]> git.proxmox.com Git - qemu.git/blobdiff - block/qcow2.c
block: move include files to include/block/
[qemu.git] / block / qcow2.c
index 2c1cd0a446d098d0eb815f22a2919f3d3450d3ee..205d910a52d018ce7e502d9dd9cd8ecb62111dd9 100644 (file)
  * THE SOFTWARE.
  */
 #include "qemu-common.h"
-#include "block_int.h"
+#include "block/block_int.h"
 #include "module.h"
 #include <zlib.h>
-#include "aes.h"
+#include "block/aes.h"
 #include "block/qcow2.h"
 #include "qemu-error.h"
-#include "qerror.h"
+#include "qapi/qmp/qerror.h"
 #include "trace.h"
 
 /*
@@ -52,6 +52,7 @@ typedef struct {
     uint32_t magic;
     uint32_t len;
 } QCowExtension;
+
 #define  QCOW2_EXT_MAGIC_END 0
 #define  QCOW2_EXT_MAGIC_BACKING_FORMAT 0xE2792ACA
 #define  QCOW2_EXT_MAGIC_FEATURE_TABLE 0x6803f857
@@ -214,6 +215,76 @@ static void report_unsupported_feature(BlockDriverState *bs,
     }
 }
 
+/*
+ * Sets the dirty bit and flushes afterwards if necessary.
+ *
+ * The incompatible_features bit is only set if the image file header was
+ * updated successfully.  Therefore it is not required to check the return
+ * value of this function.
+ */
+int qcow2_mark_dirty(BlockDriverState *bs)
+{
+    BDRVQcowState *s = bs->opaque;
+    uint64_t val;
+    int ret;
+
+    assert(s->qcow_version >= 3);
+
+    if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
+        return 0; /* already dirty */
+    }
+
+    val = cpu_to_be64(s->incompatible_features | QCOW2_INCOMPAT_DIRTY);
+    ret = bdrv_pwrite(bs->file, offsetof(QCowHeader, incompatible_features),
+                      &val, sizeof(val));
+    if (ret < 0) {
+        return ret;
+    }
+    ret = bdrv_flush(bs->file);
+    if (ret < 0) {
+        return ret;
+    }
+
+    /* Only treat image as dirty if the header was updated successfully */
+    s->incompatible_features |= QCOW2_INCOMPAT_DIRTY;
+    return 0;
+}
+
+/*
+ * Clears the dirty bit and flushes before if necessary.  Only call this
+ * function when there are no pending requests, it does not guard against
+ * concurrent requests dirtying the image.
+ */
+static int qcow2_mark_clean(BlockDriverState *bs)
+{
+    BDRVQcowState *s = bs->opaque;
+
+    if (s->incompatible_features & QCOW2_INCOMPAT_DIRTY) {
+        int ret = bdrv_flush(bs);
+        if (ret < 0) {
+            return ret;
+        }
+
+        s->incompatible_features &= ~QCOW2_INCOMPAT_DIRTY;
+        return qcow2_update_header(bs);
+    }
+    return 0;
+}
+
+static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
+                       BdrvCheckMode fix)
+{
+    int ret = qcow2_check_refcounts(bs, result, fix);
+    if (ret < 0) {
+        return ret;
+    }
+
+    if (fix && result->check_errors == 0 && result->corruptions == 0) {
+        return qcow2_mark_clean(bs);
+    }
+    return ret;
+}
+
 static int qcow2_open(BlockDriverState *bs, int flags)
 {
     BDRVQcowState *s = bs->opaque;
@@ -287,12 +358,13 @@ static int qcow2_open(BlockDriverState *bs, int flags)
     s->compatible_features      = header.compatible_features;
     s->autoclear_features       = header.autoclear_features;
 
-    if (s->incompatible_features != 0) {
+    if (s->incompatible_features & ~QCOW2_INCOMPAT_MASK) {
         void *feature_table = NULL;
         qcow2_read_extensions(bs, header.header_length, ext_end,
                               &feature_table);
         report_unsupported_feature(bs, feature_table,
-                                   s->incompatible_features);
+                                   s->incompatible_features &
+                                   ~QCOW2_INCOMPAT_MASK);
         ret = -ENOTSUP;
         goto fail;
     }
@@ -412,10 +484,21 @@ static int qcow2_open(BlockDriverState *bs, int flags)
     /* Initialise locks */
     qemu_co_mutex_init(&s->lock);
 
+    /* Repair image if dirty */
+    if (!(flags & BDRV_O_CHECK) && !bs->read_only &&
+        (s->incompatible_features & QCOW2_INCOMPAT_DIRTY)) {
+        BdrvCheckResult result = {0};
+
+        ret = qcow2_check(bs, &result, BDRV_FIX_ERRORS);
+        if (ret < 0) {
+            goto fail;
+        }
+    }
+
 #ifdef DEBUG_ALLOC
     {
         BdrvCheckResult result = {0};
-        qcow2_check_refcounts(bs, &result);
+        qcow2_check_refcounts(bs, &result, 0);
     }
 #endif
     return ret;
@@ -476,6 +559,14 @@ static int qcow2_set_key(BlockDriverState *bs, const char *key)
     return 0;
 }
 
+/* We have nothing to do for QCOW2 reopen, stubs just return
+ * success */
+static int qcow2_reopen_prepare(BDRVReopenState *state,
+                                BlockReopenQueue *queue, Error **errp)
+{
+    return 0;
+}
+
 static int coroutine_fn qcow2_co_is_allocated(BlockDriverState *bs,
         int64_t sector_num, int nb_sectors, int *pnum)
 {
@@ -508,7 +599,7 @@ int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
     else
         n1 = bs->total_sectors - sector_num;
 
-    qemu_iovec_memset_skip(qiov, 0, 512 * (nb_sectors - n1), 512 * n1);
+    qemu_iovec_memset(qiov, 512 * n1, 0, 512 * (nb_sectors - n1));
 
     return n1;
 }
@@ -547,7 +638,7 @@ static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
         index_in_cluster = sector_num & (s->cluster_sectors - 1);
 
         qemu_iovec_reset(&hd_qiov);
-        qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
+        qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
             cur_nr_sectors * 512);
 
         switch (ret) {
@@ -569,7 +660,7 @@ static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
                 }
             } else {
                 /* Note: in this case, no need to wait */
-                qemu_iovec_memset(&hd_qiov, 0, 512 * cur_nr_sectors);
+                qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
             }
             break;
 
@@ -578,7 +669,7 @@ static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
                 ret = -EIO;
                 goto fail;
             }
-            qemu_iovec_memset(&hd_qiov, 0, 512 * cur_nr_sectors);
+            qemu_iovec_memset(&hd_qiov, 0, 0, 512 * cur_nr_sectors);
             break;
 
         case QCOW2_CLUSTER_COMPRESSED:
@@ -588,7 +679,7 @@ static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
                 goto fail;
             }
 
-            qemu_iovec_from_buffer(&hd_qiov,
+            qemu_iovec_from_buf(&hd_qiov, 0,
                 s->cluster_cache + index_in_cluster * 512,
                 512 * cur_nr_sectors);
             break;
@@ -628,11 +719,8 @@ static coroutine_fn int qcow2_co_readv(BlockDriverState *bs, int64_t sector_num,
             if (s->crypt_method) {
                 qcow2_encrypt_sectors(s, sector_num,  cluster_data,
                     cluster_data, cur_nr_sectors, 0, &s->aes_decrypt_key);
-                qemu_iovec_reset(&hd_qiov);
-                qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
-                    cur_nr_sectors * 512);
-                qemu_iovec_from_buffer(&hd_qiov, cluster_data,
-                    512 * cur_nr_sectors);
+                qemu_iovec_from_buf(qiov, bytes_done,
+                    cluster_data, 512 * cur_nr_sectors);
             }
             break;
 
@@ -657,21 +745,6 @@ fail:
     return ret;
 }
 
-static void run_dependent_requests(BDRVQcowState *s, QCowL2Meta *m)
-{
-    /* Take the request off the list of running requests */
-    if (m->nb_clusters != 0) {
-        QLIST_REMOVE(m, next_in_flight);
-    }
-
-    /* Restart all dependent requests */
-    if (!qemu_co_queue_empty(&m->dependent_requests)) {
-        qemu_co_mutex_unlock(&s->lock);
-        qemu_co_queue_restart_all(&m->dependent_requests);
-        qemu_co_mutex_lock(&s->lock);
-    }
-}
-
 static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
                            int64_t sector_num,
                            int remaining_sectors,
@@ -686,15 +759,11 @@ static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
     QEMUIOVector hd_qiov;
     uint64_t bytes_done = 0;
     uint8_t *cluster_data = NULL;
-    QCowL2Meta l2meta = {
-        .nb_clusters = 0,
-    };
+    QCowL2Meta *l2meta;
 
     trace_qcow2_writev_start_req(qemu_coroutine_self(), sector_num,
                                  remaining_sectors);
 
-    qemu_co_queue_init(&l2meta.dependent_requests);
-
     qemu_iovec_init(&hd_qiov, qiov->niov);
 
     s->cluster_cache_offset = -1; /* disable compressed cache */
@@ -703,6 +772,8 @@ static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
 
     while (remaining_sectors != 0) {
 
+        l2meta = NULL;
+
         trace_qcow2_writev_start_part(qemu_coroutine_self());
         index_in_cluster = sector_num & (s->cluster_sectors - 1);
         n_end = index_in_cluster + remaining_sectors;
@@ -712,16 +783,15 @@ static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
         }
 
         ret = qcow2_alloc_cluster_offset(bs, sector_num << 9,
-            index_in_cluster, n_end, &cur_nr_sectors, &l2meta);
+            index_in_cluster, n_end, &cur_nr_sectors, &cluster_offset, &l2meta);
         if (ret < 0) {
             goto fail;
         }
 
-        cluster_offset = l2meta.cluster_offset;
         assert((cluster_offset & 511) == 0);
 
         qemu_iovec_reset(&hd_qiov);
-        qemu_iovec_copy(&hd_qiov, qiov, bytes_done,
+        qemu_iovec_concat(&hd_qiov, qiov, bytes_done,
             cur_nr_sectors * 512);
 
         if (s->crypt_method) {
@@ -732,7 +802,7 @@ static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
 
             assert(hd_qiov.size <=
                    QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size);
-            qemu_iovec_to_buffer(&hd_qiov, cluster_data);
+            qemu_iovec_to_buf(&hd_qiov, 0, cluster_data, hd_qiov.size);
 
             qcow2_encrypt_sectors(s, sector_num, cluster_data,
                 cluster_data, cur_nr_sectors, 1, &s->aes_encrypt_key);
@@ -742,8 +812,8 @@ static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
                 cur_nr_sectors * 512);
         }
 
-        BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
         qemu_co_mutex_unlock(&s->lock);
+        BLKDBG_EVENT(bs->file, BLKDBG_WRITE_AIO);
         trace_qcow2_writev_data(qemu_coroutine_self(),
                                 (cluster_offset >> 9) + index_in_cluster);
         ret = bdrv_co_writev(bs->file,
@@ -754,12 +824,24 @@ static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
             goto fail;
         }
 
-        ret = qcow2_alloc_cluster_link_l2(bs, &l2meta);
-        if (ret < 0) {
-            goto fail;
-        }
+        if (l2meta != NULL) {
+            ret = qcow2_alloc_cluster_link_l2(bs, l2meta);
+            if (ret < 0) {
+                goto fail;
+            }
+
+            /* Take the request off the list of running requests */
+            if (l2meta->nb_clusters != 0) {
+                QLIST_REMOVE(l2meta, next_in_flight);
+            }
 
-        run_dependent_requests(s, &l2meta);
+            qemu_co_mutex_unlock(&s->lock);
+            qemu_co_queue_restart_all(&l2meta->dependent_requests);
+            qemu_co_mutex_lock(&s->lock);
+
+            g_free(l2meta);
+            l2meta = NULL;
+        }
 
         remaining_sectors -= cur_nr_sectors;
         sector_num += cur_nr_sectors;
@@ -769,10 +851,16 @@ static coroutine_fn int qcow2_co_writev(BlockDriverState *bs,
     ret = 0;
 
 fail:
-    run_dependent_requests(s, &l2meta);
-
     qemu_co_mutex_unlock(&s->lock);
 
+    if (l2meta != NULL) {
+        if (l2meta->nb_clusters != 0) {
+            QLIST_REMOVE(l2meta, next_in_flight);
+        }
+        qemu_co_queue_restart_all(&l2meta->dependent_requests);
+        g_free(l2meta);
+    }
+
     qemu_iovec_destroy(&hd_qiov);
     qemu_vfree(cluster_data);
     trace_qcow2_writev_done_req(qemu_coroutine_self(), ret);
@@ -788,6 +876,8 @@ static void qcow2_close(BlockDriverState *bs)
     qcow2_cache_flush(bs, s->l2_table_cache);
     qcow2_cache_flush(bs, s->refcount_block_cache);
 
+    qcow2_mark_clean(bs);
+
     qcow2_cache_destroy(bs, s->l2_table_cache);
     qcow2_cache_destroy(bs, s->refcount_block_cache);
 
@@ -952,7 +1042,16 @@ int qcow2_update_header(BlockDriverState *bs)
 
     /* Feature table */
     Qcow2Feature features[] = {
-        /* no feature defined yet */
+        {
+            .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
+            .bit  = QCOW2_INCOMPAT_DIRTY_BITNR,
+            .name = "dirty bit",
+        },
+        {
+            .type = QCOW2_FEAT_TYPE_COMPATIBLE,
+            .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
+            .name = "lazy refcounts",
+        },
     };
 
     ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
@@ -992,6 +1091,7 @@ int qcow2_update_header(BlockDriverState *bs)
             goto fail;
         }
 
+        /* Using strncpy is ok here, since buf is not NUL-terminated. */
         strncpy(buf, bs->backing_file, buflen);
 
         header->backing_file_offset = cpu_to_be64(buf - ((char*) header));
@@ -1023,31 +1123,33 @@ static int preallocate(BlockDriverState *bs)
 {
     uint64_t nb_sectors;
     uint64_t offset;
+    uint64_t host_offset = 0;
     int num;
     int ret;
-    QCowL2Meta meta;
+    QCowL2Meta *meta;
 
     nb_sectors = bdrv_getlength(bs) >> 9;
     offset = 0;
-    qemu_co_queue_init(&meta.dependent_requests);
-    meta.cluster_offset = 0;
 
     while (nb_sectors) {
         num = MIN(nb_sectors, INT_MAX >> 9);
-        ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num, &meta);
+        ret = qcow2_alloc_cluster_offset(bs, offset, 0, num, &num,
+                                         &host_offset, &meta);
         if (ret < 0) {
             return ret;
         }
 
-        ret = qcow2_alloc_cluster_link_l2(bs, &meta);
+        ret = qcow2_alloc_cluster_link_l2(bs, meta);
         if (ret < 0) {
-            qcow2_free_any_clusters(bs, meta.cluster_offset, meta.nb_clusters);
+            qcow2_free_any_clusters(bs, meta->alloc_offset, meta->nb_clusters);
             return ret;
         }
 
         /* There are no dependent requests, but we need to remove our request
          * from the list of in-flight requests */
-        run_dependent_requests(bs->opaque, &meta);
+        if (meta != NULL) {
+            QLIST_REMOVE(meta, next_in_flight);
+        }
 
         /* TODO Preallocate data if requested */
 
@@ -1060,10 +1162,10 @@ static int preallocate(BlockDriverState *bs)
      * all of the allocated clusters (otherwise we get failing reads after
      * EOF). Extend the image to the last allocated sector.
      */
-    if (meta.cluster_offset != 0) {
+    if (host_offset != 0) {
         uint8_t buf[512];
         memset(buf, 0, 512);
-        ret = bdrv_write(bs->file, (meta.cluster_offset >> 9) + num - 1, buf, 1);
+        ret = bdrv_write(bs->file, (host_offset >> 9) + num - 1, buf, 1);
         if (ret < 0) {
             return ret;
         }
@@ -1135,6 +1237,11 @@ static int qcow2_create2(const char *filename, int64_t total_size,
         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
     }
 
+    if (flags & BLOCK_FLAG_LAZY_REFCOUNTS) {
+        header.compatible_features |=
+            cpu_to_be64(QCOW2_COMPAT_LAZY_REFCOUNTS);
+    }
+
     ret = bdrv_pwrite(bs, 0, &header, sizeof(header));
     if (ret < 0) {
         goto out;
@@ -1248,6 +1355,8 @@ static int qcow2_create(const char *filename, QEMUOptionParameter *options)
                     options->value.s);
                 return -EINVAL;
             }
+        } else if (!strcmp(options->name, BLOCK_OPT_LAZY_REFCOUNTS)) {
+            flags |= options->value.n ? BLOCK_FLAG_LAZY_REFCOUNTS : 0;
         }
         options++;
     }
@@ -1258,6 +1367,12 @@ static int qcow2_create(const char *filename, QEMUOptionParameter *options)
         return -EINVAL;
     }
 
+    if (version < 3 && (flags & BLOCK_FLAG_LAZY_REFCOUNTS)) {
+        fprintf(stderr, "Lazy refcounts only supported with compatibility "
+                "level 1.1 and above (use compat=1.1 or greater)\n");
+        return -EINVAL;
+    }
+
     return qcow2_create2(filename, sectors, backing_file, backing_fmt, flags,
                          cluster_size, prealloc, options, version);
 }
@@ -1444,10 +1559,12 @@ static coroutine_fn int qcow2_co_flush_to_os(BlockDriverState *bs)
         return ret;
     }
 
-    ret = qcow2_cache_flush(bs, s->refcount_block_cache);
-    if (ret < 0) {
-        qemu_co_mutex_unlock(&s->lock);
-        return ret;
+    if (qcow2_need_accurate_refcounts(s)) {
+        ret = qcow2_cache_flush(bs, s->refcount_block_cache);
+        if (ret < 0) {
+            qemu_co_mutex_unlock(&s->lock);
+            return ret;
+        }
     }
     qemu_co_mutex_unlock(&s->lock);
 
@@ -1467,13 +1584,6 @@ static int qcow2_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
     return 0;
 }
 
-
-static int qcow2_check(BlockDriverState *bs, BdrvCheckResult *result,
-                       BdrvCheckMode fix)
-{
-    return qcow2_check_refcounts(bs, result, fix);
-}
-
 #if 0
 static void dump_refcounts(BlockDriverState *bs)
 {
@@ -1562,6 +1672,11 @@ static QEMUOptionParameter qcow2_create_options[] = {
         .type = OPT_STRING,
         .help = "Preallocation mode (allowed values: off, metadata)"
     },
+    {
+        .name = BLOCK_OPT_LAZY_REFCOUNTS,
+        .type = OPT_FLAG,
+        .help = "Postpone refcount updates",
+    },
     { NULL }
 };
 
@@ -1571,6 +1686,7 @@ static BlockDriver bdrv_qcow2 = {
     .bdrv_probe         = qcow2_probe,
     .bdrv_open          = qcow2_open,
     .bdrv_close         = qcow2_close,
+    .bdrv_reopen_prepare  = qcow2_reopen_prepare,
     .bdrv_create        = qcow2_create,
     .bdrv_co_is_allocated = qcow2_co_is_allocated,
     .bdrv_set_key       = qcow2_set_key,