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/qdict.h"
20 #include "qemu/ratelimit.h"
21 #include "sysemu/block-backend.h"
22 #include "block/copy-on-read.h"
26 * Maximum chunk size to feed to copy-on-read. This should be
27 * large enough to process multiple clusters in a single call, so
28 * that populating contiguous regions of the image is efficient.
30 STREAM_CHUNK = 512 * 1024, /* in bytes */
33 typedef struct StreamBlockJob {
36 BlockDriverState *base_overlay; /* COW overlay (stream from this) */
37 BlockDriverState *above_base; /* Node directly above the base */
38 BlockDriverState *cor_filter_bs;
39 BlockDriverState *target_bs;
40 BlockdevOnError on_error;
41 char *backing_file_str;
45 static int coroutine_fn stream_populate(BlockBackend *blk,
46 int64_t offset, uint64_t bytes)
48 assert(bytes < SIZE_MAX);
50 return blk_co_preadv(blk, offset, bytes, NULL, BDRV_REQ_PREFETCH);
53 static int stream_prepare(Job *job)
55 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
56 BlockDriverState *unfiltered_bs = bdrv_skip_filters(s->target_bs);
57 BlockDriverState *base;
58 BlockDriverState *unfiltered_base;
59 Error *local_err = NULL;
62 /* We should drop filter at this point, as filter hold the backing chain */
63 bdrv_cor_filter_drop(s->cor_filter_bs);
64 s->cor_filter_bs = NULL;
67 * bdrv_set_backing_hd() requires that unfiltered_bs is drained. Drain
68 * already here and use bdrv_set_backing_hd_drained() instead because
69 * the polling during drained_begin() might change the graph, and if we do
70 * this only later, we may end up working with the wrong base node (or it
71 * might even have gone away by the time we want to use it).
73 bdrv_drained_begin(unfiltered_bs);
75 base = bdrv_filter_or_cow_bs(s->above_base);
76 unfiltered_base = bdrv_skip_filters(base);
78 if (bdrv_cow_child(unfiltered_bs)) {
79 const char *base_id = NULL, *base_fmt = NULL;
80 if (unfiltered_base) {
81 base_id = s->backing_file_str ?: unfiltered_base->filename;
82 if (unfiltered_base->drv) {
83 base_fmt = unfiltered_base->drv->format_name;
87 bdrv_set_backing_hd_drained(unfiltered_bs, base, &local_err);
90 * This call will do I/O, so the graph can change again from here on.
91 * We have already completed the graph change, so we are not in danger
92 * of operating on the wrong node any more if this happens.
94 ret = bdrv_change_backing_file(unfiltered_bs, base_id, base_fmt, false);
96 error_report_err(local_err);
103 bdrv_drained_end(unfiltered_bs);
107 static void stream_clean(Job *job)
109 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
111 if (s->cor_filter_bs) {
112 bdrv_cor_filter_drop(s->cor_filter_bs);
113 s->cor_filter_bs = NULL;
119 /* Reopen the image back in read-only mode if necessary */
120 if (s->bs_read_only) {
121 /* Give up write permissions before making it read-only */
122 bdrv_reopen_set_read_only(s->target_bs, true, NULL);
125 g_free(s->backing_file_str);
128 static int coroutine_fn stream_run(Job *job, Error **errp)
130 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
131 BlockDriverState *unfiltered_bs = bdrv_skip_filters(s->target_bs);
134 uint64_t delay_ns = 0;
136 int64_t n = 0; /* bytes */
138 if (unfiltered_bs == s->base_overlay) {
139 /* Nothing to stream */
143 len = bdrv_getlength(s->target_bs);
147 job_progress_set_remaining(&s->common.job, len);
149 for ( ; offset < len; offset += n) {
153 /* Note that even when no rate limit is applied we need to yield
154 * with no pending I/O here so that bdrv_drain_all() returns.
156 job_sleep_ns(&s->common.job, delay_ns);
157 if (job_is_cancelled(&s->common.job)) {
163 ret = bdrv_is_allocated(unfiltered_bs, offset, STREAM_CHUNK, &n);
165 /* Allocated in the top, no need to copy. */
166 } else if (ret >= 0) {
167 /* Copy if allocated in the intermediate images. Limit to the
168 * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE). */
169 ret = bdrv_is_allocated_above(bdrv_cow_bs(unfiltered_bs),
170 s->base_overlay, true,
172 /* Finish early if end of backing file has been reached */
173 if (ret == 0 && n == 0) {
179 trace_stream_one_iteration(s, offset, n, ret);
181 ret = stream_populate(s->blk, offset, n);
184 BlockErrorAction action =
185 block_job_error_action(&s->common, s->on_error, true, -ret);
186 if (action == BLOCK_ERROR_ACTION_STOP) {
193 if (action == BLOCK_ERROR_ACTION_REPORT) {
198 /* Publish progress */
199 job_progress_update(&s->common.job, n);
201 delay_ns = block_job_ratelimit_get_delay(&s->common, n);
207 /* Do not remove the backing file if an error was there but ignored. */
211 static const BlockJobDriver stream_job_driver = {
213 .instance_size = sizeof(StreamBlockJob),
214 .job_type = JOB_TYPE_STREAM,
215 .free = block_job_free,
217 .prepare = stream_prepare,
218 .clean = stream_clean,
219 .user_resume = block_job_user_resume,
223 void stream_start(const char *job_id, BlockDriverState *bs,
224 BlockDriverState *base, const char *backing_file_str,
225 BlockDriverState *bottom,
226 int creation_flags, int64_t speed,
227 BlockdevOnError on_error,
228 const char *filter_node_name,
231 StreamBlockJob *s = NULL;
232 BlockDriverState *iter;
234 int basic_flags = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
235 BlockDriverState *base_overlay;
236 BlockDriverState *cor_filter_bs = NULL;
237 BlockDriverState *above_base;
243 assert(!(base && bottom));
244 assert(!(backing_file_str && bottom));
248 * New simple interface. The code is written in terms of old interface
249 * with @base parameter (still, it doesn't freeze link to base, so in
250 * this mean old code is correct for new interface). So, for now, just
251 * emulate base_overlay and above_base. Still, when old interface
252 * finally removed, we should refactor code to use only "bottom", but
253 * not "*base*" things.
255 assert(!bottom->drv->is_filter);
256 base_overlay = above_base = bottom;
258 base_overlay = bdrv_find_overlay(bs, base);
260 error_setg(errp, "'%s' is not in the backing chain of '%s'",
261 base->node_name, bs->node_name);
266 * Find the node directly above @base. @base_overlay is a COW overlay,
267 * so it must have a bdrv_cow_child(), but it is the immediate overlay
268 * of @base, so between the two there can only be filters.
270 above_base = base_overlay;
271 if (bdrv_cow_bs(above_base) != base) {
272 above_base = bdrv_cow_bs(above_base);
273 while (bdrv_filter_bs(above_base) != base) {
274 above_base = bdrv_filter_bs(above_base);
279 /* Make sure that the image is opened in read-write mode */
280 bs_read_only = bdrv_is_read_only(bs);
283 /* Hold the chain during reopen */
284 if (bdrv_freeze_backing_chain(bs, above_base, errp) < 0) {
288 ret = bdrv_reopen_set_read_only(bs, false, errp);
290 /* failure, or cor-filter will hold the chain */
291 bdrv_unfreeze_backing_chain(bs, above_base);
300 qdict_put_str(opts, "driver", "copy-on-read");
301 qdict_put_str(opts, "file", bdrv_get_node_name(bs));
302 /* Pass the base_overlay node name as 'bottom' to COR driver */
303 qdict_put_str(opts, "bottom", base_overlay->node_name);
304 if (filter_node_name) {
305 qdict_put_str(opts, "node-name", filter_node_name);
308 cor_filter_bs = bdrv_insert_node(bs, opts, BDRV_O_RDWR, errp);
309 if (!cor_filter_bs) {
313 if (!filter_node_name) {
314 cor_filter_bs->implicit = true;
317 s = block_job_create(job_id, &stream_job_driver, NULL, cor_filter_bs,
319 speed, creation_flags, NULL, NULL, errp);
324 s->blk = blk_new_with_bs(cor_filter_bs, BLK_PERM_CONSISTENT_READ,
325 basic_flags | BLK_PERM_WRITE, errp);
330 * Disable request queuing in the BlockBackend to avoid deadlocks on drain:
331 * The job reports that it's busy until it reaches a pause point.
333 blk_set_disable_request_queuing(s->blk, true);
334 blk_set_allow_aio_context_change(s->blk, true);
337 * Prevent concurrent jobs trying to modify the graph structure here, we
338 * already have our own plans. Also don't allow resize as the image size is
339 * queried only at the job start and then cached.
341 if (block_job_add_bdrv(&s->common, "active node", bs, 0,
342 basic_flags | BLK_PERM_WRITE, errp)) {
346 /* Block all intermediate nodes between bs and base, because they will
347 * disappear from the chain after this operation. The streaming job reads
348 * every block only once, assuming that it doesn't change, so forbid writes
349 * and resizes. Reassign the base node pointer because the backing BS of the
350 * bottom node might change after the call to bdrv_reopen_set_read_only()
351 * due to parallel block jobs running.
352 * above_base node might change after the call to
353 * bdrv_reopen_set_read_only() due to parallel block jobs running.
355 base = bdrv_filter_or_cow_bs(above_base);
356 for (iter = bdrv_filter_or_cow_bs(bs); iter != base;
357 iter = bdrv_filter_or_cow_bs(iter))
359 ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
366 s->base_overlay = base_overlay;
367 s->above_base = above_base;
368 s->backing_file_str = g_strdup(backing_file_str);
369 s->cor_filter_bs = cor_filter_bs;
371 s->bs_read_only = bs_read_only;
373 s->on_error = on_error;
374 trace_stream_start(bs, base, s);
375 job_start(&s->common.job);
380 job_early_fail(&s->common.job);
383 bdrv_cor_filter_drop(cor_filter_bs);
386 bdrv_reopen_set_read_only(bs, true, NULL);