[PATCH] IPC namespace - shm
Kirill Korotaev [Mon, 2 Oct 2006 09:18:22 +0000 (02:18 -0700)]
IPC namespace support for IPC shm code.

Signed-off-by: Pavel Emelianiov <xemul@openvz.org>
Signed-off-by: Kirill Korotaev <dev@openvz.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

ipc/shm.c

index 940b0c9..bfbd317 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>
@@ -32,6 +36,7 @@
 #include <linux/ptrace.h>
 #include <linux/seq_file.h>
 #include <linux/mutex.h>
+#include <linux/nsproxy.h>
 
 #include <asm/uaccess.h>
 
 static struct file_operations shm_file_operations;
 static struct vm_operations_struct shm_vm_ops;
 
-static struct ipc_ids shm_ids;
+static struct ipc_ids init_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_lock(ns, id)               \
+       ((struct shmid_kernel*)ipc_lock(&shm_ids(ns),id))
+#define shm_unlock(shp)                        \
+       ipc_unlock(&(shp)->shm_perm)
+#define shm_get(ns, id)                        \
+       ((struct shmid_kernel*)ipc_get(&shm_ids(ns),id))
+#define shm_buildid(ns, id, seq)       \
+       ipc_buildid(&shm_ids(ns), id, seq)
 
-static int newseg (key_t key, int shmflg, size_t size);
+static int newseg (struct ipc_namespace *ns, 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 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;
+static void __ipc_init __shm_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids)
+{
+       ns->ids[IPC_SHM_IDS] = ids;
+       ns->shm_ctlmax = SHMMAX;
+       ns->shm_ctlall = SHMALL;
+       ns->shm_ctlmni = SHMMNI;
+       ns->shm_tot = 0;
+       ipc_init_ids(ids, 1);
+}
+
+static void do_shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *shp)
+{
+       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);
+}
+
+#ifdef CONFIG_IPC_NS
+int shm_init_ns(struct ipc_namespace *ns)
+{
+       struct ipc_ids *ids;
+
+       ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL);
+       if (ids == NULL)
+               return -ENOMEM;
 
-static int shm_tot; /* total number of shared memory pages */
+       __shm_init_ns(ns, ids);
+       return 0;
+}
+
+void shm_exit_ns(struct ipc_namespace *ns)
+{
+       int i;
+       struct shmid_kernel *shp;
+
+       mutex_lock(&shm_ids(ns).mutex);
+       for (i = 0; i <= shm_ids(ns).max_id; i++) {
+               shp = shm_lock(ns, i);
+               if (shp == NULL)
+                       continue;
+
+               do_shm_rmid(ns, shp);
+       }
+       mutex_unlock(&shm_ids(ns).mutex);
+
+       kfree(ns->ids[IPC_SHM_IDS]);
+       ns->ids[IPC_SHM_IDS] = NULL;
+}
+#endif
 
 void __init shm_init (void)
 {
-       ipc_init_ids(&shm_ids, 1);
+       __shm_init_ns(&init_ipc_ns, &init_shm_ids);
        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)
+static inline int shm_checkid(struct ipc_namespace *ns,
+               struct shmid_kernel *s, int id)
 {
-       if (ipc_checkid(&shm_ids,&s->shm_perm,id))
+       if (ipc_checkid(&shm_ids(ns), &s->shm_perm, id))
                return -EIDRM;
        return 0;
 }
 
-static inline struct shmid_kernel *shm_rmid(int id)
+static inline struct shmid_kernel *shm_rmid(struct ipc_namespace *ns, int id)
 {
-       return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
+       return (struct shmid_kernel *)ipc_rmid(&shm_ids(ns), id);
 }
 
-static inline int shm_addid(struct shmid_kernel *shp)
+static inline int shm_addid(struct ipc_namespace *ns, struct shmid_kernel *shp)
 {
-       return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
+       return ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 }
 
 
 
-static inline void shm_inc (int id) {
+static inline void shm_inc(struct ipc_namespace *ns, int id)
+{
        struct shmid_kernel *shp;
 
-       shp = shm_lock(id);
+       shp = shm_lock(ns, id);
        BUG_ON(!shp);
        shp->shm_atim = get_seconds();
        shp->shm_lprid = current->tgid;
@@ -100,10 +161,13 @@ static inline void shm_inc (int id) {
        shm_unlock(shp);
 }
 
+#define shm_file_ns(file) (*((struct ipc_namespace **)&(file)->private_data))
+
 /* This is called by fork, once for every shm attach. */
-static void shm_open (struct vm_area_struct *shmd)
+static void shm_open(struct vm_area_struct *shmd)
 {
-       shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
+       shm_inc(shm_file_ns(shmd->vm_file),
+                       shmd->vm_file->f_dentry->d_inode->i_ino);
 }
 
 /*
@@ -114,10 +178,10 @@ static void shm_open (struct vm_area_struct *shmd)
  * It has to be called with shp and shm_ids.mutex 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->id);
        shm_unlock(shp);
        if (!is_file_hugepages(shp->shm_file))
                shmem_lock(shp->shm_file, 0, shp->mlock_user);
@@ -140,20 +204,23 @@ static void shm_close (struct vm_area_struct *shmd)
        struct file * file = shmd->vm_file;
        int id = file->f_dentry->d_inode->i_ino;
        struct shmid_kernel *shp;
+       struct ipc_namespace *ns;
 
-       mutex_lock(&shm_ids.mutex);
+       ns = shm_file_ns(file);
+
+       mutex_lock(&shm_ids(ns).mutex);
        /* remove from the list of attaches of the shm segment */
-       shp = shm_lock(id);
+       shp = shm_lock(ns, id);
        BUG_ON(!shp);
        shp->shm_lprid = current->tgid;
        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);
+       mutex_unlock(&shm_ids(ns).mutex);
 }
 
 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
@@ -165,14 +232,25 @@ static int shm_mmap(struct file * file, struct vm_area_struct * vma)
                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);
+               shm_inc(shm_file_ns(file), file->f_dentry->d_inode->i_ino);
        }
 
        return ret;
 }
 
+static int shm_release(struct inode *ino, struct file *file)
+{
+       struct ipc_namespace *ns;
+
+       ns = shm_file_ns(file);
+       put_ipc_ns(ns);
+       shm_file_ns(file) = NULL;
+       return 0;
+}
+
 static struct file_operations shm_file_operations = {
-       .mmap   = shm_mmap,
+       .mmap           = shm_mmap,
+       .release        = shm_release,
 #ifndef CONFIG_MMU
        .get_unmapped_area = shmem_get_unmapped_area,
 #endif
@@ -188,7 +266,7 @@ static struct vm_operations_struct shm_vm_ops = {
 #endif
 };
 
-static int newseg (key_t key, int shmflg, size_t size)
+static int newseg (struct ipc_namespace *ns, key_t key, int shmflg, size_t size)
 {
        int error;
        struct shmid_kernel *shp;
@@ -197,10 +275,10 @@ 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));
@@ -239,7 +317,7 @@ static int newseg (key_t key, int shmflg, size_t size)
                goto no_file;
 
        error = -ENOSPC;
-       id = shm_addid(shp);
+       id = shm_addid(ns, shp);
        if(id == -1) 
                goto no_id;
 
@@ -249,15 +327,17 @@ static int newseg (key_t key, int shmflg, size_t size)
        shp->shm_ctim = get_seconds();
        shp->shm_segsz = size;
        shp->shm_nattch = 0;
-       shp->id = shm_buildid(id,shp->shm_perm.seq);
+       shp->id = shm_buildid(ns, id, shp->shm_perm.seq);
        shp->shm_file = file;
        file->f_dentry->d_inode->i_ino = shp->id;
 
+       shm_file_ns(file) = get_ipc_ns(ns);
+
        /* Hugetlb ops would have already been assigned. */
        if (!(shmflg & SHM_HUGETLB))
                file->f_op = &shm_file_operations;
 
-       shm_tot += numpages;
+       ns->shm_tot += numpages;
        shm_unlock(shp);
        return shp->id;
 
@@ -273,33 +353,36 @@ asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 {
        struct shmid_kernel *shp;
        int err, id = 0;
+       struct ipc_namespace *ns;
+
+       ns = current->nsproxy->ipc_ns;
 
-       mutex_lock(&shm_ids.mutex);
+       mutex_lock(&shm_ids(ns).mutex);
        if (key == IPC_PRIVATE) {
-               err = newseg(key, shmflg, size);
-       } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
+               err = newseg(ns, key, shmflg, size);
+       } else if ((id = ipc_findkey(&shm_ids(ns), key)) == -1) {
                if (!(shmflg & IPC_CREAT))
                        err = -ENOENT;
                else
-                       err = newseg(key, shmflg, size);
+                       err = newseg(ns, key, shmflg, size);
        } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
                err = -EEXIST;
        } else {
-               shp = shm_lock(id);
+               shp = shm_lock(ns, 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);
+                       int shmid = shm_buildid(ns, id, shp->shm_perm.seq);
                        err = security_shm_associate(shp, shmflg);
                        if (!err)
                                err = shmid;
                }
                shm_unlock(shp);
        }
-       mutex_unlock(&shm_ids.mutex);
+       mutex_unlock(&shm_ids(ns).mutex);
 
        return err;
 }
@@ -395,18 +478,19 @@ static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminf
        }
 }
 
-static void shm_get_stat(unsigned long *rss, unsigned long *swp) 
+static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
+               unsigned long *swp)
 {
        int i;
 
        *rss = 0;
        *swp = 0;
 
-       for (i = 0; i <= shm_ids.max_id; i++) {
+       for (i = 0; i <= shm_ids(ns).max_id; i++) {
                struct shmid_kernel *shp;
                struct inode *inode;
 
-               shp = shm_get(i);
+               shp = shm_get(ns, i);
                if(!shp)
                        continue;
 
@@ -430,6 +514,7 @@ 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 +522,7 @@ 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 +534,15 @@ 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;
+               err= shm_ids(ns).max_id;
                if(err<0)
                        err = 0;
                goto out;
@@ -470,14 +556,14 @@ 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;
+               mutex_lock(&shm_ids(ns).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 = shm_ids(ns).max_id;
+               mutex_unlock(&shm_ids(ns).mutex);
                if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
                        err = -EFAULT;
                        goto out;
@@ -492,17 +578,17 @@ 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);
+               shp = shm_lock(ns, shmid);
                if(shp==NULL) {
                        err = -EINVAL;
                        goto out;
                } else if(cmd==SHM_STAT) {
                        err = -EINVAL;
-                       if (shmid > shm_ids.max_id)
+                       if (shmid > shm_ids(ns).max_id)
                                goto out_unlock;
-                       result = shm_buildid(shmid, shp->shm_perm.seq);
+                       result = shm_buildid(ns, shmid, shp->shm_perm.seq);
                } else {
-                       err = shm_checkid(shp,shmid);
+                       err = shm_checkid(ns, shp,shmid);
                        if(err)
                                goto out_unlock;
                        result = 0;
@@ -534,12 +620,12 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
        case SHM_LOCK:
        case SHM_UNLOCK:
        {
-               shp = shm_lock(shmid);
+               shp = shm_lock(ns, shmid);
                if(shp==NULL) {
                        err = -EINVAL;
                        goto out;
                }
-               err = shm_checkid(shp,shmid);
+               err = shm_checkid(ns, shp,shmid);
                if(err)
                        goto out_unlock;
 
@@ -590,12 +676,12 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
                 *      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);
+               mutex_lock(&shm_ids(ns).mutex);
+               shp = shm_lock(ns, shmid);
                err = -EINVAL;
                if (shp == NULL) 
                        goto out_up;
-               err = shm_checkid(shp, shmid);
+               err = shm_checkid(ns, shp, shmid);
                if(err)
                        goto out_unlock_up;
 
@@ -614,14 +700,8 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
                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);
+               do_shm_rmid(ns, shp);
+               mutex_unlock(&shm_ids(ns).mutex);
                goto out;
        }
 
@@ -631,12 +711,12 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
                        err = -EFAULT;
                        goto out;
                }
-               mutex_lock(&shm_ids.mutex);
-               shp = shm_lock(shmid);
+               mutex_lock(&shm_ids(ns).mutex);
+               shp = shm_lock(ns, shmid);
                err=-EINVAL;
                if(shp==NULL)
                        goto out_up;
-               err = shm_checkid(shp,shmid);
+               err = shm_checkid(ns, shp,shmid);
                if(err)
                        goto out_unlock_up;
                err = audit_ipc_obj(&(shp->shm_perm));
@@ -673,7 +753,7 @@ asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 out_unlock_up:
        shm_unlock(shp);
 out_up:
-       mutex_unlock(&shm_ids.mutex);
+       mutex_unlock(&shm_ids(ns).mutex);
        goto out;
 out_unlock:
        shm_unlock(shp);
@@ -699,6 +779,7 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
        unsigned long prot;
        int acc_mode;
        void *user_addr;
+       struct ipc_namespace *ns;
 
        if (shmid < 0) {
                err = -EINVAL;
@@ -737,12 +818,13 @@ 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);
+       ns = current->nsproxy->ipc_ns;
+       shp = shm_lock(ns, shmid);
        if(shp == NULL) {
                err = -EINVAL;
                goto out;
        }
-       err = shm_checkid(shp,shmid);
+       err = shm_checkid(ns, shp,shmid);
        if (err) {
                shm_unlock(shp);
                goto out;
@@ -783,16 +865,16 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 invalid:
        up_write(&current->mm->mmap_sem);
 
-       mutex_lock(&shm_ids.mutex);
-       shp = shm_lock(shmid);
+       mutex_lock(&shm_ids(ns).mutex);
+       shp = shm_lock(ns, shmid);
        BUG_ON(!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);
+       mutex_unlock(&shm_ids(ns).mutex);
 
        *raddr = (unsigned long) user_addr;
        err = 0;