Linux Audio

Check our new training course

Loading...
v4.10.11
   1/*
   2*  Copyright (c) 2004 The Regents of the University of Michigan.
   3*  Copyright (c) 2012 Jeff Layton <jlayton@redhat.com>
   4*  All rights reserved.
   5*
   6*  Andy Adamson <andros@citi.umich.edu>
   7*
   8*  Redistribution and use in source and binary forms, with or without
   9*  modification, are permitted provided that the following conditions
  10*  are met:
  11*
  12*  1. Redistributions of source code must retain the above copyright
  13*     notice, this list of conditions and the following disclaimer.
  14*  2. Redistributions in binary form must reproduce the above copyright
  15*     notice, this list of conditions and the following disclaimer in the
  16*     documentation and/or other materials provided with the distribution.
  17*  3. Neither the name of the University nor the names of its
  18*     contributors may be used to endorse or promote products derived
  19*     from this software without specific prior written permission.
  20*
  21*  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22*  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23*  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24*  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25*  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26*  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27*  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28*  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29*  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30*  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31*  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32*
  33*/
  34
  35#include <crypto/hash.h>
  36#include <linux/file.h>
  37#include <linux/slab.h>
  38#include <linux/namei.h>
  39#include <linux/sched.h>
  40#include <linux/fs.h>
  41#include <linux/module.h>
  42#include <net/net_namespace.h>
  43#include <linux/sunrpc/rpc_pipe_fs.h>
  44#include <linux/sunrpc/clnt.h>
  45#include <linux/nfsd/cld.h>
  46
  47#include "nfsd.h"
  48#include "state.h"
  49#include "vfs.h"
  50#include "netns.h"
  51
  52#define NFSDDBG_FACILITY                NFSDDBG_PROC
  53
  54/* Declarations */
  55struct nfsd4_client_tracking_ops {
  56	int (*init)(struct net *);
  57	void (*exit)(struct net *);
  58	void (*create)(struct nfs4_client *);
  59	void (*remove)(struct nfs4_client *);
  60	int (*check)(struct nfs4_client *);
  61	void (*grace_done)(struct nfsd_net *);
 
 
  62};
  63
 
 
 
  64/* Globals */
  65static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
  66
  67static int
  68nfs4_save_creds(const struct cred **original_creds)
  69{
  70	struct cred *new;
  71
  72	new = prepare_creds();
  73	if (!new)
  74		return -ENOMEM;
  75
  76	new->fsuid = GLOBAL_ROOT_UID;
  77	new->fsgid = GLOBAL_ROOT_GID;
  78	*original_creds = override_creds(new);
  79	put_cred(new);
  80	return 0;
  81}
  82
  83static void
  84nfs4_reset_creds(const struct cred *original)
  85{
  86	revert_creds(original);
  87}
  88
  89static void
  90md5_to_hex(char *out, char *md5)
  91{
  92	int i;
  93
  94	for (i=0; i<16; i++) {
  95		unsigned char c = md5[i];
  96
  97		*out++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
  98		*out++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
  99	}
 100	*out = '\0';
 101}
 102
 103static int
 104nfs4_make_rec_clidname(char *dname, const struct xdr_netobj *clname)
 105{
 106	struct xdr_netobj cksum;
 107	struct crypto_shash *tfm;
 108	int status;
 109
 110	dprintk("NFSD: nfs4_make_rec_clidname for %.*s\n",
 111			clname->len, clname->data);
 112	tfm = crypto_alloc_shash("md5", 0, 0);
 113	if (IS_ERR(tfm)) {
 114		status = PTR_ERR(tfm);
 115		goto out_no_tfm;
 116	}
 117
 118	cksum.len = crypto_shash_digestsize(tfm);
 119	cksum.data = kmalloc(cksum.len, GFP_KERNEL);
 120	if (cksum.data == NULL) {
 121		status = -ENOMEM;
 122 		goto out;
 123	}
 124
 125	{
 126		SHASH_DESC_ON_STACK(desc, tfm);
 127
 128		desc->tfm = tfm;
 129		desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
 130
 131		status = crypto_shash_digest(desc, clname->data, clname->len,
 132					     cksum.data);
 133		shash_desc_zero(desc);
 134	}
 135
 136	if (status)
 137		goto out;
 138
 139	md5_to_hex(dname, cksum.data);
 140
 141	status = 0;
 142out:
 143	kfree(cksum.data);
 144	crypto_free_shash(tfm);
 145out_no_tfm:
 146	return status;
 147}
 148
 149/*
 150 * If we had an error generating the recdir name for the legacy tracker
 151 * then warn the admin. If the error doesn't appear to be transient,
 152 * then disable recovery tracking.
 153 */
 154static void
 155legacy_recdir_name_error(struct nfs4_client *clp, int error)
 156{
 157	printk(KERN_ERR "NFSD: unable to generate recoverydir "
 158			"name (%d).\n", error);
 159
 160	/*
 161	 * if the algorithm just doesn't exist, then disable the recovery
 162	 * tracker altogether. The crypto libs will generally return this if
 163	 * FIPS is enabled as well.
 164	 */
 165	if (error == -ENOENT) {
 166		printk(KERN_ERR "NFSD: disabling legacy clientid tracking. "
 167			"Reboot recovery will not function correctly!\n");
 168		nfsd4_client_tracking_exit(clp->net);
 169	}
 170}
 171
 172static void
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 173nfsd4_create_clid_dir(struct nfs4_client *clp)
 174{
 175	const struct cred *original_cred;
 176	char dname[HEXDIR_LEN];
 177	struct dentry *dir, *dentry;
 178	struct nfs4_client_reclaim *crp;
 179	int status;
 180	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 181
 182	if (test_and_set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 183		return;
 184	if (!nn->rec_file)
 185		return;
 186
 187	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
 188	if (status)
 189		return legacy_recdir_name_error(clp, status);
 190
 191	status = nfs4_save_creds(&original_cred);
 192	if (status < 0)
 193		return;
 194
 195	status = mnt_want_write_file(nn->rec_file);
 196	if (status)
 197		goto out_creds;
 198
 199	dir = nn->rec_file->f_path.dentry;
 200	/* lock the parent */
 201	inode_lock(d_inode(dir));
 202
 203	dentry = lookup_one_len(dname, dir, HEXDIR_LEN-1);
 204	if (IS_ERR(dentry)) {
 205		status = PTR_ERR(dentry);
 206		goto out_unlock;
 207	}
 208	if (d_really_is_positive(dentry))
 209		/*
 210		 * In the 4.1 case, where we're called from
 211		 * reclaim_complete(), records from the previous reboot
 212		 * may still be left, so this is OK.
 213		 *
 214		 * In the 4.0 case, we should never get here; but we may
 215		 * as well be forgiving and just succeed silently.
 216		 */
 217		goto out_put;
 218	status = vfs_mkdir(d_inode(dir), dentry, S_IRWXU);
 219out_put:
 220	dput(dentry);
 221out_unlock:
 222	inode_unlock(d_inode(dir));
 223	if (status == 0) {
 224		if (nn->in_grace) {
 225			crp = nfs4_client_to_reclaim(dname, nn);
 226			if (crp)
 227				crp->cr_clp = clp;
 228		}
 229		vfs_fsync(nn->rec_file, 0);
 230	} else {
 231		printk(KERN_ERR "NFSD: failed to write recovery record"
 232				" (err %d); please check that %s exists"
 233				" and is writeable", status,
 234				user_recovery_dirname);
 235	}
 236	mnt_drop_write_file(nn->rec_file);
 237out_creds:
 238	nfs4_reset_creds(original_cred);
 239}
 240
 241typedef int (recdir_func)(struct dentry *, struct dentry *, struct nfsd_net *);
 242
 243struct name_list {
 244	char name[HEXDIR_LEN];
 245	struct list_head list;
 246};
 247
 248struct nfs4_dir_ctx {
 249	struct dir_context ctx;
 250	struct list_head names;
 251};
 252
 253static int
 254nfsd4_build_namelist(struct dir_context *__ctx, const char *name, int namlen,
 255		loff_t offset, u64 ino, unsigned int d_type)
 256{
 257	struct nfs4_dir_ctx *ctx =
 258		container_of(__ctx, struct nfs4_dir_ctx, ctx);
 259	struct name_list *entry;
 260
 261	if (namlen != HEXDIR_LEN - 1)
 262		return 0;
 263	entry = kmalloc(sizeof(struct name_list), GFP_KERNEL);
 264	if (entry == NULL)
 265		return -ENOMEM;
 266	memcpy(entry->name, name, HEXDIR_LEN - 1);
 267	entry->name[HEXDIR_LEN - 1] = '\0';
 268	list_add(&entry->list, &ctx->names);
 269	return 0;
 270}
 271
 272static int
 273nfsd4_list_rec_dir(recdir_func *f, struct nfsd_net *nn)
 274{
 275	const struct cred *original_cred;
 276	struct dentry *dir = nn->rec_file->f_path.dentry;
 277	struct nfs4_dir_ctx ctx = {
 278		.ctx.actor = nfsd4_build_namelist,
 279		.names = LIST_HEAD_INIT(ctx.names)
 280	};
 281	struct name_list *entry, *tmp;
 282	int status;
 283
 284	status = nfs4_save_creds(&original_cred);
 285	if (status < 0)
 286		return status;
 287
 288	status = vfs_llseek(nn->rec_file, 0, SEEK_SET);
 289	if (status < 0) {
 290		nfs4_reset_creds(original_cred);
 291		return status;
 292	}
 293
 294	status = iterate_dir(nn->rec_file, &ctx.ctx);
 295	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 296
 297	list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
 298		if (!status) {
 299			struct dentry *dentry;
 300			dentry = lookup_one_len(entry->name, dir, HEXDIR_LEN-1);
 301			if (IS_ERR(dentry)) {
 302				status = PTR_ERR(dentry);
 303				break;
 304			}
 305			status = f(dir, dentry, nn);
 306			dput(dentry);
 307		}
 308		list_del(&entry->list);
 309		kfree(entry);
 310	}
 311	inode_unlock(d_inode(dir));
 312	nfs4_reset_creds(original_cred);
 313
 314	list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
 315		dprintk("NFSD: %s. Left entry %s\n", __func__, entry->name);
 316		list_del(&entry->list);
 317		kfree(entry);
 318	}
 319	return status;
 320}
 321
 322static int
 323nfsd4_unlink_clid_dir(char *name, int namlen, struct nfsd_net *nn)
 324{
 325	struct dentry *dir, *dentry;
 326	int status;
 327
 328	dprintk("NFSD: nfsd4_unlink_clid_dir. name %.*s\n", namlen, name);
 329
 330	dir = nn->rec_file->f_path.dentry;
 331	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 332	dentry = lookup_one_len(name, dir, namlen);
 333	if (IS_ERR(dentry)) {
 334		status = PTR_ERR(dentry);
 335		goto out_unlock;
 336	}
 337	status = -ENOENT;
 338	if (d_really_is_negative(dentry))
 339		goto out;
 340	status = vfs_rmdir(d_inode(dir), dentry);
 341out:
 342	dput(dentry);
 343out_unlock:
 344	inode_unlock(d_inode(dir));
 345	return status;
 346}
 347
 348static void
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 349nfsd4_remove_clid_dir(struct nfs4_client *clp)
 350{
 351	const struct cred *original_cred;
 352	struct nfs4_client_reclaim *crp;
 353	char dname[HEXDIR_LEN];
 354	int status;
 355	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 356
 357	if (!nn->rec_file || !test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 358		return;
 359
 360	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
 361	if (status)
 362		return legacy_recdir_name_error(clp, status);
 363
 364	status = mnt_want_write_file(nn->rec_file);
 365	if (status)
 366		goto out;
 367	clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
 368
 369	status = nfs4_save_creds(&original_cred);
 370	if (status < 0)
 371		goto out_drop_write;
 372
 373	status = nfsd4_unlink_clid_dir(dname, HEXDIR_LEN-1, nn);
 374	nfs4_reset_creds(original_cred);
 375	if (status == 0) {
 376		vfs_fsync(nn->rec_file, 0);
 377		if (nn->in_grace) {
 378			/* remove reclaim record */
 379			crp = nfsd4_find_reclaim_client(dname, nn);
 380			if (crp)
 381				nfs4_remove_reclaim_record(crp, nn);
 382		}
 383	}
 384out_drop_write:
 385	mnt_drop_write_file(nn->rec_file);
 386out:
 387	if (status)
 388		printk("NFSD: Failed to remove expired client state directory"
 389				" %.*s\n", HEXDIR_LEN, dname);
 390}
 391
 392static int
 393purge_old(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
 394{
 395	int status;
 
 396
 397	if (nfs4_has_reclaimed_state(child->d_name.name, nn))
 
 
 
 398		return 0;
 
 
 
 
 
 
 
 
 
 
 399
 400	status = vfs_rmdir(d_inode(parent), child);
 401	if (status)
 402		printk("failed to remove client recovery directory %pd\n",
 403				child);
 
 
 
 404	/* Keep trying, success or failure: */
 405	return 0;
 406}
 407
 408static void
 409nfsd4_recdir_purge_old(struct nfsd_net *nn)
 410{
 411	int status;
 412
 413	nn->in_grace = false;
 414	if (!nn->rec_file)
 415		return;
 416	status = mnt_want_write_file(nn->rec_file);
 417	if (status)
 418		goto out;
 419	status = nfsd4_list_rec_dir(purge_old, nn);
 420	if (status == 0)
 421		vfs_fsync(nn->rec_file, 0);
 422	mnt_drop_write_file(nn->rec_file);
 423out:
 424	nfs4_release_reclaim(nn);
 425	if (status)
 426		printk("nfsd4: failed to purge old clients from recovery"
 427			" directory %pD\n", nn->rec_file);
 428}
 429
 430static int
 431load_recdir(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
 432{
 
 
 
 433	if (child->d_name.len != HEXDIR_LEN - 1) {
 434		printk("nfsd4: illegal name %pd in recovery directory\n",
 435				child);
 436		/* Keep trying; maybe the others are OK: */
 437		return 0;
 438	}
 439	nfs4_client_to_reclaim(child->d_name.name, nn);
 
 
 
 
 
 
 
 
 
 440	return 0;
 441}
 442
 443static int
 444nfsd4_recdir_load(struct net *net) {
 445	int status;
 446	struct nfsd_net *nn =  net_generic(net, nfsd_net_id);
 447
 448	if (!nn->rec_file)
 449		return 0;
 450
 451	status = nfsd4_list_rec_dir(load_recdir, nn);
 452	if (status)
 453		printk("nfsd4: failed loading clients from recovery"
 454			" directory %pD\n", nn->rec_file);
 455	return status;
 456}
 457
 458/*
 459 * Hold reference to the recovery directory.
 460 */
 461
 462static int
 463nfsd4_init_recdir(struct net *net)
 464{
 465	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 466	const struct cred *original_cred;
 467	int status;
 468
 469	printk("NFSD: Using %s as the NFSv4 state recovery directory\n",
 470			user_recovery_dirname);
 471
 472	BUG_ON(nn->rec_file);
 473
 474	status = nfs4_save_creds(&original_cred);
 475	if (status < 0) {
 476		printk("NFSD: Unable to change credentials to find recovery"
 477		       " directory: error %d\n",
 478		       status);
 479		return status;
 480	}
 481
 482	nn->rec_file = filp_open(user_recovery_dirname, O_RDONLY | O_DIRECTORY, 0);
 483	if (IS_ERR(nn->rec_file)) {
 484		printk("NFSD: unable to find recovery directory %s\n",
 485				user_recovery_dirname);
 486		status = PTR_ERR(nn->rec_file);
 487		nn->rec_file = NULL;
 488	}
 489
 490	nfs4_reset_creds(original_cred);
 491	if (!status)
 492		nn->in_grace = true;
 493	return status;
 494}
 495
 496static void
 497nfsd4_shutdown_recdir(struct net *net)
 498{
 499	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 500
 501	if (!nn->rec_file)
 502		return;
 503	fput(nn->rec_file);
 504	nn->rec_file = NULL;
 505}
 506
 507static int
 508nfs4_legacy_state_init(struct net *net)
 509{
 510	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 511	int i;
 512
 513	nn->reclaim_str_hashtbl = kmalloc(sizeof(struct list_head) *
 514					  CLIENT_HASH_SIZE, GFP_KERNEL);
 
 515	if (!nn->reclaim_str_hashtbl)
 516		return -ENOMEM;
 517
 518	for (i = 0; i < CLIENT_HASH_SIZE; i++)
 519		INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
 520	nn->reclaim_str_hashtbl_size = 0;
 521
 522	return 0;
 523}
 524
 525static void
 526nfs4_legacy_state_shutdown(struct net *net)
 527{
 528	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 529
 530	kfree(nn->reclaim_str_hashtbl);
 531}
 532
 533static int
 534nfsd4_load_reboot_recovery_data(struct net *net)
 535{
 536	int status;
 537
 538	status = nfsd4_init_recdir(net);
 539	if (status)
 540		return status;
 541
 542	status = nfsd4_recdir_load(net);
 543	if (status)
 544		nfsd4_shutdown_recdir(net);
 545
 546	return status;
 547}
 548
 549static int
 550nfsd4_legacy_tracking_init(struct net *net)
 551{
 552	int status;
 553
 554	/* XXX: The legacy code won't work in a container */
 555	if (net != &init_net) {
 556		pr_warn("NFSD: attempt to initialize legacy client tracking in a container ignored.\n");
 557		return -EINVAL;
 558	}
 559
 560	status = nfs4_legacy_state_init(net);
 561	if (status)
 562		return status;
 563
 564	status = nfsd4_load_reboot_recovery_data(net);
 565	if (status)
 566		goto err;
 
 567	return 0;
 568
 569err:
 570	nfs4_legacy_state_shutdown(net);
 571	return status;
 572}
 573
 574static void
 575nfsd4_legacy_tracking_exit(struct net *net)
 576{
 577	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 578
 579	nfs4_release_reclaim(nn);
 580	nfsd4_shutdown_recdir(net);
 581	nfs4_legacy_state_shutdown(net);
 582}
 583
 584/*
 585 * Change the NFSv4 recovery directory to recdir.
 586 */
 587int
 588nfs4_reset_recoverydir(char *recdir)
 589{
 590	int status;
 591	struct path path;
 592
 593	status = kern_path(recdir, LOOKUP_FOLLOW, &path);
 594	if (status)
 595		return status;
 596	status = -ENOTDIR;
 597	if (d_is_dir(path.dentry)) {
 598		strcpy(user_recovery_dirname, recdir);
 599		status = 0;
 600	}
 601	path_put(&path);
 602	return status;
 603}
 604
 605char *
 606nfs4_recoverydir(void)
 607{
 608	return user_recovery_dirname;
 609}
 610
 611static int
 612nfsd4_check_legacy_client(struct nfs4_client *clp)
 613{
 614	int status;
 615	char dname[HEXDIR_LEN];
 616	struct nfs4_client_reclaim *crp;
 617	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 
 618
 619	/* did we already find that this client is stable? */
 620	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 621		return 0;
 622
 623	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
 624	if (status) {
 625		legacy_recdir_name_error(clp, status);
 626		return status;
 627	}
 628
 629	/* look for it in the reclaim hashtable otherwise */
 630	crp = nfsd4_find_reclaim_client(dname, nn);
 
 
 
 
 
 
 
 
 631	if (crp) {
 632		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
 633		crp->cr_clp = clp;
 634		return 0;
 635	}
 636
 
 637	return -ENOENT;
 638}
 639
 640static const struct nfsd4_client_tracking_ops nfsd4_legacy_tracking_ops = {
 641	.init		= nfsd4_legacy_tracking_init,
 642	.exit		= nfsd4_legacy_tracking_exit,
 643	.create		= nfsd4_create_clid_dir,
 644	.remove		= nfsd4_remove_clid_dir,
 645	.check		= nfsd4_check_legacy_client,
 646	.grace_done	= nfsd4_recdir_purge_old,
 
 
 647};
 648
 649/* Globals */
 650#define NFSD_PIPE_DIR		"nfsd"
 651#define NFSD_CLD_PIPE		"cld"
 652
 653/* per-net-ns structure for holding cld upcall info */
 654struct cld_net {
 655	struct rpc_pipe		*cn_pipe;
 656	spinlock_t		 cn_lock;
 657	struct list_head	 cn_list;
 658	unsigned int		 cn_xid;
 
 
 659};
 660
 661struct cld_upcall {
 662	struct list_head	 cu_list;
 663	struct cld_net		*cu_net;
 664	struct task_struct	*cu_task;
 665	struct cld_msg		 cu_msg;
 
 
 
 
 666};
 667
 668static int
 669__cld_pipe_upcall(struct rpc_pipe *pipe, struct cld_msg *cmsg)
 670{
 671	int ret;
 672	struct rpc_pipe_msg msg;
 
 
 
 673
 674	memset(&msg, 0, sizeof(msg));
 675	msg.data = cmsg;
 676	msg.len = sizeof(*cmsg);
 677
 678	/*
 679	 * Set task state before we queue the upcall. That prevents
 680	 * wake_up_process in the downcall from racing with schedule.
 681	 */
 682	set_current_state(TASK_UNINTERRUPTIBLE);
 683	ret = rpc_queue_upcall(pipe, &msg);
 684	if (ret < 0) {
 685		set_current_state(TASK_RUNNING);
 686		goto out;
 687	}
 688
 689	schedule();
 690
 691	if (msg.errno < 0)
 692		ret = msg.errno;
 693out:
 694	return ret;
 695}
 696
 697static int
 698cld_pipe_upcall(struct rpc_pipe *pipe, struct cld_msg *cmsg)
 699{
 700	int ret;
 701
 702	/*
 703	 * -EAGAIN occurs when pipe is closed and reopened while there are
 704	 *  upcalls queued.
 705	 */
 706	do {
 707		ret = __cld_pipe_upcall(pipe, cmsg);
 708	} while (ret == -EAGAIN);
 709
 710	return ret;
 711}
 712
 713static ssize_t
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 714cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
 715{
 716	struct cld_upcall *tmp, *cup;
 717	struct cld_msg __user *cmsg = (struct cld_msg __user *)src;
 
 718	uint32_t xid;
 719	struct nfsd_net *nn = net_generic(file_inode(filp)->i_sb->s_fs_info,
 720						nfsd_net_id);
 721	struct cld_net *cn = nn->cld_net;
 
 722
 723	if (mlen != sizeof(*cmsg)) {
 724		dprintk("%s: got %zu bytes, expected %zu\n", __func__, mlen,
 725			sizeof(*cmsg));
 726		return -EINVAL;
 727	}
 728
 729	/* copy just the xid so we can try to find that */
 730	if (copy_from_user(&xid, &cmsg->cm_xid, sizeof(xid)) != 0) {
 731		dprintk("%s: error when copying xid from userspace", __func__);
 732		return -EFAULT;
 733	}
 734
 
 
 
 
 
 
 
 
 
 
 735	/* walk the list and find corresponding xid */
 736	cup = NULL;
 737	spin_lock(&cn->cn_lock);
 738	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
 739		if (get_unaligned(&tmp->cu_msg.cm_xid) == xid) {
 740			cup = tmp;
 741			list_del_init(&cup->cu_list);
 
 742			break;
 743		}
 744	}
 745	spin_unlock(&cn->cn_lock);
 746
 747	/* couldn't find upcall? */
 748	if (!cup) {
 749		dprintk("%s: couldn't find upcall -- xid=%u\n", __func__, xid);
 750		return -EINVAL;
 751	}
 752
 753	if (copy_from_user(&cup->cu_msg, src, mlen) != 0)
 
 
 
 754		return -EFAULT;
 755
 756	wake_up_process(cup->cu_task);
 757	return mlen;
 758}
 759
 760static void
 761cld_pipe_destroy_msg(struct rpc_pipe_msg *msg)
 762{
 763	struct cld_msg *cmsg = msg->data;
 764	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall,
 765						 cu_msg);
 766
 767	/* errno >= 0 means we got a downcall */
 768	if (msg->errno >= 0)
 769		return;
 770
 771	wake_up_process(cup->cu_task);
 772}
 773
 774static const struct rpc_pipe_ops cld_upcall_ops = {
 775	.upcall		= rpc_pipe_generic_upcall,
 776	.downcall	= cld_pipe_downcall,
 777	.destroy_msg	= cld_pipe_destroy_msg,
 778};
 779
 780static struct dentry *
 781nfsd4_cld_register_sb(struct super_block *sb, struct rpc_pipe *pipe)
 782{
 783	struct dentry *dir, *dentry;
 784
 785	dir = rpc_d_lookup_sb(sb, NFSD_PIPE_DIR);
 786	if (dir == NULL)
 787		return ERR_PTR(-ENOENT);
 788	dentry = rpc_mkpipe_dentry(dir, NFSD_CLD_PIPE, NULL, pipe);
 789	dput(dir);
 790	return dentry;
 791}
 792
 793static void
 794nfsd4_cld_unregister_sb(struct rpc_pipe *pipe)
 795{
 796	if (pipe->dentry)
 797		rpc_unlink(pipe->dentry);
 798}
 799
 800static struct dentry *
 801nfsd4_cld_register_net(struct net *net, struct rpc_pipe *pipe)
 802{
 803	struct super_block *sb;
 804	struct dentry *dentry;
 805
 806	sb = rpc_get_sb_net(net);
 807	if (!sb)
 808		return NULL;
 809	dentry = nfsd4_cld_register_sb(sb, pipe);
 810	rpc_put_sb_net(net);
 811	return dentry;
 812}
 813
 814static void
 815nfsd4_cld_unregister_net(struct net *net, struct rpc_pipe *pipe)
 816{
 817	struct super_block *sb;
 818
 819	sb = rpc_get_sb_net(net);
 820	if (sb) {
 821		nfsd4_cld_unregister_sb(pipe);
 822		rpc_put_sb_net(net);
 823	}
 824}
 825
 826/* Initialize rpc_pipefs pipe for communication with client tracking daemon */
 827static int
 828nfsd4_init_cld_pipe(struct net *net)
 829{
 830	int ret;
 831	struct dentry *dentry;
 832	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 833	struct cld_net *cn;
 834
 835	if (nn->cld_net)
 836		return 0;
 837
 838	cn = kzalloc(sizeof(*cn), GFP_KERNEL);
 839	if (!cn) {
 840		ret = -ENOMEM;
 841		goto err;
 842	}
 843
 844	cn->cn_pipe = rpc_mkpipe_data(&cld_upcall_ops, RPC_PIPE_WAIT_FOR_OPEN);
 845	if (IS_ERR(cn->cn_pipe)) {
 846		ret = PTR_ERR(cn->cn_pipe);
 847		goto err;
 848	}
 849	spin_lock_init(&cn->cn_lock);
 850	INIT_LIST_HEAD(&cn->cn_list);
 851
 852	dentry = nfsd4_cld_register_net(net, cn->cn_pipe);
 853	if (IS_ERR(dentry)) {
 854		ret = PTR_ERR(dentry);
 855		goto err_destroy_data;
 856	}
 857
 858	cn->cn_pipe->dentry = dentry;
 
 859	nn->cld_net = cn;
 860	return 0;
 861
 862err_destroy_data:
 863	rpc_destroy_pipe_data(cn->cn_pipe);
 864err:
 865	kfree(cn);
 866	printk(KERN_ERR "NFSD: unable to create nfsdcld upcall pipe (%d)\n",
 867			ret);
 868	return ret;
 869}
 870
 
 
 
 
 
 
 
 
 
 
 
 871static void
 872nfsd4_remove_cld_pipe(struct net *net)
 873{
 874	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 875	struct cld_net *cn = nn->cld_net;
 876
 877	nfsd4_cld_unregister_net(net, cn->cn_pipe);
 878	rpc_destroy_pipe_data(cn->cn_pipe);
 
 
 879	kfree(nn->cld_net);
 880	nn->cld_net = NULL;
 881}
 882
 883static struct cld_upcall *
 884alloc_cld_upcall(struct cld_net *cn)
 885{
 886	struct cld_upcall *new, *tmp;
 
 887
 888	new = kzalloc(sizeof(*new), GFP_KERNEL);
 889	if (!new)
 890		return new;
 891
 892	/* FIXME: hard cap on number in flight? */
 893restart_search:
 894	spin_lock(&cn->cn_lock);
 895	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
 896		if (tmp->cu_msg.cm_xid == cn->cn_xid) {
 897			cn->cn_xid++;
 898			spin_unlock(&cn->cn_lock);
 899			goto restart_search;
 900		}
 901	}
 902	new->cu_task = current;
 903	new->cu_msg.cm_vers = CLD_UPCALL_VERSION;
 904	put_unaligned(cn->cn_xid++, &new->cu_msg.cm_xid);
 905	new->cu_net = cn;
 906	list_add(&new->cu_list, &cn->cn_list);
 907	spin_unlock(&cn->cn_lock);
 908
 909	dprintk("%s: allocated xid %u\n", __func__, new->cu_msg.cm_xid);
 910
 911	return new;
 912}
 913
 914static void
 915free_cld_upcall(struct cld_upcall *victim)
 916{
 917	struct cld_net *cn = victim->cu_net;
 918
 919	spin_lock(&cn->cn_lock);
 920	list_del(&victim->cu_list);
 921	spin_unlock(&cn->cn_lock);
 922	kfree(victim);
 923}
 924
 925/* Ask daemon to create a new record */
 926static void
 927nfsd4_cld_create(struct nfs4_client *clp)
 928{
 929	int ret;
 930	struct cld_upcall *cup;
 931	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 932	struct cld_net *cn = nn->cld_net;
 933
 934	/* Don't upcall if it's already stored */
 935	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 936		return;
 937
 938	cup = alloc_cld_upcall(cn);
 939	if (!cup) {
 940		ret = -ENOMEM;
 941		goto out_err;
 942	}
 943
 944	cup->cu_msg.cm_cmd = Cld_Create;
 945	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
 946	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
 947			clp->cl_name.len);
 948
 949	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
 950	if (!ret) {
 951		ret = cup->cu_msg.cm_status;
 952		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
 953	}
 954
 955	free_cld_upcall(cup);
 956out_err:
 957	if (ret)
 958		printk(KERN_ERR "NFSD: Unable to create client "
 959				"record on stable storage: %d\n", ret);
 960}
 961
 962/* Ask daemon to create a new record */
 963static void
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 964nfsd4_cld_remove(struct nfs4_client *clp)
 965{
 966	int ret;
 967	struct cld_upcall *cup;
 968	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 969	struct cld_net *cn = nn->cld_net;
 970
 971	/* Don't upcall if it's already removed */
 972	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 973		return;
 974
 975	cup = alloc_cld_upcall(cn);
 976	if (!cup) {
 977		ret = -ENOMEM;
 978		goto out_err;
 979	}
 980
 981	cup->cu_msg.cm_cmd = Cld_Remove;
 982	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
 983	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
 984			clp->cl_name.len);
 985
 986	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
 987	if (!ret) {
 988		ret = cup->cu_msg.cm_status;
 989		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
 990	}
 991
 992	free_cld_upcall(cup);
 993out_err:
 994	if (ret)
 995		printk(KERN_ERR "NFSD: Unable to remove client "
 996				"record from stable storage: %d\n", ret);
 997}
 998
 999/* Check for presence of a record, and update its timestamp */
 
 
 
 
 
1000static int
1001nfsd4_cld_check(struct nfs4_client *clp)
1002{
1003	int ret;
1004	struct cld_upcall *cup;
1005	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1006	struct cld_net *cn = nn->cld_net;
1007
1008	/* Don't upcall if one was already stored during this grace pd */
1009	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1010		return 0;
1011
1012	cup = alloc_cld_upcall(cn);
1013	if (!cup) {
1014		printk(KERN_ERR "NFSD: Unable to check client record on "
1015				"stable storage: %d\n", -ENOMEM);
1016		return -ENOMEM;
1017	}
1018
1019	cup->cu_msg.cm_cmd = Cld_Check;
1020	cup->cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1021	memcpy(cup->cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1022			clp->cl_name.len);
1023
1024	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
1025	if (!ret) {
1026		ret = cup->cu_msg.cm_status;
1027		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1028	}
1029
1030	free_cld_upcall(cup);
1031	return ret;
1032}
1033
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1034static void
1035nfsd4_cld_grace_done(struct nfsd_net *nn)
1036{
1037	int ret;
1038	struct cld_upcall *cup;
1039	struct cld_net *cn = nn->cld_net;
1040
1041	cup = alloc_cld_upcall(cn);
1042	if (!cup) {
1043		ret = -ENOMEM;
1044		goto out_err;
1045	}
1046
1047	cup->cu_msg.cm_cmd = Cld_GraceDone;
1048	cup->cu_msg.cm_u.cm_gracetime = (int64_t)nn->boot_time;
1049	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_msg);
1050	if (!ret)
1051		ret = cup->cu_msg.cm_status;
1052
1053	free_cld_upcall(cup);
1054out_err:
 
1055	if (ret)
1056		printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1057}
1058
1059static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops = {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1060	.init		= nfsd4_init_cld_pipe,
1061	.exit		= nfsd4_remove_cld_pipe,
1062	.create		= nfsd4_cld_create,
1063	.remove		= nfsd4_cld_remove,
 
 
 
 
 
 
 
 
 
 
 
 
1064	.check		= nfsd4_cld_check,
1065	.grace_done	= nfsd4_cld_grace_done,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1066};
1067
1068/* upcall via usermodehelper */
1069static char cltrack_prog[PATH_MAX] = "/sbin/nfsdcltrack";
1070module_param_string(cltrack_prog, cltrack_prog, sizeof(cltrack_prog),
1071			S_IRUGO|S_IWUSR);
1072MODULE_PARM_DESC(cltrack_prog, "Path to the nfsdcltrack upcall program");
1073
1074static bool cltrack_legacy_disable;
1075module_param(cltrack_legacy_disable, bool, S_IRUGO|S_IWUSR);
1076MODULE_PARM_DESC(cltrack_legacy_disable,
1077		"Disable legacy recoverydir conversion. Default: false");
1078
1079#define LEGACY_TOPDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_TOPDIR="
1080#define LEGACY_RECDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_RECDIR="
1081#define HAS_SESSION_ENV_PREFIX "NFSDCLTRACK_CLIENT_HAS_SESSION="
1082#define GRACE_START_ENV_PREFIX "NFSDCLTRACK_GRACE_START="
1083
1084static char *
1085nfsd4_cltrack_legacy_topdir(void)
1086{
1087	int copied;
1088	size_t len;
1089	char *result;
1090
1091	if (cltrack_legacy_disable)
1092		return NULL;
1093
1094	len = strlen(LEGACY_TOPDIR_ENV_PREFIX) +
1095		strlen(nfs4_recoverydir()) + 1;
1096
1097	result = kmalloc(len, GFP_KERNEL);
1098	if (!result)
1099		return result;
1100
1101	copied = snprintf(result, len, LEGACY_TOPDIR_ENV_PREFIX "%s",
1102				nfs4_recoverydir());
1103	if (copied >= len) {
1104		/* just return nothing if output was truncated */
1105		kfree(result);
1106		return NULL;
1107	}
1108
1109	return result;
1110}
1111
1112static char *
1113nfsd4_cltrack_legacy_recdir(const struct xdr_netobj *name)
1114{
1115	int copied;
1116	size_t len;
1117	char *result;
1118
1119	if (cltrack_legacy_disable)
1120		return NULL;
1121
1122	/* +1 is for '/' between "topdir" and "recdir" */
1123	len = strlen(LEGACY_RECDIR_ENV_PREFIX) +
1124		strlen(nfs4_recoverydir()) + 1 + HEXDIR_LEN;
1125
1126	result = kmalloc(len, GFP_KERNEL);
1127	if (!result)
1128		return result;
1129
1130	copied = snprintf(result, len, LEGACY_RECDIR_ENV_PREFIX "%s/",
1131				nfs4_recoverydir());
1132	if (copied > (len - HEXDIR_LEN)) {
1133		/* just return nothing if output will be truncated */
1134		kfree(result);
1135		return NULL;
1136	}
1137
1138	copied = nfs4_make_rec_clidname(result + copied, name);
1139	if (copied) {
1140		kfree(result);
1141		return NULL;
1142	}
1143
1144	return result;
1145}
1146
1147static char *
1148nfsd4_cltrack_client_has_session(struct nfs4_client *clp)
1149{
1150	int copied;
1151	size_t len;
1152	char *result;
1153
1154	/* prefix + Y/N character + terminating NULL */
1155	len = strlen(HAS_SESSION_ENV_PREFIX) + 1 + 1;
1156
1157	result = kmalloc(len, GFP_KERNEL);
1158	if (!result)
1159		return result;
1160
1161	copied = snprintf(result, len, HAS_SESSION_ENV_PREFIX "%c",
1162				clp->cl_minorversion ? 'Y' : 'N');
1163	if (copied >= len) {
1164		/* just return nothing if output was truncated */
1165		kfree(result);
1166		return NULL;
1167	}
1168
1169	return result;
1170}
1171
1172static char *
1173nfsd4_cltrack_grace_start(time_t grace_start)
1174{
1175	int copied;
1176	size_t len;
1177	char *result;
1178
1179	/* prefix + max width of int64_t string + terminating NULL */
1180	len = strlen(GRACE_START_ENV_PREFIX) + 22 + 1;
1181
1182	result = kmalloc(len, GFP_KERNEL);
1183	if (!result)
1184		return result;
1185
1186	copied = snprintf(result, len, GRACE_START_ENV_PREFIX "%ld",
1187				grace_start);
1188	if (copied >= len) {
1189		/* just return nothing if output was truncated */
1190		kfree(result);
1191		return NULL;
1192	}
1193
1194	return result;
1195}
1196
1197static int
1198nfsd4_umh_cltrack_upcall(char *cmd, char *arg, char *env0, char *env1)
1199{
1200	char *envp[3];
1201	char *argv[4];
1202	int ret;
1203
1204	if (unlikely(!cltrack_prog[0])) {
1205		dprintk("%s: cltrack_prog is disabled\n", __func__);
1206		return -EACCES;
1207	}
1208
1209	dprintk("%s: cmd: %s\n", __func__, cmd);
1210	dprintk("%s: arg: %s\n", __func__, arg ? arg : "(null)");
1211	dprintk("%s: env0: %s\n", __func__, env0 ? env0 : "(null)");
1212	dprintk("%s: env1: %s\n", __func__, env1 ? env1 : "(null)");
1213
1214	envp[0] = env0;
1215	envp[1] = env1;
1216	envp[2] = NULL;
1217
1218	argv[0] = (char *)cltrack_prog;
1219	argv[1] = cmd;
1220	argv[2] = arg;
1221	argv[3] = NULL;
1222
1223	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1224	/*
1225	 * Disable the upcall mechanism if we're getting an ENOENT or EACCES
1226	 * error. The admin can re-enable it on the fly by using sysfs
1227	 * once the problem has been fixed.
1228	 */
1229	if (ret == -ENOENT || ret == -EACCES) {
1230		dprintk("NFSD: %s was not found or isn't executable (%d). "
1231			"Setting cltrack_prog to blank string!",
1232			cltrack_prog, ret);
1233		cltrack_prog[0] = '\0';
1234	}
1235	dprintk("%s: %s return value: %d\n", __func__, cltrack_prog, ret);
1236
1237	return ret;
1238}
1239
1240static char *
1241bin_to_hex_dup(const unsigned char *src, int srclen)
1242{
1243	int i;
1244	char *buf, *hex;
1245
1246	/* +1 for terminating NULL */
1247	buf = kmalloc((srclen * 2) + 1, GFP_KERNEL);
1248	if (!buf)
1249		return buf;
1250
1251	hex = buf;
1252	for (i = 0; i < srclen; i++) {
1253		sprintf(hex, "%2.2x", *src++);
1254		hex += 2;
1255	}
1256	return buf;
1257}
1258
1259static int
1260nfsd4_umh_cltrack_init(struct net *net)
1261{
1262	int ret;
1263	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1264	char *grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1265
1266	/* XXX: The usermode helper s not working in container yet. */
1267	if (net != &init_net) {
1268		pr_warn("NFSD: attempt to initialize umh client tracking in a container ignored.\n");
1269		kfree(grace_start);
1270		return -EINVAL;
1271	}
1272
1273	ret = nfsd4_umh_cltrack_upcall("init", NULL, grace_start, NULL);
1274	kfree(grace_start);
 
 
1275	return ret;
1276}
1277
1278static void
1279nfsd4_cltrack_upcall_lock(struct nfs4_client *clp)
1280{
1281	wait_on_bit_lock(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK,
1282			 TASK_UNINTERRUPTIBLE);
1283}
1284
1285static void
1286nfsd4_cltrack_upcall_unlock(struct nfs4_client *clp)
1287{
1288	smp_mb__before_atomic();
1289	clear_bit(NFSD4_CLIENT_UPCALL_LOCK, &clp->cl_flags);
1290	smp_mb__after_atomic();
1291	wake_up_bit(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK);
1292}
1293
1294static void
1295nfsd4_umh_cltrack_create(struct nfs4_client *clp)
1296{
1297	char *hexid, *has_session, *grace_start;
1298	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1299
1300	/*
1301	 * With v4.0 clients, there's little difference in outcome between a
1302	 * create and check operation, and we can end up calling into this
1303	 * function multiple times per client (once for each openowner). So,
1304	 * for v4.0 clients skip upcalling once the client has been recorded
1305	 * on stable storage.
1306	 *
1307	 * For v4.1+ clients, the outcome of the two operations is different,
1308	 * so we must ensure that we upcall for the create operation. v4.1+
1309	 * clients call this on RECLAIM_COMPLETE though, so we should only end
1310	 * up doing a single create upcall per client.
1311	 */
1312	if (clp->cl_minorversion == 0 &&
1313	    test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1314		return;
1315
1316	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1317	if (!hexid) {
1318		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1319		return;
1320	}
1321
1322	has_session = nfsd4_cltrack_client_has_session(clp);
1323	grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1324
1325	nfsd4_cltrack_upcall_lock(clp);
1326	if (!nfsd4_umh_cltrack_upcall("create", hexid, has_session, grace_start))
1327		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1328	nfsd4_cltrack_upcall_unlock(clp);
1329
1330	kfree(has_session);
1331	kfree(grace_start);
1332	kfree(hexid);
1333}
1334
1335static void
1336nfsd4_umh_cltrack_remove(struct nfs4_client *clp)
1337{
1338	char *hexid;
1339
1340	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1341		return;
1342
1343	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1344	if (!hexid) {
1345		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1346		return;
1347	}
1348
1349	nfsd4_cltrack_upcall_lock(clp);
1350	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags) &&
1351	    nfsd4_umh_cltrack_upcall("remove", hexid, NULL, NULL) == 0)
1352		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1353	nfsd4_cltrack_upcall_unlock(clp);
1354
1355	kfree(hexid);
1356}
1357
1358static int
1359nfsd4_umh_cltrack_check(struct nfs4_client *clp)
1360{
1361	int ret;
1362	char *hexid, *has_session, *legacy;
1363
1364	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1365		return 0;
1366
1367	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1368	if (!hexid) {
1369		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1370		return -ENOMEM;
1371	}
1372
1373	has_session = nfsd4_cltrack_client_has_session(clp);
1374	legacy = nfsd4_cltrack_legacy_recdir(&clp->cl_name);
1375
1376	nfsd4_cltrack_upcall_lock(clp);
1377	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags)) {
1378		ret = 0;
1379	} else {
1380		ret = nfsd4_umh_cltrack_upcall("check", hexid, has_session, legacy);
1381		if (ret == 0)
1382			set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1383	}
1384	nfsd4_cltrack_upcall_unlock(clp);
1385	kfree(has_session);
1386	kfree(legacy);
1387	kfree(hexid);
1388
1389	return ret;
1390}
1391
1392static void
1393nfsd4_umh_cltrack_grace_done(struct nfsd_net *nn)
1394{
1395	char *legacy;
1396	char timestr[22]; /* FIXME: better way to determine max size? */
1397
1398	sprintf(timestr, "%ld", nn->boot_time);
1399	legacy = nfsd4_cltrack_legacy_topdir();
1400	nfsd4_umh_cltrack_upcall("gracedone", timestr, legacy, NULL);
1401	kfree(legacy);
1402}
1403
1404static const struct nfsd4_client_tracking_ops nfsd4_umh_tracking_ops = {
1405	.init		= nfsd4_umh_cltrack_init,
1406	.exit		= NULL,
1407	.create		= nfsd4_umh_cltrack_create,
1408	.remove		= nfsd4_umh_cltrack_remove,
1409	.check		= nfsd4_umh_cltrack_check,
1410	.grace_done	= nfsd4_umh_cltrack_grace_done,
 
 
1411};
1412
1413int
1414nfsd4_client_tracking_init(struct net *net)
1415{
1416	int status;
1417	struct path path;
1418	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1419
1420	/* just run the init if it the method is already decided */
1421	if (nn->client_tracking_ops)
1422		goto do_init;
1423
 
 
 
 
 
 
 
 
 
 
 
 
1424	/*
1425	 * First, try a UMH upcall. It should succeed or fail quickly, so
1426	 * there's little harm in trying that first.
1427	 */
1428	nn->client_tracking_ops = &nfsd4_umh_tracking_ops;
1429	status = nn->client_tracking_ops->init(net);
1430	if (!status)
1431		return status;
1432
1433	/*
1434	 * See if the recoverydir exists and is a directory. If it is,
1435	 * then use the legacy ops.
1436	 */
1437	nn->client_tracking_ops = &nfsd4_legacy_tracking_ops;
1438	status = kern_path(nfs4_recoverydir(), LOOKUP_FOLLOW, &path);
1439	if (!status) {
1440		status = d_is_dir(path.dentry);
1441		path_put(&path);
1442		if (status)
1443			goto do_init;
 
 
1444	}
1445
1446	/* Finally, try to use nfsdcld */
1447	nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
1448	printk(KERN_WARNING "NFSD: the nfsdcld client tracking upcall will be "
1449			"removed in 3.10. Please transition to using "
1450			"nfsdcltrack.\n");
1451do_init:
1452	status = nn->client_tracking_ops->init(net);
 
1453	if (status) {
1454		printk(KERN_WARNING "NFSD: Unable to initialize client "
1455				    "recovery tracking! (%d)\n", status);
1456		nn->client_tracking_ops = NULL;
1457	}
1458	return status;
1459}
1460
1461void
1462nfsd4_client_tracking_exit(struct net *net)
1463{
1464	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1465
1466	if (nn->client_tracking_ops) {
1467		if (nn->client_tracking_ops->exit)
1468			nn->client_tracking_ops->exit(net);
1469		nn->client_tracking_ops = NULL;
1470	}
1471}
1472
1473void
1474nfsd4_client_record_create(struct nfs4_client *clp)
1475{
1476	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1477
1478	if (nn->client_tracking_ops)
1479		nn->client_tracking_ops->create(clp);
1480}
1481
1482void
1483nfsd4_client_record_remove(struct nfs4_client *clp)
1484{
1485	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1486
1487	if (nn->client_tracking_ops)
1488		nn->client_tracking_ops->remove(clp);
1489}
1490
1491int
1492nfsd4_client_record_check(struct nfs4_client *clp)
1493{
1494	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1495
1496	if (nn->client_tracking_ops)
1497		return nn->client_tracking_ops->check(clp);
1498
1499	return -EOPNOTSUPP;
1500}
1501
1502void
1503nfsd4_record_grace_done(struct nfsd_net *nn)
1504{
1505	if (nn->client_tracking_ops)
1506		nn->client_tracking_ops->grace_done(nn);
1507}
1508
1509static int
1510rpc_pipefs_event(struct notifier_block *nb, unsigned long event, void *ptr)
1511{
1512	struct super_block *sb = ptr;
1513	struct net *net = sb->s_fs_info;
1514	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1515	struct cld_net *cn = nn->cld_net;
1516	struct dentry *dentry;
1517	int ret = 0;
1518
1519	if (!try_module_get(THIS_MODULE))
1520		return 0;
1521
1522	if (!cn) {
1523		module_put(THIS_MODULE);
1524		return 0;
1525	}
1526
1527	switch (event) {
1528	case RPC_PIPEFS_MOUNT:
1529		dentry = nfsd4_cld_register_sb(sb, cn->cn_pipe);
1530		if (IS_ERR(dentry)) {
1531			ret = PTR_ERR(dentry);
1532			break;
1533		}
1534		cn->cn_pipe->dentry = dentry;
1535		break;
1536	case RPC_PIPEFS_UMOUNT:
1537		if (cn->cn_pipe->dentry)
1538			nfsd4_cld_unregister_sb(cn->cn_pipe);
1539		break;
1540	default:
1541		ret = -ENOTSUPP;
1542		break;
1543	}
1544	module_put(THIS_MODULE);
1545	return ret;
1546}
1547
1548static struct notifier_block nfsd4_cld_block = {
1549	.notifier_call = rpc_pipefs_event,
1550};
1551
1552int
1553register_cld_notifier(void)
1554{
1555	return rpc_pipefs_notifier_register(&nfsd4_cld_block);
1556}
1557
1558void
1559unregister_cld_notifier(void)
1560{
1561	rpc_pipefs_notifier_unregister(&nfsd4_cld_block);
1562}
v5.4
   1/*
   2*  Copyright (c) 2004 The Regents of the University of Michigan.
   3*  Copyright (c) 2012 Jeff Layton <jlayton@redhat.com>
   4*  All rights reserved.
   5*
   6*  Andy Adamson <andros@citi.umich.edu>
   7*
   8*  Redistribution and use in source and binary forms, with or without
   9*  modification, are permitted provided that the following conditions
  10*  are met:
  11*
  12*  1. Redistributions of source code must retain the above copyright
  13*     notice, this list of conditions and the following disclaimer.
  14*  2. Redistributions in binary form must reproduce the above copyright
  15*     notice, this list of conditions and the following disclaimer in the
  16*     documentation and/or other materials provided with the distribution.
  17*  3. Neither the name of the University nor the names of its
  18*     contributors may be used to endorse or promote products derived
  19*     from this software without specific prior written permission.
  20*
  21*  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22*  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23*  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24*  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25*  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26*  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27*  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28*  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29*  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30*  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31*  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32*
  33*/
  34
  35#include <crypto/hash.h>
  36#include <linux/file.h>
  37#include <linux/slab.h>
  38#include <linux/namei.h>
  39#include <linux/sched.h>
  40#include <linux/fs.h>
  41#include <linux/module.h>
  42#include <net/net_namespace.h>
  43#include <linux/sunrpc/rpc_pipe_fs.h>
  44#include <linux/sunrpc/clnt.h>
  45#include <linux/nfsd/cld.h>
  46
  47#include "nfsd.h"
  48#include "state.h"
  49#include "vfs.h"
  50#include "netns.h"
  51
  52#define NFSDDBG_FACILITY                NFSDDBG_PROC
  53
  54/* Declarations */
  55struct nfsd4_client_tracking_ops {
  56	int (*init)(struct net *);
  57	void (*exit)(struct net *);
  58	void (*create)(struct nfs4_client *);
  59	void (*remove)(struct nfs4_client *);
  60	int (*check)(struct nfs4_client *);
  61	void (*grace_done)(struct nfsd_net *);
  62	uint8_t version;
  63	size_t msglen;
  64};
  65
  66static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops;
  67static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v2;
  68
  69/* Globals */
  70static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
  71
  72static int
  73nfs4_save_creds(const struct cred **original_creds)
  74{
  75	struct cred *new;
  76
  77	new = prepare_creds();
  78	if (!new)
  79		return -ENOMEM;
  80
  81	new->fsuid = GLOBAL_ROOT_UID;
  82	new->fsgid = GLOBAL_ROOT_GID;
  83	*original_creds = override_creds(new);
  84	put_cred(new);
  85	return 0;
  86}
  87
  88static void
  89nfs4_reset_creds(const struct cred *original)
  90{
  91	revert_creds(original);
  92}
  93
  94static void
  95md5_to_hex(char *out, char *md5)
  96{
  97	int i;
  98
  99	for (i=0; i<16; i++) {
 100		unsigned char c = md5[i];
 101
 102		*out++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
 103		*out++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
 104	}
 105	*out = '\0';
 106}
 107
 108static int
 109nfs4_make_rec_clidname(char *dname, const struct xdr_netobj *clname)
 110{
 111	struct xdr_netobj cksum;
 112	struct crypto_shash *tfm;
 113	int status;
 114
 115	dprintk("NFSD: nfs4_make_rec_clidname for %.*s\n",
 116			clname->len, clname->data);
 117	tfm = crypto_alloc_shash("md5", 0, 0);
 118	if (IS_ERR(tfm)) {
 119		status = PTR_ERR(tfm);
 120		goto out_no_tfm;
 121	}
 122
 123	cksum.len = crypto_shash_digestsize(tfm);
 124	cksum.data = kmalloc(cksum.len, GFP_KERNEL);
 125	if (cksum.data == NULL) {
 126		status = -ENOMEM;
 127 		goto out;
 128	}
 129
 130	{
 131		SHASH_DESC_ON_STACK(desc, tfm);
 132
 133		desc->tfm = tfm;
 
 134
 135		status = crypto_shash_digest(desc, clname->data, clname->len,
 136					     cksum.data);
 137		shash_desc_zero(desc);
 138	}
 139
 140	if (status)
 141		goto out;
 142
 143	md5_to_hex(dname, cksum.data);
 144
 145	status = 0;
 146out:
 147	kfree(cksum.data);
 148	crypto_free_shash(tfm);
 149out_no_tfm:
 150	return status;
 151}
 152
 153/*
 154 * If we had an error generating the recdir name for the legacy tracker
 155 * then warn the admin. If the error doesn't appear to be transient,
 156 * then disable recovery tracking.
 157 */
 158static void
 159legacy_recdir_name_error(struct nfs4_client *clp, int error)
 160{
 161	printk(KERN_ERR "NFSD: unable to generate recoverydir "
 162			"name (%d).\n", error);
 163
 164	/*
 165	 * if the algorithm just doesn't exist, then disable the recovery
 166	 * tracker altogether. The crypto libs will generally return this if
 167	 * FIPS is enabled as well.
 168	 */
 169	if (error == -ENOENT) {
 170		printk(KERN_ERR "NFSD: disabling legacy clientid tracking. "
 171			"Reboot recovery will not function correctly!\n");
 172		nfsd4_client_tracking_exit(clp->net);
 173	}
 174}
 175
 176static void
 177__nfsd4_create_reclaim_record_grace(struct nfs4_client *clp,
 178		const char *dname, int len, struct nfsd_net *nn)
 179{
 180	struct xdr_netobj name;
 181	struct xdr_netobj princhash = { .len = 0, .data = NULL };
 182	struct nfs4_client_reclaim *crp;
 183
 184	name.data = kmemdup(dname, len, GFP_KERNEL);
 185	if (!name.data) {
 186		dprintk("%s: failed to allocate memory for name.data!\n",
 187			__func__);
 188		return;
 189	}
 190	name.len = len;
 191	crp = nfs4_client_to_reclaim(name, princhash, nn);
 192	if (!crp) {
 193		kfree(name.data);
 194		return;
 195	}
 196	crp->cr_clp = clp;
 197}
 198
 199static void
 200nfsd4_create_clid_dir(struct nfs4_client *clp)
 201{
 202	const struct cred *original_cred;
 203	char dname[HEXDIR_LEN];
 204	struct dentry *dir, *dentry;
 
 205	int status;
 206	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 207
 208	if (test_and_set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 209		return;
 210	if (!nn->rec_file)
 211		return;
 212
 213	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
 214	if (status)
 215		return legacy_recdir_name_error(clp, status);
 216
 217	status = nfs4_save_creds(&original_cred);
 218	if (status < 0)
 219		return;
 220
 221	status = mnt_want_write_file(nn->rec_file);
 222	if (status)
 223		goto out_creds;
 224
 225	dir = nn->rec_file->f_path.dentry;
 226	/* lock the parent */
 227	inode_lock(d_inode(dir));
 228
 229	dentry = lookup_one_len(dname, dir, HEXDIR_LEN-1);
 230	if (IS_ERR(dentry)) {
 231		status = PTR_ERR(dentry);
 232		goto out_unlock;
 233	}
 234	if (d_really_is_positive(dentry))
 235		/*
 236		 * In the 4.1 case, where we're called from
 237		 * reclaim_complete(), records from the previous reboot
 238		 * may still be left, so this is OK.
 239		 *
 240		 * In the 4.0 case, we should never get here; but we may
 241		 * as well be forgiving and just succeed silently.
 242		 */
 243		goto out_put;
 244	status = vfs_mkdir(d_inode(dir), dentry, S_IRWXU);
 245out_put:
 246	dput(dentry);
 247out_unlock:
 248	inode_unlock(d_inode(dir));
 249	if (status == 0) {
 250		if (nn->in_grace)
 251			__nfsd4_create_reclaim_record_grace(clp, dname,
 252					HEXDIR_LEN, nn);
 
 
 253		vfs_fsync(nn->rec_file, 0);
 254	} else {
 255		printk(KERN_ERR "NFSD: failed to write recovery record"
 256				" (err %d); please check that %s exists"
 257				" and is writeable", status,
 258				user_recovery_dirname);
 259	}
 260	mnt_drop_write_file(nn->rec_file);
 261out_creds:
 262	nfs4_reset_creds(original_cred);
 263}
 264
 265typedef int (recdir_func)(struct dentry *, struct dentry *, struct nfsd_net *);
 266
 267struct name_list {
 268	char name[HEXDIR_LEN];
 269	struct list_head list;
 270};
 271
 272struct nfs4_dir_ctx {
 273	struct dir_context ctx;
 274	struct list_head names;
 275};
 276
 277static int
 278nfsd4_build_namelist(struct dir_context *__ctx, const char *name, int namlen,
 279		loff_t offset, u64 ino, unsigned int d_type)
 280{
 281	struct nfs4_dir_ctx *ctx =
 282		container_of(__ctx, struct nfs4_dir_ctx, ctx);
 283	struct name_list *entry;
 284
 285	if (namlen != HEXDIR_LEN - 1)
 286		return 0;
 287	entry = kmalloc(sizeof(struct name_list), GFP_KERNEL);
 288	if (entry == NULL)
 289		return -ENOMEM;
 290	memcpy(entry->name, name, HEXDIR_LEN - 1);
 291	entry->name[HEXDIR_LEN - 1] = '\0';
 292	list_add(&entry->list, &ctx->names);
 293	return 0;
 294}
 295
 296static int
 297nfsd4_list_rec_dir(recdir_func *f, struct nfsd_net *nn)
 298{
 299	const struct cred *original_cred;
 300	struct dentry *dir = nn->rec_file->f_path.dentry;
 301	struct nfs4_dir_ctx ctx = {
 302		.ctx.actor = nfsd4_build_namelist,
 303		.names = LIST_HEAD_INIT(ctx.names)
 304	};
 305	struct name_list *entry, *tmp;
 306	int status;
 307
 308	status = nfs4_save_creds(&original_cred);
 309	if (status < 0)
 310		return status;
 311
 312	status = vfs_llseek(nn->rec_file, 0, SEEK_SET);
 313	if (status < 0) {
 314		nfs4_reset_creds(original_cred);
 315		return status;
 316	}
 317
 318	status = iterate_dir(nn->rec_file, &ctx.ctx);
 319	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 320
 321	list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
 322		if (!status) {
 323			struct dentry *dentry;
 324			dentry = lookup_one_len(entry->name, dir, HEXDIR_LEN-1);
 325			if (IS_ERR(dentry)) {
 326				status = PTR_ERR(dentry);
 327				break;
 328			}
 329			status = f(dir, dentry, nn);
 330			dput(dentry);
 331		}
 332		list_del(&entry->list);
 333		kfree(entry);
 334	}
 335	inode_unlock(d_inode(dir));
 336	nfs4_reset_creds(original_cred);
 337
 338	list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
 339		dprintk("NFSD: %s. Left entry %s\n", __func__, entry->name);
 340		list_del(&entry->list);
 341		kfree(entry);
 342	}
 343	return status;
 344}
 345
 346static int
 347nfsd4_unlink_clid_dir(char *name, int namlen, struct nfsd_net *nn)
 348{
 349	struct dentry *dir, *dentry;
 350	int status;
 351
 352	dprintk("NFSD: nfsd4_unlink_clid_dir. name %.*s\n", namlen, name);
 353
 354	dir = nn->rec_file->f_path.dentry;
 355	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 356	dentry = lookup_one_len(name, dir, namlen);
 357	if (IS_ERR(dentry)) {
 358		status = PTR_ERR(dentry);
 359		goto out_unlock;
 360	}
 361	status = -ENOENT;
 362	if (d_really_is_negative(dentry))
 363		goto out;
 364	status = vfs_rmdir(d_inode(dir), dentry);
 365out:
 366	dput(dentry);
 367out_unlock:
 368	inode_unlock(d_inode(dir));
 369	return status;
 370}
 371
 372static void
 373__nfsd4_remove_reclaim_record_grace(const char *dname, int len,
 374		struct nfsd_net *nn)
 375{
 376	struct xdr_netobj name;
 377	struct nfs4_client_reclaim *crp;
 378
 379	name.data = kmemdup(dname, len, GFP_KERNEL);
 380	if (!name.data) {
 381		dprintk("%s: failed to allocate memory for name.data!\n",
 382			__func__);
 383		return;
 384	}
 385	name.len = len;
 386	crp = nfsd4_find_reclaim_client(name, nn);
 387	kfree(name.data);
 388	if (crp)
 389		nfs4_remove_reclaim_record(crp, nn);
 390}
 391
 392static void
 393nfsd4_remove_clid_dir(struct nfs4_client *clp)
 394{
 395	const struct cred *original_cred;
 
 396	char dname[HEXDIR_LEN];
 397	int status;
 398	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 399
 400	if (!nn->rec_file || !test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 401		return;
 402
 403	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
 404	if (status)
 405		return legacy_recdir_name_error(clp, status);
 406
 407	status = mnt_want_write_file(nn->rec_file);
 408	if (status)
 409		goto out;
 410	clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
 411
 412	status = nfs4_save_creds(&original_cred);
 413	if (status < 0)
 414		goto out_drop_write;
 415
 416	status = nfsd4_unlink_clid_dir(dname, HEXDIR_LEN-1, nn);
 417	nfs4_reset_creds(original_cred);
 418	if (status == 0) {
 419		vfs_fsync(nn->rec_file, 0);
 420		if (nn->in_grace)
 421			__nfsd4_remove_reclaim_record_grace(dname,
 422					HEXDIR_LEN, nn);
 
 
 
 423	}
 424out_drop_write:
 425	mnt_drop_write_file(nn->rec_file);
 426out:
 427	if (status)
 428		printk("NFSD: Failed to remove expired client state directory"
 429				" %.*s\n", HEXDIR_LEN, dname);
 430}
 431
 432static int
 433purge_old(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
 434{
 435	int status;
 436	struct xdr_netobj name;
 437
 438	if (child->d_name.len != HEXDIR_LEN - 1) {
 439		printk("%s: illegal name %pd in recovery directory\n",
 440				__func__, child);
 441		/* Keep trying; maybe the others are OK: */
 442		return 0;
 443	}
 444	name.data = kmemdup_nul(child->d_name.name, child->d_name.len, GFP_KERNEL);
 445	if (!name.data) {
 446		dprintk("%s: failed to allocate memory for name.data!\n",
 447			__func__);
 448		goto out;
 449	}
 450	name.len = HEXDIR_LEN;
 451	if (nfs4_has_reclaimed_state(name, nn))
 452		goto out_free;
 453
 454	status = vfs_rmdir(d_inode(parent), child);
 455	if (status)
 456		printk("failed to remove client recovery directory %pd\n",
 457				child);
 458out_free:
 459	kfree(name.data);
 460out:
 461	/* Keep trying, success or failure: */
 462	return 0;
 463}
 464
 465static void
 466nfsd4_recdir_purge_old(struct nfsd_net *nn)
 467{
 468	int status;
 469
 470	nn->in_grace = false;
 471	if (!nn->rec_file)
 472		return;
 473	status = mnt_want_write_file(nn->rec_file);
 474	if (status)
 475		goto out;
 476	status = nfsd4_list_rec_dir(purge_old, nn);
 477	if (status == 0)
 478		vfs_fsync(nn->rec_file, 0);
 479	mnt_drop_write_file(nn->rec_file);
 480out:
 481	nfs4_release_reclaim(nn);
 482	if (status)
 483		printk("nfsd4: failed to purge old clients from recovery"
 484			" directory %pD\n", nn->rec_file);
 485}
 486
 487static int
 488load_recdir(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
 489{
 490	struct xdr_netobj name;
 491	struct xdr_netobj princhash = { .len = 0, .data = NULL };
 492
 493	if (child->d_name.len != HEXDIR_LEN - 1) {
 494		printk("%s: illegal name %pd in recovery directory\n",
 495				__func__, child);
 496		/* Keep trying; maybe the others are OK: */
 497		return 0;
 498	}
 499	name.data = kmemdup_nul(child->d_name.name, child->d_name.len, GFP_KERNEL);
 500	if (!name.data) {
 501		dprintk("%s: failed to allocate memory for name.data!\n",
 502			__func__);
 503		goto out;
 504	}
 505	name.len = HEXDIR_LEN;
 506	if (!nfs4_client_to_reclaim(name, princhash, nn))
 507		kfree(name.data);
 508out:
 509	return 0;
 510}
 511
 512static int
 513nfsd4_recdir_load(struct net *net) {
 514	int status;
 515	struct nfsd_net *nn =  net_generic(net, nfsd_net_id);
 516
 517	if (!nn->rec_file)
 518		return 0;
 519
 520	status = nfsd4_list_rec_dir(load_recdir, nn);
 521	if (status)
 522		printk("nfsd4: failed loading clients from recovery"
 523			" directory %pD\n", nn->rec_file);
 524	return status;
 525}
 526
 527/*
 528 * Hold reference to the recovery directory.
 529 */
 530
 531static int
 532nfsd4_init_recdir(struct net *net)
 533{
 534	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 535	const struct cred *original_cred;
 536	int status;
 537
 538	printk("NFSD: Using %s as the NFSv4 state recovery directory\n",
 539			user_recovery_dirname);
 540
 541	BUG_ON(nn->rec_file);
 542
 543	status = nfs4_save_creds(&original_cred);
 544	if (status < 0) {
 545		printk("NFSD: Unable to change credentials to find recovery"
 546		       " directory: error %d\n",
 547		       status);
 548		return status;
 549	}
 550
 551	nn->rec_file = filp_open(user_recovery_dirname, O_RDONLY | O_DIRECTORY, 0);
 552	if (IS_ERR(nn->rec_file)) {
 553		printk("NFSD: unable to find recovery directory %s\n",
 554				user_recovery_dirname);
 555		status = PTR_ERR(nn->rec_file);
 556		nn->rec_file = NULL;
 557	}
 558
 559	nfs4_reset_creds(original_cred);
 560	if (!status)
 561		nn->in_grace = true;
 562	return status;
 563}
 564
 565static void
 566nfsd4_shutdown_recdir(struct net *net)
 567{
 568	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 569
 570	if (!nn->rec_file)
 571		return;
 572	fput(nn->rec_file);
 573	nn->rec_file = NULL;
 574}
 575
 576static int
 577nfs4_legacy_state_init(struct net *net)
 578{
 579	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 580	int i;
 581
 582	nn->reclaim_str_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
 583						sizeof(struct list_head),
 584						GFP_KERNEL);
 585	if (!nn->reclaim_str_hashtbl)
 586		return -ENOMEM;
 587
 588	for (i = 0; i < CLIENT_HASH_SIZE; i++)
 589		INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
 590	nn->reclaim_str_hashtbl_size = 0;
 591
 592	return 0;
 593}
 594
 595static void
 596nfs4_legacy_state_shutdown(struct net *net)
 597{
 598	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 599
 600	kfree(nn->reclaim_str_hashtbl);
 601}
 602
 603static int
 604nfsd4_load_reboot_recovery_data(struct net *net)
 605{
 606	int status;
 607
 608	status = nfsd4_init_recdir(net);
 609	if (status)
 610		return status;
 611
 612	status = nfsd4_recdir_load(net);
 613	if (status)
 614		nfsd4_shutdown_recdir(net);
 615
 616	return status;
 617}
 618
 619static int
 620nfsd4_legacy_tracking_init(struct net *net)
 621{
 622	int status;
 623
 624	/* XXX: The legacy code won't work in a container */
 625	if (net != &init_net) {
 626		pr_warn("NFSD: attempt to initialize legacy client tracking in a container ignored.\n");
 627		return -EINVAL;
 628	}
 629
 630	status = nfs4_legacy_state_init(net);
 631	if (status)
 632		return status;
 633
 634	status = nfsd4_load_reboot_recovery_data(net);
 635	if (status)
 636		goto err;
 637	printk("NFSD: Using legacy client tracking operations.\n");
 638	return 0;
 639
 640err:
 641	nfs4_legacy_state_shutdown(net);
 642	return status;
 643}
 644
 645static void
 646nfsd4_legacy_tracking_exit(struct net *net)
 647{
 648	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 649
 650	nfs4_release_reclaim(nn);
 651	nfsd4_shutdown_recdir(net);
 652	nfs4_legacy_state_shutdown(net);
 653}
 654
 655/*
 656 * Change the NFSv4 recovery directory to recdir.
 657 */
 658int
 659nfs4_reset_recoverydir(char *recdir)
 660{
 661	int status;
 662	struct path path;
 663
 664	status = kern_path(recdir, LOOKUP_FOLLOW, &path);
 665	if (status)
 666		return status;
 667	status = -ENOTDIR;
 668	if (d_is_dir(path.dentry)) {
 669		strcpy(user_recovery_dirname, recdir);
 670		status = 0;
 671	}
 672	path_put(&path);
 673	return status;
 674}
 675
 676char *
 677nfs4_recoverydir(void)
 678{
 679	return user_recovery_dirname;
 680}
 681
 682static int
 683nfsd4_check_legacy_client(struct nfs4_client *clp)
 684{
 685	int status;
 686	char dname[HEXDIR_LEN];
 687	struct nfs4_client_reclaim *crp;
 688	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 689	struct xdr_netobj name;
 690
 691	/* did we already find that this client is stable? */
 692	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 693		return 0;
 694
 695	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
 696	if (status) {
 697		legacy_recdir_name_error(clp, status);
 698		return status;
 699	}
 700
 701	/* look for it in the reclaim hashtable otherwise */
 702	name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
 703	if (!name.data) {
 704		dprintk("%s: failed to allocate memory for name.data!\n",
 705			__func__);
 706		goto out_enoent;
 707	}
 708	name.len = HEXDIR_LEN;
 709	crp = nfsd4_find_reclaim_client(name, nn);
 710	kfree(name.data);
 711	if (crp) {
 712		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
 713		crp->cr_clp = clp;
 714		return 0;
 715	}
 716
 717out_enoent:
 718	return -ENOENT;
 719}
 720
 721static const struct nfsd4_client_tracking_ops nfsd4_legacy_tracking_ops = {
 722	.init		= nfsd4_legacy_tracking_init,
 723	.exit		= nfsd4_legacy_tracking_exit,
 724	.create		= nfsd4_create_clid_dir,
 725	.remove		= nfsd4_remove_clid_dir,
 726	.check		= nfsd4_check_legacy_client,
 727	.grace_done	= nfsd4_recdir_purge_old,
 728	.version	= 1,
 729	.msglen		= 0,
 730};
 731
 732/* Globals */
 733#define NFSD_PIPE_DIR		"nfsd"
 734#define NFSD_CLD_PIPE		"cld"
 735
 736/* per-net-ns structure for holding cld upcall info */
 737struct cld_net {
 738	struct rpc_pipe		*cn_pipe;
 739	spinlock_t		 cn_lock;
 740	struct list_head	 cn_list;
 741	unsigned int		 cn_xid;
 742	bool			 cn_has_legacy;
 743	struct crypto_shash	*cn_tfm;
 744};
 745
 746struct cld_upcall {
 747	struct list_head	 cu_list;
 748	struct cld_net		*cu_net;
 749	struct completion	 cu_done;
 750	union {
 751		struct cld_msg_hdr	 cu_hdr;
 752		struct cld_msg		 cu_msg;
 753		struct cld_msg_v2	 cu_msg_v2;
 754	} cu_u;
 755};
 756
 757static int
 758__cld_pipe_upcall(struct rpc_pipe *pipe, void *cmsg)
 759{
 760	int ret;
 761	struct rpc_pipe_msg msg;
 762	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall, cu_u);
 763	struct nfsd_net *nn = net_generic(pipe->dentry->d_sb->s_fs_info,
 764					  nfsd_net_id);
 765
 766	memset(&msg, 0, sizeof(msg));
 767	msg.data = cmsg;
 768	msg.len = nn->client_tracking_ops->msglen;
 769
 
 
 
 
 
 770	ret = rpc_queue_upcall(pipe, &msg);
 771	if (ret < 0) {
 
 772		goto out;
 773	}
 774
 775	wait_for_completion(&cup->cu_done);
 776
 777	if (msg.errno < 0)
 778		ret = msg.errno;
 779out:
 780	return ret;
 781}
 782
 783static int
 784cld_pipe_upcall(struct rpc_pipe *pipe, void *cmsg)
 785{
 786	int ret;
 787
 788	/*
 789	 * -EAGAIN occurs when pipe is closed and reopened while there are
 790	 *  upcalls queued.
 791	 */
 792	do {
 793		ret = __cld_pipe_upcall(pipe, cmsg);
 794	} while (ret == -EAGAIN);
 795
 796	return ret;
 797}
 798
 799static ssize_t
 800__cld_pipe_inprogress_downcall(const struct cld_msg_v2 __user *cmsg,
 801		struct nfsd_net *nn)
 802{
 803	uint8_t cmd, princhashlen;
 804	struct xdr_netobj name, princhash = { .len = 0, .data = NULL };
 805	uint16_t namelen;
 806	struct cld_net *cn = nn->cld_net;
 807
 808	if (get_user(cmd, &cmsg->cm_cmd)) {
 809		dprintk("%s: error when copying cmd from userspace", __func__);
 810		return -EFAULT;
 811	}
 812	if (cmd == Cld_GraceStart) {
 813		if (nn->client_tracking_ops->version >= 2) {
 814			const struct cld_clntinfo __user *ci;
 815
 816			ci = &cmsg->cm_u.cm_clntinfo;
 817			if (get_user(namelen, &ci->cc_name.cn_len))
 818				return -EFAULT;
 819			name.data = memdup_user(&ci->cc_name.cn_id, namelen);
 820			if (IS_ERR_OR_NULL(name.data))
 821				return -EFAULT;
 822			name.len = namelen;
 823			get_user(princhashlen, &ci->cc_princhash.cp_len);
 824			if (princhashlen > 0) {
 825				princhash.data = memdup_user(
 826						&ci->cc_princhash.cp_data,
 827						princhashlen);
 828				if (IS_ERR_OR_NULL(princhash.data))
 829					return -EFAULT;
 830				princhash.len = princhashlen;
 831			} else
 832				princhash.len = 0;
 833		} else {
 834			const struct cld_name __user *cnm;
 835
 836			cnm = &cmsg->cm_u.cm_name;
 837			if (get_user(namelen, &cnm->cn_len))
 838				return -EFAULT;
 839			name.data = memdup_user(&cnm->cn_id, namelen);
 840			if (IS_ERR_OR_NULL(name.data))
 841				return -EFAULT;
 842			name.len = namelen;
 843		}
 844		if (name.len > 5 && memcmp(name.data, "hash:", 5) == 0) {
 845			name.len = name.len - 5;
 846			memmove(name.data, name.data + 5, name.len);
 847			cn->cn_has_legacy = true;
 848		}
 849		if (!nfs4_client_to_reclaim(name, princhash, nn)) {
 850			kfree(name.data);
 851			kfree(princhash.data);
 852			return -EFAULT;
 853		}
 854		return nn->client_tracking_ops->msglen;
 855	}
 856	return -EFAULT;
 857}
 858
 859static ssize_t
 860cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
 861{
 862	struct cld_upcall *tmp, *cup;
 863	struct cld_msg_hdr __user *hdr = (struct cld_msg_hdr __user *)src;
 864	struct cld_msg_v2 __user *cmsg = (struct cld_msg_v2 __user *)src;
 865	uint32_t xid;
 866	struct nfsd_net *nn = net_generic(file_inode(filp)->i_sb->s_fs_info,
 867						nfsd_net_id);
 868	struct cld_net *cn = nn->cld_net;
 869	int16_t status;
 870
 871	if (mlen != nn->client_tracking_ops->msglen) {
 872		dprintk("%s: got %zu bytes, expected %zu\n", __func__, mlen,
 873			nn->client_tracking_ops->msglen);
 874		return -EINVAL;
 875	}
 876
 877	/* copy just the xid so we can try to find that */
 878	if (copy_from_user(&xid, &hdr->cm_xid, sizeof(xid)) != 0) {
 879		dprintk("%s: error when copying xid from userspace", __func__);
 880		return -EFAULT;
 881	}
 882
 883	/*
 884	 * copy the status so we know whether to remove the upcall from the
 885	 * list (for -EINPROGRESS, we just want to make sure the xid is
 886	 * valid, not remove the upcall from the list)
 887	 */
 888	if (get_user(status, &hdr->cm_status)) {
 889		dprintk("%s: error when copying status from userspace", __func__);
 890		return -EFAULT;
 891	}
 892
 893	/* walk the list and find corresponding xid */
 894	cup = NULL;
 895	spin_lock(&cn->cn_lock);
 896	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
 897		if (get_unaligned(&tmp->cu_u.cu_hdr.cm_xid) == xid) {
 898			cup = tmp;
 899			if (status != -EINPROGRESS)
 900				list_del_init(&cup->cu_list);
 901			break;
 902		}
 903	}
 904	spin_unlock(&cn->cn_lock);
 905
 906	/* couldn't find upcall? */
 907	if (!cup) {
 908		dprintk("%s: couldn't find upcall -- xid=%u\n", __func__, xid);
 909		return -EINVAL;
 910	}
 911
 912	if (status == -EINPROGRESS)
 913		return __cld_pipe_inprogress_downcall(cmsg, nn);
 914
 915	if (copy_from_user(&cup->cu_u.cu_msg_v2, src, mlen) != 0)
 916		return -EFAULT;
 917
 918	complete(&cup->cu_done);
 919	return mlen;
 920}
 921
 922static void
 923cld_pipe_destroy_msg(struct rpc_pipe_msg *msg)
 924{
 925	struct cld_msg *cmsg = msg->data;
 926	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall,
 927						 cu_u.cu_msg);
 928
 929	/* errno >= 0 means we got a downcall */
 930	if (msg->errno >= 0)
 931		return;
 932
 933	complete(&cup->cu_done);
 934}
 935
 936static const struct rpc_pipe_ops cld_upcall_ops = {
 937	.upcall		= rpc_pipe_generic_upcall,
 938	.downcall	= cld_pipe_downcall,
 939	.destroy_msg	= cld_pipe_destroy_msg,
 940};
 941
 942static struct dentry *
 943nfsd4_cld_register_sb(struct super_block *sb, struct rpc_pipe *pipe)
 944{
 945	struct dentry *dir, *dentry;
 946
 947	dir = rpc_d_lookup_sb(sb, NFSD_PIPE_DIR);
 948	if (dir == NULL)
 949		return ERR_PTR(-ENOENT);
 950	dentry = rpc_mkpipe_dentry(dir, NFSD_CLD_PIPE, NULL, pipe);
 951	dput(dir);
 952	return dentry;
 953}
 954
 955static void
 956nfsd4_cld_unregister_sb(struct rpc_pipe *pipe)
 957{
 958	if (pipe->dentry)
 959		rpc_unlink(pipe->dentry);
 960}
 961
 962static struct dentry *
 963nfsd4_cld_register_net(struct net *net, struct rpc_pipe *pipe)
 964{
 965	struct super_block *sb;
 966	struct dentry *dentry;
 967
 968	sb = rpc_get_sb_net(net);
 969	if (!sb)
 970		return NULL;
 971	dentry = nfsd4_cld_register_sb(sb, pipe);
 972	rpc_put_sb_net(net);
 973	return dentry;
 974}
 975
 976static void
 977nfsd4_cld_unregister_net(struct net *net, struct rpc_pipe *pipe)
 978{
 979	struct super_block *sb;
 980
 981	sb = rpc_get_sb_net(net);
 982	if (sb) {
 983		nfsd4_cld_unregister_sb(pipe);
 984		rpc_put_sb_net(net);
 985	}
 986}
 987
 988/* Initialize rpc_pipefs pipe for communication with client tracking daemon */
 989static int
 990__nfsd4_init_cld_pipe(struct net *net)
 991{
 992	int ret;
 993	struct dentry *dentry;
 994	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 995	struct cld_net *cn;
 996
 997	if (nn->cld_net)
 998		return 0;
 999
1000	cn = kzalloc(sizeof(*cn), GFP_KERNEL);
1001	if (!cn) {
1002		ret = -ENOMEM;
1003		goto err;
1004	}
1005
1006	cn->cn_pipe = rpc_mkpipe_data(&cld_upcall_ops, RPC_PIPE_WAIT_FOR_OPEN);
1007	if (IS_ERR(cn->cn_pipe)) {
1008		ret = PTR_ERR(cn->cn_pipe);
1009		goto err;
1010	}
1011	spin_lock_init(&cn->cn_lock);
1012	INIT_LIST_HEAD(&cn->cn_list);
1013
1014	dentry = nfsd4_cld_register_net(net, cn->cn_pipe);
1015	if (IS_ERR(dentry)) {
1016		ret = PTR_ERR(dentry);
1017		goto err_destroy_data;
1018	}
1019
1020	cn->cn_pipe->dentry = dentry;
1021	cn->cn_has_legacy = false;
1022	nn->cld_net = cn;
1023	return 0;
1024
1025err_destroy_data:
1026	rpc_destroy_pipe_data(cn->cn_pipe);
1027err:
1028	kfree(cn);
1029	printk(KERN_ERR "NFSD: unable to create nfsdcld upcall pipe (%d)\n",
1030			ret);
1031	return ret;
1032}
1033
1034static int
1035nfsd4_init_cld_pipe(struct net *net)
1036{
1037	int status;
1038
1039	status = __nfsd4_init_cld_pipe(net);
1040	if (!status)
1041		printk("NFSD: Using old nfsdcld client tracking operations.\n");
1042	return status;
1043}
1044
1045static void
1046nfsd4_remove_cld_pipe(struct net *net)
1047{
1048	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1049	struct cld_net *cn = nn->cld_net;
1050
1051	nfsd4_cld_unregister_net(net, cn->cn_pipe);
1052	rpc_destroy_pipe_data(cn->cn_pipe);
1053	if (cn->cn_tfm)
1054		crypto_free_shash(cn->cn_tfm);
1055	kfree(nn->cld_net);
1056	nn->cld_net = NULL;
1057}
1058
1059static struct cld_upcall *
1060alloc_cld_upcall(struct nfsd_net *nn)
1061{
1062	struct cld_upcall *new, *tmp;
1063	struct cld_net *cn = nn->cld_net;
1064
1065	new = kzalloc(sizeof(*new), GFP_KERNEL);
1066	if (!new)
1067		return new;
1068
1069	/* FIXME: hard cap on number in flight? */
1070restart_search:
1071	spin_lock(&cn->cn_lock);
1072	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
1073		if (tmp->cu_u.cu_msg.cm_xid == cn->cn_xid) {
1074			cn->cn_xid++;
1075			spin_unlock(&cn->cn_lock);
1076			goto restart_search;
1077		}
1078	}
1079	init_completion(&new->cu_done);
1080	new->cu_u.cu_msg.cm_vers = nn->client_tracking_ops->version;
1081	put_unaligned(cn->cn_xid++, &new->cu_u.cu_msg.cm_xid);
1082	new->cu_net = cn;
1083	list_add(&new->cu_list, &cn->cn_list);
1084	spin_unlock(&cn->cn_lock);
1085
1086	dprintk("%s: allocated xid %u\n", __func__, new->cu_u.cu_msg.cm_xid);
1087
1088	return new;
1089}
1090
1091static void
1092free_cld_upcall(struct cld_upcall *victim)
1093{
1094	struct cld_net *cn = victim->cu_net;
1095
1096	spin_lock(&cn->cn_lock);
1097	list_del(&victim->cu_list);
1098	spin_unlock(&cn->cn_lock);
1099	kfree(victim);
1100}
1101
1102/* Ask daemon to create a new record */
1103static void
1104nfsd4_cld_create(struct nfs4_client *clp)
1105{
1106	int ret;
1107	struct cld_upcall *cup;
1108	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1109	struct cld_net *cn = nn->cld_net;
1110
1111	/* Don't upcall if it's already stored */
1112	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1113		return;
1114
1115	cup = alloc_cld_upcall(nn);
1116	if (!cup) {
1117		ret = -ENOMEM;
1118		goto out_err;
1119	}
1120
1121	cup->cu_u.cu_msg.cm_cmd = Cld_Create;
1122	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1123	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1124			clp->cl_name.len);
1125
1126	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1127	if (!ret) {
1128		ret = cup->cu_u.cu_msg.cm_status;
1129		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1130	}
1131
1132	free_cld_upcall(cup);
1133out_err:
1134	if (ret)
1135		printk(KERN_ERR "NFSD: Unable to create client "
1136				"record on stable storage: %d\n", ret);
1137}
1138
1139/* Ask daemon to create a new record */
1140static void
1141nfsd4_cld_create_v2(struct nfs4_client *clp)
1142{
1143	int ret;
1144	struct cld_upcall *cup;
1145	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1146	struct cld_net *cn = nn->cld_net;
1147	struct cld_msg_v2 *cmsg;
1148	struct crypto_shash *tfm = cn->cn_tfm;
1149	struct xdr_netobj cksum;
1150	char *principal = NULL;
1151	SHASH_DESC_ON_STACK(desc, tfm);
1152
1153	/* Don't upcall if it's already stored */
1154	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1155		return;
1156
1157	cup = alloc_cld_upcall(nn);
1158	if (!cup) {
1159		ret = -ENOMEM;
1160		goto out_err;
1161	}
1162
1163	cmsg = &cup->cu_u.cu_msg_v2;
1164	cmsg->cm_cmd = Cld_Create;
1165	cmsg->cm_u.cm_clntinfo.cc_name.cn_len = clp->cl_name.len;
1166	memcpy(cmsg->cm_u.cm_clntinfo.cc_name.cn_id, clp->cl_name.data,
1167			clp->cl_name.len);
1168	if (clp->cl_cred.cr_raw_principal)
1169		principal = clp->cl_cred.cr_raw_principal;
1170	else if (clp->cl_cred.cr_principal)
1171		principal = clp->cl_cred.cr_principal;
1172	if (principal) {
1173		desc->tfm = tfm;
1174		cksum.len = crypto_shash_digestsize(tfm);
1175		cksum.data = kmalloc(cksum.len, GFP_KERNEL);
1176		if (cksum.data == NULL) {
1177			ret = -ENOMEM;
1178			goto out;
1179		}
1180		ret = crypto_shash_digest(desc, principal, strlen(principal),
1181					  cksum.data);
1182		shash_desc_zero(desc);
1183		if (ret) {
1184			kfree(cksum.data);
1185			goto out;
1186		}
1187		cmsg->cm_u.cm_clntinfo.cc_princhash.cp_len = cksum.len;
1188		memcpy(cmsg->cm_u.cm_clntinfo.cc_princhash.cp_data,
1189		       cksum.data, cksum.len);
1190		kfree(cksum.data);
1191	} else
1192		cmsg->cm_u.cm_clntinfo.cc_princhash.cp_len = 0;
1193
1194	ret = cld_pipe_upcall(cn->cn_pipe, cmsg);
1195	if (!ret) {
1196		ret = cmsg->cm_status;
1197		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1198	}
1199
1200out:
1201	free_cld_upcall(cup);
1202out_err:
1203	if (ret)
1204		pr_err("NFSD: Unable to create client record on stable storage: %d\n",
1205				ret);
1206}
1207
1208/* Ask daemon to create a new record */
1209static void
1210nfsd4_cld_remove(struct nfs4_client *clp)
1211{
1212	int ret;
1213	struct cld_upcall *cup;
1214	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1215	struct cld_net *cn = nn->cld_net;
1216
1217	/* Don't upcall if it's already removed */
1218	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1219		return;
1220
1221	cup = alloc_cld_upcall(nn);
1222	if (!cup) {
1223		ret = -ENOMEM;
1224		goto out_err;
1225	}
1226
1227	cup->cu_u.cu_msg.cm_cmd = Cld_Remove;
1228	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1229	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1230			clp->cl_name.len);
1231
1232	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1233	if (!ret) {
1234		ret = cup->cu_u.cu_msg.cm_status;
1235		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1236	}
1237
1238	free_cld_upcall(cup);
1239out_err:
1240	if (ret)
1241		printk(KERN_ERR "NFSD: Unable to remove client "
1242				"record from stable storage: %d\n", ret);
1243}
1244
1245/*
1246 * For older nfsdcld's that do not allow us to "slurp" the clients
1247 * from the tracking database during startup.
1248 *
1249 * Check for presence of a record, and update its timestamp
1250 */
1251static int
1252nfsd4_cld_check_v0(struct nfs4_client *clp)
1253{
1254	int ret;
1255	struct cld_upcall *cup;
1256	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1257	struct cld_net *cn = nn->cld_net;
1258
1259	/* Don't upcall if one was already stored during this grace pd */
1260	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1261		return 0;
1262
1263	cup = alloc_cld_upcall(nn);
1264	if (!cup) {
1265		printk(KERN_ERR "NFSD: Unable to check client record on "
1266				"stable storage: %d\n", -ENOMEM);
1267		return -ENOMEM;
1268	}
1269
1270	cup->cu_u.cu_msg.cm_cmd = Cld_Check;
1271	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1272	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1273			clp->cl_name.len);
1274
1275	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1276	if (!ret) {
1277		ret = cup->cu_u.cu_msg.cm_status;
1278		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1279	}
1280
1281	free_cld_upcall(cup);
1282	return ret;
1283}
1284
1285/*
1286 * For newer nfsdcld's that allow us to "slurp" the clients
1287 * from the tracking database during startup.
1288 *
1289 * Check for presence of a record in the reclaim_str_hashtbl
1290 */
1291static int
1292nfsd4_cld_check(struct nfs4_client *clp)
1293{
1294	struct nfs4_client_reclaim *crp;
1295	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1296	struct cld_net *cn = nn->cld_net;
1297	int status;
1298	char dname[HEXDIR_LEN];
1299	struct xdr_netobj name;
1300
1301	/* did we already find that this client is stable? */
1302	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1303		return 0;
1304
1305	/* look for it in the reclaim hashtable otherwise */
1306	crp = nfsd4_find_reclaim_client(clp->cl_name, nn);
1307	if (crp)
1308		goto found;
1309
1310	if (cn->cn_has_legacy) {
1311		status = nfs4_make_rec_clidname(dname, &clp->cl_name);
1312		if (status)
1313			return -ENOENT;
1314
1315		name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
1316		if (!name.data) {
1317			dprintk("%s: failed to allocate memory for name.data!\n",
1318				__func__);
1319			return -ENOENT;
1320		}
1321		name.len = HEXDIR_LEN;
1322		crp = nfsd4_find_reclaim_client(name, nn);
1323		kfree(name.data);
1324		if (crp)
1325			goto found;
1326
1327	}
1328	return -ENOENT;
1329found:
1330	crp->cr_clp = clp;
1331	return 0;
1332}
1333
1334static int
1335nfsd4_cld_check_v2(struct nfs4_client *clp)
1336{
1337	struct nfs4_client_reclaim *crp;
1338	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1339	struct cld_net *cn = nn->cld_net;
1340	int status;
1341	char dname[HEXDIR_LEN];
1342	struct xdr_netobj name;
1343	struct crypto_shash *tfm = cn->cn_tfm;
1344	struct xdr_netobj cksum;
1345	char *principal = NULL;
1346	SHASH_DESC_ON_STACK(desc, tfm);
1347
1348	/* did we already find that this client is stable? */
1349	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1350		return 0;
1351
1352	/* look for it in the reclaim hashtable otherwise */
1353	crp = nfsd4_find_reclaim_client(clp->cl_name, nn);
1354	if (crp)
1355		goto found;
1356
1357	if (cn->cn_has_legacy) {
1358		status = nfs4_make_rec_clidname(dname, &clp->cl_name);
1359		if (status)
1360			return -ENOENT;
1361
1362		name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
1363		if (!name.data) {
1364			dprintk("%s: failed to allocate memory for name.data\n",
1365					__func__);
1366			return -ENOENT;
1367		}
1368		name.len = HEXDIR_LEN;
1369		crp = nfsd4_find_reclaim_client(name, nn);
1370		kfree(name.data);
1371		if (crp)
1372			goto found;
1373
1374	}
1375	return -ENOENT;
1376found:
1377	if (crp->cr_princhash.len) {
1378		if (clp->cl_cred.cr_raw_principal)
1379			principal = clp->cl_cred.cr_raw_principal;
1380		else if (clp->cl_cred.cr_principal)
1381			principal = clp->cl_cred.cr_principal;
1382		if (principal == NULL)
1383			return -ENOENT;
1384		desc->tfm = tfm;
1385		cksum.len = crypto_shash_digestsize(tfm);
1386		cksum.data = kmalloc(cksum.len, GFP_KERNEL);
1387		if (cksum.data == NULL)
1388			return -ENOENT;
1389		status = crypto_shash_digest(desc, principal, strlen(principal),
1390					     cksum.data);
1391		shash_desc_zero(desc);
1392		if (status) {
1393			kfree(cksum.data);
1394			return -ENOENT;
1395		}
1396		if (memcmp(crp->cr_princhash.data, cksum.data,
1397				crp->cr_princhash.len)) {
1398			kfree(cksum.data);
1399			return -ENOENT;
1400		}
1401		kfree(cksum.data);
1402	}
1403	crp->cr_clp = clp;
1404	return 0;
1405}
1406
1407static int
1408nfsd4_cld_grace_start(struct nfsd_net *nn)
1409{
1410	int ret;
1411	struct cld_upcall *cup;
1412	struct cld_net *cn = nn->cld_net;
1413
1414	cup = alloc_cld_upcall(nn);
1415	if (!cup) {
1416		ret = -ENOMEM;
1417		goto out_err;
1418	}
1419
1420	cup->cu_u.cu_msg.cm_cmd = Cld_GraceStart;
1421	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1422	if (!ret)
1423		ret = cup->cu_u.cu_msg.cm_status;
1424
1425	free_cld_upcall(cup);
1426out_err:
1427	if (ret)
1428		dprintk("%s: Unable to get clients from userspace: %d\n",
1429			__func__, ret);
1430	return ret;
1431}
1432
1433/* For older nfsdcld's that need cm_gracetime */
1434static void
1435nfsd4_cld_grace_done_v0(struct nfsd_net *nn)
1436{
1437	int ret;
1438	struct cld_upcall *cup;
1439	struct cld_net *cn = nn->cld_net;
1440
1441	cup = alloc_cld_upcall(nn);
1442	if (!cup) {
1443		ret = -ENOMEM;
1444		goto out_err;
1445	}
1446
1447	cup->cu_u.cu_msg.cm_cmd = Cld_GraceDone;
1448	cup->cu_u.cu_msg.cm_u.cm_gracetime = (int64_t)nn->boot_time;
1449	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1450	if (!ret)
1451		ret = cup->cu_u.cu_msg.cm_status;
1452
1453	free_cld_upcall(cup);
1454out_err:
1455	if (ret)
1456		printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1457}
1458
1459/*
1460 * For newer nfsdcld's that do not need cm_gracetime.  We also need to call
1461 * nfs4_release_reclaim() to clear out the reclaim_str_hashtbl.
1462 */
1463static void
1464nfsd4_cld_grace_done(struct nfsd_net *nn)
1465{
1466	int ret;
1467	struct cld_upcall *cup;
1468	struct cld_net *cn = nn->cld_net;
1469
1470	cup = alloc_cld_upcall(nn);
1471	if (!cup) {
1472		ret = -ENOMEM;
1473		goto out_err;
1474	}
1475
1476	cup->cu_u.cu_msg.cm_cmd = Cld_GraceDone;
1477	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
 
1478	if (!ret)
1479		ret = cup->cu_u.cu_msg.cm_status;
1480
1481	free_cld_upcall(cup);
1482out_err:
1483	nfs4_release_reclaim(nn);
1484	if (ret)
1485		printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1486}
1487
1488static int
1489nfs4_cld_state_init(struct net *net)
1490{
1491	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1492	int i;
1493
1494	nn->reclaim_str_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
1495						sizeof(struct list_head),
1496						GFP_KERNEL);
1497	if (!nn->reclaim_str_hashtbl)
1498		return -ENOMEM;
1499
1500	for (i = 0; i < CLIENT_HASH_SIZE; i++)
1501		INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
1502	nn->reclaim_str_hashtbl_size = 0;
1503	nn->track_reclaim_completes = true;
1504	atomic_set(&nn->nr_reclaim_complete, 0);
1505
1506	return 0;
1507}
1508
1509static void
1510nfs4_cld_state_shutdown(struct net *net)
1511{
1512	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1513
1514	nn->track_reclaim_completes = false;
1515	kfree(nn->reclaim_str_hashtbl);
1516}
1517
1518static bool
1519cld_running(struct nfsd_net *nn)
1520{
1521	struct cld_net *cn = nn->cld_net;
1522	struct rpc_pipe *pipe = cn->cn_pipe;
1523
1524	return pipe->nreaders || pipe->nwriters;
1525}
1526
1527static int
1528nfsd4_cld_get_version(struct nfsd_net *nn)
1529{
1530	int ret = 0;
1531	struct cld_upcall *cup;
1532	struct cld_net *cn = nn->cld_net;
1533	uint8_t version;
1534
1535	cup = alloc_cld_upcall(nn);
1536	if (!cup) {
1537		ret = -ENOMEM;
1538		goto out_err;
1539	}
1540	cup->cu_u.cu_msg.cm_cmd = Cld_GetVersion;
1541	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg);
1542	if (!ret) {
1543		ret = cup->cu_u.cu_msg.cm_status;
1544		if (ret)
1545			goto out_free;
1546		version = cup->cu_u.cu_msg.cm_u.cm_version;
1547		dprintk("%s: userspace returned version %u\n",
1548				__func__, version);
1549		if (version < 1)
1550			version = 1;
1551		else if (version > CLD_UPCALL_VERSION)
1552			version = CLD_UPCALL_VERSION;
1553
1554		switch (version) {
1555		case 1:
1556			nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
1557			break;
1558		case 2:
1559			nn->client_tracking_ops = &nfsd4_cld_tracking_ops_v2;
1560			break;
1561		default:
1562			break;
1563		}
1564	}
1565out_free:
1566	free_cld_upcall(cup);
1567out_err:
1568	if (ret)
1569		dprintk("%s: Unable to get version from userspace: %d\n",
1570			__func__, ret);
1571	return ret;
1572}
1573
1574static int
1575nfsd4_cld_tracking_init(struct net *net)
1576{
1577	int status;
1578	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1579	bool running;
1580	int retries = 10;
1581
1582	status = nfs4_cld_state_init(net);
1583	if (status)
1584		return status;
1585
1586	status = __nfsd4_init_cld_pipe(net);
1587	if (status)
1588		goto err_shutdown;
1589	nn->cld_net->cn_tfm = crypto_alloc_shash("sha256", 0, 0);
1590	if (IS_ERR(nn->cld_net->cn_tfm)) {
1591		status = PTR_ERR(nn->cld_net->cn_tfm);
1592		goto err_remove;
1593	}
1594
1595	/*
1596	 * rpc pipe upcalls take 30 seconds to time out, so we don't want to
1597	 * queue an upcall unless we know that nfsdcld is running (because we
1598	 * want this to fail fast so that nfsd4_client_tracking_init() can try
1599	 * the next client tracking method).  nfsdcld should already be running
1600	 * before nfsd is started, so the wait here is for nfsdcld to open the
1601	 * pipefs file we just created.
1602	 */
1603	while (!(running = cld_running(nn)) && retries--)
1604		msleep(100);
1605
1606	if (!running) {
1607		status = -ETIMEDOUT;
1608		goto err_remove;
1609	}
1610
1611	status = nfsd4_cld_get_version(nn);
1612	if (status == -EOPNOTSUPP)
1613		pr_warn("NFSD: nfsdcld GetVersion upcall failed. Please upgrade nfsdcld.\n");
1614
1615	status = nfsd4_cld_grace_start(nn);
1616	if (status) {
1617		if (status == -EOPNOTSUPP)
1618			pr_warn("NFSD: nfsdcld GraceStart upcall failed. Please upgrade nfsdcld.\n");
1619		nfs4_release_reclaim(nn);
1620		goto err_remove;
1621	} else
1622		printk("NFSD: Using nfsdcld client tracking operations.\n");
1623	return 0;
1624
1625err_remove:
1626	nfsd4_remove_cld_pipe(net);
1627err_shutdown:
1628	nfs4_cld_state_shutdown(net);
1629	return status;
1630}
1631
1632static void
1633nfsd4_cld_tracking_exit(struct net *net)
1634{
1635	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1636
1637	nfs4_release_reclaim(nn);
1638	nfsd4_remove_cld_pipe(net);
1639	nfs4_cld_state_shutdown(net);
1640}
1641
1642/* For older nfsdcld's */
1643static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v0 = {
1644	.init		= nfsd4_init_cld_pipe,
1645	.exit		= nfsd4_remove_cld_pipe,
1646	.create		= nfsd4_cld_create,
1647	.remove		= nfsd4_cld_remove,
1648	.check		= nfsd4_cld_check_v0,
1649	.grace_done	= nfsd4_cld_grace_done_v0,
1650	.version	= 1,
1651	.msglen		= sizeof(struct cld_msg),
1652};
1653
1654/* For newer nfsdcld's */
1655static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops = {
1656	.init		= nfsd4_cld_tracking_init,
1657	.exit		= nfsd4_cld_tracking_exit,
1658	.create		= nfsd4_cld_create,
1659	.remove		= nfsd4_cld_remove,
1660	.check		= nfsd4_cld_check,
1661	.grace_done	= nfsd4_cld_grace_done,
1662	.version	= 1,
1663	.msglen		= sizeof(struct cld_msg),
1664};
1665
1666/* v2 create/check ops include the principal, if available */
1667static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v2 = {
1668	.init		= nfsd4_cld_tracking_init,
1669	.exit		= nfsd4_cld_tracking_exit,
1670	.create		= nfsd4_cld_create_v2,
1671	.remove		= nfsd4_cld_remove,
1672	.check		= nfsd4_cld_check_v2,
1673	.grace_done	= nfsd4_cld_grace_done,
1674	.version	= 2,
1675	.msglen		= sizeof(struct cld_msg_v2),
1676};
1677
1678/* upcall via usermodehelper */
1679static char cltrack_prog[PATH_MAX] = "/sbin/nfsdcltrack";
1680module_param_string(cltrack_prog, cltrack_prog, sizeof(cltrack_prog),
1681			S_IRUGO|S_IWUSR);
1682MODULE_PARM_DESC(cltrack_prog, "Path to the nfsdcltrack upcall program");
1683
1684static bool cltrack_legacy_disable;
1685module_param(cltrack_legacy_disable, bool, S_IRUGO|S_IWUSR);
1686MODULE_PARM_DESC(cltrack_legacy_disable,
1687		"Disable legacy recoverydir conversion. Default: false");
1688
1689#define LEGACY_TOPDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_TOPDIR="
1690#define LEGACY_RECDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_RECDIR="
1691#define HAS_SESSION_ENV_PREFIX "NFSDCLTRACK_CLIENT_HAS_SESSION="
1692#define GRACE_START_ENV_PREFIX "NFSDCLTRACK_GRACE_START="
1693
1694static char *
1695nfsd4_cltrack_legacy_topdir(void)
1696{
1697	int copied;
1698	size_t len;
1699	char *result;
1700
1701	if (cltrack_legacy_disable)
1702		return NULL;
1703
1704	len = strlen(LEGACY_TOPDIR_ENV_PREFIX) +
1705		strlen(nfs4_recoverydir()) + 1;
1706
1707	result = kmalloc(len, GFP_KERNEL);
1708	if (!result)
1709		return result;
1710
1711	copied = snprintf(result, len, LEGACY_TOPDIR_ENV_PREFIX "%s",
1712				nfs4_recoverydir());
1713	if (copied >= len) {
1714		/* just return nothing if output was truncated */
1715		kfree(result);
1716		return NULL;
1717	}
1718
1719	return result;
1720}
1721
1722static char *
1723nfsd4_cltrack_legacy_recdir(const struct xdr_netobj *name)
1724{
1725	int copied;
1726	size_t len;
1727	char *result;
1728
1729	if (cltrack_legacy_disable)
1730		return NULL;
1731
1732	/* +1 is for '/' between "topdir" and "recdir" */
1733	len = strlen(LEGACY_RECDIR_ENV_PREFIX) +
1734		strlen(nfs4_recoverydir()) + 1 + HEXDIR_LEN;
1735
1736	result = kmalloc(len, GFP_KERNEL);
1737	if (!result)
1738		return result;
1739
1740	copied = snprintf(result, len, LEGACY_RECDIR_ENV_PREFIX "%s/",
1741				nfs4_recoverydir());
1742	if (copied > (len - HEXDIR_LEN)) {
1743		/* just return nothing if output will be truncated */
1744		kfree(result);
1745		return NULL;
1746	}
1747
1748	copied = nfs4_make_rec_clidname(result + copied, name);
1749	if (copied) {
1750		kfree(result);
1751		return NULL;
1752	}
1753
1754	return result;
1755}
1756
1757static char *
1758nfsd4_cltrack_client_has_session(struct nfs4_client *clp)
1759{
1760	int copied;
1761	size_t len;
1762	char *result;
1763
1764	/* prefix + Y/N character + terminating NULL */
1765	len = strlen(HAS_SESSION_ENV_PREFIX) + 1 + 1;
1766
1767	result = kmalloc(len, GFP_KERNEL);
1768	if (!result)
1769		return result;
1770
1771	copied = snprintf(result, len, HAS_SESSION_ENV_PREFIX "%c",
1772				clp->cl_minorversion ? 'Y' : 'N');
1773	if (copied >= len) {
1774		/* just return nothing if output was truncated */
1775		kfree(result);
1776		return NULL;
1777	}
1778
1779	return result;
1780}
1781
1782static char *
1783nfsd4_cltrack_grace_start(time_t grace_start)
1784{
1785	int copied;
1786	size_t len;
1787	char *result;
1788
1789	/* prefix + max width of int64_t string + terminating NULL */
1790	len = strlen(GRACE_START_ENV_PREFIX) + 22 + 1;
1791
1792	result = kmalloc(len, GFP_KERNEL);
1793	if (!result)
1794		return result;
1795
1796	copied = snprintf(result, len, GRACE_START_ENV_PREFIX "%ld",
1797				grace_start);
1798	if (copied >= len) {
1799		/* just return nothing if output was truncated */
1800		kfree(result);
1801		return NULL;
1802	}
1803
1804	return result;
1805}
1806
1807static int
1808nfsd4_umh_cltrack_upcall(char *cmd, char *arg, char *env0, char *env1)
1809{
1810	char *envp[3];
1811	char *argv[4];
1812	int ret;
1813
1814	if (unlikely(!cltrack_prog[0])) {
1815		dprintk("%s: cltrack_prog is disabled\n", __func__);
1816		return -EACCES;
1817	}
1818
1819	dprintk("%s: cmd: %s\n", __func__, cmd);
1820	dprintk("%s: arg: %s\n", __func__, arg ? arg : "(null)");
1821	dprintk("%s: env0: %s\n", __func__, env0 ? env0 : "(null)");
1822	dprintk("%s: env1: %s\n", __func__, env1 ? env1 : "(null)");
1823
1824	envp[0] = env0;
1825	envp[1] = env1;
1826	envp[2] = NULL;
1827
1828	argv[0] = (char *)cltrack_prog;
1829	argv[1] = cmd;
1830	argv[2] = arg;
1831	argv[3] = NULL;
1832
1833	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1834	/*
1835	 * Disable the upcall mechanism if we're getting an ENOENT or EACCES
1836	 * error. The admin can re-enable it on the fly by using sysfs
1837	 * once the problem has been fixed.
1838	 */
1839	if (ret == -ENOENT || ret == -EACCES) {
1840		dprintk("NFSD: %s was not found or isn't executable (%d). "
1841			"Setting cltrack_prog to blank string!",
1842			cltrack_prog, ret);
1843		cltrack_prog[0] = '\0';
1844	}
1845	dprintk("%s: %s return value: %d\n", __func__, cltrack_prog, ret);
1846
1847	return ret;
1848}
1849
1850static char *
1851bin_to_hex_dup(const unsigned char *src, int srclen)
1852{
1853	int i;
1854	char *buf, *hex;
1855
1856	/* +1 for terminating NULL */
1857	buf = kmalloc((srclen * 2) + 1, GFP_KERNEL);
1858	if (!buf)
1859		return buf;
1860
1861	hex = buf;
1862	for (i = 0; i < srclen; i++) {
1863		sprintf(hex, "%2.2x", *src++);
1864		hex += 2;
1865	}
1866	return buf;
1867}
1868
1869static int
1870nfsd4_umh_cltrack_init(struct net *net)
1871{
1872	int ret;
1873	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1874	char *grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1875
1876	/* XXX: The usermode helper s not working in container yet. */
1877	if (net != &init_net) {
1878		pr_warn("NFSD: attempt to initialize umh client tracking in a container ignored.\n");
1879		kfree(grace_start);
1880		return -EINVAL;
1881	}
1882
1883	ret = nfsd4_umh_cltrack_upcall("init", NULL, grace_start, NULL);
1884	kfree(grace_start);
1885	if (!ret)
1886		printk("NFSD: Using UMH upcall client tracking operations.\n");
1887	return ret;
1888}
1889
1890static void
1891nfsd4_cltrack_upcall_lock(struct nfs4_client *clp)
1892{
1893	wait_on_bit_lock(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK,
1894			 TASK_UNINTERRUPTIBLE);
1895}
1896
1897static void
1898nfsd4_cltrack_upcall_unlock(struct nfs4_client *clp)
1899{
1900	smp_mb__before_atomic();
1901	clear_bit(NFSD4_CLIENT_UPCALL_LOCK, &clp->cl_flags);
1902	smp_mb__after_atomic();
1903	wake_up_bit(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK);
1904}
1905
1906static void
1907nfsd4_umh_cltrack_create(struct nfs4_client *clp)
1908{
1909	char *hexid, *has_session, *grace_start;
1910	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1911
1912	/*
1913	 * With v4.0 clients, there's little difference in outcome between a
1914	 * create and check operation, and we can end up calling into this
1915	 * function multiple times per client (once for each openowner). So,
1916	 * for v4.0 clients skip upcalling once the client has been recorded
1917	 * on stable storage.
1918	 *
1919	 * For v4.1+ clients, the outcome of the two operations is different,
1920	 * so we must ensure that we upcall for the create operation. v4.1+
1921	 * clients call this on RECLAIM_COMPLETE though, so we should only end
1922	 * up doing a single create upcall per client.
1923	 */
1924	if (clp->cl_minorversion == 0 &&
1925	    test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1926		return;
1927
1928	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1929	if (!hexid) {
1930		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1931		return;
1932	}
1933
1934	has_session = nfsd4_cltrack_client_has_session(clp);
1935	grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1936
1937	nfsd4_cltrack_upcall_lock(clp);
1938	if (!nfsd4_umh_cltrack_upcall("create", hexid, has_session, grace_start))
1939		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1940	nfsd4_cltrack_upcall_unlock(clp);
1941
1942	kfree(has_session);
1943	kfree(grace_start);
1944	kfree(hexid);
1945}
1946
1947static void
1948nfsd4_umh_cltrack_remove(struct nfs4_client *clp)
1949{
1950	char *hexid;
1951
1952	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1953		return;
1954
1955	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1956	if (!hexid) {
1957		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1958		return;
1959	}
1960
1961	nfsd4_cltrack_upcall_lock(clp);
1962	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags) &&
1963	    nfsd4_umh_cltrack_upcall("remove", hexid, NULL, NULL) == 0)
1964		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1965	nfsd4_cltrack_upcall_unlock(clp);
1966
1967	kfree(hexid);
1968}
1969
1970static int
1971nfsd4_umh_cltrack_check(struct nfs4_client *clp)
1972{
1973	int ret;
1974	char *hexid, *has_session, *legacy;
1975
1976	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1977		return 0;
1978
1979	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1980	if (!hexid) {
1981		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1982		return -ENOMEM;
1983	}
1984
1985	has_session = nfsd4_cltrack_client_has_session(clp);
1986	legacy = nfsd4_cltrack_legacy_recdir(&clp->cl_name);
1987
1988	nfsd4_cltrack_upcall_lock(clp);
1989	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags)) {
1990		ret = 0;
1991	} else {
1992		ret = nfsd4_umh_cltrack_upcall("check", hexid, has_session, legacy);
1993		if (ret == 0)
1994			set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1995	}
1996	nfsd4_cltrack_upcall_unlock(clp);
1997	kfree(has_session);
1998	kfree(legacy);
1999	kfree(hexid);
2000
2001	return ret;
2002}
2003
2004static void
2005nfsd4_umh_cltrack_grace_done(struct nfsd_net *nn)
2006{
2007	char *legacy;
2008	char timestr[22]; /* FIXME: better way to determine max size? */
2009
2010	sprintf(timestr, "%ld", nn->boot_time);
2011	legacy = nfsd4_cltrack_legacy_topdir();
2012	nfsd4_umh_cltrack_upcall("gracedone", timestr, legacy, NULL);
2013	kfree(legacy);
2014}
2015
2016static const struct nfsd4_client_tracking_ops nfsd4_umh_tracking_ops = {
2017	.init		= nfsd4_umh_cltrack_init,
2018	.exit		= NULL,
2019	.create		= nfsd4_umh_cltrack_create,
2020	.remove		= nfsd4_umh_cltrack_remove,
2021	.check		= nfsd4_umh_cltrack_check,
2022	.grace_done	= nfsd4_umh_cltrack_grace_done,
2023	.version	= 1,
2024	.msglen		= 0,
2025};
2026
2027int
2028nfsd4_client_tracking_init(struct net *net)
2029{
2030	int status;
2031	struct path path;
2032	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2033
2034	/* just run the init if it the method is already decided */
2035	if (nn->client_tracking_ops)
2036		goto do_init;
2037
2038	/* First, try to use nfsdcld */
2039	nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
2040	status = nn->client_tracking_ops->init(net);
2041	if (!status)
2042		return status;
2043	if (status != -ETIMEDOUT) {
2044		nn->client_tracking_ops = &nfsd4_cld_tracking_ops_v0;
2045		status = nn->client_tracking_ops->init(net);
2046		if (!status)
2047			return status;
2048	}
2049
2050	/*
2051	 * Next, try the UMH upcall.
 
2052	 */
2053	nn->client_tracking_ops = &nfsd4_umh_tracking_ops;
2054	status = nn->client_tracking_ops->init(net);
2055	if (!status)
2056		return status;
2057
2058	/*
2059	 * Finally, See if the recoverydir exists and is a directory.
2060	 * If it is, then use the legacy ops.
2061	 */
2062	nn->client_tracking_ops = &nfsd4_legacy_tracking_ops;
2063	status = kern_path(nfs4_recoverydir(), LOOKUP_FOLLOW, &path);
2064	if (!status) {
2065		status = d_is_dir(path.dentry);
2066		path_put(&path);
2067		if (!status) {
2068			status = -EINVAL;
2069			goto out;
2070		}
2071	}
2072
 
 
 
 
 
2073do_init:
2074	status = nn->client_tracking_ops->init(net);
2075out:
2076	if (status) {
2077		printk(KERN_WARNING "NFSD: Unable to initialize client "
2078				    "recovery tracking! (%d)\n", status);
2079		nn->client_tracking_ops = NULL;
2080	}
2081	return status;
2082}
2083
2084void
2085nfsd4_client_tracking_exit(struct net *net)
2086{
2087	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2088
2089	if (nn->client_tracking_ops) {
2090		if (nn->client_tracking_ops->exit)
2091			nn->client_tracking_ops->exit(net);
2092		nn->client_tracking_ops = NULL;
2093	}
2094}
2095
2096void
2097nfsd4_client_record_create(struct nfs4_client *clp)
2098{
2099	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2100
2101	if (nn->client_tracking_ops)
2102		nn->client_tracking_ops->create(clp);
2103}
2104
2105void
2106nfsd4_client_record_remove(struct nfs4_client *clp)
2107{
2108	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2109
2110	if (nn->client_tracking_ops)
2111		nn->client_tracking_ops->remove(clp);
2112}
2113
2114int
2115nfsd4_client_record_check(struct nfs4_client *clp)
2116{
2117	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2118
2119	if (nn->client_tracking_ops)
2120		return nn->client_tracking_ops->check(clp);
2121
2122	return -EOPNOTSUPP;
2123}
2124
2125void
2126nfsd4_record_grace_done(struct nfsd_net *nn)
2127{
2128	if (nn->client_tracking_ops)
2129		nn->client_tracking_ops->grace_done(nn);
2130}
2131
2132static int
2133rpc_pipefs_event(struct notifier_block *nb, unsigned long event, void *ptr)
2134{
2135	struct super_block *sb = ptr;
2136	struct net *net = sb->s_fs_info;
2137	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2138	struct cld_net *cn = nn->cld_net;
2139	struct dentry *dentry;
2140	int ret = 0;
2141
2142	if (!try_module_get(THIS_MODULE))
2143		return 0;
2144
2145	if (!cn) {
2146		module_put(THIS_MODULE);
2147		return 0;
2148	}
2149
2150	switch (event) {
2151	case RPC_PIPEFS_MOUNT:
2152		dentry = nfsd4_cld_register_sb(sb, cn->cn_pipe);
2153		if (IS_ERR(dentry)) {
2154			ret = PTR_ERR(dentry);
2155			break;
2156		}
2157		cn->cn_pipe->dentry = dentry;
2158		break;
2159	case RPC_PIPEFS_UMOUNT:
2160		if (cn->cn_pipe->dentry)
2161			nfsd4_cld_unregister_sb(cn->cn_pipe);
2162		break;
2163	default:
2164		ret = -ENOTSUPP;
2165		break;
2166	}
2167	module_put(THIS_MODULE);
2168	return ret;
2169}
2170
2171static struct notifier_block nfsd4_cld_block = {
2172	.notifier_call = rpc_pipefs_event,
2173};
2174
2175int
2176register_cld_notifier(void)
2177{
2178	return rpc_pipefs_notifier_register(&nfsd4_cld_block);
2179}
2180
2181void
2182unregister_cld_notifier(void)
2183{
2184	rpc_pipefs_notifier_unregister(&nfsd4_cld_block);
2185}