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 "trace.h" 17 #include "qemu/sockets.h" /* for EINPROGRESS on Windows */ 18 #include "qed.h" 19 #include "qemu/bswap.h" 20 21 typedef struct { 22 GenericCB gencb; 23 BDRVQEDState *s; 24 QEDTable *table; 25 26 struct iovec iov; 27 QEMUIOVector qiov; 28 } QEDReadTableCB; 29 30 static void qed_read_table_cb(void *opaque, int ret) 31 { 32 QEDReadTableCB *read_table_cb = opaque; 33 QEDTable *table = read_table_cb->table; 34 int noffsets = read_table_cb->qiov.size / sizeof(uint64_t); 35 int i; 36 37 /* Handle I/O error */ 38 if (ret) { 39 goto out; 40 } 41 42 /* Byteswap offsets */ 43 for (i = 0; i < noffsets; i++) { 44 table->offsets[i] = le64_to_cpu(table->offsets[i]); 45 } 46 47 out: 48 /* Completion */ 49 trace_qed_read_table_cb(read_table_cb->s, read_table_cb->table, ret); 50 gencb_complete(&read_table_cb->gencb, ret); 51 } 52 53 static void qed_read_table(BDRVQEDState *s, uint64_t offset, QEDTable *table, 54 BlockCompletionFunc *cb, void *opaque) 55 { 56 QEDReadTableCB *read_table_cb = gencb_alloc(sizeof(*read_table_cb), 57 cb, opaque); 58 QEMUIOVector *qiov = &read_table_cb->qiov; 59 60 trace_qed_read_table(s, offset, table); 61 62 read_table_cb->s = s; 63 read_table_cb->table = table; 64 read_table_cb->iov.iov_base = table->offsets, 65 read_table_cb->iov.iov_len = s->header.cluster_size * s->header.table_size, 66 67 qemu_iovec_init_external(qiov, &read_table_cb->iov, 1); 68 bdrv_aio_readv(s->bs->file, offset / BDRV_SECTOR_SIZE, qiov, 69 qiov->size / BDRV_SECTOR_SIZE, 70 qed_read_table_cb, read_table_cb); 71 } 72 73 typedef struct { 74 GenericCB gencb; 75 BDRVQEDState *s; 76 QEDTable *orig_table; 77 QEDTable *table; 78 bool flush; /* flush after write? */ 79 80 struct iovec iov; 81 QEMUIOVector qiov; 82 } QEDWriteTableCB; 83 84 static void qed_write_table_cb(void *opaque, int ret) 85 { 86 QEDWriteTableCB *write_table_cb = opaque; 87 88 trace_qed_write_table_cb(write_table_cb->s, 89 write_table_cb->orig_table, 90 write_table_cb->flush, 91 ret); 92 93 if (ret) { 94 goto out; 95 } 96 97 if (write_table_cb->flush) { 98 /* We still need to flush first */ 99 write_table_cb->flush = false; 100 bdrv_aio_flush(write_table_cb->s->bs, qed_write_table_cb, 101 write_table_cb); 102 return; 103 } 104 105 out: 106 qemu_vfree(write_table_cb->table); 107 gencb_complete(&write_table_cb->gencb, ret); 108 } 109 110 /** 111 * Write out an updated part or all of a table 112 * 113 * @s: QED state 114 * @offset: Offset of table in image file, in bytes 115 * @table: Table 116 * @index: Index of first element 117 * @n: Number of elements 118 * @flush: Whether or not to sync to disk 119 * @cb: Completion function 120 * @opaque: Argument for completion function 121 */ 122 static void qed_write_table(BDRVQEDState *s, uint64_t offset, QEDTable *table, 123 unsigned int index, unsigned int n, bool flush, 124 BlockCompletionFunc *cb, void *opaque) 125 { 126 QEDWriteTableCB *write_table_cb; 127 unsigned int sector_mask = BDRV_SECTOR_SIZE / sizeof(uint64_t) - 1; 128 unsigned int start, end, i; 129 size_t len_bytes; 130 131 trace_qed_write_table(s, offset, table, index, n); 132 133 /* Calculate indices of the first and one after last elements */ 134 start = index & ~sector_mask; 135 end = (index + n + sector_mask) & ~sector_mask; 136 137 len_bytes = (end - start) * sizeof(uint64_t); 138 139 write_table_cb = gencb_alloc(sizeof(*write_table_cb), cb, opaque); 140 write_table_cb->s = s; 141 write_table_cb->orig_table = table; 142 write_table_cb->flush = flush; 143 write_table_cb->table = qemu_blockalign(s->bs, len_bytes); 144 write_table_cb->iov.iov_base = write_table_cb->table->offsets; 145 write_table_cb->iov.iov_len = len_bytes; 146 qemu_iovec_init_external(&write_table_cb->qiov, &write_table_cb->iov, 1); 147 148 /* Byteswap table */ 149 for (i = start; i < end; i++) { 150 uint64_t le_offset = cpu_to_le64(table->offsets[i]); 151 write_table_cb->table->offsets[i - start] = le_offset; 152 } 153 154 /* Adjust for offset into table */ 155 offset += start * sizeof(uint64_t); 156 157 bdrv_aio_writev(s->bs->file, offset / BDRV_SECTOR_SIZE, 158 &write_table_cb->qiov, 159 write_table_cb->qiov.size / BDRV_SECTOR_SIZE, 160 qed_write_table_cb, write_table_cb); 161 } 162 163 /** 164 * Propagate return value from async callback 165 */ 166 static void qed_sync_cb(void *opaque, int ret) 167 { 168 *(int *)opaque = ret; 169 } 170 171 int qed_read_l1_table_sync(BDRVQEDState *s) 172 { 173 int ret = -EINPROGRESS; 174 175 qed_read_table(s, s->header.l1_table_offset, 176 s->l1_table, qed_sync_cb, &ret); 177 BDRV_POLL_WHILE(s->bs, ret == -EINPROGRESS); 178 179 return ret; 180 } 181 182 void qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n, 183 BlockCompletionFunc *cb, void *opaque) 184 { 185 BLKDBG_EVENT(s->bs->file, BLKDBG_L1_UPDATE); 186 qed_write_table(s, s->header.l1_table_offset, 187 s->l1_table, index, n, false, cb, opaque); 188 } 189 190 int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index, 191 unsigned int n) 192 { 193 int ret = -EINPROGRESS; 194 195 qed_write_l1_table(s, index, n, qed_sync_cb, &ret); 196 BDRV_POLL_WHILE(s->bs, ret == -EINPROGRESS); 197 198 return ret; 199 } 200 201 typedef struct { 202 GenericCB gencb; 203 BDRVQEDState *s; 204 uint64_t l2_offset; 205 QEDRequest *request; 206 } QEDReadL2TableCB; 207 208 static void qed_read_l2_table_cb(void *opaque, int ret) 209 { 210 QEDReadL2TableCB *read_l2_table_cb = opaque; 211 QEDRequest *request = read_l2_table_cb->request; 212 BDRVQEDState *s = read_l2_table_cb->s; 213 CachedL2Table *l2_table = request->l2_table; 214 uint64_t l2_offset = read_l2_table_cb->l2_offset; 215 216 if (ret) { 217 /* can't trust loaded L2 table anymore */ 218 qed_unref_l2_cache_entry(l2_table); 219 request->l2_table = NULL; 220 } else { 221 l2_table->offset = l2_offset; 222 223 qed_commit_l2_cache_entry(&s->l2_cache, l2_table); 224 225 /* This is guaranteed to succeed because we just committed the entry 226 * to the cache. 227 */ 228 request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset); 229 assert(request->l2_table != NULL); 230 } 231 232 gencb_complete(&read_l2_table_cb->gencb, ret); 233 } 234 235 void qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset, 236 BlockCompletionFunc *cb, void *opaque) 237 { 238 QEDReadL2TableCB *read_l2_table_cb; 239 240 qed_unref_l2_cache_entry(request->l2_table); 241 242 /* Check for cached L2 entry */ 243 request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset); 244 if (request->l2_table) { 245 cb(opaque, 0); 246 return; 247 } 248 249 request->l2_table = qed_alloc_l2_cache_entry(&s->l2_cache); 250 request->l2_table->table = qed_alloc_table(s); 251 252 read_l2_table_cb = gencb_alloc(sizeof(*read_l2_table_cb), cb, opaque); 253 read_l2_table_cb->s = s; 254 read_l2_table_cb->l2_offset = offset; 255 read_l2_table_cb->request = request; 256 257 BLKDBG_EVENT(s->bs->file, BLKDBG_L2_LOAD); 258 qed_read_table(s, offset, request->l2_table->table, 259 qed_read_l2_table_cb, read_l2_table_cb); 260 } 261 262 int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request, uint64_t offset) 263 { 264 int ret = -EINPROGRESS; 265 266 qed_read_l2_table(s, request, offset, qed_sync_cb, &ret); 267 BDRV_POLL_WHILE(s->bs, ret == -EINPROGRESS); 268 269 return ret; 270 } 271 272 void qed_write_l2_table(BDRVQEDState *s, QEDRequest *request, 273 unsigned int index, unsigned int n, bool flush, 274 BlockCompletionFunc *cb, void *opaque) 275 { 276 BLKDBG_EVENT(s->bs->file, BLKDBG_L2_UPDATE); 277 qed_write_table(s, request->l2_table->offset, 278 request->l2_table->table, index, n, flush, cb, opaque); 279 } 280 281 int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request, 282 unsigned int index, unsigned int n, bool flush) 283 { 284 int ret = -EINPROGRESS; 285 286 qed_write_l2_table(s, request, index, n, flush, qed_sync_cb, &ret); 287 BDRV_POLL_WHILE(s->bs, ret == -EINPROGRESS); 288 289 return ret; 290 } 291