BPF stable patches

All for 4.9 backported and tested.

Thanks!

Alexei Starovoitov (3):
bpf: fix bpf_tail_call() x64 JIT
bpf: introduce BPF_JIT_ALWAYS_ON config
bpf: fix 32-bit divide by zero

Daniel Borkmann (3):
bpf: arsh is not supported in 32 bit alu thus reject it
bpf: avoid false sharing of map refcount with max_entries
bpf: reject stores into ctx via st and xadd

Eric Dumazet (2):
x86: bpf_jit: small optimization in emit_bpf_tail_call()
bpf: fix divides by zero

 arch/x86/net/bpf_jit_comp.c | 13 ++++++-------
 include/linux/bpf.h         | 16 ++++++++++++----
 init/Kconfig                |  7 +++++++
 kernel/bpf/core.c           | 24 +++++++++++++++++++++---
 kernel/bpf/verifier.c       | 42 ++++++++++++++++++++++++++++++++++++++++++
 lib/test_bpf.c              | 11 +++++++----
 net/core/filter.c           | 10 ++++++----
 net/core/sysctl_net_core.c  |  6 ++++++
 net/socket.c                |  9 +++++++++
 9 files changed, 116 insertions(+), 22 deletions(-)

2.9.5

From: Eric Dumazet <[email protected]>

[ upstream commit 84ccac6e7854ebbfb56d2fc6d5bef9be49bb304c ]

Saves 4 bytes replacing following instructions :

lea rax, [rsi + rdx * 8 + offsetof(...)] mov rax, qword ptr [rax]
cmp rax, 0

by :

mov rax, [rsi + rdx * 8 + offsetof(...)] test rax, rax

---
 arch/x86/net/bpf_jit_comp.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

index 15f7436..ece29e2 100644
@@ -281,7 +281,7 @@ static void emit_bpf_tail_call(u8 **pprog)
 	EMIT4(0x48, 0x8B, 0x46,                   /* mov rax, qword ptr [rsi + 16] */
 	      offsetof(struct bpf_array, map.max_entries));
 	EMIT3(0x48, 0x39, 0xD0);                  /* cmp rax, rdx */
-#define OFFSET1 47 /* number of bytes to jump */
+#define OFFSET1 43 /* number of bytes to jump */
 	EMIT2(X86_JBE, OFFSET1);                  /* jbe out */
 	label1 = cnt;
 

@@ -290,21 +290,20 @@ static void emit_bpf_tail_call(u8 **pprog)
 	 */
 	EMIT2_off32(0x8B, 0x85, -STACKSIZE + 36); /* mov eax, dword ptr [rbp - 516] */
 	EMIT3(0x83, 0xF8, MAX_TAIL_CALL_CNT);     /* cmp eax, MAX_TAIL_CALL_CNT */
-#define OFFSET2 36
+#define OFFSET2 32
 	EMIT2(X86_JA, OFFSET2);                   /* ja out */
 	label2 = cnt;
 	EMIT3(0x83, 0xC0, 0x01);                  /* add eax, 1 */
 	EMIT2_off32(0x89, 0x85, -STACKSIZE + 36); /* mov dword ptr [rbp - 516], eax */
 
 	/* prog = array->ptrs[index]; */
-	EMIT4_off32(0x48, 0x8D, 0x84, 0xD6,       /* lea rax, [rsi + rdx * 8 + offsetof(...)] */
+	EMIT4_off32(0x48, 0x8B, 0x84, 0xD6,       /* mov rax, [rsi + rdx * 8 + offsetof(...)] */
 		    offsetof(struct bpf_array, ptrs));
-	EMIT3(0x48, 0x8B, 0x00);                  /* mov rax, qword ptr [rax] */
 
 	/* if (prog == NULL)
 	 *   goto out;
 	 */
-	EMIT4(0x48, 0x83, 0xF8, 0x00);            /* cmp rax, 0 */
+	EMIT3(0x48, 0x85, 0xC0);		  /* test rax,rax */
 #define OFFSET3 10
 	EMIT2(X86_JE, OFFSET3);                   /* je out */
 	label3 = cnt;

2.9.5

Hi,

[ replaced stable@ and greg@ by netdev@ as my question below is not
relevant to stable ]

I've just noticed this on stable@. If these 4 bytes matter, why not use cmpq with an immediate value instead, which saves 2 extra bytes ? :

- the mov above is 11 bytes total :

   0:   48 8b 84 d6 78 56 34    mov    0x12345678(%rsi,%rdx,8),%rax
   7:   12 
   8:   48 85 c0                test   %rax,%rax

- the equivalent cmp is only 9 bytes :

0: 48 83 bc d6 78 56 34 cmpq $0x0,0x12345678(%rsi,%rdx,8)
7: 12 00

And as a bonus, it doesn't even clobber rax.

Just my two cents,
Willy

Hi Willy

Please look more closely at following instructions.

We need the value later, not only testing it being zero :)

Hi Eric,

Ah OK that makes total sense then ;-)

Thanks,
willy

This is a note to let you know that I've just added the patch titled

x86: bpf_jit: small optimization in emit_bpf_tail_call()

to the 4.9-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary

The filename of the patch is:
x86-bpf_jit-small-optimization-in-emit_bpf_tail_call.patch
and it can be found in the queue-4.9 subdirectory.

If you, or anyone else, feels it should not be added to the stable tree, please let <[email protected]> know about it.

From: Daniel Borkmann <[email protected]>
Date: Mon, 29 Jan 2018 02:48:54 +0100
Subject: x86: bpf_jit: small optimization in emit_bpf_tail_call()
To: [email protected]
Cc: [email protected], [email protected], Eric Dumazet <[email protected]>, Daniel Borkmann <[email protected]>, "David S . Miller" <[email protected]>
Message-ID: <0f909080f2ef055783fc7b394e8111e0df3c4971.1517190206.git.daniel@iogearbox.net>

From: Eric Dumazet <[email protected]>

[ upstream commit 84ccac6e7854ebbfb56d2fc6d5bef9be49bb304c ]

Saves 4 bytes replacing following instructions :

lea rax, [rsi + rdx * 8 + offsetof(...)]
mov rax, qword ptr [rax]
cmp rax, 0

by :

mov rax, [rsi + rdx * 8 + offsetof(...)] test rax, rax

---
 arch/x86/net/bpf_jit_comp.c |    9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

@@ -281,7 +281,7 @@ static void emit_bpf_tail_call(u8 **ppro
 	EMIT4(0x48, 0x8B, 0x46,                   /* mov rax, qword ptr [rsi + 16] */
 	      offsetof(struct bpf_array, map.max_entries));
 	EMIT3(0x48, 0x39, 0xD0);                  /* cmp rax, rdx */
-#define OFFSET1 47 /* number of bytes to jump */
+#define OFFSET1 43 /* number of bytes to jump */
 	EMIT2(X86_JBE, OFFSET1);                  /* jbe out */
 	label1 = cnt;
 

@@ -290,21 +290,20 @@ static void emit_bpf_tail_call(u8 **ppro
 	 */
 	EMIT2_off32(0x8B, 0x85, -STACKSIZE + 36); /* mov eax, dword ptr [rbp - 516] */
 	EMIT3(0x83, 0xF8, MAX_TAIL_CALL_CNT);     /* cmp eax, MAX_TAIL_CALL_CNT */
-#define OFFSET2 36
+#define OFFSET2 32
 	EMIT2(X86_JA, OFFSET2);                   /* ja out */
 	label2 = cnt;
 	EMIT3(0x83, 0xC0, 0x01);                  /* add eax, 1 */
 	EMIT2_off32(0x89, 0x85, -STACKSIZE + 36); /* mov dword ptr [rbp - 516], eax */
 
 	/* prog = array->ptrs[index]; */
-	EMIT4_off32(0x48, 0x8D, 0x84, 0xD6,       /* lea rax, [rsi + rdx * 8 + offsetof(...)] */
+	EMIT4_off32(0x48, 0x8B, 0x84, 0xD6,       /* mov rax, [rsi + rdx * 8 + offsetof(...)] */
 		    offsetof(struct bpf_array, ptrs));
-	EMIT3(0x48, 0x8B, 0x00);                  /* mov rax, qword ptr [rax] */
 
 	/* if (prog == NULL)
 	 *   goto out;
 	 */
-	EMIT4(0x48, 0x83, 0xF8, 0x00);            /* cmp rax, 0 */
+	EMIT3(0x48, 0x85, 0xC0);		  /* test rax,rax */
 #define OFFSET3 10
 	EMIT2(X86_JE, OFFSET3);                   /* je out */
 	label3 = cnt;

Patches currently in stable-queue which might be from [email protected] are

queue-4.9/bpf-avoid-false-sharing-of-map-refcount-with-max_entries.patch queue-4.9/x86-bpf_jit-small-optimization-in-emit_bpf_tail_call.patch queue-4.9/bpf-reject-stores-into-ctx-via-st-and-xadd.patch queue-4.9/bpf-fix-32-bit-divide-by-zero.patch queue-4.9/bpf-fix-bpf_tail_call-x64-jit.patch queue-4.9/bpf-arsh-is-not-supported-in-32-bit-alu-thus-reject-it.patch queue-4.9/bpf-fix-divides-by-zero.patch queue-4.9/bpf-introduce-bpf_jit_always_on-config.patch

From: Alexei Starovoitov <[email protected]>

[ upstream commit 90caccdd8cc0215705f18b92771b449b01e2474a ]

- bpf prog_array just like all other types of bpf array accepts 32-bit index. Clarify that in the comment.
- fix x64 JIT of bpf_tail_call which was incorrectly loading 8 instead of 4 bytes
- tighten corresponding check in the interpreter to stay consistent

The JIT bug can be triggered after introduction of BPF_F_NUMA_NODE flag in commit 96eabe7a40aa in 4.14. Before that the map_flags would stay zero and though JIT code is wrong it will check bounds correctly.
Hence two fixes tags. All other JITs don't have this problem.

---
 arch/x86/net/bpf_jit_comp.c | 4 ++--
 kernel/bpf/core.c           | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

index ece29e2..7840331 100644
@@ -278,9 +278,9 @@ static void emit_bpf_tail_call(u8 **pprog)
 	/* if (index >= array->map.max_entries)
 	 *   goto out;
 	 */
-	EMIT4(0x48, 0x8B, 0x46,                   /* mov rax, qword ptr [rsi + 16] */
+	EMIT2(0x89, 0xD2);                        /* mov edx, edx */
+	EMIT3(0x39, 0x56,                         /* cmp dword ptr [rsi + 16], edx */
 	      offsetof(struct bpf_array, map.max_entries));
-	EMIT3(0x48, 0x39, 0xD0);                  /* cmp rax, rdx */
 #define OFFSET1 43 /* number of bytes to jump */
 	EMIT2(X86_JBE, OFFSET1);                  /* jbe out */
 	label1 = cnt;

index aa6d981..ab9576b 100644
@@ -715,7 +715,7 @@ static unsigned int __bpf_prog_run(void *ctx, const struct bpf_insn *insn)
 		struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
 		struct bpf_array *array = container_of(map, struct bpf_array, map);
 		struct bpf_prog *prog;
-		u64 index = BPF_R3;
+		u32 index = BPF_R3;
 
 		if (unlikely(index >= array->map.max_entries))
 			goto out;

2.9.5

This is a note to let you know that I've just added the patch titled

bpf: fix bpf_tail_call() x64 JIT

to the 4.9-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary

The filename of the patch is:

     bpf-fix-bpf_tail_call-x64-jit.patch

and it can be found in the queue-4.9 subdirectory.

If you, or anyone else, feels it should not be added to the stable tree, please let <[email protected]> know about it.

From: Daniel Borkmann <[email protected]>
Date: Mon, 29 Jan 2018 02:48:55 +0100
Subject: bpf: fix bpf_tail_call() x64 JIT
To: [email protected]
Cc: [email protected], [email protected], Alexei Starovoitov <[email protected]>, "David S . Miller" <[email protected]>
Message-ID: <b7bd813935a7bc6a5f4fe4a3f199034f571c9b70.1517190206.git.daniel@iogearbox.net>

From: Alexei Starovoitov <[email protected]>

[ upstream commit 90caccdd8cc0215705f18b92771b449b01e2474a ]

- bpf prog_array just like all other types of bpf array accepts 32-bit index. Clarify that in the comment.
- fix x64 JIT of bpf_tail_call which was incorrectly loading 8 instead of 4 bytes
- tighten corresponding check in the interpreter to stay consistent

The JIT bug can be triggered after introduction of BPF_F_NUMA_NODE flag in commit 96eabe7a40aa in 4.14. Before that the map_flags would stay zero and though JIT code is wrong it will check bounds correctly.
Hence two fixes tags. All other JITs don't have this problem.

---
 arch/x86/net/bpf_jit_comp.c |    4 ++--
 kernel/bpf/core.c           |    2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

@@ -278,9 +278,9 @@ static void emit_bpf_tail_call(u8 **ppro
 	/* if (index >= array->map.max_entries)
 	 *   goto out;
 	 */
-	EMIT4(0x48, 0x8B, 0x46,                   /* mov rax, qword ptr [rsi + 16] */
+	EMIT2(0x89, 0xD2);                        /* mov edx, edx */
+	EMIT3(0x39, 0x56,                         /* cmp dword ptr [rsi + 16], edx */
 	      offsetof(struct bpf_array, map.max_entries));
-	EMIT3(0x48, 0x39, 0xD0);                  /* cmp rax, rdx */
 #define OFFSET1 43 /* number of bytes to jump */
 	EMIT2(X86_JBE, OFFSET1);                  /* jbe out */
 	label1 = cnt;

@@ -715,7 +715,7 @@ select_insn:
 		struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
 		struct bpf_array *array = container_of(map, struct bpf_array, map);
 		struct bpf_prog *prog;
-		u64 index = BPF_R3;
+		u32 index = BPF_R3;
 
 		if (unlikely(index >= array->map.max_entries))
 			goto out;

Patches currently in stable-queue which might be from [email protected] are

queue-4.9/bpf-avoid-false-sharing-of-map-refcount-with-max_entries.patch queue-4.9/x86-bpf_jit-small-optimization-in-emit_bpf_tail_call.patch queue-4.9/bpf-reject-stores-into-ctx-via-st-and-xadd.patch queue-4.9/bpf-fix-32-bit-divide-by-zero.patch queue-4.9/bpf-fix-bpf_tail_call-x64-jit.patch queue-4.9/bpf-arsh-is-not-supported-in-32-bit-alu-thus-reject-it.patch queue-4.9/bpf-fix-divides-by-zero.patch queue-4.9/bpf-introduce-bpf_jit_always_on-config.patch

From: Alexei Starovoitov <[email protected]>

[ upstream commit 290af86629b25ffd1ed6232c4e9107da031705cb ]

The BPF interpreter has been used as part of the spectre 2 attack CVE-2017-5715.

A quote from goolge project zero blog:
"At this point, it would normally be necessary to locate gadgets in the host kernel code that can be used to actually leak data by reading from an attacker-controlled location, shifting and masking the result appropriately and then using the result of that as offset to an attacker-controlled address for a load. But piecing gadgets together and figuring out which ones work in a speculation context seems annoying. So instead, we decided to use the eBPF interpreter, which is built into the host kernel - while there is no legitimate way to invoke it from inside a VM, the presence of the code in the host kernel's text section is sufficient to make it usable for the attack, just like with ordinary ROP gadgets."

To make attacker job harder introduce BPF_JIT_ALWAYS_ON config option that removes interpreter from the kernel in favor of JIT-only mode. So far eBPF JIT is supported by:
x64, arm64, arm32, sparc64, s390, powerpc64, mips64

The start of JITed program is randomized and code page is marked as read-only. In addition "constant blinding" can be turned on with net.core.bpf_jit_harden

v2->v3:
- move __bpf_prog_ret0 under ifdef (Daniel)

v1->v2:
- fix init order, test_bpf and cBPF (Daniel's feedback)
- fix offloaded bpf (Jakub's feedback)
- add 'return 0' dummy in case something can invoke prog->bpf_func
- retarget bpf tree. For bpf-next the patch would need one extra hunk. It will be sent when the trees are merged back to net-next

Considered doing:
int bpf_jit_enable __read_mostly = BPF_EBPF_JIT_DEFAULT;
but it seems better to land the patch as-is and in bpf-next remove bpf_jit_enable global variable from all JITs, consolidate in one place and remove this jit_init() function.

---
 init/Kconfig               |  7 +++++++
 kernel/bpf/core.c          | 18 ++++++++++++++++++
 lib/test_bpf.c             | 11 +++++++----
 net/core/filter.c          |  6 ++----
 net/core/sysctl_net_core.c |  6 ++++++
 net/socket.c               |  9 +++++++++
 6 files changed, 49 insertions(+), 8 deletions(-)

index 34407f1..b331fee 100644
@@ -1609,6 +1609,13 @@ config BPF_SYSCALL
 	  Enable the bpf() system call that allows to manipulate eBPF
 	  programs and maps via file descriptors.
 
+config BPF_JIT_ALWAYS_ON
+	bool "Permanently enable BPF JIT and remove BPF interpreter"
+	depends on BPF_SYSCALL && HAVE_EBPF_JIT && BPF_JIT
+	help
+	  Enables BPF JIT and removes BPF interpreter to avoid
+	  speculative execution of BPF instructions by the interpreter
+
 config SHMEM
 	bool "Use full shmem filesystem" if EXPERT
 	default y

index ab9576b..64c4b13 100644
@@ -458,6 +458,7 @@ noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
 }
 EXPORT_SYMBOL_GPL(__bpf_call_base);
 
+#ifndef CONFIG_BPF_JIT_ALWAYS_ON
 /**
  *	__bpf_prog_run - run eBPF program on a given context
  *	@ctx: is the data we are operating on

@@ -923,6 +924,13 @@ static unsigned int __bpf_prog_run(void *ctx, const struct bpf_insn *insn)
 }
 STACK_FRAME_NON_STANDARD(__bpf_prog_run); /* jump table */
 
+#else
+static unsigned int __bpf_prog_ret0(void *ctx, const struct bpf_insn *insn)
+{
+	return 0;
+}
+#endif
+
 bool bpf_prog_array_compatible(struct bpf_array *array,
 			       const struct bpf_prog *fp)
 {

@@ -970,7 +978,11 @@ static int bpf_check_tail_call(const struct bpf_prog *fp)
  */
 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
 {
+#ifndef CONFIG_BPF_JIT_ALWAYS_ON
 	fp->bpf_func = (void *) __bpf_prog_run;
+#else
+	fp->bpf_func = (void *) __bpf_prog_ret0;
+#endif
 
 	/* eBPF JITs can rewrite the program in case constant
 	 * blinding is active. However, in case of error during

@@ -979,6 +991,12 @@ struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
 	 * be JITed, but falls back to the interpreter.
 	 */
 	fp = bpf_int_jit_compile(fp);
+#ifdef CONFIG_BPF_JIT_ALWAYS_ON
+	if (!fp->jited) {
+		*err = -ENOTSUPP;
+		return fp;
+	}
+#endif
 	bpf_prog_lock_ro(fp);
 
 	/* The tail call compatibility check can only be done at

index 2e38502..98da752 100644
@@ -5646,9 +5646,8 @@ static struct bpf_prog *generate_filter(int which, int *err)
 				return NULL;
 			}
 		}
-		/* We don't expect to fail. */
 		if (*err) {
-			pr_cont("FAIL to attach err=%d len=%d\n",
+			pr_cont("FAIL to prog_create err=%d len=%d\n",
 				*err, fprog.len);
 			return NULL;
 		}

@@ -5671,6 +5670,10 @@ static struct bpf_prog *generate_filter(int which, int *err)
 		 * checks.
 		 */
 		fp = bpf_prog_select_runtime(fp, err);
+		if (*err) {
+			pr_cont("FAIL to select_runtime err=%d\n", *err);
+			return NULL;
+		}
 		break;
 	}
 

@@ -5856,8 +5859,8 @@ static __init int test_bpf(void)
 				pass_cnt++;
 				continue;
 			}
-
-			return err;
+			err_cnt++;
+			continue;
 		}
 
 		pr_cont("jited:%u ", fp->jited);

index 4eb4ce0..6e0d17b 100644
@@ -1005,11 +1005,9 @@ static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
 		 */
 		goto out_err_free;
 
-	/* We are guaranteed to never error here with cBPF to eBPF
-	 * transitions, since there's no issue with type compatibility
-	 * checks on program arrays.
-	 */
 	fp = bpf_prog_select_runtime(fp, &err);
+	if (err)
+		goto out_err_free;
 
 	kfree(old_prog);
 	return fp;

index a7f05f0..1b46190 100644
@@ -292,7 +292,13 @@ static struct ctl_table net_core_table[] = {
 		.data		= &bpf_jit_enable,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
+#ifndef CONFIG_BPF_JIT_ALWAYS_ON
 		.proc_handler	= proc_dointvec
+#else
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &one,
+		.extra2		= &one,
+#endif
 	},
 # ifdef CONFIG_HAVE_EBPF_JIT
 	{

index 05f13b2..bd3b339 100644
@@ -2548,6 +2548,15 @@ static int __init sock_init(void)
 
 core_initcall(sock_init);	/* early initcall */
 
+static int __init jit_init(void)
+{
+#ifdef CONFIG_BPF_JIT_ALWAYS_ON
+	bpf_jit_enable = 1;
+#endif
+	return 0;
+}
+pure_initcall(jit_init);
+
 #ifdef CONFIG_PROC_FS
 void socket_seq_show(struct seq_file *seq)
 {

2.9.5

This is a note to let you know that I've just added the patch titled

bpf: introduce BPF_JIT_ALWAYS_ON config

to the 4.9-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary

The filename of the patch is:
bpf-introduce-bpf_jit_always_on-config.patch
and it can be found in the queue-4.9 subdirectory.

If you, or anyone else, feels it should not be added to the stable tree, please let <[email protected]> know about it.

From: Daniel Borkmann <[email protected]>
Date: Mon, 29 Jan 2018 02:48:56 +0100
Subject: bpf: introduce BPF_JIT_ALWAYS_ON config
To: [email protected]
Cc: [email protected], [email protected], Daniel Borkmann <[email protected]>
Message-ID: <e2902a6193fc8672f097d4181fe9573e64171d4e.1517190206.git.daniel@iogearbox.net>

From: Alexei Starovoitov <[email protected]>

[ upstream commit 290af86629b25ffd1ed6232c4e9107da031705cb ]

The BPF interpreter has been used as part of the spectre 2 attack CVE-2017-5715.

A quote from goolge project zero blog:
"At this point, it would normally be necessary to locate gadgets in the host kernel code that can be used to actually leak data by reading from an attacker-controlled location, shifting and masking the result appropriately and then using the result of that as offset to an attacker-controlled address for a load. But piecing gadgets together and figuring out which ones work in a speculation context seems annoying. So instead, we decided to use the eBPF interpreter, which is built into the host kernel - while there is no legitimate way to invoke it from inside a VM, the presence of the code in the host kernel's text section is sufficient to make it usable for the attack, just like with ordinary ROP gadgets."

To make attacker job harder introduce BPF_JIT_ALWAYS_ON config option that removes interpreter from the kernel in favor of JIT-only mode. So far eBPF JIT is supported by:
x64, arm64, arm32, sparc64, s390, powerpc64, mips64

The start of JITed program is randomized and code page is marked as read-only. In addition "constant blinding" can be turned on with net.core.bpf_jit_harden

v2->v3:
- move __bpf_prog_ret0 under ifdef (Daniel)

v1->v2:
- fix init order, test_bpf and cBPF (Daniel's feedback)
- fix offloaded bpf (Jakub's feedback)
- add 'return 0' dummy in case something can invoke prog->bpf_func
- retarget bpf tree. For bpf-next the patch would need one extra hunk. It will be sent when the trees are merged back to net-next

Considered doing:
int bpf_jit_enable __read_mostly = BPF_EBPF_JIT_DEFAULT;
but it seems better to land the patch as-is and in bpf-next remove bpf_jit_enable global variable from all JITs, consolidate in one place and remove this jit_init() function.

---
 init/Kconfig               |    7 +++++++
 kernel/bpf/core.c          |   18 ++++++++++++++++++
 lib/test_bpf.c             |   11 +++++++----
 net/core/filter.c          |    6 ++----
 net/core/sysctl_net_core.c |    6 ++++++
 net/socket.c               |    9 +++++++++
 6 files changed, 49 insertions(+), 8 deletions(-)

@@ -1609,6 +1609,13 @@ config BPF_SYSCALL
 	  Enable the bpf() system call that allows to manipulate eBPF
 	  programs and maps via file descriptors.
 
+config BPF_JIT_ALWAYS_ON
+	bool "Permanently enable BPF JIT and remove BPF interpreter"
+	depends on BPF_SYSCALL && HAVE_EBPF_JIT && BPF_JIT
+	help
+	  Enables BPF JIT and removes BPF interpreter to avoid
+	  speculative execution of BPF instructions by the interpreter
+
 config SHMEM
 	bool "Use full shmem filesystem" if EXPERT
 	default y

@@ -458,6 +458,7 @@ noinline u64 __bpf_call_base(u64 r1, u64
 }
 EXPORT_SYMBOL_GPL(__bpf_call_base);
 
+#ifndef CONFIG_BPF_JIT_ALWAYS_ON
 /**
  *	__bpf_prog_run - run eBPF program on a given context
  *	@ctx: is the data we are operating on

@@ -923,6 +924,13 @@ load_byte:
 }
 STACK_FRAME_NON_STANDARD(__bpf_prog_run); /* jump table */
 
+#else
+static unsigned int __bpf_prog_ret0(void *ctx, const struct bpf_insn *insn)
+{
+	return 0;
+}
+#endif
+
 bool bpf_prog_array_compatible(struct bpf_array *array,
 			       const struct bpf_prog *fp)
 {

@@ -970,7 +978,11 @@ static int bpf_check_tail_call(const str
  */
 struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
 {
+#ifndef CONFIG_BPF_JIT_ALWAYS_ON
 	fp->bpf_func = (void *) __bpf_prog_run;
+#else
+	fp->bpf_func = (void *) __bpf_prog_ret0;
+#endif
 
 	/* eBPF JITs can rewrite the program in case constant
 	 * blinding is active. However, in case of error during

@@ -979,6 +991,12 @@ struct bpf_prog *bpf_prog_select_runtime
 	 * be JITed, but falls back to the interpreter.
 	 */
 	fp = bpf_int_jit_compile(fp);
+#ifdef CONFIG_BPF_JIT_ALWAYS_ON
+	if (!fp->jited) {
+		*err = -ENOTSUPP;
+		return fp;
+	}
+#endif
 	bpf_prog_lock_ro(fp);
 
 	/* The tail call compatibility check can only be done at

@@ -5646,9 +5646,8 @@ static struct bpf_prog *generate_filter(
 				return NULL;
 			}
 		}
-		/* We don't expect to fail. */
 		if (*err) {
-			pr_cont("FAIL to attach err=%d len=%d\n",
+			pr_cont("FAIL to prog_create err=%d len=%d\n",
 				*err, fprog.len);
 			return NULL;
 		}

@@ -5671,6 +5670,10 @@ static struct bpf_prog *generate_filter(
 		 * checks.
 		 */
 		fp = bpf_prog_select_runtime(fp, err);
+		if (*err) {
+			pr_cont("FAIL to select_runtime err=%d\n", *err);
+			return NULL;
+		}
 		break;
 	}
 

@@ -5856,8 +5859,8 @@ static __init int test_bpf(void)
 				pass_cnt++;
 				continue;
 			}
-
-			return err;
+			err_cnt++;
+			continue;
 		}
 
 		pr_cont("jited:%u ", fp->jited);

@@ -1005,11 +1005,9 @@ static struct bpf_prog *bpf_migrate_filt
 		 */
 		goto out_err_free;
 
-	/* We are guaranteed to never error here with cBPF to eBPF
-	 * transitions, since there's no issue with type compatibility
-	 * checks on program arrays.
-	 */
 	fp = bpf_prog_select_runtime(fp, &err);
+	if (err)
+		goto out_err_free;
 
 	kfree(old_prog);
 	return fp;

@@ -292,7 +292,13 @@ static struct ctl_table net_core_table[]
 		.data		= &bpf_jit_enable,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
+#ifndef CONFIG_BPF_JIT_ALWAYS_ON
 		.proc_handler	= proc_dointvec
+#else
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &one,
+		.extra2		= &one,
+#endif
 	},
 # ifdef CONFIG_HAVE_EBPF_JIT
 	{

@@ -2548,6 +2548,15 @@ out_fs:
 
 core_initcall(sock_init);	/* early initcall */
 
+static int __init jit_init(void)
+{
+#ifdef CONFIG_BPF_JIT_ALWAYS_ON
+	bpf_jit_enable = 1;
+#endif
+	return 0;
+}
+pure_initcall(jit_init);
+
 #ifdef CONFIG_PROC_FS
 void socket_seq_show(struct seq_file *seq)
 {

Patches currently in stable-queue which might be from [email protected] are

queue-4.9/bpf-avoid-false-sharing-of-map-refcount-with-max_entries.patch queue-4.9/x86-bpf_jit-small-optimization-in-emit_bpf_tail_call.patch queue-4.9/bpf-reject-stores-into-ctx-via-st-and-xadd.patch queue-4.9/bpf-fix-32-bit-divide-by-zero.patch queue-4.9/bpf-fix-bpf_tail_call-x64-jit.patch queue-4.9/bpf-arsh-is-not-supported-in-32-bit-alu-thus-reject-it.patch queue-4.9/bpf-fix-divides-by-zero.patch queue-4.9/bpf-introduce-bpf_jit_always_on-config.patch

[ upstream commit 7891a87efc7116590eaba57acc3c422487802c6f ]

The following snippet was throwing an 'unknown opcode cc' warning in BPF interpreter:

0: (18) r0 = 0x0
2: (7b) *(u64 *)(r10 -16) = r0
3: (cc) (u32) r0 s>>= (u32) r0
4: (95) exit

Although a number of JITs do support BPF_ALU | BPF_ARSH | BPF_{K,X} generation, not all of them do and interpreter does neither. We can leave existing ones and implement it later in bpf-next for the remaining ones, but reject this properly in verifier for the time being.

---
 kernel/bpf/verifier.c | 5 +++++
 1 file changed, 5 insertions(+)

index 19c44cf..8e0bf89 100644
@@ -1843,6 +1843,11 @@ static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn)
 			return -EINVAL;
 		}
 
+		if (opcode == BPF_ARSH && BPF_CLASS(insn->code) != BPF_ALU64) {
+			verbose("BPF_ARSH not supported for 32 bit ALU\n");
+			return -EINVAL;
+		}
+
 		if ((opcode == BPF_LSH || opcode == BPF_RSH ||
 		     opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
 			int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;

2.9.5

This is a note to let you know that I've just added the patch titled

bpf: arsh is not supported in 32 bit alu thus reject it

to the 4.9-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary

The filename of the patch is:
bpf-arsh-is-not-supported-in-32-bit-alu-thus-reject-it.patch
and it can be found in the queue-4.9 subdirectory.

If you, or anyone else, feels it should not be added to the stable tree, please let <[email protected]> know about it.

From: Daniel Borkmann <[email protected]>
Date: Mon, 29 Jan 2018 02:48:57 +0100
Subject: bpf: arsh is not supported in 32 bit alu thus reject it
To: [email protected]
Cc: [email protected], [email protected], Daniel Borkmann <[email protected]>
Message-ID: <f236d9b875aa2ab63ab155cafa0f10376883da00.1517190206.git.daniel@iogearbox.net>

From: Daniel Borkmann <[email protected]>

[ upstream commit 7891a87efc7116590eaba57acc3c422487802c6f ]

The following snippet was throwing an 'unknown opcode cc' warning in BPF interpreter:

0: (18) r0 = 0x0
2: (7b) *(u64 *)(r10 -16) = r0
3: (cc) (u32) r0 s>>= (u32) r0
4: (95) exit

Although a number of JITs do support BPF_ALU | BPF_ARSH | BPF_{K,X} generation, not all of them do and interpreter does neither. We can leave existing ones and implement it later in bpf-next for the remaining ones, but reject this properly in verifier for the time being.

---
 kernel/bpf/verifier.c |    5 +++++
 1 file changed, 5 insertions(+)

@@ -1843,6 +1843,11 @@ static int check_alu_op(struct bpf_verif
 			return -EINVAL;
 		}
 
+		if (opcode == BPF_ARSH && BPF_CLASS(insn->code) != BPF_ALU64) {
+			verbose("BPF_ARSH not supported for 32 bit ALU\n");
+			return -EINVAL;
+		}
+
 		if ((opcode == BPF_LSH || opcode == BPF_RSH ||
 		     opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) {
 			int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32;

Patches currently in stable-queue which might be from [email protected] are

queue-4.9/bpf-avoid-false-sharing-of-map-refcount-with-max_entries.patch queue-4.9/x86-bpf_jit-small-optimization-in-emit_bpf_tail_call.patch queue-4.9/bpf-reject-stores-into-ctx-via-st-and-xadd.patch queue-4.9/bpf-fix-32-bit-divide-by-zero.patch queue-4.9/bpf-fix-bpf_tail_call-x64-jit.patch queue-4.9/bpf-arsh-is-not-supported-in-32-bit-alu-thus-reject-it.patch queue-4.9/bpf-fix-divides-by-zero.patch queue-4.9/bpf-introduce-bpf_jit_always_on-config.patch

[ upstream commit be95a845cc4402272994ce290e3ad928aff06cb9 ]

In addition to commit b2157399cc98 ("bpf: prevent out-of-bounds speculation") also change the layout of struct bpf_map such that false sharing of fast-path members like max_entries is avoided when the maps reference counter is altered. Therefore enforce them to be placed into separate cachelines.

pahole dump after change:

  struct bpf_map {
        const struct bpf_map_ops  * ops;                 /*     0     8 */
        struct bpf_map *           inner_map_meta;       /*     8     8 */
        void *                     security;             /*    16     8 */
        enum bpf_map_type          map_type;             /*    24     4 */
        u32                        key_size;             /*    28     4 */
        u32                        value_size;           /*    32     4 */
        u32                        max_entries;          /*    36     4 */
        u32                        map_flags;            /*    40     4 */
        u32                        pages;                /*    44     4 */
        u32                        id;                   /*    48     4 */
        int                        numa_node;            /*    52     4 */
        bool                       unpriv_array;         /*    56     1 */

/* XXX 7 bytes hole, try to pack */

/* --- cacheline 1 boundary (64 bytes) --- */
struct user_struct * user; /* 64 8 */
atomic_t refcnt; /* 72 4 */
atomic_t usercnt; /* 76 4 */
struct work_struct work; /* 80 32 */
char name[16]; /* 112 16 */
/* --- cacheline 2 boundary (128 bytes) --- */

/* size: 128, cachelines: 2, members: 17 */ /* sum members: 121, holes: 1, sum holes: 7 */
};

Now all entries in the first cacheline are read only throughout the life time of the map, set up once during map creation. Overall struct size and number of cachelines doesn't change from the reordering. struct bpf_map is usually first member and embedded in map structs in specific map implementations, so also avoid those members to sit at the end where it could potentially share the cacheline with first map values e.g. in the array since remote CPUs could trigger map updates just as well for those (easily dirtying members like max_entries intentionally as well) while having subsequent values in cache.

Quoting from Google's Project Zero blog [1]:

Additionally, at least on the Intel machine on which this was tested, bouncing modified cache lines between cores is slow, apparently because the MESI protocol is used for cache coherence [8]. Changing the reference counter of an eBPF array on one physical CPU core causes the cache line containing the reference counter to be bounced over to that CPU core, making reads of the reference counter on all other CPU cores slow until the changed reference counter has been written back to memory. Because the length and the reference counter of an eBPF array are stored in the same cache line, this also means that changing the reference counter on one physical CPU core causes reads of the eBPF array's length to be slow on other physical CPU cores (intentional false sharing).

While this doesn't 'control' the out-of-bounds speculation through masking the index as in commit b2157399cc98, triggering a manipulation of the map's reference counter is really trivial, so lets not allow to easily affect max_entries from it.

Splitting to separate cachelines also generally makes sense from a performance perspective anyway in that fast-path won't have a cache miss if the map gets pinned, reused in other progs, etc out of control path, thus also avoids unintentional false sharing.

[1] https://googleprojectzero.blogspot.ch/2018/01/reading-privileged-memory-with-side.html

---
 include/linux/bpf.h | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

index 75ffd3b..7995940 100644
@@ -36,7 +36,10 @@ struct bpf_map_ops {
 };
 
 struct bpf_map {
-	atomic_t refcnt;
+	/* 1st cacheline with read-mostly members of which some
+	 * are also accessed in fast-path (e.g. ops, max_entries).
+	 */
+	const struct bpf_map_ops *ops ____cacheline_aligned;
 	enum bpf_map_type map_type;
 	u32 key_size;
 	u32 value_size;

@@ -44,10 +47,15 @@ struct bpf_map {
 	u32 map_flags;
 	u32 pages;
 	bool unpriv_array;
-	struct user_struct *user;
-	const struct bpf_map_ops *ops;
-	struct work_struct work;
+	/* 7 bytes hole */
+
+	/* 2nd cacheline with misc members to avoid false sharing
+	 * particularly with refcounting.
+	 */
+	struct user_struct *user ____cacheline_aligned;
+	atomic_t refcnt;
 	atomic_t usercnt;
+	struct work_struct work;
 };
 
 struct bpf_map_type_list {

2.9.5

This is a note to let you know that I've just added the patch titled

bpf: avoid false sharing of map refcount with max_entries

to the 4.9-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary

The filename of the patch is:
bpf-avoid-false-sharing-of-map-refcount-with-max_entries.patch
and it can be found in the queue-4.9 subdirectory.

If you, or anyone else, feels it should not be added to the stable tree, please let <[email protected]> know about it.

From: Daniel Borkmann <[email protected]>
Date: Mon, 29 Jan 2018 02:48:58 +0100
Subject: bpf: avoid false sharing of map refcount with max_entries
To: [email protected]
Cc: [email protected], [email protected], Daniel Borkmann <[email protected]>
Message-ID: <1ce4bdf60f6950b69dd527c41fb93f92cd3cfedc.1517190207.git.daniel@iogearbox.net>

From: Daniel Borkmann <[email protected]>

[ upstream commit be95a845cc4402272994ce290e3ad928aff06cb9 ]

In addition to commit b2157399cc98 ("bpf: prevent out-of-bounds speculation") also change the layout of struct bpf_map such that false sharing of fast-path members like max_entries is avoided when the maps reference counter is altered. Therefore enforce them to be placed into separate cachelines.

pahole dump after change:

  struct bpf_map {
        const struct bpf_map_ops  * ops;                 /*     0     8 */
        struct bpf_map *           inner_map_meta;       /*     8     8 */
        void *                     security;             /*    16     8 */
        enum bpf_map_type          map_type;             /*    24     4 */
        u32                        key_size;             /*    28     4 */
        u32                        value_size;           /*    32     4 */
        u32                        max_entries;          /*    36     4 */
        u32                        map_flags;            /*    40     4 */
        u32                        pages;                /*    44     4 */
        u32                        id;                   /*    48     4 */
        int                        numa_node;            /*    52     4 */
        bool                       unpriv_array;         /*    56     1 */

/* XXX 7 bytes hole, try to pack */

/* --- cacheline 1 boundary (64 bytes) --- */
struct user_struct * user; /* 64 8 */
atomic_t refcnt; /* 72 4 */
atomic_t usercnt; /* 76 4 */
struct work_struct work; /* 80 32 */
char name[16]; /* 112 16 */
/* --- cacheline 2 boundary (128 bytes) --- */

/* size: 128, cachelines: 2, members: 17 */ /* sum members: 121, holes: 1, sum holes: 7 */
};

Now all entries in the first cacheline are read only throughout the life time of the map, set up once during map creation. Overall struct size and number of cachelines doesn't change from the reordering. struct bpf_map is usually first member and embedded in map structs in specific map implementations, so also avoid those members to sit at the end where it could potentially share the cacheline with first map values e.g. in the array since remote CPUs could trigger map updates just as well for those (easily dirtying members like max_entries intentionally as well) while having subsequent values in cache.

Quoting from Google's Project Zero blog [1]:

Additionally, at least on the Intel machine on which this was tested, bouncing modified cache lines between cores is slow, apparently because the MESI protocol is used for cache coherence [8]. Changing the reference counter of an eBPF array on one physical CPU core causes the cache line containing the reference counter to be bounced over to that CPU core, making reads of the reference counter on all other CPU cores slow until the changed reference counter has been written back to memory. Because the length and the reference counter of an eBPF array are stored in the same cache line, this also means that changing the reference counter on one physical CPU core causes reads of the eBPF array's length to be slow on other physical CPU cores (intentional false sharing).

While this doesn't 'control' the out-of-bounds speculation through masking the index as in commit b2157399cc98, triggering a manipulation of the map's reference counter is really trivial, so lets not allow to easily affect max_entries from it.

Splitting to separate cachelines also generally makes sense from a performance perspective anyway in that fast-path won't have a cache miss if the map gets pinned, reused in other progs, etc out of control path, thus also avoids unintentional false sharing.

[1] https://googleprojectzero.blogspot.ch/2018/01/reading-privileged-memory-with-side.html

---
 include/linux/bpf.h |   16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

@@ -36,7 +36,10 @@ struct bpf_map_ops {
 };
 
 struct bpf_map {
-	atomic_t refcnt;
+	/* 1st cacheline with read-mostly members of which some
+	 * are also accessed in fast-path (e.g. ops, max_entries).
+	 */
+	const struct bpf_map_ops *ops ____cacheline_aligned;
 	enum bpf_map_type map_type;
 	u32 key_size;
 	u32 value_size;

@@ -44,10 +47,15 @@ struct bpf_map {
 	u32 map_flags;
 	u32 pages;
 	bool unpriv_array;
-	struct user_struct *user;
-	const struct bpf_map_ops *ops;
-	struct work_struct work;
+	/* 7 bytes hole */
+
+	/* 2nd cacheline with misc members to avoid false sharing
+	 * particularly with refcounting.
+	 */
+	struct user_struct *user ____cacheline_aligned;
+	atomic_t refcnt;
 	atomic_t usercnt;
+	struct work_struct work;
 };
 
 struct bpf_map_type_list {

Patches currently in stable-queue which might be from [email protected] are

queue-4.9/bpf-avoid-false-sharing-of-map-refcount-with-max_entries.patch queue-4.9/x86-bpf_jit-small-optimization-in-emit_bpf_tail_call.patch queue-4.9/bpf-reject-stores-into-ctx-via-st-and-xadd.patch queue-4.9/bpf-fix-32-bit-divide-by-zero.patch queue-4.9/bpf-fix-bpf_tail_call-x64-jit.patch queue-4.9/bpf-arsh-is-not-supported-in-32-bit-alu-thus-reject-it.patch queue-4.9/bpf-fix-divides-by-zero.patch queue-4.9/bpf-introduce-bpf_jit_always_on-config.patch

From: Eric Dumazet <[email protected]>

[ upstream commit c366287ebd698ef5e3de300d90cd62ee9ee7373e ]

Divides by zero are not nice, lets avoid them if possible.

Also do_div() seems not needed when dealing with 32bit operands, but this seems a minor detail.

---
 kernel/bpf/core.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

index 64c4b13..879ca84 100644
@@ -642,7 +642,7 @@ static unsigned int __bpf_prog_run(void *ctx, const struct bpf_insn *insn)
 		DST = tmp;
 		CONT;
 	ALU_MOD_X:
-		if (unlikely(SRC == 0))
+		if (unlikely((u32)SRC == 0))
 			return 0;
 		tmp = (u32) DST;
 		DST = do_div(tmp, (u32) SRC);

@@ -661,7 +661,7 @@ static unsigned int __bpf_prog_run(void *ctx, const struct bpf_insn *insn)
 		DST = div64_u64(DST, SRC);
 		CONT;
 	ALU_DIV_X:
-		if (unlikely(SRC == 0))
+		if (unlikely((u32)SRC == 0))
 			return 0;
 		tmp = (u32) DST;
 		do_div(tmp, (u32) SRC);

2.9.5

This is a note to let you know that I've just added the patch titled

bpf: fix divides by zero

to the 4.9-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary

The filename of the patch is:
bpf-fix-divides-by-zero.patch
and it can be found in the queue-4.9 subdirectory.

If you, or anyone else, feels it should not be added to the stable tree, please let <[email protected]> know about it.

From: Daniel Borkmann <[email protected]>
Date: Mon, 29 Jan 2018 02:48:59 +0100
Subject: bpf: fix divides by zero
To: [email protected]
Cc: [email protected], [email protected], Eric Dumazet <[email protected]>
Message-ID: <aacafad0b7353aa4acc993dc74ea5168b31d85ab.1517190207.git.daniel@iogearbox.net>

From: Eric Dumazet <[email protected]>

[ upstream commit c366287ebd698ef5e3de300d90cd62ee9ee7373e ]

Divides by zero are not nice, lets avoid them if possible.

Also do_div() seems not needed when dealing with 32bit operands, but this seems a minor detail.

---
 kernel/bpf/core.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

@@ -642,7 +642,7 @@ select_insn:
 		DST = tmp;
 		CONT;
 	ALU_MOD_X:
-		if (unlikely(SRC == 0))
+		if (unlikely((u32)SRC == 0))
 			return 0;
 		tmp = (u32) DST;
 		DST = do_div(tmp, (u32) SRC);

@@ -661,7 +661,7 @@ select_insn:
 		DST = div64_u64(DST, SRC);
 		CONT;
 	ALU_DIV_X:
-		if (unlikely(SRC == 0))
+		if (unlikely((u32)SRC == 0))
 			return 0;
 		tmp = (u32) DST;
 		do_div(tmp, (u32) SRC);

Patches currently in stable-queue which might be from [email protected] are

queue-4.9/bpf-avoid-false-sharing-of-map-refcount-with-max_entries.patch queue-4.9/x86-bpf_jit-small-optimization-in-emit_bpf_tail_call.patch queue-4.9/bpf-reject-stores-into-ctx-via-st-and-xadd.patch queue-4.9/bpf-fix-32-bit-divide-by-zero.patch queue-4.9/bpf-fix-bpf_tail_call-x64-jit.patch queue-4.9/bpf-arsh-is-not-supported-in-32-bit-alu-thus-reject-it.patch queue-4.9/bpf-fix-divides-by-zero.patch queue-4.9/bpf-introduce-bpf_jit_always_on-config.patch

From: Alexei Starovoitov <[email protected]>

[ upstream commit 68fda450a7df51cff9e5a4d4a4d9d0d5f2589153 ]

due to some JITs doing if (src_reg == 0) check in 64-bit mode for div/mod operations mask upper 32-bits of src register before doing the check

---
 kernel/bpf/verifier.c | 18 ++++++++++++++++++
 net/core/filter.c     |  4 ++++
 2 files changed, 22 insertions(+)

index 8e0bf89..5c56ff9 100644
@@ -3391,6 +3391,24 @@ static int fixup_bpf_calls(struct bpf_verifier_env *env)
 
 
 	for (i = 0; i < insn_cnt; i++, insn++) {
+		if (insn->code == (BPF_ALU | BPF_MOD | BPF_X) ||
+		    insn->code == (BPF_ALU | BPF_DIV | BPF_X)) {
+			/* due to JIT bugs clear upper 32-bits of src register
+			 * before div/mod operation
+			 */
+			insn_buf[0] = BPF_MOV32_REG(insn->src_reg, insn->src_reg);
+			insn_buf[1] = *insn;
+			cnt = 2;
+			new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
+			if (!new_prog)
+				return -ENOMEM;
+
+			delta    += cnt - 1;
+			env->prog = prog = new_prog;
+			insn      = new_prog->insnsi + i + delta;
+			continue;
+		}
+
 		if (insn->code != (BPF_JMP | BPF_CALL))
 			continue;
 

index 6e0d17b..e8c89d2 100644
@@ -441,6 +441,10 @@ static int bpf_convert_filter(struct sock_filter *prog, int len,
 			    convert_bpf_extensions(fp, &insn))
 				break;
 
+			if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
+			    fp->code == (BPF_ALU | BPF_MOD | BPF_X))
+				*insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
+
 			*insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 			break;
 

2.9.5

This is a note to let you know that I've just added the patch titled

bpf: fix 32-bit divide by zero

to the 4.9-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary

The filename of the patch is:

     bpf-fix-32-bit-divide-by-zero.patch

and it can be found in the queue-4.9 subdirectory.

If you, or anyone else, feels it should not be added to the stable tree, please let <[email protected]> know about it.

From: Daniel Borkmann <[email protected]>
Date: Mon, 29 Jan 2018 02:49:00 +0100
Subject: bpf: fix 32-bit divide by zero
To: [email protected]
Cc: [email protected], [email protected], Daniel Borkmann <[email protected]>
Message-ID: <0fe1fff7b023f38a11f9bebd975e29b78e5c788b.1517190207.git.daniel@iogearbox.net>

From: Alexei Starovoitov <[email protected]>

[ upstream commit 68fda450a7df51cff9e5a4d4a4d9d0d5f2589153 ]

due to some JITs doing if (src_reg == 0) check in 64-bit mode for div/mod operations mask upper 32-bits of src register before doing the check

---
 kernel/bpf/verifier.c |   18 ++++++++++++++++++
 net/core/filter.c     |    4 ++++
 2 files changed, 22 insertions(+)

@@ -3391,6 +3391,24 @@ static int fixup_bpf_calls(struct bpf_ve
 
 
 	for (i = 0; i < insn_cnt; i++, insn++) {
+		if (insn->code == (BPF_ALU | BPF_MOD | BPF_X) ||
+		    insn->code == (BPF_ALU | BPF_DIV | BPF_X)) {
+			/* due to JIT bugs clear upper 32-bits of src register
+			 * before div/mod operation
+			 */
+			insn_buf[0] = BPF_MOV32_REG(insn->src_reg, insn->src_reg);
+			insn_buf[1] = *insn;
+			cnt = 2;
+			new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt);
+			if (!new_prog)
+				return -ENOMEM;
+
+			delta    += cnt - 1;
+			env->prog = prog = new_prog;
+			insn      = new_prog->insnsi + i + delta;
+			continue;
+		}
+
 		if (insn->code != (BPF_JMP | BPF_CALL))
 			continue;
 

@@ -441,6 +441,10 @@ do_pass:
 			    convert_bpf_extensions(fp, &insn))
 				break;
 
+			if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
+			    fp->code == (BPF_ALU | BPF_MOD | BPF_X))
+				*insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
+
 			*insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 			break;
 

Patches currently in stable-queue which might be from [email protected] are

queue-4.9/bpf-avoid-false-sharing-of-map-refcount-with-max_entries.patch queue-4.9/x86-bpf_jit-small-optimization-in-emit_bpf_tail_call.patch queue-4.9/bpf-reject-stores-into-ctx-via-st-and-xadd.patch queue-4.9/bpf-fix-32-bit-divide-by-zero.patch queue-4.9/bpf-fix-bpf_tail_call-x64-jit.patch queue-4.9/bpf-arsh-is-not-supported-in-32-bit-alu-thus-reject-it.patch queue-4.9/bpf-fix-divides-by-zero.patch queue-4.9/bpf-introduce-bpf_jit_always_on-config.patch

[ upstream commit f37a8cb84cce18762e8f86a70bd6a49a66ab964c ]

Alexei found that verifier does not reject stores into context via BPF_ST instead of BPF_STX. And while looking at it, we also should not allow XADD variant of BPF_STX.

The context rewriter is only assuming either BPF_LDX_MEM- or BPF_STX_MEM-type operations, thus reject anything other than that so that assumptions in the rewriter properly hold. Add test cases as well for BPF selftests.

---
 kernel/bpf/verifier.c | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

index 5c56ff9..076e4a0 100644
@@ -702,6 +702,13 @@ static bool is_pointer_value(struct bpf_verifier_env *env, int regno)
 	return __is_pointer_value(env->allow_ptr_leaks, &env->cur_state.regs[regno]);
 }
 
+static bool is_ctx_reg(struct bpf_verifier_env *env, int regno)
+{
+	const struct bpf_reg_state *reg = &env->cur_state.regs[regno];
+
+	return reg->type == PTR_TO_CTX;
+}
+
 static int check_ptr_alignment(struct bpf_verifier_env *env,
 			       struct bpf_reg_state *reg, int off, int size)
 {

@@ -896,6 +903,12 @@ static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn)
 		return -EACCES;
 	}
 
+	if (is_ctx_reg(env, insn->dst_reg)) {
+		verbose("BPF_XADD stores into R%d context is not allowed\n",
+			insn->dst_reg);
+		return -EACCES;
+	}
+
 	/* check whether atomic_add can read the memory */
 	err = check_mem_access(env, insn->dst_reg, insn->off,
 			       BPF_SIZE(insn->code), BPF_READ, -1);

@@ -3012,6 +3025,12 @@ static int do_check(struct bpf_verifier_env *env)
 			if (err)
 				return err;
 
+			if (is_ctx_reg(env, insn->dst_reg)) {
+				verbose("BPF_ST stores into R%d context is not allowed\n",
+					insn->dst_reg);
+				return -EACCES;
+			}
+
 			/* check that memory (dst_reg + off) is writeable */
 			err = check_mem_access(env, insn->dst_reg, insn->off,
 					       BPF_SIZE(insn->code), BPF_WRITE,

2.9.5

This is a note to let you know that I've just added the patch titled

bpf: reject stores into ctx via st and xadd

to the 4.9-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary

The filename of the patch is:
bpf-reject-stores-into-ctx-via-st-and-xadd.patch
and it can be found in the queue-4.9 subdirectory.

If you, or anyone else, feels it should not be added to the stable tree, please let <[email protected]> know about it.

From: Daniel Borkmann <[email protected]>
Date: Mon, 29 Jan 2018 02:49:01 +0100
Subject: bpf: reject stores into ctx via st and xadd
To: [email protected]
Cc: [email protected], [email protected], Daniel Borkmann <[email protected]>
Message-ID: <bf502ba70d86b457adcbd695412af739a1d036f5.1517190207.git.daniel@iogearbox.net>

From: Daniel Borkmann <[email protected]>

[ upstream commit f37a8cb84cce18762e8f86a70bd6a49a66ab964c ]

Alexei found that verifier does not reject stores into context via BPF_ST instead of BPF_STX. And while looking at it, we also should not allow XADD variant of BPF_STX.

The context rewriter is only assuming either BPF_LDX_MEM- or BPF_STX_MEM-type operations, thus reject anything other than that so that assumptions in the rewriter properly hold. Add test cases as well for BPF selftests.

---
 kernel/bpf/verifier.c |   19 +++++++++++++++++++
 1 file changed, 19 insertions(+)

@@ -702,6 +702,13 @@ static bool is_pointer_value(struct bpf_
 	return __is_pointer_value(env->allow_ptr_leaks, &env->cur_state.regs[regno]);
 }
 
+static bool is_ctx_reg(struct bpf_verifier_env *env, int regno)
+{
+	const struct bpf_reg_state *reg = &env->cur_state.regs[regno];
+
+	return reg->type == PTR_TO_CTX;
+}
+
 static int check_ptr_alignment(struct bpf_verifier_env *env,
 			       struct bpf_reg_state *reg, int off, int size)
 {

@@ -896,6 +903,12 @@ static int check_xadd(struct bpf_verifie
 		return -EACCES;
 	}
 
+	if (is_ctx_reg(env, insn->dst_reg)) {
+		verbose("BPF_XADD stores into R%d context is not allowed\n",
+			insn->dst_reg);
+		return -EACCES;
+	}
+
 	/* check whether atomic_add can read the memory */
 	err = check_mem_access(env, insn->dst_reg, insn->off,
 			       BPF_SIZE(insn->code), BPF_READ, -1);

@@ -3012,6 +3025,12 @@ static int do_check(struct bpf_verifier_
 			if (err)
 				return err;
 
+			if (is_ctx_reg(env, insn->dst_reg)) {
+				verbose("BPF_ST stores into R%d context is not allowed\n",
+					insn->dst_reg);
+				return -EACCES;
+			}
+
 			/* check that memory (dst_reg + off) is writeable */
 			err = check_mem_access(env, insn->dst_reg, insn->off,
 					       BPF_SIZE(insn->code), BPF_WRITE,

Patches currently in stable-queue which might be from [email protected] are

queue-4.9/bpf-avoid-false-sharing-of-map-refcount-with-max_entries.patch queue-4.9/x86-bpf_jit-small-optimization-in-emit_bpf_tail_call.patch queue-4.9/bpf-reject-stores-into-ctx-via-st-and-xadd.patch queue-4.9/bpf-fix-32-bit-divide-by-zero.patch queue-4.9/bpf-fix-bpf_tail_call-x64-jit.patch queue-4.9/bpf-arsh-is-not-supported-in-32-bit-alu-thus-reject-it.patch queue-4.9/bpf-fix-divides-by-zero.patch queue-4.9/bpf-introduce-bpf_jit_always_on-config.patch

Thanks for these, all now queued up.

greg k-h