Linux Audio

Check our new training course

Linux BSP upgrade and security maintenance

Need help to get security updates for your Linux BSP?
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}
v5.9
   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	status = crypto_shash_tfm_digest(tfm, clname->data, clname->len,
 131					 cksum.data);
 
 
 
 
 
 
 
 
 
 132	if (status)
 133		goto out;
 134
 135	md5_to_hex(dname, cksum.data);
 136
 137	status = 0;
 138out:
 139	kfree(cksum.data);
 140	crypto_free_shash(tfm);
 141out_no_tfm:
 142	return status;
 143}
 144
 145/*
 146 * If we had an error generating the recdir name for the legacy tracker
 147 * then warn the admin. If the error doesn't appear to be transient,
 148 * then disable recovery tracking.
 149 */
 150static void
 151legacy_recdir_name_error(struct nfs4_client *clp, int error)
 152{
 153	printk(KERN_ERR "NFSD: unable to generate recoverydir "
 154			"name (%d).\n", error);
 155
 156	/*
 157	 * if the algorithm just doesn't exist, then disable the recovery
 158	 * tracker altogether. The crypto libs will generally return this if
 159	 * FIPS is enabled as well.
 160	 */
 161	if (error == -ENOENT) {
 162		printk(KERN_ERR "NFSD: disabling legacy clientid tracking. "
 163			"Reboot recovery will not function correctly!\n");
 164		nfsd4_client_tracking_exit(clp->net);
 165	}
 166}
 167
 168static void
 169__nfsd4_create_reclaim_record_grace(struct nfs4_client *clp,
 170		const char *dname, int len, struct nfsd_net *nn)
 171{
 172	struct xdr_netobj name;
 173	struct xdr_netobj princhash = { .len = 0, .data = NULL };
 174	struct nfs4_client_reclaim *crp;
 175
 176	name.data = kmemdup(dname, len, GFP_KERNEL);
 177	if (!name.data) {
 178		dprintk("%s: failed to allocate memory for name.data!\n",
 179			__func__);
 180		return;
 181	}
 182	name.len = len;
 183	crp = nfs4_client_to_reclaim(name, princhash, nn);
 184	if (!crp) {
 185		kfree(name.data);
 186		return;
 187	}
 188	crp->cr_clp = clp;
 189}
 190
 191static void
 192nfsd4_create_clid_dir(struct nfs4_client *clp)
 193{
 194	const struct cred *original_cred;
 195	char dname[HEXDIR_LEN];
 196	struct dentry *dir, *dentry;
 
 197	int status;
 198	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 199
 200	if (test_and_set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 201		return;
 202	if (!nn->rec_file)
 203		return;
 204
 205	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
 206	if (status)
 207		return legacy_recdir_name_error(clp, status);
 208
 209	status = nfs4_save_creds(&original_cred);
 210	if (status < 0)
 211		return;
 212
 213	status = mnt_want_write_file(nn->rec_file);
 214	if (status)
 215		goto out_creds;
 216
 217	dir = nn->rec_file->f_path.dentry;
 218	/* lock the parent */
 219	inode_lock(d_inode(dir));
 220
 221	dentry = lookup_one_len(dname, dir, HEXDIR_LEN-1);
 222	if (IS_ERR(dentry)) {
 223		status = PTR_ERR(dentry);
 224		goto out_unlock;
 225	}
 226	if (d_really_is_positive(dentry))
 227		/*
 228		 * In the 4.1 case, where we're called from
 229		 * reclaim_complete(), records from the previous reboot
 230		 * may still be left, so this is OK.
 231		 *
 232		 * In the 4.0 case, we should never get here; but we may
 233		 * as well be forgiving and just succeed silently.
 234		 */
 235		goto out_put;
 236	status = vfs_mkdir(d_inode(dir), dentry, S_IRWXU);
 237out_put:
 238	dput(dentry);
 239out_unlock:
 240	inode_unlock(d_inode(dir));
 241	if (status == 0) {
 242		if (nn->in_grace)
 243			__nfsd4_create_reclaim_record_grace(clp, dname,
 244					HEXDIR_LEN, nn);
 
 
 245		vfs_fsync(nn->rec_file, 0);
 246	} else {
 247		printk(KERN_ERR "NFSD: failed to write recovery record"
 248				" (err %d); please check that %s exists"
 249				" and is writeable", status,
 250				user_recovery_dirname);
 251	}
 252	mnt_drop_write_file(nn->rec_file);
 253out_creds:
 254	nfs4_reset_creds(original_cred);
 255}
 256
 257typedef int (recdir_func)(struct dentry *, struct dentry *, struct nfsd_net *);
 258
 259struct name_list {
 260	char name[HEXDIR_LEN];
 261	struct list_head list;
 262};
 263
 264struct nfs4_dir_ctx {
 265	struct dir_context ctx;
 266	struct list_head names;
 267};
 268
 269static int
 270nfsd4_build_namelist(struct dir_context *__ctx, const char *name, int namlen,
 271		loff_t offset, u64 ino, unsigned int d_type)
 272{
 273	struct nfs4_dir_ctx *ctx =
 274		container_of(__ctx, struct nfs4_dir_ctx, ctx);
 275	struct name_list *entry;
 276
 277	if (namlen != HEXDIR_LEN - 1)
 278		return 0;
 279	entry = kmalloc(sizeof(struct name_list), GFP_KERNEL);
 280	if (entry == NULL)
 281		return -ENOMEM;
 282	memcpy(entry->name, name, HEXDIR_LEN - 1);
 283	entry->name[HEXDIR_LEN - 1] = '\0';
 284	list_add(&entry->list, &ctx->names);
 285	return 0;
 286}
 287
 288static int
 289nfsd4_list_rec_dir(recdir_func *f, struct nfsd_net *nn)
 290{
 291	const struct cred *original_cred;
 292	struct dentry *dir = nn->rec_file->f_path.dentry;
 293	struct nfs4_dir_ctx ctx = {
 294		.ctx.actor = nfsd4_build_namelist,
 295		.names = LIST_HEAD_INIT(ctx.names)
 296	};
 297	struct name_list *entry, *tmp;
 298	int status;
 299
 300	status = nfs4_save_creds(&original_cred);
 301	if (status < 0)
 302		return status;
 303
 304	status = vfs_llseek(nn->rec_file, 0, SEEK_SET);
 305	if (status < 0) {
 306		nfs4_reset_creds(original_cred);
 307		return status;
 308	}
 309
 310	status = iterate_dir(nn->rec_file, &ctx.ctx);
 311	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 312
 313	list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
 314		if (!status) {
 315			struct dentry *dentry;
 316			dentry = lookup_one_len(entry->name, dir, HEXDIR_LEN-1);
 317			if (IS_ERR(dentry)) {
 318				status = PTR_ERR(dentry);
 319				break;
 320			}
 321			status = f(dir, dentry, nn);
 322			dput(dentry);
 323		}
 324		list_del(&entry->list);
 325		kfree(entry);
 326	}
 327	inode_unlock(d_inode(dir));
 328	nfs4_reset_creds(original_cred);
 329
 330	list_for_each_entry_safe(entry, tmp, &ctx.names, list) {
 331		dprintk("NFSD: %s. Left entry %s\n", __func__, entry->name);
 332		list_del(&entry->list);
 333		kfree(entry);
 334	}
 335	return status;
 336}
 337
 338static int
 339nfsd4_unlink_clid_dir(char *name, int namlen, struct nfsd_net *nn)
 340{
 341	struct dentry *dir, *dentry;
 342	int status;
 343
 344	dprintk("NFSD: nfsd4_unlink_clid_dir. name %.*s\n", namlen, name);
 345
 346	dir = nn->rec_file->f_path.dentry;
 347	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 348	dentry = lookup_one_len(name, dir, namlen);
 349	if (IS_ERR(dentry)) {
 350		status = PTR_ERR(dentry);
 351		goto out_unlock;
 352	}
 353	status = -ENOENT;
 354	if (d_really_is_negative(dentry))
 355		goto out;
 356	status = vfs_rmdir(d_inode(dir), dentry);
 357out:
 358	dput(dentry);
 359out_unlock:
 360	inode_unlock(d_inode(dir));
 361	return status;
 362}
 363
 364static void
 365__nfsd4_remove_reclaim_record_grace(const char *dname, int len,
 366		struct nfsd_net *nn)
 367{
 368	struct xdr_netobj name;
 369	struct nfs4_client_reclaim *crp;
 370
 371	name.data = kmemdup(dname, len, GFP_KERNEL);
 372	if (!name.data) {
 373		dprintk("%s: failed to allocate memory for name.data!\n",
 374			__func__);
 375		return;
 376	}
 377	name.len = len;
 378	crp = nfsd4_find_reclaim_client(name, nn);
 379	kfree(name.data);
 380	if (crp)
 381		nfs4_remove_reclaim_record(crp, nn);
 382}
 383
 384static void
 385nfsd4_remove_clid_dir(struct nfs4_client *clp)
 386{
 387	const struct cred *original_cred;
 
 388	char dname[HEXDIR_LEN];
 389	int status;
 390	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 391
 392	if (!nn->rec_file || !test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 393		return;
 394
 395	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
 396	if (status)
 397		return legacy_recdir_name_error(clp, status);
 398
 399	status = mnt_want_write_file(nn->rec_file);
 400	if (status)
 401		goto out;
 402	clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
 403
 404	status = nfs4_save_creds(&original_cred);
 405	if (status < 0)
 406		goto out_drop_write;
 407
 408	status = nfsd4_unlink_clid_dir(dname, HEXDIR_LEN-1, nn);
 409	nfs4_reset_creds(original_cred);
 410	if (status == 0) {
 411		vfs_fsync(nn->rec_file, 0);
 412		if (nn->in_grace)
 413			__nfsd4_remove_reclaim_record_grace(dname,
 414					HEXDIR_LEN, nn);
 
 
 
 415	}
 416out_drop_write:
 417	mnt_drop_write_file(nn->rec_file);
 418out:
 419	if (status)
 420		printk("NFSD: Failed to remove expired client state directory"
 421				" %.*s\n", HEXDIR_LEN, dname);
 422}
 423
 424static int
 425purge_old(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
 426{
 427	int status;
 428	struct xdr_netobj name;
 429
 430	if (child->d_name.len != HEXDIR_LEN - 1) {
 431		printk("%s: illegal name %pd in recovery directory\n",
 432				__func__, child);
 433		/* Keep trying; maybe the others are OK: */
 434		return 0;
 435	}
 436	name.data = kmemdup_nul(child->d_name.name, child->d_name.len, GFP_KERNEL);
 437	if (!name.data) {
 438		dprintk("%s: failed to allocate memory for name.data!\n",
 439			__func__);
 440		goto out;
 441	}
 442	name.len = HEXDIR_LEN;
 443	if (nfs4_has_reclaimed_state(name, nn))
 444		goto out_free;
 445
 446	status = vfs_rmdir(d_inode(parent), child);
 447	if (status)
 448		printk("failed to remove client recovery directory %pd\n",
 449				child);
 450out_free:
 451	kfree(name.data);
 452out:
 453	/* Keep trying, success or failure: */
 454	return 0;
 455}
 456
 457static void
 458nfsd4_recdir_purge_old(struct nfsd_net *nn)
 459{
 460	int status;
 461
 462	nn->in_grace = false;
 463	if (!nn->rec_file)
 464		return;
 465	status = mnt_want_write_file(nn->rec_file);
 466	if (status)
 467		goto out;
 468	status = nfsd4_list_rec_dir(purge_old, nn);
 469	if (status == 0)
 470		vfs_fsync(nn->rec_file, 0);
 471	mnt_drop_write_file(nn->rec_file);
 472out:
 473	nfs4_release_reclaim(nn);
 474	if (status)
 475		printk("nfsd4: failed to purge old clients from recovery"
 476			" directory %pD\n", nn->rec_file);
 477}
 478
 479static int
 480load_recdir(struct dentry *parent, struct dentry *child, struct nfsd_net *nn)
 481{
 482	struct xdr_netobj name;
 483	struct xdr_netobj princhash = { .len = 0, .data = NULL };
 484
 485	if (child->d_name.len != HEXDIR_LEN - 1) {
 486		printk("%s: illegal name %pd in recovery directory\n",
 487				__func__, child);
 488		/* Keep trying; maybe the others are OK: */
 489		return 0;
 490	}
 491	name.data = kmemdup_nul(child->d_name.name, child->d_name.len, GFP_KERNEL);
 492	if (!name.data) {
 493		dprintk("%s: failed to allocate memory for name.data!\n",
 494			__func__);
 495		goto out;
 496	}
 497	name.len = HEXDIR_LEN;
 498	if (!nfs4_client_to_reclaim(name, princhash, nn))
 499		kfree(name.data);
 500out:
 501	return 0;
 502}
 503
 504static int
 505nfsd4_recdir_load(struct net *net) {
 506	int status;
 507	struct nfsd_net *nn =  net_generic(net, nfsd_net_id);
 508
 509	if (!nn->rec_file)
 510		return 0;
 511
 512	status = nfsd4_list_rec_dir(load_recdir, nn);
 513	if (status)
 514		printk("nfsd4: failed loading clients from recovery"
 515			" directory %pD\n", nn->rec_file);
 516	return status;
 517}
 518
 519/*
 520 * Hold reference to the recovery directory.
 521 */
 522
 523static int
 524nfsd4_init_recdir(struct net *net)
 525{
 526	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 527	const struct cred *original_cred;
 528	int status;
 529
 530	printk("NFSD: Using %s as the NFSv4 state recovery directory\n",
 531			user_recovery_dirname);
 532
 533	BUG_ON(nn->rec_file);
 534
 535	status = nfs4_save_creds(&original_cred);
 536	if (status < 0) {
 537		printk("NFSD: Unable to change credentials to find recovery"
 538		       " directory: error %d\n",
 539		       status);
 540		return status;
 541	}
 542
 543	nn->rec_file = filp_open(user_recovery_dirname, O_RDONLY | O_DIRECTORY, 0);
 544	if (IS_ERR(nn->rec_file)) {
 545		printk("NFSD: unable to find recovery directory %s\n",
 546				user_recovery_dirname);
 547		status = PTR_ERR(nn->rec_file);
 548		nn->rec_file = NULL;
 549	}
 550
 551	nfs4_reset_creds(original_cred);
 552	if (!status)
 553		nn->in_grace = true;
 554	return status;
 555}
 556
 557static void
 558nfsd4_shutdown_recdir(struct net *net)
 559{
 560	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 561
 562	if (!nn->rec_file)
 563		return;
 564	fput(nn->rec_file);
 565	nn->rec_file = NULL;
 566}
 567
 568static int
 569nfs4_legacy_state_init(struct net *net)
 570{
 571	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 572	int i;
 573
 574	nn->reclaim_str_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
 575						sizeof(struct list_head),
 576						GFP_KERNEL);
 577	if (!nn->reclaim_str_hashtbl)
 578		return -ENOMEM;
 579
 580	for (i = 0; i < CLIENT_HASH_SIZE; i++)
 581		INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
 582	nn->reclaim_str_hashtbl_size = 0;
 583
 584	return 0;
 585}
 586
 587static void
 588nfs4_legacy_state_shutdown(struct net *net)
 589{
 590	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 591
 592	kfree(nn->reclaim_str_hashtbl);
 593}
 594
 595static int
 596nfsd4_load_reboot_recovery_data(struct net *net)
 597{
 598	int status;
 599
 600	status = nfsd4_init_recdir(net);
 601	if (status)
 602		return status;
 603
 604	status = nfsd4_recdir_load(net);
 605	if (status)
 606		nfsd4_shutdown_recdir(net);
 607
 608	return status;
 609}
 610
 611static int
 612nfsd4_legacy_tracking_init(struct net *net)
 613{
 614	int status;
 615
 616	/* XXX: The legacy code won't work in a container */
 617	if (net != &init_net) {
 618		pr_warn("NFSD: attempt to initialize legacy client tracking in a container ignored.\n");
 619		return -EINVAL;
 620	}
 621
 622	status = nfs4_legacy_state_init(net);
 623	if (status)
 624		return status;
 625
 626	status = nfsd4_load_reboot_recovery_data(net);
 627	if (status)
 628		goto err;
 629	printk("NFSD: Using legacy client tracking operations.\n");
 630	return 0;
 631
 632err:
 633	nfs4_legacy_state_shutdown(net);
 634	return status;
 635}
 636
 637static void
 638nfsd4_legacy_tracking_exit(struct net *net)
 639{
 640	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 641
 642	nfs4_release_reclaim(nn);
 643	nfsd4_shutdown_recdir(net);
 644	nfs4_legacy_state_shutdown(net);
 645}
 646
 647/*
 648 * Change the NFSv4 recovery directory to recdir.
 649 */
 650int
 651nfs4_reset_recoverydir(char *recdir)
 652{
 653	int status;
 654	struct path path;
 655
 656	status = kern_path(recdir, LOOKUP_FOLLOW, &path);
 657	if (status)
 658		return status;
 659	status = -ENOTDIR;
 660	if (d_is_dir(path.dentry)) {
 661		strcpy(user_recovery_dirname, recdir);
 662		status = 0;
 663	}
 664	path_put(&path);
 665	return status;
 666}
 667
 668char *
 669nfs4_recoverydir(void)
 670{
 671	return user_recovery_dirname;
 672}
 673
 674static int
 675nfsd4_check_legacy_client(struct nfs4_client *clp)
 676{
 677	int status;
 678	char dname[HEXDIR_LEN];
 679	struct nfs4_client_reclaim *crp;
 680	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 681	struct xdr_netobj name;
 682
 683	/* did we already find that this client is stable? */
 684	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
 685		return 0;
 686
 687	status = nfs4_make_rec_clidname(dname, &clp->cl_name);
 688	if (status) {
 689		legacy_recdir_name_error(clp, status);
 690		return status;
 691	}
 692
 693	/* look for it in the reclaim hashtable otherwise */
 694	name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
 695	if (!name.data) {
 696		dprintk("%s: failed to allocate memory for name.data!\n",
 697			__func__);
 698		goto out_enoent;
 699	}
 700	name.len = HEXDIR_LEN;
 701	crp = nfsd4_find_reclaim_client(name, nn);
 702	kfree(name.data);
 703	if (crp) {
 704		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
 705		crp->cr_clp = clp;
 706		return 0;
 707	}
 708
 709out_enoent:
 710	return -ENOENT;
 711}
 712
 713static const struct nfsd4_client_tracking_ops nfsd4_legacy_tracking_ops = {
 714	.init		= nfsd4_legacy_tracking_init,
 715	.exit		= nfsd4_legacy_tracking_exit,
 716	.create		= nfsd4_create_clid_dir,
 717	.remove		= nfsd4_remove_clid_dir,
 718	.check		= nfsd4_check_legacy_client,
 719	.grace_done	= nfsd4_recdir_purge_old,
 720	.version	= 1,
 721	.msglen		= 0,
 722};
 723
 724/* Globals */
 725#define NFSD_PIPE_DIR		"nfsd"
 726#define NFSD_CLD_PIPE		"cld"
 727
 728/* per-net-ns structure for holding cld upcall info */
 729struct cld_net {
 730	struct rpc_pipe		*cn_pipe;
 731	spinlock_t		 cn_lock;
 732	struct list_head	 cn_list;
 733	unsigned int		 cn_xid;
 734	bool			 cn_has_legacy;
 735	struct crypto_shash	*cn_tfm;
 736};
 737
 738struct cld_upcall {
 739	struct list_head	 cu_list;
 740	struct cld_net		*cu_net;
 741	struct completion	 cu_done;
 742	union {
 743		struct cld_msg_hdr	 cu_hdr;
 744		struct cld_msg		 cu_msg;
 745		struct cld_msg_v2	 cu_msg_v2;
 746	} cu_u;
 747};
 748
 749static int
 750__cld_pipe_upcall(struct rpc_pipe *pipe, void *cmsg, struct nfsd_net *nn)
 751{
 752	int ret;
 753	struct rpc_pipe_msg msg;
 754	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall, cu_u);
 755
 756	memset(&msg, 0, sizeof(msg));
 757	msg.data = cmsg;
 758	msg.len = nn->client_tracking_ops->msglen;
 759
 
 
 
 
 
 760	ret = rpc_queue_upcall(pipe, &msg);
 761	if (ret < 0) {
 
 762		goto out;
 763	}
 764
 765	wait_for_completion(&cup->cu_done);
 766
 767	if (msg.errno < 0)
 768		ret = msg.errno;
 769out:
 770	return ret;
 771}
 772
 773static int
 774cld_pipe_upcall(struct rpc_pipe *pipe, void *cmsg, struct nfsd_net *nn)
 775{
 776	int ret;
 777
 778	/*
 779	 * -EAGAIN occurs when pipe is closed and reopened while there are
 780	 *  upcalls queued.
 781	 */
 782	do {
 783		ret = __cld_pipe_upcall(pipe, cmsg, nn);
 784	} while (ret == -EAGAIN);
 785
 786	return ret;
 787}
 788
 789static ssize_t
 790__cld_pipe_inprogress_downcall(const struct cld_msg_v2 __user *cmsg,
 791		struct nfsd_net *nn)
 792{
 793	uint8_t cmd, princhashlen;
 794	struct xdr_netobj name, princhash = { .len = 0, .data = NULL };
 795	uint16_t namelen;
 796	struct cld_net *cn = nn->cld_net;
 797
 798	if (get_user(cmd, &cmsg->cm_cmd)) {
 799		dprintk("%s: error when copying cmd from userspace", __func__);
 800		return -EFAULT;
 801	}
 802	if (cmd == Cld_GraceStart) {
 803		if (nn->client_tracking_ops->version >= 2) {
 804			const struct cld_clntinfo __user *ci;
 805
 806			ci = &cmsg->cm_u.cm_clntinfo;
 807			if (get_user(namelen, &ci->cc_name.cn_len))
 808				return -EFAULT;
 809			name.data = memdup_user(&ci->cc_name.cn_id, namelen);
 810			if (IS_ERR_OR_NULL(name.data))
 811				return -EFAULT;
 812			name.len = namelen;
 813			get_user(princhashlen, &ci->cc_princhash.cp_len);
 814			if (princhashlen > 0) {
 815				princhash.data = memdup_user(
 816						&ci->cc_princhash.cp_data,
 817						princhashlen);
 818				if (IS_ERR_OR_NULL(princhash.data))
 819					return -EFAULT;
 820				princhash.len = princhashlen;
 821			} else
 822				princhash.len = 0;
 823		} else {
 824			const struct cld_name __user *cnm;
 825
 826			cnm = &cmsg->cm_u.cm_name;
 827			if (get_user(namelen, &cnm->cn_len))
 828				return -EFAULT;
 829			name.data = memdup_user(&cnm->cn_id, namelen);
 830			if (IS_ERR_OR_NULL(name.data))
 831				return -EFAULT;
 832			name.len = namelen;
 833		}
 834		if (name.len > 5 && memcmp(name.data, "hash:", 5) == 0) {
 835			name.len = name.len - 5;
 836			memmove(name.data, name.data + 5, name.len);
 837			cn->cn_has_legacy = true;
 838		}
 839		if (!nfs4_client_to_reclaim(name, princhash, nn)) {
 840			kfree(name.data);
 841			kfree(princhash.data);
 842			return -EFAULT;
 843		}
 844		return nn->client_tracking_ops->msglen;
 845	}
 846	return -EFAULT;
 847}
 848
 849static ssize_t
 850cld_pipe_downcall(struct file *filp, const char __user *src, size_t mlen)
 851{
 852	struct cld_upcall *tmp, *cup;
 853	struct cld_msg_hdr __user *hdr = (struct cld_msg_hdr __user *)src;
 854	struct cld_msg_v2 __user *cmsg = (struct cld_msg_v2 __user *)src;
 855	uint32_t xid;
 856	struct nfsd_net *nn = net_generic(file_inode(filp)->i_sb->s_fs_info,
 857						nfsd_net_id);
 858	struct cld_net *cn = nn->cld_net;
 859	int16_t status;
 860
 861	if (mlen != nn->client_tracking_ops->msglen) {
 862		dprintk("%s: got %zu bytes, expected %zu\n", __func__, mlen,
 863			nn->client_tracking_ops->msglen);
 864		return -EINVAL;
 865	}
 866
 867	/* copy just the xid so we can try to find that */
 868	if (copy_from_user(&xid, &hdr->cm_xid, sizeof(xid)) != 0) {
 869		dprintk("%s: error when copying xid from userspace", __func__);
 870		return -EFAULT;
 871	}
 872
 873	/*
 874	 * copy the status so we know whether to remove the upcall from the
 875	 * list (for -EINPROGRESS, we just want to make sure the xid is
 876	 * valid, not remove the upcall from the list)
 877	 */
 878	if (get_user(status, &hdr->cm_status)) {
 879		dprintk("%s: error when copying status from userspace", __func__);
 880		return -EFAULT;
 881	}
 882
 883	/* walk the list and find corresponding xid */
 884	cup = NULL;
 885	spin_lock(&cn->cn_lock);
 886	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
 887		if (get_unaligned(&tmp->cu_u.cu_hdr.cm_xid) == xid) {
 888			cup = tmp;
 889			if (status != -EINPROGRESS)
 890				list_del_init(&cup->cu_list);
 891			break;
 892		}
 893	}
 894	spin_unlock(&cn->cn_lock);
 895
 896	/* couldn't find upcall? */
 897	if (!cup) {
 898		dprintk("%s: couldn't find upcall -- xid=%u\n", __func__, xid);
 899		return -EINVAL;
 900	}
 901
 902	if (status == -EINPROGRESS)
 903		return __cld_pipe_inprogress_downcall(cmsg, nn);
 904
 905	if (copy_from_user(&cup->cu_u.cu_msg_v2, src, mlen) != 0)
 906		return -EFAULT;
 907
 908	complete(&cup->cu_done);
 909	return mlen;
 910}
 911
 912static void
 913cld_pipe_destroy_msg(struct rpc_pipe_msg *msg)
 914{
 915	struct cld_msg *cmsg = msg->data;
 916	struct cld_upcall *cup = container_of(cmsg, struct cld_upcall,
 917						 cu_u.cu_msg);
 918
 919	/* errno >= 0 means we got a downcall */
 920	if (msg->errno >= 0)
 921		return;
 922
 923	complete(&cup->cu_done);
 924}
 925
 926static const struct rpc_pipe_ops cld_upcall_ops = {
 927	.upcall		= rpc_pipe_generic_upcall,
 928	.downcall	= cld_pipe_downcall,
 929	.destroy_msg	= cld_pipe_destroy_msg,
 930};
 931
 932static struct dentry *
 933nfsd4_cld_register_sb(struct super_block *sb, struct rpc_pipe *pipe)
 934{
 935	struct dentry *dir, *dentry;
 936
 937	dir = rpc_d_lookup_sb(sb, NFSD_PIPE_DIR);
 938	if (dir == NULL)
 939		return ERR_PTR(-ENOENT);
 940	dentry = rpc_mkpipe_dentry(dir, NFSD_CLD_PIPE, NULL, pipe);
 941	dput(dir);
 942	return dentry;
 943}
 944
 945static void
 946nfsd4_cld_unregister_sb(struct rpc_pipe *pipe)
 947{
 948	if (pipe->dentry)
 949		rpc_unlink(pipe->dentry);
 950}
 951
 952static struct dentry *
 953nfsd4_cld_register_net(struct net *net, struct rpc_pipe *pipe)
 954{
 955	struct super_block *sb;
 956	struct dentry *dentry;
 957
 958	sb = rpc_get_sb_net(net);
 959	if (!sb)
 960		return NULL;
 961	dentry = nfsd4_cld_register_sb(sb, pipe);
 962	rpc_put_sb_net(net);
 963	return dentry;
 964}
 965
 966static void
 967nfsd4_cld_unregister_net(struct net *net, struct rpc_pipe *pipe)
 968{
 969	struct super_block *sb;
 970
 971	sb = rpc_get_sb_net(net);
 972	if (sb) {
 973		nfsd4_cld_unregister_sb(pipe);
 974		rpc_put_sb_net(net);
 975	}
 976}
 977
 978/* Initialize rpc_pipefs pipe for communication with client tracking daemon */
 979static int
 980__nfsd4_init_cld_pipe(struct net *net)
 981{
 982	int ret;
 983	struct dentry *dentry;
 984	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
 985	struct cld_net *cn;
 986
 987	if (nn->cld_net)
 988		return 0;
 989
 990	cn = kzalloc(sizeof(*cn), GFP_KERNEL);
 991	if (!cn) {
 992		ret = -ENOMEM;
 993		goto err;
 994	}
 995
 996	cn->cn_pipe = rpc_mkpipe_data(&cld_upcall_ops, RPC_PIPE_WAIT_FOR_OPEN);
 997	if (IS_ERR(cn->cn_pipe)) {
 998		ret = PTR_ERR(cn->cn_pipe);
 999		goto err;
1000	}
1001	spin_lock_init(&cn->cn_lock);
1002	INIT_LIST_HEAD(&cn->cn_list);
1003
1004	dentry = nfsd4_cld_register_net(net, cn->cn_pipe);
1005	if (IS_ERR(dentry)) {
1006		ret = PTR_ERR(dentry);
1007		goto err_destroy_data;
1008	}
1009
1010	cn->cn_pipe->dentry = dentry;
1011	cn->cn_has_legacy = false;
1012	nn->cld_net = cn;
1013	return 0;
1014
1015err_destroy_data:
1016	rpc_destroy_pipe_data(cn->cn_pipe);
1017err:
1018	kfree(cn);
1019	printk(KERN_ERR "NFSD: unable to create nfsdcld upcall pipe (%d)\n",
1020			ret);
1021	return ret;
1022}
1023
1024static int
1025nfsd4_init_cld_pipe(struct net *net)
1026{
1027	int status;
1028
1029	status = __nfsd4_init_cld_pipe(net);
1030	if (!status)
1031		printk("NFSD: Using old nfsdcld client tracking operations.\n");
1032	return status;
1033}
1034
1035static void
1036nfsd4_remove_cld_pipe(struct net *net)
1037{
1038	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1039	struct cld_net *cn = nn->cld_net;
1040
1041	nfsd4_cld_unregister_net(net, cn->cn_pipe);
1042	rpc_destroy_pipe_data(cn->cn_pipe);
1043	if (cn->cn_tfm)
1044		crypto_free_shash(cn->cn_tfm);
1045	kfree(nn->cld_net);
1046	nn->cld_net = NULL;
1047}
1048
1049static struct cld_upcall *
1050alloc_cld_upcall(struct nfsd_net *nn)
1051{
1052	struct cld_upcall *new, *tmp;
1053	struct cld_net *cn = nn->cld_net;
1054
1055	new = kzalloc(sizeof(*new), GFP_KERNEL);
1056	if (!new)
1057		return new;
1058
1059	/* FIXME: hard cap on number in flight? */
1060restart_search:
1061	spin_lock(&cn->cn_lock);
1062	list_for_each_entry(tmp, &cn->cn_list, cu_list) {
1063		if (tmp->cu_u.cu_msg.cm_xid == cn->cn_xid) {
1064			cn->cn_xid++;
1065			spin_unlock(&cn->cn_lock);
1066			goto restart_search;
1067		}
1068	}
1069	init_completion(&new->cu_done);
1070	new->cu_u.cu_msg.cm_vers = nn->client_tracking_ops->version;
1071	put_unaligned(cn->cn_xid++, &new->cu_u.cu_msg.cm_xid);
1072	new->cu_net = cn;
1073	list_add(&new->cu_list, &cn->cn_list);
1074	spin_unlock(&cn->cn_lock);
1075
1076	dprintk("%s: allocated xid %u\n", __func__, new->cu_u.cu_msg.cm_xid);
1077
1078	return new;
1079}
1080
1081static void
1082free_cld_upcall(struct cld_upcall *victim)
1083{
1084	struct cld_net *cn = victim->cu_net;
1085
1086	spin_lock(&cn->cn_lock);
1087	list_del(&victim->cu_list);
1088	spin_unlock(&cn->cn_lock);
1089	kfree(victim);
1090}
1091
1092/* Ask daemon to create a new record */
1093static void
1094nfsd4_cld_create(struct nfs4_client *clp)
1095{
1096	int ret;
1097	struct cld_upcall *cup;
1098	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1099	struct cld_net *cn = nn->cld_net;
1100
1101	/* Don't upcall if it's already stored */
1102	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1103		return;
1104
1105	cup = alloc_cld_upcall(nn);
1106	if (!cup) {
1107		ret = -ENOMEM;
1108		goto out_err;
1109	}
1110
1111	cup->cu_u.cu_msg.cm_cmd = Cld_Create;
1112	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1113	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1114			clp->cl_name.len);
1115
1116	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
1117	if (!ret) {
1118		ret = cup->cu_u.cu_msg.cm_status;
1119		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1120	}
1121
1122	free_cld_upcall(cup);
1123out_err:
1124	if (ret)
1125		printk(KERN_ERR "NFSD: Unable to create client "
1126				"record on stable storage: %d\n", ret);
1127}
1128
1129/* Ask daemon to create a new record */
1130static void
1131nfsd4_cld_create_v2(struct nfs4_client *clp)
1132{
1133	int ret;
1134	struct cld_upcall *cup;
1135	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1136	struct cld_net *cn = nn->cld_net;
1137	struct cld_msg_v2 *cmsg;
1138	struct crypto_shash *tfm = cn->cn_tfm;
1139	struct xdr_netobj cksum;
1140	char *principal = NULL;
1141
1142	/* Don't upcall if it's already stored */
1143	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1144		return;
1145
1146	cup = alloc_cld_upcall(nn);
1147	if (!cup) {
1148		ret = -ENOMEM;
1149		goto out_err;
1150	}
1151
1152	cmsg = &cup->cu_u.cu_msg_v2;
1153	cmsg->cm_cmd = Cld_Create;
1154	cmsg->cm_u.cm_clntinfo.cc_name.cn_len = clp->cl_name.len;
1155	memcpy(cmsg->cm_u.cm_clntinfo.cc_name.cn_id, clp->cl_name.data,
1156			clp->cl_name.len);
1157	if (clp->cl_cred.cr_raw_principal)
1158		principal = clp->cl_cred.cr_raw_principal;
1159	else if (clp->cl_cred.cr_principal)
1160		principal = clp->cl_cred.cr_principal;
1161	if (principal) {
1162		cksum.len = crypto_shash_digestsize(tfm);
1163		cksum.data = kmalloc(cksum.len, GFP_KERNEL);
1164		if (cksum.data == NULL) {
1165			ret = -ENOMEM;
1166			goto out;
1167		}
1168		ret = crypto_shash_tfm_digest(tfm, principal, strlen(principal),
1169					      cksum.data);
1170		if (ret) {
1171			kfree(cksum.data);
1172			goto out;
1173		}
1174		cmsg->cm_u.cm_clntinfo.cc_princhash.cp_len = cksum.len;
1175		memcpy(cmsg->cm_u.cm_clntinfo.cc_princhash.cp_data,
1176		       cksum.data, cksum.len);
1177		kfree(cksum.data);
1178	} else
1179		cmsg->cm_u.cm_clntinfo.cc_princhash.cp_len = 0;
1180
1181	ret = cld_pipe_upcall(cn->cn_pipe, cmsg, nn);
1182	if (!ret) {
1183		ret = cmsg->cm_status;
1184		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1185	}
1186
1187out:
1188	free_cld_upcall(cup);
1189out_err:
1190	if (ret)
1191		pr_err("NFSD: Unable to create client record on stable storage: %d\n",
1192				ret);
1193}
1194
1195/* Ask daemon to create a new record */
1196static void
1197nfsd4_cld_remove(struct nfs4_client *clp)
1198{
1199	int ret;
1200	struct cld_upcall *cup;
1201	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1202	struct cld_net *cn = nn->cld_net;
1203
1204	/* Don't upcall if it's already removed */
1205	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1206		return;
1207
1208	cup = alloc_cld_upcall(nn);
1209	if (!cup) {
1210		ret = -ENOMEM;
1211		goto out_err;
1212	}
1213
1214	cup->cu_u.cu_msg.cm_cmd = Cld_Remove;
1215	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1216	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1217			clp->cl_name.len);
1218
1219	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
1220	if (!ret) {
1221		ret = cup->cu_u.cu_msg.cm_status;
1222		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1223	}
1224
1225	free_cld_upcall(cup);
1226out_err:
1227	if (ret)
1228		printk(KERN_ERR "NFSD: Unable to remove client "
1229				"record from stable storage: %d\n", ret);
1230}
1231
1232/*
1233 * For older nfsdcld's that do not allow us to "slurp" the clients
1234 * from the tracking database during startup.
1235 *
1236 * Check for presence of a record, and update its timestamp
1237 */
1238static int
1239nfsd4_cld_check_v0(struct nfs4_client *clp)
1240{
1241	int ret;
1242	struct cld_upcall *cup;
1243	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1244	struct cld_net *cn = nn->cld_net;
1245
1246	/* Don't upcall if one was already stored during this grace pd */
1247	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1248		return 0;
1249
1250	cup = alloc_cld_upcall(nn);
1251	if (!cup) {
1252		printk(KERN_ERR "NFSD: Unable to check client record on "
1253				"stable storage: %d\n", -ENOMEM);
1254		return -ENOMEM;
1255	}
1256
1257	cup->cu_u.cu_msg.cm_cmd = Cld_Check;
1258	cup->cu_u.cu_msg.cm_u.cm_name.cn_len = clp->cl_name.len;
1259	memcpy(cup->cu_u.cu_msg.cm_u.cm_name.cn_id, clp->cl_name.data,
1260			clp->cl_name.len);
1261
1262	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
1263	if (!ret) {
1264		ret = cup->cu_u.cu_msg.cm_status;
1265		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1266	}
1267
1268	free_cld_upcall(cup);
1269	return ret;
1270}
1271
1272/*
1273 * For newer nfsdcld's that allow us to "slurp" the clients
1274 * from the tracking database during startup.
1275 *
1276 * Check for presence of a record in the reclaim_str_hashtbl
1277 */
1278static int
1279nfsd4_cld_check(struct nfs4_client *clp)
1280{
1281	struct nfs4_client_reclaim *crp;
1282	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1283	struct cld_net *cn = nn->cld_net;
1284	int status;
1285	char dname[HEXDIR_LEN];
1286	struct xdr_netobj name;
1287
1288	/* did we already find that this client is stable? */
1289	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1290		return 0;
1291
1292	/* look for it in the reclaim hashtable otherwise */
1293	crp = nfsd4_find_reclaim_client(clp->cl_name, nn);
1294	if (crp)
1295		goto found;
1296
1297	if (cn->cn_has_legacy) {
1298		status = nfs4_make_rec_clidname(dname, &clp->cl_name);
1299		if (status)
1300			return -ENOENT;
1301
1302		name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
1303		if (!name.data) {
1304			dprintk("%s: failed to allocate memory for name.data!\n",
1305				__func__);
1306			return -ENOENT;
1307		}
1308		name.len = HEXDIR_LEN;
1309		crp = nfsd4_find_reclaim_client(name, nn);
1310		kfree(name.data);
1311		if (crp)
1312			goto found;
1313
1314	}
1315	return -ENOENT;
1316found:
1317	crp->cr_clp = clp;
1318	return 0;
1319}
1320
1321static int
1322nfsd4_cld_check_v2(struct nfs4_client *clp)
1323{
1324	struct nfs4_client_reclaim *crp;
1325	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1326	struct cld_net *cn = nn->cld_net;
1327	int status;
1328	char dname[HEXDIR_LEN];
1329	struct xdr_netobj name;
1330	struct crypto_shash *tfm = cn->cn_tfm;
1331	struct xdr_netobj cksum;
1332	char *principal = NULL;
1333
1334	/* did we already find that this client is stable? */
1335	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1336		return 0;
1337
1338	/* look for it in the reclaim hashtable otherwise */
1339	crp = nfsd4_find_reclaim_client(clp->cl_name, nn);
1340	if (crp)
1341		goto found;
1342
1343	if (cn->cn_has_legacy) {
1344		status = nfs4_make_rec_clidname(dname, &clp->cl_name);
1345		if (status)
1346			return -ENOENT;
1347
1348		name.data = kmemdup(dname, HEXDIR_LEN, GFP_KERNEL);
1349		if (!name.data) {
1350			dprintk("%s: failed to allocate memory for name.data\n",
1351					__func__);
1352			return -ENOENT;
1353		}
1354		name.len = HEXDIR_LEN;
1355		crp = nfsd4_find_reclaim_client(name, nn);
1356		kfree(name.data);
1357		if (crp)
1358			goto found;
1359
1360	}
1361	return -ENOENT;
1362found:
1363	if (crp->cr_princhash.len) {
1364		if (clp->cl_cred.cr_raw_principal)
1365			principal = clp->cl_cred.cr_raw_principal;
1366		else if (clp->cl_cred.cr_principal)
1367			principal = clp->cl_cred.cr_principal;
1368		if (principal == NULL)
1369			return -ENOENT;
1370		cksum.len = crypto_shash_digestsize(tfm);
1371		cksum.data = kmalloc(cksum.len, GFP_KERNEL);
1372		if (cksum.data == NULL)
1373			return -ENOENT;
1374		status = crypto_shash_tfm_digest(tfm, principal,
1375						 strlen(principal), cksum.data);
1376		if (status) {
1377			kfree(cksum.data);
1378			return -ENOENT;
1379		}
1380		if (memcmp(crp->cr_princhash.data, cksum.data,
1381				crp->cr_princhash.len)) {
1382			kfree(cksum.data);
1383			return -ENOENT;
1384		}
1385		kfree(cksum.data);
1386	}
1387	crp->cr_clp = clp;
1388	return 0;
1389}
1390
1391static int
1392nfsd4_cld_grace_start(struct nfsd_net *nn)
1393{
1394	int ret;
1395	struct cld_upcall *cup;
1396	struct cld_net *cn = nn->cld_net;
1397
1398	cup = alloc_cld_upcall(nn);
1399	if (!cup) {
1400		ret = -ENOMEM;
1401		goto out_err;
1402	}
1403
1404	cup->cu_u.cu_msg.cm_cmd = Cld_GraceStart;
1405	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
1406	if (!ret)
1407		ret = cup->cu_u.cu_msg.cm_status;
1408
1409	free_cld_upcall(cup);
1410out_err:
1411	if (ret)
1412		dprintk("%s: Unable to get clients from userspace: %d\n",
1413			__func__, ret);
1414	return ret;
1415}
1416
1417/* For older nfsdcld's that need cm_gracetime */
1418static void
1419nfsd4_cld_grace_done_v0(struct nfsd_net *nn)
1420{
1421	int ret;
1422	struct cld_upcall *cup;
1423	struct cld_net *cn = nn->cld_net;
1424
1425	cup = alloc_cld_upcall(nn);
1426	if (!cup) {
1427		ret = -ENOMEM;
1428		goto out_err;
1429	}
1430
1431	cup->cu_u.cu_msg.cm_cmd = Cld_GraceDone;
1432	cup->cu_u.cu_msg.cm_u.cm_gracetime = nn->boot_time;
1433	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
1434	if (!ret)
1435		ret = cup->cu_u.cu_msg.cm_status;
1436
1437	free_cld_upcall(cup);
1438out_err:
1439	if (ret)
1440		printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1441}
1442
1443/*
1444 * For newer nfsdcld's that do not need cm_gracetime.  We also need to call
1445 * nfs4_release_reclaim() to clear out the reclaim_str_hashtbl.
1446 */
1447static void
1448nfsd4_cld_grace_done(struct nfsd_net *nn)
1449{
1450	int ret;
1451	struct cld_upcall *cup;
1452	struct cld_net *cn = nn->cld_net;
1453
1454	cup = alloc_cld_upcall(nn);
1455	if (!cup) {
1456		ret = -ENOMEM;
1457		goto out_err;
1458	}
1459
1460	cup->cu_u.cu_msg.cm_cmd = Cld_GraceDone;
1461	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
 
1462	if (!ret)
1463		ret = cup->cu_u.cu_msg.cm_status;
1464
1465	free_cld_upcall(cup);
1466out_err:
1467	nfs4_release_reclaim(nn);
1468	if (ret)
1469		printk(KERN_ERR "NFSD: Unable to end grace period: %d\n", ret);
1470}
1471
1472static int
1473nfs4_cld_state_init(struct net *net)
1474{
1475	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1476	int i;
1477
1478	nn->reclaim_str_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
1479						sizeof(struct list_head),
1480						GFP_KERNEL);
1481	if (!nn->reclaim_str_hashtbl)
1482		return -ENOMEM;
1483
1484	for (i = 0; i < CLIENT_HASH_SIZE; i++)
1485		INIT_LIST_HEAD(&nn->reclaim_str_hashtbl[i]);
1486	nn->reclaim_str_hashtbl_size = 0;
1487	nn->track_reclaim_completes = true;
1488	atomic_set(&nn->nr_reclaim_complete, 0);
1489
1490	return 0;
1491}
1492
1493static void
1494nfs4_cld_state_shutdown(struct net *net)
1495{
1496	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1497
1498	nn->track_reclaim_completes = false;
1499	kfree(nn->reclaim_str_hashtbl);
1500}
1501
1502static bool
1503cld_running(struct nfsd_net *nn)
1504{
1505	struct cld_net *cn = nn->cld_net;
1506	struct rpc_pipe *pipe = cn->cn_pipe;
1507
1508	return pipe->nreaders || pipe->nwriters;
1509}
1510
1511static int
1512nfsd4_cld_get_version(struct nfsd_net *nn)
1513{
1514	int ret = 0;
1515	struct cld_upcall *cup;
1516	struct cld_net *cn = nn->cld_net;
1517	uint8_t version;
1518
1519	cup = alloc_cld_upcall(nn);
1520	if (!cup) {
1521		ret = -ENOMEM;
1522		goto out_err;
1523	}
1524	cup->cu_u.cu_msg.cm_cmd = Cld_GetVersion;
1525	ret = cld_pipe_upcall(cn->cn_pipe, &cup->cu_u.cu_msg, nn);
1526	if (!ret) {
1527		ret = cup->cu_u.cu_msg.cm_status;
1528		if (ret)
1529			goto out_free;
1530		version = cup->cu_u.cu_msg.cm_u.cm_version;
1531		dprintk("%s: userspace returned version %u\n",
1532				__func__, version);
1533		if (version < 1)
1534			version = 1;
1535		else if (version > CLD_UPCALL_VERSION)
1536			version = CLD_UPCALL_VERSION;
1537
1538		switch (version) {
1539		case 1:
1540			nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
1541			break;
1542		case 2:
1543			nn->client_tracking_ops = &nfsd4_cld_tracking_ops_v2;
1544			break;
1545		default:
1546			break;
1547		}
1548	}
1549out_free:
1550	free_cld_upcall(cup);
1551out_err:
1552	if (ret)
1553		dprintk("%s: Unable to get version from userspace: %d\n",
1554			__func__, ret);
1555	return ret;
1556}
1557
1558static int
1559nfsd4_cld_tracking_init(struct net *net)
1560{
1561	int status;
1562	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1563	bool running;
1564	int retries = 10;
1565	struct crypto_shash *tfm;
1566
1567	status = nfs4_cld_state_init(net);
1568	if (status)
1569		return status;
1570
1571	status = __nfsd4_init_cld_pipe(net);
1572	if (status)
1573		goto err_shutdown;
1574
1575	/*
1576	 * rpc pipe upcalls take 30 seconds to time out, so we don't want to
1577	 * queue an upcall unless we know that nfsdcld is running (because we
1578	 * want this to fail fast so that nfsd4_client_tracking_init() can try
1579	 * the next client tracking method).  nfsdcld should already be running
1580	 * before nfsd is started, so the wait here is for nfsdcld to open the
1581	 * pipefs file we just created.
1582	 */
1583	while (!(running = cld_running(nn)) && retries--)
1584		msleep(100);
1585
1586	if (!running) {
1587		status = -ETIMEDOUT;
1588		goto err_remove;
1589	}
1590	tfm = crypto_alloc_shash("sha256", 0, 0);
1591	if (IS_ERR(tfm)) {
1592		status = PTR_ERR(tfm);
1593		goto err_remove;
1594	}
1595	nn->cld_net->cn_tfm = tfm;
1596
1597	status = nfsd4_cld_get_version(nn);
1598	if (status == -EOPNOTSUPP)
1599		pr_warn("NFSD: nfsdcld GetVersion upcall failed. Please upgrade nfsdcld.\n");
1600
1601	status = nfsd4_cld_grace_start(nn);
1602	if (status) {
1603		if (status == -EOPNOTSUPP)
1604			pr_warn("NFSD: nfsdcld GraceStart upcall failed. Please upgrade nfsdcld.\n");
1605		nfs4_release_reclaim(nn);
1606		goto err_remove;
1607	} else
1608		printk("NFSD: Using nfsdcld client tracking operations.\n");
1609	return 0;
1610
1611err_remove:
1612	nfsd4_remove_cld_pipe(net);
1613err_shutdown:
1614	nfs4_cld_state_shutdown(net);
1615	return status;
1616}
1617
1618static void
1619nfsd4_cld_tracking_exit(struct net *net)
1620{
1621	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1622
1623	nfs4_release_reclaim(nn);
1624	nfsd4_remove_cld_pipe(net);
1625	nfs4_cld_state_shutdown(net);
1626}
1627
1628/* For older nfsdcld's */
1629static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v0 = {
1630	.init		= nfsd4_init_cld_pipe,
1631	.exit		= nfsd4_remove_cld_pipe,
1632	.create		= nfsd4_cld_create,
1633	.remove		= nfsd4_cld_remove,
1634	.check		= nfsd4_cld_check_v0,
1635	.grace_done	= nfsd4_cld_grace_done_v0,
1636	.version	= 1,
1637	.msglen		= sizeof(struct cld_msg),
1638};
1639
1640/* For newer nfsdcld's */
1641static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops = {
1642	.init		= nfsd4_cld_tracking_init,
1643	.exit		= nfsd4_cld_tracking_exit,
1644	.create		= nfsd4_cld_create,
1645	.remove		= nfsd4_cld_remove,
1646	.check		= nfsd4_cld_check,
1647	.grace_done	= nfsd4_cld_grace_done,
1648	.version	= 1,
1649	.msglen		= sizeof(struct cld_msg),
1650};
1651
1652/* v2 create/check ops include the principal, if available */
1653static const struct nfsd4_client_tracking_ops nfsd4_cld_tracking_ops_v2 = {
1654	.init		= nfsd4_cld_tracking_init,
1655	.exit		= nfsd4_cld_tracking_exit,
1656	.create		= nfsd4_cld_create_v2,
1657	.remove		= nfsd4_cld_remove,
1658	.check		= nfsd4_cld_check_v2,
1659	.grace_done	= nfsd4_cld_grace_done,
1660	.version	= 2,
1661	.msglen		= sizeof(struct cld_msg_v2),
1662};
1663
1664/* upcall via usermodehelper */
1665static char cltrack_prog[PATH_MAX] = "/sbin/nfsdcltrack";
1666module_param_string(cltrack_prog, cltrack_prog, sizeof(cltrack_prog),
1667			S_IRUGO|S_IWUSR);
1668MODULE_PARM_DESC(cltrack_prog, "Path to the nfsdcltrack upcall program");
1669
1670static bool cltrack_legacy_disable;
1671module_param(cltrack_legacy_disable, bool, S_IRUGO|S_IWUSR);
1672MODULE_PARM_DESC(cltrack_legacy_disable,
1673		"Disable legacy recoverydir conversion. Default: false");
1674
1675#define LEGACY_TOPDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_TOPDIR="
1676#define LEGACY_RECDIR_ENV_PREFIX "NFSDCLTRACK_LEGACY_RECDIR="
1677#define HAS_SESSION_ENV_PREFIX "NFSDCLTRACK_CLIENT_HAS_SESSION="
1678#define GRACE_START_ENV_PREFIX "NFSDCLTRACK_GRACE_START="
1679
1680static char *
1681nfsd4_cltrack_legacy_topdir(void)
1682{
1683	int copied;
1684	size_t len;
1685	char *result;
1686
1687	if (cltrack_legacy_disable)
1688		return NULL;
1689
1690	len = strlen(LEGACY_TOPDIR_ENV_PREFIX) +
1691		strlen(nfs4_recoverydir()) + 1;
1692
1693	result = kmalloc(len, GFP_KERNEL);
1694	if (!result)
1695		return result;
1696
1697	copied = snprintf(result, len, LEGACY_TOPDIR_ENV_PREFIX "%s",
1698				nfs4_recoverydir());
1699	if (copied >= len) {
1700		/* just return nothing if output was truncated */
1701		kfree(result);
1702		return NULL;
1703	}
1704
1705	return result;
1706}
1707
1708static char *
1709nfsd4_cltrack_legacy_recdir(const struct xdr_netobj *name)
1710{
1711	int copied;
1712	size_t len;
1713	char *result;
1714
1715	if (cltrack_legacy_disable)
1716		return NULL;
1717
1718	/* +1 is for '/' between "topdir" and "recdir" */
1719	len = strlen(LEGACY_RECDIR_ENV_PREFIX) +
1720		strlen(nfs4_recoverydir()) + 1 + HEXDIR_LEN;
1721
1722	result = kmalloc(len, GFP_KERNEL);
1723	if (!result)
1724		return result;
1725
1726	copied = snprintf(result, len, LEGACY_RECDIR_ENV_PREFIX "%s/",
1727				nfs4_recoverydir());
1728	if (copied > (len - HEXDIR_LEN)) {
1729		/* just return nothing if output will be truncated */
1730		kfree(result);
1731		return NULL;
1732	}
1733
1734	copied = nfs4_make_rec_clidname(result + copied, name);
1735	if (copied) {
1736		kfree(result);
1737		return NULL;
1738	}
1739
1740	return result;
1741}
1742
1743static char *
1744nfsd4_cltrack_client_has_session(struct nfs4_client *clp)
1745{
1746	int copied;
1747	size_t len;
1748	char *result;
1749
1750	/* prefix + Y/N character + terminating NULL */
1751	len = strlen(HAS_SESSION_ENV_PREFIX) + 1 + 1;
1752
1753	result = kmalloc(len, GFP_KERNEL);
1754	if (!result)
1755		return result;
1756
1757	copied = snprintf(result, len, HAS_SESSION_ENV_PREFIX "%c",
1758				clp->cl_minorversion ? 'Y' : 'N');
1759	if (copied >= len) {
1760		/* just return nothing if output was truncated */
1761		kfree(result);
1762		return NULL;
1763	}
1764
1765	return result;
1766}
1767
1768static char *
1769nfsd4_cltrack_grace_start(time64_t grace_start)
1770{
1771	int copied;
1772	size_t len;
1773	char *result;
1774
1775	/* prefix + max width of int64_t string + terminating NULL */
1776	len = strlen(GRACE_START_ENV_PREFIX) + 22 + 1;
1777
1778	result = kmalloc(len, GFP_KERNEL);
1779	if (!result)
1780		return result;
1781
1782	copied = snprintf(result, len, GRACE_START_ENV_PREFIX "%lld",
1783				grace_start);
1784	if (copied >= len) {
1785		/* just return nothing if output was truncated */
1786		kfree(result);
1787		return NULL;
1788	}
1789
1790	return result;
1791}
1792
1793static int
1794nfsd4_umh_cltrack_upcall(char *cmd, char *arg, char *env0, char *env1)
1795{
1796	char *envp[3];
1797	char *argv[4];
1798	int ret;
1799
1800	if (unlikely(!cltrack_prog[0])) {
1801		dprintk("%s: cltrack_prog is disabled\n", __func__);
1802		return -EACCES;
1803	}
1804
1805	dprintk("%s: cmd: %s\n", __func__, cmd);
1806	dprintk("%s: arg: %s\n", __func__, arg ? arg : "(null)");
1807	dprintk("%s: env0: %s\n", __func__, env0 ? env0 : "(null)");
1808	dprintk("%s: env1: %s\n", __func__, env1 ? env1 : "(null)");
1809
1810	envp[0] = env0;
1811	envp[1] = env1;
1812	envp[2] = NULL;
1813
1814	argv[0] = (char *)cltrack_prog;
1815	argv[1] = cmd;
1816	argv[2] = arg;
1817	argv[3] = NULL;
1818
1819	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1820	/*
1821	 * Disable the upcall mechanism if we're getting an ENOENT or EACCES
1822	 * error. The admin can re-enable it on the fly by using sysfs
1823	 * once the problem has been fixed.
1824	 */
1825	if (ret == -ENOENT || ret == -EACCES) {
1826		dprintk("NFSD: %s was not found or isn't executable (%d). "
1827			"Setting cltrack_prog to blank string!",
1828			cltrack_prog, ret);
1829		cltrack_prog[0] = '\0';
1830	}
1831	dprintk("%s: %s return value: %d\n", __func__, cltrack_prog, ret);
1832
1833	return ret;
1834}
1835
1836static char *
1837bin_to_hex_dup(const unsigned char *src, int srclen)
1838{
1839	char *buf;
 
1840
1841	/* +1 for terminating NULL */
1842	buf = kzalloc((srclen * 2) + 1, GFP_KERNEL);
1843	if (!buf)
1844		return buf;
1845
1846	bin2hex(buf, src, srclen);
 
 
 
 
1847	return buf;
1848}
1849
1850static int
1851nfsd4_umh_cltrack_init(struct net *net)
1852{
1853	int ret;
1854	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1855	char *grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1856
1857	/* XXX: The usermode helper s not working in container yet. */
1858	if (net != &init_net) {
1859		pr_warn("NFSD: attempt to initialize umh client tracking in a container ignored.\n");
1860		kfree(grace_start);
1861		return -EINVAL;
1862	}
1863
1864	ret = nfsd4_umh_cltrack_upcall("init", NULL, grace_start, NULL);
1865	kfree(grace_start);
1866	if (!ret)
1867		printk("NFSD: Using UMH upcall client tracking operations.\n");
1868	return ret;
1869}
1870
1871static void
1872nfsd4_cltrack_upcall_lock(struct nfs4_client *clp)
1873{
1874	wait_on_bit_lock(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK,
1875			 TASK_UNINTERRUPTIBLE);
1876}
1877
1878static void
1879nfsd4_cltrack_upcall_unlock(struct nfs4_client *clp)
1880{
1881	smp_mb__before_atomic();
1882	clear_bit(NFSD4_CLIENT_UPCALL_LOCK, &clp->cl_flags);
1883	smp_mb__after_atomic();
1884	wake_up_bit(&clp->cl_flags, NFSD4_CLIENT_UPCALL_LOCK);
1885}
1886
1887static void
1888nfsd4_umh_cltrack_create(struct nfs4_client *clp)
1889{
1890	char *hexid, *has_session, *grace_start;
1891	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1892
1893	/*
1894	 * With v4.0 clients, there's little difference in outcome between a
1895	 * create and check operation, and we can end up calling into this
1896	 * function multiple times per client (once for each openowner). So,
1897	 * for v4.0 clients skip upcalling once the client has been recorded
1898	 * on stable storage.
1899	 *
1900	 * For v4.1+ clients, the outcome of the two operations is different,
1901	 * so we must ensure that we upcall for the create operation. v4.1+
1902	 * clients call this on RECLAIM_COMPLETE though, so we should only end
1903	 * up doing a single create upcall per client.
1904	 */
1905	if (clp->cl_minorversion == 0 &&
1906	    test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1907		return;
1908
1909	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1910	if (!hexid) {
1911		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1912		return;
1913	}
1914
1915	has_session = nfsd4_cltrack_client_has_session(clp);
1916	grace_start = nfsd4_cltrack_grace_start(nn->boot_time);
1917
1918	nfsd4_cltrack_upcall_lock(clp);
1919	if (!nfsd4_umh_cltrack_upcall("create", hexid, has_session, grace_start))
1920		set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1921	nfsd4_cltrack_upcall_unlock(clp);
1922
1923	kfree(has_session);
1924	kfree(grace_start);
1925	kfree(hexid);
1926}
1927
1928static void
1929nfsd4_umh_cltrack_remove(struct nfs4_client *clp)
1930{
1931	char *hexid;
1932
1933	if (!test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1934		return;
1935
1936	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1937	if (!hexid) {
1938		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1939		return;
1940	}
1941
1942	nfsd4_cltrack_upcall_lock(clp);
1943	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags) &&
1944	    nfsd4_umh_cltrack_upcall("remove", hexid, NULL, NULL) == 0)
1945		clear_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1946	nfsd4_cltrack_upcall_unlock(clp);
1947
1948	kfree(hexid);
1949}
1950
1951static int
1952nfsd4_umh_cltrack_check(struct nfs4_client *clp)
1953{
1954	int ret;
1955	char *hexid, *has_session, *legacy;
1956
1957	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags))
1958		return 0;
1959
1960	hexid = bin_to_hex_dup(clp->cl_name.data, clp->cl_name.len);
1961	if (!hexid) {
1962		dprintk("%s: can't allocate memory for upcall!\n", __func__);
1963		return -ENOMEM;
1964	}
1965
1966	has_session = nfsd4_cltrack_client_has_session(clp);
1967	legacy = nfsd4_cltrack_legacy_recdir(&clp->cl_name);
1968
1969	nfsd4_cltrack_upcall_lock(clp);
1970	if (test_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags)) {
1971		ret = 0;
1972	} else {
1973		ret = nfsd4_umh_cltrack_upcall("check", hexid, has_session, legacy);
1974		if (ret == 0)
1975			set_bit(NFSD4_CLIENT_STABLE, &clp->cl_flags);
1976	}
1977	nfsd4_cltrack_upcall_unlock(clp);
1978	kfree(has_session);
1979	kfree(legacy);
1980	kfree(hexid);
1981
1982	return ret;
1983}
1984
1985static void
1986nfsd4_umh_cltrack_grace_done(struct nfsd_net *nn)
1987{
1988	char *legacy;
1989	char timestr[22]; /* FIXME: better way to determine max size? */
1990
1991	sprintf(timestr, "%lld", nn->boot_time);
1992	legacy = nfsd4_cltrack_legacy_topdir();
1993	nfsd4_umh_cltrack_upcall("gracedone", timestr, legacy, NULL);
1994	kfree(legacy);
1995}
1996
1997static const struct nfsd4_client_tracking_ops nfsd4_umh_tracking_ops = {
1998	.init		= nfsd4_umh_cltrack_init,
1999	.exit		= NULL,
2000	.create		= nfsd4_umh_cltrack_create,
2001	.remove		= nfsd4_umh_cltrack_remove,
2002	.check		= nfsd4_umh_cltrack_check,
2003	.grace_done	= nfsd4_umh_cltrack_grace_done,
2004	.version	= 1,
2005	.msglen		= 0,
2006};
2007
2008int
2009nfsd4_client_tracking_init(struct net *net)
2010{
2011	int status;
2012	struct path path;
2013	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2014
2015	/* just run the init if it the method is already decided */
2016	if (nn->client_tracking_ops)
2017		goto do_init;
2018
2019	/* First, try to use nfsdcld */
2020	nn->client_tracking_ops = &nfsd4_cld_tracking_ops;
2021	status = nn->client_tracking_ops->init(net);
2022	if (!status)
2023		return status;
2024	if (status != -ETIMEDOUT) {
2025		nn->client_tracking_ops = &nfsd4_cld_tracking_ops_v0;
2026		status = nn->client_tracking_ops->init(net);
2027		if (!status)
2028			return status;
2029	}
2030
2031	/*
2032	 * Next, try the UMH upcall.
 
2033	 */
2034	nn->client_tracking_ops = &nfsd4_umh_tracking_ops;
2035	status = nn->client_tracking_ops->init(net);
2036	if (!status)
2037		return status;
2038
2039	/*
2040	 * Finally, See if the recoverydir exists and is a directory.
2041	 * If it is, then use the legacy ops.
2042	 */
2043	nn->client_tracking_ops = &nfsd4_legacy_tracking_ops;
2044	status = kern_path(nfs4_recoverydir(), LOOKUP_FOLLOW, &path);
2045	if (!status) {
2046		status = d_is_dir(path.dentry);
2047		path_put(&path);
2048		if (!status) {
2049			status = -EINVAL;
2050			goto out;
2051		}
2052	}
2053
 
 
 
 
 
2054do_init:
2055	status = nn->client_tracking_ops->init(net);
2056out:
2057	if (status) {
2058		printk(KERN_WARNING "NFSD: Unable to initialize client "
2059				    "recovery tracking! (%d)\n", status);
2060		nn->client_tracking_ops = NULL;
2061	}
2062	return status;
2063}
2064
2065void
2066nfsd4_client_tracking_exit(struct net *net)
2067{
2068	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2069
2070	if (nn->client_tracking_ops) {
2071		if (nn->client_tracking_ops->exit)
2072			nn->client_tracking_ops->exit(net);
2073		nn->client_tracking_ops = NULL;
2074	}
2075}
2076
2077void
2078nfsd4_client_record_create(struct nfs4_client *clp)
2079{
2080	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2081
2082	if (nn->client_tracking_ops)
2083		nn->client_tracking_ops->create(clp);
2084}
2085
2086void
2087nfsd4_client_record_remove(struct nfs4_client *clp)
2088{
2089	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2090
2091	if (nn->client_tracking_ops)
2092		nn->client_tracking_ops->remove(clp);
2093}
2094
2095int
2096nfsd4_client_record_check(struct nfs4_client *clp)
2097{
2098	struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2099
2100	if (nn->client_tracking_ops)
2101		return nn->client_tracking_ops->check(clp);
2102
2103	return -EOPNOTSUPP;
2104}
2105
2106void
2107nfsd4_record_grace_done(struct nfsd_net *nn)
2108{
2109	if (nn->client_tracking_ops)
2110		nn->client_tracking_ops->grace_done(nn);
2111}
2112
2113static int
2114rpc_pipefs_event(struct notifier_block *nb, unsigned long event, void *ptr)
2115{
2116	struct super_block *sb = ptr;
2117	struct net *net = sb->s_fs_info;
2118	struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2119	struct cld_net *cn = nn->cld_net;
2120	struct dentry *dentry;
2121	int ret = 0;
2122
2123	if (!try_module_get(THIS_MODULE))
2124		return 0;
2125
2126	if (!cn) {
2127		module_put(THIS_MODULE);
2128		return 0;
2129	}
2130
2131	switch (event) {
2132	case RPC_PIPEFS_MOUNT:
2133		dentry = nfsd4_cld_register_sb(sb, cn->cn_pipe);
2134		if (IS_ERR(dentry)) {
2135			ret = PTR_ERR(dentry);
2136			break;
2137		}
2138		cn->cn_pipe->dentry = dentry;
2139		break;
2140	case RPC_PIPEFS_UMOUNT:
2141		if (cn->cn_pipe->dentry)
2142			nfsd4_cld_unregister_sb(cn->cn_pipe);
2143		break;
2144	default:
2145		ret = -ENOTSUPP;
2146		break;
2147	}
2148	module_put(THIS_MODULE);
2149	return ret;
2150}
2151
2152static struct notifier_block nfsd4_cld_block = {
2153	.notifier_call = rpc_pipefs_event,
2154};
2155
2156int
2157register_cld_notifier(void)
2158{
2159	return rpc_pipefs_notifier_register(&nfsd4_cld_block);
2160}
2161
2162void
2163unregister_cld_notifier(void)
2164{
2165	rpc_pipefs_notifier_unregister(&nfsd4_cld_block);
2166}