[virtio-9p] Use preadv/pwritev instead of readv/writev

readv & writev, read & write respectively from the current offset
of the file & hence their use has to be preceeded by a call to lseek.
preadv/writev can be used instead, as they take the offset as an argument.
This saves one system call( lseek ).
In case preadv is not supported, it is implemented by an lseek
followed by a readv. Depending upon the configuration of QEMU, the
appropriate read & write methods are selected. This patch also fixes the
zero byte read/write bug & obviates the need to apply a fix for that bug separately.

Signed-off-by: M. Mohan Kumar <mohan@in.ibm.com>
Signed-off-by: Sanchit Garg <sancgarg@linux.vnet.ibm.com>
Signed-off-by: Venkateswararao Jujjuri <jvrao@linux.vnet.ibm.com>

3 files changed, 67 insertions, 94 deletions

diff --git a/hw/file-op-9p.h b/hw/file-op-9p.h
index bf867b924c..21d60b5855 100644
--- a/hw/file-op-9p.h
+++ b/hw/file-op-9p.h
@@ -80,9 +80,8 @@ typedef struct FileOperations
     off_t (*telldir)(FsContext *, DIR *);
     struct dirent *(*readdir)(FsContext *, DIR *);
     void (*seekdir)(FsContext *, DIR *, off_t);
-    ssize_t (*readv)(FsContext *, int, const struct iovec *, int);
-    ssize_t (*writev)(FsContext *, int, const struct iovec *, int);
-    off_t (*lseek)(FsContext *, int, off_t, int);
+    ssize_t (*preadv)(FsContext *, int, const struct iovec *, int, off_t);
+    ssize_t (*pwritev)(FsContext *, int, const struct iovec *, int, off_t);
     int (*mkdir)(FsContext *, const char *, FsCred *);
     int (*fstat)(FsContext *, int, struct stat *);
     int (*rename)(FsContext *, const char *, const char *);
diff --git a/hw/virtio-9p-local.c b/hw/virtio-9p-local.c
index ee630334b8..0d520201b4 100644
--- a/hw/virtio-9p-local.c
+++ b/hw/virtio-9p-local.c
@@ -168,21 +168,34 @@ static void local_seekdir(FsContext *ctx, DIR *dir, off_t off)
     return seekdir(dir, off);
 }
 
-static ssize_t local_readv(FsContext *ctx, int fd, const struct iovec *iov,
-                            int iovcnt)
+static ssize_t local_preadv(FsContext *ctx, int fd, const struct iovec *iov,
+                            int iovcnt, off_t offset)
 {
-    return readv(fd, iov, iovcnt);
-}
-
-static off_t local_lseek(FsContext *ctx, int fd, off_t offset, int whence)
-{
-    return lseek(fd, offset, whence);
+#ifdef CONFIG_PREADV
+    return preadv(fd, iov, iovcnt, offset);
+#else
+    int err = lseek(fd, offset, SEEK_SET);
+    if (err == -1) {
+        return err;
+    } else {
+        return readv(fd, iov, iovcnt);
+    }
+#endif
 }
 
-static ssize_t local_writev(FsContext *ctx, int fd, const struct iovec *iov,
-                            int iovcnt)
+static ssize_t local_pwritev(FsContext *ctx, int fd, const struct iovec *iov,
+                            int iovcnt, off_t offset)
 {
-    return writev(fd, iov, iovcnt);
+#ifdef CONFIG_PREADV
+    return pwritev(fd, iov, iovcnt, offset);
+#else
+    int err = lseek(fd, offset, SEEK_SET);
+    if (err == -1) {
+        return err;
+    } else {
+        return writev(fd, iov, iovcnt);
+    }
+#endif
 }
 
 static int local_chmod(FsContext *fs_ctx, const char *path, FsCred *credp)
@@ -523,9 +536,8 @@ FileOperations local_ops = {
     .telldir = local_telldir,
     .readdir = local_readdir,
     .seekdir = local_seekdir,
-    .readv = local_readv,
-    .lseek = local_lseek,
-    .writev = local_writev,
+    .preadv = local_preadv,
+    .pwritev = local_pwritev,
     .chmod = local_chmod,
     .mknod = local_mknod,
     .mkdir = local_mkdir,
diff --git a/hw/virtio-9p.c b/hw/virtio-9p.c
index 867fcfa095..0e8772257b 100644
--- a/hw/virtio-9p.c
+++ b/hw/virtio-9p.c
@@ -135,21 +135,16 @@ static void v9fs_do_seekdir(V9fsState *s, DIR *dir, off_t off)
     return s->ops->seekdir(&s->ctx, dir, off);
 }
 
-static int v9fs_do_readv(V9fsState *s, int fd, const struct iovec *iov,
-                            int iovcnt)
+static int v9fs_do_preadv(V9fsState *s, int fd, const struct iovec *iov,
+                            int iovcnt, int64_t offset)
 {
-    return s->ops->readv(&s->ctx, fd, iov, iovcnt);
+    return s->ops->preadv(&s->ctx, fd, iov, iovcnt, offset);
 }
 
-static off_t v9fs_do_lseek(V9fsState *s, int fd, off_t offset, int whence)
+static int v9fs_do_pwritev(V9fsState *s, int fd, const struct iovec *iov,
+                       int iovcnt, int64_t offset)
 {
-    return s->ops->lseek(&s->ctx, fd, offset, whence);
-}
-
-static int v9fs_do_writev(V9fsState *s, int fd, const struct iovec *iov,
-                       int iovcnt)
-{
-    return s->ops->writev(&s->ctx, fd, iov, iovcnt);
+    return s->ops->pwritev(&s->ctx, fd, iov, iovcnt, offset);
 }
 
 static int v9fs_do_chmod(V9fsState *s, V9fsString *path, mode_t mode)
@@ -1975,7 +1970,7 @@ static void v9fs_read_post_rewinddir(V9fsState *s, V9fsReadState *vs,
     return;
 }
 
-static void v9fs_read_post_readv(V9fsState *s, V9fsReadState *vs, ssize_t err)
+static void v9fs_read_post_preadv(V9fsState *s, V9fsReadState *vs, ssize_t err)
 {
     if (err  < 0) {
         /* IO error return the error */
@@ -1989,12 +1984,16 @@ static void v9fs_read_post_readv(V9fsState *s, V9fsReadState *vs, ssize_t err)
             if (0) {
                 print_sg(vs->sg, vs->cnt);
             }
-            vs->len = v9fs_do_readv(s, vs->fidp->fs.fd, vs->sg, vs->cnt);
+            vs->len = v9fs_do_preadv(s, vs->fidp->fs.fd, vs->sg, vs->cnt,
+                      vs->off);
+            if (vs->len > 0) {
+                vs->off += vs->len;
+            }
         } while (vs->len == -1 && errno == EINTR);
         if (vs->len == -1) {
             err  = -errno;
         }
-        v9fs_read_post_readv(s, vs, err);
+        v9fs_read_post_preadv(s, vs, err);
         return;
     }
     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->total);
@@ -2006,32 +2005,6 @@ out:
     qemu_free(vs);
 }
 
-static void v9fs_read_post_lseek(V9fsState *s, V9fsReadState *vs, ssize_t err)
-{
-    if (err == -1) {
-        err = -errno;
-        goto out;
-    }
-    vs->sg = cap_sg(vs->sg, vs->count, &vs->cnt);
-
-    if (vs->total < vs->count) {
-        do {
-            if (0) {
-                print_sg(vs->sg, vs->cnt);
-            }
-            vs->len = v9fs_do_readv(s, vs->fidp->fs.fd, vs->sg, vs->cnt);
-        } while (vs->len == -1 && errno == EINTR);
-        if (vs->len == -1) {
-            err  = -errno;
-        }
-        v9fs_read_post_readv(s, vs, err);
-        return;
-    }
-out:
-    complete_pdu(s, vs->pdu, err);
-    qemu_free(vs);
-}
-
 static void v9fs_xattr_read(V9fsState *s, V9fsReadState *vs)
 {
     ssize_t err = 0;
@@ -2089,8 +2062,16 @@ static void v9fs_read(V9fsState *s, V9fsPDU *pdu)
     } else if (vs->fidp->fid_type == P9_FID_FILE) {
         vs->sg = vs->iov;
         pdu_marshal(vs->pdu, vs->offset + 4, "v", vs->sg, &vs->cnt);
-        err = v9fs_do_lseek(s, vs->fidp->fs.fd, vs->off, SEEK_SET);
-        v9fs_read_post_lseek(s, vs, err);
+        vs->sg = cap_sg(vs->sg, vs->count, &vs->cnt);
+        if (vs->total <= vs->count) {
+            vs->len = v9fs_do_preadv(s, vs->fidp->fs.fd, vs->sg, vs->cnt,
+                                    vs->off);
+            if (vs->len > 0) {
+                vs->off += vs->len;
+            }
+            err = vs->len;
+            v9fs_read_post_preadv(s, vs, err);
+        }
         return;
     } else if (vs->fidp->fid_type == P9_FID_XATTR) {
         v9fs_xattr_read(s, vs);
@@ -2224,7 +2205,7 @@ out:
     return;
 }
 
-static void v9fs_write_post_writev(V9fsState *s, V9fsWriteState *vs,
+static void v9fs_write_post_pwritev(V9fsState *s, V9fsWriteState *vs,
                                    ssize_t err)
 {
     if (err  < 0) {
@@ -2239,49 +2220,25 @@ static void v9fs_write_post_writev(V9fsState *s, V9fsWriteState *vs,
             if (0) {
                 print_sg(vs->sg, vs->cnt);
             }
-            vs->len =  v9fs_do_writev(s, vs->fidp->fs.fd, vs->sg, vs->cnt);
+            vs->len = v9fs_do_pwritev(s, vs->fidp->fs.fd, vs->sg, vs->cnt,
+                      vs->off);
+            if (vs->len > 0) {
+                vs->off += vs->len;
+            }
         } while (vs->len == -1 && errno == EINTR);
         if (vs->len == -1) {
             err  = -errno;
         }
-        v9fs_write_post_writev(s, vs, err);
+        v9fs_write_post_pwritev(s, vs, err);
         return;
     }
     vs->offset += pdu_marshal(vs->pdu, vs->offset, "d", vs->total);
-
     err = vs->offset;
 out:
     complete_pdu(s, vs->pdu, err);
     qemu_free(vs);
 }
 
-static void v9fs_write_post_lseek(V9fsState *s, V9fsWriteState *vs, ssize_t err)
-{
-    if (err == -1) {
-        err = -errno;
-        goto out;
-    }
-    vs->sg = cap_sg(vs->sg, vs->count, &vs->cnt);
-
-    if (vs->total < vs->count) {
-        do {
-            if (0) {
-                print_sg(vs->sg, vs->cnt);
-            }
-            vs->len = v9fs_do_writev(s, vs->fidp->fs.fd, vs->sg, vs->cnt);
-        } while (vs->len == -1 && errno == EINTR);
-        if (vs->len == -1) {
-            err  = -errno;
-        }
-        v9fs_write_post_writev(s, vs, err);
-        return;
-    }
-
-out:
-    complete_pdu(s, vs->pdu, err);
-    qemu_free(vs);
-}
-
 static void v9fs_xattr_write(V9fsState *s, V9fsWriteState *vs)
 {
     int i, to_copy;
@@ -2362,11 +2319,16 @@ static void v9fs_write(V9fsState *s, V9fsPDU *pdu)
         err = -EINVAL;
         goto out;
     }
-    err = v9fs_do_lseek(s, vs->fidp->fs.fd, vs->off, SEEK_SET);
-
-    v9fs_write_post_lseek(s, vs, err);
+    vs->sg = cap_sg(vs->sg, vs->count, &vs->cnt);
+    if (vs->total <= vs->count) {
+        vs->len = v9fs_do_pwritev(s, vs->fidp->fs.fd, vs->sg, vs->cnt, vs->off);
+        if (vs->len > 0) {
+            vs->off += vs->len;
+        }
+        err = vs->len;
+        v9fs_write_post_pwritev(s, vs, err);
+    }
     return;
-
 out:
     complete_pdu(s, vs->pdu, err);
     qemu_free(vs);