Subfiling VFD (#1883)

* Added support for vector I/O calls to the VFD layer, and
associated test code.  Note that this includes the optimization
to allow shortened sizes and types arrays to allow more space
efficient representations of vectors in which all entries are
of the same size and/or type.  See the Selection I/o RFC for
further details.

Tested serial and parallel, debug and production on Charis.
       serial and parallel debug only on Jelly.

* ran code formatter

quick serial build and test on jelly

* Add H5FD_read_selection() and H5FD_write_selection().  Currently only
translate to scalar calls.  Fix const buf in H5FD_write_vector().

* Format source

* Fix comments

* Add selection I/O to chunk code, used when: not using chunk cache, no
datatype conversion, no I/O filters, no page buffer, not using collective
I/O.  Requires global variable H5_use_selection_io_g be set to TRUE.
Implemented selection to vector I/O transaltion at the file driver
layer.

* Fix formatting unrelated to previous change to stop github from
complaining.

* Add full API support for selection I/O.  Add tests for this.

* Implement selection I/O for contiguous datasets.  Fix bug in selection
I/O translation.  Add const qualifiers to some internal selection I/O
routines to maintain const-correctness while avoiding memcpys.

* Added vector read / write support to the MPIO VFD, with associated
test code (see testpar/t_vfd.c).

Note that this implementation does NOT support vector entries of
size greater than 2 GB.  This must be repaired before release,
but it should be good enough for correctness testing.

As MPIO requires vector I/O requests to be sorted in increasing
address order, also added a vector sort utility in H5FDint.c  This
function is tested in passing by the MPIO vector I/O extension.

In passing, repaired a bug in size / type vector extension management
in H5FD_read/write_vector()

Tested parallel debug and production on charis and Jelly.

* Ran source code formatter

* Add support for independent parallel I/O with selection I/O.  Add
HDF5_USE_SELECTION_IO env var to control selection I/O (default off).

* Implement parallel collective support for selection I/O.

* Fix comments and run formatter.

* Update selection IO branch with develop (#1215)

Merged branch 'develop' into selection_io

* Sync with develop (#1262)

Updated the branch with develop changes.

* Implement big I/O support for vector I/O requests in the MPIO file
driver.

* Free arrays in H5FD__mpio_read/write_vector() as soon as they're not
needed, to cut down on memory usage during I/O.

* Address comments from code review.  Fix const warnings with
H5S_SEL_ITER_INIT().

* Committing clang-format changes

* Feature/subfiling (#1464)

* Initial checkin of merged sub-filing VFD.

Passes regression tests (debug/shared/paralle) on Jelly.
However, bugs and many compiler warnings remain -- not suitable
for merge to develop.

* Minor mods to src/H5FDsubfile_mpi.c to address errors reported by autogen.sh

* Code formatting run -- no test

* Merged my subfiling code fixes into the new selection_io_branch

* Forgot to add the FindMERCURY.cmake file. This will probably disappear soon

* attempting to make a more reliable subfile file open which doesn't return errors. For some unknown reason, the regular posix open will occasionally fail to create a subfile.  Some better error handling for file close has been added.

* added NULL option for H5FD_subfiling_config_t in H5Pset_fapl_subfiling (#1034)

* NULL option automatically stacks IOC VFD for subfiling and returns a valid fapl.
* added doxygen subfiling APIs

* Various fixes which allow the IOR benchmark to run correctly

* Lots of updates including the packaging up of the mercury_util source files to enable easier builds for our Benchmarking

* Interim checkin of selection_io_with_subfiling_vfd branch

    Moddified testpar/t_vfd.c to test the subfiling vfd with default configuration.
    Must update this code to run with a variety of configurations -- most particularly
    multiple IO concentrators, and stripe depth small enough to test the other IO
    concentrators.

    testpar/t_vfd.c exposed a large number of race condidtions -- symtoms included:

      1) Crashes (usually seg faults)

      2) Heap corruption

      3) Stack corruption

      4) Double frees of heap space

      5) Hangs

      6) Out of order execution of I/O requests / violations of POSIX semantics

      7) Swapped write requests

        Items 1 - 4 turned out to be primarily caused by file close issues --
    specifically, the main I/O concentrator thread and its pool of worker threads
    were not being shut down properly on file close.  Addressing this issue in
    combination with some other minor fixes seems to have addressed these issues.

        Items 5 & 6 appear to have been caused by issue of I/O requests to the
    thread pool in an order that did not maintain POSIX semantics.  A rewrite of
    the I/O request dispatch code appears to have solved these issues.

        Item 7 seems to have been caused by multiple write requests from a given
    rank being read by the wrong worker thread.  Code to issue "unique" tags for
    each write request via the ACK message appears to have cleaned this up.

        Note that the code is still in poor condtition.  A partial list of known
    defects includes:

     a) Race condiditon on file close that allows superblock writes to arrive
        at the I/O concentrator after it has been shutdown.  This defect is
        most evident when testpar/t_subfiling_vfd is run with 8 ranks.

     b) No error reporting from I/O concentrators -- must design and implement
        this.  For now, mostly just asserts, which suggests that it should be
        run in debug mode.

     c) Much commented out and/or un-used code.

     d) Code orgnaization

     e) Build system with bits of Mercury is awkward -- think of shifting
        to pthreads with our own thread pool code.

     f) Need to add native support for vector and selection I/O to the subfiling
        VFD.

     g) Need to review, and posibly rework configuration code.

     h) Need to store subfile configuration data in a superblock extension message,
        and add code to use this data on file open.

     i) Test code is inadequate -- expect more issues as it is extended.

        In particular, there is no unit test code for the I/O request dispatch code.
        While I think it is correct at present, we need test code to verify this.

        Similarly, we need to test with multiple I/O concentrators and much smaller
        stripe depth.

    My actual code changes were limited to:

          src/H5FDioc.c
          src/H5FDioc_threads.c
          src/H5FDsubfile_int.c
          src/H5FDsubfile_mpi.c
          src/H5FDsubfiling.c
          src/H5FDsubfiling.h
          src/H5FDsubfiling_priv.h
          testpar/t_subfiling_vfd.c
          testpar/t_vfd.c

    I'm not sure what is going on with the deletions in src/mercury/src/util.

    Tested parallel/debug on Charis and Jelly

* subfiling with selection IO (#1219)

Merged branch 'selection_io' into subfiling branch.

* Subfile name fixes (#1250)

* fixed subfiling naming convention, and added leading zero to rank names.

* Merge branch 'selection_io' into selection_io_with_subfiling_vfd (#1265)

* Added script to join subfiles into a single HDF5 file (#1350)

* Modified  H5FD__subfiling_query() to report that the sub-filing VFD supports MPI
This exposed issues with truncate and get EOF in the sub-filing VFD.
I believe I have addressed these issues (get EOF not as fully tested as it should be), howeer,
it exposed race conditions resulting in hangs.  As of this writing, I have not been able
to chase these down.

Note that the tests that expose these race conditions are in testpar/t_subfiling_vfd.c, and
are currently skipped.  Unskip these tests to reproduce the race conditions.

tested (to the extent possible) debug/parallel on charis and jelly.

* Committing clang-format changes

* fixed H5MM_free

Co-authored-by: mainzer <mainzer#hdfgroup.org>
Co-authored-by: jrmainzer <72230804+jrmainzer@users.noreply.github.com>
Co-authored-by: Richard Warren <Richard.Warren@hdfgroup.org>
Co-authored-by: Richard.Warren <richard.warren@jelly.ad.hdfgroup.org>
Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com>

* Move Subfiling VFD components into H5FDsubfiling source directory

* Update Autotools build and add H5_HAVE_SUBFILING_VFD macro to H5pubconf.h

* Tidy up CMake build of subfiling sources

* Merge branch 'develop' into feature/subfiling (#1539)

Merge branch 'develop' into feature/subfiling

* Add VFD interface version field to Subfiling and IOC VFDs

* Merge branch 'develop' into feature/subfiling (#1557)

Merge branch 'develop' into feature/subfiling

* Merge branch 'develop' into feature/subfiling (#1563)

Merge branch 'develop' into feature/subfiling

* Tidy up merge artifacts after rebase on develop

* Fix incorrect variable in mirror VFD utils CMake

* Ensure VFD values are always defined

* Add subfiling to CMake VFD_LIST if built

* Mark MPI I/O driver self-initialization global as static

* Add Subfiling VFD to predefined VFDs for HDF5_DRIVER env. variable

* Initial progress towards separating private vs. public subfiling code

* include libgen.h in t_vfd tests for correct dirname/basename

* Committing clang-format changes

* removed mercury option, included subfiling header path (#1577)

Added subfiling status to configure output, installed h5fuse.sh to build directory for use in future tests.

* added check for stdatomic.h (#1578)

* added check for stdatomic.h with subfiling
* added H5_HAVE_SUBFILING_VFD for cmake

* fix old-style-definition warning (#1582)

* fix old-style-definition warning

* added test for enable parallel with subfiling VFD (#1586)

Fails if subfiling VFD is not used with parallel support.

* Subfiling/IOC VFD fixes and tidying (#1619)

* Rename CMake option for Subfiling VFD to be consistent with other VFDs

* Miscellaneous Subfiling fixes

Add error message for unset MPI communicator

Support dynamic loading of subfiling VFD with default configuration

* Temporary fix for subfile name issue

* Added subfile checks (#1634)

* added subfile checks

* Feature/subfiling (#1655)

* Subfiling/IOC VFD cleanup

Fix misuse of MPI_COMM_WORLD in IOC VFD

Propagate Subfiling FAPL MPI settings down to IOC FAPL in default
configuration case

Cleanup IOC VFD debugging code

Change sprintf to snprintf in a few places

* Major work on separating Subfiling and IOC VFDs from each other

* Re-write async_completion func to not overuse stack

* Replace usage of MPI_COMM_WORLD with file's actual MPI communicator

* Refactor H5FDsubfile_mpi.c

* Remove empty file H5FDsubfile_mpi.c

* Separate IOC VFD errors to its own error stack

* Committing clang-format changes

* Remove H5TRACE macros from H5FDioc.c

* Integrate H5FDioc_threads.c with IOC error stack

* Fix for subfile name generation

Use number of I/O concentrators from existing subfiling configuration file, if one exists

* Add temporary barrier in "Get EOF" operation to prevent races on EOF

Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com>

* Fix for retrieval of machine Host ID

* Default to MPI_COMM_WORLD if no MPI params set

* added libs rt and pthreads (#1673)

* added libs rt and pthreads

* Feature/subfiling (#1689)

* More tidying of IOC VFD and subfiling debug code

* Remove old unused log file code

* Clear FID from active file map on failure

* Fix bug in generation of subfile names when truncating file

* Change subfile names to start from 1 instead of 0

* Use long long for user-specified stripe size from environment variable

* Skip 0-sized I/Os in low-level IOC I/O routines

* Don't update EOF on read

* Convert printed warning about data size mismatch to assertion

* Don't add base file address to I/O addresses twice

Base address should already be applied as part of H5FDwrite/read_vector calls

* Account for 0-sized I/O vector entries in subfile write/read functions

* Rewrite init_indep_io for clarity

* Correction for IOC wraparound calculations

* Some corrections to iovec calculations

* Remove temporary barrier on EOF retrieval

* Complete work request queue entry on error instead of skipping over

* Account for stripe size wraparound for sf_col_offset calculation

* Committing clang-format changes

Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com>

* Re-write and fix bugs in I/O vector filling routines (#1703)

* Rewrite I/O vector filling routines for clarity

* Fix bug with iovec_fill_last when last I/O size is 0

* added subfiling_dir line read (#1714)

* added subfiling_dir line read and use it
* shellcheck fixes

* I/O request dispatch logic update (#1731)

Short-circuit I/O request dispatch when head of I/O queue is an
in-progress get EOF or truncate operation. This prevents an issue where
a write operation can be dispatched alongside a get EOF/truncate
operation, whereas all I/O requests are supposed to be ineligible for
dispatch until the get EOF/truncate is completed

* h5fuse.sh.in clean-up (#1757)

* Added command-line options

* Committing clang-format changes

* Align with changes from develop

* Mimic MPI I/O VFD for EOF handling

* Initialize context_id field for work request objects

* Use logfile for some debugging information

* Use atomic store to set IOC ready flag

* Use separate communicator for sending file EOF data

Minor IOC cleanup

* Use H5_subfile_fid_to_context to get context ID for file in Subfiling
VFD

* IOVEC calculation fixes

* Updates for debugging code

* Minor fixes for threaded code

* Committing clang-format changes

* Use separate MPI communicator for barrier operations

* Committing clang-format changes

* Rewrite EOF routine to use nonblocking MPI communication

* Committing clang-format changes

* Always dispatch I/O work requests in IOC main loop

* Return distinct MPI communicator to library when requested

* Minor warning cleanup

* Committing clang-format changes

* Generate h5fuse.sh from h5fuse.sh.in in CMake

* Send truncate messages to correct IOC rank

* Committing clang-format changes

* Miscellaneous cleanup

Post some MPI receives before sends

Free some duplicated MPI communicator/Info objects

Remove unnecessary extra MPI_Barrier

* Warning cleanup

* Fix for leaked MPI communicator

* Retrieve file EOF on single rank and bcast it

* Fixes for a few failure paths

* Cleanup of IOC file opens

* Committing clang-format changes

* Use plan MPI_Send for send of EOF messages

* Always check MPI thread support level during Subfiling init

* Committing clang-format changes

* Handle a hang on failure when IOCs can't open subfiles

* Committing clang-format changes

* Refactor file open status consensus check

* Committing clang-format changes

* Fix for MPI_Comm_free being called after MPI_Finalize

* Fix VFD test by setting MPI params before setting subfiling on FAPL

* Update Subfiling VFD error handling and error stack usage

* Improvements for Subfiling logfiles

* Remove prototypes for currently unused routines

* Disable I/O queue stat collecting by default

* Remove unused serialization mutex variable

* Update VFD testing to take subfiling VFD into account

* Fix usage of global subfiling application layout object

* Minor fixes for failure pathways

* Keep track of the number of failures in an IOC I/O queue

* Make sure not to exceed MPI_TAG_UB value for data communication messages

* Committing clang-format changes

* Update for rename of some H5FD 'ctl' opcodes

* Always include Subfiling's public header files in hdf5.h

* Remove old unused code and comments

* Implement support for per-file I/O queues

Allows the subfiling VFD to have multiple HDF5 files open simultaneously

* Use simple MPI_Iprobe over unnecessary MPI_Improbe

* Committing clang-format changes

* Update HDF5 testing to query driver for H5FD_FEAT_DEFAULT_VFD_COMPATIBLE
flag

* Fix a few bugs related to file multi-opens

* Avoid calling MPI routines if subfiling gets reinitialized

* Fix issue when files are closed in a random order

* Update HDF5 testing to query VFD for "using MPI" feature flag

* Register atexit handler in subfiling VFD to call MPI_Finalize after HDF5
closes

* Fail for collective I/O requests until support is implemented

* Correct VOL test function prototypes

* Minor cleanup of old code and comments

* Update mercury dependency

* Cleanup of subfiling configuration structure

* Committing clang-format changes

* Build system updates for Subfiling VFD

* Fix possible hang on failure in t_vfd tests caused by mismatched
MPI_Barrier calls

* Copy subfiling IOC fapl in "fapl get" method

* Mirror subfiling superblock writes to stub file for legacy POSIX-y HDF5
applications

* Allow collective I/O for MPI_BYTE types and rank 0 bcast strategy

* Committing clang-format changes

* Use different scheme for subfiling write message MPI tag calculations

* Committing clang-format changes

* Avoid performing fstat calls on all MPI ranks

* Add MPI_Barrier before finalizing IOC threads

* Use try_lock in I/O queue dispatch to minimize contention from worker threads

* Use simple Waitall for nonblocking I/O waits

* Add configurable IOC main thread delay and try_lock option to I/O queue dispatch

* Fix bug that could cause serialization of non-overlapping I/O requests

* Temporarily treat collective subfiling vector I/O calls as independent

* Removed unused mercury bits

* Add stubs for subfiling and IOC file delete callback

* Update VFD testing for Subfiling VFD

* Work around HDF5 metadata cache bug for Subfiling VFD when MPI Comm size
= 1

* Committing clang-format changes

Co-authored-by: mainzer <mainzer#hdfgroup.org>
Co-authored-by: Neil Fortner <nfortne2@hdfgroup.org>
Co-authored-by: Scot Breitenfeld <brtnfld@hdfgroup.org>
Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: jrmainzer <72230804+jrmainzer@users.noreply.github.com>
Co-authored-by: Richard Warren <Richard.Warren@hdfgroup.org>
Co-authored-by: Richard.Warren <richard.warren@jelly.ad.hdfgroup.org>

1 files changed, 257 insertions, 0 deletions

diff --git a/src/H5FDsubfiling/H5subfiling_common.h b/src/H5FDsubfiling/H5subfiling_common.h
new file mode 100644
index 0000000..cfcbf4a
--- /dev/null
+++ b/src/H5FDsubfiling/H5subfiling_common.h
@@ -0,0 +1,257 @@
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
+ * Copyright by The HDF Group.                                               *
+ * All rights reserved.                                                      *
+ *                                                                           *
+ * This file is part of HDF5.  The full HDF5 copyright notice, including     *
+ * terms governing use, modification, and redistribution, is contained in    *
+ * the COPYING file, which can be found at the root of the source code       *
+ * distribution tree, or in https://support.hdfgroup.org/ftp/HDF5/releases.  *
+ * If you do not have access to either file, you may request a copy from     *
+ * help@hdfgroup.org.                                                        *
+ * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+/*
+ * Header file for shared code between the HDF5 Subfiling VFD and IOC VFD
+ */
+
+#ifndef H5_SUBFILING_COMMON_H
+#define H5_SUBFILING_COMMON_H
+
+#include <stdatomic.h>
+
+#include "H5private.h"
+#include "H5Iprivate.h"
+
+/* TODO: needed for ioc_selection_t, which also needs to be public */
+#include "H5FDioc.h"
+
+/*
+ * Some definitions for debugging the Subfiling feature
+ */
+/* #define H5_SUBFILING_DEBUG */
+
+/*
+ * The following is our basic template for a subfile filename.
+ * Note that eventually we shouldn't use 0_of_N since we
+ * intend to use the user defined HDF5 filename for a
+ * zeroth subfile as well as for all metadata.
+ */
+#define SF_FILENAME_TEMPLATE ".subfile_%" PRIu64 "_%0*d_of_%d"
+
+/*
+ * The following is our basic template for a subfiling
+ * configuration filename.
+ */
+#define SF_CONFIG_FILENAME_TEMPLATE ".subfile_%" PRIu64 ".config"
+
+/*
+ * Environment variables interpreted by the HDF5 subfiling feature
+ */
+#define H5_IOC_SELECTION_CRITERIA "H5_IOC_SELECTION_CRITERIA"
+#define H5_IOC_COUNT_PER_NODE     "H5_IOC_COUNT_PER_NODE"
+#define H5_IOC_STRIPE_SIZE        "H5_IOC_STRIPE_SIZE"
+#define H5_IOC_SUBFILE_PREFIX     "H5_IOC_SUBFILE_PREFIX"
+
+#define H5FD_DEFAULT_STRIPE_DEPTH (32 * 1024 * 1024)
+
+/*
+ * MPI Tags are 32 bits, we treat them as unsigned
+ * to allow the use of the available bits for RPC
+ * selections, i.e. a message from the VFD read or write functions
+ * to an IO Concentrator.  The messages themselves are in general
+ * ONLY 3 int64_t values which define a) the data size to be read
+ * or written, b) the file offset where the data will be read from
+ * or stored, and c) the context_id allows the IO concentrator to
+ * locate the IO context for the new IO transaction.
+ *
+ *    0000
+ *    0001 READ_OP  (Independent)
+ *    0010 WRITE_OP (Independent)
+ *    0011 /////////
+ *    0100 CLOSE_OP (Independent)
+ *    -----
+ *    1000
+ *    1001 COLLECTIVE_READ
+ *    1010 COLLECTIVE_WRITE
+ *    1011 /////////
+ *    1100 COLLECTIVE_CLOSE
+ *
+ *   31    28      24      20      16      12       8       4       0|
+ *   +-------+-------+-------+-------+-------+-------+-------+-------+
+ *   |       |       |              ACKS             |      OP       |
+ *   +-------+-------+-------+-------+-------+-------+-------+-------+
+ *
+ */
+
+/* Bit 3 SET indicates collectives */
+#define COLL_FUNC (0x1 << 3)
+
+#define ACK_PART  (0x01 << 8)
+#define DATA_PART (0x02 << 8)
+#define READY     (0x04 << 8)
+#define COMPLETED (0x08 << 8)
+
+#define INT32_MASK 0x07FFFFFFFFFFFFFFF
+
+#define READ_INDEP  (READ_OP)
+#define READ_COLL   (COLL_FUNC | READ_OP)
+#define WRITE_INDEP (WRITE_OP)
+#define WRITE_COLL  (COLL_FUNC | WRITE_OP)
+
+#define GET_EOF_COMPLETED (COMPLETED | GET_EOF_OP)
+
+#define SET_LOGGING (LOGGING_OP)
+
+/* MPI tag values for data communicator */
+#define WRITE_INDEP_ACK 0
+#define READ_INDEP_DATA 1
+#define WRITE_TAG_BASE  2
+
+/*
+ * Object type definitions for subfiling objects.
+ * Used when generating a new subfiling object ID
+ * or accessing the cache of stored subfiling
+ * objects.
+ */
+typedef enum {
+    SF_BADID    = (-1),
+    SF_TOPOLOGY = 1,
+    SF_CONTEXT  = 2,
+    SF_NTYPES /* number of subfiling object types, MUST BE LAST */
+} sf_obj_type_t;
+
+/* The following are the basic 'op codes' used when
+ * constructing a RPC message for IO Concentrators.
+ * These are defined in the low 8 bits of the
+ * message.
+ */
+typedef enum io_ops {
+    READ_OP    = 1,
+    WRITE_OP   = 2,
+    OPEN_OP    = 3,
+    CLOSE_OP   = 4,
+    TRUNC_OP   = 5,
+    GET_EOF_OP = 6,
+    FINI_OP    = 8,
+    LOGGING_OP = 16
+} io_op_t;
+
+/* Every application rank will record their MPI rank
+ * and hostid as a structure.  These eventually get
+ * communicated to MPI rank zero(0) and sorted before
+ * being broadcast. The resulting sorted vector
+ * provides a basis for determining which MPI ranks
+ * will host an IO Concentrator (IOC), e.g. For
+ * default behavior, we choose the first vector entry
+ * associated with a "new" hostid.
+ */
+typedef struct {
+    long rank;
+    long hostid;
+} layout_t;
+
+/* This typedef defines a fixed process layout which
+ * can be reused for any number of file open operations
+ */
+typedef struct app_layout_t {
+    long      hostid;      /* value returned by gethostid()  */
+    layout_t *layout;      /* Vector of {rank,hostid} values */
+    int *     node_ranks;  /* ranks extracted from sorted layout */
+    int       node_count;  /* Total nodes (different hostids) */
+    int       node_index;  /* My node: index into node_ranks */
+    int       local_peers; /* How may local peers on my node */
+    int       world_rank;  /* My MPI rank                    */
+    int       world_size;  /* Total number of MPI ranks      */
+} app_layout_t;
+
+/*  This typedef defines things related to IOC selections */
+typedef struct topology {
+    app_layout_t *  app_layout;         /* Pointer to our layout struct   */
+    bool            rank_is_ioc;        /* Indicates that we host an IOC  */
+    int             subfile_rank;       /* Valid only if rank_is_ioc      */
+    int             n_io_concentrators; /* Number of IO concentrators  */
+    int *           io_concentrators;   /* Vector of ranks which are IOCs */
+    int *           subfile_fd;         /* file descriptor (if IOC)       */
+    ioc_selection_t selection_type;     /* Cache our IOC selection criteria */
+} sf_topology_t;
+
+typedef struct {
+    int64_t        sf_context_id;           /* Generated context ID which embeds the cache index */
+    uint64_t       h5_file_id;              /* GUID (basically the inode value) */
+    int            sf_fid;                  /* value returned by open(file,..)  */
+    size_t         sf_write_count;          /* Statistics: write_count  */
+    size_t         sf_read_count;           /* Statistics: read_count  */
+    haddr_t        sf_eof;                  /* File eof */
+    int64_t        sf_stripe_size;          /* Stripe-depth */
+    int64_t        sf_blocksize_per_stripe; /* Stripe-depth X n_IOCs  */
+    int64_t        sf_base_addr;            /* For an IOC, our base address         */
+    MPI_Comm       sf_msg_comm;             /* MPI comm used to send RPC msg        */
+    MPI_Comm       sf_data_comm;            /* MPI comm used to move data           */
+    MPI_Comm       sf_eof_comm;             /* MPI comm used to communicate EOF     */
+    MPI_Comm       sf_barrier_comm;         /* MPI comm used for barrier operations */
+    MPI_Comm       sf_group_comm;           /* Not used: for IOC collectives        */
+    MPI_Comm       sf_intercomm;            /* Not used: for msgs to all IOC        */
+    int            sf_group_size;           /* IOC count (in sf_group_comm)         */
+    int            sf_group_rank;           /* IOC rank  (in sf_group_comm)         */
+    int            sf_intercomm_root;       /* Not used: for IOC comms              */
+    char *         subfile_prefix;          /* If subfiles are node-local           */
+    char *         sf_filename;             /* A generated subfile name             */
+    char *         h5_filename;             /* The user supplied file name          */
+    void *         ioc_data;                /* Private data for underlying IOC      */
+    sf_topology_t *topology;                /* pointer to our topology              */
+
+#ifdef H5_SUBFILING_DEBUG
+    char  sf_logfile_name[PATH_MAX];
+    FILE *sf_logfile;
+#endif
+
+} subfiling_context_t;
+
+/* The following is a somewhat augmented input (by the IOC) which captures
+ * the basic RPC from a 'source'.   The fields are filled out to allow
+ * an easy gathering of statistics by the IO Concentrator.
+ */
+typedef struct {
+    /* {Datasize, Offset, FileID} */
+    int64_t header[3];        /* The basic RPC input plus       */
+    int     tag;              /* the supplied OPCODE tag        */
+    int     source;           /* Rank of who sent the message   */
+    int     subfile_rank;     /* The IOC rank                   */
+    int64_t context_id;       /* context to be used to complete */
+    double  start_time;       /* the request, + time of receipt */
+                              /* from which we calc Time(queued) */
+    void *buffer;             /* for writes, we keep the buffer */
+                              /* around for awhile...           */
+    volatile int in_progress; /* Not used!               */
+    volatile int serialize;   /* worker thread needs to wait while true */
+    volatile int dependents;  //* If current work item has dependents */
+    int          depend_id;   /* work queue index of the dependent */
+} sf_work_request_t;
+
+extern int sf_verbose_flag;
+
+extern app_layout_t *sf_app_layout;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+H5_DLL herr_t H5_open_subfiles(const char *base_filename, uint64_t h5_file_id,
+                               ioc_selection_t ioc_selection_type, int file_acc_flags, MPI_Comm file_comm,
+                               int64_t *context_id_out);
+H5_DLL herr_t H5_close_subfiles(int64_t subfiling_context_id);
+
+H5_DLL int64_t H5_new_subfiling_object_id(sf_obj_type_t obj_type, int64_t index_val);
+H5_DLL void *  H5_get_subfiling_object(int64_t object_id);
+H5_DLL int64_t H5_subfile_fid_to_context(uint64_t h5_fid);
+H5_DLL herr_t  H5_free_subfiling_object(int64_t object_id);
+
+H5_DLL void H5_subfiling_log(int64_t sf_context_id, const char *fmt, ...);
+
+void set_verbose_flag(int subfile_rank, int new_value);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* H5_SUBFILING_COMMON_H */