xref: /openbmc/qemu/block.c (revision 07a32d6b)
1 /*
2  * QEMU System Emulator block driver
3  *
4  * Copyright (c) 2003 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 #include "config-host.h"
25 #include "qemu-common.h"
26 #include "trace.h"
27 #include "monitor/monitor.h"
28 #include "block/block_int.h"
29 #include "block/blockjob.h"
30 #include "qemu/module.h"
31 #include "qapi/qmp/qjson.h"
32 #include "sysemu/sysemu.h"
33 #include "qemu/notify.h"
34 #include "block/coroutine.h"
35 #include "block/qapi.h"
36 #include "qmp-commands.h"
37 #include "qemu/timer.h"
38 
39 #ifdef CONFIG_BSD
40 #include <sys/types.h>
41 #include <sys/stat.h>
42 #include <sys/ioctl.h>
43 #include <sys/queue.h>
44 #ifndef __DragonFly__
45 #include <sys/disk.h>
46 #endif
47 #endif
48 
49 #ifdef _WIN32
50 #include <windows.h>
51 #endif
52 
53 struct BdrvDirtyBitmap {
54     HBitmap *bitmap;
55     QLIST_ENTRY(BdrvDirtyBitmap) list;
56 };
57 
58 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
59 
60 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load);
61 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
62         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
63         BlockDriverCompletionFunc *cb, void *opaque);
64 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
65         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
66         BlockDriverCompletionFunc *cb, void *opaque);
67 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
68                                          int64_t sector_num, int nb_sectors,
69                                          QEMUIOVector *iov);
70 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
71                                          int64_t sector_num, int nb_sectors,
72                                          QEMUIOVector *iov);
73 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
74     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
75     BdrvRequestFlags flags);
76 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
77     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
78     BdrvRequestFlags flags);
79 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
80                                                int64_t sector_num,
81                                                QEMUIOVector *qiov,
82                                                int nb_sectors,
83                                                BdrvRequestFlags flags,
84                                                BlockDriverCompletionFunc *cb,
85                                                void *opaque,
86                                                bool is_write);
87 static void coroutine_fn bdrv_co_do_rw(void *opaque);
88 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
89     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags);
90 
91 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
92     QTAILQ_HEAD_INITIALIZER(bdrv_states);
93 
94 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
95     QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
96 
97 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
98     QLIST_HEAD_INITIALIZER(bdrv_drivers);
99 
100 /* If non-zero, use only whitelisted block drivers */
101 static int use_bdrv_whitelist;
102 
103 #ifdef _WIN32
104 static int is_windows_drive_prefix(const char *filename)
105 {
106     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
107              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
108             filename[1] == ':');
109 }
110 
111 int is_windows_drive(const char *filename)
112 {
113     if (is_windows_drive_prefix(filename) &&
114         filename[2] == '\0')
115         return 1;
116     if (strstart(filename, "\\\\.\\", NULL) ||
117         strstart(filename, "//./", NULL))
118         return 1;
119     return 0;
120 }
121 #endif
122 
123 /* throttling disk I/O limits */
124 void bdrv_set_io_limits(BlockDriverState *bs,
125                         ThrottleConfig *cfg)
126 {
127     int i;
128 
129     throttle_config(&bs->throttle_state, cfg);
130 
131     for (i = 0; i < 2; i++) {
132         qemu_co_enter_next(&bs->throttled_reqs[i]);
133     }
134 }
135 
136 /* this function drain all the throttled IOs */
137 static bool bdrv_start_throttled_reqs(BlockDriverState *bs)
138 {
139     bool drained = false;
140     bool enabled = bs->io_limits_enabled;
141     int i;
142 
143     bs->io_limits_enabled = false;
144 
145     for (i = 0; i < 2; i++) {
146         while (qemu_co_enter_next(&bs->throttled_reqs[i])) {
147             drained = true;
148         }
149     }
150 
151     bs->io_limits_enabled = enabled;
152 
153     return drained;
154 }
155 
156 void bdrv_io_limits_disable(BlockDriverState *bs)
157 {
158     bs->io_limits_enabled = false;
159 
160     bdrv_start_throttled_reqs(bs);
161 
162     throttle_destroy(&bs->throttle_state);
163 }
164 
165 static void bdrv_throttle_read_timer_cb(void *opaque)
166 {
167     BlockDriverState *bs = opaque;
168     qemu_co_enter_next(&bs->throttled_reqs[0]);
169 }
170 
171 static void bdrv_throttle_write_timer_cb(void *opaque)
172 {
173     BlockDriverState *bs = opaque;
174     qemu_co_enter_next(&bs->throttled_reqs[1]);
175 }
176 
177 /* should be called before bdrv_set_io_limits if a limit is set */
178 void bdrv_io_limits_enable(BlockDriverState *bs)
179 {
180     assert(!bs->io_limits_enabled);
181     throttle_init(&bs->throttle_state,
182                   bdrv_get_aio_context(bs),
183                   QEMU_CLOCK_VIRTUAL,
184                   bdrv_throttle_read_timer_cb,
185                   bdrv_throttle_write_timer_cb,
186                   bs);
187     bs->io_limits_enabled = true;
188 }
189 
190 /* This function makes an IO wait if needed
191  *
192  * @nb_sectors: the number of sectors of the IO
193  * @is_write:   is the IO a write
194  */
195 static void bdrv_io_limits_intercept(BlockDriverState *bs,
196                                      unsigned int bytes,
197                                      bool is_write)
198 {
199     /* does this io must wait */
200     bool must_wait = throttle_schedule_timer(&bs->throttle_state, is_write);
201 
202     /* if must wait or any request of this type throttled queue the IO */
203     if (must_wait ||
204         !qemu_co_queue_empty(&bs->throttled_reqs[is_write])) {
205         qemu_co_queue_wait(&bs->throttled_reqs[is_write]);
206     }
207 
208     /* the IO will be executed, do the accounting */
209     throttle_account(&bs->throttle_state, is_write, bytes);
210 
211 
212     /* if the next request must wait -> do nothing */
213     if (throttle_schedule_timer(&bs->throttle_state, is_write)) {
214         return;
215     }
216 
217     /* else queue next request for execution */
218     qemu_co_queue_next(&bs->throttled_reqs[is_write]);
219 }
220 
221 size_t bdrv_opt_mem_align(BlockDriverState *bs)
222 {
223     if (!bs || !bs->drv) {
224         /* 4k should be on the safe side */
225         return 4096;
226     }
227 
228     return bs->bl.opt_mem_alignment;
229 }
230 
231 /* check if the path starts with "<protocol>:" */
232 static int path_has_protocol(const char *path)
233 {
234     const char *p;
235 
236 #ifdef _WIN32
237     if (is_windows_drive(path) ||
238         is_windows_drive_prefix(path)) {
239         return 0;
240     }
241     p = path + strcspn(path, ":/\\");
242 #else
243     p = path + strcspn(path, ":/");
244 #endif
245 
246     return *p == ':';
247 }
248 
249 int path_is_absolute(const char *path)
250 {
251 #ifdef _WIN32
252     /* specific case for names like: "\\.\d:" */
253     if (is_windows_drive(path) || is_windows_drive_prefix(path)) {
254         return 1;
255     }
256     return (*path == '/' || *path == '\\');
257 #else
258     return (*path == '/');
259 #endif
260 }
261 
262 /* if filename is absolute, just copy it to dest. Otherwise, build a
263    path to it by considering it is relative to base_path. URL are
264    supported. */
265 void path_combine(char *dest, int dest_size,
266                   const char *base_path,
267                   const char *filename)
268 {
269     const char *p, *p1;
270     int len;
271 
272     if (dest_size <= 0)
273         return;
274     if (path_is_absolute(filename)) {
275         pstrcpy(dest, dest_size, filename);
276     } else {
277         p = strchr(base_path, ':');
278         if (p)
279             p++;
280         else
281             p = base_path;
282         p1 = strrchr(base_path, '/');
283 #ifdef _WIN32
284         {
285             const char *p2;
286             p2 = strrchr(base_path, '\\');
287             if (!p1 || p2 > p1)
288                 p1 = p2;
289         }
290 #endif
291         if (p1)
292             p1++;
293         else
294             p1 = base_path;
295         if (p1 > p)
296             p = p1;
297         len = p - base_path;
298         if (len > dest_size - 1)
299             len = dest_size - 1;
300         memcpy(dest, base_path, len);
301         dest[len] = '\0';
302         pstrcat(dest, dest_size, filename);
303     }
304 }
305 
306 void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz)
307 {
308     if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) {
309         pstrcpy(dest, sz, bs->backing_file);
310     } else {
311         path_combine(dest, sz, bs->filename, bs->backing_file);
312     }
313 }
314 
315 void bdrv_register(BlockDriver *bdrv)
316 {
317     /* Block drivers without coroutine functions need emulation */
318     if (!bdrv->bdrv_co_readv) {
319         bdrv->bdrv_co_readv = bdrv_co_readv_em;
320         bdrv->bdrv_co_writev = bdrv_co_writev_em;
321 
322         /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if
323          * the block driver lacks aio we need to emulate that too.
324          */
325         if (!bdrv->bdrv_aio_readv) {
326             /* add AIO emulation layer */
327             bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
328             bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
329         }
330     }
331 
332     QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
333 }
334 
335 /* create a new block device (by default it is empty) */
336 BlockDriverState *bdrv_new(const char *device_name, Error **errp)
337 {
338     BlockDriverState *bs;
339     int i;
340 
341     if (bdrv_find(device_name)) {
342         error_setg(errp, "Device with id '%s' already exists",
343                    device_name);
344         return NULL;
345     }
346     if (bdrv_find_node(device_name)) {
347         error_setg(errp, "Device with node-name '%s' already exists",
348                    device_name);
349         return NULL;
350     }
351 
352     bs = g_malloc0(sizeof(BlockDriverState));
353     QLIST_INIT(&bs->dirty_bitmaps);
354     pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
355     if (device_name[0] != '\0') {
356         QTAILQ_INSERT_TAIL(&bdrv_states, bs, device_list);
357     }
358     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
359         QLIST_INIT(&bs->op_blockers[i]);
360     }
361     bdrv_iostatus_disable(bs);
362     notifier_list_init(&bs->close_notifiers);
363     notifier_with_return_list_init(&bs->before_write_notifiers);
364     qemu_co_queue_init(&bs->throttled_reqs[0]);
365     qemu_co_queue_init(&bs->throttled_reqs[1]);
366     bs->refcnt = 1;
367     bs->aio_context = qemu_get_aio_context();
368 
369     return bs;
370 }
371 
372 void bdrv_add_close_notifier(BlockDriverState *bs, Notifier *notify)
373 {
374     notifier_list_add(&bs->close_notifiers, notify);
375 }
376 
377 BlockDriver *bdrv_find_format(const char *format_name)
378 {
379     BlockDriver *drv1;
380     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
381         if (!strcmp(drv1->format_name, format_name)) {
382             return drv1;
383         }
384     }
385     return NULL;
386 }
387 
388 static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only)
389 {
390     static const char *whitelist_rw[] = {
391         CONFIG_BDRV_RW_WHITELIST
392     };
393     static const char *whitelist_ro[] = {
394         CONFIG_BDRV_RO_WHITELIST
395     };
396     const char **p;
397 
398     if (!whitelist_rw[0] && !whitelist_ro[0]) {
399         return 1;               /* no whitelist, anything goes */
400     }
401 
402     for (p = whitelist_rw; *p; p++) {
403         if (!strcmp(drv->format_name, *p)) {
404             return 1;
405         }
406     }
407     if (read_only) {
408         for (p = whitelist_ro; *p; p++) {
409             if (!strcmp(drv->format_name, *p)) {
410                 return 1;
411             }
412         }
413     }
414     return 0;
415 }
416 
417 BlockDriver *bdrv_find_whitelisted_format(const char *format_name,
418                                           bool read_only)
419 {
420     BlockDriver *drv = bdrv_find_format(format_name);
421     return drv && bdrv_is_whitelisted(drv, read_only) ? drv : NULL;
422 }
423 
424 typedef struct CreateCo {
425     BlockDriver *drv;
426     char *filename;
427     QemuOpts *opts;
428     int ret;
429     Error *err;
430 } CreateCo;
431 
432 static void coroutine_fn bdrv_create_co_entry(void *opaque)
433 {
434     Error *local_err = NULL;
435     int ret;
436 
437     CreateCo *cco = opaque;
438     assert(cco->drv);
439 
440     ret = cco->drv->bdrv_create(cco->filename, cco->opts, &local_err);
441     if (local_err) {
442         error_propagate(&cco->err, local_err);
443     }
444     cco->ret = ret;
445 }
446 
447 int bdrv_create(BlockDriver *drv, const char* filename,
448                 QemuOpts *opts, Error **errp)
449 {
450     int ret;
451 
452     Coroutine *co;
453     CreateCo cco = {
454         .drv = drv,
455         .filename = g_strdup(filename),
456         .opts = opts,
457         .ret = NOT_DONE,
458         .err = NULL,
459     };
460 
461     if (!drv->bdrv_create) {
462         error_setg(errp, "Driver '%s' does not support image creation", drv->format_name);
463         ret = -ENOTSUP;
464         goto out;
465     }
466 
467     if (qemu_in_coroutine()) {
468         /* Fast-path if already in coroutine context */
469         bdrv_create_co_entry(&cco);
470     } else {
471         co = qemu_coroutine_create(bdrv_create_co_entry);
472         qemu_coroutine_enter(co, &cco);
473         while (cco.ret == NOT_DONE) {
474             qemu_aio_wait();
475         }
476     }
477 
478     ret = cco.ret;
479     if (ret < 0) {
480         if (cco.err) {
481             error_propagate(errp, cco.err);
482         } else {
483             error_setg_errno(errp, -ret, "Could not create image");
484         }
485     }
486 
487 out:
488     g_free(cco.filename);
489     return ret;
490 }
491 
492 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp)
493 {
494     BlockDriver *drv;
495     Error *local_err = NULL;
496     int ret;
497 
498     drv = bdrv_find_protocol(filename, true);
499     if (drv == NULL) {
500         error_setg(errp, "Could not find protocol for file '%s'", filename);
501         return -ENOENT;
502     }
503 
504     ret = bdrv_create(drv, filename, opts, &local_err);
505     if (local_err) {
506         error_propagate(errp, local_err);
507     }
508     return ret;
509 }
510 
511 int bdrv_refresh_limits(BlockDriverState *bs)
512 {
513     BlockDriver *drv = bs->drv;
514 
515     memset(&bs->bl, 0, sizeof(bs->bl));
516 
517     if (!drv) {
518         return 0;
519     }
520 
521     /* Take some limits from the children as a default */
522     if (bs->file) {
523         bdrv_refresh_limits(bs->file);
524         bs->bl.opt_transfer_length = bs->file->bl.opt_transfer_length;
525         bs->bl.opt_mem_alignment = bs->file->bl.opt_mem_alignment;
526     } else {
527         bs->bl.opt_mem_alignment = 512;
528     }
529 
530     if (bs->backing_hd) {
531         bdrv_refresh_limits(bs->backing_hd);
532         bs->bl.opt_transfer_length =
533             MAX(bs->bl.opt_transfer_length,
534                 bs->backing_hd->bl.opt_transfer_length);
535         bs->bl.opt_mem_alignment =
536             MAX(bs->bl.opt_mem_alignment,
537                 bs->backing_hd->bl.opt_mem_alignment);
538     }
539 
540     /* Then let the driver override it */
541     if (drv->bdrv_refresh_limits) {
542         return drv->bdrv_refresh_limits(bs);
543     }
544 
545     return 0;
546 }
547 
548 /*
549  * Create a uniquely-named empty temporary file.
550  * Return 0 upon success, otherwise a negative errno value.
551  */
552 int get_tmp_filename(char *filename, int size)
553 {
554 #ifdef _WIN32
555     char temp_dir[MAX_PATH];
556     /* GetTempFileName requires that its output buffer (4th param)
557        have length MAX_PATH or greater.  */
558     assert(size >= MAX_PATH);
559     return (GetTempPath(MAX_PATH, temp_dir)
560             && GetTempFileName(temp_dir, "qem", 0, filename)
561             ? 0 : -GetLastError());
562 #else
563     int fd;
564     const char *tmpdir;
565     tmpdir = getenv("TMPDIR");
566     if (!tmpdir) {
567         tmpdir = "/var/tmp";
568     }
569     if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) {
570         return -EOVERFLOW;
571     }
572     fd = mkstemp(filename);
573     if (fd < 0) {
574         return -errno;
575     }
576     if (close(fd) != 0) {
577         unlink(filename);
578         return -errno;
579     }
580     return 0;
581 #endif
582 }
583 
584 /*
585  * Detect host devices. By convention, /dev/cdrom[N] is always
586  * recognized as a host CDROM.
587  */
588 static BlockDriver *find_hdev_driver(const char *filename)
589 {
590     int score_max = 0, score;
591     BlockDriver *drv = NULL, *d;
592 
593     QLIST_FOREACH(d, &bdrv_drivers, list) {
594         if (d->bdrv_probe_device) {
595             score = d->bdrv_probe_device(filename);
596             if (score > score_max) {
597                 score_max = score;
598                 drv = d;
599             }
600         }
601     }
602 
603     return drv;
604 }
605 
606 BlockDriver *bdrv_find_protocol(const char *filename,
607                                 bool allow_protocol_prefix)
608 {
609     BlockDriver *drv1;
610     char protocol[128];
611     int len;
612     const char *p;
613 
614     /* TODO Drivers without bdrv_file_open must be specified explicitly */
615 
616     /*
617      * XXX(hch): we really should not let host device detection
618      * override an explicit protocol specification, but moving this
619      * later breaks access to device names with colons in them.
620      * Thanks to the brain-dead persistent naming schemes on udev-
621      * based Linux systems those actually are quite common.
622      */
623     drv1 = find_hdev_driver(filename);
624     if (drv1) {
625         return drv1;
626     }
627 
628     if (!path_has_protocol(filename) || !allow_protocol_prefix) {
629         return bdrv_find_format("file");
630     }
631 
632     p = strchr(filename, ':');
633     assert(p != NULL);
634     len = p - filename;
635     if (len > sizeof(protocol) - 1)
636         len = sizeof(protocol) - 1;
637     memcpy(protocol, filename, len);
638     protocol[len] = '\0';
639     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
640         if (drv1->protocol_name &&
641             !strcmp(drv1->protocol_name, protocol)) {
642             return drv1;
643         }
644     }
645     return NULL;
646 }
647 
648 static int find_image_format(BlockDriverState *bs, const char *filename,
649                              BlockDriver **pdrv, Error **errp)
650 {
651     int score, score_max;
652     BlockDriver *drv1, *drv;
653     uint8_t buf[2048];
654     int ret = 0;
655 
656     /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
657     if (bs->sg || !bdrv_is_inserted(bs) || bdrv_getlength(bs) == 0) {
658         drv = bdrv_find_format("raw");
659         if (!drv) {
660             error_setg(errp, "Could not find raw image format");
661             ret = -ENOENT;
662         }
663         *pdrv = drv;
664         return ret;
665     }
666 
667     ret = bdrv_pread(bs, 0, buf, sizeof(buf));
668     if (ret < 0) {
669         error_setg_errno(errp, -ret, "Could not read image for determining its "
670                          "format");
671         *pdrv = NULL;
672         return ret;
673     }
674 
675     score_max = 0;
676     drv = NULL;
677     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
678         if (drv1->bdrv_probe) {
679             score = drv1->bdrv_probe(buf, ret, filename);
680             if (score > score_max) {
681                 score_max = score;
682                 drv = drv1;
683             }
684         }
685     }
686     if (!drv) {
687         error_setg(errp, "Could not determine image format: No compatible "
688                    "driver found");
689         ret = -ENOENT;
690     }
691     *pdrv = drv;
692     return ret;
693 }
694 
695 /**
696  * Set the current 'total_sectors' value
697  */
698 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
699 {
700     BlockDriver *drv = bs->drv;
701 
702     /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
703     if (bs->sg)
704         return 0;
705 
706     /* query actual device if possible, otherwise just trust the hint */
707     if (drv->bdrv_getlength) {
708         int64_t length = drv->bdrv_getlength(bs);
709         if (length < 0) {
710             return length;
711         }
712         hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE);
713     }
714 
715     bs->total_sectors = hint;
716     return 0;
717 }
718 
719 /**
720  * Set open flags for a given discard mode
721  *
722  * Return 0 on success, -1 if the discard mode was invalid.
723  */
724 int bdrv_parse_discard_flags(const char *mode, int *flags)
725 {
726     *flags &= ~BDRV_O_UNMAP;
727 
728     if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) {
729         /* do nothing */
730     } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) {
731         *flags |= BDRV_O_UNMAP;
732     } else {
733         return -1;
734     }
735 
736     return 0;
737 }
738 
739 /**
740  * Set open flags for a given cache mode
741  *
742  * Return 0 on success, -1 if the cache mode was invalid.
743  */
744 int bdrv_parse_cache_flags(const char *mode, int *flags)
745 {
746     *flags &= ~BDRV_O_CACHE_MASK;
747 
748     if (!strcmp(mode, "off") || !strcmp(mode, "none")) {
749         *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB;
750     } else if (!strcmp(mode, "directsync")) {
751         *flags |= BDRV_O_NOCACHE;
752     } else if (!strcmp(mode, "writeback")) {
753         *flags |= BDRV_O_CACHE_WB;
754     } else if (!strcmp(mode, "unsafe")) {
755         *flags |= BDRV_O_CACHE_WB;
756         *flags |= BDRV_O_NO_FLUSH;
757     } else if (!strcmp(mode, "writethrough")) {
758         /* this is the default */
759     } else {
760         return -1;
761     }
762 
763     return 0;
764 }
765 
766 /**
767  * The copy-on-read flag is actually a reference count so multiple users may
768  * use the feature without worrying about clobbering its previous state.
769  * Copy-on-read stays enabled until all users have called to disable it.
770  */
771 void bdrv_enable_copy_on_read(BlockDriverState *bs)
772 {
773     bs->copy_on_read++;
774 }
775 
776 void bdrv_disable_copy_on_read(BlockDriverState *bs)
777 {
778     assert(bs->copy_on_read > 0);
779     bs->copy_on_read--;
780 }
781 
782 /*
783  * Returns the flags that a temporary snapshot should get, based on the
784  * originally requested flags (the originally requested image will have flags
785  * like a backing file)
786  */
787 static int bdrv_temp_snapshot_flags(int flags)
788 {
789     return (flags & ~BDRV_O_SNAPSHOT) | BDRV_O_TEMPORARY;
790 }
791 
792 /*
793  * Returns the flags that bs->file should get, based on the given flags for
794  * the parent BDS
795  */
796 static int bdrv_inherited_flags(int flags)
797 {
798     /* Enable protocol handling, disable format probing for bs->file */
799     flags |= BDRV_O_PROTOCOL;
800 
801     /* Our block drivers take care to send flushes and respect unmap policy,
802      * so we can enable both unconditionally on lower layers. */
803     flags |= BDRV_O_CACHE_WB | BDRV_O_UNMAP;
804 
805     /* Clear flags that only apply to the top layer */
806     flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
807 
808     return flags;
809 }
810 
811 /*
812  * Returns the flags that bs->backing_hd should get, based on the given flags
813  * for the parent BDS
814  */
815 static int bdrv_backing_flags(int flags)
816 {
817     /* backing files always opened read-only */
818     flags &= ~(BDRV_O_RDWR | BDRV_O_COPY_ON_READ);
819 
820     /* snapshot=on is handled on the top layer */
821     flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
822 
823     return flags;
824 }
825 
826 static int bdrv_open_flags(BlockDriverState *bs, int flags)
827 {
828     int open_flags = flags | BDRV_O_CACHE_WB;
829 
830     /*
831      * Clear flags that are internal to the block layer before opening the
832      * image.
833      */
834     open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
835 
836     /*
837      * Snapshots should be writable.
838      */
839     if (flags & BDRV_O_TEMPORARY) {
840         open_flags |= BDRV_O_RDWR;
841     }
842 
843     return open_flags;
844 }
845 
846 static void bdrv_assign_node_name(BlockDriverState *bs,
847                                   const char *node_name,
848                                   Error **errp)
849 {
850     if (!node_name) {
851         return;
852     }
853 
854     /* empty string node name is invalid */
855     if (node_name[0] == '\0') {
856         error_setg(errp, "Empty node name");
857         return;
858     }
859 
860     /* takes care of avoiding namespaces collisions */
861     if (bdrv_find(node_name)) {
862         error_setg(errp, "node-name=%s is conflicting with a device id",
863                    node_name);
864         return;
865     }
866 
867     /* takes care of avoiding duplicates node names */
868     if (bdrv_find_node(node_name)) {
869         error_setg(errp, "Duplicate node name");
870         return;
871     }
872 
873     /* copy node name into the bs and insert it into the graph list */
874     pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
875     QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
876 }
877 
878 /*
879  * Common part for opening disk images and files
880  *
881  * Removes all processed options from *options.
882  */
883 static int bdrv_open_common(BlockDriverState *bs, BlockDriverState *file,
884     QDict *options, int flags, BlockDriver *drv, Error **errp)
885 {
886     int ret, open_flags;
887     const char *filename;
888     const char *node_name = NULL;
889     Error *local_err = NULL;
890 
891     assert(drv != NULL);
892     assert(bs->file == NULL);
893     assert(options != NULL && bs->options != options);
894 
895     if (file != NULL) {
896         filename = file->filename;
897     } else {
898         filename = qdict_get_try_str(options, "filename");
899     }
900 
901     if (drv->bdrv_needs_filename && !filename) {
902         error_setg(errp, "The '%s' block driver requires a file name",
903                    drv->format_name);
904         return -EINVAL;
905     }
906 
907     trace_bdrv_open_common(bs, filename ?: "", flags, drv->format_name);
908 
909     node_name = qdict_get_try_str(options, "node-name");
910     bdrv_assign_node_name(bs, node_name, &local_err);
911     if (local_err) {
912         error_propagate(errp, local_err);
913         return -EINVAL;
914     }
915     qdict_del(options, "node-name");
916 
917     /* bdrv_open() with directly using a protocol as drv. This layer is already
918      * opened, so assign it to bs (while file becomes a closed BlockDriverState)
919      * and return immediately. */
920     if (file != NULL && drv->bdrv_file_open) {
921         bdrv_swap(file, bs);
922         return 0;
923     }
924 
925     bs->open_flags = flags;
926     bs->guest_block_size = 512;
927     bs->request_alignment = 512;
928     bs->zero_beyond_eof = true;
929     open_flags = bdrv_open_flags(bs, flags);
930     bs->read_only = !(open_flags & BDRV_O_RDWR);
931 
932     if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) {
933         error_setg(errp,
934                    !bs->read_only && bdrv_is_whitelisted(drv, true)
935                         ? "Driver '%s' can only be used for read-only devices"
936                         : "Driver '%s' is not whitelisted",
937                    drv->format_name);
938         return -ENOTSUP;
939     }
940 
941     assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */
942     if (flags & BDRV_O_COPY_ON_READ) {
943         if (!bs->read_only) {
944             bdrv_enable_copy_on_read(bs);
945         } else {
946             error_setg(errp, "Can't use copy-on-read on read-only device");
947             return -EINVAL;
948         }
949     }
950 
951     if (filename != NULL) {
952         pstrcpy(bs->filename, sizeof(bs->filename), filename);
953     } else {
954         bs->filename[0] = '\0';
955     }
956 
957     bs->drv = drv;
958     bs->opaque = g_malloc0(drv->instance_size);
959 
960     bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
961 
962     /* Open the image, either directly or using a protocol */
963     if (drv->bdrv_file_open) {
964         assert(file == NULL);
965         assert(!drv->bdrv_needs_filename || filename != NULL);
966         ret = drv->bdrv_file_open(bs, options, open_flags, &local_err);
967     } else {
968         if (file == NULL) {
969             error_setg(errp, "Can't use '%s' as a block driver for the "
970                        "protocol level", drv->format_name);
971             ret = -EINVAL;
972             goto free_and_fail;
973         }
974         bs->file = file;
975         ret = drv->bdrv_open(bs, options, open_flags, &local_err);
976     }
977 
978     if (ret < 0) {
979         if (local_err) {
980             error_propagate(errp, local_err);
981         } else if (bs->filename[0]) {
982             error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename);
983         } else {
984             error_setg_errno(errp, -ret, "Could not open image");
985         }
986         goto free_and_fail;
987     }
988 
989     ret = refresh_total_sectors(bs, bs->total_sectors);
990     if (ret < 0) {
991         error_setg_errno(errp, -ret, "Could not refresh total sector count");
992         goto free_and_fail;
993     }
994 
995     bdrv_refresh_limits(bs);
996     assert(bdrv_opt_mem_align(bs) != 0);
997     assert((bs->request_alignment != 0) || bs->sg);
998     return 0;
999 
1000 free_and_fail:
1001     bs->file = NULL;
1002     g_free(bs->opaque);
1003     bs->opaque = NULL;
1004     bs->drv = NULL;
1005     return ret;
1006 }
1007 
1008 /*
1009  * Opens a file using a protocol (file, host_device, nbd, ...)
1010  *
1011  * options is an indirect pointer to a QDict of options to pass to the block
1012  * drivers, or pointer to NULL for an empty set of options. If this function
1013  * takes ownership of the QDict reference, it will set *options to NULL;
1014  * otherwise, it will contain unused/unrecognized options after this function
1015  * returns. Then, the caller is responsible for freeing it. If it intends to
1016  * reuse the QDict, QINCREF() should be called beforehand.
1017  */
1018 static int bdrv_file_open(BlockDriverState *bs, const char *filename,
1019                           QDict **options, int flags, Error **errp)
1020 {
1021     BlockDriver *drv;
1022     const char *drvname;
1023     bool parse_filename = false;
1024     Error *local_err = NULL;
1025     int ret;
1026 
1027     /* Fetch the file name from the options QDict if necessary */
1028     if (!filename) {
1029         filename = qdict_get_try_str(*options, "filename");
1030     } else if (filename && !qdict_haskey(*options, "filename")) {
1031         qdict_put(*options, "filename", qstring_from_str(filename));
1032         parse_filename = true;
1033     } else {
1034         error_setg(errp, "Can't specify 'file' and 'filename' options at the "
1035                    "same time");
1036         ret = -EINVAL;
1037         goto fail;
1038     }
1039 
1040     /* Find the right block driver */
1041     drvname = qdict_get_try_str(*options, "driver");
1042     if (drvname) {
1043         drv = bdrv_find_format(drvname);
1044         if (!drv) {
1045             error_setg(errp, "Unknown driver '%s'", drvname);
1046         }
1047         qdict_del(*options, "driver");
1048     } else if (filename) {
1049         drv = bdrv_find_protocol(filename, parse_filename);
1050         if (!drv) {
1051             error_setg(errp, "Unknown protocol");
1052         }
1053     } else {
1054         error_setg(errp, "Must specify either driver or file");
1055         drv = NULL;
1056     }
1057 
1058     if (!drv) {
1059         /* errp has been set already */
1060         ret = -ENOENT;
1061         goto fail;
1062     }
1063 
1064     /* Parse the filename and open it */
1065     if (drv->bdrv_parse_filename && parse_filename) {
1066         drv->bdrv_parse_filename(filename, *options, &local_err);
1067         if (local_err) {
1068             error_propagate(errp, local_err);
1069             ret = -EINVAL;
1070             goto fail;
1071         }
1072 
1073         if (!drv->bdrv_needs_filename) {
1074             qdict_del(*options, "filename");
1075         } else {
1076             filename = qdict_get_str(*options, "filename");
1077         }
1078     }
1079 
1080     if (!drv->bdrv_file_open) {
1081         ret = bdrv_open(&bs, filename, NULL, *options, flags, drv, &local_err);
1082         *options = NULL;
1083     } else {
1084         ret = bdrv_open_common(bs, NULL, *options, flags, drv, &local_err);
1085     }
1086     if (ret < 0) {
1087         error_propagate(errp, local_err);
1088         goto fail;
1089     }
1090 
1091     bs->growable = 1;
1092     return 0;
1093 
1094 fail:
1095     return ret;
1096 }
1097 
1098 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd)
1099 {
1100 
1101     if (bs->backing_hd) {
1102         assert(bs->backing_blocker);
1103         bdrv_op_unblock_all(bs->backing_hd, bs->backing_blocker);
1104     } else if (backing_hd) {
1105         error_setg(&bs->backing_blocker,
1106                    "device is used as backing hd of '%s'",
1107                    bs->device_name);
1108     }
1109 
1110     bs->backing_hd = backing_hd;
1111     if (!backing_hd) {
1112         error_free(bs->backing_blocker);
1113         bs->backing_blocker = NULL;
1114         goto out;
1115     }
1116     bs->open_flags &= ~BDRV_O_NO_BACKING;
1117     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_hd->filename);
1118     pstrcpy(bs->backing_format, sizeof(bs->backing_format),
1119             backing_hd->drv ? backing_hd->drv->format_name : "");
1120 
1121     bdrv_op_block_all(bs->backing_hd, bs->backing_blocker);
1122     /* Otherwise we won't be able to commit due to check in bdrv_commit */
1123     bdrv_op_unblock(bs->backing_hd, BLOCK_OP_TYPE_COMMIT,
1124                     bs->backing_blocker);
1125 out:
1126     bdrv_refresh_limits(bs);
1127 }
1128 
1129 /*
1130  * Opens the backing file for a BlockDriverState if not yet open
1131  *
1132  * options is a QDict of options to pass to the block drivers, or NULL for an
1133  * empty set of options. The reference to the QDict is transferred to this
1134  * function (even on failure), so if the caller intends to reuse the dictionary,
1135  * it needs to use QINCREF() before calling bdrv_file_open.
1136  */
1137 int bdrv_open_backing_file(BlockDriverState *bs, QDict *options, Error **errp)
1138 {
1139     char *backing_filename = g_malloc0(PATH_MAX);
1140     int ret = 0;
1141     BlockDriver *back_drv = NULL;
1142     BlockDriverState *backing_hd;
1143     Error *local_err = NULL;
1144 
1145     if (bs->backing_hd != NULL) {
1146         QDECREF(options);
1147         goto free_exit;
1148     }
1149 
1150     /* NULL means an empty set of options */
1151     if (options == NULL) {
1152         options = qdict_new();
1153     }
1154 
1155     bs->open_flags &= ~BDRV_O_NO_BACKING;
1156     if (qdict_haskey(options, "file.filename")) {
1157         backing_filename[0] = '\0';
1158     } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
1159         QDECREF(options);
1160         goto free_exit;
1161     } else {
1162         bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX);
1163     }
1164 
1165     backing_hd = bdrv_new("", errp);
1166 
1167     if (bs->backing_format[0] != '\0') {
1168         back_drv = bdrv_find_format(bs->backing_format);
1169     }
1170 
1171     assert(bs->backing_hd == NULL);
1172     ret = bdrv_open(&backing_hd,
1173                     *backing_filename ? backing_filename : NULL, NULL, options,
1174                     bdrv_backing_flags(bs->open_flags), back_drv, &local_err);
1175     if (ret < 0) {
1176         bdrv_unref(backing_hd);
1177         backing_hd = NULL;
1178         bs->open_flags |= BDRV_O_NO_BACKING;
1179         error_setg(errp, "Could not open backing file: %s",
1180                    error_get_pretty(local_err));
1181         error_free(local_err);
1182         goto free_exit;
1183     }
1184     bdrv_set_backing_hd(bs, backing_hd);
1185 
1186 free_exit:
1187     g_free(backing_filename);
1188     return ret;
1189 }
1190 
1191 /*
1192  * Opens a disk image whose options are given as BlockdevRef in another block
1193  * device's options.
1194  *
1195  * If allow_none is true, no image will be opened if filename is false and no
1196  * BlockdevRef is given. *pbs will remain unchanged and 0 will be returned.
1197  *
1198  * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
1199  * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1200  * itself, all options starting with "${bdref_key}." are considered part of the
1201  * BlockdevRef.
1202  *
1203  * The BlockdevRef will be removed from the options QDict.
1204  *
1205  * To conform with the behavior of bdrv_open(), *pbs has to be NULL.
1206  */
1207 int bdrv_open_image(BlockDriverState **pbs, const char *filename,
1208                     QDict *options, const char *bdref_key, int flags,
1209                     bool allow_none, Error **errp)
1210 {
1211     QDict *image_options;
1212     int ret;
1213     char *bdref_key_dot;
1214     const char *reference;
1215 
1216     assert(pbs);
1217     assert(*pbs == NULL);
1218 
1219     bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1220     qdict_extract_subqdict(options, &image_options, bdref_key_dot);
1221     g_free(bdref_key_dot);
1222 
1223     reference = qdict_get_try_str(options, bdref_key);
1224     if (!filename && !reference && !qdict_size(image_options)) {
1225         if (allow_none) {
1226             ret = 0;
1227         } else {
1228             error_setg(errp, "A block device must be specified for \"%s\"",
1229                        bdref_key);
1230             ret = -EINVAL;
1231         }
1232         QDECREF(image_options);
1233         goto done;
1234     }
1235 
1236     ret = bdrv_open(pbs, filename, reference, image_options, flags, NULL, errp);
1237 
1238 done:
1239     qdict_del(options, bdref_key);
1240     return ret;
1241 }
1242 
1243 void bdrv_append_temp_snapshot(BlockDriverState *bs, int flags, Error **errp)
1244 {
1245     /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1246     char *tmp_filename = g_malloc0(PATH_MAX + 1);
1247     int64_t total_size;
1248     BlockDriver *bdrv_qcow2;
1249     QemuOpts *opts = NULL;
1250     QDict *snapshot_options;
1251     BlockDriverState *bs_snapshot;
1252     Error *local_err;
1253     int ret;
1254 
1255     /* if snapshot, we create a temporary backing file and open it
1256        instead of opening 'filename' directly */
1257 
1258     /* Get the required size from the image */
1259     total_size = bdrv_getlength(bs);
1260     if (total_size < 0) {
1261         error_setg_errno(errp, -total_size, "Could not get image size");
1262         goto out;
1263     }
1264     total_size &= BDRV_SECTOR_MASK;
1265 
1266     /* Create the temporary image */
1267     ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
1268     if (ret < 0) {
1269         error_setg_errno(errp, -ret, "Could not get temporary filename");
1270         goto out;
1271     }
1272 
1273     bdrv_qcow2 = bdrv_find_format("qcow2");
1274     opts = qemu_opts_create(bdrv_qcow2->create_opts, NULL, 0,
1275                             &error_abort);
1276     qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size);
1277     ret = bdrv_create(bdrv_qcow2, tmp_filename, opts, &local_err);
1278     qemu_opts_del(opts);
1279     if (ret < 0) {
1280         error_setg_errno(errp, -ret, "Could not create temporary overlay "
1281                          "'%s': %s", tmp_filename,
1282                          error_get_pretty(local_err));
1283         error_free(local_err);
1284         goto out;
1285     }
1286 
1287     /* Prepare a new options QDict for the temporary file */
1288     snapshot_options = qdict_new();
1289     qdict_put(snapshot_options, "file.driver",
1290               qstring_from_str("file"));
1291     qdict_put(snapshot_options, "file.filename",
1292               qstring_from_str(tmp_filename));
1293 
1294     bs_snapshot = bdrv_new("", &error_abort);
1295 
1296     ret = bdrv_open(&bs_snapshot, NULL, NULL, snapshot_options,
1297                     flags, bdrv_qcow2, &local_err);
1298     if (ret < 0) {
1299         error_propagate(errp, local_err);
1300         goto out;
1301     }
1302 
1303     bdrv_append(bs_snapshot, bs);
1304 
1305 out:
1306     g_free(tmp_filename);
1307 }
1308 
1309 static QDict *parse_json_filename(const char *filename, Error **errp)
1310 {
1311     QObject *options_obj;
1312     QDict *options;
1313     int ret;
1314 
1315     ret = strstart(filename, "json:", &filename);
1316     assert(ret);
1317 
1318     options_obj = qobject_from_json(filename);
1319     if (!options_obj) {
1320         error_setg(errp, "Could not parse the JSON options");
1321         return NULL;
1322     }
1323 
1324     if (qobject_type(options_obj) != QTYPE_QDICT) {
1325         qobject_decref(options_obj);
1326         error_setg(errp, "Invalid JSON object given");
1327         return NULL;
1328     }
1329 
1330     options = qobject_to_qdict(options_obj);
1331     qdict_flatten(options);
1332 
1333     return options;
1334 }
1335 
1336 /*
1337  * Opens a disk image (raw, qcow2, vmdk, ...)
1338  *
1339  * options is a QDict of options to pass to the block drivers, or NULL for an
1340  * empty set of options. The reference to the QDict belongs to the block layer
1341  * after the call (even on failure), so if the caller intends to reuse the
1342  * dictionary, it needs to use QINCREF() before calling bdrv_open.
1343  *
1344  * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
1345  * If it is not NULL, the referenced BDS will be reused.
1346  *
1347  * The reference parameter may be used to specify an existing block device which
1348  * should be opened. If specified, neither options nor a filename may be given,
1349  * nor can an existing BDS be reused (that is, *pbs has to be NULL).
1350  */
1351 int bdrv_open(BlockDriverState **pbs, const char *filename,
1352               const char *reference, QDict *options, int flags,
1353               BlockDriver *drv, Error **errp)
1354 {
1355     int ret;
1356     BlockDriverState *file = NULL, *bs;
1357     const char *drvname;
1358     Error *local_err = NULL;
1359     int snapshot_flags = 0;
1360 
1361     assert(pbs);
1362 
1363     if (reference) {
1364         bool options_non_empty = options ? qdict_size(options) : false;
1365         QDECREF(options);
1366 
1367         if (*pbs) {
1368             error_setg(errp, "Cannot reuse an existing BDS when referencing "
1369                        "another block device");
1370             return -EINVAL;
1371         }
1372 
1373         if (filename || options_non_empty) {
1374             error_setg(errp, "Cannot reference an existing block device with "
1375                        "additional options or a new filename");
1376             return -EINVAL;
1377         }
1378 
1379         bs = bdrv_lookup_bs(reference, reference, errp);
1380         if (!bs) {
1381             return -ENODEV;
1382         }
1383         bdrv_ref(bs);
1384         *pbs = bs;
1385         return 0;
1386     }
1387 
1388     if (*pbs) {
1389         bs = *pbs;
1390     } else {
1391         bs = bdrv_new("", &error_abort);
1392     }
1393 
1394     /* NULL means an empty set of options */
1395     if (options == NULL) {
1396         options = qdict_new();
1397     }
1398 
1399     if (filename && g_str_has_prefix(filename, "json:")) {
1400         QDict *json_options = parse_json_filename(filename, &local_err);
1401         if (local_err) {
1402             ret = -EINVAL;
1403             goto fail;
1404         }
1405 
1406         /* Options given in the filename have lower priority than options
1407          * specified directly */
1408         qdict_join(options, json_options, false);
1409         QDECREF(json_options);
1410         filename = NULL;
1411     }
1412 
1413     bs->options = options;
1414     options = qdict_clone_shallow(options);
1415 
1416     if (flags & BDRV_O_PROTOCOL) {
1417         assert(!drv);
1418         ret = bdrv_file_open(bs, filename, &options, flags & ~BDRV_O_PROTOCOL,
1419                              &local_err);
1420         if (!ret) {
1421             drv = bs->drv;
1422             goto done;
1423         } else if (bs->drv) {
1424             goto close_and_fail;
1425         } else {
1426             goto fail;
1427         }
1428     }
1429 
1430     /* Open image file without format layer */
1431     if (flags & BDRV_O_RDWR) {
1432         flags |= BDRV_O_ALLOW_RDWR;
1433     }
1434     if (flags & BDRV_O_SNAPSHOT) {
1435         snapshot_flags = bdrv_temp_snapshot_flags(flags);
1436         flags = bdrv_backing_flags(flags);
1437     }
1438 
1439     assert(file == NULL);
1440     ret = bdrv_open_image(&file, filename, options, "file",
1441                           bdrv_inherited_flags(flags),
1442                           true, &local_err);
1443     if (ret < 0) {
1444         goto fail;
1445     }
1446 
1447     /* Find the right image format driver */
1448     drvname = qdict_get_try_str(options, "driver");
1449     if (drvname) {
1450         drv = bdrv_find_format(drvname);
1451         qdict_del(options, "driver");
1452         if (!drv) {
1453             error_setg(errp, "Invalid driver: '%s'", drvname);
1454             ret = -EINVAL;
1455             goto fail;
1456         }
1457     }
1458 
1459     if (!drv) {
1460         if (file) {
1461             ret = find_image_format(file, filename, &drv, &local_err);
1462         } else {
1463             error_setg(errp, "Must specify either driver or file");
1464             ret = -EINVAL;
1465             goto fail;
1466         }
1467     }
1468 
1469     if (!drv) {
1470         goto fail;
1471     }
1472 
1473     /* Open the image */
1474     ret = bdrv_open_common(bs, file, options, flags, drv, &local_err);
1475     if (ret < 0) {
1476         goto fail;
1477     }
1478 
1479     if (file && (bs->file != file)) {
1480         bdrv_unref(file);
1481         file = NULL;
1482     }
1483 
1484     /* If there is a backing file, use it */
1485     if ((flags & BDRV_O_NO_BACKING) == 0) {
1486         QDict *backing_options;
1487 
1488         qdict_extract_subqdict(options, &backing_options, "backing.");
1489         ret = bdrv_open_backing_file(bs, backing_options, &local_err);
1490         if (ret < 0) {
1491             goto close_and_fail;
1492         }
1493     }
1494 
1495     /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
1496      * temporary snapshot afterwards. */
1497     if (snapshot_flags) {
1498         bdrv_append_temp_snapshot(bs, snapshot_flags, &local_err);
1499         if (local_err) {
1500             error_propagate(errp, local_err);
1501             goto close_and_fail;
1502         }
1503     }
1504 
1505 
1506 done:
1507     /* Check if any unknown options were used */
1508     if (options && (qdict_size(options) != 0)) {
1509         const QDictEntry *entry = qdict_first(options);
1510         if (flags & BDRV_O_PROTOCOL) {
1511             error_setg(errp, "Block protocol '%s' doesn't support the option "
1512                        "'%s'", drv->format_name, entry->key);
1513         } else {
1514             error_setg(errp, "Block format '%s' used by device '%s' doesn't "
1515                        "support the option '%s'", drv->format_name,
1516                        bs->device_name, entry->key);
1517         }
1518 
1519         ret = -EINVAL;
1520         goto close_and_fail;
1521     }
1522 
1523     if (!bdrv_key_required(bs)) {
1524         bdrv_dev_change_media_cb(bs, true);
1525     } else if (!runstate_check(RUN_STATE_PRELAUNCH)
1526                && !runstate_check(RUN_STATE_INMIGRATE)
1527                && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
1528         error_setg(errp,
1529                    "Guest must be stopped for opening of encrypted image");
1530         ret = -EBUSY;
1531         goto close_and_fail;
1532     }
1533 
1534     QDECREF(options);
1535     *pbs = bs;
1536     return 0;
1537 
1538 fail:
1539     if (file != NULL) {
1540         bdrv_unref(file);
1541     }
1542     QDECREF(bs->options);
1543     QDECREF(options);
1544     bs->options = NULL;
1545     if (!*pbs) {
1546         /* If *pbs is NULL, a new BDS has been created in this function and
1547            needs to be freed now. Otherwise, it does not need to be closed,
1548            since it has not really been opened yet. */
1549         bdrv_unref(bs);
1550     }
1551     if (local_err) {
1552         error_propagate(errp, local_err);
1553     }
1554     return ret;
1555 
1556 close_and_fail:
1557     /* See fail path, but now the BDS has to be always closed */
1558     if (*pbs) {
1559         bdrv_close(bs);
1560     } else {
1561         bdrv_unref(bs);
1562     }
1563     QDECREF(options);
1564     if (local_err) {
1565         error_propagate(errp, local_err);
1566     }
1567     return ret;
1568 }
1569 
1570 typedef struct BlockReopenQueueEntry {
1571      bool prepared;
1572      BDRVReopenState state;
1573      QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1574 } BlockReopenQueueEntry;
1575 
1576 /*
1577  * Adds a BlockDriverState to a simple queue for an atomic, transactional
1578  * reopen of multiple devices.
1579  *
1580  * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1581  * already performed, or alternatively may be NULL a new BlockReopenQueue will
1582  * be created and initialized. This newly created BlockReopenQueue should be
1583  * passed back in for subsequent calls that are intended to be of the same
1584  * atomic 'set'.
1585  *
1586  * bs is the BlockDriverState to add to the reopen queue.
1587  *
1588  * flags contains the open flags for the associated bs
1589  *
1590  * returns a pointer to bs_queue, which is either the newly allocated
1591  * bs_queue, or the existing bs_queue being used.
1592  *
1593  */
1594 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
1595                                     BlockDriverState *bs, int flags)
1596 {
1597     assert(bs != NULL);
1598 
1599     BlockReopenQueueEntry *bs_entry;
1600     if (bs_queue == NULL) {
1601         bs_queue = g_new0(BlockReopenQueue, 1);
1602         QSIMPLEQ_INIT(bs_queue);
1603     }
1604 
1605     /* bdrv_open() masks this flag out */
1606     flags &= ~BDRV_O_PROTOCOL;
1607 
1608     if (bs->file) {
1609         bdrv_reopen_queue(bs_queue, bs->file, bdrv_inherited_flags(flags));
1610     }
1611 
1612     bs_entry = g_new0(BlockReopenQueueEntry, 1);
1613     QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1614 
1615     bs_entry->state.bs = bs;
1616     bs_entry->state.flags = flags;
1617 
1618     return bs_queue;
1619 }
1620 
1621 /*
1622  * Reopen multiple BlockDriverStates atomically & transactionally.
1623  *
1624  * The queue passed in (bs_queue) must have been built up previous
1625  * via bdrv_reopen_queue().
1626  *
1627  * Reopens all BDS specified in the queue, with the appropriate
1628  * flags.  All devices are prepared for reopen, and failure of any
1629  * device will cause all device changes to be abandonded, and intermediate
1630  * data cleaned up.
1631  *
1632  * If all devices prepare successfully, then the changes are committed
1633  * to all devices.
1634  *
1635  */
1636 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
1637 {
1638     int ret = -1;
1639     BlockReopenQueueEntry *bs_entry, *next;
1640     Error *local_err = NULL;
1641 
1642     assert(bs_queue != NULL);
1643 
1644     bdrv_drain_all();
1645 
1646     QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1647         if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
1648             error_propagate(errp, local_err);
1649             goto cleanup;
1650         }
1651         bs_entry->prepared = true;
1652     }
1653 
1654     /* If we reach this point, we have success and just need to apply the
1655      * changes
1656      */
1657     QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1658         bdrv_reopen_commit(&bs_entry->state);
1659     }
1660 
1661     ret = 0;
1662 
1663 cleanup:
1664     QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
1665         if (ret && bs_entry->prepared) {
1666             bdrv_reopen_abort(&bs_entry->state);
1667         }
1668         g_free(bs_entry);
1669     }
1670     g_free(bs_queue);
1671     return ret;
1672 }
1673 
1674 
1675 /* Reopen a single BlockDriverState with the specified flags. */
1676 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
1677 {
1678     int ret = -1;
1679     Error *local_err = NULL;
1680     BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, bdrv_flags);
1681 
1682     ret = bdrv_reopen_multiple(queue, &local_err);
1683     if (local_err != NULL) {
1684         error_propagate(errp, local_err);
1685     }
1686     return ret;
1687 }
1688 
1689 
1690 /*
1691  * Prepares a BlockDriverState for reopen. All changes are staged in the
1692  * 'opaque' field of the BDRVReopenState, which is used and allocated by
1693  * the block driver layer .bdrv_reopen_prepare()
1694  *
1695  * bs is the BlockDriverState to reopen
1696  * flags are the new open flags
1697  * queue is the reopen queue
1698  *
1699  * Returns 0 on success, non-zero on error.  On error errp will be set
1700  * as well.
1701  *
1702  * On failure, bdrv_reopen_abort() will be called to clean up any data.
1703  * It is the responsibility of the caller to then call the abort() or
1704  * commit() for any other BDS that have been left in a prepare() state
1705  *
1706  */
1707 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
1708                         Error **errp)
1709 {
1710     int ret = -1;
1711     Error *local_err = NULL;
1712     BlockDriver *drv;
1713 
1714     assert(reopen_state != NULL);
1715     assert(reopen_state->bs->drv != NULL);
1716     drv = reopen_state->bs->drv;
1717 
1718     /* if we are to stay read-only, do not allow permission change
1719      * to r/w */
1720     if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
1721         reopen_state->flags & BDRV_O_RDWR) {
1722         error_set(errp, QERR_DEVICE_IS_READ_ONLY,
1723                   reopen_state->bs->device_name);
1724         goto error;
1725     }
1726 
1727 
1728     ret = bdrv_flush(reopen_state->bs);
1729     if (ret) {
1730         error_set(errp, ERROR_CLASS_GENERIC_ERROR, "Error (%s) flushing drive",
1731                   strerror(-ret));
1732         goto error;
1733     }
1734 
1735     if (drv->bdrv_reopen_prepare) {
1736         ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
1737         if (ret) {
1738             if (local_err != NULL) {
1739                 error_propagate(errp, local_err);
1740             } else {
1741                 error_setg(errp, "failed while preparing to reopen image '%s'",
1742                            reopen_state->bs->filename);
1743             }
1744             goto error;
1745         }
1746     } else {
1747         /* It is currently mandatory to have a bdrv_reopen_prepare()
1748          * handler for each supported drv. */
1749         error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1750                   drv->format_name, reopen_state->bs->device_name,
1751                  "reopening of file");
1752         ret = -1;
1753         goto error;
1754     }
1755 
1756     ret = 0;
1757 
1758 error:
1759     return ret;
1760 }
1761 
1762 /*
1763  * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1764  * makes them final by swapping the staging BlockDriverState contents into
1765  * the active BlockDriverState contents.
1766  */
1767 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
1768 {
1769     BlockDriver *drv;
1770 
1771     assert(reopen_state != NULL);
1772     drv = reopen_state->bs->drv;
1773     assert(drv != NULL);
1774 
1775     /* If there are any driver level actions to take */
1776     if (drv->bdrv_reopen_commit) {
1777         drv->bdrv_reopen_commit(reopen_state);
1778     }
1779 
1780     /* set BDS specific flags now */
1781     reopen_state->bs->open_flags         = reopen_state->flags;
1782     reopen_state->bs->enable_write_cache = !!(reopen_state->flags &
1783                                               BDRV_O_CACHE_WB);
1784     reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
1785 
1786     bdrv_refresh_limits(reopen_state->bs);
1787 }
1788 
1789 /*
1790  * Abort the reopen, and delete and free the staged changes in
1791  * reopen_state
1792  */
1793 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
1794 {
1795     BlockDriver *drv;
1796 
1797     assert(reopen_state != NULL);
1798     drv = reopen_state->bs->drv;
1799     assert(drv != NULL);
1800 
1801     if (drv->bdrv_reopen_abort) {
1802         drv->bdrv_reopen_abort(reopen_state);
1803     }
1804 }
1805 
1806 
1807 void bdrv_close(BlockDriverState *bs)
1808 {
1809     if (bs->job) {
1810         block_job_cancel_sync(bs->job);
1811     }
1812     bdrv_drain_all(); /* complete I/O */
1813     bdrv_flush(bs);
1814     bdrv_drain_all(); /* in case flush left pending I/O */
1815     notifier_list_notify(&bs->close_notifiers, bs);
1816 
1817     if (bs->drv) {
1818         if (bs->backing_hd) {
1819             BlockDriverState *backing_hd = bs->backing_hd;
1820             bdrv_set_backing_hd(bs, NULL);
1821             bdrv_unref(backing_hd);
1822         }
1823         bs->drv->bdrv_close(bs);
1824         g_free(bs->opaque);
1825         bs->opaque = NULL;
1826         bs->drv = NULL;
1827         bs->copy_on_read = 0;
1828         bs->backing_file[0] = '\0';
1829         bs->backing_format[0] = '\0';
1830         bs->total_sectors = 0;
1831         bs->encrypted = 0;
1832         bs->valid_key = 0;
1833         bs->sg = 0;
1834         bs->growable = 0;
1835         bs->zero_beyond_eof = false;
1836         QDECREF(bs->options);
1837         bs->options = NULL;
1838 
1839         if (bs->file != NULL) {
1840             bdrv_unref(bs->file);
1841             bs->file = NULL;
1842         }
1843     }
1844 
1845     bdrv_dev_change_media_cb(bs, false);
1846 
1847     /*throttling disk I/O limits*/
1848     if (bs->io_limits_enabled) {
1849         bdrv_io_limits_disable(bs);
1850     }
1851 }
1852 
1853 void bdrv_close_all(void)
1854 {
1855     BlockDriverState *bs;
1856 
1857     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1858         AioContext *aio_context = bdrv_get_aio_context(bs);
1859 
1860         aio_context_acquire(aio_context);
1861         bdrv_close(bs);
1862         aio_context_release(aio_context);
1863     }
1864 }
1865 
1866 /* Check if any requests are in-flight (including throttled requests) */
1867 static bool bdrv_requests_pending(BlockDriverState *bs)
1868 {
1869     if (!QLIST_EMPTY(&bs->tracked_requests)) {
1870         return true;
1871     }
1872     if (!qemu_co_queue_empty(&bs->throttled_reqs[0])) {
1873         return true;
1874     }
1875     if (!qemu_co_queue_empty(&bs->throttled_reqs[1])) {
1876         return true;
1877     }
1878     if (bs->file && bdrv_requests_pending(bs->file)) {
1879         return true;
1880     }
1881     if (bs->backing_hd && bdrv_requests_pending(bs->backing_hd)) {
1882         return true;
1883     }
1884     return false;
1885 }
1886 
1887 /*
1888  * Wait for pending requests to complete across all BlockDriverStates
1889  *
1890  * This function does not flush data to disk, use bdrv_flush_all() for that
1891  * after calling this function.
1892  *
1893  * Note that completion of an asynchronous I/O operation can trigger any
1894  * number of other I/O operations on other devices---for example a coroutine
1895  * can be arbitrarily complex and a constant flow of I/O can come until the
1896  * coroutine is complete.  Because of this, it is not possible to have a
1897  * function to drain a single device's I/O queue.
1898  */
1899 void bdrv_drain_all(void)
1900 {
1901     /* Always run first iteration so any pending completion BHs run */
1902     bool busy = true;
1903     BlockDriverState *bs;
1904 
1905     while (busy) {
1906         busy = false;
1907 
1908         QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1909             AioContext *aio_context = bdrv_get_aio_context(bs);
1910             bool bs_busy;
1911 
1912             aio_context_acquire(aio_context);
1913             bdrv_start_throttled_reqs(bs);
1914             bs_busy = bdrv_requests_pending(bs);
1915             bs_busy |= aio_poll(aio_context, bs_busy);
1916             aio_context_release(aio_context);
1917 
1918             busy |= bs_busy;
1919         }
1920     }
1921 }
1922 
1923 /* make a BlockDriverState anonymous by removing from bdrv_state and
1924  * graph_bdrv_state list.
1925    Also, NULL terminate the device_name to prevent double remove */
1926 void bdrv_make_anon(BlockDriverState *bs)
1927 {
1928     if (bs->device_name[0] != '\0') {
1929         QTAILQ_REMOVE(&bdrv_states, bs, device_list);
1930     }
1931     bs->device_name[0] = '\0';
1932     if (bs->node_name[0] != '\0') {
1933         QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
1934     }
1935     bs->node_name[0] = '\0';
1936 }
1937 
1938 static void bdrv_rebind(BlockDriverState *bs)
1939 {
1940     if (bs->drv && bs->drv->bdrv_rebind) {
1941         bs->drv->bdrv_rebind(bs);
1942     }
1943 }
1944 
1945 static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
1946                                      BlockDriverState *bs_src)
1947 {
1948     /* move some fields that need to stay attached to the device */
1949 
1950     /* dev info */
1951     bs_dest->dev_ops            = bs_src->dev_ops;
1952     bs_dest->dev_opaque         = bs_src->dev_opaque;
1953     bs_dest->dev                = bs_src->dev;
1954     bs_dest->guest_block_size   = bs_src->guest_block_size;
1955     bs_dest->copy_on_read       = bs_src->copy_on_read;
1956 
1957     bs_dest->enable_write_cache = bs_src->enable_write_cache;
1958 
1959     /* i/o throttled req */
1960     memcpy(&bs_dest->throttle_state,
1961            &bs_src->throttle_state,
1962            sizeof(ThrottleState));
1963     bs_dest->throttled_reqs[0]  = bs_src->throttled_reqs[0];
1964     bs_dest->throttled_reqs[1]  = bs_src->throttled_reqs[1];
1965     bs_dest->io_limits_enabled  = bs_src->io_limits_enabled;
1966 
1967     /* r/w error */
1968     bs_dest->on_read_error      = bs_src->on_read_error;
1969     bs_dest->on_write_error     = bs_src->on_write_error;
1970 
1971     /* i/o status */
1972     bs_dest->iostatus_enabled   = bs_src->iostatus_enabled;
1973     bs_dest->iostatus           = bs_src->iostatus;
1974 
1975     /* dirty bitmap */
1976     bs_dest->dirty_bitmaps      = bs_src->dirty_bitmaps;
1977 
1978     /* reference count */
1979     bs_dest->refcnt             = bs_src->refcnt;
1980 
1981     /* job */
1982     bs_dest->job                = bs_src->job;
1983 
1984     /* keep the same entry in bdrv_states */
1985     pstrcpy(bs_dest->device_name, sizeof(bs_dest->device_name),
1986             bs_src->device_name);
1987     bs_dest->device_list = bs_src->device_list;
1988     memcpy(bs_dest->op_blockers, bs_src->op_blockers,
1989            sizeof(bs_dest->op_blockers));
1990 }
1991 
1992 /*
1993  * Swap bs contents for two image chains while they are live,
1994  * while keeping required fields on the BlockDriverState that is
1995  * actually attached to a device.
1996  *
1997  * This will modify the BlockDriverState fields, and swap contents
1998  * between bs_new and bs_old. Both bs_new and bs_old are modified.
1999  *
2000  * bs_new is required to be anonymous.
2001  *
2002  * This function does not create any image files.
2003  */
2004 void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
2005 {
2006     BlockDriverState tmp;
2007 
2008     /* The code needs to swap the node_name but simply swapping node_list won't
2009      * work so first remove the nodes from the graph list, do the swap then
2010      * insert them back if needed.
2011      */
2012     if (bs_new->node_name[0] != '\0') {
2013         QTAILQ_REMOVE(&graph_bdrv_states, bs_new, node_list);
2014     }
2015     if (bs_old->node_name[0] != '\0') {
2016         QTAILQ_REMOVE(&graph_bdrv_states, bs_old, node_list);
2017     }
2018 
2019     /* bs_new must be anonymous and shouldn't have anything fancy enabled */
2020     assert(bs_new->device_name[0] == '\0');
2021     assert(QLIST_EMPTY(&bs_new->dirty_bitmaps));
2022     assert(bs_new->job == NULL);
2023     assert(bs_new->dev == NULL);
2024     assert(bs_new->io_limits_enabled == false);
2025     assert(!throttle_have_timer(&bs_new->throttle_state));
2026 
2027     tmp = *bs_new;
2028     *bs_new = *bs_old;
2029     *bs_old = tmp;
2030 
2031     /* there are some fields that should not be swapped, move them back */
2032     bdrv_move_feature_fields(&tmp, bs_old);
2033     bdrv_move_feature_fields(bs_old, bs_new);
2034     bdrv_move_feature_fields(bs_new, &tmp);
2035 
2036     /* bs_new shouldn't be in bdrv_states even after the swap!  */
2037     assert(bs_new->device_name[0] == '\0');
2038 
2039     /* Check a few fields that should remain attached to the device */
2040     assert(bs_new->dev == NULL);
2041     assert(bs_new->job == NULL);
2042     assert(bs_new->io_limits_enabled == false);
2043     assert(!throttle_have_timer(&bs_new->throttle_state));
2044 
2045     /* insert the nodes back into the graph node list if needed */
2046     if (bs_new->node_name[0] != '\0') {
2047         QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs_new, node_list);
2048     }
2049     if (bs_old->node_name[0] != '\0') {
2050         QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs_old, node_list);
2051     }
2052 
2053     bdrv_rebind(bs_new);
2054     bdrv_rebind(bs_old);
2055 }
2056 
2057 /*
2058  * Add new bs contents at the top of an image chain while the chain is
2059  * live, while keeping required fields on the top layer.
2060  *
2061  * This will modify the BlockDriverState fields, and swap contents
2062  * between bs_new and bs_top. Both bs_new and bs_top are modified.
2063  *
2064  * bs_new is required to be anonymous.
2065  *
2066  * This function does not create any image files.
2067  */
2068 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
2069 {
2070     bdrv_swap(bs_new, bs_top);
2071 
2072     /* The contents of 'tmp' will become bs_top, as we are
2073      * swapping bs_new and bs_top contents. */
2074     bdrv_set_backing_hd(bs_top, bs_new);
2075 }
2076 
2077 static void bdrv_delete(BlockDriverState *bs)
2078 {
2079     assert(!bs->dev);
2080     assert(!bs->job);
2081     assert(bdrv_op_blocker_is_empty(bs));
2082     assert(!bs->refcnt);
2083     assert(QLIST_EMPTY(&bs->dirty_bitmaps));
2084 
2085     bdrv_close(bs);
2086 
2087     /* remove from list, if necessary */
2088     bdrv_make_anon(bs);
2089 
2090     g_free(bs);
2091 }
2092 
2093 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
2094 /* TODO change to DeviceState *dev when all users are qdevified */
2095 {
2096     if (bs->dev) {
2097         return -EBUSY;
2098     }
2099     bs->dev = dev;
2100     bdrv_iostatus_reset(bs);
2101     return 0;
2102 }
2103 
2104 /* TODO qdevified devices don't use this, remove when devices are qdevified */
2105 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
2106 {
2107     if (bdrv_attach_dev(bs, dev) < 0) {
2108         abort();
2109     }
2110 }
2111 
2112 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
2113 /* TODO change to DeviceState *dev when all users are qdevified */
2114 {
2115     assert(bs->dev == dev);
2116     bs->dev = NULL;
2117     bs->dev_ops = NULL;
2118     bs->dev_opaque = NULL;
2119     bs->guest_block_size = 512;
2120 }
2121 
2122 /* TODO change to return DeviceState * when all users are qdevified */
2123 void *bdrv_get_attached_dev(BlockDriverState *bs)
2124 {
2125     return bs->dev;
2126 }
2127 
2128 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
2129                       void *opaque)
2130 {
2131     bs->dev_ops = ops;
2132     bs->dev_opaque = opaque;
2133 }
2134 
2135 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
2136                                enum MonitorEvent ev,
2137                                BlockErrorAction action, bool is_read)
2138 {
2139     QObject *data;
2140     const char *action_str;
2141 
2142     switch (action) {
2143     case BDRV_ACTION_REPORT:
2144         action_str = "report";
2145         break;
2146     case BDRV_ACTION_IGNORE:
2147         action_str = "ignore";
2148         break;
2149     case BDRV_ACTION_STOP:
2150         action_str = "stop";
2151         break;
2152     default:
2153         abort();
2154     }
2155 
2156     data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
2157                               bdrv->device_name,
2158                               action_str,
2159                               is_read ? "read" : "write");
2160     monitor_protocol_event(ev, data);
2161 
2162     qobject_decref(data);
2163 }
2164 
2165 static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected)
2166 {
2167     QObject *data;
2168 
2169     data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }",
2170                               bdrv_get_device_name(bs), ejected);
2171     monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data);
2172 
2173     qobject_decref(data);
2174 }
2175 
2176 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
2177 {
2178     if (bs->dev_ops && bs->dev_ops->change_media_cb) {
2179         bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
2180         bs->dev_ops->change_media_cb(bs->dev_opaque, load);
2181         if (tray_was_closed) {
2182             /* tray open */
2183             bdrv_emit_qmp_eject_event(bs, true);
2184         }
2185         if (load) {
2186             /* tray close */
2187             bdrv_emit_qmp_eject_event(bs, false);
2188         }
2189     }
2190 }
2191 
2192 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
2193 {
2194     return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
2195 }
2196 
2197 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
2198 {
2199     if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
2200         bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
2201     }
2202 }
2203 
2204 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
2205 {
2206     if (bs->dev_ops && bs->dev_ops->is_tray_open) {
2207         return bs->dev_ops->is_tray_open(bs->dev_opaque);
2208     }
2209     return false;
2210 }
2211 
2212 static void bdrv_dev_resize_cb(BlockDriverState *bs)
2213 {
2214     if (bs->dev_ops && bs->dev_ops->resize_cb) {
2215         bs->dev_ops->resize_cb(bs->dev_opaque);
2216     }
2217 }
2218 
2219 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
2220 {
2221     if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
2222         return bs->dev_ops->is_medium_locked(bs->dev_opaque);
2223     }
2224     return false;
2225 }
2226 
2227 /*
2228  * Run consistency checks on an image
2229  *
2230  * Returns 0 if the check could be completed (it doesn't mean that the image is
2231  * free of errors) or -errno when an internal error occurred. The results of the
2232  * check are stored in res.
2233  */
2234 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
2235 {
2236     if (bs->drv->bdrv_check == NULL) {
2237         return -ENOTSUP;
2238     }
2239 
2240     memset(res, 0, sizeof(*res));
2241     return bs->drv->bdrv_check(bs, res, fix);
2242 }
2243 
2244 #define COMMIT_BUF_SECTORS 2048
2245 
2246 /* commit COW file into the raw image */
2247 int bdrv_commit(BlockDriverState *bs)
2248 {
2249     BlockDriver *drv = bs->drv;
2250     int64_t sector, total_sectors, length, backing_length;
2251     int n, ro, open_flags;
2252     int ret = 0;
2253     uint8_t *buf = NULL;
2254     char filename[PATH_MAX];
2255 
2256     if (!drv)
2257         return -ENOMEDIUM;
2258 
2259     if (!bs->backing_hd) {
2260         return -ENOTSUP;
2261     }
2262 
2263     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT, NULL) ||
2264         bdrv_op_is_blocked(bs->backing_hd, BLOCK_OP_TYPE_COMMIT, NULL)) {
2265         return -EBUSY;
2266     }
2267 
2268     ro = bs->backing_hd->read_only;
2269     /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
2270     pstrcpy(filename, sizeof(filename), bs->backing_hd->filename);
2271     open_flags =  bs->backing_hd->open_flags;
2272 
2273     if (ro) {
2274         if (bdrv_reopen(bs->backing_hd, open_flags | BDRV_O_RDWR, NULL)) {
2275             return -EACCES;
2276         }
2277     }
2278 
2279     length = bdrv_getlength(bs);
2280     if (length < 0) {
2281         ret = length;
2282         goto ro_cleanup;
2283     }
2284 
2285     backing_length = bdrv_getlength(bs->backing_hd);
2286     if (backing_length < 0) {
2287         ret = backing_length;
2288         goto ro_cleanup;
2289     }
2290 
2291     /* If our top snapshot is larger than the backing file image,
2292      * grow the backing file image if possible.  If not possible,
2293      * we must return an error */
2294     if (length > backing_length) {
2295         ret = bdrv_truncate(bs->backing_hd, length);
2296         if (ret < 0) {
2297             goto ro_cleanup;
2298         }
2299     }
2300 
2301     total_sectors = length >> BDRV_SECTOR_BITS;
2302     buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
2303 
2304     for (sector = 0; sector < total_sectors; sector += n) {
2305         ret = bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n);
2306         if (ret < 0) {
2307             goto ro_cleanup;
2308         }
2309         if (ret) {
2310             ret = bdrv_read(bs, sector, buf, n);
2311             if (ret < 0) {
2312                 goto ro_cleanup;
2313             }
2314 
2315             ret = bdrv_write(bs->backing_hd, sector, buf, n);
2316             if (ret < 0) {
2317                 goto ro_cleanup;
2318             }
2319         }
2320     }
2321 
2322     if (drv->bdrv_make_empty) {
2323         ret = drv->bdrv_make_empty(bs);
2324         if (ret < 0) {
2325             goto ro_cleanup;
2326         }
2327         bdrv_flush(bs);
2328     }
2329 
2330     /*
2331      * Make sure all data we wrote to the backing device is actually
2332      * stable on disk.
2333      */
2334     if (bs->backing_hd) {
2335         bdrv_flush(bs->backing_hd);
2336     }
2337 
2338     ret = 0;
2339 ro_cleanup:
2340     g_free(buf);
2341 
2342     if (ro) {
2343         /* ignoring error return here */
2344         bdrv_reopen(bs->backing_hd, open_flags & ~BDRV_O_RDWR, NULL);
2345     }
2346 
2347     return ret;
2348 }
2349 
2350 int bdrv_commit_all(void)
2351 {
2352     BlockDriverState *bs;
2353 
2354     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
2355         AioContext *aio_context = bdrv_get_aio_context(bs);
2356 
2357         aio_context_acquire(aio_context);
2358         if (bs->drv && bs->backing_hd) {
2359             int ret = bdrv_commit(bs);
2360             if (ret < 0) {
2361                 aio_context_release(aio_context);
2362                 return ret;
2363             }
2364         }
2365         aio_context_release(aio_context);
2366     }
2367     return 0;
2368 }
2369 
2370 /**
2371  * Remove an active request from the tracked requests list
2372  *
2373  * This function should be called when a tracked request is completing.
2374  */
2375 static void tracked_request_end(BdrvTrackedRequest *req)
2376 {
2377     if (req->serialising) {
2378         req->bs->serialising_in_flight--;
2379     }
2380 
2381     QLIST_REMOVE(req, list);
2382     qemu_co_queue_restart_all(&req->wait_queue);
2383 }
2384 
2385 /**
2386  * Add an active request to the tracked requests list
2387  */
2388 static void tracked_request_begin(BdrvTrackedRequest *req,
2389                                   BlockDriverState *bs,
2390                                   int64_t offset,
2391                                   unsigned int bytes, bool is_write)
2392 {
2393     *req = (BdrvTrackedRequest){
2394         .bs = bs,
2395         .offset         = offset,
2396         .bytes          = bytes,
2397         .is_write       = is_write,
2398         .co             = qemu_coroutine_self(),
2399         .serialising    = false,
2400         .overlap_offset = offset,
2401         .overlap_bytes  = bytes,
2402     };
2403 
2404     qemu_co_queue_init(&req->wait_queue);
2405 
2406     QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
2407 }
2408 
2409 static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
2410 {
2411     int64_t overlap_offset = req->offset & ~(align - 1);
2412     unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align)
2413                                - overlap_offset;
2414 
2415     if (!req->serialising) {
2416         req->bs->serialising_in_flight++;
2417         req->serialising = true;
2418     }
2419 
2420     req->overlap_offset = MIN(req->overlap_offset, overlap_offset);
2421     req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes);
2422 }
2423 
2424 /**
2425  * Round a region to cluster boundaries
2426  */
2427 void bdrv_round_to_clusters(BlockDriverState *bs,
2428                             int64_t sector_num, int nb_sectors,
2429                             int64_t *cluster_sector_num,
2430                             int *cluster_nb_sectors)
2431 {
2432     BlockDriverInfo bdi;
2433 
2434     if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
2435         *cluster_sector_num = sector_num;
2436         *cluster_nb_sectors = nb_sectors;
2437     } else {
2438         int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
2439         *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
2440         *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
2441                                             nb_sectors, c);
2442     }
2443 }
2444 
2445 static int bdrv_get_cluster_size(BlockDriverState *bs)
2446 {
2447     BlockDriverInfo bdi;
2448     int ret;
2449 
2450     ret = bdrv_get_info(bs, &bdi);
2451     if (ret < 0 || bdi.cluster_size == 0) {
2452         return bs->request_alignment;
2453     } else {
2454         return bdi.cluster_size;
2455     }
2456 }
2457 
2458 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
2459                                      int64_t offset, unsigned int bytes)
2460 {
2461     /*        aaaa   bbbb */
2462     if (offset >= req->overlap_offset + req->overlap_bytes) {
2463         return false;
2464     }
2465     /* bbbb   aaaa        */
2466     if (req->overlap_offset >= offset + bytes) {
2467         return false;
2468     }
2469     return true;
2470 }
2471 
2472 static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
2473 {
2474     BlockDriverState *bs = self->bs;
2475     BdrvTrackedRequest *req;
2476     bool retry;
2477     bool waited = false;
2478 
2479     if (!bs->serialising_in_flight) {
2480         return false;
2481     }
2482 
2483     do {
2484         retry = false;
2485         QLIST_FOREACH(req, &bs->tracked_requests, list) {
2486             if (req == self || (!req->serialising && !self->serialising)) {
2487                 continue;
2488             }
2489             if (tracked_request_overlaps(req, self->overlap_offset,
2490                                          self->overlap_bytes))
2491             {
2492                 /* Hitting this means there was a reentrant request, for
2493                  * example, a block driver issuing nested requests.  This must
2494                  * never happen since it means deadlock.
2495                  */
2496                 assert(qemu_coroutine_self() != req->co);
2497 
2498                 /* If the request is already (indirectly) waiting for us, or
2499                  * will wait for us as soon as it wakes up, then just go on
2500                  * (instead of producing a deadlock in the former case). */
2501                 if (!req->waiting_for) {
2502                     self->waiting_for = req;
2503                     qemu_co_queue_wait(&req->wait_queue);
2504                     self->waiting_for = NULL;
2505                     retry = true;
2506                     waited = true;
2507                     break;
2508                 }
2509             }
2510         }
2511     } while (retry);
2512 
2513     return waited;
2514 }
2515 
2516 /*
2517  * Return values:
2518  * 0        - success
2519  * -EINVAL  - backing format specified, but no file
2520  * -ENOSPC  - can't update the backing file because no space is left in the
2521  *            image file header
2522  * -ENOTSUP - format driver doesn't support changing the backing file
2523  */
2524 int bdrv_change_backing_file(BlockDriverState *bs,
2525     const char *backing_file, const char *backing_fmt)
2526 {
2527     BlockDriver *drv = bs->drv;
2528     int ret;
2529 
2530     /* Backing file format doesn't make sense without a backing file */
2531     if (backing_fmt && !backing_file) {
2532         return -EINVAL;
2533     }
2534 
2535     if (drv->bdrv_change_backing_file != NULL) {
2536         ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
2537     } else {
2538         ret = -ENOTSUP;
2539     }
2540 
2541     if (ret == 0) {
2542         pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2543         pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2544     }
2545     return ret;
2546 }
2547 
2548 /*
2549  * Finds the image layer in the chain that has 'bs' as its backing file.
2550  *
2551  * active is the current topmost image.
2552  *
2553  * Returns NULL if bs is not found in active's image chain,
2554  * or if active == bs.
2555  */
2556 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
2557                                     BlockDriverState *bs)
2558 {
2559     BlockDriverState *overlay = NULL;
2560     BlockDriverState *intermediate;
2561 
2562     assert(active != NULL);
2563     assert(bs != NULL);
2564 
2565     /* if bs is the same as active, then by definition it has no overlay
2566      */
2567     if (active == bs) {
2568         return NULL;
2569     }
2570 
2571     intermediate = active;
2572     while (intermediate->backing_hd) {
2573         if (intermediate->backing_hd == bs) {
2574             overlay = intermediate;
2575             break;
2576         }
2577         intermediate = intermediate->backing_hd;
2578     }
2579 
2580     return overlay;
2581 }
2582 
2583 typedef struct BlkIntermediateStates {
2584     BlockDriverState *bs;
2585     QSIMPLEQ_ENTRY(BlkIntermediateStates) entry;
2586 } BlkIntermediateStates;
2587 
2588 
2589 /*
2590  * Drops images above 'base' up to and including 'top', and sets the image
2591  * above 'top' to have base as its backing file.
2592  *
2593  * Requires that the overlay to 'top' is opened r/w, so that the backing file
2594  * information in 'bs' can be properly updated.
2595  *
2596  * E.g., this will convert the following chain:
2597  * bottom <- base <- intermediate <- top <- active
2598  *
2599  * to
2600  *
2601  * bottom <- base <- active
2602  *
2603  * It is allowed for bottom==base, in which case it converts:
2604  *
2605  * base <- intermediate <- top <- active
2606  *
2607  * to
2608  *
2609  * base <- active
2610  *
2611  * Error conditions:
2612  *  if active == top, that is considered an error
2613  *
2614  */
2615 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
2616                            BlockDriverState *base)
2617 {
2618     BlockDriverState *intermediate;
2619     BlockDriverState *base_bs = NULL;
2620     BlockDriverState *new_top_bs = NULL;
2621     BlkIntermediateStates *intermediate_state, *next;
2622     int ret = -EIO;
2623 
2624     QSIMPLEQ_HEAD(states_to_delete, BlkIntermediateStates) states_to_delete;
2625     QSIMPLEQ_INIT(&states_to_delete);
2626 
2627     if (!top->drv || !base->drv) {
2628         goto exit;
2629     }
2630 
2631     new_top_bs = bdrv_find_overlay(active, top);
2632 
2633     if (new_top_bs == NULL) {
2634         /* we could not find the image above 'top', this is an error */
2635         goto exit;
2636     }
2637 
2638     /* special case of new_top_bs->backing_hd already pointing to base - nothing
2639      * to do, no intermediate images */
2640     if (new_top_bs->backing_hd == base) {
2641         ret = 0;
2642         goto exit;
2643     }
2644 
2645     intermediate = top;
2646 
2647     /* now we will go down through the list, and add each BDS we find
2648      * into our deletion queue, until we hit the 'base'
2649      */
2650     while (intermediate) {
2651         intermediate_state = g_malloc0(sizeof(BlkIntermediateStates));
2652         intermediate_state->bs = intermediate;
2653         QSIMPLEQ_INSERT_TAIL(&states_to_delete, intermediate_state, entry);
2654 
2655         if (intermediate->backing_hd == base) {
2656             base_bs = intermediate->backing_hd;
2657             break;
2658         }
2659         intermediate = intermediate->backing_hd;
2660     }
2661     if (base_bs == NULL) {
2662         /* something went wrong, we did not end at the base. safely
2663          * unravel everything, and exit with error */
2664         goto exit;
2665     }
2666 
2667     /* success - we can delete the intermediate states, and link top->base */
2668     ret = bdrv_change_backing_file(new_top_bs, base_bs->filename,
2669                                    base_bs->drv ? base_bs->drv->format_name : "");
2670     if (ret) {
2671         goto exit;
2672     }
2673     bdrv_set_backing_hd(new_top_bs, base_bs);
2674 
2675     QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2676         /* so that bdrv_close() does not recursively close the chain */
2677         bdrv_set_backing_hd(intermediate_state->bs, NULL);
2678         bdrv_unref(intermediate_state->bs);
2679     }
2680     ret = 0;
2681 
2682 exit:
2683     QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2684         g_free(intermediate_state);
2685     }
2686     return ret;
2687 }
2688 
2689 
2690 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
2691                                    size_t size)
2692 {
2693     int64_t len;
2694 
2695     if (size > INT_MAX) {
2696         return -EIO;
2697     }
2698 
2699     if (!bdrv_is_inserted(bs))
2700         return -ENOMEDIUM;
2701 
2702     if (bs->growable)
2703         return 0;
2704 
2705     len = bdrv_getlength(bs);
2706 
2707     if (offset < 0)
2708         return -EIO;
2709 
2710     if ((offset > len) || (len - offset < size))
2711         return -EIO;
2712 
2713     return 0;
2714 }
2715 
2716 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
2717                               int nb_sectors)
2718 {
2719     if (nb_sectors < 0 || nb_sectors > INT_MAX / BDRV_SECTOR_SIZE) {
2720         return -EIO;
2721     }
2722 
2723     return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
2724                                    nb_sectors * BDRV_SECTOR_SIZE);
2725 }
2726 
2727 typedef struct RwCo {
2728     BlockDriverState *bs;
2729     int64_t offset;
2730     QEMUIOVector *qiov;
2731     bool is_write;
2732     int ret;
2733     BdrvRequestFlags flags;
2734 } RwCo;
2735 
2736 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
2737 {
2738     RwCo *rwco = opaque;
2739 
2740     if (!rwco->is_write) {
2741         rwco->ret = bdrv_co_do_preadv(rwco->bs, rwco->offset,
2742                                       rwco->qiov->size, rwco->qiov,
2743                                       rwco->flags);
2744     } else {
2745         rwco->ret = bdrv_co_do_pwritev(rwco->bs, rwco->offset,
2746                                        rwco->qiov->size, rwco->qiov,
2747                                        rwco->flags);
2748     }
2749 }
2750 
2751 /*
2752  * Process a vectored synchronous request using coroutines
2753  */
2754 static int bdrv_prwv_co(BlockDriverState *bs, int64_t offset,
2755                         QEMUIOVector *qiov, bool is_write,
2756                         BdrvRequestFlags flags)
2757 {
2758     Coroutine *co;
2759     RwCo rwco = {
2760         .bs = bs,
2761         .offset = offset,
2762         .qiov = qiov,
2763         .is_write = is_write,
2764         .ret = NOT_DONE,
2765         .flags = flags,
2766     };
2767 
2768     /**
2769      * In sync call context, when the vcpu is blocked, this throttling timer
2770      * will not fire; so the I/O throttling function has to be disabled here
2771      * if it has been enabled.
2772      */
2773     if (bs->io_limits_enabled) {
2774         fprintf(stderr, "Disabling I/O throttling on '%s' due "
2775                         "to synchronous I/O.\n", bdrv_get_device_name(bs));
2776         bdrv_io_limits_disable(bs);
2777     }
2778 
2779     if (qemu_in_coroutine()) {
2780         /* Fast-path if already in coroutine context */
2781         bdrv_rw_co_entry(&rwco);
2782     } else {
2783         AioContext *aio_context = bdrv_get_aio_context(bs);
2784 
2785         co = qemu_coroutine_create(bdrv_rw_co_entry);
2786         qemu_coroutine_enter(co, &rwco);
2787         while (rwco.ret == NOT_DONE) {
2788             aio_poll(aio_context, true);
2789         }
2790     }
2791     return rwco.ret;
2792 }
2793 
2794 /*
2795  * Process a synchronous request using coroutines
2796  */
2797 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
2798                       int nb_sectors, bool is_write, BdrvRequestFlags flags)
2799 {
2800     QEMUIOVector qiov;
2801     struct iovec iov = {
2802         .iov_base = (void *)buf,
2803         .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
2804     };
2805 
2806     if (nb_sectors < 0 || nb_sectors > INT_MAX / BDRV_SECTOR_SIZE) {
2807         return -EINVAL;
2808     }
2809 
2810     qemu_iovec_init_external(&qiov, &iov, 1);
2811     return bdrv_prwv_co(bs, sector_num << BDRV_SECTOR_BITS,
2812                         &qiov, is_write, flags);
2813 }
2814 
2815 /* return < 0 if error. See bdrv_write() for the return codes */
2816 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
2817               uint8_t *buf, int nb_sectors)
2818 {
2819     return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0);
2820 }
2821 
2822 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2823 int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num,
2824                           uint8_t *buf, int nb_sectors)
2825 {
2826     bool enabled;
2827     int ret;
2828 
2829     enabled = bs->io_limits_enabled;
2830     bs->io_limits_enabled = false;
2831     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
2832     bs->io_limits_enabled = enabled;
2833     return ret;
2834 }
2835 
2836 /* Return < 0 if error. Important errors are:
2837   -EIO         generic I/O error (may happen for all errors)
2838   -ENOMEDIUM   No media inserted.
2839   -EINVAL      Invalid sector number or nb_sectors
2840   -EACCES      Trying to write a read-only device
2841 */
2842 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
2843                const uint8_t *buf, int nb_sectors)
2844 {
2845     return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
2846 }
2847 
2848 int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num,
2849                       int nb_sectors, BdrvRequestFlags flags)
2850 {
2851     return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true,
2852                       BDRV_REQ_ZERO_WRITE | flags);
2853 }
2854 
2855 /*
2856  * Completely zero out a block device with the help of bdrv_write_zeroes.
2857  * The operation is sped up by checking the block status and only writing
2858  * zeroes to the device if they currently do not return zeroes. Optional
2859  * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
2860  *
2861  * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
2862  */
2863 int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags)
2864 {
2865     int64_t target_size;
2866     int64_t ret, nb_sectors, sector_num = 0;
2867     int n;
2868 
2869     target_size = bdrv_getlength(bs);
2870     if (target_size < 0) {
2871         return target_size;
2872     }
2873     target_size /= BDRV_SECTOR_SIZE;
2874 
2875     for (;;) {
2876         nb_sectors = target_size - sector_num;
2877         if (nb_sectors <= 0) {
2878             return 0;
2879         }
2880         if (nb_sectors > INT_MAX) {
2881             nb_sectors = INT_MAX;
2882         }
2883         ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n);
2884         if (ret < 0) {
2885             error_report("error getting block status at sector %" PRId64 ": %s",
2886                          sector_num, strerror(-ret));
2887             return ret;
2888         }
2889         if (ret & BDRV_BLOCK_ZERO) {
2890             sector_num += n;
2891             continue;
2892         }
2893         ret = bdrv_write_zeroes(bs, sector_num, n, flags);
2894         if (ret < 0) {
2895             error_report("error writing zeroes at sector %" PRId64 ": %s",
2896                          sector_num, strerror(-ret));
2897             return ret;
2898         }
2899         sector_num += n;
2900     }
2901 }
2902 
2903 int bdrv_pread(BlockDriverState *bs, int64_t offset, void *buf, int bytes)
2904 {
2905     QEMUIOVector qiov;
2906     struct iovec iov = {
2907         .iov_base = (void *)buf,
2908         .iov_len = bytes,
2909     };
2910     int ret;
2911 
2912     if (bytes < 0) {
2913         return -EINVAL;
2914     }
2915 
2916     qemu_iovec_init_external(&qiov, &iov, 1);
2917     ret = bdrv_prwv_co(bs, offset, &qiov, false, 0);
2918     if (ret < 0) {
2919         return ret;
2920     }
2921 
2922     return bytes;
2923 }
2924 
2925 int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov)
2926 {
2927     int ret;
2928 
2929     ret = bdrv_prwv_co(bs, offset, qiov, true, 0);
2930     if (ret < 0) {
2931         return ret;
2932     }
2933 
2934     return qiov->size;
2935 }
2936 
2937 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
2938                 const void *buf, int bytes)
2939 {
2940     QEMUIOVector qiov;
2941     struct iovec iov = {
2942         .iov_base   = (void *) buf,
2943         .iov_len    = bytes,
2944     };
2945 
2946     if (bytes < 0) {
2947         return -EINVAL;
2948     }
2949 
2950     qemu_iovec_init_external(&qiov, &iov, 1);
2951     return bdrv_pwritev(bs, offset, &qiov);
2952 }
2953 
2954 /*
2955  * Writes to the file and ensures that no writes are reordered across this
2956  * request (acts as a barrier)
2957  *
2958  * Returns 0 on success, -errno in error cases.
2959  */
2960 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
2961     const void *buf, int count)
2962 {
2963     int ret;
2964 
2965     ret = bdrv_pwrite(bs, offset, buf, count);
2966     if (ret < 0) {
2967         return ret;
2968     }
2969 
2970     /* No flush needed for cache modes that already do it */
2971     if (bs->enable_write_cache) {
2972         bdrv_flush(bs);
2973     }
2974 
2975     return 0;
2976 }
2977 
2978 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
2979         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
2980 {
2981     /* Perform I/O through a temporary buffer so that users who scribble over
2982      * their read buffer while the operation is in progress do not end up
2983      * modifying the image file.  This is critical for zero-copy guest I/O
2984      * where anything might happen inside guest memory.
2985      */
2986     void *bounce_buffer;
2987 
2988     BlockDriver *drv = bs->drv;
2989     struct iovec iov;
2990     QEMUIOVector bounce_qiov;
2991     int64_t cluster_sector_num;
2992     int cluster_nb_sectors;
2993     size_t skip_bytes;
2994     int ret;
2995 
2996     /* Cover entire cluster so no additional backing file I/O is required when
2997      * allocating cluster in the image file.
2998      */
2999     bdrv_round_to_clusters(bs, sector_num, nb_sectors,
3000                            &cluster_sector_num, &cluster_nb_sectors);
3001 
3002     trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
3003                                    cluster_sector_num, cluster_nb_sectors);
3004 
3005     iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
3006     iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
3007     qemu_iovec_init_external(&bounce_qiov, &iov, 1);
3008 
3009     ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
3010                              &bounce_qiov);
3011     if (ret < 0) {
3012         goto err;
3013     }
3014 
3015     if (drv->bdrv_co_write_zeroes &&
3016         buffer_is_zero(bounce_buffer, iov.iov_len)) {
3017         ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
3018                                       cluster_nb_sectors, 0);
3019     } else {
3020         /* This does not change the data on the disk, it is not necessary
3021          * to flush even in cache=writethrough mode.
3022          */
3023         ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
3024                                   &bounce_qiov);
3025     }
3026 
3027     if (ret < 0) {
3028         /* It might be okay to ignore write errors for guest requests.  If this
3029          * is a deliberate copy-on-read then we don't want to ignore the error.
3030          * Simply report it in all cases.
3031          */
3032         goto err;
3033     }
3034 
3035     skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
3036     qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes,
3037                         nb_sectors * BDRV_SECTOR_SIZE);
3038 
3039 err:
3040     qemu_vfree(bounce_buffer);
3041     return ret;
3042 }
3043 
3044 /*
3045  * Forwards an already correctly aligned request to the BlockDriver. This
3046  * handles copy on read and zeroing after EOF; any other features must be
3047  * implemented by the caller.
3048  */
3049 static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs,
3050     BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
3051     int64_t align, QEMUIOVector *qiov, int flags)
3052 {
3053     BlockDriver *drv = bs->drv;
3054     int ret;
3055 
3056     int64_t sector_num = offset >> BDRV_SECTOR_BITS;
3057     unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
3058 
3059     assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
3060     assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
3061 
3062     /* Handle Copy on Read and associated serialisation */
3063     if (flags & BDRV_REQ_COPY_ON_READ) {
3064         /* If we touch the same cluster it counts as an overlap.  This
3065          * guarantees that allocating writes will be serialized and not race
3066          * with each other for the same cluster.  For example, in copy-on-read
3067          * it ensures that the CoR read and write operations are atomic and
3068          * guest writes cannot interleave between them. */
3069         mark_request_serialising(req, bdrv_get_cluster_size(bs));
3070     }
3071 
3072     wait_serialising_requests(req);
3073 
3074     if (flags & BDRV_REQ_COPY_ON_READ) {
3075         int pnum;
3076 
3077         ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum);
3078         if (ret < 0) {
3079             goto out;
3080         }
3081 
3082         if (!ret || pnum != nb_sectors) {
3083             ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
3084             goto out;
3085         }
3086     }
3087 
3088     /* Forward the request to the BlockDriver */
3089     if (!(bs->zero_beyond_eof && bs->growable)) {
3090         ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
3091     } else {
3092         /* Read zeros after EOF of growable BDSes */
3093         int64_t len, total_sectors, max_nb_sectors;
3094 
3095         len = bdrv_getlength(bs);
3096         if (len < 0) {
3097             ret = len;
3098             goto out;
3099         }
3100 
3101         total_sectors = DIV_ROUND_UP(len, BDRV_SECTOR_SIZE);
3102         max_nb_sectors = ROUND_UP(MAX(0, total_sectors - sector_num),
3103                                   align >> BDRV_SECTOR_BITS);
3104         if (max_nb_sectors > 0) {
3105             ret = drv->bdrv_co_readv(bs, sector_num,
3106                                      MIN(nb_sectors, max_nb_sectors), qiov);
3107         } else {
3108             ret = 0;
3109         }
3110 
3111         /* Reading beyond end of file is supposed to produce zeroes */
3112         if (ret == 0 && total_sectors < sector_num + nb_sectors) {
3113             uint64_t offset = MAX(0, total_sectors - sector_num);
3114             uint64_t bytes = (sector_num + nb_sectors - offset) *
3115                               BDRV_SECTOR_SIZE;
3116             qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes);
3117         }
3118     }
3119 
3120 out:
3121     return ret;
3122 }
3123 
3124 /*
3125  * Handle a read request in coroutine context
3126  */
3127 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
3128     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
3129     BdrvRequestFlags flags)
3130 {
3131     BlockDriver *drv = bs->drv;
3132     BdrvTrackedRequest req;
3133 
3134     /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
3135     uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
3136     uint8_t *head_buf = NULL;
3137     uint8_t *tail_buf = NULL;
3138     QEMUIOVector local_qiov;
3139     bool use_local_qiov = false;
3140     int ret;
3141 
3142     if (!drv) {
3143         return -ENOMEDIUM;
3144     }
3145     if (bdrv_check_byte_request(bs, offset, bytes)) {
3146         return -EIO;
3147     }
3148 
3149     if (bs->copy_on_read) {
3150         flags |= BDRV_REQ_COPY_ON_READ;
3151     }
3152 
3153     /* throttling disk I/O */
3154     if (bs->io_limits_enabled) {
3155         bdrv_io_limits_intercept(bs, bytes, false);
3156     }
3157 
3158     /* Align read if necessary by padding qiov */
3159     if (offset & (align - 1)) {
3160         head_buf = qemu_blockalign(bs, align);
3161         qemu_iovec_init(&local_qiov, qiov->niov + 2);
3162         qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
3163         qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3164         use_local_qiov = true;
3165 
3166         bytes += offset & (align - 1);
3167         offset = offset & ~(align - 1);
3168     }
3169 
3170     if ((offset + bytes) & (align - 1)) {
3171         if (!use_local_qiov) {
3172             qemu_iovec_init(&local_qiov, qiov->niov + 1);
3173             qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3174             use_local_qiov = true;
3175         }
3176         tail_buf = qemu_blockalign(bs, align);
3177         qemu_iovec_add(&local_qiov, tail_buf,
3178                        align - ((offset + bytes) & (align - 1)));
3179 
3180         bytes = ROUND_UP(bytes, align);
3181     }
3182 
3183     tracked_request_begin(&req, bs, offset, bytes, false);
3184     ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align,
3185                               use_local_qiov ? &local_qiov : qiov,
3186                               flags);
3187     tracked_request_end(&req);
3188 
3189     if (use_local_qiov) {
3190         qemu_iovec_destroy(&local_qiov);
3191         qemu_vfree(head_buf);
3192         qemu_vfree(tail_buf);
3193     }
3194 
3195     return ret;
3196 }
3197 
3198 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
3199     int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
3200     BdrvRequestFlags flags)
3201 {
3202     if (nb_sectors < 0 || nb_sectors > (UINT_MAX >> BDRV_SECTOR_BITS)) {
3203         return -EINVAL;
3204     }
3205 
3206     return bdrv_co_do_preadv(bs, sector_num << BDRV_SECTOR_BITS,
3207                              nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
3208 }
3209 
3210 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
3211     int nb_sectors, QEMUIOVector *qiov)
3212 {
3213     trace_bdrv_co_readv(bs, sector_num, nb_sectors);
3214 
3215     return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
3216 }
3217 
3218 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
3219     int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
3220 {
3221     trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
3222 
3223     return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
3224                             BDRV_REQ_COPY_ON_READ);
3225 }
3226 
3227 /* if no limit is specified in the BlockLimits use a default
3228  * of 32768 512-byte sectors (16 MiB) per request.
3229  */
3230 #define MAX_WRITE_ZEROES_DEFAULT 32768
3231 
3232 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
3233     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
3234 {
3235     BlockDriver *drv = bs->drv;
3236     QEMUIOVector qiov;
3237     struct iovec iov = {0};
3238     int ret = 0;
3239 
3240     int max_write_zeroes = bs->bl.max_write_zeroes ?
3241                            bs->bl.max_write_zeroes : MAX_WRITE_ZEROES_DEFAULT;
3242 
3243     while (nb_sectors > 0 && !ret) {
3244         int num = nb_sectors;
3245 
3246         /* Align request.  Block drivers can expect the "bulk" of the request
3247          * to be aligned.
3248          */
3249         if (bs->bl.write_zeroes_alignment
3250             && num > bs->bl.write_zeroes_alignment) {
3251             if (sector_num % bs->bl.write_zeroes_alignment != 0) {
3252                 /* Make a small request up to the first aligned sector.  */
3253                 num = bs->bl.write_zeroes_alignment;
3254                 num -= sector_num % bs->bl.write_zeroes_alignment;
3255             } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) {
3256                 /* Shorten the request to the last aligned sector.  num cannot
3257                  * underflow because num > bs->bl.write_zeroes_alignment.
3258                  */
3259                 num -= (sector_num + num) % bs->bl.write_zeroes_alignment;
3260             }
3261         }
3262 
3263         /* limit request size */
3264         if (num > max_write_zeroes) {
3265             num = max_write_zeroes;
3266         }
3267 
3268         ret = -ENOTSUP;
3269         /* First try the efficient write zeroes operation */
3270         if (drv->bdrv_co_write_zeroes) {
3271             ret = drv->bdrv_co_write_zeroes(bs, sector_num, num, flags);
3272         }
3273 
3274         if (ret == -ENOTSUP) {
3275             /* Fall back to bounce buffer if write zeroes is unsupported */
3276             iov.iov_len = num * BDRV_SECTOR_SIZE;
3277             if (iov.iov_base == NULL) {
3278                 iov.iov_base = qemu_blockalign(bs, num * BDRV_SECTOR_SIZE);
3279                 memset(iov.iov_base, 0, num * BDRV_SECTOR_SIZE);
3280             }
3281             qemu_iovec_init_external(&qiov, &iov, 1);
3282 
3283             ret = drv->bdrv_co_writev(bs, sector_num, num, &qiov);
3284 
3285             /* Keep bounce buffer around if it is big enough for all
3286              * all future requests.
3287              */
3288             if (num < max_write_zeroes) {
3289                 qemu_vfree(iov.iov_base);
3290                 iov.iov_base = NULL;
3291             }
3292         }
3293 
3294         sector_num += num;
3295         nb_sectors -= num;
3296     }
3297 
3298     qemu_vfree(iov.iov_base);
3299     return ret;
3300 }
3301 
3302 /*
3303  * Forwards an already correctly aligned write request to the BlockDriver.
3304  */
3305 static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs,
3306     BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
3307     QEMUIOVector *qiov, int flags)
3308 {
3309     BlockDriver *drv = bs->drv;
3310     bool waited;
3311     int ret;
3312 
3313     int64_t sector_num = offset >> BDRV_SECTOR_BITS;
3314     unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
3315 
3316     assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
3317     assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
3318 
3319     waited = wait_serialising_requests(req);
3320     assert(!waited || !req->serialising);
3321     assert(req->overlap_offset <= offset);
3322     assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
3323 
3324     ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req);
3325 
3326     if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF &&
3327         !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_write_zeroes &&
3328         qemu_iovec_is_zero(qiov)) {
3329         flags |= BDRV_REQ_ZERO_WRITE;
3330         if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) {
3331             flags |= BDRV_REQ_MAY_UNMAP;
3332         }
3333     }
3334 
3335     if (ret < 0) {
3336         /* Do nothing, write notifier decided to fail this request */
3337     } else if (flags & BDRV_REQ_ZERO_WRITE) {
3338         BLKDBG_EVENT(bs, BLKDBG_PWRITEV_ZERO);
3339         ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors, flags);
3340     } else {
3341         BLKDBG_EVENT(bs, BLKDBG_PWRITEV);
3342         ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
3343     }
3344     BLKDBG_EVENT(bs, BLKDBG_PWRITEV_DONE);
3345 
3346     if (ret == 0 && !bs->enable_write_cache) {
3347         ret = bdrv_co_flush(bs);
3348     }
3349 
3350     bdrv_set_dirty(bs, sector_num, nb_sectors);
3351 
3352     if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
3353         bs->wr_highest_sector = sector_num + nb_sectors - 1;
3354     }
3355     if (bs->growable && ret >= 0) {
3356         bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors);
3357     }
3358 
3359     return ret;
3360 }
3361 
3362 /*
3363  * Handle a write request in coroutine context
3364  */
3365 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
3366     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
3367     BdrvRequestFlags flags)
3368 {
3369     BdrvTrackedRequest req;
3370     /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
3371     uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
3372     uint8_t *head_buf = NULL;
3373     uint8_t *tail_buf = NULL;
3374     QEMUIOVector local_qiov;
3375     bool use_local_qiov = false;
3376     int ret;
3377 
3378     if (!bs->drv) {
3379         return -ENOMEDIUM;
3380     }
3381     if (bs->read_only) {
3382         return -EACCES;
3383     }
3384     if (bdrv_check_byte_request(bs, offset, bytes)) {
3385         return -EIO;
3386     }
3387 
3388     /* throttling disk I/O */
3389     if (bs->io_limits_enabled) {
3390         bdrv_io_limits_intercept(bs, bytes, true);
3391     }
3392 
3393     /*
3394      * Align write if necessary by performing a read-modify-write cycle.
3395      * Pad qiov with the read parts and be sure to have a tracked request not
3396      * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
3397      */
3398     tracked_request_begin(&req, bs, offset, bytes, true);
3399 
3400     if (offset & (align - 1)) {
3401         QEMUIOVector head_qiov;
3402         struct iovec head_iov;
3403 
3404         mark_request_serialising(&req, align);
3405         wait_serialising_requests(&req);
3406 
3407         head_buf = qemu_blockalign(bs, align);
3408         head_iov = (struct iovec) {
3409             .iov_base   = head_buf,
3410             .iov_len    = align,
3411         };
3412         qemu_iovec_init_external(&head_qiov, &head_iov, 1);
3413 
3414         BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_HEAD);
3415         ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align,
3416                                   align, &head_qiov, 0);
3417         if (ret < 0) {
3418             goto fail;
3419         }
3420         BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
3421 
3422         qemu_iovec_init(&local_qiov, qiov->niov + 2);
3423         qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
3424         qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3425         use_local_qiov = true;
3426 
3427         bytes += offset & (align - 1);
3428         offset = offset & ~(align - 1);
3429     }
3430 
3431     if ((offset + bytes) & (align - 1)) {
3432         QEMUIOVector tail_qiov;
3433         struct iovec tail_iov;
3434         size_t tail_bytes;
3435         bool waited;
3436 
3437         mark_request_serialising(&req, align);
3438         waited = wait_serialising_requests(&req);
3439         assert(!waited || !use_local_qiov);
3440 
3441         tail_buf = qemu_blockalign(bs, align);
3442         tail_iov = (struct iovec) {
3443             .iov_base   = tail_buf,
3444             .iov_len    = align,
3445         };
3446         qemu_iovec_init_external(&tail_qiov, &tail_iov, 1);
3447 
3448         BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_TAIL);
3449         ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align,
3450                                   align, &tail_qiov, 0);
3451         if (ret < 0) {
3452             goto fail;
3453         }
3454         BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
3455 
3456         if (!use_local_qiov) {
3457             qemu_iovec_init(&local_qiov, qiov->niov + 1);
3458             qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3459             use_local_qiov = true;
3460         }
3461 
3462         tail_bytes = (offset + bytes) & (align - 1);
3463         qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes);
3464 
3465         bytes = ROUND_UP(bytes, align);
3466     }
3467 
3468     ret = bdrv_aligned_pwritev(bs, &req, offset, bytes,
3469                                use_local_qiov ? &local_qiov : qiov,
3470                                flags);
3471 
3472 fail:
3473     tracked_request_end(&req);
3474 
3475     if (use_local_qiov) {
3476         qemu_iovec_destroy(&local_qiov);
3477     }
3478     qemu_vfree(head_buf);
3479     qemu_vfree(tail_buf);
3480 
3481     return ret;
3482 }
3483 
3484 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
3485     int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
3486     BdrvRequestFlags flags)
3487 {
3488     if (nb_sectors < 0 || nb_sectors > (INT_MAX >> BDRV_SECTOR_BITS)) {
3489         return -EINVAL;
3490     }
3491 
3492     return bdrv_co_do_pwritev(bs, sector_num << BDRV_SECTOR_BITS,
3493                               nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
3494 }
3495 
3496 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
3497     int nb_sectors, QEMUIOVector *qiov)
3498 {
3499     trace_bdrv_co_writev(bs, sector_num, nb_sectors);
3500 
3501     return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
3502 }
3503 
3504 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
3505                                       int64_t sector_num, int nb_sectors,
3506                                       BdrvRequestFlags flags)
3507 {
3508     trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors, flags);
3509 
3510     if (!(bs->open_flags & BDRV_O_UNMAP)) {
3511         flags &= ~BDRV_REQ_MAY_UNMAP;
3512     }
3513 
3514     return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
3515                              BDRV_REQ_ZERO_WRITE | flags);
3516 }
3517 
3518 /**
3519  * Truncate file to 'offset' bytes (needed only for file protocols)
3520  */
3521 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
3522 {
3523     BlockDriver *drv = bs->drv;
3524     int ret;
3525     if (!drv)
3526         return -ENOMEDIUM;
3527     if (!drv->bdrv_truncate)
3528         return -ENOTSUP;
3529     if (bs->read_only)
3530         return -EACCES;
3531     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3532         return -EBUSY;
3533     }
3534     ret = drv->bdrv_truncate(bs, offset);
3535     if (ret == 0) {
3536         ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3537         bdrv_dev_resize_cb(bs);
3538     }
3539     return ret;
3540 }
3541 
3542 /**
3543  * Length of a allocated file in bytes. Sparse files are counted by actual
3544  * allocated space. Return < 0 if error or unknown.
3545  */
3546 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3547 {
3548     BlockDriver *drv = bs->drv;
3549     if (!drv) {
3550         return -ENOMEDIUM;
3551     }
3552     if (drv->bdrv_get_allocated_file_size) {
3553         return drv->bdrv_get_allocated_file_size(bs);
3554     }
3555     if (bs->file) {
3556         return bdrv_get_allocated_file_size(bs->file);
3557     }
3558     return -ENOTSUP;
3559 }
3560 
3561 /**
3562  * Length of a file in bytes. Return < 0 if error or unknown.
3563  */
3564 int64_t bdrv_getlength(BlockDriverState *bs)
3565 {
3566     BlockDriver *drv = bs->drv;
3567     if (!drv)
3568         return -ENOMEDIUM;
3569 
3570     if (drv->has_variable_length) {
3571         int ret = refresh_total_sectors(bs, bs->total_sectors);
3572         if (ret < 0) {
3573             return ret;
3574         }
3575     }
3576     return bs->total_sectors * BDRV_SECTOR_SIZE;
3577 }
3578 
3579 /* return 0 as number of sectors if no device present or error */
3580 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3581 {
3582     int64_t length;
3583     length = bdrv_getlength(bs);
3584     if (length < 0)
3585         length = 0;
3586     else
3587         length = length >> BDRV_SECTOR_BITS;
3588     *nb_sectors_ptr = length;
3589 }
3590 
3591 void bdrv_set_on_error(BlockDriverState *bs, BlockdevOnError on_read_error,
3592                        BlockdevOnError on_write_error)
3593 {
3594     bs->on_read_error = on_read_error;
3595     bs->on_write_error = on_write_error;
3596 }
3597 
3598 BlockdevOnError bdrv_get_on_error(BlockDriverState *bs, bool is_read)
3599 {
3600     return is_read ? bs->on_read_error : bs->on_write_error;
3601 }
3602 
3603 BlockErrorAction bdrv_get_error_action(BlockDriverState *bs, bool is_read, int error)
3604 {
3605     BlockdevOnError on_err = is_read ? bs->on_read_error : bs->on_write_error;
3606 
3607     switch (on_err) {
3608     case BLOCKDEV_ON_ERROR_ENOSPC:
3609         return (error == ENOSPC) ? BDRV_ACTION_STOP : BDRV_ACTION_REPORT;
3610     case BLOCKDEV_ON_ERROR_STOP:
3611         return BDRV_ACTION_STOP;
3612     case BLOCKDEV_ON_ERROR_REPORT:
3613         return BDRV_ACTION_REPORT;
3614     case BLOCKDEV_ON_ERROR_IGNORE:
3615         return BDRV_ACTION_IGNORE;
3616     default:
3617         abort();
3618     }
3619 }
3620 
3621 /* This is done by device models because, while the block layer knows
3622  * about the error, it does not know whether an operation comes from
3623  * the device or the block layer (from a job, for example).
3624  */
3625 void bdrv_error_action(BlockDriverState *bs, BlockErrorAction action,
3626                        bool is_read, int error)
3627 {
3628     assert(error >= 0);
3629     bdrv_emit_qmp_error_event(bs, QEVENT_BLOCK_IO_ERROR, action, is_read);
3630     if (action == BDRV_ACTION_STOP) {
3631         vm_stop(RUN_STATE_IO_ERROR);
3632         bdrv_iostatus_set_err(bs, error);
3633     }
3634 }
3635 
3636 int bdrv_is_read_only(BlockDriverState *bs)
3637 {
3638     return bs->read_only;
3639 }
3640 
3641 int bdrv_is_sg(BlockDriverState *bs)
3642 {
3643     return bs->sg;
3644 }
3645 
3646 int bdrv_enable_write_cache(BlockDriverState *bs)
3647 {
3648     return bs->enable_write_cache;
3649 }
3650 
3651 void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce)
3652 {
3653     bs->enable_write_cache = wce;
3654 
3655     /* so a reopen() will preserve wce */
3656     if (wce) {
3657         bs->open_flags |= BDRV_O_CACHE_WB;
3658     } else {
3659         bs->open_flags &= ~BDRV_O_CACHE_WB;
3660     }
3661 }
3662 
3663 int bdrv_is_encrypted(BlockDriverState *bs)
3664 {
3665     if (bs->backing_hd && bs->backing_hd->encrypted)
3666         return 1;
3667     return bs->encrypted;
3668 }
3669 
3670 int bdrv_key_required(BlockDriverState *bs)
3671 {
3672     BlockDriverState *backing_hd = bs->backing_hd;
3673 
3674     if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
3675         return 1;
3676     return (bs->encrypted && !bs->valid_key);
3677 }
3678 
3679 int bdrv_set_key(BlockDriverState *bs, const char *key)
3680 {
3681     int ret;
3682     if (bs->backing_hd && bs->backing_hd->encrypted) {
3683         ret = bdrv_set_key(bs->backing_hd, key);
3684         if (ret < 0)
3685             return ret;
3686         if (!bs->encrypted)
3687             return 0;
3688     }
3689     if (!bs->encrypted) {
3690         return -EINVAL;
3691     } else if (!bs->drv || !bs->drv->bdrv_set_key) {
3692         return -ENOMEDIUM;
3693     }
3694     ret = bs->drv->bdrv_set_key(bs, key);
3695     if (ret < 0) {
3696         bs->valid_key = 0;
3697     } else if (!bs->valid_key) {
3698         bs->valid_key = 1;
3699         /* call the change callback now, we skipped it on open */
3700         bdrv_dev_change_media_cb(bs, true);
3701     }
3702     return ret;
3703 }
3704 
3705 const char *bdrv_get_format_name(BlockDriverState *bs)
3706 {
3707     return bs->drv ? bs->drv->format_name : NULL;
3708 }
3709 
3710 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3711                          void *opaque)
3712 {
3713     BlockDriver *drv;
3714     int count = 0;
3715     const char **formats = NULL;
3716 
3717     QLIST_FOREACH(drv, &bdrv_drivers, list) {
3718         if (drv->format_name) {
3719             bool found = false;
3720             int i = count;
3721             while (formats && i && !found) {
3722                 found = !strcmp(formats[--i], drv->format_name);
3723             }
3724 
3725             if (!found) {
3726                 formats = g_realloc(formats, (count + 1) * sizeof(char *));
3727                 formats[count++] = drv->format_name;
3728                 it(opaque, drv->format_name);
3729             }
3730         }
3731     }
3732     g_free(formats);
3733 }
3734 
3735 /* This function is to find block backend bs */
3736 BlockDriverState *bdrv_find(const char *name)
3737 {
3738     BlockDriverState *bs;
3739 
3740     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3741         if (!strcmp(name, bs->device_name)) {
3742             return bs;
3743         }
3744     }
3745     return NULL;
3746 }
3747 
3748 /* This function is to find a node in the bs graph */
3749 BlockDriverState *bdrv_find_node(const char *node_name)
3750 {
3751     BlockDriverState *bs;
3752 
3753     assert(node_name);
3754 
3755     QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3756         if (!strcmp(node_name, bs->node_name)) {
3757             return bs;
3758         }
3759     }
3760     return NULL;
3761 }
3762 
3763 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3764 BlockDeviceInfoList *bdrv_named_nodes_list(void)
3765 {
3766     BlockDeviceInfoList *list, *entry;
3767     BlockDriverState *bs;
3768 
3769     list = NULL;
3770     QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3771         entry = g_malloc0(sizeof(*entry));
3772         entry->value = bdrv_block_device_info(bs);
3773         entry->next = list;
3774         list = entry;
3775     }
3776 
3777     return list;
3778 }
3779 
3780 BlockDriverState *bdrv_lookup_bs(const char *device,
3781                                  const char *node_name,
3782                                  Error **errp)
3783 {
3784     BlockDriverState *bs = NULL;
3785 
3786     if (device) {
3787         bs = bdrv_find(device);
3788 
3789         if (bs) {
3790             return bs;
3791         }
3792     }
3793 
3794     if (node_name) {
3795         bs = bdrv_find_node(node_name);
3796 
3797         if (bs) {
3798             return bs;
3799         }
3800     }
3801 
3802     error_setg(errp, "Cannot find device=%s nor node_name=%s",
3803                      device ? device : "",
3804                      node_name ? node_name : "");
3805     return NULL;
3806 }
3807 
3808 BlockDriverState *bdrv_next(BlockDriverState *bs)
3809 {
3810     if (!bs) {
3811         return QTAILQ_FIRST(&bdrv_states);
3812     }
3813     return QTAILQ_NEXT(bs, device_list);
3814 }
3815 
3816 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
3817 {
3818     BlockDriverState *bs;
3819 
3820     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3821         it(opaque, bs);
3822     }
3823 }
3824 
3825 const char *bdrv_get_device_name(BlockDriverState *bs)
3826 {
3827     return bs->device_name;
3828 }
3829 
3830 int bdrv_get_flags(BlockDriverState *bs)
3831 {
3832     return bs->open_flags;
3833 }
3834 
3835 int bdrv_flush_all(void)
3836 {
3837     BlockDriverState *bs;
3838     int result = 0;
3839 
3840     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3841         AioContext *aio_context = bdrv_get_aio_context(bs);
3842         int ret;
3843 
3844         aio_context_acquire(aio_context);
3845         ret = bdrv_flush(bs);
3846         if (ret < 0 && !result) {
3847             result = ret;
3848         }
3849         aio_context_release(aio_context);
3850     }
3851 
3852     return result;
3853 }
3854 
3855 int bdrv_has_zero_init_1(BlockDriverState *bs)
3856 {
3857     return 1;
3858 }
3859 
3860 int bdrv_has_zero_init(BlockDriverState *bs)
3861 {
3862     assert(bs->drv);
3863 
3864     /* If BS is a copy on write image, it is initialized to
3865        the contents of the base image, which may not be zeroes.  */
3866     if (bs->backing_hd) {
3867         return 0;
3868     }
3869     if (bs->drv->bdrv_has_zero_init) {
3870         return bs->drv->bdrv_has_zero_init(bs);
3871     }
3872 
3873     /* safe default */
3874     return 0;
3875 }
3876 
3877 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3878 {
3879     BlockDriverInfo bdi;
3880 
3881     if (bs->backing_hd) {
3882         return false;
3883     }
3884 
3885     if (bdrv_get_info(bs, &bdi) == 0) {
3886         return bdi.unallocated_blocks_are_zero;
3887     }
3888 
3889     return false;
3890 }
3891 
3892 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
3893 {
3894     BlockDriverInfo bdi;
3895 
3896     if (bs->backing_hd || !(bs->open_flags & BDRV_O_UNMAP)) {
3897         return false;
3898     }
3899 
3900     if (bdrv_get_info(bs, &bdi) == 0) {
3901         return bdi.can_write_zeroes_with_unmap;
3902     }
3903 
3904     return false;
3905 }
3906 
3907 typedef struct BdrvCoGetBlockStatusData {
3908     BlockDriverState *bs;
3909     BlockDriverState *base;
3910     int64_t sector_num;
3911     int nb_sectors;
3912     int *pnum;
3913     int64_t ret;
3914     bool done;
3915 } BdrvCoGetBlockStatusData;
3916 
3917 /*
3918  * Returns true iff the specified sector is present in the disk image. Drivers
3919  * not implementing the functionality are assumed to not support backing files,
3920  * hence all their sectors are reported as allocated.
3921  *
3922  * If 'sector_num' is beyond the end of the disk image the return value is 0
3923  * and 'pnum' is set to 0.
3924  *
3925  * 'pnum' is set to the number of sectors (including and immediately following
3926  * the specified sector) that are known to be in the same
3927  * allocated/unallocated state.
3928  *
3929  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
3930  * beyond the end of the disk image it will be clamped.
3931  */
3932 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,
3933                                                      int64_t sector_num,
3934                                                      int nb_sectors, int *pnum)
3935 {
3936     int64_t length;
3937     int64_t n;
3938     int64_t ret, ret2;
3939 
3940     length = bdrv_getlength(bs);
3941     if (length < 0) {
3942         return length;
3943     }
3944 
3945     if (sector_num >= (length >> BDRV_SECTOR_BITS)) {
3946         *pnum = 0;
3947         return 0;
3948     }
3949 
3950     n = bs->total_sectors - sector_num;
3951     if (n < nb_sectors) {
3952         nb_sectors = n;
3953     }
3954 
3955     if (!bs->drv->bdrv_co_get_block_status) {
3956         *pnum = nb_sectors;
3957         ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED;
3958         if (bs->drv->protocol_name) {
3959             ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE);
3960         }
3961         return ret;
3962     }
3963 
3964     ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum);
3965     if (ret < 0) {
3966         *pnum = 0;
3967         return ret;
3968     }
3969 
3970     if (ret & BDRV_BLOCK_RAW) {
3971         assert(ret & BDRV_BLOCK_OFFSET_VALID);
3972         return bdrv_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
3973                                      *pnum, pnum);
3974     }
3975 
3976     if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) {
3977         ret |= BDRV_BLOCK_ALLOCATED;
3978     }
3979 
3980     if (!(ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO)) {
3981         if (bdrv_unallocated_blocks_are_zero(bs)) {
3982             ret |= BDRV_BLOCK_ZERO;
3983         } else if (bs->backing_hd) {
3984             BlockDriverState *bs2 = bs->backing_hd;
3985             int64_t length2 = bdrv_getlength(bs2);
3986             if (length2 >= 0 && sector_num >= (length2 >> BDRV_SECTOR_BITS)) {
3987                 ret |= BDRV_BLOCK_ZERO;
3988             }
3989         }
3990     }
3991 
3992     if (bs->file &&
3993         (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
3994         (ret & BDRV_BLOCK_OFFSET_VALID)) {
3995         ret2 = bdrv_co_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
3996                                         *pnum, pnum);
3997         if (ret2 >= 0) {
3998             /* Ignore errors.  This is just providing extra information, it
3999              * is useful but not necessary.
4000              */
4001             ret |= (ret2 & BDRV_BLOCK_ZERO);
4002         }
4003     }
4004 
4005     return ret;
4006 }
4007 
4008 /* Coroutine wrapper for bdrv_get_block_status() */
4009 static void coroutine_fn bdrv_get_block_status_co_entry(void *opaque)
4010 {
4011     BdrvCoGetBlockStatusData *data = opaque;
4012     BlockDriverState *bs = data->bs;
4013 
4014     data->ret = bdrv_co_get_block_status(bs, data->sector_num, data->nb_sectors,
4015                                          data->pnum);
4016     data->done = true;
4017 }
4018 
4019 /*
4020  * Synchronous wrapper around bdrv_co_get_block_status().
4021  *
4022  * See bdrv_co_get_block_status() for details.
4023  */
4024 int64_t bdrv_get_block_status(BlockDriverState *bs, int64_t sector_num,
4025                               int nb_sectors, int *pnum)
4026 {
4027     Coroutine *co;
4028     BdrvCoGetBlockStatusData data = {
4029         .bs = bs,
4030         .sector_num = sector_num,
4031         .nb_sectors = nb_sectors,
4032         .pnum = pnum,
4033         .done = false,
4034     };
4035 
4036     if (qemu_in_coroutine()) {
4037         /* Fast-path if already in coroutine context */
4038         bdrv_get_block_status_co_entry(&data);
4039     } else {
4040         AioContext *aio_context = bdrv_get_aio_context(bs);
4041 
4042         co = qemu_coroutine_create(bdrv_get_block_status_co_entry);
4043         qemu_coroutine_enter(co, &data);
4044         while (!data.done) {
4045             aio_poll(aio_context, true);
4046         }
4047     }
4048     return data.ret;
4049 }
4050 
4051 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num,
4052                                    int nb_sectors, int *pnum)
4053 {
4054     int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum);
4055     if (ret < 0) {
4056         return ret;
4057     }
4058     return (ret & BDRV_BLOCK_ALLOCATED);
4059 }
4060 
4061 /*
4062  * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
4063  *
4064  * Return true if the given sector is allocated in any image between
4065  * BASE and TOP (inclusive).  BASE can be NULL to check if the given
4066  * sector is allocated in any image of the chain.  Return false otherwise.
4067  *
4068  * 'pnum' is set to the number of sectors (including and immediately following
4069  *  the specified sector) that are known to be in the same
4070  *  allocated/unallocated state.
4071  *
4072  */
4073 int bdrv_is_allocated_above(BlockDriverState *top,
4074                             BlockDriverState *base,
4075                             int64_t sector_num,
4076                             int nb_sectors, int *pnum)
4077 {
4078     BlockDriverState *intermediate;
4079     int ret, n = nb_sectors;
4080 
4081     intermediate = top;
4082     while (intermediate && intermediate != base) {
4083         int pnum_inter;
4084         ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors,
4085                                 &pnum_inter);
4086         if (ret < 0) {
4087             return ret;
4088         } else if (ret) {
4089             *pnum = pnum_inter;
4090             return 1;
4091         }
4092 
4093         /*
4094          * [sector_num, nb_sectors] is unallocated on top but intermediate
4095          * might have
4096          *
4097          * [sector_num+x, nr_sectors] allocated.
4098          */
4099         if (n > pnum_inter &&
4100             (intermediate == top ||
4101              sector_num + pnum_inter < intermediate->total_sectors)) {
4102             n = pnum_inter;
4103         }
4104 
4105         intermediate = intermediate->backing_hd;
4106     }
4107 
4108     *pnum = n;
4109     return 0;
4110 }
4111 
4112 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
4113 {
4114     if (bs->backing_hd && bs->backing_hd->encrypted)
4115         return bs->backing_file;
4116     else if (bs->encrypted)
4117         return bs->filename;
4118     else
4119         return NULL;
4120 }
4121 
4122 void bdrv_get_backing_filename(BlockDriverState *bs,
4123                                char *filename, int filename_size)
4124 {
4125     pstrcpy(filename, filename_size, bs->backing_file);
4126 }
4127 
4128 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
4129                           const uint8_t *buf, int nb_sectors)
4130 {
4131     BlockDriver *drv = bs->drv;
4132     if (!drv)
4133         return -ENOMEDIUM;
4134     if (!drv->bdrv_write_compressed)
4135         return -ENOTSUP;
4136     if (bdrv_check_request(bs, sector_num, nb_sectors))
4137         return -EIO;
4138 
4139     assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4140 
4141     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
4142 }
4143 
4144 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4145 {
4146     BlockDriver *drv = bs->drv;
4147     if (!drv)
4148         return -ENOMEDIUM;
4149     if (!drv->bdrv_get_info)
4150         return -ENOTSUP;
4151     memset(bdi, 0, sizeof(*bdi));
4152     return drv->bdrv_get_info(bs, bdi);
4153 }
4154 
4155 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
4156 {
4157     BlockDriver *drv = bs->drv;
4158     if (drv && drv->bdrv_get_specific_info) {
4159         return drv->bdrv_get_specific_info(bs);
4160     }
4161     return NULL;
4162 }
4163 
4164 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
4165                       int64_t pos, int size)
4166 {
4167     QEMUIOVector qiov;
4168     struct iovec iov = {
4169         .iov_base   = (void *) buf,
4170         .iov_len    = size,
4171     };
4172 
4173     qemu_iovec_init_external(&qiov, &iov, 1);
4174     return bdrv_writev_vmstate(bs, &qiov, pos);
4175 }
4176 
4177 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
4178 {
4179     BlockDriver *drv = bs->drv;
4180 
4181     if (!drv) {
4182         return -ENOMEDIUM;
4183     } else if (drv->bdrv_save_vmstate) {
4184         return drv->bdrv_save_vmstate(bs, qiov, pos);
4185     } else if (bs->file) {
4186         return bdrv_writev_vmstate(bs->file, qiov, pos);
4187     }
4188 
4189     return -ENOTSUP;
4190 }
4191 
4192 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
4193                       int64_t pos, int size)
4194 {
4195     BlockDriver *drv = bs->drv;
4196     if (!drv)
4197         return -ENOMEDIUM;
4198     if (drv->bdrv_load_vmstate)
4199         return drv->bdrv_load_vmstate(bs, buf, pos, size);
4200     if (bs->file)
4201         return bdrv_load_vmstate(bs->file, buf, pos, size);
4202     return -ENOTSUP;
4203 }
4204 
4205 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
4206 {
4207     if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4208         return;
4209     }
4210 
4211     bs->drv->bdrv_debug_event(bs, event);
4212 }
4213 
4214 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4215                           const char *tag)
4216 {
4217     while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4218         bs = bs->file;
4219     }
4220 
4221     if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4222         return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4223     }
4224 
4225     return -ENOTSUP;
4226 }
4227 
4228 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4229 {
4230     while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4231         bs = bs->file;
4232     }
4233 
4234     if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4235         return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4236     }
4237 
4238     return -ENOTSUP;
4239 }
4240 
4241 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4242 {
4243     while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
4244         bs = bs->file;
4245     }
4246 
4247     if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4248         return bs->drv->bdrv_debug_resume(bs, tag);
4249     }
4250 
4251     return -ENOTSUP;
4252 }
4253 
4254 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4255 {
4256     while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4257         bs = bs->file;
4258     }
4259 
4260     if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4261         return bs->drv->bdrv_debug_is_suspended(bs, tag);
4262     }
4263 
4264     return false;
4265 }
4266 
4267 int bdrv_is_snapshot(BlockDriverState *bs)
4268 {
4269     return !!(bs->open_flags & BDRV_O_SNAPSHOT);
4270 }
4271 
4272 /* backing_file can either be relative, or absolute, or a protocol.  If it is
4273  * relative, it must be relative to the chain.  So, passing in bs->filename
4274  * from a BDS as backing_file should not be done, as that may be relative to
4275  * the CWD rather than the chain. */
4276 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4277         const char *backing_file)
4278 {
4279     char *filename_full = NULL;
4280     char *backing_file_full = NULL;
4281     char *filename_tmp = NULL;
4282     int is_protocol = 0;
4283     BlockDriverState *curr_bs = NULL;
4284     BlockDriverState *retval = NULL;
4285 
4286     if (!bs || !bs->drv || !backing_file) {
4287         return NULL;
4288     }
4289 
4290     filename_full     = g_malloc(PATH_MAX);
4291     backing_file_full = g_malloc(PATH_MAX);
4292     filename_tmp      = g_malloc(PATH_MAX);
4293 
4294     is_protocol = path_has_protocol(backing_file);
4295 
4296     for (curr_bs = bs; curr_bs->backing_hd; curr_bs = curr_bs->backing_hd) {
4297 
4298         /* If either of the filename paths is actually a protocol, then
4299          * compare unmodified paths; otherwise make paths relative */
4300         if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4301             if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4302                 retval = curr_bs->backing_hd;
4303                 break;
4304             }
4305         } else {
4306             /* If not an absolute filename path, make it relative to the current
4307              * image's filename path */
4308             path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4309                          backing_file);
4310 
4311             /* We are going to compare absolute pathnames */
4312             if (!realpath(filename_tmp, filename_full)) {
4313                 continue;
4314             }
4315 
4316             /* We need to make sure the backing filename we are comparing against
4317              * is relative to the current image filename (or absolute) */
4318             path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4319                          curr_bs->backing_file);
4320 
4321             if (!realpath(filename_tmp, backing_file_full)) {
4322                 continue;
4323             }
4324 
4325             if (strcmp(backing_file_full, filename_full) == 0) {
4326                 retval = curr_bs->backing_hd;
4327                 break;
4328             }
4329         }
4330     }
4331 
4332     g_free(filename_full);
4333     g_free(backing_file_full);
4334     g_free(filename_tmp);
4335     return retval;
4336 }
4337 
4338 int bdrv_get_backing_file_depth(BlockDriverState *bs)
4339 {
4340     if (!bs->drv) {
4341         return 0;
4342     }
4343 
4344     if (!bs->backing_hd) {
4345         return 0;
4346     }
4347 
4348     return 1 + bdrv_get_backing_file_depth(bs->backing_hd);
4349 }
4350 
4351 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
4352 {
4353     BlockDriverState *curr_bs = NULL;
4354 
4355     if (!bs) {
4356         return NULL;
4357     }
4358 
4359     curr_bs = bs;
4360 
4361     while (curr_bs->backing_hd) {
4362         curr_bs = curr_bs->backing_hd;
4363     }
4364     return curr_bs;
4365 }
4366 
4367 /**************************************************************/
4368 /* async I/Os */
4369 
4370 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
4371                                  QEMUIOVector *qiov, int nb_sectors,
4372                                  BlockDriverCompletionFunc *cb, void *opaque)
4373 {
4374     trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
4375 
4376     return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
4377                                  cb, opaque, false);
4378 }
4379 
4380 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
4381                                   QEMUIOVector *qiov, int nb_sectors,
4382                                   BlockDriverCompletionFunc *cb, void *opaque)
4383 {
4384     trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
4385 
4386     return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
4387                                  cb, opaque, true);
4388 }
4389 
4390 BlockDriverAIOCB *bdrv_aio_write_zeroes(BlockDriverState *bs,
4391         int64_t sector_num, int nb_sectors, BdrvRequestFlags flags,
4392         BlockDriverCompletionFunc *cb, void *opaque)
4393 {
4394     trace_bdrv_aio_write_zeroes(bs, sector_num, nb_sectors, flags, opaque);
4395 
4396     return bdrv_co_aio_rw_vector(bs, sector_num, NULL, nb_sectors,
4397                                  BDRV_REQ_ZERO_WRITE | flags,
4398                                  cb, opaque, true);
4399 }
4400 
4401 
4402 typedef struct MultiwriteCB {
4403     int error;
4404     int num_requests;
4405     int num_callbacks;
4406     struct {
4407         BlockDriverCompletionFunc *cb;
4408         void *opaque;
4409         QEMUIOVector *free_qiov;
4410     } callbacks[];
4411 } MultiwriteCB;
4412 
4413 static void multiwrite_user_cb(MultiwriteCB *mcb)
4414 {
4415     int i;
4416 
4417     for (i = 0; i < mcb->num_callbacks; i++) {
4418         mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
4419         if (mcb->callbacks[i].free_qiov) {
4420             qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
4421         }
4422         g_free(mcb->callbacks[i].free_qiov);
4423     }
4424 }
4425 
4426 static void multiwrite_cb(void *opaque, int ret)
4427 {
4428     MultiwriteCB *mcb = opaque;
4429 
4430     trace_multiwrite_cb(mcb, ret);
4431 
4432     if (ret < 0 && !mcb->error) {
4433         mcb->error = ret;
4434     }
4435 
4436     mcb->num_requests--;
4437     if (mcb->num_requests == 0) {
4438         multiwrite_user_cb(mcb);
4439         g_free(mcb);
4440     }
4441 }
4442 
4443 static int multiwrite_req_compare(const void *a, const void *b)
4444 {
4445     const BlockRequest *req1 = a, *req2 = b;
4446 
4447     /*
4448      * Note that we can't simply subtract req2->sector from req1->sector
4449      * here as that could overflow the return value.
4450      */
4451     if (req1->sector > req2->sector) {
4452         return 1;
4453     } else if (req1->sector < req2->sector) {
4454         return -1;
4455     } else {
4456         return 0;
4457     }
4458 }
4459 
4460 /*
4461  * Takes a bunch of requests and tries to merge them. Returns the number of
4462  * requests that remain after merging.
4463  */
4464 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
4465     int num_reqs, MultiwriteCB *mcb)
4466 {
4467     int i, outidx;
4468 
4469     // Sort requests by start sector
4470     qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
4471 
4472     // Check if adjacent requests touch the same clusters. If so, combine them,
4473     // filling up gaps with zero sectors.
4474     outidx = 0;
4475     for (i = 1; i < num_reqs; i++) {
4476         int merge = 0;
4477         int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
4478 
4479         // Handle exactly sequential writes and overlapping writes.
4480         if (reqs[i].sector <= oldreq_last) {
4481             merge = 1;
4482         }
4483 
4484         if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
4485             merge = 0;
4486         }
4487 
4488         if (merge) {
4489             size_t size;
4490             QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
4491             qemu_iovec_init(qiov,
4492                 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
4493 
4494             // Add the first request to the merged one. If the requests are
4495             // overlapping, drop the last sectors of the first request.
4496             size = (reqs[i].sector - reqs[outidx].sector) << 9;
4497             qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size);
4498 
4499             // We should need to add any zeros between the two requests
4500             assert (reqs[i].sector <= oldreq_last);
4501 
4502             // Add the second request
4503             qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size);
4504 
4505             reqs[outidx].nb_sectors = qiov->size >> 9;
4506             reqs[outidx].qiov = qiov;
4507 
4508             mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
4509         } else {
4510             outidx++;
4511             reqs[outidx].sector     = reqs[i].sector;
4512             reqs[outidx].nb_sectors = reqs[i].nb_sectors;
4513             reqs[outidx].qiov       = reqs[i].qiov;
4514         }
4515     }
4516 
4517     return outidx + 1;
4518 }
4519 
4520 /*
4521  * Submit multiple AIO write requests at once.
4522  *
4523  * On success, the function returns 0 and all requests in the reqs array have
4524  * been submitted. In error case this function returns -1, and any of the
4525  * requests may or may not be submitted yet. In particular, this means that the
4526  * callback will be called for some of the requests, for others it won't. The
4527  * caller must check the error field of the BlockRequest to wait for the right
4528  * callbacks (if error != 0, no callback will be called).
4529  *
4530  * The implementation may modify the contents of the reqs array, e.g. to merge
4531  * requests. However, the fields opaque and error are left unmodified as they
4532  * are used to signal failure for a single request to the caller.
4533  */
4534 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
4535 {
4536     MultiwriteCB *mcb;
4537     int i;
4538 
4539     /* don't submit writes if we don't have a medium */
4540     if (bs->drv == NULL) {
4541         for (i = 0; i < num_reqs; i++) {
4542             reqs[i].error = -ENOMEDIUM;
4543         }
4544         return -1;
4545     }
4546 
4547     if (num_reqs == 0) {
4548         return 0;
4549     }
4550 
4551     // Create MultiwriteCB structure
4552     mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
4553     mcb->num_requests = 0;
4554     mcb->num_callbacks = num_reqs;
4555 
4556     for (i = 0; i < num_reqs; i++) {
4557         mcb->callbacks[i].cb = reqs[i].cb;
4558         mcb->callbacks[i].opaque = reqs[i].opaque;
4559     }
4560 
4561     // Check for mergable requests
4562     num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
4563 
4564     trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
4565 
4566     /* Run the aio requests. */
4567     mcb->num_requests = num_reqs;
4568     for (i = 0; i < num_reqs; i++) {
4569         bdrv_co_aio_rw_vector(bs, reqs[i].sector, reqs[i].qiov,
4570                               reqs[i].nb_sectors, reqs[i].flags,
4571                               multiwrite_cb, mcb,
4572                               true);
4573     }
4574 
4575     return 0;
4576 }
4577 
4578 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
4579 {
4580     acb->aiocb_info->cancel(acb);
4581 }
4582 
4583 /**************************************************************/
4584 /* async block device emulation */
4585 
4586 typedef struct BlockDriverAIOCBSync {
4587     BlockDriverAIOCB common;
4588     QEMUBH *bh;
4589     int ret;
4590     /* vector translation state */
4591     QEMUIOVector *qiov;
4592     uint8_t *bounce;
4593     int is_write;
4594 } BlockDriverAIOCBSync;
4595 
4596 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
4597 {
4598     BlockDriverAIOCBSync *acb =
4599         container_of(blockacb, BlockDriverAIOCBSync, common);
4600     qemu_bh_delete(acb->bh);
4601     acb->bh = NULL;
4602     qemu_aio_release(acb);
4603 }
4604 
4605 static const AIOCBInfo bdrv_em_aiocb_info = {
4606     .aiocb_size         = sizeof(BlockDriverAIOCBSync),
4607     .cancel             = bdrv_aio_cancel_em,
4608 };
4609 
4610 static void bdrv_aio_bh_cb(void *opaque)
4611 {
4612     BlockDriverAIOCBSync *acb = opaque;
4613 
4614     if (!acb->is_write)
4615         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
4616     qemu_vfree(acb->bounce);
4617     acb->common.cb(acb->common.opaque, acb->ret);
4618     qemu_bh_delete(acb->bh);
4619     acb->bh = NULL;
4620     qemu_aio_release(acb);
4621 }
4622 
4623 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
4624                                             int64_t sector_num,
4625                                             QEMUIOVector *qiov,
4626                                             int nb_sectors,
4627                                             BlockDriverCompletionFunc *cb,
4628                                             void *opaque,
4629                                             int is_write)
4630 
4631 {
4632     BlockDriverAIOCBSync *acb;
4633 
4634     acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque);
4635     acb->is_write = is_write;
4636     acb->qiov = qiov;
4637     acb->bounce = qemu_blockalign(bs, qiov->size);
4638     acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_aio_bh_cb, acb);
4639 
4640     if (is_write) {
4641         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
4642         acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
4643     } else {
4644         acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
4645     }
4646 
4647     qemu_bh_schedule(acb->bh);
4648 
4649     return &acb->common;
4650 }
4651 
4652 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
4653         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4654         BlockDriverCompletionFunc *cb, void *opaque)
4655 {
4656     return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
4657 }
4658 
4659 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
4660         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4661         BlockDriverCompletionFunc *cb, void *opaque)
4662 {
4663     return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
4664 }
4665 
4666 
4667 typedef struct BlockDriverAIOCBCoroutine {
4668     BlockDriverAIOCB common;
4669     BlockRequest req;
4670     bool is_write;
4671     bool *done;
4672     QEMUBH* bh;
4673 } BlockDriverAIOCBCoroutine;
4674 
4675 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
4676 {
4677     AioContext *aio_context = bdrv_get_aio_context(blockacb->bs);
4678     BlockDriverAIOCBCoroutine *acb =
4679         container_of(blockacb, BlockDriverAIOCBCoroutine, common);
4680     bool done = false;
4681 
4682     acb->done = &done;
4683     while (!done) {
4684         aio_poll(aio_context, true);
4685     }
4686 }
4687 
4688 static const AIOCBInfo bdrv_em_co_aiocb_info = {
4689     .aiocb_size         = sizeof(BlockDriverAIOCBCoroutine),
4690     .cancel             = bdrv_aio_co_cancel_em,
4691 };
4692 
4693 static void bdrv_co_em_bh(void *opaque)
4694 {
4695     BlockDriverAIOCBCoroutine *acb = opaque;
4696 
4697     acb->common.cb(acb->common.opaque, acb->req.error);
4698 
4699     if (acb->done) {
4700         *acb->done = true;
4701     }
4702 
4703     qemu_bh_delete(acb->bh);
4704     qemu_aio_release(acb);
4705 }
4706 
4707 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
4708 static void coroutine_fn bdrv_co_do_rw(void *opaque)
4709 {
4710     BlockDriverAIOCBCoroutine *acb = opaque;
4711     BlockDriverState *bs = acb->common.bs;
4712 
4713     if (!acb->is_write) {
4714         acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
4715             acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4716     } else {
4717         acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
4718             acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4719     }
4720 
4721     acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
4722     qemu_bh_schedule(acb->bh);
4723 }
4724 
4725 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
4726                                                int64_t sector_num,
4727                                                QEMUIOVector *qiov,
4728                                                int nb_sectors,
4729                                                BdrvRequestFlags flags,
4730                                                BlockDriverCompletionFunc *cb,
4731                                                void *opaque,
4732                                                bool is_write)
4733 {
4734     Coroutine *co;
4735     BlockDriverAIOCBCoroutine *acb;
4736 
4737     acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4738     acb->req.sector = sector_num;
4739     acb->req.nb_sectors = nb_sectors;
4740     acb->req.qiov = qiov;
4741     acb->req.flags = flags;
4742     acb->is_write = is_write;
4743     acb->done = NULL;
4744 
4745     co = qemu_coroutine_create(bdrv_co_do_rw);
4746     qemu_coroutine_enter(co, acb);
4747 
4748     return &acb->common;
4749 }
4750 
4751 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
4752 {
4753     BlockDriverAIOCBCoroutine *acb = opaque;
4754     BlockDriverState *bs = acb->common.bs;
4755 
4756     acb->req.error = bdrv_co_flush(bs);
4757     acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
4758     qemu_bh_schedule(acb->bh);
4759 }
4760 
4761 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
4762         BlockDriverCompletionFunc *cb, void *opaque)
4763 {
4764     trace_bdrv_aio_flush(bs, opaque);
4765 
4766     Coroutine *co;
4767     BlockDriverAIOCBCoroutine *acb;
4768 
4769     acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4770     acb->done = NULL;
4771 
4772     co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
4773     qemu_coroutine_enter(co, acb);
4774 
4775     return &acb->common;
4776 }
4777 
4778 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
4779 {
4780     BlockDriverAIOCBCoroutine *acb = opaque;
4781     BlockDriverState *bs = acb->common.bs;
4782 
4783     acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
4784     acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
4785     qemu_bh_schedule(acb->bh);
4786 }
4787 
4788 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
4789         int64_t sector_num, int nb_sectors,
4790         BlockDriverCompletionFunc *cb, void *opaque)
4791 {
4792     Coroutine *co;
4793     BlockDriverAIOCBCoroutine *acb;
4794 
4795     trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
4796 
4797     acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4798     acb->req.sector = sector_num;
4799     acb->req.nb_sectors = nb_sectors;
4800     acb->done = NULL;
4801     co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
4802     qemu_coroutine_enter(co, acb);
4803 
4804     return &acb->common;
4805 }
4806 
4807 void bdrv_init(void)
4808 {
4809     module_call_init(MODULE_INIT_BLOCK);
4810 }
4811 
4812 void bdrv_init_with_whitelist(void)
4813 {
4814     use_bdrv_whitelist = 1;
4815     bdrv_init();
4816 }
4817 
4818 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
4819                    BlockDriverCompletionFunc *cb, void *opaque)
4820 {
4821     BlockDriverAIOCB *acb;
4822 
4823     acb = g_slice_alloc(aiocb_info->aiocb_size);
4824     acb->aiocb_info = aiocb_info;
4825     acb->bs = bs;
4826     acb->cb = cb;
4827     acb->opaque = opaque;
4828     return acb;
4829 }
4830 
4831 void qemu_aio_release(void *p)
4832 {
4833     BlockDriverAIOCB *acb = p;
4834     g_slice_free1(acb->aiocb_info->aiocb_size, acb);
4835 }
4836 
4837 /**************************************************************/
4838 /* Coroutine block device emulation */
4839 
4840 typedef struct CoroutineIOCompletion {
4841     Coroutine *coroutine;
4842     int ret;
4843 } CoroutineIOCompletion;
4844 
4845 static void bdrv_co_io_em_complete(void *opaque, int ret)
4846 {
4847     CoroutineIOCompletion *co = opaque;
4848 
4849     co->ret = ret;
4850     qemu_coroutine_enter(co->coroutine, NULL);
4851 }
4852 
4853 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
4854                                       int nb_sectors, QEMUIOVector *iov,
4855                                       bool is_write)
4856 {
4857     CoroutineIOCompletion co = {
4858         .coroutine = qemu_coroutine_self(),
4859     };
4860     BlockDriverAIOCB *acb;
4861 
4862     if (is_write) {
4863         acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
4864                                        bdrv_co_io_em_complete, &co);
4865     } else {
4866         acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
4867                                       bdrv_co_io_em_complete, &co);
4868     }
4869 
4870     trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
4871     if (!acb) {
4872         return -EIO;
4873     }
4874     qemu_coroutine_yield();
4875 
4876     return co.ret;
4877 }
4878 
4879 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
4880                                          int64_t sector_num, int nb_sectors,
4881                                          QEMUIOVector *iov)
4882 {
4883     return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
4884 }
4885 
4886 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
4887                                          int64_t sector_num, int nb_sectors,
4888                                          QEMUIOVector *iov)
4889 {
4890     return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
4891 }
4892 
4893 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
4894 {
4895     RwCo *rwco = opaque;
4896 
4897     rwco->ret = bdrv_co_flush(rwco->bs);
4898 }
4899 
4900 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
4901 {
4902     int ret;
4903 
4904     if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
4905         return 0;
4906     }
4907 
4908     /* Write back cached data to the OS even with cache=unsafe */
4909     BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
4910     if (bs->drv->bdrv_co_flush_to_os) {
4911         ret = bs->drv->bdrv_co_flush_to_os(bs);
4912         if (ret < 0) {
4913             return ret;
4914         }
4915     }
4916 
4917     /* But don't actually force it to the disk with cache=unsafe */
4918     if (bs->open_flags & BDRV_O_NO_FLUSH) {
4919         goto flush_parent;
4920     }
4921 
4922     BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
4923     if (bs->drv->bdrv_co_flush_to_disk) {
4924         ret = bs->drv->bdrv_co_flush_to_disk(bs);
4925     } else if (bs->drv->bdrv_aio_flush) {
4926         BlockDriverAIOCB *acb;
4927         CoroutineIOCompletion co = {
4928             .coroutine = qemu_coroutine_self(),
4929         };
4930 
4931         acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
4932         if (acb == NULL) {
4933             ret = -EIO;
4934         } else {
4935             qemu_coroutine_yield();
4936             ret = co.ret;
4937         }
4938     } else {
4939         /*
4940          * Some block drivers always operate in either writethrough or unsafe
4941          * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4942          * know how the server works (because the behaviour is hardcoded or
4943          * depends on server-side configuration), so we can't ensure that
4944          * everything is safe on disk. Returning an error doesn't work because
4945          * that would break guests even if the server operates in writethrough
4946          * mode.
4947          *
4948          * Let's hope the user knows what he's doing.
4949          */
4950         ret = 0;
4951     }
4952     if (ret < 0) {
4953         return ret;
4954     }
4955 
4956     /* Now flush the underlying protocol.  It will also have BDRV_O_NO_FLUSH
4957      * in the case of cache=unsafe, so there are no useless flushes.
4958      */
4959 flush_parent:
4960     return bdrv_co_flush(bs->file);
4961 }
4962 
4963 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4964 {
4965     Error *local_err = NULL;
4966     int ret;
4967 
4968     if (!bs->drv)  {
4969         return;
4970     }
4971 
4972     if (bs->drv->bdrv_invalidate_cache) {
4973         bs->drv->bdrv_invalidate_cache(bs, &local_err);
4974     } else if (bs->file) {
4975         bdrv_invalidate_cache(bs->file, &local_err);
4976     }
4977     if (local_err) {
4978         error_propagate(errp, local_err);
4979         return;
4980     }
4981 
4982     ret = refresh_total_sectors(bs, bs->total_sectors);
4983     if (ret < 0) {
4984         error_setg_errno(errp, -ret, "Could not refresh total sector count");
4985         return;
4986     }
4987 }
4988 
4989 void bdrv_invalidate_cache_all(Error **errp)
4990 {
4991     BlockDriverState *bs;
4992     Error *local_err = NULL;
4993 
4994     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
4995         AioContext *aio_context = bdrv_get_aio_context(bs);
4996 
4997         aio_context_acquire(aio_context);
4998         bdrv_invalidate_cache(bs, &local_err);
4999         aio_context_release(aio_context);
5000         if (local_err) {
5001             error_propagate(errp, local_err);
5002             return;
5003         }
5004     }
5005 }
5006 
5007 void bdrv_clear_incoming_migration_all(void)
5008 {
5009     BlockDriverState *bs;
5010 
5011     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
5012         AioContext *aio_context = bdrv_get_aio_context(bs);
5013 
5014         aio_context_acquire(aio_context);
5015         bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
5016         aio_context_release(aio_context);
5017     }
5018 }
5019 
5020 int bdrv_flush(BlockDriverState *bs)
5021 {
5022     Coroutine *co;
5023     RwCo rwco = {
5024         .bs = bs,
5025         .ret = NOT_DONE,
5026     };
5027 
5028     if (qemu_in_coroutine()) {
5029         /* Fast-path if already in coroutine context */
5030         bdrv_flush_co_entry(&rwco);
5031     } else {
5032         AioContext *aio_context = bdrv_get_aio_context(bs);
5033 
5034         co = qemu_coroutine_create(bdrv_flush_co_entry);
5035         qemu_coroutine_enter(co, &rwco);
5036         while (rwco.ret == NOT_DONE) {
5037             aio_poll(aio_context, true);
5038         }
5039     }
5040 
5041     return rwco.ret;
5042 }
5043 
5044 typedef struct DiscardCo {
5045     BlockDriverState *bs;
5046     int64_t sector_num;
5047     int nb_sectors;
5048     int ret;
5049 } DiscardCo;
5050 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
5051 {
5052     DiscardCo *rwco = opaque;
5053 
5054     rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
5055 }
5056 
5057 /* if no limit is specified in the BlockLimits use a default
5058  * of 32768 512-byte sectors (16 MiB) per request.
5059  */
5060 #define MAX_DISCARD_DEFAULT 32768
5061 
5062 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
5063                                  int nb_sectors)
5064 {
5065     int max_discard;
5066 
5067     if (!bs->drv) {
5068         return -ENOMEDIUM;
5069     } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
5070         return -EIO;
5071     } else if (bs->read_only) {
5072         return -EROFS;
5073     }
5074 
5075     bdrv_reset_dirty(bs, sector_num, nb_sectors);
5076 
5077     /* Do nothing if disabled.  */
5078     if (!(bs->open_flags & BDRV_O_UNMAP)) {
5079         return 0;
5080     }
5081 
5082     if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) {
5083         return 0;
5084     }
5085 
5086     max_discard = bs->bl.max_discard ?  bs->bl.max_discard : MAX_DISCARD_DEFAULT;
5087     while (nb_sectors > 0) {
5088         int ret;
5089         int num = nb_sectors;
5090 
5091         /* align request */
5092         if (bs->bl.discard_alignment &&
5093             num >= bs->bl.discard_alignment &&
5094             sector_num % bs->bl.discard_alignment) {
5095             if (num > bs->bl.discard_alignment) {
5096                 num = bs->bl.discard_alignment;
5097             }
5098             num -= sector_num % bs->bl.discard_alignment;
5099         }
5100 
5101         /* limit request size */
5102         if (num > max_discard) {
5103             num = max_discard;
5104         }
5105 
5106         if (bs->drv->bdrv_co_discard) {
5107             ret = bs->drv->bdrv_co_discard(bs, sector_num, num);
5108         } else {
5109             BlockDriverAIOCB *acb;
5110             CoroutineIOCompletion co = {
5111                 .coroutine = qemu_coroutine_self(),
5112             };
5113 
5114             acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
5115                                             bdrv_co_io_em_complete, &co);
5116             if (acb == NULL) {
5117                 return -EIO;
5118             } else {
5119                 qemu_coroutine_yield();
5120                 ret = co.ret;
5121             }
5122         }
5123         if (ret && ret != -ENOTSUP) {
5124             return ret;
5125         }
5126 
5127         sector_num += num;
5128         nb_sectors -= num;
5129     }
5130     return 0;
5131 }
5132 
5133 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
5134 {
5135     Coroutine *co;
5136     DiscardCo rwco = {
5137         .bs = bs,
5138         .sector_num = sector_num,
5139         .nb_sectors = nb_sectors,
5140         .ret = NOT_DONE,
5141     };
5142 
5143     if (qemu_in_coroutine()) {
5144         /* Fast-path if already in coroutine context */
5145         bdrv_discard_co_entry(&rwco);
5146     } else {
5147         AioContext *aio_context = bdrv_get_aio_context(bs);
5148 
5149         co = qemu_coroutine_create(bdrv_discard_co_entry);
5150         qemu_coroutine_enter(co, &rwco);
5151         while (rwco.ret == NOT_DONE) {
5152             aio_poll(aio_context, true);
5153         }
5154     }
5155 
5156     return rwco.ret;
5157 }
5158 
5159 /**************************************************************/
5160 /* removable device support */
5161 
5162 /**
5163  * Return TRUE if the media is present
5164  */
5165 int bdrv_is_inserted(BlockDriverState *bs)
5166 {
5167     BlockDriver *drv = bs->drv;
5168 
5169     if (!drv)
5170         return 0;
5171     if (!drv->bdrv_is_inserted)
5172         return 1;
5173     return drv->bdrv_is_inserted(bs);
5174 }
5175 
5176 /**
5177  * Return whether the media changed since the last call to this
5178  * function, or -ENOTSUP if we don't know.  Most drivers don't know.
5179  */
5180 int bdrv_media_changed(BlockDriverState *bs)
5181 {
5182     BlockDriver *drv = bs->drv;
5183 
5184     if (drv && drv->bdrv_media_changed) {
5185         return drv->bdrv_media_changed(bs);
5186     }
5187     return -ENOTSUP;
5188 }
5189 
5190 /**
5191  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5192  */
5193 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
5194 {
5195     BlockDriver *drv = bs->drv;
5196 
5197     if (drv && drv->bdrv_eject) {
5198         drv->bdrv_eject(bs, eject_flag);
5199     }
5200 
5201     if (bs->device_name[0] != '\0') {
5202         bdrv_emit_qmp_eject_event(bs, eject_flag);
5203     }
5204 }
5205 
5206 /**
5207  * Lock or unlock the media (if it is locked, the user won't be able
5208  * to eject it manually).
5209  */
5210 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
5211 {
5212     BlockDriver *drv = bs->drv;
5213 
5214     trace_bdrv_lock_medium(bs, locked);
5215 
5216     if (drv && drv->bdrv_lock_medium) {
5217         drv->bdrv_lock_medium(bs, locked);
5218     }
5219 }
5220 
5221 /* needed for generic scsi interface */
5222 
5223 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
5224 {
5225     BlockDriver *drv = bs->drv;
5226 
5227     if (drv && drv->bdrv_ioctl)
5228         return drv->bdrv_ioctl(bs, req, buf);
5229     return -ENOTSUP;
5230 }
5231 
5232 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
5233         unsigned long int req, void *buf,
5234         BlockDriverCompletionFunc *cb, void *opaque)
5235 {
5236     BlockDriver *drv = bs->drv;
5237 
5238     if (drv && drv->bdrv_aio_ioctl)
5239         return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
5240     return NULL;
5241 }
5242 
5243 void bdrv_set_guest_block_size(BlockDriverState *bs, int align)
5244 {
5245     bs->guest_block_size = align;
5246 }
5247 
5248 void *qemu_blockalign(BlockDriverState *bs, size_t size)
5249 {
5250     return qemu_memalign(bdrv_opt_mem_align(bs), size);
5251 }
5252 
5253 /*
5254  * Check if all memory in this vector is sector aligned.
5255  */
5256 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
5257 {
5258     int i;
5259     size_t alignment = bdrv_opt_mem_align(bs);
5260 
5261     for (i = 0; i < qiov->niov; i++) {
5262         if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
5263             return false;
5264         }
5265         if (qiov->iov[i].iov_len % alignment) {
5266             return false;
5267         }
5268     }
5269 
5270     return true;
5271 }
5272 
5273 BdrvDirtyBitmap *bdrv_create_dirty_bitmap(BlockDriverState *bs, int granularity,
5274                                           Error **errp)
5275 {
5276     int64_t bitmap_size;
5277     BdrvDirtyBitmap *bitmap;
5278 
5279     assert((granularity & (granularity - 1)) == 0);
5280 
5281     granularity >>= BDRV_SECTOR_BITS;
5282     assert(granularity);
5283     bitmap_size = bdrv_getlength(bs);
5284     if (bitmap_size < 0) {
5285         error_setg_errno(errp, -bitmap_size, "could not get length of device");
5286         errno = -bitmap_size;
5287         return NULL;
5288     }
5289     bitmap_size >>= BDRV_SECTOR_BITS;
5290     bitmap = g_malloc0(sizeof(BdrvDirtyBitmap));
5291     bitmap->bitmap = hbitmap_alloc(bitmap_size, ffs(granularity) - 1);
5292     QLIST_INSERT_HEAD(&bs->dirty_bitmaps, bitmap, list);
5293     return bitmap;
5294 }
5295 
5296 void bdrv_release_dirty_bitmap(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
5297 {
5298     BdrvDirtyBitmap *bm, *next;
5299     QLIST_FOREACH_SAFE(bm, &bs->dirty_bitmaps, list, next) {
5300         if (bm == bitmap) {
5301             QLIST_REMOVE(bitmap, list);
5302             hbitmap_free(bitmap->bitmap);
5303             g_free(bitmap);
5304             return;
5305         }
5306     }
5307 }
5308 
5309 BlockDirtyInfoList *bdrv_query_dirty_bitmaps(BlockDriverState *bs)
5310 {
5311     BdrvDirtyBitmap *bm;
5312     BlockDirtyInfoList *list = NULL;
5313     BlockDirtyInfoList **plist = &list;
5314 
5315     QLIST_FOREACH(bm, &bs->dirty_bitmaps, list) {
5316         BlockDirtyInfo *info = g_malloc0(sizeof(BlockDirtyInfo));
5317         BlockDirtyInfoList *entry = g_malloc0(sizeof(BlockDirtyInfoList));
5318         info->count = bdrv_get_dirty_count(bs, bm);
5319         info->granularity =
5320             ((int64_t) BDRV_SECTOR_SIZE << hbitmap_granularity(bm->bitmap));
5321         entry->value = info;
5322         *plist = entry;
5323         plist = &entry->next;
5324     }
5325 
5326     return list;
5327 }
5328 
5329 int bdrv_get_dirty(BlockDriverState *bs, BdrvDirtyBitmap *bitmap, int64_t sector)
5330 {
5331     if (bitmap) {
5332         return hbitmap_get(bitmap->bitmap, sector);
5333     } else {
5334         return 0;
5335     }
5336 }
5337 
5338 void bdrv_dirty_iter_init(BlockDriverState *bs,
5339                           BdrvDirtyBitmap *bitmap, HBitmapIter *hbi)
5340 {
5341     hbitmap_iter_init(hbi, bitmap->bitmap, 0);
5342 }
5343 
5344 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector,
5345                     int nr_sectors)
5346 {
5347     BdrvDirtyBitmap *bitmap;
5348     QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
5349         hbitmap_set(bitmap->bitmap, cur_sector, nr_sectors);
5350     }
5351 }
5352 
5353 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector, int nr_sectors)
5354 {
5355     BdrvDirtyBitmap *bitmap;
5356     QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
5357         hbitmap_reset(bitmap->bitmap, cur_sector, nr_sectors);
5358     }
5359 }
5360 
5361 int64_t bdrv_get_dirty_count(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
5362 {
5363     return hbitmap_count(bitmap->bitmap);
5364 }
5365 
5366 /* Get a reference to bs */
5367 void bdrv_ref(BlockDriverState *bs)
5368 {
5369     bs->refcnt++;
5370 }
5371 
5372 /* Release a previously grabbed reference to bs.
5373  * If after releasing, reference count is zero, the BlockDriverState is
5374  * deleted. */
5375 void bdrv_unref(BlockDriverState *bs)
5376 {
5377     assert(bs->refcnt > 0);
5378     if (--bs->refcnt == 0) {
5379         bdrv_delete(bs);
5380     }
5381 }
5382 
5383 struct BdrvOpBlocker {
5384     Error *reason;
5385     QLIST_ENTRY(BdrvOpBlocker) list;
5386 };
5387 
5388 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
5389 {
5390     BdrvOpBlocker *blocker;
5391     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5392     if (!QLIST_EMPTY(&bs->op_blockers[op])) {
5393         blocker = QLIST_FIRST(&bs->op_blockers[op]);
5394         if (errp) {
5395             error_setg(errp, "Device '%s' is busy: %s",
5396                        bs->device_name, error_get_pretty(blocker->reason));
5397         }
5398         return true;
5399     }
5400     return false;
5401 }
5402 
5403 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
5404 {
5405     BdrvOpBlocker *blocker;
5406     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5407 
5408     blocker = g_malloc0(sizeof(BdrvOpBlocker));
5409     blocker->reason = reason;
5410     QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
5411 }
5412 
5413 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
5414 {
5415     BdrvOpBlocker *blocker, *next;
5416     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5417     QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
5418         if (blocker->reason == reason) {
5419             QLIST_REMOVE(blocker, list);
5420             g_free(blocker);
5421         }
5422     }
5423 }
5424 
5425 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
5426 {
5427     int i;
5428     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5429         bdrv_op_block(bs, i, reason);
5430     }
5431 }
5432 
5433 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
5434 {
5435     int i;
5436     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5437         bdrv_op_unblock(bs, i, reason);
5438     }
5439 }
5440 
5441 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
5442 {
5443     int i;
5444 
5445     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5446         if (!QLIST_EMPTY(&bs->op_blockers[i])) {
5447             return false;
5448         }
5449     }
5450     return true;
5451 }
5452 
5453 void bdrv_iostatus_enable(BlockDriverState *bs)
5454 {
5455     bs->iostatus_enabled = true;
5456     bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
5457 }
5458 
5459 /* The I/O status is only enabled if the drive explicitly
5460  * enables it _and_ the VM is configured to stop on errors */
5461 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
5462 {
5463     return (bs->iostatus_enabled &&
5464            (bs->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
5465             bs->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
5466             bs->on_read_error == BLOCKDEV_ON_ERROR_STOP));
5467 }
5468 
5469 void bdrv_iostatus_disable(BlockDriverState *bs)
5470 {
5471     bs->iostatus_enabled = false;
5472 }
5473 
5474 void bdrv_iostatus_reset(BlockDriverState *bs)
5475 {
5476     if (bdrv_iostatus_is_enabled(bs)) {
5477         bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
5478         if (bs->job) {
5479             block_job_iostatus_reset(bs->job);
5480         }
5481     }
5482 }
5483 
5484 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
5485 {
5486     assert(bdrv_iostatus_is_enabled(bs));
5487     if (bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
5488         bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
5489                                          BLOCK_DEVICE_IO_STATUS_FAILED;
5490     }
5491 }
5492 
5493 void
5494 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
5495         enum BlockAcctType type)
5496 {
5497     assert(type < BDRV_MAX_IOTYPE);
5498 
5499     cookie->bytes = bytes;
5500     cookie->start_time_ns = get_clock();
5501     cookie->type = type;
5502 }
5503 
5504 void
5505 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
5506 {
5507     assert(cookie->type < BDRV_MAX_IOTYPE);
5508 
5509     bs->nr_bytes[cookie->type] += cookie->bytes;
5510     bs->nr_ops[cookie->type]++;
5511     bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
5512 }
5513 
5514 void bdrv_img_create(const char *filename, const char *fmt,
5515                      const char *base_filename, const char *base_fmt,
5516                      char *options, uint64_t img_size, int flags,
5517                      Error **errp, bool quiet)
5518 {
5519     QemuOptsList *create_opts = NULL;
5520     QemuOpts *opts = NULL;
5521     const char *backing_fmt, *backing_file;
5522     int64_t size;
5523     BlockDriver *drv, *proto_drv;
5524     BlockDriver *backing_drv = NULL;
5525     Error *local_err = NULL;
5526     int ret = 0;
5527 
5528     /* Find driver and parse its options */
5529     drv = bdrv_find_format(fmt);
5530     if (!drv) {
5531         error_setg(errp, "Unknown file format '%s'", fmt);
5532         return;
5533     }
5534 
5535     proto_drv = bdrv_find_protocol(filename, true);
5536     if (!proto_drv) {
5537         error_setg(errp, "Unknown protocol '%s'", filename);
5538         return;
5539     }
5540 
5541     create_opts = qemu_opts_append(create_opts, drv->create_opts);
5542     create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
5543 
5544     /* Create parameter list with default values */
5545     opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
5546     qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size);
5547 
5548     /* Parse -o options */
5549     if (options) {
5550         if (qemu_opts_do_parse(opts, options, NULL) != 0) {
5551             error_setg(errp, "Invalid options for file format '%s'", fmt);
5552             goto out;
5553         }
5554     }
5555 
5556     if (base_filename) {
5557         if (qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename)) {
5558             error_setg(errp, "Backing file not supported for file format '%s'",
5559                        fmt);
5560             goto out;
5561         }
5562     }
5563 
5564     if (base_fmt) {
5565         if (qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt)) {
5566             error_setg(errp, "Backing file format not supported for file "
5567                              "format '%s'", fmt);
5568             goto out;
5569         }
5570     }
5571 
5572     backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5573     if (backing_file) {
5574         if (!strcmp(filename, backing_file)) {
5575             error_setg(errp, "Error: Trying to create an image with the "
5576                              "same filename as the backing file");
5577             goto out;
5578         }
5579     }
5580 
5581     backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5582     if (backing_fmt) {
5583         backing_drv = bdrv_find_format(backing_fmt);
5584         if (!backing_drv) {
5585             error_setg(errp, "Unknown backing file format '%s'",
5586                        backing_fmt);
5587             goto out;
5588         }
5589     }
5590 
5591     // The size for the image must always be specified, with one exception:
5592     // If we are using a backing file, we can obtain the size from there
5593     size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
5594     if (size == -1) {
5595         if (backing_file) {
5596             BlockDriverState *bs;
5597             uint64_t size;
5598             char buf[32];
5599             int back_flags;
5600 
5601             /* backing files always opened read-only */
5602             back_flags =
5603                 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
5604 
5605             bs = NULL;
5606             ret = bdrv_open(&bs, backing_file, NULL, NULL, back_flags,
5607                             backing_drv, &local_err);
5608             if (ret < 0) {
5609                 error_setg_errno(errp, -ret, "Could not open '%s': %s",
5610                                  backing_file,
5611                                  error_get_pretty(local_err));
5612                 error_free(local_err);
5613                 local_err = NULL;
5614                 goto out;
5615             }
5616             bdrv_get_geometry(bs, &size);
5617             size *= 512;
5618 
5619             snprintf(buf, sizeof(buf), "%" PRId64, size);
5620             qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size);
5621 
5622             bdrv_unref(bs);
5623         } else {
5624             error_setg(errp, "Image creation needs a size parameter");
5625             goto out;
5626         }
5627     }
5628 
5629     if (!quiet) {
5630         printf("Formatting '%s', fmt=%s ", filename, fmt);
5631         qemu_opts_print(opts);
5632         puts("");
5633     }
5634 
5635     ret = bdrv_create(drv, filename, opts, &local_err);
5636 
5637     if (ret == -EFBIG) {
5638         /* This is generally a better message than whatever the driver would
5639          * deliver (especially because of the cluster_size_hint), since that
5640          * is most probably not much different from "image too large". */
5641         const char *cluster_size_hint = "";
5642         if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
5643             cluster_size_hint = " (try using a larger cluster size)";
5644         }
5645         error_setg(errp, "The image size is too large for file format '%s'"
5646                    "%s", fmt, cluster_size_hint);
5647         error_free(local_err);
5648         local_err = NULL;
5649     }
5650 
5651 out:
5652     qemu_opts_del(opts);
5653     qemu_opts_free(create_opts);
5654     if (local_err) {
5655         error_propagate(errp, local_err);
5656     }
5657 }
5658 
5659 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
5660 {
5661     return bs->aio_context;
5662 }
5663 
5664 void bdrv_detach_aio_context(BlockDriverState *bs)
5665 {
5666     if (!bs->drv) {
5667         return;
5668     }
5669 
5670     if (bs->io_limits_enabled) {
5671         throttle_detach_aio_context(&bs->throttle_state);
5672     }
5673     if (bs->drv->bdrv_detach_aio_context) {
5674         bs->drv->bdrv_detach_aio_context(bs);
5675     }
5676     if (bs->file) {
5677         bdrv_detach_aio_context(bs->file);
5678     }
5679     if (bs->backing_hd) {
5680         bdrv_detach_aio_context(bs->backing_hd);
5681     }
5682 
5683     bs->aio_context = NULL;
5684 }
5685 
5686 void bdrv_attach_aio_context(BlockDriverState *bs,
5687                              AioContext *new_context)
5688 {
5689     if (!bs->drv) {
5690         return;
5691     }
5692 
5693     bs->aio_context = new_context;
5694 
5695     if (bs->backing_hd) {
5696         bdrv_attach_aio_context(bs->backing_hd, new_context);
5697     }
5698     if (bs->file) {
5699         bdrv_attach_aio_context(bs->file, new_context);
5700     }
5701     if (bs->drv->bdrv_attach_aio_context) {
5702         bs->drv->bdrv_attach_aio_context(bs, new_context);
5703     }
5704     if (bs->io_limits_enabled) {
5705         throttle_attach_aio_context(&bs->throttle_state, new_context);
5706     }
5707 }
5708 
5709 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
5710 {
5711     bdrv_drain_all(); /* ensure there are no in-flight requests */
5712 
5713     bdrv_detach_aio_context(bs);
5714 
5715     /* This function executes in the old AioContext so acquire the new one in
5716      * case it runs in a different thread.
5717      */
5718     aio_context_acquire(new_context);
5719     bdrv_attach_aio_context(bs, new_context);
5720     aio_context_release(new_context);
5721 }
5722 
5723 void bdrv_add_before_write_notifier(BlockDriverState *bs,
5724                                     NotifierWithReturn *notifier)
5725 {
5726     notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
5727 }
5728 
5729 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts)
5730 {
5731     if (!bs->drv->bdrv_amend_options) {
5732         return -ENOTSUP;
5733     }
5734     return bs->drv->bdrv_amend_options(bs, opts);
5735 }
5736 
5737 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5738  * of block filter and by bdrv_is_first_non_filter.
5739  * It is used to test if the given bs is the candidate or recurse more in the
5740  * node graph.
5741  */
5742 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5743                                       BlockDriverState *candidate)
5744 {
5745     /* return false if basic checks fails */
5746     if (!bs || !bs->drv) {
5747         return false;
5748     }
5749 
5750     /* the code reached a non block filter driver -> check if the bs is
5751      * the same as the candidate. It's the recursion termination condition.
5752      */
5753     if (!bs->drv->is_filter) {
5754         return bs == candidate;
5755     }
5756     /* Down this path the driver is a block filter driver */
5757 
5758     /* If the block filter recursion method is defined use it to recurse down
5759      * the node graph.
5760      */
5761     if (bs->drv->bdrv_recurse_is_first_non_filter) {
5762         return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5763     }
5764 
5765     /* the driver is a block filter but don't allow to recurse -> return false
5766      */
5767     return false;
5768 }
5769 
5770 /* This function checks if the candidate is the first non filter bs down it's
5771  * bs chain. Since we don't have pointers to parents it explore all bs chains
5772  * from the top. Some filters can choose not to pass down the recursion.
5773  */
5774 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5775 {
5776     BlockDriverState *bs;
5777 
5778     /* walk down the bs forest recursively */
5779     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
5780         bool perm;
5781 
5782         /* try to recurse in this top level bs */
5783         perm = bdrv_recurse_is_first_non_filter(bs, candidate);
5784 
5785         /* candidate is the first non filter */
5786         if (perm) {
5787             return true;
5788         }
5789     }
5790 
5791     return false;
5792 }
5793