xref: /openbmc/qemu/block/file-posix.c (revision dc5bd18f)
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 
25 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu/cutils.h"
28 #include "qemu/error-report.h"
29 #include "block/block_int.h"
30 #include "qemu/module.h"
31 #include "qemu/option.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/qmp/qdict.h"
37 #include "qapi/qmp/qstring.h"
38 
39 #include "scsi/pr-manager.h"
40 #include "scsi/constants.h"
41 
42 #if defined(__APPLE__) && (__MACH__)
43 #include <paths.h>
44 #include <sys/param.h>
45 #include <IOKit/IOKitLib.h>
46 #include <IOKit/IOBSD.h>
47 #include <IOKit/storage/IOMediaBSDClient.h>
48 #include <IOKit/storage/IOMedia.h>
49 #include <IOKit/storage/IOCDMedia.h>
50 //#include <IOKit/storage/IOCDTypes.h>
51 #include <IOKit/storage/IODVDMedia.h>
52 #include <CoreFoundation/CoreFoundation.h>
53 #endif
54 
55 #ifdef __sun__
56 #define _POSIX_PTHREAD_SEMANTICS 1
57 #include <sys/dkio.h>
58 #endif
59 #ifdef __linux__
60 #include <sys/ioctl.h>
61 #include <sys/param.h>
62 #include <linux/cdrom.h>
63 #include <linux/fd.h>
64 #include <linux/fs.h>
65 #include <linux/hdreg.h>
66 #include <scsi/sg.h>
67 #ifdef __s390__
68 #include <asm/dasd.h>
69 #endif
70 #ifndef FS_NOCOW_FL
71 #define FS_NOCOW_FL                     0x00800000 /* Do not cow file */
72 #endif
73 #endif
74 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
75 #include <linux/falloc.h>
76 #endif
77 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
78 #include <sys/disk.h>
79 #include <sys/cdio.h>
80 #endif
81 
82 #ifdef __OpenBSD__
83 #include <sys/ioctl.h>
84 #include <sys/disklabel.h>
85 #include <sys/dkio.h>
86 #endif
87 
88 #ifdef __NetBSD__
89 #include <sys/ioctl.h>
90 #include <sys/disklabel.h>
91 #include <sys/dkio.h>
92 #include <sys/disk.h>
93 #endif
94 
95 #ifdef __DragonFly__
96 #include <sys/ioctl.h>
97 #include <sys/diskslice.h>
98 #endif
99 
100 #ifdef CONFIG_XFS
101 #include <xfs/xfs.h>
102 #endif
103 
104 //#define DEBUG_BLOCK
105 
106 #ifdef DEBUG_BLOCK
107 # define DEBUG_BLOCK_PRINT 1
108 #else
109 # define DEBUG_BLOCK_PRINT 0
110 #endif
111 #define DPRINTF(fmt, ...) \
112 do { \
113     if (DEBUG_BLOCK_PRINT) { \
114         printf(fmt, ## __VA_ARGS__); \
115     } \
116 } while (0)
117 
118 /* OS X does not have O_DSYNC */
119 #ifndef O_DSYNC
120 #ifdef O_SYNC
121 #define O_DSYNC O_SYNC
122 #elif defined(O_FSYNC)
123 #define O_DSYNC O_FSYNC
124 #endif
125 #endif
126 
127 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
128 #ifndef O_DIRECT
129 #define O_DIRECT O_DSYNC
130 #endif
131 
132 #define FTYPE_FILE   0
133 #define FTYPE_CD     1
134 
135 #define MAX_BLOCKSIZE	4096
136 
137 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
138  * leaving a few more bytes for its future use. */
139 #define RAW_LOCK_PERM_BASE             100
140 #define RAW_LOCK_SHARED_BASE           200
141 
142 typedef struct BDRVRawState {
143     int fd;
144     int lock_fd;
145     bool use_lock;
146     int type;
147     int open_flags;
148     size_t buf_align;
149 
150     /* The current permissions. */
151     uint64_t perm;
152     uint64_t shared_perm;
153 
154 #ifdef CONFIG_XFS
155     bool is_xfs:1;
156 #endif
157     bool has_discard:1;
158     bool has_write_zeroes:1;
159     bool discard_zeroes:1;
160     bool use_linux_aio:1;
161     bool page_cache_inconsistent:1;
162     bool has_fallocate;
163     bool needs_alignment;
164 
165     PRManager *pr_mgr;
166 } BDRVRawState;
167 
168 typedef struct BDRVRawReopenState {
169     int fd;
170     int open_flags;
171 } BDRVRawReopenState;
172 
173 static int fd_open(BlockDriverState *bs);
174 static int64_t raw_getlength(BlockDriverState *bs);
175 
176 typedef struct RawPosixAIOData {
177     BlockDriverState *bs;
178     int aio_fildes;
179     union {
180         struct iovec *aio_iov;
181         void *aio_ioctl_buf;
182     };
183     int aio_niov;
184     uint64_t aio_nbytes;
185 #define aio_ioctl_cmd   aio_nbytes /* for QEMU_AIO_IOCTL */
186     off_t aio_offset;
187     int aio_type;
188 } RawPosixAIOData;
189 
190 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
191 static int cdrom_reopen(BlockDriverState *bs);
192 #endif
193 
194 #if defined(__NetBSD__)
195 static int raw_normalize_devicepath(const char **filename)
196 {
197     static char namebuf[PATH_MAX];
198     const char *dp, *fname;
199     struct stat sb;
200 
201     fname = *filename;
202     dp = strrchr(fname, '/');
203     if (lstat(fname, &sb) < 0) {
204         fprintf(stderr, "%s: stat failed: %s\n",
205             fname, strerror(errno));
206         return -errno;
207     }
208 
209     if (!S_ISBLK(sb.st_mode)) {
210         return 0;
211     }
212 
213     if (dp == NULL) {
214         snprintf(namebuf, PATH_MAX, "r%s", fname);
215     } else {
216         snprintf(namebuf, PATH_MAX, "%.*s/r%s",
217             (int)(dp - fname), fname, dp + 1);
218     }
219     fprintf(stderr, "%s is a block device", fname);
220     *filename = namebuf;
221     fprintf(stderr, ", using %s\n", *filename);
222 
223     return 0;
224 }
225 #else
226 static int raw_normalize_devicepath(const char **filename)
227 {
228     return 0;
229 }
230 #endif
231 
232 /*
233  * Get logical block size via ioctl. On success store it in @sector_size_p.
234  */
235 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
236 {
237     unsigned int sector_size;
238     bool success = false;
239     int i;
240 
241     errno = ENOTSUP;
242     static const unsigned long ioctl_list[] = {
243 #ifdef BLKSSZGET
244         BLKSSZGET,
245 #endif
246 #ifdef DKIOCGETBLOCKSIZE
247         DKIOCGETBLOCKSIZE,
248 #endif
249 #ifdef DIOCGSECTORSIZE
250         DIOCGSECTORSIZE,
251 #endif
252     };
253 
254     /* Try a few ioctls to get the right size */
255     for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) {
256         if (ioctl(fd, ioctl_list[i], &sector_size) >= 0) {
257             *sector_size_p = sector_size;
258             success = true;
259         }
260     }
261 
262     return success ? 0 : -errno;
263 }
264 
265 /**
266  * Get physical block size of @fd.
267  * On success, store it in @blk_size and return 0.
268  * On failure, return -errno.
269  */
270 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
271 {
272 #ifdef BLKPBSZGET
273     if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
274         return -errno;
275     }
276     return 0;
277 #else
278     return -ENOTSUP;
279 #endif
280 }
281 
282 /* Check if read is allowed with given memory buffer and length.
283  *
284  * This function is used to check O_DIRECT memory buffer and request alignment.
285  */
286 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
287 {
288     ssize_t ret = pread(fd, buf, len, 0);
289 
290     if (ret >= 0) {
291         return true;
292     }
293 
294 #ifdef __linux__
295     /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads.  Ignore
296      * other errors (e.g. real I/O error), which could happen on a failed
297      * drive, since we only care about probing alignment.
298      */
299     if (errno != EINVAL) {
300         return true;
301     }
302 #endif
303 
304     return false;
305 }
306 
307 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
308 {
309     BDRVRawState *s = bs->opaque;
310     char *buf;
311     size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize());
312 
313     /* For SCSI generic devices the alignment is not really used.
314        With buffered I/O, we don't have any restrictions. */
315     if (bdrv_is_sg(bs) || !s->needs_alignment) {
316         bs->bl.request_alignment = 1;
317         s->buf_align = 1;
318         return;
319     }
320 
321     bs->bl.request_alignment = 0;
322     s->buf_align = 0;
323     /* Let's try to use the logical blocksize for the alignment. */
324     if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) {
325         bs->bl.request_alignment = 0;
326     }
327 #ifdef CONFIG_XFS
328     if (s->is_xfs) {
329         struct dioattr da;
330         if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
331             bs->bl.request_alignment = da.d_miniosz;
332             /* The kernel returns wrong information for d_mem */
333             /* s->buf_align = da.d_mem; */
334         }
335     }
336 #endif
337 
338     /* If we could not get the sizes so far, we can only guess them */
339     if (!s->buf_align) {
340         size_t align;
341         buf = qemu_memalign(max_align, 2 * max_align);
342         for (align = 512; align <= max_align; align <<= 1) {
343             if (raw_is_io_aligned(fd, buf + align, max_align)) {
344                 s->buf_align = align;
345                 break;
346             }
347         }
348         qemu_vfree(buf);
349     }
350 
351     if (!bs->bl.request_alignment) {
352         size_t align;
353         buf = qemu_memalign(s->buf_align, max_align);
354         for (align = 512; align <= max_align; align <<= 1) {
355             if (raw_is_io_aligned(fd, buf, align)) {
356                 bs->bl.request_alignment = align;
357                 break;
358             }
359         }
360         qemu_vfree(buf);
361     }
362 
363     if (!s->buf_align || !bs->bl.request_alignment) {
364         error_setg(errp, "Could not find working O_DIRECT alignment");
365         error_append_hint(errp, "Try cache.direct=off\n");
366     }
367 }
368 
369 static void raw_parse_flags(int bdrv_flags, int *open_flags)
370 {
371     assert(open_flags != NULL);
372 
373     *open_flags |= O_BINARY;
374     *open_flags &= ~O_ACCMODE;
375     if (bdrv_flags & BDRV_O_RDWR) {
376         *open_flags |= O_RDWR;
377     } else {
378         *open_flags |= O_RDONLY;
379     }
380 
381     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
382      * and O_DIRECT for no caching. */
383     if ((bdrv_flags & BDRV_O_NOCACHE)) {
384         *open_flags |= O_DIRECT;
385     }
386 }
387 
388 static void raw_parse_filename(const char *filename, QDict *options,
389                                Error **errp)
390 {
391     bdrv_parse_filename_strip_prefix(filename, "file:", options);
392 }
393 
394 static QemuOptsList raw_runtime_opts = {
395     .name = "raw",
396     .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
397     .desc = {
398         {
399             .name = "filename",
400             .type = QEMU_OPT_STRING,
401             .help = "File name of the image",
402         },
403         {
404             .name = "aio",
405             .type = QEMU_OPT_STRING,
406             .help = "host AIO implementation (threads, native)",
407         },
408         {
409             .name = "locking",
410             .type = QEMU_OPT_STRING,
411             .help = "file locking mode (on/off/auto, default: auto)",
412         },
413         {
414             .name = "pr-manager",
415             .type = QEMU_OPT_STRING,
416             .help = "id of persistent reservation manager object (default: none)",
417         },
418         { /* end of list */ }
419     },
420 };
421 
422 static int raw_open_common(BlockDriverState *bs, QDict *options,
423                            int bdrv_flags, int open_flags, Error **errp)
424 {
425     BDRVRawState *s = bs->opaque;
426     QemuOpts *opts;
427     Error *local_err = NULL;
428     const char *filename = NULL;
429     const char *str;
430     BlockdevAioOptions aio, aio_default;
431     int fd, ret;
432     struct stat st;
433     OnOffAuto locking;
434 
435     opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
436     qemu_opts_absorb_qdict(opts, options, &local_err);
437     if (local_err) {
438         error_propagate(errp, local_err);
439         ret = -EINVAL;
440         goto fail;
441     }
442 
443     filename = qemu_opt_get(opts, "filename");
444 
445     ret = raw_normalize_devicepath(&filename);
446     if (ret != 0) {
447         error_setg_errno(errp, -ret, "Could not normalize device path");
448         goto fail;
449     }
450 
451     aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO)
452                   ? BLOCKDEV_AIO_OPTIONS_NATIVE
453                   : BLOCKDEV_AIO_OPTIONS_THREADS;
454     aio = qapi_enum_parse(&BlockdevAioOptions_lookup,
455                           qemu_opt_get(opts, "aio"),
456                           aio_default, &local_err);
457     if (local_err) {
458         error_propagate(errp, local_err);
459         ret = -EINVAL;
460         goto fail;
461     }
462     s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE);
463 
464     locking = qapi_enum_parse(&OnOffAuto_lookup,
465                               qemu_opt_get(opts, "locking"),
466                               ON_OFF_AUTO_AUTO, &local_err);
467     if (local_err) {
468         error_propagate(errp, local_err);
469         ret = -EINVAL;
470         goto fail;
471     }
472     switch (locking) {
473     case ON_OFF_AUTO_ON:
474         s->use_lock = true;
475         if (!qemu_has_ofd_lock()) {
476             fprintf(stderr,
477                     "File lock requested but OFD locking syscall is "
478                     "unavailable, falling back to POSIX file locks.\n"
479                     "Due to the implementation, locks can be lost "
480                     "unexpectedly.\n");
481         }
482         break;
483     case ON_OFF_AUTO_OFF:
484         s->use_lock = false;
485         break;
486     case ON_OFF_AUTO_AUTO:
487         s->use_lock = qemu_has_ofd_lock();
488         break;
489     default:
490         abort();
491     }
492 
493     str = qemu_opt_get(opts, "pr-manager");
494     if (str) {
495         s->pr_mgr = pr_manager_lookup(str, &local_err);
496         if (local_err) {
497             error_propagate(errp, local_err);
498             ret = -EINVAL;
499             goto fail;
500         }
501     }
502 
503     s->open_flags = open_flags;
504     raw_parse_flags(bdrv_flags, &s->open_flags);
505 
506     s->fd = -1;
507     fd = qemu_open(filename, s->open_flags, 0644);
508     if (fd < 0) {
509         ret = -errno;
510         error_setg_errno(errp, errno, "Could not open '%s'", filename);
511         if (ret == -EROFS) {
512             ret = -EACCES;
513         }
514         goto fail;
515     }
516     s->fd = fd;
517 
518     s->lock_fd = -1;
519     if (s->use_lock) {
520         fd = qemu_open(filename, s->open_flags);
521         if (fd < 0) {
522             ret = -errno;
523             error_setg_errno(errp, errno, "Could not open '%s' for locking",
524                              filename);
525             qemu_close(s->fd);
526             goto fail;
527         }
528         s->lock_fd = fd;
529     }
530     s->perm = 0;
531     s->shared_perm = BLK_PERM_ALL;
532 
533 #ifdef CONFIG_LINUX_AIO
534      /* Currently Linux does AIO only for files opened with O_DIRECT */
535     if (s->use_linux_aio && !(s->open_flags & O_DIRECT)) {
536         error_setg(errp, "aio=native was specified, but it requires "
537                          "cache.direct=on, which was not specified.");
538         ret = -EINVAL;
539         goto fail;
540     }
541 #else
542     if (s->use_linux_aio) {
543         error_setg(errp, "aio=native was specified, but is not supported "
544                          "in this build.");
545         ret = -EINVAL;
546         goto fail;
547     }
548 #endif /* !defined(CONFIG_LINUX_AIO) */
549 
550     s->has_discard = true;
551     s->has_write_zeroes = true;
552     if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
553         s->needs_alignment = true;
554     }
555 
556     if (fstat(s->fd, &st) < 0) {
557         ret = -errno;
558         error_setg_errno(errp, errno, "Could not stat file");
559         goto fail;
560     }
561     if (S_ISREG(st.st_mode)) {
562         s->discard_zeroes = true;
563         s->has_fallocate = true;
564     }
565     if (S_ISBLK(st.st_mode)) {
566 #ifdef BLKDISCARDZEROES
567         unsigned int arg;
568         if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
569             s->discard_zeroes = true;
570         }
571 #endif
572 #ifdef __linux__
573         /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache.  Do
574          * not rely on the contents of discarded blocks unless using O_DIRECT.
575          * Same for BLKZEROOUT.
576          */
577         if (!(bs->open_flags & BDRV_O_NOCACHE)) {
578             s->discard_zeroes = false;
579             s->has_write_zeroes = false;
580         }
581 #endif
582     }
583 #ifdef __FreeBSD__
584     if (S_ISCHR(st.st_mode)) {
585         /*
586          * The file is a char device (disk), which on FreeBSD isn't behind
587          * a pager, so force all requests to be aligned. This is needed
588          * so QEMU makes sure all IO operations on the device are aligned
589          * to sector size, or else FreeBSD will reject them with EINVAL.
590          */
591         s->needs_alignment = true;
592     }
593 #endif
594 
595 #ifdef CONFIG_XFS
596     if (platform_test_xfs_fd(s->fd)) {
597         s->is_xfs = true;
598     }
599 #endif
600 
601     bs->supported_zero_flags = s->discard_zeroes ? BDRV_REQ_MAY_UNMAP : 0;
602     ret = 0;
603 fail:
604     if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
605         unlink(filename);
606     }
607     qemu_opts_del(opts);
608     return ret;
609 }
610 
611 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
612                     Error **errp)
613 {
614     BDRVRawState *s = bs->opaque;
615 
616     s->type = FTYPE_FILE;
617     return raw_open_common(bs, options, flags, 0, errp);
618 }
619 
620 typedef enum {
621     RAW_PL_PREPARE,
622     RAW_PL_COMMIT,
623     RAW_PL_ABORT,
624 } RawPermLockOp;
625 
626 #define PERM_FOREACH(i) \
627     for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
628 
629 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
630  * file; if @unlock == true, also unlock the unneeded bytes.
631  * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
632  */
633 static int raw_apply_lock_bytes(BDRVRawState *s,
634                                 uint64_t perm_lock_bits,
635                                 uint64_t shared_perm_lock_bits,
636                                 bool unlock, Error **errp)
637 {
638     int ret;
639     int i;
640 
641     PERM_FOREACH(i) {
642         int off = RAW_LOCK_PERM_BASE + i;
643         if (perm_lock_bits & (1ULL << i)) {
644             ret = qemu_lock_fd(s->lock_fd, off, 1, false);
645             if (ret) {
646                 error_setg(errp, "Failed to lock byte %d", off);
647                 return ret;
648             }
649         } else if (unlock) {
650             ret = qemu_unlock_fd(s->lock_fd, off, 1);
651             if (ret) {
652                 error_setg(errp, "Failed to unlock byte %d", off);
653                 return ret;
654             }
655         }
656     }
657     PERM_FOREACH(i) {
658         int off = RAW_LOCK_SHARED_BASE + i;
659         if (shared_perm_lock_bits & (1ULL << i)) {
660             ret = qemu_lock_fd(s->lock_fd, off, 1, false);
661             if (ret) {
662                 error_setg(errp, "Failed to lock byte %d", off);
663                 return ret;
664             }
665         } else if (unlock) {
666             ret = qemu_unlock_fd(s->lock_fd, off, 1);
667             if (ret) {
668                 error_setg(errp, "Failed to unlock byte %d", off);
669                 return ret;
670             }
671         }
672     }
673     return 0;
674 }
675 
676 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
677 static int raw_check_lock_bytes(BDRVRawState *s,
678                                 uint64_t perm, uint64_t shared_perm,
679                                 Error **errp)
680 {
681     int ret;
682     int i;
683 
684     PERM_FOREACH(i) {
685         int off = RAW_LOCK_SHARED_BASE + i;
686         uint64_t p = 1ULL << i;
687         if (perm & p) {
688             ret = qemu_lock_fd_test(s->lock_fd, off, 1, true);
689             if (ret) {
690                 char *perm_name = bdrv_perm_names(p);
691                 error_setg(errp,
692                            "Failed to get \"%s\" lock",
693                            perm_name);
694                 g_free(perm_name);
695                 error_append_hint(errp,
696                                   "Is another process using the image?\n");
697                 return ret;
698             }
699         }
700     }
701     PERM_FOREACH(i) {
702         int off = RAW_LOCK_PERM_BASE + i;
703         uint64_t p = 1ULL << i;
704         if (!(shared_perm & p)) {
705             ret = qemu_lock_fd_test(s->lock_fd, off, 1, true);
706             if (ret) {
707                 char *perm_name = bdrv_perm_names(p);
708                 error_setg(errp,
709                            "Failed to get shared \"%s\" lock",
710                            perm_name);
711                 g_free(perm_name);
712                 error_append_hint(errp,
713                                   "Is another process using the image?\n");
714                 return ret;
715             }
716         }
717     }
718     return 0;
719 }
720 
721 static int raw_handle_perm_lock(BlockDriverState *bs,
722                                 RawPermLockOp op,
723                                 uint64_t new_perm, uint64_t new_shared,
724                                 Error **errp)
725 {
726     BDRVRawState *s = bs->opaque;
727     int ret = 0;
728     Error *local_err = NULL;
729 
730     if (!s->use_lock) {
731         return 0;
732     }
733 
734     if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) {
735         return 0;
736     }
737 
738     assert(s->lock_fd > 0);
739 
740     switch (op) {
741     case RAW_PL_PREPARE:
742         ret = raw_apply_lock_bytes(s, s->perm | new_perm,
743                                    ~s->shared_perm | ~new_shared,
744                                    false, errp);
745         if (!ret) {
746             ret = raw_check_lock_bytes(s, new_perm, new_shared, errp);
747             if (!ret) {
748                 return 0;
749             }
750         }
751         op = RAW_PL_ABORT;
752         /* fall through to unlock bytes. */
753     case RAW_PL_ABORT:
754         raw_apply_lock_bytes(s, s->perm, ~s->shared_perm, true, &local_err);
755         if (local_err) {
756             /* Theoretically the above call only unlocks bytes and it cannot
757              * fail. Something weird happened, report it.
758              */
759             error_report_err(local_err);
760         }
761         break;
762     case RAW_PL_COMMIT:
763         raw_apply_lock_bytes(s, new_perm, ~new_shared, true, &local_err);
764         if (local_err) {
765             /* Theoretically the above call only unlocks bytes and it cannot
766              * fail. Something weird happened, report it.
767              */
768             error_report_err(local_err);
769         }
770         break;
771     }
772     return ret;
773 }
774 
775 static int raw_reopen_prepare(BDRVReopenState *state,
776                               BlockReopenQueue *queue, Error **errp)
777 {
778     BDRVRawState *s;
779     BDRVRawReopenState *rs;
780     int ret = 0;
781     Error *local_err = NULL;
782 
783     assert(state != NULL);
784     assert(state->bs != NULL);
785 
786     s = state->bs->opaque;
787 
788     state->opaque = g_new0(BDRVRawReopenState, 1);
789     rs = state->opaque;
790 
791     if (s->type == FTYPE_CD) {
792         rs->open_flags |= O_NONBLOCK;
793     }
794 
795     raw_parse_flags(state->flags, &rs->open_flags);
796 
797     rs->fd = -1;
798 
799     int fcntl_flags = O_APPEND | O_NONBLOCK;
800 #ifdef O_NOATIME
801     fcntl_flags |= O_NOATIME;
802 #endif
803 
804 #ifdef O_ASYNC
805     /* Not all operating systems have O_ASYNC, and those that don't
806      * will not let us track the state into rs->open_flags (typically
807      * you achieve the same effect with an ioctl, for example I_SETSIG
808      * on Solaris). But we do not use O_ASYNC, so that's fine.
809      */
810     assert((s->open_flags & O_ASYNC) == 0);
811 #endif
812 
813     if ((rs->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
814         /* dup the original fd */
815         rs->fd = qemu_dup(s->fd);
816         if (rs->fd >= 0) {
817             ret = fcntl_setfl(rs->fd, rs->open_flags);
818             if (ret) {
819                 qemu_close(rs->fd);
820                 rs->fd = -1;
821             }
822         }
823     }
824 
825     /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
826     if (rs->fd == -1) {
827         const char *normalized_filename = state->bs->filename;
828         ret = raw_normalize_devicepath(&normalized_filename);
829         if (ret < 0) {
830             error_setg_errno(errp, -ret, "Could not normalize device path");
831         } else {
832             assert(!(rs->open_flags & O_CREAT));
833             rs->fd = qemu_open(normalized_filename, rs->open_flags);
834             if (rs->fd == -1) {
835                 error_setg_errno(errp, errno, "Could not reopen file");
836                 ret = -1;
837             }
838         }
839     }
840 
841     /* Fail already reopen_prepare() if we can't get a working O_DIRECT
842      * alignment with the new fd. */
843     if (rs->fd != -1) {
844         raw_probe_alignment(state->bs, rs->fd, &local_err);
845         if (local_err) {
846             qemu_close(rs->fd);
847             rs->fd = -1;
848             error_propagate(errp, local_err);
849             ret = -EINVAL;
850         }
851     }
852 
853     return ret;
854 }
855 
856 static void raw_reopen_commit(BDRVReopenState *state)
857 {
858     BDRVRawReopenState *rs = state->opaque;
859     BDRVRawState *s = state->bs->opaque;
860 
861     s->open_flags = rs->open_flags;
862 
863     qemu_close(s->fd);
864     s->fd = rs->fd;
865 
866     g_free(state->opaque);
867     state->opaque = NULL;
868 }
869 
870 
871 static void raw_reopen_abort(BDRVReopenState *state)
872 {
873     BDRVRawReopenState *rs = state->opaque;
874 
875      /* nothing to do if NULL, we didn't get far enough */
876     if (rs == NULL) {
877         return;
878     }
879 
880     if (rs->fd >= 0) {
881         qemu_close(rs->fd);
882         rs->fd = -1;
883     }
884     g_free(state->opaque);
885     state->opaque = NULL;
886 }
887 
888 static int hdev_get_max_transfer_length(BlockDriverState *bs, int fd)
889 {
890 #ifdef BLKSECTGET
891     int max_bytes = 0;
892     short max_sectors = 0;
893     if (bs->sg && ioctl(fd, BLKSECTGET, &max_bytes) == 0) {
894         return max_bytes;
895     } else if (!bs->sg && ioctl(fd, BLKSECTGET, &max_sectors) == 0) {
896         return max_sectors << BDRV_SECTOR_BITS;
897     } else {
898         return -errno;
899     }
900 #else
901     return -ENOSYS;
902 #endif
903 }
904 
905 static int hdev_get_max_segments(const struct stat *st)
906 {
907 #ifdef CONFIG_LINUX
908     char buf[32];
909     const char *end;
910     char *sysfspath;
911     int ret;
912     int fd = -1;
913     long max_segments;
914 
915     sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
916                                 major(st->st_rdev), minor(st->st_rdev));
917     fd = open(sysfspath, O_RDONLY);
918     if (fd == -1) {
919         ret = -errno;
920         goto out;
921     }
922     do {
923         ret = read(fd, buf, sizeof(buf) - 1);
924     } while (ret == -1 && errno == EINTR);
925     if (ret < 0) {
926         ret = -errno;
927         goto out;
928     } else if (ret == 0) {
929         ret = -EIO;
930         goto out;
931     }
932     buf[ret] = 0;
933     /* The file is ended with '\n', pass 'end' to accept that. */
934     ret = qemu_strtol(buf, &end, 10, &max_segments);
935     if (ret == 0 && end && *end == '\n') {
936         ret = max_segments;
937     }
938 
939 out:
940     if (fd != -1) {
941         close(fd);
942     }
943     g_free(sysfspath);
944     return ret;
945 #else
946     return -ENOTSUP;
947 #endif
948 }
949 
950 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
951 {
952     BDRVRawState *s = bs->opaque;
953     struct stat st;
954 
955     if (!fstat(s->fd, &st)) {
956         if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode)) {
957             int ret = hdev_get_max_transfer_length(bs, s->fd);
958             if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) {
959                 bs->bl.max_transfer = pow2floor(ret);
960             }
961             ret = hdev_get_max_segments(&st);
962             if (ret > 0) {
963                 bs->bl.max_transfer = MIN(bs->bl.max_transfer,
964                                           ret * getpagesize());
965             }
966         }
967     }
968 
969     raw_probe_alignment(bs, s->fd, errp);
970     bs->bl.min_mem_alignment = s->buf_align;
971     bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
972 }
973 
974 static int check_for_dasd(int fd)
975 {
976 #ifdef BIODASDINFO2
977     struct dasd_information2_t info = {0};
978 
979     return ioctl(fd, BIODASDINFO2, &info);
980 #else
981     return -1;
982 #endif
983 }
984 
985 /**
986  * Try to get @bs's logical and physical block size.
987  * On success, store them in @bsz and return zero.
988  * On failure, return negative errno.
989  */
990 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
991 {
992     BDRVRawState *s = bs->opaque;
993     int ret;
994 
995     /* If DASD, get blocksizes */
996     if (check_for_dasd(s->fd) < 0) {
997         return -ENOTSUP;
998     }
999     ret = probe_logical_blocksize(s->fd, &bsz->log);
1000     if (ret < 0) {
1001         return ret;
1002     }
1003     return probe_physical_blocksize(s->fd, &bsz->phys);
1004 }
1005 
1006 /**
1007  * Try to get @bs's geometry: cyls, heads, sectors.
1008  * On success, store them in @geo and return 0.
1009  * On failure return -errno.
1010  * (Allows block driver to assign default geometry values that guest sees)
1011  */
1012 #ifdef __linux__
1013 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1014 {
1015     BDRVRawState *s = bs->opaque;
1016     struct hd_geometry ioctl_geo = {0};
1017 
1018     /* If DASD, get its geometry */
1019     if (check_for_dasd(s->fd) < 0) {
1020         return -ENOTSUP;
1021     }
1022     if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
1023         return -errno;
1024     }
1025     /* HDIO_GETGEO may return success even though geo contains zeros
1026        (e.g. certain multipath setups) */
1027     if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
1028         return -ENOTSUP;
1029     }
1030     /* Do not return a geometry for partition */
1031     if (ioctl_geo.start != 0) {
1032         return -ENOTSUP;
1033     }
1034     geo->heads = ioctl_geo.heads;
1035     geo->sectors = ioctl_geo.sectors;
1036     geo->cylinders = ioctl_geo.cylinders;
1037 
1038     return 0;
1039 }
1040 #else /* __linux__ */
1041 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1042 {
1043     return -ENOTSUP;
1044 }
1045 #endif
1046 
1047 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
1048 {
1049     int ret;
1050 
1051     ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
1052     if (ret == -1) {
1053         return -errno;
1054     }
1055 
1056     return 0;
1057 }
1058 
1059 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
1060 {
1061     BDRVRawState *s = aiocb->bs->opaque;
1062     int ret;
1063 
1064     if (s->page_cache_inconsistent) {
1065         return -EIO;
1066     }
1067 
1068     ret = qemu_fdatasync(aiocb->aio_fildes);
1069     if (ret == -1) {
1070         /* There is no clear definition of the semantics of a failing fsync(),
1071          * so we may have to assume the worst. The sad truth is that this
1072          * assumption is correct for Linux. Some pages are now probably marked
1073          * clean in the page cache even though they are inconsistent with the
1074          * on-disk contents. The next fdatasync() call would succeed, but no
1075          * further writeback attempt will be made. We can't get back to a state
1076          * in which we know what is on disk (we would have to rewrite
1077          * everything that was touched since the last fdatasync() at least), so
1078          * make bdrv_flush() fail permanently. Given that the behaviour isn't
1079          * really defined, I have little hope that other OSes are doing better.
1080          *
1081          * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1082          * cache. */
1083         if ((s->open_flags & O_DIRECT) == 0) {
1084             s->page_cache_inconsistent = true;
1085         }
1086         return -errno;
1087     }
1088     return 0;
1089 }
1090 
1091 #ifdef CONFIG_PREADV
1092 
1093 static bool preadv_present = true;
1094 
1095 static ssize_t
1096 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1097 {
1098     return preadv(fd, iov, nr_iov, offset);
1099 }
1100 
1101 static ssize_t
1102 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1103 {
1104     return pwritev(fd, iov, nr_iov, offset);
1105 }
1106 
1107 #else
1108 
1109 static bool preadv_present = false;
1110 
1111 static ssize_t
1112 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1113 {
1114     return -ENOSYS;
1115 }
1116 
1117 static ssize_t
1118 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1119 {
1120     return -ENOSYS;
1121 }
1122 
1123 #endif
1124 
1125 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
1126 {
1127     ssize_t len;
1128 
1129     do {
1130         if (aiocb->aio_type & QEMU_AIO_WRITE)
1131             len = qemu_pwritev(aiocb->aio_fildes,
1132                                aiocb->aio_iov,
1133                                aiocb->aio_niov,
1134                                aiocb->aio_offset);
1135          else
1136             len = qemu_preadv(aiocb->aio_fildes,
1137                               aiocb->aio_iov,
1138                               aiocb->aio_niov,
1139                               aiocb->aio_offset);
1140     } while (len == -1 && errno == EINTR);
1141 
1142     if (len == -1) {
1143         return -errno;
1144     }
1145     return len;
1146 }
1147 
1148 /*
1149  * Read/writes the data to/from a given linear buffer.
1150  *
1151  * Returns the number of bytes handles or -errno in case of an error. Short
1152  * reads are only returned if the end of the file is reached.
1153  */
1154 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
1155 {
1156     ssize_t offset = 0;
1157     ssize_t len;
1158 
1159     while (offset < aiocb->aio_nbytes) {
1160         if (aiocb->aio_type & QEMU_AIO_WRITE) {
1161             len = pwrite(aiocb->aio_fildes,
1162                          (const char *)buf + offset,
1163                          aiocb->aio_nbytes - offset,
1164                          aiocb->aio_offset + offset);
1165         } else {
1166             len = pread(aiocb->aio_fildes,
1167                         buf + offset,
1168                         aiocb->aio_nbytes - offset,
1169                         aiocb->aio_offset + offset);
1170         }
1171         if (len == -1 && errno == EINTR) {
1172             continue;
1173         } else if (len == -1 && errno == EINVAL &&
1174                    (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
1175                    !(aiocb->aio_type & QEMU_AIO_WRITE) &&
1176                    offset > 0) {
1177             /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1178              * after a short read.  Assume that O_DIRECT short reads only occur
1179              * at EOF.  Therefore this is a short read, not an I/O error.
1180              */
1181             break;
1182         } else if (len == -1) {
1183             offset = -errno;
1184             break;
1185         } else if (len == 0) {
1186             break;
1187         }
1188         offset += len;
1189     }
1190 
1191     return offset;
1192 }
1193 
1194 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
1195 {
1196     ssize_t nbytes;
1197     char *buf;
1198 
1199     if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
1200         /*
1201          * If there is just a single buffer, and it is properly aligned
1202          * we can just use plain pread/pwrite without any problems.
1203          */
1204         if (aiocb->aio_niov == 1) {
1205              return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
1206         }
1207         /*
1208          * We have more than one iovec, and all are properly aligned.
1209          *
1210          * Try preadv/pwritev first and fall back to linearizing the
1211          * buffer if it's not supported.
1212          */
1213         if (preadv_present) {
1214             nbytes = handle_aiocb_rw_vector(aiocb);
1215             if (nbytes == aiocb->aio_nbytes ||
1216                 (nbytes < 0 && nbytes != -ENOSYS)) {
1217                 return nbytes;
1218             }
1219             preadv_present = false;
1220         }
1221 
1222         /*
1223          * XXX(hch): short read/write.  no easy way to handle the reminder
1224          * using these interfaces.  For now retry using plain
1225          * pread/pwrite?
1226          */
1227     }
1228 
1229     /*
1230      * Ok, we have to do it the hard way, copy all segments into
1231      * a single aligned buffer.
1232      */
1233     buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
1234     if (buf == NULL) {
1235         return -ENOMEM;
1236     }
1237 
1238     if (aiocb->aio_type & QEMU_AIO_WRITE) {
1239         char *p = buf;
1240         int i;
1241 
1242         for (i = 0; i < aiocb->aio_niov; ++i) {
1243             memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
1244             p += aiocb->aio_iov[i].iov_len;
1245         }
1246         assert(p - buf == aiocb->aio_nbytes);
1247     }
1248 
1249     nbytes = handle_aiocb_rw_linear(aiocb, buf);
1250     if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
1251         char *p = buf;
1252         size_t count = aiocb->aio_nbytes, copy;
1253         int i;
1254 
1255         for (i = 0; i < aiocb->aio_niov && count; ++i) {
1256             copy = count;
1257             if (copy > aiocb->aio_iov[i].iov_len) {
1258                 copy = aiocb->aio_iov[i].iov_len;
1259             }
1260             memcpy(aiocb->aio_iov[i].iov_base, p, copy);
1261             assert(count >= copy);
1262             p     += copy;
1263             count -= copy;
1264         }
1265         assert(count == 0);
1266     }
1267     qemu_vfree(buf);
1268 
1269     return nbytes;
1270 }
1271 
1272 #ifdef CONFIG_XFS
1273 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
1274 {
1275     struct xfs_flock64 fl;
1276     int err;
1277 
1278     memset(&fl, 0, sizeof(fl));
1279     fl.l_whence = SEEK_SET;
1280     fl.l_start = offset;
1281     fl.l_len = bytes;
1282 
1283     if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
1284         err = errno;
1285         DPRINTF("cannot write zero range (%s)\n", strerror(errno));
1286         return -err;
1287     }
1288 
1289     return 0;
1290 }
1291 
1292 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
1293 {
1294     struct xfs_flock64 fl;
1295     int err;
1296 
1297     memset(&fl, 0, sizeof(fl));
1298     fl.l_whence = SEEK_SET;
1299     fl.l_start = offset;
1300     fl.l_len = bytes;
1301 
1302     if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
1303         err = errno;
1304         DPRINTF("cannot punch hole (%s)\n", strerror(errno));
1305         return -err;
1306     }
1307 
1308     return 0;
1309 }
1310 #endif
1311 
1312 static int translate_err(int err)
1313 {
1314     if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1315         err == -ENOTTY) {
1316         err = -ENOTSUP;
1317     }
1318     return err;
1319 }
1320 
1321 #ifdef CONFIG_FALLOCATE
1322 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1323 {
1324     do {
1325         if (fallocate(fd, mode, offset, len) == 0) {
1326             return 0;
1327         }
1328     } while (errno == EINTR);
1329     return translate_err(-errno);
1330 }
1331 #endif
1332 
1333 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1334 {
1335     int ret = -ENOTSUP;
1336     BDRVRawState *s = aiocb->bs->opaque;
1337 
1338     if (!s->has_write_zeroes) {
1339         return -ENOTSUP;
1340     }
1341 
1342 #ifdef BLKZEROOUT
1343     do {
1344         uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1345         if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1346             return 0;
1347         }
1348     } while (errno == EINTR);
1349 
1350     ret = translate_err(-errno);
1351 #endif
1352 
1353     if (ret == -ENOTSUP) {
1354         s->has_write_zeroes = false;
1355     }
1356     return ret;
1357 }
1358 
1359 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
1360 {
1361 #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
1362     BDRVRawState *s = aiocb->bs->opaque;
1363 #endif
1364 #ifdef CONFIG_FALLOCATE
1365     int64_t len;
1366 #endif
1367 
1368     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1369         return handle_aiocb_write_zeroes_block(aiocb);
1370     }
1371 
1372 #ifdef CONFIG_XFS
1373     if (s->is_xfs) {
1374         return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
1375     }
1376 #endif
1377 
1378 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1379     if (s->has_write_zeroes) {
1380         int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1381                                aiocb->aio_offset, aiocb->aio_nbytes);
1382         if (ret == 0 || ret != -ENOTSUP) {
1383             return ret;
1384         }
1385         s->has_write_zeroes = false;
1386     }
1387 #endif
1388 
1389 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1390     if (s->has_discard && s->has_fallocate) {
1391         int ret = do_fallocate(s->fd,
1392                                FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1393                                aiocb->aio_offset, aiocb->aio_nbytes);
1394         if (ret == 0) {
1395             ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1396             if (ret == 0 || ret != -ENOTSUP) {
1397                 return ret;
1398             }
1399             s->has_fallocate = false;
1400         } else if (ret != -ENOTSUP) {
1401             return ret;
1402         } else {
1403             s->has_discard = false;
1404         }
1405     }
1406 #endif
1407 
1408 #ifdef CONFIG_FALLOCATE
1409     /* Last resort: we are trying to extend the file with zeroed data. This
1410      * can be done via fallocate(fd, 0) */
1411     len = bdrv_getlength(aiocb->bs);
1412     if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) {
1413         int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1414         if (ret == 0 || ret != -ENOTSUP) {
1415             return ret;
1416         }
1417         s->has_fallocate = false;
1418     }
1419 #endif
1420 
1421     return -ENOTSUP;
1422 }
1423 
1424 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
1425 {
1426     int ret = -EOPNOTSUPP;
1427     BDRVRawState *s = aiocb->bs->opaque;
1428 
1429     if (!s->has_discard) {
1430         return -ENOTSUP;
1431     }
1432 
1433     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1434 #ifdef BLKDISCARD
1435         do {
1436             uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1437             if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1438                 return 0;
1439             }
1440         } while (errno == EINTR);
1441 
1442         ret = -errno;
1443 #endif
1444     } else {
1445 #ifdef CONFIG_XFS
1446         if (s->is_xfs) {
1447             return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1448         }
1449 #endif
1450 
1451 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1452         ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1453                            aiocb->aio_offset, aiocb->aio_nbytes);
1454 #endif
1455     }
1456 
1457     ret = translate_err(ret);
1458     if (ret == -ENOTSUP) {
1459         s->has_discard = false;
1460     }
1461     return ret;
1462 }
1463 
1464 static int aio_worker(void *arg)
1465 {
1466     RawPosixAIOData *aiocb = arg;
1467     ssize_t ret = 0;
1468 
1469     switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
1470     case QEMU_AIO_READ:
1471         ret = handle_aiocb_rw(aiocb);
1472         if (ret >= 0 && ret < aiocb->aio_nbytes) {
1473             iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
1474                       0, aiocb->aio_nbytes - ret);
1475 
1476             ret = aiocb->aio_nbytes;
1477         }
1478         if (ret == aiocb->aio_nbytes) {
1479             ret = 0;
1480         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1481             ret = -EINVAL;
1482         }
1483         break;
1484     case QEMU_AIO_WRITE:
1485         ret = handle_aiocb_rw(aiocb);
1486         if (ret == aiocb->aio_nbytes) {
1487             ret = 0;
1488         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1489             ret = -EINVAL;
1490         }
1491         break;
1492     case QEMU_AIO_FLUSH:
1493         ret = handle_aiocb_flush(aiocb);
1494         break;
1495     case QEMU_AIO_IOCTL:
1496         ret = handle_aiocb_ioctl(aiocb);
1497         break;
1498     case QEMU_AIO_DISCARD:
1499         ret = handle_aiocb_discard(aiocb);
1500         break;
1501     case QEMU_AIO_WRITE_ZEROES:
1502         ret = handle_aiocb_write_zeroes(aiocb);
1503         break;
1504     default:
1505         fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1506         ret = -EINVAL;
1507         break;
1508     }
1509 
1510     g_free(aiocb);
1511     return ret;
1512 }
1513 
1514 static int paio_submit_co(BlockDriverState *bs, int fd,
1515                           int64_t offset, QEMUIOVector *qiov,
1516                           int bytes, 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 == bytes);
1532     }
1533 
1534     trace_paio_submit_co(offset, bytes, type);
1535     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1536     return thread_pool_submit_co(pool, aio_worker, acb);
1537 }
1538 
1539 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
1540         int64_t offset, QEMUIOVector *qiov, int bytes,
1541         BlockCompletionFunc *cb, void *opaque, int type)
1542 {
1543     RawPosixAIOData *acb = g_new(RawPosixAIOData, 1);
1544     ThreadPool *pool;
1545 
1546     acb->bs = bs;
1547     acb->aio_type = type;
1548     acb->aio_fildes = fd;
1549 
1550     acb->aio_nbytes = bytes;
1551     acb->aio_offset = offset;
1552 
1553     if (qiov) {
1554         acb->aio_iov = qiov->iov;
1555         acb->aio_niov = qiov->niov;
1556         assert(qiov->size == acb->aio_nbytes);
1557     }
1558 
1559     trace_paio_submit(acb, opaque, offset, bytes, type);
1560     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1561     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1562 }
1563 
1564 static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset,
1565                                    uint64_t bytes, QEMUIOVector *qiov, int type)
1566 {
1567     BDRVRawState *s = bs->opaque;
1568 
1569     if (fd_open(bs) < 0)
1570         return -EIO;
1571 
1572     /*
1573      * Check if the underlying device requires requests to be aligned,
1574      * and if the request we are trying to submit is aligned or not.
1575      * If this is the case tell the low-level driver that it needs
1576      * to copy the buffer.
1577      */
1578     if (s->needs_alignment) {
1579         if (!bdrv_qiov_is_aligned(bs, qiov)) {
1580             type |= QEMU_AIO_MISALIGNED;
1581 #ifdef CONFIG_LINUX_AIO
1582         } else if (s->use_linux_aio) {
1583             LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1584             assert(qiov->size == bytes);
1585             return laio_co_submit(bs, aio, s->fd, offset, qiov, type);
1586 #endif
1587         }
1588     }
1589 
1590     return paio_submit_co(bs, s->fd, offset, qiov, bytes, type);
1591 }
1592 
1593 static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset,
1594                                       uint64_t bytes, QEMUIOVector *qiov,
1595                                       int flags)
1596 {
1597     return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ);
1598 }
1599 
1600 static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
1601                                        uint64_t bytes, QEMUIOVector *qiov,
1602                                        int flags)
1603 {
1604     assert(flags == 0);
1605     return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE);
1606 }
1607 
1608 static void raw_aio_plug(BlockDriverState *bs)
1609 {
1610 #ifdef CONFIG_LINUX_AIO
1611     BDRVRawState *s = bs->opaque;
1612     if (s->use_linux_aio) {
1613         LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1614         laio_io_plug(bs, aio);
1615     }
1616 #endif
1617 }
1618 
1619 static void raw_aio_unplug(BlockDriverState *bs)
1620 {
1621 #ifdef CONFIG_LINUX_AIO
1622     BDRVRawState *s = bs->opaque;
1623     if (s->use_linux_aio) {
1624         LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1625         laio_io_unplug(bs, aio);
1626     }
1627 #endif
1628 }
1629 
1630 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
1631         BlockCompletionFunc *cb, void *opaque)
1632 {
1633     BDRVRawState *s = bs->opaque;
1634 
1635     if (fd_open(bs) < 0)
1636         return NULL;
1637 
1638     return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1639 }
1640 
1641 static void raw_close(BlockDriverState *bs)
1642 {
1643     BDRVRawState *s = bs->opaque;
1644 
1645     if (s->fd >= 0) {
1646         qemu_close(s->fd);
1647         s->fd = -1;
1648     }
1649     if (s->lock_fd >= 0) {
1650         qemu_close(s->lock_fd);
1651         s->lock_fd = -1;
1652     }
1653 }
1654 
1655 /**
1656  * Truncates the given regular file @fd to @offset and, when growing, fills the
1657  * new space according to @prealloc.
1658  *
1659  * Returns: 0 on success, -errno on failure.
1660  */
1661 static int raw_regular_truncate(int fd, int64_t offset, PreallocMode prealloc,
1662                                 Error **errp)
1663 {
1664     int result = 0;
1665     int64_t current_length = 0;
1666     char *buf = NULL;
1667     struct stat st;
1668 
1669     if (fstat(fd, &st) < 0) {
1670         result = -errno;
1671         error_setg_errno(errp, -result, "Could not stat file");
1672         return result;
1673     }
1674 
1675     current_length = st.st_size;
1676     if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1677         error_setg(errp, "Cannot use preallocation for shrinking files");
1678         return -ENOTSUP;
1679     }
1680 
1681     switch (prealloc) {
1682 #ifdef CONFIG_POSIX_FALLOCATE
1683     case PREALLOC_MODE_FALLOC:
1684         /*
1685          * Truncating before posix_fallocate() makes it about twice slower on
1686          * file systems that do not support fallocate(), trying to check if a
1687          * block is allocated before allocating it, so don't do that here.
1688          */
1689         result = -posix_fallocate(fd, current_length, offset - current_length);
1690         if (result != 0) {
1691             /* posix_fallocate() doesn't set errno. */
1692             error_setg_errno(errp, -result,
1693                              "Could not preallocate new data");
1694         }
1695         goto out;
1696 #endif
1697     case PREALLOC_MODE_FULL:
1698     {
1699         int64_t num = 0, left = offset - current_length;
1700 
1701         /*
1702          * Knowing the final size from the beginning could allow the file
1703          * system driver to do less allocations and possibly avoid
1704          * fragmentation of the file.
1705          */
1706         if (ftruncate(fd, offset) != 0) {
1707             result = -errno;
1708             error_setg_errno(errp, -result, "Could not resize file");
1709             goto out;
1710         }
1711 
1712         buf = g_malloc0(65536);
1713 
1714         result = lseek(fd, current_length, SEEK_SET);
1715         if (result < 0) {
1716             result = -errno;
1717             error_setg_errno(errp, -result,
1718                              "Failed to seek to the old end of file");
1719             goto out;
1720         }
1721 
1722         while (left > 0) {
1723             num = MIN(left, 65536);
1724             result = write(fd, buf, num);
1725             if (result < 0) {
1726                 result = -errno;
1727                 error_setg_errno(errp, -result,
1728                                  "Could not write zeros for preallocation");
1729                 goto out;
1730             }
1731             left -= result;
1732         }
1733         if (result >= 0) {
1734             result = fsync(fd);
1735             if (result < 0) {
1736                 result = -errno;
1737                 error_setg_errno(errp, -result,
1738                                  "Could not flush file to disk");
1739                 goto out;
1740             }
1741         }
1742         goto out;
1743     }
1744     case PREALLOC_MODE_OFF:
1745         if (ftruncate(fd, offset) != 0) {
1746             result = -errno;
1747             error_setg_errno(errp, -result, "Could not resize file");
1748         }
1749         return result;
1750     default:
1751         result = -ENOTSUP;
1752         error_setg(errp, "Unsupported preallocation mode: %s",
1753                    PreallocMode_str(prealloc));
1754         return result;
1755     }
1756 
1757 out:
1758     if (result < 0) {
1759         if (ftruncate(fd, current_length) < 0) {
1760             error_report("Failed to restore old file length: %s",
1761                          strerror(errno));
1762         }
1763     }
1764 
1765     g_free(buf);
1766     return result;
1767 }
1768 
1769 static int raw_truncate(BlockDriverState *bs, int64_t offset,
1770                         PreallocMode prealloc, Error **errp)
1771 {
1772     BDRVRawState *s = bs->opaque;
1773     struct stat st;
1774     int ret;
1775 
1776     if (fstat(s->fd, &st)) {
1777         ret = -errno;
1778         error_setg_errno(errp, -ret, "Failed to fstat() the file");
1779         return ret;
1780     }
1781 
1782     if (S_ISREG(st.st_mode)) {
1783         return raw_regular_truncate(s->fd, offset, prealloc, errp);
1784     }
1785 
1786     if (prealloc != PREALLOC_MODE_OFF) {
1787         error_setg(errp, "Preallocation mode '%s' unsupported for this "
1788                    "non-regular file", PreallocMode_str(prealloc));
1789         return -ENOTSUP;
1790     }
1791 
1792     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1793         if (offset > raw_getlength(bs)) {
1794             error_setg(errp, "Cannot grow device files");
1795             return -EINVAL;
1796         }
1797     } else {
1798         error_setg(errp, "Resizing this file is not supported");
1799         return -ENOTSUP;
1800     }
1801 
1802     return 0;
1803 }
1804 
1805 #ifdef __OpenBSD__
1806 static int64_t raw_getlength(BlockDriverState *bs)
1807 {
1808     BDRVRawState *s = bs->opaque;
1809     int fd = s->fd;
1810     struct stat st;
1811 
1812     if (fstat(fd, &st))
1813         return -errno;
1814     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1815         struct disklabel dl;
1816 
1817         if (ioctl(fd, DIOCGDINFO, &dl))
1818             return -errno;
1819         return (uint64_t)dl.d_secsize *
1820             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1821     } else
1822         return st.st_size;
1823 }
1824 #elif defined(__NetBSD__)
1825 static int64_t raw_getlength(BlockDriverState *bs)
1826 {
1827     BDRVRawState *s = bs->opaque;
1828     int fd = s->fd;
1829     struct stat st;
1830 
1831     if (fstat(fd, &st))
1832         return -errno;
1833     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1834         struct dkwedge_info dkw;
1835 
1836         if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1837             return dkw.dkw_size * 512;
1838         } else {
1839             struct disklabel dl;
1840 
1841             if (ioctl(fd, DIOCGDINFO, &dl))
1842                 return -errno;
1843             return (uint64_t)dl.d_secsize *
1844                 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1845         }
1846     } else
1847         return st.st_size;
1848 }
1849 #elif defined(__sun__)
1850 static int64_t raw_getlength(BlockDriverState *bs)
1851 {
1852     BDRVRawState *s = bs->opaque;
1853     struct dk_minfo minfo;
1854     int ret;
1855     int64_t size;
1856 
1857     ret = fd_open(bs);
1858     if (ret < 0) {
1859         return ret;
1860     }
1861 
1862     /*
1863      * Use the DKIOCGMEDIAINFO ioctl to read the size.
1864      */
1865     ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1866     if (ret != -1) {
1867         return minfo.dki_lbsize * minfo.dki_capacity;
1868     }
1869 
1870     /*
1871      * There are reports that lseek on some devices fails, but
1872      * irc discussion said that contingency on contingency was overkill.
1873      */
1874     size = lseek(s->fd, 0, SEEK_END);
1875     if (size < 0) {
1876         return -errno;
1877     }
1878     return size;
1879 }
1880 #elif defined(CONFIG_BSD)
1881 static int64_t raw_getlength(BlockDriverState *bs)
1882 {
1883     BDRVRawState *s = bs->opaque;
1884     int fd = s->fd;
1885     int64_t size;
1886     struct stat sb;
1887 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1888     int reopened = 0;
1889 #endif
1890     int ret;
1891 
1892     ret = fd_open(bs);
1893     if (ret < 0)
1894         return ret;
1895 
1896 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1897 again:
1898 #endif
1899     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1900 #ifdef DIOCGMEDIASIZE
1901 	if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1902 #elif defined(DIOCGPART)
1903         {
1904                 struct partinfo pi;
1905                 if (ioctl(fd, DIOCGPART, &pi) == 0)
1906                         size = pi.media_size;
1907                 else
1908                         size = 0;
1909         }
1910         if (size == 0)
1911 #endif
1912 #if defined(__APPLE__) && defined(__MACH__)
1913         {
1914             uint64_t sectors = 0;
1915             uint32_t sector_size = 0;
1916 
1917             if (ioctl(fd, DKIOCGETBLOCKCOUNT, &sectors) == 0
1918                && ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) == 0) {
1919                 size = sectors * sector_size;
1920             } else {
1921                 size = lseek(fd, 0LL, SEEK_END);
1922                 if (size < 0) {
1923                     return -errno;
1924                 }
1925             }
1926         }
1927 #else
1928         size = lseek(fd, 0LL, SEEK_END);
1929         if (size < 0) {
1930             return -errno;
1931         }
1932 #endif
1933 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1934         switch(s->type) {
1935         case FTYPE_CD:
1936             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1937             if (size == 2048LL * (unsigned)-1)
1938                 size = 0;
1939             /* XXX no disc?  maybe we need to reopen... */
1940             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1941                 reopened = 1;
1942                 goto again;
1943             }
1944         }
1945 #endif
1946     } else {
1947         size = lseek(fd, 0, SEEK_END);
1948         if (size < 0) {
1949             return -errno;
1950         }
1951     }
1952     return size;
1953 }
1954 #else
1955 static int64_t raw_getlength(BlockDriverState *bs)
1956 {
1957     BDRVRawState *s = bs->opaque;
1958     int ret;
1959     int64_t size;
1960 
1961     ret = fd_open(bs);
1962     if (ret < 0) {
1963         return ret;
1964     }
1965 
1966     size = lseek(s->fd, 0, SEEK_END);
1967     if (size < 0) {
1968         return -errno;
1969     }
1970     return size;
1971 }
1972 #endif
1973 
1974 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1975 {
1976     struct stat st;
1977     BDRVRawState *s = bs->opaque;
1978 
1979     if (fstat(s->fd, &st) < 0) {
1980         return -errno;
1981     }
1982     return (int64_t)st.st_blocks * 512;
1983 }
1984 
1985 static int coroutine_fn raw_co_create_opts(const char *filename, QemuOpts *opts,
1986                                            Error **errp)
1987 {
1988     int fd;
1989     int result = 0;
1990     int64_t total_size = 0;
1991     bool nocow = false;
1992     PreallocMode prealloc;
1993     char *buf = NULL;
1994     Error *local_err = NULL;
1995 
1996     strstart(filename, "file:", &filename);
1997 
1998     /* Read out options */
1999     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2000                           BDRV_SECTOR_SIZE);
2001     nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
2002     buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2003     prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
2004                                PREALLOC_MODE_OFF, &local_err);
2005     g_free(buf);
2006     if (local_err) {
2007         error_propagate(errp, local_err);
2008         result = -EINVAL;
2009         goto out;
2010     }
2011 
2012     fd = qemu_open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY,
2013                    0644);
2014     if (fd < 0) {
2015         result = -errno;
2016         error_setg_errno(errp, -result, "Could not create file");
2017         goto out;
2018     }
2019 
2020     if (nocow) {
2021 #ifdef __linux__
2022         /* Set NOCOW flag to solve performance issue on fs like btrfs.
2023          * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2024          * will be ignored since any failure of this operation should not
2025          * block the left work.
2026          */
2027         int attr;
2028         if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
2029             attr |= FS_NOCOW_FL;
2030             ioctl(fd, FS_IOC_SETFLAGS, &attr);
2031         }
2032 #endif
2033     }
2034 
2035     result = raw_regular_truncate(fd, total_size, prealloc, errp);
2036     if (result < 0) {
2037         goto out_close;
2038     }
2039 
2040 out_close:
2041     if (qemu_close(fd) != 0 && result == 0) {
2042         result = -errno;
2043         error_setg_errno(errp, -result, "Could not close the new file");
2044     }
2045 out:
2046     return result;
2047 }
2048 
2049 /*
2050  * Find allocation range in @bs around offset @start.
2051  * May change underlying file descriptor's file offset.
2052  * If @start is not in a hole, store @start in @data, and the
2053  * beginning of the next hole in @hole, and return 0.
2054  * If @start is in a non-trailing hole, store @start in @hole and the
2055  * beginning of the next non-hole in @data, and return 0.
2056  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2057  * If we can't find out, return a negative errno other than -ENXIO.
2058  */
2059 static int find_allocation(BlockDriverState *bs, off_t start,
2060                            off_t *data, off_t *hole)
2061 {
2062 #if defined SEEK_HOLE && defined SEEK_DATA
2063     BDRVRawState *s = bs->opaque;
2064     off_t offs;
2065 
2066     /*
2067      * SEEK_DATA cases:
2068      * D1. offs == start: start is in data
2069      * D2. offs > start: start is in a hole, next data at offs
2070      * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2071      *                              or start is beyond EOF
2072      *     If the latter happens, the file has been truncated behind
2073      *     our back since we opened it.  All bets are off then.
2074      *     Treating like a trailing hole is simplest.
2075      * D4. offs < 0, errno != ENXIO: we learned nothing
2076      */
2077     offs = lseek(s->fd, start, SEEK_DATA);
2078     if (offs < 0) {
2079         return -errno;          /* D3 or D4 */
2080     }
2081     assert(offs >= start);
2082 
2083     if (offs > start) {
2084         /* D2: in hole, next data at offs */
2085         *hole = start;
2086         *data = offs;
2087         return 0;
2088     }
2089 
2090     /* D1: in data, end not yet known */
2091 
2092     /*
2093      * SEEK_HOLE cases:
2094      * H1. offs == start: start is in a hole
2095      *     If this happens here, a hole has been dug behind our back
2096      *     since the previous lseek().
2097      * H2. offs > start: either start is in data, next hole at offs,
2098      *                   or start is in trailing hole, EOF at offs
2099      *     Linux treats trailing holes like any other hole: offs ==
2100      *     start.  Solaris seeks to EOF instead: offs > start (blech).
2101      *     If that happens here, a hole has been dug behind our back
2102      *     since the previous lseek().
2103      * H3. offs < 0, errno = ENXIO: start is beyond EOF
2104      *     If this happens, the file has been truncated behind our
2105      *     back since we opened it.  Treat it like a trailing hole.
2106      * H4. offs < 0, errno != ENXIO: we learned nothing
2107      *     Pretend we know nothing at all, i.e. "forget" about D1.
2108      */
2109     offs = lseek(s->fd, start, SEEK_HOLE);
2110     if (offs < 0) {
2111         return -errno;          /* D1 and (H3 or H4) */
2112     }
2113     assert(offs >= start);
2114 
2115     if (offs > start) {
2116         /*
2117          * D1 and H2: either in data, next hole at offs, or it was in
2118          * data but is now in a trailing hole.  In the latter case,
2119          * all bets are off.  Treating it as if it there was data all
2120          * the way to EOF is safe, so simply do that.
2121          */
2122         *data = start;
2123         *hole = offs;
2124         return 0;
2125     }
2126 
2127     /* D1 and H1 */
2128     return -EBUSY;
2129 #else
2130     return -ENOTSUP;
2131 #endif
2132 }
2133 
2134 /*
2135  * Returns the allocation status of the specified offset.
2136  *
2137  * The block layer guarantees 'offset' and 'bytes' are within bounds.
2138  *
2139  * 'pnum' is set to the number of bytes (including and immediately following
2140  * the specified offset) that are known to be in the same
2141  * allocated/unallocated state.
2142  *
2143  * 'bytes' is the max value 'pnum' should be set to.
2144  */
2145 static int coroutine_fn raw_co_block_status(BlockDriverState *bs,
2146                                             bool want_zero,
2147                                             int64_t offset,
2148                                             int64_t bytes, int64_t *pnum,
2149                                             int64_t *map,
2150                                             BlockDriverState **file)
2151 {
2152     off_t data = 0, hole = 0;
2153     int ret;
2154 
2155     ret = fd_open(bs);
2156     if (ret < 0) {
2157         return ret;
2158     }
2159 
2160     if (!want_zero) {
2161         *pnum = bytes;
2162         *map = offset;
2163         *file = bs;
2164         return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
2165     }
2166 
2167     ret = find_allocation(bs, offset, &data, &hole);
2168     if (ret == -ENXIO) {
2169         /* Trailing hole */
2170         *pnum = bytes;
2171         ret = BDRV_BLOCK_ZERO;
2172     } else if (ret < 0) {
2173         /* No info available, so pretend there are no holes */
2174         *pnum = bytes;
2175         ret = BDRV_BLOCK_DATA;
2176     } else if (data == offset) {
2177         /* On a data extent, compute bytes to the end of the extent,
2178          * possibly including a partial sector at EOF. */
2179         *pnum = MIN(bytes, hole - offset);
2180         ret = BDRV_BLOCK_DATA;
2181     } else {
2182         /* On a hole, compute bytes to the beginning of the next extent.  */
2183         assert(hole == offset);
2184         *pnum = MIN(bytes, data - offset);
2185         ret = BDRV_BLOCK_ZERO;
2186     }
2187     *map = offset;
2188     *file = bs;
2189     return ret | BDRV_BLOCK_OFFSET_VALID;
2190 }
2191 
2192 static coroutine_fn BlockAIOCB *raw_aio_pdiscard(BlockDriverState *bs,
2193     int64_t offset, int bytes,
2194     BlockCompletionFunc *cb, void *opaque)
2195 {
2196     BDRVRawState *s = bs->opaque;
2197 
2198     return paio_submit(bs, s->fd, offset, NULL, bytes,
2199                        cb, opaque, QEMU_AIO_DISCARD);
2200 }
2201 
2202 static int coroutine_fn raw_co_pwrite_zeroes(
2203     BlockDriverState *bs, int64_t offset,
2204     int bytes, BdrvRequestFlags flags)
2205 {
2206     BDRVRawState *s = bs->opaque;
2207 
2208     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2209         return paio_submit_co(bs, s->fd, offset, NULL, bytes,
2210                               QEMU_AIO_WRITE_ZEROES);
2211     } else if (s->discard_zeroes) {
2212         return paio_submit_co(bs, s->fd, offset, NULL, bytes,
2213                               QEMU_AIO_DISCARD);
2214     }
2215     return -ENOTSUP;
2216 }
2217 
2218 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2219 {
2220     BDRVRawState *s = bs->opaque;
2221 
2222     bdi->unallocated_blocks_are_zero = s->discard_zeroes;
2223     return 0;
2224 }
2225 
2226 static QemuOptsList raw_create_opts = {
2227     .name = "raw-create-opts",
2228     .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
2229     .desc = {
2230         {
2231             .name = BLOCK_OPT_SIZE,
2232             .type = QEMU_OPT_SIZE,
2233             .help = "Virtual disk size"
2234         },
2235         {
2236             .name = BLOCK_OPT_NOCOW,
2237             .type = QEMU_OPT_BOOL,
2238             .help = "Turn off copy-on-write (valid only on btrfs)"
2239         },
2240         {
2241             .name = BLOCK_OPT_PREALLOC,
2242             .type = QEMU_OPT_STRING,
2243             .help = "Preallocation mode (allowed values: off, falloc, full)"
2244         },
2245         { /* end of list */ }
2246     }
2247 };
2248 
2249 static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared,
2250                           Error **errp)
2251 {
2252     return raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp);
2253 }
2254 
2255 static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared)
2256 {
2257     BDRVRawState *s = bs->opaque;
2258     raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL);
2259     s->perm = perm;
2260     s->shared_perm = shared;
2261 }
2262 
2263 static void raw_abort_perm_update(BlockDriverState *bs)
2264 {
2265     raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
2266 }
2267 
2268 BlockDriver bdrv_file = {
2269     .format_name = "file",
2270     .protocol_name = "file",
2271     .instance_size = sizeof(BDRVRawState),
2272     .bdrv_needs_filename = true,
2273     .bdrv_probe = NULL, /* no probe for protocols */
2274     .bdrv_parse_filename = raw_parse_filename,
2275     .bdrv_file_open = raw_open,
2276     .bdrv_reopen_prepare = raw_reopen_prepare,
2277     .bdrv_reopen_commit = raw_reopen_commit,
2278     .bdrv_reopen_abort = raw_reopen_abort,
2279     .bdrv_close = raw_close,
2280     .bdrv_co_create_opts = raw_co_create_opts,
2281     .bdrv_has_zero_init = bdrv_has_zero_init_1,
2282     .bdrv_co_block_status = raw_co_block_status,
2283     .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes,
2284 
2285     .bdrv_co_preadv         = raw_co_preadv,
2286     .bdrv_co_pwritev        = raw_co_pwritev,
2287     .bdrv_aio_flush = raw_aio_flush,
2288     .bdrv_aio_pdiscard = raw_aio_pdiscard,
2289     .bdrv_refresh_limits = raw_refresh_limits,
2290     .bdrv_io_plug = raw_aio_plug,
2291     .bdrv_io_unplug = raw_aio_unplug,
2292 
2293     .bdrv_truncate = raw_truncate,
2294     .bdrv_getlength = raw_getlength,
2295     .bdrv_get_info = raw_get_info,
2296     .bdrv_get_allocated_file_size
2297                         = raw_get_allocated_file_size,
2298     .bdrv_check_perm = raw_check_perm,
2299     .bdrv_set_perm   = raw_set_perm,
2300     .bdrv_abort_perm_update = raw_abort_perm_update,
2301     .create_opts = &raw_create_opts,
2302 };
2303 
2304 /***********************************************/
2305 /* host device */
2306 
2307 #if defined(__APPLE__) && defined(__MACH__)
2308 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
2309                                 CFIndex maxPathSize, int flags);
2310 static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
2311 {
2312     kern_return_t kernResult = KERN_FAILURE;
2313     mach_port_t     masterPort;
2314     CFMutableDictionaryRef  classesToMatch;
2315     const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
2316     char *mediaType = NULL;
2317 
2318     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
2319     if ( KERN_SUCCESS != kernResult ) {
2320         printf( "IOMasterPort returned %d\n", kernResult );
2321     }
2322 
2323     int index;
2324     for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
2325         classesToMatch = IOServiceMatching(matching_array[index]);
2326         if (classesToMatch == NULL) {
2327             error_report("IOServiceMatching returned NULL for %s",
2328                          matching_array[index]);
2329             continue;
2330         }
2331         CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
2332                              kCFBooleanTrue);
2333         kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch,
2334                                                   mediaIterator);
2335         if (kernResult != KERN_SUCCESS) {
2336             error_report("Note: IOServiceGetMatchingServices returned %d",
2337                          kernResult);
2338             continue;
2339         }
2340 
2341         /* If a match was found, leave the loop */
2342         if (*mediaIterator != 0) {
2343             DPRINTF("Matching using %s\n", matching_array[index]);
2344             mediaType = g_strdup(matching_array[index]);
2345             break;
2346         }
2347     }
2348     return mediaType;
2349 }
2350 
2351 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
2352                          CFIndex maxPathSize, int flags)
2353 {
2354     io_object_t     nextMedia;
2355     kern_return_t   kernResult = KERN_FAILURE;
2356     *bsdPath = '\0';
2357     nextMedia = IOIteratorNext( mediaIterator );
2358     if ( nextMedia )
2359     {
2360         CFTypeRef   bsdPathAsCFString;
2361     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
2362         if ( bsdPathAsCFString ) {
2363             size_t devPathLength;
2364             strcpy( bsdPath, _PATH_DEV );
2365             if (flags & BDRV_O_NOCACHE) {
2366                 strcat(bsdPath, "r");
2367             }
2368             devPathLength = strlen( bsdPath );
2369             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
2370                 kernResult = KERN_SUCCESS;
2371             }
2372             CFRelease( bsdPathAsCFString );
2373         }
2374         IOObjectRelease( nextMedia );
2375     }
2376 
2377     return kernResult;
2378 }
2379 
2380 /* Sets up a real cdrom for use in QEMU */
2381 static bool setup_cdrom(char *bsd_path, Error **errp)
2382 {
2383     int index, num_of_test_partitions = 2, fd;
2384     char test_partition[MAXPATHLEN];
2385     bool partition_found = false;
2386 
2387     /* look for a working partition */
2388     for (index = 0; index < num_of_test_partitions; index++) {
2389         snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
2390                  index);
2391         fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE);
2392         if (fd >= 0) {
2393             partition_found = true;
2394             qemu_close(fd);
2395             break;
2396         }
2397     }
2398 
2399     /* if a working partition on the device was not found */
2400     if (partition_found == false) {
2401         error_setg(errp, "Failed to find a working partition on disc");
2402     } else {
2403         DPRINTF("Using %s as optical disc\n", test_partition);
2404         pstrcpy(bsd_path, MAXPATHLEN, test_partition);
2405     }
2406     return partition_found;
2407 }
2408 
2409 /* Prints directions on mounting and unmounting a device */
2410 static void print_unmounting_directions(const char *file_name)
2411 {
2412     error_report("If device %s is mounted on the desktop, unmount"
2413                  " it first before using it in QEMU", file_name);
2414     error_report("Command to unmount device: diskutil unmountDisk %s",
2415                  file_name);
2416     error_report("Command to mount device: diskutil mountDisk %s", file_name);
2417 }
2418 
2419 #endif /* defined(__APPLE__) && defined(__MACH__) */
2420 
2421 static int hdev_probe_device(const char *filename)
2422 {
2423     struct stat st;
2424 
2425     /* allow a dedicated CD-ROM driver to match with a higher priority */
2426     if (strstart(filename, "/dev/cdrom", NULL))
2427         return 50;
2428 
2429     if (stat(filename, &st) >= 0 &&
2430             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
2431         return 100;
2432     }
2433 
2434     return 0;
2435 }
2436 
2437 static int check_hdev_writable(BDRVRawState *s)
2438 {
2439 #if defined(BLKROGET)
2440     /* Linux block devices can be configured "read-only" using blockdev(8).
2441      * This is independent of device node permissions and therefore open(2)
2442      * with O_RDWR succeeds.  Actual writes fail with EPERM.
2443      *
2444      * bdrv_open() is supposed to fail if the disk is read-only.  Explicitly
2445      * check for read-only block devices so that Linux block devices behave
2446      * properly.
2447      */
2448     struct stat st;
2449     int readonly = 0;
2450 
2451     if (fstat(s->fd, &st)) {
2452         return -errno;
2453     }
2454 
2455     if (!S_ISBLK(st.st_mode)) {
2456         return 0;
2457     }
2458 
2459     if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
2460         return -errno;
2461     }
2462 
2463     if (readonly) {
2464         return -EACCES;
2465     }
2466 #endif /* defined(BLKROGET) */
2467     return 0;
2468 }
2469 
2470 static void hdev_parse_filename(const char *filename, QDict *options,
2471                                 Error **errp)
2472 {
2473     bdrv_parse_filename_strip_prefix(filename, "host_device:", options);
2474 }
2475 
2476 static bool hdev_is_sg(BlockDriverState *bs)
2477 {
2478 
2479 #if defined(__linux__)
2480 
2481     BDRVRawState *s = bs->opaque;
2482     struct stat st;
2483     struct sg_scsi_id scsiid;
2484     int sg_version;
2485     int ret;
2486 
2487     if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) {
2488         return false;
2489     }
2490 
2491     ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version);
2492     if (ret < 0) {
2493         return false;
2494     }
2495 
2496     ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid);
2497     if (ret >= 0) {
2498         DPRINTF("SG device found: type=%d, version=%d\n",
2499             scsiid.scsi_type, sg_version);
2500         return true;
2501     }
2502 
2503 #endif
2504 
2505     return false;
2506 }
2507 
2508 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
2509                      Error **errp)
2510 {
2511     BDRVRawState *s = bs->opaque;
2512     Error *local_err = NULL;
2513     int ret;
2514 
2515 #if defined(__APPLE__) && defined(__MACH__)
2516     /*
2517      * Caution: while qdict_get_str() is fine, getting non-string types
2518      * would require more care.  When @options come from -blockdev or
2519      * blockdev_add, its members are typed according to the QAPI
2520      * schema, but when they come from -drive, they're all QString.
2521      */
2522     const char *filename = qdict_get_str(options, "filename");
2523     char bsd_path[MAXPATHLEN] = "";
2524     bool error_occurred = false;
2525 
2526     /* If using a real cdrom */
2527     if (strcmp(filename, "/dev/cdrom") == 0) {
2528         char *mediaType = NULL;
2529         kern_return_t ret_val;
2530         io_iterator_t mediaIterator = 0;
2531 
2532         mediaType = FindEjectableOpticalMedia(&mediaIterator);
2533         if (mediaType == NULL) {
2534             error_setg(errp, "Please make sure your CD/DVD is in the optical"
2535                        " drive");
2536             error_occurred = true;
2537             goto hdev_open_Mac_error;
2538         }
2539 
2540         ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
2541         if (ret_val != KERN_SUCCESS) {
2542             error_setg(errp, "Could not get BSD path for optical drive");
2543             error_occurred = true;
2544             goto hdev_open_Mac_error;
2545         }
2546 
2547         /* If a real optical drive was not found */
2548         if (bsd_path[0] == '\0') {
2549             error_setg(errp, "Failed to obtain bsd path for optical drive");
2550             error_occurred = true;
2551             goto hdev_open_Mac_error;
2552         }
2553 
2554         /* If using a cdrom disc and finding a partition on the disc failed */
2555         if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
2556             setup_cdrom(bsd_path, errp) == false) {
2557             print_unmounting_directions(bsd_path);
2558             error_occurred = true;
2559             goto hdev_open_Mac_error;
2560         }
2561 
2562         qdict_put_str(options, "filename", bsd_path);
2563 
2564 hdev_open_Mac_error:
2565         g_free(mediaType);
2566         if (mediaIterator) {
2567             IOObjectRelease(mediaIterator);
2568         }
2569         if (error_occurred) {
2570             return -ENOENT;
2571         }
2572     }
2573 #endif /* defined(__APPLE__) && defined(__MACH__) */
2574 
2575     s->type = FTYPE_FILE;
2576 
2577     ret = raw_open_common(bs, options, flags, 0, &local_err);
2578     if (ret < 0) {
2579         error_propagate(errp, local_err);
2580 #if defined(__APPLE__) && defined(__MACH__)
2581         if (*bsd_path) {
2582             filename = bsd_path;
2583         }
2584         /* if a physical device experienced an error while being opened */
2585         if (strncmp(filename, "/dev/", 5) == 0) {
2586             print_unmounting_directions(filename);
2587         }
2588 #endif /* defined(__APPLE__) && defined(__MACH__) */
2589         return ret;
2590     }
2591 
2592     /* Since this does ioctl the device must be already opened */
2593     bs->sg = hdev_is_sg(bs);
2594 
2595     if (flags & BDRV_O_RDWR) {
2596         ret = check_hdev_writable(s);
2597         if (ret < 0) {
2598             raw_close(bs);
2599             error_setg_errno(errp, -ret, "The device is not writable");
2600             return ret;
2601         }
2602     }
2603 
2604     return ret;
2605 }
2606 
2607 #if defined(__linux__)
2608 
2609 static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
2610         unsigned long int req, void *buf,
2611         BlockCompletionFunc *cb, void *opaque)
2612 {
2613     BDRVRawState *s = bs->opaque;
2614     RawPosixAIOData *acb;
2615     ThreadPool *pool;
2616 
2617     if (fd_open(bs) < 0)
2618         return NULL;
2619 
2620     if (req == SG_IO && s->pr_mgr) {
2621         struct sg_io_hdr *io_hdr = buf;
2622         if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT ||
2623             io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) {
2624             return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs),
2625                                       s->fd, io_hdr, cb, opaque);
2626         }
2627     }
2628 
2629     acb = g_new(RawPosixAIOData, 1);
2630     acb->bs = bs;
2631     acb->aio_type = QEMU_AIO_IOCTL;
2632     acb->aio_fildes = s->fd;
2633     acb->aio_offset = 0;
2634     acb->aio_ioctl_buf = buf;
2635     acb->aio_ioctl_cmd = req;
2636     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
2637     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
2638 }
2639 #endif /* linux */
2640 
2641 static int fd_open(BlockDriverState *bs)
2642 {
2643     BDRVRawState *s = bs->opaque;
2644 
2645     /* this is just to ensure s->fd is sane (its called by io ops) */
2646     if (s->fd >= 0)
2647         return 0;
2648     return -EIO;
2649 }
2650 
2651 static coroutine_fn BlockAIOCB *hdev_aio_pdiscard(BlockDriverState *bs,
2652     int64_t offset, int bytes,
2653     BlockCompletionFunc *cb, void *opaque)
2654 {
2655     BDRVRawState *s = bs->opaque;
2656 
2657     if (fd_open(bs) < 0) {
2658         return NULL;
2659     }
2660     return paio_submit(bs, s->fd, offset, NULL, bytes,
2661                        cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2662 }
2663 
2664 static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs,
2665     int64_t offset, int bytes, BdrvRequestFlags flags)
2666 {
2667     BDRVRawState *s = bs->opaque;
2668     int rc;
2669 
2670     rc = fd_open(bs);
2671     if (rc < 0) {
2672         return rc;
2673     }
2674     if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2675         return paio_submit_co(bs, s->fd, offset, NULL, bytes,
2676                               QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2677     } else if (s->discard_zeroes) {
2678         return paio_submit_co(bs, s->fd, offset, NULL, bytes,
2679                               QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2680     }
2681     return -ENOTSUP;
2682 }
2683 
2684 static int coroutine_fn hdev_co_create_opts(const char *filename, QemuOpts *opts,
2685                                             Error **errp)
2686 {
2687     int fd;
2688     int ret = 0;
2689     struct stat stat_buf;
2690     int64_t total_size = 0;
2691     bool has_prefix;
2692 
2693     /* This function is used by both protocol block drivers and therefore either
2694      * of these prefixes may be given.
2695      * The return value has to be stored somewhere, otherwise this is an error
2696      * due to -Werror=unused-value. */
2697     has_prefix =
2698         strstart(filename, "host_device:", &filename) ||
2699         strstart(filename, "host_cdrom:" , &filename);
2700 
2701     (void)has_prefix;
2702 
2703     ret = raw_normalize_devicepath(&filename);
2704     if (ret < 0) {
2705         error_setg_errno(errp, -ret, "Could not normalize device path");
2706         return ret;
2707     }
2708 
2709     /* Read out options */
2710     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2711                           BDRV_SECTOR_SIZE);
2712 
2713     fd = qemu_open(filename, O_WRONLY | O_BINARY);
2714     if (fd < 0) {
2715         ret = -errno;
2716         error_setg_errno(errp, -ret, "Could not open device");
2717         return ret;
2718     }
2719 
2720     if (fstat(fd, &stat_buf) < 0) {
2721         ret = -errno;
2722         error_setg_errno(errp, -ret, "Could not stat device");
2723     } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2724         error_setg(errp,
2725                    "The given file is neither a block nor a character device");
2726         ret = -ENODEV;
2727     } else if (lseek(fd, 0, SEEK_END) < total_size) {
2728         error_setg(errp, "Device is too small");
2729         ret = -ENOSPC;
2730     }
2731 
2732     if (!ret && total_size) {
2733         uint8_t buf[BDRV_SECTOR_SIZE] = { 0 };
2734         int64_t zero_size = MIN(BDRV_SECTOR_SIZE, total_size);
2735         if (lseek(fd, 0, SEEK_SET) == -1) {
2736             ret = -errno;
2737         } else {
2738             ret = qemu_write_full(fd, buf, zero_size);
2739             ret = ret == zero_size ? 0 : -errno;
2740         }
2741     }
2742     qemu_close(fd);
2743     return ret;
2744 }
2745 
2746 static BlockDriver bdrv_host_device = {
2747     .format_name        = "host_device",
2748     .protocol_name        = "host_device",
2749     .instance_size      = sizeof(BDRVRawState),
2750     .bdrv_needs_filename = true,
2751     .bdrv_probe_device  = hdev_probe_device,
2752     .bdrv_parse_filename = hdev_parse_filename,
2753     .bdrv_file_open     = hdev_open,
2754     .bdrv_close         = raw_close,
2755     .bdrv_reopen_prepare = raw_reopen_prepare,
2756     .bdrv_reopen_commit  = raw_reopen_commit,
2757     .bdrv_reopen_abort   = raw_reopen_abort,
2758     .bdrv_co_create_opts = hdev_co_create_opts,
2759     .create_opts         = &raw_create_opts,
2760     .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes,
2761 
2762     .bdrv_co_preadv         = raw_co_preadv,
2763     .bdrv_co_pwritev        = raw_co_pwritev,
2764     .bdrv_aio_flush	= raw_aio_flush,
2765     .bdrv_aio_pdiscard   = hdev_aio_pdiscard,
2766     .bdrv_refresh_limits = raw_refresh_limits,
2767     .bdrv_io_plug = raw_aio_plug,
2768     .bdrv_io_unplug = raw_aio_unplug,
2769 
2770     .bdrv_truncate      = raw_truncate,
2771     .bdrv_getlength	= raw_getlength,
2772     .bdrv_get_info = raw_get_info,
2773     .bdrv_get_allocated_file_size
2774                         = raw_get_allocated_file_size,
2775     .bdrv_check_perm = raw_check_perm,
2776     .bdrv_set_perm   = raw_set_perm,
2777     .bdrv_abort_perm_update = raw_abort_perm_update,
2778     .bdrv_probe_blocksizes = hdev_probe_blocksizes,
2779     .bdrv_probe_geometry = hdev_probe_geometry,
2780 
2781     /* generic scsi device */
2782 #ifdef __linux__
2783     .bdrv_aio_ioctl     = hdev_aio_ioctl,
2784 #endif
2785 };
2786 
2787 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2788 static void cdrom_parse_filename(const char *filename, QDict *options,
2789                                  Error **errp)
2790 {
2791     bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options);
2792 }
2793 #endif
2794 
2795 #ifdef __linux__
2796 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2797                       Error **errp)
2798 {
2799     BDRVRawState *s = bs->opaque;
2800 
2801     s->type = FTYPE_CD;
2802 
2803     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2804     return raw_open_common(bs, options, flags, O_NONBLOCK, errp);
2805 }
2806 
2807 static int cdrom_probe_device(const char *filename)
2808 {
2809     int fd, ret;
2810     int prio = 0;
2811     struct stat st;
2812 
2813     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2814     if (fd < 0) {
2815         goto out;
2816     }
2817     ret = fstat(fd, &st);
2818     if (ret == -1 || !S_ISBLK(st.st_mode)) {
2819         goto outc;
2820     }
2821 
2822     /* Attempt to detect via a CDROM specific ioctl */
2823     ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2824     if (ret >= 0)
2825         prio = 100;
2826 
2827 outc:
2828     qemu_close(fd);
2829 out:
2830     return prio;
2831 }
2832 
2833 static bool cdrom_is_inserted(BlockDriverState *bs)
2834 {
2835     BDRVRawState *s = bs->opaque;
2836     int ret;
2837 
2838     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2839     return ret == CDS_DISC_OK;
2840 }
2841 
2842 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2843 {
2844     BDRVRawState *s = bs->opaque;
2845 
2846     if (eject_flag) {
2847         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2848             perror("CDROMEJECT");
2849     } else {
2850         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2851             perror("CDROMEJECT");
2852     }
2853 }
2854 
2855 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2856 {
2857     BDRVRawState *s = bs->opaque;
2858 
2859     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2860         /*
2861          * Note: an error can happen if the distribution automatically
2862          * mounts the CD-ROM
2863          */
2864         /* perror("CDROM_LOCKDOOR"); */
2865     }
2866 }
2867 
2868 static BlockDriver bdrv_host_cdrom = {
2869     .format_name        = "host_cdrom",
2870     .protocol_name      = "host_cdrom",
2871     .instance_size      = sizeof(BDRVRawState),
2872     .bdrv_needs_filename = true,
2873     .bdrv_probe_device	= cdrom_probe_device,
2874     .bdrv_parse_filename = cdrom_parse_filename,
2875     .bdrv_file_open     = cdrom_open,
2876     .bdrv_close         = raw_close,
2877     .bdrv_reopen_prepare = raw_reopen_prepare,
2878     .bdrv_reopen_commit  = raw_reopen_commit,
2879     .bdrv_reopen_abort   = raw_reopen_abort,
2880     .bdrv_co_create_opts = hdev_co_create_opts,
2881     .create_opts         = &raw_create_opts,
2882 
2883 
2884     .bdrv_co_preadv         = raw_co_preadv,
2885     .bdrv_co_pwritev        = raw_co_pwritev,
2886     .bdrv_aio_flush	= raw_aio_flush,
2887     .bdrv_refresh_limits = raw_refresh_limits,
2888     .bdrv_io_plug = raw_aio_plug,
2889     .bdrv_io_unplug = raw_aio_unplug,
2890 
2891     .bdrv_truncate      = raw_truncate,
2892     .bdrv_getlength      = raw_getlength,
2893     .has_variable_length = true,
2894     .bdrv_get_allocated_file_size
2895                         = raw_get_allocated_file_size,
2896 
2897     /* removable device support */
2898     .bdrv_is_inserted   = cdrom_is_inserted,
2899     .bdrv_eject         = cdrom_eject,
2900     .bdrv_lock_medium   = cdrom_lock_medium,
2901 
2902     /* generic scsi device */
2903     .bdrv_aio_ioctl     = hdev_aio_ioctl,
2904 };
2905 #endif /* __linux__ */
2906 
2907 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2908 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2909                       Error **errp)
2910 {
2911     BDRVRawState *s = bs->opaque;
2912     Error *local_err = NULL;
2913     int ret;
2914 
2915     s->type = FTYPE_CD;
2916 
2917     ret = raw_open_common(bs, options, flags, 0, &local_err);
2918     if (ret) {
2919         error_propagate(errp, local_err);
2920         return ret;
2921     }
2922 
2923     /* make sure the door isn't locked at this time */
2924     ioctl(s->fd, CDIOCALLOW);
2925     return 0;
2926 }
2927 
2928 static int cdrom_probe_device(const char *filename)
2929 {
2930     if (strstart(filename, "/dev/cd", NULL) ||
2931             strstart(filename, "/dev/acd", NULL))
2932         return 100;
2933     return 0;
2934 }
2935 
2936 static int cdrom_reopen(BlockDriverState *bs)
2937 {
2938     BDRVRawState *s = bs->opaque;
2939     int fd;
2940 
2941     /*
2942      * Force reread of possibly changed/newly loaded disc,
2943      * FreeBSD seems to not notice sometimes...
2944      */
2945     if (s->fd >= 0)
2946         qemu_close(s->fd);
2947     fd = qemu_open(bs->filename, s->open_flags, 0644);
2948     if (fd < 0) {
2949         s->fd = -1;
2950         return -EIO;
2951     }
2952     s->fd = fd;
2953 
2954     /* make sure the door isn't locked at this time */
2955     ioctl(s->fd, CDIOCALLOW);
2956     return 0;
2957 }
2958 
2959 static bool cdrom_is_inserted(BlockDriverState *bs)
2960 {
2961     return raw_getlength(bs) > 0;
2962 }
2963 
2964 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2965 {
2966     BDRVRawState *s = bs->opaque;
2967 
2968     if (s->fd < 0)
2969         return;
2970 
2971     (void) ioctl(s->fd, CDIOCALLOW);
2972 
2973     if (eject_flag) {
2974         if (ioctl(s->fd, CDIOCEJECT) < 0)
2975             perror("CDIOCEJECT");
2976     } else {
2977         if (ioctl(s->fd, CDIOCCLOSE) < 0)
2978             perror("CDIOCCLOSE");
2979     }
2980 
2981     cdrom_reopen(bs);
2982 }
2983 
2984 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2985 {
2986     BDRVRawState *s = bs->opaque;
2987 
2988     if (s->fd < 0)
2989         return;
2990     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2991         /*
2992          * Note: an error can happen if the distribution automatically
2993          * mounts the CD-ROM
2994          */
2995         /* perror("CDROM_LOCKDOOR"); */
2996     }
2997 }
2998 
2999 static BlockDriver bdrv_host_cdrom = {
3000     .format_name        = "host_cdrom",
3001     .protocol_name      = "host_cdrom",
3002     .instance_size      = sizeof(BDRVRawState),
3003     .bdrv_needs_filename = true,
3004     .bdrv_probe_device	= cdrom_probe_device,
3005     .bdrv_parse_filename = cdrom_parse_filename,
3006     .bdrv_file_open     = cdrom_open,
3007     .bdrv_close         = raw_close,
3008     .bdrv_reopen_prepare = raw_reopen_prepare,
3009     .bdrv_reopen_commit  = raw_reopen_commit,
3010     .bdrv_reopen_abort   = raw_reopen_abort,
3011     .bdrv_co_create_opts = hdev_co_create_opts,
3012     .create_opts        = &raw_create_opts,
3013 
3014     .bdrv_co_preadv         = raw_co_preadv,
3015     .bdrv_co_pwritev        = raw_co_pwritev,
3016     .bdrv_aio_flush	= raw_aio_flush,
3017     .bdrv_refresh_limits = raw_refresh_limits,
3018     .bdrv_io_plug = raw_aio_plug,
3019     .bdrv_io_unplug = raw_aio_unplug,
3020 
3021     .bdrv_truncate      = raw_truncate,
3022     .bdrv_getlength      = raw_getlength,
3023     .has_variable_length = true,
3024     .bdrv_get_allocated_file_size
3025                         = raw_get_allocated_file_size,
3026 
3027     /* removable device support */
3028     .bdrv_is_inserted   = cdrom_is_inserted,
3029     .bdrv_eject         = cdrom_eject,
3030     .bdrv_lock_medium   = cdrom_lock_medium,
3031 };
3032 #endif /* __FreeBSD__ */
3033 
3034 static void bdrv_file_init(void)
3035 {
3036     /*
3037      * Register all the drivers.  Note that order is important, the driver
3038      * registered last will get probed first.
3039      */
3040     bdrv_register(&bdrv_file);
3041     bdrv_register(&bdrv_host_device);
3042 #ifdef __linux__
3043     bdrv_register(&bdrv_host_cdrom);
3044 #endif
3045 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3046     bdrv_register(&bdrv_host_cdrom);
3047 #endif
3048 }
3049 
3050 block_init(bdrv_file_init);
3051