xref: /openbmc/qemu/include/block/block_int.h (revision 56e2cd24)
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 #ifndef BLOCK_INT_H
25 #define BLOCK_INT_H
26 
27 #include "block/accounting.h"
28 #include "block/block.h"
29 #include "qemu/option.h"
30 #include "qemu/queue.h"
31 #include "qemu/coroutine.h"
32 #include "qemu/timer.h"
33 #include "qapi-types.h"
34 #include "qemu/hbitmap.h"
35 #include "block/snapshot.h"
36 #include "qemu/main-loop.h"
37 #include "qemu/throttle.h"
38 
39 #define BLOCK_FLAG_ENCRYPT          1
40 #define BLOCK_FLAG_LAZY_REFCOUNTS   8
41 
42 #define BLOCK_OPT_SIZE              "size"
43 #define BLOCK_OPT_ENCRYPT           "encryption"
44 #define BLOCK_OPT_COMPAT6           "compat6"
45 #define BLOCK_OPT_HWVERSION         "hwversion"
46 #define BLOCK_OPT_BACKING_FILE      "backing_file"
47 #define BLOCK_OPT_BACKING_FMT       "backing_fmt"
48 #define BLOCK_OPT_CLUSTER_SIZE      "cluster_size"
49 #define BLOCK_OPT_TABLE_SIZE        "table_size"
50 #define BLOCK_OPT_PREALLOC          "preallocation"
51 #define BLOCK_OPT_SUBFMT            "subformat"
52 #define BLOCK_OPT_COMPAT_LEVEL      "compat"
53 #define BLOCK_OPT_LAZY_REFCOUNTS    "lazy_refcounts"
54 #define BLOCK_OPT_ADAPTER_TYPE      "adapter_type"
55 #define BLOCK_OPT_REDUNDANCY        "redundancy"
56 #define BLOCK_OPT_NOCOW             "nocow"
57 #define BLOCK_OPT_OBJECT_SIZE       "object_size"
58 #define BLOCK_OPT_REFCOUNT_BITS     "refcount_bits"
59 
60 #define BLOCK_PROBE_BUF_SIZE        512
61 
62 enum BdrvTrackedRequestType {
63     BDRV_TRACKED_READ,
64     BDRV_TRACKED_WRITE,
65     BDRV_TRACKED_DISCARD,
66 };
67 
68 typedef struct BdrvTrackedRequest {
69     BlockDriverState *bs;
70     int64_t offset;
71     unsigned int bytes;
72     enum BdrvTrackedRequestType type;
73 
74     bool serialising;
75     int64_t overlap_offset;
76     unsigned int overlap_bytes;
77 
78     QLIST_ENTRY(BdrvTrackedRequest) list;
79     Coroutine *co; /* owner, used for deadlock detection */
80     CoQueue wait_queue; /* coroutines blocked on this request */
81 
82     struct BdrvTrackedRequest *waiting_for;
83 } BdrvTrackedRequest;
84 
85 struct BlockDriver {
86     const char *format_name;
87     int instance_size;
88 
89     /* set to true if the BlockDriver is a block filter */
90     bool is_filter;
91     /* for snapshots block filter like Quorum can implement the
92      * following recursive callback.
93      * It's purpose is to recurse on the filter children while calling
94      * bdrv_recurse_is_first_non_filter on them.
95      * For a sample implementation look in the future Quorum block filter.
96      */
97     bool (*bdrv_recurse_is_first_non_filter)(BlockDriverState *bs,
98                                              BlockDriverState *candidate);
99 
100     int (*bdrv_probe)(const uint8_t *buf, int buf_size, const char *filename);
101     int (*bdrv_probe_device)(const char *filename);
102 
103     /* Any driver implementing this callback is expected to be able to handle
104      * NULL file names in its .bdrv_open() implementation */
105     void (*bdrv_parse_filename)(const char *filename, QDict *options, Error **errp);
106     /* Drivers not implementing bdrv_parse_filename nor bdrv_open should have
107      * this field set to true, except ones that are defined only by their
108      * child's bs.
109      * An example of the last type will be the quorum block driver.
110      */
111     bool bdrv_needs_filename;
112 
113     /* Set if a driver can support backing files */
114     bool supports_backing;
115 
116     /* For handling image reopen for split or non-split files */
117     int (*bdrv_reopen_prepare)(BDRVReopenState *reopen_state,
118                                BlockReopenQueue *queue, Error **errp);
119     void (*bdrv_reopen_commit)(BDRVReopenState *reopen_state);
120     void (*bdrv_reopen_abort)(BDRVReopenState *reopen_state);
121     void (*bdrv_join_options)(QDict *options, QDict *old_options);
122 
123     int (*bdrv_open)(BlockDriverState *bs, QDict *options, int flags,
124                      Error **errp);
125     int (*bdrv_file_open)(BlockDriverState *bs, QDict *options, int flags,
126                           Error **errp);
127     void (*bdrv_close)(BlockDriverState *bs);
128     int (*bdrv_create)(const char *filename, QemuOpts *opts, Error **errp);
129     int (*bdrv_set_key)(BlockDriverState *bs, const char *key);
130     int (*bdrv_make_empty)(BlockDriverState *bs);
131 
132     void (*bdrv_refresh_filename)(BlockDriverState *bs, QDict *options);
133 
134     /* aio */
135     BlockAIOCB *(*bdrv_aio_readv)(BlockDriverState *bs,
136         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
137         BlockCompletionFunc *cb, void *opaque);
138     BlockAIOCB *(*bdrv_aio_writev)(BlockDriverState *bs,
139         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
140         BlockCompletionFunc *cb, void *opaque);
141     BlockAIOCB *(*bdrv_aio_flush)(BlockDriverState *bs,
142         BlockCompletionFunc *cb, void *opaque);
143     BlockAIOCB *(*bdrv_aio_pdiscard)(BlockDriverState *bs,
144         int64_t offset, int count,
145         BlockCompletionFunc *cb, void *opaque);
146 
147     int coroutine_fn (*bdrv_co_readv)(BlockDriverState *bs,
148         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
149     int coroutine_fn (*bdrv_co_preadv)(BlockDriverState *bs,
150         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
151     int coroutine_fn (*bdrv_co_writev)(BlockDriverState *bs,
152         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
153     int coroutine_fn (*bdrv_co_writev_flags)(BlockDriverState *bs,
154         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, int flags);
155     int coroutine_fn (*bdrv_co_pwritev)(BlockDriverState *bs,
156         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
157 
158     /*
159      * Efficiently zero a region of the disk image.  Typically an image format
160      * would use a compact metadata representation to implement this.  This
161      * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
162      * will be called instead.
163      */
164     int coroutine_fn (*bdrv_co_pwrite_zeroes)(BlockDriverState *bs,
165         int64_t offset, int count, BdrvRequestFlags flags);
166     int coroutine_fn (*bdrv_co_pdiscard)(BlockDriverState *bs,
167         int64_t offset, int count);
168     int64_t coroutine_fn (*bdrv_co_get_block_status)(BlockDriverState *bs,
169         int64_t sector_num, int nb_sectors, int *pnum,
170         BlockDriverState **file);
171 
172     /*
173      * Invalidate any cached meta-data.
174      */
175     void (*bdrv_invalidate_cache)(BlockDriverState *bs, Error **errp);
176     int (*bdrv_inactivate)(BlockDriverState *bs);
177 
178     /*
179      * Flushes all data for all layers by calling bdrv_co_flush for underlying
180      * layers, if needed. This function is needed for deterministic
181      * synchronization of the flush finishing callback.
182      */
183     int coroutine_fn (*bdrv_co_flush)(BlockDriverState *bs);
184 
185     /*
186      * Flushes all data that was already written to the OS all the way down to
187      * the disk (for example file-posix.c calls fsync()).
188      */
189     int coroutine_fn (*bdrv_co_flush_to_disk)(BlockDriverState *bs);
190 
191     /*
192      * Flushes all internal caches to the OS. The data may still sit in a
193      * writeback cache of the host OS, but it will survive a crash of the qemu
194      * process.
195      */
196     int coroutine_fn (*bdrv_co_flush_to_os)(BlockDriverState *bs);
197 
198     const char *protocol_name;
199     int (*bdrv_truncate)(BlockDriverState *bs, int64_t offset);
200 
201     int64_t (*bdrv_getlength)(BlockDriverState *bs);
202     bool has_variable_length;
203     int64_t (*bdrv_get_allocated_file_size)(BlockDriverState *bs);
204 
205     int coroutine_fn (*bdrv_co_pwritev_compressed)(BlockDriverState *bs,
206         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov);
207 
208     int (*bdrv_snapshot_create)(BlockDriverState *bs,
209                                 QEMUSnapshotInfo *sn_info);
210     int (*bdrv_snapshot_goto)(BlockDriverState *bs,
211                               const char *snapshot_id);
212     int (*bdrv_snapshot_delete)(BlockDriverState *bs,
213                                 const char *snapshot_id,
214                                 const char *name,
215                                 Error **errp);
216     int (*bdrv_snapshot_list)(BlockDriverState *bs,
217                               QEMUSnapshotInfo **psn_info);
218     int (*bdrv_snapshot_load_tmp)(BlockDriverState *bs,
219                                   const char *snapshot_id,
220                                   const char *name,
221                                   Error **errp);
222     int (*bdrv_get_info)(BlockDriverState *bs, BlockDriverInfo *bdi);
223     ImageInfoSpecific *(*bdrv_get_specific_info)(BlockDriverState *bs);
224 
225     int coroutine_fn (*bdrv_save_vmstate)(BlockDriverState *bs,
226                                           QEMUIOVector *qiov,
227                                           int64_t pos);
228     int coroutine_fn (*bdrv_load_vmstate)(BlockDriverState *bs,
229                                           QEMUIOVector *qiov,
230                                           int64_t pos);
231 
232     int (*bdrv_change_backing_file)(BlockDriverState *bs,
233         const char *backing_file, const char *backing_fmt);
234 
235     /* removable device specific */
236     bool (*bdrv_is_inserted)(BlockDriverState *bs);
237     int (*bdrv_media_changed)(BlockDriverState *bs);
238     void (*bdrv_eject)(BlockDriverState *bs, bool eject_flag);
239     void (*bdrv_lock_medium)(BlockDriverState *bs, bool locked);
240 
241     /* to control generic scsi devices */
242     BlockAIOCB *(*bdrv_aio_ioctl)(BlockDriverState *bs,
243         unsigned long int req, void *buf,
244         BlockCompletionFunc *cb, void *opaque);
245     int coroutine_fn (*bdrv_co_ioctl)(BlockDriverState *bs,
246                                       unsigned long int req, void *buf);
247 
248     /* List of options for creating images, terminated by name == NULL */
249     QemuOptsList *create_opts;
250 
251     /*
252      * Returns 0 for completed check, -errno for internal errors.
253      * The check results are stored in result.
254      */
255     int (*bdrv_check)(BlockDriverState* bs, BdrvCheckResult *result,
256         BdrvCheckMode fix);
257 
258     int (*bdrv_amend_options)(BlockDriverState *bs, QemuOpts *opts,
259                               BlockDriverAmendStatusCB *status_cb,
260                               void *cb_opaque);
261 
262     void (*bdrv_debug_event)(BlockDriverState *bs, BlkdebugEvent event);
263 
264     /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
265     int (*bdrv_debug_breakpoint)(BlockDriverState *bs, const char *event,
266         const char *tag);
267     int (*bdrv_debug_remove_breakpoint)(BlockDriverState *bs,
268         const char *tag);
269     int (*bdrv_debug_resume)(BlockDriverState *bs, const char *tag);
270     bool (*bdrv_debug_is_suspended)(BlockDriverState *bs, const char *tag);
271 
272     void (*bdrv_refresh_limits)(BlockDriverState *bs, Error **errp);
273 
274     /*
275      * Returns 1 if newly created images are guaranteed to contain only
276      * zeros, 0 otherwise.
277      */
278     int (*bdrv_has_zero_init)(BlockDriverState *bs);
279 
280     /* Remove fd handlers, timers, and other event loop callbacks so the event
281      * loop is no longer in use.  Called with no in-flight requests and in
282      * depth-first traversal order with parents before child nodes.
283      */
284     void (*bdrv_detach_aio_context)(BlockDriverState *bs);
285 
286     /* Add fd handlers, timers, and other event loop callbacks so I/O requests
287      * can be processed again.  Called with no in-flight requests and in
288      * depth-first traversal order with child nodes before parent nodes.
289      */
290     void (*bdrv_attach_aio_context)(BlockDriverState *bs,
291                                     AioContext *new_context);
292 
293     /* io queue for linux-aio */
294     void (*bdrv_io_plug)(BlockDriverState *bs);
295     void (*bdrv_io_unplug)(BlockDriverState *bs);
296 
297     /**
298      * Try to get @bs's logical and physical block size.
299      * On success, store them in @bsz and return zero.
300      * On failure, return negative errno.
301      */
302     int (*bdrv_probe_blocksizes)(BlockDriverState *bs, BlockSizes *bsz);
303     /**
304      * Try to get @bs's geometry (cyls, heads, sectors)
305      * On success, store them in @geo and return 0.
306      * On failure return -errno.
307      * Only drivers that want to override guest geometry implement this
308      * callback; see hd_geometry_guess().
309      */
310     int (*bdrv_probe_geometry)(BlockDriverState *bs, HDGeometry *geo);
311 
312     /**
313      * Drain and stop any internal sources of requests in the driver, and
314      * remain so until next I/O callback (e.g. bdrv_co_writev) is called.
315      */
316     void (*bdrv_drain)(BlockDriverState *bs);
317 
318     void (*bdrv_add_child)(BlockDriverState *parent, BlockDriverState *child,
319                            Error **errp);
320     void (*bdrv_del_child)(BlockDriverState *parent, BdrvChild *child,
321                            Error **errp);
322 
323     /**
324      * Informs the block driver that a permission change is intended. The
325      * driver checks whether the change is permissible and may take other
326      * preparations for the change (e.g. get file system locks). This operation
327      * is always followed either by a call to either .bdrv_set_perm or
328      * .bdrv_abort_perm_update.
329      *
330      * Checks whether the requested set of cumulative permissions in @perm
331      * can be granted for accessing @bs and whether no other users are using
332      * permissions other than those given in @shared (both arguments take
333      * BLK_PERM_* bitmasks).
334      *
335      * If both conditions are met, 0 is returned. Otherwise, -errno is returned
336      * and errp is set to an error describing the conflict.
337      */
338     int (*bdrv_check_perm)(BlockDriverState *bs, uint64_t perm,
339                            uint64_t shared, Error **errp);
340 
341     /**
342      * Called to inform the driver that the set of cumulative set of used
343      * permissions for @bs has changed to @perm, and the set of sharable
344      * permission to @shared. The driver can use this to propagate changes to
345      * its children (i.e. request permissions only if a parent actually needs
346      * them).
347      *
348      * This function is only invoked after bdrv_check_perm(), so block drivers
349      * may rely on preparations made in their .bdrv_check_perm implementation.
350      */
351     void (*bdrv_set_perm)(BlockDriverState *bs, uint64_t perm, uint64_t shared);
352 
353     /*
354      * Called to inform the driver that after a previous bdrv_check_perm()
355      * call, the permission update is not performed and any preparations made
356      * for it (e.g. taken file locks) need to be undone.
357      *
358      * This function can be called even for nodes that never saw a
359      * bdrv_check_perm() call. It is a no-op then.
360      */
361     void (*bdrv_abort_perm_update)(BlockDriverState *bs);
362 
363     /**
364      * Returns in @nperm and @nshared the permissions that the driver for @bs
365      * needs on its child @c, based on the cumulative permissions requested by
366      * the parents in @parent_perm and @parent_shared.
367      *
368      * If @c is NULL, return the permissions for attaching a new child for the
369      * given @role.
370      */
371      void (*bdrv_child_perm)(BlockDriverState *bs, BdrvChild *c,
372                              const BdrvChildRole *role,
373                              uint64_t parent_perm, uint64_t parent_shared,
374                              uint64_t *nperm, uint64_t *nshared);
375 
376     QLIST_ENTRY(BlockDriver) list;
377 };
378 
379 typedef struct BlockLimits {
380     /* Alignment requirement, in bytes, for offset/length of I/O
381      * requests. Must be a power of 2 less than INT_MAX; defaults to
382      * 1 for drivers with modern byte interfaces, and to 512
383      * otherwise. */
384     uint32_t request_alignment;
385 
386     /* Maximum number of bytes that can be discarded at once (since it
387      * is signed, it must be < 2G, if set). Must be multiple of
388      * pdiscard_alignment, but need not be power of 2. May be 0 if no
389      * inherent 32-bit limit */
390     int32_t max_pdiscard;
391 
392     /* Optimal alignment for discard requests in bytes. A power of 2
393      * is best but not mandatory.  Must be a multiple of
394      * bl.request_alignment, and must be less than max_pdiscard if
395      * that is set. May be 0 if bl.request_alignment is good enough */
396     uint32_t pdiscard_alignment;
397 
398     /* Maximum number of bytes that can zeroized at once (since it is
399      * signed, it must be < 2G, if set). Must be multiple of
400      * pwrite_zeroes_alignment. May be 0 if no inherent 32-bit limit */
401     int32_t max_pwrite_zeroes;
402 
403     /* Optimal alignment for write zeroes requests in bytes. A power
404      * of 2 is best but not mandatory.  Must be a multiple of
405      * bl.request_alignment, and must be less than max_pwrite_zeroes
406      * if that is set. May be 0 if bl.request_alignment is good
407      * enough */
408     uint32_t pwrite_zeroes_alignment;
409 
410     /* Optimal transfer length in bytes.  A power of 2 is best but not
411      * mandatory.  Must be a multiple of bl.request_alignment, or 0 if
412      * no preferred size */
413     uint32_t opt_transfer;
414 
415     /* Maximal transfer length in bytes.  Need not be power of 2, but
416      * must be multiple of opt_transfer and bl.request_alignment, or 0
417      * for no 32-bit limit.  For now, anything larger than INT_MAX is
418      * clamped down. */
419     uint32_t max_transfer;
420 
421     /* memory alignment, in bytes so that no bounce buffer is needed */
422     size_t min_mem_alignment;
423 
424     /* memory alignment, in bytes, for bounce buffer */
425     size_t opt_mem_alignment;
426 
427     /* maximum number of iovec elements */
428     int max_iov;
429 } BlockLimits;
430 
431 typedef struct BdrvOpBlocker BdrvOpBlocker;
432 
433 typedef struct BdrvAioNotifier {
434     void (*attached_aio_context)(AioContext *new_context, void *opaque);
435     void (*detach_aio_context)(void *opaque);
436 
437     void *opaque;
438     bool deleted;
439 
440     QLIST_ENTRY(BdrvAioNotifier) list;
441 } BdrvAioNotifier;
442 
443 struct BdrvChildRole {
444     /* If true, bdrv_replace_node() doesn't change the node this BdrvChild
445      * points to. */
446     bool stay_at_node;
447 
448     void (*inherit_options)(int *child_flags, QDict *child_options,
449                             int parent_flags, QDict *parent_options);
450 
451     void (*change_media)(BdrvChild *child, bool load);
452     void (*resize)(BdrvChild *child);
453 
454     /* Returns a name that is supposedly more useful for human users than the
455      * node name for identifying the node in question (in particular, a BB
456      * name), or NULL if the parent can't provide a better name. */
457     const char* (*get_name)(BdrvChild *child);
458 
459     /* Returns a malloced string that describes the parent of the child for a
460      * human reader. This could be a node-name, BlockBackend name, qdev ID or
461      * QOM path of the device owning the BlockBackend, job type and ID etc. The
462      * caller is responsible for freeing the memory. */
463     char* (*get_parent_desc)(BdrvChild *child);
464 
465     /*
466      * If this pair of functions is implemented, the parent doesn't issue new
467      * requests after returning from .drained_begin() until .drained_end() is
468      * called.
469      *
470      * Note that this can be nested. If drained_begin() was called twice, new
471      * I/O is allowed only after drained_end() was called twice, too.
472      */
473     void (*drained_begin)(BdrvChild *child);
474     void (*drained_end)(BdrvChild *child);
475 
476     void (*attach)(BdrvChild *child);
477     void (*detach)(BdrvChild *child);
478 };
479 
480 extern const BdrvChildRole child_file;
481 extern const BdrvChildRole child_format;
482 extern const BdrvChildRole child_backing;
483 
484 struct BdrvChild {
485     BlockDriverState *bs;
486     char *name;
487     const BdrvChildRole *role;
488     void *opaque;
489 
490     /**
491      * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
492      */
493     uint64_t perm;
494 
495     /**
496      * Permissions that can still be granted to other users of @bs while this
497      * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
498      */
499     uint64_t shared_perm;
500 
501     QLIST_ENTRY(BdrvChild) next;
502     QLIST_ENTRY(BdrvChild) next_parent;
503 };
504 
505 /*
506  * Note: the function bdrv_append() copies and swaps contents of
507  * BlockDriverStates, so if you add new fields to this struct, please
508  * inspect bdrv_append() to determine if the new fields need to be
509  * copied as well.
510  */
511 struct BlockDriverState {
512     /* Protected by big QEMU lock or read-only after opening.  No special
513      * locking needed during I/O...
514      */
515     int open_flags; /* flags used to open the file, re-used for re-open */
516     bool read_only; /* if true, the media is read only */
517     bool encrypted; /* if true, the media is encrypted */
518     bool valid_key; /* if true, a valid encryption key has been set */
519     bool sg;        /* if true, the device is a /dev/sg* */
520     bool probed;    /* if true, format was probed rather than specified */
521 
522     BlockDriver *drv; /* NULL means no media */
523     void *opaque;
524 
525     AioContext *aio_context; /* event loop used for fd handlers, timers, etc */
526     /* long-running tasks intended to always use the same AioContext as this
527      * BDS may register themselves in this list to be notified of changes
528      * regarding this BDS's context */
529     QLIST_HEAD(, BdrvAioNotifier) aio_notifiers;
530     bool walking_aio_notifiers; /* to make removal during iteration safe */
531 
532     char filename[PATH_MAX];
533     char backing_file[PATH_MAX]; /* if non zero, the image is a diff of
534                                     this file image */
535     char backing_format[16]; /* if non-zero and backing_file exists */
536 
537     QDict *full_open_options;
538     char exact_filename[PATH_MAX];
539 
540     BdrvChild *backing;
541     BdrvChild *file;
542 
543     /* I/O Limits */
544     BlockLimits bl;
545 
546     /* Flags honored during pwrite (so far: BDRV_REQ_FUA) */
547     unsigned int supported_write_flags;
548     /* Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
549      * BDRV_REQ_MAY_UNMAP) */
550     unsigned int supported_zero_flags;
551 
552     /* the following member gives a name to every node on the bs graph. */
553     char node_name[32];
554     /* element of the list of named nodes building the graph */
555     QTAILQ_ENTRY(BlockDriverState) node_list;
556     /* element of the list of all BlockDriverStates (all_bdrv_states) */
557     QTAILQ_ENTRY(BlockDriverState) bs_list;
558     /* element of the list of monitor-owned BDS */
559     QTAILQ_ENTRY(BlockDriverState) monitor_list;
560     int refcnt;
561 
562     /* operation blockers */
563     QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX];
564 
565     /* long-running background operation */
566     BlockJob *job;
567 
568     /* The node that this node inherited default options from (and a reopen on
569      * which can affect this node by changing these defaults). This is always a
570      * parent node of this node. */
571     BlockDriverState *inherits_from;
572     QLIST_HEAD(, BdrvChild) children;
573     QLIST_HEAD(, BdrvChild) parents;
574 
575     QDict *options;
576     QDict *explicit_options;
577     BlockdevDetectZeroesOptions detect_zeroes;
578 
579     /* The error object in use for blocking operations on backing_hd */
580     Error *backing_blocker;
581 
582     /* Protected by AioContext lock */
583 
584     /* If true, copy read backing sectors into image.  Can be >1 if more
585      * than one client has requested copy-on-read.
586      */
587     int copy_on_read;
588 
589     /* If we are reading a disk image, give its size in sectors.
590      * Generally read-only; it is written to by load_vmstate and save_vmstate,
591      * but the block layer is quiescent during those.
592      */
593     int64_t total_sectors;
594 
595     /* Callback before write request is processed */
596     NotifierWithReturnList before_write_notifiers;
597 
598     /* number of in-flight requests; overall and serialising */
599     unsigned int in_flight;
600     unsigned int serialising_in_flight;
601 
602     bool wakeup;
603 
604     /* Offset after the highest byte written to */
605     uint64_t wr_highest_offset;
606 
607     /* threshold limit for writes, in bytes. "High water mark". */
608     uint64_t write_threshold_offset;
609     NotifierWithReturn write_threshold_notifier;
610 
611     /* counter for nested bdrv_io_plug */
612     unsigned io_plugged;
613 
614     QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
615     CoQueue flush_queue;                  /* Serializing flush queue */
616     bool active_flush_req;                /* Flush request in flight? */
617     unsigned int write_gen;               /* Current data generation */
618     unsigned int flushed_gen;             /* Flushed write generation */
619 
620     QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
621 
622     /* do we need to tell the quest if we have a volatile write cache? */
623     int enable_write_cache;
624 
625     int quiesce_counter;
626 };
627 
628 struct BlockBackendRootState {
629     int open_flags;
630     bool read_only;
631     BlockdevDetectZeroesOptions detect_zeroes;
632 };
633 
634 typedef enum BlockMirrorBackingMode {
635     /* Reuse the existing backing chain from the source for the target.
636      * - sync=full: Set backing BDS to NULL.
637      * - sync=top:  Use source's backing BDS.
638      * - sync=none: Use source as the backing BDS. */
639     MIRROR_SOURCE_BACKING_CHAIN,
640 
641     /* Open the target's backing chain completely anew */
642     MIRROR_OPEN_BACKING_CHAIN,
643 
644     /* Do not change the target's backing BDS after job completion */
645     MIRROR_LEAVE_BACKING_CHAIN,
646 } BlockMirrorBackingMode;
647 
648 static inline BlockDriverState *backing_bs(BlockDriverState *bs)
649 {
650     return bs->backing ? bs->backing->bs : NULL;
651 }
652 
653 
654 /* Essential block drivers which must always be statically linked into qemu, and
655  * which therefore can be accessed without using bdrv_find_format() */
656 extern BlockDriver bdrv_file;
657 extern BlockDriver bdrv_raw;
658 extern BlockDriver bdrv_qcow2;
659 
660 int coroutine_fn bdrv_co_preadv(BdrvChild *child,
661     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
662     BdrvRequestFlags flags);
663 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
664     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
665     BdrvRequestFlags flags);
666 
667 int get_tmp_filename(char *filename, int size);
668 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
669                             const char *filename);
670 
671 
672 /**
673  * bdrv_add_before_write_notifier:
674  *
675  * Register a callback that is invoked before write requests are processed but
676  * after any throttling or waiting for overlapping requests.
677  */
678 void bdrv_add_before_write_notifier(BlockDriverState *bs,
679                                     NotifierWithReturn *notifier);
680 
681 /**
682  * bdrv_detach_aio_context:
683  *
684  * May be called from .bdrv_detach_aio_context() to detach children from the
685  * current #AioContext.  This is only needed by block drivers that manage their
686  * own children.  Both ->file and ->backing are automatically handled and
687  * block drivers should not call this function on them explicitly.
688  */
689 void bdrv_detach_aio_context(BlockDriverState *bs);
690 
691 /**
692  * bdrv_attach_aio_context:
693  *
694  * May be called from .bdrv_attach_aio_context() to attach children to the new
695  * #AioContext.  This is only needed by block drivers that manage their own
696  * children.  Both ->file and ->backing are automatically handled and block
697  * drivers should not call this function on them explicitly.
698  */
699 void bdrv_attach_aio_context(BlockDriverState *bs,
700                              AioContext *new_context);
701 
702 /**
703  * bdrv_add_aio_context_notifier:
704  *
705  * If a long-running job intends to be always run in the same AioContext as a
706  * certain BDS, it may use this function to be notified of changes regarding the
707  * association of the BDS to an AioContext.
708  *
709  * attached_aio_context() is called after the target BDS has been attached to a
710  * new AioContext; detach_aio_context() is called before the target BDS is being
711  * detached from its old AioContext.
712  */
713 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
714         void (*attached_aio_context)(AioContext *new_context, void *opaque),
715         void (*detach_aio_context)(void *opaque), void *opaque);
716 
717 /**
718  * bdrv_remove_aio_context_notifier:
719  *
720  * Unsubscribe of change notifications regarding the BDS's AioContext. The
721  * parameters given here have to be the same as those given to
722  * bdrv_add_aio_context_notifier().
723  */
724 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
725                                       void (*aio_context_attached)(AioContext *,
726                                                                    void *),
727                                       void (*aio_context_detached)(void *),
728                                       void *opaque);
729 
730 /**
731  * bdrv_wakeup:
732  * @bs: The BlockDriverState for which an I/O operation has been completed.
733  *
734  * Wake up the main thread if it is waiting on BDRV_POLL_WHILE.  During
735  * synchronous I/O on a BlockDriverState that is attached to another
736  * I/O thread, the main thread lets the I/O thread's event loop run,
737  * waiting for the I/O operation to complete.  A bdrv_wakeup will wake
738  * up the main thread if necessary.
739  *
740  * Manual calls to bdrv_wakeup are rarely necessary, because
741  * bdrv_dec_in_flight already calls it.
742  */
743 void bdrv_wakeup(BlockDriverState *bs);
744 
745 #ifdef _WIN32
746 int is_windows_drive(const char *filename);
747 #endif
748 
749 /**
750  * stream_start:
751  * @job_id: The id of the newly-created job, or %NULL to use the
752  * device name of @bs.
753  * @bs: Block device to operate on.
754  * @base: Block device that will become the new base, or %NULL to
755  * flatten the whole backing file chain onto @bs.
756  * @backing_file_str: The file name that will be written to @bs as the
757  * the new backing file if the job completes. Ignored if @base is %NULL.
758  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
759  * @on_error: The action to take upon error.
760  * @errp: Error object.
761  *
762  * Start a streaming operation on @bs.  Clusters that are unallocated
763  * in @bs, but allocated in any image between @base and @bs (both
764  * exclusive) will be written to @bs.  At the end of a successful
765  * streaming job, the backing file of @bs will be changed to
766  * @backing_file_str in the written image and to @base in the live
767  * BlockDriverState.
768  */
769 void stream_start(const char *job_id, BlockDriverState *bs,
770                   BlockDriverState *base, const char *backing_file_str,
771                   int64_t speed, BlockdevOnError on_error, Error **errp);
772 
773 /**
774  * commit_start:
775  * @job_id: The id of the newly-created job, or %NULL to use the
776  * device name of @bs.
777  * @bs: Active block device.
778  * @top: Top block device to be committed.
779  * @base: Block device that will be written into, and become the new top.
780  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
781  * @on_error: The action to take upon error.
782  * @backing_file_str: String to use as the backing file in @top's overlay
783  * @filter_node_name: The node name that should be assigned to the filter
784  * driver that the commit job inserts into the graph above @top. NULL means
785  * that a node name should be autogenerated.
786  * @errp: Error object.
787  *
788  */
789 void commit_start(const char *job_id, BlockDriverState *bs,
790                   BlockDriverState *base, BlockDriverState *top, int64_t speed,
791                   BlockdevOnError on_error, const char *backing_file_str,
792                   const char *filter_node_name, Error **errp);
793 /**
794  * commit_active_start:
795  * @job_id: The id of the newly-created job, or %NULL to use the
796  * device name of @bs.
797  * @bs: Active block device to be committed.
798  * @base: Block device that will be written into, and become the new top.
799  * @creation_flags: Flags that control the behavior of the Job lifetime.
800  *                  See @BlockJobCreateFlags
801  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
802  * @on_error: The action to take upon error.
803  * @filter_node_name: The node name that should be assigned to the filter
804  * driver that the commit job inserts into the graph above @bs. NULL means that
805  * a node name should be autogenerated.
806  * @cb: Completion function for the job.
807  * @opaque: Opaque pointer value passed to @cb.
808  * @errp: Error object.
809  * @auto_complete: Auto complete the job.
810  *
811  */
812 void commit_active_start(const char *job_id, BlockDriverState *bs,
813                          BlockDriverState *base, int creation_flags,
814                          int64_t speed, BlockdevOnError on_error,
815                          const char *filter_node_name,
816                          BlockCompletionFunc *cb, void *opaque, Error **errp,
817                          bool auto_complete);
818 /*
819  * mirror_start:
820  * @job_id: The id of the newly-created job, or %NULL to use the
821  * device name of @bs.
822  * @bs: Block device to operate on.
823  * @target: Block device to write to.
824  * @replaces: Block graph node name to replace once the mirror is done. Can
825  *            only be used when full mirroring is selected.
826  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
827  * @granularity: The chosen granularity for the dirty bitmap.
828  * @buf_size: The amount of data that can be in flight at one time.
829  * @mode: Whether to collapse all images in the chain to the target.
830  * @backing_mode: How to establish the target's backing chain after completion.
831  * @on_source_error: The action to take upon error reading from the source.
832  * @on_target_error: The action to take upon error writing to the target.
833  * @unmap: Whether to unmap target where source sectors only contain zeroes.
834  * @filter_node_name: The node name that should be assigned to the filter
835  * driver that the mirror job inserts into the graph above @bs. NULL means that
836  * a node name should be autogenerated.
837  * @errp: Error object.
838  *
839  * Start a mirroring operation on @bs.  Clusters that are allocated
840  * in @bs will be written to @target until the job is cancelled or
841  * manually completed.  At the end of a successful mirroring job,
842  * @bs will be switched to read from @target.
843  */
844 void mirror_start(const char *job_id, BlockDriverState *bs,
845                   BlockDriverState *target, const char *replaces,
846                   int64_t speed, uint32_t granularity, int64_t buf_size,
847                   MirrorSyncMode mode, BlockMirrorBackingMode backing_mode,
848                   BlockdevOnError on_source_error,
849                   BlockdevOnError on_target_error,
850                   bool unmap, const char *filter_node_name, Error **errp);
851 
852 /*
853  * backup_job_create:
854  * @job_id: The id of the newly-created job, or %NULL to use the
855  * device name of @bs.
856  * @bs: Block device to operate on.
857  * @target: Block device to write to.
858  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
859  * @sync_mode: What parts of the disk image should be copied to the destination.
860  * @sync_bitmap: The dirty bitmap if sync_mode is MIRROR_SYNC_MODE_INCREMENTAL.
861  * @on_source_error: The action to take upon error reading from the source.
862  * @on_target_error: The action to take upon error writing to the target.
863  * @creation_flags: Flags that control the behavior of the Job lifetime.
864  *                  See @BlockJobCreateFlags
865  * @cb: Completion function for the job.
866  * @opaque: Opaque pointer value passed to @cb.
867  * @txn: Transaction that this job is part of (may be NULL).
868  *
869  * Create a backup operation on @bs.  Clusters in @bs are written to @target
870  * until the job is cancelled or manually completed.
871  */
872 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
873                             BlockDriverState *target, int64_t speed,
874                             MirrorSyncMode sync_mode,
875                             BdrvDirtyBitmap *sync_bitmap,
876                             bool compress,
877                             BlockdevOnError on_source_error,
878                             BlockdevOnError on_target_error,
879                             int creation_flags,
880                             BlockCompletionFunc *cb, void *opaque,
881                             BlockJobTxn *txn, Error **errp);
882 
883 void hmp_drive_add_node(Monitor *mon, const char *optstr);
884 
885 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
886                                   const char *child_name,
887                                   const BdrvChildRole *child_role,
888                                   uint64_t perm, uint64_t shared_perm,
889                                   void *opaque, Error **errp);
890 void bdrv_root_unref_child(BdrvChild *child);
891 
892 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
893                             Error **errp);
894 
895 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
896  * block filters: Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED and RESIZE to
897  * all children */
898 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
899                                const BdrvChildRole *role,
900                                uint64_t perm, uint64_t shared,
901                                uint64_t *nperm, uint64_t *nshared);
902 
903 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
904  * (non-raw) image formats: Like above for bs->backing, but for bs->file it
905  * requires WRITE | RESIZE for read-write images, always requires
906  * CONSISTENT_READ and doesn't share WRITE. */
907 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
908                                const BdrvChildRole *role,
909                                uint64_t perm, uint64_t shared,
910                                uint64_t *nperm, uint64_t *nshared);
911 
912 const char *bdrv_get_parent_name(const BlockDriverState *bs);
913 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp);
914 bool blk_dev_has_removable_media(BlockBackend *blk);
915 bool blk_dev_has_tray(BlockBackend *blk);
916 void blk_dev_eject_request(BlockBackend *blk, bool force);
917 bool blk_dev_is_tray_open(BlockBackend *blk);
918 bool blk_dev_is_medium_locked(BlockBackend *blk);
919 
920 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector, int64_t nr_sect);
921 bool bdrv_requests_pending(BlockDriverState *bs);
922 
923 void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap **out);
924 void bdrv_undo_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap *in);
925 
926 void bdrv_inc_in_flight(BlockDriverState *bs);
927 void bdrv_dec_in_flight(BlockDriverState *bs);
928 
929 void blockdev_close_all_bdrv_states(void);
930 
931 #endif /* BLOCK_INT_H */
932