Bug#25807: LOAD DATA INFILE does not work with Pipes

Showstopper and regression against 5.0.24.

Previously, we ignored seek() errors (see Bug#22828) and let seek()s
against pipes fail.  Now, since we check that a seek didn't fail,
and return without reading, this bug popped up.

This restores the behavior for file-ish objects that could never be 
seek()ed.


mysys/mf_iocache.c:
  If we detect early that the file is not tell()able, then we should
  assume that it's also not seek()able and therefore we should never
  set the (poorly named) "seek_not_done" flag so that we don't immedi-
  ately try to seek() when reading later.
  
  The problem was that tell() was returning -1, so when we read later, 
  we needlessly tried to seek to position  (unsigned long) -1 . 
  
  Also, if we think we're supposed to seek to a position in a file and 
  the file is un-tell()able, then abort.

BitKeeper/etc/collapsed

@@ -39,3 +39,5 @@
 45ba4faf2oqu6eR8fqecR3LfSNcYUg
 45ba5238-NKl80QVXzdGo8hO9M75Xg
 45c0fdfb2mz6NdOIsLenJtf6_ZelTA
+45d1ffcd-r3v8A7uh92hQaMfQM9UPQ
+45d21437Vg_-i4uOWyvzYWHESXDP6A

mysys/mf_iocache.c

@@ -157,6 +157,7 @@ int init_io_cache(IO_CACHE *info, File file, uint cachesize,
 		  pbool use_async_io, myf cache_myflags)
 {
   uint min_cache;
+  my_off_t pos;
   my_off_t end_of_file= ~(my_off_t) 0;
   DBUG_ENTER("init_io_cache");
   DBUG_PRINT("enter",("cache: 0x%lx  type: %d  pos: %ld",
@@ -169,7 +170,25 @@ int init_io_cache(IO_CACHE *info, File file, uint cachesize,
   info->arg = 0;
   info->alloced_buffer = 0;
   info->buffer=0;
-  info->seek_not_done= test(file >= 0 && seek_offset != my_tell(file, MYF(0)));
+
+  pos= my_tell(file, MYF(0));
+  if ((pos == (my_off_t) -1) && (my_errno == ESPIPE))
+  {
+    /* 
+      This kind of object doesn't support seek() or tell().  Don't set a flag
+      that will make us again try to seek() later and fail.
+    */
+    info->seek_not_done= 0;
+    /*
+      Additionally, if we're supposed to start somewhere other than the
+      the beginning of whatever this file is, then somebody made a bad
+      assumption.
+    */
+    DBUG_ASSERT(seek_offset == 0);
+  }
+  else
+    info->seek_not_done= test(file >= 0 && seek_offset != pos);
+
   info->disk_writes= 0;
 #ifdef THREAD
   info->share=0;
@@ -454,13 +473,23 @@ int _my_b_read(register IO_CACHE *info, byte *Buffer, uint Count)
   */
   if (info->seek_not_done)
   {
-    if (my_seek(info->file,pos_in_file,MY_SEEK_SET,MYF(0)) 
-        == MY_FILEPOS_ERROR)
+    if ((my_seek(info->file,pos_in_file,MY_SEEK_SET,MYF(0)) 
+        != MY_FILEPOS_ERROR))
     {
-        info->error= -1;
-        DBUG_RETURN(1);
+      /* No error, reset seek_not_done flag. */
+      info->seek_not_done= 0;
+    }
+    else
+    {
+      /*
+        If the seek failed and the error number is ESPIPE, it is because
+        info->file is a pipe or socket or FIFO.  We never should have tried
+        to seek on that.  See Bugs#25807 and #22828 for more info.
+      */
+      DBUG_ASSERT(my_errno != ESPIPE);
+      info->error= -1;
+      DBUG_RETURN(1);
     }
-    info->seek_not_done=0;
   }
 
   diff_length=(uint) (pos_in_file & (IO_SIZE-1));