mm: dump_page: add debugfs file for dumping page state by pfn
Subsystems (4)
PATCH 1/1 mm: dump_page: add debugfs file for dumping page state by pfn
Tool 'page-types' could list pages mapped by process or file cache pages, but it shows only limited amount of state exported via procfs.
Let's employ existing helper dump_page() to reach remaining information:
writing pfn into /sys/kernel/debug/dump_page dumps state into kernel log.
# echo 0x37c43c > /sys/kernel/debug/dump_page # dmesg | tail -6 page:ffffcb0b0df10f00 refcount:1 mapcount:0 mapping:000000007755d3d9 index:0x30 0xffffffffae4239e0 name:"libGeoIP.so.1.6.9" flags: 0x200000000020014(uptodate|lru|mappedtodisk) raw: 0200000000020014 ffffcb0b187fd288 ffffcb0b189e6248 ffff9528a04afe10 raw: 0000000000000030 0000000000000000 00000001ffffffff 0000000000000000 page dumped because: debugfs request
With CONFIG_PAGE_OWNER=y shows also stacks for last page alloc and free:
page:ffffea0018fff480 refcount:1 mapcount:1 mapping:0000000000000000 index:0x7f9f28f62 anon flags: 0x100000000080034(uptodate|lru|active|swapbacked)
raw: 0100000000080034 ffffea00184140c8 ffffea0018517d88 ffff8886076ba161 raw: 00000007f9f28f62 0000000000000000 0000000100000000 ffff888bfc79f000 page dumped because: debugfs request page->mem_cgroup:ffff888bfc79f000 page_owner tracks the page as allocated page last allocated via order 0, migratetype Movable, gfp_mask 0x100dca(GFP_HIGHUSER_MOVABLE|__GFP_ZERO) prep_new_page+0x139/0x1a0 get_page_from_freelist+0xde9/0x14e0 __alloc_pages_nodemask+0x18b/0x360 alloc_pages_vma+0x7c/0x270 __handle_mm_fault+0xd40/0x12b0 handle_mm_fault+0xe7/0x1e0 do_page_fault+0x2d5/0x610 page_fault+0x2f/0x40 page last free stack trace: free_pcp_prepare+0x11e/0x1c0 free_unref_page_list+0x71/0x180 release_pages+0x31e/0x480 tlb_flush_mmu+0x44/0x150 tlb_finish_mmu+0x3d/0x70 exit_mmap+0xdd/0x1a0 mmput+0x70/0x140 do_exit+0x33f/0xc40 do_group_exit+0x3a/0xa0 __x64_sys_exit_group+0x14/0x20 do_syscall_64+0x48/0x130 entry_SYSCALL_64_after_hwframe+0x44/0xa9
--- Documentation/admin-guide/mm/pagemap.rst | 3 +++ Documentation/vm/page_owner.rst | 10 ++++++++++ mm/debug.c | 27 +++++++++++++++++++++++++++ 3 files changed, 40 insertions(+)
Documentation/admin-guide/mm/pagemap.rst +3 -0
@@ -205,3 +205,6 @@ Before Linux 3.11 pagemap bits 55-60 were used for "page-shift" (which is always 12 at most architectures). Since Linux 3.11 their meaning changes after first clear of soft-dirty bits. Since Linux 4.2 they are used for flags unconditionally. + +Page state could be dumped into kernel log by writing pfn in text form +into /sys/kernel/debug/dump_page.
Documentation/vm/page_owner.rst +10 -0
@@ -88,3 +88,13 @@ Usage See the result about who allocated each page in the ``sorted_page_owner.txt``. + +Notes +===== + +To lookup pages in file cache or mapped in process you could use interface +pagemap documented in Documentation/admin-guide/mm/pagemap.rst or tool +page-types in the tools/vm directory. + +Page state could be dumped into kernel log by writing pfn in text form +into /sys/kernel/debug/dump_page.
mm/debug.c +27 -0
@@ -14,6 +14,7 @@#include <linux/migrate.h> #include <linux/page_owner.h> #include <linux/ctype.h> +#include <linux/debugfs.h> #include "internal.h"
@@ -147,6 +148,32 @@ void dump_page(struct page *page, const char *reason) }EXPORT_SYMBOL (dump_page); +#ifdef CONFIG_DEBUG_FS +static int dump_page_set (void *data,u64 pfn) +{ +struct page *page; + +if (!capable (CAP_SYS_ADMIN )) +return -EPERM ; + + page =pfn_to_online_page (pfn); +if (!page) +return -ENXIO ; + +dump_page (page,"debugfs request" ); +return 0 ; +} +DEFINE_DEBUGFS_ATTRIBUTE (dump_page_fops,NULL , dump_page_set,"%llx\n" ); + +static int __init dump_page_debugfs (void ) +{ +debugfs_create_file_unsafe ("dump_page" ,0200 ,NULL ,NULL , + &dump_page_fops); +return 0 ; +} +late_initcall (dump_page_debugfs); +#endif /* CONFIG_DEBUG_FS */ +#ifdef CONFIG_DEBUG_VM void dump_vma (const struct vm_area_struct *vma)
Quoted 47 lines · Konstantin Khlebnikov
Tool 'page-types' could list pages mapped by process or file cache pages, but it shows only limited amount of state exported via procfs.
Let's employ existing helper
dump_page()to reach remaining information:
writing pfn into/sys/kernel/debug/dump_pagedumps state into kernel log.# echo 0x37c43c > /sys/kernel/debug/dump_page # dmesg | tail -6 page:ffffcb0b0df10f00 refcount:1 mapcount:0 mapping:000000007755d3d9 index:0x30 0xffffffffae4239e0 name:"libGeoIP.so.1.6.9" flags: 0x200000000020014(uptodate|lru|mappedtodisk) raw: 0200000000020014 ffffcb0b187fd288 ffffcb0b189e6248 ffff9528a04afe10 raw: 0000000000000030 0000000000000000 00000001ffffffff 0000000000000000 page dumped because: debugfs requestWith
CONFIG_PAGE_OWNER=yshows also stacks for last page alloc and free:page:ffffea0018fff480 refcount:1 mapcount:1 mapping:0000000000000000 index:0x7f9f28f62 anon flags: 0x100000000080034(uptodate|lru|active|swapbacked)
raw: 0100000000080034 ffffea00184140c8 ffffea0018517d88 ffff8886076ba161 raw: 00000007f9f28f62 0000000000000000 0000000100000000 ffff888bfc79f000 page dumped because: debugfs request page->mem_cgroup:ffff888bfc79f000 page_owner tracks the page as allocated page last allocated via order 0, migratetype Movable, gfp_mask 0x100dca(GFP_HIGHUSER_MOVABLE|__GFP_ZERO) prep_new_page+0x139/0x1a0 get_page_from_freelist+0xde9/0x14e0 __alloc_pages_nodemask+0x18b/0x360 alloc_pages_vma+0x7c/0x270 __handle_mm_fault+0xd40/0x12b0 handle_mm_fault+0xe7/0x1e0 do_page_fault+0x2d5/0x610 page_fault+0x2f/0x40 page last free stack trace: free_pcp_prepare+0x11e/0x1c0 free_unref_page_list+0x71/0x180 release_pages+0x31e/0x480 tlb_flush_mmu+0x44/0x150 tlb_finish_mmu+0x3d/0x70 exit_mmap+0xdd/0x1a0 mmput+0x70/0x140 do_exit+0x33f/0xc40 do_group_exit+0x3a/0xa0 __x64_sys_exit_group+0x14/0x20 do_syscall_64+0x48/0x130 entry_SYSCALL_64_after_hwframe+0x44/0xa9
Quoted 1 line
Signed-off-by Konstantin Khlebnikov <[email protected]>
Looks useful to me:
Signature · 1 line
Kirill A. Shutemov
Quoted 6 lines · Konstantin Khlebnikov
Tool 'page-types' could list pages mapped by process or file cache pages, but it shows only limited amount of state exported via procfs.
Let's employ existing helper
dump_page()to reach remaining information:
writing pfn into/sys/kernel/debug/dump_pagedumps state into kernel log.
Quoted 8 lines
# echo 0x37c43c > /sys/kernel/debug/dump_page # dmesg | tail -6 page:ffffcb0b0df10f00 refcount:1 mapcount:0 mapping:000000007755d3d9 index:0x30 0xffffffffae4239e0 name:"libGeoIP.so.1.6.9" flags: 0x200000000020014(uptodate|lru|mappedtodisk) raw: 0200000000020014 ffffcb0b187fd288 ffffcb0b189e6248 ffff9528a04afe10 raw: 0000000000000030 0000000000000000 00000001ffffffff 0000000000000000 page dumped because: debugfs request
This makes me deeply uncomfortable. We're using %px, and %lx (for the 'raw' lines) so we actually get to see kernel addresses. We've rationalised this in the past as being acceptable because you're already in an "assert triggered" kind of situation. Now you're adding a way for any process with CAP_SYS_ADMIN to get kernel addresses dumped into the syslog.
I think we need a different function for this, or we need to re-audit dump_page() for exposing kernel pointers, and not expose the raw data in struct page.
Quoted 23 lines · Matthew Wilcox
Quoted 6 lines · Konstantin Khlebnikov
Tool 'page-types' could list pages mapped by process or file cache pages, but it shows only limited amount of state exported via procfs.
Let's employ existing helper
dump_page()to reach remaining information:
writing pfn into/sys/kernel/debug/dump_pagedumps state into kernel log.Quoted 8 lines
# echo 0x37c43c > /sys/kernel/debug/dump_page # dmesg | tail -6 page:ffffcb0b0df10f00 refcount:1 mapcount:0 mapping:000000007755d3d9 index:0x30 0xffffffffae4239e0 name:"libGeoIP.so.1.6.9" flags: 0x200000000020014(uptodate|lru|mappedtodisk) raw: 0200000000020014 ffffcb0b187fd288 ffffcb0b189e6248 ffff9528a04afe10 raw: 0000000000000030 0000000000000000 00000001ffffffff 0000000000000000 page dumped because: debugfs requestThis makes me deeply uncomfortable. We're using %px, and %lx (for the 'raw' lines) so we actually get to see kernel addresses. We've rationalised this in the past as being acceptable because you're already in an "assert triggered" kind of situation. Now you're adding a way for any process with
CAP_SYS_ADMINto get kernel addresses dumped into the syslog.
Quoted 4 lines
I think we need a different function for this, or we need to re-audit
dump_page()for exposing kernel pointers, and not expose the raw data in struct page.
It's better to add switch for disabling paranoia if bad things happening. I.e. keep everything safe by default (or whatever sysctl/config set) and flip the switch when needed.
b
Quoted 26 lines · Konstantin Khlebnikov
Quoted 23 lines · Matthew Wilcox
Quoted 6 lines · Konstantin Khlebnikov
Tool 'page-types' could list pages mapped by process or file cache pages, but it shows only limited amount of state exported via procfs.
Let's employ existing helper
dump_page()to reach remaining information:
writing pfn into/sys/kernel/debug/dump_pagedumps state into kernel log.Quoted 8 lines
# echo 0x37c43c > /sys/kernel/debug/dump_page # dmesg | tail -6 page:ffffcb0b0df10f00 refcount:1 mapcount:0 mapping:000000007755d3d9 index:0x30 0xffffffffae4239e0 name:"libGeoIP.so.1.6.9" flags: 0x200000000020014(uptodate|lru|mappedtodisk) raw: 0200000000020014 ffffcb0b187fd288 ffffcb0b189e6248 ffff9528a04afe10 raw: 0000000000000030 0000000000000000 00000001ffffffff 0000000000000000 page dumped because: debugfs requestThis makes me deeply uncomfortable. We're using %px, and %lx (for the 'raw' lines) so we actually get to see kernel addresses. We've rationalised this in the past as being acceptable because you're already in an "assert triggered" kind of situation. Now you're adding a way for any process with
CAP_SYS_ADMINto get kernel addresses dumped into the syslog.
Quoted 8 lines
Quoted 4 lines
I think we need a different function for this, or we need to re-audit
dump_page()for exposing kernel pointers, and not expose the raw data in struct page.It's better to add switch for disabling paranoia if bad things happening. I.e. keep everything safe by default (or whatever sysctl/config set) and flip the switch when needed.
Also I'm ok to seal this interface if kernel in mode of serious paranoia.
Quoted 5 lines · Konstantin Khlebnikov
Tool 'page-types' could list pages mapped by process or file cache pages, but it shows only limited amount of state exported via procfs.
Let's employ existing helper
dump_page()to reach remaining information:
writing pfn into/sys/kernel/debug/dump_pagedumps state into kernel log.
Yeah that's indeed useful, however I'm less sure if kernel log is the proper way to extract the data. For example IIRC with the page_owner file can "seek to pfn" to dump it, although that makes it somewhat harder to use.
Or we could write pfn to one file and read the dump from another one? But that's not atomic.
Perhaps if we could do something like "cat /sys/kernel/debug/dump_page/<pfn>" without all the pfns being actually listed in the dump_page directory with "ls"? Is that possible?
Quoted 123 lines
# echo 0x37c43c > /sys/kernel/debug/dump_page # dmesg | tail -6 page:ffffcb0b0df10f00 refcount:1 mapcount:0 mapping:000000007755d3d9 index:0x30 0xffffffffae4239e0 name:"libGeoIP.so.1.6.9" flags: 0x200000000020014(uptodate|lru|mappedtodisk) raw: 0200000000020014 ffffcb0b187fd288 ffffcb0b189e6248 ffff9528a04afe10 raw: 0000000000000030 0000000000000000 00000001ffffffff 0000000000000000 page dumped because: debugfs requestWith
CONFIG_PAGE_OWNER=yshows also stacks for last page alloc and free:page:ffffea0018fff480 refcount:1 mapcount:1 mapping:0000000000000000 index:0x7f9f28f62 anon flags: 0x100000000080034(uptodate|lru|active|swapbacked)
raw: 0100000000080034 ffffea00184140c8 ffffea0018517d88 ffff8886076ba161 raw: 00000007f9f28f62 0000000000000000 0000000100000000 ffff888bfc79f000 page dumped because: debugfs request page->mem_cgroup:ffff888bfc79f000 page_owner tracks the page as allocated page last allocated via order 0, migratetype Movable, gfp_mask 0x100dca(GFP_HIGHUSER_MOVABLE|__GFP_ZERO) prep_new_page+0x139/0x1a0 get_page_from_freelist+0xde9/0x14e0 __alloc_pages_nodemask+0x18b/0x360 alloc_pages_vma+0x7c/0x270 __handle_mm_fault+0xd40/0x12b0 handle_mm_fault+0xe7/0x1e0 do_page_fault+0x2d5/0x610 page_fault+0x2f/0x40 page last free stack trace: free_pcp_prepare+0x11e/0x1c0 free_unref_page_list+0x71/0x180 release_pages+0x31e/0x480 tlb_flush_mmu+0x44/0x150 tlb_finish_mmu+0x3d/0x70 exit_mmap+0xdd/0x1a0 mmput+0x70/0x140 do_exit+0x33f/0xc40 do_group_exit+0x3a/0xa0 __x64_sys_exit_group+0x14/0x20 do_syscall_64+0x48/0x130 entry_SYSCALL_64_after_hwframe+0x44/0xa9Signed-off-by Konstantin Khlebnikov <[email protected]>--- Documentation/admin-guide/mm/pagemap.rst | 3 +++ Documentation/vm/page_owner.rst | 10 ++++++++++ mm/debug.c | 27 +++++++++++++++++++++++++++ 3 files changed, 40 insertions(+)Documentation/admin-guide/mm/pagemap.rst +3
-0@@ -205,3 +205,6 @@ Before Linux 3.11 pagemap bits 55-60 were used for "page-shift" (which is always 12 at most architectures). Since Linux 3.11 their meaning changes after first clear of soft-dirty bits. Since Linux 4.2 they are used for flags unconditionally. + +Page state could be dumped into kernel log by writing pfn in text form +into /sys/kernel/debug/dump_page.Documentation/vm/page_owner.rst +10
-0@@ -88,3 +88,13 @@ Usage See the result about who allocated each page in the ``sorted_page_owner.txt``. + +Notes +===== + +To lookup pages in file cache or mapped in process you could use interface +pagemap documented in Documentation/admin-guide/mm/pagemap.rst or tool +page-types in the tools/vm directory. + +Page state could be dumped into kernel log by writing pfn in text form +into /sys/kernel/debug/dump_page.mm/debug.c +27
-0@@ -14,6 +14,7 @@#include <linux/migrate.h> #include <linux/page_owner.h> #include <linux/ctype.h> +#include <linux/debugfs.h> #include "internal.h" @@ -147,6 +148,32 @@ void dump_page(struct page *page, const char *reason) }EXPORT_SYMBOL (dump_page); +#ifdef CONFIG_DEBUG_FS +static int dump_page_set (void *data,u64 pfn) +{ +struct page *page; + +if (!capable (CAP_SYS_ADMIN )) +return -EPERM ; + + page =pfn_to_online_page (pfn); +if (!page) +return -ENXIO ; + +dump_page (page,"debugfs request" ); +return 0 ; +} +DEFINE_DEBUGFS_ATTRIBUTE (dump_page_fops,NULL , dump_page_set,"%llx\n" ); + +static int __init dump_page_debugfs (void ) +{ +debugfs_create_file_unsafe ("dump_page" ,0200 ,NULL ,NULL , + &dump_page_fops); +return 0 ; +} +late_initcall (dump_page_debugfs); +#endif /* CONFIG_DEBUG_FS */ +#ifdef CONFIG_DEBUG_VM void dump_vma (const struct vm_area_struct *vma)
Quoted 11 lines · Vlastimil Babka
Quoted 5 lines · Konstantin Khlebnikov
Tool 'page-types' could list pages mapped by process or file cache pages, but it shows only limited amount of state exported via procfs.
Let's employ existing helper
dump_page()to reach remaining information:
writing pfn into/sys/kernel/debug/dump_pagedumps state into kernel log.Yeah that's indeed useful, however I'm less sure if kernel log is the proper way to extract the data. For example IIRC with the
page_ownerfile can "seek to pfn" to dump it, although that makes it somewhat harder to use.
Quoted 6 lines
Or we could write pfn to one file and read the dump from another one? But that's not atomic.
Perhaps if we could do something like "cat /sys/kernel/debug/dump_page/<pfn>" without all the pfns being actually listed in the
dump_pagedirectory with "ls"? Is that possible?
Too much code for me. =)
This could be kind of ftrace tracer which iterates over pages and dumps them, but anyway looks ridiculously overengineered.
This one hack connects existing 'pagemap' with existing 'dump_page', so almost free.
For complicated cases there is gdb and special tool drgn https://github.com/osandov/drgn
Writing script which parses all that stuff from kernel log isn't big deal either. I have one with 100+ lines regexp for all kinds of kernel splats. Will publish when find time for polishing.
Quoted 125 lines
Quoted 124 lines
# echo 0x37c43c > /sys/kernel/debug/dump_page # dmesg | tail -6 page:ffffcb0b0df10f00 refcount:1 mapcount:0 mapping:000000007755d3d9 index:0x30 0xffffffffae4239e0 name:"libGeoIP.so.1.6.9" flags: 0x200000000020014(uptodate|lru|mappedtodisk) raw: 0200000000020014 ffffcb0b187fd288 ffffcb0b189e6248 ffff9528a04afe10 raw: 0000000000000030 0000000000000000 00000001ffffffff 0000000000000000 page dumped because: debugfs requestWith
CONFIG_PAGE_OWNER=yshows also stacks for last page alloc and free:page:ffffea0018fff480 refcount:1 mapcount:1 mapping:0000000000000000 index:0x7f9f28f62 anon flags: 0x100000000080034(uptodate|lru|active|swapbacked)
raw: 0100000000080034 ffffea00184140c8 ffffea0018517d88 ffff8886076ba161 raw: 00000007f9f28f62 0000000000000000 0000000100000000 ffff888bfc79f000 page dumped because: debugfs request page->mem_cgroup:ffff888bfc79f000 page_owner tracks the page as allocated page last allocated via order 0, migratetype Movable, gfp_mask 0x100dca(GFP_HIGHUSER_MOVABLE|__GFP_ZERO) prep_new_page+0x139/0x1a0 get_page_from_freelist+0xde9/0x14e0 __alloc_pages_nodemask+0x18b/0x360 alloc_pages_vma+0x7c/0x270 __handle_mm_fault+0xd40/0x12b0 handle_mm_fault+0xe7/0x1e0 do_page_fault+0x2d5/0x610 page_fault+0x2f/0x40 page last free stack trace: free_pcp_prepare+0x11e/0x1c0 free_unref_page_list+0x71/0x180 release_pages+0x31e/0x480 tlb_flush_mmu+0x44/0x150 tlb_finish_mmu+0x3d/0x70 exit_mmap+0xdd/0x1a0 mmput+0x70/0x140 do_exit+0x33f/0xc40 do_group_exit+0x3a/0xa0 __x64_sys_exit_group+0x14/0x20 do_syscall_64+0x48/0x130 entry_SYSCALL_64_after_hwframe+0x44/0xa9Signed-off-by Konstantin Khlebnikov <[email protected]>--- Documentation/admin-guide/mm/pagemap.rst | 3 +++ Documentation/vm/page_owner.rst | 10 ++++++++++ mm/debug.c | 27 +++++++++++++++++++++++++++ 3 files changed, 40 insertions(+)Documentation/admin-guide/mm/pagemap.rst +3
-0@@ -205,3 +205,6 @@ Before Linux 3.11 pagemap bits 55-60 were used for "page-shift" (which is always 12 at most architectures). Since Linux 3.11 their meaning changes after first clear of soft-dirty bits. Since Linux 4.2 they are used for flags unconditionally. + +Page state could be dumped into kernel log by writing pfn in text form +into /sys/kernel/debug/dump_page.Documentation/vm/page_owner.rst +10
-0@@ -88,3 +88,13 @@ Usage See the result about who allocated each page in the ``sorted_page_owner.txt``. + +Notes +===== + +To lookup pages in file cache or mapped in process you could use interface +pagemap documented in Documentation/admin-guide/mm/pagemap.rst or tool +page-types in the tools/vm directory. + +Page state could be dumped into kernel log by writing pfn in text form +into /sys/kernel/debug/dump_page.mm/debug.c +27
-0@@ -14,6 +14,7 @@#include <linux/migrate.h> #include <linux/page_owner.h> #include <linux/ctype.h> +#include <linux/debugfs.h> #include "internal.h" @@ -147,6 +148,32 @@ void dump_page(struct page *page, const char *reason) }EXPORT_SYMBOL (dump_page); +#ifdef CONFIG_DEBUG_FS +static int dump_page_set (void *data,u64 pfn) +{ +struct page *page; + +if (!capable (CAP_SYS_ADMIN )) +return -EPERM ; + + page =pfn_to_online_page (pfn); +if (!page) +return -ENXIO ; + +dump_page (page,"debugfs request" ); +return 0 ; +} +DEFINE_DEBUGFS_ATTRIBUTE (dump_page_fops,NULL , dump_page_set,"%llx\n" ); + +static int __init dump_page_debugfs (void ) +{ +debugfs_create_file_unsafe ("dump_page" ,0200 ,NULL ,NULL , + &dump_page_fops); +return 0 ; +} +late_initcall (dump_page_debugfs); +#endif /* CONFIG_DEBUG_FS */ +#ifdef CONFIG_DEBUG_VM void dump_vma (const struct vm_area_struct *vma)