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