xref: /openbmc/qemu/block/commit.c (revision d341d9f3)
1 /*
2  * Live block commit
3  *
4  * Copyright Red Hat, Inc. 2012
5  *
6  * Authors:
7  *  Jeff Cody   <jcody@redhat.com>
8  *  Based on stream.c by Stefan Hajnoczi
9  *
10  * This work is licensed under the terms of the GNU LGPL, version 2 or later.
11  * See the COPYING.LIB file in the top-level directory.
12  *
13  */
14 
15 #include "qemu/osdep.h"
16 #include "trace.h"
17 #include "block/block_int.h"
18 #include "block/blockjob.h"
19 #include "qapi/qmp/qerror.h"
20 #include "qemu/ratelimit.h"
21 #include "sysemu/block-backend.h"
22 
23 enum {
24     /*
25      * Size of data buffer for populating the image file.  This should be large
26      * enough to process multiple clusters in a single call, so that populating
27      * contiguous regions of the image is efficient.
28      */
29     COMMIT_BUFFER_SIZE = 512 * 1024, /* in bytes */
30 };
31 
32 #define SLICE_TIME 100000000ULL /* ns */
33 
34 typedef struct CommitBlockJob {
35     BlockJob common;
36     RateLimit limit;
37     BlockDriverState *active;
38     BlockDriverState *top;
39     BlockDriverState *base;
40     BlockdevOnError on_error;
41     int base_flags;
42     int orig_overlay_flags;
43     char *backing_file_str;
44 } CommitBlockJob;
45 
46 static int coroutine_fn commit_populate(BlockDriverState *bs,
47                                         BlockDriverState *base,
48                                         int64_t sector_num, int nb_sectors,
49                                         void *buf)
50 {
51     int ret = 0;
52 
53     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
54     if (ret) {
55         return ret;
56     }
57 
58     ret = bdrv_write(base, sector_num, buf, nb_sectors);
59     if (ret) {
60         return ret;
61     }
62 
63     return 0;
64 }
65 
66 typedef struct {
67     int ret;
68 } CommitCompleteData;
69 
70 static void commit_complete(BlockJob *job, void *opaque)
71 {
72     CommitBlockJob *s = container_of(job, CommitBlockJob, common);
73     CommitCompleteData *data = opaque;
74     BlockDriverState *active = s->active;
75     BlockDriverState *top = s->top;
76     BlockDriverState *base = s->base;
77     BlockDriverState *overlay_bs;
78     int ret = data->ret;
79 
80     if (!block_job_is_cancelled(&s->common) && ret == 0) {
81         /* success */
82         ret = bdrv_drop_intermediate(active, top, base, s->backing_file_str);
83     }
84 
85     /* restore base open flags here if appropriate (e.g., change the base back
86      * to r/o). These reopens do not need to be atomic, since we won't abort
87      * even on failure here */
88     if (s->base_flags != bdrv_get_flags(base)) {
89         bdrv_reopen(base, s->base_flags, NULL);
90     }
91     overlay_bs = bdrv_find_overlay(active, top);
92     if (overlay_bs && s->orig_overlay_flags != bdrv_get_flags(overlay_bs)) {
93         bdrv_reopen(overlay_bs, s->orig_overlay_flags, NULL);
94     }
95     g_free(s->backing_file_str);
96     block_job_completed(&s->common, ret);
97     g_free(data);
98 }
99 
100 static void coroutine_fn commit_run(void *opaque)
101 {
102     CommitBlockJob *s = opaque;
103     CommitCompleteData *data;
104     BlockDriverState *top = s->top;
105     BlockDriverState *base = s->base;
106     int64_t sector_num, end;
107     int ret = 0;
108     int n = 0;
109     void *buf = NULL;
110     int bytes_written = 0;
111     int64_t base_len;
112 
113     ret = s->common.len = bdrv_getlength(top);
114 
115 
116     if (s->common.len < 0) {
117         goto out;
118     }
119 
120     ret = base_len = bdrv_getlength(base);
121     if (base_len < 0) {
122         goto out;
123     }
124 
125     if (base_len < s->common.len) {
126         ret = bdrv_truncate(base, s->common.len);
127         if (ret) {
128             goto out;
129         }
130     }
131 
132     end = s->common.len >> BDRV_SECTOR_BITS;
133     buf = qemu_blockalign(top, COMMIT_BUFFER_SIZE);
134 
135     for (sector_num = 0; sector_num < end; sector_num += n) {
136         uint64_t delay_ns = 0;
137         bool copy;
138 
139 wait:
140         /* Note that even when no rate limit is applied we need to yield
141          * with no pending I/O here so that bdrv_drain_all() returns.
142          */
143         block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
144         if (block_job_is_cancelled(&s->common)) {
145             break;
146         }
147         /* Copy if allocated above the base */
148         ret = bdrv_is_allocated_above(top, base, sector_num,
149                                       COMMIT_BUFFER_SIZE / BDRV_SECTOR_SIZE,
150                                       &n);
151         copy = (ret == 1);
152         trace_commit_one_iteration(s, sector_num, n, ret);
153         if (copy) {
154             if (s->common.speed) {
155                 delay_ns = ratelimit_calculate_delay(&s->limit, n);
156                 if (delay_ns > 0) {
157                     goto wait;
158                 }
159             }
160             ret = commit_populate(top, base, sector_num, n, buf);
161             bytes_written += n * BDRV_SECTOR_SIZE;
162         }
163         if (ret < 0) {
164             if (s->on_error == BLOCKDEV_ON_ERROR_STOP ||
165                 s->on_error == BLOCKDEV_ON_ERROR_REPORT||
166                 (s->on_error == BLOCKDEV_ON_ERROR_ENOSPC && ret == -ENOSPC)) {
167                 goto out;
168             } else {
169                 n = 0;
170                 continue;
171             }
172         }
173         /* Publish progress */
174         s->common.offset += n * BDRV_SECTOR_SIZE;
175     }
176 
177     ret = 0;
178 
179 out:
180     qemu_vfree(buf);
181 
182     data = g_malloc(sizeof(*data));
183     data->ret = ret;
184     block_job_defer_to_main_loop(&s->common, commit_complete, data);
185 }
186 
187 static void commit_set_speed(BlockJob *job, int64_t speed, Error **errp)
188 {
189     CommitBlockJob *s = container_of(job, CommitBlockJob, common);
190 
191     if (speed < 0) {
192         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
193         return;
194     }
195     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
196 }
197 
198 static const BlockJobDriver commit_job_driver = {
199     .instance_size = sizeof(CommitBlockJob),
200     .job_type      = BLOCK_JOB_TYPE_COMMIT,
201     .set_speed     = commit_set_speed,
202 };
203 
204 void commit_start(BlockDriverState *bs, BlockDriverState *base,
205                   BlockDriverState *top, int64_t speed,
206                   BlockdevOnError on_error, BlockCompletionFunc *cb,
207                   void *opaque, const char *backing_file_str, Error **errp)
208 {
209     CommitBlockJob *s;
210     BlockReopenQueue *reopen_queue = NULL;
211     int orig_overlay_flags;
212     int orig_base_flags;
213     BlockDriverState *overlay_bs;
214     Error *local_err = NULL;
215 
216     if ((on_error == BLOCKDEV_ON_ERROR_STOP ||
217          on_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
218         (!bs->blk || !blk_iostatus_is_enabled(bs->blk))) {
219         error_setg(errp, "Invalid parameter combination");
220         return;
221     }
222 
223     assert(top != bs);
224     if (top == base) {
225         error_setg(errp, "Invalid files for merge: top and base are the same");
226         return;
227     }
228 
229     overlay_bs = bdrv_find_overlay(bs, top);
230 
231     if (overlay_bs == NULL) {
232         error_setg(errp, "Could not find overlay image for %s:", top->filename);
233         return;
234     }
235 
236     orig_base_flags    = bdrv_get_flags(base);
237     orig_overlay_flags = bdrv_get_flags(overlay_bs);
238 
239     /* convert base & overlay_bs to r/w, if necessary */
240     if (!(orig_overlay_flags & BDRV_O_RDWR)) {
241         reopen_queue = bdrv_reopen_queue(reopen_queue, overlay_bs, NULL,
242                                          orig_overlay_flags | BDRV_O_RDWR);
243     }
244     if (!(orig_base_flags & BDRV_O_RDWR)) {
245         reopen_queue = bdrv_reopen_queue(reopen_queue, base, NULL,
246                                          orig_base_flags | BDRV_O_RDWR);
247     }
248     if (reopen_queue) {
249         bdrv_reopen_multiple(reopen_queue, &local_err);
250         if (local_err != NULL) {
251             error_propagate(errp, local_err);
252             return;
253         }
254     }
255 
256 
257     s = block_job_create(&commit_job_driver, bs, speed, cb, opaque, errp);
258     if (!s) {
259         return;
260     }
261 
262     s->base   = base;
263     s->top    = top;
264     s->active = bs;
265 
266     s->base_flags          = orig_base_flags;
267     s->orig_overlay_flags  = orig_overlay_flags;
268 
269     s->backing_file_str = g_strdup(backing_file_str);
270 
271     s->on_error = on_error;
272     s->common.co = qemu_coroutine_create(commit_run);
273 
274     trace_commit_start(bs, base, top, s, s->common.co, opaque);
275     qemu_coroutine_enter(s->common.co, s);
276 }
277