block: fix deadlock in bdrv_co_flush
[qemu/kevin.git] / docs / bitmaps.md
bloba2e8d511637623e3a128e9b6bb9717516d12ff9f
1 <!--
2 Copyright 2015 John Snow <jsnow@redhat.com> and Red Hat, Inc.
3 All rights reserved.
5 This file is licensed via The FreeBSD Documentation License, the full text of
6 which is included at the end of this document.
7 -->
9 # Dirty Bitmaps and Incremental Backup
11 * Dirty Bitmaps are objects that track which data needs to be backed up for the
12   next incremental backup.
14 * Dirty bitmaps can be created at any time and attached to any node
15   (not just complete drives.)
17 ## Dirty Bitmap Names
19 * A dirty bitmap's name is unique to the node, but bitmaps attached to different
20   nodes can share the same name.
22 * Dirty bitmaps created for internal use by QEMU may be anonymous and have no
23   name, but any user-created bitmaps may not be. There can be any number of
24   anonymous bitmaps per node.
26 * The name of a user-created bitmap must not be empty ("").
28 ## Bitmap Modes
30 * A Bitmap can be "frozen," which means that it is currently in-use by a backup
31   operation and cannot be deleted, renamed, written to, reset,
32   etc.
34 * The normal operating mode for a bitmap is "active."
36 ## Basic QMP Usage
38 ### Supported Commands ###
40 * block-dirty-bitmap-add
41 * block-dirty-bitmap-remove
42 * block-dirty-bitmap-clear
44 ### Creation
46 * To create a new bitmap, enabled, on the drive with id=drive0:
48 ```json
49 { "execute": "block-dirty-bitmap-add",
50   "arguments": {
51     "node": "drive0",
52     "name": "bitmap0"
53   }
55 ```
57 * This bitmap will have a default granularity that matches the cluster size of
58   its associated drive, if available, clamped to between [4KiB, 64KiB].
59   The current default for qcow2 is 64KiB.
61 * To create a new bitmap that tracks changes in 32KiB segments:
63 ```json
64 { "execute": "block-dirty-bitmap-add",
65   "arguments": {
66     "node": "drive0",
67     "name": "bitmap0",
68     "granularity": 32768
69   }
71 ```
73 ### Deletion
75 * Bitmaps that are frozen cannot be deleted.
77 * Deleting the bitmap does not impact any other bitmaps attached to the same
78   node, nor does it affect any backups already created from this node.
80 * Because bitmaps are only unique to the node to which they are attached,
81   you must specify the node/drive name here, too.
83 ```json
84 { "execute": "block-dirty-bitmap-remove",
85   "arguments": {
86     "node": "drive0",
87     "name": "bitmap0"
88   }
90 ```
92 ### Resetting
94 * Resetting a bitmap will clear all information it holds.
96 * An incremental backup created from an empty bitmap will copy no data,
97   as if nothing has changed.
99 ```json
100 { "execute": "block-dirty-bitmap-clear",
101   "arguments": {
102     "node": "drive0",
103     "name": "bitmap0"
104   }
108 ## Transactions
110 ### Justification
112 Bitmaps can be safely modified when the VM is paused or halted by using
113 the basic QMP commands. For instance, you might perform the following actions:
115 1. Boot the VM in a paused state.
116 2. Create a full drive backup of drive0.
117 3. Create a new bitmap attached to drive0.
118 4. Resume execution of the VM.
119 5. Incremental backups are ready to be created.
121 At this point, the bitmap and drive backup would be correctly in sync,
122 and incremental backups made from this point forward would be correctly aligned
123 to the full drive backup.
125 This is not particularly useful if we decide we want to start incremental
126 backups after the VM has been running for a while, for which we will need to
127 perform actions such as the following:
129 1. Boot the VM and begin execution.
130 2. Using a single transaction, perform the following operations:
131     * Create bitmap0.
132     * Create a full drive backup of drive0.
133 3. Incremental backups are now ready to be created.
135 ### Supported Bitmap Transactions
137 * block-dirty-bitmap-add
138 * block-dirty-bitmap-clear
140 The usages are identical to their respective QMP commands, but see below
141 for examples.
143 ### Example: New Incremental Backup
145 As outlined in the justification, perhaps we want to create a new incremental
146 backup chain attached to a drive.
148 ```json
149 { "execute": "transaction",
150   "arguments": {
151     "actions": [
152       {"type": "block-dirty-bitmap-add",
153        "data": {"node": "drive0", "name": "bitmap0"} },
154       {"type": "drive-backup",
155        "data": {"device": "drive0", "target": "/path/to/full_backup.img",
156                 "sync": "full", "format": "qcow2"} }
157     ]
158   }
162 ### Example: New Incremental Backup Anchor Point
164 Maybe we just want to create a new full backup with an existing bitmap and
165 want to reset the bitmap to track the new chain.
167 ```json
168 { "execute": "transaction",
169   "arguments": {
170     "actions": [
171       {"type": "block-dirty-bitmap-clear",
172        "data": {"node": "drive0", "name": "bitmap0"} },
173       {"type": "drive-backup",
174        "data": {"device": "drive0", "target": "/path/to/new_full_backup.img",
175                 "sync": "full", "format": "qcow2"} }
176     ]
177   }
181 ## Incremental Backups
183 The star of the show.
185 **Nota Bene!** Only incremental backups of entire drives are supported for now.
186 So despite the fact that you can attach a bitmap to any arbitrary node, they are
187 only currently useful when attached to the root node. This is because
188 drive-backup only supports drives/devices instead of arbitrary nodes.
190 ### Example: First Incremental Backup
192 1. Create a full backup and sync it to the dirty bitmap, as in the transactional
193 examples above; or with the VM offline, manually create a full copy and then
194 create a new bitmap before the VM begins execution.
196     * Let's assume the full backup is named 'full_backup.img'.
197     * Let's assume the bitmap you created is 'bitmap0' attached to 'drive0'.
199 2. Create a destination image for the incremental backup that utilizes the
200 full backup as a backing image.
202     * Let's assume it is named 'incremental.0.img'.
204     ```sh
205     # qemu-img create -f qcow2 incremental.0.img -b full_backup.img -F qcow2
206     ```
208 3. Issue the incremental backup command:
210     ```json
211     { "execute": "drive-backup",
212       "arguments": {
213         "device": "drive0",
214         "bitmap": "bitmap0",
215         "target": "incremental.0.img",
216         "format": "qcow2",
217         "sync": "incremental",
218         "mode": "existing"
219       }
220     }
221     ```
223 ### Example: Second Incremental Backup
225 1. Create a new destination image for the incremental backup that points to the
226    previous one, e.g.: 'incremental.1.img'
228     ```sh
229     # qemu-img create -f qcow2 incremental.1.img -b incremental.0.img -F qcow2
230     ```
232 2. Issue a new incremental backup command. The only difference here is that we
233    have changed the target image below.
235     ```json
236     { "execute": "drive-backup",
237       "arguments": {
238         "device": "drive0",
239         "bitmap": "bitmap0",
240         "target": "incremental.1.img",
241         "format": "qcow2",
242         "sync": "incremental",
243         "mode": "existing"
244       }
245     }
246     ```
248 ## Errors
250 * In the event of an error that occurs after a backup job is successfully
251   launched, either by a direct QMP command or a QMP transaction, the user
252   will receive a BLOCK_JOB_COMPLETE event with a failure message, accompanied
253   by a BLOCK_JOB_ERROR event.
255 * In the case of an event being cancelled, the user will receive a
256   BLOCK_JOB_CANCELLED event instead of a pair of COMPLETE and ERROR events.
258 * In either case, the incremental backup data contained within the bitmap is
259   safely rolled back, and the data within the bitmap is not lost. The image
260   file created for the failed attempt can be safely deleted.
262 * Once the underlying problem is fixed (e.g. more storage space is freed up),
263   you can simply retry the incremental backup command with the same bitmap.
265 ### Example
267 1. Create a target image:
269     ```sh
270     # qemu-img create -f qcow2 incremental.0.img -b full_backup.img -F qcow2
271     ```
273 2. Attempt to create an incremental backup via QMP:
275     ```json
276     { "execute": "drive-backup",
277       "arguments": {
278         "device": "drive0",
279         "bitmap": "bitmap0",
280         "target": "incremental.0.img",
281         "format": "qcow2",
282         "sync": "incremental",
283         "mode": "existing"
284       }
285     }
286     ```
288 3. Receive an event notifying us of failure:
290     ```json
291     { "timestamp": { "seconds": 1424709442, "microseconds": 844524 },
292       "data": { "speed": 0, "offset": 0, "len": 67108864,
293                 "error": "No space left on device",
294                 "device": "drive1", "type": "backup" },
295       "event": "BLOCK_JOB_COMPLETED" }
296     ```
298 4. Delete the failed incremental, and re-create the image.
300     ```sh
301     # rm incremental.0.img
302     # qemu-img create -f qcow2 incremental.0.img -b full_backup.img -F qcow2
303     ```
305 5. Retry the command after fixing the underlying problem,
306    such as freeing up space on the backup volume:
308     ```json
309     { "execute": "drive-backup",
310       "arguments": {
311         "device": "drive0",
312         "bitmap": "bitmap0",
313         "target": "incremental.0.img",
314         "format": "qcow2",
315         "sync": "incremental",
316         "mode": "existing"
317       }
318     }
319     ```
321 6. Receive confirmation that the job completed successfully:
323     ```json
324     { "timestamp": { "seconds": 1424709668, "microseconds": 526525 },
325       "data": { "device": "drive1", "type": "backup",
326                 "speed": 0, "len": 67108864, "offset": 67108864},
327       "event": "BLOCK_JOB_COMPLETED" }
328     ```
330 ### Partial Transactional Failures
332 * Sometimes, a transaction will succeed in launching and return success,
333   but then later the backup jobs themselves may fail. It is possible that
334   a management application may have to deal with a partial backup failure
335   after a successful transaction.
337 * If multiple backup jobs are specified in a single transaction, when one of
338   them fails, it will not interact with the other backup jobs in any way.
340 * The job(s) that succeeded will clear the dirty bitmap associated with the
341   operation, but the job(s) that failed will not. It is not "safe" to delete
342   any incremental backups that were created successfully in this scenario,
343   even though others failed.
345 #### Example
347 * QMP example highlighting two backup jobs:
349     ```json
350     { "execute": "transaction",
351       "arguments": {
352         "actions": [
353           { "type": "drive-backup",
354             "data": { "device": "drive0", "bitmap": "bitmap0",
355                       "format": "qcow2", "mode": "existing",
356                       "sync": "incremental", "target": "d0-incr-1.qcow2" } },
357           { "type": "drive-backup",
358             "data": { "device": "drive1", "bitmap": "bitmap1",
359                       "format": "qcow2", "mode": "existing",
360                       "sync": "incremental", "target": "d1-incr-1.qcow2" } },
361         ]
362       }
363     }
364     ```
366 * QMP example response, highlighting one success and one failure:
367     * Acknowledgement that the Transaction was accepted and jobs were launched:
368         ```json
369         { "return": {} }
370         ```
372     * Later, QEMU sends notice that the first job was completed:
373         ```json
374         { "timestamp": { "seconds": 1447192343, "microseconds": 615698 },
375           "data": { "device": "drive0", "type": "backup",
376                      "speed": 0, "len": 67108864, "offset": 67108864 },
377           "event": "BLOCK_JOB_COMPLETED"
378         }
379         ```
381     * Later yet, QEMU sends notice that the second job has failed:
382         ```json
383         { "timestamp": { "seconds": 1447192399, "microseconds": 683015 },
384           "data": { "device": "drive1", "action": "report",
385                     "operation": "read" },
386           "event": "BLOCK_JOB_ERROR" }
387         ```
389         ```json
390         { "timestamp": { "seconds": 1447192399, "microseconds": 685853 },
391           "data": { "speed": 0, "offset": 0, "len": 67108864,
392                     "error": "Input/output error",
393                     "device": "drive1", "type": "backup" },
394           "event": "BLOCK_JOB_COMPLETED" }
396 * In the above example, "d0-incr-1.qcow2" is valid and must be kept,
397   but "d1-incr-1.qcow2" is invalid and should be deleted. If a VM-wide
398   incremental backup of all drives at a point-in-time is to be made,
399   new backups for both drives will need to be made, taking into account
400   that a new incremental backup for drive0 needs to be based on top of
401   "d0-incr-1.qcow2."
403 ### Grouped Completion Mode
405 * While jobs launched by transactions normally complete or fail on their own,
406   it is possible to instruct them to complete or fail together as a group.
408 * QMP transactions take an optional properties structure that can affect
409   the semantics of the transaction.
411 * The "completion-mode" transaction property can be either "individual"
412   which is the default, legacy behavior described above, or "grouped,"
413   a new behavior detailed below.
415 * Delayed Completion: In grouped completion mode, no jobs will report
416   success until all jobs are ready to report success.
418 * Grouped failure: If any job fails in grouped completion mode, all remaining
419   jobs will be cancelled. Any incremental backups will restore their dirty
420   bitmap objects as if no backup command was ever issued.
422     * Regardless of if QEMU reports a particular incremental backup job as
423       CANCELLED or as an ERROR, the in-memory bitmap will be restored.
425 #### Example
427 * Here's the same example scenario from above with the new property:
429     ```json
430     { "execute": "transaction",
431       "arguments": {
432         "actions": [
433           { "type": "drive-backup",
434             "data": { "device": "drive0", "bitmap": "bitmap0",
435                       "format": "qcow2", "mode": "existing",
436                       "sync": "incremental", "target": "d0-incr-1.qcow2" } },
437           { "type": "drive-backup",
438             "data": { "device": "drive1", "bitmap": "bitmap1",
439                       "format": "qcow2", "mode": "existing",
440                       "sync": "incremental", "target": "d1-incr-1.qcow2" } },
441         ],
442         "properties": {
443           "completion-mode": "grouped"
444         }
445       }
446     }
447     ```
449 * QMP example response, highlighting a failure for drive2:
450     * Acknowledgement that the Transaction was accepted and jobs were launched:
451         ```json
452         { "return": {} }
453         ```
455     * Later, QEMU sends notice that the second job has errored out,
456       but that the first job was also cancelled:
457         ```json
458         { "timestamp": { "seconds": 1447193702, "microseconds": 632377 },
459           "data": { "device": "drive1", "action": "report",
460                     "operation": "read" },
461           "event": "BLOCK_JOB_ERROR" }
462         ```
464         ```json
465         { "timestamp": { "seconds": 1447193702, "microseconds": 640074 },
466           "data": { "speed": 0, "offset": 0, "len": 67108864,
467                     "error": "Input/output error",
468                     "device": "drive1", "type": "backup" },
469           "event": "BLOCK_JOB_COMPLETED" }
470         ```
472         ```json
473         { "timestamp": { "seconds": 1447193702, "microseconds": 640163 },
474           "data": { "device": "drive0", "type": "backup", "speed": 0,
475                     "len": 67108864, "offset": 16777216 },
476           "event": "BLOCK_JOB_CANCELLED" }
477         ```
479 <!--
480 The FreeBSD Documentation License
482 Redistribution and use in source (Markdown) and 'compiled' forms (SGML, HTML,
483 PDF, PostScript, RTF and so forth) with or without modification, are permitted
484 provided that the following conditions are met:
486 Redistributions of source code (Markdown) must retain the above copyright
487 notice, this list of conditions and the following disclaimer of this file
488 unmodified.
490 Redistributions in compiled form (transformed to other DTDs, converted to PDF,
491 PostScript, RTF and other formats) must reproduce the above copyright notice,
492 this list of conditions and the following disclaimer in the documentation and/or
493 other materials provided with the distribution.
495 THIS DOCUMENTATION IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
496 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
497 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR  PURPOSE ARE
498 DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS  BE LIABLE
499 FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
500 DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
501 SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
502 CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
503 OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
504 THIS DOCUMENTATION, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.