[PATCH] Support residual io counts from io engines

We need this for requeuing support, the network engine makes this
pretty apparent (it's not unusual to see short tranfers there).

Basically we add an xfer_buf and xfer_buflen member to the io_u,
and these are the fields that the io engine MUST use. That allows
fio to increment and reset these appropriately, and simply requeue
the io_u for service of the next part of it.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

diff --git a/engines/libaio.c b/engines/libaio.c
index da43f18ce1571b7c729ceb35fce02ec1aae00e12..9a644f6fe4d07d17be9997a7cc94c882c275cd3e 100644 (file)
--- a/engines/libaio.c
+++ b/engines/libaio.c
@@ -25,9 +25,9 @@ static int fio_libaio_prep(struct thread_data fio_unused *td, struct io_u *io_u)
        struct fio_file *f = io_u->file;
 
        if (io_u->ddir == DDIR_READ)
-               io_prep_pread(&io_u->iocb, f->fd, io_u->buf, io_u->buflen, io_u->offset);
+               io_prep_pread(&io_u->iocb, f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
        else if (io_u->ddir == DDIR_WRITE)
-               io_prep_pwrite(&io_u->iocb, f->fd, io_u->buf, io_u->buflen, io_u->offset);
+               io_prep_pwrite(&io_u->iocb, f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
        else if (io_u->ddir == DDIR_SYNC)
                io_prep_fsync(&io_u->iocb, f->fd);
        else
@@ -87,7 +87,7 @@ static int fio_libaio_queue(struct thread_data *td, struct io_u *io_u)
        } while (1);
 
        if (ret <= 0) {
-               io_u->resid = io_u->buflen;
+               io_u->resid = io_u->xfer_buflen;
                io_u->error = -ret;
                return 1;
        }

diff --git a/engines/mmap.c b/engines/mmap.c
index 20dcfd2264cb3afc981e371c9c89aa2c28c1a447..dc1fd33a2927fc8163140875ee12dbd0e6f22684 100644 (file)
--- a/engines/mmap.c
+++ b/engines/mmap.c
@@ -48,9 +48,9 @@ static int fio_mmapio_queue(struct thread_data *td, struct io_u *io_u)
        struct mmapio_data *sd = td->io_ops->data;
 
        if (io_u->ddir == DDIR_READ)
-               memcpy(io_u->buf, f->mmap + real_off, io_u->buflen);
+               memcpy(io_u->xfer_buf, f->mmap + real_off, io_u->xfer_buflen);
        else if (io_u->ddir == DDIR_WRITE)
-               memcpy(f->mmap + real_off, io_u->buf, io_u->buflen);
+               memcpy(f->mmap + real_off, io_u->xfer_buf, io_u->xfer_buflen);
        else if (io_u->ddir == DDIR_SYNC) {
                if (msync(f->mmap, f->file_size, MS_SYNC))
                        io_u->error = errno;
@@ -60,9 +60,9 @@ static int fio_mmapio_queue(struct thread_data *td, struct io_u *io_u)
         * not really direct, but should drop the pages from the cache
         */
        if (td->odirect && io_u->ddir != DDIR_SYNC) {
-               if (msync(f->mmap + real_off, io_u->buflen, MS_SYNC) < 0)
+               if (msync(f->mmap + real_off, io_u->xfer_buflen, MS_SYNC) < 0)
                        io_u->error = errno;
-               if (madvise(f->mmap + real_off, io_u->buflen,  MADV_DONTNEED) < 0)
+               if (madvise(f->mmap + real_off, io_u->xfer_buflen,  MADV_DONTNEED) < 0)
                        io_u->error = errno;
        }
 

diff --git a/engines/net.c b/engines/net.c
index b1a4cdc21e4092d39b907526a25078cb184de4e1..43026e5f747469c50093587d234732f874e801fe 100644 (file)
--- a/engines/net.c
+++ b/engines/net.c
@@ -77,17 +77,18 @@ static int fio_netio_queue(struct thread_data *td, struct io_u *io_u)
 {
        struct net_data *nd = td->io_ops->data;
        struct fio_file *f = io_u->file;
-       unsigned int ret = 0;
+       int ret = 0;
 
        if (io_u->ddir == DDIR_WRITE)
-               ret = write(f->fd, io_u->buf, io_u->buflen);
+               ret = write(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
        else if (io_u->ddir == DDIR_READ)
-               ret = read(f->fd, io_u->buf, io_u->buflen);
+               ret = read(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
 
-       if (ret != io_u->buflen) {
+       if (ret != (int) io_u->xfer_buflen) {
                if (ret > 0) {
-                       io_u->resid = io_u->buflen - ret;
-                       io_u->error = EIO;
+                       io_u->resid = io_u->xfer_buflen - ret;
+                       io_u->error = 0;
+                       return ret;
                } else
                        io_u->error = errno;
        }

diff --git a/engines/posixaio.c b/engines/posixaio.c
index ef4d78ebe8a58c03465eb33cdc58aa7f90b52d7c..71601fdd31c8df937cc08498c04d308875135d5e 100644 (file)
--- a/engines/posixaio.c
+++ b/engines/posixaio.c
@@ -67,8 +67,8 @@ static int fio_posixaio_prep(struct thread_data fio_unused *td,
        struct fio_file *f = io_u->file;
 
        aiocb->aio_fildes = f->fd;
-       aiocb->aio_buf = io_u->buf;
-       aiocb->aio_nbytes = io_u->buflen;
+       aiocb->aio_buf = io_u->xfer_buf;
+       aiocb->aio_nbytes = io_u->xfer_buflen;
        aiocb->aio_offset = io_u->offset;
 
        io_u->seen = 0;

diff --git a/engines/sg.c b/engines/sg.c
index 3ea1e289604ba6159bf9e3c3f92b3b8d0000ee92..862cd609b45f5870ed2d53e2ef9901feff75ae76 100644 (file)
--- a/engines/sg.c
+++ b/engines/sg.c
@@ -40,8 +40,8 @@ static void sgio_hdr_init(struct sgio_data *sd, struct sg_io_hdr *hdr,
        hdr->usr_ptr = io_u;
 
        if (fs) {
-               hdr->dxferp = io_u->buf;
-               hdr->dxfer_len = io_u->buflen;
+               hdr->dxferp = io_u->xfer_buf;
+               hdr->dxfer_len = io_u->xfer_buflen;
        }
 }
 
@@ -160,7 +160,7 @@ static int fio_sgio_prep(struct thread_data *td, struct io_u *io_u)
        struct sgio_data *sd = td->io_ops->data;
        int nr_blocks, lba;
 
-       if (io_u->buflen & (sd->bs - 1)) {
+       if (io_u->xfer_buflen & (sd->bs - 1)) {
                log_err("read/write not sector aligned\n");
                return EINVAL;
        }
@@ -183,7 +183,7 @@ static int fio_sgio_prep(struct thread_data *td, struct io_u *io_u)
        }
 
        if (hdr->dxfer_direction != SG_DXFER_NONE) {
-               nr_blocks = io_u->buflen / sd->bs;
+               nr_blocks = io_u->xfer_buflen / sd->bs;
                lba = io_u->offset / sd->bs;
                hdr->cmdp[2] = (unsigned char) ((lba >> 24) & 0xff);
                hdr->cmdp[3] = (unsigned char) ((lba >> 16) & 0xff);

diff --git a/engines/splice.c b/engines/splice.c
index fa4a6ee1db21ff192d811f1a95dfacc05de35483..026a82bb053b538ac61f83ed721f8ca4448fb8a1 100644 (file)
--- a/engines/splice.c
+++ b/engines/splice.c
@@ -57,8 +57,8 @@ static int fio_splice_read(struct thread_data *td, struct io_u *io_u)
        void *p;
 
        offset = io_u->offset;
-       buflen = io_u->buflen;
-       p = io_u->buf;
+       buflen = io_u->xfer_buflen;
+       p = io_u->xfer_buf;
        while (buflen) {
                int this_len = buflen;
 
@@ -85,7 +85,7 @@ static int fio_splice_read(struct thread_data *td, struct io_u *io_u)
                }
        }
 
-       return io_u->buflen;
+       return io_u->xfer_buflen;
 }
 
 /*
@@ -97,8 +97,8 @@ static int fio_splice_write(struct thread_data *td, struct io_u *io_u)
        struct spliceio_data *sd = td->io_ops->data;
        struct iovec iov[1] = {
                {
-                       .iov_base = io_u->buf,
-                       .iov_len = io_u->buflen,
+                       .iov_base = io_u->xfer_buf,
+                       .iov_len = io_u->xfer_buflen,
                }
        };
        struct pollfd pfd = { .fd = sd->pipe[1], .events = POLLOUT, };
@@ -126,13 +126,13 @@ static int fio_splice_write(struct thread_data *td, struct io_u *io_u)
                }
        }
 
-       return io_u->buflen;
+       return io_u->xfer_buflen;
 }
 
 static int fio_spliceio_queue(struct thread_data *td, struct io_u *io_u)
 {
        struct spliceio_data *sd = td->io_ops->data;
-       unsigned int ret;
+       int ret;
 
        if (io_u->ddir == DDIR_READ)
                ret = fio_splice_read(td, io_u);
@@ -141,10 +141,11 @@ static int fio_spliceio_queue(struct thread_data *td, struct io_u *io_u)
        else
                ret = fsync(io_u->file->fd);
 
-       if (ret != io_u->buflen) {
+       if (ret != (int) io_u->xfer_buflen) {
                if (ret > 0) {
-                       io_u->resid = io_u->buflen - ret;
-                       io_u->error = ENODATA;
+                       io_u->resid = io_u->xfer_buflen - ret;
+                       io_u->error = 0;
+                       return ret;
                } else
                        io_u->error = errno;
        }

diff --git a/engines/sync.c b/engines/sync.c
index c7ddd4c52956125bd408e4c6f2a0d86f95908041..94dd71049f0cb1cf85f355ba210754150b1eec64 100644 (file)
--- a/engines/sync.c
+++ b/engines/sync.c
@@ -60,19 +60,20 @@ static int fio_syncio_queue(struct thread_data *td, struct io_u *io_u)
 {
        struct syncio_data *sd = td->io_ops->data;
        struct fio_file *f = io_u->file;
-       unsigned int ret;
+       int ret;
 
        if (io_u->ddir == DDIR_READ)
-               ret = read(f->fd, io_u->buf, io_u->buflen);
+               ret = read(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
        else if (io_u->ddir == DDIR_WRITE)
-               ret = write(f->fd, io_u->buf, io_u->buflen);
+               ret = write(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
        else
                ret = fsync(f->fd);
 
-       if (ret != io_u->buflen) {
+       if (ret != (int) io_u->xfer_buflen) {
                if (ret > 0) {
-                       io_u->resid = io_u->buflen - ret;
-                       io_u->error = EIO;
+                       io_u->resid = io_u->xfer_buflen - ret;
+                       io_u->error = 0;
+                       return ret;
                } else
                        io_u->error = errno;
        }

diff --git a/fio.c b/fio.c
index bce13a72306fccc62d84020ee4d57a3a2d8d8722..127314e3590bd1efd79908eeb5363c5580afef86 100644 (file)
--- a/fio.c
+++ b/fio.c
@@ -406,11 +406,22 @@ static void do_io(struct thread_data *td)
 
                memcpy(&s, &io_u->start_time, sizeof(s));
 
+requeue:
                ret = td_io_queue(td, io_u);
                if (ret) {
-                       td_verror(td, io_u->error);
-                       put_io_u(td, io_u);
-                       break;
+                       if (ret > 0 && (io_u->xfer_buflen != io_u->resid) &&
+                           io_u->resid) {
+                               /*
+                                * short read/write. requeue.
+                                */
+                               io_u->xfer_buflen = io_u->resid;
+                               io_u->xfer_buf += ret;
+                               goto requeue;
+                       } else {
+                               td_verror(td, io_u->error);
+                               put_io_u(td, io_u);
+                               break;
+                       }
                }
 
                add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time));

diff --git a/fio.h b/fio.h
index d635e99d0454f47241e199df671be79ddc7758d6..56ed248551a5f6158dedf99eb0e4033f2bf0bf2d 100644 (file)
--- a/fio.h
+++ b/fio.h
@@ -75,6 +75,9 @@ struct io_u {
        unsigned int buflen;
        unsigned long long offset;
 
+       void *xfer_buf;
+       unsigned int xfer_buflen;
+
        unsigned int resid;
        unsigned int error;
 

diff --git a/io_u.c b/io_u.c
index 5e53b81443cda688bdb42163dc2a5af4cc28fe04..da6fe8f3af0a55f2f0a26680f257b78f6996d0de 100644 (file)
--- a/io_u.c
+++ b/io_u.c
@@ -322,6 +322,8 @@ struct io_u *get_io_u(struct thread_data *td, struct fio_file *f)
                return NULL;
        }
 
+       io_u->xfer_buf = io_u->buf;
+       io_u->xfer_buflen = io_u->buflen;
        fio_gettime(&io_u->start_time, NULL);
        return io_u;
 }