1/*
   2 * POSIX message queues filesystem for Linux.
   3 *
   4 * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
   5 *                          Michal Wronski          (michal.wronski@gmail.com)
   6 *
   7 * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
   8 * Lockless receive & send, fd based notify:
   9 *			    Manfred Spraul	    (manfred@colorfullife.com)
  10 *
  11 * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
  12 *
  13 * This file is released under the GPL.
  14 */
  15
  16#include <linux/capability.h>
  17#include <linux/init.h>
  18#include <linux/pagemap.h>
  19#include <linux/file.h>
  20#include <linux/mount.h>
  21#include <linux/fs_context.h>
  22#include <linux/namei.h>
  23#include <linux/sysctl.h>
  24#include <linux/poll.h>
  25#include <linux/mqueue.h>
  26#include <linux/msg.h>
  27#include <linux/skbuff.h>
  28#include <linux/vmalloc.h>
  29#include <linux/netlink.h>
  30#include <linux/syscalls.h>
  31#include <linux/audit.h>
  32#include <linux/signal.h>
  33#include <linux/mutex.h>
  34#include <linux/nsproxy.h>
  35#include <linux/pid.h>
  36#include <linux/ipc_namespace.h>
  37#include <linux/user_namespace.h>
  38#include <linux/slab.h>
  39#include <linux/sched/wake_q.h>
  40#include <linux/sched/signal.h>
  41#include <linux/sched/user.h>
  42
  43#include <net/sock.h>
  44#include "util.h"
  45
  46struct mqueue_fs_context {
  47	struct ipc_namespace	*ipc_ns;
  48	bool			 newns;	/* Set if newly created ipc namespace */
  49};
  50
  51#define MQUEUE_MAGIC	0x19800202
  52#define DIRENT_SIZE	20
  53#define FILENT_SIZE	80
  54
  55#define SEND		0
  56#define RECV		1
  57
  58#define STATE_NONE	0
  59#define STATE_READY	1
  60
  61struct posix_msg_tree_node {
  62	struct rb_node		rb_node;
  63	struct list_head	msg_list;
  64	int			priority;
  65};
  66
  67/*
  68 * Locking:
  69 *
  70 * Accesses to a message queue are synchronized by acquiring info->lock.
  71 *
  72 * There are two notable exceptions:
  73 * - The actual wakeup of a sleeping task is performed using the wake_q
  74 *   framework. info->lock is already released when wake_up_q is called.
  75 * - The exit codepaths after sleeping check ext_wait_queue->state without
  76 *   any locks. If it is STATE_READY, then the syscall is completed without
  77 *   acquiring info->lock.
  78 *
  79 * MQ_BARRIER:
  80 * To achieve proper release/acquire memory barrier pairing, the state is set to
  81 * STATE_READY with smp_store_release(), and it is read with READ_ONCE followed
  82 * by smp_acquire__after_ctrl_dep(). In addition, wake_q_add_safe() is used.
  83 *
  84 * This prevents the following races:
  85 *
  86 * 1) With the simple wake_q_add(), the task could be gone already before
  87 *    the increase of the reference happens
  88 * Thread A
  89 *				Thread B
  90 * WRITE_ONCE(wait.state, STATE_NONE);
  91 * schedule_hrtimeout()
  92 *				wake_q_add(A)
  93 *				if (cmpxchg()) // success
  94 *				   ->state = STATE_READY (reordered)
  95 * <timeout returns>
  96 * if (wait.state == STATE_READY) return;
  97 * sysret to user space
  98 * sys_exit()
  99 *				get_task_struct() // UaF
 100 *
 101 * Solution: Use wake_q_add_safe() and perform the get_task_struct() before
 102 * the smp_store_release() that does ->state = STATE_READY.
 103 *
 104 * 2) Without proper _release/_acquire barriers, the woken up task
 105 *    could read stale data
 106 *
 107 * Thread A
 108 *				Thread B
 109 * do_mq_timedreceive
 110 * WRITE_ONCE(wait.state, STATE_NONE);
 111 * schedule_hrtimeout()
 112 *				state = STATE_READY;
 113 * <timeout returns>
 114 * if (wait.state == STATE_READY) return;
 115 * msg_ptr = wait.msg;		// Access to stale data!
 116 *				receiver->msg = message; (reordered)
 117 *
 118 * Solution: use _release and _acquire barriers.
 119 *
 120 * 3) There is intentionally no barrier when setting current->state
 121 *    to TASK_INTERRUPTIBLE: spin_unlock(&info->lock) provides the
 122 *    release memory barrier, and the wakeup is triggered when holding
 123 *    info->lock, i.e. spin_lock(&info->lock) provided a pairing
 124 *    acquire memory barrier.
 125 */
 126
 127struct ext_wait_queue {		/* queue of sleeping tasks */
 128	struct task_struct *task;
 129	struct list_head list;
 130	struct msg_msg *msg;	/* ptr of loaded message */
 131	int state;		/* one of STATE_* values */
 132};
 133
 134struct mqueue_inode_info {
 135	spinlock_t lock;
 136	struct inode vfs_inode;
 137	wait_queue_head_t wait_q;
 138
 139	struct rb_root msg_tree;
 140	struct rb_node *msg_tree_rightmost;
 141	struct posix_msg_tree_node *node_cache;
 142	struct mq_attr attr;
 143
 144	struct sigevent notify;
 145	struct pid *notify_owner;
 146	u32 notify_self_exec_id;
 147	struct user_namespace *notify_user_ns;
 148	struct ucounts *ucounts;	/* user who created, for accounting */
 149	struct sock *notify_sock;
 150	struct sk_buff *notify_cookie;
 151
 152	/* for tasks waiting for free space and messages, respectively */
 153	struct ext_wait_queue e_wait_q[2];
 154
 155	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
 156};
 157
 158static struct file_system_type mqueue_fs_type;
 159static const struct inode_operations mqueue_dir_inode_operations;
 160static const struct file_operations mqueue_file_operations;
 161static const struct super_operations mqueue_super_ops;
 162static const struct fs_context_operations mqueue_fs_context_ops;
 163static void remove_notification(struct mqueue_inode_info *info);
 164
 165static struct kmem_cache *mqueue_inode_cachep;
 166
 167static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
 168{
 169	return container_of(inode, struct mqueue_inode_info, vfs_inode);
 170}
 171
 172/*
 173 * This routine should be called with the mq_lock held.
 174 */
 175static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
 176{
 177	return get_ipc_ns(inode->i_sb->s_fs_info);
 178}
 179
 180static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
 181{
 182	struct ipc_namespace *ns;
 183
 184	spin_lock(&mq_lock);
 185	ns = __get_ns_from_inode(inode);
 186	spin_unlock(&mq_lock);
 187	return ns;
 188}
 189
 190/* Auxiliary functions to manipulate messages' list */
 191static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
 192{
 193	struct rb_node **p, *parent = NULL;
 194	struct posix_msg_tree_node *leaf;
 195	bool rightmost = true;
 196
 197	p = &info->msg_tree.rb_node;
 198	while (*p) {
 199		parent = *p;
 200		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
 201
 202		if (likely(leaf->priority == msg->m_type))
 203			goto insert_msg;
 204		else if (msg->m_type < leaf->priority) {
 205			p = &(*p)->rb_left;
 206			rightmost = false;
 207		} else
 208			p = &(*p)->rb_right;
 209	}
 210	if (info->node_cache) {
 211		leaf = info->node_cache;
 212		info->node_cache = NULL;
 213	} else {
 214		leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
 215		if (!leaf)
 216			return -ENOMEM;
 217		INIT_LIST_HEAD(&leaf->msg_list);
 218	}
 219	leaf->priority = msg->m_type;
 220
 221	if (rightmost)
 222		info->msg_tree_rightmost = &leaf->rb_node;
 223
 224	rb_link_node(&leaf->rb_node, parent, p);
 225	rb_insert_color(&leaf->rb_node, &info->msg_tree);
 226insert_msg:
 227	info->attr.mq_curmsgs++;
 228	info->qsize += msg->m_ts;
 229	list_add_tail(&msg->m_list, &leaf->msg_list);
 230	return 0;
 231}
 232
 233static inline void msg_tree_erase(struct posix_msg_tree_node *leaf,
 234				  struct mqueue_inode_info *info)
 235{
 236	struct rb_node *node = &leaf->rb_node;
 237
 238	if (info->msg_tree_rightmost == node)
 239		info->msg_tree_rightmost = rb_prev(node);
 240
 241	rb_erase(node, &info->msg_tree);
 242	if (info->node_cache)
 243		kfree(leaf);
 244	else
 245		info->node_cache = leaf;
 246}
 247
 248static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 249{
 250	struct rb_node *parent = NULL;
 251	struct posix_msg_tree_node *leaf;
 252	struct msg_msg *msg;
 253
 254try_again:
 255	/*
 256	 * During insert, low priorities go to the left and high to the
 257	 * right.  On receive, we want the highest priorities first, so
 258	 * walk all the way to the right.
 259	 */
 260	parent = info->msg_tree_rightmost;
 261	if (!parent) {
 262		if (info->attr.mq_curmsgs) {
 263			pr_warn_once("Inconsistency in POSIX message queue, "
 264				     "no tree element, but supposedly messages "
 265				     "should exist!\n");
 266			info->attr.mq_curmsgs = 0;
 267		}
 268		return NULL;
 269	}
 270	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
 271	if (unlikely(list_empty(&leaf->msg_list))) {
 272		pr_warn_once("Inconsistency in POSIX message queue, "
 273			     "empty leaf node but we haven't implemented "
 274			     "lazy leaf delete!\n");
 275		msg_tree_erase(leaf, info);
 276		goto try_again;
 277	} else {
 278		msg = list_first_entry(&leaf->msg_list,
 279				       struct msg_msg, m_list);
 280		list_del(&msg->m_list);
 281		if (list_empty(&leaf->msg_list)) {
 282			msg_tree_erase(leaf, info);
 283		}
 284	}
 285	info->attr.mq_curmsgs--;
 286	info->qsize -= msg->m_ts;
 287	return msg;
 288}
 289
 290static struct inode *mqueue_get_inode(struct super_block *sb,
 291		struct ipc_namespace *ipc_ns, umode_t mode,
 292		struct mq_attr *attr)
 293{
 294	struct inode *inode;
 295	int ret = -ENOMEM;
 296
 297	inode = new_inode(sb);
 298	if (!inode)
 299		goto err;
 300
 301	inode->i_ino = get_next_ino();
 302	inode->i_mode = mode;
 303	inode->i_uid = current_fsuid();
 304	inode->i_gid = current_fsgid();
 305	inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
 306
 307	if (S_ISREG(mode)) {
 308		struct mqueue_inode_info *info;
 309		unsigned long mq_bytes, mq_treesize;
 310
 311		inode->i_fop = &mqueue_file_operations;
 312		inode->i_size = FILENT_SIZE;
 313		/* mqueue specific info */
 314		info = MQUEUE_I(inode);
 315		spin_lock_init(&info->lock);
 316		init_waitqueue_head(&info->wait_q);
 317		INIT_LIST_HEAD(&info->e_wait_q[0].list);
 318		INIT_LIST_HEAD(&info->e_wait_q[1].list);
 319		info->notify_owner = NULL;
 320		info->notify_user_ns = NULL;
 321		info->qsize = 0;
 322		info->ucounts = NULL;	/* set when all is ok */
 323		info->msg_tree = RB_ROOT;
 324		info->msg_tree_rightmost = NULL;
 325		info->node_cache = NULL;
 326		memset(&info->attr, 0, sizeof(info->attr));
 327		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
 328					   ipc_ns->mq_msg_default);
 329		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
 330					    ipc_ns->mq_msgsize_default);
 331		if (attr) {
 332			info->attr.mq_maxmsg = attr->mq_maxmsg;
 333			info->attr.mq_msgsize = attr->mq_msgsize;
 334		}
 335		/*
 336		 * We used to allocate a static array of pointers and account
 337		 * the size of that array as well as one msg_msg struct per
 338		 * possible message into the queue size. That's no longer
 339		 * accurate as the queue is now an rbtree and will grow and
 340		 * shrink depending on usage patterns.  We can, however, still
 341		 * account one msg_msg struct per message, but the nodes are
 342		 * allocated depending on priority usage, and most programs
 343		 * only use one, or a handful, of priorities.  However, since
 344		 * this is pinned memory, we need to assume worst case, so
 345		 * that means the min(mq_maxmsg, max_priorities) * struct
 346		 * posix_msg_tree_node.
 347		 */
 348
 349		ret = -EINVAL;
 350		if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
 351			goto out_inode;
 352		if (capable(CAP_SYS_RESOURCE)) {
 353			if (info->attr.mq_maxmsg > HARD_MSGMAX ||
 354			    info->attr.mq_msgsize > HARD_MSGSIZEMAX)
 355				goto out_inode;
 356		} else {
 357			if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
 358					info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
 359				goto out_inode;
 360		}
 361		ret = -EOVERFLOW;
 362		/* check for overflow */
 363		if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
 364			goto out_inode;
 365		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
 366			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
 367			sizeof(struct posix_msg_tree_node);
 368		mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
 369		if (mq_bytes + mq_treesize < mq_bytes)
 370			goto out_inode;
 371		mq_bytes += mq_treesize;
 372		info->ucounts = get_ucounts(current_ucounts());
 373		if (info->ucounts) {
 374			long msgqueue;
 375
 376			spin_lock(&mq_lock);
 377			msgqueue = inc_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes);
 378			if (msgqueue == LONG_MAX || msgqueue > rlimit(RLIMIT_MSGQUEUE)) {
 379				dec_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes);
 380				spin_unlock(&mq_lock);
 381				put_ucounts(info->ucounts);
 382				info->ucounts = NULL;
 383				/* mqueue_evict_inode() releases info->messages */
 384				ret = -EMFILE;
 385				goto out_inode;
 386			}
 387			spin_unlock(&mq_lock);
 388		}
 389	} else if (S_ISDIR(mode)) {
 390		inc_nlink(inode);
 391		/* Some things misbehave if size == 0 on a directory */
 392		inode->i_size = 2 * DIRENT_SIZE;
 393		inode->i_op = &mqueue_dir_inode_operations;
 394		inode->i_fop = &simple_dir_operations;
 395	}
 396
 397	return inode;
 398out_inode:
 399	iput(inode);
 400err:
 401	return ERR_PTR(ret);
 402}
 403
 404static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
 405{
 406	struct inode *inode;
 407	struct ipc_namespace *ns = sb->s_fs_info;
 408
 409	sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
 410	sb->s_blocksize = PAGE_SIZE;
 411	sb->s_blocksize_bits = PAGE_SHIFT;
 412	sb->s_magic = MQUEUE_MAGIC;
 413	sb->s_op = &mqueue_super_ops;
 414
 415	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 416	if (IS_ERR(inode))
 417		return PTR_ERR(inode);
 418
 419	sb->s_root = d_make_root(inode);
 420	if (!sb->s_root)
 421		return -ENOMEM;
 422	return 0;
 423}
 424
 425static int mqueue_get_tree(struct fs_context *fc)
 426{
 427	struct mqueue_fs_context *ctx = fc->fs_private;
 428
 429	/*
 430	 * With a newly created ipc namespace, we don't need to do a search
 431	 * for an ipc namespace match, but we still need to set s_fs_info.
 432	 */
 433	if (ctx->newns) {
 434		fc->s_fs_info = ctx->ipc_ns;
 435		return get_tree_nodev(fc, mqueue_fill_super);
 436	}
 437	return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns);
 438}
 439
 440static void mqueue_fs_context_free(struct fs_context *fc)
 441{
 442	struct mqueue_fs_context *ctx = fc->fs_private;
 443
 444	put_ipc_ns(ctx->ipc_ns);
 445	kfree(ctx);
 446}
 447
 448static int mqueue_init_fs_context(struct fs_context *fc)
 449{
 450	struct mqueue_fs_context *ctx;
 451
 452	ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
 453	if (!ctx)
 454		return -ENOMEM;
 455
 456	ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
 457	put_user_ns(fc->user_ns);
 458	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
 459	fc->fs_private = ctx;
 460	fc->ops = &mqueue_fs_context_ops;
 461	return 0;
 462}
 463
 464/*
 465 * mq_init_ns() is currently the only caller of mq_create_mount().
 466 * So the ns parameter is always a newly created ipc namespace.
 467 */
 468static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
 469{
 470	struct mqueue_fs_context *ctx;
 471	struct fs_context *fc;
 472	struct vfsmount *mnt;
 473
 474	fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
 475	if (IS_ERR(fc))
 476		return ERR_CAST(fc);
 477
 478	ctx = fc->fs_private;
 479	ctx->newns = true;
 480	put_ipc_ns(ctx->ipc_ns);
 481	ctx->ipc_ns = get_ipc_ns(ns);
 482	put_user_ns(fc->user_ns);
 483	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
 484
 485	mnt = fc_mount(fc);
 486	put_fs_context(fc);
 487	return mnt;
 488}
 489
 490static void init_once(void *foo)
 491{
 492	struct mqueue_inode_info *p = foo;
 493
 494	inode_init_once(&p->vfs_inode);
 495}
 496
 497static struct inode *mqueue_alloc_inode(struct super_block *sb)
 498{
 499	struct mqueue_inode_info *ei;
 500
 501	ei = alloc_inode_sb(sb, mqueue_inode_cachep, GFP_KERNEL);
 502	if (!ei)
 503		return NULL;
 504	return &ei->vfs_inode;
 505}
 506
 507static void mqueue_free_inode(struct inode *inode)
 508{
 509	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 510}
 511
 512static void mqueue_evict_inode(struct inode *inode)
 513{
 514	struct mqueue_inode_info *info;
 515	struct ipc_namespace *ipc_ns;
 516	struct msg_msg *msg, *nmsg;
 517	LIST_HEAD(tmp_msg);
 518
 519	clear_inode(inode);
 520
 521	if (S_ISDIR(inode->i_mode))
 522		return;
 523
 524	ipc_ns = get_ns_from_inode(inode);
 525	info = MQUEUE_I(inode);
 526	spin_lock(&info->lock);
 527	while ((msg = msg_get(info)) != NULL)
 528		list_add_tail(&msg->m_list, &tmp_msg);
 529	kfree(info->node_cache);
 530	spin_unlock(&info->lock);
 531
 532	list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
 533		list_del(&msg->m_list);
 534		free_msg(msg);
 535	}
 536
 537	if (info->ucounts) {
 538		unsigned long mq_bytes, mq_treesize;
 539
 540		/* Total amount of bytes accounted for the mqueue */
 541		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
 542			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
 543			sizeof(struct posix_msg_tree_node);
 544
 545		mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
 546					  info->attr.mq_msgsize);
 547
 548		spin_lock(&mq_lock);
 549		dec_rlimit_ucounts(info->ucounts, UCOUNT_RLIMIT_MSGQUEUE, mq_bytes);
 550		/*
 551		 * get_ns_from_inode() ensures that the
 552		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 553		 * to which we now hold a reference, or it is NULL.
 554		 * We can't put it here under mq_lock, though.
 555		 */
 556		if (ipc_ns)
 557			ipc_ns->mq_queues_count--;
 558		spin_unlock(&mq_lock);
 559		put_ucounts(info->ucounts);
 560		info->ucounts = NULL;
 561	}
 562	if (ipc_ns)
 563		put_ipc_ns(ipc_ns);
 564}
 565
 566static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
 567{
 568	struct inode *dir = dentry->d_parent->d_inode;
 569	struct inode *inode;
 570	struct mq_attr *attr = arg;
 571	int error;
 572	struct ipc_namespace *ipc_ns;
 573
 574	spin_lock(&mq_lock);
 575	ipc_ns = __get_ns_from_inode(dir);
 576	if (!ipc_ns) {
 577		error = -EACCES;
 578		goto out_unlock;
 579	}
 580
 581	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 582	    !capable(CAP_SYS_RESOURCE)) {
 583		error = -ENOSPC;
 584		goto out_unlock;
 585	}
 586	ipc_ns->mq_queues_count++;
 587	spin_unlock(&mq_lock);
 588
 589	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 590	if (IS_ERR(inode)) {
 591		error = PTR_ERR(inode);
 592		spin_lock(&mq_lock);
 593		ipc_ns->mq_queues_count--;
 594		goto out_unlock;
 595	}
 596
 597	put_ipc_ns(ipc_ns);
 598	dir->i_size += DIRENT_SIZE;
 599	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
 600
 601	d_instantiate(dentry, inode);
 602	dget(dentry);
 603	return 0;
 604out_unlock:
 605	spin_unlock(&mq_lock);
 606	if (ipc_ns)
 607		put_ipc_ns(ipc_ns);
 608	return error;
 609}
 610
 611static int mqueue_create(struct user_namespace *mnt_userns, struct inode *dir,
 612			 struct dentry *dentry, umode_t mode, bool excl)
 613{
 614	return mqueue_create_attr(dentry, mode, NULL);
 615}
 616
 617static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 618{
 619	struct inode *inode = d_inode(dentry);
 620
 621	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
 622	dir->i_size -= DIRENT_SIZE;
 623	drop_nlink(inode);
 624	dput(dentry);
 625	return 0;
 626}
 627
 628/*
 629*	This is routine for system read from queue file.
 630*	To avoid mess with doing here some sort of mq_receive we allow
 631*	to read only queue size & notification info (the only values
 632*	that are interesting from user point of view and aren't accessible
 633*	through std routines)
 634*/
 635static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 636				size_t count, loff_t *off)
 637{
 638	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 
 639	char buffer[FILENT_SIZE];
 640	ssize_t ret;
 641
 642	spin_lock(&info->lock);
 643	snprintf(buffer, sizeof(buffer),
 644			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 645			info->qsize,
 646			info->notify_owner ? info->notify.sigev_notify : 0,
 647			(info->notify_owner &&
 648			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
 649				info->notify.sigev_signo : 0,
 650			pid_vnr(info->notify_owner));
 651	spin_unlock(&info->lock);
 652	buffer[sizeof(buffer)-1] = '\0';
 653
 654	ret = simple_read_from_buffer(u_data, count, off, buffer,
 655				strlen(buffer));
 656	if (ret <= 0)
 657		return ret;
 658
 659	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
 660	return ret;
 661}
 662
 663static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 664{
 665	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 666
 667	spin_lock(&info->lock);
 668	if (task_tgid(current) == info->notify_owner)
 669		remove_notification(info);
 670
 671	spin_unlock(&info->lock);
 672	return 0;
 673}
 674
 675static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 676{
 677	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 678	__poll_t retval = 0;
 679
 680	poll_wait(filp, &info->wait_q, poll_tab);
 681
 682	spin_lock(&info->lock);
 683	if (info->attr.mq_curmsgs)
 684		retval = EPOLLIN | EPOLLRDNORM;
 685
 686	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 687		retval |= EPOLLOUT | EPOLLWRNORM;
 688	spin_unlock(&info->lock);
 689
 690	return retval;
 691}
 692
 693/* Adds current to info->e_wait_q[sr] before element with smaller prio */
 694static void wq_add(struct mqueue_inode_info *info, int sr,
 695			struct ext_wait_queue *ewp)
 696{
 697	struct ext_wait_queue *walk;
 698
 699	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 700		if (walk->task->prio <= current->prio) {
 701			list_add_tail(&ewp->list, &walk->list);
 702			return;
 703		}
 704	}
 705	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 706}
 707
 708/*
 709 * Puts current task to sleep. Caller must hold queue lock. After return
 710 * lock isn't held.
 711 * sr: SEND or RECV
 712 */
 713static int wq_sleep(struct mqueue_inode_info *info, int sr,
 714		    ktime_t *timeout, struct ext_wait_queue *ewp)
 715	__releases(&info->lock)
 716{
 717	int retval;
 718	signed long time;
 719
 720	wq_add(info, sr, ewp);
 721
 722	for (;;) {
 723		/* memory barrier not required, we hold info->lock */
 724		__set_current_state(TASK_INTERRUPTIBLE);
 725
 726		spin_unlock(&info->lock);
 727		time = schedule_hrtimeout_range_clock(timeout, 0,
 728			HRTIMER_MODE_ABS, CLOCK_REALTIME);
 729
 730		if (READ_ONCE(ewp->state) == STATE_READY) {
 731			/* see MQ_BARRIER for purpose/pairing */
 732			smp_acquire__after_ctrl_dep();
 733			retval = 0;
 734			goto out;
 735		}
 736		spin_lock(&info->lock);
 737
 738		/* we hold info->lock, so no memory barrier required */
 739		if (READ_ONCE(ewp->state) == STATE_READY) {
 740			retval = 0;
 741			goto out_unlock;
 742		}
 743		if (signal_pending(current)) {
 744			retval = -ERESTARTSYS;
 745			break;
 746		}
 747		if (time == 0) {
 748			retval = -ETIMEDOUT;
 749			break;
 750		}
 751	}
 752	list_del(&ewp->list);
 753out_unlock:
 754	spin_unlock(&info->lock);
 755out:
 756	return retval;
 757}
 758
 759/*
 760 * Returns waiting task that should be serviced first or NULL if none exists
 761 */
 762static struct ext_wait_queue *wq_get_first_waiter(
 763		struct mqueue_inode_info *info, int sr)
 764{
 765	struct list_head *ptr;
 766
 767	ptr = info->e_wait_q[sr].list.prev;
 768	if (ptr == &info->e_wait_q[sr].list)
 769		return NULL;
 770	return list_entry(ptr, struct ext_wait_queue, list);
 771}
 772
 773
 774static inline void set_cookie(struct sk_buff *skb, char code)
 775{
 776	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 777}
 778
 779/*
 780 * The next function is only to split too long sys_mq_timedsend
 781 */
 782static void __do_notify(struct mqueue_inode_info *info)
 783{
 784	/* notification
 785	 * invoked when there is registered process and there isn't process
 786	 * waiting synchronously for message AND state of queue changed from
 787	 * empty to not empty. Here we are sure that no one is waiting
 788	 * synchronously. */
 789	if (info->notify_owner &&
 790	    info->attr.mq_curmsgs == 1) {
 791		switch (info->notify.sigev_notify) {
 792		case SIGEV_NONE:
 793			break;
 794		case SIGEV_SIGNAL: {
 795			struct kernel_siginfo sig_i;
 796			struct task_struct *task;
 797
 798			/* do_mq_notify() accepts sigev_signo == 0, why?? */
 799			if (!info->notify.sigev_signo)
 800				break;
 801
 802			clear_siginfo(&sig_i);
 803			sig_i.si_signo = info->notify.sigev_signo;
 804			sig_i.si_errno = 0;
 805			sig_i.si_code = SI_MESGQ;
 806			sig_i.si_value = info->notify.sigev_value;
 807			rcu_read_lock();
 808			/* map current pid/uid into info->owner's namespaces */
 809			sig_i.si_pid = task_tgid_nr_ns(current,
 810						ns_of_pid(info->notify_owner));
 811			sig_i.si_uid = from_kuid_munged(info->notify_user_ns,
 812						current_uid());
 813			/*
 814			 * We can't use kill_pid_info(), this signal should
 815			 * bypass check_kill_permission(). It is from kernel
 816			 * but si_fromuser() can't know this.
 817			 * We do check the self_exec_id, to avoid sending
 818			 * signals to programs that don't expect them.
 819			 */
 820			task = pid_task(info->notify_owner, PIDTYPE_TGID);
 821			if (task && task->self_exec_id ==
 822						info->notify_self_exec_id) {
 823				do_send_sig_info(info->notify.sigev_signo,
 824						&sig_i, task, PIDTYPE_TGID);
 825			}
 826			rcu_read_unlock();
 827			break;
 828		}
 829		case SIGEV_THREAD:
 830			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 831			netlink_sendskb(info->notify_sock, info->notify_cookie);
 832			break;
 833		}
 834		/* after notification unregisters process */
 835		put_pid(info->notify_owner);
 836		put_user_ns(info->notify_user_ns);
 837		info->notify_owner = NULL;
 838		info->notify_user_ns = NULL;
 839	}
 840	wake_up(&info->wait_q);
 841}
 842
 843static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
 844			   struct timespec64 *ts)
 845{
 846	if (get_timespec64(ts, u_abs_timeout))
 847		return -EFAULT;
 848	if (!timespec64_valid(ts))
 849		return -EINVAL;
 850	return 0;
 851}
 852
 853static void remove_notification(struct mqueue_inode_info *info)
 854{
 855	if (info->notify_owner != NULL &&
 856	    info->notify.sigev_notify == SIGEV_THREAD) {
 857		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 858		netlink_sendskb(info->notify_sock, info->notify_cookie);
 859	}
 860	put_pid(info->notify_owner);
 861	put_user_ns(info->notify_user_ns);
 862	info->notify_owner = NULL;
 863	info->notify_user_ns = NULL;
 864}
 865
 866static int prepare_open(struct dentry *dentry, int oflag, int ro,
 867			umode_t mode, struct filename *name,
 868			struct mq_attr *attr)
 869{
 870	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 871						  MAY_READ | MAY_WRITE };
 872	int acc;
 873
 874	if (d_really_is_negative(dentry)) {
 875		if (!(oflag & O_CREAT))
 876			return -ENOENT;
 877		if (ro)
 878			return ro;
 879		audit_inode_parent_hidden(name, dentry->d_parent);
 880		return vfs_mkobj(dentry, mode & ~current_umask(),
 881				  mqueue_create_attr, attr);
 882	}
 883	/* it already existed */
 884	audit_inode(name, dentry, 0);
 885	if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
 886		return -EEXIST;
 887	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
 888		return -EINVAL;
 889	acc = oflag2acc[oflag & O_ACCMODE];
 890	return inode_permission(&init_user_ns, d_inode(dentry), acc);
 891}
 892
 893static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
 894		      struct mq_attr *attr)
 895{
 896	struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
 897	struct dentry *root = mnt->mnt_root;
 898	struct filename *name;
 899	struct path path;
 900	int fd, error;
 901	int ro;
 902
 903	audit_mq_open(oflag, mode, attr);
 904
 905	if (IS_ERR(name = getname(u_name)))
 
 906		return PTR_ERR(name);
 907
 908	fd = get_unused_fd_flags(O_CLOEXEC);
 909	if (fd < 0)
 910		goto out_putname;
 911
 912	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
 913	inode_lock(d_inode(root));
 914	path.dentry = lookup_one_len(name->name, root, strlen(name->name));
 915	if (IS_ERR(path.dentry)) {
 916		error = PTR_ERR(path.dentry);
 917		goto out_putfd;
 918	}
 919	path.mnt = mntget(mnt);
 920	error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
 921	if (!error) {
 922		struct file *file = dentry_open(&path, oflag, current_cred());
 923		if (!IS_ERR(file))
 924			fd_install(fd, file);
 925		else
 926			error = PTR_ERR(file);
 927	}
 928	path_put(&path);
 929out_putfd:
 930	if (error) {
 931		put_unused_fd(fd);
 932		fd = error;
 933	}
 934	inode_unlock(d_inode(root));
 935	if (!ro)
 936		mnt_drop_write(mnt);
 937out_putname:
 938	putname(name);
 939	return fd;
 940}
 941
 942SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
 943		struct mq_attr __user *, u_attr)
 944{
 945	struct mq_attr attr;
 946	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 947		return -EFAULT;
 948
 949	return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
 950}
 951
 952SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 953{
 954	int err;
 955	struct filename *name;
 956	struct dentry *dentry;
 957	struct inode *inode = NULL;
 958	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 959	struct vfsmount *mnt = ipc_ns->mq_mnt;
 960
 961	name = getname(u_name);
 962	if (IS_ERR(name))
 963		return PTR_ERR(name);
 964
 965	audit_inode_parent_hidden(name, mnt->mnt_root);
 966	err = mnt_want_write(mnt);
 967	if (err)
 968		goto out_name;
 969	inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
 970	dentry = lookup_one_len(name->name, mnt->mnt_root,
 971				strlen(name->name));
 972	if (IS_ERR(dentry)) {
 973		err = PTR_ERR(dentry);
 974		goto out_unlock;
 975	}
 976
 977	inode = d_inode(dentry);
 978	if (!inode) {
 979		err = -ENOENT;
 980	} else {
 981		ihold(inode);
 982		err = vfs_unlink(&init_user_ns, d_inode(dentry->d_parent),
 983				 dentry, NULL);
 984	}
 985	dput(dentry);
 986
 987out_unlock:
 988	inode_unlock(d_inode(mnt->mnt_root));
 989	iput(inode);
 990	mnt_drop_write(mnt);
 991out_name:
 992	putname(name);
 993
 994	return err;
 995}
 996
 997/* Pipelined send and receive functions.
 998 *
 999 * If a receiver finds no waiting message, then it registers itself in the
1000 * list of waiting receivers. A sender checks that list before adding the new
1001 * message into the message array. If there is a waiting receiver, then it
1002 * bypasses the message array and directly hands the message over to the
1003 * receiver. The receiver accepts the message and returns without grabbing the
1004 * queue spinlock:
1005 *
1006 * - Set pointer to message.
1007 * - Queue the receiver task for later wakeup (without the info->lock).
1008 * - Update its state to STATE_READY. Now the receiver can continue.
1009 * - Wake up the process after the lock is dropped. Should the process wake up
1010 *   before this wakeup (due to a timeout or a signal) it will either see
1011 *   STATE_READY and continue or acquire the lock to check the state again.
1012 *
1013 * The same algorithm is used for senders.
1014 */
1015
1016static inline void __pipelined_op(struct wake_q_head *wake_q,
1017				  struct mqueue_inode_info *info,
1018				  struct ext_wait_queue *this)
1019{
1020	struct task_struct *task;
1021
1022	list_del(&this->list);
1023	task = get_task_struct(this->task);
1024
1025	/* see MQ_BARRIER for purpose/pairing */
1026	smp_store_release(&this->state, STATE_READY);
1027	wake_q_add_safe(wake_q, task);
1028}
1029
1030/* pipelined_send() - send a message directly to the task waiting in
1031 * sys_mq_timedreceive() (without inserting message into a queue).
1032 */
1033static inline void pipelined_send(struct wake_q_head *wake_q,
1034				  struct mqueue_inode_info *info,
1035				  struct msg_msg *message,
1036				  struct ext_wait_queue *receiver)
1037{
1038	receiver->msg = message;
1039	__pipelined_op(wake_q, info, receiver);
1040}
1041
1042/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
1043 * gets its message and put to the queue (we have one free place for sure). */
1044static inline void pipelined_receive(struct wake_q_head *wake_q,
1045				     struct mqueue_inode_info *info)
1046{
1047	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
1048
1049	if (!sender) {
1050		/* for poll */
1051		wake_up_interruptible(&info->wait_q);
1052		return;
1053	}
1054	if (msg_insert(sender->msg, info))
1055		return;
1056
1057	__pipelined_op(wake_q, info, sender);
1058}
1059
1060static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
1061		size_t msg_len, unsigned int msg_prio,
1062		struct timespec64 *ts)
1063{
1064	struct fd f;
1065	struct inode *inode;
1066	struct ext_wait_queue wait;
1067	struct ext_wait_queue *receiver;
1068	struct msg_msg *msg_ptr;
1069	struct mqueue_inode_info *info;
1070	ktime_t expires, *timeout = NULL;
1071	struct posix_msg_tree_node *new_leaf = NULL;
1072	int ret = 0;
1073	DEFINE_WAKE_Q(wake_q);
1074
1075	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
1076		return -EINVAL;
1077
1078	if (ts) {
1079		expires = timespec64_to_ktime(*ts);
1080		timeout = &expires;
1081	}
1082
1083	audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
1084
1085	f = fdget(mqdes);
1086	if (unlikely(!f.file)) {
1087		ret = -EBADF;
1088		goto out;
1089	}
1090
1091	inode = file_inode(f.file);
1092	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1093		ret = -EBADF;
1094		goto out_fput;
1095	}
1096	info = MQUEUE_I(inode);
1097	audit_file(f.file);
1098
1099	if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1100		ret = -EBADF;
1101		goto out_fput;
1102	}
1103
1104	if (unlikely(msg_len > info->attr.mq_msgsize)) {
1105		ret = -EMSGSIZE;
1106		goto out_fput;
1107	}
1108
1109	/* First try to allocate memory, before doing anything with
1110	 * existing queues. */
1111	msg_ptr = load_msg(u_msg_ptr, msg_len);
1112	if (IS_ERR(msg_ptr)) {
1113		ret = PTR_ERR(msg_ptr);
1114		goto out_fput;
1115	}
1116	msg_ptr->m_ts = msg_len;
1117	msg_ptr->m_type = msg_prio;
1118
1119	/*
1120	 * msg_insert really wants us to have a valid, spare node struct so
1121	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1122	 * fall back to that if necessary.
1123	 */
1124	if (!info->node_cache)
1125		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1126
1127	spin_lock(&info->lock);
1128
1129	if (!info->node_cache && new_leaf) {
1130		/* Save our speculative allocation into the cache */
1131		INIT_LIST_HEAD(&new_leaf->msg_list);
1132		info->node_cache = new_leaf;
1133		new_leaf = NULL;
1134	} else {
1135		kfree(new_leaf);
1136	}
1137
1138	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1139		if (f.file->f_flags & O_NONBLOCK) {
1140			ret = -EAGAIN;
1141		} else {
1142			wait.task = current;
1143			wait.msg = (void *) msg_ptr;
1144
1145			/* memory barrier not required, we hold info->lock */
1146			WRITE_ONCE(wait.state, STATE_NONE);
1147			ret = wq_sleep(info, SEND, timeout, &wait);
1148			/*
1149			 * wq_sleep must be called with info->lock held, and
1150			 * returns with the lock released
1151			 */
1152			goto out_free;
1153		}
1154	} else {
1155		receiver = wq_get_first_waiter(info, RECV);
1156		if (receiver) {
1157			pipelined_send(&wake_q, info, msg_ptr, receiver);
1158		} else {
1159			/* adds message to the queue */
1160			ret = msg_insert(msg_ptr, info);
1161			if (ret)
1162				goto out_unlock;
1163			__do_notify(info);
1164		}
1165		inode->i_atime = inode->i_mtime = inode->i_ctime =
1166				current_time(inode);
1167	}
1168out_unlock:
1169	spin_unlock(&info->lock);
1170	wake_up_q(&wake_q);
1171out_free:
1172	if (ret)
1173		free_msg(msg_ptr);
1174out_fput:
1175	fdput(f);
1176out:
1177	return ret;
1178}
1179
1180static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1181		size_t msg_len, unsigned int __user *u_msg_prio,
1182		struct timespec64 *ts)
1183{
1184	ssize_t ret;
1185	struct msg_msg *msg_ptr;
1186	struct fd f;
1187	struct inode *inode;
1188	struct mqueue_inode_info *info;
1189	struct ext_wait_queue wait;
1190	ktime_t expires, *timeout = NULL;
1191	struct posix_msg_tree_node *new_leaf = NULL;
1192
1193	if (ts) {
1194		expires = timespec64_to_ktime(*ts);
1195		timeout = &expires;
1196	}
1197
1198	audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1199
1200	f = fdget(mqdes);
1201	if (unlikely(!f.file)) {
1202		ret = -EBADF;
1203		goto out;
1204	}
1205
1206	inode = file_inode(f.file);
1207	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1208		ret = -EBADF;
1209		goto out_fput;
1210	}
1211	info = MQUEUE_I(inode);
1212	audit_file(f.file);
1213
1214	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1215		ret = -EBADF;
1216		goto out_fput;
1217	}
1218
1219	/* checks if buffer is big enough */
1220	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1221		ret = -EMSGSIZE;
1222		goto out_fput;
1223	}
1224
1225	/*
1226	 * msg_insert really wants us to have a valid, spare node struct so
1227	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1228	 * fall back to that if necessary.
1229	 */
1230	if (!info->node_cache)
1231		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1232
1233	spin_lock(&info->lock);
1234
1235	if (!info->node_cache && new_leaf) {
1236		/* Save our speculative allocation into the cache */
1237		INIT_LIST_HEAD(&new_leaf->msg_list);
1238		info->node_cache = new_leaf;
1239	} else {
1240		kfree(new_leaf);
1241	}
1242
1243	if (info->attr.mq_curmsgs == 0) {
1244		if (f.file->f_flags & O_NONBLOCK) {
1245			spin_unlock(&info->lock);
1246			ret = -EAGAIN;
1247		} else {
1248			wait.task = current;
1249
1250			/* memory barrier not required, we hold info->lock */
1251			WRITE_ONCE(wait.state, STATE_NONE);
1252			ret = wq_sleep(info, RECV, timeout, &wait);
1253			msg_ptr = wait.msg;
1254		}
1255	} else {
1256		DEFINE_WAKE_Q(wake_q);
1257
1258		msg_ptr = msg_get(info);
1259
1260		inode->i_atime = inode->i_mtime = inode->i_ctime =
1261				current_time(inode);
1262
1263		/* There is now free space in queue. */
1264		pipelined_receive(&wake_q, info);
1265		spin_unlock(&info->lock);
1266		wake_up_q(&wake_q);
1267		ret = 0;
1268	}
1269	if (ret == 0) {
1270		ret = msg_ptr->m_ts;
1271
1272		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1273			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1274			ret = -EFAULT;
1275		}
1276		free_msg(msg_ptr);
1277	}
1278out_fput:
1279	fdput(f);
1280out:
1281	return ret;
1282}
1283
1284SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1285		size_t, msg_len, unsigned int, msg_prio,
1286		const struct __kernel_timespec __user *, u_abs_timeout)
1287{
1288	struct timespec64 ts, *p = NULL;
1289	if (u_abs_timeout) {
1290		int res = prepare_timeout(u_abs_timeout, &ts);
1291		if (res)
1292			return res;
1293		p = &ts;
1294	}
1295	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1296}
1297
1298SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1299		size_t, msg_len, unsigned int __user *, u_msg_prio,
1300		const struct __kernel_timespec __user *, u_abs_timeout)
1301{
1302	struct timespec64 ts, *p = NULL;
1303	if (u_abs_timeout) {
1304		int res = prepare_timeout(u_abs_timeout, &ts);
1305		if (res)
1306			return res;
1307		p = &ts;
1308	}
1309	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1310}
1311
1312/*
1313 * Notes: the case when user wants us to deregister (with NULL as pointer)
1314 * and he isn't currently owner of notification, will be silently discarded.
1315 * It isn't explicitly defined in the POSIX.
1316 */
1317static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
1318{
1319	int ret;
1320	struct fd f;
1321	struct sock *sock;
1322	struct inode *inode;
1323	struct mqueue_inode_info *info;
1324	struct sk_buff *nc;
1325
1326	audit_mq_notify(mqdes, notification);
1327
1328	nc = NULL;
1329	sock = NULL;
1330	if (notification != NULL) {
1331		if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1332			     notification->sigev_notify != SIGEV_SIGNAL &&
1333			     notification->sigev_notify != SIGEV_THREAD))
1334			return -EINVAL;
1335		if (notification->sigev_notify == SIGEV_SIGNAL &&
1336			!valid_signal(notification->sigev_signo)) {
1337			return -EINVAL;
1338		}
1339		if (notification->sigev_notify == SIGEV_THREAD) {
1340			long timeo;
1341
1342			/* create the notify skb */
1343			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1344			if (!nc)
1345				return -ENOMEM;
1346
1347			if (copy_from_user(nc->data,
1348					notification->sigev_value.sival_ptr,
1349					NOTIFY_COOKIE_LEN)) {
1350				ret = -EFAULT;
1351				goto free_skb;
1352			}
1353
1354			/* TODO: add a header? */
1355			skb_put(nc, NOTIFY_COOKIE_LEN);
1356			/* and attach it to the socket */
1357retry:
1358			f = fdget(notification->sigev_signo);
1359			if (!f.file) {
1360				ret = -EBADF;
1361				goto out;
1362			}
1363			sock = netlink_getsockbyfilp(f.file);
1364			fdput(f);
1365			if (IS_ERR(sock)) {
1366				ret = PTR_ERR(sock);
1367				goto free_skb;
1368			}
1369
1370			timeo = MAX_SCHEDULE_TIMEOUT;
1371			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1372			if (ret == 1) {
1373				sock = NULL;
1374				goto retry;
1375			}
1376			if (ret)
1377				return ret;
1378		}
1379	}
1380
1381	f = fdget(mqdes);
1382	if (!f.file) {
1383		ret = -EBADF;
1384		goto out;
1385	}
1386
1387	inode = file_inode(f.file);
1388	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1389		ret = -EBADF;
1390		goto out_fput;
1391	}
1392	info = MQUEUE_I(inode);
1393
1394	ret = 0;
1395	spin_lock(&info->lock);
1396	if (notification == NULL) {
1397		if (info->notify_owner == task_tgid(current)) {
1398			remove_notification(info);
1399			inode->i_atime = inode->i_ctime = current_time(inode);
 
1400		}
1401	} else if (info->notify_owner != NULL) {
1402		ret = -EBUSY;
1403	} else {
1404		switch (notification->sigev_notify) {
1405		case SIGEV_NONE:
1406			info->notify.sigev_notify = SIGEV_NONE;
1407			break;
1408		case SIGEV_THREAD:
1409			info->notify_sock = sock;
1410			info->notify_cookie = nc;
1411			sock = NULL;
1412			nc = NULL;
1413			info->notify.sigev_notify = SIGEV_THREAD;
1414			break;
1415		case SIGEV_SIGNAL:
1416			info->notify.sigev_signo = notification->sigev_signo;
1417			info->notify.sigev_value = notification->sigev_value;
1418			info->notify.sigev_notify = SIGEV_SIGNAL;
1419			info->notify_self_exec_id = current->self_exec_id;
1420			break;
1421		}
1422
1423		info->notify_owner = get_pid(task_tgid(current));
1424		info->notify_user_ns = get_user_ns(current_user_ns());
1425		inode->i_atime = inode->i_ctime = current_time(inode);
1426	}
1427	spin_unlock(&info->lock);
1428out_fput:
1429	fdput(f);
1430out:
1431	if (sock)
1432		netlink_detachskb(sock, nc);
1433	else
1434free_skb:
1435		dev_kfree_skb(nc);
1436
1437	return ret;
1438}
1439
1440SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1441		const struct sigevent __user *, u_notification)
1442{
1443	struct sigevent n, *p = NULL;
1444	if (u_notification) {
1445		if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1446			return -EFAULT;
1447		p = &n;
1448	}
1449	return do_mq_notify(mqdes, p);
1450}
1451
1452static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1453{
1454	struct fd f;
1455	struct inode *inode;
1456	struct mqueue_inode_info *info;
1457
1458	if (new && (new->mq_flags & (~O_NONBLOCK)))
1459		return -EINVAL;
1460
1461	f = fdget(mqdes);
1462	if (!f.file)
1463		return -EBADF;
1464
1465	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1466		fdput(f);
1467		return -EBADF;
1468	}
1469
1470	inode = file_inode(f.file);
1471	info = MQUEUE_I(inode);
1472
1473	spin_lock(&info->lock);
1474
1475	if (old) {
1476		*old = info->attr;
1477		old->mq_flags = f.file->f_flags & O_NONBLOCK;
1478	}
1479	if (new) {
1480		audit_mq_getsetattr(mqdes, new);
1481		spin_lock(&f.file->f_lock);
1482		if (new->mq_flags & O_NONBLOCK)
1483			f.file->f_flags |= O_NONBLOCK;
1484		else
1485			f.file->f_flags &= ~O_NONBLOCK;
1486		spin_unlock(&f.file->f_lock);
1487
1488		inode->i_atime = inode->i_ctime = current_time(inode);
1489	}
1490
1491	spin_unlock(&info->lock);
1492	fdput(f);
1493	return 0;
1494}
1495
1496SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1497		const struct mq_attr __user *, u_mqstat,
1498		struct mq_attr __user *, u_omqstat)
1499{
1500	int ret;
1501	struct mq_attr mqstat, omqstat;
1502	struct mq_attr *new = NULL, *old = NULL;
1503
1504	if (u_mqstat) {
1505		new = &mqstat;
1506		if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1507			return -EFAULT;
1508	}
1509	if (u_omqstat)
1510		old = &omqstat;
1511
1512	ret = do_mq_getsetattr(mqdes, new, old);
1513	if (ret || !old)
1514		return ret;
1515
1516	if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1517		return -EFAULT;
1518	return 0;
1519}
1520
1521#ifdef CONFIG_COMPAT
1522
1523struct compat_mq_attr {
1524	compat_long_t mq_flags;      /* message queue flags		     */
1525	compat_long_t mq_maxmsg;     /* maximum number of messages	     */
1526	compat_long_t mq_msgsize;    /* maximum message size		     */
1527	compat_long_t mq_curmsgs;    /* number of messages currently queued  */
1528	compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1529};
1530
1531static inline int get_compat_mq_attr(struct mq_attr *attr,
1532			const struct compat_mq_attr __user *uattr)
1533{
1534	struct compat_mq_attr v;
1535
1536	if (copy_from_user(&v, uattr, sizeof(*uattr)))
1537		return -EFAULT;
1538
1539	memset(attr, 0, sizeof(*attr));
1540	attr->mq_flags = v.mq_flags;
1541	attr->mq_maxmsg = v.mq_maxmsg;
1542	attr->mq_msgsize = v.mq_msgsize;
1543	attr->mq_curmsgs = v.mq_curmsgs;
1544	return 0;
1545}
1546
1547static inline int put_compat_mq_attr(const struct mq_attr *attr,
1548			struct compat_mq_attr __user *uattr)
1549{
1550	struct compat_mq_attr v;
1551
1552	memset(&v, 0, sizeof(v));
1553	v.mq_flags = attr->mq_flags;
1554	v.mq_maxmsg = attr->mq_maxmsg;
1555	v.mq_msgsize = attr->mq_msgsize;
1556	v.mq_curmsgs = attr->mq_curmsgs;
1557	if (copy_to_user(uattr, &v, sizeof(*uattr)))
1558		return -EFAULT;
1559	return 0;
1560}
1561
1562COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1563		       int, oflag, compat_mode_t, mode,
1564		       struct compat_mq_attr __user *, u_attr)
1565{
1566	struct mq_attr attr, *p = NULL;
1567	if (u_attr && oflag & O_CREAT) {
1568		p = &attr;
1569		if (get_compat_mq_attr(&attr, u_attr))
1570			return -EFAULT;
1571	}
1572	return do_mq_open(u_name, oflag, mode, p);
1573}
1574
1575COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1576		       const struct compat_sigevent __user *, u_notification)
1577{
1578	struct sigevent n, *p = NULL;
1579	if (u_notification) {
1580		if (get_compat_sigevent(&n, u_notification))
1581			return -EFAULT;
1582		if (n.sigev_notify == SIGEV_THREAD)
1583			n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1584		p = &n;
1585	}
1586	return do_mq_notify(mqdes, p);
1587}
1588
1589COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1590		       const struct compat_mq_attr __user *, u_mqstat,
1591		       struct compat_mq_attr __user *, u_omqstat)
1592{
1593	int ret;
1594	struct mq_attr mqstat, omqstat;
1595	struct mq_attr *new = NULL, *old = NULL;
1596
1597	if (u_mqstat) {
1598		new = &mqstat;
1599		if (get_compat_mq_attr(new, u_mqstat))
1600			return -EFAULT;
1601	}
1602	if (u_omqstat)
1603		old = &omqstat;
1604
1605	ret = do_mq_getsetattr(mqdes, new, old);
1606	if (ret || !old)
1607		return ret;
1608
1609	if (put_compat_mq_attr(old, u_omqstat))
1610		return -EFAULT;
1611	return 0;
1612}
1613#endif
1614
1615#ifdef CONFIG_COMPAT_32BIT_TIME
1616static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1617				   struct timespec64 *ts)
1618{
1619	if (get_old_timespec32(ts, p))
1620		return -EFAULT;
1621	if (!timespec64_valid(ts))
1622		return -EINVAL;
1623	return 0;
1624}
1625
1626SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1627		const char __user *, u_msg_ptr,
1628		unsigned int, msg_len, unsigned int, msg_prio,
1629		const struct old_timespec32 __user *, u_abs_timeout)
1630{
1631	struct timespec64 ts, *p = NULL;
1632	if (u_abs_timeout) {
1633		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1634		if (res)
1635			return res;
1636		p = &ts;
1637	}
1638	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1639}
1640
1641SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1642		char __user *, u_msg_ptr,
1643		unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1644		const struct old_timespec32 __user *, u_abs_timeout)
1645{
1646	struct timespec64 ts, *p = NULL;
1647	if (u_abs_timeout) {
1648		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1649		if (res)
1650			return res;
1651		p = &ts;
1652	}
1653	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1654}
1655#endif
1656
1657static const struct inode_operations mqueue_dir_inode_operations = {
1658	.lookup = simple_lookup,
1659	.create = mqueue_create,
1660	.unlink = mqueue_unlink,
1661};
1662
1663static const struct file_operations mqueue_file_operations = {
1664	.flush = mqueue_flush_file,
1665	.poll = mqueue_poll_file,
1666	.read = mqueue_read_file,
1667	.llseek = default_llseek,
1668};
1669
1670static const struct super_operations mqueue_super_ops = {
1671	.alloc_inode = mqueue_alloc_inode,
1672	.free_inode = mqueue_free_inode,
1673	.evict_inode = mqueue_evict_inode,
1674	.statfs = simple_statfs,
1675};
1676
1677static const struct fs_context_operations mqueue_fs_context_ops = {
1678	.free		= mqueue_fs_context_free,
1679	.get_tree	= mqueue_get_tree,
1680};
1681
1682static struct file_system_type mqueue_fs_type = {
1683	.name			= "mqueue",
1684	.init_fs_context	= mqueue_init_fs_context,
1685	.kill_sb		= kill_litter_super,
1686	.fs_flags		= FS_USERNS_MOUNT,
1687};
1688
1689int mq_init_ns(struct ipc_namespace *ns)
1690{
1691	struct vfsmount *m;
1692
1693	ns->mq_queues_count  = 0;
1694	ns->mq_queues_max    = DFLT_QUEUESMAX;
1695	ns->mq_msg_max       = DFLT_MSGMAX;
1696	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1697	ns->mq_msg_default   = DFLT_MSG;
1698	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1699
1700	m = mq_create_mount(ns);
1701	if (IS_ERR(m))
1702		return PTR_ERR(m);
1703	ns->mq_mnt = m;
1704	return 0;
1705}
1706
1707void mq_clear_sbinfo(struct ipc_namespace *ns)
1708{
1709	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1710}
1711
1712void mq_put_mnt(struct ipc_namespace *ns)
1713{
1714	kern_unmount(ns->mq_mnt);
1715}
1716
1717static int __init init_mqueue_fs(void)
1718{
1719	int error;
1720
1721	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1722				sizeof(struct mqueue_inode_info), 0,
1723				SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1724	if (mqueue_inode_cachep == NULL)
1725		return -ENOMEM;
1726
1727	if (!setup_mq_sysctls(&init_ipc_ns)) {
1728		pr_warn("sysctl registration failed\n");
1729		error = -ENOMEM;
1730		goto out_kmem;
1731	}
1732
1733	error = register_filesystem(&mqueue_fs_type);
1734	if (error)
1735		goto out_sysctl;
1736
1737	spin_lock_init(&mq_lock);
1738
1739	error = mq_init_ns(&init_ipc_ns);
1740	if (error)
1741		goto out_filesystem;
1742
1743	return 0;
1744
1745out_filesystem:
1746	unregister_filesystem(&mqueue_fs_type);
1747out_sysctl:
1748	retire_mq_sysctls(&init_ipc_ns);
1749out_kmem:
1750	kmem_cache_destroy(mqueue_inode_cachep);
1751	return error;
1752}
1753
1754device_initcall(init_mqueue_fs);