4 * Copyright IBM, Corp. 2011
9 * This work is licensed under the terms of the GNU LGPL, version 2 or later.
10 * See the COPYING.LIB file in the top-level directory.
15 #include "block_int.h"
19 * Size of data buffer for populating the image file. This should be large
20 * enough to process multiple clusters in a single call, so that populating
21 * contiguous regions of the image is efficient.
23 STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */
26 #define SLICE_TIME 100000000ULL /* ns */
29 int64_t next_slice_time;
34 static int64_t ratelimit_calculate_delay(RateLimit *limit, uint64_t n)
37 int64_t now = qemu_get_clock_ns(rt_clock);
39 if (limit->next_slice_time < now) {
40 limit->next_slice_time = now + SLICE_TIME;
41 limit->dispatched = 0;
43 if (limit->dispatched + n > limit->slice_quota) {
44 delay_ns = limit->next_slice_time - now;
46 limit->dispatched += n;
51 static void ratelimit_set_speed(RateLimit *limit, uint64_t speed)
53 limit->slice_quota = speed / (1000000000ULL / SLICE_TIME);
56 typedef struct StreamBlockJob {
59 BlockDriverState *base;
60 char backing_file_id[1024];
63 static int coroutine_fn stream_populate(BlockDriverState *bs,
64 int64_t sector_num, int nb_sectors,
69 .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
73 qemu_iovec_init_external(&qiov, &iov, 1);
75 /* Copy-on-read the unallocated clusters */
76 return bdrv_co_copy_on_readv(bs, sector_num, nb_sectors, &qiov);
79 static void close_unused_images(BlockDriverState *top, BlockDriverState *base,
82 BlockDriverState *intermediate;
83 intermediate = top->backing_hd;
85 while (intermediate) {
86 BlockDriverState *unused;
89 if (intermediate == base) {
93 unused = intermediate;
94 intermediate = intermediate->backing_hd;
95 unused->backing_hd = NULL;
98 top->backing_hd = base;
100 pstrcpy(top->backing_file, sizeof(top->backing_file), "");
101 pstrcpy(top->backing_format, sizeof(top->backing_format), "");
103 pstrcpy(top->backing_file, sizeof(top->backing_file), base_id);
105 pstrcpy(top->backing_format, sizeof(top->backing_format),
106 base->drv->format_name);
113 * Given an image chain: [BASE] -> [INTER1] -> [INTER2] -> [TOP]
115 * Return true if the given sector is allocated in top.
116 * Return false if the given sector is allocated in intermediate images.
117 * Return true otherwise.
119 * 'pnum' is set to the number of sectors (including and immediately following
120 * the specified sector) that are known to be in the same
121 * allocated/unallocated state.
124 static int coroutine_fn is_allocated_base(BlockDriverState *top,
125 BlockDriverState *base,
127 int nb_sectors, int *pnum)
129 BlockDriverState *intermediate;
132 ret = bdrv_co_is_allocated(top, sector_num, nb_sectors, &n);
139 * Is the unallocated chunk [sector_num, n] also
140 * unallocated between base and top?
142 intermediate = top->backing_hd;
144 while (intermediate) {
148 if (intermediate == base) {
152 ret = bdrv_co_is_allocated(intermediate, sector_num, nb_sectors,
162 * [sector_num, nb_sectors] is unallocated on top but intermediate
165 * [sector_num+x, nr_sectors] allocated.
167 if (n > pnum_inter) {
171 intermediate = intermediate->backing_hd;
177 static void coroutine_fn stream_run(void *opaque)
179 StreamBlockJob *s = opaque;
180 BlockDriverState *bs = s->common.bs;
181 BlockDriverState *base = s->base;
182 int64_t sector_num, end;
187 s->common.len = bdrv_getlength(bs);
188 if (s->common.len < 0) {
189 block_job_complete(&s->common, s->common.len);
193 end = s->common.len >> BDRV_SECTOR_BITS;
194 buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
196 /* Turn on copy-on-read for the whole block device so that guest read
197 * requests help us make progress. Only do this when copying the entire
198 * backing chain since the copy-on-read operation does not take base into
202 bdrv_enable_copy_on_read(bs);
205 for (sector_num = 0; sector_num < end; sector_num += n) {
207 if (block_job_is_cancelled(&s->common)) {
211 s->common.busy = true;
213 ret = is_allocated_base(bs, base, sector_num,
214 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
216 ret = bdrv_co_is_allocated(bs, sector_num,
217 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE,
220 trace_stream_one_iteration(s, sector_num, n, ret);
222 if (s->common.speed) {
223 uint64_t delay_ns = ratelimit_calculate_delay(&s->limit, n);
225 s->common.busy = false;
226 co_sleep_ns(rt_clock, delay_ns);
228 /* Recheck cancellation and that sectors are unallocated */
232 ret = stream_populate(bs, sector_num, n, buf);
239 /* Publish progress */
240 s->common.offset += n * BDRV_SECTOR_SIZE;
242 /* Note that even when no rate limit is applied we need to yield
243 * with no pending I/O here so that qemu_aio_flush() returns.
245 s->common.busy = false;
246 co_sleep_ns(rt_clock, 0);
250 bdrv_disable_copy_on_read(bs);
253 if (!block_job_is_cancelled(&s->common) && sector_num == end && ret == 0) {
254 const char *base_id = NULL;
256 base_id = s->backing_file_id;
258 ret = bdrv_change_backing_file(bs, base_id, NULL);
259 close_unused_images(bs, base, base_id);
263 block_job_complete(&s->common, ret);
266 static int stream_set_speed(BlockJob *job, int64_t value)
268 StreamBlockJob *s = container_of(job, StreamBlockJob, common);
273 ratelimit_set_speed(&s->limit, value / BDRV_SECTOR_SIZE);
277 static BlockJobType stream_job_type = {
278 .instance_size = sizeof(StreamBlockJob),
279 .job_type = "stream",
280 .set_speed = stream_set_speed,
283 int stream_start(BlockDriverState *bs, BlockDriverState *base,
284 const char *base_id, BlockDriverCompletionFunc *cb,
290 s = block_job_create(&stream_job_type, bs, cb, opaque);
292 return -EBUSY; /* bs must already be in use */
297 pstrcpy(s->backing_file_id, sizeof(s->backing_file_id), base_id);
300 co = qemu_coroutine_create(stream_run);
301 trace_stream_start(bs, base, s, co, opaque);
302 qemu_coroutine_enter(co, s);