]> nv-tegra.nvidia Code Review - linux-3.10.git/blobdiff - ipc/shm.c
ipc: clamp with min()
[linux-3.10.git] / ipc / shm.c
index 867e5d6a55c23ec64361db7e9eb2e1cf93b4f559..cb858df061d37185cf96b7dadd6fe8d5d3e2b869 100644 (file)
--- a/ipc/shm.c
+++ b/ipc/shm.c
@@ -54,7 +54,7 @@ struct shm_file_data {
 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
 
 static const struct file_operations shm_file_operations;
-static struct vm_operations_struct shm_vm_ops;
+static const struct vm_operations_struct shm_vm_ops;
 
 #define shm_ids(ns)    ((ns)->ids[IPC_SHM_IDS])
 
@@ -74,8 +74,9 @@ void shm_init_ns(struct ipc_namespace *ns)
        ns->shm_ctlmax = SHMMAX;
        ns->shm_ctlall = SHMALL;
        ns->shm_ctlmni = SHMMNI;
+       ns->shm_rmid_forced = 0;
        ns->shm_tot = 0;
-       ipc_init_ids(&ns->ids[IPC_SHM_IDS]);
+       ipc_init_ids(&shm_ids(ns));
 }
 
 /*
@@ -100,14 +101,26 @@ static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
 void shm_exit_ns(struct ipc_namespace *ns)
 {
        free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
+       idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 }
 #endif
 
-void __init shm_init (void)
+static int __init ipc_ns_init(void)
 {
        shm_init_ns(&init_ipc_ns);
+       return 0;
+}
+
+pure_initcall(ipc_ns_init);
+
+void __init shm_init (void)
+{
        ipc_init_proc_interface("sysvipc/shm",
-                               "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
+#if BITS_PER_LONG <= 32
+                               "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
+#else
+                               "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
+#endif
                                IPC_SHM_IDS, sysvipc_shm_proc_show);
 }
 
@@ -125,6 +138,12 @@ static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
        return container_of(ipcp, struct shmid_kernel, shm_perm);
 }
 
+static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
+{
+       rcu_read_lock();
+       spin_lock(&ipcp->shm_perm.lock);
+}
+
 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
                                                int id)
 {
@@ -173,14 +192,31 @@ static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
        shm_unlock(shp);
        if (!is_file_hugepages(shp->shm_file))
                shmem_lock(shp->shm_file, 0, shp->mlock_user);
-       else
-               user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
+       else if (shp->mlock_user)
+               user_shm_unlock(file_inode(shp->shm_file)->i_size,
                                                shp->mlock_user);
        fput (shp->shm_file);
        security_shm_free(shp);
        ipc_rcu_putref(shp);
 }
 
+/*
+ * shm_may_destroy - identifies whether shm segment should be destroyed now
+ *
+ * Returns true if and only if there are no active users of the segment and
+ * one of the following is true:
+ *
+ * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
+ *
+ * 2) sysctl kernel.shm_rmid_forced is set to 1.
+ */
+static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
+{
+       return (shp->shm_nattch == 0) &&
+              (ns->shm_rmid_forced ||
+               (shp->shm_perm.mode & SHM_DEST));
+}
+
 /*
  * remove the attach descriptor vma.
  * free memory for segment if it is marked destroyed.
@@ -201,14 +237,90 @@ static void shm_close(struct vm_area_struct *vma)
        shp->shm_lprid = task_tgid_vnr(current);
        shp->shm_dtim = get_seconds();
        shp->shm_nattch--;
-       if(shp->shm_nattch == 0 &&
-          shp->shm_perm.mode & SHM_DEST)
+       if (shm_may_destroy(ns, shp))
                shm_destroy(ns, shp);
        else
                shm_unlock(shp);
        up_write(&shm_ids(ns).rw_mutex);
 }
 
+/* Called with ns->shm_ids(ns).rw_mutex locked */
+static int shm_try_destroy_current(int id, void *p, void *data)
+{
+       struct ipc_namespace *ns = data;
+       struct kern_ipc_perm *ipcp = p;
+       struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
+
+       if (shp->shm_creator != current)
+               return 0;
+
+       /*
+        * Mark it as orphaned to destroy the segment when
+        * kernel.shm_rmid_forced is changed.
+        * It is noop if the following shm_may_destroy() returns true.
+        */
+       shp->shm_creator = NULL;
+
+       /*
+        * Don't even try to destroy it.  If shm_rmid_forced=0 and IPC_RMID
+        * is not set, it shouldn't be deleted here.
+        */
+       if (!ns->shm_rmid_forced)
+               return 0;
+
+       if (shm_may_destroy(ns, shp)) {
+               shm_lock_by_ptr(shp);
+               shm_destroy(ns, shp);
+       }
+       return 0;
+}
+
+/* Called with ns->shm_ids(ns).rw_mutex locked */
+static int shm_try_destroy_orphaned(int id, void *p, void *data)
+{
+       struct ipc_namespace *ns = data;
+       struct kern_ipc_perm *ipcp = p;
+       struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
+
+       /*
+        * We want to destroy segments without users and with already
+        * exit'ed originating process.
+        *
+        * As shp->* are changed under rw_mutex, it's safe to skip shp locking.
+        */
+       if (shp->shm_creator != NULL)
+               return 0;
+
+       if (shm_may_destroy(ns, shp)) {
+               shm_lock_by_ptr(shp);
+               shm_destroy(ns, shp);
+       }
+       return 0;
+}
+
+void shm_destroy_orphaned(struct ipc_namespace *ns)
+{
+       down_write(&shm_ids(ns).rw_mutex);
+       if (shm_ids(ns).in_use)
+               idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
+       up_write(&shm_ids(ns).rw_mutex);
+}
+
+
+void exit_shm(struct task_struct *task)
+{
+       struct ipc_namespace *ns = task->nsproxy->ipc_ns;
+
+       if (shm_ids(ns).in_use == 0)
+               return;
+
+       /* Destroy all already created segments, but not mapped yet */
+       down_write(&shm_ids(ns).rw_mutex);
+       if (shm_ids(ns).in_use)
+               idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
+       up_write(&shm_ids(ns).rw_mutex);
+}
+
 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
        struct file *file = vma->vm_file;
@@ -272,46 +384,60 @@ static int shm_release(struct inode *ino, struct file *file)
        return 0;
 }
 
-static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
+static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 {
-       int (*fsync) (struct file *, struct dentry *, int datasync);
        struct shm_file_data *sfd = shm_file_data(file);
-       int ret = -EINVAL;
 
-       fsync = sfd->file->f_op->fsync;
-       if (fsync)
-               ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
-       return ret;
+       if (!sfd->file->f_op->fsync)
+               return -EINVAL;
+       return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 }
 
-static unsigned long shm_get_unmapped_area(struct file *file,
-       unsigned long addr, unsigned long len, unsigned long pgoff,
-       unsigned long flags)
+static long shm_fallocate(struct file *file, int mode, loff_t offset,
+                         loff_t len)
 {
        struct shm_file_data *sfd = shm_file_data(file);
-       return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
+
+       if (!sfd->file->f_op->fallocate)
+               return -EOPNOTSUPP;
+       return sfd->file->f_op->fallocate(file, mode, offset, len);
 }
 
-int is_file_shm_hugepages(struct file *file)
+static unsigned long shm_get_unmapped_area(struct file *file,
+       unsigned long addr, unsigned long len, unsigned long pgoff,
+       unsigned long flags)
 {
-       int ret = 0;
-
-       if (file->f_op == &shm_file_operations) {
-               struct shm_file_data *sfd;
-               sfd = shm_file_data(file);
-               ret = is_file_hugepages(sfd->file);
-       }
-       return ret;
+       struct shm_file_data *sfd = shm_file_data(file);
+       return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
+                                               pgoff, flags);
 }
 
 static const struct file_operations shm_file_operations = {
        .mmap           = shm_mmap,
        .fsync          = shm_fsync,
        .release        = shm_release,
+#ifndef CONFIG_MMU
+       .get_unmapped_area      = shm_get_unmapped_area,
+#endif
+       .llseek         = noop_llseek,
+       .fallocate      = shm_fallocate,
+};
+
+static const struct file_operations shm_file_operations_huge = {
+       .mmap           = shm_mmap,
+       .fsync          = shm_fsync,
+       .release        = shm_release,
        .get_unmapped_area      = shm_get_unmapped_area,
+       .llseek         = noop_llseek,
+       .fallocate      = shm_fallocate,
 };
 
-static struct vm_operations_struct shm_vm_ops = {
+int is_file_shm_hugepages(struct file *file)
+{
+       return file->f_op == &shm_file_operations_huge;
+}
+
+static const struct vm_operations_struct shm_vm_ops = {
        .open   = shm_open,     /* callback for a new vm-area open */
        .close  = shm_close,    /* callback for when the vm-area is released */
        .fault  = shm_fault,
@@ -340,6 +466,7 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
        struct file * file;
        char name[13];
        int id;
+       vm_flags_t acctflag = 0;
 
        if (size < SHMMIN || size > ns->shm_ctlmax)
                return -EINVAL;
@@ -364,18 +491,20 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 
        sprintf (name, "SYSV%08x", key);
        if (shmflg & SHM_HUGETLB) {
-               /* hugetlb_file_setup takes care of mlock user accounting */
-               file = hugetlb_file_setup(name, size);
-               shp->mlock_user = current->user;
+               /* hugetlb_file_setup applies strict accounting */
+               if (shmflg & SHM_NORESERVE)
+                       acctflag = VM_NORESERVE;
+               file = hugetlb_file_setup(name, 0, size, acctflag,
+                                 &shp->mlock_user, HUGETLB_SHMFS_INODE,
+                               (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
        } else {
-               int acctflag = VM_ACCOUNT;
                /*
                 * Do not allow no accounting for OVERCOMMIT_NEVER, even
                 * if it's asked for.
                 */
                if  ((shmflg & SHM_NORESERVE) &&
                                sysctl_overcommit_memory != OVERCOMMIT_NEVER)
-                       acctflag = 0;
+                       acctflag = VM_NORESERVE;
                file = shmem_file_setup(name, size, acctflag);
        }
        error = PTR_ERR(file);
@@ -395,11 +524,12 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
        shp->shm_segsz = size;
        shp->shm_nattch = 0;
        shp->shm_file = file;
+       shp->shm_creator = current;
        /*
         * shmid gets reported as "inode#" in /proc/pid/maps.
         * proc-ps tools use this. Changing this will break them.
         */
-       file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
+       file_inode(file)->i_ino = shp->shm_perm.id;
 
        ns->shm_tot += numpages;
        error = shp->shm_perm.id;
@@ -407,6 +537,8 @@ static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
        return error;
 
 no_id:
+       if (is_file_hugepages(file) && shp->mlock_user)
+               user_shm_unlock(size, shp->mlock_user);
        fput(file);
 no_file:
        security_shm_free(shp);
@@ -440,7 +572,7 @@ static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
        return 0;
 }
 
-asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
+SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 {
        struct ipc_namespace *ns;
        struct ipc_ops shm_ops;
@@ -468,6 +600,7 @@ static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_
            {
                struct shmid_ds out;
 
+               memset(&out, 0, sizeof(out));
                ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
                out.shm_segsz   = in->shm_segsz;
                out.shm_atime   = in->shm_atime;
@@ -536,6 +669,34 @@ static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminf
        }
 }
 
+/*
+ * Calculate and add used RSS and swap pages of a shm.
+ * Called with shm_ids.rw_mutex held as a reader
+ */
+static void shm_add_rss_swap(struct shmid_kernel *shp,
+       unsigned long *rss_add, unsigned long *swp_add)
+{
+       struct inode *inode;
+
+       inode = file_inode(shp->shm_file);
+
+       if (is_file_hugepages(shp->shm_file)) {
+               struct address_space *mapping = inode->i_mapping;
+               struct hstate *h = hstate_file(shp->shm_file);
+               *rss_add += pages_per_huge_page(h) * mapping->nrpages;
+       } else {
+#ifdef CONFIG_SHMEM
+               struct shmem_inode_info *info = SHMEM_I(inode);
+               spin_lock(&info->lock);
+               *rss_add += inode->i_mapping->nrpages;
+               *swp_add += info->swapped;
+               spin_unlock(&info->lock);
+#else
+               *rss_add += inode->i_mapping->nrpages;
+#endif
+       }
+}
+
 /*
  * Called with shm_ids.rw_mutex held as a reader
  */
@@ -551,26 +712,15 @@ static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
        in_use = shm_ids(ns).in_use;
 
        for (total = 0, next_id = 0; total < in_use; next_id++) {
+               struct kern_ipc_perm *ipc;
                struct shmid_kernel *shp;
-               struct inode *inode;
 
-               shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
-               if (shp == NULL)
+               ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
+               if (ipc == NULL)
                        continue;
+               shp = container_of(ipc, struct shmid_kernel, shm_perm);
 
-               inode = shp->shm_file->f_path.dentry->d_inode;
-
-               if (is_file_hugepages(shp->shm_file)) {
-                       struct address_space *mapping = inode->i_mapping;
-                       struct hstate *h = hstate_file(shp->shm_file);
-                       *rss += pages_per_huge_page(h) * mapping->nrpages;
-               } else {
-                       struct shmem_inode_info *info = SHMEM_I(inode);
-                       spin_lock(&info->lock);
-                       *rss += inode->i_mapping->nrpages;
-                       *swp += info->swapped;
-                       spin_unlock(&info->lock);
-               }
+               shm_add_rss_swap(shp, rss, swp);
 
                total++;
        }
@@ -594,7 +744,8 @@ static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
                        return -EFAULT;
        }
 
-       ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
+       ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
+                              &shmid64.shm_perm, 0);
        if (IS_ERR(ipcp))
                return PTR_ERR(ipcp);
 
@@ -608,7 +759,9 @@ static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
                do_shm_rmid(ns, ipcp);
                goto out_up;
        case IPC_SET:
-               ipc_update_perm(&shmid64.shm_perm, ipcp);
+               err = ipc_update_perm(&shmid64.shm_perm, ipcp);
+               if (err)
+                       goto out_unlock;
                shp->shm_ctim = get_seconds();
                break;
        default:
@@ -621,7 +774,7 @@ out_up:
        return err;
 }
 
-asmlinkage long sys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
+SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
 {
        struct shmid_kernel *shp;
        int err, version;
@@ -644,7 +797,7 @@ asmlinkage long sys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
                if (err)
                        return err;
 
-               memset(&shminfo,0,sizeof(shminfo));
+               memset(&shminfo, 0, sizeof(shminfo));
                shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
                shminfo.shmmax = ns->shm_ctlmax;
                shminfo.shmall = ns->shm_ctlall;
@@ -669,7 +822,7 @@ asmlinkage long sys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
                if (err)
                        return err;
 
-               memset(&shm_info,0,sizeof(shm_info));
+               memset(&shm_info, 0, sizeof(shm_info));
                down_read(&shm_ids(ns).rw_mutex);
                shm_info.used_ids = shm_ids(ns).in_use;
                shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
@@ -678,7 +831,7 @@ asmlinkage long sys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
                shm_info.swap_successes = 0;
                err = ipc_get_maxid(&shm_ids(ns));
                up_read(&shm_ids(ns).rw_mutex);
-               if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
+               if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
                        err = -EFAULT;
                        goto out;
                }
@@ -692,11 +845,6 @@ asmlinkage long sys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
                struct shmid64_ds tbuf;
                int result;
 
-               if (!buf) {
-                       err = -EFAULT;
-                       goto out;
-               }
-
                if (cmd == SHM_STAT) {
                        shp = shm_lock(ns, shmid);
                        if (IS_ERR(shp)) {
@@ -712,8 +860,8 @@ asmlinkage long sys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
                        }
                        result = 0;
                }
-               err=-EACCES;
-               if (ipcperms (&shp->shm_perm, S_IRUGO))
+               err = -EACCES;
+               if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
                        goto out_unlock;
                err = security_shm_shmctl(shp, cmd);
                if (err)
@@ -737,9 +885,7 @@ asmlinkage long sys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
        case SHM_LOCK:
        case SHM_UNLOCK:
        {
-               struct file *uninitialized_var(shm_file);
-
-               lru_add_drain_all();  /* drain pagevecs to lru lists */
+               struct file *shm_file;
 
                shp = shm_lock_check(ns, shmid);
                if (IS_ERR(shp)) {
@@ -747,39 +893,46 @@ asmlinkage long sys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
                        goto out;
                }
 
-               err = audit_ipc_obj(&(shp->shm_perm));
-               if (err)
-                       goto out_unlock;
+               audit_ipc_obj(&(shp->shm_perm));
 
-               if (!capable(CAP_IPC_LOCK)) {
+               if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
+                       kuid_t euid = current_euid();
                        err = -EPERM;
-                       if (current->euid != shp->shm_perm.uid &&
-                           current->euid != shp->shm_perm.cuid)
+                       if (!uid_eq(euid, shp->shm_perm.uid) &&
+                           !uid_eq(euid, shp->shm_perm.cuid))
                                goto out_unlock;
-                       if (cmd == SHM_LOCK &&
-                           !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
+                       if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
                                goto out_unlock;
                }
 
                err = security_shm_shmctl(shp, cmd);
                if (err)
                        goto out_unlock;
-               
-               if(cmd==SHM_LOCK) {
-                       struct user_struct * user = current->user;
-                       if (!is_file_hugepages(shp->shm_file)) {
-                               err = shmem_lock(shp->shm_file, 1, user);
-                               if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
-                                       shp->shm_perm.mode |= SHM_LOCKED;
-                                       shp->mlock_user = user;
-                               }
+
+               shm_file = shp->shm_file;
+               if (is_file_hugepages(shm_file))
+                       goto out_unlock;
+
+               if (cmd == SHM_LOCK) {
+                       struct user_struct *user = current_user();
+                       err = shmem_lock(shm_file, 1, user);
+                       if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
+                               shp->shm_perm.mode |= SHM_LOCKED;
+                               shp->mlock_user = user;
                        }
-               } else if (!is_file_hugepages(shp->shm_file)) {
-                       shmem_lock(shp->shm_file, 0, shp->mlock_user);
-                       shp->shm_perm.mode &= ~SHM_LOCKED;
-                       shp->mlock_user = NULL;
+                       goto out_unlock;
                }
+
+               /* SHM_UNLOCK */
+               if (!(shp->shm_perm.mode & SHM_LOCKED))
+                       goto out_unlock;
+               shmem_lock(shm_file, 0, shp->mlock_user);
+               shp->shm_perm.mode &= ~SHM_LOCKED;
+               shp->mlock_user = NULL;
+               get_file(shm_file);
                shm_unlock(shp);
+               shmem_unlock_mapping(shm_file->f_mapping);
+               fput(shm_file);
                goto out;
        }
        case IPC_RMID:
@@ -803,7 +956,8 @@ out:
  * "raddr" thing points to kernel space, and there has to be a wrapper around
  * this.
  */
-long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
+long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
+             unsigned long shmlba)
 {
        struct shmid_kernel *shp;
        unsigned long addr;
@@ -813,19 +967,19 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
        unsigned long flags;
        unsigned long prot;
        int acc_mode;
-       unsigned long user_addr;
        struct ipc_namespace *ns;
        struct shm_file_data *sfd;
        struct path path;
        fmode_t f_mode;
+       unsigned long populate = 0;
 
        err = -EINVAL;
        if (shmid < 0)
                goto out;
        else if ((addr = (ulong)shmaddr)) {
-               if (addr & (SHMLBA-1)) {
+               if (addr & (shmlba - 1)) {
                        if (shmflg & SHM_RND)
-                               addr &= ~(SHMLBA-1);       /* round down */
+                               addr &= ~(shmlba - 1);     /* round down */
                        else
 #ifndef __ARCH_FORCE_SHMLBA
                                if (addr & ~PAGE_MASK)
@@ -866,15 +1020,15 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
        }
 
        err = -EACCES;
-       if (ipcperms(&shp->shm_perm, acc_mode))
+       if (ipcperms(ns, &shp->shm_perm, acc_mode))
                goto out_unlock;
 
        err = security_shm_shmat(shp, shmaddr, shmflg);
        if (err)
                goto out_unlock;
 
-       path.dentry = dget(shp->shm_file->f_path.dentry);
-       path.mnt    = shp->shm_file->f_path.mnt;
+       path = shp->shm_file->f_path;
+       path_get(&path);
        shp->shm_nattch++;
        size = i_size_read(path.dentry->d_inode);
        shm_unlock(shp);
@@ -884,8 +1038,12 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
        if (!sfd)
                goto out_put_dentry;
 
-       file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
-       if (!file)
+       file = alloc_file(&path, f_mode,
+                         is_file_hugepages(shp->shm_file) ?
+                               &shm_file_operations_huge :
+                               &shm_file_operations);
+       err = PTR_ERR(file);
+       if (IS_ERR(file))
                goto out_free;
 
        file->private_data = sfd;
@@ -895,6 +1053,10 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
        sfd->file = shp->shm_file;
        sfd->vm_ops = NULL;
 
+       err = security_mmap_file(file, prot, flags);
+       if (err)
+               goto out_fput;
+
        down_write(&current->mm->mmap_sem);
        if (addr && !(shmflg & SHM_REMAP)) {
                err = -EINVAL;
@@ -909,14 +1071,17 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
                        goto invalid;
        }
                
-       user_addr = do_mmap (file, addr, size, prot, flags, 0);
-       *raddr = user_addr;
+       addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
+       *raddr = addr;
        err = 0;
-       if (IS_ERR_VALUE(user_addr))
-               err = (long)user_addr;
+       if (IS_ERR_VALUE(addr))
+               err = (long)addr;
 invalid:
        up_write(&current->mm->mmap_sem);
+       if (populate)
+               mm_populate(addr, populate);
 
+out_fput:
        fput(file);
 
 out_nattch:
@@ -924,8 +1089,7 @@ out_nattch:
        shp = shm_lock(ns, shmid);
        BUG_ON(IS_ERR(shp));
        shp->shm_nattch--;
-       if(shp->shm_nattch == 0 &&
-          shp->shm_perm.mode & SHM_DEST)
+       if (shm_may_destroy(ns, shp))
                shm_destroy(ns, shp);
        else
                shm_unlock(shp);
@@ -941,16 +1105,16 @@ out_unlock:
 out_free:
        kfree(sfd);
 out_put_dentry:
-       dput(path.dentry);
+       path_put(&path);
        goto out_nattch;
 }
 
-asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
+SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
 {
        unsigned long ret;
        long err;
 
-       err = do_shmat(shmid, shmaddr, shmflg, &ret);
+       err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
        if (err)
                return err;
        force_successful_syscall_return();
@@ -961,13 +1125,16 @@ asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
  * detach and kill segment if marked destroyed.
  * The work is done in shm_close.
  */
-asmlinkage long sys_shmdt(char __user *shmaddr)
+SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
 {
        struct mm_struct *mm = current->mm;
-       struct vm_area_struct *vma, *next;
+       struct vm_area_struct *vma;
        unsigned long addr = (unsigned long)shmaddr;
-       loff_t size = 0;
        int retval = -EINVAL;
+#ifdef CONFIG_MMU
+       loff_t size = 0;
+       struct vm_area_struct *next;
+#endif
 
        if (addr & ~PAGE_MASK)
                return retval;
@@ -996,6 +1163,7 @@ asmlinkage long sys_shmdt(char __user *shmaddr)
         */
        vma = find_vma(mm, addr);
 
+#ifdef CONFIG_MMU
        while (vma) {
                next = vma->vm_next;
 
@@ -1008,7 +1176,7 @@ asmlinkage long sys_shmdt(char __user *shmaddr)
                        (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 
 
-                       size = vma->vm_file->f_path.dentry->d_inode->i_size;
+                       size = file_inode(vma->vm_file)->i_size;
                        do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
                        /*
                         * We discovered the size of the shm segment, so
@@ -1026,7 +1194,7 @@ asmlinkage long sys_shmdt(char __user *shmaddr)
        /*
         * We need look no further than the maximum address a fragment
         * could possibly have landed at. Also cast things to loff_t to
-        * prevent overflows and make comparisions vs. equal-width types.
+        * prevent overflows and make comparisons vs. equal-width types.
         */
        size = PAGE_ALIGN(size);
        while (vma && (loff_t)(vma->vm_end - addr) <= size) {
@@ -1040,6 +1208,17 @@ asmlinkage long sys_shmdt(char __user *shmaddr)
                vma = next;
        }
 
+#else /* CONFIG_MMU */
+       /* under NOMMU conditions, the exact address to be destroyed must be
+        * given */
+       retval = -EINVAL;
+       if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
+               do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
+               retval = 0;
+       }
+
+#endif
+
        up_write(&mm->mmap_sem);
        return retval;
 }
@@ -1047,7 +1226,11 @@ asmlinkage long sys_shmdt(char __user *shmaddr)
 #ifdef CONFIG_PROC_FS
 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
 {
+       struct user_namespace *user_ns = seq_user_ns(s);
        struct shmid_kernel *shp = it;
+       unsigned long rss = 0, swp = 0;
+
+       shm_add_rss_swap(shp, &rss, &swp);
 
 #if BITS_PER_LONG <= 32
 #define SIZE_SPEC "%10lu"
@@ -1057,7 +1240,8 @@ static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
 
        return seq_printf(s,
                          "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
-                         "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
+                         "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
+                         SIZE_SPEC " " SIZE_SPEC "\n",
                          shp->shm_perm.key,
                          shp->shm_perm.id,
                          shp->shm_perm.mode,
@@ -1065,12 +1249,14 @@ static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
                          shp->shm_cprid,
                          shp->shm_lprid,
                          shp->shm_nattch,
-                         shp->shm_perm.uid,
-                         shp->shm_perm.gid,
-                         shp->shm_perm.cuid,
-                         shp->shm_perm.cgid,
+                         from_kuid_munged(user_ns, shp->shm_perm.uid),
+                         from_kgid_munged(user_ns, shp->shm_perm.gid),
+                         from_kuid_munged(user_ns, shp->shm_perm.cuid),
+                         from_kgid_munged(user_ns, shp->shm_perm.cgid),
                          shp->shm_atim,
                          shp->shm_dtim,
-                         shp->shm_ctim);
+                         shp->shm_ctim,
+                         rss * PAGE_SIZE,
+                         swp * PAGE_SIZE);
 }
 #endif