os_rw.c

/*-
 * See the file LICENSE for redistribution information.
 *
 * Copyright (c) 1997-2005
 *      Sleepycat Software.  All rights reserved.
 *
 * $Id: os_rw.c,v 12.4 2005/08/10 15:47:28 bostic Exp $
 */

#include "db_config.h"

#ifndef NO_SYSTEM_INCLUDES
#include <sys/types.h>

#include <string.h>
#endif

#include "db_int.h"

#ifdef HAVE_FILESYSTEM_NOTZERO
static int __os_zerofill __P((DB_ENV *, DB_FH *));
#endif
static int __os_physwrite __P((DB_ENV *, DB_FH *, void *, size_t, size_t *));

/*
 * __os_io --
 *      Do an I/O.
 */
int
__os_io(dbenv, op, fhp, pgno, pagesize, buf, niop)
        DB_ENV *dbenv;
        int op;
        DB_FH *fhp;
        db_pgno_t pgno;
        u_int32_t pagesize;
        u_int8_t *buf;
        size_t *niop;
{
        int ret;

        if (__os_is_winnt()) {
                ULONG64 off = (ULONG64)pagesize * pgno;
                OVERLAPPED over;
                DWORD nbytes;
                over.Offset = (DWORD)(off & 0xffffffff);
                over.OffsetHigh = (DWORD)(off >> 32);
                over.hEvent = 0; /* we don't want asynchronous notifications */

                switch (op) {
                case DB_IO_READ:
                        if (DB_GLOBAL(j_read) != NULL)
                                goto slow;
                        if (!ReadFile(fhp->handle,
                            buf, (DWORD)pagesize, &nbytes, &over))
                                goto slow;
                        break;
                case DB_IO_WRITE:
                        if (DB_GLOBAL(j_write) != NULL)
                                goto slow;
#ifdef HAVE_FILESYSTEM_NOTZERO
                        if (__os_fs_notzero())
                                goto slow;
#endif
                        if (!WriteFile(fhp->handle,
                            buf, (DWORD)pagesize, &nbytes, &over))
                                goto slow;
                        break;
                }
                if (nbytes == pagesize) {
                        *niop = (size_t)nbytes;
                        return (0);
                }
        }

slow:   MUTEX_LOCK(dbenv, fhp->mtx_fh);

        if ((ret = __os_seek(dbenv, fhp,
            pagesize, pgno, 0, 0, DB_OS_SEEK_SET)) != 0)
                goto err;

        switch (op) {
        case DB_IO_READ:
                ret = __os_read(dbenv, fhp, buf, pagesize, niop);
                break;
        case DB_IO_WRITE:
                ret = __os_write(dbenv, fhp, buf, pagesize, niop);
                break;
        }

err:    MUTEX_UNLOCK(dbenv, fhp->mtx_fh);

        return (ret);
}

/*
 * __os_read --
 *      Read from a file handle.
 */
int
__os_read(dbenv, fhp, addr, len, nrp)
        DB_ENV *dbenv;
        DB_FH *fhp;
        void *addr;
        size_t len;
        size_t *nrp;
{
        size_t offset, nr;
        DWORD count;
        int ret;
        u_int8_t *taddr;

        ret = 0;

        if (DB_GLOBAL(j_read) != NULL) {
                *nrp = len;
                if (DB_GLOBAL(j_read)(fhp->fd, addr, len) != (ssize_t)len) {
                        ret = __os_get_errno();
                        __db_err(dbenv, "read: %#lx, %lu: %s",
                            P_TO_ULONG(addr), (u_long)len, strerror(ret));
                }
                return (ret);
        }

        ret = 0;
        for (taddr = addr,
            offset = 0; offset < len; taddr += nr, offset += nr) {
                RETRY_CHK((!ReadFile(fhp->handle,
                    taddr, (DWORD)(len - offset), &count, NULL)), ret);
                if (count == 0 || ret != 0)
                        break;
                nr = (size_t)count;
        }
        *nrp = taddr - (u_int8_t *)addr;
        if (ret != 0)
                __db_err(dbenv, "read: 0x%lx, %lu: %s",
                    P_TO_ULONG(taddr), (u_long)len - offset, strerror(ret));
        return (ret);
}

/*
 * __os_write --
 *      Write to a file handle.
 */
int
__os_write(dbenv, fhp, addr, len, nwp)
        DB_ENV *dbenv;
        DB_FH *fhp;
        void *addr;
        size_t len;
        size_t *nwp;
{
        int ret;

#ifdef HAVE_FILESYSTEM_NOTZERO
        /* Zero-fill as necessary. */
        if (__os_fs_notzero() && (ret = __os_zerofill(dbenv, fhp)) != 0)
                return (ret);
#endif
        return (__os_physwrite(dbenv, fhp, addr, len, nwp));
}

/*
 * __os_physwrite --
 *      Physical write to a file handle.
 */
static int
__os_physwrite(dbenv, fhp, addr, len, nwp)
        DB_ENV *dbenv;
        DB_FH *fhp;
        void *addr;
        size_t len;
        size_t *nwp;
{
        size_t offset, nw;
        DWORD count;
        int ret;
        u_int8_t *taddr;

        /*
         * Make a last "panic" check.  Imagine a thread of control running in
         * Berkeley DB, going to sleep.  Another thread of control decides to
         * run recovery because the environment is broken.  The first thing
         * recovery does is panic the existing environment, but we only check
         * the panic flag when crossing the public API.  If the sleeping thread
         * wakes up and writes something, we could have two threads of control
         * writing the log files at the same time.  So, before writing, make a
         * last panic check.  Obviously, there's still a window, but it's very,
         * very small.
         */
        PANIC_CHECK(dbenv);

        if (DB_GLOBAL(j_write) != NULL) {
                *nwp = len;
                if (DB_GLOBAL(j_write)(fhp->fd, addr, len) != (ssize_t)len) {
                        ret = __os_get_errno();
                        __db_err(dbenv, "write: %#lx, %lu: %s",
                            P_TO_ULONG(addr), (u_long)len, strerror(ret));
                }
                return (ret);
        }

        ret = 0;
        for (taddr = addr,
            offset = 0; offset < len; taddr += nw, offset += nw) {
                RETRY_CHK((!WriteFile(fhp->handle,
                    taddr, (DWORD)(len - offset), &count, NULL)), ret);
                if (ret != 0)
                        break;
                nw = (size_t)count;
        }
        *nwp = len;
        if (ret != 0)
                __db_err(dbenv, "write: %#lx, %lu: %s",
                    P_TO_ULONG(taddr), (u_long)len - offset, strerror(ret));
        return (ret);
}

#ifdef HAVE_FILESYSTEM_NOTZERO
/*
 * __os_zerofill --
 *      Zero out bytes in the file.
 *
 *      Pages allocated by writing pages past end-of-file are not zeroed,
 *      on some systems.  Recovery could theoretically be fooled by a page
 *      showing up that contained garbage.  In order to avoid this, we
 *      have to write the pages out to disk, and flush them.  The reason
 *      for the flush is because if we don't sync, the allocation of another
 *      page subsequent to this one might reach the disk first, and if we
 *      crashed at the right moment, leave us with this page as the one
 *      allocated by writing a page past it in the file.
 */
static int
__os_zerofill(dbenv, fhp)
        DB_ENV *dbenv;
        DB_FH *fhp;
{
        unsigned __int64 stat_offset, write_offset;
        size_t blen, nw;
        u_int32_t bytes, mbytes;
        int group_sync, need_free, ret;
        u_int8_t buf[8 * 1024], *bp;

        /* Calculate the byte offset of the next write. */
        write_offset = (unsigned __int64)fhp->pgno * fhp->pgsize + fhp->offset;

        /* Stat the file. */
        if ((ret = __os_ioinfo(dbenv, NULL, fhp, &mbytes, &bytes, NULL)) != 0)
                return (ret);
        stat_offset = (unsigned __int64)mbytes * MEGABYTE + bytes;

        /* Check if the file is large enough. */
        if (stat_offset >= write_offset)
                return (0);

        /* Get a large buffer if we're writing lots of data. */
#undef  ZF_LARGE_WRITE
#define ZF_LARGE_WRITE  (64 * 1024)
        if (write_offset - stat_offset > ZF_LARGE_WRITE) {
                if ((ret = __os_calloc(dbenv, 1, ZF_LARGE_WRITE, &bp)) != 0)
                            return (ret);
                blen = ZF_LARGE_WRITE;
                need_free = 1;
        } else {
                bp = buf;
                blen = sizeof(buf);
                need_free = 0;
                memset(buf, 0, sizeof(buf));
        }

        /* Seek to the current end of the file. */
        if ((ret = __os_seek(
            dbenv, fhp, MEGABYTE, mbytes, bytes, 0, DB_OS_SEEK_SET)) != 0)
                goto err;

        /*
         * Hash is the only access method that allocates groups of pages.  Hash
         * uses the existence of the last page in a group to signify the entire
         * group is OK; so, write all the pages but the last one in the group,
         * flush them to disk, then write the last one to disk and flush it.
         */
        for (group_sync = 0; stat_offset < write_offset; group_sync = 1) {
                if (write_offset - stat_offset <= blen) {
                        blen = (size_t)(write_offset - stat_offset);
                        if (group_sync && (ret = __os_fsync(dbenv, fhp)) != 0)
                                goto err;
                }
                if ((ret = __os_physwrite(dbenv, fhp, bp, blen, &nw)) != 0)
                        goto err;
                stat_offset += blen;
        }
        if ((ret = __os_fsync(dbenv, fhp)) != 0)
                goto err;

        /* Seek back to where we started. */
        mbytes = (u_int32_t)(write_offset / MEGABYTE);
        bytes = (u_int32_t)(write_offset % MEGABYTE);
        ret = __os_seek(dbenv, fhp, MEGABYTE, mbytes, bytes, 0, DB_OS_SEEK_SET);

err:    if (need_free)
                __os_free(dbenv, bp);
        return (ret);
}
#endif

---