uml: cover stubs with a VMA

Give the stubs a VMA.  This allows the removal of a truly nasty kludge to make
sure that mm->nr_ptes was correct in exit_mmap.  The underlying problem was
always that the stubs, which have ptes, and thus allocated a page table,
weren't covered by a VMA.

This patch fixes that by using install_special_mapping in arch_dup_mmap and
activate_context to create the VMA.  The stubs have to be moved, since
shift_arg_pages seems to assume that the stack is the only VMA present at that
point during exec, and uses vma_adjust to fiddle its VMA.  However, that
extends the stub VMA by the amount removed from the stack VMA.

To avoid this problem, the stubs were moved to a different fixed location at
the start of the address space.

The init_stub_pte calls were moved from init_new_context to arch_dup_mmap
because I was occasionally seeing arch_dup_mmap not being called, causing
exit_mmap to die.  Rather than figure out what was really happening, I decided
it was cleaner to just move the calls so that there's no doubt that both the
pte and VMA creation happen, no matter what.  arch_exit_mmap is used to clear
the stub ptes at exit time.

The STUB_* constants in as-layout.h no longer depend on UM_TASK_SIZE, that
that definition is removed, along with the comments complaining about gcc.

Because the stubs are no longer at the top of the address space, some care is
needed while flushing TLBs.  update_pte_range checks for addresses in the stub
range and skips them.  flush_thread now issues two unmaps, one for the range
before STUB_START and one for the range after STUB_END.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

arch/um/include/as-layout.h

@@ -29,21 +29,10 @@
 #define _AC(X, Y)	__AC(X, Y)
 #endif
 
-/*
+#define STUB_START _AC(, 0x100000)
- * The "- 1"'s are to avoid gcc complaining about integer overflows
+#define STUB_CODE _AC((unsigned long), STUB_START)
- * and unrepresentable decimal constants.  With 3-level page tables,
+#define STUB_DATA _AC((unsigned long), STUB_CODE + UM_KERN_PAGE_SIZE)
- * TASK_SIZE is 0x80000000, which gets turned into its signed decimal
+#define STUB_END _AC((unsigned long), STUB_DATA + UM_KERN_PAGE_SIZE)
- * equivalent in asm-offsets.s.  gcc then complains about that being
- * unsigned only in C90.  To avoid that, UM_TASK_SIZE is defined as
- * TASK_SIZE - 1.  To compensate, we need to add the 1 back here.
- * However, adding it back to UM_TASK_SIZE produces more gcc
- * complaints.  So, I adjust the thing being subtracted from
- * UM_TASK_SIZE instead.  Bah.
- */
-#define STUB_CODE _AC((unsigned long), \
-		      UM_TASK_SIZE - (2 * UM_KERN_PAGE_SIZE - 1))
-#define STUB_DATA _AC((unsigned long), UM_TASK_SIZE - (UM_KERN_PAGE_SIZE - 1))
-#define STUB_START _AC(, STUB_CODE)
 
 #ifndef __ASSEMBLY__
 

arch/um/include/common-offsets.h

@@ -39,6 +39,3 @@ DEFINE(UM_HZ, HZ);
 DEFINE(UM_USEC_PER_SEC, USEC_PER_SEC);
 DEFINE(UM_NSEC_PER_SEC, NSEC_PER_SEC);
 DEFINE(UM_NSEC_PER_USEC, NSEC_PER_USEC);
-
-/* See as-layout.h for an explanation of the "- 1".  Bah. */
-DEFINE(UM_TASK_SIZE, TASK_SIZE - 1);

arch/um/kernel/exec.c

@@ -19,12 +19,13 @@
 void flush_thread(void)
 {
 	void *data = NULL;
-	unsigned long end = proc_mm ? TASK_SIZE : STUB_START;
 	int ret;
 
 	arch_flush_thread(&current->thread.arch);
 
-	ret = unmap(&current->mm->context.id, 0, end, 1, &data);
+	ret = unmap(&current->mm->context.id, 0, STUB_START, 0, &data);
+	ret = ret || unmap(&current->mm->context.id, STUB_END,
+			   TASK_SIZE - STUB_END, 1, &data);
 	if (ret) {
 		printk(KERN_ERR "flush_thread - clearing address space failed, "
 		       "err = %d\n", ret);

arch/um/kernel/skas/mmu.c

@@ -34,25 +34,6 @@ static int init_stub_pte(struct mm_struct *mm, unsigned long proc,
 	if (!pte)
 		goto out_pte;
 
-	/*
-	 * There's an interaction between the skas0 stub pages, stack
-	 * randomization, and the BUG at the end of exit_mmap.  exit_mmap
-	 * checks that the number of page tables freed is the same as had
-	 * been allocated.  If the stack is on the last page table page,
-	 * then the stack pte page will be freed, and if not, it won't.  To
-	 * avoid having to know where the stack is, or if the process mapped
-	 * something at the top of its address space for some other reason,
-	 * we set TASK_SIZE to end at the start of the last page table.
-	 * This keeps exit_mmap off the last page, but introduces a leak
-	 * of that page.  So, we hang onto it here and free it in
-	 * destroy_context_skas.
-	 */
-
-	mm->context.last_page_table = pmd_page_vaddr(*pmd);
-#ifdef CONFIG_3_LEVEL_PGTABLES
-	mm->context.last_pmd = (unsigned long) __va(pud_val(*pud));
-#endif
-
 	*pte = mk_pte(virt_to_page(kernel), __pgprot(_PAGE_PRESENT));
 	*pte = pte_mkread(*pte);
 	return 0;
@@ -76,24 +57,6 @@ int init_new_context(struct task_struct *task, struct mm_struct *mm)
 		stack = get_zeroed_page(GFP_KERNEL);
 		if (stack == 0)
 			goto out;
-
-		/*
-		 * This zeros the entry that pgd_alloc didn't, needed since
-		 * we are about to reinitialize it, and want mm.nr_ptes to
-		 * be accurate.
-		 */
-		mm->pgd[USER_PTRS_PER_PGD] = __pgd(0);
-
-		ret = init_stub_pte(mm, STUB_CODE,
-				    (unsigned long) &__syscall_stub_start);
-		if (ret)
-			goto out_free;
-
-		ret = init_stub_pte(mm, STUB_DATA, stack);
-		if (ret)
-			goto out_free;
-
-		mm->nr_ptes--;
 	}
 
 	to_mm->id.stack = stack;
@@ -137,6 +100,64 @@ int init_new_context(struct task_struct *task, struct mm_struct *mm)
 	return ret;
 }
 
+void arch_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm)
+{
+	struct page **pages;
+	int err, ret;
+
+	if (!skas_needs_stub)
+		return;
+
+	ret = init_stub_pte(mm, STUB_CODE,
+			    (unsigned long) &__syscall_stub_start);
+	if (ret)
+		goto out;
+
+	ret = init_stub_pte(mm, STUB_DATA, mm->context.id.stack);
+	if (ret)
+		goto out;
+
+	pages = kmalloc(2 * sizeof(struct page *), GFP_KERNEL);
+	if (pages == NULL) {
+		printk(KERN_ERR "arch_dup_mmap failed to allocate 2 page "
+		       "pointers\n");
+		goto out;
+	}
+
+	pages[0] = virt_to_page(&__syscall_stub_start);
+	pages[1] = virt_to_page(mm->context.id.stack);
+
+	/* dup_mmap already holds mmap_sem */
+	err = install_special_mapping(mm, STUB_START, STUB_END - STUB_START,
+				      VM_READ | VM_MAYREAD | VM_EXEC |
+				      VM_MAYEXEC | VM_DONTCOPY, pages);
+	if (err) {
+		printk(KERN_ERR "install_special_mapping returned %d\n", err);
+		goto out_free;
+	}
+	return;
+
+out_free:
+	kfree(pages);
+out:
+	force_sigsegv(SIGSEGV, current);
+}
+
+void arch_exit_mmap(struct mm_struct *mm)
+{
+	pte_t *pte;
+
+	pte = virt_to_pte(mm, STUB_CODE);
+	if (pte != NULL)
+		pte_clear(mm, STUB_CODE, pte);
+
+	pte = virt_to_pte(mm, STUB_DATA);
+	if (pte == NULL)
+		return;
+
+	pte_clear(mm, STUB_DATA, pte);
+}
+
 void destroy_context(struct mm_struct *mm)
 {
 	struct mm_context *mmu = &mm->context;
@@ -146,15 +167,8 @@ void destroy_context(struct mm_struct *mm)
 	else
 		os_kill_ptraced_process(mmu->id.u.pid, 1);
 
-	if (!proc_mm || !ptrace_faultinfo) {
+	if (skas_needs_stub)
 		free_page(mmu->id.stack);
-		pte_lock_deinit(virt_to_page(mmu->last_page_table));
-		pte_free_kernel(mm, (pte_t *) mmu->last_page_table);
-		dec_zone_page_state(virt_to_page(mmu->last_page_table), NR_PAGETABLE);
-#ifdef CONFIG_3_LEVEL_PGTABLES
-		pmd_free(mm, (pmd_t *) mmu->last_pmd);
-#endif
-	}
 
 	free_ldt(mmu);
 }

arch/um/kernel/tlb.c

@@ -184,6 +184,9 @@ static inline int update_pte_range(pmd_t *pmd, unsigned long addr,
 
 	pte = pte_offset_kernel(pmd, addr);
 	do {
+		if ((addr >= STUB_START) && (addr < STUB_END))
+			continue;
+
 		r = pte_read(*pte);
 		w = pte_write(*pte);
 		x = pte_exec(*pte);
@@ -486,9 +489,6 @@ void __flush_tlb_one(unsigned long addr)
 static void fix_range(struct mm_struct *mm, unsigned long start_addr,
 		      unsigned long end_addr, int force)
 {
-	if (!proc_mm && (end_addr > STUB_START))
-		end_addr = STUB_START;
-
 	fix_range_common(mm, start_addr, end_addr, force);
 }
 
@@ -502,8 +502,6 @@ void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
 
 void flush_tlb_mm(struct mm_struct *mm)
 {
-	unsigned long end;
-
 	/*
 	 * Don't bother flushing if this address space is about to be
 	 * destroyed.
@@ -511,8 +509,7 @@ void flush_tlb_mm(struct mm_struct *mm)
 	if (atomic_read(&mm->mm_users) == 0)
 		return;
 
-	end = proc_mm ? TASK_SIZE : STUB_START;
+	fix_range(mm, 0, TASK_SIZE, 0);
-	fix_range(mm, 0, end, 0);
 }
 
 void force_flush_all(void)

include/asm-um/mmu_context.h

@@ -6,11 +6,12 @@
 #ifndef __UM_MMU_CONTEXT_H
 #define __UM_MMU_CONTEXT_H
 
-#include <asm-generic/mm_hooks.h>
-
 #include "linux/sched.h"
 #include "um_mmu.h"
 
+extern void arch_dup_mmap(struct mm_struct *oldmm, struct mm_struct *mm);
+extern void arch_exit_mmap(struct mm_struct *mm);
+
 #define get_mmu_context(task) do ; while(0)
 #define activate_context(tsk) do ; while(0)
 
@@ -30,6 +31,8 @@ static inline void activate_mm(struct mm_struct *old, struct mm_struct *new)
 	 */
 	if (old != new && (current->flags & PF_BORROWED_MM))
 		__switch_mm(&new->context.id);
+
+	arch_dup_mmap(old, new);
 }
 
 static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next,