Back to home page

LXR

 
 

    


0001 /*
0002  * POSIX message queues filesystem for Linux.
0003  *
0004  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
0005  *                          Michal Wronski          (michal.wronski@gmail.com)
0006  *
0007  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
0008  * Lockless receive & send, fd based notify:
0009  *              Manfred Spraul      (manfred@colorfullife.com)
0010  *
0011  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
0012  *
0013  * This file is released under the GPL.
0014  */
0015 
0016 #include <linux/capability.h>
0017 #include <linux/init.h>
0018 #include <linux/pagemap.h>
0019 #include <linux/file.h>
0020 #include <linux/mount.h>
0021 #include <linux/namei.h>
0022 #include <linux/sysctl.h>
0023 #include <linux/poll.h>
0024 #include <linux/mqueue.h>
0025 #include <linux/msg.h>
0026 #include <linux/skbuff.h>
0027 #include <linux/vmalloc.h>
0028 #include <linux/netlink.h>
0029 #include <linux/syscalls.h>
0030 #include <linux/audit.h>
0031 #include <linux/signal.h>
0032 #include <linux/mutex.h>
0033 #include <linux/nsproxy.h>
0034 #include <linux/pid.h>
0035 #include <linux/ipc_namespace.h>
0036 #include <linux/user_namespace.h>
0037 #include <linux/slab.h>
0038 
0039 #include <net/sock.h>
0040 #include "util.h"
0041 
0042 #define MQUEUE_MAGIC    0x19800202
0043 #define DIRENT_SIZE 20
0044 #define FILENT_SIZE 80
0045 
0046 #define SEND        0
0047 #define RECV        1
0048 
0049 #define STATE_NONE  0
0050 #define STATE_READY 1
0051 
0052 struct posix_msg_tree_node {
0053     struct rb_node      rb_node;
0054     struct list_head    msg_list;
0055     int         priority;
0056 };
0057 
0058 struct ext_wait_queue {     /* queue of sleeping tasks */
0059     struct task_struct *task;
0060     struct list_head list;
0061     struct msg_msg *msg;    /* ptr of loaded message */
0062     int state;      /* one of STATE_* values */
0063 };
0064 
0065 struct mqueue_inode_info {
0066     spinlock_t lock;
0067     struct inode vfs_inode;
0068     wait_queue_head_t wait_q;
0069 
0070     struct rb_root msg_tree;
0071     struct posix_msg_tree_node *node_cache;
0072     struct mq_attr attr;
0073 
0074     struct sigevent notify;
0075     struct pid *notify_owner;
0076     struct user_namespace *notify_user_ns;
0077     struct user_struct *user;   /* user who created, for accounting */
0078     struct sock *notify_sock;
0079     struct sk_buff *notify_cookie;
0080 
0081     /* for tasks waiting for free space and messages, respectively */
0082     struct ext_wait_queue e_wait_q[2];
0083 
0084     unsigned long qsize; /* size of queue in memory (sum of all msgs) */
0085 };
0086 
0087 static const struct inode_operations mqueue_dir_inode_operations;
0088 static const struct file_operations mqueue_file_operations;
0089 static const struct super_operations mqueue_super_ops;
0090 static void remove_notification(struct mqueue_inode_info *info);
0091 
0092 static struct kmem_cache *mqueue_inode_cachep;
0093 
0094 static struct ctl_table_header *mq_sysctl_table;
0095 
0096 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
0097 {
0098     return container_of(inode, struct mqueue_inode_info, vfs_inode);
0099 }
0100 
0101 /*
0102  * This routine should be called with the mq_lock held.
0103  */
0104 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
0105 {
0106     return get_ipc_ns(inode->i_sb->s_fs_info);
0107 }
0108 
0109 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
0110 {
0111     struct ipc_namespace *ns;
0112 
0113     spin_lock(&mq_lock);
0114     ns = __get_ns_from_inode(inode);
0115     spin_unlock(&mq_lock);
0116     return ns;
0117 }
0118 
0119 /* Auxiliary functions to manipulate messages' list */
0120 static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
0121 {
0122     struct rb_node **p, *parent = NULL;
0123     struct posix_msg_tree_node *leaf;
0124 
0125     p = &info->msg_tree.rb_node;
0126     while (*p) {
0127         parent = *p;
0128         leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
0129 
0130         if (likely(leaf->priority == msg->m_type))
0131             goto insert_msg;
0132         else if (msg->m_type < leaf->priority)
0133             p = &(*p)->rb_left;
0134         else
0135             p = &(*p)->rb_right;
0136     }
0137     if (info->node_cache) {
0138         leaf = info->node_cache;
0139         info->node_cache = NULL;
0140     } else {
0141         leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
0142         if (!leaf)
0143             return -ENOMEM;
0144         INIT_LIST_HEAD(&leaf->msg_list);
0145     }
0146     leaf->priority = msg->m_type;
0147     rb_link_node(&leaf->rb_node, parent, p);
0148     rb_insert_color(&leaf->rb_node, &info->msg_tree);
0149 insert_msg:
0150     info->attr.mq_curmsgs++;
0151     info->qsize += msg->m_ts;
0152     list_add_tail(&msg->m_list, &leaf->msg_list);
0153     return 0;
0154 }
0155 
0156 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
0157 {
0158     struct rb_node **p, *parent = NULL;
0159     struct posix_msg_tree_node *leaf;
0160     struct msg_msg *msg;
0161 
0162 try_again:
0163     p = &info->msg_tree.rb_node;
0164     while (*p) {
0165         parent = *p;
0166         /*
0167          * During insert, low priorities go to the left and high to the
0168          * right.  On receive, we want the highest priorities first, so
0169          * walk all the way to the right.
0170          */
0171         p = &(*p)->rb_right;
0172     }
0173     if (!parent) {
0174         if (info->attr.mq_curmsgs) {
0175             pr_warn_once("Inconsistency in POSIX message queue, "
0176                      "no tree element, but supposedly messages "
0177                      "should exist!\n");
0178             info->attr.mq_curmsgs = 0;
0179         }
0180         return NULL;
0181     }
0182     leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
0183     if (unlikely(list_empty(&leaf->msg_list))) {
0184         pr_warn_once("Inconsistency in POSIX message queue, "
0185                  "empty leaf node but we haven't implemented "
0186                  "lazy leaf delete!\n");
0187         rb_erase(&leaf->rb_node, &info->msg_tree);
0188         if (info->node_cache) {
0189             kfree(leaf);
0190         } else {
0191             info->node_cache = leaf;
0192         }
0193         goto try_again;
0194     } else {
0195         msg = list_first_entry(&leaf->msg_list,
0196                        struct msg_msg, m_list);
0197         list_del(&msg->m_list);
0198         if (list_empty(&leaf->msg_list)) {
0199             rb_erase(&leaf->rb_node, &info->msg_tree);
0200             if (info->node_cache) {
0201                 kfree(leaf);
0202             } else {
0203                 info->node_cache = leaf;
0204             }
0205         }
0206     }
0207     info->attr.mq_curmsgs--;
0208     info->qsize -= msg->m_ts;
0209     return msg;
0210 }
0211 
0212 static struct inode *mqueue_get_inode(struct super_block *sb,
0213         struct ipc_namespace *ipc_ns, umode_t mode,
0214         struct mq_attr *attr)
0215 {
0216     struct user_struct *u = current_user();
0217     struct inode *inode;
0218     int ret = -ENOMEM;
0219 
0220     inode = new_inode(sb);
0221     if (!inode)
0222         goto err;
0223 
0224     inode->i_ino = get_next_ino();
0225     inode->i_mode = mode;
0226     inode->i_uid = current_fsuid();
0227     inode->i_gid = current_fsgid();
0228     inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
0229 
0230     if (S_ISREG(mode)) {
0231         struct mqueue_inode_info *info;
0232         unsigned long mq_bytes, mq_treesize;
0233 
0234         inode->i_fop = &mqueue_file_operations;
0235         inode->i_size = FILENT_SIZE;
0236         /* mqueue specific info */
0237         info = MQUEUE_I(inode);
0238         spin_lock_init(&info->lock);
0239         init_waitqueue_head(&info->wait_q);
0240         INIT_LIST_HEAD(&info->e_wait_q[0].list);
0241         INIT_LIST_HEAD(&info->e_wait_q[1].list);
0242         info->notify_owner = NULL;
0243         info->notify_user_ns = NULL;
0244         info->qsize = 0;
0245         info->user = NULL;  /* set when all is ok */
0246         info->msg_tree = RB_ROOT;
0247         info->node_cache = NULL;
0248         memset(&info->attr, 0, sizeof(info->attr));
0249         info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
0250                        ipc_ns->mq_msg_default);
0251         info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
0252                         ipc_ns->mq_msgsize_default);
0253         if (attr) {
0254             info->attr.mq_maxmsg = attr->mq_maxmsg;
0255             info->attr.mq_msgsize = attr->mq_msgsize;
0256         }
0257         /*
0258          * We used to allocate a static array of pointers and account
0259          * the size of that array as well as one msg_msg struct per
0260          * possible message into the queue size. That's no longer
0261          * accurate as the queue is now an rbtree and will grow and
0262          * shrink depending on usage patterns.  We can, however, still
0263          * account one msg_msg struct per message, but the nodes are
0264          * allocated depending on priority usage, and most programs
0265          * only use one, or a handful, of priorities.  However, since
0266          * this is pinned memory, we need to assume worst case, so
0267          * that means the min(mq_maxmsg, max_priorities) * struct
0268          * posix_msg_tree_node.
0269          */
0270         mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
0271             min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
0272             sizeof(struct posix_msg_tree_node);
0273 
0274         mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
0275                       info->attr.mq_msgsize);
0276 
0277         spin_lock(&mq_lock);
0278         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
0279             u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
0280             spin_unlock(&mq_lock);
0281             /* mqueue_evict_inode() releases info->messages */
0282             ret = -EMFILE;
0283             goto out_inode;
0284         }
0285         u->mq_bytes += mq_bytes;
0286         spin_unlock(&mq_lock);
0287 
0288         /* all is ok */
0289         info->user = get_uid(u);
0290     } else if (S_ISDIR(mode)) {
0291         inc_nlink(inode);
0292         /* Some things misbehave if size == 0 on a directory */
0293         inode->i_size = 2 * DIRENT_SIZE;
0294         inode->i_op = &mqueue_dir_inode_operations;
0295         inode->i_fop = &simple_dir_operations;
0296     }
0297 
0298     return inode;
0299 out_inode:
0300     iput(inode);
0301 err:
0302     return ERR_PTR(ret);
0303 }
0304 
0305 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
0306 {
0307     struct inode *inode;
0308     struct ipc_namespace *ns = sb->s_fs_info;
0309 
0310     sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
0311     sb->s_blocksize = PAGE_SIZE;
0312     sb->s_blocksize_bits = PAGE_SHIFT;
0313     sb->s_magic = MQUEUE_MAGIC;
0314     sb->s_op = &mqueue_super_ops;
0315 
0316     inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
0317     if (IS_ERR(inode))
0318         return PTR_ERR(inode);
0319 
0320     sb->s_root = d_make_root(inode);
0321     if (!sb->s_root)
0322         return -ENOMEM;
0323     return 0;
0324 }
0325 
0326 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
0327              int flags, const char *dev_name,
0328              void *data)
0329 {
0330     struct ipc_namespace *ns;
0331     if (flags & MS_KERNMOUNT) {
0332         ns = data;
0333         data = NULL;
0334     } else {
0335         ns = current->nsproxy->ipc_ns;
0336     }
0337     return mount_ns(fs_type, flags, data, ns, ns->user_ns, mqueue_fill_super);
0338 }
0339 
0340 static void init_once(void *foo)
0341 {
0342     struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
0343 
0344     inode_init_once(&p->vfs_inode);
0345 }
0346 
0347 static struct inode *mqueue_alloc_inode(struct super_block *sb)
0348 {
0349     struct mqueue_inode_info *ei;
0350 
0351     ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
0352     if (!ei)
0353         return NULL;
0354     return &ei->vfs_inode;
0355 }
0356 
0357 static void mqueue_i_callback(struct rcu_head *head)
0358 {
0359     struct inode *inode = container_of(head, struct inode, i_rcu);
0360     kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
0361 }
0362 
0363 static void mqueue_destroy_inode(struct inode *inode)
0364 {
0365     call_rcu(&inode->i_rcu, mqueue_i_callback);
0366 }
0367 
0368 static void mqueue_evict_inode(struct inode *inode)
0369 {
0370     struct mqueue_inode_info *info;
0371     struct user_struct *user;
0372     unsigned long mq_bytes, mq_treesize;
0373     struct ipc_namespace *ipc_ns;
0374     struct msg_msg *msg;
0375 
0376     clear_inode(inode);
0377 
0378     if (S_ISDIR(inode->i_mode))
0379         return;
0380 
0381     ipc_ns = get_ns_from_inode(inode);
0382     info = MQUEUE_I(inode);
0383     spin_lock(&info->lock);
0384     while ((msg = msg_get(info)) != NULL)
0385         free_msg(msg);
0386     kfree(info->node_cache);
0387     spin_unlock(&info->lock);
0388 
0389     /* Total amount of bytes accounted for the mqueue */
0390     mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
0391         min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
0392         sizeof(struct posix_msg_tree_node);
0393 
0394     mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
0395                   info->attr.mq_msgsize);
0396 
0397     user = info->user;
0398     if (user) {
0399         spin_lock(&mq_lock);
0400         user->mq_bytes -= mq_bytes;
0401         /*
0402          * get_ns_from_inode() ensures that the
0403          * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
0404          * to which we now hold a reference, or it is NULL.
0405          * We can't put it here under mq_lock, though.
0406          */
0407         if (ipc_ns)
0408             ipc_ns->mq_queues_count--;
0409         spin_unlock(&mq_lock);
0410         free_uid(user);
0411     }
0412     if (ipc_ns)
0413         put_ipc_ns(ipc_ns);
0414 }
0415 
0416 static int mqueue_create(struct inode *dir, struct dentry *dentry,
0417                 umode_t mode, bool excl)
0418 {
0419     struct inode *inode;
0420     struct mq_attr *attr = dentry->d_fsdata;
0421     int error;
0422     struct ipc_namespace *ipc_ns;
0423 
0424     spin_lock(&mq_lock);
0425     ipc_ns = __get_ns_from_inode(dir);
0426     if (!ipc_ns) {
0427         error = -EACCES;
0428         goto out_unlock;
0429     }
0430 
0431     if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
0432         !capable(CAP_SYS_RESOURCE)) {
0433         error = -ENOSPC;
0434         goto out_unlock;
0435     }
0436     ipc_ns->mq_queues_count++;
0437     spin_unlock(&mq_lock);
0438 
0439     inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
0440     if (IS_ERR(inode)) {
0441         error = PTR_ERR(inode);
0442         spin_lock(&mq_lock);
0443         ipc_ns->mq_queues_count--;
0444         goto out_unlock;
0445     }
0446 
0447     put_ipc_ns(ipc_ns);
0448     dir->i_size += DIRENT_SIZE;
0449     dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
0450 
0451     d_instantiate(dentry, inode);
0452     dget(dentry);
0453     return 0;
0454 out_unlock:
0455     spin_unlock(&mq_lock);
0456     if (ipc_ns)
0457         put_ipc_ns(ipc_ns);
0458     return error;
0459 }
0460 
0461 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
0462 {
0463     struct inode *inode = d_inode(dentry);
0464 
0465     dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
0466     dir->i_size -= DIRENT_SIZE;
0467     drop_nlink(inode);
0468     dput(dentry);
0469     return 0;
0470 }
0471 
0472 /*
0473 *   This is routine for system read from queue file.
0474 *   To avoid mess with doing here some sort of mq_receive we allow
0475 *   to read only queue size & notification info (the only values
0476 *   that are interesting from user point of view and aren't accessible
0477 *   through std routines)
0478 */
0479 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
0480                 size_t count, loff_t *off)
0481 {
0482     struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
0483     char buffer[FILENT_SIZE];
0484     ssize_t ret;
0485 
0486     spin_lock(&info->lock);
0487     snprintf(buffer, sizeof(buffer),
0488             "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
0489             info->qsize,
0490             info->notify_owner ? info->notify.sigev_notify : 0,
0491             (info->notify_owner &&
0492              info->notify.sigev_notify == SIGEV_SIGNAL) ?
0493                 info->notify.sigev_signo : 0,
0494             pid_vnr(info->notify_owner));
0495     spin_unlock(&info->lock);
0496     buffer[sizeof(buffer)-1] = '\0';
0497 
0498     ret = simple_read_from_buffer(u_data, count, off, buffer,
0499                 strlen(buffer));
0500     if (ret <= 0)
0501         return ret;
0502 
0503     file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
0504     return ret;
0505 }
0506 
0507 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
0508 {
0509     struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
0510 
0511     spin_lock(&info->lock);
0512     if (task_tgid(current) == info->notify_owner)
0513         remove_notification(info);
0514 
0515     spin_unlock(&info->lock);
0516     return 0;
0517 }
0518 
0519 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
0520 {
0521     struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
0522     int retval = 0;
0523 
0524     poll_wait(filp, &info->wait_q, poll_tab);
0525 
0526     spin_lock(&info->lock);
0527     if (info->attr.mq_curmsgs)
0528         retval = POLLIN | POLLRDNORM;
0529 
0530     if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
0531         retval |= POLLOUT | POLLWRNORM;
0532     spin_unlock(&info->lock);
0533 
0534     return retval;
0535 }
0536 
0537 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
0538 static void wq_add(struct mqueue_inode_info *info, int sr,
0539             struct ext_wait_queue *ewp)
0540 {
0541     struct ext_wait_queue *walk;
0542 
0543     ewp->task = current;
0544 
0545     list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
0546         if (walk->task->static_prio <= current->static_prio) {
0547             list_add_tail(&ewp->list, &walk->list);
0548             return;
0549         }
0550     }
0551     list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
0552 }
0553 
0554 /*
0555  * Puts current task to sleep. Caller must hold queue lock. After return
0556  * lock isn't held.
0557  * sr: SEND or RECV
0558  */
0559 static int wq_sleep(struct mqueue_inode_info *info, int sr,
0560             ktime_t *timeout, struct ext_wait_queue *ewp)
0561 {
0562     int retval;
0563     signed long time;
0564 
0565     wq_add(info, sr, ewp);
0566 
0567     for (;;) {
0568         __set_current_state(TASK_INTERRUPTIBLE);
0569 
0570         spin_unlock(&info->lock);
0571         time = schedule_hrtimeout_range_clock(timeout, 0,
0572             HRTIMER_MODE_ABS, CLOCK_REALTIME);
0573 
0574         if (ewp->state == STATE_READY) {
0575             retval = 0;
0576             goto out;
0577         }
0578         spin_lock(&info->lock);
0579         if (ewp->state == STATE_READY) {
0580             retval = 0;
0581             goto out_unlock;
0582         }
0583         if (signal_pending(current)) {
0584             retval = -ERESTARTSYS;
0585             break;
0586         }
0587         if (time == 0) {
0588             retval = -ETIMEDOUT;
0589             break;
0590         }
0591     }
0592     list_del(&ewp->list);
0593 out_unlock:
0594     spin_unlock(&info->lock);
0595 out:
0596     return retval;
0597 }
0598 
0599 /*
0600  * Returns waiting task that should be serviced first or NULL if none exists
0601  */
0602 static struct ext_wait_queue *wq_get_first_waiter(
0603         struct mqueue_inode_info *info, int sr)
0604 {
0605     struct list_head *ptr;
0606 
0607     ptr = info->e_wait_q[sr].list.prev;
0608     if (ptr == &info->e_wait_q[sr].list)
0609         return NULL;
0610     return list_entry(ptr, struct ext_wait_queue, list);
0611 }
0612 
0613 
0614 static inline void set_cookie(struct sk_buff *skb, char code)
0615 {
0616     ((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
0617 }
0618 
0619 /*
0620  * The next function is only to split too long sys_mq_timedsend
0621  */
0622 static void __do_notify(struct mqueue_inode_info *info)
0623 {
0624     /* notification
0625      * invoked when there is registered process and there isn't process
0626      * waiting synchronously for message AND state of queue changed from
0627      * empty to not empty. Here we are sure that no one is waiting
0628      * synchronously. */
0629     if (info->notify_owner &&
0630         info->attr.mq_curmsgs == 1) {
0631         struct siginfo sig_i;
0632         switch (info->notify.sigev_notify) {
0633         case SIGEV_NONE:
0634             break;
0635         case SIGEV_SIGNAL:
0636             /* sends signal */
0637 
0638             sig_i.si_signo = info->notify.sigev_signo;
0639             sig_i.si_errno = 0;
0640             sig_i.si_code = SI_MESGQ;
0641             sig_i.si_value = info->notify.sigev_value;
0642             /* map current pid/uid into info->owner's namespaces */
0643             rcu_read_lock();
0644             sig_i.si_pid = task_tgid_nr_ns(current,
0645                         ns_of_pid(info->notify_owner));
0646             sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
0647             rcu_read_unlock();
0648 
0649             kill_pid_info(info->notify.sigev_signo,
0650                       &sig_i, info->notify_owner);
0651             break;
0652         case SIGEV_THREAD:
0653             set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
0654             netlink_sendskb(info->notify_sock, info->notify_cookie);
0655             break;
0656         }
0657         /* after notification unregisters process */
0658         put_pid(info->notify_owner);
0659         put_user_ns(info->notify_user_ns);
0660         info->notify_owner = NULL;
0661         info->notify_user_ns = NULL;
0662     }
0663     wake_up(&info->wait_q);
0664 }
0665 
0666 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
0667                ktime_t *expires, struct timespec *ts)
0668 {
0669     if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
0670         return -EFAULT;
0671     if (!timespec_valid(ts))
0672         return -EINVAL;
0673 
0674     *expires = timespec_to_ktime(*ts);
0675     return 0;
0676 }
0677 
0678 static void remove_notification(struct mqueue_inode_info *info)
0679 {
0680     if (info->notify_owner != NULL &&
0681         info->notify.sigev_notify == SIGEV_THREAD) {
0682         set_cookie(info->notify_cookie, NOTIFY_REMOVED);
0683         netlink_sendskb(info->notify_sock, info->notify_cookie);
0684     }
0685     put_pid(info->notify_owner);
0686     put_user_ns(info->notify_user_ns);
0687     info->notify_owner = NULL;
0688     info->notify_user_ns = NULL;
0689 }
0690 
0691 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
0692 {
0693     int mq_treesize;
0694     unsigned long total_size;
0695 
0696     if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
0697         return -EINVAL;
0698     if (capable(CAP_SYS_RESOURCE)) {
0699         if (attr->mq_maxmsg > HARD_MSGMAX ||
0700             attr->mq_msgsize > HARD_MSGSIZEMAX)
0701             return -EINVAL;
0702     } else {
0703         if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
0704                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
0705             return -EINVAL;
0706     }
0707     /* check for overflow */
0708     if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
0709         return -EOVERFLOW;
0710     mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
0711         min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
0712         sizeof(struct posix_msg_tree_node);
0713     total_size = attr->mq_maxmsg * attr->mq_msgsize;
0714     if (total_size + mq_treesize < total_size)
0715         return -EOVERFLOW;
0716     return 0;
0717 }
0718 
0719 /*
0720  * Invoked when creating a new queue via sys_mq_open
0721  */
0722 static struct file *do_create(struct ipc_namespace *ipc_ns, struct inode *dir,
0723             struct path *path, int oflag, umode_t mode,
0724             struct mq_attr *attr)
0725 {
0726     const struct cred *cred = current_cred();
0727     int ret;
0728 
0729     if (attr) {
0730         ret = mq_attr_ok(ipc_ns, attr);
0731         if (ret)
0732             return ERR_PTR(ret);
0733         /* store for use during create */
0734         path->dentry->d_fsdata = attr;
0735     } else {
0736         struct mq_attr def_attr;
0737 
0738         def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
0739                      ipc_ns->mq_msg_default);
0740         def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
0741                       ipc_ns->mq_msgsize_default);
0742         ret = mq_attr_ok(ipc_ns, &def_attr);
0743         if (ret)
0744             return ERR_PTR(ret);
0745     }
0746 
0747     mode &= ~current_umask();
0748     ret = vfs_create(dir, path->dentry, mode, true);
0749     path->dentry->d_fsdata = NULL;
0750     if (ret)
0751         return ERR_PTR(ret);
0752     return dentry_open(path, oflag, cred);
0753 }
0754 
0755 /* Opens existing queue */
0756 static struct file *do_open(struct path *path, int oflag)
0757 {
0758     static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
0759                           MAY_READ | MAY_WRITE };
0760     int acc;
0761     if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
0762         return ERR_PTR(-EINVAL);
0763     acc = oflag2acc[oflag & O_ACCMODE];
0764     if (inode_permission(d_inode(path->dentry), acc))
0765         return ERR_PTR(-EACCES);
0766     return dentry_open(path, oflag, current_cred());
0767 }
0768 
0769 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
0770         struct mq_attr __user *, u_attr)
0771 {
0772     struct path path;
0773     struct file *filp;
0774     struct filename *name;
0775     struct mq_attr attr;
0776     int fd, error;
0777     struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
0778     struct vfsmount *mnt = ipc_ns->mq_mnt;
0779     struct dentry *root = mnt->mnt_root;
0780     int ro;
0781 
0782     if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
0783         return -EFAULT;
0784 
0785     audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
0786 
0787     if (IS_ERR(name = getname(u_name)))
0788         return PTR_ERR(name);
0789 
0790     fd = get_unused_fd_flags(O_CLOEXEC);
0791     if (fd < 0)
0792         goto out_putname;
0793 
0794     ro = mnt_want_write(mnt);   /* we'll drop it in any case */
0795     error = 0;
0796     inode_lock(d_inode(root));
0797     path.dentry = lookup_one_len(name->name, root, strlen(name->name));
0798     if (IS_ERR(path.dentry)) {
0799         error = PTR_ERR(path.dentry);
0800         goto out_putfd;
0801     }
0802     path.mnt = mntget(mnt);
0803 
0804     if (oflag & O_CREAT) {
0805         if (d_really_is_positive(path.dentry)) {    /* entry already exists */
0806             audit_inode(name, path.dentry, 0);
0807             if (oflag & O_EXCL) {
0808                 error = -EEXIST;
0809                 goto out;
0810             }
0811             filp = do_open(&path, oflag);
0812         } else {
0813             if (ro) {
0814                 error = ro;
0815                 goto out;
0816             }
0817             audit_inode_parent_hidden(name, root);
0818             filp = do_create(ipc_ns, d_inode(root),
0819                         &path, oflag, mode,
0820                         u_attr ? &attr : NULL);
0821         }
0822     } else {
0823         if (d_really_is_negative(path.dentry)) {
0824             error = -ENOENT;
0825             goto out;
0826         }
0827         audit_inode(name, path.dentry, 0);
0828         filp = do_open(&path, oflag);
0829     }
0830 
0831     if (!IS_ERR(filp))
0832         fd_install(fd, filp);
0833     else
0834         error = PTR_ERR(filp);
0835 out:
0836     path_put(&path);
0837 out_putfd:
0838     if (error) {
0839         put_unused_fd(fd);
0840         fd = error;
0841     }
0842     inode_unlock(d_inode(root));
0843     if (!ro)
0844         mnt_drop_write(mnt);
0845 out_putname:
0846     putname(name);
0847     return fd;
0848 }
0849 
0850 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
0851 {
0852     int err;
0853     struct filename *name;
0854     struct dentry *dentry;
0855     struct inode *inode = NULL;
0856     struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
0857     struct vfsmount *mnt = ipc_ns->mq_mnt;
0858 
0859     name = getname(u_name);
0860     if (IS_ERR(name))
0861         return PTR_ERR(name);
0862 
0863     audit_inode_parent_hidden(name, mnt->mnt_root);
0864     err = mnt_want_write(mnt);
0865     if (err)
0866         goto out_name;
0867     inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
0868     dentry = lookup_one_len(name->name, mnt->mnt_root,
0869                 strlen(name->name));
0870     if (IS_ERR(dentry)) {
0871         err = PTR_ERR(dentry);
0872         goto out_unlock;
0873     }
0874 
0875     inode = d_inode(dentry);
0876     if (!inode) {
0877         err = -ENOENT;
0878     } else {
0879         ihold(inode);
0880         err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
0881     }
0882     dput(dentry);
0883 
0884 out_unlock:
0885     inode_unlock(d_inode(mnt->mnt_root));
0886     if (inode)
0887         iput(inode);
0888     mnt_drop_write(mnt);
0889 out_name:
0890     putname(name);
0891 
0892     return err;
0893 }
0894 
0895 /* Pipelined send and receive functions.
0896  *
0897  * If a receiver finds no waiting message, then it registers itself in the
0898  * list of waiting receivers. A sender checks that list before adding the new
0899  * message into the message array. If there is a waiting receiver, then it
0900  * bypasses the message array and directly hands the message over to the
0901  * receiver. The receiver accepts the message and returns without grabbing the
0902  * queue spinlock:
0903  *
0904  * - Set pointer to message.
0905  * - Queue the receiver task for later wakeup (without the info->lock).
0906  * - Update its state to STATE_READY. Now the receiver can continue.
0907  * - Wake up the process after the lock is dropped. Should the process wake up
0908  *   before this wakeup (due to a timeout or a signal) it will either see
0909  *   STATE_READY and continue or acquire the lock to check the state again.
0910  *
0911  * The same algorithm is used for senders.
0912  */
0913 
0914 /* pipelined_send() - send a message directly to the task waiting in
0915  * sys_mq_timedreceive() (without inserting message into a queue).
0916  */
0917 static inline void pipelined_send(struct wake_q_head *wake_q,
0918                   struct mqueue_inode_info *info,
0919                   struct msg_msg *message,
0920                   struct ext_wait_queue *receiver)
0921 {
0922     receiver->msg = message;
0923     list_del(&receiver->list);
0924     wake_q_add(wake_q, receiver->task);
0925     /*
0926      * Rely on the implicit cmpxchg barrier from wake_q_add such
0927      * that we can ensure that updating receiver->state is the last
0928      * write operation: As once set, the receiver can continue,
0929      * and if we don't have the reference count from the wake_q,
0930      * yet, at that point we can later have a use-after-free
0931      * condition and bogus wakeup.
0932      */
0933     receiver->state = STATE_READY;
0934 }
0935 
0936 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
0937  * gets its message and put to the queue (we have one free place for sure). */
0938 static inline void pipelined_receive(struct wake_q_head *wake_q,
0939                      struct mqueue_inode_info *info)
0940 {
0941     struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
0942 
0943     if (!sender) {
0944         /* for poll */
0945         wake_up_interruptible(&info->wait_q);
0946         return;
0947     }
0948     if (msg_insert(sender->msg, info))
0949         return;
0950 
0951     list_del(&sender->list);
0952     wake_q_add(wake_q, sender->task);
0953     sender->state = STATE_READY;
0954 }
0955 
0956 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
0957         size_t, msg_len, unsigned int, msg_prio,
0958         const struct timespec __user *, u_abs_timeout)
0959 {
0960     struct fd f;
0961     struct inode *inode;
0962     struct ext_wait_queue wait;
0963     struct ext_wait_queue *receiver;
0964     struct msg_msg *msg_ptr;
0965     struct mqueue_inode_info *info;
0966     ktime_t expires, *timeout = NULL;
0967     struct timespec ts;
0968     struct posix_msg_tree_node *new_leaf = NULL;
0969     int ret = 0;
0970     DEFINE_WAKE_Q(wake_q);
0971 
0972     if (u_abs_timeout) {
0973         int res = prepare_timeout(u_abs_timeout, &expires, &ts);
0974         if (res)
0975             return res;
0976         timeout = &expires;
0977     }
0978 
0979     if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
0980         return -EINVAL;
0981 
0982     audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
0983 
0984     f = fdget(mqdes);
0985     if (unlikely(!f.file)) {
0986         ret = -EBADF;
0987         goto out;
0988     }
0989 
0990     inode = file_inode(f.file);
0991     if (unlikely(f.file->f_op != &mqueue_file_operations)) {
0992         ret = -EBADF;
0993         goto out_fput;
0994     }
0995     info = MQUEUE_I(inode);
0996     audit_file(f.file);
0997 
0998     if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
0999         ret = -EBADF;
1000         goto out_fput;
1001     }
1002 
1003     if (unlikely(msg_len > info->attr.mq_msgsize)) {
1004         ret = -EMSGSIZE;
1005         goto out_fput;
1006     }
1007 
1008     /* First try to allocate memory, before doing anything with
1009      * existing queues. */
1010     msg_ptr = load_msg(u_msg_ptr, msg_len);
1011     if (IS_ERR(msg_ptr)) {
1012         ret = PTR_ERR(msg_ptr);
1013         goto out_fput;
1014     }
1015     msg_ptr->m_ts = msg_len;
1016     msg_ptr->m_type = msg_prio;
1017 
1018     /*
1019      * msg_insert really wants us to have a valid, spare node struct so
1020      * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1021      * fall back to that if necessary.
1022      */
1023     if (!info->node_cache)
1024         new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1025 
1026     spin_lock(&info->lock);
1027 
1028     if (!info->node_cache && new_leaf) {
1029         /* Save our speculative allocation into the cache */
1030         INIT_LIST_HEAD(&new_leaf->msg_list);
1031         info->node_cache = new_leaf;
1032         new_leaf = NULL;
1033     } else {
1034         kfree(new_leaf);
1035     }
1036 
1037     if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1038         if (f.file->f_flags & O_NONBLOCK) {
1039             ret = -EAGAIN;
1040         } else {
1041             wait.task = current;
1042             wait.msg = (void *) msg_ptr;
1043             wait.state = STATE_NONE;
1044             ret = wq_sleep(info, SEND, timeout, &wait);
1045             /*
1046              * wq_sleep must be called with info->lock held, and
1047              * returns with the lock released
1048              */
1049             goto out_free;
1050         }
1051     } else {
1052         receiver = wq_get_first_waiter(info, RECV);
1053         if (receiver) {
1054             pipelined_send(&wake_q, info, msg_ptr, receiver);
1055         } else {
1056             /* adds message to the queue */
1057             ret = msg_insert(msg_ptr, info);
1058             if (ret)
1059                 goto out_unlock;
1060             __do_notify(info);
1061         }
1062         inode->i_atime = inode->i_mtime = inode->i_ctime =
1063                 current_time(inode);
1064     }
1065 out_unlock:
1066     spin_unlock(&info->lock);
1067     wake_up_q(&wake_q);
1068 out_free:
1069     if (ret)
1070         free_msg(msg_ptr);
1071 out_fput:
1072     fdput(f);
1073 out:
1074     return ret;
1075 }
1076 
1077 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1078         size_t, msg_len, unsigned int __user *, u_msg_prio,
1079         const struct timespec __user *, u_abs_timeout)
1080 {
1081     ssize_t ret;
1082     struct msg_msg *msg_ptr;
1083     struct fd f;
1084     struct inode *inode;
1085     struct mqueue_inode_info *info;
1086     struct ext_wait_queue wait;
1087     ktime_t expires, *timeout = NULL;
1088     struct timespec ts;
1089     struct posix_msg_tree_node *new_leaf = NULL;
1090 
1091     if (u_abs_timeout) {
1092         int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1093         if (res)
1094             return res;
1095         timeout = &expires;
1096     }
1097 
1098     audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1099 
1100     f = fdget(mqdes);
1101     if (unlikely(!f.file)) {
1102         ret = -EBADF;
1103         goto out;
1104     }
1105 
1106     inode = file_inode(f.file);
1107     if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1108         ret = -EBADF;
1109         goto out_fput;
1110     }
1111     info = MQUEUE_I(inode);
1112     audit_file(f.file);
1113 
1114     if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1115         ret = -EBADF;
1116         goto out_fput;
1117     }
1118 
1119     /* checks if buffer is big enough */
1120     if (unlikely(msg_len < info->attr.mq_msgsize)) {
1121         ret = -EMSGSIZE;
1122         goto out_fput;
1123     }
1124 
1125     /*
1126      * msg_insert really wants us to have a valid, spare node struct so
1127      * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1128      * fall back to that if necessary.
1129      */
1130     if (!info->node_cache)
1131         new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1132 
1133     spin_lock(&info->lock);
1134 
1135     if (!info->node_cache && new_leaf) {
1136         /* Save our speculative allocation into the cache */
1137         INIT_LIST_HEAD(&new_leaf->msg_list);
1138         info->node_cache = new_leaf;
1139     } else {
1140         kfree(new_leaf);
1141     }
1142 
1143     if (info->attr.mq_curmsgs == 0) {
1144         if (f.file->f_flags & O_NONBLOCK) {
1145             spin_unlock(&info->lock);
1146             ret = -EAGAIN;
1147         } else {
1148             wait.task = current;
1149             wait.state = STATE_NONE;
1150             ret = wq_sleep(info, RECV, timeout, &wait);
1151             msg_ptr = wait.msg;
1152         }
1153     } else {
1154         DEFINE_WAKE_Q(wake_q);
1155 
1156         msg_ptr = msg_get(info);
1157 
1158         inode->i_atime = inode->i_mtime = inode->i_ctime =
1159                 current_time(inode);
1160 
1161         /* There is now free space in queue. */
1162         pipelined_receive(&wake_q, info);
1163         spin_unlock(&info->lock);
1164         wake_up_q(&wake_q);
1165         ret = 0;
1166     }
1167     if (ret == 0) {
1168         ret = msg_ptr->m_ts;
1169 
1170         if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1171             store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1172             ret = -EFAULT;
1173         }
1174         free_msg(msg_ptr);
1175     }
1176 out_fput:
1177     fdput(f);
1178 out:
1179     return ret;
1180 }
1181 
1182 /*
1183  * Notes: the case when user wants us to deregister (with NULL as pointer)
1184  * and he isn't currently owner of notification, will be silently discarded.
1185  * It isn't explicitly defined in the POSIX.
1186  */
1187 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1188         const struct sigevent __user *, u_notification)
1189 {
1190     int ret;
1191     struct fd f;
1192     struct sock *sock;
1193     struct inode *inode;
1194     struct sigevent notification;
1195     struct mqueue_inode_info *info;
1196     struct sk_buff *nc;
1197 
1198     if (u_notification) {
1199         if (copy_from_user(&notification, u_notification,
1200                     sizeof(struct sigevent)))
1201             return -EFAULT;
1202     }
1203 
1204     audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1205 
1206     nc = NULL;
1207     sock = NULL;
1208     if (u_notification != NULL) {
1209         if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1210                  notification.sigev_notify != SIGEV_SIGNAL &&
1211                  notification.sigev_notify != SIGEV_THREAD))
1212             return -EINVAL;
1213         if (notification.sigev_notify == SIGEV_SIGNAL &&
1214             !valid_signal(notification.sigev_signo)) {
1215             return -EINVAL;
1216         }
1217         if (notification.sigev_notify == SIGEV_THREAD) {
1218             long timeo;
1219 
1220             /* create the notify skb */
1221             nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1222             if (!nc) {
1223                 ret = -ENOMEM;
1224                 goto out;
1225             }
1226             if (copy_from_user(nc->data,
1227                     notification.sigev_value.sival_ptr,
1228                     NOTIFY_COOKIE_LEN)) {
1229                 ret = -EFAULT;
1230                 goto out;
1231             }
1232 
1233             /* TODO: add a header? */
1234             skb_put(nc, NOTIFY_COOKIE_LEN);
1235             /* and attach it to the socket */
1236 retry:
1237             f = fdget(notification.sigev_signo);
1238             if (!f.file) {
1239                 ret = -EBADF;
1240                 goto out;
1241             }
1242             sock = netlink_getsockbyfilp(f.file);
1243             fdput(f);
1244             if (IS_ERR(sock)) {
1245                 ret = PTR_ERR(sock);
1246                 sock = NULL;
1247                 goto out;
1248             }
1249 
1250             timeo = MAX_SCHEDULE_TIMEOUT;
1251             ret = netlink_attachskb(sock, nc, &timeo, NULL);
1252             if (ret == 1)
1253                 goto retry;
1254             if (ret) {
1255                 sock = NULL;
1256                 nc = NULL;
1257                 goto out;
1258             }
1259         }
1260     }
1261 
1262     f = fdget(mqdes);
1263     if (!f.file) {
1264         ret = -EBADF;
1265         goto out;
1266     }
1267 
1268     inode = file_inode(f.file);
1269     if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1270         ret = -EBADF;
1271         goto out_fput;
1272     }
1273     info = MQUEUE_I(inode);
1274 
1275     ret = 0;
1276     spin_lock(&info->lock);
1277     if (u_notification == NULL) {
1278         if (info->notify_owner == task_tgid(current)) {
1279             remove_notification(info);
1280             inode->i_atime = inode->i_ctime = current_time(inode);
1281         }
1282     } else if (info->notify_owner != NULL) {
1283         ret = -EBUSY;
1284     } else {
1285         switch (notification.sigev_notify) {
1286         case SIGEV_NONE:
1287             info->notify.sigev_notify = SIGEV_NONE;
1288             break;
1289         case SIGEV_THREAD:
1290             info->notify_sock = sock;
1291             info->notify_cookie = nc;
1292             sock = NULL;
1293             nc = NULL;
1294             info->notify.sigev_notify = SIGEV_THREAD;
1295             break;
1296         case SIGEV_SIGNAL:
1297             info->notify.sigev_signo = notification.sigev_signo;
1298             info->notify.sigev_value = notification.sigev_value;
1299             info->notify.sigev_notify = SIGEV_SIGNAL;
1300             break;
1301         }
1302 
1303         info->notify_owner = get_pid(task_tgid(current));
1304         info->notify_user_ns = get_user_ns(current_user_ns());
1305         inode->i_atime = inode->i_ctime = current_time(inode);
1306     }
1307     spin_unlock(&info->lock);
1308 out_fput:
1309     fdput(f);
1310 out:
1311     if (sock)
1312         netlink_detachskb(sock, nc);
1313     else if (nc)
1314         dev_kfree_skb(nc);
1315 
1316     return ret;
1317 }
1318 
1319 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1320         const struct mq_attr __user *, u_mqstat,
1321         struct mq_attr __user *, u_omqstat)
1322 {
1323     int ret;
1324     struct mq_attr mqstat, omqstat;
1325     struct fd f;
1326     struct inode *inode;
1327     struct mqueue_inode_info *info;
1328 
1329     if (u_mqstat != NULL) {
1330         if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1331             return -EFAULT;
1332         if (mqstat.mq_flags & (~O_NONBLOCK))
1333             return -EINVAL;
1334     }
1335 
1336     f = fdget(mqdes);
1337     if (!f.file) {
1338         ret = -EBADF;
1339         goto out;
1340     }
1341 
1342     inode = file_inode(f.file);
1343     if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1344         ret = -EBADF;
1345         goto out_fput;
1346     }
1347     info = MQUEUE_I(inode);
1348 
1349     spin_lock(&info->lock);
1350 
1351     omqstat = info->attr;
1352     omqstat.mq_flags = f.file->f_flags & O_NONBLOCK;
1353     if (u_mqstat) {
1354         audit_mq_getsetattr(mqdes, &mqstat);
1355         spin_lock(&f.file->f_lock);
1356         if (mqstat.mq_flags & O_NONBLOCK)
1357             f.file->f_flags |= O_NONBLOCK;
1358         else
1359             f.file->f_flags &= ~O_NONBLOCK;
1360         spin_unlock(&f.file->f_lock);
1361 
1362         inode->i_atime = inode->i_ctime = current_time(inode);
1363     }
1364 
1365     spin_unlock(&info->lock);
1366 
1367     ret = 0;
1368     if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1369                         sizeof(struct mq_attr)))
1370         ret = -EFAULT;
1371 
1372 out_fput:
1373     fdput(f);
1374 out:
1375     return ret;
1376 }
1377 
1378 static const struct inode_operations mqueue_dir_inode_operations = {
1379     .lookup = simple_lookup,
1380     .create = mqueue_create,
1381     .unlink = mqueue_unlink,
1382 };
1383 
1384 static const struct file_operations mqueue_file_operations = {
1385     .flush = mqueue_flush_file,
1386     .poll = mqueue_poll_file,
1387     .read = mqueue_read_file,
1388     .llseek = default_llseek,
1389 };
1390 
1391 static const struct super_operations mqueue_super_ops = {
1392     .alloc_inode = mqueue_alloc_inode,
1393     .destroy_inode = mqueue_destroy_inode,
1394     .evict_inode = mqueue_evict_inode,
1395     .statfs = simple_statfs,
1396 };
1397 
1398 static struct file_system_type mqueue_fs_type = {
1399     .name = "mqueue",
1400     .mount = mqueue_mount,
1401     .kill_sb = kill_litter_super,
1402     .fs_flags = FS_USERNS_MOUNT,
1403 };
1404 
1405 int mq_init_ns(struct ipc_namespace *ns)
1406 {
1407     ns->mq_queues_count  = 0;
1408     ns->mq_queues_max    = DFLT_QUEUESMAX;
1409     ns->mq_msg_max       = DFLT_MSGMAX;
1410     ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1411     ns->mq_msg_default   = DFLT_MSG;
1412     ns->mq_msgsize_default  = DFLT_MSGSIZE;
1413 
1414     ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1415     if (IS_ERR(ns->mq_mnt)) {
1416         int err = PTR_ERR(ns->mq_mnt);
1417         ns->mq_mnt = NULL;
1418         return err;
1419     }
1420     return 0;
1421 }
1422 
1423 void mq_clear_sbinfo(struct ipc_namespace *ns)
1424 {
1425     ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1426 }
1427 
1428 void mq_put_mnt(struct ipc_namespace *ns)
1429 {
1430     kern_unmount(ns->mq_mnt);
1431 }
1432 
1433 static int __init init_mqueue_fs(void)
1434 {
1435     int error;
1436 
1437     mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1438                 sizeof(struct mqueue_inode_info), 0,
1439                 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1440     if (mqueue_inode_cachep == NULL)
1441         return -ENOMEM;
1442 
1443     /* ignore failures - they are not fatal */
1444     mq_sysctl_table = mq_register_sysctl_table();
1445 
1446     error = register_filesystem(&mqueue_fs_type);
1447     if (error)
1448         goto out_sysctl;
1449 
1450     spin_lock_init(&mq_lock);
1451 
1452     error = mq_init_ns(&init_ipc_ns);
1453     if (error)
1454         goto out_filesystem;
1455 
1456     return 0;
1457 
1458 out_filesystem:
1459     unregister_filesystem(&mqueue_fs_type);
1460 out_sysctl:
1461     if (mq_sysctl_table)
1462         unregister_sysctl_table(mq_sysctl_table);
1463     kmem_cache_destroy(mqueue_inode_cachep);
1464     return error;
1465 }
1466 
1467 device_initcall(init_mqueue_fs);