[PATCH] i386: Make COMPAT_VDSO runtime selectable.

Now that relocation of the VDSO for COMPAT_VDSO users is done at runtime rather than compile time, it is possible to enable/disable compat mode at runtime. This patch allows you to enable COMPAT_VDSO mode with "vdso=2" on the kernel command line, or via sysctl. (Switching on a running system shouldn't be done lightly; any process which was relying on the compat VDSO will be upset if it goes away.) The COMPAT_VDSO config option still exists, but if enabled it just makes vdso_enabled default to VDSO_COMPAT. +From: Hugh Dickins <hugh@veritas.com> Fix oops from i386-make-compat_vdso-runtime-selectable.patch. Even mingetty at system startup finds it easy to trigger an oops while reading /proc/PID/maps: though it has a good hold on the mm itself, that cannot stop exit_mm() from resetting tsk->mm to NULL. (It is usually show_map()'s call to get_gate_vma() which oopses, and I expect we could change that to check priv->tail_vma instead; but no matter, even m_start()'s call just after get_task_mm() is racy.) Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com> Signed-off-by: Andi Kleen <ak@suse.de> Cc: Zachary Amsden <zach@vmware.com> Cc: "Jan Beulich" <JBeulich@novell.com> Cc: Eric W. Biederman <ebiederm@xmission.com> Cc: Andi Kleen <ak@suse.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland McGrath <roland@redhat.com>

[PATCH] i386: Make COMPAT_VDSO runtime selectable.
Now that relocation of the VDSO for COMPAT_VDSO users is done at runtime rather than compile time, it is possible to enable/disable compat mode at runtime. This patch allows you to enable COMPAT_VDSO mode with "vdso=2" on the kernel command line, or via sysctl. (Switching on a running system shouldn't be done lightly; any process which was relying on the compat VDSO will be upset if it goes away.) The COMPAT_VDSO config option still exists, but if enabled it just makes vdso_enabled default to VDSO_COMPAT. +From: Hugh Dickins <hugh@veritas.com> Fix oops from i386-make-compat_vdso-runtime-selectable.patch. Even mingetty at system startup finds it easy to trigger an oops while reading /proc/PID/maps: though it has a good hold on the mm itself, that cannot stop exit_mm() from resetting tsk->mm to NULL. (It is usually show_map()'s call to get_gate_vma() which oopses, and I expect we could change that to check priv->tail_vma instead; but no matter, even m_start()'s call just after get_task_mm() is racy.) Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com> Signed-off-by: Andi Kleen <ak@suse.de> Cc: Zachary Amsden <zach@vmware.com> Cc: "Jan Beulich" <JBeulich@novell.com> Cc: Eric W. Biederman <ebiederm@xmission.com> Cc: Andi Kleen <ak@suse.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: Roland McGrath <roland@redhat.com>
Jeremy Fitzhardinge · Andi Kleen
1 parent d4f7a2c18e
Showing 3 changed files with 95 additions and 53 deletions Side-by-side Diff
Documentation/kernel-parameters.txt
arch/i386/kernel/sysenter.c
include/asm-i386/page.h
@@ -1820,6 +1820,7 @@
 			[USBHID] The interval which mice are to be polled at.
  
 	vdso=		[IA-32,SH]
+			vdso=2: enable compat VDSO (default with COMPAT_VDSO)
 			vdso=1: enable VDSO (default)
 			vdso=0: disable VDSO mapping
  
@@ -23,16 +23,25 @@
 #include <asm/pgtable.h>
 #include <asm/unistd.h>
 #include <asm/elf.h>
+#include <asm/tlbflush.h>
  
+enum {
+	VDSO_DISABLED = 0,
+	VDSO_ENABLED = 1,
+	VDSO_COMPAT = 2,
+};
+
+#ifdef CONFIG_COMPAT_VDSO
+#define VDSO_DEFAULT	VDSO_COMPAT
+#else
+#define VDSO_DEFAULT	VDSO_ENABLED
+#endif
+
 /*
  * Should the kernel map a VDSO page into processes and pass its
  * address down to glibc upon exec()?
  */
-#ifdef CONFIG_PARAVIRT
-unsigned int __read_mostly vdso_enabled = 0;
-#else
-unsigned int __read_mostly vdso_enabled = 1;
-#endif
+unsigned int __read_mostly vdso_enabled = VDSO_DEFAULT;
  
 EXPORT_SYMBOL_GPL(vdso_enabled);
  
@@ -47,7 +56,6 @@
  
 extern asmlinkage void sysenter_entry(void);
  
-#ifdef CONFIG_COMPAT_VDSO
 static __init void reloc_symtab(Elf32_Ehdr *ehdr,
 				unsigned offset, unsigned size)
 {
@@ -164,11 +172,6 @@
 				     shdr[i].sh_size);
 	}
 }
-#else
-static inline void relocate_vdso(Elf32_Ehdr *ehdr)
-{
-}
-#endif	/* COMPAT_VDSO */
  
 void enable_sep_cpu(void)
 {
@@ -188,6 +191,25 @@
 	put_cpu();	
 }
  
+static struct vm_area_struct gate_vma;
+
+static int __init gate_vma_init(void)
+{
+	gate_vma.vm_mm = NULL;
+	gate_vma.vm_start = FIXADDR_USER_START;
+	gate_vma.vm_end = FIXADDR_USER_END;
+	gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
+	gate_vma.vm_page_prot = __P101;
+	/*
+	 * Make sure the vDSO gets into every core dump.
+	 * Dumping its contents makes post-mortem fully interpretable later
+	 * without matching up the same kernel and hardware config to see
+	 * what PC values meant.
+	 */
+	gate_vma.vm_flags |= VM_ALWAYSDUMP;
+	return 0;
+}
+
 /*
  * These symbols are defined by vsyscall.o to mark the bounds
  * of the ELF DSO images included therein.
@@ -196,6 +218,22 @@
 extern const char vsyscall_sysenter_start, vsyscall_sysenter_end;
 static struct page *syscall_pages[1];
  
+static void map_compat_vdso(int map)
+{
+	static int vdso_mapped;
+
+	if (map == vdso_mapped)
+		return;
+
+	vdso_mapped = map;
+
+	__set_fixmap(FIX_VDSO, page_to_pfn(syscall_pages[0]) << PAGE_SHIFT,
+		     map ? PAGE_READONLY_EXEC : PAGE_NONE);
+
+	/* flush stray tlbs */
+	flush_tlb_all();
+}
+
 int __init sysenter_setup(void)
 {
 	void *syscall_page = (void *)get_zeroed_page(GFP_ATOMIC);
  
@@ -204,10 +242,9 @@
  
 	syscall_pages[0] = virt_to_page(syscall_page);
  
-#ifdef CONFIG_COMPAT_VDSO
-	__set_fixmap(FIX_VDSO, __pa(syscall_page), PAGE_READONLY_EXEC);
+	gate_vma_init();
+
 	printk("Compat vDSO mapped to %08lx.\n", __fix_to_virt(FIX_VDSO));
-#endif
  
 	if (!boot_cpu_has(X86_FEATURE_SEP)) {
 		vsyscall = &vsyscall_int80_start;
  
  
  
  
  
  
@@ -226,42 +263,57 @@
 /* Defined in vsyscall-sysenter.S */
 extern void SYSENTER_RETURN;
  
-#ifdef __HAVE_ARCH_GATE_AREA
 /* Setup a VMA at program startup for the vsyscall page */
 int arch_setup_additional_pages(struct linux_binprm *bprm, int exstack)
 {
 	struct mm_struct *mm = current->mm;
 	unsigned long addr;
 	int ret;
+	bool compat;
  
 	down_write(&mm->mmap_sem);
-	addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
-	if (IS_ERR_VALUE(addr)) {
-		ret = addr;
-		goto up_fail;
-	}
  
-	/*
-	 * MAYWRITE to allow gdb to COW and set breakpoints
-	 *
-	 * Make sure the vDSO gets into every core dump.
-	 * Dumping its contents makes post-mortem fully interpretable later
-	 * without matching up the same kernel and hardware config to see
-	 * what PC values meant.
-	 */
-	ret = install_special_mapping(mm, addr, PAGE_SIZE,
-				      VM_READ|VM_EXEC|
-				      VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
-				      VM_ALWAYSDUMP,
-				      syscall_pages);
-	if (ret)
-		goto up_fail;
+	/* Test compat mode once here, in case someone
+	   changes it via sysctl */
+	compat = (vdso_enabled == VDSO_COMPAT);
  
+	map_compat_vdso(compat);
+
+	if (compat)
+		addr = VDSO_HIGH_BASE;
+	else {
+		addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
+		if (IS_ERR_VALUE(addr)) {
+			ret = addr;
+			goto up_fail;
+		}
+
+		/*
+		 * MAYWRITE to allow gdb to COW and set breakpoints
+		 *
+		 * Make sure the vDSO gets into every core dump.
+		 * Dumping its contents makes post-mortem fully
+		 * interpretable later without matching up the same
+		 * kernel and hardware config to see what PC values
+		 * meant.
+		 */
+		ret = install_special_mapping(mm, addr, PAGE_SIZE,
+					      VM_READ|VM_EXEC|
+					      VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC|
+					      VM_ALWAYSDUMP,
+					      syscall_pages);
+
+		if (ret)
+			goto up_fail;
+	}
+
 	current->mm->context.vdso = (void *)addr;
 	current_thread_info()->sysenter_return =
-				    (void *)VDSO_SYM(&SYSENTER_RETURN);
-up_fail:
+		(void *)VDSO_SYM(&SYSENTER_RETURN);
+
+  up_fail:
 	up_write(&mm->mmap_sem);
+
 	return ret;
 }
  
@@ -274,6 +326,11 @@
  
 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
 {
+	struct mm_struct *mm = tsk->mm;
+
+	/* Check to see if this task was created in compat vdso mode */
+	if (mm && mm->context.vdso == (void *)VDSO_HIGH_BASE)
+		return &gate_vma;
 	return NULL;
 }
  
@@ -286,18 +343,4 @@
 {
 	return 0;
 }
-#else  /* !__HAVE_ARCH_GATE_AREA */
-int arch_setup_additional_pages(struct linux_binprm *bprm, int exstack)
-{
-	/*
-	 * If not creating userspace VMA, simply set vdso to point to
-	 * fixmap page.
-	 */
-	current->mm->context.vdso = (void *)VDSO_HIGH_BASE;
-	current_thread_info()->sysenter_return =
-		(void *)VDSO_SYM(&SYSENTER_RETURN);
-
-	return 0;
-}
-#endif	/* __HAVE_ARCH_GATE_AREA */
@@ -143,9 +143,7 @@
 #include <asm-generic/memory_model.h>
 #include <asm-generic/page.h>
  
-#ifndef CONFIG_COMPAT_VDSO
 #define __HAVE_ARCH_GATE_AREA 1
-#endif
 #endif /* __KERNEL__ */
  
 #endif /* _I386_PAGE_H */
...	...	@@ -1820,6 +1820,7 @@
1820	1820	[USBHID] The interval which mice are to be polled at.
1821	1821
1822	1822	vdso= [IA-32,SH]
	1823	+ vdso=2: enable compat VDSO (default with COMPAT_VDSO)
1823	1824	vdso=1: enable VDSO (default)
1824	1825	vdso=0: disable VDSO mapping
1825	1826
...	...	@@ -23,16 +23,25 @@
23	23	#include <asm/pgtable.h>
24	24	#include <asm/unistd.h>
25	25	#include <asm/elf.h>
	26	+#include <asm/tlbflush.h>
26	27
	28	+enum {
	29	+ VDSO_DISABLED = 0,
	30	+ VDSO_ENABLED = 1,
	31	+ VDSO_COMPAT = 2,
	32	+};
	33	+
	34	+#ifdef CONFIG_COMPAT_VDSO
	35	+#define VDSO_DEFAULT VDSO_COMPAT
	36	+#else
	37	+#define VDSO_DEFAULT VDSO_ENABLED
	38	+#endif
	39	+
27	40	/*
28	41	* Should the kernel map a VDSO page into processes and pass its
29	42	* address down to glibc upon exec()?
30	43	*/
31		-#ifdef CONFIG_PARAVIRT
32		-unsigned int __read_mostly vdso_enabled = 0;
33		-#else
34		-unsigned int __read_mostly vdso_enabled = 1;
35		-#endif
	44	+unsigned int __read_mostly vdso_enabled = VDSO_DEFAULT;
36	45
37	46	EXPORT_SYMBOL_GPL(vdso_enabled);
38	47
...	...	@@ -47,7 +56,6 @@
47	56
48	57	extern asmlinkage void sysenter_entry(void);
49	58
50		-#ifdef CONFIG_COMPAT_VDSO
51	59	static __init void reloc_symtab(Elf32_Ehdr *ehdr,
52	60	unsigned offset, unsigned size)
53	61	{
...	...	@@ -164,11 +172,6 @@
164	172	shdr[i].sh_size);
165	173	}
166	174	}
167		-#else
168		-static inline void relocate_vdso(Elf32_Ehdr *ehdr)
169		-{
170		-}
171		-#endif /* COMPAT_VDSO */
172	175
173	176	void enable_sep_cpu(void)
174	177	{
...	...	@@ -188,6 +191,25 @@
188	191	put_cpu();
189	192	}
190	193
	194	+static struct vm_area_struct gate_vma;
	195	+
	196	+static int __init gate_vma_init(void)
	197	+{
	198	+ gate_vma.vm_mm = NULL;
	199	+ gate_vma.vm_start = FIXADDR_USER_START;
	200	+ gate_vma.vm_end = FIXADDR_USER_END;
	201	+ gate_vma.vm_flags = VM_READ \| VM_MAYREAD \| VM_EXEC \| VM_MAYEXEC;
	202	+ gate_vma.vm_page_prot = __P101;
	203	+ /*
	204	+ * Make sure the vDSO gets into every core dump.
	205	+ * Dumping its contents makes post-mortem fully interpretable later
	206	+ * without matching up the same kernel and hardware config to see
	207	+ * what PC values meant.
	208	+ */
	209	+ gate_vma.vm_flags \|= VM_ALWAYSDUMP;
	210	+ return 0;
	211	+}
	212	+
191	213	/*
192	214	* These symbols are defined by vsyscall.o to mark the bounds
193	215	* of the ELF DSO images included therein.
...	...	@@ -196,6 +218,22 @@
196	218	extern const char vsyscall_sysenter_start, vsyscall_sysenter_end;
197	219	static struct page *syscall_pages[1];
198	220
	221	+static void map_compat_vdso(int map)
	222	+{
	223	+ static int vdso_mapped;
	224	+
	225	+ if (map == vdso_mapped)
	226	+ return;
	227	+
	228	+ vdso_mapped = map;
	229	+
	230	+ __set_fixmap(FIX_VDSO, page_to_pfn(syscall_pages[0]) << PAGE_SHIFT,
	231	+ map ? PAGE_READONLY_EXEC : PAGE_NONE);
	232	+
	233	+ /* flush stray tlbs */
	234	+ flush_tlb_all();
	235	+}
	236	+
199	237	int __init sysenter_setup(void)
200	238	{
201	239	void syscall_page = (void )get_zeroed_page(GFP_ATOMIC);
202	240
...	...	@@ -204,10 +242,9 @@
204	242
205	243	syscall_pages[0] = virt_to_page(syscall_page);
206	244
207		-#ifdef CONFIG_COMPAT_VDSO
208		- __set_fixmap(FIX_VDSO, __pa(syscall_page), PAGE_READONLY_EXEC);
	245	+ gate_vma_init();
	246	+
209	247	printk("Compat vDSO mapped to %08lx.\n", __fix_to_virt(FIX_VDSO));
210		-#endif
211	248
212	249	if (!boot_cpu_has(X86_FEATURE_SEP)) {
213	250	vsyscall = &vsyscall_int80_start;
214	251
215	252
216	253
217	254
218	255
219	256
...	...	@@ -226,42 +263,57 @@
226	263	/* Defined in vsyscall-sysenter.S */
227	264	extern void SYSENTER_RETURN;
228	265
229		-#ifdef __HAVE_ARCH_GATE_AREA
230	266	/* Setup a VMA at program startup for the vsyscall page */
231	267	int arch_setup_additional_pages(struct linux_binprm *bprm, int exstack)
232	268	{
233	269	struct mm_struct *mm = current->mm;
234	270	unsigned long addr;
235	271	int ret;
	272	+ bool compat;
236	273
237	274	down_write(&mm->mmap_sem);
238		- addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
239		- if (IS_ERR_VALUE(addr)) {
240		- ret = addr;
241		- goto up_fail;
242		- }
243	275
244		- /*
245		- * MAYWRITE to allow gdb to COW and set breakpoints
246		- *
247		- * Make sure the vDSO gets into every core dump.
248		- * Dumping its contents makes post-mortem fully interpretable later
249		- * without matching up the same kernel and hardware config to see
250		- * what PC values meant.
251		- */
252		- ret = install_special_mapping(mm, addr, PAGE_SIZE,
253		- VM_READ\|VM_EXEC\|
254		- VM_MAYREAD\|VM_MAYWRITE\|VM_MAYEXEC\|
255		- VM_ALWAYSDUMP,
256		- syscall_pages);
257		- if (ret)
258		- goto up_fail;
	276	+ /* Test compat mode once here, in case someone
	277	+ changes it via sysctl */
	278	+ compat = (vdso_enabled == VDSO_COMPAT);
259	279
	280	+ map_compat_vdso(compat);
	281	+
	282	+ if (compat)
	283	+ addr = VDSO_HIGH_BASE;
	284	+ else {
	285	+ addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
	286	+ if (IS_ERR_VALUE(addr)) {
	287	+ ret = addr;
	288	+ goto up_fail;
	289	+ }
	290	+
	291	+ /*
	292	+ * MAYWRITE to allow gdb to COW and set breakpoints
	293	+ *
	294	+ * Make sure the vDSO gets into every core dump.
	295	+ * Dumping its contents makes post-mortem fully
	296	+ * interpretable later without matching up the same
	297	+ * kernel and hardware config to see what PC values
	298	+ * meant.
	299	+ */
	300	+ ret = install_special_mapping(mm, addr, PAGE_SIZE,
	301	+ VM_READ\|VM_EXEC\|
	302	+ VM_MAYREAD\|VM_MAYWRITE\|VM_MAYEXEC\|
	303	+ VM_ALWAYSDUMP,
	304	+ syscall_pages);
	305	+
	306	+ if (ret)
	307	+ goto up_fail;
	308	+ }
	309	+
260	310	current->mm->context.vdso = (void *)addr;
261	311	current_thread_info()->sysenter_return =
262		- (void *)VDSO_SYM(&SYSENTER_RETURN);
263		-up_fail:
	312	+ (void *)VDSO_SYM(&SYSENTER_RETURN);
	313	+
	314	+ up_fail:
264	315	up_write(&mm->mmap_sem);
	316	+
265	317	return ret;
266	318	}
267	319
...	...	@@ -274,6 +326,11 @@
274	326
275	327	struct vm_area_struct get_gate_vma(struct task_struct tsk)
276	328	{
	329	+ struct mm_struct *mm = tsk->mm;
	330	+
	331	+ /* Check to see if this task was created in compat vdso mode */
	332	+ if (mm && mm->context.vdso == (void *)VDSO_HIGH_BASE)
	333	+ return &gate_vma;
277	334	return NULL;
278	335	}
279	336
...	...	@@ -286,18 +343,4 @@
286	343	{
287	344	return 0;
288	345	}
289		-#else /* !__HAVE_ARCH_GATE_AREA */
290		-int arch_setup_additional_pages(struct linux_binprm *bprm, int exstack)
291		-{
292		- /*
293		- * If not creating userspace VMA, simply set vdso to point to
294		- * fixmap page.
295		- */
296		- current->mm->context.vdso = (void *)VDSO_HIGH_BASE;
297		- current_thread_info()->sysenter_return =
298		- (void *)VDSO_SYM(&SYSENTER_RETURN);
299		-
300		- return 0;
301		-}
302		-#endif /* __HAVE_ARCH_GATE_AREA */
...	...	@@ -143,9 +143,7 @@
143	143	#include <asm-generic/memory_model.h>
144	144	#include <asm-generic/page.h>
145	145
146		-#ifndef CONFIG_COMPAT_VDSO
147	146	#define __HAVE_ARCH_GATE_AREA 1
148		-#endif
149	147	#endif /* __KERNEL__ */
150	148
151	149	#endif /* _I386_PAGE_H */