| Message ID | 20190412030322.15494-1-bpoirier@suse.com |
|---|---|
| State | Changes Requested |
| Delegated to: | BPF Maintainers |
| Headers | show |
| Series | [v2,bpf-next,1/2] bpftool: Use print_entry_error() in case of ENOENT when dumping | expand |
2019-04-12 12:03 UTC+0900 ~ Benjamin Poirier <bpoirier@suse.com> > Commit bf598a8f0f77 ("bpftool: Improve handling of ENOENT on map dumps") > used print_entry_plain() in case of ENOENT because that function provided > the desired formatting. However, that commit actually introduces dead code. > Per-cpu maps are zero-filled. When reading them, it's all or nothing. There > will never be a case where some cpus have an entry and others don't. > > The truth is that ENOENT is an error case. So rework print_entry_error() to > provide the desired formatting. > > Note that this commit changes the output format in case of errors other > than ENOENT. > > example before: > key: > 00 00 00 00 > value: > No space left on device > [...] > > example after: > key: 00 00 00 00 > value: > No space left on device > [...] > > The ENOENT case is unchanged: > key: 00 00 00 00 value: 14 5b 00 00 00 00 00 00 > key: 01 00 00 00 value: <no entry> > [...] > > Signed-off-by: Benjamin Poirier <bpoirier@suse.com> > --- > tools/bpf/bpftool/map.c | 34 ++++++++++++++-------------------- > 1 file changed, 14 insertions(+), 20 deletions(-) > > diff --git a/tools/bpf/bpftool/map.c b/tools/bpf/bpftool/map.c > index e96903078991..71840faaeab5 100644 > --- a/tools/bpf/bpftool/map.c > +++ b/tools/bpf/bpftool/map.c > @@ -261,20 +261,19 @@ static void print_entry_json(struct bpf_map_info *info, unsigned char *key, > } > > static void print_entry_error(struct bpf_map_info *info, unsigned char *key, > - const char *value) > + const char *value, bool single_line) Nit: if you respin the series, could you rename "value" into "error_msg" or something like this to better indicate we never pass an actual map value to the function? > { > - int value_size = strlen(value); > - bool single_line, break_names; > + bool break_names; > > - break_names = info->key_size > 16 || value_size > 16; > - single_line = info->key_size + value_size <= 24 && !break_names; > + break_names = info->key_size > 16; > + single_line = single_line && !break_names; If I understand correctly, this will also change formatting when error message is short (shorter than 16 characters: the "value" line will now be unconditionally split, even for short error messages (other than "<no entry>")). Why removing the condition on value_size > 16? (This is just a remark, I am not against changing it.) > > printf("key:%c", break_names ? '\n' : ' '); > fprint_hex(stdout, key, info->key_size, " "); > > printf(single_line ? " " : "\n"); > > - printf("value:%c%s", break_names ? '\n' : ' ', value); > + printf("value:%c%s", single_line ? ' ' : '\n', value); > > printf("\n"); > } [...]
On 2019/04/12 11:28, Quentin Monnet wrote: [...] > > > { > > - int value_size = strlen(value); > > - bool single_line, break_names; > > + bool break_names; > > - break_names = info->key_size > 16 || value_size > 16; > > - single_line = info->key_size + value_size <= 24 && !break_names; > > + break_names = info->key_size > 16; > > + single_line = single_line && !break_names; > > If I understand correctly, this will also change formatting when error > message is short (shorter than 16 characters: the "value" line will now be > unconditionally split, even for short error messages (other than "<no > entry>")). Why removing the condition on value_size > 16? (This is just a > remark, I am not against changing it.) > With this patch, the error messages from bpftool ("<no entry>", "<cannot read>"), which are chosen to be short, appear on the same line and the messages from strerror appear on a separate line. Because those latter messages are from a source external to bpftool and their length is unknown ahead of time, I felt it led to a more predictable output to consistently put them on their own line. To be honest, I don't think the formatting in those print_entry_* functions should change according to the length in any case. I think the key and value for each entry should always be on the same line for ease of grepping. A followup patch maybe...
On Sat, 13 Apr 2019 07:49:24 +0900, Benjamin Poirier wrote: > To be honest, I don't think the formatting in those print_entry_* > functions should change according to the length in any case. I think the > key and value for each entry should always be on the same line for ease > of grepping. A followup patch maybe... No, no, please never grep bpftool output. We have JSON output, please use JQ: http://manpages.ubuntu.com/manpages/bionic/man1/jq.1.html
On Fri, 12 Apr 2019 12:03:21 +0900, Benjamin Poirier wrote: > diff --git a/tools/bpf/bpftool/map.c b/tools/bpf/bpftool/map.c > index e96903078991..71840faaeab5 100644 > --- a/tools/bpf/bpftool/map.c > +++ b/tools/bpf/bpftool/map.c > @@ -261,20 +261,19 @@ static void print_entry_json(struct bpf_map_info *info, unsigned char *key, > } > > static void print_entry_error(struct bpf_map_info *info, unsigned char *key, > - const char *value) > + const char *value, bool single_line) > { > - int value_size = strlen(value); > - bool single_line, break_names; > + bool break_names; I'd rather you kept the strlen() and 16 char limitation logic. > - break_names = info->key_size > 16 || value_size > 16; > - single_line = info->key_size + value_size <= 24 && !break_names; > + break_names = info->key_size > 16; > + single_line = single_line && !break_names; > > printf("key:%c", break_names ? '\n' : ' '); > fprint_hex(stdout, key, info->key_size, " "); > > printf(single_line ? " " : "\n"); > > - printf("value:%c%s", break_names ? '\n' : ' ', value); > + printf("value:%c%s", single_line ? ' ' : '\n', value); > > printf("\n"); > }
On 2019/04/12 16:53, Jakub Kicinski wrote: > On Sat, 13 Apr 2019 07:49:24 +0900, Benjamin Poirier wrote: > > To be honest, I don't think the formatting in those print_entry_* > > functions should change according to the length in any case. I think the > > key and value for each entry should always be on the same line for ease > > of grepping. A followup patch maybe... > > No, no, please never grep bpftool output. We have JSON output, > please use JQ: http://manpages.ubuntu.com/manpages/bionic/man1/jq.1.html > Fair enough. I didn't know about jq, thanks.
diff --git a/tools/bpf/bpftool/map.c b/tools/bpf/bpftool/map.c index e96903078991..71840faaeab5 100644 --- a/tools/bpf/bpftool/map.c +++ b/tools/bpf/bpftool/map.c @@ -261,20 +261,19 @@ static void print_entry_json(struct bpf_map_info *info, unsigned char *key, } static void print_entry_error(struct bpf_map_info *info, unsigned char *key, - const char *value) + const char *value, bool single_line) { - int value_size = strlen(value); - bool single_line, break_names; + bool break_names; - break_names = info->key_size > 16 || value_size > 16; - single_line = info->key_size + value_size <= 24 && !break_names; + break_names = info->key_size > 16; + single_line = single_line && !break_names; printf("key:%c", break_names ? '\n' : ' '); fprint_hex(stdout, key, info->key_size, " "); printf(single_line ? " " : "\n"); - printf("value:%c%s", break_names ? '\n' : ' ', value); + printf("value:%c%s", single_line ? ' ' : '\n', value); printf("\n"); } @@ -298,11 +297,7 @@ static void print_entry_plain(struct bpf_map_info *info, unsigned char *key, if (info->value_size) { printf("value:%c", break_names ? '\n' : ' '); - if (value) - fprint_hex(stdout, value, info->value_size, - " "); - else - printf("<no entry>"); + fprint_hex(stdout, value, info->value_size, " "); } printf("\n"); @@ -321,11 +316,8 @@ static void print_entry_plain(struct bpf_map_info *info, unsigned char *key, for (i = 0; i < n; i++) { printf("value (CPU %02d):%c", i, info->value_size > 16 ? '\n' : ' '); - if (value) - fprint_hex(stdout, value + i * step, - info->value_size, " "); - else - printf("<no entry>"); + fprint_hex(stdout, value + i * step, + info->value_size, " "); printf("\n"); } } @@ -722,11 +714,13 @@ static int dump_map_elem(int fd, void *key, void *value, jsonw_string_field(json_wtr, "error", strerror(lookup_errno)); jsonw_end_object(json_wtr); } else { + const char *msg = NULL; + if (errno == ENOENT) - print_entry_plain(map_info, key, NULL); - else - print_entry_error(map_info, key, - strerror(lookup_errno)); + msg = "<no entry>"; + + print_entry_error(map_info, key, + msg ? : strerror(lookup_errno), !!msg); } return 0;
Commit bf598a8f0f77 ("bpftool: Improve handling of ENOENT on map dumps") used print_entry_plain() in case of ENOENT because that function provided the desired formatting. However, that commit actually introduces dead code. Per-cpu maps are zero-filled. When reading them, it's all or nothing. There will never be a case where some cpus have an entry and others don't. The truth is that ENOENT is an error case. So rework print_entry_error() to provide the desired formatting. Note that this commit changes the output format in case of errors other than ENOENT. example before: key: 00 00 00 00 value: No space left on device [...] example after: key: 00 00 00 00 value: No space left on device [...] The ENOENT case is unchanged: key: 00 00 00 00 value: 14 5b 00 00 00 00 00 00 key: 01 00 00 00 value: <no entry> [...] Signed-off-by: Benjamin Poirier <bpoirier@suse.com> --- tools/bpf/bpftool/map.c | 34 ++++++++++++++-------------------- 1 file changed, 14 insertions(+), 20 deletions(-)