[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-changelog] [linux-2.6.18-xen] blktap2: a completely rewritten blktap implementation



# HG changeset patch
# User Keir Fraser <keir.fraser@xxxxxxxxxx>
# Date 1243333396 -3600
# Node ID eba6fe6d8d53168cc5c3c8b6b209646a02ffd580
# Parent  f3a935eb30e0f6f1e42f7055e9cd32837fd34eac
blktap2: a completely rewritten blktap implementation

Benefits to blktap2 over the old version of blktap:

* Isolation from xenstore - Blktap devices are now created directly on
   the linux dom0 command line, rather than being spawned in response
   to XenStore events.  This is handy for debugging, makes blktap
   generally easier to work with, and is a step toward a generic
   user-level block device implementation that is not Xen-specific.

* Improved tapdisk infrastructure: simpler request forwarding, new
   request scheduler, request merging, more efficient use of AIO.

* Improved tapdisk error handling and memory management.  No
   allocations on the block data path, IO retry logic to protect
   guests
   transient block device failures.  This has been tested and is known
   to work on weird environments such as NFS soft mounts.

* Pause and snapshot of live virtual disks (see xmsnap script).

* VHD support.  The VHD code in this release has been rigorously
   tested, and represents a very mature implementation of the VHD
   image
   format.

* No more duplication of mechanism with blkback.  The blktap kernel
   module has changed dramatically from the original blktap.  Blkback
   is now always used to talk to Xen guests, blktap just presents a
   Linux gendisk that blkback can export.  This is done while
   preserving the zero-copy data path from domU to physical device.

These patches deprecate the old blktap code, which can hopefully be
removed from the tree completely at some point in the future.

Signed-off-by: Jake Wires <jake.wires@xxxxxxxxxx>
Signed-off-by: Dutch Meyer <dmeyer@xxxxxxxxx>
---
 drivers/xen/Makefile                  |    1 
 drivers/xen/blkback/Makefile          |    2 
 drivers/xen/blkback/blkback-pagemap.c |   91 ++
 drivers/xen/blkback/blkback-pagemap.h |   17 
 drivers/xen/blkback/blkback.c         |    8 
 drivers/xen/blkback/common.h          |    6 
 drivers/xen/blktap/blktap.c           |   26 
 drivers/xen/blktap2/Makefile          |    3 
 drivers/xen/blktap2/blktap.h          |  244 +++++++
 drivers/xen/blktap2/control.c         |  277 ++++++++
 drivers/xen/blktap2/device.c          | 1132 ++++++++++++++++++++++++++++++++++
 drivers/xen/blktap2/request.c         |  297 ++++++++
 drivers/xen/blktap2/ring.c            |  613 ++++++++++++++++++
 drivers/xen/blktap2/sysfs.c           |  425 ++++++++++++
 drivers/xen/blktap2/wait_queue.c      |   40 +
 include/linux/mm.h                    |    7 
 include/linux/page-flags.h            |   10 
 mm/memory.c                           |    4 
 mm/mmap.c                             |   12 
 19 files changed, 3199 insertions(+), 16 deletions(-)

diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/Makefile
--- a/drivers/xen/Makefile      Tue May 26 09:53:55 2009 +0100
+++ b/drivers/xen/Makefile      Tue May 26 11:23:16 2009 +0100
@@ -8,6 +8,7 @@ obj-$(CONFIG_XEN_BALLOON)               += balloon/
 obj-$(CONFIG_XEN_BALLOON)              += balloon/
 obj-$(CONFIG_XEN_BLKDEV_BACKEND)       += blkback/
 obj-$(CONFIG_XEN_BLKDEV_TAP)           += blktap/
+obj-$(CONFIG_XEN_BLKDEV_TAP)            += blktap2/
 obj-$(CONFIG_XEN_NETDEV_BACKEND)       += netback/
 obj-$(CONFIG_XEN_TPMDEV_BACKEND)       += tpmback/
 obj-$(CONFIG_XEN_BLKDEV_FRONTEND)      += blkfront/
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blkback/Makefile
--- a/drivers/xen/blkback/Makefile      Tue May 26 09:53:55 2009 +0100
+++ b/drivers/xen/blkback/Makefile      Tue May 26 11:23:16 2009 +0100
@@ -1,3 +1,3 @@ obj-$(CONFIG_XEN_BLKDEV_BACKEND) := blkb
 obj-$(CONFIG_XEN_BLKDEV_BACKEND) := blkbk.o
 
-blkbk-y        := blkback.o xenbus.o interface.o vbd.o
+blkbk-y        := blkback.o xenbus.o interface.o vbd.o blkback-pagemap.o
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blkback/blkback-pagemap.c
--- /dev/null   Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/xen/blkback/blkback-pagemap.c     Tue May 26 11:23:16 2009 +0100
@@ -0,0 +1,91 @@
+#include "common.h"
+#include "blkback-pagemap.h"
+
+static int blkback_pagemap_size;
+static struct blkback_pagemap *blkback_pagemap;
+
+static inline int
+blkback_pagemap_entry_clear(struct blkback_pagemap *map)
+{
+       static struct blkback_pagemap zero;
+       return !memcmp(map, &zero, sizeof(zero));
+}
+
+int
+blkback_pagemap_init(int pages)
+{
+       blkback_pagemap = kzalloc(pages * sizeof(struct blkback_pagemap),
+                                 GFP_KERNEL);
+       if (!blkback_pagemap)
+               return -ENOMEM;
+
+       blkback_pagemap_size = pages;
+       return 0;
+}
+
+void
+blkback_pagemap_set(int idx, struct page *page,
+                   domid_t domid, busid_t busid, grant_ref_t gref)
+{
+       struct blkback_pagemap *entry;
+
+       BUG_ON(!blkback_pagemap);
+       BUG_ON(idx >= blkback_pagemap_size);
+
+       SetPageBlkback(page);
+       set_page_private(page, idx);
+
+       entry = blkback_pagemap + idx;
+       if (!blkback_pagemap_entry_clear(entry)) {
+               printk("overwriting pagemap %d: d %u b %u g %u\n",
+                      idx, entry->domid, entry->busid, entry->gref);
+               BUG();
+       }
+
+       entry->domid = domid;
+       entry->busid = busid;
+       entry->gref  = gref;
+}
+
+void
+blkback_pagemap_clear(struct page *page)
+{
+       int idx;
+       struct blkback_pagemap *entry;
+
+       idx = (int)page_private(page);
+
+       BUG_ON(!blkback_pagemap);
+       BUG_ON(!PageBlkback(page));
+       BUG_ON(idx >= blkback_pagemap_size);
+
+       entry = blkback_pagemap + idx;
+       if (blkback_pagemap_entry_clear(entry)) {
+               printk("clearing empty pagemap %d\n", idx);
+               BUG();
+       }
+
+       memset(entry, 0, sizeof(*entry));
+}
+
+struct blkback_pagemap
+blkback_pagemap_read(struct page *page)
+{
+       int idx;
+       struct blkback_pagemap *entry;
+
+       idx = (int)page_private(page);
+
+       BUG_ON(!blkback_pagemap);
+       BUG_ON(!PageBlkback(page));
+       BUG_ON(idx >= blkback_pagemap_size);
+
+       entry = blkback_pagemap + idx;
+       if (blkback_pagemap_entry_clear(entry)) {
+               printk("reading empty pagemap %d\n", idx);
+               BUG();
+       }
+
+       return *entry;
+}
+EXPORT_SYMBOL(blkback_pagemap_read);
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blkback/blkback-pagemap.h
--- /dev/null   Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/xen/blkback/blkback-pagemap.h     Tue May 26 11:23:16 2009 +0100
@@ -0,0 +1,17 @@
+#ifndef _BLKBACK_PAGEMAP_H_
+#define _BLKBACK_PAGEMAP_H_
+
+#include <xen/interface/xen.h>
+#include <xen/interface/grant_table.h>
+
+typedef unsigned int busid_t;
+
+struct blkback_pagemap {
+       domid_t          domid;
+       busid_t          busid;
+       grant_ref_t      gref;
+};
+
+struct blkback_pagemap blkback_pagemap_read(struct page *);
+
+#endif
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blkback/blkback.c
--- a/drivers/xen/blkback/blkback.c     Tue May 26 09:53:55 2009 +0100
+++ b/drivers/xen/blkback/blkback.c     Tue May 26 11:23:16 2009 +0100
@@ -173,6 +173,7 @@ static void fast_flush_area(pending_req_
                handle = pending_handle(req, i);
                if (handle == BLKBACK_INVALID_HANDLE)
                        continue;
+               blkback_pagemap_clear(virt_to_page(vaddr(req, i)));
                gnttab_set_unmap_op(&unmap[invcount], vaddr(req, i),
                                    GNTMAP_host_map, handle);
                pending_handle(req, i) = BLKBACK_INVALID_HANDLE;
@@ -464,6 +465,10 @@ static void dispatch_rw_block_io(blkif_t
                        FOREIGN_FRAME(map[i].dev_bus_addr >> PAGE_SHIFT));
                seg[i].buf  = map[i].dev_bus_addr | 
                        (req->seg[i].first_sect << 9);
+               blkback_pagemap_set(vaddr_pagenr(pending_req, i),
+                                   virt_to_page(vaddr(pending_req, i)),
+                                   blkif->domid, req->handle,
+                                   req->seg[i].gref);
        }
 
        if (ret)
@@ -625,6 +630,9 @@ static int __init blkif_init(void)
                                        mmap_pages, GFP_KERNEL);
        pending_pages         = alloc_empty_pages_and_pagevec(mmap_pages);
 
+       if (blkback_pagemap_init(mmap_pages))
+               goto out_of_memory;
+
        if (!pending_reqs || !pending_grant_handles || !pending_pages)
                goto out_of_memory;
 
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blkback/common.h
--- a/drivers/xen/blkback/common.h      Tue May 26 09:53:55 2009 +0100
+++ b/drivers/xen/blkback/common.h      Tue May 26 11:23:16 2009 +0100
@@ -43,6 +43,8 @@
 #include <xen/gnttab.h>
 #include <xen/driver_util.h>
 #include <xen/xenbus.h>
+#include "blkback-pagemap.h"
+
 
 #define DPRINTK(_f, _a...)                     \
        pr_debug("(file=%s, line=%d) " _f,      \
@@ -136,4 +138,8 @@ int blkback_barrier(struct xenbus_transa
 int blkback_barrier(struct xenbus_transaction xbt,
                    struct backend_info *be, int state);
 
+int blkback_pagemap_init(int);
+void blkback_pagemap_set(int, struct page *, domid_t, busid_t, grant_ref_t);
+void blkback_pagemap_clear(struct page *);
+
 #endif /* __BLKIF__BACKEND__COMMON_H__ */
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blktap/blktap.c
--- a/drivers/xen/blktap/blktap.c       Tue May 26 09:53:55 2009 +0100
+++ b/drivers/xen/blktap/blktap.c       Tue May 26 11:23:16 2009 +0100
@@ -116,7 +116,7 @@ typedef struct tap_blkif {
                                        [req id, idx] tuple                  */
        blkif_t *blkif;               /*Associate blkif with tapdev          */
        struct domid_translate_ext trans; /*Translation from domid to bus.   */
-       struct page **map;            /*Mapping page */
+       struct vm_foreign_map foreign_map;    /*Mapping page */
 } tap_blkif_t;
 
 static struct tap_blkif *tapfds[MAX_TAP_DEV];
@@ -347,7 +347,7 @@ static pte_t blktap_clear_pte(struct vm_
        kvaddr = idx_to_kaddr(mmap_idx, pending_idx, seg);
        pg = pfn_to_page(__pa(kvaddr) >> PAGE_SHIFT);
        ClearPageReserved(pg);
-       info->map[offset + RING_PAGES] = NULL;
+       info->foreign_map.map[offset + RING_PAGES] = NULL;
 
        khandle = &pending_handle(mmap_idx, pending_idx, seg);
 
@@ -396,7 +396,7 @@ static void blktap_vma_open(struct vm_ar
 
        info = vma->vm_file->private_data;
        vma->vm_private_data =
-               &info->map[(vma->vm_start - info->rings_vstart) >> PAGE_SHIFT];
+               &info->foreign_map.map[(vma->vm_start - info->rings_vstart) >> 
PAGE_SHIFT];
 }
 
 /* tricky part
@@ -418,7 +418,7 @@ static void blktap_vma_close(struct vm_a
 
        info = vma->vm_file->private_data;
        next->vm_private_data =
-               &info->map[(next->vm_start - info->rings_vstart) >> PAGE_SHIFT];
+               &info->foreign_map.map[(next->vm_start - info->rings_vstart) >> 
PAGE_SHIFT];
 }
 
 static struct vm_operations_struct blktap_vm_ops = {
@@ -642,8 +642,8 @@ static int blktap_release(struct inode *
 
        mmput(info->mm);
        info->mm = NULL;
-       kfree(info->map);
-       info->map = NULL;
+       kfree(info->foreign_map.map);
+       info->foreign_map.map = NULL;
 
        /* Free the ring page. */
        ClearPageReserved(virt_to_page(info->ufe_ring.sring));
@@ -726,14 +726,14 @@ static int blktap_mmap(struct file *filp
        }
 
        /* Mark this VM as containing foreign pages, and set up mappings. */
-       info->map = kzalloc(((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) *
-                           sizeof(*info->map), GFP_KERNEL);
-       if (info->map == NULL) {
+       info->foreign_map.map = kzalloc(((vma->vm_end - vma->vm_start) >> 
PAGE_SHIFT) *
+                           sizeof(*info->foreign_map.map), GFP_KERNEL);
+       if (info->foreign_map.map == NULL) {
                WPRINTK("Couldn't alloc VM_FOREIGN map.\n");
                goto fail;
        }
 
-       vma->vm_private_data = info->map;
+       vma->vm_private_data = info->foreign_map.map;
        vma->vm_flags |= VM_FOREIGN;
        vma->vm_flags |= VM_DONTCOPY;
 
@@ -1238,7 +1238,7 @@ static int blktap_read_ufe_ring(tap_blki
                        pg = pfn_to_page(__pa(kvaddr) >> PAGE_SHIFT);
                        ClearPageReserved(pg);
                        offset = (uvaddr - info->rings_vstart) >> PAGE_SHIFT;
-                       info->map[offset] = NULL;
+                       info->foreign_map.map[offset] = NULL;
                }
                fast_flush_area(pending_req, pending_idx, usr_idx, info->minor);
                info->idx_map[usr_idx] = INVALID_REQ;
@@ -1530,7 +1530,7 @@ static void dispatch_rw_block_io(blkif_t
                                                          >> PAGE_SHIFT));
                        offset = (uvaddr - info->rings_vstart) >> PAGE_SHIFT;
                        pg = pfn_to_page(__pa(kvaddr) >> PAGE_SHIFT);
-                       info->map[offset] = pg;
+                       info->foreign_map.map[offset] = pg;
                }
        } else {
                for (i = 0; i < nseg; i++) {
@@ -1557,7 +1557,7 @@ static void dispatch_rw_block_io(blkif_t
 
                        offset = (uvaddr - info->rings_vstart) >> PAGE_SHIFT;
                        pg = pfn_to_page(__pa(kvaddr) >> PAGE_SHIFT);
-                       info->map[offset] = pg;
+                       info->foreign_map.map[offset] = pg;
                }
        }
 
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blktap2/Makefile
--- /dev/null   Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/xen/blktap2/Makefile      Tue May 26 11:23:16 2009 +0100
@@ -0,0 +1,3 @@
+obj-y := blktap.o
+
+blktap-objs := control.o ring.o wait_queue.o device.o request.o sysfs.o
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blktap2/blktap.h
--- /dev/null   Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/xen/blktap2/blktap.h      Tue May 26 11:23:16 2009 +0100
@@ -0,0 +1,244 @@
+#ifndef _BLKTAP_H_
+#define _BLKTAP_H_
+
+#include <linux/fs.h>
+#include <linux/poll.h>
+#include <linux/cdev.h>
+#include <xen/blkif.h>
+#include <xen/gnttab.h>
+
+//#define ENABLE_PASSTHROUGH
+
+extern int blktap_debug_level;
+
+#define BTPRINTK(level, tag, force, _f, _a...)                         \
+       do {                                                            \
+               if (blktap_debug_level > level &&                       \
+                   (force || printk_ratelimit()))                      \
+                       printk(tag "%s: " _f, __func__, ##_a);          \
+       } while (0)
+
+#define BTDBG(_f, _a...)             BTPRINTK(8, KERN_DEBUG, 1, _f, ##_a)
+#define BTINFO(_f, _a...)            BTPRINTK(0, KERN_INFO, 0, _f, ##_a)
+#define BTWARN(_f, _a...)            BTPRINTK(0, KERN_WARNING, 0, _f, ##_a)
+#define BTERR(_f, _a...)             BTPRINTK(0, KERN_ERR, 0, _f, ##_a)
+
+#define MAX_BLKTAP_DEVICE            256
+
+#define BLKTAP_CONTROL               1
+#define BLKTAP_RING_FD               2
+#define BLKTAP_RING_VMA              3
+#define BLKTAP_DEVICE                4
+#define BLKTAP_SYSFS                 5
+#define BLKTAP_PAUSE_REQUESTED       6
+#define BLKTAP_PAUSED                7
+#define BLKTAP_SHUTDOWN_REQUESTED    8
+#define BLKTAP_PASSTHROUGH           9
+#define BLKTAP_DEFERRED              10
+
+/* blktap IOCTLs: */
+#define BLKTAP2_IOCTL_KICK_FE        1
+#define BLKTAP2_IOCTL_ALLOC_TAP             200
+#define BLKTAP2_IOCTL_FREE_TAP       201
+#define BLKTAP2_IOCTL_CREATE_DEVICE  202
+#define BLKTAP2_IOCTL_SET_PARAMS     203
+#define BLKTAP2_IOCTL_PAUSE          204
+#define BLKTAP2_IOCTL_REOPEN         205
+#define BLKTAP2_IOCTL_RESUME         206
+
+#define BLKTAP2_MAX_MESSAGE_LEN      256
+
+#define BLKTAP2_RING_MESSAGE_PAUSE   1
+#define BLKTAP2_RING_MESSAGE_RESUME  2
+#define BLKTAP2_RING_MESSAGE_CLOSE   3
+
+#define BLKTAP_REQUEST_FREE          0
+#define BLKTAP_REQUEST_PENDING       1
+
+/*
+ * The maximum number of requests that can be outstanding at any time
+ * is determined by
+ *
+ *   [mmap_alloc * MAX_PENDING_REQS * BLKIF_MAX_SEGMENTS_PER_REQUEST] 
+ *
+ * where mmap_alloc < MAX_DYNAMIC_MEM.
+ *
+ * TODO:
+ * mmap_alloc is initialised to 2 and should be adjustable on the fly via
+ * sysfs.
+ */
+#define BLK_RING_SIZE          __RING_SIZE((blkif_sring_t *)0, PAGE_SIZE)
+#define MAX_DYNAMIC_MEM                BLK_RING_SIZE
+#define MAX_PENDING_REQS       BLK_RING_SIZE
+#define MMAP_PAGES (MAX_PENDING_REQS * BLKIF_MAX_SEGMENTS_PER_REQUEST)
+#define MMAP_VADDR(_start, _req, _seg)                                 \
+        (_start +                                                       \
+         ((_req) * BLKIF_MAX_SEGMENTS_PER_REQUEST * PAGE_SIZE) +        \
+         ((_seg) * PAGE_SIZE))
+
+#define blktap_get(_b) (atomic_inc(&(_b)->refcnt))
+#define blktap_put(_b)                                 \
+       do {                                            \
+               if (atomic_dec_and_test(&(_b)->refcnt)) \
+                       wake_up(&(_b)->wq);             \
+       } while (0)
+
+struct blktap;
+
+struct grant_handle_pair {
+       grant_handle_t                 kernel;
+       grant_handle_t                 user;
+};
+#define INVALID_GRANT_HANDLE           0xFFFF
+
+struct blktap_handle {
+       unsigned int                   ring;
+       unsigned int                   device;
+       unsigned int                   minor;
+};
+
+struct blktap_params {
+       char                           name[BLKTAP2_MAX_MESSAGE_LEN];
+       unsigned long long             capacity;
+       unsigned long                  sector_size;
+};
+
+struct blktap_device {
+       int                            users;
+       spinlock_t                     lock;
+       struct gendisk                *gd;
+
+#ifdef ENABLE_PASSTHROUGH
+       struct block_device           *bdev;
+#endif
+};
+
+struct blktap_ring {
+       struct vm_area_struct         *vma;
+       blkif_front_ring_t             ring;
+       struct vm_foreign_map          foreign_map;
+       unsigned long                  ring_vstart;
+       unsigned long                  user_vstart;
+
+       int                            response;
+
+       wait_queue_head_t              poll_wait;
+
+       dev_t                          devno;
+       struct class_device           *dev;
+       atomic_t                       sysfs_refcnt;
+       struct mutex                   sysfs_mutex;
+};
+
+struct blktap_statistics {
+       unsigned long                  st_print;
+       int                            st_rd_req;
+       int                            st_wr_req;
+       int                            st_oo_req;
+       int                            st_rd_sect;
+       int                            st_wr_sect;
+       s64                            st_rd_cnt;
+       s64                            st_rd_sum_usecs;
+       s64                            st_rd_max_usecs;
+       s64                            st_wr_cnt;
+       s64                            st_wr_sum_usecs;
+       s64                            st_wr_max_usecs; 
+};
+
+struct blktap_request {
+       uint64_t                       id;
+       uint16_t                       usr_idx;
+
+       uint8_t                        status;
+       atomic_t                       pendcnt;
+       uint8_t                        nr_pages;
+       unsigned short                 operation;
+
+       struct timeval                 time;
+       struct grant_handle_pair       handles[BLKIF_MAX_SEGMENTS_PER_REQUEST];
+       struct list_head               free_list;
+};
+
+struct blktap {
+       int                            minor;
+       pid_t                          pid;
+       atomic_t                       refcnt;
+       unsigned long                  dev_inuse;
+
+       struct blktap_params           params;
+
+       struct rw_semaphore            tap_sem;
+
+       struct blktap_ring             ring;
+       struct blktap_device           device;
+
+       int                            pending_cnt;
+       struct blktap_request         *pending_requests[MAX_PENDING_REQS];
+
+       wait_queue_head_t              wq;
+       struct list_head               deferred_queue;
+
+       struct blktap_statistics       stats;
+};
+
+extern struct blktap *blktaps[MAX_BLKTAP_DEVICE];
+
+static inline int
+blktap_active(struct blktap *tap)
+{
+       return test_bit(BLKTAP_RING_VMA, &tap->dev_inuse);
+}
+
+static inline int
+blktap_validate_params(struct blktap *tap, struct blktap_params *params)
+{
+       /* TODO: sanity check */
+       params->name[sizeof(params->name) - 1] = '\0';
+       BTINFO("%s: capacity: %llu, sector-size: %lu\n",
+              params->name, params->capacity, params->sector_size);
+       return 0;
+}
+
+int blktap_control_destroy_device(struct blktap *);
+
+int blktap_ring_init(int *);
+int blktap_ring_free(void);
+int blktap_ring_create(struct blktap *);
+int blktap_ring_destroy(struct blktap *);
+int blktap_ring_pause(struct blktap *);
+int blktap_ring_resume(struct blktap *);
+void blktap_ring_kick_user(struct blktap *);
+
+int blktap_sysfs_init(void);
+void blktap_sysfs_free(void);
+int blktap_sysfs_create(struct blktap *);
+int blktap_sysfs_destroy(struct blktap *);
+
+int blktap_device_init(int *);
+void blktap_device_free(void);
+int blktap_device_create(struct blktap *);
+int blktap_device_destroy(struct blktap *);
+int blktap_device_pause(struct blktap *);
+int blktap_device_resume(struct blktap *);
+void blktap_device_restart(struct blktap *);
+void blktap_device_finish_request(struct blktap *,
+                                 blkif_response_t *,
+                                 struct blktap_request *);
+void blktap_device_fail_pending_requests(struct blktap *);
+#ifdef ENABLE_PASSTHROUGH
+int blktap_device_enable_passthrough(struct blktap *,
+                                    unsigned, unsigned);
+#endif
+
+void blktap_defer(struct blktap *);
+void blktap_run_deferred(void);
+
+int blktap_request_pool_init(void);
+void blktap_request_pool_free(void);
+int blktap_request_pool_grow(void);
+int blktap_request_pool_shrink(void);
+struct blktap_request *blktap_request_allocate(struct blktap *);
+void blktap_request_free(struct blktap *, struct blktap_request *);
+unsigned long request_to_kaddr(struct blktap_request *, int);
+
+#endif
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blktap2/control.c
--- /dev/null   Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/xen/blktap2/control.c     Tue May 26 11:23:16 2009 +0100
@@ -0,0 +1,277 @@
+#include <linux/module.h>
+#include <linux/miscdevice.h>
+
+#include "blktap.h"
+
+static DEFINE_SPINLOCK(blktap_control_lock);
+struct blktap *blktaps[MAX_BLKTAP_DEVICE];
+
+static int ring_major;
+static int device_major;
+static int blktap_control_registered;
+
+static void
+blktap_control_initialize_tap(struct blktap *tap)
+{
+       int minor = tap->minor;
+
+       memset(tap, 0, sizeof(*tap));
+       set_bit(BLKTAP_CONTROL, &tap->dev_inuse);
+       init_rwsem(&tap->tap_sem);
+       init_waitqueue_head(&tap->wq);
+       atomic_set(&tap->refcnt, 0);
+
+       tap->minor = minor;
+}
+
+static struct blktap *
+blktap_control_create_tap(void)
+{
+       int minor;
+       struct blktap *tap;
+
+       tap = kmalloc(sizeof(*tap), GFP_KERNEL);
+       if (unlikely(!tap))
+               return NULL;
+
+       blktap_control_initialize_tap(tap);
+
+       spin_lock_irq(&blktap_control_lock);
+       for (minor = 0; minor < MAX_BLKTAP_DEVICE; minor++)
+               if (!blktaps[minor])
+                       break;
+
+       if (minor == MAX_BLKTAP_DEVICE) {
+               kfree(tap);
+               tap = NULL;
+               goto out;
+       }
+
+       tap->minor = minor;
+       blktaps[minor] = tap;
+
+out:
+       spin_unlock_irq(&blktap_control_lock);
+       return tap;
+}
+
+static struct blktap *
+blktap_control_allocate_tap(void)
+{
+       int err, minor;
+       struct blktap *tap;
+
+       /*
+        * This is called only from the ioctl, which
+        * means we should always have interrupts enabled.
+        */
+       BUG_ON(irqs_disabled());
+
+       spin_lock_irq(&blktap_control_lock);
+
+       for (minor = 0; minor < MAX_BLKTAP_DEVICE; minor++) {
+               tap = blktaps[minor];
+               if (!tap)
+                       goto found;
+
+               if (!tap->dev_inuse) {
+                       blktap_control_initialize_tap(tap);
+                       goto found;
+               }
+       }
+
+       tap = NULL;
+
+found:
+       spin_unlock_irq(&blktap_control_lock);
+
+       if (!tap) {
+               tap = blktap_control_create_tap();
+               if (!tap)
+                       return NULL;
+       }
+
+       err = blktap_ring_create(tap);
+       if (err) {
+               BTERR("ring creation failed: %d\n", err);
+               clear_bit(BLKTAP_CONTROL, &tap->dev_inuse);
+               return NULL;
+       }
+
+       BTINFO("allocated tap %p\n", tap);
+       return tap;
+}
+
+static int
+blktap_control_ioctl(struct inode *inode, struct file *filp,
+                    unsigned int cmd, unsigned long arg)
+{
+       unsigned long dev;
+       struct blktap *tap;
+
+       switch (cmd) {
+       case BLKTAP2_IOCTL_ALLOC_TAP: {
+               struct blktap_handle h;
+
+               tap = blktap_control_allocate_tap();
+               if (!tap) {
+                       BTERR("error allocating device\n");
+                       return -ENOMEM;
+               }
+
+               h.ring   = ring_major;
+               h.device = device_major;
+               h.minor  = tap->minor;
+
+               if (copy_to_user((struct blktap_handle __user *)arg,
+                                &h, sizeof(h))) {
+                       blktap_control_destroy_device(tap);
+                       return -EFAULT;
+               }
+
+               return 0;
+       }
+
+       case BLKTAP2_IOCTL_FREE_TAP:
+               dev = arg;
+
+               if (dev > MAX_BLKTAP_DEVICE || !blktaps[dev])
+                       return -EINVAL;
+
+               blktap_control_destroy_device(blktaps[dev]);
+               return 0;
+       }
+
+       return -ENOIOCTLCMD;
+}
+
+static struct file_operations blktap_control_file_operations = {
+       .owner    = THIS_MODULE,
+       .ioctl    = blktap_control_ioctl,
+};
+
+static struct miscdevice blktap_misc = {
+       .minor    = MISC_DYNAMIC_MINOR,
+       .name     = "blktap-control",
+       .fops     = &blktap_control_file_operations,
+};
+
+int
+blktap_control_destroy_device(struct blktap *tap)
+{
+       int err;
+       unsigned long inuse;
+
+       if (!tap)
+               return 0;
+
+       set_bit(BLKTAP_SHUTDOWN_REQUESTED, &tap->dev_inuse);
+
+       for (;;) {
+               inuse = tap->dev_inuse;
+               err   = blktap_device_destroy(tap);
+               if (err)
+                       goto wait;
+
+               inuse = tap->dev_inuse;
+               err   = blktap_ring_destroy(tap);
+               if (err)
+                       goto wait;
+
+               inuse = tap->dev_inuse;
+               err   = blktap_sysfs_destroy(tap);
+               if (err)
+                       goto wait;
+
+               break;
+
+       wait:
+               BTDBG("inuse: 0x%lx, dev_inuse: 0x%lx\n",
+                     inuse, tap->dev_inuse);
+               if (wait_event_interruptible(tap->wq, tap->dev_inuse != inuse))
+                       break;
+       }
+
+       clear_bit(BLKTAP_SHUTDOWN_REQUESTED, &tap->dev_inuse);
+
+       if (tap->dev_inuse == (1UL << BLKTAP_CONTROL)) {
+               err = 0;
+               clear_bit(BLKTAP_CONTROL, &tap->dev_inuse);
+       }
+
+       return err;
+}
+
+static int
+blktap_control_init(void)
+{
+       int err;
+
+       err = misc_register(&blktap_misc);
+       if (err) {
+               BTERR("misc_register failed for control device");
+               return err;
+       }
+
+       blktap_control_registered = 1;
+       return 0;
+}
+
+static void
+blktap_control_free(void)
+{
+       int i;
+
+       for (i = 0; i < MAX_BLKTAP_DEVICE; i++)
+               blktap_control_destroy_device(blktaps[i]);
+
+       if (blktap_control_registered)
+               if (misc_deregister(&blktap_misc) < 0)
+                       BTERR("misc_deregister failed for control device");
+}
+
+static void
+blktap_exit(void)
+{
+       blktap_control_free();
+       blktap_ring_free();
+       blktap_sysfs_free();
+       blktap_device_free();
+       blktap_request_pool_free();
+}
+
+static int __init
+blktap_init(void)
+{
+       int err;
+
+       err = blktap_request_pool_init();
+       if (err)
+               return err;
+
+       err = blktap_device_init(&device_major);
+       if (err)
+               goto fail;
+
+       err = blktap_ring_init(&ring_major);
+       if (err)
+               goto fail;
+
+       err = blktap_sysfs_init();
+       if (err)
+               goto fail;
+
+       err = blktap_control_init();
+       if (err)
+               goto fail;
+
+       return 0;
+
+fail:
+       blktap_exit();
+       return err;
+}
+
+module_init(blktap_init);
+module_exit(blktap_exit);
+MODULE_LICENSE("Dual BSD/GPL");
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blktap2/device.c
--- /dev/null   Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/xen/blktap2/device.c      Tue May 26 11:23:16 2009 +0100
@@ -0,0 +1,1132 @@
+#include <linux/fs.h>
+#include <linux/blkdev.h>
+#include <linux/cdrom.h>
+#include <linux/hdreg.h>
+#include <linux/module.h>
+
+#include <scsi/scsi.h>
+#include <scsi/scsi_ioctl.h>
+
+#include <xen/xenbus.h>
+#include <xen/interface/io/blkif.h>
+
+#include "blktap.h"
+
+#ifdef CONFIG_XEN_BLKDEV_BACKEND
+#include "../blkback/blkback-pagemap.h"
+#else
+struct blkback_pagemap { };
+#define blkback_pagemap_read(page) BUG();
+#endif
+
+#if 0
+#define DPRINTK_IOCTL(_f, _a...) printk(KERN_ALERT _f, ## _a)
+#else
+#define DPRINTK_IOCTL(_f, _a...) ((void)0)
+#endif
+
+struct blktap_grant_table {
+       int cnt;
+       struct gnttab_map_grant_ref grants[BLKIF_MAX_SEGMENTS_PER_REQUEST * 2];
+};
+
+static int blktap_device_major;
+
+static inline struct blktap *
+dev_to_blktap(struct blktap_device *dev)
+{
+       return container_of(dev, struct blktap, device);
+}
+
+static int
+blktap_device_open(struct inode *inode, struct file *filep)
+{
+       struct blktap *tap;
+       struct blktap_device *dev = inode->i_bdev->bd_disk->private_data;
+
+       if (!dev)
+               return -ENOENT;
+
+       tap = dev_to_blktap(dev);
+       if (!blktap_active(tap) ||
+           test_bit(BLKTAP_SHUTDOWN_REQUESTED, &tap->dev_inuse))
+               return -ENOENT;
+
+       dev->users++;
+
+       return 0;
+}
+
+static int
+blktap_device_release(struct inode *inode, struct file *filep)
+{
+       struct blktap_device *dev = inode->i_bdev->bd_disk->private_data;
+       struct blktap *tap = dev_to_blktap(dev);
+
+       dev->users--;
+       if (test_bit(BLKTAP_SHUTDOWN_REQUESTED, &tap->dev_inuse))
+               blktap_device_destroy(tap);
+
+       return 0;
+}
+
+static int
+blktap_device_getgeo(struct block_device *bd, struct hd_geometry *hg)
+{
+       /* We don't have real geometry info, but let's at least return
+          values consistent with the size of the device */
+       sector_t nsect = get_capacity(bd->bd_disk);
+       sector_t cylinders = nsect;
+
+       hg->heads = 0xff;
+       hg->sectors = 0x3f;
+       sector_div(cylinders, hg->heads * hg->sectors);
+       hg->cylinders = cylinders;
+       if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
+               hg->cylinders = 0xffff;
+       return 0;
+}
+
+static int
+blktap_device_ioctl(struct inode *inode, struct file *filep,
+                   unsigned command, unsigned long argument)
+{
+       int i;
+
+       DPRINTK_IOCTL("command: 0x%x, argument: 0x%lx, dev: 0x%04x\n",
+                     command, (long)argument, inode->i_rdev);
+
+       switch (command) {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,16)
+       case HDIO_GETGEO: {
+               struct block_device *bd = inode->i_bdev;
+               struct hd_geometry geo;
+               int ret;
+
+                if (!argument)
+                        return -EINVAL;
+
+               geo.start = get_start_sect(bd);
+               ret = blktap_device_getgeo(bd, &geo);
+               if (ret)
+                       return ret;
+
+               if (copy_to_user((struct hd_geometry __user *)argument, &geo,
+                                sizeof(geo)))
+                        return -EFAULT;
+
+                return 0;
+       }
+#endif
+       case CDROMMULTISESSION:
+               BTDBG("FIXME: support multisession CDs later\n");
+               for (i = 0; i < sizeof(struct cdrom_multisession); i++)
+                       if (put_user(0, (char __user *)(argument + i)))
+                               return -EFAULT;
+               return 0;
+
+       case SCSI_IOCTL_GET_IDLUN:
+               if (!access_ok(VERIFY_WRITE, argument, 
+                       sizeof(struct scsi_idlun)))
+                       return -EFAULT;
+
+               /* return 0 for now. */
+               __put_user(0, &((struct scsi_idlun __user *)argument)->dev_id);
+               __put_user(0, 
+                       &((struct scsi_idlun __user 
*)argument)->host_unique_id);
+               return 0;
+
+       default:
+               /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
+                 command);*/
+               return -EINVAL; /* same return as native Linux */
+       }
+
+       return 0;
+}
+
+static struct block_device_operations blktap_device_file_operations = {
+       .owner     = THIS_MODULE,
+       .open      = blktap_device_open,
+       .release   = blktap_device_release,
+       .ioctl     = blktap_device_ioctl,
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,16)
+       .getgeo    = blktap_device_getgeo
+#endif
+};
+
+static int
+blktap_map_uaddr_fn(pte_t *ptep, struct page *pmd_page,
+                   unsigned long addr, void *data)
+{
+       pte_t *pte = (pte_t *)data;
+
+       BTDBG("ptep %p -> %012llx\n", ptep, pte_val(*pte));
+       set_pte(ptep, *pte);
+       xen_invlpg(addr);
+       return 0;
+}
+
+static int
+blktap_map_uaddr(struct mm_struct *mm, unsigned long address, pte_t pte)
+{
+       return apply_to_page_range(mm, address,
+                                  PAGE_SIZE, blktap_map_uaddr_fn, &pte);
+}
+
+static int
+blktap_umap_uaddr_fn(pte_t *ptep, struct page *pmd_page,
+                    unsigned long addr, void *data)
+{
+       struct mm_struct *mm = (struct mm_struct *)data;
+
+       BTDBG("ptep %p\n", ptep);
+       pte_clear(mm, addr, ptep);
+       xen_invlpg(addr);
+       return 0;
+}
+
+static int
+blktap_umap_uaddr(struct mm_struct *mm, unsigned long address)
+{
+       return apply_to_page_range(mm, address,
+                                  PAGE_SIZE, blktap_umap_uaddr_fn, mm);
+}
+
+static void
+blktap_device_end_dequeued_request(struct blktap_device *dev,
+                                  struct request *req, int uptodate)
+{
+       int ret;
+
+       ret = end_that_request_first(req, uptodate, req->hard_nr_sectors);
+       BUG_ON(ret);
+
+       spin_lock_irq(&dev->lock);
+       end_that_request_last(req, uptodate);
+       spin_unlock_irq(&dev->lock);
+}
+
+/*
+ * tap->tap_sem held on entry
+ */
+static void
+blktap_device_fast_flush(struct blktap *tap, struct blktap_request *request)
+{
+       uint64_t ptep;
+       int ret, usr_idx;
+       unsigned int i, cnt;
+       struct page **map, *page;
+       struct blktap_ring *ring;
+       struct grant_handle_pair *khandle;
+       unsigned long kvaddr, uvaddr, offset;
+       struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST * 2];
+
+       cnt     = 0;
+       ring    = &tap->ring;
+       usr_idx = request->usr_idx;
+       map     = ring->foreign_map.map;
+
+       if (!ring->vma)
+               return;
+
+       if (xen_feature(XENFEAT_auto_translated_physmap))
+               zap_page_range(ring->vma, 
+                              MMAP_VADDR(ring->user_vstart, usr_idx, 0),
+                              request->nr_pages << PAGE_SHIFT, NULL);
+
+       for (i = 0; i < request->nr_pages; i++) {
+               kvaddr = request_to_kaddr(request, i);
+               uvaddr = MMAP_VADDR(ring->user_vstart, usr_idx, i);
+
+               khandle = request->handles + i;
+
+               if (khandle->kernel != INVALID_GRANT_HANDLE) {
+                       gnttab_set_unmap_op(&unmap[cnt], kvaddr,
+                                           GNTMAP_host_map, khandle->kernel);
+                       cnt++;
+                       set_phys_to_machine(__pa(kvaddr) >> PAGE_SHIFT,
+                                           INVALID_P2M_ENTRY);
+               }
+
+               if (khandle->user != INVALID_GRANT_HANDLE) {
+                       BUG_ON(xen_feature(XENFEAT_auto_translated_physmap));
+                       if (create_lookup_pte_addr(ring->vma->vm_mm,
+                                                  uvaddr, &ptep) != 0) {
+                               BTERR("Couldn't get a pte addr!\n");
+                               return;
+                       }
+
+                       gnttab_set_unmap_op(&unmap[cnt], ptep,
+                                           GNTMAP_host_map
+                                           | GNTMAP_application_map
+                                           | GNTMAP_contains_pte,
+                                           khandle->user);
+                       cnt++;
+               }
+
+               offset = (uvaddr - ring->vma->vm_start) >> PAGE_SHIFT;
+
+               BTDBG("offset: 0x%08lx, page: %p, request: %p, usr_idx: %d, "
+                     "seg: %d, kvaddr: 0x%08lx, khandle: %u, uvaddr: "
+                     "0x%08lx, handle: %u\n", offset, map[offset], request,
+                     usr_idx, i, kvaddr, khandle->kernel, uvaddr,
+                     khandle->user);
+
+               page = map[offset];
+               if (page) {
+                       ClearPageReserved(map[offset]);
+                       if (PageBlkback(page)) {
+                               ClearPageBlkback(page);
+                               set_page_private(page, 0);
+                       }
+               }
+               map[offset] = NULL;
+
+               khandle->kernel = INVALID_GRANT_HANDLE;
+               khandle->user   = INVALID_GRANT_HANDLE;
+       }
+
+       if (cnt) {
+               ret = HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
+                                               unmap, cnt);
+               BUG_ON(ret);
+       }
+
+       if (!xen_feature(XENFEAT_auto_translated_physmap))
+               zap_page_range(ring->vma, 
+                              MMAP_VADDR(ring->user_vstart, usr_idx, 0), 
+                              request->nr_pages << PAGE_SHIFT, NULL);
+}
+
+/*
+ * tap->tap_sem held on entry
+ */
+static void
+blktap_unmap(struct blktap *tap, struct blktap_request *request)
+{
+       int i, usr_idx;
+       unsigned long kvaddr;
+
+       usr_idx = request->usr_idx;
+       down_write(&tap->ring.vma->vm_mm->mmap_sem);
+
+       for (i = 0; i < request->nr_pages; i++) {
+               BTDBG("request: %p, seg: %d, kvaddr: 0x%08lx, khandle: %u, "
+                     "uvaddr: 0x%08lx, uhandle: %u\n", request, i,
+                     request_to_kaddr(request, i),
+                     request->handles[i].kernel,
+                     MMAP_VADDR(tap->ring.user_vstart, usr_idx, i),
+                     request->handles[i].user);
+
+               if (request->handles[i].kernel == INVALID_GRANT_HANDLE) {
+                       kvaddr = request_to_kaddr(request, i);
+                       blktap_umap_uaddr(&init_mm, kvaddr);
+                       set_phys_to_machine(__pa(kvaddr) >> PAGE_SHIFT,
+                                           INVALID_P2M_ENTRY);
+               }
+       }
+
+       blktap_device_fast_flush(tap, request);
+       up_write(&tap->ring.vma->vm_mm->mmap_sem);
+}
+
+/*
+ * called if the tapdisk process dies unexpectedly.
+ * fail and release any pending requests and disable queue.
+ */
+void
+blktap_device_fail_pending_requests(struct blktap *tap)
+{
+       int usr_idx;
+       struct request *req;
+       struct blktap_device *dev;
+       struct blktap_request *request;
+
+       if (!test_bit(BLKTAP_DEVICE, &tap->dev_inuse))
+               return;
+
+       down_write(&tap->tap_sem);
+
+       dev = &tap->device;
+       for (usr_idx = 0; usr_idx < MAX_PENDING_REQS; usr_idx++) {
+               request = tap->pending_requests[usr_idx];
+               if (!request || request->status != BLKTAP_REQUEST_PENDING)
+                       continue;
+
+               BTERR("%u:%u: failing pending %s of %d pages\n",
+                     blktap_device_major, tap->minor,
+                     (request->operation == BLKIF_OP_READ ?
+                      "read" : "write"), request->nr_pages);
+
+               blktap_unmap(tap, request);
+               req = (struct request *)(unsigned long)request->id;
+               blktap_device_end_dequeued_request(dev, req, 0);
+               blktap_request_free(tap, request);
+       }
+
+       up_write(&tap->tap_sem);
+
+       spin_lock_irq(&dev->lock);
+
+       /* fail any future requests */
+       dev->gd->queue->queuedata = NULL;
+       blk_start_queue(dev->gd->queue);
+
+       spin_unlock_irq(&dev->lock);
+}
+
+/*
+ * tap->tap_sem held on entry
+ */
+void
+blktap_device_finish_request(struct blktap *tap,
+                            blkif_response_t *res,
+                            struct blktap_request *request)
+{
+       int uptodate;
+       struct request *req;
+       struct blktap_device *dev;
+
+       dev = &tap->device;
+
+       blktap_unmap(tap, request);
+
+       req = (struct request *)(unsigned long)request->id;
+       uptodate = (res->status == BLKIF_RSP_OKAY);
+
+       BTDBG("req %p res status %d operation %d/%d id %lld\n", req,
+               res->status, res->operation, request->operation, res->id);
+
+       switch (request->operation) {
+       case BLKIF_OP_READ:
+       case BLKIF_OP_WRITE:
+               if (unlikely(res->status != BLKIF_RSP_OKAY))
+                       BTERR("Bad return from device data "
+                               "request: %x\n", res->status);
+               blktap_device_end_dequeued_request(dev, req, uptodate);
+               break;
+       default:
+               BUG();
+       }
+
+       blktap_request_free(tap, request);
+}
+
+static int
+blktap_prep_foreign(struct blktap *tap,
+                   struct blktap_request *request,
+                   blkif_request_t *blkif_req,
+                   unsigned int seg, struct page *page,
+                   struct blktap_grant_table *table)
+{
+       uint64_t ptep;
+       uint32_t flags;
+       struct page *tap_page;
+       struct blktap_ring *ring;
+       struct blkback_pagemap map;
+       unsigned long uvaddr, kvaddr;
+
+       ring = &tap->ring;
+       map  = blkback_pagemap_read(page);
+       blkif_req->seg[seg].gref = map.gref;
+
+       uvaddr = MMAP_VADDR(ring->user_vstart, request->usr_idx, seg);
+       kvaddr = request_to_kaddr(request, seg);
+       flags  = GNTMAP_host_map |
+               (request->operation == BLKIF_OP_WRITE ? GNTMAP_readonly : 0);
+
+       gnttab_set_map_op(&table->grants[table->cnt],
+                         kvaddr, flags, map.gref, map.domid);
+       table->cnt++;
+
+       /* enable chained tap devices */
+       tap_page = pfn_to_page(__pa(kvaddr) >> PAGE_SHIFT);
+       set_page_private(tap_page, page_private(page));
+       SetPageBlkback(tap_page);
+
+       if (xen_feature(XENFEAT_auto_translated_physmap))
+               return 0;
+
+       if (create_lookup_pte_addr(ring->vma->vm_mm, uvaddr, &ptep)) {
+               BTERR("couldn't get a pte addr!\n");
+               return -1;
+       }
+
+       flags |= GNTMAP_application_map | GNTMAP_contains_pte;
+       gnttab_set_map_op(&table->grants[table->cnt],
+                         ptep, flags, map.gref, map.domid);
+       table->cnt++;
+
+       return 0;
+}
+
+static int
+blktap_map_foreign(struct blktap *tap,
+                  struct blktap_request *request,
+                  blkif_request_t *blkif_req,
+                  struct blktap_grant_table *table)
+{
+       struct page *page;
+       int i, grant, err, usr_idx;
+       struct blktap_ring *ring;
+       unsigned long uvaddr, kvaddr, foreign_mfn;
+
+       if (!table->cnt)
+               return 0;
+
+       err = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref,
+                                       table->grants, table->cnt);
+       BUG_ON(err);
+
+       grant   = 0;
+       usr_idx = request->usr_idx;
+       ring    = &tap->ring;
+
+       for (i = 0; i < request->nr_pages; i++) {
+               if (!blkif_req->seg[i].gref)
+                       continue;
+
+               uvaddr = MMAP_VADDR(ring->user_vstart, usr_idx, i);
+               kvaddr = request_to_kaddr(request, i);
+
+               if (unlikely(table->grants[grant].status)) {
+                       BTERR("invalid kernel buffer: could not remap it\n");
+                       err |= 1;
+                       table->grants[grant].handle = INVALID_GRANT_HANDLE;
+               }
+
+               request->handles[i].kernel = table->grants[grant].handle;
+               foreign_mfn = table->grants[grant].dev_bus_addr >> PAGE_SHIFT;
+               grant++;
+
+               if (xen_feature(XENFEAT_auto_translated_physmap))
+                       goto done;
+
+               if (unlikely(table->grants[grant].status)) {
+                       BTERR("invalid user buffer: could not remap it\n");
+                       err |= 1;
+                       table->grants[grant].handle = INVALID_GRANT_HANDLE;
+               }
+
+               request->handles[i].user = table->grants[grant].handle;
+               grant++;
+
+       done:
+               if (err)
+                       continue;
+
+               page = pfn_to_page(__pa(kvaddr) >> PAGE_SHIFT);
+
+               if (!xen_feature(XENFEAT_auto_translated_physmap))
+                       set_phys_to_machine(__pa(kvaddr) >> PAGE_SHIFT,
+                                           FOREIGN_FRAME(foreign_mfn));
+               else if (vm_insert_page(ring->vma, uvaddr, page))
+                       err |= 1;
+
+               BTDBG("pending_req: %p, seg: %d, page: %p, "
+                     "kvaddr: 0x%08lx, khandle: %u, uvaddr: 0x%08lx, "
+                     "uhandle: %u\n", request, i, page,
+                     kvaddr, request->handles[i].kernel,                      
+                     uvaddr, request->handles[i].user);
+       }
+
+       return err;
+}
+
+static void
+blktap_map(struct blktap *tap,
+          struct blktap_request *request,
+          unsigned int seg, struct page *page)
+{
+       pte_t pte;
+       int usr_idx;
+       struct blktap_ring *ring;
+       unsigned long uvaddr, kvaddr;
+
+       ring    = &tap->ring;
+       usr_idx = request->usr_idx;
+       uvaddr  = MMAP_VADDR(ring->user_vstart, usr_idx, seg);
+       kvaddr  = request_to_kaddr(request, seg);
+
+       pte = mk_pte(page, ring->vma->vm_page_prot);
+       blktap_map_uaddr(ring->vma->vm_mm, uvaddr, pte_mkwrite(pte));
+       blktap_map_uaddr(&init_mm, kvaddr, mk_pte(page, PAGE_KERNEL));
+
+       set_phys_to_machine(__pa(kvaddr) >> PAGE_SHIFT, pte_mfn(pte));
+       request->handles[seg].kernel = INVALID_GRANT_HANDLE;
+       request->handles[seg].user   = INVALID_GRANT_HANDLE;
+
+       BTDBG("pending_req: %p, seg: %d, page: %p, kvaddr: 0x%08lx, "
+             "uvaddr: 0x%08lx\n", request, seg, page, kvaddr,
+             uvaddr);
+}
+
+static int
+blktap_device_process_request(struct blktap *tap,
+                             struct blktap_request *request,
+                             struct request *req)
+{
+       struct bio *bio;
+       struct page *page;
+       struct bio_vec *bvec;
+       int idx, usr_idx, err;
+       struct blktap_ring *ring;
+       struct blktap_grant_table table;
+       unsigned int fsect, lsect, nr_sects;
+       unsigned long offset, uvaddr, kvaddr;
+       struct blkif_request blkif_req, *target;
+
+       err = -1;
+       memset(&table, 0, sizeof(table));
+
+       if (!blktap_active(tap))
+               goto out;
+
+       ring    = &tap->ring;
+       usr_idx = request->usr_idx;
+       blkif_req.id = usr_idx;
+       blkif_req.sector_number = (blkif_sector_t)req->sector;
+       blkif_req.handle = 0;
+       blkif_req.operation = rq_data_dir(req) ?
+               BLKIF_OP_WRITE : BLKIF_OP_READ;
+
+       request->id        = (unsigned long)req;
+       request->operation = blkif_req.operation;
+       request->status    = BLKTAP_REQUEST_PENDING;
+       do_gettimeofday(&request->time);
+
+       nr_sects = 0;
+       request->nr_pages = 0;
+       blkif_req.nr_segments = 0;
+       rq_for_each_bio(bio, req) {
+               bio_for_each_segment(bvec, bio, idx) {
+                       BUG_ON(blkif_req.nr_segments ==
+                              BLKIF_MAX_SEGMENTS_PER_REQUEST);
+
+                       fsect     = bvec->bv_offset >> 9;
+                       lsect     = fsect + (bvec->bv_len >> 9) - 1;
+                       nr_sects += bvec->bv_len >> 9;
+
+                       blkif_req.seg[blkif_req.nr_segments] =
+                               (struct blkif_request_segment) {
+                               .gref       = 0,
+                               .first_sect = fsect,
+                               .last_sect  = lsect };
+
+                       if (PageBlkback(bvec->bv_page)) {
+                               /* foreign page -- use xen */
+                               if (blktap_prep_foreign(tap,
+                                                       request,
+                                                       &blkif_req,
+                                                       blkif_req.nr_segments,
+                                                       bvec->bv_page,
+                                                       &table))
+                                       goto out;
+                       } else {
+                               /* do it the old fashioned way */
+                               blktap_map(tap,
+                                          request,
+                                          blkif_req.nr_segments,
+                                          bvec->bv_page);
+                       }
+
+                       uvaddr = MMAP_VADDR(ring->user_vstart,
+                                           usr_idx, blkif_req.nr_segments);
+                       kvaddr = request_to_kaddr(request,
+                                                 blkif_req.nr_segments);
+                       offset = (uvaddr - ring->vma->vm_start) >> PAGE_SHIFT;
+                       page   = pfn_to_page(__pa(kvaddr) >> PAGE_SHIFT);
+                       ring->foreign_map.map[offset] = page;
+                       SetPageReserved(page);
+
+                       BTDBG("mapped uaddr %08lx to page %p pfn 0x%lx\n",
+                             uvaddr, page, __pa(kvaddr) >> PAGE_SHIFT);
+                       BTDBG("offset: 0x%08lx, pending_req: %p, seg: %d, "
+                             "page: %p, kvaddr: 0x%08lx, uvaddr: 0x%08lx\n",
+                             offset, request, blkif_req.nr_segments,
+                             page, kvaddr, uvaddr);
+
+                       blkif_req.nr_segments++;
+                       request->nr_pages++;
+               }
+       }
+
+       if (blktap_map_foreign(tap, request, &blkif_req, &table))
+               goto out;
+
+       /* Finally, write the request message to the user ring. */
+       target = RING_GET_REQUEST(&ring->ring, ring->ring.req_prod_pvt);
+       memcpy(target, &blkif_req, sizeof(blkif_req));
+       target->id = request->usr_idx;
+       wmb(); /* blktap_poll() reads req_prod_pvt asynchronously */
+       ring->ring.req_prod_pvt++;
+
+       if (rq_data_dir(req)) {
+               tap->stats.st_wr_sect += nr_sects;
+               tap->stats.st_wr_req++;
+       } else {
+               tap->stats.st_rd_sect += nr_sects;
+               tap->stats.st_rd_req++;
+       }
+
+       err = 0;
+
+out:
+       if (err)
+               blktap_device_fast_flush(tap, request);
+       return err;
+}
+
+#ifdef ENABLE_PASSTHROUGH
+#define rq_for_each_bio_safe(_bio, _tmp, _req)                         \
+       if ((_req)->bio)                                                \
+               for (_bio = (_req)->bio;                                \
+                    _bio && ((_tmp = _bio->bi_next) || 1);             \
+                    _bio = _tmp)
+
+static void
+blktap_device_forward_request(struct blktap *tap, struct request *req)
+{
+       struct bio *bio, *tmp;
+       struct blktap_device *dev;
+
+       dev = &tap->device;
+
+       rq_for_each_bio_safe(bio, tmp, req) {
+               bio->bi_bdev = dev->bdev;
+               submit_bio(bio->bi_rw, bio);
+       }
+}
+
+static void
+blktap_device_close_bdev(struct blktap *tap)
+{
+       struct blktap_device *dev;
+
+       dev = &tap->device;
+
+       if (dev->bdev)
+               blkdev_put(dev->bdev);
+
+       dev->bdev = NULL;
+       clear_bit(BLKTAP_PASSTHROUGH, &tap->dev_inuse);
+}
+
+static int
+blktap_device_open_bdev(struct blktap *tap, u32 pdev)
+{
+       struct block_device *bdev;
+       struct blktap_device *dev;
+
+       dev = &tap->device;
+
+       bdev = open_by_devnum(pdev, FMODE_WRITE);
+       if (IS_ERR(bdev)) {
+               BTERR("opening device %x:%x failed: %ld\n",
+                     MAJOR(pdev), MINOR(pdev), PTR_ERR(bdev));
+               return PTR_ERR(bdev);
+       }
+
+       if (!bdev->bd_disk) {
+               BTERR("device %x:%x doesn't exist\n",
+                     MAJOR(pdev), MINOR(pdev));
+               blkdev_put(dev->bdev);
+               return -ENOENT;
+       }
+
+       dev->bdev = bdev;
+       set_bit(BLKTAP_PASSTHROUGH, &tap->dev_inuse);
+
+       /* TODO: readjust queue parameters */
+
+       BTINFO("set device %d to passthrough on %x:%x\n",
+              tap->minor, MAJOR(pdev), MINOR(pdev));
+
+       return 0;
+}
+
+int
+blktap_device_enable_passthrough(struct blktap *tap,
+                                unsigned major, unsigned minor)
+{
+       u32 pdev;
+       struct blktap_device *dev;
+
+       dev  = &tap->device;
+       pdev = MKDEV(major, minor);
+
+       if (!test_bit(BLKTAP_PAUSED, &tap->dev_inuse))
+               return -EINVAL;
+
+       if (dev->bdev) {
+               if (pdev)
+                       return -EINVAL;
+               blktap_device_close_bdev(tap);
+               return 0;
+       }
+
+       return blktap_device_open_bdev(tap, pdev);
+}
+#endif
+
+/*
+ * dev->lock held on entry
+ */
+static void
+blktap_device_run_queue(struct blktap *tap)
+{
+       int queued, err;
+       request_queue_t *rq;
+       struct request *req;
+       struct blktap_ring *ring;
+       struct blktap_device *dev;
+       struct blktap_request *request;
+
+       queued = 0;
+       ring   = &tap->ring;
+       dev    = &tap->device;
+       rq     = dev->gd->queue;
+
+       BTDBG("running queue for %d\n", tap->minor);
+
+       while ((req = elv_next_request(rq)) != NULL) {
+               if (!blk_fs_request(req)) {
+                       end_request(req, 0);
+                       continue;
+               }
+
+               if (blk_barrier_rq(req)) {
+                       end_request(req, 0);
+                       continue;
+               }
+
+#ifdef ENABLE_PASSTHROUGH
+               if (test_bit(BLKTAP_PASSTHROUGH, &tap->dev_inuse)) {
+                       blkdev_dequeue_request(req);
+                       blktap_device_forward_request(tap, req);
+                       continue;
+               }
+#endif
+
+               if (RING_FULL(&ring->ring)) {
+               wait:
+                       /* Avoid pointless unplugs. */
+                       blk_stop_queue(rq);
+                       blktap_defer(tap);
+                       break;
+               }
+
+               request = blktap_request_allocate(tap);
+               if (!request) {
+                       tap->stats.st_oo_req++;
+                       goto wait;
+               }
+
+               BTDBG("req %p: dev %d cmd %p, sec 0x%llx, (0x%x/0x%lx) "
+                     "buffer:%p [%s], pending: %p\n", req, tap->minor,
+                     req->cmd, req->sector, req->current_nr_sectors,
+                     req->nr_sectors, req->buffer,
+                     rq_data_dir(req) ? "write" : "read", request);
+
+               blkdev_dequeue_request(req);
+
+               spin_unlock_irq(&dev->lock);
+               down_read(&tap->tap_sem);
+
+               err = blktap_device_process_request(tap, request, req);
+               if (!err)
+                       queued++;
+               else {
+                       blktap_device_end_dequeued_request(dev, req, 0);
+                       blktap_request_free(tap, request);
+               }
+
+               up_read(&tap->tap_sem);
+               spin_lock_irq(&dev->lock);
+       }
+
+       if (queued)
+               blktap_ring_kick_user(tap);
+}
+
+/*
+ * dev->lock held on entry
+ */
+static void
+blktap_device_do_request(request_queue_t *rq)
+{
+       struct request *req;
+       struct blktap *tap;
+       struct blktap_device *dev;
+
+       dev = rq->queuedata;
+       if (!dev)
+               goto fail;
+
+       tap = dev_to_blktap(dev);
+       if (!blktap_active(tap))
+               goto fail;
+
+       if (test_bit(BLKTAP_PAUSED, &tap->dev_inuse) ||
+           test_bit(BLKTAP_PAUSE_REQUESTED, &tap->dev_inuse)) {
+               blktap_defer(tap);
+               return;
+       }
+
+       blktap_device_run_queue(tap);
+       return;
+
+fail:
+       while ((req = elv_next_request(rq))) {
+               BTERR("device closed: failing secs %llu - %llu\n",
+                     req->sector, req->sector + req->nr_sectors);
+               end_request(req, 0);
+       }
+}
+
+void
+blktap_device_restart(struct blktap *tap)
+{
+       struct blktap_device *dev;
+
+       dev = &tap->device;
+       if (!dev->gd || !dev->gd->queue)
+               return;
+
+       if (blktap_active(tap) && RING_FULL(&tap->ring.ring)) {
+               blktap_defer(tap);
+               return;
+       }
+
+       if (test_bit(BLKTAP_PAUSED, &tap->dev_inuse) ||
+           test_bit(BLKTAP_PAUSE_REQUESTED, &tap->dev_inuse)) {
+               blktap_defer(tap);
+               return;
+       }
+
+       spin_lock_irq(&dev->lock);
+
+       /* Re-enable calldowns. */
+       if (blk_queue_stopped(dev->gd->queue))
+               blk_start_queue(dev->gd->queue);
+
+       /* Kick things off immediately. */
+       blktap_device_do_request(dev->gd->queue);
+
+       spin_unlock_irq(&dev->lock);
+}
+
+static void
+blktap_device_configure(struct blktap *tap)
+{
+       struct request_queue *rq;
+       struct blktap_device *dev = &tap->device;
+
+       if (!test_bit(BLKTAP_DEVICE, &tap->dev_inuse) || !dev->gd)
+               return;
+
+       dev = &tap->device;
+       rq  = dev->gd->queue;
+
+       spin_lock_irq(&dev->lock);
+
+       set_capacity(dev->gd, tap->params.capacity);
+
+       /* Hard sector size and max sectors impersonate the equiv. hardware. */
+       blk_queue_hardsect_size(rq, tap->params.sector_size);
+       blk_queue_max_sectors(rq, 512);
+
+       /* Each segment in a request is up to an aligned page in size. */
+       blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
+       blk_queue_max_segment_size(rq, PAGE_SIZE);
+
+       /* Ensure a merged request will fit in a single I/O ring slot. */
+       blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
+       blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
+
+       /* Make sure buffer addresses are sector-aligned. */
+       blk_queue_dma_alignment(rq, 511);
+
+       spin_unlock_irq(&dev->lock);
+}
+
+int
+blktap_device_resume(struct blktap *tap)
+{
+       int err;
+
+       if (!test_bit(BLKTAP_DEVICE, &tap->dev_inuse) || !blktap_active(tap))
+               return -ENODEV;
+
+       if (!test_bit(BLKTAP_PAUSED, &tap->dev_inuse))
+               return 0;
+
+       err = blktap_ring_resume(tap);
+       if (err)
+               return err;
+
+       /* device size may have changed */
+       blktap_device_configure(tap);
+
+       BTDBG("restarting device\n");
+       blktap_device_restart(tap);
+
+       return 0;
+}
+
+int
+blktap_device_pause(struct blktap *tap)
+{
+       unsigned long flags;
+       struct blktap_device *dev = &tap->device;
+
+       if (!test_bit(BLKTAP_DEVICE, &tap->dev_inuse) || !blktap_active(tap))
+               return -ENODEV;
+
+       if (test_bit(BLKTAP_PAUSED, &tap->dev_inuse))
+               return 0;
+
+       spin_lock_irqsave(&dev->lock, flags);
+
+       blk_stop_queue(dev->gd->queue);
+       set_bit(BLKTAP_PAUSE_REQUESTED, &tap->dev_inuse);
+
+       spin_unlock_irqrestore(&dev->lock, flags);
+
+       return blktap_ring_pause(tap);
+}
+
+int
+blktap_device_destroy(struct blktap *tap)
+{
+       struct blktap_device *dev = &tap->device;
+
+       if (!test_bit(BLKTAP_DEVICE, &tap->dev_inuse))
+               return 0;
+
+       BTINFO("destroy device %d users %d\n", tap->minor, dev->users);
+
+       if (dev->users)
+               return -EBUSY;
+
+       spin_lock_irq(&dev->lock);
+       /* No more blktap_device_do_request(). */
+       blk_stop_queue(dev->gd->queue);
+       clear_bit(BLKTAP_DEVICE, &tap->dev_inuse);
+       spin_unlock_irq(&dev->lock);
+
+#ifdef ENABLE_PASSTHROUGH
+       if (dev->bdev)
+               blktap_device_close_bdev(tap);
+#endif
+
+       del_gendisk(dev->gd);
+       put_disk(dev->gd);
+       blk_cleanup_queue(dev->gd->queue);
+
+       dev->gd = NULL;
+
+       wake_up(&tap->wq);
+
+       return 0;
+}
+
+int
+blktap_device_create(struct blktap *tap)
+{
+       int minor, err;
+       struct gendisk *gd;
+       struct request_queue *rq;
+       struct blktap_device *dev;
+
+       gd    = NULL;
+       rq    = NULL;
+       dev   = &tap->device;
+       minor = tap->minor;
+
+       if (test_bit(BLKTAP_DEVICE, &tap->dev_inuse))
+               return -EEXIST;
+
+       if (blktap_validate_params(tap, &tap->params))
+               return -EINVAL;
+
+       BTINFO("minor %d sectors %Lu sector-size %lu\n",
+              minor, tap->params.capacity, tap->params.sector_size);
+
+       err = -ENODEV;
+
+       gd = alloc_disk(1);
+       if (!gd)
+               goto error;
+
+       if (minor < 26)
+               sprintf(gd->disk_name, "tapdev%c", 'a' + minor);
+       else
+               sprintf(gd->disk_name, "tapdev%c%c",
+                       'a' + ((minor / 26) - 1), 'a' + (minor % 26));
+
+       gd->major = blktap_device_major;
+       gd->first_minor = minor;
+       gd->fops = &blktap_device_file_operations;
+       gd->private_data = dev;
+
+       spin_lock_init(&dev->lock);
+       rq = blk_init_queue(blktap_device_do_request, &dev->lock);
+       if (!rq)
+               goto error;
+
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,10)
+       elevator_init(rq, "noop");
+#else
+       elevator_init(rq, &elevator_noop);
+#endif
+
+       gd->queue     = rq;
+       rq->queuedata = dev;
+       dev->gd       = gd;
+
+       set_bit(BLKTAP_DEVICE, &tap->dev_inuse);
+       blktap_device_configure(tap);
+
+       add_disk(gd);
+
+       err = 0;
+       goto out;
+
+ error:
+       if (gd)
+               del_gendisk(gd);
+       if (rq)
+               blk_cleanup_queue(rq);
+
+ out:
+       BTINFO("creation of %u:%u: %d\n", blktap_device_major, tap->minor, err);
+       return err;
+}
+
+int
+blktap_device_init(int *maj)
+{
+       int major;
+
+       /* Dynamically allocate a major for this device */
+       major = register_blkdev(0, "tapdev");
+       if (major < 0) {
+               BTERR("Couldn't register blktap device\n");
+               return -ENOMEM;
+       }       
+
+       blktap_device_major = *maj = major;
+       BTINFO("blktap device major %d\n", major);
+
+       return 0;
+}
+
+void
+blktap_device_free(void)
+{
+       if (blktap_device_major)
+               if (unregister_blkdev(blktap_device_major, "tapdev"))
+                       BTERR("blktap device unregister failed\n");
+}
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blktap2/request.c
--- /dev/null   Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/xen/blktap2/request.c     Tue May 26 11:23:16 2009 +0100
@@ -0,0 +1,297 @@
+#include <linux/spinlock.h>
+#include <xen/balloon.h>
+
+#include "blktap.h"
+
+#define MAX_BUCKETS                      8
+#define BUCKET_SIZE                      MAX_PENDING_REQS
+
+#define BLKTAP_POOL_CLOSING              1
+
+struct blktap_request_bucket;
+
+struct blktap_request_handle {
+       int                              slot;
+       uint8_t                          inuse;
+       struct blktap_request            request;
+       struct blktap_request_bucket    *bucket;
+};
+
+struct blktap_request_bucket {
+       atomic_t                         reqs_in_use;
+       struct blktap_request_handle     handles[BUCKET_SIZE];
+       struct page                    **foreign_pages;
+};
+
+struct blktap_request_pool {
+       spinlock_t                       lock;
+       uint8_t                          status;
+       struct list_head                 free_list;
+       atomic_t                         reqs_in_use;
+       wait_queue_head_t                wait_queue;
+       struct blktap_request_bucket    *buckets[MAX_BUCKETS];
+};
+
+static struct blktap_request_pool pool;
+
+static inline struct blktap_request_handle *
+blktap_request_to_handle(struct blktap_request *req)
+{
+       return container_of(req, struct blktap_request_handle, request);
+}
+
+static void
+blktap_request_pool_init_request(struct blktap_request *request)
+{
+       int i;
+
+       request->usr_idx  = -1;
+       request->nr_pages = 0;
+       request->status   = BLKTAP_REQUEST_FREE;
+       INIT_LIST_HEAD(&request->free_list);
+       for (i = 0; i < ARRAY_SIZE(request->handles); i++) {
+               request->handles[i].user   = INVALID_GRANT_HANDLE;
+               request->handles[i].kernel = INVALID_GRANT_HANDLE;
+       }
+}
+
+static int
+blktap_request_pool_allocate_bucket(void)
+{
+       int i, idx;
+       unsigned long flags;
+       struct blktap_request *request;
+       struct blktap_request_handle *handle;
+       struct blktap_request_bucket *bucket;
+
+       bucket = kzalloc(sizeof(struct blktap_request_bucket), GFP_KERNEL);
+       if (!bucket)
+               goto fail;
+
+       bucket->foreign_pages = alloc_empty_pages_and_pagevec(MMAP_PAGES);
+       if (!bucket->foreign_pages)
+               goto fail;
+
+       spin_lock_irqsave(&pool.lock, flags);
+
+       idx = -1;
+       for (i = 0; i < MAX_BUCKETS; i++) {
+               if (!pool.buckets[i]) {
+                       idx = i;
+                       pool.buckets[idx] = bucket;
+                       break;
+               }
+       }
+
+       if (idx == -1) {
+               spin_unlock_irqrestore(&pool.lock, flags);
+               goto fail;
+       }
+
+       for (i = 0; i < BUCKET_SIZE; i++) {
+               handle  = bucket->handles + i;
+               request = &handle->request;
+
+               handle->slot   = i;
+               handle->inuse  = 0;
+               handle->bucket = bucket;
+
+               blktap_request_pool_init_request(request);
+               list_add_tail(&request->free_list, &pool.free_list);
+       }
+
+       spin_unlock_irqrestore(&pool.lock, flags);
+
+       return 0;
+
+fail:
+       if (bucket && bucket->foreign_pages)
+               free_empty_pages_and_pagevec(bucket->foreign_pages, MMAP_PAGES);
+       kfree(bucket);
+       return -ENOMEM;
+}
+
+static void
+blktap_request_pool_free_bucket(struct blktap_request_bucket *bucket)
+{
+       if (!bucket)
+               return;
+
+       BTDBG("freeing bucket %p\n", bucket);
+
+       free_empty_pages_and_pagevec(bucket->foreign_pages, MMAP_PAGES);
+       kfree(bucket);
+}
+
+unsigned long
+request_to_kaddr(struct blktap_request *req, int seg)
+{
+       struct blktap_request_handle *handle = blktap_request_to_handle(req);
+       int idx = handle->slot * BLKIF_MAX_SEGMENTS_PER_REQUEST + seg;
+       unsigned long pfn = page_to_pfn(handle->bucket->foreign_pages[idx]);
+       return (unsigned long)pfn_to_kaddr(pfn);
+}
+
+int
+blktap_request_pool_shrink(void)
+{
+       int i, err;
+       unsigned long flags;
+       struct blktap_request_bucket *bucket;
+
+       err = -EAGAIN;
+
+       spin_lock_irqsave(&pool.lock, flags);
+
+       /* always keep at least one bucket */
+       for (i = 1; i < MAX_BUCKETS; i++) {
+               bucket = pool.buckets[i];
+               if (!bucket)
+                       continue;
+
+               if (atomic_read(&bucket->reqs_in_use))
+                       continue;
+
+               blktap_request_pool_free_bucket(bucket);
+               pool.buckets[i] = NULL;
+               err = 0;
+               break;
+       }
+
+       spin_unlock_irqrestore(&pool.lock, flags);
+
+       return err;
+}
+
+int
+blktap_request_pool_grow(void)
+{
+       return blktap_request_pool_allocate_bucket();
+}
+
+struct blktap_request *
+blktap_request_allocate(struct blktap *tap)
+{
+       int i;
+       uint16_t usr_idx;
+       unsigned long flags;
+       struct blktap_request *request;
+
+       usr_idx = -1;
+       request = NULL;
+
+       spin_lock_irqsave(&pool.lock, flags);
+
+       if (pool.status == BLKTAP_POOL_CLOSING)
+               goto out;
+
+       for (i = 0; i < ARRAY_SIZE(tap->pending_requests); i++)
+               if (!tap->pending_requests[i]) {
+                       usr_idx = i;
+                       break;
+               }
+
+       if (usr_idx == (uint16_t)-1)
+               goto out;
+
+       if (!list_empty(&pool.free_list)) {
+               request = list_entry(pool.free_list.next,
+                                    struct blktap_request, free_list);
+               list_del(&request->free_list);
+       }
+
+       if (request) {
+               struct blktap_request_handle *handle;
+
+               atomic_inc(&pool.reqs_in_use);
+
+               handle = blktap_request_to_handle(request);
+               atomic_inc(&handle->bucket->reqs_in_use);
+               handle->inuse = 1;
+
+               request->usr_idx = usr_idx;
+
+               tap->pending_requests[usr_idx] = request;
+               tap->pending_cnt++;
+       }
+
+out:
+       spin_unlock_irqrestore(&pool.lock, flags);
+       return request;
+}
+
+void
+blktap_request_free(struct blktap *tap, struct blktap_request *request)
+{
+       int free;
+       unsigned long flags;
+       struct blktap_request_handle *handle;
+
+       BUG_ON(request->usr_idx >= ARRAY_SIZE(tap->pending_requests));
+       handle = blktap_request_to_handle(request);
+
+       spin_lock_irqsave(&pool.lock, flags);
+
+       handle->inuse = 0;
+       tap->pending_requests[request->usr_idx] = NULL;
+       blktap_request_pool_init_request(request);
+       list_add(&request->free_list, &pool.free_list);
+       atomic_dec(&handle->bucket->reqs_in_use);
+       free = atomic_dec_and_test(&pool.reqs_in_use);
+
+       spin_unlock_irqrestore(&pool.lock, flags);
+
+       if (--tap->pending_cnt == 0)
+               wake_up_interruptible(&tap->wq);
+
+       if (free)
+               wake_up(&pool.wait_queue);
+}
+
+void
+blktap_request_pool_free(void)
+{
+       int i;
+       unsigned long flags;
+
+       spin_lock_irqsave(&pool.lock, flags);
+
+       pool.status = BLKTAP_POOL_CLOSING;
+       while (atomic_read(&pool.reqs_in_use)) {
+               spin_unlock_irqrestore(&pool.lock, flags);
+               wait_event(pool.wait_queue, !atomic_read(&pool.reqs_in_use));
+               spin_lock_irqsave(&pool.lock, flags);
+       }
+
+       for (i = 0; i < MAX_BUCKETS; i++) {
+               blktap_request_pool_free_bucket(pool.buckets[i]);
+               pool.buckets[i] = NULL;
+       }
+
+       spin_unlock_irqrestore(&pool.lock, flags);
+}
+
+int
+blktap_request_pool_init(void)
+{
+       int i, err;
+
+       memset(&pool, 0, sizeof(pool));
+
+       spin_lock_init(&pool.lock);
+       INIT_LIST_HEAD(&pool.free_list);
+       atomic_set(&pool.reqs_in_use, 0);
+       init_waitqueue_head(&pool.wait_queue);
+
+       for (i = 0; i < 2; i++) {
+               err = blktap_request_pool_allocate_bucket();
+               if (err)
+                       goto fail;
+       }
+
+       return 0;
+
+fail:
+       blktap_request_pool_free();
+       return err;
+}
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blktap2/ring.c
--- /dev/null   Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/xen/blktap2/ring.c        Tue May 26 11:23:16 2009 +0100
@@ -0,0 +1,613 @@
+#include <linux/module.h>
+#include <linux/signal.h>
+
+#include "blktap.h"
+
+static int blktap_ring_major;
+
+static inline struct blktap *
+vma_to_blktap(struct vm_area_struct *vma)
+{
+       struct vm_foreign_map *m = vma->vm_private_data;
+       struct blktap_ring *r = container_of(m, struct blktap_ring, 
foreign_map);
+       return container_of(r, struct blktap, ring);
+}
+
+ /* 
+  * BLKTAP - immediately before the mmap area,
+  * we have a bunch of pages reserved for shared memory rings.
+  */
+#define RING_PAGES 1
+
+static int
+blktap_read_ring(struct blktap *tap)
+{
+       /* This is called to read responses from the ring. */
+       int usr_idx;
+       RING_IDX rc, rp;
+       blkif_response_t res;
+       struct blktap_ring *ring;
+       struct blktap_request *request;
+
+       down_read(&tap->tap_sem);
+
+       ring = &tap->ring;
+       if (!ring->vma) {
+               up_read(&tap->tap_sem);
+               return 0;
+       }
+
+       /* for each outstanding message on the ring  */
+       rp = ring->ring.sring->rsp_prod;
+       rmb();
+
+       for (rc = ring->ring.rsp_cons; rc != rp; rc++) {
+               memcpy(&res, RING_GET_RESPONSE(&ring->ring, rc), sizeof(res));
+               mb(); /* rsp_cons read by RING_FULL() in do_block_io_op(). */
+               ++ring->ring.rsp_cons;
+
+               usr_idx = (int)res.id;
+               if (usr_idx >= MAX_PENDING_REQS ||
+                   !tap->pending_requests[usr_idx]) {
+                       BTWARN("Request %d/%d invalid [%x], tapdisk %d%p\n",
+                              rc, rp, usr_idx, tap->pid, ring->vma);
+                       continue;
+               }
+
+               request = tap->pending_requests[usr_idx];
+               BTDBG("request %p response #%d id %x\n", request, rc, usr_idx);
+               blktap_device_finish_request(tap, &res, request);
+       }
+
+       up_read(&tap->tap_sem);
+
+       blktap_run_deferred();
+
+       return 0;
+}
+
+static struct page *
+blktap_ring_nopage(struct vm_area_struct *vma,
+                  unsigned long address, int *type)
+{
+       /*
+        * if the page has not been mapped in by the driver then return
+        * NOPAGE_SIGBUS to the domain.
+        */
+
+       return NOPAGE_SIGBUS;
+}
+
+static pte_t
+blktap_ring_clear_pte(struct vm_area_struct *vma,
+                     unsigned long uvaddr,
+                     pte_t *ptep, int is_fullmm)
+{
+       pte_t copy;
+       struct blktap *tap;
+       unsigned long kvaddr;
+       struct page **map, *page;
+       struct blktap_ring *ring;
+       struct blktap_request *request;
+       struct grant_handle_pair *khandle;
+       struct gnttab_unmap_grant_ref unmap[2];
+       int offset, seg, usr_idx, count = 0;
+
+       tap  = vma_to_blktap(vma);
+       ring = &tap->ring;
+       map  = ring->foreign_map.map;
+       BUG_ON(!map);   /* TODO Should this be changed to if statement? */
+
+       /*
+        * Zap entry if the address is before the start of the grant
+        * mapped region.
+        */
+       if (uvaddr < ring->user_vstart)
+               return ptep_get_and_clear_full(vma->vm_mm, uvaddr,
+                                              ptep, is_fullmm);
+
+       offset  = (int)((uvaddr - ring->user_vstart) >> PAGE_SHIFT);
+       usr_idx = offset / BLKIF_MAX_SEGMENTS_PER_REQUEST;
+       seg     = offset % BLKIF_MAX_SEGMENTS_PER_REQUEST;
+
+       offset  = (int)((uvaddr - vma->vm_start) >> PAGE_SHIFT);
+       page    = map[offset];
+       if (page) {
+               ClearPageReserved(page);
+               if (PageBlkback(page)) {
+                       ClearPageBlkback(page);
+                       set_page_private(page, 0);
+               }
+       }
+       map[offset] = NULL;
+
+       request = tap->pending_requests[usr_idx];
+       kvaddr  = request_to_kaddr(request, seg);
+       khandle = request->handles + seg;
+
+       if (khandle->kernel != INVALID_GRANT_HANDLE) {
+               gnttab_set_unmap_op(&unmap[count], kvaddr, 
+                                   GNTMAP_host_map, khandle->kernel);
+               count++;
+
+               set_phys_to_machine(__pa(kvaddr) >> PAGE_SHIFT, 
+                                   INVALID_P2M_ENTRY);
+       }
+
+
+       if (khandle->user != INVALID_GRANT_HANDLE) {
+               BUG_ON(xen_feature(XENFEAT_auto_translated_physmap));
+
+               copy = *ptep;
+               gnttab_set_unmap_op(&unmap[count], virt_to_machine(ptep), 
+                                   GNTMAP_host_map 
+                                   | GNTMAP_application_map 
+                                   | GNTMAP_contains_pte,
+                                   khandle->user);
+               count++;
+       } else
+               copy = ptep_get_and_clear_full(vma->vm_mm, uvaddr, ptep,
+                                              is_fullmm);
+
+       if (count)
+               if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref,
+                                             unmap, count))
+                       BUG();
+
+       khandle->kernel = INVALID_GRANT_HANDLE;
+       khandle->user   = INVALID_GRANT_HANDLE;
+
+       return copy;
+}
+
+static void
+blktap_ring_vm_unmap(struct vm_area_struct *vma)
+{
+       struct blktap *tap = vma_to_blktap(vma);
+
+       down_write(&tap->tap_sem);
+       clear_bit(BLKTAP_RING_VMA, &tap->dev_inuse);
+       clear_bit(BLKTAP_PAUSED, &tap->dev_inuse);
+       clear_bit(BLKTAP_PAUSE_REQUESTED, &tap->dev_inuse);
+       up_write(&tap->tap_sem);
+}
+
+static void
+blktap_ring_vm_close(struct vm_area_struct *vma)
+{
+       struct blktap *tap = vma_to_blktap(vma);
+       struct blktap_ring *ring = &tap->ring;
+
+       blktap_ring_vm_unmap(vma);                 /* fail future requests */
+       blktap_device_fail_pending_requests(tap);  /* fail pending requests */
+       blktap_device_restart(tap);                /* fail deferred requests */
+
+       down_write(&tap->tap_sem);
+
+       zap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, NULL);
+
+       kfree(ring->foreign_map.map);
+       ring->foreign_map.map = NULL;
+
+       /* Free the ring page. */
+       ClearPageReserved(virt_to_page(ring->ring.sring));
+       free_page((unsigned long)ring->ring.sring);
+
+       BTINFO("unmapping ring %d\n", tap->minor);
+       ring->ring.sring = NULL;
+       ring->vma = NULL;
+
+       up_write(&tap->tap_sem);
+
+       wake_up(&tap->wq);
+}
+
+static struct vm_operations_struct blktap_ring_vm_operations = {
+       .close    = blktap_ring_vm_close,
+       .unmap    = blktap_ring_vm_unmap,
+       .nopage   = blktap_ring_nopage,
+       .zap_pte  = blktap_ring_clear_pte,
+};
+
+static int
+blktap_ring_open(struct inode *inode, struct file *filp)
+{
+       int idx;
+       struct blktap *tap;
+
+       idx = iminor(inode);
+       if (idx < 0 || idx > MAX_BLKTAP_DEVICE || blktaps[idx] == NULL) {
+               BTERR("unable to open device blktap%d\n", idx);
+               return -ENODEV;
+       }
+
+       tap = blktaps[idx];
+
+       BTINFO("opening device blktap%d\n", idx);
+
+       if (!test_bit(BLKTAP_CONTROL, &tap->dev_inuse))
+               return -ENODEV;
+
+       /* Only one process can access ring at a time */
+       if (test_and_set_bit(BLKTAP_RING_FD, &tap->dev_inuse))
+               return -EBUSY;
+
+       filp->private_data = tap;
+       BTINFO("opened device %d\n", tap->minor);
+
+       return 0;
+}
+
+static int
+blktap_ring_release(struct inode *inode, struct file *filp)
+{
+       struct blktap *tap = filp->private_data;
+
+       BTINFO("freeing device %d\n", tap->minor);
+       clear_bit(BLKTAP_RING_FD, &tap->dev_inuse);
+       filp->private_data = NULL;
+       wake_up(&tap->wq);      
+       return 0;
+}
+
+/* Note on mmap:
+ * We need to map pages to user space in a way that will allow the block
+ * subsystem set up direct IO to them.  This couldn't be done before, because
+ * there isn't really a sane way to translate a user virtual address down to a 
+ * physical address when the page belongs to another domain.
+ *
+ * My first approach was to map the page in to kernel memory, add an entry
+ * for it in the physical frame list (using alloc_lomem_region as in blkback)
+ * and then attempt to map that page up to user space.  This is disallowed
+ * by xen though, which realizes that we don't really own the machine frame
+ * underlying the physical page.
+ *
+ * The new approach is to provide explicit support for this in xen linux.
+ * The VMA now has a flag, VM_FOREIGN, to indicate that it contains pages
+ * mapped from other vms.  vma->vm_private_data is set up as a mapping 
+ * from pages to actual page structs.  There is a new clause in get_user_pages
+ * that does the right thing for this sort of mapping.
+ */
+static int
+blktap_ring_mmap(struct file *filp, struct vm_area_struct *vma)
+{
+       int size, err;
+       struct page **map;
+       struct blktap *tap;
+       blkif_sring_t *sring;
+       struct blktap_ring *ring;
+
+       tap   = filp->private_data;
+       ring  = &tap->ring;
+       map   = NULL;
+       sring = NULL;
+
+       if (!tap || test_and_set_bit(BLKTAP_RING_VMA, &tap->dev_inuse))
+               return -ENOMEM;
+
+       size = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
+       if (size != (MMAP_PAGES + RING_PAGES)) {
+               BTERR("you _must_ map exactly %lu pages!\n",
+                     MMAP_PAGES + RING_PAGES);
+               return -EAGAIN;
+       }
+
+       /* Allocate the fe ring. */
+       sring = (blkif_sring_t *)get_zeroed_page(GFP_KERNEL);
+       if (!sring) {
+               BTERR("Couldn't alloc sring.\n");
+               goto fail_mem;
+       }
+
+       map = kzalloc(size * sizeof(struct page *), GFP_KERNEL);
+       if (!map) {
+               BTERR("Couldn't alloc VM_FOREIGN map.\n");
+               goto fail_mem;
+       }
+
+       SetPageReserved(virt_to_page(sring));
+    
+       SHARED_RING_INIT(sring);
+       FRONT_RING_INIT(&ring->ring, sring, PAGE_SIZE);
+
+       ring->ring_vstart = vma->vm_start;
+       ring->user_vstart = ring->ring_vstart + (RING_PAGES << PAGE_SHIFT);
+
+       /* Map the ring pages to the start of the region and reserve it. */
+       if (xen_feature(XENFEAT_auto_translated_physmap))
+               err = vm_insert_page(vma, vma->vm_start,
+                                    virt_to_page(ring->ring.sring));
+       else
+               err = remap_pfn_range(vma, vma->vm_start,
+                                     __pa(ring->ring.sring) >> PAGE_SHIFT,
+                                     PAGE_SIZE, vma->vm_page_prot);
+       if (err) {
+               BTERR("Mapping user ring failed: %d\n", err);
+               goto fail;
+       }
+
+       /* Mark this VM as containing foreign pages, and set up mappings. */
+       ring->foreign_map.map = map;
+       vma->vm_private_data = &ring->foreign_map;
+       vma->vm_flags |= VM_FOREIGN;
+       vma->vm_flags |= VM_DONTCOPY;
+       vma->vm_flags |= VM_RESERVED;
+       vma->vm_ops = &blktap_ring_vm_operations;
+
+#ifdef CONFIG_X86
+       vma->vm_mm->context.has_foreign_mappings = 1;
+#endif
+
+       tap->pid = current->pid;
+       BTINFO("blktap: mapping pid is %d\n", tap->pid);
+
+       ring->vma = vma;
+       return 0;
+
+ fail:
+       /* Clear any active mappings. */
+       zap_page_range(vma, vma->vm_start, 
+                      vma->vm_end - vma->vm_start, NULL);
+       ClearPageReserved(virt_to_page(sring));
+ fail_mem:
+       free_page((unsigned long)sring);
+       kfree(map);
+
+       return -ENOMEM;
+}
+
+static inline void
+blktap_ring_set_message(struct blktap *tap, int msg)
+{
+       struct blktap_ring *ring = &tap->ring;
+
+       down_read(&tap->tap_sem);
+       if (ring->ring.sring)
+               ring->ring.sring->pad[0] = msg;
+       up_read(&tap->tap_sem);
+}
+
+static int
+blktap_ring_ioctl(struct inode *inode, struct file *filp,
+                 unsigned int cmd, unsigned long arg)
+{
+       struct blktap_params params;
+       struct blktap *tap = filp->private_data;
+
+       BTDBG("%d: cmd: %u, arg: %lu\n", tap->minor, cmd, arg);
+
+       switch(cmd) {
+       case BLKTAP2_IOCTL_KICK_FE:
+               /* There are fe messages to process. */
+               return blktap_read_ring(tap);
+
+       case BLKTAP2_IOCTL_CREATE_DEVICE:
+               if (!arg)
+                       return -EINVAL;
+
+               if (copy_from_user(&params, (struct blktap_params __user *)arg,
+                                  sizeof(params))) {
+                       BTERR("failed to get params\n");
+                       return -EFAULT;
+               }
+
+               if (blktap_validate_params(tap, &params)) {
+                       BTERR("invalid params\n");
+                       return -EINVAL;
+               }
+
+               tap->params = params;
+               return blktap_device_create(tap);
+
+       case BLKTAP2_IOCTL_SET_PARAMS:
+               if (!arg)
+                       return -EINVAL;
+
+               if (!test_bit(BLKTAP_PAUSED, &tap->dev_inuse))
+                       return -EINVAL;
+
+               if (copy_from_user(&params, (struct blktap_params __user *)arg,
+                                  sizeof(params))) {
+                       BTERR("failed to get params\n");
+                       return -EFAULT;
+               }
+
+               if (blktap_validate_params(tap, &params)) {
+                       BTERR("invalid params\n");
+                       return -EINVAL;
+               }
+
+               tap->params = params;
+               return 0;
+
+       case BLKTAP2_IOCTL_PAUSE:
+               if (!test_bit(BLKTAP_PAUSE_REQUESTED, &tap->dev_inuse))
+                       return -EINVAL;
+
+               set_bit(BLKTAP_PAUSED, &tap->dev_inuse);
+               clear_bit(BLKTAP_PAUSE_REQUESTED, &tap->dev_inuse);
+
+               blktap_ring_set_message(tap, 0);
+               wake_up_interruptible(&tap->wq);
+
+               return 0;
+
+
+       case BLKTAP2_IOCTL_REOPEN:
+               if (!test_bit(BLKTAP_PAUSED, &tap->dev_inuse))
+                       return -EINVAL;
+
+               if (!arg)
+                       return -EINVAL;
+
+               if (copy_to_user((char __user *)arg,
+                                tap->params.name,
+                                strlen(tap->params.name) + 1))
+                       return -EFAULT;
+
+               blktap_ring_set_message(tap, 0);
+               wake_up_interruptible(&tap->wq);
+
+               return 0;
+
+       case BLKTAP2_IOCTL_RESUME:
+               if (!test_bit(BLKTAP_PAUSED, &tap->dev_inuse))
+                       return -EINVAL;
+
+               tap->ring.response = (int)arg;
+               if (!tap->ring.response)
+                       clear_bit(BLKTAP_PAUSED, &tap->dev_inuse);
+
+               blktap_ring_set_message(tap, 0);
+               wake_up_interruptible(&tap->wq);
+
+               return 0;
+       }
+
+       return -ENOIOCTLCMD;
+}
+
+static unsigned int blktap_ring_poll(struct file *filp, poll_table *wait)
+{
+       struct blktap *tap = filp->private_data;
+       struct blktap_ring *ring = &tap->ring;
+
+       poll_wait(filp, &ring->poll_wait, wait);
+       if (ring->ring.sring->pad[0] != 0 ||
+           ring->ring.req_prod_pvt != ring->ring.sring->req_prod) {
+               RING_PUSH_REQUESTS(&ring->ring);
+               return POLLIN | POLLRDNORM;
+       }
+
+       return 0;
+}
+
+static struct file_operations blktap_ring_file_operations = {
+       .owner    = THIS_MODULE,
+       .open     = blktap_ring_open,
+       .release  = blktap_ring_release,
+       .ioctl    = blktap_ring_ioctl,
+       .mmap     = blktap_ring_mmap,
+       .poll     = blktap_ring_poll,
+};
+
+void
+blktap_ring_kick_user(struct blktap *tap)
+{
+       wake_up_interruptible(&tap->ring.poll_wait);
+}
+
+int
+blktap_ring_resume(struct blktap *tap)
+{
+       int err;
+       struct blktap_ring *ring = &tap->ring;
+
+       if (!blktap_active(tap))
+               return -ENODEV;
+
+       if (!test_bit(BLKTAP_PAUSED, &tap->dev_inuse))
+               return -EINVAL;
+
+       /* set shared flag for resume */
+       ring->response = 0;
+
+       blktap_ring_set_message(tap, BLKTAP2_RING_MESSAGE_RESUME);
+       blktap_ring_kick_user(tap);
+
+       wait_event_interruptible(tap->wq, ring->response ||
+                                !test_bit(BLKTAP_PAUSED, &tap->dev_inuse));
+
+       err = ring->response;
+       ring->response = 0;
+
+       BTDBG("err: %d\n", err);
+
+       if (err)
+               return err;
+
+       if (test_bit(BLKTAP_PAUSED, &tap->dev_inuse))
+               return -EAGAIN;
+
+       return 0;
+}
+
+int
+blktap_ring_pause(struct blktap *tap)
+{
+       if (!blktap_active(tap))
+               return -ENODEV;
+
+       if (!test_bit(BLKTAP_PAUSE_REQUESTED, &tap->dev_inuse))
+               return -EINVAL;
+
+       BTDBG("draining queue\n");
+       wait_event_interruptible(tap->wq, !tap->pending_cnt);
+       if (tap->pending_cnt)
+               return -EAGAIN;
+
+       blktap_ring_set_message(tap, BLKTAP2_RING_MESSAGE_PAUSE);
+       blktap_ring_kick_user(tap);
+
+       BTDBG("waiting for tapdisk response\n");
+       wait_event_interruptible(tap->wq, test_bit(BLKTAP_PAUSED, 
&tap->dev_inuse));
+       if (!test_bit(BLKTAP_PAUSED, &tap->dev_inuse))
+               return -EAGAIN;
+
+       return 0;
+}
+
+int
+blktap_ring_destroy(struct blktap *tap)
+{
+       if (!test_bit(BLKTAP_RING_FD, &tap->dev_inuse) &&
+           !test_bit(BLKTAP_RING_VMA, &tap->dev_inuse))
+               return 0;
+
+       BTDBG("sending tapdisk close message\n");
+       blktap_ring_set_message(tap, BLKTAP2_RING_MESSAGE_CLOSE);
+       blktap_ring_kick_user(tap);
+
+       return -EAGAIN;
+}
+
+static void
+blktap_ring_initialize(struct blktap_ring *ring, int minor)
+{
+       memset(ring, 0, sizeof(*ring));
+       init_waitqueue_head(&ring->poll_wait);
+       ring->devno = MKDEV(blktap_ring_major, minor);
+}
+
+int
+blktap_ring_create(struct blktap *tap)
+{
+       struct blktap_ring *ring = &tap->ring;
+       blktap_ring_initialize(ring, tap->minor);
+       return blktap_sysfs_create(tap);
+}
+
+int
+blktap_ring_init(int *major)
+{
+       int err;
+
+       err = register_chrdev(0, "blktap2", &blktap_ring_file_operations);
+       if (err < 0) {
+               BTERR("error registering blktap ring device: %d\n", err);
+               return err;
+       }
+
+       blktap_ring_major = *major = err;
+       BTINFO("blktap ring major: %d\n", blktap_ring_major);
+       return 0;
+}
+
+int
+blktap_ring_free(void)
+{
+       if (blktap_ring_major)
+               unregister_chrdev(blktap_ring_major, "blktap2");
+
+       return 0;
+}
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blktap2/sysfs.c
--- /dev/null   Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/xen/blktap2/sysfs.c       Tue May 26 11:23:16 2009 +0100
@@ -0,0 +1,425 @@
+#include <linux/types.h>
+#include <linux/device.h>
+#include <linux/module.h>
+
+#include "blktap.h"
+
+int blktap_debug_level = 1;
+
+static struct class *class;
+static DECLARE_WAIT_QUEUE_HEAD(sysfs_wq);
+
+static inline void
+blktap_sysfs_get(struct blktap *tap)
+{
+       atomic_inc(&tap->ring.sysfs_refcnt);
+}
+
+static inline void
+blktap_sysfs_put(struct blktap *tap)
+{
+       if (atomic_dec_and_test(&tap->ring.sysfs_refcnt))
+               wake_up(&sysfs_wq);
+}
+
+static inline void
+blktap_sysfs_enter(struct blktap *tap)
+{
+       blktap_sysfs_get(tap);               /* pin sysfs device */
+       mutex_lock(&tap->ring.sysfs_mutex);  /* serialize sysfs operations */
+}
+
+static inline void
+blktap_sysfs_exit(struct blktap *tap)
+{
+       mutex_unlock(&tap->ring.sysfs_mutex);
+       blktap_sysfs_put(tap);
+}
+
+static ssize_t blktap_sysfs_pause_device(struct class_device *, const char *, 
size_t);
+CLASS_DEVICE_ATTR(pause, S_IWUSR, NULL, blktap_sysfs_pause_device);
+static ssize_t blktap_sysfs_resume_device(struct class_device *, const char *, 
size_t);
+CLASS_DEVICE_ATTR(resume, S_IWUSR, NULL, blktap_sysfs_resume_device);
+
+static ssize_t
+blktap_sysfs_set_name(struct class_device *dev, const char *buf, size_t size)
+{
+       int err;
+       struct blktap *tap = (struct blktap *)dev->class_data;
+
+       blktap_sysfs_enter(tap);
+
+       if (!tap->ring.dev ||
+           test_bit(BLKTAP_SHUTDOWN_REQUESTED, &tap->dev_inuse)) {
+               err = -ENODEV;
+               goto out;
+       }
+
+       if (!test_bit(BLKTAP_PAUSED, &tap->dev_inuse)) {
+               err = -EPERM;
+               goto out;
+       }
+
+       if (size > BLKTAP2_MAX_MESSAGE_LEN) {
+               err = -ENAMETOOLONG;
+               goto out;
+       }
+
+       if (strnlen(buf, BLKTAP2_MAX_MESSAGE_LEN) >= BLKTAP2_MAX_MESSAGE_LEN) {
+               err = -EINVAL;
+               goto out;
+       }
+
+       snprintf(tap->params.name, sizeof(tap->params.name) - 1, "%s", buf);
+       err = size;
+
+out:
+       blktap_sysfs_exit(tap); 
+       return err;
+}
+
+static ssize_t
+blktap_sysfs_get_name(struct class_device *dev, char *buf)
+{
+       ssize_t size;
+       struct blktap *tap = (struct blktap *)dev->class_data;
+
+       blktap_sysfs_enter(tap);
+
+       if (!tap->ring.dev)
+               size = -ENODEV;
+       else if (tap->params.name[0])
+               size = sprintf(buf, "%s\n", tap->params.name);
+       else
+               size = sprintf(buf, "%d\n", tap->minor);
+
+       blktap_sysfs_exit(tap);
+
+       return size;
+}
+CLASS_DEVICE_ATTR(name, S_IRUSR | S_IWUSR,
+                 blktap_sysfs_get_name, blktap_sysfs_set_name);
+
+static ssize_t
+blktap_sysfs_remove_device(struct class_device *dev,
+                          const char *buf, size_t size)
+{
+       int err;
+       struct blktap *tap = (struct blktap *)dev->class_data;
+
+       if (!tap->ring.dev)
+               return size;
+
+       if (test_and_set_bit(BLKTAP_SHUTDOWN_REQUESTED, &tap->dev_inuse))
+               return -EBUSY;
+
+       err = blktap_control_destroy_device(tap);
+
+       return (err ? : size);
+}
+CLASS_DEVICE_ATTR(remove, S_IWUSR, NULL, blktap_sysfs_remove_device);
+
+static ssize_t
+blktap_sysfs_pause_device(struct class_device *dev,
+                         const char *buf, size_t size)
+{
+       int err;
+       struct blktap *tap = (struct blktap *)dev->class_data;
+
+       blktap_sysfs_enter(tap);
+
+       BTDBG("pausing %u:%u: dev_inuse: %lu\n",
+             MAJOR(tap->ring.devno), MINOR(tap->ring.devno), tap->dev_inuse);
+
+       if (!tap->ring.dev ||
+           test_bit(BLKTAP_SHUTDOWN_REQUESTED, &tap->dev_inuse)) {
+               err = -ENODEV;
+               goto out;
+       }
+
+       if (test_bit(BLKTAP_PAUSE_REQUESTED, &tap->dev_inuse)) {
+               err = -EBUSY;
+               goto out;
+       }
+
+       if (test_bit(BLKTAP_PAUSED, &tap->dev_inuse)) {
+               err = 0;
+               goto out;
+       }
+
+       err = blktap_device_pause(tap);
+       if (!err) {
+               class_device_remove_file(dev, &class_device_attr_pause);
+               class_device_create_file(dev, &class_device_attr_resume);
+       }
+
+out:
+       blktap_sysfs_exit(tap);
+
+       return (err ? err : size);
+}
+
+static ssize_t
+blktap_sysfs_resume_device(struct class_device *dev,
+                          const char *buf, size_t size)
+{
+       int err;
+       struct blktap *tap = (struct blktap *)dev->class_data;
+
+       blktap_sysfs_enter(tap);
+
+       if (!tap->ring.dev ||
+           test_bit(BLKTAP_SHUTDOWN_REQUESTED, &tap->dev_inuse)) {
+               err = -ENODEV;
+               goto out;
+       }
+
+       if (!test_bit(BLKTAP_PAUSED, &tap->dev_inuse)) {
+               err = -EINVAL;
+               goto out;
+       }
+
+       err = blktap_device_resume(tap);
+       if (!err) {
+               class_device_remove_file(dev, &class_device_attr_resume);
+               class_device_create_file(dev, &class_device_attr_pause);
+       }
+
+out:
+       blktap_sysfs_exit(tap);
+
+       BTDBG("returning %d\n", (err ? err : size));
+       return (err ? err : size);
+}
+
+#ifdef ENABLE_PASSTHROUGH
+static ssize_t
+blktap_sysfs_enable_passthrough(struct class_device *dev,
+                               const char *buf, size_t size)
+{
+       int err;
+       unsigned major, minor;
+       struct blktap *tap = (struct blktap *)dev->class_data;
+
+       BTINFO("passthrough request enabled\n");
+
+       blktap_sysfs_enter(tap);
+
+       if (!tap->ring.dev ||
+           test_bit(BLKTAP_SHUTDOWN_REQUESTED, &tap->dev_inuse)) {
+               err = -ENODEV;
+               goto out;
+       }
+
+       if (!test_bit(BLKTAP_PAUSED, &tap->dev_inuse)) {
+               err = -EINVAL;
+               goto out;
+       }
+
+       if (test_bit(BLKTAP_PASSTHROUGH, &tap->dev_inuse)) {
+               err = -EINVAL;
+               goto out;
+       }
+
+       err = sscanf(buf, "%x:%x", &major, &minor);
+       if (err != 2) {
+               err = -EINVAL;
+               goto out;
+       }
+
+       err = blktap_device_enable_passthrough(tap, major, minor);
+
+out:
+       blktap_sysfs_exit(tap);
+       BTDBG("returning %d\n", (err ? err : size));
+       return (err ? err : size);
+}
+#endif
+
+static ssize_t
+blktap_sysfs_debug_device(struct class_device *dev, char *buf)
+{
+       char *tmp;
+       int i, ret;
+       struct blktap *tap = (struct blktap *)dev->class_data;
+
+       tmp = buf;
+       blktap_sysfs_get(tap);
+
+       if (!tap->ring.dev) {
+               ret = sprintf(tmp, "no device\n");
+               goto out;
+       }
+
+       tmp += sprintf(tmp, "%s (%u:%u), refcnt: %d, dev_inuse: 0x%08lx\n",
+                      tap->params.name, MAJOR(tap->ring.devno),
+                      MINOR(tap->ring.devno), atomic_read(&tap->refcnt),
+                      tap->dev_inuse);
+       tmp += sprintf(tmp, "capacity: 0x%llx, sector size: 0x%lx, "
+                      "device users: %d\n", tap->params.capacity,
+                      tap->params.sector_size, tap->device.users);
+
+       down_read(&tap->tap_sem);
+
+       tmp += sprintf(tmp, "pending requests: %d\n", tap->pending_cnt);
+       for (i = 0; i < MAX_PENDING_REQS; i++) {
+               struct blktap_request *req = tap->pending_requests[i];
+               if (!req)
+                       continue;
+
+               tmp += sprintf(tmp, "req %d: id: %llu, usr_idx: %d, "
+                              "status: 0x%02x, pendcnt: %d, "
+                              "nr_pages: %u, op: %d, time: %lu:%lu\n",
+                              i, req->id, req->usr_idx,
+                              req->status, atomic_read(&req->pendcnt),
+                              req->nr_pages, req->operation, req->time.tv_sec,
+                              req->time.tv_usec);
+       }
+
+       up_read(&tap->tap_sem);
+       ret = (tmp - buf) + 1;
+
+out:
+       blktap_sysfs_put(tap);
+       BTDBG("%s\n", buf);
+
+       return ret;
+}
+CLASS_DEVICE_ATTR(debug, S_IRUSR, blktap_sysfs_debug_device, NULL);
+
+int
+blktap_sysfs_create(struct blktap *tap)
+{
+       struct blktap_ring *ring;
+       struct class_device *dev;
+
+       if (!class)
+               return -ENODEV;
+
+       ring = &tap->ring;
+
+       dev = class_device_create(class, NULL, ring->devno,
+                                 NULL, "blktap%d", tap->minor);
+       if (IS_ERR(dev))
+               return PTR_ERR(dev);
+
+       ring->dev       = dev;
+       dev->class_data = tap;
+
+       mutex_init(&ring->sysfs_mutex);
+       atomic_set(&ring->sysfs_refcnt, 0);
+       set_bit(BLKTAP_SYSFS, &tap->dev_inuse);
+
+       class_device_create_file(dev, &class_device_attr_name);
+       class_device_create_file(dev, &class_device_attr_remove);
+       class_device_create_file(dev, &class_device_attr_pause);
+       class_device_create_file(dev, &class_device_attr_debug);
+
+       return 0;
+}
+
+int
+blktap_sysfs_destroy(struct blktap *tap)
+{
+       struct blktap_ring *ring;
+       struct class_device *dev;
+
+       ring = &tap->ring;
+       dev  = ring->dev;
+       if (!class || !dev)
+               return 0;
+
+       ring->dev = NULL;
+       if (wait_event_interruptible(sysfs_wq,
+                                    !atomic_read(&tap->ring.sysfs_refcnt)))
+               return -EAGAIN;
+
+       /* XXX: is it safe to remove the class from a sysfs attribute? */
+       class_device_remove_file(dev, &class_device_attr_name);
+       class_device_remove_file(dev, &class_device_attr_remove);
+       class_device_remove_file(dev, &class_device_attr_pause);
+       class_device_remove_file(dev, &class_device_attr_resume);
+       class_device_remove_file(dev, &class_device_attr_debug);
+       class_device_destroy(class, ring->devno);
+
+       clear_bit(BLKTAP_SYSFS, &tap->dev_inuse);
+
+       return 0;
+}
+
+static ssize_t
+blktap_sysfs_show_verbosity(struct class *class, char *buf)
+{
+       return sprintf(buf, "%d\n", blktap_debug_level);
+}
+
+static ssize_t
+blktap_sysfs_set_verbosity(struct class *class, const char *buf, size_t size)
+{
+       int level;
+
+       if (sscanf(buf, "%d", &level) == 1) {
+               blktap_debug_level = level;
+               return size;
+       }
+
+       return -EINVAL;
+}
+CLASS_ATTR(verbosity, S_IRUSR | S_IWUSR,
+          blktap_sysfs_show_verbosity, blktap_sysfs_set_verbosity);
+
+static ssize_t
+blktap_sysfs_show_devices(struct class *class, char *buf)
+{
+       int i, ret;
+       struct blktap *tap;
+
+       ret = 0;
+       for (i = 0; i < MAX_BLKTAP_DEVICE; i++) {
+               tap = blktaps[i];
+               if (!tap)
+                       continue;
+
+               if (!test_bit(BLKTAP_DEVICE, &tap->dev_inuse))
+                       continue;
+
+               ret += sprintf(buf + ret, "%d ", tap->minor);
+               ret += snprintf(buf + ret, sizeof(tap->params.name) - 1,
+                               tap->params.name);
+               ret += sprintf(buf + ret, "\n");
+       }
+
+       return ret;
+}
+CLASS_ATTR(devices, S_IRUSR, blktap_sysfs_show_devices, NULL);
+
+void
+blktap_sysfs_free(void)
+{
+       if (!class)
+               return;
+
+       class_remove_file(class, &class_attr_verbosity);
+       class_remove_file(class, &class_attr_devices);
+
+       class_destroy(class);
+}
+
+int
+blktap_sysfs_init(void)
+{
+       struct class *cls;
+
+       if (class)
+               return -EEXIST;
+
+       cls = class_create(THIS_MODULE, "blktap2");
+       if (IS_ERR(cls))
+               return PTR_ERR(cls);
+
+       class_create_file(cls, &class_attr_verbosity);
+       class_create_file(cls, &class_attr_devices);
+
+       class = cls;
+       return 0;
+}
diff -r f3a935eb30e0 -r eba6fe6d8d53 drivers/xen/blktap2/wait_queue.c
--- /dev/null   Thu Jan 01 00:00:00 1970 +0000
+++ b/drivers/xen/blktap2/wait_queue.c  Tue May 26 11:23:16 2009 +0100
@@ -0,0 +1,40 @@
+#include <linux/list.h>
+#include <linux/spinlock.h>
+
+#include "blktap.h"
+
+static LIST_HEAD(deferred_work_queue);
+static DEFINE_SPINLOCK(deferred_work_lock);
+
+void
+blktap_run_deferred(void)
+{
+       LIST_HEAD(queue);
+       struct blktap *tap;
+       unsigned long flags;
+
+       spin_lock_irqsave(&deferred_work_lock, flags);
+       list_splice_init(&deferred_work_queue, &queue);
+       list_for_each_entry(tap, &queue, deferred_queue)
+               clear_bit(BLKTAP_DEFERRED, &tap->dev_inuse);
+       spin_unlock_irqrestore(&deferred_work_lock, flags);
+
+       while (!list_empty(&queue)) {
+               tap = list_entry(queue.next, struct blktap, deferred_queue);
+               list_del_init(&tap->deferred_queue);
+               blktap_device_restart(tap);
+       }
+}
+
+void
+blktap_defer(struct blktap *tap)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&deferred_work_lock, flags);
+       if (!test_bit(BLKTAP_DEFERRED, &tap->dev_inuse)) {
+               set_bit(BLKTAP_DEFERRED, &tap->dev_inuse);
+               list_add_tail(&tap->deferred_queue, &deferred_work_queue);
+       }
+       spin_unlock_irqrestore(&deferred_work_lock, flags);
+}
diff -r f3a935eb30e0 -r eba6fe6d8d53 include/linux/mm.h
--- a/include/linux/mm.h        Tue May 26 09:53:55 2009 +0100
+++ b/include/linux/mm.h        Tue May 26 11:23:16 2009 +0100
@@ -166,6 +166,9 @@ extern unsigned int kobjsize(const void 
 #define VM_INSERTPAGE  0x02000000      /* The vma has had "vm_insert_page()" 
done on it */
 #ifdef CONFIG_XEN
 #define VM_FOREIGN     0x04000000      /* Has pages belonging to another VM */
+struct vm_foreign_map {
+        struct page **map;
+};
 #endif
 #define VM_ALWAYSDUMP  0x08000000      /* Always include in core dumps */
 
@@ -210,6 +213,10 @@ struct vm_operations_struct {
         * original value of @ptep. */
        pte_t (*zap_pte)(struct vm_area_struct *vma, 
                         unsigned long addr, pte_t *ptep, int is_fullmm);
+
+        /* called before close() to indicate no more pages should be mapped */
+        void (*unmap)(struct vm_area_struct *area);
+
 #ifdef CONFIG_NUMA
        int (*set_policy)(struct vm_area_struct *vma, struct mempolicy *new);
        struct mempolicy *(*get_policy)(struct vm_area_struct *vma,
diff -r f3a935eb30e0 -r eba6fe6d8d53 include/linux/page-flags.h
--- a/include/linux/page-flags.h        Tue May 26 09:53:55 2009 +0100
+++ b/include/linux/page-flags.h        Tue May 26 11:23:16 2009 +0100
@@ -99,6 +99,16 @@
 #endif
 
 #define PG_foreign             20      /* Page is owned by foreign allocator. 
*/
+
+#define PG_netback              21      /* Page is owned by netback */
+#define PageNetback(page)       test_bit(PG_netback, &(page)->flags)
+#define SetPageNetback(page)    set_bit(PG_netback, &(page)->flags)
+#define ClearPageNetback(page)  clear_bit(PG_netback, &(page)->flags)
+
+#define PG_blkback              22      /* Page is owned by blkback */
+#define PageBlkback(page)       test_bit(PG_blkback, &(page)->flags)
+#define SetPageBlkback(page)    set_bit(PG_blkback, &(page)->flags)
+#define ClearPageBlkback(page)  clear_bit(PG_blkback, &(page)->flags)
 
 /*
  * Manipulation of page state flags
diff -r f3a935eb30e0 -r eba6fe6d8d53 mm/memory.c
--- a/mm/memory.c       Tue May 26 09:53:55 2009 +0100
+++ b/mm/memory.c       Tue May 26 11:23:16 2009 +0100
@@ -1045,7 +1045,9 @@ int get_user_pages(struct task_struct *t
 
 #ifdef CONFIG_XEN
                if (vma && (vma->vm_flags & VM_FOREIGN)) {
-                       struct page **map = vma->vm_private_data;
+                       struct vm_foreign_map *foreign_map =
+                               vma->vm_private_data;
+                       struct page **map = foreign_map->map;
                        int offset = (start - vma->vm_start) >> PAGE_SHIFT;
                        if (map[offset] != NULL) {
                                if (pages) {
diff -r f3a935eb30e0 -r eba6fe6d8d53 mm/mmap.c
--- a/mm/mmap.c Tue May 26 09:53:55 2009 +0100
+++ b/mm/mmap.c Tue May 26 11:23:16 2009 +0100
@@ -1687,6 +1687,12 @@ static void unmap_region(struct mm_struc
        tlb_finish_mmu(tlb, start, end);
 }
 
+static inline void unmap_vma(struct vm_area_struct *vma)
+{
+       if (unlikely(vma->vm_ops && vma->vm_ops->unmap))
+               vma->vm_ops->unmap(vma);
+}
+
 /*
  * Create a list of vma's touched by the unmap, removing them from the mm's
  * vma list as we go..
@@ -1702,6 +1708,7 @@ detach_vmas_to_be_unmapped(struct mm_str
        insertion_point = (prev ? &prev->vm_next : &mm->mmap);
        do {
                rb_erase(&vma->vm_rb, &mm->mm_rb);
+               unmap_vma(vma);
                mm->map_count--;
                tail_vma = vma;
                vma = vma->vm_next;
@@ -1959,13 +1966,16 @@ void exit_mmap(struct mm_struct *mm)
 void exit_mmap(struct mm_struct *mm)
 {
        struct mmu_gather *tlb;
-       struct vm_area_struct *vma = mm->mmap;
+       struct vm_area_struct *vma_tmp, *vma = mm->mmap;
        unsigned long nr_accounted = 0;
        unsigned long end;
 
 #ifdef arch_exit_mmap
        arch_exit_mmap(mm);
 #endif
+
+       for (vma_tmp = mm->mmap; vma_tmp; vma_tmp = vma_tmp->vm_next)
+               unmap_vma(vma_tmp);
 
        lru_add_drain();
        flush_cache_mm(mm);

_______________________________________________
Xen-changelog mailing list
Xen-changelog@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-changelog


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.