Updated sec2, log, and core VFDs to use pread/pwrite when

available (can be controlled via a configure/CMake option)
HDFGroup · Jan 31, 2019 · f833001 · f833001
1 parent 9149d5a
commit f833001
Show file tree

Hide file tree

Showing 8 changed files with 139 additions and 23 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -481,6 +481,15 @@ if (HDF5_MEMORY_ALLOC_SANITY_CHECK)
   set (H5_MEMORY_ALLOC_SANITY_CHECK 1)
 endif ()
 
+#-----------------------------------------------------------------------------
+# Option to enable/disable using pread/pwrite for VFDs
+#-----------------------------------------------------------------------------
+option (HDF5_ENABLE_PREADWRITE "Use pread/pwrite in sec2/log/core VFDs in place of read/write (when available)" ON)
+mark_as_advanced (HDF5_ENABLE_PREADWRITE)
+if (HDF5_ENABLE_PREADWRITE AND ${HDF_PREFIX}_HAVE_PREAD AND ${HDF_PREFIX}_HAVE_PWRITE)
+  set (H5_HAVE_PREADWRITE 1)
+endif ()
+
 #-----------------------------------------------------------------------------
 # Option to use deprecated public API symbols
 #-----------------------------------------------------------------------------

diff --git a/config/cmake/H5pubconf.h.in b/config/cmake/H5pubconf.h.in
@@ -267,6 +267,9 @@
 /* Define if we have parallel support */
 #cmakedefine H5_HAVE_PARALLEL @H5_HAVE_PARALLEL@
 
+/* Define if both pread and pwrite exist. */
+#cmakedefine H5_HAVE_PREADWRITE @H5_HAVE_PREADWRITE@
+
 /* Define to 1 if you have the <pthread.h> header file. */
 #cmakedefine H5_HAVE_PTHREAD_H @H5_HAVE_PTHREAD_H@
 

diff --git a/config/cmake_ext_mod/ConfigureChecks.cmake b/config/cmake_ext_mod/ConfigureChecks.cmake
@@ -274,10 +274,12 @@ if (NOT WINDOWS)
   # functionality so clock_gettime and CLOCK_MONOTONIC are defined
   # correctly. This was later updated to 200112L so that
   # posix_memalign() is visible for the direct VFD code on Linux
-  # systems.
+  # systems. Even later, this was changed to 200809L to support
+  # pread/pwrite in VFDs.
+  #
   # POSIX feature information can be found in the gcc manual at:
   # http://www.gnu.org/s/libc/manual/html_node/Feature-Test-Macros.html
-  set (HDF_EXTRA_C_FLAGS -D_POSIX_C_SOURCE=200112L)
+  set (HDF_EXTRA_C_FLAGS -D_POSIX_C_SOURCE=200809L)
 
   # Need to add this so that O_DIRECT is visible for the direct
   # VFD on Linux systems.
@@ -506,6 +508,8 @@ CHECK_FUNCTION_EXISTS (lround            ${HDF_PREFIX}_HAVE_LROUND)
 CHECK_FUNCTION_EXISTS (lroundf           ${HDF_PREFIX}_HAVE_LROUNDF)
 CHECK_FUNCTION_EXISTS (lstat             ${HDF_PREFIX}_HAVE_LSTAT)
 
+CHECK_FUNCTION_EXISTS (pread             ${HDF_PREFIX}_HAVE_PREAD)
+CHECK_FUNCTION_EXISTS (pwrite            ${HDF_PREFIX}_HAVE_PWRITE)
 CHECK_FUNCTION_EXISTS (rand_r            ${HDF_PREFIX}_HAVE_RAND_R)
 CHECK_FUNCTION_EXISTS (random            ${HDF_PREFIX}_HAVE_RANDOM)
 CHECK_FUNCTION_EXISTS (round             ${HDF_PREFIX}_HAVE_ROUND)

diff --git a/configure.ac b/configure.ac
@@ -1130,11 +1130,12 @@ case "$host_cpu-$host_vendor-$host_os" in
     ## functionality so clock_gettime and CLOCK_MONOTONIC are defined
     ## correctly. This was later updated to 200112L so that
     ## posix_memalign() is visible for the direct VFD code on Linux
-    ## systems.
+    ## systems. Even later, this was changed to 200809L to support
+    ## pread/pwrite in VFDs.
     ##
     ## POSIX feature information can be found in the gcc manual at:
     ## http://www.gnu.org/s/libc/manual/html_node/Feature-Test-Macros.html
-    H5_CPPFLAGS="-D_POSIX_C_SOURCE=200112L $H5_CPPFLAGS"
+    H5_CPPFLAGS="-D_POSIX_C_SOURCE=200809L $H5_CPPFLAGS"
 
     ## Need to add this so that O_DIRECT is visible for the direct
     ## VFD on Linux systems.
@@ -3265,6 +3266,47 @@ case "X-$STRICT_FORMAT_CHECKS" in
 esac
 
 
+## ----------------------------------------------------------------------
+## Enable use of pread/pwrite instead of read/write in certain VFDs.
+##
+AC_SUBST([PREADWRITE])
+
+## Check these first to avoid interspersed output in the AC_ARG_ENABLE line
+## below. (Probably overkill to check for both, but we'll be extra careful)
+PREADWRITE_HAVE_BOTH=yes
+AC_CHECK_FUNC([pread], [], [PREADWRITE_HAVE_BOTH=no])
+AC_CHECK_FUNC([pwrite], [], [PREADWRITE_HAVE_BOTH=no])
+
+AC_MSG_CHECKING([whether to use pread/pwrite instead of read/write in certain VFDs])
+AC_ARG_ENABLE([preadwrite],
+              [AS_HELP_STRING([--enable-preadwrite],
+                              [Enable using pread/pwrite instead of read/write in sec2/log/core VFDs.
+                              [default=yes if pread/pwrite are present]])],
+              [PREADWRITE=$enableval])
+
+## Set the default level.
+if test "X-$PREADWRITE" = X- ; then
+  PREADWRITE=yes
+fi
+
+case "X-$PREADWRITE" in
+  X-yes)
+      if test "X-$PREADWRITE_HAVE_BOTH" = "X-yes"; then
+        AC_DEFINE([HAVE_PREADWRITE], [1], [Define if both pread and pwrite exist.])
+        AC_MSG_RESULT([yes])
+      else
+        AC_MSG_RESULT([no])
+      fi
+    ;;
+  X-no)
+      AC_MSG_RESULT([no])
+    ;;
+  *)
+      AC_MSG_ERROR([Unrecognized value: $PREADWRITE])
+    ;;
+esac
+
+
 ## ----------------------------------------------------------------------
 ## Enable embedded library information
 ##

diff --git a/src/H5FDcore.c b/src/H5FDcore.c
@@ -348,14 +348,17 @@ H5FD__core_write_to_bstore(H5FD_core_t *file, haddr_t addr, size_t size)
 
     HDassert(file);
 
-    /* Write to backing store */
-    if((off_t)addr != HDlseek(file->fd, (off_t)addr, SEEK_SET))
+#ifndef H5_HAVE_PREADWRITE
+    /* Seek to the correct location (if we don't have pwrite) */
+    if((HDoff_t)addr != HDlseek(file->fd, (off_t)addr, SEEK_SET))
         HGOTO_ERROR(H5E_IO, H5E_SEEKERROR, FAIL, "error seeking in backing store")
+#endif /* H5_HAVE_PREADWRITE */
 
     while (size > 0) {
 
         h5_posix_io_t       bytes_in        = 0;    /* # of bytes to write  */
         h5_posix_io_ret_t   bytes_wrote     = -1;   /* # of bytes written   */
+        HDoff_t             offset          = (HDoff_t)addr;
 
         /* Trying to write more bytes than the return type can handle is
          * undefined behavior in POSIX.
@@ -366,15 +369,21 @@ H5FD__core_write_to_bstore(H5FD_core_t *file, haddr_t addr, size_t size)
             bytes_in = (h5_posix_io_t)size;
 
         do {
+#ifdef H5_HAVE_PREADWRITE
+            bytes_wrote = HDpwrite(file->fd, ptr, bytes_in, offset);
+            offset += bytes_wrote;
+#else
             bytes_wrote = HDwrite(file->fd, ptr, bytes_in);
+#endif /* H5_HAVE_PREADWRITE */
         } while(-1 == bytes_wrote && EINTR == errno);
 
         if(-1 == bytes_wrote) { /* error */
             int myerrno = errno;
             time_t mytime = HDtime(NULL);
-            HDoff_t myoffset = HDlseek(file->fd, (HDoff_t)0, SEEK_CUR);
 
-            HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "write to backing store failed: time = %s, filename = '%s', file descriptor = %d, errno = %d, error message = '%s', ptr = %p, total write size = %llu, bytes this sub-write = %llu, bytes actually written = %llu, offset = %llu", HDctime(&mytime), file->name, file->fd, myerrno, HDstrerror(myerrno), ptr, (unsigned long long)size, (unsigned long long)bytes_in, (unsigned long long)bytes_wrote, (unsigned long long)myoffset);
+            offset = HDlseek(file->fd, (HDoff_t)0, SEEK_CUR);
+
+            HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "write to backing store failed: time = %s, filename = '%s', file descriptor = %d, errno = %d, error message = '%s', ptr = %p, total write size = %llu, bytes this sub-write = %llu, bytes actually written = %llu, offset = %llu", HDctime(&mytime), file->name, file->fd, myerrno, HDstrerror(myerrno), ptr, (unsigned long long)size, (unsigned long long)bytes_in, (unsigned long long)bytes_wrote, (unsigned long long)offset);
         } /* end if */
 
         HDassert(bytes_wrote > 0);
@@ -852,6 +861,7 @@ H5FD__core_open(const char *name, unsigned flags, hid_t fapl_id, haddr_t maxaddr
                 while(size > 0) {
                     h5_posix_io_t       bytes_in        = 0;    /* # of bytes to read       */
                     h5_posix_io_ret_t   bytes_read      = -1;   /* # of bytes actually read */
+                    HDoff_t             offset          = (HDoff_t)0;
 
                     /* Trying to read more bytes than the return type can handle is
                      * undefined behavior in POSIX.
@@ -862,15 +872,21 @@ H5FD__core_open(const char *name, unsigned flags, hid_t fapl_id, haddr_t maxaddr
                         bytes_in = (h5_posix_io_t)size;
 
                     do {
+#ifdef H5_HAVE_PREADWRITE
+                        bytes_read = HDpread(file->fd, mem, bytes_in, offset);
+                        offset += bytes_read;
+#else
                         bytes_read = HDread(file->fd, mem, bytes_in);
+#endif /* H5_HAVE_PREADWRITE */
                     } while(-1 == bytes_read && EINTR == errno);
 
                     if(-1 == bytes_read) { /* error */
                         int myerrno = errno;
                         time_t mytime = HDtime(NULL);
-                        HDoff_t myoffset = HDlseek(file->fd, (HDoff_t)0, SEEK_CUR);
 
-                        HGOTO_ERROR(H5E_IO, H5E_READERROR, NULL, "file read failed: time = %s, filename = '%s', file descriptor = %d, errno = %d, error message = '%s', file->mem = %p, total read size = %llu, bytes this sub-read = %llu, bytes actually read = %llu, offset = %llu", HDctime(&mytime), file->name, file->fd, myerrno, HDstrerror(myerrno), file->mem, (unsigned long long)size, (unsigned long long)bytes_in, (unsigned long long)bytes_read, (unsigned long long)myoffset);
+                        offset = HDlseek(file->fd, (HDoff_t)0, SEEK_CUR);
+
+                        HGOTO_ERROR(H5E_IO, H5E_READERROR, NULL, "file read failed: time = %s, filename = '%s', file descriptor = %d, errno = %d, error message = '%s', file->mem = %p, total read size = %llu, bytes this sub-read = %llu, bytes actually read = %llu, offset = %llu", HDctime(&mytime), file->name, file->fd, myerrno, HDstrerror(myerrno), file->mem, (unsigned long long)size, (unsigned long long)bytes_in, (unsigned long long)bytes_read, (unsigned long long)offset);
                     } /* end if */
 
                     HDassert(bytes_read >= 0);

diff --git a/src/H5FDlog.c b/src/H5FDlog.c
@@ -1191,7 +1191,8 @@ H5FD_log_read(H5FD_t *_file, H5FD_mem_t type, hid_t H5_ATTR_UNUSED dxpl_id, hadd
         } /* end if */
     } /* end if */
 
-    /* Seek to the correct location */
+#ifndef H5_HAVE_PREADWRITE
+    /* Seek to the correct location (if we don't have pread) */
     if(addr != file->pos || OP_READ != file->op) {
 #ifdef H5_HAVE_GETTIMEOFDAY
         if(file->fa.flags & H5FD_LOG_TIME_SEEK)
@@ -1234,6 +1235,7 @@ H5FD_log_read(H5FD_t *_file, H5FD_mem_t type, hid_t H5_ATTR_UNUSED dxpl_id, hadd
 #endif /* H5_HAVE_GETTIMEOFDAY */
         } /* end if */
     } /* end if */
+#endif /* H5_HAVE_PREADWRITE */
 
     /*
      * Read data, being careful of interrupted system calls, partial results,
@@ -1247,6 +1249,7 @@ H5FD_log_read(H5FD_t *_file, H5FD_mem_t type, hid_t H5_ATTR_UNUSED dxpl_id, hadd
 
         h5_posix_io_t       bytes_in        = 0;    /* # of bytes to read       */
         h5_posix_io_ret_t   bytes_read      = -1;   /* # of bytes actually read */ 
+        HDoff_t             offset          = (HDoff_t)addr;
 
         /* Trying to read more bytes than the return type can handle is
          * undefined behavior in POSIX.
@@ -1257,18 +1260,24 @@ H5FD_log_read(H5FD_t *_file, H5FD_mem_t type, hid_t H5_ATTR_UNUSED dxpl_id, hadd
             bytes_in = (h5_posix_io_t)size;
 
         do {
+#ifdef H5_HAVE_PREADWRITE
+            bytes_read = HDpread(file->fd, buf, bytes_in, offset);
+            offset += bytes_read;
+#else
             bytes_read = HDread(file->fd, buf, bytes_in);
+#endif /* H5_HAVE_PREADWRITE */
         } while(-1 == bytes_read && EINTR == errno);
 
         if(-1 == bytes_read) { /* error */
             int myerrno = errno;
             time_t mytime = HDtime(NULL);
-            HDoff_t myoffset = HDlseek(file->fd, (HDoff_t)0, SEEK_CUR);
+
+            offset = HDlseek(file->fd, (HDoff_t)0, SEEK_CUR);
 
             if(file->fa.flags & H5FD_LOG_LOC_READ)
                 HDfprintf(file->logfp, "Error! Reading: %10a-%10a (%10Zu bytes)\n", orig_addr, (orig_addr + orig_size) - 1, orig_size);
 
-            HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "file read failed: time = %s, filename = '%s', file descriptor = %d, errno = %d, error message = '%s', buf = %p, total read size = %llu, bytes this sub-read = %llu, bytes actually read = %llu, offset = %llu", HDctime(&mytime), file->filename, file->fd, myerrno, HDstrerror(myerrno), buf, (unsigned long long)size, (unsigned long long)bytes_in, (unsigned long long)bytes_read, (unsigned long long)myoffset);
+            HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "file read failed: time = %s, filename = '%s', file descriptor = %d, errno = %d, error message = '%s', buf = %p, total read size = %llu, bytes this sub-read = %llu, bytes actually read = %llu, offset = %llu", HDctime(&mytime), file->filename, file->fd, myerrno, HDstrerror(myerrno), buf, (unsigned long long)size, (unsigned long long)bytes_in, (unsigned long long)bytes_read, (unsigned long long)offset);
         } /* end if */
 
         if(0 == bytes_read) {
@@ -1398,7 +1407,8 @@ H5FD_log_write(H5FD_t *_file, H5FD_mem_t type, hid_t H5_ATTR_UNUSED dxpl_id, had
             file->nwrite[tmp_addr++]++;
     } /* end if */
 
-    /* Seek to the correct location */
+#ifndef H5_HAVE_PREADWRITE
+    /* Seek to the correct location (if we don't have pwrite) */
     if(addr != file->pos || OP_WRITE != file->op) {
 #ifdef H5_HAVE_GETTIMEOFDAY
         if(file->fa.flags & H5FD_LOG_TIME_SEEK)
@@ -1441,6 +1451,7 @@ H5FD_log_write(H5FD_t *_file, H5FD_mem_t type, hid_t H5_ATTR_UNUSED dxpl_id, had
 #endif /* H5_HAVE_GETTIMEOFDAY */
         } /* end if */
     } /* end if */
+#endif /* H5_HAVE_PREADWRITE */
 
     /*
      * Write the data, being careful of interrupted system calls and partial
@@ -1454,6 +1465,7 @@ H5FD_log_write(H5FD_t *_file, H5FD_mem_t type, hid_t H5_ATTR_UNUSED dxpl_id, had
 
         h5_posix_io_t       bytes_in        = 0;    /* # of bytes to write  */
         h5_posix_io_ret_t   bytes_wrote     = -1;   /* # of bytes written   */ 
+        HDoff_t             offset          = (HDoff_t)addr;
 
         /* Trying to write more bytes than the return type can handle is
          * undefined behavior in POSIX.
@@ -1464,18 +1476,24 @@ H5FD_log_write(H5FD_t *_file, H5FD_mem_t type, hid_t H5_ATTR_UNUSED dxpl_id, had
             bytes_in = (h5_posix_io_t)size;
 
         do {
+#ifdef H5_HAVE_PREADWRITE
+            bytes_wrote = HDpwrite(file->fd, buf, bytes_in, offset);
+            offset += bytes_wrote;
+#else
             bytes_wrote = HDwrite(file->fd, buf, bytes_in);
+#endif /* H5_HAVE_PREADWRITE */
         } while(-1 == bytes_wrote && EINTR == errno);
 
         if(-1 == bytes_wrote) { /* error */
             int myerrno = errno;
             time_t mytime = HDtime(NULL);
-            HDoff_t myoffset = HDlseek(file->fd, (HDoff_t)0, SEEK_CUR);
+
+            offset = HDlseek(file->fd, (HDoff_t)0, SEEK_CUR);
 
             if(file->fa.flags & H5FD_LOG_LOC_WRITE)
                 HDfprintf(file->logfp, "Error! Writing: %10a-%10a (%10Zu bytes)\n", orig_addr, (orig_addr + orig_size) - 1, orig_size);
 
-            HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "file write failed: time = %s, filename = '%s', file descriptor = %d, errno = %d, error message = '%s', buf = %p, total write size = %llu, bytes this sub-write = %llu, bytes actually written = %llu, offset = %llu", HDctime(&mytime), file->filename, file->fd, myerrno, HDstrerror(myerrno), buf, (unsigned long long)size, (unsigned long long)bytes_in, (unsigned long long)bytes_wrote, (unsigned long long)myoffset);
+            HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "file write failed: time = %s, filename = '%s', file descriptor = %d, errno = %d, error message = '%s', buf = %p, total write size = %llu, bytes this sub-write = %llu, bytes actually written = %llu, offset = %llu", HDctime(&mytime), file->filename, file->fd, myerrno, HDstrerror(myerrno), buf, (unsigned long long)size, (unsigned long long)bytes_in, (unsigned long long)bytes_wrote, (unsigned long long)offset);
         } /* end if */
 
         HDassert(bytes_wrote > 0);