Skip to contentSkip to conversation

kexec: Move vmcoreinfo out of the kernel's .bss section

PATCH Superseded Xunlei Pang 7 participants 18 messages Mar 20, 2017 – Mar 24, 2017

Subsystems (6)
x86 PowerPC Driver core Live update +2 more subsystems

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/3 kexec: Move vmcoreinfo out of the kernel's .bss section

As Eric said,
"what we need to do is move the variable vmcoreinfo_note out of the kernel's .bss section. And modify the code to regenerate and keep this information in something like the control page.

Definitely something like this needs a page all to itself, and ideally far away from any other kernel data structures. I clearly was not watching closely the data someone decided to keep this silly thing in the kernel's .bss section."

This patch allocates extra pages for these vmcoreinfo_XXX variables, one advantage is that it enhances some safety of vmcoreinfo, because vmcoreinfo now is kept far away from other kernel data structures.

---
 arch/ia64/kernel/machine_kexec.c |  5 -----
 arch/x86/kernel/crash.c          |  2 +-
 include/linux/kexec.h            |  2 +-
 kernel/kexec_core.c              | 29 ++++++++++++++++++++++++-----
 kernel/ksysfs.c                  |  2 +-
 5 files changed, 27 insertions(+), 13 deletions(-)
arch/ia64/kernel/machine_kexec.c +0 -5
index 599507b..c14815d 100644
@@ -163,8 +163,3 @@ void arch_crash_save_vmcoreinfo(void)
 #endif
 }
 
-phys_addr_t paddr_vmcoreinfo_note(void)
-{
-	return ia64_tpa((unsigned long)(char *)&vmcoreinfo_note);
-}
-
arch/x86/kernel/crash.c +1 -1
index 3741461..4d35fbb 100644
@@ -456,7 +456,7 @@ static int prepare_elf64_headers(struct crash_elf_data *ced,
 	bufp += sizeof(Elf64_Phdr);
 	phdr->p_type = PT_NOTE;
 	phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
-	phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
+	phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
 	(ehdr->e_phnum)++;
 
 #ifdef CONFIG_X86_64
include/linux/kexec.h +1 -1
index e98e546..f1c601b 100644
@@ -317,7 +317,7 @@ extern void *kexec_purgatory_get_symbol_addr(struct kimage *image,
 extern struct resource crashk_low_res;
 typedef u32 note_buf_t[KEXEC_NOTE_BYTES/4];
 extern note_buf_t __percpu *crash_notes;
-extern u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
+extern u32 *vmcoreinfo_note;
 extern size_t vmcoreinfo_size;
 extern size_t vmcoreinfo_max_size;
 
kernel/kexec_core.c +24 -5
index bfe62d5..e3a4bda 100644
@@ -52,10 +52,10 @@
 note_buf_t __percpu *crash_notes;
 
 /* vmcoreinfo stuff */
-static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
-u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
+static unsigned char *vmcoreinfo_data;
 size_t vmcoreinfo_size;
-size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
+size_t vmcoreinfo_max_size = VMCOREINFO_BYTES;
+u32 *vmcoreinfo_note;
 
 /* Flag to indicate we are going to kexec a new kernel */
 bool kexec_in_progress = false;
@@ -1369,6 +1369,9 @@ static void update_vmcoreinfo_note(void)
 
 void crash_save_vmcoreinfo(void)
 {
+	if (!vmcoreinfo_note)
+		return;
+
 	vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
 	update_vmcoreinfo_note();
 }
@@ -1397,13 +1400,29 @@ void vmcoreinfo_append_str(const char *fmt, ...)
 void __weak arch_crash_save_vmcoreinfo(void)
 {}
 
-phys_addr_t __weak paddr_vmcoreinfo_note(void)
+phys_addr_t paddr_vmcoreinfo_note(void)
 {
-	return __pa_symbol((unsigned long)(char *)&vmcoreinfo_note);
+	return __pa(vmcoreinfo_note);
 }
 
 static int __init crash_save_vmcoreinfo_init(void)
 {
+	/* One page should be enough for VMCOREINFO_BYTES under all archs */
+	vmcoreinfo_data = (unsigned char *)get_zeroed_page(GFP_KERNEL);
+	if (!vmcoreinfo_data) {
+		pr_warn("Memory allocation for vmcoreinfo_data failed\n");
+		return -ENOMEM;
+	}
+
+	vmcoreinfo_note = alloc_pages_exact(VMCOREINFO_NOTE_SIZE,
+						GFP_KERNEL | __GFP_ZERO);
+	if (!vmcoreinfo_note) {
+		free_page((unsigned long)vmcoreinfo_data);
+		vmcoreinfo_data = NULL;
+		pr_warn("Memory allocation for vmcoreinfo_note failed\n");
+		return -ENOMEM;
+	}
+
 	VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
 	VMCOREINFO_PAGESIZE(PAGE_SIZE);
 
kernel/ksysfs.c +1 -1
index ee1bc1b..9de6fcc 100644
@@ -130,7 +130,7 @@ static ssize_t vmcoreinfo_show(struct kobject *kobj,
 {
 	phys_addr_t vmcore_base = paddr_vmcoreinfo_note();
 	return sprintf(buf, "%pa %x\n", &vmcore_base,
-		       (unsigned int)sizeof(vmcoreinfo_note));
+			(unsigned int)VMCOREINFO_NOTE_SIZE);
 }
 KERNEL_ATTR_RO(vmcoreinfo);
 
Signature · 1 line
1.8.3.1
#1 lore raw

PATCH v3 2/3 powerpc/fadump: Use the correct VMCOREINFO_NOTE_SIZE for phdr

vmcoreinfo_max_size stands for the vmcoreinfo_data, the correct one we should use is vmcoreinfo_note whose total size is VMCOREINFO_NOTE_SIZE.

Like explained in commit 77019967f06b ("kdump: fix exported size of vmcoreinfo note"), it does not affect the actual function, we better fix it, also this change should be safe and backward compatible.

After this, we can get rid of variable vmcoreinfo_max_size, let's use the macro VMCOREINFO_BYTES instead, fewer variables means more safety for vmcoreinfo operation.

---
 arch/powerpc/kernel/fadump.c | 3 +--
 include/linux/kexec.h        | 1 -
 kernel/kexec_core.c          | 3 +--
 3 files changed, 2 insertions(+), 5 deletions(-)
arch/powerpc/kernel/fadump.c +1 -2
index 8ff0dd4..b8e15cf 100644
@@ -906,8 +906,7 @@ static int fadump_create_elfcore_headers(char *bufp)
 
 	phdr->p_paddr	= fadump_relocate(paddr_vmcoreinfo_note());
 	phdr->p_offset	= phdr->p_paddr;
-	phdr->p_memsz	= vmcoreinfo_max_size;
-	phdr->p_filesz	= vmcoreinfo_max_size;
+	phdr->p_memsz	= phdr->p_filesz = VMCOREINFO_NOTE_SIZE;
 
 	/* Increment number of program headers. */
 	(elf->e_phnum)++;
include/linux/kexec.h +0 -1
index f1c601b..6918fda 100644
@@ -319,7 +319,6 @@ extern void *kexec_purgatory_get_symbol_addr(struct kimage *image,
 extern note_buf_t __percpu *crash_notes;
 extern u32 *vmcoreinfo_note;
 extern size_t vmcoreinfo_size;
-extern size_t vmcoreinfo_max_size;
 
 /* flag to track if kexec reboot is in progress */
 extern bool kexec_in_progress;
kernel/kexec_core.c +1 -2
index e3a4bda..e503b48 100644
@@ -54,7 +54,6 @@
 /* vmcoreinfo stuff */
 static unsigned char *vmcoreinfo_data;
 size_t vmcoreinfo_size;
-size_t vmcoreinfo_max_size = VMCOREINFO_BYTES;
 u32 *vmcoreinfo_note;
 
 /* Flag to indicate we are going to kexec a new kernel */
@@ -1386,7 +1385,7 @@ void vmcoreinfo_append_str(const char *fmt, ...)
 	r = vscnprintf(buf, sizeof(buf), fmt, args);
 	va_end(args);
 
-	r = min(r, vmcoreinfo_max_size - vmcoreinfo_size);
+	r = min(r, VMCOREINFO_BYTES - vmcoreinfo_size);
 
 	memcpy(&vmcoreinfo_data[vmcoreinfo_size], buf, r);
 
Signature · 7 lines
1.8.3.1


_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec
#2 lore raw

PATCH v3 3/3 kdump: Relocate vmcoreinfo to the crash memory range

In reply to Xunlei Pang

Currently vmcoreinfo data is updated at boot time subsys_initcall(), it has the risk of being modified by some wrong code during system is running.

As a result, vmcore dumped may contain the wrong vmcoreinfo. Later on, when using "crash", "makedumpfile", etc utility to parse this vmcore, we probably will get "Segmentation fault" or other unexpected errors.

E.g. 1) wrong code overwrites vmcoreinfo_data; 2) further crashes the system; 3) trigger kdump, then we obviously will fail to recognize the crash context correctly due to the corrupted vmcoreinfo.

Now except for vmcoreinfo, all the crash data is well protected(including the cpu note which is fully updated in the crash path, thus its correctness is guaranteed). Given that vmcoreinfo data is a large chunk, we better protect it as well.

To solve this, we relocate and copy vmcoreinfo_data to the crash memory when kdump is loading via kexec syscalls. Because the whole crash memory will be protected by existing arch_kexec_protect_crashkres() mechanism, we naturally protect vmcoreinfo_data from write(even read) access under kernel direct mapping after kdump is loaded.

Since kdump is usually loaded at the very early stage after boot, we can trust the correctness of the vmcoreinfo data copied.

On the other hand, we still need to operate the vmcoreinfo safe copy when crash happens to generate vmcoreinfo_note again, we rely on vmap() to map out a new kernel virtual address and update to use this new one instead in the following crash_save_vmcoreinfo().

BTW, we do not touch vmcoreinfo_note, because it will be fully updated using the protected vmcoreinfo_data after crash which is surely correct just like the cpu crash note.

---
 include/linux/kexec.h |  3 +++
 kernel/kexec.c        |  3 +++
 kernel/kexec_core.c   | 52 +++++++++++++++++++++++++++++++++++++++++++++++++++
 kernel/kexec_file.c   |  3 +++
 4 files changed, 61 insertions(+)
include/linux/kexec.h +3 -0
index 6918fda..fae2fc6 100644
@@ -187,6 +187,8 @@ struct kimage {
 	unsigned long start;
 	struct page *control_code_page;
 	struct page *swap_page;
+	void *vmcoreinfo_data_copy; /* locates in the crash memory */
+	size_t vmcoreinfo_size_copy;
 
 	unsigned long nr_segments;
 	struct kexec_segment segment[KEXEC_SEGMENT_MAX];
@@ -243,6 +245,7 @@ extern asmlinkage long sys_kexec_load(unsigned long entry,
 extern int kernel_kexec(void);
 extern struct page *kimage_alloc_control_pages(struct kimage *image,
 						unsigned int order);
+extern int kimage_crash_copy_vmcoreinfo(struct kimage *image);
 extern int kexec_load_purgatory(struct kimage *image, unsigned long min,
 				unsigned long max, int top_down,
 				unsigned long *load_addr);
kernel/kexec.c +3 -0
index 980936a..e0c4dea 100644
@@ -93,6 +93,9 @@ static int kimage_alloc_init(struct kimage **rimage, unsigned long entry,
 			pr_err("Could not allocate swap buffer\n");
 			goto out_free_control_pages;
 		}
+	} else {
+		if (kimage_crash_copy_vmcoreinfo(image) < 0)
+			goto out_free_image;
 	}
 
 	*rimage = image;
kernel/kexec_core.c +52 -0

70 lines not on this page · raw message

kernel/kexec_file.c +3 -0
index b56a558..6bb3e4d 100644
@@ -244,6 +244,9 @@ void kimage_file_post_load_cleanup(struct kimage *image)
 			pr_err("Could not allocate swap buffer\n");
 			goto out_free_control_pages;
 		}
+	} else {
+		if (kimage_crash_copy_vmcoreinfo(image) < 0)
+			goto out_free_post_load_bufs;
 	}
 
 	*rimage = image;
Signature · 7 lines
1.8.3.1


_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec
#3 lore raw
In reply to Xunlei Pang
Quoted 5 lines · Xunlei Pang

As Eric said,
"what we need to do is move the variable vmcoreinfo_note out of the kernel's .bss section. And modify the code to regenerate and keep this information in something like the control page.

Quoted 8 lines

Definitely something like this needs a page all to itself, and ideally far away from any other kernel data structures. I clearly was not watching closely the data someone decided to keep this silly thing in the kernel's .bss section."

This patch allocates extra pages for these vmcoreinfo_XXX variables, one advantage is that it enhances some safety of vmcoreinfo, because vmcoreinfo now is kept far away from other kernel data structures.

Can you preceed this patch with a patch that removes CRASHTIME from vmcoreinfo? If someone actually cares we can add a separate note that holds a 64bit crashtime in the per cpu notes.

As we are looking at reliability concerns removing CRASHTIME should make everything in vmcoreinfo a boot time constant. Which should simplify everything considerably.

Which means we only need to worry abou the per-cpu notes being written at the time of a crash.

Quoted 121 lines
---
 arch/ia64/kernel/machine_kexec.c |  5 -----
 arch/x86/kernel/crash.c          |  2 +-
 include/linux/kexec.h            |  2 +-
 kernel/kexec_core.c              | 29 ++++++++++++++++++++++++-----
 kernel/ksysfs.c                  |  2 +-
 5 files changed, 27 insertions(+), 13 deletions(-)
arch/ia64/kernel/machine_kexec.c +0 -5
index 599507b..c14815d 100644
@@ -163,8 +163,3 @@ void arch_crash_save_vmcoreinfo(void)
 #endif
 }
 
-phys_addr_t paddr_vmcoreinfo_note(void)
-{
-	return ia64_tpa((unsigned long)(char *)&vmcoreinfo_note);
-}
-
arch/x86/kernel/crash.c +1 -1
index 3741461..4d35fbb 100644
@@ -456,7 +456,7 @@ static int prepare_elf64_headers(struct crash_elf_data *ced,
 	bufp += sizeof(Elf64_Phdr);
 	phdr->p_type = PT_NOTE;
 	phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
-	phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
+	phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
 	(ehdr->e_phnum)++;
 
 #ifdef CONFIG_X86_64
include/linux/kexec.h +1 -1
index e98e546..f1c601b 100644
@@ -317,7 +317,7 @@ extern void *kexec_purgatory_get_symbol_addr(struct kimage *image,
 extern struct resource crashk_low_res;
 typedef u32 note_buf_t[KEXEC_NOTE_BYTES/4];
 extern note_buf_t __percpu *crash_notes;
-extern u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
+extern u32 *vmcoreinfo_note;
 extern size_t vmcoreinfo_size;
 extern size_t vmcoreinfo_max_size;
 
kernel/kexec_core.c +24 -5
index bfe62d5..e3a4bda 100644
@@ -52,10 +52,10 @@
 note_buf_t __percpu *crash_notes;
 
 /* vmcoreinfo stuff */
-static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
-u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
+static unsigned char *vmcoreinfo_data;
 size_t vmcoreinfo_size;
-size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
+size_t vmcoreinfo_max_size = VMCOREINFO_BYTES;
+u32 *vmcoreinfo_note;
 
 /* Flag to indicate we are going to kexec a new kernel */
 bool kexec_in_progress = false;
@@ -1369,6 +1369,9 @@ static void update_vmcoreinfo_note(void)
 
 void crash_save_vmcoreinfo(void)
 {
+	if (!vmcoreinfo_note)
+		return;
+
 	vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
 	update_vmcoreinfo_note();
 }
@@ -1397,13 +1400,29 @@ void vmcoreinfo_append_str(const char *fmt, ...)
 void __weak arch_crash_save_vmcoreinfo(void)
 {}
 
-phys_addr_t __weak paddr_vmcoreinfo_note(void)
+phys_addr_t paddr_vmcoreinfo_note(void)
 {
-	return __pa_symbol((unsigned long)(char *)&vmcoreinfo_note);
+	return __pa(vmcoreinfo_note);
 }
 
 static int __init crash_save_vmcoreinfo_init(void)
 {
+	/* One page should be enough for VMCOREINFO_BYTES under all archs */
+	vmcoreinfo_data = (unsigned char *)get_zeroed_page(GFP_KERNEL);
+	if (!vmcoreinfo_data) {
+		pr_warn("Memory allocation for vmcoreinfo_data failed\n");
+		return -ENOMEM;
+	}
+
+	vmcoreinfo_note = alloc_pages_exact(VMCOREINFO_NOTE_SIZE,
+						GFP_KERNEL | __GFP_ZERO);
+	if (!vmcoreinfo_note) {
+		free_page((unsigned long)vmcoreinfo_data);
+		vmcoreinfo_data = NULL;
+		pr_warn("Memory allocation for vmcoreinfo_note failed\n");
+		return -ENOMEM;
+	}
+
 	VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
 	VMCOREINFO_PAGESIZE(PAGE_SIZE);
 
kernel/ksysfs.c +1 -1
index ee1bc1b..9de6fcc 100644
@@ -130,7 +130,7 @@ static ssize_t vmcoreinfo_show(struct kobject *kobj,
 {
 	phys_addr_t vmcore_base = paddr_vmcoreinfo_note();
 	return sprintf(buf, "%pa %x\n", &vmcore_base,
-		       (unsigned int)sizeof(vmcoreinfo_note));
+			(unsigned int)VMCOREINFO_NOTE_SIZE);
 }
 KERNEL_ATTR_RO(vmcoreinfo);
#4 lore raw
Quoted 12 lines · Eric W. Biederman
Quoted 10 lines · Xunlei Pang

As Eric said,
"what we need to do is move the variable vmcoreinfo_note out of the kernel's .bss section. And modify the code to regenerate and keep this information in something like the control page.

Definitely something like this needs a page all to itself, and ideally far away from any other kernel data structures. I clearly was not watching closely the data someone decided to keep this silly thing in the kernel's .bss section."

Quoted 7 lines
Quoted 3 lines

This patch allocates extra pages for these vmcoreinfo_XXX variables, one advantage is that it enhances some safety of vmcoreinfo, because vmcoreinfo now is kept far away from other kernel data structures.

Can you preceed this patch with a patch that removes CRASHTIME from vmcoreinfo? If someone actually cares we can add a separate note that holds a 64bit crashtime in the per cpu notes.

I think makedumpfile is using it, but I also vote to remove the CRASHTIME. It is better not to do this while crashing and a makedumpfile userspace patch is needed to drop the use of it.

Quoted 4 lines

As we are looking at reliability concerns removing CRASHTIME should make everything in vmcoreinfo a boot time constant. Which should simplify everything considerably.

It is a nice improvement..

Quoted 122 lines

Which means we only need to worry abou the per-cpu notes being written at the time of a crash.

Quoted 118 lines
---
 arch/ia64/kernel/machine_kexec.c |  5 -----
 arch/x86/kernel/crash.c          |  2 +-
 include/linux/kexec.h            |  2 +-
 kernel/kexec_core.c              | 29 ++++++++++++++++++++++++-----
 kernel/ksysfs.c                  |  2 +-
 5 files changed, 27 insertions(+), 13 deletions(-)
arch/ia64/kernel/machine_kexec.c +0 -5
index 599507b..c14815d 100644
@@ -163,8 +163,3 @@ void arch_crash_save_vmcoreinfo(void)
 #endif
 }
 
-phys_addr_t paddr_vmcoreinfo_note(void)
-{
-	return ia64_tpa((unsigned long)(char *)&vmcoreinfo_note);
-}
-
arch/x86/kernel/crash.c +1 -1
index 3741461..4d35fbb 100644
@@ -456,7 +456,7 @@ static int prepare_elf64_headers(struct crash_elf_data *ced,
 	bufp += sizeof(Elf64_Phdr);
 	phdr->p_type = PT_NOTE;
 	phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
-	phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
+	phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
 	(ehdr->e_phnum)++;
 
 #ifdef CONFIG_X86_64
include/linux/kexec.h +1 -1
index e98e546..f1c601b 100644
@@ -317,7 +317,7 @@ extern void *kexec_purgatory_get_symbol_addr(struct kimage *image,
 extern struct resource crashk_low_res;
 typedef u32 note_buf_t[KEXEC_NOTE_BYTES/4];
 extern note_buf_t __percpu *crash_notes;
-extern u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
+extern u32 *vmcoreinfo_note;
 extern size_t vmcoreinfo_size;
 extern size_t vmcoreinfo_max_size;
 
kernel/kexec_core.c +24 -5
index bfe62d5..e3a4bda 100644
@@ -52,10 +52,10 @@
 note_buf_t __percpu *crash_notes;
 
 /* vmcoreinfo stuff */
-static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
-u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
+static unsigned char *vmcoreinfo_data;
 size_t vmcoreinfo_size;
-size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
+size_t vmcoreinfo_max_size = VMCOREINFO_BYTES;
+u32 *vmcoreinfo_note;
 
 /* Flag to indicate we are going to kexec a new kernel */
 bool kexec_in_progress = false;
@@ -1369,6 +1369,9 @@ static void update_vmcoreinfo_note(void)
 
 void crash_save_vmcoreinfo(void)
 {
+	if (!vmcoreinfo_note)
+		return;
+
 	vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
 	update_vmcoreinfo_note();
 }
@@ -1397,13 +1400,29 @@ void vmcoreinfo_append_str(const char *fmt, ...)
 void __weak arch_crash_save_vmcoreinfo(void)
 {}
 
-phys_addr_t __weak paddr_vmcoreinfo_note(void)
+phys_addr_t paddr_vmcoreinfo_note(void)
 {
-	return __pa_symbol((unsigned long)(char *)&vmcoreinfo_note);
+	return __pa(vmcoreinfo_note);
 }
 
 static int __init crash_save_vmcoreinfo_init(void)
 {
+	/* One page should be enough for VMCOREINFO_BYTES under all archs */
+	vmcoreinfo_data = (unsigned char *)get_zeroed_page(GFP_KERNEL);
+	if (!vmcoreinfo_data) {
+		pr_warn("Memory allocation for vmcoreinfo_data failed\n");
+		return -ENOMEM;
+	}
+
+	vmcoreinfo_note = alloc_pages_exact(VMCOREINFO_NOTE_SIZE,
+						GFP_KERNEL | __GFP_ZERO);
+	if (!vmcoreinfo_note) {
+		free_page((unsigned long)vmcoreinfo_data);
+		vmcoreinfo_data = NULL;
+		pr_warn("Memory allocation for vmcoreinfo_note failed\n");
+		return -ENOMEM;
+	}
+
 	VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
 	VMCOREINFO_PAGESIZE(PAGE_SIZE);
 
kernel/ksysfs.c +0 -1
index ee1bc1b..9de6fcc 100644
@@ -130,7 +130,7 @@ static ssize_t vmcoreinfo_show(struct kobject *kobj,
 {
 	phys_addr_t vmcore_base = paddr_vmcoreinfo_note();
 	return sprintf(buf, "%pa %x\n", &vmcore_base,
-		       (unsigned int)sizeof(vmcoreinfo_note));
Quoted 8 lines

kernel/ksysfs.c static ssize_t vmcoreinfo_show(struct kobject *kobj, · line 135

Quoted 3 lines
kernel/ksysfs.c +1 -0
+			(unsigned int)VMCOREINFO_NOTE_SIZE);
 }
 KERNEL_ATTR_RO(vmcoreinfo);

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

Thanks
Dave

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#5 lore raw
Quoted 26 lines · Dave Young
Quoted 12 lines · Eric W. Biederman
Quoted 10 lines · Xunlei Pang

As Eric said,
"what we need to do is move the variable vmcoreinfo_note out of the kernel's .bss section. And modify the code to regenerate and keep this information in something like the control page.

Definitely something like this needs a page all to itself, and ideally far away from any other kernel data structures. I clearly was not watching closely the data someone decided to keep this silly thing in the kernel's .bss section."

Quoted 7 lines
Quoted 3 lines

This patch allocates extra pages for these vmcoreinfo_XXX variables, one advantage is that it enhances some safety of vmcoreinfo, because vmcoreinfo now is kept far away from other kernel data structures.

Can you preceed this patch with a patch that removes CRASHTIME from vmcoreinfo? If someone actually cares we can add a separate note that holds a 64bit crashtime in the per cpu notes.

I think makedumpfile is using it, but I also vote to remove the CRASHTIME. It is better not to do this while crashing and a makedumpfile userspace patch is needed to drop the use of it.

Quoted 1 line
Quoted 5 lines
Quoted 3 lines

As we are looking at reliability concerns removing CRASHTIME should make everything in vmcoreinfo a boot time constant. Which should simplify everything considerably.

It is a nice improvement..

We also need to take a close look at what s390 is doing with vmcoreinfo. As apparently it is reading it in a different kind of crashdump process.

Eric

#6 lore raw
Quoted 28 lines · Eric W. Biederman
Quoted 26 lines · Dave Young
Quoted 12 lines · Eric W. Biederman
Quoted 10 lines · Xunlei Pang

As Eric said,
"what we need to do is move the variable vmcoreinfo_note out of the kernel's .bss section. And modify the code to regenerate and keep this information in something like the control page.

Definitely something like this needs a page all to itself, and ideally far away from any other kernel data structures. I clearly was not watching closely the data someone decided to keep this silly thing in the kernel's .bss section."

Quoted 7 lines
Quoted 3 lines

This patch allocates extra pages for these vmcoreinfo_XXX variables, one advantage is that it enhances some safety of vmcoreinfo, because vmcoreinfo now is kept far away from other kernel data structures.

Can you preceed this patch with a patch that removes CRASHTIME from vmcoreinfo? If someone actually cares we can add a separate note that holds a 64bit crashtime in the per cpu notes.

I think makedumpfile is using it, but I also vote to remove the CRASHTIME. It is better not to do this while crashing and a makedumpfile userspace patch is needed to drop the use of it.

Quoted 1 line
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

As we are looking at reliability concerns removing CRASHTIME should make everything in vmcoreinfo a boot time constant. Which should simplify everything considerably.

It is a nice improvement..

We also need to take a close look at what s390 is doing with vmcoreinfo. As apparently it is reading it in a different kind of crashdump process.

Yes, need careful review from s390 and maybe ppc64 especially about patch 2/3, better to have comments from IBM about s390 dump tool and ppc fadump. Added more cc.

Thanks
Dave

#7 lore raw
Quoted 19 lines · Dave Young
Quoted 18 lines · Eric W. Biederman
Quoted 16 lines · Dave Young
Quoted 15 lines · Eric W. Biederman
Quoted 13 lines · Xunlei Pang

As Eric said,
"what we need to do is move the variable vmcoreinfo_note out of the kernel's .bss section. And modify the code to regenerate and keep this information in something like the control page.

Definitely something like this needs a page all to itself, and ideally far away from any other kernel data structures. I clearly was not watching closely the data someone decided to keep this silly thing in the kernel's .bss section."

This patch allocates extra pages for these vmcoreinfo_XXX variables, one advantage is that it enhances some safety of vmcoreinfo, because vmcoreinfo now is kept far away from other kernel data structures.

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 3 lines

Can you preceed this patch with a patch that removes CRASHTIME from vmcoreinfo? If someone actually cares we can add a separate note that holds a 64bit crashtime in the per cpu notes.

I think makedumpfile is using it, but I also vote to remove the CRASHTIME. It is better not to do this while crashing and a makedumpfile userspace patch is needed to drop the use of it.

By moving the CRASHTIME info to the cpu note of crashed cpu may be a good way. In kdump kernel, notes of vmcore elfhdr will be merged into one big note section, I don't know how makedumpfile or crash handle the big note section? If they process the note in some order, breakage will definitely happen...

There is also a fadump may be affected.

Regards,
Xunlei

Quoted 17 lines
Quoted 6 lines
Quoted 4 lines
Quoted 3 lines

As we are looking at reliability concerns removing CRASHTIME should make everything in vmcoreinfo a boot time constant. Which should simplify everything considerably.

It is a nice improvement..

We also need to take a close look at what s390 is doing with vmcoreinfo. As apparently it is reading it in a different kind of crashdump process.

Yes, need careful review from s390 and maybe ppc64 especially about patch 2/3, better to have comments from IBM about s390 dump tool and ppc fadump. Added more cc.

Thanks
Dave

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#8 lore raw

Hi Xunlei,

Quoted 27 lines · Xunlei Pang
Quoted 19 lines · Dave Young
Quoted 18 lines · Eric W. Biederman
Quoted 16 lines · Dave Young
Quoted 15 lines · Eric W. Biederman
Quoted 13 lines · Xunlei Pang

As Eric said,
"what we need to do is move the variable vmcoreinfo_note out of the kernel's .bss section. And modify the code to regenerate and keep this information in something like the control page.

Definitely something like this needs a page all to itself, and ideally far away from any other kernel data structures. I clearly was not watching closely the data someone decided to keep this silly thing in the kernel's .bss section."

This patch allocates extra pages for these vmcoreinfo_XXX variables, one advantage is that it enhances some safety of vmcoreinfo, because vmcoreinfo now is kept far away from other kernel data structures.

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 3 lines

Can you preceed this patch with a patch that removes CRASHTIME from vmcoreinfo? If someone actually cares we can add a separate note that holds a 64bit crashtime in the per cpu notes.

I think makedumpfile is using it, but I also vote to remove the CRASHTIME. It is better not to do this while crashing and a makedumpfile userspace patch is needed to drop the use of it.

Quoted 7 lines

By moving the CRASHTIME info to the cpu note of crashed cpu may be a good way. In kdump kernel, notes of vmcore elfhdr will be merged into one big note section, I don't know how makedumpfile or crash handle the big note section? If they process the note in some order, breakage will definitely happen...

There is also a fadump may be affected.

Would like to keep a tab of such change as fadump builds cpu notes differently and such change may have an impact on it considering it depends on the same tools - crash, makedumpfile..

Thanks
Hari

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#9 lore raw
In reply to Dave Young

Hi Dave,

Quoted 29 lines · Dave Young
Quoted 28 lines · Eric W. Biederman
Quoted 26 lines · Dave Young
Quoted 12 lines · Eric W. Biederman
Quoted 10 lines · Xunlei Pang

As Eric said,
"what we need to do is move the variable vmcoreinfo_note out of the kernel's .bss section. And modify the code to regenerate and keep this information in something like the control page.

Definitely something like this needs a page all to itself, and ideally far away from any other kernel data structures. I clearly was not watching closely the data someone decided to keep this silly thing in the kernel's .bss section."

Quoted 6 lines
Quoted 3 lines

This patch allocates extra pages for these vmcoreinfo_XXX variables, one advantage is that it enhances some safety of vmcoreinfo, because vmcoreinfo now is kept far away from other kernel data structures.

Can you preceed this patch with a patch that removes CRASHTIME from vmcoreinfo? If someone actually cares we can add a separate note that holds a 64bit crashtime in the per cpu notes.

I think makedumpfile is using it, but I also vote to remove the CRASHTIME. It is better not to do this while crashing and a makedumpfile userspace patch is needed to drop the use of it.

Quoted 3 lines

As we are looking at reliability concerns removing CRASHTIME should make everything in vmcoreinfo a boot time constant. Which should simplify everything considerably.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

It is a nice improvement..

We also need to take a close look at what s390 is doing with vmcoreinfo. As apparently it is reading it in a different kind of crashdump process.

Yes, need careful review from s390 and maybe ppc64 especially about patch 2/3, better to have comments from IBM about s390 dump tool and ppc fadump. Added more cc.

w.r.t powerpc/fadump, this patch-set works fine..

Thanks
Hari

#10 lore raw
In reply to Dave Young

Am Wed, 22 Mar 2017 12:30:04 +0800
schrieb Dave Young <[email protected]>:

Quoted 3 lines · Eric W. Biederman
Quoted 2 lines

Dave Young <[email protected]> writes:

[snip]

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines

I think makedumpfile is using it, but I also vote to remove the CRASHTIME. It is better not to do this while crashing and a makedumpfile userspace patch is needed to drop the use of it.

Quoted 4 lines

As we are looking at reliability concerns removing CRASHTIME should make everything in vmcoreinfo a boot time constant. Which should simplify everything considerably.

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

It is a nice improvement..

We also need to take a close look at what s390 is doing with vmcoreinfo. As apparently it is reading it in a different kind of crashdump process.

Yes, need careful review from s390 and maybe ppc64 especially about patch 2/3, better to have comments from IBM about s390 dump tool and ppc fadump. Added more cc.

On s390 we have at least an issue with patch 1/3. For stand-alone dump and also because we create the ELF header for kdump in the new kernel we save the pointer to the vmcoreinfo note in the old kernel on a defined memory address in our absolute zero lowcore.

This is done in arch/s390/kernel/setup.c:

static void __init setup_vmcoreinfo(void)

{
        mem_assign_absolute(S390_lowcore.vmcore_info, paddr_vmcoreinfo_note());
}

Since with patch 1/3 paddr_vmcoreinfo_note() returns NULL at this point in time we have a problem here.

To solve this - I think - we could move the initialization to arch/s390/kernel/machine_kexec.c:

void arch_crash_save_vmcoreinfo(void)

{
        VMCOREINFO_SYMBOL(lowcore_ptr);
        VMCOREINFO_SYMBOL(high_memory);
        VMCOREINFO_LENGTH(lowcore_ptr, NR_CPUS);
        mem_assign_absolute(S390_lowcore.vmcore_info, paddr_vmcoreinfo_note());
}

Probably related to this is my observation that patch 3/3 leads to an empty VMCOREINFO note for kdump on s390. The note is there ...

# readelf -n /var/crash/127.0.0.1-2017-03-22-21:14:39/vmcore | grep VMCORE
  VMCOREINFO           0x0000068e       Unknown note type: (0x00000000)

But it contains only zeros.

Unfortunately I have not yet understood the reason for this.

Michael

#11 lore raw
Quoted 47 lines · Michael Holzheu

Am Wed, 22 Mar 2017 12:30:04 +0800
schrieb Dave Young <[email protected]>:

Quoted 3 lines · Eric W. Biederman
Quoted 2 lines

Dave Young <[email protected]> writes:

[snip]

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

I think makedumpfile is using it, but I also vote to remove the CRASHTIME. It is better not to do this while crashing and a makedumpfile userspace patch is needed to drop the use of it.

Quoted 3 lines

As we are looking at reliability concerns removing CRASHTIME should make everything in vmcoreinfo a boot time constant. Which should simplify everything considerably.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

It is a nice improvement..

We also need to take a close look at what s390 is doing with vmcoreinfo. As apparently it is reading it in a different kind of crashdump process.

Yes, need careful review from s390 and maybe ppc64 especially about patch 2/3, better to have comments from IBM about s390 dump tool and ppc fadump. Added more cc.

On s390 we have at least an issue with patch 1/3. For stand-alone dump and also because we create the ELF header for kdump in the new kernel we save the pointer to the vmcoreinfo note in the old kernel on a defined memory address in our absolute zero lowcore.

This is done in arch/s390/kernel/setup.c:

static void __init setup_vmcoreinfo(void)

{
        mem_assign_absolute(S390_lowcore.vmcore_info, paddr_vmcoreinfo_note());
}

Since with patch 1/3 paddr_vmcoreinfo_note() returns NULL at this point in time we have a problem here.

To solve this - I think - we could move the initialization to arch/s390/kernel/machine_kexec.c:

void arch_crash_save_vmcoreinfo(void)

{
        VMCOREINFO_SYMBOL(lowcore_ptr);
        VMCOREINFO_SYMBOL(high_memory);
        VMCOREINFO_LENGTH(lowcore_ptr, NR_CPUS);
        mem_assign_absolute(S390_lowcore.vmcore_info, paddr_vmcoreinfo_note());
}
Quoted 7 lines

Probably related to this is my observation that patch 3/3 leads to an empty VMCOREINFO note for kdump on s390. The note is there ...

# readelf -n /var/crash/127.0.0.1-2017-03-22-21:14:39/vmcore | grep VMCORE
  VMCOREINFO           0x0000068e       Unknown note type: (0x00000000)

But it contains only zeros.

Yes, this is a good catch, I will do more tests.

Thanks,
Xunlei

Quoted 10 lines

Unfortunately I have not yet understood the reason for this.

Michael

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#12 lore raw

Am Thu, 23 Mar 2017 17:23:53 +0800
schrieb Xunlei Pang <[email protected]>:

Quoted 51 lines · Michael Holzheu
Quoted 50 lines

Am Wed, 22 Mar 2017 12:30:04 +0800
schrieb Dave Young <[email protected]>:

Quoted 3 lines · Eric W. Biederman
Quoted 2 lines

Dave Young <[email protected]> writes:

[snip]

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

I think makedumpfile is using it, but I also vote to remove the CRASHTIME. It is better not to do this while crashing and a makedumpfile userspace patch is needed to drop the use of it.

Quoted 3 lines

As we are looking at reliability concerns removing CRASHTIME should make everything in vmcoreinfo a boot time constant. Which should simplify everything considerably.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

It is a nice improvement..

We also need to take a close look at what s390 is doing with vmcoreinfo. As apparently it is reading it in a different kind of crashdump process.

Yes, need careful review from s390 and maybe ppc64 especially about patch 2/3, better to have comments from IBM about s390 dump tool and ppc fadump. Added more cc.

On s390 we have at least an issue with patch 1/3. For stand-alone dump and also because we create the ELF header for kdump in the new kernel we save the pointer to the vmcoreinfo note in the old kernel on a defined memory address in our absolute zero lowcore.

This is done in arch/s390/kernel/setup.c:

static void __init setup_vmcoreinfo(void)

{
        mem_assign_absolute(S390_lowcore.vmcore_info, paddr_vmcoreinfo_note());
}

Since with patch 1/3 paddr_vmcoreinfo_note() returns NULL at this point in time we have a problem here.

To solve this - I think - we could move the initialization to arch/s390/kernel/machine_kexec.c:

void arch_crash_save_vmcoreinfo(void)

{
        VMCOREINFO_SYMBOL(lowcore_ptr);
        VMCOREINFO_SYMBOL(high_memory);
        VMCOREINFO_LENGTH(lowcore_ptr, NR_CPUS);
        mem_assign_absolute(S390_lowcore.vmcore_info, paddr_vmcoreinfo_note());
}

Probably related to this is my observation that patch 3/3 leads to an empty VMCOREINFO note for kdump on s390. The note is there ...

Quoted 6 lines
Quoted 4 lines
# readelf -n /var/crash/127.0.0.1-2017-03-22-21:14:39/vmcore | grep VMCORE
  VMCOREINFO           0x0000068e       Unknown note type: (0x00000000)

But it contains only zeros.

Yes, this is a good catch, I will do more tests.

Hello Xunlei,

After spending some time on this, I now understood the problem:

In patch 3/3 you copy vmcoreinfo into the control page before machine_kexec_prepare() is called. For s390 we give back all the crashkernel memory to the hypervisor before the new crashkernel is loaded:

/*
* Give back memory to hypervisor before new kdump is loaded */
static int machine_kexec_prepare_kdump(void)
{

#ifdef CONFIG_CRASH_DUMP
        if (MACHINE_IS_VM)
                diag10_range(PFN_DOWN(crashk_res.start),
                             PFN_DOWN(crashk_res.end - crashk_res.start + 1));
        return 0;
#else
        return -EINVAL;

#endif
}

So after machine_kexec_prepare_kdump() the contents of your control page is gone and therefore the vmcorinfo ELF note contains only zeros.

If you call kimage_crash_copy_vmcoreinfo() after machine_kexec_prepare_kdump() the problem should be solved for s390.

Regards
Michael

#13 lore raw
Quoted 80 lines · Michael Holzheu

Am Thu, 23 Mar 2017 17:23:53 +0800
schrieb Xunlei Pang <[email protected]>:

Quoted 48 lines · Michael Holzheu
Quoted 47 lines

Am Wed, 22 Mar 2017 12:30:04 +0800
schrieb Dave Young <[email protected]>:

Quoted 3 lines · Eric W. Biederman
Quoted 2 lines

Dave Young <[email protected]> writes:

[snip]

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

I think makedumpfile is using it, but I also vote to remove the CRASHTIME. It is better not to do this while crashing and a makedumpfile userspace patch is needed to drop the use of it.

Quoted 3 lines

As we are looking at reliability concerns removing CRASHTIME should make everything in vmcoreinfo a boot time constant. Which should simplify everything considerably.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

It is a nice improvement..

We also need to take a close look at what s390 is doing with vmcoreinfo. As apparently it is reading it in a different kind of crashdump process.

Yes, need careful review from s390 and maybe ppc64 especially about patch 2/3, better to have comments from IBM about s390 dump tool and ppc fadump. Added more cc.

On s390 we have at least an issue with patch 1/3. For stand-alone dump and also because we create the ELF header for kdump in the new kernel we save the pointer to the vmcoreinfo note in the old kernel on a defined memory address in our absolute zero lowcore.

This is done in arch/s390/kernel/setup.c:

static void __init setup_vmcoreinfo(void)

{
        mem_assign_absolute(S390_lowcore.vmcore_info, paddr_vmcoreinfo_note());
}

Since with patch 1/3 paddr_vmcoreinfo_note() returns NULL at this point in time we have a problem here.

To solve this - I think - we could move the initialization to arch/s390/kernel/machine_kexec.c:

void arch_crash_save_vmcoreinfo(void)

{
        VMCOREINFO_SYMBOL(lowcore_ptr);
        VMCOREINFO_SYMBOL(high_memory);
        VMCOREINFO_LENGTH(lowcore_ptr, NR_CPUS);
        mem_assign_absolute(S390_lowcore.vmcore_info, paddr_vmcoreinfo_note());
}
Quoted 8 lines
Quoted 7 lines

Probably related to this is my observation that patch 3/3 leads to an empty VMCOREINFO note for kdump on s390. The note is there ...

# readelf -n /var/crash/127.0.0.1-2017-03-22-21:14:39/vmcore | grep VMCORE
  VMCOREINFO           0x0000068e       Unknown note type: (0x00000000)

But it contains only zeros.

Yes, this is a good catch, I will do more tests.

Hello Xunlei,

After spending some time on this, I now understood the problem:

In patch 3/3 you copy vmcoreinfo into the control page before machine_kexec_prepare() is called. For s390 we give back all the crashkernel memory to the hypervisor before the new crashkernel is loaded:

/*
* Give back memory to hypervisor before new kdump is loaded */
static int machine_kexec_prepare_kdump(void)
{

#ifdef CONFIG_CRASH_DUMP
        if (MACHINE_IS_VM)
                diag10_range(PFN_DOWN(crashk_res.start),
                             PFN_DOWN(crashk_res.end - crashk_res.start + 1));
        return 0;
#else
        return -EINVAL;
Quoted 8 lines

#endif
}

So after machine_kexec_prepare_kdump() the contents of your control page is gone and therefore the vmcorinfo ELF note contains only zeros.

If you call kimage_crash_copy_vmcoreinfo() after machine_kexec_prepare_kdump() the problem should be solved for s390.

Will update, thanks for finding the root cause.

Regards,
Xunlei

#14 lore raw
In reply to Eric W. Biederman
Quoted 12 lines · Eric W. Biederman
Quoted 10 lines · Xunlei Pang

As Eric said,
"what we need to do is move the variable vmcoreinfo_note out of the kernel's .bss section. And modify the code to regenerate and keep this information in something like the control page.

Definitely something like this needs a page all to itself, and ideally far away from any other kernel data structures. I clearly was not watching closely the data someone decided to keep this silly thing in the kernel's .bss section."

Quoted 6 lines
Quoted 3 lines

This patch allocates extra pages for these vmcoreinfo_XXX variables, one advantage is that it enhances some safety of vmcoreinfo, because vmcoreinfo now is kept far away from other kernel data structures.

Can you preceed this patch with a patch that removes CRASHTIME from vmcoreinfo? If someone actually cares we can add a separate note that holds a 64bit crashtime in the per cpu notes.

Hi Eric,

Thanks for your review, I took some time and did some investigation.

Removing "CRASHTIME=X" from vmcoreinfo_note will break user-space tools. For example, makedumpfile gets vmcoreinfo note information by reading "/sys/kernel/vmcoreinfo" its PA, then get its "VA = PA | PAGE_OFFSET", and then get the timestamp. This operates in the first kernel even before kdump is loaded.

Actually, even moving vmcoreinfo_note[] into the crash memory, it may have problems, for example, on s390 system the crash memory range will be unmapped, so I guess it may cause some risks.

Additionally, there is no available way for us to allocate a page from the crash memory during kernel initialization, we only can achieve this during the kexec syscalls. There is not a neat way to implement a function to allocate pages from the crash memory during kernel initialization without some hack code added, because user-space tools(like kexec-tools) can allocate the crash segment by their own ways from the crash memory.

That's why I only copy vmcoreinfo_data[] into the crash memory, and not touch vmcoreinfo_note, so vmcoreinfo_data is well protected in the crash memory copy, then in crash_save_vmcoreinfo(), we copy this guaranteed copy into vmcoreinfo_note[], so the correctness of vmcoreinfo_note[] is guaranteed. This is what [PATCH v3 3/3] does.

The current crash_save_vmcoreinfo() only involves memory(memcpy) operations even for get_seconds(no locks), the only risk I can think of now is that vmcoreinfo_note pointer may be corrupted. If it is a concern, I guess we can put it into struct kimage" just like vmcoreinfo_XXX_copy in this patch. After all if kimage structure was corrupted when crash happens, we can do nothing but have to accept the fate.

So does it really deserve to eliminate crash_save_vmcoreinfo()?

Regards,
Xunlei

Quoted 117 lines

As we are looking at reliability concerns removing CRASHTIME should make everything in vmcoreinfo a boot time constant. Which should simplify everything considerably.

Which means we only need to worry abou the per-cpu notes being written at the time of a crash.

Quoted 109 lines
---
 arch/ia64/kernel/machine_kexec.c |  5 -----
 arch/x86/kernel/crash.c          |  2 +-
 include/linux/kexec.h            |  2 +-
 kernel/kexec_core.c              | 29 ++++++++++++++++++++++++-----
 kernel/ksysfs.c                  |  2 +-
 5 files changed, 27 insertions(+), 13 deletions(-)
arch/ia64/kernel/machine_kexec.c +0 -5
index 599507b..c14815d 100644
@@ -163,8 +163,3 @@ void arch_crash_save_vmcoreinfo(void)
 #endif
 }
 
-phys_addr_t paddr_vmcoreinfo_note(void)
-{
-	return ia64_tpa((unsigned long)(char *)&vmcoreinfo_note);
-}
-
arch/x86/kernel/crash.c +1 -1
index 3741461..4d35fbb 100644
@@ -456,7 +456,7 @@ static int prepare_elf64_headers(struct crash_elf_data *ced,
 	bufp += sizeof(Elf64_Phdr);
 	phdr->p_type = PT_NOTE;
 	phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
-	phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
+	phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
 	(ehdr->e_phnum)++;
 
 #ifdef CONFIG_X86_64
include/linux/kexec.h +1 -1
index e98e546..f1c601b 100644
@@ -317,7 +317,7 @@ extern void *kexec_purgatory_get_symbol_addr(struct kimage *image,
 extern struct resource crashk_low_res;
 typedef u32 note_buf_t[KEXEC_NOTE_BYTES/4];
 extern note_buf_t __percpu *crash_notes;
-extern u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
+extern u32 *vmcoreinfo_note;
 extern size_t vmcoreinfo_size;
 extern size_t vmcoreinfo_max_size;
 
kernel/kexec_core.c +24 -5
index bfe62d5..e3a4bda 100644
@@ -52,10 +52,10 @@
 note_buf_t __percpu *crash_notes;
 
 /* vmcoreinfo stuff */
-static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
-u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
+static unsigned char *vmcoreinfo_data;
 size_t vmcoreinfo_size;
-size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
+size_t vmcoreinfo_max_size = VMCOREINFO_BYTES;
+u32 *vmcoreinfo_note;
 
 /* Flag to indicate we are going to kexec a new kernel */
 bool kexec_in_progress = false;
@@ -1369,6 +1369,9 @@ static void update_vmcoreinfo_note(void)
 
 void crash_save_vmcoreinfo(void)
 {
+	if (!vmcoreinfo_note)
+		return;
+
 	vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
 	update_vmcoreinfo_note();
 }
@@ -1397,13 +1400,29 @@ void vmcoreinfo_append_str(const char *fmt, ...)
 void __weak arch_crash_save_vmcoreinfo(void)
 {}
 
-phys_addr_t __weak paddr_vmcoreinfo_note(void)
+phys_addr_t paddr_vmcoreinfo_note(void)
 {
-	return __pa_symbol((unsigned long)(char *)&vmcoreinfo_note);
+	return __pa(vmcoreinfo_note);
 }
 
 static int __init crash_save_vmcoreinfo_init(void)
 {
+	/* One page should be enough for VMCOREINFO_BYTES under all archs */
+	vmcoreinfo_data = (unsigned char *)get_zeroed_page(GFP_KERNEL);
+	if (!vmcoreinfo_data) {
+		pr_warn("Memory allocation for vmcoreinfo_data failed\n");
+		return -ENOMEM;
+	}
+
+	vmcoreinfo_note = alloc_pages_exact(VMCOREINFO_NOTE_SIZE,
+						GFP_KERNEL | __GFP_ZERO);
+	if (!vmcoreinfo_note) {
+		free_page((unsigned long)vmcoreinfo_data);
+		vmcoreinfo_data = NULL;
+		pr_warn("Memory allocation for vmcoreinfo_note failed\n");
+		return -ENOMEM;
+	}
+
 	VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
 	VMCOREINFO_PAGESIZE(PAGE_SIZE);
 
Quoted 12 lines

kernel/ksysfs.c static ssize_t vmcoreinfo_show(struct kobject *kobj, · line 135

Quoted 12 lines
kernel/ksysfs.c +1 -1
index ee1bc1b..9de6fcc 100644
@@ -130,7 +130,7 @@ static ssize_t vmcoreinfo_show(struct kobject *kobj,
 {
 	phys_addr_t vmcore_base = paddr_vmcoreinfo_note();
 	return sprintf(buf, "%pa %x\n", &vmcore_base,
-		       (unsigned int)sizeof(vmcoreinfo_note));
+			(unsigned int)VMCOREINFO_NOTE_SIZE);
 }
 KERNEL_ATTR_RO(vmcoreinfo);

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#15 lore raw
Quoted 21 lines · Xunlei Pang
Quoted 12 lines · Eric W. Biederman
Quoted 10 lines · Xunlei Pang

As Eric said,
"what we need to do is move the variable vmcoreinfo_note out of the kernel's .bss section. And modify the code to regenerate and keep this information in something like the control page.

Definitely something like this needs a page all to itself, and ideally far away from any other kernel data structures. I clearly was not watching closely the data someone decided to keep this silly thing in the kernel's .bss section."

Quoted 6 lines
Quoted 3 lines

This patch allocates extra pages for these vmcoreinfo_XXX variables, one advantage is that it enhances some safety of vmcoreinfo, because vmcoreinfo now is kept far away from other kernel data structures.

Can you preceed this patch with a patch that removes CRASHTIME from vmcoreinfo? If someone actually cares we can add a separate note that holds a 64bit crashtime in the per cpu notes.

Hi Eric,

Quoted 7 lines

Thanks for your review, I took some time and did some investigation.

Removing "CRASHTIME=X" from vmcoreinfo_note will break user-space tools. For example, makedumpfile gets vmcoreinfo note information by reading "/sys/kernel/vmcoreinfo" its PA, then get its "VA = PA | PAGE_OFFSET", and then get the timestamp. This operates in the first kernel even before kdump is loaded.

Think more, this is not a problem for "makedumpfile --mem-usage", as the system doesn't have "CRASHTIME" before crash. But still we may have the following concerns.

Quoted 155 lines

Actually, even moving vmcoreinfo_note[] into the crash memory, it may have problems, for example, on s390 system the crash memory range will be unmapped, so I guess it may cause some risks.

Additionally, there is no available way for us to allocate a page from the crash memory during kernel initialization, we only can achieve this during the kexec syscalls. There is not a neat way to implement a function to allocate pages from the crash memory during kernel initialization without some hack code added, because user-space tools(like kexec-tools) can allocate the crash segment by their own ways from the crash memory.

That's why I only copy vmcoreinfo_data[] into the crash memory, and not touch vmcoreinfo_note, so vmcoreinfo_data is well protected in the crash memory copy, then in crash_save_vmcoreinfo(), we copy this guaranteed copy into vmcoreinfo_note[], so the correctness of vmcoreinfo_note[] is guaranteed. This is what [PATCH v3 3/3] does.

The current crash_save_vmcoreinfo() only involves memory(memcpy) operations even for get_seconds(no locks), the only risk I can think of now is that vmcoreinfo_note pointer may be corrupted. If it is a concern, I guess we can put it into struct kimage" just like vmcoreinfo_XXX_copy in this patch. After all if kimage structure was corrupted when crash happens, we can do nothing but have to accept the fate.

So does it really deserve to eliminate crash_save_vmcoreinfo()?

Regards,
Xunlei

Quoted 125 lines

As we are looking at reliability concerns removing CRASHTIME should make everything in vmcoreinfo a boot time constant. Which should simplify everything considerably.

Which means we only need to worry abou the per-cpu notes being written at the time of a crash.

Quoted 118 lines
---
 arch/ia64/kernel/machine_kexec.c |  5 -----
 arch/x86/kernel/crash.c          |  2 +-
 include/linux/kexec.h            |  2 +-
 kernel/kexec_core.c              | 29 ++++++++++++++++++++++++-----
 kernel/ksysfs.c                  |  2 +-
 5 files changed, 27 insertions(+), 13 deletions(-)
arch/ia64/kernel/machine_kexec.c +0 -5
index 599507b..c14815d 100644
@@ -163,8 +163,3 @@ void arch_crash_save_vmcoreinfo(void)
 #endif
 }
 
-phys_addr_t paddr_vmcoreinfo_note(void)
-{
-	return ia64_tpa((unsigned long)(char *)&vmcoreinfo_note);
-}
-
arch/x86/kernel/crash.c +1 -1
index 3741461..4d35fbb 100644
@@ -456,7 +456,7 @@ static int prepare_elf64_headers(struct crash_elf_data *ced,
 	bufp += sizeof(Elf64_Phdr);
 	phdr->p_type = PT_NOTE;
 	phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
-	phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
+	phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
 	(ehdr->e_phnum)++;
 
 #ifdef CONFIG_X86_64
include/linux/kexec.h +1 -1
index e98e546..f1c601b 100644
@@ -317,7 +317,7 @@ extern void *kexec_purgatory_get_symbol_addr(struct kimage *image,
 extern struct resource crashk_low_res;
 typedef u32 note_buf_t[KEXEC_NOTE_BYTES/4];
 extern note_buf_t __percpu *crash_notes;
-extern u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
+extern u32 *vmcoreinfo_note;
 extern size_t vmcoreinfo_size;
 extern size_t vmcoreinfo_max_size;
 
kernel/kexec_core.c +24 -5
index bfe62d5..e3a4bda 100644
@@ -52,10 +52,10 @@
 note_buf_t __percpu *crash_notes;
 
 /* vmcoreinfo stuff */
-static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
-u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
+static unsigned char *vmcoreinfo_data;
 size_t vmcoreinfo_size;
-size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
+size_t vmcoreinfo_max_size = VMCOREINFO_BYTES;
+u32 *vmcoreinfo_note;
 
 /* Flag to indicate we are going to kexec a new kernel */
 bool kexec_in_progress = false;
@@ -1369,6 +1369,9 @@ static void update_vmcoreinfo_note(void)
 
 void crash_save_vmcoreinfo(void)
 {
+	if (!vmcoreinfo_note)
+		return;
+
 	vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
 	update_vmcoreinfo_note();
 }
@@ -1397,13 +1400,29 @@ void vmcoreinfo_append_str(const char *fmt, ...)
 void __weak arch_crash_save_vmcoreinfo(void)
 {}
 
-phys_addr_t __weak paddr_vmcoreinfo_note(void)
+phys_addr_t paddr_vmcoreinfo_note(void)
 {
-	return __pa_symbol((unsigned long)(char *)&vmcoreinfo_note);
+	return __pa(vmcoreinfo_note);
 }
 
 static int __init crash_save_vmcoreinfo_init(void)
 {
+	/* One page should be enough for VMCOREINFO_BYTES under all archs */
+	vmcoreinfo_data = (unsigned char *)get_zeroed_page(GFP_KERNEL);
+	if (!vmcoreinfo_data) {
+		pr_warn("Memory allocation for vmcoreinfo_data failed\n");
+		return -ENOMEM;
+	}
+
+	vmcoreinfo_note = alloc_pages_exact(VMCOREINFO_NOTE_SIZE,
+						GFP_KERNEL | __GFP_ZERO);
+	if (!vmcoreinfo_note) {
+		free_page((unsigned long)vmcoreinfo_data);
+		vmcoreinfo_data = NULL;
+		pr_warn("Memory allocation for vmcoreinfo_note failed\n");
+		return -ENOMEM;
+	}
+
 	VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
 	VMCOREINFO_PAGESIZE(PAGE_SIZE);
 
kernel/ksysfs.c +0 -1
index ee1bc1b..9de6fcc 100644
@@ -130,7 +130,7 @@ static ssize_t vmcoreinfo_show(struct kobject *kobj,
 {
 	phys_addr_t vmcore_base = paddr_vmcoreinfo_note();
 	return sprintf(buf, "%pa %x\n", &vmcore_base,
-		       (unsigned int)sizeof(vmcoreinfo_note));
Quoted 8 lines
Quoted 3 lines

kernel/ksysfs.c static ssize_t vmcoreinfo_show(struct kobject *kobj, · line 135

Quoted 3 lines
kernel/ksysfs.c +1 -0
+			(unsigned int)VMCOREINFO_NOTE_SIZE);
 }
 KERNEL_ATTR_RO(vmcoreinfo);

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#16 lore raw
In reply to Xunlei Pang
Quoted 21 lines · Xunlei Pang
Quoted 12 lines · Eric W. Biederman
Quoted 10 lines · Xunlei Pang

As Eric said,
"what we need to do is move the variable vmcoreinfo_note out of the kernel's .bss section. And modify the code to regenerate and keep this information in something like the control page.

Definitely something like this needs a page all to itself, and ideally far away from any other kernel data structures. I clearly was not watching closely the data someone decided to keep this silly thing in the kernel's .bss section."

Quoted 6 lines
Quoted 3 lines

This patch allocates extra pages for these vmcoreinfo_XXX variables, one advantage is that it enhances some safety of vmcoreinfo, because vmcoreinfo now is kept far away from other kernel data structures.

Can you preceed this patch with a patch that removes CRASHTIME from vmcoreinfo? If someone actually cares we can add a separate note that holds a 64bit crashtime in the per cpu notes.

Hi Eric,

Quoted 7 lines

Thanks for your review, I took some time and did some investigation.

Removing "CRASHTIME=X" from vmcoreinfo_note will break user-space tools. For example, makedumpfile gets vmcoreinfo note information by reading "/sys/kernel/vmcoreinfo" its PA, then get its "VA = PA | PAGE_OFFSET", and then get the timestamp. This operates in the first kernel even before kdump is loaded.

Think more, this is not a problem for "makedumpfile --mem-usage", as the system doesn't have "CRASHTIME" before crash. But still we may have the following concerns.

Quoted 155 lines

Actually, even moving vmcoreinfo_note[] into the crash memory, it may have problems, for example, on s390 system the crash memory range will be unmapped, so I guess it may cause some risks.

Additionally, there is no available way for us to allocate a page from the crash memory during kernel initialization, we only can achieve this during the kexec syscalls. There is not a neat way to implement a function to allocate pages from the crash memory during kernel initialization without some hack code added, because user-space tools(like kexec-tools) can allocate the crash segment by their own ways from the crash memory.

That's why I only copy vmcoreinfo_data[] into the crash memory, and not touch vmcoreinfo_note, so vmcoreinfo_data is well protected in the crash memory copy, then in crash_save_vmcoreinfo(), we copy this guaranteed copy into vmcoreinfo_note[], so the correctness of vmcoreinfo_note[] is guaranteed. This is what [PATCH v3 3/3] does.

The current crash_save_vmcoreinfo() only involves memory(memcpy) operations even for get_seconds(no locks), the only risk I can think of now is that vmcoreinfo_note pointer may be corrupted. If it is a concern, I guess we can put it into struct kimage" just like vmcoreinfo_XXX_copy in this patch. After all if kimage structure was corrupted when crash happens, we can do nothing but have to accept the fate.

So does it really deserve to eliminate crash_save_vmcoreinfo()?

Regards,
Xunlei

Quoted 125 lines

As we are looking at reliability concerns removing CRASHTIME should make everything in vmcoreinfo a boot time constant. Which should simplify everything considerably.

Which means we only need to worry abou the per-cpu notes being written at the time of a crash.

Quoted 118 lines
---
 arch/ia64/kernel/machine_kexec.c |  5 -----
 arch/x86/kernel/crash.c          |  2 +-
 include/linux/kexec.h            |  2 +-
 kernel/kexec_core.c              | 29 ++++++++++++++++++++++++-----
 kernel/ksysfs.c                  |  2 +-
 5 files changed, 27 insertions(+), 13 deletions(-)
arch/ia64/kernel/machine_kexec.c +0 -5
index 599507b..c14815d 100644
@@ -163,8 +163,3 @@ void arch_crash_save_vmcoreinfo(void)
 #endif
 }
 
-phys_addr_t paddr_vmcoreinfo_note(void)
-{
-	return ia64_tpa((unsigned long)(char *)&vmcoreinfo_note);
-}
-
arch/x86/kernel/crash.c +1 -1
index 3741461..4d35fbb 100644
@@ -456,7 +456,7 @@ static int prepare_elf64_headers(struct crash_elf_data *ced,
 	bufp += sizeof(Elf64_Phdr);
 	phdr->p_type = PT_NOTE;
 	phdr->p_offset = phdr->p_paddr = paddr_vmcoreinfo_note();
-	phdr->p_filesz = phdr->p_memsz = sizeof(vmcoreinfo_note);
+	phdr->p_filesz = phdr->p_memsz = VMCOREINFO_NOTE_SIZE;
 	(ehdr->e_phnum)++;
 
 #ifdef CONFIG_X86_64
include/linux/kexec.h +1 -1
index e98e546..f1c601b 100644
@@ -317,7 +317,7 @@ extern void *kexec_purgatory_get_symbol_addr(struct kimage *image,
 extern struct resource crashk_low_res;
 typedef u32 note_buf_t[KEXEC_NOTE_BYTES/4];
 extern note_buf_t __percpu *crash_notes;
-extern u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
+extern u32 *vmcoreinfo_note;
 extern size_t vmcoreinfo_size;
 extern size_t vmcoreinfo_max_size;
 
kernel/kexec_core.c +24 -5
index bfe62d5..e3a4bda 100644
@@ -52,10 +52,10 @@
 note_buf_t __percpu *crash_notes;
 
 /* vmcoreinfo stuff */
-static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
-u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
+static unsigned char *vmcoreinfo_data;
 size_t vmcoreinfo_size;
-size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
+size_t vmcoreinfo_max_size = VMCOREINFO_BYTES;
+u32 *vmcoreinfo_note;
 
 /* Flag to indicate we are going to kexec a new kernel */
 bool kexec_in_progress = false;
@@ -1369,6 +1369,9 @@ static void update_vmcoreinfo_note(void)
 
 void crash_save_vmcoreinfo(void)
 {
+	if (!vmcoreinfo_note)
+		return;
+
 	vmcoreinfo_append_str("CRASHTIME=%ld\n", get_seconds());
 	update_vmcoreinfo_note();
 }
@@ -1397,13 +1400,29 @@ void vmcoreinfo_append_str(const char *fmt, ...)
 void __weak arch_crash_save_vmcoreinfo(void)
 {}
 
-phys_addr_t __weak paddr_vmcoreinfo_note(void)
+phys_addr_t paddr_vmcoreinfo_note(void)
 {
-	return __pa_symbol((unsigned long)(char *)&vmcoreinfo_note);
+	return __pa(vmcoreinfo_note);
 }
 
 static int __init crash_save_vmcoreinfo_init(void)
 {
+	/* One page should be enough for VMCOREINFO_BYTES under all archs */
+	vmcoreinfo_data = (unsigned char *)get_zeroed_page(GFP_KERNEL);
+	if (!vmcoreinfo_data) {
+		pr_warn("Memory allocation for vmcoreinfo_data failed\n");
+		return -ENOMEM;
+	}
+
+	vmcoreinfo_note = alloc_pages_exact(VMCOREINFO_NOTE_SIZE,
+						GFP_KERNEL | __GFP_ZERO);
+	if (!vmcoreinfo_note) {
+		free_page((unsigned long)vmcoreinfo_data);
+		vmcoreinfo_data = NULL;
+		pr_warn("Memory allocation for vmcoreinfo_note failed\n");
+		return -ENOMEM;
+	}
+
 	VMCOREINFO_OSRELEASE(init_uts_ns.name.release);
 	VMCOREINFO_PAGESIZE(PAGE_SIZE);
 
kernel/ksysfs.c +0 -1
index ee1bc1b..9de6fcc 100644
@@ -130,7 +130,7 @@ static ssize_t vmcoreinfo_show(struct kobject *kobj,
 {
 	phys_addr_t vmcore_base = paddr_vmcoreinfo_note();
 	return sprintf(buf, "%pa %x\n", &vmcore_base,
-		       (unsigned int)sizeof(vmcoreinfo_note));
Quoted 8 lines
Quoted 3 lines

kernel/ksysfs.c static ssize_t vmcoreinfo_show(struct kobject *kobj, · line 135

Quoted 3 lines
kernel/ksysfs.c +1 -0
+			(unsigned int)VMCOREINFO_NOTE_SIZE);
 }
 KERNEL_ATTR_RO(vmcoreinfo);

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#17 lore raw
In reply to Xunlei Pang
Quoted 5 lines · On Mon, 20 Mar 2017 13:50:31 +0800 Xunlei Pang <[email protected]> wrote:

As Eric said,
"what we need to do is move the variable vmcoreinfo_note out of the kernel's .bss section. And modify the code to regenerate and keep this information in something like the control page.

Quoted 8 lines

Definitely something like this needs a page all to itself, and ideally far away from any other kernel data structures. I clearly was not watching closely the data someone decided to keep this silly thing in the kernel's .bss section."

This patch allocates extra pages for these vmcoreinfo_XXX variables, one advantage is that it enhances some safety of vmcoreinfo, because vmcoreinfo now is kept far away from other kernel data structures.

Yes, I like this patch set very much now. Thank you!

Petr T