[dm-devel] Latest dm-userspace kernel code
Dan Smith
danms at us.ibm.com
Mon Sep 18 14:52:48 UTC 2006
This is the latest dm-userspace kernel code. I have removed the
in-kernel remap cache and cleaned up a lot of code as a result.
Performance of the map cache/no cache versions seem to be about the
same.
I also tweaked the message protocol to be more rigid. It now looks
more like the ringbuffer messages, which should allow an easier
transition to a ringbuffer approach. This is not completely clean,
but I wanted to post it for visibility now that it seems to be stable
for me.
I will send the userspace example code shortly...
Comments appreciated.
--
Dan Smith
IBM Linux Technology Center
Open Hypervisor Team
email: danms at us.ibm.com
Signed-off-by: Dan Smith <danms at us.ibm.com>
diff -Naur linux-2.6.17.13-orig/drivers/md/dm-user.h linux-2.6.17.13-dmu/drivers/md/dm-user.h
--- linux-2.6.17.13-orig/drivers/md/dm-user.h 1969-12-31 16:00:00.000000000 -0800
+++ linux-2.6.17.13-dmu/drivers/md/dm-user.h 2006-09-15 14:20:48.000000000 -0700
@@ -0,0 +1,142 @@
+/*
+ * Copyright (C) International Business Machines Corp., 2006
+ * Author: Dan Smith <danms at us.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; under version 2 of the License.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ *
+ */
+
+#ifndef __DM_USER_H
+#define __DM_USER_H
+
+#include <linux/hardirq.h>
+
+#define DMU_KEY_LEN 256
+
+extern struct target_type userspace_target;
+extern mempool_t *request_pool;
+extern dev_t dmu_dev;
+extern spinlock_t devices_lock;
+extern struct list_head devices;
+
+/*
+ * A block device that we can send bios to
+ */
+struct target_device {
+ struct list_head list; /* Our place in the targets list */
+ struct block_device *bdev; /* The target block_device */
+ struct kref users; /* Self-destructing reference count */
+};
+
+/*
+ * A dm-userspace device, which consists of multiple targets sharing a
+ * common key
+ */
+struct dmu_device {
+ struct list_head list; /* Our place in the devices list */
+
+ spinlock_t lock; /* Protects all the fields below */
+
+ struct list_head tx_requests; /* Requests to send to userspace */
+ struct list_head rx_requests; /* Requests waiting for reply */
+ struct list_head cp_requests; /* Requests waiting to be copied */
+
+ struct list_head target_devs; /* List of devices we can target */
+
+ void *transport_private; /* Private data for userspace comms */
+
+ char key[DMU_KEY_LEN]; /* Unique name string for device */
+ struct kref users; /* Self-destructing reference count */
+
+ wait_queue_head_t wqueue; /* To block while waiting for reqs */
+
+ uint64_t block_size; /* Block size for this device */
+ uint64_t block_mask; /* Mask for offset in block */
+ unsigned int block_shift; /* Shift to convert to/from block */
+
+ struct kcopyd_client *kcopy; /* Interface to kcopyd */
+
+ /* FIXME: Remove after debugging */
+ atomic_t t_reqs;
+ atomic_t r_reqs;
+ atomic_t f_reqs;
+ atomic_t total;
+};
+
+struct dmu_request {
+ struct list_head list; /* Our place on the request queue */
+ struct dmu_device *dev; /* The DMU device that owns us */
+
+ int type; /* Type of request */
+ uint32_t flags; /* Attribute flags */
+ uint64_t id; /* Unique ID for sync with userspace */
+ union {
+ uint64_t block; /* The block in question */
+ } u;
+
+ struct list_head deps; /* Requests depending on this one */
+ struct bio *bio; /* The bio this request represents */
+
+ struct work_struct task; /* Async task to run for this req */
+
+ struct dmu_msg_map_response response; /* FIXME: Clean this up */
+};
+
+
+/* Find and grab a reference to a target device */
+struct target_device *find_target(struct dmu_device *dev,
+ dev_t devno);
+/* Character device transport functions */
+int register_chardev_transport(struct dmu_device *dev);
+void unregister_chardev_transport(struct dmu_device *dev);
+int init_chardev_transport(void);
+void cleanup_chardev_transport(void);
+void write_chardev_transport_info(struct dmu_device *dev,
+ char *buf, unsigned int maxlen);
+
+/* Return the block number for @sector */
+static inline u64 dmu_block(struct dmu_device *dev,
+ sector_t sector)
+{
+ return sector >> dev->block_shift;
+}
+
+/* Return the sector offset in a block for @sector */
+static inline u64 dmu_sector_offset(struct dmu_device *dev,
+ sector_t sector)
+{
+ return sector & dev->block_mask;
+}
+
+/* Return the starting sector for @block */
+static inline u64 dmu_sector(struct dmu_device *dev,
+ uint64_t block)
+{
+ return block << dev->block_shift;
+}
+
+/* Increase the usage count for @dev */
+static inline void get_dev(struct dmu_device *dev)
+{
+ kref_get(&dev->users);
+}
+
+/* Decrease the usage count for @dev */
+void destroy_dmu_device(struct kref *ref);
+static inline void put_dev(struct dmu_device *dev)
+{
+ kref_put(&dev->users, destroy_dmu_device);
+}
+
+#endif
diff -Naur linux-2.6.17.13-orig/drivers/md/dm-userspace.c linux-2.6.17.13-dmu/drivers/md/dm-userspace.c
--- linux-2.6.17.13-orig/drivers/md/dm-userspace.c 1969-12-31 16:00:00.000000000 -0800
+++ linux-2.6.17.13-dmu/drivers/md/dm-userspace.c 2006-09-15 14:20:48.000000000 -0700
@@ -0,0 +1,575 @@
+/*
+ * Copyright (C) International Business Machines Corp., 2006
+ * Author: Dan Smith <danms at us.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; under version 2 of the License.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/blkdev.h>
+#include <linux/bio.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/list.h>
+#include <linux/fs.h>
+#include <linux/cdev.h>
+#include <linux/types.h>
+#include <linux/poll.h>
+
+#include <linux/dm-userspace.h>
+
+#include "dm.h"
+#include "dm-bio-list.h"
+#include "kcopyd.h"
+#include "dm-user.h"
+
+#define DMU_COPY_PAGES 256
+
+#define DM_MSG_PREFIX "dm-userspace"
+
+static kmem_cache_t *request_cache;
+mempool_t *request_pool;
+
+spinlock_t devices_lock;
+LIST_HEAD(devices);
+
+/* Device number for the control device */
+dev_t dmu_dev;
+
+/* Add a request to a device's request queue */
+static void add_tx_request(struct dmu_device *dev,
+ struct dmu_request *req)
+{
+
+ req->id = (uint64_t)(unsigned int)req;
+
+ spin_lock(&dev->lock);
+ list_add_tail(&req->list, &dev->tx_requests);
+ spin_unlock(&dev->lock);
+
+ wake_up(&dev->wqueue);
+}
+
+/* Handle the cleanup of a request once its endio has fired. In
+ * general, the endio has been allowed to complete and all this does
+ * is clean up the request (or wait, if necessary). In the case of a
+ * SYNC request, we may need to send a message to userspace
+ */
+static void endio_worker(void *data)
+{
+ struct dmu_request *req = data;
+ enum {RESCHED, REQ_SYNC, FREE_REQ} task;
+
+ spin_lock(&req->dev->lock);
+ if (dmu_get_flag(&req->flags, DMU_FLAG_SYNC))
+ task = REQ_SYNC;
+ else if (!list_empty(&req->list))
+ task = RESCHED;
+ else if (list_empty(&req->list))
+ task = FREE_REQ;
+ spin_unlock(&req->dev->lock);
+
+ if (task == REQ_SYNC) {
+ req->type = DM_USERSPACE_SYNC_COMPLETE;
+ add_tx_request(req->dev, req);
+ atomic_inc(&req->dev->t_reqs);
+ } else if (task == RESCHED) {
+ PREPARE_WORK(&req->task, endio_worker, req);
+ schedule_work(&req->task);
+ } else if (task == FREE_REQ) {
+ atomic_dec(&req->dev->f_reqs);
+ mempool_free(req, request_pool);
+ }
+}
+
+/* Return an already-bound target device */
+struct target_device *find_target(struct dmu_device *dev,
+ dev_t devno)
+{
+ struct target_device *target, *match = NULL;
+
+ spin_lock(&dev->lock);
+ list_for_each_entry(target, &dev->target_devs, list) {
+ if (target->bdev->bd_dev == devno) {
+ match = target;
+ break;
+ }
+ }
+ spin_unlock(&dev->lock);
+
+ return match;
+}
+
+/* Find a new target device and bind it to our device */
+static struct target_device *get_target(struct dmu_device *dev,
+ dev_t devno)
+{
+ struct target_device *target;
+ struct block_device *bdev;
+
+ target = find_target(dev, devno);
+ if (target)
+ return target;
+
+ bdev = open_by_devnum(devno, FMODE_READ | FMODE_WRITE);
+ if (IS_ERR(bdev)) {
+ DMERR("Unable to lookup device %x", devno);
+ return NULL;
+ }
+
+ target = kmalloc(sizeof(*target), GFP_KERNEL);
+ if (!target) {
+ DMERR("Unable to alloc new target device");
+ return NULL;
+ }
+
+ target->bdev = bdev;
+ INIT_LIST_HEAD(&target->list);
+
+ if (in_interrupt())
+ printk("%s in irq\n", __FUNCTION__);
+
+ spin_lock(&dev->lock);
+ list_add_tail(&target->list, &dev->target_devs);
+ spin_unlock(&dev->lock);
+
+ return target;
+}
+
+/* Caller must hold dev->lock */
+static void put_target(struct dmu_device *dev,
+ struct target_device *target)
+{
+ list_del(&target->list);
+
+ bd_release(target->bdev);
+ blkdev_put(target->bdev);
+
+ kfree(target);
+}
+
+void destroy_dmu_device(struct kref *ref)
+{
+ struct dmu_device *dev;
+ struct list_head *cursor, *next;
+
+ dev = container_of(ref, struct dmu_device, users);
+
+ spin_lock(&devices_lock);
+ list_del(&dev->list);
+ spin_unlock(&devices_lock);
+
+ list_for_each_safe(cursor, next, &dev->target_devs) {
+ struct target_device *target;
+
+ target = list_entry(cursor,
+ struct target_device,
+ list);
+
+ put_target(dev, target);
+ }
+
+ list_for_each_safe(cursor, next, &dev->tx_requests) {
+ struct dmu_request *req;
+
+ req = list_entry(cursor,
+ struct dmu_request,
+ list);
+
+ DMERR("Failing unsent bio");
+ bio_io_error(req->bio, req->bio->bi_size);
+
+ list_del(&req->list);
+
+ mempool_free(req, request_pool);
+ }
+
+ list_for_each_safe(cursor, next, &dev->rx_requests) {
+ struct dmu_request *req;
+
+ req = list_entry(cursor,
+ struct dmu_request,
+ list);
+
+ DMERR("Failing bio");
+ req->flags = 0;
+ bio_io_error(req->bio, req->bio->bi_size);
+
+ list_del(&req->list);
+
+ mempool_free(req, request_pool);
+ }
+
+ list_for_each_safe(cursor, next, &dev->cp_requests) {
+ struct dmu_request *req;
+
+ req = list_entry(cursor,
+ struct dmu_request,
+ list);
+
+ DMERR("Failing bio");
+ req->flags = 0;
+ bio_io_error(req->bio, req->bio->bi_size);
+
+ list_del(&req->list);
+
+ mempool_free(req, request_pool);
+ }
+
+ kcopyd_client_destroy(dev->kcopy);
+ unregister_chardev_transport(dev);
+
+ kfree(dev);
+}
+
+static int init_dmu_device(struct dmu_device *dev, u32 block_size)
+{
+ int ret;
+
+ init_waitqueue_head(&dev->wqueue);
+ INIT_LIST_HEAD(&dev->list);
+ INIT_LIST_HEAD(&dev->target_devs);
+ kref_init(&dev->users);
+ spin_lock_init(&dev->lock);
+
+ INIT_LIST_HEAD(&dev->tx_requests);
+ INIT_LIST_HEAD(&dev->rx_requests);
+ INIT_LIST_HEAD(&dev->cp_requests);
+
+ dev->block_size = block_size;
+ dev->block_mask = block_size - 1;
+ dev->block_shift = ffs(block_size) - 1;
+
+ printk("%s: %llu %llu %u\n", dev->key,
+ dev->block_size, dev->block_mask, dev->block_shift);
+
+ atomic_set(&dev->t_reqs, 0);
+ atomic_set(&dev->r_reqs, 0);
+ atomic_set(&dev->f_reqs, 0);
+ atomic_set(&dev->total, 0);
+
+ ret = kcopyd_client_create(DMU_COPY_PAGES, &dev->kcopy);
+ if (ret) {
+ DMERR("Failed to initialize kcopyd client");
+ return 0;
+ }
+
+ return 1;
+}
+
+static struct dmu_device *new_dmu_device(char *key,
+ struct dm_target *ti,
+ u32 block_size)
+{
+ struct dmu_device *dev;
+ int ret;
+
+ dev = kmalloc(sizeof(*dev), GFP_KERNEL);
+ if (dev == NULL) {
+ DMERR("Failed to allocate new userspace device");
+ return NULL;
+ }
+
+ if (!init_dmu_device(dev, block_size))
+ goto bad1;
+
+ snprintf(dev->key, DMU_KEY_LEN, "%s", key);
+
+ ret = register_chardev_transport(dev);
+ if (!ret)
+ goto bad2;
+
+ spin_lock(&devices_lock);
+ list_add(&dev->list, &devices);
+ spin_unlock(&devices_lock);
+
+ return dev;
+
+ bad2:
+ put_dev(dev);
+ bad1:
+ kfree(dev);
+ DMERR("Failed to create device");
+ return NULL;
+}
+
+static struct dmu_device *find_dmu_device(const char *key)
+{
+ struct dmu_device *dev;
+ struct dmu_device *match = NULL;
+
+ spin_lock(&devices_lock);
+
+ list_for_each_entry(dev, &devices, list) {
+ spin_lock(&dev->lock);
+ if (strncmp(dev->key, key, DMU_KEY_LEN) == 0) {
+ match = dev;
+ spin_unlock(&dev->lock);
+ break;
+ }
+ spin_unlock(&dev->lock);
+ }
+
+ spin_unlock(&devices_lock);
+
+ return match;
+}
+
+static int dmu_ctr(struct dm_target *ti, unsigned int argc, char **argv)
+{
+ uint64_t block_size;
+ struct dmu_device *dev;
+ char *device_key;
+ char *block_size_param;
+ int target_idx = 2;
+
+ if (argc < 3) {
+ ti->error = "Invalid argument count";
+ return -EINVAL;
+ }
+
+ device_key = argv[0];
+ block_size_param = argv[1];
+
+ block_size = simple_strtoul(block_size_param, NULL, 10) / 512;
+
+ dev = find_dmu_device(device_key);
+ if (dev == NULL) {
+ dev = new_dmu_device(device_key,
+ ti,
+ block_size);
+ if (dev == NULL) {
+ ti->error = "Failed to create device";
+ goto bad;
+ }
+ } else {
+ get_dev(dev);
+ }
+
+ spin_lock(&dev->lock);
+ if (dev->block_size != block_size) {
+ ti->error = "Invalid block size";
+ goto bad;
+ }
+ spin_unlock(&dev->lock);
+
+ /* Resolve target devices */
+ do {
+ int maj, min;
+ sscanf(argv[target_idx], "%i:%i", &maj, &min);
+ if (!get_target(dev, MKDEV(maj, min))) {
+ DMERR("Failed to find target device %i:%i (%s)",
+ maj, min, argv[target_idx]);
+ goto out;
+ }
+ } while (++target_idx < argc);
+
+ ti->private = dev;
+ ti->split_io = block_size;
+
+ return 0;
+
+ bad:
+ if (dev) {
+ spin_unlock(&dev->lock);
+ }
+ out:
+ if (dev) {
+ put_dev(dev);
+ }
+
+ return -EINVAL;
+}
+
+static void dmu_dtr(struct dm_target *ti)
+{
+ struct dmu_device *dev = (struct dmu_device *) ti->private;
+
+ put_dev(dev);
+}
+
+static int dmu_map(struct dm_target *ti, struct bio *bio,
+ union map_info *map_context)
+{
+ struct dmu_device *dev = (struct dmu_device *) ti->private;
+ struct dmu_request *req;
+
+ req = mempool_alloc(request_pool, GFP_NOIO);
+ if (!req) {
+ DMERR("Failed to alloc request");
+ return -1;
+ }
+
+ atomic_inc(&dev->t_reqs);
+ atomic_inc(&dev->total);
+
+ map_context->ptr = req;
+
+ /* FIXME: Need an allocation function here */
+ req->type = DM_USERSPACE_MAP_BLOCK_REQ;
+ req->dev = dev;
+ req->bio = bio;
+ req->u.block = dmu_block(dev, bio->bi_sector);
+ req->flags = 0;
+ INIT_LIST_HEAD(&req->deps);
+
+ if (bio_rw(bio))
+ dmu_set_flag(&req->flags, DMU_FLAG_WR);
+
+ add_tx_request(dev, req);
+
+ return 0;
+}
+
+static int dmu_status(struct dm_target *ti, status_type_t type,
+ char *result, unsigned int maxlen)
+{
+ struct dmu_device *dev = (struct dmu_device *) ti->private;
+
+ /* FIXME: Remove after debug */
+ spin_lock(&dev->lock);
+ printk("Requests: %u t:%u r:%u f:%u (%c%c)\n",
+ atomic_read(&dev->total),
+ atomic_read(&dev->t_reqs),
+ atomic_read(&dev->r_reqs),
+ atomic_read(&dev->f_reqs),
+ list_empty(&dev->tx_requests) ? ' ':'T',
+ list_empty(&dev->rx_requests) ? ' ':'R');
+ spin_unlock(&dev->lock);
+
+ switch (type) {
+ case STATUSTYPE_INFO:
+ write_chardev_transport_info(dev, result, maxlen);
+ break;
+
+ case STATUSTYPE_TABLE:
+ snprintf(result, maxlen, "%s %llu",
+ dev->key,
+ dev->block_size * 512);
+ break;
+ }
+
+ return 0;
+}
+
+static int dmu_end_io(struct dm_target *ti, struct bio *bio,
+ int error, union map_info *map_context)
+{
+ struct dmu_request *req = map_context->ptr;
+ int ret = 0;
+
+ if (error) {
+ DMERR("Error in dmu_end_io");
+ return -1;
+ }
+
+ INIT_WORK(&req->task, endio_worker, req);
+ schedule_work(&req->task);
+
+ if (dmu_get_flag(&req->flags, DMU_FLAG_SYNC))
+ ret = 1;
+
+ return ret;
+}
+
+struct target_type userspace_target = {
+ .name = "userspace",
+ .version = {0, 1, 0},
+ .module = THIS_MODULE,
+ .ctr = dmu_ctr,
+ .dtr = dmu_dtr,
+ .map = dmu_map,
+ .status = dmu_status,
+ .end_io = dmu_end_io
+};
+
+int __init dm_userspace_init(void)
+{
+ int r = dm_register_target(&userspace_target);
+ if (r < 0) {
+ DMERR("Register failed %d", r);
+ return 0;
+ }
+
+ spin_lock_init(&devices_lock);
+
+ request_cache =
+ kmem_cache_create("dm-userspace-requests",
+ sizeof(struct dmu_request),
+ __alignof__ (struct dmu_request),
+ 0, NULL, NULL);
+ if (!request_cache) {
+ DMERR("Failed to allocate request cache");
+ goto bad;
+ }
+
+ request_pool = mempool_create(64,
+ mempool_alloc_slab, mempool_free_slab,
+ request_cache);
+ if (!request_pool) {
+ DMERR("Failed to allocate request pool");
+ goto bad2;
+ }
+
+ r = init_chardev_transport();
+ if (!r)
+ goto bad3;
+
+ return 1;
+
+ bad3:
+ mempool_destroy(request_pool);
+ bad2:
+ kmem_cache_destroy(request_cache);
+ bad:
+ dm_unregister_target(&userspace_target);
+
+ return 0;
+}
+
+void __exit dm_userspace_exit(void)
+{
+ int r;
+ struct list_head *cursor, *next;
+ struct dmu_device *dev;
+
+ spin_lock(&devices_lock);
+
+ list_for_each_safe(cursor, next, &devices) {
+ dev = list_entry(cursor, struct dmu_device, list);
+ list_del(cursor);
+ destroy_dmu_device(&dev->users);
+ DMERR("Destroying hanging device %s", dev->key);
+ }
+
+ spin_unlock(&devices_lock);
+
+ cleanup_chardev_transport();
+
+ mempool_destroy(request_pool);
+ kmem_cache_destroy(request_cache);
+
+ r = dm_unregister_target(&userspace_target);
+ if (r < 0)
+ DMERR("unregister failed %d", r);
+}
+
+module_init(dm_userspace_init);
+module_exit(dm_userspace_exit);
+
+MODULE_DESCRIPTION(DM_NAME " userspace target");
+MODULE_AUTHOR("Dan Smith");
+MODULE_LICENSE("GPL");
diff -Naur linux-2.6.17.13-orig/drivers/md/dm-userspace-chardev.c linux-2.6.17.13-dmu/drivers/md/dm-userspace-chardev.c
--- linux-2.6.17.13-orig/drivers/md/dm-userspace-chardev.c 1969-12-31 16:00:00.000000000 -0800
+++ linux-2.6.17.13-dmu/drivers/md/dm-userspace-chardev.c 2006-09-15 14:20:48.000000000 -0700
@@ -0,0 +1,614 @@
+/*
+ * Copyright (C) International Business Machines Corp., 2006
+ * Author: Dan Smith <danms at us.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; under version 2 of the License.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ *
+ */
+
+#include <linux/spinlock.h>
+#include <linux/blkdev.h>
+#include <linux/mempool.h>
+#include <linux/dm-userspace.h>
+#include <linux/list.h>
+#include <linux/sched.h>
+#include <linux/wait.h>
+#include <linux/poll.h>
+#include <linux/fs.h>
+#include <linux/cdev.h>
+#include <asm/uaccess.h>
+
+#include "dm.h"
+#include "dm-bio-list.h"
+#include "kcopyd.h"
+#include "dm-user.h"
+
+#define DM_MSG_PREFIX "dm-userspace"
+
+/* This allows for a cleaner separation between the dm-userspace
+ * device-mapper target, and the userspace transport used. Right now,
+ * only a chardev transport exists, but it's possible that there could
+ * be more in the future
+ */
+struct chardev_transport {
+ struct cdev cdev;
+ dev_t ctl_dev;
+ struct dmu_device *parent;
+};
+
+static void add_rx_request(struct dmu_device *dev,
+ struct dmu_request *req)
+{
+ spin_lock(&dev->lock);
+ list_add_tail(&req->list, &dev->rx_requests);
+ spin_unlock(&dev->lock);
+}
+
+static struct dmu_request *find_rx_request(struct dmu_device *dev,
+ uint64_t id)
+{
+ struct dmu_request *req, *next, *match = NULL;
+
+ spin_lock(&dev->lock);
+ list_for_each_entry_safe(req, next, &dev->rx_requests, list) {
+ if (req->id == id) {
+ list_del_init(&req->list);
+ match = req;
+ break;
+ }
+ }
+ spin_unlock(&dev->lock);
+
+ return match;
+}
+
+static int have_pending_requests(struct dmu_device *dev)
+{
+ return atomic_read(&dev->t_reqs) != 0;
+}
+
+static int send_userspace_message(uint8_t __user *buffer,
+ struct dmu_request *req)
+{
+ int ret = 0;
+ struct dmu_msg msg;
+
+ memset(&msg, 0, sizeof(msg));
+
+ msg.hdr.id = req->id;
+
+ switch (req->type) {
+ case DM_USERSPACE_GET_VERSION:
+ msg.hdr.msg_type = req->type;
+ msg.payload.ver.kernel_ver =
+ userspace_target.version[0] << 16 |
+ userspace_target.version[1] << 8 |
+ userspace_target.version[2];
+
+ break;
+
+ case DM_USERSPACE_MAP_BLOCK_REQ:
+ msg.hdr.msg_type = req->type;
+ msg.payload.map_req.org_block = req->u.block;
+ dmu_cpy_flag(&msg.payload.map_req.flags,
+ req->flags, DMU_FLAG_RD);
+ dmu_cpy_flag(&msg.payload.map_req.flags,
+ req->flags, DMU_FLAG_WR);
+
+ break;
+
+ case DM_USERSPACE_SYNC_COMPLETE:
+ case DM_USERSPACE_INVAL_COMPLETE:
+ case DM_USERSPACE_INVAL_FAILED:
+ msg.hdr.msg_type = DM_USERSPACE_STATUS;
+ msg.payload.status.status = req->type;
+ msg.payload.status.id_of_op = req->id;
+
+ break;
+
+ default:
+ DMWARN("Unknown outgoing message type %i", req->type);
+ ret = 0;
+ }
+
+ if (copy_to_user(buffer, &msg, sizeof(msg)))
+ return -EFAULT;
+
+ ret = sizeof(msg);
+
+ if ((msg.hdr.msg_type != DM_USERSPACE_MAP_BLOCK_REQ) &&
+ (msg.hdr.msg_type != DM_USERSPACE_STATUS)) {
+ printk("Sending weird message type %u\n", msg.hdr.msg_type);
+ }
+
+ if (req->type == DM_USERSPACE_GET_VERSION) {
+ mempool_free(req, request_pool);
+ }
+
+ return ret;
+}
+
+struct dmu_request *pluck_next_request(struct dmu_device *dev)
+{
+ struct dmu_request *req = NULL;
+
+ spin_lock(&dev->lock);
+ if (!list_empty(&dev->tx_requests)) {
+ req = list_entry(dev->tx_requests.next,
+ struct dmu_request, list);
+
+ list_del_init(&req->list);
+ atomic_dec(&dev->t_reqs);
+ }
+ spin_unlock(&dev->lock);
+
+ if (req && ((req->type == DM_USERSPACE_MAP_BLOCK_REQ) ||
+ (req->type == DM_USERSPACE_SYNC_COMPLETE))) {
+ add_rx_request(dev, req);
+ atomic_inc(&dev->r_reqs);
+ }
+
+ return req;
+}
+
+ssize_t dmu_ctl_read(struct file *file, char __user *buffer,
+ size_t size, loff_t *offset)
+{
+
+ struct dmu_device *dev = (struct dmu_device *)file->private_data;
+ struct dmu_request *req = NULL;
+ int ret = 0, r;
+
+ if (!capable(CAP_SYS_ADMIN))
+ return -EACCES;
+
+ if (size < sizeof(struct dmu_msg)) {
+ DMERR("Userspace buffer too small for a single message");
+ return 0;
+ }
+
+ while (!have_pending_requests(dev)) {
+ if (file->f_flags & O_NONBLOCK) {
+ return 0;
+ }
+
+ if (wait_event_interruptible(dev->wqueue,
+ have_pending_requests(dev)))
+ return -ERESTARTSYS;
+ }
+
+ while (ret < size) {
+ if ((size - ret) < sizeof(struct dmu_msg))
+ break;
+
+ req = pluck_next_request(dev);
+ if (!req)
+ break;
+
+ r = send_userspace_message((void *)(buffer + ret), req);
+ if (r == 0)
+ continue;
+ else if (r < 0)
+ return r;
+
+ ret += r;
+ }
+
+ if (ret < sizeof(struct dmu_msg))
+ DMINFO("Sent 0 requests to userspace");
+
+ return ret;
+}
+
+/*
+ * Adds the request to the front of the queue so it's picked up first
+ */
+static void add_urgent_request(struct dmu_device *dev,
+ struct dmu_request *req)
+{
+
+ if (in_interrupt())
+ printk("%s in irq\n", __FUNCTION__);
+
+ spin_lock(&dev->lock);
+ list_add(&req->list, &dev->tx_requests);
+ spin_unlock(&dev->lock);
+
+ wake_up(&dev->wqueue);
+}
+
+static int version_request(struct dmu_msg_version *msg,
+ struct dmu_device *dev, uint32_t id)
+{
+ struct dmu_request *req;
+
+ req = mempool_alloc(request_pool, GFP_NOIO);
+ if (!req) {
+ DMERR("Failed to alloc version response");
+ return 0;
+ }
+
+ atomic_inc(&dev->t_reqs);
+
+ req->dev = dev;
+ req->type = DM_USERSPACE_GET_VERSION;
+ add_urgent_request(dev, req);
+
+ return 1;
+}
+
+static struct dmu_request *pluck_dep_req(struct dmu_request *req)
+{
+ struct dmu_request *dreq = NULL;
+
+ spin_lock(&req->dev->lock);
+ if (list_empty(&req->deps)) {
+ /* Delete from cp_requests */
+ list_del_init(&req->list);
+ } else {
+ /* Get next dependent request */
+ dreq = list_entry(req->deps.next, struct dmu_request, list);
+ list_del_init(&dreq->list);
+ }
+ spin_unlock(&req->dev->lock);
+
+ return dreq;
+}
+
+static void flush_block(int read_err, unsigned int write_err, void *data)
+{
+ struct dmu_request *req = data;
+ struct dmu_request *dreq;
+
+ generic_make_request(req->bio);
+
+ while ((dreq = pluck_dep_req(req)))
+ generic_make_request(dreq->bio);
+}
+
+static void copy_block(struct dmu_device *dev,
+ struct block_device *src_dev,
+ struct block_device *dst_dev,
+ struct dmu_request *req,
+ uint64_t org_block,
+ uint64_t new_block,
+ int64_t offset)
+{
+ struct io_region src, dst;
+
+ src.bdev = src_dev;
+ src.sector = org_block << dev->block_shift;
+ src.count = dev->block_size;
+
+ dst.bdev = dst_dev;
+ dst.sector = new_block << dev->block_shift;
+ dst.sector += offset;
+ dst.count = dev->block_size;
+
+ kcopyd_copy(dev->kcopy, &src, 1, &dst, 0, flush_block, req);
+}
+
+/*
+ * Queues @req with a waiting request to the same block, if one
+ * exists. Returns nonzero if queued.
+ */
+static int queue_dependent_request(struct dmu_request *req)
+{
+ struct dmu_request *dreq = NULL;
+ int found = 0;
+
+ spin_lock(&req->dev->lock);
+ list_for_each_entry(dreq, &req->dev->cp_requests, list) {
+ if (dmu_block(req->dev, req->bio->bi_sector) ==
+ dmu_block(req->dev, dreq->bio->bi_sector)) {
+ list_add_tail(&req->list, &dreq->deps);
+ found = 1;
+ break;
+ }
+ }
+ spin_unlock(&req->dev->lock);
+
+ return found;
+}
+
+static void map_worker(void *data)
+{
+ struct dmu_request *req = data;
+ struct dmu_msg_map_response *msg = &req->response;
+ struct dmu_device *dev = req->dev;
+ struct target_device *src_dev, *dst_dev;
+
+ if (dmu_get_flag(&msg->flags, DMU_FLAG_COPY_FIRST)) {
+ src_dev = find_target(dev, MKDEV(msg->src_maj, msg->src_min));
+ if (!src_dev) {
+ DMERR("Failed to find src device %i:%i\n",
+ msg->src_maj, msg->src_min);
+ goto fail;
+ }
+ } else
+ src_dev = NULL;
+
+ dst_dev = find_target(dev, MKDEV(msg->dst_maj, msg->dst_min));
+ if (!dst_dev) {
+ DMERR("Failed to find dest device %i:%i\n",
+ msg->dst_maj, msg->dst_min);
+ goto fail;
+ }
+
+ /* Remap the bio */
+ req->bio->bi_sector = dmu_sector(dev, msg->new_block) +
+ dmu_sector_offset(dev, req->bio->bi_sector) +
+ msg->offset;
+ req->bio->bi_bdev = dst_dev->bdev;
+
+ dmu_cpy_flag(&req->flags, msg->flags, DMU_FLAG_SYNC);
+
+ atomic_dec(&dev->r_reqs);
+ atomic_inc(&dev->f_reqs);
+
+ if (!queue_dependent_request(req)) {
+ if (dmu_get_flag(&msg->flags, DMU_FLAG_COPY_FIRST)) {
+ spin_lock(&dev->lock);
+ list_add(&req->list, &req->dev->cp_requests);
+ spin_unlock(&dev->lock);
+
+ copy_block(dev, src_dev->bdev, dst_dev->bdev, req,
+ msg->org_block, msg->new_block,
+ msg->offset);
+ } else {
+ flush_block(0, 0, req);
+ }
+ }
+
+ return;
+
+ fail:
+ bio_io_error(req->bio, req->bio->bi_size);
+}
+
+static void do_map_bio(struct dmu_device *dev,
+ struct dmu_msg_map_response *msg)
+{
+ struct dmu_request *req;
+
+ req = find_rx_request(dev, msg->id_of_req);
+ if (!req) {
+ DMERR("Unable to complete unknown map: %llu\n",
+ msg->id_of_req);
+ return;
+ }
+
+ memcpy(&req->response, msg, sizeof(req->response));
+
+ INIT_WORK(&req->task, map_worker, req);
+ schedule_work(&req->task);
+}
+
+static void do_sync_complete(struct dmu_device *dev, uint32_t id_of_op) {
+ struct dmu_request *req;
+
+ req = find_rx_request(dev, id_of_op);
+ if (!req) {
+ DMERR("Unable to complete unknown request: %u\n",
+ id_of_op);
+ return;
+ }
+
+ atomic_dec(&req->dev->r_reqs);
+
+ dmu_clr_flag(&req->flags, DMU_FLAG_SYNC);
+
+ req->bio->bi_end_io(req->bio, 0, 0);
+}
+
+ssize_t dmu_ctl_write(struct file *file, const char __user *buffer,
+ size_t size, loff_t *offset)
+{
+ struct dmu_device *dev = (struct dmu_device *)file->private_data;
+ int ret = 0;
+ struct dmu_msg msg;
+
+ if (!capable(CAP_SYS_ADMIN))
+ return -EACCES;
+
+ while ((ret + sizeof(msg)) <= size) {
+ if (copy_from_user(&msg, buffer+ret, sizeof(msg))) {
+ DMERR("%s copy_from_user failed!", __FUNCTION__);
+ ret = -EFAULT;
+ goto out;
+ }
+
+ ret += sizeof(msg);
+
+ switch (msg.hdr.msg_type) {
+
+ case DM_USERSPACE_GET_VERSION:
+ version_request(&msg.payload.ver, dev, msg.hdr.id);
+ break;
+
+ case DM_USERSPACE_MAP_BLOCK_RESP:
+ do_map_bio(dev, &msg.payload.map_rsp);
+ break;
+
+ case DM_USERSPACE_MAP_FAILED:
+ /* FIXME: Do something here */
+ DMERR("Userspace map failed");
+ break;
+
+ case DM_USERSPACE_STATUS:
+ if (msg.payload.status.status ==
+ DM_USERSPACE_SYNC_COMPLETE) {
+ do_sync_complete(dev,
+ msg.payload.status.id_of_op);
+ } else {
+ printk("Unknown status type %u\n",
+ msg.payload.status.status);
+ }
+ break;
+
+ default:
+ DMWARN("Unknown incoming request type: %i",
+ msg.hdr.msg_type);
+ }
+ }
+ out:
+ if (ret < sizeof(msg)) {
+ DMINFO("Received 0 responses from userspace");
+ }
+
+ return ret;
+}
+
+int dmu_ctl_open(struct inode *inode, struct file *file)
+{
+ struct chardev_transport *t;
+ struct dmu_device *dev;
+
+ if (!capable(CAP_SYS_ADMIN))
+ return -EACCES;
+
+ t = container_of(inode->i_cdev, struct chardev_transport, cdev);
+ dev = t->parent;
+
+ get_dev(dev);
+
+ file->private_data = dev;
+
+ return 0;
+}
+
+int dmu_ctl_release(struct inode *inode, struct file *file)
+{
+ struct dmu_device *dev;
+
+ dev = (struct dmu_device *)file->private_data;
+
+ put_dev(dev);
+
+ return 0;
+}
+
+unsigned dmu_ctl_poll(struct file *file, poll_table *wait)
+{
+ struct dmu_device *dev = (struct dmu_device *)file->private_data;
+ unsigned mask = 0;
+
+ poll_wait(file, &dev->wqueue, wait);
+
+ if (have_pending_requests(dev))
+ mask |= POLLIN | POLLRDNORM;
+
+ return mask;
+}
+
+static struct file_operations ctl_fops = {
+ .open = dmu_ctl_open,
+ .release = dmu_ctl_release,
+ .read = dmu_ctl_read,
+ .write = dmu_ctl_write,
+ .poll = dmu_ctl_poll,
+ .owner = THIS_MODULE,
+};
+
+static int get_free_minor(void)
+{
+ struct dmu_device *dev;
+ int minor = 0;
+
+ spin_lock(&devices_lock);
+
+ while (1) {
+ list_for_each_entry(dev, &devices, list) {
+ struct chardev_transport *t = dev->transport_private;
+ if (MINOR(t->ctl_dev) == minor)
+ goto dupe;
+ }
+ break;
+ dupe:
+ minor++;
+ }
+
+ spin_unlock(&devices_lock);
+
+ return minor;
+}
+
+int register_chardev_transport(struct dmu_device *dev)
+{
+ struct chardev_transport *t;
+ int ret;
+
+ dev->transport_private = kmalloc(sizeof(struct chardev_transport),
+ GFP_KERNEL);
+ t = dev->transport_private;
+
+ if (!t) {
+ DMERR("Failed to allocate chardev transport");
+ goto bad;
+ }
+
+ t->ctl_dev = MKDEV(MAJOR(dmu_dev), get_free_minor());
+ t->parent = dev;
+
+ cdev_init(&t->cdev, &ctl_fops);
+ t->cdev.owner = THIS_MODULE;
+ t->cdev.ops = &ctl_fops;
+
+ ret = cdev_add(&t->cdev, t->ctl_dev, 1);
+ if (ret < 0) {
+ DMERR("Failed to register control device %d:%d",
+ MAJOR(t->ctl_dev), MINOR(t->ctl_dev));
+ goto bad;
+ }
+
+ return 1;
+
+ bad:
+ kfree(t);
+ return 0;
+}
+
+void unregister_chardev_transport(struct dmu_device *dev)
+{
+ struct chardev_transport *t = dev->transport_private;
+
+ cdev_del(&t->cdev);
+ kfree(t);
+}
+
+int init_chardev_transport(void)
+{
+ int r;
+
+ r = alloc_chrdev_region(&dmu_dev, 0, 10, "dm-userspace");
+ if (r) {
+ DMERR("Failed to allocate chardev region");
+ return 0;
+ } else
+ return 1;
+}
+
+void cleanup_chardev_transport(void)
+{
+ unregister_chrdev_region(dmu_dev, 10);
+}
+
+void write_chardev_transport_info(struct dmu_device *dev,
+ char *buf, unsigned int maxlen)
+{
+ struct chardev_transport *t = dev->transport_private;
+
+ snprintf(buf, maxlen, "%x:%x",
+ MAJOR(t->ctl_dev), MINOR(t->ctl_dev));
+}
diff -Naur linux-2.6.17.13-orig/drivers/md/Kconfig linux-2.6.17.13-dmu/drivers/md/Kconfig
--- linux-2.6.17.13-orig/drivers/md/Kconfig 2006-09-08 20:23:25.000000000 -0700
+++ linux-2.6.17.13-dmu/drivers/md/Kconfig 2006-09-15 14:20:48.000000000 -0700
@@ -237,6 +237,12 @@
---help---
Allow volume managers to take writeable snapshots of a device.
+config DM_USERSPACE
+ tristate "Userspace target (EXPERIMENTAL)"
+ depends on BLK_DEV_DM && EXPERIMENTAL
+ ---help---
+ A target that provides a userspace interface to device-mapper
+
config DM_MIRROR
tristate "Mirror target (EXPERIMENTAL)"
depends on BLK_DEV_DM && EXPERIMENTAL
diff -Naur linux-2.6.17.13-orig/drivers/md/Makefile linux-2.6.17.13-dmu/drivers/md/Makefile
--- linux-2.6.17.13-orig/drivers/md/Makefile 2006-09-08 20:23:25.000000000 -0700
+++ linux-2.6.17.13-dmu/drivers/md/Makefile 2006-09-15 14:20:48.000000000 -0700
@@ -14,6 +14,7 @@
raid6altivec1.o raid6altivec2.o raid6altivec4.o \
raid6altivec8.o \
raid6mmx.o raid6sse1.o raid6sse2.o
+dm-user-objs := dm-userspace.o dm-userspace-chardev.o
hostprogs-y := mktables
# Note: link order is important. All raid personalities
@@ -37,6 +38,7 @@
obj-$(CONFIG_DM_SNAPSHOT) += dm-snapshot.o
obj-$(CONFIG_DM_MIRROR) += dm-mirror.o
obj-$(CONFIG_DM_ZERO) += dm-zero.o
+obj-$(CONFIG_DM_USERSPACE) += dm-user.o
quiet_cmd_unroll = UNROLL $@
cmd_unroll = $(PERL) $(srctree)/$(src)/unroll.pl $(UNROLL) \
diff -Naur linux-2.6.17.13-orig/include/linux/dm-userspace.h linux-2.6.17.13-dmu/include/linux/dm-userspace.h
--- linux-2.6.17.13-orig/include/linux/dm-userspace.h 1969-12-31 16:00:00.000000000 -0800
+++ linux-2.6.17.13-dmu/include/linux/dm-userspace.h 2006-09-15 14:20:48.000000000 -0700
@@ -0,0 +1,151 @@
+/*
+ * Copyright (C) International Business Machines Corp., 2006
+ * Author: Dan Smith <danms at us.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; under version 2 of the License.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ *
+ */
+
+#ifndef __DM_USERSPACE_H
+#define __DM_USERSPACE_H
+
+#include <linux/types.h>
+
+/*
+ * Message Types
+ */
+#define DM_USERSPACE_GET_VERSION 1
+#define DM_USERSPACE_MAP_BLOCK_REQ 2
+#define DM_USERSPACE_MAP_BLOCK_RESP 3
+#define DM_USERSPACE_MAP_FAILED 4
+#define DM_USERSPACE_MAP_INVALIDATE 5
+#define DM_USERSPACE_STATUS 6
+
+/*
+ * Status codes
+ */
+#define DM_USERSPACE_INVAL_COMPLETE 101
+#define DM_USERSPACE_INVAL_FAILED 102
+#define DM_USERSPACE_SYNC_COMPLETE 103
+
+/*
+ * Flags and associated macros
+ */
+#define DMU_FLAG_VALID 1
+#define DMU_FLAG_RD 2
+#define DMU_FLAG_WR 4
+#define DMU_FLAG_COPY_FIRST 8
+#define DMU_FLAG_TEMPORARY 16
+#define DMU_FLAG_DONE 32
+#define DMU_FLAG_SYNC 64
+#define DMU_FLAG_WAITING 128
+
+static int dmu_get_flag(uint32_t *flags, uint32_t flag)
+{
+ return (*flags & flag) != 0;
+}
+
+static void dmu_set_flag(uint32_t *flags, uint32_t flag)
+{
+ *flags |= flag;
+}
+
+static void dmu_clr_flag(uint32_t *flags, uint32_t flag)
+{
+ *flags &= (~flag);
+}
+
+static void dmu_cpy_flag(uint32_t *flags, uint32_t src, uint32_t flag)
+{
+ *flags = (*flags & ~flag) | (src & flag);
+}
+
+/*
+ * This message header is sent in front of every message, in both
+ * directions
+ */
+struct dmu_msg_header {
+ uint64_t id;
+ uint32_t msg_type;
+ uint32_t payload_len;
+};
+
+/* DM_USERSPACE_GET_VERSION */
+struct dmu_msg_version {
+ uint32_t userspace_ver;
+ uint32_t kernel_ver;
+};
+
+/* For status codes */
+struct dmu_msg_status {
+ uint64_t id_of_op;
+ uint32_t status;
+};
+
+/* DM_USERSPACE_MAP_BLOCK_REQ */
+struct dmu_msg_map_request {
+ uint64_t org_block;
+
+ uint32_t flags;
+};
+
+/* DM_USERSPACE_MAP_BLOCK_RESP
+ * DM_USERSPACE_MAP_BLOCK_FAILED
+ */
+struct dmu_msg_map_response {
+ uint64_t org_block;
+ uint64_t new_block;
+ int64_t offset;
+
+ uint64_t id_of_req;
+ uint32_t flags;
+
+ uint32_t src_maj;
+ uint32_t src_min;
+
+ uint32_t dst_maj;
+ uint32_t dst_min;
+};
+
+/* A full message */
+struct dmu_msg {
+ struct dmu_msg_header hdr;
+ union {
+ struct dmu_msg_version ver;
+ struct dmu_msg_status status;
+ struct dmu_msg_map_request map_req;
+ struct dmu_msg_map_response map_rsp;
+ } payload;
+};
+
+static inline int dmu_get_msg_len(int type)
+{
+ switch (type) {
+ case DM_USERSPACE_GET_VERSION:
+ return sizeof(struct dmu_msg_version);
+ case DM_USERSPACE_INVAL_COMPLETE:
+ case DM_USERSPACE_INVAL_FAILED:
+ case DM_USERSPACE_STATUS:
+ return sizeof(struct dmu_msg_status);
+ case DM_USERSPACE_MAP_BLOCK_REQ:
+ return sizeof(struct dmu_msg_map_request);
+ case DM_USERSPACE_MAP_BLOCK_RESP:
+ case DM_USERSPACE_MAP_FAILED:
+ return sizeof(struct dmu_msg_map_response);
+ default:
+ return -1;
+ };
+}
+
+#endif
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 190 bytes
Desc: not available
URL: <http://listman.redhat.com/archives/dm-devel/attachments/20060918/2258ef5b/attachment.sig>
More information about the dm-devel
mailing list