Skip to contentSkip to conversation

Enable ASan in GHA

PATCH Superseded David Marchand 3 participants 18 messages Apr 15, 2022 – May 11, 2022 Arm/Amlogic Meson SoC support

lore
Share & tools
lore
#0 lore raw

cover Enable ASan in GHA

This series fixes two issues that were revealed while running unit tests with ASan in GHA.

There are still some outstanding issues for which bz have been created. The last patch enables ASan in GHA, skipping tests which have issues.

Signature · 18 lines
David Marchand

David Marchand (3):
  test/mem: disable ASan when accessing unallocated mem
  mem: fix ASan shadow for remapped memory segments
  ci: build some job with ASan

 .ci/linux-build.sh           |   8 ++
 .github/workflows/build.yml  |   3 +-
 app/test/meson.build         | 208 ++++++++++++++++++-----------------
 app/test/test_memory.c       |   5 +
 lib/eal/common/malloc_elem.h |  12 +-
 lib/eal/common/malloc_heap.c |  12 +-
 lib/eal/include/rte_common.h |  13 +++
 7 files changed, 152 insertions(+), 109 deletions(-)

-- 
2.23.0
#1 lore raw

PATCH 1/3 test/mem: disable ASan when accessing unallocated mem

As described in bugzilla, ASan reports accesses to all memory segment as invalid, since those parts have not been allocated.
Move __rte_no_asan to rte_common.h and disable ASan on a part of the test.

Bugzilla ID: 880

Fixes 6cc51b1293ce ("mem: instrument allocator for ASan")
---
 app/test/test_memory.c       |  5 +++++
 lib/eal/common/malloc_elem.h | 10 ++--------
 lib/eal/include/rte_common.h | 13 +++++++++++++
 3 files changed, 20 insertions(+), 8 deletions(-)
app/test/test_memory.c +5 -0
index 140ac3f3cf..440e5ef838 100644
@@ -25,6 +25,11 @@
  * - Try to read all memory; it should not segfault.
  */
 
+/*
+ * ASan complains about accessing unallocated memory.
+ * See: https://bugs.dpdk.org/show_bug.cgi?id=880
+ */
+__rte_no_asan
 static int
 check_mem(const struct rte_memseg_list *msl __rte_unused,
 		const struct rte_memseg *ms, void *arg __rte_unused)
lib/eal/common/malloc_elem.h +2 -8
index f2aa98821b..228f178418 100644
@@ -7,6 +7,8 @@
 
 #include <stdbool.h>
 
+#include <rte_common.h>
+
 #define MIN_DATA_SIZE (RTE_CACHE_LINE_SIZE)
 
 /* dummy definition of struct so we can use pointers to it in malloc_elem struct */
@@ -131,12 +133,6 @@ malloc_elem_cookies_ok(const struct malloc_elem *elem)
 #define ASAN_MEM_TO_SHADOW(mem) \
 	RTE_PTR_ADD(ASAN_MEM_SHIFT(mem), ASAN_SHADOW_OFFSET)
 
-#if defined(__clang__)
-#define __rte_no_asan __attribute__((no_sanitize("address", "hwaddress")))
-#else
-#define __rte_no_asan __attribute__((no_sanitize_address))
-#endif
-
 __rte_no_asan
 static inline void
 asan_set_shadow(void *addr, char val)
@@ -276,8 +272,6 @@ old_malloc_size(struct malloc_elem *elem)
 
 #else /* !RTE_MALLOC_ASAN */
 
-#define __rte_no_asan
-
 static inline void
 asan_set_freezone(void *ptr __rte_unused, size_t size __rte_unused) { }
 
lib/eal/include/rte_common.h +13 -0
index 67587025ab..d56a7570c0 100644
@@ -267,6 +267,19 @@ static void __attribute__((destructor(RTE_PRIO(prio)), used)) func(void)
  */
 #define __rte_cold __attribute__((cold))
 
+/**
+ * Disable AddressSanitizer on some code
+ */
+#ifdef RTE_MALLOC_ASAN
+#ifdef RTE_CC_CLANG
+#define __rte_no_asan __attribute__((no_sanitize("address", "hwaddress")))
+#else
+#define __rte_no_asan __attribute__((no_sanitize_address))
+#endif
+#else /* ! RTE_MALLOC_ASAN */
+#define __rte_no_asan
+#endif
+
 /*********** Macros for pointer arithmetic ********/
 
 /**
Signature · 1 line
2.23.0
Acked-by on v1 1/3
#2 lore raw

Re: [PATCH 1/3] test/mem: disable ASan when accessing unallocated mem

Quoted 10 lines · David Marchand

As described in bugzilla, ASan reports accesses to all memory segment as invalid, since those parts have not been allocated.
Move __rte_no_asan to rte_common.h and disable ASan on a part of the test.

Bugzilla ID: 880

Fixes 6cc51b1293ce ("mem: instrument allocator for ASan")

---

Signature · 2 lines
Thanks,
Anatoly
#3 lore raw

PATCH 2/3 mem: fix ASan shadow for remapped memory segments

In reply to David Marchand

When releasing some memory, the allocator can choose to return some pages to the OS. At the same time, this memory was poisoned in ASAn shadow. Doing the latter made it impossible to remap this same page later.
On the other hand, without this poison, the OS would pagefault in any case for this page.

Remove the poisoning for unmapped pages.

Bugzilla ID: 994

Fixes 6cc51b1293ce ("mem: instrument allocator for ASan")
---
 lib/eal/common/malloc_elem.h |  4 ++++
 lib/eal/common/malloc_heap.c | 12 +++++++++++-
 2 files changed, 15 insertions(+), 1 deletion(-)
lib/eal/common/malloc_elem.h +4 -0
index 228f178418..b859003722 100644
@@ -272,6 +272,10 @@ old_malloc_size(struct malloc_elem *elem)
 
 #else /* !RTE_MALLOC_ASAN */
 
+static inline void
+asan_set_zone(void *ptr __rte_unused, size_t len __rte_unused,
+	uint32_t val __rte_unused) { }
+
 static inline void
 asan_set_freezone(void *ptr __rte_unused, size_t size __rte_unused) { }
 
lib/eal/common/malloc_heap.c +11 -1
index 6c572b6f2c..5913d9f862 100644
@@ -860,6 +860,7 @@ malloc_heap_free(struct malloc_elem *elem)
 	size_t len, aligned_len, page_sz;
 	struct rte_memseg_list *msl;
 	unsigned int i, n_segs, before_space, after_space;
+	bool unmapped_pages = false;
 	int ret;
 	const struct internal_config *internal_conf =
 		eal_get_internal_configuration();
@@ -999,6 +1000,13 @@ malloc_heap_free(struct malloc_elem *elem)
 
 		/* don't care if any of this fails */
 		malloc_heap_free_pages(aligned_start, aligned_len);
+		/*
+		 * Clear any poisoning in ASan for the associated pages so that
+		 * next time EAL maps those pages, the allocator can access
+		 * them.
+		 */
+		asan_set_zone(aligned_start, aligned_len, 0x00);
+		unmapped_pages = true;
 
 		request_sync();
 	} else {
@@ -1032,7 +1040,9 @@ malloc_heap_free(struct malloc_elem *elem)
 
 	rte_mcfg_mem_write_unlock();
 free_unlock:
-	asan_set_freezone(asan_ptr, asan_data_len);
+	/* Poison memory range if belonging to some still mapped pages. */
+	if (!unmapped_pages)
+		asan_set_freezone(asan_ptr, asan_data_len);
 
 	rte_spinlock_unlock(&(heap->lock));
 	return ret;
Signature · 1 line
2.23.0
#4 lore raw

Re: [PATCH 2/3] mem: fix ASan shadow for remapped memory segments

Quoted 61 lines · David Marchand

When releasing some memory, the allocator can choose to return some pages to the OS. At the same time, this memory was poisoned in ASAn shadow. Doing the latter made it impossible to remap this same page later.
On the other hand, without this poison, the OS would pagefault in any case for this page.

Remove the poisoning for unmapped pages.

Bugzilla ID: 994

Fixes 6cc51b1293ce ("mem: instrument allocator for ASan")
---
  lib/eal/common/malloc_elem.h |  4 ++++
  lib/eal/common/malloc_heap.c | 12 +++++++++++-
  2 files changed, 15 insertions(+), 1 deletion(-)
lib/eal/common/malloc_elem.h +4 -0
index 228f178418..b859003722 100644
@@ -272,6 +272,10 @@ old_malloc_size(struct malloc_elem *elem)
 
  #else /* !RTE_MALLOC_ASAN */
 
+static inline void
+asan_set_zone(void *ptr __rte_unused, size_t len __rte_unused,
+	uint32_t val __rte_unused) { }
+
  static inline void
  asan_set_freezone(void *ptr __rte_unused, size_t size __rte_unused) { }
 
lib/eal/common/malloc_heap.c +8 -0
index 6c572b6f2c..5913d9f862 100644
@@ -860,6 +860,7 @@ malloc_heap_free(struct malloc_elem *elem)
  	size_t len, aligned_len, page_sz;
  	struct rte_memseg_list *msl;
  	unsigned int i, n_segs, before_space, after_space;
+	bool unmapped_pages = false;
  	int ret;
  	const struct internal_config *internal_conf =
  		eal_get_internal_configuration();
@@ -999,6 +1000,13 @@ malloc_heap_free(struct malloc_elem *elem)
 
  		/* don't care if any of this fails */
  		malloc_heap_free_pages(aligned_start, aligned_len);
+		/*
+		 * Clear any poisoning in ASan for the associated pages so that
+		 * next time EAL maps those pages, the allocator can access
+		 * them.
+		 */
+		asan_set_zone(aligned_start, aligned_len, 0x00);
+		unmapped_pages = true;
 
  		request_sync();
  	} else {

lib/eal/common/malloc_heap.c malloc_heap_free(struct malloc_elem *elem) · line 1048

Quoted 10 lines
lib/eal/common/malloc_heap.c +3 -1
@@ -1032,7 +1040,9 @@ malloc_heap_free(struct malloc_elem *elem)
 
  	rte_mcfg_mem_write_unlock();
  free_unlock:
-	asan_set_freezone(asan_ptr, asan_data_len);
+	/* Poison memory range if belonging to some still mapped pages. */
+	if (!unmapped_pages)
+		asan_set_freezone(asan_ptr, asan_data_len);
 
  	rte_spinlock_unlock(&(heap->lock));
  	return ret;

I suspect the patch should be a little more complicated than that. When we unmap pages, we don't necessarily unmap the entire malloc element, it could be that we have a freed allocation like so:

| malloc header | free space | unmapped space | free space | next malloc header |

So, i think the freezone should be set from asan_ptr till aligned_start, and then from (aligned_start + aligned_len) till (asan_ptr + asan_data_len). Does that make sense?

Signature · 2 lines
Thanks,
Anatoly
#5 lore raw

Re: [PATCH 2/3] mem: fix ASan shadow for remapped memory segments

Quoted 78 lines · Anatoly
Quoted 61 lines · David Marchand

When releasing some memory, the allocator can choose to return some pages to the OS. At the same time, this memory was poisoned in ASAn shadow. Doing the latter made it impossible to remap this same page later.
On the other hand, without this poison, the OS would pagefault in any case for this page.

Remove the poisoning for unmapped pages.

Bugzilla ID: 994

Fixes 6cc51b1293ce ("mem: instrument allocator for ASan")
---
  lib/eal/common/malloc_elem.h |  4 ++++
  lib/eal/common/malloc_heap.c | 12 +++++++++++-
  2 files changed, 15 insertions(+), 1 deletion(-)
lib/eal/common/malloc_elem.h +4 -0
index 228f178418..b859003722 100644
@@ -272,6 +272,10 @@ old_malloc_size(struct malloc_elem *elem)
 
  #else /* !RTE_MALLOC_ASAN */
 
+static inline void
+asan_set_zone(void *ptr __rte_unused, size_t len __rte_unused,
+     uint32_t val __rte_unused) { }
+
  static inline void
  asan_set_freezone(void *ptr __rte_unused, size_t size __rte_unused) { }
 
lib/eal/common/malloc_heap.c +8 -0
index 6c572b6f2c..5913d9f862 100644
@@ -860,6 +860,7 @@ malloc_heap_free(struct malloc_elem *elem)
      size_t len, aligned_len, page_sz;
      struct rte_memseg_list *msl;
      unsigned int i, n_segs, before_space, after_space;
+     bool unmapped_pages = false;
      int ret;
      const struct internal_config *internal_conf =
              eal_get_internal_configuration();
@@ -999,6 +1000,13 @@ malloc_heap_free(struct malloc_elem *elem)
 
              /* don't care if any of this fails */
              malloc_heap_free_pages(aligned_start, aligned_len);
+             /*
+              * Clear any poisoning in ASan for the associated pages so that
+              * next time EAL maps those pages, the allocator can access
+              * them.
+              */
+             asan_set_zone(aligned_start, aligned_len, 0x00);
+             unmapped_pages = true;
 
              request_sync();
      } else {

lib/eal/common/malloc_heap.c malloc_heap_free(struct malloc_elem *elem) · line 1048

Quoted 10 lines
lib/eal/common/malloc_heap.c +3 -1
@@ -1032,7 +1040,9 @@ malloc_heap_free(struct malloc_elem *elem)
 
      rte_mcfg_mem_write_unlock();
  free_unlock:
-     asan_set_freezone(asan_ptr, asan_data_len);
+     /* Poison memory range if belonging to some still mapped pages. */
+     if (!unmapped_pages)
+             asan_set_freezone(asan_ptr, asan_data_len);
 
      rte_spinlock_unlock(&(heap->lock));
      return ret;

I suspect the patch should be a little more complicated than that. When we unmap pages, we don't necessarily unmap the entire malloc element, it could be that we have a freed allocation like so:

Quoted 6 lines
| malloc header | free space | unmapped space | free space | next malloc
header |

So, i think the freezone should be set from asan_ptr till aligned_start, and then from (aligned_start + aligned_len) till (asan_ptr + asan_data_len). Does that make sense?

(btw, I get a bounce for Zhihong mail address, is he not working at Intel anymore?)

To be honest, I don't understand if we can get to this situation :-) (especially the free space after the unmapped region). But I guess you mean something like (on top of current patch):

@@ -1040,9 +1040,25 @@ malloc_heap_free(struct malloc_elem *elem)

        rte_mcfg_mem_write_unlock();
 free_unlock:
-       /* Poison memory range if belonging to some still mapped pages. */
-       if (!unmapped_pages)
+       if (!unmapped_pages) {
                asan_set_freezone(asan_ptr, asan_data_len);

+ } else {
+ /*
+ * We may be in a situation where we unmapped pages like this:
+ * malloc header | free space | unmapped space | free
space | malloc header
+ */

+               void *free1_start = asan_ptr;
+               void *free1_end = aligned_start;
+               void *free2_start = RTE_PTR_ADD(aligned_start, aligned_len);
+               void *free2_end = RTE_PTR_ADD(asan_ptr, asan_data_len);

+
+ if (free1_start < free1_end)
+ asan_set_freezone(free1_start,
+ RTE_PTR_DIFF(free1_end, free1_start));
+ if (free2_start < free2_end)
+ asan_set_freezone(free2_start,
+ RTE_PTR_DIFF(free2_end, free2_start));
+ }

        rte_spinlock_unlock(&(heap->lock));
        return ret;
Signature · 1 line
David Marchand
#6 lore raw

Re: [PATCH 2/3] mem: fix ASan shadow for remapped memory segments

Quoted 114 lines · David Marchand
Quoted 78 lines · Anatoly
Quoted 61 lines · David Marchand

When releasing some memory, the allocator can choose to return some pages to the OS. At the same time, this memory was poisoned in ASAn shadow. Doing the latter made it impossible to remap this same page later.
On the other hand, without this poison, the OS would pagefault in any case for this page.

Remove the poisoning for unmapped pages.

Bugzilla ID: 994

Fixes 6cc51b1293ce ("mem: instrument allocator for ASan")
---
  lib/eal/common/malloc_elem.h |  4 ++++
  lib/eal/common/malloc_heap.c | 12 +++++++++++-
  2 files changed, 15 insertions(+), 1 deletion(-)
lib/eal/common/malloc_elem.h +4 -0
index 228f178418..b859003722 100644
@@ -272,6 +272,10 @@ old_malloc_size(struct malloc_elem *elem)
 
  #else /* !RTE_MALLOC_ASAN */
 
+static inline void
+asan_set_zone(void *ptr __rte_unused, size_t len __rte_unused,
+     uint32_t val __rte_unused) { }
+
  static inline void
  asan_set_freezone(void *ptr __rte_unused, size_t size __rte_unused) { }
 
lib/eal/common/malloc_heap.c +8 -0
index 6c572b6f2c..5913d9f862 100644
@@ -860,6 +860,7 @@ malloc_heap_free(struct malloc_elem *elem)
      size_t len, aligned_len, page_sz;
      struct rte_memseg_list *msl;
      unsigned int i, n_segs, before_space, after_space;
+     bool unmapped_pages = false;
      int ret;
      const struct internal_config *internal_conf =
              eal_get_internal_configuration();
@@ -999,6 +1000,13 @@ malloc_heap_free(struct malloc_elem *elem)
 
              /* don't care if any of this fails */
              malloc_heap_free_pages(aligned_start, aligned_len);
+             /*
+              * Clear any poisoning in ASan for the associated pages so that
+              * next time EAL maps those pages, the allocator can access
+              * them.
+              */
+             asan_set_zone(aligned_start, aligned_len, 0x00);
+             unmapped_pages = true;
 
              request_sync();
      } else {

lib/eal/common/malloc_heap.c malloc_heap_free(struct malloc_elem *elem) · line 1048

Quoted 10 lines
lib/eal/common/malloc_heap.c +3 -1
@@ -1032,7 +1040,9 @@ malloc_heap_free(struct malloc_elem *elem)
 
      rte_mcfg_mem_write_unlock();
  free_unlock:
-     asan_set_freezone(asan_ptr, asan_data_len);
+     /* Poison memory range if belonging to some still mapped pages. */
+     if (!unmapped_pages)
+             asan_set_freezone(asan_ptr, asan_data_len);
 
      rte_spinlock_unlock(&(heap->lock));
      return ret;

I suspect the patch should be a little more complicated than that. When we unmap pages, we don't necessarily unmap the entire malloc element, it could be that we have a freed allocation like so:

Quoted 6 lines
| malloc header | free space | unmapped space | free space | next malloc
header |

So, i think the freezone should be set from asan_ptr till aligned_start, and then from (aligned_start + aligned_len) till (asan_ptr + asan_data_len). Does that make sense?

(btw, I get a bounce for Zhihong mail address, is he not working at Intel anymore?)

To be honest, I don't understand if we can get to this situation :-) (especially the free space after the unmapped region). But I guess you mean something like (on top of current patch):

(quoted diff) +5 -2
@@ -1040,9 +1040,25 @@ malloc_heap_free(struct malloc_elem *elem)
 
        rte_mcfg_mem_write_unlock();
 free_unlock:
-       /* Poison memory range if belonging to some still mapped pages. */
-       if (!unmapped_pages)
+       if (!unmapped_pages) {
                asan_set_freezone(asan_ptr, asan_data_len);
+       } else {
+               /*
+                * We may be in a situation where we unmapped pages like this:
+                * malloc header | free space | unmapped space | free

space | malloc header

(quoted diff) +6 -0
+                */
+               void *free1_start = asan_ptr;
+               void *free1_end = aligned_start;
+               void *free2_start = RTE_PTR_ADD(aligned_start, aligned_len);
+               void *free2_end = RTE_PTR_ADD(asan_ptr, asan_data_len);
+

Review

Quoted 10 lines
(quoted diff) +7 -0
+               if (free1_start < free1_end)
+                       asan_set_freezone(free1_start,
+                               RTE_PTR_DIFF(free1_end, free1_start));
+               if (free2_start < free2_end)
+                       asan_set_freezone(free2_start,
+                               RTE_PTR_DIFF(free2_end, free2_start));
+       }
 
        rte_spinlock_unlock(&(heap->lock));
        return ret;

But I get a splat for func_reentrancy_autotest:

=================================================================
==4098809==ERROR: AddressSanitizer: heap-use-after-free on address 0x7fa00fa00030 at pc 0x0000035779fe bp 0x7ffe01ed0cf0 sp 0x7ffe01ed0ce8
READ of size 1 at 0x7fa00fa00030 thread T0

    #0 0x35779fd in malloc_elem_join_adjacent_free
../lib/eal/common/malloc_elem.c:539
    #1 0x3577bc5 in malloc_elem_free ../lib/eal/common/malloc_elem.c:586
    #2 0x357bd25 in malloc_heap_free ../lib/eal/common/malloc_heap.c:886
    #3 0x357e032 in mem_free ../lib/eal/common/rte_malloc.c:37
    #4 0x357e032 in rte_free ../lib/eal/common/rte_malloc.c:44
    #5 0x336a547 in rte_lpm_free ../lib/lpm/rte_lpm.c:281
    #6 0x2cb3488 in lpm_create_free ../app/test/test_func_reentrancy.c:395
    #7 0x2cb47ed in launch_test ../app/test/test_func_reentrancy.c:455
    #8 0x2cb47ed in test_func_reentrancy ../app/test/test_func_reentrancy.c:502
    #9 0x2b0dd75 in cmd_autotest_parsed ../app/test/commands.c:68
    #10 0x34e5dce in cmdline_parse ../lib/cmdline/cmdline_parse.c:287
    #11 0x34e31af in cmdline_valid_buffer ../lib/cmdline/cmdline.c:24
    #12 0x34eba9f in rdline_char_in ../lib/cmdline/cmdline_rdline.c:444
    #13 0x34e329b in cmdline_in ../lib/cmdline/cmdline.c:146
    #14 0x34e329b in cmdline_in ../lib/cmdline/cmdline.c:135
    #15 0x40d8a5 in main ../app/test/test.c:217
    #16 0x7fa41009b55f in __libc_start_call_main
../sysdeps/nptl/libc_start_call_main.h:58
    #17 0x7fa41009b60b in __libc_start_main_impl ../csu/libc-start.c:409
    #18 0x2b0dc64 in _start

(/home/dmarchan/dpdk/build-gcc-asan/app/test/dpdk-test+0x2b0dc64)

Address 0x7fa00fa00030 is a wild pointer.
SUMMARY: AddressSanitizer: heap-use-after-free ../lib/eal/common/malloc_elem.c:539 in malloc_elem_join_adjacent_free Shadow bytes around the buggy address:
0x0ff481f37fb0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 0x0ff481f37fc0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 0x0ff481f37fd0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 0x0ff481f37fe0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 0x0ff481f37ff0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
=>0x0ff481f38000: fd fd fd fd fd fd[fd]fd fd fd fd fd fd fd fd fd
0x0ff481f38010: fd fd fd fd fd fd fd fd fd fd fd fd fd fd fd fd 0x0ff481f38020: fd fd fd fd fd fd fd fd fd fd fd fd fd fd fd fd 0x0ff481f38030: fd fd fd fd fd fd fd fd fd fd fd fd fd fd fd fd 0x0ff481f38040: fd fd fd fd fd fd fd fd fd fd fd fd fd fd fd fd 0x0ff481f38050: fd fd fd fd fd fd fd fd fd fd fd fd fd fd fd fd
Shadow byte legend (one shadow byte represents 8 application bytes):

Signature · 1 line
David Marchand
#7 lore raw

Re: [PATCH 2/3] mem: fix ASan shadow for remapped memory segments

In reply to David Marchand
Quoted 114 lines · David Marchand
Quoted 78 lines · Anatoly
Quoted 61 lines · David Marchand

When releasing some memory, the allocator can choose to return some pages to the OS. At the same time, this memory was poisoned in ASAn shadow. Doing the latter made it impossible to remap this same page later.
On the other hand, without this poison, the OS would pagefault in any case for this page.

Remove the poisoning for unmapped pages.

Bugzilla ID: 994

Fixes 6cc51b1293ce ("mem: instrument allocator for ASan")
---
   lib/eal/common/malloc_elem.h |  4 ++++
   lib/eal/common/malloc_heap.c | 12 +++++++++++-
   2 files changed, 15 insertions(+), 1 deletion(-)
lib/eal/common/malloc_elem.h +4 -0
index 228f178418..b859003722 100644
@@ -272,6 +272,10 @@ old_malloc_size(struct malloc_elem *elem)
 
   #else /* !RTE_MALLOC_ASAN */
 
+static inline void
+asan_set_zone(void *ptr __rte_unused, size_t len __rte_unused,
+     uint32_t val __rte_unused) { }
+
   static inline void
   asan_set_freezone(void *ptr __rte_unused, size_t size __rte_unused) { }
 
lib/eal/common/malloc_heap.c +8 -0
index 6c572b6f2c..5913d9f862 100644
@@ -860,6 +860,7 @@ malloc_heap_free(struct malloc_elem *elem)
       size_t len, aligned_len, page_sz;
       struct rte_memseg_list *msl;
       unsigned int i, n_segs, before_space, after_space;
+     bool unmapped_pages = false;
       int ret;
       const struct internal_config *internal_conf =
               eal_get_internal_configuration();
@@ -999,6 +1000,13 @@ malloc_heap_free(struct malloc_elem *elem)
 
               /* don't care if any of this fails */
               malloc_heap_free_pages(aligned_start, aligned_len);
+             /*
+              * Clear any poisoning in ASan for the associated pages so that
+              * next time EAL maps those pages, the allocator can access
+              * them.
+              */
+             asan_set_zone(aligned_start, aligned_len, 0x00);
+             unmapped_pages = true;
 
               request_sync();
       } else {

lib/eal/common/malloc_heap.c malloc_heap_free(struct malloc_elem *elem) · line 1048

Quoted 10 lines
lib/eal/common/malloc_heap.c +3 -1
@@ -1032,7 +1040,9 @@ malloc_heap_free(struct malloc_elem *elem)
 
       rte_mcfg_mem_write_unlock();
   free_unlock:
-     asan_set_freezone(asan_ptr, asan_data_len);
+     /* Poison memory range if belonging to some still mapped pages. */
+     if (!unmapped_pages)
+             asan_set_freezone(asan_ptr, asan_data_len);
 
       rte_spinlock_unlock(&(heap->lock));
       return ret;

I suspect the patch should be a little more complicated than that. When we unmap pages, we don't necessarily unmap the entire malloc element, it could be that we have a freed allocation like so:

Quoted 6 lines
| malloc header | free space | unmapped space | free space | next malloc
header |

So, i think the freezone should be set from asan_ptr till aligned_start, and then from (aligned_start + aligned_len) till (asan_ptr + asan_data_len). Does that make sense?

(btw, I get a bounce for Zhihong mail address, is he not working at Intel anymore?)

To be honest, I don't understand if we can get to this situation :-) (especially the free space after the unmapped region). But I guess you mean something like (on top of current patch):

(quoted diff) +5 -2
@@ -1040,9 +1040,25 @@ malloc_heap_free(struct malloc_elem *elem)
 
         rte_mcfg_mem_write_unlock();
  free_unlock:
-       /* Poison memory range if belonging to some still mapped pages. */
-       if (!unmapped_pages)
+       if (!unmapped_pages) {
                 asan_set_freezone(asan_ptr, asan_data_len);
+       } else {
+               /*
+                * We may be in a situation where we unmapped pages like this:
+                * malloc header | free space | unmapped space | free

space | malloc header

(quoted diff) +6 -0
+                */
+               void *free1_start = asan_ptr;
+               void *free1_end = aligned_start;
+               void *free2_start = RTE_PTR_ADD(aligned_start, aligned_len);
+               void *free2_end = RTE_PTR_ADD(asan_ptr, asan_data_len);
+

Review

Quoted 10 lines
(quoted diff) +7 -0
+               if (free1_start < free1_end)
+                       asan_set_freezone(free1_start,
+                               RTE_PTR_DIFF(free1_end, free1_start));
+               if (free2_start < free2_end)
+                       asan_set_freezone(free2_start,
+                               RTE_PTR_DIFF(free2_end, free2_start));
+       }
 
         rte_spinlock_unlock(&(heap->lock));
         return ret;

Something like that, yes. I will have to think through this a bit more, especially in light of your func_reentrancy splat :)

Signature · 2 lines
Thanks,
Anatoly
#8 lore raw

Re: [PATCH 2/3] mem: fix ASan shadow for remapped memory segments

Quoted 128 lines · Anatoly
Quoted 127 lines · David Marchand
Quoted 81 lines · Anatoly
Quoted 72 lines · David Marchand

When releasing some memory, the allocator can choose to return some pages to the OS. At the same time, this memory was poisoned in ASAn shadow. Doing the latter made it impossible to remap this same page later.
On the other hand, without this poison, the OS would pagefault in any case for this page.

Remove the poisoning for unmapped pages.

Bugzilla ID: 994

Fixes 6cc51b1293ce ("mem: instrument allocator for ASan")

---
   lib/eal/common/malloc_elem.h |  4 ++++
   lib/eal/common/malloc_heap.c | 12 +++++++++++-

   2 files changed, 15 insertions(+), 1 deletion(-)
lib/eal/common/malloc_elem.h +4 -0
index 228f178418..b859003722 100644
@@ -272,6 +272,10 @@ old_malloc_size(struct malloc_elem *elem)
 
    #else /* !RTE_MALLOC_ASAN */
 
+static inline void
+asan_set_zone(void *ptr __rte_unused, size_t len __rte_unused,
+     uint32_t val __rte_unused) { }
+

   static inline void
   asan_set_freezone(void *ptr __rte_unused, size_t size __rte_unused) { }

lib/eal/common/malloc_heap.c +1 -0
index 6c572b6f2c..5913d9f862 100644
@@ -860,6 +860,7 @@ malloc_heap_free(struct malloc_elem *elem)
        size_t len, aligned_len, page_sz;
        struct rte_memseg_list *msl;
        unsigned int i, n_segs, before_space, after_space;
+     bool unmapped_pages = false;
        int ret;

       const struct internal_config *internal_conf = eal_get_internal_configuration();

lib/eal/common/malloc_heap.c +2 -0
@@ -999,6 +1000,13 @@ malloc_heap_free(struct malloc_elem *elem)
 
                /* don't care if any of this fails */
                malloc_heap_free_pages(aligned_start, aligned_len);
+             /*
+              * Clear any poisoning in ASan for the associated 

pages so that
+              * next time EAL maps those pages, the allocator can access
+              * them.
+              */
+             asan_set_zone(aligned_start, aligned_len, 0x00); +             unmapped_pages = true;

               request_sync();
       } else {

lib/eal/common/malloc_heap.c +0 -0
@@ -1032,7 +1040,9 @@ malloc_heap_free(struct malloc_elem *elem)
 
        rte_mcfg_mem_write_unlock();

   free_unlock:
-     asan_set_freezone(asan_ptr, asan_data_len);
+     /* Poison memory range if belonging to some still mapped pages. */
+     if (!unmapped_pages)
+             asan_set_freezone(asan_ptr, asan_data_len);

Quoted 2 lines

       rte_spinlock_unlock(&(heap->lock));
       return ret;

I suspect the patch should be a little more complicated than that. When we unmap pages, we don't necessarily unmap the entire malloc element, it could be that we have a freed allocation like so:

Quoted 6 lines
| malloc header | free space | unmapped space | free space | next malloc
header |

So, i think the freezone should be set from asan_ptr till aligned_start, and then from (aligned_start + aligned_len) till (asan_ptr + asan_data_len). Does that make sense?

(btw, I get a bounce for Zhihong mail address, is he not working at Intel anymore?)

To be honest, I don't understand if we can get to this situation :-) (especially the free space after the unmapped region). But I guess you mean something like (on top of current patch):

lib/eal/common/malloc_heap.c +0 -0
@@ -1040,9 +1040,25 @@ malloc_heap_free(struct malloc_elem *elem)
 
          rte_mcfg_mem_write_unlock();

  free_unlock:
-       /* Poison memory range if belonging to some still mapped pages. */
-       if (!unmapped_pages)
+       if (!unmapped_pages) {
                 asan_set_freezone(asan_ptr, asan_data_len); +       } else {
+               /*
+                * We may be in a situation where we unmapped pages like this:
+                * malloc header | free space | unmapped space | free space | malloc header
+                */
+               void *free1_start = asan_ptr;
+               void *free1_end = aligned_start;
+               void *free2_start = RTE_PTR_ADD(aligned_start, aligned_len);
+               void *free2_end = RTE_PTR_ADD(asan_ptr, asan_data_len); +
+               if (free1_start < free1_end)
+                       asan_set_freezone(free1_start,
+                               RTE_PTR_DIFF(free1_end, free1_start)); +               if (free2_start < free2_end)
+                       asan_set_freezone(free2_start,
+                               RTE_PTR_DIFF(free2_end, free2_start)); +       }

Quoted 6 lines
Quoted 3 lines

         rte_spinlock_unlock(&(heap->lock));
         return ret;

Something like that, yes. I will have to think through this a bit more, especially in light of your func_reentrancy splat :)

So, the reason splat in func_reentrancy test happens is as follows: the above patch is sorta correct (i have a different one but does the same thing), but incomplete. What happens then is when we add new memory, we are integrating it into our existing malloc heap, which triggers malloc_elem_join_adjacent_free() which will trigger a write into old header space being merged, which may be marked as "freed". So, again we are hit with our internal allocator messing with ASan.

To properly fix this is to answer the following question: what is the goal of having ASan support in DPDK? Is it there to catch bugs *in the allocator*, or can we just trust that our allocator code is correct, and only concern ourselves with user-allocated areas of the code? Because it seems like the best way to address this issue would be to just avoid triggering ASan checks for certain allocator-internal actions: this way, we don't need to care what allocator itself does, just what user code does. As in, IIRC there was a compiler attribute that disables ASan checks for a specific function: perhaps we could just wrap certain access in that and be done with it?

What do you think?

Signature · 2 lines
Thanks,
Anatoly
#9 lore raw

Re: [PATCH 2/3] mem: fix ASan shadow for remapped memory segments

Quoted 38 lines · Anatoly
Quoted 30 lines
Quoted 30 lines
lib/eal/common/malloc_heap.c +0 -1
@@ -1040,9 +1040,25 @@ malloc_heap_free(struct malloc_elem *elem)
 
         rte_mcfg_mem_write_unlock();
  free_unlock:
-       /* Poison memory range if belonging to some still mapped

pages. */

(quoted diff) +4 -1
-       if (!unmapped_pages)
+       if (!unmapped_pages) {
                 asan_set_freezone(asan_ptr, asan_data_len);
+       } else {
+               /*
+                * We may be in a situation where we unmapped pages

like this:
+ * malloc header | free space | unmapped space | free
space | malloc header

(quoted diff) +4 -0
+                */
+               void *free1_start = asan_ptr;
+               void *free1_end = aligned_start;
+               void *free2_start = RTE_PTR_ADD(aligned_start,

aligned_len);

(quoted diff) +9 -0
+               void *free2_end = RTE_PTR_ADD(asan_ptr, asan_data_len);
+
+               if (free1_start < free1_end)
+                       asan_set_freezone(free1_start,
+                               RTE_PTR_DIFF(free1_end, free1_start));
+               if (free2_start < free2_end)
+                       asan_set_freezone(free2_start,
+                               RTE_PTR_DIFF(free2_end, free2_start));
+       }
Quoted 7 lines
Quoted 3 lines
         rte_spinlock_unlock(&(heap->lock));
         return ret;

Something like that, yes. I will have to think through this a bit more, especially in light of your func_reentrancy splat :)

Quoted 7 lines

So, the reason splat in func_reentrancy test happens is as follows: the above patch is sorta correct (i have a different one but does the same thing), but incomplete. What happens then is when we add new memory, we are integrating it into our existing malloc heap, which triggers malloc_elem_join_adjacent_free() which will trigger a write into old header space being merged, which may be marked as "freed". So, again we are hit with our internal allocator messing with ASan.

I ended up with the same conclusion.
Thanks for confirming.

Quoted 5 lines

To properly fix this is to answer the following question: what is the goal of having ASan support in DPDK? Is it there to catch bugs *in the allocator*, or can we just trust that our allocator code is correct, and only concern ourselves with user-allocated areas of the code? Because it

The best would be to handle both.
I don't think clang disables ASan for the instrumentations on malloc.

Quoted 8 lines

seems like the best way to address this issue would be to just avoid triggering ASan checks for certain allocator-internal actions: this way, we don't need to care what allocator itself does, just what user code does. As in, IIRC there was a compiler attribute that disables ASan checks for a specific function: perhaps we could just wrap certain access in that and be done with it?

What do you think?

It is tempting because it is the easiest way to avoid the issue. Though, by waiving those checks in the allocator, does it leave the ASan shadow in a consistent state?

Signature · 1 line
David Marchand
#10 lore raw

Re: [PATCH 2/3] mem: fix ASan shadow for remapped memory segments

Quoted 53 lines · David Marchand
Quoted 38 lines · Anatoly
Quoted 30 lines
Quoted 30 lines
lib/eal/common/malloc_heap.c +0 -1
@@ -1040,9 +1040,25 @@ malloc_heap_free(struct malloc_elem *elem)
 
          rte_mcfg_mem_write_unlock();
   free_unlock:
-       /* Poison memory range if belonging to some still mapped

pages. */

(quoted diff) +4 -1
-       if (!unmapped_pages)
+       if (!unmapped_pages) {
                  asan_set_freezone(asan_ptr, asan_data_len);
+       } else {
+               /*
+                * We may be in a situation where we unmapped pages

like this:
+ * malloc header | free space | unmapped space | free
space | malloc header

(quoted diff) +4 -0
+                */
+               void *free1_start = asan_ptr;
+               void *free1_end = aligned_start;
+               void *free2_start = RTE_PTR_ADD(aligned_start,

aligned_len);

(quoted diff) +9 -0
+               void *free2_end = RTE_PTR_ADD(asan_ptr, asan_data_len);
+
+               if (free1_start < free1_end)
+                       asan_set_freezone(free1_start,
+                               RTE_PTR_DIFF(free1_end, free1_start));
+               if (free2_start < free2_end)
+                       asan_set_freezone(free2_start,
+                               RTE_PTR_DIFF(free2_end, free2_start));
+       }
Quoted 7 lines
Quoted 3 lines
          rte_spinlock_unlock(&(heap->lock));
          return ret;

Something like that, yes. I will have to think through this a bit more, especially in light of your func_reentrancy splat :)

Quoted 7 lines

So, the reason splat in func_reentrancy test happens is as follows: the above patch is sorta correct (i have a different one but does the same thing), but incomplete. What happens then is when we add new memory, we are integrating it into our existing malloc heap, which triggers malloc_elem_join_adjacent_free() which will trigger a write into old header space being merged, which may be marked as "freed". So, again we are hit with our internal allocator messing with ASan.

I ended up with the same conclusion.
Thanks for confirming.

Quoted 1 line
Quoted 7 lines
Quoted 4 lines

To properly fix this is to answer the following question: what is the goal of having ASan support in DPDK? Is it there to catch bugs *in the allocator*, or can we just trust that our allocator code is correct, and only concern ourselves with user-allocated areas of the code? Because it

The best would be to handle both.
I don't think clang disables ASan for the instrumentations on malloc.

I've actually prototyped these changes a bit. We use memset in a few places, and that one can't be disabled as far as i can tell (not without blacklisting memset for entire DPDK).

Quoted 9 lines
Quoted 9 lines

seems like the best way to address this issue would be to just avoid triggering ASan checks for certain allocator-internal actions: this way, we don't need to care what allocator itself does, just what user code does. As in, IIRC there was a compiler attribute that disables ASan checks for a specific function: perhaps we could just wrap certain access in that and be done with it?

Quoted 6 lines
Quoted 1 line

What do you think?

It is tempting because it is the easiest way to avoid the issue. Though, by waiving those checks in the allocator, does it leave the ASan shadow in a consistent state?

The "consistent state" is kinda difficult to achieve because there is no "default" state for memory - sometimes it comes as available (0x00), sometimes it is marked as already freed (0xFF). So, coming into a malloc function, we don't know whether the memory we're about to mess with is 0x00 or 0xFF.

What we could do is mark every malloc header with 0xFF regardless of its status, and leave the rest to "regular" zoning. This would be strange from ASan's point of view (because we're marking memory as "freed" when it wasn't ever allocated), but at least this would be consistent :D

Signature · 2 lines
Thanks,
Anatoly
#11 lore raw

Re: [PATCH 2/3] mem: fix ASan shadow for remapped memory segments

Quoted 76 lines · Anatoly
Quoted 54 lines · David Marchand
Quoted 39 lines · Anatoly
Quoted 31 lines
Quoted 31 lines
(quoted diff) +0 -0
@@ -1040,9 +1040,25 @@ malloc_heap_free(struct malloc_elem *elem)
 
           rte_mcfg_mem_write_unlock();

   free_unlock:
-       /* Poison memory range if belonging to some still mapped pages. */
-       if (!unmapped_pages)
+       if (!unmapped_pages) {
                  asan_set_freezone(asan_ptr, asan_data_len); +       } else {
+               /*
+                * We may be in a situation where we unmapped pages like this:
+                * malloc header | free space | unmapped space | free space | malloc header
+                */
+               void *free1_start = asan_ptr;
+               void *free1_end = aligned_start;
+               void *free2_start = RTE_PTR_ADD(aligned_start, aligned_len);
+               void *free2_end = RTE_PTR_ADD(asan_ptr, asan_data_len);
+
+               if (free1_start < free1_end)
+                       asan_set_freezone(free1_start,
+                               RTE_PTR_DIFF(free1_end, free1_start)); +               if (free2_start < free2_end)
+                       asan_set_freezone(free2_start,
+                               RTE_PTR_DIFF(free2_end, free2_start)); +       }

Quoted 7 lines
Quoted 3 lines

          rte_spinlock_unlock(&(heap->lock));
          return ret;

Something like that, yes. I will have to think through this a bit more, especially in light of your func_reentrancy splat :)

Quoted 7 lines

So, the reason splat in func_reentrancy test happens is as follows: the above patch is sorta correct (i have a different one but does the same thing), but incomplete. What happens then is when we add new memory, we are integrating it into our existing malloc heap, which triggers malloc_elem_join_adjacent_free() which will trigger a write into old header space being merged, which may be marked as "freed". So, again we are hit with our internal allocator messing with ASan.

I ended up with the same conclusion.
Thanks for confirming.

Quoted 1 line
Quoted 7 lines
Quoted 4 lines

To properly fix this is to answer the following question: what is the goal of having ASan support in DPDK? Is it there to catch bugs *in the allocator*, or can we just trust that our allocator code is correct, and only concern ourselves with user-allocated areas of the code? Because it

The best would be to handle both.
I don't think clang disables ASan for the instrumentations on malloc.

I've actually prototyped these changes a bit. We use memset in a few places, and that one can't be disabled as far as i can tell (not without blacklisting memset for entire DPDK).

Quoted 11 lines
Quoted 10 lines

seems like the best way to address this issue would be to just avoid triggering ASan checks for certain allocator-internal actions: this way, we don't need to care what allocator itself does, just what user code does. As in, IIRC there was a compiler attribute that disables ASan checks for a specific function: perhaps we could just wrap certain access in that and be done with it?

What do you think?

Quoted 9 lines
Quoted 4 lines

It is tempting because it is the easiest way to avoid the issue. Though, by waiving those checks in the allocator, does it leave the ASan shadow in a consistent state?

The "consistent state" is kinda difficult to achieve because there is no "default" state for memory - sometimes it comes as available (0x00), sometimes it is marked as already freed (0xFF). So, coming into a malloc function, we don't know whether the memory we're about to mess with is 0x00 or 0xFF.

What we could do is mark every malloc header with 0xFF regardless of its status, and leave the rest to "regular" zoning. This would be strange from ASan's point of view (because we're marking memory as "freed" when it wasn't ever allocated), but at least this would be consistent :D

I've been prototyping a solution for this, but I keep bumping into our dual usage of ASan: ASan doesn't differentiate between allocator-internal accesses, and user code accesses. Therefore, we can't either, so either we start marking areas as "accessible" even when they shouldn't be (such as unallocated areas that correspond to malloc headers), or we only use ASan to mark user-available areas and forego its usage inside the allocator entirely.

Right now, the best I can think of is the combination of approaches discussed earlier: that is, we mark all malloc element header areas as "available" unconditionally (thereby sacrificing part of the protection ASan provides us - because we can't prevent ASan from complaining about accesses from inside the allocator without losing our ability to detect cases where user accidentally accesses a malloc element), and we also mark unmapped memory as "available" (because writing to it will trigger a fault anyway).

I haven't yet figured out the cleanest solution (we miss asan zoning for headers somewhere), but at least i got func reentrancy test to pass :D

Signature · 2 lines
Thanks,
Anatoly
#12 lore raw

PATCH 3/3 ci: build some job with ASan

In reply to David Marchand

Enable ASan, this can greatly help identify leaks and buffer overflows. Running all unit tests is not possible at the moment: skip unit tests who have known issues with ASan.

---
 .ci/linux-build.sh          |   8 ++
 .github/workflows/build.yml |   3 +-
 app/test/meson.build        | 208 +++++++++++++++++++-----------------
 3 files changed, 118 insertions(+), 101 deletions(-)
.ci/linux-build.sh +8 -0
index 774a1441bf..93706c0131 100755
@@ -95,6 +95,14 @@ if [ "$MINI" = "true" ]; then
     OPTS="$OPTS -Denable_drivers=net/null"
     OPTS="$OPTS -Ddisable_libs=*"
 fi
+
+if [ "$ASAN" = "true" ]; then
+    OPTS="$OPTS -Db_sanitize=address"
+    if [ "${CC%%clang}" != "$CC" ]; then
+        OPTS="$OPTS -Db_lundef=false"
+    fi
+fi
+
 meson build --werror $OPTS
 ninja -C build
 
.github/workflows/build.yml +2 -1
index 22daaabb91..45871e76ed 100644
@@ -16,6 +16,7 @@ jobs:
     env:
       AARCH64: ${{ matrix.config.cross == 'aarch64' }}
       ABI_CHECKS: ${{ contains(matrix.config.checks, 'abi') }}
+      ASAN: ${{ contains(matrix.config.checks, 'asan') }}
       BUILD_32BIT: ${{ matrix.config.cross == 'i386' }}
       BUILD_DOCS: ${{ contains(matrix.config.checks, 'doc') }}
       CC: ccache ${{ matrix.config.compiler }}
@@ -47,7 +48,7 @@ jobs:
           - os: ubuntu-18.04
             compiler: clang
             library: shared
-            checks: doc+tests
+            checks: asan+doc+tests
           - os: ubuntu-18.04
             compiler: gcc
             library: static
app/test/meson.build +108 -100

265 lines not on this page · raw message

Signature · 1 line
2.23.0
#13 lore raw

v2 cover Enable ASan in GHA

In reply to David Marchand

Now that rte_malloc instrumentations are fixed, we can enable ASan in GHA.
There are still some unit tests (relying on multiprocess) that can't reliably run with ASan enabled. Those unit tests are skipped.

Signature · 20 lines
David Marchand

Changes since v1:
- dropped patch 2 in favor of Anatoly fix,
- rebased the last patch after other unit tests fixes have been merged,

David Marchand (2):
  test/mem: disable ASan when accessing unallocated mem
  ci: build some job with ASan

 .ci/linux-build.sh           |   8 ++
 .github/workflows/build.yml  |   3 +-
 app/test/meson.build         | 208 ++++++++++++++++++-----------------
 app/test/test_memory.c       |   5 +
 lib/eal/common/malloc_elem.h |  10 +-
 lib/eal/include/rte_common.h |  13 +++
 6 files changed, 138 insertions(+), 109 deletions(-)

-- 
2.23.0
#14 lore raw

PATCH v2 1/2 test/mem: disable ASan when accessing unallocated mem

As described in bugzilla, ASan reports accesses to all memory segment as invalid, since those parts have not been allocated.
Move __rte_no_asan to rte_common.h and disable ASan on a part of the test.

Bugzilla ID: 880

Fixes 6cc51b1293ce ("mem: instrument allocator for ASan")
---
 app/test/test_memory.c       |  5 +++++
 lib/eal/common/malloc_elem.h | 10 ++--------
 lib/eal/include/rte_common.h | 13 +++++++++++++
 3 files changed, 20 insertions(+), 8 deletions(-)
app/test/test_memory.c +5 -0
index 140ac3f3cf..440e5ef838 100644
@@ -25,6 +25,11 @@
  * - Try to read all memory; it should not segfault.
  */
 
+/*
+ * ASan complains about accessing unallocated memory.
+ * See: https://bugs.dpdk.org/show_bug.cgi?id=880
+ */
+__rte_no_asan
 static int
 check_mem(const struct rte_memseg_list *msl __rte_unused,
 		const struct rte_memseg *ms, void *arg __rte_unused)
lib/eal/common/malloc_elem.h +2 -8
index c5f65895e1..952ce7343b 100644
@@ -7,6 +7,8 @@
 
 #include <stdbool.h>
 
+#include <rte_common.h>
+
 #define MIN_DATA_SIZE (RTE_CACHE_LINE_SIZE)
 
 /* dummy definition of struct so we can use pointers to it in malloc_elem struct */
@@ -131,12 +133,6 @@ malloc_elem_cookies_ok(const struct malloc_elem *elem)
 #define ASAN_MEM_TO_SHADOW(mem) \
 	RTE_PTR_ADD(ASAN_MEM_SHIFT(mem), ASAN_SHADOW_OFFSET)
 
-#if defined(__clang__)
-#define __rte_no_asan __attribute__((no_sanitize("address", "hwaddress")))
-#else
-#define __rte_no_asan __attribute__((no_sanitize_address))
-#endif
-
 __rte_no_asan
 static inline void
 asan_set_shadow(void *addr, char val)
@@ -276,8 +272,6 @@ old_malloc_size(struct malloc_elem *elem)
 
 #else /* !RTE_MALLOC_ASAN */
 
-#define __rte_no_asan
-
 static inline void
 asan_set_zone(void *ptr __rte_unused, size_t len __rte_unused,
 		uint32_t val __rte_unused) { }
lib/eal/include/rte_common.h +13 -0
index 67587025ab..d56a7570c0 100644
@@ -267,6 +267,19 @@ static void __attribute__((destructor(RTE_PRIO(prio)), used)) func(void)
  */
 #define __rte_cold __attribute__((cold))
 
+/**
+ * Disable AddressSanitizer on some code
+ */
+#ifdef RTE_MALLOC_ASAN
+#ifdef RTE_CC_CLANG
+#define __rte_no_asan __attribute__((no_sanitize("address", "hwaddress")))
+#else
+#define __rte_no_asan __attribute__((no_sanitize_address))
+#endif
+#else /* ! RTE_MALLOC_ASAN */
+#define __rte_no_asan
+#endif
+
 /*********** Macros for pointer arithmetic ********/
 
 /**
Signature · 1 line
2.23.0
#15 lore raw

PATCH v2 2/2 ci: build some job with ASan

In reply to David Marchand

Enable ASan, this can greatly help identify leaks and buffer overflows. Running unit tests relying on multiprocess is unreliable with ASan enabled, so skip them.

---
Changes since v1:
- reinstated table_autotest in "ASan-safe" list of ut,

---
 .ci/linux-build.sh          |   8 ++
 .github/workflows/build.yml |   3 +-
 app/test/meson.build        | 208 +++++++++++++++++++-----------------
 3 files changed, 118 insertions(+), 101 deletions(-)
.ci/linux-build.sh +8 -0
index 774a1441bf..93706c0131 100755
@@ -95,6 +95,14 @@ if [ "$MINI" = "true" ]; then
     OPTS="$OPTS -Denable_drivers=net/null"
     OPTS="$OPTS -Ddisable_libs=*"
 fi
+
+if [ "$ASAN" = "true" ]; then
+    OPTS="$OPTS -Db_sanitize=address"
+    if [ "${CC%%clang}" != "$CC" ]; then
+        OPTS="$OPTS -Db_lundef=false"
+    fi
+fi
+
 meson build --werror $OPTS
 ninja -C build
 
.github/workflows/build.yml +2 -1
index 22daaabb91..45871e76ed 100644
@@ -16,6 +16,7 @@ jobs:
     env:
       AARCH64: ${{ matrix.config.cross == 'aarch64' }}
       ABI_CHECKS: ${{ contains(matrix.config.checks, 'abi') }}
+      ASAN: ${{ contains(matrix.config.checks, 'asan') }}
       BUILD_32BIT: ${{ matrix.config.cross == 'i386' }}
       BUILD_DOCS: ${{ contains(matrix.config.checks, 'doc') }}
       CC: ccache ${{ matrix.config.compiler }}
@@ -47,7 +48,7 @@ jobs:
           - os: ubuntu-18.04
             compiler: clang
             library: shared
-            checks: doc+tests
+            checks: asan+doc+tests
           - os: ubuntu-18.04
             compiler: gcc
             library: static
app/test/meson.build +108 -100

265 lines not on this page · raw message

Signature · 1 line
2.23.0
Acked-by on v2 2/2
#16 lore raw

Re: [PATCH v2 2/2] ci: build some job with ASan

Quoted 8 lines · David Marchand

Enable ASan, this can greatly help identify leaks and buffer overflows. Running unit tests relying on multiprocess is unreliable with ASan enabled, so skip them.

---
Changes since v1:
- reinstated table_autotest in "ASan-safe" list of ut,

This is great!

#17 lore raw
In reply to David Marchand
Quoted 9 lines · David Marchand

Now that rte_malloc instrumentations are fixed, we can enable ASan in GHA.
There are still some unit tests (relying on multiprocess) that can't reliably run with ASan enabled. Those unit tests are skipped.

--
David Marchand

Quoted 8 lines

Changes since v1:
- dropped patch 2 in favor of Anatoly fix,
- rebased the last patch after other unit tests fixes have been merged,

David Marchand (2):
test/mem: disable ASan when accessing unallocated mem
ci: build some job with ASan

Series applied.
Thanks for the reviews.

Signature · 1 line
David Marchand