| 1 | <HTML>
|
|---|
| 2 | <HEAD>
|
|---|
| 3 | <TITLE>Using the Garbage Collector as Leak Detector</title>
|
|---|
| 4 | </head>
|
|---|
| 5 | <BODY>
|
|---|
| 6 | <H1>Using the Garbage Collector as Leak Detector</h1>
|
|---|
| 7 | The garbage collector may be used as a leak detector.
|
|---|
| 8 | In this case, the primary function of the collector is to report
|
|---|
| 9 | objects that were allocated (typically with <TT>GC_MALLOC</tt>),
|
|---|
| 10 | not deallocated (normally with <TT>GC_FREE</tt>), but are
|
|---|
| 11 | no longer accessible. Since the object is no longer accessible,
|
|---|
| 12 | there in normally no way to deallocate the object at a later time;
|
|---|
| 13 | thus it can safely be assumed that the object has been "leaked".
|
|---|
| 14 | <P>
|
|---|
| 15 | This is substantially different from counting leak detectors,
|
|---|
| 16 | which simply verify that all allocated objects are eventually
|
|---|
| 17 | deallocated. A garbage-collector based leak detector can provide
|
|---|
| 18 | somewhat more precise information when an object was leaked.
|
|---|
| 19 | More importantly, it does not report objects that are never
|
|---|
| 20 | deallocated because they are part of "permanent" data structures.
|
|---|
| 21 | Thus it does not require all objects to be deallocated at process
|
|---|
| 22 | exit time, a potentially useless activity that often triggers
|
|---|
| 23 | large amounts of paging.
|
|---|
| 24 | <P>
|
|---|
| 25 | All non-ancient versions of the garbage collector provide
|
|---|
| 26 | leak detection support. Version 5.3 adds the following
|
|---|
| 27 | features:
|
|---|
| 28 | <OL>
|
|---|
| 29 | <LI> Leak detection mode can be initiated at run-time by
|
|---|
| 30 | setting GC_find_leak instead of building the collector with FIND_LEAK
|
|---|
| 31 | defined. This variable should be set to a nonzero value
|
|---|
| 32 | at program startup.
|
|---|
| 33 | <LI> Leaked objects should be reported and then correctly garbage collected.
|
|---|
| 34 | Prior versions either reported leaks or functioned as a garbage collector.
|
|---|
| 35 | </ol>
|
|---|
| 36 | For the rest of this description we will give instructions that work
|
|---|
| 37 | with any reasonable version of the collector.
|
|---|
| 38 | <P>
|
|---|
| 39 | To use the collector as a leak detector, follow the following steps:
|
|---|
| 40 | <OL>
|
|---|
| 41 | <LI> Build the collector with -DFIND_LEAK. Otherwise use default
|
|---|
| 42 | build options.
|
|---|
| 43 | <LI> Change the program so that all allocation and deallocation goes
|
|---|
| 44 | through the garbage collector.
|
|---|
| 45 | <LI> Arrange to call <TT>GC_gcollect</tt> at appropriate points to check
|
|---|
| 46 | for leaks.
|
|---|
| 47 | (For sufficiently long running programs, this will happen implicitly,
|
|---|
| 48 | but probably not with sufficient frequency.)
|
|---|
| 49 | </ol>
|
|---|
| 50 | The second step can usually be accomplished with the
|
|---|
| 51 | <TT>-DREDIRECT_MALLOC=GC_malloc</tt> option when the collector is built,
|
|---|
| 52 | or by defining <TT>malloc</tt>, <TT>calloc</tt>,
|
|---|
| 53 | <TT>realloc</tt> and <TT>free</tt>
|
|---|
| 54 | to call the corresponding garbage collector functions.
|
|---|
| 55 | But this, by itself, will not yield very informative diagnostics,
|
|---|
| 56 | since the collector does not keep track of information about
|
|---|
| 57 | how objects were allocated. The error reports will include
|
|---|
| 58 | only object addresses.
|
|---|
| 59 | <P>
|
|---|
| 60 | For more precise error reports, as much of the program as possible
|
|---|
| 61 | should use the all uppercase variants of these functions, after
|
|---|
| 62 | defining <TT>GC_DEBUG</tt>, and then including <TT>gc.h</tt>.
|
|---|
| 63 | In this environment <TT>GC_MALLOC</tt> is a macro which causes
|
|---|
| 64 | at least the file name and line number at the allocation point to
|
|---|
| 65 | be saved as part of the object. Leak reports will then also include
|
|---|
| 66 | this information.
|
|---|
| 67 | <P>
|
|---|
| 68 | Many collector features (<I>e.g</i> stubborn objects, finalization,
|
|---|
| 69 | and disappearing links) are less useful in this context, and are not
|
|---|
| 70 | fully supported. Their use will usually generate additional bogus
|
|---|
| 71 | leak reports, since the collector itself drops some associated objects.
|
|---|
| 72 | <P>
|
|---|
| 73 | The same is generally true of thread support. However, as of 6.0alpha4,
|
|---|
| 74 | correct leak reports should be generated with linuxthreads.
|
|---|
| 75 | <P>
|
|---|
| 76 | On a few platforms (currently Solaris/SPARC, Irix, and, with -DSAVE_CALL_CHAIN,
|
|---|
| 77 | Linux/X86), <TT>GC_MALLOC</tt>
|
|---|
| 78 | also causes some more information about its call stack to be saved
|
|---|
| 79 | in the object. Such information is reproduced in the error
|
|---|
| 80 | reports in very non-symbolic form, but it can be very useful with the
|
|---|
| 81 | aid of a debugger.
|
|---|
| 82 | <H2>An Example</h2>
|
|---|
| 83 | The following header file <TT>leak_detector.h</tt> is included in the
|
|---|
| 84 | "include" subdirectory of the distribution:
|
|---|
| 85 | <PRE>
|
|---|
| 86 | #define GC_DEBUG
|
|---|
| 87 | #include "gc.h"
|
|---|
| 88 | #define malloc(n) GC_MALLOC(n)
|
|---|
| 89 | #define calloc(m,n) GC_MALLOC((m)*(n))
|
|---|
| 90 | #define free(p) GC_FREE(p)
|
|---|
| 91 | #define realloc(p,n) GC_REALLOC((p),(n))
|
|---|
| 92 | #define CHECK_LEAKS() GC_gcollect()
|
|---|
| 93 | </pre>
|
|---|
| 94 | <P>
|
|---|
| 95 | Assume the collector has been built with -DFIND_LEAK. (For very
|
|---|
| 96 | new versions of the collector, we could instead add the statement
|
|---|
| 97 | <TT>GC_find_leak = 1</tt> as the first statement in <TT>main</tt>.
|
|---|
| 98 | <P>
|
|---|
| 99 | The program to be tested for leaks can then look like:
|
|---|
| 100 | <PRE>
|
|---|
| 101 | #include "leak_detector.h"
|
|---|
| 102 |
|
|---|
| 103 | main() {
|
|---|
| 104 | int *p[10];
|
|---|
| 105 | int i;
|
|---|
| 106 | /* GC_find_leak = 1; for new collector versions not */
|
|---|
| 107 | /* compiled with -DFIND_LEAK. */
|
|---|
| 108 | for (i = 0; i < 10; ++i) {
|
|---|
| 109 | p[i] = malloc(sizeof(int)+i);
|
|---|
| 110 | }
|
|---|
| 111 | for (i = 1; i < 10; ++i) {
|
|---|
| 112 | free(p[i]);
|
|---|
| 113 | }
|
|---|
| 114 | for (i = 0; i < 9; ++i) {
|
|---|
| 115 | p[i] = malloc(sizeof(int)+i);
|
|---|
| 116 | }
|
|---|
| 117 | CHECK_LEAKS();
|
|---|
| 118 | }
|
|---|
| 119 | </pre>
|
|---|
| 120 | <P>
|
|---|
| 121 | On an Intel X86 Linux system this produces on the stderr stream:
|
|---|
| 122 | <PRE>
|
|---|
| 123 | Leaked composite object at 0x806dff0 (leak_test.c:8, sz=4)
|
|---|
| 124 | </pre>
|
|---|
| 125 | (On most unmentioned operating systems, the output is similar to this.
|
|---|
| 126 | If the collector had been built on Linux/X86 with -DSAVE_CALL_CHAIN,
|
|---|
| 127 | the output would be closer to the Solaris example. For this to work,
|
|---|
| 128 | the program should not be compiled with -fomit_frame_pointer.)
|
|---|
| 129 | <P>
|
|---|
| 130 | On Irix it reports
|
|---|
| 131 | <PRE>
|
|---|
| 132 | Leaked composite object at 0x10040fe0 (leak_test.c:8, sz=4)
|
|---|
| 133 | Caller at allocation:
|
|---|
| 134 | ##PC##= 0x10004910
|
|---|
| 135 | </pre>
|
|---|
| 136 | and on Solaris the error report is
|
|---|
| 137 | <PRE>
|
|---|
| 138 | Leaked composite object at 0xef621fc8 (leak_test.c:8, sz=4)
|
|---|
| 139 | Call chain at allocation:
|
|---|
| 140 | args: 4 (0x4), 200656 (0x30FD0)
|
|---|
| 141 | ##PC##= 0x14ADC
|
|---|
| 142 | args: 1 (0x1), -268436012 (0xEFFFFDD4)
|
|---|
| 143 | ##PC##= 0x14A64
|
|---|
| 144 | </pre>
|
|---|
| 145 | In the latter two cases some additional information is given about
|
|---|
| 146 | how malloc was called when the leaked object was allocated. For
|
|---|
| 147 | Solaris, the first line specifies the arguments to <TT>GC_debug_malloc</tt>
|
|---|
| 148 | (the actual allocation routine), The second the program counter inside
|
|---|
| 149 | main, the third the arguments to <TT>main</tt>, and finally the program
|
|---|
| 150 | counter inside the caller to main (i.e. in the C startup code).
|
|---|
| 151 | <P>
|
|---|
| 152 | In the Irix case, only the address inside the caller to main is given.
|
|---|
| 153 | <P>
|
|---|
| 154 | In many cases, a debugger is needed to interpret the additional information.
|
|---|
| 155 | On systems supporting the "adb" debugger, the <TT>callprocs</tt> script
|
|---|
| 156 | can be used to replace program counter values with symbolic names.
|
|---|
| 157 | As of version 6.1, the collector tries to generate symbolic names for
|
|---|
| 158 | call stacks if it knows how to do so on the platform. This is true on
|
|---|
| 159 | Linux/X86, but not on most other platforms.
|
|---|
| 160 | <H2>Simplified leak detection under Linux</h2>
|
|---|
| 161 | Since version 6.1, it should be possible to run the collector in leak
|
|---|
| 162 | detection mode on a program a.out under Linux/X86 as follows:
|
|---|
| 163 | <OL>
|
|---|
| 164 | <LI> Ensure that a.out is a single-threaded executable. This doesn't yet work
|
|---|
| 165 | for multithreaded programs.
|
|---|
| 166 | <LI> If possible, ensure that the addr2line program is installed in
|
|---|
| 167 | /usr/bin. (It comes with RedHat Linux.)
|
|---|
| 168 | <LI> If possible, compile a.out with full debug information.
|
|---|
| 169 | This will improve the quality of the leak reports. With this approach, it is
|
|---|
| 170 | no longer necessary to call GC_ routines explicitly, though that can also
|
|---|
| 171 | improve the quality of the leak reports.
|
|---|
| 172 | <LI> Build the collector and install it in directory <I>foo</i> as follows:
|
|---|
| 173 | <UL>
|
|---|
| 174 | <LI> configure --prefix=<I>foo</i> --enable-full-debug --enable-redirect-malloc
|
|---|
| 175 | --disable-threads
|
|---|
| 176 | <LI> make
|
|---|
| 177 | <LI> make install
|
|---|
| 178 | </ul>
|
|---|
| 179 | <LI> Set environment variables as follows:
|
|---|
| 180 | <UL>
|
|---|
| 181 | <LI> LD_PRELOAD=<I>foo</i>/lib/libgc.so
|
|---|
| 182 | <LI> GC_FIND_LEAK
|
|---|
| 183 | <LI> You may also want to set GC_PRINT_STATS (to confirm that the collector
|
|---|
| 184 | is running) and/or GC_LOOP_ON_ABORT (to facilitate debugging from another
|
|---|
| 185 | window if something goes wrong).
|
|---|
| 186 | </ul
|
|---|
| 187 | <LI> Simply run a.out as you normally would. Note that if you run anything
|
|---|
| 188 | else (<I>e.g.</i> your editor) with those environment variables set,
|
|---|
| 189 | it will also be leak tested. This may or may not be useful and/or
|
|---|
| 190 | embarrassing. It can generate
|
|---|
| 191 | mountains of leak reports if the application wasn't designed to avoid leaks,
|
|---|
| 192 | <I>e.g.</i> because it's always short-lived.
|
|---|
| 193 | </ol>
|
|---|
| 194 | This has not yet been thropughly tested on large applications, but it's known
|
|---|
| 195 | to do the right thing on at least some small ones.
|
|---|
| 196 | </body>
|
|---|
| 197 | </html>
|
|---|
