rbd: use the higher level librbd instead of just librados
[qemu-dev-zwu.git] / blockdev.c
blob1502575acb9d359aec4c66940b7cc28d2ea2b9ef
1 /*
2 * QEMU host block devices
4 * Copyright (c) 2003-2008 Fabrice Bellard
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 */
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 "sysemu.h"
17 #include "hw/qdev.h"
18 #include "block_int.h"
20 static QTAILQ_HEAD(drivelist, DriveInfo) drives = QTAILQ_HEAD_INITIALIZER(drives);
22 static const char *const if_name[IF_COUNT] = {
23 [IF_NONE] = "none",
24 [IF_IDE] = "ide",
25 [IF_SCSI] = "scsi",
26 [IF_FLOPPY] = "floppy",
27 [IF_PFLASH] = "pflash",
28 [IF_MTD] = "mtd",
29 [IF_SD] = "sd",
30 [IF_VIRTIO] = "virtio",
31 [IF_XEN] = "xen",
34 static const int if_max_devs[IF_COUNT] = {
36 * Do not change these numbers! They govern how drive option
37 * index maps to unit and bus. That mapping is ABI.
39 * All controllers used to imlement if=T drives need to support
40 * if_max_devs[T] units, for any T with if_max_devs[T] != 0.
41 * Otherwise, some index values map to "impossible" bus, unit
42 * values.
44 * For instance, if you change [IF_SCSI] to 255, -drive
45 * if=scsi,index=12 no longer means bus=1,unit=5, but
46 * bus=0,unit=12. With an lsi53c895a controller (7 units max),
47 * the drive can't be set up. Regression.
49 [IF_IDE] = 2,
50 [IF_SCSI] = 7,
54 * We automatically delete the drive when a device using it gets
55 * unplugged. Questionable feature, but we can't just drop it.
56 * Device models call blockdev_mark_auto_del() to schedule the
57 * automatic deletion, and generic qdev code calls blockdev_auto_del()
58 * when deletion is actually safe.
60 void blockdev_mark_auto_del(BlockDriverState *bs)
62 DriveInfo *dinfo = drive_get_by_blockdev(bs);
64 if (dinfo) {
65 dinfo->auto_del = 1;
69 void blockdev_auto_del(BlockDriverState *bs)
71 DriveInfo *dinfo = drive_get_by_blockdev(bs);
73 if (dinfo && dinfo->auto_del) {
74 drive_put_ref(dinfo);
78 static int drive_index_to_bus_id(BlockInterfaceType type, int index)
80 int max_devs = if_max_devs[type];
81 return max_devs ? index / max_devs : 0;
84 static int drive_index_to_unit_id(BlockInterfaceType type, int index)
86 int max_devs = if_max_devs[type];
87 return max_devs ? index % max_devs : index;
90 QemuOpts *drive_def(const char *optstr)
92 return qemu_opts_parse(qemu_find_opts("drive"), optstr, 0);
95 QemuOpts *drive_add(BlockInterfaceType type, int index, const char *file,
96 const char *optstr)
98 QemuOpts *opts;
99 char buf[32];
101 opts = drive_def(optstr);
102 if (!opts) {
103 return NULL;
105 if (type != IF_DEFAULT) {
106 qemu_opt_set(opts, "if", if_name[type]);
108 if (index >= 0) {
109 snprintf(buf, sizeof(buf), "%d", index);
110 qemu_opt_set(opts, "index", buf);
112 if (file)
113 qemu_opt_set(opts, "file", file);
114 return opts;
117 DriveInfo *drive_get(BlockInterfaceType type, int bus, int unit)
119 DriveInfo *dinfo;
121 /* seek interface, bus and unit */
123 QTAILQ_FOREACH(dinfo, &drives, next) {
124 if (dinfo->type == type &&
125 dinfo->bus == bus &&
126 dinfo->unit == unit)
127 return dinfo;
130 return NULL;
133 DriveInfo *drive_get_by_index(BlockInterfaceType type, int index)
135 return drive_get(type,
136 drive_index_to_bus_id(type, index),
137 drive_index_to_unit_id(type, index));
140 int drive_get_max_bus(BlockInterfaceType type)
142 int max_bus;
143 DriveInfo *dinfo;
145 max_bus = -1;
146 QTAILQ_FOREACH(dinfo, &drives, next) {
147 if(dinfo->type == type &&
148 dinfo->bus > max_bus)
149 max_bus = dinfo->bus;
151 return max_bus;
154 /* Get a block device. This should only be used for single-drive devices
155 (e.g. SD/Floppy/MTD). Multi-disk devices (scsi/ide) should use the
156 appropriate bus. */
157 DriveInfo *drive_get_next(BlockInterfaceType type)
159 static int next_block_unit[IF_COUNT];
161 return drive_get(type, 0, next_block_unit[type]++);
164 DriveInfo *drive_get_by_blockdev(BlockDriverState *bs)
166 DriveInfo *dinfo;
168 QTAILQ_FOREACH(dinfo, &drives, next) {
169 if (dinfo->bdrv == bs) {
170 return dinfo;
173 return NULL;
176 static void bdrv_format_print(void *opaque, const char *name)
178 error_printf(" %s", name);
181 static void drive_uninit(DriveInfo *dinfo)
183 qemu_opts_del(dinfo->opts);
184 bdrv_delete(dinfo->bdrv);
185 qemu_free(dinfo->id);
186 QTAILQ_REMOVE(&drives, dinfo, next);
187 qemu_free(dinfo);
190 void drive_put_ref(DriveInfo *dinfo)
192 assert(dinfo->refcount);
193 if (--dinfo->refcount == 0) {
194 drive_uninit(dinfo);
198 void drive_get_ref(DriveInfo *dinfo)
200 dinfo->refcount++;
203 static int parse_block_error_action(const char *buf, int is_read)
205 if (!strcmp(buf, "ignore")) {
206 return BLOCK_ERR_IGNORE;
207 } else if (!is_read && !strcmp(buf, "enospc")) {
208 return BLOCK_ERR_STOP_ENOSPC;
209 } else if (!strcmp(buf, "stop")) {
210 return BLOCK_ERR_STOP_ANY;
211 } else if (!strcmp(buf, "report")) {
212 return BLOCK_ERR_REPORT;
213 } else {
214 error_report("'%s' invalid %s error action",
215 buf, is_read ? "read" : "write");
216 return -1;
220 DriveInfo *drive_init(QemuOpts *opts, int default_to_scsi)
222 const char *buf;
223 const char *file = NULL;
224 char devname[128];
225 const char *serial;
226 const char *mediastr = "";
227 BlockInterfaceType type;
228 enum { MEDIA_DISK, MEDIA_CDROM } media;
229 int bus_id, unit_id;
230 int cyls, heads, secs, translation;
231 BlockDriver *drv = NULL;
232 int max_devs;
233 int index;
234 int ro = 0;
235 int bdrv_flags = 0;
236 int on_read_error, on_write_error;
237 const char *devaddr;
238 DriveInfo *dinfo;
239 int snapshot = 0;
240 int ret;
242 translation = BIOS_ATA_TRANSLATION_AUTO;
244 if (default_to_scsi) {
245 type = IF_SCSI;
246 pstrcpy(devname, sizeof(devname), "scsi");
247 } else {
248 type = IF_IDE;
249 pstrcpy(devname, sizeof(devname), "ide");
251 media = MEDIA_DISK;
253 /* extract parameters */
254 bus_id = qemu_opt_get_number(opts, "bus", 0);
255 unit_id = qemu_opt_get_number(opts, "unit", -1);
256 index = qemu_opt_get_number(opts, "index", -1);
258 cyls = qemu_opt_get_number(opts, "cyls", 0);
259 heads = qemu_opt_get_number(opts, "heads", 0);
260 secs = qemu_opt_get_number(opts, "secs", 0);
262 snapshot = qemu_opt_get_bool(opts, "snapshot", 0);
263 ro = qemu_opt_get_bool(opts, "readonly", 0);
265 file = qemu_opt_get(opts, "file");
266 serial = qemu_opt_get(opts, "serial");
268 if ((buf = qemu_opt_get(opts, "if")) != NULL) {
269 pstrcpy(devname, sizeof(devname), buf);
270 for (type = 0; type < IF_COUNT && strcmp(buf, if_name[type]); type++)
272 if (type == IF_COUNT) {
273 error_report("unsupported bus type '%s'", buf);
274 return NULL;
277 max_devs = if_max_devs[type];
279 if (cyls || heads || secs) {
280 if (cyls < 1 || (type == IF_IDE && cyls > 16383)) {
281 error_report("invalid physical cyls number");
282 return NULL;
284 if (heads < 1 || (type == IF_IDE && heads > 16)) {
285 error_report("invalid physical heads number");
286 return NULL;
288 if (secs < 1 || (type == IF_IDE && secs > 63)) {
289 error_report("invalid physical secs number");
290 return NULL;
294 if ((buf = qemu_opt_get(opts, "trans")) != NULL) {
295 if (!cyls) {
296 error_report("'%s' trans must be used with cyls,heads and secs",
297 buf);
298 return NULL;
300 if (!strcmp(buf, "none"))
301 translation = BIOS_ATA_TRANSLATION_NONE;
302 else if (!strcmp(buf, "lba"))
303 translation = BIOS_ATA_TRANSLATION_LBA;
304 else if (!strcmp(buf, "auto"))
305 translation = BIOS_ATA_TRANSLATION_AUTO;
306 else {
307 error_report("'%s' invalid translation type", buf);
308 return NULL;
312 if ((buf = qemu_opt_get(opts, "media")) != NULL) {
313 if (!strcmp(buf, "disk")) {
314 media = MEDIA_DISK;
315 } else if (!strcmp(buf, "cdrom")) {
316 if (cyls || secs || heads) {
317 error_report("'%s' invalid physical CHS format", buf);
318 return NULL;
320 media = MEDIA_CDROM;
321 } else {
322 error_report("'%s' invalid media", buf);
323 return NULL;
327 if ((buf = qemu_opt_get(opts, "cache")) != NULL) {
328 if (!strcmp(buf, "off") || !strcmp(buf, "none")) {
329 bdrv_flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
330 } else if (!strcmp(buf, "writeback")) {
331 bdrv_flags |= BDRV_O_CACHE_WB;
332 } else if (!strcmp(buf, "unsafe")) {
333 bdrv_flags |= BDRV_O_CACHE_WB;
334 bdrv_flags |= BDRV_O_NO_FLUSH;
335 } else if (!strcmp(buf, "writethrough")) {
336 /* this is the default */
337 } else {
338 error_report("invalid cache option");
339 return NULL;
343 #ifdef CONFIG_LINUX_AIO
344 if ((buf = qemu_opt_get(opts, "aio")) != NULL) {
345 if (!strcmp(buf, "native")) {
346 bdrv_flags |= BDRV_O_NATIVE_AIO;
347 } else if (!strcmp(buf, "threads")) {
348 /* this is the default */
349 } else {
350 error_report("invalid aio option");
351 return NULL;
354 #endif
356 if ((buf = qemu_opt_get(opts, "format")) != NULL) {
357 if (strcmp(buf, "?") == 0) {
358 error_printf("Supported formats:");
359 bdrv_iterate_format(bdrv_format_print, NULL);
360 error_printf("\n");
361 return NULL;
363 drv = bdrv_find_whitelisted_format(buf);
364 if (!drv) {
365 error_report("'%s' invalid format", buf);
366 return NULL;
370 on_write_error = BLOCK_ERR_STOP_ENOSPC;
371 if ((buf = qemu_opt_get(opts, "werror")) != NULL) {
372 if (type != IF_IDE && type != IF_SCSI && type != IF_VIRTIO && type != IF_NONE) {
373 error_report("werror is not supported by this bus type");
374 return NULL;
377 on_write_error = parse_block_error_action(buf, 0);
378 if (on_write_error < 0) {
379 return NULL;
383 on_read_error = BLOCK_ERR_REPORT;
384 if ((buf = qemu_opt_get(opts, "rerror")) != NULL) {
385 if (type != IF_IDE && type != IF_VIRTIO && type != IF_SCSI && type != IF_NONE) {
386 error_report("rerror is not supported by this bus type");
387 return NULL;
390 on_read_error = parse_block_error_action(buf, 1);
391 if (on_read_error < 0) {
392 return NULL;
396 if ((devaddr = qemu_opt_get(opts, "addr")) != NULL) {
397 if (type != IF_VIRTIO) {
398 error_report("addr is not supported by this bus type");
399 return NULL;
403 /* compute bus and unit according index */
405 if (index != -1) {
406 if (bus_id != 0 || unit_id != -1) {
407 error_report("index cannot be used with bus and unit");
408 return NULL;
410 bus_id = drive_index_to_bus_id(type, index);
411 unit_id = drive_index_to_unit_id(type, index);
414 /* if user doesn't specify a unit_id,
415 * try to find the first free
418 if (unit_id == -1) {
419 unit_id = 0;
420 while (drive_get(type, bus_id, unit_id) != NULL) {
421 unit_id++;
422 if (max_devs && unit_id >= max_devs) {
423 unit_id -= max_devs;
424 bus_id++;
429 /* check unit id */
431 if (max_devs && unit_id >= max_devs) {
432 error_report("unit %d too big (max is %d)",
433 unit_id, max_devs - 1);
434 return NULL;
438 * catch multiple definitions
441 if (drive_get(type, bus_id, unit_id) != NULL) {
442 error_report("drive with bus=%d, unit=%d (index=%d) exists",
443 bus_id, unit_id, index);
444 return NULL;
447 /* init */
449 dinfo = qemu_mallocz(sizeof(*dinfo));
450 if ((buf = qemu_opts_id(opts)) != NULL) {
451 dinfo->id = qemu_strdup(buf);
452 } else {
453 /* no id supplied -> create one */
454 dinfo->id = qemu_mallocz(32);
455 if (type == IF_IDE || type == IF_SCSI)
456 mediastr = (media == MEDIA_CDROM) ? "-cd" : "-hd";
457 if (max_devs)
458 snprintf(dinfo->id, 32, "%s%i%s%i",
459 devname, bus_id, mediastr, unit_id);
460 else
461 snprintf(dinfo->id, 32, "%s%s%i",
462 devname, mediastr, unit_id);
464 dinfo->bdrv = bdrv_new(dinfo->id);
465 dinfo->devaddr = devaddr;
466 dinfo->type = type;
467 dinfo->bus = bus_id;
468 dinfo->unit = unit_id;
469 dinfo->opts = opts;
470 dinfo->refcount = 1;
471 if (serial)
472 strncpy(dinfo->serial, serial, sizeof(dinfo->serial) - 1);
473 QTAILQ_INSERT_TAIL(&drives, dinfo, next);
475 bdrv_set_on_error(dinfo->bdrv, on_read_error, on_write_error);
477 switch(type) {
478 case IF_IDE:
479 case IF_SCSI:
480 case IF_XEN:
481 case IF_NONE:
482 switch(media) {
483 case MEDIA_DISK:
484 if (cyls != 0) {
485 bdrv_set_geometry_hint(dinfo->bdrv, cyls, heads, secs);
486 bdrv_set_translation_hint(dinfo->bdrv, translation);
488 break;
489 case MEDIA_CDROM:
490 bdrv_set_removable(dinfo->bdrv, 1);
491 dinfo->media_cd = 1;
492 break;
494 break;
495 case IF_SD:
496 /* FIXME: This isn't really a floppy, but it's a reasonable
497 approximation. */
498 case IF_FLOPPY:
499 bdrv_set_removable(dinfo->bdrv, 1);
500 break;
501 case IF_PFLASH:
502 case IF_MTD:
503 break;
504 case IF_VIRTIO:
505 /* add virtio block device */
506 opts = qemu_opts_create(qemu_find_opts("device"), NULL, 0);
507 qemu_opt_set(opts, "driver", "virtio-blk");
508 qemu_opt_set(opts, "drive", dinfo->id);
509 if (devaddr)
510 qemu_opt_set(opts, "addr", devaddr);
511 break;
512 default:
513 abort();
515 if (!file || !*file) {
516 return dinfo;
518 if (snapshot) {
519 /* always use cache=unsafe with snapshot */
520 bdrv_flags &= ~BDRV_O_CACHE_MASK;
521 bdrv_flags |= (BDRV_O_SNAPSHOT|BDRV_O_CACHE_WB|BDRV_O_NO_FLUSH);
524 if (media == MEDIA_CDROM) {
525 /* CDROM is fine for any interface, don't check. */
526 ro = 1;
527 } else if (ro == 1) {
528 if (type != IF_SCSI && type != IF_VIRTIO && type != IF_FLOPPY && type != IF_NONE) {
529 error_report("readonly not supported by this bus type");
530 goto err;
534 bdrv_flags |= ro ? 0 : BDRV_O_RDWR;
536 ret = bdrv_open(dinfo->bdrv, file, bdrv_flags, drv);
537 if (ret < 0) {
538 error_report("could not open disk image %s: %s",
539 file, strerror(-ret));
540 goto err;
543 if (bdrv_key_required(dinfo->bdrv))
544 autostart = 0;
545 return dinfo;
547 err:
548 bdrv_delete(dinfo->bdrv);
549 qemu_free(dinfo->id);
550 QTAILQ_REMOVE(&drives, dinfo, next);
551 qemu_free(dinfo);
552 return NULL;
555 void do_commit(Monitor *mon, const QDict *qdict)
557 const char *device = qdict_get_str(qdict, "device");
558 BlockDriverState *bs;
560 if (!strcmp(device, "all")) {
561 bdrv_commit_all();
562 } else {
563 bs = bdrv_find(device);
564 if (!bs) {
565 qerror_report(QERR_DEVICE_NOT_FOUND, device);
566 return;
568 bdrv_commit(bs);
572 int do_snapshot_blkdev(Monitor *mon, const QDict *qdict, QObject **ret_data)
574 const char *device = qdict_get_str(qdict, "device");
575 const char *filename = qdict_get_try_str(qdict, "snapshot_file");
576 const char *format = qdict_get_try_str(qdict, "format");
577 BlockDriverState *bs;
578 BlockDriver *drv, *old_drv, *proto_drv;
579 int ret = 0;
580 int flags;
581 char old_filename[1024];
583 if (!filename) {
584 qerror_report(QERR_MISSING_PARAMETER, "snapshot_file");
585 ret = -1;
586 goto out;
589 bs = bdrv_find(device);
590 if (!bs) {
591 qerror_report(QERR_DEVICE_NOT_FOUND, device);
592 ret = -1;
593 goto out;
596 pstrcpy(old_filename, sizeof(old_filename), bs->filename);
598 old_drv = bs->drv;
599 flags = bs->open_flags;
601 if (!format) {
602 format = "qcow2";
605 drv = bdrv_find_format(format);
606 if (!drv) {
607 qerror_report(QERR_INVALID_BLOCK_FORMAT, format);
608 ret = -1;
609 goto out;
612 proto_drv = bdrv_find_protocol(filename);
613 if (!proto_drv) {
614 qerror_report(QERR_INVALID_BLOCK_FORMAT, format);
615 ret = -1;
616 goto out;
619 ret = bdrv_img_create(filename, format, bs->filename,
620 bs->drv->format_name, NULL, -1, flags);
621 if (ret) {
622 goto out;
625 qemu_aio_flush();
626 bdrv_flush(bs);
628 bdrv_close(bs);
629 ret = bdrv_open(bs, filename, flags, drv);
631 * If reopening the image file we just created fails, fall back
632 * and try to re-open the original image. If that fails too, we
633 * are in serious trouble.
635 if (ret != 0) {
636 ret = bdrv_open(bs, old_filename, flags, old_drv);
637 if (ret != 0) {
638 qerror_report(QERR_OPEN_FILE_FAILED, old_filename);
639 } else {
640 qerror_report(QERR_OPEN_FILE_FAILED, filename);
643 out:
644 if (ret) {
645 ret = -1;
648 return ret;
651 static int eject_device(Monitor *mon, BlockDriverState *bs, int force)
653 if (!force) {
654 if (!bdrv_is_removable(bs)) {
655 qerror_report(QERR_DEVICE_NOT_REMOVABLE,
656 bdrv_get_device_name(bs));
657 return -1;
659 if (bdrv_is_locked(bs)) {
660 qerror_report(QERR_DEVICE_LOCKED, bdrv_get_device_name(bs));
661 return -1;
664 bdrv_close(bs);
665 return 0;
668 int do_eject(Monitor *mon, const QDict *qdict, QObject **ret_data)
670 BlockDriverState *bs;
671 int force = qdict_get_try_bool(qdict, "force", 0);
672 const char *filename = qdict_get_str(qdict, "device");
674 bs = bdrv_find(filename);
675 if (!bs) {
676 qerror_report(QERR_DEVICE_NOT_FOUND, filename);
677 return -1;
679 return eject_device(mon, bs, force);
682 int do_block_set_passwd(Monitor *mon, const QDict *qdict,
683 QObject **ret_data)
685 BlockDriverState *bs;
686 int err;
688 bs = bdrv_find(qdict_get_str(qdict, "device"));
689 if (!bs) {
690 qerror_report(QERR_DEVICE_NOT_FOUND, qdict_get_str(qdict, "device"));
691 return -1;
694 err = bdrv_set_key(bs, qdict_get_str(qdict, "password"));
695 if (err == -EINVAL) {
696 qerror_report(QERR_DEVICE_NOT_ENCRYPTED, bdrv_get_device_name(bs));
697 return -1;
698 } else if (err < 0) {
699 qerror_report(QERR_INVALID_PASSWORD);
700 return -1;
703 return 0;
706 int do_change_block(Monitor *mon, const char *device,
707 const char *filename, const char *fmt)
709 BlockDriverState *bs;
710 BlockDriver *drv = NULL;
711 int bdrv_flags;
713 bs = bdrv_find(device);
714 if (!bs) {
715 qerror_report(QERR_DEVICE_NOT_FOUND, device);
716 return -1;
718 if (fmt) {
719 drv = bdrv_find_whitelisted_format(fmt);
720 if (!drv) {
721 qerror_report(QERR_INVALID_BLOCK_FORMAT, fmt);
722 return -1;
725 if (eject_device(mon, bs, 0) < 0) {
726 return -1;
728 bdrv_flags = bdrv_is_read_only(bs) ? 0 : BDRV_O_RDWR;
729 bdrv_flags |= bdrv_is_snapshot(bs) ? BDRV_O_SNAPSHOT : 0;
730 if (bdrv_open(bs, filename, bdrv_flags, drv) < 0) {
731 qerror_report(QERR_OPEN_FILE_FAILED, filename);
732 return -1;
734 return monitor_read_bdrv_key_start(mon, bs, NULL, NULL);
737 int do_drive_del(Monitor *mon, const QDict *qdict, QObject **ret_data)
739 const char *id = qdict_get_str(qdict, "id");
740 BlockDriverState *bs;
742 bs = bdrv_find(id);
743 if (!bs) {
744 qerror_report(QERR_DEVICE_NOT_FOUND, id);
745 return -1;
747 if (bdrv_in_use(bs)) {
748 qerror_report(QERR_DEVICE_IN_USE, id);
749 return -1;
752 /* quiesce block driver; prevent further io */
753 qemu_aio_flush();
754 bdrv_flush(bs);
755 bdrv_close(bs);
757 /* if we have a device associated with this BlockDriverState (bs->peer)
758 * then we need to make the drive anonymous until the device
759 * can be removed. If this is a drive with no device backing
760 * then we can just get rid of the block driver state right here.
762 if (bs->peer) {
763 bdrv_make_anon(bs);
764 } else {
765 drive_uninit(drive_get_by_blockdev(bs));
768 return 0;
772 * XXX: replace the QERR_UNDEFINED_ERROR errors with real values once the
773 * existing QERR_ macro mess is cleaned up. A good example for better
774 * error reports can be found in the qemu-img resize code.
776 int do_block_resize(Monitor *mon, const QDict *qdict, QObject **ret_data)
778 const char *device = qdict_get_str(qdict, "device");
779 int64_t size = qdict_get_int(qdict, "size");
780 BlockDriverState *bs;
782 bs = bdrv_find(device);
783 if (!bs) {
784 qerror_report(QERR_DEVICE_NOT_FOUND, device);
785 return -1;
788 if (size < 0) {
789 qerror_report(QERR_UNDEFINED_ERROR);
790 return -1;
793 if (bdrv_truncate(bs, size)) {
794 qerror_report(QERR_UNDEFINED_ERROR);
795 return -1;
798 return 0;