xref: /openbmc/qemu/block/file-posix.c (revision 80adf54e)
1 /*
2  * Block driver for RAW files (posix)
3  *
4  * Copyright (c) 2006 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 "qemu/osdep.h"
25 #include "qapi/error.h"
26 #include "qemu/cutils.h"
27 #include "qemu/error-report.h"
28 #include "block/block_int.h"
29 #include "qemu/module.h"
30 #include "trace.h"
31 #include "block/thread-pool.h"
32 #include "qemu/iov.h"
33 #include "block/raw-aio.h"
34 #include "qapi/util.h"
35 #include "qapi/qmp/qstring.h"
36 
37 #if defined(__APPLE__) && (__MACH__)
38 #include <paths.h>
39 #include <sys/param.h>
40 #include <IOKit/IOKitLib.h>
41 #include <IOKit/IOBSD.h>
42 #include <IOKit/storage/IOMediaBSDClient.h>
43 #include <IOKit/storage/IOMedia.h>
44 #include <IOKit/storage/IOCDMedia.h>
45 //#include <IOKit/storage/IOCDTypes.h>
46 #include <IOKit/storage/IODVDMedia.h>
47 #include <CoreFoundation/CoreFoundation.h>
48 #endif
49 
50 #ifdef __sun__
51 #define _POSIX_PTHREAD_SEMANTICS 1
52 #include <sys/dkio.h>
53 #endif
54 #ifdef __linux__
55 #include <sys/ioctl.h>
56 #include <sys/param.h>
57 #include <linux/cdrom.h>
58 #include <linux/fd.h>
59 #include <linux/fs.h>
60 #include <linux/hdreg.h>
61 #include <scsi/sg.h>
62 #ifdef __s390__
63 #include <asm/dasd.h>
64 #endif
65 #ifndef FS_NOCOW_FL
66 #define FS_NOCOW_FL                     0x00800000 /* Do not cow file */
67 #endif
68 #endif
69 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
70 #include <linux/falloc.h>
71 #endif
72 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
73 #include <sys/disk.h>
74 #include <sys/cdio.h>
75 #endif
76 
77 #ifdef __OpenBSD__
78 #include <sys/ioctl.h>
79 #include <sys/disklabel.h>
80 #include <sys/dkio.h>
81 #endif
82 
83 #ifdef __NetBSD__
84 #include <sys/ioctl.h>
85 #include <sys/disklabel.h>
86 #include <sys/dkio.h>
87 #include <sys/disk.h>
88 #endif
89 
90 #ifdef __DragonFly__
91 #include <sys/ioctl.h>
92 #include <sys/diskslice.h>
93 #endif
94 
95 #ifdef CONFIG_XFS
96 #include <xfs/xfs.h>
97 #endif
98 
99 //#define DEBUG_BLOCK
100 
101 #ifdef DEBUG_BLOCK
102 # define DEBUG_BLOCK_PRINT 1
103 #else
104 # define DEBUG_BLOCK_PRINT 0
105 #endif
106 #define DPRINTF(fmt, ...) \
107 do { \
108     if (DEBUG_BLOCK_PRINT) { \
109         printf(fmt, ## __VA_ARGS__); \
110     } \
111 } while (0)
112 
113 /* OS X does not have O_DSYNC */
114 #ifndef O_DSYNC
115 #ifdef O_SYNC
116 #define O_DSYNC O_SYNC
117 #elif defined(O_FSYNC)
118 #define O_DSYNC O_FSYNC
119 #endif
120 #endif
121 
122 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
123 #ifndef O_DIRECT
124 #define O_DIRECT O_DSYNC
125 #endif
126 
127 #define FTYPE_FILE   0
128 #define FTYPE_CD     1
129 
130 #define MAX_BLOCKSIZE	4096
131 
132 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
133  * leaving a few more bytes for its future use. */
134 #define RAW_LOCK_PERM_BASE             100
135 #define RAW_LOCK_SHARED_BASE           200
136 
137 typedef struct BDRVRawState {
138     int fd;
139     int lock_fd;
140     bool use_lock;
141     int type;
142     int open_flags;
143     size_t buf_align;
144 
145     /* The current permissions. */
146     uint64_t perm;
147     uint64_t shared_perm;
148 
149 #ifdef CONFIG_XFS
150     bool is_xfs:1;
151 #endif
152     bool has_discard:1;
153     bool has_write_zeroes:1;
154     bool discard_zeroes:1;
155     bool use_linux_aio:1;
156     bool page_cache_inconsistent:1;
157     bool has_fallocate;
158     bool needs_alignment;
159 } BDRVRawState;
160 
161 typedef struct BDRVRawReopenState {
162     int fd;
163     int open_flags;
164 } BDRVRawReopenState;
165 
166 static int fd_open(BlockDriverState *bs);
167 static int64_t raw_getlength(BlockDriverState *bs);
168 
169 typedef struct RawPosixAIOData {
170     BlockDriverState *bs;
171     int aio_fildes;
172     union {
173         struct iovec *aio_iov;
174         void *aio_ioctl_buf;
175     };
176     int aio_niov;
177     uint64_t aio_nbytes;
178 #define aio_ioctl_cmd   aio_nbytes /* for QEMU_AIO_IOCTL */
179     off_t aio_offset;
180     int aio_type;
181 } RawPosixAIOData;
182 
183 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
184 static int cdrom_reopen(BlockDriverState *bs);
185 #endif
186 
187 #if defined(__NetBSD__)
188 static int raw_normalize_devicepath(const char **filename)
189 {
190     static char namebuf[PATH_MAX];
191     const char *dp, *fname;
192     struct stat sb;
193 
194     fname = *filename;
195     dp = strrchr(fname, '/');
196     if (lstat(fname, &sb) < 0) {
197         fprintf(stderr, "%s: stat failed: %s\n",
198             fname, strerror(errno));
199         return -errno;
200     }
201 
202     if (!S_ISBLK(sb.st_mode)) {
203         return 0;
204     }
205 
206     if (dp == NULL) {
207         snprintf(namebuf, PATH_MAX, "r%s", fname);
208     } else {
209         snprintf(namebuf, PATH_MAX, "%.*s/r%s",
210             (int)(dp - fname), fname, dp + 1);
211     }
212     fprintf(stderr, "%s is a block device", fname);
213     *filename = namebuf;
214     fprintf(stderr, ", using %s\n", *filename);
215 
216     return 0;
217 }
218 #else
219 static int raw_normalize_devicepath(const char **filename)
220 {
221     return 0;
222 }
223 #endif
224 
225 /*
226  * Get logical block size via ioctl. On success store it in @sector_size_p.
227  */
228 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
229 {
230     unsigned int sector_size;
231     bool success = false;
232     int i;
233 
234     errno = ENOTSUP;
235     static const unsigned long ioctl_list[] = {
236 #ifdef BLKSSZGET
237         BLKSSZGET,
238 #endif
239 #ifdef DKIOCGETBLOCKSIZE
240         DKIOCGETBLOCKSIZE,
241 #endif
242 #ifdef DIOCGSECTORSIZE
243         DIOCGSECTORSIZE,
244 #endif
245     };
246 
247     /* Try a few ioctls to get the right size */
248     for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) {
249         if (ioctl(fd, ioctl_list[i], &sector_size) >= 0) {
250             *sector_size_p = sector_size;
251             success = true;
252         }
253     }
254 
255     return success ? 0 : -errno;
256 }
257 
258 /**
259  * Get physical block size of @fd.
260  * On success, store it in @blk_size and return 0.
261  * On failure, return -errno.
262  */
263 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
264 {
265 #ifdef BLKPBSZGET
266     if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
267         return -errno;
268     }
269     return 0;
270 #else
271     return -ENOTSUP;
272 #endif
273 }
274 
275 /* Check if read is allowed with given memory buffer and length.
276  *
277  * This function is used to check O_DIRECT memory buffer and request alignment.
278  */
279 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
280 {
281     ssize_t ret = pread(fd, buf, len, 0);
282 
283     if (ret >= 0) {
284         return true;
285     }
286 
287 #ifdef __linux__
288     /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads.  Ignore
289      * other errors (e.g. real I/O error), which could happen on a failed
290      * drive, since we only care about probing alignment.
291      */
292     if (errno != EINVAL) {
293         return true;
294     }
295 #endif
296 
297     return false;
298 }
299 
300 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
301 {
302     BDRVRawState *s = bs->opaque;
303     char *buf;
304     size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize());
305 
306     /* For SCSI generic devices the alignment is not really used.
307        With buffered I/O, we don't have any restrictions. */
308     if (bdrv_is_sg(bs) || !s->needs_alignment) {
309         bs->bl.request_alignment = 1;
310         s->buf_align = 1;
311         return;
312     }
313 
314     bs->bl.request_alignment = 0;
315     s->buf_align = 0;
316     /* Let's try to use the logical blocksize for the alignment. */
317     if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) {
318         bs->bl.request_alignment = 0;
319     }
320 #ifdef CONFIG_XFS
321     if (s->is_xfs) {
322         struct dioattr da;
323         if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
324             bs->bl.request_alignment = da.d_miniosz;
325             /* The kernel returns wrong information for d_mem */
326             /* s->buf_align = da.d_mem; */
327         }
328     }
329 #endif
330 
331     /* If we could not get the sizes so far, we can only guess them */
332     if (!s->buf_align) {
333         size_t align;
334         buf = qemu_memalign(max_align, 2 * max_align);
335         for (align = 512; align <= max_align; align <<= 1) {
336             if (raw_is_io_aligned(fd, buf + align, max_align)) {
337                 s->buf_align = align;
338                 break;
339             }
340         }
341         qemu_vfree(buf);
342     }
343 
344     if (!bs->bl.request_alignment) {
345         size_t align;
346         buf = qemu_memalign(s->buf_align, max_align);
347         for (align = 512; align <= max_align; align <<= 1) {
348             if (raw_is_io_aligned(fd, buf, align)) {
349                 bs->bl.request_alignment = align;
350                 break;
351             }
352         }
353         qemu_vfree(buf);
354     }
355 
356     if (!s->buf_align || !bs->bl.request_alignment) {
357         error_setg(errp, "Could not find working O_DIRECT alignment");
358         error_append_hint(errp, "Try cache.direct=off\n");
359     }
360 }
361 
362 static void raw_parse_flags(int bdrv_flags, int *open_flags)
363 {
364     assert(open_flags != NULL);
365 
366     *open_flags |= O_BINARY;
367     *open_flags &= ~O_ACCMODE;
368     if (bdrv_flags & BDRV_O_RDWR) {
369         *open_flags |= O_RDWR;
370     } else {
371         *open_flags |= O_RDONLY;
372     }
373 
374     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
375      * and O_DIRECT for no caching. */
376     if ((bdrv_flags & BDRV_O_NOCACHE)) {
377         *open_flags |= O_DIRECT;
378     }
379 }
380 
381 static void raw_parse_filename(const char *filename, QDict *options,
382                                Error **errp)
383 {
384     bdrv_parse_filename_strip_prefix(filename, "file:", options);
385 }
386 
387 static QemuOptsList raw_runtime_opts = {
388     .name = "raw",
389     .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
390     .desc = {
391         {
392             .name = "filename",
393             .type = QEMU_OPT_STRING,
394             .help = "File name of the image",
395         },
396         {
397             .name = "aio",
398             .type = QEMU_OPT_STRING,
399             .help = "host AIO implementation (threads, native)",
400         },
401         {
402             .name = "locking",
403             .type = QEMU_OPT_STRING,
404             .help = "file locking mode (on/off/auto, default: auto)",
405         },
406         { /* end of list */ }
407     },
408 };
409 
410 static int raw_open_common(BlockDriverState *bs, QDict *options,
411                            int bdrv_flags, int open_flags, Error **errp)
412 {
413     BDRVRawState *s = bs->opaque;
414     QemuOpts *opts;
415     Error *local_err = NULL;
416     const char *filename = NULL;
417     BlockdevAioOptions aio, aio_default;
418     int fd, ret;
419     struct stat st;
420     OnOffAuto locking;
421 
422     opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
423     qemu_opts_absorb_qdict(opts, options, &local_err);
424     if (local_err) {
425         error_propagate(errp, local_err);
426         ret = -EINVAL;
427         goto fail;
428     }
429 
430     filename = qemu_opt_get(opts, "filename");
431 
432     ret = raw_normalize_devicepath(&filename);
433     if (ret != 0) {
434         error_setg_errno(errp, -ret, "Could not normalize device path");
435         goto fail;
436     }
437 
438     aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO)
439                   ? BLOCKDEV_AIO_OPTIONS_NATIVE
440                   : BLOCKDEV_AIO_OPTIONS_THREADS;
441     aio = qapi_enum_parse(BlockdevAioOptions_lookup, qemu_opt_get(opts, "aio"),
442                           BLOCKDEV_AIO_OPTIONS__MAX, aio_default, &local_err);
443     if (local_err) {
444         error_propagate(errp, local_err);
445         ret = -EINVAL;
446         goto fail;
447     }
448     s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE);
449 
450     locking = qapi_enum_parse(OnOffAuto_lookup, qemu_opt_get(opts, "locking"),
451                               ON_OFF_AUTO__MAX, ON_OFF_AUTO_AUTO, &local_err);
452     if (local_err) {
453         error_propagate(errp, local_err);
454         ret = -EINVAL;
455         goto fail;
456     }
457     switch (locking) {
458     case ON_OFF_AUTO_ON:
459         s->use_lock = true;
460         if (!qemu_has_ofd_lock()) {
461             fprintf(stderr,
462                     "File lock requested but OFD locking syscall is "
463                     "unavailable, falling back to POSIX file locks.\n"
464                     "Due to the implementation, locks can be lost "
465                     "unexpectedly.\n");
466         }
467         break;
468     case ON_OFF_AUTO_OFF:
469         s->use_lock = false;
470         break;
471     case ON_OFF_AUTO_AUTO:
472         s->use_lock = qemu_has_ofd_lock();
473         break;
474     default:
475         abort();
476     }
477 
478     s->open_flags = open_flags;
479     raw_parse_flags(bdrv_flags, &s->open_flags);
480 
481     s->fd = -1;
482     fd = qemu_open(filename, s->open_flags, 0644);
483     if (fd < 0) {
484         ret = -errno;
485         error_setg_errno(errp, errno, "Could not open '%s'", filename);
486         if (ret == -EROFS) {
487             ret = -EACCES;
488         }
489         goto fail;
490     }
491     s->fd = fd;
492 
493     s->lock_fd = -1;
494     if (s->use_lock) {
495         fd = qemu_open(filename, s->open_flags);
496         if (fd < 0) {
497             ret = -errno;
498             error_setg_errno(errp, errno, "Could not open '%s' for locking",
499                              filename);
500             qemu_close(s->fd);
501             goto fail;
502         }
503         s->lock_fd = fd;
504     }
505     s->perm = 0;
506     s->shared_perm = BLK_PERM_ALL;
507 
508 #ifdef CONFIG_LINUX_AIO
509      /* Currently Linux does AIO only for files opened with O_DIRECT */
510     if (s->use_linux_aio && !(s->open_flags & O_DIRECT)) {
511         error_setg(errp, "aio=native was specified, but it requires "
512                          "cache.direct=on, which was not specified.");
513         ret = -EINVAL;
514         goto fail;
515     }
516 #else
517     if (s->use_linux_aio) {
518         error_setg(errp, "aio=native was specified, but is not supported "
519                          "in this build.");
520         ret = -EINVAL;
521         goto fail;
522     }
523 #endif /* !defined(CONFIG_LINUX_AIO) */
524 
525     s->has_discard = true;
526     s->has_write_zeroes = true;
527     bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
528     if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
529         s->needs_alignment = true;
530     }
531 
532     if (fstat(s->fd, &st) < 0) {
533         ret = -errno;
534         error_setg_errno(errp, errno, "Could not stat file");
535         goto fail;
536     }
537     if (S_ISREG(st.st_mode)) {
538         s->discard_zeroes = true;
539         s->has_fallocate = true;
540     }
541     if (S_ISBLK(st.st_mode)) {
542 #ifdef BLKDISCARDZEROES
543         unsigned int arg;
544         if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
545             s->discard_zeroes = true;
546         }
547 #endif
548 #ifdef __linux__
549         /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache.  Do
550          * not rely on the contents of discarded blocks unless using O_DIRECT.
551          * Same for BLKZEROOUT.
552          */
553         if (!(bs->open_flags & BDRV_O_NOCACHE)) {
554             s->discard_zeroes = false;
555             s->has_write_zeroes = false;
556         }
557 #endif
558     }
559 #ifdef __FreeBSD__
560     if (S_ISCHR(st.st_mode)) {
561         /*
562          * The file is a char device (disk), which on FreeBSD isn't behind
563          * a pager, so force all requests to be aligned. This is needed
564          * so QEMU makes sure all IO operations on the device are aligned
565          * to sector size, or else FreeBSD will reject them with EINVAL.
566          */
567         s->needs_alignment = true;
568     }
569 #endif
570 
571 #ifdef CONFIG_XFS
572     if (platform_test_xfs_fd(s->fd)) {
573         s->is_xfs = true;
574     }
575 #endif
576 
577     ret = 0;
578 fail:
579     if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
580         unlink(filename);
581     }
582     qemu_opts_del(opts);
583     return ret;
584 }
585 
586 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
587                     Error **errp)
588 {
589     BDRVRawState *s = bs->opaque;
590 
591     s->type = FTYPE_FILE;
592     return raw_open_common(bs, options, flags, 0, errp);
593 }
594 
595 typedef enum {
596     RAW_PL_PREPARE,
597     RAW_PL_COMMIT,
598     RAW_PL_ABORT,
599 } RawPermLockOp;
600 
601 #define PERM_FOREACH(i) \
602     for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
603 
604 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
605  * file; if @unlock == true, also unlock the unneeded bytes.
606  * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
607  */
608 static int raw_apply_lock_bytes(BDRVRawState *s,
609                                 uint64_t perm_lock_bits,
610                                 uint64_t shared_perm_lock_bits,
611                                 bool unlock, Error **errp)
612 {
613     int ret;
614     int i;
615 
616     PERM_FOREACH(i) {
617         int off = RAW_LOCK_PERM_BASE + i;
618         if (perm_lock_bits & (1ULL << i)) {
619             ret = qemu_lock_fd(s->lock_fd, off, 1, false);
620             if (ret) {
621                 error_setg(errp, "Failed to lock byte %d", off);
622                 return ret;
623             }
624         } else if (unlock) {
625             ret = qemu_unlock_fd(s->lock_fd, off, 1);
626             if (ret) {
627                 error_setg(errp, "Failed to unlock byte %d", off);
628                 return ret;
629             }
630         }
631     }
632     PERM_FOREACH(i) {
633         int off = RAW_LOCK_SHARED_BASE + i;
634         if (shared_perm_lock_bits & (1ULL << i)) {
635             ret = qemu_lock_fd(s->lock_fd, off, 1, false);
636             if (ret) {
637                 error_setg(errp, "Failed to lock byte %d", off);
638                 return ret;
639             }
640         } else if (unlock) {
641             ret = qemu_unlock_fd(s->lock_fd, off, 1);
642             if (ret) {
643                 error_setg(errp, "Failed to unlock byte %d", off);
644                 return ret;
645             }
646         }
647     }
648     return 0;
649 }
650 
651 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
652 static int raw_check_lock_bytes(BDRVRawState *s,
653                                 uint64_t perm, uint64_t shared_perm,
654                                 Error **errp)
655 {
656     int ret;
657     int i;
658 
659     PERM_FOREACH(i) {
660         int off = RAW_LOCK_SHARED_BASE + i;
661         uint64_t p = 1ULL << i;
662         if (perm & p) {
663             ret = qemu_lock_fd_test(s->lock_fd, off, 1, true);
664             if (ret) {
665                 char *perm_name = bdrv_perm_names(p);
666                 error_setg(errp,
667                            "Failed to get \"%s\" lock",
668                            perm_name);
669                 g_free(perm_name);
670                 error_append_hint(errp,
671                                   "Is another process using the image?\n");
672                 return ret;
673             }
674         }
675     }
676     PERM_FOREACH(i) {
677         int off = RAW_LOCK_PERM_BASE + i;
678         uint64_t p = 1ULL << i;
679         if (!(shared_perm & p)) {
680             ret = qemu_lock_fd_test(s->lock_fd, off, 1, true);
681             if (ret) {
682                 char *perm_name = bdrv_perm_names(p);
683                 error_setg(errp,
684                            "Failed to get shared \"%s\" lock",
685                            perm_name);
686                 g_free(perm_name);
687                 error_append_hint(errp,
688                                   "Is another process using the image?\n");
689                 return ret;
690             }
691         }
692     }
693     return 0;
694 }
695 
696 static int raw_handle_perm_lock(BlockDriverState *bs,
697                                 RawPermLockOp op,
698                                 uint64_t new_perm, uint64_t new_shared,
699                                 Error **errp)
700 {
701     BDRVRawState *s = bs->opaque;
702     int ret = 0;
703     Error *local_err = NULL;
704 
705     if (!s->use_lock) {
706         return 0;
707     }
708 
709     if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) {
710         return 0;
711     }
712 
713     assert(s->lock_fd > 0);
714 
715     switch (op) {
716     case RAW_PL_PREPARE:
717         ret = raw_apply_lock_bytes(s, s->perm | new_perm,
718                                    ~s->shared_perm | ~new_shared,
719                                    false, errp);
720         if (!ret) {
721             ret = raw_check_lock_bytes(s, new_perm, new_shared, errp);
722             if (!ret) {
723                 return 0;
724             }
725         }
726         op = RAW_PL_ABORT;
727         /* fall through to unlock bytes. */
728     case RAW_PL_ABORT:
729         raw_apply_lock_bytes(s, s->perm, ~s->shared_perm, true, &local_err);
730         if (local_err) {
731             /* Theoretically the above call only unlocks bytes and it cannot
732              * fail. Something weird happened, report it.
733              */
734             error_report_err(local_err);
735         }
736         break;
737     case RAW_PL_COMMIT:
738         raw_apply_lock_bytes(s, new_perm, ~new_shared, true, &local_err);
739         if (local_err) {
740             /* Theoretically the above call only unlocks bytes and it cannot
741              * fail. Something weird happened, report it.
742              */
743             error_report_err(local_err);
744         }
745         break;
746     }
747     return ret;
748 }
749 
750 static int raw_reopen_prepare(BDRVReopenState *state,
751                               BlockReopenQueue *queue, Error **errp)
752 {
753     BDRVRawState *s;
754     BDRVRawReopenState *rs;
755     int ret = 0;
756     Error *local_err = NULL;
757 
758     assert(state != NULL);
759     assert(state->bs != NULL);
760 
761     s = state->bs->opaque;
762 
763     state->opaque = g_new0(BDRVRawReopenState, 1);
764     rs = state->opaque;
765 
766     if (s->type == FTYPE_CD) {
767         rs->open_flags |= O_NONBLOCK;
768     }
769 
770     raw_parse_flags(state->flags, &rs->open_flags);
771 
772     rs->fd = -1;
773 
774     int fcntl_flags = O_APPEND | O_NONBLOCK;
775 #ifdef O_NOATIME
776     fcntl_flags |= O_NOATIME;
777 #endif
778 
779 #ifdef O_ASYNC
780     /* Not all operating systems have O_ASYNC, and those that don't
781      * will not let us track the state into rs->open_flags (typically
782      * you achieve the same effect with an ioctl, for example I_SETSIG
783      * on Solaris). But we do not use O_ASYNC, so that's fine.
784      */
785     assert((s->open_flags & O_ASYNC) == 0);
786 #endif
787 
788     if ((rs->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
789         /* dup the original fd */
790         rs->fd = qemu_dup(s->fd);
791         if (rs->fd >= 0) {
792             ret = fcntl_setfl(rs->fd, rs->open_flags);
793             if (ret) {
794                 qemu_close(rs->fd);
795                 rs->fd = -1;
796             }
797         }
798     }
799 
800     /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
801     if (rs->fd == -1) {
802         const char *normalized_filename = state->bs->filename;
803         ret = raw_normalize_devicepath(&normalized_filename);
804         if (ret < 0) {
805             error_setg_errno(errp, -ret, "Could not normalize device path");
806         } else {
807             assert(!(rs->open_flags & O_CREAT));
808             rs->fd = qemu_open(normalized_filename, rs->open_flags);
809             if (rs->fd == -1) {
810                 error_setg_errno(errp, errno, "Could not reopen file");
811                 ret = -1;
812             }
813         }
814     }
815 
816     /* Fail already reopen_prepare() if we can't get a working O_DIRECT
817      * alignment with the new fd. */
818     if (rs->fd != -1) {
819         raw_probe_alignment(state->bs, rs->fd, &local_err);
820         if (local_err) {
821             qemu_close(rs->fd);
822             rs->fd = -1;
823             error_propagate(errp, local_err);
824             ret = -EINVAL;
825         }
826     }
827 
828     return ret;
829 }
830 
831 static void raw_reopen_commit(BDRVReopenState *state)
832 {
833     BDRVRawReopenState *rs = state->opaque;
834     BDRVRawState *s = state->bs->opaque;
835 
836     s->open_flags = rs->open_flags;
837 
838     qemu_close(s->fd);
839     s->fd = rs->fd;
840 
841     g_free(state->opaque);
842     state->opaque = NULL;
843 }
844 
845 
846 static void raw_reopen_abort(BDRVReopenState *state)
847 {
848     BDRVRawReopenState *rs = state->opaque;
849 
850      /* nothing to do if NULL, we didn't get far enough */
851     if (rs == NULL) {
852         return;
853     }
854 
855     if (rs->fd >= 0) {
856         qemu_close(rs->fd);
857         rs->fd = -1;
858     }
859     g_free(state->opaque);
860     state->opaque = NULL;
861 }
862 
863 static int hdev_get_max_transfer_length(BlockDriverState *bs, int fd)
864 {
865 #ifdef BLKSECTGET
866     int max_bytes = 0;
867     short max_sectors = 0;
868     if (bs->sg && ioctl(fd, BLKSECTGET, &max_bytes) == 0) {
869         return max_bytes;
870     } else if (!bs->sg && ioctl(fd, BLKSECTGET, &max_sectors) == 0) {
871         return max_sectors << BDRV_SECTOR_BITS;
872     } else {
873         return -errno;
874     }
875 #else
876     return -ENOSYS;
877 #endif
878 }
879 
880 static int hdev_get_max_segments(const struct stat *st)
881 {
882 #ifdef CONFIG_LINUX
883     char buf[32];
884     const char *end;
885     char *sysfspath;
886     int ret;
887     int fd = -1;
888     long max_segments;
889 
890     sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
891                                 major(st->st_rdev), minor(st->st_rdev));
892     fd = open(sysfspath, O_RDONLY);
893     if (fd == -1) {
894         ret = -errno;
895         goto out;
896     }
897     do {
898         ret = read(fd, buf, sizeof(buf) - 1);
899     } while (ret == -1 && errno == EINTR);
900     if (ret < 0) {
901         ret = -errno;
902         goto out;
903     } else if (ret == 0) {
904         ret = -EIO;
905         goto out;
906     }
907     buf[ret] = 0;
908     /* The file is ended with '\n', pass 'end' to accept that. */
909     ret = qemu_strtol(buf, &end, 10, &max_segments);
910     if (ret == 0 && end && *end == '\n') {
911         ret = max_segments;
912     }
913 
914 out:
915     if (fd != -1) {
916         close(fd);
917     }
918     g_free(sysfspath);
919     return ret;
920 #else
921     return -ENOTSUP;
922 #endif
923 }
924 
925 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
926 {
927     BDRVRawState *s = bs->opaque;
928     struct stat st;
929 
930     if (!fstat(s->fd, &st)) {
931         if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode)) {
932             int ret = hdev_get_max_transfer_length(bs, s->fd);
933             if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) {
934                 bs->bl.max_transfer = pow2floor(ret);
935             }
936             ret = hdev_get_max_segments(&st);
937             if (ret > 0) {
938                 bs->bl.max_transfer = MIN(bs->bl.max_transfer,
939                                           ret * getpagesize());
940             }
941         }
942     }
943 
944     raw_probe_alignment(bs, s->fd, errp);
945     bs->bl.min_mem_alignment = s->buf_align;
946     bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
947 }
948 
949 static int check_for_dasd(int fd)
950 {
951 #ifdef BIODASDINFO2
952     struct dasd_information2_t info = {0};
953 
954     return ioctl(fd, BIODASDINFO2, &info);
955 #else
956     return -1;
957 #endif
958 }
959 
960 /**
961  * Try to get @bs's logical and physical block size.
962  * On success, store them in @bsz and return zero.
963  * On failure, return negative errno.
964  */
965 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
966 {
967     BDRVRawState *s = bs->opaque;
968     int ret;
969 
970     /* If DASD, get blocksizes */
971     if (check_for_dasd(s->fd) < 0) {
972         return -ENOTSUP;
973     }
974     ret = probe_logical_blocksize(s->fd, &bsz->log);
975     if (ret < 0) {
976         return ret;
977     }
978     return probe_physical_blocksize(s->fd, &bsz->phys);
979 }
980 
981 /**
982  * Try to get @bs's geometry: cyls, heads, sectors.
983  * On success, store them in @geo and return 0.
984  * On failure return -errno.
985  * (Allows block driver to assign default geometry values that guest sees)
986  */
987 #ifdef __linux__
988 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
989 {
990     BDRVRawState *s = bs->opaque;
991     struct hd_geometry ioctl_geo = {0};
992 
993     /* If DASD, get its geometry */
994     if (check_for_dasd(s->fd) < 0) {
995         return -ENOTSUP;
996     }
997     if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
998         return -errno;
999     }
1000     /* HDIO_GETGEO may return success even though geo contains zeros
1001        (e.g. certain multipath setups) */
1002     if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
1003         return -ENOTSUP;
1004     }
1005     /* Do not return a geometry for partition */
1006     if (ioctl_geo.start != 0) {
1007         return -ENOTSUP;
1008     }
1009     geo->heads = ioctl_geo.heads;
1010     geo->sectors = ioctl_geo.sectors;
1011     geo->cylinders = ioctl_geo.cylinders;
1012 
1013     return 0;
1014 }
1015 #else /* __linux__ */
1016 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1017 {
1018     return -ENOTSUP;
1019 }
1020 #endif
1021 
1022 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
1023 {
1024     int ret;
1025 
1026     ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
1027     if (ret == -1) {
1028         return -errno;
1029     }
1030 
1031     return 0;
1032 }
1033 
1034 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
1035 {
1036     BDRVRawState *s = aiocb->bs->opaque;
1037     int ret;
1038 
1039     if (s->page_cache_inconsistent) {
1040         return -EIO;
1041     }
1042 
1043     ret = qemu_fdatasync(aiocb->aio_fildes);
1044     if (ret == -1) {
1045         /* There is no clear definition of the semantics of a failing fsync(),
1046          * so we may have to assume the worst. The sad truth is that this
1047          * assumption is correct for Linux. Some pages are now probably marked
1048          * clean in the page cache even though they are inconsistent with the
1049          * on-disk contents. The next fdatasync() call would succeed, but no
1050          * further writeback attempt will be made. We can't get back to a state
1051          * in which we know what is on disk (we would have to rewrite
1052          * everything that was touched since the last fdatasync() at least), so
1053          * make bdrv_flush() fail permanently. Given that the behaviour isn't
1054          * really defined, I have little hope that other OSes are doing better.
1055          *
1056          * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1057          * cache. */
1058         if ((s->open_flags & O_DIRECT) == 0) {
1059             s->page_cache_inconsistent = true;
1060         }
1061         return -errno;
1062     }
1063     return 0;
1064 }
1065 
1066 #ifdef CONFIG_PREADV
1067 
1068 static bool preadv_present = true;
1069 
1070 static ssize_t
1071 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1072 {
1073     return preadv(fd, iov, nr_iov, offset);
1074 }
1075 
1076 static ssize_t
1077 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1078 {
1079     return pwritev(fd, iov, nr_iov, offset);
1080 }
1081 
1082 #else
1083 
1084 static bool preadv_present = false;
1085 
1086 static ssize_t
1087 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1088 {
1089     return -ENOSYS;
1090 }
1091 
1092 static ssize_t
1093 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1094 {
1095     return -ENOSYS;
1096 }
1097 
1098 #endif
1099 
1100 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
1101 {
1102     ssize_t len;
1103 
1104     do {
1105         if (aiocb->aio_type & QEMU_AIO_WRITE)
1106             len = qemu_pwritev(aiocb->aio_fildes,
1107                                aiocb->aio_iov,
1108                                aiocb->aio_niov,
1109                                aiocb->aio_offset);
1110          else
1111             len = qemu_preadv(aiocb->aio_fildes,
1112                               aiocb->aio_iov,
1113                               aiocb->aio_niov,
1114                               aiocb->aio_offset);
1115     } while (len == -1 && errno == EINTR);
1116 
1117     if (len == -1) {
1118         return -errno;
1119     }
1120     return len;
1121 }
1122 
1123 /*
1124  * Read/writes the data to/from a given linear buffer.
1125  *
1126  * Returns the number of bytes handles or -errno in case of an error. Short
1127  * reads are only returned if the end of the file is reached.
1128  */
1129 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
1130 {
1131     ssize_t offset = 0;
1132     ssize_t len;
1133 
1134     while (offset < aiocb->aio_nbytes) {
1135         if (aiocb->aio_type & QEMU_AIO_WRITE) {
1136             len = pwrite(aiocb->aio_fildes,
1137                          (const char *)buf + offset,
1138                          aiocb->aio_nbytes - offset,
1139                          aiocb->aio_offset + offset);
1140         } else {
1141             len = pread(aiocb->aio_fildes,
1142                         buf + offset,
1143                         aiocb->aio_nbytes - offset,
1144                         aiocb->aio_offset + offset);
1145         }
1146         if (len == -1 && errno == EINTR) {
1147             continue;
1148         } else if (len == -1 && errno == EINVAL &&
1149                    (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
1150                    !(aiocb->aio_type & QEMU_AIO_WRITE) &&
1151                    offset > 0) {
1152             /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1153              * after a short read.  Assume that O_DIRECT short reads only occur
1154              * at EOF.  Therefore this is a short read, not an I/O error.
1155              */
1156             break;
1157         } else if (len == -1) {
1158             offset = -errno;
1159             break;
1160         } else if (len == 0) {
1161             break;
1162         }
1163         offset += len;
1164     }
1165 
1166     return offset;
1167 }
1168 
1169 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
1170 {
1171     ssize_t nbytes;
1172     char *buf;
1173 
1174     if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
1175         /*
1176          * If there is just a single buffer, and it is properly aligned
1177          * we can just use plain pread/pwrite without any problems.
1178          */
1179         if (aiocb->aio_niov == 1) {
1180              return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
1181         }
1182         /*
1183          * We have more than one iovec, and all are properly aligned.
1184          *
1185          * Try preadv/pwritev first and fall back to linearizing the
1186          * buffer if it's not supported.
1187          */
1188         if (preadv_present) {
1189             nbytes = handle_aiocb_rw_vector(aiocb);
1190             if (nbytes == aiocb->aio_nbytes ||
1191                 (nbytes < 0 && nbytes != -ENOSYS)) {
1192                 return nbytes;
1193             }
1194             preadv_present = false;
1195         }
1196 
1197         /*
1198          * XXX(hch): short read/write.  no easy way to handle the reminder
1199          * using these interfaces.  For now retry using plain
1200          * pread/pwrite?
1201          */
1202     }
1203 
1204     /*
1205      * Ok, we have to do it the hard way, copy all segments into
1206      * a single aligned buffer.
1207      */
1208     buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
1209     if (buf == NULL) {
1210         return -ENOMEM;
1211     }
1212 
1213     if (aiocb->aio_type & QEMU_AIO_WRITE) {
1214         char *p = buf;
1215         int i;
1216 
1217         for (i = 0; i < aiocb->aio_niov; ++i) {
1218             memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
1219             p += aiocb->aio_iov[i].iov_len;
1220         }
1221         assert(p - buf == aiocb->aio_nbytes);
1222     }
1223 
1224     nbytes = handle_aiocb_rw_linear(aiocb, buf);
1225     if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
1226         char *p = buf;
1227         size_t count = aiocb->aio_nbytes, copy;
1228         int i;
1229 
1230         for (i = 0; i < aiocb->aio_niov && count; ++i) {
1231             copy = count;
1232             if (copy > aiocb->aio_iov[i].iov_len) {
1233                 copy = aiocb->aio_iov[i].iov_len;
1234             }
1235             memcpy(aiocb->aio_iov[i].iov_base, p, copy);
1236             assert(count >= copy);
1237             p     += copy;
1238             count -= copy;
1239         }
1240         assert(count == 0);
1241     }
1242     qemu_vfree(buf);
1243 
1244     return nbytes;
1245 }
1246 
1247 #ifdef CONFIG_XFS
1248 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
1249 {
1250     struct xfs_flock64 fl;
1251     int err;
1252 
1253     memset(&fl, 0, sizeof(fl));
1254     fl.l_whence = SEEK_SET;
1255     fl.l_start = offset;
1256     fl.l_len = bytes;
1257 
1258     if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
1259         err = errno;
1260         DPRINTF("cannot write zero range (%s)\n", strerror(errno));
1261         return -err;
1262     }
1263 
1264     return 0;
1265 }
1266 
1267 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
1268 {
1269     struct xfs_flock64 fl;
1270     int err;
1271 
1272     memset(&fl, 0, sizeof(fl));
1273     fl.l_whence = SEEK_SET;
1274     fl.l_start = offset;
1275     fl.l_len = bytes;
1276 
1277     if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
1278         err = errno;
1279         DPRINTF("cannot punch hole (%s)\n", strerror(errno));
1280         return -err;
1281     }
1282 
1283     return 0;
1284 }
1285 #endif
1286 
1287 static int translate_err(int err)
1288 {
1289     if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1290         err == -ENOTTY) {
1291         err = -ENOTSUP;
1292     }
1293     return err;
1294 }
1295 
1296 #ifdef CONFIG_FALLOCATE
1297 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1298 {
1299     do {
1300         if (fallocate(fd, mode, offset, len) == 0) {
1301             return 0;
1302         }
1303     } while (errno == EINTR);
1304     return translate_err(-errno);
1305 }
1306 #endif
1307 
1308 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1309 {
1310     int ret = -ENOTSUP;
1311     BDRVRawState *s = aiocb->bs->opaque;
1312 
1313     if (!s->has_write_zeroes) {
1314         return -ENOTSUP;
1315     }
1316 
1317 #ifdef BLKZEROOUT
1318     do {
1319         uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1320         if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1321             return 0;
1322         }
1323     } while (errno == EINTR);
1324 
1325     ret = translate_err(-errno);
1326 #endif
1327 
1328     if (ret == -ENOTSUP) {
1329         s->has_write_zeroes = false;
1330     }
1331     return ret;
1332 }
1333 
1334 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
1335 {
1336 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1337     BDRVRawState *s = aiocb->bs->opaque;
1338 #endif
1339 #ifdef CONFIG_FALLOCATE
1340     int64_t len;
1341 #endif
1342 
1343     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1344         return handle_aiocb_write_zeroes_block(aiocb);
1345     }
1346 
1347 #ifdef CONFIG_XFS
1348     if (s->is_xfs) {
1349         return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
1350     }
1351 #endif
1352 
1353 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1354     if (s->has_write_zeroes) {
1355         int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1356                                aiocb->aio_offset, aiocb->aio_nbytes);
1357         if (ret == 0 || ret != -ENOTSUP) {
1358             return ret;
1359         }
1360         s->has_write_zeroes = false;
1361     }
1362 #endif
1363 
1364 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1365     if (s->has_discard && s->has_fallocate) {
1366         int ret = do_fallocate(s->fd,
1367                                FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1368                                aiocb->aio_offset, aiocb->aio_nbytes);
1369         if (ret == 0) {
1370             ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1371             if (ret == 0 || ret != -ENOTSUP) {
1372                 return ret;
1373             }
1374             s->has_fallocate = false;
1375         } else if (ret != -ENOTSUP) {
1376             return ret;
1377         } else {
1378             s->has_discard = false;
1379         }
1380     }
1381 #endif
1382 
1383 #ifdef CONFIG_FALLOCATE
1384     /* Last resort: we are trying to extend the file with zeroed data. This
1385      * can be done via fallocate(fd, 0) */
1386     len = bdrv_getlength(aiocb->bs);
1387     if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) {
1388         int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1389         if (ret == 0 || ret != -ENOTSUP) {
1390             return ret;
1391         }
1392         s->has_fallocate = false;
1393     }
1394 #endif
1395 
1396     return -ENOTSUP;
1397 }
1398 
1399 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
1400 {
1401     int ret = -EOPNOTSUPP;
1402     BDRVRawState *s = aiocb->bs->opaque;
1403 
1404     if (!s->has_discard) {
1405         return -ENOTSUP;
1406     }
1407 
1408     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1409 #ifdef BLKDISCARD
1410         do {
1411             uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1412             if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1413                 return 0;
1414             }
1415         } while (errno == EINTR);
1416 
1417         ret = -errno;
1418 #endif
1419     } else {
1420 #ifdef CONFIG_XFS
1421         if (s->is_xfs) {
1422             return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1423         }
1424 #endif
1425 
1426 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1427         ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1428                            aiocb->aio_offset, aiocb->aio_nbytes);
1429 #endif
1430     }
1431 
1432     ret = translate_err(ret);
1433     if (ret == -ENOTSUP) {
1434         s->has_discard = false;
1435     }
1436     return ret;
1437 }
1438 
1439 static int aio_worker(void *arg)
1440 {
1441     RawPosixAIOData *aiocb = arg;
1442     ssize_t ret = 0;
1443 
1444     switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
1445     case QEMU_AIO_READ:
1446         ret = handle_aiocb_rw(aiocb);
1447         if (ret >= 0 && ret < aiocb->aio_nbytes) {
1448             iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
1449                       0, aiocb->aio_nbytes - ret);
1450 
1451             ret = aiocb->aio_nbytes;
1452         }
1453         if (ret == aiocb->aio_nbytes) {
1454             ret = 0;
1455         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1456             ret = -EINVAL;
1457         }
1458         break;
1459     case QEMU_AIO_WRITE:
1460         ret = handle_aiocb_rw(aiocb);
1461         if (ret == aiocb->aio_nbytes) {
1462             ret = 0;
1463         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1464             ret = -EINVAL;
1465         }
1466         break;
1467     case QEMU_AIO_FLUSH:
1468         ret = handle_aiocb_flush(aiocb);
1469         break;
1470     case QEMU_AIO_IOCTL:
1471         ret = handle_aiocb_ioctl(aiocb);
1472         break;
1473     case QEMU_AIO_DISCARD:
1474         ret = handle_aiocb_discard(aiocb);
1475         break;
1476     case QEMU_AIO_WRITE_ZEROES:
1477         ret = handle_aiocb_write_zeroes(aiocb);
1478         break;
1479     default:
1480         fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1481         ret = -EINVAL;
1482         break;
1483     }
1484 
1485     g_free(aiocb);
1486     return ret;
1487 }
1488 
1489 static int paio_submit_co(BlockDriverState *bs, int fd,
1490                           int64_t offset, QEMUIOVector *qiov,
1491                           int bytes, int type)
1492 {
1493     RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1494     ThreadPool *pool;
1495 
1496     acb->bs = bs;
1497     acb->aio_type = type;
1498     acb->aio_fildes = fd;
1499 
1500     acb->aio_nbytes = bytes;
1501     acb->aio_offset = offset;
1502 
1503     if (qiov) {
1504         acb->aio_iov = qiov->iov;
1505         acb->aio_niov = qiov->niov;
1506         assert(qiov->size == bytes);
1507     }
1508 
1509     trace_paio_submit_co(offset, bytes, type);
1510     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1511     return thread_pool_submit_co(pool, aio_worker, acb);
1512 }
1513 
1514 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
1515         int64_t offset, QEMUIOVector *qiov, int bytes,
1516         BlockCompletionFunc *cb, void *opaque, int type)
1517 {
1518     RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1519     ThreadPool *pool;
1520 
1521     acb->bs = bs;
1522     acb->aio_type = type;
1523     acb->aio_fildes = fd;
1524 
1525     acb->aio_nbytes = bytes;
1526     acb->aio_offset = offset;
1527 
1528     if (qiov) {
1529         acb->aio_iov = qiov->iov;
1530         acb->aio_niov = qiov->niov;
1531         assert(qiov->size == acb->aio_nbytes);
1532     }
1533 
1534     trace_paio_submit(acb, opaque, offset, bytes, type);
1535     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1536     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1537 }
1538 
1539 static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset,
1540                                    uint64_t bytes, QEMUIOVector *qiov, int type)
1541 {
1542     BDRVRawState *s = bs->opaque;
1543 
1544     if (fd_open(bs) < 0)
1545         return -EIO;
1546 
1547     /*
1548      * Check if the underlying device requires requests to be aligned,
1549      * and if the request we are trying to submit is aligned or not.
1550      * If this is the case tell the low-level driver that it needs
1551      * to copy the buffer.
1552      */
1553     if (s->needs_alignment) {
1554         if (!bdrv_qiov_is_aligned(bs, qiov)) {
1555             type |= QEMU_AIO_MISALIGNED;
1556 #ifdef CONFIG_LINUX_AIO
1557         } else if (s->use_linux_aio) {
1558             LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1559             assert(qiov->size == bytes);
1560             return laio_co_submit(bs, aio, s->fd, offset, qiov, type);
1561 #endif
1562         }
1563     }
1564 
1565     return paio_submit_co(bs, s->fd, offset, qiov, bytes, type);
1566 }
1567 
1568 static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset,
1569                                       uint64_t bytes, QEMUIOVector *qiov,
1570                                       int flags)
1571 {
1572     return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ);
1573 }
1574 
1575 static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
1576                                        uint64_t bytes, QEMUIOVector *qiov,
1577                                        int flags)
1578 {
1579     assert(flags == 0);
1580     return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE);
1581 }
1582 
1583 static void raw_aio_plug(BlockDriverState *bs)
1584 {
1585 #ifdef CONFIG_LINUX_AIO
1586     BDRVRawState *s = bs->opaque;
1587     if (s->use_linux_aio) {
1588         LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1589         laio_io_plug(bs, aio);
1590     }
1591 #endif
1592 }
1593 
1594 static void raw_aio_unplug(BlockDriverState *bs)
1595 {
1596 #ifdef CONFIG_LINUX_AIO
1597     BDRVRawState *s = bs->opaque;
1598     if (s->use_linux_aio) {
1599         LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1600         laio_io_unplug(bs, aio);
1601     }
1602 #endif
1603 }
1604 
1605 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
1606         BlockCompletionFunc *cb, void *opaque)
1607 {
1608     BDRVRawState *s = bs->opaque;
1609 
1610     if (fd_open(bs) < 0)
1611         return NULL;
1612 
1613     return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1614 }
1615 
1616 static void raw_close(BlockDriverState *bs)
1617 {
1618     BDRVRawState *s = bs->opaque;
1619 
1620     if (s->fd >= 0) {
1621         qemu_close(s->fd);
1622         s->fd = -1;
1623     }
1624     if (s->lock_fd >= 0) {
1625         qemu_close(s->lock_fd);
1626         s->lock_fd = -1;
1627     }
1628 }
1629 
1630 /**
1631  * Truncates the given regular file @fd to @offset and, when growing, fills the
1632  * new space according to @prealloc.
1633  *
1634  * Returns: 0 on success, -errno on failure.
1635  */
1636 static int raw_regular_truncate(int fd, int64_t offset, PreallocMode prealloc,
1637                                 Error **errp)
1638 {
1639     int result = 0;
1640     int64_t current_length = 0;
1641     char *buf = NULL;
1642     struct stat st;
1643 
1644     if (fstat(fd, &st) < 0) {
1645         result = -errno;
1646         error_setg_errno(errp, -result, "Could not stat file");
1647         return result;
1648     }
1649 
1650     current_length = st.st_size;
1651     if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1652         error_setg(errp, "Cannot use preallocation for shrinking files");
1653         return -ENOTSUP;
1654     }
1655 
1656     switch (prealloc) {
1657 #ifdef CONFIG_POSIX_FALLOCATE
1658     case PREALLOC_MODE_FALLOC:
1659         /*
1660          * Truncating before posix_fallocate() makes it about twice slower on
1661          * file systems that do not support fallocate(), trying to check if a
1662          * block is allocated before allocating it, so don't do that here.
1663          */
1664         result = -posix_fallocate(fd, current_length, offset - current_length);
1665         if (result != 0) {
1666             /* posix_fallocate() doesn't set errno. */
1667             error_setg_errno(errp, -result,
1668                              "Could not preallocate new data");
1669         }
1670         goto out;
1671 #endif
1672     case PREALLOC_MODE_FULL:
1673     {
1674         int64_t num = 0, left = offset - current_length;
1675 
1676         /*
1677          * Knowing the final size from the beginning could allow the file
1678          * system driver to do less allocations and possibly avoid
1679          * fragmentation of the file.
1680          */
1681         if (ftruncate(fd, offset) != 0) {
1682             result = -errno;
1683             error_setg_errno(errp, -result, "Could not resize file");
1684             goto out;
1685         }
1686 
1687         buf = g_malloc0(65536);
1688 
1689         result = lseek(fd, current_length, SEEK_SET);
1690         if (result < 0) {
1691             result = -errno;
1692             error_setg_errno(errp, -result,
1693                              "Failed to seek to the old end of file");
1694             goto out;
1695         }
1696 
1697         while (left > 0) {
1698             num = MIN(left, 65536);
1699             result = write(fd, buf, num);
1700             if (result < 0) {
1701                 result = -errno;
1702                 error_setg_errno(errp, -result,
1703                                  "Could not write zeros for preallocation");
1704                 goto out;
1705             }
1706             left -= result;
1707         }
1708         if (result >= 0) {
1709             result = fsync(fd);
1710             if (result < 0) {
1711                 result = -errno;
1712                 error_setg_errno(errp, -result,
1713                                  "Could not flush file to disk");
1714                 goto out;
1715             }
1716         }
1717         goto out;
1718     }
1719     case PREALLOC_MODE_OFF:
1720         if (ftruncate(fd, offset) != 0) {
1721             result = -errno;
1722             error_setg_errno(errp, -result, "Could not resize file");
1723         }
1724         return result;
1725     default:
1726         result = -ENOTSUP;
1727         error_setg(errp, "Unsupported preallocation mode: %s",
1728                    PreallocMode_lookup[prealloc]);
1729         return result;
1730     }
1731 
1732 out:
1733     if (result < 0) {
1734         if (ftruncate(fd, current_length) < 0) {
1735             error_report("Failed to restore old file length: %s",
1736                          strerror(errno));
1737         }
1738     }
1739 
1740     g_free(buf);
1741     return result;
1742 }
1743 
1744 static int raw_truncate(BlockDriverState *bs, int64_t offset,
1745                         PreallocMode prealloc, Error **errp)
1746 {
1747     BDRVRawState *s = bs->opaque;
1748     struct stat st;
1749     int ret;
1750 
1751     if (fstat(s->fd, &st)) {
1752         ret = -errno;
1753         error_setg_errno(errp, -ret, "Failed to fstat() the file");
1754         return ret;
1755     }
1756 
1757     if (S_ISREG(st.st_mode)) {
1758         return raw_regular_truncate(s->fd, offset, prealloc, errp);
1759     }
1760 
1761     if (prealloc != PREALLOC_MODE_OFF) {
1762         error_setg(errp, "Preallocation mode '%s' unsupported for this "
1763                    "non-regular file", PreallocMode_lookup[prealloc]);
1764         return -ENOTSUP;
1765     }
1766 
1767     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1768         if (offset > raw_getlength(bs)) {
1769             error_setg(errp, "Cannot grow device files");
1770             return -EINVAL;
1771         }
1772     } else {
1773         error_setg(errp, "Resizing this file is not supported");
1774         return -ENOTSUP;
1775     }
1776 
1777     return 0;
1778 }
1779 
1780 #ifdef __OpenBSD__
1781 static int64_t raw_getlength(BlockDriverState *bs)
1782 {
1783     BDRVRawState *s = bs->opaque;
1784     int fd = s->fd;
1785     struct stat st;
1786 
1787     if (fstat(fd, &st))
1788         return -errno;
1789     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1790         struct disklabel dl;
1791 
1792         if (ioctl(fd, DIOCGDINFO, &dl))
1793             return -errno;
1794         return (uint64_t)dl.d_secsize *
1795             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1796     } else
1797         return st.st_size;
1798 }
1799 #elif defined(__NetBSD__)
1800 static int64_t raw_getlength(BlockDriverState *bs)
1801 {
1802     BDRVRawState *s = bs->opaque;
1803     int fd = s->fd;
1804     struct stat st;
1805 
1806     if (fstat(fd, &st))
1807         return -errno;
1808     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1809         struct dkwedge_info dkw;
1810 
1811         if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1812             return dkw.dkw_size * 512;
1813         } else {
1814             struct disklabel dl;
1815 
1816             if (ioctl(fd, DIOCGDINFO, &dl))
1817                 return -errno;
1818             return (uint64_t)dl.d_secsize *
1819                 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1820         }
1821     } else
1822         return st.st_size;
1823 }
1824 #elif defined(__sun__)
1825 static int64_t raw_getlength(BlockDriverState *bs)
1826 {
1827     BDRVRawState *s = bs->opaque;
1828     struct dk_minfo minfo;
1829     int ret;
1830     int64_t size;
1831 
1832     ret = fd_open(bs);
1833     if (ret < 0) {
1834         return ret;
1835     }
1836 
1837     /*
1838      * Use the DKIOCGMEDIAINFO ioctl to read the size.
1839      */
1840     ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1841     if (ret != -1) {
1842         return minfo.dki_lbsize * minfo.dki_capacity;
1843     }
1844 
1845     /*
1846      * There are reports that lseek on some devices fails, but
1847      * irc discussion said that contingency on contingency was overkill.
1848      */
1849     size = lseek(s->fd, 0, SEEK_END);
1850     if (size < 0) {
1851         return -errno;
1852     }
1853     return size;
1854 }
1855 #elif defined(CONFIG_BSD)
1856 static int64_t raw_getlength(BlockDriverState *bs)
1857 {
1858     BDRVRawState *s = bs->opaque;
1859     int fd = s->fd;
1860     int64_t size;
1861     struct stat sb;
1862 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1863     int reopened = 0;
1864 #endif
1865     int ret;
1866 
1867     ret = fd_open(bs);
1868     if (ret < 0)
1869         return ret;
1870 
1871 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1872 again:
1873 #endif
1874     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1875 #ifdef DIOCGMEDIASIZE
1876 	if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1877 #elif defined(DIOCGPART)
1878         {
1879                 struct partinfo pi;
1880                 if (ioctl(fd, DIOCGPART, &pi) == 0)
1881                         size = pi.media_size;
1882                 else
1883                         size = 0;
1884         }
1885         if (size == 0)
1886 #endif
1887 #if defined(__APPLE__) && defined(__MACH__)
1888         {
1889             uint64_t sectors = 0;
1890             uint32_t sector_size = 0;
1891 
1892             if (ioctl(fd, DKIOCGETBLOCKCOUNT, &sectors) == 0
1893                && ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) == 0) {
1894                 size = sectors * sector_size;
1895             } else {
1896                 size = lseek(fd, 0LL, SEEK_END);
1897                 if (size < 0) {
1898                     return -errno;
1899                 }
1900             }
1901         }
1902 #else
1903         size = lseek(fd, 0LL, SEEK_END);
1904         if (size < 0) {
1905             return -errno;
1906         }
1907 #endif
1908 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1909         switch(s->type) {
1910         case FTYPE_CD:
1911             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1912             if (size == 2048LL * (unsigned)-1)
1913                 size = 0;
1914             /* XXX no disc?  maybe we need to reopen... */
1915             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1916                 reopened = 1;
1917                 goto again;
1918             }
1919         }
1920 #endif
1921     } else {
1922         size = lseek(fd, 0, SEEK_END);
1923         if (size < 0) {
1924             return -errno;
1925         }
1926     }
1927     return size;
1928 }
1929 #else
1930 static int64_t raw_getlength(BlockDriverState *bs)
1931 {
1932     BDRVRawState *s = bs->opaque;
1933     int ret;
1934     int64_t size;
1935 
1936     ret = fd_open(bs);
1937     if (ret < 0) {
1938         return ret;
1939     }
1940 
1941     size = lseek(s->fd, 0, SEEK_END);
1942     if (size < 0) {
1943         return -errno;
1944     }
1945     return size;
1946 }
1947 #endif
1948 
1949 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1950 {
1951     struct stat st;
1952     BDRVRawState *s = bs->opaque;
1953 
1954     if (fstat(s->fd, &st) < 0) {
1955         return -errno;
1956     }
1957     return (int64_t)st.st_blocks * 512;
1958 }
1959 
1960 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
1961 {
1962     int fd;
1963     int result = 0;
1964     int64_t total_size = 0;
1965     bool nocow = false;
1966     PreallocMode prealloc;
1967     char *buf = NULL;
1968     Error *local_err = NULL;
1969 
1970     strstart(filename, "file:", &filename);
1971 
1972     /* Read out options */
1973     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1974                           BDRV_SECTOR_SIZE);
1975     nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
1976     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1977     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
1978                                PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
1979                                &local_err);
1980     g_free(buf);
1981     if (local_err) {
1982         error_propagate(errp, local_err);
1983         result = -EINVAL;
1984         goto out;
1985     }
1986 
1987     fd = qemu_open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY,
1988                    0644);
1989     if (fd < 0) {
1990         result = -errno;
1991         error_setg_errno(errp, -result, "Could not create file");
1992         goto out;
1993     }
1994 
1995     if (nocow) {
1996 #ifdef __linux__
1997         /* Set NOCOW flag to solve performance issue on fs like btrfs.
1998          * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1999          * will be ignored since any failure of this operation should not
2000          * block the left work.
2001          */
2002         int attr;
2003         if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
2004             attr |= FS_NOCOW_FL;
2005             ioctl(fd, FS_IOC_SETFLAGS, &attr);
2006         }
2007 #endif
2008     }
2009 
2010     result = raw_regular_truncate(fd, total_size, prealloc, errp);
2011     if (result < 0) {
2012         goto out_close;
2013     }
2014 
2015 out_close:
2016     if (qemu_close(fd) != 0 && result == 0) {
2017         result = -errno;
2018         error_setg_errno(errp, -result, "Could not close the new file");
2019     }
2020 out:
2021     return result;
2022 }
2023 
2024 /*
2025  * Find allocation range in @bs around offset @start.
2026  * May change underlying file descriptor's file offset.
2027  * If @start is not in a hole, store @start in @data, and the
2028  * beginning of the next hole in @hole, and return 0.
2029  * If @start is in a non-trailing hole, store @start in @hole and the
2030  * beginning of the next non-hole in @data, and return 0.
2031  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2032  * If we can't find out, return a negative errno other than -ENXIO.
2033  */
2034 static int find_allocation(BlockDriverState *bs, off_t start,
2035                            off_t *data, off_t *hole)
2036 {
2037 #if defined SEEK_HOLE && defined SEEK_DATA
2038     BDRVRawState *s = bs->opaque;
2039     off_t offs;
2040 
2041     /*
2042      * SEEK_DATA cases:
2043      * D1. offs == start: start is in data
2044      * D2. offs > start: start is in a hole, next data at offs
2045      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2046      *                              or start is beyond EOF
2047      *     If the latter happens, the file has been truncated behind
2048      *     our back since we opened it.  All bets are off then.
2049      *     Treating like a trailing hole is simplest.
2050      * D4. offs < 0, errno != ENXIO: we learned nothing
2051      */
2052     offs = lseek(s->fd, start, SEEK_DATA);
2053     if (offs < 0) {
2054         return -errno;          /* D3 or D4 */
2055     }
2056     assert(offs >= start);
2057 
2058     if (offs > start) {
2059         /* D2: in hole, next data at offs */
2060         *hole = start;
2061         *data = offs;
2062         return 0;
2063     }
2064 
2065     /* D1: in data, end not yet known */
2066 
2067     /*
2068      * SEEK_HOLE cases:
2069      * H1. offs == start: start is in a hole
2070      *     If this happens here, a hole has been dug behind our back
2071      *     since the previous lseek().
2072      * H2. offs > start: either start is in data, next hole at offs,
2073      *                   or start is in trailing hole, EOF at offs
2074      *     Linux treats trailing holes like any other hole: offs ==
2075      *     start.  Solaris seeks to EOF instead: offs > start (blech).
2076      *     If that happens here, a hole has been dug behind our back
2077      *     since the previous lseek().
2078      * H3. offs < 0, errno = ENXIO: start is beyond EOF
2079      *     If this happens, the file has been truncated behind our
2080      *     back since we opened it.  Treat it like a trailing hole.
2081      * H4. offs < 0, errno != ENXIO: we learned nothing
2082      *     Pretend we know nothing at all, i.e. "forget" about D1.
2083      */
2084     offs = lseek(s->fd, start, SEEK_HOLE);
2085     if (offs < 0) {
2086         return -errno;          /* D1 and (H3 or H4) */
2087     }
2088     assert(offs >= start);
2089 
2090     if (offs > start) {
2091         /*
2092          * D1 and H2: either in data, next hole at offs, or it was in
2093          * data but is now in a trailing hole.  In the latter case,
2094          * all bets are off.  Treating it as if it there was data all
2095          * the way to EOF is safe, so simply do that.
2096          */
2097         *data = start;
2098         *hole = offs;
2099         return 0;
2100     }
2101 
2102     /* D1 and H1 */
2103     return -EBUSY;
2104 #else
2105     return -ENOTSUP;
2106 #endif
2107 }
2108 
2109 /*
2110  * Returns the allocation status of the specified sectors.
2111  *
2112  * If 'sector_num' is beyond the end of the disk image the return value is 0
2113  * and 'pnum' is set to 0.
2114  *
2115  * 'pnum' is set to the number of sectors (including and immediately following
2116  * the specified sector) that are known to be in the same
2117  * allocated/unallocated state.
2118  *
2119  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
2120  * beyond the end of the disk image it will be clamped.
2121  */
2122 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
2123                                                     int64_t sector_num,
2124                                                     int nb_sectors, int *pnum,
2125                                                     BlockDriverState **file)
2126 {
2127     off_t start, data = 0, hole = 0;
2128     int64_t total_size;
2129     int ret;
2130 
2131     ret = fd_open(bs);
2132     if (ret < 0) {
2133         return ret;
2134     }
2135 
2136     start = sector_num * BDRV_SECTOR_SIZE;
2137     total_size = bdrv_getlength(bs);
2138     if (total_size < 0) {
2139         return total_size;
2140     } else if (start >= total_size) {
2141         *pnum = 0;
2142         return 0;
2143     } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
2144         nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
2145     }
2146 
2147     ret = find_allocation(bs, start, &data, &hole);
2148     if (ret == -ENXIO) {
2149         /* Trailing hole */
2150         *pnum = nb_sectors;
2151         ret = BDRV_BLOCK_ZERO;
2152     } else if (ret < 0) {
2153         /* No info available, so pretend there are no holes */
2154         *pnum = nb_sectors;
2155         ret = BDRV_BLOCK_DATA;
2156     } else if (data == start) {
2157         /* On a data extent, compute sectors to the end of the extent,
2158          * possibly including a partial sector at EOF. */
2159         *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
2160         ret = BDRV_BLOCK_DATA;
2161     } else {
2162         /* On a hole, compute sectors to the beginning of the next extent.  */
2163         assert(hole == start);
2164         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
2165         ret = BDRV_BLOCK_ZERO;
2166     }
2167     *file = bs;
2168     return ret | BDRV_BLOCK_OFFSET_VALID | start;
2169 }
2170 
2171 static coroutine_fn BlockAIOCB *raw_aio_pdiscard(BlockDriverState *bs,
2172     int64_t offset, int bytes,
2173     BlockCompletionFunc *cb, void *opaque)
2174 {
2175     BDRVRawState *s = bs->opaque;
2176 
2177     return paio_submit(bs, s->fd, offset, NULL, bytes,
2178                        cb, opaque, QEMU_AIO_DISCARD);
2179 }
2180 
2181 static int coroutine_fn raw_co_pwrite_zeroes(
2182     BlockDriverState *bs, int64_t offset,
2183     int bytes, BdrvRequestFlags flags)
2184 {
2185     BDRVRawState *s = bs->opaque;
2186 
2187     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2188         return paio_submit_co(bs, s->fd, offset, NULL, bytes,
2189                               QEMU_AIO_WRITE_ZEROES);
2190     } else if (s->discard_zeroes) {
2191         return paio_submit_co(bs, s->fd, offset, NULL, bytes,
2192                               QEMU_AIO_DISCARD);
2193     }
2194     return -ENOTSUP;
2195 }
2196 
2197 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2198 {
2199     BDRVRawState *s = bs->opaque;
2200 
2201     bdi->unallocated_blocks_are_zero = s->discard_zeroes;
2202     bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
2203     return 0;
2204 }
2205 
2206 static QemuOptsList raw_create_opts = {
2207     .name = "raw-create-opts",
2208     .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
2209     .desc = {
2210         {
2211             .name = BLOCK_OPT_SIZE,
2212             .type = QEMU_OPT_SIZE,
2213             .help = "Virtual disk size"
2214         },
2215         {
2216             .name = BLOCK_OPT_NOCOW,
2217             .type = QEMU_OPT_BOOL,
2218             .help = "Turn off copy-on-write (valid only on btrfs)"
2219         },
2220         {
2221             .name = BLOCK_OPT_PREALLOC,
2222             .type = QEMU_OPT_STRING,
2223             .help = "Preallocation mode (allowed values: off, falloc, full)"
2224         },
2225         { /* end of list */ }
2226     }
2227 };
2228 
2229 static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared,
2230                           Error **errp)
2231 {
2232     return raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp);
2233 }
2234 
2235 static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared)
2236 {
2237     BDRVRawState *s = bs->opaque;
2238     raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL);
2239     s->perm = perm;
2240     s->shared_perm = shared;
2241 }
2242 
2243 static void raw_abort_perm_update(BlockDriverState *bs)
2244 {
2245     raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
2246 }
2247 
2248 BlockDriver bdrv_file = {
2249     .format_name = "file",
2250     .protocol_name = "file",
2251     .instance_size = sizeof(BDRVRawState),
2252     .bdrv_needs_filename = true,
2253     .bdrv_probe = NULL, /* no probe for protocols */
2254     .bdrv_parse_filename = raw_parse_filename,
2255     .bdrv_file_open = raw_open,
2256     .bdrv_reopen_prepare = raw_reopen_prepare,
2257     .bdrv_reopen_commit = raw_reopen_commit,
2258     .bdrv_reopen_abort = raw_reopen_abort,
2259     .bdrv_close = raw_close,
2260     .bdrv_create = raw_create,
2261     .bdrv_has_zero_init = bdrv_has_zero_init_1,
2262     .bdrv_co_get_block_status = raw_co_get_block_status,
2263     .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes,
2264 
2265     .bdrv_co_preadv         = raw_co_preadv,
2266     .bdrv_co_pwritev        = raw_co_pwritev,
2267     .bdrv_aio_flush = raw_aio_flush,
2268     .bdrv_aio_pdiscard = raw_aio_pdiscard,
2269     .bdrv_refresh_limits = raw_refresh_limits,
2270     .bdrv_io_plug = raw_aio_plug,
2271     .bdrv_io_unplug = raw_aio_unplug,
2272 
2273     .bdrv_truncate = raw_truncate,
2274     .bdrv_getlength = raw_getlength,
2275     .bdrv_get_info = raw_get_info,
2276     .bdrv_get_allocated_file_size
2277                         = raw_get_allocated_file_size,
2278     .bdrv_check_perm = raw_check_perm,
2279     .bdrv_set_perm   = raw_set_perm,
2280     .bdrv_abort_perm_update = raw_abort_perm_update,
2281     .create_opts = &raw_create_opts,
2282 };
2283 
2284 /***********************************************/
2285 /* host device */
2286 
2287 #if defined(__APPLE__) && defined(__MACH__)
2288 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
2289                                 CFIndex maxPathSize, int flags);
2290 static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
2291 {
2292     kern_return_t kernResult = KERN_FAILURE;
2293     mach_port_t     masterPort;
2294     CFMutableDictionaryRef  classesToMatch;
2295     const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
2296     char *mediaType = NULL;
2297 
2298     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
2299     if ( KERN_SUCCESS != kernResult ) {
2300         printf( "IOMasterPort returned %d\n", kernResult );
2301     }
2302 
2303     int index;
2304     for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
2305         classesToMatch = IOServiceMatching(matching_array[index]);
2306         if (classesToMatch == NULL) {
2307             error_report("IOServiceMatching returned NULL for %s",
2308                          matching_array[index]);
2309             continue;
2310         }
2311         CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
2312                              kCFBooleanTrue);
2313         kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch,
2314                                                   mediaIterator);
2315         if (kernResult != KERN_SUCCESS) {
2316             error_report("Note: IOServiceGetMatchingServices returned %d",
2317                          kernResult);
2318             continue;
2319         }
2320 
2321         /* If a match was found, leave the loop */
2322         if (*mediaIterator != 0) {
2323             DPRINTF("Matching using %s\n", matching_array[index]);
2324             mediaType = g_strdup(matching_array[index]);
2325             break;
2326         }
2327     }
2328     return mediaType;
2329 }
2330 
2331 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
2332                          CFIndex maxPathSize, int flags)
2333 {
2334     io_object_t     nextMedia;
2335     kern_return_t   kernResult = KERN_FAILURE;
2336     *bsdPath = '\0';
2337     nextMedia = IOIteratorNext( mediaIterator );
2338     if ( nextMedia )
2339     {
2340         CFTypeRef   bsdPathAsCFString;
2341     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
2342         if ( bsdPathAsCFString ) {
2343             size_t devPathLength;
2344             strcpy( bsdPath, _PATH_DEV );
2345             if (flags & BDRV_O_NOCACHE) {
2346                 strcat(bsdPath, "r");
2347             }
2348             devPathLength = strlen( bsdPath );
2349             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
2350                 kernResult = KERN_SUCCESS;
2351             }
2352             CFRelease( bsdPathAsCFString );
2353         }
2354         IOObjectRelease( nextMedia );
2355     }
2356 
2357     return kernResult;
2358 }
2359 
2360 /* Sets up a real cdrom for use in QEMU */
2361 static bool setup_cdrom(char *bsd_path, Error **errp)
2362 {
2363     int index, num_of_test_partitions = 2, fd;
2364     char test_partition[MAXPATHLEN];
2365     bool partition_found = false;
2366 
2367     /* look for a working partition */
2368     for (index = 0; index < num_of_test_partitions; index++) {
2369         snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
2370                  index);
2371         fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE);
2372         if (fd >= 0) {
2373             partition_found = true;
2374             qemu_close(fd);
2375             break;
2376         }
2377     }
2378 
2379     /* if a working partition on the device was not found */
2380     if (partition_found == false) {
2381         error_setg(errp, "Failed to find a working partition on disc");
2382     } else {
2383         DPRINTF("Using %s as optical disc\n", test_partition);
2384         pstrcpy(bsd_path, MAXPATHLEN, test_partition);
2385     }
2386     return partition_found;
2387 }
2388 
2389 /* Prints directions on mounting and unmounting a device */
2390 static void print_unmounting_directions(const char *file_name)
2391 {
2392     error_report("If device %s is mounted on the desktop, unmount"
2393                  " it first before using it in QEMU", file_name);
2394     error_report("Command to unmount device: diskutil unmountDisk %s",
2395                  file_name);
2396     error_report("Command to mount device: diskutil mountDisk %s", file_name);
2397 }
2398 
2399 #endif /* defined(__APPLE__) && defined(__MACH__) */
2400 
2401 static int hdev_probe_device(const char *filename)
2402 {
2403     struct stat st;
2404 
2405     /* allow a dedicated CD-ROM driver to match with a higher priority */
2406     if (strstart(filename, "/dev/cdrom", NULL))
2407         return 50;
2408 
2409     if (stat(filename, &st) >= 0 &&
2410             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
2411         return 100;
2412     }
2413 
2414     return 0;
2415 }
2416 
2417 static int check_hdev_writable(BDRVRawState *s)
2418 {
2419 #if defined(BLKROGET)
2420     /* Linux block devices can be configured "read-only" using blockdev(8).
2421      * This is independent of device node permissions and therefore open(2)
2422      * with O_RDWR succeeds.  Actual writes fail with EPERM.
2423      *
2424      * bdrv_open() is supposed to fail if the disk is read-only.  Explicitly
2425      * check for read-only block devices so that Linux block devices behave
2426      * properly.
2427      */
2428     struct stat st;
2429     int readonly = 0;
2430 
2431     if (fstat(s->fd, &st)) {
2432         return -errno;
2433     }
2434 
2435     if (!S_ISBLK(st.st_mode)) {
2436         return 0;
2437     }
2438 
2439     if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
2440         return -errno;
2441     }
2442 
2443     if (readonly) {
2444         return -EACCES;
2445     }
2446 #endif /* defined(BLKROGET) */
2447     return 0;
2448 }
2449 
2450 static void hdev_parse_filename(const char *filename, QDict *options,
2451                                 Error **errp)
2452 {
2453     bdrv_parse_filename_strip_prefix(filename, "host_device:", options);
2454 }
2455 
2456 static bool hdev_is_sg(BlockDriverState *bs)
2457 {
2458 
2459 #if defined(__linux__)
2460 
2461     BDRVRawState *s = bs->opaque;
2462     struct stat st;
2463     struct sg_scsi_id scsiid;
2464     int sg_version;
2465     int ret;
2466 
2467     if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) {
2468         return false;
2469     }
2470 
2471     ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version);
2472     if (ret < 0) {
2473         return false;
2474     }
2475 
2476     ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid);
2477     if (ret >= 0) {
2478         DPRINTF("SG device found: type=%d, version=%d\n",
2479             scsiid.scsi_type, sg_version);
2480         return true;
2481     }
2482 
2483 #endif
2484 
2485     return false;
2486 }
2487 
2488 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
2489                      Error **errp)
2490 {
2491     BDRVRawState *s = bs->opaque;
2492     Error *local_err = NULL;
2493     int ret;
2494 
2495 #if defined(__APPLE__) && defined(__MACH__)
2496     /*
2497      * Caution: while qdict_get_str() is fine, getting non-string types
2498      * would require more care.  When @options come from -blockdev or
2499      * blockdev_add, its members are typed according to the QAPI
2500      * schema, but when they come from -drive, they're all QString.
2501      */
2502     const char *filename = qdict_get_str(options, "filename");
2503     char bsd_path[MAXPATHLEN] = "";
2504     bool error_occurred = false;
2505 
2506     /* If using a real cdrom */
2507     if (strcmp(filename, "/dev/cdrom") == 0) {
2508         char *mediaType = NULL;
2509         kern_return_t ret_val;
2510         io_iterator_t mediaIterator = 0;
2511 
2512         mediaType = FindEjectableOpticalMedia(&mediaIterator);
2513         if (mediaType == NULL) {
2514             error_setg(errp, "Please make sure your CD/DVD is in the optical"
2515                        " drive");
2516             error_occurred = true;
2517             goto hdev_open_Mac_error;
2518         }
2519 
2520         ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
2521         if (ret_val != KERN_SUCCESS) {
2522             error_setg(errp, "Could not get BSD path for optical drive");
2523             error_occurred = true;
2524             goto hdev_open_Mac_error;
2525         }
2526 
2527         /* If a real optical drive was not found */
2528         if (bsd_path[0] == '\0') {
2529             error_setg(errp, "Failed to obtain bsd path for optical drive");
2530             error_occurred = true;
2531             goto hdev_open_Mac_error;
2532         }
2533 
2534         /* If using a cdrom disc and finding a partition on the disc failed */
2535         if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
2536             setup_cdrom(bsd_path, errp) == false) {
2537             print_unmounting_directions(bsd_path);
2538             error_occurred = true;
2539             goto hdev_open_Mac_error;
2540         }
2541 
2542         qdict_put_str(options, "filename", bsd_path);
2543 
2544 hdev_open_Mac_error:
2545         g_free(mediaType);
2546         if (mediaIterator) {
2547             IOObjectRelease(mediaIterator);
2548         }
2549         if (error_occurred) {
2550             return -ENOENT;
2551         }
2552     }
2553 #endif /* defined(__APPLE__) && defined(__MACH__) */
2554 
2555     s->type = FTYPE_FILE;
2556 
2557     ret = raw_open_common(bs, options, flags, 0, &local_err);
2558     if (ret < 0) {
2559         error_propagate(errp, local_err);
2560 #if defined(__APPLE__) && defined(__MACH__)
2561         if (*bsd_path) {
2562             filename = bsd_path;
2563         }
2564         /* if a physical device experienced an error while being opened */
2565         if (strncmp(filename, "/dev/", 5) == 0) {
2566             print_unmounting_directions(filename);
2567         }
2568 #endif /* defined(__APPLE__) && defined(__MACH__) */
2569         return ret;
2570     }
2571 
2572     /* Since this does ioctl the device must be already opened */
2573     bs->sg = hdev_is_sg(bs);
2574 
2575     if (flags & BDRV_O_RDWR) {
2576         ret = check_hdev_writable(s);
2577         if (ret < 0) {
2578             raw_close(bs);
2579             error_setg_errno(errp, -ret, "The device is not writable");
2580             return ret;
2581         }
2582     }
2583 
2584     return ret;
2585 }
2586 
2587 #if defined(__linux__)
2588 
2589 static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
2590         unsigned long int req, void *buf,
2591         BlockCompletionFunc *cb, void *opaque)
2592 {
2593     BDRVRawState *s = bs->opaque;
2594     RawPosixAIOData *acb;
2595     ThreadPool *pool;
2596 
2597     if (fd_open(bs) < 0)
2598         return NULL;
2599 
2600     acb = g_new(RawPosixAIOData, 1);
2601     acb->bs = bs;
2602     acb->aio_type = QEMU_AIO_IOCTL;
2603     acb->aio_fildes = s->fd;
2604     acb->aio_offset = 0;
2605     acb->aio_ioctl_buf = buf;
2606     acb->aio_ioctl_cmd = req;
2607     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
2608     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
2609 }
2610 #endif /* linux */
2611 
2612 static int fd_open(BlockDriverState *bs)
2613 {
2614     BDRVRawState *s = bs->opaque;
2615 
2616     /* this is just to ensure s->fd is sane (its called by io ops) */
2617     if (s->fd >= 0)
2618         return 0;
2619     return -EIO;
2620 }
2621 
2622 static coroutine_fn BlockAIOCB *hdev_aio_pdiscard(BlockDriverState *bs,
2623     int64_t offset, int bytes,
2624     BlockCompletionFunc *cb, void *opaque)
2625 {
2626     BDRVRawState *s = bs->opaque;
2627 
2628     if (fd_open(bs) < 0) {
2629         return NULL;
2630     }
2631     return paio_submit(bs, s->fd, offset, NULL, bytes,
2632                        cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2633 }
2634 
2635 static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs,
2636     int64_t offset, int bytes, BdrvRequestFlags flags)
2637 {
2638     BDRVRawState *s = bs->opaque;
2639     int rc;
2640 
2641     rc = fd_open(bs);
2642     if (rc < 0) {
2643         return rc;
2644     }
2645     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2646         return paio_submit_co(bs, s->fd, offset, NULL, bytes,
2647                               QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2648     } else if (s->discard_zeroes) {
2649         return paio_submit_co(bs, s->fd, offset, NULL, bytes,
2650                               QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2651     }
2652     return -ENOTSUP;
2653 }
2654 
2655 static int hdev_create(const char *filename, QemuOpts *opts,
2656                        Error **errp)
2657 {
2658     int fd;
2659     int ret = 0;
2660     struct stat stat_buf;
2661     int64_t total_size = 0;
2662     bool has_prefix;
2663 
2664     /* This function is used by both protocol block drivers and therefore either
2665      * of these prefixes may be given.
2666      * The return value has to be stored somewhere, otherwise this is an error
2667      * due to -Werror=unused-value. */
2668     has_prefix =
2669         strstart(filename, "host_device:", &filename) ||
2670         strstart(filename, "host_cdrom:" , &filename);
2671 
2672     (void)has_prefix;
2673 
2674     ret = raw_normalize_devicepath(&filename);
2675     if (ret < 0) {
2676         error_setg_errno(errp, -ret, "Could not normalize device path");
2677         return ret;
2678     }
2679 
2680     /* Read out options */
2681     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2682                           BDRV_SECTOR_SIZE);
2683 
2684     fd = qemu_open(filename, O_WRONLY | O_BINARY);
2685     if (fd < 0) {
2686         ret = -errno;
2687         error_setg_errno(errp, -ret, "Could not open device");
2688         return ret;
2689     }
2690 
2691     if (fstat(fd, &stat_buf) < 0) {
2692         ret = -errno;
2693         error_setg_errno(errp, -ret, "Could not stat device");
2694     } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2695         error_setg(errp,
2696                    "The given file is neither a block nor a character device");
2697         ret = -ENODEV;
2698     } else if (lseek(fd, 0, SEEK_END) < total_size) {
2699         error_setg(errp, "Device is too small");
2700         ret = -ENOSPC;
2701     }
2702 
2703     qemu_close(fd);
2704     return ret;
2705 }
2706 
2707 static BlockDriver bdrv_host_device = {
2708     .format_name        = "host_device",
2709     .protocol_name        = "host_device",
2710     .instance_size      = sizeof(BDRVRawState),
2711     .bdrv_needs_filename = true,
2712     .bdrv_probe_device  = hdev_probe_device,
2713     .bdrv_parse_filename = hdev_parse_filename,
2714     .bdrv_file_open     = hdev_open,
2715     .bdrv_close         = raw_close,
2716     .bdrv_reopen_prepare = raw_reopen_prepare,
2717     .bdrv_reopen_commit  = raw_reopen_commit,
2718     .bdrv_reopen_abort   = raw_reopen_abort,
2719     .bdrv_create         = hdev_create,
2720     .create_opts         = &raw_create_opts,
2721     .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes,
2722 
2723     .bdrv_co_preadv         = raw_co_preadv,
2724     .bdrv_co_pwritev        = raw_co_pwritev,
2725     .bdrv_aio_flush	= raw_aio_flush,
2726     .bdrv_aio_pdiscard   = hdev_aio_pdiscard,
2727     .bdrv_refresh_limits = raw_refresh_limits,
2728     .bdrv_io_plug = raw_aio_plug,
2729     .bdrv_io_unplug = raw_aio_unplug,
2730 
2731     .bdrv_truncate      = raw_truncate,
2732     .bdrv_getlength	= raw_getlength,
2733     .bdrv_get_info = raw_get_info,
2734     .bdrv_get_allocated_file_size
2735                         = raw_get_allocated_file_size,
2736     .bdrv_check_perm = raw_check_perm,
2737     .bdrv_set_perm   = raw_set_perm,
2738     .bdrv_abort_perm_update = raw_abort_perm_update,
2739     .bdrv_probe_blocksizes = hdev_probe_blocksizes,
2740     .bdrv_probe_geometry = hdev_probe_geometry,
2741 
2742     /* generic scsi device */
2743 #ifdef __linux__
2744     .bdrv_aio_ioctl     = hdev_aio_ioctl,
2745 #endif
2746 };
2747 
2748 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2749 static void cdrom_parse_filename(const char *filename, QDict *options,
2750                                  Error **errp)
2751 {
2752     bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options);
2753 }
2754 #endif
2755 
2756 #ifdef __linux__
2757 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2758                       Error **errp)
2759 {
2760     BDRVRawState *s = bs->opaque;
2761 
2762     s->type = FTYPE_CD;
2763 
2764     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2765     return raw_open_common(bs, options, flags, O_NONBLOCK, errp);
2766 }
2767 
2768 static int cdrom_probe_device(const char *filename)
2769 {
2770     int fd, ret;
2771     int prio = 0;
2772     struct stat st;
2773 
2774     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2775     if (fd < 0) {
2776         goto out;
2777     }
2778     ret = fstat(fd, &st);
2779     if (ret == -1 || !S_ISBLK(st.st_mode)) {
2780         goto outc;
2781     }
2782 
2783     /* Attempt to detect via a CDROM specific ioctl */
2784     ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2785     if (ret >= 0)
2786         prio = 100;
2787 
2788 outc:
2789     qemu_close(fd);
2790 out:
2791     return prio;
2792 }
2793 
2794 static bool cdrom_is_inserted(BlockDriverState *bs)
2795 {
2796     BDRVRawState *s = bs->opaque;
2797     int ret;
2798 
2799     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2800     return ret == CDS_DISC_OK;
2801 }
2802 
2803 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2804 {
2805     BDRVRawState *s = bs->opaque;
2806 
2807     if (eject_flag) {
2808         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2809             perror("CDROMEJECT");
2810     } else {
2811         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2812             perror("CDROMEJECT");
2813     }
2814 }
2815 
2816 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2817 {
2818     BDRVRawState *s = bs->opaque;
2819 
2820     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2821         /*
2822          * Note: an error can happen if the distribution automatically
2823          * mounts the CD-ROM
2824          */
2825         /* perror("CDROM_LOCKDOOR"); */
2826     }
2827 }
2828 
2829 static BlockDriver bdrv_host_cdrom = {
2830     .format_name        = "host_cdrom",
2831     .protocol_name      = "host_cdrom",
2832     .instance_size      = sizeof(BDRVRawState),
2833     .bdrv_needs_filename = true,
2834     .bdrv_probe_device	= cdrom_probe_device,
2835     .bdrv_parse_filename = cdrom_parse_filename,
2836     .bdrv_file_open     = cdrom_open,
2837     .bdrv_close         = raw_close,
2838     .bdrv_reopen_prepare = raw_reopen_prepare,
2839     .bdrv_reopen_commit  = raw_reopen_commit,
2840     .bdrv_reopen_abort   = raw_reopen_abort,
2841     .bdrv_create         = hdev_create,
2842     .create_opts         = &raw_create_opts,
2843 
2844 
2845     .bdrv_co_preadv         = raw_co_preadv,
2846     .bdrv_co_pwritev        = raw_co_pwritev,
2847     .bdrv_aio_flush	= raw_aio_flush,
2848     .bdrv_refresh_limits = raw_refresh_limits,
2849     .bdrv_io_plug = raw_aio_plug,
2850     .bdrv_io_unplug = raw_aio_unplug,
2851 
2852     .bdrv_truncate      = raw_truncate,
2853     .bdrv_getlength      = raw_getlength,
2854     .has_variable_length = true,
2855     .bdrv_get_allocated_file_size
2856                         = raw_get_allocated_file_size,
2857 
2858     /* removable device support */
2859     .bdrv_is_inserted   = cdrom_is_inserted,
2860     .bdrv_eject         = cdrom_eject,
2861     .bdrv_lock_medium   = cdrom_lock_medium,
2862 
2863     /* generic scsi device */
2864     .bdrv_aio_ioctl     = hdev_aio_ioctl,
2865 };
2866 #endif /* __linux__ */
2867 
2868 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2869 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2870                       Error **errp)
2871 {
2872     BDRVRawState *s = bs->opaque;
2873     Error *local_err = NULL;
2874     int ret;
2875 
2876     s->type = FTYPE_CD;
2877 
2878     ret = raw_open_common(bs, options, flags, 0, &local_err);
2879     if (ret) {
2880         error_propagate(errp, local_err);
2881         return ret;
2882     }
2883 
2884     /* make sure the door isn't locked at this time */
2885     ioctl(s->fd, CDIOCALLOW);
2886     return 0;
2887 }
2888 
2889 static int cdrom_probe_device(const char *filename)
2890 {
2891     if (strstart(filename, "/dev/cd", NULL) ||
2892             strstart(filename, "/dev/acd", NULL))
2893         return 100;
2894     return 0;
2895 }
2896 
2897 static int cdrom_reopen(BlockDriverState *bs)
2898 {
2899     BDRVRawState *s = bs->opaque;
2900     int fd;
2901 
2902     /*
2903      * Force reread of possibly changed/newly loaded disc,
2904      * FreeBSD seems to not notice sometimes...
2905      */
2906     if (s->fd >= 0)
2907         qemu_close(s->fd);
2908     fd = qemu_open(bs->filename, s->open_flags, 0644);
2909     if (fd < 0) {
2910         s->fd = -1;
2911         return -EIO;
2912     }
2913     s->fd = fd;
2914 
2915     /* make sure the door isn't locked at this time */
2916     ioctl(s->fd, CDIOCALLOW);
2917     return 0;
2918 }
2919 
2920 static bool cdrom_is_inserted(BlockDriverState *bs)
2921 {
2922     return raw_getlength(bs) > 0;
2923 }
2924 
2925 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2926 {
2927     BDRVRawState *s = bs->opaque;
2928 
2929     if (s->fd < 0)
2930         return;
2931 
2932     (void) ioctl(s->fd, CDIOCALLOW);
2933 
2934     if (eject_flag) {
2935         if (ioctl(s->fd, CDIOCEJECT) < 0)
2936             perror("CDIOCEJECT");
2937     } else {
2938         if (ioctl(s->fd, CDIOCCLOSE) < 0)
2939             perror("CDIOCCLOSE");
2940     }
2941 
2942     cdrom_reopen(bs);
2943 }
2944 
2945 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2946 {
2947     BDRVRawState *s = bs->opaque;
2948 
2949     if (s->fd < 0)
2950         return;
2951     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2952         /*
2953          * Note: an error can happen if the distribution automatically
2954          * mounts the CD-ROM
2955          */
2956         /* perror("CDROM_LOCKDOOR"); */
2957     }
2958 }
2959 
2960 static BlockDriver bdrv_host_cdrom = {
2961     .format_name        = "host_cdrom",
2962     .protocol_name      = "host_cdrom",
2963     .instance_size      = sizeof(BDRVRawState),
2964     .bdrv_needs_filename = true,
2965     .bdrv_probe_device	= cdrom_probe_device,
2966     .bdrv_parse_filename = cdrom_parse_filename,
2967     .bdrv_file_open     = cdrom_open,
2968     .bdrv_close         = raw_close,
2969     .bdrv_reopen_prepare = raw_reopen_prepare,
2970     .bdrv_reopen_commit  = raw_reopen_commit,
2971     .bdrv_reopen_abort   = raw_reopen_abort,
2972     .bdrv_create        = hdev_create,
2973     .create_opts        = &raw_create_opts,
2974 
2975     .bdrv_co_preadv         = raw_co_preadv,
2976     .bdrv_co_pwritev        = raw_co_pwritev,
2977     .bdrv_aio_flush	= raw_aio_flush,
2978     .bdrv_refresh_limits = raw_refresh_limits,
2979     .bdrv_io_plug = raw_aio_plug,
2980     .bdrv_io_unplug = raw_aio_unplug,
2981 
2982     .bdrv_truncate      = raw_truncate,
2983     .bdrv_getlength      = raw_getlength,
2984     .has_variable_length = true,
2985     .bdrv_get_allocated_file_size
2986                         = raw_get_allocated_file_size,
2987 
2988     /* removable device support */
2989     .bdrv_is_inserted   = cdrom_is_inserted,
2990     .bdrv_eject         = cdrom_eject,
2991     .bdrv_lock_medium   = cdrom_lock_medium,
2992 };
2993 #endif /* __FreeBSD__ */
2994 
2995 static void bdrv_file_init(void)
2996 {
2997     /*
2998      * Register all the drivers.  Note that order is important, the driver
2999      * registered last will get probed first.
3000      */
3001     bdrv_register(&bdrv_file);
3002     bdrv_register(&bdrv_host_device);
3003 #ifdef __linux__
3004     bdrv_register(&bdrv_host_cdrom);
3005 #endif
3006 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3007     bdrv_register(&bdrv_host_cdrom);
3008 #endif
3009 }
3010 
3011 block_init(bdrv_file_init);
3012