unlink.c 14.3 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13
/*
 *  linux/fs/nfs/unlink.c
 *
 * nfs sillydelete handling
 *
 */

#include <linux/slab.h>
#include <linux/string.h>
#include <linux/dcache.h>
#include <linux/sunrpc/sched.h>
#include <linux/sunrpc/clnt.h>
#include <linux/nfs_fs.h>
14 15
#include <linux/sched.h>
#include <linux/wait.h>
16
#include <linux/namei.h>
L
Linus Torvalds 已提交
17

18
#include "internal.h"
19
#include "nfs4_fs.h"
20 21
#include "iostat.h"
#include "delegation.h"
22

L
Linus Torvalds 已提交
23
/**
24
 * nfs_free_unlinkdata - release data from a sillydelete operation.
L
Linus Torvalds 已提交
25 26 27
 * @data: pointer to unlink structure.
 */
static void
28
nfs_free_unlinkdata(struct nfs_unlinkdata *data)
L
Linus Torvalds 已提交
29
{
30 31 32 33
	iput(data->dir);
	put_rpccred(data->cred);
	kfree(data->args.name.name);
	kfree(data);
L
Linus Torvalds 已提交
34 35 36 37 38 39 40 41
}

#define NAME_ALLOC_LEN(len)	((len+16) & ~15)
/**
 * nfs_copy_dname - copy dentry name to data structure
 * @dentry: pointer to dentry
 * @data: nfs_unlinkdata
 */
42
static int nfs_copy_dname(struct dentry *dentry, struct nfs_unlinkdata *data)
L
Linus Torvalds 已提交
43 44 45 46
{
	char		*str;
	int		len = dentry->d_name.len;

47
	str = kmemdup(dentry->d_name.name, NAME_ALLOC_LEN(len), GFP_KERNEL);
L
Linus Torvalds 已提交
48
	if (!str)
49 50 51 52
		return -ENOMEM;
	data->args.name.len = len;
	data->args.name.name = str;
	return 0;
L
Linus Torvalds 已提交
53 54
}

T
Trond Myklebust 已提交
55 56 57 58 59 60 61 62 63 64 65 66 67 68
static void nfs_free_dname(struct nfs_unlinkdata *data)
{
	kfree(data->args.name.name);
	data->args.name.name = NULL;
	data->args.name.len = 0;
}

static void nfs_dec_sillycount(struct inode *dir)
{
	struct nfs_inode *nfsi = NFS_I(dir);
	if (atomic_dec_return(&nfsi->silly_count) == 1)
		wake_up(&nfsi->waitqueue);
}

L
Linus Torvalds 已提交
69 70 71 72 73 74
/**
 * nfs_async_unlink_done - Sillydelete post-processing
 * @task: rpc_task of the sillydelete
 *
 * Do the directory attribute update.
 */
75
static void nfs_async_unlink_done(struct rpc_task *task, void *calldata)
L
Linus Torvalds 已提交
76
{
77 78 79 80
	struct nfs_unlinkdata *data = calldata;
	struct inode *dir = data->dir;

	if (!NFS_PROTO(dir)->unlink_done(task, dir))
81
		rpc_restart_call_prepare(task);
L
Linus Torvalds 已提交
82 83 84 85 86 87 88 89 90
}

/**
 * nfs_async_unlink_release - Release the sillydelete data.
 * @task: rpc_task of the sillydelete
 *
 * We need to call nfs_put_unlinkdata as a 'tk_release' task since the
 * rpc_task would be freed too.
 */
91
static void nfs_async_unlink_release(void *calldata)
L
Linus Torvalds 已提交
92
{
93
	struct nfs_unlinkdata	*data = calldata;
94
	struct super_block *sb = data->dir->i_sb;
T
Trond Myklebust 已提交
95 96

	nfs_dec_sillycount(data->dir);
97
	nfs_free_unlinkdata(data);
98
	nfs_sb_deactive(sb);
L
Linus Torvalds 已提交
99 100
}

101
static void nfs_unlink_prepare(struct rpc_task *task, void *calldata)
102 103
{
	struct nfs_unlinkdata *data = calldata;
104
	NFS_PROTO(data->dir)->unlink_rpc_prepare(task, data);
105 106
}

107 108 109
static const struct rpc_call_ops nfs_unlink_ops = {
	.rpc_call_done = nfs_async_unlink_done,
	.rpc_release = nfs_async_unlink_release,
110
	.rpc_call_prepare = nfs_unlink_prepare,
111 112
};

T
Trond Myklebust 已提交
113
static int nfs_do_call_unlink(struct dentry *parent, struct inode *dir, struct nfs_unlinkdata *data)
114
{
115 116 117 118 119
	struct rpc_message msg = {
		.rpc_argp = &data->args,
		.rpc_resp = &data->res,
		.rpc_cred = data->cred,
	};
T
Trond Myklebust 已提交
120
	struct rpc_task_setup task_setup_data = {
121
		.rpc_message = &msg,
T
Trond Myklebust 已提交
122 123
		.callback_ops = &nfs_unlink_ops,
		.callback_data = data,
124
		.workqueue = nfsiod_workqueue,
T
Trond Myklebust 已提交
125 126
		.flags = RPC_TASK_ASYNC,
	};
127
	struct rpc_task *task;
T
Trond Myklebust 已提交
128 129 130 131
	struct dentry *alias;

	alias = d_lookup(parent, &data->args.name);
	if (alias != NULL) {
132
		int ret;
133
		void *devname_garbage = NULL;
134

T
Trond Myklebust 已提交
135 136 137 138 139
		/*
		 * Hey, we raced with lookup... See if we need to transfer
		 * the sillyrename information to the aliased dentry.
		 */
		nfs_free_dname(data);
140
		ret = nfs_copy_dname(alias, data);
T
Trond Myklebust 已提交
141
		spin_lock(&alias->d_lock);
142
		if (ret == 0 && alias->d_inode != NULL &&
143
		    !(alias->d_flags & DCACHE_NFSFS_RENAMED)) {
144
			devname_garbage = alias->d_fsdata;
T
Trond Myklebust 已提交
145
			alias->d_fsdata = data;
T
Trond Myklebust 已提交
146
			alias->d_flags |= DCACHE_NFSFS_RENAMED;
T
Trond Myklebust 已提交
147
			ret = 1;
148 149
		} else
			ret = 0;
T
Trond Myklebust 已提交
150 151 152
		spin_unlock(&alias->d_lock);
		nfs_dec_sillycount(dir);
		dput(alias);
153 154 155 156 157
		/*
		 * If we'd displaced old cached devname, free it.  At that
		 * point dentry is definitely not a root, so we won't need
		 * that anymore.
		 */
158
		kfree(devname_garbage);
T
Trond Myklebust 已提交
159 160 161 162 163 164 165
		return ret;
	}
	data->dir = igrab(dir);
	if (!data->dir) {
		nfs_dec_sillycount(dir);
		return 0;
	}
166
	nfs_sb_active(dir->i_sb);
T
Trond Myklebust 已提交
167
	data->args.fh = NFS_FH(dir);
168
	nfs_fattr_init(data->res.dir_attr);
T
Trond Myklebust 已提交
169

170
	NFS_PROTO(dir)->unlink_setup(&msg, dir);
T
Trond Myklebust 已提交
171

172
	task_setup_data.rpc_client = NFS_CLIENT(dir);
T
Trond Myklebust 已提交
173
	task = rpc_run_task(&task_setup_data);
T
Trond Myklebust 已提交
174
	if (!IS_ERR(task))
175
		rpc_put_task_async(task);
T
Trond Myklebust 已提交
176 177 178 179 180
	return 1;
}

static int nfs_call_unlink(struct dentry *dentry, struct nfs_unlinkdata *data)
{
181 182
	struct dentry *parent;
	struct inode *dir;
T
Trond Myklebust 已提交
183
	int ret = 0;
184 185 186 187 188


	parent = dget_parent(dentry);
	if (parent == NULL)
		goto out_free;
T
Trond Myklebust 已提交
189 190 191 192 193 194 195 196 197 198 199 200 201
	dir = parent->d_inode;
	/* Non-exclusive lock protects against concurrent lookup() calls */
	spin_lock(&dir->i_lock);
	if (atomic_inc_not_zero(&NFS_I(dir)->silly_count) == 0) {
		/* Deferred delete */
		hlist_add_head(&data->list, &NFS_I(dir)->silly_list);
		spin_unlock(&dir->i_lock);
		ret = 1;
		goto out_dput;
	}
	spin_unlock(&dir->i_lock);
	ret = nfs_do_call_unlink(parent, dir, data);
out_dput:
202
	dput(parent);
T
Trond Myklebust 已提交
203 204 205
out_free:
	return ret;
}
206

T
Trond Myklebust 已提交
207 208 209
void nfs_block_sillyrename(struct dentry *dentry)
{
	struct nfs_inode *nfsi = NFS_I(dentry->d_inode);
210

T
Trond Myklebust 已提交
211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232
	wait_event(nfsi->waitqueue, atomic_cmpxchg(&nfsi->silly_count, 1, 0) == 1);
}

void nfs_unblock_sillyrename(struct dentry *dentry)
{
	struct inode *dir = dentry->d_inode;
	struct nfs_inode *nfsi = NFS_I(dir);
	struct nfs_unlinkdata *data;

	atomic_inc(&nfsi->silly_count);
	spin_lock(&dir->i_lock);
	while (!hlist_empty(&nfsi->silly_list)) {
		if (!atomic_inc_not_zero(&nfsi->silly_count))
			break;
		data = hlist_entry(nfsi->silly_list.first, struct nfs_unlinkdata, list);
		hlist_del(&data->list);
		spin_unlock(&dir->i_lock);
		if (nfs_do_call_unlink(dentry, dir, data) == 0)
			nfs_free_unlinkdata(data);
		spin_lock(&dir->i_lock);
	}
	spin_unlock(&dir->i_lock);
233 234
}

L
Linus Torvalds 已提交
235 236
/**
 * nfs_async_unlink - asynchronous unlinking of a file
237
 * @dir: parent directory of dentry
L
Linus Torvalds 已提交
238 239
 * @dentry: dentry to unlink
 */
240
static int
241
nfs_async_unlink(struct inode *dir, struct dentry *dentry)
L
Linus Torvalds 已提交
242
{
243 244
	struct nfs_unlinkdata *data;
	int status = -ENOMEM;
245
	void *devname_garbage = NULL;
L
Linus Torvalds 已提交
246

247
	data = kzalloc(sizeof(*data), GFP_KERNEL);
248
	if (data == NULL)
L
Linus Torvalds 已提交
249 250
		goto out;

251
	data->cred = rpc_lookup_cred();
L
Linus Torvalds 已提交
252 253 254 255
	if (IS_ERR(data->cred)) {
		status = PTR_ERR(data->cred);
		goto out_free;
	}
256
	data->res.dir_attr = &data->dir_attr;
L
Linus Torvalds 已提交
257

258
	status = -EBUSY;
L
Linus Torvalds 已提交
259
	spin_lock(&dentry->d_lock);
260 261
	if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
		goto out_unlock;
L
Linus Torvalds 已提交
262
	dentry->d_flags |= DCACHE_NFSFS_RENAMED;
263
	devname_garbage = dentry->d_fsdata;
264
	dentry->d_fsdata = data;
L
Linus Torvalds 已提交
265
	spin_unlock(&dentry->d_lock);
266 267 268 269 270
	/*
	 * If we'd displaced old cached devname, free it.  At that
	 * point dentry is definitely not a root, so we won't need
	 * that anymore.
	 */
271
	kfree(devname_garbage);
272 273 274 275
	return 0;
out_unlock:
	spin_unlock(&dentry->d_lock);
	put_rpccred(data->cred);
L
Linus Torvalds 已提交
276 277
out_free:
	kfree(data);
278
out:
L
Linus Torvalds 已提交
279 280 281 282 283 284
	return status;
}

/**
 * nfs_complete_unlink - Initialize completion of the sillydelete
 * @dentry: dentry to delete
285
 * @inode: inode
L
Linus Torvalds 已提交
286 287 288 289 290 291
 *
 * Since we're most likely to be called by dentry_iput(), we
 * only use the dentry to find the sillydelete. We then copy the name
 * into the qstr.
 */
void
292
nfs_complete_unlink(struct dentry *dentry, struct inode *inode)
L
Linus Torvalds 已提交
293
{
294
	struct nfs_unlinkdata	*data = NULL;
L
Linus Torvalds 已提交
295 296

	spin_lock(&dentry->d_lock);
297 298 299
	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
		dentry->d_flags &= ~DCACHE_NFSFS_RENAMED;
		data = dentry->d_fsdata;
300
		dentry->d_fsdata = NULL;
301
	}
L
Linus Torvalds 已提交
302
	spin_unlock(&dentry->d_lock);
303 304 305

	if (data != NULL && (NFS_STALE(inode) || !nfs_call_unlink(dentry, data)))
		nfs_free_unlinkdata(data);
L
Linus Torvalds 已提交
306
}
307

308 309 310 311 312 313 314 315 316
/* Cancel a queued async unlink. Called when a sillyrename run fails. */
static void
nfs_cancel_async_unlink(struct dentry *dentry)
{
	spin_lock(&dentry->d_lock);
	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
		struct nfs_unlinkdata *data = dentry->d_fsdata;

		dentry->d_flags &= ~DCACHE_NFSFS_RENAMED;
317
		dentry->d_fsdata = NULL;
318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336
		spin_unlock(&dentry->d_lock);
		nfs_free_unlinkdata(data);
		return;
	}
	spin_unlock(&dentry->d_lock);
}

/**
 * nfs_async_rename_done - Sillyrename post-processing
 * @task: rpc_task of the sillyrename
 * @calldata: nfs_renamedata for the sillyrename
 *
 * Do the directory attribute updates and the d_move
 */
static void nfs_async_rename_done(struct rpc_task *task, void *calldata)
{
	struct nfs_renamedata *data = calldata;
	struct inode *old_dir = data->old_dir;
	struct inode *new_dir = data->new_dir;
337 338
	struct dentry *old_dentry = data->old_dentry;
	struct dentry *new_dentry = data->new_dentry;
339 340

	if (!NFS_PROTO(old_dir)->rename_done(task, old_dir, new_dir)) {
341
		rpc_restart_call_prepare(task);
342 343 344 345
		return;
	}

	if (task->tk_status != 0) {
346
		nfs_cancel_async_unlink(old_dentry);
347 348 349
		return;
	}

350 351
	d_drop(old_dentry);
	d_drop(new_dentry);
352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377
}

/**
 * nfs_async_rename_release - Release the sillyrename data.
 * @calldata: the struct nfs_renamedata to be released
 */
static void nfs_async_rename_release(void *calldata)
{
	struct nfs_renamedata	*data = calldata;
	struct super_block *sb = data->old_dir->i_sb;

	if (data->old_dentry->d_inode)
		nfs_mark_for_revalidate(data->old_dentry->d_inode);

	dput(data->old_dentry);
	dput(data->new_dentry);
	iput(data->old_dir);
	iput(data->new_dir);
	nfs_sb_deactive(sb);
	put_rpccred(data->cred);
	kfree(data);
}

static void nfs_rename_prepare(struct rpc_task *task, void *calldata)
{
	struct nfs_renamedata *data = calldata;
378
	NFS_PROTO(data->old_dir)->rename_rpc_prepare(task, data);
379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409
}

static const struct rpc_call_ops nfs_rename_ops = {
	.rpc_call_done = nfs_async_rename_done,
	.rpc_release = nfs_async_rename_release,
	.rpc_call_prepare = nfs_rename_prepare,
};

/**
 * nfs_async_rename - perform an asynchronous rename operation
 * @old_dir: directory that currently holds the dentry to be renamed
 * @new_dir: target directory for the rename
 * @old_dentry: original dentry to be renamed
 * @new_dentry: dentry to which the old_dentry should be renamed
 *
 * It's expected that valid references to the dentries and inodes are held
 */
static struct rpc_task *
nfs_async_rename(struct inode *old_dir, struct inode *new_dir,
		 struct dentry *old_dentry, struct dentry *new_dentry)
{
	struct nfs_renamedata *data;
	struct rpc_message msg = { };
	struct rpc_task_setup task_setup_data = {
		.rpc_message = &msg,
		.callback_ops = &nfs_rename_ops,
		.workqueue = nfsiod_workqueue,
		.rpc_client = NFS_CLIENT(old_dir),
		.flags = RPC_TASK_ASYNC,
	};

410
	data = kzalloc(sizeof(*data), GFP_KERNEL);
411 412
	if (data == NULL)
		return ERR_PTR(-ENOMEM);
413
	task_setup_data.callback_data = data;
414 415 416

	data->cred = rpc_lookup_cred();
	if (IS_ERR(data->cred)) {
417
		struct rpc_task *task = ERR_CAST(data->cred);
418 419 420 421 422 423 424 425 426 427
		kfree(data);
		return task;
	}

	msg.rpc_argp = &data->args;
	msg.rpc_resp = &data->res;
	msg.rpc_cred = data->cred;

	/* set up nfs_renamedata */
	data->old_dir = old_dir;
428
	ihold(old_dir);
429
	data->new_dir = new_dir;
430
	ihold(new_dir);
431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449
	data->old_dentry = dget(old_dentry);
	data->new_dentry = dget(new_dentry);
	nfs_fattr_init(&data->old_fattr);
	nfs_fattr_init(&data->new_fattr);

	/* set up nfs_renameargs */
	data->args.old_dir = NFS_FH(old_dir);
	data->args.old_name = &old_dentry->d_name;
	data->args.new_dir = NFS_FH(new_dir);
	data->args.new_name = &new_dentry->d_name;

	/* set up nfs_renameres */
	data->res.old_fattr = &data->old_fattr;
	data->res.new_fattr = &data->new_fattr;

	nfs_sb_active(old_dir->i_sb);

	NFS_PROTO(data->old_dir)->rename_setup(&msg, old_dir);

450
	return rpc_run_task(&task_setup_data);
451 452
}

453 454 455 456 457 458 459 460 461 462 463 464
/**
 * nfs_sillyrename - Perform a silly-rename of a dentry
 * @dir: inode of directory that contains dentry
 * @dentry: dentry to be sillyrenamed
 *
 * NFSv2/3 is stateless and the server doesn't know when the client is
 * holding a file open. To prevent application problems when a file is
 * unlinked while it's still open, the client performs a "silly-rename".
 * That is, it renames the file to a hidden file in the same directory,
 * and only performs the unlink once the last reference to it is put.
 *
 * The final cleanup is done during dentry_iput.
465 466 467 468 469 470 471 472
 *
 * (Note: NFSv4 is stateful, and has opens, so in theory an NFSv4 server
 * could take responsibility for keeping open files referenced.  The server
 * would also need to ensure that opened-but-deleted files were kept over
 * reboots.  However, we may not assume a server does so.  (RFC 5661
 * does provide an OPEN4_RESULT_PRESERVE_UNLINKED flag that a server can
 * use to advertise that it does this; some day we may take advantage of
 * it.))
473 474 475 476 477 478 479 480 481 482
 */
int
nfs_sillyrename(struct inode *dir, struct dentry *dentry)
{
	static unsigned int sillycounter;
	const int      fileidsize  = sizeof(NFS_FILEID(dentry->d_inode))*2;
	const int      countersize = sizeof(sillycounter)*2;
	const int      slen        = sizeof(".nfs")+fileidsize+countersize-1;
	char           silly[slen+1];
	struct dentry *sdentry;
483
	struct rpc_task *task;
484 485 486 487
	int            error = -EIO;

	dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
		dentry->d_parent->d_name.name, dentry->d_name.name,
N
Nick Piggin 已提交
488
		dentry->d_count);
489 490 491 492 493 494 495 496 497 498 499 500 501 502
	nfs_inc_stats(dir, NFSIOS_SILLYRENAME);

	/*
	 * We don't allow a dentry to be silly-renamed twice.
	 */
	error = -EBUSY;
	if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
		goto out;

	sprintf(silly, ".nfs%*.*Lx",
		fileidsize, fileidsize,
		(unsigned long long)NFS_FILEID(dentry->d_inode));

	/* Return delegation in anticipation of the rename */
503
	NFS_PROTO(dentry->d_inode)->return_delegation(dentry->d_inode);
504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524

	sdentry = NULL;
	do {
		char *suffix = silly + slen - countersize;

		dput(sdentry);
		sillycounter++;
		sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);

		dfprintk(VFS, "NFS: trying to rename %s to %s\n",
				dentry->d_name.name, silly);

		sdentry = lookup_one_len(silly, dentry->d_parent, slen);
		/*
		 * N.B. Better to return EBUSY here ... it could be
		 * dangerous to delete the file while it's in use.
		 */
		if (IS_ERR(sdentry))
			goto out;
	} while (sdentry->d_inode != NULL); /* need negative lookup */

525 526 527 528 529 530 531
	/* queue unlink first. Can't do this from rpc_release as it
	 * has to allocate memory
	 */
	error = nfs_async_unlink(dir, dentry);
	if (error)
		goto out_dput;

532 533 534 535 536 537 538 539
	/* populate unlinkdata with the right dname */
	error = nfs_copy_dname(sdentry,
				(struct nfs_unlinkdata *)dentry->d_fsdata);
	if (error) {
		nfs_cancel_async_unlink(dentry);
		goto out_dput;
	}

540 541 542 543 544 545
	/* run the rename task, undo unlink if it fails */
	task = nfs_async_rename(dir, dir, dentry, sdentry);
	if (IS_ERR(task)) {
		error = -EBUSY;
		nfs_cancel_async_unlink(dentry);
		goto out_dput;
546
	}
547 548 549 550 551 552 553

	/* wait for the RPC task to complete, unless a SIGKILL intervenes */
	error = rpc_wait_for_completion_task(task);
	if (error == 0)
		error = task->tk_status;
	rpc_put_task(task);
out_dput:
554 555 556 557
	dput(sdentry);
out:
	return error;
}