Skip to contentSkip to conversation

makedumpfile: Fix off-by-one issue in exclude_nodata_pages()

PATCH In review Mikhail Zaslonko 2 participants 2 messages Nov 4, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 makedumpfile: Fix off-by-one issue in exclude_nodata_pages()

When building a dump bitmap (2nd bitmap) for the ELF dump, the last pfn of the cycle is always ignored in exclude_nodata_pages() function due to off-by-one error on cycle boundary check. Thus, the respective bit of the bitmap is never cleared.
That can lead to the error when such a pfn should not be dumpable (e.g. the last pfn of the ELF-load of zero filesize). Based on the bit in the bitmap the page is treated as dumpable in write_elf_pages_cyclic() function and the follow on error is triggered in write_elf_load_segment() function due to the failing sanity check of paddr_to_offset2():

   $ makedumpfile -E dump.elf dump.elf.E
   Checking for memory holes                         : [100.0 %] |
   write_elf_load_segment: Can't convert physaddr(7ffff000) to an offset.
   makedumpfile Failed.
---
 makedumpfile.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
makedumpfile.c +1 -1
index de0973f..4a00011 100644
@@ -4740,7 +4740,7 @@ exclude_nodata_pages(struct cycle *cycle)
 		if (pfn < cycle->start_pfn)
 			pfn = cycle->start_pfn;
 		if (pfn_end >= cycle->end_pfn)
-			pfn_end = cycle->end_pfn - 1;
+			pfn_end = cycle->end_pfn;
 		while (pfn < pfn_end) {
 			clear_bit_on_2nd_bitmap(pfn, cycle);
 			++pfn;
Signature · 7 lines
2.17.1


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

Hi Mikhail,

Quoted 10 lines

-----Original Message-----
When building a dump bitmap (2nd bitmap) for the ELF dump, the last pfn of the cycle is always ignored in exclude_nodata_pages() function due to off-by-one error on cycle boundary check. Thus, the respective bit of the bitmap is never cleared.
That can lead to the error when such a pfn should not be dumpable (e.g. the last pfn of the ELF-load of zero filesize). Based on the bit in the bitmap the page is treated as dumpable in write_elf_pages_cyclic() function and the follow on error is triggered in write_elf_load_segment() function due to the failing sanity check of paddr_to_offset2():

Good catch. I could reproduce this issue with a vmcore on hand and confirm that the patch fixed it. Looks good to me, applied.

Thanks,
Kazu

Quoted 28 lines
   $ makedumpfile -E dump.elf dump.elf.E
   Checking for memory holes                         : [100.0 %] |
   write_elf_load_segment: Can't convert physaddr(7ffff000) to an offset.
   makedumpfile Failed.
---
 makedumpfile.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
makedumpfile.c +1 -2
index de0973f..4a00011 100644
@@ -4740,7 +4740,7 @@ exclude_nodata_pages(struct cycle *cycle)
 		if (pfn < cycle->start_pfn)
 			pfn = cycle->start_pfn;
 		if (pfn_end >= cycle->end_pfn)
-			pfn_end = cycle->end_pfn - 1;
+			pfn_end = cycle->end_pfn;
 		while (pfn < pfn_end) {
 			clear_bit_on_2nd_bitmap(pfn, cycle);
 			++pfn;
--

2.17.1

Quoted 4 lines

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

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