xref: /openbmc/qemu/block.c (revision 462f5bcf69d3cae4a93ad55bee59516c15236a55)
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 "block/block_int.h"
28 #include "block/blockjob.h"
29 #include "qemu/module.h"
30 #include "qapi/qmp/qjson.h"
31 #include "sysemu/sysemu.h"
32 #include "qemu/notify.h"
33 #include "block/coroutine.h"
34 #include "block/qapi.h"
35 #include "qmp-commands.h"
36 #include "qemu/timer.h"
37 #include "qapi-event.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  * Fills in default options for opening images and converts the legacy
1010  * filename/flags pair to option QDict entries.
1011  */
1012 static int bdrv_fill_options(QDict **options, const char *filename, int flags,
1013                              Error **errp)
1014 {
1015     const char *drvname;
1016     bool protocol = flags & BDRV_O_PROTOCOL;
1017     bool parse_filename = false;
1018     Error *local_err = NULL;
1019     BlockDriver *drv;
1020 
1021     if (!protocol) {
1022         return 0;
1023     }
1024 
1025     /* Fetch the file name from the options QDict if necessary */
1026     if (filename) {
1027         if (!qdict_haskey(*options, "filename")) {
1028             qdict_put(*options, "filename", qstring_from_str(filename));
1029             parse_filename = true;
1030         } else {
1031             error_setg(errp, "Can't specify 'file' and 'filename' options at "
1032                              "the same time");
1033             return -EINVAL;
1034         }
1035     }
1036 
1037     /* Find the right block driver */
1038     filename = qdict_get_try_str(*options, "filename");
1039     drvname = qdict_get_try_str(*options, "driver");
1040 
1041     if (!drvname) {
1042         if (filename) {
1043             drv = bdrv_find_protocol(filename, parse_filename);
1044             if (!drv) {
1045                 error_setg(errp, "Unknown protocol");
1046                 return -EINVAL;
1047             }
1048 
1049             drvname = drv->format_name;
1050             qdict_put(*options, "driver", qstring_from_str(drvname));
1051         } else {
1052             error_setg(errp, "Must specify either driver or file");
1053             return -EINVAL;
1054         }
1055     }
1056 
1057     drv = bdrv_find_format(drvname);
1058     if (!drv) {
1059         error_setg(errp, "Unknown driver '%s'", drvname);
1060         return -ENOENT;
1061     }
1062 
1063     /* Driver-specific filename parsing */
1064     if (drv->bdrv_parse_filename && parse_filename) {
1065         drv->bdrv_parse_filename(filename, *options, &local_err);
1066         if (local_err) {
1067             error_propagate(errp, local_err);
1068             return -EINVAL;
1069         }
1070 
1071         if (!drv->bdrv_needs_filename) {
1072             qdict_del(*options, "filename");
1073         }
1074     }
1075 
1076     return 0;
1077 }
1078 
1079 /*
1080  * Opens a file using a protocol (file, host_device, nbd, ...)
1081  *
1082  * options is an indirect pointer to a QDict of options to pass to the block
1083  * drivers, or pointer to NULL for an empty set of options. If this function
1084  * takes ownership of the QDict reference, it will set *options to NULL;
1085  * otherwise, it will contain unused/unrecognized options after this function
1086  * returns. Then, the caller is responsible for freeing it. If it intends to
1087  * reuse the QDict, QINCREF() should be called beforehand.
1088  */
1089 static int bdrv_file_open(BlockDriverState *bs, QDict **options, int flags,
1090                           Error **errp)
1091 {
1092     BlockDriver *drv;
1093     const char *filename;
1094     const char *drvname;
1095     Error *local_err = NULL;
1096     int ret;
1097 
1098     filename = qdict_get_try_str(*options, "filename");
1099     drvname = qdict_get_str(*options, "driver");
1100 
1101     drv = bdrv_find_format(drvname);
1102     assert(drv);
1103     qdict_del(*options, "driver");
1104 
1105     /* Open the file */
1106     if (!drv->bdrv_file_open) {
1107         ret = bdrv_open(&bs, filename, NULL, *options, flags, drv, &local_err);
1108         *options = NULL;
1109     } else {
1110         ret = bdrv_open_common(bs, NULL, *options, flags, drv, &local_err);
1111     }
1112     if (ret < 0) {
1113         error_propagate(errp, local_err);
1114         goto fail;
1115     }
1116 
1117     bs->growable = 1;
1118     return 0;
1119 
1120 fail:
1121     return ret;
1122 }
1123 
1124 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd)
1125 {
1126 
1127     if (bs->backing_hd) {
1128         assert(bs->backing_blocker);
1129         bdrv_op_unblock_all(bs->backing_hd, bs->backing_blocker);
1130     } else if (backing_hd) {
1131         error_setg(&bs->backing_blocker,
1132                    "device is used as backing hd of '%s'",
1133                    bs->device_name);
1134     }
1135 
1136     bs->backing_hd = backing_hd;
1137     if (!backing_hd) {
1138         error_free(bs->backing_blocker);
1139         bs->backing_blocker = NULL;
1140         goto out;
1141     }
1142     bs->open_flags &= ~BDRV_O_NO_BACKING;
1143     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_hd->filename);
1144     pstrcpy(bs->backing_format, sizeof(bs->backing_format),
1145             backing_hd->drv ? backing_hd->drv->format_name : "");
1146 
1147     bdrv_op_block_all(bs->backing_hd, bs->backing_blocker);
1148     /* Otherwise we won't be able to commit due to check in bdrv_commit */
1149     bdrv_op_unblock(bs->backing_hd, BLOCK_OP_TYPE_COMMIT,
1150                     bs->backing_blocker);
1151 out:
1152     bdrv_refresh_limits(bs);
1153 }
1154 
1155 /*
1156  * Opens the backing file for a BlockDriverState if not yet open
1157  *
1158  * options is a QDict of options to pass to the block drivers, or NULL for an
1159  * empty set of options. The reference to the QDict is transferred to this
1160  * function (even on failure), so if the caller intends to reuse the dictionary,
1161  * it needs to use QINCREF() before calling bdrv_file_open.
1162  */
1163 int bdrv_open_backing_file(BlockDriverState *bs, QDict *options, Error **errp)
1164 {
1165     char *backing_filename = g_malloc0(PATH_MAX);
1166     int ret = 0;
1167     BlockDriver *back_drv = NULL;
1168     BlockDriverState *backing_hd;
1169     Error *local_err = NULL;
1170 
1171     if (bs->backing_hd != NULL) {
1172         QDECREF(options);
1173         goto free_exit;
1174     }
1175 
1176     /* NULL means an empty set of options */
1177     if (options == NULL) {
1178         options = qdict_new();
1179     }
1180 
1181     bs->open_flags &= ~BDRV_O_NO_BACKING;
1182     if (qdict_haskey(options, "file.filename")) {
1183         backing_filename[0] = '\0';
1184     } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
1185         QDECREF(options);
1186         goto free_exit;
1187     } else {
1188         bdrv_get_full_backing_filename(bs, backing_filename, PATH_MAX);
1189     }
1190 
1191     backing_hd = bdrv_new("", errp);
1192 
1193     if (bs->backing_format[0] != '\0') {
1194         back_drv = bdrv_find_format(bs->backing_format);
1195     }
1196 
1197     assert(bs->backing_hd == NULL);
1198     ret = bdrv_open(&backing_hd,
1199                     *backing_filename ? backing_filename : NULL, NULL, options,
1200                     bdrv_backing_flags(bs->open_flags), back_drv, &local_err);
1201     if (ret < 0) {
1202         bdrv_unref(backing_hd);
1203         backing_hd = NULL;
1204         bs->open_flags |= BDRV_O_NO_BACKING;
1205         error_setg(errp, "Could not open backing file: %s",
1206                    error_get_pretty(local_err));
1207         error_free(local_err);
1208         goto free_exit;
1209     }
1210     bdrv_set_backing_hd(bs, backing_hd);
1211 
1212 free_exit:
1213     g_free(backing_filename);
1214     return ret;
1215 }
1216 
1217 /*
1218  * Opens a disk image whose options are given as BlockdevRef in another block
1219  * device's options.
1220  *
1221  * If allow_none is true, no image will be opened if filename is false and no
1222  * BlockdevRef is given. *pbs will remain unchanged and 0 will be returned.
1223  *
1224  * bdrev_key specifies the key for the image's BlockdevRef in the options QDict.
1225  * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
1226  * itself, all options starting with "${bdref_key}." are considered part of the
1227  * BlockdevRef.
1228  *
1229  * The BlockdevRef will be removed from the options QDict.
1230  *
1231  * To conform with the behavior of bdrv_open(), *pbs has to be NULL.
1232  */
1233 int bdrv_open_image(BlockDriverState **pbs, const char *filename,
1234                     QDict *options, const char *bdref_key, int flags,
1235                     bool allow_none, Error **errp)
1236 {
1237     QDict *image_options;
1238     int ret;
1239     char *bdref_key_dot;
1240     const char *reference;
1241 
1242     assert(pbs);
1243     assert(*pbs == NULL);
1244 
1245     bdref_key_dot = g_strdup_printf("%s.", bdref_key);
1246     qdict_extract_subqdict(options, &image_options, bdref_key_dot);
1247     g_free(bdref_key_dot);
1248 
1249     reference = qdict_get_try_str(options, bdref_key);
1250     if (!filename && !reference && !qdict_size(image_options)) {
1251         if (allow_none) {
1252             ret = 0;
1253         } else {
1254             error_setg(errp, "A block device must be specified for \"%s\"",
1255                        bdref_key);
1256             ret = -EINVAL;
1257         }
1258         QDECREF(image_options);
1259         goto done;
1260     }
1261 
1262     ret = bdrv_open(pbs, filename, reference, image_options, flags, NULL, errp);
1263 
1264 done:
1265     qdict_del(options, bdref_key);
1266     return ret;
1267 }
1268 
1269 void bdrv_append_temp_snapshot(BlockDriverState *bs, int flags, Error **errp)
1270 {
1271     /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */
1272     char *tmp_filename = g_malloc0(PATH_MAX + 1);
1273     int64_t total_size;
1274     BlockDriver *bdrv_qcow2;
1275     QemuOpts *opts = NULL;
1276     QDict *snapshot_options;
1277     BlockDriverState *bs_snapshot;
1278     Error *local_err;
1279     int ret;
1280 
1281     /* if snapshot, we create a temporary backing file and open it
1282        instead of opening 'filename' directly */
1283 
1284     /* Get the required size from the image */
1285     total_size = bdrv_getlength(bs);
1286     if (total_size < 0) {
1287         error_setg_errno(errp, -total_size, "Could not get image size");
1288         goto out;
1289     }
1290     total_size &= BDRV_SECTOR_MASK;
1291 
1292     /* Create the temporary image */
1293     ret = get_tmp_filename(tmp_filename, PATH_MAX + 1);
1294     if (ret < 0) {
1295         error_setg_errno(errp, -ret, "Could not get temporary filename");
1296         goto out;
1297     }
1298 
1299     bdrv_qcow2 = bdrv_find_format("qcow2");
1300     opts = qemu_opts_create(bdrv_qcow2->create_opts, NULL, 0,
1301                             &error_abort);
1302     qemu_opt_set_number(opts, BLOCK_OPT_SIZE, total_size);
1303     ret = bdrv_create(bdrv_qcow2, tmp_filename, opts, &local_err);
1304     qemu_opts_del(opts);
1305     if (ret < 0) {
1306         error_setg_errno(errp, -ret, "Could not create temporary overlay "
1307                          "'%s': %s", tmp_filename,
1308                          error_get_pretty(local_err));
1309         error_free(local_err);
1310         goto out;
1311     }
1312 
1313     /* Prepare a new options QDict for the temporary file */
1314     snapshot_options = qdict_new();
1315     qdict_put(snapshot_options, "file.driver",
1316               qstring_from_str("file"));
1317     qdict_put(snapshot_options, "file.filename",
1318               qstring_from_str(tmp_filename));
1319 
1320     bs_snapshot = bdrv_new("", &error_abort);
1321 
1322     ret = bdrv_open(&bs_snapshot, NULL, NULL, snapshot_options,
1323                     flags, bdrv_qcow2, &local_err);
1324     if (ret < 0) {
1325         error_propagate(errp, local_err);
1326         goto out;
1327     }
1328 
1329     bdrv_append(bs_snapshot, bs);
1330 
1331 out:
1332     g_free(tmp_filename);
1333 }
1334 
1335 static QDict *parse_json_filename(const char *filename, Error **errp)
1336 {
1337     QObject *options_obj;
1338     QDict *options;
1339     int ret;
1340 
1341     ret = strstart(filename, "json:", &filename);
1342     assert(ret);
1343 
1344     options_obj = qobject_from_json(filename);
1345     if (!options_obj) {
1346         error_setg(errp, "Could not parse the JSON options");
1347         return NULL;
1348     }
1349 
1350     if (qobject_type(options_obj) != QTYPE_QDICT) {
1351         qobject_decref(options_obj);
1352         error_setg(errp, "Invalid JSON object given");
1353         return NULL;
1354     }
1355 
1356     options = qobject_to_qdict(options_obj);
1357     qdict_flatten(options);
1358 
1359     return options;
1360 }
1361 
1362 /*
1363  * Opens a disk image (raw, qcow2, vmdk, ...)
1364  *
1365  * options is a QDict of options to pass to the block drivers, or NULL for an
1366  * empty set of options. The reference to the QDict belongs to the block layer
1367  * after the call (even on failure), so if the caller intends to reuse the
1368  * dictionary, it needs to use QINCREF() before calling bdrv_open.
1369  *
1370  * If *pbs is NULL, a new BDS will be created with a pointer to it stored there.
1371  * If it is not NULL, the referenced BDS will be reused.
1372  *
1373  * The reference parameter may be used to specify an existing block device which
1374  * should be opened. If specified, neither options nor a filename may be given,
1375  * nor can an existing BDS be reused (that is, *pbs has to be NULL).
1376  */
1377 int bdrv_open(BlockDriverState **pbs, const char *filename,
1378               const char *reference, QDict *options, int flags,
1379               BlockDriver *drv, Error **errp)
1380 {
1381     int ret;
1382     BlockDriverState *file = NULL, *bs;
1383     const char *drvname;
1384     Error *local_err = NULL;
1385     int snapshot_flags = 0;
1386 
1387     assert(pbs);
1388 
1389     if (reference) {
1390         bool options_non_empty = options ? qdict_size(options) : false;
1391         QDECREF(options);
1392 
1393         if (*pbs) {
1394             error_setg(errp, "Cannot reuse an existing BDS when referencing "
1395                        "another block device");
1396             return -EINVAL;
1397         }
1398 
1399         if (filename || options_non_empty) {
1400             error_setg(errp, "Cannot reference an existing block device with "
1401                        "additional options or a new filename");
1402             return -EINVAL;
1403         }
1404 
1405         bs = bdrv_lookup_bs(reference, reference, errp);
1406         if (!bs) {
1407             return -ENODEV;
1408         }
1409         bdrv_ref(bs);
1410         *pbs = bs;
1411         return 0;
1412     }
1413 
1414     if (*pbs) {
1415         bs = *pbs;
1416     } else {
1417         bs = bdrv_new("", &error_abort);
1418     }
1419 
1420     /* NULL means an empty set of options */
1421     if (options == NULL) {
1422         options = qdict_new();
1423     }
1424 
1425     if (filename && g_str_has_prefix(filename, "json:")) {
1426         QDict *json_options = parse_json_filename(filename, &local_err);
1427         if (local_err) {
1428             ret = -EINVAL;
1429             goto fail;
1430         }
1431 
1432         /* Options given in the filename have lower priority than options
1433          * specified directly */
1434         qdict_join(options, json_options, false);
1435         QDECREF(json_options);
1436         filename = NULL;
1437     }
1438 
1439     ret = bdrv_fill_options(&options, filename, flags, &local_err);
1440     if (local_err) {
1441         goto fail;
1442     }
1443 
1444     bs->options = options;
1445     options = qdict_clone_shallow(options);
1446 
1447     if (flags & BDRV_O_PROTOCOL) {
1448         assert(!drv);
1449         ret = bdrv_file_open(bs, &options, flags & ~BDRV_O_PROTOCOL,
1450                              &local_err);
1451         if (!ret) {
1452             drv = bs->drv;
1453             goto done;
1454         } else if (bs->drv) {
1455             goto close_and_fail;
1456         } else {
1457             goto fail;
1458         }
1459     }
1460 
1461     /* Open image file without format layer */
1462     if (flags & BDRV_O_RDWR) {
1463         flags |= BDRV_O_ALLOW_RDWR;
1464     }
1465     if (flags & BDRV_O_SNAPSHOT) {
1466         snapshot_flags = bdrv_temp_snapshot_flags(flags);
1467         flags = bdrv_backing_flags(flags);
1468     }
1469 
1470     assert(file == NULL);
1471     ret = bdrv_open_image(&file, filename, options, "file",
1472                           bdrv_inherited_flags(flags),
1473                           true, &local_err);
1474     if (ret < 0) {
1475         goto fail;
1476     }
1477 
1478     /* Find the right image format driver */
1479     drvname = qdict_get_try_str(options, "driver");
1480     if (drvname) {
1481         drv = bdrv_find_format(drvname);
1482         qdict_del(options, "driver");
1483         if (!drv) {
1484             error_setg(errp, "Invalid driver: '%s'", drvname);
1485             ret = -EINVAL;
1486             goto fail;
1487         }
1488     }
1489 
1490     if (!drv) {
1491         if (file) {
1492             ret = find_image_format(file, filename, &drv, &local_err);
1493         } else {
1494             error_setg(errp, "Must specify either driver or file");
1495             ret = -EINVAL;
1496             goto fail;
1497         }
1498     }
1499 
1500     if (!drv) {
1501         goto fail;
1502     }
1503 
1504     /* Open the image */
1505     ret = bdrv_open_common(bs, file, options, flags, drv, &local_err);
1506     if (ret < 0) {
1507         goto fail;
1508     }
1509 
1510     if (file && (bs->file != file)) {
1511         bdrv_unref(file);
1512         file = NULL;
1513     }
1514 
1515     /* If there is a backing file, use it */
1516     if ((flags & BDRV_O_NO_BACKING) == 0) {
1517         QDict *backing_options;
1518 
1519         qdict_extract_subqdict(options, &backing_options, "backing.");
1520         ret = bdrv_open_backing_file(bs, backing_options, &local_err);
1521         if (ret < 0) {
1522             goto close_and_fail;
1523         }
1524     }
1525 
1526     /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
1527      * temporary snapshot afterwards. */
1528     if (snapshot_flags) {
1529         bdrv_append_temp_snapshot(bs, snapshot_flags, &local_err);
1530         if (local_err) {
1531             error_propagate(errp, local_err);
1532             goto close_and_fail;
1533         }
1534     }
1535 
1536 
1537 done:
1538     /* Check if any unknown options were used */
1539     if (options && (qdict_size(options) != 0)) {
1540         const QDictEntry *entry = qdict_first(options);
1541         if (flags & BDRV_O_PROTOCOL) {
1542             error_setg(errp, "Block protocol '%s' doesn't support the option "
1543                        "'%s'", drv->format_name, entry->key);
1544         } else {
1545             error_setg(errp, "Block format '%s' used by device '%s' doesn't "
1546                        "support the option '%s'", drv->format_name,
1547                        bs->device_name, entry->key);
1548         }
1549 
1550         ret = -EINVAL;
1551         goto close_and_fail;
1552     }
1553 
1554     if (!bdrv_key_required(bs)) {
1555         bdrv_dev_change_media_cb(bs, true);
1556     } else if (!runstate_check(RUN_STATE_PRELAUNCH)
1557                && !runstate_check(RUN_STATE_INMIGRATE)
1558                && !runstate_check(RUN_STATE_PAUSED)) { /* HACK */
1559         error_setg(errp,
1560                    "Guest must be stopped for opening of encrypted image");
1561         ret = -EBUSY;
1562         goto close_and_fail;
1563     }
1564 
1565     QDECREF(options);
1566     *pbs = bs;
1567     return 0;
1568 
1569 fail:
1570     if (file != NULL) {
1571         bdrv_unref(file);
1572     }
1573     QDECREF(bs->options);
1574     QDECREF(options);
1575     bs->options = NULL;
1576     if (!*pbs) {
1577         /* If *pbs is NULL, a new BDS has been created in this function and
1578            needs to be freed now. Otherwise, it does not need to be closed,
1579            since it has not really been opened yet. */
1580         bdrv_unref(bs);
1581     }
1582     if (local_err) {
1583         error_propagate(errp, local_err);
1584     }
1585     return ret;
1586 
1587 close_and_fail:
1588     /* See fail path, but now the BDS has to be always closed */
1589     if (*pbs) {
1590         bdrv_close(bs);
1591     } else {
1592         bdrv_unref(bs);
1593     }
1594     QDECREF(options);
1595     if (local_err) {
1596         error_propagate(errp, local_err);
1597     }
1598     return ret;
1599 }
1600 
1601 typedef struct BlockReopenQueueEntry {
1602      bool prepared;
1603      BDRVReopenState state;
1604      QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry;
1605 } BlockReopenQueueEntry;
1606 
1607 /*
1608  * Adds a BlockDriverState to a simple queue for an atomic, transactional
1609  * reopen of multiple devices.
1610  *
1611  * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT
1612  * already performed, or alternatively may be NULL a new BlockReopenQueue will
1613  * be created and initialized. This newly created BlockReopenQueue should be
1614  * passed back in for subsequent calls that are intended to be of the same
1615  * atomic 'set'.
1616  *
1617  * bs is the BlockDriverState to add to the reopen queue.
1618  *
1619  * flags contains the open flags for the associated bs
1620  *
1621  * returns a pointer to bs_queue, which is either the newly allocated
1622  * bs_queue, or the existing bs_queue being used.
1623  *
1624  */
1625 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
1626                                     BlockDriverState *bs, int flags)
1627 {
1628     assert(bs != NULL);
1629 
1630     BlockReopenQueueEntry *bs_entry;
1631     if (bs_queue == NULL) {
1632         bs_queue = g_new0(BlockReopenQueue, 1);
1633         QSIMPLEQ_INIT(bs_queue);
1634     }
1635 
1636     /* bdrv_open() masks this flag out */
1637     flags &= ~BDRV_O_PROTOCOL;
1638 
1639     if (bs->file) {
1640         bdrv_reopen_queue(bs_queue, bs->file, bdrv_inherited_flags(flags));
1641     }
1642 
1643     bs_entry = g_new0(BlockReopenQueueEntry, 1);
1644     QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry);
1645 
1646     bs_entry->state.bs = bs;
1647     bs_entry->state.flags = flags;
1648 
1649     return bs_queue;
1650 }
1651 
1652 /*
1653  * Reopen multiple BlockDriverStates atomically & transactionally.
1654  *
1655  * The queue passed in (bs_queue) must have been built up previous
1656  * via bdrv_reopen_queue().
1657  *
1658  * Reopens all BDS specified in the queue, with the appropriate
1659  * flags.  All devices are prepared for reopen, and failure of any
1660  * device will cause all device changes to be abandonded, and intermediate
1661  * data cleaned up.
1662  *
1663  * If all devices prepare successfully, then the changes are committed
1664  * to all devices.
1665  *
1666  */
1667 int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
1668 {
1669     int ret = -1;
1670     BlockReopenQueueEntry *bs_entry, *next;
1671     Error *local_err = NULL;
1672 
1673     assert(bs_queue != NULL);
1674 
1675     bdrv_drain_all();
1676 
1677     QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1678         if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) {
1679             error_propagate(errp, local_err);
1680             goto cleanup;
1681         }
1682         bs_entry->prepared = true;
1683     }
1684 
1685     /* If we reach this point, we have success and just need to apply the
1686      * changes
1687      */
1688     QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
1689         bdrv_reopen_commit(&bs_entry->state);
1690     }
1691 
1692     ret = 0;
1693 
1694 cleanup:
1695     QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
1696         if (ret && bs_entry->prepared) {
1697             bdrv_reopen_abort(&bs_entry->state);
1698         }
1699         g_free(bs_entry);
1700     }
1701     g_free(bs_queue);
1702     return ret;
1703 }
1704 
1705 
1706 /* Reopen a single BlockDriverState with the specified flags. */
1707 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp)
1708 {
1709     int ret = -1;
1710     Error *local_err = NULL;
1711     BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, bdrv_flags);
1712 
1713     ret = bdrv_reopen_multiple(queue, &local_err);
1714     if (local_err != NULL) {
1715         error_propagate(errp, local_err);
1716     }
1717     return ret;
1718 }
1719 
1720 
1721 /*
1722  * Prepares a BlockDriverState for reopen. All changes are staged in the
1723  * 'opaque' field of the BDRVReopenState, which is used and allocated by
1724  * the block driver layer .bdrv_reopen_prepare()
1725  *
1726  * bs is the BlockDriverState to reopen
1727  * flags are the new open flags
1728  * queue is the reopen queue
1729  *
1730  * Returns 0 on success, non-zero on error.  On error errp will be set
1731  * as well.
1732  *
1733  * On failure, bdrv_reopen_abort() will be called to clean up any data.
1734  * It is the responsibility of the caller to then call the abort() or
1735  * commit() for any other BDS that have been left in a prepare() state
1736  *
1737  */
1738 int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
1739                         Error **errp)
1740 {
1741     int ret = -1;
1742     Error *local_err = NULL;
1743     BlockDriver *drv;
1744 
1745     assert(reopen_state != NULL);
1746     assert(reopen_state->bs->drv != NULL);
1747     drv = reopen_state->bs->drv;
1748 
1749     /* if we are to stay read-only, do not allow permission change
1750      * to r/w */
1751     if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
1752         reopen_state->flags & BDRV_O_RDWR) {
1753         error_set(errp, QERR_DEVICE_IS_READ_ONLY,
1754                   reopen_state->bs->device_name);
1755         goto error;
1756     }
1757 
1758 
1759     ret = bdrv_flush(reopen_state->bs);
1760     if (ret) {
1761         error_set(errp, ERROR_CLASS_GENERIC_ERROR, "Error (%s) flushing drive",
1762                   strerror(-ret));
1763         goto error;
1764     }
1765 
1766     if (drv->bdrv_reopen_prepare) {
1767         ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err);
1768         if (ret) {
1769             if (local_err != NULL) {
1770                 error_propagate(errp, local_err);
1771             } else {
1772                 error_setg(errp, "failed while preparing to reopen image '%s'",
1773                            reopen_state->bs->filename);
1774             }
1775             goto error;
1776         }
1777     } else {
1778         /* It is currently mandatory to have a bdrv_reopen_prepare()
1779          * handler for each supported drv. */
1780         error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
1781                   drv->format_name, reopen_state->bs->device_name,
1782                  "reopening of file");
1783         ret = -1;
1784         goto error;
1785     }
1786 
1787     ret = 0;
1788 
1789 error:
1790     return ret;
1791 }
1792 
1793 /*
1794  * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and
1795  * makes them final by swapping the staging BlockDriverState contents into
1796  * the active BlockDriverState contents.
1797  */
1798 void bdrv_reopen_commit(BDRVReopenState *reopen_state)
1799 {
1800     BlockDriver *drv;
1801 
1802     assert(reopen_state != NULL);
1803     drv = reopen_state->bs->drv;
1804     assert(drv != NULL);
1805 
1806     /* If there are any driver level actions to take */
1807     if (drv->bdrv_reopen_commit) {
1808         drv->bdrv_reopen_commit(reopen_state);
1809     }
1810 
1811     /* set BDS specific flags now */
1812     reopen_state->bs->open_flags         = reopen_state->flags;
1813     reopen_state->bs->enable_write_cache = !!(reopen_state->flags &
1814                                               BDRV_O_CACHE_WB);
1815     reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
1816 
1817     bdrv_refresh_limits(reopen_state->bs);
1818 }
1819 
1820 /*
1821  * Abort the reopen, and delete and free the staged changes in
1822  * reopen_state
1823  */
1824 void bdrv_reopen_abort(BDRVReopenState *reopen_state)
1825 {
1826     BlockDriver *drv;
1827 
1828     assert(reopen_state != NULL);
1829     drv = reopen_state->bs->drv;
1830     assert(drv != NULL);
1831 
1832     if (drv->bdrv_reopen_abort) {
1833         drv->bdrv_reopen_abort(reopen_state);
1834     }
1835 }
1836 
1837 
1838 void bdrv_close(BlockDriverState *bs)
1839 {
1840     if (bs->job) {
1841         block_job_cancel_sync(bs->job);
1842     }
1843     bdrv_drain_all(); /* complete I/O */
1844     bdrv_flush(bs);
1845     bdrv_drain_all(); /* in case flush left pending I/O */
1846     notifier_list_notify(&bs->close_notifiers, bs);
1847 
1848     if (bs->drv) {
1849         if (bs->backing_hd) {
1850             BlockDriverState *backing_hd = bs->backing_hd;
1851             bdrv_set_backing_hd(bs, NULL);
1852             bdrv_unref(backing_hd);
1853         }
1854         bs->drv->bdrv_close(bs);
1855         g_free(bs->opaque);
1856         bs->opaque = NULL;
1857         bs->drv = NULL;
1858         bs->copy_on_read = 0;
1859         bs->backing_file[0] = '\0';
1860         bs->backing_format[0] = '\0';
1861         bs->total_sectors = 0;
1862         bs->encrypted = 0;
1863         bs->valid_key = 0;
1864         bs->sg = 0;
1865         bs->growable = 0;
1866         bs->zero_beyond_eof = false;
1867         QDECREF(bs->options);
1868         bs->options = NULL;
1869 
1870         if (bs->file != NULL) {
1871             bdrv_unref(bs->file);
1872             bs->file = NULL;
1873         }
1874     }
1875 
1876     bdrv_dev_change_media_cb(bs, false);
1877 
1878     /*throttling disk I/O limits*/
1879     if (bs->io_limits_enabled) {
1880         bdrv_io_limits_disable(bs);
1881     }
1882 }
1883 
1884 void bdrv_close_all(void)
1885 {
1886     BlockDriverState *bs;
1887 
1888     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1889         AioContext *aio_context = bdrv_get_aio_context(bs);
1890 
1891         aio_context_acquire(aio_context);
1892         bdrv_close(bs);
1893         aio_context_release(aio_context);
1894     }
1895 }
1896 
1897 /* Check if any requests are in-flight (including throttled requests) */
1898 static bool bdrv_requests_pending(BlockDriverState *bs)
1899 {
1900     if (!QLIST_EMPTY(&bs->tracked_requests)) {
1901         return true;
1902     }
1903     if (!qemu_co_queue_empty(&bs->throttled_reqs[0])) {
1904         return true;
1905     }
1906     if (!qemu_co_queue_empty(&bs->throttled_reqs[1])) {
1907         return true;
1908     }
1909     if (bs->file && bdrv_requests_pending(bs->file)) {
1910         return true;
1911     }
1912     if (bs->backing_hd && bdrv_requests_pending(bs->backing_hd)) {
1913         return true;
1914     }
1915     return false;
1916 }
1917 
1918 /*
1919  * Wait for pending requests to complete across all BlockDriverStates
1920  *
1921  * This function does not flush data to disk, use bdrv_flush_all() for that
1922  * after calling this function.
1923  *
1924  * Note that completion of an asynchronous I/O operation can trigger any
1925  * number of other I/O operations on other devices---for example a coroutine
1926  * can be arbitrarily complex and a constant flow of I/O can come until the
1927  * coroutine is complete.  Because of this, it is not possible to have a
1928  * function to drain a single device's I/O queue.
1929  */
1930 void bdrv_drain_all(void)
1931 {
1932     /* Always run first iteration so any pending completion BHs run */
1933     bool busy = true;
1934     BlockDriverState *bs;
1935 
1936     while (busy) {
1937         busy = false;
1938 
1939         QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
1940             AioContext *aio_context = bdrv_get_aio_context(bs);
1941             bool bs_busy;
1942 
1943             aio_context_acquire(aio_context);
1944             bdrv_start_throttled_reqs(bs);
1945             bs_busy = bdrv_requests_pending(bs);
1946             bs_busy |= aio_poll(aio_context, bs_busy);
1947             aio_context_release(aio_context);
1948 
1949             busy |= bs_busy;
1950         }
1951     }
1952 }
1953 
1954 /* make a BlockDriverState anonymous by removing from bdrv_state and
1955  * graph_bdrv_state list.
1956    Also, NULL terminate the device_name to prevent double remove */
1957 void bdrv_make_anon(BlockDriverState *bs)
1958 {
1959     if (bs->device_name[0] != '\0') {
1960         QTAILQ_REMOVE(&bdrv_states, bs, device_list);
1961     }
1962     bs->device_name[0] = '\0';
1963     if (bs->node_name[0] != '\0') {
1964         QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list);
1965     }
1966     bs->node_name[0] = '\0';
1967 }
1968 
1969 static void bdrv_rebind(BlockDriverState *bs)
1970 {
1971     if (bs->drv && bs->drv->bdrv_rebind) {
1972         bs->drv->bdrv_rebind(bs);
1973     }
1974 }
1975 
1976 static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
1977                                      BlockDriverState *bs_src)
1978 {
1979     /* move some fields that need to stay attached to the device */
1980 
1981     /* dev info */
1982     bs_dest->dev_ops            = bs_src->dev_ops;
1983     bs_dest->dev_opaque         = bs_src->dev_opaque;
1984     bs_dest->dev                = bs_src->dev;
1985     bs_dest->guest_block_size   = bs_src->guest_block_size;
1986     bs_dest->copy_on_read       = bs_src->copy_on_read;
1987 
1988     bs_dest->enable_write_cache = bs_src->enable_write_cache;
1989 
1990     /* i/o throttled req */
1991     memcpy(&bs_dest->throttle_state,
1992            &bs_src->throttle_state,
1993            sizeof(ThrottleState));
1994     bs_dest->throttled_reqs[0]  = bs_src->throttled_reqs[0];
1995     bs_dest->throttled_reqs[1]  = bs_src->throttled_reqs[1];
1996     bs_dest->io_limits_enabled  = bs_src->io_limits_enabled;
1997 
1998     /* r/w error */
1999     bs_dest->on_read_error      = bs_src->on_read_error;
2000     bs_dest->on_write_error     = bs_src->on_write_error;
2001 
2002     /* i/o status */
2003     bs_dest->iostatus_enabled   = bs_src->iostatus_enabled;
2004     bs_dest->iostatus           = bs_src->iostatus;
2005 
2006     /* dirty bitmap */
2007     bs_dest->dirty_bitmaps      = bs_src->dirty_bitmaps;
2008 
2009     /* reference count */
2010     bs_dest->refcnt             = bs_src->refcnt;
2011 
2012     /* job */
2013     bs_dest->job                = bs_src->job;
2014 
2015     /* keep the same entry in bdrv_states */
2016     pstrcpy(bs_dest->device_name, sizeof(bs_dest->device_name),
2017             bs_src->device_name);
2018     bs_dest->device_list = bs_src->device_list;
2019     memcpy(bs_dest->op_blockers, bs_src->op_blockers,
2020            sizeof(bs_dest->op_blockers));
2021 }
2022 
2023 /*
2024  * Swap bs contents for two image chains while they are live,
2025  * while keeping required fields on the BlockDriverState that is
2026  * actually attached to a device.
2027  *
2028  * This will modify the BlockDriverState fields, and swap contents
2029  * between bs_new and bs_old. Both bs_new and bs_old are modified.
2030  *
2031  * bs_new is required to be anonymous.
2032  *
2033  * This function does not create any image files.
2034  */
2035 void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
2036 {
2037     BlockDriverState tmp;
2038 
2039     /* The code needs to swap the node_name but simply swapping node_list won't
2040      * work so first remove the nodes from the graph list, do the swap then
2041      * insert them back if needed.
2042      */
2043     if (bs_new->node_name[0] != '\0') {
2044         QTAILQ_REMOVE(&graph_bdrv_states, bs_new, node_list);
2045     }
2046     if (bs_old->node_name[0] != '\0') {
2047         QTAILQ_REMOVE(&graph_bdrv_states, bs_old, node_list);
2048     }
2049 
2050     /* bs_new must be anonymous and shouldn't have anything fancy enabled */
2051     assert(bs_new->device_name[0] == '\0');
2052     assert(QLIST_EMPTY(&bs_new->dirty_bitmaps));
2053     assert(bs_new->job == NULL);
2054     assert(bs_new->dev == NULL);
2055     assert(bs_new->io_limits_enabled == false);
2056     assert(!throttle_have_timer(&bs_new->throttle_state));
2057 
2058     tmp = *bs_new;
2059     *bs_new = *bs_old;
2060     *bs_old = tmp;
2061 
2062     /* there are some fields that should not be swapped, move them back */
2063     bdrv_move_feature_fields(&tmp, bs_old);
2064     bdrv_move_feature_fields(bs_old, bs_new);
2065     bdrv_move_feature_fields(bs_new, &tmp);
2066 
2067     /* bs_new shouldn't be in bdrv_states even after the swap!  */
2068     assert(bs_new->device_name[0] == '\0');
2069 
2070     /* Check a few fields that should remain attached to the device */
2071     assert(bs_new->dev == NULL);
2072     assert(bs_new->job == NULL);
2073     assert(bs_new->io_limits_enabled == false);
2074     assert(!throttle_have_timer(&bs_new->throttle_state));
2075 
2076     /* insert the nodes back into the graph node list if needed */
2077     if (bs_new->node_name[0] != '\0') {
2078         QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs_new, node_list);
2079     }
2080     if (bs_old->node_name[0] != '\0') {
2081         QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs_old, node_list);
2082     }
2083 
2084     bdrv_rebind(bs_new);
2085     bdrv_rebind(bs_old);
2086 }
2087 
2088 /*
2089  * Add new bs contents at the top of an image chain while the chain is
2090  * live, while keeping required fields on the top layer.
2091  *
2092  * This will modify the BlockDriverState fields, and swap contents
2093  * between bs_new and bs_top. Both bs_new and bs_top are modified.
2094  *
2095  * bs_new is required to be anonymous.
2096  *
2097  * This function does not create any image files.
2098  */
2099 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
2100 {
2101     bdrv_swap(bs_new, bs_top);
2102 
2103     /* The contents of 'tmp' will become bs_top, as we are
2104      * swapping bs_new and bs_top contents. */
2105     bdrv_set_backing_hd(bs_top, bs_new);
2106 }
2107 
2108 static void bdrv_delete(BlockDriverState *bs)
2109 {
2110     assert(!bs->dev);
2111     assert(!bs->job);
2112     assert(bdrv_op_blocker_is_empty(bs));
2113     assert(!bs->refcnt);
2114     assert(QLIST_EMPTY(&bs->dirty_bitmaps));
2115 
2116     bdrv_close(bs);
2117 
2118     /* remove from list, if necessary */
2119     bdrv_make_anon(bs);
2120 
2121     g_free(bs);
2122 }
2123 
2124 int bdrv_attach_dev(BlockDriverState *bs, void *dev)
2125 /* TODO change to DeviceState *dev when all users are qdevified */
2126 {
2127     if (bs->dev) {
2128         return -EBUSY;
2129     }
2130     bs->dev = dev;
2131     bdrv_iostatus_reset(bs);
2132     return 0;
2133 }
2134 
2135 /* TODO qdevified devices don't use this, remove when devices are qdevified */
2136 void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev)
2137 {
2138     if (bdrv_attach_dev(bs, dev) < 0) {
2139         abort();
2140     }
2141 }
2142 
2143 void bdrv_detach_dev(BlockDriverState *bs, void *dev)
2144 /* TODO change to DeviceState *dev when all users are qdevified */
2145 {
2146     assert(bs->dev == dev);
2147     bs->dev = NULL;
2148     bs->dev_ops = NULL;
2149     bs->dev_opaque = NULL;
2150     bs->guest_block_size = 512;
2151 }
2152 
2153 /* TODO change to return DeviceState * when all users are qdevified */
2154 void *bdrv_get_attached_dev(BlockDriverState *bs)
2155 {
2156     return bs->dev;
2157 }
2158 
2159 void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops,
2160                       void *opaque)
2161 {
2162     bs->dev_ops = ops;
2163     bs->dev_opaque = opaque;
2164 }
2165 
2166 static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load)
2167 {
2168     if (bs->dev_ops && bs->dev_ops->change_media_cb) {
2169         bool tray_was_closed = !bdrv_dev_is_tray_open(bs);
2170         bs->dev_ops->change_media_cb(bs->dev_opaque, load);
2171         if (tray_was_closed) {
2172             /* tray open */
2173             qapi_event_send_device_tray_moved(bdrv_get_device_name(bs),
2174                                               true, &error_abort);
2175         }
2176         if (load) {
2177             /* tray close */
2178             qapi_event_send_device_tray_moved(bdrv_get_device_name(bs),
2179                                               false, &error_abort);
2180         }
2181     }
2182 }
2183 
2184 bool bdrv_dev_has_removable_media(BlockDriverState *bs)
2185 {
2186     return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb);
2187 }
2188 
2189 void bdrv_dev_eject_request(BlockDriverState *bs, bool force)
2190 {
2191     if (bs->dev_ops && bs->dev_ops->eject_request_cb) {
2192         bs->dev_ops->eject_request_cb(bs->dev_opaque, force);
2193     }
2194 }
2195 
2196 bool bdrv_dev_is_tray_open(BlockDriverState *bs)
2197 {
2198     if (bs->dev_ops && bs->dev_ops->is_tray_open) {
2199         return bs->dev_ops->is_tray_open(bs->dev_opaque);
2200     }
2201     return false;
2202 }
2203 
2204 static void bdrv_dev_resize_cb(BlockDriverState *bs)
2205 {
2206     if (bs->dev_ops && bs->dev_ops->resize_cb) {
2207         bs->dev_ops->resize_cb(bs->dev_opaque);
2208     }
2209 }
2210 
2211 bool bdrv_dev_is_medium_locked(BlockDriverState *bs)
2212 {
2213     if (bs->dev_ops && bs->dev_ops->is_medium_locked) {
2214         return bs->dev_ops->is_medium_locked(bs->dev_opaque);
2215     }
2216     return false;
2217 }
2218 
2219 /*
2220  * Run consistency checks on an image
2221  *
2222  * Returns 0 if the check could be completed (it doesn't mean that the image is
2223  * free of errors) or -errno when an internal error occurred. The results of the
2224  * check are stored in res.
2225  */
2226 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix)
2227 {
2228     if (bs->drv->bdrv_check == NULL) {
2229         return -ENOTSUP;
2230     }
2231 
2232     memset(res, 0, sizeof(*res));
2233     return bs->drv->bdrv_check(bs, res, fix);
2234 }
2235 
2236 #define COMMIT_BUF_SECTORS 2048
2237 
2238 /* commit COW file into the raw image */
2239 int bdrv_commit(BlockDriverState *bs)
2240 {
2241     BlockDriver *drv = bs->drv;
2242     int64_t sector, total_sectors, length, backing_length;
2243     int n, ro, open_flags;
2244     int ret = 0;
2245     uint8_t *buf = NULL;
2246     char filename[PATH_MAX];
2247 
2248     if (!drv)
2249         return -ENOMEDIUM;
2250 
2251     if (!bs->backing_hd) {
2252         return -ENOTSUP;
2253     }
2254 
2255     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_COMMIT, NULL) ||
2256         bdrv_op_is_blocked(bs->backing_hd, BLOCK_OP_TYPE_COMMIT, NULL)) {
2257         return -EBUSY;
2258     }
2259 
2260     ro = bs->backing_hd->read_only;
2261     /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */
2262     pstrcpy(filename, sizeof(filename), bs->backing_hd->filename);
2263     open_flags =  bs->backing_hd->open_flags;
2264 
2265     if (ro) {
2266         if (bdrv_reopen(bs->backing_hd, open_flags | BDRV_O_RDWR, NULL)) {
2267             return -EACCES;
2268         }
2269     }
2270 
2271     length = bdrv_getlength(bs);
2272     if (length < 0) {
2273         ret = length;
2274         goto ro_cleanup;
2275     }
2276 
2277     backing_length = bdrv_getlength(bs->backing_hd);
2278     if (backing_length < 0) {
2279         ret = backing_length;
2280         goto ro_cleanup;
2281     }
2282 
2283     /* If our top snapshot is larger than the backing file image,
2284      * grow the backing file image if possible.  If not possible,
2285      * we must return an error */
2286     if (length > backing_length) {
2287         ret = bdrv_truncate(bs->backing_hd, length);
2288         if (ret < 0) {
2289             goto ro_cleanup;
2290         }
2291     }
2292 
2293     total_sectors = length >> BDRV_SECTOR_BITS;
2294     buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE);
2295 
2296     for (sector = 0; sector < total_sectors; sector += n) {
2297         ret = bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n);
2298         if (ret < 0) {
2299             goto ro_cleanup;
2300         }
2301         if (ret) {
2302             ret = bdrv_read(bs, sector, buf, n);
2303             if (ret < 0) {
2304                 goto ro_cleanup;
2305             }
2306 
2307             ret = bdrv_write(bs->backing_hd, sector, buf, n);
2308             if (ret < 0) {
2309                 goto ro_cleanup;
2310             }
2311         }
2312     }
2313 
2314     if (drv->bdrv_make_empty) {
2315         ret = drv->bdrv_make_empty(bs);
2316         if (ret < 0) {
2317             goto ro_cleanup;
2318         }
2319         bdrv_flush(bs);
2320     }
2321 
2322     /*
2323      * Make sure all data we wrote to the backing device is actually
2324      * stable on disk.
2325      */
2326     if (bs->backing_hd) {
2327         bdrv_flush(bs->backing_hd);
2328     }
2329 
2330     ret = 0;
2331 ro_cleanup:
2332     g_free(buf);
2333 
2334     if (ro) {
2335         /* ignoring error return here */
2336         bdrv_reopen(bs->backing_hd, open_flags & ~BDRV_O_RDWR, NULL);
2337     }
2338 
2339     return ret;
2340 }
2341 
2342 int bdrv_commit_all(void)
2343 {
2344     BlockDriverState *bs;
2345 
2346     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
2347         AioContext *aio_context = bdrv_get_aio_context(bs);
2348 
2349         aio_context_acquire(aio_context);
2350         if (bs->drv && bs->backing_hd) {
2351             int ret = bdrv_commit(bs);
2352             if (ret < 0) {
2353                 aio_context_release(aio_context);
2354                 return ret;
2355             }
2356         }
2357         aio_context_release(aio_context);
2358     }
2359     return 0;
2360 }
2361 
2362 /**
2363  * Remove an active request from the tracked requests list
2364  *
2365  * This function should be called when a tracked request is completing.
2366  */
2367 static void tracked_request_end(BdrvTrackedRequest *req)
2368 {
2369     if (req->serialising) {
2370         req->bs->serialising_in_flight--;
2371     }
2372 
2373     QLIST_REMOVE(req, list);
2374     qemu_co_queue_restart_all(&req->wait_queue);
2375 }
2376 
2377 /**
2378  * Add an active request to the tracked requests list
2379  */
2380 static void tracked_request_begin(BdrvTrackedRequest *req,
2381                                   BlockDriverState *bs,
2382                                   int64_t offset,
2383                                   unsigned int bytes, bool is_write)
2384 {
2385     *req = (BdrvTrackedRequest){
2386         .bs = bs,
2387         .offset         = offset,
2388         .bytes          = bytes,
2389         .is_write       = is_write,
2390         .co             = qemu_coroutine_self(),
2391         .serialising    = false,
2392         .overlap_offset = offset,
2393         .overlap_bytes  = bytes,
2394     };
2395 
2396     qemu_co_queue_init(&req->wait_queue);
2397 
2398     QLIST_INSERT_HEAD(&bs->tracked_requests, req, list);
2399 }
2400 
2401 static void mark_request_serialising(BdrvTrackedRequest *req, uint64_t align)
2402 {
2403     int64_t overlap_offset = req->offset & ~(align - 1);
2404     unsigned int overlap_bytes = ROUND_UP(req->offset + req->bytes, align)
2405                                - overlap_offset;
2406 
2407     if (!req->serialising) {
2408         req->bs->serialising_in_flight++;
2409         req->serialising = true;
2410     }
2411 
2412     req->overlap_offset = MIN(req->overlap_offset, overlap_offset);
2413     req->overlap_bytes = MAX(req->overlap_bytes, overlap_bytes);
2414 }
2415 
2416 /**
2417  * Round a region to cluster boundaries
2418  */
2419 void bdrv_round_to_clusters(BlockDriverState *bs,
2420                             int64_t sector_num, int nb_sectors,
2421                             int64_t *cluster_sector_num,
2422                             int *cluster_nb_sectors)
2423 {
2424     BlockDriverInfo bdi;
2425 
2426     if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) {
2427         *cluster_sector_num = sector_num;
2428         *cluster_nb_sectors = nb_sectors;
2429     } else {
2430         int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE;
2431         *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c);
2432         *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num +
2433                                             nb_sectors, c);
2434     }
2435 }
2436 
2437 static int bdrv_get_cluster_size(BlockDriverState *bs)
2438 {
2439     BlockDriverInfo bdi;
2440     int ret;
2441 
2442     ret = bdrv_get_info(bs, &bdi);
2443     if (ret < 0 || bdi.cluster_size == 0) {
2444         return bs->request_alignment;
2445     } else {
2446         return bdi.cluster_size;
2447     }
2448 }
2449 
2450 static bool tracked_request_overlaps(BdrvTrackedRequest *req,
2451                                      int64_t offset, unsigned int bytes)
2452 {
2453     /*        aaaa   bbbb */
2454     if (offset >= req->overlap_offset + req->overlap_bytes) {
2455         return false;
2456     }
2457     /* bbbb   aaaa        */
2458     if (req->overlap_offset >= offset + bytes) {
2459         return false;
2460     }
2461     return true;
2462 }
2463 
2464 static bool coroutine_fn wait_serialising_requests(BdrvTrackedRequest *self)
2465 {
2466     BlockDriverState *bs = self->bs;
2467     BdrvTrackedRequest *req;
2468     bool retry;
2469     bool waited = false;
2470 
2471     if (!bs->serialising_in_flight) {
2472         return false;
2473     }
2474 
2475     do {
2476         retry = false;
2477         QLIST_FOREACH(req, &bs->tracked_requests, list) {
2478             if (req == self || (!req->serialising && !self->serialising)) {
2479                 continue;
2480             }
2481             if (tracked_request_overlaps(req, self->overlap_offset,
2482                                          self->overlap_bytes))
2483             {
2484                 /* Hitting this means there was a reentrant request, for
2485                  * example, a block driver issuing nested requests.  This must
2486                  * never happen since it means deadlock.
2487                  */
2488                 assert(qemu_coroutine_self() != req->co);
2489 
2490                 /* If the request is already (indirectly) waiting for us, or
2491                  * will wait for us as soon as it wakes up, then just go on
2492                  * (instead of producing a deadlock in the former case). */
2493                 if (!req->waiting_for) {
2494                     self->waiting_for = req;
2495                     qemu_co_queue_wait(&req->wait_queue);
2496                     self->waiting_for = NULL;
2497                     retry = true;
2498                     waited = true;
2499                     break;
2500                 }
2501             }
2502         }
2503     } while (retry);
2504 
2505     return waited;
2506 }
2507 
2508 /*
2509  * Return values:
2510  * 0        - success
2511  * -EINVAL  - backing format specified, but no file
2512  * -ENOSPC  - can't update the backing file because no space is left in the
2513  *            image file header
2514  * -ENOTSUP - format driver doesn't support changing the backing file
2515  */
2516 int bdrv_change_backing_file(BlockDriverState *bs,
2517     const char *backing_file, const char *backing_fmt)
2518 {
2519     BlockDriver *drv = bs->drv;
2520     int ret;
2521 
2522     /* Backing file format doesn't make sense without a backing file */
2523     if (backing_fmt && !backing_file) {
2524         return -EINVAL;
2525     }
2526 
2527     if (drv->bdrv_change_backing_file != NULL) {
2528         ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
2529     } else {
2530         ret = -ENOTSUP;
2531     }
2532 
2533     if (ret == 0) {
2534         pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: "");
2535         pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: "");
2536     }
2537     return ret;
2538 }
2539 
2540 /*
2541  * Finds the image layer in the chain that has 'bs' as its backing file.
2542  *
2543  * active is the current topmost image.
2544  *
2545  * Returns NULL if bs is not found in active's image chain,
2546  * or if active == bs.
2547  */
2548 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
2549                                     BlockDriverState *bs)
2550 {
2551     BlockDriverState *overlay = NULL;
2552     BlockDriverState *intermediate;
2553 
2554     assert(active != NULL);
2555     assert(bs != NULL);
2556 
2557     /* if bs is the same as active, then by definition it has no overlay
2558      */
2559     if (active == bs) {
2560         return NULL;
2561     }
2562 
2563     intermediate = active;
2564     while (intermediate->backing_hd) {
2565         if (intermediate->backing_hd == bs) {
2566             overlay = intermediate;
2567             break;
2568         }
2569         intermediate = intermediate->backing_hd;
2570     }
2571 
2572     return overlay;
2573 }
2574 
2575 typedef struct BlkIntermediateStates {
2576     BlockDriverState *bs;
2577     QSIMPLEQ_ENTRY(BlkIntermediateStates) entry;
2578 } BlkIntermediateStates;
2579 
2580 
2581 /*
2582  * Drops images above 'base' up to and including 'top', and sets the image
2583  * above 'top' to have base as its backing file.
2584  *
2585  * Requires that the overlay to 'top' is opened r/w, so that the backing file
2586  * information in 'bs' can be properly updated.
2587  *
2588  * E.g., this will convert the following chain:
2589  * bottom <- base <- intermediate <- top <- active
2590  *
2591  * to
2592  *
2593  * bottom <- base <- active
2594  *
2595  * It is allowed for bottom==base, in which case it converts:
2596  *
2597  * base <- intermediate <- top <- active
2598  *
2599  * to
2600  *
2601  * base <- active
2602  *
2603  * Error conditions:
2604  *  if active == top, that is considered an error
2605  *
2606  */
2607 int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top,
2608                            BlockDriverState *base)
2609 {
2610     BlockDriverState *intermediate;
2611     BlockDriverState *base_bs = NULL;
2612     BlockDriverState *new_top_bs = NULL;
2613     BlkIntermediateStates *intermediate_state, *next;
2614     int ret = -EIO;
2615 
2616     QSIMPLEQ_HEAD(states_to_delete, BlkIntermediateStates) states_to_delete;
2617     QSIMPLEQ_INIT(&states_to_delete);
2618 
2619     if (!top->drv || !base->drv) {
2620         goto exit;
2621     }
2622 
2623     new_top_bs = bdrv_find_overlay(active, top);
2624 
2625     if (new_top_bs == NULL) {
2626         /* we could not find the image above 'top', this is an error */
2627         goto exit;
2628     }
2629 
2630     /* special case of new_top_bs->backing_hd already pointing to base - nothing
2631      * to do, no intermediate images */
2632     if (new_top_bs->backing_hd == base) {
2633         ret = 0;
2634         goto exit;
2635     }
2636 
2637     intermediate = top;
2638 
2639     /* now we will go down through the list, and add each BDS we find
2640      * into our deletion queue, until we hit the 'base'
2641      */
2642     while (intermediate) {
2643         intermediate_state = g_malloc0(sizeof(BlkIntermediateStates));
2644         intermediate_state->bs = intermediate;
2645         QSIMPLEQ_INSERT_TAIL(&states_to_delete, intermediate_state, entry);
2646 
2647         if (intermediate->backing_hd == base) {
2648             base_bs = intermediate->backing_hd;
2649             break;
2650         }
2651         intermediate = intermediate->backing_hd;
2652     }
2653     if (base_bs == NULL) {
2654         /* something went wrong, we did not end at the base. safely
2655          * unravel everything, and exit with error */
2656         goto exit;
2657     }
2658 
2659     /* success - we can delete the intermediate states, and link top->base */
2660     ret = bdrv_change_backing_file(new_top_bs, base_bs->filename,
2661                                    base_bs->drv ? base_bs->drv->format_name : "");
2662     if (ret) {
2663         goto exit;
2664     }
2665     bdrv_set_backing_hd(new_top_bs, base_bs);
2666 
2667     QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2668         /* so that bdrv_close() does not recursively close the chain */
2669         bdrv_set_backing_hd(intermediate_state->bs, NULL);
2670         bdrv_unref(intermediate_state->bs);
2671     }
2672     ret = 0;
2673 
2674 exit:
2675     QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) {
2676         g_free(intermediate_state);
2677     }
2678     return ret;
2679 }
2680 
2681 
2682 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
2683                                    size_t size)
2684 {
2685     int64_t len;
2686 
2687     if (size > INT_MAX) {
2688         return -EIO;
2689     }
2690 
2691     if (!bdrv_is_inserted(bs))
2692         return -ENOMEDIUM;
2693 
2694     if (bs->growable)
2695         return 0;
2696 
2697     len = bdrv_getlength(bs);
2698 
2699     if (offset < 0)
2700         return -EIO;
2701 
2702     if ((offset > len) || (len - offset < size))
2703         return -EIO;
2704 
2705     return 0;
2706 }
2707 
2708 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
2709                               int nb_sectors)
2710 {
2711     if (nb_sectors < 0 || nb_sectors > INT_MAX / BDRV_SECTOR_SIZE) {
2712         return -EIO;
2713     }
2714 
2715     return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
2716                                    nb_sectors * BDRV_SECTOR_SIZE);
2717 }
2718 
2719 typedef struct RwCo {
2720     BlockDriverState *bs;
2721     int64_t offset;
2722     QEMUIOVector *qiov;
2723     bool is_write;
2724     int ret;
2725     BdrvRequestFlags flags;
2726 } RwCo;
2727 
2728 static void coroutine_fn bdrv_rw_co_entry(void *opaque)
2729 {
2730     RwCo *rwco = opaque;
2731 
2732     if (!rwco->is_write) {
2733         rwco->ret = bdrv_co_do_preadv(rwco->bs, rwco->offset,
2734                                       rwco->qiov->size, rwco->qiov,
2735                                       rwco->flags);
2736     } else {
2737         rwco->ret = bdrv_co_do_pwritev(rwco->bs, rwco->offset,
2738                                        rwco->qiov->size, rwco->qiov,
2739                                        rwco->flags);
2740     }
2741 }
2742 
2743 /*
2744  * Process a vectored synchronous request using coroutines
2745  */
2746 static int bdrv_prwv_co(BlockDriverState *bs, int64_t offset,
2747                         QEMUIOVector *qiov, bool is_write,
2748                         BdrvRequestFlags flags)
2749 {
2750     Coroutine *co;
2751     RwCo rwco = {
2752         .bs = bs,
2753         .offset = offset,
2754         .qiov = qiov,
2755         .is_write = is_write,
2756         .ret = NOT_DONE,
2757         .flags = flags,
2758     };
2759 
2760     /**
2761      * In sync call context, when the vcpu is blocked, this throttling timer
2762      * will not fire; so the I/O throttling function has to be disabled here
2763      * if it has been enabled.
2764      */
2765     if (bs->io_limits_enabled) {
2766         fprintf(stderr, "Disabling I/O throttling on '%s' due "
2767                         "to synchronous I/O.\n", bdrv_get_device_name(bs));
2768         bdrv_io_limits_disable(bs);
2769     }
2770 
2771     if (qemu_in_coroutine()) {
2772         /* Fast-path if already in coroutine context */
2773         bdrv_rw_co_entry(&rwco);
2774     } else {
2775         AioContext *aio_context = bdrv_get_aio_context(bs);
2776 
2777         co = qemu_coroutine_create(bdrv_rw_co_entry);
2778         qemu_coroutine_enter(co, &rwco);
2779         while (rwco.ret == NOT_DONE) {
2780             aio_poll(aio_context, true);
2781         }
2782     }
2783     return rwco.ret;
2784 }
2785 
2786 /*
2787  * Process a synchronous request using coroutines
2788  */
2789 static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf,
2790                       int nb_sectors, bool is_write, BdrvRequestFlags flags)
2791 {
2792     QEMUIOVector qiov;
2793     struct iovec iov = {
2794         .iov_base = (void *)buf,
2795         .iov_len = nb_sectors * BDRV_SECTOR_SIZE,
2796     };
2797 
2798     if (nb_sectors < 0 || nb_sectors > INT_MAX / BDRV_SECTOR_SIZE) {
2799         return -EINVAL;
2800     }
2801 
2802     qemu_iovec_init_external(&qiov, &iov, 1);
2803     return bdrv_prwv_co(bs, sector_num << BDRV_SECTOR_BITS,
2804                         &qiov, is_write, flags);
2805 }
2806 
2807 /* return < 0 if error. See bdrv_write() for the return codes */
2808 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
2809               uint8_t *buf, int nb_sectors)
2810 {
2811     return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0);
2812 }
2813 
2814 /* Just like bdrv_read(), but with I/O throttling temporarily disabled */
2815 int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num,
2816                           uint8_t *buf, int nb_sectors)
2817 {
2818     bool enabled;
2819     int ret;
2820 
2821     enabled = bs->io_limits_enabled;
2822     bs->io_limits_enabled = false;
2823     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
2824     bs->io_limits_enabled = enabled;
2825     return ret;
2826 }
2827 
2828 /* Return < 0 if error. Important errors are:
2829   -EIO         generic I/O error (may happen for all errors)
2830   -ENOMEDIUM   No media inserted.
2831   -EINVAL      Invalid sector number or nb_sectors
2832   -EACCES      Trying to write a read-only device
2833 */
2834 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
2835                const uint8_t *buf, int nb_sectors)
2836 {
2837     return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0);
2838 }
2839 
2840 int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num,
2841                       int nb_sectors, BdrvRequestFlags flags)
2842 {
2843     return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true,
2844                       BDRV_REQ_ZERO_WRITE | flags);
2845 }
2846 
2847 /*
2848  * Completely zero out a block device with the help of bdrv_write_zeroes.
2849  * The operation is sped up by checking the block status and only writing
2850  * zeroes to the device if they currently do not return zeroes. Optional
2851  * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP).
2852  *
2853  * Returns < 0 on error, 0 on success. For error codes see bdrv_write().
2854  */
2855 int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags)
2856 {
2857     int64_t target_size;
2858     int64_t ret, nb_sectors, sector_num = 0;
2859     int n;
2860 
2861     target_size = bdrv_getlength(bs);
2862     if (target_size < 0) {
2863         return target_size;
2864     }
2865     target_size /= BDRV_SECTOR_SIZE;
2866 
2867     for (;;) {
2868         nb_sectors = target_size - sector_num;
2869         if (nb_sectors <= 0) {
2870             return 0;
2871         }
2872         if (nb_sectors > INT_MAX) {
2873             nb_sectors = INT_MAX;
2874         }
2875         ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n);
2876         if (ret < 0) {
2877             error_report("error getting block status at sector %" PRId64 ": %s",
2878                          sector_num, strerror(-ret));
2879             return ret;
2880         }
2881         if (ret & BDRV_BLOCK_ZERO) {
2882             sector_num += n;
2883             continue;
2884         }
2885         ret = bdrv_write_zeroes(bs, sector_num, n, flags);
2886         if (ret < 0) {
2887             error_report("error writing zeroes at sector %" PRId64 ": %s",
2888                          sector_num, strerror(-ret));
2889             return ret;
2890         }
2891         sector_num += n;
2892     }
2893 }
2894 
2895 int bdrv_pread(BlockDriverState *bs, int64_t offset, void *buf, int bytes)
2896 {
2897     QEMUIOVector qiov;
2898     struct iovec iov = {
2899         .iov_base = (void *)buf,
2900         .iov_len = bytes,
2901     };
2902     int ret;
2903 
2904     if (bytes < 0) {
2905         return -EINVAL;
2906     }
2907 
2908     qemu_iovec_init_external(&qiov, &iov, 1);
2909     ret = bdrv_prwv_co(bs, offset, &qiov, false, 0);
2910     if (ret < 0) {
2911         return ret;
2912     }
2913 
2914     return bytes;
2915 }
2916 
2917 int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov)
2918 {
2919     int ret;
2920 
2921     ret = bdrv_prwv_co(bs, offset, qiov, true, 0);
2922     if (ret < 0) {
2923         return ret;
2924     }
2925 
2926     return qiov->size;
2927 }
2928 
2929 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
2930                 const void *buf, int bytes)
2931 {
2932     QEMUIOVector qiov;
2933     struct iovec iov = {
2934         .iov_base   = (void *) buf,
2935         .iov_len    = bytes,
2936     };
2937 
2938     if (bytes < 0) {
2939         return -EINVAL;
2940     }
2941 
2942     qemu_iovec_init_external(&qiov, &iov, 1);
2943     return bdrv_pwritev(bs, offset, &qiov);
2944 }
2945 
2946 /*
2947  * Writes to the file and ensures that no writes are reordered across this
2948  * request (acts as a barrier)
2949  *
2950  * Returns 0 on success, -errno in error cases.
2951  */
2952 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
2953     const void *buf, int count)
2954 {
2955     int ret;
2956 
2957     ret = bdrv_pwrite(bs, offset, buf, count);
2958     if (ret < 0) {
2959         return ret;
2960     }
2961 
2962     /* No flush needed for cache modes that already do it */
2963     if (bs->enable_write_cache) {
2964         bdrv_flush(bs);
2965     }
2966 
2967     return 0;
2968 }
2969 
2970 static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs,
2971         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
2972 {
2973     /* Perform I/O through a temporary buffer so that users who scribble over
2974      * their read buffer while the operation is in progress do not end up
2975      * modifying the image file.  This is critical for zero-copy guest I/O
2976      * where anything might happen inside guest memory.
2977      */
2978     void *bounce_buffer;
2979 
2980     BlockDriver *drv = bs->drv;
2981     struct iovec iov;
2982     QEMUIOVector bounce_qiov;
2983     int64_t cluster_sector_num;
2984     int cluster_nb_sectors;
2985     size_t skip_bytes;
2986     int ret;
2987 
2988     /* Cover entire cluster so no additional backing file I/O is required when
2989      * allocating cluster in the image file.
2990      */
2991     bdrv_round_to_clusters(bs, sector_num, nb_sectors,
2992                            &cluster_sector_num, &cluster_nb_sectors);
2993 
2994     trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors,
2995                                    cluster_sector_num, cluster_nb_sectors);
2996 
2997     iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE;
2998     iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len);
2999     qemu_iovec_init_external(&bounce_qiov, &iov, 1);
3000 
3001     ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors,
3002                              &bounce_qiov);
3003     if (ret < 0) {
3004         goto err;
3005     }
3006 
3007     if (drv->bdrv_co_write_zeroes &&
3008         buffer_is_zero(bounce_buffer, iov.iov_len)) {
3009         ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num,
3010                                       cluster_nb_sectors, 0);
3011     } else {
3012         /* This does not change the data on the disk, it is not necessary
3013          * to flush even in cache=writethrough mode.
3014          */
3015         ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors,
3016                                   &bounce_qiov);
3017     }
3018 
3019     if (ret < 0) {
3020         /* It might be okay to ignore write errors for guest requests.  If this
3021          * is a deliberate copy-on-read then we don't want to ignore the error.
3022          * Simply report it in all cases.
3023          */
3024         goto err;
3025     }
3026 
3027     skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE;
3028     qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes,
3029                         nb_sectors * BDRV_SECTOR_SIZE);
3030 
3031 err:
3032     qemu_vfree(bounce_buffer);
3033     return ret;
3034 }
3035 
3036 /*
3037  * Forwards an already correctly aligned request to the BlockDriver. This
3038  * handles copy on read and zeroing after EOF; any other features must be
3039  * implemented by the caller.
3040  */
3041 static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs,
3042     BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
3043     int64_t align, QEMUIOVector *qiov, int flags)
3044 {
3045     BlockDriver *drv = bs->drv;
3046     int ret;
3047 
3048     int64_t sector_num = offset >> BDRV_SECTOR_BITS;
3049     unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
3050 
3051     assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
3052     assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
3053 
3054     /* Handle Copy on Read and associated serialisation */
3055     if (flags & BDRV_REQ_COPY_ON_READ) {
3056         /* If we touch the same cluster it counts as an overlap.  This
3057          * guarantees that allocating writes will be serialized and not race
3058          * with each other for the same cluster.  For example, in copy-on-read
3059          * it ensures that the CoR read and write operations are atomic and
3060          * guest writes cannot interleave between them. */
3061         mark_request_serialising(req, bdrv_get_cluster_size(bs));
3062     }
3063 
3064     wait_serialising_requests(req);
3065 
3066     if (flags & BDRV_REQ_COPY_ON_READ) {
3067         int pnum;
3068 
3069         ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum);
3070         if (ret < 0) {
3071             goto out;
3072         }
3073 
3074         if (!ret || pnum != nb_sectors) {
3075             ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov);
3076             goto out;
3077         }
3078     }
3079 
3080     /* Forward the request to the BlockDriver */
3081     if (!(bs->zero_beyond_eof && bs->growable)) {
3082         ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov);
3083     } else {
3084         /* Read zeros after EOF of growable BDSes */
3085         int64_t len, total_sectors, max_nb_sectors;
3086 
3087         len = bdrv_getlength(bs);
3088         if (len < 0) {
3089             ret = len;
3090             goto out;
3091         }
3092 
3093         total_sectors = DIV_ROUND_UP(len, BDRV_SECTOR_SIZE);
3094         max_nb_sectors = ROUND_UP(MAX(0, total_sectors - sector_num),
3095                                   align >> BDRV_SECTOR_BITS);
3096         if (max_nb_sectors > 0) {
3097             ret = drv->bdrv_co_readv(bs, sector_num,
3098                                      MIN(nb_sectors, max_nb_sectors), qiov);
3099         } else {
3100             ret = 0;
3101         }
3102 
3103         /* Reading beyond end of file is supposed to produce zeroes */
3104         if (ret == 0 && total_sectors < sector_num + nb_sectors) {
3105             uint64_t offset = MAX(0, total_sectors - sector_num);
3106             uint64_t bytes = (sector_num + nb_sectors - offset) *
3107                               BDRV_SECTOR_SIZE;
3108             qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes);
3109         }
3110     }
3111 
3112 out:
3113     return ret;
3114 }
3115 
3116 /*
3117  * Handle a read request in coroutine context
3118  */
3119 static int coroutine_fn bdrv_co_do_preadv(BlockDriverState *bs,
3120     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
3121     BdrvRequestFlags flags)
3122 {
3123     BlockDriver *drv = bs->drv;
3124     BdrvTrackedRequest req;
3125 
3126     /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
3127     uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
3128     uint8_t *head_buf = NULL;
3129     uint8_t *tail_buf = NULL;
3130     QEMUIOVector local_qiov;
3131     bool use_local_qiov = false;
3132     int ret;
3133 
3134     if (!drv) {
3135         return -ENOMEDIUM;
3136     }
3137     if (bdrv_check_byte_request(bs, offset, bytes)) {
3138         return -EIO;
3139     }
3140 
3141     if (bs->copy_on_read) {
3142         flags |= BDRV_REQ_COPY_ON_READ;
3143     }
3144 
3145     /* throttling disk I/O */
3146     if (bs->io_limits_enabled) {
3147         bdrv_io_limits_intercept(bs, bytes, false);
3148     }
3149 
3150     /* Align read if necessary by padding qiov */
3151     if (offset & (align - 1)) {
3152         head_buf = qemu_blockalign(bs, align);
3153         qemu_iovec_init(&local_qiov, qiov->niov + 2);
3154         qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
3155         qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3156         use_local_qiov = true;
3157 
3158         bytes += offset & (align - 1);
3159         offset = offset & ~(align - 1);
3160     }
3161 
3162     if ((offset + bytes) & (align - 1)) {
3163         if (!use_local_qiov) {
3164             qemu_iovec_init(&local_qiov, qiov->niov + 1);
3165             qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3166             use_local_qiov = true;
3167         }
3168         tail_buf = qemu_blockalign(bs, align);
3169         qemu_iovec_add(&local_qiov, tail_buf,
3170                        align - ((offset + bytes) & (align - 1)));
3171 
3172         bytes = ROUND_UP(bytes, align);
3173     }
3174 
3175     tracked_request_begin(&req, bs, offset, bytes, false);
3176     ret = bdrv_aligned_preadv(bs, &req, offset, bytes, align,
3177                               use_local_qiov ? &local_qiov : qiov,
3178                               flags);
3179     tracked_request_end(&req);
3180 
3181     if (use_local_qiov) {
3182         qemu_iovec_destroy(&local_qiov);
3183         qemu_vfree(head_buf);
3184         qemu_vfree(tail_buf);
3185     }
3186 
3187     return ret;
3188 }
3189 
3190 static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs,
3191     int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
3192     BdrvRequestFlags flags)
3193 {
3194     if (nb_sectors < 0 || nb_sectors > (UINT_MAX >> BDRV_SECTOR_BITS)) {
3195         return -EINVAL;
3196     }
3197 
3198     return bdrv_co_do_preadv(bs, sector_num << BDRV_SECTOR_BITS,
3199                              nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
3200 }
3201 
3202 int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num,
3203     int nb_sectors, QEMUIOVector *qiov)
3204 {
3205     trace_bdrv_co_readv(bs, sector_num, nb_sectors);
3206 
3207     return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0);
3208 }
3209 
3210 int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs,
3211     int64_t sector_num, int nb_sectors, QEMUIOVector *qiov)
3212 {
3213     trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors);
3214 
3215     return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov,
3216                             BDRV_REQ_COPY_ON_READ);
3217 }
3218 
3219 /* if no limit is specified in the BlockLimits use a default
3220  * of 32768 512-byte sectors (16 MiB) per request.
3221  */
3222 #define MAX_WRITE_ZEROES_DEFAULT 32768
3223 
3224 static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs,
3225     int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
3226 {
3227     BlockDriver *drv = bs->drv;
3228     QEMUIOVector qiov;
3229     struct iovec iov = {0};
3230     int ret = 0;
3231 
3232     int max_write_zeroes = bs->bl.max_write_zeroes ?
3233                            bs->bl.max_write_zeroes : MAX_WRITE_ZEROES_DEFAULT;
3234 
3235     while (nb_sectors > 0 && !ret) {
3236         int num = nb_sectors;
3237 
3238         /* Align request.  Block drivers can expect the "bulk" of the request
3239          * to be aligned.
3240          */
3241         if (bs->bl.write_zeroes_alignment
3242             && num > bs->bl.write_zeroes_alignment) {
3243             if (sector_num % bs->bl.write_zeroes_alignment != 0) {
3244                 /* Make a small request up to the first aligned sector.  */
3245                 num = bs->bl.write_zeroes_alignment;
3246                 num -= sector_num % bs->bl.write_zeroes_alignment;
3247             } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) {
3248                 /* Shorten the request to the last aligned sector.  num cannot
3249                  * underflow because num > bs->bl.write_zeroes_alignment.
3250                  */
3251                 num -= (sector_num + num) % bs->bl.write_zeroes_alignment;
3252             }
3253         }
3254 
3255         /* limit request size */
3256         if (num > max_write_zeroes) {
3257             num = max_write_zeroes;
3258         }
3259 
3260         ret = -ENOTSUP;
3261         /* First try the efficient write zeroes operation */
3262         if (drv->bdrv_co_write_zeroes) {
3263             ret = drv->bdrv_co_write_zeroes(bs, sector_num, num, flags);
3264         }
3265 
3266         if (ret == -ENOTSUP) {
3267             /* Fall back to bounce buffer if write zeroes is unsupported */
3268             iov.iov_len = num * BDRV_SECTOR_SIZE;
3269             if (iov.iov_base == NULL) {
3270                 iov.iov_base = qemu_blockalign(bs, num * BDRV_SECTOR_SIZE);
3271                 memset(iov.iov_base, 0, num * BDRV_SECTOR_SIZE);
3272             }
3273             qemu_iovec_init_external(&qiov, &iov, 1);
3274 
3275             ret = drv->bdrv_co_writev(bs, sector_num, num, &qiov);
3276 
3277             /* Keep bounce buffer around if it is big enough for all
3278              * all future requests.
3279              */
3280             if (num < max_write_zeroes) {
3281                 qemu_vfree(iov.iov_base);
3282                 iov.iov_base = NULL;
3283             }
3284         }
3285 
3286         sector_num += num;
3287         nb_sectors -= num;
3288     }
3289 
3290     qemu_vfree(iov.iov_base);
3291     return ret;
3292 }
3293 
3294 /*
3295  * Forwards an already correctly aligned write request to the BlockDriver.
3296  */
3297 static int coroutine_fn bdrv_aligned_pwritev(BlockDriverState *bs,
3298     BdrvTrackedRequest *req, int64_t offset, unsigned int bytes,
3299     QEMUIOVector *qiov, int flags)
3300 {
3301     BlockDriver *drv = bs->drv;
3302     bool waited;
3303     int ret;
3304 
3305     int64_t sector_num = offset >> BDRV_SECTOR_BITS;
3306     unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS;
3307 
3308     assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0);
3309     assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0);
3310 
3311     waited = wait_serialising_requests(req);
3312     assert(!waited || !req->serialising);
3313     assert(req->overlap_offset <= offset);
3314     assert(offset + bytes <= req->overlap_offset + req->overlap_bytes);
3315 
3316     ret = notifier_with_return_list_notify(&bs->before_write_notifiers, req);
3317 
3318     if (!ret && bs->detect_zeroes != BLOCKDEV_DETECT_ZEROES_OPTIONS_OFF &&
3319         !(flags & BDRV_REQ_ZERO_WRITE) && drv->bdrv_co_write_zeroes &&
3320         qemu_iovec_is_zero(qiov)) {
3321         flags |= BDRV_REQ_ZERO_WRITE;
3322         if (bs->detect_zeroes == BLOCKDEV_DETECT_ZEROES_OPTIONS_UNMAP) {
3323             flags |= BDRV_REQ_MAY_UNMAP;
3324         }
3325     }
3326 
3327     if (ret < 0) {
3328         /* Do nothing, write notifier decided to fail this request */
3329     } else if (flags & BDRV_REQ_ZERO_WRITE) {
3330         BLKDBG_EVENT(bs, BLKDBG_PWRITEV_ZERO);
3331         ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors, flags);
3332     } else {
3333         BLKDBG_EVENT(bs, BLKDBG_PWRITEV);
3334         ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov);
3335     }
3336     BLKDBG_EVENT(bs, BLKDBG_PWRITEV_DONE);
3337 
3338     if (ret == 0 && !bs->enable_write_cache) {
3339         ret = bdrv_co_flush(bs);
3340     }
3341 
3342     bdrv_set_dirty(bs, sector_num, nb_sectors);
3343 
3344     if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
3345         bs->wr_highest_sector = sector_num + nb_sectors - 1;
3346     }
3347     if (bs->growable && ret >= 0) {
3348         bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors);
3349     }
3350 
3351     return ret;
3352 }
3353 
3354 /*
3355  * Handle a write request in coroutine context
3356  */
3357 static int coroutine_fn bdrv_co_do_pwritev(BlockDriverState *bs,
3358     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
3359     BdrvRequestFlags flags)
3360 {
3361     BdrvTrackedRequest req;
3362     /* TODO Lift BDRV_SECTOR_SIZE restriction in BlockDriver interface */
3363     uint64_t align = MAX(BDRV_SECTOR_SIZE, bs->request_alignment);
3364     uint8_t *head_buf = NULL;
3365     uint8_t *tail_buf = NULL;
3366     QEMUIOVector local_qiov;
3367     bool use_local_qiov = false;
3368     int ret;
3369 
3370     if (!bs->drv) {
3371         return -ENOMEDIUM;
3372     }
3373     if (bs->read_only) {
3374         return -EACCES;
3375     }
3376     if (bdrv_check_byte_request(bs, offset, bytes)) {
3377         return -EIO;
3378     }
3379 
3380     /* throttling disk I/O */
3381     if (bs->io_limits_enabled) {
3382         bdrv_io_limits_intercept(bs, bytes, true);
3383     }
3384 
3385     /*
3386      * Align write if necessary by performing a read-modify-write cycle.
3387      * Pad qiov with the read parts and be sure to have a tracked request not
3388      * only for bdrv_aligned_pwritev, but also for the reads of the RMW cycle.
3389      */
3390     tracked_request_begin(&req, bs, offset, bytes, true);
3391 
3392     if (offset & (align - 1)) {
3393         QEMUIOVector head_qiov;
3394         struct iovec head_iov;
3395 
3396         mark_request_serialising(&req, align);
3397         wait_serialising_requests(&req);
3398 
3399         head_buf = qemu_blockalign(bs, align);
3400         head_iov = (struct iovec) {
3401             .iov_base   = head_buf,
3402             .iov_len    = align,
3403         };
3404         qemu_iovec_init_external(&head_qiov, &head_iov, 1);
3405 
3406         BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_HEAD);
3407         ret = bdrv_aligned_preadv(bs, &req, offset & ~(align - 1), align,
3408                                   align, &head_qiov, 0);
3409         if (ret < 0) {
3410             goto fail;
3411         }
3412         BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_AFTER_HEAD);
3413 
3414         qemu_iovec_init(&local_qiov, qiov->niov + 2);
3415         qemu_iovec_add(&local_qiov, head_buf, offset & (align - 1));
3416         qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3417         use_local_qiov = true;
3418 
3419         bytes += offset & (align - 1);
3420         offset = offset & ~(align - 1);
3421     }
3422 
3423     if ((offset + bytes) & (align - 1)) {
3424         QEMUIOVector tail_qiov;
3425         struct iovec tail_iov;
3426         size_t tail_bytes;
3427         bool waited;
3428 
3429         mark_request_serialising(&req, align);
3430         waited = wait_serialising_requests(&req);
3431         assert(!waited || !use_local_qiov);
3432 
3433         tail_buf = qemu_blockalign(bs, align);
3434         tail_iov = (struct iovec) {
3435             .iov_base   = tail_buf,
3436             .iov_len    = align,
3437         };
3438         qemu_iovec_init_external(&tail_qiov, &tail_iov, 1);
3439 
3440         BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_TAIL);
3441         ret = bdrv_aligned_preadv(bs, &req, (offset + bytes) & ~(align - 1), align,
3442                                   align, &tail_qiov, 0);
3443         if (ret < 0) {
3444             goto fail;
3445         }
3446         BLKDBG_EVENT(bs, BLKDBG_PWRITEV_RMW_AFTER_TAIL);
3447 
3448         if (!use_local_qiov) {
3449             qemu_iovec_init(&local_qiov, qiov->niov + 1);
3450             qemu_iovec_concat(&local_qiov, qiov, 0, qiov->size);
3451             use_local_qiov = true;
3452         }
3453 
3454         tail_bytes = (offset + bytes) & (align - 1);
3455         qemu_iovec_add(&local_qiov, tail_buf + tail_bytes, align - tail_bytes);
3456 
3457         bytes = ROUND_UP(bytes, align);
3458     }
3459 
3460     ret = bdrv_aligned_pwritev(bs, &req, offset, bytes,
3461                                use_local_qiov ? &local_qiov : qiov,
3462                                flags);
3463 
3464 fail:
3465     tracked_request_end(&req);
3466 
3467     if (use_local_qiov) {
3468         qemu_iovec_destroy(&local_qiov);
3469     }
3470     qemu_vfree(head_buf);
3471     qemu_vfree(tail_buf);
3472 
3473     return ret;
3474 }
3475 
3476 static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs,
3477     int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
3478     BdrvRequestFlags flags)
3479 {
3480     if (nb_sectors < 0 || nb_sectors > (INT_MAX >> BDRV_SECTOR_BITS)) {
3481         return -EINVAL;
3482     }
3483 
3484     return bdrv_co_do_pwritev(bs, sector_num << BDRV_SECTOR_BITS,
3485                               nb_sectors << BDRV_SECTOR_BITS, qiov, flags);
3486 }
3487 
3488 int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num,
3489     int nb_sectors, QEMUIOVector *qiov)
3490 {
3491     trace_bdrv_co_writev(bs, sector_num, nb_sectors);
3492 
3493     return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0);
3494 }
3495 
3496 int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs,
3497                                       int64_t sector_num, int nb_sectors,
3498                                       BdrvRequestFlags flags)
3499 {
3500     trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors, flags);
3501 
3502     if (!(bs->open_flags & BDRV_O_UNMAP)) {
3503         flags &= ~BDRV_REQ_MAY_UNMAP;
3504     }
3505 
3506     return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL,
3507                              BDRV_REQ_ZERO_WRITE | flags);
3508 }
3509 
3510 /**
3511  * Truncate file to 'offset' bytes (needed only for file protocols)
3512  */
3513 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
3514 {
3515     BlockDriver *drv = bs->drv;
3516     int ret;
3517     if (!drv)
3518         return -ENOMEDIUM;
3519     if (!drv->bdrv_truncate)
3520         return -ENOTSUP;
3521     if (bs->read_only)
3522         return -EACCES;
3523     if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_RESIZE, NULL)) {
3524         return -EBUSY;
3525     }
3526     ret = drv->bdrv_truncate(bs, offset);
3527     if (ret == 0) {
3528         ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
3529         bdrv_dev_resize_cb(bs);
3530     }
3531     return ret;
3532 }
3533 
3534 /**
3535  * Length of a allocated file in bytes. Sparse files are counted by actual
3536  * allocated space. Return < 0 if error or unknown.
3537  */
3538 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs)
3539 {
3540     BlockDriver *drv = bs->drv;
3541     if (!drv) {
3542         return -ENOMEDIUM;
3543     }
3544     if (drv->bdrv_get_allocated_file_size) {
3545         return drv->bdrv_get_allocated_file_size(bs);
3546     }
3547     if (bs->file) {
3548         return bdrv_get_allocated_file_size(bs->file);
3549     }
3550     return -ENOTSUP;
3551 }
3552 
3553 /**
3554  * Length of a file in bytes. Return < 0 if error or unknown.
3555  */
3556 int64_t bdrv_getlength(BlockDriverState *bs)
3557 {
3558     BlockDriver *drv = bs->drv;
3559     if (!drv)
3560         return -ENOMEDIUM;
3561 
3562     if (drv->has_variable_length) {
3563         int ret = refresh_total_sectors(bs, bs->total_sectors);
3564         if (ret < 0) {
3565             return ret;
3566         }
3567     }
3568     return bs->total_sectors * BDRV_SECTOR_SIZE;
3569 }
3570 
3571 /* return 0 as number of sectors if no device present or error */
3572 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
3573 {
3574     int64_t length;
3575     length = bdrv_getlength(bs);
3576     if (length < 0)
3577         length = 0;
3578     else
3579         length = length >> BDRV_SECTOR_BITS;
3580     *nb_sectors_ptr = length;
3581 }
3582 
3583 void bdrv_set_on_error(BlockDriverState *bs, BlockdevOnError on_read_error,
3584                        BlockdevOnError on_write_error)
3585 {
3586     bs->on_read_error = on_read_error;
3587     bs->on_write_error = on_write_error;
3588 }
3589 
3590 BlockdevOnError bdrv_get_on_error(BlockDriverState *bs, bool is_read)
3591 {
3592     return is_read ? bs->on_read_error : bs->on_write_error;
3593 }
3594 
3595 BlockErrorAction bdrv_get_error_action(BlockDriverState *bs, bool is_read, int error)
3596 {
3597     BlockdevOnError on_err = is_read ? bs->on_read_error : bs->on_write_error;
3598 
3599     switch (on_err) {
3600     case BLOCKDEV_ON_ERROR_ENOSPC:
3601         return (error == ENOSPC) ?
3602                BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
3603     case BLOCKDEV_ON_ERROR_STOP:
3604         return BLOCK_ERROR_ACTION_STOP;
3605     case BLOCKDEV_ON_ERROR_REPORT:
3606         return BLOCK_ERROR_ACTION_REPORT;
3607     case BLOCKDEV_ON_ERROR_IGNORE:
3608         return BLOCK_ERROR_ACTION_IGNORE;
3609     default:
3610         abort();
3611     }
3612 }
3613 
3614 /* This is done by device models because, while the block layer knows
3615  * about the error, it does not know whether an operation comes from
3616  * the device or the block layer (from a job, for example).
3617  */
3618 void bdrv_error_action(BlockDriverState *bs, BlockErrorAction action,
3619                        bool is_read, int error)
3620 {
3621     assert(error >= 0);
3622 
3623     if (action == BLOCK_ERROR_ACTION_STOP) {
3624         /* First set the iostatus, so that "info block" returns an iostatus
3625          * that matches the events raised so far (an additional error iostatus
3626          * is fine, but not a lost one).
3627          */
3628         bdrv_iostatus_set_err(bs, error);
3629 
3630         /* Then raise the request to stop the VM and the event.
3631          * qemu_system_vmstop_request_prepare has two effects.  First,
3632          * it ensures that the STOP event always comes after the
3633          * BLOCK_IO_ERROR event.  Second, it ensures that even if management
3634          * can observe the STOP event and do a "cont" before the STOP
3635          * event is issued, the VM will not stop.  In this case, vm_start()
3636          * also ensures that the STOP/RESUME pair of events is emitted.
3637          */
3638         qemu_system_vmstop_request_prepare();
3639         qapi_event_send_block_io_error(bdrv_get_device_name(bs),
3640                                        is_read ? IO_OPERATION_TYPE_READ :
3641                                        IO_OPERATION_TYPE_WRITE,
3642                                        action, &error_abort);
3643         qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
3644     } else {
3645         qapi_event_send_block_io_error(bdrv_get_device_name(bs),
3646                                        is_read ? IO_OPERATION_TYPE_READ :
3647                                        IO_OPERATION_TYPE_WRITE,
3648                                        action, &error_abort);
3649     }
3650 }
3651 
3652 int bdrv_is_read_only(BlockDriverState *bs)
3653 {
3654     return bs->read_only;
3655 }
3656 
3657 int bdrv_is_sg(BlockDriverState *bs)
3658 {
3659     return bs->sg;
3660 }
3661 
3662 int bdrv_enable_write_cache(BlockDriverState *bs)
3663 {
3664     return bs->enable_write_cache;
3665 }
3666 
3667 void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce)
3668 {
3669     bs->enable_write_cache = wce;
3670 
3671     /* so a reopen() will preserve wce */
3672     if (wce) {
3673         bs->open_flags |= BDRV_O_CACHE_WB;
3674     } else {
3675         bs->open_flags &= ~BDRV_O_CACHE_WB;
3676     }
3677 }
3678 
3679 int bdrv_is_encrypted(BlockDriverState *bs)
3680 {
3681     if (bs->backing_hd && bs->backing_hd->encrypted)
3682         return 1;
3683     return bs->encrypted;
3684 }
3685 
3686 int bdrv_key_required(BlockDriverState *bs)
3687 {
3688     BlockDriverState *backing_hd = bs->backing_hd;
3689 
3690     if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
3691         return 1;
3692     return (bs->encrypted && !bs->valid_key);
3693 }
3694 
3695 int bdrv_set_key(BlockDriverState *bs, const char *key)
3696 {
3697     int ret;
3698     if (bs->backing_hd && bs->backing_hd->encrypted) {
3699         ret = bdrv_set_key(bs->backing_hd, key);
3700         if (ret < 0)
3701             return ret;
3702         if (!bs->encrypted)
3703             return 0;
3704     }
3705     if (!bs->encrypted) {
3706         return -EINVAL;
3707     } else if (!bs->drv || !bs->drv->bdrv_set_key) {
3708         return -ENOMEDIUM;
3709     }
3710     ret = bs->drv->bdrv_set_key(bs, key);
3711     if (ret < 0) {
3712         bs->valid_key = 0;
3713     } else if (!bs->valid_key) {
3714         bs->valid_key = 1;
3715         /* call the change callback now, we skipped it on open */
3716         bdrv_dev_change_media_cb(bs, true);
3717     }
3718     return ret;
3719 }
3720 
3721 const char *bdrv_get_format_name(BlockDriverState *bs)
3722 {
3723     return bs->drv ? bs->drv->format_name : NULL;
3724 }
3725 
3726 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
3727                          void *opaque)
3728 {
3729     BlockDriver *drv;
3730     int count = 0;
3731     const char **formats = NULL;
3732 
3733     QLIST_FOREACH(drv, &bdrv_drivers, list) {
3734         if (drv->format_name) {
3735             bool found = false;
3736             int i = count;
3737             while (formats && i && !found) {
3738                 found = !strcmp(formats[--i], drv->format_name);
3739             }
3740 
3741             if (!found) {
3742                 formats = g_realloc(formats, (count + 1) * sizeof(char *));
3743                 formats[count++] = drv->format_name;
3744                 it(opaque, drv->format_name);
3745             }
3746         }
3747     }
3748     g_free(formats);
3749 }
3750 
3751 /* This function is to find block backend bs */
3752 BlockDriverState *bdrv_find(const char *name)
3753 {
3754     BlockDriverState *bs;
3755 
3756     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3757         if (!strcmp(name, bs->device_name)) {
3758             return bs;
3759         }
3760     }
3761     return NULL;
3762 }
3763 
3764 /* This function is to find a node in the bs graph */
3765 BlockDriverState *bdrv_find_node(const char *node_name)
3766 {
3767     BlockDriverState *bs;
3768 
3769     assert(node_name);
3770 
3771     QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3772         if (!strcmp(node_name, bs->node_name)) {
3773             return bs;
3774         }
3775     }
3776     return NULL;
3777 }
3778 
3779 /* Put this QMP function here so it can access the static graph_bdrv_states. */
3780 BlockDeviceInfoList *bdrv_named_nodes_list(void)
3781 {
3782     BlockDeviceInfoList *list, *entry;
3783     BlockDriverState *bs;
3784 
3785     list = NULL;
3786     QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) {
3787         entry = g_malloc0(sizeof(*entry));
3788         entry->value = bdrv_block_device_info(bs);
3789         entry->next = list;
3790         list = entry;
3791     }
3792 
3793     return list;
3794 }
3795 
3796 BlockDriverState *bdrv_lookup_bs(const char *device,
3797                                  const char *node_name,
3798                                  Error **errp)
3799 {
3800     BlockDriverState *bs = NULL;
3801 
3802     if (device) {
3803         bs = bdrv_find(device);
3804 
3805         if (bs) {
3806             return bs;
3807         }
3808     }
3809 
3810     if (node_name) {
3811         bs = bdrv_find_node(node_name);
3812 
3813         if (bs) {
3814             return bs;
3815         }
3816     }
3817 
3818     error_setg(errp, "Cannot find device=%s nor node_name=%s",
3819                      device ? device : "",
3820                      node_name ? node_name : "");
3821     return NULL;
3822 }
3823 
3824 BlockDriverState *bdrv_next(BlockDriverState *bs)
3825 {
3826     if (!bs) {
3827         return QTAILQ_FIRST(&bdrv_states);
3828     }
3829     return QTAILQ_NEXT(bs, device_list);
3830 }
3831 
3832 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
3833 {
3834     BlockDriverState *bs;
3835 
3836     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3837         it(opaque, bs);
3838     }
3839 }
3840 
3841 const char *bdrv_get_device_name(BlockDriverState *bs)
3842 {
3843     return bs->device_name;
3844 }
3845 
3846 int bdrv_get_flags(BlockDriverState *bs)
3847 {
3848     return bs->open_flags;
3849 }
3850 
3851 int bdrv_flush_all(void)
3852 {
3853     BlockDriverState *bs;
3854     int result = 0;
3855 
3856     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
3857         AioContext *aio_context = bdrv_get_aio_context(bs);
3858         int ret;
3859 
3860         aio_context_acquire(aio_context);
3861         ret = bdrv_flush(bs);
3862         if (ret < 0 && !result) {
3863             result = ret;
3864         }
3865         aio_context_release(aio_context);
3866     }
3867 
3868     return result;
3869 }
3870 
3871 int bdrv_has_zero_init_1(BlockDriverState *bs)
3872 {
3873     return 1;
3874 }
3875 
3876 int bdrv_has_zero_init(BlockDriverState *bs)
3877 {
3878     assert(bs->drv);
3879 
3880     /* If BS is a copy on write image, it is initialized to
3881        the contents of the base image, which may not be zeroes.  */
3882     if (bs->backing_hd) {
3883         return 0;
3884     }
3885     if (bs->drv->bdrv_has_zero_init) {
3886         return bs->drv->bdrv_has_zero_init(bs);
3887     }
3888 
3889     /* safe default */
3890     return 0;
3891 }
3892 
3893 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs)
3894 {
3895     BlockDriverInfo bdi;
3896 
3897     if (bs->backing_hd) {
3898         return false;
3899     }
3900 
3901     if (bdrv_get_info(bs, &bdi) == 0) {
3902         return bdi.unallocated_blocks_are_zero;
3903     }
3904 
3905     return false;
3906 }
3907 
3908 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs)
3909 {
3910     BlockDriverInfo bdi;
3911 
3912     if (bs->backing_hd || !(bs->open_flags & BDRV_O_UNMAP)) {
3913         return false;
3914     }
3915 
3916     if (bdrv_get_info(bs, &bdi) == 0) {
3917         return bdi.can_write_zeroes_with_unmap;
3918     }
3919 
3920     return false;
3921 }
3922 
3923 typedef struct BdrvCoGetBlockStatusData {
3924     BlockDriverState *bs;
3925     BlockDriverState *base;
3926     int64_t sector_num;
3927     int nb_sectors;
3928     int *pnum;
3929     int64_t ret;
3930     bool done;
3931 } BdrvCoGetBlockStatusData;
3932 
3933 /*
3934  * Returns true iff the specified sector is present in the disk image. Drivers
3935  * not implementing the functionality are assumed to not support backing files,
3936  * hence all their sectors are reported as allocated.
3937  *
3938  * If 'sector_num' is beyond the end of the disk image the return value is 0
3939  * and 'pnum' is set to 0.
3940  *
3941  * 'pnum' is set to the number of sectors (including and immediately following
3942  * the specified sector) that are known to be in the same
3943  * allocated/unallocated state.
3944  *
3945  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
3946  * beyond the end of the disk image it will be clamped.
3947  */
3948 static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs,
3949                                                      int64_t sector_num,
3950                                                      int nb_sectors, int *pnum)
3951 {
3952     int64_t length;
3953     int64_t n;
3954     int64_t ret, ret2;
3955 
3956     length = bdrv_getlength(bs);
3957     if (length < 0) {
3958         return length;
3959     }
3960 
3961     if (sector_num >= (length >> BDRV_SECTOR_BITS)) {
3962         *pnum = 0;
3963         return 0;
3964     }
3965 
3966     n = bs->total_sectors - sector_num;
3967     if (n < nb_sectors) {
3968         nb_sectors = n;
3969     }
3970 
3971     if (!bs->drv->bdrv_co_get_block_status) {
3972         *pnum = nb_sectors;
3973         ret = BDRV_BLOCK_DATA | BDRV_BLOCK_ALLOCATED;
3974         if (bs->drv->protocol_name) {
3975             ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE);
3976         }
3977         return ret;
3978     }
3979 
3980     ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum);
3981     if (ret < 0) {
3982         *pnum = 0;
3983         return ret;
3984     }
3985 
3986     if (ret & BDRV_BLOCK_RAW) {
3987         assert(ret & BDRV_BLOCK_OFFSET_VALID);
3988         return bdrv_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
3989                                      *pnum, pnum);
3990     }
3991 
3992     if (ret & (BDRV_BLOCK_DATA | BDRV_BLOCK_ZERO)) {
3993         ret |= BDRV_BLOCK_ALLOCATED;
3994     }
3995 
3996     if (!(ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO)) {
3997         if (bdrv_unallocated_blocks_are_zero(bs)) {
3998             ret |= BDRV_BLOCK_ZERO;
3999         } else if (bs->backing_hd) {
4000             BlockDriverState *bs2 = bs->backing_hd;
4001             int64_t length2 = bdrv_getlength(bs2);
4002             if (length2 >= 0 && sector_num >= (length2 >> BDRV_SECTOR_BITS)) {
4003                 ret |= BDRV_BLOCK_ZERO;
4004             }
4005         }
4006     }
4007 
4008     if (bs->file &&
4009         (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) &&
4010         (ret & BDRV_BLOCK_OFFSET_VALID)) {
4011         ret2 = bdrv_co_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS,
4012                                         *pnum, pnum);
4013         if (ret2 >= 0) {
4014             /* Ignore errors.  This is just providing extra information, it
4015              * is useful but not necessary.
4016              */
4017             ret |= (ret2 & BDRV_BLOCK_ZERO);
4018         }
4019     }
4020 
4021     return ret;
4022 }
4023 
4024 /* Coroutine wrapper for bdrv_get_block_status() */
4025 static void coroutine_fn bdrv_get_block_status_co_entry(void *opaque)
4026 {
4027     BdrvCoGetBlockStatusData *data = opaque;
4028     BlockDriverState *bs = data->bs;
4029 
4030     data->ret = bdrv_co_get_block_status(bs, data->sector_num, data->nb_sectors,
4031                                          data->pnum);
4032     data->done = true;
4033 }
4034 
4035 /*
4036  * Synchronous wrapper around bdrv_co_get_block_status().
4037  *
4038  * See bdrv_co_get_block_status() for details.
4039  */
4040 int64_t bdrv_get_block_status(BlockDriverState *bs, int64_t sector_num,
4041                               int nb_sectors, int *pnum)
4042 {
4043     Coroutine *co;
4044     BdrvCoGetBlockStatusData data = {
4045         .bs = bs,
4046         .sector_num = sector_num,
4047         .nb_sectors = nb_sectors,
4048         .pnum = pnum,
4049         .done = false,
4050     };
4051 
4052     if (qemu_in_coroutine()) {
4053         /* Fast-path if already in coroutine context */
4054         bdrv_get_block_status_co_entry(&data);
4055     } else {
4056         AioContext *aio_context = bdrv_get_aio_context(bs);
4057 
4058         co = qemu_coroutine_create(bdrv_get_block_status_co_entry);
4059         qemu_coroutine_enter(co, &data);
4060         while (!data.done) {
4061             aio_poll(aio_context, true);
4062         }
4063     }
4064     return data.ret;
4065 }
4066 
4067 int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num,
4068                                    int nb_sectors, int *pnum)
4069 {
4070     int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum);
4071     if (ret < 0) {
4072         return ret;
4073     }
4074     return (ret & BDRV_BLOCK_ALLOCATED);
4075 }
4076 
4077 /*
4078  * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP]
4079  *
4080  * Return true if the given sector is allocated in any image between
4081  * BASE and TOP (inclusive).  BASE can be NULL to check if the given
4082  * sector is allocated in any image of the chain.  Return false otherwise.
4083  *
4084  * 'pnum' is set to the number of sectors (including and immediately following
4085  *  the specified sector) that are known to be in the same
4086  *  allocated/unallocated state.
4087  *
4088  */
4089 int bdrv_is_allocated_above(BlockDriverState *top,
4090                             BlockDriverState *base,
4091                             int64_t sector_num,
4092                             int nb_sectors, int *pnum)
4093 {
4094     BlockDriverState *intermediate;
4095     int ret, n = nb_sectors;
4096 
4097     intermediate = top;
4098     while (intermediate && intermediate != base) {
4099         int pnum_inter;
4100         ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors,
4101                                 &pnum_inter);
4102         if (ret < 0) {
4103             return ret;
4104         } else if (ret) {
4105             *pnum = pnum_inter;
4106             return 1;
4107         }
4108 
4109         /*
4110          * [sector_num, nb_sectors] is unallocated on top but intermediate
4111          * might have
4112          *
4113          * [sector_num+x, nr_sectors] allocated.
4114          */
4115         if (n > pnum_inter &&
4116             (intermediate == top ||
4117              sector_num + pnum_inter < intermediate->total_sectors)) {
4118             n = pnum_inter;
4119         }
4120 
4121         intermediate = intermediate->backing_hd;
4122     }
4123 
4124     *pnum = n;
4125     return 0;
4126 }
4127 
4128 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
4129 {
4130     if (bs->backing_hd && bs->backing_hd->encrypted)
4131         return bs->backing_file;
4132     else if (bs->encrypted)
4133         return bs->filename;
4134     else
4135         return NULL;
4136 }
4137 
4138 void bdrv_get_backing_filename(BlockDriverState *bs,
4139                                char *filename, int filename_size)
4140 {
4141     pstrcpy(filename, filename_size, bs->backing_file);
4142 }
4143 
4144 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
4145                           const uint8_t *buf, int nb_sectors)
4146 {
4147     BlockDriver *drv = bs->drv;
4148     if (!drv)
4149         return -ENOMEDIUM;
4150     if (!drv->bdrv_write_compressed)
4151         return -ENOTSUP;
4152     if (bdrv_check_request(bs, sector_num, nb_sectors))
4153         return -EIO;
4154 
4155     assert(QLIST_EMPTY(&bs->dirty_bitmaps));
4156 
4157     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
4158 }
4159 
4160 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
4161 {
4162     BlockDriver *drv = bs->drv;
4163     if (!drv)
4164         return -ENOMEDIUM;
4165     if (!drv->bdrv_get_info)
4166         return -ENOTSUP;
4167     memset(bdi, 0, sizeof(*bdi));
4168     return drv->bdrv_get_info(bs, bdi);
4169 }
4170 
4171 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
4172 {
4173     BlockDriver *drv = bs->drv;
4174     if (drv && drv->bdrv_get_specific_info) {
4175         return drv->bdrv_get_specific_info(bs);
4176     }
4177     return NULL;
4178 }
4179 
4180 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
4181                       int64_t pos, int size)
4182 {
4183     QEMUIOVector qiov;
4184     struct iovec iov = {
4185         .iov_base   = (void *) buf,
4186         .iov_len    = size,
4187     };
4188 
4189     qemu_iovec_init_external(&qiov, &iov, 1);
4190     return bdrv_writev_vmstate(bs, &qiov, pos);
4191 }
4192 
4193 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos)
4194 {
4195     BlockDriver *drv = bs->drv;
4196 
4197     if (!drv) {
4198         return -ENOMEDIUM;
4199     } else if (drv->bdrv_save_vmstate) {
4200         return drv->bdrv_save_vmstate(bs, qiov, pos);
4201     } else if (bs->file) {
4202         return bdrv_writev_vmstate(bs->file, qiov, pos);
4203     }
4204 
4205     return -ENOTSUP;
4206 }
4207 
4208 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
4209                       int64_t pos, int size)
4210 {
4211     BlockDriver *drv = bs->drv;
4212     if (!drv)
4213         return -ENOMEDIUM;
4214     if (drv->bdrv_load_vmstate)
4215         return drv->bdrv_load_vmstate(bs, buf, pos, size);
4216     if (bs->file)
4217         return bdrv_load_vmstate(bs->file, buf, pos, size);
4218     return -ENOTSUP;
4219 }
4220 
4221 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
4222 {
4223     if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
4224         return;
4225     }
4226 
4227     bs->drv->bdrv_debug_event(bs, event);
4228 }
4229 
4230 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
4231                           const char *tag)
4232 {
4233     while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) {
4234         bs = bs->file;
4235     }
4236 
4237     if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) {
4238         return bs->drv->bdrv_debug_breakpoint(bs, event, tag);
4239     }
4240 
4241     return -ENOTSUP;
4242 }
4243 
4244 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag)
4245 {
4246     while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) {
4247         bs = bs->file;
4248     }
4249 
4250     if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) {
4251         return bs->drv->bdrv_debug_remove_breakpoint(bs, tag);
4252     }
4253 
4254     return -ENOTSUP;
4255 }
4256 
4257 int bdrv_debug_resume(BlockDriverState *bs, const char *tag)
4258 {
4259     while (bs && (!bs->drv || !bs->drv->bdrv_debug_resume)) {
4260         bs = bs->file;
4261     }
4262 
4263     if (bs && bs->drv && bs->drv->bdrv_debug_resume) {
4264         return bs->drv->bdrv_debug_resume(bs, tag);
4265     }
4266 
4267     return -ENOTSUP;
4268 }
4269 
4270 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag)
4271 {
4272     while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) {
4273         bs = bs->file;
4274     }
4275 
4276     if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) {
4277         return bs->drv->bdrv_debug_is_suspended(bs, tag);
4278     }
4279 
4280     return false;
4281 }
4282 
4283 int bdrv_is_snapshot(BlockDriverState *bs)
4284 {
4285     return !!(bs->open_flags & BDRV_O_SNAPSHOT);
4286 }
4287 
4288 /* backing_file can either be relative, or absolute, or a protocol.  If it is
4289  * relative, it must be relative to the chain.  So, passing in bs->filename
4290  * from a BDS as backing_file should not be done, as that may be relative to
4291  * the CWD rather than the chain. */
4292 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
4293         const char *backing_file)
4294 {
4295     char *filename_full = NULL;
4296     char *backing_file_full = NULL;
4297     char *filename_tmp = NULL;
4298     int is_protocol = 0;
4299     BlockDriverState *curr_bs = NULL;
4300     BlockDriverState *retval = NULL;
4301 
4302     if (!bs || !bs->drv || !backing_file) {
4303         return NULL;
4304     }
4305 
4306     filename_full     = g_malloc(PATH_MAX);
4307     backing_file_full = g_malloc(PATH_MAX);
4308     filename_tmp      = g_malloc(PATH_MAX);
4309 
4310     is_protocol = path_has_protocol(backing_file);
4311 
4312     for (curr_bs = bs; curr_bs->backing_hd; curr_bs = curr_bs->backing_hd) {
4313 
4314         /* If either of the filename paths is actually a protocol, then
4315          * compare unmodified paths; otherwise make paths relative */
4316         if (is_protocol || path_has_protocol(curr_bs->backing_file)) {
4317             if (strcmp(backing_file, curr_bs->backing_file) == 0) {
4318                 retval = curr_bs->backing_hd;
4319                 break;
4320             }
4321         } else {
4322             /* If not an absolute filename path, make it relative to the current
4323              * image's filename path */
4324             path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4325                          backing_file);
4326 
4327             /* We are going to compare absolute pathnames */
4328             if (!realpath(filename_tmp, filename_full)) {
4329                 continue;
4330             }
4331 
4332             /* We need to make sure the backing filename we are comparing against
4333              * is relative to the current image filename (or absolute) */
4334             path_combine(filename_tmp, PATH_MAX, curr_bs->filename,
4335                          curr_bs->backing_file);
4336 
4337             if (!realpath(filename_tmp, backing_file_full)) {
4338                 continue;
4339             }
4340 
4341             if (strcmp(backing_file_full, filename_full) == 0) {
4342                 retval = curr_bs->backing_hd;
4343                 break;
4344             }
4345         }
4346     }
4347 
4348     g_free(filename_full);
4349     g_free(backing_file_full);
4350     g_free(filename_tmp);
4351     return retval;
4352 }
4353 
4354 int bdrv_get_backing_file_depth(BlockDriverState *bs)
4355 {
4356     if (!bs->drv) {
4357         return 0;
4358     }
4359 
4360     if (!bs->backing_hd) {
4361         return 0;
4362     }
4363 
4364     return 1 + bdrv_get_backing_file_depth(bs->backing_hd);
4365 }
4366 
4367 BlockDriverState *bdrv_find_base(BlockDriverState *bs)
4368 {
4369     BlockDriverState *curr_bs = NULL;
4370 
4371     if (!bs) {
4372         return NULL;
4373     }
4374 
4375     curr_bs = bs;
4376 
4377     while (curr_bs->backing_hd) {
4378         curr_bs = curr_bs->backing_hd;
4379     }
4380     return curr_bs;
4381 }
4382 
4383 /**************************************************************/
4384 /* async I/Os */
4385 
4386 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
4387                                  QEMUIOVector *qiov, int nb_sectors,
4388                                  BlockDriverCompletionFunc *cb, void *opaque)
4389 {
4390     trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque);
4391 
4392     return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
4393                                  cb, opaque, false);
4394 }
4395 
4396 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
4397                                   QEMUIOVector *qiov, int nb_sectors,
4398                                   BlockDriverCompletionFunc *cb, void *opaque)
4399 {
4400     trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque);
4401 
4402     return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0,
4403                                  cb, opaque, true);
4404 }
4405 
4406 BlockDriverAIOCB *bdrv_aio_write_zeroes(BlockDriverState *bs,
4407         int64_t sector_num, int nb_sectors, BdrvRequestFlags flags,
4408         BlockDriverCompletionFunc *cb, void *opaque)
4409 {
4410     trace_bdrv_aio_write_zeroes(bs, sector_num, nb_sectors, flags, opaque);
4411 
4412     return bdrv_co_aio_rw_vector(bs, sector_num, NULL, nb_sectors,
4413                                  BDRV_REQ_ZERO_WRITE | flags,
4414                                  cb, opaque, true);
4415 }
4416 
4417 
4418 typedef struct MultiwriteCB {
4419     int error;
4420     int num_requests;
4421     int num_callbacks;
4422     struct {
4423         BlockDriverCompletionFunc *cb;
4424         void *opaque;
4425         QEMUIOVector *free_qiov;
4426     } callbacks[];
4427 } MultiwriteCB;
4428 
4429 static void multiwrite_user_cb(MultiwriteCB *mcb)
4430 {
4431     int i;
4432 
4433     for (i = 0; i < mcb->num_callbacks; i++) {
4434         mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
4435         if (mcb->callbacks[i].free_qiov) {
4436             qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
4437         }
4438         g_free(mcb->callbacks[i].free_qiov);
4439     }
4440 }
4441 
4442 static void multiwrite_cb(void *opaque, int ret)
4443 {
4444     MultiwriteCB *mcb = opaque;
4445 
4446     trace_multiwrite_cb(mcb, ret);
4447 
4448     if (ret < 0 && !mcb->error) {
4449         mcb->error = ret;
4450     }
4451 
4452     mcb->num_requests--;
4453     if (mcb->num_requests == 0) {
4454         multiwrite_user_cb(mcb);
4455         g_free(mcb);
4456     }
4457 }
4458 
4459 static int multiwrite_req_compare(const void *a, const void *b)
4460 {
4461     const BlockRequest *req1 = a, *req2 = b;
4462 
4463     /*
4464      * Note that we can't simply subtract req2->sector from req1->sector
4465      * here as that could overflow the return value.
4466      */
4467     if (req1->sector > req2->sector) {
4468         return 1;
4469     } else if (req1->sector < req2->sector) {
4470         return -1;
4471     } else {
4472         return 0;
4473     }
4474 }
4475 
4476 /*
4477  * Takes a bunch of requests and tries to merge them. Returns the number of
4478  * requests that remain after merging.
4479  */
4480 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
4481     int num_reqs, MultiwriteCB *mcb)
4482 {
4483     int i, outidx;
4484 
4485     // Sort requests by start sector
4486     qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
4487 
4488     // Check if adjacent requests touch the same clusters. If so, combine them,
4489     // filling up gaps with zero sectors.
4490     outidx = 0;
4491     for (i = 1; i < num_reqs; i++) {
4492         int merge = 0;
4493         int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
4494 
4495         // Handle exactly sequential writes and overlapping writes.
4496         if (reqs[i].sector <= oldreq_last) {
4497             merge = 1;
4498         }
4499 
4500         if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
4501             merge = 0;
4502         }
4503 
4504         if (merge) {
4505             size_t size;
4506             QEMUIOVector *qiov = g_malloc0(sizeof(*qiov));
4507             qemu_iovec_init(qiov,
4508                 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
4509 
4510             // Add the first request to the merged one. If the requests are
4511             // overlapping, drop the last sectors of the first request.
4512             size = (reqs[i].sector - reqs[outidx].sector) << 9;
4513             qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size);
4514 
4515             // We should need to add any zeros between the two requests
4516             assert (reqs[i].sector <= oldreq_last);
4517 
4518             // Add the second request
4519             qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size);
4520 
4521             reqs[outidx].nb_sectors = qiov->size >> 9;
4522             reqs[outidx].qiov = qiov;
4523 
4524             mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
4525         } else {
4526             outidx++;
4527             reqs[outidx].sector     = reqs[i].sector;
4528             reqs[outidx].nb_sectors = reqs[i].nb_sectors;
4529             reqs[outidx].qiov       = reqs[i].qiov;
4530         }
4531     }
4532 
4533     return outidx + 1;
4534 }
4535 
4536 /*
4537  * Submit multiple AIO write requests at once.
4538  *
4539  * On success, the function returns 0 and all requests in the reqs array have
4540  * been submitted. In error case this function returns -1, and any of the
4541  * requests may or may not be submitted yet. In particular, this means that the
4542  * callback will be called for some of the requests, for others it won't. The
4543  * caller must check the error field of the BlockRequest to wait for the right
4544  * callbacks (if error != 0, no callback will be called).
4545  *
4546  * The implementation may modify the contents of the reqs array, e.g. to merge
4547  * requests. However, the fields opaque and error are left unmodified as they
4548  * are used to signal failure for a single request to the caller.
4549  */
4550 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
4551 {
4552     MultiwriteCB *mcb;
4553     int i;
4554 
4555     /* don't submit writes if we don't have a medium */
4556     if (bs->drv == NULL) {
4557         for (i = 0; i < num_reqs; i++) {
4558             reqs[i].error = -ENOMEDIUM;
4559         }
4560         return -1;
4561     }
4562 
4563     if (num_reqs == 0) {
4564         return 0;
4565     }
4566 
4567     // Create MultiwriteCB structure
4568     mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
4569     mcb->num_requests = 0;
4570     mcb->num_callbacks = num_reqs;
4571 
4572     for (i = 0; i < num_reqs; i++) {
4573         mcb->callbacks[i].cb = reqs[i].cb;
4574         mcb->callbacks[i].opaque = reqs[i].opaque;
4575     }
4576 
4577     // Check for mergable requests
4578     num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
4579 
4580     trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs);
4581 
4582     /* Run the aio requests. */
4583     mcb->num_requests = num_reqs;
4584     for (i = 0; i < num_reqs; i++) {
4585         bdrv_co_aio_rw_vector(bs, reqs[i].sector, reqs[i].qiov,
4586                               reqs[i].nb_sectors, reqs[i].flags,
4587                               multiwrite_cb, mcb,
4588                               true);
4589     }
4590 
4591     return 0;
4592 }
4593 
4594 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
4595 {
4596     acb->aiocb_info->cancel(acb);
4597 }
4598 
4599 /**************************************************************/
4600 /* async block device emulation */
4601 
4602 typedef struct BlockDriverAIOCBSync {
4603     BlockDriverAIOCB common;
4604     QEMUBH *bh;
4605     int ret;
4606     /* vector translation state */
4607     QEMUIOVector *qiov;
4608     uint8_t *bounce;
4609     int is_write;
4610 } BlockDriverAIOCBSync;
4611 
4612 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
4613 {
4614     BlockDriverAIOCBSync *acb =
4615         container_of(blockacb, BlockDriverAIOCBSync, common);
4616     qemu_bh_delete(acb->bh);
4617     acb->bh = NULL;
4618     qemu_aio_release(acb);
4619 }
4620 
4621 static const AIOCBInfo bdrv_em_aiocb_info = {
4622     .aiocb_size         = sizeof(BlockDriverAIOCBSync),
4623     .cancel             = bdrv_aio_cancel_em,
4624 };
4625 
4626 static void bdrv_aio_bh_cb(void *opaque)
4627 {
4628     BlockDriverAIOCBSync *acb = opaque;
4629 
4630     if (!acb->is_write)
4631         qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size);
4632     qemu_vfree(acb->bounce);
4633     acb->common.cb(acb->common.opaque, acb->ret);
4634     qemu_bh_delete(acb->bh);
4635     acb->bh = NULL;
4636     qemu_aio_release(acb);
4637 }
4638 
4639 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
4640                                             int64_t sector_num,
4641                                             QEMUIOVector *qiov,
4642                                             int nb_sectors,
4643                                             BlockDriverCompletionFunc *cb,
4644                                             void *opaque,
4645                                             int is_write)
4646 
4647 {
4648     BlockDriverAIOCBSync *acb;
4649 
4650     acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque);
4651     acb->is_write = is_write;
4652     acb->qiov = qiov;
4653     acb->bounce = qemu_blockalign(bs, qiov->size);
4654     acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_aio_bh_cb, acb);
4655 
4656     if (is_write) {
4657         qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size);
4658         acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
4659     } else {
4660         acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
4661     }
4662 
4663     qemu_bh_schedule(acb->bh);
4664 
4665     return &acb->common;
4666 }
4667 
4668 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
4669         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4670         BlockDriverCompletionFunc *cb, void *opaque)
4671 {
4672     return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
4673 }
4674 
4675 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
4676         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
4677         BlockDriverCompletionFunc *cb, void *opaque)
4678 {
4679     return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
4680 }
4681 
4682 
4683 typedef struct BlockDriverAIOCBCoroutine {
4684     BlockDriverAIOCB common;
4685     BlockRequest req;
4686     bool is_write;
4687     bool *done;
4688     QEMUBH* bh;
4689 } BlockDriverAIOCBCoroutine;
4690 
4691 static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb)
4692 {
4693     AioContext *aio_context = bdrv_get_aio_context(blockacb->bs);
4694     BlockDriverAIOCBCoroutine *acb =
4695         container_of(blockacb, BlockDriverAIOCBCoroutine, common);
4696     bool done = false;
4697 
4698     acb->done = &done;
4699     while (!done) {
4700         aio_poll(aio_context, true);
4701     }
4702 }
4703 
4704 static const AIOCBInfo bdrv_em_co_aiocb_info = {
4705     .aiocb_size         = sizeof(BlockDriverAIOCBCoroutine),
4706     .cancel             = bdrv_aio_co_cancel_em,
4707 };
4708 
4709 static void bdrv_co_em_bh(void *opaque)
4710 {
4711     BlockDriverAIOCBCoroutine *acb = opaque;
4712 
4713     acb->common.cb(acb->common.opaque, acb->req.error);
4714 
4715     if (acb->done) {
4716         *acb->done = true;
4717     }
4718 
4719     qemu_bh_delete(acb->bh);
4720     qemu_aio_release(acb);
4721 }
4722 
4723 /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */
4724 static void coroutine_fn bdrv_co_do_rw(void *opaque)
4725 {
4726     BlockDriverAIOCBCoroutine *acb = opaque;
4727     BlockDriverState *bs = acb->common.bs;
4728 
4729     if (!acb->is_write) {
4730         acb->req.error = bdrv_co_do_readv(bs, acb->req.sector,
4731             acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4732     } else {
4733         acb->req.error = bdrv_co_do_writev(bs, acb->req.sector,
4734             acb->req.nb_sectors, acb->req.qiov, acb->req.flags);
4735     }
4736 
4737     acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
4738     qemu_bh_schedule(acb->bh);
4739 }
4740 
4741 static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs,
4742                                                int64_t sector_num,
4743                                                QEMUIOVector *qiov,
4744                                                int nb_sectors,
4745                                                BdrvRequestFlags flags,
4746                                                BlockDriverCompletionFunc *cb,
4747                                                void *opaque,
4748                                                bool is_write)
4749 {
4750     Coroutine *co;
4751     BlockDriverAIOCBCoroutine *acb;
4752 
4753     acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4754     acb->req.sector = sector_num;
4755     acb->req.nb_sectors = nb_sectors;
4756     acb->req.qiov = qiov;
4757     acb->req.flags = flags;
4758     acb->is_write = is_write;
4759     acb->done = NULL;
4760 
4761     co = qemu_coroutine_create(bdrv_co_do_rw);
4762     qemu_coroutine_enter(co, acb);
4763 
4764     return &acb->common;
4765 }
4766 
4767 static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque)
4768 {
4769     BlockDriverAIOCBCoroutine *acb = opaque;
4770     BlockDriverState *bs = acb->common.bs;
4771 
4772     acb->req.error = bdrv_co_flush(bs);
4773     acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
4774     qemu_bh_schedule(acb->bh);
4775 }
4776 
4777 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
4778         BlockDriverCompletionFunc *cb, void *opaque)
4779 {
4780     trace_bdrv_aio_flush(bs, opaque);
4781 
4782     Coroutine *co;
4783     BlockDriverAIOCBCoroutine *acb;
4784 
4785     acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4786     acb->done = NULL;
4787 
4788     co = qemu_coroutine_create(bdrv_aio_flush_co_entry);
4789     qemu_coroutine_enter(co, acb);
4790 
4791     return &acb->common;
4792 }
4793 
4794 static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque)
4795 {
4796     BlockDriverAIOCBCoroutine *acb = opaque;
4797     BlockDriverState *bs = acb->common.bs;
4798 
4799     acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors);
4800     acb->bh = aio_bh_new(bdrv_get_aio_context(bs), bdrv_co_em_bh, acb);
4801     qemu_bh_schedule(acb->bh);
4802 }
4803 
4804 BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs,
4805         int64_t sector_num, int nb_sectors,
4806         BlockDriverCompletionFunc *cb, void *opaque)
4807 {
4808     Coroutine *co;
4809     BlockDriverAIOCBCoroutine *acb;
4810 
4811     trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque);
4812 
4813     acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque);
4814     acb->req.sector = sector_num;
4815     acb->req.nb_sectors = nb_sectors;
4816     acb->done = NULL;
4817     co = qemu_coroutine_create(bdrv_aio_discard_co_entry);
4818     qemu_coroutine_enter(co, acb);
4819 
4820     return &acb->common;
4821 }
4822 
4823 void bdrv_init(void)
4824 {
4825     module_call_init(MODULE_INIT_BLOCK);
4826 }
4827 
4828 void bdrv_init_with_whitelist(void)
4829 {
4830     use_bdrv_whitelist = 1;
4831     bdrv_init();
4832 }
4833 
4834 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
4835                    BlockDriverCompletionFunc *cb, void *opaque)
4836 {
4837     BlockDriverAIOCB *acb;
4838 
4839     acb = g_slice_alloc(aiocb_info->aiocb_size);
4840     acb->aiocb_info = aiocb_info;
4841     acb->bs = bs;
4842     acb->cb = cb;
4843     acb->opaque = opaque;
4844     return acb;
4845 }
4846 
4847 void qemu_aio_release(void *p)
4848 {
4849     BlockDriverAIOCB *acb = p;
4850     g_slice_free1(acb->aiocb_info->aiocb_size, acb);
4851 }
4852 
4853 /**************************************************************/
4854 /* Coroutine block device emulation */
4855 
4856 typedef struct CoroutineIOCompletion {
4857     Coroutine *coroutine;
4858     int ret;
4859 } CoroutineIOCompletion;
4860 
4861 static void bdrv_co_io_em_complete(void *opaque, int ret)
4862 {
4863     CoroutineIOCompletion *co = opaque;
4864 
4865     co->ret = ret;
4866     qemu_coroutine_enter(co->coroutine, NULL);
4867 }
4868 
4869 static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num,
4870                                       int nb_sectors, QEMUIOVector *iov,
4871                                       bool is_write)
4872 {
4873     CoroutineIOCompletion co = {
4874         .coroutine = qemu_coroutine_self(),
4875     };
4876     BlockDriverAIOCB *acb;
4877 
4878     if (is_write) {
4879         acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors,
4880                                        bdrv_co_io_em_complete, &co);
4881     } else {
4882         acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors,
4883                                       bdrv_co_io_em_complete, &co);
4884     }
4885 
4886     trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb);
4887     if (!acb) {
4888         return -EIO;
4889     }
4890     qemu_coroutine_yield();
4891 
4892     return co.ret;
4893 }
4894 
4895 static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs,
4896                                          int64_t sector_num, int nb_sectors,
4897                                          QEMUIOVector *iov)
4898 {
4899     return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false);
4900 }
4901 
4902 static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs,
4903                                          int64_t sector_num, int nb_sectors,
4904                                          QEMUIOVector *iov)
4905 {
4906     return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true);
4907 }
4908 
4909 static void coroutine_fn bdrv_flush_co_entry(void *opaque)
4910 {
4911     RwCo *rwco = opaque;
4912 
4913     rwco->ret = bdrv_co_flush(rwco->bs);
4914 }
4915 
4916 int coroutine_fn bdrv_co_flush(BlockDriverState *bs)
4917 {
4918     int ret;
4919 
4920     if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
4921         return 0;
4922     }
4923 
4924     /* Write back cached data to the OS even with cache=unsafe */
4925     BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS);
4926     if (bs->drv->bdrv_co_flush_to_os) {
4927         ret = bs->drv->bdrv_co_flush_to_os(bs);
4928         if (ret < 0) {
4929             return ret;
4930         }
4931     }
4932 
4933     /* But don't actually force it to the disk with cache=unsafe */
4934     if (bs->open_flags & BDRV_O_NO_FLUSH) {
4935         goto flush_parent;
4936     }
4937 
4938     BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK);
4939     if (bs->drv->bdrv_co_flush_to_disk) {
4940         ret = bs->drv->bdrv_co_flush_to_disk(bs);
4941     } else if (bs->drv->bdrv_aio_flush) {
4942         BlockDriverAIOCB *acb;
4943         CoroutineIOCompletion co = {
4944             .coroutine = qemu_coroutine_self(),
4945         };
4946 
4947         acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co);
4948         if (acb == NULL) {
4949             ret = -EIO;
4950         } else {
4951             qemu_coroutine_yield();
4952             ret = co.ret;
4953         }
4954     } else {
4955         /*
4956          * Some block drivers always operate in either writethrough or unsafe
4957          * mode and don't support bdrv_flush therefore. Usually qemu doesn't
4958          * know how the server works (because the behaviour is hardcoded or
4959          * depends on server-side configuration), so we can't ensure that
4960          * everything is safe on disk. Returning an error doesn't work because
4961          * that would break guests even if the server operates in writethrough
4962          * mode.
4963          *
4964          * Let's hope the user knows what he's doing.
4965          */
4966         ret = 0;
4967     }
4968     if (ret < 0) {
4969         return ret;
4970     }
4971 
4972     /* Now flush the underlying protocol.  It will also have BDRV_O_NO_FLUSH
4973      * in the case of cache=unsafe, so there are no useless flushes.
4974      */
4975 flush_parent:
4976     return bdrv_co_flush(bs->file);
4977 }
4978 
4979 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp)
4980 {
4981     Error *local_err = NULL;
4982     int ret;
4983 
4984     if (!bs->drv)  {
4985         return;
4986     }
4987 
4988     if (bs->drv->bdrv_invalidate_cache) {
4989         bs->drv->bdrv_invalidate_cache(bs, &local_err);
4990     } else if (bs->file) {
4991         bdrv_invalidate_cache(bs->file, &local_err);
4992     }
4993     if (local_err) {
4994         error_propagate(errp, local_err);
4995         return;
4996     }
4997 
4998     ret = refresh_total_sectors(bs, bs->total_sectors);
4999     if (ret < 0) {
5000         error_setg_errno(errp, -ret, "Could not refresh total sector count");
5001         return;
5002     }
5003 }
5004 
5005 void bdrv_invalidate_cache_all(Error **errp)
5006 {
5007     BlockDriverState *bs;
5008     Error *local_err = NULL;
5009 
5010     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
5011         AioContext *aio_context = bdrv_get_aio_context(bs);
5012 
5013         aio_context_acquire(aio_context);
5014         bdrv_invalidate_cache(bs, &local_err);
5015         aio_context_release(aio_context);
5016         if (local_err) {
5017             error_propagate(errp, local_err);
5018             return;
5019         }
5020     }
5021 }
5022 
5023 void bdrv_clear_incoming_migration_all(void)
5024 {
5025     BlockDriverState *bs;
5026 
5027     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
5028         AioContext *aio_context = bdrv_get_aio_context(bs);
5029 
5030         aio_context_acquire(aio_context);
5031         bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING);
5032         aio_context_release(aio_context);
5033     }
5034 }
5035 
5036 int bdrv_flush(BlockDriverState *bs)
5037 {
5038     Coroutine *co;
5039     RwCo rwco = {
5040         .bs = bs,
5041         .ret = NOT_DONE,
5042     };
5043 
5044     if (qemu_in_coroutine()) {
5045         /* Fast-path if already in coroutine context */
5046         bdrv_flush_co_entry(&rwco);
5047     } else {
5048         AioContext *aio_context = bdrv_get_aio_context(bs);
5049 
5050         co = qemu_coroutine_create(bdrv_flush_co_entry);
5051         qemu_coroutine_enter(co, &rwco);
5052         while (rwco.ret == NOT_DONE) {
5053             aio_poll(aio_context, true);
5054         }
5055     }
5056 
5057     return rwco.ret;
5058 }
5059 
5060 typedef struct DiscardCo {
5061     BlockDriverState *bs;
5062     int64_t sector_num;
5063     int nb_sectors;
5064     int ret;
5065 } DiscardCo;
5066 static void coroutine_fn bdrv_discard_co_entry(void *opaque)
5067 {
5068     DiscardCo *rwco = opaque;
5069 
5070     rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors);
5071 }
5072 
5073 /* if no limit is specified in the BlockLimits use a default
5074  * of 32768 512-byte sectors (16 MiB) per request.
5075  */
5076 #define MAX_DISCARD_DEFAULT 32768
5077 
5078 int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num,
5079                                  int nb_sectors)
5080 {
5081     int max_discard;
5082 
5083     if (!bs->drv) {
5084         return -ENOMEDIUM;
5085     } else if (bdrv_check_request(bs, sector_num, nb_sectors)) {
5086         return -EIO;
5087     } else if (bs->read_only) {
5088         return -EROFS;
5089     }
5090 
5091     bdrv_reset_dirty(bs, sector_num, nb_sectors);
5092 
5093     /* Do nothing if disabled.  */
5094     if (!(bs->open_flags & BDRV_O_UNMAP)) {
5095         return 0;
5096     }
5097 
5098     if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) {
5099         return 0;
5100     }
5101 
5102     max_discard = bs->bl.max_discard ?  bs->bl.max_discard : MAX_DISCARD_DEFAULT;
5103     while (nb_sectors > 0) {
5104         int ret;
5105         int num = nb_sectors;
5106 
5107         /* align request */
5108         if (bs->bl.discard_alignment &&
5109             num >= bs->bl.discard_alignment &&
5110             sector_num % bs->bl.discard_alignment) {
5111             if (num > bs->bl.discard_alignment) {
5112                 num = bs->bl.discard_alignment;
5113             }
5114             num -= sector_num % bs->bl.discard_alignment;
5115         }
5116 
5117         /* limit request size */
5118         if (num > max_discard) {
5119             num = max_discard;
5120         }
5121 
5122         if (bs->drv->bdrv_co_discard) {
5123             ret = bs->drv->bdrv_co_discard(bs, sector_num, num);
5124         } else {
5125             BlockDriverAIOCB *acb;
5126             CoroutineIOCompletion co = {
5127                 .coroutine = qemu_coroutine_self(),
5128             };
5129 
5130             acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors,
5131                                             bdrv_co_io_em_complete, &co);
5132             if (acb == NULL) {
5133                 return -EIO;
5134             } else {
5135                 qemu_coroutine_yield();
5136                 ret = co.ret;
5137             }
5138         }
5139         if (ret && ret != -ENOTSUP) {
5140             return ret;
5141         }
5142 
5143         sector_num += num;
5144         nb_sectors -= num;
5145     }
5146     return 0;
5147 }
5148 
5149 int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
5150 {
5151     Coroutine *co;
5152     DiscardCo rwco = {
5153         .bs = bs,
5154         .sector_num = sector_num,
5155         .nb_sectors = nb_sectors,
5156         .ret = NOT_DONE,
5157     };
5158 
5159     if (qemu_in_coroutine()) {
5160         /* Fast-path if already in coroutine context */
5161         bdrv_discard_co_entry(&rwco);
5162     } else {
5163         AioContext *aio_context = bdrv_get_aio_context(bs);
5164 
5165         co = qemu_coroutine_create(bdrv_discard_co_entry);
5166         qemu_coroutine_enter(co, &rwco);
5167         while (rwco.ret == NOT_DONE) {
5168             aio_poll(aio_context, true);
5169         }
5170     }
5171 
5172     return rwco.ret;
5173 }
5174 
5175 /**************************************************************/
5176 /* removable device support */
5177 
5178 /**
5179  * Return TRUE if the media is present
5180  */
5181 int bdrv_is_inserted(BlockDriverState *bs)
5182 {
5183     BlockDriver *drv = bs->drv;
5184 
5185     if (!drv)
5186         return 0;
5187     if (!drv->bdrv_is_inserted)
5188         return 1;
5189     return drv->bdrv_is_inserted(bs);
5190 }
5191 
5192 /**
5193  * Return whether the media changed since the last call to this
5194  * function, or -ENOTSUP if we don't know.  Most drivers don't know.
5195  */
5196 int bdrv_media_changed(BlockDriverState *bs)
5197 {
5198     BlockDriver *drv = bs->drv;
5199 
5200     if (drv && drv->bdrv_media_changed) {
5201         return drv->bdrv_media_changed(bs);
5202     }
5203     return -ENOTSUP;
5204 }
5205 
5206 /**
5207  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
5208  */
5209 void bdrv_eject(BlockDriverState *bs, bool eject_flag)
5210 {
5211     BlockDriver *drv = bs->drv;
5212 
5213     if (drv && drv->bdrv_eject) {
5214         drv->bdrv_eject(bs, eject_flag);
5215     }
5216 
5217     if (bs->device_name[0] != '\0') {
5218         qapi_event_send_device_tray_moved(bdrv_get_device_name(bs),
5219                                           eject_flag, &error_abort);
5220     }
5221 }
5222 
5223 /**
5224  * Lock or unlock the media (if it is locked, the user won't be able
5225  * to eject it manually).
5226  */
5227 void bdrv_lock_medium(BlockDriverState *bs, bool locked)
5228 {
5229     BlockDriver *drv = bs->drv;
5230 
5231     trace_bdrv_lock_medium(bs, locked);
5232 
5233     if (drv && drv->bdrv_lock_medium) {
5234         drv->bdrv_lock_medium(bs, locked);
5235     }
5236 }
5237 
5238 /* needed for generic scsi interface */
5239 
5240 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
5241 {
5242     BlockDriver *drv = bs->drv;
5243 
5244     if (drv && drv->bdrv_ioctl)
5245         return drv->bdrv_ioctl(bs, req, buf);
5246     return -ENOTSUP;
5247 }
5248 
5249 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
5250         unsigned long int req, void *buf,
5251         BlockDriverCompletionFunc *cb, void *opaque)
5252 {
5253     BlockDriver *drv = bs->drv;
5254 
5255     if (drv && drv->bdrv_aio_ioctl)
5256         return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
5257     return NULL;
5258 }
5259 
5260 void bdrv_set_guest_block_size(BlockDriverState *bs, int align)
5261 {
5262     bs->guest_block_size = align;
5263 }
5264 
5265 void *qemu_blockalign(BlockDriverState *bs, size_t size)
5266 {
5267     return qemu_memalign(bdrv_opt_mem_align(bs), size);
5268 }
5269 
5270 /*
5271  * Check if all memory in this vector is sector aligned.
5272  */
5273 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
5274 {
5275     int i;
5276     size_t alignment = bdrv_opt_mem_align(bs);
5277 
5278     for (i = 0; i < qiov->niov; i++) {
5279         if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
5280             return false;
5281         }
5282         if (qiov->iov[i].iov_len % alignment) {
5283             return false;
5284         }
5285     }
5286 
5287     return true;
5288 }
5289 
5290 BdrvDirtyBitmap *bdrv_create_dirty_bitmap(BlockDriverState *bs, int granularity,
5291                                           Error **errp)
5292 {
5293     int64_t bitmap_size;
5294     BdrvDirtyBitmap *bitmap;
5295 
5296     assert((granularity & (granularity - 1)) == 0);
5297 
5298     granularity >>= BDRV_SECTOR_BITS;
5299     assert(granularity);
5300     bitmap_size = bdrv_getlength(bs);
5301     if (bitmap_size < 0) {
5302         error_setg_errno(errp, -bitmap_size, "could not get length of device");
5303         errno = -bitmap_size;
5304         return NULL;
5305     }
5306     bitmap_size >>= BDRV_SECTOR_BITS;
5307     bitmap = g_malloc0(sizeof(BdrvDirtyBitmap));
5308     bitmap->bitmap = hbitmap_alloc(bitmap_size, ffs(granularity) - 1);
5309     QLIST_INSERT_HEAD(&bs->dirty_bitmaps, bitmap, list);
5310     return bitmap;
5311 }
5312 
5313 void bdrv_release_dirty_bitmap(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
5314 {
5315     BdrvDirtyBitmap *bm, *next;
5316     QLIST_FOREACH_SAFE(bm, &bs->dirty_bitmaps, list, next) {
5317         if (bm == bitmap) {
5318             QLIST_REMOVE(bitmap, list);
5319             hbitmap_free(bitmap->bitmap);
5320             g_free(bitmap);
5321             return;
5322         }
5323     }
5324 }
5325 
5326 BlockDirtyInfoList *bdrv_query_dirty_bitmaps(BlockDriverState *bs)
5327 {
5328     BdrvDirtyBitmap *bm;
5329     BlockDirtyInfoList *list = NULL;
5330     BlockDirtyInfoList **plist = &list;
5331 
5332     QLIST_FOREACH(bm, &bs->dirty_bitmaps, list) {
5333         BlockDirtyInfo *info = g_malloc0(sizeof(BlockDirtyInfo));
5334         BlockDirtyInfoList *entry = g_malloc0(sizeof(BlockDirtyInfoList));
5335         info->count = bdrv_get_dirty_count(bs, bm);
5336         info->granularity =
5337             ((int64_t) BDRV_SECTOR_SIZE << hbitmap_granularity(bm->bitmap));
5338         entry->value = info;
5339         *plist = entry;
5340         plist = &entry->next;
5341     }
5342 
5343     return list;
5344 }
5345 
5346 int bdrv_get_dirty(BlockDriverState *bs, BdrvDirtyBitmap *bitmap, int64_t sector)
5347 {
5348     if (bitmap) {
5349         return hbitmap_get(bitmap->bitmap, sector);
5350     } else {
5351         return 0;
5352     }
5353 }
5354 
5355 void bdrv_dirty_iter_init(BlockDriverState *bs,
5356                           BdrvDirtyBitmap *bitmap, HBitmapIter *hbi)
5357 {
5358     hbitmap_iter_init(hbi, bitmap->bitmap, 0);
5359 }
5360 
5361 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector,
5362                     int nr_sectors)
5363 {
5364     BdrvDirtyBitmap *bitmap;
5365     QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
5366         hbitmap_set(bitmap->bitmap, cur_sector, nr_sectors);
5367     }
5368 }
5369 
5370 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector, int nr_sectors)
5371 {
5372     BdrvDirtyBitmap *bitmap;
5373     QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) {
5374         hbitmap_reset(bitmap->bitmap, cur_sector, nr_sectors);
5375     }
5376 }
5377 
5378 int64_t bdrv_get_dirty_count(BlockDriverState *bs, BdrvDirtyBitmap *bitmap)
5379 {
5380     return hbitmap_count(bitmap->bitmap);
5381 }
5382 
5383 /* Get a reference to bs */
5384 void bdrv_ref(BlockDriverState *bs)
5385 {
5386     bs->refcnt++;
5387 }
5388 
5389 /* Release a previously grabbed reference to bs.
5390  * If after releasing, reference count is zero, the BlockDriverState is
5391  * deleted. */
5392 void bdrv_unref(BlockDriverState *bs)
5393 {
5394     assert(bs->refcnt > 0);
5395     if (--bs->refcnt == 0) {
5396         bdrv_delete(bs);
5397     }
5398 }
5399 
5400 struct BdrvOpBlocker {
5401     Error *reason;
5402     QLIST_ENTRY(BdrvOpBlocker) list;
5403 };
5404 
5405 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp)
5406 {
5407     BdrvOpBlocker *blocker;
5408     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5409     if (!QLIST_EMPTY(&bs->op_blockers[op])) {
5410         blocker = QLIST_FIRST(&bs->op_blockers[op]);
5411         if (errp) {
5412             error_setg(errp, "Device '%s' is busy: %s",
5413                        bs->device_name, error_get_pretty(blocker->reason));
5414         }
5415         return true;
5416     }
5417     return false;
5418 }
5419 
5420 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason)
5421 {
5422     BdrvOpBlocker *blocker;
5423     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5424 
5425     blocker = g_malloc0(sizeof(BdrvOpBlocker));
5426     blocker->reason = reason;
5427     QLIST_INSERT_HEAD(&bs->op_blockers[op], blocker, list);
5428 }
5429 
5430 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason)
5431 {
5432     BdrvOpBlocker *blocker, *next;
5433     assert((int) op >= 0 && op < BLOCK_OP_TYPE_MAX);
5434     QLIST_FOREACH_SAFE(blocker, &bs->op_blockers[op], list, next) {
5435         if (blocker->reason == reason) {
5436             QLIST_REMOVE(blocker, list);
5437             g_free(blocker);
5438         }
5439     }
5440 }
5441 
5442 void bdrv_op_block_all(BlockDriverState *bs, Error *reason)
5443 {
5444     int i;
5445     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5446         bdrv_op_block(bs, i, reason);
5447     }
5448 }
5449 
5450 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason)
5451 {
5452     int i;
5453     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5454         bdrv_op_unblock(bs, i, reason);
5455     }
5456 }
5457 
5458 bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
5459 {
5460     int i;
5461 
5462     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
5463         if (!QLIST_EMPTY(&bs->op_blockers[i])) {
5464             return false;
5465         }
5466     }
5467     return true;
5468 }
5469 
5470 void bdrv_iostatus_enable(BlockDriverState *bs)
5471 {
5472     bs->iostatus_enabled = true;
5473     bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
5474 }
5475 
5476 /* The I/O status is only enabled if the drive explicitly
5477  * enables it _and_ the VM is configured to stop on errors */
5478 bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
5479 {
5480     return (bs->iostatus_enabled &&
5481            (bs->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
5482             bs->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
5483             bs->on_read_error == BLOCKDEV_ON_ERROR_STOP));
5484 }
5485 
5486 void bdrv_iostatus_disable(BlockDriverState *bs)
5487 {
5488     bs->iostatus_enabled = false;
5489 }
5490 
5491 void bdrv_iostatus_reset(BlockDriverState *bs)
5492 {
5493     if (bdrv_iostatus_is_enabled(bs)) {
5494         bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
5495         if (bs->job) {
5496             block_job_iostatus_reset(bs->job);
5497         }
5498     }
5499 }
5500 
5501 void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
5502 {
5503     assert(bdrv_iostatus_is_enabled(bs));
5504     if (bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
5505         bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
5506                                          BLOCK_DEVICE_IO_STATUS_FAILED;
5507     }
5508 }
5509 
5510 void
5511 bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes,
5512         enum BlockAcctType type)
5513 {
5514     assert(type < BDRV_MAX_IOTYPE);
5515 
5516     cookie->bytes = bytes;
5517     cookie->start_time_ns = get_clock();
5518     cookie->type = type;
5519 }
5520 
5521 void
5522 bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie)
5523 {
5524     assert(cookie->type < BDRV_MAX_IOTYPE);
5525 
5526     bs->nr_bytes[cookie->type] += cookie->bytes;
5527     bs->nr_ops[cookie->type]++;
5528     bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns;
5529 }
5530 
5531 void bdrv_img_create(const char *filename, const char *fmt,
5532                      const char *base_filename, const char *base_fmt,
5533                      char *options, uint64_t img_size, int flags,
5534                      Error **errp, bool quiet)
5535 {
5536     QemuOptsList *create_opts = NULL;
5537     QemuOpts *opts = NULL;
5538     const char *backing_fmt, *backing_file;
5539     int64_t size;
5540     BlockDriver *drv, *proto_drv;
5541     BlockDriver *backing_drv = NULL;
5542     Error *local_err = NULL;
5543     int ret = 0;
5544 
5545     /* Find driver and parse its options */
5546     drv = bdrv_find_format(fmt);
5547     if (!drv) {
5548         error_setg(errp, "Unknown file format '%s'", fmt);
5549         return;
5550     }
5551 
5552     proto_drv = bdrv_find_protocol(filename, true);
5553     if (!proto_drv) {
5554         error_setg(errp, "Unknown protocol '%s'", filename);
5555         return;
5556     }
5557 
5558     create_opts = qemu_opts_append(create_opts, drv->create_opts);
5559     create_opts = qemu_opts_append(create_opts, proto_drv->create_opts);
5560 
5561     /* Create parameter list with default values */
5562     opts = qemu_opts_create(create_opts, NULL, 0, &error_abort);
5563     qemu_opt_set_number(opts, BLOCK_OPT_SIZE, img_size);
5564 
5565     /* Parse -o options */
5566     if (options) {
5567         if (qemu_opts_do_parse(opts, options, NULL) != 0) {
5568             error_setg(errp, "Invalid options for file format '%s'", fmt);
5569             goto out;
5570         }
5571     }
5572 
5573     if (base_filename) {
5574         if (qemu_opt_set(opts, BLOCK_OPT_BACKING_FILE, base_filename)) {
5575             error_setg(errp, "Backing file not supported for file format '%s'",
5576                        fmt);
5577             goto out;
5578         }
5579     }
5580 
5581     if (base_fmt) {
5582         if (qemu_opt_set(opts, BLOCK_OPT_BACKING_FMT, base_fmt)) {
5583             error_setg(errp, "Backing file format not supported for file "
5584                              "format '%s'", fmt);
5585             goto out;
5586         }
5587     }
5588 
5589     backing_file = qemu_opt_get(opts, BLOCK_OPT_BACKING_FILE);
5590     if (backing_file) {
5591         if (!strcmp(filename, backing_file)) {
5592             error_setg(errp, "Error: Trying to create an image with the "
5593                              "same filename as the backing file");
5594             goto out;
5595         }
5596     }
5597 
5598     backing_fmt = qemu_opt_get(opts, BLOCK_OPT_BACKING_FMT);
5599     if (backing_fmt) {
5600         backing_drv = bdrv_find_format(backing_fmt);
5601         if (!backing_drv) {
5602             error_setg(errp, "Unknown backing file format '%s'",
5603                        backing_fmt);
5604             goto out;
5605         }
5606     }
5607 
5608     // The size for the image must always be specified, with one exception:
5609     // If we are using a backing file, we can obtain the size from there
5610     size = qemu_opt_get_size(opts, BLOCK_OPT_SIZE, 0);
5611     if (size == -1) {
5612         if (backing_file) {
5613             BlockDriverState *bs;
5614             uint64_t size;
5615             int back_flags;
5616 
5617             /* backing files always opened read-only */
5618             back_flags =
5619                 flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
5620 
5621             bs = NULL;
5622             ret = bdrv_open(&bs, backing_file, NULL, NULL, back_flags,
5623                             backing_drv, &local_err);
5624             if (ret < 0) {
5625                 error_setg_errno(errp, -ret, "Could not open '%s': %s",
5626                                  backing_file,
5627                                  error_get_pretty(local_err));
5628                 error_free(local_err);
5629                 local_err = NULL;
5630                 goto out;
5631             }
5632             bdrv_get_geometry(bs, &size);
5633             size *= 512;
5634 
5635             qemu_opt_set_number(opts, BLOCK_OPT_SIZE, size);
5636 
5637             bdrv_unref(bs);
5638         } else {
5639             error_setg(errp, "Image creation needs a size parameter");
5640             goto out;
5641         }
5642     }
5643 
5644     if (!quiet) {
5645         printf("Formatting '%s', fmt=%s ", filename, fmt);
5646         qemu_opts_print(opts);
5647         puts("");
5648     }
5649 
5650     ret = bdrv_create(drv, filename, opts, &local_err);
5651 
5652     if (ret == -EFBIG) {
5653         /* This is generally a better message than whatever the driver would
5654          * deliver (especially because of the cluster_size_hint), since that
5655          * is most probably not much different from "image too large". */
5656         const char *cluster_size_hint = "";
5657         if (qemu_opt_get_size(opts, BLOCK_OPT_CLUSTER_SIZE, 0)) {
5658             cluster_size_hint = " (try using a larger cluster size)";
5659         }
5660         error_setg(errp, "The image size is too large for file format '%s'"
5661                    "%s", fmt, cluster_size_hint);
5662         error_free(local_err);
5663         local_err = NULL;
5664     }
5665 
5666 out:
5667     qemu_opts_del(opts);
5668     qemu_opts_free(create_opts);
5669     if (local_err) {
5670         error_propagate(errp, local_err);
5671     }
5672 }
5673 
5674 AioContext *bdrv_get_aio_context(BlockDriverState *bs)
5675 {
5676     return bs->aio_context;
5677 }
5678 
5679 void bdrv_detach_aio_context(BlockDriverState *bs)
5680 {
5681     if (!bs->drv) {
5682         return;
5683     }
5684 
5685     if (bs->io_limits_enabled) {
5686         throttle_detach_aio_context(&bs->throttle_state);
5687     }
5688     if (bs->drv->bdrv_detach_aio_context) {
5689         bs->drv->bdrv_detach_aio_context(bs);
5690     }
5691     if (bs->file) {
5692         bdrv_detach_aio_context(bs->file);
5693     }
5694     if (bs->backing_hd) {
5695         bdrv_detach_aio_context(bs->backing_hd);
5696     }
5697 
5698     bs->aio_context = NULL;
5699 }
5700 
5701 void bdrv_attach_aio_context(BlockDriverState *bs,
5702                              AioContext *new_context)
5703 {
5704     if (!bs->drv) {
5705         return;
5706     }
5707 
5708     bs->aio_context = new_context;
5709 
5710     if (bs->backing_hd) {
5711         bdrv_attach_aio_context(bs->backing_hd, new_context);
5712     }
5713     if (bs->file) {
5714         bdrv_attach_aio_context(bs->file, new_context);
5715     }
5716     if (bs->drv->bdrv_attach_aio_context) {
5717         bs->drv->bdrv_attach_aio_context(bs, new_context);
5718     }
5719     if (bs->io_limits_enabled) {
5720         throttle_attach_aio_context(&bs->throttle_state, new_context);
5721     }
5722 }
5723 
5724 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context)
5725 {
5726     bdrv_drain_all(); /* ensure there are no in-flight requests */
5727 
5728     bdrv_detach_aio_context(bs);
5729 
5730     /* This function executes in the old AioContext so acquire the new one in
5731      * case it runs in a different thread.
5732      */
5733     aio_context_acquire(new_context);
5734     bdrv_attach_aio_context(bs, new_context);
5735     aio_context_release(new_context);
5736 }
5737 
5738 void bdrv_add_before_write_notifier(BlockDriverState *bs,
5739                                     NotifierWithReturn *notifier)
5740 {
5741     notifier_with_return_list_add(&bs->before_write_notifiers, notifier);
5742 }
5743 
5744 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts)
5745 {
5746     if (!bs->drv->bdrv_amend_options) {
5747         return -ENOTSUP;
5748     }
5749     return bs->drv->bdrv_amend_options(bs, opts);
5750 }
5751 
5752 /* This function will be called by the bdrv_recurse_is_first_non_filter method
5753  * of block filter and by bdrv_is_first_non_filter.
5754  * It is used to test if the given bs is the candidate or recurse more in the
5755  * node graph.
5756  */
5757 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
5758                                       BlockDriverState *candidate)
5759 {
5760     /* return false if basic checks fails */
5761     if (!bs || !bs->drv) {
5762         return false;
5763     }
5764 
5765     /* the code reached a non block filter driver -> check if the bs is
5766      * the same as the candidate. It's the recursion termination condition.
5767      */
5768     if (!bs->drv->is_filter) {
5769         return bs == candidate;
5770     }
5771     /* Down this path the driver is a block filter driver */
5772 
5773     /* If the block filter recursion method is defined use it to recurse down
5774      * the node graph.
5775      */
5776     if (bs->drv->bdrv_recurse_is_first_non_filter) {
5777         return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate);
5778     }
5779 
5780     /* the driver is a block filter but don't allow to recurse -> return false
5781      */
5782     return false;
5783 }
5784 
5785 /* This function checks if the candidate is the first non filter bs down it's
5786  * bs chain. Since we don't have pointers to parents it explore all bs chains
5787  * from the top. Some filters can choose not to pass down the recursion.
5788  */
5789 bool bdrv_is_first_non_filter(BlockDriverState *candidate)
5790 {
5791     BlockDriverState *bs;
5792 
5793     /* walk down the bs forest recursively */
5794     QTAILQ_FOREACH(bs, &bdrv_states, device_list) {
5795         bool perm;
5796 
5797         /* try to recurse in this top level bs */
5798         perm = bdrv_recurse_is_first_non_filter(bs, candidate);
5799 
5800         /* candidate is the first non filter */
5801         if (perm) {
5802             return true;
5803         }
5804     }
5805 
5806     return false;
5807 }
5808