Commit 095a9c58ce12afeeb90c225ed087017c307bc0fd
1 parent
52d893ec
qcow2: Aggregate same type clusters (Laurent Vivier)
Modify get_cluster_offset(), alloc_cluster_offset() to specify how many clusters we want. Signed-off-by: Laurent Vivier <Laurent.Vivier@bull.net> Signed-off-by: Anthony Liguori <aliguori@us.ibm.com> git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@5006 c046a42c-6fe2-441c-8c8c-71466251a162
Showing
1 changed file
with
181 additions
and
53 deletions
block-qcow2.c
... | ... | @@ -52,6 +52,8 @@ |
52 | 52 | #define QCOW_CRYPT_NONE 0 |
53 | 53 | #define QCOW_CRYPT_AES 1 |
54 | 54 | |
55 | +#define QCOW_MAX_CRYPT_CLUSTERS 32 | |
56 | + | |
55 | 57 | /* indicate that the refcount of the referenced cluster is exactly one. */ |
56 | 58 | #define QCOW_OFLAG_COPIED (1LL << 63) |
57 | 59 | /* indicate that the cluster is compressed (they never have the copied flag) */ |
... | ... | @@ -263,7 +265,8 @@ static int qcow_open(BlockDriverState *bs, const char *filename, int flags) |
263 | 265 | if (!s->cluster_cache) |
264 | 266 | goto fail; |
265 | 267 | /* one more sector for decompressed data alignment */ |
266 | - s->cluster_data = qemu_malloc(s->cluster_size + 512); | |
268 | + s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size | |
269 | + + 512); | |
267 | 270 | if (!s->cluster_data) |
268 | 271 | goto fail; |
269 | 272 | s->cluster_cache_offset = -1; |
... | ... | @@ -612,43 +615,98 @@ static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index) |
612 | 615 | * For a given offset of the disk image, return cluster offset in |
613 | 616 | * qcow2 file. |
614 | 617 | * |
618 | + * on entry, *num is the number of contiguous clusters we'd like to | |
619 | + * access following offset. | |
620 | + * | |
621 | + * on exit, *num is the number of contiguous clusters we can read. | |
622 | + * | |
615 | 623 | * Return 1, if the offset is found |
616 | 624 | * Return 0, otherwise. |
617 | 625 | * |
618 | 626 | */ |
619 | 627 | |
620 | -static uint64_t get_cluster_offset(BlockDriverState *bs, uint64_t offset) | |
628 | +static uint64_t get_cluster_offset(BlockDriverState *bs, | |
629 | + uint64_t offset, int *num) | |
621 | 630 | { |
622 | 631 | BDRVQcowState *s = bs->opaque; |
623 | 632 | int l1_index, l2_index; |
624 | - uint64_t l2_offset, *l2_table, cluster_offset; | |
633 | + uint64_t l2_offset, *l2_table, cluster_offset, next; | |
634 | + int l1_bits; | |
635 | + int index_in_cluster, nb_available, nb_needed; | |
636 | + | |
637 | + index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1); | |
638 | + nb_needed = *num + index_in_cluster; | |
639 | + | |
640 | + l1_bits = s->l2_bits + s->cluster_bits; | |
641 | + | |
642 | + /* compute how many bytes there are between the offset and | |
643 | + * and the end of the l1 entry | |
644 | + */ | |
645 | + | |
646 | + nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1)); | |
647 | + | |
648 | + /* compute the number of available sectors */ | |
649 | + | |
650 | + nb_available = (nb_available >> 9) + index_in_cluster; | |
651 | + | |
652 | + cluster_offset = 0; | |
625 | 653 | |
626 | 654 | /* seek the the l2 offset in the l1 table */ |
627 | 655 | |
628 | - l1_index = offset >> (s->l2_bits + s->cluster_bits); | |
656 | + l1_index = offset >> l1_bits; | |
629 | 657 | if (l1_index >= s->l1_size) |
630 | - return 0; | |
658 | + goto out; | |
631 | 659 | |
632 | 660 | l2_offset = s->l1_table[l1_index]; |
633 | 661 | |
634 | 662 | /* seek the l2 table of the given l2 offset */ |
635 | 663 | |
636 | 664 | if (!l2_offset) |
637 | - return 0; | |
665 | + goto out; | |
638 | 666 | |
639 | 667 | /* load the l2 table in memory */ |
640 | 668 | |
641 | 669 | l2_offset &= ~QCOW_OFLAG_COPIED; |
642 | 670 | l2_table = l2_load(bs, l2_offset); |
643 | 671 | if (l2_table == NULL) |
644 | - return 0; | |
672 | + goto out; | |
645 | 673 | |
646 | 674 | /* find the cluster offset for the given disk offset */ |
647 | 675 | |
648 | 676 | l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1); |
649 | 677 | cluster_offset = be64_to_cpu(l2_table[l2_index]); |
678 | + nb_available = s->cluster_sectors; | |
679 | + l2_index++; | |
680 | + | |
681 | + if (!cluster_offset) { | |
650 | 682 | |
651 | - return cluster_offset & ~QCOW_OFLAG_COPIED; | |
683 | + /* how many empty clusters ? */ | |
684 | + | |
685 | + while (nb_available < nb_needed && !l2_table[l2_index]) { | |
686 | + l2_index++; | |
687 | + nb_available += s->cluster_sectors; | |
688 | + } | |
689 | + } else { | |
690 | + | |
691 | + /* how many allocated clusters ? */ | |
692 | + | |
693 | + cluster_offset &= ~QCOW_OFLAG_COPIED; | |
694 | + while (nb_available < nb_needed) { | |
695 | + next = be64_to_cpu(l2_table[l2_index]) & ~QCOW_OFLAG_COPIED; | |
696 | + if (next != cluster_offset + (nb_available << 9)) | |
697 | + break; | |
698 | + l2_index++; | |
699 | + nb_available += s->cluster_sectors; | |
700 | + } | |
701 | + } | |
702 | + | |
703 | +out: | |
704 | + if (nb_available > nb_needed) | |
705 | + nb_available = nb_needed; | |
706 | + | |
707 | + *num = nb_available - index_in_cluster; | |
708 | + | |
709 | + return cluster_offset; | |
652 | 710 | } |
653 | 711 | |
654 | 712 | /* |
... | ... | @@ -659,13 +717,10 @@ static uint64_t get_cluster_offset(BlockDriverState *bs, uint64_t offset) |
659 | 717 | */ |
660 | 718 | |
661 | 719 | static void free_any_clusters(BlockDriverState *bs, |
662 | - uint64_t cluster_offset) | |
720 | + uint64_t cluster_offset, int nb_clusters) | |
663 | 721 | { |
664 | 722 | BDRVQcowState *s = bs->opaque; |
665 | 723 | |
666 | - if (cluster_offset == 0) | |
667 | - return; | |
668 | - | |
669 | 724 | /* free the cluster */ |
670 | 725 | |
671 | 726 | if (cluster_offset & QCOW_OFLAG_COMPRESSED) { |
... | ... | @@ -677,7 +732,9 @@ static void free_any_clusters(BlockDriverState *bs, |
677 | 732 | return; |
678 | 733 | } |
679 | 734 | |
680 | - free_clusters(bs, cluster_offset, s->cluster_size); | |
735 | + free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits); | |
736 | + | |
737 | + return; | |
681 | 738 | } |
682 | 739 | |
683 | 740 | /* |
... | ... | @@ -768,7 +825,8 @@ static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs, |
768 | 825 | if (cluster_offset & QCOW_OFLAG_COPIED) |
769 | 826 | return cluster_offset & ~QCOW_OFLAG_COPIED; |
770 | 827 | |
771 | - free_any_clusters(bs, cluster_offset); | |
828 | + if (cluster_offset) | |
829 | + free_any_clusters(bs, cluster_offset, 1); | |
772 | 830 | |
773 | 831 | cluster_offset = alloc_bytes(bs, compressed_size); |
774 | 832 | nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) - |
... | ... | @@ -806,68 +864,136 @@ static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs, |
806 | 864 | |
807 | 865 | static uint64_t alloc_cluster_offset(BlockDriverState *bs, |
808 | 866 | uint64_t offset, |
809 | - int n_start, int n_end) | |
867 | + int n_start, int n_end, | |
868 | + int *num) | |
810 | 869 | { |
811 | 870 | BDRVQcowState *s = bs->opaque; |
812 | 871 | int l2_index, ret; |
813 | 872 | uint64_t l2_offset, *l2_table, cluster_offset; |
873 | + int nb_available, nb_clusters, i; | |
874 | + uint64_t start_sect, current; | |
814 | 875 | |
815 | 876 | ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index); |
816 | 877 | if (ret == 0) |
817 | 878 | return 0; |
818 | 879 | |
880 | + nb_clusters = ((n_end << 9) + s->cluster_size - 1) >> | |
881 | + s->cluster_bits; | |
882 | + if (nb_clusters > s->l2_size - l2_index) | |
883 | + nb_clusters = s->l2_size - l2_index; | |
884 | + | |
819 | 885 | cluster_offset = be64_to_cpu(l2_table[l2_index]); |
820 | - if (cluster_offset & QCOW_OFLAG_COPIED) | |
821 | - return cluster_offset & ~QCOW_OFLAG_COPIED; | |
822 | 886 | |
823 | - free_any_clusters(bs, cluster_offset); | |
887 | + /* We keep all QCOW_OFLAG_COPIED clusters */ | |
888 | + | |
889 | + if (cluster_offset & QCOW_OFLAG_COPIED) { | |
890 | + | |
891 | + for (i = 1; i < nb_clusters; i++) { | |
892 | + current = be64_to_cpu(l2_table[l2_index + i]); | |
893 | + if (cluster_offset + (i << s->cluster_bits) != current) | |
894 | + break; | |
895 | + } | |
896 | + nb_clusters = i; | |
897 | + | |
898 | + nb_available = nb_clusters << (s->cluster_bits - 9); | |
899 | + if (nb_available > n_end) | |
900 | + nb_available = n_end; | |
901 | + | |
902 | + cluster_offset &= ~QCOW_OFLAG_COPIED; | |
903 | + | |
904 | + goto out; | |
905 | + } | |
906 | + | |
907 | + /* for the moment, multiple compressed clusters are not managed */ | |
908 | + | |
909 | + if (cluster_offset & QCOW_OFLAG_COMPRESSED) | |
910 | + nb_clusters = 1; | |
911 | + | |
912 | + /* how many empty or how many to free ? */ | |
913 | + | |
914 | + if (!cluster_offset) { | |
915 | + | |
916 | + /* how many free clusters ? */ | |
917 | + | |
918 | + i = 1; | |
919 | + while (i < nb_clusters && | |
920 | + l2_table[l2_index + i] == 0) { | |
921 | + i++; | |
922 | + } | |
923 | + nb_clusters = i; | |
924 | + | |
925 | + } else { | |
926 | + | |
927 | + /* how many contiguous clusters ? */ | |
928 | + | |
929 | + for (i = 1; i < nb_clusters; i++) { | |
930 | + current = be64_to_cpu(l2_table[l2_index + i]); | |
931 | + if (cluster_offset + (i << s->cluster_bits) != current) | |
932 | + break; | |
933 | + } | |
934 | + nb_clusters = i; | |
935 | + | |
936 | + free_any_clusters(bs, cluster_offset, i); | |
937 | + } | |
824 | 938 | |
825 | 939 | /* allocate a new cluster */ |
826 | 940 | |
827 | - cluster_offset = alloc_clusters(bs, s->cluster_size); | |
941 | + cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size); | |
828 | 942 | |
829 | 943 | /* we must initialize the cluster content which won't be |
830 | 944 | written */ |
831 | 945 | |
832 | - if ((n_end - n_start) < s->cluster_sectors) { | |
833 | - uint64_t start_sect; | |
946 | + nb_available = nb_clusters << (s->cluster_bits - 9); | |
947 | + if (nb_available > n_end) | |
948 | + nb_available = n_end; | |
834 | 949 | |
835 | - start_sect = (offset & ~(s->cluster_size - 1)) >> 9; | |
836 | - ret = copy_sectors(bs, start_sect, | |
837 | - cluster_offset, 0, n_start); | |
950 | + /* copy content of unmodified sectors */ | |
951 | + | |
952 | + start_sect = (offset & ~(s->cluster_size - 1)) >> 9; | |
953 | + if (n_start) { | |
954 | + ret = copy_sectors(bs, start_sect, cluster_offset, 0, n_start); | |
838 | 955 | if (ret < 0) |
839 | 956 | return 0; |
840 | - ret = copy_sectors(bs, start_sect, | |
841 | - cluster_offset, n_end, s->cluster_sectors); | |
957 | + } | |
958 | + | |
959 | + if (nb_available & (s->cluster_sectors - 1)) { | |
960 | + uint64_t end = nb_available & ~(uint64_t)(s->cluster_sectors - 1); | |
961 | + ret = copy_sectors(bs, start_sect + end, | |
962 | + cluster_offset + (end << 9), | |
963 | + nb_available - end, | |
964 | + s->cluster_sectors); | |
842 | 965 | if (ret < 0) |
843 | 966 | return 0; |
844 | 967 | } |
845 | 968 | |
846 | 969 | /* update L2 table */ |
847 | 970 | |
848 | - l2_table[l2_index] = cpu_to_be64(cluster_offset | QCOW_OFLAG_COPIED); | |
971 | + for (i = 0; i < nb_clusters; i++) | |
972 | + l2_table[l2_index + i] = cpu_to_be64((cluster_offset + | |
973 | + (i << s->cluster_bits)) | | |
974 | + QCOW_OFLAG_COPIED); | |
975 | + | |
849 | 976 | if (bdrv_pwrite(s->hd, |
850 | 977 | l2_offset + l2_index * sizeof(uint64_t), |
851 | 978 | l2_table + l2_index, |
852 | - sizeof(uint64_t)) != sizeof(uint64_t)) | |
979 | + nb_clusters * sizeof(uint64_t)) != | |
980 | + nb_clusters * sizeof(uint64_t)) | |
853 | 981 | return 0; |
854 | 982 | |
983 | +out: | |
984 | + *num = nb_available - n_start; | |
985 | + | |
855 | 986 | return cluster_offset; |
856 | 987 | } |
857 | 988 | |
858 | 989 | static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num, |
859 | 990 | int nb_sectors, int *pnum) |
860 | 991 | { |
861 | - BDRVQcowState *s = bs->opaque; | |
862 | - int index_in_cluster, n; | |
863 | 992 | uint64_t cluster_offset; |
864 | 993 | |
865 | - cluster_offset = get_cluster_offset(bs, sector_num << 9); | |
866 | - index_in_cluster = sector_num & (s->cluster_sectors - 1); | |
867 | - n = s->cluster_sectors - index_in_cluster; | |
868 | - if (n > nb_sectors) | |
869 | - n = nb_sectors; | |
870 | - *pnum = n; | |
994 | + *pnum = nb_sectors; | |
995 | + cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum); | |
996 | + | |
871 | 997 | return (cluster_offset != 0); |
872 | 998 | } |
873 | 999 | |
... | ... | @@ -944,11 +1070,9 @@ static int qcow_read(BlockDriverState *bs, int64_t sector_num, |
944 | 1070 | uint64_t cluster_offset; |
945 | 1071 | |
946 | 1072 | while (nb_sectors > 0) { |
947 | - cluster_offset = get_cluster_offset(bs, sector_num << 9); | |
1073 | + n = nb_sectors; | |
1074 | + cluster_offset = get_cluster_offset(bs, sector_num << 9, &n); | |
948 | 1075 | index_in_cluster = sector_num & (s->cluster_sectors - 1); |
949 | - n = s->cluster_sectors - index_in_cluster; | |
950 | - if (n > nb_sectors) | |
951 | - n = nb_sectors; | |
952 | 1076 | if (!cluster_offset) { |
953 | 1077 | if (bs->backing_hd) { |
954 | 1078 | /* read from the base image */ |
... | ... | @@ -987,15 +1111,17 @@ static int qcow_write(BlockDriverState *bs, int64_t sector_num, |
987 | 1111 | BDRVQcowState *s = bs->opaque; |
988 | 1112 | int ret, index_in_cluster, n; |
989 | 1113 | uint64_t cluster_offset; |
1114 | + int n_end; | |
990 | 1115 | |
991 | 1116 | while (nb_sectors > 0) { |
992 | 1117 | index_in_cluster = sector_num & (s->cluster_sectors - 1); |
993 | - n = s->cluster_sectors - index_in_cluster; | |
994 | - if (n > nb_sectors) | |
995 | - n = nb_sectors; | |
1118 | + n_end = index_in_cluster + nb_sectors; | |
1119 | + if (s->crypt_method && | |
1120 | + n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) | |
1121 | + n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors; | |
996 | 1122 | cluster_offset = alloc_cluster_offset(bs, sector_num << 9, |
997 | 1123 | index_in_cluster, |
998 | - index_in_cluster + n); | |
1124 | + n_end, &n); | |
999 | 1125 | if (!cluster_offset) |
1000 | 1126 | return -1; |
1001 | 1127 | if (s->crypt_method) { |
... | ... | @@ -1068,11 +1194,9 @@ static void qcow_aio_read_cb(void *opaque, int ret) |
1068 | 1194 | } |
1069 | 1195 | |
1070 | 1196 | /* prepare next AIO request */ |
1071 | - acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9); | |
1197 | + acb->n = acb->nb_sectors; | |
1198 | + acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n); | |
1072 | 1199 | index_in_cluster = acb->sector_num & (s->cluster_sectors - 1); |
1073 | - acb->n = s->cluster_sectors - index_in_cluster; | |
1074 | - if (acb->n > acb->nb_sectors) | |
1075 | - acb->n = acb->nb_sectors; | |
1076 | 1200 | |
1077 | 1201 | if (!acb->cluster_offset) { |
1078 | 1202 | if (bs->backing_hd) { |
... | ... | @@ -1152,6 +1276,7 @@ static void qcow_aio_write_cb(void *opaque, int ret) |
1152 | 1276 | int index_in_cluster; |
1153 | 1277 | uint64_t cluster_offset; |
1154 | 1278 | const uint8_t *src_buf; |
1279 | + int n_end; | |
1155 | 1280 | |
1156 | 1281 | acb->hd_aiocb = NULL; |
1157 | 1282 | |
... | ... | @@ -1174,19 +1299,22 @@ static void qcow_aio_write_cb(void *opaque, int ret) |
1174 | 1299 | } |
1175 | 1300 | |
1176 | 1301 | index_in_cluster = acb->sector_num & (s->cluster_sectors - 1); |
1177 | - acb->n = s->cluster_sectors - index_in_cluster; | |
1178 | - if (acb->n > acb->nb_sectors) | |
1179 | - acb->n = acb->nb_sectors; | |
1302 | + n_end = index_in_cluster + acb->nb_sectors; | |
1303 | + if (s->crypt_method && | |
1304 | + n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors) | |
1305 | + n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors; | |
1306 | + | |
1180 | 1307 | cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9, |
1181 | 1308 | index_in_cluster, |
1182 | - index_in_cluster + acb->n); | |
1309 | + n_end, &acb->n); | |
1183 | 1310 | if (!cluster_offset || (cluster_offset & 511) != 0) { |
1184 | 1311 | ret = -EIO; |
1185 | 1312 | goto fail; |
1186 | 1313 | } |
1187 | 1314 | if (s->crypt_method) { |
1188 | 1315 | if (!acb->cluster_data) { |
1189 | - acb->cluster_data = qemu_mallocz(s->cluster_size); | |
1316 | + acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS * | |
1317 | + s->cluster_size); | |
1190 | 1318 | if (!acb->cluster_data) { |
1191 | 1319 | ret = -ENOMEM; |
1192 | 1320 | goto fail; | ... | ... |