_pos,
1474 FUSE_DIO_WRITE);
1475 }
1476 }
812408fb 1477 fuse_invalidate_attr(inode);
bcba24cc 1478 if (res > 0)
15316263 1479 fuse_write_update_size(inode, iocb->ki_pos);
5955102c 1480 inode_unlock(inode);
4273b793
AA		 1481
1482 return res;
1483}
1484
55752a3a
MS		 1485static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1486{
1487 struct fuse_file *ff = iocb->ki_filp->private_data;
1488
1489 if (!(ff->open_flags & FOPEN_DIRECT_IO))
1490 return fuse_cache_read_iter(iocb, to);
1491 else
1492 return fuse_direct_read_iter(iocb, to);
1493}
1494
1495static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1496{
1497 struct fuse_file *ff = iocb->ki_filp->private_data;
1498
1499 if (!(ff->open_flags & FOPEN_DIRECT_IO))
1500 return fuse_cache_write_iter(iocb, from);
1501 else
1502 return fuse_direct_write_iter(iocb, from);
1503}
1504
3be5a52b 1505static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
b6aeaded 1506{
385b1268
PE		 1507 int i;
1508
1509 for (i = 0; i < req->num_pages; i++)
1510 __free_page(req->pages[i]);
8b284dc4
MS		 1511
1512 if (req->ff)
2e64ff15 1513 fuse_file_put(req->ff, false, false);
3be5a52b
MS		 1514}
1515
1516static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
1517{
1518 struct inode *inode = req->inode;
1519 struct fuse_inode *fi = get_fuse_inode(inode);
de1414a6 1520 struct backing_dev_info *bdi = inode_to_bdi(inode);
385b1268 1521 int i;
3be5a52b
MS		 1522
1523 list_del(&req->writepages_entry);
385b1268 1524 for (i = 0; i < req->num_pages; i++) {
93f78d88 1525 dec_wb_stat(&bdi->wb, WB_WRITEBACK);
11fb9989 1526 dec_node_page_state(req->pages[i], NR_WRITEBACK_TEMP);
93f78d88 1527 wb_writeout_inc(&bdi->wb);
385b1268 1528 }
3be5a52b
MS		 1529 wake_up(&fi->page_waitq);
1530}
1531
f15ecfef 1532/* Called under fi->lock, may release and reacquire it */
6eaf4782
MP		 1533static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req,
1534 loff_t size)
f15ecfef
KT		 1535__releases(fi->lock)
1536__acquires(fi->lock)
3be5a52b 1537{
e2653bd5 1538 struct fuse_req *aux, *next;
3be5a52b 1539 struct fuse_inode *fi = get_fuse_inode(req->inode);
3be5a52b 1540 struct fuse_write_in *inarg = &req->misc.write.in;
09cbfeaf 1541 __u64 data_size = req->num_pages * PAGE_SIZE;
63825b4e 1542 bool queued;
3be5a52b 1543
385b1268
PE		 1544 if (inarg->offset + data_size <= size) {
1545 inarg->size = data_size;
3be5a52b 1546 } else if (inarg->offset < size) {
385b1268 1547 inarg->size = size - inarg->offset;
3be5a52b
MS		 1548 } else {
1549 /* Got truncated off completely */
1550 goto out_free;
b6aeaded 1551 }
3be5a52b
MS		 1552
1553 req->in.args[1].size = inarg->size;
63825b4e 1554 queued = fuse_request_queue_background(fc, req);
f15ecfef
KT		 1555 /* Fails on broken connection only */
1556 if (unlikely(!queued))
1557 goto out_free;
1558
1559 fi->writectr++;
3be5a52b
MS		 1560 return;
1561
1562 out_free:
1563 fuse_writepage_finish(fc, req);
f15ecfef 1564 spin_unlock(&fi->lock);
e2653bd5
MS		 1565
1566 /* After fuse_writepage_finish() aux request list is private */
1567 for (aux = req->misc.write.next; aux; aux = next) {
1568 next = aux->misc.write.next;
1569 aux->misc.write.next = NULL;
1570 fuse_writepage_free(fc, aux);
1571 fuse_put_request(fc, aux);
1572 }
1573
3be5a52b 1574 fuse_writepage_free(fc, req);
e9bb09dd 1575 fuse_put_request(fc, req);
f15ecfef 1576 spin_lock(&fi->lock);
b6aeaded
MS		 1577}
1578
3be5a52b
MS		 1579/*
1580 * If fi->writectr is positive (no truncate or fsync going on) send
1581 * all queued writepage requests.
1582 *
f15ecfef 1583 * Called with fi->lock
3be5a52b
MS		 1584 */
1585void fuse_flush_writepages(struct inode *inode)
f15ecfef
KT		 1586__releases(fi->lock)
1587__acquires(fi->lock)
b6aeaded 1588{
3be5a52b
MS		 1589 struct fuse_conn *fc = get_fuse_conn(inode);
1590 struct fuse_inode *fi = get_fuse_inode(inode);
6eaf4782 1591 size_t crop = i_size_read(inode);
3be5a52b
MS		 1592 struct fuse_req *req;
1593
1594 while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
1595 req = list_entry(fi->queued_writes.next, struct fuse_req, list);
1596 list_del_init(&req->list);
6eaf4782 1597 fuse_send_writepage(fc, req, crop);
3be5a52b
MS		 1598 }
1599}
1600
1601static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
1602{
1603 struct inode *inode = req->inode;
1604 struct fuse_inode *fi = get_fuse_inode(inode);
1605
1606 mapping_set_error(inode->i_mapping, req->out.h.error);
f15ecfef 1607 spin_lock(&fi->lock);
8b284dc4 1608 while (req->misc.write.next) {
6eaf4782
MP		 1609 struct fuse_conn *fc = get_fuse_conn(inode);
1610 struct fuse_write_in *inarg = &req->misc.write.in;
8b284dc4
MS		 1611 struct fuse_req *next = req->misc.write.next;
1612 req->misc.write.next = next->misc.write.next;
1613 next->misc.write.next = NULL;
ce128de6 1614 next->ff = fuse_file_get(req->ff);
8b284dc4 1615 list_add(&next->writepages_entry, &fi->writepages);
6eaf4782
MP		 1616
1617 /*
1618 * Skip fuse_flush_writepages() to make it easy to crop requests
1619 * based on primary request size.
1620 *
1621 * 1st case (trivial): there are no concurrent activities using
1622 * fuse_set/release_nowrite. Then we're on safe side because
1623 * fuse_flush_writepages() would call fuse_send_writepage()
1624 * anyway.
1625 *
1626 * 2nd case: someone called fuse_set_nowrite and it is waiting
1627 * now for completion of all in-flight requests. This happens
1628 * rarely and no more than once per page, so this should be
1629 * okay.
1630 *
1631 * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle
1632 * of fuse_set_nowrite..fuse_release_nowrite section. The fact
1633 * that fuse_set_nowrite returned implies that all in-flight
1634 * requests were completed along with all of their secondary
1635 * requests. Further primary requests are blocked by negative
1636 * writectr. Hence there cannot be any in-flight requests and
1637 * no invocations of fuse_writepage_end() while we're in
1638 * fuse_set_nowrite..fuse_release_nowrite section.
1639 */
1640 fuse_send_writepage(fc, next, inarg->offset + inarg->size);
8b284dc4 1641 }
3be5a52b
MS		 1642 fi->writectr--;
1643 fuse_writepage_finish(fc, req);
f15ecfef 1644 spin_unlock(&fi->lock);
3be5a52b
MS		 1645 fuse_writepage_free(fc, req);
1646}
1647
1e18bda8
MS		 1648static struct fuse_file *__fuse_write_file_get(struct fuse_conn *fc,
1649 struct fuse_inode *fi)
adcadfa8 1650{
72523425 1651 struct fuse_file *ff = NULL;
adcadfa8 1652
f15ecfef 1653 spin_lock(&fi->lock);
1e18bda8 1654 if (!list_empty(&fi->write_files)) {
72523425
MS		 1655 ff = list_entry(fi->write_files.next, struct fuse_file,
1656 write_entry);
1657 fuse_file_get(ff);
1658 }
f15ecfef 1659 spin_unlock(&fi->lock);
adcadfa8
PE		 1660
1661 return ff;
1662}
1663
1e18bda8
MS		 1664static struct fuse_file *fuse_write_file_get(struct fuse_conn *fc,
1665 struct fuse_inode *fi)
1666{
1667 struct fuse_file *ff = __fuse_write_file_get(fc, fi);
1668 WARN_ON(!ff);
1669 return ff;
1670}
1671
1672int fuse_write_inode(struct inode *inode, struct writeback_control *wbc)
1673{
1674 struct fuse_conn *fc = get_fuse_conn(inode);
1675 struct fuse_inode *fi = get_fuse_inode(inode);
1676 struct fuse_file *ff;
1677 int err;
1678
1679 ff = __fuse_write_file_get(fc, fi);
ab9e13f7 1680 err = fuse_flush_times(inode, ff);
1e18bda8 1681 if (ff)
2e64ff15 1682 fuse_file_put(ff, false, false);
1e18bda8
MS		 1683
1684 return err;
1685}
1686
3be5a52b
MS		 1687static int fuse_writepage_locked(struct page *page)
1688{
1689 struct address_space *mapping = page->mapping;
1690 struct inode *inode = mapping->host;
1691 struct fuse_conn *fc = get_fuse_conn(inode);
1692 struct fuse_inode *fi = get_fuse_inode(inode);
1693 struct fuse_req *req;
3be5a52b 1694 struct page *tmp_page;
72523425 1695 int error = -ENOMEM;
3be5a52b
MS		 1696
1697 set_page_writeback(page);
1698
4250c066 1699 req = fuse_request_alloc_nofs(1);
3be5a52b
MS		 1700 if (!req)
1701 goto err;
1702
825d6d33
MS		 1703 /* writeback always goes to bg_queue */
1704 __set_bit(FR_BACKGROUND, &req->flags);
3be5a52b
MS		 1705 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1706 if (!tmp_page)
1707 goto err_free;
1708
72523425 1709 error = -EIO;
26d614df 1710 req->ff = fuse_write_file_get(fc, fi);
72523425 1711 if (!req->ff)
27f1b363 1712 goto err_nofile;
72523425 1713
adcadfa8 1714 fuse_write_fill(req, req->ff, page_offset(page), 0);
3be5a52b
MS		 1715
1716 copy_highpage(tmp_page, page);
2d698b07 1717 req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
8b284dc4 1718 req->misc.write.next = NULL;
f4975c67 1719 req->in.argpages = 1;
3be5a52b
MS		 1720 req->num_pages = 1;
1721 req->pages[0] = tmp_page;
b2430d75 1722 req->page_descs[0].offset = 0;
85f40aec 1723 req->page_descs[0].length = PAGE_SIZE;
3be5a52b
MS		 1724 req->end = fuse_writepage_end;
1725 req->inode = inode;
1726
93f78d88 1727 inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
11fb9989 1728 inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP);
3be5a52b 1729
f15ecfef 1730 spin_lock(&fi->lock);
3be5a52b
MS		 1731 list_add(&req->writepages_entry, &fi->writepages);
1732 list_add_tail(&req->list, &fi->queued_writes);
1733 fuse_flush_writepages(inode);
f15ecfef 1734 spin_unlock(&fi->lock);
3be5a52b 1735
4a4ac4eb
MP		 1736 end_page_writeback(page);
1737
3be5a52b
MS		 1738 return 0;
1739
27f1b363
MP		 1740err_nofile:
1741 __free_page(tmp_page);
3be5a52b
MS		 1742err_free:
1743 fuse_request_free(req);
1744err:
9183976e 1745 mapping_set_error(page->mapping, error);
3be5a52b 1746 end_page_writeback(page);
72523425 1747 return error;
3be5a52b
MS		 1748}
1749
1750static int fuse_writepage(struct page *page, struct writeback_control *wbc)
1751{
1752 int err;
1753
ff17be08
MS		 1754 if (fuse_page_is_writeback(page->mapping->host, page->index)) {
1755 /*
1756 * ->writepages() should be called for sync() and friends. We
1757 * should only get here on direct reclaim and then we are
1758 * allowed to skip a page which is already in flight
1759 */
1760 WARN_ON(wbc->sync_mode == WB_SYNC_ALL);
1761
1762 redirty_page_for_writepage(wbc, page);
1763 return 0;
1764 }
1765
3be5a52b
MS		 1766 err = fuse_writepage_locked(page);
1767 unlock_page(page);
1768
1769 return err;
1770}
1771
26d614df
PE		 1772struct fuse_fill_wb_data {
1773 struct fuse_req *req;
1774 struct fuse_file *ff;
1775 struct inode *inode;
2d033eaa 1776 struct page **orig_pages;
26d614df
PE		 1777};
1778
1779static void fuse_writepages_send(struct fuse_fill_wb_data *data)
1780{
1781 struct fuse_req *req = data->req;
1782 struct inode *inode = data->inode;
26d614df 1783 struct fuse_inode *fi = get_fuse_inode(inode);
2d033eaa
MP		 1784 int num_pages = req->num_pages;
1785 int i;
26d614df
PE		 1786
1787 req->ff = fuse_file_get(data->ff);
f15ecfef 1788 spin_lock(&fi->lock);
26d614df
PE		 1789 list_add_tail(&req->list, &fi->queued_writes);
1790 fuse_flush_writepages(inode);
f15ecfef 1791 spin_unlock(&fi->lock);
2d033eaa
MP		 1792
1793 for (i = 0; i < num_pages; i++)
1794 end_page_writeback(data->orig_pages[i]);
26d614df
PE		 1795}
1796
7f305ca1
MS		 1797/*
1798 * First recheck under fi->lock if the offending offset is still under
419234d5
MS		 1799 * writeback. If yes, then iterate auxiliary write requests, to see if there's
1800 * one already added for a page at this offset. If there's none, then insert
1801 * this new request onto the auxiliary list, otherwise reuse the existing one by
7f305ca1
MS		 1802 * copying the new page contents over to the old temporary page.
1803 */
8b284dc4
MS		 1804static bool fuse_writepage_in_flight(struct fuse_req *new_req,
1805 struct page *page)
1806{
1807 struct fuse_conn *fc = get_fuse_conn(new_req->inode);
1808 struct fuse_inode *fi = get_fuse_inode(new_req->inode);
1809 struct fuse_req *tmp;
1810 struct fuse_req *old_req;
8b284dc4 1811
7f305ca1 1812 WARN_ON(new_req->num_pages != 0);
8b284dc4 1813
f15ecfef 1814 spin_lock(&fi->lock);
8b284dc4 1815 list_del(&new_req->writepages_entry);
2fe93bd4
MS		 1816 old_req = fuse_find_writeback(fi, page->index, page->index);
1817 if (!old_req) {
f6011081 1818 list_add(&new_req->writepages_entry, &fi->writepages);
f15ecfef 1819 spin_unlock(&fi->lock);
2fe93bd4 1820 return false;
f6011081 1821 }
8b284dc4 1822
f6011081 1823 new_req->num_pages = 1;
419234d5 1824 for (tmp = old_req->misc.write.next; tmp; tmp = tmp->misc.write.next) {
7f305ca1
MS		 1825 pgoff_t curr_index;
1826
1827 WARN_ON(tmp->inode != new_req->inode);
09cbfeaf 1828 curr_index = tmp->misc.write.in.offset >> PAGE_SHIFT;
419234d5
MS		 1829 if (curr_index == page->index) {
1830 WARN_ON(tmp->num_pages != 1);
1831 WARN_ON(!test_bit(FR_PENDING, &tmp->flags));
c5de16cc 1832 swap(tmp->pages[0], new_req->pages[0]);
7f305ca1 1833 break;
8b284dc4
MS		 1834 }
1835 }
1836
7f305ca1
MS		 1837 if (!tmp) {
1838 new_req->misc.write.next = old_req->misc.write.next;
1839 old_req->misc.write.next = new_req;
1840 }
41b6e41f 1841
f15ecfef 1842 spin_unlock(&fi->lock);
7f305ca1
MS		 1843
1844 if (tmp) {
1845 struct backing_dev_info *bdi = inode_to_bdi(new_req->inode);
8b284dc4 1846
93f78d88 1847 dec_wb_stat(&bdi->wb, WB_WRITEBACK);
a2ebba82 1848 dec_node_page_state(new_req->pages[0], NR_WRITEBACK_TEMP);
93f78d88 1849 wb_writeout_inc(&bdi->wb);
8b284dc4
MS		 1850 fuse_writepage_free(fc, new_req);
1851 fuse_request_free(new_req);
8b284dc4 1852 }
7f305ca1 1853
2fe93bd4 1854 return true;
8b284dc4
MS		 1855}
1856
26d614df
PE		 1857static int fuse_writepages_fill(struct page *page,
1858 struct writeback_control *wbc, void *_data)
1859{
1860 struct fuse_fill_wb_data *data = _data;
1861 struct fuse_req *req = data->req;
1862 struct inode *inode = data->inode;
f15ecfef 1863 struct fuse_inode *fi = get_fuse_inode(inode);
26d614df
PE		 1864 struct fuse_conn *fc = get_fuse_conn(inode);
1865 struct page *tmp_page;
8b284dc4 1866 bool is_writeback;
26d614df
PE		 1867 int err;
1868
1869 if (!data->ff) {
1870 err = -EIO;
1871 data->ff = fuse_write_file_get(fc, get_fuse_inode(inode));
1872 if (!data->ff)
1873 goto out_unlock;
1874 }
1875
8b284dc4
MS		 1876 /*
1877 * Being under writeback is unlikely but possible. For example direct
1878 * read to an mmaped fuse file will set the page dirty twice; once when
1879 * the pages are faulted with get_user_pages(), and then after the read
1880 * completed.
1881 */
1882 is_writeback = fuse_page_is_writeback(inode, page->index);
1883
1884 if (req && req->num_pages &&
5da784cc 1885 (is_writeback || req->num_pages == fc->max_pages ||
09cbfeaf 1886 (req->num_pages + 1) * PAGE_SIZE > fc->max_write ||
8b284dc4
MS		 1887 data->orig_pages[req->num_pages - 1]->index + 1 != page->index)) {
1888 fuse_writepages_send(data);
1889 data->req = NULL;
e52a8250
MS		 1890 } else if (req && req->num_pages == req->max_pages) {
1891 if (!fuse_req_realloc_pages(fc, req, GFP_NOFS)) {
1892 fuse_writepages_send(data);
1893 req = data->req = NULL;
1894 }
26d614df 1895 }
e52a8250 1896
26d614df
PE		 1897 err = -ENOMEM;
1898 tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1899 if (!tmp_page)
1900 goto out_unlock;
1901
1902 /*
1903 * The page must not be redirtied until the writeout is completed
1904 * (i.e. userspace has sent a reply to the write request). Otherwise
1905 * there could be more than one temporary page instance for each real
1906 * page.
1907 *
1908 * This is ensured by holding the page lock in page_mkwrite() while
1909 * checking fuse_page_is_writeback(). We already hold the page lock
1910 * since clear_page_dirty_for_io() and keep it held until we add the
1911 * request to the fi->writepages list and increment req->num_pages.
1912 * After this fuse_page_is_writeback() will indicate that the page is
1913 * under writeback, so we can release the page lock.
1914 */
1915 if (data->req == NULL) {
1916 struct fuse_inode *fi = get_fuse_inode(inode);
1917
1918 err = -ENOMEM;
e52a8250 1919 req = fuse_request_alloc_nofs(FUSE_REQ_INLINE_PAGES);
26d614df
PE		 1920 if (!req) {
1921 __free_page(tmp_page);
1922 goto out_unlock;
1923 }
1924
1925 fuse_write_fill(req, data->ff, page_offset(page), 0);
1926 req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
8b284dc4 1927 req->misc.write.next = NULL;
26d614df 1928 req->in.argpages = 1;
825d6d33 1929 __set_bit(FR_BACKGROUND, &req->flags);
26d614df
PE		 1930 req->num_pages = 0;
1931 req->end = fuse_writepage_end;
1932 req->inode = inode;
1933
f15ecfef 1934 spin_lock(&fi->lock);
26d614df 1935 list_add(&req->writepages_entry, &fi->writepages);
f15ecfef 1936 spin_unlock(&fi->lock);
26d614df
PE		 1937
1938 data->req = req;
1939 }
1940 set_page_writeback(page);
1941
1942 copy_highpage(tmp_page, page);
1943 req->pages[req->num_pages] = tmp_page;
1944 req->page_descs[req->num_pages].offset = 0;
1945 req->page_descs[req->num_pages].length = PAGE_SIZE;
1946
93f78d88 1947 inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
11fb9989 1948 inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP);
8b284dc4
MS		 1949
1950 err = 0;
1951 if (is_writeback && fuse_writepage_in_flight(req, page)) {
1952 end_page_writeback(page);
1953 data->req = NULL;
1954 goto out_unlock;
1955 }
2d033eaa 1956 data->orig_pages[req->num_pages] = page;
26d614df
PE		 1957
1958 /*
f15ecfef 1959 * Protected by fi->lock against concurrent access by
26d614df
PE		 1960 * fuse_page_is_writeback().
1961 */
f15ecfef 1962 spin_lock(&fi->lock);
26d614df 1963 req->num_pages++;
f15ecfef 1964 spin_unlock(&fi->lock);
26d614df 1965
26d614df
PE		 1966out_unlock:
1967 unlock_page(page);
1968
1969 return err;
1970}
1971
1972static int fuse_writepages(struct address_space *mapping,
1973 struct writeback_control *wbc)
1974{
1975 struct inode *inode = mapping->host;
5da784cc 1976 struct fuse_conn *fc = get_fuse_conn(inode);
26d614df
PE		 1977 struct fuse_fill_wb_data data;
1978 int err;
1979
1980 err = -EIO;
1981 if (is_bad_inode(inode))
1982 goto out;
1983
1984 data.inode = inode;
1985 data.req = NULL;
1986 data.ff = NULL;
1987
2d033eaa 1988 err = -ENOMEM;
5da784cc 1989 data.orig_pages = kcalloc(fc->max_pages,
f2b3455e 1990 sizeof(struct page *),
2d033eaa
MP		 1991 GFP_NOFS);
1992 if (!data.orig_pages)
1993 goto out;
1994
26d614df
PE		 1995 err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data);
1996 if (data.req) {
1997 /* Ignore errors if we can write at least one page */
1998 BUG_ON(!data.req->num_pages);
1999 fuse_writepages_send(&data);
2000 err = 0;
2001 }
2002 if (data.ff)
2e64ff15 2003 fuse_file_put(data.ff, false, false);
2d033eaa
MP		 2004
2005 kfree(data.orig_pages);
26d614df
PE		 2006out:
2007 return err;
2008}
2009
6b12c1b3
PE		 2010/*
2011 * It's worthy to make sure that space is reserved on disk for the write,
2012 * but how to implement it without killing performance need more thinking.
2013 */
2014static int fuse_write_begin(struct file *file, struct address_space *mapping,
2015 loff_t pos, unsigned len, unsigned flags,
2016 struct page **pagep, void **fsdata)
2017{
09cbfeaf 2018 pgoff_t index = pos >> PAGE_SHIFT;
a455589f 2019 struct fuse_conn *fc = get_fuse_conn(file_inode(file));
6b12c1b3
PE		 2020 struct page *page;
2021 loff_t fsize;
2022 int err = -ENOMEM;
2023
2024 WARN_ON(!fc->writeback_cache);
2025
2026 page = grab_cache_page_write_begin(mapping, index, flags);
2027 if (!page)
2028 goto error;
2029
2030 fuse_wait_on_page_writeback(mapping->host, page->index);
2031
09cbfeaf 2032 if (PageUptodate(page) || len == PAGE_SIZE)
6b12c1b3
PE		 2033 goto success;
2034 /*
2035 * Check if the start this page comes after the end of file, in which
2036 * case the readpage can be optimized away.
2037 */
2038 fsize = i_size_read(mapping->host);
09cbfeaf
KS		 2039 if (fsize <= (pos & PAGE_MASK)) {
2040 size_t off = pos & ~PAGE_MASK;
6b12c1b3
PE		 2041 if (off)
2042 zero_user_segment(page, 0, off);
2043 goto success;
2044 }
2045 err = fuse_do_readpage(file, page);
2046 if (err)
2047 goto cleanup;
2048success:
2049 *pagep = page;
2050 return 0;
2051
2052cleanup:
2053 unlock_page(page);
09cbfeaf 2054 put_page(page);
6b12c1b3
PE		 2055error:
2056 return err;
2057}
2058
2059static int fuse_write_end(struct file *file, struct address_space *mapping,
2060 loff_t pos, unsigned len, unsigned copied,
2061 struct page *page, void *fsdata)
2062{
2063 struct inode *inode = page->mapping->host;
2064
59c3b76c
MS		 2065 /* Haven't copied anything? Skip zeroing, size extending, dirtying. */
2066 if (!copied)
2067 goto unlock;
2068
6b12c1b3
PE		 2069 if (!PageUptodate(page)) {
2070 /* Zero any unwritten bytes at the end of the page */
09cbfeaf 2071 size_t endoff = (pos + copied) & ~PAGE_MASK;
6b12c1b3 2072 if (endoff)
09cbfeaf 2073 zero_user_segment(page, endoff, PAGE_SIZE);
6b12c1b3
PE		 2074 SetPageUptodate(page);
2075 }
2076
2077 fuse_write_update_size(inode, pos + copied);
2078 set_page_dirty(page);
59c3b76c
MS		 2079
2080unlock:
6b12c1b3 2081 unlock_page(page);
09cbfeaf 2082 put_page(page);
6b12c1b3
PE		 2083
2084 return copied;
2085}
2086
3be5a52b
MS		 2087static int fuse_launder_page(struct page *page)
2088{
2089 int err = 0;
2090 if (clear_page_dirty_for_io(page)) {
2091 struct inode *inode = page->mapping->host;
2092 err = fuse_writepage_locked(page);
2093 if (!err)
2094 fuse_wait_on_page_writeback(inode, page->index);
2095 }
2096 return err;
2097}
2098
2099/*
2100 * Write back dirty pages now, because there may not be any suitable
2101 * open files later
2102 */
2103static void fuse_vma_close(struct vm_area_struct *vma)
2104{
2105 filemap_write_and_wait(vma->vm_file->f_mapping);
2106}
2107
2108/*
2109 * Wait for writeback against this page to complete before allowing it
2110 * to be marked dirty again, and hence written back again, possibly
2111 * before the previous writepage completed.
2112 *
2113 * Block here, instead of in ->writepage(), so that the userspace fs
2114 * can only block processes actually operating on the filesystem.
2115 *
2116 * Otherwise unprivileged userspace fs would be able to block
2117 * unrelated:
2118 *
2119 * - page migration
2120 * - sync(2)
2121 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
2122 */
46fb504a 2123static vm_fault_t fuse_page_mkwrite(struct vm_fault *vmf)
3be5a52b 2124{
c2ec175c 2125 struct page *page = vmf->page;
11bac800 2126 struct inode *inode = file_inode(vmf->vma->vm_file);
cca24370 2127
11bac800 2128 file_update_time(vmf->vma->vm_file);
cca24370
MS		 2129 lock_page(page);
2130 if (page->mapping != inode->i_mapping) {
2131 unlock_page(page);
2132 return VM_FAULT_NOPAGE;
2133 }
3be5a52b
MS		 2134
2135 fuse_wait_on_page_writeback(inode, page->index);
cca24370 2136 return VM_FAULT_LOCKED;
3be5a52b
MS		 2137}
2138
f0f37e2f 2139static const struct vm_operations_struct fuse_file_vm_ops = {
3be5a52b
MS		 2140 .close = fuse_vma_close,
2141 .fault = filemap_fault,
f1820361 2142 .map_pages = filemap_map_pages,
3be5a52b
MS		 2143 .page_mkwrite = fuse_page_mkwrite,
2144};
2145
2146static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
2147{
55752a3a
MS		 2148 struct fuse_file *ff = file->private_data;
2149
2150 if (ff->open_flags & FOPEN_DIRECT_IO) {
2151 /* Can't provide the coherency needed for MAP_SHARED */
2152 if (vma->vm_flags & VM_MAYSHARE)
2153 return -ENODEV;
2154
2155 invalidate_inode_pages2(file->f_mapping);
2156
2157 return generic_file_mmap(file, vma);
2158 }
2159
650b22b9
PE		 2160 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2161 fuse_link_write_file(file);
2162
3be5a52b
MS		 2163 file_accessed(file);
2164 vma->vm_ops = &fuse_file_vm_ops;
b6aeaded
MS		 2165 return 0;
2166}
2167
0b6e9ea0
SF		 2168static int convert_fuse_file_lock(struct fuse_conn *fc,
2169 const struct fuse_file_lock *ffl,
71421259
MS		 2170 struct file_lock *fl)
2171{
2172 switch (ffl->type) {
2173 case F_UNLCK:
2174 break;
2175
2176 case F_RDLCK:
2177 case F_WRLCK:
2178 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
2179 ffl->end < ffl->start)
2180 return -EIO;
2181
2182 fl->fl_start = ffl->start;
2183 fl->fl_end = ffl->end;
0b6e9ea0
SF		 2184
2185 /*
9d5b86ac
BC		 2186 * Convert pid into init's pid namespace. The locks API will
2187 * translate it into the caller's pid namespace.
0b6e9ea0
SF		 2188 */
2189 rcu_read_lock();
9d5b86ac 2190 fl->fl_pid = pid_nr_ns(find_pid_ns(ffl->pid, fc->pid_ns), &init_pid_ns);
0b6e9ea0 2191 rcu_read_unlock();
71421259
MS		 2192 break;
2193
2194 default:
2195 return -EIO;
2196 }
2197 fl->fl_type = ffl->type;
2198 return 0;
2199}
2200
7078187a 2201static void fuse_lk_fill(struct fuse_args *args, struct file *file,
a9ff4f87 2202 const struct file_lock *fl, int opcode, pid_t pid,
7078187a 2203 int flock, struct fuse_lk_in *inarg)
71421259 2204{
6131ffaa 2205 struct inode *inode = file_inode(file);
9c8ef561 2206 struct fuse_conn *fc = get_fuse_conn(inode);
71421259 2207 struct fuse_file *ff = file->private_data;
7078187a
MS		 2208
2209 memset(inarg, 0, sizeof(*inarg));
2210 inarg->fh = ff->fh;
2211 inarg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
2212 inarg->lk.start = fl->fl_start;
2213 inarg->lk.end = fl->fl_end;
2214 inarg->lk.type = fl->fl_type;
2215 inarg->lk.pid = pid;
a9ff4f87 2216 if (flock)
7078187a
MS		 2217 inarg->lk_flags |= FUSE_LK_FLOCK;
2218 args->in.h.opcode = opcode;
2219 args->in.h.nodeid = get_node_id(inode);
2220 args->in.numargs = 1;
2221 args->in.args[0].size = sizeof(*inarg);
2222 args->in.args[0].value = inarg;
71421259
MS		 2223}
2224
2225static int fuse_getlk(struct file *file, struct file_lock *fl)
2226{
6131ffaa 2227 struct inode *inode = file_inode(file);
71421259 2228 struct fuse_conn *fc = get_fuse_conn(inode);
7078187a
MS		 2229 FUSE_ARGS(args);
2230 struct fuse_lk_in inarg;
71421259
MS		 2231 struct fuse_lk_out outarg;
2232 int err;
2233
7078187a
MS		 2234 fuse_lk_fill(&args, file, fl, FUSE_GETLK, 0, 0, &inarg);
2235 args.out.numargs = 1;
2236 args.out.args[0].size = sizeof(outarg);
2237 args.out.args[0].value = &outarg;
2238 err = fuse_simple_request(fc, &args);
71421259 2239 if (!err)
0b6e9ea0 2240 err = convert_fuse_file_lock(fc, &outarg.lk, fl);
71421259
MS		 2241
2242 return err;
2243}
2244
a9ff4f87 2245static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
71421259 2246{
6131ffaa 2247 struct inode *inode = file_inode(file);
71421259 2248 struct fuse_conn *fc = get_fuse_conn(inode);
7078187a
MS		 2249 FUSE_ARGS(args);
2250 struct fuse_lk_in inarg;
71421259 2251 int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
0b6e9ea0
SF		 2252 struct pid *pid = fl->fl_type != F_UNLCK ? task_tgid(current) : NULL;
2253 pid_t pid_nr = pid_nr_ns(pid, fc->pid_ns);
71421259
MS		 2254 int err;
2255
8fb47a4f 2256 if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
48e90761
MS		 2257 /* NLM needs asynchronous locks, which we don't support yet */
2258 return -ENOLCK;
2259 }
2260
71421259 2261 /* Unlock on close is handled by the flush method */
50f2112c 2262 if ((fl->fl_flags & FL_CLOSE_POSIX) == FL_CLOSE_POSIX)
71421259
MS		 2263 return 0;
2264
0b6e9ea0 2265 fuse_lk_fill(&args, file, fl, opcode, pid_nr, flock, &inarg);
7078187a 2266 err = fuse_simple_request(fc, &args);
71421259 2267
a4d27e75
MS		 2268 /* locking is restartable */
2269 if (err == -EINTR)
2270 err = -ERESTARTSYS;
7078187a 2271
71421259
MS		 2272 return err;
2273}
2274
2275static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
2276{
6131ffaa 2277 struct inode *inode = file_inode(file);
71421259
MS		 2278 struct fuse_conn *fc = get_fuse_conn(inode);
2279 int err;
2280
48e90761
MS		 2281 if (cmd == F_CANCELLK) {
2282 err = 0;
2283 } else if (cmd == F_GETLK) {
71421259 2284 if (fc->no_lock) {
9d6a8c5c 2285 posix_test_lock(file, fl);
71421259
MS		 2286 err = 0;
2287 } else
2288 err = fuse_getlk(file, fl);
2289 } else {
2290 if (fc->no_lock)
48e90761 2291 err = posix_lock_file(file, fl, NULL);
71421259 2292 else
a9ff4f87 2293 err = fuse_setlk(file, fl, 0);
71421259
MS		 2294 }
2295 return err;
2296}
2297
a9ff4f87
MS		 2298static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
2299{
6131ffaa 2300 struct inode *inode = file_inode(file);
a9ff4f87
MS		 2301 struct fuse_conn *fc = get_fuse_conn(inode);
2302 int err;
2303
37fb3a30 2304 if (fc->no_flock) {
4f656367 2305 err = locks_lock_file_wait(file, fl);
a9ff4f87 2306 } else {
37fb3a30
MS		 2307 struct fuse_file *ff = file->private_data;
2308
a9ff4f87 2309 /* emulate flock with POSIX locks */
37fb3a30 2310 ff->flock = true;
a9ff4f87
MS		 2311 err = fuse_setlk(file, fl, 1);
2312 }
2313
2314 return err;
2315}
2316
b2d2272f
MS		 2317static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
2318{
2319 struct inode *inode = mapping->host;
2320 struct fuse_conn *fc = get_fuse_conn(inode);
7078187a 2321 FUSE_ARGS(args);
b2d2272f
MS		 2322 struct fuse_bmap_in inarg;
2323 struct fuse_bmap_out outarg;
2324 int err;
2325
2326 if (!inode->i_sb->s_bdev || fc->no_bmap)
2327 return 0;
2328
b2d2272f
MS		 2329 memset(&inarg, 0, sizeof(inarg));
2330 inarg.block = block;
2331 inarg.blocksize = inode->i_sb->s_blocksize;
7078187a
MS		 2332 args.in.h.opcode = FUSE_BMAP;
2333 args.in.h.nodeid = get_node_id(inode);
2334 args.in.numargs = 1;
2335 args.in.args[0].size = sizeof(inarg);
2336 args.in.args[0].value = &inarg;
2337 args.out.numargs = 1;
2338 args.out.args[0].size = sizeof(outarg);
2339 args.out.args[0].value = &outarg;
2340 err = fuse_simple_request(fc, &args);
b2d2272f
MS		 2341 if (err == -ENOSYS)
2342 fc->no_bmap = 1;
2343
2344 return err ? 0 : outarg.block;
2345}
2346
0b5da8db
R		 2347static loff_t fuse_lseek(struct file *file, loff_t offset, int whence)
2348{
2349 struct inode *inode = file->f_mapping->host;
2350 struct fuse_conn *fc = get_fuse_conn(inode);
2351 struct fuse_file *ff = file->private_data;
2352 FUSE_ARGS(args);
2353 struct fuse_lseek_in inarg = {
2354 .fh = ff->fh,
2355 .offset = offset,
2356 .whence = whence
2357 };
2358 struct fuse_lseek_out outarg;
2359 int err;
2360
2361 if (fc->no_lseek)
2362 goto fallback;
2363
2364 args.in.h.opcode = FUSE_LSEEK;
2365 args.in.h.nodeid = ff->nodeid;
2366 args.in.numargs = 1;
2367 args.in.args[0].size = sizeof(inarg);
2368 args.in.args[0].value = &inarg;
2369 args.out.numargs = 1;
2370 args.out.args[0].size = sizeof(outarg);
2371 args.out.args[0].value = &outarg;
2372 err = fuse_simple_request(fc, &args);
2373 if (err) {
2374 if (err == -ENOSYS) {
2375 fc->no_lseek = 1;
2376 goto fallback;
2377 }
2378 return err;
2379 }
2380
2381 return vfs_setpos(file, outarg.offset, inode->i_sb->s_maxbytes);
2382
2383fallback:
5b97eeac 2384 err = fuse_update_attributes(inode, file);
0b5da8db
R		 2385 if (!err)
2386 return generic_file_llseek(file, offset, whence);
2387 else
2388 return err;
2389}
2390
965c8e59 2391static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence)
5559b8f4
MS		 2392{
2393 loff_t retval;
6131ffaa 2394 struct inode *inode = file_inode(file);
5559b8f4 2395
0b5da8db
R		 2396 switch (whence) {
2397 case SEEK_SET:
2398 case SEEK_CUR:
2399 /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */
965c8e59 2400 retval = generic_file_llseek(file, offset, whence);
0b5da8db
R		 2401 break;
2402 case SEEK_END:
5955102c 2403 inode_lock(inode);
5b97eeac 2404 retval = fuse_update_attributes(inode, file);
0b5da8db
R		 2405 if (!retval)
2406 retval = generic_file_llseek(file, offset, whence);
5955102c 2407 inode_unlock(inode);
0b5da8db
R		 2408 break;
2409 case SEEK_HOLE:
2410 case SEEK_DATA:
5955102c 2411 inode_lock(inode);
0b5da8db 2412 retval = fuse_lseek(file, offset, whence);
5955102c 2413 inode_unlock(inode);
0b5da8db
R		 2414 break;
2415 default:
2416 retval = -EINVAL;
2417 }
c07c3d19 2418
5559b8f4
MS		 2419 return retval;
2420}
2421
d9d318d3
MS		 2422/*
2423 * CUSE servers compiled on 32bit broke on 64bit kernels because the
2424 * ABI was defined to be 'struct iovec' which is different on 32bit
2425 * and 64bit. Fortunately we can determine which structure the server
2426 * used from the size of the reply.
2427 */
1baa26b2
MS		 2428static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
2429 size_t transferred, unsigned count,
2430 bool is_compat)
d9d318d3
MS		 2431{
2432#ifdef CONFIG_COMPAT
2433 if (count * sizeof(struct compat_iovec) == transferred) {
2434 struct compat_iovec *ciov = src;
2435 unsigned i;
2436
2437 /*
2438 * With this interface a 32bit server cannot support
2439 * non-compat (i.e. ones coming from 64bit apps) ioctl
2440 * requests
2441 */
2442 if (!is_compat)
2443 return -EINVAL;
2444
2445 for (i = 0; i < count; i++) {
2446 dst[i].iov_base = compat_ptr(ciov[i].iov_base);
2447 dst[i].iov_len = ciov[i].iov_len;
2448 }
2449 return 0;
2450 }
2451#endif
2452
2453 if (count * sizeof(struct iovec) != transferred)
2454 return -EIO;
2455
2456 memcpy(dst, src, transferred);
2457 return 0;
2458}
2459
7572777e 2460/* Make sure iov_length() won't overflow */
5da784cc
CS		 2461static int fuse_verify_ioctl_iov(struct fuse_conn *fc, struct iovec *iov,
2462 size_t count)
7572777e
MS		 2463{
2464 size_t n;
5da784cc 2465 u32 max = fc->max_pages << PAGE_SHIFT;
7572777e 2466
fb6ccff6 2467 for (n = 0; n < count; n++, iov++) {
7572777e
MS		 2468 if (iov->iov_len > (size_t) max)
2469 return -ENOMEM;
2470 max -= iov->iov_len;
2471 }
2472 return 0;
2473}
2474
1baa26b2
MS		 2475static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
2476 void *src, size_t transferred, unsigned count,
2477 bool is_compat)
2478{
2479 unsigned i;
2480 struct fuse_ioctl_iovec *fiov = src;
2481
2482 if (fc->minor < 16) {
2483 return fuse_copy_ioctl_iovec_old(dst, src, transferred,
2484 count, is_compat);
2485 }
2486
2487 if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
2488 return -EIO;
2489
2490 for (i = 0; i < count; i++) {
2491 /* Did the server supply an inappropriate value? */
2492 if (fiov[i].base != (unsigned long) fiov[i].base ||
2493 fiov[i].len != (unsigned long) fiov[i].len)
2494 return -EIO;
2495
2496 dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
2497 dst[i].iov_len = (size_t) fiov[i].len;
2498
2499#ifdef CONFIG_COMPAT
2500 if (is_compat &&
2501 (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
2502 (compat_size_t) dst[i].iov_len != fiov[i].len))
2503 return -EIO;
2504#endif
2505 }
2506
2507 return 0;
2508}
2509
2510
59efec7b
TH		 2511/*
2512 * For ioctls, there is no generic way to determine how much memory
2513 * needs to be read and/or written. Furthermore, ioctls are allowed
2514 * to dereference the passed pointer, so the parameter requires deep
2515 * copying but FUSE has no idea whatsoever about what to copy in or
2516 * out.
2517 *
2518 * This is solved by allowing FUSE server to retry ioctl with
2519 * necessary in/out iovecs. Let's assume the ioctl implementation
2520 * needs to read in the following structure.
2521 *
2522 * struct a {
2523 * char *buf;
2524 * size_t buflen;
2525 * }
2526 *
2527 * On the first callout to FUSE server, inarg->in_size and
2528 * inarg->out_size will be NULL; then, the server completes the ioctl
2529 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
2530 * the actual iov array to
2531 *
2532 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } }
2533 *
2534 * which tells FUSE to copy in the requested area and retry the ioctl.
2535 * On the second round, the server has access to the structure and
2536 * from that it can tell what to look for next, so on the invocation,
2537 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
2538 *
2539 * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) },
2540 * { .iov_base = a.buf, .iov_len = a.buflen } }
2541 *
2542 * FUSE will copy both struct a and the pointed buffer from the
2543 * process doing the ioctl and retry ioctl with both struct a and the
2544 * buffer.
2545 *
2546 * This time, FUSE server has everything it needs and completes ioctl
2547 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
2548 *
2549 * Copying data out works the same way.
2550 *
2551 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
2552 * automatically initializes in and out iovs by decoding @cmd with
2553 * _IOC_* macros and the server is not allowed to request RETRY. This
2554 * limits ioctl data transfers to well-formed ioctls and is the forced
2555 * behavior for all FUSE servers.
2556 */
08cbf542
TH		 2557long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
2558 unsigned int flags)
59efec7b 2559{
59efec7b 2560 struct fuse_file *ff = file->private_data;
d36f2487 2561 struct fuse_conn *fc = ff->fc;
59efec7b
TH		 2562 struct fuse_ioctl_in inarg = {
2563 .fh = ff->fh,
2564 .cmd = cmd,
2565 .arg = arg,
2566 .flags = flags
2567 };
2568 struct fuse_ioctl_out outarg;
2569 struct fuse_req *req = NULL;
2570 struct page **pages = NULL;
8ac83505 2571 struct iovec *iov_page = NULL;
59efec7b
TH		 2572 struct iovec *in_iov = NULL, *out_iov = NULL;
2573 unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages;
acbe5fda
MS		 2574 size_t in_size, out_size, transferred, c;
2575 int err, i;
2576 struct iov_iter ii;
59efec7b 2577
1baa26b2
MS		 2578#if BITS_PER_LONG == 32
2579 inarg.flags |= FUSE_IOCTL_32BIT;
2580#else
2581 if (flags & FUSE_IOCTL_COMPAT)
2582 inarg.flags |= FUSE_IOCTL_32BIT;
2583#endif
2584
59efec7b 2585 /* assume all the iovs returned by client always fits in a page */
1baa26b2 2586 BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
59efec7b 2587
59efec7b 2588 err = -ENOMEM;
5da784cc 2589 pages = kcalloc(fc->max_pages, sizeof(pages[0]), GFP_KERNEL);
8ac83505 2590 iov_page = (struct iovec *) __get_free_page(GFP_KERNEL);
59efec7b
TH		 2591 if (!pages || !iov_page)
2592 goto out;
2593
2594 /*
2595 * If restricted, initialize IO parameters as encoded in @cmd.
2596 * RETRY from server is not allowed.
2597 */
2598 if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
8ac83505 2599 struct iovec *iov = iov_page;
59efec7b 2600
c9f0523d 2601 iov->iov_base = (void __user *)arg;
59efec7b
TH		 2602 iov->iov_len = _IOC_SIZE(cmd);
2603
2604 if (_IOC_DIR(cmd) & _IOC_WRITE) {
2605 in_iov = iov;
2606 in_iovs = 1;
2607 }
2608
2609 if (_IOC_DIR(cmd) & _IOC_READ) {
2610 out_iov = iov;
2611 out_iovs = 1;
2612 }
2613 }
2614
2615 retry:
2616 inarg.in_size = in_size = iov_length(in_iov, in_iovs);
2617 inarg.out_size = out_size = iov_length(out_iov, out_iovs);
2618
2619 /*
2620 * Out data can be used either for actual out data or iovs,
2621 * make sure there always is at least one page.
2622 */
2623 out_size = max_t(size_t, out_size, PAGE_SIZE);
2624 max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
2625
2626 /* make sure there are enough buffer pages and init request with them */
2627 err = -ENOMEM;
5da784cc 2628 if (max_pages > fc->max_pages)
59efec7b
TH		 2629 goto out;
2630 while (num_pages < max_pages) {
2631 pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2632 if (!pages[num_pages])
2633 goto out;
2634 num_pages++;
2635 }
2636
54b96670 2637 req = fuse_get_req(fc, num_pages);
59efec7b
TH		 2638 if (IS_ERR(req)) {
2639 err = PTR_ERR(req);
2640 req = NULL;
2641 goto out;
2642 }
2643 memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages);
2644 req->num_pages = num_pages;
7c190c8b 2645 fuse_page_descs_length_init(req, 0, req->num_pages);
59efec7b
TH		 2646
2647 /* okay, let's send it to the client */
2648 req->in.h.opcode = FUSE_IOCTL;
d36f2487 2649 req->in.h.nodeid = ff->nodeid;
59efec7b
TH		 2650 req->in.numargs = 1;
2651 req->in.args[0].size = sizeof(inarg);
2652 req->in.args[0].value = &inarg;
2653 if (in_size) {
2654 req->in.numargs++;
2655 req->in.args[1].size = in_size;
2656 req->in.argpages = 1;
2657
acbe5fda
MS		 2658 err = -EFAULT;
2659 iov_iter_init(&ii, WRITE, in_iov, in_iovs, in_size);
2660 for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= num_pages); i++) {
2661 c = copy_page_from_iter(pages[i], 0, PAGE_SIZE, &ii);
2662 if (c != PAGE_SIZE && iov_iter_count(&ii))
2663 goto out;
2664 }
59efec7b
TH		 2665 }
2666
2667 req->out.numargs = 2;
2668 req->out.args[0].size = sizeof(outarg);
2669 req->out.args[0].value = &outarg;
2670 req->out.args[1].size = out_size;
2671 req->out.argpages = 1;
2672 req->out.argvar = 1;
2673
b93f858a 2674 fuse_request_send(fc, req);
59efec7b
TH		 2675 err = req->out.h.error;
2676 transferred = req->out.args[1].size;
2677 fuse_put_request(fc, req);
2678 req = NULL;
2679 if (err)
2680 goto out;
2681
2682 /* did it ask for retry? */
2683 if (outarg.flags & FUSE_IOCTL_RETRY) {
8ac83505 2684 void *vaddr;
59efec7b
TH		 2685
2686 /* no retry if in restricted mode */
2687 err = -EIO;
2688 if (!(flags & FUSE_IOCTL_UNRESTRICTED))
2689 goto out;
2690
2691 in_iovs = outarg.in_iovs;
2692 out_iovs = outarg.out_iovs;
2693
2694 /*
2695 * Make sure things are in boundary, separate checks
2696 * are to protect against overflow.
2697 */
2698 err = -ENOMEM;
2699 if (in_iovs > FUSE_IOCTL_MAX_IOV ||
2700 out_iovs > FUSE_IOCTL_MAX_IOV ||
2701 in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
2702 goto out;
2703
2408f6ef 2704 vaddr = kmap_atomic(pages[0]);
1baa26b2 2705 err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr,
d9d318d3
MS		 2706 transferred, in_iovs + out_iovs,
2707 (flags & FUSE_IOCTL_COMPAT) != 0);
2408f6ef 2708 kunmap_atomic(vaddr);
d9d318d3
MS		 2709 if (err)
2710 goto out;
59efec7b 2711
8ac83505 2712 in_iov = iov_page;
59efec7b
TH		 2713 out_iov = in_iov + in_iovs;
2714
5da784cc 2715 err = fuse_verify_ioctl_iov(fc, in_iov, in_iovs);
7572777e
MS		 2716 if (err)
2717 goto out;
2718
5da784cc 2719 err = fuse_verify_ioctl_iov(fc, out_iov, out_iovs);
7572777e
MS		 2720 if (err)
2721 goto out;
2722
59efec7b
TH		 2723 goto retry;
2724 }
2725
2726 err = -EIO;
2727 if (transferred > inarg.out_size)
2728 goto out;
2729
acbe5fda
MS		 2730 err = -EFAULT;
2731 iov_iter_init(&ii, READ, out_iov, out_iovs, transferred);
2732 for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= num_pages); i++) {
2733 c = copy_page_to_iter(pages[i], 0, PAGE_SIZE, &ii);
2734 if (c != PAGE_SIZE && iov_iter_count(&ii))
2735 goto out;
2736 }
2737 err = 0;
59efec7b
TH		 2738 out:
2739 if (req)
2740 fuse_put_request(fc, req);
8ac83505 2741 free_page((unsigned long) iov_page);
59efec7b
TH		 2742 while (num_pages)
2743 __free_page(pages[--num_pages]);
2744 kfree(pages);
2745
2746 return err ? err : outarg.result;
2747}
08cbf542 2748EXPORT_SYMBOL_GPL(fuse_do_ioctl);
59efec7b 2749
b18da0c5
MS		 2750long fuse_ioctl_common(struct file *file, unsigned int cmd,
2751 unsigned long arg, unsigned int flags)
d36f2487 2752{
6131ffaa 2753 struct inode *inode = file_inode(file);
d36f2487
MS		 2754 struct fuse_conn *fc = get_fuse_conn(inode);
2755
c2132c1b 2756 if (!fuse_allow_current_process(fc))
d36f2487
MS		 2757 return -EACCES;
2758
2759 if (is_bad_inode(inode))
2760 return -EIO;
2761
2762 return fuse_do_ioctl(file, cmd, arg, flags);
2763}
2764
59efec7b
TH		 2765static long fuse_file_ioctl(struct file *file, unsigned int cmd,
2766 unsigned long arg)
2767{
b18da0c5 2768 return fuse_ioctl_common(file, cmd, arg, 0);
59efec7b
TH		 2769}
2770
2771static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
2772 unsigned long arg)
2773{
b18da0c5 2774 return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
59efec7b
TH		 2775}
2776
95668a69
TH		 2777/*
2778 * All files which have been polled are linked to RB tree
2779 * fuse_conn->polled_files which is indexed by kh. Walk the tree and
2780 * find the matching one.
2781 */
2782static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh,
2783 struct rb_node **parent_out)
2784{
2785 struct rb_node **link = &fc->polled_files.rb_node;
2786 struct rb_node *last = NULL;
2787
2788 while (*link) {
2789 struct fuse_file *ff;
2790
2791 last = *link;
2792 ff = rb_entry(last, struct fuse_file, polled_node);
2793
2794 if (kh < ff->kh)
2795 link = &last->rb_left;
2796 else if (kh > ff->kh)
2797 link = &last->rb_right;
2798 else
2799 return link;
2800 }
2801
2802 if (parent_out)
2803 *parent_out = last;
2804 return link;
2805}
2806
2807/*
2808 * The file is about to be polled. Make sure it's on the polled_files
2809 * RB tree. Note that files once added to the polled_files tree are
2810 * not removed before the file is released. This is because a file
2811 * polled once is likely to be polled again.
2812 */
2813static void fuse_register_polled_file(struct fuse_conn *fc,
2814 struct fuse_file *ff)
2815{
2816 spin_lock(&fc->lock);
2817 if (RB_EMPTY_NODE(&ff->polled_node)) {
f3846266 2818 struct rb_node **link, *uninitialized_var(parent);
95668a69
TH		 2819
2820 link = fuse_find_polled_node(fc, ff->kh, &parent);
2821 BUG_ON(*link);
2822 rb_link_node(&ff->polled_node, parent, link);
2823 rb_insert_color(&ff->polled_node, &fc->polled_files);
2824 }
2825 spin_unlock(&fc->lock);
2826}
2827
076ccb76 2828__poll_t fuse_file_poll(struct file *file, poll_table *wait)
95668a69 2829{
95668a69 2830 struct fuse_file *ff = file->private_data;
797759aa 2831 struct fuse_conn *fc = ff->fc;
95668a69
TH		 2832 struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh };
2833 struct fuse_poll_out outarg;
7078187a 2834 FUSE_ARGS(args);
95668a69
TH		 2835 int err;
2836
2837 if (fc->no_poll)
2838 return DEFAULT_POLLMASK;
2839
2840 poll_wait(file, &ff->poll_wait, wait);
c71d227f 2841 inarg.events = mangle_poll(poll_requested_events(wait));
95668a69
TH		 2842
2843 /*
2844 * Ask for notification iff there's someone waiting for it.
2845 * The client may ignore the flag and always notify.
2846 */
2847 if (waitqueue_active(&ff->poll_wait)) {
2848 inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY;
2849 fuse_register_polled_file(fc, ff);
2850 }
2851
7078187a
MS		 2852 args.in.h.opcode = FUSE_POLL;
2853 args.in.h.nodeid = ff->nodeid;
2854 args.in.numargs = 1;
2855 args.in.args[0].size = sizeof(inarg);
2856 args.in.args[0].value = &inarg;
2857 args.out.numargs = 1;
2858 args.out.args[0].size = sizeof(outarg);
2859 args.out.args[0].value = &outarg;
2860 err = fuse_simple_request(fc, &args);
95668a69
TH		 2861
2862 if (!err)
c71d227f 2863 return demangle_poll(outarg.revents);
95668a69
TH		 2864 if (err == -ENOSYS) {
2865 fc->no_poll = 1;
2866 return DEFAULT_POLLMASK;
2867 }
a9a08845 2868 return EPOLLERR;
95668a69 2869}
08cbf542 2870EXPORT_SYMBOL_GPL(fuse_file_poll);
95668a69
TH		 2871
2872/*
2873 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
2874 * wakes up the poll waiters.
2875 */
2876int fuse_notify_poll_wakeup(struct fuse_conn *fc,
2877 struct fuse_notify_poll_wakeup_out *outarg)
2878{
2879 u64 kh = outarg->kh;
2880 struct rb_node **link;
2881
2882 spin_lock(&fc->lock);
2883
2884 link = fuse_find_polled_node(fc, kh, NULL);
2885 if (*link) {
2886 struct fuse_file *ff;
2887
2888 ff = rb_entry(*link, struct fuse_file, polled_node);
2889 wake_up_interruptible_sync(&ff->poll_wait);
2890 }
2891
2892 spin_unlock(&fc->lock);
2893 return 0;
2894}
2895
efb9fa9e
MP		 2896static void fuse_do_truncate(struct file *file)
2897{
2898 struct inode *inode = file->f_mapping->host;
2899 struct iattr attr;
2900
2901 attr.ia_valid = ATTR_SIZE;
2902 attr.ia_size = i_size_read(inode);
2903
2904 attr.ia_file = file;
2905 attr.ia_valid |= ATTR_FILE;
2906
62490330 2907 fuse_do_setattr(file_dentry(file), &attr, file);
efb9fa9e
MP		 2908}
2909
5da784cc 2910static inline loff_t fuse_round_up(struct fuse_conn *fc, loff_t off)
e5c5f05d 2911{
5da784cc 2912 return round_up(off, fc->max_pages << PAGE_SHIFT);
e5c5f05d
MP		 2913}
2914
4273b793 2915static ssize_t
c8b8e32d 2916fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
4273b793 2917{
9d5722b7 2918 DECLARE_COMPLETION_ONSTACK(wait);
4273b793 2919 ssize_t ret = 0;
60b9df7a
MS		 2920 struct file *file = iocb->ki_filp;
2921 struct fuse_file *ff = file->private_data;
e5c5f05d 2922 bool async_dio = ff->fc->async_dio;
4273b793 2923 loff_t pos = 0;
bcba24cc
MP		 2924 struct inode *inode;
2925 loff_t i_size;
a6cbcd4a 2926 size_t count = iov_iter_count(iter);
c8b8e32d 2927 loff_t offset = iocb->ki_pos;
36cf66ed 2928 struct fuse_io_priv *io;
4273b793 2929
4273b793 2930 pos = offset;
bcba24cc
MP		 2931 inode = file->f_mapping->host;
2932 i_size = i_size_read(inode);
4273b793 2933
6f673763 2934 if ((iov_iter_rw(iter) == READ) && (offset > i_size))
9fe55eea
SW		 2935 return 0;
2936
439ee5f0 2937 /* optimization for short read */
6f673763 2938 if (async_dio && iov_iter_rw(iter) != WRITE && offset + count > i_size) {
439ee5f0
MP		 2939 if (offset >= i_size)
2940 return 0;
5da784cc 2941 iov_iter_truncate(iter, fuse_round_up(ff->fc, i_size - offset));
6b775b18 2942 count = iov_iter_count(iter);
439ee5f0
MP		 2943 }
2944
bcba24cc 2945 io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL);
36cf66ed
MP		 2946 if (!io)
2947 return -ENOMEM;
bcba24cc 2948 spin_lock_init(&io->lock);
744742d6 2949 kref_init(&io->refcnt);
bcba24cc
MP		 2950 io->reqs = 1;
2951 io->bytes = -1;
2952 io->size = 0;
2953 io->offset = offset;
6f673763 2954 io->write = (iov_iter_rw(iter) == WRITE);
bcba24cc 2955 io->err = 0;
bcba24cc
MP		 2956 /*
2957 * By default, we want to optimize all I/Os with async request
60b9df7a 2958 * submission to the client filesystem if supported.
bcba24cc 2959 */
e5c5f05d 2960 io->async = async_dio;
bcba24cc 2961 io->iocb = iocb;
7879c4e5 2962 io->blocking = is_sync_kiocb(iocb);
bcba24cc
MP		 2963
2964 /*
7879c4e5
AS		 2965 * We cannot asynchronously extend the size of a file.
2966 * In such case the aio will behave exactly like sync io.
bcba24cc 2967 */
7879c4e5
AS		 2968 if ((offset + count > i_size) && iov_iter_rw(iter) == WRITE)
2969 io->blocking = true;
4273b793 2970
7879c4e5 2971 if (io->async && io->blocking) {
744742d6
SF		 2972 /*
2973 * Additional reference to keep io around after
2974 * calling fuse_aio_complete()
2975 */
2976 kref_get(&io->refcnt);
9d5722b7 2977 io->done = &wait;
744742d6 2978 }
9d5722b7 2979
6f673763 2980 if (iov_iter_rw(iter) == WRITE) {
6b775b18 2981 ret = fuse_direct_io(io, iter, &pos, FUSE_DIO_WRITE);
812408fb
AV		 2982 fuse_invalidate_attr(inode);
2983 } else {
d22a943f 2984 ret = __fuse_direct_read(io, iter, &pos);
812408fb 2985 }
36cf66ed 2986
bcba24cc 2987 if (io->async) {
ebacb812
LC		 2988 bool blocking = io->blocking;
2989
bcba24cc
MP		 2990 fuse_aio_complete(io, ret < 0 ? ret : 0, -1);
2991
2992 /* we have a non-extending, async request, so return */
ebacb812 2993 if (!blocking)
bcba24cc
MP		 2994 return -EIOCBQUEUED;
2995
9d5722b7
CH		 2996 wait_for_completion(&wait);
2997 ret = fuse_get_res_by_io(io);
bcba24cc
MP		 2998 }
2999
744742d6 3000 kref_put(&io->refcnt, fuse_io_release);
9d5722b7 3001
6f673763 3002 if (iov_iter_rw(iter) == WRITE) {
efb9fa9e
MP		 3003 if (ret > 0)
3004 fuse_write_update_size(inode, pos);
3005 else if (ret < 0 && offset + count > i_size)
3006 fuse_do_truncate(file);
3007 }
4273b793
AA		 3008
3009 return ret;
3010}
3011
cdadb11c
MS		 3012static long fuse_file_fallocate(struct file *file, int mode, loff_t offset,
3013 loff_t length)
05ba1f08
AP		 3014{
3015 struct fuse_file *ff = file->private_data;
1c68271c 3016 struct inode *inode = file_inode(file);
0ab08f57 3017 struct fuse_inode *fi = get_fuse_inode(inode);
05ba1f08 3018 struct fuse_conn *fc = ff->fc;
7078187a 3019 FUSE_ARGS(args);
05ba1f08
AP		 3020 struct fuse_fallocate_in inarg = {
3021 .fh = ff->fh,
3022 .offset = offset,
3023 .length = length,
3024 .mode = mode
3025 };
3026 int err;
14c14414
MP		 3027 bool lock_inode = !(mode & FALLOC_FL_KEEP_SIZE) ||
3028 (mode & FALLOC_FL_PUNCH_HOLE);
05ba1f08 3029
4adb8302
MS		 3030 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
3031 return -EOPNOTSUPP;
3032
519c6040
MS		 3033 if (fc->no_fallocate)
3034 return -EOPNOTSUPP;
3035
14c14414 3036 if (lock_inode) {
5955102c 3037 inode_lock(inode);
bde52788
MP		 3038 if (mode & FALLOC_FL_PUNCH_HOLE) {
3039 loff_t endbyte = offset + length - 1;
3040 err = filemap_write_and_wait_range(inode->i_mapping,
3041 offset, endbyte);
3042 if (err)
3043 goto out;
3044
3045 fuse_sync_writes(inode);
3046 }
3634a632
BF		 3047 }
3048
0ab08f57
MP		 3049 if (!(mode & FALLOC_FL_KEEP_SIZE))
3050 set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
3051
7078187a
MS		 3052 args.in.h.opcode = FUSE_FALLOCATE;
3053 args.in.h.nodeid = ff->nodeid;
3054 args.in.numargs = 1;
3055 args.in.args[0].size = sizeof(inarg);
3056 args.in.args[0].value = &inarg;
3057 err = fuse_simple_request(fc, &args);
519c6040
MS		 3058 if (err == -ENOSYS) {
3059 fc->no_fallocate = 1;
3060 err = -EOPNOTSUPP;
3061 }
bee6c307
BF		 3062 if (err)
3063 goto out;
3064
3065 /* we could have extended the file */
b0aa7606
MP		 3066 if (!(mode & FALLOC_FL_KEEP_SIZE)) {
3067 bool changed = fuse_write_update_size(inode, offset + length);
3068
93d2269d
MS		 3069 if (changed && fc->writeback_cache)
3070 file_update_time(file);
b0aa7606 3071 }
bee6c307
BF		 3072
3073 if (mode & FALLOC_FL_PUNCH_HOLE)
3074 truncate_pagecache_range(inode, offset, offset + length - 1);
3075
3076 fuse_invalidate_attr(inode);
3077
3634a632 3078out:
0ab08f57
MP		 3079 if (!(mode & FALLOC_FL_KEEP_SIZE))
3080 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
3081
bde52788 3082 if (lock_inode)
5955102c 3083 inode_unlock(inode);
3634a632 3084
05ba1f08
AP		 3085 return err;
3086}
05ba1f08 3087
88bc7d50
NV		 3088static ssize_t fuse_copy_file_range(struct file *file_in, loff_t pos_in,
3089 struct file *file_out, loff_t pos_out,
3090 size_t len, unsigned int flags)
3091{
3092 struct fuse_file *ff_in = file_in->private_data;
3093 struct fuse_file *ff_out = file_out->private_data;
3094 struct inode *inode_out = file_inode(file_out);
3095 struct fuse_inode *fi_out = get_fuse_inode(inode_out);
3096 struct fuse_conn *fc = ff_in->fc;
3097 FUSE_ARGS(args);
3098 struct fuse_copy_file_range_in inarg = {
3099 .fh_in = ff_in->fh,
3100 .off_in = pos_in,
3101 .nodeid_out = ff_out->nodeid,
3102 .fh_out = ff_out->fh,
3103 .off_out = pos_out,
3104 .len = len,
3105 .flags = flags
3106 };
3107 struct fuse_write_out outarg;
3108 ssize_t err;
3109 /* mark unstable when write-back is not used, and file_out gets
3110 * extended */
3111 bool is_unstable = (!fc->writeback_cache) &&
3112 ((pos_out + len) > inode_out->i_size);
3113
3114 if (fc->no_copy_file_range)
3115 return -EOPNOTSUPP;
3116
3117 inode_lock(inode_out);
3118
3119 if (fc->writeback_cache) {
3120 err = filemap_write_and_wait_range(inode_out->i_mapping,
3121 pos_out, pos_out + len);
3122 if (err)
3123 goto out;
3124
3125 fuse_sync_writes(inode_out);
3126 }
3127
3128 if (is_unstable)
3129 set_bit(FUSE_I_SIZE_UNSTABLE, &fi_out->state);
3130
3131 args.in.h.opcode = FUSE_COPY_FILE_RANGE;
3132 args.in.h.nodeid = ff_in->nodeid;
3133 args.in.numargs = 1;
3134 args.in.args[0].size = sizeof(inarg);
3135 args.in.args[0].value = &inarg;
3136 args.out.numargs = 1;
3137 args.out.args[0].size = sizeof(outarg);
3138 args.out.args[0].value = &outarg;
3139 err = fuse_simple_request(fc, &args);
3140 if (err == -ENOSYS) {
3141 fc->no_copy_file_range = 1;
3142 err = -EOPNOTSUPP;
3143 }
3144 if (err)
3145 goto out;
3146
3147 if (fc->writeback_cache) {
3148 fuse_write_update_size(inode_out, pos_out + outarg.size);
3149 file_update_time(file_out);
3150 }
3151
3152 fuse_invalidate_attr(inode_out);
3153
3154 err = outarg.size;
3155out:
3156 if (is_unstable)
3157 clear_bit(FUSE_I_SIZE_UNSTABLE, &fi_out->state);
3158
3159 inode_unlock(inode_out);
3160
3161 return err;
3162}
3163
4b6f5d20 3164static const struct file_operations fuse_file_operations = {
5559b8f4 3165 .llseek = fuse_file_llseek,
37c20f16 3166 .read_iter = fuse_file_read_iter,
84c3d55c 3167 .write_iter = fuse_file_write_iter,
b6aeaded
MS		 3168 .mmap = fuse_file_mmap,
3169 .open = fuse_open,
3170 .flush = fuse_flush,
3171 .release = fuse_release,
3172 .fsync = fuse_fsync,
71421259 3173 .lock = fuse_file_lock,
a9ff4f87 3174 .flock = fuse_file_flock,
3c3db095
MS		 3175 .splice_read = generic_file_splice_read,
3176 .splice_write = iter_file_splice_write,
59efec7b
TH		 3177 .unlocked_ioctl = fuse_file_ioctl,
3178 .compat_ioctl = fuse_file_compat_ioctl,
95668a69 3179 .poll = fuse_file_poll,
05ba1f08 3180 .fallocate = fuse_file_fallocate,
d4136d60 3181 .copy_file_range = fuse_copy_file_range,
413ef8cb
MS		 3182};
3183
f5e54d6e 3184static const struct address_space_operations fuse_file_aops = {
b6aeaded 3185 .readpage = fuse_readpage,
3be5a52b 3186 .writepage = fuse_writepage,
26d614df 3187 .writepages = fuse_writepages,
3be5a52b 3188 .launder_page = fuse_launder_page,
db50b96c 3189 .readpages = fuse_readpages,
3be5a52b 3190 .set_page_dirty = __set_page_dirty_nobuffers,
b2d2272f 3191 .bmap = fuse_bmap,
4273b793 3192 .direct_IO = fuse_direct_IO,
6b12c1b3
PE		 3193 .write_begin = fuse_write_begin,
3194 .write_end = fuse_write_end,
b6aeaded
MS		 3195};
3196
3197void fuse_init_file_inode(struct inode *inode)
3198{
ab2257e9
MS		 3199 struct fuse_inode *fi = get_fuse_inode(inode);
3200
45323fb7
MS		 3201 inode->i_fop = &fuse_file_operations;
3202 inode->i_data.a_ops = &fuse_file_aops;
ab2257e9
MS		 3203
3204 INIT_LIST_HEAD(&fi->write_files);
3205 INIT_LIST_HEAD(&fi->queued_writes);
3206 fi->writectr = 0;
3207 init_waitqueue_head(&fi->page_waitq);
3208 INIT_LIST_HEAD(&fi->writepages);
b6aeaded 3209}
Linux 6.x block layer and io_uring tree(s)