xref: /qemu/block/qed-table.c (revision b2a3cbb8)
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 #include "qemu/memalign.h"
21 
22 /* Called with table_lock held.  */
23 static int coroutine_fn qed_read_table(BDRVQEDState *s, uint64_t offset,
24                                        QEDTable *table)
25 {
26     unsigned int bytes = s->header.cluster_size * s->header.table_size;
27 
28     int noffsets;
29     int i, ret;
30 
31     trace_qed_read_table(s, offset, table);
32 
33     qemu_co_mutex_unlock(&s->table_lock);
34     ret = bdrv_co_pread(s->bs->file, offset, bytes, table->offsets, 0);
35     qemu_co_mutex_lock(&s->table_lock);
36     if (ret < 0) {
37         goto out;
38     }
39 
40     /* Byteswap offsets */
41     noffsets = bytes / sizeof(uint64_t);
42     for (i = 0; i < noffsets; i++) {
43         table->offsets[i] = le64_to_cpu(table->offsets[i]);
44     }
45 
46     ret = 0;
47 out:
48     /* Completion */
49     trace_qed_read_table_cb(s, table, ret);
50     return ret;
51 }
52 
53 /**
54  * Write out an updated part or all of a table
55  *
56  * @s:          QED state
57  * @offset:     Offset of table in image file, in bytes
58  * @table:      Table
59  * @index:      Index of first element
60  * @n:          Number of elements
61  * @flush:      Whether or not to sync to disk
62  *
63  * Called with table_lock held.
64  */
65 static int coroutine_fn qed_write_table(BDRVQEDState *s, uint64_t offset,
66                                         QEDTable *table, unsigned int index,
67                                         unsigned int n, bool flush)
68 {
69     unsigned int sector_mask = BDRV_SECTOR_SIZE / sizeof(uint64_t) - 1;
70     unsigned int start, end, i;
71     QEDTable *new_table;
72     size_t len_bytes;
73     int ret;
74 
75     trace_qed_write_table(s, offset, table, index, n);
76 
77     /* Calculate indices of the first and one after last elements */
78     start = index & ~sector_mask;
79     end = (index + n + sector_mask) & ~sector_mask;
80 
81     len_bytes = (end - start) * sizeof(uint64_t);
82 
83     new_table = qemu_blockalign(s->bs, len_bytes);
84 
85     /* Byteswap table */
86     for (i = start; i < end; i++) {
87         uint64_t le_offset = cpu_to_le64(table->offsets[i]);
88         new_table->offsets[i - start] = le_offset;
89     }
90 
91     /* Adjust for offset into table */
92     offset += start * sizeof(uint64_t);
93 
94     qemu_co_mutex_unlock(&s->table_lock);
95     ret = bdrv_co_pwrite(s->bs->file, offset, len_bytes, new_table->offsets, 0);
96     qemu_co_mutex_lock(&s->table_lock);
97     trace_qed_write_table_cb(s, table, flush, ret);
98     if (ret < 0) {
99         goto out;
100     }
101 
102     if (flush) {
103         ret = bdrv_co_flush(s->bs);
104         if (ret < 0) {
105             goto out;
106         }
107     }
108 
109     ret = 0;
110 out:
111     qemu_vfree(new_table);
112     return ret;
113 }
114 
115 int coroutine_fn qed_read_l1_table_sync(BDRVQEDState *s)
116 {
117     return qed_read_table(s, s->header.l1_table_offset, s->l1_table);
118 }
119 
120 /* Called with table_lock held.  */
121 int coroutine_fn qed_write_l1_table(BDRVQEDState *s, unsigned int index,
122                                     unsigned int n)
123 {
124     BLKDBG_EVENT(s->bs->file, BLKDBG_L1_UPDATE);
125     return qed_write_table(s, s->header.l1_table_offset,
126                            s->l1_table, index, n, false);
127 }
128 
129 int coroutine_fn qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
130                                          unsigned int n)
131 {
132     return qed_write_l1_table(s, index, n);
133 }
134 
135 /* Called with table_lock held.  */
136 int coroutine_fn qed_read_l2_table(BDRVQEDState *s, QEDRequest *request,
137                                    uint64_t offset)
138 {
139     int ret;
140 
141     qed_unref_l2_cache_entry(request->l2_table);
142 
143     /* Check for cached L2 entry */
144     request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset);
145     if (request->l2_table) {
146         return 0;
147     }
148 
149     request->l2_table = qed_alloc_l2_cache_entry(&s->l2_cache);
150     request->l2_table->table = qed_alloc_table(s);
151 
152     BLKDBG_EVENT(s->bs->file, BLKDBG_L2_LOAD);
153     ret = qed_read_table(s, offset, request->l2_table->table);
154 
155     if (ret) {
156         /* can't trust loaded L2 table anymore */
157         qed_unref_l2_cache_entry(request->l2_table);
158         request->l2_table = NULL;
159     } else {
160         request->l2_table->offset = offset;
161 
162         qed_commit_l2_cache_entry(&s->l2_cache, request->l2_table);
163 
164         /* This is guaranteed to succeed because we just committed the entry
165          * to the cache.
166          */
167         request->l2_table = qed_find_l2_cache_entry(&s->l2_cache, offset);
168         assert(request->l2_table != NULL);
169     }
170 
171     return ret;
172 }
173 
174 int coroutine_fn qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
175                                         uint64_t offset)
176 {
177     return qed_read_l2_table(s, request, offset);
178 }
179 
180 /* Called with table_lock held.  */
181 int coroutine_fn qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
182                                     unsigned int index, unsigned int n,
183                                     bool flush)
184 {
185     BLKDBG_EVENT(s->bs->file, BLKDBG_L2_UPDATE);
186     return qed_write_table(s, request->l2_table->offset,
187                            request->l2_table->table, index, n, flush);
188 }
189 
190 int coroutine_fn qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
191                                          unsigned int index, unsigned int n,
192                                          bool flush)
193 {
194     return qed_write_l2_table(s, request, index, n, flush);
195 }
196