xref: /qemu/blockdev.c (revision 6c1fdcf9)
1 /*
2  * QEMU host block devices
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  *
6  * This work is licensed under the terms of the GNU GPL, version 2 or
7  * later.  See the COPYING file in the top-level directory.
8  */
9 
10 #include "block.h"
11 #include "blockdev.h"
12 #include "monitor.h"
13 #include "qerror.h"
14 #include "qemu-option.h"
15 #include "qemu-config.h"
16 #include "qemu-objects.h"
17 #include "sysemu.h"
18 #include "block_int.h"
19 #include "qmp-commands.h"
20 #include "trace.h"
21 #include "arch_init.h"
22 
23 static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives);
24 
25 static const char *const if_name[IF_COUNT] = {
26     [IF_NONE] = "none",
27     [IF_IDE] = "ide",
28     [IF_SCSI] = "scsi",
29     [IF_FLOPPY] = "floppy",
30     [IF_PFLASH] = "pflash",
31     [IF_MTD] = "mtd",
32     [IF_SD] = "sd",
33     [IF_VIRTIO] = "virtio",
34     [IF_XEN] = "xen",
35 };
36 
37 static const int if_max_devs[IF_COUNT] = {
38     /*
39      * Do not change these numbers!  They govern how drive option
40      * index maps to unit and bus.  That mapping is ABI.
41      *
42      * All controllers used to imlement if=T drives need to support
43      * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
44      * Otherwise, some index values map to "impossible" bus, unit
45      * values.
46      *
47      * For instance, if you change [IF_SCSI] to 255, -drive
48      * if=scsi,index=12 no longer means bus=1,unit=5, but
49      * bus=0,unit=12.  With an lsi53c895a controller (7 units max),
50      * the drive can't be set up.  Regression.
51      */
52     [IF_IDE] = 2,
53     [IF_SCSI] = 7,
54 };
55 
56 /*
57  * We automatically delete the drive when a device using it gets
58  * unplugged.  Questionable feature, but we can't just drop it.
59  * Device models call blockdev_mark_auto_del() to schedule the
60  * automatic deletion, and generic qdev code calls blockdev_auto_del()
61  * when deletion is actually safe.
62  */
63 void blockdev_mark_auto_del(BlockDriverState *bs)
64 {
65     DriveInfo *dinfo = drive_get_by_blockdev(bs);
66 
67     if (dinfo) {
68         dinfo->auto_del = 1;
69     }
70 }
71 
72 void blockdev_auto_del(BlockDriverState *bs)
73 {
74     DriveInfo *dinfo = drive_get_by_blockdev(bs);
75 
76     if (dinfo && dinfo->auto_del) {
77         drive_put_ref(dinfo);
78     }
79 }
80 
81 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
82 {
83     int max_devs = if_max_devs[type];
84     return max_devs ? index / max_devs : 0;
85 }
86 
87 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
88 {
89     int max_devs = if_max_devs[type];
90     return max_devs ? index % max_devs : index;
91 }
92 
93 QemuOpts *drive_def(const char *optstr)
94 {
95     return qemu_opts_parse(qemu_find_opts("drive"), optstr, 0);
96 }
97 
98 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
99                     const char *optstr)
100 {
101     QemuOpts *opts;
102     char buf[32];
103 
104     opts = drive_def(optstr);
105     if (!opts) {
106         return NULL;
107     }
108     if (type != IF_DEFAULT) {
109         qemu_opt_set(opts, "if", if_name[type]);
110     }
111     if (index >= 0) {
112         snprintf(buf, sizeof(buf), "%d", index);
113         qemu_opt_set(opts, "index", buf);
114     }
115     if (file)
116         qemu_opt_set(opts, "file", file);
117     return opts;
118 }
119 
120 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
121 {
122     DriveInfo *dinfo;
123 
124     /* seek interface, bus and unit */
125 
126     QTAILQ_FOREACH(dinfo, &drives, next) {
127         if (dinfo->type == type &&
128 	    dinfo->bus == bus &&
129 	    dinfo->unit == unit)
130             return dinfo;
131     }
132 
133     return NULL;
134 }
135 
136 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
137 {
138     return drive_get(type,
139                      drive_index_to_bus_id(type, index),
140                      drive_index_to_unit_id(type, index));
141 }
142 
143 int drive_get_max_bus(BlockInterfaceType type)
144 {
145     int max_bus;
146     DriveInfo *dinfo;
147 
148     max_bus = -1;
149     QTAILQ_FOREACH(dinfo, &drives, next) {
150         if(dinfo->type == type &&
151            dinfo->bus > max_bus)
152             max_bus = dinfo->bus;
153     }
154     return max_bus;
155 }
156 
157 /* Get a block device.  This should only be used for single-drive devices
158    (e.g. SD/Floppy/MTD).  Multi-disk devices (scsi/ide) should use the
159    appropriate bus.  */
160 DriveInfo *drive_get_next(BlockInterfaceType type)
161 {
162     static int next_block_unit[IF_COUNT];
163 
164     return drive_get(type, 0, next_block_unit[type]++);
165 }
166 
167 DriveInfo *drive_get_by_blockdev(BlockDriverState *bs)
168 {
169     DriveInfo *dinfo;
170 
171     QTAILQ_FOREACH(dinfo, &drives, next) {
172         if (dinfo->bdrv == bs) {
173             return dinfo;
174         }
175     }
176     return NULL;
177 }
178 
179 static void bdrv_format_print(void *opaque, const char *name)
180 {
181     error_printf(" %s", name);
182 }
183 
184 static void drive_uninit(DriveInfo *dinfo)
185 {
186     qemu_opts_del(dinfo->opts);
187     bdrv_delete(dinfo->bdrv);
188     g_free(dinfo->id);
189     QTAILQ_REMOVE(&drives, dinfo, next);
190     g_free(dinfo);
191 }
192 
193 void drive_put_ref(DriveInfo *dinfo)
194 {
195     assert(dinfo->refcount);
196     if (--dinfo->refcount == 0) {
197         drive_uninit(dinfo);
198     }
199 }
200 
201 void drive_get_ref(DriveInfo *dinfo)
202 {
203     dinfo->refcount++;
204 }
205 
206 typedef struct {
207     QEMUBH *bh;
208     DriveInfo *dinfo;
209 } DrivePutRefBH;
210 
211 static void drive_put_ref_bh(void *opaque)
212 {
213     DrivePutRefBH *s = opaque;
214 
215     drive_put_ref(s->dinfo);
216     qemu_bh_delete(s->bh);
217     g_free(s);
218 }
219 
220 /*
221  * Release a drive reference in a BH
222  *
223  * It is not possible to use drive_put_ref() from a callback function when the
224  * callers still need the drive.  In such cases we schedule a BH to release the
225  * reference.
226  */
227 static void drive_put_ref_bh_schedule(DriveInfo *dinfo)
228 {
229     DrivePutRefBH *s;
230 
231     s = g_new(DrivePutRefBH, 1);
232     s->bh = qemu_bh_new(drive_put_ref_bh, s);
233     s->dinfo = dinfo;
234     qemu_bh_schedule(s->bh);
235 }
236 
237 static int parse_block_error_action(const char *buf, int is_read)
238 {
239     if (!strcmp(buf, "ignore")) {
240         return BLOCK_ERR_IGNORE;
241     } else if (!is_read && !strcmp(buf, "enospc")) {
242         return BLOCK_ERR_STOP_ENOSPC;
243     } else if (!strcmp(buf, "stop")) {
244         return BLOCK_ERR_STOP_ANY;
245     } else if (!strcmp(buf, "report")) {
246         return BLOCK_ERR_REPORT;
247     } else {
248         error_report("'%s' invalid %s error action",
249                      buf, is_read ? "read" : "write");
250         return -1;
251     }
252 }
253 
254 static bool do_check_io_limits(BlockIOLimit *io_limits)
255 {
256     bool bps_flag;
257     bool iops_flag;
258 
259     assert(io_limits);
260 
261     bps_flag  = (io_limits->bps[BLOCK_IO_LIMIT_TOTAL] != 0)
262                  && ((io_limits->bps[BLOCK_IO_LIMIT_READ] != 0)
263                  || (io_limits->bps[BLOCK_IO_LIMIT_WRITE] != 0));
264     iops_flag = (io_limits->iops[BLOCK_IO_LIMIT_TOTAL] != 0)
265                  && ((io_limits->iops[BLOCK_IO_LIMIT_READ] != 0)
266                  || (io_limits->iops[BLOCK_IO_LIMIT_WRITE] != 0));
267     if (bps_flag || iops_flag) {
268         return false;
269     }
270 
271     return true;
272 }
273 
274 DriveInfo *drive_init(QemuOpts *opts, int default_to_scsi)
275 {
276     const char *buf;
277     const char *file = NULL;
278     char devname[128];
279     const char *serial;
280     const char *mediastr = "";
281     BlockInterfaceType type;
282     enum { MEDIA_DISK, MEDIA_CDROM } media;
283     int bus_id, unit_id;
284     int cyls, heads, secs, translation;
285     BlockDriver *drv = NULL;
286     int max_devs;
287     int index;
288     int ro = 0;
289     int bdrv_flags = 0;
290     int on_read_error, on_write_error;
291     const char *devaddr;
292     DriveInfo *dinfo;
293     BlockIOLimit io_limits;
294     int snapshot = 0;
295     bool copy_on_read;
296     int ret;
297 
298     translation = BIOS_ATA_TRANSLATION_AUTO;
299     media = MEDIA_DISK;
300 
301     /* extract parameters */
302     bus_id  = qemu_opt_get_number(opts, "bus", 0);
303     unit_id = qemu_opt_get_number(opts, "unit", -1);
304     index   = qemu_opt_get_number(opts, "index", -1);
305 
306     cyls  = qemu_opt_get_number(opts, "cyls", 0);
307     heads = qemu_opt_get_number(opts, "heads", 0);
308     secs  = qemu_opt_get_number(opts, "secs", 0);
309 
310     snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
311     ro = qemu_opt_get_bool(opts, "readonly", 0);
312     copy_on_read = qemu_opt_get_bool(opts, "copy-on-read", false);
313 
314     file = qemu_opt_get(opts, "file");
315     serial = qemu_opt_get(opts, "serial");
316 
317     if ((buf = qemu_opt_get(opts, "if")) != NULL) {
318         pstrcpy(devname, sizeof(devname), buf);
319         for (type = 0; type < IF_COUNT && strcmp(buf, if_name[type]); type++)
320             ;
321         if (type == IF_COUNT) {
322             error_report("unsupported bus type '%s'", buf);
323             return NULL;
324 	}
325     } else {
326         type = default_to_scsi ? IF_SCSI : IF_IDE;
327         pstrcpy(devname, sizeof(devname), if_name[type]);
328     }
329 
330     max_devs = if_max_devs[type];
331 
332     if (cyls || heads || secs) {
333         if (cyls < 1 || (type == IF_IDE && cyls > 16383)) {
334             error_report("invalid physical cyls number");
335 	    return NULL;
336 	}
337         if (heads < 1 || (type == IF_IDE && heads > 16)) {
338             error_report("invalid physical heads number");
339 	    return NULL;
340 	}
341         if (secs < 1 || (type == IF_IDE && secs > 63)) {
342             error_report("invalid physical secs number");
343 	    return NULL;
344 	}
345     }
346 
347     if ((buf = qemu_opt_get(opts, "trans")) != NULL) {
348         if (!cyls) {
349             error_report("'%s' trans must be used with cyls, heads and secs",
350                          buf);
351             return NULL;
352         }
353         if (!strcmp(buf, "none"))
354             translation = BIOS_ATA_TRANSLATION_NONE;
355         else if (!strcmp(buf, "lba"))
356             translation = BIOS_ATA_TRANSLATION_LBA;
357         else if (!strcmp(buf, "auto"))
358             translation = BIOS_ATA_TRANSLATION_AUTO;
359 	else {
360             error_report("'%s' invalid translation type", buf);
361 	    return NULL;
362 	}
363     }
364 
365     if ((buf = qemu_opt_get(opts, "media")) != NULL) {
366         if (!strcmp(buf, "disk")) {
367 	    media = MEDIA_DISK;
368 	} else if (!strcmp(buf, "cdrom")) {
369             if (cyls || secs || heads) {
370                 error_report("CHS can't be set with media=%s", buf);
371 	        return NULL;
372             }
373 	    media = MEDIA_CDROM;
374 	} else {
375 	    error_report("'%s' invalid media", buf);
376 	    return NULL;
377 	}
378     }
379 
380     if ((buf = qemu_opt_get(opts, "cache")) != NULL) {
381         if (bdrv_parse_cache_flags(buf, &bdrv_flags) != 0) {
382             error_report("invalid cache option");
383             return NULL;
384         }
385     }
386 
387 #ifdef CONFIG_LINUX_AIO
388     if ((buf = qemu_opt_get(opts, "aio")) != NULL) {
389         if (!strcmp(buf, "native")) {
390             bdrv_flags |= BDRV_O_NATIVE_AIO;
391         } else if (!strcmp(buf, "threads")) {
392             /* this is the default */
393         } else {
394            error_report("invalid aio option");
395            return NULL;
396         }
397     }
398 #endif
399 
400     if ((buf = qemu_opt_get(opts, "format")) != NULL) {
401        if (strcmp(buf, "?") == 0) {
402            error_printf("Supported formats:");
403            bdrv_iterate_format(bdrv_format_print, NULL);
404            error_printf("\n");
405            return NULL;
406         }
407         drv = bdrv_find_whitelisted_format(buf);
408         if (!drv) {
409             error_report("'%s' invalid format", buf);
410             return NULL;
411         }
412     }
413 
414     /* disk I/O throttling */
415     io_limits.bps[BLOCK_IO_LIMIT_TOTAL]  =
416                            qemu_opt_get_number(opts, "bps", 0);
417     io_limits.bps[BLOCK_IO_LIMIT_READ]   =
418                            qemu_opt_get_number(opts, "bps_rd", 0);
419     io_limits.bps[BLOCK_IO_LIMIT_WRITE]  =
420                            qemu_opt_get_number(opts, "bps_wr", 0);
421     io_limits.iops[BLOCK_IO_LIMIT_TOTAL] =
422                            qemu_opt_get_number(opts, "iops", 0);
423     io_limits.iops[BLOCK_IO_LIMIT_READ]  =
424                            qemu_opt_get_number(opts, "iops_rd", 0);
425     io_limits.iops[BLOCK_IO_LIMIT_WRITE] =
426                            qemu_opt_get_number(opts, "iops_wr", 0);
427 
428     if (!do_check_io_limits(&io_limits)) {
429         error_report("bps(iops) and bps_rd/bps_wr(iops_rd/iops_wr) "
430                      "cannot be used at the same time");
431         return NULL;
432     }
433 
434     on_write_error = BLOCK_ERR_STOP_ENOSPC;
435     if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
436         if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO && type != IF_NONE) {
437             error_report("werror is not supported by this bus type");
438             return NULL;
439         }
440 
441         on_write_error = parse_block_error_action(buf, 0);
442         if (on_write_error < 0) {
443             return NULL;
444         }
445     }
446 
447     on_read_error = BLOCK_ERR_REPORT;
448     if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
449         if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI && type != IF_NONE) {
450             error_report("rerror is not supported by this bus type");
451             return NULL;
452         }
453 
454         on_read_error = parse_block_error_action(buf, 1);
455         if (on_read_error < 0) {
456             return NULL;
457         }
458     }
459 
460     if ((devaddr = qemu_opt_get(opts, "addr")) != NULL) {
461         if (type != IF_VIRTIO) {
462             error_report("addr is not supported by this bus type");
463             return NULL;
464         }
465     }
466 
467     /* compute bus and unit according index */
468 
469     if (index != -1) {
470         if (bus_id != 0 || unit_id != -1) {
471             error_report("index cannot be used with bus and unit");
472             return NULL;
473         }
474         bus_id = drive_index_to_bus_id(type, index);
475         unit_id = drive_index_to_unit_id(type, index);
476     }
477 
478     /* if user doesn't specify a unit_id,
479      * try to find the first free
480      */
481 
482     if (unit_id == -1) {
483        unit_id = 0;
484        while (drive_get(type, bus_id, unit_id) != NULL) {
485            unit_id++;
486            if (max_devs && unit_id >= max_devs) {
487                unit_id -= max_devs;
488                bus_id++;
489            }
490        }
491     }
492 
493     /* check unit id */
494 
495     if (max_devs && unit_id >= max_devs) {
496         error_report("unit %d too big (max is %d)",
497                      unit_id, max_devs - 1);
498         return NULL;
499     }
500 
501     /*
502      * catch multiple definitions
503      */
504 
505     if (drive_get(type, bus_id, unit_id) != NULL) {
506         error_report("drive with bus=%d, unit=%d (index=%d) exists",
507                      bus_id, unit_id, index);
508         return NULL;
509     }
510 
511     /* init */
512 
513     dinfo = g_malloc0(sizeof(*dinfo));
514     if ((buf = qemu_opts_id(opts)) != NULL) {
515         dinfo->id = g_strdup(buf);
516     } else {
517         /* no id supplied -> create one */
518         dinfo->id = g_malloc0(32);
519         if (type == IF_IDE || type == IF_SCSI)
520             mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
521         if (max_devs)
522             snprintf(dinfo->id, 32, "%s%i%s%i",
523                      devname, bus_id, mediastr, unit_id);
524         else
525             snprintf(dinfo->id, 32, "%s%s%i",
526                      devname, mediastr, unit_id);
527     }
528     dinfo->bdrv = bdrv_new(dinfo->id);
529     dinfo->devaddr = devaddr;
530     dinfo->type = type;
531     dinfo->bus = bus_id;
532     dinfo->unit = unit_id;
533     dinfo->opts = opts;
534     dinfo->refcount = 1;
535     if (serial)
536         strncpy(dinfo->serial, serial, sizeof(dinfo->serial) - 1);
537     QTAILQ_INSERT_TAIL(&drives, dinfo, next);
538 
539     bdrv_set_on_error(dinfo->bdrv, on_read_error, on_write_error);
540 
541     /* disk I/O throttling */
542     bdrv_set_io_limits(dinfo->bdrv, &io_limits);
543 
544     switch(type) {
545     case IF_IDE:
546     case IF_SCSI:
547     case IF_XEN:
548     case IF_NONE:
549         switch(media) {
550 	case MEDIA_DISK:
551             if (cyls != 0) {
552                 bdrv_set_geometry_hint(dinfo->bdrv, cyls, heads, secs);
553                 bdrv_set_translation_hint(dinfo->bdrv, translation);
554             }
555 	    break;
556 	case MEDIA_CDROM:
557             dinfo->media_cd = 1;
558 	    break;
559 	}
560         break;
561     case IF_SD:
562     case IF_FLOPPY:
563     case IF_PFLASH:
564     case IF_MTD:
565         break;
566     case IF_VIRTIO:
567         /* add virtio block device */
568         opts = qemu_opts_create(qemu_find_opts("device"), NULL, 0);
569         if (arch_type == QEMU_ARCH_S390X) {
570             qemu_opt_set(opts, "driver", "virtio-blk-s390");
571         } else {
572             qemu_opt_set(opts, "driver", "virtio-blk-pci");
573         }
574         qemu_opt_set(opts, "drive", dinfo->id);
575         if (devaddr)
576             qemu_opt_set(opts, "addr", devaddr);
577         break;
578     default:
579         abort();
580     }
581     if (!file || !*file) {
582         return dinfo;
583     }
584     if (snapshot) {
585         /* always use cache=unsafe with snapshot */
586         bdrv_flags &= ~BDRV_O_CACHE_MASK;
587         bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
588     }
589 
590     if (copy_on_read) {
591         bdrv_flags |= BDRV_O_COPY_ON_READ;
592     }
593 
594     if (media == MEDIA_CDROM) {
595         /* CDROM is fine for any interface, don't check.  */
596         ro = 1;
597     } else if (ro == 1) {
598         if (type != IF_SCSI && type != IF_VIRTIO && type != IF_FLOPPY &&
599             type != IF_NONE && type != IF_PFLASH) {
600             error_report("readonly not supported by this bus type");
601             goto err;
602         }
603     }
604 
605     bdrv_flags |= ro ? 0 : BDRV_O_RDWR;
606 
607     ret = bdrv_open(dinfo->bdrv, file, bdrv_flags, drv);
608     if (ret < 0) {
609         error_report("could not open disk image %s: %s",
610                      file, strerror(-ret));
611         goto err;
612     }
613 
614     if (bdrv_key_required(dinfo->bdrv))
615         autostart = 0;
616     return dinfo;
617 
618 err:
619     bdrv_delete(dinfo->bdrv);
620     g_free(dinfo->id);
621     QTAILQ_REMOVE(&drives, dinfo, next);
622     g_free(dinfo);
623     return NULL;
624 }
625 
626 void do_commit(Monitor *mon, const QDict *qdict)
627 {
628     const char *device = qdict_get_str(qdict, "device");
629     BlockDriverState *bs;
630 
631     if (!strcmp(device, "all")) {
632         bdrv_commit_all();
633     } else {
634         int ret;
635 
636         bs = bdrv_find(device);
637         if (!bs) {
638             qerror_report(QERR_DEVICE_NOT_FOUND, device);
639             return;
640         }
641         ret = bdrv_commit(bs);
642         if (ret == -EBUSY) {
643             qerror_report(QERR_DEVICE_IN_USE, device);
644             return;
645         }
646     }
647 }
648 
649 void qmp_blockdev_snapshot_sync(const char *device, const char *snapshot_file,
650                                 bool has_format, const char *format,
651                                 Error **errp)
652 {
653     BlockDriverState *bs;
654     BlockDriver *drv, *old_drv, *proto_drv;
655     int ret = 0;
656     int flags;
657     char old_filename[1024];
658 
659     bs = bdrv_find(device);
660     if (!bs) {
661         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
662         return;
663     }
664     if (bdrv_in_use(bs)) {
665         error_set(errp, QERR_DEVICE_IN_USE, device);
666         return;
667     }
668 
669     pstrcpy(old_filename, sizeof(old_filename), bs->filename);
670 
671     old_drv = bs->drv;
672     flags = bs->open_flags;
673 
674     if (!has_format) {
675         format = "qcow2";
676     }
677 
678     drv = bdrv_find_format(format);
679     if (!drv) {
680         error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
681         return;
682     }
683 
684     proto_drv = bdrv_find_protocol(snapshot_file);
685     if (!proto_drv) {
686         error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
687         return;
688     }
689 
690     ret = bdrv_img_create(snapshot_file, format, bs->filename,
691                           bs->drv->format_name, NULL, -1, flags);
692     if (ret) {
693         error_set(errp, QERR_UNDEFINED_ERROR);
694         return;
695     }
696 
697     bdrv_drain_all();
698     bdrv_flush(bs);
699 
700     bdrv_close(bs);
701     ret = bdrv_open(bs, snapshot_file, flags, drv);
702     /*
703      * If reopening the image file we just created fails, fall back
704      * and try to re-open the original image. If that fails too, we
705      * are in serious trouble.
706      */
707     if (ret != 0) {
708         ret = bdrv_open(bs, old_filename, flags, old_drv);
709         if (ret != 0) {
710             error_set(errp, QERR_OPEN_FILE_FAILED, old_filename);
711         } else {
712             error_set(errp, QERR_OPEN_FILE_FAILED, snapshot_file);
713         }
714     }
715 }
716 
717 
718 /* New and old BlockDriverState structs for group snapshots */
719 typedef struct BlkGroupSnapshotStates {
720     BlockDriverState *old_bs;
721     BlockDriverState *new_bs;
722     QSIMPLEQ_ENTRY(BlkGroupSnapshotStates) entry;
723 } BlkGroupSnapshotStates;
724 
725 /*
726  * 'Atomic' group snapshots.  The snapshots are taken as a set, and if any fail
727  *  then we do not pivot any of the devices in the group, and abandon the
728  *  snapshots
729  */
730 void qmp_blockdev_group_snapshot_sync(SnapshotDevList *dev_list,
731                                       Error **errp)
732 {
733     int ret = 0;
734     SnapshotDevList *dev_entry = dev_list;
735     SnapshotDev *dev_info = NULL;
736     BlkGroupSnapshotStates *states;
737     BlockDriver *proto_drv;
738     BlockDriver *drv;
739     int flags;
740     const char *format;
741     const char *snapshot_file;
742 
743     QSIMPLEQ_HEAD(snap_bdrv_states, BlkGroupSnapshotStates) snap_bdrv_states;
744     QSIMPLEQ_INIT(&snap_bdrv_states);
745 
746     /* drain all i/o before any snapshots */
747     bdrv_drain_all();
748 
749     /* We don't do anything in this loop that commits us to the snapshot */
750     while (NULL != dev_entry) {
751         dev_info = dev_entry->value;
752         dev_entry = dev_entry->next;
753 
754         states = g_malloc0(sizeof(BlkGroupSnapshotStates));
755         QSIMPLEQ_INSERT_TAIL(&snap_bdrv_states, states, entry);
756 
757         states->old_bs = bdrv_find(dev_info->device);
758 
759         if (!states->old_bs) {
760             error_set(errp, QERR_DEVICE_NOT_FOUND, dev_info->device);
761             goto delete_and_fail;
762         }
763 
764         if (bdrv_in_use(states->old_bs)) {
765             error_set(errp, QERR_DEVICE_IN_USE, dev_info->device);
766             goto delete_and_fail;
767         }
768 
769         if (!bdrv_is_read_only(states->old_bs) &&
770              bdrv_is_inserted(states->old_bs)) {
771 
772             if (bdrv_flush(states->old_bs)) {
773                 error_set(errp, QERR_IO_ERROR);
774                 goto delete_and_fail;
775             }
776         }
777 
778         snapshot_file = dev_info->snapshot_file;
779 
780         flags = states->old_bs->open_flags;
781 
782         if (!dev_info->has_format) {
783             format = "qcow2";
784         } else {
785             format = dev_info->format;
786         }
787 
788         drv = bdrv_find_format(format);
789         if (!drv) {
790             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
791             goto delete_and_fail;
792         }
793 
794         proto_drv = bdrv_find_protocol(snapshot_file);
795         if (!proto_drv) {
796             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
797             goto delete_and_fail;
798         }
799 
800         /* create new image w/backing file */
801         ret = bdrv_img_create(snapshot_file, format,
802                               states->old_bs->filename,
803                               drv->format_name, NULL, -1, flags);
804         if (ret) {
805             error_set(errp, QERR_OPEN_FILE_FAILED, snapshot_file);
806             goto delete_and_fail;
807         }
808 
809         /* We will manually add the backing_hd field to the bs later */
810         states->new_bs = bdrv_new("");
811         ret = bdrv_open(states->new_bs, snapshot_file,
812                         flags | BDRV_O_NO_BACKING, drv);
813         if (ret != 0) {
814             error_set(errp, QERR_OPEN_FILE_FAILED, snapshot_file);
815             goto delete_and_fail;
816         }
817     }
818 
819 
820     /* Now we are going to do the actual pivot.  Everything up to this point
821      * is reversible, but we are committed at this point */
822     QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) {
823         /* This removes our old bs from the bdrv_states, and adds the new bs */
824         bdrv_append(states->new_bs, states->old_bs);
825     }
826 
827     /* success */
828     goto exit;
829 
830 delete_and_fail:
831     /*
832     * failure, and it is all-or-none; abandon each new bs, and keep using
833     * the original bs for all images
834     */
835     QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) {
836         if (states->new_bs) {
837              bdrv_delete(states->new_bs);
838         }
839     }
840 exit:
841     QSIMPLEQ_FOREACH(states, &snap_bdrv_states, entry) {
842         g_free(states);
843     }
844     return;
845 }
846 
847 
848 static void eject_device(BlockDriverState *bs, int force, Error **errp)
849 {
850     if (bdrv_in_use(bs)) {
851         error_set(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs));
852         return;
853     }
854     if (!bdrv_dev_has_removable_media(bs)) {
855         error_set(errp, QERR_DEVICE_NOT_REMOVABLE, bdrv_get_device_name(bs));
856         return;
857     }
858 
859     if (bdrv_dev_is_medium_locked(bs) && !bdrv_dev_is_tray_open(bs)) {
860         bdrv_dev_eject_request(bs, force);
861         if (!force) {
862             error_set(errp, QERR_DEVICE_LOCKED, bdrv_get_device_name(bs));
863             return;
864         }
865     }
866 
867     bdrv_close(bs);
868 }
869 
870 void qmp_eject(const char *device, bool has_force, bool force, Error **errp)
871 {
872     BlockDriverState *bs;
873 
874     bs = bdrv_find(device);
875     if (!bs) {
876         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
877         return;
878     }
879 
880     eject_device(bs, force, errp);
881 }
882 
883 void qmp_block_passwd(const char *device, const char *password, Error **errp)
884 {
885     BlockDriverState *bs;
886     int err;
887 
888     bs = bdrv_find(device);
889     if (!bs) {
890         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
891         return;
892     }
893 
894     err = bdrv_set_key(bs, password);
895     if (err == -EINVAL) {
896         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
897         return;
898     } else if (err < 0) {
899         error_set(errp, QERR_INVALID_PASSWORD);
900         return;
901     }
902 }
903 
904 static void qmp_bdrv_open_encrypted(BlockDriverState *bs, const char *filename,
905                                     int bdrv_flags, BlockDriver *drv,
906                                     const char *password, Error **errp)
907 {
908     if (bdrv_open(bs, filename, bdrv_flags, drv) < 0) {
909         error_set(errp, QERR_OPEN_FILE_FAILED, filename);
910         return;
911     }
912 
913     if (bdrv_key_required(bs)) {
914         if (password) {
915             if (bdrv_set_key(bs, password) < 0) {
916                 error_set(errp, QERR_INVALID_PASSWORD);
917             }
918         } else {
919             error_set(errp, QERR_DEVICE_ENCRYPTED, bdrv_get_device_name(bs),
920                       bdrv_get_encrypted_filename(bs));
921         }
922     } else if (password) {
923         error_set(errp, QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
924     }
925 }
926 
927 void qmp_change_blockdev(const char *device, const char *filename,
928                          bool has_format, const char *format, Error **errp)
929 {
930     BlockDriverState *bs;
931     BlockDriver *drv = NULL;
932     int bdrv_flags;
933     Error *err = NULL;
934 
935     bs = bdrv_find(device);
936     if (!bs) {
937         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
938         return;
939     }
940 
941     if (format) {
942         drv = bdrv_find_whitelisted_format(format);
943         if (!drv) {
944             error_set(errp, QERR_INVALID_BLOCK_FORMAT, format);
945             return;
946         }
947     }
948 
949     eject_device(bs, 0, &err);
950     if (error_is_set(&err)) {
951         error_propagate(errp, err);
952         return;
953     }
954 
955     bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
956     bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
957 
958     qmp_bdrv_open_encrypted(bs, filename, bdrv_flags, drv, NULL, errp);
959 }
960 
961 /* throttling disk I/O limits */
962 void qmp_block_set_io_throttle(const char *device, int64_t bps, int64_t bps_rd,
963                                int64_t bps_wr, int64_t iops, int64_t iops_rd,
964                                int64_t iops_wr, Error **errp)
965 {
966     BlockIOLimit io_limits;
967     BlockDriverState *bs;
968 
969     bs = bdrv_find(device);
970     if (!bs) {
971         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
972         return;
973     }
974 
975     io_limits.bps[BLOCK_IO_LIMIT_TOTAL] = bps;
976     io_limits.bps[BLOCK_IO_LIMIT_READ]  = bps_rd;
977     io_limits.bps[BLOCK_IO_LIMIT_WRITE] = bps_wr;
978     io_limits.iops[BLOCK_IO_LIMIT_TOTAL]= iops;
979     io_limits.iops[BLOCK_IO_LIMIT_READ] = iops_rd;
980     io_limits.iops[BLOCK_IO_LIMIT_WRITE]= iops_wr;
981 
982     if (!do_check_io_limits(&io_limits)) {
983         error_set(errp, QERR_INVALID_PARAMETER_COMBINATION);
984         return;
985     }
986 
987     bs->io_limits = io_limits;
988     bs->slice_time = BLOCK_IO_SLICE_TIME;
989 
990     if (!bs->io_limits_enabled && bdrv_io_limits_enabled(bs)) {
991         bdrv_io_limits_enable(bs);
992     } else if (bs->io_limits_enabled && !bdrv_io_limits_enabled(bs)) {
993         bdrv_io_limits_disable(bs);
994     } else {
995         if (bs->block_timer) {
996             qemu_mod_timer(bs->block_timer, qemu_get_clock_ns(vm_clock));
997         }
998     }
999 }
1000 
1001 int do_drive_del(Monitor *mon, const QDict *qdict, QObject **ret_data)
1002 {
1003     const char *id = qdict_get_str(qdict, "id");
1004     BlockDriverState *bs;
1005 
1006     bs = bdrv_find(id);
1007     if (!bs) {
1008         qerror_report(QERR_DEVICE_NOT_FOUND, id);
1009         return -1;
1010     }
1011     if (bdrv_in_use(bs)) {
1012         qerror_report(QERR_DEVICE_IN_USE, id);
1013         return -1;
1014     }
1015 
1016     /* quiesce block driver; prevent further io */
1017     bdrv_drain_all();
1018     bdrv_flush(bs);
1019     bdrv_close(bs);
1020 
1021     /* if we have a device attached to this BlockDriverState
1022      * then we need to make the drive anonymous until the device
1023      * can be removed.  If this is a drive with no device backing
1024      * then we can just get rid of the block driver state right here.
1025      */
1026     if (bdrv_get_attached_dev(bs)) {
1027         bdrv_make_anon(bs);
1028     } else {
1029         drive_uninit(drive_get_by_blockdev(bs));
1030     }
1031 
1032     return 0;
1033 }
1034 
1035 void qmp_block_resize(const char *device, int64_t size, Error **errp)
1036 {
1037     BlockDriverState *bs;
1038 
1039     bs = bdrv_find(device);
1040     if (!bs) {
1041         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1042         return;
1043     }
1044 
1045     if (size < 0) {
1046         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "size", "a >0 size");
1047         return;
1048     }
1049 
1050     switch (bdrv_truncate(bs, size)) {
1051     case 0:
1052         break;
1053     case -ENOMEDIUM:
1054         error_set(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
1055         break;
1056     case -ENOTSUP:
1057         error_set(errp, QERR_UNSUPPORTED);
1058         break;
1059     case -EACCES:
1060         error_set(errp, QERR_DEVICE_IS_READ_ONLY, device);
1061         break;
1062     case -EBUSY:
1063         error_set(errp, QERR_DEVICE_IN_USE, device);
1064         break;
1065     default:
1066         error_set(errp, QERR_UNDEFINED_ERROR);
1067         break;
1068     }
1069 }
1070 
1071 static QObject *qobject_from_block_job(BlockJob *job)
1072 {
1073     return qobject_from_jsonf("{ 'type': %s,"
1074                               "'device': %s,"
1075                               "'len': %" PRId64 ","
1076                               "'offset': %" PRId64 ","
1077                               "'speed': %" PRId64 " }",
1078                               job->job_type->job_type,
1079                               bdrv_get_device_name(job->bs),
1080                               job->len,
1081                               job->offset,
1082                               job->speed);
1083 }
1084 
1085 static void block_stream_cb(void *opaque, int ret)
1086 {
1087     BlockDriverState *bs = opaque;
1088     QObject *obj;
1089 
1090     trace_block_stream_cb(bs, bs->job, ret);
1091 
1092     assert(bs->job);
1093     obj = qobject_from_block_job(bs->job);
1094     if (ret < 0) {
1095         QDict *dict = qobject_to_qdict(obj);
1096         qdict_put(dict, "error", qstring_from_str(strerror(-ret)));
1097     }
1098 
1099     if (block_job_is_cancelled(bs->job)) {
1100         monitor_protocol_event(QEVENT_BLOCK_JOB_CANCELLED, obj);
1101     } else {
1102         monitor_protocol_event(QEVENT_BLOCK_JOB_COMPLETED, obj);
1103     }
1104     qobject_decref(obj);
1105 
1106     drive_put_ref_bh_schedule(drive_get_by_blockdev(bs));
1107 }
1108 
1109 void qmp_block_stream(const char *device, bool has_base,
1110                       const char *base, Error **errp)
1111 {
1112     BlockDriverState *bs;
1113     BlockDriverState *base_bs = NULL;
1114     int ret;
1115 
1116     bs = bdrv_find(device);
1117     if (!bs) {
1118         error_set(errp, QERR_DEVICE_NOT_FOUND, device);
1119         return;
1120     }
1121 
1122     if (base) {
1123         base_bs = bdrv_find_backing_image(bs, base);
1124         if (base_bs == NULL) {
1125             error_set(errp, QERR_BASE_NOT_FOUND, base);
1126             return;
1127         }
1128     }
1129 
1130     ret = stream_start(bs, base_bs, base, block_stream_cb, bs);
1131     if (ret < 0) {
1132         switch (ret) {
1133         case -EBUSY:
1134             error_set(errp, QERR_DEVICE_IN_USE, device);
1135             return;
1136         default:
1137             error_set(errp, QERR_NOT_SUPPORTED);
1138             return;
1139         }
1140     }
1141 
1142     /* Grab a reference so hotplug does not delete the BlockDriverState from
1143      * underneath us.
1144      */
1145     drive_get_ref(drive_get_by_blockdev(bs));
1146 
1147     trace_qmp_block_stream(bs, bs->job);
1148 }
1149 
1150 static BlockJob *find_block_job(const char *device)
1151 {
1152     BlockDriverState *bs;
1153 
1154     bs = bdrv_find(device);
1155     if (!bs || !bs->job) {
1156         return NULL;
1157     }
1158     return bs->job;
1159 }
1160 
1161 void qmp_block_job_set_speed(const char *device, int64_t value, Error **errp)
1162 {
1163     BlockJob *job = find_block_job(device);
1164 
1165     if (!job) {
1166         error_set(errp, QERR_DEVICE_NOT_ACTIVE, device);
1167         return;
1168     }
1169 
1170     if (block_job_set_speed(job, value) < 0) {
1171         error_set(errp, QERR_NOT_SUPPORTED);
1172     }
1173 }
1174 
1175 void qmp_block_job_cancel(const char *device, Error **errp)
1176 {
1177     BlockJob *job = find_block_job(device);
1178 
1179     if (!job) {
1180         error_set(errp, QERR_DEVICE_NOT_ACTIVE, device);
1181         return;
1182     }
1183 
1184     trace_qmp_block_job_cancel(job);
1185     block_job_cancel(job);
1186 }
1187 
1188 static void do_qmp_query_block_jobs_one(void *opaque, BlockDriverState *bs)
1189 {
1190     BlockJobInfoList **prev = opaque;
1191     BlockJob *job = bs->job;
1192 
1193     if (job) {
1194         BlockJobInfoList *elem;
1195         BlockJobInfo *info = g_new(BlockJobInfo, 1);
1196         *info = (BlockJobInfo){
1197             .type   = g_strdup(job->job_type->job_type),
1198             .device = g_strdup(bdrv_get_device_name(bs)),
1199             .len    = job->len,
1200             .offset = job->offset,
1201             .speed  = job->speed,
1202         };
1203 
1204         elem = g_new0(BlockJobInfoList, 1);
1205         elem->value = info;
1206 
1207         (*prev)->next = elem;
1208         *prev = elem;
1209     }
1210 }
1211 
1212 BlockJobInfoList *qmp_query_block_jobs(Error **errp)
1213 {
1214     /* Dummy is a fake list element for holding the head pointer */
1215     BlockJobInfoList dummy = {};
1216     BlockJobInfoList *prev = &dummy;
1217     bdrv_iterate(do_qmp_query_block_jobs_one, &prev);
1218     return dummy.next;
1219 }
1220