darshan-posix.c

/*
 *  (C) 2009 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

#include "darshan-runtime-config.h"
#include <stdio.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <stdarg.h>
#include <string.h>
#include <time.h>
#include <stdlib.h>
#include <errno.h>
#include <sys/uio.h>
#include <sys/mman.h>
#include <search.h>
#include <assert.h>
#define __USE_GNU
#include <pthread.h>

#include "darshan.h"

#ifndef HAVE_OFF64_T
typedef int64_t off64_t;
#endif

extern char* __progname_full;

#ifdef DARSHAN_PRELOAD
#define __USE_GNU
#include <dlfcn.h>
#include <stdlib.h>

#define DARSHAN_FORWARD_DECL(name,ret,args) \
  ret (*__real_ ## name)args = NULL;

#define DARSHAN_DECL(__name) __name

#define DARSHAN_MPI_CALL(func) __real_ ## func

#define MAP_OR_FAIL(func) \
    if (!(__real_ ## func)) \
    { \
        __real_ ## func = dlsym(RTLD_NEXT, #func); \
        if(!(__real_ ## func)) { \
           fprintf(stderr, "Darshan failed to map symbol: %s\n", #func); \
           exit(1); \
       } \
    }


extern double (*__real_PMPI_Wtime)(void);

#else

#define DARSHAN_FORWARD_DECL(name,ret,args) \
  extern ret __real_ ## name args;

#define DARSHAN_DECL(__name) __wrap_ ## __name

#define MAP_OR_FAIL(func)

#define DARSHAN_MPI_CALL(func) func

#endif

DARSHAN_FORWARD_DECL(creat, int, (const char* path, mode_t mode));
DARSHAN_FORWARD_DECL(creat64, int, (const char* path, mode_t mode));
DARSHAN_FORWARD_DECL(open, int, (const char *path, int flags, ...));
DARSHAN_FORWARD_DECL(open64, int, (const char *path, int flags, ...));
DARSHAN_FORWARD_DECL(close, int, (int fd));
DARSHAN_FORWARD_DECL(write, ssize_t, (int fd, const void *buf, size_t count));
DARSHAN_FORWARD_DECL(read, ssize_t, (int fd, void *buf, size_t count));
DARSHAN_FORWARD_DECL(lseek, off_t, (int fd, off_t offset, int whence));
DARSHAN_FORWARD_DECL(lseek64, off64_t, (int fd, off64_t offset, int whence));
DARSHAN_FORWARD_DECL(pread, ssize_t, (int fd, void *buf, size_t count, off_t offset));
DARSHAN_FORWARD_DECL(pread64, ssize_t, (int fd, void *buf, size_t count, off64_t offset));
DARSHAN_FORWARD_DECL(pwrite, ssize_t, (int fd, const void *buf, size_t count, off_t offset));
DARSHAN_FORWARD_DECL(pwrite64, ssize_t, (int fd, const void *buf, size_t count, off64_t offset
));
DARSHAN_FORWARD_DECL(readv, ssize_t, (int fd, const struct iovec *iov, int iovcnt));
DARSHAN_FORWARD_DECL(writev, ssize_t, (int fd, const struct iovec *iov, int iovcnt));
DARSHAN_FORWARD_DECL(__fxstat, int, (int vers, int fd, struct stat *buf));
DARSHAN_FORWARD_DECL(__fxstat64, int, (int vers, int fd, struct stat64 *buf));
DARSHAN_FORWARD_DECL(__lxstat, int, (int vers, const char* path, struct stat *buf));
DARSHAN_FORWARD_DECL(__lxstat64, int, (int vers, const char* path, struct stat64 *buf));
DARSHAN_FORWARD_DECL(__xstat, int, (int vers, const char* path, struct stat *buf));
DARSHAN_FORWARD_DECL(__xstat64, int, (int vers, const char* path, struct stat64 *buf));
DARSHAN_FORWARD_DECL(mmap, void*, (void *addr, size_t length, int prot, int flags, int fd, off_t offset));
DARSHAN_FORWARD_DECL(mmap64, void*, (void *addr, size_t length, int prot, int flags, int fd, off64_t offset));
DARSHAN_FORWARD_DECL(fopen, FILE*, (const char *path, const char *mode));
DARSHAN_FORWARD_DECL(fopen64, FILE*, (const char *path, const char *mode));
DARSHAN_FORWARD_DECL(fclose, int, (FILE *fp));
DARSHAN_FORWARD_DECL(fread, size_t, (void *ptr, size_t size, size_t nmemb, FILE *stream));
DARSHAN_FORWARD_DECL(fwrite, size_t, (const void *ptr, size_t size, size_t nmemb, FILE *stream));
DARSHAN_FORWARD_DECL(fseek, int, (FILE *stream, long offset, int whence));
DARSHAN_FORWARD_DECL(fsync, int, (int fd));
DARSHAN_FORWARD_DECL(fdatasync, int, (int fd));

pthread_mutex_t cp_mutex = PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP;
struct darshan_job_runtime* darshan_global_job = NULL;
static int my_rank = -1;
static struct stat64 cp_stat_buf;
static int darshan_mem_alignment = 1;

/* these are paths that we will not trace */
static char* exclusions[] = {
"/etc/",
"/dev/",
"/usr/",
"/bin/",
"/boot/",
"/lib/",
"/opt/",
"/sbin/",
"/sys/",
"/proc/",
NULL
};

static double posix_wtime(void);

static void cp_access_counter(struct darshan_file_runtime* file, ssize_t size,     enum cp_counter_type type);

#define CP_RECORD_WRITE(__ret, __fd, __count, __pwrite_flag, __pwrite_offset, __aligned, __stream_flag, __tm1, __tm2) do{ \
    size_t stride; \
    int64_t this_offset; \
    int64_t file_alignment; \
    struct darshan_file_runtime* file; \
    double __elapsed = __tm2-__tm1; \
    if(__ret < 0) break; \
    file = darshan_file_by_fd(__fd); \
    if(!file) break; \
    if(__pwrite_flag) \
        this_offset = __pwrite_offset; \
    else \
        this_offset = file->offset; \
    file_alignment = CP_VALUE(file, CP_FILE_ALIGNMENT); \
    if(this_offset > file->last_byte_written) \
        CP_INC(file, CP_SEQ_WRITES, 1); \
    if(this_offset == (file->last_byte_written + 1)) \
        CP_INC(file, CP_CONSEC_WRITES, 1); \
    if(this_offset > 0 && this_offset > file->last_byte_written \
        && file->last_byte_written != 0) \
        stride = this_offset - file->last_byte_written - 1; \
    else \
        stride = 0; \
    file->last_byte_written = this_offset + __ret - 1; \
    file->offset = this_offset + __ret; \
    CP_MAX(file, CP_MAX_BYTE_WRITTEN, (this_offset + __ret -1)); \
    CP_INC(file, CP_BYTES_WRITTEN, __ret); \
    if(__stream_flag) \
        CP_INC(file, CP_POSIX_FWRITES, 1); \
    else \
        CP_INC(file, CP_POSIX_WRITES, 1); \
    CP_BUCKET_INC(file, CP_SIZE_WRITE_0_100, __ret); \
    cp_access_counter(file, stride, CP_COUNTER_STRIDE); \
    if(!__aligned) \
        CP_INC(file, CP_MEM_NOT_ALIGNED, 1); \
    if(file_alignment && (this_offset % file_alignment) != 0) \
        CP_INC(file, CP_FILE_NOT_ALIGNED, 1); \
    cp_access_counter(file, __ret, CP_COUNTER_ACCESS); \
    if(file->last_io_type == CP_READ) \
        CP_INC(file, CP_RW_SWITCHES, 1); \
    file->last_io_type = CP_WRITE; \
    CP_F_INC(file, CP_F_POSIX_WRITE_TIME, (__elapsed)); \
    if(CP_F_VALUE(file, CP_F_WRITE_START_TIMESTAMP) == 0) \
        CP_F_SET(file, CP_F_WRITE_START_TIMESTAMP, __tm1); \
    CP_F_SET(file, CP_F_WRITE_END_TIMESTAMP, __tm2); \
    if(CP_F_VALUE(file, CP_F_MAX_WRITE_TIME) < __elapsed){ \
        CP_F_SET(file, CP_F_MAX_WRITE_TIME, __elapsed); \
        CP_SET(file, CP_MAX_WRITE_TIME_SIZE, __ret); } \
} while(0)

#define CP_RECORD_READ(__ret, __fd, __count, __pread_flag, __pread_offset, __aligned, __stream_flag, __tm1, __tm2) do{ \
    size_t stride; \
    int64_t this_offset; \
    struct darshan_file_runtime* file; \
    int64_t file_alignment; \
    double __elapsed = __tm2-__tm1; \
    if(__ret < 0) break; \
    file = darshan_file_by_fd(__fd); \
    if(!file) break; \
    if(__pread_flag)\
        this_offset = __pread_offset; \
    else \
        this_offset = file->offset; \
    file_alignment = CP_VALUE(file, CP_FILE_ALIGNMENT); \
    if(this_offset > file->last_byte_read) \
        CP_INC(file, CP_SEQ_READS, 1); \
    if(this_offset == (file->last_byte_read + 1)) \
        CP_INC(file, CP_CONSEC_READS, 1); \
    if(this_offset > 0 && this_offset > file->last_byte_read \
        && file->last_byte_read != 0) \
        stride = this_offset - file->last_byte_read - 1; \
    else \
        stride = 0; \
    file->last_byte_read = this_offset + __ret - 1; \
    CP_MAX(file, CP_MAX_BYTE_READ, (this_offset + __ret -1)); \
    file->offset = this_offset + __ret; \
    CP_INC(file, CP_BYTES_READ, __ret); \
    if(__stream_flag)\
        CP_INC(file, CP_POSIX_FREADS, 1); \
    else\
        CP_INC(file, CP_POSIX_READS, 1); \
    CP_BUCKET_INC(file, CP_SIZE_READ_0_100, __ret); \
    cp_access_counter(file, stride, CP_COUNTER_STRIDE); \
    if(!__aligned) \
        CP_INC(file, CP_MEM_NOT_ALIGNED, 1); \
    if(file_alignment && (this_offset % file_alignment) != 0) \
        CP_INC(file, CP_FILE_NOT_ALIGNED, 1); \
    cp_access_counter(file, __ret, CP_COUNTER_ACCESS); \
    if(file->last_io_type == CP_WRITE) \
        CP_INC(file, CP_RW_SWITCHES, 1); \
    file->last_io_type = CP_READ; \
    CP_F_INC(file, CP_F_POSIX_READ_TIME, (__elapsed)); \
    if(CP_F_VALUE(file, CP_F_READ_START_TIMESTAMP) == 0) \
        CP_F_SET(file, CP_F_READ_START_TIMESTAMP, __tm1); \
    CP_F_SET(file, CP_F_READ_END_TIMESTAMP, __tm2); \
    if(CP_F_VALUE(file, CP_F_MAX_READ_TIME) < __elapsed){ \
        CP_F_SET(file, CP_F_MAX_READ_TIME, __elapsed); \
        CP_SET(file, CP_MAX_READ_TIME_SIZE, __ret); } \
} while(0)

#define CP_LOOKUP_RECORD_STAT(__path, __statbuf, __tm1, __tm2) do { \
    char* exclude; \
    int tmp_index = 0; \
    struct darshan_file_runtime* file; \
    while((exclude = exclusions[tmp_index])) { \
        if(!(strncmp(exclude, __path, strlen(exclude)))) \
            break; \
        tmp_index++; \
    } \
    if(exclude) break; \
    file = darshan_file_by_name(__path); \
    if (file) \
    { \
        CP_RECORD_STAT(file, __statbuf, __tm1, __tm2); \
    } \
} while(0)

    
#define CP_RECORD_STAT(__file, __statbuf, __tm1, __tm2) do { \
    if(!CP_VALUE((__file), CP_FILE_ALIGNMENT)){ \
        CP_SET((__file), CP_DEVICE, (__statbuf)->st_dev); \
        CP_SET((__file), CP_FILE_ALIGNMENT, (__statbuf)->st_blksize); \
        CP_SET((__file), CP_SIZE_AT_OPEN, (__statbuf)->st_size); \
    }\
    (__file)->log_file->rank = my_rank; \
    CP_F_INC(__file, CP_F_POSIX_META_TIME, (__tm2-__tm1)); \
    CP_INC(__file, CP_POSIX_STATS, 1); \
} while(0)

#define CP_RECORD_OPEN(__ret, __path, __mode, __stream_flag, __tm1, __tm2) do { \
    struct darshan_file_runtime* file; \
    char* exclude; \
    int tmp_index = 0; \
    int hash_index; \
    if(__ret < 0) break; \
    while((exclude = exclusions[tmp_index])) { \
        if(!(strncmp(exclude, __path, strlen(exclude)))) \
            break; \
        tmp_index++; \
    } \
    if(exclude) break; \
    file = darshan_file_by_name(__path); \
    if(!file) break; \
    if(file->fd != -1) break; /* TODO: handle multiple concurrent opens */ \
    file->fd = __ret; \
    if(!CP_VALUE(file, CP_FILE_ALIGNMENT)){ \
        if(fstat64(file->fd, &cp_stat_buf) == 0) {\
            CP_SET(file, CP_DEVICE, cp_stat_buf.st_dev); \
            CP_SET(file, CP_FILE_ALIGNMENT, cp_stat_buf.st_blksize); \
            CP_SET(file, CP_SIZE_AT_OPEN, cp_stat_buf.st_size); \
        }\
    }\
    file->log_file->rank = my_rank; \
    if(__mode) \
        CP_SET(file, CP_MODE, __mode); \
    file->offset = 0; \
    file->last_byte_written = 0; \
    file->last_byte_read = 0; \
    if(__stream_flag)\
        CP_INC(file, CP_POSIX_FOPENS, 1); \
    else \
        CP_INC(file, CP_POSIX_OPENS, 1); \
    if(CP_F_VALUE(file, CP_F_OPEN_TIMESTAMP) == 0) \
        CP_F_SET(file, CP_F_OPEN_TIMESTAMP, posix_wtime()); \
    CP_F_INC(file, CP_F_POSIX_META_TIME, (__tm2-__tm1)); \
    hash_index = file->fd & CP_HASH_MASK; \
    file->fd_prev = NULL; \
    file->fd_next = darshan_global_job->fd_table[hash_index]; \
    if(file->fd_next) \
        file->fd_next->fd_prev = file; \
    darshan_global_job->fd_table[hash_index] = file; \
} while (0)

int DARSHAN_DECL(close)(int fd)
{
    struct darshan_file_runtime* file;
    int hash_index;
    int tmp_fd = fd;
    double tm1, tm2;
    int ret;

    MAP_OR_FAIL(close);

    tm1 = darshan_wtime();
    ret = __real_close(fd);
    tm2 = darshan_wtime();

    CP_LOCK();
    file = darshan_file_by_fd(tmp_fd);
    if(file)
    {
        file->fd = -1;
        file->last_byte_written = 0;
        file->last_byte_read = 0;
        CP_F_SET(file, CP_F_CLOSE_TIMESTAMP, posix_wtime());
        CP_F_INC(file, CP_F_POSIX_META_TIME, (tm2-tm1));
        if(file->fd_prev == NULL)
        {
            /* head of fd hash table list */
            hash_index = tmp_fd & CP_HASH_MASK;
            darshan_global_job->fd_table[hash_index] = file->fd_next;
            if(file->fd_next)
                file->fd_next->fd_prev = NULL;
        }
        else
        {
            if(file->fd_prev)
                file->fd_prev->fd_next = file->fd_next;
            if(file->fd_next)
                file->fd_next->fd_prev = file->fd_prev;
        }
        file->fd_prev = NULL;
        file->fd_next = NULL;
        darshan_global_job->darshan_mru_file = file; /* in case we open it again */
    }
    CP_UNLOCK();

    return(ret);
}

int DARSHAN_DECL(fclose)(FILE *fp)
{
    struct darshan_file_runtime* file;
    int hash_index;
    int tmp_fd = fileno(fp);
    double tm1, tm2;
    int ret;

    MAP_OR_FAIL(fclose);

    tm1 = darshan_wtime();
    ret = __real_fclose(fp);
    tm2 = darshan_wtime();
    
    CP_LOCK();
    file = darshan_file_by_fd(tmp_fd);
    if(file)
    {
        file->fd = -1;
        file->last_byte_written = 0;
        file->last_byte_read = 0;
        CP_F_SET(file, CP_F_CLOSE_TIMESTAMP, posix_wtime());
        CP_F_INC(file, CP_F_POSIX_META_TIME, (tm2-tm1));
        if(file->fd_prev == NULL)
        {
            /* head of fd hash table list */
            hash_index = tmp_fd & CP_HASH_MASK;
            darshan_global_job->fd_table[hash_index] = file->fd_next;
            if(file->fd_next)
                file->fd_next->fd_prev = NULL;
        }
        else
        {
            if(file->fd_prev)
                file->fd_prev->fd_next = file->fd_next;
            if(file->fd_next)
                file->fd_next->fd_prev = file->fd_prev;
        }
        file->fd_prev = NULL;
        file->fd_next = NULL;
        darshan_global_job->darshan_mru_file = file; /* in case we open it again */
    }
    CP_UNLOCK();

    return(ret);
}


int DARSHAN_DECL(fsync)(int fd)
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

    MAP_OR_FAIL(fsync);

    tm1 = darshan_wtime();
    ret = __real_fsync(fd);
    tm2 = darshan_wtime();

    if(ret < 0)
        return(ret);

    CP_LOCK();
    file = darshan_file_by_fd(fd);
    if(file)
    {
        CP_F_INC(file, CP_F_POSIX_WRITE_TIME, (tm2-tm1));
        CP_INC(file, CP_POSIX_FSYNCS, 1);
    }
    CP_UNLOCK();

    return(ret);
}

int DARSHAN_DECL(fdatasync)(int fd)
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

    MAP_OR_FAIL(fdatasync);

    tm1 = darshan_wtime();
    ret = __real_fdatasync(fd);
    tm2 = darshan_wtime();
    if(ret < 0)
        return(ret);

    CP_LOCK();
    file = darshan_file_by_fd(fd);
    if(file)
    {
        CP_F_INC(file, CP_F_POSIX_WRITE_TIME, (tm2-tm1));
        CP_INC(file, CP_POSIX_FDSYNCS, 1);
    }
    CP_UNLOCK();

    return(ret);
}


void* DARSHAN_DECL(mmap64)(void *addr, size_t length, int prot, int flags,
    int fd, off64_t offset)
{
    void* ret;
    struct darshan_file_runtime* file;

    MAP_OR_FAIL(mmap64);

    ret = __real_mmap64(addr, length, prot, flags, fd, offset);
    if(ret == MAP_FAILED)
        return(ret);

    CP_LOCK();
    file = darshan_file_by_fd(fd);
    if(file)
    {
        CP_INC(file, CP_POSIX_MMAPS, 1);
    }
    CP_UNLOCK();

    return(ret);
}


void* DARSHAN_DECL(mmap)(void *addr, size_t length, int prot, int flags,
    int fd, off_t offset)
{
    void* ret;
    struct darshan_file_runtime* file;

    MAP_OR_FAIL(mmap);

    ret = __real_mmap(addr, length, prot, flags, fd, offset);
    if(ret == MAP_FAILED)
        return(ret);

    CP_LOCK();
    file = darshan_file_by_fd(fd);
    if(file)
    {
        CP_INC(file, CP_POSIX_MMAPS, 1);
    }
    CP_UNLOCK();

    return(ret);
}

int DARSHAN_DECL(creat)(const char* path, mode_t mode)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(creat);

    tm1 = darshan_wtime();
    ret = __real_creat(path, mode);
    tm2 = darshan_wtime();

    CP_LOCK();
    CP_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

int DARSHAN_DECL(creat64)(const char* path, mode_t mode)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(creat64);

    tm1 = darshan_wtime();
    ret = __real_creat64(path, mode);
    tm2 = darshan_wtime();

    CP_LOCK();
    CP_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

int DARSHAN_DECL(open64)(const char* path, int flags, ...)
{
    int mode = 0;
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(open64);

    if (flags & O_CREAT) 
    {
        va_list arg;
        va_start(arg, flags);
        mode = va_arg(arg, int);
        va_end(arg);

        tm1 = darshan_wtime();
        ret = __real_open64(path, flags, mode);
        tm2 = darshan_wtime();
    }
    else
    {
        tm1 = darshan_wtime();
        ret = __real_open64(path, flags);
        tm2 = darshan_wtime();
    }

    CP_LOCK();
    CP_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

int DARSHAN_DECL(open)(const char *path, int flags, ...)
{
    int mode = 0;
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(open);

    if (flags & O_CREAT) 
    {
        va_list arg;
        va_start(arg, flags);
        mode = va_arg(arg, int);
        va_end(arg);

        tm1 = darshan_wtime();
        ret = __real_open(path, flags, mode);
        tm2 = darshan_wtime();
    }
    else
    {
        tm1 = darshan_wtime();
        ret = __real_open(path, flags);
        tm2 = darshan_wtime();
    }

    CP_LOCK();
    CP_RECORD_OPEN(ret, path, mode, 0, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

FILE* DARSHAN_DECL(fopen64)(const char *path, const char *mode)
{
    FILE* ret;
    int fd;
    double tm1, tm2;

    MAP_OR_FAIL(fopen64);

    tm1 = darshan_wtime();
    ret = __real_fopen64(path, mode);
    tm2 = darshan_wtime();
    if(ret == 0)
        fd = -1;
    else
        fd = fileno(ret);

    CP_LOCK();
    CP_RECORD_OPEN(fd, path, 0, 1, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

FILE* DARSHAN_DECL(fopen)(const char *path, const char *mode)
{
    FILE* ret;
    int fd;
    double tm1, tm2;

    MAP_OR_FAIL(fopen);

    tm1 = darshan_wtime();
    ret = __real_fopen(path, mode);
    tm2 = darshan_wtime();
    if(ret == 0)
        fd = -1;
    else
        fd = fileno(ret);

    CP_LOCK();
    CP_RECORD_OPEN(fd, path, 0, 1, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

int DARSHAN_DECL(__xstat64)(int vers, const char *path, struct stat64 *buf)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(__xstat64);

    tm1 = darshan_wtime();
    ret = __real___xstat64(vers, path, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    CP_LOCK();
    CP_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

int DARSHAN_DECL(__lxstat64)(int vers, const char *path, struct stat64 *buf)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(__lxstat64);

    tm1 = darshan_wtime();
    ret = __real___lxstat64(vers, path, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    CP_LOCK();
    CP_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

int DARSHAN_DECL(__fxstat64)(int vers, int fd, struct stat64 *buf)
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

    MAP_OR_FAIL(__fxstat64);

    tm1 = darshan_wtime();
    ret = __real___fxstat64(vers, fd, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    /* skip logging if this was triggered internally */
    if(buf == &cp_stat_buf)
        return(ret);

    CP_LOCK();
    file = darshan_file_by_fd(fd);
    if(file)
    {
        CP_RECORD_STAT(file, buf, tm1, tm2);
    }
    CP_UNLOCK();

    return(ret);
}


int DARSHAN_DECL(__xstat)(int vers, const char *path, struct stat *buf)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(__xstat);

    tm1 = darshan_wtime();
    ret = __real___xstat(vers, path, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    CP_LOCK();
    CP_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

int DARSHAN_DECL(__lxstat)(int vers, const char *path, struct stat *buf)
{
    int ret;
    double tm1, tm2;

    MAP_OR_FAIL(__lxstat);

    tm1 = darshan_wtime();
    ret = __real___lxstat(vers, path, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    CP_LOCK();
    CP_LOOKUP_RECORD_STAT(path, buf, tm1, tm2);
    CP_UNLOCK();

    return(ret);
}

int DARSHAN_DECL(__fxstat)(int vers, int fd, struct stat *buf)
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

    MAP_OR_FAIL(__fxstat);

    tm1 = darshan_wtime();
    ret = __real___fxstat(vers, fd, buf);
    tm2 = darshan_wtime();
    if(ret < 0 || !S_ISREG(buf->st_mode))
        return(ret);

    /* skip logging if this was triggered internally */
    if((void*)buf == (void*)&cp_stat_buf)
        return(ret);

    CP_LOCK();
    file = darshan_file_by_fd(fd);
    if(file)
    {
        CP_RECORD_STAT(file, buf, tm1, tm2);
    }
    CP_UNLOCK();

    return(ret);
}

ssize_t DARSHAN_DECL(pread64)(int fd, void *buf, size_t count, off64_t offset)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(pread64);

    if((unsigned long)buf % darshan_mem_alignment == 0)
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_pread64(fd, buf, count, offset);
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_READ(ret, fd, count, 1, offset, aligned_flag, 0, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

ssize_t DARSHAN_DECL(pread)(int fd, void *buf, size_t count, off_t offset)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(pread);

    if((unsigned long)buf % darshan_mem_alignment == 0)
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_pread(fd, buf, count, offset);
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_READ(ret, fd, count, 1, offset, aligned_flag, 0, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}


ssize_t DARSHAN_DECL(pwrite)(int fd, const void *buf, size_t count, off_t offset)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(pwrite);

    if((unsigned long)buf % darshan_mem_alignment == 0)
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_pwrite(fd, buf, count, offset);
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_WRITE(ret, fd, count, 1, offset, aligned_flag, 0, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

ssize_t DARSHAN_DECL(pwrite64)(int fd, const void *buf, size_t count, off64_t offset)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(pwrite64);

    if((unsigned long)buf % darshan_mem_alignment == 0)
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_pwrite64(fd, buf, count, offset);
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_WRITE(ret, fd, count, 1, offset, aligned_flag, 0, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

ssize_t DARSHAN_DECL(readv)(int fd, const struct iovec *iov, int iovcnt)
{
    ssize_t ret;
    int aligned_flag = 1;
    int i;
    double tm1, tm2;

    MAP_OR_FAIL(readv);

    for(i=0; i<iovcnt; i++)
    {
        if(((unsigned long)iov[i].iov_base % darshan_mem_alignment) != 0)
            aligned_flag = 0;
    }

    tm1 = darshan_wtime();
    ret = __real_readv(fd, iov, iovcnt);
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_READ(ret, fd, count, 0, 0, aligned_flag, 0, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

ssize_t DARSHAN_DECL(writev)(int fd, const struct iovec *iov, int iovcnt)
{
    ssize_t ret;
    int aligned_flag = 1;
    int i;
    double tm1, tm2;

    MAP_OR_FAIL(writev);

    for(i=0; i<iovcnt; i++)
    {
        if(!((unsigned long)iov[i].iov_base % darshan_mem_alignment == 0))
            aligned_flag = 0;
    }

    tm1 = darshan_wtime();
    ret = __real_writev(fd, iov, iovcnt);
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_WRITE(ret, fd, count, 0, 0, aligned_flag, 0, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

size_t DARSHAN_DECL(fread)(void *ptr, size_t size, size_t nmemb, FILE *stream)
{
    size_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(fread);

    if((unsigned long)ptr % darshan_mem_alignment == 0)
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_fread(ptr, size, nmemb, stream);
    tm2 = darshan_wtime();
    CP_LOCK();
    if(ret > 0)
        CP_RECORD_READ(size*ret, fileno(stream), (size*nmemb), 0, 0, aligned_flag, 1, tm1, tm2);
    else
        CP_RECORD_READ(ret, fileno(stream), (size*nmemb), 0, 0, aligned_flag, 1, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

ssize_t DARSHAN_DECL(read)(int fd, void *buf, size_t count)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(read);

    if((unsigned long)buf % darshan_mem_alignment == 0)
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_read(fd, buf, count);
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_READ(ret, fd, count, 0, 0, aligned_flag, 0, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

ssize_t DARSHAN_DECL(write)(int fd, const void *buf, size_t count)
{
    ssize_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(write);

    if((unsigned long)buf % darshan_mem_alignment == 0)
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_write(fd, buf, count);
    tm2 = darshan_wtime();
    CP_LOCK();
    CP_RECORD_WRITE(ret, fd, count, 0, 0, aligned_flag, 0, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

size_t DARSHAN_DECL(fwrite)(const void *ptr, size_t size, size_t nmemb, FILE *stream)
{
    size_t ret;
    int aligned_flag = 0;
    double tm1, tm2;

    MAP_OR_FAIL(fwrite);

    if((unsigned long)ptr % darshan_mem_alignment == 0)
        aligned_flag = 1;

    tm1 = darshan_wtime();
    ret = __real_fwrite(ptr, size, nmemb, stream);
    tm2 = darshan_wtime();
    CP_LOCK();
    if(ret > 0)
        CP_RECORD_WRITE(size*ret, fileno(stream), (size*nmemb), 0, 0, aligned_flag, 1, tm1, tm2);
    else
        CP_RECORD_WRITE(ret, fileno(stream), 0, 0, 0, aligned_flag, 1, tm1, tm2);
    CP_UNLOCK();
    return(ret);
}

off64_t DARSHAN_DECL(lseek64)(int fd, off64_t offset, int whence)
{
    off64_t ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

    MAP_OR_FAIL(lseek64);

    tm1 = darshan_wtime();
    ret = __real_lseek64(fd, offset, whence);
    tm2 = darshan_wtime();
    if(ret >= 0)
    {
        CP_LOCK();
        file = darshan_file_by_fd(fd);
        if(file)
        {
            file->offset = ret;
            CP_F_INC(file, CP_F_POSIX_META_TIME, (tm2-tm1));
            CP_INC(file, CP_POSIX_SEEKS, 1);
        }
        CP_UNLOCK();
    }
    return(ret);
}

off_t DARSHAN_DECL(lseek)(int fd, off_t offset, int whence)
{
    off_t ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

    MAP_OR_FAIL(lseek);

    tm1 = darshan_wtime();
    ret = __real_lseek(fd, offset, whence);
    tm2 = darshan_wtime();
    if(ret >= 0)
    {
        CP_LOCK();
        file = darshan_file_by_fd(fd);
        if(file)
        {
            file->offset = ret;
            CP_F_INC(file, CP_F_POSIX_META_TIME, (tm2-tm1));
            CP_INC(file, CP_POSIX_SEEKS, 1);
        }
        CP_UNLOCK();
    }
    return(ret);
}

int DARSHAN_DECL(fseek)(FILE *stream, long offset, int whence)
{
    int ret;
    struct darshan_file_runtime* file;
    double tm1, tm2;

    MAP_OR_FAIL(fseek);

    tm1 = darshan_wtime();
    ret = __real_fseek(stream, offset, whence);
    tm2 = darshan_wtime();
    if(ret >= 0)
    {
        CP_LOCK();
        file = darshan_file_by_fd(fileno(stream));
        if(file)
        {
            file->offset = ret;
            CP_F_INC(file, CP_F_POSIX_META_TIME, (tm2-tm1));
            CP_INC(file, CP_POSIX_FSEEKS, 1);
        }
        CP_UNLOCK();
    }
    return(ret);
}

void darshan_finalize(struct darshan_job_runtime* job)
{
    if(!job)
    {
        return;
    }

    free(job);
}

void darshan_initialize(int argc, char** argv,  int nprocs, int rank)
{
    int i;
    char* disable;
    char* disable_timing;
    char* envstr;
    char* truncate_string = "<TRUNCATED>";
    int truncate_offset;
    int chars_left = 0;
    int ret;
    int tmpval;

    disable = getenv("DARSHAN_DISABLE");
    if(disable)
    {
        /* turn off tracing */
        return;
    }

    disable_timing = getenv("DARSHAN_DISABLE_TIMING");

    if(darshan_global_job != NULL)
    {
        return;
    }

    #if (__CP_MEM_ALIGNMENT < 1)
        #error Darshan must be configured with a positive value for --with-mem-align
    #endif
    envstr = getenv("DARSHAN_MEMALIGN");
    if (envstr)
    {
        ret = sscanf(envstr, "%d", &tmpval);
        /* silently ignore if the env variable is set poorly */
        if(ret == 1 && tmpval > 0)
        {
            darshan_mem_alignment = tmpval;
        }
    }
    else
    {
        darshan_mem_alignment = __CP_MEM_ALIGNMENT;
    }

    /* avoid floating point errors on faulty input */
    if (darshan_mem_alignment < 1)
    {
        darshan_mem_alignment = 1;
    }

    /* allocate structure to track darshan_global_job information */
    darshan_global_job = malloc(sizeof(*darshan_global_job));
    if(!darshan_global_job)
    {
        return;
    }
    memset(darshan_global_job, 0, sizeof(*darshan_global_job));

    if(disable_timing)
    {
        darshan_global_job->flags |= CP_FLAG_NOTIMING;
    }

    /* set up file records */
    for(i=0; i<CP_MAX_FILES; i++)
    {
        darshan_global_job->file_runtime_array[i].log_file = 
            &darshan_global_job->file_array[i];
        darshan_global_job->file_runtime_array[i].fd = -1;
        darshan_global_job->file_runtime_array[i].ncid = -1;
        darshan_global_job->file_runtime_array[i].hid = -1;
        darshan_global_job->file_runtime_array[i].fh = MPI_FILE_NULL;
    }

    strcpy(darshan_global_job->log_job.version_string, CP_VERSION);
    darshan_global_job->log_job.magic_nr = CP_MAGIC_NR;
    darshan_global_job->log_job.uid = getuid();
    darshan_global_job->log_job.start_time = time(NULL);
    darshan_global_job->log_job.nprocs = nprocs;
    my_rank = rank;

    /* record exe and arguments */
    for(i=0; i<argc; i++)
    {
        chars_left = CP_EXE_LEN-strlen(darshan_global_job->exe);
        strncat(darshan_global_job->exe, argv[i], chars_left);
        if(i < (argc-1))
        {
            chars_left = CP_EXE_LEN-strlen(darshan_global_job->exe);
            strncat(darshan_global_job->exe, " ", chars_left);
        }
    }

    /* if we don't see any arguments, then use glibc symbol to get
     * program name at least (this happens in fortran)
     */
    if(argc == 0)
    {
        chars_left = CP_EXE_LEN-strlen(darshan_global_job->exe);
        strncat(darshan_global_job->exe, __progname_full, chars_left);
        chars_left = CP_EXE_LEN-strlen(darshan_global_job->exe);
        strncat(darshan_global_job->exe, " <unknown args>", chars_left);
    }

    if(chars_left == 0)
    {
        /* we ran out of room; mark that string was truncated */
        truncate_offset = CP_EXE_LEN - strlen(truncate_string);
        sprintf(&darshan_global_job->exe[truncate_offset], "%s", 
            truncate_string);
    }
}

/* darshan_condense()
 *
 * collapses all file statistics into a single unified set of counters; used
 * when we have opened too many files to track independently
 */
void darshan_condense(void)
{
    struct darshan_file_runtime* base_file;
    struct darshan_file_runtime* iter_file;
    int i;
    int j;

    if(!darshan_global_job)
        return;

    base_file = &darshan_global_job->file_runtime_array[0];

    /* iterate through files */
    for(j=1; j<darshan_global_job->file_count; j++)
    {
        iter_file = &darshan_global_job->file_runtime_array[j];

        /* iterate through records */
        for(i=0; i<CP_NUM_INDICES; i++)
        {
            switch(i)
            {
                /* NOTE: several fields cease to make sense if the records
                 * have been condensed.  Just let them get summed anyway.
                 */
                /* TODO: double check this */

                /* keep up with global maxes in case they are helpful */
                case CP_MAX_BYTE_READ:
                case CP_MAX_BYTE_WRITTEN:
                    CP_MAX(base_file, i, CP_VALUE(iter_file, i));
                    break;

                /* do nothing with these; they are handled in the floating
                 * point loop 
                 */
                case CP_MAX_WRITE_TIME_SIZE:
                case CP_MAX_READ_TIME_SIZE:
                    break;

                /* pick one */
                case CP_DEVICE: