[Qemu-devel] [PATCH FOR REVIEW] migration to/from file (v3 aio)

[Uri Lublin <uril@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1034 bytes --]

Hello,

Attached for review is an implementation of migration to/from a file.

migrate-to-file: uses posix-aio-compat and FdMigration
migrate-from-file: uses qemu-fopen directly.

This patch is incomplete as currently it uses polling of aio-completion (we fake 
EAGAIN, and select finds the fd writeable, so file_write will be called again).

As Anthony suggested I plan to implement s->set_fd in FdMigration code, and use 
signal based aio-completion notifications (SIGUSR1).
Anthony mentioned we need only a single aio operation at a time. Although I'm 
not sure what the reasoning for that, I modified the code to support a single 
operation (my previous implementation used multiple aio operations).

Differences from v2:
    use aio writes (posix-aio-compat) instead of select and/or write.
    compile migration-to-file only if AIO is configured.

An alternative (v4) would be to use a single thread and a pipe which is also 
asynchronous, much easier to implement and possibly faster (need to be measured).

Regards,
     Uri.

[-- Attachment #2: 0001-Adding-migration-to-from-file-v3-aio.patch --]
[-- Type: text/x-patch, Size: 9170 bytes --]

>From 34e539ca774d14aa9950fbc4404148d7a5e49101 Mon Sep 17 00:00:00 2001
From: Uri Lublin <uril@redhat.com>
Date: Thu, 12 Feb 2009 11:04:59 +0200
Subject: [PATCH] Adding migration to/from file (v3 - aio)

Migration to file, reuses migration-to-fd.
Migration from file, uses qemu-fopen directly.

The saved state-file should be used only once and removed (or used
with -snapshot, or a the disk-image should be copied), as the
disk image is not saved, only the VM state.

I recommend to stop the VM before migrating its state to a file.
This version supports live migration to file. Can be used later for
VM-checkpointing.

An advantage migration-to-file over savevm/loadvm is that for the latter
a qcow2 is a requirement, while the former works for any image-format.

This version uses posix-aio-compat (which uses worker threads).
Additional memory is allocated to hold the data, so it's not
changed while written to file.

Signed-off-by: Uri Lublin <uril@redhat.com>
---
 Makefile         |    3 +
 migration-file.c |  245 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
 migration.c      |    8 ++
 migration.h      |    6 ++
 4 files changed, 262 insertions(+), 0 deletions(-)
 create mode 100644 migration-file.c

diff --git a/Makefile b/Makefile
index 4f7a55a..60170d1 100644
--- a/Makefile
+++ b/Makefile
@@ -95,6 +95,9 @@ ifdef CONFIG_WIN32
 OBJS+=tap-win32.o
 else
 OBJS+=migration-exec.o
+ifdef CONFIG_AIO
+OBJS+=migration-file.o
+endif
 endif
 
 AUDIO_OBJS = audio.o noaudio.o wavaudio.o mixeng.o
diff --git a/migration-file.c b/migration-file.c
new file mode 100644
index 0000000..6d3dbf5
--- /dev/null
+++ b/migration-file.c
@@ -0,0 +1,245 @@
+/*
+ * QEMU live migration
+ *
+ * Copyright IBM, Corp. 2008
+ *           Red Hat, Inc. 2009
+ *
+ * Authors:
+ *  Anthony Liguori   <aliguori@us.ibm.com>
+ *  Uri Lublin        <uril@redhat.com>
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2.  See
+ * the COPYING file in the top-level directory.
+ *
+ */
+
+#include "qemu-common.h"
+#include "migration.h"
+#include "sysemu.h"
+#include "console.h"
+#include "block.h"
+#include "posix-aio-compat.h"
+#include "buffered_file.h"
+
+//#define DEBUG_MIGRATION_FILE
+
+#ifdef DEBUG_MIGRATION_FILE
+#define dprintf(fmt, ...) \
+    do { printf("migration-file: " fmt, ## __VA_ARGS__); } while (0)
+#else
+#define dprintf(fmt, ...) \
+    do { } while (0)
+#endif
+
+#define MF_AIO_BUF_SIZE 32768
+
+typedef struct FileMigrationState_s {
+    FdMigrationState fds;
+    off_t write_offset;
+    struct qemu_paiocb aiocb;
+    int aio_in_use;
+} FileMigrationState;
+
+
+static int file_errno(FdMigrationState *fds)
+{
+    return errno;
+}
+
+static void mig_file_check_aiocb(FileMigrationState *s)
+{
+    ssize_t ret;
+
+    if (s->aio_in_use) {
+        ret = qemu_paio_return(&s->aiocb);
+        dprintf("mig_file_check_aiocb: aiocb=%p size=%ld ret=%ld\n",
+                &s->aiocb, s->aiocb.aio_nbytes, ret);
+
+        if (ret == -EINPROGRESS)
+            return;
+        if (ret == s->aiocb.aio_nbytes) {
+            s->aio_in_use = 0;
+        }
+        else {
+            dprintf("posix aio write failed, returned %ld expected %ld\n",
+                    ret, s->aiocb.aio_nbytes);
+            if (s->fds.state == MIG_STATE_ACTIVE)
+                s->fds.state = MIG_STATE_ERROR;
+        }
+    }
+}
+
+static struct qemu_paiocb* mig_file_get_aiocb(FileMigrationState *s)
+{
+    struct qemu_paiocb *paiocb = NULL;
+
+    if (s->aio_in_use == 0) {
+        s->aio_in_use = 1;
+        paiocb = &s->aiocb;
+        if (paiocb->aio_buf == NULL)
+            paiocb->aio_buf = qemu_malloc(MF_AIO_BUF_SIZE);
+    }
+
+    return paiocb;
+}
+
+static void free_aio_buf(FileMigrationState *s)
+{
+    struct qemu_paiocb *paiocb = &s->aiocb;
+
+    if (paiocb->aio_buf) {
+        qemu_free(paiocb->aio_buf);
+        paiocb->aio_buf = NULL;
+    }
+}
+
+static int write_async(FileMigrationState *s, const void *buf, size_t size)
+{
+    struct qemu_paiocb *paiocb;
+
+    mig_file_check_aiocb(s);
+
+    paiocb = mig_file_get_aiocb(s);
+    if (paiocb == NULL) {
+        dprintf("write_async: returning EAGAIN\n");
+        return EAGAIN;
+    }
+
+    if (size > MF_AIO_BUF_SIZE) {
+        fprintf(stderr, "mig2file: write_async: size=%ld > aio_max=%d\n",
+                size, MF_AIO_BUF_SIZE);
+        return -EINVAL;
+    }
+
+    memcpy(paiocb->aio_buf, buf, size);
+
+    paiocb->aio_fildes = s->fds.fd;
+    paiocb->aio_nbytes = size;
+    paiocb->ev_signo = 0; /* No notification is needed */
+    paiocb->aio_offset = s->write_offset;
+
+    s->write_offset += size;
+
+    return qemu_paio_write(paiocb);
+}
+
+static int file_write(FdMigrationState *fds, const void * buf, size_t size)
+{
+    FileMigrationState *s = (FileMigrationState*)fds;
+    int ret = 0;
+    ssize_t offset = 0, len = MF_AIO_BUF_SIZE;
+
+    while (offset < size) {
+        if (size - offset < len)
+            len = size - offset;
+        ret = write_async(s, buf + offset, len);
+        if (ret == EAGAIN) {
+            if (offset == 0) {
+                errno = EAGAIN;
+                return -1;
+            }
+            break;
+        }
+        if (ret < 0)
+            return ret;
+        offset += len;
+    }
+
+    dprintf("file_write: size=%ld ret=%d returning %ld\n", size, ret, offset);
+
+    return offset;
+}
+
+static int file_close(FdMigrationState *fds)
+{
+    FileMigrationState *s = (FileMigrationState*)fds;
+    int i;
+
+    for (i=0; i<30; i++) {
+        dprintf("file_close: still in_use %d/%d\n", i, 30);
+        mig_file_check_aiocb(s);
+        if (s->aio_in_use == 0)
+            break;
+        sleep(1);
+    }
+    close(fds->fd);
+
+    free_aio_buf(s);
+
+    return 0;
+}
+
+MigrationState *file_start_outgoing_migration(const char *filename,
+                                              int64_t bandwidth_limit,
+                                              int async)
+{
+    FdMigrationState *fds;
+    FileMigrationState *s;
+    int fd;
+
+    s = qemu_mallocz(sizeof(*s));
+
+    fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0600);
+    if (fd < 0) {
+        perror("file_migration: failed to open filename");
+        term_printf("file_migration: failed to open filename %s\n", filename);
+        goto err;
+    }
+
+    fds = &s->fds;
+
+    fds->fd = fd;
+    fds->close = file_close;
+    fds->get_error = file_errno;
+    fds->write = file_write;
+    fds->mig_state.cancel = migrate_fd_cancel;
+    fds->mig_state.get_status = migrate_fd_get_status;
+    fds->mig_state.release = migrate_fd_release;
+
+    fds->state = MIG_STATE_ACTIVE;
+    fds->detach = !async;
+    fds->bandwidth_limit = bandwidth_limit;
+
+
+    if (fds->detach == 1) {
+        monitor_suspend();
+        fds->detach = 2;
+    }
+
+    migrate_fd_connect(fds);
+    return &fds->mig_state;
+
+err:
+    qemu_free(s);
+    return NULL;
+}
+
+int file_start_incoming_migration(const char *filename)
+{
+    int ret;
+    QEMUFile *f;
+
+    dprintf("Starting incoming file migration from '%s'\n", filename);
+    f = qemu_fopen(filename, "rb");
+    if(f == NULL) {
+        perror("failed to open file");
+        term_printf("failed to open file %s\n", filename);
+        return -errno;
+    }
+
+    vm_stop(0); /* just in case */
+    ret = qemu_loadvm_state(f);
+    if (ret < 0) {
+        fprintf(stderr, "in_file_mig: load of migration failed\n");
+        goto err;
+    }
+    qemu_announce_self();
+    dprintf("successfully loaded vm state\n");
+    vm_start();
+    qemu_fclose(f);
+    return 0;
+
+err:
+    qemu_fclose(f);
+    return -errno;
+}
diff --git a/migration.c b/migration.c
index 0ef777a..234dcf6 100644
--- a/migration.c
+++ b/migration.c
@@ -43,6 +43,10 @@ void qemu_start_incoming_migration(const char *uri)
 #if !defined(WIN32)
     else if (strstart(uri, "exec:", &p))
         exec_start_incoming_migration(p);
+#ifdef CONFIG_AIO
+    else if (strstart(uri, "file:", &p))
+        file_start_incoming_migration(p);
+#endif
 #endif
     else
         fprintf(stderr, "unknown migration protocol: %s\n", uri);
@@ -58,6 +62,10 @@ void do_migrate(int detach, const char *uri)
 #if !defined(WIN32)
     else if (strstart(uri, "exec:", &p))
         s = exec_start_outgoing_migration(p, max_throttle, detach);
+#ifdef CONFIG_AIO
+    else if (strstart(uri, "file:", &p))
+        s = file_start_outgoing_migration(p, max_throttle, detach);
+#endif
 #endif
     else
         term_printf("unknown migration protocol: %s\n", uri);
diff --git a/migration.h b/migration.h
index d9771ad..ae78792 100644
--- a/migration.h
+++ b/migration.h
@@ -67,6 +67,12 @@ MigrationState *tcp_start_outgoing_migration(const char *host_port,
 					     int64_t bandwidth_limit,
 					     int detach);
 
+
+int file_start_incoming_migration(const char *filename);
+MigrationState *file_start_outgoing_migration(const char *filename,
+                                              int64_t bandwidth_limit,
+                                              int detach);
+
 void migrate_fd_error(FdMigrationState *s);
 
 void migrate_fd_cleanup(FdMigrationState *s);
-- 
1.6.0.6