Revived GUEST_BASE support for usermode emulation targets [v5]
[qemu] / block.c
diff --git a/block.c b/block.c
index 3cb3cc2..c7e0dcb 100644 (file)
--- a/block.c
+++ b/block.c
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  * THE SOFTWARE.
  */
-#include "qemu-common.h"
-#ifndef QEMU_IMG
-#include "console.h"
+#include "config-host.h"
+#ifdef HOST_BSD
+/* include native header before sys-queue.h */
+#include <sys/queue.h>
 #endif
+
+#include "qemu-common.h"
+#include "monitor.h"
 #include "block_int.h"
+#include "module.h"
 
-#ifdef _BSD
+#ifdef HOST_BSD
 #include <sys/types.h>
 #include <sys/stat.h>
 #include <sys/ioctl.h>
-#include <sys/queue.h>
+#ifndef __DragonFly__
 #include <sys/disk.h>
 #endif
+#endif
+
+#ifdef _WIN32
+#include <windows.h>
+#endif
 
 #define SECTOR_BITS 9
 #define SECTOR_SIZE (1 << SECTOR_BITS)
 
-typedef struct BlockDriverAIOCBSync {
-    BlockDriverAIOCB common;
-    QEMUBH *bh;
-    int ret;
-} BlockDriverAIOCBSync;
-
-static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
-        int64_t sector_num, uint8_t *buf, int nb_sectors,
+static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque);
-static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
-        int64_t sector_num, const uint8_t *buf, int nb_sectors,
+static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque);
-static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb);
 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
                         uint8_t *buf, int nb_sectors);
 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
                          const uint8_t *buf, int nb_sectors);
 
 BlockDriverState *bdrv_first;
+
 static BlockDriver *first_drv;
 
 int path_is_absolute(const char *path)
@@ -123,16 +127,13 @@ void path_combine(char *dest, int dest_size,
     }
 }
 
-
-static void bdrv_register(BlockDriver *bdrv)
+void bdrv_register(BlockDriver *bdrv)
 {
-    if (!bdrv->bdrv_aio_read) {
+    if (!bdrv->bdrv_aio_readv) {
         /* add AIO emulation layer */
-        bdrv->bdrv_aio_read = bdrv_aio_read_em;
-        bdrv->bdrv_aio_write = bdrv_aio_write_em;
-        bdrv->bdrv_aio_cancel = bdrv_aio_cancel_em;
-        bdrv->aiocb_size = sizeof(BlockDriverAIOCBSync);
-    } else if (!bdrv->bdrv_read && !bdrv->bdrv_pread) {
+        bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
+        bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
+    } else if (!bdrv->bdrv_read) {
         /* add synchronous IO emulation layer */
         bdrv->bdrv_read = bdrv_read_em;
         bdrv->bdrv_write = bdrv_write_em;
@@ -147,8 +148,6 @@ BlockDriverState *bdrv_new(const char *device_name)
     BlockDriverState **pbs, *bs;
 
     bs = qemu_mallocz(sizeof(BlockDriverState));
-    if(!bs)
-        return NULL;
     pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
     if (device_name[0] != '\0') {
         /* insert at the end */
@@ -170,13 +169,13 @@ BlockDriver *bdrv_find_format(const char *format_name)
     return NULL;
 }
 
-int bdrv_create(BlockDriver *drv,
-                const char *filename, int64_t size_in_sectors,
-                const char *backing_file, int flags)
+int bdrv_create(BlockDriver *drv, const char* filename,
+    QEMUOptionParameter *options)
 {
     if (!drv->bdrv_create)
         return -ENOTSUP;
-    return drv->bdrv_create(filename, size_in_sectors, backing_file, flags);
+
+    return drv->bdrv_create(filename, options);
 }
 
 #ifdef _WIN32
@@ -191,8 +190,12 @@ void get_tmp_filename(char *filename, int size)
 void get_tmp_filename(char *filename, int size)
 {
     int fd;
+    const char *tmpdir;
     /* XXX: race condition possible */
-    pstrcpy(filename, size, "/tmp/vl.XXXXXX");
+    tmpdir = getenv("TMPDIR");
+    if (!tmpdir)
+        tmpdir = "/tmp";
+    snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
     fd = mkstemp(filename);
     close(fd);
 }
@@ -206,7 +209,7 @@ static int is_windows_drive_prefix(const char *filename)
             filename[1] == ':');
 }
 
-static int is_windows_drive(const char *filename)
+int is_windows_drive(const char *filename)
 {
     if (is_windows_drive_prefix(filename) &&
         filename[2] == '\0')
@@ -228,11 +231,11 @@ static BlockDriver *find_protocol(const char *filename)
 #ifdef _WIN32
     if (is_windows_drive(filename) ||
         is_windows_drive_prefix(filename))
-        return &bdrv_raw;
+        return bdrv_find_format("raw");
 #endif
     p = strchr(filename, ':');
     if (!p)
-        return &bdrv_raw;
+        return bdrv_find_format("raw");
     len = p - filename;
     if (len > sizeof(protocol) - 1)
         len = sizeof(protocol) - 1;
@@ -246,8 +249,28 @@ static BlockDriver *find_protocol(const char *filename)
     return NULL;
 }
 
-/* XXX: force raw format if block or character device ? It would
-   simplify the BSD case */
+/*
+ * Detect host devices. By convention, /dev/cdrom[N] is always
+ * recognized as a host CDROM.
+ */
+static BlockDriver *find_hdev_driver(const char *filename)
+{
+    int score_max = 0, score;
+    BlockDriver *drv = NULL, *d;
+
+    for (d = first_drv; d; d = d->next) {
+        if (d->bdrv_probe_device) {
+            score = d->bdrv_probe_device(filename);
+            if (score > score_max) {
+                score_max = score;
+                drv = d;
+            }
+        }
+    }
+
+    return drv;
+}
+
 static BlockDriver *find_image_format(const char *filename)
 {
     int ret, score, score_max;
@@ -255,26 +278,9 @@ static BlockDriver *find_image_format(const char *filename)
     uint8_t buf[2048];
     BlockDriverState *bs;
 
-    /* detect host devices. By convention, /dev/cdrom[N] is always
-       recognized as a host CDROM */
-    if (strstart(filename, "/dev/cdrom", NULL))
-        return &bdrv_host_device;
-#ifdef _WIN32
-    if (is_windows_drive(filename))
-        return &bdrv_host_device;
-#else
-    {
-        struct stat st;
-        if (stat(filename, &st) >= 0 &&
-            (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
-            return &bdrv_host_device;
-        }
-    }
-#endif
-
     drv = find_protocol(filename);
     /* no need to test disk image formats for vvfat */
-    if (drv == &bdrv_vvfat)
+    if (drv && strcmp(drv->format_name, "vvfat") == 0)
         return drv;
 
     ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
@@ -305,13 +311,12 @@ int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
     int ret;
 
     bs = bdrv_new("");
-    if (!bs)
-        return -ENOMEM;
     ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
     if (ret < 0) {
         bdrv_delete(bs);
         return ret;
     }
+    bs->growable = 1;
     *pbs = bs;
     return 0;
 }
@@ -331,67 +336,96 @@ int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
     bs->read_only = 0;
     bs->is_temporary = 0;
     bs->encrypted = 0;
+    bs->valid_key = 0;
+    /* buffer_alignment defaulted to 512, drivers can change this value */
+    bs->buffer_alignment = 512;
 
     if (flags & BDRV_O_SNAPSHOT) {
         BlockDriverState *bs1;
         int64_t total_size;
+        int is_protocol = 0;
+        BlockDriver *bdrv_qcow2;
+        QEMUOptionParameter *options;
 
         /* if snapshot, we create a temporary backing file and open it
            instead of opening 'filename' directly */
 
         /* if there is a backing file, use it */
         bs1 = bdrv_new("");
-        if (!bs1) {
-            return -ENOMEM;
-        }
-        if (bdrv_open(bs1, filename, 0) < 0) {
+        ret = bdrv_open2(bs1, filename, 0, drv);
+        if (ret < 0) {
             bdrv_delete(bs1);
-            return -1;
+            return ret;
         }
         total_size = bdrv_getlength(bs1) >> SECTOR_BITS;
+
+        if (bs1->drv && bs1->drv->protocol_name)
+            is_protocol = 1;
+
         bdrv_delete(bs1);
 
         get_tmp_filename(tmp_filename, sizeof(tmp_filename));
-        realpath(filename, backing_filename);
-        if (bdrv_create(&bdrv_qcow2, tmp_filename,
-                        total_size, backing_filename, 0) < 0) {
-            return -1;
+
+        /* Real path is meaningless for protocols */
+        if (is_protocol)
+            snprintf(backing_filename, sizeof(backing_filename),
+                     "%s", filename);
+        else
+            realpath(filename, backing_filename);
+
+        bdrv_qcow2 = bdrv_find_format("qcow2");
+        options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
+
+        set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size * 512);
+        set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
+        if (drv) {
+            set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
+                drv->format_name);
+        }
+
+        ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
+        if (ret < 0) {
+            return ret;
         }
+
         filename = tmp_filename;
+        drv = bdrv_qcow2;
         bs->is_temporary = 1;
     }
 
     pstrcpy(bs->filename, sizeof(bs->filename), filename);
     if (flags & BDRV_O_FILE) {
         drv = find_protocol(filename);
-        if (!drv)
-            return -ENOENT;
-    } else {
+    } else if (!drv) {
+        drv = find_hdev_driver(filename);
         if (!drv) {
             drv = find_image_format(filename);
-            if (!drv)
-                return -1;
         }
     }
+    if (!drv) {
+        ret = -ENOENT;
+        goto unlink_and_fail;
+    }
     bs->drv = drv;
     bs->opaque = qemu_mallocz(drv->instance_size);
-    if (bs->opaque == NULL && drv->instance_size > 0)
-        return -1;
     /* Note: for compatibility, we open disk image files as RDWR, and
        RDONLY as fallback */
     if (!(flags & BDRV_O_FILE))
-        open_flags = BDRV_O_RDWR;
+        open_flags = BDRV_O_RDWR | (flags & BDRV_O_CACHE_MASK);
     else
         open_flags = flags & ~(BDRV_O_FILE | BDRV_O_SNAPSHOT);
     ret = drv->bdrv_open(bs, filename, open_flags);
-    if (ret == -EACCES && !(flags & BDRV_O_FILE)) {
-        ret = drv->bdrv_open(bs, filename, BDRV_O_RDONLY);
+    if ((ret == -EACCES || ret == -EPERM) && !(flags & BDRV_O_FILE)) {
+        ret = drv->bdrv_open(bs, filename, open_flags & ~BDRV_O_RDWR);
         bs->read_only = 1;
     }
     if (ret < 0) {
         qemu_free(bs->opaque);
         bs->opaque = NULL;
         bs->drv = NULL;
+    unlink_and_fail:
+        if (bs->is_temporary)
+            unlink(filename);
         return ret;
     }
     if (drv->bdrv_getlength) {
@@ -404,23 +438,26 @@ int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
 #endif
     if (bs->backing_file[0] != '\0') {
         /* if there is a backing file, use it */
+        BlockDriver *back_drv = NULL;
         bs->backing_hd = bdrv_new("");
-        if (!bs->backing_hd) {
-        fail:
-            bdrv_close(bs);
-            return -ENOMEM;
-        }
         path_combine(backing_filename, sizeof(backing_filename),
                      filename, bs->backing_file);
-        if (bdrv_open(bs->backing_hd, backing_filename, 0) < 0)
-            goto fail;
+        if (bs->backing_format[0] != '\0')
+            back_drv = bdrv_find_format(bs->backing_format);
+        ret = bdrv_open2(bs->backing_hd, backing_filename, open_flags,
+                         back_drv);
+        if (ret < 0) {
+            bdrv_close(bs);
+            return ret;
+        }
     }
 
-    /* call the change callback */
-    bs->media_changed = 1;
-    if (bs->change_cb)
-        bs->change_cb(bs->change_opaque);
-
+    if (!bdrv_key_required(bs)) {
+        /* call the change callback */
+        bs->media_changed = 1;
+        if (bs->change_cb)
+            bs->change_cb(bs->change_opaque);
+    }
     return 0;
 }
 
@@ -448,11 +485,32 @@ void bdrv_close(BlockDriverState *bs)
 
 void bdrv_delete(BlockDriverState *bs)
 {
-    /* XXX: remove the driver list */
+    BlockDriverState **pbs;
+
+    pbs = &bdrv_first;
+    while (*pbs != bs && *pbs != NULL)
+        pbs = &(*pbs)->next;
+    if (*pbs == bs)
+        *pbs = bs->next;
+
     bdrv_close(bs);
     qemu_free(bs);
 }
 
+/*
+ * Run consistency checks on an image
+ *
+ * Returns the number of errors or -errno when an internal error occurs
+ */
+int bdrv_check(BlockDriverState *bs)
+{
+    if (bs->drv->bdrv_check == NULL) {
+        return -ENOTSUP;
+    }
+
+    return bs->drv->bdrv_check(bs);
+}
+
 /* commit COW file into the raw image */
 int bdrv_commit(BlockDriverState *bs)
 {
@@ -496,6 +554,34 @@ int bdrv_commit(BlockDriverState *bs)
     return 0;
 }
 
+static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
+                                   size_t size)
+{
+    int64_t len;
+
+    if (!bdrv_is_inserted(bs))
+        return -ENOMEDIUM;
+
+    if (bs->growable)
+        return 0;
+
+    len = bdrv_getlength(bs);
+
+    if (offset < 0)
+        return -EIO;
+
+    if ((offset > len) || (len - offset < size))
+        return -EIO;
+
+    return 0;
+}
+
+static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
+                              int nb_sectors)
+{
+    return bdrv_check_byte_request(bs, sector_num * 512, nb_sectors * 512);
+}
+
 /* return < 0 if error. See bdrv_write() for the return codes */
 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
               uint8_t *buf, int nb_sectors)
@@ -504,31 +590,10 @@ int bdrv_read(BlockDriverState *bs, int64_t sector_num,
 
     if (!drv)
         return -ENOMEDIUM;
+    if (bdrv_check_request(bs, sector_num, nb_sectors))
+        return -EIO;
 
-    if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
-            memcpy(buf, bs->boot_sector_data, 512);
-        sector_num++;
-        nb_sectors--;
-        buf += 512;
-        if (nb_sectors == 0)
-            return 0;
-    }
-    if (drv->bdrv_pread) {
-        int ret, len;
-        len = nb_sectors * 512;
-        ret = drv->bdrv_pread(bs, sector_num * 512, buf, len);
-        if (ret < 0)
-            return ret;
-        else if (ret != len)
-            return -EINVAL;
-        else {
-           bs->rd_bytes += (unsigned) len;
-           bs->rd_ops ++;
-            return 0;
-       }
-    } else {
-        return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
-    }
+    return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
 }
 
 /* Return < 0 if error. Important errors are:
@@ -545,29 +610,14 @@ int bdrv_write(BlockDriverState *bs, int64_t sector_num,
         return -ENOMEDIUM;
     if (bs->read_only)
         return -EACCES;
-    if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
-        memcpy(bs->boot_sector_data, buf, 512);
-    }
-    if (drv->bdrv_pwrite) {
-        int ret, len;
-        len = nb_sectors * 512;
-        ret = drv->bdrv_pwrite(bs, sector_num * 512, buf, len);
-        if (ret < 0)
-            return ret;
-        else if (ret != len)
-            return -EIO;
-        else {
-           bs->wr_bytes += (unsigned) len;
-           bs->wr_ops ++;
-            return 0;
-       }
-    } else {
-        return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
-    }
+    if (bdrv_check_request(bs, sector_num, nb_sectors))
+        return -EIO;
+
+    return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
 }
 
-static int bdrv_pread_em(BlockDriverState *bs, int64_t offset,
-                         uint8_t *buf, int count1)
+int bdrv_pread(BlockDriverState *bs, int64_t offset,
+               void *buf, int count1)
 {
     uint8_t tmp_buf[SECTOR_SIZE];
     int len, nb_sectors, count;
@@ -610,8 +660,8 @@ static int bdrv_pread_em(BlockDriverState *bs, int64_t offset,
     return count1;
 }
 
-static int bdrv_pwrite_em(BlockDriverState *bs, int64_t offset,
-                          const uint8_t *buf, int count1)
+int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
+                const void *buf, int count1)
 {
     uint8_t tmp_buf[SECTOR_SIZE];
     int len, nb_sectors, count;
@@ -659,36 +709,6 @@ static int bdrv_pwrite_em(BlockDriverState *bs, int64_t offset,
 }
 
 /**
- * Read with byte offsets (needed only for file protocols)
- */
-int bdrv_pread(BlockDriverState *bs, int64_t offset,
-               void *buf1, int count1)
-{
-    BlockDriver *drv = bs->drv;
-
-    if (!drv)
-        return -ENOMEDIUM;
-    if (!drv->bdrv_pread)
-        return bdrv_pread_em(bs, offset, buf1, count1);
-    return drv->bdrv_pread(bs, offset, buf1, count1);
-}
-
-/**
- * Write with byte offsets (needed only for file protocols)
- */
-int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
-                const void *buf1, int count1)
-{
-    BlockDriver *drv = bs->drv;
-
-    if (!drv)
-        return -ENOMEDIUM;
-    if (!drv->bdrv_pwrite)
-        return bdrv_pwrite_em(bs, offset, buf1, count1);
-    return drv->bdrv_pwrite(bs, offset, buf1, count1);
-}
-
-/**
  * Truncate file to 'offset' bytes (needed only for file protocols)
  */
 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
@@ -717,7 +737,7 @@ int64_t bdrv_getlength(BlockDriverState *bs)
 }
 
 /* return 0 as number of sectors if no device present or error */
-void bdrv_get_geometry(BlockDriverState *bs, int64_t *nb_sectors_ptr)
+void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
 {
     int64_t length;
     length = bdrv_getlength(bs);
@@ -728,14 +748,120 @@ void bdrv_get_geometry(BlockDriverState *bs, int64_t *nb_sectors_ptr)
     *nb_sectors_ptr = length;
 }
 
-/* force a given boot sector. */
-void bdrv_set_boot_sector(BlockDriverState *bs, const uint8_t *data, int size)
+struct partition {
+        uint8_t boot_ind;           /* 0x80 - active */
+        uint8_t head;               /* starting head */
+        uint8_t sector;             /* starting sector */
+        uint8_t cyl;                /* starting cylinder */
+        uint8_t sys_ind;            /* What partition type */
+        uint8_t end_head;           /* end head */
+        uint8_t end_sector;         /* end sector */
+        uint8_t end_cyl;            /* end cylinder */
+        uint32_t start_sect;        /* starting sector counting from 0 */
+        uint32_t nr_sects;          /* nr of sectors in partition */
+} __attribute__((packed));
+
+/* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
+static int guess_disk_lchs(BlockDriverState *bs,
+                           int *pcylinders, int *pheads, int *psectors)
+{
+    uint8_t buf[512];
+    int ret, i, heads, sectors, cylinders;
+    struct partition *p;
+    uint32_t nr_sects;
+    uint64_t nb_sectors;
+
+    bdrv_get_geometry(bs, &nb_sectors);
+
+    ret = bdrv_read(bs, 0, buf, 1);
+    if (ret < 0)
+        return -1;
+    /* test msdos magic */
+    if (buf[510] != 0x55 || buf[511] != 0xaa)
+        return -1;
+    for(i = 0; i < 4; i++) {
+        p = ((struct partition *)(buf + 0x1be)) + i;
+        nr_sects = le32_to_cpu(p->nr_sects);
+        if (nr_sects && p->end_head) {
+            /* We make the assumption that the partition terminates on
+               a cylinder boundary */
+            heads = p->end_head + 1;
+            sectors = p->end_sector & 63;
+            if (sectors == 0)
+                continue;
+            cylinders = nb_sectors / (heads * sectors);
+            if (cylinders < 1 || cylinders > 16383)
+                continue;
+            *pheads = heads;
+            *psectors = sectors;
+            *pcylinders = cylinders;
+#if 0
+            printf("guessed geometry: LCHS=%d %d %d\n",
+                   cylinders, heads, sectors);
+#endif
+            return 0;
+        }
+    }
+    return -1;
+}
+
+void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
 {
-    bs->boot_sector_enabled = 1;
-    if (size > 512)
-        size = 512;
-    memcpy(bs->boot_sector_data, data, size);
-    memset(bs->boot_sector_data + size, 0, 512 - size);
+    int translation, lba_detected = 0;
+    int cylinders, heads, secs;
+    uint64_t nb_sectors;
+
+    /* if a geometry hint is available, use it */
+    bdrv_get_geometry(bs, &nb_sectors);
+    bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
+    translation = bdrv_get_translation_hint(bs);
+    if (cylinders != 0) {
+        *pcyls = cylinders;
+        *pheads = heads;
+        *psecs = secs;
+    } else {
+        if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
+            if (heads > 16) {
+                /* if heads > 16, it means that a BIOS LBA
+                   translation was active, so the default
+                   hardware geometry is OK */
+                lba_detected = 1;
+                goto default_geometry;
+            } else {
+                *pcyls = cylinders;
+                *pheads = heads;
+                *psecs = secs;
+                /* disable any translation to be in sync with
+                   the logical geometry */
+                if (translation == BIOS_ATA_TRANSLATION_AUTO) {
+                    bdrv_set_translation_hint(bs,
+                                              BIOS_ATA_TRANSLATION_NONE);
+                }
+            }
+        } else {
+        default_geometry:
+            /* if no geometry, use a standard physical disk geometry */
+            cylinders = nb_sectors / (16 * 63);
+
+            if (cylinders > 16383)
+                cylinders = 16383;
+            else if (cylinders < 2)
+                cylinders = 2;
+            *pcyls = cylinders;
+            *pheads = 16;
+            *psecs = 63;
+            if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
+                if ((*pcyls * *pheads) <= 131072) {
+                    bdrv_set_translation_hint(bs,
+                                              BIOS_ATA_TRANSLATION_LARGE);
+                } else {
+                    bdrv_set_translation_hint(bs,
+                                              BIOS_ATA_TRANSLATION_LBA);
+                }
+            }
+        }
+        bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
+    }
 }
 
 void bdrv_set_geometry_hint(BlockDriverState *bs,
@@ -786,6 +912,11 @@ int bdrv_is_read_only(BlockDriverState *bs)
     return bs->read_only;
 }
 
+int bdrv_is_sg(BlockDriverState *bs)
+{
+    return bs->sg;
+}
+
 /* XXX: no longer used */
 void bdrv_set_change_cb(BlockDriverState *bs,
                         void (*change_cb)(void *opaque), void *opaque)
@@ -801,6 +932,15 @@ int bdrv_is_encrypted(BlockDriverState *bs)
     return bs->encrypted;
 }
 
+int bdrv_key_required(BlockDriverState *bs)
+{
+    BlockDriverState *backing_hd = bs->backing_hd;
+
+    if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
+        return 1;
+    return (bs->encrypted && !bs->valid_key);
+}
+
 int bdrv_set_key(BlockDriverState *bs, const char *key)
 {
     int ret;
@@ -813,7 +953,17 @@ int bdrv_set_key(BlockDriverState *bs, const char *key)
     }
     if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
         return -1;
-    return bs->drv->bdrv_set_key(bs, key);
+    ret = bs->drv->bdrv_set_key(bs, key);
+    if (ret < 0) {
+        bs->valid_key = 0;
+    } else if (!bs->valid_key) {
+        bs->valid_key = 1;
+        /* call the change callback now, we skipped it on open */
+        bs->media_changed = 1;
+        if (bs->change_cb)
+            bs->change_cb(bs->change_opaque);
+    }
+    return ret;
 }
 
 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
@@ -846,12 +996,12 @@ BlockDriverState *bdrv_find(const char *name)
     return NULL;
 }
 
-void bdrv_iterate(void (*it)(void *opaque, const char *name), void *opaque)
+void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
 {
     BlockDriverState *bs;
 
     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
-        it(opaque, bs->device_name);
+        it(opaque, bs);
     }
 }
 
@@ -862,71 +1012,117 @@ const char *bdrv_get_device_name(BlockDriverState *bs)
 
 void bdrv_flush(BlockDriverState *bs)
 {
+    if (!bs->drv)
+        return;
     if (bs->drv->bdrv_flush)
         bs->drv->bdrv_flush(bs);
     if (bs->backing_hd)
         bdrv_flush(bs->backing_hd);
 }
 
-#ifndef QEMU_IMG
-void bdrv_info(void)
+void bdrv_flush_all(void)
+{
+    BlockDriverState *bs;
+
+    for (bs = bdrv_first; bs != NULL; bs = bs->next)
+        if (bs->drv && !bdrv_is_read_only(bs) && 
+            (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
+            bdrv_flush(bs);
+}
+
+/*
+ * Returns true iff the specified sector is present in the disk image. Drivers
+ * not implementing the functionality are assumed to not support backing files,
+ * hence all their sectors are reported as allocated.
+ *
+ * 'pnum' is set to the number of sectors (including and immediately following
+ * the specified sector) that are known to be in the same
+ * allocated/unallocated state.
+ *
+ * 'nb_sectors' is the max value 'pnum' should be set to.
+ */
+int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
+       int *pnum)
+{
+    int64_t n;
+    if (!bs->drv->bdrv_is_allocated) {
+        if (sector_num >= bs->total_sectors) {
+            *pnum = 0;
+            return 0;
+        }
+        n = bs->total_sectors - sector_num;
+        *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
+        return 1;
+    }
+    return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
+}
+
+void bdrv_info(Monitor *mon)
 {
     BlockDriverState *bs;
 
     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
-        term_printf("%s:", bs->device_name);
-        term_printf(" type=");
+        monitor_printf(mon, "%s:", bs->device_name);
+        monitor_printf(mon, " type=");
         switch(bs->type) {
         case BDRV_TYPE_HD:
-            term_printf("hd");
+            monitor_printf(mon, "hd");
             break;
         case BDRV_TYPE_CDROM:
-            term_printf("cdrom");
+            monitor_printf(mon, "cdrom");
             break;
         case BDRV_TYPE_FLOPPY:
-            term_printf("floppy");
+            monitor_printf(mon, "floppy");
             break;
         }
-        term_printf(" removable=%d", bs->removable);
+        monitor_printf(mon, " removable=%d", bs->removable);
         if (bs->removable) {
-            term_printf(" locked=%d", bs->locked);
+            monitor_printf(mon, " locked=%d", bs->locked);
         }
         if (bs->drv) {
-            term_printf(" file=");
-           term_print_filename(bs->filename);
+            monitor_printf(mon, " file=");
+            monitor_print_filename(mon, bs->filename);
             if (bs->backing_file[0] != '\0') {
-                term_printf(" backing_file=");
-               term_print_filename(bs->backing_file);
-           }
-            term_printf(" ro=%d", bs->read_only);
-            term_printf(" drv=%s", bs->drv->format_name);
-            if (bs->encrypted)
-                term_printf(" encrypted");
+                monitor_printf(mon, " backing_file=");
+                monitor_print_filename(mon, bs->backing_file);
+            }
+            monitor_printf(mon, " ro=%d", bs->read_only);
+            monitor_printf(mon, " drv=%s", bs->drv->format_name);
+            monitor_printf(mon, " encrypted=%d", bdrv_is_encrypted(bs));
         } else {
-            term_printf(" [not inserted]");
+            monitor_printf(mon, " [not inserted]");
         }
-        term_printf("\n");
+        monitor_printf(mon, "\n");
     }
 }
 
 /* The "info blockstats" command. */
-void bdrv_info_stats (void)
+void bdrv_info_stats(Monitor *mon)
 {
     BlockDriverState *bs;
 
     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
-       term_printf ("%s:"
-                    " rd_bytes=%" PRIu64
-                    " wr_bytes=%" PRIu64
-                    " rd_operations=%" PRIu64
-                    " wr_operations=%" PRIu64
-                    "\n",
-                    bs->device_name,
-                    bs->rd_bytes, bs->wr_bytes,
-                    bs->rd_ops, bs->wr_ops);
+        monitor_printf(mon, "%s:"
+                       " rd_bytes=%" PRIu64
+                       " wr_bytes=%" PRIu64
+                       " rd_operations=%" PRIu64
+                       " wr_operations=%" PRIu64
+                       "\n",
+                       bs->device_name,
+                       bs->rd_bytes, bs->wr_bytes,
+                       bs->rd_ops, bs->wr_ops);
     }
 }
-#endif
+
+const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
+{
+    if (bs->backing_hd && bs->backing_hd->encrypted)
+        return bs->backing_file;
+    else if (bs->encrypted)
+        return bs->filename;
+    else
+        return NULL;
+}
 
 void bdrv_get_backing_filename(BlockDriverState *bs,
                                char *filename, int filename_size)
@@ -946,6 +1142,8 @@ int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
         return -ENOMEDIUM;
     if (!drv->bdrv_write_compressed)
         return -ENOTSUP;
+    if (bdrv_check_request(bs, sector_num, nb_sectors))
+        return -EIO;
     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
 }
 
@@ -960,6 +1158,26 @@ int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
     return drv->bdrv_get_info(bs, bdi);
 }
 
+int bdrv_put_buffer(BlockDriverState *bs, const uint8_t *buf, int64_t pos, int size)
+{
+    BlockDriver *drv = bs->drv;
+    if (!drv)
+        return -ENOMEDIUM;
+    if (!drv->bdrv_put_buffer)
+        return -ENOTSUP;
+    return drv->bdrv_put_buffer(bs, buf, pos, size);
+}
+
+int bdrv_get_buffer(BlockDriverState *bs, uint8_t *buf, int64_t pos, int size)
+{
+    BlockDriver *drv = bs->drv;
+    if (!drv)
+        return -ENOMEDIUM;
+    if (!drv->bdrv_get_buffer)
+        return -ENOTSUP;
+    return drv->bdrv_get_buffer(bs, buf, pos, size);
+}
+
 /**************************************************************/
 /* handling of snapshots */
 
@@ -1083,25 +1301,20 @@ char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
 /**************************************************************/
 /* async I/Os */
 
-BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
-                                uint8_t *buf, int nb_sectors,
-                                BlockDriverCompletionFunc *cb, void *opaque)
+BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
+                                 QEMUIOVector *qiov, int nb_sectors,
+                                 BlockDriverCompletionFunc *cb, void *opaque)
 {
     BlockDriver *drv = bs->drv;
     BlockDriverAIOCB *ret;
 
     if (!drv)
         return NULL;
+    if (bdrv_check_request(bs, sector_num, nb_sectors))
+        return NULL;
 
-    /* XXX: we assume that nb_sectors == 0 is suppored by the async read */
-    if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
-        memcpy(buf, bs->boot_sector_data, 512);
-        sector_num++;
-        nb_sectors--;
-        buf += 512;
-    }
-
-    ret = drv->bdrv_aio_read(bs, sector_num, buf, nb_sectors, cb, opaque);
+    ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
+                              cb, opaque);
 
     if (ret) {
        /* Update stats even though technically transfer has not happened. */
@@ -1112,9 +1325,9 @@ BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
     return ret;
 }
 
-BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,
-                                 const uint8_t *buf, int nb_sectors,
-                                 BlockDriverCompletionFunc *cb, void *opaque)
+BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
+                                  QEMUIOVector *qiov, int nb_sectors,
+                                  BlockDriverCompletionFunc *cb, void *opaque)
 {
     BlockDriver *drv = bs->drv;
     BlockDriverAIOCB *ret;
@@ -1123,11 +1336,11 @@ BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,
         return NULL;
     if (bs->read_only)
         return NULL;
-    if (sector_num == 0 && bs->boot_sector_enabled && nb_sectors > 0) {
-        memcpy(bs->boot_sector_data, buf, 512);
-    }
+    if (bdrv_check_request(bs, sector_num, nb_sectors))
+        return NULL;
 
-    ret = drv->bdrv_aio_write(bs, sector_num, buf, nb_sectors, cb, opaque);
+    ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
+                               cb, opaque);
 
     if (ret) {
        /* Update stats even though technically transfer has not happened. */
@@ -1140,86 +1353,91 @@ BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,
 
 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
 {
-    BlockDriver *drv = acb->bs->drv;
-
-    drv->bdrv_aio_cancel(acb);
+    acb->pool->cancel(acb);
 }
 
 
 /**************************************************************/
 /* async block device emulation */
 
-#ifdef QEMU_IMG
-static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
-        int64_t sector_num, uint8_t *buf, int nb_sectors,
-        BlockDriverCompletionFunc *cb, void *opaque)
-{
+typedef struct BlockDriverAIOCBSync {
+    BlockDriverAIOCB common;
+    QEMUBH *bh;
     int ret;
-    ret = bdrv_read(bs, sector_num, buf, nb_sectors);
-    cb(opaque, ret);
-    return NULL;
-}
+    /* vector translation state */
+    QEMUIOVector *qiov;
+    uint8_t *bounce;
+    int is_write;
+} BlockDriverAIOCBSync;
 
-static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
-        int64_t sector_num, const uint8_t *buf, int nb_sectors,
-        BlockDriverCompletionFunc *cb, void *opaque)
+static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
 {
-    int ret;
-    ret = bdrv_write(bs, sector_num, buf, nb_sectors);
-    cb(opaque, ret);
-    return NULL;
+    BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
+    qemu_bh_cancel(acb->bh);
+    qemu_aio_release(acb);
 }
 
-static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb)
-{
-}
-#else
+static AIOPool bdrv_em_aio_pool = {
+    .aiocb_size         = sizeof(BlockDriverAIOCBSync),
+    .cancel             = bdrv_aio_cancel_em,
+};
+
 static void bdrv_aio_bh_cb(void *opaque)
 {
     BlockDriverAIOCBSync *acb = opaque;
+
+    if (!acb->is_write)
+        qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
+    qemu_vfree(acb->bounce);
     acb->common.cb(acb->common.opaque, acb->ret);
+
     qemu_aio_release(acb);
 }
 
-static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
-        int64_t sector_num, uint8_t *buf, int nb_sectors,
-        BlockDriverCompletionFunc *cb, void *opaque)
+static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
+                                            int64_t sector_num,
+                                            QEMUIOVector *qiov,
+                                            int nb_sectors,
+                                            BlockDriverCompletionFunc *cb,
+                                            void *opaque,
+                                            int is_write)
+
 {
     BlockDriverAIOCBSync *acb;
-    int ret;
 
-    acb = qemu_aio_get(bs, cb, opaque);
+    acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
+    acb->is_write = is_write;
+    acb->qiov = qiov;
+    acb->bounce = qemu_blockalign(bs, qiov->size);
+
     if (!acb->bh)
         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
-    ret = bdrv_read(bs, sector_num, buf, nb_sectors);
-    acb->ret = ret;
+
+    if (is_write) {
+        qemu_iovec_to_buffer(acb->qiov, acb->bounce);
+        acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
+    } else {
+        acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
+    }
+
     qemu_bh_schedule(acb->bh);
+
     return &acb->common;
 }
 
-static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
-        int64_t sector_num, const uint8_t *buf, int nb_sectors,
+static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque)
 {
-    BlockDriverAIOCBSync *acb;
-    int ret;
-
-    acb = qemu_aio_get(bs, cb, opaque);
-    if (!acb->bh)
-        acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
-    ret = bdrv_write(bs, sector_num, buf, nb_sectors);
-    acb->ret = ret;
-    qemu_bh_schedule(acb->bh);
-    return &acb->common;
+    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
 }
 
-static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
+static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
+        BlockDriverCompletionFunc *cb, void *opaque)
 {
-    BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
-    qemu_bh_cancel(acb->bh);
-    qemu_aio_release(acb);
+    return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
 }
-#endif /* !QEMU_IMG */
 
 /**************************************************************/
 /* sync block device emulation */
@@ -1236,19 +1454,22 @@ static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
 {
     int async_ret;
     BlockDriverAIOCB *acb;
+    struct iovec iov;
+    QEMUIOVector qiov;
 
     async_ret = NOT_DONE;
-    qemu_aio_wait_start();
-    acb = bdrv_aio_read(bs, sector_num, buf, nb_sectors,
-                        bdrv_rw_em_cb, &async_ret);
-    if (acb == NULL) {
-        qemu_aio_wait_end();
+    iov.iov_base = (void *)buf;
+    iov.iov_len = nb_sectors * 512;
+    qemu_iovec_init_external(&qiov, &iov, 1);
+    acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
+        bdrv_rw_em_cb, &async_ret);
+    if (acb == NULL)
         return -1;
-    }
+
     while (async_ret == NOT_DONE) {
         qemu_aio_wait();
     }
-    qemu_aio_wait_end();
+
     return async_ret;
 }
 
@@ -1257,54 +1478,39 @@ static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
 {
     int async_ret;
     BlockDriverAIOCB *acb;
+    struct iovec iov;
+    QEMUIOVector qiov;
 
     async_ret = NOT_DONE;
-    qemu_aio_wait_start();
-    acb = bdrv_aio_write(bs, sector_num, buf, nb_sectors,
-                         bdrv_rw_em_cb, &async_ret);
-    if (acb == NULL) {
-        qemu_aio_wait_end();
+    iov.iov_base = (void *)buf;
+    iov.iov_len = nb_sectors * 512;
+    qemu_iovec_init_external(&qiov, &iov, 1);
+    acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
+        bdrv_rw_em_cb, &async_ret);
+    if (acb == NULL)
         return -1;
-    }
     while (async_ret == NOT_DONE) {
         qemu_aio_wait();
     }
-    qemu_aio_wait_end();
     return async_ret;
 }
 
 void bdrv_init(void)
 {
-    bdrv_register(&bdrv_raw);
-    bdrv_register(&bdrv_host_device);
-#ifndef _WIN32
-    bdrv_register(&bdrv_cow);
-#endif
-    bdrv_register(&bdrv_qcow);
-    bdrv_register(&bdrv_vmdk);
-    bdrv_register(&bdrv_cloop);
-    bdrv_register(&bdrv_dmg);
-    bdrv_register(&bdrv_bochs);
-    bdrv_register(&bdrv_vpc);
-    bdrv_register(&bdrv_vvfat);
-    bdrv_register(&bdrv_qcow2);
-    bdrv_register(&bdrv_parallels);
+    module_call_init(MODULE_INIT_BLOCK);
 }
 
-void *qemu_aio_get(BlockDriverState *bs, BlockDriverCompletionFunc *cb,
-                   void *opaque)
+void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
+                   BlockDriverCompletionFunc *cb, void *opaque)
 {
-    BlockDriver *drv;
     BlockDriverAIOCB *acb;
 
-    drv = bs->drv;
-    if (drv->free_aiocb) {
-        acb = drv->free_aiocb;
-        drv->free_aiocb = acb->next;
+    if (pool->free_aiocb) {
+        acb = pool->free_aiocb;
+        pool->free_aiocb = acb->next;
     } else {
-        acb = qemu_mallocz(drv->aiocb_size);
-        if (!acb)
-            return NULL;
+        acb = qemu_mallocz(pool->aiocb_size);
+        acb->pool = pool;
     }
     acb->bs = bs;
     acb->cb = cb;
@@ -1314,10 +1520,10 @@ void *qemu_aio_get(BlockDriverState *bs, BlockDriverCompletionFunc *cb,
 
 void qemu_aio_release(void *p)
 {
-    BlockDriverAIOCB *acb = p;
-    BlockDriver *drv = acb->bs->drv;
-    acb->next = drv->free_aiocb;
-    drv->free_aiocb = acb;
+    BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
+    AIOPool *pool = acb->pool;
+    acb->next = pool->free_aiocb;
+    pool->free_aiocb = acb;
 }
 
 /**************************************************************/
@@ -1394,3 +1600,30 @@ void bdrv_set_locked(BlockDriverState *bs, int locked)
         drv->bdrv_set_locked(bs, locked);
     }
 }
+
+/* needed for generic scsi interface */
+
+int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
+{
+    BlockDriver *drv = bs->drv;
+
+    if (drv && drv->bdrv_ioctl)
+        return drv->bdrv_ioctl(bs, req, buf);
+    return -ENOTSUP;
+}
+
+BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
+        unsigned long int req, void *buf,
+        BlockDriverCompletionFunc *cb, void *opaque)
+{
+    BlockDriver *drv = bs->drv;
+
+    if (drv && drv->bdrv_aio_ioctl)
+        return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
+    return NULL;
+}
+
+void *qemu_blockalign(BlockDriverState *bs, size_t size)
+{
+    return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
+}