跳过导航链接 | |
退出打印视图 | |
![]() |
手册页第 5 部分:标准、环境和宏 Oracle Solaris 11.1 Information Library (简体中文) |
- overview of concepts related to POSIX thread cancellation
|
Thread cancellation allows a thread to terminate the execution of any application thread in the process. Cancellation is useful when further operations of one or more threads are undesirable or unnecessary.
An example of a situation that could benefit from using cancellation is an asynchronously-generated cancel condition such as a user requesting to close or exit some running operation. Another example is the completion of a task undertaken by a number of threads, such as solving a maze. While many threads search for the solution, one of the threads might solve the puzzle while the others continue to operate. Since they are serving no purpose at that point, they should all be canceled.
Planning and programming for most cancellations follow this pattern:
Identify which threads you want to cancel, and insert pthread_cancel(3C) statements.
Identify system-defined cancellation points where a thread that might be canceled could have changed system or program state that should be restored. See the Cancellation Points for a list.
When a thread changes the system or program state just before a cancellation point, and should restore that state before the thread is canceled, place a cleanup handler before the cancellation point with pthread_cleanup_push(3C). Wherever a thread restores the changed state, pop the cleanup handler from the cleanup stack with pthread_cleanup_pop(3C).
Know whether the threads you are canceling call into cancel-unsafe libraries, and disable cancellation with pthread_setcancelstate(3C) before the call into the library. See Cancellation State and Cancel-Safe.
To cancel a thread in a procedure that contains no cancellation points, insert your own cancellation points with pthread_testcancel(3C). This function creates cancellation points by testing for pending cancellations and performing those cancellations if they are found. Push and pop cleanup handlers around the cancellation point, if necessary (see Step 3, above).
The system defines certain points at which cancellation can occur (cancellation points), and you can create additional cancellation points in your application with pthread_testcancel().
The following cancellation points are defined by the system (system-defined cancellation points): creat(2), aio_suspend(3C), close(2), creat(2), getmsg(2), getpmsg(2), lockf(3C), mq_receive(3C), mq_send(3C), msgrcv(2), msgsnd(2), msync(3C), nanosleep(3C), open(2), pause(2), poll(2), pread(2), pthread_cond_timedwait(3C), pthread_cond_wait(3C), pthread_join(3C), pthread_testcancel(3C), putmsg(2), putpmsg(2), pwrite(2), read(2), readv(2), select(3C), sem_wait(3C), sigpause(3C), sigwaitinfo(3C), sigsuspend(2), sigtimedwait(3C), sigwait(2), sleep(3C), sync(2), system(3C), tcdrain(3C), usleep(3C), wait(3C), waitid(2), wait3(3C), waitpid(3C), write(2), writev(2), and fcntl(2), when specifying F_SETLKW as the command.
When cancellation is asynchronous, cancellation can occur at any time (before, during, or after the execution of the function defined as the cancellation point). When cancellation is deferred (the default case), cancellation occurs only within the scope of a function defined as a cancellation point (after the function is called and before the function returns). See Cancellation Type for more information about deferred and asynchronous cancellation.
Choosing where to place cancellation points and understanding how cancellation affects your program depend upon your understanding of both your application and of cancellation mechanics.
Typically, any call that might require a long wait should be a cancellation point. Operations need to check for pending cancellation requests when the operation is about to block indefinitely. This includes threads waiting in pthread_cond_wait() and pthread_cond_timedwait(), threads waiting for the termination of another thread in pthread_join(), and threads blocked on sigwait().
A mutex is explicitly not a cancellation point and should be held for only the minimal essential time.
Most of the dangers in performing cancellations deal with properly restoring invariants and freeing shared resources. For example, a carelessly canceled thread might leave a mutex in a locked state, leading to a deadlock. Or it might leave a region of memory allocated with no way to identify it and therefore no way to free it.
When a thread is canceled, it should release resources and clean up the state that is shared with other threads. So, whenever a thread that might be canceled changes the state of the system or of the program, be sure to push a cleanup handler with pthread_cleanup_push(3C) before the cancellation point.
When a thread is canceled, all the currently-stacked cleanup handlers are executed in last-in-first-out (LIFO) order. Each handler is run in the scope in which it was pushed. When the last cleanup handler returns, the thread-specific data destructor functions are called. Thread execution terminates when the last destructor function returns.
When, in the normal course of the program, an uncanceled thread restores state that it had previously changed, be sure to pop the cleanup handler (that you had set up where the change took place) using pthread_cleanup_pop(3C). That way, if the thread is canceled later, only currently-changed state will be restored by the handlers that are left in the stack.
The pthread_cleanup_push() and pthread_cleanup_pop() functions can be implemented as macros. The application must ensure that they appear as statements, and in pairs within the same lexical scope (that is, the pthread_cleanup_push() macro can be thought to expand to a token list whose first token is '{' with pthread_cleanup_pop() expanding to a token list whose last token is the corresponding '}').
The effect of the use of return, break, continue, and goto to prematurely leave a code block described by a pair of pthread_cleanup_push() and pthread_cleanup_pop() function calls is undefined.
Most programmers will use only the default cancellation state of PTHREAD_CANCEL_ENABLE, but can choose to change the state by using pthread_setcancelstate(3C), which determines whether a thread is cancelable at all. With the default state of PTHREAD_CANCEL_ENABLE, cancellation is enabled and the thread is cancelable at points determined by its cancellation type. See Cancellation Type.
If the state is PTHREAD_CANCEL_DISABLE, cancellation is disabled, the thread is not cancelable at any point, and all cancellation requests to it are held pending.
You might want to disable cancellation before a call to a cancel-unsafe library, restoring the old cancel state when the call returns from the library. See Cancel-Safe for explanations of cancel safety.
A thread's cancellation type is set with pthread_setcanceltype(3C), and determines whether the thread can be canceled anywhere in its execution or only at cancellation points.
With the default type of PTHREAD_CANCEL_DEFERRED, the thread is cancelable only at cancellation points, and then only when cancellation is enabled.
If the type is PTHREAD_CANCEL_ASYNCHRONOUS, the thread is cancelable at any point in its execution (assuming, of course, that cancellation is enabled). Try to limit regions of asynchronous cancellation to sequences with no external dependencies that could result in dangling resources or unresolved state conditions. Using asynchronous cancellation is discouraged because of the danger involved in trying to guarantee correct cleanup handling at absolutely every point in the program.
|
With the arrival of POSIX cancellation, the Cancel-Safe level has been added to the list of MT-Safety levels. See attributes(5). An application or library is Cancel-Safe whenever it has arranged for cleanup handlers to restore system or program state wherever cancellation can occur. The application or library is specifically Deferred-Cancel-Safe when it is Cancel-Safe for threads whose cancellation type is PTHREAD_CANCEL_DEFERRED. See Cancellation State. It is specifically Asynchronous-Cancel-Safe when it is Cancel-Safe for threads whose cancellation type is PTHREAD_CANCEL_ASYNCHRONOUS.
It is easier to arrange for deferred cancel safety, as this requires system and program state protection only around cancellation points. In general, expect that most applications and libraries are not Asynchronous-Cancel-Safe.
The cancellation functions described in this manual page are available for POSIX threads, only (the Solaris threads interfaces do not provide cancellation functions).
示例 1 Cancellation example
The following short C++ example shows the pushing/popping of cancellation handlers, the disabling/enabling of cancellation, the use of pthread_testcancel(), and so on. The free_res() cancellation handler in this example is a dummy function that simply prints a message, but that would free resources in a real application. The function f2() is called from the main thread, and goes deep into its call stack by calling itself recursively.
Before f2() starts running, the newly created thread has probably posted a cancellation on the main thread since the main thread calls thr_yield() right after creating thread2. Because cancellation was initially disabled in the main thread, through a call to pthread_setcancelstate(), the call to f2() from main() continues and constructs X at each recursive call, even though the main thread has a pending cancellation.
When f2() is called for the fifty-first time (when "i == 50"), f2() enables cancellation by calling pthread_setcancelstate(). It then establishes a cancellation point for itself by calling pthread_testcancel(). (Because a cancellation is pending, a call to a cancellation point such as read(2) or write(2) would also cancel the caller here.)
After the main() thread is canceled at the fifty-first iteration, all the cleanup handlers that were pushed are called in sequence; this is indicated by the calls to free_res() and the calls to the destructor for X. At each level, the C++ runtime calls the destructor for X and then the cancellation handler, free_res(). The print messages from free_res() and X's destructor show the sequence of calls.
At the end, the main thread is joined by thread2. Because the main thread was canceled, its return status from pthread_join() is PTHREAD_CANCELED. After the status is printed, thread2 returns, killing the process (since it is the last thread in the process).
#include <pthread.h> #include <sched.h> extern "C" void thr_yield(void); extern "C" void printf(...); struct X { int x; X(int i){x = i; printf("X(%d) constructed.\n", i);} ~X(){ printf("X(%d) destroyed.\n", x);} }; void free_res(void *i) { printf("Freeing `%d`\n",i); } char* f2(int i) { try { X dummy(i); pthread_cleanup_push(free_res, (void *)i); if (i == 50) { pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL); pthread_testcancel(); } f2(i+1); pthread_cleanup_pop(0); } catch (int) { printf("Error: In handler.\n"); } return "f2"; } void * thread2(void *tid) { void *sts; printf("I am new thread :%d\n", pthread_self()); pthread_cancel((pthread_t)tid); pthread_join((pthread_t)tid, &sts); printf("main thread cancelled due to %d\n", sts); return (sts); } main() { pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL); pthread_create(NULL, NULL, thread2, (void *)pthread_self()); thr_yield(); printf("Returned from %s\n",f2(0)); }
See attributes(5) for descriptions of the following attributes:
|
read(2), sigwait(2), write(2), Intro(3), condition(5), pthread_cleanup_pop(3C), pthread_cleanup_push(3C), pthread_exit(3C), pthread_join(3C), pthread_setcancelstate(3C), pthread_setcanceltype(3C), pthread_testcancel(3C), setjmp(3C), attributes(5), standards(5)