1 /* 2 * Image streaming 3 * 4 * Copyright IBM, Corp. 2011 5 * 6 * Authors: 7 * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com> 8 * 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. 11 * 12 */ 13 14 #include "trace.h" 15 #include "block_int.h" 16 #include "qemu/ratelimit.h" 17 18 enum { 19 /* 20 * Size of data buffer for populating the image file. This should be large 21 * enough to process multiple clusters in a single call, so that populating 22 * contiguous regions of the image is efficient. 23 */ 24 STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */ 25 }; 26 27 #define SLICE_TIME 100000000ULL /* ns */ 28 29 typedef struct StreamBlockJob { 30 BlockJob common; 31 RateLimit limit; 32 BlockDriverState *base; 33 char backing_file_id[1024]; 34 } StreamBlockJob; 35 36 static int coroutine_fn stream_populate(BlockDriverState *bs, 37 int64_t sector_num, int nb_sectors, 38 void *buf) 39 { 40 struct iovec iov = { 41 .iov_base = buf, 42 .iov_len = nb_sectors * BDRV_SECTOR_SIZE, 43 }; 44 QEMUIOVector qiov; 45 46 qemu_iovec_init_external(&qiov, &iov, 1); 47 48 /* Copy-on-read the unallocated clusters */ 49 return bdrv_co_copy_on_readv(bs, sector_num, nb_sectors, &qiov); 50 } 51 52 static void close_unused_images(BlockDriverState *top, BlockDriverState *base, 53 const char *base_id) 54 { 55 BlockDriverState *intermediate; 56 intermediate = top->backing_hd; 57 58 while (intermediate) { 59 BlockDriverState *unused; 60 61 /* reached base */ 62 if (intermediate == base) { 63 break; 64 } 65 66 unused = intermediate; 67 intermediate = intermediate->backing_hd; 68 unused->backing_hd = NULL; 69 bdrv_delete(unused); 70 } 71 top->backing_hd = base; 72 } 73 74 static void coroutine_fn stream_run(void *opaque) 75 { 76 StreamBlockJob *s = opaque; 77 BlockDriverState *bs = s->common.bs; 78 BlockDriverState *base = s->base; 79 int64_t sector_num, end; 80 int ret = 0; 81 int n = 0; 82 void *buf; 83 84 s->common.len = bdrv_getlength(bs); 85 if (s->common.len < 0) { 86 block_job_complete(&s->common, s->common.len); 87 return; 88 } 89 90 end = s->common.len >> BDRV_SECTOR_BITS; 91 buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE); 92 93 /* Turn on copy-on-read for the whole block device so that guest read 94 * requests help us make progress. Only do this when copying the entire 95 * backing chain since the copy-on-read operation does not take base into 96 * account. 97 */ 98 if (!base) { 99 bdrv_enable_copy_on_read(bs); 100 } 101 102 for (sector_num = 0; sector_num < end; sector_num += n) { 103 uint64_t delay_ns = 0; 104 bool copy; 105 106 wait: 107 /* Note that even when no rate limit is applied we need to yield 108 * with no pending I/O here so that qemu_aio_flush() returns. 109 */ 110 block_job_sleep_ns(&s->common, rt_clock, delay_ns); 111 if (block_job_is_cancelled(&s->common)) { 112 break; 113 } 114 115 ret = bdrv_co_is_allocated(bs, sector_num, 116 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n); 117 if (ret == 1) { 118 /* Allocated in the top, no need to copy. */ 119 copy = false; 120 } else { 121 /* Copy if allocated in the intermediate images. Limit to the 122 * known-unallocated area [sector_num, sector_num+n). */ 123 ret = bdrv_co_is_allocated_above(bs->backing_hd, base, 124 sector_num, n, &n); 125 copy = (ret == 1); 126 } 127 trace_stream_one_iteration(s, sector_num, n, ret); 128 if (ret >= 0 && copy) { 129 if (s->common.speed) { 130 delay_ns = ratelimit_calculate_delay(&s->limit, n); 131 if (delay_ns > 0) { 132 goto wait; 133 } 134 } 135 ret = stream_populate(bs, sector_num, n, buf); 136 } 137 if (ret < 0) { 138 break; 139 } 140 ret = 0; 141 142 /* Publish progress */ 143 s->common.offset += n * BDRV_SECTOR_SIZE; 144 } 145 146 if (!base) { 147 bdrv_disable_copy_on_read(bs); 148 } 149 150 if (!block_job_is_cancelled(&s->common) && sector_num == end && ret == 0) { 151 const char *base_id = NULL, *base_fmt = NULL; 152 if (base) { 153 base_id = s->backing_file_id; 154 if (base->drv) { 155 base_fmt = base->drv->format_name; 156 } 157 } 158 ret = bdrv_change_backing_file(bs, base_id, base_fmt); 159 close_unused_images(bs, base, base_id); 160 } 161 162 qemu_vfree(buf); 163 block_job_complete(&s->common, ret); 164 } 165 166 static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp) 167 { 168 StreamBlockJob *s = container_of(job, StreamBlockJob, common); 169 170 if (speed < 0) { 171 error_set(errp, QERR_INVALID_PARAMETER, "speed"); 172 return; 173 } 174 ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME); 175 } 176 177 static BlockJobType stream_job_type = { 178 .instance_size = sizeof(StreamBlockJob), 179 .job_type = "stream", 180 .set_speed = stream_set_speed, 181 }; 182 183 void stream_start(BlockDriverState *bs, BlockDriverState *base, 184 const char *base_id, int64_t speed, 185 BlockDriverCompletionFunc *cb, 186 void *opaque, Error **errp) 187 { 188 StreamBlockJob *s; 189 190 s = block_job_create(&stream_job_type, bs, speed, cb, opaque, errp); 191 if (!s) { 192 return; 193 } 194 195 s->base = base; 196 if (base_id) { 197 pstrcpy(s->backing_file_id, sizeof(s->backing_file_id), base_id); 198 } 199 200 s->common.co = qemu_coroutine_create(stream_run); 201 trace_stream_start(bs, base, s, s->common.co, opaque); 202 qemu_coroutine_enter(s->common.co, s); 203 } 204