xref: /openbmc/qemu/block/file-posix.c (revision 4a09d0bb)
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 "qemu/timer.h"
29 #include "qemu/log.h"
30 #include "block/block_int.h"
31 #include "qemu/module.h"
32 #include "trace.h"
33 #include "block/thread-pool.h"
34 #include "qemu/iov.h"
35 #include "block/raw-aio.h"
36 #include "qapi/util.h"
37 #include "qapi/qmp/qstring.h"
38 
39 #if defined(__APPLE__) && (__MACH__)
40 #include <paths.h>
41 #include <sys/param.h>
42 #include <IOKit/IOKitLib.h>
43 #include <IOKit/IOBSD.h>
44 #include <IOKit/storage/IOMediaBSDClient.h>
45 #include <IOKit/storage/IOMedia.h>
46 #include <IOKit/storage/IOCDMedia.h>
47 //#include <IOKit/storage/IOCDTypes.h>
48 #include <IOKit/storage/IODVDMedia.h>
49 #include <CoreFoundation/CoreFoundation.h>
50 #endif
51 
52 #ifdef __sun__
53 #define _POSIX_PTHREAD_SEMANTICS 1
54 #include <sys/dkio.h>
55 #endif
56 #ifdef __linux__
57 #include <sys/ioctl.h>
58 #include <sys/param.h>
59 #include <linux/cdrom.h>
60 #include <linux/fd.h>
61 #include <linux/fs.h>
62 #include <linux/hdreg.h>
63 #include <scsi/sg.h>
64 #ifdef __s390__
65 #include <asm/dasd.h>
66 #endif
67 #ifndef FS_NOCOW_FL
68 #define FS_NOCOW_FL                     0x00800000 /* Do not cow file */
69 #endif
70 #endif
71 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
72 #include <linux/falloc.h>
73 #endif
74 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
75 #include <sys/disk.h>
76 #include <sys/cdio.h>
77 #endif
78 
79 #ifdef __OpenBSD__
80 #include <sys/ioctl.h>
81 #include <sys/disklabel.h>
82 #include <sys/dkio.h>
83 #endif
84 
85 #ifdef __NetBSD__
86 #include <sys/ioctl.h>
87 #include <sys/disklabel.h>
88 #include <sys/dkio.h>
89 #include <sys/disk.h>
90 #endif
91 
92 #ifdef __DragonFly__
93 #include <sys/ioctl.h>
94 #include <sys/diskslice.h>
95 #endif
96 
97 #ifdef CONFIG_XFS
98 #include <xfs/xfs.h>
99 #endif
100 
101 //#define DEBUG_BLOCK
102 
103 #ifdef DEBUG_BLOCK
104 # define DEBUG_BLOCK_PRINT 1
105 #else
106 # define DEBUG_BLOCK_PRINT 0
107 #endif
108 #define DPRINTF(fmt, ...) \
109 do { \
110     if (DEBUG_BLOCK_PRINT) { \
111         printf(fmt, ## __VA_ARGS__); \
112     } \
113 } while (0)
114 
115 /* OS X does not have O_DSYNC */
116 #ifndef O_DSYNC
117 #ifdef O_SYNC
118 #define O_DSYNC O_SYNC
119 #elif defined(O_FSYNC)
120 #define O_DSYNC O_FSYNC
121 #endif
122 #endif
123 
124 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
125 #ifndef O_DIRECT
126 #define O_DIRECT O_DSYNC
127 #endif
128 
129 #define FTYPE_FILE   0
130 #define FTYPE_CD     1
131 
132 #define MAX_BLOCKSIZE	4096
133 
134 typedef struct BDRVRawState {
135     int fd;
136     int type;
137     int open_flags;
138     size_t buf_align;
139 
140 #ifdef CONFIG_XFS
141     bool is_xfs:1;
142 #endif
143     bool has_discard:1;
144     bool has_write_zeroes:1;
145     bool discard_zeroes:1;
146     bool use_linux_aio:1;
147     bool has_fallocate;
148     bool needs_alignment;
149 } BDRVRawState;
150 
151 typedef struct BDRVRawReopenState {
152     int fd;
153     int open_flags;
154 } BDRVRawReopenState;
155 
156 static int fd_open(BlockDriverState *bs);
157 static int64_t raw_getlength(BlockDriverState *bs);
158 
159 typedef struct RawPosixAIOData {
160     BlockDriverState *bs;
161     int aio_fildes;
162     union {
163         struct iovec *aio_iov;
164         void *aio_ioctl_buf;
165     };
166     int aio_niov;
167     uint64_t aio_nbytes;
168 #define aio_ioctl_cmd   aio_nbytes /* for QEMU_AIO_IOCTL */
169     off_t aio_offset;
170     int aio_type;
171 } RawPosixAIOData;
172 
173 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
174 static int cdrom_reopen(BlockDriverState *bs);
175 #endif
176 
177 #if defined(__NetBSD__)
178 static int raw_normalize_devicepath(const char **filename)
179 {
180     static char namebuf[PATH_MAX];
181     const char *dp, *fname;
182     struct stat sb;
183 
184     fname = *filename;
185     dp = strrchr(fname, '/');
186     if (lstat(fname, &sb) < 0) {
187         fprintf(stderr, "%s: stat failed: %s\n",
188             fname, strerror(errno));
189         return -errno;
190     }
191 
192     if (!S_ISBLK(sb.st_mode)) {
193         return 0;
194     }
195 
196     if (dp == NULL) {
197         snprintf(namebuf, PATH_MAX, "r%s", fname);
198     } else {
199         snprintf(namebuf, PATH_MAX, "%.*s/r%s",
200             (int)(dp - fname), fname, dp + 1);
201     }
202     fprintf(stderr, "%s is a block device", fname);
203     *filename = namebuf;
204     fprintf(stderr, ", using %s\n", *filename);
205 
206     return 0;
207 }
208 #else
209 static int raw_normalize_devicepath(const char **filename)
210 {
211     return 0;
212 }
213 #endif
214 
215 /*
216  * Get logical block size via ioctl. On success store it in @sector_size_p.
217  */
218 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
219 {
220     unsigned int sector_size;
221     bool success = false;
222 
223     errno = ENOTSUP;
224 
225     /* Try a few ioctls to get the right size */
226 #ifdef BLKSSZGET
227     if (ioctl(fd, BLKSSZGET, &sector_size) >= 0) {
228         *sector_size_p = sector_size;
229         success = true;
230     }
231 #endif
232 #ifdef DKIOCGETBLOCKSIZE
233     if (ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) >= 0) {
234         *sector_size_p = sector_size;
235         success = true;
236     }
237 #endif
238 #ifdef DIOCGSECTORSIZE
239     if (ioctl(fd, DIOCGSECTORSIZE, &sector_size) >= 0) {
240         *sector_size_p = sector_size;
241         success = true;
242     }
243 #endif
244 
245     return success ? 0 : -errno;
246 }
247 
248 /**
249  * Get physical block size of @fd.
250  * On success, store it in @blk_size and return 0.
251  * On failure, return -errno.
252  */
253 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
254 {
255 #ifdef BLKPBSZGET
256     if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
257         return -errno;
258     }
259     return 0;
260 #else
261     return -ENOTSUP;
262 #endif
263 }
264 
265 /* Check if read is allowed with given memory buffer and length.
266  *
267  * This function is used to check O_DIRECT memory buffer and request alignment.
268  */
269 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
270 {
271     ssize_t ret = pread(fd, buf, len, 0);
272 
273     if (ret >= 0) {
274         return true;
275     }
276 
277 #ifdef __linux__
278     /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads.  Ignore
279      * other errors (e.g. real I/O error), which could happen on a failed
280      * drive, since we only care about probing alignment.
281      */
282     if (errno != EINVAL) {
283         return true;
284     }
285 #endif
286 
287     return false;
288 }
289 
290 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
291 {
292     BDRVRawState *s = bs->opaque;
293     char *buf;
294     size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize());
295 
296     /* For SCSI generic devices the alignment is not really used.
297        With buffered I/O, we don't have any restrictions. */
298     if (bdrv_is_sg(bs) || !s->needs_alignment) {
299         bs->bl.request_alignment = 1;
300         s->buf_align = 1;
301         return;
302     }
303 
304     bs->bl.request_alignment = 0;
305     s->buf_align = 0;
306     /* Let's try to use the logical blocksize for the alignment. */
307     if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) {
308         bs->bl.request_alignment = 0;
309     }
310 #ifdef CONFIG_XFS
311     if (s->is_xfs) {
312         struct dioattr da;
313         if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
314             bs->bl.request_alignment = da.d_miniosz;
315             /* The kernel returns wrong information for d_mem */
316             /* s->buf_align = da.d_mem; */
317         }
318     }
319 #endif
320 
321     /* If we could not get the sizes so far, we can only guess them */
322     if (!s->buf_align) {
323         size_t align;
324         buf = qemu_memalign(max_align, 2 * max_align);
325         for (align = 512; align <= max_align; align <<= 1) {
326             if (raw_is_io_aligned(fd, buf + align, max_align)) {
327                 s->buf_align = align;
328                 break;
329             }
330         }
331         qemu_vfree(buf);
332     }
333 
334     if (!bs->bl.request_alignment) {
335         size_t align;
336         buf = qemu_memalign(s->buf_align, max_align);
337         for (align = 512; align <= max_align; align <<= 1) {
338             if (raw_is_io_aligned(fd, buf, align)) {
339                 bs->bl.request_alignment = align;
340                 break;
341             }
342         }
343         qemu_vfree(buf);
344     }
345 
346     if (!s->buf_align || !bs->bl.request_alignment) {
347         error_setg(errp, "Could not find working O_DIRECT alignment");
348         error_append_hint(errp, "Try cache.direct=off\n");
349     }
350 }
351 
352 static void raw_parse_flags(int bdrv_flags, int *open_flags)
353 {
354     assert(open_flags != NULL);
355 
356     *open_flags |= O_BINARY;
357     *open_flags &= ~O_ACCMODE;
358     if (bdrv_flags & BDRV_O_RDWR) {
359         *open_flags |= O_RDWR;
360     } else {
361         *open_flags |= O_RDONLY;
362     }
363 
364     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
365      * and O_DIRECT for no caching. */
366     if ((bdrv_flags & BDRV_O_NOCACHE)) {
367         *open_flags |= O_DIRECT;
368     }
369 }
370 
371 static void raw_parse_filename(const char *filename, QDict *options,
372                                Error **errp)
373 {
374     /* The filename does not have to be prefixed by the protocol name, since
375      * "file" is the default protocol; therefore, the return value of this
376      * function call can be ignored. */
377     strstart(filename, "file:", &filename);
378 
379     qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
380 }
381 
382 static QemuOptsList raw_runtime_opts = {
383     .name = "raw",
384     .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
385     .desc = {
386         {
387             .name = "filename",
388             .type = QEMU_OPT_STRING,
389             .help = "File name of the image",
390         },
391         {
392             .name = "aio",
393             .type = QEMU_OPT_STRING,
394             .help = "host AIO implementation (threads, native)",
395         },
396         { /* end of list */ }
397     },
398 };
399 
400 static int raw_open_common(BlockDriverState *bs, QDict *options,
401                            int bdrv_flags, int open_flags, Error **errp)
402 {
403     BDRVRawState *s = bs->opaque;
404     QemuOpts *opts;
405     Error *local_err = NULL;
406     const char *filename = NULL;
407     BlockdevAioOptions aio, aio_default;
408     int fd, ret;
409     struct stat st;
410 
411     opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
412     qemu_opts_absorb_qdict(opts, options, &local_err);
413     if (local_err) {
414         error_propagate(errp, local_err);
415         ret = -EINVAL;
416         goto fail;
417     }
418 
419     filename = qemu_opt_get(opts, "filename");
420 
421     ret = raw_normalize_devicepath(&filename);
422     if (ret != 0) {
423         error_setg_errno(errp, -ret, "Could not normalize device path");
424         goto fail;
425     }
426 
427     aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO)
428                   ? BLOCKDEV_AIO_OPTIONS_NATIVE
429                   : BLOCKDEV_AIO_OPTIONS_THREADS;
430     aio = qapi_enum_parse(BlockdevAioOptions_lookup, qemu_opt_get(opts, "aio"),
431                           BLOCKDEV_AIO_OPTIONS__MAX, aio_default, &local_err);
432     if (local_err) {
433         error_propagate(errp, local_err);
434         ret = -EINVAL;
435         goto fail;
436     }
437     s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE);
438 
439     s->open_flags = open_flags;
440     raw_parse_flags(bdrv_flags, &s->open_flags);
441 
442     s->fd = -1;
443     fd = qemu_open(filename, s->open_flags, 0644);
444     if (fd < 0) {
445         ret = -errno;
446         error_setg_errno(errp, errno, "Could not open '%s'", filename);
447         if (ret == -EROFS) {
448             ret = -EACCES;
449         }
450         goto fail;
451     }
452     s->fd = fd;
453 
454 #ifdef CONFIG_LINUX_AIO
455      /* Currently Linux does AIO only for files opened with O_DIRECT */
456     if (s->use_linux_aio && !(s->open_flags & O_DIRECT)) {
457         error_setg(errp, "aio=native was specified, but it requires "
458                          "cache.direct=on, which was not specified.");
459         ret = -EINVAL;
460         goto fail;
461     }
462 #else
463     if (s->use_linux_aio) {
464         error_setg(errp, "aio=native was specified, but is not supported "
465                          "in this build.");
466         ret = -EINVAL;
467         goto fail;
468     }
469 #endif /* !defined(CONFIG_LINUX_AIO) */
470 
471     s->has_discard = true;
472     s->has_write_zeroes = true;
473     bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
474     if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
475         s->needs_alignment = true;
476     }
477 
478     if (fstat(s->fd, &st) < 0) {
479         ret = -errno;
480         error_setg_errno(errp, errno, "Could not stat file");
481         goto fail;
482     }
483     if (S_ISREG(st.st_mode)) {
484         s->discard_zeroes = true;
485         s->has_fallocate = true;
486     }
487     if (S_ISBLK(st.st_mode)) {
488 #ifdef BLKDISCARDZEROES
489         unsigned int arg;
490         if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
491             s->discard_zeroes = true;
492         }
493 #endif
494 #ifdef __linux__
495         /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache.  Do
496          * not rely on the contents of discarded blocks unless using O_DIRECT.
497          * Same for BLKZEROOUT.
498          */
499         if (!(bs->open_flags & BDRV_O_NOCACHE)) {
500             s->discard_zeroes = false;
501             s->has_write_zeroes = false;
502         }
503 #endif
504     }
505 #ifdef __FreeBSD__
506     if (S_ISCHR(st.st_mode)) {
507         /*
508          * The file is a char device (disk), which on FreeBSD isn't behind
509          * a pager, so force all requests to be aligned. This is needed
510          * so QEMU makes sure all IO operations on the device are aligned
511          * to sector size, or else FreeBSD will reject them with EINVAL.
512          */
513         s->needs_alignment = true;
514     }
515 #endif
516 
517 #ifdef CONFIG_XFS
518     if (platform_test_xfs_fd(s->fd)) {
519         s->is_xfs = true;
520     }
521 #endif
522 
523     ret = 0;
524 fail:
525     if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
526         unlink(filename);
527     }
528     qemu_opts_del(opts);
529     return ret;
530 }
531 
532 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
533                     Error **errp)
534 {
535     BDRVRawState *s = bs->opaque;
536 
537     s->type = FTYPE_FILE;
538     return raw_open_common(bs, options, flags, 0, errp);
539 }
540 
541 static int raw_reopen_prepare(BDRVReopenState *state,
542                               BlockReopenQueue *queue, Error **errp)
543 {
544     BDRVRawState *s;
545     BDRVRawReopenState *rs;
546     int ret = 0;
547     Error *local_err = NULL;
548 
549     assert(state != NULL);
550     assert(state->bs != NULL);
551 
552     s = state->bs->opaque;
553 
554     state->opaque = g_new0(BDRVRawReopenState, 1);
555     rs = state->opaque;
556 
557     if (s->type == FTYPE_CD) {
558         rs->open_flags |= O_NONBLOCK;
559     }
560 
561     raw_parse_flags(state->flags, &rs->open_flags);
562 
563     rs->fd = -1;
564 
565     int fcntl_flags = O_APPEND | O_NONBLOCK;
566 #ifdef O_NOATIME
567     fcntl_flags |= O_NOATIME;
568 #endif
569 
570 #ifdef O_ASYNC
571     /* Not all operating systems have O_ASYNC, and those that don't
572      * will not let us track the state into rs->open_flags (typically
573      * you achieve the same effect with an ioctl, for example I_SETSIG
574      * on Solaris). But we do not use O_ASYNC, so that's fine.
575      */
576     assert((s->open_flags & O_ASYNC) == 0);
577 #endif
578 
579     if ((rs->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
580         /* dup the original fd */
581         rs->fd = qemu_dup(s->fd);
582         if (rs->fd >= 0) {
583             ret = fcntl_setfl(rs->fd, rs->open_flags);
584             if (ret) {
585                 qemu_close(rs->fd);
586                 rs->fd = -1;
587             }
588         }
589     }
590 
591     /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
592     if (rs->fd == -1) {
593         const char *normalized_filename = state->bs->filename;
594         ret = raw_normalize_devicepath(&normalized_filename);
595         if (ret < 0) {
596             error_setg_errno(errp, -ret, "Could not normalize device path");
597         } else {
598             assert(!(rs->open_flags & O_CREAT));
599             rs->fd = qemu_open(normalized_filename, rs->open_flags);
600             if (rs->fd == -1) {
601                 error_setg_errno(errp, errno, "Could not reopen file");
602                 ret = -1;
603             }
604         }
605     }
606 
607     /* Fail already reopen_prepare() if we can't get a working O_DIRECT
608      * alignment with the new fd. */
609     if (rs->fd != -1) {
610         raw_probe_alignment(state->bs, rs->fd, &local_err);
611         if (local_err) {
612             qemu_close(rs->fd);
613             rs->fd = -1;
614             error_propagate(errp, local_err);
615             ret = -EINVAL;
616         }
617     }
618 
619     return ret;
620 }
621 
622 static void raw_reopen_commit(BDRVReopenState *state)
623 {
624     BDRVRawReopenState *rs = state->opaque;
625     BDRVRawState *s = state->bs->opaque;
626 
627     s->open_flags = rs->open_flags;
628 
629     qemu_close(s->fd);
630     s->fd = rs->fd;
631 
632     g_free(state->opaque);
633     state->opaque = NULL;
634 }
635 
636 
637 static void raw_reopen_abort(BDRVReopenState *state)
638 {
639     BDRVRawReopenState *rs = state->opaque;
640 
641      /* nothing to do if NULL, we didn't get far enough */
642     if (rs == NULL) {
643         return;
644     }
645 
646     if (rs->fd >= 0) {
647         qemu_close(rs->fd);
648         rs->fd = -1;
649     }
650     g_free(state->opaque);
651     state->opaque = NULL;
652 }
653 
654 static int hdev_get_max_transfer_length(BlockDriverState *bs, int fd)
655 {
656 #ifdef BLKSECTGET
657     int max_bytes = 0;
658     short max_sectors = 0;
659     if (bs->sg && ioctl(fd, BLKSECTGET, &max_bytes) == 0) {
660         return max_bytes;
661     } else if (!bs->sg && ioctl(fd, BLKSECTGET, &max_sectors) == 0) {
662         return max_sectors << BDRV_SECTOR_BITS;
663     } else {
664         return -errno;
665     }
666 #else
667     return -ENOSYS;
668 #endif
669 }
670 
671 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
672 {
673     BDRVRawState *s = bs->opaque;
674     struct stat st;
675 
676     if (!fstat(s->fd, &st)) {
677         if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode)) {
678             int ret = hdev_get_max_transfer_length(bs, s->fd);
679             if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) {
680                 bs->bl.max_transfer = pow2floor(ret);
681             }
682         }
683     }
684 
685     raw_probe_alignment(bs, s->fd, errp);
686     bs->bl.min_mem_alignment = s->buf_align;
687     bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
688 }
689 
690 static int check_for_dasd(int fd)
691 {
692 #ifdef BIODASDINFO2
693     struct dasd_information2_t info = {0};
694 
695     return ioctl(fd, BIODASDINFO2, &info);
696 #else
697     return -1;
698 #endif
699 }
700 
701 /**
702  * Try to get @bs's logical and physical block size.
703  * On success, store them in @bsz and return zero.
704  * On failure, return negative errno.
705  */
706 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
707 {
708     BDRVRawState *s = bs->opaque;
709     int ret;
710 
711     /* If DASD, get blocksizes */
712     if (check_for_dasd(s->fd) < 0) {
713         return -ENOTSUP;
714     }
715     ret = probe_logical_blocksize(s->fd, &bsz->log);
716     if (ret < 0) {
717         return ret;
718     }
719     return probe_physical_blocksize(s->fd, &bsz->phys);
720 }
721 
722 /**
723  * Try to get @bs's geometry: cyls, heads, sectors.
724  * On success, store them in @geo and return 0.
725  * On failure return -errno.
726  * (Allows block driver to assign default geometry values that guest sees)
727  */
728 #ifdef __linux__
729 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
730 {
731     BDRVRawState *s = bs->opaque;
732     struct hd_geometry ioctl_geo = {0};
733 
734     /* If DASD, get its geometry */
735     if (check_for_dasd(s->fd) < 0) {
736         return -ENOTSUP;
737     }
738     if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
739         return -errno;
740     }
741     /* HDIO_GETGEO may return success even though geo contains zeros
742        (e.g. certain multipath setups) */
743     if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
744         return -ENOTSUP;
745     }
746     /* Do not return a geometry for partition */
747     if (ioctl_geo.start != 0) {
748         return -ENOTSUP;
749     }
750     geo->heads = ioctl_geo.heads;
751     geo->sectors = ioctl_geo.sectors;
752     geo->cylinders = ioctl_geo.cylinders;
753 
754     return 0;
755 }
756 #else /* __linux__ */
757 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
758 {
759     return -ENOTSUP;
760 }
761 #endif
762 
763 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
764 {
765     int ret;
766 
767     ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
768     if (ret == -1) {
769         return -errno;
770     }
771 
772     return 0;
773 }
774 
775 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
776 {
777     int ret;
778 
779     ret = qemu_fdatasync(aiocb->aio_fildes);
780     if (ret == -1) {
781         return -errno;
782     }
783     return 0;
784 }
785 
786 #ifdef CONFIG_PREADV
787 
788 static bool preadv_present = true;
789 
790 static ssize_t
791 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
792 {
793     return preadv(fd, iov, nr_iov, offset);
794 }
795 
796 static ssize_t
797 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
798 {
799     return pwritev(fd, iov, nr_iov, offset);
800 }
801 
802 #else
803 
804 static bool preadv_present = false;
805 
806 static ssize_t
807 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
808 {
809     return -ENOSYS;
810 }
811 
812 static ssize_t
813 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
814 {
815     return -ENOSYS;
816 }
817 
818 #endif
819 
820 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
821 {
822     ssize_t len;
823 
824     do {
825         if (aiocb->aio_type & QEMU_AIO_WRITE)
826             len = qemu_pwritev(aiocb->aio_fildes,
827                                aiocb->aio_iov,
828                                aiocb->aio_niov,
829                                aiocb->aio_offset);
830          else
831             len = qemu_preadv(aiocb->aio_fildes,
832                               aiocb->aio_iov,
833                               aiocb->aio_niov,
834                               aiocb->aio_offset);
835     } while (len == -1 && errno == EINTR);
836 
837     if (len == -1) {
838         return -errno;
839     }
840     return len;
841 }
842 
843 /*
844  * Read/writes the data to/from a given linear buffer.
845  *
846  * Returns the number of bytes handles or -errno in case of an error. Short
847  * reads are only returned if the end of the file is reached.
848  */
849 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
850 {
851     ssize_t offset = 0;
852     ssize_t len;
853 
854     while (offset < aiocb->aio_nbytes) {
855         if (aiocb->aio_type & QEMU_AIO_WRITE) {
856             len = pwrite(aiocb->aio_fildes,
857                          (const char *)buf + offset,
858                          aiocb->aio_nbytes - offset,
859                          aiocb->aio_offset + offset);
860         } else {
861             len = pread(aiocb->aio_fildes,
862                         buf + offset,
863                         aiocb->aio_nbytes - offset,
864                         aiocb->aio_offset + offset);
865         }
866         if (len == -1 && errno == EINTR) {
867             continue;
868         } else if (len == -1 && errno == EINVAL &&
869                    (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
870                    !(aiocb->aio_type & QEMU_AIO_WRITE) &&
871                    offset > 0) {
872             /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
873              * after a short read.  Assume that O_DIRECT short reads only occur
874              * at EOF.  Therefore this is a short read, not an I/O error.
875              */
876             break;
877         } else if (len == -1) {
878             offset = -errno;
879             break;
880         } else if (len == 0) {
881             break;
882         }
883         offset += len;
884     }
885 
886     return offset;
887 }
888 
889 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
890 {
891     ssize_t nbytes;
892     char *buf;
893 
894     if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
895         /*
896          * If there is just a single buffer, and it is properly aligned
897          * we can just use plain pread/pwrite without any problems.
898          */
899         if (aiocb->aio_niov == 1) {
900              return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
901         }
902         /*
903          * We have more than one iovec, and all are properly aligned.
904          *
905          * Try preadv/pwritev first and fall back to linearizing the
906          * buffer if it's not supported.
907          */
908         if (preadv_present) {
909             nbytes = handle_aiocb_rw_vector(aiocb);
910             if (nbytes == aiocb->aio_nbytes ||
911                 (nbytes < 0 && nbytes != -ENOSYS)) {
912                 return nbytes;
913             }
914             preadv_present = false;
915         }
916 
917         /*
918          * XXX(hch): short read/write.  no easy way to handle the reminder
919          * using these interfaces.  For now retry using plain
920          * pread/pwrite?
921          */
922     }
923 
924     /*
925      * Ok, we have to do it the hard way, copy all segments into
926      * a single aligned buffer.
927      */
928     buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
929     if (buf == NULL) {
930         return -ENOMEM;
931     }
932 
933     if (aiocb->aio_type & QEMU_AIO_WRITE) {
934         char *p = buf;
935         int i;
936 
937         for (i = 0; i < aiocb->aio_niov; ++i) {
938             memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
939             p += aiocb->aio_iov[i].iov_len;
940         }
941         assert(p - buf == aiocb->aio_nbytes);
942     }
943 
944     nbytes = handle_aiocb_rw_linear(aiocb, buf);
945     if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
946         char *p = buf;
947         size_t count = aiocb->aio_nbytes, copy;
948         int i;
949 
950         for (i = 0; i < aiocb->aio_niov && count; ++i) {
951             copy = count;
952             if (copy > aiocb->aio_iov[i].iov_len) {
953                 copy = aiocb->aio_iov[i].iov_len;
954             }
955             memcpy(aiocb->aio_iov[i].iov_base, p, copy);
956             assert(count >= copy);
957             p     += copy;
958             count -= copy;
959         }
960         assert(count == 0);
961     }
962     qemu_vfree(buf);
963 
964     return nbytes;
965 }
966 
967 #ifdef CONFIG_XFS
968 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
969 {
970     struct xfs_flock64 fl;
971     int err;
972 
973     memset(&fl, 0, sizeof(fl));
974     fl.l_whence = SEEK_SET;
975     fl.l_start = offset;
976     fl.l_len = bytes;
977 
978     if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
979         err = errno;
980         DPRINTF("cannot write zero range (%s)\n", strerror(errno));
981         return -err;
982     }
983 
984     return 0;
985 }
986 
987 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
988 {
989     struct xfs_flock64 fl;
990     int err;
991 
992     memset(&fl, 0, sizeof(fl));
993     fl.l_whence = SEEK_SET;
994     fl.l_start = offset;
995     fl.l_len = bytes;
996 
997     if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
998         err = errno;
999         DPRINTF("cannot punch hole (%s)\n", strerror(errno));
1000         return -err;
1001     }
1002 
1003     return 0;
1004 }
1005 #endif
1006 
1007 static int translate_err(int err)
1008 {
1009     if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1010         err == -ENOTTY) {
1011         err = -ENOTSUP;
1012     }
1013     return err;
1014 }
1015 
1016 #ifdef CONFIG_FALLOCATE
1017 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1018 {
1019     do {
1020         if (fallocate(fd, mode, offset, len) == 0) {
1021             return 0;
1022         }
1023     } while (errno == EINTR);
1024     return translate_err(-errno);
1025 }
1026 #endif
1027 
1028 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1029 {
1030     int ret = -ENOTSUP;
1031     BDRVRawState *s = aiocb->bs->opaque;
1032 
1033     if (!s->has_write_zeroes) {
1034         return -ENOTSUP;
1035     }
1036 
1037 #ifdef BLKZEROOUT
1038     do {
1039         uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1040         if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1041             return 0;
1042         }
1043     } while (errno == EINTR);
1044 
1045     ret = translate_err(-errno);
1046 #endif
1047 
1048     if (ret == -ENOTSUP) {
1049         s->has_write_zeroes = false;
1050     }
1051     return ret;
1052 }
1053 
1054 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
1055 {
1056 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1057     BDRVRawState *s = aiocb->bs->opaque;
1058 #endif
1059 
1060     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1061         return handle_aiocb_write_zeroes_block(aiocb);
1062     }
1063 
1064 #ifdef CONFIG_XFS
1065     if (s->is_xfs) {
1066         return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
1067     }
1068 #endif
1069 
1070 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1071     if (s->has_write_zeroes) {
1072         int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1073                                aiocb->aio_offset, aiocb->aio_nbytes);
1074         if (ret == 0 || ret != -ENOTSUP) {
1075             return ret;
1076         }
1077         s->has_write_zeroes = false;
1078     }
1079 #endif
1080 
1081 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1082     if (s->has_discard && s->has_fallocate) {
1083         int ret = do_fallocate(s->fd,
1084                                FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1085                                aiocb->aio_offset, aiocb->aio_nbytes);
1086         if (ret == 0) {
1087             ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1088             if (ret == 0 || ret != -ENOTSUP) {
1089                 return ret;
1090             }
1091             s->has_fallocate = false;
1092         } else if (ret != -ENOTSUP) {
1093             return ret;
1094         } else {
1095             s->has_discard = false;
1096         }
1097     }
1098 #endif
1099 
1100 #ifdef CONFIG_FALLOCATE
1101     if (s->has_fallocate && aiocb->aio_offset >= bdrv_getlength(aiocb->bs)) {
1102         int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1103         if (ret == 0 || ret != -ENOTSUP) {
1104             return ret;
1105         }
1106         s->has_fallocate = false;
1107     }
1108 #endif
1109 
1110     return -ENOTSUP;
1111 }
1112 
1113 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
1114 {
1115     int ret = -EOPNOTSUPP;
1116     BDRVRawState *s = aiocb->bs->opaque;
1117 
1118     if (!s->has_discard) {
1119         return -ENOTSUP;
1120     }
1121 
1122     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1123 #ifdef BLKDISCARD
1124         do {
1125             uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1126             if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1127                 return 0;
1128             }
1129         } while (errno == EINTR);
1130 
1131         ret = -errno;
1132 #endif
1133     } else {
1134 #ifdef CONFIG_XFS
1135         if (s->is_xfs) {
1136             return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1137         }
1138 #endif
1139 
1140 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1141         ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1142                            aiocb->aio_offset, aiocb->aio_nbytes);
1143 #endif
1144     }
1145 
1146     ret = translate_err(ret);
1147     if (ret == -ENOTSUP) {
1148         s->has_discard = false;
1149     }
1150     return ret;
1151 }
1152 
1153 static int aio_worker(void *arg)
1154 {
1155     RawPosixAIOData *aiocb = arg;
1156     ssize_t ret = 0;
1157 
1158     switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
1159     case QEMU_AIO_READ:
1160         ret = handle_aiocb_rw(aiocb);
1161         if (ret >= 0 && ret < aiocb->aio_nbytes) {
1162             iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
1163                       0, aiocb->aio_nbytes - ret);
1164 
1165             ret = aiocb->aio_nbytes;
1166         }
1167         if (ret == aiocb->aio_nbytes) {
1168             ret = 0;
1169         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1170             ret = -EINVAL;
1171         }
1172         break;
1173     case QEMU_AIO_WRITE:
1174         ret = handle_aiocb_rw(aiocb);
1175         if (ret == aiocb->aio_nbytes) {
1176             ret = 0;
1177         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1178             ret = -EINVAL;
1179         }
1180         break;
1181     case QEMU_AIO_FLUSH:
1182         ret = handle_aiocb_flush(aiocb);
1183         break;
1184     case QEMU_AIO_IOCTL:
1185         ret = handle_aiocb_ioctl(aiocb);
1186         break;
1187     case QEMU_AIO_DISCARD:
1188         ret = handle_aiocb_discard(aiocb);
1189         break;
1190     case QEMU_AIO_WRITE_ZEROES:
1191         ret = handle_aiocb_write_zeroes(aiocb);
1192         break;
1193     default:
1194         fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1195         ret = -EINVAL;
1196         break;
1197     }
1198 
1199     g_free(aiocb);
1200     return ret;
1201 }
1202 
1203 static int paio_submit_co(BlockDriverState *bs, int fd,
1204                           int64_t offset, QEMUIOVector *qiov,
1205                           int count, int type)
1206 {
1207     RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1208     ThreadPool *pool;
1209 
1210     acb->bs = bs;
1211     acb->aio_type = type;
1212     acb->aio_fildes = fd;
1213 
1214     acb->aio_nbytes = count;
1215     acb->aio_offset = offset;
1216 
1217     if (qiov) {
1218         acb->aio_iov = qiov->iov;
1219         acb->aio_niov = qiov->niov;
1220         assert(qiov->size == count);
1221     }
1222 
1223     trace_paio_submit_co(offset, count, type);
1224     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1225     return thread_pool_submit_co(pool, aio_worker, acb);
1226 }
1227 
1228 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
1229         int64_t offset, QEMUIOVector *qiov, int count,
1230         BlockCompletionFunc *cb, void *opaque, int type)
1231 {
1232     RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1233     ThreadPool *pool;
1234 
1235     acb->bs = bs;
1236     acb->aio_type = type;
1237     acb->aio_fildes = fd;
1238 
1239     acb->aio_nbytes = count;
1240     acb->aio_offset = offset;
1241 
1242     if (qiov) {
1243         acb->aio_iov = qiov->iov;
1244         acb->aio_niov = qiov->niov;
1245         assert(qiov->size == acb->aio_nbytes);
1246     }
1247 
1248     trace_paio_submit(acb, opaque, offset, count, type);
1249     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1250     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1251 }
1252 
1253 static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset,
1254                                    uint64_t bytes, QEMUIOVector *qiov, int type)
1255 {
1256     BDRVRawState *s = bs->opaque;
1257 
1258     if (fd_open(bs) < 0)
1259         return -EIO;
1260 
1261     /*
1262      * Check if the underlying device requires requests to be aligned,
1263      * and if the request we are trying to submit is aligned or not.
1264      * If this is the case tell the low-level driver that it needs
1265      * to copy the buffer.
1266      */
1267     if (s->needs_alignment) {
1268         if (!bdrv_qiov_is_aligned(bs, qiov)) {
1269             type |= QEMU_AIO_MISALIGNED;
1270 #ifdef CONFIG_LINUX_AIO
1271         } else if (s->use_linux_aio) {
1272             LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1273             assert(qiov->size == bytes);
1274             return laio_co_submit(bs, aio, s->fd, offset, qiov, type);
1275 #endif
1276         }
1277     }
1278 
1279     return paio_submit_co(bs, s->fd, offset, qiov, bytes, type);
1280 }
1281 
1282 static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset,
1283                                       uint64_t bytes, QEMUIOVector *qiov,
1284                                       int flags)
1285 {
1286     return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ);
1287 }
1288 
1289 static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
1290                                        uint64_t bytes, QEMUIOVector *qiov,
1291                                        int flags)
1292 {
1293     assert(flags == 0);
1294     return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE);
1295 }
1296 
1297 static void raw_aio_plug(BlockDriverState *bs)
1298 {
1299 #ifdef CONFIG_LINUX_AIO
1300     BDRVRawState *s = bs->opaque;
1301     if (s->use_linux_aio) {
1302         LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1303         laio_io_plug(bs, aio);
1304     }
1305 #endif
1306 }
1307 
1308 static void raw_aio_unplug(BlockDriverState *bs)
1309 {
1310 #ifdef CONFIG_LINUX_AIO
1311     BDRVRawState *s = bs->opaque;
1312     if (s->use_linux_aio) {
1313         LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1314         laio_io_unplug(bs, aio);
1315     }
1316 #endif
1317 }
1318 
1319 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
1320         BlockCompletionFunc *cb, void *opaque)
1321 {
1322     BDRVRawState *s = bs->opaque;
1323 
1324     if (fd_open(bs) < 0)
1325         return NULL;
1326 
1327     return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1328 }
1329 
1330 static void raw_close(BlockDriverState *bs)
1331 {
1332     BDRVRawState *s = bs->opaque;
1333 
1334     if (s->fd >= 0) {
1335         qemu_close(s->fd);
1336         s->fd = -1;
1337     }
1338 }
1339 
1340 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1341 {
1342     BDRVRawState *s = bs->opaque;
1343     struct stat st;
1344 
1345     if (fstat(s->fd, &st)) {
1346         return -errno;
1347     }
1348 
1349     if (S_ISREG(st.st_mode)) {
1350         if (ftruncate(s->fd, offset) < 0) {
1351             return -errno;
1352         }
1353     } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1354        if (offset > raw_getlength(bs)) {
1355            return -EINVAL;
1356        }
1357     } else {
1358         return -ENOTSUP;
1359     }
1360 
1361     return 0;
1362 }
1363 
1364 #ifdef __OpenBSD__
1365 static int64_t raw_getlength(BlockDriverState *bs)
1366 {
1367     BDRVRawState *s = bs->opaque;
1368     int fd = s->fd;
1369     struct stat st;
1370 
1371     if (fstat(fd, &st))
1372         return -errno;
1373     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1374         struct disklabel dl;
1375 
1376         if (ioctl(fd, DIOCGDINFO, &dl))
1377             return -errno;
1378         return (uint64_t)dl.d_secsize *
1379             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1380     } else
1381         return st.st_size;
1382 }
1383 #elif defined(__NetBSD__)
1384 static int64_t raw_getlength(BlockDriverState *bs)
1385 {
1386     BDRVRawState *s = bs->opaque;
1387     int fd = s->fd;
1388     struct stat st;
1389 
1390     if (fstat(fd, &st))
1391         return -errno;
1392     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1393         struct dkwedge_info dkw;
1394 
1395         if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1396             return dkw.dkw_size * 512;
1397         } else {
1398             struct disklabel dl;
1399 
1400             if (ioctl(fd, DIOCGDINFO, &dl))
1401                 return -errno;
1402             return (uint64_t)dl.d_secsize *
1403                 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1404         }
1405     } else
1406         return st.st_size;
1407 }
1408 #elif defined(__sun__)
1409 static int64_t raw_getlength(BlockDriverState *bs)
1410 {
1411     BDRVRawState *s = bs->opaque;
1412     struct dk_minfo minfo;
1413     int ret;
1414     int64_t size;
1415 
1416     ret = fd_open(bs);
1417     if (ret < 0) {
1418         return ret;
1419     }
1420 
1421     /*
1422      * Use the DKIOCGMEDIAINFO ioctl to read the size.
1423      */
1424     ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1425     if (ret != -1) {
1426         return minfo.dki_lbsize * minfo.dki_capacity;
1427     }
1428 
1429     /*
1430      * There are reports that lseek on some devices fails, but
1431      * irc discussion said that contingency on contingency was overkill.
1432      */
1433     size = lseek(s->fd, 0, SEEK_END);
1434     if (size < 0) {
1435         return -errno;
1436     }
1437     return size;
1438 }
1439 #elif defined(CONFIG_BSD)
1440 static int64_t raw_getlength(BlockDriverState *bs)
1441 {
1442     BDRVRawState *s = bs->opaque;
1443     int fd = s->fd;
1444     int64_t size;
1445     struct stat sb;
1446 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1447     int reopened = 0;
1448 #endif
1449     int ret;
1450 
1451     ret = fd_open(bs);
1452     if (ret < 0)
1453         return ret;
1454 
1455 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1456 again:
1457 #endif
1458     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1459 #ifdef DIOCGMEDIASIZE
1460 	if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1461 #elif defined(DIOCGPART)
1462         {
1463                 struct partinfo pi;
1464                 if (ioctl(fd, DIOCGPART, &pi) == 0)
1465                         size = pi.media_size;
1466                 else
1467                         size = 0;
1468         }
1469         if (size == 0)
1470 #endif
1471 #if defined(__APPLE__) && defined(__MACH__)
1472         {
1473             uint64_t sectors = 0;
1474             uint32_t sector_size = 0;
1475 
1476             if (ioctl(fd, DKIOCGETBLOCKCOUNT, &sectors) == 0
1477                && ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) == 0) {
1478                 size = sectors * sector_size;
1479             } else {
1480                 size = lseek(fd, 0LL, SEEK_END);
1481                 if (size < 0) {
1482                     return -errno;
1483                 }
1484             }
1485         }
1486 #else
1487         size = lseek(fd, 0LL, SEEK_END);
1488         if (size < 0) {
1489             return -errno;
1490         }
1491 #endif
1492 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1493         switch(s->type) {
1494         case FTYPE_CD:
1495             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1496             if (size == 2048LL * (unsigned)-1)
1497                 size = 0;
1498             /* XXX no disc?  maybe we need to reopen... */
1499             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1500                 reopened = 1;
1501                 goto again;
1502             }
1503         }
1504 #endif
1505     } else {
1506         size = lseek(fd, 0, SEEK_END);
1507         if (size < 0) {
1508             return -errno;
1509         }
1510     }
1511     return size;
1512 }
1513 #else
1514 static int64_t raw_getlength(BlockDriverState *bs)
1515 {
1516     BDRVRawState *s = bs->opaque;
1517     int ret;
1518     int64_t size;
1519 
1520     ret = fd_open(bs);
1521     if (ret < 0) {
1522         return ret;
1523     }
1524 
1525     size = lseek(s->fd, 0, SEEK_END);
1526     if (size < 0) {
1527         return -errno;
1528     }
1529     return size;
1530 }
1531 #endif
1532 
1533 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1534 {
1535     struct stat st;
1536     BDRVRawState *s = bs->opaque;
1537 
1538     if (fstat(s->fd, &st) < 0) {
1539         return -errno;
1540     }
1541     return (int64_t)st.st_blocks * 512;
1542 }
1543 
1544 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
1545 {
1546     int fd;
1547     int result = 0;
1548     int64_t total_size = 0;
1549     bool nocow = false;
1550     PreallocMode prealloc;
1551     char *buf = NULL;
1552     Error *local_err = NULL;
1553 
1554     strstart(filename, "file:", &filename);
1555 
1556     /* Read out options */
1557     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1558                           BDRV_SECTOR_SIZE);
1559     nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
1560     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1561     prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
1562                                PREALLOC_MODE__MAX, PREALLOC_MODE_OFF,
1563                                &local_err);
1564     g_free(buf);
1565     if (local_err) {
1566         error_propagate(errp, local_err);
1567         result = -EINVAL;
1568         goto out;
1569     }
1570 
1571     fd = qemu_open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY,
1572                    0644);
1573     if (fd < 0) {
1574         result = -errno;
1575         error_setg_errno(errp, -result, "Could not create file");
1576         goto out;
1577     }
1578 
1579     if (nocow) {
1580 #ifdef __linux__
1581         /* Set NOCOW flag to solve performance issue on fs like btrfs.
1582          * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1583          * will be ignored since any failure of this operation should not
1584          * block the left work.
1585          */
1586         int attr;
1587         if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
1588             attr |= FS_NOCOW_FL;
1589             ioctl(fd, FS_IOC_SETFLAGS, &attr);
1590         }
1591 #endif
1592     }
1593 
1594     if (ftruncate(fd, total_size) != 0) {
1595         result = -errno;
1596         error_setg_errno(errp, -result, "Could not resize file");
1597         goto out_close;
1598     }
1599 
1600     switch (prealloc) {
1601 #ifdef CONFIG_POSIX_FALLOCATE
1602     case PREALLOC_MODE_FALLOC:
1603         /* posix_fallocate() doesn't set errno. */
1604         result = -posix_fallocate(fd, 0, total_size);
1605         if (result != 0) {
1606             error_setg_errno(errp, -result,
1607                              "Could not preallocate data for the new file");
1608         }
1609         break;
1610 #endif
1611     case PREALLOC_MODE_FULL:
1612     {
1613         int64_t num = 0, left = total_size;
1614         buf = g_malloc0(65536);
1615 
1616         while (left > 0) {
1617             num = MIN(left, 65536);
1618             result = write(fd, buf, num);
1619             if (result < 0) {
1620                 result = -errno;
1621                 error_setg_errno(errp, -result,
1622                                  "Could not write to the new file");
1623                 break;
1624             }
1625             left -= result;
1626         }
1627         if (result >= 0) {
1628             result = fsync(fd);
1629             if (result < 0) {
1630                 result = -errno;
1631                 error_setg_errno(errp, -result,
1632                                  "Could not flush new file to disk");
1633             }
1634         }
1635         g_free(buf);
1636         break;
1637     }
1638     case PREALLOC_MODE_OFF:
1639         break;
1640     default:
1641         result = -EINVAL;
1642         error_setg(errp, "Unsupported preallocation mode: %s",
1643                    PreallocMode_lookup[prealloc]);
1644         break;
1645     }
1646 
1647 out_close:
1648     if (qemu_close(fd) != 0 && result == 0) {
1649         result = -errno;
1650         error_setg_errno(errp, -result, "Could not close the new file");
1651     }
1652 out:
1653     return result;
1654 }
1655 
1656 /*
1657  * Find allocation range in @bs around offset @start.
1658  * May change underlying file descriptor's file offset.
1659  * If @start is not in a hole, store @start in @data, and the
1660  * beginning of the next hole in @hole, and return 0.
1661  * If @start is in a non-trailing hole, store @start in @hole and the
1662  * beginning of the next non-hole in @data, and return 0.
1663  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
1664  * If we can't find out, return a negative errno other than -ENXIO.
1665  */
1666 static int find_allocation(BlockDriverState *bs, off_t start,
1667                            off_t *data, off_t *hole)
1668 {
1669 #if defined SEEK_HOLE && defined SEEK_DATA
1670     BDRVRawState *s = bs->opaque;
1671     off_t offs;
1672 
1673     /*
1674      * SEEK_DATA cases:
1675      * D1. offs == start: start is in data
1676      * D2. offs > start: start is in a hole, next data at offs
1677      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
1678      *                              or start is beyond EOF
1679      *     If the latter happens, the file has been truncated behind
1680      *     our back since we opened it.  All bets are off then.
1681      *     Treating like a trailing hole is simplest.
1682      * D4. offs < 0, errno != ENXIO: we learned nothing
1683      */
1684     offs = lseek(s->fd, start, SEEK_DATA);
1685     if (offs < 0) {
1686         return -errno;          /* D3 or D4 */
1687     }
1688     assert(offs >= start);
1689 
1690     if (offs > start) {
1691         /* D2: in hole, next data at offs */
1692         *hole = start;
1693         *data = offs;
1694         return 0;
1695     }
1696 
1697     /* D1: in data, end not yet known */
1698 
1699     /*
1700      * SEEK_HOLE cases:
1701      * H1. offs == start: start is in a hole
1702      *     If this happens here, a hole has been dug behind our back
1703      *     since the previous lseek().
1704      * H2. offs > start: either start is in data, next hole at offs,
1705      *                   or start is in trailing hole, EOF at offs
1706      *     Linux treats trailing holes like any other hole: offs ==
1707      *     start.  Solaris seeks to EOF instead: offs > start (blech).
1708      *     If that happens here, a hole has been dug behind our back
1709      *     since the previous lseek().
1710      * H3. offs < 0, errno = ENXIO: start is beyond EOF
1711      *     If this happens, the file has been truncated behind our
1712      *     back since we opened it.  Treat it like a trailing hole.
1713      * H4. offs < 0, errno != ENXIO: we learned nothing
1714      *     Pretend we know nothing at all, i.e. "forget" about D1.
1715      */
1716     offs = lseek(s->fd, start, SEEK_HOLE);
1717     if (offs < 0) {
1718         return -errno;          /* D1 and (H3 or H4) */
1719     }
1720     assert(offs >= start);
1721 
1722     if (offs > start) {
1723         /*
1724          * D1 and H2: either in data, next hole at offs, or it was in
1725          * data but is now in a trailing hole.  In the latter case,
1726          * all bets are off.  Treating it as if it there was data all
1727          * the way to EOF is safe, so simply do that.
1728          */
1729         *data = start;
1730         *hole = offs;
1731         return 0;
1732     }
1733 
1734     /* D1 and H1 */
1735     return -EBUSY;
1736 #else
1737     return -ENOTSUP;
1738 #endif
1739 }
1740 
1741 /*
1742  * Returns the allocation status of the specified sectors.
1743  *
1744  * If 'sector_num' is beyond the end of the disk image the return value is 0
1745  * and 'pnum' is set to 0.
1746  *
1747  * 'pnum' is set to the number of sectors (including and immediately following
1748  * the specified sector) that are known to be in the same
1749  * allocated/unallocated state.
1750  *
1751  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
1752  * beyond the end of the disk image it will be clamped.
1753  */
1754 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1755                                                     int64_t sector_num,
1756                                                     int nb_sectors, int *pnum,
1757                                                     BlockDriverState **file)
1758 {
1759     off_t start, data = 0, hole = 0;
1760     int64_t total_size;
1761     int ret;
1762 
1763     ret = fd_open(bs);
1764     if (ret < 0) {
1765         return ret;
1766     }
1767 
1768     start = sector_num * BDRV_SECTOR_SIZE;
1769     total_size = bdrv_getlength(bs);
1770     if (total_size < 0) {
1771         return total_size;
1772     } else if (start >= total_size) {
1773         *pnum = 0;
1774         return 0;
1775     } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1776         nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1777     }
1778 
1779     ret = find_allocation(bs, start, &data, &hole);
1780     if (ret == -ENXIO) {
1781         /* Trailing hole */
1782         *pnum = nb_sectors;
1783         ret = BDRV_BLOCK_ZERO;
1784     } else if (ret < 0) {
1785         /* No info available, so pretend there are no holes */
1786         *pnum = nb_sectors;
1787         ret = BDRV_BLOCK_DATA;
1788     } else if (data == start) {
1789         /* On a data extent, compute sectors to the end of the extent,
1790          * possibly including a partial sector at EOF. */
1791         *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE));
1792         ret = BDRV_BLOCK_DATA;
1793     } else {
1794         /* On a hole, compute sectors to the beginning of the next extent.  */
1795         assert(hole == start);
1796         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1797         ret = BDRV_BLOCK_ZERO;
1798     }
1799     *file = bs;
1800     return ret | BDRV_BLOCK_OFFSET_VALID | start;
1801 }
1802 
1803 static coroutine_fn BlockAIOCB *raw_aio_pdiscard(BlockDriverState *bs,
1804     int64_t offset, int count,
1805     BlockCompletionFunc *cb, void *opaque)
1806 {
1807     BDRVRawState *s = bs->opaque;
1808 
1809     return paio_submit(bs, s->fd, offset, NULL, count,
1810                        cb, opaque, QEMU_AIO_DISCARD);
1811 }
1812 
1813 static int coroutine_fn raw_co_pwrite_zeroes(
1814     BlockDriverState *bs, int64_t offset,
1815     int count, BdrvRequestFlags flags)
1816 {
1817     BDRVRawState *s = bs->opaque;
1818 
1819     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1820         return paio_submit_co(bs, s->fd, offset, NULL, count,
1821                               QEMU_AIO_WRITE_ZEROES);
1822     } else if (s->discard_zeroes) {
1823         return paio_submit_co(bs, s->fd, offset, NULL, count,
1824                               QEMU_AIO_DISCARD);
1825     }
1826     return -ENOTSUP;
1827 }
1828 
1829 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1830 {
1831     BDRVRawState *s = bs->opaque;
1832 
1833     bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1834     bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1835     return 0;
1836 }
1837 
1838 static QemuOptsList raw_create_opts = {
1839     .name = "raw-create-opts",
1840     .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
1841     .desc = {
1842         {
1843             .name = BLOCK_OPT_SIZE,
1844             .type = QEMU_OPT_SIZE,
1845             .help = "Virtual disk size"
1846         },
1847         {
1848             .name = BLOCK_OPT_NOCOW,
1849             .type = QEMU_OPT_BOOL,
1850             .help = "Turn off copy-on-write (valid only on btrfs)"
1851         },
1852         {
1853             .name = BLOCK_OPT_PREALLOC,
1854             .type = QEMU_OPT_STRING,
1855             .help = "Preallocation mode (allowed values: off, falloc, full)"
1856         },
1857         { /* end of list */ }
1858     }
1859 };
1860 
1861 BlockDriver bdrv_file = {
1862     .format_name = "file",
1863     .protocol_name = "file",
1864     .instance_size = sizeof(BDRVRawState),
1865     .bdrv_needs_filename = true,
1866     .bdrv_probe = NULL, /* no probe for protocols */
1867     .bdrv_parse_filename = raw_parse_filename,
1868     .bdrv_file_open = raw_open,
1869     .bdrv_reopen_prepare = raw_reopen_prepare,
1870     .bdrv_reopen_commit = raw_reopen_commit,
1871     .bdrv_reopen_abort = raw_reopen_abort,
1872     .bdrv_close = raw_close,
1873     .bdrv_create = raw_create,
1874     .bdrv_has_zero_init = bdrv_has_zero_init_1,
1875     .bdrv_co_get_block_status = raw_co_get_block_status,
1876     .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes,
1877 
1878     .bdrv_co_preadv         = raw_co_preadv,
1879     .bdrv_co_pwritev        = raw_co_pwritev,
1880     .bdrv_aio_flush = raw_aio_flush,
1881     .bdrv_aio_pdiscard = raw_aio_pdiscard,
1882     .bdrv_refresh_limits = raw_refresh_limits,
1883     .bdrv_io_plug = raw_aio_plug,
1884     .bdrv_io_unplug = raw_aio_unplug,
1885 
1886     .bdrv_truncate = raw_truncate,
1887     .bdrv_getlength = raw_getlength,
1888     .bdrv_get_info = raw_get_info,
1889     .bdrv_get_allocated_file_size
1890                         = raw_get_allocated_file_size,
1891 
1892     .create_opts = &raw_create_opts,
1893 };
1894 
1895 /***********************************************/
1896 /* host device */
1897 
1898 #if defined(__APPLE__) && defined(__MACH__)
1899 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
1900                                 CFIndex maxPathSize, int flags);
1901 static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
1902 {
1903     kern_return_t kernResult = KERN_FAILURE;
1904     mach_port_t     masterPort;
1905     CFMutableDictionaryRef  classesToMatch;
1906     const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
1907     char *mediaType = NULL;
1908 
1909     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1910     if ( KERN_SUCCESS != kernResult ) {
1911         printf( "IOMasterPort returned %d\n", kernResult );
1912     }
1913 
1914     int index;
1915     for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
1916         classesToMatch = IOServiceMatching(matching_array[index]);
1917         if (classesToMatch == NULL) {
1918             error_report("IOServiceMatching returned NULL for %s",
1919                          matching_array[index]);
1920             continue;
1921         }
1922         CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
1923                              kCFBooleanTrue);
1924         kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch,
1925                                                   mediaIterator);
1926         if (kernResult != KERN_SUCCESS) {
1927             error_report("Note: IOServiceGetMatchingServices returned %d",
1928                          kernResult);
1929             continue;
1930         }
1931 
1932         /* If a match was found, leave the loop */
1933         if (*mediaIterator != 0) {
1934             DPRINTF("Matching using %s\n", matching_array[index]);
1935             mediaType = g_strdup(matching_array[index]);
1936             break;
1937         }
1938     }
1939     return mediaType;
1940 }
1941 
1942 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
1943                          CFIndex maxPathSize, int flags)
1944 {
1945     io_object_t     nextMedia;
1946     kern_return_t   kernResult = KERN_FAILURE;
1947     *bsdPath = '\0';
1948     nextMedia = IOIteratorNext( mediaIterator );
1949     if ( nextMedia )
1950     {
1951         CFTypeRef   bsdPathAsCFString;
1952     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1953         if ( bsdPathAsCFString ) {
1954             size_t devPathLength;
1955             strcpy( bsdPath, _PATH_DEV );
1956             if (flags & BDRV_O_NOCACHE) {
1957                 strcat(bsdPath, "r");
1958             }
1959             devPathLength = strlen( bsdPath );
1960             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1961                 kernResult = KERN_SUCCESS;
1962             }
1963             CFRelease( bsdPathAsCFString );
1964         }
1965         IOObjectRelease( nextMedia );
1966     }
1967 
1968     return kernResult;
1969 }
1970 
1971 /* Sets up a real cdrom for use in QEMU */
1972 static bool setup_cdrom(char *bsd_path, Error **errp)
1973 {
1974     int index, num_of_test_partitions = 2, fd;
1975     char test_partition[MAXPATHLEN];
1976     bool partition_found = false;
1977 
1978     /* look for a working partition */
1979     for (index = 0; index < num_of_test_partitions; index++) {
1980         snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
1981                  index);
1982         fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE);
1983         if (fd >= 0) {
1984             partition_found = true;
1985             qemu_close(fd);
1986             break;
1987         }
1988     }
1989 
1990     /* if a working partition on the device was not found */
1991     if (partition_found == false) {
1992         error_setg(errp, "Failed to find a working partition on disc");
1993     } else {
1994         DPRINTF("Using %s as optical disc\n", test_partition);
1995         pstrcpy(bsd_path, MAXPATHLEN, test_partition);
1996     }
1997     return partition_found;
1998 }
1999 
2000 /* Prints directions on mounting and unmounting a device */
2001 static void print_unmounting_directions(const char *file_name)
2002 {
2003     error_report("If device %s is mounted on the desktop, unmount"
2004                  " it first before using it in QEMU", file_name);
2005     error_report("Command to unmount device: diskutil unmountDisk %s",
2006                  file_name);
2007     error_report("Command to mount device: diskutil mountDisk %s", file_name);
2008 }
2009 
2010 #endif /* defined(__APPLE__) && defined(__MACH__) */
2011 
2012 static int hdev_probe_device(const char *filename)
2013 {
2014     struct stat st;
2015 
2016     /* allow a dedicated CD-ROM driver to match with a higher priority */
2017     if (strstart(filename, "/dev/cdrom", NULL))
2018         return 50;
2019 
2020     if (stat(filename, &st) >= 0 &&
2021             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
2022         return 100;
2023     }
2024 
2025     return 0;
2026 }
2027 
2028 static int check_hdev_writable(BDRVRawState *s)
2029 {
2030 #if defined(BLKROGET)
2031     /* Linux block devices can be configured "read-only" using blockdev(8).
2032      * This is independent of device node permissions and therefore open(2)
2033      * with O_RDWR succeeds.  Actual writes fail with EPERM.
2034      *
2035      * bdrv_open() is supposed to fail if the disk is read-only.  Explicitly
2036      * check for read-only block devices so that Linux block devices behave
2037      * properly.
2038      */
2039     struct stat st;
2040     int readonly = 0;
2041 
2042     if (fstat(s->fd, &st)) {
2043         return -errno;
2044     }
2045 
2046     if (!S_ISBLK(st.st_mode)) {
2047         return 0;
2048     }
2049 
2050     if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
2051         return -errno;
2052     }
2053 
2054     if (readonly) {
2055         return -EACCES;
2056     }
2057 #endif /* defined(BLKROGET) */
2058     return 0;
2059 }
2060 
2061 static void hdev_parse_filename(const char *filename, QDict *options,
2062                                 Error **errp)
2063 {
2064     /* The prefix is optional, just as for "file". */
2065     strstart(filename, "host_device:", &filename);
2066 
2067     qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2068 }
2069 
2070 static bool hdev_is_sg(BlockDriverState *bs)
2071 {
2072 
2073 #if defined(__linux__)
2074 
2075     BDRVRawState *s = bs->opaque;
2076     struct stat st;
2077     struct sg_scsi_id scsiid;
2078     int sg_version;
2079     int ret;
2080 
2081     if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) {
2082         return false;
2083     }
2084 
2085     ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version);
2086     if (ret < 0) {
2087         return false;
2088     }
2089 
2090     ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid);
2091     if (ret >= 0) {
2092         DPRINTF("SG device found: type=%d, version=%d\n",
2093             scsiid.scsi_type, sg_version);
2094         return true;
2095     }
2096 
2097 #endif
2098 
2099     return false;
2100 }
2101 
2102 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
2103                      Error **errp)
2104 {
2105     BDRVRawState *s = bs->opaque;
2106     Error *local_err = NULL;
2107     int ret;
2108 
2109 #if defined(__APPLE__) && defined(__MACH__)
2110     const char *filename = qdict_get_str(options, "filename");
2111     char bsd_path[MAXPATHLEN] = "";
2112     bool error_occurred = false;
2113 
2114     /* If using a real cdrom */
2115     if (strcmp(filename, "/dev/cdrom") == 0) {
2116         char *mediaType = NULL;
2117         kern_return_t ret_val;
2118         io_iterator_t mediaIterator = 0;
2119 
2120         mediaType = FindEjectableOpticalMedia(&mediaIterator);
2121         if (mediaType == NULL) {
2122             error_setg(errp, "Please make sure your CD/DVD is in the optical"
2123                        " drive");
2124             error_occurred = true;
2125             goto hdev_open_Mac_error;
2126         }
2127 
2128         ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
2129         if (ret_val != KERN_SUCCESS) {
2130             error_setg(errp, "Could not get BSD path for optical drive");
2131             error_occurred = true;
2132             goto hdev_open_Mac_error;
2133         }
2134 
2135         /* If a real optical drive was not found */
2136         if (bsd_path[0] == '\0') {
2137             error_setg(errp, "Failed to obtain bsd path for optical drive");
2138             error_occurred = true;
2139             goto hdev_open_Mac_error;
2140         }
2141 
2142         /* If using a cdrom disc and finding a partition on the disc failed */
2143         if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
2144             setup_cdrom(bsd_path, errp) == false) {
2145             print_unmounting_directions(bsd_path);
2146             error_occurred = true;
2147             goto hdev_open_Mac_error;
2148         }
2149 
2150         qdict_put(options, "filename", qstring_from_str(bsd_path));
2151 
2152 hdev_open_Mac_error:
2153         g_free(mediaType);
2154         if (mediaIterator) {
2155             IOObjectRelease(mediaIterator);
2156         }
2157         if (error_occurred) {
2158             return -ENOENT;
2159         }
2160     }
2161 #endif /* defined(__APPLE__) && defined(__MACH__) */
2162 
2163     s->type = FTYPE_FILE;
2164 
2165     ret = raw_open_common(bs, options, flags, 0, &local_err);
2166     if (ret < 0) {
2167         error_propagate(errp, local_err);
2168 #if defined(__APPLE__) && defined(__MACH__)
2169         if (*bsd_path) {
2170             filename = bsd_path;
2171         }
2172         /* if a physical device experienced an error while being opened */
2173         if (strncmp(filename, "/dev/", 5) == 0) {
2174             print_unmounting_directions(filename);
2175         }
2176 #endif /* defined(__APPLE__) && defined(__MACH__) */
2177         return ret;
2178     }
2179 
2180     /* Since this does ioctl the device must be already opened */
2181     bs->sg = hdev_is_sg(bs);
2182 
2183     if (flags & BDRV_O_RDWR) {
2184         ret = check_hdev_writable(s);
2185         if (ret < 0) {
2186             raw_close(bs);
2187             error_setg_errno(errp, -ret, "The device is not writable");
2188             return ret;
2189         }
2190     }
2191 
2192     return ret;
2193 }
2194 
2195 #if defined(__linux__)
2196 
2197 static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
2198         unsigned long int req, void *buf,
2199         BlockCompletionFunc *cb, void *opaque)
2200 {
2201     BDRVRawState *s = bs->opaque;
2202     RawPosixAIOData *acb;
2203     ThreadPool *pool;
2204 
2205     if (fd_open(bs) < 0)
2206         return NULL;
2207 
2208     acb = g_new(RawPosixAIOData, 1);
2209     acb->bs = bs;
2210     acb->aio_type = QEMU_AIO_IOCTL;
2211     acb->aio_fildes = s->fd;
2212     acb->aio_offset = 0;
2213     acb->aio_ioctl_buf = buf;
2214     acb->aio_ioctl_cmd = req;
2215     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
2216     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
2217 }
2218 #endif /* linux */
2219 
2220 static int fd_open(BlockDriverState *bs)
2221 {
2222     BDRVRawState *s = bs->opaque;
2223 
2224     /* this is just to ensure s->fd is sane (its called by io ops) */
2225     if (s->fd >= 0)
2226         return 0;
2227     return -EIO;
2228 }
2229 
2230 static coroutine_fn BlockAIOCB *hdev_aio_pdiscard(BlockDriverState *bs,
2231     int64_t offset, int count,
2232     BlockCompletionFunc *cb, void *opaque)
2233 {
2234     BDRVRawState *s = bs->opaque;
2235 
2236     if (fd_open(bs) < 0) {
2237         return NULL;
2238     }
2239     return paio_submit(bs, s->fd, offset, NULL, count,
2240                        cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2241 }
2242 
2243 static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs,
2244     int64_t offset, int count, BdrvRequestFlags flags)
2245 {
2246     BDRVRawState *s = bs->opaque;
2247     int rc;
2248 
2249     rc = fd_open(bs);
2250     if (rc < 0) {
2251         return rc;
2252     }
2253     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2254         return paio_submit_co(bs, s->fd, offset, NULL, count,
2255                               QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2256     } else if (s->discard_zeroes) {
2257         return paio_submit_co(bs, s->fd, offset, NULL, count,
2258                               QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2259     }
2260     return -ENOTSUP;
2261 }
2262 
2263 static int hdev_create(const char *filename, QemuOpts *opts,
2264                        Error **errp)
2265 {
2266     int fd;
2267     int ret = 0;
2268     struct stat stat_buf;
2269     int64_t total_size = 0;
2270     bool has_prefix;
2271 
2272     /* This function is used by both protocol block drivers and therefore either
2273      * of these prefixes may be given.
2274      * The return value has to be stored somewhere, otherwise this is an error
2275      * due to -Werror=unused-value. */
2276     has_prefix =
2277         strstart(filename, "host_device:", &filename) ||
2278         strstart(filename, "host_cdrom:" , &filename);
2279 
2280     (void)has_prefix;
2281 
2282     ret = raw_normalize_devicepath(&filename);
2283     if (ret < 0) {
2284         error_setg_errno(errp, -ret, "Could not normalize device path");
2285         return ret;
2286     }
2287 
2288     /* Read out options */
2289     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2290                           BDRV_SECTOR_SIZE);
2291 
2292     fd = qemu_open(filename, O_WRONLY | O_BINARY);
2293     if (fd < 0) {
2294         ret = -errno;
2295         error_setg_errno(errp, -ret, "Could not open device");
2296         return ret;
2297     }
2298 
2299     if (fstat(fd, &stat_buf) < 0) {
2300         ret = -errno;
2301         error_setg_errno(errp, -ret, "Could not stat device");
2302     } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2303         error_setg(errp,
2304                    "The given file is neither a block nor a character device");
2305         ret = -ENODEV;
2306     } else if (lseek(fd, 0, SEEK_END) < total_size) {
2307         error_setg(errp, "Device is too small");
2308         ret = -ENOSPC;
2309     }
2310 
2311     qemu_close(fd);
2312     return ret;
2313 }
2314 
2315 static BlockDriver bdrv_host_device = {
2316     .format_name        = "host_device",
2317     .protocol_name        = "host_device",
2318     .instance_size      = sizeof(BDRVRawState),
2319     .bdrv_needs_filename = true,
2320     .bdrv_probe_device  = hdev_probe_device,
2321     .bdrv_parse_filename = hdev_parse_filename,
2322     .bdrv_file_open     = hdev_open,
2323     .bdrv_close         = raw_close,
2324     .bdrv_reopen_prepare = raw_reopen_prepare,
2325     .bdrv_reopen_commit  = raw_reopen_commit,
2326     .bdrv_reopen_abort   = raw_reopen_abort,
2327     .bdrv_create         = hdev_create,
2328     .create_opts         = &raw_create_opts,
2329     .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes,
2330 
2331     .bdrv_co_preadv         = raw_co_preadv,
2332     .bdrv_co_pwritev        = raw_co_pwritev,
2333     .bdrv_aio_flush	= raw_aio_flush,
2334     .bdrv_aio_pdiscard   = hdev_aio_pdiscard,
2335     .bdrv_refresh_limits = raw_refresh_limits,
2336     .bdrv_io_plug = raw_aio_plug,
2337     .bdrv_io_unplug = raw_aio_unplug,
2338 
2339     .bdrv_truncate      = raw_truncate,
2340     .bdrv_getlength	= raw_getlength,
2341     .bdrv_get_info = raw_get_info,
2342     .bdrv_get_allocated_file_size
2343                         = raw_get_allocated_file_size,
2344     .bdrv_probe_blocksizes = hdev_probe_blocksizes,
2345     .bdrv_probe_geometry = hdev_probe_geometry,
2346 
2347     /* generic scsi device */
2348 #ifdef __linux__
2349     .bdrv_aio_ioctl     = hdev_aio_ioctl,
2350 #endif
2351 };
2352 
2353 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2354 static void cdrom_parse_filename(const char *filename, QDict *options,
2355                                  Error **errp)
2356 {
2357     /* The prefix is optional, just as for "file". */
2358     strstart(filename, "host_cdrom:", &filename);
2359 
2360     qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2361 }
2362 #endif
2363 
2364 #ifdef __linux__
2365 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2366                       Error **errp)
2367 {
2368     BDRVRawState *s = bs->opaque;
2369 
2370     s->type = FTYPE_CD;
2371 
2372     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2373     return raw_open_common(bs, options, flags, O_NONBLOCK, errp);
2374 }
2375 
2376 static int cdrom_probe_device(const char *filename)
2377 {
2378     int fd, ret;
2379     int prio = 0;
2380     struct stat st;
2381 
2382     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2383     if (fd < 0) {
2384         goto out;
2385     }
2386     ret = fstat(fd, &st);
2387     if (ret == -1 || !S_ISBLK(st.st_mode)) {
2388         goto outc;
2389     }
2390 
2391     /* Attempt to detect via a CDROM specific ioctl */
2392     ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2393     if (ret >= 0)
2394         prio = 100;
2395 
2396 outc:
2397     qemu_close(fd);
2398 out:
2399     return prio;
2400 }
2401 
2402 static bool cdrom_is_inserted(BlockDriverState *bs)
2403 {
2404     BDRVRawState *s = bs->opaque;
2405     int ret;
2406 
2407     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2408     return ret == CDS_DISC_OK;
2409 }
2410 
2411 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2412 {
2413     BDRVRawState *s = bs->opaque;
2414 
2415     if (eject_flag) {
2416         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2417             perror("CDROMEJECT");
2418     } else {
2419         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2420             perror("CDROMEJECT");
2421     }
2422 }
2423 
2424 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2425 {
2426     BDRVRawState *s = bs->opaque;
2427 
2428     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2429         /*
2430          * Note: an error can happen if the distribution automatically
2431          * mounts the CD-ROM
2432          */
2433         /* perror("CDROM_LOCKDOOR"); */
2434     }
2435 }
2436 
2437 static BlockDriver bdrv_host_cdrom = {
2438     .format_name        = "host_cdrom",
2439     .protocol_name      = "host_cdrom",
2440     .instance_size      = sizeof(BDRVRawState),
2441     .bdrv_needs_filename = true,
2442     .bdrv_probe_device	= cdrom_probe_device,
2443     .bdrv_parse_filename = cdrom_parse_filename,
2444     .bdrv_file_open     = cdrom_open,
2445     .bdrv_close         = raw_close,
2446     .bdrv_reopen_prepare = raw_reopen_prepare,
2447     .bdrv_reopen_commit  = raw_reopen_commit,
2448     .bdrv_reopen_abort   = raw_reopen_abort,
2449     .bdrv_create         = hdev_create,
2450     .create_opts         = &raw_create_opts,
2451 
2452 
2453     .bdrv_co_preadv         = raw_co_preadv,
2454     .bdrv_co_pwritev        = raw_co_pwritev,
2455     .bdrv_aio_flush	= raw_aio_flush,
2456     .bdrv_refresh_limits = raw_refresh_limits,
2457     .bdrv_io_plug = raw_aio_plug,
2458     .bdrv_io_unplug = raw_aio_unplug,
2459 
2460     .bdrv_truncate      = raw_truncate,
2461     .bdrv_getlength      = raw_getlength,
2462     .has_variable_length = true,
2463     .bdrv_get_allocated_file_size
2464                         = raw_get_allocated_file_size,
2465 
2466     /* removable device support */
2467     .bdrv_is_inserted   = cdrom_is_inserted,
2468     .bdrv_eject         = cdrom_eject,
2469     .bdrv_lock_medium   = cdrom_lock_medium,
2470 
2471     /* generic scsi device */
2472     .bdrv_aio_ioctl     = hdev_aio_ioctl,
2473 };
2474 #endif /* __linux__ */
2475 
2476 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2477 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2478                       Error **errp)
2479 {
2480     BDRVRawState *s = bs->opaque;
2481     Error *local_err = NULL;
2482     int ret;
2483 
2484     s->type = FTYPE_CD;
2485 
2486     ret = raw_open_common(bs, options, flags, 0, &local_err);
2487     if (ret) {
2488         error_propagate(errp, local_err);
2489         return ret;
2490     }
2491 
2492     /* make sure the door isn't locked at this time */
2493     ioctl(s->fd, CDIOCALLOW);
2494     return 0;
2495 }
2496 
2497 static int cdrom_probe_device(const char *filename)
2498 {
2499     if (strstart(filename, "/dev/cd", NULL) ||
2500             strstart(filename, "/dev/acd", NULL))
2501         return 100;
2502     return 0;
2503 }
2504 
2505 static int cdrom_reopen(BlockDriverState *bs)
2506 {
2507     BDRVRawState *s = bs->opaque;
2508     int fd;
2509 
2510     /*
2511      * Force reread of possibly changed/newly loaded disc,
2512      * FreeBSD seems to not notice sometimes...
2513      */
2514     if (s->fd >= 0)
2515         qemu_close(s->fd);
2516     fd = qemu_open(bs->filename, s->open_flags, 0644);
2517     if (fd < 0) {
2518         s->fd = -1;
2519         return -EIO;
2520     }
2521     s->fd = fd;
2522 
2523     /* make sure the door isn't locked at this time */
2524     ioctl(s->fd, CDIOCALLOW);
2525     return 0;
2526 }
2527 
2528 static bool cdrom_is_inserted(BlockDriverState *bs)
2529 {
2530     return raw_getlength(bs) > 0;
2531 }
2532 
2533 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2534 {
2535     BDRVRawState *s = bs->opaque;
2536 
2537     if (s->fd < 0)
2538         return;
2539 
2540     (void) ioctl(s->fd, CDIOCALLOW);
2541 
2542     if (eject_flag) {
2543         if (ioctl(s->fd, CDIOCEJECT) < 0)
2544             perror("CDIOCEJECT");
2545     } else {
2546         if (ioctl(s->fd, CDIOCCLOSE) < 0)
2547             perror("CDIOCCLOSE");
2548     }
2549 
2550     cdrom_reopen(bs);
2551 }
2552 
2553 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2554 {
2555     BDRVRawState *s = bs->opaque;
2556 
2557     if (s->fd < 0)
2558         return;
2559     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2560         /*
2561          * Note: an error can happen if the distribution automatically
2562          * mounts the CD-ROM
2563          */
2564         /* perror("CDROM_LOCKDOOR"); */
2565     }
2566 }
2567 
2568 static BlockDriver bdrv_host_cdrom = {
2569     .format_name        = "host_cdrom",
2570     .protocol_name      = "host_cdrom",
2571     .instance_size      = sizeof(BDRVRawState),
2572     .bdrv_needs_filename = true,
2573     .bdrv_probe_device	= cdrom_probe_device,
2574     .bdrv_parse_filename = cdrom_parse_filename,
2575     .bdrv_file_open     = cdrom_open,
2576     .bdrv_close         = raw_close,
2577     .bdrv_reopen_prepare = raw_reopen_prepare,
2578     .bdrv_reopen_commit  = raw_reopen_commit,
2579     .bdrv_reopen_abort   = raw_reopen_abort,
2580     .bdrv_create        = hdev_create,
2581     .create_opts        = &raw_create_opts,
2582 
2583     .bdrv_co_preadv         = raw_co_preadv,
2584     .bdrv_co_pwritev        = raw_co_pwritev,
2585     .bdrv_aio_flush	= raw_aio_flush,
2586     .bdrv_refresh_limits = raw_refresh_limits,
2587     .bdrv_io_plug = raw_aio_plug,
2588     .bdrv_io_unplug = raw_aio_unplug,
2589 
2590     .bdrv_truncate      = raw_truncate,
2591     .bdrv_getlength      = raw_getlength,
2592     .has_variable_length = true,
2593     .bdrv_get_allocated_file_size
2594                         = raw_get_allocated_file_size,
2595 
2596     /* removable device support */
2597     .bdrv_is_inserted   = cdrom_is_inserted,
2598     .bdrv_eject         = cdrom_eject,
2599     .bdrv_lock_medium   = cdrom_lock_medium,
2600 };
2601 #endif /* __FreeBSD__ */
2602 
2603 static void bdrv_file_init(void)
2604 {
2605     /*
2606      * Register all the drivers.  Note that order is important, the driver
2607      * registered last will get probed first.
2608      */
2609     bdrv_register(&bdrv_file);
2610     bdrv_register(&bdrv_host_device);
2611 #ifdef __linux__
2612     bdrv_register(&bdrv_host_cdrom);
2613 #endif
2614 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2615     bdrv_register(&bdrv_host_cdrom);
2616 #endif
2617 }
2618 
2619 block_init(bdrv_file_init);
2620