From: Jiri Olsa <jolsa@kernel.org> Date: 2021-01-19 22:18:58
hi,
kpatch guys hit an issue with pahole over their vmlinux, which
contains many (over 100000) sections, pahole crashes.
With so many sections, ELF is using extended section index table,
which is used to hold values for some of the indexes and extra
code is needed to retrieve them.
This patchset adds the support for pahole to properly read string
table index and symbol's section index, which are used in btf_encoder.
This patchset also adds support for libbpf to properly parse .BTF
section on such object.
This patchset based on previously posted fix [1].
thanks,
jirka
[1] https://lore.kernel.org/bpf/20210113102509.1338601-1-jolsa@kernel.org/
---
dwarves:
Jiri Olsa (2):
elf_symtab: Add support for SHN_XINDEX index to elf_section_by_name
bpf_encoder: Translate SHN_XINDEX in symbol's st_shndx values
btf_encoder.c | 18 ++++++++++++++++++
dutil.c | 8 ++++++--
elf_symtab.c | 31 ++++++++++++++++++++++++++++++-
elf_symtab.h | 1 +
4 files changed, 55 insertions(+), 3 deletions(-)
libbpf:
Jiri Olsa (1):
libbpf: Use string table index from index table if needed
tools/lib/bpf/btf.c | 14 ++++++++++++--
1 file changed, 12 insertions(+), 2 deletions(-)
From: Jiri Olsa <jolsa@kernel.org> Date: 2021-01-19 22:13:46
For very large ELF objects (with many sections), we could
get special value SHN_XINDEX (65535) for elf object's string
table index - e_shstrndx.
In such case we need to call elf_getshdrstrndx to get the
proper string table index.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
tools/lib/bpf/btf.c | 14 ++++++++++++--
1 file changed, 12 insertions(+), 2 deletions(-)
@@ -863,6 +863,7 @@ static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,Elf_Scn*scn=NULL;Elf*elf=NULL;GElf_Ehdrehdr;+size_tshstrndx;if(elf_version(EV_CURRENT)==EV_NONE){pr_warn("failed to init libelf for %s\n",path);
@@ -887,7 +888,16 @@ static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,pr_warn("failed to get EHDR from %s\n",path);gotodone;}-if(!elf_rawdata(elf_getscn(elf,ehdr.e_shstrndx),NULL)){++/*+*Getstringtableindexfromextendedsectionindex+*tableifneeded.+*/+shstrndx=ehdr.e_shstrndx;+if(shstrndx==SHN_XINDEX&&elf_getshdrstrndx(elf,&shstrndx))+gotodone;++if(!elf_rawdata(elf_getscn(elf,shstrndx),NULL)){pr_warn("failed to get e_shstrndx from %s\n",path);gotodone;}
@@ -902,7 +912,7 @@ static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,idx,path);gotodone;}-name=elf_strptr(elf,ehdr.e_shstrndx,sh.sh_name);+name=elf_strptr(elf,shstrndx,sh.sh_name);if(!name){pr_warn("failed to get section(%d) name from %s\n",idx,path);
From: Jiri Olsa <jolsa@kernel.org> Date: 2021-01-19 22:15:26
For very large ELF objects (with many sections), we could
get special value SHN_XINDEX (65535) for symbol's st_shndx.
This patch is adding code to detect the optional extended
section index table and use it to resolve symbol's section
index id needed.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
btf_encoder.c | 18 ++++++++++++++++++
elf_symtab.c | 31 ++++++++++++++++++++++++++++++-
elf_symtab.h | 1 +
3 files changed, 49 insertions(+), 1 deletion(-)
@@ -609,6 +609,24 @@ static int collect_symbols(struct btf_elf *btfe, bool collect_percpu_vars)/* search within symtab for percpu variables */elf_symtab__for_each_symbol(btfe->symtab,core_id,sym){+structelf_symtab*symtab=btfe->symtab;++/*+*Symbol'sst_shndxneedstobetranslatedwiththe+*extendedsectionindextable.+*/+if(sym.st_shndx==SHN_XINDEX){+Elf32_Wordxndx;++if(!symtab->syms_shndx){+fprintf(stderr,"SHN_XINDEX found, but no symtab shndx section.\n");+return-1;+}++if(gelf_getsymshndx(symtab->syms,symtab->syms_shndx,core_id,&sym,&xndx))+sym.st_shndx=xndx;+}+if(collect_percpu_vars&&collect_percpu_var(btfe,&sym))return-1;if(collect_function(btfe,&sym))
@@ -49,6 +53,31 @@ struct elf_symtab *elf_symtab__new(const char *name, Elf *elf, GElf_Ehdr *ehdr)if(symtab->symstrs==NULL)gotoout_free_name;+/*+*The.symtabsectionhasoptionalextendedsectionindex+*table,loaditsdatasoitcanbeusedtoresolvesymbol's+*sectionindex.+**/+if(xindex>0){+GElf_Shdrshdr_shndx;+Elf_Scn*sec_shndx;++sec_shndx=elf_getscn(elf,xindex);+if(sec_shndx==NULL)+gotoout_free_name;++if(gelf_getshdr(sec_shndx,&shdr_shndx)==NULL)+gotoout_free_name;++/* Extra check to verify it belongs to the .symtab */+if(index!=shdr_shndx.sh_link)+gotoout_free_name;++symtab->syms_shndx=elf_getdata(elf_getscn(elf,xindex),NULL);+if(symtab->syms_shndx==NULL)+gotoout_free_name;+}+symtab->nr_syms=shdr.sh_size/shdr.sh_entsize;returnsymtab;
From: Jiri Olsa <jolsa@kernel.org> Date: 2021-01-19 22:16:53
In case the elf's header e_shstrndx contains SHN_XINDEX,
we need to call elf_getshdrstrndx to get the proper
string table index.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
dutil.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
From: Joe Lawrence <joe.lawrence@redhat.com> Date: 2021-01-19 23:20:42
On Tue, Jan 19, 2021 at 11:12:17PM +0100, Jiri Olsa wrote:
hi,
kpatch guys hit an issue with pahole over their vmlinux, which
contains many (over 100000) sections, pahole crashes.
FWIW this is probably only going to be problem when building the kernel
with -f[function|data]-sections and tipping over 65536 sections. We
only use these option to determine code deltas, but other users
(FG-ASLR, LTO?) may need to actually build runtime code.
With so many sections, ELF is using extended section index table,
which is used to hold values for some of the indexes and extra
code is needed to retrieve them.
This patchset adds the support for pahole to properly read string
table index and symbol's section index, which are used in btf_encoder.
This patchset also adds support for libbpf to properly parse .BTF
section on such object.
This patchset based on previously posted fix [1].
Hi Jiri,
Thanks for posting a potential fix, here's what I saw when running it:
1-Installed your scratch build:
% rpm -q --whatprovides $(which pahole)
dwarves-1.19-2.el8.x86_64
2-From the kernel build:
LD vmlinux.o
MODPOST vmlinux.o
BTF .btf.vmlinux.bin.o
scripts/link-vmlinux.sh: line 127: 1851330 Segmentation fault (core dumped) LLVM_OBJCOPY=${OBJCOPY} ${PAHOLE} -J ${1}
objcopy: --change-section-vma .BTF=0x0000000000000000 never used
objcopy: --change-section-lma .BTF=0x0000000000000000 never used
objcopy: error: the input file '.btf.vmlinux.bin' is empty
Failed to generate BTF for vmlinux
Try to disable CONFIG_DEBUG_INFO_BTF
make: *** [Makefile:1050: vmlinux] Error 1
3-coredump backtrace:
...
Core was generated by `pahole -J .tmp_vmlinux.btf'.
...
(gdb) bt
#0 0x00007fc0e81e31c0 in __memcpy_ssse3 () from /lib64/libc.so.6
#1 0x00007fc0e8b6700a in memcpy (__len=306248, __src=<optimized out>, __dest=0x7fc0e91a0010) at /usr/include/bits/string_fortified.h:34
#2 get_vmlinux_addrs (btfe=<optimized out>, pcount=<synthetic pointer>, paddrs=<synthetic pointer>, fl=<synthetic pointer>)
at /usr/src/debug/dwarves-1.19-2.el8.x86_64/btf_encoder.c:167
#3 setup_functions (fl=<synthetic pointer>, btfe=<optimized out>) at /usr/src/debug/dwarves-1.19-2.el8.x86_64/btf_encoder.c:251
#4 collect_symbols (collect_percpu_vars=<optimized out>, btfe=<optimized out>)
at /usr/src/debug/dwarves-1.19-2.el8.x86_64/btf_encoder.c:645
#5 cu__encode_btf (cu=<optimized out>, verbose=0, force=false, skip_encoding_vars=<optimized out>)
at /usr/src/debug/dwarves-1.19-2.el8.x86_64/btf_encoder.c:694
#6 0x000055eb2e4b6cc5 in pahole_stealer (cu=0x55eb2ecbb920, conf_load=<optimized out>)
at /usr/src/debug/dwarves-1.19-2.el8.x86_64/pahole.c:2402
#7 0x00007fc0e8b6d2db in finalize_cu_immediately (conf=0x55eb2e6bc0e0 <conf_load>, dcu=0x7ffd88fda9d0, cu=0x55eb2ecbb920,
cus=0x55eb2ecbb5d0) at /usr/src/debug/dwarves-1.19-2.el8.x86_64/dwarf_loader.c:2477
#8 cus__load_module (cus=cus@entry=0x55eb2ecbb5d0, conf=0x55eb2e6bc0e0 <conf_load>, mod=mod@entry=0x55eb2ecbb5f0, dw=0x55eb2ecbe6a0,
elf=elf@entry=0x7fc0ad941010, filename=0x7ffd8905b98d ".tmp_vmlinux.btf")
at /usr/src/debug/dwarves-1.19-2.el8.x86_64/dwarf_loader.c:2477
#9 0x00007fc0e8b6d5c5 in cus__process_dwflmod (dwflmod=0x55eb2ecbb5f0, userdata=<optimized out>, name=<optimized out>,
base=<optimized out>, arg=0x7ffd8905ab00) at /usr/src/debug/dwarves-1.19-2.el8.x86_64/dwarf_loader.c:2522
#10 0x00007fc0e88f9c71 in dwfl_getmodules () from /lib64/libdw.so.1
#11 0x00007fc0e8b69cdc in cus__process_file (filename=<optimized out>, fd=5, conf=0x55eb2e6bc0e0 <conf_load>, cus=0x55eb2ecbb5d0)
at /usr/src/debug/dwarves-1.19-2.el8.x86_64/dwarf_loader.c:2575
#12 dwarf__load_file (cus=0x55eb2ecbb5d0, conf=0x55eb2e6bc0e0 <conf_load>, filename=<optimized out>)
at /usr/src/debug/dwarves-1.19-2.el8.x86_64/dwarf_loader.c:2592
#13 0x00007fc0e8b5cb82 in cus__load_file (cus=cus@entry=0x55eb2ecbb5d0, conf=conf@entry=0x55eb2e6bc0e0 <conf_load>,
filename=0x7ffd8905b98d ".tmp_vmlinux.btf") at /usr/src/debug/dwarves-1.19-2.el8.x86_64/dwarves.c:1963
#14 0x00007fc0e8b5ce29 in cus__load_files (cus=0x55eb2ecbb5d0, conf=0x55eb2e6bc0e0 <conf_load>, filenames=0x7ffd8905aea8)
at /usr/src/debug/dwarves-1.19-2.el8.x86_64/dwarves.c:2324
#15 0x000055eb2e4b371e in main (argc=3, argv=0x7ffd8905ae98) at /usr/src/debug/dwarves-1.19-2.el8.x86_64/pahole.c:2760
I uploaded a gzipped core file here:
http://people.redhat.com/~jolawren/coredump.gz
If it's easier to get you setup on a repro system, let me know and I can
do that.
Thanks,
-- Joe
On Tue, Jan 19, 2021 at 2:15 PM Jiri Olsa [off-list ref] wrote:
quoted hunk
For very large ELF objects (with many sections), we could
get special value SHN_XINDEX (65535) for elf object's string
table index - e_shstrndx.
In such case we need to call elf_getshdrstrndx to get the
proper string table index.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
tools/lib/bpf/btf.c | 14 ++++++++++++--
1 file changed, 12 insertions(+), 2 deletions(-)
On Tue, Jan 19, 2021 at 2:16 PM Jiri Olsa [off-list ref] wrote:
quoted hunk
In case the elf's header e_shstrndx contains SHN_XINDEX,
we need to call elf_getshdrstrndx to get the proper
string table index.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
dutil.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
On Tue, Jan 19, 2021 at 2:16 PM Jiri Olsa [off-list ref] wrote:
quoted hunk
For very large ELF objects (with many sections), we could
get special value SHN_XINDEX (65535) for symbol's st_shndx.
This patch is adding code to detect the optional extended
section index table and use it to resolve symbol's section
index id needed.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
btf_encoder.c | 18 ++++++++++++++++++
elf_symtab.c | 31 ++++++++++++++++++++++++++++++-
elf_symtab.h | 1 +
3 files changed, 49 insertions(+), 1 deletion(-)
@@ -609,6 +609,24 @@ static int collect_symbols(struct btf_elf *btfe, bool collect_percpu_vars)/* search within symtab for percpu variables */elf_symtab__for_each_symbol(btfe->symtab,core_id,sym){
How about we introduce elf_symtab__for_each_symbol variant that uses
gelf_getsymshndx undercover and returns a real section index as the
4th macro argument?
+ struct elf_symtab *symtab = btfe->symtab;
+
+ /*
+ * Symbol's st_shndx needs to be translated with the
+ * extended section index table.
+ */
+ if (sym.st_shndx == SHN_XINDEX) {
+ Elf32_Word xndx;
+
+ if (!symtab->syms_shndx) {
+ fprintf(stderr, "SHN_XINDEX found, but no symtab shndx section.\n");
+ return -1;
+ }
+
+ if (gelf_getsymshndx(symtab->syms, symtab->syms_shndx, core_id, &sym, &xndx))
+ sym.st_shndx = xndx;
This bit makes me really nervous and it looks very unclean, which is
why I think providing correct section index as part of iterator macro
is better approach. And all this code would just go away.
quoted hunk
+ }
+
if (collect_percpu_vars && collect_percpu_var(btfe, &sym))
return -1;
if (collect_function(btfe, &sym))
move this closer to where you check (xindex > 0)? Can you please leave
a small comment that this returns extended section index table's
section index (I know, this is horrible), if it exists. It's
notoriously hard to find anything about libelf's API, especially for
obscure APIs like this.
@@ -49,6 +53,31 @@ struct elf_symtab *elf_symtab__new(const char *name, Elf *elf, GElf_Ehdr *ehdr) if (symtab->symstrs == NULL) goto out_free_name;+ /*+ * The .symtab section has optional extended section index+ * table, load its data so it can be used to resolve symbol's+ * section index.+ **/+ if (xindex > 0) {+ GElf_Shdr shdr_shndx;+ Elf_Scn *sec_shndx;++ sec_shndx = elf_getscn(elf, xindex);+ if (sec_shndx == NULL)+ goto out_free_name;++ if (gelf_getshdr(sec_shndx, &shdr_shndx) == NULL)+ goto out_free_name;++ /* Extra check to verify it belongs to the .symtab */+ if (index != shdr_shndx.sh_link)+ goto out_free_name;
you can also verify that section type is SHT_SYMTAB_SHNDX
my mind breaks looking at all those shndxs, especially in this case
where it's not a section number, but rather data. How about we call it
something like symtab->syms_sec_idx_table or something similar but
human-comprehensible?
From: Jiri Olsa <hidden> Date: 2021-01-20 12:40:01
On Tue, Jan 19, 2021 at 05:23:00PM -0800, Andrii Nakryiko wrote:
On Tue, Jan 19, 2021 at 2:16 PM Jiri Olsa [off-list ref] wrote:
quoted
In case the elf's header e_shstrndx contains SHN_XINDEX,
we need to call elf_getshdrstrndx to get the proper
string table index.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
dutil.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
From: Jiri Olsa <hidden> Date: 2021-01-20 12:40:21
On Tue, Jan 19, 2021 at 06:03:28PM -0800, Andrii Nakryiko wrote:
On Tue, Jan 19, 2021 at 2:16 PM Jiri Olsa [off-list ref] wrote:
quoted
For very large ELF objects (with many sections), we could
get special value SHN_XINDEX (65535) for symbol's st_shndx.
This patch is adding code to detect the optional extended
section index table and use it to resolve symbol's section
index id needed.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
btf_encoder.c | 18 ++++++++++++++++++
elf_symtab.c | 31 ++++++++++++++++++++++++++++++-
elf_symtab.h | 1 +
3 files changed, 49 insertions(+), 1 deletion(-)
@@ -609,6 +609,24 @@ static int collect_symbols(struct btf_elf *btfe, bool collect_percpu_vars)/* search within symtab for percpu variables */elf_symtab__for_each_symbol(btfe->symtab,core_id,sym){
How about we introduce elf_symtab__for_each_symbol variant that uses
gelf_getsymshndx undercover and returns a real section index as the
4th macro argument?
ok, that's good idea
quoted
+ struct elf_symtab *symtab = btfe->symtab;
+
+ /*
+ * Symbol's st_shndx needs to be translated with the
+ * extended section index table.
+ */
+ if (sym.st_shndx == SHN_XINDEX) {
+ Elf32_Word xndx;
+
+ if (!symtab->syms_shndx) {
+ fprintf(stderr, "SHN_XINDEX found, but no symtab shndx section.\n");
+ return -1;
+ }
+
+ if (gelf_getsymshndx(symtab->syms, symtab->syms_shndx, core_id, &sym, &xndx))
+ sym.st_shndx = xndx;
This bit makes me really nervous and it looks very unclean, which is
why I think providing correct section index as part of iterator macro
is better approach. And all this code would just go away.
ok
quoted
+ }
+
if (collect_percpu_vars && collect_percpu_var(btfe, &sym))
return -1;
if (collect_function(btfe, &sym))
move this closer to where you check (xindex > 0)? Can you please leave
a small comment that this returns extended section index table's
section index (I know, this is horrible), if it exists. It's
notoriously hard to find anything about libelf's API, especially for
obscure APIs like this.
it's here because 'sec' gets overwitten shortly on with string table
I'll add some comment
@@ -49,6 +53,31 @@ struct elf_symtab *elf_symtab__new(const char *name, Elf *elf, GElf_Ehdr *ehdr) if (symtab->symstrs == NULL) goto out_free_name;+ /*+ * The .symtab section has optional extended section index+ * table, load its data so it can be used to resolve symbol's+ * section index.+ **/+ if (xindex > 0) {+ GElf_Shdr shdr_shndx;+ Elf_Scn *sec_shndx;++ sec_shndx = elf_getscn(elf, xindex);+ if (sec_shndx == NULL)+ goto out_free_name;++ if (gelf_getshdr(sec_shndx, &shdr_shndx) == NULL)+ goto out_free_name;++ /* Extra check to verify it belongs to the .symtab */+ if (index != shdr_shndx.sh_link)+ goto out_free_name;
you can also verify that section type is SHT_SYMTAB_SHNDX
my mind breaks looking at all those shndxs, especially in this case
you better not look at elfutils code then ;-)
where it's not a section number, but rather data. How about we call it
something like symtab->syms_sec_idx_table or something similar but
human-comprehensible?
From: Jiri Olsa <hidden> Date: 2021-01-20 12:44:42
On Tue, Jan 19, 2021 at 05:22:08PM -0800, Andrii Nakryiko wrote:
On Tue, Jan 19, 2021 at 2:15 PM Jiri Olsa [off-list ref] wrote:
quoted
For very large ELF objects (with many sections), we could
get special value SHN_XINDEX (65535) for elf object's string
table index - e_shstrndx.
In such case we need to call elf_getshdrstrndx to get the
proper string table index.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
tools/lib/bpf/btf.c | 14 ++++++++++++--
1 file changed, 12 insertions(+), 2 deletions(-)
From: Arnaldo Carvalho de Melo <acme@kernel.org> Date: 2021-01-20 12:44:43
Em Tue, Jan 19, 2021 at 05:22:08PM -0800, Andrii Nakryiko escreveu:
On Tue, Jan 19, 2021 at 2:15 PM Jiri Olsa [off-list ref] wrote:
quoted
For very large ELF objects (with many sections), we could
get special value SHN_XINDEX (65535) for elf object's string
table index - e_shstrndx.
In such case we need to call elf_getshdrstrndx to get the
proper string table index.
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
---
tools/lib/bpf/btf.c | 14 ++++++++++++--
1 file changed, 12 insertions(+), 2 deletions(-)
@@ -863,6 +863,7 @@ static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,Elf_Scn*scn=NULL;Elf*elf=NULL;GElf_Ehdrehdr;+size_tshstrndx;if(elf_version(EV_CURRENT)==EV_NONE){pr_warn("failed to init libelf for %s\n",path);
@@ -887,7 +888,16 @@ static struct btf *btf_parse_elf(const char *path, struct btf *base_btf,pr_warn("failed to get EHDR from %s\n",path);gotodone;}-if(!elf_rawdata(elf_getscn(elf,ehdr.e_shstrndx),NULL)){++/*+*Getstringtableindexfromextendedsectionindex+*tableifneeded.+*/+shstrndx=ehdr.e_shstrndx;+if(shstrndx==SHN_XINDEX&&elf_getshdrstrndx(elf,&shstrndx))+gotodone;
just use elf_getshdrstrndx() unconditionally, it works for extended
and non-extended numbering (see libbpf.c).
Yeah, we even have this in tools/perf/util/symbol-elf.c:
#ifndef HAVE_ELF_GETSHDRSTRNDX_SUPPORT
static int elf_getshdrstrndx(Elf *elf __maybe_unused, size_t *dst __maybe_unused)
{
pr_err("%s: update your libelf to > 0.140, this one lacks elf_getshdrstrndx().\n", __func__);
return -1;
}
#endif
And a feature detection for that:
[acme@five perf]$ cat tools/build/feature/test-libelf-getshdrstrndx.c
// SPDX-License-Identifier: GPL-2.0
#include <libelf.h>
int main(void)
{
size_t dst;
return elf_getshdrstrndx(0, &dst);
}
[acme@five perf]$
Your implementation would provide a good fallback tho to avoid requiring
updating to 0.140 :-)
- Arnaldo
quoted
+
+ if (!elf_rawdata(elf_getscn(elf, shstrndx), NULL)) {
pr_warn("failed to get e_shstrndx from %s\n", path);
goto done;
}
@@ -902,7 +912,7 @@ static struct btf *btf_parse_elf(const char *path, struct btf *base_btf, idx, path); goto done; }- name = elf_strptr(elf, ehdr.e_shstrndx, sh.sh_name);+ name = elf_strptr(elf, shstrndx, sh.sh_name); if (!name) { pr_warn("failed to get section(%d) name from %s\n", idx, path);--