Re: [PATCH 1/2] diskdump/netdump: fix segmentation fault caused by failure of stopping CPUs

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Hatayama-san,

thank you for the patches.

On 2023/05/26 19:45, HATAYAMA Daisuke wrote:
> There's no NMI on ARM. Hence, stopping the non-panicking CPUs from the
> panicking CPU via IPI can fail easily if interrupts are being masked
> in those moment. Moreover, crash_notes are not initialized for such
> unstopped CPUs and the corresponding NT_PRSTATUS notes are not
> attached to vmcore. However, crash utility never takes it
> consideration such uninitialized crash_notes and then ends with
> mapping different NT_PRSTATUS to actually unstopped CPUs. This corrupt
> mapping can result crash utility into segmentation fault in the
> operations where register values in NT_PRSTATUS notes are used.
> 
> For example:
> 
>      crash> bt 1408
> 	PID: 1408     TASK: ffff000003e22200  CPU: 2    COMMAND: "repro"
> 	Segmentation fault (core dumped)
> 
> 	crash> help -D
> 	diskdump_data:
> 			  filename: 127.0.0.1-2023-05-26-02:21:27/vmcore-ld1
> 				 flags: 46 (KDUMP_CMPRS_LOCAL|ERROR_EXCLUDED|LZO_SUPPORTED)
> 	...snip...
> 			   notes_buf: 1815df0
> 	  num_vmcoredd_notes: 0
> 	  num_prstatus_notes: 5
> 				notes[0]: 1815df0 (NT_PRSTATUS)
> 						  si.signo: 0  si.code: 0  si.errno: 0
> 	...snip...
> 						  PSTATE: 80400005   FPVALID: 00000000
> 				notes[4]: 1808f10 (NT_PRSTATUS)
> 	Segmentation fault (core dumped)
> 
> To fix this issue, let's map NT_PRSTATUS to some CPU only if the
> corresponding crash_notes is checked to be initialized.
> 
> Signed-off-by: HATAYAMA Daisuke <d.hatayama@xxxxxxxxxxx>
> ---
>   defs.h     |  1 +
>   diskdump.c | 51 ++++++++++++++++++++++++++++++++++++++++++++++++++-
>   netdump.c  |  5 ++++-
>   3 files changed, 55 insertions(+), 2 deletions(-)
> 
> diff --git a/defs.h b/defs.h
> index 12ad6aa..72129a1 100644
> --- a/defs.h
> +++ b/defs.h
> @@ -7111,6 +7111,7 @@ int dumpfile_is_split(void);
>   void show_split_dumpfiles(void);
>   void x86_process_elf_notes(void *, unsigned long);
>   void *diskdump_get_prstatus_percpu(int);
> +int have_crash_notes(int cpu);
>   void map_cpus_to_prstatus_kdump_cmprs(void);
>   void diskdump_display_regs(int, FILE *);
>   void process_elf32_notes(void *, ulong);
> diff --git a/diskdump.c b/diskdump.c
> index cf5f5d9..11d29d3 100644
> --- a/diskdump.c
> +++ b/diskdump.c
> @@ -101,6 +101,55 @@ int dumpfile_is_split(void)
>   	return KDUMP_SPLIT();
>   }
>   
> +int have_crash_notes(int cpu)
> +{
> +	ulong crash_notes, notes_ptr;
> +	char *buf, *p;
> +	Elf64_Nhdr *note = NULL;
> +
> +	if (!readmem(symbol_value("crash_notes"),
> +		     KVADDR,
> +		     &crash_notes,
> +		     sizeof(crash_notes),
> +		     "crash_notes",
> +		     RETURN_ON_ERROR)) {
> +		error(WARNING, "cannot read \"crash_notes\"\n");
> +		return FALSE;
> +	}
> +
> +	if (symbol_exists("__per_cpu_offset"))

According to the crash sources, I think this can be replaced with:

     if ((kt->flags & SMP) && (kt->flags & PER_CPU_OFF))

> +		notes_ptr = crash_notes + kt->__per_cpu_offset[cpu];
> +	else
> +		notes_ptr = crash_notes;
> +
> +	buf = GETBUF(SIZE(note_buf));
> +
> +	if (!readmem(notes_ptr,
> +		     KVADDR,
> +		     buf,
> +		     SIZE(note_buf),
> +		     "note_buf_t",
> +		     RETURN_ON_ERROR)) {
> +		error(WARNING, "cpu %d: cannot read NT_PRSTATUS note\n", cpu);
> +		return FALSE;
> +	}
> +
> +	note = (Elf64_Nhdr *)buf;
> +	p = buf + sizeof(Elf64_Nhdr);
> +
> +	if (note->n_type != NT_PRSTATUS) {
> +		error(WARNING, "cpu %d: invalid NT_PRSTATUS note (n_type != NT_PRSTATUS)\n", cpu);
> +		return FALSE;
> +	}
> +
> +	if (!STRNEQ(p, "CORE")) {
> +		error(WARNING, "cpu %d: invalid NT_PRSTATUS note (name != \"CORE\")\n", cpu);
> +		return FALSE;
> +	}
> +
> +	return TRUE;
> +}
> +
>   void
>   map_cpus_to_prstatus_kdump_cmprs(void)
>   {
> @@ -131,7 +180,7 @@ map_cpus_to_prstatus_kdump_cmprs(void)
>   	nrcpus = (kt->kernel_NR_CPUS ? kt->kernel_NR_CPUS : NR_CPUS);
>   
>   	for (i = 0, j = 0; i < nrcpus; i++) {
> -		if (in_cpu_map(ONLINE_MAP, i)) {
> +		if (in_cpu_map(ONLINE_MAP, i) && have_crash_notes(i)) {

I've seen a user using a kernel without the crash_notes symbol before 
e.g. [1].  It might be better to call have_crash_notes() only if it exists?

[1] 
https://github.com/crash-utility/crash/commit/4b34197508578bb43639e6d169fb91fb0489fa2b

Thanks,
Kazu

>   			dd->nt_prstatus_percpu[i] = nt_ptr[j++];
>   			dd->num_prstatus_notes =
>   				MAX(dd->num_prstatus_notes, i+1);
> diff --git a/netdump.c b/netdump.c
> index 01af145..b272984 100644
> --- a/netdump.c
> +++ b/netdump.c
> @@ -99,8 +99,11 @@ map_cpus_to_prstatus(void)
>   	nrcpus = (kt->kernel_NR_CPUS ? kt->kernel_NR_CPUS : NR_CPUS);
>   
>   	for (i = 0, j = 0; i < nrcpus; i++) {
> -		if (in_cpu_map(ONLINE_MAP, i))
> +		if (in_cpu_map(ONLINE_MAP, i) && have_crash_notes(i)) {
>   			nd->nt_prstatus_percpu[i] = nt_ptr[j++];
> +			nd->num_prstatus_notes =
> +				MAX(nd->num_prstatus_notes, i+1);
> +		}
>   	}
>   
>   	FREEBUF(nt_ptr);
--
Crash-utility mailing list
Crash-utility@xxxxxxxxxx
https://listman.redhat.com/mailman/listinfo/crash-utility
Contribution Guidelines: https://github.com/crash-utility/crash/wiki




[Index of Archives]     [Fedora Development]     [Fedora Desktop]     [Fedora SELinux]     [Yosemite News]     [KDE Users]     [Fedora Tools]

 

Powered by Linux