diff options
Diffstat (limited to 'block/qcow2-cluster.c')
| -rw-r--r-- | block/qcow2-cluster.c | 509 |
1 files changed, 360 insertions, 149 deletions
diff --git a/block/qcow2-cluster.c b/block/qcow2-cluster.c index d72d063e6d..c71470a3db 100644 --- a/block/qcow2-cluster.c +++ b/block/qcow2-cluster.c @@ -759,31 +759,50 @@ out: * Check if there already is an AIO write request in flight which allocates * the same cluster. In this case we need to wait until the previous * request has completed and updated the L2 table accordingly. + * + * Returns: + * 0 if there was no dependency. *cur_bytes indicates the number of + * bytes from guest_offset that can be read before the next + * dependency must be processed (or the request is complete) + * + * -EAGAIN if we had to wait for another request, previously gathered + * information on cluster allocation may be invalid now. The caller + * must start over anyway, so consider *cur_bytes undefined. */ static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, - unsigned int *nb_clusters) + uint64_t *cur_bytes, QCowL2Meta **m) { BDRVQcowState *s = bs->opaque; QCowL2Meta *old_alloc; + uint64_t bytes = *cur_bytes; QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) { - uint64_t start = guest_offset >> s->cluster_bits; - uint64_t end = start + *nb_clusters; - uint64_t old_start = old_alloc->offset >> s->cluster_bits; - uint64_t old_end = old_start + old_alloc->nb_clusters; + uint64_t start = guest_offset; + uint64_t end = start + bytes; + uint64_t old_start = l2meta_cow_start(old_alloc); + uint64_t old_end = l2meta_cow_end(old_alloc); - if (end < old_start || start > old_end) { + if (end <= old_start || start >= old_end) { /* No intersection */ } else { if (start < old_start) { /* Stop at the start of a running allocation */ - *nb_clusters = old_start - start; + bytes = old_start - start; } else { - *nb_clusters = 0; + bytes = 0; + } + + /* Stop if already an l2meta exists. After yielding, it wouldn't + * be valid any more, so we'd have to clean up the old L2Metas + * and deal with requests depending on them before starting to + * gather new ones. Not worth the trouble. */ + if (bytes == 0 && *m) { + *cur_bytes = 0; + return 0; } - if (*nb_clusters == 0) { + if (bytes == 0) { /* Wait for the dependency to complete. We need to recheck * the free/allocated clusters when we continue. */ qemu_co_mutex_unlock(&s->lock); @@ -794,14 +813,118 @@ static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset, } } - if (!*nb_clusters) { - abort(); - } + /* Make sure that existing clusters and new allocations are only used up to + * the next dependency if we shortened the request above */ + *cur_bytes = bytes; return 0; } /* + * Checks how many already allocated clusters that don't require a copy on + * write there are at the given guest_offset (up to *bytes). If + * *host_offset is not zero, only physically contiguous clusters beginning at + * this host offset are counted. + * + * Note that guest_offset may not be cluster aligned. In this case, the + * returned *host_offset points to exact byte referenced by guest_offset and + * therefore isn't cluster aligned as well. + * + * Returns: + * 0: if no allocated clusters are available at the given offset. + * *bytes is normally unchanged. It is set to 0 if the cluster + * is allocated and doesn't need COW, but doesn't have the right + * physical offset. + * + * 1: if allocated clusters that don't require a COW are available at + * the requested offset. *bytes may have decreased and describes + * the length of the area that can be written to. + * + * -errno: in error cases + */ +static int handle_copied(BlockDriverState *bs, uint64_t guest_offset, + uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) +{ + BDRVQcowState *s = bs->opaque; + int l2_index; + uint64_t cluster_offset; + uint64_t *l2_table; + unsigned int nb_clusters; + unsigned int keep_clusters; + int ret, pret; + + trace_qcow2_handle_copied(qemu_coroutine_self(), guest_offset, *host_offset, + *bytes); + + assert(*host_offset == 0 || offset_into_cluster(s, guest_offset) + == offset_into_cluster(s, *host_offset)); + + /* + * Calculate the number of clusters to look for. We stop at L2 table + * boundaries to keep things simple. + */ + nb_clusters = + size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); + + l2_index = offset_to_l2_index(s, guest_offset); + nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); + + /* Find L2 entry for the first involved cluster */ + ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); + if (ret < 0) { + return ret; + } + + cluster_offset = be64_to_cpu(l2_table[l2_index]); + + /* Check how many clusters are already allocated and don't need COW */ + if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL + && (cluster_offset & QCOW_OFLAG_COPIED)) + { + /* If a specific host_offset is required, check it */ + bool offset_matches = + (cluster_offset & L2E_OFFSET_MASK) == *host_offset; + + if (*host_offset != 0 && !offset_matches) { + *bytes = 0; + ret = 0; + goto out; + } + + /* We keep all QCOW_OFLAG_COPIED clusters */ + keep_clusters = + count_contiguous_clusters(nb_clusters, s->cluster_size, + &l2_table[l2_index], 0, + QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); + assert(keep_clusters <= nb_clusters); + + *bytes = MIN(*bytes, + keep_clusters * s->cluster_size + - offset_into_cluster(s, guest_offset)); + + ret = 1; + } else { + ret = 0; + } + + /* Cleanup */ +out: + pret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); + if (pret < 0) { + return pret; + } + + /* Only return a host offset if we actually made progress. Otherwise we + * would make requirements for handle_alloc() that it can't fulfill */ + if (ret) { + *host_offset = (cluster_offset & L2E_OFFSET_MASK) + + offset_into_cluster(s, guest_offset); + } + + return ret; +} + +/* * Allocates new clusters for the given guest_offset. * * At most *nb_clusters are allocated, and on return *nb_clusters is updated to @@ -824,16 +947,10 @@ static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, uint64_t *host_offset, unsigned int *nb_clusters) { BDRVQcowState *s = bs->opaque; - int ret; trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset, *host_offset, *nb_clusters); - ret = handle_dependencies(bs, guest_offset, nb_clusters); - if (ret < 0) { - return ret; - } - /* Allocate new clusters */ trace_qcow2_cluster_alloc_phys(qemu_coroutine_self()); if (*host_offset == 0) { @@ -845,7 +962,7 @@ static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, *host_offset = cluster_offset; return 0; } else { - ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); + int ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters); if (ret < 0) { return ret; } @@ -855,6 +972,151 @@ static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset, } /* + * Allocates new clusters for an area that either is yet unallocated or needs a + * copy on write. If *host_offset is non-zero, clusters are only allocated if + * the new allocation can match the specified host offset. + * + * Note that guest_offset may not be cluster aligned. In this case, the + * returned *host_offset points to exact byte referenced by guest_offset and + * therefore isn't cluster aligned as well. + * + * Returns: + * 0: if no clusters could be allocated. *bytes is set to 0, + * *host_offset is left unchanged. + * + * 1: if new clusters were allocated. *bytes may be decreased if the + * new allocation doesn't cover all of the requested area. + * *host_offset is updated to contain the host offset of the first + * newly allocated cluster. + * + * -errno: in error cases + */ +static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, + uint64_t *host_offset, uint64_t *bytes, QCowL2Meta **m) +{ + BDRVQcowState *s = bs->opaque; + int l2_index; + uint64_t *l2_table; + uint64_t entry; + unsigned int nb_clusters; + int ret; + + uint64_t alloc_cluster_offset; + + trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset, + *bytes); + assert(*bytes > 0); + + /* + * Calculate the number of clusters to look for. We stop at L2 table + * boundaries to keep things simple. + */ + nb_clusters = + size_to_clusters(s, offset_into_cluster(s, guest_offset) + *bytes); + + l2_index = offset_to_l2_index(s, guest_offset); + nb_clusters = MIN(nb_clusters, s->l2_size - l2_index); + + /* Find L2 entry for the first involved cluster */ + ret = get_cluster_table(bs, guest_offset, &l2_table, &l2_index); + if (ret < 0) { + return ret; + } + + entry = be64_to_cpu(l2_table[l2_index]); + + /* For the moment, overwrite compressed clusters one by one */ + if (entry & QCOW_OFLAG_COMPRESSED) { + nb_clusters = 1; + } else { + nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, l2_index); + } + + /* This function is only called when there were no non-COW clusters, so if + * we can't find any unallocated or COW clusters either, something is + * wrong with our code. */ + assert(nb_clusters > 0); + + ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); + if (ret < 0) { + return ret; + } + + /* Allocate, if necessary at a given offset in the image file */ + alloc_cluster_offset = start_of_cluster(s, *host_offset); + ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset, + &nb_clusters); + if (ret < 0) { + goto fail; + } + + /* Can't extend contiguous allocation */ + if (nb_clusters == 0) { + *bytes = 0; + return 0; + } + + /* + * Save info needed for meta data update. + * + * requested_sectors: Number of sectors from the start of the first + * newly allocated cluster to the end of the (possibly shortened + * before) write request. + * + * avail_sectors: Number of sectors from the start of the first + * newly allocated to the end of the last newly allocated cluster. + * + * nb_sectors: The number of sectors from the start of the first + * newly allocated cluster to the end of the area that the write + * request actually writes to (excluding COW at the end) + */ + int requested_sectors = + (*bytes + offset_into_cluster(s, guest_offset)) + >> BDRV_SECTOR_BITS; + int avail_sectors = nb_clusters + << (s->cluster_bits - BDRV_SECTOR_BITS); + int alloc_n_start = offset_into_cluster(s, guest_offset) + >> BDRV_SECTOR_BITS; + int nb_sectors = MIN(requested_sectors, avail_sectors); + QCowL2Meta *old_m = *m; + + *m = g_malloc0(sizeof(**m)); + + **m = (QCowL2Meta) { + .next = old_m, + + .alloc_offset = alloc_cluster_offset, + .offset = start_of_cluster(s, guest_offset), + .nb_clusters = nb_clusters, + .nb_available = nb_sectors, + + .cow_start = { + .offset = 0, + .nb_sectors = alloc_n_start, + }, + .cow_end = { + .offset = nb_sectors * BDRV_SECTOR_SIZE, + .nb_sectors = avail_sectors - nb_sectors, + }, + }; + qemu_co_queue_init(&(*m)->dependent_requests); + QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); + + *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset); + *bytes = MIN(*bytes, (nb_sectors * BDRV_SECTOR_SIZE) + - offset_into_cluster(s, guest_offset)); + assert(*bytes != 0); + + return 1; + +fail: + if (*m && (*m)->nb_clusters > 0) { + QLIST_REMOVE(*m, next_in_flight); + } + return ret; +} + +/* * alloc_cluster_offset * * For a given offset on the virtual disk, find the cluster offset in qcow2 @@ -877,161 +1139,110 @@ int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset, int n_start, int n_end, int *num, uint64_t *host_offset, QCowL2Meta **m) { BDRVQcowState *s = bs->opaque; - int l2_index, ret, sectors; - uint64_t *l2_table; - unsigned int nb_clusters, keep_clusters; + uint64_t start, remaining; uint64_t cluster_offset; + uint64_t cur_bytes; + int ret; trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset, n_start, n_end); - /* Find L2 entry for the first involved cluster */ -again: - ret = get_cluster_table(bs, offset, &l2_table, &l2_index); - if (ret < 0) { - return ret; - } - - /* - * Calculate the number of clusters to look for. We stop at L2 table - * boundaries to keep things simple. - */ - nb_clusters = MIN(size_to_clusters(s, n_end << BDRV_SECTOR_BITS), - s->l2_size - l2_index); + assert(n_start * BDRV_SECTOR_SIZE == offset_into_cluster(s, offset)); + offset = start_of_cluster(s, offset); - cluster_offset = be64_to_cpu(l2_table[l2_index]); +again: + start = offset + (n_start << BDRV_SECTOR_BITS); + remaining = (n_end - n_start) << BDRV_SECTOR_BITS; + cluster_offset = 0; + *host_offset = 0; + cur_bytes = 0; + *m = NULL; - /* - * Check how many clusters are already allocated and don't need COW, and how - * many need a new allocation. - */ - if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL - && (cluster_offset & QCOW_OFLAG_COPIED)) - { - /* We keep all QCOW_OFLAG_COPIED clusters */ - keep_clusters = - count_contiguous_clusters(nb_clusters, s->cluster_size, - &l2_table[l2_index], 0, - QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO); - assert(keep_clusters <= nb_clusters); - nb_clusters -= keep_clusters; - } else { - keep_clusters = 0; - cluster_offset = 0; - } + while (true) { - if (nb_clusters > 0) { - /* For the moment, overwrite compressed clusters one by one */ - uint64_t entry = be64_to_cpu(l2_table[l2_index + keep_clusters]); - if (entry & QCOW_OFLAG_COMPRESSED) { - nb_clusters = 1; - } else { - nb_clusters = count_cow_clusters(s, nb_clusters, l2_table, - l2_index + keep_clusters); + if (!*host_offset) { + *host_offset = start_of_cluster(s, cluster_offset); } - } - cluster_offset &= L2E_OFFSET_MASK; + assert(remaining >= cur_bytes); - /* - * The L2 table isn't used any more after this. As long as the cache works - * synchronously, it's important to release it before calling - * do_alloc_cluster_offset, which may yield if we need to wait for another - * request to complete. If we still had the reference, we could use up the - * whole cache with sleeping requests. - */ - ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table); - if (ret < 0) { - return ret; - } + start += cur_bytes; + remaining -= cur_bytes; + cluster_offset += cur_bytes; - /* If there is something left to allocate, do that now */ - if (nb_clusters > 0) { - uint64_t alloc_offset; - uint64_t alloc_cluster_offset; - uint64_t keep_bytes = keep_clusters * s->cluster_size; - - /* Calculate start and size of allocation */ - alloc_offset = offset + keep_bytes; - - if (keep_clusters == 0) { - alloc_cluster_offset = 0; - } else { - alloc_cluster_offset = cluster_offset + keep_bytes; + if (remaining == 0) { + break; } - /* Allocate, if necessary at a given offset in the image file */ - ret = do_alloc_cluster_offset(bs, alloc_offset, &alloc_cluster_offset, - &nb_clusters); + cur_bytes = remaining; + + /* + * Now start gathering as many contiguous clusters as possible: + * + * 1. Check for overlaps with in-flight allocations + * + * a) Overlap not in the first cluster -> shorten this request and + * let the caller handle the rest in its next loop iteration. + * + * b) Real overlaps of two requests. Yield and restart the search + * for contiguous clusters (the situation could have changed + * while we were sleeping) + * + * c) TODO: Request starts in the same cluster as the in-flight + * allocation ends. Shorten the COW of the in-fight allocation, + * set cluster_offset to write to the same cluster and set up + * the right synchronisation between the in-flight request and + * the new one. + */ + ret = handle_dependencies(bs, start, &cur_bytes, m); if (ret == -EAGAIN) { + /* Currently handle_dependencies() doesn't yield if we already had + * an allocation. If it did, we would have to clean up the L2Meta + * structs before starting over. */ + assert(*m == NULL); goto again; } else if (ret < 0) { - goto fail; + return ret; + } else if (cur_bytes == 0) { + break; + } else { + /* handle_dependencies() may have decreased cur_bytes (shortened + * the allocations below) so that the next dependency is processed + * correctly during the next loop iteration. */ } - /* save info needed for meta data update */ - if (nb_clusters > 0) { - /* - * requested_sectors: Number of sectors from the start of the first - * newly allocated cluster to the end of the (possibly shortened - * before) write request. - * - * avail_sectors: Number of sectors from the start of the first - * newly allocated to the end of the last newly allocated cluster. - * - * nb_sectors: The number of sectors from the start of the first - * newly allocated cluster to the end of the aread that the write - * request actually writes to (excluding COW at the end) - */ - int requested_sectors = n_end - keep_clusters * s->cluster_sectors; - int avail_sectors = nb_clusters - << (s->cluster_bits - BDRV_SECTOR_BITS); - int alloc_n_start = keep_clusters == 0 ? n_start : 0; - int nb_sectors = MIN(requested_sectors, avail_sectors); - - if (keep_clusters == 0) { - cluster_offset = alloc_cluster_offset; - } - - *m = g_malloc0(sizeof(**m)); - - **m = (QCowL2Meta) { - .alloc_offset = alloc_cluster_offset, - .offset = alloc_offset & ~(s->cluster_size - 1), - .nb_clusters = nb_clusters, - .nb_available = nb_sectors, - - .cow_start = { - .offset = 0, - .nb_sectors = alloc_n_start, - }, - .cow_end = { - .offset = nb_sectors * BDRV_SECTOR_SIZE, - .nb_sectors = avail_sectors - nb_sectors, - }, - }; - qemu_co_queue_init(&(*m)->dependent_requests); - QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight); + /* + * 2. Count contiguous COPIED clusters. + */ + ret = handle_copied(bs, start, &cluster_offset, &cur_bytes, m); + if (ret < 0) { + return ret; + } else if (ret) { + continue; + } else if (cur_bytes == 0) { + break; } - } - /* Some cleanup work */ - sectors = (keep_clusters + nb_clusters) << (s->cluster_bits - 9); - if (sectors > n_end) { - sectors = n_end; + /* + * 3. If the request still hasn't completed, allocate new clusters, + * considering any cluster_offset of steps 1c or 2. + */ + ret = handle_alloc(bs, start, &cluster_offset, &cur_bytes, m); + if (ret < 0) { + return ret; + } else if (ret) { + continue; + } else { + assert(cur_bytes == 0); + break; + } } - assert(sectors > n_start); - *num = sectors - n_start; - *host_offset = cluster_offset; + *num = (n_end - n_start) - (remaining >> BDRV_SECTOR_BITS); + assert(*num > 0); + assert(*host_offset != 0); return 0; - -fail: - if (*m && (*m)->nb_clusters > 0) { - QLIST_REMOVE(*m, next_in_flight); - } - return ret; } static int decompress_buffer(uint8_t *out_buf, int out_buf_size, |