ASYNC_start_job.pod 10.7 KB
Newer Older
M
Matt Caswell 已提交
1 2 3 4
=pod

=head1 NAME

5 6 7 8
ASYNC_init, ASYNC_cleanup, ASYNC_init_thread, ASYNC_cleanup_thread,
ASYNC_start_job, ASYNC_pause_job, ASYNC_in_job, ASYNC_get_wait_fd,
ASYNC_get_current_job, ASYNC_wake, ASYNC_clear_wake, ASYNC_block_pause,
ASYNC_unblock_pause - asynchronous job management functions
M
Matt Caswell 已提交
9 10 11 12 13

=head1 SYNOPSIS

 #include <openssl/async.h>

14 15 16 17 18
 int ASYNC_init(int init_thread, size_t max_size, size_t init_size);
 void ASYNC_cleanup(int cleanupthread);

 int ASYNC_init_thread(size_t max_size, size_t init_size);
 void ASYNC_cleanup_thread(void);
M
Matt Caswell 已提交
19 20 21 22 23 24 25 26 27

 int ASYNC_start_job(ASYNC_JOB **job, int *ret, int (*func)(void *),
                     void *args, size_t size);
 int ASYNC_pause_job(void);

 int ASYNC_get_wait_fd(ASYNC_JOB *job);
 ASYNC_JOB *ASYNC_get_current_job(void);
 void ASYNC_wake(ASYNC_JOB *job);
 void ASYNC_clear_wake(ASYNC_JOB *job);
28 29
 void ASYNC_block_pause(void);
 void ASYNC_unblock_pause(void);
M
Matt Caswell 已提交
30 31 32 33 34 35 36 37 38 39 40 41 42

=head1 DESCRIPTION

OpenSSL implements asynchronous capabilities through an ASYNC_JOB. This
represents code that can be started and executes until some event occurs. At
that point the code can be paused and control returns to user code until some
subsequent event indicates that the job can be resumed.

The creation of an ASYNC_JOB is a relatively expensive operation. Therefore, for
efficiency reasons, jobs can be created up front and reused many times. They are
held in a pool until they are needed, at which point they are removed from the
pool, used, and then returned to the pool when the job completes. Before using
any of the asynchronous job functions, user code should first call
43 44 45 46 47 48 49 50 51 52 53
ASYNC_init(). If the user application is multi-threaded, then
ASYNC_init_thread() should be called for each thread that will initiate
asynchronous jobs. If the B<init_thread> parameter to ASYNC_init() is non-zero
then ASYNC_init_thread is automatically called for the current thread. Before
user code exits it should free up resources for each thread that was initialised
using ASYNC_cleanup_thread(). No asynchronous jobs must be outstanding for the thread
when ASYNC_cleanup_thread() is called. Failing to ensure this will result in memory
leaks. Additionally an application should call ASYNC_cleanup() when all
asynchronous work is complete across all threads. If B<cleanupthread> is
non-zero then ASYNC_cleanup_thread() is automatically called for the current
thread.
54 55 56 57 58 59

The B<max_size> argument limits the number of ASYNC_JOBs that will be held in
the pool. If B<max_size> is set to 0 then no upper limit is set. When an
ASYNC_JOB is needed but there are none available in the pool already then one
will be automatically created, as long as the total of ASYNC_JOBs managed by the
pool does not exceed B<max_size>. When the pool is first initialised
60 61 62
B<init_size> ASYNC_JOBs will be created immediately. If ASYNC_init_thread() is
not called before the pool is first used then it will be called automatically
with a B<max_size> of 0 (no upper limit) and an B<init_size> of 0 (no ASYNC_JOBs
63
created up front). If a pool is created in this way it must still be cleaned up
64
with an explicit call to ASYNC_cleanup_thread().
M
Matt Caswell 已提交
65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92

An asynchronous job is started by calling the ASYNC_start_job() function.
Initially B<*job> should be NULL. B<ret> should point to a location where the
return value of the asynchronous function should be stored on completion of the
job. B<func> represents the function that should be started asynchronously. The
data pointed to by B<args> and of size B<size> will be copied and then passed as
an argument to B<func> when the job starts. ASYNC_start_job will return one of
the following values:

=over 4

=item B<ASYNC_ERR>

An error occurred trying to start the job. Check the OpenSSL error queue (e.g.
see L<ERR_print_errors(3)>) for more details.

=item B<ASYNC_NO_JOBS>

There are no jobs currently available in the pool. This call can be retried
again at a later time.

=item B<ASYNC_PAUSE>

The job was successfully started but was "paused" before it completed (see
ASYNC_pause_job() below). A handle to the job is placed in B<*job>. Other work
can be performed (if desired) and the job restarted at a later time. To restart
a job call ASYNC_start_job() again passing the job handle in B<*job>. The
B<func>, B<args> and B<size> parameters will be ignored when restarting a job.
93 94
When restarting a job ASYNC_start_job() B<must> be called from the same thread
that the job was originally started from.
M
Matt Caswell 已提交
95 96 97 98 99 100 101 102

=item B<ASYNC_FINISH>

The job completed. B<*job> will be NULL and the return value from B<func> will
be placed in B<*ret>.

=back

103 104 105 106
At any one time there can be a maximum of one job actively running per thread
(you can have many that are paused). ASYNC_get_current_job() can be used to get
a pointer to the currently executing ASYNC_JOB. If no job is currently executing
then this will return NULL.
M
Matt Caswell 已提交
107 108 109 110 111 112 113 114 115 116 117 118 119 120

If executing within the context of a job (i.e. having been called directly or
indirectly by the function "func" passed as an argument to ASYNC_start_job())
then ASYNC_pause_job() will immediately return control to the calling
application with ASYNC_PAUSE returned from the ASYNC_start_job() call. A
subsequent call to ASYNC_start_job passing in the relevant ASYNC_JOB in the
B<*job> parameter will resume execution from the ASYNC_pause_job() call. If
ASYNC_pause_job() is called whilst not within the context of a job then no
action is taken and ASYNC_pause_job() returns immediately.

Every ASYNC_JOB has a "wait" file descriptor associated with it. Calling
ASYNC_get_wait_fd() and passing in a pointer to an ASYNC_JOB in the B<job>
parameter will return the wait file descriptor associated with that job. This
file descriptor can be used to signal that the job should be resumed.
121 122 123 124
Applications can wait for the file descriptor to be ready for "read" using a
system function call such as select or poll (being ready for "read" indicates
that the job should be resumed). Applications can signal that a job is ready to
resume using ASYNC_wake() or clear an existing signal using ASYNC_clear_wake().
M
Matt Caswell 已提交
125 126 127 128 129 130 131

An example of typical usage might be an async capable engine. User code would
initiate cryptographic operations. The engine would initiate those operations
asynchronously and then call ASYNC_pause_job() to return control to the user
code. The user code can then perform other tasks or wait for the job to be ready
by calling "select" or other similar function on the wait file descriptor. The
engine can signal to the user code that the job should be resumed using
132
ASYNC_wake(). Once resumed the engine would clear the wake signal by calling
M
Matt Caswell 已提交
133 134
ASYNC_clear_wake().

135 136 137 138 139 140 141 142 143 144 145 146 147 148
The ASYNC_block_pause() function will prevent the currently active job from
pausing. The block will remain in place until a subsequent call to
ASYNC_unblock_pause(). These functions can be nested, e.g. if you call
ASYNC_block_pause() twice then you must call ASYNC_unblock_pause() twice in
order to reenable pausing. If these functions are called while there is no
currently active job then they have no effect. This functionality can be useful
to avoid deadlock scenarios. For example during the execution of an ASYNC_JOB an
application aquires a lock. It then calls some cryptographic function which
invokes ASYNC_pause_job(). This returns control back to the code that created
the ASYNC_JOB. If that code then attempts to aquire the same lock before
resuming the original job then a deadlock can occur. By calling
ASYNC_block_pause() immediately after aquiring the lock and
ASYNC_unblock_pause() immediately before releasing it then this situation cannot
occur.
M
Matt Caswell 已提交
149 150 151

=head1 RETURN VALUES

152
ASYNC_init and ASYNC_init_thread return 1 on success or 0 otherwise.
M
Matt Caswell 已提交
153 154 155 156

ASYNC_start_job returns one of ASYNC_ERR, ASYNC_NO_JOBS, ASYNC_PAUSE or
ASYNC_FINISH as described above.

157 158 159
ASYNC_pause_job returns 0 if an error occured or 1 on success. If called when
not within the context of an ASYNC_JOB then this is counted as success so 1 is
returned.
M
Matt Caswell 已提交
160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219

ASYNC_get_wait_fd returns the "wait" file descriptor associated with the
ASYNC_JOB provided as an argument.

ASYNC_get_current_job returns a pointer to the currently executing ASYNC_JOB or
NULL if not within the context of a job.

=head1 EXAMPLE

The following example demonstrates how to use most of the core async APIs:

 #include <stdio.h>
 #include <openssl/async.h>

 int jobfunc(void *arg)
 {
     ASYNC_JOB *currjob;
     unsigned char *msg;

     currjob = ASYNC_get_current_job();
     if (currjob != NULL) {
         printf("Executing within a job\n");
     } else {
         printf("Not executing within a job - should not happen\n");
         return 0;
     }

     msg = (unsigned char *)arg;
     printf("Passed in message is: %s\n", msg);

     /*
      * Normally some external event would cause this to happen at some
      * later point - but we do it here for demo purposes, i.e.
      * immediately signalling that the job is ready to be woken up after
      * we return to main via ASYNC_pause_job().
      */
     ASYNC_wake(currjob);

     /* Return control back to main */
     ASYNC_pause_job();

     /* Clear the wake signal */
     ASYNC_clear_wake(currjob);

     printf ("Resumed the job after a pause\n");

     return 1;
 }

 int main(void)
 {
     ASYNC_JOB *job = NULL;
     int ret, waitfd;
     fd_set waitfdset;
     unsigned char msg[13] = "Hello world!";

     /*
      * We're only expecting 1 job to be used here so we're only creating
      * a pool of 1
      */
220
     if (!ASYNC_init(1, 1, 1)) {
M
Matt Caswell 已提交
221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250
         printf("Error creating pool\n");
         goto end;
     }

     printf("Starting...\n");

     for (;;) {
         switch(ASYNC_start_job(&job, &ret, jobfunc, msg, sizeof(msg))) {
         case ASYNC_ERR:
         case ASYNC_NO_JOBS:
                 printf("An error occurred\n");
                 goto end;
         case ASYNC_PAUSE:
                 printf("Job was paused\n");
                 break;
         case ASYNC_FINISH:
                 printf("Job finished with return value %d\n", ret);
                 goto end;
         }

         /* Wait for the job to be woken */
         printf("Waiting for the job to be woken up\n");
         waitfd = ASYNC_get_wait_fd(job);
         FD_ZERO(&waitfdset);
         FD_SET(waitfd, &waitfdset);
         select(waitfd + 1, &waitfdset, NULL, NULL, NULL);
     }

 end:
     printf("Finishing\n");
251
     ASYNC_cleanup(1);
M
Matt Caswell 已提交
252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272

     return 0;
 }

The expected output from executing the above example program is:

 Starting...
 Executing within a job
 Passed in message is: Hello world!
 Job was paused
 Waiting for the job to be woken up
 Resumed the job after a pause
 Job finished with return value 1
 Finishing

=head1 SEE ALSO

L<crypto(3)>, L<ERR_print_errors(3)>

=head1 HISTORY

273 274 275
ASYNC_init, ASYNC_init_thread, ASYNC_cleanup, ASYNC_cleanup_thread,
ASYNC_start_job, ASYNC_pause_job, ASYNC_get_wait_fd, ASYNC_get_current_job,
ASYNC_wake, ASYNC_clear_wake were first added to OpenSSL 1.1.0.
M
Matt Caswell 已提交
276 277

=cut