upgrade to fedora-2.6.12-1.1398.FC4 + vserver 2.0.rc7
[linux-2.6.git] / fs / exec.c
index dd40d30..ac3f2ac 100644 (file)
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -48,6 +48,8 @@
 #include <linux/syscalls.h>
 #include <linux/rmap.h>
 #include <linux/vs_memory.h>
+#include <linux/acct.h>
+#include <linux/vs_memory.h>
 
 #include <asm/uaccess.h>
 #include <asm/mmu_context.h>
 
 int core_uses_pid;
 char core_pattern[65] = "core";
-int suid_dumpable = 0;
-
-EXPORT_SYMBOL(suid_dumpable);
 /* The maximal length of core_pattern is also specified in sysctl.c */
 
 static struct linux_binfmt *formats;
-static rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
+static DEFINE_RWLOCK(binfmt_lock);
 
 int register_binfmt(struct linux_binfmt * fmt)
 {
@@ -189,6 +188,7 @@ static int count(char __user * __user * argv, int max)
                        argv++;
                        if(++i > max)
                                return -E2BIG;
+                       cond_resched();
                }
        }
        return i;
@@ -199,7 +199,8 @@ static int count(char __user * __user * argv, int max)
  * memory to free pages in kernel mem. These are in a format ready
  * to be put directly into the top of new user memory.
  */
-int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
+static int copy_strings(int argc, char __user * __user * argv,
+                       struct linux_binprm *bprm)
 {
        struct page *kmapped_page = NULL;
        char *kaddr = NULL;
@@ -304,6 +305,7 @@ void install_arg_page(struct vm_area_struct *vma,
 {
        struct mm_struct *mm = vma->vm_mm;
        pgd_t * pgd;
+       pud_t * pud;
        pmd_t * pmd;
        pte_t * pte;
 
@@ -314,7 +316,10 @@ void install_arg_page(struct vm_area_struct *vma,
        pgd = pgd_offset(mm, address);
 
        spin_lock(&mm->page_table_lock);
-       pmd = pmd_alloc(mm, pgd, address);
+       pud = pud_alloc(mm, pgd, address);
+       if (!pud)
+               goto out;
+       pmd = pmd_alloc(mm, pud, address);
        if (!pmd)
                goto out;
        pte = pte_alloc_map(mm, pmd, address);
@@ -324,10 +329,9 @@ void install_arg_page(struct vm_area_struct *vma,
                pte_unmap(pte);
                goto out;
        }
-       // mm->rss++;
-       vx_rsspages_inc(mm);
+       inc_mm_counter(mm, rss);
        lru_cache_add_active(page);
-       set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(
+       set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
                                        page, vma->vm_page_prot))));
        page_add_anon_rmap(page, vma, address);
        pte_unmap(pte);
@@ -344,7 +348,9 @@ out_sig:
 
 #define EXTRA_STACK_VM_PAGES   20      /* random */
 
-int setup_arg_pages(struct linux_binprm *bprm, int executable_stack)
+int setup_arg_pages(struct linux_binprm *bprm,
+                   unsigned long stack_top,
+                   int executable_stack)
 {
        unsigned long stack_base;
        struct vm_area_struct *mpnt;
@@ -385,7 +391,7 @@ int setup_arg_pages(struct linux_binprm *bprm, int executable_stack)
        stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
        if (stack_base > (1 << 30))
                stack_base = 1 << 30;
-       stack_base = PAGE_ALIGN(STACK_TOP - stack_base);
+       stack_base = PAGE_ALIGN(stack_top - stack_base);
 
        /* Adjust bprm->p to point to the end of the strings. */
        bprm->p = stack_base + PAGE_SIZE * i - offset;
@@ -397,15 +403,11 @@ int setup_arg_pages(struct linux_binprm *bprm, int executable_stack)
        while (i < MAX_ARG_PAGES)
                bprm->page[i++] = NULL;
 #else
-#ifdef __HAVE_ARCH_ALIGN_STACK
-       stack_base = arch_align_stack(STACK_TOP - MAX_ARG_PAGES*PAGE_SIZE);
+       stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
        stack_base = PAGE_ALIGN(stack_base);
-#else
-       stack_base = STACK_TOP - MAX_ARG_PAGES * PAGE_SIZE;
-#endif
        bprm->p += stack_base;
        mm->arg_start = bprm->p;
-       arg_size = STACK_TOP - (PAGE_MASK & (unsigned long) mm->arg_start);
+       arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
 #endif
 
        arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
@@ -433,7 +435,7 @@ int setup_arg_pages(struct linux_binprm *bprm, int executable_stack)
                mpnt->vm_start = stack_base;
                mpnt->vm_end = stack_base + arg_size;
 #else
-               mpnt->vm_end = STACK_TOP;
+               mpnt->vm_end = stack_top;
                mpnt->vm_start = mpnt->vm_end - arg_size;
 #endif
                /* Adjust stack execute permissions; explicitly enable
@@ -452,7 +454,6 @@ int setup_arg_pages(struct linux_binprm *bprm, int executable_stack)
                        kmem_cache_free(vm_area_cachep, mpnt);
                        return ret;
                }
-               // mm->stack_vm = mm->total_vm = vma_pages(mpnt);
                vx_vmpages_sub(mm, mm->total_vm - vma_pages(mpnt));
                mm->stack_vm = mm->total_vm;
        }
@@ -555,6 +556,21 @@ static int exec_mmap(struct mm_struct *mm)
        old_mm = current->mm;
        mm_release(tsk, old_mm);
 
+       if (old_mm) {
+               /*
+                * Make sure that if there is a core dump in progress
+                * for the old mm, we get out and die instead of going
+                * through with the exec.  We must hold mmap_sem around
+                * checking core_waiters and changing tsk->mm.  The
+                * core-inducing thread will increment core_waiters for
+                * each thread whose ->mm == old_mm.
+                */
+               down_read(&old_mm->mmap_sem);
+               if (unlikely(old_mm->core_waiters)) {
+                       up_read(&old_mm->mmap_sem);
+                       return -EINTR;
+               }
+       }
        task_lock(tsk);
        active_mm = tsk->active_mm;
        tsk->mm = mm;
@@ -563,6 +579,7 @@ static int exec_mmap(struct mm_struct *mm)
        task_unlock(tsk);
        arch_pick_mmap_layout(mm);
        if (old_mm) {
+               up_read(&old_mm->mmap_sem);
                if (active_mm != old_mm) BUG();
                mmput(old_mm);
                return 0;
@@ -607,7 +624,7 @@ static inline int de_thread(struct task_struct *tsk)
         */
        read_lock(&tasklist_lock);
        spin_lock_irq(lock);
-       if (sig->group_exit) {
+       if (sig->flags & SIGNAL_GROUP_EXIT) {
                /*
                 * Another group action in progress, just
                 * return so that the signal is processed.
@@ -617,7 +634,6 @@ static inline int de_thread(struct task_struct *tsk)
                kmem_cache_free(sighand_cachep, newsighand);
                return -EAGAIN;
        }
-       sig->group_exit = 1;
        zap_other_threads(current);
        read_unlock(&tasklist_lock);
 
@@ -625,7 +641,7 @@ static inline int de_thread(struct task_struct *tsk)
         * Account for the thread group leader hanging around:
         */
        count = 2;
-       if (current->pid == current->tgid)
+       if (thread_group_leader(current))
                count = 1;
        while (atomic_read(&sig->count) > count) {
                sig->group_exit_task = current;
@@ -637,6 +653,7 @@ static inline int de_thread(struct task_struct *tsk)
        }
        sig->group_exit_task = NULL;
        sig->notify_count = 0;
+       sig->real_timer.data = (unsigned long)current;
        spin_unlock_irq(lock);
 
        /*
@@ -644,7 +661,7 @@ static inline int de_thread(struct task_struct *tsk)
         * do is to wait for the thread group leader to become inactive,
         * and to assume its PID:
         */
-       if (current->pid != current->tgid) {
+       if (!thread_group_leader(current)) {
                struct task_struct *leader = current->group_leader, *parent;
                struct dentry *proc_dentry1, *proc_dentry2;
                unsigned long exit_state, ptrace;
@@ -675,6 +692,14 @@ static inline int de_thread(struct task_struct *tsk)
                 */
                ptrace = leader->ptrace;
                parent = leader->parent;
+               if (unlikely(ptrace) && unlikely(parent == current)) {
+                       /*
+                        * Joker was ptracing his own group leader,
+                        * and now he wants to be his own parent!
+                        * We can't have that.
+                        */
+                       ptrace = 0;
+               }
 
                ptrace_unlink(current);
                ptrace_unlink(leader);
@@ -715,7 +740,7 @@ static inline int de_thread(struct task_struct *tsk)
         * Now there are really no other threads at all,
         * so it's safe to stop telling them to kill themselves.
         */
-       sig->group_exit = 0;
+       sig->flags = 0;
 
 no_thread_group:
        BUG_ON(atomic_read(&sig->count) != 1);
@@ -751,7 +776,7 @@ no_thread_group:
 
        if (!thread_group_empty(current))
                BUG();
-       if (current->tgid != current->pid)
+       if (!thread_group_leader(current))
                BUG();
        return 0;
 }
@@ -845,13 +870,12 @@ int flush_old_exec(struct linux_binprm * bprm)
 
        if (current->euid == current->uid && current->egid == current->gid)
                current->mm->dumpable = 1;
-       else
-               current->mm->dumpable = suid_dumpable;
-               
        name = bprm->filename;
+
+       /* Copies the binary name from after last slash */
        for (i=0; (ch = *(name++)) != '\0';) {
                if (ch == '/')
-                       i = 0;
+                       i = 0; /* overwrite what we wrote */
                else
                        if (i < (sizeof(tcomm) - 1))
                                tcomm[i++] = ch;
@@ -859,14 +883,14 @@ int flush_old_exec(struct linux_binprm * bprm)
        tcomm[i] = '\0';
        set_task_comm(current, tcomm);
 
-       current->flags &= ~PF_RELOCEXEC;
+       current->flags &= ~PF_RANDOMIZE;
        flush_thread();
 
        if (bprm->e_uid != current->euid || bprm->e_gid != current->egid || 
            permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
            (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
                suid_keys(current);
-               current->mm->dumpable = suid_dumpable;
+               current->mm->dumpable = 0;
        }
 
        /* An exec changes our domain. We are no longer part of the thread
@@ -970,6 +994,7 @@ void compute_creds(struct linux_binprm *bprm)
        unsafe = unsafe_exec(current);
        security_bprm_apply_creds(bprm, unsafe);
        task_unlock(current);
+       security_bprm_post_apply_creds(bprm);
 }
 
 EXPORT_SYMBOL(compute_creds);
@@ -1173,6 +1198,8 @@ int do_execve(char * filename,
 
                /* execve success */
                security_bprm_free(bprm);
+               acct_update_integrals(current);
+               update_mem_hiwater(current);
                kfree(bprm);
                return retval;
        }
@@ -1366,53 +1393,18 @@ static void zap_threads (struct mm_struct *mm)
 
        read_unlock(&tasklist_lock);
 
-       while (unlikely(traced)) {
+       if (unlikely(traced)) {
                /*
                 * We are zapping a thread and the thread it ptraces.
-                * The tracee won't come out of TASK_TRACED state until
-                * its ptracer detaches.  That happens when the ptracer
-                * dies, but it synchronizes with us and so won't get
-                * that far until we finish the core dump.  If we're
-                * waiting for the tracee to synchronize but it stays
-                * blocked in TASK_TRACED, then we deadlock.  So, for
-                * this weirdo case we have to do another round with
-                * tasklist_lock write-locked to __ptrace_unlink the
-                * children that might cause this deadlock.  That will
-                * wake them up to process their pending SIGKILL.
-                *
-                * First, give everyone we just killed a chance to run
-                * so they can all get into the coredump synchronization.
-                * That should leave only the TASK_TRACED stragglers for
-                * us to wake up.  If a ptracer is still running, we'll
-                * have to come around again after letting it finish.
+                * If the tracee went into a ptrace stop for exit tracing,
+                * we could deadlock since the tracer is waiting for this
+                * coredump to finish.  Detach them so they can both die.
                 */
-               yield();
-               traced = 0;
                write_lock_irq(&tasklist_lock);
                do_each_thread(g,p) {
-                       if (mm != p->mm || p == tsk ||
-                           !p->ptrace || p->parent->mm != mm)
-                               continue;
-                       if ((p->parent->flags & (PF_SIGNALED|PF_EXITING)) ||
-                           (p->parent->state & (TASK_TRACED|TASK_STOPPED))) {
-                               /*
-                                * The parent is in the process of exiting
-                                * itself, or else it's stopped right now.
-                                * It cannot be in a ptrace call, and would
-                                * have to read_lock tasklist_lock before
-                                * it could start one, so we are safe here.
-                                */
+                       if (mm == p->mm && p != tsk &&
+                           p->ptrace && p->parent->mm == mm) {
                                __ptrace_unlink(p);
-                       } else {
-                               /*
-                                * Blargh!  The ptracer is not dying
-                                * yet, so we cannot be sure that it
-                                * isn't in the middle of a ptrace call.
-                                * We'll have to let it run to get into
-                                * coredump_wait and come around another
-                                * time to detach its tracee.
-                                */
-                               traced = 1;
                        }
                } while_each_thread(g,p);
                write_unlock_irq(&tasklist_lock);
@@ -1446,8 +1438,6 @@ int do_coredump(long signr, int exit_code, struct pt_regs * regs)
        struct inode * inode;
        struct file * file;
        int retval = 0;
-       int fsuid = current->fsuid;
-       int flag = 0;
 
        binfmt = current->binfmt;
        if (!binfmt || !binfmt->core_dump)
@@ -1459,22 +1449,21 @@ int do_coredump(long signr, int exit_code, struct pt_regs * regs)
                up_write(&mm->mmap_sem);
                goto fail;
        }
-
-       /*
-        *      We cannot trust fsuid as being the "true" uid of the
-        *      process nor do we know its entire history. We only know it
-        *      was tainted so we dump it as root in mode 2.
-        */
-       if (mm->dumpable == 2) {        /* Setuid core dump mode */
-               flag = O_EXCL;          /* Stop rewrite attacks */
-               current->fsuid = 0;     /* Dump root private */
-       }
        mm->dumpable = 0;
        init_completion(&mm->core_done);
-       current->signal->group_exit = 1;
+       spin_lock_irq(&current->sighand->siglock);
+       current->signal->flags = SIGNAL_GROUP_EXIT;
        current->signal->group_exit_code = exit_code;
+       spin_unlock_irq(&current->sighand->siglock);
        coredump_wait(mm);
 
+       /*
+        * Clear any false indication of pending signals that might
+        * be seen by the filesystem code called to write the core file.
+        */
+       current->signal->group_stop_count = 0;
+       clear_thread_flag(TIF_SIGPENDING);
+
        if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
                goto fail_unlock;
 
@@ -1485,7 +1474,7 @@ int do_coredump(long signr, int exit_code, struct pt_regs * regs)
        lock_kernel();
        format_corename(corename, core_pattern, signr);
        unlock_kernel();
-       file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
+       file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
        if (IS_ERR(file))
                goto fail_unlock;
        inode = file->f_dentry->d_inode;
@@ -1510,7 +1499,6 @@ int do_coredump(long signr, int exit_code, struct pt_regs * regs)
 close_fail:
        filp_close(file, NULL);
 fail_unlock:
-       current->fsuid = fsuid;
        complete_all(&mm->core_done);
 fail:
        return retval;