AT_ENTROPY1 and AT_ENTROPY2 values for include/linux/auxvec.h

Hello glibc and kernel maintainers,

could you please add two AT_ entries to include/linux/auxvec.h in the upstream kernel and the respective elf/elf.h definitions in glibc.

#define AT_ENTROPY1 24 /* kernel entropy in auxiliary vector */
#define AT_ENTROPY2 25 /* second field for kernel entropy */

The accompanying kernel patch follows, it is basically a config option to give userland entropy.
We are using the appended patch for Gentoo Hardened.
You don't have to put it into the upstream linux kernel, it is only there for explaining what i need the AT_ values for. Also the patch contains a subtle "bug" because the repetive call to get_random_int()
seems to yield the same result when done immediately after another:

$$ od -tx8 /proc/self/auxv
0000000 ffffe40000000020 ffffe00000000021
0000020 0383f9ff00000010 0000100000000006
0000040 0000006400000011 8000003400000003
0000060 0000002000000004 0000000a00000005
0000100 b7ef200000000007 0000000000000008
0000120 800013f000000009 000000000000000b
0000140 000000000000000c 000000000000000d
0000160 000000000000000e 0000000000000017
0000200 0c1f4d8100000018 0c1f4d8100000019
0000220 bfa7da8b0000000f 0000000000000000
0000240

However, this is just a proof of concept, all i'm asking for is to get the two numbers into the upstream kernel and glibc auxvec.h to avoid future clashes with the numbers, thx in advance.

-- patch --

diff -Nru linux-2.6.21.5.ORIG/fs/binfmt_elf.c linux-2.6.21.5/fs/binfmt_elf.c
@@ -201,6 +201,13 @@

	NEW_AUX_ENT(AT_GID, tsk->gid);
	NEW_AUX_ENT(AT_EGID, tsk->egid);

NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
+#if defined(CONFIG_AT_ENTROPY1)
+ NEW_AUX_ENT(AT_ENTROPY1, get_random_int()); +#endif
+#if defined(CONFIG_AT_ENTROPY2)
+ NEW_AUX_ENT(AT_ENTROPY2, get_random_int()); +#endif
+

	if (k_platform) {
		NEW_AUX_ENT(AT_PLATFORM,
			    (elf_addr_t)(unsigned long)u_platform);

diff -Nru linux-2.6.21.5.ORIG/include/linux/auxvec.h linux-2.6.21.5/include/linux/auxvec.h
@@ -26,6 +26,9 @@
 

#define AT_SECURE 23 /* secure mode boolean */

-#define AT_VECTOR_SIZE 44 /* Size of auxiliary table. */
+#define AT_ENTROPY1 24 /* kernel entropy in auxiliary vector */
+#define AT_ENTROPY2 25 /* second field for kernel entropy */
+
+#define AT_VECTOR_SIZE 46 /* Size of auxiliary table. */

#endif /* _LINUX_AUXVEC_H */

diff -Nru linux-2.6.21.5.ORIG/security/Kconfig linux-2.6.21.5/security/Kconfig
@@ -4,6 +4,24 @@
 

menu "Security options"

+config AT_ENTROPY1
+ bool "Enable kernel entropy in the auxiliary vector"
+ help

+    This option provides support for kernel-generated entropy handed
+    to processes via the auxiliary vector table in /proc/self/auxv.
+    Most notably userlands with heavy usage of stack smashing protection
+    benefit from kernel entropy because /dev/urandom is less likely to
+    become depleted due to the extensive need for randomized SSP __guard.
+
+config AT_ENTROPY2
+  bool "Enable additional kernel entropy in the auxiliary vector"
+  depends on AT_ENTROPY1
+  help
+    This option provides support for an additional field of kernel-generated
+    entropy for userland processes.  It can be used for an extension to SSP
+    where a second guard value is necessary for XOR encoding control data
+    residing in the stack frame of userland processes.

+
config KEYS
bool "Enable access key retention support" help

-- /patch --

also available at

http://dev.gentoo.org/~pappy/ssp/AT_ENTROPY/linux-2.6.21.5_AT_ENTROPY.patch

Thanks again,

Alex

while I'm not per se against adding such aux vectors, I think it's a bad mistake to make them a config option (and 2 options at that!!)

Stack protector is very widely deployed today (Fedora,RHEL,SLES,OpenSUSE,Gentoo etc), so I can sure see the point of helping it a bit... but I'd like to see a little more data on how the current approach isn't sufficient.

Well our glibc patch (still under development, here comes a snippet) will conditionally check for the auxv and if not "fall back" to normal SSP entropy:

-  /* Set up the stack checker's canary.  */
-  uintptr_t stack_chk_guard = _dl_setup_stack_chk_guard ();
+  /* Set up the stack checker's canary, optional kernel entropy  */
+  uintptr_t stack_chk_guard;

+
+ if (GLRO(dl_entropy) != 0) {
+ stack_chk_guard = GLRO(dl_entropy);
+ }
+ else {
+ stack_chk_guard = _dl_setup_stack_chk_guard ();
+ }

Hence the config option for the kernel- it's philosophy at Gentoo to make choices available to users how they want their systems to behave, even on the expense of added complexity and need to "understand" how things work in the first place.

If you add the aux vector behaviour as a default, there would surely be a backport of the default behaviour to the more "Gentoo flavoured" choice-based system of activating/deactivating it to your own needs.

For your second question, why two config options: entropy is a precious resource and many critical apps rely on it.
Until SSPx (http://dev.gentoo.org/~pappy/sspx/ssxp.pdf) is not available, we don't want to force users to have two vector entries with one containing unneeded (wasted?) entropy. SSP already works good, so if this stuff makes it's way into the kernel or kernel and glibc folks tell me which numbers i can use, i can make my patches and add the logic to Gentoo kernel sources and glibc.

Sincere thanks,

Alex

While I am not opposed to choice, I am opposed to having too finegrained kernel config options. In your view, every single kernel patch would be a config option... I much rather have config options for "important" big changes, not for something this small. Another argument is that this kind of userspace interface is better off being always there or never; making this variable serves no-one.

if you want to mail me at work (you don't), use arjan (at) linux.intel.com
Test the interaction between Linux and your BIOS via http://www.linuxfirmwarekit.org

Hi again,

as said, i'm not voting for getting the specific example patch into kernel source.
All i came for and would like to see is the two numbers into the upstream header file to avoid future namespace clashes (as said in the original mail).

I honestly don't know how to follow the discussion to implement this kind of entropy given via auxv to userland- if you (the kernel devs) think it makes sense the way you want it you as upstream devs can do it all the way you like it- i can always change it with a patch for my own needs in my own sources. I can't say anything at all about all of this because i don't know what is important from your point of view as kernel maintainers and what is not. I understand your concerns and know that we are talking from different perspectives, thus it is probably normal that our interests and technical solutions vary.

However, as said, the only thing important for me is to not use "arbitrary" chosen numbers for the AT_ENTROPY1 and 2 values and later have problems when my "private" numbers suddenly are assigned to a different thing by "official" kernel devs. That would be bad for me and this is what i want to avoid in the first place.

So please tell me what i can do about the requirements and the time and what the usual process is for asking how to get the numbers into the kernel so i can start using them for my own work.

Thanks in advance,

Alex

Hi,

http://dev.gentoo.org/~pappy/kernel/linux-2.6.21.5-get_urandom_long-AT_ENTROPY.patch

this patch adds the function drivers/char/random.c:get_random_long() and adds an AT_ENTROPY field in the auxv without config option (the config option was removed as suggested by Arjan on LKML).

README: get_random_long() and AT_ENTROPY support for auxv
NAME: Alexander Gabert
EMAIL: [email protected]

diff -Nru linux-2.6.21.5.ORIG/drivers/char/random.c linux-2.6.21.5/drivers/char/random.c
@@ -1654,6 +1654,53 @@

}

/*
+ * get_random_long() returns a randomized unsigned long word.
+ * It recycles it's entropy cache for a given time period and
+ * uses half_md4_transform to generate a unique return value.
+ * Every REKEY_INTERVAL the cache is reloaded with fresh
+ * randomization data using get_random_bytes().
+ * This function is not intended for strong cryptographic routines.
+ */
+unsigned long get_random_long(void)
+{
+ /* remember the last time we refreshed the cache with random entropy */
+ static time_t rekey_time;
+
+ time_t t;
+
+ /*

+   * the following data in the buffer is unchanged during REKEY_INTERVAL:
+   * |----|----|KKKK|KKKK|KKKK|KKKK|KKKK|KKKK|----|----|----|----|
+   * ___0____1____2____3____4____5____6____7____8____9___10___11__
+   *
+   * the following data is updated during the first half_md4_transform call
+   * |----|YYYY|----|----|----|----|----|----|ZZZZ|ZZZZ|ZZZZ|ZZZZ|
+   * ___0____1____2____3____4____5____6____7____8____9___10___11__
+   *
+   * the following data is updated during the second half_md4_transform
+   * |XXXX|----|----|----|----|----|----|----|ZZZZ|ZZZZ|ZZZZ|ZZZZ|
+   * ___0____1____2____3____4____5____6____7____8____9___10___11__
+   */

+ static __u32 entropycache[12];
+
+ /* get the current time in seconds */
+ t = get_seconds();
+
+ /* check for REKEY_INTERVAL */
+ if (t && (!rekey_time || ((t - rekey_time) > REKEY_INTERVAL))) {

+    rekey_time = t;
+    /* refresh with random entropy */
+    get_random_bytes(entropycache, sizeof(entropycache));
+  }
+
+  /* transform the buffer to a new state, thus generating new return value */
+  entropycache[1] = half_md4_transform(entropycache+8, entropycache);
+  entropycache[0] = half_md4_transform(entropycache+8, entropycache);

+
+ return *(unsigned long *)entropycache;
+}
+
+/*
* randomize_range() returns a start address such that *
* [...... <range> .....]

diff -Nru linux-2.6.21.5.ORIG/fs/binfmt_elf.c linux-2.6.21.5/fs/binfmt_elf.c
@@ -201,6 +201,7 @@
    NEW_AUX_ENT(AT_GID, tsk->gid);
    NEW_AUX_ENT(AT_EGID, tsk->egid);
     NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
+  NEW_AUX_ENT(AT_ENTROPY, get_random_long());
    if (k_platform) {
        NEW_AUX_ENT(AT_PLATFORM,
                (elf_addr_t)(unsigned long)u_platform);

diff -Nru linux-2.6.21.5.ORIG/include/linux/auxvec.h linux-2.6.21.5/include/linux/auxvec.h
@@ -26,6 +26,8 @@
 

#define AT_SECURE 23 /* secure mode boolean */

-#define AT_VECTOR_SIZE 44 /* Size of auxiliary table. */ +#define AT_ENTROPY 24 /* kernel entropy in auxv */ +
+#define AT_VECTOR_SIZE 45 /* Size of auxiliary table in. */

#endif /* _LINUX_AUXVEC_H */

diff -Nru linux-2.6.21.5.ORIG/include/linux/random.h linux-2.6.21.5/include/linux/random.h
@@ -67,6 +67,9 @@

#endif

unsigned int get_random_int(void);
+
+unsigned long get_random_long(void);
+
unsigned long randomize_range(unsigned long start, unsigned long end, unsigned long len);

u32 random32(void);

your patch has some whitespace damage it seems...
other than that I can agree with the approach.

if you want to mail me at work (you don't), use arjan (at) linux.intel.com
Test the interaction between Linux and your BIOS via http://www.linuxfirmwarekit.org

It's totally whitespace-damaged, and I wonder why the "int" vs "long" is so important. Also, what are the performance implications for execve?

Linus

Hello Alexander

One ought to cc: the /dev/random maintainer.

What's the point of this?

Mathematics is the supreme nostalgia of our time.

Hi Linus,
hi LKML,

i would like to thank LKML and especially Eric (thanks for the per_cpu macro tips and design guidelines!) and the other contributors to this idea.

This time the patch is rather big because it also removes get_random_int() and introduces get_random_long() throughout the kernel.
The function get_random_int was primarily used for setting up the randomization of memory used for process stacks and initializing a stack canary in a tsk. However, using get_random_long works exactly the same way to generate randomization deltas for this purpose and i have applied it to all locations where get_random_int was previously used.

Now to explain the reasoning for replacing get_random_int...
My findings with get_random_int was that when i called it two times very fast after another because of setting up two SSP randomized guard values for the same process, it returned the same, yet randomized, integer number.

I learned that get_random_int uses the current process pid and the current jiffies for feeding the half_md4_transform function that's following deeper in the code. However, when called in a relatively short timeframe (lesser than a jiffie can increase it's count), this information given to the underlying "randomization" generation is not changing, which means when you call get_random_int two or more times in short time, it will use the same data for generating the randomized data, which is the pid of the process and current jiffies. The returned randomized integer numbers are the same because the pid and the jiffies used as the input for the randomization function did not change in such a short time. Sorry that i cannot explain it more detailed or in better english, you should be able to see it from the code looking at get_random_int and how it's working by calling the other function and looking at the input to the function. Also remember that the comment in the function is probably a bit outdated because rekeying is happening not every 1 second any more.

This is the url for the patch:
http://dev.gentoo.org/~pappy/kernel/linux-2.6.21.5-get_random_long-AT_ENTROPY.patch

Thanks for the attention and thanks again to all posters for your help!

-Alex

This is the inlined patch, i changed my vimrc and really hope i'm not making a whitespace error again:

diff -Nru linux-2.6.21.5.ORIG/arch/i386/kernel/process.c linux-2.6.21.5/arch/i386/kernel/process.c
@@ -924,6 +924,6 @@

unsigned long arch_align_stack(unsigned long sp)

{
	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
-		sp -= get_random_int() % 8192;
+		sp -= get_random_long() % 8192;
	return sp & ~0xf;
}

diff -Nru linux-2.6.21.5.ORIG/arch/i386/mm/mmap.c linux-2.6.21.5/arch/i386/mm/mmap.c
@@ -42,7 +42,7 @@

	unsigned long random_factor = 0;

	if (current->flags & PF_RANDOMIZE)

- random_factor = get_random_int() % (1024*1024);
+ random_factor = get_random_long() % (1024*1024);

	if (gap < MIN_GAP)
		gap = MIN_GAP;

diff -Nru linux-2.6.21.5.ORIG/arch/sparc64/kernel/sys_sparc.c linux-2.6.21.5/arch/sparc64/kernel/sys_sparc.c
@@ -362,7 +362,7 @@

	unsigned long random_factor = 0UL;

	if (current->flags & PF_RANDOMIZE) {
-		random_factor = get_random_int();
+		random_factor = get_random_long();
		if (test_thread_flag(TIF_32BIT))
			random_factor &= ((1 * 1024 * 1024) - 1);
		else

diff -Nru linux-2.6.21.5.ORIG/arch/um/kernel/process.c linux-2.6.21.5/arch/um/kernel/process.c
@@ -478,7 +478,7 @@

unsigned long arch_align_stack(unsigned long sp)

{
	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
-		sp -= get_random_int() % 8192;
+		sp -= get_random_long() % 8192;
	return sp & ~0xf;
}

#endif

diff -Nru linux-2.6.21.5.ORIG/arch/x86_64/ia32/mmap32.c linux-2.6.21.5/arch/x86_64/ia32/mmap32.c
@@ -44,7 +44,7 @@

	unsigned long random_factor = 0;

	if (current->flags & PF_RANDOMIZE)

- random_factor = get_random_int() % (1024*1024);
+ random_factor = get_random_long() % (1024*1024);

	if (gap < MIN_GAP)
		gap = MIN_GAP;

diff -Nru linux-2.6.21.5.ORIG/arch/x86_64/kernel/functionlist linux-2.6.21.5/arch/x86_64/kernel/functionlist
@@ -862,7 +862,7 @@

*(.text.__ide_dma_end)
*(.text.hrtimer_get_remaining)
*(.text.get_task_mm)
-*(.text.get_random_int)
+*(.text.get_random_long)
*(.text.free_pipe_info)
*(.text.filemap_write_and_wait_range)
*(.text.exit_thread)

diff -Nru linux-2.6.21.5.ORIG/arch/x86_64/kernel/process.c linux-2.6.21.5/arch/x86_64/kernel/process.c
@@ -885,6 +885,6 @@

unsigned long arch_align_stack(unsigned long sp)

{
	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
-		sp -= get_random_int() % 8192;
+		sp -= get_random_long() % 8192;
	return sp & ~0xf;
}

diff -Nru linux-2.6.21.5.ORIG/arch/x86_64/mm/mmap.c linux-2.6.21.5/arch/x86_64/mm/mmap.c
@@ -18,9 +18,9 @@

	if (current->flags & PF_RANDOMIZE) {
		/* Add 28bit randomness which is about 40bits of address space
		   because mmap base has to be page aligned.

- or ~1/128 of the total user VM
- (total user address space is 47bits) */
- unsigned rnd = get_random_int() & 0xfffffff;
+ or ~1/128 of the total user VM
+ (total user address space is 47bits) */

+		unsigned rnd = get_random_long() & 0xfffffff;
		mm->mmap_base += ((unsigned long)rnd) << PAGE_SHIFT;
	}
	mm->get_unmapped_area = arch_get_unmapped_area;

diff -Nru linux-2.6.21.5.ORIG/drivers/char/random.c linux-2.6.21.5/drivers/char/random.c
@@ -437,6 +437,9 @@

.pool = nonblocking_pool_data
};

+static u32 entropycache_shared[8] __read_mostly;
+static DEFINE_PER_CPU(unsigned long , entropycache_pcpu)[16 / sizeof(unsigned long)]; +
/*
* This function adds a byte into the entropy "pool". It does not
* update the entropy estimate. The caller should call
@@ -1455,6 +1458,10 @@

	keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
	smp_wmb();
	ip_cnt++;

+
+ /* initialize the shared part of the entropy cache */
+ get_random_bytes(entropycache_shared, sizeof(entropycache_shared));
+

	schedule_delayed_work(&rekey_work, REKEY_INTERVAL);
}

@@ -1637,20 +1644,17 @@

/*
- * Get a random word for internal kernel use only. Similar to urandom but
- * with the goal of minimal entropy pool depletion. As a result, the random
- * value is not cryptographically secure but for several uses the cost of
- * depleting entropy is too high
+ * get_random_long() returns a randomized unsigned long word.
+ * The entropy cache is refreshed periodically in rekey_seq_generator.
+ * This function uses half_md4_transform to generate a unique return value
+ * and is not intended for strong cryptographic routines.
*/
-unsigned int get_random_int(void)
+unsigned long get_random_long(void)
{
- /*
- * Use IP's RNG. It suits our purpose perfectly: it re-keys itself
- * every second, from the entropy pool (and thus creates a limited
- * drain on it), and uses halfMD4Transform within the second. We
- * also mix it with jiffies and the PID:
- */
- return secure_ip_id((__force __be32)(current->pid + jiffies));
+ half_md4_transform((u32 *)__get_cpu_var(entropycache_pcpu),
+ entropycache_shared);
+
+ return __get_cpu_var(entropycache_pcpu)[0];
}

/*
@@ -1669,5 +1673,5 @@

	if (end <= start + len)
		return 0;

- return PAGE_ALIGN(get_random_int() % range + start);
+ return PAGE_ALIGN(get_random_long() % range + start);
}

diff -Nru linux-2.6.21.5.ORIG/fs/binfmt_elf.c linux-2.6.21.5/fs/binfmt_elf.c
@@ -200,7 +200,8 @@

	NEW_AUX_ENT(AT_EUID, tsk->euid);
	NEW_AUX_ENT(AT_GID, tsk->gid);
	NEW_AUX_ENT(AT_EGID, tsk->egid);
- 	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
+	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
+	NEW_AUX_ENT(AT_ENTROPY, get_random_long());
	if (k_platform) {
		NEW_AUX_ENT(AT_PLATFORM,
			    (elf_addr_t)(unsigned long)u_platform);

@@ -516,7 +517,7 @@

	if ((current->flags & PF_RANDOMIZE) &&
		!(current->personality & ADDR_NO_RANDOMIZE)) {
-		random_variable = get_random_int() & STACK_RND_MASK;
+		random_variable = get_random_long() & STACK_RND_MASK;
		random_variable <<= PAGE_SHIFT;
	}

#ifdef CONFIG_STACK_GROWSUP

diff -Nru linux-2.6.21.5.ORIG/include/linux/auxvec.h linux-2.6.21.5/include/linux/auxvec.h
@@ -24,8 +24,9 @@

#define AT_HWCAP 16 /* arch dependent hints at CPU capabilities */
#define AT_CLKTCK 17 /* frequency at which times() increments */

-#define AT_SECURE 23 /* secure mode boolean */
+#define AT_SECURE 23 /* secure mode boolean */
+#define AT_ENTROPY 24 /* kernel entropy in auxv */

-#define AT_VECTOR_SIZE  44 /* Size of auxiliary table.  */
+#define AT_VECTOR_SIZE  45 /* Size of auxiliary table.  */

#endif /* _LINUX_AUXVEC_H */

diff -Nru linux-2.6.21.5.ORIG/include/linux/random.h linux-2.6.21.5/include/linux/random.h
@@ -66,7 +66,7 @@

extern const struct file_operations random_fops, urandom_fops; #endif

-unsigned int get_random_int(void);
+unsigned long get_random_long(void);
unsigned long randomize_range(unsigned long start, unsigned long end, unsigned long len);

u32 random32(void);

diff -Nru linux-2.6.21.5.ORIG/kernel/fork.c linux-2.6.21.5/kernel/fork.c
@@ -180,7 +180,7 @@

	setup_thread_stack(tsk, orig);

#ifdef CONFIG_CC_STACKPROTECTOR
- tsk->stack_canary = get_random_int();
+ tsk->stack_canary = get_random_long();
#endif

	/* One for us, one for whoever does the "release_task()" (usually parent) */

diff -Nru linux-2.6.21.5.ORIG/arch/i386/kernel/process.c linux-2.6.21.5/arch/i386/kernel/process.c
@@ -924,6 +924,6 @@
 unsigned long arch_align_stack(unsigned long sp)
 {
 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
-		sp -= get_random_int() % 8192;
+		sp -= get_random_long() % 8192;
 	return sp & ~0xf;
 }

diff -Nru linux-2.6.21.5.ORIG/arch/i386/mm/mmap.c linux-2.6.21.5/arch/i386/mm/mmap.c
@@ -42,7 +42,7 @@
 	unsigned long random_factor = 0;
 
 	if (current->flags & PF_RANDOMIZE)
-		random_factor = get_random_int() % (1024*1024);
+		random_factor = get_random_long() % (1024*1024);
 
 	if (gap < MIN_GAP)
 		gap = MIN_GAP;

diff -Nru linux-2.6.21.5.ORIG/arch/sparc64/kernel/sys_sparc.c linux-2.6.21.5/arch/sparc64/kernel/sys_sparc.c
@@ -362,7 +362,7 @@
 	unsigned long random_factor = 0UL;
 
 	if (current->flags & PF_RANDOMIZE) {
-		random_factor = get_random_int();
+		random_factor = get_random_long();
 		if (test_thread_flag(TIF_32BIT))
 			random_factor &= ((1 * 1024 * 1024) - 1);
 		else

diff -Nru linux-2.6.21.5.ORIG/arch/um/kernel/process.c linux-2.6.21.5/arch/um/kernel/process.c
@@ -478,7 +478,7 @@
 unsigned long arch_align_stack(unsigned long sp)
 {
 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
-		sp -= get_random_int() % 8192;
+		sp -= get_random_long() % 8192;
 	return sp & ~0xf;
 }
 #endif

diff -Nru linux-2.6.21.5.ORIG/arch/x86_64/ia32/mmap32.c linux-2.6.21.5/arch/x86_64/ia32/mmap32.c
@@ -44,7 +44,7 @@
 	unsigned long random_factor = 0;
 
 	if (current->flags & PF_RANDOMIZE)
-		random_factor = get_random_int() % (1024*1024);
+		random_factor = get_random_long() % (1024*1024);
 
 	if (gap < MIN_GAP)
 		gap = MIN_GAP;

diff -Nru linux-2.6.21.5.ORIG/arch/x86_64/kernel/functionlist linux-2.6.21.5/arch/x86_64/kernel/functionlist
@@ -862,7 +862,7 @@
 *(.text.__ide_dma_end)
 *(.text.hrtimer_get_remaining)
 *(.text.get_task_mm)
-*(.text.get_random_int)
+*(.text.get_random_long)
 *(.text.free_pipe_info)
 *(.text.filemap_write_and_wait_range)
 *(.text.exit_thread)

diff -Nru linux-2.6.21.5.ORIG/arch/x86_64/kernel/process.c linux-2.6.21.5/arch/x86_64/kernel/process.c
@@ -885,6 +885,6 @@
 unsigned long arch_align_stack(unsigned long sp)
 {
 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
-		sp -= get_random_int() % 8192;
+		sp -= get_random_long() % 8192;
 	return sp & ~0xf;
 }

diff -Nru linux-2.6.21.5.ORIG/arch/x86_64/mm/mmap.c linux-2.6.21.5/arch/x86_64/mm/mmap.c
@@ -18,9 +18,9 @@
 	if (current->flags & PF_RANDOMIZE) {
 		/* Add 28bit randomness which is about 40bits of address space
 		   because mmap base has to be page aligned.
- 		   or ~1/128 of the total user VM
-	   	   (total user address space is 47bits) */
-		unsigned rnd = get_random_int() & 0xfffffff;
+		   or ~1/128 of the total user VM
+		   (total user address space is 47bits) */
+		unsigned rnd = get_random_long() & 0xfffffff;
 		mm->mmap_base += ((unsigned long)rnd) << PAGE_SHIFT;
 	}
 	mm->get_unmapped_area = arch_get_unmapped_area;

diff -Nru linux-2.6.21.5.ORIG/drivers/char/random.c linux-2.6.21.5/drivers/char/random.c
@@ -437,6 +437,9 @@
 	.pool = nonblocking_pool_data
 };
 
+static u32 entropycache_shared[8] __read_mostly;
+static DEFINE_PER_CPU(unsigned long , entropycache_pcpu)[16 / sizeof(unsigned long)];
+
 /*
  * This function adds a byte into the entropy "pool".  It does not
  * update the entropy estimate.  The caller should call

@@ -1455,6 +1458,10 @@
 	keyptr->count = (ip_cnt & COUNT_MASK) << HASH_BITS;
 	smp_wmb();
 	ip_cnt++;
+
+	/* initialize the shared part of the entropy cache */
+	get_random_bytes(entropycache_shared, sizeof(entropycache_shared));
+
 	schedule_delayed_work(&rekey_work, REKEY_INTERVAL);
 }
 

@@ -1637,20 +1644,17 @@
 
 
 /*
- * Get a random word for internal kernel use only. Similar to urandom but
- * with the goal of minimal entropy pool depletion. As a result, the random
- * value is not cryptographically secure but for several uses the cost of
- * depleting entropy is too high
+ * get_random_long() returns a randomized unsigned long word.
+ * The entropy cache is refreshed periodically in rekey_seq_generator.
+ * This function uses half_md4_transform to generate a unique return value
+ * and is not intended for strong cryptographic routines.
  */
-unsigned int get_random_int(void)
+unsigned long get_random_long(void)
 {
-	/*
-	 * Use IP's RNG. It suits our purpose perfectly: it re-keys itself
-	 * every second, from the entropy pool (and thus creates a limited
-	 * drain on it), and uses halfMD4Transform within the second. We
-	 * also mix it with jiffies and the PID:
-	 */
-	return secure_ip_id((__force __be32)(current->pid + jiffies));
+	half_md4_transform((u32 *)__get_cpu_var(entropycache_pcpu),
+				entropycache_shared);
+
+	return __get_cpu_var(entropycache_pcpu)[0];
 }
 
 /*

@@ -1669,5 +1673,5 @@
 
 	if (end <= start + len)
 		return 0;
-	return PAGE_ALIGN(get_random_int() % range + start);
+	return PAGE_ALIGN(get_random_long() % range + start);
 }

diff -Nru linux-2.6.21.5.ORIG/fs/binfmt_elf.c linux-2.6.21.5/fs/binfmt_elf.c
@@ -200,7 +200,8 @@
 	NEW_AUX_ENT(AT_EUID, tsk->euid);
 	NEW_AUX_ENT(AT_GID, tsk->gid);
 	NEW_AUX_ENT(AT_EGID, tsk->egid);
- 	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
+	NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
+	NEW_AUX_ENT(AT_ENTROPY, get_random_long());
 	if (k_platform) {
 		NEW_AUX_ENT(AT_PLATFORM,
 			    (elf_addr_t)(unsigned long)u_platform);

@@ -516,7 +517,7 @@
 
 	if ((current->flags & PF_RANDOMIZE) &&
 		!(current->personality & ADDR_NO_RANDOMIZE)) {
-		random_variable = get_random_int() & STACK_RND_MASK;
+		random_variable = get_random_long() & STACK_RND_MASK;
 		random_variable <<= PAGE_SHIFT;
 	}
 #ifdef CONFIG_STACK_GROWSUP

diff -Nru linux-2.6.21.5.ORIG/include/linux/auxvec.h linux-2.6.21.5/include/linux/auxvec.h
@@ -24,8 +24,9 @@
 #define AT_HWCAP  16    /* arch dependent hints at CPU capabilities */
 #define AT_CLKTCK 17	/* frequency at which times() increments */
 
-#define AT_SECURE 23   /* secure mode boolean */
+#define AT_SECURE 23	/* secure mode boolean */
+#define AT_ENTROPY 24	/* kernel entropy in auxv */
 
-#define AT_VECTOR_SIZE  44 /* Size of auxiliary table.  */
+#define AT_VECTOR_SIZE  45 /* Size of auxiliary table.  */
 
 #endif /* _LINUX_AUXVEC_H */

diff -Nru linux-2.6.21.5.ORIG/include/linux/random.h linux-2.6.21.5/include/linux/random.h
@@ -66,7 +66,7 @@
 extern const struct file_operations random_fops, urandom_fops;
 #endif
 
-unsigned int get_random_int(void);
+unsigned long get_random_long(void);
 unsigned long randomize_range(unsigned long start, unsigned long end, unsigned long len);
 
 u32 random32(void);

diff -Nru linux-2.6.21.5.ORIG/kernel/fork.c linux-2.6.21.5/kernel/fork.c
@@ -180,7 +180,7 @@
 	setup_thread_stack(tsk, orig);
 
 #ifdef CONFIG_CC_STACKPROTECTOR
-	tsk->stack_canary = get_random_int();
+	tsk->stack_canary = get_random_long();
 #endif
 
 	/* One for us, one for whoever does the "release_task()" (usually parent) */

Stop right there. You still haven't answered my original question. What is the point of this exercise in the first place, please?

Am I right in thinking you have three unrelated patches here?

- something to do with aux vector headers
- something to do with get_random_int repeating itself
- sweeping change of get_random_int to get_random_long for no obvious reason

These should be three completely separate patches.

Send me a patch that fixes the above and nothing else, please.

Then we can talk about another patch to change things to get_random_long.

Mathematics is the supreme nostalgia of our time.

the primary goal is to pass a random value to userspace at process start; this to save glibc from having to open /dev/urandom on ever program start (which it does now for all apps compiled with -fstack-protector, which in various distros is "everything").

and this is because Alexander wants 2 and not 1 random int to be passed for his own glibc proposed change (combined with get_random_int() being designed for only 4 bytes per process ;-)

if you want to mail me at work (you don't), use arjan (at) linux.intel.com
Test the interaction between Linux and your BIOS via http://www.linuxfirmwarekit.org

Interesting.

What are our requirements here? Defending against local attackers who can build exploits on the fly probably means something stronger than get_random_int.

Sure. Still, separate patches.

Mathematics is the supreme nostalgia of our time.

There are 2 ways to compile -fstack-protector supporting glibc actually, only one opens /dev/urandom on every program initialization, the other computes the stack guard from some bits of the stack address (so indirectly depends on get_random_int() in stack randomization). Nevertheless, having one random long (32-bit for 32-bit arches, 64-bit otherwise) in aux vector would be useful.

Jakub

Hi Matt,
sorry for not answering your questions in the first place, i hope this did not mean to make a bad impression

I don't think so but you may be right nonetheless if my opinion.

Adding the new field

Found while adding the new field and testing it.

It is needed for properly initializing a SSP guard which is (afaik) a long value.

Probably ... but bear in mind that the goal is still the same: allowing glibc to use SSP with /proc/self/auxv instead of fopen(/dev/urandom) as it is now.
Effectively saving three syscalls (open,read,close) and making life easier for glibc because randomization "generated" in the kernel does not deplete /dev/urandom too much for high coverage SSP userlands (i.e. Gentoo Hardened).

I can imagine that Redhat would do the same with the SSP implementation in glibc, i think if this patch moves into kernel, they will bring out a glibc patch that is checking for AT_ENTROPY and using the opening of /dev/urandom for retrieving randomized data as a fallback for machines where such a kernel is not available. This is a win-win situation for both sides- the kernel wins because the pressure on /dev/urandom is released a bit (applicable to SSP environments) and the glibc wins because it has a reliable, fast, cheap and easy to use source for randomization.

Thank you,

Alex

The point is: the way we do development here is to break things down into a series of simple, obvious pieces and send them individually. Each patch should build and run when applied on top of the earlier patches. Then each part can be reviewed, tested, and applied on its own. It also makes figuring out what happened in the future and tracking down bugs much easier.

So send three patches in this order:

- fix get_random_int
- rename it
- add auxv bits

Also, you might want to drop Linus from the cc: list unless he's chimed in on this topic already. The likely path here is for the first two patches to go through me and the last to go through Arjan, then on to Andrew Morton for his -mm tree, and then to Linus after a bit.

Mathematics is the supreme nostalgia of our time.