pid namespaces: allow cloning of new namespace
[deliverable/linux.git] / kernel / fork.c
index 33f12f48684a0053a82d72506c1d7ec1e7ca6a15..f252784f93308077d483bca888593ec7fad36774 100644 (file)
@@ -29,7 +29,7 @@
 #include <linux/nsproxy.h>
 #include <linux/capability.h>
 #include <linux/cpu.h>
-#include <linux/cpuset.h>
+#include <linux/cgroup.h>
 #include <linux/security.h>
 #include <linux/swap.h>
 #include <linux/syscalls.h>
@@ -107,6 +107,7 @@ static struct kmem_cache *mm_cachep;
 
 void free_task(struct task_struct *tsk)
 {
+       prop_local_destroy_single(&tsk->dirties);
        free_thread_info(tsk->stack);
        rt_mutex_debug_task_free(tsk);
        free_task_struct(tsk);
@@ -163,6 +164,7 @@ static struct task_struct *dup_task_struct(struct task_struct *orig)
 {
        struct task_struct *tsk;
        struct thread_info *ti;
+       int err;
 
        prepare_to_copy(orig);
 
@@ -178,6 +180,14 @@ static struct task_struct *dup_task_struct(struct task_struct *orig)
 
        *tsk = *orig;
        tsk->stack = ti;
+
+       err = prop_local_init_single(&tsk->dirties);
+       if (err) {
+               free_thread_info(ti);
+               free_task_struct(tsk);
+               return NULL;
+       }
+
        setup_thread_stack(tsk, orig);
 
 #ifdef CONFIG_CC_STACKPROTECTOR
@@ -258,7 +268,7 @@ static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
                        get_file(file);
                        if (tmp->vm_flags & VM_DENYWRITE)
                                atomic_dec(&inode->i_writecount);
-      
+
                        /* insert tmp into the share list, just after mpnt */
                        spin_lock(&file->f_mapping->i_mmap_lock);
                        tmp->vm_truncate_count = mpnt->vm_truncate_count;
@@ -321,7 +331,7 @@ static inline void mm_free_pgd(struct mm_struct * mm)
 #define mm_free_pgd(mm)
 #endif /* CONFIG_MMU */
 
- __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
+__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
 
 #define allocate_mm()  (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
 #define free_mm(mm)    (kmem_cache_free(mm_cachep, (mm)))
@@ -728,8 +738,8 @@ static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
        /* compute the remainder to be cleared */
        size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
 
-       /* This is long word aligned thus could use a optimized version */ 
-       memset(new_fds, 0, size); 
+       /* This is long word aligned thus could use a optimized version */
+       memset(new_fds, 0, size);
 
        if (new_fdt->max_fds > open_files) {
                int left = (new_fdt->max_fds-open_files)/8;
@@ -877,6 +887,8 @@ static inline int copy_signal(unsigned long clone_flags, struct task_struct * ts
        sig->tty_old_pgrp = NULL;
 
        sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
+       sig->gtime = cputime_zero;
+       sig->cgtime = cputime_zero;
        sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
        sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
        sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
@@ -930,6 +942,7 @@ static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
        if (!(clone_flags & CLONE_PTRACE))
                p->ptrace = 0;
        p->flags = new_flags;
+       clear_freeze_flag(p);
 }
 
 asmlinkage long sys_set_tid_address(int __user *tidptr)
@@ -960,12 +973,12 @@ static struct task_struct *copy_process(unsigned long clone_flags,
                                        unsigned long stack_start,
                                        struct pt_regs *regs,
                                        unsigned long stack_size,
-                                       int __user *parent_tidptr,
                                        int __user *child_tidptr,
                                        struct pid *pid)
 {
        int retval;
        struct task_struct *p = NULL;
+       int cgroup_callbacks_done = 0;
 
        if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
                return ERR_PTR(-EINVAL);
@@ -1029,12 +1042,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
        p->did_exec = 0;
        delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
        copy_flags(clone_flags, p);
-       p->pid = pid_nr(pid);
-       retval = -EFAULT;
-       if (clone_flags & CLONE_PARENT_SETTID)
-               if (put_user(p->pid, parent_tidptr))
-                       goto bad_fork_cleanup_delays_binfmt;
-
        INIT_LIST_HEAD(&p->children);
        INIT_LIST_HEAD(&p->sibling);
        p->vfork_done = NULL;
@@ -1045,6 +1052,9 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 
        p->utime = cputime_zero;
        p->stime = cputime_zero;
+       p->gtime = cputime_zero;
+       p->utimescaled = cputime_zero;
+       p->stimescaled = cputime_zero;
 
 #ifdef CONFIG_TASK_XACCT
        p->rchar = 0;           /* I/O counter: bytes read */
@@ -1055,28 +1065,29 @@ static struct task_struct *copy_process(unsigned long clone_flags,
        task_io_accounting_init(p);
        acct_clear_integrals(p);
 
-       p->it_virt_expires = cputime_zero;
+       p->it_virt_expires = cputime_zero;
        p->it_prof_expires = cputime_zero;
-       p->it_sched_expires = 0;
-       INIT_LIST_HEAD(&p->cpu_timers[0]);
-       INIT_LIST_HEAD(&p->cpu_timers[1]);
-       INIT_LIST_HEAD(&p->cpu_timers[2]);
+       p->it_sched_expires = 0;
+       INIT_LIST_HEAD(&p->cpu_timers[0]);
+       INIT_LIST_HEAD(&p->cpu_timers[1]);
+       INIT_LIST_HEAD(&p->cpu_timers[2]);
 
        p->lock_depth = -1;             /* -1 = no lock */
        do_posix_clock_monotonic_gettime(&p->start_time);
        p->real_start_time = p->start_time;
        monotonic_to_bootbased(&p->real_start_time);
+#ifdef CONFIG_SECURITY
        p->security = NULL;
+#endif
        p->io_context = NULL;
-       p->io_wait = NULL;
        p->audit_context = NULL;
-       cpuset_fork(p);
+       cgroup_fork(p);
 #ifdef CONFIG_NUMA
        p->mempolicy = mpol_copy(p->mempolicy);
        if (IS_ERR(p->mempolicy)) {
                retval = PTR_ERR(p->mempolicy);
                p->mempolicy = NULL;
-               goto bad_fork_cleanup_cpuset;
+               goto bad_fork_cleanup_cgroup;
        }
        mpol_fix_fork_child_flag(p);
 #endif
@@ -1109,10 +1120,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
        p->blocked_on = NULL; /* not blocked yet */
 #endif
 
-       p->tgid = p->pid;
-       if (clone_flags & CLONE_THREAD)
-               p->tgid = current->tgid;
-
        if ((retval = security_task_alloc(p)))
                goto bad_fork_cleanup_policy;
        if ((retval = audit_alloc(p)))
@@ -1138,18 +1145,31 @@ static struct task_struct *copy_process(unsigned long clone_flags,
        if (retval)
                goto bad_fork_cleanup_namespaces;
 
+       if (pid != &init_struct_pid) {
+               retval = -ENOMEM;
+               pid = alloc_pid(task_active_pid_ns(p));
+               if (!pid)
+                       goto bad_fork_cleanup_namespaces;
+       }
+
+       p->pid = pid_nr(pid);
+       p->tgid = p->pid;
+       if (clone_flags & CLONE_THREAD)
+               p->tgid = current->tgid;
+
        p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
        /*
         * Clear TID on mm_release()?
         */
        p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
+#ifdef CONFIG_FUTEX
        p->robust_list = NULL;
 #ifdef CONFIG_COMPAT
        p->compat_robust_list = NULL;
 #endif
        INIT_LIST_HEAD(&p->pi_state_list);
        p->pi_state_cache = NULL;
-
+#endif
        /*
         * sigaltstack should be cleared when sharing the same VM
         */
@@ -1186,6 +1206,12 @@ static struct task_struct *copy_process(unsigned long clone_flags,
        /* Perform scheduler related setup. Assign this task to a CPU. */
        sched_fork(p, clone_flags);
 
+       /* Now that the task is set up, run cgroup callbacks if
+        * necessary. We need to run them before the task is visible
+        * on the tasklist. */
+       cgroup_fork_callbacks(p);
+       cgroup_callbacks_done = 1;
+
        /* Need tasklist lock for parent etc handling! */
        write_lock_irq(&tasklist_lock);
 
@@ -1223,12 +1249,12 @@ static struct task_struct *copy_process(unsigned long clone_flags,
         * A fatal signal pending means that current will exit, so the new
         * thread can't slip out of an OOM kill (or normal SIGKILL).
         */
-       recalc_sigpending();
+       recalc_sigpending();
        if (signal_pending(current)) {
                spin_unlock(&current->sighand->siglock);
                write_unlock_irq(&tasklist_lock);
                retval = -ERESTARTNOINTR;
-               goto bad_fork_cleanup_namespaces;
+               goto bad_fork_free_pid;
        }
 
        if (clone_flags & CLONE_THREAD) {
@@ -1257,11 +1283,22 @@ static struct task_struct *copy_process(unsigned long clone_flags,
                        __ptrace_link(p, current->parent);
 
                if (thread_group_leader(p)) {
-                       p->signal->tty = current->signal->tty;
-                       p->signal->pgrp = process_group(current);
-                       set_signal_session(p->signal, process_session(current));
-                       attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
-                       attach_pid(p, PIDTYPE_SID, task_session(current));
+                       if (clone_flags & CLONE_NEWPID) {
+                               p->nsproxy->pid_ns->child_reaper = p;
+                               p->signal->tty = NULL;
+                               p->signal->pgrp = p->pid;
+                               set_task_session(p, p->pid);
+                               attach_pid(p, PIDTYPE_PGID, pid);
+                               attach_pid(p, PIDTYPE_SID, pid);
+                       } else {
+                               p->signal->tty = current->signal->tty;
+                               p->signal->pgrp = task_pgrp_nr(current);
+                               set_task_session(p, task_session_nr(current));
+                               attach_pid(p, PIDTYPE_PGID,
+                                               task_pgrp(current));
+                               attach_pid(p, PIDTYPE_SID,
+                                               task_session(current));
+                       }
 
                        list_add_tail_rcu(&p->tasks, &init_task.tasks);
                        __get_cpu_var(process_counts)++;
@@ -1274,8 +1311,12 @@ static struct task_struct *copy_process(unsigned long clone_flags,
        spin_unlock(&current->sighand->siglock);
        write_unlock_irq(&tasklist_lock);
        proc_fork_connector(p);
+       cgroup_post_fork(p);
        return p;
 
+bad_fork_free_pid:
+       if (pid != &init_struct_pid)
+               free_pid(pid);
 bad_fork_cleanup_namespaces:
        exit_task_namespaces(p);
 bad_fork_cleanup_keys:
@@ -1300,10 +1341,9 @@ bad_fork_cleanup_security:
 bad_fork_cleanup_policy:
 #ifdef CONFIG_NUMA
        mpol_free(p->mempolicy);
-bad_fork_cleanup_cpuset:
+bad_fork_cleanup_cgroup:
 #endif
-       cpuset_exit(p);
-bad_fork_cleanup_delays_binfmt:
+       cgroup_exit(p, cgroup_callbacks_done);
        delayacct_tsk_free(p);
        if (p->binfmt)
                module_put(p->binfmt->module);
@@ -1330,7 +1370,7 @@ struct task_struct * __cpuinit fork_idle(int cpu)
        struct task_struct *task;
        struct pt_regs regs;
 
-       task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL,
+       task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
                                &init_struct_pid);
        if (!IS_ERR(task))
                init_idle(task, cpu);
@@ -1369,19 +1409,16 @@ long do_fork(unsigned long clone_flags,
 {
        struct task_struct *p;
        int trace = 0;
-       struct pid *pid = alloc_pid();
        long nr;
 
-       if (!pid)
-               return -EAGAIN;
-       nr = pid->nr;
        if (unlikely(current->ptrace)) {
                trace = fork_traceflag (clone_flags);
                if (trace)
                        clone_flags |= CLONE_PTRACE;
        }
 
-       p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
+       p = copy_process(clone_flags, stack_start, regs, stack_size,
+                       child_tidptr, NULL);
        /*
         * Do this prior waking up the new thread - the thread pointer
         * might get invalid after that point, if the thread exits quickly.
@@ -1389,6 +1426,17 @@ long do_fork(unsigned long clone_flags,
        if (!IS_ERR(p)) {
                struct completion vfork;
 
+               /*
+                * this is enough to call pid_nr_ns here, but this if
+                * improves optimisation of regular fork()
+                */
+               nr = (clone_flags & CLONE_NEWPID) ?
+                       task_pid_nr_ns(p, current->nsproxy->pid_ns) :
+                               task_pid_vnr(p);
+
+               if (clone_flags & CLONE_PARENT_SETTID)
+                       put_user(nr, parent_tidptr);
+
                if (clone_flags & CLONE_VFORK) {
                        p->vfork_done = &vfork;
                        init_completion(&vfork);
@@ -1422,7 +1470,6 @@ long do_fork(unsigned long clone_flags,
                        }
                }
        } else {
-               free_pid(pid);
                nr = PTR_ERR(p);
        }
        return nr;
@@ -1432,8 +1479,7 @@ long do_fork(unsigned long clone_flags,
 #define ARCH_MIN_MMSTRUCT_ALIGN 0
 #endif
 
-static void sighand_ctor(void *data, struct kmem_cache *cachep,
-                       unsigned long flags)
+static void sighand_ctor(struct kmem_cache *cachep, void *data)
 {
        struct sighand_struct *sighand = data;
 
@@ -1600,7 +1646,7 @@ asmlinkage long sys_unshare(unsigned long unshare_flags)
        struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
        struct files_struct *fd, *new_fd = NULL;
        struct sem_undo_list *new_ulist = NULL;
-       struct nsproxy *new_nsproxy = NULL, *old_nsproxy = NULL;
+       struct nsproxy *new_nsproxy = NULL;
 
        check_unshare_flags(&unshare_flags);
 
@@ -1608,7 +1654,8 @@ asmlinkage long sys_unshare(unsigned long unshare_flags)
        err = -EINVAL;
        if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
                                CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
-                               CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER))
+                               CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
+                               CLONE_NEWNET))
                goto bad_unshare_out;
 
        if ((err = unshare_thread(unshare_flags)))
@@ -1629,14 +1676,13 @@ asmlinkage long sys_unshare(unsigned long unshare_flags)
 
        if (new_fs ||  new_mm || new_fd || new_ulist || new_nsproxy) {
 
-               task_lock(current);
-
                if (new_nsproxy) {
-                       old_nsproxy = current->nsproxy;
-                       current->nsproxy = new_nsproxy;
-                       new_nsproxy = old_nsproxy;
+                       switch_task_namespaces(current, new_nsproxy);
+                       new_nsproxy = NULL;
                }
 
+               task_lock(current);
+
                if (new_fs) {
                        fs = current->fs;
                        current->fs = new_fs;
This page took 0.029832 seconds and 5 git commands to generate.