Skip to contentSkip to conversation

Handle memmap and mem kernel options in boot stage kaslr

PATCH In review Baoquan He 4 participants 12 messages May 13, 2017 – May 24, 2017

Subsystems (2)
lore
Share & tools
lore

Baoquan He

author
#0 lore raw

v5 cover Handle memmap and mem kernel options in boot stage kaslr

People reported kernel panic occurs during system boots up with mem boot option. After checking code, several problems are found about memmap= and mem= in boot stage kaslr.

*) In commit f28442497b5c ("x86/boot: Fix KASLR and memmap= collision"), only one memmap
entry is considered and only the last one if multiple memmap entries are specified.

*) mem= and memmap=nn[KMG] are not considered yet. They are used to limit max address
of system. Kernel can't be randomized to be above the limit.

*) kernel-parameters.txt doesn't tell the updated behaviour of memmap=.

This patchset tries to solve above issues, and it sits on top of tip:x86/boot branch.

Changelog
v4->v5:
1. Change patch log according to Thomas's comment.

2. Put "Fall through" to the right place in parse_memmap() according to Kees's suggestion.

v3->v4:
1. Code improved patch 1/3 according to Kees's suggestion.

2. Add 'Fall through' in switch case of parse_memmap() which is suggestd by Kees.

v2->v3:
No functionality change in this round.
1. Use local static variable insted of global variable mem_avoid_memmap_index in patch 1/3.

2. Fix a typo in patch 3/3.

v1->v2:
1. The original patch 1/4 has been put in tip:x86/boot and no update, so it's not included in this post.

2. Use patch log Ingo reorganized.

3. lib/ctype.c and lib/cmdline.c are needed for kaslr.c, while those EXPORT_SYMBOL(x) contained caused failure of build on 32-bit allmodconfig:
......
ld: -r and -shared may not be used together
scripts/Makefile.build:294: recipe for target 'arch/x86/boot/compressed/kaslr.o' failed
......
Disabling the symbol exporting removes the build failure.

4. Use dynamic allocation to allocate memory to contain copied kernel cmdline buffer, it's implemented in include/linux/decompress/mm.h.

Baoquan He (3):
KASLR: Parse all memmap entries in command line
KASLR: Handle memory limit specified by memmap and mem option Documentation/kernel-parameters.txt: Update 'memmap=' option
description

 Documentation/admin-guide/kernel-parameters.txt |   9 ++
 arch/x86/boot/compressed/cmdline.c              |   2 +-
 arch/x86/boot/compressed/kaslr.c                | 190 ++++++++++++++++--------
 arch/x86/boot/string.c                          |   8 +
 4 files changed, 143 insertions(+), 66 deletions(-)
Signature · 1 line
2.5.5
author
#1 lore raw

PATCH v5 1/3 KASLR: Parse all memmap entries in command line

In commit:

f28442497b5c ("x86/boot: Fix KASLR and memmap= collision")

... the memmap= option is parsed so that KASLR can avoid those reserved regions. It uses cmdline_find_option() to get the value if memmap= is specified, however the problem is that cmdline_find_option() can only find the last entry if multiple memmap entries are provided. This is not correct.

Address this by checking each command line token for a "memmap=" match and parse each instance instead of using cmdline_find_option().

---
 arch/x86/boot/compressed/cmdline.c |   2 +-
 arch/x86/boot/compressed/kaslr.c   | 136 ++++++++++++++++++++++---------------
 arch/x86/boot/string.c             |   8 +++
 3 files changed, 91 insertions(+), 55 deletions(-)
arch/x86/boot/compressed/cmdline.c +1 -1
index 73ccf63..9dc1ce6 100644
@@ -13,7 +13,7 @@ static inline char rdfs8(addr_t addr)
 	return *((char *)(fs + addr));
 }
 #include "../cmdline.c"
-static unsigned long get_cmd_line_ptr(void)
+unsigned long get_cmd_line_ptr(void)
 {
 	unsigned long cmd_line_ptr = boot_params->hdr.cmd_line_ptr;
 
arch/x86/boot/compressed/kaslr.c +82 -54

200 lines not on this page · raw message

arch/x86/boot/string.c +8 -0
index 5457b02..630e366 100644
@@ -122,6 +122,14 @@ unsigned long long simple_strtoull(const char *cp, char **endp, unsigned int bas
 	return result;
 }
 
+long simple_strtol(const char *cp, char **endp, unsigned int base)
+{
+	if (*cp == '-')
+		return -simple_strtoull(cp + 1, endp, base);
+
+	return simple_strtoull(cp, endp, base);
+}
+
 /**
  * strlen - Find the length of a string
  * @s: The string to be sized
Signature · 1 line
2.5.5
#2 lore raw

[tip:x86/boot] x86/KASLR: Parse all 'memmap=' boot option entries

x86/KASLR: Parse all 'memmap=' boot option entries
Commit-ID:  d52e7d5a952c5e35783f96e8c5b7fcffbb0d7c60
Gitweb:     http://git.kernel.org/tip/d52e7d5a952c5e35783f96e8c5b7fcffbb0d7c60
Author:     Baoquan He <[email protected]>
AuthorDate: Sat, 13 May 2017 13:46:28 +0800
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 24 May 2017 09:50:27 +0200

x86/KASLR: Parse all 'memmap=' boot option entries

In commit:

f28442497b5c ("x86/boot: Fix KASLR and memmap= collision")

... the memmap= option is parsed so that KASLR can avoid those reserved regions. It uses cmdline_find_option() to get the value if memmap= is specified, however the problem is that cmdline_find_option() can only find the last entry if multiple memmap entries are provided. This is not correct.

Address this by checking each command line token for a "memmap=" match and parse each instance instead of using cmdline_find_option().

---
 arch/x86/boot/compressed/cmdline.c |   2 +-
 arch/x86/boot/compressed/kaslr.c   | 136 ++++++++++++++++++++++---------------
 arch/x86/boot/string.c             |   8 +++
 3 files changed, 91 insertions(+), 55 deletions(-)
arch/x86/boot/compressed/cmdline.c +1 -1
index 73ccf63..9dc1ce6 100644
@@ -13,7 +13,7 @@ static inline char rdfs8(addr_t addr)
 	return *((char *)(fs + addr));
 }
 #include "../cmdline.c"
-static unsigned long get_cmd_line_ptr(void)
+unsigned long get_cmd_line_ptr(void)
 {
 	unsigned long cmd_line_ptr = boot_params->hdr.cmd_line_ptr;
 
arch/x86/boot/compressed/kaslr.c +82 -54

200 lines not on this page · raw message

arch/x86/boot/string.c +8 -0
index 5457b02..630e366 100644
@@ -122,6 +122,14 @@ unsigned long long simple_strtoull(const char *cp, char **endp, unsigned int bas
 	return result;
 }
 
+long simple_strtol(const char *cp, char **endp, unsigned int base)
+{
+	if (*cp == '-')
+		return -simple_strtoull(cp + 1, endp, base);
+
+	return simple_strtoull(cp, endp, base);
+}
+
 /**
  * strlen - Find the length of a string
  * @s: The string to be sized
author
#3 lore raw

PATCH v5 2/3 KASLR: Handle memory limit specified by memmap and mem option

In reply to Baoquan He

Option mem= will limit the max address a system can use and any memory region above the limit will be removed.

Furthermore, memmap=nn[KMG] which has no offset specified has the same behaviour as mem=.

KASLR needs to consider this when choosing the random position for decompressing the kernel. Do it now.

---
 arch/x86/boot/compressed/kaslr.c | 68 +++++++++++++++++++++++++++++-----------
 1 file changed, 50 insertions(+), 18 deletions(-)
arch/x86/boot/compressed/kaslr.c +50 -18
index 106e13b..e0eba12 100644
@@ -88,6 +88,10 @@ struct mem_vector {
 static bool memmap_too_large;
 
 
+/* Store memory limit specified by "mem=nn[KMG]" or "memmap=nn[KMG]" */
+unsigned long long mem_limit = ULLONG_MAX;
+
+
 enum mem_avoid_index {
 	MEM_AVOID_ZO_RANGE = 0,
 	MEM_AVOID_INITRD,
@@ -138,16 +142,23 @@ parse_memmap(char *p, unsigned long long *start, unsigned long long *size)
 		return -EINVAL;
 
 	switch (*p) {
-	case '@':
-		/* Skip this region, usable */
-		*start = 0;
-		*size = 0;
-		return 0;
 	case '#':
 	case '$':
 	case '!':
 		*start = memparse(p + 1, &p);
 		return 0;
+	case '@':
+		/* memmap=nn@ss specifies usable region, should be skipped */
+		*size = 0;
+		/* Fall through */
+	default:
+		/*
+		 * If w/o offset, only size specified, memmap=nn[KMG] has the
+		 * same behaviour as mem=nn[KMG]. It limits the max address
+		 * system can use. Region above the limit should be avoided.
+		 */
+		*start = 0;
+		return 0;
 	}
 
 	return -EINVAL;
@@ -173,9 +184,14 @@ static void mem_avoid_memmap(char *str)
 		if (rc < 0)
 			break;
 		str = k;
-		/* A usable region that should not be skipped */
-		if (size == 0)
+
+		if (start == 0) {
+			/* Store the specified memory limit if size > 0 */
+			if (size > 0)
+				mem_limit = size;
+
 			continue;
+		}
 
 		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].start = start;
 		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].size = size;
@@ -187,19 +203,15 @@ static void mem_avoid_memmap(char *str)
 		memmap_too_large = true;
 }
 
-
-/*
- * handle_mem_memmap will also cover 'mem=' issue in next patch. Will remove
- * this note later.
- */
 static int handle_mem_memmap(void)
 {
 	char *args = (char *)get_cmd_line_ptr();
 	size_t len = strlen((char *)args);
 	char *tmp_cmdline;
 	char *param, *val;
+	u64 mem_size;
 
-	if (!strstr(args, "memmap="))
+	if (!strstr(args, "memmap=") && !strstr(args, "mem="))
 		return 0;
 
 	tmp_cmdline = malloc(len + 1);
@@ -222,8 +234,20 @@ static int handle_mem_memmap(void)
 			return -1;
 		}
 
-		if (!strcmp(param, "memmap"))
+		if (!strcmp(param, "memmap")) {
 			mem_avoid_memmap(val);
+		} else if (!strcmp(param, "mem")) {
+			char *p = val;
+
+			if (!strcmp(p, "nopentium"))
+				continue;
+			mem_size = memparse(p, &p);
+			if (mem_size == 0) {
+				free(tmp_cmdline);
+				return -EINVAL;
+			}
+			mem_limit = mem_size;
+		}
 	}
 
 	free(tmp_cmdline);
@@ -460,7 +484,8 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 {
 	struct mem_vector region, overlap;
 	struct slot_area slot_area;
-	unsigned long start_orig;
+	unsigned long start_orig, end;
+	struct boot_e820_entry cur_entry;
 
 	/* Skip non-RAM entries. */
 	if (entry->type != E820_TYPE_RAM)
@@ -474,8 +499,15 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 	if (entry->addr + entry->size < minimum)
 		return;
 
-	region.start = entry->addr;
-	region.size = entry->size;
+	/* Ignore entries above memory limit */
+	end = min(entry->size + entry->addr, mem_limit);
+	if (entry->addr >= end)
+		return;
+	cur_entry.addr = entry->addr;
+	cur_entry.size = end - entry->addr;
+
+	region.start = cur_entry.addr;
+	region.size = cur_entry.size;
 
 	/* Give up if slot area array is full. */
 	while (slot_area_index < MAX_SLOT_AREA) {
@@ -489,7 +521,7 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 		region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN);
 
 		/* Did we raise the address above this e820 region? */
-		if (region.start > entry->addr + entry->size)
+		if (region.start > cur_entry.addr + cur_entry.size)
 			return;
 
 		/* Reduce size by any delta from the original address. */
Signature · 1 line
2.5.5
#4 lore raw

Re: [PATCH v5 2/3] KASLR: Handle memory limit specified by memmap and mem option

Hi Baoquan,

Quoted 60 lines · Baoquan He

Option mem= will limit the max address a system can use and any memory region above the limit will be removed.

Furthermore, memmap=nn[KMG] which has no offset specified has the same behaviour as mem=.

KASLR needs to consider this when choosing the random position for decompressing the kernel. Do it now.

---
 arch/x86/boot/compressed/kaslr.c | 68 +++++++++++++++++++++++++++++-----------
 1 file changed, 50 insertions(+), 18 deletions(-)
arch/x86/boot/compressed/kaslr.c +16 -5
index 106e13b..e0eba12 100644
@@ -88,6 +88,10 @@ struct mem_vector {
 static bool memmap_too_large;
 
 
+/* Store memory limit specified by "mem=nn[KMG]" or "memmap=nn[KMG]" */
+unsigned long long mem_limit = ULLONG_MAX;
+
+
 enum mem_avoid_index {
 	MEM_AVOID_ZO_RANGE = 0,
 	MEM_AVOID_INITRD,
@@ -138,16 +142,23 @@ parse_memmap(char *p, unsigned long long *start, unsigned long long *size)
 		return -EINVAL;
 
 	switch (*p) {
-	case '@':
-		/* Skip this region, usable */
-		*start = 0;
-		*size = 0;
-		return 0;
 	case '#':
 	case '$':
 	case '!':
 		*start = memparse(p + 1, &p);
 		return 0;
+	case '@':
+		/* memmap=nn@ss specifies usable region, should be skipped */
+		*size = 0;
+		/* Fall through */
+	default:
+		/*
+		 * If w/o offset, only size specified, memmap=nn[KMG] has the
+		 * same behaviour as mem=nn[KMG]. It limits the max address
+		 * system can use. Region above the limit should be avoided.
+		 */
+		*start = 0;
+		return 0;
 	}
 
 	return -EINVAL;

arch/x86/boot/compressed/kaslr.c static void mem_avoid_memmap(char *str) · line 191

Quoted 10 lines
arch/x86/boot/compressed/kaslr.c +5 -2
@@ -173,9 +184,14 @@ static void mem_avoid_memmap(char *str)
 		if (rc < 0)
 			break;
 		str = k;
-		/* A usable region that should not be skipped */
-		if (size == 0)
+
+		if (start == 0) {
+			/* Store the specified memory limit if size > 0 */
+			if (size > 0)
+				mem_limit = size;

Baoquan,

I am not sure about setting the value of mem_limit to mem_size directly.

If the command line has both the "memmap" and "mem", such as
... mem=2G memmap=4G ...

...in that code, the mem_limit may be 4G not 2G.

In my opinion, How about following:

mem_limit = mem_limit > mem_size ? mem_size : mem_limit;

Quoted 36 lines
arch/x86/boot/compressed/kaslr.c +6 -7
+
 			continue;
+		}
 
 		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].start = start;
 		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].size = size;
@@ -187,19 +203,15 @@ static void mem_avoid_memmap(char *str)
 		memmap_too_large = true;
 }
 
-
-/*
- * handle_mem_memmap will also cover 'mem=' issue in next patch. Will remove
- * this note later.
- */
 static int handle_mem_memmap(void)
 {
 	char *args = (char *)get_cmd_line_ptr();
 	size_t len = strlen((char *)args);
 	char *tmp_cmdline;
 	char *param, *val;
+	u64 mem_size;
 
-	if (!strstr(args, "memmap="))
+	if (!strstr(args, "memmap=") && !strstr(args, "mem="))
 		return 0;
 
 	tmp_cmdline = malloc(len + 1);
@@ -222,8 +234,20 @@ static int handle_mem_memmap(void)
 			return -1;
 		}
 
-		if (!strcmp(param, "memmap"))
+		if (!strcmp(param, "memmap")) {
 			mem_avoid_memmap(val);
+		} else if (!strcmp(param, "mem")) {

arch/x86/boot/compressed/kaslr.c static int handle_mem_memmap(void) · line 249

Quoted 10 lines
arch/x86/boot/compressed/kaslr.c +10 -0
+			char *p = val;
+
+			if (!strcmp(p, "nopentium"))
+				continue;
+			mem_size = memparse(p, &p);
+			if (mem_size == 0) {
+				free(tmp_cmdline);
+				return -EINVAL;
+			}
+			mem_limit = mem_size;

The same as above.

Thanks,
Liyang.

Quoted 42 lines
arch/x86/boot/compressed/kaslr.c +13 -4
+		}
 	}
 
 	free(tmp_cmdline);
@@ -460,7 +484,8 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 {
 	struct mem_vector region, overlap;
 	struct slot_area slot_area;
-	unsigned long start_orig;
+	unsigned long start_orig, end;
+	struct boot_e820_entry cur_entry;
 
 	/* Skip non-RAM entries. */
 	if (entry->type != E820_TYPE_RAM)
@@ -474,8 +499,15 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 	if (entry->addr + entry->size < minimum)
 		return;
 
-	region.start = entry->addr;
-	region.size = entry->size;
+	/* Ignore entries above memory limit */
+	end = min(entry->size + entry->addr, mem_limit);
+	if (entry->addr >= end)
+		return;
+	cur_entry.addr = entry->addr;
+	cur_entry.size = end - entry->addr;
+
+	region.start = cur_entry.addr;
+	region.size = cur_entry.size;
 
 	/* Give up if slot area array is full. */
 	while (slot_area_index < MAX_SLOT_AREA) {
@@ -489,7 +521,7 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 		region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN);
 
 		/* Did we raise the address above this e820 region? */
-		if (region.start > entry->addr + entry->size)
+		if (region.start > cur_entry.addr + cur_entry.size)
 			return;
 
 		/* Reduce size by any delta from the original address. */
author
#5 lore raw

Re: [PATCH v5 2/3] KASLR: Handle memory limit specified by memmap and mem option

Quoted 76 lines · Dou Liyang

Hi Baoquan,

Quoted 60 lines · Baoquan He

Option mem= will limit the max address a system can use and any memory region above the limit will be removed.

Furthermore, memmap=nn[KMG] which has no offset specified has the same behaviour as mem=.

KASLR needs to consider this when choosing the random position for decompressing the kernel. Do it now.

---
 arch/x86/boot/compressed/kaslr.c | 68 +++++++++++++++++++++++++++++-----------
 1 file changed, 50 insertions(+), 18 deletions(-)
arch/x86/boot/compressed/kaslr.c +16 -5
index 106e13b..e0eba12 100644
@@ -88,6 +88,10 @@ struct mem_vector {
 static bool memmap_too_large;
 
 
+/* Store memory limit specified by "mem=nn[KMG]" or "memmap=nn[KMG]" */
+unsigned long long mem_limit = ULLONG_MAX;
+
+
 enum mem_avoid_index {
 	MEM_AVOID_ZO_RANGE = 0,
 	MEM_AVOID_INITRD,
@@ -138,16 +142,23 @@ parse_memmap(char *p, unsigned long long *start, unsigned long long *size)
 		return -EINVAL;
 
 	switch (*p) {
-	case '@':
-		/* Skip this region, usable */
-		*start = 0;
-		*size = 0;
-		return 0;
 	case '#':
 	case '$':
 	case '!':
 		*start = memparse(p + 1, &p);
 		return 0;
+	case '@':
+		/* memmap=nn@ss specifies usable region, should be skipped */
+		*size = 0;
+		/* Fall through */
+	default:
+		/*
+		 * If w/o offset, only size specified, memmap=nn[KMG] has the
+		 * same behaviour as mem=nn[KMG]. It limits the max address
+		 * system can use. Region above the limit should be avoided.
+		 */
+		*start = 0;
+		return 0;
 	}
 
 	return -EINVAL;

arch/x86/boot/compressed/kaslr.c static void mem_avoid_memmap(char *str) · line 191

Quoted 10 lines
arch/x86/boot/compressed/kaslr.c +5 -2
@@ -173,9 +184,14 @@ static void mem_avoid_memmap(char *str)
 		if (rc < 0)
 			break;
 		str = k;
-		/* A usable region that should not be skipped */
-		if (size == 0)
+
+		if (start == 0) {
+			/* Store the specified memory limit if size > 0 */
+			if (size > 0)
+				mem_limit = size;

Baoquan,

Quoted 6 lines

I am not sure about setting the value of mem_limit to mem_size directly.

If the command line has both the "memmap" and "mem", such as
... mem=2G memmap=4G ...

...in that code, the mem_limit may be 4G not 2G.

No, could you tell why you want to add both "memmap=nnKMG" and "mem=" at the same time? As you sid, what if I add "mem=4G mem=2G mem=1G"?

Quoted 101 lines

In my opinion, How about following:

mem_limit = mem_limit > mem_size ? mem_size : mem_limit;

Quoted 36 lines
arch/x86/boot/compressed/kaslr.c +6 -7
+
 			continue;
+		}
 
 		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].start = start;
 		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].size = size;
@@ -187,19 +203,15 @@ static void mem_avoid_memmap(char *str)
 		memmap_too_large = true;
 }
 
-
-/*
- * handle_mem_memmap will also cover 'mem=' issue in next patch. Will remove
- * this note later.
- */
 static int handle_mem_memmap(void)
 {
 	char *args = (char *)get_cmd_line_ptr();
 	size_t len = strlen((char *)args);
 	char *tmp_cmdline;
 	char *param, *val;
+	u64 mem_size;
 
-	if (!strstr(args, "memmap="))
+	if (!strstr(args, "memmap=") && !strstr(args, "mem="))
 		return 0;
 
 	tmp_cmdline = malloc(len + 1);
@@ -222,8 +234,20 @@ static int handle_mem_memmap(void)
 			return -1;
 		}
 
-		if (!strcmp(param, "memmap"))
+		if (!strcmp(param, "memmap")) {
 			mem_avoid_memmap(val);
+		} else if (!strcmp(param, "mem")) {

arch/x86/boot/compressed/kaslr.c static int handle_mem_memmap(void) · line 249

Quoted 10 lines
arch/x86/boot/compressed/kaslr.c +10 -0
+			char *p = val;
+
+			if (!strcmp(p, "nopentium"))
+				continue;
+			mem_size = memparse(p, &p);
+			if (mem_size == 0) {
+				free(tmp_cmdline);
+				return -EINVAL;
+			}
+			mem_limit = mem_size;

The same as above.

Thanks,
Liyang.

Quoted 42 lines
arch/x86/boot/compressed/kaslr.c +13 -4
+		}
 	}
 
 	free(tmp_cmdline);
@@ -460,7 +484,8 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 {
 	struct mem_vector region, overlap;
 	struct slot_area slot_area;
-	unsigned long start_orig;
+	unsigned long start_orig, end;
+	struct boot_e820_entry cur_entry;
 
 	/* Skip non-RAM entries. */
 	if (entry->type != E820_TYPE_RAM)
@@ -474,8 +499,15 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 	if (entry->addr + entry->size < minimum)
 		return;
 
-	region.start = entry->addr;
-	region.size = entry->size;
+	/* Ignore entries above memory limit */
+	end = min(entry->size + entry->addr, mem_limit);
+	if (entry->addr >= end)
+		return;
+	cur_entry.addr = entry->addr;
+	cur_entry.size = end - entry->addr;
+
+	region.start = cur_entry.addr;
+	region.size = cur_entry.size;
 
 	/* Give up if slot area array is full. */
 	while (slot_area_index < MAX_SLOT_AREA) {
@@ -489,7 +521,7 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 		region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN);
 
 		/* Did we raise the address above this e820 region? */
-		if (region.start > entry->addr + entry->size)
+		if (region.start > cur_entry.addr + cur_entry.size)
 			return;
 
 		/* Reduce size by any delta from the original address. */
#6 lore raw

Re: [PATCH v5 2/3] KASLR: Handle memory limit specified by memmap and mem option

Quoted 79 lines · Baoquan He
Quoted 78 lines · Dou Liyang

Hi Baoquan,

Quoted 60 lines · Baoquan He

Option mem= will limit the max address a system can use and any memory region above the limit will be removed.

Furthermore, memmap=nn[KMG] which has no offset specified has the same behaviour as mem=.

KASLR needs to consider this when choosing the random position for decompressing the kernel. Do it now.

---
 arch/x86/boot/compressed/kaslr.c | 68 +++++++++++++++++++++++++++++-----------
 1 file changed, 50 insertions(+), 18 deletions(-)
arch/x86/boot/compressed/kaslr.c +16 -5
index 106e13b..e0eba12 100644
@@ -88,6 +88,10 @@ struct mem_vector {
 static bool memmap_too_large;
 
 
+/* Store memory limit specified by "mem=nn[KMG]" or "memmap=nn[KMG]" */
+unsigned long long mem_limit = ULLONG_MAX;
+
+
 enum mem_avoid_index {
 	MEM_AVOID_ZO_RANGE = 0,
 	MEM_AVOID_INITRD,
@@ -138,16 +142,23 @@ parse_memmap(char *p, unsigned long long *start, unsigned long long *size)
 		return -EINVAL;
 
 	switch (*p) {
-	case '@':
-		/* Skip this region, usable */
-		*start = 0;
-		*size = 0;
-		return 0;
 	case '#':
 	case '$':
 	case '!':
 		*start = memparse(p + 1, &p);
 		return 0;
+	case '@':
+		/* memmap=nn@ss specifies usable region, should be skipped */
+		*size = 0;
+		/* Fall through */
+	default:
+		/*
+		 * If w/o offset, only size specified, memmap=nn[KMG] has the
+		 * same behaviour as mem=nn[KMG]. It limits the max address
+		 * system can use. Region above the limit should be avoided.
+		 */
+		*start = 0;
+		return 0;
 	}
 
 	return -EINVAL;

arch/x86/boot/compressed/kaslr.c static void mem_avoid_memmap(char *str) · line 191

Quoted 10 lines
arch/x86/boot/compressed/kaslr.c +5 -2
@@ -173,9 +184,14 @@ static void mem_avoid_memmap(char *str)
 		if (rc < 0)
 			break;
 		str = k;
-		/* A usable region that should not be skipped */
-		if (size == 0)
+
+		if (start == 0) {
+			/* Store the specified memory limit if size > 0 */
+			if (size > 0)
+				mem_limit = size;

Baoquan,

I am not sure about setting the value of mem_limit to mem_size directly.

Quoted 7 lines
Quoted 4 lines

If the command line has both the "memmap" and "mem", such as
... mem=2G memmap=4G ...

...in that code, the mem_limit may be 4G not 2G.

No, could you tell why you want to add both "memmap=nnKMG" and "mem=" at the same time? As you sid, what if I add "mem=4G mem=2G mem=1G"?

Just for testing :)

Ok, thanks, I see. We should be responsible for our command line. don't need to consider with these situations in kernel.

Thanks,
Liyang.

Quoted 105 lines
Quoted 102 lines

In my opinion, How about following:

mem_limit = mem_limit > mem_size ? mem_size : mem_limit;

Quoted 36 lines
arch/x86/boot/compressed/kaslr.c +6 -7
+
 			continue;
+		}
 
 		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].start = start;
 		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].size = size;
@@ -187,19 +203,15 @@ static void mem_avoid_memmap(char *str)
 		memmap_too_large = true;
 }
 
-
-/*
- * handle_mem_memmap will also cover 'mem=' issue in next patch. Will remove
- * this note later.
- */
 static int handle_mem_memmap(void)
 {
 	char *args = (char *)get_cmd_line_ptr();
 	size_t len = strlen((char *)args);
 	char *tmp_cmdline;
 	char *param, *val;
+	u64 mem_size;
 
-	if (!strstr(args, "memmap="))
+	if (!strstr(args, "memmap=") && !strstr(args, "mem="))
 		return 0;
 
 	tmp_cmdline = malloc(len + 1);
@@ -222,8 +234,20 @@ static int handle_mem_memmap(void)
 			return -1;
 		}
 
-		if (!strcmp(param, "memmap"))
+		if (!strcmp(param, "memmap")) {
 			mem_avoid_memmap(val);
+		} else if (!strcmp(param, "mem")) {

arch/x86/boot/compressed/kaslr.c static int handle_mem_memmap(void) · line 249

Quoted 10 lines
arch/x86/boot/compressed/kaslr.c +10 -0
+			char *p = val;
+
+			if (!strcmp(p, "nopentium"))
+				continue;
+			mem_size = memparse(p, &p);
+			if (mem_size == 0) {
+				free(tmp_cmdline);
+				return -EINVAL;
+			}
+			mem_limit = mem_size;

The same as above.

Thanks,
Liyang.

Quoted 42 lines
arch/x86/boot/compressed/kaslr.c +13 -4
+		}
 	}
 
 	free(tmp_cmdline);
@@ -460,7 +484,8 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 {
 	struct mem_vector region, overlap;
 	struct slot_area slot_area;
-	unsigned long start_orig;
+	unsigned long start_orig, end;
+	struct boot_e820_entry cur_entry;
 
 	/* Skip non-RAM entries. */
 	if (entry->type != E820_TYPE_RAM)
@@ -474,8 +499,15 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 	if (entry->addr + entry->size < minimum)
 		return;
 
-	region.start = entry->addr;
-	region.size = entry->size;
+	/* Ignore entries above memory limit */
+	end = min(entry->size + entry->addr, mem_limit);
+	if (entry->addr >= end)
+		return;
+	cur_entry.addr = entry->addr;
+	cur_entry.size = end - entry->addr;
+
+	region.start = cur_entry.addr;
+	region.size = cur_entry.size;
 
 	/* Give up if slot area array is full. */
 	while (slot_area_index < MAX_SLOT_AREA) {
@@ -489,7 +521,7 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 		region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN);
 
 		/* Did we raise the address above this e820 region? */
-		if (region.start > entry->addr + entry->size)
+		if (region.start > cur_entry.addr + cur_entry.size)
 			return;
 
 		/* Reduce size by any delta from the original address. */
author
#7 lore raw

Re: [PATCH v5 2/3] KASLR: Handle memory limit specified by memmap and mem option

Quoted 44 lines · Dou Liyang
Quoted 41 lines · Baoquan He
Quoted 39 lines · Dou Liyang

Hi Baoquan,

Quoted 17 lines · Baoquan He

Option mem= will limit the max address a system can use and any memory region above the limit will be removed.

Furthermore, memmap=nn[KMG] which has no offset specified has the same behaviour as mem=.

KASLR needs to consider this when choosing the random position for decompressing the kernel. Do it now.

---
 arch/x86/boot/compressed/kaslr.c | 68 +++++++++++++++++++++++++++++-----------
 1 file changed, 50 insertions(+), 18 deletions(-)
arch/x86/boot/compressed/kaslr.c +0 -0
 	return -EINVAL;

arch/x86/boot/compressed/kaslr.c static void mem_avoid_memmap(char *str) · line 191

Quoted 10 lines
arch/x86/boot/compressed/kaslr.c +5 -2
@@ -173,9 +184,14 @@ static void mem_avoid_memmap(char *str)
 		if (rc < 0)
 			break;
 		str = k;
-		/* A usable region that should not be skipped */
-		if (size == 0)
+
+		if (start == 0) {
+			/* Store the specified memory limit if size > 0 */
+			if (size > 0)
+				mem_limit = size;

Baoquan,

I am not sure about setting the value of mem_limit to mem_size directly.

If the command line has both the "memmap" and "mem", such as
... mem=2G memmap=4G ...

...in that code, the mem_limit may be 4G not 2G.

Quoted 7 lines
Quoted 2 lines

No, could you tell why you want to add both "memmap=nnKMG" and "mem=" at the same time? As you sid, what if I add "mem=4G mem=2G mem=1G"?

Just for testing :)

Ok, thanks, I see. We should be responsible for our command line. don't need to consider with these situations in kernel.

Thanks for testing and looking into this patchset, Liyang. I understand your worry, this patch is trying to have the same behaviour as in arch/x86/kernel/e820.c. Seems the liability of specifying them right lays on users' shoulder.

Thanks
Baoquan

#8 lore raw

[tip:x86/boot] x86/KASLR: Handle the memory limit specified by the 'memmap=' and 'mem=' boot options

In reply to Baoquan He
x86/KASLR: Handle the memory limit specified by the 'memmap=' and 'mem=' boot options
Commit-ID:  4cdba14f84c9102c4434384731cd61018b970d59
Gitweb:     http://git.kernel.org/tip/4cdba14f84c9102c4434384731cd61018b970d59
Author:     Baoquan He <[email protected]>
AuthorDate: Sat, 13 May 2017 13:46:29 +0800
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 24 May 2017 09:50:27 +0200

x86/KASLR: Handle the memory limit specified by the 'memmap=' and 'mem=' boot options

The 'mem=' boot option limits the max address a system can use - any memory region above the limit will be removed.

Furthermore, the 'memmap=nn[KMG]' variant (with no offset specified) has the same behaviour as 'mem='.

KASLR needs to consider this when choosing the random position for decompressing the kernel. Do it.

---
 arch/x86/boot/compressed/kaslr.c | 68 +++++++++++++++++++++++++++++-----------
 1 file changed, 50 insertions(+), 18 deletions(-)
arch/x86/boot/compressed/kaslr.c +50 -18
index 106e13b..e0eba12 100644
@@ -88,6 +88,10 @@ struct mem_vector {
 static bool memmap_too_large;
 
 
+/* Store memory limit specified by "mem=nn[KMG]" or "memmap=nn[KMG]" */
+unsigned long long mem_limit = ULLONG_MAX;
+
+
 enum mem_avoid_index {
 	MEM_AVOID_ZO_RANGE = 0,
 	MEM_AVOID_INITRD,
@@ -138,16 +142,23 @@ parse_memmap(char *p, unsigned long long *start, unsigned long long *size)
 		return -EINVAL;
 
 	switch (*p) {
-	case '@':
-		/* Skip this region, usable */
-		*start = 0;
-		*size = 0;
-		return 0;
 	case '#':
 	case '$':
 	case '!':
 		*start = memparse(p + 1, &p);
 		return 0;
+	case '@':
+		/* memmap=nn@ss specifies usable region, should be skipped */
+		*size = 0;
+		/* Fall through */
+	default:
+		/*
+		 * If w/o offset, only size specified, memmap=nn[KMG] has the
+		 * same behaviour as mem=nn[KMG]. It limits the max address
+		 * system can use. Region above the limit should be avoided.
+		 */
+		*start = 0;
+		return 0;
 	}
 
 	return -EINVAL;
@@ -173,9 +184,14 @@ static void mem_avoid_memmap(char *str)
 		if (rc < 0)
 			break;
 		str = k;
-		/* A usable region that should not be skipped */
-		if (size == 0)
+
+		if (start == 0) {
+			/* Store the specified memory limit if size > 0 */
+			if (size > 0)
+				mem_limit = size;
+
 			continue;
+		}
 
 		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].start = start;
 		mem_avoid[MEM_AVOID_MEMMAP_BEGIN + i].size = size;
@@ -187,19 +203,15 @@ static void mem_avoid_memmap(char *str)
 		memmap_too_large = true;
 }
 
-
-/*
- * handle_mem_memmap will also cover 'mem=' issue in next patch. Will remove
- * this note later.
- */
 static int handle_mem_memmap(void)
 {
 	char *args = (char *)get_cmd_line_ptr();
 	size_t len = strlen((char *)args);
 	char *tmp_cmdline;
 	char *param, *val;
+	u64 mem_size;
 
-	if (!strstr(args, "memmap="))
+	if (!strstr(args, "memmap=") && !strstr(args, "mem="))
 		return 0;
 
 	tmp_cmdline = malloc(len + 1);
@@ -222,8 +234,20 @@ static int handle_mem_memmap(void)
 			return -1;
 		}
 
-		if (!strcmp(param, "memmap"))
+		if (!strcmp(param, "memmap")) {
 			mem_avoid_memmap(val);
+		} else if (!strcmp(param, "mem")) {
+			char *p = val;
+
+			if (!strcmp(p, "nopentium"))
+				continue;
+			mem_size = memparse(p, &p);
+			if (mem_size == 0) {
+				free(tmp_cmdline);
+				return -EINVAL;
+			}
+			mem_limit = mem_size;
+		}
 	}
 
 	free(tmp_cmdline);
@@ -460,7 +484,8 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 {
 	struct mem_vector region, overlap;
 	struct slot_area slot_area;
-	unsigned long start_orig;
+	unsigned long start_orig, end;
+	struct boot_e820_entry cur_entry;
 
 	/* Skip non-RAM entries. */
 	if (entry->type != E820_TYPE_RAM)
@@ -474,8 +499,15 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 	if (entry->addr + entry->size < minimum)
 		return;
 
-	region.start = entry->addr;
-	region.size = entry->size;
+	/* Ignore entries above memory limit */
+	end = min(entry->size + entry->addr, mem_limit);
+	if (entry->addr >= end)
+		return;
+	cur_entry.addr = entry->addr;
+	cur_entry.size = end - entry->addr;
+
+	region.start = cur_entry.addr;
+	region.size = cur_entry.size;
 
 	/* Give up if slot area array is full. */
 	while (slot_area_index < MAX_SLOT_AREA) {
@@ -489,7 +521,7 @@ static void process_e820_entry(struct boot_e820_entry *entry,
 		region.start = ALIGN(region.start, CONFIG_PHYSICAL_ALIGN);
 
 		/* Did we raise the address above this e820 region? */
-		if (region.start > entry->addr + entry->size)
+		if (region.start > cur_entry.addr + cur_entry.size)
 			return;
 
 		/* Reduce size by any delta from the original address. */
author
#9 lore raw

PATCH v5 3/3 Documentation/kernel-parameters.txt: Update 'memmap=' option description

In reply to Baoquan He

In commit:

9710f581bb4c ("x86, mm: Let "memmap=" take more entries one time")

... 'memmap=' was changed to adopt multiple, comma delimited values in a single entry, so update the related description.

In the special case of only specifying size value without an offset, like memmap=nn[KMG], memmap behaves similarly to mem=nn[KMG], so update it too here.

Furthermore, for memmap=nn[KMG]$ss[KMG], an escape character needs be added before '$' for some bootloaders. E.g in grub2, if we specify memmap=100M$5G as suggested by the documentation, "memmap=100MG" gets passed to the kernel.

Clarify all this.

---
 Documentation/admin-guide/kernel-parameters.txt | 9 +++++++++
 1 file changed, 9 insertions(+)
Documentation/admin-guide/kernel-parameters.txt +9 -0
index 15f79c2..60d69aa 100644
@@ -2127,6 +2127,12 @@
 	memmap=nn[KMG]@ss[KMG]
 			[KNL] Force usage of a specific region of memory.
 			Region of memory to be used is from ss to ss+nn.
+			If @ss[KMG] is omitted, it equals to mem=nn[KMG]
+			which limits max address as nn[KMG].
+			Multiple different options can be put into one entry
+			with comma delimited to save space:
+			Example:
+				memmap=100M@2G,100M#3G,1G!1024G
 
 	memmap=nn[KMG]#ss[KMG]
 			[KNL,ACPI] Mark specific memory as ACPI data.
@@ -2139,6 +2145,9 @@
 			         memmap=64K$0x18690000
 			         or
 			         memmap=0x10000$0x18690000
+			Some bootloaders may need escape character before '$',
+			like in grub2, otherwise '$' and the following number
+			will be eaten.
 
 	memmap=nn[KMG]!ss[KMG]
 			[KNL,X86] Mark specific memory as protected.
Signature · 1 line
2.5.5
#10 lore raw

[tip:x86/boot] Documentation/kernel-parameters.txt: Update 'memmap=' boot option description

Documentation/kernel-parameters.txt: Update 'memmap=' boot option description
Commit-ID:  8fcc9bc3eaa2ef8345e2b4b22e3a88804ac46337
Gitweb:     http://git.kernel.org/tip/8fcc9bc3eaa2ef8345e2b4b22e3a88804ac46337
Author:     Baoquan He <[email protected]>
AuthorDate: Sat, 13 May 2017 13:46:30 +0800
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 24 May 2017 09:50:27 +0200

Documentation/kernel-parameters.txt: Update 'memmap=' boot option description

In commit:

9710f581bb4c ("x86, mm: Let "memmap=" take more entries one time")

... 'memmap=' was changed to adopt multiple, comma delimited values in a single entry, so update the related description.

In the special case of only specifying size value without an offset, like memmap=nn[KMG], memmap behaves similarly to mem=nn[KMG], so update it too here.

Furthermore, for memmap=nn[KMG]$ss[KMG], an escape character needs be added before '$' for some bootloaders. E.g in grub2, if we specify memmap=100M$5G as suggested by the documentation, "memmap=100MG" gets passed to the kernel.

Clarify all this.

---
 Documentation/admin-guide/kernel-parameters.txt | 9 +++++++++
 1 file changed, 9 insertions(+)
Documentation/admin-guide/kernel-parameters.txt +9 -0
index 15f79c2..4e4c340 100644
@@ -2127,6 +2127,12 @@
 	memmap=nn[KMG]@ss[KMG]
 			[KNL] Force usage of a specific region of memory.
 			Region of memory to be used is from ss to ss+nn.
+			If @ss[KMG] is omitted, it is equivalent to mem=nn[KMG],
+			which limits max address to nn[KMG].
+			Multiple different regions can be specified,
+			comma delimited.
+			Example:
+				memmap=100M@2G,100M#3G,1G!1024G
 
 	memmap=nn[KMG]#ss[KMG]
 			[KNL,ACPI] Mark specific memory as ACPI data.
@@ -2139,6 +2145,9 @@
 			         memmap=64K$0x18690000
 			         or
 			         memmap=0x10000$0x18690000
+			Some bootloaders may need an escape character before '$',
+			like Grub2, otherwise '$' and the following number
+			will be eaten.
 
 	memmap=nn[KMG]!ss[KMG]
 			[KNL,X86] Mark specific memory as protected.
Acked-by on the v5 series
#11 lore raw
In reply to Baoquan He
Quoted 7 lines · Baoquan He

People reported kernel panic occurs during system boots up with mem boot option. After checking code, several problems are found about memmap= and mem= in boot stage kaslr.

*) In commit f28442497b5c ("x86/boot: Fix KASLR and memmap= collision"), only one memmap
entry is considered and only the last one if multiple memmap entries are specified.

Quoted 7 lines

*) mem= and memmap=nn[KMG] are not considered yet. They are used to limit max address
of system. Kernel can't be randomized to be above the limit.

*) kernel-parameters.txt doesn't tell the updated behaviour of memmap=.

This patchset tries to solve above issues, and it sits on top of tip:x86/boot branch.

Thanks for keeping at this. I think they look good now!

-Kees

Quoted 53 lines

Changelog
v4->v5:
1. Change patch log according to Thomas's comment.

2. Put "Fall through" to the right place in parse_memmap() according to Kees's suggestion.

v3->v4:
1. Code improved patch 1/3 according to Kees's suggestion.

2. Add 'Fall through' in switch case of parse_memmap() which is suggestd by Kees.

v2->v3:
No functionality change in this round.
1. Use local static variable insted of global variable mem_avoid_memmap_index in patch 1/3.

2. Fix a typo in patch 3/3.

v1->v2:
1. The original patch 1/4 has been put in tip:x86/boot and no update, so it's not included in this post.

2. Use patch log Ingo reorganized.

3. lib/ctype.c and lib/cmdline.c are needed for kaslr.c, while those EXPORT_SYMBOL(x) contained caused failure of build on 32-bit allmodconfig:
......
ld: -r and -shared may not be used together
scripts/Makefile.build:294: recipe for target 'arch/x86/boot/compressed/kaslr.o' failed
......
Disabling the symbol exporting removes the build failure.

4. Use dynamic allocation to allocate memory to contain copied kernel cmdline buffer, it's implemented in include/linux/decompress/mm.h.

Baoquan He (3):
KASLR: Parse all memmap entries in command line
KASLR: Handle memory limit specified by memmap and mem option Documentation/kernel-parameters.txt: Update 'memmap=' option
description

 Documentation/admin-guide/kernel-parameters.txt |   9 ++
 arch/x86/boot/compressed/cmdline.c              |   2 +-
 arch/x86/boot/compressed/kaslr.c                | 190 ++++++++++++++++--------
 arch/x86/boot/string.c                          |   8 +
 4 files changed, 143 insertions(+), 66 deletions(-)

--
2.5.5

Signature · 2 lines
Kees Cook
Pixel Security