[BACK]Return to block.c CVS log [TXT] [DIR] Up to [Qemu by Fabrice Bellard] / qemu

Diff for /qemu/block.c between versions 1.1.1.9 and 1.1.1.23

version 1.1.1.9, 2018/04/24 17:01:52 version 1.1.1.23, 2018/04/24 19:34:33
Line 22 Line 22
  * THE SOFTWARE.   * THE SOFTWARE.
  */   */
 #include "config-host.h"  #include "config-host.h"
 #ifdef _BSD  
 /* include native header before sys-queue.h */  
 #include <sys/queue.h>  
 #endif  
   
 #include "qemu-common.h"  #include "qemu-common.h"
 #include "console.h"  #include "trace.h"
   #include "monitor.h"
 #include "block_int.h"  #include "block_int.h"
   #include "module.h"
   #include "qjson.h"
   #include "qemu-coroutine.h"
   #include "qmp-commands.h"
   #include "qemu-timer.h"
   
 #ifdef _BSD  #ifdef CONFIG_BSD
 #include <sys/types.h>  #include <sys/types.h>
 #include <sys/stat.h>  #include <sys/stat.h>
 #include <sys/ioctl.h>  #include <sys/ioctl.h>
   #include <sys/queue.h>
   #ifndef __DragonFly__
 #include <sys/disk.h>  #include <sys/disk.h>
 #endif  #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,  typedef enum {
         int64_t sector_num, uint8_t *buf, int nb_sectors,      BDRV_REQ_COPY_ON_READ = 0x1,
       BDRV_REQ_ZERO_WRITE   = 0x2,
   } BdrvRequestFlags;
   
   static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
   static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
           int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
         BlockDriverCompletionFunc *cb, void *opaque);          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);          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,
       BdrvRequestFlags flags);
   static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
       int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
       BdrvRequestFlags flags);
   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 int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
       int64_t sector_num, int nb_sectors);
   
   static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
           bool is_write, double elapsed_time, uint64_t *wait);
   static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
           double elapsed_time, uint64_t *wait);
   static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
           bool is_write, int64_t *wait);
   
   static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
       QTAILQ_HEAD_INITIALIZER(bdrv_states);
   
 BlockDriverState *bdrv_first;  static QLIST_HEAD(, BlockDriver) bdrv_drivers =
       QLIST_HEAD_INITIALIZER(bdrv_drivers);
   
 static BlockDriver *first_drv;  /* The device to use for VM snapshots */
   static BlockDriverState *bs_snapshots;
   
 int path_is_absolute(const char *path)  /* 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
   
   /* throttling disk I/O limits */
   void bdrv_io_limits_disable(BlockDriverState *bs)
   {
       bs->io_limits_enabled = false;
   
       while (qemu_co_queue_next(&bs->throttled_reqs));
   
       if (bs->block_timer) {
           qemu_del_timer(bs->block_timer);
           qemu_free_timer(bs->block_timer);
           bs->block_timer = NULL;
       }
   
       bs->slice_start = 0;
       bs->slice_end   = 0;
       bs->slice_time  = 0;
       memset(&bs->io_base, 0, sizeof(bs->io_base));
   }
   
   static void bdrv_block_timer(void *opaque)
   {
       BlockDriverState *bs = opaque;
   
       qemu_co_queue_next(&bs->throttled_reqs);
   }
   
   void bdrv_io_limits_enable(BlockDriverState *bs)
   {
       qemu_co_queue_init(&bs->throttled_reqs);
       bs->block_timer = qemu_new_timer_ns(vm_clock, bdrv_block_timer, bs);
       bs->slice_time  = 5 * BLOCK_IO_SLICE_TIME;
       bs->slice_start = qemu_get_clock_ns(vm_clock);
       bs->slice_end   = bs->slice_start + bs->slice_time;
       memset(&bs->io_base, 0, sizeof(bs->io_base));
       bs->io_limits_enabled = true;
   }
   
   bool bdrv_io_limits_enabled(BlockDriverState *bs)
   {
       BlockIOLimit *io_limits = &bs->io_limits;
       return io_limits->bps[BLOCK_IO_LIMIT_READ]
            || io_limits->bps[BLOCK_IO_LIMIT_WRITE]
            || io_limits->bps[BLOCK_IO_LIMIT_TOTAL]
            || io_limits->iops[BLOCK_IO_LIMIT_READ]
            || io_limits->iops[BLOCK_IO_LIMIT_WRITE]
            || io_limits->iops[BLOCK_IO_LIMIT_TOTAL];
   }
   
   static void bdrv_io_limits_intercept(BlockDriverState *bs,
                                        bool is_write, int nb_sectors)
   {
       int64_t wait_time = -1;
   
       if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
           qemu_co_queue_wait(&bs->throttled_reqs);
       }
   
       /* In fact, we hope to keep each request's timing, in FIFO mode. The next
        * throttled requests will not be dequeued until the current request is
        * allowed to be serviced. So if the current request still exceeds the
        * limits, it will be inserted to the head. All requests followed it will
        * be still in throttled_reqs queue.
        */
   
       while (bdrv_exceed_io_limits(bs, nb_sectors, is_write, &wait_time)) {
           qemu_mod_timer(bs->block_timer,
                          wait_time + qemu_get_clock_ns(vm_clock));
           qemu_co_queue_wait_insert_head(&bs->throttled_reqs);
       }
   
       qemu_co_queue_next(&bs->throttled_reqs);
   }
   
   /* check if the path starts with "<protocol>:" */
   static int path_has_protocol(const char *path)
 {  {
     const char *p;      const char *p;
   
 #ifdef _WIN32  #ifdef _WIN32
     /* specific case for names like: "\\.\d:" */      if (is_windows_drive(path) ||
     if (*path == '/' || *path == '\\')          is_windows_drive_prefix(path)) {
         return 1;          return 0;
       }
       p = path + strcspn(path, ":/\\");
   #else
       p = path + strcspn(path, ":/");
 #endif  #endif
     p = strchr(path, ':');  
     if (p)      return *p == ':';
         p++;  }
     else  
         p = path;  int path_is_absolute(const char *path)
   {
 #ifdef _WIN32  #ifdef _WIN32
     return (*p == '/' || *p == '\\');      /* specific case for names like: "\\.\d:" */
       if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
           return 1;
       }
       return (*path == '/' || *path == '\\');
 #else  #else
     return (*p == '/');      return (*path == '/');
 #endif  #endif
 }  }
   
Line 127  void path_combine(char *dest, int dest_s Line 270  void path_combine(char *dest, int dest_s
     }      }
 }  }
   
   void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz)
   {
       if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) {
           pstrcpy(dest, sz, bs->backing_file);
       } else {
           path_combine(dest, sz, bs->filename, bs->backing_file);
       }
   }
   
 static void bdrv_register(BlockDriver *bdrv)  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) */  /* create a new block device (by default it is empty) */
 BlockDriverState *bdrv_new(const char *device_name)  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);      pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
     if (device_name[0] != '\0') {      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;      return bs;
 }  }
   
 BlockDriver *bdrv_find_format(const char *format_name)  BlockDriver *bdrv_find_format(const char *format_name)
 {  {
     BlockDriver *drv1;      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 drv1;
           }
     }      }
     return NULL;      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;
   }
   
   typedef struct CreateCo {
       BlockDriver *drv;
       char *filename;
       QEMUOptionParameter *options;
       int ret;
   } CreateCo;
   
   static void coroutine_fn bdrv_create_co_entry(void *opaque)
   {
       CreateCo *cco = opaque;
       assert(cco->drv);
   
       cco->ret = cco->drv->bdrv_create(cco->filename, cco->options);
   }
   
   int bdrv_create(BlockDriver *drv, const char* filename,
       QEMUOptionParameter *options)
 {  {
     if (!drv->bdrv_create)      int ret;
   
       Coroutine *co;
       CreateCo cco = {
           .drv = drv,
           .filename = g_strdup(filename),
           .options = options,
           .ret = NOT_DONE,
       };
   
       if (!drv->bdrv_create) {
         return -ENOTSUP;          return -ENOTSUP;
     return drv->bdrv_create(filename, size_in_sectors, backing_file, flags);      }
   
       if (qemu_in_coroutine()) {
           /* Fast-path if already in coroutine context */
           bdrv_create_co_entry(&cco);
       } else {
           co = qemu_coroutine_create(bdrv_create_co_entry);
           qemu_coroutine_enter(co, &cco);
           while (cco.ret == NOT_DONE) {
               qemu_aio_wait();
           }
       }
   
       ret = cco.ret;
       g_free(cco.filename);
   
       return ret;
 }  }
   
 #ifdef _WIN32  int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
 void get_tmp_filename(char *filename, int size)  
 {  {
     char temp_dir[MAX_PATH];      BlockDriver *drv;
   
       drv = bdrv_find_protocol(filename);
       if (drv == NULL) {
           return -ENOENT;
       }
   
     GetTempPath(MAX_PATH, temp_dir);      return bdrv_create(drv, filename, options);
     GetTempFileName(temp_dir, "qem", 0, filename);  
 }  }
 #else  
 void get_tmp_filename(char *filename, int size)  /*
    * Create a uniquely-named empty temporary file.
    * Return 0 upon success, otherwise a negative errno value.
    */
   int get_tmp_filename(char *filename, int size)
 {  {
   #ifdef _WIN32
       char temp_dir[MAX_PATH];
       /* GetTempFileName requires that its output buffer (4th param)
          have length MAX_PATH or greater.  */
       assert(size >= MAX_PATH);
       return (GetTempPath(MAX_PATH, temp_dir)
               && GetTempFileName(temp_dir, "qem", 0, filename)
               ? 0 : -GetLastError());
   #else
     int fd;      int fd;
     const char *tmpdir;      const char *tmpdir;
     /* XXX: race condition possible */  
     tmpdir = getenv("TMPDIR");      tmpdir = getenv("TMPDIR");
     if (!tmpdir)      if (!tmpdir)
         tmpdir = "/tmp";          tmpdir = "/tmp";
     snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);      if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
           return -EOVERFLOW;
       }
     fd = mkstemp(filename);      fd = mkstemp(filename);
     close(fd);      if (fd < 0 || close(fd)) {
 }          return -errno;
       }
       return 0;
 #endif  #endif
   
 #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] == ':');  
 }  }
   
 static int is_windows_drive(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)
 {  {
     if (is_windows_drive_prefix(filename) &&      int score_max = 0, score;
         filename[2] == '\0')      BlockDriver *drv = NULL, *d;
         return 1;  
     if (strstart(filename, "\\\\.\\", NULL) ||      QLIST_FOREACH(d, &bdrv_drivers, list) {
         strstart(filename, "//./", NULL))          if (d->bdrv_probe_device) {
         return 1;              score = d->bdrv_probe_device(filename);
     return 0;              if (score > score_max) {
                   score_max = score;
                   drv = d;
               }
           }
       }
   
       return drv;
 }  }
 #endif  
   
 static BlockDriver *find_protocol(const char *filename)  BlockDriver *bdrv_find_protocol(const char *filename)
 {  {
     BlockDriver *drv1;      BlockDriver *drv1;
     char protocol[128];      char protocol[128];
     int len;      int len;
     const char *p;      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, ':');      p = strchr(filename, ':');
     if (!p)      assert(p != NULL);
         return &bdrv_raw;  
     len = p - filename;      len = p - filename;
     if (len > sizeof(protocol) - 1)      if (len > sizeof(protocol) - 1)
         len = sizeof(protocol) - 1;          len = sizeof(protocol) - 1;
     memcpy(protocol, filename, len);      memcpy(protocol, filename, len);
     protocol[len] = '\0';      protocol[len] = '\0';
     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {      QLIST_FOREACH(drv1, &bdrv_drivers, list) {
         if (drv1->protocol_name &&          if (drv1->protocol_name &&
             !strcmp(drv1->protocol_name, protocol))              !strcmp(drv1->protocol_name, protocol)) {
             return drv1;              return drv1;
           }
     }      }
     return NULL;      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;      int ret, score, score_max;
     BlockDriver *drv1, *drv;      BlockDriver *drv1, *drv;
     uint8_t buf[2048];      uint8_t buf[2048];
     BlockDriverState *bs;      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));      ret = bdrv_pread(bs, 0, buf, sizeof(buf));
     bdrv_delete(bs);      bdrv_delete(bs);
     if (ret < 0) {      if (ret < 0) {
         return NULL;          *pdrv = NULL;
           return ret;
     }      }
   
     score_max = 0;      score_max = 0;
     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {      drv = NULL;
       QLIST_FOREACH(drv1, &bdrv_drivers, list) {
         if (drv1->bdrv_probe) {          if (drv1->bdrv_probe) {
             score = drv1->bdrv_probe(buf, ret, filename);              score = drv1->bdrv_probe(buf, ret, filename);
             if (score > score_max) {              if (score > score_max) {
Line 302  static BlockDriver *find_image_format(co Line 544  static BlockDriver *find_image_format(co
             }              }
         }          }
     }      }
     return drv;      if (!drv) {
           ret = -ENOENT;
       }
       *pdrv = drv;
       return ret;
 }  }
   
 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)  /**
    * Set the current 'total_sectors' value
    */
   static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
 {  {
     BlockDriverState *bs;      BlockDriver *drv = bs->drv;
     int ret;  
   
     bs = bdrv_new("");      /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
     ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);      if (bs->sg)
     if (ret < 0) {          return 0;
         bdrv_delete(bs);  
         return ret;      /* 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->growable = 1;  
     *pbs = bs;      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;      return 0;
 }  }
   
 int bdrv_open(BlockDriverState *bs, const char *filename, int flags)  /**
    * The copy-on-read flag is actually a reference count so multiple users may
    * use the feature without worrying about clobbering its previous state.
    * Copy-on-read stays enabled until all users have called to disable it.
    */
   void bdrv_enable_copy_on_read(BlockDriverState *bs)
   {
       bs->copy_on_read++;
   }
   
   void bdrv_disable_copy_on_read(BlockDriverState *bs)
 {  {
     return bdrv_open2(bs, filename, flags, NULL);      assert(bs->copy_on_read > 0);
       bs->copy_on_read--;
 }  }
   
 int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,  /*
                BlockDriver *drv)   * 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;      int ret, open_flags;
     char tmp_filename[PATH_MAX];  
     char backing_filename[PATH_MAX];  
   
     bs->read_only = 0;      assert(drv != NULL);
     bs->is_temporary = 0;      assert(bs->file == NULL);
     bs->encrypted = 0;  
     bs->valid_key = 0;  
   
     if (flags & BDRV_O_SNAPSHOT) {      trace_bdrv_open_common(bs, filename, flags, drv->format_name);
         BlockDriverState *bs1;  
         int64_t total_size;  
         int is_protocol = 0;  
   
         /* if snapshot, we create a temporary backing file and open it      bs->open_flags = flags;
            instead of opening 'filename' directly */      bs->buffer_alignment = 512;
   
         /* if there is a backing file, use it */      assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
         bs1 = bdrv_new("");      if ((flags & BDRV_O_RDWR) && (flags & BDRV_O_COPY_ON_READ)) {
         ret = bdrv_open(bs1, filename, 0);          bdrv_enable_copy_on_read(bs);
         if (ret < 0) {      }
             bdrv_delete(bs1);  
             return ret;  
         }  
         total_size = bdrv_getlength(bs1) >> SECTOR_BITS;  
   
         if (bs1->drv && bs1->drv->protocol_name)      pstrcpy(bs->filename, sizeof(bs->filename), filename);
             is_protocol = 1;  
   
         bdrv_delete(bs1);      if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
           return -ENOTSUP;
       }
   
         get_tmp_filename(tmp_filename, sizeof(tmp_filename));      bs->drv = drv;
       bs->opaque = g_malloc0(drv->instance_size);
   
         /* Real path is meaningless for protocols */      bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
         if (is_protocol)  
             snprintf(backing_filename, sizeof(backing_filename),  
                      "%s", filename);  
         else  
             realpath(filename, backing_filename);  
   
         ret = bdrv_create(&bdrv_qcow2, tmp_filename,      /*
                           total_size, backing_filename, 0);       * Clear flags that are internal to the block layer before opening the
         if (ret < 0) {       * image.
             return ret;       */
       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);
         }          }
         filename = tmp_filename;  
         bs->is_temporary = 1;  
     }      }
   
     pstrcpy(bs->filename, sizeof(bs->filename), filename);      if (ret < 0) {
     if (flags & BDRV_O_FILE) {          goto free_and_fail;
         drv = find_protocol(filename);  
     } else if (!drv) {  
         drv = find_image_format(filename);  
     }      }
     if (!drv) {  
         ret = -ENOENT;      ret = refresh_total_sectors(bs, bs->total_sectors);
         goto unlink_and_fail;      if (ret < 0) {
     }          goto free_and_fail;
     bs->drv = drv;  
     bs->opaque = qemu_mallocz(drv->instance_size);  
     /* Note: for compatibility, we open disk image files as RDWR, and  
        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);  
         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  #ifndef _WIN32
     if (bs->is_temporary) {      if (bs->is_temporary) {
         unlink(filename);          unlink(filename);
     }      }
 #endif  #endif
     if (bs->backing_file[0] != '\0') {      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_open_common(bs, filename, flags, drv);
       if (ret < 0) {
           bdrv_delete(bs);
           return ret;
       }
       bs->growable = 1;
       *pbs = bs;
       return 0;
   }
   
   /*
    * Opens a disk image (raw, qcow2, vmdk, ...)
    */
   int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
                 BlockDriver *drv)
   {
       int ret;
       char tmp_filename[PATH_MAX];
   
       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 */          /* if there is a backing file, use it */
           bs1 = bdrv_new("");
           ret = bdrv_open(bs1, filename, 0, drv);
           if (ret < 0) {
               bdrv_delete(bs1);
               return ret;
           }
           total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
   
           if (bs1->drv && bs1->drv->protocol_name)
               is_protocol = 1;
   
           bdrv_delete(bs1);
   
           ret = get_tmp_filename(tmp_filename, sizeof(tmp_filename));
           if (ret < 0) {
               return ret;
           }
   
           /* Real path is meaningless for protocols */
           if (is_protocol)
               snprintf(backing_filename, sizeof(backing_filename),
                        "%s", filename);
           else if (!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, options);
           free_option_parameters(options);
           if (ret < 0) {
               return ret;
           }
   
           filename = tmp_filename;
           drv = bdrv_qcow2;
           bs->is_temporary = 1;
       }
   
       /* Find the right image format driver */
       if (!drv) {
           ret = find_image_format(filename, &drv);
       }
   
       if (!drv) {
           goto unlink_and_fail;
       }
   
       /* Open the image */
       ret = bdrv_open_common(bs, filename, flags, drv);
       if (ret < 0) {
           goto unlink_and_fail;
       }
   
       /* 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("");          bs->backing_hd = bdrv_new("");
         path_combine(backing_filename, sizeof(backing_filename),          bdrv_get_full_backing_filename(bs, backing_filename,
                      filename, bs->backing_file);                                         sizeof(backing_filename));
         ret = bdrv_open(bs->backing_hd, backing_filename, open_flags);  
           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) {          if (ret < 0) {
             bdrv_close(bs);              bdrv_close(bs);
             return ret;              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);  
       /* throttling disk I/O limits */
       if (bs->io_limits_enabled) {
           bdrv_io_limits_enable(bs);
       }
   
     return 0;      return 0;
   
   unlink_and_fail:
       if (bs->is_temporary) {
           unlink(filename);
       }
       return ret;
 }  }
   
 void bdrv_close(BlockDriverState *bs)  void bdrv_close(BlockDriverState *bs)
 {  {
       bdrv_flush(bs);
     if (bs->drv) {      if (bs->drv) {
         if (bs->backing_hd)          if (bs->job) {
               block_job_cancel_sync(bs->job);
           }
           bdrv_drain_all();
   
           if (bs == bs_snapshots) {
               bs_snapshots = NULL;
           }
           if (bs->backing_hd) {
             bdrv_delete(bs->backing_hd);              bdrv_delete(bs->backing_hd);
               bs->backing_hd = NULL;
           }
         bs->drv->bdrv_close(bs);          bs->drv->bdrv_close(bs);
         qemu_free(bs->opaque);          g_free(bs->opaque);
 #ifdef _WIN32  #ifdef _WIN32
         if (bs->is_temporary) {          if (bs->is_temporary) {
             unlink(bs->filename);              unlink(bs->filename);
Line 452  void bdrv_close(BlockDriverState *bs) Line 883  void bdrv_close(BlockDriverState *bs)
 #endif  #endif
         bs->opaque = NULL;          bs->opaque = NULL;
         bs->drv = NULL;          bs->drv = NULL;
           bs->copy_on_read = 0;
           bs->backing_file[0] = '\0';
           bs->backing_format[0] = '\0';
           bs->total_sectors = 0;
           bs->encrypted = 0;
           bs->valid_key = 0;
           bs->sg = 0;
           bs->growable = 0;
   
           if (bs->file != NULL) {
               bdrv_delete(bs->file);
               bs->file = NULL;
           }
   
           bdrv_dev_change_media_cb(bs, false);
       }
   
         /* call the change callback */      /*throttling disk I/O limits*/
         bs->media_changed = 1;      if (bs->io_limits_enabled) {
         if (bs->change_cb)          bdrv_io_limits_disable(bs);
             bs->change_cb(bs->change_opaque);  
     }      }
 }  }
   
 void bdrv_delete(BlockDriverState *bs)  void bdrv_close_all(void)
 {  {
     BlockDriverState **pbs;      BlockDriverState *bs;
   
     pbs = &bdrv_first;  
     while (*pbs != bs && *pbs != NULL)  
         pbs = &(*pbs)->next;  
     if (*pbs == bs)  
         *pbs = bs->next;  
   
     bdrv_close(bs);      QTAILQ_FOREACH(bs, &bdrv_states, list) {
     qemu_free(bs);          bdrv_close(bs);
       }
 }  }
   
 /* commit COW file into the raw image */  /*
 int bdrv_commit(BlockDriverState *bs)   * Wait for pending requests to complete across all BlockDriverStates
    *
    * This function does not flush data to disk, use bdrv_flush_all() for that
    * after calling this function.
    *
    * Note that completion of an asynchronous I/O operation can trigger any
    * number of other I/O operations on other devices---for example a coroutine
    * can be arbitrarily complex and a constant flow of I/O can come until the
    * coroutine is complete.  Because of this, it is not possible to have a
    * function to drain a single device's I/O queue.
    */
   void bdrv_drain_all(void)
 {  {
     BlockDriver *drv = bs->drv;      BlockDriverState *bs;
     int64_t i, total_sectors;      bool busy;
     int n, j;  
     unsigned char sector[512];  
   
     if (!drv)      do {
         return -ENOMEDIUM;          busy = qemu_aio_wait();
   
     if (bs->read_only) {          /* FIXME: We do not have timer support here, so this is effectively
         return -EACCES;           * a busy wait.
     }           */
           QTAILQ_FOREACH(bs, &bdrv_states, list) {
               if (!qemu_co_queue_empty(&bs->throttled_reqs)) {
                   qemu_co_queue_restart_all(&bs->throttled_reqs);
                   busy = true;
               }
           }
       } while (busy);
   
     if (!bs->backing_hd) {      /* If requests are still pending there is a bug somewhere */
         return -ENOTSUP;      QTAILQ_FOREACH(bs, &bdrv_states, list) {
           assert(QLIST_EMPTY(&bs->tracked_requests));
           assert(qemu_co_queue_empty(&bs->throttled_reqs));
     }      }
   }
   
     total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;  /* make a BlockDriverState anonymous by removing from bdrv_state list.
     for (i = 0; i < total_sectors;) {     Also, NULL terminate the device_name to prevent double remove */
         if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {  void bdrv_make_anon(BlockDriverState *bs)
             for(j = 0; j < n; j++) {  {
                 if (bdrv_read(bs, i, sector, 1) != 0) {      if (bs->device_name[0] != '\0') {
                     return -EIO;          QTAILQ_REMOVE(&bdrv_states, bs, list);
                 }      }
       bs->device_name[0] = '\0';
   }
   
                 if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {  static void bdrv_rebind(BlockDriverState *bs)
                     return -EIO;  {
                 }      if (bs->drv && bs->drv->bdrv_rebind) {
                 i++;          bs->drv->bdrv_rebind(bs);
             }  
         } else {  
             i += n;  
         }  
     }      }
   }
   
     if (drv->bdrv_make_empty)  /*
         return drv->bdrv_make_empty(bs);   * Add new bs contents at the top of an image chain while the chain is
    * live, while keeping required fields on the top layer.
    *
    * This will modify the BlockDriverState fields, and swap contents
    * between bs_new and bs_top. Both bs_new and bs_top are modified.
    *
    * bs_new is required to be anonymous.
    *
    * This function does not create any image files.
    */
   void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
   {
       BlockDriverState tmp;
   
     return 0;      /* bs_new must be anonymous */
       assert(bs_new->device_name[0] == '\0');
   
       tmp = *bs_new;
   
       /* there are some fields that need to stay on the top layer: */
       tmp.open_flags        = bs_top->open_flags;
   
       /* dev info */
       tmp.dev_ops           = bs_top->dev_ops;
       tmp.dev_opaque        = bs_top->dev_opaque;
       tmp.dev               = bs_top->dev;
       tmp.buffer_alignment  = bs_top->buffer_alignment;
       tmp.copy_on_read      = bs_top->copy_on_read;
   
       /* i/o timing parameters */
       tmp.slice_time        = bs_top->slice_time;
       tmp.slice_start       = bs_top->slice_start;
       tmp.slice_end         = bs_top->slice_end;
       tmp.io_limits         = bs_top->io_limits;
       tmp.io_base           = bs_top->io_base;
       tmp.throttled_reqs    = bs_top->throttled_reqs;
       tmp.block_timer       = bs_top->block_timer;
       tmp.io_limits_enabled = bs_top->io_limits_enabled;
   
       /* geometry */
       tmp.cyls              = bs_top->cyls;
       tmp.heads             = bs_top->heads;
       tmp.secs              = bs_top->secs;
       tmp.translation       = bs_top->translation;
   
       /* r/w error */
       tmp.on_read_error     = bs_top->on_read_error;
       tmp.on_write_error    = bs_top->on_write_error;
   
       /* i/o status */
       tmp.iostatus_enabled  = bs_top->iostatus_enabled;
       tmp.iostatus          = bs_top->iostatus;
   
       /* keep the same entry in bdrv_states */
       pstrcpy(tmp.device_name, sizeof(tmp.device_name), bs_top->device_name);
       tmp.list = bs_top->list;
   
       /* The contents of 'tmp' will become bs_top, as we are
        * swapping bs_new and bs_top contents. */
       tmp.backing_hd = bs_new;
       pstrcpy(tmp.backing_file, sizeof(tmp.backing_file), bs_top->filename);
       bdrv_get_format(bs_top, tmp.backing_format, sizeof(tmp.backing_format));
   
       /* swap contents of the fixed new bs and the current top */
       *bs_new = *bs_top;
       *bs_top = tmp;
   
       /* device_name[] was carried over from the old bs_top.  bs_new
        * shouldn't be in bdrv_states, so we need to make device_name[]
        * reflect the anonymity of bs_new
        */
       bs_new->device_name[0] = '\0';
   
       /* clear the copied fields in the new backing file */
       bdrv_detach_dev(bs_new, bs_new->dev);
   
       qemu_co_queue_init(&bs_new->throttled_reqs);
       memset(&bs_new->io_base,   0, sizeof(bs_new->io_base));
       memset(&bs_new->io_limits, 0, sizeof(bs_new->io_limits));
       bdrv_iostatus_disable(bs_new);
   
       /* we don't use bdrv_io_limits_disable() for this, because we don't want
        * to affect or delete the block_timer, as it has been moved to bs_top */
       bs_new->io_limits_enabled = false;
       bs_new->block_timer       = NULL;
       bs_new->slice_time        = 0;
       bs_new->slice_start       = 0;
       bs_new->slice_end         = 0;
   
       bdrv_rebind(bs_new);
       bdrv_rebind(bs_top);
 }  }
   
 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,  void bdrv_delete(BlockDriverState *bs)
                                    size_t size)  
 {  {
     int64_t len;      assert(!bs->dev);
       assert(!bs->job);
     if (!bdrv_is_inserted(bs))      assert(!bs->in_use);
         return -ENOMEDIUM;  
   
     if (bs->growable)      /* remove from list, if necessary */
         return 0;      bdrv_make_anon(bs);
   
     len = bdrv_getlength(bs);      bdrv_close(bs);
   
     if ((offset + size) > len)      assert(bs != bs_snapshots);
         return -EIO;      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;      return 0;
 }  }
   
 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,  /* TODO qdevified devices don't use this, remove when devices are qdevified */
                               int nb_sectors)  void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
 {  {
     int64_t offset;      if (bdrv_attach_dev(bs, dev) < 0) {
           abort();
       }
   }
   
     /* Deal with byte accesses */  void bdrv_detach_dev(BlockDriverState *bs, void *dev)
     if (sector_num < 0)  /* TODO change to DeviceState *dev when all users are qdevified */
         offset = -sector_num;  {
     else      assert(bs->dev == dev);
         offset = sector_num * 512;      bs->dev = NULL;
       bs->dev_ops = NULL;
       bs->dev_opaque = NULL;
       bs->buffer_alignment = 512;
   }
   
     return bdrv_check_byte_request(bs, offset, nb_sectors * 512);  /* TODO change to return DeviceState * when all users are qdevified */
   void *bdrv_get_attached_dev(BlockDriverState *bs)
   {
       return bs->dev;
 }  }
   
 /* return < 0 if error. See bdrv_write() for the return codes */  void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
 int bdrv_read(BlockDriverState *bs, int64_t sector_num,                        void *opaque)
               uint8_t *buf, int nb_sectors)  
 {  {
     BlockDriver *drv = bs->drv;      bs->dev_ops = ops;
       bs->dev_opaque = opaque;
       if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) {
           bs_snapshots = NULL;
       }
   }
   
     if (!drv)  void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
         return -ENOMEDIUM;                                 BlockQMPEventAction action, int is_read)
     if (bdrv_check_request(bs, sector_num, nb_sectors))  {
         return -EIO;      QObject *data;
       const char *action_str;
   
     if (drv->bdrv_pread) {      switch (action) {
         int ret, len;      case BDRV_ACTION_REPORT:
         len = nb_sectors * 512;          action_str = "report";
         ret = drv->bdrv_pread(bs, sector_num * 512, buf, len);          break;
         if (ret < 0)      case BDRV_ACTION_IGNORE:
             return ret;          action_str = "ignore";
         else if (ret != len)          break;
             return -EINVAL;      case BDRV_ACTION_STOP:
         else {          action_str = "stop";
             bs->rd_bytes += (unsigned) len;          break;
             bs->rd_ops ++;      default:
             return 0;          abort();
         }  
     } else {  
         return drv->bdrv_read(bs, sector_num, buf, nb_sectors);  
     }      }
   
       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);
 }  }
   
 /* Return < 0 if error. Important errors are:  static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
   -EIO         generic I/O error (may happen for all errors)  
   -ENOMEDIUM   No media inserted.  
   -EINVAL      Invalid sector number or nb_sectors  
   -EACCES      Trying to write a read-only device  
 */  
 int bdrv_write(BlockDriverState *bs, int64_t sector_num,  
                const uint8_t *buf, int nb_sectors)  
 {  {
     BlockDriver *drv = bs->drv;      QObject *data;
     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) {      data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
         int ret, len, count = 0;                                bdrv_get_device_name(bs), ejected);
         len = nb_sectors * 512;      monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
         do {  
             ret = drv->bdrv_pwrite(bs, sector_num * 512, buf, len - count);      qobject_decref(data);
             if (ret < 0) {  }
                 printf("bdrv_write ret=%d\n", ret);  
                 return ret;  static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
             }  {
             count += ret;      if (bs->dev_ops && bs->dev_ops->change_media_cb) {
             buf += ret;          bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
         } while (count != len);          bs->dev_ops->change_media_cb(bs->dev_opaque, load);
         bs->wr_bytes += (unsigned) len;          if (tray_was_closed) {
         bs->wr_ops ++;              /* tray open */
         return 0;              bdrv_emit_qmp_eject_event(bs, true);
           }
           if (load) {
               /* tray close */
               bdrv_emit_qmp_eject_event(bs, false);
           }
     }      }
     return drv->bdrv_write(bs, sector_num, buf, nb_sectors);  
 }  }
   
 static int bdrv_pread_em(BlockDriverState *bs, int64_t offset,  bool bdrv_dev_has_removable_media(BlockDriverState *bs)
                          uint8_t *buf, int count1)  
 {  {
     uint8_t tmp_buf[SECTOR_SIZE];      return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
     int len, nb_sectors, count;  }
     int64_t sector_num;  
   
     count = count1;  void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
     /* first read to align to sector start */  {
     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);      if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
     if (len > count)          bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
         len = count;  
     sector_num = offset >> SECTOR_BITS;  
     if (len > 0) {  
         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)  
             return -EIO;  
         memcpy(buf, tmp_buf + (offset & (SECTOR_SIZE - 1)), len);  
         count -= len;  
         if (count == 0)  
             return count1;  
         sector_num++;  
         buf += len;  
     }      }
   }
   
     /* read the sectors "in place" */  bool bdrv_dev_is_tray_open(BlockDriverState *bs)
     nb_sectors = count >> SECTOR_BITS;  {
     if (nb_sectors > 0) {      if (bs->dev_ops && bs->dev_ops->is_tray_open) {
         if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)          return bs->dev_ops->is_tray_open(bs->dev_opaque);
             return -EIO;  
         sector_num += nb_sectors;  
         len = nb_sectors << SECTOR_BITS;  
         buf += len;  
         count -= len;  
     }      }
       return false;
   }
   
     /* add data from the last sector */  static void bdrv_dev_resize_cb(BlockDriverState *bs)
     if (count > 0) {  {
         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)      if (bs->dev_ops && bs->dev_ops->resize_cb) {
             return -EIO;          bs->dev_ops->resize_cb(bs->dev_opaque);
         memcpy(buf, tmp_buf, count);  
     }      }
     return count1;  
 }  }
   
 static int bdrv_pwrite_em(BlockDriverState *bs, int64_t offset,  bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
                           const uint8_t *buf, int count1)  
 {  {
     uint8_t tmp_buf[SECTOR_SIZE];      if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
     int len, nb_sectors, count;          return bs->dev_ops->is_medium_locked(bs->dev_opaque);
     int64_t sector_num;      }
       return false;
   }
   
     count = count1;  /*
     /* first write to align to sector start */   * Run consistency checks on an image
     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);   *
     if (len > count)   * Returns 0 if the check could be completed (it doesn't mean that the image is
         len = count;   * free of errors) or -errno when an internal error occurred. The results of the
     sector_num = offset >> SECTOR_BITS;   * 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;
       }
   
       if (bs->backing_hd->keep_read_only) {
           return -EACCES;
       }
   
       if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) {
           return -EBUSY;
       }
   
       backing_drv = bs->backing_hd->drv;
       ro = bs->backing_hd->read_only;
       strncpy(filename, bs->backing_hd->filename, sizeof(filename));
       open_flags =  bs->backing_hd->open_flags;
   
       if (ro) {
           /* 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;
       }
   
       total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
       buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
   
       for (sector = 0; sector < total_sectors; sector += n) {
           if (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;
   }
   
   int bdrv_commit_all(void)
   {
       BlockDriverState *bs;
   
       QTAILQ_FOREACH(bs, &bdrv_states, list) {
           int ret = bdrv_commit(bs);
           if (ret < 0) {
               return ret;
           }
       }
       return 0;
   }
   
   struct BdrvTrackedRequest {
       BlockDriverState *bs;
       int64_t sector_num;
       int nb_sectors;
       bool is_write;
       QLIST_ENTRY(BdrvTrackedRequest) list;
       Coroutine *co; /* owner, used for deadlock detection */
       CoQueue wait_queue; /* coroutines blocked on this request */
   };
   
   /**
    * Remove an active request from the tracked requests list
    *
    * This function should be called when a tracked request is completing.
    */
   static void tracked_request_end(BdrvTrackedRequest *req)
   {
       QLIST_REMOVE(req, list);
       qemu_co_queue_restart_all(&req->wait_queue);
   }
   
   /**
    * Add an active request to the tracked requests list
    */
   static void tracked_request_begin(BdrvTrackedRequest *req,
                                     BlockDriverState *bs,
                                     int64_t sector_num,
                                     int nb_sectors, bool is_write)
   {
       *req = (BdrvTrackedRequest){
           .bs = bs,
           .sector_num = sector_num,
           .nb_sectors = nb_sectors,
           .is_write = is_write,
           .co = qemu_coroutine_self(),
       };
   
       qemu_co_queue_init(&req->wait_queue);
   
       QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
   }
   
   /**
    * Round a region to cluster boundaries
    */
   static void round_to_clusters(BlockDriverState *bs,
                                 int64_t sector_num, int nb_sectors,
                                 int64_t *cluster_sector_num,
                                 int *cluster_nb_sectors)
   {
       BlockDriverInfo bdi;
   
       if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
           *cluster_sector_num = sector_num;
           *cluster_nb_sectors = nb_sectors;
       } else {
           int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
           *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
           *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
                                               nb_sectors, c);
       }
   }
   
   static bool tracked_request_overlaps(BdrvTrackedRequest *req,
                                        int64_t sector_num, int nb_sectors) {
       /*        aaaa   bbbb */
       if (sector_num >= req->sector_num + req->nb_sectors) {
           return false;
       }
       /* bbbb   aaaa        */
       if (req->sector_num >= sector_num + nb_sectors) {
           return false;
       }
       return true;
   }
   
   static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs,
           int64_t sector_num, int nb_sectors)
   {
       BdrvTrackedRequest *req;
       int64_t cluster_sector_num;
       int cluster_nb_sectors;
       bool retry;
   
       /* If we touch the same cluster it counts as an overlap.  This guarantees
        * that allocating writes will be serialized and not race with each other
        * for the same cluster.  For example, in copy-on-read it ensures that the
        * CoR read and write operations are atomic and guest writes cannot
        * interleave between them.
        */
       round_to_clusters(bs, sector_num, nb_sectors,
                         &cluster_sector_num, &cluster_nb_sectors);
   
       do {
           retry = false;
           QLIST_FOREACH(req, &bs->tracked_requests, list) {
               if (tracked_request_overlaps(req, cluster_sector_num,
                                            cluster_nb_sectors)) {
                   /* Hitting this means there was a reentrant request, for
                    * example, a block driver issuing nested requests.  This must
                    * never happen since it means deadlock.
                    */
                   assert(qemu_coroutine_self() != req->co);
   
                   qemu_co_queue_wait(&req->wait_queue);
                   retry = true;
                   break;
               }
           }
       } while (retry);
   }
   
   /*
    * 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;
       int ret;
   
       /* Backing file format doesn't make sense without a backing file */
       if (backing_fmt && !backing_file) {
           return -EINVAL;
       }
   
       if (drv->bdrv_change_backing_file != NULL) {
           ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
       } else {
           ret = -ENOTSUP;
       }
   
       if (ret == 0) {
           pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
           pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
       }
       return ret;
   }
   
   static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
                                      size_t size)
   {
       int64_t len;
   
       if (!bdrv_is_inserted(bs))
           return -ENOMEDIUM;
   
       if (bs->growable)
           return 0;
   
       len = bdrv_getlength(bs);
   
       if (offset < 0)
           return -EIO;
   
       if ((offset > len) || (len - offset < size))
           return -EIO;
   
       return 0;
   }
   
   static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
                                 int nb_sectors)
   {
       return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
                                      nb_sectors * BDRV_SECTOR_SIZE);
   }
   
   typedef struct RwCo {
       BlockDriverState *bs;
       int64_t sector_num;
       int nb_sectors;
       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, 0);
       } else {
           rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num,
                                         rwco->nb_sectors, rwco->qiov, 0);
       }
   }
   
   /*
    * 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);
   
       /**
        * In sync call context, when the vcpu is blocked, this throttling timer
        * will not fire; so the I/O throttling function has to be disabled here
        * if it has been enabled.
        */
       if (bs->io_limits_enabled) {
           fprintf(stderr, "Disabling I/O throttling on '%s' due "
                           "to synchronous I/O.\n", bdrv_get_device_name(bs));
           bdrv_io_limits_disable(bs);
       }
   
       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)
   {
       return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false);
   }
   
   #define BITS_PER_LONG  (sizeof(unsigned long) * 8)
   
   static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
                                int nb_sectors, int dirty)
   {
       int64_t start, end;
       unsigned long val, idx, bit;
   
       start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
       end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
   
       for (; start <= end; start++) {
           idx = start / BITS_PER_LONG;
           bit = start % BITS_PER_LONG;
           val = bs->dirty_bitmap[idx];
           if (dirty) {
               if (!(val & (1UL << bit))) {
                   bs->dirty_count++;
                   val |= 1UL << bit;
               }
           } else {
               if (val & (1UL << bit)) {
                   bs->dirty_count--;
                   val &= ~(1UL << bit);
               }
           }
           bs->dirty_bitmap[idx] = val;
       }
   }
   
   /* Return < 0 if error. Important errors are:
     -EIO         generic I/O error (may happen for all errors)
     -ENOMEDIUM   No media inserted.
     -EINVAL      Invalid sector number or nb_sectors
     -EACCES      Trying to write a read-only device
   */
   int bdrv_write(BlockDriverState *bs, int64_t sector_num,
                  const uint8_t *buf, int nb_sectors)
   {
       return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true);
   }
   
   int bdrv_pread(BlockDriverState *bs, int64_t offset,
                  void *buf, int count1)
   {
       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 = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
       if (len > count)
           len = count;
       sector_num = offset >> BDRV_SECTOR_BITS;
     if (len > 0) {      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(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)          count -= len;
             return -EIO;          if (count == 0)
               return count1;
           sector_num++;
           buf += len;
       }
   
       /* read the sectors "in place" */
       nb_sectors = count >> BDRV_SECTOR_BITS;
       if (nb_sectors > 0) {
           if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
               return ret;
           sector_num += nb_sectors;
           len = nb_sectors << BDRV_SECTOR_BITS;
           buf += len;
           count -= len;
       }
   
       /* add data from the last sector */
       if (count > 0) {
           if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
               return ret;
           memcpy(buf, tmp_buf, count);
       }
       return count1;
   }
   
   int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
                   const void *buf, int count1)
   {
       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 = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
       if (len > count)
           len = count;
       sector_num = offset >> BDRV_SECTOR_BITS;
       if (len > 0) {
           if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
               return ret;
           memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
           if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
               return ret;
         count -= len;          count -= len;
         if (count == 0)          if (count == 0)
             return count1;              return count1;
Line 686  static int bdrv_pwrite_em(BlockDriverSta Line 1721  static int bdrv_pwrite_em(BlockDriverSta
     }      }
   
     /* write the sectors "in place" */      /* write the sectors "in place" */
     nb_sectors = count >> SECTOR_BITS;      nb_sectors = count >> BDRV_SECTOR_BITS;
     if (nb_sectors > 0) {      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;          sector_num += nb_sectors;
         len = nb_sectors << SECTOR_BITS;          len = nb_sectors << BDRV_SECTOR_BITS;
         buf += len;          buf += len;
         count -= len;          count -= len;
     }      }
   
     /* add data from the last sector */      /* add data from the last sector */
     if (count > 0) {      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);          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;      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;
   }
   
   static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
           int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
 {  {
       /* Perform I/O through a temporary buffer so that users who scribble over
        * their read buffer while the operation is in progress do not end up
        * modifying the image file.  This is critical for zero-copy guest I/O
        * where anything might happen inside guest memory.
        */
       void *bounce_buffer;
   
     BlockDriver *drv = bs->drv;      BlockDriver *drv = bs->drv;
       struct iovec iov;
       QEMUIOVector bounce_qiov;
       int64_t cluster_sector_num;
       int cluster_nb_sectors;
       size_t skip_bytes;
       int ret;
   
     if (!drv)      /* Cover entire cluster so no additional backing file I/O is required when
         return -ENOMEDIUM;       * allocating cluster in the image file.
     if (bdrv_check_byte_request(bs, offset, count1))       */
         return -EIO;      round_to_clusters(bs, sector_num, nb_sectors,
                         &cluster_sector_num, &cluster_nb_sectors);
   
       trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
                                      cluster_sector_num, cluster_nb_sectors);
   
       iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
       iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
       qemu_iovec_init_external(&bounce_qiov, &iov, 1);
   
       ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
                                &bounce_qiov);
       if (ret < 0) {
           goto err;
       }
   
       if (drv->bdrv_co_write_zeroes &&
           buffer_is_zero(bounce_buffer, iov.iov_len)) {
           ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
                                         cluster_nb_sectors);
       } else {
           ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
                                     &bounce_qiov);
       }
   
       if (ret < 0) {
           /* It might be okay to ignore write errors for guest requests.  If this
            * is a deliberate copy-on-read then we don't want to ignore the error.
            * Simply report it in all cases.
            */
           goto err;
       }
   
       skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
       qemu_iovec_from_buffer(qiov, bounce_buffer + skip_bytes,
                              nb_sectors * BDRV_SECTOR_SIZE);
   
   err:
       qemu_vfree(bounce_buffer);
       return ret;
   }
   
   /*
    * 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,
       BdrvRequestFlags flags)
   {
       BlockDriver *drv = bs->drv;
       BdrvTrackedRequest req;
       int ret;
   
       if (!drv) {
           return -ENOMEDIUM;
       }
       if (bdrv_check_request(bs, sector_num, nb_sectors)) {
           return -EIO;
       }
   
       /* throttling disk read I/O */
       if (bs->io_limits_enabled) {
           bdrv_io_limits_intercept(bs, false, nb_sectors);
       }
   
       if (bs->copy_on_read) {
           flags |= BDRV_REQ_COPY_ON_READ;
       }
       if (flags & BDRV_REQ_COPY_ON_READ) {
           bs->copy_on_read_in_flight++;
       }
   
       if (bs->copy_on_read_in_flight) {
           wait_for_overlapping_requests(bs, sector_num, nb_sectors);
       }
   
       tracked_request_begin(&req, bs, sector_num, nb_sectors, false);
   
       if (flags & BDRV_REQ_COPY_ON_READ) {
           int pnum;
   
           ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum);
           if (ret < 0) {
               goto out;
           }
   
           if (!ret || pnum != nb_sectors) {
               ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
               goto out;
           }
       }
   
       ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
   
   out:
       tracked_request_end(&req);
   
       if (flags & BDRV_REQ_COPY_ON_READ) {
           bs->copy_on_read_in_flight--;
       }
   
       return ret;
   }
   
   int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
       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, 0);
   }
   
   int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
       int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
   {
       trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
   
       return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
                               BDRV_REQ_COPY_ON_READ);
   }
   
   static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
       int64_t sector_num, int nb_sectors)
   {
       BlockDriver *drv = bs->drv;
       QEMUIOVector qiov;
       struct iovec iov;
       int ret;
   
       /* TODO Emulate only part of misaligned requests instead of letting block
        * drivers return -ENOTSUP and emulate everything */
   
       /* First try the efficient write zeroes operation */
       if (drv->bdrv_co_write_zeroes) {
           ret = drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
           if (ret != -ENOTSUP) {
               return ret;
           }
       }
   
       /* Fall back to bounce buffer if write zeroes is unsupported */
       iov.iov_len  = nb_sectors * BDRV_SECTOR_SIZE;
       iov.iov_base = qemu_blockalign(bs, iov.iov_len);
       memset(iov.iov_base, 0, iov.iov_len);
       qemu_iovec_init_external(&qiov, &iov, 1);
   
       ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, &qiov);
   
       qemu_vfree(iov.iov_base);
       return ret;
   }
   
   /*
    * Handle a write request in coroutine context
    */
   static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
       int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
       BdrvRequestFlags flags)
   {
       BlockDriver *drv = bs->drv;
       BdrvTrackedRequest req;
       int ret;
   
       if (!bs->drv) {
           return -ENOMEDIUM;
       }
       if (bs->read_only) {
           return -EACCES;
       }
       if (bdrv_check_request(bs, sector_num, nb_sectors)) {
           return -EIO;
       }
   
       /* throttling disk write I/O */
       if (bs->io_limits_enabled) {
           bdrv_io_limits_intercept(bs, true, nb_sectors);
       }
   
       if (bs->copy_on_read_in_flight) {
           wait_for_overlapping_requests(bs, sector_num, nb_sectors);
       }
   
       tracked_request_begin(&req, bs, sector_num, nb_sectors, true);
   
       if (flags & BDRV_REQ_ZERO_WRITE) {
           ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors);
       } else {
           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;
       }
   
       tracked_request_end(&req);
   
       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);
   
       return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
   }
   
   int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
                                         int64_t sector_num, int nb_sectors)
   {
       trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors);
   
       return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
                                BDRV_REQ_ZERO_WRITE);
   }
   
   /**
    * Truncate file to 'offset' bytes (needed only for file protocols)
    */
   int bdrv_truncate(BlockDriverState *bs, int64_t offset)
   {
       BlockDriver *drv = bs->drv;
       int ret;
       if (!drv)
           return -ENOMEDIUM;
       if (!drv->bdrv_truncate)
           return -ENOTSUP;
       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;
   }
   
   /**
    * Length of a file in bytes. Return < 0 if error or unknown.
    */
   int64_t bdrv_getlength(BlockDriverState *bs)
   {
       BlockDriver *drv = bs->drv;
       if (!drv)
           return -ENOMEDIUM;
   
       if (bs->growable || bdrv_dev_has_removable_media(bs)) {
           if (drv->bdrv_getlength) {
               return drv->bdrv_getlength(bs);
           }
       }
       return bs->total_sectors * BDRV_SECTOR_SIZE;
   }
   
   /* return 0 as number of sectors if no device present or error */
   void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
   {
       int64_t length;
       length = bdrv_getlength(bs);
       if (length < 0)
           length = 0;
       else
           length = length >> BDRV_SECTOR_BITS;
       *nb_sectors_ptr = length;
   }
   
   struct partition {
           uint8_t boot_ind;           /* 0x80 - active */
           uint8_t head;               /* starting head */
           uint8_t sector;             /* starting sector */
           uint8_t cyl;                /* starting cylinder */
           uint8_t sys_ind;            /* What partition type */
           uint8_t end_head;           /* end head */
           uint8_t end_sector;         /* end sector */
           uint8_t end_cyl;            /* end cylinder */
           uint32_t start_sect;        /* starting sector counting from 0 */
           uint32_t nr_sects;          /* nr of sectors in partition */
   } 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[BDRV_SECTOR_SIZE];
       int ret, i, heads, sectors, cylinders;
       struct partition *p;
       uint32_t nr_sects;
       uint64_t nb_sectors;
       bool enabled;
   
       bdrv_get_geometry(bs, &nb_sectors);
   
       /**
        * The function will be invoked during startup not only in sync I/O mode,
        * but also in async I/O mode. So the I/O throttling function has to
        * be disabled temporarily here, not permanently.
        */
       enabled = bs->io_limits_enabled;
       bs->io_limits_enabled = false;
       ret = bdrv_read(bs, 0, buf, 1);
       bs->io_limits_enabled = enabled;
       if (ret < 0)
           return -1;
       /* test msdos magic */
       if (buf[510] != 0x55 || buf[511] != 0xaa)
           return -1;
       for(i = 0; i < 4; i++) {
           p = ((struct partition *)(buf + 0x1be)) + i;
           nr_sects = le32_to_cpu(p->nr_sects);
           if (nr_sects && p->end_head) {
               /* We make the assumption that the partition terminates on
                  a cylinder boundary */
               heads = p->end_head + 1;
               sectors = p->end_sector & 63;
               if (sectors == 0)
                   continue;
               cylinders = nb_sectors / (heads * sectors);
               if (cylinders < 1 || cylinders > 16383)
                   continue;
               *pheads = heads;
               *psectors = sectors;
               *pcylinders = cylinders;
   #if 0
               printf("guessed geometry: LCHS=%d %d %d\n",
                      cylinders, heads, sectors);
   #endif
               return 0;
           }
       }
       return -1;
   }
   
   void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
   {
       int translation, lba_detected = 0;
       int cylinders, heads, secs;
       uint64_t nb_sectors;
   
       /* if a geometry hint is available, use it */
       bdrv_get_geometry(bs, &nb_sectors);
       bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
       translation = bdrv_get_translation_hint(bs);
       if (cylinders != 0) {
           *pcyls = cylinders;
           *pheads = heads;
           *psecs = secs;
       } else {
           if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
               if (heads > 16) {
                   /* if heads > 16, it means that a BIOS LBA
                      translation was active, so the default
                      hardware geometry is OK */
                   lba_detected = 1;
                   goto default_geometry;
               } else {
                   *pcyls = cylinders;
                   *pheads = heads;
                   *psecs = secs;
                   /* disable any translation to be in sync with
                      the logical geometry */
                   if (translation == BIOS_ATA_TRANSLATION_AUTO) {
                       bdrv_set_translation_hint(bs,
                                                 BIOS_ATA_TRANSLATION_NONE);
                   }
               }
           } else {
           default_geometry:
               /* if no geometry, use a standard physical disk geometry */
               cylinders = nb_sectors / (16 * 63);
   
               if (cylinders > 16383)
                   cylinders = 16383;
               else if (cylinders < 2)
                   cylinders = 2;
               *pcyls = cylinders;
               *pheads = 16;
               *psecs = 63;
               if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
                   if ((*pcyls * *pheads) <= 131072) {
                       bdrv_set_translation_hint(bs,
                                                 BIOS_ATA_TRANSLATION_LARGE);
                   } else {
                       bdrv_set_translation_hint(bs,
                                                 BIOS_ATA_TRANSLATION_LBA);
                   }
               }
           }
           bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
       }
   }
   
   void bdrv_set_geometry_hint(BlockDriverState *bs,
                               int cyls, int heads, int secs)
   {
       bs->cyls = cyls;
       bs->heads = heads;
       bs->secs = secs;
   }
   
   void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
   {
       bs->translation = translation;
   }
   
   void bdrv_get_geometry_hint(BlockDriverState *bs,
                               int *pcyls, int *pheads, int *psecs)
   {
       *pcyls = bs->cyls;
       *pheads = bs->heads;
       *psecs = bs->secs;
   }
   
   /* throttling disk io limits */
   void bdrv_set_io_limits(BlockDriverState *bs,
                           BlockIOLimit *io_limits)
   {
       bs->io_limits = *io_limits;
       bs->io_limits_enabled = bdrv_io_limits_enabled(bs);
   }
   
   /* Recognize floppy formats */
   typedef struct FDFormat {
       FDriveType drive;
       uint8_t last_sect;
       uint8_t max_track;
       uint8_t max_head;
       FDriveRate rate;
   } 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_RATE_500K, },
       { FDRIVE_DRV_144, 20, 80, 1, FDRIVE_RATE_500K, },
       { FDRIVE_DRV_144, 21, 80, 1, FDRIVE_RATE_500K, },
       { FDRIVE_DRV_144, 21, 82, 1, FDRIVE_RATE_500K, },
       { FDRIVE_DRV_144, 21, 83, 1, FDRIVE_RATE_500K, },
       { FDRIVE_DRV_144, 22, 80, 1, FDRIVE_RATE_500K, },
       { FDRIVE_DRV_144, 23, 80, 1, FDRIVE_RATE_500K, },
       { FDRIVE_DRV_144, 24, 80, 1, FDRIVE_RATE_500K, },
       /* 2.88 MB 3"1/2 floppy disks */
       { FDRIVE_DRV_288, 36, 80, 1, FDRIVE_RATE_1M, },
       { FDRIVE_DRV_288, 39, 80, 1, FDRIVE_RATE_1M, },
       { FDRIVE_DRV_288, 40, 80, 1, FDRIVE_RATE_1M, },
       { FDRIVE_DRV_288, 44, 80, 1, FDRIVE_RATE_1M, },
       { FDRIVE_DRV_288, 48, 80, 1, FDRIVE_RATE_1M, },
       /* 720 kB 3"1/2 floppy disks */
       { FDRIVE_DRV_144,  9, 80, 1, FDRIVE_RATE_250K, },
       { FDRIVE_DRV_144, 10, 80, 1, FDRIVE_RATE_250K, },
       { FDRIVE_DRV_144, 10, 82, 1, FDRIVE_RATE_250K, },
       { FDRIVE_DRV_144, 10, 83, 1, FDRIVE_RATE_250K, },
       { FDRIVE_DRV_144, 13, 80, 1, FDRIVE_RATE_250K, },
       { FDRIVE_DRV_144, 14, 80, 1, FDRIVE_RATE_250K, },
       /* 1.2 MB 5"1/4 floppy disks */
       { FDRIVE_DRV_120, 15, 80, 1, FDRIVE_RATE_500K, },
       { FDRIVE_DRV_120, 18, 80, 1, FDRIVE_RATE_500K, },
       { FDRIVE_DRV_120, 18, 82, 1, FDRIVE_RATE_500K, },
       { FDRIVE_DRV_120, 18, 83, 1, FDRIVE_RATE_500K, },
       { FDRIVE_DRV_120, 20, 80, 1, FDRIVE_RATE_500K, },
       /* 720 kB 5"1/4 floppy disks */
       { FDRIVE_DRV_120,  9, 80, 1, FDRIVE_RATE_250K, },
       { FDRIVE_DRV_120, 11, 80, 1, FDRIVE_RATE_250K, },
       /* 360 kB 5"1/4 floppy disks */
       { FDRIVE_DRV_120,  9, 40, 1, FDRIVE_RATE_300K, },
       { FDRIVE_DRV_120,  9, 40, 0, FDRIVE_RATE_300K, },
       { FDRIVE_DRV_120, 10, 41, 1, FDRIVE_RATE_300K, },
       { FDRIVE_DRV_120, 10, 42, 1, FDRIVE_RATE_300K, },
       /* 320 kB 5"1/4 floppy disks */
       { FDRIVE_DRV_120,  8, 40, 1, FDRIVE_RATE_250K, },
       { FDRIVE_DRV_120,  8, 40, 0, FDRIVE_RATE_250K, },
       /* 360 kB must match 5"1/4 better than 3"1/2... */
       { FDRIVE_DRV_144,  9, 80, 0, FDRIVE_RATE_250K, },
       /* end */
       { FDRIVE_DRV_NONE, -1, -1, 0, 0, },
   };
   
   void bdrv_get_floppy_geometry_hint(BlockDriverState *bs, int *nb_heads,
                                      int *max_track, int *last_sect,
                                      FDriveType drive_in, FDriveType *drive,
                                      FDriveRate *rate)
   {
       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 */
           *rate = FDRIVE_RATE_500K;
       } 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;
           *rate = parse->rate;
       }
   }
   
   int bdrv_get_translation_hint(BlockDriverState *bs)
   {
       return bs->translation;
   }
   
   void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
                          BlockErrorAction on_write_error)
   {
       bs->on_read_error = on_read_error;
       bs->on_write_error = on_write_error;
   }
   
   BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
   {
       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)
   {
       return bs->sg;
   }
   
   int bdrv_enable_write_cache(BlockDriverState *bs)
   {
       return bs->enable_write_cache;
   }
   
   int bdrv_is_encrypted(BlockDriverState *bs)
   {
       if (bs->backing_hd && bs->backing_hd->encrypted)
           return 1;
       return bs->encrypted;
   }
   
   int bdrv_key_required(BlockDriverState *bs)
   {
       BlockDriverState *backing_hd = bs->backing_hd;
   
       if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
           return 1;
       return (bs->encrypted && !bs->valid_key);
   }
   
   int bdrv_set_key(BlockDriverState *bs, const char *key)
   {
       int ret;
       if (bs->backing_hd && bs->backing_hd->encrypted) {
           ret = bdrv_set_key(bs->backing_hd, key);
           if (ret < 0)
               return ret;
           if (!bs->encrypted)
               return 0;
       }
       if (!bs->encrypted) {
           return -EINVAL;
       } else if (!bs->drv || !bs->drv->bdrv_set_key) {
           return -ENOMEDIUM;
       }
       ret = bs->drv->bdrv_set_key(bs, key);
       if (ret < 0) {
           bs->valid_key = 0;
       } else if (!bs->valid_key) {
           bs->valid_key = 1;
           /* call the change callback now, we skipped it on open */
           bdrv_dev_change_media_cb(bs, true);
       }
       return ret;
   }
   
   void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
   {
       if (!bs->drv) {
           buf[0] = '\0';
       } else {
           pstrcpy(buf, buf_size, bs->drv->format_name);
       }
   }
   
   void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
                            void *opaque)
   {
       BlockDriver *drv;
   
       QLIST_FOREACH(drv, &bdrv_drivers, list) {
           it(opaque, drv->format_name);
       }
   }
   
   BlockDriverState *bdrv_find(const char *name)
   {
       BlockDriverState *bs;
   
       QTAILQ_FOREACH(bs, &bdrv_states, list) {
           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;
   
       QTAILQ_FOREACH(bs, &bdrv_states, list) {
           it(opaque, bs);
       }
   }
   
   const char *bdrv_get_device_name(BlockDriverState *bs)
   {
       return bs->device_name;
   }
   
   void bdrv_flush_all(void)
   {
       BlockDriverState *bs;
   
       QTAILQ_FOREACH(bs, &bdrv_states, list) {
           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;
   }
   
   typedef struct BdrvCoIsAllocatedData {
       BlockDriverState *bs;
       int64_t sector_num;
       int nb_sectors;
       int *pnum;
       int ret;
       bool done;
   } BdrvCoIsAllocatedData;
   
   /*
    * Returns true iff the specified sector is present in the disk image. Drivers
    * not implementing the functionality are assumed to not support backing files,
    * hence all their sectors are reported as allocated.
    *
    * If 'sector_num' is beyond the end of the disk image the return value is 0
    * and 'pnum' is set to 0.
    *
    * 'pnum' is set to the number of sectors (including and immediately following
    * the specified sector) that are known to be in the same
    * allocated/unallocated state.
    *
    * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
    * beyond the end of the disk image it will be clamped.
    */
   int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num,
                                         int nb_sectors, int *pnum)
   {
       int64_t n;
   
       if (sector_num >= bs->total_sectors) {
           *pnum = 0;
           return 0;
       }
   
       n = bs->total_sectors - sector_num;
       if (n < nb_sectors) {
           nb_sectors = n;
       }
   
       if (!bs->drv->bdrv_co_is_allocated) {
           *pnum = nb_sectors;
           return 1;
       }
   
       return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum);
   }
   
   /* Coroutine wrapper for bdrv_is_allocated() */
   static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque)
   {
       BdrvCoIsAllocatedData *data = opaque;
       BlockDriverState *bs = data->bs;
   
       data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors,
                                        data->pnum);
       data->done = true;
   }
   
   /*
    * Synchronous wrapper around bdrv_co_is_allocated().
    *
    * See bdrv_co_is_allocated() for details.
    */
   int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
                         int *pnum)
   {
       Coroutine *co;
       BdrvCoIsAllocatedData data = {
           .bs = bs,
           .sector_num = sector_num,
           .nb_sectors = nb_sectors,
           .pnum = pnum,
           .done = false,
       };
   
       co = qemu_coroutine_create(bdrv_is_allocated_co_entry);
       qemu_coroutine_enter(co, &data);
       while (!data.done) {
           qemu_aio_wait();
       }
       return data.ret;
   }
   
   BlockInfoList *qmp_query_block(Error **errp)
   {
       BlockInfoList *head = NULL, *cur_item = NULL;
       BlockDriverState *bs;
   
       QTAILQ_FOREACH(bs, &bdrv_states, list) {
           BlockInfoList *info = g_malloc0(sizeof(*info));
   
           info->value = g_malloc0(sizeof(*info->value));
           info->value->device = g_strdup(bs->device_name);
           info->value->type = g_strdup("unknown");
           info->value->locked = bdrv_dev_is_medium_locked(bs);
           info->value->removable = bdrv_dev_has_removable_media(bs);
   
           if (bdrv_dev_has_removable_media(bs)) {
               info->value->has_tray_open = true;
               info->value->tray_open = bdrv_dev_is_tray_open(bs);
           }
   
           if (bdrv_iostatus_is_enabled(bs)) {
               info->value->has_io_status = true;
               info->value->io_status = bs->iostatus;
           }
   
           if (bs->drv) {
               info->value->has_inserted = true;
               info->value->inserted = g_malloc0(sizeof(*info->value->inserted));
               info->value->inserted->file = g_strdup(bs->filename);
               info->value->inserted->ro = bs->read_only;
               info->value->inserted->drv = g_strdup(bs->drv->format_name);
               info->value->inserted->encrypted = bs->encrypted;
               if (bs->backing_file[0]) {
                   info->value->inserted->has_backing_file = true;
                   info->value->inserted->backing_file = g_strdup(bs->backing_file);
               }
   
               if (bs->io_limits_enabled) {
                   info->value->inserted->bps =
                                  bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
                   info->value->inserted->bps_rd =
                                  bs->io_limits.bps[BLOCK_IO_LIMIT_READ];
                   info->value->inserted->bps_wr =
                                  bs->io_limits.bps[BLOCK_IO_LIMIT_WRITE];
                   info->value->inserted->iops =
                                  bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
                   info->value->inserted->iops_rd =
                                  bs->io_limits.iops[BLOCK_IO_LIMIT_READ];
                   info->value->inserted->iops_wr =
                                  bs->io_limits.iops[BLOCK_IO_LIMIT_WRITE];
               }
           }
   
           /* XXX: waiting for the qapi to support GSList */
           if (!cur_item) {
               head = cur_item = info;
           } else {
               cur_item->next = info;
               cur_item = info;
           }
       }
   
       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;
   }
   
   BlockStatsList *qmp_query_blockstats(Error **errp)
   {
       BlockStatsList *head = NULL, *cur_item = NULL;
       BlockDriverState *bs;
   
       QTAILQ_FOREACH(bs, &bdrv_states, list) {
           BlockStatsList *info = g_malloc0(sizeof(*info));
           info->value = qmp_query_blockstat(bs, NULL);
   
           /* XXX: waiting for the qapi to support GSList */
           if (!cur_item) {
               head = cur_item = info;
           } else {
               cur_item->next = info;
               cur_item = info;
           }
       }
   
       return head;
   }
   
   const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
   {
       if (bs->backing_hd && bs->backing_hd->encrypted)
           return bs->backing_file;
       else if (bs->encrypted)
           return bs->filename;
       else
           return NULL;
   }
   
   void bdrv_get_backing_filename(BlockDriverState *bs,
                                  char *filename, int filename_size)
   {
       pstrcpy(filename, filename_size, bs->backing_file);
   }
   
   int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
                             const uint8_t *buf, int nb_sectors)
   {
       BlockDriver *drv = bs->drv;
       if (!drv)
           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);
   }
   
   int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
   {
       BlockDriver *drv = bs->drv;
       if (!drv)
           return -ENOMEDIUM;
       if (!drv->bdrv_get_info)
           return -ENOTSUP;
       memset(bdi, 0, sizeof(*bdi));
       return drv->bdrv_get_info(bs, bdi);
   }
   
   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_save_vmstate)
           return drv->bdrv_save_vmstate(bs, buf, pos, size);
       if (bs->file)
           return bdrv_save_vmstate(bs->file, buf, pos, size);
       return -ENOTSUP;
   }
   
   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_load_vmstate)
           return drv->bdrv_load_vmstate(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)
           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)
           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)
   {
       BlockDriver *drv = bs->drv;
       if (!drv)
           return -ENOMEDIUM;
       if (drv->bdrv_snapshot_delete)
           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,
                          QEMUSnapshotInfo **psn_info)
   {
       BlockDriver *drv = bs->drv;
       if (!drv)
           return -ENOMEDIUM;
       if (drv->bdrv_snapshot_list)
           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;
   }
   
   BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
           const char *backing_file)
   {
       if (!bs->drv) {
           return NULL;
       }
   
       if (bs->backing_hd) {
           if (strcmp(bs->backing_file, backing_file) == 0) {
               return bs->backing_hd;
           } else {
               return bdrv_find_backing_image(bs->backing_hd, backing_file);
           }
       }
   
       return NULL;
   }
   
   #define NB_SUFFIXES 4
   
   char *get_human_readable_size(char *buf, int buf_size, int64_t size)
   {
       static const char suffixes[NB_SUFFIXES] = "KMGT";
       int64_t base;
       int i;
   
       if (size <= 999) {
           snprintf(buf, buf_size, "%" PRId64, size);
       } else {
           base = 1024;
           for(i = 0; i < NB_SUFFIXES; i++) {
               if (size < (10 * base)) {
                   snprintf(buf, buf_size, "%0.1f%c",
                            (double)size / base,
                            suffixes[i]);
                   break;
               } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
                   snprintf(buf, buf_size, "%" PRId64 "%c",
                            ((size + (base >> 1)) / base),
                            suffixes[i]);
                   break;
               }
               base = base * 1024;
           }
       }
       return buf;
   }
   
   char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
   {
       char buf1[128], date_buf[128], clock_buf[128];
   #ifdef _WIN32
       struct tm *ptm;
   #else
       struct tm tm;
   #endif
       time_t ti;
       int64_t secs;
   
       if (!sn) {
           snprintf(buf, buf_size,
                    "%-10s%-20s%7s%20s%15s",
                    "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
       } else {
           ti = sn->date_sec;
   #ifdef _WIN32
           ptm = localtime(&ti);
           strftime(date_buf, sizeof(date_buf),
                    "%Y-%m-%d %H:%M:%S", ptm);
   #else
           localtime_r(&ti, &tm);
           strftime(date_buf, sizeof(date_buf),
                    "%Y-%m-%d %H:%M:%S", &tm);
   #endif
           secs = sn->vm_clock_nsec / 1000000000;
           snprintf(clock_buf, sizeof(clock_buf),
                    "%02d:%02d:%02d.%03d",
                    (int)(secs / 3600),
                    (int)((secs / 60) % 60),
                    (int)(secs % 60),
                    (int)((sn->vm_clock_nsec / 1000000) % 1000));
           snprintf(buf, buf_size,
                    "%-10s%-20s%7s%20s%15s",
                    sn->id_str, sn->name,
                    get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
                    date_buf,
                    clock_buf);
       }
       return buf;
   }
   
   /**************************************************************/
   /* async I/Os */
   
   BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
                                    QEMUIOVector *qiov, int nb_sectors,
                                    BlockDriverCompletionFunc *cb, void *opaque)
   {
       trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
   
       return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
                                    cb, opaque, false);
   }
   
   BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
                                     QEMUIOVector *qiov, int nb_sectors,
                                     BlockDriverCompletionFunc *cb, void *opaque)
   {
       trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
   
       return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors,
                                    cb, opaque, true);
   }
   
   
   typedef struct MultiwriteCB {
       int error;
       int num_requests;
       int num_callbacks;
       struct {
           BlockDriverCompletionFunc *cb;
           void *opaque;
           QEMUIOVector *free_qiov;
       } callbacks[];
   } MultiwriteCB;
   
   static void multiwrite_user_cb(MultiwriteCB *mcb)
   {
       int i;
   
       for (i = 0; i < mcb->num_callbacks; i++) {
           mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
           if (mcb->callbacks[i].free_qiov) {
               qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
           }
           g_free(mcb->callbacks[i].free_qiov);
       }
   }
   
   static void multiwrite_cb(void *opaque, int ret)
   {
       MultiwriteCB *mcb = opaque;
   
       trace_multiwrite_cb(mcb, ret);
   
       if (ret < 0 && !mcb->error) {
           mcb->error = ret;
       }
   
       mcb->num_requests--;
       if (mcb->num_requests == 0) {
           multiwrite_user_cb(mcb);
           g_free(mcb);
       }
   }
   
   static int multiwrite_req_compare(const void *a, const void *b)
   {
       const BlockRequest *req1 = a, *req2 = b;
   
       /*
        * 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;
       }
   }
   
   /*
    * Takes a bunch of requests and tries to merge them. Returns the number of
    * requests that remain after merging.
    */
   static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
       int num_reqs, MultiwriteCB *mcb)
   {
       int i, outidx;
   
       // Sort requests by start sector
       qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
   
       // 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;
   
           // Handle exactly sequential writes and overlapping writes.
           if (reqs[i].sector <= oldreq_last) {
               merge = 1;
           }
   
           if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
               merge = 0;
           }
   
     if (!drv->bdrv_pread)          if (merge) {
         return bdrv_pread_em(bs, offset, buf1, count1);              size_t size;
     return drv->bdrv_pread(bs, offset, buf1, count1);              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 should need to add any zeros between the two requests
               assert (reqs[i].sector <= oldreq_last);
   
 /**              // Add the second request
  * Write with byte offsets (needed only for file protocols)              qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
  */  
 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,  
                 const void *buf1, int count1)  
 {  
     BlockDriver *drv = bs->drv;  
   
     if (!drv)              reqs[outidx].nb_sectors = qiov->size >> 9;
         return -ENOMEDIUM;              reqs[outidx].qiov = qiov;
     if (bdrv_check_byte_request(bs, offset, count1))  
         return -EIO;  
   
     if (!drv->bdrv_pwrite)              mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
         return bdrv_pwrite_em(bs, offset, buf1, count1);          } else {
     return drv->bdrv_pwrite(bs, offset, buf1, count1);              outidx++;
 }              reqs[outidx].sector     = reqs[i].sector;
               reqs[outidx].nb_sectors = reqs[i].nb_sectors;
               reqs[outidx].qiov       = reqs[i].qiov;
           }
       }
   
 /**      return outidx + 1;
  * Truncate file to 'offset' bytes (needed only for file protocols)  
  */  
 int bdrv_truncate(BlockDriverState *bs, int64_t offset)  
 {  
     BlockDriver *drv = bs->drv;  
     if (!drv)  
         return -ENOMEDIUM;  
     if (!drv->bdrv_truncate)  
         return -ENOTSUP;  
     return drv->bdrv_truncate(bs, offset);  
 }  }
   
 /**  /*
  * Length of a file in bytes. Return < 0 if error or unknown.   * Submit multiple AIO write requests at once.
    *
    * 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.
  */   */
 int64_t bdrv_getlength(BlockDriverState *bs)  int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
 {  {
     BlockDriver *drv = bs->drv;      MultiwriteCB *mcb;
     if (!drv)      int i;
         return -ENOMEDIUM;  
     if (!drv->bdrv_getlength) {      /* don't submit writes if we don't have a medium */
         /* legacy mode */      if (bs->drv == NULL) {
         return bs->total_sectors * SECTOR_SIZE;          for (i = 0; i < num_reqs; i++) {
               reqs[i].error = -ENOMEDIUM;
           }
           return -1;
       }
   
       if (num_reqs == 0) {
           return 0;
     }      }
     return drv->bdrv_getlength(bs);  
       // Create MultiwriteCB structure
       mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
       mcb->num_requests = 0;
       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. */
       mcb->num_requests = num_reqs;
       for (i = 0; i < num_reqs; i++) {
           bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
               reqs[i].nb_sectors, multiwrite_cb, mcb);
       }
   
       return 0;
 }  }
   
 /* return 0 as number of sectors if no device present or error */  void bdrv_aio_cancel(BlockDriverAIOCB *acb)
 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)  
 {  {
     int64_t length;      acb->pool->cancel(acb);
     length = bdrv_getlength(bs);  
     if (length < 0)  
         length = 0;  
     else  
         length = length >> SECTOR_BITS;  
     *nb_sectors_ptr = length;  
 }  }
   
 struct partition {  /* block I/O throttling */
         uint8_t boot_ind;           /* 0x80 - active */  static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors,
         uint8_t head;               /* starting head */                   bool is_write, double elapsed_time, uint64_t *wait)
         uint8_t sector;             /* starting sector */  
         uint8_t cyl;                /* starting cylinder */  
         uint8_t sys_ind;            /* What partition type */  
         uint8_t end_head;           /* end head */  
         uint8_t end_sector;         /* end sector */  
         uint8_t end_cyl;            /* end cylinder */  
         uint32_t start_sect;        /* starting sector counting from 0 */  
         uint32_t nr_sects;          /* nr of sectors in partition */  
 } __attribute__((packed));  
   
 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */  
 static int guess_disk_lchs(BlockDriverState *bs,  
                            int *pcylinders, int *pheads, int *psectors)  
 {  {
     uint8_t buf[512];      uint64_t bps_limit = 0;
     int ret, i, heads, sectors, cylinders;      double   bytes_limit, bytes_base, bytes_res;
     struct partition *p;      double   slice_time, wait_time;
     uint32_t nr_sects;  
     uint64_t nb_sectors;  
   
     bdrv_get_geometry(bs, &nb_sectors);      if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
           bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL];
       } else if (bs->io_limits.bps[is_write]) {
           bps_limit = bs->io_limits.bps[is_write];
       } else {
           if (wait) {
               *wait = 0;
           }
   
     ret = bdrv_read(bs, 0, buf, 1);          return false;
     if (ret < 0)      }
         return -1;  
     /* test msdos magic */      slice_time = bs->slice_end - bs->slice_start;
     if (buf[510] != 0x55 || buf[511] != 0xaa)      slice_time /= (NANOSECONDS_PER_SECOND);
         return -1;      bytes_limit = bps_limit * slice_time;
     for(i = 0; i < 4; i++) {      bytes_base  = bs->nr_bytes[is_write] - bs->io_base.bytes[is_write];
         p = ((struct partition *)(buf + 0x1be)) + i;      if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) {
         nr_sects = le32_to_cpu(p->nr_sects);          bytes_base += bs->nr_bytes[!is_write] - bs->io_base.bytes[!is_write];
         if (nr_sects && p->end_head) {      }
             /* We make the assumption that the partition terminates on  
                a cylinder boundary */      /* bytes_base: the bytes of data which have been read/written; and
             heads = p->end_head + 1;       *             it is obtained from the history statistic info.
             sectors = p->end_sector & 63;       * bytes_res: the remaining bytes of data which need to be read/written.
             if (sectors == 0)       * (bytes_base + bytes_res) / bps_limit: used to calcuate
                 continue;       *             the total time for completing reading/writting all data.
             cylinders = nb_sectors / (heads * sectors);       */
             if (cylinders < 1 || cylinders > 16383)      bytes_res   = (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
                 continue;  
             *pheads = heads;      if (bytes_base + bytes_res <= bytes_limit) {
             *psectors = sectors;          if (wait) {
             *pcylinders = cylinders;              *wait = 0;
 #if 0  
             printf("guessed geometry: LCHS=%d %d %d\n",  
                    cylinders, heads, sectors);  
 #endif  
             return 0;  
         }          }
   
           return false;
     }      }
     return -1;  
       /* Calc approx time to dispatch */
       wait_time = (bytes_base + bytes_res) / bps_limit - elapsed_time;
   
       /* When the I/O rate at runtime exceeds the limits,
        * bs->slice_end need to be extended in order that the current statistic
        * info can be kept until the timer fire, so it is increased and tuned
        * based on the result of experiment.
        */
       bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
       bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
       if (wait) {
           *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
       }
   
       return true;
 }  }
   
 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)  static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write,
                                double elapsed_time, uint64_t *wait)
 {  {
     int translation, lba_detected = 0;      uint64_t iops_limit = 0;
     int cylinders, heads, secs;      double   ios_limit, ios_base;
     uint64_t nb_sectors;      double   slice_time, wait_time;
   
     /* if a geometry hint is available, use it */      if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
     bdrv_get_geometry(bs, &nb_sectors);          iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL];
     bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);      } else if (bs->io_limits.iops[is_write]) {
     translation = bdrv_get_translation_hint(bs);          iops_limit = bs->io_limits.iops[is_write];
     if (cylinders != 0) {  
         *pcyls = cylinders;  
         *pheads = heads;  
         *psecs = secs;  
     } else {      } else {
         if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {          if (wait) {
             if (heads > 16) {              *wait = 0;
                 /* if heads > 16, it means that a BIOS LBA          }
                    translation was active, so the default  
                    hardware geometry is OK */  
                 lba_detected = 1;  
                 goto default_geometry;  
             } else {  
                 *pcyls = cylinders;  
                 *pheads = heads;  
                 *psecs = secs;  
                 /* disable any translation to be in sync with  
                    the logical geometry */  
                 if (translation == BIOS_ATA_TRANSLATION_AUTO) {  
                     bdrv_set_translation_hint(bs,  
                                               BIOS_ATA_TRANSLATION_NONE);  
                 }  
             }  
         } else {  
         default_geometry:  
             /* if no geometry, use a standard physical disk geometry */  
             cylinders = nb_sectors / (16 * 63);  
   
             if (cylinders > 16383)          return false;
                 cylinders = 16383;      }
             else if (cylinders < 2)  
                 cylinders = 2;      slice_time = bs->slice_end - bs->slice_start;
             *pcyls = cylinders;      slice_time /= (NANOSECONDS_PER_SECOND);
             *pheads = 16;      ios_limit  = iops_limit * slice_time;
             *psecs = 63;      ios_base   = bs->nr_ops[is_write] - bs->io_base.ios[is_write];
             if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {      if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) {
                 if ((*pcyls * *pheads) <= 131072) {          ios_base += bs->nr_ops[!is_write] - bs->io_base.ios[!is_write];
                     bdrv_set_translation_hint(bs,      }
                                               BIOS_ATA_TRANSLATION_LARGE);  
                 } else {      if (ios_base + 1 <= ios_limit) {
                     bdrv_set_translation_hint(bs,          if (wait) {
                                               BIOS_ATA_TRANSLATION_LBA);              *wait = 0;
                 }  
             }  
         }          }
         bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);  
           return false;
     }      }
 }  
   
 void bdrv_set_geometry_hint(BlockDriverState *bs,      /* Calc approx time to dispatch */
                             int cyls, int heads, int secs)      wait_time = (ios_base + 1) / iops_limit;
 {      if (wait_time > elapsed_time) {
     bs->cyls = cyls;          wait_time = wait_time - elapsed_time;
     bs->heads = heads;      } else {
     bs->secs = secs;          wait_time = 0;
 }      }
   
 void bdrv_set_type_hint(BlockDriverState *bs, int type)      bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10;
 {      bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME;
     bs->type = type;      if (wait) {
     bs->removable = ((type == BDRV_TYPE_CDROM ||          *wait = wait_time * BLOCK_IO_SLICE_TIME * 10;
                       type == BDRV_TYPE_FLOPPY));      }
 }  
   
 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)      return true;
 {  
     bs->translation = translation;  
 }  }
   
 void bdrv_get_geometry_hint(BlockDriverState *bs,  static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors,
                             int *pcyls, int *pheads, int *psecs)                             bool is_write, int64_t *wait)
 {  {
     *pcyls = bs->cyls;      int64_t  now, max_wait;
     *pheads = bs->heads;      uint64_t bps_wait = 0, iops_wait = 0;
     *psecs = bs->secs;      double   elapsed_time;
 }      int      bps_ret, iops_ret;
   
 int bdrv_get_type_hint(BlockDriverState *bs)      now = qemu_get_clock_ns(vm_clock);
 {      if ((bs->slice_start < now)
     return bs->type;          && (bs->slice_end > now)) {
 }          bs->slice_end = now + bs->slice_time;
       } else {
           bs->slice_time  =  5 * BLOCK_IO_SLICE_TIME;
           bs->slice_start = now;
           bs->slice_end   = now + bs->slice_time;
   
           bs->io_base.bytes[is_write]  = bs->nr_bytes[is_write];
           bs->io_base.bytes[!is_write] = bs->nr_bytes[!is_write];
   
           bs->io_base.ios[is_write]    = bs->nr_ops[is_write];
           bs->io_base.ios[!is_write]   = bs->nr_ops[!is_write];
       }
   
       elapsed_time  = now - bs->slice_start;
       elapsed_time  /= (NANOSECONDS_PER_SECOND);
   
       bps_ret  = bdrv_exceed_bps_limits(bs, nb_sectors,
                                         is_write, elapsed_time, &bps_wait);
       iops_ret = bdrv_exceed_iops_limits(bs, is_write,
                                         elapsed_time, &iops_wait);
       if (bps_ret || iops_ret) {
           max_wait = bps_wait > iops_wait ? bps_wait : iops_wait;
           if (wait) {
               *wait = max_wait;
           }
   
 int bdrv_get_translation_hint(BlockDriverState *bs)          now = qemu_get_clock_ns(vm_clock);
 {          if (bs->slice_end < now + max_wait) {
     return bs->translation;              bs->slice_end = now + max_wait;
 }          }
   
 int bdrv_is_removable(BlockDriverState *bs)          return true;
 {      }
     return bs->removable;  
 }  
   
 int bdrv_is_read_only(BlockDriverState *bs)      if (wait) {
 {          *wait = 0;
     return bs->read_only;      }
 }  
   
 int bdrv_is_sg(BlockDriverState *bs)      return false;
 {  
     return bs->sg;  
 }  }
   
 /* XXX: no longer used */  /**************************************************************/
 void bdrv_set_change_cb(BlockDriverState *bs,  /* async block device emulation */
                         void (*change_cb)(void *opaque), void *opaque)  
 {  
     bs->change_cb = change_cb;  
     bs->change_opaque = opaque;  
 }  
   
 int bdrv_is_encrypted(BlockDriverState *bs)  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)
 {  {
     if (bs->backing_hd && bs->backing_hd->encrypted)      BlockDriverAIOCBSync *acb =
         return 1;          container_of(blockacb, BlockDriverAIOCBSync, common);
     return bs->encrypted;      qemu_bh_delete(acb->bh);
       acb->bh = NULL;
       qemu_aio_release(acb);
 }  }
   
 int bdrv_key_required(BlockDriverState *bs)  static AIOPool bdrv_em_aio_pool = {
       .aiocb_size         = sizeof(BlockDriverAIOCBSync),
       .cancel             = bdrv_aio_cancel_em,
   };
   
   static void bdrv_aio_bh_cb(void *opaque)
 {  {
     BlockDriverState *backing_hd = bs->backing_hd;      BlockDriverAIOCBSync *acb = opaque;
   
     if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)      if (!acb->is_write)
         return 1;          qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
     return (bs->encrypted && !bs->valid_key);      qemu_vfree(acb->bounce);
       acb->common.cb(acb->common.opaque, acb->ret);
       qemu_bh_delete(acb->bh);
       acb->bh = NULL;
       qemu_aio_release(acb);
 }  }
   
 int bdrv_set_key(BlockDriverState *bs, const char *key)  static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
                                               int64_t sector_num,
                                               QEMUIOVector *qiov,
                                               int nb_sectors,
                                               BlockDriverCompletionFunc *cb,
                                               void *opaque,
                                               int is_write)
   
 {  {
     int ret;      BlockDriverAIOCBSync *acb;
     if (bs->backing_hd && bs->backing_hd->encrypted) {  
         ret = bdrv_set_key(bs->backing_hd, key);      acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
         if (ret < 0)      acb->is_write = is_write;
             return ret;      acb->qiov = qiov;
         if (!bs->encrypted)      acb->bounce = qemu_blockalign(bs, qiov->size);
             return 0;      acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
   
       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);
     }      }
     if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)  
         return -1;      qemu_bh_schedule(acb->bh);
     ret = bs->drv->bdrv_set_key(bs, key);  
     bs->valid_key = (ret == 0);      return &acb->common;
     return ret;  
 }  }
   
 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)  static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
           int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
           BlockDriverCompletionFunc *cb, void *opaque)
 {  {
     if (!bs->drv) {      return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
         buf[0] = '\0';  
     } else {  
         pstrcpy(buf, buf_size, bs->drv->format_name);  
     }  
 }  }
   
 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),  static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
                          void *opaque)          int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
           BlockDriverCompletionFunc *cb, void *opaque)
 {  {
     BlockDriver *drv;      return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
   }
   
     for (drv = first_drv; drv != NULL; drv = drv->next) {  
         it(opaque, drv->format_name);  typedef struct BlockDriverAIOCBCoroutine {
     }      BlockDriverAIOCB common;
       BlockRequest req;
       bool is_write;
       QEMUBH* bh;
   } BlockDriverAIOCBCoroutine;
   
   static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
   {
       qemu_aio_flush();
 }  }
   
 BlockDriverState *bdrv_find(const char *name)  static AIOPool bdrv_em_co_aio_pool = {
       .aiocb_size         = sizeof(BlockDriverAIOCBCoroutine),
       .cancel             = bdrv_aio_co_cancel_em,
   };
   
   static void bdrv_co_em_bh(void *opaque)
 {  {
     BlockDriverState *bs;      BlockDriverAIOCBCoroutine *acb = opaque;
   
     for (bs = bdrv_first; bs != NULL; bs = bs->next) {      acb->common.cb(acb->common.opaque, acb->req.error);
         if (!strcmp(name, bs->device_name))      qemu_bh_delete(acb->bh);
             return bs;      qemu_aio_release(acb);
     }  
     return NULL;  
 }  }
   
 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)  /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
   static void coroutine_fn bdrv_co_do_rw(void *opaque)
 {  {
     BlockDriverState *bs;      BlockDriverAIOCBCoroutine *acb = opaque;
       BlockDriverState *bs = acb->common.bs;
   
     for (bs = bdrv_first; bs != NULL; bs = bs->next) {      if (!acb->is_write) {
         it(opaque, bs);          acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
               acb->req.nb_sectors, acb->req.qiov, 0);
       } else {
           acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
               acb->req.nb_sectors, acb->req.qiov, 0);
     }      }
   
       acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
       qemu_bh_schedule(acb->bh);
 }  }
   
 const char *bdrv_get_device_name(BlockDriverState *bs)  static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
 {                                                 int64_t sector_num,
     return bs->device_name;                                                 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;
 }  }
   
 void bdrv_flush(BlockDriverState *bs)  static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
 {  {
     if (bs->drv->bdrv_flush)      BlockDriverAIOCBCoroutine *acb = opaque;
         bs->drv->bdrv_flush(bs);      BlockDriverState *bs = acb->common.bs;
     if (bs->backing_hd)  
         bdrv_flush(bs->backing_hd);      acb->req.error = bdrv_co_flush(bs);
       acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
       qemu_bh_schedule(acb->bh);
 }  }
   
 void bdrv_flush_all(void)  BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
           BlockDriverCompletionFunc *cb, void *opaque)
 {  {
     BlockDriverState *bs;      trace_bdrv_aio_flush(bs, opaque);
   
     for (bs = bdrv_first; bs != NULL; bs = bs->next)      Coroutine *co;
         if (bs->drv && !bdrv_is_read_only(bs) &&       BlockDriverAIOCBCoroutine *acb;
             (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))  
             bdrv_flush(bs);      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)
  * Returns true iff the specified sector is present in the disk image. Drivers  
  * not implementing the functionality are assumed to not support backing files,  
  * hence all their sectors are reported as allocated.  
  *  
  * 'pnum' is set to the number of sectors (including and immediately following  
  * the specified sector) that are known to be in the same  
  * allocated/unallocated state.  
  *  
  * 'nb_sectors' is the max value 'pnum' should be set to.  
  */  
 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,  
         int *pnum)  
 {  {
     int64_t n;      BlockDriverAIOCBCoroutine *acb = opaque;
     if (!bs->drv->bdrv_is_allocated) {      BlockDriverState *bs = acb->common.bs;
         if (sector_num >= bs->total_sectors) {  
             *pnum = 0;      acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
             return 0;      acb->bh = qemu_bh_new(bdrv_co_em_bh, acb);
         }      qemu_bh_schedule(acb->bh);
         n = bs->total_sectors - sector_num;  
         *pnum = (n < nb_sectors) ? (n) : (nb_sectors);  
         return 1;  
     }  
     return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);  
 }  }
   
 void bdrv_info(void)  BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
           int64_t sector_num, int nb_sectors,
           BlockDriverCompletionFunc *cb, void *opaque)
 {  {
     BlockDriverState *bs;      Coroutine *co;
       BlockDriverAIOCBCoroutine *acb;
   
     for (bs = bdrv_first; bs != NULL; bs = bs->next) {      trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
         term_printf("%s:", bs->device_name);  
         term_printf(" type=");      acb = qemu_aio_get(&bdrv_em_co_aio_pool, bs, cb, opaque);
         switch(bs->type) {      acb->req.sector = sector_num;
         case BDRV_TYPE_HD:      acb->req.nb_sectors = nb_sectors;
             term_printf("hd");      co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
             break;      qemu_coroutine_enter(co, acb);
         case BDRV_TYPE_CDROM:  
             term_printf("cdrom");      return &acb->common;
             break;  
         case BDRV_TYPE_FLOPPY:  
             term_printf("floppy");  
             break;  
         }  
         term_printf(" removable=%d", bs->removable);  
         if (bs->removable) {  
             term_printf(" locked=%d", bs->locked);  
         }  
         if (bs->drv) {  
             term_printf(" file=");  
             term_print_filename(bs->filename);  
             if (bs->backing_file[0] != '\0') {  
                 term_printf(" backing_file=");  
                 term_print_filename(bs->backing_file);  
             }  
             term_printf(" ro=%d", bs->read_only);  
             term_printf(" drv=%s", bs->drv->format_name);  
             term_printf(" encrypted=%d", bdrv_is_encrypted(bs));  
         } else {  
             term_printf(" [not inserted]");  
         }  
         term_printf("\n");  
     }  
 }  }
   
 /* The "info blockstats" command. */  void bdrv_init(void)
 void bdrv_info_stats (void)  
 {  {
     BlockDriverState *bs;      module_call_init(MODULE_INIT_BLOCK);
   
     for (bs = bdrv_first; bs != NULL; bs = bs->next) {  
         term_printf ("%s:"  
                      " rd_bytes=%" PRIu64  
                      " wr_bytes=%" PRIu64  
                      " rd_operations=%" PRIu64  
                      " wr_operations=%" PRIu64  
                      "\n",  
                      bs->device_name,  
                      bs->rd_bytes, bs->wr_bytes,  
                      bs->rd_ops, bs->wr_ops);  
     }  
 }  }
   
 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)  void bdrv_init_with_whitelist(void)
 {  {
     if (bs->backing_hd && bs->backing_hd->encrypted)      use_bdrv_whitelist = 1;
         return bs->backing_file;      bdrv_init();
     else if (bs->encrypted)  
         return bs->filename;  
     else  
         return NULL;  
 }  }
   
 void bdrv_get_backing_filename(BlockDriverState *bs,  void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
                                char *filename, int filename_size)                     BlockDriverCompletionFunc *cb, void *opaque)
 {  {
     if (!bs->backing_hd) {      BlockDriverAIOCB *acb;
         pstrcpy(filename, filename_size, "");  
       if (pool->free_aiocb) {
           acb = pool->free_aiocb;
           pool->free_aiocb = acb->next;
     } else {      } else {
         pstrcpy(filename, filename_size, bs->backing_file);          acb = g_malloc0(pool->aiocb_size);
           acb->pool = pool;
     }      }
       acb->bs = bs;
       acb->cb = cb;
       acb->opaque = opaque;
       return acb;
 }  }
   
 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,  void qemu_aio_release(void *p)
                           const uint8_t *buf, int nb_sectors)  
 {  {
     BlockDriver *drv = bs->drv;      BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
     if (!drv)      AIOPool *pool = acb->pool;
         return -ENOMEDIUM;      acb->next = pool->free_aiocb;
     if (!drv->bdrv_write_compressed)      pool->free_aiocb = acb;
         return -ENOTSUP;  
     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);  
 }  }
   
 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)  /**************************************************************/
 {  /* Coroutine block device emulation */
     BlockDriver *drv = bs->drv;  
     if (!drv)  typedef struct CoroutineIOCompletion {
         return -ENOMEDIUM;      Coroutine *coroutine;
     if (!drv->bdrv_get_info)      int ret;
         return -ENOTSUP;  } CoroutineIOCompletion;
     memset(bdi, 0, sizeof(*bdi));  
     return drv->bdrv_get_info(bs, bdi);  
 }  
   
 int bdrv_put_buffer(BlockDriverState *bs, const uint8_t *buf, int64_t pos, int size)  static void bdrv_co_io_em_complete(void *opaque, int ret)
 {  {
     BlockDriver *drv = bs->drv;      CoroutineIOCompletion *co = opaque;
     if (!drv)  
         return -ENOMEDIUM;      co->ret = ret;
     if (!drv->bdrv_put_buffer)      qemu_coroutine_enter(co->coroutine, NULL);
         return -ENOTSUP;  
     return drv->bdrv_put_buffer(bs, buf, pos, size);  
 }  }
   
 int bdrv_get_buffer(BlockDriverState *bs, uint8_t *buf, int64_t pos, int size)  static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
                                         int nb_sectors, QEMUIOVector *iov,
                                         bool is_write)
 {  {
     BlockDriver *drv = bs->drv;      CoroutineIOCompletion co = {
     if (!drv)          .coroutine = qemu_coroutine_self(),
         return -ENOMEDIUM;      };
     if (!drv->bdrv_get_buffer)      BlockDriverAIOCB *acb;
         return -ENOTSUP;  
     return drv->bdrv_get_buffer(bs, buf, pos, size);  
 }  
   
 /**************************************************************/      if (is_write) {
 /* handling of snapshots */          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);
       }
   
 int bdrv_snapshot_create(BlockDriverState *bs,      trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
                          QEMUSnapshotInfo *sn_info)      if (!acb) {
 {          return -EIO;
     BlockDriver *drv = bs->drv;      }
     if (!drv)      qemu_coroutine_yield();
         return -ENOMEDIUM;  
     if (!drv->bdrv_snapshot_create)      return co.ret;
         return -ENOTSUP;  
     return drv->bdrv_snapshot_create(bs, sn_info);  
 }  }
   
 int bdrv_snapshot_goto(BlockDriverState *bs,  static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
                        const char *snapshot_id)                                           int64_t sector_num, int nb_sectors,
                                            QEMUIOVector *iov)
 {  {
     BlockDriver *drv = bs->drv;      return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
     if (!drv)  
         return -ENOMEDIUM;  
     if (!drv->bdrv_snapshot_goto)  
         return -ENOTSUP;  
     return drv->bdrv_snapshot_goto(bs, snapshot_id);  
 }  }
   
 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)  static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
                                            int64_t sector_num, int nb_sectors,
                                            QEMUIOVector *iov)
 {  {
     BlockDriver *drv = bs->drv;      return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
     if (!drv)  
         return -ENOMEDIUM;  
     if (!drv->bdrv_snapshot_delete)  
         return -ENOTSUP;  
     return drv->bdrv_snapshot_delete(bs, snapshot_id);  
 }  }
   
 int bdrv_snapshot_list(BlockDriverState *bs,  static void coroutine_fn bdrv_flush_co_entry(void *opaque)
                        QEMUSnapshotInfo **psn_info)  
 {  {
     BlockDriver *drv = bs->drv;      RwCo *rwco = opaque;
     if (!drv)  
         return -ENOMEDIUM;  
     if (!drv->bdrv_snapshot_list)  
         return -ENOTSUP;  
     return drv->bdrv_snapshot_list(bs, psn_info);  
 }  
   
 #define NB_SUFFIXES 4      rwco->ret = bdrv_co_flush(rwco->bs);
   }
   
 char *get_human_readable_size(char *buf, int buf_size, int64_t size)  int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
 {  {
     static const char suffixes[NB_SUFFIXES] = "KMGT";      int ret;
     int64_t base;  
     int i;  
   
     if (size <= 999) {      if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
         snprintf(buf, buf_size, "%" PRId64, size);          return 0;
     } else {      }
         base = 1024;  
         for(i = 0; i < NB_SUFFIXES; i++) {      /* Write back cached data to the OS even with cache=unsafe */
             if (size < (10 * base)) {      if (bs->drv->bdrv_co_flush_to_os) {
                 snprintf(buf, buf_size, "%0.1f%c",          ret = bs->drv->bdrv_co_flush_to_os(bs);
                          (double)size / base,          if (ret < 0) {
                          suffixes[i]);              return ret;
                 break;  
             } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {  
                 snprintf(buf, buf_size, "%" PRId64 "%c",  
                          ((size + (base >> 1)) / base),  
                          suffixes[i]);  
                 break;  
             }  
             base = base * 1024;  
         }          }
     }      }
     return buf;  
 }  
   
 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)      /* But don't actually force it to the disk with cache=unsafe */
 {      if (bs->open_flags & BDRV_O_NO_FLUSH) {
     char buf1[128], date_buf[128], clock_buf[128];          return 0;
 #ifdef _WIN32      }
     struct tm *ptm;  
 #else  
     struct tm tm;  
 #endif  
     time_t ti;  
     int64_t secs;  
   
     if (!sn) {      if (bs->drv->bdrv_co_flush_to_disk) {
         snprintf(buf, buf_size,          ret = bs->drv->bdrv_co_flush_to_disk(bs);
                  "%-10s%-20s%7s%20s%15s",      } else if (bs->drv->bdrv_aio_flush) {
                  "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");          BlockDriverAIOCB *acb;
           CoroutineIOCompletion co = {
               .coroutine = qemu_coroutine_self(),
           };
   
           acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
           if (acb == NULL) {
               ret = -EIO;
           } else {
               qemu_coroutine_yield();
               ret = co.ret;
           }
     } else {      } else {
         ti = sn->date_sec;          /*
 #ifdef _WIN32           * Some block drivers always operate in either writethrough or unsafe
         ptm = localtime(&ti);           * mode and don't support bdrv_flush therefore. Usually qemu doesn't
         strftime(date_buf, sizeof(date_buf),           * know how the server works (because the behaviour is hardcoded or
                  "%Y-%m-%d %H:%M:%S", ptm);           * depends on server-side configuration), so we can't ensure that
 #else           * everything is safe on disk. Returning an error doesn't work because
         localtime_r(&ti, &tm);           * that would break guests even if the server operates in writethrough
         strftime(date_buf, sizeof(date_buf),           * mode.
                  "%Y-%m-%d %H:%M:%S", &tm);           *
 #endif           * Let's hope the user knows what he's doing.
         secs = sn->vm_clock_nsec / 1000000000;           */
         snprintf(clock_buf, sizeof(clock_buf),          ret = 0;
                  "%02d:%02d:%02d.%03d",  
                  (int)(secs / 3600),  
                  (int)((secs / 60) % 60),  
                  (int)(secs % 60),  
                  (int)((sn->vm_clock_nsec / 1000000) % 1000));  
         snprintf(buf, buf_size,  
                  "%-10s%-20s%7s%20s%15s",  
                  sn->id_str, sn->name,  
                  get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),  
                  date_buf,  
                  clock_buf);  
     }      }
     return buf;      if (ret < 0) {
           return ret;
       }
   
       /* Now flush the underlying protocol.  It will also have BDRV_O_NO_FLUSH
        * in the case of cache=unsafe, so there are no useless flushes.
        */
       return bdrv_co_flush(bs->file);
 }  }
   
   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)
 /* async I/Os */  {
       BlockDriverState *bs;
   
 typedef struct VectorTranslationState {      QTAILQ_FOREACH(bs, &bdrv_states, list) {
     QEMUIOVector *iov;          bdrv_invalidate_cache(bs);
     uint8_t *bounce;      }
     int is_write;  }
     BlockDriverAIOCB *aiocb;  
     BlockDriverAIOCB *this_aiocb;  
 } VectorTranslationState;  
   
 static void bdrv_aio_rw_vector_cb(void *opaque, int ret)  void bdrv_clear_incoming_migration_all(void)
 {  {
     VectorTranslationState *s = opaque;      BlockDriverState *bs;
   
     if (!s->is_write) {      QTAILQ_FOREACH(bs, &bdrv_states, list) {
         qemu_iovec_from_buffer(s->iov, s->bounce, s->iov->size);          bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
     }      }
     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,  int bdrv_flush(BlockDriverState *bs)
                                             int64_t sector_num,  
                                             QEMUIOVector *iov,  
                                             int nb_sectors,  
                                             BlockDriverCompletionFunc *cb,  
                                             void *opaque,  
                                             int is_write)  
   
 {  {
     VectorTranslationState *s = qemu_mallocz(sizeof(*s));      Coroutine *co;
     BlockDriverAIOCB *aiocb = qemu_aio_get(bs, cb, opaque);      RwCo rwco = {
           .bs = bs,
           .ret = NOT_DONE,
       };
   
     s->this_aiocb = aiocb;      if (qemu_in_coroutine()) {
     s->iov = iov;          /* Fast-path if already in coroutine context */
     s->bounce = qemu_memalign(512, nb_sectors * 512);          bdrv_flush_co_entry(&rwco);
     s->is_write = is_write;  
     if (is_write) {  
         qemu_iovec_to_buffer(s->iov, s->bounce);  
         s->aiocb = bdrv_aio_write(bs, sector_num, s->bounce, nb_sectors,  
                                   bdrv_aio_rw_vector_cb, s);  
     } else {      } else {
         s->aiocb = bdrv_aio_read(bs, sector_num, s->bounce, nb_sectors,          co = qemu_coroutine_create(bdrv_flush_co_entry);
                                  bdrv_aio_rw_vector_cb, s);          qemu_coroutine_enter(co, &rwco);
           while (rwco.ret == NOT_DONE) {
               qemu_aio_wait();
           }
     }      }
     return aiocb;  
       return rwco.ret;
 }  }
   
 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,  static void coroutine_fn bdrv_discard_co_entry(void *opaque)
                                  QEMUIOVector *iov, int nb_sectors,  
                                  BlockDriverCompletionFunc *cb, void *opaque)  
 {  {
     if (bdrv_check_request(bs, sector_num, nb_sectors))      RwCo *rwco = opaque;
         return NULL;  
   
     return bdrv_aio_rw_vector(bs, sector_num, iov, nb_sectors,      rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
                               cb, opaque, 0);  
 }  }
   
 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,  int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
                                   QEMUIOVector *iov, int nb_sectors,                                   int nb_sectors)
                                   BlockDriverCompletionFunc *cb, void *opaque)  
 {  {
     if (bdrv_check_request(bs, sector_num, nb_sectors))      if (!bs->drv) {
         return NULL;          return -ENOMEDIUM;
       } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
     return bdrv_aio_rw_vector(bs, sector_num, iov, nb_sectors,          return -EIO;
                               cb, opaque, 1);      } 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;
       }
 }  }
   
 BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,  int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
                                 uint8_t *buf, int nb_sectors,  
                                 BlockDriverCompletionFunc *cb, void *opaque)  
 {  {
     BlockDriver *drv = bs->drv;      Coroutine *co;
     BlockDriverAIOCB *ret;      RwCo rwco = {
           .bs = bs,
     if (!drv)          .sector_num = sector_num,
         return NULL;          .nb_sectors = nb_sectors,
     if (bdrv_check_request(bs, sector_num, nb_sectors))          .ret = NOT_DONE,
         return NULL;      };
   
     ret = drv->bdrv_aio_read(bs, sector_num, buf, nb_sectors, cb, opaque);      if (qemu_in_coroutine()) {
           /* Fast-path if already in coroutine context */
     if (ret) {          bdrv_discard_co_entry(&rwco);
         /* Update stats even though technically transfer has not happened. */      } else {
         bs->rd_bytes += (unsigned) nb_sectors * SECTOR_SIZE;          co = qemu_coroutine_create(bdrv_discard_co_entry);
         bs->rd_ops ++;          qemu_coroutine_enter(co, &rwco);
           while (rwco.ret == NOT_DONE) {
               qemu_aio_wait();
           }
     }      }
   
     return ret;      return rwco.ret;
 }  }
   
 BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,  /**************************************************************/
                                  const uint8_t *buf, int nb_sectors,  /* removable device support */
                                  BlockDriverCompletionFunc *cb, void *opaque)  
   /**
    * Return TRUE if the media is present
    */
   int bdrv_is_inserted(BlockDriverState *bs)
 {  {
     BlockDriver *drv = bs->drv;      BlockDriver *drv = bs->drv;
     BlockDriverAIOCB *ret;  
   
     if (!drv)      if (!drv)
         return NULL;          return 0;
     if (bs->read_only)      if (!drv->bdrv_is_inserted)
         return NULL;          return 1;
     if (bdrv_check_request(bs, sector_num, nb_sectors))      return drv->bdrv_is_inserted(bs);
         return NULL;  }
   
     ret = drv->bdrv_aio_write(bs, sector_num, buf, nb_sectors, cb, opaque);  /**
    * Return whether the media changed since the last call to this
    * function, or -ENOTSUP if we don't know.  Most drivers don't know.
    */
   int bdrv_media_changed(BlockDriverState *bs)
   {
       BlockDriver *drv = bs->drv;
   
     if (ret) {      if (drv && drv->bdrv_media_changed) {
         /* Update stats even though technically transfer has not happened. */          return drv->bdrv_media_changed(bs);
         bs->wr_bytes += (unsigned) nb_sectors * SECTOR_SIZE;  
         bs->wr_ops ++;  
     }      }
       return -ENOTSUP;
     return ret;  
 }  }
   
 void bdrv_aio_cancel(BlockDriverAIOCB *acb)  /**
    * If eject_flag is TRUE, eject the media. Otherwise, close the tray
    */
   void bdrv_eject(BlockDriverState *bs, bool eject_flag)
 {  {
     BlockDriver *drv = acb->bs->drv;      BlockDriver *drv = bs->drv;
   
     if (acb->cb == bdrv_aio_rw_vector_cb) {      if (drv && drv->bdrv_eject) {
         VectorTranslationState *s = acb->opaque;          drv->bdrv_eject(bs, eject_flag);
         acb = s->aiocb;  
     }      }
   
     drv->bdrv_aio_cancel(acb);      if (bs->device_name[0] != '\0') {
           bdrv_emit_qmp_eject_event(bs, eject_flag);
       }
 }  }
   
   /**
    * Lock or unlock the media (if it is locked, the user won't be able
    * to eject it manually).
    */
   void bdrv_lock_medium(BlockDriverState *bs, bool locked)
   {
       BlockDriver *drv = bs->drv;
   
 /**************************************************************/      trace_bdrv_lock_medium(bs, locked);
 /* async block device emulation */  
   
 static void bdrv_aio_bh_cb(void *opaque)      if (drv && drv->bdrv_lock_medium) {
 {          drv->bdrv_lock_medium(bs, locked);
     BlockDriverAIOCBSync *acb = opaque;      }
     acb->common.cb(acb->common.opaque, acb->ret);  
     qemu_aio_release(acb);  
 }  }
   
 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,  /* needed for generic scsi interface */
         int64_t sector_num, uint8_t *buf, int nb_sectors,  
         BlockDriverCompletionFunc *cb, void *opaque)  int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 {  {
     BlockDriverAIOCBSync *acb;      BlockDriver *drv = bs->drv;
     int ret;  
   
     acb = qemu_aio_get(bs, cb, opaque);      if (drv && drv->bdrv_ioctl)
     if (!acb->bh)          return drv->bdrv_ioctl(bs, req, buf);
         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);      return -ENOTSUP;
     ret = bdrv_read(bs, sector_num, buf, nb_sectors);  
     acb->ret = ret;  
     qemu_bh_schedule(acb->bh);  
     return &acb->common;  
 }  }
   
 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,  BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
         int64_t sector_num, const uint8_t *buf, int nb_sectors,          unsigned long int req, void *buf,
         BlockDriverCompletionFunc *cb, void *opaque)          BlockDriverCompletionFunc *cb, void *opaque)
 {  {
     BlockDriverAIOCBSync *acb;      BlockDriver *drv = bs->drv;
     int ret;  
   
     acb = qemu_aio_get(bs, cb, opaque);      if (drv && drv->bdrv_aio_ioctl)
     if (!acb->bh)          return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);      return NULL;
     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)  void bdrv_set_buffer_alignment(BlockDriverState *bs, int align)
 {  {
     BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;      bs->buffer_alignment = align;
     qemu_bh_cancel(acb->bh);  
     qemu_aio_release(acb);  
 }  }
   
 /**************************************************************/  void *qemu_blockalign(BlockDriverState *bs, size_t size)
 /* sync block device emulation */  
   
 static void bdrv_rw_em_cb(void *opaque, int ret)  
 {  {
     *(int *)opaque = ret;      return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
 }  }
   
 #define NOT_DONE 0x7fffffff  void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
   
 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,  
                         uint8_t *buf, int nb_sectors)  
 {  {
     int async_ret;      int64_t bitmap_size;
     BlockDriverAIOCB *acb;  
   
     async_ret = NOT_DONE;      bs->dirty_count = 0;
     acb = bdrv_aio_read(bs, sector_num, buf, nb_sectors,      if (enable) {
                         bdrv_rw_em_cb, &async_ret);          if (!bs->dirty_bitmap) {
     if (acb == NULL)              bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
         return -1;                      BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG - 1;
               bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG;
   
     while (async_ret == NOT_DONE) {              bs->dirty_bitmap = g_new0(unsigned long, bitmap_size);
         qemu_aio_wait();          }
       } else {
           if (bs->dirty_bitmap) {
               g_free(bs->dirty_bitmap);
               bs->dirty_bitmap = NULL;
           }
     }      }
   
     return async_ret;  
 }  }
   
 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,  int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
                          const uint8_t *buf, int nb_sectors)  
 {  {
     int async_ret;      int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
     BlockDriverAIOCB *acb;  
   
     async_ret = NOT_DONE;      if (bs->dirty_bitmap &&
     acb = bdrv_aio_write(bs, sector_num, buf, nb_sectors,          (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
                          bdrv_rw_em_cb, &async_ret);          return !!(bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
     if (acb == NULL)              (1UL << (chunk % (sizeof(unsigned long) * 8))));
         return -1;      } else {
     while (async_ret == NOT_DONE) {          return 0;
         qemu_aio_wait();  
     }      }
     return async_ret;  
 }  }
   
 void bdrv_init(void)  void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
                         int nr_sectors)
 {  {
     bdrv_register(&bdrv_raw);      set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
     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,  int64_t bdrv_get_dirty_count(BlockDriverState *bs)
                    void *opaque)  
 {  {
     BlockDriver *drv;      return bs->dirty_count;
     BlockDriverAIOCB *acb;  }
   
     drv = bs->drv;  void bdrv_set_in_use(BlockDriverState *bs, int in_use)
     if (drv->free_aiocb) {  {
         acb = drv->free_aiocb;      assert(bs->in_use != in_use);
         drv->free_aiocb = acb->next;      bs->in_use = in_use;
     } else {  }
         acb = qemu_mallocz(drv->aiocb_size);  
   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;
     }      }
     acb->bs = bs;  
     acb->cb = cb;  
     acb->opaque = opaque;  
     return acb;  
 }  }
   
 void qemu_aio_release(void *p)  /* 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)
 {  {
     BlockDriverAIOCB *acb = p;      if (bdrv_iostatus_is_enabled(bs) &&
     BlockDriver *drv = acb->bs->drv;          bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
     acb->next = drv->free_aiocb;          assert(error >= 0);
     drv->free_aiocb = acb;          bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
                                            BLOCK_DEVICE_IO_STATUS_FAILED;
       }
 }  }
   
 /**************************************************************/  void
 /* removable device support */  bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
           enum BlockAcctType type)
   {
       assert(type < BDRV_MAX_IOTYPE);
   
 /**      cookie->bytes = bytes;
  * Return TRUE if the media is present      cookie->start_time_ns = get_clock();
  */      cookie->type = type;
 int bdrv_is_inserted(BlockDriverState *bs)  }
   
   void
   bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
 {  {
     BlockDriver *drv = bs->drv;      assert(cookie->type < BDRV_MAX_IOTYPE);
     int ret;  
     if (!drv)      bs->nr_bytes[cookie->type] += cookie->bytes;
         return 0;      bs->nr_ops[cookie->type]++;
     if (!drv->bdrv_is_inserted)      bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
         return 1;  
     ret = drv->bdrv_is_inserted(bs);  
     return ret;  
 }  }
   
 /**  int bdrv_img_create(const char *filename, const char *fmt,
  * Return TRUE if the media changed since the last call to this                      const char *base_filename, const char *base_fmt,
  * function. It is currently only used for floppy disks                      char *options, uint64_t img_size, int flags)
  */  
 int bdrv_media_changed(BlockDriverState *bs)  
 {  {
     BlockDriver *drv = bs->drv;      QEMUOptionParameter *param = NULL, *create_options = NULL;
     int ret;      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];
               int back_flags;
   
               /* backing files always opened read-only */
               back_flags =
                   flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
   
               bs = bdrv_new("");
   
               ret = bdrv_open(bs, backing_file->value.s, back_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);
       }
   
     if (!drv || !drv->bdrv_media_changed)  
         ret = -ENOTSUP;  
     else  
         ret = drv->bdrv_media_changed(bs);  
     if (ret == -ENOTSUP)  
         ret = bs->media_changed;  
     bs->media_changed = 0;  
     return ret;      return ret;
 }  }
   
 /**  void *block_job_create(const BlockJobType *job_type, BlockDriverState *bs,
  * If eject_flag is TRUE, eject the media. Otherwise, close the tray                         int64_t speed, BlockDriverCompletionFunc *cb,
  */                         void *opaque, Error **errp)
 void bdrv_eject(BlockDriverState *bs, int eject_flag)  
 {  {
     BlockDriver *drv = bs->drv;      BlockJob *job;
     int ret;  
   
     if (!drv || !drv->bdrv_eject) {      if (bs->job || bdrv_in_use(bs)) {
         ret = -ENOTSUP;          error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
     } else {          return NULL;
         ret = drv->bdrv_eject(bs, eject_flag);  
     }      }
     if (ret == -ENOTSUP) {      bdrv_set_in_use(bs, 1);
         if (eject_flag)  
             bdrv_close(bs);      job = g_malloc0(job_type->instance_size);
       job->job_type      = job_type;
       job->bs            = bs;
       job->cb            = cb;
       job->opaque        = opaque;
       job->busy          = true;
       bs->job = job;
   
       /* Only set speed when necessary to avoid NotSupported error */
       if (speed != 0) {
           Error *local_err = NULL;
   
           block_job_set_speed(job, speed, &local_err);
           if (error_is_set(&local_err)) {
               bs->job = NULL;
               g_free(job);
               bdrv_set_in_use(bs, 0);
               error_propagate(errp, local_err);
               return NULL;
           }
     }      }
       return job;
 }  }
   
 int bdrv_is_locked(BlockDriverState *bs)  void block_job_complete(BlockJob *job, int ret)
 {  {
     return bs->locked;      BlockDriverState *bs = job->bs;
   
       assert(bs->job == job);
       job->cb(job->opaque, ret);
       bs->job = NULL;
       g_free(job);
       bdrv_set_in_use(bs, 0);
 }  }
   
 /**  void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp)
  * 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)  
 {  {
     BlockDriver *drv = bs->drv;      Error *local_err = NULL;
   
     bs->locked = locked;      if (!job->job_type->set_speed) {
     if (drv && drv->bdrv_set_locked) {          error_set(errp, QERR_NOT_SUPPORTED);
         drv->bdrv_set_locked(bs, locked);          return;
     }      }
       job->job_type->set_speed(job, speed, &local_err);
       if (error_is_set(&local_err)) {
           error_propagate(errp, local_err);
           return;
       }
   
       job->speed = speed;
 }  }
   
 /* needed for generic scsi interface */  void block_job_cancel(BlockJob *job)
   {
       job->cancelled = true;
       if (job->co && !job->busy) {
           qemu_coroutine_enter(job->co, NULL);
       }
   }
   
 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)  bool block_job_is_cancelled(BlockJob *job)
 {  {
     BlockDriver *drv = bs->drv;      return job->cancelled;
   }
   
     if (drv && drv->bdrv_ioctl)  struct BlockCancelData {
         return drv->bdrv_ioctl(bs, req, buf);      BlockJob *job;
     return -ENOTSUP;      BlockDriverCompletionFunc *cb;
       void *opaque;
       bool cancelled;
       int ret;
   };
   
   static void block_job_cancel_cb(void *opaque, int ret)
   {
       struct BlockCancelData *data = opaque;
   
       data->cancelled = block_job_is_cancelled(data->job);
       data->ret = ret;
       data->cb(data->opaque, ret);
   }
   
   int block_job_cancel_sync(BlockJob *job)
   {
       struct BlockCancelData data;
       BlockDriverState *bs = job->bs;
   
       assert(bs->job == job);
   
       /* Set up our own callback to store the result and chain to
        * the original callback.
        */
       data.job = job;
       data.cb = job->cb;
       data.opaque = job->opaque;
       data.ret = -EINPROGRESS;
       job->cb = block_job_cancel_cb;
       job->opaque = &data;
       block_job_cancel(job);
       while (data.ret == -EINPROGRESS) {
           qemu_aio_wait();
       }
       return (data.cancelled && data.ret == 0) ? -ECANCELED : data.ret;
   }
   
   void block_job_sleep_ns(BlockJob *job, QEMUClock *clock, int64_t ns)
   {
       /* Check cancellation *before* setting busy = false, too!  */
       if (!block_job_is_cancelled(job)) {
           job->busy = false;
           co_sleep_ns(clock, ns);
           job->busy = true;
       }
 }  }
Removed from v.1.1.1.9  
changed lines
  Added in v.1.1.1.23

unix.superglobalmegacorp.com