]> Pileus Git - ~andy/linux/blobdiff - kernel/sys.c
Merge git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile
[~andy/linux] / kernel / sys.c
index 6df42624e454aeb236ab1c9413d6fdf5f676365f..9ff89cb9657a681d15714447d7d3c64939fbbe50 100644 (file)
@@ -36,6 +36,8 @@
 #include <linux/personality.h>
 #include <linux/ptrace.h>
 #include <linux/fs_struct.h>
+#include <linux/file.h>
+#include <linux/mount.h>
 #include <linux/gfp.h>
 #include <linux/syscore_ops.h>
 #include <linux/version.h>
@@ -1378,8 +1380,8 @@ SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
                memcpy(u->nodename, tmp, len);
                memset(u->nodename + len, 0, sizeof(u->nodename) - len);
                errno = 0;
+               uts_proc_notify(UTS_PROC_HOSTNAME);
        }
-       uts_proc_notify(UTS_PROC_HOSTNAME);
        up_write(&uts_sem);
        return errno;
 }
@@ -1429,8 +1431,8 @@ SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
                memcpy(u->domainname, tmp, len);
                memset(u->domainname + len, 0, sizeof(u->domainname) - len);
                errno = 0;
+               uts_proc_notify(UTS_PROC_DOMAINNAME);
        }
-       uts_proc_notify(UTS_PROC_DOMAINNAME);
        up_write(&uts_sem);
        return errno;
 }
@@ -1784,77 +1786,102 @@ SYSCALL_DEFINE1(umask, int, mask)
 }
 
 #ifdef CONFIG_CHECKPOINT_RESTORE
+static bool vma_flags_mismatch(struct vm_area_struct *vma,
+                              unsigned long required,
+                              unsigned long banned)
+{
+       return (vma->vm_flags & required) != required ||
+               (vma->vm_flags & banned);
+}
+
+static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
+{
+       struct file *exe_file;
+       struct dentry *dentry;
+       int err;
+
+       /*
+        * Setting new mm::exe_file is only allowed when no VM_EXECUTABLE vma's
+        * remain. So perform a quick test first.
+        */
+       if (mm->num_exe_file_vmas)
+               return -EBUSY;
+
+       exe_file = fget(fd);
+       if (!exe_file)
+               return -EBADF;
+
+       dentry = exe_file->f_path.dentry;
+
+       /*
+        * Because the original mm->exe_file points to executable file, make
+        * sure that this one is executable as well, to avoid breaking an
+        * overall picture.
+        */
+       err = -EACCES;
+       if (!S_ISREG(dentry->d_inode->i_mode)   ||
+           exe_file->f_path.mnt->mnt_flags & MNT_NOEXEC)
+               goto exit;
+
+       err = inode_permission(dentry->d_inode, MAY_EXEC);
+       if (err)
+               goto exit;
+
+       /*
+        * The symlink can be changed only once, just to disallow arbitrary
+        * transitions malicious software might bring in. This means one
+        * could make a snapshot over all processes running and monitor
+        * /proc/pid/exe changes to notice unusual activity if needed.
+        */
+       down_write(&mm->mmap_sem);
+       if (likely(!mm->exe_file))
+               set_mm_exe_file(mm, exe_file);
+       else
+               err = -EBUSY;
+       up_write(&mm->mmap_sem);
+
+exit:
+       fput(exe_file);
+       return err;
+}
+
 static int prctl_set_mm(int opt, unsigned long addr,
                        unsigned long arg4, unsigned long arg5)
 {
        unsigned long rlim = rlimit(RLIMIT_DATA);
-       unsigned long vm_req_flags;
-       unsigned long vm_bad_flags;
-       struct vm_area_struct *vma;
-       int error = 0;
        struct mm_struct *mm = current->mm;
+       struct vm_area_struct *vma;
+       int error;
 
-       if (arg4 | arg5)
+       if (arg5 || (arg4 && opt != PR_SET_MM_AUXV))
                return -EINVAL;
 
        if (!capable(CAP_SYS_RESOURCE))
                return -EPERM;
 
+       if (opt == PR_SET_MM_EXE_FILE)
+               return prctl_set_mm_exe_file(mm, (unsigned int)addr);
+
        if (addr >= TASK_SIZE)
                return -EINVAL;
 
+       error = -EINVAL;
+
        down_read(&mm->mmap_sem);
        vma = find_vma(mm, addr);
 
-       if (opt != PR_SET_MM_START_BRK && opt != PR_SET_MM_BRK) {
-               /* It must be existing VMA */
-               if (!vma || vma->vm_start > addr)
-                       goto out;
-       }
-
-       error = -EINVAL;
        switch (opt) {
        case PR_SET_MM_START_CODE:
+               mm->start_code = addr;
+               break;
        case PR_SET_MM_END_CODE:
-               vm_req_flags = VM_READ | VM_EXEC;
-               vm_bad_flags = VM_WRITE | VM_MAYSHARE;
-
-               if ((vma->vm_flags & vm_req_flags) != vm_req_flags ||
-                   (vma->vm_flags & vm_bad_flags))
-                       goto out;
-
-               if (opt == PR_SET_MM_START_CODE)
-                       mm->start_code = addr;
-               else
-                       mm->end_code = addr;
+               mm->end_code = addr;
                break;
-
        case PR_SET_MM_START_DATA:
-       case PR_SET_MM_END_DATA:
-               vm_req_flags = VM_READ | VM_WRITE;
-               vm_bad_flags = VM_EXEC | VM_MAYSHARE;
-
-               if ((vma->vm_flags & vm_req_flags) != vm_req_flags ||
-                   (vma->vm_flags & vm_bad_flags))
-                       goto out;
-
-               if (opt == PR_SET_MM_START_DATA)
-                       mm->start_data = addr;
-               else
-                       mm->end_data = addr;
+               mm->start_data = addr;
                break;
-
-       case PR_SET_MM_START_STACK:
-
-#ifdef CONFIG_STACK_GROWSUP
-               vm_req_flags = VM_READ | VM_WRITE | VM_GROWSUP;
-#else
-               vm_req_flags = VM_READ | VM_WRITE | VM_GROWSDOWN;
-#endif
-               if ((vma->vm_flags & vm_req_flags) != vm_req_flags)
-                       goto out;
-
-               mm->start_stack = addr;
+       case PR_SET_MM_END_DATA:
+               mm->end_data = addr;
                break;
 
        case PR_SET_MM_START_BRK:
@@ -1881,16 +1908,77 @@ static int prctl_set_mm(int opt, unsigned long addr,
                mm->brk = addr;
                break;
 
+       /*
+        * If command line arguments and environment
+        * are placed somewhere else on stack, we can
+        * set them up here, ARG_START/END to setup
+        * command line argumets and ENV_START/END
+        * for environment.
+        */
+       case PR_SET_MM_START_STACK:
+       case PR_SET_MM_ARG_START:
+       case PR_SET_MM_ARG_END:
+       case PR_SET_MM_ENV_START:
+       case PR_SET_MM_ENV_END:
+               if (!vma) {
+                       error = -EFAULT;
+                       goto out;
+               }
+#ifdef CONFIG_STACK_GROWSUP
+               if (vma_flags_mismatch(vma, VM_READ | VM_WRITE | VM_GROWSUP, 0))
+#else
+               if (vma_flags_mismatch(vma, VM_READ | VM_WRITE | VM_GROWSDOWN, 0))
+#endif
+                       goto out;
+               if (opt == PR_SET_MM_START_STACK)
+                       mm->start_stack = addr;
+               else if (opt == PR_SET_MM_ARG_START)
+                       mm->arg_start = addr;
+               else if (opt == PR_SET_MM_ARG_END)
+                       mm->arg_end = addr;
+               else if (opt == PR_SET_MM_ENV_START)
+                       mm->env_start = addr;
+               else if (opt == PR_SET_MM_ENV_END)
+                       mm->env_end = addr;
+               break;
+
+       /*
+        * This doesn't move auxiliary vector itself
+        * since it's pinned to mm_struct, but allow
+        * to fill vector with new values. It's up
+        * to a caller to provide sane values here
+        * otherwise user space tools which use this
+        * vector might be unhappy.
+        */
+       case PR_SET_MM_AUXV: {
+               unsigned long user_auxv[AT_VECTOR_SIZE];
+
+               if (arg4 > sizeof(user_auxv))
+                       goto out;
+               up_read(&mm->mmap_sem);
+
+               if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
+                       return -EFAULT;
+
+               /* Make sure the last entry is always AT_NULL */
+               user_auxv[AT_VECTOR_SIZE - 2] = 0;
+               user_auxv[AT_VECTOR_SIZE - 1] = 0;
+
+               BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
+
+               task_lock(current);
+               memcpy(mm->saved_auxv, user_auxv, arg4);
+               task_unlock(current);
+
+               return 0;
+       }
        default:
-               error = -EINVAL;
                goto out;
        }
 
        error = 0;
-
 out:
        up_read(&mm->mmap_sem);
-
        return error;
 }
 #else /* CONFIG_CHECKPOINT_RESTORE */
@@ -2114,7 +2202,6 @@ int orderly_poweroff(bool force)
                NULL
        };
        int ret = -ENOMEM;
-       struct subprocess_info *info;
 
        if (argv == NULL) {
                printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
@@ -2122,18 +2209,16 @@ int orderly_poweroff(bool force)
                goto out;
        }
 
-       info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
-       if (info == NULL) {
-               argv_free(argv);
-               goto out;
-       }
-
-       call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
+       ret = call_usermodehelper_fns(argv[0], argv, envp, UMH_NO_WAIT,
+                                     NULL, argv_cleanup, NULL);
+out:
+       if (likely(!ret))
+               return 0;
 
-       ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
+       if (ret == -ENOMEM)
+               argv_free(argv);
 
-  out:
-       if (ret && force) {
+       if (force) {
                printk(KERN_WARNING "Failed to start orderly shutdown: "
                       "forcing the issue\n");