file.c 75.3 KB
Newer Older
M
Miklos Szeredi 已提交
1 2
/*
  FUSE: Filesystem in Userspace
M
Miklos Szeredi 已提交
3
  Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
M
Miklos Szeredi 已提交
4 5 6 7 8 9 10 11 12 13

  This program can be distributed under the terms of the GNU GPL.
  See the file COPYING.
*/

#include "fuse_i.h"

#include <linux/pagemap.h>
#include <linux/slab.h>
#include <linux/kernel.h>
A
Alexey Dobriyan 已提交
14
#include <linux/sched.h>
15
#include <linux/sched/signal.h>
16
#include <linux/module.h>
17
#include <linux/compat.h>
18
#include <linux/swap.h>
19
#include <linux/falloc.h>
20
#include <linux/uio.h>
M
Miklos Szeredi 已提交
21

22
static const struct file_operations fuse_direct_io_file_operations;
23

24 25
static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
			  int opcode, struct fuse_open_out *outargp)
M
Miklos Szeredi 已提交
26 27
{
	struct fuse_open_in inarg;
28
	FUSE_ARGS(args);
M
Miklos Szeredi 已提交
29 30

	memset(&inarg, 0, sizeof(inarg));
31 32 33
	inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY);
	if (!fc->atomic_o_trunc)
		inarg.flags &= ~O_TRUNC;
34 35 36 37 38 39 40 41
	args.in.h.opcode = opcode;
	args.in.h.nodeid = nodeid;
	args.in.numargs = 1;
	args.in.args[0].size = sizeof(inarg);
	args.in.args[0].value = &inarg;
	args.out.numargs = 1;
	args.out.args[0].size = sizeof(*outargp);
	args.out.args[0].value = outargp;
M
Miklos Szeredi 已提交
42

43
	return fuse_simple_request(fc, &args);
M
Miklos Szeredi 已提交
44 45
}

T
Tejun Heo 已提交
46
struct fuse_file *fuse_file_alloc(struct fuse_conn *fc)
M
Miklos Szeredi 已提交
47 48
{
	struct fuse_file *ff;
T
Tejun Heo 已提交
49

50
	ff = kzalloc(sizeof(struct fuse_file), GFP_KERNEL);
T
Tejun Heo 已提交
51 52 53
	if (unlikely(!ff))
		return NULL;

54
	ff->fc = fc;
55
	ff->reserved_req = fuse_request_alloc(0);
T
Tejun Heo 已提交
56 57 58
	if (unlikely(!ff->reserved_req)) {
		kfree(ff);
		return NULL;
M
Miklos Szeredi 已提交
59
	}
T
Tejun Heo 已提交
60 61

	INIT_LIST_HEAD(&ff->write_entry);
62
	refcount_set(&ff->count, 1);
T
Tejun Heo 已提交
63 64 65 66 67 68 69
	RB_CLEAR_NODE(&ff->polled_node);
	init_waitqueue_head(&ff->poll_wait);

	spin_lock(&fc->lock);
	ff->kh = ++fc->khctr;
	spin_unlock(&fc->lock);

M
Miklos Szeredi 已提交
70 71 72 73 74
	return ff;
}

void fuse_file_free(struct fuse_file *ff)
{
75
	fuse_request_free(ff->reserved_req);
M
Miklos Szeredi 已提交
76 77 78
	kfree(ff);
}

79
static struct fuse_file *fuse_file_get(struct fuse_file *ff)
80
{
81
	refcount_inc(&ff->count);
82 83 84
	return ff;
}

M
Miklos Szeredi 已提交
85 86
static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
{
87
	iput(req->misc.release.inode);
M
Miklos Szeredi 已提交
88 89
}

90
static void fuse_file_put(struct fuse_file *ff, bool sync, bool isdir)
91
{
92
	if (refcount_dec_and_test(&ff->count)) {
93
		struct fuse_req *req = ff->reserved_req;
94

95
		if (ff->fc->no_open && !isdir) {
96 97 98 99
			/*
			 * Drop the release request when client does not
			 * implement 'open'
			 */
M
Miklos Szeredi 已提交
100
			__clear_bit(FR_BACKGROUND, &req->flags);
101
			iput(req->misc.release.inode);
102 103
			fuse_put_request(ff->fc, req);
		} else if (sync) {
M
Miklos Szeredi 已提交
104
			__set_bit(FR_FORCE, &req->flags);
M
Miklos Szeredi 已提交
105
			__clear_bit(FR_BACKGROUND, &req->flags);
106
			fuse_request_send(ff->fc, req);
107
			iput(req->misc.release.inode);
108 109 110
			fuse_put_request(ff->fc, req);
		} else {
			req->end = fuse_release_end;
M
Miklos Szeredi 已提交
111
			__set_bit(FR_BACKGROUND, &req->flags);
112 113
			fuse_request_send_background(ff->fc, req);
		}
114 115 116 117
		kfree(ff);
	}
}

118 119
int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
		 bool isdir)
120 121 122 123 124 125 126 127
{
	struct fuse_file *ff;
	int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;

	ff = fuse_file_alloc(fc);
	if (!ff)
		return -ENOMEM;

128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144
	ff->fh = 0;
	ff->open_flags = FOPEN_KEEP_CACHE; /* Default for no-open */
	if (!fc->no_open || isdir) {
		struct fuse_open_out outarg;
		int err;

		err = fuse_send_open(fc, nodeid, file, opcode, &outarg);
		if (!err) {
			ff->fh = outarg.fh;
			ff->open_flags = outarg.open_flags;

		} else if (err != -ENOSYS || isdir) {
			fuse_file_free(ff);
			return err;
		} else {
			fc->no_open = 1;
		}
145 146 147
	}

	if (isdir)
148
		ff->open_flags &= ~FOPEN_DIRECT_IO;
149 150

	ff->nodeid = nodeid;
151
	file->private_data = ff;
152 153 154

	return 0;
}
155
EXPORT_SYMBOL_GPL(fuse_do_open);
156

157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172
static void fuse_link_write_file(struct file *file)
{
	struct inode *inode = file_inode(file);
	struct fuse_conn *fc = get_fuse_conn(inode);
	struct fuse_inode *fi = get_fuse_inode(inode);
	struct fuse_file *ff = file->private_data;
	/*
	 * file may be written through mmap, so chain it onto the
	 * inodes's write_file list
	 */
	spin_lock(&fc->lock);
	if (list_empty(&ff->write_entry))
		list_add(&ff->write_entry, &fi->write_files);
	spin_unlock(&fc->lock);
}

173
void fuse_finish_open(struct inode *inode, struct file *file)
M
Miklos Szeredi 已提交
174
{
175
	struct fuse_file *ff = file->private_data;
176
	struct fuse_conn *fc = get_fuse_conn(inode);
177 178

	if (ff->open_flags & FOPEN_DIRECT_IO)
M
Miklos Szeredi 已提交
179
		file->f_op = &fuse_direct_io_file_operations;
180
	if (!(ff->open_flags & FOPEN_KEEP_CACHE))
M
Miklos Szeredi 已提交
181
		invalidate_inode_pages2(inode->i_mapping);
182 183 184
	if (ff->open_flags & FOPEN_STREAM)
		stream_open(inode, file);
	else if (ff->open_flags & FOPEN_NONSEEKABLE)
T
Tejun Heo 已提交
185
		nonseekable_open(inode, file);
186 187 188 189 190 191 192 193
	if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) {
		struct fuse_inode *fi = get_fuse_inode(inode);

		spin_lock(&fc->lock);
		fi->attr_version = ++fc->attr_version;
		i_size_write(inode, 0);
		spin_unlock(&fc->lock);
		fuse_invalidate_attr(inode);
194 195
		if (fc->writeback_cache)
			file_update_time(file);
196
	}
P
Pavel Emelyanov 已提交
197 198
	if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache)
		fuse_link_write_file(file);
M
Miklos Szeredi 已提交
199 200
}

201
int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
M
Miklos Szeredi 已提交
202
{
T
Tejun Heo 已提交
203
	struct fuse_conn *fc = get_fuse_conn(inode);
M
Miklos Szeredi 已提交
204
	int err;
205 206 207
	bool lock_inode = (file->f_flags & O_TRUNC) &&
			  fc->atomic_o_trunc &&
			  fc->writeback_cache;
M
Miklos Szeredi 已提交
208 209 210 211 212

	err = generic_file_open(inode, file);
	if (err)
		return err;

213
	if (lock_inode)
A
Al Viro 已提交
214
		inode_lock(inode);
215

216
	err = fuse_do_open(fc, get_node_id(inode), file, isdir);
M
Miklos Szeredi 已提交
217

218 219
	if (!err)
		fuse_finish_open(inode, file);
220

221
	if (lock_inode)
A
Al Viro 已提交
222
		inode_unlock(inode);
223 224

	return err;
M
Miklos Szeredi 已提交
225 226
}

227
static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode)
228
{
229
	struct fuse_conn *fc = ff->fc;
230
	struct fuse_req *req = ff->reserved_req;
231
	struct fuse_release_in *inarg = &req->misc.release.in;
M
Miklos Szeredi 已提交
232

233 234 235 236 237 238
	spin_lock(&fc->lock);
	list_del(&ff->write_entry);
	if (!RB_EMPTY_NODE(&ff->polled_node))
		rb_erase(&ff->polled_node, &fc->polled_files);
	spin_unlock(&fc->lock);

239
	wake_up_interruptible_all(&ff->poll_wait);
240

M
Miklos Szeredi 已提交
241
	inarg->fh = ff->fh;
M
Miklos Szeredi 已提交
242
	inarg->flags = flags;
243
	req->in.h.opcode = opcode;
244
	req->in.h.nodeid = ff->nodeid;
M
Miklos Szeredi 已提交
245 246 247
	req->in.numargs = 1;
	req->in.args[0].size = sizeof(struct fuse_release_in);
	req->in.args[0].value = inarg;
M
Miklos Szeredi 已提交
248 249
}

250
void fuse_release_common(struct file *file, bool isdir)
M
Miklos Szeredi 已提交
251
{
252 253
	struct fuse_file *ff = file->private_data;
	struct fuse_req *req = ff->reserved_req;
254
	int opcode = isdir ? FUSE_RELEASEDIR : FUSE_RELEASE;
T
Tejun Heo 已提交
255

256
	fuse_prepare_release(ff, file->f_flags, opcode);
T
Tejun Heo 已提交
257

M
Miklos Szeredi 已提交
258 259 260 261 262 263
	if (ff->flock) {
		struct fuse_release_in *inarg = &req->misc.release.in;
		inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK;
		inarg->lock_owner = fuse_lock_owner_id(ff->fc,
						       (fl_owner_t) file);
	}
264 265
	/* Hold inode until release is finished */
	req->misc.release.inode = igrab(file_inode(file));
T
Tejun Heo 已提交
266 267 268 269 270

	/*
	 * Normally this will send the RELEASE request, however if
	 * some asynchronous READ or WRITE requests are outstanding,
	 * the sending will be delayed.
271 272 273 274
	 *
	 * Make the release synchronous if this is a fuseblk mount,
	 * synchronous RELEASE is allowed (and desirable) in this case
	 * because the server can be trusted not to screw up.
T
Tejun Heo 已提交
275
	 */
276
	fuse_file_put(ff, ff->fc->destroy_req != NULL, isdir);
M
Miklos Szeredi 已提交
277 278
}

279 280
static int fuse_open(struct inode *inode, struct file *file)
{
281
	return fuse_open_common(inode, file, false);
282 283 284 285
}

static int fuse_release(struct inode *inode, struct file *file)
{
P
Pavel Emelyanov 已提交
286 287 288 289
	struct fuse_conn *fc = get_fuse_conn(inode);

	/* see fuse_vma_close() for !writeback_cache case */
	if (fc->writeback_cache)
M
Miklos Szeredi 已提交
290
		write_inode_now(inode, 1);
M
Maxim Patlasov 已提交
291

292
	fuse_release_common(file, false);
293 294 295 296 297 298 299

	/* return value is ignored by VFS */
	return 0;
}

void fuse_sync_release(struct fuse_file *ff, int flags)
{
300
	WARN_ON(refcount_read(&ff->count) > 1);
301
	fuse_prepare_release(ff, flags, FUSE_RELEASE);
302 303 304 305
	/*
	 * iput(NULL) is a no-op and since the refcount is 1 and everything's
	 * synchronous, we are fine with not doing igrab() here"
	 */
306
	fuse_file_put(ff, true, false);
307
}
308
EXPORT_SYMBOL_GPL(fuse_sync_release);
309

310
/*
311 312
 * Scramble the ID space with XTEA, so that the value of the files_struct
 * pointer is not exposed to userspace.
313
 */
314
u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
315
{
316 317 318 319 320 321 322 323 324 325 326 327 328 329
	u32 *k = fc->scramble_key;
	u64 v = (unsigned long) id;
	u32 v0 = v;
	u32 v1 = v >> 32;
	u32 sum = 0;
	int i;

	for (i = 0; i < 32; i++) {
		v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
		sum += 0x9E3779B9;
		v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
	}

	return (u64) v0 + ((u64) v1 << 32);
330 331
}

M
Miklos Szeredi 已提交
332
/*
333
 * Check if any page in a range is under writeback
M
Miklos Szeredi 已提交
334 335 336 337
 *
 * This is currently done by walking the list of writepage requests
 * for the inode, which can be pretty inefficient.
 */
338 339
static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from,
				   pgoff_t idx_to)
M
Miklos Szeredi 已提交
340 341 342 343 344 345 346 347 348 349 350
{
	struct fuse_conn *fc = get_fuse_conn(inode);
	struct fuse_inode *fi = get_fuse_inode(inode);
	struct fuse_req *req;
	bool found = false;

	spin_lock(&fc->lock);
	list_for_each_entry(req, &fi->writepages, writepages_entry) {
		pgoff_t curr_index;

		BUG_ON(req->inode != inode);
351
		curr_index = req->misc.write.in.offset >> PAGE_SHIFT;
352 353
		if (idx_from < curr_index + req->num_pages &&
		    curr_index <= idx_to) {
M
Miklos Szeredi 已提交
354 355 356 357 358 359 360 361 362
			found = true;
			break;
		}
	}
	spin_unlock(&fc->lock);

	return found;
}

363 364 365 366 367
static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
{
	return fuse_range_is_writeback(inode, index, index);
}

M
Miklos Szeredi 已提交
368 369 370 371 372 373 374 375 376 377 378 379 380 381
/*
 * Wait for page writeback to be completed.
 *
 * Since fuse doesn't rely on the VM writeback tracking, this has to
 * use some other means.
 */
static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
{
	struct fuse_inode *fi = get_fuse_inode(inode);

	wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
	return 0;
}

382 383 384 385 386 387 388 389 390 391 392 393 394 395 396
/*
 * Wait for all pending writepages on the inode to finish.
 *
 * This is currently done by blocking further writes with FUSE_NOWRITE
 * and waiting for all sent writes to complete.
 *
 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
 * could conflict with truncation.
 */
static void fuse_sync_writes(struct inode *inode)
{
	fuse_set_nowrite(inode);
	fuse_release_nowrite(inode);
}

397
static int fuse_flush(struct file *file, fl_owner_t id)
M
Miklos Szeredi 已提交
398
{
A
Al Viro 已提交
399
	struct inode *inode = file_inode(file);
M
Miklos Szeredi 已提交
400 401 402 403 404 405
	struct fuse_conn *fc = get_fuse_conn(inode);
	struct fuse_file *ff = file->private_data;
	struct fuse_req *req;
	struct fuse_flush_in inarg;
	int err;

406 407 408
	if (is_bad_inode(inode))
		return -EIO;

M
Miklos Szeredi 已提交
409 410 411
	if (fc->no_flush)
		return 0;

M
Miklos Szeredi 已提交
412
	err = write_inode_now(inode, 1);
413 414 415
	if (err)
		return err;

A
Al Viro 已提交
416
	inode_lock(inode);
417
	fuse_sync_writes(inode);
A
Al Viro 已提交
418
	inode_unlock(inode);
419

420
	err = filemap_check_errors(file->f_mapping);
421 422 423
	if (err)
		return err;

M
Maxim Patlasov 已提交
424
	req = fuse_get_req_nofail_nopages(fc, file);
M
Miklos Szeredi 已提交
425 426
	memset(&inarg, 0, sizeof(inarg));
	inarg.fh = ff->fh;
427
	inarg.lock_owner = fuse_lock_owner_id(fc, id);
M
Miklos Szeredi 已提交
428 429 430 431 432
	req->in.h.opcode = FUSE_FLUSH;
	req->in.h.nodeid = get_node_id(inode);
	req->in.numargs = 1;
	req->in.args[0].size = sizeof(inarg);
	req->in.args[0].value = &inarg;
M
Miklos Szeredi 已提交
433
	__set_bit(FR_FORCE, &req->flags);
434
	fuse_request_send(fc, req);
M
Miklos Szeredi 已提交
435 436 437 438 439 440 441 442 443
	err = req->out.h.error;
	fuse_put_request(fc, req);
	if (err == -ENOSYS) {
		fc->no_flush = 1;
		err = 0;
	}
	return err;
}

444 445
int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
		      int datasync, int isdir)
M
Miklos Szeredi 已提交
446
{
447
	struct inode *inode = file->f_mapping->host;
M
Miklos Szeredi 已提交
448 449
	struct fuse_conn *fc = get_fuse_conn(inode);
	struct fuse_file *ff = file->private_data;
450
	FUSE_ARGS(args);
M
Miklos Szeredi 已提交
451 452 453
	struct fuse_fsync_in inarg;
	int err;

454 455 456
	if (is_bad_inode(inode))
		return -EIO;

A
Al Viro 已提交
457
	inode_lock(inode);
458

M
Miklos Szeredi 已提交
459 460 461 462 463
	/*
	 * Start writeback against all dirty pages of the inode, then
	 * wait for all outstanding writes, before sending the FSYNC
	 * request.
	 */
464
	err = file_write_and_wait_range(file, start, end);
M
Miklos Szeredi 已提交
465
	if (err)
466
		goto out;
M
Miklos Szeredi 已提交
467 468

	fuse_sync_writes(inode);
469 470 471

	/*
	 * Due to implementation of fuse writeback
472
	 * file_write_and_wait_range() does not catch errors.
473 474
	 * We have to do this directly after fuse_sync_writes()
	 */
475
	err = file_check_and_advance_wb_err(file);
476 477 478
	if (err)
		goto out;

M
Miklos Szeredi 已提交
479 480 481
	err = sync_inode_metadata(inode, 1);
	if (err)
		goto out;
M
Miklos Szeredi 已提交
482

M
Miklos Szeredi 已提交
483 484
	if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
		goto out;
M
Maxim Patlasov 已提交
485

M
Miklos Szeredi 已提交
486 487 488
	memset(&inarg, 0, sizeof(inarg));
	inarg.fh = ff->fh;
	inarg.fsync_flags = datasync ? 1 : 0;
489 490 491 492 493 494
	args.in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
	args.in.h.nodeid = get_node_id(inode);
	args.in.numargs = 1;
	args.in.args[0].size = sizeof(inarg);
	args.in.args[0].value = &inarg;
	err = fuse_simple_request(fc, &args);
M
Miklos Szeredi 已提交
495
	if (err == -ENOSYS) {
496 497 498 499
		if (isdir)
			fc->no_fsyncdir = 1;
		else
			fc->no_fsync = 1;
M
Miklos Szeredi 已提交
500 501
		err = 0;
	}
502
out:
A
Al Viro 已提交
503
	inode_unlock(inode);
M
Miklos Szeredi 已提交
504 505 506
	return err;
}

507 508
static int fuse_fsync(struct file *file, loff_t start, loff_t end,
		      int datasync)
509
{
510
	return fuse_fsync_common(file, start, end, datasync, 0);
511 512
}

513 514
void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos,
		    size_t count, int opcode)
M
Miklos Szeredi 已提交
515
{
516
	struct fuse_read_in *inarg = &req->misc.read.in;
517
	struct fuse_file *ff = file->private_data;
M
Miklos Szeredi 已提交
518

519 520 521
	inarg->fh = ff->fh;
	inarg->offset = pos;
	inarg->size = count;
522
	inarg->flags = file->f_flags;
523
	req->in.h.opcode = opcode;
524
	req->in.h.nodeid = ff->nodeid;
M
Miklos Szeredi 已提交
525 526
	req->in.numargs = 1;
	req->in.args[0].size = sizeof(struct fuse_read_in);
527
	req->in.args[0].value = inarg;
M
Miklos Szeredi 已提交
528 529 530 531 532
	req->out.argvar = 1;
	req->out.numargs = 1;
	req->out.args[0].size = count;
}

533
static void fuse_release_user_pages(struct fuse_req *req, bool should_dirty)
534 535 536 537 538
{
	unsigned i;

	for (i = 0; i < req->num_pages; i++) {
		struct page *page = req->pages[i];
539
		if (should_dirty)
540 541 542 543 544
			set_page_dirty_lock(page);
		put_page(page);
	}
}

545 546 547 548 549
static void fuse_io_release(struct kref *kref)
{
	kfree(container_of(kref, struct fuse_io_priv, refcnt));
}

550 551 552 553 554 555 556 557 558 559 560
static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io)
{
	if (io->err)
		return io->err;

	if (io->bytes >= 0 && io->write)
		return -EIO;

	return io->bytes < 0 ? io->size : io->bytes;
}

M
Maxim Patlasov 已提交
561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587
/**
 * In case of short read, the caller sets 'pos' to the position of
 * actual end of fuse request in IO request. Otherwise, if bytes_requested
 * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1.
 *
 * An example:
 * User requested DIO read of 64K. It was splitted into two 32K fuse requests,
 * both submitted asynchronously. The first of them was ACKed by userspace as
 * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The
 * second request was ACKed as short, e.g. only 1K was read, resulting in
 * pos == 33K.
 *
 * Thus, when all fuse requests are completed, the minimal non-negative 'pos'
 * will be equal to the length of the longest contiguous fragment of
 * transferred data starting from the beginning of IO request.
 */
static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
{
	int left;

	spin_lock(&io->lock);
	if (err)
		io->err = io->err ? : err;
	else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes))
		io->bytes = pos;

	left = --io->reqs;
588
	if (!left && io->blocking)
589
		complete(io->done);
M
Maxim Patlasov 已提交
590 591
	spin_unlock(&io->lock);

592
	if (!left && !io->blocking) {
593
		ssize_t res = fuse_get_res_by_io(io);
M
Maxim Patlasov 已提交
594

595 596 597 598
		if (res >= 0) {
			struct inode *inode = file_inode(io->iocb->ki_filp);
			struct fuse_conn *fc = get_fuse_conn(inode);
			struct fuse_inode *fi = get_fuse_inode(inode);
M
Maxim Patlasov 已提交
599

600 601 602
			spin_lock(&fc->lock);
			fi->attr_version = ++fc->attr_version;
			spin_unlock(&fc->lock);
M
Maxim Patlasov 已提交
603 604
		}

605
		io->iocb->ki_complete(io->iocb, res, 0);
M
Maxim Patlasov 已提交
606
	}
607 608

	kref_put(&io->refcnt, fuse_io_release);
M
Maxim Patlasov 已提交
609 610 611 612 613 614 615
}

static void fuse_aio_complete_req(struct fuse_conn *fc, struct fuse_req *req)
{
	struct fuse_io_priv *io = req->io;
	ssize_t pos = -1;

616
	fuse_release_user_pages(req, io->should_dirty);
M
Maxim Patlasov 已提交
617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634

	if (io->write) {
		if (req->misc.write.in.size != req->misc.write.out.size)
			pos = req->misc.write.in.offset - io->offset +
				req->misc.write.out.size;
	} else {
		if (req->misc.read.in.size != req->out.args[0].size)
			pos = req->misc.read.in.offset - io->offset +
				req->out.args[0].size;
	}

	fuse_aio_complete(io, req->out.h.error, pos);
}

static size_t fuse_async_req_send(struct fuse_conn *fc, struct fuse_req *req,
		size_t num_bytes, struct fuse_io_priv *io)
{
	spin_lock(&io->lock);
635
	kref_get(&io->refcnt);
M
Maxim Patlasov 已提交
636 637 638 639 640 641 642
	io->size += num_bytes;
	io->reqs++;
	spin_unlock(&io->lock);

	req->io = io;
	req->end = fuse_aio_complete_req;

643
	__fuse_get_request(req);
M
Maxim Patlasov 已提交
644 645 646 647 648
	fuse_request_send_background(fc, req);

	return num_bytes;
}

649
static size_t fuse_send_read(struct fuse_req *req, struct fuse_io_priv *io,
650
			     loff_t pos, size_t count, fl_owner_t owner)
651
{
652
	struct file *file = io->iocb->ki_filp;
653 654
	struct fuse_file *ff = file->private_data;
	struct fuse_conn *fc = ff->fc;
655

656
	fuse_read_fill(req, file, pos, count, FUSE_READ);
657
	if (owner != NULL) {
658
		struct fuse_read_in *inarg = &req->misc.read.in;
659 660 661 662

		inarg->read_flags |= FUSE_READ_LOCKOWNER;
		inarg->lock_owner = fuse_lock_owner_id(fc, owner);
	}
663 664 665 666

	if (io->async)
		return fuse_async_req_send(fc, req, count, io);

667
	fuse_request_send(fc, req);
668
	return req->out.args[0].size;
669 670
}

671 672 673 674 675 676 677
static void fuse_read_update_size(struct inode *inode, loff_t size,
				  u64 attr_ver)
{
	struct fuse_conn *fc = get_fuse_conn(inode);
	struct fuse_inode *fi = get_fuse_inode(inode);

	spin_lock(&fc->lock);
678 679
	if (attr_ver == fi->attr_version && size < inode->i_size &&
	    !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
680 681 682 683 684 685
		fi->attr_version = ++fc->attr_version;
		i_size_write(inode, size);
	}
	spin_unlock(&fc->lock);
}

686 687 688 689
static void fuse_short_read(struct fuse_req *req, struct inode *inode,
			    u64 attr_ver)
{
	size_t num_read = req->out.args[0].size;
P
Pavel Emelyanov 已提交
690 691 692 693 694 695 696 697 698
	struct fuse_conn *fc = get_fuse_conn(inode);

	if (fc->writeback_cache) {
		/*
		 * A hole in a file. Some data after the hole are in page cache,
		 * but have not reached the client fs yet. So, the hole is not
		 * present there.
		 */
		int i;
699 700
		int start_idx = num_read >> PAGE_SHIFT;
		size_t off = num_read & (PAGE_SIZE - 1);
701

P
Pavel Emelyanov 已提交
702
		for (i = start_idx; i < req->num_pages; i++) {
703
			zero_user_segment(req->pages[i], off, PAGE_SIZE);
P
Pavel Emelyanov 已提交
704 705 706 707 708 709
			off = 0;
		}
	} else {
		loff_t pos = page_offset(req->pages[0]) + num_read;
		fuse_read_update_size(inode, pos, attr_ver);
	}
710 711
}

712
static int fuse_do_readpage(struct file *file, struct page *page)
M
Miklos Szeredi 已提交
713
{
714 715
	struct kiocb iocb;
	struct fuse_io_priv io;
M
Miklos Szeredi 已提交
716 717
	struct inode *inode = page->mapping->host;
	struct fuse_conn *fc = get_fuse_conn(inode);
718
	struct fuse_req *req;
719 720
	size_t num_read;
	loff_t pos = page_offset(page);
721
	size_t count = PAGE_SIZE;
722
	u64 attr_ver;
723 724
	int err;

M
Miklos Szeredi 已提交
725
	/*
L
Lucas De Marchi 已提交
726
	 * Page writeback can extend beyond the lifetime of the
M
Miklos Szeredi 已提交
727 728 729 730 731
	 * page-cache page, so make sure we read a properly synced
	 * page.
	 */
	fuse_wait_on_page_writeback(inode, page->index);

M
Maxim Patlasov 已提交
732
	req = fuse_get_req(fc, 1);
733
	if (IS_ERR(req))
734
		return PTR_ERR(req);
M
Miklos Szeredi 已提交
735

736 737
	attr_ver = fuse_get_attr_version(fc);

M
Miklos Szeredi 已提交
738
	req->out.page_zeroing = 1;
739
	req->out.argpages = 1;
M
Miklos Szeredi 已提交
740 741
	req->num_pages = 1;
	req->pages[0] = page;
742
	req->page_descs[0].length = count;
743 744
	init_sync_kiocb(&iocb, file);
	io = (struct fuse_io_priv) FUSE_IO_PRIV_SYNC(&iocb);
745
	num_read = fuse_send_read(req, &io, pos, count, NULL);
M
Miklos Szeredi 已提交
746
	err = req->out.h.error;
747 748 749 750 751 752

	if (!err) {
		/*
		 * Short read means EOF.  If file size is larger, truncate it
		 */
		if (num_read < count)
753
			fuse_short_read(req, inode, attr_ver);
754

M
Miklos Szeredi 已提交
755
		SetPageUptodate(page);
756 757
	}

758
	fuse_put_request(fc, req);
759 760 761 762 763 764 765 766 767 768 769 770 771 772

	return err;
}

static int fuse_readpage(struct file *file, struct page *page)
{
	struct inode *inode = page->mapping->host;
	int err;

	err = -EIO;
	if (is_bad_inode(inode))
		goto out;

	err = fuse_do_readpage(file, page);
773
	fuse_invalidate_atime(inode);
M
Miklos Szeredi 已提交
774 775 776 777 778
 out:
	unlock_page(page);
	return err;
}

779
static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
780
{
781
	int i;
782 783
	size_t count = req->misc.read.in.size;
	size_t num_read = req->out.args[0].size;
784
	struct address_space *mapping = NULL;
785

786 787
	for (i = 0; mapping == NULL && i < req->num_pages; i++)
		mapping = req->pages[i]->mapping;
788

789 790 791 792 793 794
	if (mapping) {
		struct inode *inode = mapping->host;

		/*
		 * Short read means EOF. If file size is larger, truncate it
		 */
795 796
		if (!req->out.h.error && num_read < count)
			fuse_short_read(req, inode, req->misc.read.attr_ver);
797

798
		fuse_invalidate_atime(inode);
799
	}
800

801 802 803 804
	for (i = 0; i < req->num_pages; i++) {
		struct page *page = req->pages[i];
		if (!req->out.h.error)
			SetPageUptodate(page);
805 806
		else
			SetPageError(page);
807
		unlock_page(page);
808
		put_page(page);
809
	}
810
	if (req->ff)
811
		fuse_file_put(req->ff, false, false);
812 813
}

814
static void fuse_send_readpages(struct fuse_req *req, struct file *file)
815
{
816 817
	struct fuse_file *ff = file->private_data;
	struct fuse_conn *fc = ff->fc;
818
	loff_t pos = page_offset(req->pages[0]);
819
	size_t count = req->num_pages << PAGE_SHIFT;
820 821

	req->out.argpages = 1;
822
	req->out.page_zeroing = 1;
823
	req->out.page_replace = 1;
824
	fuse_read_fill(req, file, pos, count, FUSE_READ);
825
	req->misc.read.attr_ver = fuse_get_attr_version(fc);
826
	if (fc->async_read) {
827
		req->ff = fuse_file_get(ff);
828
		req->end = fuse_readpages_end;
829
		fuse_request_send_background(fc, req);
830
	} else {
831
		fuse_request_send(fc, req);
832
		fuse_readpages_end(fc, req);
833
		fuse_put_request(fc, req);
834
	}
835 836
}

837
struct fuse_fill_data {
838
	struct fuse_req *req;
839
	struct file *file;
840
	struct inode *inode;
M
Maxim Patlasov 已提交
841
	unsigned nr_pages;
842 843 844 845
};

static int fuse_readpages_fill(void *_data, struct page *page)
{
846
	struct fuse_fill_data *data = _data;
847 848 849 850
	struct fuse_req *req = data->req;
	struct inode *inode = data->inode;
	struct fuse_conn *fc = get_fuse_conn(inode);

M
Miklos Szeredi 已提交
851 852
	fuse_wait_on_page_writeback(inode, page->index);

853 854
	if (req->num_pages &&
	    (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
855
	     (req->num_pages + 1) * PAGE_SIZE > fc->max_read ||
856
	     req->pages[req->num_pages - 1]->index + 1 != page->index)) {
M
Maxim Patlasov 已提交
857 858
		int nr_alloc = min_t(unsigned, data->nr_pages,
				     FUSE_MAX_PAGES_PER_REQ);
859
		fuse_send_readpages(req, data->file);
860 861 862 863 864 865
		if (fc->async_read)
			req = fuse_get_req_for_background(fc, nr_alloc);
		else
			req = fuse_get_req(fc, nr_alloc);

		data->req = req;
866
		if (IS_ERR(req)) {
867
			unlock_page(page);
868
			return PTR_ERR(req);
869 870
		}
	}
M
Maxim Patlasov 已提交
871 872

	if (WARN_ON(req->num_pages >= req->max_pages)) {
873
		unlock_page(page);
M
Maxim Patlasov 已提交
874 875 876 877
		fuse_put_request(fc, req);
		return -EIO;
	}

878
	get_page(page);
879
	req->pages[req->num_pages] = page;
880
	req->page_descs[req->num_pages].length = PAGE_SIZE;
M
Miklos Szeredi 已提交
881
	req->num_pages++;
M
Maxim Patlasov 已提交
882
	data->nr_pages--;
883 884 885 886 887 888 889 890
	return 0;
}

static int fuse_readpages(struct file *file, struct address_space *mapping,
			  struct list_head *pages, unsigned nr_pages)
{
	struct inode *inode = mapping->host;
	struct fuse_conn *fc = get_fuse_conn(inode);
891
	struct fuse_fill_data data;
892
	int err;
M
Maxim Patlasov 已提交
893
	int nr_alloc = min_t(unsigned, nr_pages, FUSE_MAX_PAGES_PER_REQ);
894

895
	err = -EIO;
896
	if (is_bad_inode(inode))
897
		goto out;
898

899
	data.file = file;
900
	data.inode = inode;
901 902 903 904
	if (fc->async_read)
		data.req = fuse_get_req_for_background(fc, nr_alloc);
	else
		data.req = fuse_get_req(fc, nr_alloc);
M
Maxim Patlasov 已提交
905
	data.nr_pages = nr_pages;
906
	err = PTR_ERR(data.req);
907
	if (IS_ERR(data.req))
908
		goto out;
909 910

	err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
911 912
	if (!err) {
		if (data.req->num_pages)
913
			fuse_send_readpages(data.req, file);
914 915 916
		else
			fuse_put_request(fc, data.req);
	}
917
out:
918
	return err;
919 920
}

921
static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
M
Miklos Szeredi 已提交
922 923
{
	struct inode *inode = iocb->ki_filp->f_mapping->host;
924
	struct fuse_conn *fc = get_fuse_conn(inode);
M
Miklos Szeredi 已提交
925

926 927 928 929 930 931
	/*
	 * In auto invalidate mode, always update attributes on read.
	 * Otherwise, only update if we attempt to read past EOF (to ensure
	 * i_size is up to date).
	 */
	if (fc->auto_inval_data ||
932
	    (iocb->ki_pos + iov_iter_count(to) > i_size_read(inode))) {
M
Miklos Szeredi 已提交
933
		int err;
M
Miklos Szeredi 已提交
934
		err = fuse_update_attributes(inode, iocb->ki_filp);
M
Miklos Szeredi 已提交
935 936 937 938
		if (err)
			return err;
	}

939
	return generic_file_read_iter(iocb, to);
M
Miklos Szeredi 已提交
940 941
}

942
static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff,
943
			    loff_t pos, size_t count)
M
Miklos Szeredi 已提交
944
{
945 946
	struct fuse_write_in *inarg = &req->misc.write.in;
	struct fuse_write_out *outarg = &req->misc.write.out;
M
Miklos Szeredi 已提交
947

948 949 950
	inarg->fh = ff->fh;
	inarg->offset = pos;
	inarg->size = count;
M
Miklos Szeredi 已提交
951
	req->in.h.opcode = FUSE_WRITE;
952
	req->in.h.nodeid = ff->nodeid;
M
Miklos Szeredi 已提交
953
	req->in.numargs = 2;
954
	if (ff->fc->minor < 9)
955 956 957
		req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
	else
		req->in.args[0].size = sizeof(struct fuse_write_in);
958
	req->in.args[0].value = inarg;
M
Miklos Szeredi 已提交
959 960 961
	req->in.args[1].size = count;
	req->out.numargs = 1;
	req->out.args[0].size = sizeof(struct fuse_write_out);
962 963 964
	req->out.args[0].value = outarg;
}

965
static size_t fuse_send_write(struct fuse_req *req, struct fuse_io_priv *io,
966
			      loff_t pos, size_t count, fl_owner_t owner)
967
{
968 969
	struct kiocb *iocb = io->iocb;
	struct file *file = iocb->ki_filp;
970 971
	struct fuse_file *ff = file->private_data;
	struct fuse_conn *fc = ff->fc;
972 973
	struct fuse_write_in *inarg = &req->misc.write.in;

974
	fuse_write_fill(req, ff, pos, count);
975
	inarg->flags = file->f_flags;
976 977 978 979
	if (iocb->ki_flags & IOCB_DSYNC)
		inarg->flags |= O_DSYNC;
	if (iocb->ki_flags & IOCB_SYNC)
		inarg->flags |= O_SYNC;
980 981 982 983
	if (owner != NULL) {
		inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
		inarg->lock_owner = fuse_lock_owner_id(fc, owner);
	}
984 985 986 987

	if (io->async)
		return fuse_async_req_send(fc, req, count, io);

988
	fuse_request_send(fc, req);
989
	return req->misc.write.out.size;
M
Miklos Szeredi 已提交
990 991
}

M
Maxim Patlasov 已提交
992
bool fuse_write_update_size(struct inode *inode, loff_t pos)
993 994 995
{
	struct fuse_conn *fc = get_fuse_conn(inode);
	struct fuse_inode *fi = get_fuse_inode(inode);
M
Maxim Patlasov 已提交
996
	bool ret = false;
997 998 999

	spin_lock(&fc->lock);
	fi->attr_version = ++fc->attr_version;
M
Maxim Patlasov 已提交
1000
	if (pos > inode->i_size) {
1001
		i_size_write(inode, pos);
M
Maxim Patlasov 已提交
1002 1003
		ret = true;
	}
1004
	spin_unlock(&fc->lock);
M
Maxim Patlasov 已提交
1005 1006

	return ret;
1007 1008
}

1009
static size_t fuse_send_write_pages(struct fuse_req *req, struct kiocb *iocb,
N
Nick Piggin 已提交
1010 1011 1012 1013 1014 1015
				    struct inode *inode, loff_t pos,
				    size_t count)
{
	size_t res;
	unsigned offset;
	unsigned i;
1016
	struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
N
Nick Piggin 已提交
1017 1018 1019 1020

	for (i = 0; i < req->num_pages; i++)
		fuse_wait_on_page_writeback(inode, req->pages[i]->index);

1021
	res = fuse_send_write(req, &io, pos, count, NULL);
N
Nick Piggin 已提交
1022

1023
	offset = req->page_descs[0].offset;
N
Nick Piggin 已提交
1024 1025 1026 1027
	count = res;
	for (i = 0; i < req->num_pages; i++) {
		struct page *page = req->pages[i];

1028
		if (!req->out.h.error && !offset && count >= PAGE_SIZE)
N
Nick Piggin 已提交
1029 1030
			SetPageUptodate(page);

1031 1032
		if (count > PAGE_SIZE - offset)
			count -= PAGE_SIZE - offset;
N
Nick Piggin 已提交
1033 1034 1035 1036 1037
		else
			count = 0;
		offset = 0;

		unlock_page(page);
1038
		put_page(page);
N
Nick Piggin 已提交
1039 1040 1041 1042 1043 1044 1045 1046 1047 1048
	}

	return res;
}

static ssize_t fuse_fill_write_pages(struct fuse_req *req,
			       struct address_space *mapping,
			       struct iov_iter *ii, loff_t pos)
{
	struct fuse_conn *fc = get_fuse_conn(mapping->host);
1049
	unsigned offset = pos & (PAGE_SIZE - 1);
N
Nick Piggin 已提交
1050 1051 1052
	size_t count = 0;
	int err;

1053
	req->in.argpages = 1;
1054
	req->page_descs[0].offset = offset;
N
Nick Piggin 已提交
1055 1056 1057 1058

	do {
		size_t tmp;
		struct page *page;
1059 1060
		pgoff_t index = pos >> PAGE_SHIFT;
		size_t bytes = min_t(size_t, PAGE_SIZE - offset,
N
Nick Piggin 已提交
1061 1062 1063 1064 1065 1066 1067 1068 1069 1070
				     iov_iter_count(ii));

		bytes = min_t(size_t, bytes, fc->max_write - count);

 again:
		err = -EFAULT;
		if (iov_iter_fault_in_readable(ii, bytes))
			break;

		err = -ENOMEM;
1071
		page = grab_cache_page_write_begin(mapping, index, 0);
N
Nick Piggin 已提交
1072 1073 1074
		if (!page)
			break;

1075 1076 1077
		if (mapping_writably_mapped(mapping))
			flush_dcache_page(page);

N
Nick Piggin 已提交
1078 1079 1080
		tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes);
		flush_dcache_page(page);

1081
		iov_iter_advance(ii, tmp);
N
Nick Piggin 已提交
1082 1083
		if (!tmp) {
			unlock_page(page);
1084
			put_page(page);
N
Nick Piggin 已提交
1085 1086 1087 1088 1089 1090
			bytes = min(bytes, iov_iter_single_seg_count(ii));
			goto again;
		}

		err = 0;
		req->pages[req->num_pages] = page;
1091
		req->page_descs[req->num_pages].length = tmp;
N
Nick Piggin 已提交
1092 1093 1094 1095 1096
		req->num_pages++;

		count += tmp;
		pos += tmp;
		offset += tmp;
1097
		if (offset == PAGE_SIZE)
N
Nick Piggin 已提交
1098 1099
			offset = 0;

1100 1101
		if (!fc->big_writes)
			break;
N
Nick Piggin 已提交
1102
	} while (iov_iter_count(ii) && count < fc->max_write &&
1103
		 req->num_pages < req->max_pages && offset == 0);
N
Nick Piggin 已提交
1104 1105 1106 1107

	return count > 0 ? count : err;
}

1108 1109 1110
static inline unsigned fuse_wr_pages(loff_t pos, size_t len)
{
	return min_t(unsigned,
1111 1112
		     ((pos + len - 1) >> PAGE_SHIFT) -
		     (pos >> PAGE_SHIFT) + 1,
1113 1114 1115
		     FUSE_MAX_PAGES_PER_REQ);
}

1116
static ssize_t fuse_perform_write(struct kiocb *iocb,
N
Nick Piggin 已提交
1117 1118 1119 1120 1121
				  struct address_space *mapping,
				  struct iov_iter *ii, loff_t pos)
{
	struct inode *inode = mapping->host;
	struct fuse_conn *fc = get_fuse_conn(inode);
1122
	struct fuse_inode *fi = get_fuse_inode(inode);
N
Nick Piggin 已提交
1123 1124 1125 1126 1127 1128
	int err = 0;
	ssize_t res = 0;

	if (is_bad_inode(inode))
		return -EIO;

1129 1130 1131
	if (inode->i_size < pos + iov_iter_count(ii))
		set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);

N
Nick Piggin 已提交
1132 1133 1134
	do {
		struct fuse_req *req;
		ssize_t count;
1135
		unsigned nr_pages = fuse_wr_pages(pos, iov_iter_count(ii));
N
Nick Piggin 已提交
1136

1137
		req = fuse_get_req(fc, nr_pages);
N
Nick Piggin 已提交
1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148
		if (IS_ERR(req)) {
			err = PTR_ERR(req);
			break;
		}

		count = fuse_fill_write_pages(req, mapping, ii, pos);
		if (count <= 0) {
			err = count;
		} else {
			size_t num_written;

1149
			num_written = fuse_send_write_pages(req, iocb, inode,
N
Nick Piggin 已提交
1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166
							    pos, count);
			err = req->out.h.error;
			if (!err) {
				res += num_written;
				pos += num_written;

				/* break out of the loop on short write */
				if (num_written != count)
					err = -EIO;
			}
		}
		fuse_put_request(fc, req);
	} while (!err && iov_iter_count(ii));

	if (res > 0)
		fuse_write_update_size(inode, pos);

1167
	clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
N
Nick Piggin 已提交
1168 1169 1170 1171 1172
	fuse_invalidate_attr(inode);

	return res > 0 ? res : err;
}

A
Al Viro 已提交
1173
static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
N
Nick Piggin 已提交
1174 1175 1176 1177
{
	struct file *file = iocb->ki_filp;
	struct address_space *mapping = file->f_mapping;
	ssize_t written = 0;
A
Anand Avati 已提交
1178
	ssize_t written_buffered = 0;
N
Nick Piggin 已提交
1179 1180
	struct inode *inode = mapping->host;
	ssize_t err;
A
Anand Avati 已提交
1181
	loff_t endbyte = 0;
N
Nick Piggin 已提交
1182

P
Pavel Emelyanov 已提交
1183 1184
	if (get_fuse_conn(inode)->writeback_cache) {
		/* Update size (EOF optimization) and mode (SUID clearing) */
M
Miklos Szeredi 已提交
1185
		err = fuse_update_attributes(mapping->host, file);
P
Pavel Emelyanov 已提交
1186 1187 1188
		if (err)
			return err;

A
Al Viro 已提交
1189
		return generic_file_write_iter(iocb, from);
P
Pavel Emelyanov 已提交
1190 1191
	}

A
Al Viro 已提交
1192
	inode_lock(inode);
N
Nick Piggin 已提交
1193 1194

	/* We can write back this queue in page reclaim */
1195
	current->backing_dev_info = inode_to_bdi(inode);
N
Nick Piggin 已提交
1196

1197 1198
	err = generic_write_checks(iocb, from);
	if (err <= 0)
N
Nick Piggin 已提交
1199 1200
		goto out;

1201
	err = file_remove_privs(file);
N
Nick Piggin 已提交
1202 1203 1204
	if (err)
		goto out;

1205 1206 1207
	err = file_update_time(file);
	if (err)
		goto out;
N
Nick Piggin 已提交
1208

1209
	if (iocb->ki_flags & IOCB_DIRECT) {
1210
		loff_t pos = iocb->ki_pos;
1211
		written = generic_file_direct_write(iocb, from);
A
Al Viro 已提交
1212
		if (written < 0 || !iov_iter_count(from))
A
Anand Avati 已提交
1213 1214 1215
			goto out;

		pos += written;
N
Nick Piggin 已提交
1216

1217
		written_buffered = fuse_perform_write(iocb, mapping, from, pos);
A
Anand Avati 已提交
1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229
		if (written_buffered < 0) {
			err = written_buffered;
			goto out;
		}
		endbyte = pos + written_buffered - 1;

		err = filemap_write_and_wait_range(file->f_mapping, pos,
						   endbyte);
		if (err)
			goto out;

		invalidate_mapping_pages(file->f_mapping,
1230 1231
					 pos >> PAGE_SHIFT,
					 endbyte >> PAGE_SHIFT);
A
Anand Avati 已提交
1232 1233 1234 1235

		written += written_buffered;
		iocb->ki_pos = pos + written_buffered;
	} else {
1236
		written = fuse_perform_write(iocb, mapping, from, iocb->ki_pos);
A
Anand Avati 已提交
1237
		if (written >= 0)
1238
			iocb->ki_pos += written;
A
Anand Avati 已提交
1239
	}
N
Nick Piggin 已提交
1240 1241
out:
	current->backing_dev_info = NULL;
A
Al Viro 已提交
1242
	inode_unlock(inode);
1243 1244
	if (written > 0)
		written = generic_write_sync(iocb, written);
N
Nick Piggin 已提交
1245 1246 1247 1248

	return written ? written : err;
}

1249 1250
static inline void fuse_page_descs_length_init(struct fuse_req *req,
		unsigned index, unsigned nr_pages)
1251 1252 1253
{
	int i;

1254
	for (i = index; i < index + nr_pages; i++)
1255 1256 1257 1258
		req->page_descs[i].length = PAGE_SIZE -
			req->page_descs[i].offset;
}

1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269
static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii)
{
	return (unsigned long)ii->iov->iov_base + ii->iov_offset;
}

static inline size_t fuse_get_frag_size(const struct iov_iter *ii,
					size_t max_size)
{
	return min(iov_iter_single_seg_count(ii), max_size);
}

1270
static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii,
1271
			       size_t *nbytesp, int write)
M
Miklos Szeredi 已提交
1272
{
1273
	size_t nbytes = 0;  /* # bytes already packed in req */
1274
	ssize_t ret = 0;
1275

1276
	/* Special case for kernel I/O: can copy directly into the buffer */
A
Al Viro 已提交
1277
	if (ii->type & ITER_KVEC) {
1278 1279 1280
		unsigned long user_addr = fuse_get_user_addr(ii);
		size_t frag_size = fuse_get_frag_size(ii, *nbytesp);

1281 1282 1283 1284 1285
		if (write)
			req->in.args[1].value = (void *) user_addr;
		else
			req->out.args[0].value = (void *) user_addr;

1286 1287
		iov_iter_advance(ii, frag_size);
		*nbytesp = frag_size;
1288 1289
		return 0;
	}
M
Miklos Szeredi 已提交
1290

1291
	while (nbytes < *nbytesp && req->num_pages < req->max_pages) {
1292
		unsigned npages;
A
Al Viro 已提交
1293
		size_t start;
1294
		ret = iov_iter_get_pages(ii, &req->pages[req->num_pages],
1295
					*nbytesp - nbytes,
1296 1297
					req->max_pages - req->num_pages,
					&start);
1298
		if (ret < 0)
1299
			break;
1300

A
Al Viro 已提交
1301 1302
		iov_iter_advance(ii, ret);
		nbytes += ret;
1303

A
Al Viro 已提交
1304 1305
		ret += start;
		npages = (ret + PAGE_SIZE - 1) / PAGE_SIZE;
1306

A
Al Viro 已提交
1307
		req->page_descs[req->num_pages].offset = start;
1308 1309 1310 1311
		fuse_page_descs_length_init(req, req->num_pages, npages);

		req->num_pages += npages;
		req->page_descs[req->num_pages - 1].length -=
A
Al Viro 已提交
1312
			(PAGE_SIZE - ret) & (PAGE_SIZE - 1);
1313
	}
1314 1315 1316 1317 1318 1319

	if (write)
		req->in.argpages = 1;
	else
		req->out.argpages = 1;

1320
	*nbytesp = nbytes;
1321

1322
	return ret < 0 ? ret : 0;
M
Miklos Szeredi 已提交
1323 1324
}

1325 1326
static inline int fuse_iter_npages(const struct iov_iter *ii_p)
{
A
Al Viro 已提交
1327
	return iov_iter_npages(ii_p, FUSE_MAX_PAGES_PER_REQ);
1328 1329
}

A
Al Viro 已提交
1330 1331
ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
		       loff_t *ppos, int flags)
M
Miklos Szeredi 已提交
1332
{
1333 1334
	int write = flags & FUSE_DIO_WRITE;
	int cuse = flags & FUSE_DIO_CUSE;
1335
	struct file *file = io->iocb->ki_filp;
1336
	struct inode *inode = file->f_mapping->host;
1337 1338
	struct fuse_file *ff = file->private_data;
	struct fuse_conn *fc = ff->fc;
M
Miklos Szeredi 已提交
1339 1340
	size_t nmax = write ? fc->max_write : fc->max_read;
	loff_t pos = *ppos;
A
Al Viro 已提交
1341
	size_t count = iov_iter_count(iter);
1342 1343
	pgoff_t idx_from = pos >> PAGE_SHIFT;
	pgoff_t idx_to = (pos + count - 1) >> PAGE_SHIFT;
M
Miklos Szeredi 已提交
1344
	ssize_t res = 0;
1345
	struct fuse_req *req;
1346
	int err = 0;
1347

1348
	if (io->async)
A
Al Viro 已提交
1349
		req = fuse_get_req_for_background(fc, fuse_iter_npages(iter));
1350
	else
A
Al Viro 已提交
1351
		req = fuse_get_req(fc, fuse_iter_npages(iter));
1352 1353
	if (IS_ERR(req))
		return PTR_ERR(req);
M
Miklos Szeredi 已提交
1354

1355 1356
	if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) {
		if (!write)
A
Al Viro 已提交
1357
			inode_lock(inode);
1358 1359
		fuse_sync_writes(inode);
		if (!write)
A
Al Viro 已提交
1360
			inode_unlock(inode);
1361 1362
	}

1363
	io->should_dirty = !write && iter_is_iovec(iter);
M
Miklos Szeredi 已提交
1364 1365
	while (count) {
		size_t nres;
1366
		fl_owner_t owner = current->files;
1367
		size_t nbytes = min(count, nmax);
1368 1369
		err = fuse_get_user_pages(req, iter, &nbytes, write);
		if (err && !nbytes)
M
Miklos Szeredi 已提交
1370
			break;
1371

M
Miklos Szeredi 已提交
1372
		if (write)
1373
			nres = fuse_send_write(req, io, pos, nbytes, owner);
M
Miklos Szeredi 已提交
1374
		else
1375
			nres = fuse_send_read(req, io, pos, nbytes, owner);
1376

1377
		if (!io->async)
1378
			fuse_release_user_pages(req, io->should_dirty);
M
Miklos Szeredi 已提交
1379
		if (req->out.h.error) {
1380
			err = req->out.h.error;
M
Miklos Szeredi 已提交
1381 1382
			break;
		} else if (nres > nbytes) {
1383 1384
			res = 0;
			err = -EIO;
M
Miklos Szeredi 已提交
1385 1386 1387 1388 1389 1390 1391
			break;
		}
		count -= nres;
		res += nres;
		pos += nres;
		if (nres != nbytes)
			break;
1392 1393
		if (count) {
			fuse_put_request(fc, req);
1394 1395
			if (io->async)
				req = fuse_get_req_for_background(fc,
A
Al Viro 已提交
1396
					fuse_iter_npages(iter));
1397
			else
A
Al Viro 已提交
1398
				req = fuse_get_req(fc, fuse_iter_npages(iter));
1399 1400 1401
			if (IS_ERR(req))
				break;
		}
M
Miklos Szeredi 已提交
1402
	}
1403 1404
	if (!IS_ERR(req))
		fuse_put_request(fc, req);
1405
	if (res > 0)
M
Miklos Szeredi 已提交
1406 1407
		*ppos = pos;

1408
	return res > 0 ? res : err;
M
Miklos Szeredi 已提交
1409
}
1410
EXPORT_SYMBOL_GPL(fuse_direct_io);
M
Miklos Szeredi 已提交
1411

1412
static ssize_t __fuse_direct_read(struct fuse_io_priv *io,
A
Al Viro 已提交
1413 1414
				  struct iov_iter *iter,
				  loff_t *ppos)
M
Miklos Szeredi 已提交
1415
{
1416
	ssize_t res;
1417
	struct inode *inode = file_inode(io->iocb->ki_filp);
1418 1419 1420 1421

	if (is_bad_inode(inode))
		return -EIO;

A
Al Viro 已提交
1422
	res = fuse_direct_io(io, iter, ppos, 0);
1423 1424 1425 1426

	fuse_invalidate_attr(inode);

	return res;
M
Miklos Szeredi 已提交
1427 1428
}

1429
static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to)
1430
{
1431
	struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
1432
	return __fuse_direct_read(&io, to, &iocb->ki_pos);
1433 1434
}

1435
static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from)
A
Anand Avati 已提交
1436
{
1437 1438
	struct inode *inode = file_inode(iocb->ki_filp);
	struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb);
1439
	ssize_t res;
A
Anand Avati 已提交
1440 1441 1442 1443 1444

	if (is_bad_inode(inode))
		return -EIO;

	/* Don't allow parallel writes to the same file */
A
Al Viro 已提交
1445
	inode_lock(inode);
1446 1447
	res = generic_write_checks(iocb, from);
	if (res > 0)
1448 1449
		res = fuse_direct_io(&io, from, &iocb->ki_pos, FUSE_DIO_WRITE);
	fuse_invalidate_attr(inode);
1450
	if (res > 0)
1451
		fuse_write_update_size(inode, iocb->ki_pos);
A
Al Viro 已提交
1452
	inode_unlock(inode);
A
Anand Avati 已提交
1453 1454 1455 1456

	return res;
}

M
Miklos Szeredi 已提交
1457
static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
M
Miklos Szeredi 已提交
1458
{
1459 1460 1461 1462
	int i;

	for (i = 0; i < req->num_pages; i++)
		__free_page(req->pages[i]);
1463 1464

	if (req->ff)
1465
		fuse_file_put(req->ff, false, false);
M
Miklos Szeredi 已提交
1466 1467 1468 1469 1470 1471
}

static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
{
	struct inode *inode = req->inode;
	struct fuse_inode *fi = get_fuse_inode(inode);
1472
	struct backing_dev_info *bdi = inode_to_bdi(inode);
1473
	int i;
M
Miklos Szeredi 已提交
1474 1475

	list_del(&req->writepages_entry);
1476
	for (i = 0; i < req->num_pages; i++) {
1477
		dec_wb_stat(&bdi->wb, WB_WRITEBACK);
1478
		dec_node_page_state(req->pages[i], NR_WRITEBACK_TEMP);
1479
		wb_writeout_inc(&bdi->wb);
1480
	}
M
Miklos Szeredi 已提交
1481 1482 1483 1484
	wake_up(&fi->page_waitq);
}

/* Called under fc->lock, may release and reacquire it */
1485 1486
static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req,
				loff_t size)
M
Miklos Szeredi 已提交
1487 1488
__releases(fc->lock)
__acquires(fc->lock)
M
Miklos Szeredi 已提交
1489 1490 1491
{
	struct fuse_inode *fi = get_fuse_inode(req->inode);
	struct fuse_write_in *inarg = &req->misc.write.in;
1492
	__u64 data_size = req->num_pages * PAGE_SIZE;
M
Miklos Szeredi 已提交
1493 1494 1495 1496

	if (!fc->connected)
		goto out_free;

1497 1498
	if (inarg->offset + data_size <= size) {
		inarg->size = data_size;
M
Miklos Szeredi 已提交
1499
	} else if (inarg->offset < size) {
1500
		inarg->size = size - inarg->offset;
M
Miklos Szeredi 已提交
1501 1502 1503
	} else {
		/* Got truncated off completely */
		goto out_free;
M
Miklos Szeredi 已提交
1504
	}
M
Miklos Szeredi 已提交
1505 1506 1507

	req->in.args[1].size = inarg->size;
	fi->writectr++;
1508
	fuse_request_send_background_locked(fc, req);
M
Miklos Szeredi 已提交
1509 1510 1511 1512 1513 1514
	return;

 out_free:
	fuse_writepage_finish(fc, req);
	spin_unlock(&fc->lock);
	fuse_writepage_free(fc, req);
1515
	fuse_put_request(fc, req);
M
Miklos Szeredi 已提交
1516
	spin_lock(&fc->lock);
M
Miklos Szeredi 已提交
1517 1518
}

M
Miklos Szeredi 已提交
1519 1520 1521 1522 1523 1524 1525
/*
 * If fi->writectr is positive (no truncate or fsync going on) send
 * all queued writepage requests.
 *
 * Called with fc->lock
 */
void fuse_flush_writepages(struct inode *inode)
M
Miklos Szeredi 已提交
1526 1527
__releases(fc->lock)
__acquires(fc->lock)
M
Miklos Szeredi 已提交
1528
{
M
Miklos Szeredi 已提交
1529 1530
	struct fuse_conn *fc = get_fuse_conn(inode);
	struct fuse_inode *fi = get_fuse_inode(inode);
M
Miklos Szeredi 已提交
1531
	loff_t crop = i_size_read(inode);
M
Miklos Szeredi 已提交
1532 1533 1534 1535 1536
	struct fuse_req *req;

	while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
		req = list_entry(fi->queued_writes.next, struct fuse_req, list);
		list_del_init(&req->list);
1537
		fuse_send_writepage(fc, req, crop);
M
Miklos Szeredi 已提交
1538 1539 1540 1541 1542 1543 1544 1545 1546 1547
	}
}

static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
{
	struct inode *inode = req->inode;
	struct fuse_inode *fi = get_fuse_inode(inode);

	mapping_set_error(inode->i_mapping, req->out.h.error);
	spin_lock(&fc->lock);
1548
	while (req->misc.write.next) {
1549 1550
		struct fuse_conn *fc = get_fuse_conn(inode);
		struct fuse_write_in *inarg = &req->misc.write.in;
1551 1552 1553
		struct fuse_req *next = req->misc.write.next;
		req->misc.write.next = next->misc.write.next;
		next->misc.write.next = NULL;
1554
		next->ff = fuse_file_get(req->ff);
1555
		list_add(&next->writepages_entry, &fi->writepages);
1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580

		/*
		 * Skip fuse_flush_writepages() to make it easy to crop requests
		 * based on primary request size.
		 *
		 * 1st case (trivial): there are no concurrent activities using
		 * fuse_set/release_nowrite.  Then we're on safe side because
		 * fuse_flush_writepages() would call fuse_send_writepage()
		 * anyway.
		 *
		 * 2nd case: someone called fuse_set_nowrite and it is waiting
		 * now for completion of all in-flight requests.  This happens
		 * rarely and no more than once per page, so this should be
		 * okay.
		 *
		 * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle
		 * of fuse_set_nowrite..fuse_release_nowrite section.  The fact
		 * that fuse_set_nowrite returned implies that all in-flight
		 * requests were completed along with all of their secondary
		 * requests.  Further primary requests are blocked by negative
		 * writectr.  Hence there cannot be any in-flight requests and
		 * no invocations of fuse_writepage_end() while we're in
		 * fuse_set_nowrite..fuse_release_nowrite section.
		 */
		fuse_send_writepage(fc, next, inarg->offset + inarg->size);
1581
	}
M
Miklos Szeredi 已提交
1582 1583 1584 1585 1586 1587
	fi->writectr--;
	fuse_writepage_finish(fc, req);
	spin_unlock(&fc->lock);
	fuse_writepage_free(fc, req);
}

M
Miklos Szeredi 已提交
1588 1589
static struct fuse_file *__fuse_write_file_get(struct fuse_conn *fc,
					       struct fuse_inode *fi)
1590
{
M
Miklos Szeredi 已提交
1591
	struct fuse_file *ff = NULL;
1592 1593

	spin_lock(&fc->lock);
M
Miklos Szeredi 已提交
1594
	if (!list_empty(&fi->write_files)) {
M
Miklos Szeredi 已提交
1595 1596 1597 1598
		ff = list_entry(fi->write_files.next, struct fuse_file,
				write_entry);
		fuse_file_get(ff);
	}
1599 1600 1601 1602 1603
	spin_unlock(&fc->lock);

	return ff;
}

M
Miklos Szeredi 已提交
1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619
static struct fuse_file *fuse_write_file_get(struct fuse_conn *fc,
					     struct fuse_inode *fi)
{
	struct fuse_file *ff = __fuse_write_file_get(fc, fi);
	WARN_ON(!ff);
	return ff;
}

int fuse_write_inode(struct inode *inode, struct writeback_control *wbc)
{
	struct fuse_conn *fc = get_fuse_conn(inode);
	struct fuse_inode *fi = get_fuse_inode(inode);
	struct fuse_file *ff;
	int err;

	ff = __fuse_write_file_get(fc, fi);
1620
	err = fuse_flush_times(inode, ff);
M
Miklos Szeredi 已提交
1621
	if (ff)
1622
		fuse_file_put(ff, false, false);
M
Miklos Szeredi 已提交
1623 1624 1625 1626

	return err;
}

M
Miklos Szeredi 已提交
1627 1628 1629 1630 1631 1632 1633 1634
static int fuse_writepage_locked(struct page *page)
{
	struct address_space *mapping = page->mapping;
	struct inode *inode = mapping->host;
	struct fuse_conn *fc = get_fuse_conn(inode);
	struct fuse_inode *fi = get_fuse_inode(inode);
	struct fuse_req *req;
	struct page *tmp_page;
M
Miklos Szeredi 已提交
1635
	int error = -ENOMEM;
M
Miklos Szeredi 已提交
1636 1637 1638

	set_page_writeback(page);

1639
	req = fuse_request_alloc_nofs(1);
M
Miklos Szeredi 已提交
1640 1641 1642
	if (!req)
		goto err;

M
Miklos Szeredi 已提交
1643 1644
	/* writeback always goes to bg_queue */
	__set_bit(FR_BACKGROUND, &req->flags);
M
Miklos Szeredi 已提交
1645 1646 1647 1648
	tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
	if (!tmp_page)
		goto err_free;

M
Miklos Szeredi 已提交
1649
	error = -EIO;
1650
	req->ff = fuse_write_file_get(fc, fi);
M
Miklos Szeredi 已提交
1651
	if (!req->ff)
1652
		goto err_nofile;
M
Miklos Szeredi 已提交
1653

1654
	fuse_write_fill(req, req->ff, page_offset(page), 0);
M
Miklos Szeredi 已提交
1655 1656

	copy_highpage(tmp_page, page);
1657
	req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
1658
	req->misc.write.next = NULL;
1659
	req->in.argpages = 1;
M
Miklos Szeredi 已提交
1660 1661
	req->num_pages = 1;
	req->pages[0] = tmp_page;
1662
	req->page_descs[0].offset = 0;
1663
	req->page_descs[0].length = PAGE_SIZE;
M
Miklos Szeredi 已提交
1664 1665 1666
	req->end = fuse_writepage_end;
	req->inode = inode;

1667
	inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
1668
	inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP);
M
Miklos Szeredi 已提交
1669 1670 1671 1672 1673 1674 1675

	spin_lock(&fc->lock);
	list_add(&req->writepages_entry, &fi->writepages);
	list_add_tail(&req->list, &fi->queued_writes);
	fuse_flush_writepages(inode);
	spin_unlock(&fc->lock);

1676 1677
	end_page_writeback(page);

M
Miklos Szeredi 已提交
1678 1679
	return 0;

1680 1681
err_nofile:
	__free_page(tmp_page);
M
Miklos Szeredi 已提交
1682 1683 1684
err_free:
	fuse_request_free(req);
err:
1685
	mapping_set_error(page->mapping, error);
M
Miklos Szeredi 已提交
1686
	end_page_writeback(page);
M
Miklos Szeredi 已提交
1687
	return error;
M
Miklos Szeredi 已提交
1688 1689 1690 1691 1692 1693
}

static int fuse_writepage(struct page *page, struct writeback_control *wbc)
{
	int err;

1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705
	if (fuse_page_is_writeback(page->mapping->host, page->index)) {
		/*
		 * ->writepages() should be called for sync() and friends.  We
		 * should only get here on direct reclaim and then we are
		 * allowed to skip a page which is already in flight
		 */
		WARN_ON(wbc->sync_mode == WB_SYNC_ALL);

		redirty_page_for_writepage(wbc, page);
		return 0;
	}

M
Miklos Szeredi 已提交
1706 1707 1708 1709 1710 1711
	err = fuse_writepage_locked(page);
	unlock_page(page);

	return err;
}

1712 1713 1714 1715
struct fuse_fill_wb_data {
	struct fuse_req *req;
	struct fuse_file *ff;
	struct inode *inode;
1716
	struct page **orig_pages;
1717 1718 1719 1720 1721 1722 1723 1724
};

static void fuse_writepages_send(struct fuse_fill_wb_data *data)
{
	struct fuse_req *req = data->req;
	struct inode *inode = data->inode;
	struct fuse_conn *fc = get_fuse_conn(inode);
	struct fuse_inode *fi = get_fuse_inode(inode);
1725 1726
	int num_pages = req->num_pages;
	int i;
1727 1728 1729 1730 1731 1732

	req->ff = fuse_file_get(data->ff);
	spin_lock(&fc->lock);
	list_add_tail(&req->list, &fi->queued_writes);
	fuse_flush_writepages(inode);
	spin_unlock(&fc->lock);
1733 1734 1735

	for (i = 0; i < num_pages; i++)
		end_page_writeback(data->orig_pages[i]);
1736 1737
}

1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753
static bool fuse_writepage_in_flight(struct fuse_req *new_req,
				     struct page *page)
{
	struct fuse_conn *fc = get_fuse_conn(new_req->inode);
	struct fuse_inode *fi = get_fuse_inode(new_req->inode);
	struct fuse_req *tmp;
	struct fuse_req *old_req;
	bool found = false;
	pgoff_t curr_index;

	BUG_ON(new_req->num_pages != 0);

	spin_lock(&fc->lock);
	list_del(&new_req->writepages_entry);
	list_for_each_entry(old_req, &fi->writepages, writepages_entry) {
		BUG_ON(old_req->inode != new_req->inode);
1754
		curr_index = old_req->misc.write.in.offset >> PAGE_SHIFT;
1755 1756 1757 1758 1759 1760
		if (curr_index <= page->index &&
		    page->index < curr_index + old_req->num_pages) {
			found = true;
			break;
		}
	}
1761 1762
	if (!found) {
		list_add(&new_req->writepages_entry, &fi->writepages);
1763
		goto out_unlock;
1764
	}
1765

1766
	new_req->num_pages = 1;
1767 1768
	for (tmp = old_req; tmp != NULL; tmp = tmp->misc.write.next) {
		BUG_ON(tmp->inode != new_req->inode);
1769
		curr_index = tmp->misc.write.in.offset >> PAGE_SHIFT;
1770 1771 1772 1773 1774 1775
		if (tmp->num_pages == 1 &&
		    curr_index == page->index) {
			old_req = tmp;
		}
	}

M
Miklos Szeredi 已提交
1776
	if (old_req->num_pages == 1 && test_bit(FR_PENDING, &old_req->flags)) {
1777
		struct backing_dev_info *bdi = inode_to_bdi(page->mapping->host);
1778

1779 1780 1781
		copy_highpage(old_req->pages[0], page);
		spin_unlock(&fc->lock);

1782
		dec_wb_stat(&bdi->wb, WB_WRITEBACK);
1783
		dec_node_page_state(new_req->pages[0], NR_WRITEBACK_TEMP);
1784
		wb_writeout_inc(&bdi->wb);
1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797
		fuse_writepage_free(fc, new_req);
		fuse_request_free(new_req);
		goto out;
	} else {
		new_req->misc.write.next = old_req->misc.write.next;
		old_req->misc.write.next = new_req;
	}
out_unlock:
	spin_unlock(&fc->lock);
out:
	return found;
}

1798 1799 1800 1801 1802 1803 1804 1805
static int fuse_writepages_fill(struct page *page,
		struct writeback_control *wbc, void *_data)
{
	struct fuse_fill_wb_data *data = _data;
	struct fuse_req *req = data->req;
	struct inode *inode = data->inode;
	struct fuse_conn *fc = get_fuse_conn(inode);
	struct page *tmp_page;
1806
	bool is_writeback;
1807 1808 1809 1810 1811 1812 1813 1814 1815
	int err;

	if (!data->ff) {
		err = -EIO;
		data->ff = fuse_write_file_get(fc, get_fuse_inode(inode));
		if (!data->ff)
			goto out_unlock;
	}

1816 1817 1818 1819 1820 1821 1822 1823 1824 1825
	/*
	 * Being under writeback is unlikely but possible.  For example direct
	 * read to an mmaped fuse file will set the page dirty twice; once when
	 * the pages are faulted with get_user_pages(), and then after the read
	 * completed.
	 */
	is_writeback = fuse_page_is_writeback(inode, page->index);

	if (req && req->num_pages &&
	    (is_writeback || req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
1826
	     (req->num_pages + 1) * PAGE_SIZE > fc->max_write ||
1827 1828 1829
	     data->orig_pages[req->num_pages - 1]->index + 1 != page->index)) {
		fuse_writepages_send(data);
		data->req = NULL;
1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860
	}
	err = -ENOMEM;
	tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
	if (!tmp_page)
		goto out_unlock;

	/*
	 * The page must not be redirtied until the writeout is completed
	 * (i.e. userspace has sent a reply to the write request).  Otherwise
	 * there could be more than one temporary page instance for each real
	 * page.
	 *
	 * This is ensured by holding the page lock in page_mkwrite() while
	 * checking fuse_page_is_writeback().  We already hold the page lock
	 * since clear_page_dirty_for_io() and keep it held until we add the
	 * request to the fi->writepages list and increment req->num_pages.
	 * After this fuse_page_is_writeback() will indicate that the page is
	 * under writeback, so we can release the page lock.
	 */
	if (data->req == NULL) {
		struct fuse_inode *fi = get_fuse_inode(inode);

		err = -ENOMEM;
		req = fuse_request_alloc_nofs(FUSE_MAX_PAGES_PER_REQ);
		if (!req) {
			__free_page(tmp_page);
			goto out_unlock;
		}

		fuse_write_fill(req, data->ff, page_offset(page), 0);
		req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
1861
		req->misc.write.next = NULL;
1862
		req->in.argpages = 1;
M
Miklos Szeredi 已提交
1863
		__set_bit(FR_BACKGROUND, &req->flags);
1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880
		req->num_pages = 0;
		req->end = fuse_writepage_end;
		req->inode = inode;

		spin_lock(&fc->lock);
		list_add(&req->writepages_entry, &fi->writepages);
		spin_unlock(&fc->lock);

		data->req = req;
	}
	set_page_writeback(page);

	copy_highpage(tmp_page, page);
	req->pages[req->num_pages] = tmp_page;
	req->page_descs[req->num_pages].offset = 0;
	req->page_descs[req->num_pages].length = PAGE_SIZE;

1881
	inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK);
1882
	inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP);
1883 1884 1885 1886 1887 1888 1889

	err = 0;
	if (is_writeback && fuse_writepage_in_flight(req, page)) {
		end_page_writeback(page);
		data->req = NULL;
		goto out_unlock;
	}
1890
	data->orig_pages[req->num_pages] = page;
1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920

	/*
	 * Protected by fc->lock against concurrent access by
	 * fuse_page_is_writeback().
	 */
	spin_lock(&fc->lock);
	req->num_pages++;
	spin_unlock(&fc->lock);

out_unlock:
	unlock_page(page);

	return err;
}

static int fuse_writepages(struct address_space *mapping,
			   struct writeback_control *wbc)
{
	struct inode *inode = mapping->host;
	struct fuse_fill_wb_data data;
	int err;

	err = -EIO;
	if (is_bad_inode(inode))
		goto out;

	data.inode = inode;
	data.req = NULL;
	data.ff = NULL;

1921
	err = -ENOMEM;
1922 1923
	data.orig_pages = kcalloc(FUSE_MAX_PAGES_PER_REQ,
				  sizeof(struct page *),
1924 1925 1926 1927
				  GFP_NOFS);
	if (!data.orig_pages)
		goto out;

1928 1929 1930 1931 1932 1933 1934 1935
	err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data);
	if (data.req) {
		/* Ignore errors if we can write at least one page */
		BUG_ON(!data.req->num_pages);
		fuse_writepages_send(&data);
		err = 0;
	}
	if (data.ff)
1936
		fuse_file_put(data.ff, false, false);
1937 1938

	kfree(data.orig_pages);
1939 1940 1941 1942
out:
	return err;
}

1943 1944 1945 1946 1947 1948 1949 1950
/*
 * It's worthy to make sure that space is reserved on disk for the write,
 * but how to implement it without killing performance need more thinking.
 */
static int fuse_write_begin(struct file *file, struct address_space *mapping,
		loff_t pos, unsigned len, unsigned flags,
		struct page **pagep, void **fsdata)
{
1951
	pgoff_t index = pos >> PAGE_SHIFT;
A
Al Viro 已提交
1952
	struct fuse_conn *fc = get_fuse_conn(file_inode(file));
1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964
	struct page *page;
	loff_t fsize;
	int err = -ENOMEM;

	WARN_ON(!fc->writeback_cache);

	page = grab_cache_page_write_begin(mapping, index, flags);
	if (!page)
		goto error;

	fuse_wait_on_page_writeback(mapping->host, page->index);

1965
	if (PageUptodate(page) || len == PAGE_SIZE)
1966 1967 1968 1969 1970 1971
		goto success;
	/*
	 * Check if the start this page comes after the end of file, in which
	 * case the readpage can be optimized away.
	 */
	fsize = i_size_read(mapping->host);
1972 1973
	if (fsize <= (pos & PAGE_MASK)) {
		size_t off = pos & ~PAGE_MASK;
1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986
		if (off)
			zero_user_segment(page, 0, off);
		goto success;
	}
	err = fuse_do_readpage(file, page);
	if (err)
		goto cleanup;
success:
	*pagep = page;
	return 0;

cleanup:
	unlock_page(page);
1987
	put_page(page);
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997
error:
	return err;
}

static int fuse_write_end(struct file *file, struct address_space *mapping,
		loff_t pos, unsigned len, unsigned copied,
		struct page *page, void *fsdata)
{
	struct inode *inode = page->mapping->host;

1998 1999 2000 2001
	/* Haven't copied anything?  Skip zeroing, size extending, dirtying. */
	if (!copied)
		goto unlock;

2002 2003
	if (!PageUptodate(page)) {
		/* Zero any unwritten bytes at the end of the page */
2004
		size_t endoff = (pos + copied) & ~PAGE_MASK;
2005
		if (endoff)
2006
			zero_user_segment(page, endoff, PAGE_SIZE);
2007 2008 2009 2010 2011
		SetPageUptodate(page);
	}

	fuse_write_update_size(inode, pos + copied);
	set_page_dirty(page);
2012 2013

unlock:
2014
	unlock_page(page);
2015
	put_page(page);
2016 2017 2018 2019

	return copied;
}

M
Miklos Szeredi 已提交
2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055
static int fuse_launder_page(struct page *page)
{
	int err = 0;
	if (clear_page_dirty_for_io(page)) {
		struct inode *inode = page->mapping->host;
		err = fuse_writepage_locked(page);
		if (!err)
			fuse_wait_on_page_writeback(inode, page->index);
	}
	return err;
}

/*
 * Write back dirty pages now, because there may not be any suitable
 * open files later
 */
static void fuse_vma_close(struct vm_area_struct *vma)
{
	filemap_write_and_wait(vma->vm_file->f_mapping);
}

/*
 * Wait for writeback against this page to complete before allowing it
 * to be marked dirty again, and hence written back again, possibly
 * before the previous writepage completed.
 *
 * Block here, instead of in ->writepage(), so that the userspace fs
 * can only block processes actually operating on the filesystem.
 *
 * Otherwise unprivileged userspace fs would be able to block
 * unrelated:
 *
 * - page migration
 * - sync(2)
 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
 */
2056
static vm_fault_t fuse_page_mkwrite(struct vm_fault *vmf)
M
Miklos Szeredi 已提交
2057
{
2058
	struct page *page = vmf->page;
2059
	struct inode *inode = file_inode(vmf->vma->vm_file);
M
Miklos Szeredi 已提交
2060

2061
	file_update_time(vmf->vma->vm_file);
M
Miklos Szeredi 已提交
2062 2063 2064 2065 2066
	lock_page(page);
	if (page->mapping != inode->i_mapping) {
		unlock_page(page);
		return VM_FAULT_NOPAGE;
	}
M
Miklos Szeredi 已提交
2067 2068

	fuse_wait_on_page_writeback(inode, page->index);
M
Miklos Szeredi 已提交
2069
	return VM_FAULT_LOCKED;
M
Miklos Szeredi 已提交
2070 2071
}

2072
static const struct vm_operations_struct fuse_file_vm_ops = {
M
Miklos Szeredi 已提交
2073 2074
	.close		= fuse_vma_close,
	.fault		= filemap_fault,
2075
	.map_pages	= filemap_map_pages,
M
Miklos Szeredi 已提交
2076 2077 2078 2079 2080
	.page_mkwrite	= fuse_page_mkwrite,
};

static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
{
2081 2082 2083
	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
		fuse_link_write_file(file);

M
Miklos Szeredi 已提交
2084 2085
	file_accessed(file);
	vma->vm_ops = &fuse_file_vm_ops;
M
Miklos Szeredi 已提交
2086 2087 2088
	return 0;
}

2089 2090 2091 2092 2093 2094
static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma)
{
	/* Can't provide the coherency needed for MAP_SHARED */
	if (vma->vm_flags & VM_MAYSHARE)
		return -ENODEV;

2095 2096
	invalidate_inode_pages2(file->f_mapping);

2097 2098 2099
	return generic_file_mmap(file, vma);
}

2100 2101
static int convert_fuse_file_lock(struct fuse_conn *fc,
				  const struct fuse_file_lock *ffl,
2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115
				  struct file_lock *fl)
{
	switch (ffl->type) {
	case F_UNLCK:
		break;

	case F_RDLCK:
	case F_WRLCK:
		if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
		    ffl->end < ffl->start)
			return -EIO;

		fl->fl_start = ffl->start;
		fl->fl_end = ffl->end;
2116 2117

		/*
2118 2119
		 * Convert pid into init's pid namespace.  The locks API will
		 * translate it into the caller's pid namespace.
2120 2121
		 */
		rcu_read_lock();
2122
		fl->fl_pid = pid_nr_ns(find_pid_ns(ffl->pid, fc->pid_ns), &init_pid_ns);
2123
		rcu_read_unlock();
2124 2125 2126 2127 2128 2129 2130 2131 2132
		break;

	default:
		return -EIO;
	}
	fl->fl_type = ffl->type;
	return 0;
}

2133
static void fuse_lk_fill(struct fuse_args *args, struct file *file,
2134
			 const struct file_lock *fl, int opcode, pid_t pid,
2135
			 int flock, struct fuse_lk_in *inarg)
2136
{
A
Al Viro 已提交
2137
	struct inode *inode = file_inode(file);
2138
	struct fuse_conn *fc = get_fuse_conn(inode);
2139
	struct fuse_file *ff = file->private_data;
2140 2141 2142 2143 2144 2145 2146 2147

	memset(inarg, 0, sizeof(*inarg));
	inarg->fh = ff->fh;
	inarg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
	inarg->lk.start = fl->fl_start;
	inarg->lk.end = fl->fl_end;
	inarg->lk.type = fl->fl_type;
	inarg->lk.pid = pid;
2148
	if (flock)
2149 2150 2151 2152 2153 2154
		inarg->lk_flags |= FUSE_LK_FLOCK;
	args->in.h.opcode = opcode;
	args->in.h.nodeid = get_node_id(inode);
	args->in.numargs = 1;
	args->in.args[0].size = sizeof(*inarg);
	args->in.args[0].value = inarg;
2155 2156 2157 2158
}

static int fuse_getlk(struct file *file, struct file_lock *fl)
{
A
Al Viro 已提交
2159
	struct inode *inode = file_inode(file);
2160
	struct fuse_conn *fc = get_fuse_conn(inode);
2161 2162
	FUSE_ARGS(args);
	struct fuse_lk_in inarg;
2163 2164 2165
	struct fuse_lk_out outarg;
	int err;

2166 2167 2168 2169 2170
	fuse_lk_fill(&args, file, fl, FUSE_GETLK, 0, 0, &inarg);
	args.out.numargs = 1;
	args.out.args[0].size = sizeof(outarg);
	args.out.args[0].value = &outarg;
	err = fuse_simple_request(fc, &args);
2171
	if (!err)
2172
		err = convert_fuse_file_lock(fc, &outarg.lk, fl);
2173 2174 2175 2176

	return err;
}

2177
static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
2178
{
A
Al Viro 已提交
2179
	struct inode *inode = file_inode(file);
2180
	struct fuse_conn *fc = get_fuse_conn(inode);
2181 2182
	FUSE_ARGS(args);
	struct fuse_lk_in inarg;
2183
	int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
2184 2185
	struct pid *pid = fl->fl_type != F_UNLCK ? task_tgid(current) : NULL;
	pid_t pid_nr = pid_nr_ns(pid, fc->pid_ns);
2186 2187
	int err;

J
J. Bruce Fields 已提交
2188
	if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
M
Miklos Szeredi 已提交
2189 2190 2191 2192
		/* NLM needs asynchronous locks, which we don't support yet */
		return -ENOLCK;
	}

2193
	/* Unlock on close is handled by the flush method */
2194
	if ((fl->fl_flags & FL_CLOSE_POSIX) == FL_CLOSE_POSIX)
2195 2196
		return 0;

2197
	fuse_lk_fill(&args, file, fl, opcode, pid_nr, flock, &inarg);
2198
	err = fuse_simple_request(fc, &args);
2199

2200 2201 2202
	/* locking is restartable */
	if (err == -EINTR)
		err = -ERESTARTSYS;
2203

2204 2205 2206 2207 2208
	return err;
}

static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
{
A
Al Viro 已提交
2209
	struct inode *inode = file_inode(file);
2210 2211 2212
	struct fuse_conn *fc = get_fuse_conn(inode);
	int err;

M
Miklos Szeredi 已提交
2213 2214 2215
	if (cmd == F_CANCELLK) {
		err = 0;
	} else if (cmd == F_GETLK) {
2216
		if (fc->no_lock) {
2217
			posix_test_lock(file, fl);
2218 2219 2220 2221 2222
			err = 0;
		} else
			err = fuse_getlk(file, fl);
	} else {
		if (fc->no_lock)
M
Miklos Szeredi 已提交
2223
			err = posix_lock_file(file, fl, NULL);
2224
		else
2225
			err = fuse_setlk(file, fl, 0);
2226 2227 2228 2229
	}
	return err;
}

2230 2231
static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
{
A
Al Viro 已提交
2232
	struct inode *inode = file_inode(file);
2233 2234 2235
	struct fuse_conn *fc = get_fuse_conn(inode);
	int err;

M
Miklos Szeredi 已提交
2236
	if (fc->no_flock) {
2237
		err = locks_lock_file_wait(file, fl);
2238
	} else {
M
Miklos Szeredi 已提交
2239 2240
		struct fuse_file *ff = file->private_data;

2241
		/* emulate flock with POSIX locks */
M
Miklos Szeredi 已提交
2242
		ff->flock = true;
2243 2244 2245 2246 2247 2248
		err = fuse_setlk(file, fl, 1);
	}

	return err;
}

M
Miklos Szeredi 已提交
2249 2250 2251 2252
static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
{
	struct inode *inode = mapping->host;
	struct fuse_conn *fc = get_fuse_conn(inode);
2253
	FUSE_ARGS(args);
M
Miklos Szeredi 已提交
2254 2255 2256 2257 2258 2259 2260 2261 2262 2263
	struct fuse_bmap_in inarg;
	struct fuse_bmap_out outarg;
	int err;

	if (!inode->i_sb->s_bdev || fc->no_bmap)
		return 0;

	memset(&inarg, 0, sizeof(inarg));
	inarg.block = block;
	inarg.blocksize = inode->i_sb->s_blocksize;
2264 2265 2266 2267 2268 2269 2270 2271 2272
	args.in.h.opcode = FUSE_BMAP;
	args.in.h.nodeid = get_node_id(inode);
	args.in.numargs = 1;
	args.in.args[0].size = sizeof(inarg);
	args.in.args[0].value = &inarg;
	args.out.numargs = 1;
	args.out.args[0].size = sizeof(outarg);
	args.out.args[0].value = &outarg;
	err = fuse_simple_request(fc, &args);
M
Miklos Szeredi 已提交
2273 2274 2275 2276 2277 2278
	if (err == -ENOSYS)
		fc->no_bmap = 1;

	return err ? 0 : outarg.block;
}

2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315
static loff_t fuse_lseek(struct file *file, loff_t offset, int whence)
{
	struct inode *inode = file->f_mapping->host;
	struct fuse_conn *fc = get_fuse_conn(inode);
	struct fuse_file *ff = file->private_data;
	FUSE_ARGS(args);
	struct fuse_lseek_in inarg = {
		.fh = ff->fh,
		.offset = offset,
		.whence = whence
	};
	struct fuse_lseek_out outarg;
	int err;

	if (fc->no_lseek)
		goto fallback;

	args.in.h.opcode = FUSE_LSEEK;
	args.in.h.nodeid = ff->nodeid;
	args.in.numargs = 1;
	args.in.args[0].size = sizeof(inarg);
	args.in.args[0].value = &inarg;
	args.out.numargs = 1;
	args.out.args[0].size = sizeof(outarg);
	args.out.args[0].value = &outarg;
	err = fuse_simple_request(fc, &args);
	if (err) {
		if (err == -ENOSYS) {
			fc->no_lseek = 1;
			goto fallback;
		}
		return err;
	}

	return vfs_setpos(file, outarg.offset, inode->i_sb->s_maxbytes);

fallback:
M
Miklos Szeredi 已提交
2316
	err = fuse_update_attributes(inode, file);
2317 2318 2319 2320 2321 2322
	if (!err)
		return generic_file_llseek(file, offset, whence);
	else
		return err;
}

2323
static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence)
M
Miklos Szeredi 已提交
2324 2325
{
	loff_t retval;
A
Al Viro 已提交
2326
	struct inode *inode = file_inode(file);
M
Miklos Szeredi 已提交
2327

2328 2329 2330 2331
	switch (whence) {
	case SEEK_SET:
	case SEEK_CUR:
		 /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */
2332
		retval = generic_file_llseek(file, offset, whence);
2333 2334
		break;
	case SEEK_END:
A
Al Viro 已提交
2335
		inode_lock(inode);
M
Miklos Szeredi 已提交
2336
		retval = fuse_update_attributes(inode, file);
2337 2338
		if (!retval)
			retval = generic_file_llseek(file, offset, whence);
A
Al Viro 已提交
2339
		inode_unlock(inode);
2340 2341 2342
		break;
	case SEEK_HOLE:
	case SEEK_DATA:
A
Al Viro 已提交
2343
		inode_lock(inode);
2344
		retval = fuse_lseek(file, offset, whence);
A
Al Viro 已提交
2345
		inode_unlock(inode);
2346 2347 2348 2349
		break;
	default:
		retval = -EINVAL;
	}
2350

M
Miklos Szeredi 已提交
2351 2352 2353
	return retval;
}

2354 2355 2356 2357 2358 2359
/*
 * CUSE servers compiled on 32bit broke on 64bit kernels because the
 * ABI was defined to be 'struct iovec' which is different on 32bit
 * and 64bit.  Fortunately we can determine which structure the server
 * used from the size of the reply.
 */
M
Miklos Szeredi 已提交
2360 2361 2362
static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
				     size_t transferred, unsigned count,
				     bool is_compat)
2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391
{
#ifdef CONFIG_COMPAT
	if (count * sizeof(struct compat_iovec) == transferred) {
		struct compat_iovec *ciov = src;
		unsigned i;

		/*
		 * With this interface a 32bit server cannot support
		 * non-compat (i.e. ones coming from 64bit apps) ioctl
		 * requests
		 */
		if (!is_compat)
			return -EINVAL;

		for (i = 0; i < count; i++) {
			dst[i].iov_base = compat_ptr(ciov[i].iov_base);
			dst[i].iov_len = ciov[i].iov_len;
		}
		return 0;
	}
#endif

	if (count * sizeof(struct iovec) != transferred)
		return -EIO;

	memcpy(dst, src, transferred);
	return 0;
}

M
Miklos Szeredi 已提交
2392 2393 2394 2395 2396 2397
/* Make sure iov_length() won't overflow */
static int fuse_verify_ioctl_iov(struct iovec *iov, size_t count)
{
	size_t n;
	u32 max = FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT;

2398
	for (n = 0; n < count; n++, iov++) {
M
Miklos Szeredi 已提交
2399 2400 2401 2402 2403 2404 2405
		if (iov->iov_len > (size_t) max)
			return -ENOMEM;
		max -= iov->iov_len;
	}
	return 0;
}

M
Miklos Szeredi 已提交
2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441
static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
				 void *src, size_t transferred, unsigned count,
				 bool is_compat)
{
	unsigned i;
	struct fuse_ioctl_iovec *fiov = src;

	if (fc->minor < 16) {
		return fuse_copy_ioctl_iovec_old(dst, src, transferred,
						 count, is_compat);
	}

	if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
		return -EIO;

	for (i = 0; i < count; i++) {
		/* Did the server supply an inappropriate value? */
		if (fiov[i].base != (unsigned long) fiov[i].base ||
		    fiov[i].len != (unsigned long) fiov[i].len)
			return -EIO;

		dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
		dst[i].iov_len = (size_t) fiov[i].len;

#ifdef CONFIG_COMPAT
		if (is_compat &&
		    (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
		     (compat_size_t) dst[i].iov_len != fiov[i].len))
			return -EIO;
#endif
	}

	return 0;
}


T
Tejun Heo 已提交
2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487
/*
 * For ioctls, there is no generic way to determine how much memory
 * needs to be read and/or written.  Furthermore, ioctls are allowed
 * to dereference the passed pointer, so the parameter requires deep
 * copying but FUSE has no idea whatsoever about what to copy in or
 * out.
 *
 * This is solved by allowing FUSE server to retry ioctl with
 * necessary in/out iovecs.  Let's assume the ioctl implementation
 * needs to read in the following structure.
 *
 * struct a {
 *	char	*buf;
 *	size_t	buflen;
 * }
 *
 * On the first callout to FUSE server, inarg->in_size and
 * inarg->out_size will be NULL; then, the server completes the ioctl
 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
 * the actual iov array to
 *
 * { { .iov_base = inarg.arg,	.iov_len = sizeof(struct a) } }
 *
 * which tells FUSE to copy in the requested area and retry the ioctl.
 * On the second round, the server has access to the structure and
 * from that it can tell what to look for next, so on the invocation,
 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
 *
 * { { .iov_base = inarg.arg,	.iov_len = sizeof(struct a)	},
 *   { .iov_base = a.buf,	.iov_len = a.buflen		} }
 *
 * FUSE will copy both struct a and the pointed buffer from the
 * process doing the ioctl and retry ioctl with both struct a and the
 * buffer.
 *
 * This time, FUSE server has everything it needs and completes ioctl
 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
 *
 * Copying data out works the same way.
 *
 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
 * automatically initializes in and out iovs by decoding @cmd with
 * _IOC_* macros and the server is not allowed to request RETRY.  This
 * limits ioctl data transfers to well-formed ioctls and is the forced
 * behavior for all FUSE servers.
 */
2488 2489
long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
		   unsigned int flags)
T
Tejun Heo 已提交
2490 2491
{
	struct fuse_file *ff = file->private_data;
2492
	struct fuse_conn *fc = ff->fc;
T
Tejun Heo 已提交
2493 2494 2495 2496 2497 2498 2499 2500 2501
	struct fuse_ioctl_in inarg = {
		.fh = ff->fh,
		.cmd = cmd,
		.arg = arg,
		.flags = flags
	};
	struct fuse_ioctl_out outarg;
	struct fuse_req *req = NULL;
	struct page **pages = NULL;
M
Miklos Szeredi 已提交
2502
	struct iovec *iov_page = NULL;
T
Tejun Heo 已提交
2503 2504
	struct iovec *in_iov = NULL, *out_iov = NULL;
	unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages;
2505 2506 2507
	size_t in_size, out_size, transferred, c;
	int err, i;
	struct iov_iter ii;
T
Tejun Heo 已提交
2508

M
Miklos Szeredi 已提交
2509 2510 2511 2512 2513 2514 2515
#if BITS_PER_LONG == 32
	inarg.flags |= FUSE_IOCTL_32BIT;
#else
	if (flags & FUSE_IOCTL_COMPAT)
		inarg.flags |= FUSE_IOCTL_32BIT;
#endif

T
Tejun Heo 已提交
2516
	/* assume all the iovs returned by client always fits in a page */
M
Miklos Szeredi 已提交
2517
	BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
T
Tejun Heo 已提交
2518 2519

	err = -ENOMEM;
2520
	pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, sizeof(pages[0]), GFP_KERNEL);
M
Miklos Szeredi 已提交
2521
	iov_page = (struct iovec *) __get_free_page(GFP_KERNEL);
T
Tejun Heo 已提交
2522 2523 2524 2525 2526 2527 2528 2529
	if (!pages || !iov_page)
		goto out;

	/*
	 * If restricted, initialize IO parameters as encoded in @cmd.
	 * RETRY from server is not allowed.
	 */
	if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
M
Miklos Szeredi 已提交
2530
		struct iovec *iov = iov_page;
T
Tejun Heo 已提交
2531

M
Miklos Szeredi 已提交
2532
		iov->iov_base = (void __user *)arg;
T
Tejun Heo 已提交
2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567
		iov->iov_len = _IOC_SIZE(cmd);

		if (_IOC_DIR(cmd) & _IOC_WRITE) {
			in_iov = iov;
			in_iovs = 1;
		}

		if (_IOC_DIR(cmd) & _IOC_READ) {
			out_iov = iov;
			out_iovs = 1;
		}
	}

 retry:
	inarg.in_size = in_size = iov_length(in_iov, in_iovs);
	inarg.out_size = out_size = iov_length(out_iov, out_iovs);

	/*
	 * Out data can be used either for actual out data or iovs,
	 * make sure there always is at least one page.
	 */
	out_size = max_t(size_t, out_size, PAGE_SIZE);
	max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);

	/* make sure there are enough buffer pages and init request with them */
	err = -ENOMEM;
	if (max_pages > FUSE_MAX_PAGES_PER_REQ)
		goto out;
	while (num_pages < max_pages) {
		pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
		if (!pages[num_pages])
			goto out;
		num_pages++;
	}

M
Maxim Patlasov 已提交
2568
	req = fuse_get_req(fc, num_pages);
T
Tejun Heo 已提交
2569 2570 2571 2572 2573 2574 2575
	if (IS_ERR(req)) {
		err = PTR_ERR(req);
		req = NULL;
		goto out;
	}
	memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages);
	req->num_pages = num_pages;
2576
	fuse_page_descs_length_init(req, 0, req->num_pages);
T
Tejun Heo 已提交
2577 2578 2579

	/* okay, let's send it to the client */
	req->in.h.opcode = FUSE_IOCTL;
2580
	req->in.h.nodeid = ff->nodeid;
T
Tejun Heo 已提交
2581 2582 2583 2584 2585 2586 2587 2588
	req->in.numargs = 1;
	req->in.args[0].size = sizeof(inarg);
	req->in.args[0].value = &inarg;
	if (in_size) {
		req->in.numargs++;
		req->in.args[1].size = in_size;
		req->in.argpages = 1;

2589 2590 2591 2592 2593 2594 2595
		err = -EFAULT;
		iov_iter_init(&ii, WRITE, in_iov, in_iovs, in_size);
		for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= num_pages); i++) {
			c = copy_page_from_iter(pages[i], 0, PAGE_SIZE, &ii);
			if (c != PAGE_SIZE && iov_iter_count(&ii))
				goto out;
		}
T
Tejun Heo 已提交
2596 2597 2598 2599 2600 2601 2602 2603 2604
	}

	req->out.numargs = 2;
	req->out.args[0].size = sizeof(outarg);
	req->out.args[0].value = &outarg;
	req->out.args[1].size = out_size;
	req->out.argpages = 1;
	req->out.argvar = 1;

2605
	fuse_request_send(fc, req);
T
Tejun Heo 已提交
2606 2607 2608 2609 2610 2611 2612 2613 2614
	err = req->out.h.error;
	transferred = req->out.args[1].size;
	fuse_put_request(fc, req);
	req = NULL;
	if (err)
		goto out;

	/* did it ask for retry? */
	if (outarg.flags & FUSE_IOCTL_RETRY) {
M
Miklos Szeredi 已提交
2615
		void *vaddr;
T
Tejun Heo 已提交
2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634

		/* no retry if in restricted mode */
		err = -EIO;
		if (!(flags & FUSE_IOCTL_UNRESTRICTED))
			goto out;

		in_iovs = outarg.in_iovs;
		out_iovs = outarg.out_iovs;

		/*
		 * Make sure things are in boundary, separate checks
		 * are to protect against overflow.
		 */
		err = -ENOMEM;
		if (in_iovs > FUSE_IOCTL_MAX_IOV ||
		    out_iovs > FUSE_IOCTL_MAX_IOV ||
		    in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
			goto out;

2635
		vaddr = kmap_atomic(pages[0]);
M
Miklos Szeredi 已提交
2636
		err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr,
2637 2638
					    transferred, in_iovs + out_iovs,
					    (flags & FUSE_IOCTL_COMPAT) != 0);
2639
		kunmap_atomic(vaddr);
2640 2641
		if (err)
			goto out;
T
Tejun Heo 已提交
2642

M
Miklos Szeredi 已提交
2643
		in_iov = iov_page;
T
Tejun Heo 已提交
2644 2645
		out_iov = in_iov + in_iovs;

M
Miklos Szeredi 已提交
2646 2647 2648 2649 2650 2651 2652 2653
		err = fuse_verify_ioctl_iov(in_iov, in_iovs);
		if (err)
			goto out;

		err = fuse_verify_ioctl_iov(out_iov, out_iovs);
		if (err)
			goto out;

T
Tejun Heo 已提交
2654 2655 2656 2657 2658 2659 2660
		goto retry;
	}

	err = -EIO;
	if (transferred > inarg.out_size)
		goto out;

2661 2662 2663 2664 2665 2666 2667 2668
	err = -EFAULT;
	iov_iter_init(&ii, READ, out_iov, out_iovs, transferred);
	for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= num_pages); i++) {
		c = copy_page_to_iter(pages[i], 0, PAGE_SIZE, &ii);
		if (c != PAGE_SIZE && iov_iter_count(&ii))
			goto out;
	}
	err = 0;
T
Tejun Heo 已提交
2669 2670 2671
 out:
	if (req)
		fuse_put_request(fc, req);
M
Miklos Szeredi 已提交
2672
	free_page((unsigned long) iov_page);
T
Tejun Heo 已提交
2673 2674 2675 2676 2677 2678
	while (num_pages)
		__free_page(pages[--num_pages]);
	kfree(pages);

	return err ? err : outarg.result;
}
2679
EXPORT_SYMBOL_GPL(fuse_do_ioctl);
T
Tejun Heo 已提交
2680

2681 2682
long fuse_ioctl_common(struct file *file, unsigned int cmd,
		       unsigned long arg, unsigned int flags)
2683
{
A
Al Viro 已提交
2684
	struct inode *inode = file_inode(file);
2685 2686
	struct fuse_conn *fc = get_fuse_conn(inode);

2687
	if (!fuse_allow_current_process(fc))
2688 2689 2690 2691 2692 2693 2694 2695
		return -EACCES;

	if (is_bad_inode(inode))
		return -EIO;

	return fuse_do_ioctl(file, cmd, arg, flags);
}

T
Tejun Heo 已提交
2696 2697 2698
static long fuse_file_ioctl(struct file *file, unsigned int cmd,
			    unsigned long arg)
{
2699
	return fuse_ioctl_common(file, cmd, arg, 0);
T
Tejun Heo 已提交
2700 2701 2702 2703 2704
}

static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
				   unsigned long arg)
{
2705
	return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
T
Tejun Heo 已提交
2706 2707
}

T
Tejun Heo 已提交
2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748
/*
 * All files which have been polled are linked to RB tree
 * fuse_conn->polled_files which is indexed by kh.  Walk the tree and
 * find the matching one.
 */
static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh,
					      struct rb_node **parent_out)
{
	struct rb_node **link = &fc->polled_files.rb_node;
	struct rb_node *last = NULL;

	while (*link) {
		struct fuse_file *ff;

		last = *link;
		ff = rb_entry(last, struct fuse_file, polled_node);

		if (kh < ff->kh)
			link = &last->rb_left;
		else if (kh > ff->kh)
			link = &last->rb_right;
		else
			return link;
	}

	if (parent_out)
		*parent_out = last;
	return link;
}

/*
 * The file is about to be polled.  Make sure it's on the polled_files
 * RB tree.  Note that files once added to the polled_files tree are
 * not removed before the file is released.  This is because a file
 * polled once is likely to be polled again.
 */
static void fuse_register_polled_file(struct fuse_conn *fc,
				      struct fuse_file *ff)
{
	spin_lock(&fc->lock);
	if (RB_EMPTY_NODE(&ff->polled_node)) {
2749
		struct rb_node **link, *uninitialized_var(parent);
T
Tejun Heo 已提交
2750 2751 2752 2753 2754 2755 2756 2757 2758

		link = fuse_find_polled_node(fc, ff->kh, &parent);
		BUG_ON(*link);
		rb_link_node(&ff->polled_node, parent, link);
		rb_insert_color(&ff->polled_node, &fc->polled_files);
	}
	spin_unlock(&fc->lock);
}

A
Al Viro 已提交
2759
__poll_t fuse_file_poll(struct file *file, poll_table *wait)
T
Tejun Heo 已提交
2760 2761
{
	struct fuse_file *ff = file->private_data;
2762
	struct fuse_conn *fc = ff->fc;
T
Tejun Heo 已提交
2763 2764
	struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh };
	struct fuse_poll_out outarg;
2765
	FUSE_ARGS(args);
T
Tejun Heo 已提交
2766 2767 2768 2769 2770 2771
	int err;

	if (fc->no_poll)
		return DEFAULT_POLLMASK;

	poll_wait(file, &ff->poll_wait, wait);
2772
	inarg.events = mangle_poll(poll_requested_events(wait));
T
Tejun Heo 已提交
2773 2774 2775 2776 2777 2778 2779 2780 2781 2782

	/*
	 * Ask for notification iff there's someone waiting for it.
	 * The client may ignore the flag and always notify.
	 */
	if (waitqueue_active(&ff->poll_wait)) {
		inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY;
		fuse_register_polled_file(fc, ff);
	}

2783 2784 2785 2786 2787 2788 2789 2790 2791
	args.in.h.opcode = FUSE_POLL;
	args.in.h.nodeid = ff->nodeid;
	args.in.numargs = 1;
	args.in.args[0].size = sizeof(inarg);
	args.in.args[0].value = &inarg;
	args.out.numargs = 1;
	args.out.args[0].size = sizeof(outarg);
	args.out.args[0].value = &outarg;
	err = fuse_simple_request(fc, &args);
T
Tejun Heo 已提交
2792 2793

	if (!err)
2794
		return demangle_poll(outarg.revents);
T
Tejun Heo 已提交
2795 2796 2797 2798
	if (err == -ENOSYS) {
		fc->no_poll = 1;
		return DEFAULT_POLLMASK;
	}
2799
	return EPOLLERR;
T
Tejun Heo 已提交
2800
}
2801
EXPORT_SYMBOL_GPL(fuse_file_poll);
T
Tejun Heo 已提交
2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826

/*
 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
 * wakes up the poll waiters.
 */
int fuse_notify_poll_wakeup(struct fuse_conn *fc,
			    struct fuse_notify_poll_wakeup_out *outarg)
{
	u64 kh = outarg->kh;
	struct rb_node **link;

	spin_lock(&fc->lock);

	link = fuse_find_polled_node(fc, kh, NULL);
	if (*link) {
		struct fuse_file *ff;

		ff = rb_entry(*link, struct fuse_file, polled_node);
		wake_up_interruptible_sync(&ff->poll_wait);
	}

	spin_unlock(&fc->lock);
	return 0;
}

2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837
static void fuse_do_truncate(struct file *file)
{
	struct inode *inode = file->f_mapping->host;
	struct iattr attr;

	attr.ia_valid = ATTR_SIZE;
	attr.ia_size = i_size_read(inode);

	attr.ia_file = file;
	attr.ia_valid |= ATTR_FILE;

2838
	fuse_do_setattr(file_dentry(file), &attr, file);
2839 2840
}

2841 2842 2843 2844 2845
static inline loff_t fuse_round_up(loff_t off)
{
	return round_up(off, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
}

A
Anand Avati 已提交
2846
static ssize_t
2847
fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
A
Anand Avati 已提交
2848
{
2849
	DECLARE_COMPLETION_ONSTACK(wait);
A
Anand Avati 已提交
2850
	ssize_t ret = 0;
2851 2852
	struct file *file = iocb->ki_filp;
	struct fuse_file *ff = file->private_data;
2853
	bool async_dio = ff->fc->async_dio;
A
Anand Avati 已提交
2854
	loff_t pos = 0;
2855 2856
	struct inode *inode;
	loff_t i_size;
2857
	size_t count = iov_iter_count(iter);
2858
	loff_t offset = iocb->ki_pos;
2859
	struct fuse_io_priv *io;
A
Anand Avati 已提交
2860 2861

	pos = offset;
2862 2863
	inode = file->f_mapping->host;
	i_size = i_size_read(inode);
A
Anand Avati 已提交
2864

2865
	if ((iov_iter_rw(iter) == READ) && (offset > i_size))
2866 2867
		return 0;

2868
	/* optimization for short read */
2869
	if (async_dio && iov_iter_rw(iter) != WRITE && offset + count > i_size) {
2870 2871
		if (offset >= i_size)
			return 0;
2872 2873
		iov_iter_truncate(iter, fuse_round_up(i_size - offset));
		count = iov_iter_count(iter);
2874 2875
	}

2876
	io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL);
2877 2878
	if (!io)
		return -ENOMEM;
2879
	spin_lock_init(&io->lock);
2880
	kref_init(&io->refcnt);
2881 2882 2883 2884
	io->reqs = 1;
	io->bytes = -1;
	io->size = 0;
	io->offset = offset;
2885
	io->write = (iov_iter_rw(iter) == WRITE);
2886 2887 2888
	io->err = 0;
	/*
	 * By default, we want to optimize all I/Os with async request
2889
	 * submission to the client filesystem if supported.
2890
	 */
2891
	io->async = async_dio;
2892
	io->iocb = iocb;
2893
	io->blocking = is_sync_kiocb(iocb);
2894 2895

	/*
2896 2897
	 * We cannot asynchronously extend the size of a file.
	 * In such case the aio will behave exactly like sync io.
2898
	 */
2899 2900
	if ((offset + count > i_size) && iov_iter_rw(iter) == WRITE)
		io->blocking = true;
A
Anand Avati 已提交
2901

2902
	if (io->async && io->blocking) {
2903 2904 2905 2906 2907
		/*
		 * Additional reference to keep io around after
		 * calling fuse_aio_complete()
		 */
		kref_get(&io->refcnt);
2908
		io->done = &wait;
2909
	}
2910

2911
	if (iov_iter_rw(iter) == WRITE) {
2912
		ret = fuse_direct_io(io, iter, &pos, FUSE_DIO_WRITE);
2913 2914
		fuse_invalidate_attr(inode);
	} else {
A
Al Viro 已提交
2915
		ret = __fuse_direct_read(io, iter, &pos);
2916
	}
2917

2918
	if (io->async) {
2919 2920
		bool blocking = io->blocking;

2921 2922 2923
		fuse_aio_complete(io, ret < 0 ? ret : 0, -1);

		/* we have a non-extending, async request, so return */
2924
		if (!blocking)
2925 2926
			return -EIOCBQUEUED;

2927 2928
		wait_for_completion(&wait);
		ret = fuse_get_res_by_io(io);
2929 2930
	}

2931
	kref_put(&io->refcnt, fuse_io_release);
2932

2933
	if (iov_iter_rw(iter) == WRITE) {
2934 2935 2936 2937 2938
		if (ret > 0)
			fuse_write_update_size(inode, pos);
		else if (ret < 0 && offset + count > i_size)
			fuse_do_truncate(file);
	}
A
Anand Avati 已提交
2939 2940 2941 2942

	return ret;
}

2943 2944
static long fuse_file_fallocate(struct file *file, int mode, loff_t offset,
				loff_t length)
A
Anatol Pomozov 已提交
2945 2946
{
	struct fuse_file *ff = file->private_data;
2947
	struct inode *inode = file_inode(file);
2948
	struct fuse_inode *fi = get_fuse_inode(inode);
A
Anatol Pomozov 已提交
2949
	struct fuse_conn *fc = ff->fc;
2950
	FUSE_ARGS(args);
A
Anatol Pomozov 已提交
2951 2952 2953 2954 2955 2956 2957
	struct fuse_fallocate_in inarg = {
		.fh = ff->fh,
		.offset = offset,
		.length = length,
		.mode = mode
	};
	int err;
2958 2959
	bool lock_inode = !(mode & FALLOC_FL_KEEP_SIZE) ||
			   (mode & FALLOC_FL_PUNCH_HOLE);
A
Anatol Pomozov 已提交
2960

M
Miklos Szeredi 已提交
2961 2962 2963
	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
		return -EOPNOTSUPP;

2964 2965 2966
	if (fc->no_fallocate)
		return -EOPNOTSUPP;

2967
	if (lock_inode) {
A
Al Viro 已提交
2968
		inode_lock(inode);
2969 2970 2971 2972 2973 2974 2975 2976 2977
		if (mode & FALLOC_FL_PUNCH_HOLE) {
			loff_t endbyte = offset + length - 1;
			err = filemap_write_and_wait_range(inode->i_mapping,
							   offset, endbyte);
			if (err)
				goto out;

			fuse_sync_writes(inode);
		}
2978 2979
	}

2980 2981 2982 2983 2984 2985 2986
	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
	    offset + length > i_size_read(inode)) {
		err = inode_newsize_ok(inode, offset + length);
		if (err)
			return err;
	}

2987 2988 2989
	if (!(mode & FALLOC_FL_KEEP_SIZE))
		set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);

2990 2991 2992 2993 2994 2995
	args.in.h.opcode = FUSE_FALLOCATE;
	args.in.h.nodeid = ff->nodeid;
	args.in.numargs = 1;
	args.in.args[0].size = sizeof(inarg);
	args.in.args[0].value = &inarg;
	err = fuse_simple_request(fc, &args);
2996 2997 2998 2999
	if (err == -ENOSYS) {
		fc->no_fallocate = 1;
		err = -EOPNOTSUPP;
	}
3000 3001 3002 3003
	if (err)
		goto out;

	/* we could have extended the file */
M
Maxim Patlasov 已提交
3004 3005 3006
	if (!(mode & FALLOC_FL_KEEP_SIZE)) {
		bool changed = fuse_write_update_size(inode, offset + length);

3007 3008
		if (changed && fc->writeback_cache)
			file_update_time(file);
M
Maxim Patlasov 已提交
3009
	}
3010 3011 3012 3013 3014 3015

	if (mode & FALLOC_FL_PUNCH_HOLE)
		truncate_pagecache_range(inode, offset, offset + length - 1);

	fuse_invalidate_attr(inode);

3016
out:
3017 3018 3019
	if (!(mode & FALLOC_FL_KEEP_SIZE))
		clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);

3020
	if (lock_inode)
A
Al Viro 已提交
3021
		inode_unlock(inode);
3022

A
Anatol Pomozov 已提交
3023 3024 3025
	return err;
}

3026
static const struct file_operations fuse_file_operations = {
M
Miklos Szeredi 已提交
3027
	.llseek		= fuse_file_llseek,
3028
	.read_iter	= fuse_file_read_iter,
A
Al Viro 已提交
3029
	.write_iter	= fuse_file_write_iter,
M
Miklos Szeredi 已提交
3030 3031 3032 3033 3034
	.mmap		= fuse_file_mmap,
	.open		= fuse_open,
	.flush		= fuse_flush,
	.release	= fuse_release,
	.fsync		= fuse_fsync,
3035
	.lock		= fuse_file_lock,
3036
	.flock		= fuse_file_flock,
3037
	.splice_read	= generic_file_splice_read,
T
Tejun Heo 已提交
3038 3039
	.unlocked_ioctl	= fuse_file_ioctl,
	.compat_ioctl	= fuse_file_compat_ioctl,
T
Tejun Heo 已提交
3040
	.poll		= fuse_file_poll,
A
Anatol Pomozov 已提交
3041
	.fallocate	= fuse_file_fallocate,
M
Miklos Szeredi 已提交
3042 3043
};

3044
static const struct file_operations fuse_direct_io_file_operations = {
M
Miklos Szeredi 已提交
3045
	.llseek		= fuse_file_llseek,
3046 3047
	.read_iter	= fuse_direct_read_iter,
	.write_iter	= fuse_direct_write_iter,
3048
	.mmap		= fuse_direct_mmap,
M
Miklos Szeredi 已提交
3049 3050 3051 3052
	.open		= fuse_open,
	.flush		= fuse_flush,
	.release	= fuse_release,
	.fsync		= fuse_fsync,
3053
	.lock		= fuse_file_lock,
3054
	.flock		= fuse_file_flock,
T
Tejun Heo 已提交
3055 3056
	.unlocked_ioctl	= fuse_file_ioctl,
	.compat_ioctl	= fuse_file_compat_ioctl,
T
Tejun Heo 已提交
3057
	.poll		= fuse_file_poll,
A
Anatol Pomozov 已提交
3058
	.fallocate	= fuse_file_fallocate,
3059
	/* no splice_read */
M
Miklos Szeredi 已提交
3060 3061
};

3062
static const struct address_space_operations fuse_file_aops  = {
M
Miklos Szeredi 已提交
3063
	.readpage	= fuse_readpage,
M
Miklos Szeredi 已提交
3064
	.writepage	= fuse_writepage,
3065
	.writepages	= fuse_writepages,
M
Miklos Szeredi 已提交
3066
	.launder_page	= fuse_launder_page,
3067
	.readpages	= fuse_readpages,
M
Miklos Szeredi 已提交
3068
	.set_page_dirty	= __set_page_dirty_nobuffers,
M
Miklos Szeredi 已提交
3069
	.bmap		= fuse_bmap,
A
Anand Avati 已提交
3070
	.direct_IO	= fuse_direct_IO,
3071 3072
	.write_begin	= fuse_write_begin,
	.write_end	= fuse_write_end,
M
Miklos Szeredi 已提交
3073 3074 3075 3076
};

void fuse_init_file_inode(struct inode *inode)
{
3077 3078
	inode->i_fop = &fuse_file_operations;
	inode->i_data.a_ops = &fuse_file_aops;
M
Miklos Szeredi 已提交
3079
}