]> git.proxmox.com Git - qemu.git/blobdiff - block/stream.c
Merge remote-tracking branch 'luiz/queue/qmp' into staging
[qemu.git] / block / stream.c
index 0efe1adfd5e1302dddf3378a48d4eafbe2560504..d6df06f35a1157a399ced8dc03a3ef319f19979f 100644 (file)
@@ -12,7 +12,9 @@
  */
 
 #include "trace.h"
-#include "block_int.h"
+#include "block/block_int.h"
+#include "block/blockjob.h"
+#include "qemu/ratelimit.h"
 
 enum {
     /*
@@ -25,38 +27,11 @@ enum {
 
 #define SLICE_TIME 100000000ULL /* ns */
 
-typedef struct {
-    int64_t next_slice_time;
-    uint64_t slice_quota;
-    uint64_t dispatched;
-} RateLimit;
-
-static int64_t ratelimit_calculate_delay(RateLimit *limit, uint64_t n)
-{
-    int64_t delay_ns = 0;
-    int64_t now = qemu_get_clock_ns(rt_clock);
-
-    if (limit->next_slice_time < now) {
-        limit->next_slice_time = now + SLICE_TIME;
-        limit->dispatched = 0;
-    }
-    if (limit->dispatched + n > limit->slice_quota) {
-        delay_ns = limit->next_slice_time - now;
-    } else {
-        limit->dispatched += n;
-    }
-    return delay_ns;
-}
-
-static void ratelimit_set_speed(RateLimit *limit, uint64_t speed)
-{
-    limit->slice_quota = speed / (1000000000ULL / SLICE_TIME);
-}
-
 typedef struct StreamBlockJob {
     BlockJob common;
     RateLimit limit;
     BlockDriverState *base;
+    BlockdevOnError on_error;
     char backing_file_id[1024];
 } StreamBlockJob;
 
@@ -96,82 +71,6 @@ static void close_unused_images(BlockDriverState *top, BlockDriverState *base,
         bdrv_delete(unused);
     }
     top->backing_hd = base;
-
-    pstrcpy(top->backing_file, sizeof(top->backing_file), "");
-    pstrcpy(top->backing_format, sizeof(top->backing_format), "");
-    if (base_id) {
-        pstrcpy(top->backing_file, sizeof(top->backing_file), base_id);
-        if (base->drv) {
-            pstrcpy(top->backing_format, sizeof(top->backing_format),
-                    base->drv->format_name);
-        }
-    }
-
-}
-
-/*
- * Given an image chain: [BASE] -> [INTER1] -> [INTER2] -> [TOP]
- *
- * Return true if the given sector is allocated in top.
- * Return false if the given sector is allocated in intermediate images.
- * Return true otherwise.
- *
- * '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.
- *
- */
-static int coroutine_fn is_allocated_base(BlockDriverState *top,
-                                          BlockDriverState *base,
-                                          int64_t sector_num,
-                                          int nb_sectors, int *pnum)
-{
-    BlockDriverState *intermediate;
-    int ret, n;
-
-    ret = bdrv_co_is_allocated(top, sector_num, nb_sectors, &n);
-    if (ret) {
-        *pnum = n;
-        return ret;
-    }
-
-    /*
-     * Is the unallocated chunk [sector_num, n] also
-     * unallocated between base and top?
-     */
-    intermediate = top->backing_hd;
-
-    while (intermediate) {
-        int pnum_inter;
-
-        /* reached base */
-        if (intermediate == base) {
-            *pnum = n;
-            return 1;
-        }
-        ret = bdrv_co_is_allocated(intermediate, sector_num, nb_sectors,
-                                   &pnum_inter);
-        if (ret < 0) {
-            return ret;
-        } else if (ret) {
-            *pnum = pnum_inter;
-            return 0;
-        }
-
-        /*
-         * [sector_num, nb_sectors] is unallocated on top but intermediate
-         * might have
-         *
-         * [sector_num+x, nr_sectors] allocated.
-         */
-        if (n > pnum_inter) {
-            n = pnum_inter;
-        }
-
-        intermediate = intermediate->backing_hd;
-    }
-
-    return 1;
 }
 
 static void coroutine_fn stream_run(void *opaque)
@@ -180,13 +79,14 @@ static void coroutine_fn stream_run(void *opaque)
     BlockDriverState *bs = s->common.bs;
     BlockDriverState *base = s->base;
     int64_t sector_num, end;
+    int error = 0;
     int ret = 0;
-    int n;
+    int n = 0;
     void *buf;
 
     s->common.len = bdrv_getlength(bs);
     if (s->common.len < 0) {
-        block_job_complete(&s->common, s->common.len);
+        block_job_completed(&s->common, s->common.len);
         return;
     }
 
@@ -203,75 +103,99 @@ static void coroutine_fn stream_run(void *opaque)
     }
 
     for (sector_num = 0; sector_num < end; sector_num += n) {
-retry:
+        uint64_t delay_ns = 0;
+        bool copy;
+
+wait:
+        /* Note that even when no rate limit is applied we need to yield
+         * with no pending I/O here so that bdrv_drain_all() returns.
+         */
+        block_job_sleep_ns(&s->common, rt_clock, delay_ns);
         if (block_job_is_cancelled(&s->common)) {
             break;
         }
 
-        s->common.busy = true;
-        if (base) {
-            ret = is_allocated_base(bs, base, sector_num,
-                                    STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
+        ret = bdrv_co_is_allocated(bs, sector_num,
+                                   STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
+        if (ret == 1) {
+            /* Allocated in the top, no need to copy.  */
+            copy = false;
         } else {
-            ret = bdrv_co_is_allocated(bs, sector_num,
-                                       STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE,
-                                       &n);
+            /* Copy if allocated in the intermediate images.  Limit to the
+             * known-unallocated area [sector_num, sector_num+n).  */
+            ret = bdrv_co_is_allocated_above(bs->backing_hd, base,
+                                             sector_num, n, &n);
+
+            /* Finish early if end of backing file has been reached */
+            if (ret == 0 && n == 0) {
+                n = end - sector_num;
+            }
+
+            copy = (ret == 1);
         }
         trace_stream_one_iteration(s, sector_num, n, ret);
-        if (ret == 0) {
+        if (ret >= 0 && copy) {
             if (s->common.speed) {
-                uint64_t delay_ns = ratelimit_calculate_delay(&s->limit, n);
+                delay_ns = ratelimit_calculate_delay(&s->limit, n);
                 if (delay_ns > 0) {
-                    s->common.busy = false;
-                    co_sleep_ns(rt_clock, delay_ns);
-
-                    /* Recheck cancellation and that sectors are unallocated */
-                    goto retry;
+                    goto wait;
                 }
             }
             ret = stream_populate(bs, sector_num, n, buf);
         }
         if (ret < 0) {
-            break;
+            BlockErrorAction action =
+                block_job_error_action(&s->common, s->common.bs, s->on_error,
+                                       true, -ret);
+            if (action == BDRV_ACTION_STOP) {
+                n = 0;
+                continue;
+            }
+            if (error == 0) {
+                error = ret;
+            }
+            if (action == BDRV_ACTION_REPORT) {
+                break;
+            }
         }
         ret = 0;
 
         /* Publish progress */
         s->common.offset += n * BDRV_SECTOR_SIZE;
-
-        /* Note that even when no rate limit is applied we need to yield
-         * with no pending I/O here so that qemu_aio_flush() returns.
-         */
-        s->common.busy = false;
-        co_sleep_ns(rt_clock, 0);
     }
 
     if (!base) {
         bdrv_disable_copy_on_read(bs);
     }
 
+    /* Do not remove the backing file if an error was there but ignored.  */
+    ret = error;
+
     if (!block_job_is_cancelled(&s->common) && sector_num == end && ret == 0) {
-        const char *base_id = NULL;
+        const char *base_id = NULL, *base_fmt = NULL;
         if (base) {
             base_id = s->backing_file_id;
+            if (base->drv) {
+                base_fmt = base->drv->format_name;
+            }
         }
-        ret = bdrv_change_backing_file(bs, base_id, NULL);
+        ret = bdrv_change_backing_file(bs, base_id, base_fmt);
         close_unused_images(bs, base, base_id);
     }
 
     qemu_vfree(buf);
-    block_job_complete(&s->common, ret);
+    block_job_completed(&s->common, ret);
 }
 
-static int stream_set_speed(BlockJob *job, int64_t value)
+static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp)
 {
     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
 
-    if (value < 0) {
-        return -EINVAL;
+    if (speed < 0) {
+        error_set(errp, QERR_INVALID_PARAMETER, "speed");
+        return;
     }
-    ratelimit_set_speed(&s->limit, value / BDRV_SECTOR_SIZE);
-    return 0;
+    ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
 }
 
 static BlockJobType stream_job_type = {
@@ -280,16 +204,24 @@ static BlockJobType stream_job_type = {
     .set_speed     = stream_set_speed,
 };
 
-int stream_start(BlockDriverState *bs, BlockDriverState *base,
-                 const char *base_id, BlockDriverCompletionFunc *cb,
-                 void *opaque)
+void stream_start(BlockDriverState *bs, BlockDriverState *base,
+                  const char *base_id, int64_t speed,
+                  BlockdevOnError on_error,
+                  BlockDriverCompletionFunc *cb,
+                  void *opaque, Error **errp)
 {
     StreamBlockJob *s;
-    Coroutine *co;
 
-    s = block_job_create(&stream_job_type, bs, cb, opaque);
+    if ((on_error == BLOCKDEV_ON_ERROR_STOP ||
+         on_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
+        !bdrv_iostatus_is_enabled(bs)) {
+        error_set(errp, QERR_INVALID_PARAMETER, "on-error");
+        return;
+    }
+
+    s = block_job_create(&stream_job_type, bs, speed, cb, opaque, errp);
     if (!s) {
-        return -EBUSY; /* bs must already be in use */
+        return;
     }
 
     s->base = base;
@@ -297,8 +229,8 @@ int stream_start(BlockDriverState *bs, BlockDriverState *base,
         pstrcpy(s->backing_file_id, sizeof(s->backing_file_id), base_id);
     }
 
-    co = qemu_coroutine_create(stream_run);
-    trace_stream_start(bs, base, s, co, opaque);
-    qemu_coroutine_enter(co, s);
-    return 0;
+    s->on_error = on_error;
+    s->common.co = qemu_coroutine_create(stream_run);
+    trace_stream_start(bs, base, s, s->common.co, opaque);
+    qemu_coroutine_enter(s->common.co, s);
 }