Introduction to segmentation fault handling

Intro to SegmentationIntro to Segmentation
Fault Handling in LinuxFault Handling in Linux
By Khanh Ngo-DuyBy Khanh Ngo-Duy
Khanhnd@elarion.comKhanhnd@elarion.com

SeminarSeminar
What is Segmentation Fault (Segfault)What is Segmentation Fault (Segfault)
Examples and ScreenshotsExamples and Screenshots
Tips to get Segfault informationTips to get Segfault information
Segfault Debugging TechniquesSegfault Debugging Techniques

What is SegmentationWhat is Segmentation
Fault?Fault?
Segmentation Fault (Segfault)Segmentation Fault (Segfault) oror accessaccess
violationviolation is a particular error condition that canis a particular error condition that can
occur during the operation of computeroccur during the operation of computer
softwaresoftware
A Segfault occurs when a program attempts toA Segfault occurs when a program attempts to
access a memory location that is not allowedaccess a memory location that is not allowed
to access, or attempts to access a memoryto access, or attempts to access a memory
location in a way that is not allowedlocation in a way that is not allowed
Write to a read-only locationWrite to a read-only location
To overwrite part of the operating system or protected memoryTo overwrite part of the operating system or protected memory
locationslocations
Access to invalid memory location. e.g : memorry address NULL, -1 . . .Access to invalid memory location. e.g : memorry address NULL, -1 . . .
etc . . .etc . . .

Examples andExamples and
ScreenshotsScreenshots (1 of 3)(1 of 3)
Write to Read-Only
memory address

Write to Invalid
memory address
(NULL = 0x00)

Stack overflow

Tips to get SegfaultTips to get Segfault
informationinformation (1 of 7)(1 of 7)
Generally, when Segfault occurs, very lessGenerally, when Segfault occurs, very less
information is provided (see previous slides)information is provided (see previous slides)
•
→→ Very hard to debugVery hard to debug

UseUse dmesgdmesg to show information that saved byto show information that saved by
the Kernel when any application crashesthe Kernel when any application crashes
Last Segfault information

How to readHow to read dmesgdmesg outputs ?outputs ?
S eg fault[19960]: s eg fault at 7fffff7feff8 ip 400480 s p 7fffff7ff000 error 6 in
S eg fault[400000+1000]
Application
name
Reason
it crashed
Address
caused fault
Instruction
Pointer address
Additional
error code
Stack
Pointer address
Other values ???
I DON'T KNOW...

AddAdd -g-g when compiling the source code. Thewhen compiling the source code. The
compiler will add debugging symbols into thecompiler will add debugging symbols into the
binarybinary
Will provide more useful information when debugging with gdbWill provide more useful information when debugging with gdb
Compiled binary size will be largers (debugging symbols are added)Compiled binary size will be largers (debugging symbols are added)
Application runs slower, takes more RAMApplication runs slower, takes more RAM
Maybe, some other drawbacksMaybe, some other drawbacks

AddAdd -g-g when compiling the source code. Thewhen compiling the source code. The
compiler will add debugging symbols into thecompiler will add debugging symbols into the
binarybinary
Will provide more useful information when debugging withWill provide more useful information when debugging with gdbgdb
WithoutWithout -g-g,, gccgcc still adds some minimal debugging informationstill adds some minimal debugging information
Compiled binary size will be largers (debugging symbols are added)Compiled binary size will be largers (debugging symbols are added)
Application runs slower, takes more RAMApplication runs slower, takes more RAM
Maybe, some other drawbacksMaybe, some other drawbacks

UseUse nmnm to view the symbols in the binary fileto view the symbols in the binary file
Address of symbol, symbol type, symbol name can be listedAddress of symbol, symbol type, symbol name can be listed
Give us chance to know the Segfault occurred with what symbolGive us chance to know the Segfault occurred with what symbol
$man nm$man nm for more informationfor more information
on the usageon the usage

UseUse lddldd to view the shared library dependenciesto view the shared library dependencies
Show shared library name, starting address of libraryShow shared library name, starting address of library
We know Segfault occurred in our application or in shared libraryWe know Segfault occurred in our application or in shared library
$man ldd$man ldd for more information how to usefor more information how to use lddldd

Segfault DebuggingSegfault Debugging
TechniquesTechniques
UsingUsing gdbgdb – The GNU Debugger– The GNU Debugger
Core dump fileCore dump file andand gdbgdb
objdumpobjdump

Using gdb – The GNU DebuggerUsing gdb – The GNU Debugger (1 of 6)(1 of 6)
gdbgdb supports:supports:
Starting programs , attaching to running programs or debuggingStarting programs , attaching to running programs or debugging
crashed programscrashed programs
Debugging locally or remotely (viaDebugging locally or remotely (via gdbservergdbserver))
Setting breakpoints and watchpointsSetting breakpoints and watchpoints
Examining variables, registers and call stackExamining variables, registers and call stack
Changing data and calling functionsChanging data and calling functions
Automating debug tasksAutomating debug tasks
Multi threaded programsMulti threaded programs

In order to effectively debug program, addIn order to effectively debug program, add -g-g
when compiling withwhen compiling with gccgcc
Load a program into gdb:Load a program into gdb:
$gdb$gdb programprogram
Once you are in gdb, you can run the programOnce you are in gdb, you can run the program
(gdb)(gdb)runrun [parameters to program][parameters to program]

To stop program, pressTo stop program, press Ctrl+CCtrl+C

To quit gdb, execute commandTo quit gdb, execute command qq

← Step 1: Load the program into gdb
← Step 2: execute the program
← Step 3: Quit from gdb
← gdb detects Segfault
but very less info (-g is
not add when compiling)

← add -g when compile
← Step 1: Load
← Step 3: Quit from gdb
← gdb detects Segfault, shows
the lines which caused Segfault
Line 6, in main(), file: Segfault.c
← Step 2: Run

Is thisIs this usefuluseful andand easyeasy ?? YES!!!YES!!! ButBut why?why?
Because ofBecause of -g-g → we can see the→ we can see the file namefile name,, function namefunction name andand line numberline number
Because of the source code is available → can see the exactly line of codeBecause of the source code is available → can see the exactly line of code
If there is no source code → we can see the filename, function name, lineIf there is no source code → we can see the filename, function name, line
number but NOT contents of the line cause Segfault ← No problem, stillnumber but NOT contents of the line cause Segfault ← No problem, still
GOOD! :-)GOOD! :-)
Because this situation is simple, sometime you can NOT use thisBecause this situation is simple, sometime you can NOT use this
technique! See next...technique! See next...

This technique canThis technique can ON LYON LY be used when:be used when:
You know for sure, Segfault will occurYou know for sure, Segfault will occur
Only when testing. When in production time, you can NOT → gdb causesOnly when testing. When in production time, you can NOT → gdb causes
manymany side effectsside effects: slow down the running, running is not stable etc …: slow down the running, running is not stable etc …
Even when testing, if application is soEven when testing, if application is so BigBig oror ComplicatedComplicated (many(many
threads, many resources)threads, many resources) →→ gdb can not handlegdb can not handle
To be able to debug when your application is in production mode and notTo be able to debug when your application is in production mode and not
able to reduce the Segfault ? See the next techniques → ....able to reduce the Segfault ? See the next techniques → ....

core dump file and gdbcore dump file and gdb (1 of 5)(1 of 5)
A core dump consists of the recorded state ofA core dump consists of the recorded state of
the working memory of a computer program atthe working memory of a computer program at
a specific time, generally when the programa specific time, generally when the program
has terminated abnormally (crashed)has terminated abnormally (crashed)
Core dump file might contain: processor registers, which may include theCore dump file might contain: processor registers, which may include the
program counter and stack pointer, memory management information, andprogram counter and stack pointer, memory management information, and
other processor and operating system flags and informationother processor and operating system flags and information
Core dumps are disabled by default on some Linux distributionsCore dumps are disabled by default on some Linux distributions
To force the core dump generation, you can using command lineTo force the core dump generation, you can using command line
$ulimit$ulimit -c <limit size of core file>-c <limit size of core file>
To force the core dump generation, you can also insert code to yourTo force the core dump generation, you can also insert code to your
application to request generating the core dump when it crashesapplication to request generating the core dump when it crashes
To disable the core dump just setTo disable the core dump just set <limit size of core file><limit size of core file> toto 00

← Enable core dump, limit to 1024 MB, just once
← core file is generated when app crashes
← It is here!

Once you have core dump, what to do ?Once you have core dump, what to do ?
→→ Just load it into gdb and seeJust load it into gdb and see
$gdb$gdb <application name> <core file name><application name> <core file name>

← Load the application and core file
← gdb reads core file and
shows the results as if the
Application has just run and
crashed, actually core dump
just shows the actual HISTORY

Is this better than the previous technique ?Is this better than the previous technique ?
Yes, because I could not reproduce theYes, because I could not reproduce the
Segfault, however, core file shows meSegfault, however, core file shows me
Though it is good, still some disadvantagesThough it is good, still some disadvantages
Core file may grow very large if your application uses much memory, soCore file may grow very large if your application uses much memory, so
sometime you simply can not use this methodsometime you simply can not use this method
In case of complicated application, thereIn case of complicated application, there might be some side effectsmight be some side effects whenwhen
forcing core dump → your application might run unstableforcing core dump → your application might run unstable

What I read till now, justWhat I read till now, just s ide effec tss ide effec ts .. Is there anyIs there any
else? I don't want to risk the production system! →else? I don't want to risk the production system! →
YES. There is, see the last techniques ..YES. There is, see the last techniques ....

objdumpobjdump (1 of 9)(1 of 9)
AdvantagesAdvantages
No need to addNo need to add -g-g → do not affect the memory and reduce the size of binary→ do not affect the memory and reduce the size of binary
filefile
No need to generating core dump → no side effects, do not take disk spaceNo need to generating core dump → no side effects, do not take disk space
Actually, you do not need to do anything, what will come will come, and youActually, you do not need to do anything, what will come will come, and you
will solve it!will solve it!
Disadvantages ?Disadvantages ?
You need a little knowledge aboutYou need a little knowledge about assemblyassembly language :-) → don't be scare, stilllanguage :-) → don't be scare, still
easy!easy!
If adding optimization flag toIf adding optimization flag to gccgcc ((-O-O,, -O2-O2,, -O3-O3) it will be a little hard you to) it will be a little hard you to
read assembly code laterread assembly code later

First of all, what you need is the output of dmesgFirst of all, what you need is the output of dmesg
(very first slide told you).(very first slide told you).
Note theNote the “Address caused fault”“Address caused fault” andand “Instruction pointer address”“Instruction pointer address”
Use the tool namedUse the tool named objdumpobjdump to generate information from your applicationto generate information from your application
Output of objdump should be redirected to a file, we need this file later!Output of objdump should be redirected to a file, we need this file later!
$objdump$objdump -DCl <application name> > <output file>-DCl <application name> > <output file>

OK, so my fault address isOK, so my fault address is 0x40058c0x40058c andand
instruction pointer isinstruction pointer is 0x40048c0x40048c
myDumpmyDump contains the assembly code of my appcontains the assembly code of my app
Now I will see at what line of code, my appNow I will see at what line of code, my app
crashed → just find where iscrashed → just find where is 0x40048c0x40048c inin
myDumpmyDump
$grep$grep -n -A 100 -B 100 “40048c” ./myDump-n -A 100 -B 100 “40048c” ./myDump
What it does it just find the line having 40048c in ./myDump, and also showWhat it does it just find the line having 40048c in ./myDump, and also show
100 more lines after the found line, and 100 lines before the found line. You100 more lines after the found line, and 100 lines before the found line. You
can customize the grep command as you want ;)can customize the grep command as you want ;)

← Step 1: Find instruction pointer address
← Step 2: Found, This caused segfault
← Step 3: Look above to see the code
that caused segfault in what function?
Here it is in main()

Now you know the code that cause Segfault inNow you know the code that cause Segfault in
assemblyassembly..
What to do is open your source code (in C, C++What to do is open your source code (in C, C++
…) to see the appropriate line of code…) to see the appropriate line of code
corresponding to that Assembly code, you willcorresponding to that Assembly code, you will
figure out what caused Segfault :-)figure out what caused Segfault :-)

0x48 = 'H'

You are done now! Bravo !!!You are done now! Bravo !!!
Just with “Just with “Instruction PointerInstruction Pointer””, you know where, you know where
caused Segfault exactlycaused Segfault exactly
How about the “How about the “Address caused faultAddress caused fault””
((0x40058c0x40058c), we have not used it, haven't we?), we have not used it, haven't we?
No we don't. BUT, till now I can say the line caused Segfault isNo we don't. BUT, till now I can say the line caused Segfault is *s = 'H';*s = 'H';
And, the address of variableAnd, the address of variable ss at that time is 0x40058cat that time is 0x40058c
Meaningless to know this?Meaningless to know this? NO! There is sometime you will need it to knowNO! There is sometime you will need it to know
the root cause, see the next slidethe root cause, see the next slide

Sometimes, the “Sometimes, the “Address caused faultAddress caused fault” tell you” tell you
the root cause. See the following example, wethe root cause. See the following example, we
can say that, value ofcan say that, value of ss isis N U LLN U LL

Thanks for watchingThanks for watchingIf you see it useful → clap your hands :-)If you see it useful → clap your hands :-)

Introduction to segmentation fault handling

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Introduction to segmentation fault handling (20)

Recently uploaded (20)

Introduction to segmentation fault handling