Re: libvirt, libxl and QDISKs

[Marek Marczykowski <marmarek@invisiblethingslab.com>]

[-- Attachment #1.1.1: Type: text/plain, Size: 3371 bytes --]

On 26.04.2013 12:10, Stefano Stabellini wrote:
> On Fri, 26 Apr 2013, Jim Fehlig wrote:
>>     if (l_disk->driverName) {
>>         if (STREQ(l_disk->driverName, "tap") ||
>>             STREQ(l_disk->driverName, "tap2")) {
>>             switch (l_disk->format) {
>>             case VIR_STORAGE_FILE_QCOW:
>>                 x_disk->format = LIBXL_DISK_FORMAT_QCOW;
>>                 x_disk->backend = LIBXL_DISK_BACKEND_QDISK;
>>                 break;
>>             case VIR_STORAGE_FILE_QCOW2:
>>                 x_disk->format = LIBXL_DISK_FORMAT_QCOW2;
>>                 x_disk->backend = LIBXL_DISK_BACKEND_QDISK;
>>                 break;
>>             case VIR_STORAGE_FILE_VHD:
>>                 x_disk->format = LIBXL_DISK_FORMAT_VHD;
>>                 x_disk->backend = LIBXL_DISK_BACKEND_TAP;
>>                 break;
>>             case VIR_STORAGE_FILE_NONE:
>>                 /* No subtype specified, default to raw/tap */
>>             case VIR_STORAGE_FILE_RAW:
>>                 x_disk->format = LIBXL_DISK_FORMAT_RAW;
>>                 x_disk->backend = LIBXL_DISK_BACKEND_TAP;
>>                 break;
>>             default:
>>                 virReportError(VIR_ERR_INTERNAL_ERROR,
>>                                _("libxenlight does not support disk
>> driver %s"),
>>                               
>> virStorageFileFormatTypeToString(l_disk->format));
>>                 return -1;
>>             }
>>         } else if (STREQ(l_disk->driverName, "file")) {
>>             x_disk->format = LIBXL_DISK_FORMAT_RAW;
>>             x_disk->backend = LIBXL_DISK_BACKEND_TAP;
>>         } else if (STREQ(l_disk->driverName, "phy")) {
>>             x_disk->format = LIBXL_DISK_FORMAT_RAW;
>>             x_disk->backend = LIBXL_DISK_BACKEND_PHY;
>>         } else {
>>             virReportError(VIR_ERR_INTERNAL_ERROR,
>>                            _("libxenlight does not support disk driver %s"),
>>                            l_disk->driverName);
>>             return -1;
>>         }
>>     } else {
>>         /*
>>          * If driverName is not specified, default to raw as per
>>          * xl-disk-configuration.txt in the xen documentation and let
>>          * libxl pick a suitable backend.
>>          */
>>         x_disk->format = LIBXL_DISK_FORMAT_RAW;
>>         x_disk->backend = LIBXL_DISK_BACKEND_UNKNOWN;
>>     }
> 
> It looks like the defaults are the same of libxl.
> 
> However the mapping of RAW to TAP (libxl does the same) has always been
> a bit dubious to me: now that upstream QEMU is used with HVM guests too
> by libxl, there is no reason to use blktap over QEMU for raw files any
> more.

What about old good loop+phy based backend for file disk images? I don't want
whole qemu in dom0 for PV domains, only for handling simple disk backend.
Additionally sparse images + loop + phy + mount -o discard in domU makes the
images "auto shrinking". Don't know if qemu is able to do this.

Attached patch, which I currently use for that. If it is close to something
that would be accepted, I will send it in new thread.

BTW I don't want qemu in dom0 for HVM also, but this is another story
(modified stubdom etc) and probably not acceptable in vanilla xen.

-- 
Best Regards / Pozdrawiam,
Marek Marczykowski
Invisible Things Lab

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1.1.2: 0001-libxl-allow-PHY-backend-for-files-allocate-loop-devi.patch --]
[-- Type: text/x-patch; name="0001-libxl-allow-PHY-backend-for-files-allocate-loop-devi.patch", Size: 8660 bytes --]

From cbd5831fd5a6a00429957acb6b1d58d7c645f219 Mon Sep 17 00:00:00 2001
From: Marek Marczykowski <marmarek@invisiblethingslab.com>
Date: Tue, 16 Apr 2013 22:20:25 +0200
Subject: [PATCH] libxl: allow PHY backend for files - allocate loop device
Organization: Invisible Things Lab
Cc: Marek Marczykowski <marmarek@invisiblethingslab.com>

This is implementation of /etc/xen/scripts/block behaviour, but without
calling the script, which should be much faster.

Signed-off-by: Marek Marczykowski <marmarek@invisiblethingslab.com>
---
 tools/libxl/libxl.c          |  13 ++++-
 tools/libxl/libxl_device.c   |   8 +++
 tools/libxl/libxl_internal.h |   9 +++
 tools/libxl/libxl_linux.c    | 129 ++++++++++++++++++++++++++++++++++++++++++-
 tools/libxl/libxl_netbsd.c   |  11 ++++
 5 files changed, 168 insertions(+), 2 deletions(-)

diff --git a/tools/libxl/libxl.c b/tools/libxl/libxl.c
index a813a88..4af5591 100644
--- a/tools/libxl/libxl.c
+++ b/tools/libxl/libxl.c
@@ -2086,7 +2086,18 @@ static void device_disk_add(libxl__egc *egc, uint32_t domid,
                  */
                 if (!disk->script) {
                     int major, minor;
-                    libxl__device_physdisk_major_minor(dev, &major, &minor);
+                    char *node;
+
+                    node = libxl__loopdev_setup(gc, dev);
+                    if (!node) {
+                        /* error already reported by libxl__setup_loopdev */
+                        goto out;
+                    }
+                    if (node != dev) {
+                        flexarray_append_pair(back, "node", node);
+                    }
+
+                    libxl__device_physdisk_major_minor(node, &major, &minor);
                     flexarray_append_pair(back, "physical-device",
                             libxl__sprintf(gc, "%x:%x", major, minor));
                 }
diff --git a/tools/libxl/libxl_device.c b/tools/libxl/libxl_device.c
index d622826..936961e 100644
--- a/tools/libxl/libxl_device.c
+++ b/tools/libxl/libxl_device.c
@@ -883,6 +883,7 @@ static void device_hotplug(libxl__egc *egc, libxl__ao_device *aodev)
 {
     STATE_AO_GC(aodev->ao);
     char *be_path = libxl__device_backend_path(gc, aodev->dev);
+    char *node;
     char **args = NULL, **env = NULL;
     int rc = 0;
     int hotplug;
@@ -906,6 +907,13 @@ static void device_hotplug(libxl__egc *egc, libxl__ao_device *aodev)
     switch (hotplug) {
     case 0:
         /* no hotplug script to execute */
+        /* check if we have loop device allocated - if so, try to release it */
+        node = libxl__xs_read(gc, XBT_NULL, GCSPRINTF("%s/node", be_path));
+        if (node) {
+            /* ignore errors: if device is already released - no problem, if
+             * EBUSY - can be shared */
+            libxl__loopdev_cleanup(gc, node);
+        }
         goto out;
     case 1:
         /* execute hotplug script */
diff --git a/tools/libxl/libxl_internal.h b/tools/libxl/libxl_internal.h
index 3ba3a21..573e3d1 100644
--- a/tools/libxl/libxl_internal.h
+++ b/tools/libxl/libxl_internal.h
@@ -1050,6 +1050,15 @@ static inline void libxl__domaindeathcheck_stop(libxl__gc *gc,
  */
 _hidden int libxl__try_phy_backend(mode_t st_mode);
 
+/*
+ * libxl__setup_loopdev - Setup loop device if needed.
+ *
+ * Return its path on success or NULL in case of failure. If no loop device is
+ * needed, return original filename.
+ */
+_hidden char *libxl__loopdev_setup(libxl__gc *gc, char *filename);
+
+_hidden void libxl__loopdev_cleanup(libxl__gc *gc, char *devname);
 
 _hidden char *libxl__devid_to_localdev(libxl__gc *gc, int devid);
 
diff --git a/tools/libxl/libxl_linux.c b/tools/libxl/libxl_linux.c
index 37815eb..115332a 100644
--- a/tools/libxl/libxl_linux.c
+++ b/tools/libxl/libxl_linux.c
@@ -16,16 +16,143 @@
 #include "libxl_osdeps.h" /* must come before any other headers */
 
 #include "libxl_internal.h"
+#include <linux/loop.h>
  
 int libxl__try_phy_backend(mode_t st_mode)
 {
-    if (!S_ISBLK(st_mode)) {
+    if (!S_ISBLK(st_mode) && !S_ISREG(st_mode)) {
         return 0;
     }
 
     return 1;
 }
 
+char *libxl__loopdev_setup(libxl__gc *gc, char *filename) {
+    struct stat st;
+    DIR *sysblock;
+    int dir_fd;
+    int control_fd = -1;
+    int loop_fd = -1;
+    int file_fd = -1;
+    int devnr;
+    struct dirent *d;
+    struct loop_info64  info;
+    char *loop_name = NULL;
+    char *ret = NULL;
+
+    if (stat(filename, &st) < 0) {
+        LOG(ERROR, "unable to stat %s", filename);
+        return NULL;
+    }
+
+    /* do not do anything if filename already is block device */
+    if (S_ISBLK(st.st_mode))
+        return filename;
+
+    /* first check if file has already loop device assigned according to
+     *
+     * losetup implementation readdir on /sys/block + stat on
+     * loop/backing_file is faster than open each loop dev
+     */
+    sysblock = opendir("/sys/block");
+    if (!sysblock) {
+        LOG(ERROR, "unable to open /sys/block");
+        return NULL;
+    }
+
+    dir_fd = dirfd(sysblock);
+
+    while ((d = readdir(sysblock))) {
+        char name[256];
+        struct stat buf;
+
+        if (strcmp(d->d_name, ".") == 0
+                || strcmp(d->d_name, "..") == 0
+                || strncmp(d->d_name, "loop", 4) != 0)
+            continue;
+        snprintf(name, sizeof(name), "%s/loop/backing_file", d->d_name);
+        if (fstatat(dir_fd, name, &buf, 0) == 0) {
+            snprintf(name, sizeof(name), "/dev/%s", d->d_name);
+            loop_fd = open(name, O_RDWR);
+            if (loop_fd < 0) {
+                LOG(ERROR, "unable to open %s", name);
+                goto out;
+            }
+            if (ioctl(loop_fd, LOOP_GET_STATUS64, &info) < 0) {
+                LOG(ERROR, "unable to get %s info: %s", name, strerror(errno));
+                goto out;
+            }
+            if (info.lo_inode == st.st_ino && info.lo_device == st.st_dev) {
+                /* found */
+                ret = libxl__strdup(gc, name);
+                goto out;
+            }
+        }
+    }
+
+    /* not found, so allocate new one */
+    file_fd = open(filename, O_RDWR);
+    if (file_fd < 0) {
+        LOG(ERROR, "unable to open %s: %s", filename, strerror(errno));
+        goto out;
+    }
+
+    control_fd = open("/dev/loop-control", O_RDWR);
+    if (control_fd < 0) {
+        LOG(ERROR, "unable to open /dev/loop-control: %s", strerror(errno));
+        goto out;
+    }
+    devnr = ioctl(control_fd, LOOP_CTL_GET_FREE);
+    if (devnr < 0) {
+        LOG(ERROR, "unable to get free loop device: %s", strerror(errno));
+        goto out;
+    }
+    loop_name = libxl__sprintf(gc, "/dev/loop%d", devnr);
+    loop_fd = open(loop_name, O_RDWR);
+    if (loop_fd < 0) {
+        LOG(ERROR, "unable to open %s: %s", loop_name, strerror(errno));
+        goto out;
+    }
+    if (ioctl(loop_fd, LOOP_SET_FD, file_fd) < 0) {
+        LOG(ERROR, "unable to set loop backing store: %s", strerror(errno));
+        goto out;
+    }
+    ret = loop_name;
+
+out:
+    if (file_fd >= 0)
+        close(file_fd);
+    if (control_fd >= 0)
+        close(control_fd);
+    if (loop_fd >= 0)
+        close(loop_fd);
+    if (sysblock)
+        closedir(sysblock);
+
+    return ret;
+}
+
+void libxl__loopdev_cleanup(libxl__gc *gc, char *devname)
+{
+    int loop_fd;
+
+    loop_fd = open(devname, O_RDWR);
+    if (loop_fd < 0) {
+        LOG(ERROR, "unable to open device %s: %s", devname, strerror(errno));
+        goto out;
+    }
+
+    if (ioctl(loop_fd, LOOP_CLR_FD) < 0) {
+        LOG(ERROR, "unable to release device %s: %s", devname, strerror(errno));
+        goto out;
+    }
+
+out:
+    if (loop_fd >= 0)
+        close(loop_fd);
+}
+
+
 #define EXT_SHIFT 28
 #define EXTENDED (1<<EXT_SHIFT)
 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
diff --git a/tools/libxl/libxl_netbsd.c b/tools/libxl/libxl_netbsd.c
index 898e160..1e71572 100644
--- a/tools/libxl/libxl_netbsd.c
+++ b/tools/libxl/libxl_netbsd.c
@@ -25,6 +25,17 @@ int libxl__try_phy_backend(mode_t st_mode)
     return 0;
 }
 
+char *libxl__loopdev_setup(libxl__gc *gc, char *filename)
+{
+    /* files supported natively */
+    return filename;
+}
+
+void libxl__loopdev_cleanup(libxl__gc *gc, char *devname)
+{
+    /* nothing to do */
+}
+
 char *libxl__devid_to_localdev(libxl__gc *gc, int devid)
 {
     /* TODO */
-- 
1.8.1.4


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 553 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel