dir.c 47.5 KB
Newer Older
1 2
/*
  FUSE: Filesystem in Userspace
Miklos Szeredi's avatar
Miklos Szeredi committed
3
  Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
4 5 6 7 8 9 10 11 12 13 14

  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/file.h>
#include <linux/sched.h>
#include <linux/namei.h>
15
#include <linux/slab.h>
16

Al Viro's avatar
Al Viro committed
17
static bool fuse_use_readdirplus(struct inode *dir, struct dir_context *ctx)
18 19 20 21 22 23
{
	struct fuse_conn *fc = get_fuse_conn(dir);
	struct fuse_inode *fi = get_fuse_inode(dir);

	if (!fc->do_readdirplus)
		return false;
24 25
	if (!fc->readdirplus_auto)
		return true;
26 27
	if (test_and_clear_bit(FUSE_I_ADVISE_RDPLUS, &fi->state))
		return true;
Al Viro's avatar
Al Viro committed
28
	if (ctx->pos == 0)
29 30 31 32 33 34 35 36 37 38 39
		return true;
	return false;
}

static void fuse_advise_use_readdirplus(struct inode *dir)
{
	struct fuse_inode *fi = get_fuse_inode(dir);

	set_bit(FUSE_I_ADVISE_RDPLUS, &fi->state);
}

40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66
#if BITS_PER_LONG >= 64
static inline void fuse_dentry_settime(struct dentry *entry, u64 time)
{
	entry->d_time = time;
}

static inline u64 fuse_dentry_time(struct dentry *entry)
{
	return entry->d_time;
}
#else
/*
 * On 32 bit archs store the high 32 bits of time in d_fsdata
 */
static void fuse_dentry_settime(struct dentry *entry, u64 time)
{
	entry->d_time = time;
	entry->d_fsdata = (void *) (unsigned long) (time >> 32);
}

static u64 fuse_dentry_time(struct dentry *entry)
{
	return (u64) entry->d_time +
		((u64) (unsigned long) entry->d_fsdata << 32);
}
#endif

67 68 69 70 71 72 73 74 75
/*
 * FUSE caches dentries and attributes with separate timeout.  The
 * time in jiffies until the dentry/attributes are valid is stored in
 * dentry->d_time and fuse_inode->i_time respectively.
 */

/*
 * Calculate the time in jiffies until a dentry/attributes are valid
 */
76
static u64 time_to_jiffies(unsigned long sec, unsigned long nsec)
77
{
78 79
	if (sec || nsec) {
		struct timespec ts = {sec, nsec};
80
		return get_jiffies_64() + timespec_to_jiffies(&ts);
81
	} else
82
		return 0;
83 84
}

85 86 87 88
/*
 * Set dentry and possibly attribute timeouts from the lookup/mk*
 * replies
 */
89 90
static void fuse_change_entry_timeout(struct dentry *entry,
				      struct fuse_entry_out *o)
91
{
92 93
	fuse_dentry_settime(entry,
		time_to_jiffies(o->entry_valid, o->entry_valid_nsec));
94 95 96 97 98 99 100 101 102 103
}

static u64 attr_timeout(struct fuse_attr_out *o)
{
	return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
}

static u64 entry_attr_timeout(struct fuse_entry_out *o)
{
	return time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
104 105
}

106 107 108 109
/*
 * Mark the attributes as stale, so that at the next call to
 * ->getattr() they will be fetched from userspace
 */
110 111
void fuse_invalidate_attr(struct inode *inode)
{
112
	get_fuse_inode(inode)->i_time = 0;
113 114
}

115 116 117 118 119 120 121 122 123 124
/**
 * Mark the attributes as stale due to an atime change.  Avoid the invalidate if
 * atime is not used.
 */
void fuse_invalidate_atime(struct inode *inode)
{
	if (!IS_RDONLY(inode))
		fuse_invalidate_attr(inode);
}

125 126 127 128 129 130 131 132
/*
 * Just mark the entry as stale, so that a next attempt to look it up
 * will result in a new lookup call to userspace
 *
 * This is called when a dentry is about to become negative and the
 * timeout is unknown (unlink, rmdir, rename and in some cases
 * lookup)
 */
133
void fuse_invalidate_entry_cache(struct dentry *entry)
134
{
135
	fuse_dentry_settime(entry, 0);
136 137
}

138 139 140 141
/*
 * Same as fuse_invalidate_entry_cache(), but also try to remove the
 * dentry from the hash
 */
142 143 144 145
static void fuse_invalidate_entry(struct dentry *entry)
{
	d_invalidate(entry);
	fuse_invalidate_entry_cache(entry);
146 147
}

148
static void fuse_lookup_init(struct fuse_conn *fc, struct fuse_args *args,
149
			     u64 nodeid, const struct qstr *name,
150 151
			     struct fuse_entry_out *outarg)
{
152
	memset(outarg, 0, sizeof(struct fuse_entry_out));
153 154 155 156 157 158
	args->in.h.opcode = FUSE_LOOKUP;
	args->in.h.nodeid = nodeid;
	args->in.numargs = 1;
	args->in.args[0].size = name->len + 1;
	args->in.args[0].value = name->name;
	args->out.numargs = 1;
159
	args->out.args[0].size = sizeof(struct fuse_entry_out);
160
	args->out.args[0].value = outarg;
161 162
}

163
u64 fuse_get_attr_version(struct fuse_conn *fc)
164 165 166 167 168 169 170 171 172 173 174 175 176 177
{
	u64 curr_version;

	/*
	 * The spin lock isn't actually needed on 64bit archs, but we
	 * don't yet care too much about such optimizations.
	 */
	spin_lock(&fc->lock);
	curr_version = fc->attr_version;
	spin_unlock(&fc->lock);

	return curr_version;
}

178 179 180 181 182 183 184 185 186
/*
 * Check whether the dentry is still valid
 *
 * If the entry validity timeout has expired and the dentry is
 * positive, try to redo the lookup.  If the lookup results in a
 * different inode, then let the VFS invalidate the dentry and redo
 * the lookup once more.  If the lookup results in the same inode,
 * then refresh the attributes, timeouts and mark the dentry valid.
 */
187
static int fuse_dentry_revalidate(struct dentry *entry, unsigned int flags)
188
{
189
	struct inode *inode;
190 191
	struct dentry *parent;
	struct fuse_conn *fc;
192
	struct fuse_inode *fi;
193
	int ret;
194

195
	inode = d_inode_rcu(entry);
196
	if (inode && is_bad_inode(inode))
197
		goto invalid;
198 199
	else if (time_before64(fuse_dentry_time(entry), get_jiffies_64()) ||
		 (flags & LOOKUP_REVAL)) {
200
		struct fuse_entry_out outarg;
201
		FUSE_ARGS(args);
202
		struct fuse_forget_link *forget;
203
		u64 attr_version;
204

205
		/* For negative dentries, always do a fresh lookup */
206
		if (!inode)
207
			goto invalid;
208

209
		ret = -ECHILD;
210
		if (flags & LOOKUP_RCU)
211
			goto out;
212

213
		fc = get_fuse_conn(inode);
214

215
		forget = fuse_alloc_forget();
216 217
		ret = -ENOMEM;
		if (!forget)
218
			goto out;
219

220
		attr_version = fuse_get_attr_version(fc);
221

222
		parent = dget_parent(entry);
223
		fuse_lookup_init(fc, &args, get_node_id(d_inode(parent)),
224
				 &entry->d_name, &outarg);
225
		ret = fuse_simple_request(fc, &args);
226
		dput(parent);
227
		/* Zero nodeid is same as -ENOENT */
228 229 230
		if (!ret && !outarg.nodeid)
			ret = -ENOENT;
		if (!ret) {
231
			fi = get_fuse_inode(inode);
232
			if (outarg.nodeid != get_node_id(inode)) {
233
				fuse_queue_forget(fc, forget, outarg.nodeid, 1);
234
				goto invalid;
235
			}
236
			spin_lock(&fc->lock);
Miklos Szeredi's avatar
Miklos Szeredi committed
237
			fi->nlookup++;
238
			spin_unlock(&fc->lock);
239
		}
240
		kfree(forget);
241 242 243
		if (ret == -ENOMEM)
			goto out;
		if (ret || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
244
			goto invalid;
245

246 247 248 249
		fuse_change_attributes(inode, &outarg.attr,
				       entry_attr_timeout(&outarg),
				       attr_version);
		fuse_change_entry_timeout(entry, &outarg);
250
	} else if (inode) {
251 252 253 254 255
		fi = get_fuse_inode(inode);
		if (flags & LOOKUP_RCU) {
			if (test_bit(FUSE_I_INIT_RDPLUS, &fi->state))
				return -ECHILD;
		} else if (test_and_clear_bit(FUSE_I_INIT_RDPLUS, &fi->state)) {
256
			parent = dget_parent(entry);
257
			fuse_advise_use_readdirplus(d_inode(parent));
258 259
			dput(parent);
		}
260
	}
261 262 263 264 265 266 267
	ret = 1;
out:
	return ret;

invalid:
	ret = 0;
	goto out;
268 269
}

270
static int invalid_nodeid(u64 nodeid)
271 272 273 274
{
	return !nodeid || nodeid == FUSE_ROOT_ID;
}

275
const struct dentry_operations fuse_dentry_operations = {
276 277 278
	.d_revalidate	= fuse_dentry_revalidate,
};

279
int fuse_valid_type(int m)
280 281 282 283 284
{
	return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
		S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
}

285
int fuse_lookup_name(struct super_block *sb, u64 nodeid, const struct qstr *name,
286
		     struct fuse_entry_out *outarg, struct inode **inode)
287
{
288
	struct fuse_conn *fc = get_fuse_conn_super(sb);
289
	FUSE_ARGS(args);
290
	struct fuse_forget_link *forget;
291
	u64 attr_version;
292
	int err;
293

294 295 296 297
	*inode = NULL;
	err = -ENAMETOOLONG;
	if (name->len > FUSE_NAME_MAX)
		goto out;
298 299


300 301
	forget = fuse_alloc_forget();
	err = -ENOMEM;
302
	if (!forget)
303
		goto out;
304

305
	attr_version = fuse_get_attr_version(fc);
306

307 308
	fuse_lookup_init(fc, &args, nodeid, name, outarg);
	err = fuse_simple_request(fc, &args);
309
	/* Zero nodeid is same as -ENOENT, but with valid timeout */
310 311 312 313 314 315 316 317 318 319 320 321 322 323
	if (err || !outarg->nodeid)
		goto out_put_forget;

	err = -EIO;
	if (!outarg->nodeid)
		goto out_put_forget;
	if (!fuse_valid_type(outarg->attr.mode))
		goto out_put_forget;

	*inode = fuse_iget(sb, outarg->nodeid, outarg->generation,
			   &outarg->attr, entry_attr_timeout(outarg),
			   attr_version);
	err = -ENOMEM;
	if (!*inode) {
324
		fuse_queue_forget(fc, forget, outarg->nodeid, 1);
325
		goto out;
326
	}
327 328 329
	err = 0;

 out_put_forget:
330
	kfree(forget);
331 332 333 334 335
 out:
	return err;
}

static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
336
				  unsigned int flags)
337 338 339 340 341 342 343
{
	int err;
	struct fuse_entry_out outarg;
	struct inode *inode;
	struct dentry *newent;
	bool outarg_valid = true;

344
	fuse_lock_inode(dir);
345 346
	err = fuse_lookup_name(dir->i_sb, get_node_id(dir), &entry->d_name,
			       &outarg, &inode);
347
	fuse_unlock_inode(dir);
348 349 350 351 352 353 354 355 356 357
	if (err == -ENOENT) {
		outarg_valid = false;
		err = 0;
	}
	if (err)
		goto out_err;

	err = -EIO;
	if (inode && get_node_id(inode) == FUSE_ROOT_ID)
		goto out_iput;
358

359
	newent = d_splice_alias(inode, entry);
360 361 362
	err = PTR_ERR(newent);
	if (IS_ERR(newent))
		goto out_err;
363

364
	entry = newent ? newent : entry;
365
	if (outarg_valid)
366
		fuse_change_entry_timeout(entry, &outarg);
367 368
	else
		fuse_invalidate_entry_cache(entry);
369

370
	fuse_advise_use_readdirplus(dir);
371
	return newent;
372 373 374 375 376

 out_iput:
	iput(inode);
 out_err:
	return ERR_PTR(err);
377 378
}

379 380 381 382 383 384
/*
 * Atomic create+open operation
 *
 * If the filesystem doesn't support this, then fall back to separate
 * 'mknod' + 'open' requests.
 */
385
static int fuse_create_open(struct inode *dir, struct dentry *entry,
Al Viro's avatar
Al Viro committed
386
			    struct file *file, unsigned flags,
387
			    umode_t mode, int *opened)
388 389 390 391
{
	int err;
	struct inode *inode;
	struct fuse_conn *fc = get_fuse_conn(dir);
392
	FUSE_ARGS(args);
393
	struct fuse_forget_link *forget;
394
	struct fuse_create_in inarg;
395 396 397 398
	struct fuse_open_out outopen;
	struct fuse_entry_out outentry;
	struct fuse_file *ff;

399 400 401
	/* Userspace expects S_IFREG in create mode */
	BUG_ON((mode & S_IFMT) != S_IFREG);

402
	forget = fuse_alloc_forget();
403
	err = -ENOMEM;
404
	if (!forget)
405
		goto out_err;
406

407
	err = -ENOMEM;
Tejun Heo's avatar
Tejun Heo committed
408
	ff = fuse_file_alloc(fc);
409
	if (!ff)
410
		goto out_put_forget_req;
411

412 413 414
	if (!fc->dont_mask)
		mode &= ~current_umask();

415 416
	flags &= ~O_NOCTTY;
	memset(&inarg, 0, sizeof(inarg));
417
	memset(&outentry, 0, sizeof(outentry));
418 419
	inarg.flags = flags;
	inarg.mode = mode;
420
	inarg.umask = current_umask();
421 422 423
	args.in.h.opcode = FUSE_CREATE;
	args.in.h.nodeid = get_node_id(dir);
	args.in.numargs = 2;
424
	args.in.args[0].size = sizeof(inarg);
425 426 427 428
	args.in.args[0].value = &inarg;
	args.in.args[1].size = entry->d_name.len + 1;
	args.in.args[1].value = entry->d_name.name;
	args.out.numargs = 2;
429
	args.out.args[0].size = sizeof(outentry);
430 431 432 433
	args.out.args[0].value = &outentry;
	args.out.args[1].size = sizeof(outopen);
	args.out.args[1].value = &outopen;
	err = fuse_simple_request(fc, &args);
434
	if (err)
435 436 437
		goto out_free_ff;

	err = -EIO;
438
	if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
439 440
		goto out_free_ff;

441 442 443
	ff->fh = outopen.fh;
	ff->nodeid = outentry.nodeid;
	ff->open_flags = outopen.open_flags;
444
	inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
445
			  &outentry.attr, entry_attr_timeout(&outentry), 0);
446 447
	if (!inode) {
		flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
448
		fuse_sync_release(ff, flags);
449
		fuse_queue_forget(fc, forget, outentry.nodeid, 1);
450 451
		err = -ENOMEM;
		goto out_err;
452
	}
453
	kfree(forget);
454
	d_instantiate(entry, inode);
455
	fuse_change_entry_timeout(entry, &outentry);
456
	fuse_invalidate_attr(dir);
Al Viro's avatar
Al Viro committed
457 458
	err = finish_open(file, entry, generic_file_open, opened);
	if (err) {
459
		fuse_sync_release(ff, flags);
460 461 462
	} else {
		file->private_data = fuse_file_get(ff);
		fuse_finish_open(inode, file);
463
	}
464
	return err;
465

466
out_free_ff:
467
	fuse_file_free(ff);
468
out_put_forget_req:
469
	kfree(forget);
470
out_err:
471
	return err;
472 473 474
}

static int fuse_mknod(struct inode *, struct dentry *, umode_t, dev_t);
475
static int fuse_atomic_open(struct inode *dir, struct dentry *entry,
Al Viro's avatar
Al Viro committed
476
			    struct file *file, unsigned flags,
477
			    umode_t mode, int *opened)
478 479 480 481 482
{
	int err;
	struct fuse_conn *fc = get_fuse_conn(dir);
	struct dentry *res = NULL;

483
	if (d_in_lookup(entry)) {
484
		res = fuse_lookup(dir, entry, 0);
485
		if (IS_ERR(res))
486
			return PTR_ERR(res);
487 488 489 490 491

		if (res)
			entry = res;
	}

492
	if (!(flags & O_CREAT) || d_really_is_positive(entry))
493 494 495
		goto no_open;

	/* Only creates */
496
	*opened |= FILE_CREATED;
497 498 499 500

	if (fc->no_create)
		goto mknod;

Al Viro's avatar
Al Viro committed
501
	err = fuse_create_open(dir, entry, file, flags, mode, opened);
502
	if (err == -ENOSYS) {
503 504 505 506 507
		fc->no_create = 1;
		goto mknod;
	}
out_dput:
	dput(res);
508
	return err;
509 510 511

mknod:
	err = fuse_mknod(dir, entry, mode, 0);
512
	if (err)
513 514
		goto out_dput;
no_open:
515
	return finish_no_open(file, res);
516 517
}

518 519 520
/*
 * Code shared between mknod, mkdir, symlink and link
 */
521
static int create_new_entry(struct fuse_conn *fc, struct fuse_args *args,
522
			    struct inode *dir, struct dentry *entry,
Al Viro's avatar
Al Viro committed
523
			    umode_t mode)
524 525 526 527
{
	struct fuse_entry_out outarg;
	struct inode *inode;
	int err;
528
	struct fuse_forget_link *forget;
529

530
	forget = fuse_alloc_forget();
531
	if (!forget)
532
		return -ENOMEM;
533

534
	memset(&outarg, 0, sizeof(outarg));
535 536
	args->in.h.nodeid = get_node_id(dir);
	args->out.numargs = 1;
537
	args->out.args[0].size = sizeof(outarg);
538 539
	args->out.args[0].value = &outarg;
	err = fuse_simple_request(fc, args);
540 541 542
	if (err)
		goto out_put_forget_req;

543 544
	err = -EIO;
	if (invalid_nodeid(outarg.nodeid))
545
		goto out_put_forget_req;
546 547

	if ((outarg.attr.mode ^ mode) & S_IFMT)
548
		goto out_put_forget_req;
549

550
	inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
551
			  &outarg.attr, entry_attr_timeout(&outarg), 0);
552
	if (!inode) {
553
		fuse_queue_forget(fc, forget, outarg.nodeid, 1);
554 555
		return -ENOMEM;
	}
556
	kfree(forget);
557

558 559 560
	err = d_instantiate_no_diralias(entry, inode);
	if (err)
		return err;
561

562
	fuse_change_entry_timeout(entry, &outarg);
563 564
	fuse_invalidate_attr(dir);
	return 0;
565

566
 out_put_forget_req:
567
	kfree(forget);
568
	return err;
569 570
}

Al Viro's avatar
Al Viro committed
571
static int fuse_mknod(struct inode *dir, struct dentry *entry, umode_t mode,
572 573 574 575
		      dev_t rdev)
{
	struct fuse_mknod_in inarg;
	struct fuse_conn *fc = get_fuse_conn(dir);
576
	FUSE_ARGS(args);
577

578 579 580
	if (!fc->dont_mask)
		mode &= ~current_umask();

581 582 583
	memset(&inarg, 0, sizeof(inarg));
	inarg.mode = mode;
	inarg.rdev = new_encode_dev(rdev);
584
	inarg.umask = current_umask();
585 586
	args.in.h.opcode = FUSE_MKNOD;
	args.in.numargs = 2;
587
	args.in.args[0].size = sizeof(inarg);
588 589 590 591
	args.in.args[0].value = &inarg;
	args.in.args[1].size = entry->d_name.len + 1;
	args.in.args[1].value = entry->d_name.name;
	return create_new_entry(fc, &args, dir, entry, mode);
592 593
}

Al Viro's avatar
Al Viro committed
594
static int fuse_create(struct inode *dir, struct dentry *entry, umode_t mode,
595
		       bool excl)
596 597 598 599
{
	return fuse_mknod(dir, entry, mode, 0);
}

600
static int fuse_mkdir(struct inode *dir, struct dentry *entry, umode_t mode)
601 602 603
{
	struct fuse_mkdir_in inarg;
	struct fuse_conn *fc = get_fuse_conn(dir);
604
	FUSE_ARGS(args);
605

606 607 608
	if (!fc->dont_mask)
		mode &= ~current_umask();

609 610
	memset(&inarg, 0, sizeof(inarg));
	inarg.mode = mode;
611
	inarg.umask = current_umask();
612 613 614 615 616 617 618
	args.in.h.opcode = FUSE_MKDIR;
	args.in.numargs = 2;
	args.in.args[0].size = sizeof(inarg);
	args.in.args[0].value = &inarg;
	args.in.args[1].size = entry->d_name.len + 1;
	args.in.args[1].value = entry->d_name.name;
	return create_new_entry(fc, &args, dir, entry, S_IFDIR);
619 620 621 622 623 624 625
}

static int fuse_symlink(struct inode *dir, struct dentry *entry,
			const char *link)
{
	struct fuse_conn *fc = get_fuse_conn(dir);
	unsigned len = strlen(link) + 1;
626
	FUSE_ARGS(args);
627

628 629 630 631 632 633 634
	args.in.h.opcode = FUSE_SYMLINK;
	args.in.numargs = 2;
	args.in.args[0].size = entry->d_name.len + 1;
	args.in.args[0].value = entry->d_name.name;
	args.in.args[1].size = len;
	args.in.args[1].value = link;
	return create_new_entry(fc, &args, dir, entry, S_IFLNK);
635 636
}

637 638 639 640 641 642 643 644
static inline void fuse_update_ctime(struct inode *inode)
{
	if (!IS_NOCMTIME(inode)) {
		inode->i_ctime = current_fs_time(inode->i_sb);
		mark_inode_dirty_sync(inode);
	}
}

645 646 647 648
static int fuse_unlink(struct inode *dir, struct dentry *entry)
{
	int err;
	struct fuse_conn *fc = get_fuse_conn(dir);
649 650 651 652 653 654 655 656
	FUSE_ARGS(args);

	args.in.h.opcode = FUSE_UNLINK;
	args.in.h.nodeid = get_node_id(dir);
	args.in.numargs = 1;
	args.in.args[0].size = entry->d_name.len + 1;
	args.in.args[0].value = entry->d_name.name;
	err = fuse_simple_request(fc, &args);
657
	if (!err) {
658
		struct inode *inode = d_inode(entry);
659
		struct fuse_inode *fi = get_fuse_inode(inode);
660

661 662
		spin_lock(&fc->lock);
		fi->attr_version = ++fc->attr_version;
663 664 665 666 667 668 669 670
		/*
		 * If i_nlink == 0 then unlink doesn't make sense, yet this can
		 * happen if userspace filesystem is careless.  It would be
		 * difficult to enforce correct nlink usage so just ignore this
		 * condition here
		 */
		if (inode->i_nlink > 0)
			drop_nlink(inode);
671
		spin_unlock(&fc->lock);
672 673
		fuse_invalidate_attr(inode);
		fuse_invalidate_attr(dir);
674
		fuse_invalidate_entry_cache(entry);
675
		fuse_update_ctime(inode);
676 677 678 679 680 681 682 683 684
	} else if (err == -EINTR)
		fuse_invalidate_entry(entry);
	return err;
}

static int fuse_rmdir(struct inode *dir, struct dentry *entry)
{
	int err;
	struct fuse_conn *fc = get_fuse_conn(dir);
685 686 687 688 689 690 691 692
	FUSE_ARGS(args);

	args.in.h.opcode = FUSE_RMDIR;
	args.in.h.nodeid = get_node_id(dir);
	args.in.numargs = 1;
	args.in.args[0].size = entry->d_name.len + 1;
	args.in.args[0].value = entry->d_name.name;
	err = fuse_simple_request(fc, &args);
693
	if (!err) {
694
		clear_nlink(d_inode(entry));
695
		fuse_invalidate_attr(dir);
696
		fuse_invalidate_entry_cache(entry);
697 698 699 700 701
	} else if (err == -EINTR)
		fuse_invalidate_entry(entry);
	return err;
}

702 703 704
static int fuse_rename_common(struct inode *olddir, struct dentry *oldent,
			      struct inode *newdir, struct dentry *newent,
			      unsigned int flags, int opcode, size_t argsize)
705 706
{
	int err;
707
	struct fuse_rename2_in inarg;
708
	struct fuse_conn *fc = get_fuse_conn(olddir);
709
	FUSE_ARGS(args);
710

711
	memset(&inarg, 0, argsize);
712
	inarg.newdir = get_node_id(newdir);
713
	inarg.flags = flags;
714 715 716 717 718 719 720 721 722 723
	args.in.h.opcode = opcode;
	args.in.h.nodeid = get_node_id(olddir);
	args.in.numargs = 3;
	args.in.args[0].size = argsize;
	args.in.args[0].value = &inarg;
	args.in.args[1].size = oldent->d_name.len + 1;
	args.in.args[1].value = oldent->d_name.name;
	args.in.args[2].size = newent->d_name.len + 1;
	args.in.args[2].value = newent->d_name.name;
	err = fuse_simple_request(fc, &args);
724
	if (!err) {
725
		/* ctime changes */
726 727
		fuse_invalidate_attr(d_inode(oldent));
		fuse_update_ctime(d_inode(oldent));
728

729
		if (flags & RENAME_EXCHANGE) {
730 731
			fuse_invalidate_attr(d_inode(newent));
			fuse_update_ctime(d_inode(newent));
732 733
		}

734 735 736
		fuse_invalidate_attr(olddir);
		if (olddir != newdir)
			fuse_invalidate_attr(newdir);
737 738

		/* newent will end up negative */
739 740
		if (!(flags & RENAME_EXCHANGE) && d_really_is_positive(newent)) {
			fuse_invalidate_attr(d_inode(newent));
741
			fuse_invalidate_entry_cache(newent);
742
			fuse_update_ctime(d_inode(newent));
743
		}
744 745 746 747 748 749 750
	} else if (err == -EINTR) {
		/* If request was interrupted, DEITY only knows if the
		   rename actually took place.  If the invalidation
		   fails (e.g. some process has CWD under the renamed
		   directory), then there can be inconsistency between
		   the dcache and the real filesystem.  Tough luck. */
		fuse_invalidate_entry(oldent);
751
		if (d_really_is_positive(newent))
752 753 754 755 756 757
			fuse_invalidate_entry(newent);
	}

	return err;
}

758 759 760 761 762 763 764 765 766 767
static int fuse_rename2(struct inode *olddir, struct dentry *oldent,
			struct inode *newdir, struct dentry *newent,
			unsigned int flags)
{
	struct fuse_conn *fc = get_fuse_conn(olddir);
	int err;

	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
		return -EINVAL;

768 769 770
	if (flags) {
		if (fc->no_rename2 || fc->minor < 23)
			return -EINVAL;
771

772 773 774 775 776 777 778 779 780 781 782
		err = fuse_rename_common(olddir, oldent, newdir, newent, flags,
					 FUSE_RENAME2,
					 sizeof(struct fuse_rename2_in));
		if (err == -ENOSYS) {
			fc->no_rename2 = 1;
			err = -EINVAL;
		}
	} else {
		err = fuse_rename_common(olddir, oldent, newdir, newent, 0,
					 FUSE_RENAME,
					 sizeof(struct fuse_rename_in));
783
	}
784

785
	return err;
786
}
787

788 789 790 791 792
static int fuse_link(struct dentry *entry, struct inode *newdir,
		     struct dentry *newent)
{
	int err;
	struct fuse_link_in inarg;
793
	struct inode *inode = d_inode(entry);
794
	struct fuse_conn *fc = get_fuse_conn(inode);
795
	FUSE_ARGS(args);
796 797 798

	memset(&inarg, 0, sizeof(inarg));
	inarg.oldnodeid = get_node_id(inode);
799 800 801 802 803 804 805
	args.in.h.opcode = FUSE_LINK;
	args.in.numargs = 2;
	args.in.args[0].size = sizeof(inarg);
	args.in.args[0].value = &inarg;
	args.in.args[1].size = newent->d_name.len + 1;
	args.in.args[1].value = newent->d_name.name;
	err = create_new_entry(fc, &args, newdir, newent, inode->i_mode);
806 807 808 809 810 811
	/* Contrary to "normal" filesystems it can happen that link
	   makes two "logical" inodes point to the same "physical"
	   inode.  We invalidate the attributes of the old one, so it
	   will reflect changes in the backing inode (link count,
	   etc.)
	*/
812 813 814 815 816 817 818
	if (!err) {
		struct fuse_inode *fi = get_fuse_inode(inode);

		spin_lock(&fc->lock);
		fi->attr_version = ++fc->attr_version;
		inc_nlink(inode);
		spin_unlock(&fc->lock);
819
		fuse_invalidate_attr(inode);
820
		fuse_update_ctime(inode);
821 822 823
	} else if (err == -EINTR) {
		fuse_invalidate_attr(inode);
	}
824 825 826
	return err;
}

827 828 829
static void fuse_fillattr(struct inode *inode, struct fuse_attr *attr,
			  struct kstat *stat)
{
830
	unsigned int blkbits;
831 832 833
	struct fuse_conn *fc = get_fuse_conn(inode);

	/* see the comment in fuse_change_attributes() */
834
	if (fc->writeback_cache && S_ISREG(inode->i_mode)) {
835
		attr->size = i_size_read(inode);
836 837
		attr->mtime = inode->i_mtime.tv_sec;
		attr->mtimensec = inode->i_mtime.tv_nsec;
838 839
		attr->ctime = inode->i_ctime.tv_sec;
		attr->ctimensec = inode->i_ctime.tv_nsec;
840
	}
841

842 843 844 845
	stat->dev = inode->i_sb->s_dev;
	stat->ino = attr->ino;
	stat->mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
	stat->nlink = attr->nlink;
846 847
	stat->uid = make_kuid(&init_user_ns, attr->uid);
	stat->gid = make_kgid(&init_user_ns, attr->gid);
848 849 850 851 852 853 854 855 856
	stat->rdev = inode->i_rdev;
	stat->atime.tv_sec = attr->atime;
	stat->atime.tv_nsec = attr->atimensec;
	stat->mtime.tv_sec = attr->mtime;
	stat->mtime.tv_nsec = attr->mtimensec;
	stat->ctime.tv_sec = attr->ctime;
	stat->ctime.tv_nsec = attr->ctimensec;
	stat->size = attr->size;
	stat->blocks = attr->blocks;
857 858 859 860 861 862 863

	if (attr->blksize != 0)
		blkbits = ilog2(attr->blksize);
	else
		blkbits = inode->i_sb->s_blocksize_bits;

	stat->blksize = 1 << blkbits;
864 865
}

866 867
static int fuse_do_getattr(struct inode *inode, struct kstat *stat,
			   struct file *file)
868 869
{
	int err;
870 871
	struct fuse_getattr_in inarg;
	struct fuse_attr_out outarg;
872
	struct fuse_conn *fc = get_fuse_conn(inode);
873
	FUSE_ARGS(args);
874 875
	u64 attr_version;

876
	attr_version = fuse_get_attr_version(fc);
877

878
	memset(&inarg, 0, sizeof(inarg));
879
	memset(&outarg, 0, sizeof(outarg));
880 881 882 883 884 885 886
	/* Directories have separate file-handle space */
	if (file && S_ISREG(inode->i_mode)) {
		struct fuse_file *ff = file->private_data;

		inarg.getattr_flags |= FUSE_GETATTR_FH;
		inarg.fh = ff->fh;
	}
887 888 889 890 891 892
	args.in.h.opcode = FUSE_GETATTR;
	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;
893
	args.out.args[0].size = sizeof(outarg);
894 895
	args.out.args[0].value = &outarg;
	err = fuse_simple_request(fc, &args);
896
	if (!err) {
897
		if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
898 899 900
			make_bad_inode(inode);
			err = -EIO;
		} else {
901 902
			fuse_change_attributes(inode, &outarg.attr,
					       attr_timeout(&outarg),
903 904
					       attr_version);
			if (stat)
905
				fuse_fillattr(inode, &outarg.attr, stat);
906 907 908 909 910
		}
	}
	return err;
}

911 912 913 914 915 916 917
int fuse_update_attributes(struct inode *inode, struct kstat *stat,
			   struct file *file, bool *refreshed)
{
	struct fuse_inode *fi = get_fuse_inode(inode);
	int err;
	bool r;

918
	if (time_before64(fi->i_time, get_jiffies_64())) {
919 920 921 922 923 924 925 926
		r = true;
		err = fuse_do_getattr(inode, stat, file);
	} else {
		r = false;
		err = 0;
		if (stat) {
			generic_fillattr(inode, stat);
			stat->mode = fi->orig_i_mode;
927
			stat->ino = fi->orig_ino;
928 929 930 931 932 933 934 935 936
		}
	}

	if (refreshed != NULL)
		*refreshed = r;

	return err;
}

937
int fuse_reverse_inval_entry(struct super_block *sb, u64 parent_nodeid,
938
			     u64 child_nodeid, struct qstr *name)
939 940 941 942 943 944 945 946 947 948
{
	int err = -ENOTDIR;
	struct inode *parent;
	struct dentry *dir;
	struct dentry *entry;

	parent = ilookup5(sb, parent_nodeid, fuse_inode_eq, &parent_nodeid);
	if (!parent)
		return -ENOENT;

Al Viro's avatar
Al Viro committed
949
	inode_lock(parent);
950 951 952 953 954 955 956 957
	if (!S_ISDIR(parent->i_mode))
		goto unlock;

	err = -ENOENT;
	dir = d_find_alias(parent);
	if (!dir)
		goto unlock;

958
	name->hash = full_name_hash(dir, name->name, name->len);
959 960 961 962 963 964 965
	entry = d_lookup(dir, name);
	dput(dir);
	if (!entry)
		goto unlock;

	fuse_invalidate_attr(parent);
	fuse_invalidate_entry(entry);
966

967
	if (child_nodeid != 0 && d_really_is_positive(entry)) {
Al Viro's avatar
Al Viro committed
968
		inode_lock(d_inode(entry));
969
		if (get_node_id(d_inode(entry)) != child_nodeid) {
970 971 972 973 974 975 976
			err = -ENOENT;
			goto badentry;
		}
		if (d_mountpoint(entry)) {
			err = -EBUSY;
			goto badentry;
		}
977
		if (d_is_dir(entry)) {
978 979 980 981 982
			shrink_dcache_parent(entry);
			if (!simple_empty(entry)) {
				err = -ENOTEMPTY;
				goto badentry;
			}
983
			d_inode(entry)->i_flags |= S_DEAD;
984 985
		}
		dont_mount(entry);
986
		clear_nlink(d_inode(entry));
987 988
		err = 0;
 badentry:
Al Viro's avatar
Al Viro committed
989
		inode_unlock(d_inode(entry));
990 991 992 993 994
		if (!err)
			d_delete(entry);
	} else {
		err = 0;
	}
995 996 997
	dput(entry);

 unlock:
Al Viro's avatar
Al Viro committed
998
	inode_unlock(parent);
999 1000 1001 1002
	iput(parent);
	return err;
}

1003 1004
/*
 * Calling into a user-controlled filesystem gives the filesystem
1005
 * daemon ptrace-like capabilities over the current process.  This
1006 1007 1008 1009 1010 1011 1012 1013 1014 1015
 * means, that the filesystem daemon is able to record the exact
 * filesystem operations performed, and can also control the behavior
 * of the requester process in otherwise impossible ways.  For example
 * it can delay the operation for arbitrary length of time allowing
 * DoS against the requester.
 *
 * For this reason only those processes can call into the filesystem,
 * for which the owner of the mount has ptrace privilege.  This
 * excludes processes started by other users, suid or sgid processes.
 */
1016
int fuse_allow_current_process(struct fuse_conn *fc)
1017
{
1018
	const struct cred *cred;
1019

1020
	if (fc->flags & FUSE_ALLOW_OTHER)
1021 1022
		return 1;

1023
	cred = current_cred();
1024 1025 1026 1027 1028 1029
	if (uid_eq(cred->euid, fc->user_id) &&
	    uid_eq(cred->suid, fc->user_id) &&
	    uid_eq(cred->uid,  fc->user_id) &&
	    gid_eq(cred->egid, fc->group_id) &&
	    gid_eq(cred->sgid, fc->group_id) &&
	    gid_eq(cred->gid,  fc->group_id))
1030
		return 1;
1031

1032
	return 0;
1033 1034
}

1035 1036 1037
static int fuse_access(struct inode *inode, int mask)
{
	struct fuse_conn *fc = get_fuse_conn(inode);
1038
	FUSE_ARGS(args);
1039 1040 1041
	struct fuse_access_in inarg;
	int err;

1042 1043
	BUG_ON(mask & MAY_NOT_BLOCK);

1044 1045 1046 1047
	if (fc->no_access)
		return 0;

	memset(&inarg, 0, sizeof(inarg));
1048
	inarg.mask = mask & (MAY_READ | MAY_WRITE | MAY_EXEC);
1049 1050 1051 1052 1053 1054
	args.in.h.opcode = FUSE_ACCESS;
	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);
1055 1056 1057 1058 1059 1060 1061
	if (err == -ENOSYS) {
		fc->no_access = 1;
		err = 0;
	}
	return err;
}

1062
static int fuse_perm_getattr(struct inode *inode, int mask)
1063
{
1064
	if (mask & MAY_NOT_BLOCK)
1065 1066 1067 1068 1069
		return -ECHILD;

	return fuse_do_getattr(inode, NULL, NULL);
}

1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082
/*
 * Check permission.  The two basic access models of FUSE are:
 *
 * 1) Local access checking ('default_permissions' mount option) based
 * on file mode.  This is the plain old disk filesystem permission
 * modell.
 *
 * 2) "Remote" access checking, where server is responsible for
 * checking permission in each inode operation.  An exception to this
 * is if ->permission() was invoked from sys_access() in which case an
 * access request is sent.  Execute permission is still checked
 * locally based on file mode.
 */
1083
static int fuse_permission(struct inode *inode, int mask)
1084 1085
{
	struct fuse_conn *fc = get_fuse_conn(inode);
1086 1087
	bool refreshed = false;
	int err = 0;
1088

1089
	if (!fuse_allow_current_process(fc))
1090
		return -EACCES;
1091 1092

	/*
1093
	 * If attributes are needed, refresh them before proceeding
1094
	 */
1095 1096
	if ((fc->flags & FUSE_DEFAULT_PERMISSIONS) ||
	    ((mask & MAY_EXEC) && S_ISREG(inode->i_mode))) {
1097 1098
		struct fuse_inode *fi = get_fuse_inode(inode);

1099
		if (time_before64(fi->i_time, get_jiffies_64())) {
1100 1101
			refreshed = true;

1102
			err = fuse_perm_getattr(inode, mask);
1103 1104 1105
			if (err)
				return err;
		}
1106 1107 1108
	}

	if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
1109
		err = generic_permission(inode, mask);
1110 1111 1112 1113

		/* If permission is denied, try to refresh file
		   attributes.  This is also needed, because the root
		   node will at first have no permissions */
1114
		if (err == -EACCES && !refreshed) {
1115
			err = fuse_perm_getattr(inode, mask);
1116
			if (!err)
1117
				err = generic_permission(inode, mask);
1118 1119
		}

1120 1121 1122 1123
		/* Note: the opposite of the above test does not
		   exist.  So if permissions are revoked this won't be
		   noticed immediately, only after the attribute
		   timeout has expired */
Eric Paris's avatar
Eric Paris committed
1124
	} else if (mask & (MAY_ACCESS | MAY_CHDIR)) {
1125 1126 1127 1128 1129 1130
		err = fuse_access(inode, mask);
	} else if ((mask & MAY_EXEC) && S_ISREG(inode->i_mode)) {
		if (!(inode->i_mode & S_IXUGO)) {
			if (refreshed)
				return -EACCES;

1131
			err = fuse_perm_getattr(inode, mask);
1132 1133 1134
			if (!err && !(inode->i_mode & S_IXUGO))
				return -EACCES;
		}
1135
	}
1136
	return err;
1137 1138 1139
}

static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
Al Viro's avatar
Al Viro committed
1140
			 struct dir_context *ctx)
1141 1142 1143 1144 1145 1146 1147 1148
{
	while (nbytes >= FUSE_NAME_OFFSET) {
		struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
		size_t reclen = FUSE_DIRENT_SIZE(dirent);
		if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
			return -EIO;
		if (reclen > nbytes)
			break;
1149 1150
		if (memchr(dirent->name, '/', dirent->namelen) != NULL)
			return -EIO;
1151

Al Viro's avatar
Al Viro committed
1152 1153
		if (!dir_emit(ctx, dirent->name, dirent->namelen,
			       dirent->ino, dirent->type))
1154 1155 1156 1157
			break;

		buf += reclen;
		nbytes -= reclen;
Al Viro's avatar
Al Viro committed
1158
		ctx->pos = dirent->off;
1159 1160 1161 1162 1163
	}

	return 0;
}

1164 1165 1166 1167 1168 1169 1170 1171 1172 1173
static int fuse_direntplus_link(struct file *file,
				struct fuse_direntplus *direntplus,
				u64 attr_version)
{
	struct fuse_entry_out *o = &direntplus->entry_out;
	struct fuse_dirent *dirent = &direntplus->dirent;
	struct dentry *parent = file->f_path.dentry;
	struct qstr name = QSTR_INIT(dirent->name, dirent->namelen);
	struct dentry *dentry;
	struct dentry *alias;
1174
	struct inode *dir = d_inode(parent);
1175 1176
	struct fuse_conn *fc;
	struct inode *inode;
1177
	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199

	if (!o->nodeid) {
		/*
		 * Unlike in the case of fuse_lookup, zero nodeid does not mean
		 * ENOENT. Instead, it only means the userspace filesystem did
		 * not want to return attributes/handle for this entry.
		 *
		 * So do nothing.
		 */
		return 0;
	}

	if (name.name[0] == '.') {
		/*
		 * We could potentially refresh the attributes of the directory
		 * and its parent?
		 */
		if (name.len == 1)
			return 0;
		if (name.name[1] == '.' && name.len == 2)
			return 0;
	}
1200 1201 1202 1203 1204 1205

	if (invalid_nodeid(o->nodeid))
		return -EIO;
	if (!fuse_valid_type(o->attr.mode))
		return -EIO;

1206 1207
	fc = get_fuse_conn(dir);

1208
	name.hash = full_name_hash(parent, name.name, name.len);
1209
	dentry = d_lookup(parent, &name);
1210 1211 1212 1213 1214 1215 1216 1217
	if (!dentry) {
retry:
		dentry = d_alloc_parallel(parent, &name, &wq);
		if (IS_ERR(dentry))
			return PTR_ERR(dentry);
	}
	if (!d_in_lookup(dentry)) {
		struct fuse_inode *fi;
1218
		inode = d_inode(dentry);
1219 1220 1221
		if (!inode ||
		    get_node_id(inode) != o->nodeid ||
		    ((o->attr.mode ^ inode->i_mode) & S_IFMT)) {
1222
			d_invalidate(dentry);
1223 1224 1225 1226 1227 1228
			dput(dentry);
			goto retry;
		}
		if (is_bad_inode(inode)) {
			dput(dentry);
			return -EIO;
1229 1230
		}

1231 1232 1233 1234
		fi = get_fuse_inode(inode);
		spin_lock(&fc->lock);
		fi->nlookup++;
		spin_unlock(&fc->lock);
1235

1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248
		fuse_change_attributes(inode, &o->attr,
				       entry_attr_timeout(o),
				       attr_version);
		/*
		 * The other branch comes via fuse_iget()
		 * which bumps nlookup inside
		 */
	} else {
		inode = fuse_iget(dir->i_sb, o->nodeid, o->generation,
				  &o->attr, entry_attr_timeout(o),
				  attr_version);
		if (!inode)
			inode = ERR_PTR(-ENOMEM);
1249

1250