qemu/block.c - diff

Return to block.c CVS log

Up to [Qemu by Fabrice Bellard] / qemu

Diff for /qemu/block.c between versions 1.1.1.9 and 1.1.1.22

-version 1.1.1.9, 2018/04/24 17:01:52
+version 1.1.1.22, 2018/04/24 19:17:20
  Line 22
   * THE SOFTWARE.
   */
  #include "config-host.h"
- #ifdef _BSD
- /* include native header before sys-queue.h */
- #include <sys/queue.h>
- #endif
  #include "qemu-common.h"
- #include "console.h"
+ #include "trace.h"
+ #include "monitor.h"
  #include "block_int.h"
+ #include "module.h"
+ #include "qjson.h"
+ #include "qemu-coroutine.h"
+ #include "qmp-commands.h"
- #ifdef _BSD
+ #ifdef CONFIG_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
- #define SECTOR_BITS 9
+ #ifdef _WIN32
- #define SECTOR_SIZE (1 << SECTOR_BITS)
+ #include <windows.h>
+ #endif
- typedef struct BlockDriverAIOCBSync {
+ #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
-     BlockDriverAIOCB common;
-     QEMUBH *bh;
-     int ret;
- } BlockDriverAIOCBSync;
- static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
+ static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
-         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,
+ static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
-         int64_t sector_num, const uint8_t *buf, int nb_sectors,
+         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
          BlockDriverCompletionFunc *cb, void *opaque);
- static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb);
+ static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
- static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
+                                          int64_t sector_num, int nb_sectors,
-                         uint8_t *buf, int nb_sectors);
+                                          QEMUIOVector *iov);
- static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
+ static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
-                          const uint8_t *buf, int nb_sectors);
+                                          int64_t sector_num, int nb_sectors,
+                                          QEMUIOVector *iov);
+ static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
+     int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
+ static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
+     int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
+ static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
+                                                int64_t sector_num,
+                                                QEMUIOVector *qiov,
+                                                int nb_sectors,
+                                                BlockDriverCompletionFunc *cb,
+                                                void *opaque,
+                                                bool is_write);
+ static void coroutine_fn bdrv_co_do_rw(void *opaque);
+ static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
+     QTAILQ_HEAD_INITIALIZER(bdrv_states);
+ static QLIST_HEAD(, BlockDriver) bdrv_drivers =
+     QLIST_HEAD_INITIALIZER(bdrv_drivers);
+ /* The device to use for VM snapshots */
+ static BlockDriverState *bs_snapshots;
+ /* If non-zero, use only whitelisted block drivers */
+ static int use_bdrv_whitelist;
+ #ifdef _WIN32
+ static int is_windows_drive_prefix(const char *filename)
+ {
+     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
+              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
+             filename[1] == ':');
+ }
+ int is_windows_drive(const char *filename)
+ {
+     if (is_windows_drive_prefix(filename) &&
+         filename[2] == '\0')
+         return 1;
+     if (strstart(filename, "\\\\.\\", NULL) ||
+         strstart(filename, "//./", NULL))
+         return 1;
+     return 0;
+ }
+ #endif
- BlockDriverState *bdrv_first;
+ /* check if the path starts with "<protocol>:" */
+ static int path_has_protocol(const char *path)
+ {
+ #ifdef _WIN32
+     if (is_windows_drive(path) ||
+         is_windows_drive_prefix(path)) {
+         return 0;
+     }
+ #endif
- static BlockDriver *first_drv;
+     return strchr(path, ':') != NULL;
+ }
  int path_is_absolute(const char *path)
  {
- Line 127  void path_combine(char *dest, int dest_s
+ Line 182  void path_combine(char *dest, int dest_s
      }
  }
+ void bdrv_register(BlockDriver *bdrv)
- static void bdrv_register(BlockDriver *bdrv)
  {
-     if (!bdrv->bdrv_aio_read) {
+     /* Block drivers without coroutine functions need emulation */
-         /* add AIO emulation layer */
+     if (!bdrv->bdrv_co_readv) {
-         bdrv->bdrv_aio_read = bdrv_aio_read_em;
+         bdrv->bdrv_co_readv = bdrv_co_readv_em;
-         bdrv->bdrv_aio_write = bdrv_aio_write_em;
+         bdrv->bdrv_co_writev = bdrv_co_writev_em;
-         bdrv->bdrv_aio_cancel = bdrv_aio_cancel_em;
-         bdrv->aiocb_size = sizeof(BlockDriverAIOCBSync);
+         /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
-     } else if (!bdrv->bdrv_read && !bdrv->bdrv_pread) {
+          * the block driver lacks aio we need to emulate that too.
-         /* add synchronous IO emulation layer */
+          */
-         bdrv->bdrv_read = bdrv_read_em;
+         if (!bdrv->bdrv_aio_readv) {
-         bdrv->bdrv_write = bdrv_write_em;
+             /* add AIO emulation layer */
+             bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
+             bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
+         }
      }
-     bdrv->next = first_drv;
-     first_drv = bdrv;
+     QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
  }
  /* create a new block device (by default it is empty) */
  BlockDriverState *bdrv_new(const char *device_name)
  {
-     BlockDriverState **pbs, *bs;
+     BlockDriverState *bs;
-     bs = qemu_mallocz(sizeof(BlockDriverState));
+     bs = g_malloc0(sizeof(BlockDriverState));
      pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
      if (device_name[0] != '\0') {
-         /* insert at the end */
+         QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
-         pbs = &bdrv_first;
-         while (*pbs != NULL)
-             pbs = &(*pbs)->next;
-         *pbs = bs;
      }
+     bdrv_iostatus_disable(bs);
      return bs;
  }
  BlockDriver *bdrv_find_format(const char *format_name)
  {
      BlockDriver *drv1;
-     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
+     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
-         if (!strcmp(drv1->format_name, format_name))
+         if (!strcmp(drv1->format_name, format_name)) {
              return drv1;
+         }
      }
      return NULL;
  }
- int bdrv_create(BlockDriver *drv,
+ static int bdrv_is_whitelisted(BlockDriver *drv)
-                 const char *filename, int64_t size_in_sectors,
+ {
-                 const char *backing_file, int flags)
+     static const char *whitelist[] = {
+         CONFIG_BDRV_WHITELIST
+     };
+     const char **p;
+     if (!whitelist[0])
+         return 1;               /* no whitelist, anything goes */
+     for (p = whitelist; *p; p++) {
+         if (!strcmp(drv->format_name, *p)) {
+             return 1;
+         }
+     }
+     return 0;
+ }
+ BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
+ {
+     BlockDriver *drv = bdrv_find_format(format_name);
+     return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
+ }
+ 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);
+ }
+ int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
+ {
+     BlockDriver *drv;
+     drv = bdrv_find_protocol(filename);
+     if (drv == NULL) {
+         return -ENOENT;
+     }
+     return bdrv_create(drv, filename, options);
  }
  #ifdef _WIN32
- Line 204  void get_tmp_filename(char *filename, in
+ Line 295  void get_tmp_filename(char *filename, in
  }
  #endif
- #ifdef _WIN32
+ /*
- static int is_windows_drive_prefix(const char *filename)
+  * Detect host devices. By convention, /dev/cdrom[N] is always
+  * recognized as a host CDROM.
+  */
+ static BlockDriver *find_hdev_driver(const char *filename)
  {
-     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
+     int score_max = 0, score;
-              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
+     BlockDriver *drv = NULL, *d;
-             filename[1] == ':');
- }
- static int is_windows_drive(const char *filename)
+     QLIST_FOREACH(d, &bdrv_drivers, list) {
- {
+         if (d->bdrv_probe_device) {
-     if (is_windows_drive_prefix(filename) &&
+             score = d->bdrv_probe_device(filename);
-         filename[2] == '\0')
+             if (score > score_max) {
-         return 1;
+                 score_max = score;
-     if (strstart(filename, "\\\\.\\", NULL) ||
+                 drv = d;
-         strstart(filename, "//./", NULL))
+             }
-         return 1;
+         }
-     return 0;
+     }
+     return drv;
  }
- #endif
- static BlockDriver *find_protocol(const char *filename)
+ BlockDriver *bdrv_find_protocol(const char *filename)
  {
      BlockDriver *drv1;
      char protocol[128];
      int len;
      const char *p;
- #ifdef _WIN32
+     /* TODO Drivers without bdrv_file_open must be specified explicitly */
-     if (is_windows_drive(filename) ||
-         is_windows_drive_prefix(filename))
+     /*
-         return &bdrv_raw;
+      * XXX(hch): we really should not let host device detection
- #endif
+      * override an explicit protocol specification, but moving this
+      * later breaks access to device names with colons in them.
+      * Thanks to the brain-dead persistent naming schemes on udev-
+      * based Linux systems those actually are quite common.
+      */
+     drv1 = find_hdev_driver(filename);
+     if (drv1) {
+         return drv1;
+     }
+     if (!path_has_protocol(filename)) {
+         return bdrv_find_format("file");
+     }
      p = strchr(filename, ':');
-     if (!p)
+     assert(p != NULL);
-         return &bdrv_raw;
      len = p - filename;
      if (len > sizeof(protocol) - 1)
          len = sizeof(protocol) - 1;
      memcpy(protocol, filename, len);
      protocol[len] = '\0';
-     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
+     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
          if (drv1->protocol_name &&
-             !strcmp(drv1->protocol_name, protocol))
+             !strcmp(drv1->protocol_name, protocol)) {
              return drv1;
+         }
      }
      return NULL;
  }
- /* XXX: force raw format if block or character device ? It would
+ static int find_image_format(const char *filename, BlockDriver **pdrv)
-    simplify the BSD case */
- static BlockDriver *find_image_format(const char *filename)
  {
      int ret, score, score_max;
      BlockDriver *drv1, *drv;
      uint8_t buf[2048];
      BlockDriverState *bs;
-     /* detect host devices. By convention, /dev/cdrom[N] is always
+     ret = bdrv_file_open(&bs, filename, 0);
-        recognized as a host CDROM */
+     if (ret < 0) {
-     if (strstart(filename, "/dev/cdrom", NULL))
+         *pdrv = NULL;
-         return &bdrv_host_device;
+         return ret;
- #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);
+     /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
-     /* no need to test disk image formats for vvfat */
+     if (bs->sg || !bdrv_is_inserted(bs)) {
-     if (drv == &bdrv_vvfat)
+         bdrv_delete(bs);
-         return drv;
+         drv = bdrv_find_format("raw");
+         if (!drv) {
+             ret = -ENOENT;
+         }
+         *pdrv = drv;
+         return ret;
+     }
-     ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
-     if (ret < 0)
-         return NULL;
      ret = bdrv_pread(bs, 0, buf, sizeof(buf));
      bdrv_delete(bs);
      if (ret < 0) {
-         return NULL;
+         *pdrv = NULL;
+         return ret;
      }
      score_max = 0;
-     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
+     drv = NULL;
+     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
          if (drv1->bdrv_probe) {
              score = drv1->bdrv_probe(buf, ret, filename);
              if (score > score_max) {
- Line 302  static BlockDriver *find_image_format(co
+ Line 399  static BlockDriver *find_image_format(co
              }
          }
      }
-     return drv;
+     if (!drv) {
+         ret = -ENOENT;
+     }
+     *pdrv = drv;
+     return ret;
+ }
+ /**
+  * Set the current 'total_sectors' value
+  */
+ static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
+ {
+     BlockDriver *drv = bs->drv;
+     /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
+     if (bs->sg)
+         return 0;
+     /* query actual device if possible, otherwise just trust the hint */
+     if (drv->bdrv_getlength) {
+         int64_t length = drv->bdrv_getlength(bs);
+         if (length < 0) {
+             return length;
+         }
+         hint = length >> BDRV_SECTOR_BITS;
+     }
+     bs->total_sectors = hint;
+     return 0;
+ }
+ /**
+  * Set open flags for a given cache mode
+  *
+  * Return 0 on success, -1 if the cache mode was invalid.
+  */
+ int bdrv_parse_cache_flags(const char *mode, int *flags)
+ {
+     *flags &= ~BDRV_O_CACHE_MASK;
+     if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
+         *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
+     } else if (!strcmp(mode, "directsync")) {
+         *flags |= BDRV_O_NOCACHE;
+     } else if (!strcmp(mode, "writeback")) {
+         *flags |= BDRV_O_CACHE_WB;
+     } else if (!strcmp(mode, "unsafe")) {
+         *flags |= BDRV_O_CACHE_WB;
+         *flags |= BDRV_O_NO_FLUSH;
+     } else if (!strcmp(mode, "writethrough")) {
+         /* this is the default */
+     } else {
+         return -1;
+     }
+     return 0;
+ }
+ /*
+  * Common part for opening disk images and files
+  */
+ static int bdrv_open_common(BlockDriverState *bs, const char *filename,
+     int flags, BlockDriver *drv)
+ {
+     int ret, open_flags;
+     assert(drv != NULL);
+     trace_bdrv_open_common(bs, filename, flags, drv->format_name);
+     bs->file = NULL;
+     bs->total_sectors = 0;
+     bs->encrypted = 0;
+     bs->valid_key = 0;
+     bs->sg = 0;
+     bs->open_flags = flags;
+     bs->growable = 0;
+     bs->buffer_alignment = 512;
+     pstrcpy(bs->filename, sizeof(bs->filename), filename);
+     bs->backing_file[0] = '\0';
+     if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
+         return -ENOTSUP;
+     }
+     bs->drv = drv;
+     bs->opaque = g_malloc0(drv->instance_size);
+     bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
+     /*
+      * Clear flags that are internal to the block layer before opening the
+      * image.
+      */
+     open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
+     /*
+      * Snapshots should be writable.
+      */
+     if (bs->is_temporary) {
+         open_flags |= BDRV_O_RDWR;
+     }
+     bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
+     /* Open the image, either directly or using a protocol */
+     if (drv->bdrv_file_open) {
+         ret = drv->bdrv_file_open(bs, filename, open_flags);
+     } else {
+         ret = bdrv_file_open(&bs->file, filename, open_flags);
+         if (ret >= 0) {
+             ret = drv->bdrv_open(bs, open_flags);
+         }
+     }
+     if (ret < 0) {
+         goto free_and_fail;
+     }
+     ret = refresh_total_sectors(bs, bs->total_sectors);
+     if (ret < 0) {
+         goto free_and_fail;
+     }
+ #ifndef _WIN32
+     if (bs->is_temporary) {
+         unlink(filename);
+     }
+ #endif
+     return 0;
+ free_and_fail:
+     if (bs->file) {
+         bdrv_delete(bs->file);
+         bs->file = NULL;
+     }
+     g_free(bs->opaque);
+     bs->opaque = NULL;
+     bs->drv = NULL;
+     return ret;
  }
+ /*
+  * Opens a file using a protocol (file, host_device, nbd, ...)
+  */
  int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
  {
      BlockDriverState *bs;
+     BlockDriver *drv;
      int ret;
+     drv = bdrv_find_protocol(filename);
+     if (!drv) {
+         return -ENOENT;
+     }
      bs = bdrv_new("");
-     ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
+     ret = bdrv_open_common(bs, filename, flags, drv);
      if (ret < 0) {
          bdrv_delete(bs);
          return ret;
- Line 321  int bdrv_file_open(BlockDriverState **pb
+ Line 567  int bdrv_file_open(BlockDriverState **pb
      return 0;
  }
- int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
+ /*
- {
+  * Opens a disk image (raw, qcow2, vmdk, ...)
-     return bdrv_open2(bs, filename, flags, NULL);
+  */
- }
+ int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
+               BlockDriver *drv)
- int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
-                BlockDriver *drv)
  {
-     int ret, open_flags;
+     int ret;
      char tmp_filename[PATH_MAX];
-     char backing_filename[PATH_MAX];
-     bs->read_only = 0;
-     bs->is_temporary = 0;
-     bs->encrypted = 0;
-     bs->valid_key = 0;
      if (flags & BDRV_O_SNAPSHOT) {
          BlockDriverState *bs1;
          int64_t total_size;
          int is_protocol = 0;
+         BlockDriver *bdrv_qcow2;
+         QEMUOptionParameter *options;
+         char backing_filename[PATH_MAX];
          /* 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("");
-         ret = bdrv_open(bs1, filename, 0);
+         ret = bdrv_open(bs1, filename, 0, drv);
          if (ret < 0) {
              bdrv_delete(bs1);
              return ret;
          }
-         total_size = bdrv_getlength(bs1) >> SECTOR_BITS;
+         total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
          if (bs1->drv && bs1->drv->protocol_name)
              is_protocol = 1;
- Line 366  int bdrv_open2(BlockDriverState *bs, con
+ Line 607  int bdrv_open2(BlockDriverState *bs, con
          if (is_protocol)
              snprintf(backing_filename, sizeof(backing_filename),
                       "%s", filename);
-         else
+         else if (!realpath(filename, backing_filename))
-             realpath(filename, backing_filename);
+             return -errno;
+         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);
+         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,
+         ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
-                           total_size, backing_filename, 0);
+         free_option_parameters(options);
          if (ret < 0) {
              return ret;
          }
          filename = tmp_filename;
+         drv = bdrv_qcow2;
          bs->is_temporary = 1;
      }
-     pstrcpy(bs->filename, sizeof(bs->filename), filename);
+     /* Find the right image format driver */
-     if (flags & BDRV_O_FILE) {
+     if (!drv) {
-         drv = find_protocol(filename);
+         ret = find_image_format(filename, &drv);
-     } else if (!drv) {
-         drv = find_image_format(filename);
      }
      if (!drv) {
-         ret = -ENOENT;
          goto unlink_and_fail;
      }
-     bs->drv = drv;
-     bs->opaque = qemu_mallocz(drv->instance_size);
+     /* Open the image */
-     /* Note: for compatibility, we open disk image files as RDWR, and
+     ret = bdrv_open_common(bs, filename, flags, drv);
-        RDONLY as fallback */
-     if (!(flags & BDRV_O_FILE))
-         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 || 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);
+         goto unlink_and_fail;
-         bs->opaque = NULL;
-         bs->drv = NULL;
-     unlink_and_fail:
-         if (bs->is_temporary)
-             unlink(filename);
-         return ret;
-     }
-     if (drv->bdrv_getlength) {
-         bs->total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
-     }
- #ifndef _WIN32
-     if (bs->is_temporary) {
-         unlink(filename);
      }
- #endif
-     if (bs->backing_file[0] != '\0') {
+     /* If there is a backing file, use it */
-         /* if there is a backing file, use it */
+     if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
+         char backing_filename[PATH_MAX];
+         int back_flags;
+         BlockDriver *back_drv = NULL;
          bs->backing_hd = bdrv_new("");
-         path_combine(backing_filename, sizeof(backing_filename),
-                      filename, bs->backing_file);
+         if (path_has_protocol(bs->backing_file)) {
-         ret = bdrv_open(bs->backing_hd, backing_filename, open_flags);
+             pstrcpy(backing_filename, sizeof(backing_filename),
+                     bs->backing_file);
+         } else {
+             path_combine(backing_filename, sizeof(backing_filename),
+                          filename, bs->backing_file);
+         }
+         if (bs->backing_format[0] != '\0') {
+             back_drv = bdrv_find_format(bs->backing_format);
+         }
+         /* backing files always opened read-only */
+         back_flags =
+             flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
+         ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
          if (ret < 0) {
              bdrv_close(bs);
              return ret;
          }
+         if (bs->is_temporary) {
+             bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
+         } else {
+             /* base image inherits from "parent" */
+             bs->backing_hd->keep_read_only = bs->keep_read_only;
+         }
      }
-     /* call the change callback */
+     if (!bdrv_key_required(bs)) {
-     bs->media_changed = 1;
+         bdrv_dev_change_media_cb(bs, true);
-     if (bs->change_cb)
+     }
-         bs->change_cb(bs->change_opaque);
      return 0;
+ unlink_and_fail:
+     if (bs->is_temporary) {
+         unlink(filename);
+     }
+     return ret;
  }
  void bdrv_close(BlockDriverState *bs)
  {
      if (bs->drv) {
-         if (bs->backing_hd)
+         if (bs == bs_snapshots) {
+             bs_snapshots = NULL;
+         }
+         if (bs->backing_hd) {
              bdrv_delete(bs->backing_hd);
+             bs->backing_hd = NULL;
+         }
          bs->drv->bdrv_close(bs);
-         qemu_free(bs->opaque);
+         g_free(bs->opaque);
  #ifdef _WIN32
          if (bs->is_temporary) {
              unlink(bs->filename);
- Line 453  void bdrv_close(BlockDriverState *bs)
+ Line 716  void bdrv_close(BlockDriverState *bs)
          bs->opaque = NULL;
          bs->drv = NULL;
-         /* call the change callback */
+         if (bs->file != NULL) {
-         bs->media_changed = 1;
+             bdrv_close(bs->file);
-         if (bs->change_cb)
+         }
-             bs->change_cb(bs->change_opaque);
+         bdrv_dev_change_media_cb(bs, false);
      }
  }
- void bdrv_delete(BlockDriverState *bs)
+ void bdrv_close_all(void)
  {
-     BlockDriverState **pbs;
+     BlockDriverState *bs;
-     pbs = &bdrv_first;
+     QTAILQ_FOREACH(bs, &bdrv_states, list) {
-     while (*pbs != bs && *pbs != NULL)
+         bdrv_close(bs);
-         pbs = &(*pbs)->next;
+     }
-     if (*pbs == bs)
+ }
-         *pbs = bs->next;
-     bdrv_close(bs);
+ /* make a BlockDriverState anonymous by removing from bdrv_state list.
-     qemu_free(bs);
+    Also, NULL terminate the device_name to prevent double remove */
+ void bdrv_make_anon(BlockDriverState *bs)
+ {
+     if (bs->device_name[0] != '\0') {
+         QTAILQ_REMOVE(&bdrv_states, bs, list);
+     }
+     bs->device_name[0] = '\0';
  }
- /* commit COW file into the raw image */
+ void bdrv_delete(BlockDriverState *bs)
- int bdrv_commit(BlockDriverState *bs)
  {
-     BlockDriver *drv = bs->drv;
+     assert(!bs->dev);
-     int64_t i, total_sectors;
-     int n, j;
-     unsigned char sector[512];
-     if (!drv)
+     /* remove from list, if necessary */
-         return -ENOMEDIUM;
+     bdrv_make_anon(bs);
-     if (bs->read_only) {
+     bdrv_close(bs);
-         return -EACCES;
+     if (bs->file != NULL) {
+         bdrv_delete(bs->file);
+     }
+     assert(bs != bs_snapshots);
+     g_free(bs);
+ }
+ int bdrv_attach_dev(BlockDriverState *bs, void *dev)
+ /* TODO change to DeviceState *dev when all users are qdevified */
+ {
+     if (bs->dev) {
+         return -EBUSY;
+     }
+     bs->dev = dev;
+     bdrv_iostatus_reset(bs);
+     return 0;
+ }
+ /* TODO qdevified devices don't use this, remove when devices are qdevified */
+ void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
+ {
+     if (bdrv_attach_dev(bs, dev) < 0) {
+         abort();
+     }
+ }
+ void bdrv_detach_dev(BlockDriverState *bs, void *dev)
+ /* TODO change to DeviceState *dev when all users are qdevified */
+ {
+     assert(bs->dev == dev);
+     bs->dev = NULL;
+     bs->dev_ops = NULL;
+     bs->dev_opaque = NULL;
+     bs->buffer_alignment = 512;
+ }
+ /* TODO change to return DeviceState * when all users are qdevified */
+ void *bdrv_get_attached_dev(BlockDriverState *bs)
+ {
+     return bs->dev;
+ }
+ void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
+                       void *opaque)
+ {
+     bs->dev_ops = ops;
+     bs->dev_opaque = opaque;
+     if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
+         bs_snapshots = NULL;
+     }
+ }
+ static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
+ {
+     if (bs->dev_ops && bs->dev_ops->change_media_cb) {
+         bs->dev_ops->change_media_cb(bs->dev_opaque, load);
+     }
+ }
+ bool bdrv_dev_has_removable_media(BlockDriverState *bs)
+ {
+     return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
+ }
+ void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
+ {
+     if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
+         bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
+     }
+ }
+ bool bdrv_dev_is_tray_open(BlockDriverState *bs)
+ {
+     if (bs->dev_ops && bs->dev_ops->is_tray_open) {
+         return bs->dev_ops->is_tray_open(bs->dev_opaque);
+     }
+     return false;
+ }
+ static void bdrv_dev_resize_cb(BlockDriverState *bs)
+ {
+     if (bs->dev_ops && bs->dev_ops->resize_cb) {
+         bs->dev_ops->resize_cb(bs->dev_opaque);
+     }
+ }
+ bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
+ {
+     if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
+         return bs->dev_ops->is_medium_locked(bs->dev_opaque);
+     }
+     return false;
+ }
+ /*
+  * Run consistency checks on an image
+  *
+  * Returns 0 if the check could be completed (it doesn't mean that the image is
+  * free of errors) or -errno when an internal error occurred. The results of the
+  * check are stored in res.
+  */
+ int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res)
+ {
+     if (bs->drv->bdrv_check == NULL) {
+         return -ENOTSUP;
      }
+     memset(res, 0, sizeof(*res));
+     return bs->drv->bdrv_check(bs, res);
+ }
+ #define COMMIT_BUF_SECTORS 2048
+ /* commit COW file into the raw image */
+ int bdrv_commit(BlockDriverState *bs)
+ {
+     BlockDriver *drv = bs->drv;
+     BlockDriver *backing_drv;
+     int64_t sector, total_sectors;
+     int n, ro, open_flags;
+     int ret = 0, rw_ret = 0;
+     uint8_t *buf;
+     char filename[1024];
+     BlockDriverState *bs_rw, *bs_ro;
+     if (!drv)
+         return -ENOMEDIUM;
      if (!bs->backing_hd) {
          return -ENOTSUP;
      }
-     total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
+     if (bs->backing_hd->keep_read_only) {
-     for (i = 0; i < total_sectors;) {
+         return -EACCES;
-         if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
+     }
-             for(j = 0; j < n; j++) {
-                 if (bdrv_read(bs, i, sector, 1) != 0) {
-                     return -EIO;
-                 }
-                 if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
+     backing_drv = bs->backing_hd->drv;
-                     return -EIO;
+     ro = bs->backing_hd->read_only;
-                 }
+     strncpy(filename, bs->backing_hd->filename, sizeof(filename));
-                 i++;
+     open_flags =  bs->backing_hd->open_flags;
-             }
-         } else {
+     if (ro) {
-             i += n;
+         /* re-open as RW */
+         bdrv_delete(bs->backing_hd);
+         bs->backing_hd = NULL;
+         bs_rw = bdrv_new("");
+         rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR,
+             backing_drv);
+         if (rw_ret < 0) {
+             bdrv_delete(bs_rw);
+             /* try to re-open read-only */
+             bs_ro = bdrv_new("");
+             ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
+                 backing_drv);
+             if (ret < 0) {
+                 bdrv_delete(bs_ro);
+                 /* drive not functional anymore */
+                 bs->drv = NULL;
+                 return ret;
+             }
+             bs->backing_hd = bs_ro;
+             return rw_ret;
          }
+         bs->backing_hd = bs_rw;
      }
-     if (drv->bdrv_make_empty)
+     total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
-         return drv->bdrv_make_empty(bs);
+     buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
-     return 0;
+     for (sector = 0; sector < total_sectors; sector += n) {
+         if (drv->bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) {
+             if (bdrv_read(bs, sector, buf, n) != 0) {
+                 ret = -EIO;
+                 goto ro_cleanup;
+             }
+             if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) {
+                 ret = -EIO;
+                 goto ro_cleanup;
+             }
+         }
+     }
+     if (drv->bdrv_make_empty) {
+         ret = drv->bdrv_make_empty(bs);
+         bdrv_flush(bs);
+     }
+     /*
+      * Make sure all data we wrote to the backing device is actually
+      * stable on disk.
+      */
+     if (bs->backing_hd)
+         bdrv_flush(bs->backing_hd);
+ ro_cleanup:
+     g_free(buf);
+     if (ro) {
+         /* re-open as RO */
+         bdrv_delete(bs->backing_hd);
+         bs->backing_hd = NULL;
+         bs_ro = bdrv_new("");
+         ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR,
+             backing_drv);
+         if (ret < 0) {
+             bdrv_delete(bs_ro);
+             /* drive not functional anymore */
+             bs->drv = NULL;
+             return ret;
+         }
+         bs->backing_hd = bs_ro;
+         bs->backing_hd->keep_read_only = 0;
+     }
+     return ret;
+ }
+ void bdrv_commit_all(void)
+ {
+     BlockDriverState *bs;
+     QTAILQ_FOREACH(bs, &bdrv_states, list) {
+         bdrv_commit(bs);
+     }
+ }
+ /*
+  * Return values:
+  * 0        - success
+  * -EINVAL  - backing format specified, but no file
+  * -ENOSPC  - can't update the backing file because no space is left in the
+  *            image file header
+  * -ENOTSUP - format driver doesn't support changing the backing file
+  */
+ int bdrv_change_backing_file(BlockDriverState *bs,
+     const char *backing_file, const char *backing_fmt)
+ {
+     BlockDriver *drv = bs->drv;
+     if (drv->bdrv_change_backing_file != NULL) {
+         return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
+     } else {
+         return -ENOTSUP;
+     }
  }
  static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
- Line 530  static int bdrv_check_byte_request(Block
+ Line 1013  static int bdrv_check_byte_request(Block
      len = bdrv_getlength(bs);
-     if ((offset + size) > len)
+     if (offset < 0)
+         return -EIO;
+     if ((offset > len) || (len - offset < size))
          return -EIO;
      return 0;
- Line 539  static int bdrv_check_byte_request(Block
+ Line 1025  static int bdrv_check_byte_request(Block
  static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
                                int nb_sectors)
  {
-     int64_t offset;
+     return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
+                                    nb_sectors * BDRV_SECTOR_SIZE);
+ }
-     /* Deal with byte accesses */
+ typedef struct RwCo {
-     if (sector_num < 0)
+     BlockDriverState *bs;
-         offset = -sector_num;
+     int64_t sector_num;
-     else
+     int nb_sectors;
-         offset = sector_num * 512;
+     QEMUIOVector *qiov;
+     bool is_write;
+     int ret;
+ } RwCo;
+ static void coroutine_fn bdrv_rw_co_entry(void *opaque)
+ {
+     RwCo *rwco = opaque;
+     if (!rwco->is_write) {
+         rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num,
+                                      rwco->nb_sectors, rwco->qiov);
+     } else {
+         rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
+                                       rwco->nb_sectors, rwco->qiov);
+     }
+ }
-     return bdrv_check_byte_request(bs, offset, nb_sectors * 512);
+ /*
+  * Process a synchronous request using coroutines
+  */
+ static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
+                       int nb_sectors, bool is_write)
+ {
+     QEMUIOVector qiov;
+     struct iovec iov = {
+         .iov_base = (void *)buf,
+         .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
+     };
+     Coroutine *co;
+     RwCo rwco = {
+         .bs = bs,
+         .sector_num = sector_num,
+         .nb_sectors = nb_sectors,
+         .qiov = &qiov,
+         .is_write = is_write,
+         .ret = NOT_DONE,
+     };
+     qemu_iovec_init_external(&qiov, &iov, 1);
+     if (qemu_in_coroutine()) {
+         /* Fast-path if already in coroutine context */
+         bdrv_rw_co_entry(&rwco);
+     } else {
+         co = qemu_coroutine_create(bdrv_rw_co_entry);
+         qemu_coroutine_enter(co, &rwco);
+         while (rwco.ret == NOT_DONE) {
+             qemu_aio_wait();
+         }
+     }
+     return rwco.ret;
  }
  /* 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)
  {
-     BlockDriver *drv = bs->drv;
+     return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false);
+ }
-     if (!drv)
+ static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
-         return -ENOMEDIUM;
+                              int nb_sectors, int dirty)
-     if (bdrv_check_request(bs, sector_num, nb_sectors))
+ {
-         return -EIO;
+     int64_t start, end;
+     unsigned long val, idx, bit;
-     if (drv->bdrv_pread) {
+     start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
-         int ret, len;
+     end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
-         len = nb_sectors * 512;
-         ret = drv->bdrv_pread(bs, sector_num * 512, buf, len);
+     for (; start <= end; start++) {
-         if (ret < 0)
+         idx = start / (sizeof(unsigned long) * 8);
-             return ret;
+         bit = start % (sizeof(unsigned long) * 8);
-         else if (ret != len)
+         val = bs->dirty_bitmap[idx];
-             return -EINVAL;
+         if (dirty) {
-         else {
+             if (!(val & (1UL << bit))) {
-             bs->rd_bytes += (unsigned) len;
+                 bs->dirty_count++;
-             bs->rd_ops ++;
+                 val |= 1UL << bit;
-             return 0;
+             }
-         }
+         } else {
-     } else {
+             if (val & (1UL << bit)) {
-         return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
+                 bs->dirty_count--;
+                 val &= ~(1UL << bit);
+             }
+         }
+         bs->dirty_bitmap[idx] = val;
      }
  }
- Line 588  int bdrv_read(BlockDriverState *bs, int6
+ Line 1131  int bdrv_read(BlockDriverState *bs, int6
  int bdrv_write(BlockDriverState *bs, int64_t sector_num,
                 const uint8_t *buf, int nb_sectors)
  {
-     BlockDriver *drv = bs->drv;
+     return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true);
-     if (!bs->drv)
-         return -ENOMEDIUM;
-     if (bs->read_only)
-         return -EACCES;
-     if (bdrv_check_request(bs, sector_num, nb_sectors))
-         return -EIO;
-     if (drv->bdrv_pwrite) {
-         int ret, len, count = 0;
-         len = nb_sectors * 512;
-         do {
-             ret = drv->bdrv_pwrite(bs, sector_num * 512, buf, len - count);
-             if (ret < 0) {
-                 printf("bdrv_write ret=%d\n", ret);
-                 return ret;
-             }
-             count += ret;
-             buf += ret;
-         } while (count != len);
-         bs->wr_bytes += (unsigned) len;
-         bs->wr_ops ++;
-         return 0;
-     }
-     return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
  }
- static int bdrv_pread_em(BlockDriverState *bs, int64_t offset,
+ int bdrv_pread(BlockDriverState *bs, int64_t offset,
-                          uint8_t *buf, int count1)
+                void *buf, int count1)
  {
-     uint8_t tmp_buf[SECTOR_SIZE];
+     uint8_t tmp_buf[BDRV_SECTOR_SIZE];
      int len, nb_sectors, count;
      int64_t sector_num;
+     int ret;
      count = count1;
      /* first read to align to sector start */
-     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
+     len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
      if (len > count)
          len = count;
-     sector_num = offset >> SECTOR_BITS;
+     sector_num = offset >> BDRV_SECTOR_BITS;
      if (len > 0) {
-         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
+         if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
-             return -EIO;
+             return ret;
-         memcpy(buf, tmp_buf + (offset & (SECTOR_SIZE - 1)), len);
+         memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
          count -= len;
          if (count == 0)
              return count1;
- Line 640  static int bdrv_pread_em(BlockDriverStat
+ Line 1160  static int bdrv_pread_em(BlockDriverStat
      }
      /* read the sectors "in place" */
-     nb_sectors = count >> SECTOR_BITS;
+     nb_sectors = count >> BDRV_SECTOR_BITS;
      if (nb_sectors > 0) {
-         if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
+         if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
-             return -EIO;
+             return ret;
          sector_num += nb_sectors;
-         len = nb_sectors << SECTOR_BITS;
+         len = nb_sectors << BDRV_SECTOR_BITS;
          buf += len;
          count -= len;
      }
      /* add data from the last sector */
      if (count > 0) {
-         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
+         if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
-             return -EIO;
+             return ret;
          memcpy(buf, tmp_buf, count);
      }
      return count1;
  }
- static int bdrv_pwrite_em(BlockDriverState *bs, int64_t offset,
+ int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
-                           const uint8_t *buf, int count1)
+                 const void *buf, int count1)
  {
-     uint8_t tmp_buf[SECTOR_SIZE];
+     uint8_t tmp_buf[BDRV_SECTOR_SIZE];
      int len, nb_sectors, count;
      int64_t sector_num;
+     int ret;
      count = count1;
      /* first write to align to sector start */
-     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
+     len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
      if (len > count)
          len = count;
-     sector_num = offset >> SECTOR_BITS;
+     sector_num = offset >> BDRV_SECTOR_BITS;
      if (len > 0) {
-         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
+         if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
-             return -EIO;
+             return ret;
-         memcpy(tmp_buf + (offset & (SECTOR_SIZE - 1)), buf, len);
+         memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
-         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
+         if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
-             return -EIO;
+             return ret;
          count -= len;
          if (count == 0)
              return count1;
- Line 686  static int bdrv_pwrite_em(BlockDriverSta
+ Line 1207  static int bdrv_pwrite_em(BlockDriverSta
      }
      /* write the sectors "in place" */
-     nb_sectors = count >> SECTOR_BITS;
+     nb_sectors = count >> BDRV_SECTOR_BITS;
      if (nb_sectors > 0) {
-         if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
+         if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
-             return -EIO;
+             return ret;
          sector_num += nb_sectors;
-         len = nb_sectors << SECTOR_BITS;
+         len = nb_sectors << BDRV_SECTOR_BITS;
          buf += len;
          count -= len;
      }
      /* add data from the last sector */
      if (count > 0) {
-         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
+         if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
-             return -EIO;
+             return ret;
          memcpy(tmp_buf, buf, count);
-         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
+         if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
-             return -EIO;
+             return ret;
      }
      return count1;
  }
- /**
+ /*
-  * Read with byte offsets (needed only for file protocols)
+  * Writes to the file and ensures that no writes are reordered across this
+  * request (acts as a barrier)
+  *
+  * Returns 0 on success, -errno in error cases.
   */
- int bdrv_pread(BlockDriverState *bs, int64_t offset,
+ int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
-                void *buf1, int count1)
+     const void *buf, int count)
+ {
+     int ret;
+     ret = bdrv_pwrite(bs, offset, buf, count);
+     if (ret < 0) {
+         return ret;
+     }
+     /* No flush needed for cache modes that use O_DSYNC */
+     if ((bs->open_flags & BDRV_O_CACHE_WB) != 0) {
+         bdrv_flush(bs);
+     }
+     return 0;
+ }
+ /*
+  * Handle a read request in coroutine context
+  */
+ static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
+     int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
  {
      BlockDriver *drv = bs->drv;
-     if (!drv)
+     if (!drv) {
          return -ENOMEDIUM;
-     if (bdrv_check_byte_request(bs, offset, count1))
+     }
+     if (bdrv_check_request(bs, sector_num, nb_sectors)) {
          return -EIO;
+     }
-     if (!drv->bdrv_pread)
+     return drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
-         return bdrv_pread_em(bs, offset, buf1, count1);
-     return drv->bdrv_pread(bs, offset, buf1, count1);
  }
- /**
+ int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
-  * Write with byte offsets (needed only for file protocols)
+     int nb_sectors, QEMUIOVector *qiov)
+ {
+     trace_bdrv_co_readv(bs, sector_num, nb_sectors);
+     return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov);
+ }
+ /*
+  * Handle a write request in coroutine context
   */
- int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
+ static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
-                 const void *buf1, int count1)
+     int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
  {
      BlockDriver *drv = bs->drv;
+     int ret;
-     if (!drv)
+     if (!bs->drv) {
          return -ENOMEDIUM;
-     if (bdrv_check_byte_request(bs, offset, count1))
+     }
+     if (bs->read_only) {
+         return -EACCES;
+     }
+     if (bdrv_check_request(bs, sector_num, nb_sectors)) {
          return -EIO;
+     }
+     ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
+     if (bs->dirty_bitmap) {
+         set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
+     }
+     if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
+         bs->wr_highest_sector = sector_num + nb_sectors - 1;
+     }
+     return ret;
+ }
+ int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
+     int nb_sectors, QEMUIOVector *qiov)
+ {
+     trace_bdrv_co_writev(bs, sector_num, nb_sectors);
-     if (!drv->bdrv_pwrite)
+     return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov);
-         return bdrv_pwrite_em(bs, offset, buf1, count1);
-     return drv->bdrv_pwrite(bs, offset, buf1, count1);
  }
  /**
- Line 749  int bdrv_pwrite(BlockDriverState *bs, in
+ Line 1324  int bdrv_pwrite(BlockDriverState *bs, in
  int bdrv_truncate(BlockDriverState *bs, int64_t offset)
  {
      BlockDriver *drv = bs->drv;
+     int ret;
      if (!drv)
          return -ENOMEDIUM;
      if (!drv->bdrv_truncate)
          return -ENOTSUP;
-     return drv->bdrv_truncate(bs, offset);
+     if (bs->read_only)
+         return -EACCES;
+     if (bdrv_in_use(bs))
+         return -EBUSY;
+     ret = drv->bdrv_truncate(bs, offset);
+     if (ret == 0) {
+         ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
+         bdrv_dev_resize_cb(bs);
+     }
+     return ret;
+ }
+ /**
+  * Length of a allocated file in bytes. Sparse files are counted by actual
+  * allocated space. Return < 0 if error or unknown.
+  */
+ int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
+ {
+     BlockDriver *drv = bs->drv;
+     if (!drv) {
+         return -ENOMEDIUM;
+     }
+     if (drv->bdrv_get_allocated_file_size) {
+         return drv->bdrv_get_allocated_file_size(bs);
+     }
+     if (bs->file) {
+         return bdrv_get_allocated_file_size(bs->file);
+     }
+     return -ENOTSUP;
  }
  /**
- Line 764  int64_t bdrv_getlength(BlockDriverState
+ Line 1368  int64_t bdrv_getlength(BlockDriverState
      BlockDriver *drv = bs->drv;
      if (!drv)
          return -ENOMEDIUM;
-     if (!drv->bdrv_getlength) {
-         /* legacy mode */
+     if (bs->growable || bdrv_dev_has_removable_media(bs)) {
-         return bs->total_sectors * SECTOR_SIZE;
+         if (drv->bdrv_getlength) {
+             return drv->bdrv_getlength(bs);
+         }
      }
-     return drv->bdrv_getlength(bs);
+     return bs->total_sectors * BDRV_SECTOR_SIZE;
  }
  /* return 0 as number of sectors if no device present or error */
- Line 779  void bdrv_get_geometry(BlockDriverState
+ Line 1385  void bdrv_get_geometry(BlockDriverState
      if (length < 0)
          length = 0;
      else
-         length = length >> SECTOR_BITS;
+         length = length >> BDRV_SECTOR_BITS;
      *nb_sectors_ptr = length;
  }
- Line 794  struct partition {
+ Line 1400  struct partition {
          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));
+ } QEMU_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];
+     uint8_t buf[BDRV_SECTOR_SIZE];
      int ret, i, heads, sectors, cylinders;
      struct partition *p;
      uint32_t nr_sects;
- Line 907  void bdrv_set_geometry_hint(BlockDriverS
+ Line 1513  void bdrv_set_geometry_hint(BlockDriverS
      bs->secs = secs;
  }
- void bdrv_set_type_hint(BlockDriverState *bs, int type)
- {
-     bs->type = type;
-     bs->removable = ((type == BDRV_TYPE_CDROM ||
-                       type == BDRV_TYPE_FLOPPY));
- }
  void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
  {
      bs->translation = translation;
- Line 927  void bdrv_get_geometry_hint(BlockDriverS
+ Line 1526  void bdrv_get_geometry_hint(BlockDriverS
      *psecs = bs->secs;
  }
- int bdrv_get_type_hint(BlockDriverState *bs)
+ /* Recognize floppy formats */
- {
+ typedef struct FDFormat {
-     return bs->type;
+     FDriveType drive;
+     uint8_t last_sect;
+     uint8_t max_track;
+     uint8_t max_head;
+ } FDFormat;
+ static const FDFormat fd_formats[] = {
+     /* First entry is default format */
+     /* 1.44 MB 3"1/2 floppy disks */
+     { FDRIVE_DRV_144, 18, 80, 1, },
+     { FDRIVE_DRV_144, 20, 80, 1, },
+     { FDRIVE_DRV_144, 21, 80, 1, },
+     { FDRIVE_DRV_144, 21, 82, 1, },
+     { FDRIVE_DRV_144, 21, 83, 1, },
+     { FDRIVE_DRV_144, 22, 80, 1, },
+     { FDRIVE_DRV_144, 23, 80, 1, },
+     { FDRIVE_DRV_144, 24, 80, 1, },
+     /* 2.88 MB 3"1/2 floppy disks */
+     { FDRIVE_DRV_288, 36, 80, 1, },
+     { FDRIVE_DRV_288, 39, 80, 1, },
+     { FDRIVE_DRV_288, 40, 80, 1, },
+     { FDRIVE_DRV_288, 44, 80, 1, },
+     { FDRIVE_DRV_288, 48, 80, 1, },
+     /* 720 kB 3"1/2 floppy disks */
+     { FDRIVE_DRV_144,  9, 80, 1, },
+     { FDRIVE_DRV_144, 10, 80, 1, },
+     { FDRIVE_DRV_144, 10, 82, 1, },
+     { FDRIVE_DRV_144, 10, 83, 1, },
+     { FDRIVE_DRV_144, 13, 80, 1, },
+     { FDRIVE_DRV_144, 14, 80, 1, },
+     /* 1.2 MB 5"1/4 floppy disks */
+     { FDRIVE_DRV_120, 15, 80, 1, },
+     { FDRIVE_DRV_120, 18, 80, 1, },
+     { FDRIVE_DRV_120, 18, 82, 1, },
+     { FDRIVE_DRV_120, 18, 83, 1, },
+     { FDRIVE_DRV_120, 20, 80, 1, },
+     /* 720 kB 5"1/4 floppy disks */
+     { FDRIVE_DRV_120,  9, 80, 1, },
+     { FDRIVE_DRV_120, 11, 80, 1, },
+     /* 360 kB 5"1/4 floppy disks */
+     { FDRIVE_DRV_120,  9, 40, 1, },
+     { FDRIVE_DRV_120,  9, 40, 0, },
+     { FDRIVE_DRV_120, 10, 41, 1, },
+     { FDRIVE_DRV_120, 10, 42, 1, },
+     /* 320 kB 5"1/4 floppy disks */
+     { FDRIVE_DRV_120,  8, 40, 1, },
+     { FDRIVE_DRV_120,  8, 40, 0, },
+     /* 360 kB must match 5"1/4 better than 3"1/2... */
+     { FDRIVE_DRV_144,  9, 80, 0, },
+     /* end */
+     { FDRIVE_DRV_NONE, -1, -1, 0, },
+ };
+ void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
+                                    int *max_track, int *last_sect,
+                                    FDriveType drive_in, FDriveType *drive)
+ {
+     const FDFormat *parse;
+     uint64_t nb_sectors, size;
+     int i, first_match, match;
+     bdrv_get_geometry_hint(bs, nb_heads, max_track, last_sect);
+     if (*nb_heads != 0 && *max_track != 0 && *last_sect != 0) {
+         /* User defined disk */
+     } else {
+         bdrv_get_geometry(bs, &nb_sectors);
+         match = -1;
+         first_match = -1;
+         for (i = 0; ; i++) {
+             parse = &fd_formats[i];
+             if (parse->drive == FDRIVE_DRV_NONE) {
+                 break;
+             }
+             if (drive_in == parse->drive ||
+                 drive_in == FDRIVE_DRV_NONE) {
+                 size = (parse->max_head + 1) * parse->max_track *
+                     parse->last_sect;
+                 if (nb_sectors == size) {
+                     match = i;
+                     break;
+                 }
+                 if (first_match == -1) {
+                     first_match = i;
+                 }
+             }
+         }
+         if (match == -1) {
+             if (first_match == -1) {
+                 match = 1;
+             } else {
+                 match = first_match;
+             }
+             parse = &fd_formats[match];
+         }
+         *nb_heads = parse->max_head + 1;
+         *max_track = parse->max_track;
+         *last_sect = parse->last_sect;
+         *drive = parse->drive;
+     }
  }
  int bdrv_get_translation_hint(BlockDriverState *bs)
- Line 937  int bdrv_get_translation_hint(BlockDrive
+ Line 1634  int bdrv_get_translation_hint(BlockDrive
      return bs->translation;
  }
- int bdrv_is_removable(BlockDriverState *bs)
+ void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
+                        BlockErrorAction on_write_error)
  {
-     return bs->removable;
+     bs->on_read_error = on_read_error;
+     bs->on_write_error = on_write_error;
  }
- int bdrv_is_read_only(BlockDriverState *bs)
+ BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
  {
-     return bs->read_only;
+     return is_read ? bs->on_read_error : bs->on_write_error;
+ }
+ int bdrv_is_read_only(BlockDriverState *bs)
+ {
+     return bs->read_only;
  }
  int bdrv_is_sg(BlockDriverState *bs)
- Line 952  int bdrv_is_sg(BlockDriverState *bs)
+ Line 1656  int bdrv_is_sg(BlockDriverState *bs)
      return bs->sg;
  }
- /* XXX: no longer used */
+ int bdrv_enable_write_cache(BlockDriverState *bs)
- void bdrv_set_change_cb(BlockDriverState *bs,
-                         void (*change_cb)(void *opaque), void *opaque)
  {
-     bs->change_cb = change_cb;
+     return bs->enable_write_cache;
-     bs->change_opaque = opaque;
  }
  int bdrv_is_encrypted(BlockDriverState *bs)
- Line 986  int bdrv_set_key(BlockDriverState *bs, c
+ Line 1687  int bdrv_set_key(BlockDriverState *bs, c
          if (!bs->encrypted)
              return 0;
      }
-     if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
+     if (!bs->encrypted) {
-         return -1;
+         return -EINVAL;
+     } else if (!bs->drv || !bs->drv->bdrv_set_key) {
+         return -ENOMEDIUM;
+     }
      ret = bs->drv->bdrv_set_key(bs, key);
-     bs->valid_key = (ret == 0);
+     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 */
+         bdrv_dev_change_media_cb(bs, true);
+     }
      return ret;
  }
- Line 1007  void bdrv_iterate_format(void (*it)(void
+ Line 1717  void bdrv_iterate_format(void (*it)(void
  {
      BlockDriver *drv;
-     for (drv = first_drv; drv != NULL; drv = drv->next) {
+     QLIST_FOREACH(drv, &bdrv_drivers, list) {
          it(opaque, drv->format_name);
      }
  }
- Line 1016  BlockDriverState *bdrv_find(const char *
+ Line 1726  BlockDriverState *bdrv_find(const char *
  {
      BlockDriverState *bs;
-     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
+     QTAILQ_FOREACH(bs, &bdrv_states, list) {
-         if (!strcmp(name, bs->device_name))
+         if (!strcmp(name, bs->device_name)) {
              return bs;
+         }
      }
      return NULL;
  }
+ BlockDriverState *bdrv_next(BlockDriverState *bs)
+ {
+     if (!bs) {
+         return QTAILQ_FIRST(&bdrv_states);
+     }
+     return QTAILQ_NEXT(bs, list);
+ }
  void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
  {
      BlockDriverState *bs;
-     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
+     QTAILQ_FOREACH(bs, &bdrv_states, list) {
          it(opaque, bs);
      }
  }
- Line 1037  const char *bdrv_get_device_name(BlockDr
+ Line 1756  const char *bdrv_get_device_name(BlockDr
      return bs->device_name;
  }
- void bdrv_flush(BlockDriverState *bs)
- {
-     if (bs->drv->bdrv_flush)
-         bs->drv->bdrv_flush(bs);
-     if (bs->backing_hd)
-         bdrv_flush(bs->backing_hd);
- }
  void bdrv_flush_all(void)
  {
      BlockDriverState *bs;
-     for (bs = bdrv_first; bs != NULL; bs = bs->next)
+     QTAILQ_FOREACH(bs, &bdrv_states, list) {
-         if (bs->drv && !bdrv_is_read_only(bs) &&
+         if (!bdrv_is_read_only(bs) && bdrv_is_inserted(bs)) {
-             (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
              bdrv_flush(bs);
+         }
+     }
+ }
+ int bdrv_has_zero_init(BlockDriverState *bs)
+ {
+     assert(bs->drv);
+     if (bs->drv->bdrv_has_zero_init) {
+         return bs->drv->bdrv_has_zero_init(bs);
+     }
+     return 1;
  }
  /*
- Line 1082  int bdrv_is_allocated(BlockDriverState *
+ Line 1805  int bdrv_is_allocated(BlockDriverState *
      return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
  }
- void bdrv_info(void)
+ void bdrv_mon_event(const BlockDriverState *bdrv,
+                     BlockMonEventAction action, int is_read)
+ {
+     QObject *data;
+     const char *action_str;
+     switch (action) {
+     case BDRV_ACTION_REPORT:
+         action_str = "report";
+         break;
+     case BDRV_ACTION_IGNORE:
+         action_str = "ignore";
+         break;
+     case BDRV_ACTION_STOP:
+         action_str = "stop";
+         break;
+     default:
+         abort();
+     }
+     data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
+                               bdrv->device_name,
+                               action_str,
+                               is_read ? "read" : "write");
+     monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
+     qobject_decref(data);
+ }
+ BlockInfoList *qmp_query_block(Error **errp)
  {
+     BlockInfoList *head = NULL, *cur_item = NULL;
      BlockDriverState *bs;
-     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
+     QTAILQ_FOREACH(bs, &bdrv_states, list) {
-         term_printf("%s:", bs->device_name);
+         BlockInfoList *info = g_malloc0(sizeof(*info));
-         term_printf(" type=");
-         switch(bs->type) {
+         info->value = g_malloc0(sizeof(*info->value));
-         case BDRV_TYPE_HD:
+         info->value->device = g_strdup(bs->device_name);
-             term_printf("hd");
+         info->value->type = g_strdup("unknown");
-             break;
+         info->value->locked = bdrv_dev_is_medium_locked(bs);
-         case BDRV_TYPE_CDROM:
+         info->value->removable = bdrv_dev_has_removable_media(bs);
-             term_printf("cdrom");
-             break;
+         if (bdrv_dev_has_removable_media(bs)) {
-         case BDRV_TYPE_FLOPPY:
+             info->value->has_tray_open = true;
-             term_printf("floppy");
+             info->value->tray_open = bdrv_dev_is_tray_open(bs);
-             break;
+         }
-         }
-         term_printf(" removable=%d", bs->removable);
+         if (bdrv_iostatus_is_enabled(bs)) {
-         if (bs->removable) {
+             info->value->has_io_status = true;
-             term_printf(" locked=%d", bs->locked);
+             info->value->io_status = bs->iostatus;
          }
          if (bs->drv) {
-             term_printf(" file=");
+             info->value->has_inserted = true;
-             term_print_filename(bs->filename);
+             info->value->inserted = g_malloc0(sizeof(*info->value->inserted));
-             if (bs->backing_file[0] != '\0') {
+             info->value->inserted->file = g_strdup(bs->filename);
-                 term_printf(" backing_file=");
+             info->value->inserted->ro = bs->read_only;
-                 term_print_filename(bs->backing_file);
+             info->value->inserted->drv = g_strdup(bs->drv->format_name);
-             }
+             info->value->inserted->encrypted = bs->encrypted;
-             term_printf(" ro=%d", bs->read_only);
+             if (bs->backing_file[0]) {
-             term_printf(" drv=%s", bs->drv->format_name);
+                 info->value->inserted->has_backing_file = true;
-             term_printf(" encrypted=%d", bdrv_is_encrypted(bs));
+                 info->value->inserted->backing_file = g_strdup(bs->backing_file);
+             }
+         }
+         /* XXX: waiting for the qapi to support GSList */
+         if (!cur_item) {
+             head = cur_item = info;
          } else {
-             term_printf(" [not inserted]");
+             cur_item->next = info;
+             cur_item = info;
          }
-         term_printf("\n");
      }
+     return head;
+ }
+ /* Consider exposing this as a full fledged QMP command */
+ static BlockStats *qmp_query_blockstat(const BlockDriverState *bs, Error **errp)
+ {
+     BlockStats *s;
+     s = g_malloc0(sizeof(*s));
+     if (bs->device_name[0]) {
+         s->has_device = true;
+         s->device = g_strdup(bs->device_name);
+     }
+     s->stats = g_malloc0(sizeof(*s->stats));
+     s->stats->rd_bytes = bs->nr_bytes[BDRV_ACCT_READ];
+     s->stats->wr_bytes = bs->nr_bytes[BDRV_ACCT_WRITE];
+     s->stats->rd_operations = bs->nr_ops[BDRV_ACCT_READ];
+     s->stats->wr_operations = bs->nr_ops[BDRV_ACCT_WRITE];
+     s->stats->wr_highest_offset = bs->wr_highest_sector * BDRV_SECTOR_SIZE;
+     s->stats->flush_operations = bs->nr_ops[BDRV_ACCT_FLUSH];
+     s->stats->wr_total_time_ns = bs->total_time_ns[BDRV_ACCT_WRITE];
+     s->stats->rd_total_time_ns = bs->total_time_ns[BDRV_ACCT_READ];
+     s->stats->flush_total_time_ns = bs->total_time_ns[BDRV_ACCT_FLUSH];
+     if (bs->file) {
+         s->has_parent = true;
+         s->parent = qmp_query_blockstat(bs->file, NULL);
+     }
+     return s;
  }
- /* The "info blockstats" command. */
+ BlockStatsList *qmp_query_blockstats(Error **errp)
- void bdrv_info_stats (void)
  {
+     BlockStatsList *head = NULL, *cur_item = NULL;
      BlockDriverState *bs;
-     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
+     QTAILQ_FOREACH(bs, &bdrv_states, list) {
-         term_printf ("%s:"
+         BlockStatsList *info = g_malloc0(sizeof(*info));
-                      " rd_bytes=%" PRIu64
+         info->value = qmp_query_blockstat(bs, NULL);
-                      " wr_bytes=%" PRIu64
-                      " rd_operations=%" PRIu64
+         /* XXX: waiting for the qapi to support GSList */
-                      " wr_operations=%" PRIu64
+         if (!cur_item) {
-                      "\n",
+             head = cur_item = info;
-                      bs->device_name,
+         } else {
-                      bs->rd_bytes, bs->wr_bytes,
+             cur_item->next = info;
-                      bs->rd_ops, bs->wr_ops);
+             cur_item = info;
+         }
      }
+     return head;
  }
  const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
- Line 1152  const char *bdrv_get_encrypted_filename(
+ Line 1948  const char *bdrv_get_encrypted_filename(
  void bdrv_get_backing_filename(BlockDriverState *bs,
                                 char *filename, int filename_size)
  {
-     if (!bs->backing_hd) {
+     pstrcpy(filename, filename_size, bs->backing_file);
-         pstrcpy(filename, filename_size, "");
-     } else {
-         pstrcpy(filename, filename_size, bs->backing_file);
-     }
  }
  int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
- Line 1167  int bdrv_write_compressed(BlockDriverSta
+ Line 1959  int bdrv_write_compressed(BlockDriverSta
          return -ENOMEDIUM;
      if (!drv->bdrv_write_compressed)
          return -ENOTSUP;
+     if (bdrv_check_request(bs, sector_num, nb_sectors))
+         return -EIO;
+     if (bs->dirty_bitmap) {
+         set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
+     }
      return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
  }
- Line 1181  int bdrv_get_info(BlockDriverState *bs,
+ Line 1980  int bdrv_get_info(BlockDriverState *bs,
      return drv->bdrv_get_info(bs, bdi);
  }
- int bdrv_put_buffer(BlockDriverState *bs, const uint8_t *buf, int64_t pos, int size)
+ int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
+                       int64_t pos, int size)
  {
      BlockDriver *drv = bs->drv;
      if (!drv)
          return -ENOMEDIUM;
-     if (!drv->bdrv_put_buffer)
+     if (drv->bdrv_save_vmstate)
-         return -ENOTSUP;
+         return drv->bdrv_save_vmstate(bs, buf, pos, size);
-     return drv->bdrv_put_buffer(bs, buf, pos, size);
+     if (bs->file)
+         return bdrv_save_vmstate(bs->file, buf, pos, size);
+     return -ENOTSUP;
  }
- int bdrv_get_buffer(BlockDriverState *bs, uint8_t *buf, int64_t pos, int size)
+ int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
+                       int64_t pos, int size)
  {
      BlockDriver *drv = bs->drv;
      if (!drv)
          return -ENOMEDIUM;
-     if (!drv->bdrv_get_buffer)
+     if (drv->bdrv_load_vmstate)
-         return -ENOTSUP;
+         return drv->bdrv_load_vmstate(bs, buf, pos, size);
-     return drv->bdrv_get_buffer(bs, buf, pos, size);
+     if (bs->file)
+         return bdrv_load_vmstate(bs->file, buf, pos, size);
+     return -ENOTSUP;
+ }
+ void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
+ {
+     BlockDriver *drv = bs->drv;
+     if (!drv || !drv->bdrv_debug_event) {
+         return;
+     }
+     return drv->bdrv_debug_event(bs, event);
  }
  /**************************************************************/
  /* handling of snapshots */
+ int bdrv_can_snapshot(BlockDriverState *bs)
+ {
+     BlockDriver *drv = bs->drv;
+     if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
+         return 0;
+     }
+     if (!drv->bdrv_snapshot_create) {
+         if (bs->file != NULL) {
+             return bdrv_can_snapshot(bs->file);
+         }
+         return 0;
+     }
+     return 1;
+ }
+ int bdrv_is_snapshot(BlockDriverState *bs)
+ {
+     return !!(bs->open_flags & BDRV_O_SNAPSHOT);
+ }
+ BlockDriverState *bdrv_snapshots(void)
+ {
+     BlockDriverState *bs;
+     if (bs_snapshots) {
+         return bs_snapshots;
+     }
+     bs = NULL;
+     while ((bs = bdrv_next(bs))) {
+         if (bdrv_can_snapshot(bs)) {
+             bs_snapshots = bs;
+             return bs;
+         }
+     }
+     return NULL;
+ }
  int bdrv_snapshot_create(BlockDriverState *bs,
                           QEMUSnapshotInfo *sn_info)
  {
      BlockDriver *drv = bs->drv;
      if (!drv)
          return -ENOMEDIUM;
-     if (!drv->bdrv_snapshot_create)
+     if (drv->bdrv_snapshot_create)
-         return -ENOTSUP;
+         return drv->bdrv_snapshot_create(bs, sn_info);
-     return drv->bdrv_snapshot_create(bs, sn_info);
+     if (bs->file)
+         return bdrv_snapshot_create(bs->file, sn_info);
+     return -ENOTSUP;
  }
  int bdrv_snapshot_goto(BlockDriverState *bs,
                         const char *snapshot_id)
  {
      BlockDriver *drv = bs->drv;
+     int ret, open_ret;
      if (!drv)
          return -ENOMEDIUM;
-     if (!drv->bdrv_snapshot_goto)
+     if (drv->bdrv_snapshot_goto)
-         return -ENOTSUP;
+         return drv->bdrv_snapshot_goto(bs, snapshot_id);
-     return drv->bdrv_snapshot_goto(bs, snapshot_id);
+     if (bs->file) {
+         drv->bdrv_close(bs);
+         ret = bdrv_snapshot_goto(bs->file, snapshot_id);
+         open_ret = drv->bdrv_open(bs, bs->open_flags);
+         if (open_ret < 0) {
+             bdrv_delete(bs->file);
+             bs->drv = NULL;
+             return open_ret;
+         }
+         return ret;
+     }
+     return -ENOTSUP;
  }
  int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
- Line 1231  int bdrv_snapshot_delete(BlockDriverStat
+ Line 2105  int bdrv_snapshot_delete(BlockDriverStat
      BlockDriver *drv = bs->drv;
      if (!drv)
          return -ENOMEDIUM;
-     if (!drv->bdrv_snapshot_delete)
+     if (drv->bdrv_snapshot_delete)
-         return -ENOTSUP;
+         return drv->bdrv_snapshot_delete(bs, snapshot_id);
-     return drv->bdrv_snapshot_delete(bs, snapshot_id);
+     if (bs->file)
+         return bdrv_snapshot_delete(bs->file, snapshot_id);
+     return -ENOTSUP;
  }
  int bdrv_snapshot_list(BlockDriverState *bs,
- Line 1242  int bdrv_snapshot_list(BlockDriverState
+ Line 2118  int bdrv_snapshot_list(BlockDriverState
      BlockDriver *drv = bs->drv;
      if (!drv)
          return -ENOMEDIUM;
-     if (!drv->bdrv_snapshot_list)
+     if (drv->bdrv_snapshot_list)
-         return -ENOTSUP;
+         return drv->bdrv_snapshot_list(bs, psn_info);
-     return drv->bdrv_snapshot_list(bs, psn_info);
+     if (bs->file)
+         return bdrv_snapshot_list(bs->file, psn_info);
+     return -ENOTSUP;
+ }
+ int bdrv_snapshot_load_tmp(BlockDriverState *bs,
+         const char *snapshot_name)
+ {
+     BlockDriver *drv = bs->drv;
+     if (!drv) {
+         return -ENOMEDIUM;
+     }
+     if (!bs->read_only) {
+         return -EINVAL;
+     }
+     if (drv->bdrv_snapshot_load_tmp) {
+         return drv->bdrv_snapshot_load_tmp(bs, snapshot_name);
+     }
+     return -ENOTSUP;
  }
  #define NB_SUFFIXES 4
- Line 1320  char *bdrv_snapshot_dump(char *buf, int
+ Line 2214  char *bdrv_snapshot_dump(char *buf, int
      return buf;
  }
  /**************************************************************/
  /* async I/Os */
- typedef struct VectorTranslationState {
+ BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
-     QEMUIOVector *iov;
+                                  QEMUIOVector *qiov, int nb_sectors,
-     uint8_t *bounce;
+                                  BlockDriverCompletionFunc *cb, void *opaque)
-     int is_write;
+ {
-     BlockDriverAIOCB *aiocb;
+     trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
-     BlockDriverAIOCB *this_aiocb;
- } VectorTranslationState;
+     return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
+                                  cb, opaque, false);
+ }
- static void bdrv_aio_rw_vector_cb(void *opaque, int ret)
+ BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
+                                   QEMUIOVector *qiov, int nb_sectors,
+                                   BlockDriverCompletionFunc *cb, void *opaque)
  {
-     VectorTranslationState *s = opaque;
+     trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
-     if (!s->is_write) {
+     return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
-         qemu_iovec_from_buffer(s->iov, s->bounce, s->iov->size);
+                                  cb, opaque, true);
-     }
-     qemu_vfree(s->bounce);
-     s->this_aiocb->cb(s->this_aiocb->opaque, ret);
-     qemu_aio_release(s->this_aiocb);
  }
- static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
-                                             int64_t sector_num,
-                                             QEMUIOVector *iov,
-                                             int nb_sectors,
-                                             BlockDriverCompletionFunc *cb,
-                                             void *opaque,
-                                             int is_write)
+ typedef struct MultiwriteCB {
+     int error;
+     int num_requests;
+     int num_callbacks;
+     struct {
+         BlockDriverCompletionFunc *cb;
+         void *opaque;
+         QEMUIOVector *free_qiov;
+         void *free_buf;
+     } callbacks[];
+ } MultiwriteCB;
+ static void multiwrite_user_cb(MultiwriteCB *mcb)
  {
-     VectorTranslationState *s = qemu_mallocz(sizeof(*s));
+     int i;
-     BlockDriverAIOCB *aiocb = qemu_aio_get(bs, cb, opaque);
-     s->this_aiocb = aiocb;
+     for (i = 0; i < mcb->num_callbacks; i++) {
-     s->iov = iov;
+         mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
-     s->bounce = qemu_memalign(512, nb_sectors * 512);
+         if (mcb->callbacks[i].free_qiov) {
-     s->is_write = is_write;
+             qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
-     if (is_write) {
+         }
-         qemu_iovec_to_buffer(s->iov, s->bounce);
+         g_free(mcb->callbacks[i].free_qiov);
-         s->aiocb = bdrv_aio_write(bs, sector_num, s->bounce, nb_sectors,
+         qemu_vfree(mcb->callbacks[i].free_buf);
-                                   bdrv_aio_rw_vector_cb, s);
-     } else {
-         s->aiocb = bdrv_aio_read(bs, sector_num, s->bounce, nb_sectors,
-                                  bdrv_aio_rw_vector_cb, s);
      }
-     return aiocb;
  }
- BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
+ static void multiwrite_cb(void *opaque, int ret)
-                                  QEMUIOVector *iov, int nb_sectors,
-                                  BlockDriverCompletionFunc *cb, void *opaque)
  {
-     if (bdrv_check_request(bs, sector_num, nb_sectors))
+     MultiwriteCB *mcb = opaque;
-         return NULL;
+     trace_multiwrite_cb(mcb, ret);
+     if (ret < 0 && !mcb->error) {
+         mcb->error = ret;
+     }
-     return bdrv_aio_rw_vector(bs, sector_num, iov, nb_sectors,
+     mcb->num_requests--;
-                               cb, opaque, 0);
+     if (mcb->num_requests == 0) {
+         multiwrite_user_cb(mcb);
+         g_free(mcb);
+     }
  }
- BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
+ static int multiwrite_req_compare(const void *a, const void *b)
-                                   QEMUIOVector *iov, int nb_sectors,
-                                   BlockDriverCompletionFunc *cb, void *opaque)
  {
-     if (bdrv_check_request(bs, sector_num, nb_sectors))
+     const BlockRequest *req1 = a, *req2 = b;
-         return NULL;
-     return bdrv_aio_rw_vector(bs, sector_num, iov, nb_sectors,
+     /*
-                               cb, opaque, 1);
+      * Note that we can't simply subtract req2->sector from req1->sector
+      * here as that could overflow the return value.
+      */
+     if (req1->sector > req2->sector) {
+         return 1;
+     } else if (req1->sector < req2->sector) {
+         return -1;
+     } else {
+         return 0;
+     }
  }
- BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
+ /*
-                                 uint8_t *buf, int nb_sectors,
+  * Takes a bunch of requests and tries to merge them. Returns the number of
-                                 BlockDriverCompletionFunc *cb, void *opaque)
+  * requests that remain after merging.
+  */
+ static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
+     int num_reqs, MultiwriteCB *mcb)
  {
-     BlockDriver *drv = bs->drv;
+     int i, outidx;
-     BlockDriverAIOCB *ret;
-     if (!drv)
+     // Sort requests by start sector
-         return NULL;
+     qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
-     if (bdrv_check_request(bs, sector_num, nb_sectors))
-         return NULL;
+     // Check if adjacent requests touch the same clusters. If so, combine them,
+     // filling up gaps with zero sectors.
+     outidx = 0;
+     for (i = 1; i < num_reqs; i++) {
+         int merge = 0;
+         int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
+         // This handles the cases that are valid for all block drivers, namely
+         // exactly sequential writes and overlapping writes.
+         if (reqs[i].sector <= oldreq_last) {
+             merge = 1;
+         }
+         // The block driver may decide that it makes sense to combine requests
+         // even if there is a gap of some sectors between them. In this case,
+         // the gap is filled with zeros (therefore only applicable for yet
+         // unused space in format like qcow2).
+         if (!merge && bs->drv->bdrv_merge_requests) {
+             merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
+         }
+         if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
+             merge = 0;
+         }
-     ret = drv->bdrv_aio_read(bs, sector_num, buf, nb_sectors, cb, opaque);
+         if (merge) {
+             size_t size;
+             QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
+             qemu_iovec_init(qiov,
+                 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
+             // Add the first request to the merged one. If the requests are
+             // overlapping, drop the last sectors of the first request.
+             size = (reqs[i].sector - reqs[outidx].sector) << 9;
+             qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
+             // We might need to add some zeros between the two requests
+             if (reqs[i].sector > oldreq_last) {
+                 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
+                 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
+                 memset(buf, 0, zero_bytes);
+                 qemu_iovec_add(qiov, buf, zero_bytes);
+                 mcb->callbacks[i].free_buf = buf;
+             }
+             // Add the second request
+             qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
-     if (ret) {
+             reqs[outidx].nb_sectors = qiov->size >> 9;
-         /* Update stats even though technically transfer has not happened. */
+             reqs[outidx].qiov = qiov;
-         bs->rd_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
-         bs->rd_ops ++;
+             mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
+         } else {
+             outidx++;
+             reqs[outidx].sector     = reqs[i].sector;
+             reqs[outidx].nb_sectors = reqs[i].nb_sectors;
+             reqs[outidx].qiov       = reqs[i].qiov;
+         }
      }
-     return ret;
+     return outidx + 1;
  }
- BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,
+ /*
-                                  const uint8_t *buf, int nb_sectors,
+  * Submit multiple AIO write requests at once.
-                                  BlockDriverCompletionFunc *cb, void *opaque)
+  *
+  * On success, the function returns 0 and all requests in the reqs array have
+  * been submitted. In error case this function returns -1, and any of the
+  * requests may or may not be submitted yet. In particular, this means that the
+  * callback will be called for some of the requests, for others it won't. The
+  * caller must check the error field of the BlockRequest to wait for the right
+  * callbacks (if error != 0, no callback will be called).
+  *
+  * The implementation may modify the contents of the reqs array, e.g. to merge
+  * requests. However, the fields opaque and error are left unmodified as they
+  * are used to signal failure for a single request to the caller.
+  */
+ int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
  {
-     BlockDriver *drv = bs->drv;
+     BlockDriverAIOCB *acb;
-     BlockDriverAIOCB *ret;
+     MultiwriteCB *mcb;
+     int i;
-     if (!drv)
+     /* don't submit writes if we don't have a medium */
-         return NULL;
+     if (bs->drv == NULL) {
-     if (bs->read_only)
+         for (i = 0; i < num_reqs; i++) {
-         return NULL;
+             reqs[i].error = -ENOMEDIUM;
-     if (bdrv_check_request(bs, sector_num, nb_sectors))
+         }
-         return NULL;
+         return -1;
+     }
-     ret = drv->bdrv_aio_write(bs, sector_num, buf, nb_sectors, cb, opaque);
+     if (num_reqs == 0) {
+         return 0;
+     }
-     if (ret) {
+     // Create MultiwriteCB structure
-         /* Update stats even though technically transfer has not happened. */
+     mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
-         bs->wr_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
+     mcb->num_requests = 0;
-         bs->wr_ops ++;
+     mcb->num_callbacks = num_reqs;
+     for (i = 0; i < num_reqs; i++) {
+         mcb->callbacks[i].cb = reqs[i].cb;
+         mcb->callbacks[i].opaque = reqs[i].opaque;
+     }
+     // Check for mergable requests
+     num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
+     trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
+     /*
+      * Run the aio requests. As soon as one request can't be submitted
+      * successfully, fail all requests that are not yet submitted (we must
+      * return failure for all requests anyway)
+      *
+      * num_requests cannot be set to the right value immediately: If
+      * bdrv_aio_writev fails for some request, num_requests would be too high
+      * and therefore multiwrite_cb() would never recognize the multiwrite
+      * request as completed. We also cannot use the loop variable i to set it
+      * when the first request fails because the callback may already have been
+      * called for previously submitted requests. Thus, num_requests must be
+      * incremented for each request that is submitted.
+      *
+      * The problem that callbacks may be called early also means that we need
+      * to take care that num_requests doesn't become 0 before all requests are
+      * submitted - multiwrite_cb() would consider the multiwrite request
+      * completed. A dummy request that is "completed" by a manual call to
+      * multiwrite_cb() takes care of this.
+      */
+     mcb->num_requests = 1;
+     // Run the aio requests
+     for (i = 0; i < num_reqs; i++) {
+         mcb->num_requests++;
+         acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
+             reqs[i].nb_sectors, multiwrite_cb, mcb);
+         if (acb == NULL) {
+             // We can only fail the whole thing if no request has been
+             // submitted yet. Otherwise we'll wait for the submitted AIOs to
+             // complete and report the error in the callback.
+             if (i == 0) {
+                 trace_bdrv_aio_multiwrite_earlyfail(mcb);
+                 goto fail;
+             } else {
+                 trace_bdrv_aio_multiwrite_latefail(mcb, i);
+                 multiwrite_cb(mcb, -EIO);
+                 break;
+             }
+         }
      }
-     return ret;
+     /* Complete the dummy request */
- }
+     multiwrite_cb(mcb, 0);
- void bdrv_aio_cancel(BlockDriverAIOCB *acb)
+     return 0;
- {
-     BlockDriver *drv = acb->bs->drv;
-     if (acb->cb == bdrv_aio_rw_vector_cb) {
+ fail:
-         VectorTranslationState *s = acb->opaque;
+     for (i = 0; i < mcb->num_callbacks; i++) {
-         acb = s->aiocb;
+         reqs[i].error = -EIO;
      }
+     g_free(mcb);
+     return -1;
+ }
-     drv->bdrv_aio_cancel(acb);
+ void bdrv_aio_cancel(BlockDriverAIOCB *acb)
+ {
+     acb->pool->cancel(acb);
  }
  /**************************************************************/
  /* async block device emulation */
+ typedef struct BlockDriverAIOCBSync {
+     BlockDriverAIOCB common;
+     QEMUBH *bh;
+     int ret;
+     /* vector translation state */
+     QEMUIOVector *qiov;
+     uint8_t *bounce;
+     int is_write;
+ } BlockDriverAIOCBSync;
+ static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
+ {
+     BlockDriverAIOCBSync *acb =
+         container_of(blockacb, BlockDriverAIOCBSync, common);
+     qemu_bh_delete(acb->bh);
+     acb->bh = NULL;
+     qemu_aio_release(acb);
+ }
+ 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_bh_delete(acb->bh);
+     acb->bh = NULL;
      qemu_aio_release(acb);
  }
- static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
+ static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
-         int64_t sector_num, uint8_t *buf, int nb_sectors,
+                                             int64_t sector_num,
-         BlockDriverCompletionFunc *cb, void *opaque)
+                                             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 = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
+     } else {
+         acb->ret = bs->drv->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,
+ static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
-         int64_t sector_num, const uint8_t *buf, int nb_sectors,
+         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
          BlockDriverCompletionFunc *cb, void *opaque)
  {
-     BlockDriverAIOCBSync *acb;
+     return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
-     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;
  }
- 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;
+     return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
-     qemu_bh_cancel(acb->bh);
-     qemu_aio_release(acb);
  }
- /**************************************************************/
- /* sync block device emulation */
- static void bdrv_rw_em_cb(void *opaque, int ret)
+ typedef struct BlockDriverAIOCBCoroutine {
+     BlockDriverAIOCB common;
+     BlockRequest req;
+     bool is_write;
+     QEMUBH* bh;
+ } BlockDriverAIOCBCoroutine;
+ static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
  {
-     *(int *)opaque = ret;
+     qemu_aio_flush();
  }
- #define NOT_DONE 0x7fffffff
+ static AIOPool bdrv_em_co_aio_pool = {
+     .aiocb_size         = sizeof(BlockDriverAIOCBCoroutine),
+     .cancel             = bdrv_aio_co_cancel_em,
+ };
- static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
+ static void bdrv_co_em_bh(void *opaque)
-                         uint8_t *buf, int nb_sectors)
  {
-     int async_ret;
+     BlockDriverAIOCBCoroutine *acb = opaque;
-     BlockDriverAIOCB *acb;
-     async_ret = NOT_DONE;
+     acb->common.cb(acb->common.opaque, acb->req.error);
-     acb = bdrv_aio_read(bs, sector_num, buf, nb_sectors,
+     qemu_bh_delete(acb->bh);
-                         bdrv_rw_em_cb, &async_ret);
+     qemu_aio_release(acb);
-     if (acb == NULL)
+ }
-         return -1;
+ /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
+ static void coroutine_fn bdrv_co_do_rw(void *opaque)
+ {
+     BlockDriverAIOCBCoroutine *acb = opaque;
+     BlockDriverState *bs = acb->common.bs;
-     while (async_ret == NOT_DONE) {
+     if (!acb->is_write) {
-         qemu_aio_wait();
+         acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
+             acb->req.nb_sectors, acb->req.qiov);
+     } else {
+         acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
+             acb->req.nb_sectors, acb->req.qiov);
      }
-     return async_ret;
+     acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
+     qemu_bh_schedule(acb->bh);
+ }
+ static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
+                                                int64_t sector_num,
+                                                QEMUIOVector *qiov,
+                                                int nb_sectors,
+                                                BlockDriverCompletionFunc *cb,
+                                                void *opaque,
+                                                bool is_write)
+ {
+     Coroutine *co;
+     BlockDriverAIOCBCoroutine *acb;
+     acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
+     acb->req.sector = sector_num;
+     acb->req.nb_sectors = nb_sectors;
+     acb->req.qiov = qiov;
+     acb->is_write = is_write;
+     co = qemu_coroutine_create(bdrv_co_do_rw);
+     qemu_coroutine_enter(co, acb);
+     return &acb->common;
  }
- static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
+ static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
-                          const uint8_t *buf, int nb_sectors)
  {
-     int async_ret;
+     BlockDriverAIOCBCoroutine *acb = opaque;
-     BlockDriverAIOCB *acb;
+     BlockDriverState *bs = acb->common.bs;
-     async_ret = NOT_DONE;
+     acb->req.error = bdrv_co_flush(bs);
-     acb = bdrv_aio_write(bs, sector_num, buf, nb_sectors,
+     acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
-                          bdrv_rw_em_cb, &async_ret);
+     qemu_bh_schedule(acb->bh);
-     if (acb == NULL)
+ }
-         return -1;
-     while (async_ret == NOT_DONE) {
+ BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
-         qemu_aio_wait();
+         BlockDriverCompletionFunc *cb, void *opaque)
-     }
+ {
-     return async_ret;
+     trace_bdrv_aio_flush(bs, opaque);
+     Coroutine *co;
+     BlockDriverAIOCBCoroutine *acb;
+     acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
+     co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
+     qemu_coroutine_enter(co, acb);
+     return &acb->common;
+ }
+ static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
+ {
+     BlockDriverAIOCBCoroutine *acb = opaque;
+     BlockDriverState *bs = acb->common.bs;
+     acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
+     acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
+     qemu_bh_schedule(acb->bh);
+ }
+ BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
+         int64_t sector_num, int nb_sectors,
+         BlockDriverCompletionFunc *cb, void *opaque)
+ {
+     Coroutine *co;
+     BlockDriverAIOCBCoroutine *acb;
+     trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
+     acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
+     acb->req.sector = sector_num;
+     acb->req.nb_sectors = nb_sectors;
+     co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
+     qemu_coroutine_enter(co, acb);
+     return &acb->common;
  }
  void bdrv_init(void)
  {
-     bdrv_register(&bdrv_raw);
+     module_call_init(MODULE_INIT_BLOCK);
-     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);
-     bdrv_register(&bdrv_nbd);
  }
- void *qemu_aio_get(BlockDriverState *bs, BlockDriverCompletionFunc *cb,
+ void bdrv_init_with_whitelist(void)
-                    void *opaque)
+ {
+     use_bdrv_whitelist = 1;
+     bdrv_init();
+ }
+ void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
+                    BlockDriverCompletionFunc *cb, void *opaque)
  {
-     BlockDriver *drv;
      BlockDriverAIOCB *acb;
-     drv = bs->drv;
+     if (pool->free_aiocb) {
-     if (drv->free_aiocb) {
+         acb = pool->free_aiocb;
-         acb = drv->free_aiocb;
+         pool->free_aiocb = acb->next;
-         drv->free_aiocb = acb->next;
      } else {
-         acb = qemu_mallocz(drv->aiocb_size);
+         acb = g_malloc0(pool->aiocb_size);
+         acb->pool = pool;
      }
      acb->bs = bs;
      acb->cb = cb;
- Line 1589  void *qemu_aio_get(BlockDriverState *bs,
+ Line 2718  void *qemu_aio_get(BlockDriverState *bs,
  void qemu_aio_release(void *p)
  {
-     BlockDriverAIOCB *acb = p;
+     BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
-     BlockDriver *drv = acb->bs->drv;
+     AIOPool *pool = acb->pool;
-     acb->next = drv->free_aiocb;
+     acb->next = pool->free_aiocb;
-     drv->free_aiocb = acb;
+     pool->free_aiocb = acb;
+ }
+ /**************************************************************/
+ /* Coroutine block device emulation */
+ typedef struct CoroutineIOCompletion {
+     Coroutine *coroutine;
+     int ret;
+ } CoroutineIOCompletion;
+ static void bdrv_co_io_em_complete(void *opaque, int ret)
+ {
+     CoroutineIOCompletion *co = opaque;
+     co->ret = ret;
+     qemu_coroutine_enter(co->coroutine, NULL);
+ }
+ static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
+                                       int nb_sectors, QEMUIOVector *iov,
+                                       bool is_write)
+ {
+     CoroutineIOCompletion co = {
+         .coroutine = qemu_coroutine_self(),
+     };
+     BlockDriverAIOCB *acb;
+     if (is_write) {
+         acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
+                                        bdrv_co_io_em_complete, &co);
+     } else {
+         acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
+                                       bdrv_co_io_em_complete, &co);
+     }
+     trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
+     if (!acb) {
+         return -EIO;
+     }
+     qemu_coroutine_yield();
+     return co.ret;
+ }
+ static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
+                                          int64_t sector_num, int nb_sectors,
+                                          QEMUIOVector *iov)
+ {
+     return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
+ }
+ static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
+                                          int64_t sector_num, int nb_sectors,
+                                          QEMUIOVector *iov)
+ {
+     return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
+ }
+ static void coroutine_fn bdrv_flush_co_entry(void *opaque)
+ {
+     RwCo *rwco = opaque;
+     rwco->ret = bdrv_co_flush(rwco->bs);
+ }
+ int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
+ {
+     int ret;
+     if (!bs->drv) {
+         return 0;
+     }
+     /* Write back cached data to the OS even with cache=unsafe */
+     if (bs->drv->bdrv_co_flush_to_os) {
+         ret = bs->drv->bdrv_co_flush_to_os(bs);
+         if (ret < 0) {
+             return ret;
+         }
+     }
+     /* But don't actually force it to the disk with cache=unsafe */
+     if (bs->open_flags & BDRV_O_NO_FLUSH) {
+         return 0;
+     }
+     if (bs->drv->bdrv_co_flush_to_disk) {
+         return bs->drv->bdrv_co_flush_to_disk(bs);
+     } else if (bs->drv->bdrv_aio_flush) {
+         BlockDriverAIOCB *acb;
+         CoroutineIOCompletion co = {
+             .coroutine = qemu_coroutine_self(),
+         };
+         acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
+         if (acb == NULL) {
+             return -EIO;
+         } else {
+             qemu_coroutine_yield();
+             return co.ret;
+         }
+     } else {
+         /*
+          * Some block drivers always operate in either writethrough or unsafe
+          * mode and don't support bdrv_flush therefore. Usually qemu doesn't
+          * know how the server works (because the behaviour is hardcoded or
+          * depends on server-side configuration), so we can't ensure that
+          * everything is safe on disk. Returning an error doesn't work because
+          * that would break guests even if the server operates in writethrough
+          * mode.
+          *
+          * Let's hope the user knows what he's doing.
+          */
+         return 0;
+     }
+ }
+ void bdrv_invalidate_cache(BlockDriverState *bs)
+ {
+     if (bs->drv && bs->drv->bdrv_invalidate_cache) {
+         bs->drv->bdrv_invalidate_cache(bs);
+     }
+ }
+ void bdrv_invalidate_cache_all(void)
+ {
+     BlockDriverState *bs;
+     QTAILQ_FOREACH(bs, &bdrv_states, list) {
+         bdrv_invalidate_cache(bs);
+     }
+ }
+ int bdrv_flush(BlockDriverState *bs)
+ {
+     Coroutine *co;
+     RwCo rwco = {
+         .bs = bs,
+         .ret = NOT_DONE,
+     };
+     if (qemu_in_coroutine()) {
+         /* Fast-path if already in coroutine context */
+         bdrv_flush_co_entry(&rwco);
+     } else {
+         co = qemu_coroutine_create(bdrv_flush_co_entry);
+         qemu_coroutine_enter(co, &rwco);
+         while (rwco.ret == NOT_DONE) {
+             qemu_aio_wait();
+         }
+     }
+     return rwco.ret;
+ }
+ static void coroutine_fn bdrv_discard_co_entry(void *opaque)
+ {
+     RwCo *rwco = opaque;
+     rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
+ }
+ int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
+                                  int nb_sectors)
+ {
+     if (!bs->drv) {
+         return -ENOMEDIUM;
+     } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
+         return -EIO;
+     } else if (bs->read_only) {
+         return -EROFS;
+     } else if (bs->drv->bdrv_co_discard) {
+         return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors);
+     } else if (bs->drv->bdrv_aio_discard) {
+         BlockDriverAIOCB *acb;
+         CoroutineIOCompletion co = {
+             .coroutine = qemu_coroutine_self(),
+         };
+         acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
+                                         bdrv_co_io_em_complete, &co);
+         if (acb == NULL) {
+             return -EIO;
+         } else {
+             qemu_coroutine_yield();
+             return co.ret;
+         }
+     } else {
+         return 0;
+     }
+ }
+ int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
+ {
+     Coroutine *co;
+     RwCo rwco = {
+         .bs = bs,
+         .sector_num = sector_num,
+         .nb_sectors = nb_sectors,
+         .ret = NOT_DONE,
+     };
+     if (qemu_in_coroutine()) {
+         /* Fast-path if already in coroutine context */
+         bdrv_discard_co_entry(&rwco);
+     } else {
+         co = qemu_coroutine_create(bdrv_discard_co_entry);
+         qemu_coroutine_enter(co, &rwco);
+         while (rwco.ret == NOT_DONE) {
+             qemu_aio_wait();
+         }
+     }
+     return rwco.ret;
  }
  /**************************************************************/
- Line 1604  void qemu_aio_release(void *p)
+ Line 2947  void qemu_aio_release(void *p)
  int bdrv_is_inserted(BlockDriverState *bs)
  {
      BlockDriver *drv = bs->drv;
-     int ret;
      if (!drv)
          return 0;
      if (!drv->bdrv_is_inserted)
          return 1;
-     ret = drv->bdrv_is_inserted(bs);
+     return drv->bdrv_is_inserted(bs);
-     return ret;
  }
  /**
-  * Return TRUE if the media changed since the last call to this
+  * Return whether the media changed since the last call to this
-  * function. It is currently only used for floppy disks
+  * function, or -ENOTSUP if we don't know.  Most drivers don't know.
   */
  int bdrv_media_changed(BlockDriverState *bs)
  {
      BlockDriver *drv = bs->drv;
-     int ret;
-     if (!drv || !drv->bdrv_media_changed)
+     if (drv && drv->bdrv_media_changed) {
-         ret = -ENOTSUP;
+         return drv->bdrv_media_changed(bs);
-     else
+     }
-         ret = drv->bdrv_media_changed(bs);
+     return -ENOTSUP;
-     if (ret == -ENOTSUP)
-         ret = bs->media_changed;
-     bs->media_changed = 0;
-     return ret;
  }
  /**
- Line 1638  int bdrv_media_changed(BlockDriverState
+ Line 2975  int bdrv_media_changed(BlockDriverState
  void bdrv_eject(BlockDriverState *bs, int eject_flag)
  {
      BlockDriver *drv = bs->drv;
-     int ret;
-     if (!drv || !drv->bdrv_eject) {
+     if (drv && drv->bdrv_eject) {
-         ret = -ENOTSUP;
+         drv->bdrv_eject(bs, eject_flag);
-     } else {
-         ret = drv->bdrv_eject(bs, eject_flag);
-     }
-     if (ret == -ENOTSUP) {
-         if (eject_flag)
-             bdrv_close(bs);
      }
  }
- int bdrv_is_locked(BlockDriverState *bs)
- {
-     return bs->locked;
- }
  /**
   * Lock or unlock the media (if it is locked, the user won't be able
   * to eject it manually).
   */
- void bdrv_set_locked(BlockDriverState *bs, int locked)
+ void bdrv_lock_medium(BlockDriverState *bs, bool locked)
  {
      BlockDriver *drv = bs->drv;
-     bs->locked = locked;
+     trace_bdrv_lock_medium(bs, locked);
-     if (drv && drv->bdrv_set_locked) {
-         drv->bdrv_set_locked(bs, locked);
+     if (drv && drv->bdrv_lock_medium) {
+         drv->bdrv_lock_medium(bs, locked);
      }
  }
- Line 1680  int bdrv_ioctl(BlockDriverState *bs, uns
+ Line 3006  int bdrv_ioctl(BlockDriverState *bs, uns
          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 bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
+ {
+     bs->buffer_alignment = align;
+ }
+ void *qemu_blockalign(BlockDriverState *bs, size_t size)
+ {
+     return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
+ }
+ void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
+ {
+     int64_t bitmap_size;
+     bs->dirty_count = 0;
+     if (enable) {
+         if (!bs->dirty_bitmap) {
+             bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
+                     BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
+             bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
+             bs->dirty_bitmap = g_malloc0(bitmap_size);
+         }
+     } else {
+         if (bs->dirty_bitmap) {
+             g_free(bs->dirty_bitmap);
+             bs->dirty_bitmap = NULL;
+         }
+     }
+ }
+ int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
+ {
+     int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
+     if (bs->dirty_bitmap &&
+         (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
+         return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
+             (1UL << (chunk % (sizeof(unsigned long) * 8))));
+     } else {
+         return 0;
+     }
+ }
+ void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
+                       int nr_sectors)
+ {
+     set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
+ }
+ int64_t bdrv_get_dirty_count(BlockDriverState *bs)
+ {
+     return bs->dirty_count;
+ }
+ void bdrv_set_in_use(BlockDriverState *bs, int in_use)
+ {
+     assert(bs->in_use != in_use);
+     bs->in_use = in_use;
+ }
+ int bdrv_in_use(BlockDriverState *bs)
+ {
+     return bs->in_use;
+ }
+ void bdrv_iostatus_enable(BlockDriverState *bs)
+ {
+     bs->iostatus_enabled = true;
+     bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
+ }
+ /* The I/O status is only enabled if the drive explicitly
+  * enables it _and_ the VM is configured to stop on errors */
+ bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
+ {
+     return (bs->iostatus_enabled &&
+            (bs->on_write_error == BLOCK_ERR_STOP_ENOSPC ||
+             bs->on_write_error == BLOCK_ERR_STOP_ANY    ||
+             bs->on_read_error == BLOCK_ERR_STOP_ANY));
+ }
+ void bdrv_iostatus_disable(BlockDriverState *bs)
+ {
+     bs->iostatus_enabled = false;
+ }
+ void bdrv_iostatus_reset(BlockDriverState *bs)
+ {
+     if (bdrv_iostatus_is_enabled(bs)) {
+         bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
+     }
+ }
+ /* XXX: Today this is set by device models because it makes the implementation
+    quite simple. However, the block layer knows about the error, so it's
+    possible to implement this without device models being involved */
+ void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
+ {
+     if (bdrv_iostatus_is_enabled(bs) &&
+         bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
+         assert(error >= 0);
+         bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
+                                          BLOCK_DEVICE_IO_STATUS_FAILED;
+     }
+ }
+ void
+ bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
+         enum BlockAcctType type)
+ {
+     assert(type < BDRV_MAX_IOTYPE);
+     cookie->bytes = bytes;
+     cookie->start_time_ns = get_clock();
+     cookie->type = type;
+ }
+ void
+ bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
+ {
+     assert(cookie->type < BDRV_MAX_IOTYPE);
+     bs->nr_bytes[cookie->type] += cookie->bytes;
+     bs->nr_ops[cookie->type]++;
+     bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
+ }
+ int bdrv_img_create(const char *filename, const char *fmt,
+                     const char *base_filename, const char *base_fmt,
+                     char *options, uint64_t img_size, int flags)
+ {
+     QEMUOptionParameter *param = NULL, *create_options = NULL;
+     QEMUOptionParameter *backing_fmt, *backing_file, *size;
+     BlockDriverState *bs = NULL;
+     BlockDriver *drv, *proto_drv;
+     BlockDriver *backing_drv = NULL;
+     int ret = 0;
+     /* Find driver and parse its options */
+     drv = bdrv_find_format(fmt);
+     if (!drv) {
+         error_report("Unknown file format '%s'", fmt);
+         ret = -EINVAL;
+         goto out;
+     }
+     proto_drv = bdrv_find_protocol(filename);
+     if (!proto_drv) {
+         error_report("Unknown protocol '%s'", filename);
+         ret = -EINVAL;
+         goto out;
+     }
+     create_options = append_option_parameters(create_options,
+                                               drv->create_options);
+     create_options = append_option_parameters(create_options,
+                                               proto_drv->create_options);
+     /* Create parameter list with default values */
+     param = parse_option_parameters("", create_options, param);
+     set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size);
+     /* Parse -o options */
+     if (options) {
+         param = parse_option_parameters(options, create_options, param);
+         if (param == NULL) {
+             error_report("Invalid options for file format '%s'.", fmt);
+             ret = -EINVAL;
+             goto out;
+         }
+     }
+     if (base_filename) {
+         if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE,
+                                  base_filename)) {
+             error_report("Backing file not supported for file format '%s'",
+                          fmt);
+             ret = -EINVAL;
+             goto out;
+         }
+     }
+     if (base_fmt) {
+         if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) {
+             error_report("Backing file format not supported for file "
+                          "format '%s'", fmt);
+             ret = -EINVAL;
+             goto out;
+         }
+     }
+     backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE);
+     if (backing_file && backing_file->value.s) {
+         if (!strcmp(filename, backing_file->value.s)) {
+             error_report("Error: Trying to create an image with the "
+                          "same filename as the backing file");
+             ret = -EINVAL;
+             goto out;
+         }
+     }
+     backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT);
+     if (backing_fmt && backing_fmt->value.s) {
+         backing_drv = bdrv_find_format(backing_fmt->value.s);
+         if (!backing_drv) {
+             error_report("Unknown backing file format '%s'",
+                          backing_fmt->value.s);
+             ret = -EINVAL;
+             goto out;
+         }
+     }
+     // The size for the image must always be specified, with one exception:
+     // If we are using a backing file, we can obtain the size from there
+     size = get_option_parameter(param, BLOCK_OPT_SIZE);
+     if (size && size->value.n == -1) {
+         if (backing_file && backing_file->value.s) {
+             uint64_t size;
+             char buf[32];
+             bs = bdrv_new("");
+             ret = bdrv_open(bs, backing_file->value.s, flags, backing_drv);
+             if (ret < 0) {
+                 error_report("Could not open '%s'", backing_file->value.s);
+                 goto out;
+             }
+             bdrv_get_geometry(bs, &size);
+             size *= 512;
+             snprintf(buf, sizeof(buf), "%" PRId64, size);
+             set_option_parameter(param, BLOCK_OPT_SIZE, buf);
+         } else {
+             error_report("Image creation needs a size parameter");
+             ret = -EINVAL;
+             goto out;
+         }
+     }
+     printf("Formatting '%s', fmt=%s ", filename, fmt);
+     print_option_parameters(param);
+     puts("");
+     ret = bdrv_create(drv, filename, param);
+     if (ret < 0) {
+         if (ret == -ENOTSUP) {
+             error_report("Formatting or formatting option not supported for "
+                          "file format '%s'", fmt);
+         } else if (ret == -EFBIG) {
+             error_report("The image size is too large for file format '%s'",
+                          fmt);
+         } else {
+             error_report("%s: error while creating %s: %s", filename, fmt,
+                          strerror(-ret));
+         }
+     }
+ out:
+     free_option_parameters(create_options);
+     free_option_parameters(param);
+     if (bs) {
+         bdrv_delete(bs);
+     }
+     return ret;
+ }

unix.superglobalmegacorp.com

Removed from v.1.1.1.9
changed lines
	Added in v.1.1.1.22