Clarify some of the I/O issues.

Add case of using select() and blocking I/O with
BIOs and why you shouldn't (thanks Bodo!).

2 files changed, 28 insertions, 35 deletions

diff --git a/doc/crypto/BIO_read.pod b/doc/crypto/BIO_read.pod
index 6c001a3092..e7eb5ea045 100644
--- a/doc/crypto/BIO_read.pod
+++ b/doc/crypto/BIO_read.pod
@@ -38,16 +38,28 @@ the operation is not implemented in the specific BIO type.
 =head1 NOTES
 
 A 0 or -1 return is not necessarily an indication of an error. In
-particular when the source/sink is non-blocking or of a certain type (for
-example an SSL BIO can retry even if the underlying connection is blocking)
+particular when the source/sink is non-blocking or of a certain type
 it may merely be an indication that no data is currently available and that
-the application should retry the operation later. L<BIO_should_retry(3)|BIO_should_retry(3)>
-can be called to determine the precise cause.
+the application should retry the operation later.
+
+One technique sometimes used with blocking sockets is to use a system call
+(such as select(), poll() or eqivalent) to determine when data is available
+and then call read() to read the data. The eqivalent with BIOs (that is call
+select() on the underlying I/O structure and then call BIO_read() to
+read the data) should B<not> be used because a single call to BIO_read()
+can cause several reads (and writes in the case of SSL BIOs) on the underlying
+I/O structure and may block as a result. Instead select() (or equivalent)
+should be combined with non blocking I/O so successive reads will request
+a retry instead of blocking.
+
+See the L<BIO_should_retry(3)|BIO_should_retry(3)> for details of how to
+determine the cause of a retry and other I/O issues.
 
 If the BIO_gets() function is not supported by a BIO then it possible to
 work around this by adding a buffering BIO L<BIO_f_buffer(3)|BIO_f_buffer(3)>
 to the chain.
 
 =head1 SEE ALSO
+L<BIO_should_retry(3)|BIO_should_retry(3)>
 
 TBA
diff --git a/doc/crypto/BIO_should_retry.pod b/doc/crypto/BIO_should_retry.pod
index ab67a46114..6d291b1888 100644
--- a/doc/crypto/BIO_should_retry.pod
+++ b/doc/crypto/BIO_should_retry.pod
@@ -46,7 +46,7 @@ reason other than reading or writing is the cause of the condition.
 BIO_get_retry_reason() returns a mask of the cause of a retry condition
 consisting of the values B<BIO_FLAGS_READ>, B<BIO_FLAGS_WRITE>,
 B<BIO_FLAGS_IO_SPECIAL> though current BIO types will only set one of
-these (Q: is this correct?).
+these.
 
 BIO_get_retry_BIO() determines the precise reason for the special
 condition, it returns the BIO that caused this condition and if 
@@ -55,7 +55,7 @@ the reason code and the action that should be taken depends on
 the type of BIO that resulted in this condition.
 
 BIO_get_retry_reason() returns the reason for a special condition if
-pass the relevant BIO, for example as returned by BIO_get_retry_BIO().
+passed the relevant BIO, for example as returned by BIO_get_retry_BIO().
 
 =head1 NOTES
 
@@ -68,27 +68,17 @@ has reached EOF. Some BIO types may place additional information on
 the error queue. For more details see the individual BIO type manual
 pages.
 
-If the underlying I/O structure is in a blocking mode then most BIO
-types will not signal a retry condition, because the underlying I/O
+If the underlying I/O structure is in a blocking mode almost all current
+BIO types will not request a retry, because the underlying I/O
 calls will not. If the application knows that the BIO type will never
 signal a retry then it need not call BIO_should_retry() after a failed
 BIO I/O call. This is typically done with file BIOs.
 
-The presence of an SSL BIO is an exception to this rule: it can
-request a retry because the handshake process is underway (either
-initially or due to a session renegotiation) even if the underlying
-I/O structure (for example a socket) is in a blocking mode.
-
-The action an application should take after a BIO has signalled that a
-retry is required depends on the BIO that caused the retry.
-
-If the underlying I/O structure is in a blocking mode then the BIO
-call can be retried immediately. That is something like this can be
-done:
-
- do {
-    len = BIO_read(bio, buf, len);
- } while((len <= 0) && BIO_should_retry(bio));
+SSL BIOs are the only current exception to this rule: they can request a
+retry even if the underlying I/O structure is blocking, if a handshake
+occurs during a call to BIO_read(). An application can retry the failed
+call immediately or avoid this situation by setting SSL_MODE_AUTO_RETRY
+on the underlying SSL structure.
 
 While an application may retry a failed non blocking call immediately
 this is likely to be very inefficient because the call will fail
@@ -100,18 +90,9 @@ For example if the cause is ultimately a socket and BIO_should_read()
 is true then a call to select() may be made to wait until data is
 available and then retry the BIO operation. By combining the retry
 conditions of several non blocking BIOs in a single select() call
-it is possible to service several BIOs in a single thread. 
-
-The cause of the retry condition may not be the same as the call that
-made it: for example if BIO_write() fails BIO_should_read() can be
-true. One possible reason for this is that an SSL handshake is taking
-place.
-
-Even if data is read from the underlying I/O structure this does not
-imply that the next BIO I/O call will succeed. For example if an
-encryption BIO reads only a fraction of a block it will not be
-able to pass any data to the application until a complete block has
-been read.
+it is possible to service several BIOs in a single thread, though
+the performance may be poor if SSL BIOs are present because long delays
+can occur during the initial handshake process. 
 
 It is possible for a BIO to block indefinitely if the underlying I/O
 structure cannot process or return any data. This depends on the behaviour of