ipc: use simple_read_from_buffer()
[linux-2.6.git] / ipc / shm.c
index 940b0c9..e77ec69 100644 (file)
--- a/ipc/shm.c
+++ b/ipc/shm.c
  *
  * support for audit of ipc object properties and permission changes
  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
+ *
+ * namespaces support
+ * OpenVZ, SWsoft Inc.
+ * Pavel Emelianov <xemul@openvz.org>
  */
 
 #include <linux/slab.h>
 #include <linux/capability.h>
 #include <linux/ptrace.h>
 #include <linux/seq_file.h>
-#include <linux/mutex.h>
+#include <linux/rwsem.h>
+#include <linux/nsproxy.h>
+#include <linux/mount.h>
+#include <linux/ipc_namespace.h>
 
 #include <asm/uaccess.h>
 
 #include "util.h"
 
-static struct file_operations shm_file_operations;
+struct shm_file_data {
+       int id;
+       struct ipc_namespace *ns;
+       struct file *file;
+       const struct vm_operations_struct *vm_ops;
+};
+
+#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 struct ipc_ids shm_ids;
+#define shm_ids(ns)    ((ns)->ids[IPC_SHM_IDS])
 
-#define shm_lock(id)   ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
-#define shm_unlock(shp)        ipc_unlock(&(shp)->shm_perm)
-#define shm_get(id)    ((struct shmid_kernel*)ipc_get(&shm_ids,id))
-#define shm_buildid(id, seq) \
-       ipc_buildid(&shm_ids, id, seq)
+#define shm_unlock(shp)                        \
+       ipc_unlock(&(shp)->shm_perm)
 
-static int newseg (key_t key, int shmflg, size_t size);
-static void shm_open (struct vm_area_struct *shmd);
-static void shm_close (struct vm_area_struct *shmd);
+static int newseg(struct ipc_namespace *, struct ipc_params *);
+static void shm_open(struct vm_area_struct *vma);
+static void shm_close(struct vm_area_struct *vma);
+static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
 #ifdef CONFIG_PROC_FS
 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
 #endif
 
-size_t shm_ctlmax = SHMMAX;
-size_t         shm_ctlall = SHMALL;
-int    shm_ctlmni = SHMMNI;
+void shm_init_ns(struct ipc_namespace *ns)
+{
+       ns->shm_ctlmax = SHMMAX;
+       ns->shm_ctlall = SHMALL;
+       ns->shm_ctlmni = SHMMNI;
+       ns->shm_tot = 0;
+       ipc_init_ids(&ns->ids[IPC_SHM_IDS]);
+}
+
+/*
+ * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
+ * Only shm_ids.rw_mutex remains locked on exit.
+ */
+static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
+{
+       struct shmid_kernel *shp;
+       shp = container_of(ipcp, struct shmid_kernel, shm_perm);
+
+       if (shp->shm_nattch){
+               shp->shm_perm.mode |= SHM_DEST;
+               /* Do not find it any more */
+               shp->shm_perm.key = IPC_PRIVATE;
+               shm_unlock(shp);
+       } else
+               shm_destroy(ns, shp);
+}
 
-static int shm_tot; /* total number of shared memory pages */
+#ifdef CONFIG_IPC_NS
+void shm_exit_ns(struct ipc_namespace *ns)
+{
+       free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
+}
+#endif
 
 void __init shm_init (void)
 {
-       ipc_init_ids(&shm_ids, 1);
+       shm_init_ns(&init_ipc_ns);
        ipc_init_proc_interface("sysvipc/shm",
                                "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
-                               &shm_ids,
-                               sysvipc_shm_proc_show);
+                               IPC_SHM_IDS, sysvipc_shm_proc_show);
 }
 
-static inline int shm_checkid(struct shmid_kernel *s, int id)
+/*
+ * shm_lock_(check_) routines are called in the paths where the rw_mutex
+ * is not necessarily held.
+ */
+static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 {
-       if (ipc_checkid(&shm_ids,&s->shm_perm,id))
-               return -EIDRM;
-       return 0;
+       struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
+
+       if (IS_ERR(ipcp))
+               return (struct shmid_kernel *)ipcp;
+
+       return container_of(ipcp, struct shmid_kernel, shm_perm);
 }
 
-static inline struct shmid_kernel *shm_rmid(int id)
+static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
+                                               int id)
 {
-       return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
+       struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
+
+       if (IS_ERR(ipcp))
+               return (struct shmid_kernel *)ipcp;
+
+       return container_of(ipcp, struct shmid_kernel, shm_perm);
 }
 
-static inline int shm_addid(struct shmid_kernel *shp)
+static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 {
-       return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
+       ipc_rmid(&shm_ids(ns), &s->shm_perm);
 }
 
 
-
-static inline void shm_inc (int id) {
+/* This is called by fork, once for every shm attach. */
+static void shm_open(struct vm_area_struct *vma)
+{
+       struct file *file = vma->vm_file;
+       struct shm_file_data *sfd = shm_file_data(file);
        struct shmid_kernel *shp;
 
-       shp = shm_lock(id);
-       BUG_ON(!shp);
+       shp = shm_lock(sfd->ns, sfd->id);
+       BUG_ON(IS_ERR(shp));
        shp->shm_atim = get_seconds();
-       shp->shm_lprid = current->tgid;
+       shp->shm_lprid = task_tgid_vnr(current);
        shp->shm_nattch++;
        shm_unlock(shp);
 }
 
-/* This is called by fork, once for every shm attach. */
-static void shm_open (struct vm_area_struct *shmd)
-{
-       shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
-}
-
 /*
  * shm_destroy - free the struct shmid_kernel
  *
+ * @ns: namespace
  * @shp: struct to free
  *
- * It has to be called with shp and shm_ids.mutex locked,
+ * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
  * but returns with shp unlocked and freed.
  */
-static void shm_destroy (struct shmid_kernel *shp)
+static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 {
-       shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
-       shm_rmid (shp->id);
+       ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
+       shm_rmid(ns, 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_dentry->d_inode->i_size,
+               user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
                                                shp->mlock_user);
        fput (shp->shm_file);
        security_shm_free(shp);
@@ -130,66 +182,158 @@ static void shm_destroy (struct shmid_kernel *shp)
 }
 
 /*
- * remove the attach descriptor shmd.
+ * remove the attach descriptor vma.
  * free memory for segment if it is marked destroyed.
  * The descriptor has already been removed from the current->mm->mmap list
  * and will later be kfree()d.
  */
-static void shm_close (struct vm_area_struct *shmd)
+static void shm_close(struct vm_area_struct *vma)
 {
-       struct file * file = shmd->vm_file;
-       int id = file->f_dentry->d_inode->i_ino;
+       struct file * file = vma->vm_file;
+       struct shm_file_data *sfd = shm_file_data(file);
        struct shmid_kernel *shp;
+       struct ipc_namespace *ns = sfd->ns;
 
-       mutex_lock(&shm_ids.mutex);
+       down_write(&shm_ids(ns).rw_mutex);
        /* remove from the list of attaches of the shm segment */
-       shp = shm_lock(id);
-       BUG_ON(!shp);
-       shp->shm_lprid = current->tgid;
+       shp = shm_lock(ns, sfd->id);
+       BUG_ON(IS_ERR(shp));
+       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)
-               shm_destroy (shp);
+               shm_destroy(ns, shp);
        else
                shm_unlock(shp);
-       mutex_unlock(&shm_ids.mutex);
+       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;
+       struct shm_file_data *sfd = shm_file_data(file);
+
+       return sfd->vm_ops->fault(vma, vmf);
+}
+
+#ifdef CONFIG_NUMA
+static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
+{
+       struct file *file = vma->vm_file;
+       struct shm_file_data *sfd = shm_file_data(file);
+       int err = 0;
+       if (sfd->vm_ops->set_policy)
+               err = sfd->vm_ops->set_policy(vma, new);
+       return err;
+}
+
+static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
+                                       unsigned long addr)
+{
+       struct file *file = vma->vm_file;
+       struct shm_file_data *sfd = shm_file_data(file);
+       struct mempolicy *pol = NULL;
+
+       if (sfd->vm_ops->get_policy)
+               pol = sfd->vm_ops->get_policy(vma, addr);
+       else if (vma->vm_policy)
+               pol = vma->vm_policy;
+
+       return pol;
 }
+#endif
 
 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 {
+       struct shm_file_data *sfd = shm_file_data(file);
        int ret;
 
-       ret = shmem_mmap(file, vma);
-       if (ret == 0) {
-               vma->vm_ops = &shm_vm_ops;
-               if (!(vma->vm_flags & VM_WRITE))
-                       vma->vm_flags &= ~VM_MAYWRITE;
-               shm_inc(file->f_dentry->d_inode->i_ino);
-       }
+       ret = sfd->file->f_op->mmap(sfd->file, vma);
+       if (ret != 0)
+               return ret;
+       sfd->vm_ops = vma->vm_ops;
+#ifdef CONFIG_MMU
+       BUG_ON(!sfd->vm_ops->fault);
+#endif
+       vma->vm_ops = &shm_vm_ops;
+       shm_open(vma);
 
        return ret;
 }
 
-static struct file_operations shm_file_operations = {
-       .mmap   = shm_mmap,
-#ifndef CONFIG_MMU
-       .get_unmapped_area = shmem_get_unmapped_area,
-#endif
+static int shm_release(struct inode *ino, struct file *file)
+{
+       struct shm_file_data *sfd = shm_file_data(file);
+
+       put_ipc_ns(sfd->ns);
+       shm_file_data(file) = NULL;
+       kfree(sfd);
+       return 0;
+}
+
+static int shm_fsync(struct file *file, struct dentry *dentry, 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;
+}
+
+static unsigned long shm_get_unmapped_area(struct file *file,
+       unsigned long addr, unsigned long len, unsigned long pgoff,
+       unsigned long flags)
+{
+       struct shm_file_data *sfd = shm_file_data(file);
+       return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
+}
+
+int is_file_shm_hugepages(struct file *file)
+{
+       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;
+}
+
+static const struct file_operations shm_file_operations = {
+       .mmap           = shm_mmap,
+       .fsync          = shm_fsync,
+       .release        = shm_release,
+       .get_unmapped_area      = shm_get_unmapped_area,
 };
 
 static 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 */
-       .nopage = shmem_nopage,
-#if defined(CONFIG_NUMA) && defined(CONFIG_SHMEM)
-       .set_policy = shmem_set_policy,
-       .get_policy = shmem_get_policy,
+       .fault  = shm_fault,
+#if defined(CONFIG_NUMA)
+       .set_policy = shm_set_policy,
+       .get_policy = shm_get_policy,
 #endif
 };
 
-static int newseg (key_t key, int shmflg, size_t size)
+/**
+ * newseg - Create a new shared memory segment
+ * @ns: namespace
+ * @params: ptr to the structure that contains key, size and shmflg
+ *
+ * Called with shm_ids.rw_mutex held as a writer.
+ */
+
+static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 {
+       key_t key = params->key;
+       int shmflg = params->flg;
+       size_t size = params->u.size;
        int error;
        struct shmid_kernel *shp;
        int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
@@ -197,10 +341,10 @@ static int newseg (key_t key, int shmflg, size_t size)
        char name[13];
        int id;
 
-       if (size < SHMMIN || size > shm_ctlmax)
+       if (size < SHMMIN || size > ns->shm_ctlmax)
                return -EINVAL;
 
-       if (shm_tot + numpages >= shm_ctlall)
+       if (ns->shm_tot + numpages > ns->shm_ctlall)
                return -ENOSPC;
 
        shp = ipc_rcu_alloc(sizeof(*shp));
@@ -218,9 +362,10 @@ static int newseg (key_t key, int shmflg, size_t size)
                return error;
        }
 
+       sprintf (name, "SYSV%08x", key);
        if (shmflg & SHM_HUGETLB) {
-               /* hugetlb_zero_setup takes care of mlock user accounting */
-               file = hugetlb_zero_setup(size);
+               /* hugetlb_file_setup takes care of mlock user accounting */
+               file = hugetlb_file_setup(name, size);
                shp->mlock_user = current->user;
        } else {
                int acctflag = VM_ACCOUNT;
@@ -231,35 +376,35 @@ static int newseg (key_t key, int shmflg, size_t size)
                if  ((shmflg & SHM_NORESERVE) &&
                                sysctl_overcommit_memory != OVERCOMMIT_NEVER)
                        acctflag = 0;
-               sprintf (name, "SYSV%08x", key);
                file = shmem_file_setup(name, size, acctflag);
        }
        error = PTR_ERR(file);
        if (IS_ERR(file))
                goto no_file;
 
-       error = -ENOSPC;
-       id = shm_addid(shp);
-       if(id == -1) 
+       id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
+       if (id < 0) {
+               error = id;
                goto no_id;
+       }
 
-       shp->shm_cprid = current->tgid;
+       shp->shm_cprid = task_tgid_vnr(current);
        shp->shm_lprid = 0;
        shp->shm_atim = shp->shm_dtim = 0;
        shp->shm_ctim = get_seconds();
        shp->shm_segsz = size;
        shp->shm_nattch = 0;
-       shp->id = shm_buildid(id,shp->shm_perm.seq);
        shp->shm_file = file;
-       file->f_dentry->d_inode->i_ino = shp->id;
-
-       /* Hugetlb ops would have already been assigned. */
-       if (!(shmflg & SHM_HUGETLB))
-               file->f_op = &shm_file_operations;
+       /*
+        * 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;
 
-       shm_tot += numpages;
+       ns->shm_tot += numpages;
+       error = shp->shm_perm.id;
        shm_unlock(shp);
-       return shp->id;
+       return error;
 
 no_id:
        fput(file);
@@ -269,39 +414,49 @@ no_file:
        return error;
 }
 
-asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
+/*
+ * Called with shm_ids.rw_mutex and ipcp locked.
+ */
+static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 {
        struct shmid_kernel *shp;
-       int err, id = 0;
-
-       mutex_lock(&shm_ids.mutex);
-       if (key == IPC_PRIVATE) {
-               err = newseg(key, shmflg, size);
-       } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
-               if (!(shmflg & IPC_CREAT))
-                       err = -ENOENT;
-               else
-                       err = newseg(key, shmflg, size);
-       } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
-               err = -EEXIST;
-       } else {
-               shp = shm_lock(id);
-               BUG_ON(shp==NULL);
-               if (shp->shm_segsz < size)
-                       err = -EINVAL;
-               else if (ipcperms(&shp->shm_perm, shmflg))
-                       err = -EACCES;
-               else {
-                       int shmid = shm_buildid(id, shp->shm_perm.seq);
-                       err = security_shm_associate(shp, shmflg);
-                       if (!err)
-                               err = shmid;
-               }
-               shm_unlock(shp);
-       }
-       mutex_unlock(&shm_ids.mutex);
 
-       return err;
+       shp = container_of(ipcp, struct shmid_kernel, shm_perm);
+       return security_shm_associate(shp, shmflg);
+}
+
+/*
+ * Called with shm_ids.rw_mutex and ipcp locked.
+ */
+static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
+                               struct ipc_params *params)
+{
+       struct shmid_kernel *shp;
+
+       shp = container_of(ipcp, struct shmid_kernel, shm_perm);
+       if (shp->shm_segsz < params->u.size)
+               return -EINVAL;
+
+       return 0;
+}
+
+asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
+{
+       struct ipc_namespace *ns;
+       struct ipc_ops shm_ops;
+       struct ipc_params shm_params;
+
+       ns = current->nsproxy->ipc_ns;
+
+       shm_ops.getnew = newseg;
+       shm_ops.associate = shm_security;
+       shm_ops.more_checks = shm_more_checks;
+
+       shm_params.key = key;
+       shm_params.flg = shmflg;
+       shm_params.u.size = size;
+
+       return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 }
 
 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
@@ -329,28 +484,14 @@ static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_
        }
 }
 
-struct shm_setbuf {
-       uid_t   uid;
-       gid_t   gid;
-       mode_t  mode;
-};     
-
-static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
+static inline unsigned long
+copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 {
        switch(version) {
        case IPC_64:
-           {
-               struct shmid64_ds tbuf;
-
-               if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
+               if (copy_from_user(out, buf, sizeof(*out)))
                        return -EFAULT;
-
-               out->uid        = tbuf.shm_perm.uid;
-               out->gid        = tbuf.shm_perm.gid;
-               out->mode       = tbuf.shm_perm.mode;
-
                return 0;
-           }
        case IPC_OLD:
            {
                struct shmid_ds tbuf_old;
@@ -358,9 +499,9 @@ static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __
                if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
                        return -EFAULT;
 
-               out->uid        = tbuf_old.shm_perm.uid;
-               out->gid        = tbuf_old.shm_perm.gid;
-               out->mode       = tbuf_old.shm_perm.mode;
+               out->shm_perm.uid       = tbuf_old.shm_perm.uid;
+               out->shm_perm.gid       = tbuf_old.shm_perm.gid;
+               out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 
                return 0;
            }
@@ -395,26 +536,34 @@ static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminf
        }
 }
 
-static void shm_get_stat(unsigned long *rss, unsigned long *swp) 
+/*
+ * Called with shm_ids.rw_mutex held as a reader
+ */
+static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
+               unsigned long *swp)
 {
-       int i;
+       int next_id;
+       int total, in_use;
 
        *rss = 0;
        *swp = 0;
 
-       for (i = 0; i <= shm_ids.max_id; i++) {
+       in_use = shm_ids(ns).in_use;
+
+       for (total = 0, next_id = 0; total < in_use; next_id++) {
                struct shmid_kernel *shp;
                struct inode *inode;
 
-               shp = shm_get(i);
-               if(!shp)
+               shp = idr_find(&shm_ids(ns).ipcs_idr, next_id);
+               if (shp == NULL)
                        continue;
 
-               inode = shp->shm_file->f_dentry->d_inode;
+               inode = shp->shm_file->f_path.dentry->d_inode;
 
                if (is_file_hugepages(shp->shm_file)) {
                        struct address_space *mapping = inode->i_mapping;
-                       *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
+                       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);
@@ -422,14 +571,61 @@ static void shm_get_stat(unsigned long *rss, unsigned long *swp)
                        *swp += info->swapped;
                        spin_unlock(&info->lock);
                }
+
+               total++;
+       }
+}
+
+/*
+ * This function handles some shmctl commands which require the rw_mutex
+ * to be held in write mode.
+ * NOTE: no locks must be held, the rw_mutex is taken inside this function.
+ */
+static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
+                      struct shmid_ds __user *buf, int version)
+{
+       struct kern_ipc_perm *ipcp;
+       struct shmid64_ds shmid64;
+       struct shmid_kernel *shp;
+       int err;
+
+       if (cmd == IPC_SET) {
+               if (copy_shmid_from_user(&shmid64, buf, version))
+                       return -EFAULT;
+       }
+
+       ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
+       if (IS_ERR(ipcp))
+               return PTR_ERR(ipcp);
+
+       shp = container_of(ipcp, struct shmid_kernel, shm_perm);
+
+       err = security_shm_shmctl(shp, cmd);
+       if (err)
+               goto out_unlock;
+       switch (cmd) {
+       case IPC_RMID:
+               do_shm_rmid(ns, ipcp);
+               goto out_up;
+       case IPC_SET:
+               ipc_update_perm(&shmid64.shm_perm, ipcp);
+               shp->shm_ctim = get_seconds();
+               break;
+       default:
+               err = -EINVAL;
        }
+out_unlock:
+       shm_unlock(shp);
+out_up:
+       up_write(&shm_ids(ns).rw_mutex);
+       return err;
 }
 
-asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
+asmlinkage long sys_shmctl(int shmid, int cmd, struct shmid_ds __user *buf)
 {
-       struct shm_setbuf setbuf;
        struct shmid_kernel *shp;
        int err, version;
+       struct ipc_namespace *ns;
 
        if (cmd < 0 || shmid < 0) {
                err = -EINVAL;
@@ -437,6 +633,7 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
        }
 
        version = ipc_parse_version(&cmd);
+       ns = current->nsproxy->ipc_ns;
 
        switch (cmd) { /* replace with proc interface ? */
        case IPC_INFO:
@@ -448,15 +645,18 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
                        return err;
 
                memset(&shminfo,0,sizeof(shminfo));
-               shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
-               shminfo.shmmax = shm_ctlmax;
-               shminfo.shmall = shm_ctlall;
+               shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
+               shminfo.shmmax = ns->shm_ctlmax;
+               shminfo.shmall = ns->shm_ctlall;
 
                shminfo.shmmin = SHMMIN;
                if(copy_shminfo_to_user (buf, &shminfo, version))
                        return -EFAULT;
-               /* reading a integer is always atomic */
-               err= shm_ids.max_id;
+
+               down_read(&shm_ids(ns).rw_mutex);
+               err = ipc_get_maxid(&shm_ids(ns));
+               up_read(&shm_ids(ns).rw_mutex);
+
                if(err<0)
                        err = 0;
                goto out;
@@ -470,14 +670,14 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
                        return err;
 
                memset(&shm_info,0,sizeof(shm_info));
-               mutex_lock(&shm_ids.mutex);
-               shm_info.used_ids = shm_ids.in_use;
-               shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
-               shm_info.shm_tot = shm_tot;
+               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);
+               shm_info.shm_tot = ns->shm_tot;
                shm_info.swap_attempts = 0;
                shm_info.swap_successes = 0;
-               err = shm_ids.max_id;
-               mutex_unlock(&shm_ids.mutex);
+               err = ipc_get_maxid(&shm_ids(ns));
+               up_read(&shm_ids(ns).rw_mutex);
                if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
                        err = -EFAULT;
                        goto out;
@@ -491,20 +691,25 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
        {
                struct shmid64_ds tbuf;
                int result;
-               memset(&tbuf, 0, sizeof(tbuf));
-               shp = shm_lock(shmid);
-               if(shp==NULL) {
-                       err = -EINVAL;
+
+               if (!buf) {
+                       err = -EFAULT;
                        goto out;
-               } else if(cmd==SHM_STAT) {
-                       err = -EINVAL;
-                       if (shmid > shm_ids.max_id)
-                               goto out_unlock;
-                       result = shm_buildid(shmid, shp->shm_perm.seq);
+               }
+
+               if (cmd == SHM_STAT) {
+                       shp = shm_lock(ns, shmid);
+                       if (IS_ERR(shp)) {
+                               err = PTR_ERR(shp);
+                               goto out;
+                       }
+                       result = shp->shm_perm.id;
                } else {
-                       err = shm_checkid(shp,shmid);
-                       if(err)
-                               goto out_unlock;
+                       shp = shm_lock_check(ns, shmid);
+                       if (IS_ERR(shp)) {
+                               err = PTR_ERR(shp);
+                               goto out;
+                       }
                        result = 0;
                }
                err=-EACCES;
@@ -513,6 +718,7 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
                err = security_shm_shmctl(shp, cmd);
                if (err)
                        goto out_unlock;
+               memset(&tbuf, 0, sizeof(tbuf));
                kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
                tbuf.shm_segsz  = shp->shm_segsz;
                tbuf.shm_atime  = shp->shm_atim;
@@ -520,10 +726,7 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
                tbuf.shm_ctime  = shp->shm_ctim;
                tbuf.shm_cpid   = shp->shm_cprid;
                tbuf.shm_lpid   = shp->shm_lprid;
-               if (!is_file_hugepages(shp->shm_file))
-                       tbuf.shm_nattch = shp->shm_nattch;
-               else
-                       tbuf.shm_nattch = file_count(shp->shm_file) - 1;
+               tbuf.shm_nattch = shp->shm_nattch;
                shm_unlock(shp);
                if(copy_shmid_to_user (buf, &tbuf, version))
                        err = -EFAULT;
@@ -534,14 +737,11 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
        case SHM_LOCK:
        case SHM_UNLOCK:
        {
-               shp = shm_lock(shmid);
-               if(shp==NULL) {
-                       err = -EINVAL;
+               shp = shm_lock_check(ns, shmid);
+               if (IS_ERR(shp)) {
+                       err = PTR_ERR(shp);
                        goto out;
                }
-               err = shm_checkid(shp,shmid);
-               if(err)
-                       goto out_unlock;
 
                err = audit_ipc_obj(&(shp->shm_perm));
                if (err)
@@ -565,7 +765,7 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
                        struct user_struct * user = current->user;
                        if (!is_file_hugepages(shp->shm_file)) {
                                err = shmem_lock(shp->shm_file, 1, user);
-                               if (!err) {
+                               if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
                                        shp->shm_perm.mode |= SHM_LOCKED;
                                        shp->mlock_user = user;
                                }
@@ -579,102 +779,13 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
                goto out;
        }
        case IPC_RMID:
-       {
-               /*
-                *      We cannot simply remove the file. The SVID states
-                *      that the block remains until the last person
-                *      detaches from it, then is deleted. A shmat() on
-                *      an RMID segment is legal in older Linux and if 
-                *      we change it apps break...
-                *
-                *      Instead we set a destroyed flag, and then blow
-                *      the name away when the usage hits zero.
-                */
-               mutex_lock(&shm_ids.mutex);
-               shp = shm_lock(shmid);
-               err = -EINVAL;
-               if (shp == NULL) 
-                       goto out_up;
-               err = shm_checkid(shp, shmid);
-               if(err)
-                       goto out_unlock_up;
-
-               err = audit_ipc_obj(&(shp->shm_perm));
-               if (err)
-                       goto out_unlock_up;
-
-               if (current->euid != shp->shm_perm.uid &&
-                   current->euid != shp->shm_perm.cuid && 
-                   !capable(CAP_SYS_ADMIN)) {
-                       err=-EPERM;
-                       goto out_unlock_up;
-               }
-
-               err = security_shm_shmctl(shp, cmd);
-               if (err)
-                       goto out_unlock_up;
-
-               if (shp->shm_nattch){
-                       shp->shm_perm.mode |= SHM_DEST;
-                       /* Do not find it any more */
-                       shp->shm_perm.key = IPC_PRIVATE;
-                       shm_unlock(shp);
-               } else
-                       shm_destroy (shp);
-               mutex_unlock(&shm_ids.mutex);
-               goto out;
-       }
-
        case IPC_SET:
-       {
-               if (copy_shmid_from_user (&setbuf, buf, version)) {
-                       err = -EFAULT;
-                       goto out;
-               }
-               mutex_lock(&shm_ids.mutex);
-               shp = shm_lock(shmid);
-               err=-EINVAL;
-               if(shp==NULL)
-                       goto out_up;
-               err = shm_checkid(shp,shmid);
-               if(err)
-                       goto out_unlock_up;
-               err = audit_ipc_obj(&(shp->shm_perm));
-               if (err)
-                       goto out_unlock_up;
-               err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
-               if (err)
-                       goto out_unlock_up;
-               err=-EPERM;
-               if (current->euid != shp->shm_perm.uid &&
-                   current->euid != shp->shm_perm.cuid && 
-                   !capable(CAP_SYS_ADMIN)) {
-                       goto out_unlock_up;
-               }
-
-               err = security_shm_shmctl(shp, cmd);
-               if (err)
-                       goto out_unlock_up;
-               
-               shp->shm_perm.uid = setbuf.uid;
-               shp->shm_perm.gid = setbuf.gid;
-               shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
-                       | (setbuf.mode & S_IRWXUGO);
-               shp->shm_ctim = get_seconds();
-               break;
-       }
-
+               err = shmctl_down(ns, shmid, cmd, buf, version);
+               return err;
        default:
-               err = -EINVAL;
-               goto out;
+               return -EINVAL;
        }
 
-       err = 0;
-out_unlock_up:
-       shm_unlock(shp);
-out_up:
-       mutex_unlock(&shm_ids.mutex);
-       goto out;
 out_unlock:
        shm_unlock(shp);
 out:
@@ -698,12 +809,16 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
        unsigned long flags;
        unsigned long prot;
        int acc_mode;
-       void *user_addr;
-
-       if (shmid < 0) {
-               err = -EINVAL;
+       unsigned long user_addr;
+       struct ipc_namespace *ns;
+       struct shm_file_data *sfd;
+       struct path path;
+       mode_t f_mode;
+
+       err = -EINVAL;
+       if (shmid < 0)
                goto out;
-       } else if ((addr = (ulong)shmaddr)) {
+       else if ((addr = (ulong)shmaddr)) {
                if (addr & (SHMLBA-1)) {
                        if (shmflg & SHM_RND)
                                addr &= ~(SHMLBA-1);       /* round down */
@@ -711,12 +826,12 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 #ifndef __ARCH_FORCE_SHMLBA
                                if (addr & ~PAGE_MASK)
 #endif
-                                       return -EINVAL;
+                                       goto out;
                }
                flags = MAP_SHARED | MAP_FIXED;
        } else {
                if ((shmflg & SHM_REMAP))
-                       return -EINVAL;
+                       goto out;
 
                flags = MAP_SHARED;
        }
@@ -724,9 +839,11 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
        if (shmflg & SHM_RDONLY) {
                prot = PROT_READ;
                acc_mode = S_IRUGO;
+               f_mode = FMODE_READ;
        } else {
                prot = PROT_READ | PROT_WRITE;
                acc_mode = S_IRUGO | S_IWUGO;
+               f_mode = FMODE_READ | FMODE_WRITE;
        }
        if (shmflg & SHM_EXEC) {
                prot |= PROT_EXEC;
@@ -737,36 +854,46 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
         * We cannot rely on the fs check since SYSV IPC does have an
         * additional creator id...
         */
-       shp = shm_lock(shmid);
-       if(shp == NULL) {
-               err = -EINVAL;
-               goto out;
-       }
-       err = shm_checkid(shp,shmid);
-       if (err) {
-               shm_unlock(shp);
-               goto out;
-       }
-       if (ipcperms(&shp->shm_perm, acc_mode)) {
-               shm_unlock(shp);
-               err = -EACCES;
+       ns = current->nsproxy->ipc_ns;
+       shp = shm_lock_check(ns, shmid);
+       if (IS_ERR(shp)) {
+               err = PTR_ERR(shp);
                goto out;
        }
 
+       err = -EACCES;
+       if (ipcperms(&shp->shm_perm, acc_mode))
+               goto out_unlock;
+
        err = security_shm_shmat(shp, shmaddr, shmflg);
-       if (err) {
-               shm_unlock(shp);
-               return err;
-       }
-               
-       file = shp->shm_file;
-       size = i_size_read(file->f_dentry->d_inode);
+       if (err)
+               goto out_unlock;
+
+       path.dentry = dget(shp->shm_file->f_path.dentry);
+       path.mnt    = shp->shm_file->f_path.mnt;
        shp->shm_nattch++;
+       size = i_size_read(path.dentry->d_inode);
        shm_unlock(shp);
 
+       err = -ENOMEM;
+       sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
+       if (!sfd)
+               goto out_put_dentry;
+
+       file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
+       if (!file)
+               goto out_free;
+
+       file->private_data = sfd;
+       file->f_mapping = shp->shm_file->f_mapping;
+       sfd->id = shp->shm_perm.id;
+       sfd->ns = get_ipc_ns(ns);
+       sfd->file = shp->shm_file;
+       sfd->vm_ops = NULL;
+
        down_write(&current->mm->mmap_sem);
        if (addr && !(shmflg & SHM_REMAP)) {
-               user_addr = ERR_PTR(-EINVAL);
+               err = -EINVAL;
                if (find_vma_intersection(current->mm, addr, addr + size))
                        goto invalid;
                /*
@@ -778,28 +905,40 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
                        goto invalid;
        }
                
-       user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
-
+       user_addr = do_mmap (file, addr, size, prot, flags, 0);
+       *raddr = user_addr;
+       err = 0;
+       if (IS_ERR_VALUE(user_addr))
+               err = (long)user_addr;
 invalid:
        up_write(&current->mm->mmap_sem);
 
-       mutex_lock(&shm_ids.mutex);
-       shp = shm_lock(shmid);
-       BUG_ON(!shp);
+       fput(file);
+
+out_nattch:
+       down_write(&shm_ids(ns).rw_mutex);
+       shp = shm_lock(ns, shmid);
+       BUG_ON(IS_ERR(shp));
        shp->shm_nattch--;
        if(shp->shm_nattch == 0 &&
           shp->shm_perm.mode & SHM_DEST)
-               shm_destroy (shp);
+               shm_destroy(ns, shp);
        else
                shm_unlock(shp);
-       mutex_unlock(&shm_ids.mutex);
+       up_write(&shm_ids(ns).rw_mutex);
 
-       *raddr = (unsigned long) user_addr;
-       err = 0;
-       if (IS_ERR(user_addr))
-               err = PTR_ERR(user_addr);
 out:
        return err;
+
+out_unlock:
+       shm_unlock(shp);
+       goto out;
+
+out_free:
+       kfree(sfd);
+out_put_dentry:
+       dput(path.dentry);
+       goto out_nattch;
 }
 
 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
@@ -861,11 +1000,11 @@ asmlinkage long sys_shmdt(char __user *shmaddr)
                 * a fragment created by mprotect() and/or munmap(), or it
                 * otherwise it starts at this address with no hassles.
                 */
-               if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
+               if ((vma->vm_ops == &shm_vm_ops) &&
                        (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 
 
-                       size = vma->vm_file->f_dentry->d_inode->i_size;
+                       size = vma->vm_file->f_path.dentry->d_inode->i_size;
                        do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
                        /*
                         * We discovered the size of the shm segment, so
@@ -890,7 +1029,7 @@ asmlinkage long sys_shmdt(char __user *shmaddr)
                next = vma->vm_next;
 
                /* finding a matching vma now does not alter retval */
-               if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
+               if ((vma->vm_ops == &shm_vm_ops) &&
                        (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 
                        do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
@@ -905,23 +1044,23 @@ asmlinkage long sys_shmdt(char __user *shmaddr)
 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
 {
        struct shmid_kernel *shp = it;
-       char *format;
 
-#define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
-#define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
+#if BITS_PER_LONG <= 32
+#define SIZE_SPEC "%10lu"
+#else
+#define SIZE_SPEC "%21lu"
+#endif
 
-       if (sizeof(size_t) <= sizeof(int))
-               format = SMALL_STRING;
-       else
-               format = BIG_STRING;
-       return seq_printf(s, format,
+       return seq_printf(s,
+                         "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
+                         "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
                          shp->shm_perm.key,
-                         shp->id,
+                         shp->shm_perm.id,
                          shp->shm_perm.mode,
                          shp->shm_segsz,
                          shp->shm_cprid,
                          shp->shm_lprid,
-                         is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
+                         shp->shm_nattch,
                          shp->shm_perm.uid,
                          shp->shm_perm.gid,
                          shp->shm_perm.cuid,