| 1 | /* |
| 2 | * QEMU Enhanced Disk Format Table I/O |
| 3 | * |
| 4 | * Copyright IBM, Corp. 2010 |
| 5 | * |
| 6 | * Authors: |
| 7 | * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com> |
| 8 | * Anthony Liguori <aliguori@us.ibm.com> |
| 9 | * |
| 10 | * This work is licensed under the terms of the GNU LGPL, version 2 or later. |
| 11 | * See the COPYING.LIB file in the top-level directory. |
| 12 | * |
| 13 | */ |
| 14 | |
| 15 | #include "qemu/osdep.h" |
| 16 | #include "block/block-io.h" |
| 17 | #include "trace.h" |
| 18 | #include "qemu/sockets.h" /* for EINPROGRESS on Windows */ |
| 19 | #include "qed.h" |
| 20 | #include "qemu/bswap.h" |
| 21 | #include "qemu/memalign.h" |
| 22 | |
| 23 | /* Called with table_lock held. */ |
| 24 | static int coroutine_fn GRAPH_RDLOCK |
| 25 | qed_read_table(BDRVQEDState *s, uint64_t offset, QEDTable *table) |
| 26 | { |
| 27 | unsigned int bytes = s->header.cluster_size * s->header.table_size; |
| 28 | |
| 29 | int noffsets; |
| 30 | int i, ret; |
| 31 | |
| 32 | trace_qed_read_table(s, offset, table); |
| 33 | |
| 34 | qemu_co_mutex_unlock(&s->table_lock); |
| 35 | ret = bdrv_co_pread(s->bs->file, offset, bytes, table->offsets, 0); |
| 36 | qemu_co_mutex_lock(&s->table_lock); |
| 37 | if (ret < 0) { |
| 38 | goto out; |
| 39 | } |
| 40 | |
| 41 | /* Byteswap offsets */ |
| 42 | noffsets = bytes / sizeof(uint64_t); |
| 43 | for (i = 0; i < noffsets; i++) { |
| 44 | table->offsets[i] = le64_to_cpu(table->offsets[i]); |
| 45 | } |
| 46 | |
| 47 | ret = 0; |
| 48 | out: |
| 49 | /* Completion */ |
| 50 | trace_qed_read_table_cb(s, table, ret); |
| 51 | return ret; |
| 52 | } |
| 53 | |
| 54 | /** |
| 55 | * Write out an updated part or all of a table |
| 56 | * |
| 57 | * @s: QED state |
| 58 | * @offset: Offset of table in image file, in bytes |
| 59 | * @table: Table |
| 60 | * @index: Index of first element |
| 61 | * @n: Number of elements |
| 62 | * @flush: Whether or not to sync to disk |
| 63 | * |
| 64 | * Called with table_lock held. |
| 65 | */ |
| 66 | static int coroutine_fn GRAPH_RDLOCK |
| 67 | qed_write_table(BDRVQEDState *s, uint64_t offset, QEDTable *table, |
| 68 | unsigned int index, unsigned int n, bool flush) |
| 69 | { |
| 70 | unsigned int sector_mask = BDRV_SECTOR_SIZE / sizeof(uint64_t) - 1; |
| 71 | unsigned int start, end, i; |
| 72 | QEDTable *new_table; |
| 73 | size_t len_bytes; |
| 74 | int ret; |
| 75 | |
| 76 | trace_qed_write_table(s, offset, table, index, n); |
| 77 | |
| 78 | /* Calculate indices of the first and one after last elements */ |
| 79 | start = index & ~sector_mask; |
| 80 | end = (index + n + sector_mask) & ~sector_mask; |
| 81 | |
| 82 | len_bytes = (end - start) * sizeof(uint64_t); |
| 83 | |
| 84 | new_table = qemu_blockalign(s->bs, len_bytes); |
| 85 | |
| 86 | /* Byteswap table */ |
| 87 | for (i = start; i < end; i++) { |
| 88 | uint64_t le_offset = cpu_to_le64(table->offsets[i]); |
| 89 | new_table->offsets[i - start] = le_offset; |
| 90 | } |
| 91 | |
| 92 | /* Adjust for offset into table */ |
| 93 | offset += start * sizeof(uint64_t); |
| 94 | |
| 95 | qemu_co_mutex_unlock(&s->table_lock); |
| 96 | ret = bdrv_co_pwrite(s->bs->file, offset, len_bytes, new_table->offsets, 0); |
| 97 | qemu_co_mutex_lock(&s->table_lock); |
| 98 | trace_qed_write_table_cb(s, table, flush, ret); |
| 99 | if (ret < 0) { |
| 100 | goto out; |
| 101 | } |
| 102 | |
| 103 | if (flush) { |
| 104 | ret = bdrv_co_flush(s->bs); |
| 105 | if (ret < 0) { |
| 106 | goto out; |
| 107 | } |
| 108 | } |
| 109 | |
| 110 | ret = 0; |
| 111 | out: |
| 112 | qemu_vfree(new_table); |
| 113 | return ret; |
| 114 | } |
| 115 | |
| 116 | int coroutine_fn qed_read_l1_table_sync(BDRVQEDState *s) |
| 117 | { |
| 118 | return qed_read_table(s, s->header.l1_table_offset, s->l1_table); |
| 119 | } |
| 120 | |
| 121 | /* Called with table_lock held. */ |
| 122 | int coroutine_fn qed_write_l1_table(BDRVQEDState *s, unsigned int index, |
| 123 | unsigned int n) |
| 124 | { |
| 125 | BLKDBG_CO_EVENT(s->bs->file, BLKDBG_L1_UPDATE); |
| 126 | return qed_write_table(s, s->header.l1_table_offset, |
| 127 | s->l1_table, index, n, false); |
| 128 | } |
| 129 | |
| 130 | int coroutine_fn qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index, |
| 131 | unsigned int n) |
| 132 | { |
| 133 | return qed_write_l1_table(s, index, n); |
| 134 | } |
| 135 | |
| 136 | /* Called with table_lock held. */ |
| 137 | int coroutine_fn qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, |
| 138 | uint64_t offset) |
| 139 | { |
| 140 | int ret; |
| 141 | |
| 142 | qed_unref_l2_cache_entry(request->l2_table); |
| 143 | |
| 144 | /* Check for cached L2 entry */ |
| 145 | request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset); |
| 146 | if (request->l2_table) { |
| 147 | return 0; |
| 148 | } |
| 149 | |
| 150 | request->l2_table = qed_alloc_l2_cache_entry(&s->l2_cache); |
| 151 | request->l2_table->table = qed_alloc_table(s); |
| 152 | |
| 153 | BLKDBG_CO_EVENT(s->bs->file, BLKDBG_L2_LOAD); |
| 154 | ret = qed_read_table(s, offset, request->l2_table->table); |
| 155 | |
| 156 | if (ret) { |
| 157 | /* can't trust loaded L2 table anymore */ |
| 158 | qed_unref_l2_cache_entry(request->l2_table); |
| 159 | request->l2_table = NULL; |
| 160 | } else { |
| 161 | request->l2_table->offset = offset; |
| 162 | |
| 163 | qed_commit_l2_cache_entry(&s->l2_cache, request->l2_table); |
| 164 | |
| 165 | /* This is guaranteed to succeed because we just committed the entry |
| 166 | * to the cache. |
| 167 | */ |
| 168 | request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset); |
| 169 | assert(request->l2_table != NULL); |
| 170 | } |
| 171 | |
| 172 | return ret; |
| 173 | } |
| 174 | |
| 175 | int coroutine_fn qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request, |
| 176 | uint64_t offset) |
| 177 | { |
| 178 | return qed_read_l2_table(s, request, offset); |
| 179 | } |
| 180 | |
| 181 | /* Called with table_lock held. */ |
| 182 | int coroutine_fn qed_write_l2_table(BDRVQEDState *s, QEDRequest *request, |
| 183 | unsigned int index, unsigned int n, |
| 184 | bool flush) |
| 185 | { |
| 186 | BLKDBG_CO_EVENT(s->bs->file, BLKDBG_L2_UPDATE); |
| 187 | return qed_write_table(s, request->l2_table->offset, |
| 188 | request->l2_table->table, index, n, flush); |
| 189 | } |
| 190 | |
| 191 | int coroutine_fn qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request, |
| 192 | unsigned int index, unsigned int n, |
| 193 | bool flush) |
| 194 | { |
| 195 | return qed_write_l2_table(s, request, index, n, flush); |
| 196 | } |