perf hists browser: Add NULL pointer check to prevent crash
PATCH 1/2 perf hists browser: Add NULL pointer check to prevent crash
Before this patch we can trigger a segfault by following steps:
Step 1: perf report
Step 2: Use UP/DOWN to select an entry, don't press 'ENTER'
Step 3: Use '/' to filter symbols, use a filter which returns
empty result
Step 4: Press 'ENTER' (notice here that the old selection is still
there. This is another problem)
Step 5: Press 'ENTER' to annotate that symbol
Step 6: Press 'LEFT' to go out.
Result: segfault:
perf: Segmentation fault
-------- backtrace -------- /home/wangnan/perf[0x53e568] /lib64/libc.so.6(+0x3545f)[0x7fba75d3245f] /home/wangnan/perf[0x537516] /home/wangnan/perf[0x533fef] /home/wangnan/perf[0x53b347] /home/wangnan/perf(perf_evlist__tui_browse_hists+0x96)[0x53d206] /home/wangnan/perf(cmd_report+0x1b9f)[0x442c7f] /home/wangnan/perf[0x47efa2] /home/wangnan/perf(main+0x5f5)[0x432fa5] /lib64/libc.so.6(__libc_start_main+0xf4)[0x7fba75d1ebd4] /home/wangnan/perf[0x4330d4]
This is because in this case 'nd' could be NULL in ui_browser__hists_seek(), but that function never check it.
This patch adds checker for potential NULL pointer in that function. After this patch the above steps won't segfault again.
--- tools/perf/ui/browsers/hists.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
tools/perf/ui/browsers/hists.c +6 -2
@@ -1280,8 +1280,10 @@ static void ui_browser__hists_seek(struct ui_browser *browser,* Moves not relative to the first visible entry invalidates its * row_offset: */ - h = rb_entry(browser->top, struct hist_entry, rb_node); - h->row_offset =0 ; +if (browser->top) { + h = rb_entry(browser->top, struct hist_entry, rb_node); + h->row_offset =0 ; + }/* * Here we have to check if nd is expanded (+), if it is we can't go
@@ -1297,6 +1299,8 @@ static void ui_browser__hists_seek(struct ui_browser *browser,* and stop when we printed enough lines to fill the screen. */ do_offset: +if (!nd) +return ;if (offset >0 ) { do { h = rb_entry(nd, struct hist_entry, rb_node);
Signature · 1 line
1.8.3.4
PATCH 2/2 perf hists browser: Reset selection when refresh
With following steps:
Step 1: perf report
Step 2: Use UP/DOWN to select an entry, don't press 'ENTER'
Step 3: Use '/' to filter symbols, use a filter which returns
empty result
Step 4: Press 'ENTER'
We see that, even if we have filter all symbols (and the main interface is empty), pressing 'ENTER' still select one symbol. This behavior surprise user. This patch resets browser->selection in hist_browser__refresh() and let it choose default selection. In this case browser->selection keeps NULL so user won't see annotation item in menu.
---
Note that if this patch is applied before 1/2 then the steps listed in commit message in 1/2 won't trigger segfault. However I believe patch 1/1 is still required.
--- tools/perf/ui/browsers/hists.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
tools/perf/ui/browsers/hists.c +4 -2
@@ -1192,6 +1192,7 @@ static unsigned int hist_browser__refresh(struct ui_browser *browser) }ui_browser__hists_init_top (browser); + hb->selection =NULL ;for (nd = browser->top; nd; nd =rb_next (nd)) { struct hist_entry *h = rb_entry(nd, struct hist_entry, rb_node);
@@ -2102,7 +2103,7 @@ static int perf_evsel__hists_browse(struct perf_evsel *evsel, int nr_events,if (browser->he_selection !=NULL ) { thread =hist_browser__selected_thread (browser); - map = browser->selection->map; + map = browser->selection ? browser->selection->map :NULL ; socked_id = browser->he_selection->socket; }switch (key) {
@@ -2321,7 +2322,8 @@ skip_annotation: nr_options +=add_script_opt (browser, &actions[nr_options], &options[nr_options], -NULL , browser->selection->sym); +NULL , browser->selection ? + browser->selection->sym :NULL ); } nr_options +=add_script_opt (browser, &actions[nr_options], &options[nr_options],NULL ,NULL );
Signature · 1 line
1.8.3.4
Namhyung Kim
maintainerRe: [PATCH 2/2] perf hists browser: Reset selection when refresh
Quoted 36 lines · Wang Nan
With following steps:
Step 1: perf report
Step 2: Use UP/DOWN to select an entry, don't press 'ENTER'
Step 3: Use '/' to filter symbols, use a filter which returns
empty resultStep 4: Press 'ENTER'
We see that, even if we have filter all symbols (and the main interface is empty), pressing 'ENTER' still select one symbol. This behavior surprise user. This patch resets
browser->selectioninhist_browser__refresh()and let it choose default selection. In this casebrowser->selectionkeeps NULL so user won't see annotation item in menu.Signed-off-by Wang Nan <[email protected]>Cc Arnaldo Carvalho de Melo <[email protected]>Cc Namhyung Kim <[email protected]>---
Note that if this patch is applied before 1/2 then the steps listed in commit message in 1/2 won't trigger segfault. However I believe patch 1/1 is still required.
--- tools/perf/ui/browsers/hists.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)tools/perf/ui/browsers/hists.c +0
-0
tools/perf/ui/browsers/hists.c static unsigned int hist_browser__refresh(struct ui_browser *browser) · line 1195
Quoted 4 lines
tools/perf/ui/browsers/hists.c +1
-0@@ -1192,6 +1192,7 @@ static unsigned int hist_browser__refresh(struct ui_browser *browser) }ui_browser__hists_init_top (browser); + hb->selection =NULL ;
This code assumes that hb->selection is not NULL if hb->he_selection is not NULL. So I think that the right (and simple) fix is to reset hb->he_selection rather than hb->selection (or both). It'll make the change below unnecessary IMHO.
Thanks,
Namhyung
Quoted 25 lines
tools/perf/ui/browsers/hists.c +3
-2for (nd = browser->top; nd; nd =rb_next (nd)) { struct hist_entry *h = rb_entry(nd, struct hist_entry, rb_node);@@ -2102,7 +2103,7 @@ static int perf_evsel__hists_browse(struct perf_evsel *evsel, int nr_events,if (browser->he_selection !=NULL ) { thread =hist_browser__selected_thread (browser); - map = browser->selection->map; + map = browser->selection ? browser->selection->map :NULL ; socked_id = browser->he_selection->socket; }switch (key) {@@ -2321,7 +2322,8 @@ skip_annotation: nr_options +=add_script_opt (browser, &actions[nr_options], &options[nr_options], -NULL , browser->selection->sym); +NULL , browser->selection ? + browser->selection->sym :NULL ); } nr_options +=add_script_opt (browser, &actions[nr_options], &options[nr_options],NULL ,NULL );Signature · 1 line
1.8.3.4
Re: [PATCH 2/2] perf hists browser: Reset selection when refresh
Quoted 39 lines · Namhyung Kim
Quoted 38 lines · Wang Nan
With following steps:
Step 1: perf report
Step 2: Use UP/DOWN to select an entry, don't press 'ENTER'
Step 3: Use '/' to filter symbols, use a filter which returns
empty resultStep 4: Press 'ENTER'
We see that, even if we have filter all symbols (and the main interface is empty), pressing 'ENTER' still select one symbol. This behavior surprise user. This patch resets
browser->selectioninhist_browser__refresh()and let it choose default selection. In this casebrowser->selectionkeeps NULL so user won't see annotation item in menu.Signed-off-by Wang Nan <[email protected]>Cc Arnaldo Carvalho de Melo <[email protected]>Cc Namhyung Kim <[email protected]>---
Note that if this patch is applied before 1/2 then the steps listed in commit message in 1/2 won't trigger segfault. However I believe patch 1/1 is still required.
--- tools/perf/ui/browsers/hists.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)tools/perf/ui/browsers/hists.c +0
-0@@ -1192,6 +1192,7 @@ static unsigned int hist_browser__refresh(struct ui_browser *browser) }
Quoted 6 lines
tools/perf/ui/browsers/hists.c static unsigned int hist_browser__refresh(struct ui_browser *browser) · line 1195
Quoted 2 lines
tools/perf/ui/browsers/hists.c +1
-0ui_browser__hists_init_top (browser); + hb->selection =NULL ;This code assumes that
hb->selectionis not NULL ifhb->he_selectionis not NULL. So I think that the right (and simple) fix is to resethb->he_selectionrather thanhb->selection(or both). It'll make the change below unnecessary IMHO.
No. Only set hb->he_selection causes crash:
Step 0: user 'perf record ls' create a perf.data without callchain; Step 1: perf report
Step 2: choose a annotable entry, don't press 'ENTER'
Step 3: use '/' and enter a filter like 'asdfasdf' which ensure no entry
would be left
Step 3: Press 'ENTER' twice
Crash:
# ./perf report perf: Segmentation fault -------- backtrace -------- ./perf[0x53e588] /tmp/oxygen_root/lib64/libc.so.6(+0x3545f)[0x7f2b4d6da45f] ./perf[0x539e03] ./perf(perf_evlist__tui_browse_hists+0x96)[0x53d226] ./perf(cmd_report+0x1b9f)[0x442c7f] ./perf[0x47efc2] ./perf(main+0x5f5)[0x432fa5] /tmp/oxygen_root/lib64/libc.so.6(__libc_start_main+0xf4)[0x7f2b4d6c6bd4] ./perf[0x4330d4]
GDB result:
Program received signal SIGSEGV, Segmentation fault.hist_browser__toggle_fold (browser=0x1f71160) at ui/browsers/hists.c:352
(gdb) bt
#0 hist_browser__toggle_fold (browser=0x1f71160) at ui/browsers/hists.c:352
#1 hist_browser__run (help=0x649038 "For a higher level overview, try: perf report --sort comm,dso", browser=0x1f71160) at ui/ browsers/hists.c:539
#2 perf_evsel__hists_browse (evsel=0x19ef140, nr_events=nr_events@entry=1, helpline=helpline@entry=0x649038 "For a higher leve l overview, try: perf report --sort comm,dso", left_exits=left_exits@entry=false, hbt=hbt@entry=0x0, min_pcnt=<optimized out>, env=0x19ed850) at ui/browsers/hists.c:2101
#3 0x000000000053d227 in perf_evlist__tui_browse_hists (evlist=0x19ee730, help=help@entry=0x649038 "For a higher level overvie w, try: perf report --sort comm,dso", hbt=hbt@entry=0x0, min_pcnt=<optimized out>, env=<optimized out>) at ui/browsers/hists.c:
2555
#4 0x0000000000442c80 in report__browse_hists (rep=0x7fffffffca20) at builtin-report.c:440 #5 __cmd_report (rep=0x7fffffffca20) at builtin-report.c:576 #6 cmd_report (argc=0, argv=0x7fffffffe590, prefix=<optimized out>) at builtin-report.c:962 #7 0x000000000047efc3 in run_builtin (p=p@entry=0x8ff9e0 <commands+192>, argc=argc@entry=1, argv=argv@entry=0x7fffffffe590) at
perf.c:387
#8 0x0000000000432fa6 in handle_internal_command (argv=0x7fffffffe590, argc=1) at perf.c:448 #9 run_argv (argv=0x7fffffffe310, argcp=0x7fffffffe31c) at perf.c:492
#10 main (argc=1, argv=0x7fffffffe590) at perf.c:609
But setting both of them to NULL in hist_browser__refresh() can avoid this crash.
So we have two choice:
1. In hist_browser__refresh() let's set both selection and he_selection to NULL;
By this way after step 3 we won't see annotation options. The
context menu
(by pressing ENTER or 'm') is:
Run scripts for all samples
Switch to another data file in PWD
Exit
2. In hist_browser__toggle_fold() let's check both he amd ms.
By this way we still get annotation and map options in context menu:
Annotate __strcoll_l
Browse map details
Run scripts for all samples
Switch to another data file in PWD
Exit
I'm not sure which one is better because I don't really understand this part of code. But for me the second result is surprising because I have filtered all entries out in my interface, and I believe I should select nothing, so pressing 'ENTER' or 'm' I shall not get annotation option because I don't know which entry would be annotated.
Thank you.
Namhyung Kim
maintainerRe: [PATCH 2/2] perf hists browser: Reset selection when refresh
Hi,
Quoted 70 lines · Wangnan (F)
Quoted 32 lines · Namhyung Kim
Quoted 31 lines · Wang Nan
With following steps:
Step 1: perf report
Step 2: Use UP/DOWN to select an entry, don't press 'ENTER'
Step 3: Use '/' to filter symbols, use a filter which returns
empty resultStep 4: Press 'ENTER'
We see that, even if we have filter all symbols (and the main interface is empty), pressing 'ENTER' still select one symbol. This behavior surprise user. This patch resets
browser->selectioninhist_browser__refresh()and let it choose default selection. In this casebrowser->selectionkeeps NULL so user won't see annotation item in menu.Signed-off-by Wang Nan <[email protected]>Cc Arnaldo Carvalho de Melo <[email protected]>Cc Namhyung Kim <[email protected]>---
Note that if this patch is applied before 1/2 then the steps listed in commit message in 1/2 won't trigger segfault. However I believe patch 1/1 is still required.
--- tools/perf/ui/browsers/hists.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)Quoted 12 lines
tools/perf/ui/browsers/hists.c static unsigned int hist_browser__refresh(struct ui_browser *browser) · line 1194
Quoted 8 lines
tools/perf/ui/browsers/hists.c +1
-0@@ -1192,6 +1192,7 @@ static unsigned int hist_browser__refresh(struct ui_browser *browser) }ui_browser__hists_init_top (browser); + hb->selection =NULL ;This code assumes that
hb->selectionis not NULL ifhb->he_selectionis not NULL. So I think that the right (and simple) fix is to resethb->he_selectionrather thanhb->selection(or both). It'll make the change below unnecessary IMHO.No. Only set
hb->he_selectioncauses crash:Step 0: user 'perf record ls' create a perf.data without callchain; Step 1: perf report
Step 2: choose a annotable entry, don't press 'ENTER'
Step 3: use '/' and enter a filter like 'asdfasdf' which ensure no entry
would be left
Step 3: Press 'ENTER' twiceCrash:
# ./perf report perf: Segmentation fault -------- backtrace -------- ./perf[0x53e588] /tmp/oxygen_root/lib64/libc.so.6(+0x3545f)[0x7f2b4d6da45f] ./perf[0x539e03] ./perf(perf_evlist__tui_browse_hists+0x96)[0x53d226] ./perf(cmd_report+0x1b9f)[0x442c7f] ./perf[0x47efc2] ./perf(main+0x5f5)[0x432fa5] /tmp/oxygen_root/lib64/libc.so.6(__libc_start_main+0xf4)[0x7f2b4d6c6bd4] ./perf[0x4330d4]
Quoted 4 lines
GDB result:
Program received signal SIGSEGV, Segmentation fault.
hist_browser__toggle_fold(browser=0x1f71160) atui/browsers/hists.c:352
So this is same crash you already fixed. Now I think that it'd be better to check hb->he_selection instead of hb->selection in that patch for the sake of consistency.
Quoted 56 lines
(gdb) bt
#0hist_browser__toggle_fold(browser=0x1f71160) atui/browsers/hists.c:352#1hist_browser__run(help=0x649038 "For a higher level overview, try: perf report --sort comm,dso", browser=0x1f71160) at ui/browsers/hists.c:539
#2perf_evsel__hists_browse(evsel=0x19ef140, nr_events=nr_events@entry=1, helpline=helpline@entry=0x649038 "For a higher leve
l overview, try: perf report --sort comm,dso", left_exits=left_exits@entry=false, hbt=hbt@entry=0x0, min_pcnt=<optimized out>,
env=0x19ed850) atui/browsers/hists.c:2101
#30x000000000053d227inperf_evlist__tui_browse_hists(evlist=0x19ee730, help=help@entry=0x649038 "For a higher level overvie
w, try: perf report --sort comm,dso", hbt=hbt@entry=0x0, min_pcnt=<optimized out>, env=<optimized out>) atui/browsers/hists.c:
2555
#40x0000000000442c80inreport__browse_hists(rep=0x7fffffffca20) at builtin-report.c:440
#5__cmd_report(rep=0x7fffffffca20) at builtin-report.c:576
#6cmd_report(argc=0, argv=0x7fffffffe590, prefix=<optimized out>) at builtin-report.c:962
#70x000000000047efc3inrun_builtin(p=p@entry=0x8ff9e0 <commands+192>, argc=argc@entry=1, argv=argv@entry=0x7fffffffe590) at
perf.c:387
#80x0000000000432fa6inhandle_internal_command(argv=0x7fffffffe590, argc=1) at perf.c:448
#9run_argv(argv=0x7fffffffe310, argcp=0x7fffffffe31c) at perf.c:492 #10 main (argc=1, argv=0x7fffffffe590) at perf.c:609But setting both of them to NULL in
hist_browser__refresh()can avoid this crash.So we have two choice:
1. In
hist_browser__refresh()let's set both selection andhe_selectionto NULL;By this way after step 3 we won't see annotation options. The context
menu
(by pressing ENTER or 'm') is:Run scripts for all samples
Switch to another data file in PWD
Exit2. In
hist_browser__toggle_fold()let's check both he amd ms.By this way we still get annotation and map options in context menu:
Annotate
__strcoll_l
Browse map details
Run scripts for all samples
Switch to another data file in PWD
Exit
Quoted 8 lines
I'm not sure which one is better because I don't really understand this part of
code. But for me the second result is surprising because I have filtered all entries out in my interface, and I believe I should select nothing, so pressing
'ENTER' or 'm' I shall not get annotation option because I don't know which entry
would be annotated.
Agreed. I also think that the choice 1 is the right thing.
Thanks,
Namhyung
Namhyung Kim
maintainerQuoted 11 lines · Wang Nan
Before this patch we can trigger a segfault by following steps:
Step 1: perf report
Step 2: Use UP/DOWN to select an entry, don't press 'ENTER'
Step 3: Use '/' to filter symbols, use a filter which returns
empty resultStep 4: Press 'ENTER' (notice here that the old selection is still
there. This is another problem)
I guess your data file doesn't contain callchains. Otherwise 'ENTER' key will toggle it and not show context menu. We now support 'm' key to show context menu in any case.
Quoted 39 lines
Step 5: Press 'ENTER' to annotate that symbol
Step 6: Press 'LEFT' to go out.
Result: segfault:
perf: Segmentation fault
-------- backtrace -------- /home/wangnan/perf[0x53e568] /lib64/libc.so.6(+0x3545f)[0x7fba75d3245f] /home/wangnan/perf[0x537516] /home/wangnan/perf[0x533fef] /home/wangnan/perf[0x53b347] /home/wangnan/perf(perf_evlist__tui_browse_hists+0x96)[0x53d206] /home/wangnan/perf(cmd_report+0x1b9f)[0x442c7f] /home/wangnan/perf[0x47efa2] /home/wangnan/perf(main+0x5f5)[0x432fa5] /lib64/libc.so.6(__libc_start_main+0xf4)[0x7fba75d1ebd4] /home/wangnan/perf[0x4330d4]This is because in this case 'nd' could be NULL in
ui_browser__hists_seek(), but that function never check it.This patch adds checker for potential NULL pointer in that function. After this patch the above steps won't segfault again.
Signed-off-by Wang Nan <[email protected]>Cc Arnaldo Carvalho de Melo <[email protected]>Cc Namhyung Kim <[email protected]>--- tools/perf/ui/browsers/hists.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)tools/perf/ui/browsers/hists.c +0
-0
tools/perf/ui/browsers/hists.c static void ui_browser__hists_seek(struct ui_browser *browser, · line 1286
Quoted 9 lines
tools/perf/ui/browsers/hists.c +4
-2@@ -1280,8 +1280,10 @@ static void ui_browser__hists_seek(struct ui_browser *browser,* Moves not relative to the first visible entry invalidates its * row_offset: */ - h = rb_entry(browser->top, struct hist_entry, rb_node); - h->row_offset =0 ; +if (browser->top) { + h = rb_entry(browser->top, struct hist_entry, rb_node); + h->row_offset =0 ; + }
Did you have a problem with this code? It seems that ui_browser__hists_init_top() ensures browser->top is initialized.
tools/perf/ui/browsers/hists.c static void ui_browser__hists_seek(struct ui_browser *browser, · line 1303
Quoted 9 lines
tools/perf/ui/browsers/hists.c +2
-0/* * Here we have to check if nd is expanded (+), if it is we can't go @@ -1297,6 +1299,8 @@ static void ui_browser__hists_seek(struct ui_browser *browser,* and stop when we printed enough lines to fill the screen. */ do_offset: +if (!nd) +return ;
So I guess checking 'nd' being NULL here is enough. And please add a blank line for readability.
Thanks,
Namhyung
Quoted 6 lines
tools/perf/ui/browsers/hists.c +0
-0if (offset >0 ) { do { h = rb_entry(nd, struct hist_entry, rb_node);Signature · 1 line
1.8.3.4
Quoted 7 lines · Namhyung Kim
Quoted 6 lines · Wang Nan
Before this patch we can trigger a segfault by following steps:
Step 1: perf report
Step 2: Use UP/DOWN to select an entry, don't press 'ENTER'
Quoted 7 lines
Quoted 5 lines
Step 3: Use '/' to filter symbols, use a filter which returns
empty resultStep 4: Press 'ENTER' (notice here that the old selection is still
there. This is another problem)I guess your data file doesn't contain callchains. Otherwise 'ENTER' key will toggle it and not show context menu.
Yes. With callchain information, this step should be replaced by:
Toggle an entry and select a leaf entry which can be annotated with UP and DOWN, don't press 'ENTER'.
I get similar result.
Quoted 37 lines
We now support 'm' key
to show context menu in any case.Quoted 33 lines
Step 5: Press 'ENTER' to annotate that symbol
Step 6: Press 'LEFT' to go out.
Result: segfault:
perf: Segmentation fault
-------- backtrace -------- /home/wangnan/perf[0x53e568] /lib64/libc.so.6(+0x3545f)[0x7fba75d3245f] /home/wangnan/perf[0x537516] /home/wangnan/perf[0x533fef] /home/wangnan/perf[0x53b347] /home/wangnan/perf(perf_evlist__tui_browse_hists+0x96)[0x53d206] /home/wangnan/perf(cmd_report+0x1b9f)[0x442c7f] /home/wangnan/perf[0x47efa2] /home/wangnan/perf(main+0x5f5)[0x432fa5] /lib64/libc.so.6(__libc_start_main+0xf4)[0x7fba75d1ebd4] /home/wangnan/perf[0x4330d4]This is because in this case 'nd' could be NULL in
ui_browser__hists_seek(), but that function never check it.This patch adds checker for potential NULL pointer in that function. After this patch the above steps won't segfault again.
Signed-off-by Wang Nan <[email protected]>Cc Arnaldo Carvalho de Melo <[email protected]>Cc Namhyung Kim <[email protected]>--- tools/perf/ui/browsers/hists.c | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-)
Quoted 17 lines
Quoted 5 lines
tools/perf/ui/browsers/hists.c +0
-0tools/perf/ui/browsers/hists.c static void ui_browser__hists_seek(struct ui_browser *browser, · line 1286
Quoted 9 lines
tools/perf/ui/browsers/hists.c +4
-2@@ -1280,8 +1280,10 @@ static void ui_browser__hists_seek(struct ui_browser *browser,* Moves not relative to the first visible entry invalidates its * row_offset: */ - h = rb_entry(browser->top, struct hist_entry, rb_node); - h->row_offset =0 ; +if (browser->top) { + h = rb_entry(browser->top, struct hist_entry, rb_node); + h->row_offset =0 ; + }Did you have a problem with this code? It seems that
ui_browser__hists_init_top()ensuresbrowser->topis initialized.
No. Add it for safty because I see in following code that browser->top can be NULL.
Will be removed in next version.
Thank you.