export/fuse: Let permissions be adjustable
[qemu.git] / block / stream.c
blob97bee482dce8f4bc9e4d37c7a53baa9a0c2ff7a5
1 /*
2 * Image streaming
4 * Copyright IBM, Corp. 2011
6 * Authors:
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"
15 #include "trace.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 "qapi/qmp/qdict.h"
21 #include "qemu/ratelimit.h"
22 #include "sysemu/block-backend.h"
23 #include "block/copy-on-read.h"
25 enum {
27 * Maximum chunk size to feed to copy-on-read. This should be
28 * large enough to process multiple clusters in a single call, so
29 * that populating contiguous regions of the image is efficient.
31 STREAM_CHUNK = 512 * 1024, /* in bytes */
34 typedef struct StreamBlockJob {
35 BlockJob common;
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;
42 bool bs_read_only;
43 } StreamBlockJob;
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 = bdrv_filter_or_cow_bs(s->above_base);
58 BlockDriverState *unfiltered_base = bdrv_skip_filters(base);
59 Error *local_err = NULL;
60 int ret = 0;
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;
66 if (bdrv_cow_child(unfiltered_bs)) {
67 const char *base_id = NULL, *base_fmt = NULL;
68 if (unfiltered_base) {
69 base_id = s->backing_file_str ?: unfiltered_base->filename;
70 if (unfiltered_base->drv) {
71 base_fmt = unfiltered_base->drv->format_name;
74 bdrv_set_backing_hd(unfiltered_bs, base, &local_err);
75 ret = bdrv_change_backing_file(unfiltered_bs, base_id, base_fmt, false);
76 if (local_err) {
77 error_report_err(local_err);
78 return -EPERM;
82 return ret;
85 static void stream_clean(Job *job)
87 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
88 BlockJob *bjob = &s->common;
90 if (s->cor_filter_bs) {
91 bdrv_cor_filter_drop(s->cor_filter_bs);
92 s->cor_filter_bs = NULL;
95 /* Reopen the image back in read-only mode if necessary */
96 if (s->bs_read_only) {
97 /* Give up write permissions before making it read-only */
98 blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort);
99 bdrv_reopen_set_read_only(s->target_bs, true, NULL);
102 g_free(s->backing_file_str);
105 static int coroutine_fn stream_run(Job *job, Error **errp)
107 StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
108 BlockBackend *blk = s->common.blk;
109 BlockDriverState *unfiltered_bs = bdrv_skip_filters(s->target_bs);
110 int64_t len;
111 int64_t offset = 0;
112 uint64_t delay_ns = 0;
113 int error = 0;
114 int64_t n = 0; /* bytes */
116 if (unfiltered_bs == s->base_overlay) {
117 /* Nothing to stream */
118 return 0;
121 len = bdrv_getlength(s->target_bs);
122 if (len < 0) {
123 return len;
125 job_progress_set_remaining(&s->common.job, len);
127 for ( ; offset < len; offset += n) {
128 bool copy;
129 int ret;
131 /* Note that even when no rate limit is applied we need to yield
132 * with no pending I/O here so that bdrv_drain_all() returns.
134 job_sleep_ns(&s->common.job, delay_ns);
135 if (job_is_cancelled(&s->common.job)) {
136 break;
139 copy = false;
141 ret = bdrv_is_allocated(unfiltered_bs, offset, STREAM_CHUNK, &n);
142 if (ret == 1) {
143 /* Allocated in the top, no need to copy. */
144 } else if (ret >= 0) {
145 /* Copy if allocated in the intermediate images. Limit to the
146 * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE). */
147 ret = bdrv_is_allocated_above(bdrv_cow_bs(unfiltered_bs),
148 s->base_overlay, true,
149 offset, n, &n);
150 /* Finish early if end of backing file has been reached */
151 if (ret == 0 && n == 0) {
152 n = len - offset;
155 copy = (ret > 0);
157 trace_stream_one_iteration(s, offset, n, ret);
158 if (copy) {
159 ret = stream_populate(blk, offset, n);
161 if (ret < 0) {
162 BlockErrorAction action =
163 block_job_error_action(&s->common, s->on_error, true, -ret);
164 if (action == BLOCK_ERROR_ACTION_STOP) {
165 n = 0;
166 continue;
168 if (error == 0) {
169 error = ret;
171 if (action == BLOCK_ERROR_ACTION_REPORT) {
172 break;
176 /* Publish progress */
177 job_progress_update(&s->common.job, n);
178 if (copy) {
179 delay_ns = block_job_ratelimit_get_delay(&s->common, n);
180 } else {
181 delay_ns = 0;
185 /* Do not remove the backing file if an error was there but ignored. */
186 return error;
189 static const BlockJobDriver stream_job_driver = {
190 .job_driver = {
191 .instance_size = sizeof(StreamBlockJob),
192 .job_type = JOB_TYPE_STREAM,
193 .free = block_job_free,
194 .run = stream_run,
195 .prepare = stream_prepare,
196 .clean = stream_clean,
197 .user_resume = block_job_user_resume,
201 void stream_start(const char *job_id, BlockDriverState *bs,
202 BlockDriverState *base, const char *backing_file_str,
203 BlockDriverState *bottom,
204 int creation_flags, int64_t speed,
205 BlockdevOnError on_error,
206 const char *filter_node_name,
207 Error **errp)
209 StreamBlockJob *s = NULL;
210 BlockDriverState *iter;
211 bool bs_read_only;
212 int basic_flags = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
213 BlockDriverState *base_overlay;
214 BlockDriverState *cor_filter_bs = NULL;
215 BlockDriverState *above_base;
216 QDict *opts;
217 int ret;
219 assert(!(base && bottom));
220 assert(!(backing_file_str && bottom));
222 if (bottom) {
224 * New simple interface. The code is written in terms of old interface
225 * with @base parameter (still, it doesn't freeze link to base, so in
226 * this mean old code is correct for new interface). So, for now, just
227 * emulate base_overlay and above_base. Still, when old interface
228 * finally removed, we should refactor code to use only "bottom", but
229 * not "*base*" things.
231 assert(!bottom->drv->is_filter);
232 base_overlay = above_base = bottom;
233 } else {
234 base_overlay = bdrv_find_overlay(bs, base);
235 if (!base_overlay) {
236 error_setg(errp, "'%s' is not in the backing chain of '%s'",
237 base->node_name, bs->node_name);
238 return;
242 * Find the node directly above @base. @base_overlay is a COW overlay,
243 * so it must have a bdrv_cow_child(), but it is the immediate overlay
244 * of @base, so between the two there can only be filters.
246 above_base = base_overlay;
247 if (bdrv_cow_bs(above_base) != base) {
248 above_base = bdrv_cow_bs(above_base);
249 while (bdrv_filter_bs(above_base) != base) {
250 above_base = bdrv_filter_bs(above_base);
255 /* Make sure that the image is opened in read-write mode */
256 bs_read_only = bdrv_is_read_only(bs);
257 if (bs_read_only) {
258 int ret;
259 /* Hold the chain during reopen */
260 if (bdrv_freeze_backing_chain(bs, above_base, errp) < 0) {
261 return;
264 ret = bdrv_reopen_set_read_only(bs, false, errp);
266 /* failure, or cor-filter will hold the chain */
267 bdrv_unfreeze_backing_chain(bs, above_base);
269 if (ret < 0) {
270 return;
274 opts = qdict_new();
276 qdict_put_str(opts, "driver", "copy-on-read");
277 qdict_put_str(opts, "file", bdrv_get_node_name(bs));
278 /* Pass the base_overlay node name as 'bottom' to COR driver */
279 qdict_put_str(opts, "bottom", base_overlay->node_name);
280 if (filter_node_name) {
281 qdict_put_str(opts, "node-name", filter_node_name);
284 cor_filter_bs = bdrv_insert_node(bs, opts, BDRV_O_RDWR, errp);
285 if (!cor_filter_bs) {
286 goto fail;
289 if (!filter_node_name) {
290 cor_filter_bs->implicit = true;
293 s = block_job_create(job_id, &stream_job_driver, NULL, cor_filter_bs,
294 BLK_PERM_CONSISTENT_READ,
295 basic_flags | BLK_PERM_WRITE,
296 speed, creation_flags, NULL, NULL, errp);
297 if (!s) {
298 goto fail;
302 * Prevent concurrent jobs trying to modify the graph structure here, we
303 * already have our own plans. Also don't allow resize as the image size is
304 * queried only at the job start and then cached.
306 if (block_job_add_bdrv(&s->common, "active node", bs, 0,
307 basic_flags | BLK_PERM_WRITE, errp)) {
308 goto fail;
311 /* Block all intermediate nodes between bs and base, because they will
312 * disappear from the chain after this operation. The streaming job reads
313 * every block only once, assuming that it doesn't change, so forbid writes
314 * and resizes. Reassign the base node pointer because the backing BS of the
315 * bottom node might change after the call to bdrv_reopen_set_read_only()
316 * due to parallel block jobs running.
317 * above_base node might change after the call to
318 * bdrv_reopen_set_read_only() due to parallel block jobs running.
320 base = bdrv_filter_or_cow_bs(above_base);
321 for (iter = bdrv_filter_or_cow_bs(bs); iter != base;
322 iter = bdrv_filter_or_cow_bs(iter))
324 ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
325 basic_flags, errp);
326 if (ret < 0) {
327 goto fail;
331 s->base_overlay = base_overlay;
332 s->above_base = above_base;
333 s->backing_file_str = g_strdup(backing_file_str);
334 s->cor_filter_bs = cor_filter_bs;
335 s->target_bs = bs;
336 s->bs_read_only = bs_read_only;
338 s->on_error = on_error;
339 trace_stream_start(bs, base, s);
340 job_start(&s->common.job);
341 return;
343 fail:
344 if (s) {
345 job_early_fail(&s->common.job);
347 if (cor_filter_bs) {
348 bdrv_cor_filter_drop(cor_filter_bs);
350 if (bs_read_only) {
351 bdrv_reopen_set_read_only(bs, true, NULL);