4 * Copyright IBM, Corp. 2011
7 * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
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.
14 #include "qemu/osdep.h"
16 #include "block/block_int.h"
17 #include "block/blockjob_int.h"
18 #include "qapi/error.h"
19 #include "qapi/qmp/qerror.h"
20 #include "qemu/ratelimit.h"
21 #include "sysemu/block-backend.h"
25 * Maximum chunk size to feed to copy-on-read. This should be
26 * large enough to process multiple clusters in a single call, so
27 * that populating contiguous regions of the image is efficient.
29 STREAM_CHUNK = 512 * 1024, /* in bytes */
32 typedef struct StreamBlockJob {
34 BlockDriverState *base_overlay; /* COW overlay (stream from this) */
35 BlockDriverState *above_base; /* Node directly above the base */
36 BlockdevOnError on_error;
37 char *backing_file_str;
42 static int coroutine_fn stream_populate(BlockBackend *blk,
43 int64_t offset, uint64_t bytes)
45 assert(bytes < SIZE_MAX);
47 return blk_co_preadv(blk, offset, bytes, NULL,
48 BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
51 static void stream_abort(Job *job)
53 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
55 if (s->chain_frozen) {
56 BlockJob *bjob = &s->common;
57 bdrv_unfreeze_backing_chain(blk_bs(bjob->blk), s->above_base);
61 static int stream_prepare(Job *job)
63 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
64 BlockJob *bjob = &s->common;
65 BlockDriverState *bs = blk_bs(bjob->blk);
66 BlockDriverState *unfiltered_bs = bdrv_skip_filters(bs);
67 BlockDriverState *base = bdrv_filter_or_cow_bs(s->above_base);
68 BlockDriverState *unfiltered_base = bdrv_skip_filters(base);
69 Error *local_err = NULL;
72 bdrv_unfreeze_backing_chain(bs, s->above_base);
73 s->chain_frozen = false;
75 if (bdrv_cow_child(unfiltered_bs)) {
76 const char *base_id = NULL, *base_fmt = NULL;
77 if (unfiltered_base) {
78 base_id = s->backing_file_str ?: unfiltered_base->filename;
79 if (unfiltered_base->drv) {
80 base_fmt = unfiltered_base->drv->format_name;
83 bdrv_set_backing_hd(unfiltered_bs, base, &local_err);
84 ret = bdrv_change_backing_file(unfiltered_bs, base_id, base_fmt, false);
86 error_report_err(local_err);
94 static void stream_clean(Job *job)
96 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
97 BlockJob *bjob = &s->common;
98 BlockDriverState *bs = blk_bs(bjob->blk);
100 /* Reopen the image back in read-only mode if necessary */
101 if (s->bs_read_only) {
102 /* Give up write permissions before making it read-only */
103 blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort);
104 bdrv_reopen_set_read_only(bs, true, NULL);
107 g_free(s->backing_file_str);
110 static int coroutine_fn stream_run(Job *job, Error **errp)
112 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
113 BlockBackend *blk = s->common.blk;
114 BlockDriverState *bs = blk_bs(blk);
115 BlockDriverState *unfiltered_bs = bdrv_skip_filters(bs);
116 bool enable_cor = !bdrv_cow_child(s->base_overlay);
119 uint64_t delay_ns = 0;
121 int64_t n = 0; /* bytes */
123 if (unfiltered_bs == s->base_overlay) {
124 /* Nothing to stream */
128 len = bdrv_getlength(bs);
132 job_progress_set_remaining(&s->common.job, len);
134 /* Turn on copy-on-read for the whole block device so that guest read
135 * requests help us make progress. Only do this when copying the entire
136 * backing chain since the copy-on-read operation does not take base into
140 bdrv_enable_copy_on_read(bs);
143 for ( ; offset < len; offset += n) {
147 /* Note that even when no rate limit is applied we need to yield
148 * with no pending I/O here so that bdrv_drain_all() returns.
150 job_sleep_ns(&s->common.job, delay_ns);
151 if (job_is_cancelled(&s->common.job)) {
157 ret = bdrv_is_allocated(unfiltered_bs, offset, STREAM_CHUNK, &n);
159 /* Allocated in the top, no need to copy. */
160 } else if (ret >= 0) {
161 /* Copy if allocated in the intermediate images. Limit to the
162 * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE). */
163 ret = bdrv_is_allocated_above(bdrv_cow_bs(unfiltered_bs),
164 s->base_overlay, true,
166 /* Finish early if end of backing file has been reached */
167 if (ret == 0 && n == 0) {
173 trace_stream_one_iteration(s, offset, n, ret);
175 ret = stream_populate(blk, offset, n);
178 BlockErrorAction action =
179 block_job_error_action(&s->common, s->on_error, true, -ret);
180 if (action == BLOCK_ERROR_ACTION_STOP) {
187 if (action == BLOCK_ERROR_ACTION_REPORT) {
192 /* Publish progress */
193 job_progress_update(&s->common.job, n);
195 delay_ns = block_job_ratelimit_get_delay(&s->common, n);
202 bdrv_disable_copy_on_read(bs);
205 /* Do not remove the backing file if an error was there but ignored. */
209 static const BlockJobDriver stream_job_driver = {
211 .instance_size = sizeof(StreamBlockJob),
212 .job_type = JOB_TYPE_STREAM,
213 .free = block_job_free,
215 .prepare = stream_prepare,
216 .abort = stream_abort,
217 .clean = stream_clean,
218 .user_resume = block_job_user_resume,
222 void stream_start(const char *job_id, BlockDriverState *bs,
223 BlockDriverState *base, const char *backing_file_str,
224 BlockDriverState *bottom,
225 int creation_flags, int64_t speed,
226 BlockdevOnError on_error,
227 const char *filter_node_name,
231 BlockDriverState *iter;
233 int basic_flags = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
234 BlockDriverState *base_overlay;
235 BlockDriverState *above_base;
237 assert(!(base && bottom));
238 assert(!(backing_file_str && bottom));
242 * New simple interface. The code is written in terms of old interface
243 * with @base parameter (still, it doesn't freeze link to base, so in
244 * this mean old code is correct for new interface). So, for now, just
245 * emulate base_overlay and above_base. Still, when old interface
246 * finally removed, we should refactor code to use only "bottom", but
247 * not "*base*" things.
249 assert(!bottom->drv->is_filter);
250 base_overlay = above_base = bottom;
252 base_overlay = bdrv_find_overlay(bs, base);
254 error_setg(errp, "'%s' is not in the backing chain of '%s'",
255 base->node_name, bs->node_name);
260 * Find the node directly above @base. @base_overlay is a COW overlay,
261 * so it must have a bdrv_cow_child(), but it is the immediate overlay
262 * of @base, so between the two there can only be filters.
264 above_base = base_overlay;
265 if (bdrv_cow_bs(above_base) != base) {
266 above_base = bdrv_cow_bs(above_base);
267 while (bdrv_filter_bs(above_base) != base) {
268 above_base = bdrv_filter_bs(above_base);
273 if (bdrv_freeze_backing_chain(bs, above_base, errp) < 0) {
277 /* Make sure that the image is opened in read-write mode */
278 bs_read_only = bdrv_is_read_only(bs);
280 if (bdrv_reopen_set_read_only(bs, false, errp) != 0) {
281 bs_read_only = false;
286 /* Prevent concurrent jobs trying to modify the graph structure here, we
287 * already have our own plans. Also don't allow resize as the image size is
288 * queried only at the job start and then cached. */
289 s = block_job_create(job_id, &stream_job_driver, NULL, bs,
290 basic_flags | BLK_PERM_GRAPH_MOD,
291 basic_flags | BLK_PERM_WRITE,
292 speed, creation_flags, NULL, NULL, errp);
297 /* Block all intermediate nodes between bs and base, because they will
298 * disappear from the chain after this operation. The streaming job reads
299 * every block only once, assuming that it doesn't change, so forbid writes
300 * and resizes. Reassign the base node pointer because the backing BS of the
301 * bottom node might change after the call to bdrv_reopen_set_read_only()
302 * due to parallel block jobs running.
303 * above_base node might change after the call to
304 * bdrv_reopen_set_read_only() due to parallel block jobs running.
306 base = bdrv_filter_or_cow_bs(above_base);
307 for (iter = bdrv_filter_or_cow_bs(bs); iter != base;
308 iter = bdrv_filter_or_cow_bs(iter))
310 block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
311 basic_flags, &error_abort);
314 s->base_overlay = base_overlay;
315 s->above_base = above_base;
316 s->backing_file_str = g_strdup(backing_file_str);
317 s->bs_read_only = bs_read_only;
318 s->chain_frozen = true;
320 s->on_error = on_error;
321 trace_stream_start(bs, base, s);
322 job_start(&s->common.job);
327 bdrv_reopen_set_read_only(bs, true, NULL);
329 bdrv_unfreeze_backing_chain(bs, above_base);