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 "qemu-common.h"
27 #include "qapi/error.h"
28 #include "qemu/cutils.h"
29 #include "qemu/error-report.h"
30 #include "block/block_int.h"
31 #include "qemu/module.h"
32 #include "qemu/option.h"
33 #include "trace.h"
34 #include "block/thread-pool.h"
35 #include "qemu/iov.h"
36 #include "block/raw-aio.h"
37 #include "qapi/qmp/qdict.h"
38 #include "qapi/qmp/qstring.h"
39
40 #include "scsi/pr-manager.h"
41 #include "scsi/constants.h"
42
43 #if defined(__APPLE__) && (__MACH__)
44 #include <paths.h>
45 #include <sys/param.h>
46 #include <IOKit/IOKitLib.h>
47 #include <IOKit/IOBSD.h>
48 #include <IOKit/storage/IOMediaBSDClient.h>
49 #include <IOKit/storage/IOMedia.h>
50 #include <IOKit/storage/IOCDMedia.h>
51 //#include <IOKit/storage/IOCDTypes.h>
52 #include <IOKit/storage/IODVDMedia.h>
53 #include <CoreFoundation/CoreFoundation.h>
54 #endif
55
56 #ifdef __sun__
57 #define _POSIX_PTHREAD_SEMANTICS 1
58 #include <sys/dkio.h>
59 #endif
60 #ifdef __linux__
61 #include <sys/ioctl.h>
62 #include <sys/param.h>
63 #include <sys/syscall.h>
64 #include <linux/cdrom.h>
65 #include <linux/fd.h>
66 #include <linux/fs.h>
67 #include <linux/hdreg.h>
68 #include <scsi/sg.h>
69 #ifdef __s390__
70 #include <asm/dasd.h>
71 #endif
72 #ifndef FS_NOCOW_FL
73 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
74 #endif
75 #endif
76 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
77 #include <linux/falloc.h>
78 #endif
79 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
80 #include <sys/disk.h>
81 #include <sys/cdio.h>
82 #endif
83
84 #ifdef __OpenBSD__
85 #include <sys/ioctl.h>
86 #include <sys/disklabel.h>
87 #include <sys/dkio.h>
88 #endif
89
90 #ifdef __NetBSD__
91 #include <sys/ioctl.h>
92 #include <sys/disklabel.h>
93 #include <sys/dkio.h>
94 #include <sys/disk.h>
95 #endif
96
97 #ifdef __DragonFly__
98 #include <sys/ioctl.h>
99 #include <sys/diskslice.h>
100 #endif
101
102 #ifdef CONFIG_XFS
103 #include <xfs/xfs.h>
104 #endif
105
106 #include "trace.h"
107
108 /* OS X does not have O_DSYNC */
109 #ifndef O_DSYNC
110 #ifdef O_SYNC
111 #define O_DSYNC O_SYNC
112 #elif defined(O_FSYNC)
113 #define O_DSYNC O_FSYNC
114 #endif
115 #endif
116
117 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
118 #ifndef O_DIRECT
119 #define O_DIRECT O_DSYNC
120 #endif
121
122 #define FTYPE_FILE 0
123 #define FTYPE_CD 1
124
125 #define MAX_BLOCKSIZE 4096
126
127 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
128 * leaving a few more bytes for its future use. */
129 #define RAW_LOCK_PERM_BASE 100
130 #define RAW_LOCK_SHARED_BASE 200
131
132 typedef struct BDRVRawState {
133 int fd;
134 bool use_lock;
135 int type;
136 int open_flags;
137 size_t buf_align;
138
139 /* The current permissions. */
140 uint64_t perm;
141 uint64_t shared_perm;
142
143 /* The perms bits whose corresponding bytes are already locked in
144 * s->fd. */
145 uint64_t locked_perm;
146 uint64_t locked_shared_perm;
147
148 int perm_change_fd;
149 int perm_change_flags;
150 BDRVReopenState *reopen_state;
151
152 #ifdef CONFIG_XFS
153 bool is_xfs:1;
154 #endif
155 bool has_discard:1;
156 bool has_write_zeroes:1;
157 bool discard_zeroes:1;
158 bool use_linux_aio:1;
159 bool use_linux_io_uring:1;
160 bool page_cache_inconsistent:1;
161 bool has_fallocate;
162 bool needs_alignment;
163 bool drop_cache;
164 bool check_cache_dropped;
165 struct {
166 uint64_t discard_nb_ok;
167 uint64_t discard_nb_failed;
168 uint64_t discard_bytes_ok;
169 } stats;
170
171 PRManager *pr_mgr;
172 } BDRVRawState;
173
174 typedef struct BDRVRawReopenState {
175 int fd;
176 int open_flags;
177 bool drop_cache;
178 bool check_cache_dropped;
179 } BDRVRawReopenState;
180
181 static int fd_open(BlockDriverState *bs);
182 static int64_t raw_getlength(BlockDriverState *bs);
183
184 typedef struct RawPosixAIOData {
185 BlockDriverState *bs;
186 int aio_type;
187 int aio_fildes;
188
189 off_t aio_offset;
190 uint64_t aio_nbytes;
191
192 union {
193 struct {
194 struct iovec *iov;
195 int niov;
196 } io;
197 struct {
198 uint64_t cmd;
199 void *buf;
200 } ioctl;
201 struct {
202 int aio_fd2;
203 off_t aio_offset2;
204 } copy_range;
205 struct {
206 PreallocMode prealloc;
207 Error **errp;
208 } truncate;
209 };
210 } RawPosixAIOData;
211
212 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
213 static int cdrom_reopen(BlockDriverState *bs);
214 #endif
215
216 #if defined(__NetBSD__)
raw_normalize_devicepath(const char ** filename,Error ** errp)217 static int raw_normalize_devicepath(const char **filename, Error **errp)
218 {
219 static char namebuf[PATH_MAX];
220 const char *dp, *fname;
221 struct stat sb;
222
223 fname = *filename;
224 dp = strrchr(fname, '/');
225 if (lstat(fname, &sb) < 0) {
226 error_setg_file_open(errp, errno, fname);
227 return -errno;
228 }
229
230 if (!S_ISBLK(sb.st_mode)) {
231 return 0;
232 }
233
234 if (dp == NULL) {
235 snprintf(namebuf, PATH_MAX, "r%s", fname);
236 } else {
237 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
238 (int)(dp - fname), fname, dp + 1);
239 }
240 *filename = namebuf;
241 warn_report("%s is a block device, using %s", fname, *filename);
242
243 return 0;
244 }
245 #else
raw_normalize_devicepath(const char ** filename,Error ** errp)246 static int raw_normalize_devicepath(const char **filename, Error **errp)
247 {
248 return 0;
249 }
250 #endif
251
252 /*
253 * Get logical block size via ioctl. On success store it in @sector_size_p.
254 */
probe_logical_blocksize(int fd,unsigned int * sector_size_p)255 static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
256 {
257 unsigned int sector_size;
258 bool success = false;
259 int i;
260
261 errno = ENOTSUP;
262 static const unsigned long ioctl_list[] = {
263 #ifdef BLKSSZGET
264 BLKSSZGET,
265 #endif
266 #ifdef DKIOCGETBLOCKSIZE
267 DKIOCGETBLOCKSIZE,
268 #endif
269 #ifdef DIOCGSECTORSIZE
270 DIOCGSECTORSIZE,
271 #endif
272 };
273
274 /* Try a few ioctls to get the right size */
275 for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) {
276 if (ioctl(fd, ioctl_list[i], §or_size) >= 0) {
277 *sector_size_p = sector_size;
278 success = true;
279 }
280 }
281
282 return success ? 0 : -errno;
283 }
284
285 /**
286 * Get physical block size of @fd.
287 * On success, store it in @blk_size and return 0.
288 * On failure, return -errno.
289 */
probe_physical_blocksize(int fd,unsigned int * blk_size)290 static int probe_physical_blocksize(int fd, unsigned int *blk_size)
291 {
292 #ifdef BLKPBSZGET
293 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
294 return -errno;
295 }
296 return 0;
297 #else
298 return -ENOTSUP;
299 #endif
300 }
301
302 /* Check if read is allowed with given memory buffer and length.
303 *
304 * This function is used to check O_DIRECT memory buffer and request alignment.
305 */
raw_is_io_aligned(int fd,void * buf,size_t len)306 static bool raw_is_io_aligned(int fd, void *buf, size_t len)
307 {
308 ssize_t ret = pread(fd, buf, len, 0);
309
310 if (ret >= 0) {
311 return true;
312 }
313
314 #ifdef __linux__
315 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
316 * other errors (e.g. real I/O error), which could happen on a failed
317 * drive, since we only care about probing alignment.
318 */
319 if (errno != EINVAL) {
320 return true;
321 }
322 #endif
323
324 return false;
325 }
326
raw_probe_alignment(BlockDriverState * bs,int fd,Error ** errp)327 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
328 {
329 BDRVRawState *s = bs->opaque;
330 char *buf;
331 size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size);
332 size_t alignments[] = {1, 512, 1024, 2048, 4096};
333
334 /* For SCSI generic devices the alignment is not really used.
335 With buffered I/O, we don't have any restrictions. */
336 if (bdrv_is_sg(bs) || !s->needs_alignment) {
337 bs->bl.request_alignment = 1;
338 s->buf_align = 1;
339 return;
340 }
341
342 bs->bl.request_alignment = 0;
343 s->buf_align = 0;
344 /* Let's try to use the logical blocksize for the alignment. */
345 if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) {
346 bs->bl.request_alignment = 0;
347 }
348 #ifdef CONFIG_XFS
349 if (s->is_xfs) {
350 struct dioattr da;
351 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
352 bs->bl.request_alignment = da.d_miniosz;
353 /* The kernel returns wrong information for d_mem */
354 /* s->buf_align = da.d_mem; */
355 }
356 }
357 #endif
358
359 /*
360 * If we could not get the sizes so far, we can only guess them. First try
361 * to detect request alignment, since it is more likely to succeed. Then
362 * try to detect buf_align, which cannot be detected in some cases (e.g.
363 * Gluster). If buf_align cannot be detected, we fallback to the value of
364 * request_alignment.
365 */
366
367 if (!bs->bl.request_alignment) {
368 int i;
369 size_t align;
370 buf = qemu_memalign(max_align, max_align);
371 for (i = 0; i < ARRAY_SIZE(alignments); i++) {
372 align = alignments[i];
373 if (raw_is_io_aligned(fd, buf, align)) {
374 /* Fallback to safe value. */
375 bs->bl.request_alignment = (align != 1) ? align : max_align;
376 break;
377 }
378 }
379 qemu_vfree(buf);
380 }
381
382 if (!s->buf_align) {
383 int i;
384 size_t align;
385 buf = qemu_memalign(max_align, 2 * max_align);
386 for (i = 0; i < ARRAY_SIZE(alignments); i++) {
387 align = alignments[i];
388 if (raw_is_io_aligned(fd, buf + align, max_align)) {
389 /* Fallback to request_alignment. */
390 s->buf_align = (align != 1) ? align : bs->bl.request_alignment;
391 break;
392 }
393 }
394 qemu_vfree(buf);
395 }
396
397 if (!s->buf_align || !bs->bl.request_alignment) {
398 error_setg(errp, "Could not find working O_DIRECT alignment");
399 error_append_hint(errp, "Try cache.direct=off\n");
400 }
401 }
402
raw_parse_flags(int bdrv_flags,int * open_flags,bool has_writers)403 static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers)
404 {
405 bool read_write = false;
406 assert(open_flags != NULL);
407
408 *open_flags |= O_BINARY;
409 *open_flags &= ~O_ACCMODE;
410
411 if (bdrv_flags & BDRV_O_AUTO_RDONLY) {
412 read_write = has_writers;
413 } else if (bdrv_flags & BDRV_O_RDWR) {
414 read_write = true;
415 }
416
417 if (read_write) {
418 *open_flags |= O_RDWR;
419 } else {
420 *open_flags |= O_RDONLY;
421 }
422
423 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
424 * and O_DIRECT for no caching. */
425 if ((bdrv_flags & BDRV_O_NOCACHE)) {
426 *open_flags |= O_DIRECT;
427 }
428 }
429
raw_parse_filename(const char * filename,QDict * options,Error ** errp)430 static void raw_parse_filename(const char *filename, QDict *options,
431 Error **errp)
432 {
433 bdrv_parse_filename_strip_prefix(filename, "file:", options);
434 }
435
436 static QemuOptsList raw_runtime_opts = {
437 .name = "raw",
438 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
439 .desc = {
440 {
441 .name = "filename",
442 .type = QEMU_OPT_STRING,
443 .help = "File name of the image",
444 },
445 {
446 .name = "aio",
447 .type = QEMU_OPT_STRING,
448 .help = "host AIO implementation (threads, native, io_uring)",
449 },
450 {
451 .name = "locking",
452 .type = QEMU_OPT_STRING,
453 .help = "file locking mode (on/off/auto, default: auto)",
454 },
455 {
456 .name = "pr-manager",
457 .type = QEMU_OPT_STRING,
458 .help = "id of persistent reservation manager object (default: none)",
459 },
460 #if defined(__linux__)
461 {
462 .name = "drop-cache",
463 .type = QEMU_OPT_BOOL,
464 .help = "invalidate page cache during live migration (default: on)",
465 },
466 #endif
467 {
468 .name = "x-check-cache-dropped",
469 .type = QEMU_OPT_BOOL,
470 .help = "check that page cache was dropped on live migration (default: off)"
471 },
472 { /* end of list */ }
473 },
474 };
475
476 static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL };
477
raw_open_common(BlockDriverState * bs,QDict * options,int bdrv_flags,int open_flags,bool device,Error ** errp)478 static int raw_open_common(BlockDriverState *bs, QDict *options,
479 int bdrv_flags, int open_flags,
480 bool device, Error **errp)
481 {
482 BDRVRawState *s = bs->opaque;
483 QemuOpts *opts;
484 Error *local_err = NULL;
485 const char *filename = NULL;
486 const char *str;
487 BlockdevAioOptions aio, aio_default;
488 int fd, ret;
489 struct stat st;
490 OnOffAuto locking;
491
492 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
493 qemu_opts_absorb_qdict(opts, options, &local_err);
494 if (local_err) {
495 error_propagate(errp, local_err);
496 ret = -EINVAL;
497 goto fail;
498 }
499
500 filename = qemu_opt_get(opts, "filename");
501
502 ret = raw_normalize_devicepath(&filename, errp);
503 if (ret != 0) {
504 goto fail;
505 }
506
507 if (bdrv_flags & BDRV_O_NATIVE_AIO) {
508 aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE;
509 #ifdef CONFIG_LINUX_IO_URING
510 } else if (bdrv_flags & BDRV_O_IO_URING) {
511 aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING;
512 #endif
513 } else {
514 aio_default = BLOCKDEV_AIO_OPTIONS_THREADS;
515 }
516
517 aio = qapi_enum_parse(&BlockdevAioOptions_lookup,
518 qemu_opt_get(opts, "aio"),
519 aio_default, &local_err);
520 if (local_err) {
521 error_propagate(errp, local_err);
522 ret = -EINVAL;
523 goto fail;
524 }
525
526 s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE);
527 #ifdef CONFIG_LINUX_IO_URING
528 s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING);
529 #endif
530
531 locking = qapi_enum_parse(&OnOffAuto_lookup,
532 qemu_opt_get(opts, "locking"),
533 ON_OFF_AUTO_AUTO, &local_err);
534 if (local_err) {
535 error_propagate(errp, local_err);
536 ret = -EINVAL;
537 goto fail;
538 }
539 switch (locking) {
540 case ON_OFF_AUTO_ON:
541 s->use_lock = true;
542 if (!qemu_has_ofd_lock()) {
543 warn_report("File lock requested but OFD locking syscall is "
544 "unavailable, falling back to POSIX file locks");
545 error_printf("Due to the implementation, locks can be lost "
546 "unexpectedly.\n");
547 }
548 break;
549 case ON_OFF_AUTO_OFF:
550 s->use_lock = false;
551 break;
552 case ON_OFF_AUTO_AUTO:
553 s->use_lock = qemu_has_ofd_lock();
554 break;
555 default:
556 abort();
557 }
558
559 str = qemu_opt_get(opts, "pr-manager");
560 if (str) {
561 s->pr_mgr = pr_manager_lookup(str, &local_err);
562 if (local_err) {
563 error_propagate(errp, local_err);
564 ret = -EINVAL;
565 goto fail;
566 }
567 }
568
569 s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true);
570 s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped",
571 false);
572
573 s->open_flags = open_flags;
574 raw_parse_flags(bdrv_flags, &s->open_flags, false);
575
576 s->fd = -1;
577 fd = qemu_open(filename, s->open_flags, 0644);
578 ret = fd < 0 ? -errno : 0;
579
580 if (ret < 0) {
581 error_setg_file_open(errp, -ret, filename);
582 if (ret == -EROFS) {
583 ret = -EACCES;
584 }
585 goto fail;
586 }
587 s->fd = fd;
588
589 s->perm = 0;
590 s->shared_perm = BLK_PERM_ALL;
591
592 #ifdef CONFIG_LINUX_AIO
593 /* Currently Linux does AIO only for files opened with O_DIRECT */
594 if (s->use_linux_aio) {
595 if (!(s->open_flags & O_DIRECT)) {
596 error_setg(errp, "aio=native was specified, but it requires "
597 "cache.direct=on, which was not specified.");
598 ret = -EINVAL;
599 goto fail;
600 }
601 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) {
602 error_prepend(errp, "Unable to use native AIO: ");
603 goto fail;
604 }
605 }
606 #else
607 if (s->use_linux_aio) {
608 error_setg(errp, "aio=native was specified, but is not supported "
609 "in this build.");
610 ret = -EINVAL;
611 goto fail;
612 }
613 #endif /* !defined(CONFIG_LINUX_AIO) */
614
615 #ifdef CONFIG_LINUX_IO_URING
616 if (s->use_linux_io_uring) {
617 if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs), errp)) {
618 error_prepend(errp, "Unable to use io_uring: ");
619 goto fail;
620 }
621 }
622 #else
623 if (s->use_linux_io_uring) {
624 error_setg(errp, "aio=io_uring was specified, but is not supported "
625 "in this build.");
626 ret = -EINVAL;
627 goto fail;
628 }
629 #endif /* !defined(CONFIG_LINUX_IO_URING) */
630
631 s->has_discard = true;
632 s->has_write_zeroes = true;
633 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
634 s->needs_alignment = true;
635 }
636
637 if (fstat(s->fd, &st) < 0) {
638 ret = -errno;
639 error_setg_errno(errp, errno, "Could not stat file");
640 goto fail;
641 }
642
643 if (!device) {
644 if (S_ISBLK(st.st_mode)) {
645 warn_report("Opening a block device as a file using the '%s' "
646 "driver is deprecated", bs->drv->format_name);
647 } else if (S_ISCHR(st.st_mode)) {
648 warn_report("Opening a character device as a file using the '%s' "
649 "driver is deprecated", bs->drv->format_name);
650 } else if (!S_ISREG(st.st_mode)) {
651 error_setg(errp, "A regular file was expected by the '%s' driver, "
652 "but something else was given", bs->drv->format_name);
653 ret = -EINVAL;
654 goto fail;
655 } else {
656 s->discard_zeroes = true;
657 s->has_fallocate = true;
658 }
659 } else {
660 if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
661 error_setg(errp, "'%s' driver expects either "
662 "a character or block device", bs->drv->format_name);
663 ret = -EINVAL;
664 goto fail;
665 }
666 }
667
668 if (S_ISBLK(st.st_mode)) {
669 #ifdef BLKDISCARDZEROES
670 unsigned int arg;
671 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
672 s->discard_zeroes = true;
673 }
674 #endif
675 #ifdef __linux__
676 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
677 * not rely on the contents of discarded blocks unless using O_DIRECT.
678 * Same for BLKZEROOUT.
679 */
680 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
681 s->discard_zeroes = false;
682 s->has_write_zeroes = false;
683 }
684 #endif
685 }
686 #ifdef __FreeBSD__
687 if (S_ISCHR(st.st_mode)) {
688 /*
689 * The file is a char device (disk), which on FreeBSD isn't behind
690 * a pager, so force all requests to be aligned. This is needed
691 * so QEMU makes sure all IO operations on the device are aligned
692 * to sector size, or else FreeBSD will reject them with EINVAL.
693 */
694 s->needs_alignment = true;
695 }
696 #endif
697
698 #ifdef CONFIG_XFS
699 if (platform_test_xfs_fd(s->fd)) {
700 s->is_xfs = true;
701 }
702 #endif
703
704 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK;
705 ret = 0;
706 fail:
707 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
708 unlink(filename);
709 }
710 qemu_opts_del(opts);
711 return ret;
712 }
713
raw_open(BlockDriverState * bs,QDict * options,int flags,Error ** errp)714 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
715 Error **errp)
716 {
717 BDRVRawState *s = bs->opaque;
718
719 s->type = FTYPE_FILE;
720 return raw_open_common(bs, options, flags, 0, false, errp);
721 }
722
723 typedef enum {
724 RAW_PL_PREPARE,
725 RAW_PL_COMMIT,
726 RAW_PL_ABORT,
727 } RawPermLockOp;
728
729 #define PERM_FOREACH(i) \
730 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
731
732 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
733 * file; if @unlock == true, also unlock the unneeded bytes.
734 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
735 */
raw_apply_lock_bytes(BDRVRawState * s,int fd,uint64_t perm_lock_bits,uint64_t shared_perm_lock_bits,bool unlock,Error ** errp)736 static int raw_apply_lock_bytes(BDRVRawState *s, int fd,
737 uint64_t perm_lock_bits,
738 uint64_t shared_perm_lock_bits,
739 bool unlock, Error **errp)
740 {
741 int ret;
742 int i;
743 uint64_t locked_perm, locked_shared_perm;
744
745 if (s) {
746 locked_perm = s->locked_perm;
747 locked_shared_perm = s->locked_shared_perm;
748 } else {
749 /*
750 * We don't have the previous bits, just lock/unlock for each of the
751 * requested bits.
752 */
753 if (unlock) {
754 locked_perm = BLK_PERM_ALL;
755 locked_shared_perm = BLK_PERM_ALL;
756 } else {
757 locked_perm = 0;
758 locked_shared_perm = 0;
759 }
760 }
761
762 PERM_FOREACH(i) {
763 int off = RAW_LOCK_PERM_BASE + i;
764 uint64_t bit = (1ULL << i);
765 if ((perm_lock_bits & bit) && !(locked_perm & bit)) {
766 ret = qemu_lock_fd(fd, off, 1, false);
767 if (ret) {
768 error_setg(errp, "Failed to lock byte %d", off);
769 return ret;
770 } else if (s) {
771 s->locked_perm |= bit;
772 }
773 } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) {
774 ret = qemu_unlock_fd(fd, off, 1);
775 if (ret) {
776 error_setg(errp, "Failed to unlock byte %d", off);
777 return ret;
778 } else if (s) {
779 s->locked_perm &= ~bit;
780 }
781 }
782 }
783 PERM_FOREACH(i) {
784 int off = RAW_LOCK_SHARED_BASE + i;
785 uint64_t bit = (1ULL << i);
786 if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) {
787 ret = qemu_lock_fd(fd, off, 1, false);
788 if (ret) {
789 error_setg(errp, "Failed to lock byte %d", off);
790 return ret;
791 } else if (s) {
792 s->locked_shared_perm |= bit;
793 }
794 } else if (unlock && (locked_shared_perm & bit) &&
795 !(shared_perm_lock_bits & bit)) {
796 ret = qemu_unlock_fd(fd, off, 1);
797 if (ret) {
798 error_setg(errp, "Failed to unlock byte %d", off);
799 return ret;
800 } else if (s) {
801 s->locked_shared_perm &= ~bit;
802 }
803 }
804 }
805 return 0;
806 }
807
808 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
raw_check_lock_bytes(int fd,uint64_t perm,uint64_t shared_perm,Error ** errp)809 static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm,
810 Error **errp)
811 {
812 int ret;
813 int i;
814
815 PERM_FOREACH(i) {
816 int off = RAW_LOCK_SHARED_BASE + i;
817 uint64_t p = 1ULL << i;
818 if (perm & p) {
819 ret = qemu_lock_fd_test(fd, off, 1, true);
820 if (ret) {
821 char *perm_name = bdrv_perm_names(p);
822 error_setg(errp,
823 "Failed to get \"%s\" lock",
824 perm_name);
825 g_free(perm_name);
826 return ret;
827 }
828 }
829 }
830 PERM_FOREACH(i) {
831 int off = RAW_LOCK_PERM_BASE + i;
832 uint64_t p = 1ULL << i;
833 if (!(shared_perm & p)) {
834 ret = qemu_lock_fd_test(fd, off, 1, true);
835 if (ret) {
836 char *perm_name = bdrv_perm_names(p);
837 error_setg(errp,
838 "Failed to get shared \"%s\" lock",
839 perm_name);
840 g_free(perm_name);
841 return ret;
842 }
843 }
844 }
845 return 0;
846 }
847
raw_handle_perm_lock(BlockDriverState * bs,RawPermLockOp op,uint64_t new_perm,uint64_t new_shared,Error ** errp)848 static int raw_handle_perm_lock(BlockDriverState *bs,
849 RawPermLockOp op,
850 uint64_t new_perm, uint64_t new_shared,
851 Error **errp)
852 {
853 BDRVRawState *s = bs->opaque;
854 int ret = 0;
855 Error *local_err = NULL;
856
857 if (!s->use_lock) {
858 return 0;
859 }
860
861 if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) {
862 return 0;
863 }
864
865 switch (op) {
866 case RAW_PL_PREPARE:
867 if ((s->perm | new_perm) == s->perm &&
868 (s->shared_perm & new_shared) == s->shared_perm)
869 {
870 /*
871 * We are going to unlock bytes, it should not fail. If it fail due
872 * to some fs-dependent permission-unrelated reasons (which occurs
873 * sometimes on NFS and leads to abort in bdrv_replace_child) we
874 * can't prevent such errors by any check here. And we ignore them
875 * anyway in ABORT and COMMIT.
876 */
877 return 0;
878 }
879 ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm,
880 ~s->shared_perm | ~new_shared,
881 false, errp);
882 if (!ret) {
883 ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp);
884 if (!ret) {
885 return 0;
886 }
887 error_append_hint(errp,
888 "Is another process using the image [%s]?\n",
889 bs->filename);
890 }
891 /* fall through to unlock bytes. */
892 case RAW_PL_ABORT:
893 raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm,
894 true, &local_err);
895 if (local_err) {
896 /* Theoretically the above call only unlocks bytes and it cannot
897 * fail. Something weird happened, report it.
898 */
899 warn_report_err(local_err);
900 }
901 break;
902 case RAW_PL_COMMIT:
903 raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared,
904 true, &local_err);
905 if (local_err) {
906 /* Theoretically the above call only unlocks bytes and it cannot
907 * fail. Something weird happened, report it.
908 */
909 warn_report_err(local_err);
910 }
911 break;
912 }
913 return ret;
914 }
915
raw_reconfigure_getfd(BlockDriverState * bs,int flags,int * open_flags,uint64_t perm,bool force_dup,Error ** errp)916 static int raw_reconfigure_getfd(BlockDriverState *bs, int flags,
917 int *open_flags, uint64_t perm, bool force_dup,
918 Error **errp)
919 {
920 BDRVRawState *s = bs->opaque;
921 int fd = -1;
922 int ret;
923 bool has_writers = perm &
924 (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE);
925 int fcntl_flags = O_APPEND | O_NONBLOCK;
926 #ifdef O_NOATIME
927 fcntl_flags |= O_NOATIME;
928 #endif
929
930 *open_flags = 0;
931 if (s->type == FTYPE_CD) {
932 *open_flags |= O_NONBLOCK;
933 }
934
935 raw_parse_flags(flags, open_flags, has_writers);
936
937 #ifdef O_ASYNC
938 /* Not all operating systems have O_ASYNC, and those that don't
939 * will not let us track the state into rs->open_flags (typically
940 * you achieve the same effect with an ioctl, for example I_SETSIG
941 * on Solaris). But we do not use O_ASYNC, so that's fine.
942 */
943 assert((s->open_flags & O_ASYNC) == 0);
944 #endif
945
946 if (!force_dup && *open_flags == s->open_flags) {
947 /* We're lucky, the existing fd is fine */
948 return s->fd;
949 }
950
951 if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
952 /* dup the original fd */
953 fd = qemu_dup(s->fd);
954 if (fd >= 0) {
955 ret = fcntl_setfl(fd, *open_flags);
956 if (ret) {
957 qemu_close(fd);
958 fd = -1;
959 }
960 }
961 }
962
963 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
964 if (fd == -1) {
965 const char *normalized_filename = bs->filename;
966 ret = raw_normalize_devicepath(&normalized_filename, errp);
967 if (ret >= 0) {
968 assert(!(*open_flags & O_CREAT));
969 fd = qemu_open(normalized_filename, *open_flags);
970 if (fd == -1) {
971 error_setg_errno(errp, errno, "Could not reopen file");
972 return -1;
973 }
974 }
975 }
976
977 return fd;
978 }
979
raw_reopen_prepare(BDRVReopenState * state,BlockReopenQueue * queue,Error ** errp)980 static int raw_reopen_prepare(BDRVReopenState *state,
981 BlockReopenQueue *queue, Error **errp)
982 {
983 BDRVRawState *s;
984 BDRVRawReopenState *rs;
985 QemuOpts *opts;
986 int ret;
987 Error *local_err = NULL;
988
989 assert(state != NULL);
990 assert(state->bs != NULL);
991
992 s = state->bs->opaque;
993
994 state->opaque = g_new0(BDRVRawReopenState, 1);
995 rs = state->opaque;
996
997 /* Handle options changes */
998 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
999 qemu_opts_absorb_qdict(opts, state->options, &local_err);
1000 if (local_err) {
1001 error_propagate(errp, local_err);
1002 ret = -EINVAL;
1003 goto out;
1004 }
1005
1006 rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true);
1007 rs->check_cache_dropped =
1008 qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false);
1009
1010 /* This driver's reopen function doesn't currently allow changing
1011 * other options, so let's put them back in the original QDict and
1012 * bdrv_reopen_prepare() will detect changes and complain. */
1013 qemu_opts_to_qdict(opts, state->options);
1014
1015 rs->fd = raw_reconfigure_getfd(state->bs, state->flags, &rs->open_flags,
1016 state->perm, true, &local_err);
1017 if (local_err) {
1018 error_propagate(errp, local_err);
1019 ret = -1;
1020 goto out;
1021 }
1022
1023 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
1024 * alignment with the new fd. */
1025 if (rs->fd != -1) {
1026 raw_probe_alignment(state->bs, rs->fd, &local_err);
1027 if (local_err) {
1028 error_propagate(errp, local_err);
1029 ret = -EINVAL;
1030 goto out_fd;
1031 }
1032 }
1033
1034 s->reopen_state = state;
1035 ret = 0;
1036 out_fd:
1037 if (ret < 0) {
1038 qemu_close(rs->fd);
1039 rs->fd = -1;
1040 }
1041 out:
1042 qemu_opts_del(opts);
1043 return ret;
1044 }
1045
raw_reopen_commit(BDRVReopenState * state)1046 static void raw_reopen_commit(BDRVReopenState *state)
1047 {
1048 BDRVRawReopenState *rs = state->opaque;
1049 BDRVRawState *s = state->bs->opaque;
1050
1051 s->drop_cache = rs->drop_cache;
1052 s->check_cache_dropped = rs->check_cache_dropped;
1053 s->open_flags = rs->open_flags;
1054
1055 qemu_close(s->fd);
1056 s->fd = rs->fd;
1057
1058 g_free(state->opaque);
1059 state->opaque = NULL;
1060
1061 assert(s->reopen_state == state);
1062 s->reopen_state = NULL;
1063 }
1064
1065
raw_reopen_abort(BDRVReopenState * state)1066 static void raw_reopen_abort(BDRVReopenState *state)
1067 {
1068 BDRVRawReopenState *rs = state->opaque;
1069 BDRVRawState *s = state->bs->opaque;
1070
1071 /* nothing to do if NULL, we didn't get far enough */
1072 if (rs == NULL) {
1073 return;
1074 }
1075
1076 if (rs->fd >= 0) {
1077 qemu_close(rs->fd);
1078 rs->fd = -1;
1079 }
1080 g_free(state->opaque);
1081 state->opaque = NULL;
1082
1083 assert(s->reopen_state == state);
1084 s->reopen_state = NULL;
1085 }
1086
sg_get_max_transfer_length(int fd)1087 static int sg_get_max_transfer_length(int fd)
1088 {
1089 #ifdef BLKSECTGET
1090 int max_bytes = 0;
1091
1092 if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) {
1093 return max_bytes;
1094 } else {
1095 return -errno;
1096 }
1097 #else
1098 return -ENOSYS;
1099 #endif
1100 }
1101
sg_get_max_segments(int fd)1102 static int sg_get_max_segments(int fd)
1103 {
1104 #ifdef CONFIG_LINUX
1105 char buf[32];
1106 const char *end;
1107 char *sysfspath = NULL;
1108 int ret;
1109 int sysfd = -1;
1110 long max_segments;
1111 struct stat st;
1112
1113 if (fstat(fd, &st)) {
1114 ret = -errno;
1115 goto out;
1116 }
1117
1118 sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1119 major(st.st_rdev), minor(st.st_rdev));
1120 sysfd = open(sysfspath, O_RDONLY);
1121 if (sysfd == -1) {
1122 ret = -errno;
1123 goto out;
1124 }
1125 do {
1126 ret = read(sysfd, buf, sizeof(buf) - 1);
1127 } while (ret == -1 && errno == EINTR);
1128 if (ret < 0) {
1129 ret = -errno;
1130 goto out;
1131 } else if (ret == 0) {
1132 ret = -EIO;
1133 goto out;
1134 }
1135 buf[ret] = 0;
1136 /* The file is ended with '\n', pass 'end' to accept that. */
1137 ret = qemu_strtol(buf, &end, 10, &max_segments);
1138 if (ret == 0 && end && *end == '\n') {
1139 ret = max_segments;
1140 }
1141
1142 out:
1143 if (sysfd != -1) {
1144 close(sysfd);
1145 }
1146 g_free(sysfspath);
1147 return ret;
1148 #else
1149 return -ENOTSUP;
1150 #endif
1151 }
1152
raw_refresh_limits(BlockDriverState * bs,Error ** errp)1153 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
1154 {
1155 BDRVRawState *s = bs->opaque;
1156
1157 if (bs->sg) {
1158 int ret = sg_get_max_transfer_length(s->fd);
1159
1160 if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) {
1161 bs->bl.max_transfer = pow2floor(ret);
1162 }
1163
1164 ret = sg_get_max_segments(s->fd);
1165 if (ret > 0) {
1166 bs->bl.max_transfer = MIN(bs->bl.max_transfer,
1167 ret * qemu_real_host_page_size);
1168 }
1169 }
1170
1171 raw_probe_alignment(bs, s->fd, errp);
1172 bs->bl.min_mem_alignment = s->buf_align;
1173 bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size);
1174 }
1175
check_for_dasd(int fd)1176 static int check_for_dasd(int fd)
1177 {
1178 #ifdef BIODASDINFO2
1179 struct dasd_information2_t info = {0};
1180
1181 return ioctl(fd, BIODASDINFO2, &info);
1182 #else
1183 return -1;
1184 #endif
1185 }
1186
1187 /**
1188 * Try to get @bs's logical and physical block size.
1189 * On success, store them in @bsz and return zero.
1190 * On failure, return negative errno.
1191 */
hdev_probe_blocksizes(BlockDriverState * bs,BlockSizes * bsz)1192 static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
1193 {
1194 BDRVRawState *s = bs->opaque;
1195 int ret;
1196
1197 /* If DASD, get blocksizes */
1198 if (check_for_dasd(s->fd) < 0) {
1199 return -ENOTSUP;
1200 }
1201 ret = probe_logical_blocksize(s->fd, &bsz->log);
1202 if (ret < 0) {
1203 return ret;
1204 }
1205 return probe_physical_blocksize(s->fd, &bsz->phys);
1206 }
1207
1208 /**
1209 * Try to get @bs's geometry: cyls, heads, sectors.
1210 * On success, store them in @geo and return 0.
1211 * On failure return -errno.
1212 * (Allows block driver to assign default geometry values that guest sees)
1213 */
1214 #ifdef __linux__
hdev_probe_geometry(BlockDriverState * bs,HDGeometry * geo)1215 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1216 {
1217 BDRVRawState *s = bs->opaque;
1218 struct hd_geometry ioctl_geo = {0};
1219
1220 /* If DASD, get its geometry */
1221 if (check_for_dasd(s->fd) < 0) {
1222 return -ENOTSUP;
1223 }
1224 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
1225 return -errno;
1226 }
1227 /* HDIO_GETGEO may return success even though geo contains zeros
1228 (e.g. certain multipath setups) */
1229 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
1230 return -ENOTSUP;
1231 }
1232 /* Do not return a geometry for partition */
1233 if (ioctl_geo.start != 0) {
1234 return -ENOTSUP;
1235 }
1236 geo->heads = ioctl_geo.heads;
1237 geo->sectors = ioctl_geo.sectors;
1238 geo->cylinders = ioctl_geo.cylinders;
1239
1240 return 0;
1241 }
1242 #else /* __linux__ */
hdev_probe_geometry(BlockDriverState * bs,HDGeometry * geo)1243 static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1244 {
1245 return -ENOTSUP;
1246 }
1247 #endif
1248
1249 #if defined(__linux__)
handle_aiocb_ioctl(void * opaque)1250 static int handle_aiocb_ioctl(void *opaque)
1251 {
1252 RawPosixAIOData *aiocb = opaque;
1253 int ret;
1254
1255 ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf);
1256 if (ret == -1) {
1257 return -errno;
1258 }
1259
1260 return 0;
1261 }
1262 #endif /* linux */
1263
handle_aiocb_flush(void * opaque)1264 static int handle_aiocb_flush(void *opaque)
1265 {
1266 RawPosixAIOData *aiocb = opaque;
1267 BDRVRawState *s = aiocb->bs->opaque;
1268 int ret;
1269
1270 if (s->page_cache_inconsistent) {
1271 return -EIO;
1272 }
1273
1274 ret = qemu_fdatasync(aiocb->aio_fildes);
1275 if (ret == -1) {
1276 /* There is no clear definition of the semantics of a failing fsync(),
1277 * so we may have to assume the worst. The sad truth is that this
1278 * assumption is correct for Linux. Some pages are now probably marked
1279 * clean in the page cache even though they are inconsistent with the
1280 * on-disk contents. The next fdatasync() call would succeed, but no
1281 * further writeback attempt will be made. We can't get back to a state
1282 * in which we know what is on disk (we would have to rewrite
1283 * everything that was touched since the last fdatasync() at least), so
1284 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1285 * really defined, I have little hope that other OSes are doing better.
1286 *
1287 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1288 * cache. */
1289 if ((s->open_flags & O_DIRECT) == 0) {
1290 s->page_cache_inconsistent = true;
1291 }
1292 return -errno;
1293 }
1294 return 0;
1295 }
1296
1297 #ifdef CONFIG_PREADV
1298
1299 static bool preadv_present = true;
1300
1301 static ssize_t
qemu_preadv(int fd,const struct iovec * iov,int nr_iov,off_t offset)1302 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1303 {
1304 return preadv(fd, iov, nr_iov, offset);
1305 }
1306
1307 static ssize_t
qemu_pwritev(int fd,const struct iovec * iov,int nr_iov,off_t offset)1308 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1309 {
1310 return pwritev(fd, iov, nr_iov, offset);
1311 }
1312
1313 #else
1314
1315 static bool preadv_present = false;
1316
1317 static ssize_t
qemu_preadv(int fd,const struct iovec * iov,int nr_iov,off_t offset)1318 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1319 {
1320 return -ENOSYS;
1321 }
1322
1323 static ssize_t
qemu_pwritev(int fd,const struct iovec * iov,int nr_iov,off_t offset)1324 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1325 {
1326 return -ENOSYS;
1327 }
1328
1329 #endif
1330
handle_aiocb_rw_vector(RawPosixAIOData * aiocb)1331 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
1332 {
1333 ssize_t len;
1334
1335 do {
1336 if (aiocb->aio_type & QEMU_AIO_WRITE)
1337 len = qemu_pwritev(aiocb->aio_fildes,
1338 aiocb->io.iov,
1339 aiocb->io.niov,
1340 aiocb->aio_offset);
1341 else
1342 len = qemu_preadv(aiocb->aio_fildes,
1343 aiocb->io.iov,
1344 aiocb->io.niov,
1345 aiocb->aio_offset);
1346 } while (len == -1 && errno == EINTR);
1347
1348 if (len == -1) {
1349 return -errno;
1350 }
1351 return len;
1352 }
1353
1354 /*
1355 * Read/writes the data to/from a given linear buffer.
1356 *
1357 * Returns the number of bytes handles or -errno in case of an error. Short
1358 * reads are only returned if the end of the file is reached.
1359 */
handle_aiocb_rw_linear(RawPosixAIOData * aiocb,char * buf)1360 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
1361 {
1362 ssize_t offset = 0;
1363 ssize_t len;
1364
1365 while (offset < aiocb->aio_nbytes) {
1366 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1367 len = pwrite(aiocb->aio_fildes,
1368 (const char *)buf + offset,
1369 aiocb->aio_nbytes - offset,
1370 aiocb->aio_offset + offset);
1371 } else {
1372 len = pread(aiocb->aio_fildes,
1373 buf + offset,
1374 aiocb->aio_nbytes - offset,
1375 aiocb->aio_offset + offset);
1376 }
1377 if (len == -1 && errno == EINTR) {
1378 continue;
1379 } else if (len == -1 && errno == EINVAL &&
1380 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
1381 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
1382 offset > 0) {
1383 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1384 * after a short read. Assume that O_DIRECT short reads only occur
1385 * at EOF. Therefore this is a short read, not an I/O error.
1386 */
1387 break;
1388 } else if (len == -1) {
1389 offset = -errno;
1390 break;
1391 } else if (len == 0) {
1392 break;
1393 }
1394 offset += len;
1395 }
1396
1397 return offset;
1398 }
1399
handle_aiocb_rw(void * opaque)1400 static int handle_aiocb_rw(void *opaque)
1401 {
1402 RawPosixAIOData *aiocb = opaque;
1403 ssize_t nbytes;
1404 char *buf;
1405
1406 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
1407 /*
1408 * If there is just a single buffer, and it is properly aligned
1409 * we can just use plain pread/pwrite without any problems.
1410 */
1411 if (aiocb->io.niov == 1) {
1412 nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base);
1413 goto out;
1414 }
1415 /*
1416 * We have more than one iovec, and all are properly aligned.
1417 *
1418 * Try preadv/pwritev first and fall back to linearizing the
1419 * buffer if it's not supported.
1420 */
1421 if (preadv_present) {
1422 nbytes = handle_aiocb_rw_vector(aiocb);
1423 if (nbytes == aiocb->aio_nbytes ||
1424 (nbytes < 0 && nbytes != -ENOSYS)) {
1425 goto out;
1426 }
1427 preadv_present = false;
1428 }
1429
1430 /*
1431 * XXX(hch): short read/write. no easy way to handle the reminder
1432 * using these interfaces. For now retry using plain
1433 * pread/pwrite?
1434 */
1435 }
1436
1437 /*
1438 * Ok, we have to do it the hard way, copy all segments into
1439 * a single aligned buffer.
1440 */
1441 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
1442 if (buf == NULL) {
1443 nbytes = -ENOMEM;
1444 goto out;
1445 }
1446
1447 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1448 char *p = buf;
1449 int i;
1450
1451 for (i = 0; i < aiocb->io.niov; ++i) {
1452 memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len);
1453 p += aiocb->io.iov[i].iov_len;
1454 }
1455 assert(p - buf == aiocb->aio_nbytes);
1456 }
1457
1458 nbytes = handle_aiocb_rw_linear(aiocb, buf);
1459 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
1460 char *p = buf;
1461 size_t count = aiocb->aio_nbytes, copy;
1462 int i;
1463
1464 for (i = 0; i < aiocb->io.niov && count; ++i) {
1465 copy = count;
1466 if (copy > aiocb->io.iov[i].iov_len) {
1467 copy = aiocb->io.iov[i].iov_len;
1468 }
1469 memcpy(aiocb->io.iov[i].iov_base, p, copy);
1470 assert(count >= copy);
1471 p += copy;
1472 count -= copy;
1473 }
1474 assert(count == 0);
1475 }
1476 qemu_vfree(buf);
1477
1478 out:
1479 if (nbytes == aiocb->aio_nbytes) {
1480 return 0;
1481 } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) {
1482 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1483 return -EINVAL;
1484 } else {
1485 iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes,
1486 0, aiocb->aio_nbytes - nbytes);
1487 return 0;
1488 }
1489 } else {
1490 assert(nbytes < 0);
1491 return nbytes;
1492 }
1493 }
1494
translate_err(int err)1495 static int translate_err(int err)
1496 {
1497 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1498 err == -ENOTTY) {
1499 err = -ENOTSUP;
1500 }
1501 return err;
1502 }
1503
1504 #ifdef CONFIG_FALLOCATE
do_fallocate(int fd,int mode,off_t offset,off_t len)1505 static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1506 {
1507 do {
1508 if (fallocate(fd, mode, offset, len) == 0) {
1509 return 0;
1510 }
1511 } while (errno == EINTR);
1512 return translate_err(-errno);
1513 }
1514 #endif
1515
handle_aiocb_write_zeroes_block(RawPosixAIOData * aiocb)1516 static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
1517 {
1518 int ret = -ENOTSUP;
1519 BDRVRawState *s = aiocb->bs->opaque;
1520
1521 if (!s->has_write_zeroes) {
1522 return -ENOTSUP;
1523 }
1524
1525 #ifdef BLKZEROOUT
1526 /* The BLKZEROOUT implementation in the kernel doesn't set
1527 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1528 * fallbacks. */
1529 if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) {
1530 do {
1531 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1532 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1533 return 0;
1534 }
1535 } while (errno == EINTR);
1536
1537 ret = translate_err(-errno);
1538 if (ret == -ENOTSUP) {
1539 s->has_write_zeroes = false;
1540 }
1541 }
1542 #endif
1543
1544 return ret;
1545 }
1546
handle_aiocb_write_zeroes(void * opaque)1547 static int handle_aiocb_write_zeroes(void *opaque)
1548 {
1549 RawPosixAIOData *aiocb = opaque;
1550 #ifdef CONFIG_FALLOCATE
1551 BDRVRawState *s = aiocb->bs->opaque;
1552 int64_t len;
1553 #endif
1554
1555 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1556 return handle_aiocb_write_zeroes_block(aiocb);
1557 }
1558
1559 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1560 if (s->has_write_zeroes) {
1561 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1562 aiocb->aio_offset, aiocb->aio_nbytes);
1563 if (ret == -EINVAL) {
1564 /*
1565 * Allow falling back to pwrite for file systems that
1566 * do not support fallocate() for an unaligned byte range.
1567 */
1568 return -ENOTSUP;
1569 }
1570 if (ret == 0 || ret != -ENOTSUP) {
1571 return ret;
1572 }
1573 s->has_write_zeroes = false;
1574 }
1575 #endif
1576
1577 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1578 if (s->has_discard && s->has_fallocate) {
1579 int ret = do_fallocate(s->fd,
1580 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1581 aiocb->aio_offset, aiocb->aio_nbytes);
1582 if (ret == 0) {
1583 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1584 if (ret == 0 || ret != -ENOTSUP) {
1585 return ret;
1586 }
1587 s->has_fallocate = false;
1588 } else if (ret != -ENOTSUP) {
1589 return ret;
1590 } else {
1591 s->has_discard = false;
1592 }
1593 }
1594 #endif
1595
1596 #ifdef CONFIG_FALLOCATE
1597 /* Last resort: we are trying to extend the file with zeroed data. This
1598 * can be done via fallocate(fd, 0) */
1599 len = bdrv_getlength(aiocb->bs);
1600 if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) {
1601 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1602 if (ret == 0 || ret != -ENOTSUP) {
1603 return ret;
1604 }
1605 s->has_fallocate = false;
1606 }
1607 #endif
1608
1609 return -ENOTSUP;
1610 }
1611
handle_aiocb_write_zeroes_unmap(void * opaque)1612 static int handle_aiocb_write_zeroes_unmap(void *opaque)
1613 {
1614 RawPosixAIOData *aiocb = opaque;
1615 BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque;
1616 int ret;
1617
1618 /* First try to write zeros and unmap at the same time */
1619
1620 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1621 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1622 aiocb->aio_offset, aiocb->aio_nbytes);
1623 if (ret != -ENOTSUP) {
1624 return ret;
1625 }
1626 #endif
1627
1628 /* If we couldn't manage to unmap while guaranteed that the area reads as
1629 * all-zero afterwards, just write zeroes without unmapping */
1630 ret = handle_aiocb_write_zeroes(aiocb);
1631 return ret;
1632 }
1633
1634 #ifndef HAVE_COPY_FILE_RANGE
copy_file_range(int in_fd,off_t * in_off,int out_fd,off_t * out_off,size_t len,unsigned int flags)1635 static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd,
1636 off_t *out_off, size_t len, unsigned int flags)
1637 {
1638 #ifdef __NR_copy_file_range
1639 return syscall(__NR_copy_file_range, in_fd, in_off, out_fd,
1640 out_off, len, flags);
1641 #else
1642 errno = ENOSYS;
1643 return -1;
1644 #endif
1645 }
1646 #endif
1647
handle_aiocb_copy_range(void * opaque)1648 static int handle_aiocb_copy_range(void *opaque)
1649 {
1650 RawPosixAIOData *aiocb = opaque;
1651 uint64_t bytes = aiocb->aio_nbytes;
1652 off_t in_off = aiocb->aio_offset;
1653 off_t out_off = aiocb->copy_range.aio_offset2;
1654
1655 while (bytes) {
1656 ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off,
1657 aiocb->copy_range.aio_fd2, &out_off,
1658 bytes, 0);
1659 trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off,
1660 aiocb->copy_range.aio_fd2, out_off, bytes,
1661 0, ret);
1662 if (ret == 0) {
1663 /* No progress (e.g. when beyond EOF), let the caller fall back to
1664 * buffer I/O. */
1665 return -ENOSPC;
1666 }
1667 if (ret < 0) {
1668 switch (errno) {
1669 case ENOSYS:
1670 return -ENOTSUP;
1671 case EINTR:
1672 continue;
1673 default:
1674 return -errno;
1675 }
1676 }
1677 bytes -= ret;
1678 }
1679 return 0;
1680 }
1681
handle_aiocb_discard(void * opaque)1682 static int handle_aiocb_discard(void *opaque)
1683 {
1684 RawPosixAIOData *aiocb = opaque;
1685 int ret = -EOPNOTSUPP;
1686 BDRVRawState *s = aiocb->bs->opaque;
1687
1688 if (!s->has_discard) {
1689 return -ENOTSUP;
1690 }
1691
1692 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1693 #ifdef BLKDISCARD
1694 do {
1695 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1696 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1697 return 0;
1698 }
1699 } while (errno == EINTR);
1700
1701 ret = -errno;
1702 #endif
1703 } else {
1704 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1705 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1706 aiocb->aio_offset, aiocb->aio_nbytes);
1707 #endif
1708 }
1709
1710 ret = translate_err(ret);
1711 if (ret == -ENOTSUP) {
1712 s->has_discard = false;
1713 }
1714 return ret;
1715 }
1716
1717 /*
1718 * Help alignment probing by allocating the first block.
1719 *
1720 * When reading with direct I/O from unallocated area on Gluster backed by XFS,
1721 * reading succeeds regardless of request length. In this case we fallback to
1722 * safe alignment which is not optimal. Allocating the first block avoids this
1723 * fallback.
1724 *
1725 * fd may be opened with O_DIRECT, but we don't know the buffer alignment or
1726 * request alignment, so we use safe values.
1727 *
1728 * Returns: 0 on success, -errno on failure. Since this is an optimization,
1729 * caller may ignore failures.
1730 */
allocate_first_block(int fd,size_t max_size)1731 static int allocate_first_block(int fd, size_t max_size)
1732 {
1733 size_t write_size = (max_size < MAX_BLOCKSIZE)
1734 ? BDRV_SECTOR_SIZE
1735 : MAX_BLOCKSIZE;
1736 size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size);
1737 void *buf;
1738 ssize_t n;
1739 int ret;
1740
1741 buf = qemu_memalign(max_align, write_size);
1742 memset(buf, 0, write_size);
1743
1744 do {
1745 n = pwrite(fd, buf, write_size, 0);
1746 } while (n == -1 && errno == EINTR);
1747
1748 ret = (n == -1) ? -errno : 0;
1749
1750 qemu_vfree(buf);
1751 return ret;
1752 }
1753
handle_aiocb_truncate(void * opaque)1754 static int handle_aiocb_truncate(void *opaque)
1755 {
1756 RawPosixAIOData *aiocb = opaque;
1757 int result = 0;
1758 int64_t current_length = 0;
1759 char *buf = NULL;
1760 struct stat st;
1761 int fd = aiocb->aio_fildes;
1762 int64_t offset = aiocb->aio_offset;
1763 PreallocMode prealloc = aiocb->truncate.prealloc;
1764 Error **errp = aiocb->truncate.errp;
1765
1766 if (fstat(fd, &st) < 0) {
1767 result = -errno;
1768 error_setg_errno(errp, -result, "Could not stat file");
1769 return result;
1770 }
1771
1772 current_length = st.st_size;
1773 if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
1774 error_setg(errp, "Cannot use preallocation for shrinking files");
1775 return -ENOTSUP;
1776 }
1777
1778 switch (prealloc) {
1779 #ifdef CONFIG_POSIX_FALLOCATE
1780 case PREALLOC_MODE_FALLOC:
1781 /*
1782 * Truncating before posix_fallocate() makes it about twice slower on
1783 * file systems that do not support fallocate(), trying to check if a
1784 * block is allocated before allocating it, so don't do that here.
1785 */
1786 if (offset != current_length) {
1787 result = -posix_fallocate(fd, current_length,
1788 offset - current_length);
1789 if (result != 0) {
1790 /* posix_fallocate() doesn't set errno. */
1791 error_setg_errno(errp, -result,
1792 "Could not preallocate new data");
1793 } else if (current_length == 0) {
1794 /*
1795 * posix_fallocate() uses fallocate() if the filesystem
1796 * supports it, or fallback to manually writing zeroes. If
1797 * fallocate() was used, unaligned reads from the fallocated
1798 * area in raw_probe_alignment() will succeed, hence we need to
1799 * allocate the first block.
1800 *
1801 * Optimize future alignment probing; ignore failures.
1802 */
1803 allocate_first_block(fd, offset);
1804 }
1805 } else {
1806 result = 0;
1807 }
1808 goto out;
1809 #endif
1810 case PREALLOC_MODE_FULL:
1811 {
1812 int64_t num = 0, left = offset - current_length;
1813 off_t seek_result;
1814
1815 /*
1816 * Knowing the final size from the beginning could allow the file
1817 * system driver to do less allocations and possibly avoid
1818 * fragmentation of the file.
1819 */
1820 if (ftruncate(fd, offset) != 0) {
1821 result = -errno;
1822 error_setg_errno(errp, -result, "Could not resize file");
1823 goto out;
1824 }
1825
1826 buf = g_malloc0(65536);
1827
1828 seek_result = lseek(fd, current_length, SEEK_SET);
1829 if (seek_result < 0) {
1830 result = -errno;
1831 error_setg_errno(errp, -result,
1832 "Failed to seek to the old end of file");
1833 goto out;
1834 }
1835
1836 while (left > 0) {
1837 num = MIN(left, 65536);
1838 result = write(fd, buf, num);
1839 if (result < 0) {
1840 if (errno == EINTR) {
1841 continue;
1842 }
1843 result = -errno;
1844 error_setg_errno(errp, -result,
1845 "Could not write zeros for preallocation");
1846 goto out;
1847 }
1848 left -= result;
1849 }
1850 if (result >= 0) {
1851 result = fsync(fd);
1852 if (result < 0) {
1853 result = -errno;
1854 error_setg_errno(errp, -result,
1855 "Could not flush file to disk");
1856 goto out;
1857 }
1858 }
1859 goto out;
1860 }
1861 case PREALLOC_MODE_OFF:
1862 if (ftruncate(fd, offset) != 0) {
1863 result = -errno;
1864 error_setg_errno(errp, -result, "Could not resize file");
1865 } else if (current_length == 0 && offset > current_length) {
1866 /* Optimize future alignment probing; ignore failures. */
1867 allocate_first_block(fd, offset);
1868 }
1869 return result;
1870 default:
1871 result = -ENOTSUP;
1872 error_setg(errp, "Unsupported preallocation mode: %s",
1873 PreallocMode_str(prealloc));
1874 return result;
1875 }
1876
1877 out:
1878 if (result < 0) {
1879 if (ftruncate(fd, current_length) < 0) {
1880 error_report("Failed to restore old file length: %s",
1881 strerror(errno));
1882 }
1883 }
1884
1885 g_free(buf);
1886 return result;
1887 }
1888
raw_thread_pool_submit(BlockDriverState * bs,ThreadPoolFunc func,void * arg)1889 static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs,
1890 ThreadPoolFunc func, void *arg)
1891 {
1892 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1893 ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1894 return thread_pool_submit_co(pool, func, arg);
1895 }
1896
raw_co_prw(BlockDriverState * bs,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov,int type)1897 static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset,
1898 uint64_t bytes, QEMUIOVector *qiov, int type)
1899 {
1900 BDRVRawState *s = bs->opaque;
1901 RawPosixAIOData acb;
1902
1903 if (fd_open(bs) < 0)
1904 return -EIO;
1905
1906 /*
1907 * When using O_DIRECT, the request must be aligned to be able to use
1908 * either libaio or io_uring interface. If not fail back to regular thread
1909 * pool read/write code which emulates this for us if we
1910 * set QEMU_AIO_MISALIGNED.
1911 */
1912 if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) {
1913 type |= QEMU_AIO_MISALIGNED;
1914 #ifdef CONFIG_LINUX_IO_URING
1915 } else if (s->use_linux_io_uring) {
1916 LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs));
1917 assert(qiov->size == bytes);
1918 return luring_co_submit(bs, aio, s->fd, offset, qiov, type);
1919 #endif
1920 #ifdef CONFIG_LINUX_AIO
1921 } else if (s->use_linux_aio) {
1922 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1923 assert(qiov->size == bytes);
1924 return laio_co_submit(bs, aio, s->fd, offset, qiov, type);
1925 #endif
1926 }
1927
1928 acb = (RawPosixAIOData) {
1929 .bs = bs,
1930 .aio_fildes = s->fd,
1931 .aio_type = type,
1932 .aio_offset = offset,
1933 .aio_nbytes = bytes,
1934 .io = {
1935 .iov = qiov->iov,
1936 .niov = qiov->niov,
1937 },
1938 };
1939
1940 assert(qiov->size == bytes);
1941 return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb);
1942 }
1943
raw_co_preadv(BlockDriverState * bs,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov,int flags)1944 static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset,
1945 uint64_t bytes, QEMUIOVector *qiov,
1946 int flags)
1947 {
1948 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ);
1949 }
1950
raw_co_pwritev(BlockDriverState * bs,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov,int flags)1951 static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
1952 uint64_t bytes, QEMUIOVector *qiov,
1953 int flags)
1954 {
1955 assert(flags == 0);
1956 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE);
1957 }
1958
raw_aio_plug(BlockDriverState * bs)1959 static void raw_aio_plug(BlockDriverState *bs)
1960 {
1961 BDRVRawState __attribute__((unused)) *s = bs->opaque;
1962 #ifdef CONFIG_LINUX_AIO
1963 if (s->use_linux_aio) {
1964 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1965 laio_io_plug(bs, aio);
1966 }
1967 #endif
1968 #ifdef CONFIG_LINUX_IO_URING
1969 if (s->use_linux_io_uring) {
1970 LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs));
1971 luring_io_plug(bs, aio);
1972 }
1973 #endif
1974 }
1975
raw_aio_unplug(BlockDriverState * bs)1976 static void raw_aio_unplug(BlockDriverState *bs)
1977 {
1978 BDRVRawState __attribute__((unused)) *s = bs->opaque;
1979 #ifdef CONFIG_LINUX_AIO
1980 if (s->use_linux_aio) {
1981 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1982 laio_io_unplug(bs, aio);
1983 }
1984 #endif
1985 #ifdef CONFIG_LINUX_IO_URING
1986 if (s->use_linux_io_uring) {
1987 LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs));
1988 luring_io_unplug(bs, aio);
1989 }
1990 #endif
1991 }
1992
raw_co_flush_to_disk(BlockDriverState * bs)1993 static int raw_co_flush_to_disk(BlockDriverState *bs)
1994 {
1995 BDRVRawState *s = bs->opaque;
1996 RawPosixAIOData acb;
1997 int ret;
1998
1999 ret = fd_open(bs);
2000 if (ret < 0) {
2001 return ret;
2002 }
2003
2004 acb = (RawPosixAIOData) {
2005 .bs = bs,
2006 .aio_fildes = s->fd,
2007 .aio_type = QEMU_AIO_FLUSH,
2008 };
2009
2010 #ifdef CONFIG_LINUX_IO_URING
2011 if (s->use_linux_io_uring) {
2012 LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs));
2013 return luring_co_submit(bs, aio, s->fd, 0, NULL, QEMU_AIO_FLUSH);
2014 }
2015 #endif
2016 return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb);
2017 }
2018
raw_aio_attach_aio_context(BlockDriverState * bs,AioContext * new_context)2019 static void raw_aio_attach_aio_context(BlockDriverState *bs,
2020 AioContext *new_context)
2021 {
2022 BDRVRawState __attribute__((unused)) *s = bs->opaque;
2023 #ifdef CONFIG_LINUX_AIO
2024 if (s->use_linux_aio) {
2025 Error *local_err = NULL;
2026 if (!aio_setup_linux_aio(new_context, &local_err)) {
2027 error_reportf_err(local_err, "Unable to use native AIO, "
2028 "falling back to thread pool: ");
2029 s->use_linux_aio = false;
2030 }
2031 }
2032 #endif
2033 #ifdef CONFIG_LINUX_IO_URING
2034 if (s->use_linux_io_uring) {
2035 Error *local_err;
2036 if (!aio_setup_linux_io_uring(new_context, &local_err)) {
2037 error_reportf_err(local_err, "Unable to use linux io_uring, "
2038 "falling back to thread pool: ");
2039 s->use_linux_io_uring = false;
2040 }
2041 }
2042 #endif
2043 }
2044
raw_close(BlockDriverState * bs)2045 static void raw_close(BlockDriverState *bs)
2046 {
2047 BDRVRawState *s = bs->opaque;
2048
2049 if (s->fd >= 0) {
2050 qemu_close(s->fd);
2051 s->fd = -1;
2052 }
2053 }
2054
2055 /**
2056 * Truncates the given regular file @fd to @offset and, when growing, fills the
2057 * new space according to @prealloc.
2058 *
2059 * Returns: 0 on success, -errno on failure.
2060 */
2061 static int coroutine_fn
raw_regular_truncate(BlockDriverState * bs,int fd,int64_t offset,PreallocMode prealloc,Error ** errp)2062 raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset,
2063 PreallocMode prealloc, Error **errp)
2064 {
2065 RawPosixAIOData acb;
2066
2067 acb = (RawPosixAIOData) {
2068 .bs = bs,
2069 .aio_fildes = fd,
2070 .aio_type = QEMU_AIO_TRUNCATE,
2071 .aio_offset = offset,
2072 .truncate = {
2073 .prealloc = prealloc,
2074 .errp = errp,
2075 },
2076 };
2077
2078 return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb);
2079 }
2080
raw_co_truncate(BlockDriverState * bs,int64_t offset,bool exact,PreallocMode prealloc,Error ** errp)2081 static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset,
2082 bool exact, PreallocMode prealloc,
2083 Error **errp)
2084 {
2085 BDRVRawState *s = bs->opaque;
2086 struct stat st;
2087 int ret;
2088
2089 if (fstat(s->fd, &st)) {
2090 ret = -errno;
2091 error_setg_errno(errp, -ret, "Failed to fstat() the file");
2092 return ret;
2093 }
2094
2095 if (S_ISREG(st.st_mode)) {
2096 /* Always resizes to the exact @offset */
2097 return raw_regular_truncate(bs, s->fd, offset, prealloc, errp);
2098 }
2099
2100 if (prealloc != PREALLOC_MODE_OFF) {
2101 error_setg(errp, "Preallocation mode '%s' unsupported for this "
2102 "non-regular file", PreallocMode_str(prealloc));
2103 return -ENOTSUP;
2104 }
2105
2106 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2107 int64_t cur_length = raw_getlength(bs);
2108
2109 if (offset != cur_length && exact) {
2110 error_setg(errp, "Cannot resize device files");
2111 return -ENOTSUP;
2112 } else if (offset > cur_length) {
2113 error_setg(errp, "Cannot grow device files");
2114 return -EINVAL;
2115 }
2116 } else {
2117 error_setg(errp, "Resizing this file is not supported");
2118 return -ENOTSUP;
2119 }
2120
2121 return 0;
2122 }
2123
2124 #ifdef __OpenBSD__
raw_getlength(BlockDriverState * bs)2125 static int64_t raw_getlength(BlockDriverState *bs)
2126 {
2127 BDRVRawState *s = bs->opaque;
2128 int fd = s->fd;
2129 struct stat st;
2130
2131 if (fstat(fd, &st))
2132 return -errno;
2133 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2134 struct disklabel dl;
2135
2136 if (ioctl(fd, DIOCGDINFO, &dl))
2137 return -errno;
2138 return (uint64_t)dl.d_secsize *
2139 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
2140 } else
2141 return st.st_size;
2142 }
2143 #elif defined(__NetBSD__)
raw_getlength(BlockDriverState * bs)2144 static int64_t raw_getlength(BlockDriverState *bs)
2145 {
2146 BDRVRawState *s = bs->opaque;
2147 int fd = s->fd;
2148 struct stat st;
2149
2150 if (fstat(fd, &st))
2151 return -errno;
2152 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2153 struct dkwedge_info dkw;
2154
2155 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
2156 return dkw.dkw_size * 512;
2157 } else {
2158 struct disklabel dl;
2159
2160 if (ioctl(fd, DIOCGDINFO, &dl))
2161 return -errno;
2162 return (uint64_t)dl.d_secsize *
2163 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
2164 }
2165 } else
2166 return st.st_size;
2167 }
2168 #elif defined(__sun__)
raw_getlength(BlockDriverState * bs)2169 static int64_t raw_getlength(BlockDriverState *bs)
2170 {
2171 BDRVRawState *s = bs->opaque;
2172 struct dk_minfo minfo;
2173 int ret;
2174 int64_t size;
2175
2176 ret = fd_open(bs);
2177 if (ret < 0) {
2178 return ret;
2179 }
2180
2181 /*
2182 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2183 */
2184 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
2185 if (ret != -1) {
2186 return minfo.dki_lbsize * minfo.dki_capacity;
2187 }
2188
2189 /*
2190 * There are reports that lseek on some devices fails, but
2191 * irc discussion said that contingency on contingency was overkill.
2192 */
2193 size = lseek(s->fd, 0, SEEK_END);
2194 if (size < 0) {
2195 return -errno;
2196 }
2197 return size;
2198 }
2199 #elif defined(CONFIG_BSD)
raw_getlength(BlockDriverState * bs)2200 static int64_t raw_getlength(BlockDriverState *bs)
2201 {
2202 BDRVRawState *s = bs->opaque;
2203 int fd = s->fd;
2204 int64_t size;
2205 struct stat sb;
2206 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2207 int reopened = 0;
2208 #endif
2209 int ret;
2210
2211 ret = fd_open(bs);
2212 if (ret < 0)
2213 return ret;
2214
2215 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2216 again:
2217 #endif
2218 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
2219 #ifdef DIOCGMEDIASIZE
2220 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
2221 #elif defined(DIOCGPART)
2222 {
2223 struct partinfo pi;
2224 if (ioctl(fd, DIOCGPART, &pi) == 0)
2225 size = pi.media_size;
2226 else
2227 size = 0;
2228 }
2229 if (size == 0)
2230 #endif
2231 #if defined(__APPLE__) && defined(__MACH__)
2232 {
2233 uint64_t sectors = 0;
2234 uint32_t sector_size = 0;
2235
2236 if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0
2237 && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) {
2238 size = sectors * sector_size;
2239 } else {
2240 size = lseek(fd, 0LL, SEEK_END);
2241 if (size < 0) {
2242 return -errno;
2243 }
2244 }
2245 }
2246 #else
2247 size = lseek(fd, 0LL, SEEK_END);
2248 if (size < 0) {
2249 return -errno;
2250 }
2251 #endif
2252 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2253 switch(s->type) {
2254 case FTYPE_CD:
2255 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2256 if (size == 2048LL * (unsigned)-1)
2257 size = 0;
2258 /* XXX no disc? maybe we need to reopen... */
2259 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
2260 reopened = 1;
2261 goto again;
2262 }
2263 }
2264 #endif
2265 } else {
2266 size = lseek(fd, 0, SEEK_END);
2267 if (size < 0) {
2268 return -errno;
2269 }
2270 }
2271 return size;
2272 }
2273 #else
raw_getlength(BlockDriverState * bs)2274 static int64_t raw_getlength(BlockDriverState *bs)
2275 {
2276 BDRVRawState *s = bs->opaque;
2277 int ret;
2278 int64_t size;
2279
2280 ret = fd_open(bs);
2281 if (ret < 0) {
2282 return ret;
2283 }
2284
2285 size = lseek(s->fd, 0, SEEK_END);
2286 if (size < 0) {
2287 return -errno;
2288 }
2289 return size;
2290 }
2291 #endif
2292
raw_get_allocated_file_size(BlockDriverState * bs)2293 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
2294 {
2295 struct stat st;
2296 BDRVRawState *s = bs->opaque;
2297
2298 if (fstat(s->fd, &st) < 0) {
2299 return -errno;
2300 }
2301 return (int64_t)st.st_blocks * 512;
2302 }
2303
2304 static int coroutine_fn
raw_co_create(BlockdevCreateOptions * options,Error ** errp)2305 raw_co_create(BlockdevCreateOptions *options, Error **errp)
2306 {
2307 BlockdevCreateOptionsFile *file_opts;
2308 Error *local_err = NULL;
2309 int fd;
2310 uint64_t perm, shared;
2311 int result = 0;
2312
2313 /* Validate options and set default values */
2314 assert(options->driver == BLOCKDEV_DRIVER_FILE);
2315 file_opts = &options->u.file;
2316
2317 if (!file_opts->has_nocow) {
2318 file_opts->nocow = false;
2319 }
2320 if (!file_opts->has_preallocation) {
2321 file_opts->preallocation = PREALLOC_MODE_OFF;
2322 }
2323
2324 /* Create file */
2325 fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_BINARY, 0644);
2326 if (fd < 0) {
2327 result = -errno;
2328 error_setg_errno(errp, -result, "Could not create file");
2329 goto out;
2330 }
2331
2332 /* Take permissions: We want to discard everything, so we need
2333 * BLK_PERM_WRITE; and truncation to the desired size requires
2334 * BLK_PERM_RESIZE.
2335 * On the other hand, we cannot share the RESIZE permission
2336 * because we promise that after this function, the file has the
2337 * size given in the options. If someone else were to resize it
2338 * concurrently, we could not guarantee that.
2339 * Note that after this function, we can no longer guarantee that
2340 * the file is not touched by a third party, so it may be resized
2341 * then. */
2342 perm = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2343 shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE;
2344
2345 /* Step one: Take locks */
2346 result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp);
2347 if (result < 0) {
2348 goto out_close;
2349 }
2350
2351 /* Step two: Check that nobody else has taken conflicting locks */
2352 result = raw_check_lock_bytes(fd, perm, shared, errp);
2353 if (result < 0) {
2354 error_append_hint(errp,
2355 "Is another process using the image [%s]?\n",
2356 file_opts->filename);
2357 goto out_unlock;
2358 }
2359
2360 /* Clear the file by truncating it to 0 */
2361 result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp);
2362 if (result < 0) {
2363 goto out_unlock;
2364 }
2365
2366 if (file_opts->nocow) {
2367 #ifdef __linux__
2368 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2369 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2370 * will be ignored since any failure of this operation should not
2371 * block the left work.
2372 */
2373 int attr;
2374 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
2375 attr |= FS_NOCOW_FL;
2376 ioctl(fd, FS_IOC_SETFLAGS, &attr);
2377 }
2378 #endif
2379 }
2380
2381 /* Resize and potentially preallocate the file to the desired
2382 * final size */
2383 result = raw_regular_truncate(NULL, fd, file_opts->size,
2384 file_opts->preallocation, errp);
2385 if (result < 0) {
2386 goto out_unlock;
2387 }
2388
2389 out_unlock:
2390 raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err);
2391 if (local_err) {
2392 /* The above call should not fail, and if it does, that does
2393 * not mean the whole creation operation has failed. So
2394 * report it the user for their convenience, but do not report
2395 * it to the caller. */
2396 warn_report_err(local_err);
2397 }
2398
2399 out_close:
2400 if (qemu_close(fd) != 0 && result == 0) {
2401 result = -errno;
2402 error_setg_errno(errp, -result, "Could not close the new file");
2403 }
2404 out:
2405 return result;
2406 }
2407
raw_co_create_opts(const char * filename,QemuOpts * opts,Error ** errp)2408 static int coroutine_fn raw_co_create_opts(const char *filename, QemuOpts *opts,
2409 Error **errp)
2410 {
2411 BlockdevCreateOptions options;
2412 int64_t total_size = 0;
2413 bool nocow = false;
2414 PreallocMode prealloc;
2415 char *buf = NULL;
2416 Error *local_err = NULL;
2417
2418 /* Skip file: protocol prefix */
2419 strstart(filename, "file:", &filename);
2420
2421 /* Read out options */
2422 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2423 BDRV_SECTOR_SIZE);
2424 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
2425 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2426 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
2427 PREALLOC_MODE_OFF, &local_err);
2428 g_free(buf);
2429 if (local_err) {
2430 error_propagate(errp, local_err);
2431 return -EINVAL;
2432 }
2433
2434 options = (BlockdevCreateOptions) {
2435 .driver = BLOCKDEV_DRIVER_FILE,
2436 .u.file = {
2437 .filename = (char *) filename,
2438 .size = total_size,
2439 .has_preallocation = true,
2440 .preallocation = prealloc,
2441 .has_nocow = true,
2442 .nocow = nocow,
2443 },
2444 };
2445 return raw_co_create(&options, errp);
2446 }
2447
raw_co_delete_file(BlockDriverState * bs,Error ** errp)2448 static int coroutine_fn raw_co_delete_file(BlockDriverState *bs,
2449 Error **errp)
2450 {
2451 struct stat st;
2452 int ret;
2453
2454 if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) {
2455 error_setg_errno(errp, ENOENT, "%s is not a regular file",
2456 bs->filename);
2457 return -ENOENT;
2458 }
2459
2460 ret = unlink(bs->filename);
2461 if (ret < 0) {
2462 ret = -errno;
2463 error_setg_errno(errp, -ret, "Error when deleting file %s",
2464 bs->filename);
2465 }
2466
2467 return ret;
2468 }
2469
2470 /*
2471 * Find allocation range in @bs around offset @start.
2472 * May change underlying file descriptor's file offset.
2473 * If @start is not in a hole, store @start in @data, and the
2474 * beginning of the next hole in @hole, and return 0.
2475 * If @start is in a non-trailing hole, store @start in @hole and the
2476 * beginning of the next non-hole in @data, and return 0.
2477 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2478 * If we can't find out, return a negative errno other than -ENXIO.
2479 */
find_allocation(BlockDriverState * bs,off_t start,off_t * data,off_t * hole)2480 static int find_allocation(BlockDriverState *bs, off_t start,
2481 off_t *data, off_t *hole)
2482 {
2483 #if defined SEEK_HOLE && defined SEEK_DATA
2484 BDRVRawState *s = bs->opaque;
2485 off_t offs;
2486
2487 /*
2488 * SEEK_DATA cases:
2489 * D1. offs == start: start is in data
2490 * D2. offs > start: start is in a hole, next data at offs
2491 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2492 * or start is beyond EOF
2493 * If the latter happens, the file has been truncated behind
2494 * our back since we opened it. All bets are off then.
2495 * Treating like a trailing hole is simplest.
2496 * D4. offs < 0, errno != ENXIO: we learned nothing
2497 */
2498 offs = lseek(s->fd, start, SEEK_DATA);
2499 if (offs < 0) {
2500 return -errno; /* D3 or D4 */
2501 }
2502
2503 if (offs < start) {
2504 /* This is not a valid return by lseek(). We are safe to just return
2505 * -EIO in this case, and we'll treat it like D4. */
2506 return -EIO;
2507 }
2508
2509 if (offs > start) {
2510 /* D2: in hole, next data at offs */
2511 *hole = start;
2512 *data = offs;
2513 return 0;
2514 }
2515
2516 /* D1: in data, end not yet known */
2517
2518 /*
2519 * SEEK_HOLE cases:
2520 * H1. offs == start: start is in a hole
2521 * If this happens here, a hole has been dug behind our back
2522 * since the previous lseek().
2523 * H2. offs > start: either start is in data, next hole at offs,
2524 * or start is in trailing hole, EOF at offs
2525 * Linux treats trailing holes like any other hole: offs ==
2526 * start. Solaris seeks to EOF instead: offs > start (blech).
2527 * If that happens here, a hole has been dug behind our back
2528 * since the previous lseek().
2529 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2530 * If this happens, the file has been truncated behind our
2531 * back since we opened it. Treat it like a trailing hole.
2532 * H4. offs < 0, errno != ENXIO: we learned nothing
2533 * Pretend we know nothing at all, i.e. "forget" about D1.
2534 */
2535 offs = lseek(s->fd, start, SEEK_HOLE);
2536 if (offs < 0) {
2537 return -errno; /* D1 and (H3 or H4) */
2538 }
2539
2540 if (offs < start) {
2541 /* This is not a valid return by lseek(). We are safe to just return
2542 * -EIO in this case, and we'll treat it like H4. */
2543 return -EIO;
2544 }
2545
2546 if (offs > start) {
2547 /*
2548 * D1 and H2: either in data, next hole at offs, or it was in
2549 * data but is now in a trailing hole. In the latter case,
2550 * all bets are off. Treating it as if it there was data all
2551 * the way to EOF is safe, so simply do that.
2552 */
2553 *data = start;
2554 *hole = offs;
2555 return 0;
2556 }
2557
2558 /* D1 and H1 */
2559 return -EBUSY;
2560 #else
2561 return -ENOTSUP;
2562 #endif
2563 }
2564
2565 /*
2566 * Returns the allocation status of the specified offset.
2567 *
2568 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2569 *
2570 * 'pnum' is set to the number of bytes (including and immediately following
2571 * the specified offset) that are known to be in the same
2572 * allocated/unallocated state.
2573 *
2574 * 'bytes' is the max value 'pnum' should be set to.
2575 */
raw_co_block_status(BlockDriverState * bs,bool want_zero,int64_t offset,int64_t bytes,int64_t * pnum,int64_t * map,BlockDriverState ** file)2576 static int coroutine_fn raw_co_block_status(BlockDriverState *bs,
2577 bool want_zero,
2578 int64_t offset,
2579 int64_t bytes, int64_t *pnum,
2580 int64_t *map,
2581 BlockDriverState **file)
2582 {
2583 off_t data = 0, hole = 0;
2584 int ret;
2585
2586 assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment));
2587
2588 ret = fd_open(bs);
2589 if (ret < 0) {
2590 return ret;
2591 }
2592
2593 if (!want_zero) {
2594 *pnum = bytes;
2595 *map = offset;
2596 *file = bs;
2597 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
2598 }
2599
2600 ret = find_allocation(bs, offset, &data, &hole);
2601 if (ret == -ENXIO) {
2602 /* Trailing hole */
2603 *pnum = bytes;
2604 ret = BDRV_BLOCK_ZERO;
2605 } else if (ret < 0) {
2606 /* No info available, so pretend there are no holes */
2607 *pnum = bytes;
2608 ret = BDRV_BLOCK_DATA;
2609 } else if (data == offset) {
2610 /* On a data extent, compute bytes to the end of the extent,
2611 * possibly including a partial sector at EOF. */
2612 *pnum = MIN(bytes, hole - offset);
2613
2614 /*
2615 * We are not allowed to return partial sectors, though, so
2616 * round up if necessary.
2617 */
2618 if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) {
2619 int64_t file_length = raw_getlength(bs);
2620 if (file_length > 0) {
2621 /* Ignore errors, this is just a safeguard */
2622 assert(hole == file_length);
2623 }
2624 *pnum = ROUND_UP(*pnum, bs->bl.request_alignment);
2625 }
2626
2627 ret = BDRV_BLOCK_DATA;
2628 } else {
2629 /* On a hole, compute bytes to the beginning of the next extent. */
2630 assert(hole == offset);
2631 *pnum = MIN(bytes, data - offset);
2632 ret = BDRV_BLOCK_ZERO;
2633 }
2634 *map = offset;
2635 *file = bs;
2636 return ret | BDRV_BLOCK_OFFSET_VALID;
2637 }
2638
2639 #if defined(__linux__)
2640 /* Verify that the file is not in the page cache */
check_cache_dropped(BlockDriverState * bs,Error ** errp)2641 static void check_cache_dropped(BlockDriverState *bs, Error **errp)
2642 {
2643 const size_t window_size = 128 * 1024 * 1024;
2644 BDRVRawState *s = bs->opaque;
2645 void *window = NULL;
2646 size_t length = 0;
2647 unsigned char *vec;
2648 size_t page_size;
2649 off_t offset;
2650 off_t end;
2651
2652 /* mincore(2) page status information requires 1 byte per page */
2653 page_size = sysconf(_SC_PAGESIZE);
2654 vec = g_malloc(DIV_ROUND_UP(window_size, page_size));
2655
2656 end = raw_getlength(bs);
2657
2658 for (offset = 0; offset < end; offset += window_size) {
2659 void *new_window;
2660 size_t new_length;
2661 size_t vec_end;
2662 size_t i;
2663 int ret;
2664
2665 /* Unmap previous window if size has changed */
2666 new_length = MIN(end - offset, window_size);
2667 if (new_length != length) {
2668 munmap(window, length);
2669 window = NULL;
2670 length = 0;
2671 }
2672
2673 new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE,
2674 s->fd, offset);
2675 if (new_window == MAP_FAILED) {
2676 error_setg_errno(errp, errno, "mmap failed");
2677 break;
2678 }
2679
2680 window = new_window;
2681 length = new_length;
2682
2683 ret = mincore(window, length, vec);
2684 if (ret < 0) {
2685 error_setg_errno(errp, errno, "mincore failed");
2686 break;
2687 }
2688
2689 vec_end = DIV_ROUND_UP(length, page_size);
2690 for (i = 0; i < vec_end; i++) {
2691 if (vec[i] & 0x1) {
2692 error_setg(errp, "page cache still in use!");
2693 break;
2694 }
2695 }
2696 }
2697
2698 if (window) {
2699 munmap(window, length);
2700 }
2701
2702 g_free(vec);
2703 }
2704 #endif /* __linux__ */
2705
raw_co_invalidate_cache(BlockDriverState * bs,Error ** errp)2706 static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs,
2707 Error **errp)
2708 {
2709 BDRVRawState *s = bs->opaque;
2710 int ret;
2711
2712 ret = fd_open(bs);
2713 if (ret < 0) {
2714 error_setg_errno(errp, -ret, "The file descriptor is not open");
2715 return;
2716 }
2717
2718 if (!s->drop_cache) {
2719 return;
2720 }
2721
2722 if (s->open_flags & O_DIRECT) {
2723 return; /* No host kernel page cache */
2724 }
2725
2726 #if defined(__linux__)
2727 /* This sets the scene for the next syscall... */
2728 ret = bdrv_co_flush(bs);
2729 if (ret < 0) {
2730 error_setg_errno(errp, -ret, "flush failed");
2731 return;
2732 }
2733
2734 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2735 * process. These limitations are okay because we just fsynced the file,
2736 * we don't use mmap, and the file should not be in use by other processes.
2737 */
2738 ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED);
2739 if (ret != 0) { /* the return value is a positive errno */
2740 error_setg_errno(errp, ret, "fadvise failed");
2741 return;
2742 }
2743
2744 if (s->check_cache_dropped) {
2745 check_cache_dropped(bs, errp);
2746 }
2747 #else /* __linux__ */
2748 /* Do nothing. Live migration to a remote host with cache.direct=off is
2749 * unsupported on other host operating systems. Cache consistency issues
2750 * may occur but no error is reported here, partly because that's the
2751 * historical behavior and partly because it's hard to differentiate valid
2752 * configurations that should not cause errors.
2753 */
2754 #endif /* !__linux__ */
2755 }
2756
raw_account_discard(BDRVRawState * s,uint64_t nbytes,int ret)2757 static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret)
2758 {
2759 if (ret) {
2760 s->stats.discard_nb_failed++;
2761 } else {
2762 s->stats.discard_nb_ok++;
2763 s->stats.discard_bytes_ok += nbytes;
2764 }
2765 }
2766
2767 static coroutine_fn int
raw_do_pdiscard(BlockDriverState * bs,int64_t offset,int bytes,bool blkdev)2768 raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev)
2769 {
2770 BDRVRawState *s = bs->opaque;
2771 RawPosixAIOData acb;
2772 int ret;
2773
2774 acb = (RawPosixAIOData) {
2775 .bs = bs,
2776 .aio_fildes = s->fd,
2777 .aio_type = QEMU_AIO_DISCARD,
2778 .aio_offset = offset,
2779 .aio_nbytes = bytes,
2780 };
2781
2782 if (blkdev) {
2783 acb.aio_type |= QEMU_AIO_BLKDEV;
2784 }
2785
2786 ret = raw_thread_pool_submit(bs, handle_aiocb_discard, &acb);
2787 raw_account_discard(s, bytes, ret);
2788 return ret;
2789 }
2790
2791 static coroutine_fn int
raw_co_pdiscard(BlockDriverState * bs,int64_t offset,int bytes)2792 raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
2793 {
2794 return raw_do_pdiscard(bs, offset, bytes, false);
2795 }
2796
2797 static int coroutine_fn
raw_do_pwrite_zeroes(BlockDriverState * bs,int64_t offset,int bytes,BdrvRequestFlags flags,bool blkdev)2798 raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes,
2799 BdrvRequestFlags flags, bool blkdev)
2800 {
2801 BDRVRawState *s = bs->opaque;
2802 RawPosixAIOData acb;
2803 ThreadPoolFunc *handler;
2804
2805 #ifdef CONFIG_FALLOCATE
2806 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
2807 BdrvTrackedRequest *req;
2808 uint64_t end;
2809
2810 /*
2811 * This is a workaround for a bug in the Linux XFS driver,
2812 * where writes submitted through the AIO interface will be
2813 * discarded if they happen beyond a concurrently running
2814 * fallocate() that increases the file length (i.e., both the
2815 * write and the fallocate() happen beyond the EOF).
2816 *
2817 * To work around it, we extend the tracked request for this
2818 * zero write until INT64_MAX (effectively infinity), and mark
2819 * it as serializing.
2820 *
2821 * We have to enable this workaround for all filesystems and
2822 * AIO modes (not just XFS with aio=native), because for
2823 * remote filesystems we do not know the host configuration.
2824 */
2825
2826 req = bdrv_co_get_self_request(bs);
2827 assert(req);
2828 assert(req->type == BDRV_TRACKED_WRITE);
2829 assert(req->offset <= offset);
2830 assert(req->offset + req->bytes >= offset + bytes);
2831
2832 end = INT64_MAX & -(uint64_t)bs->bl.request_alignment;
2833 req->bytes = end - req->offset;
2834 req->overlap_bytes = req->bytes;
2835
2836 bdrv_mark_request_serialising(req, bs->bl.request_alignment);
2837 }
2838 #endif
2839
2840 acb = (RawPosixAIOData) {
2841 .bs = bs,
2842 .aio_fildes = s->fd,
2843 .aio_type = QEMU_AIO_WRITE_ZEROES,
2844 .aio_offset = offset,
2845 .aio_nbytes = bytes,
2846 };
2847
2848 if (blkdev) {
2849 acb.aio_type |= QEMU_AIO_BLKDEV;
2850 }
2851 if (flags & BDRV_REQ_NO_FALLBACK) {
2852 acb.aio_type |= QEMU_AIO_NO_FALLBACK;
2853 }
2854
2855 if (flags & BDRV_REQ_MAY_UNMAP) {
2856 acb.aio_type |= QEMU_AIO_DISCARD;
2857 handler = handle_aiocb_write_zeroes_unmap;
2858 } else {
2859 handler = handle_aiocb_write_zeroes;
2860 }
2861
2862 return raw_thread_pool_submit(bs, handler, &acb);
2863 }
2864
raw_co_pwrite_zeroes(BlockDriverState * bs,int64_t offset,int bytes,BdrvRequestFlags flags)2865 static int coroutine_fn raw_co_pwrite_zeroes(
2866 BlockDriverState *bs, int64_t offset,
2867 int bytes, BdrvRequestFlags flags)
2868 {
2869 return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false);
2870 }
2871
raw_get_info(BlockDriverState * bs,BlockDriverInfo * bdi)2872 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2873 {
2874 BDRVRawState *s = bs->opaque;
2875
2876 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
2877 return 0;
2878 }
2879
get_blockstats_specific_file(BlockDriverState * bs)2880 static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs)
2881 {
2882 BDRVRawState *s = bs->opaque;
2883 return (BlockStatsSpecificFile) {
2884 .discard_nb_ok = s->stats.discard_nb_ok,
2885 .discard_nb_failed = s->stats.discard_nb_failed,
2886 .discard_bytes_ok = s->stats.discard_bytes_ok,
2887 };
2888 }
2889
raw_get_specific_stats(BlockDriverState * bs)2890 static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs)
2891 {
2892 BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1);
2893
2894 stats->driver = BLOCKDEV_DRIVER_FILE;
2895 stats->u.file = get_blockstats_specific_file(bs);
2896
2897 return stats;
2898 }
2899
hdev_get_specific_stats(BlockDriverState * bs)2900 static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs)
2901 {
2902 BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1);
2903
2904 stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE;
2905 stats->u.host_device = get_blockstats_specific_file(bs);
2906
2907 return stats;
2908 }
2909
2910 static QemuOptsList raw_create_opts = {
2911 .name = "raw-create-opts",
2912 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
2913 .desc = {
2914 {
2915 .name = BLOCK_OPT_SIZE,
2916 .type = QEMU_OPT_SIZE,
2917 .help = "Virtual disk size"
2918 },
2919 {
2920 .name = BLOCK_OPT_NOCOW,
2921 .type = QEMU_OPT_BOOL,
2922 .help = "Turn off copy-on-write (valid only on btrfs)"
2923 },
2924 {
2925 .name = BLOCK_OPT_PREALLOC,
2926 .type = QEMU_OPT_STRING,
2927 .help = "Preallocation mode (allowed values: off"
2928 #ifdef CONFIG_POSIX_FALLOCATE
2929 ", falloc"
2930 #endif
2931 ", full)"
2932 },
2933 { /* end of list */ }
2934 }
2935 };
2936
raw_check_perm(BlockDriverState * bs,uint64_t perm,uint64_t shared,Error ** errp)2937 static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared,
2938 Error **errp)
2939 {
2940 BDRVRawState *s = bs->opaque;
2941 BDRVRawReopenState *rs = NULL;
2942 int open_flags;
2943 int ret;
2944
2945 if (s->perm_change_fd) {
2946 /*
2947 * In the context of reopen, this function may be called several times
2948 * (directly and recursively while change permissions of the parent).
2949 * This is even true for children that don't inherit from the original
2950 * reopen node, so s->reopen_state is not set.
2951 *
2952 * Ignore all but the first call.
2953 */
2954 return 0;
2955 }
2956
2957 if (s->reopen_state) {
2958 /* We already have a new file descriptor to set permissions for */
2959 assert(s->reopen_state->perm == perm);
2960 assert(s->reopen_state->shared_perm == shared);
2961 rs = s->reopen_state->opaque;
2962 s->perm_change_fd = rs->fd;
2963 s->perm_change_flags = rs->open_flags;
2964 } else {
2965 /* We may need a new fd if auto-read-only switches the mode */
2966 ret = raw_reconfigure_getfd(bs, bs->open_flags, &open_flags, perm,
2967 false, errp);
2968 if (ret < 0) {
2969 return ret;
2970 } else if (ret != s->fd) {
2971 s->perm_change_fd = ret;
2972 s->perm_change_flags = open_flags;
2973 }
2974 }
2975
2976 /* Prepare permissions on old fd to avoid conflicts between old and new,
2977 * but keep everything locked that new will need. */
2978 ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp);
2979 if (ret < 0) {
2980 goto fail;
2981 }
2982
2983 /* Copy locks to the new fd */
2984 if (s->perm_change_fd) {
2985 ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared,
2986 false, errp);
2987 if (ret < 0) {
2988 raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
2989 goto fail;
2990 }
2991 }
2992 return 0;
2993
2994 fail:
2995 if (s->perm_change_fd && !s->reopen_state) {
2996 qemu_close(s->perm_change_fd);
2997 }
2998 s->perm_change_fd = 0;
2999 return ret;
3000 }
3001
raw_set_perm(BlockDriverState * bs,uint64_t perm,uint64_t shared)3002 static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared)
3003 {
3004 BDRVRawState *s = bs->opaque;
3005
3006 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
3007 * called after .bdrv_reopen_commit) */
3008 if (s->perm_change_fd && s->fd != s->perm_change_fd) {
3009 qemu_close(s->fd);
3010 s->fd = s->perm_change_fd;
3011 s->open_flags = s->perm_change_flags;
3012 }
3013 s->perm_change_fd = 0;
3014
3015 raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL);
3016 s->perm = perm;
3017 s->shared_perm = shared;
3018 }
3019
raw_abort_perm_update(BlockDriverState * bs)3020 static void raw_abort_perm_update(BlockDriverState *bs)
3021 {
3022 BDRVRawState *s = bs->opaque;
3023
3024 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
3025 * the file descriptor. */
3026 if (s->perm_change_fd && !s->reopen_state) {
3027 qemu_close(s->perm_change_fd);
3028 }
3029 s->perm_change_fd = 0;
3030
3031 raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
3032 }
3033
raw_co_copy_range_from(BlockDriverState * bs,BdrvChild * src,uint64_t src_offset,BdrvChild * dst,uint64_t dst_offset,uint64_t bytes,BdrvRequestFlags read_flags,BdrvRequestFlags write_flags)3034 static int coroutine_fn raw_co_copy_range_from(
3035 BlockDriverState *bs, BdrvChild *src, uint64_t src_offset,
3036 BdrvChild *dst, uint64_t dst_offset, uint64_t bytes,
3037 BdrvRequestFlags read_flags, BdrvRequestFlags write_flags)
3038 {
3039 return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes,
3040 read_flags, write_flags);
3041 }
3042
raw_co_copy_range_to(BlockDriverState * bs,BdrvChild * src,uint64_t src_offset,BdrvChild * dst,uint64_t dst_offset,uint64_t bytes,BdrvRequestFlags read_flags,BdrvRequestFlags write_flags)3043 static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs,
3044 BdrvChild *src,
3045 uint64_t src_offset,
3046 BdrvChild *dst,
3047 uint64_t dst_offset,
3048 uint64_t bytes,
3049 BdrvRequestFlags read_flags,
3050 BdrvRequestFlags write_flags)
3051 {
3052 RawPosixAIOData acb;
3053 BDRVRawState *s = bs->opaque;
3054 BDRVRawState *src_s;
3055
3056 assert(dst->bs == bs);
3057 if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) {
3058 return -ENOTSUP;
3059 }
3060
3061 src_s = src->bs->opaque;
3062 if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) {
3063 return -EIO;
3064 }
3065
3066 acb = (RawPosixAIOData) {
3067 .bs = bs,
3068 .aio_type = QEMU_AIO_COPY_RANGE,
3069 .aio_fildes = src_s->fd,
3070 .aio_offset = src_offset,
3071 .aio_nbytes = bytes,
3072 .copy_range = {
3073 .aio_fd2 = s->fd,
3074 .aio_offset2 = dst_offset,
3075 },
3076 };
3077
3078 return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb);
3079 }
3080
3081 BlockDriver bdrv_file = {
3082 .format_name = "file",
3083 .protocol_name = "file",
3084 .instance_size = sizeof(BDRVRawState),
3085 .bdrv_needs_filename = true,
3086 .bdrv_probe = NULL, /* no probe for protocols */
3087 .bdrv_parse_filename = raw_parse_filename,
3088 .bdrv_file_open = raw_open,
3089 .bdrv_reopen_prepare = raw_reopen_prepare,
3090 .bdrv_reopen_commit = raw_reopen_commit,
3091 .bdrv_reopen_abort = raw_reopen_abort,
3092 .bdrv_close = raw_close,
3093 .bdrv_co_create = raw_co_create,
3094 .bdrv_co_create_opts = raw_co_create_opts,
3095 .bdrv_has_zero_init = bdrv_has_zero_init_1,
3096 .bdrv_has_zero_init_truncate = bdrv_has_zero_init_1,
3097 .bdrv_co_block_status = raw_co_block_status,
3098 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
3099 .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes,
3100 .bdrv_co_delete_file = raw_co_delete_file,
3101
3102 .bdrv_co_preadv = raw_co_preadv,
3103 .bdrv_co_pwritev = raw_co_pwritev,
3104 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3105 .bdrv_co_pdiscard = raw_co_pdiscard,
3106 .bdrv_co_copy_range_from = raw_co_copy_range_from,
3107 .bdrv_co_copy_range_to = raw_co_copy_range_to,
3108 .bdrv_refresh_limits = raw_refresh_limits,
3109 .bdrv_io_plug = raw_aio_plug,
3110 .bdrv_io_unplug = raw_aio_unplug,
3111 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
3112
3113 .bdrv_co_truncate = raw_co_truncate,
3114 .bdrv_getlength = raw_getlength,
3115 .bdrv_get_info = raw_get_info,
3116 .bdrv_get_allocated_file_size
3117 = raw_get_allocated_file_size,
3118 .bdrv_get_specific_stats = raw_get_specific_stats,
3119 .bdrv_check_perm = raw_check_perm,
3120 .bdrv_set_perm = raw_set_perm,
3121 .bdrv_abort_perm_update = raw_abort_perm_update,
3122 .create_opts = &raw_create_opts,
3123 .mutable_opts = mutable_opts,
3124 };
3125
3126 /***********************************************/
3127 /* host device */
3128
3129 #if defined(__APPLE__) && defined(__MACH__)
3130 static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
3131 CFIndex maxPathSize, int flags);
FindEjectableOpticalMedia(io_iterator_t * mediaIterator)3132 static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
3133 {
3134 kern_return_t kernResult = KERN_FAILURE;
3135 mach_port_t masterPort;
3136 CFMutableDictionaryRef classesToMatch;
3137 const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
3138 char *mediaType = NULL;
3139
3140 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
3141 if ( KERN_SUCCESS != kernResult ) {
3142 printf( "IOMasterPort returned %d\n", kernResult );
3143 }
3144
3145 int index;
3146 for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
3147 classesToMatch = IOServiceMatching(matching_array[index]);
3148 if (classesToMatch == NULL) {
3149 error_report("IOServiceMatching returned NULL for %s",
3150 matching_array[index]);
3151 continue;
3152 }
3153 CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
3154 kCFBooleanTrue);
3155 kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch,
3156 mediaIterator);
3157 if (kernResult != KERN_SUCCESS) {
3158 error_report("Note: IOServiceGetMatchingServices returned %d",
3159 kernResult);
3160 continue;
3161 }
3162
3163 /* If a match was found, leave the loop */
3164 if (*mediaIterator != 0) {
3165 trace_file_FindEjectableOpticalMedia(matching_array[index]);
3166 mediaType = g_strdup(matching_array[index]);
3167 break;
3168 }
3169 }
3170 return mediaType;
3171 }
3172
GetBSDPath(io_iterator_t mediaIterator,char * bsdPath,CFIndex maxPathSize,int flags)3173 kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
3174 CFIndex maxPathSize, int flags)
3175 {
3176 io_object_t nextMedia;
3177 kern_return_t kernResult = KERN_FAILURE;
3178 *bsdPath = '\0';
3179 nextMedia = IOIteratorNext( mediaIterator );
3180 if ( nextMedia )
3181 {
3182 CFTypeRef bsdPathAsCFString;
3183 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
3184 if ( bsdPathAsCFString ) {
3185 size_t devPathLength;
3186 strcpy( bsdPath, _PATH_DEV );
3187 if (flags & BDRV_O_NOCACHE) {
3188 strcat(bsdPath, "r");
3189 }
3190 devPathLength = strlen( bsdPath );
3191 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
3192 kernResult = KERN_SUCCESS;
3193 }
3194 CFRelease( bsdPathAsCFString );
3195 }
3196 IOObjectRelease( nextMedia );
3197 }
3198
3199 return kernResult;
3200 }
3201
3202 /* Sets up a real cdrom for use in QEMU */
setup_cdrom(char * bsd_path,Error ** errp)3203 static bool setup_cdrom(char *bsd_path, Error **errp)
3204 {
3205 int index, num_of_test_partitions = 2, fd;
3206 char test_partition[MAXPATHLEN];
3207 bool partition_found = false;
3208
3209 /* look for a working partition */
3210 for (index = 0; index < num_of_test_partitions; index++) {
3211 snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
3212 index);
3213 fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE);
3214 if (fd >= 0) {
3215 partition_found = true;
3216 qemu_close(fd);
3217 break;
3218 }
3219 }
3220
3221 /* if a working partition on the device was not found */
3222 if (partition_found == false) {
3223 error_setg(errp, "Failed to find a working partition on disc");
3224 } else {
3225 trace_file_setup_cdrom(test_partition);
3226 pstrcpy(bsd_path, MAXPATHLEN, test_partition);
3227 }
3228 return partition_found;
3229 }
3230
3231 /* Prints directions on mounting and unmounting a device */
print_unmounting_directions(const char * file_name)3232 static void print_unmounting_directions(const char *file_name)
3233 {
3234 error_report("If device %s is mounted on the desktop, unmount"
3235 " it first before using it in QEMU", file_name);
3236 error_report("Command to unmount device: diskutil unmountDisk %s",
3237 file_name);
3238 error_report("Command to mount device: diskutil mountDisk %s", file_name);
3239 }
3240
3241 #endif /* defined(__APPLE__) && defined(__MACH__) */
3242
hdev_probe_device(const char * filename)3243 static int hdev_probe_device(const char *filename)
3244 {
3245 struct stat st;
3246
3247 /* allow a dedicated CD-ROM driver to match with a higher priority */
3248 if (strstart(filename, "/dev/cdrom", NULL))
3249 return 50;
3250
3251 if (stat(filename, &st) >= 0 &&
3252 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
3253 return 100;
3254 }
3255
3256 return 0;
3257 }
3258
check_hdev_writable(BDRVRawState * s)3259 static int check_hdev_writable(BDRVRawState *s)
3260 {
3261 #if defined(BLKROGET)
3262 /* Linux block devices can be configured "read-only" using blockdev(8).
3263 * This is independent of device node permissions and therefore open(2)
3264 * with O_RDWR succeeds. Actual writes fail with EPERM.
3265 *
3266 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
3267 * check for read-only block devices so that Linux block devices behave
3268 * properly.
3269 */
3270 struct stat st;
3271 int readonly = 0;
3272
3273 if (fstat(s->fd, &st)) {
3274 return -errno;
3275 }
3276
3277 if (!S_ISBLK(st.st_mode)) {
3278 return 0;
3279 }
3280
3281 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
3282 return -errno;
3283 }
3284
3285 if (readonly) {
3286 return -EACCES;
3287 }
3288 #endif /* defined(BLKROGET) */
3289 return 0;
3290 }
3291
hdev_parse_filename(const char * filename,QDict * options,Error ** errp)3292 static void hdev_parse_filename(const char *filename, QDict *options,
3293 Error **errp)
3294 {
3295 bdrv_parse_filename_strip_prefix(filename, "host_device:", options);
3296 }
3297
hdev_is_sg(BlockDriverState * bs)3298 static bool hdev_is_sg(BlockDriverState *bs)
3299 {
3300
3301 #if defined(__linux__)
3302
3303 BDRVRawState *s = bs->opaque;
3304 struct stat st;
3305 struct sg_scsi_id scsiid;
3306 int sg_version;
3307 int ret;
3308
3309 if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) {
3310 return false;
3311 }
3312
3313 ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version);
3314 if (ret < 0) {
3315 return false;
3316 }
3317
3318 ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid);
3319 if (ret >= 0) {
3320 trace_file_hdev_is_sg(scsiid.scsi_type, sg_version);
3321 return true;
3322 }
3323
3324 #endif
3325
3326 return false;
3327 }
3328
hdev_open(BlockDriverState * bs,QDict * options,int flags,Error ** errp)3329 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
3330 Error **errp)
3331 {
3332 BDRVRawState *s = bs->opaque;
3333 Error *local_err = NULL;
3334 int ret;
3335
3336 #if defined(__APPLE__) && defined(__MACH__)
3337 /*
3338 * Caution: while qdict_get_str() is fine, getting non-string types
3339 * would require more care. When @options come from -blockdev or
3340 * blockdev_add, its members are typed according to the QAPI
3341 * schema, but when they come from -drive, they're all QString.
3342 */
3343 const char *filename = qdict_get_str(options, "filename");
3344 char bsd_path[MAXPATHLEN] = "";
3345 bool error_occurred = false;
3346
3347 /* If using a real cdrom */
3348 if (strcmp(filename, "/dev/cdrom") == 0) {
3349 char *mediaType = NULL;
3350 kern_return_t ret_val;
3351 io_iterator_t mediaIterator = 0;
3352
3353 mediaType = FindEjectableOpticalMedia(&mediaIterator);
3354 if (mediaType == NULL) {
3355 error_setg(errp, "Please make sure your CD/DVD is in the optical"
3356 " drive");
3357 error_occurred = true;
3358 goto hdev_open_Mac_error;
3359 }
3360
3361 ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
3362 if (ret_val != KERN_SUCCESS) {
3363 error_setg(errp, "Could not get BSD path for optical drive");
3364 error_occurred = true;
3365 goto hdev_open_Mac_error;
3366 }
3367
3368 /* If a real optical drive was not found */
3369 if (bsd_path[0] == '\0') {
3370 error_setg(errp, "Failed to obtain bsd path for optical drive");
3371 error_occurred = true;
3372 goto hdev_open_Mac_error;
3373 }
3374
3375 /* If using a cdrom disc and finding a partition on the disc failed */
3376 if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
3377 setup_cdrom(bsd_path, errp) == false) {
3378 print_unmounting_directions(bsd_path);
3379 error_occurred = true;
3380 goto hdev_open_Mac_error;
3381 }
3382
3383 qdict_put_str(options, "filename", bsd_path);
3384
3385 hdev_open_Mac_error:
3386 g_free(mediaType);
3387 if (mediaIterator) {
3388 IOObjectRelease(mediaIterator);
3389 }
3390 if (error_occurred) {
3391 return -ENOENT;
3392 }
3393 }
3394 #endif /* defined(__APPLE__) && defined(__MACH__) */
3395
3396 s->type = FTYPE_FILE;
3397
3398 ret = raw_open_common(bs, options, flags, 0, true, &local_err);
3399 if (ret < 0) {
3400 error_propagate(errp, local_err);
3401 #if defined(__APPLE__) && defined(__MACH__)
3402 if (*bsd_path) {
3403 filename = bsd_path;
3404 }
3405 /* if a physical device experienced an error while being opened */
3406 if (strncmp(filename, "/dev/", 5) == 0) {
3407 print_unmounting_directions(filename);
3408 }
3409 #endif /* defined(__APPLE__) && defined(__MACH__) */
3410 return ret;
3411 }
3412
3413 /* Since this does ioctl the device must be already opened */
3414 bs->sg = hdev_is_sg(bs);
3415
3416 if (flags & BDRV_O_RDWR) {
3417 ret = check_hdev_writable(s);
3418 if (ret < 0) {
3419 raw_close(bs);
3420 error_setg_errno(errp, -ret, "The device is not writable");
3421 return ret;
3422 }
3423 }
3424
3425 return ret;
3426 }
3427
3428 #if defined(__linux__)
3429 static int coroutine_fn
hdev_co_ioctl(BlockDriverState * bs,unsigned long int req,void * buf)3430 hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
3431 {
3432 BDRVRawState *s = bs->opaque;
3433 RawPosixAIOData acb;
3434 int ret;
3435
3436 ret = fd_open(bs);
3437 if (ret < 0) {
3438 return ret;
3439 }
3440
3441 if (req == SG_IO && s->pr_mgr) {
3442 struct sg_io_hdr *io_hdr = buf;
3443 if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT ||
3444 io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) {
3445 return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs),
3446 s->fd, io_hdr);
3447 }
3448 }
3449
3450 acb = (RawPosixAIOData) {
3451 .bs = bs,
3452 .aio_type = QEMU_AIO_IOCTL,
3453 .aio_fildes = s->fd,
3454 .aio_offset = 0,
3455 .ioctl = {
3456 .buf = buf,
3457 .cmd = req,
3458 },
3459 };
3460
3461 return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb);
3462 }
3463 #endif /* linux */
3464
fd_open(BlockDriverState * bs)3465 static int fd_open(BlockDriverState *bs)
3466 {
3467 BDRVRawState *s = bs->opaque;
3468
3469 /* this is just to ensure s->fd is sane (its called by io ops) */
3470 if (s->fd >= 0)
3471 return 0;
3472 return -EIO;
3473 }
3474
3475 static coroutine_fn int
hdev_co_pdiscard(BlockDriverState * bs,int64_t offset,int bytes)3476 hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
3477 {
3478 BDRVRawState *s = bs->opaque;
3479 int ret;
3480
3481 ret = fd_open(bs);
3482 if (ret < 0) {
3483 raw_account_discard(s, bytes, ret);
3484 return ret;
3485 }
3486 return raw_do_pdiscard(bs, offset, bytes, true);
3487 }
3488
hdev_co_pwrite_zeroes(BlockDriverState * bs,int64_t offset,int bytes,BdrvRequestFlags flags)3489 static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs,
3490 int64_t offset, int bytes, BdrvRequestFlags flags)
3491 {
3492 int rc;
3493
3494 rc = fd_open(bs);
3495 if (rc < 0) {
3496 return rc;
3497 }
3498
3499 return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true);
3500 }
3501
3502 static BlockDriver bdrv_host_device = {
3503 .format_name = "host_device",
3504 .protocol_name = "host_device",
3505 .instance_size = sizeof(BDRVRawState),
3506 .bdrv_needs_filename = true,
3507 .bdrv_probe_device = hdev_probe_device,
3508 .bdrv_parse_filename = hdev_parse_filename,
3509 .bdrv_file_open = hdev_open,
3510 .bdrv_close = raw_close,
3511 .bdrv_reopen_prepare = raw_reopen_prepare,
3512 .bdrv_reopen_commit = raw_reopen_commit,
3513 .bdrv_reopen_abort = raw_reopen_abort,
3514 .mutable_opts = mutable_opts,
3515 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
3516 .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes,
3517
3518 .bdrv_co_preadv = raw_co_preadv,
3519 .bdrv_co_pwritev = raw_co_pwritev,
3520 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3521 .bdrv_co_pdiscard = hdev_co_pdiscard,
3522 .bdrv_co_copy_range_from = raw_co_copy_range_from,
3523 .bdrv_co_copy_range_to = raw_co_copy_range_to,
3524 .bdrv_refresh_limits = raw_refresh_limits,
3525 .bdrv_io_plug = raw_aio_plug,
3526 .bdrv_io_unplug = raw_aio_unplug,
3527 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
3528
3529 .bdrv_co_truncate = raw_co_truncate,
3530 .bdrv_getlength = raw_getlength,
3531 .bdrv_get_info = raw_get_info,
3532 .bdrv_get_allocated_file_size
3533 = raw_get_allocated_file_size,
3534 .bdrv_get_specific_stats = hdev_get_specific_stats,
3535 .bdrv_check_perm = raw_check_perm,
3536 .bdrv_set_perm = raw_set_perm,
3537 .bdrv_abort_perm_update = raw_abort_perm_update,
3538 .bdrv_probe_blocksizes = hdev_probe_blocksizes,
3539 .bdrv_probe_geometry = hdev_probe_geometry,
3540
3541 /* generic scsi device */
3542 #ifdef __linux__
3543 .bdrv_co_ioctl = hdev_co_ioctl,
3544 #endif
3545 };
3546
3547 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
cdrom_parse_filename(const char * filename,QDict * options,Error ** errp)3548 static void cdrom_parse_filename(const char *filename, QDict *options,
3549 Error **errp)
3550 {
3551 bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options);
3552 }
3553 #endif
3554
3555 #ifdef __linux__
cdrom_open(BlockDriverState * bs,QDict * options,int flags,Error ** errp)3556 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
3557 Error **errp)
3558 {
3559 BDRVRawState *s = bs->opaque;
3560
3561 s->type = FTYPE_CD;
3562
3563 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3564 return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp);
3565 }
3566
cdrom_probe_device(const char * filename)3567 static int cdrom_probe_device(const char *filename)
3568 {
3569 int fd, ret;
3570 int prio = 0;
3571 struct stat st;
3572
3573 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
3574 if (fd < 0) {
3575 goto out;
3576 }
3577 ret = fstat(fd, &st);
3578 if (ret == -1 || !S_ISBLK(st.st_mode)) {
3579 goto outc;
3580 }
3581
3582 /* Attempt to detect via a CDROM specific ioctl */
3583 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
3584 if (ret >= 0)
3585 prio = 100;
3586
3587 outc:
3588 qemu_close(fd);
3589 out:
3590 return prio;
3591 }
3592
cdrom_is_inserted(BlockDriverState * bs)3593 static bool cdrom_is_inserted(BlockDriverState *bs)
3594 {
3595 BDRVRawState *s = bs->opaque;
3596 int ret;
3597
3598 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
3599 return ret == CDS_DISC_OK;
3600 }
3601
cdrom_eject(BlockDriverState * bs,bool eject_flag)3602 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
3603 {
3604 BDRVRawState *s = bs->opaque;
3605
3606 if (eject_flag) {
3607 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
3608 perror("CDROMEJECT");
3609 } else {
3610 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
3611 perror("CDROMEJECT");
3612 }
3613 }
3614
cdrom_lock_medium(BlockDriverState * bs,bool locked)3615 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
3616 {
3617 BDRVRawState *s = bs->opaque;
3618
3619 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
3620 /*
3621 * Note: an error can happen if the distribution automatically
3622 * mounts the CD-ROM
3623 */
3624 /* perror("CDROM_LOCKDOOR"); */
3625 }
3626 }
3627
3628 static BlockDriver bdrv_host_cdrom = {
3629 .format_name = "host_cdrom",
3630 .protocol_name = "host_cdrom",
3631 .instance_size = sizeof(BDRVRawState),
3632 .bdrv_needs_filename = true,
3633 .bdrv_probe_device = cdrom_probe_device,
3634 .bdrv_parse_filename = cdrom_parse_filename,
3635 .bdrv_file_open = cdrom_open,
3636 .bdrv_close = raw_close,
3637 .bdrv_reopen_prepare = raw_reopen_prepare,
3638 .bdrv_reopen_commit = raw_reopen_commit,
3639 .bdrv_reopen_abort = raw_reopen_abort,
3640 .mutable_opts = mutable_opts,
3641 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
3642
3643
3644 .bdrv_co_preadv = raw_co_preadv,
3645 .bdrv_co_pwritev = raw_co_pwritev,
3646 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3647 .bdrv_refresh_limits = raw_refresh_limits,
3648 .bdrv_io_plug = raw_aio_plug,
3649 .bdrv_io_unplug = raw_aio_unplug,
3650 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
3651
3652 .bdrv_co_truncate = raw_co_truncate,
3653 .bdrv_getlength = raw_getlength,
3654 .has_variable_length = true,
3655 .bdrv_get_allocated_file_size
3656 = raw_get_allocated_file_size,
3657
3658 /* removable device support */
3659 .bdrv_is_inserted = cdrom_is_inserted,
3660 .bdrv_eject = cdrom_eject,
3661 .bdrv_lock_medium = cdrom_lock_medium,
3662
3663 /* generic scsi device */
3664 .bdrv_co_ioctl = hdev_co_ioctl,
3665 };
3666 #endif /* __linux__ */
3667
3668 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
cdrom_open(BlockDriverState * bs,QDict * options,int flags,Error ** errp)3669 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
3670 Error **errp)
3671 {
3672 BDRVRawState *s = bs->opaque;
3673 Error *local_err = NULL;
3674 int ret;
3675
3676 s->type = FTYPE_CD;
3677
3678 ret = raw_open_common(bs, options, flags, 0, true, &local_err);
3679 if (ret) {
3680 error_propagate(errp, local_err);
3681 return ret;
3682 }
3683
3684 /* make sure the door isn't locked at this time */
3685 ioctl(s->fd, CDIOCALLOW);
3686 return 0;
3687 }
3688
cdrom_probe_device(const char * filename)3689 static int cdrom_probe_device(const char *filename)
3690 {
3691 if (strstart(filename, "/dev/cd", NULL) ||
3692 strstart(filename, "/dev/acd", NULL))
3693 return 100;
3694 return 0;
3695 }
3696
cdrom_reopen(BlockDriverState * bs)3697 static int cdrom_reopen(BlockDriverState *bs)
3698 {
3699 BDRVRawState *s = bs->opaque;
3700 int fd;
3701
3702 /*
3703 * Force reread of possibly changed/newly loaded disc,
3704 * FreeBSD seems to not notice sometimes...
3705 */
3706 if (s->fd >= 0)
3707 qemu_close(s->fd);
3708 fd = qemu_open(bs->filename, s->open_flags, 0644);
3709 if (fd < 0) {
3710 s->fd = -1;
3711 return -EIO;
3712 }
3713 s->fd = fd;
3714
3715 /* make sure the door isn't locked at this time */
3716 ioctl(s->fd, CDIOCALLOW);
3717 return 0;
3718 }
3719
cdrom_is_inserted(BlockDriverState * bs)3720 static bool cdrom_is_inserted(BlockDriverState *bs)
3721 {
3722 return raw_getlength(bs) > 0;
3723 }
3724
cdrom_eject(BlockDriverState * bs,bool eject_flag)3725 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
3726 {
3727 BDRVRawState *s = bs->opaque;
3728
3729 if (s->fd < 0)
3730 return;
3731
3732 (void) ioctl(s->fd, CDIOCALLOW);
3733
3734 if (eject_flag) {
3735 if (ioctl(s->fd, CDIOCEJECT) < 0)
3736 perror("CDIOCEJECT");
3737 } else {
3738 if (ioctl(s->fd, CDIOCCLOSE) < 0)
3739 perror("CDIOCCLOSE");
3740 }
3741
3742 cdrom_reopen(bs);
3743 }
3744
cdrom_lock_medium(BlockDriverState * bs,bool locked)3745 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
3746 {
3747 BDRVRawState *s = bs->opaque;
3748
3749 if (s->fd < 0)
3750 return;
3751 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
3752 /*
3753 * Note: an error can happen if the distribution automatically
3754 * mounts the CD-ROM
3755 */
3756 /* perror("CDROM_LOCKDOOR"); */
3757 }
3758 }
3759
3760 static BlockDriver bdrv_host_cdrom = {
3761 .format_name = "host_cdrom",
3762 .protocol_name = "host_cdrom",
3763 .instance_size = sizeof(BDRVRawState),
3764 .bdrv_needs_filename = true,
3765 .bdrv_probe_device = cdrom_probe_device,
3766 .bdrv_parse_filename = cdrom_parse_filename,
3767 .bdrv_file_open = cdrom_open,
3768 .bdrv_close = raw_close,
3769 .bdrv_reopen_prepare = raw_reopen_prepare,
3770 .bdrv_reopen_commit = raw_reopen_commit,
3771 .bdrv_reopen_abort = raw_reopen_abort,
3772 .mutable_opts = mutable_opts,
3773
3774 .bdrv_co_preadv = raw_co_preadv,
3775 .bdrv_co_pwritev = raw_co_pwritev,
3776 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3777 .bdrv_refresh_limits = raw_refresh_limits,
3778 .bdrv_io_plug = raw_aio_plug,
3779 .bdrv_io_unplug = raw_aio_unplug,
3780 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
3781
3782 .bdrv_co_truncate = raw_co_truncate,
3783 .bdrv_getlength = raw_getlength,
3784 .has_variable_length = true,
3785 .bdrv_get_allocated_file_size
3786 = raw_get_allocated_file_size,
3787
3788 /* removable device support */
3789 .bdrv_is_inserted = cdrom_is_inserted,
3790 .bdrv_eject = cdrom_eject,
3791 .bdrv_lock_medium = cdrom_lock_medium,
3792 };
3793 #endif /* __FreeBSD__ */
3794
bdrv_file_init(void)3795 static void bdrv_file_init(void)
3796 {
3797 /*
3798 * Register all the drivers. Note that order is important, the driver
3799 * registered last will get probed first.
3800 */
3801 bdrv_register(&bdrv_file);
3802 bdrv_register(&bdrv_host_device);
3803 #ifdef __linux__
3804 bdrv_register(&bdrv_host_cdrom);
3805 #endif
3806 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3807 bdrv_register(&bdrv_host_cdrom);
3808 #endif
3809 }
3810
3811 block_init(bdrv_file_init);
3812