blob: 13461b55da5151d6270fed331c8b119eeb8fbebd [file] [log] [blame]
The Android Open Source Projectf6c38712009-03-03 19:28:47 -08001<html>
2 <head>
3 <title>Android JNI Tips</title>
4 <link rel=stylesheet href="android.css">
5 </head>
6
7 <body>
8 <h1><a name="JNI_Tips"></a>Android JNI Tips</h1>
9<p>
10</p><p>
11</p><ul>
12<li> <a href="#What_s_JNI_">What's JNI?</a>
13</li>
14<li> <a href="#JavaVM_and_JNIEnv">JavaVM and JNIEnv</a>
15
16</li>
17<li> <a href="#jclassID_jmethodID_and_jfieldID">jclassID, jmethodID, and jfieldID</a>
18</li>
19<li> <a href="#local_vs_global_references">Local vs. Global References</a>
20</li>
21<li> <a href="#UTF_8_and_UTF_16_strings">UTF-8 and UTF-16 Strings</a>
22</li>
23<li> <a href="#Arrays">Primitive Arrays</a>
24</li>
25<li> <a href="#RegionCalls">Region Calls</a>
26</li>
27<li> <a href="#Exceptions">Exceptions</a>
28</li>
29
30<li> <a href="#Extended_checking">Extended Checking</a>
31</li>
32<li> <a href="#Native_Libraries">Native Libraries</a>
33</li>
34<li> <a href="#64bit">64-bit Considerations</a>
35</li>
36
37<li> <a href="#Unsupported">Unsupported Features</a>
38</ul>
39<p>
40<noautolink>
41</noautolink></p><p>
42</p><h2><a name="What_s_JNI_"> </a> What's JNI? </h2>
43<p>
44
45JNI is the Java Native Interface. It defines a way for code written in the
46Java programming language to interact with native
47code, e.g. functions written in C/C++. It's VM-neutral, has support for loading code from
48dynamic shared libraries, and while cumbersome at times is reasonably efficient.
49</p><p>
50You really should read through the
51<a href="http://java.sun.com/javase/6/docs/technotes/guides/jni/spec/jniTOC.html">JNI spec for J2SE 1.6</a>
52to get a sense for how JNI works and what features are available. Some
53aspects of the interface aren't immediately obvious on
54first reading, so you may find the next few sections handy.
55The more detailed <i>JNI Programmer's Guide and Specification</i> can be found
56<a href="http://java.sun.com/docs/books/jni/html/jniTOC.html">here</a>.
57</p><p>
58</p><p>
59</p><h2><a name="JavaVM_and_JNIEnv"> </a> JavaVM and JNIEnv </h2>
60<p>
61JNI defines two key data structures, "JavaVM" and "JNIEnv". Both of these are essentially
62pointers to pointers to function tables. (In the C++ version, it's a class whose sole member
63is a pointer to a function table.) The JavaVM provides the "invocation interface" functions,
64which allow you to create and destroy the VM. In theory you can have multiple VMs per process,
65but Android's VMs only allow one.
66</p><p>
67The JNIEnv provides most of the JNI functions. Your native functions all receive a JNIEnv as
68the first argument.
69</p><p>
70
71On some VMs, the JNIEnv is used for thread-local storage. For this reason, <strong>you cannot share a JNIEnv between threads</strong>.
72If a piece of code has no other way to get its JNIEnv, you should share
73the JavaVM, and use JavaVM-&gt;GetEnv to discover the thread's JNIEnv.
74</p><p>
75The C and C++ declarations of JNIEnv and JavaVM are different. "jni.h" provides different typedefs
76depending on whether it's included into ".c" or ".cpp". For this reason it's a bad idea to
77include JNIEnv arguments in header files included by both languages. (Put another way: if your
78header file requires "#ifdef __cplusplus", you may have to do some extra work if anything in
79that header refers to JNIEnv.)
80</p><p>
81</p><p>
82</p><h2><a name="jclassID_jmethodID_and_jfieldID"> jclassID, jmethodID, and jfieldID </a></h2>
83<p>
84If you want to access an object's field from native code, you would do the following:
85</p><p>
86</p><ul>
87<li> Get the class object reference for the class with <code>FindClass</code>
88</li>
89<li> Get the field ID for the field with <code>GetFieldID</code>
90</li>
91<li> Get the contents of the field with something appropriate, e.g.
92<code>GetIntField</code>
93</li>
94</ul>
95<p>
96Similarly, to call a method, you'd first get a class object reference and then a method ID. The IDs are often just
97pointers to internal VM data structures. Looking them up may require several string
98comparisons, but once you have them the actual call to get the field or invoke the method
99is very quick.
100</p><p>
101If performance is important, it's useful to look the values up once and cache the results
102in your native code. Because we are limiting ourselves to one VM per process, it's reasonable
103to store this data in a static local structure.
104</p><p>
105The class references, field IDs, and method IDs are guaranteed valid until the class is unloaded. Classes
106are only unloaded if all classes associated with a ClassLoader can be garbage collected,
107which is rare but will not be impossible in our system. The jclassID
108is a class reference and <strong>must be protected</strong> with a call
109to <code>NewGlobalRef</code> (see the next section).
110</p><p>
111If you would like to cache the IDs when a class is loaded, and automatically re-cache them
112if the class is ever unloaded and reloaded, the correct way to initialize
113the IDs is to add a piece of code that looks like this to the appropriate class:
114</p><p>
115
116</p><pre> /*
117 * We use a class initializer to allow the native code to cache some
118 * field offsets.
119 */
120
121 /*
122 * A native function that looks up and caches interesting
123 * class/field/method IDs for this class. Returns false on failure.
124 */
125 native private static boolean nativeClassInit();
126
127 /*
128 * Invoke the native initializer when the class is loaded.
129 */
130 static {
131 if (!nativeClassInit())
132 throw new RuntimeException("native init failed");
133 }
134</pre>
135<p>
136Create a nativeClassInit method in your C/C++ code that performs the ID lookups. The code
137will be executed once, when the class is initialized. If the class is ever unloaded and
138then reloaded, it will be executed again. (See the implementation of java.io.FileDescriptor
139for an example in our source tree.)
140</p><p>
141</p><p>
142</p><p>
143</p><h2><a name="local_vs_global_references"> Local vs. Global References </a></h2>
144<p>
145Every object that JNI returns is a "local reference". This means that it's valid for the
146duration of the current native method in the current thread.
147<strong>Even if the object itself continues to live on after the native method returns, the reference is not valid.</strong>
148This applies to all sub-classes of jobject, including jclass and jarray.
149(Dalvik VM will warn you about this when -Xcheck:jni is enabled.)
150</p><p>
151
152If you want to hold on to a reference for a longer period, you must use a "global" reference.
153The <code>NewGlobalRef</code> function takes the local reference as
154an argument and returns a global one:
155
156<p><pre>jobject* localRef = [...];
157jobject* globalRef;
158globalRef = env-&gt;NewGlobalRef(localRef);
159</pre>
160
161The global reference is guaranteed to be valid until you call
Andy McFadden278c12b2009-09-21 16:28:39 -0700162<code>DeleteGlobalRef</code>. This pattern is frequently used for
163cached copies of class objects obtained from <code>FindClass</code>.
164</p><p>
The Android Open Source Projectf6c38712009-03-03 19:28:47 -0800165</p><p>
166All JNI methods accept both local and global references as arguments.
Andy McFadden5aca6032009-05-06 16:48:33 -0700167It's possible for references to the same object to have different values;
168for example, the return values from consecutive calls to
169<code>NewGlobalRef</code> on the same object may be different.
170<strong>To see if two references refer to the same object,
171you must use the <code>IsSameObject</code> function.</strong> Never compare
172references with "==" in native code.
The Android Open Source Projectf6c38712009-03-03 19:28:47 -0800173</p><p>
Andy McFaddenc7659ec2009-09-18 16:14:41 -0700174One consequence of this is that you
175<strong>must not assume object references are constant</strong>
176in native code. The 32-bit value representing an object may be different
177from one invocation of a method to the next, and it's possible that two
178different objects could have the same 32-bit value at different times. Do
179not use jobjects as keys.
180</p><p>
The Android Open Source Projectf6c38712009-03-03 19:28:47 -0800181Programmers are required to "not excessively allocate" local references. In practical terms this means
182that if you're creating large numbers of local references, perhaps while running through an array of
183Objects, you should free them manually with
184<code>DeleteLocalRef</code> instead of letting JNI do it for you. The
185VM is only required to reserve slots for
18616 local references, so if you need more than that you should either delete as you go or use
187<code>EnsureLocalCapacity</code> to reserve more.
188</p><p>
189Note: method and field IDs are just 32-bit identifiers, not object
190references, and should not be passed to <code>NewGlobalRef</code>. The raw data
191pointers returned by functions like <code>GetStringUTFChars</code>
192and <code>GetByteArrayElements</code> are also not objects.
193</p><p>
194One unusual case deserves separate mention. If you attach a native
195thread to the VM with AttachCurrentThread, the code you are running will
196never "return" to the VM until the thread detaches from the VM. Any local
197references you create will have to be deleted manually unless the thread
198is about to exit or detach.
199</p><p>
200</p><p>
201</p><p>
202</p><h2><a name="UTF_8_and_UTF_16_strings"> </a> UTF-8 and UTF-16 Strings </h2>
203<p>
204The Java programming language uses UTF-16. For convenience, JNI provides methods that work with "modified UTF-8" encoding
205as well. (Some VMs use the modified UTF-8 internally to store strings; ours do not.) The
206modified encoding only supports the 8- and 16-bit forms, and stores ASCII NUL values in a 16-bit encoding.
207The nice thing about it is that you can count on having C-style zero-terminated strings,
208suitable for use with standard libc string functions. The down side is that you cannot pass
209arbitrary UTF-8 data into the VM and expect it to work correctly.
210</p><p>
211It's usually best to operate with UTF-16 strings. With our current VMs, the
212<code>GetStringChars</code> method
213does not require a copy, whereas <code>GetStringUTFChars</code> requires a malloc and a UTF conversion. Note that
214<strong>UTF-16 strings are not zero-terminated</strong>, and \u0000 is allowed,
215so you need to hang on to the string length as well as
216the string pointer.
217
218</p><p>
Andy McFadden5aca6032009-05-06 16:48:33 -0700219<strong>Don't forget to Release the strings you Get</strong>. The
220string functions return <code>jchar*</code> or <code>jbyte*</code>, which
221are C-style pointers to primitive data rather than local references. They
222are guaranteed valid until Release is called, which means they are not
The Android Open Source Projectf6c38712009-03-03 19:28:47 -0800223released when the native method returns.
224</p><p>
225</p><p>
226
227
228</p><h2><a name="Arrays"> </a> Primitive Arrays </h2>
229<p>
230JNI provides functions for accessing the contents of array objects.
231While arrays of objects must be accessed one entry at a time, arrays of
232primitives can be read and written directly as if they were declared in C.
233</p><p>
234To make the interface as efficient as possible without constraining
235the VM implementation,
236the <code>Get&lt;PrimitiveType&gt;ArrayElements</code> family of calls
237allows the VM to either return a pointer to the actual elements, or
238allocate some memory and make a copy. Either way, the raw pointer returned
239is guaranteed to be valid until the corresponding <code>Release</code> call
240is issued (which implies that, if the data wasn't copied, the array object
241will be pinned down and can't be relocated as part of compacting the heap).
242<strong>You must Release every array you Get.</strong> Also, if the Get
243call fails, you must ensure that your code doesn't try to Release a NULL
244pointer later.
245</p><p>
246You can determine whether or not the data was copied by passing in a
247non-NULL pointer for the <code>isCopy</code> argument. This is rarely
248useful.
249</p><p>
250The <code>Release</code> call takes a <code>mode</code> argument that can
251have one of three values. The actions performed by the VM depend upon
252whether it returned a pointer to the actual data or a copy of it:
253<ul>
254 <li><code>0</code>
255 <ul>
256 <li>Actual: the array object is un-pinned.
257 <li>Copy: data is copied back. The buffer with the copy is freed.
258 </ul>
259 <li><code>JNI_COMMIT</code>
260 <ul>
261 <li>Actual: does nothing.
262 <li>Copy: data is copied back. The buffer with the copy
263 <strong>is not freed</strong>.
264 </ul>
265 <li><code>JNI_ABORT</code>
266 <ul>
267 <li>Actual: the array object is un-pinned. Earlier
268 writes are <strong>not</strong> aborted.
269 <li>Copy: the buffer with the copy is freed; any changes to it are lost.
270 </ul>
271</ul>
272</p><p>
273One reason for checking the <code>isCopy</code> flag is to know if
274you need to call <code>Release</code> with <code>JNI_COMMIT</code>
275after making changes to an array -- if you're alternating between making
276changes and executing code that uses the contents of the array, you may be
277able to
278skip the no-op commit. Another possible reason for checking the flag is for
279efficient handling of <code>JNI_ABORT</code>. For example, you might want
280to get an array, modify it in place, pass pieces to other functions, and
281then discard the changes. If you know that JNI is making a new copy for
282you, there's no need to create another "editable" copy. If JNI is passing
283you the original, then you do need to make your own copy.
284</p><p>
285Some have asserted that you can skip the <code>Release</code> call if
286<code>*isCopy</code> is false. This is not the case. If no copy buffer was
287allocated, then the original memory must be pinned down and can't be moved by
288the garbage collector.
289</p><p>
290Also note that the <code>JNI_COMMIT</code> flag does NOT release the array,
291and you will need to call <code>Release</code> again with a different flag
292eventually.
293</p><p>
294</p><p>
295
296
297</p><h2><a name="RegionCalls"> Region Calls </a></h2>
298
299<p>
300There is an alternative to calls like <code>Get&lt;Type&gt;ArrayElements</code>
301and <code>GetStringChars</code> that may be very helpful when all you want
302to do is copy data in or out. Consider the following:
303<pre>
304 jbyte* data = env->GetByteArrayElements(array, NULL);
305 if (data != NULL) {
306 memcpy(buffer, data, len);
307 env->ReleaseByteArrayElements(array, data, JNI_ABORT);
308 }
309</pre>
310<p>
311This grabs the array, copies the first <code>len</code> byte
312elements out of it, and then releases the array. Depending upon the VM
313policies the <code>Get</code> call will either pin or copy the array contents.
314We copy the data (for perhaps a second time), then call Release; in this case
315we use <code>JNI_ABORT</code> so there's no chance of a third copy.
316</p><p>
317We can accomplish the same thing with this:
318<pre>
319 env->GetByteArrayRegion(array, 0, len, buffer);
320</pre>
321</p><p>
322This accomplishes the same thing, with several advantages:
323<ul>
Andy McFaddenc7659ec2009-09-18 16:14:41 -0700324 <li>Requires one JNI call instead of 2, reducing overhead.
The Android Open Source Projectf6c38712009-03-03 19:28:47 -0800325 <li>Doesn't require pinning or extra data copies.
Andy McFaddenc7659ec2009-09-18 16:14:41 -0700326 <li>Reduces the risk of programmer error -- no risk of forgetting
327 to call <code>Release</code> after something fails.
The Android Open Source Projectf6c38712009-03-03 19:28:47 -0800328</ul>
329</p><p>
330Similarly, you can use the <code>Set&lt;Type&gt;ArrayRegion</code> call
331to copy data into an array, and <code>GetStringRegion</code> or
332<code>GetStringUTFRegion</code> to copy characters out of a
333<code>String</code>.
334
335
336</p><h2><a name="Exceptions"> Exceptions </a></h2>
337<p>
338<strong>You may not call most JNI functions while an exception is pending.</strong>
339Your code is expected to notice the exception (via the function's return value,
340<code>ExceptionCheck()</code>, or <code>ExceptionOccurred()</code>) and return,
341or clear the exception and handle it.
342</p><p>
343The only JNI functions that you are allowed to call while an exception is
344pending are:
345<font size="-1"><ul>
346 <li>DeleteGlobalRef
347 <li>DeleteLocalRef
348 <li>DeleteWeakGlobalRef
349 <li>ExceptionCheck
350 <li>ExceptionClear
351 <li>ExceptionDescribe
352 <li>ExceptionOccurred
353 <li>MonitorExit
354 <li>PopLocalFrame
355 <li>PushLocalFrame
356 <li>Release<PrimitiveType>ArrayElements
357 <li>ReleasePrimitiveArrayCritical
358 <li>ReleaseStringChars
359 <li>ReleaseStringCritical
360 <li>ReleaseStringUTFChars
361</ul></font>
362</p><p>
363Note that exceptions thrown by interpreted code do not "leap over" native code,
364and C++ exceptions thrown by native code are not handled by Dalvik.
365The JNI <code>Throw</code> and <code>ThrowNew</code> instructions just
366set an exception pointer in the current thread. Upon returning to the VM from
367native code, the exception will be noted and handled appropriately.
368</p><p>
369Native code can "catch" an exception by calling <code>ExceptionCheck</code> or
370<code>ExceptionOccurred</code>, and clear it with
371<code>ExceptionClear</code>. As usual,
372discarding exceptions without handling them can lead to problems.
373</p><p>
374There are no built-in functions for manipulating the Throwable object
375itself, so if you want to (say) get the exception string you will need to
376find the Throwable class, look up the method ID for
377<code>getMessage "()Ljava/lang/String;"</code>, invoke it, and if the result
378is non-NULL use <code>GetStringUTFChars</code> to get something you can
379hand to printf or a LOG macro.
380
381</p><p>
382</p><p>
383</p><h2><a name="Extended_checking"> Extended Checking </a></h2>
384<p>
385JNI does very little error checking. Calling <code>SetFieldInt</code>
386on an Object field will succeed, even if the field is marked
387<code>private</code> and <code>final</code>. The
388goal is to minimize the overhead on the assumption that, if you've written it in native code,
389you probably did it for performance reasons.
390</p><p>
391Some VMs support extended checking with the "<code>-Xcheck:jni</code>" flag. If the flag is set, the VM
392puts a different table of functions into the JavaVM and JNIEnv pointers. These functions do
393an extended series of checks before calling the standard implementation.
394
395</p><p>
Andy McFaddenc7659ec2009-09-18 16:14:41 -0700396Some things that may be checked:
The Android Open Source Projectf6c38712009-03-03 19:28:47 -0800397</p><p>
398</p>
399<ul>
400<li> Check for null pointers where not allowed.
401<li>
402<li> Verify argument type correctness (jclass is a class object,
403jfieldID points to field data, jstring is a java.lang.String).
404</li>
405<li> Field type correctness, e.g. don't store a HashMap in a String field.
406</li>
407<li> Check to see if an exception is pending on calls where pending exceptions are not legal.
408</li>
409<li> Check for calls to inappropriate functions between Critical get/release calls.
410</li>
411<li> Check that JNIEnv structs aren't being shared between threads.
412
413</li>
414<li> Make sure local references aren't used outside their allowed lifespan.
415</li>
416<li> UTF-8 strings contain valid "modified UTF-8" data.
417</li>
418</ul>
419<p>Accessibility of methods and fields (i.e. public vs. private) is not
420checked.
421<p>
422The Dalvik VM supports the <code>-Xcheck:jni</code> flag. For a
423description of how to enable it for Android apps, see
424<a href="embedded-vm-control.html">Controlling the Embedded VM</a>.
425It's currently enabled by default in the Android emulator and on
426"engineering" device builds.
427
428</p><p>
429JNI checks can be modified with the <code>-Xjniopts</code> command-line
430flag. Currently supported values include:
431</p>
432<blockquote><dl>
433<dt>forcecopy
434<dd>When set, any function that can return a copy of the original data
435(array of primitive values, UTF-16 chars) will always do so. The buffers
436are over-allocated and surrounded with a guard pattern to help identify
437code writing outside the buffer, and the contents are erased before the
438storage is freed to trip up code that uses the data after calling Release.
Andy McFaddenc7659ec2009-09-18 16:14:41 -0700439This will have a noticeable performance impact on some applications.
The Android Open Source Projectf6c38712009-03-03 19:28:47 -0800440<dt>warnonly
441<dd>By default, JNI "warnings" cause the VM to abort. With this flag
442it continues on.
443</dl></blockquote>
444
445
446</p><p>
447</p><h2><a name="Native_Libraries"> Native Libraries </a></h2>
448<p>
449You can load native code from shared libraries with the standard
450<code>System.loadLibrary()</code> call. The
451preferred way to get at your native code is:
452</p><p>
453</p><ul>
454<li> Call <code>System.loadLibrary()</code> from a static class initializer. (See the earlier example, where one is used to call nativeClassInit().) The argument is the "undecorated" library name, e.g. to load "libfubar.so" you would pass in "fubar".
455
456</li>
457<li> Provide a native function: <code><strong>jint JNI_OnLoad(JavaVM* vm, void* reserved)</strong></code>
458</li>
459<li>In <code>JNI_OnLoad</code>, register all of your native methods. You
460should declare
461the methods "static" so the names don't take up space in the symbol table
462on the device.
463</li>
464</ul>
465<p>
466The <code>JNI_OnLoad</code> function should look something like this if
467written in C:
468</p><blockquote><pre>jint JNI_OnLoad(JavaVM* vm, void* reserved)
469{
470 JNIEnv* env;
471 if ((*vm)->GetEnv(vm, (void**) &env, JNI_VERSION_1_4) != JNI_OK)
472 return -1;
473
474 /* get class with (*env)->FindClass */
475 /* register methods with (*env)->RegisterNatives */
476
477 return JNI_VERSION_1_4;
478}
479</pre></blockquote>
480</p><p>
481You can also call <code>System.load()</code> with the full path name of the
482shared library. For Android apps, you can get the full path to the
483application's private data storage area from the context object.
484</p><p>
Andy McFadden278c12b2009-09-21 16:28:39 -0700485This is the recommended approach, but not the only approach. The VM does
486not require explicit registration, nor that you provide a
487<code>JNI_OnLoad</code> function.
488You can instead use "discovery" of native methods that are named in a
The Android Open Source Projectf6c38712009-03-03 19:28:47 -0800489specific way (see <a href="http://java.sun.com/javase/6/docs/technotes/guides/jni/spec/design.html#wp615">
Andy McFadden278c12b2009-09-21 16:28:39 -0700490 the JNI spec</a> for details), though this is less desirable.
491It requires more space in the shared object symbol table,
The Android Open Source Projectf6c38712009-03-03 19:28:47 -0800492loading is slower because it requires string searches through all of the
493loaded shared libraries, and if a method signature is wrong you won't know
494about it until the first time the method is actually used.
495</p><p>
Andy McFadden278c12b2009-09-21 16:28:39 -0700496One other note about <code>JNI_OnLoad</code>: any <code>FindClass</code>
497calls you make from there will happen in the context of the class loader
498that was used to load the shared library. Normally <code>FindClass</code>
499uses the loader associated with the method at the top of the interpreted
500stack, or if there isn't one (because the thread was just attached to
501the VM) it uses the "system" class loader.
502</p><p>
The Android Open Source Projectf6c38712009-03-03 19:28:47 -0800503
504
505</p><h2><a name="64bit"> 64-bit Considerations </a></h2>
506
507<p>
508Android is currently expected to run on 32-bit platforms. In theory it
509could be built for a 64-bit system, but that is not a goal at this time.
510For the most part this isn't something that you will need to worry about
511when interacting with native code,
512but it becomes significant if you plan to store pointers to native
513structures in integer fields in an object. To support architectures
514that use 64-bit pointers, <strong>you need to stash your native pointers in a
515<code>long</code> field rather than an <code>int</code></strong>.
516
517
518</p><h2><a name="Unsupported"> Unsupported Features </a></h2>
519<p>All JNI 1.6 features are supported, with the following exceptions:
520<ul>
521 <li><code>DefineClass</code> is not implemented. Dalvik does not use
522 Java bytecodes or class files, so passing in binary class data
523 doesn't work. Translation facilities may be added in a future
524 version of the VM.</li>
525 <li><code>NewWeakGlobalRef</code> and <code>DeleteWeakGlobalRef</code>
526 are not implemented. The
527 VM supports weak references, but not JNI "weak global" references.
528 These will be supported in a future release.</li>
529 <li><code>GetObjectRefType</code> (new in 1.6) is implemented but not fully
530 functional -- it can't always tell the difference between "local" and
531 "global" references.</li>
532</ul>
533
534</p>
535
536<address>Copyright &copy; 2008 The Android Open Source Project</address>
537
538 </body>
539</html>