| 1 | |
| 2 | * POSIX message queues filesystem for Linux.
|
| 3 | *
|
| 4 | * Copyright (C) 2003,2004 Krzysztof Benedyczak ([email protected])
|
| 5 | * Michal Wronski ([email protected])
|
| 6 | *
|
| 7 | * Spinlocks: Mohamed Abbas ([email protected])
|
| 8 | * Lockless receive & send, fd based notify:
|
| 9 | * Manfred Spraul ([email protected])
|
| 10 | *
|
| 11 | * Audit: George Wilson ([email protected])
|
| 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 |
|
| 46 | struct mqueue_fs_context {
|
| 47 | struct ipc_namespace *ipc_ns;
|
| 48 | bool newns;
|
| 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 |
|
| 61 | struct 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 |
|
| 127 | struct ext_wait_queue {
|
| 128 | struct task_struct *task;
|
| 129 | struct list_head list;
|
| 130 | struct msg_msg *msg;
|
| 131 | int state;
|
| 132 | };
|
| 133 |
|
| 134 | struct 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;
|
| 149 | struct sock *notify_sock;
|
| 150 | struct sk_buff *notify_cookie;
|
| 151 |
|
| 152 |
|
| 153 | struct ext_wait_queue e_wait_q[2];
|
| 154 |
|
| 155 | unsigned long qsize;
|
| 156 | };
|
| 157 |
|
| 158 | static struct file_system_type mqueue_fs_type;
|
| 159 | static const struct inode_operations mqueue_dir_inode_operations;
|
| 160 | static const struct file_operations mqueue_file_operations;
|
| 161 | static const struct super_operations mqueue_super_ops;
|
| 162 | static const struct fs_context_operations mqueue_fs_context_ops;
|
| 163 | static void remove_notification(struct mqueue_inode_info *info);
|
| 164 |
|
| 165 | static struct kmem_cache *mqueue_inode_cachep;
|
| 166 |
|
| 167 | static 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 |
|
| 175 | static 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 |
|
| 180 | static 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 |
|
| 191 | static 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);
|
| 226 | insert_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 |
|
| 233 | static 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 |
|
| 248 | static 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 |
|
| 254 | try_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 |
|
| 290 | static 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_atime = inode_set_ctime_current(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 |
|
| 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;
|
| 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 |
|
| 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 |
|
| 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 |
|
| 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;
|
| 398 | out_inode:
|
| 399 | iput(inode);
|
| 400 | err:
|
| 401 | return ERR_PTR(ret);
|
| 402 | }
|
| 403 |
|
| 404 | static 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 |
|
| 425 | static 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 |
|
| 440 | static 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 |
|
| 448 | static 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 |
|
| 468 | static 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 |
|
| 490 | static void init_once(void *foo)
|
| 491 | {
|
| 492 | struct mqueue_inode_info *p = foo;
|
| 493 |
|
| 494 | inode_init_once(&p->vfs_inode);
|
| 495 | }
|
| 496 |
|
| 497 | static 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 |
|
| 507 | static void mqueue_free_inode(struct inode *inode)
|
| 508 | {
|
| 509 | kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
|
| 510 | }
|
| 511 |
|
| 512 | static 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 |
|
| 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 |
|
| 566 | static 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_mtime = dir->i_atime = inode_set_ctime_current(dir);
|
| 600 |
|
| 601 | d_instantiate(dentry, inode);
|
| 602 | dget(dentry);
|
| 603 | return 0;
|
| 604 | out_unlock:
|
| 605 | spin_unlock(&mq_lock);
|
| 606 | if (ipc_ns)
|
| 607 | put_ipc_ns(ipc_ns);
|
| 608 | return error;
|
| 609 | }
|
| 610 |
|
| 611 | static int mqueue_create(struct mnt_idmap *idmap, struct inode *dir,
|
| 612 | struct dentry *dentry, umode_t mode, bool excl)
|
| 613 | {
|
| 614 | return mqueue_create_attr(dentry, mode, NULL);
|
| 615 | }
|
| 616 |
|
| 617 | static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
|
| 618 | {
|
| 619 | struct inode *inode = d_inode(dentry);
|
| 620 |
|
| 621 | dir->i_mtime = dir->i_atime = inode_set_ctime_current(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 |
|
| 635 | static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
|
| 636 | size_t count, loff_t *off)
|
| 637 | {
|
| 638 | struct inode *inode = file_inode(filp);
|
| 639 | struct mqueue_inode_info *info = MQUEUE_I(inode);
|
| 640 | char buffer[FILENT_SIZE];
|
| 641 | ssize_t ret;
|
| 642 |
|
| 643 | spin_lock(&info->lock);
|
| 644 | snprintf(buffer, sizeof(buffer),
|
| 645 | "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
|
| 646 | info->qsize,
|
| 647 | info->notify_owner ? info->notify.sigev_notify : 0,
|
| 648 | (info->notify_owner &&
|
| 649 | info->notify.sigev_notify == SIGEV_SIGNAL) ?
|
| 650 | info->notify.sigev_signo : 0,
|
| 651 | pid_vnr(info->notify_owner));
|
| 652 | spin_unlock(&info->lock);
|
| 653 | buffer[sizeof(buffer)-1] = '\0';
|
| 654 |
|
| 655 | ret = simple_read_from_buffer(u_data, count, off, buffer,
|
| 656 | strlen(buffer));
|
| 657 | if (ret <= 0)
|
| 658 | return ret;
|
| 659 |
|
| 660 | inode->i_atime = inode_set_ctime_current(inode);
|
| 661 | return ret;
|
| 662 | }
|
| 663 |
|
| 664 | static int mqueue_flush_file(struct file *filp, fl_owner_t id)
|
| 665 | {
|
| 666 | struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
|
| 667 |
|
| 668 | spin_lock(&info->lock);
|
| 669 | if (task_tgid(current) == info->notify_owner)
|
| 670 | remove_notification(info);
|
| 671 |
|
| 672 | spin_unlock(&info->lock);
|
| 673 | return 0;
|
| 674 | }
|
| 675 |
|
| 676 | static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
|
| 677 | {
|
| 678 | struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
|
| 679 | __poll_t retval = 0;
|
| 680 |
|
| 681 | poll_wait(filp, &info->wait_q, poll_tab);
|
| 682 |
|
| 683 | spin_lock(&info->lock);
|
| 684 | if (info->attr.mq_curmsgs)
|
| 685 | retval = EPOLLIN | EPOLLRDNORM;
|
| 686 |
|
| 687 | if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
|
| 688 | retval |= EPOLLOUT | EPOLLWRNORM;
|
| 689 | spin_unlock(&info->lock);
|
| 690 |
|
| 691 | return retval;
|
| 692 | }
|
| 693 |
|
| 694 |
|
| 695 | static void wq_add(struct mqueue_inode_info *info, int sr,
|
| 696 | struct ext_wait_queue *ewp)
|
| 697 | {
|
| 698 | struct ext_wait_queue *walk;
|
| 699 |
|
| 700 | list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
|
| 701 | if (walk->task->prio <= current->prio) {
|
| 702 | list_add_tail(&ewp->list, &walk->list);
|
| 703 | return;
|
| 704 | }
|
| 705 | }
|
| 706 | list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
|
| 707 | }
|
| 708 |
|
| 709 | |
| 710 | * Puts current task to sleep. Caller must hold queue lock. After return
|
| 711 | * lock isn't held.
|
| 712 | * sr: SEND or RECV
|
| 713 |
|
| 714 | static int wq_sleep(struct mqueue_inode_info *info, int sr,
|
| 715 | ktime_t *timeout, struct ext_wait_queue *ewp)
|
| 716 | __releases(&info->lock)
|
| 717 | {
|
| 718 | int retval;
|
| 719 | signed long time;
|
| 720 |
|
| 721 | wq_add(info, sr, ewp);
|
| 722 |
|
| 723 | for (;;) {
|
| 724 |
|
| 725 | __set_current_state(TASK_INTERRUPTIBLE);
|
| 726 |
|
| 727 | spin_unlock(&info->lock);
|
| 728 | time = schedule_hrtimeout_range_clock(timeout, 0,
|
| 729 | HRTIMER_MODE_ABS, CLOCK_REALTIME);
|
| 730 |
|
| 731 | if (READ_ONCE(ewp->state) == STATE_READY) {
|
| 732 |
|
| 733 | smp_acquire__after_ctrl_dep();
|
| 734 | retval = 0;
|
| 735 | goto out;
|
| 736 | }
|
| 737 | spin_lock(&info->lock);
|
| 738 |
|
| 739 |
|
| 740 | if (READ_ONCE(ewp->state) == STATE_READY) {
|
| 741 | retval = 0;
|
| 742 | goto out_unlock;
|
| 743 | }
|
| 744 | if (signal_pending(current)) {
|
| 745 | retval = -ERESTARTSYS;
|
| 746 | break;
|
| 747 | }
|
| 748 | if (time == 0) {
|
| 749 | retval = -ETIMEDOUT;
|
| 750 | break;
|
| 751 | }
|
| 752 | }
|
| 753 | list_del(&ewp->list);
|
| 754 | out_unlock:
|
| 755 | spin_unlock(&info->lock);
|
| 756 | out:
|
| 757 | return retval;
|
| 758 | }
|
| 759 |
|
| 760 | |
| 761 | * Returns waiting task that should be serviced first or NULL if none exists
|
| 762 |
|
| 763 | static struct ext_wait_queue *wq_get_first_waiter(
|
| 764 | struct mqueue_inode_info *info, int sr)
|
| 765 | {
|
| 766 | struct list_head *ptr;
|
| 767 |
|
| 768 | ptr = info->e_wait_q[sr].list.prev;
|
| 769 | if (ptr == &info->e_wait_q[sr].list)
|
| 770 | return NULL;
|
| 771 | return list_entry(ptr, struct ext_wait_queue, list);
|
| 772 | }
|
| 773 |
|
| 774 |
|
| 775 | static inline void set_cookie(struct sk_buff *skb, char code)
|
| 776 | {
|
| 777 | ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
|
| 778 | }
|
| 779 |
|
| 780 | |
| 781 | * The next function is only to split too long sys_mq_timedsend
|
| 782 |
|
| 783 | static void __do_notify(struct mqueue_inode_info *info)
|
| 784 | {
|
| 785 | |
| 786 | * invoked when there is registered process and there isn't process
|
| 787 | * waiting synchronously for message AND state of queue changed from
|
| 788 | * empty to not empty. Here we are sure that no one is waiting
|
| 789 | * synchronously.
|
| 790 | if (info->notify_owner &&
|
| 791 | info->attr.mq_curmsgs == 1) {
|
| 792 | switch (info->notify.sigev_notify) {
|
| 793 | case SIGEV_NONE:
|
| 794 | break;
|
| 795 | case SIGEV_SIGNAL: {
|
| 796 | struct kernel_siginfo sig_i;
|
| 797 | struct task_struct *task;
|
| 798 |
|
| 799 |
|
| 800 | if (!info->notify.sigev_signo)
|
| 801 | break;
|
| 802 |
|
| 803 | clear_siginfo(&sig_i);
|
| 804 | sig_i.si_signo = info->notify.sigev_signo;
|
| 805 | sig_i.si_errno = 0;
|
| 806 | sig_i.si_code = SI_MESGQ;
|
| 807 | sig_i.si_value = info->notify.sigev_value;
|
| 808 | rcu_read_lock();
|
| 809 |
|
| 810 | sig_i.si_pid = task_tgid_nr_ns(current,
|
| 811 | ns_of_pid(info->notify_owner));
|
| 812 | sig_i.si_uid = from_kuid_munged(info->notify_user_ns,
|
| 813 | current_uid());
|
| 814 | |
| 815 | * We can't use kill_pid_info(), this signal should
|
| 816 | * bypass check_kill_permission(). It is from kernel
|
| 817 | * but si_fromuser() can't know this.
|
| 818 | * We do check the self_exec_id, to avoid sending
|
| 819 | * signals to programs that don't expect them.
|
| 820 |
|
| 821 | task = pid_task(info->notify_owner, PIDTYPE_TGID);
|
| 822 | if (task && task->self_exec_id ==
|
| 823 | info->notify_self_exec_id) {
|
| 824 | do_send_sig_info(info->notify.sigev_signo,
|
| 825 | &sig_i, task, PIDTYPE_TGID);
|
| 826 | }
|
| 827 | rcu_read_unlock();
|
| 828 | break;
|
| 829 | }
|
| 830 | case SIGEV_THREAD:
|
| 831 | set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
|
| 832 | netlink_sendskb(info->notify_sock, info->notify_cookie);
|
| 833 | break;
|
| 834 | }
|
| 835 |
|
| 836 | put_pid(info->notify_owner);
|
| 837 | put_user_ns(info->notify_user_ns);
|
| 838 | info->notify_owner = NULL;
|
| 839 | info->notify_user_ns = NULL;
|
| 840 | }
|
| 841 | wake_up(&info->wait_q);
|
| 842 | }
|
| 843 |
|
| 844 | static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
|
| 845 | struct timespec64 *ts)
|
| 846 | {
|
| 847 | if (get_timespec64(ts, u_abs_timeout))
|
| 848 | return -EFAULT;
|
| 849 | if (!timespec64_valid(ts))
|
| 850 | return -EINVAL;
|
| 851 | return 0;
|
| 852 | }
|
| 853 |
|
| 854 | static void remove_notification(struct mqueue_inode_info *info)
|
| 855 | {
|
| 856 | if (info->notify_owner != NULL &&
|
| 857 | info->notify.sigev_notify == SIGEV_THREAD) {
|
| 858 | set_cookie(info->notify_cookie, NOTIFY_REMOVED);
|
| 859 | netlink_sendskb(info->notify_sock, info->notify_cookie);
|
| 860 | }
|
| 861 | put_pid(info->notify_owner);
|
| 862 | put_user_ns(info->notify_user_ns);
|
| 863 | info->notify_owner = NULL;
|
| 864 | info->notify_user_ns = NULL;
|
| 865 | }
|
| 866 |
|
| 867 | static int prepare_open(struct dentry *dentry, int oflag, int ro,
|
| 868 | umode_t mode, struct filename *name,
|
| 869 | struct mq_attr *attr)
|
| 870 | {
|
| 871 | static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
|
| 872 | MAY_READ | MAY_WRITE };
|
| 873 | int acc;
|
| 874 |
|
| 875 | if (d_really_is_negative(dentry)) {
|
| 876 | if (!(oflag & O_CREAT))
|
| 877 | return -ENOENT;
|
| 878 | if (ro)
|
| 879 | return ro;
|
| 880 | audit_inode_parent_hidden(name, dentry->d_parent);
|
| 881 | return vfs_mkobj(dentry, mode & ~current_umask(),
|
| 882 | mqueue_create_attr, attr);
|
| 883 | }
|
| 884 |
|
| 885 | audit_inode(name, dentry, 0);
|
| 886 | if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
|
| 887 | return -EEXIST;
|
| 888 | if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
|
| 889 | return -EINVAL;
|
| 890 | acc = oflag2acc[oflag & O_ACCMODE];
|
| 891 | return inode_permission(&nop_mnt_idmap, d_inode(dentry), acc);
|
| 892 | }
|
| 893 |
|
| 894 | static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
|
| 895 | struct mq_attr *attr)
|
| 896 | {
|
| 897 | struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
|
| 898 | struct dentry *root = mnt->mnt_root;
|
| 899 | struct filename *name;
|
| 900 | struct path path;
|
| 901 | int fd, error;
|
| 902 | int ro;
|
| 903 |
|
| 904 | audit_mq_open(oflag, mode, attr);
|
| 905 |
|
| 906 | if (IS_ERR(name = getname(u_name)))
|
| 907 | return PTR_ERR(name);
|
| 908 |
|
| 909 | fd = get_unused_fd_flags(O_CLOEXEC);
|
| 910 | if (fd < 0)
|
| 911 | goto out_putname;
|
| 912 |
|
| 913 | ro = mnt_want_write(mnt);
|
| 914 | inode_lock(d_inode(root));
|
| 915 | path.dentry = lookup_one_len(name->name, root, strlen(name->name));
|
| 916 | if (IS_ERR(path.dentry)) {
|
| 917 | error = PTR_ERR(path.dentry);
|
| 918 | goto out_putfd;
|
| 919 | }
|
| 920 | path.mnt = mntget(mnt);
|
| 921 | error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
|
| 922 | if (!error) {
|
| 923 | struct file *file = dentry_open(&path, oflag, current_cred());
|
| 924 | if (!IS_ERR(file))
|
| 925 | fd_install(fd, file);
|
| 926 | else
|
| 927 | error = PTR_ERR(file);
|
| 928 | }
|
| 929 | path_put(&path);
|
| 930 | out_putfd:
|
| 931 | if (error) {
|
| 932 | put_unused_fd(fd);
|
| 933 | fd = error;
|
| 934 | }
|
| 935 | inode_unlock(d_inode(root));
|
| 936 | if (!ro)
|
| 937 | mnt_drop_write(mnt);
|
| 938 | out_putname:
|
| 939 | putname(name);
|
| 940 | return fd;
|
| 941 | }
|
| 942 |
|
| 943 | SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
|
| 944 | struct mq_attr __user *, u_attr)
|
| 945 | {
|
| 946 | struct mq_attr attr;
|
| 947 | if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
|
| 948 | return -EFAULT;
|
| 949 |
|
| 950 | return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
|
| 951 | }
|
| 952 |
|
| 953 | SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
|
| 954 | {
|
| 955 | int err;
|
| 956 | struct filename *name;
|
| 957 | struct dentry *dentry;
|
| 958 | struct inode *inode = NULL;
|
| 959 | struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
|
| 960 | struct vfsmount *mnt = ipc_ns->mq_mnt;
|
| 961 |
|
| 962 | name = getname(u_name);
|
| 963 | if (IS_ERR(name))
|
| 964 | return PTR_ERR(name);
|
| 965 |
|
| 966 | audit_inode_parent_hidden(name, mnt->mnt_root);
|
| 967 | err = mnt_want_write(mnt);
|
| 968 | if (err)
|
| 969 | goto out_name;
|
| 970 | inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
|
| 971 | dentry = lookup_one_len(name->name, mnt->mnt_root,
|
| 972 | strlen(name->name));
|
| 973 | if (IS_ERR(dentry)) {
|
| 974 | err = PTR_ERR(dentry);
|
| 975 | goto out_unlock;
|
| 976 | }
|
| 977 |
|
| 978 | inode = d_inode(dentry);
|
| 979 | if (!inode) {
|
| 980 | err = -ENOENT;
|
| 981 | } else {
|
| 982 | ihold(inode);
|
| 983 | err = vfs_unlink(&nop_mnt_idmap, d_inode(dentry->d_parent),
|
| 984 | dentry, NULL);
|
| 985 | }
|
| 986 | dput(dentry);
|
| 987 |
|
| 988 | out_unlock:
|
| 989 | inode_unlock(d_inode(mnt->mnt_root));
|
| 990 | iput(inode);
|
| 991 | mnt_drop_write(mnt);
|
| 992 | out_name:
|
| 993 | putname(name);
|
| 994 |
|
| 995 | return err;
|
| 996 | }
|
| 997 |
|
| 998 | |
| 999 | *
|
| 1000 | * If a receiver finds no waiting message, then it registers itself in the
|
| 1001 | * list of waiting receivers. A sender checks that list before adding the new
|
| 1002 | * message into the message array. If there is a waiting receiver, then it
|
| 1003 | * bypasses the message array and directly hands the message over to the
|
| 1004 | * receiver. The receiver accepts the message and returns without grabbing the
|
| 1005 | * queue spinlock:
|
| 1006 | *
|
| 1007 | * - Set pointer to message.
|
| 1008 | * - Queue the receiver task for later wakeup (without the info->lock).
|
| 1009 | * - Update its state to STATE_READY. Now the receiver can continue.
|
| 1010 | * - Wake up the process after the lock is dropped. Should the process wake up
|
| 1011 | * before this wakeup (due to a timeout or a signal) it will either see
|
| 1012 | * STATE_READY and continue or acquire the lock to check the state again.
|
| 1013 | *
|
| 1014 | * The same algorithm is used for senders.
|
| 1015 |
|
| 1016 |
|
| 1017 | static inline void __pipelined_op(struct wake_q_head *wake_q,
|
| 1018 | struct mqueue_inode_info *info,
|
| 1019 | struct ext_wait_queue *this)
|
| 1020 | {
|
| 1021 | struct task_struct *task;
|
| 1022 |
|
| 1023 | list_del(&this->list);
|
| 1024 | task = get_task_struct(this->task);
|
| 1025 |
|
| 1026 |
|
| 1027 | smp_store_release(&this->state, STATE_READY);
|
| 1028 | wake_q_add_safe(wake_q, task);
|
| 1029 | }
|
| 1030 |
|
| 1031 | |
| 1032 | * sys_mq_timedreceive() (without inserting message into a queue).
|
| 1033 |
|
| 1034 | static inline void pipelined_send(struct wake_q_head *wake_q,
|
| 1035 | struct mqueue_inode_info *info,
|
| 1036 | struct msg_msg *message,
|
| 1037 | struct ext_wait_queue *receiver)
|
| 1038 | {
|
| 1039 | receiver->msg = message;
|
| 1040 | __pipelined_op(wake_q, info, receiver);
|
| 1041 | }
|
| 1042 |
|
| 1043 | |
| 1044 | * gets its message and put to the queue (we have one free place for sure).
|
| 1045 | static inline void pipelined_receive(struct wake_q_head *wake_q,
|
| 1046 | struct mqueue_inode_info *info)
|
| 1047 | {
|
| 1048 | struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
|
| 1049 |
|
| 1050 | if (!sender) {
|
| 1051 |
|
| 1052 | wake_up_interruptible(&info->wait_q);
|
| 1053 | return;
|
| 1054 | }
|
| 1055 | if (msg_insert(sender->msg, info))
|
| 1056 | return;
|
| 1057 |
|
| 1058 | __pipelined_op(wake_q, info, sender);
|
| 1059 | }
|
| 1060 |
|
| 1061 | static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
|
| 1062 | size_t msg_len, unsigned int msg_prio,
|
| 1063 | struct timespec64 *ts)
|
| 1064 | {
|
| 1065 | struct fd f;
|
| 1066 | struct inode *inode;
|
| 1067 | struct ext_wait_queue wait;
|
| 1068 | struct ext_wait_queue *receiver;
|
| 1069 | struct msg_msg *msg_ptr;
|
| 1070 | struct mqueue_inode_info *info;
|
| 1071 | ktime_t expires, *timeout = NULL;
|
| 1072 | struct posix_msg_tree_node *new_leaf = NULL;
|
| 1073 | int ret = 0;
|
| 1074 | DEFINE_WAKE_Q(wake_q);
|
| 1075 |
|
| 1076 | if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
|
| 1077 | return -EINVAL;
|
| 1078 |
|
| 1079 | if (ts) {
|
| 1080 | expires = timespec64_to_ktime(*ts);
|
| 1081 | timeout = &expires;
|
| 1082 | }
|
| 1083 |
|
| 1084 | audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
|
| 1085 |
|
| 1086 | f = fdget(mqdes);
|
| 1087 | if (unlikely(!f.file)) {
|
| 1088 | ret = -EBADF;
|
| 1089 | goto out;
|
| 1090 | }
|
| 1091 |
|
| 1092 | inode = file_inode(f.file);
|
| 1093 | if (unlikely(f.file->f_op != &mqueue_file_operations)) {
|
| 1094 | ret = -EBADF;
|
| 1095 | goto out_fput;
|
| 1096 | }
|
| 1097 | info = MQUEUE_I(inode);
|
| 1098 | audit_file(f.file);
|
| 1099 |
|
| 1100 | if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
|
| 1101 | ret = -EBADF;
|
| 1102 | goto out_fput;
|
| 1103 | }
|
| 1104 |
|
| 1105 | if (unlikely(msg_len > info->attr.mq_msgsize)) {
|
| 1106 | ret = -EMSGSIZE;
|
| 1107 | goto out_fput;
|
| 1108 | }
|
| 1109 |
|
| 1110 | |
| 1111 | * existing queues.
|
| 1112 | msg_ptr = load_msg(u_msg_ptr, msg_len);
|
| 1113 | if (IS_ERR(msg_ptr)) {
|
| 1114 | ret = PTR_ERR(msg_ptr);
|
| 1115 | goto out_fput;
|
| 1116 | }
|
| 1117 | msg_ptr->m_ts = msg_len;
|
| 1118 | msg_ptr->m_type = msg_prio;
|
| 1119 |
|
| 1120 | |
| 1121 | * msg_insert really wants us to have a valid, spare node struct so
|
| 1122 | * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
|
| 1123 | * fall back to that if necessary.
|
| 1124 |
|
| 1125 | if (!info->node_cache)
|
| 1126 | new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
|
| 1127 |
|
| 1128 | spin_lock(&info->lock);
|
| 1129 |
|
| 1130 | if (!info->node_cache && new_leaf) {
|
| 1131 |
|
| 1132 | INIT_LIST_HEAD(&new_leaf->msg_list);
|
| 1133 | info->node_cache = new_leaf;
|
| 1134 | new_leaf = NULL;
|
| 1135 | } else {
|
| 1136 | kfree(new_leaf);
|
| 1137 | }
|
| 1138 |
|
| 1139 | if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
|
| 1140 | if (f.file->f_flags & O_NONBLOCK) {
|
| 1141 | ret = -EAGAIN;
|
| 1142 | } else {
|
| 1143 | wait.task = current;
|
| 1144 | wait.msg = (void *) msg_ptr;
|
| 1145 |
|
| 1146 |
|
| 1147 | WRITE_ONCE(wait.state, STATE_NONE);
|
| 1148 | ret = wq_sleep(info, SEND, timeout, &wait);
|
| 1149 | |
| 1150 | * wq_sleep must be called with info->lock held, and
|
| 1151 | * returns with the lock released
|
| 1152 |
|
| 1153 | goto out_free;
|
| 1154 | }
|
| 1155 | } else {
|
| 1156 | receiver = wq_get_first_waiter(info, RECV);
|
| 1157 | if (receiver) {
|
| 1158 | pipelined_send(&wake_q, info, msg_ptr, receiver);
|
| 1159 | } else {
|
| 1160 |
|
| 1161 | ret = msg_insert(msg_ptr, info);
|
| 1162 | if (ret)
|
| 1163 | goto out_unlock;
|
| 1164 | __do_notify(info);
|
| 1165 | }
|
| 1166 | inode->i_atime = inode->i_mtime = inode_set_ctime_current(inode);
|
| 1167 | }
|
| 1168 | out_unlock:
|
| 1169 | spin_unlock(&info->lock);
|
| 1170 | wake_up_q(&wake_q);
|
| 1171 | out_free:
|
| 1172 | if (ret)
|
| 1173 | free_msg(msg_ptr);
|
| 1174 | out_fput:
|
| 1175 | fdput(f);
|
| 1176 | out:
|
| 1177 | return ret;
|
| 1178 | }
|
| 1179 |
|
| 1180 | static 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 |
|
| 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 |
|
| 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 |
|
| 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_set_ctime_current(inode);
|
| 1261 |
|
| 1262 |
|
| 1263 | pipelined_receive(&wake_q, info);
|
| 1264 | spin_unlock(&info->lock);
|
| 1265 | wake_up_q(&wake_q);
|
| 1266 | ret = 0;
|
| 1267 | }
|
| 1268 | if (ret == 0) {
|
| 1269 | ret = msg_ptr->m_ts;
|
| 1270 |
|
| 1271 | if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
|
| 1272 | store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
|
| 1273 | ret = -EFAULT;
|
| 1274 | }
|
| 1275 | free_msg(msg_ptr);
|
| 1276 | }
|
| 1277 | out_fput:
|
| 1278 | fdput(f);
|
| 1279 | out:
|
| 1280 | return ret;
|
| 1281 | }
|
| 1282 |
|
| 1283 | SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
|
| 1284 | size_t, msg_len, unsigned int, msg_prio,
|
| 1285 | const struct __kernel_timespec __user *, u_abs_timeout)
|
| 1286 | {
|
| 1287 | struct timespec64 ts, *p = NULL;
|
| 1288 | if (u_abs_timeout) {
|
| 1289 | int res = prepare_timeout(u_abs_timeout, &ts);
|
| 1290 | if (res)
|
| 1291 | return res;
|
| 1292 | p = &ts;
|
| 1293 | }
|
| 1294 | return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
|
| 1295 | }
|
| 1296 |
|
| 1297 | SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
|
| 1298 | size_t, msg_len, unsigned int __user *, u_msg_prio,
|
| 1299 | const struct __kernel_timespec __user *, u_abs_timeout)
|
| 1300 | {
|
| 1301 | struct timespec64 ts, *p = NULL;
|
| 1302 | if (u_abs_timeout) {
|
| 1303 | int res = prepare_timeout(u_abs_timeout, &ts);
|
| 1304 | if (res)
|
| 1305 | return res;
|
| 1306 | p = &ts;
|
| 1307 | }
|
| 1308 | return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
|
| 1309 | }
|
| 1310 |
|
| 1311 | |
| 1312 | * Notes: the case when user wants us to deregister (with NULL as pointer)
|
| 1313 | * and he isn't currently owner of notification, will be silently discarded.
|
| 1314 | * It isn't explicitly defined in the POSIX.
|
| 1315 |
|
| 1316 | static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
|
| 1317 | {
|
| 1318 | int ret;
|
| 1319 | struct fd f;
|
| 1320 | struct sock *sock;
|
| 1321 | struct inode *inode;
|
| 1322 | struct mqueue_inode_info *info;
|
| 1323 | struct sk_buff *nc;
|
| 1324 |
|
| 1325 | audit_mq_notify(mqdes, notification);
|
| 1326 |
|
| 1327 | nc = NULL;
|
| 1328 | sock = NULL;
|
| 1329 | if (notification != NULL) {
|
| 1330 | if (unlikely(notification->sigev_notify != SIGEV_NONE &&
|
| 1331 | notification->sigev_notify != SIGEV_SIGNAL &&
|
| 1332 | notification->sigev_notify != SIGEV_THREAD))
|
| 1333 | return -EINVAL;
|
| 1334 | if (notification->sigev_notify == SIGEV_SIGNAL &&
|
| 1335 | !valid_signal(notification->sigev_signo)) {
|
| 1336 | return -EINVAL;
|
| 1337 | }
|
| 1338 | if (notification->sigev_notify == SIGEV_THREAD) {
|
| 1339 | long timeo;
|
| 1340 |
|
| 1341 |
|
| 1342 | nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
|
| 1343 | if (!nc)
|
| 1344 | return -ENOMEM;
|
| 1345 |
|
| 1346 | if (copy_from_user(nc->data,
|
| 1347 | notification->sigev_value.sival_ptr,
|
| 1348 | NOTIFY_COOKIE_LEN)) {
|
| 1349 | ret = -EFAULT;
|
| 1350 | goto free_skb;
|
| 1351 | }
|
| 1352 |
|
| 1353 |
|
| 1354 | skb_put(nc, NOTIFY_COOKIE_LEN);
|
| 1355 |
|
| 1356 | retry:
|
| 1357 | f = fdget(notification->sigev_signo);
|
| 1358 | if (!f.file) {
|
| 1359 | ret = -EBADF;
|
| 1360 | goto out;
|
| 1361 | }
|
| 1362 | sock = netlink_getsockbyfilp(f.file);
|
| 1363 | fdput(f);
|
| 1364 | if (IS_ERR(sock)) {
|
| 1365 | ret = PTR_ERR(sock);
|
| 1366 | goto free_skb;
|
| 1367 | }
|
| 1368 |
|
| 1369 | timeo = MAX_SCHEDULE_TIMEOUT;
|
| 1370 | ret = netlink_attachskb(sock, nc, &timeo, NULL);
|
| 1371 | if (ret == 1) {
|
| 1372 | sock = NULL;
|
| 1373 | goto retry;
|
| 1374 | }
|
| 1375 | if (ret)
|
| 1376 | return ret;
|
| 1377 | }
|
| 1378 | }
|
| 1379 |
|
| 1380 | f = fdget(mqdes);
|
| 1381 | if (!f.file) {
|
| 1382 | ret = -EBADF;
|
| 1383 | goto out;
|
| 1384 | }
|
| 1385 |
|
| 1386 | inode = file_inode(f.file);
|
| 1387 | if (unlikely(f.file->f_op != &mqueue_file_operations)) {
|
| 1388 | ret = -EBADF;
|
| 1389 | goto out_fput;
|
| 1390 | }
|
| 1391 | info = MQUEUE_I(inode);
|
| 1392 |
|
| 1393 | ret = 0;
|
| 1394 | spin_lock(&info->lock);
|
| 1395 | if (notification == NULL) {
|
| 1396 | if (info->notify_owner == task_tgid(current)) {
|
| 1397 | remove_notification(info);
|
| 1398 | inode->i_atime = inode_set_ctime_current(inode);
|
| 1399 | }
|
| 1400 | } else if (info->notify_owner != NULL) {
|
| 1401 | ret = -EBUSY;
|
| 1402 | } else {
|
| 1403 | switch (notification->sigev_notify) {
|
| 1404 | case SIGEV_NONE:
|
| 1405 | info->notify.sigev_notify = SIGEV_NONE;
|
| 1406 | break;
|
| 1407 | case SIGEV_THREAD:
|
| 1408 | info->notify_sock = sock;
|
| 1409 | info->notify_cookie = nc;
|
| 1410 | sock = NULL;
|
| 1411 | nc = NULL;
|
| 1412 | info->notify.sigev_notify = SIGEV_THREAD;
|
| 1413 | break;
|
| 1414 | case SIGEV_SIGNAL:
|
| 1415 | info->notify.sigev_signo = notification->sigev_signo;
|
| 1416 | info->notify.sigev_value = notification->sigev_value;
|
| 1417 | info->notify.sigev_notify = SIGEV_SIGNAL;
|
| 1418 | info->notify_self_exec_id = current->self_exec_id;
|
| 1419 | break;
|
| 1420 | }
|
| 1421 |
|
| 1422 | info->notify_owner = get_pid(task_tgid(current));
|
| 1423 | info->notify_user_ns = get_user_ns(current_user_ns());
|
| 1424 | inode->i_atime = inode_set_ctime_current(inode);
|
| 1425 | }
|
| 1426 | spin_unlock(&info->lock);
|
| 1427 | out_fput:
|
| 1428 | fdput(f);
|
| 1429 | out:
|
| 1430 | if (sock)
|
| 1431 | netlink_detachskb(sock, nc);
|
| 1432 | else
|
| 1433 | free_skb:
|
| 1434 | dev_kfree_skb(nc);
|
| 1435 |
|
| 1436 | return ret;
|
| 1437 | }
|
| 1438 |
|
| 1439 | SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
|
| 1440 | const struct sigevent __user *, u_notification)
|
| 1441 | {
|
| 1442 | struct sigevent n, *p = NULL;
|
| 1443 | if (u_notification) {
|
| 1444 | if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
|
| 1445 | return -EFAULT;
|
| 1446 | p = &n;
|
| 1447 | }
|
| 1448 | return do_mq_notify(mqdes, p);
|
| 1449 | }
|
| 1450 |
|
| 1451 | static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
|
| 1452 | {
|
| 1453 | struct fd f;
|
| 1454 | struct inode *inode;
|
| 1455 | struct mqueue_inode_info *info;
|
| 1456 |
|
| 1457 | if (new && (new->mq_flags & (~O_NONBLOCK)))
|
| 1458 | return -EINVAL;
|
| 1459 |
|
| 1460 | f = fdget(mqdes);
|
| 1461 | if (!f.file)
|
| 1462 | return -EBADF;
|
| 1463 |
|
| 1464 | if (unlikely(f.file->f_op != &mqueue_file_operations)) {
|
| 1465 | fdput(f);
|
| 1466 | return -EBADF;
|
| 1467 | }
|
| 1468 |
|
| 1469 | inode = file_inode(f.file);
|
| 1470 | info = MQUEUE_I(inode);
|
| 1471 |
|
| 1472 | spin_lock(&info->lock);
|
| 1473 |
|
| 1474 | if (old) {
|
| 1475 | *old = info->attr;
|
| 1476 | old->mq_flags = f.file->f_flags & O_NONBLOCK;
|
| 1477 | }
|
| 1478 | if (new) {
|
| 1479 | audit_mq_getsetattr(mqdes, new);
|
| 1480 | spin_lock(&f.file->f_lock);
|
| 1481 | if (new->mq_flags & O_NONBLOCK)
|
| 1482 | f.file->f_flags |= O_NONBLOCK;
|
| 1483 | else
|
| 1484 | f.file->f_flags &= ~O_NONBLOCK;
|
| 1485 | spin_unlock(&f.file->f_lock);
|
| 1486 |
|
| 1487 | inode->i_atime = inode_set_ctime_current(inode);
|
| 1488 | }
|
| 1489 |
|
| 1490 | spin_unlock(&info->lock);
|
| 1491 | fdput(f);
|
| 1492 | return 0;
|
| 1493 | }
|
| 1494 |
|
| 1495 | SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
|
| 1496 | const struct mq_attr __user *, u_mqstat,
|
| 1497 | struct mq_attr __user *, u_omqstat)
|
| 1498 | {
|
| 1499 | int ret;
|
| 1500 | struct mq_attr mqstat, omqstat;
|
| 1501 | struct mq_attr *new = NULL, *old = NULL;
|
| 1502 |
|
| 1503 | if (u_mqstat) {
|
| 1504 | new = &mqstat;
|
| 1505 | if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
|
| 1506 | return -EFAULT;
|
| 1507 | }
|
| 1508 | if (u_omqstat)
|
| 1509 | old = &omqstat;
|
| 1510 |
|
| 1511 | ret = do_mq_getsetattr(mqdes, new, old);
|
| 1512 | if (ret || !old)
|
| 1513 | return ret;
|
| 1514 |
|
| 1515 | if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
|
| 1516 | return -EFAULT;
|
| 1517 | return 0;
|
| 1518 | }
|
| 1519 |
|
| 1520 | #ifdef CONFIG_COMPAT
|
| 1521 |
|
| 1522 | struct compat_mq_attr {
|
| 1523 | compat_long_t mq_flags;
|
| 1524 | compat_long_t mq_maxmsg;
|
| 1525 | compat_long_t mq_msgsize;
|
| 1526 | compat_long_t mq_curmsgs;
|
| 1527 | compat_long_t __reserved[4];
|
| 1528 | };
|
| 1529 |
|
| 1530 | static inline int get_compat_mq_attr(struct mq_attr *attr,
|
| 1531 | const struct compat_mq_attr __user *uattr)
|
| 1532 | {
|
| 1533 | struct compat_mq_attr v;
|
| 1534 |
|
| 1535 | if (copy_from_user(&v, uattr, sizeof(*uattr)))
|
| 1536 | return -EFAULT;
|
| 1537 |
|
| 1538 | memset(attr, 0, sizeof(*attr));
|
| 1539 | attr->mq_flags = v.mq_flags;
|
| 1540 | attr->mq_maxmsg = v.mq_maxmsg;
|
| 1541 | attr->mq_msgsize = v.mq_msgsize;
|
| 1542 | attr->mq_curmsgs = v.mq_curmsgs;
|
| 1543 | return 0;
|
| 1544 | }
|
| 1545 |
|
| 1546 | static inline int put_compat_mq_attr(const struct mq_attr *attr,
|
| 1547 | struct compat_mq_attr __user *uattr)
|
| 1548 | {
|
| 1549 | struct compat_mq_attr v;
|
| 1550 |
|
| 1551 | memset(&v, 0, sizeof(v));
|
| 1552 | v.mq_flags = attr->mq_flags;
|
| 1553 | v.mq_maxmsg = attr->mq_maxmsg;
|
| 1554 | v.mq_msgsize = attr->mq_msgsize;
|
| 1555 | v.mq_curmsgs = attr->mq_curmsgs;
|
| 1556 | if (copy_to_user(uattr, &v, sizeof(*uattr)))
|
| 1557 | return -EFAULT;
|
| 1558 | return 0;
|
| 1559 | }
|
| 1560 |
|
| 1561 | COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
|
| 1562 | int, oflag, compat_mode_t, mode,
|
| 1563 | struct compat_mq_attr __user *, u_attr)
|
| 1564 | {
|
| 1565 | struct mq_attr attr, *p = NULL;
|
| 1566 | if (u_attr && oflag & O_CREAT) {
|
| 1567 | p = &attr;
|
| 1568 | if (get_compat_mq_attr(&attr, u_attr))
|
| 1569 | return -EFAULT;
|
| 1570 | }
|
| 1571 | return do_mq_open(u_name, oflag, mode, p);
|
| 1572 | }
|
| 1573 |
|
| 1574 | COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
|
| 1575 | const struct compat_sigevent __user *, u_notification)
|
| 1576 | {
|
| 1577 | struct sigevent n, *p = NULL;
|
| 1578 | if (u_notification) {
|
| 1579 | if (get_compat_sigevent(&n, u_notification))
|
| 1580 | return -EFAULT;
|
| 1581 | if (n.sigev_notify == SIGEV_THREAD)
|
| 1582 | n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
|
| 1583 | p = &n;
|
| 1584 | }
|
| 1585 | return do_mq_notify(mqdes, p);
|
| 1586 | }
|
| 1587 |
|
| 1588 | COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
|
| 1589 | const struct compat_mq_attr __user *, u_mqstat,
|
| 1590 | struct compat_mq_attr __user *, u_omqstat)
|
| 1591 | {
|
| 1592 | int ret;
|
| 1593 | struct mq_attr mqstat, omqstat;
|
| 1594 | struct mq_attr *new = NULL, *old = NULL;
|
| 1595 |
|
| 1596 | if (u_mqstat) {
|
| 1597 | new = &mqstat;
|
| 1598 | if (get_compat_mq_attr(new, u_mqstat))
|
| 1599 | return -EFAULT;
|
| 1600 | }
|
| 1601 | if (u_omqstat)
|
| 1602 | old = &omqstat;
|
| 1603 |
|
| 1604 | ret = do_mq_getsetattr(mqdes, new, old);
|
| 1605 | if (ret || !old)
|
| 1606 | return ret;
|
| 1607 |
|
| 1608 | if (put_compat_mq_attr(old, u_omqstat))
|
| 1609 | return -EFAULT;
|
| 1610 | return 0;
|
| 1611 | }
|
| 1612 | #endif
|
| 1613 |
|
| 1614 | #ifdef CONFIG_COMPAT_32BIT_TIME
|
| 1615 | static int compat_prepare_timeout(const struct old_timespec32 __user *p,
|
| 1616 | struct timespec64 *ts)
|
| 1617 | {
|
| 1618 | if (get_old_timespec32(ts, p))
|
| 1619 | return -EFAULT;
|
| 1620 | if (!timespec64_valid(ts))
|
| 1621 | return -EINVAL;
|
| 1622 | return 0;
|
| 1623 | }
|
| 1624 |
|
| 1625 | SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
|
| 1626 | const char __user *, u_msg_ptr,
|
| 1627 | unsigned int, msg_len, unsigned int, msg_prio,
|
| 1628 | const struct old_timespec32 __user *, u_abs_timeout)
|
| 1629 | {
|
| 1630 | struct timespec64 ts, *p = NULL;
|
| 1631 | if (u_abs_timeout) {
|
| 1632 | int res = compat_prepare_timeout(u_abs_timeout, &ts);
|
| 1633 | if (res)
|
| 1634 | return res;
|
| 1635 | p = &ts;
|
| 1636 | }
|
| 1637 | return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
|
| 1638 | }
|
| 1639 |
|
| 1640 | SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
|
| 1641 | char __user *, u_msg_ptr,
|
| 1642 | unsigned int, msg_len, unsigned int __user *, u_msg_prio,
|
| 1643 | const struct old_timespec32 __user *, u_abs_timeout)
|
| 1644 | {
|
| 1645 | struct timespec64 ts, *p = NULL;
|
| 1646 | if (u_abs_timeout) {
|
| 1647 | int res = compat_prepare_timeout(u_abs_timeout, &ts);
|
| 1648 | if (res)
|
| 1649 | return res;
|
| 1650 | p = &ts;
|
| 1651 | }
|
| 1652 | return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
|
| 1653 | }
|
| 1654 | #endif
|
| 1655 |
|
| 1656 | static const struct inode_operations mqueue_dir_inode_operations = {
|
| 1657 | .lookup = simple_lookup,
|
| 1658 | .create = mqueue_create,
|
| 1659 | .unlink = mqueue_unlink,
|
| 1660 | };
|
| 1661 |
|
| 1662 | static const struct file_operations mqueue_file_operations = {
|
| 1663 | .flush = mqueue_flush_file,
|
| 1664 | .poll = mqueue_poll_file,
|
| 1665 | .read = mqueue_read_file,
|
| 1666 | .llseek = default_llseek,
|
| 1667 | };
|
| 1668 |
|
| 1669 | static const struct super_operations mqueue_super_ops = {
|
| 1670 | .alloc_inode = mqueue_alloc_inode,
|
| 1671 | .free_inode = mqueue_free_inode,
|
| 1672 | .evict_inode = mqueue_evict_inode,
|
| 1673 | .statfs = simple_statfs,
|
| 1674 | };
|
| 1675 |
|
| 1676 | static const struct fs_context_operations mqueue_fs_context_ops = {
|
| 1677 | .free = mqueue_fs_context_free,
|
| 1678 | .get_tree = mqueue_get_tree,
|
| 1679 | };
|
| 1680 |
|
| 1681 | static struct file_system_type mqueue_fs_type = {
|
| 1682 | .name = "mqueue",
|
| 1683 | .init_fs_context = mqueue_init_fs_context,
|
| 1684 | .kill_sb = kill_litter_super,
|
| 1685 | .fs_flags = FS_USERNS_MOUNT,
|
| 1686 | };
|
| 1687 |
|
| 1688 | int mq_init_ns(struct ipc_namespace *ns)
|
| 1689 | {
|
| 1690 | struct vfsmount *m;
|
| 1691 |
|
| 1692 | ns->mq_queues_count = 0;
|
| 1693 | ns->mq_queues_max = DFLT_QUEUESMAX;
|
| 1694 | ns->mq_msg_max = DFLT_MSGMAX;
|
| 1695 | ns->mq_msgsize_max = DFLT_MSGSIZEMAX;
|
| 1696 | ns->mq_msg_default = DFLT_MSG;
|
| 1697 | ns->mq_msgsize_default = DFLT_MSGSIZE;
|
| 1698 |
|
| 1699 | m = mq_create_mount(ns);
|
| 1700 | if (IS_ERR(m))
|
| 1701 | return PTR_ERR(m);
|
| 1702 | ns->mq_mnt = m;
|
| 1703 | return 0;
|
| 1704 | }
|
| 1705 |
|
| 1706 | void mq_clear_sbinfo(struct ipc_namespace *ns)
|
| 1707 | {
|
| 1708 | ns->mq_mnt->mnt_sb->s_fs_info = NULL;
|
| 1709 | }
|
| 1710 |
|
| 1711 | static int __init init_mqueue_fs(void)
|
| 1712 | {
|
| 1713 | int error;
|
| 1714 |
|
| 1715 | mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
|
| 1716 | sizeof(struct mqueue_inode_info), 0,
|
| 1717 | SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
|
| 1718 | if (mqueue_inode_cachep == NULL)
|
| 1719 | return -ENOMEM;
|
| 1720 |
|
| 1721 | if (!setup_mq_sysctls(&init_ipc_ns)) {
|
| 1722 | pr_warn("sysctl registration failed\n");
|
| 1723 | error = -ENOMEM;
|
| 1724 | goto out_kmem;
|
| 1725 | }
|
| 1726 |
|
| 1727 | error = register_filesystem(&mqueue_fs_type);
|
| 1728 | if (error)
|
| 1729 | goto out_sysctl;
|
| 1730 |
|
| 1731 | spin_lock_init(&mq_lock);
|
| 1732 |
|
| 1733 | error = mq_init_ns(&init_ipc_ns);
|
| 1734 | if (error)
|
| 1735 | goto out_filesystem;
|
| 1736 |
|
| 1737 | return 0;
|
| 1738 |
|
| 1739 | out_filesystem:
|
| 1740 | unregister_filesystem(&mqueue_fs_type);
|
| 1741 | out_sysctl:
|
| 1742 | retire_mq_sysctls(&init_ipc_ns);
|
| 1743 | out_kmem:
|
| 1744 | kmem_cache_destroy(mqueue_inode_cachep);
|
| 1745 | return error;
|
| 1746 | }
|
| 1747 |
|
| 1748 | device_initcall(init_mqueue_fs);
|