blob: ce8c3140896d0f01d9ef0bb798af7fdf6d1a3c14 [file] [log] [blame]
Stefan Hajnoczi75411d22010-12-06 16:08:00 +00001/*
2 * QEMU Enhanced Disk Format
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#ifndef BLOCK_QED_H
16#define BLOCK_QED_H
17
Paolo Bonzini737e1502012-12-17 18:19:44 +010018#include "block/block_int.h"
Veronia Bahaaf348b6d2016-03-20 19:16:19 +020019#include "qemu/cutils.h"
Stefan Hajnoczi75411d22010-12-06 16:08:00 +000020
21/* The layout of a QED file is as follows:
22 *
23 * +--------+----------+----------+----------+-----+
24 * | header | L1 table | cluster0 | cluster1 | ... |
25 * +--------+----------+----------+----------+-----+
26 *
27 * There is a 2-level pagetable for cluster allocation:
28 *
29 * +----------+
30 * | L1 table |
31 * +----------+
32 * ,------' | '------.
33 * +----------+ | +----------+
34 * | L2 table | ... | L2 table |
35 * +----------+ +----------+
36 * ,------' | '------.
37 * +----------+ | +----------+
38 * | Data | ... | Data |
39 * +----------+ +----------+
40 *
41 * The L1 table is fixed size and always present. L2 tables are allocated on
42 * demand. The L1 table size determines the maximum possible image size; it
43 * can be influenced using the cluster_size and table_size values.
44 *
45 * All fields are little-endian on disk.
46 */
Chunyan Liu7ab74842014-06-05 17:21:00 +080047#define QED_DEFAULT_CLUSTER_SIZE 65536
Stefan Hajnoczi75411d22010-12-06 16:08:00 +000048enum {
49 QED_MAGIC = 'Q' | 'E' << 8 | 'D' << 16 | '\0' << 24,
50
51 /* The image supports a backing file */
52 QED_F_BACKING_FILE = 0x01,
53
Stefan Hajnoczi01979a92010-12-06 16:08:03 +000054 /* The image needs a consistency check before use */
55 QED_F_NEED_CHECK = 0x02,
56
Stefan Hajnoczi75411d22010-12-06 16:08:00 +000057 /* The backing file format must not be probed, treat as raw image */
58 QED_F_BACKING_FORMAT_NO_PROBE = 0x04,
59
60 /* Feature bits must be used when the on-disk format changes */
61 QED_FEATURE_MASK = QED_F_BACKING_FILE | /* supported feature bits */
Stefan Hajnoczi01979a92010-12-06 16:08:03 +000062 QED_F_NEED_CHECK |
Stefan Hajnoczi75411d22010-12-06 16:08:00 +000063 QED_F_BACKING_FORMAT_NO_PROBE,
64 QED_COMPAT_FEATURE_MASK = 0, /* supported compat feature bits */
65 QED_AUTOCLEAR_FEATURE_MASK = 0, /* supported autoclear feature bits */
66
67 /* Data is stored in groups of sectors called clusters. Cluster size must
68 * be large to avoid keeping too much metadata. I/O requests that have
69 * sub-cluster size will require read-modify-write.
70 */
71 QED_MIN_CLUSTER_SIZE = 4 * 1024, /* in bytes */
72 QED_MAX_CLUSTER_SIZE = 64 * 1024 * 1024,
Stefan Hajnoczi75411d22010-12-06 16:08:00 +000073
74 /* Allocated clusters are tracked using a 2-level pagetable. Table size is
75 * a multiple of clusters so large maximum image sizes can be supported
76 * without jacking up the cluster size too much.
77 */
78 QED_MIN_TABLE_SIZE = 1, /* in clusters */
79 QED_MAX_TABLE_SIZE = 16,
80 QED_DEFAULT_TABLE_SIZE = 4,
Stefan Hajnoczi6f321e92011-05-09 16:45:40 +010081
82 /* Delay to flush and clean image after last allocating write completes */
83 QED_NEED_CHECK_TIMEOUT = 5, /* in seconds */
Stefan Hajnoczi75411d22010-12-06 16:08:00 +000084};
85
86typedef struct {
87 uint32_t magic; /* QED\0 */
88
89 uint32_t cluster_size; /* in bytes */
90 uint32_t table_size; /* for L1 and L2 tables, in clusters */
91 uint32_t header_size; /* in clusters */
92
93 uint64_t features; /* format feature bits */
94 uint64_t compat_features; /* compatible feature bits */
95 uint64_t autoclear_features; /* self-resetting feature bits */
96
97 uint64_t l1_table_offset; /* in bytes */
98 uint64_t image_size; /* total logical image size, in bytes */
99
100 /* if (features & QED_F_BACKING_FILE) */
101 uint32_t backing_filename_offset; /* in bytes from start of header */
102 uint32_t backing_filename_size; /* in bytes */
Jeff Cody687fb892013-09-25 12:08:51 -0400103} QEMU_PACKED QEDHeader;
Stefan Hajnoczi75411d22010-12-06 16:08:00 +0000104
105typedef struct {
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000106 uint64_t offsets[0]; /* in bytes */
107} QEDTable;
108
109/* The L2 cache is a simple write-through cache for L2 structures */
110typedef struct CachedL2Table {
111 QEDTable *table;
112 uint64_t offset; /* offset=0 indicates an invalidate entry */
113 QTAILQ_ENTRY(CachedL2Table) node;
114 int ref;
115} CachedL2Table;
116
117typedef struct {
118 QTAILQ_HEAD(, CachedL2Table) entries;
119 unsigned int n_entries;
120} L2TableCache;
121
122typedef struct QEDRequest {
123 CachedL2Table *l2_table;
124} QEDRequest;
125
Stefan Hajnoczi6e4f59b2012-02-07 13:27:27 +0000126enum {
127 QED_AIOCB_WRITE = 0x0001, /* read or write? */
Stefan Hajnoczi0e71be12012-02-07 13:27:28 +0000128 QED_AIOCB_ZERO = 0x0002, /* zero write, used with QED_AIOCB_WRITE */
Stefan Hajnoczi6e4f59b2012-02-07 13:27:27 +0000129};
130
Stefan Hajnoczieabba582010-12-06 16:08:02 +0000131typedef struct QEDAIOCB {
Markus Armbruster7c84b1b2014-10-07 13:59:14 +0200132 BlockAIOCB common;
Stefan Hajnoczieabba582010-12-06 16:08:02 +0000133 int bh_ret; /* final return status for completion bh */
134 QSIMPLEQ_ENTRY(QEDAIOCB) next; /* next request */
Stefan Hajnoczi6e4f59b2012-02-07 13:27:27 +0000135 int flags; /* QED_AIOCB_* bits ORed together */
Stefan Hajnoczieabba582010-12-06 16:08:02 +0000136 uint64_t end_pos; /* request end on block device, in bytes */
137
138 /* User scatter-gather list */
139 QEMUIOVector *qiov;
140 size_t qiov_offset; /* byte count already processed */
141
142 /* Current cluster scatter-gather list */
143 QEMUIOVector cur_qiov;
Kevin Wolff06ee3d2014-07-04 17:11:28 +0200144 QEMUIOVector *backing_qiov;
Stefan Hajnoczieabba582010-12-06 16:08:02 +0000145 uint64_t cur_pos; /* position on block device, in bytes */
146 uint64_t cur_cluster; /* cluster offset in image file */
147 unsigned int cur_nclusters; /* number of clusters being accessed */
148 int find_cluster_ret; /* used for L1/L2 update */
149
150 QEDRequest request;
151} QEDAIOCB;
152
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000153typedef struct {
Stefan Hajnoczi75411d22010-12-06 16:08:00 +0000154 BlockDriverState *bs; /* device */
155 uint64_t file_size; /* length of image file, in bytes */
156
157 QEDHeader header; /* always cpu-endian */
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000158 QEDTable *l1_table;
159 L2TableCache l2_cache; /* l2 table cache */
Stefan Hajnoczi75411d22010-12-06 16:08:00 +0000160 uint32_t table_nelems;
161 uint32_t l1_shift;
162 uint32_t l2_shift;
163 uint32_t l2_mask;
Stefan Hajnoczieabba582010-12-06 16:08:02 +0000164
165 /* Allocating write request queue */
166 QSIMPLEQ_HEAD(, QEDAIOCB) allocating_write_reqs;
Stefan Hajnoczi6f321e92011-05-09 16:45:40 +0100167 bool allocating_write_reqs_plugged;
168
169 /* Periodic flush and clear need check flag */
170 QEMUTimer *need_check_timer;
Stefan Hajnoczi75411d22010-12-06 16:08:00 +0000171} BDRVQEDState;
172
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000173enum {
174 QED_CLUSTER_FOUND, /* cluster found */
Anthony Liguori21df65b2010-12-17 15:58:22 +0000175 QED_CLUSTER_ZERO, /* zero cluster found */
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000176 QED_CLUSTER_L2, /* cluster missing in L2 */
177 QED_CLUSTER_L1, /* cluster missing in L1 */
178};
179
180/**
181 * qed_find_cluster() completion callback
182 *
183 * @opaque: User data for completion callback
184 * @ret: QED_CLUSTER_FOUND Success
185 * QED_CLUSTER_L2 Data cluster unallocated in L2
186 * QED_CLUSTER_L1 L2 unallocated in L1
187 * -errno POSIX error occurred
188 * @offset: Data cluster offset
189 * @len: Contiguous bytes starting from cluster offset
190 *
191 * This function is invoked when qed_find_cluster() completes.
192 *
193 * On success ret is QED_CLUSTER_FOUND and offset/len are a contiguous range
194 * in the image file.
195 *
196 * On failure ret is QED_CLUSTER_L2 or QED_CLUSTER_L1 for missing L2 or L1
197 * table offset, respectively. len is number of contiguous unallocated bytes.
198 */
199typedef void QEDFindClusterFunc(void *opaque, int ret, uint64_t offset, size_t len);
200
Paolo Bonzini2f47da52017-02-13 14:52:29 +0100201void qed_acquire(BDRVQEDState *s);
202void qed_release(BDRVQEDState *s);
203
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000204/**
205 * Generic callback for chaining async callbacks
206 */
207typedef struct {
Markus Armbruster097310b2014-10-07 13:59:15 +0200208 BlockCompletionFunc *cb;
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000209 void *opaque;
210} GenericCB;
211
Markus Armbruster097310b2014-10-07 13:59:15 +0200212void *gencb_alloc(size_t len, BlockCompletionFunc *cb, void *opaque);
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000213void gencb_complete(void *opaque, int ret);
214
215/**
Stefan Hajnoczib10170a2012-08-09 13:05:54 +0100216 * Header functions
217 */
218int qed_write_header_sync(BDRVQEDState *s);
219
220/**
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000221 * L2 cache functions
222 */
223void qed_init_l2_cache(L2TableCache *l2_cache);
224void qed_free_l2_cache(L2TableCache *l2_cache);
225CachedL2Table *qed_alloc_l2_cache_entry(L2TableCache *l2_cache);
226void qed_unref_l2_cache_entry(CachedL2Table *entry);
227CachedL2Table *qed_find_l2_cache_entry(L2TableCache *l2_cache, uint64_t offset);
228void qed_commit_l2_cache_entry(L2TableCache *l2_cache, CachedL2Table *l2_table);
229
230/**
231 * Table I/O functions
232 */
233int qed_read_l1_table_sync(BDRVQEDState *s);
234void qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n,
Markus Armbruster097310b2014-10-07 13:59:15 +0200235 BlockCompletionFunc *cb, void *opaque);
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000236int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
237 unsigned int n);
238int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
239 uint64_t offset);
240void qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset,
Markus Armbruster097310b2014-10-07 13:59:15 +0200241 BlockCompletionFunc *cb, void *opaque);
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000242void qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
243 unsigned int index, unsigned int n, bool flush,
Markus Armbruster097310b2014-10-07 13:59:15 +0200244 BlockCompletionFunc *cb, void *opaque);
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000245int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
246 unsigned int index, unsigned int n, bool flush);
247
248/**
249 * Cluster functions
250 */
251void qed_find_cluster(BDRVQEDState *s, QEDRequest *request, uint64_t pos,
252 size_t len, QEDFindClusterFunc *cb, void *opaque);
253
254/**
255 * Consistency check
256 */
257int qed_check(BDRVQEDState *s, BdrvCheckResult *result, bool fix);
258
259QEDTable *qed_alloc_table(BDRVQEDState *s);
260
Stefan Hajnoczi75411d22010-12-06 16:08:00 +0000261/**
262 * Round down to the start of a cluster
263 */
264static inline uint64_t qed_start_of_cluster(BDRVQEDState *s, uint64_t offset)
265{
266 return offset & ~(uint64_t)(s->header.cluster_size - 1);
267}
268
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000269static inline uint64_t qed_offset_into_cluster(BDRVQEDState *s, uint64_t offset)
270{
271 return offset & (s->header.cluster_size - 1);
272}
273
Stefan Hajnoczi19dfc442011-04-24 18:38:58 +0100274static inline uint64_t qed_bytes_to_clusters(BDRVQEDState *s, uint64_t bytes)
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000275{
276 return qed_start_of_cluster(s, bytes + (s->header.cluster_size - 1)) /
277 (s->header.cluster_size - 1);
278}
279
280static inline unsigned int qed_l1_index(BDRVQEDState *s, uint64_t pos)
281{
282 return pos >> s->l1_shift;
283}
284
285static inline unsigned int qed_l2_index(BDRVQEDState *s, uint64_t pos)
286{
287 return (pos >> s->l2_shift) & s->l2_mask;
288}
289
Stefan Hajnoczi75411d22010-12-06 16:08:00 +0000290/**
291 * Test if a cluster offset is valid
292 */
293static inline bool qed_check_cluster_offset(BDRVQEDState *s, uint64_t offset)
294{
295 uint64_t header_size = (uint64_t)s->header.header_size *
296 s->header.cluster_size;
297
298 if (offset & (s->header.cluster_size - 1)) {
299 return false;
300 }
301 return offset >= header_size && offset < s->file_size;
302}
303
304/**
305 * Test if a table offset is valid
306 */
307static inline bool qed_check_table_offset(BDRVQEDState *s, uint64_t offset)
308{
309 uint64_t end_offset = offset + (s->header.table_size - 1) *
310 s->header.cluster_size;
311
312 /* Overflow check */
313 if (end_offset <= offset) {
314 return false;
315 }
316
317 return qed_check_cluster_offset(s, offset) &&
318 qed_check_cluster_offset(s, end_offset);
319}
320
Anthony Liguori21df65b2010-12-17 15:58:22 +0000321static inline bool qed_offset_is_cluster_aligned(BDRVQEDState *s,
322 uint64_t offset)
323{
324 if (qed_offset_into_cluster(s, offset)) {
325 return false;
326 }
327 return true;
328}
329
330static inline bool qed_offset_is_unalloc_cluster(uint64_t offset)
331{
332 if (offset == 0) {
333 return true;
334 }
335 return false;
336}
337
338static inline bool qed_offset_is_zero_cluster(uint64_t offset)
339{
340 if (offset == 1) {
341 return true;
342 }
343 return false;
344}
345
Stefan Hajnoczi75411d22010-12-06 16:08:00 +0000346#endif /* BLOCK_QED_H */