blob: 8aee7a3fb8be591aa4b883182ebf0fa01cb35491 [file] [log] [blame]
bellard83f64092006-08-01 16:21:11 +00001/*
ths223d4672007-12-15 17:28:36 +00002 * Block driver for RAW files (posix)
ths5fafdf22007-09-16 21:08:06 +00003 *
bellard83f64092006-08-01 16:21:11 +00004 * Copyright (c) 2006 Fabrice Bellard
ths5fafdf22007-09-16 21:08:06 +00005 *
bellard83f64092006-08-01 16:21:11 +00006 * 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 */
Markus Armbruster922a01a2018-02-01 12:18:46 +010024
Peter Maydell80c71a22016-01-18 18:01:42 +000025#include "qemu/osdep.h"
Markus Armbrusterda34e652016-03-14 09:01:28 +010026#include "qapi/error.h"
Veronia Bahaaf348b6d2016-03-20 19:16:19 +020027#include "qemu/cutils.h"
Markus Armbrusterd49b6832015-03-17 18:29:20 +010028#include "qemu/error-report.h"
Paolo Bonzini737e1502012-12-17 18:19:44 +010029#include "block/block_int.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010030#include "qemu/module.h"
Markus Armbruster922a01a2018-02-01 12:18:46 +010031#include "qemu/option.h"
Paolo Bonzinide81a162012-05-25 11:46:27 +020032#include "trace.h"
Paolo Bonzini737e1502012-12-17 18:19:44 +010033#include "block/thread-pool.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010034#include "qemu/iov.h"
Paolo Bonzini0187f5c2016-07-04 18:33:20 +020035#include "block/raw-aio.h"
Markus Armbruster452fcdb2018-02-01 12:18:39 +010036#include "qapi/qmp/qdict.h"
Markus Armbrusterd49b6832015-03-17 18:29:20 +010037#include "qapi/qmp/qstring.h"
bellard83f64092006-08-01 16:21:11 +000038
Paolo Bonzini7c9e5272017-08-21 18:58:56 +020039#include "scsi/pr-manager.h"
40#include "scsi/constants.h"
41
Pavel Borzenkov83affaa2011-11-10 22:40:06 +040042#if defined(__APPLE__) && (__MACH__)
bellard83f64092006-08-01 16:21:11 +000043#include <paths.h>
44#include <sys/param.h>
45#include <IOKit/IOKitLib.h>
46#include <IOKit/IOBSD.h>
47#include <IOKit/storage/IOMediaBSDClient.h>
48#include <IOKit/storage/IOMedia.h>
49#include <IOKit/storage/IOCDMedia.h>
50//#include <IOKit/storage/IOCDTypes.h>
Programmingkidd0855f12016-03-21 11:41:28 -040051#include <IOKit/storage/IODVDMedia.h>
bellard83f64092006-08-01 16:21:11 +000052#include <CoreFoundation/CoreFoundation.h>
53#endif
54
55#ifdef __sun__
ths2e9671d2007-01-05 17:44:41 +000056#define _POSIX_PTHREAD_SEMANTICS 1
bellard83f64092006-08-01 16:21:11 +000057#include <sys/dkio.h>
58#endif
bellard19cb3732006-08-19 11:45:59 +000059#ifdef __linux__
60#include <sys/ioctl.h>
Bernhard Kohl05acda42010-09-06 17:06:02 +020061#include <sys/param.h>
Fam Zheng1efad062018-06-01 17:26:43 +080062#include <sys/syscall.h>
bellard19cb3732006-08-19 11:45:59 +000063#include <linux/cdrom.h>
64#include <linux/fd.h>
Paolo Bonzini55003162012-05-09 16:49:58 +020065#include <linux/fs.h>
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +010066#include <linux/hdreg.h>
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +030067#include <scsi/sg.h>
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +010068#ifdef __s390__
69#include <asm/dasd.h>
70#endif
Chunyan Liu4ab15592014-06-30 14:29:58 +080071#ifndef FS_NOCOW_FL
72#define FS_NOCOW_FL 0x00800000 /* Do not cow file */
73#endif
Paolo Bonzini55003162012-05-09 16:49:58 +020074#endif
Denis V. Lunevb953f072015-01-30 11:42:14 +030075#if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
Kusanagi Kouichi3d4fa432013-01-14 16:26:52 +010076#include <linux/falloc.h>
77#endif
Aurelien Jarnoa167ba52009-11-29 18:00:41 +010078#if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
ths1cb6c3f2006-12-21 19:14:11 +000079#include <sys/disk.h>
blueswir19f230112009-03-28 08:37:13 +000080#include <sys/cdio.h>
ths1cb6c3f2006-12-21 19:14:11 +000081#endif
bellard19cb3732006-08-19 11:45:59 +000082
blueswir1128ab2f2008-08-15 18:33:42 +000083#ifdef __OpenBSD__
84#include <sys/ioctl.h>
85#include <sys/disklabel.h>
86#include <sys/dkio.h>
87#endif
88
Christoph Eggerd1f6fd82011-05-23 14:31:17 +020089#ifdef __NetBSD__
90#include <sys/ioctl.h>
91#include <sys/disklabel.h>
92#include <sys/dkio.h>
93#include <sys/disk.h>
94#endif
95
blueswir1c5e97232009-03-07 20:06:23 +000096#ifdef __DragonFly__
97#include <sys/ioctl.h>
98#include <sys/diskslice.h>
99#endif
100
Christoph Hellwigdce512d2010-12-17 11:41:15 +0100101#ifdef CONFIG_XFS
102#include <xfs/xfs.h>
103#endif
104
pbrookfaf07962007-11-11 02:51:17 +0000105//#define DEBUG_BLOCK
Dimitris Aragiorgisbcb22552015-06-23 13:44:58 +0300106
107#ifdef DEBUG_BLOCK
108# define DEBUG_BLOCK_PRINT 1
ths8c05dbf2007-09-13 12:29:23 +0000109#else
Dimitris Aragiorgisbcb22552015-06-23 13:44:58 +0300110# define DEBUG_BLOCK_PRINT 0
ths8c05dbf2007-09-13 12:29:23 +0000111#endif
Dimitris Aragiorgisbcb22552015-06-23 13:44:58 +0300112#define DPRINTF(fmt, ...) \
113do { \
114 if (DEBUG_BLOCK_PRINT) { \
115 printf(fmt, ## __VA_ARGS__); \
116 } \
117} while (0)
ths8c05dbf2007-09-13 12:29:23 +0000118
aliguorif6465572008-10-14 18:00:38 +0000119/* OS X does not have O_DSYNC */
120#ifndef O_DSYNC
G 31c27a8b2009-07-01 13:28:32 -0400121#ifdef O_SYNC
aliguori7ab064d2008-10-14 18:14:47 +0000122#define O_DSYNC O_SYNC
G 31c27a8b2009-07-01 13:28:32 -0400123#elif defined(O_FSYNC)
124#define O_DSYNC O_FSYNC
125#endif
aliguorif6465572008-10-14 18:00:38 +0000126#endif
127
aliguori9f7965c2008-10-14 14:42:54 +0000128/* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
129#ifndef O_DIRECT
130#define O_DIRECT O_DSYNC
131#endif
132
bellard19cb3732006-08-19 11:45:59 +0000133#define FTYPE_FILE 0
134#define FTYPE_CD 1
bellard83f64092006-08-01 16:21:11 +0000135
Christoph Hellwig581b9e22010-09-12 23:43:21 +0200136#define MAX_BLOCKSIZE 4096
137
Fam Zheng244a5662017-05-03 00:35:56 +0800138/* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
139 * leaving a few more bytes for its future use. */
140#define RAW_LOCK_PERM_BASE 100
141#define RAW_LOCK_SHARED_BASE 200
142
bellard83f64092006-08-01 16:21:11 +0000143typedef struct BDRVRawState {
144 int fd;
Fam Zheng244a5662017-05-03 00:35:56 +0800145 bool use_lock;
bellard19cb3732006-08-19 11:45:59 +0000146 int type;
Christoph Hellwig0e1d8f42009-06-15 13:53:26 +0200147 int open_flags;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100148 size_t buf_align;
149
Fam Zheng244a5662017-05-03 00:35:56 +0800150 /* The current permissions. */
151 uint64_t perm;
152 uint64_t shared_perm;
153
Fam Zheng2996ffa2018-10-11 15:21:33 +0800154 /* The perms bits whose corresponding bytes are already locked in
Fam Zhengf2e3af22018-10-11 15:21:34 +0800155 * s->fd. */
Fam Zheng2996ffa2018-10-11 15:21:33 +0800156 uint64_t locked_perm;
157 uint64_t locked_shared_perm;
158
Christoph Hellwigdce512d2010-12-17 11:41:15 +0100159#ifdef CONFIG_XFS
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100160 bool is_xfs:1;
Christoph Hellwigdce512d2010-12-17 11:41:15 +0100161#endif
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100162 bool has_discard:1;
Paolo Bonzini97a2ae32013-11-22 13:39:57 +0100163 bool has_write_zeroes:1;
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100164 bool discard_zeroes:1;
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200165 bool use_linux_aio:1;
Kevin Wolfe5bcf962017-03-22 22:00:05 +0100166 bool page_cache_inconsistent:1;
Denis V. Lunevd50d8222015-01-30 11:42:15 +0300167 bool has_fallocate;
Roger Pau Monne3cad8302014-10-21 16:03:03 +0200168 bool needs_alignment;
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +0100169 bool check_cache_dropped;
Paolo Bonzini7c9e5272017-08-21 18:58:56 +0200170
171 PRManager *pr_mgr;
bellard83f64092006-08-01 16:21:11 +0000172} BDRVRawState;
173
Jeff Codyeeb6b452012-09-20 15:13:25 -0400174typedef struct BDRVRawReopenState {
175 int fd;
176 int open_flags;
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +0100177 bool check_cache_dropped;
Jeff Codyeeb6b452012-09-20 15:13:25 -0400178} BDRVRawReopenState;
179
bellard19cb3732006-08-19 11:45:59 +0000180static int fd_open(BlockDriverState *bs);
Kevin Wolf22afa7b2009-06-26 19:51:24 +0200181static int64_t raw_getlength(BlockDriverState *bs);
bellard83f64092006-08-01 16:21:11 +0000182
Paolo Bonzinide81a162012-05-25 11:46:27 +0200183typedef struct RawPosixAIOData {
184 BlockDriverState *bs;
Paolo Bonzinide81a162012-05-25 11:46:27 +0200185 int aio_type;
Kevin Wolfd57c44d2018-10-25 16:21:14 +0100186 int aio_fildes;
187
188 off_t aio_offset;
189 uint64_t aio_nbytes;
190
Kevin Wolf93f4e2f2018-06-21 18:23:16 +0200191 union {
192 struct {
Kevin Wolfd57c44d2018-10-25 16:21:14 +0100193 struct iovec *iov;
194 int niov;
195 } io;
196 struct {
197 uint64_t cmd;
198 void *buf;
199 } ioctl;
200 struct {
Kevin Wolf93f4e2f2018-06-21 18:23:16 +0200201 int aio_fd2;
202 off_t aio_offset2;
Kevin Wolfd57c44d2018-10-25 16:21:14 +0100203 } copy_range;
Kevin Wolf93f4e2f2018-06-21 18:23:16 +0200204 struct {
205 PreallocMode prealloc;
206 Error **errp;
Kevin Wolfd57c44d2018-10-25 16:21:14 +0100207 } truncate;
Kevin Wolf93f4e2f2018-06-21 18:23:16 +0200208 };
Paolo Bonzinide81a162012-05-25 11:46:27 +0200209} RawPosixAIOData;
210
Aurelien Jarnoa167ba52009-11-29 18:00:41 +0100211#if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200212static int cdrom_reopen(BlockDriverState *bs);
blueswir19f230112009-03-28 08:37:13 +0000213#endif
214
Christoph Egger1de1ae02011-05-24 11:30:29 +0200215#if defined(__NetBSD__)
Fam Zhengdb0754d2018-11-01 14:29:09 +0800216static int raw_normalize_devicepath(const char **filename, Error **errp)
Christoph Egger1de1ae02011-05-24 11:30:29 +0200217{
218 static char namebuf[PATH_MAX];
219 const char *dp, *fname;
220 struct stat sb;
221
222 fname = *filename;
223 dp = strrchr(fname, '/');
224 if (lstat(fname, &sb) < 0) {
Fam Zhengdb0754d2018-11-01 14:29:09 +0800225 error_setg_errno(errp, errno, "%s: stat failed", fname);
Christoph Egger1de1ae02011-05-24 11:30:29 +0200226 return -errno;
227 }
228
229 if (!S_ISBLK(sb.st_mode)) {
230 return 0;
231 }
232
233 if (dp == NULL) {
234 snprintf(namebuf, PATH_MAX, "r%s", fname);
235 } else {
236 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
237 (int)(dp - fname), fname, dp + 1);
238 }
Christoph Egger1de1ae02011-05-24 11:30:29 +0200239 *filename = namebuf;
Fam Zhengdb0754d2018-11-01 14:29:09 +0800240 warn_report("%s is a block device, using %s", fname, *filename);
Christoph Egger1de1ae02011-05-24 11:30:29 +0200241
242 return 0;
243}
244#else
Fam Zhengdb0754d2018-11-01 14:29:09 +0800245static int raw_normalize_devicepath(const char **filename, Error **errp)
Christoph Egger1de1ae02011-05-24 11:30:29 +0200246{
247 return 0;
248}
249#endif
250
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100251/*
252 * Get logical block size via ioctl. On success store it in @sector_size_p.
253 */
254static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
255{
256 unsigned int sector_size;
257 bool success = false;
Peter Maydell700f9ce2017-03-23 14:36:28 +0000258 int i;
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100259
260 errno = ENOTSUP;
Peter Maydell700f9ce2017-03-23 14:36:28 +0000261 static const unsigned long ioctl_list[] = {
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100262#ifdef BLKSSZGET
Peter Maydell700f9ce2017-03-23 14:36:28 +0000263 BLKSSZGET,
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100264#endif
265#ifdef DKIOCGETBLOCKSIZE
Peter Maydell700f9ce2017-03-23 14:36:28 +0000266 DKIOCGETBLOCKSIZE,
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100267#endif
268#ifdef DIOCGSECTORSIZE
Peter Maydell700f9ce2017-03-23 14:36:28 +0000269 DIOCGSECTORSIZE,
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100270#endif
Peter Maydell700f9ce2017-03-23 14:36:28 +0000271 };
272
273 /* Try a few ioctls to get the right size */
274 for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) {
275 if (ioctl(fd, ioctl_list[i], &sector_size) >= 0) {
276 *sector_size_p = sector_size;
277 success = true;
278 }
279 }
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100280
281 return success ? 0 : -errno;
282}
283
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +0100284/**
285 * Get physical block size of @fd.
286 * On success, store it in @blk_size and return 0.
287 * On failure, return -errno.
288 */
289static int probe_physical_blocksize(int fd, unsigned int *blk_size)
290{
291#ifdef BLKPBSZGET
292 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
293 return -errno;
294 }
295 return 0;
296#else
297 return -ENOTSUP;
298#endif
299}
300
Stefan Hajnoczi22d182e2015-03-05 15:38:17 -0600301/* Check if read is allowed with given memory buffer and length.
302 *
303 * This function is used to check O_DIRECT memory buffer and request alignment.
304 */
305static bool raw_is_io_aligned(int fd, void *buf, size_t len)
306{
307 ssize_t ret = pread(fd, buf, len, 0);
308
309 if (ret >= 0) {
310 return true;
311 }
312
313#ifdef __linux__
314 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
315 * other errors (e.g. real I/O error), which could happen on a failed
316 * drive, since we only care about probing alignment.
317 */
318 if (errno != EINVAL) {
319 return true;
320 }
321#endif
322
323 return false;
324}
325
Kevin Wolfdf26a352014-07-16 17:48:17 +0200326static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100327{
328 BDRVRawState *s = bs->opaque;
329 char *buf;
Denis V. Lunev459b4e62015-05-12 17:30:56 +0300330 size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize());
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100331
Dimitris Aragiorgisb192af82015-06-23 13:44:56 +0300332 /* For SCSI generic devices the alignment is not really used.
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100333 With buffered I/O, we don't have any restrictions. */
Dimitris Aragiorgisb192af82015-06-23 13:44:56 +0300334 if (bdrv_is_sg(bs) || !s->needs_alignment) {
Eric Blakea5b8dd22016-06-23 16:37:24 -0600335 bs->bl.request_alignment = 1;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100336 s->buf_align = 1;
337 return;
338 }
339
Eric Blakea5b8dd22016-06-23 16:37:24 -0600340 bs->bl.request_alignment = 0;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100341 s->buf_align = 0;
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100342 /* Let's try to use the logical blocksize for the alignment. */
Eric Blakea5b8dd22016-06-23 16:37:24 -0600343 if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) {
344 bs->bl.request_alignment = 0;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100345 }
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100346#ifdef CONFIG_XFS
347 if (s->is_xfs) {
348 struct dioattr da;
Kevin Wolfdf26a352014-07-16 17:48:17 +0200349 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
Eric Blakea5b8dd22016-06-23 16:37:24 -0600350 bs->bl.request_alignment = da.d_miniosz;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100351 /* The kernel returns wrong information for d_mem */
352 /* s->buf_align = da.d_mem; */
353 }
354 }
355#endif
356
357 /* If we could not get the sizes so far, we can only guess them */
358 if (!s->buf_align) {
359 size_t align;
Denis V. Lunev459b4e62015-05-12 17:30:56 +0300360 buf = qemu_memalign(max_align, 2 * max_align);
361 for (align = 512; align <= max_align; align <<= 1) {
362 if (raw_is_io_aligned(fd, buf + align, max_align)) {
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100363 s->buf_align = align;
364 break;
365 }
366 }
367 qemu_vfree(buf);
368 }
369
Eric Blakea5b8dd22016-06-23 16:37:24 -0600370 if (!bs->bl.request_alignment) {
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100371 size_t align;
Denis V. Lunev459b4e62015-05-12 17:30:56 +0300372 buf = qemu_memalign(s->buf_align, max_align);
373 for (align = 512; align <= max_align; align <<= 1) {
Stefan Hajnoczi22d182e2015-03-05 15:38:17 -0600374 if (raw_is_io_aligned(fd, buf, align)) {
Eric Blakea5b8dd22016-06-23 16:37:24 -0600375 bs->bl.request_alignment = align;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100376 break;
377 }
378 }
379 qemu_vfree(buf);
380 }
Kevin Wolfdf26a352014-07-16 17:48:17 +0200381
Eric Blakea5b8dd22016-06-23 16:37:24 -0600382 if (!s->buf_align || !bs->bl.request_alignment) {
Eric Blake8cc9c6e2016-06-23 16:37:25 -0600383 error_setg(errp, "Could not find working O_DIRECT alignment");
384 error_append_hint(errp, "Try cache.direct=off\n");
Kevin Wolfdf26a352014-07-16 17:48:17 +0200385 }
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100386}
387
Jeff Cody6a8dc042012-09-20 15:13:21 -0400388static void raw_parse_flags(int bdrv_flags, int *open_flags)
389{
390 assert(open_flags != NULL);
391
392 *open_flags |= O_BINARY;
393 *open_flags &= ~O_ACCMODE;
394 if (bdrv_flags & BDRV_O_RDWR) {
395 *open_flags |= O_RDWR;
396 } else {
397 *open_flags |= O_RDONLY;
398 }
399
400 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
401 * and O_DIRECT for no caching. */
402 if ((bdrv_flags & BDRV_O_NOCACHE)) {
403 *open_flags |= O_DIRECT;
404 }
Jeff Cody6a8dc042012-09-20 15:13:21 -0400405}
406
Max Reitz078896a2014-03-05 22:41:37 +0100407static void raw_parse_filename(const char *filename, QDict *options,
408 Error **errp)
409{
Max Reitz03c320d2017-05-22 21:52:16 +0200410 bdrv_parse_filename_strip_prefix(filename, "file:", options);
Max Reitz078896a2014-03-05 22:41:37 +0100411}
412
Kevin Wolfc66a6152013-04-02 10:47:40 +0200413static QemuOptsList raw_runtime_opts = {
414 .name = "raw",
415 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
416 .desc = {
417 {
418 .name = "filename",
419 .type = QEMU_OPT_STRING,
420 .help = "File name of the image",
421 },
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200422 {
423 .name = "aio",
424 .type = QEMU_OPT_STRING,
425 .help = "host AIO implementation (threads, native)",
426 },
Fam Zheng16b48d52017-05-03 00:35:50 +0800427 {
428 .name = "locking",
429 .type = QEMU_OPT_STRING,
430 .help = "file locking mode (on/off/auto, default: auto)",
431 },
Paolo Bonzini7c9e5272017-08-21 18:58:56 +0200432 {
433 .name = "pr-manager",
434 .type = QEMU_OPT_STRING,
435 .help = "id of persistent reservation manager object (default: none)",
436 },
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +0100437 {
438 .name = "x-check-cache-dropped",
439 .type = QEMU_OPT_BOOL,
440 .help = "check that page cache was dropped on live migration (default: off)"
441 },
Kevin Wolfc66a6152013-04-02 10:47:40 +0200442 { /* end of list */ }
443 },
444};
445
446static int raw_open_common(BlockDriverState *bs, QDict *options,
John Snow230ff732018-07-10 13:00:40 -0400447 int bdrv_flags, int open_flags,
448 bool device, Error **errp)
bellard83f64092006-08-01 16:21:11 +0000449{
450 BDRVRawState *s = bs->opaque;
Kevin Wolfc66a6152013-04-02 10:47:40 +0200451 QemuOpts *opts;
452 Error *local_err = NULL;
Kevin Wolf8bfea152014-04-11 19:16:36 +0200453 const char *filename = NULL;
Paolo Bonzini7c9e5272017-08-21 18:58:56 +0200454 const char *str;
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200455 BlockdevAioOptions aio, aio_default;
Christoph Hellwig0e1d8f42009-06-15 13:53:26 +0200456 int fd, ret;
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100457 struct stat st;
Fam Zheng244a5662017-05-03 00:35:56 +0800458 OnOffAuto locking;
bellard83f64092006-08-01 16:21:11 +0000459
Peter Crosthwaite87ea75d2014-01-01 18:49:17 -0800460 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
Kevin Wolfc66a6152013-04-02 10:47:40 +0200461 qemu_opts_absorb_qdict(opts, options, &local_err);
Markus Armbruster84d18f02014-01-30 15:07:28 +0100462 if (local_err) {
Max Reitze428e432013-10-11 11:37:01 +0200463 error_propagate(errp, local_err);
Kevin Wolfc66a6152013-04-02 10:47:40 +0200464 ret = -EINVAL;
465 goto fail;
466 }
467
468 filename = qemu_opt_get(opts, "filename");
469
Fam Zhengdb0754d2018-11-01 14:29:09 +0800470 ret = raw_normalize_devicepath(&filename, errp);
Christoph Egger1de1ae02011-05-24 11:30:29 +0200471 if (ret != 0) {
Kevin Wolfc66a6152013-04-02 10:47:40 +0200472 goto fail;
Christoph Egger1de1ae02011-05-24 11:30:29 +0200473 }
474
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200475 aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO)
476 ? BLOCKDEV_AIO_OPTIONS_NATIVE
477 : BLOCKDEV_AIO_OPTIONS_THREADS;
Marc-AndrƩ Lureauf7abe0e2017-08-24 10:46:10 +0200478 aio = qapi_enum_parse(&BlockdevAioOptions_lookup,
479 qemu_opt_get(opts, "aio"),
Markus Armbruster06c60b62017-08-24 10:45:57 +0200480 aio_default, &local_err);
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200481 if (local_err) {
482 error_propagate(errp, local_err);
483 ret = -EINVAL;
484 goto fail;
485 }
486 s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE);
487
Marc-AndrƩ Lureauf7abe0e2017-08-24 10:46:10 +0200488 locking = qapi_enum_parse(&OnOffAuto_lookup,
489 qemu_opt_get(opts, "locking"),
Markus Armbruster06c60b62017-08-24 10:45:57 +0200490 ON_OFF_AUTO_AUTO, &local_err);
Fam Zheng244a5662017-05-03 00:35:56 +0800491 if (local_err) {
492 error_propagate(errp, local_err);
493 ret = -EINVAL;
494 goto fail;
495 }
496 switch (locking) {
497 case ON_OFF_AUTO_ON:
498 s->use_lock = true;
Fam Zheng2b218f52017-08-11 19:44:47 +0800499 if (!qemu_has_ofd_lock()) {
Fam Zhengdb0754d2018-11-01 14:29:09 +0800500 warn_report("File lock requested but OFD locking syscall is "
501 "unavailable, falling back to POSIX file locks");
502 error_printf("Due to the implementation, locks can be lost "
503 "unexpectedly.\n");
Fam Zheng2b218f52017-08-11 19:44:47 +0800504 }
Fam Zheng244a5662017-05-03 00:35:56 +0800505 break;
506 case ON_OFF_AUTO_OFF:
507 s->use_lock = false;
508 break;
509 case ON_OFF_AUTO_AUTO:
Fam Zheng2b218f52017-08-11 19:44:47 +0800510 s->use_lock = qemu_has_ofd_lock();
Fam Zheng244a5662017-05-03 00:35:56 +0800511 break;
512 default:
513 abort();
514 }
515
Paolo Bonzini7c9e5272017-08-21 18:58:56 +0200516 str = qemu_opt_get(opts, "pr-manager");
517 if (str) {
518 s->pr_mgr = pr_manager_lookup(str, &local_err);
519 if (local_err) {
520 error_propagate(errp, local_err);
521 ret = -EINVAL;
522 goto fail;
523 }
524 }
525
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +0100526 s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped",
527 false);
528
Jeff Cody6a8dc042012-09-20 15:13:21 -0400529 s->open_flags = open_flags;
530 raw_parse_flags(bdrv_flags, &s->open_flags);
bellard83f64092006-08-01 16:21:11 +0000531
Christoph Hellwig90babde2009-06-15 13:53:38 +0200532 s->fd = -1;
Kevin Wolf40ff6d72009-12-02 12:24:42 +0100533 fd = qemu_open(filename, s->open_flags, 0644);
Kevin Wolf64107dc2018-10-08 17:27:18 +0200534 ret = fd < 0 ? -errno : 0;
535
536 if (ret == -EACCES || ret == -EROFS) {
537 /* Try to degrade to read-only, but if it doesn't work, still use the
538 * normal error message. */
539 if (bdrv_apply_auto_read_only(bs, NULL, NULL) == 0) {
540 bdrv_flags &= ~BDRV_O_RDWR;
541 raw_parse_flags(bdrv_flags, &s->open_flags);
542 assert(!(s->open_flags & O_CREAT));
543 fd = qemu_open(filename, s->open_flags);
544 ret = fd < 0 ? -errno : 0;
545 }
546 }
547
548 if (ret < 0) {
549 error_setg_errno(errp, -ret, "Could not open '%s'", filename);
Kevin Wolfc66a6152013-04-02 10:47:40 +0200550 if (ret == -EROFS) {
bellard19cb3732006-08-19 11:45:59 +0000551 ret = -EACCES;
Kevin Wolfc66a6152013-04-02 10:47:40 +0200552 }
553 goto fail;
bellard19cb3732006-08-19 11:45:59 +0000554 }
bellard83f64092006-08-01 16:21:11 +0000555 s->fd = fd;
Christoph Hellwig9ef91a62009-08-20 16:58:19 +0200556
Fam Zheng244a5662017-05-03 00:35:56 +0800557 s->perm = 0;
558 s->shared_perm = BLK_PERM_ALL;
559
Christoph Hellwig5c6c3a62009-08-20 16:58:35 +0200560#ifdef CONFIG_LINUX_AIO
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200561 /* Currently Linux does AIO only for files opened with O_DIRECT */
Nishanth Aravamudaned6e2162018-06-22 12:37:00 -0700562 if (s->use_linux_aio) {
563 if (!(s->open_flags & O_DIRECT)) {
564 error_setg(errp, "aio=native was specified, but it requires "
565 "cache.direct=on, which was not specified.");
566 ret = -EINVAL;
567 goto fail;
568 }
569 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) {
570 error_prepend(errp, "Unable to use native AIO: ");
571 goto fail;
572 }
Kevin Wolf96518252015-03-17 13:45:21 +0100573 }
Stefan Hajnoczi1501ecc2015-07-23 13:48:34 +0100574#else
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200575 if (s->use_linux_aio) {
Kevin Wolfd657c0c2015-12-15 11:35:36 +0100576 error_setg(errp, "aio=native was specified, but is not supported "
577 "in this build.");
578 ret = -EINVAL;
579 goto fail;
Stefan Hajnoczi1501ecc2015-07-23 13:48:34 +0100580 }
581#endif /* !defined(CONFIG_LINUX_AIO) */
Christoph Hellwig9ef91a62009-08-20 16:58:19 +0200582
Paolo Bonzini7ce21012013-11-22 13:39:47 +0100583 s->has_discard = true;
Paolo Bonzini97a2ae32013-11-22 13:39:57 +0100584 s->has_write_zeroes = true;
Roger Pau Monne3cad8302014-10-21 16:03:03 +0200585 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
586 s->needs_alignment = true;
587 }
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100588
589 if (fstat(s->fd, &st) < 0) {
Max Reitz01212d42014-12-02 18:32:53 +0100590 ret = -errno;
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100591 error_setg_errno(errp, errno, "Could not stat file");
592 goto fail;
593 }
John Snow230ff732018-07-10 13:00:40 -0400594
595 if (!device) {
596 if (S_ISBLK(st.st_mode)) {
597 warn_report("Opening a block device as a file using the '%s' "
598 "driver is deprecated", bs->drv->format_name);
599 } else if (S_ISCHR(st.st_mode)) {
600 warn_report("Opening a character device as a file using the '%s' "
601 "driver is deprecated", bs->drv->format_name);
602 } else if (!S_ISREG(st.st_mode)) {
603 error_setg(errp, "A regular file was expected by the '%s' driver, "
604 "but something else was given", bs->drv->format_name);
605 ret = -EINVAL;
606 goto fail;
607 } else {
608 s->discard_zeroes = true;
609 s->has_fallocate = true;
610 }
611 } else {
612 if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
613 error_setg(errp, "'%s' driver expects either "
614 "a character or block device", bs->drv->format_name);
615 ret = -EINVAL;
616 goto fail;
617 }
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100618 }
John Snow230ff732018-07-10 13:00:40 -0400619
Paolo Bonzinid0b45032013-11-22 13:39:56 +0100620 if (S_ISBLK(st.st_mode)) {
621#ifdef BLKDISCARDZEROES
622 unsigned int arg;
623 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
624 s->discard_zeroes = true;
625 }
626#endif
627#ifdef __linux__
628 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
629 * not rely on the contents of discarded blocks unless using O_DIRECT.
Paolo Bonzini97a2ae32013-11-22 13:39:57 +0100630 * Same for BLKZEROOUT.
Paolo Bonzinid0b45032013-11-22 13:39:56 +0100631 */
632 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
633 s->discard_zeroes = false;
Paolo Bonzini97a2ae32013-11-22 13:39:57 +0100634 s->has_write_zeroes = false;
Paolo Bonzinid0b45032013-11-22 13:39:56 +0100635 }
636#endif
637 }
Roger Pau Monne3cad8302014-10-21 16:03:03 +0200638#ifdef __FreeBSD__
639 if (S_ISCHR(st.st_mode)) {
640 /*
641 * The file is a char device (disk), which on FreeBSD isn't behind
642 * a pager, so force all requests to be aligned. This is needed
643 * so QEMU makes sure all IO operations on the device are aligned
644 * to sector size, or else FreeBSD will reject them with EINVAL.
645 */
646 s->needs_alignment = true;
647 }
648#endif
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100649
Christoph Hellwigdce512d2010-12-17 11:41:15 +0100650#ifdef CONFIG_XFS
651 if (platform_test_xfs_fd(s->fd)) {
Paolo Bonzini7ce21012013-11-22 13:39:47 +0100652 s->is_xfs = true;
Christoph Hellwigdce512d2010-12-17 11:41:15 +0100653 }
654#endif
655
Kevin Wolf34fa1102018-07-26 11:28:30 +0200656 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP;
Kevin Wolfc66a6152013-04-02 10:47:40 +0200657 ret = 0;
658fail:
Kevin Wolf8bfea152014-04-11 19:16:36 +0200659 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
660 unlink(filename);
661 }
Kevin Wolfc66a6152013-04-02 10:47:40 +0200662 qemu_opts_del(opts);
663 return ret;
bellard83f64092006-08-01 16:21:11 +0000664}
665
Max Reitz015a1032013-09-05 14:22:29 +0200666static int raw_open(BlockDriverState *bs, QDict *options, int flags,
667 Error **errp)
Christoph Hellwig90babde2009-06-15 13:53:38 +0200668{
669 BDRVRawState *s = bs->opaque;
670
671 s->type = FTYPE_FILE;
John Snow230ff732018-07-10 13:00:40 -0400672 return raw_open_common(bs, options, flags, 0, false, errp);
Christoph Hellwig90babde2009-06-15 13:53:38 +0200673}
674
Fam Zheng244a5662017-05-03 00:35:56 +0800675typedef enum {
676 RAW_PL_PREPARE,
677 RAW_PL_COMMIT,
678 RAW_PL_ABORT,
679} RawPermLockOp;
680
681#define PERM_FOREACH(i) \
682 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
683
684/* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
685 * file; if @unlock == true, also unlock the unneeded bytes.
686 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
687 */
Fam Zheng2996ffa2018-10-11 15:21:33 +0800688static int raw_apply_lock_bytes(BDRVRawState *s, int fd,
Fam Zheng244a5662017-05-03 00:35:56 +0800689 uint64_t perm_lock_bits,
690 uint64_t shared_perm_lock_bits,
691 bool unlock, Error **errp)
692{
693 int ret;
694 int i;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800695 uint64_t locked_perm, locked_shared_perm;
696
697 if (s) {
698 locked_perm = s->locked_perm;
699 locked_shared_perm = s->locked_shared_perm;
700 } else {
701 /*
702 * We don't have the previous bits, just lock/unlock for each of the
703 * requested bits.
704 */
705 if (unlock) {
706 locked_perm = BLK_PERM_ALL;
707 locked_shared_perm = BLK_PERM_ALL;
708 } else {
709 locked_perm = 0;
710 locked_shared_perm = 0;
711 }
712 }
Fam Zheng244a5662017-05-03 00:35:56 +0800713
714 PERM_FOREACH(i) {
715 int off = RAW_LOCK_PERM_BASE + i;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800716 uint64_t bit = (1ULL << i);
717 if ((perm_lock_bits & bit) && !(locked_perm & bit)) {
Max Reitzd0a96152018-05-09 23:53:34 +0200718 ret = qemu_lock_fd(fd, off, 1, false);
Fam Zheng244a5662017-05-03 00:35:56 +0800719 if (ret) {
720 error_setg(errp, "Failed to lock byte %d", off);
721 return ret;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800722 } else if (s) {
723 s->locked_perm |= bit;
Fam Zheng244a5662017-05-03 00:35:56 +0800724 }
Fam Zheng2996ffa2018-10-11 15:21:33 +0800725 } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) {
Max Reitzd0a96152018-05-09 23:53:34 +0200726 ret = qemu_unlock_fd(fd, off, 1);
Fam Zheng244a5662017-05-03 00:35:56 +0800727 if (ret) {
728 error_setg(errp, "Failed to unlock byte %d", off);
729 return ret;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800730 } else if (s) {
731 s->locked_perm &= ~bit;
Fam Zheng244a5662017-05-03 00:35:56 +0800732 }
733 }
734 }
735 PERM_FOREACH(i) {
736 int off = RAW_LOCK_SHARED_BASE + i;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800737 uint64_t bit = (1ULL << i);
738 if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) {
Max Reitzd0a96152018-05-09 23:53:34 +0200739 ret = qemu_lock_fd(fd, off, 1, false);
Fam Zheng244a5662017-05-03 00:35:56 +0800740 if (ret) {
741 error_setg(errp, "Failed to lock byte %d", off);
742 return ret;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800743 } else if (s) {
744 s->locked_shared_perm |= bit;
Fam Zheng244a5662017-05-03 00:35:56 +0800745 }
Fam Zheng2996ffa2018-10-11 15:21:33 +0800746 } else if (unlock && (locked_shared_perm & bit) &&
747 !(shared_perm_lock_bits & bit)) {
Max Reitzd0a96152018-05-09 23:53:34 +0200748 ret = qemu_unlock_fd(fd, off, 1);
Fam Zheng244a5662017-05-03 00:35:56 +0800749 if (ret) {
750 error_setg(errp, "Failed to unlock byte %d", off);
751 return ret;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800752 } else if (s) {
753 s->locked_shared_perm &= ~bit;
Fam Zheng244a5662017-05-03 00:35:56 +0800754 }
755 }
756 }
757 return 0;
758}
759
760/* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
Max Reitzd0a96152018-05-09 23:53:34 +0200761static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm,
Fam Zheng244a5662017-05-03 00:35:56 +0800762 Error **errp)
763{
764 int ret;
765 int i;
766
767 PERM_FOREACH(i) {
768 int off = RAW_LOCK_SHARED_BASE + i;
769 uint64_t p = 1ULL << i;
770 if (perm & p) {
Max Reitzd0a96152018-05-09 23:53:34 +0200771 ret = qemu_lock_fd_test(fd, off, 1, true);
Fam Zheng244a5662017-05-03 00:35:56 +0800772 if (ret) {
773 char *perm_name = bdrv_perm_names(p);
774 error_setg(errp,
775 "Failed to get \"%s\" lock",
776 perm_name);
777 g_free(perm_name);
Fam Zheng244a5662017-05-03 00:35:56 +0800778 return ret;
779 }
780 }
781 }
782 PERM_FOREACH(i) {
783 int off = RAW_LOCK_PERM_BASE + i;
784 uint64_t p = 1ULL << i;
785 if (!(shared_perm & p)) {
Max Reitzd0a96152018-05-09 23:53:34 +0200786 ret = qemu_lock_fd_test(fd, off, 1, true);
Fam Zheng244a5662017-05-03 00:35:56 +0800787 if (ret) {
788 char *perm_name = bdrv_perm_names(p);
789 error_setg(errp,
790 "Failed to get shared \"%s\" lock",
791 perm_name);
792 g_free(perm_name);
Fam Zheng244a5662017-05-03 00:35:56 +0800793 return ret;
794 }
795 }
796 }
797 return 0;
798}
799
800static int raw_handle_perm_lock(BlockDriverState *bs,
801 RawPermLockOp op,
802 uint64_t new_perm, uint64_t new_shared,
803 Error **errp)
804{
805 BDRVRawState *s = bs->opaque;
806 int ret = 0;
807 Error *local_err = NULL;
808
809 if (!s->use_lock) {
810 return 0;
811 }
812
813 if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) {
814 return 0;
815 }
816
Fam Zheng244a5662017-05-03 00:35:56 +0800817 switch (op) {
818 case RAW_PL_PREPARE:
Fam Zhengf2e3af22018-10-11 15:21:34 +0800819 ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm,
Fam Zheng244a5662017-05-03 00:35:56 +0800820 ~s->shared_perm | ~new_shared,
821 false, errp);
822 if (!ret) {
Fam Zhengf2e3af22018-10-11 15:21:34 +0800823 ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp);
Fam Zheng244a5662017-05-03 00:35:56 +0800824 if (!ret) {
825 return 0;
826 }
Fam Zhengb8574312018-09-25 13:05:01 +0800827 error_append_hint(errp,
828 "Is another process using the image [%s]?\n",
829 bs->filename);
Fam Zheng244a5662017-05-03 00:35:56 +0800830 }
831 op = RAW_PL_ABORT;
832 /* fall through to unlock bytes. */
833 case RAW_PL_ABORT:
Fam Zhengf2e3af22018-10-11 15:21:34 +0800834 raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm,
Max Reitzd0a96152018-05-09 23:53:34 +0200835 true, &local_err);
Fam Zheng244a5662017-05-03 00:35:56 +0800836 if (local_err) {
837 /* Theoretically the above call only unlocks bytes and it cannot
838 * fail. Something weird happened, report it.
839 */
Fam Zhengdb0754d2018-11-01 14:29:09 +0800840 warn_report_err(local_err);
Fam Zheng244a5662017-05-03 00:35:56 +0800841 }
842 break;
843 case RAW_PL_COMMIT:
Fam Zhengf2e3af22018-10-11 15:21:34 +0800844 raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared,
Max Reitzd0a96152018-05-09 23:53:34 +0200845 true, &local_err);
Fam Zheng244a5662017-05-03 00:35:56 +0800846 if (local_err) {
847 /* Theoretically the above call only unlocks bytes and it cannot
848 * fail. Something weird happened, report it.
849 */
Fam Zhengdb0754d2018-11-01 14:29:09 +0800850 warn_report_err(local_err);
Fam Zheng244a5662017-05-03 00:35:56 +0800851 }
852 break;
853 }
854 return ret;
855}
856
Jeff Codyeeb6b452012-09-20 15:13:25 -0400857static int raw_reopen_prepare(BDRVReopenState *state,
858 BlockReopenQueue *queue, Error **errp)
859{
860 BDRVRawState *s;
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800861 BDRVRawReopenState *rs;
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +0100862 QemuOpts *opts;
Jeff Codyeeb6b452012-09-20 15:13:25 -0400863 int ret = 0;
Kevin Wolfdf26a352014-07-16 17:48:17 +0200864 Error *local_err = NULL;
Jeff Codyeeb6b452012-09-20 15:13:25 -0400865
866 assert(state != NULL);
867 assert(state->bs != NULL);
868
869 s = state->bs->opaque;
870
Markus Armbruster5839e532014-08-19 10:31:08 +0200871 state->opaque = g_new0(BDRVRawReopenState, 1);
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800872 rs = state->opaque;
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +0100873 rs->fd = -1;
874
875 /* Handle options changes */
876 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
877 qemu_opts_absorb_qdict(opts, state->options, &local_err);
878 if (local_err) {
879 error_propagate(errp, local_err);
880 ret = -EINVAL;
881 goto out;
882 }
883
Alberto Garcia8d324572018-09-06 12:37:07 +0300884 rs->check_cache_dropped =
885 qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false);
886
887 /* This driver's reopen function doesn't currently allow changing
888 * other options, so let's put them back in the original QDict and
889 * bdrv_reopen_prepare() will detect changes and complain. */
890 qemu_opts_to_qdict(opts, state->options);
Jeff Codyeeb6b452012-09-20 15:13:25 -0400891
Max Reitzf7096232015-10-19 17:53:07 +0200892 if (s->type == FTYPE_CD) {
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800893 rs->open_flags |= O_NONBLOCK;
Jeff Cody1bc6b702012-11-20 10:21:10 -0500894 }
895
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800896 raw_parse_flags(state->flags, &rs->open_flags);
Jeff Codyeeb6b452012-09-20 15:13:25 -0400897
Andreas FƤrberfdf263f2013-01-31 15:40:14 +0100898 int fcntl_flags = O_APPEND | O_NONBLOCK;
Jeff Codyeeb6b452012-09-20 15:13:25 -0400899#ifdef O_NOATIME
900 fcntl_flags |= O_NOATIME;
901#endif
902
Andreas FƤrberfdf263f2013-01-31 15:40:14 +0100903#ifdef O_ASYNC
904 /* Not all operating systems have O_ASYNC, and those that don't
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800905 * will not let us track the state into rs->open_flags (typically
Andreas FƤrberfdf263f2013-01-31 15:40:14 +0100906 * you achieve the same effect with an ioctl, for example I_SETSIG
907 * on Solaris). But we do not use O_ASYNC, so that's fine.
908 */
909 assert((s->open_flags & O_ASYNC) == 0);
910#endif
911
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800912 if ((rs->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
Jeff Codyeeb6b452012-09-20 15:13:25 -0400913 /* dup the original fd */
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800914 rs->fd = qemu_dup(s->fd);
915 if (rs->fd >= 0) {
916 ret = fcntl_setfl(rs->fd, rs->open_flags);
Jeff Codyeeb6b452012-09-20 15:13:25 -0400917 if (ret) {
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800918 qemu_close(rs->fd);
919 rs->fd = -1;
Jeff Codyeeb6b452012-09-20 15:13:25 -0400920 }
921 }
922 }
923
924 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800925 if (rs->fd == -1) {
Max Reitzbdd03cd2015-08-12 17:33:31 +0200926 const char *normalized_filename = state->bs->filename;
Fam Zhengdb0754d2018-11-01 14:29:09 +0800927 ret = raw_normalize_devicepath(&normalized_filename, errp);
928 if (ret >= 0) {
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800929 assert(!(rs->open_flags & O_CREAT));
930 rs->fd = qemu_open(normalized_filename, rs->open_flags);
931 if (rs->fd == -1) {
Max Reitzbdd03cd2015-08-12 17:33:31 +0200932 error_setg_errno(errp, errno, "Could not reopen file");
933 ret = -1;
934 }
Jeff Codyeeb6b452012-09-20 15:13:25 -0400935 }
936 }
Kevin Wolfdf26a352014-07-16 17:48:17 +0200937
938 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
939 * alignment with the new fd. */
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800940 if (rs->fd != -1) {
941 raw_probe_alignment(state->bs, rs->fd, &local_err);
Kevin Wolfdf26a352014-07-16 17:48:17 +0200942 if (local_err) {
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800943 qemu_close(rs->fd);
944 rs->fd = -1;
Kevin Wolfdf26a352014-07-16 17:48:17 +0200945 error_propagate(errp, local_err);
946 ret = -EINVAL;
947 }
948 }
949
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +0100950out:
951 qemu_opts_del(opts);
Jeff Codyeeb6b452012-09-20 15:13:25 -0400952 return ret;
953}
954
Jeff Codyeeb6b452012-09-20 15:13:25 -0400955static void raw_reopen_commit(BDRVReopenState *state)
956{
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800957 BDRVRawReopenState *rs = state->opaque;
Jeff Codyeeb6b452012-09-20 15:13:25 -0400958 BDRVRawState *s = state->bs->opaque;
Fam Zhengf2e3af22018-10-11 15:21:34 +0800959 Error *local_err = NULL;
Jeff Codyeeb6b452012-09-20 15:13:25 -0400960
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +0100961 s->check_cache_dropped = rs->check_cache_dropped;
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800962 s->open_flags = rs->open_flags;
Jeff Codyeeb6b452012-09-20 15:13:25 -0400963
Fam Zhengf2e3af22018-10-11 15:21:34 +0800964 /* Copy locks to the new fd before closing the old one. */
965 raw_apply_lock_bytes(NULL, rs->fd, s->locked_perm,
Max Reitz577a1332018-11-16 17:45:25 +0100966 s->locked_shared_perm, false, &local_err);
Fam Zhengf2e3af22018-10-11 15:21:34 +0800967 if (local_err) {
968 /* shouldn't fail in a sane host, but report it just in case. */
969 error_report_err(local_err);
970 }
Jeff Codyeeb6b452012-09-20 15:13:25 -0400971 qemu_close(s->fd);
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800972 s->fd = rs->fd;
Jeff Codyeeb6b452012-09-20 15:13:25 -0400973
974 g_free(state->opaque);
975 state->opaque = NULL;
976}
977
978
979static void raw_reopen_abort(BDRVReopenState *state)
980{
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800981 BDRVRawReopenState *rs = state->opaque;
Jeff Codyeeb6b452012-09-20 15:13:25 -0400982
983 /* nothing to do if NULL, we didn't get far enough */
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800984 if (rs == NULL) {
Jeff Codyeeb6b452012-09-20 15:13:25 -0400985 return;
986 }
987
Fam Zheng4e6d13c2016-10-27 18:45:17 +0800988 if (rs->fd >= 0) {
989 qemu_close(rs->fd);
990 rs->fd = -1;
Jeff Codyeeb6b452012-09-20 15:13:25 -0400991 }
992 g_free(state->opaque);
993 state->opaque = NULL;
994}
995
Eric Farman48265252017-01-20 17:25:26 +0100996static int hdev_get_max_transfer_length(BlockDriverState *bs, int fd)
Fam Zheng6f607172016-06-03 10:07:02 +0800997{
998#ifdef BLKSECTGET
Eric Farman48265252017-01-20 17:25:26 +0100999 int max_bytes = 0;
1000 short max_sectors = 0;
1001 if (bs->sg && ioctl(fd, BLKSECTGET, &max_bytes) == 0) {
1002 return max_bytes;
1003 } else if (!bs->sg && ioctl(fd, BLKSECTGET, &max_sectors) == 0) {
1004 return max_sectors << BDRV_SECTOR_BITS;
Fam Zheng6f607172016-06-03 10:07:02 +08001005 } else {
1006 return -errno;
1007 }
1008#else
1009 return -ENOSYS;
1010#endif
1011}
1012
Fam Zheng9103f1c2017-03-08 20:08:14 +08001013static int hdev_get_max_segments(const struct stat *st)
1014{
1015#ifdef CONFIG_LINUX
1016 char buf[32];
1017 const char *end;
1018 char *sysfspath;
1019 int ret;
1020 int fd = -1;
1021 long max_segments;
1022
1023 sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1024 major(st->st_rdev), minor(st->st_rdev));
1025 fd = open(sysfspath, O_RDONLY);
1026 if (fd == -1) {
1027 ret = -errno;
1028 goto out;
1029 }
1030 do {
Stefan Hajnoczi69583492017-03-14 17:09:22 +08001031 ret = read(fd, buf, sizeof(buf) - 1);
Fam Zheng9103f1c2017-03-08 20:08:14 +08001032 } while (ret == -1 && errno == EINTR);
1033 if (ret < 0) {
1034 ret = -errno;
1035 goto out;
1036 } else if (ret == 0) {
1037 ret = -EIO;
1038 goto out;
1039 }
1040 buf[ret] = 0;
1041 /* The file is ended with '\n', pass 'end' to accept that. */
1042 ret = qemu_strtol(buf, &end, 10, &max_segments);
1043 if (ret == 0 && end && *end == '\n') {
1044 ret = max_segments;
1045 }
1046
1047out:
Fam Zhengfed414d2017-03-15 00:12:05 +08001048 if (fd != -1) {
1049 close(fd);
1050 }
Fam Zheng9103f1c2017-03-08 20:08:14 +08001051 g_free(sysfspath);
1052 return ret;
1053#else
1054 return -ENOTSUP;
1055#endif
1056}
1057
Kevin Wolf3baca892014-07-16 17:48:16 +02001058static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
Paolo Bonzinic25f53b2011-11-29 12:42:20 +01001059{
1060 BDRVRawState *s = bs->opaque;
Fam Zheng6f607172016-06-03 10:07:02 +08001061 struct stat st;
1062
1063 if (!fstat(s->fd, &st)) {
Eric Farmanc4c41a02017-01-20 17:25:27 +01001064 if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode)) {
Eric Farman48265252017-01-20 17:25:26 +01001065 int ret = hdev_get_max_transfer_length(bs, s->fd);
1066 if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) {
1067 bs->bl.max_transfer = pow2floor(ret);
Fam Zheng6f607172016-06-03 10:07:02 +08001068 }
Fam Zheng9103f1c2017-03-08 20:08:14 +08001069 ret = hdev_get_max_segments(&st);
1070 if (ret > 0) {
1071 bs->bl.max_transfer = MIN(bs->bl.max_transfer,
1072 ret * getpagesize());
1073 }
Fam Zheng6f607172016-06-03 10:07:02 +08001074 }
1075 }
Jeff Codyeeb6b452012-09-20 15:13:25 -04001076
Kevin Wolfdf26a352014-07-16 17:48:17 +02001077 raw_probe_alignment(bs, s->fd, errp);
Denis V. Lunev4196d2f2015-05-12 17:30:55 +03001078 bs->bl.min_mem_alignment = s->buf_align;
Denis V. Lunev459b4e62015-05-12 17:30:56 +03001079 bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize());
Paolo Bonzinic25f53b2011-11-29 12:42:20 +01001080}
bellard83f64092006-08-01 16:21:11 +00001081
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +01001082static int check_for_dasd(int fd)
1083{
1084#ifdef BIODASDINFO2
1085 struct dasd_information2_t info = {0};
1086
1087 return ioctl(fd, BIODASDINFO2, &info);
1088#else
1089 return -1;
1090#endif
1091}
1092
1093/**
1094 * Try to get @bs's logical and physical block size.
1095 * On success, store them in @bsz and return zero.
1096 * On failure, return negative errno.
1097 */
1098static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
1099{
1100 BDRVRawState *s = bs->opaque;
1101 int ret;
1102
1103 /* If DASD, get blocksizes */
1104 if (check_for_dasd(s->fd) < 0) {
1105 return -ENOTSUP;
1106 }
1107 ret = probe_logical_blocksize(s->fd, &bsz->log);
1108 if (ret < 0) {
1109 return ret;
1110 }
1111 return probe_physical_blocksize(s->fd, &bsz->phys);
1112}
1113
1114/**
1115 * Try to get @bs's geometry: cyls, heads, sectors.
1116 * On success, store them in @geo and return 0.
1117 * On failure return -errno.
1118 * (Allows block driver to assign default geometry values that guest sees)
1119 */
1120#ifdef __linux__
1121static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1122{
1123 BDRVRawState *s = bs->opaque;
1124 struct hd_geometry ioctl_geo = {0};
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +01001125
1126 /* If DASD, get its geometry */
1127 if (check_for_dasd(s->fd) < 0) {
1128 return -ENOTSUP;
1129 }
1130 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
1131 return -errno;
1132 }
1133 /* HDIO_GETGEO may return success even though geo contains zeros
1134 (e.g. certain multipath setups) */
1135 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
1136 return -ENOTSUP;
1137 }
1138 /* Do not return a geometry for partition */
1139 if (ioctl_geo.start != 0) {
1140 return -ENOTSUP;
1141 }
1142 geo->heads = ioctl_geo.heads;
1143 geo->sectors = ioctl_geo.sectors;
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +01001144 geo->cylinders = ioctl_geo.cylinders;
1145
1146 return 0;
1147}
1148#else /* __linux__ */
1149static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1150{
1151 return -ENOTSUP;
1152}
1153#endif
1154
Kevin Wolf03425672018-10-31 11:30:42 +01001155#if defined(__linux__)
1156static int handle_aiocb_ioctl(void *opaque)
Paolo Bonzinide81a162012-05-25 11:46:27 +02001157{
Kevin Wolf03425672018-10-31 11:30:42 +01001158 RawPosixAIOData *aiocb = opaque;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001159 int ret;
1160
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001161 ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf);
Paolo Bonzinide81a162012-05-25 11:46:27 +02001162 if (ret == -1) {
1163 return -errno;
1164 }
1165
Paolo Bonzinib608c8d2013-01-10 15:28:35 +01001166 return 0;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001167}
Kevin Wolf03425672018-10-31 11:30:42 +01001168#endif /* linux */
Paolo Bonzinide81a162012-05-25 11:46:27 +02001169
Kevin Wolf06dc9bd2018-10-25 14:18:58 +01001170static int handle_aiocb_flush(void *opaque)
Paolo Bonzinide81a162012-05-25 11:46:27 +02001171{
Kevin Wolf06dc9bd2018-10-25 14:18:58 +01001172 RawPosixAIOData *aiocb = opaque;
Kevin Wolfe5bcf962017-03-22 22:00:05 +01001173 BDRVRawState *s = aiocb->bs->opaque;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001174 int ret;
1175
Kevin Wolfe5bcf962017-03-22 22:00:05 +01001176 if (s->page_cache_inconsistent) {
1177 return -EIO;
1178 }
1179
Paolo Bonzinide81a162012-05-25 11:46:27 +02001180 ret = qemu_fdatasync(aiocb->aio_fildes);
1181 if (ret == -1) {
Kevin Wolfe5bcf962017-03-22 22:00:05 +01001182 /* There is no clear definition of the semantics of a failing fsync(),
1183 * so we may have to assume the worst. The sad truth is that this
1184 * assumption is correct for Linux. Some pages are now probably marked
1185 * clean in the page cache even though they are inconsistent with the
1186 * on-disk contents. The next fdatasync() call would succeed, but no
1187 * further writeback attempt will be made. We can't get back to a state
1188 * in which we know what is on disk (we would have to rewrite
1189 * everything that was touched since the last fdatasync() at least), so
1190 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1191 * really defined, I have little hope that other OSes are doing better.
1192 *
1193 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1194 * cache. */
1195 if ((s->open_flags & O_DIRECT) == 0) {
1196 s->page_cache_inconsistent = true;
1197 }
Paolo Bonzinide81a162012-05-25 11:46:27 +02001198 return -errno;
1199 }
1200 return 0;
1201}
1202
1203#ifdef CONFIG_PREADV
1204
1205static bool preadv_present = true;
1206
1207static ssize_t
1208qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1209{
1210 return preadv(fd, iov, nr_iov, offset);
1211}
1212
1213static ssize_t
1214qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1215{
1216 return pwritev(fd, iov, nr_iov, offset);
1217}
1218
1219#else
1220
1221static bool preadv_present = false;
1222
1223static ssize_t
1224qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1225{
1226 return -ENOSYS;
1227}
1228
1229static ssize_t
1230qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1231{
1232 return -ENOSYS;
1233}
1234
1235#endif
1236
1237static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
1238{
1239 ssize_t len;
1240
1241 do {
1242 if (aiocb->aio_type & QEMU_AIO_WRITE)
1243 len = qemu_pwritev(aiocb->aio_fildes,
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001244 aiocb->io.iov,
1245 aiocb->io.niov,
Paolo Bonzinide81a162012-05-25 11:46:27 +02001246 aiocb->aio_offset);
1247 else
1248 len = qemu_preadv(aiocb->aio_fildes,
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001249 aiocb->io.iov,
1250 aiocb->io.niov,
Paolo Bonzinide81a162012-05-25 11:46:27 +02001251 aiocb->aio_offset);
1252 } while (len == -1 && errno == EINTR);
1253
1254 if (len == -1) {
1255 return -errno;
1256 }
1257 return len;
1258}
1259
1260/*
1261 * Read/writes the data to/from a given linear buffer.
1262 *
1263 * Returns the number of bytes handles or -errno in case of an error. Short
1264 * reads are only returned if the end of the file is reached.
1265 */
1266static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
1267{
1268 ssize_t offset = 0;
1269 ssize_t len;
1270
1271 while (offset < aiocb->aio_nbytes) {
1272 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1273 len = pwrite(aiocb->aio_fildes,
1274 (const char *)buf + offset,
1275 aiocb->aio_nbytes - offset,
1276 aiocb->aio_offset + offset);
1277 } else {
1278 len = pread(aiocb->aio_fildes,
1279 buf + offset,
1280 aiocb->aio_nbytes - offset,
1281 aiocb->aio_offset + offset);
1282 }
1283 if (len == -1 && errno == EINTR) {
1284 continue;
Stefan Hajnoczi61ed73c2014-08-21 13:44:07 +01001285 } else if (len == -1 && errno == EINVAL &&
1286 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
1287 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
1288 offset > 0) {
1289 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1290 * after a short read. Assume that O_DIRECT short reads only occur
1291 * at EOF. Therefore this is a short read, not an I/O error.
1292 */
1293 break;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001294 } else if (len == -1) {
1295 offset = -errno;
1296 break;
1297 } else if (len == 0) {
1298 break;
1299 }
1300 offset += len;
1301 }
1302
1303 return offset;
1304}
1305
Kevin Wolf999e6b62018-10-25 14:18:58 +01001306static int handle_aiocb_rw(void *opaque)
Paolo Bonzinide81a162012-05-25 11:46:27 +02001307{
Kevin Wolf999e6b62018-10-25 14:18:58 +01001308 RawPosixAIOData *aiocb = opaque;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001309 ssize_t nbytes;
1310 char *buf;
1311
1312 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
1313 /*
1314 * If there is just a single buffer, and it is properly aligned
1315 * we can just use plain pread/pwrite without any problems.
1316 */
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001317 if (aiocb->io.niov == 1) {
Kevin Wolf54c7ca12018-10-31 09:43:17 +01001318 nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base);
1319 goto out;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001320 }
1321 /*
1322 * We have more than one iovec, and all are properly aligned.
1323 *
1324 * Try preadv/pwritev first and fall back to linearizing the
1325 * buffer if it's not supported.
1326 */
1327 if (preadv_present) {
1328 nbytes = handle_aiocb_rw_vector(aiocb);
1329 if (nbytes == aiocb->aio_nbytes ||
1330 (nbytes < 0 && nbytes != -ENOSYS)) {
Kevin Wolf54c7ca12018-10-31 09:43:17 +01001331 goto out;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001332 }
1333 preadv_present = false;
1334 }
1335
1336 /*
1337 * XXX(hch): short read/write. no easy way to handle the reminder
1338 * using these interfaces. For now retry using plain
1339 * pread/pwrite?
1340 */
1341 }
1342
1343 /*
1344 * Ok, we have to do it the hard way, copy all segments into
1345 * a single aligned buffer.
1346 */
Kevin Wolf50d4a852014-05-21 18:02:42 +02001347 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
1348 if (buf == NULL) {
Kevin Wolf54c7ca12018-10-31 09:43:17 +01001349 nbytes = -ENOMEM;
1350 goto out;
Kevin Wolf50d4a852014-05-21 18:02:42 +02001351 }
1352
Paolo Bonzinide81a162012-05-25 11:46:27 +02001353 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1354 char *p = buf;
1355 int i;
1356
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001357 for (i = 0; i < aiocb->io.niov; ++i) {
1358 memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len);
1359 p += aiocb->io.iov[i].iov_len;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001360 }
Kevin Wolf8eb029c2014-07-01 16:09:54 +02001361 assert(p - buf == aiocb->aio_nbytes);
Paolo Bonzinide81a162012-05-25 11:46:27 +02001362 }
1363
1364 nbytes = handle_aiocb_rw_linear(aiocb, buf);
1365 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
1366 char *p = buf;
1367 size_t count = aiocb->aio_nbytes, copy;
1368 int i;
1369
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001370 for (i = 0; i < aiocb->io.niov && count; ++i) {
Paolo Bonzinide81a162012-05-25 11:46:27 +02001371 copy = count;
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001372 if (copy > aiocb->io.iov[i].iov_len) {
1373 copy = aiocb->io.iov[i].iov_len;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001374 }
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001375 memcpy(aiocb->io.iov[i].iov_base, p, copy);
Kevin Wolf8eb029c2014-07-01 16:09:54 +02001376 assert(count >= copy);
Paolo Bonzinide81a162012-05-25 11:46:27 +02001377 p += copy;
1378 count -= copy;
1379 }
Kevin Wolf8eb029c2014-07-01 16:09:54 +02001380 assert(count == 0);
Paolo Bonzinide81a162012-05-25 11:46:27 +02001381 }
1382 qemu_vfree(buf);
1383
Kevin Wolf54c7ca12018-10-31 09:43:17 +01001384out:
1385 if (nbytes == aiocb->aio_nbytes) {
1386 return 0;
1387 } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) {
1388 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1389 return -EINVAL;
1390 } else {
1391 iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes,
1392 0, aiocb->aio_nbytes - nbytes);
1393 return 0;
1394 }
1395 } else {
1396 assert(nbytes < 0);
1397 return nbytes;
1398 }
Paolo Bonzinide81a162012-05-25 11:46:27 +02001399}
1400
Paolo Bonzini82380102013-01-14 16:26:55 +01001401#ifdef CONFIG_XFS
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001402static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
1403{
1404 struct xfs_flock64 fl;
Dimitris Aragiorgisbcb22552015-06-23 13:44:58 +03001405 int err;
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001406
1407 memset(&fl, 0, sizeof(fl));
1408 fl.l_whence = SEEK_SET;
1409 fl.l_start = offset;
1410 fl.l_len = bytes;
1411
1412 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
Dimitris Aragiorgisbcb22552015-06-23 13:44:58 +03001413 err = errno;
1414 DPRINTF("cannot write zero range (%s)\n", strerror(errno));
1415 return -err;
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001416 }
1417
1418 return 0;
1419}
1420
Paolo Bonzini82380102013-01-14 16:26:55 +01001421static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
1422{
1423 struct xfs_flock64 fl;
Dimitris Aragiorgisbcb22552015-06-23 13:44:58 +03001424 int err;
Paolo Bonzini82380102013-01-14 16:26:55 +01001425
1426 memset(&fl, 0, sizeof(fl));
1427 fl.l_whence = SEEK_SET;
1428 fl.l_start = offset;
1429 fl.l_len = bytes;
1430
1431 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
Dimitris Aragiorgisbcb22552015-06-23 13:44:58 +03001432 err = errno;
1433 DPRINTF("cannot punch hole (%s)\n", strerror(errno));
1434 return -err;
Paolo Bonzini82380102013-01-14 16:26:55 +01001435 }
1436
1437 return 0;
1438}
1439#endif
1440
Denis V. Lunev1486df02015-01-30 11:42:11 +03001441static int translate_err(int err)
1442{
1443 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1444 err == -ENOTTY) {
1445 err = -ENOTSUP;
1446 }
1447 return err;
1448}
1449
Denis V. Lunevd50d8222015-01-30 11:42:15 +03001450#ifdef CONFIG_FALLOCATE
Denis V. Lunev0b991712015-01-30 11:42:12 +03001451static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1452{
1453 do {
1454 if (fallocate(fd, mode, offset, len) == 0) {
1455 return 0;
1456 }
1457 } while (errno == EINTR);
1458 return translate_err(-errno);
1459}
1460#endif
1461
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001462static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001463{
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001464 int ret = -ENOTSUP;
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001465 BDRVRawState *s = aiocb->bs->opaque;
1466
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001467 if (!s->has_write_zeroes) {
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001468 return -ENOTSUP;
1469 }
1470
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001471#ifdef BLKZEROOUT
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001472 do {
1473 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1474 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1475 return 0;
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001476 }
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001477 } while (errno == EINTR);
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001478
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001479 ret = translate_err(-errno);
1480#endif
1481
Denis V. Lunev1486df02015-01-30 11:42:11 +03001482 if (ret == -ENOTSUP) {
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001483 s->has_write_zeroes = false;
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001484 }
1485 return ret;
1486}
1487
Kevin Wolf7154d8a2018-10-25 14:18:58 +01001488static int handle_aiocb_write_zeroes(void *opaque)
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001489{
Kevin Wolf7154d8a2018-10-25 14:18:58 +01001490 RawPosixAIOData *aiocb = opaque;
Denis V. Luneva6dcf092015-02-12 08:35:49 +03001491#if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS)
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001492 BDRVRawState *s = aiocb->bs->opaque;
Denis V. Luneva6dcf092015-02-12 08:35:49 +03001493#endif
Denis V. Lunev70d91102017-08-04 18:10:11 +03001494#ifdef CONFIG_FALLOCATE
1495 int64_t len;
1496#endif
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001497
1498 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1499 return handle_aiocb_write_zeroes_block(aiocb);
1500 }
1501
1502#ifdef CONFIG_XFS
1503 if (s->is_xfs) {
1504 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
1505 }
1506#endif
1507
Denis V. Lunevb953f072015-01-30 11:42:14 +03001508#ifdef CONFIG_FALLOCATE_ZERO_RANGE
1509 if (s->has_write_zeroes) {
1510 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1511 aiocb->aio_offset, aiocb->aio_nbytes);
1512 if (ret == 0 || ret != -ENOTSUP) {
1513 return ret;
1514 }
1515 s->has_write_zeroes = false;
1516 }
1517#endif
1518
Denis V. Lunev1cdc3232015-01-30 11:42:16 +03001519#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1520 if (s->has_discard && s->has_fallocate) {
1521 int ret = do_fallocate(s->fd,
1522 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1523 aiocb->aio_offset, aiocb->aio_nbytes);
1524 if (ret == 0) {
1525 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1526 if (ret == 0 || ret != -ENOTSUP) {
1527 return ret;
1528 }
1529 s->has_fallocate = false;
1530 } else if (ret != -ENOTSUP) {
1531 return ret;
1532 } else {
1533 s->has_discard = false;
1534 }
1535 }
1536#endif
1537
Denis V. Lunevd50d8222015-01-30 11:42:15 +03001538#ifdef CONFIG_FALLOCATE
Denis V. Lunev70d91102017-08-04 18:10:11 +03001539 /* Last resort: we are trying to extend the file with zeroed data. This
1540 * can be done via fallocate(fd, 0) */
1541 len = bdrv_getlength(aiocb->bs);
1542 if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) {
Denis V. Lunevd50d8222015-01-30 11:42:15 +03001543 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1544 if (ret == 0 || ret != -ENOTSUP) {
1545 return ret;
1546 }
1547 s->has_fallocate = false;
1548 }
1549#endif
1550
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001551 return -ENOTSUP;
1552}
1553
Kevin Wolf7154d8a2018-10-25 14:18:58 +01001554static int handle_aiocb_write_zeroes_unmap(void *opaque)
Kevin Wolf34fa1102018-07-26 11:28:30 +02001555{
Kevin Wolf7154d8a2018-10-25 14:18:58 +01001556 RawPosixAIOData *aiocb = opaque;
Kevin Wolf34fa1102018-07-26 11:28:30 +02001557 BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque;
1558 int ret;
1559
1560 /* First try to write zeros and unmap at the same time */
1561
1562#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1563 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1564 aiocb->aio_offset, aiocb->aio_nbytes);
1565 if (ret != -ENOTSUP) {
1566 return ret;
1567 }
1568#endif
1569
1570#ifdef CONFIG_XFS
1571 if (s->is_xfs) {
1572 /* xfs_discard() guarantees that the discarded area reads as all-zero
1573 * afterwards, so we can use it here. */
1574 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1575 }
1576#endif
1577
1578 /* If we couldn't manage to unmap while guaranteed that the area reads as
1579 * all-zero afterwards, just write zeroes without unmapping */
1580 ret = handle_aiocb_write_zeroes(aiocb);
1581 return ret;
1582}
1583
Fam Zheng1efad062018-06-01 17:26:43 +08001584#ifndef HAVE_COPY_FILE_RANGE
1585static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd,
1586 off_t *out_off, size_t len, unsigned int flags)
1587{
1588#ifdef __NR_copy_file_range
1589 return syscall(__NR_copy_file_range, in_fd, in_off, out_fd,
1590 out_off, len, flags);
1591#else
1592 errno = ENOSYS;
1593 return -1;
1594#endif
1595}
1596#endif
1597
Kevin Wolf58a209c2018-10-25 14:18:58 +01001598static int handle_aiocb_copy_range(void *opaque)
Fam Zheng1efad062018-06-01 17:26:43 +08001599{
Kevin Wolf58a209c2018-10-25 14:18:58 +01001600 RawPosixAIOData *aiocb = opaque;
Fam Zheng1efad062018-06-01 17:26:43 +08001601 uint64_t bytes = aiocb->aio_nbytes;
1602 off_t in_off = aiocb->aio_offset;
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001603 off_t out_off = aiocb->copy_range.aio_offset2;
Fam Zheng1efad062018-06-01 17:26:43 +08001604
1605 while (bytes) {
1606 ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off,
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001607 aiocb->copy_range.aio_fd2, &out_off,
Fam Zheng1efad062018-06-01 17:26:43 +08001608 bytes, 0);
Fam Zhengecc983a2018-07-10 14:31:16 +08001609 trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off,
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001610 aiocb->copy_range.aio_fd2, out_off, bytes,
1611 0, ret);
Fam Zhengc436e3d2018-06-29 14:03:28 +08001612 if (ret == 0) {
1613 /* No progress (e.g. when beyond EOF), let the caller fall back to
1614 * buffer I/O. */
1615 return -ENOSPC;
Fam Zheng1efad062018-06-01 17:26:43 +08001616 }
1617 if (ret < 0) {
Fam Zhengc436e3d2018-06-29 14:03:28 +08001618 switch (errno) {
1619 case ENOSYS:
Fam Zheng1efad062018-06-01 17:26:43 +08001620 return -ENOTSUP;
Fam Zhengc436e3d2018-06-29 14:03:28 +08001621 case EINTR:
1622 continue;
1623 default:
Fam Zheng1efad062018-06-01 17:26:43 +08001624 return -errno;
1625 }
1626 }
Fam Zheng1efad062018-06-01 17:26:43 +08001627 bytes -= ret;
1628 }
1629 return 0;
1630}
1631
Kevin Wolf46ee0f42018-10-25 14:18:58 +01001632static int handle_aiocb_discard(void *opaque)
Paolo Bonzini82380102013-01-14 16:26:55 +01001633{
Kevin Wolf46ee0f42018-10-25 14:18:58 +01001634 RawPosixAIOData *aiocb = opaque;
Paolo Bonzini82380102013-01-14 16:26:55 +01001635 int ret = -EOPNOTSUPP;
1636 BDRVRawState *s = aiocb->bs->opaque;
1637
Paolo Bonzini7ce21012013-11-22 13:39:47 +01001638 if (!s->has_discard) {
1639 return -ENOTSUP;
Paolo Bonzini82380102013-01-14 16:26:55 +01001640 }
1641
1642 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1643#ifdef BLKDISCARD
1644 do {
1645 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1646 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
1647 return 0;
1648 }
1649 } while (errno == EINTR);
1650
1651 ret = -errno;
1652#endif
1653 } else {
1654#ifdef CONFIG_XFS
1655 if (s->is_xfs) {
1656 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
1657 }
1658#endif
1659
1660#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
Denis V. Lunev0b991712015-01-30 11:42:12 +03001661 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1662 aiocb->aio_offset, aiocb->aio_nbytes);
Paolo Bonzini82380102013-01-14 16:26:55 +01001663#endif
1664 }
1665
Denis V. Lunev1486df02015-01-30 11:42:11 +03001666 ret = translate_err(ret);
1667 if (ret == -ENOTSUP) {
Paolo Bonzini7ce21012013-11-22 13:39:47 +01001668 s->has_discard = false;
Paolo Bonzini82380102013-01-14 16:26:55 +01001669 }
1670 return ret;
1671}
1672
Kevin Wolf29cb4c02018-10-25 14:18:58 +01001673static int handle_aiocb_truncate(void *opaque)
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02001674{
Kevin Wolf29cb4c02018-10-25 14:18:58 +01001675 RawPosixAIOData *aiocb = opaque;
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02001676 int result = 0;
1677 int64_t current_length = 0;
1678 char *buf = NULL;
1679 struct stat st;
1680 int fd = aiocb->aio_fildes;
1681 int64_t offset = aiocb->aio_offset;
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001682 PreallocMode prealloc = aiocb->truncate.prealloc;
1683 Error **errp = aiocb->truncate.errp;
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02001684
1685 if (fstat(fd, &st) < 0) {
1686 result = -errno;
1687 error_setg_errno(errp, -result, "Could not stat file");
1688 return result;
1689 }
1690
1691 current_length = st.st_size;
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001692 if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02001693 error_setg(errp, "Cannot use preallocation for shrinking files");
1694 return -ENOTSUP;
1695 }
1696
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001697 switch (prealloc) {
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02001698#ifdef CONFIG_POSIX_FALLOCATE
1699 case PREALLOC_MODE_FALLOC:
1700 /*
1701 * Truncating before posix_fallocate() makes it about twice slower on
1702 * file systems that do not support fallocate(), trying to check if a
1703 * block is allocated before allocating it, so don't do that here.
1704 */
1705 if (offset != current_length) {
1706 result = -posix_fallocate(fd, current_length,
1707 offset - current_length);
1708 if (result != 0) {
1709 /* posix_fallocate() doesn't set errno. */
1710 error_setg_errno(errp, -result,
1711 "Could not preallocate new data");
1712 }
1713 } else {
1714 result = 0;
1715 }
1716 goto out;
1717#endif
1718 case PREALLOC_MODE_FULL:
1719 {
1720 int64_t num = 0, left = offset - current_length;
1721 off_t seek_result;
1722
1723 /*
1724 * Knowing the final size from the beginning could allow the file
1725 * system driver to do less allocations and possibly avoid
1726 * fragmentation of the file.
1727 */
1728 if (ftruncate(fd, offset) != 0) {
1729 result = -errno;
1730 error_setg_errno(errp, -result, "Could not resize file");
1731 goto out;
1732 }
1733
1734 buf = g_malloc0(65536);
1735
1736 seek_result = lseek(fd, current_length, SEEK_SET);
1737 if (seek_result < 0) {
1738 result = -errno;
1739 error_setg_errno(errp, -result,
1740 "Failed to seek to the old end of file");
1741 goto out;
1742 }
1743
1744 while (left > 0) {
1745 num = MIN(left, 65536);
1746 result = write(fd, buf, num);
1747 if (result < 0) {
Fam Zhenga1c81f42018-07-27 14:53:14 +08001748 if (errno == EINTR) {
1749 continue;
1750 }
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02001751 result = -errno;
1752 error_setg_errno(errp, -result,
1753 "Could not write zeros for preallocation");
1754 goto out;
1755 }
1756 left -= result;
1757 }
1758 if (result >= 0) {
1759 result = fsync(fd);
1760 if (result < 0) {
1761 result = -errno;
1762 error_setg_errno(errp, -result,
1763 "Could not flush file to disk");
1764 goto out;
1765 }
1766 }
1767 goto out;
1768 }
1769 case PREALLOC_MODE_OFF:
1770 if (ftruncate(fd, offset) != 0) {
1771 result = -errno;
1772 error_setg_errno(errp, -result, "Could not resize file");
1773 }
1774 return result;
1775 default:
1776 result = -ENOTSUP;
1777 error_setg(errp, "Unsupported preallocation mode: %s",
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001778 PreallocMode_str(prealloc));
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02001779 return result;
1780 }
1781
1782out:
1783 if (result < 0) {
1784 if (ftruncate(fd, current_length) < 0) {
1785 error_report("Failed to restore old file length: %s",
1786 strerror(errno));
1787 }
1788 }
1789
1790 g_free(buf);
1791 return result;
1792}
1793
Kevin Wolf5d5de252018-10-26 16:53:47 +01001794static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs,
1795 ThreadPoolFunc func, void *arg)
1796{
1797 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1798 ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1799 return thread_pool_submit_co(pool, func, arg);
1800}
1801
Kevin Wolf9d52aa32016-06-03 17:36:27 +02001802static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset,
1803 uint64_t bytes, QEMUIOVector *qiov, int type)
bellard83f64092006-08-01 16:21:11 +00001804{
pbrookce1a14d2006-08-07 02:38:06 +00001805 BDRVRawState *s = bs->opaque;
Kevin Wolf999e6b62018-10-25 14:18:58 +01001806 RawPosixAIOData acb;
bellard83f64092006-08-01 16:21:11 +00001807
bellard19cb3732006-08-19 11:45:59 +00001808 if (fd_open(bs) < 0)
Kevin Wolf2174f122014-08-06 17:18:07 +02001809 return -EIO;
bellard19cb3732006-08-19 11:45:59 +00001810
aliguorif141eaf2009-04-07 18:43:24 +00001811 /*
Roger Pau Monne3cad8302014-10-21 16:03:03 +02001812 * Check if the underlying device requires requests to be aligned,
1813 * and if the request we are trying to submit is aligned or not.
1814 * If this is the case tell the low-level driver that it needs
1815 * to copy the buffer.
aliguorif141eaf2009-04-07 18:43:24 +00001816 */
Roger Pau Monne3cad8302014-10-21 16:03:03 +02001817 if (s->needs_alignment) {
Stefan Hajnoczic53b1c52013-01-11 16:41:27 +01001818 if (!bdrv_qiov_is_aligned(bs, qiov)) {
Christoph Hellwig5c6c3a62009-08-20 16:58:35 +02001819 type |= QEMU_AIO_MISALIGNED;
Stefan Weile44bd6f2009-08-28 14:39:31 +02001820#ifdef CONFIG_LINUX_AIO
Kevin Wolf0a4279d2016-09-08 15:09:01 +02001821 } else if (s->use_linux_aio) {
Paolo Bonzini0187f5c2016-07-04 18:33:20 +02001822 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
Kevin Wolf9d52aa32016-06-03 17:36:27 +02001823 assert(qiov->size == bytes);
Paolo Bonzini0187f5c2016-07-04 18:33:20 +02001824 return laio_co_submit(bs, aio, s->fd, offset, qiov, type);
Stefan Weile44bd6f2009-08-28 14:39:31 +02001825#endif
Christoph Hellwig5c6c3a62009-08-20 16:58:35 +02001826 }
Christoph Hellwig9ef91a62009-08-20 16:58:19 +02001827 }
aliguorif141eaf2009-04-07 18:43:24 +00001828
Kevin Wolf999e6b62018-10-25 14:18:58 +01001829 acb = (RawPosixAIOData) {
1830 .bs = bs,
1831 .aio_fildes = s->fd,
1832 .aio_type = type,
1833 .aio_offset = offset,
1834 .aio_nbytes = bytes,
1835 .io = {
1836 .iov = qiov->iov,
1837 .niov = qiov->niov,
1838 },
1839 };
1840
1841 assert(qiov->size == bytes);
1842 return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb);
Kevin Wolf2174f122014-08-06 17:18:07 +02001843}
1844
Kevin Wolf9d52aa32016-06-03 17:36:27 +02001845static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset,
1846 uint64_t bytes, QEMUIOVector *qiov,
1847 int flags)
Kevin Wolf2174f122014-08-06 17:18:07 +02001848{
Kevin Wolf9d52aa32016-06-03 17:36:27 +02001849 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ);
Kevin Wolf2174f122014-08-06 17:18:07 +02001850}
1851
Kevin Wolf9d52aa32016-06-03 17:36:27 +02001852static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset,
1853 uint64_t bytes, QEMUIOVector *qiov,
1854 int flags)
Kevin Wolf2174f122014-08-06 17:18:07 +02001855{
Kevin Wolf9d52aa32016-06-03 17:36:27 +02001856 assert(flags == 0);
1857 return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE);
bellard83f64092006-08-01 16:21:11 +00001858}
1859
Ming Lei1b3abdc2014-07-04 18:04:34 +08001860static void raw_aio_plug(BlockDriverState *bs)
1861{
1862#ifdef CONFIG_LINUX_AIO
Kevin Wolf0a4279d2016-09-08 15:09:01 +02001863 BDRVRawState *s = bs->opaque;
1864 if (s->use_linux_aio) {
Paolo Bonzini0187f5c2016-07-04 18:33:20 +02001865 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1866 laio_io_plug(bs, aio);
Ming Lei1b3abdc2014-07-04 18:04:34 +08001867 }
1868#endif
1869}
1870
1871static void raw_aio_unplug(BlockDriverState *bs)
1872{
1873#ifdef CONFIG_LINUX_AIO
Kevin Wolf0a4279d2016-09-08 15:09:01 +02001874 BDRVRawState *s = bs->opaque;
1875 if (s->use_linux_aio) {
Paolo Bonzini0187f5c2016-07-04 18:33:20 +02001876 LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs));
1877 laio_io_unplug(bs, aio);
Ming Lei1b3abdc2014-07-04 18:04:34 +08001878 }
1879#endif
1880}
1881
Kevin Wolf33d70fb2018-06-21 19:07:32 +02001882static int raw_co_flush_to_disk(BlockDriverState *bs)
Christoph Hellwigb2e12bc2009-09-04 19:01:49 +02001883{
1884 BDRVRawState *s = bs->opaque;
Kevin Wolf06dc9bd2018-10-25 14:18:58 +01001885 RawPosixAIOData acb;
Kevin Wolf33d70fb2018-06-21 19:07:32 +02001886 int ret;
Christoph Hellwigb2e12bc2009-09-04 19:01:49 +02001887
Kevin Wolf33d70fb2018-06-21 19:07:32 +02001888 ret = fd_open(bs);
1889 if (ret < 0) {
1890 return ret;
1891 }
Christoph Hellwigb2e12bc2009-09-04 19:01:49 +02001892
Kevin Wolf06dc9bd2018-10-25 14:18:58 +01001893 acb = (RawPosixAIOData) {
1894 .bs = bs,
1895 .aio_fildes = s->fd,
1896 .aio_type = QEMU_AIO_FLUSH,
1897 };
1898
1899 return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb);
Christoph Hellwigb2e12bc2009-09-04 19:01:49 +02001900}
1901
Nishanth Aravamudaned6e2162018-06-22 12:37:00 -07001902static void raw_aio_attach_aio_context(BlockDriverState *bs,
1903 AioContext *new_context)
1904{
1905#ifdef CONFIG_LINUX_AIO
1906 BDRVRawState *s = bs->opaque;
1907 if (s->use_linux_aio) {
1908 Error *local_err;
1909 if (!aio_setup_linux_aio(new_context, &local_err)) {
1910 error_reportf_err(local_err, "Unable to use native AIO, "
1911 "falling back to thread pool: ");
1912 s->use_linux_aio = false;
1913 }
1914 }
1915#endif
1916}
1917
bellard83f64092006-08-01 16:21:11 +00001918static void raw_close(BlockDriverState *bs)
1919{
1920 BDRVRawState *s = bs->opaque;
Stefan Hajnoczic2f34262014-05-08 16:34:47 +02001921
bellard19cb3732006-08-19 11:45:59 +00001922 if (s->fd >= 0) {
Corey Bryant2e1e79d2012-08-14 16:43:46 -04001923 qemu_close(s->fd);
bellard19cb3732006-08-19 11:45:59 +00001924 s->fd = -1;
1925 }
bellard83f64092006-08-01 16:21:11 +00001926}
1927
Max Reitzd0bc9e52017-06-13 22:20:58 +02001928/**
1929 * Truncates the given regular file @fd to @offset and, when growing, fills the
1930 * new space according to @prealloc.
1931 *
1932 * Returns: 0 on success, -errno on failure.
1933 */
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02001934static int coroutine_fn
1935raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset,
1936 PreallocMode prealloc, Error **errp)
Max Reitz9f63b072017-06-13 22:20:57 +02001937{
Kevin Wolf29cb4c02018-10-25 14:18:58 +01001938 RawPosixAIOData acb;
Max Reitzd0bc9e52017-06-13 22:20:58 +02001939
Kevin Wolf29cb4c02018-10-25 14:18:58 +01001940 acb = (RawPosixAIOData) {
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02001941 .bs = bs,
1942 .aio_fildes = fd,
1943 .aio_type = QEMU_AIO_TRUNCATE,
1944 .aio_offset = offset,
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001945 .truncate = {
1946 .prealloc = prealloc,
1947 .errp = errp,
1948 },
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02001949 };
Max Reitzd0bc9e52017-06-13 22:20:58 +02001950
Kevin Wolf29cb4c02018-10-25 14:18:58 +01001951 return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb);
Max Reitz9f63b072017-06-13 22:20:57 +02001952}
1953
Kevin Wolf061ca8a2018-06-21 17:54:35 +02001954static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset,
1955 PreallocMode prealloc, Error **errp)
bellard83f64092006-08-01 16:21:11 +00001956{
1957 BDRVRawState *s = bs->opaque;
Christoph Hellwig55b949c2011-09-21 01:10:37 +02001958 struct stat st;
Max Reitzf59adb32017-03-28 22:51:29 +02001959 int ret;
Christoph Hellwig55b949c2011-09-21 01:10:37 +02001960
1961 if (fstat(s->fd, &st)) {
Max Reitzf59adb32017-03-28 22:51:29 +02001962 ret = -errno;
1963 error_setg_errno(errp, -ret, "Failed to fstat() the file");
1964 return ret;
Christoph Hellwig55b949c2011-09-21 01:10:37 +02001965 }
1966
1967 if (S_ISREG(st.st_mode)) {
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02001968 return raw_regular_truncate(bs, s->fd, offset, prealloc, errp);
Max Reitz35d72602017-06-13 22:20:59 +02001969 }
1970
1971 if (prealloc != PREALLOC_MODE_OFF) {
1972 error_setg(errp, "Preallocation mode '%s' unsupported for this "
Markus Armbruster977c7362017-08-24 10:46:08 +02001973 "non-regular file", PreallocMode_str(prealloc));
Max Reitz35d72602017-06-13 22:20:59 +02001974 return -ENOTSUP;
1975 }
1976
1977 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
Max Reitzf59adb32017-03-28 22:51:29 +02001978 if (offset > raw_getlength(bs)) {
1979 error_setg(errp, "Cannot grow device files");
1980 return -EINVAL;
1981 }
Christoph Hellwig55b949c2011-09-21 01:10:37 +02001982 } else {
Max Reitzf59adb32017-03-28 22:51:29 +02001983 error_setg(errp, "Resizing this file is not supported");
Christoph Hellwig55b949c2011-09-21 01:10:37 +02001984 return -ENOTSUP;
1985 }
1986
bellard83f64092006-08-01 16:21:11 +00001987 return 0;
1988}
1989
blueswir1128ab2f2008-08-15 18:33:42 +00001990#ifdef __OpenBSD__
1991static int64_t raw_getlength(BlockDriverState *bs)
1992{
1993 BDRVRawState *s = bs->opaque;
1994 int fd = s->fd;
1995 struct stat st;
1996
1997 if (fstat(fd, &st))
Markus Armbrusteraa729702014-06-26 13:23:16 +02001998 return -errno;
blueswir1128ab2f2008-08-15 18:33:42 +00001999 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2000 struct disklabel dl;
2001
2002 if (ioctl(fd, DIOCGDINFO, &dl))
Markus Armbrusteraa729702014-06-26 13:23:16 +02002003 return -errno;
blueswir1128ab2f2008-08-15 18:33:42 +00002004 return (uint64_t)dl.d_secsize *
2005 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
2006 } else
2007 return st.st_size;
2008}
Christoph Eggerd1f6fd82011-05-23 14:31:17 +02002009#elif defined(__NetBSD__)
2010static int64_t raw_getlength(BlockDriverState *bs)
2011{
2012 BDRVRawState *s = bs->opaque;
2013 int fd = s->fd;
2014 struct stat st;
2015
2016 if (fstat(fd, &st))
Markus Armbrusteraa729702014-06-26 13:23:16 +02002017 return -errno;
Christoph Eggerd1f6fd82011-05-23 14:31:17 +02002018 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2019 struct dkwedge_info dkw;
2020
2021 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
2022 return dkw.dkw_size * 512;
2023 } else {
2024 struct disklabel dl;
2025
2026 if (ioctl(fd, DIOCGDINFO, &dl))
Markus Armbrusteraa729702014-06-26 13:23:16 +02002027 return -errno;
Christoph Eggerd1f6fd82011-05-23 14:31:17 +02002028 return (uint64_t)dl.d_secsize *
2029 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
2030 }
2031 } else
2032 return st.st_size;
2033}
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002034#elif defined(__sun__)
2035static int64_t raw_getlength(BlockDriverState *bs)
2036{
2037 BDRVRawState *s = bs->opaque;
2038 struct dk_minfo minfo;
2039 int ret;
Markus Armbrusteraa729702014-06-26 13:23:16 +02002040 int64_t size;
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002041
2042 ret = fd_open(bs);
2043 if (ret < 0) {
2044 return ret;
2045 }
2046
2047 /*
2048 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2049 */
2050 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
2051 if (ret != -1) {
2052 return minfo.dki_lbsize * minfo.dki_capacity;
2053 }
2054
2055 /*
2056 * There are reports that lseek on some devices fails, but
2057 * irc discussion said that contingency on contingency was overkill.
2058 */
Markus Armbrusteraa729702014-06-26 13:23:16 +02002059 size = lseek(s->fd, 0, SEEK_END);
2060 if (size < 0) {
2061 return -errno;
2062 }
2063 return size;
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002064}
2065#elif defined(CONFIG_BSD)
2066static int64_t raw_getlength(BlockDriverState *bs)
bellard83f64092006-08-01 16:21:11 +00002067{
2068 BDRVRawState *s = bs->opaque;
2069 int fd = s->fd;
2070 int64_t size;
bellard83f64092006-08-01 16:21:11 +00002071 struct stat sb;
Aurelien Jarnoa167ba52009-11-29 18:00:41 +01002072#if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
blueswir19f230112009-03-28 08:37:13 +00002073 int reopened = 0;
2074#endif
bellard19cb3732006-08-19 11:45:59 +00002075 int ret;
2076
2077 ret = fd_open(bs);
2078 if (ret < 0)
2079 return ret;
bellard83f64092006-08-01 16:21:11 +00002080
Aurelien Jarnoa167ba52009-11-29 18:00:41 +01002081#if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
blueswir19f230112009-03-28 08:37:13 +00002082again:
2083#endif
bellard83f64092006-08-01 16:21:11 +00002084 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
2085#ifdef DIOCGMEDIASIZE
Paolo Bonzini7d374352018-12-13 23:37:37 +01002086 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
blueswir1c5e97232009-03-07 20:06:23 +00002087#elif defined(DIOCGPART)
2088 {
2089 struct partinfo pi;
2090 if (ioctl(fd, DIOCGPART, &pi) == 0)
2091 size = pi.media_size;
2092 else
2093 size = 0;
2094 }
2095 if (size == 0)
bellard83f64092006-08-01 16:21:11 +00002096#endif
Pavel Borzenkov83affaa2011-11-10 22:40:06 +04002097#if defined(__APPLE__) && defined(__MACH__)
Programmingkid728dacb2015-01-19 17:12:55 -05002098 {
2099 uint64_t sectors = 0;
2100 uint32_t sector_size = 0;
2101
2102 if (ioctl(fd, DKIOCGETBLOCKCOUNT, &sectors) == 0
2103 && ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) == 0) {
2104 size = sectors * sector_size;
2105 } else {
2106 size = lseek(fd, 0LL, SEEK_END);
2107 if (size < 0) {
2108 return -errno;
2109 }
2110 }
2111 }
bellard83f64092006-08-01 16:21:11 +00002112#else
2113 size = lseek(fd, 0LL, SEEK_END);
Markus Armbrusteraa729702014-06-26 13:23:16 +02002114 if (size < 0) {
2115 return -errno;
2116 }
bellard83f64092006-08-01 16:21:11 +00002117#endif
Aurelien Jarnoa167ba52009-11-29 18:00:41 +01002118#if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
blueswir19f230112009-03-28 08:37:13 +00002119 switch(s->type) {
2120 case FTYPE_CD:
2121 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2122 if (size == 2048LL * (unsigned)-1)
2123 size = 0;
2124 /* XXX no disc? maybe we need to reopen... */
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02002125 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
blueswir19f230112009-03-28 08:37:13 +00002126 reopened = 1;
2127 goto again;
2128 }
2129 }
2130#endif
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002131 } else {
bellard83f64092006-08-01 16:21:11 +00002132 size = lseek(fd, 0, SEEK_END);
Markus Armbrusteraa729702014-06-26 13:23:16 +02002133 if (size < 0) {
2134 return -errno;
2135 }
bellard83f64092006-08-01 16:21:11 +00002136 }
bellard83f64092006-08-01 16:21:11 +00002137 return size;
2138}
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002139#else
2140static int64_t raw_getlength(BlockDriverState *bs)
2141{
2142 BDRVRawState *s = bs->opaque;
2143 int ret;
Markus Armbrusteraa729702014-06-26 13:23:16 +02002144 int64_t size;
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002145
2146 ret = fd_open(bs);
2147 if (ret < 0) {
2148 return ret;
2149 }
2150
Markus Armbrusteraa729702014-06-26 13:23:16 +02002151 size = lseek(s->fd, 0, SEEK_END);
2152 if (size < 0) {
2153 return -errno;
2154 }
2155 return size;
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002156}
blueswir1128ab2f2008-08-15 18:33:42 +00002157#endif
bellard83f64092006-08-01 16:21:11 +00002158
Fam Zheng4a1d5e12011-07-12 19:56:39 +08002159static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
2160{
2161 struct stat st;
2162 BDRVRawState *s = bs->opaque;
2163
2164 if (fstat(s->fd, &st) < 0) {
2165 return -errno;
2166 }
2167 return (int64_t)st.st_blocks * 512;
2168}
2169
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002170static int coroutine_fn
2171raw_co_create(BlockdevCreateOptions *options, Error **errp)
bellard83f64092006-08-01 16:21:11 +00002172{
Kevin Wolf927f11e2018-01-16 16:04:21 +01002173 BlockdevCreateOptionsFile *file_opts;
Max Reitz7c20c802018-07-04 16:47:51 +02002174 Error *local_err = NULL;
bellard83f64092006-08-01 16:21:11 +00002175 int fd;
Max Reitzd815efc2018-07-04 16:47:50 +02002176 uint64_t perm, shared;
Stefan Weil1e37d052009-07-11 16:43:37 +02002177 int result = 0;
bellard83f64092006-08-01 16:21:11 +00002178
Kevin Wolf927f11e2018-01-16 16:04:21 +01002179 /* Validate options and set default values */
2180 assert(options->driver == BLOCKDEV_DRIVER_FILE);
2181 file_opts = &options->u.file;
Max Reitz464d9f62014-03-05 22:41:38 +01002182
Kevin Wolf927f11e2018-01-16 16:04:21 +01002183 if (!file_opts->has_nocow) {
2184 file_opts->nocow = false;
2185 }
2186 if (!file_opts->has_preallocation) {
2187 file_opts->preallocation = PREALLOC_MODE_OFF;
Hu Tao06247422014-09-10 17:05:48 +08002188 }
bellard83f64092006-08-01 16:21:11 +00002189
Kevin Wolf927f11e2018-01-16 16:04:21 +01002190 /* Create file */
Max Reitzb8cf1912018-05-09 23:53:35 +02002191 fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_BINARY, 0644);
Stefan Weil1e37d052009-07-11 16:43:37 +02002192 if (fd < 0) {
2193 result = -errno;
Max Reitze428e432013-10-11 11:37:01 +02002194 error_setg_errno(errp, -result, "Could not create file");
Hu Tao06247422014-09-10 17:05:48 +08002195 goto out;
Stefan Weil1e37d052009-07-11 16:43:37 +02002196 }
Hu Tao06247422014-09-10 17:05:48 +08002197
Max Reitzb8cf1912018-05-09 23:53:35 +02002198 /* Take permissions: We want to discard everything, so we need
2199 * BLK_PERM_WRITE; and truncation to the desired size requires
2200 * BLK_PERM_RESIZE.
2201 * On the other hand, we cannot share the RESIZE permission
2202 * because we promise that after this function, the file has the
2203 * size given in the options. If someone else were to resize it
2204 * concurrently, we could not guarantee that.
2205 * Note that after this function, we can no longer guarantee that
2206 * the file is not touched by a third party, so it may be resized
2207 * then. */
2208 perm = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2209 shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE;
2210
2211 /* Step one: Take locks */
Fam Zheng2996ffa2018-10-11 15:21:33 +08002212 result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp);
Max Reitzb8cf1912018-05-09 23:53:35 +02002213 if (result < 0) {
2214 goto out_close;
2215 }
2216
2217 /* Step two: Check that nobody else has taken conflicting locks */
2218 result = raw_check_lock_bytes(fd, perm, shared, errp);
2219 if (result < 0) {
Fam Zhengb8574312018-09-25 13:05:01 +08002220 error_append_hint(errp,
2221 "Is another process using the image [%s]?\n",
2222 file_opts->filename);
Max Reitz7c20c802018-07-04 16:47:51 +02002223 goto out_unlock;
Max Reitzb8cf1912018-05-09 23:53:35 +02002224 }
2225
2226 /* Clear the file by truncating it to 0 */
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002227 result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp);
Max Reitzb8cf1912018-05-09 23:53:35 +02002228 if (result < 0) {
Max Reitz7c20c802018-07-04 16:47:51 +02002229 goto out_unlock;
Max Reitzb8cf1912018-05-09 23:53:35 +02002230 }
2231
Kevin Wolf927f11e2018-01-16 16:04:21 +01002232 if (file_opts->nocow) {
Hu Tao06247422014-09-10 17:05:48 +08002233#ifdef __linux__
2234 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2235 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2236 * will be ignored since any failure of this operation should not
2237 * block the left work.
2238 */
2239 int attr;
2240 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
2241 attr |= FS_NOCOW_FL;
2242 ioctl(fd, FS_IOC_SETFLAGS, &attr);
2243 }
2244#endif
2245 }
2246
Max Reitzb8cf1912018-05-09 23:53:35 +02002247 /* Resize and potentially preallocate the file to the desired
2248 * final size */
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002249 result = raw_regular_truncate(NULL, fd, file_opts->size,
2250 file_opts->preallocation, errp);
Max Reitz9f63b072017-06-13 22:20:57 +02002251 if (result < 0) {
Max Reitz7c20c802018-07-04 16:47:51 +02002252 goto out_unlock;
2253 }
2254
2255out_unlock:
Fam Zheng2996ffa2018-10-11 15:21:33 +08002256 raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err);
Max Reitz7c20c802018-07-04 16:47:51 +02002257 if (local_err) {
2258 /* The above call should not fail, and if it does, that does
2259 * not mean the whole creation operation has failed. So
2260 * report it the user for their convenience, but do not report
2261 * it to the caller. */
Fam Zhengdb0754d2018-11-01 14:29:09 +08002262 warn_report_err(local_err);
Hu Tao06247422014-09-10 17:05:48 +08002263 }
2264
Nir Soffer5a1dad92017-02-17 02:51:26 +02002265out_close:
Hu Tao06247422014-09-10 17:05:48 +08002266 if (qemu_close(fd) != 0 && result == 0) {
2267 result = -errno;
2268 error_setg_errno(errp, -result, "Could not close the new file");
2269 }
2270out:
Stefan Weil1e37d052009-07-11 16:43:37 +02002271 return result;
bellard83f64092006-08-01 16:21:11 +00002272}
2273
Kevin Wolf927f11e2018-01-16 16:04:21 +01002274static int coroutine_fn raw_co_create_opts(const char *filename, QemuOpts *opts,
2275 Error **errp)
2276{
2277 BlockdevCreateOptions options;
2278 int64_t total_size = 0;
2279 bool nocow = false;
2280 PreallocMode prealloc;
2281 char *buf = NULL;
2282 Error *local_err = NULL;
2283
2284 /* Skip file: protocol prefix */
2285 strstart(filename, "file:", &filename);
2286
2287 /* Read out options */
2288 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2289 BDRV_SECTOR_SIZE);
2290 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
2291 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
2292 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
2293 PREALLOC_MODE_OFF, &local_err);
2294 g_free(buf);
2295 if (local_err) {
2296 error_propagate(errp, local_err);
2297 return -EINVAL;
2298 }
2299
2300 options = (BlockdevCreateOptions) {
2301 .driver = BLOCKDEV_DRIVER_FILE,
2302 .u.file = {
2303 .filename = (char *) filename,
2304 .size = total_size,
2305 .has_preallocation = true,
2306 .preallocation = prealloc,
2307 .has_nocow = true,
2308 .nocow = nocow,
2309 },
2310 };
2311 return raw_co_create(&options, errp);
2312}
2313
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01002314/*
2315 * Find allocation range in @bs around offset @start.
2316 * May change underlying file descriptor's file offset.
2317 * If @start is not in a hole, store @start in @data, and the
2318 * beginning of the next hole in @hole, and return 0.
2319 * If @start is in a non-trailing hole, store @start in @hole and the
2320 * beginning of the next non-hole in @data, and return 0.
2321 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2322 * If we can't find out, return a negative errno other than -ENXIO.
2323 */
2324static int find_allocation(BlockDriverState *bs, off_t start,
2325 off_t *data, off_t *hole)
Max Reitz4f11aa82014-05-08 20:57:55 +02002326{
2327#if defined SEEK_HOLE && defined SEEK_DATA
2328 BDRVRawState *s = bs->opaque;
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01002329 off_t offs;
Max Reitz4f11aa82014-05-08 20:57:55 +02002330
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01002331 /*
2332 * SEEK_DATA cases:
2333 * D1. offs == start: start is in data
2334 * D2. offs > start: start is in a hole, next data at offs
2335 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2336 * or start is beyond EOF
2337 * If the latter happens, the file has been truncated behind
2338 * our back since we opened it. All bets are off then.
2339 * Treating like a trailing hole is simplest.
2340 * D4. offs < 0, errno != ENXIO: we learned nothing
2341 */
2342 offs = lseek(s->fd, start, SEEK_DATA);
2343 if (offs < 0) {
2344 return -errno; /* D3 or D4 */
2345 }
Jeff Codya03083a2018-04-03 00:37:26 -04002346
2347 if (offs < start) {
2348 /* This is not a valid return by lseek(). We are safe to just return
2349 * -EIO in this case, and we'll treat it like D4. */
2350 return -EIO;
2351 }
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01002352
2353 if (offs > start) {
2354 /* D2: in hole, next data at offs */
2355 *hole = start;
2356 *data = offs;
2357 return 0;
Max Reitz4f11aa82014-05-08 20:57:55 +02002358 }
2359
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01002360 /* D1: in data, end not yet known */
2361
2362 /*
2363 * SEEK_HOLE cases:
2364 * H1. offs == start: start is in a hole
2365 * If this happens here, a hole has been dug behind our back
2366 * since the previous lseek().
2367 * H2. offs > start: either start is in data, next hole at offs,
2368 * or start is in trailing hole, EOF at offs
2369 * Linux treats trailing holes like any other hole: offs ==
2370 * start. Solaris seeks to EOF instead: offs > start (blech).
2371 * If that happens here, a hole has been dug behind our back
2372 * since the previous lseek().
2373 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2374 * If this happens, the file has been truncated behind our
2375 * back since we opened it. Treat it like a trailing hole.
2376 * H4. offs < 0, errno != ENXIO: we learned nothing
2377 * Pretend we know nothing at all, i.e. "forget" about D1.
2378 */
2379 offs = lseek(s->fd, start, SEEK_HOLE);
2380 if (offs < 0) {
2381 return -errno; /* D1 and (H3 or H4) */
2382 }
Jeff Codya03083a2018-04-03 00:37:26 -04002383
2384 if (offs < start) {
2385 /* This is not a valid return by lseek(). We are safe to just return
2386 * -EIO in this case, and we'll treat it like H4. */
2387 return -EIO;
2388 }
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01002389
2390 if (offs > start) {
2391 /*
2392 * D1 and H2: either in data, next hole at offs, or it was in
2393 * data but is now in a trailing hole. In the latter case,
2394 * all bets are off. Treating it as if it there was data all
2395 * the way to EOF is safe, so simply do that.
2396 */
Max Reitz4f11aa82014-05-08 20:57:55 +02002397 *data = start;
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01002398 *hole = offs;
2399 return 0;
Max Reitz4f11aa82014-05-08 20:57:55 +02002400 }
2401
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01002402 /* D1 and H1 */
2403 return -EBUSY;
Max Reitz4f11aa82014-05-08 20:57:55 +02002404#else
2405 return -ENOTSUP;
2406#endif
2407}
2408
Paolo Bonzini55003162012-05-09 16:49:58 +02002409/*
Eric Blakea290f082018-02-13 14:26:44 -06002410 * Returns the allocation status of the specified offset.
Paolo Bonzini55003162012-05-09 16:49:58 +02002411 *
Eric Blakea290f082018-02-13 14:26:44 -06002412 * The block layer guarantees 'offset' and 'bytes' are within bounds.
Paolo Bonzini55003162012-05-09 16:49:58 +02002413 *
Eric Blakea290f082018-02-13 14:26:44 -06002414 * 'pnum' is set to the number of bytes (including and immediately following
2415 * the specified offset) that are known to be in the same
Paolo Bonzini55003162012-05-09 16:49:58 +02002416 * allocated/unallocated state.
2417 *
Eric Blakea290f082018-02-13 14:26:44 -06002418 * 'bytes' is the max value 'pnum' should be set to.
Paolo Bonzini55003162012-05-09 16:49:58 +02002419 */
Eric Blakea290f082018-02-13 14:26:44 -06002420static int coroutine_fn raw_co_block_status(BlockDriverState *bs,
2421 bool want_zero,
2422 int64_t offset,
2423 int64_t bytes, int64_t *pnum,
2424 int64_t *map,
2425 BlockDriverState **file)
Paolo Bonzini55003162012-05-09 16:49:58 +02002426{
Eric Blakea290f082018-02-13 14:26:44 -06002427 off_t data = 0, hole = 0;
Max Reitzd7f62752014-10-24 12:57:59 +02002428 int ret;
Paolo Bonzini55003162012-05-09 16:49:58 +02002429
2430 ret = fd_open(bs);
2431 if (ret < 0) {
2432 return ret;
2433 }
2434
Eric Blakea290f082018-02-13 14:26:44 -06002435 if (!want_zero) {
2436 *pnum = bytes;
2437 *map = offset;
2438 *file = bs;
2439 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
Max Reitze6d7ec32014-10-24 12:57:58 +02002440 }
Kevin Wolf94282e72012-06-19 22:02:51 +00002441
Eric Blakea290f082018-02-13 14:26:44 -06002442 ret = find_allocation(bs, offset, &data, &hole);
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01002443 if (ret == -ENXIO) {
2444 /* Trailing hole */
Eric Blakea290f082018-02-13 14:26:44 -06002445 *pnum = bytes;
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01002446 ret = BDRV_BLOCK_ZERO;
2447 } else if (ret < 0) {
2448 /* No info available, so pretend there are no holes */
Eric Blakea290f082018-02-13 14:26:44 -06002449 *pnum = bytes;
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01002450 ret = BDRV_BLOCK_DATA;
Eric Blakea290f082018-02-13 14:26:44 -06002451 } else if (data == offset) {
2452 /* On a data extent, compute bytes to the end of the extent,
Kevin Wolff4a769a2015-06-09 10:55:08 +02002453 * possibly including a partial sector at EOF. */
Eric Blakea290f082018-02-13 14:26:44 -06002454 *pnum = MIN(bytes, hole - offset);
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01002455 ret = BDRV_BLOCK_DATA;
Paolo Bonzini55003162012-05-09 16:49:58 +02002456 } else {
Eric Blakea290f082018-02-13 14:26:44 -06002457 /* On a hole, compute bytes to the beginning of the next extent. */
2458 assert(hole == offset);
2459 *pnum = MIN(bytes, data - offset);
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01002460 ret = BDRV_BLOCK_ZERO;
Paolo Bonzini55003162012-05-09 16:49:58 +02002461 }
Eric Blakea290f082018-02-13 14:26:44 -06002462 *map = offset;
Fam Zheng02650ac2016-01-26 11:58:51 +08002463 *file = bs;
Eric Blakea290f082018-02-13 14:26:44 -06002464 return ret | BDRV_BLOCK_OFFSET_VALID;
Paolo Bonzini55003162012-05-09 16:49:58 +02002465}
2466
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +01002467#if defined(__linux__)
2468/* Verify that the file is not in the page cache */
2469static void check_cache_dropped(BlockDriverState *bs, Error **errp)
2470{
2471 const size_t window_size = 128 * 1024 * 1024;
2472 BDRVRawState *s = bs->opaque;
2473 void *window = NULL;
2474 size_t length = 0;
2475 unsigned char *vec;
2476 size_t page_size;
2477 off_t offset;
2478 off_t end;
2479
2480 /* mincore(2) page status information requires 1 byte per page */
2481 page_size = sysconf(_SC_PAGESIZE);
2482 vec = g_malloc(DIV_ROUND_UP(window_size, page_size));
2483
2484 end = raw_getlength(bs);
2485
2486 for (offset = 0; offset < end; offset += window_size) {
2487 void *new_window;
2488 size_t new_length;
2489 size_t vec_end;
2490 size_t i;
2491 int ret;
2492
2493 /* Unmap previous window if size has changed */
2494 new_length = MIN(end - offset, window_size);
2495 if (new_length != length) {
2496 munmap(window, length);
2497 window = NULL;
2498 length = 0;
2499 }
2500
2501 new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE,
2502 s->fd, offset);
2503 if (new_window == MAP_FAILED) {
2504 error_setg_errno(errp, errno, "mmap failed");
2505 break;
2506 }
2507
2508 window = new_window;
2509 length = new_length;
2510
2511 ret = mincore(window, length, vec);
2512 if (ret < 0) {
2513 error_setg_errno(errp, errno, "mincore failed");
2514 break;
2515 }
2516
2517 vec_end = DIV_ROUND_UP(length, page_size);
2518 for (i = 0; i < vec_end; i++) {
2519 if (vec[i] & 0x1) {
2520 error_setg(errp, "page cache still in use!");
2521 break;
2522 }
2523 }
2524 }
2525
2526 if (window) {
2527 munmap(window, length);
2528 }
2529
2530 g_free(vec);
2531}
2532#endif /* __linux__ */
2533
Stefan Hajnoczidd577a22018-04-27 17:23:11 +01002534static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs,
2535 Error **errp)
2536{
2537 BDRVRawState *s = bs->opaque;
2538 int ret;
2539
2540 ret = fd_open(bs);
2541 if (ret < 0) {
2542 error_setg_errno(errp, -ret, "The file descriptor is not open");
2543 return;
2544 }
2545
2546 if (s->open_flags & O_DIRECT) {
2547 return; /* No host kernel page cache */
2548 }
2549
2550#if defined(__linux__)
2551 /* This sets the scene for the next syscall... */
2552 ret = bdrv_co_flush(bs);
2553 if (ret < 0) {
2554 error_setg_errno(errp, -ret, "flush failed");
2555 return;
2556 }
2557
2558 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2559 * process. These limitations are okay because we just fsynced the file,
2560 * we don't use mmap, and the file should not be in use by other processes.
2561 */
2562 ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED);
2563 if (ret != 0) { /* the return value is a positive errno */
2564 error_setg_errno(errp, ret, "fadvise failed");
2565 return;
2566 }
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +01002567
2568 if (s->check_cache_dropped) {
2569 check_cache_dropped(bs, errp);
2570 }
Stefan Hajnoczidd577a22018-04-27 17:23:11 +01002571#else /* __linux__ */
2572 /* Do nothing. Live migration to a remote host with cache.direct=off is
2573 * unsupported on other host operating systems. Cache consistency issues
2574 * may occur but no error is reported here, partly because that's the
2575 * historical behavior and partly because it's hard to differentiate valid
2576 * configurations that should not cause errors.
2577 */
2578#endif /* !__linux__ */
2579}
2580
Kevin Wolf33d70fb2018-06-21 19:07:32 +02002581static coroutine_fn int
Kevin Wolf46ee0f42018-10-25 14:18:58 +01002582raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev)
Christoph Hellwigdce512d2010-12-17 11:41:15 +01002583{
Christoph Hellwigdce512d2010-12-17 11:41:15 +01002584 BDRVRawState *s = bs->opaque;
Kevin Wolf46ee0f42018-10-25 14:18:58 +01002585 RawPosixAIOData acb;
Christoph Hellwigdce512d2010-12-17 11:41:15 +01002586
Kevin Wolf46ee0f42018-10-25 14:18:58 +01002587 acb = (RawPosixAIOData) {
2588 .bs = bs,
2589 .aio_fildes = s->fd,
2590 .aio_type = QEMU_AIO_DISCARD,
2591 .aio_offset = offset,
2592 .aio_nbytes = bytes,
2593 };
2594
2595 if (blkdev) {
2596 acb.aio_type |= QEMU_AIO_BLKDEV;
2597 }
2598
2599 return raw_thread_pool_submit(bs, handle_aiocb_discard, &acb);
2600}
2601
2602static coroutine_fn int
2603raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
2604{
2605 return raw_do_pdiscard(bs, offset, bytes, false);
Christoph Hellwigdce512d2010-12-17 11:41:15 +01002606}
Kevin Wolf0e7e1982009-05-18 16:42:10 +02002607
Kevin Wolf7154d8a2018-10-25 14:18:58 +01002608static int coroutine_fn
2609raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes,
2610 BdrvRequestFlags flags, bool blkdev)
2611{
2612 BDRVRawState *s = bs->opaque;
2613 RawPosixAIOData acb;
2614 ThreadPoolFunc *handler;
2615
2616 acb = (RawPosixAIOData) {
2617 .bs = bs,
2618 .aio_fildes = s->fd,
2619 .aio_type = QEMU_AIO_WRITE_ZEROES,
2620 .aio_offset = offset,
2621 .aio_nbytes = bytes,
2622 };
2623
2624 if (blkdev) {
2625 acb.aio_type |= QEMU_AIO_BLKDEV;
2626 }
2627
2628 if (flags & BDRV_REQ_MAY_UNMAP) {
2629 acb.aio_type |= QEMU_AIO_DISCARD;
2630 handler = handle_aiocb_write_zeroes_unmap;
2631 } else {
2632 handler = handle_aiocb_write_zeroes;
2633 }
2634
2635 return raw_thread_pool_submit(bs, handler, &acb);
2636}
2637
Eric Blake2ffa76c2016-06-01 15:10:10 -06002638static int coroutine_fn raw_co_pwrite_zeroes(
2639 BlockDriverState *bs, int64_t offset,
Manos Pitsidianakisf5a5ca72017-06-09 13:18:08 +03002640 int bytes, BdrvRequestFlags flags)
Paolo Bonzini260a82e2013-11-22 13:39:55 +01002641{
Kevin Wolf7154d8a2018-10-25 14:18:58 +01002642 return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false);
Paolo Bonzini260a82e2013-11-22 13:39:55 +01002643}
2644
2645static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2646{
2647 BDRVRawState *s = bs->opaque;
2648
2649 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
Paolo Bonzini260a82e2013-11-22 13:39:55 +01002650 return 0;
2651}
2652
Chunyan Liu6f482f72014-06-05 17:21:01 +08002653static QemuOptsList raw_create_opts = {
2654 .name = "raw-create-opts",
2655 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
2656 .desc = {
2657 {
2658 .name = BLOCK_OPT_SIZE,
2659 .type = QEMU_OPT_SIZE,
2660 .help = "Virtual disk size"
2661 },
Chunyan Liu4ab15592014-06-30 14:29:58 +08002662 {
2663 .name = BLOCK_OPT_NOCOW,
2664 .type = QEMU_OPT_BOOL,
2665 .help = "Turn off copy-on-write (valid only on btrfs)"
2666 },
Hu Tao06247422014-09-10 17:05:48 +08002667 {
2668 .name = BLOCK_OPT_PREALLOC,
2669 .type = QEMU_OPT_STRING,
2670 .help = "Preallocation mode (allowed values: off, falloc, full)"
2671 },
Chunyan Liu6f482f72014-06-05 17:21:01 +08002672 { /* end of list */ }
2673 }
Kevin Wolf0e7e1982009-05-18 16:42:10 +02002674};
2675
Fam Zheng244a5662017-05-03 00:35:56 +08002676static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared,
2677 Error **errp)
2678{
2679 return raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp);
2680}
2681
2682static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared)
2683{
2684 BDRVRawState *s = bs->opaque;
2685 raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL);
2686 s->perm = perm;
2687 s->shared_perm = shared;
2688}
2689
2690static void raw_abort_perm_update(BlockDriverState *bs)
2691{
2692 raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
2693}
2694
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03002695static int coroutine_fn raw_co_copy_range_from(
2696 BlockDriverState *bs, BdrvChild *src, uint64_t src_offset,
2697 BdrvChild *dst, uint64_t dst_offset, uint64_t bytes,
2698 BdrvRequestFlags read_flags, BdrvRequestFlags write_flags)
Fam Zheng1efad062018-06-01 17:26:43 +08002699{
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03002700 return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes,
2701 read_flags, write_flags);
Fam Zheng1efad062018-06-01 17:26:43 +08002702}
2703
2704static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs,
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03002705 BdrvChild *src,
2706 uint64_t src_offset,
2707 BdrvChild *dst,
2708 uint64_t dst_offset,
2709 uint64_t bytes,
2710 BdrvRequestFlags read_flags,
2711 BdrvRequestFlags write_flags)
Fam Zheng1efad062018-06-01 17:26:43 +08002712{
Kevin Wolf58a209c2018-10-25 14:18:58 +01002713 RawPosixAIOData acb;
Fam Zheng1efad062018-06-01 17:26:43 +08002714 BDRVRawState *s = bs->opaque;
2715 BDRVRawState *src_s;
2716
2717 assert(dst->bs == bs);
2718 if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) {
2719 return -ENOTSUP;
2720 }
2721
2722 src_s = src->bs->opaque;
Fam Zheng9f850f62018-07-02 10:58:34 +08002723 if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) {
Fam Zheng1efad062018-06-01 17:26:43 +08002724 return -EIO;
2725 }
Kevin Wolf58a209c2018-10-25 14:18:58 +01002726
2727 acb = (RawPosixAIOData) {
2728 .bs = bs,
2729 .aio_type = QEMU_AIO_COPY_RANGE,
2730 .aio_fildes = src_s->fd,
2731 .aio_offset = src_offset,
2732 .aio_nbytes = bytes,
2733 .copy_range = {
2734 .aio_fd2 = s->fd,
2735 .aio_offset2 = dst_offset,
2736 },
2737 };
2738
2739 return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb);
Fam Zheng1efad062018-06-01 17:26:43 +08002740}
2741
Max Reitz5f535a92014-12-02 18:32:41 +01002742BlockDriver bdrv_file = {
Christoph Hellwig84a12e62010-04-07 22:30:24 +02002743 .format_name = "file",
2744 .protocol_name = "file",
blueswir1856ae5c2009-04-07 17:57:09 +00002745 .instance_size = sizeof(BDRVRawState),
BenoƮt Canet030be322013-09-24 17:07:04 +02002746 .bdrv_needs_filename = true,
blueswir1856ae5c2009-04-07 17:57:09 +00002747 .bdrv_probe = NULL, /* no probe for protocols */
Max Reitz078896a2014-03-05 22:41:37 +01002748 .bdrv_parse_filename = raw_parse_filename,
Kevin Wolf66f82ce2010-04-14 14:17:38 +02002749 .bdrv_file_open = raw_open,
Jeff Codyeeb6b452012-09-20 15:13:25 -04002750 .bdrv_reopen_prepare = raw_reopen_prepare,
2751 .bdrv_reopen_commit = raw_reopen_commit,
2752 .bdrv_reopen_abort = raw_reopen_abort,
blueswir1856ae5c2009-04-07 17:57:09 +00002753 .bdrv_close = raw_close,
Kevin Wolf927f11e2018-01-16 16:04:21 +01002754 .bdrv_co_create = raw_co_create,
Stefan Hajnocziefc75e22018-01-18 13:43:45 +01002755 .bdrv_co_create_opts = raw_co_create_opts,
Peter Lieven3ac21622013-06-28 12:47:42 +02002756 .bdrv_has_zero_init = bdrv_has_zero_init_1,
Eric Blakea290f082018-02-13 14:26:44 -06002757 .bdrv_co_block_status = raw_co_block_status,
Stefan Hajnoczidd577a22018-04-27 17:23:11 +01002758 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
Eric Blake2ffa76c2016-06-01 15:10:10 -06002759 .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes,
ths3b46e622007-09-17 08:09:54 +00002760
Kevin Wolf9d52aa32016-06-03 17:36:27 +02002761 .bdrv_co_preadv = raw_co_preadv,
2762 .bdrv_co_pwritev = raw_co_pwritev,
Kevin Wolf33d70fb2018-06-21 19:07:32 +02002763 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
2764 .bdrv_co_pdiscard = raw_co_pdiscard,
Fam Zheng1efad062018-06-01 17:26:43 +08002765 .bdrv_co_copy_range_from = raw_co_copy_range_from,
2766 .bdrv_co_copy_range_to = raw_co_copy_range_to,
Paolo Bonzinic25f53b2011-11-29 12:42:20 +01002767 .bdrv_refresh_limits = raw_refresh_limits,
Ming Lei1b3abdc2014-07-04 18:04:34 +08002768 .bdrv_io_plug = raw_aio_plug,
2769 .bdrv_io_unplug = raw_aio_unplug,
Nishanth Aravamudaned6e2162018-06-22 12:37:00 -07002770 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
aliguori3c529d92008-12-12 16:41:40 +00002771
Kevin Wolf061ca8a2018-06-21 17:54:35 +02002772 .bdrv_co_truncate = raw_co_truncate,
bellard83f64092006-08-01 16:21:11 +00002773 .bdrv_getlength = raw_getlength,
Paolo Bonzini260a82e2013-11-22 13:39:55 +01002774 .bdrv_get_info = raw_get_info,
Fam Zheng4a1d5e12011-07-12 19:56:39 +08002775 .bdrv_get_allocated_file_size
2776 = raw_get_allocated_file_size,
Fam Zheng244a5662017-05-03 00:35:56 +08002777 .bdrv_check_perm = raw_check_perm,
2778 .bdrv_set_perm = raw_set_perm,
2779 .bdrv_abort_perm_update = raw_abort_perm_update,
Chunyan Liu6f482f72014-06-05 17:21:01 +08002780 .create_opts = &raw_create_opts,
bellard83f64092006-08-01 16:21:11 +00002781};
2782
bellard19cb3732006-08-19 11:45:59 +00002783/***********************************************/
2784/* host device */
2785
Pavel Borzenkov83affaa2011-11-10 22:40:06 +04002786#if defined(__APPLE__) && defined(__MACH__)
Programmingkid98caa5b2015-11-20 19:17:48 -05002787static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
2788 CFIndex maxPathSize, int flags);
Programmingkidd0855f12016-03-21 11:41:28 -04002789static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
bellard19cb3732006-08-19 11:45:59 +00002790{
Programmingkidd0855f12016-03-21 11:41:28 -04002791 kern_return_t kernResult = KERN_FAILURE;
bellard19cb3732006-08-19 11:45:59 +00002792 mach_port_t masterPort;
2793 CFMutableDictionaryRef classesToMatch;
Programmingkidd0855f12016-03-21 11:41:28 -04002794 const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
2795 char *mediaType = NULL;
bellard19cb3732006-08-19 11:45:59 +00002796
2797 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
2798 if ( KERN_SUCCESS != kernResult ) {
2799 printf( "IOMasterPort returned %d\n", kernResult );
2800 }
ths3b46e622007-09-17 08:09:54 +00002801
Programmingkidd0855f12016-03-21 11:41:28 -04002802 int index;
2803 for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
2804 classesToMatch = IOServiceMatching(matching_array[index]);
2805 if (classesToMatch == NULL) {
2806 error_report("IOServiceMatching returned NULL for %s",
2807 matching_array[index]);
2808 continue;
2809 }
2810 CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
2811 kCFBooleanTrue);
2812 kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch,
2813 mediaIterator);
2814 if (kernResult != KERN_SUCCESS) {
2815 error_report("Note: IOServiceGetMatchingServices returned %d",
2816 kernResult);
2817 continue;
2818 }
ths3b46e622007-09-17 08:09:54 +00002819
Programmingkidd0855f12016-03-21 11:41:28 -04002820 /* If a match was found, leave the loop */
2821 if (*mediaIterator != 0) {
2822 DPRINTF("Matching using %s\n", matching_array[index]);
2823 mediaType = g_strdup(matching_array[index]);
2824 break;
2825 }
2826 }
2827 return mediaType;
bellard19cb3732006-08-19 11:45:59 +00002828}
2829
Programmingkid98caa5b2015-11-20 19:17:48 -05002830kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
2831 CFIndex maxPathSize, int flags)
bellard19cb3732006-08-19 11:45:59 +00002832{
2833 io_object_t nextMedia;
2834 kern_return_t kernResult = KERN_FAILURE;
2835 *bsdPath = '\0';
2836 nextMedia = IOIteratorNext( mediaIterator );
2837 if ( nextMedia )
2838 {
2839 CFTypeRef bsdPathAsCFString;
2840 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
2841 if ( bsdPathAsCFString ) {
2842 size_t devPathLength;
2843 strcpy( bsdPath, _PATH_DEV );
Programmingkid98caa5b2015-11-20 19:17:48 -05002844 if (flags & BDRV_O_NOCACHE) {
2845 strcat(bsdPath, "r");
2846 }
bellard19cb3732006-08-19 11:45:59 +00002847 devPathLength = strlen( bsdPath );
2848 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
2849 kernResult = KERN_SUCCESS;
2850 }
2851 CFRelease( bsdPathAsCFString );
2852 }
2853 IOObjectRelease( nextMedia );
2854 }
ths3b46e622007-09-17 08:09:54 +00002855
bellard19cb3732006-08-19 11:45:59 +00002856 return kernResult;
2857}
2858
Programmingkidd0855f12016-03-21 11:41:28 -04002859/* Sets up a real cdrom for use in QEMU */
2860static bool setup_cdrom(char *bsd_path, Error **errp)
2861{
2862 int index, num_of_test_partitions = 2, fd;
2863 char test_partition[MAXPATHLEN];
2864 bool partition_found = false;
2865
2866 /* look for a working partition */
2867 for (index = 0; index < num_of_test_partitions; index++) {
2868 snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
2869 index);
2870 fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE);
2871 if (fd >= 0) {
2872 partition_found = true;
2873 qemu_close(fd);
2874 break;
2875 }
2876 }
2877
2878 /* if a working partition on the device was not found */
2879 if (partition_found == false) {
2880 error_setg(errp, "Failed to find a working partition on disc");
2881 } else {
2882 DPRINTF("Using %s as optical disc\n", test_partition);
2883 pstrcpy(bsd_path, MAXPATHLEN, test_partition);
2884 }
2885 return partition_found;
2886}
2887
2888/* Prints directions on mounting and unmounting a device */
2889static void print_unmounting_directions(const char *file_name)
2890{
2891 error_report("If device %s is mounted on the desktop, unmount"
2892 " it first before using it in QEMU", file_name);
2893 error_report("Command to unmount device: diskutil unmountDisk %s",
2894 file_name);
2895 error_report("Command to mount device: diskutil mountDisk %s", file_name);
2896}
2897
2898#endif /* defined(__APPLE__) && defined(__MACH__) */
bellard19cb3732006-08-19 11:45:59 +00002899
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02002900static int hdev_probe_device(const char *filename)
2901{
2902 struct stat st;
2903
2904 /* allow a dedicated CD-ROM driver to match with a higher priority */
2905 if (strstart(filename, "/dev/cdrom", NULL))
2906 return 50;
2907
2908 if (stat(filename, &st) >= 0 &&
2909 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
2910 return 100;
2911 }
2912
2913 return 0;
2914}
2915
Stefan Hajnoczida888d32013-02-05 12:28:33 +01002916static int check_hdev_writable(BDRVRawState *s)
2917{
2918#if defined(BLKROGET)
2919 /* Linux block devices can be configured "read-only" using blockdev(8).
2920 * This is independent of device node permissions and therefore open(2)
2921 * with O_RDWR succeeds. Actual writes fail with EPERM.
2922 *
2923 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
2924 * check for read-only block devices so that Linux block devices behave
2925 * properly.
2926 */
2927 struct stat st;
2928 int readonly = 0;
2929
2930 if (fstat(s->fd, &st)) {
2931 return -errno;
2932 }
2933
2934 if (!S_ISBLK(st.st_mode)) {
2935 return 0;
2936 }
2937
2938 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
2939 return -errno;
2940 }
2941
2942 if (readonly) {
2943 return -EACCES;
2944 }
2945#endif /* defined(BLKROGET) */
2946 return 0;
2947}
2948
Max Reitz7af803d2014-03-08 00:39:41 +01002949static void hdev_parse_filename(const char *filename, QDict *options,
2950 Error **errp)
2951{
Max Reitz03c320d2017-05-22 21:52:16 +02002952 bdrv_parse_filename_strip_prefix(filename, "host_device:", options);
Max Reitz7af803d2014-03-08 00:39:41 +01002953}
2954
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03002955static bool hdev_is_sg(BlockDriverState *bs)
2956{
2957
2958#if defined(__linux__)
2959
Kevin Wolf0d4377b2016-10-20 14:50:12 +02002960 BDRVRawState *s = bs->opaque;
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03002961 struct stat st;
2962 struct sg_scsi_id scsiid;
2963 int sg_version;
Kevin Wolf0d4377b2016-10-20 14:50:12 +02002964 int ret;
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03002965
Kevin Wolf0d4377b2016-10-20 14:50:12 +02002966 if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) {
2967 return false;
2968 }
2969
2970 ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version);
2971 if (ret < 0) {
2972 return false;
2973 }
2974
2975 ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid);
2976 if (ret >= 0) {
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03002977 DPRINTF("SG device found: type=%d, version=%d\n",
2978 scsiid.scsi_type, sg_version);
2979 return true;
2980 }
2981
2982#endif
2983
2984 return false;
2985}
2986
Max Reitz015a1032013-09-05 14:22:29 +02002987static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
2988 Error **errp)
bellard19cb3732006-08-19 11:45:59 +00002989{
2990 BDRVRawState *s = bs->opaque;
Max Reitze428e432013-10-11 11:37:01 +02002991 Error *local_err = NULL;
Stefan Hajnoczida888d32013-02-05 12:28:33 +01002992 int ret;
aliguoria76bab42008-09-22 19:17:18 +00002993
Pavel Borzenkov83affaa2011-11-10 22:40:06 +04002994#if defined(__APPLE__) && defined(__MACH__)
Markus Armbruster129c7d12017-03-30 19:43:12 +02002995 /*
2996 * Caution: while qdict_get_str() is fine, getting non-string types
2997 * would require more care. When @options come from -blockdev or
2998 * blockdev_add, its members are typed according to the QAPI
2999 * schema, but when they come from -drive, they're all QString.
3000 */
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03003001 const char *filename = qdict_get_str(options, "filename");
Programmingkidd0855f12016-03-21 11:41:28 -04003002 char bsd_path[MAXPATHLEN] = "";
3003 bool error_occurred = false;
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03003004
Programmingkidd0855f12016-03-21 11:41:28 -04003005 /* If using a real cdrom */
3006 if (strcmp(filename, "/dev/cdrom") == 0) {
3007 char *mediaType = NULL;
3008 kern_return_t ret_val;
3009 io_iterator_t mediaIterator = 0;
ths5fafdf22007-09-16 21:08:06 +00003010
Programmingkidd0855f12016-03-21 11:41:28 -04003011 mediaType = FindEjectableOpticalMedia(&mediaIterator);
3012 if (mediaType == NULL) {
3013 error_setg(errp, "Please make sure your CD/DVD is in the optical"
3014 " drive");
3015 error_occurred = true;
3016 goto hdev_open_Mac_error;
bellard19cb3732006-08-19 11:45:59 +00003017 }
ths3b46e622007-09-17 08:09:54 +00003018
Programmingkidd0855f12016-03-21 11:41:28 -04003019 ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
3020 if (ret_val != KERN_SUCCESS) {
3021 error_setg(errp, "Could not get BSD path for optical drive");
3022 error_occurred = true;
3023 goto hdev_open_Mac_error;
3024 }
3025
3026 /* If a real optical drive was not found */
3027 if (bsd_path[0] == '\0') {
3028 error_setg(errp, "Failed to obtain bsd path for optical drive");
3029 error_occurred = true;
3030 goto hdev_open_Mac_error;
3031 }
3032
3033 /* If using a cdrom disc and finding a partition on the disc failed */
3034 if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
3035 setup_cdrom(bsd_path, errp) == false) {
3036 print_unmounting_directions(bsd_path);
3037 error_occurred = true;
3038 goto hdev_open_Mac_error;
3039 }
3040
Eric Blake46f5ac22017-04-27 16:58:17 -05003041 qdict_put_str(options, "filename", bsd_path);
Programmingkidd0855f12016-03-21 11:41:28 -04003042
3043hdev_open_Mac_error:
3044 g_free(mediaType);
3045 if (mediaIterator) {
3046 IOObjectRelease(mediaIterator);
3047 }
3048 if (error_occurred) {
3049 return -ENOENT;
3050 }
bellard19cb3732006-08-19 11:45:59 +00003051 }
Programmingkidd0855f12016-03-21 11:41:28 -04003052#endif /* defined(__APPLE__) && defined(__MACH__) */
bellard19cb3732006-08-19 11:45:59 +00003053
3054 s->type = FTYPE_FILE;
Christoph Hellwig90babde2009-06-15 13:53:38 +02003055
John Snow230ff732018-07-10 13:00:40 -04003056 ret = raw_open_common(bs, options, flags, 0, true, &local_err);
Stefan Hajnoczida888d32013-02-05 12:28:33 +01003057 if (ret < 0) {
Eduardo Habkost621ff942016-06-13 18:57:56 -03003058 error_propagate(errp, local_err);
Programmingkidd0855f12016-03-21 11:41:28 -04003059#if defined(__APPLE__) && defined(__MACH__)
3060 if (*bsd_path) {
3061 filename = bsd_path;
3062 }
3063 /* if a physical device experienced an error while being opened */
3064 if (strncmp(filename, "/dev/", 5) == 0) {
3065 print_unmounting_directions(filename);
3066 }
3067#endif /* defined(__APPLE__) && defined(__MACH__) */
Stefan Hajnoczida888d32013-02-05 12:28:33 +01003068 return ret;
3069 }
3070
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03003071 /* Since this does ioctl the device must be already opened */
3072 bs->sg = hdev_is_sg(bs);
3073
Stefan Hajnoczida888d32013-02-05 12:28:33 +01003074 if (flags & BDRV_O_RDWR) {
3075 ret = check_hdev_writable(s);
3076 if (ret < 0) {
3077 raw_close(bs);
Max Reitze428e432013-10-11 11:37:01 +02003078 error_setg_errno(errp, -ret, "The device is not writable");
Stefan Hajnoczida888d32013-02-05 12:28:33 +01003079 return ret;
3080 }
3081 }
3082
3083 return ret;
bellard19cb3732006-08-19 11:45:59 +00003084}
3085
aliguori03ff3ca2008-09-15 15:51:35 +00003086#if defined(__linux__)
Kevin Wolf2f3a7ab2018-10-31 11:25:18 +01003087static int coroutine_fn
3088hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
aliguori221f7152009-03-28 17:28:41 +00003089{
aliguorif141eaf2009-04-07 18:43:24 +00003090 BDRVRawState *s = bs->opaque;
Kevin Wolf03425672018-10-31 11:30:42 +01003091 RawPosixAIOData acb;
Kevin Wolf2f3a7ab2018-10-31 11:25:18 +01003092 int ret;
aliguori221f7152009-03-28 17:28:41 +00003093
Kevin Wolf2f3a7ab2018-10-31 11:25:18 +01003094 ret = fd_open(bs);
3095 if (ret < 0) {
3096 return ret;
3097 }
Paolo Bonzinic208e8c2012-11-02 16:14:20 +01003098
Paolo Bonzini7c9e5272017-08-21 18:58:56 +02003099 if (req == SG_IO && s->pr_mgr) {
3100 struct sg_io_hdr *io_hdr = buf;
3101 if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT ||
3102 io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) {
3103 return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs),
Kevin Wolf2f3a7ab2018-10-31 11:25:18 +01003104 s->fd, io_hdr);
Paolo Bonzini7c9e5272017-08-21 18:58:56 +02003105 }
3106 }
3107
Kevin Wolf03425672018-10-31 11:30:42 +01003108 acb = (RawPosixAIOData) {
3109 .bs = bs,
3110 .aio_type = QEMU_AIO_IOCTL,
3111 .aio_fildes = s->fd,
3112 .aio_offset = 0,
3113 .ioctl = {
3114 .buf = buf,
3115 .cmd = req,
3116 },
3117 };
3118
3119 return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb);
aliguori221f7152009-03-28 17:28:41 +00003120}
Max Reitzf7096232015-10-19 17:53:07 +02003121#endif /* linux */
aliguori221f7152009-03-28 17:28:41 +00003122
blueswir19f230112009-03-28 08:37:13 +00003123static int fd_open(BlockDriverState *bs)
3124{
3125 BDRVRawState *s = bs->opaque;
3126
3127 /* this is just to ensure s->fd is sane (its called by io ops) */
3128 if (s->fd >= 0)
3129 return 0;
3130 return -EIO;
3131}
bellard19cb3732006-08-19 11:45:59 +00003132
Kevin Wolf33d70fb2018-06-21 19:07:32 +02003133static coroutine_fn int
3134hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
Andreas FƤrberc36dd8a2013-01-18 16:43:35 +01003135{
Kevin Wolf33d70fb2018-06-21 19:07:32 +02003136 int ret;
Andreas FƤrberc36dd8a2013-01-18 16:43:35 +01003137
Kevin Wolf33d70fb2018-06-21 19:07:32 +02003138 ret = fd_open(bs);
3139 if (ret < 0) {
3140 return ret;
Andreas FƤrberc36dd8a2013-01-18 16:43:35 +01003141 }
Kevin Wolf46ee0f42018-10-25 14:18:58 +01003142 return raw_do_pdiscard(bs, offset, bytes, true);
Andreas FƤrberc36dd8a2013-01-18 16:43:35 +01003143}
3144
Eric Blake2ffa76c2016-06-01 15:10:10 -06003145static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs,
Manos Pitsidianakisf5a5ca72017-06-09 13:18:08 +03003146 int64_t offset, int bytes, BdrvRequestFlags flags)
Paolo Bonzinid0b45032013-11-22 13:39:56 +01003147{
Paolo Bonzinid0b45032013-11-22 13:39:56 +01003148 int rc;
3149
3150 rc = fd_open(bs);
3151 if (rc < 0) {
3152 return rc;
3153 }
Kevin Wolf34fa1102018-07-26 11:28:30 +02003154
Kevin Wolf7154d8a2018-10-25 14:18:58 +01003155 return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true);
Paolo Bonzinid0b45032013-11-22 13:39:56 +01003156}
3157
Stefan Hajnocziefc75e22018-01-18 13:43:45 +01003158static int coroutine_fn hdev_co_create_opts(const char *filename, QemuOpts *opts,
3159 Error **errp)
aliguori93c65b42009-04-05 17:40:43 +00003160{
3161 int fd;
3162 int ret = 0;
3163 struct stat stat_buf;
Kevin Wolf0e7e1982009-05-18 16:42:10 +02003164 int64_t total_size = 0;
Max Reitzcc28c6a2014-03-08 00:39:44 +01003165 bool has_prefix;
3166
Max Reitzf7096232015-10-19 17:53:07 +02003167 /* This function is used by both protocol block drivers and therefore either
3168 * of these prefixes may be given.
Max Reitzcc28c6a2014-03-08 00:39:44 +01003169 * The return value has to be stored somewhere, otherwise this is an error
3170 * due to -Werror=unused-value. */
3171 has_prefix =
3172 strstart(filename, "host_device:", &filename) ||
Max Reitzf7096232015-10-19 17:53:07 +02003173 strstart(filename, "host_cdrom:" , &filename);
Max Reitzcc28c6a2014-03-08 00:39:44 +01003174
3175 (void)has_prefix;
aliguori93c65b42009-04-05 17:40:43 +00003176
Fam Zhengdb0754d2018-11-01 14:29:09 +08003177 ret = raw_normalize_devicepath(&filename, errp);
Max Reitzbdd03cd2015-08-12 17:33:31 +02003178 if (ret < 0) {
Max Reitzbdd03cd2015-08-12 17:33:31 +02003179 return ret;
3180 }
3181
Kevin Wolf0e7e1982009-05-18 16:42:10 +02003182 /* Read out options */
Hu Tao180e9522014-09-10 17:05:46 +08003183 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
3184 BDRV_SECTOR_SIZE);
aliguori93c65b42009-04-05 17:40:43 +00003185
Corey Bryant6165f4d2012-08-14 16:43:45 -04003186 fd = qemu_open(filename, O_WRONLY | O_BINARY);
Max Reitze428e432013-10-11 11:37:01 +02003187 if (fd < 0) {
Kevin Wolf57e69b72010-03-12 13:52:31 +01003188 ret = -errno;
Max Reitze428e432013-10-11 11:37:01 +02003189 error_setg_errno(errp, -ret, "Could not open device");
3190 return ret;
3191 }
3192
3193 if (fstat(fd, &stat_buf) < 0) {
3194 ret = -errno;
3195 error_setg_errno(errp, -ret, "Could not stat device");
3196 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
3197 error_setg(errp,
3198 "The given file is neither a block nor a character device");
Kevin Wolf57e69b72010-03-12 13:52:31 +01003199 ret = -ENODEV;
Hu Tao180e9522014-09-10 17:05:46 +08003200 } else if (lseek(fd, 0, SEEK_END) < total_size) {
Max Reitze428e432013-10-11 11:37:01 +02003201 error_setg(errp, "Device is too small");
aliguori93c65b42009-04-05 17:40:43 +00003202 ret = -ENOSPC;
Max Reitze428e432013-10-11 11:37:01 +02003203 }
aliguori93c65b42009-04-05 17:40:43 +00003204
Fam Zheng97ec9112017-09-08 17:44:57 +08003205 if (!ret && total_size) {
3206 uint8_t buf[BDRV_SECTOR_SIZE] = { 0 };
3207 int64_t zero_size = MIN(BDRV_SECTOR_SIZE, total_size);
3208 if (lseek(fd, 0, SEEK_SET) == -1) {
3209 ret = -errno;
3210 } else {
3211 ret = qemu_write_full(fd, buf, zero_size);
3212 ret = ret == zero_size ? 0 : -errno;
3213 }
3214 }
Corey Bryant2e1e79d2012-08-14 16:43:46 -04003215 qemu_close(fd);
aliguori93c65b42009-04-05 17:40:43 +00003216 return ret;
3217}
3218
Anthony Liguori5efa9d52009-05-09 17:03:42 -05003219static BlockDriver bdrv_host_device = {
Kevin Wolf0b4ce022009-10-01 12:35:49 +02003220 .format_name = "host_device",
Christoph Hellwig84a12e62010-04-07 22:30:24 +02003221 .protocol_name = "host_device",
Kevin Wolf0b4ce022009-10-01 12:35:49 +02003222 .instance_size = sizeof(BDRVRawState),
BenoƮt Canet030be322013-09-24 17:07:04 +02003223 .bdrv_needs_filename = true,
Kevin Wolf0b4ce022009-10-01 12:35:49 +02003224 .bdrv_probe_device = hdev_probe_device,
Max Reitz7af803d2014-03-08 00:39:41 +01003225 .bdrv_parse_filename = hdev_parse_filename,
Kevin Wolf66f82ce2010-04-14 14:17:38 +02003226 .bdrv_file_open = hdev_open,
Kevin Wolf0b4ce022009-10-01 12:35:49 +02003227 .bdrv_close = raw_close,
Jeff Cody1bc6b702012-11-20 10:21:10 -05003228 .bdrv_reopen_prepare = raw_reopen_prepare,
3229 .bdrv_reopen_commit = raw_reopen_commit,
3230 .bdrv_reopen_abort = raw_reopen_abort,
Stefan Hajnocziefc75e22018-01-18 13:43:45 +01003231 .bdrv_co_create_opts = hdev_co_create_opts,
Chunyan Liu6f482f72014-06-05 17:21:01 +08003232 .create_opts = &raw_create_opts,
Stefan Hajnoczidd577a22018-04-27 17:23:11 +01003233 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
Eric Blake2ffa76c2016-06-01 15:10:10 -06003234 .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes,
ths3b46e622007-09-17 08:09:54 +00003235
Kevin Wolf9d52aa32016-06-03 17:36:27 +02003236 .bdrv_co_preadv = raw_co_preadv,
3237 .bdrv_co_pwritev = raw_co_pwritev,
Kevin Wolf33d70fb2018-06-21 19:07:32 +02003238 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3239 .bdrv_co_pdiscard = hdev_co_pdiscard,
Fam Zheng1efad062018-06-01 17:26:43 +08003240 .bdrv_co_copy_range_from = raw_co_copy_range_from,
3241 .bdrv_co_copy_range_to = raw_co_copy_range_to,
Paolo Bonzinic25f53b2011-11-29 12:42:20 +01003242 .bdrv_refresh_limits = raw_refresh_limits,
Ming Lei1b3abdc2014-07-04 18:04:34 +08003243 .bdrv_io_plug = raw_aio_plug,
3244 .bdrv_io_unplug = raw_aio_unplug,
Nishanth Aravamudan042b7572018-07-18 14:12:56 -07003245 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
aliguori3c529d92008-12-12 16:41:40 +00003246
Kevin Wolf061ca8a2018-06-21 17:54:35 +02003247 .bdrv_co_truncate = raw_co_truncate,
aurel32e60f4692009-03-07 22:00:29 +00003248 .bdrv_getlength = raw_getlength,
Paolo Bonzini260a82e2013-11-22 13:39:55 +01003249 .bdrv_get_info = raw_get_info,
Fam Zheng4a1d5e12011-07-12 19:56:39 +08003250 .bdrv_get_allocated_file_size
3251 = raw_get_allocated_file_size,
Fam Zheng244a5662017-05-03 00:35:56 +08003252 .bdrv_check_perm = raw_check_perm,
3253 .bdrv_set_perm = raw_set_perm,
3254 .bdrv_abort_perm_update = raw_abort_perm_update,
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +01003255 .bdrv_probe_blocksizes = hdev_probe_blocksizes,
3256 .bdrv_probe_geometry = hdev_probe_geometry,
bellard19cb3732006-08-19 11:45:59 +00003257
ths985a03b2007-12-24 16:10:43 +00003258 /* generic scsi device */
Christoph Hellwig63ec93d2009-06-15 14:04:34 +02003259#ifdef __linux__
Kevin Wolf2f3a7ab2018-10-31 11:25:18 +01003260 .bdrv_co_ioctl = hdev_co_ioctl,
Christoph Hellwig63ec93d2009-06-15 14:04:34 +02003261#endif
bellard19cb3732006-08-19 11:45:59 +00003262};
Anthony Liguori5efa9d52009-05-09 17:03:42 -05003263
Max Reitz18fa1c42014-03-08 00:39:43 +01003264#if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3265static void cdrom_parse_filename(const char *filename, QDict *options,
3266 Error **errp)
3267{
Max Reitz03c320d2017-05-22 21:52:16 +02003268 bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options);
Max Reitz18fa1c42014-03-08 00:39:43 +01003269}
3270#endif
3271
3272#ifdef __linux__
Max Reitz015a1032013-09-05 14:22:29 +02003273static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
3274 Error **errp)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003275{
3276 BDRVRawState *s = bs->opaque;
3277
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003278 s->type = FTYPE_CD;
3279
Blue Swirl19a3da72009-06-17 18:27:44 +03003280 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
John Snow230ff732018-07-10 13:00:40 -04003281 return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp);
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003282}
3283
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02003284static int cdrom_probe_device(const char *filename)
3285{
Cole Robinson3baf7202010-01-14 11:19:40 -05003286 int fd, ret;
3287 int prio = 0;
Johannes Stezenbach343f8562011-06-29 16:25:17 +02003288 struct stat st;
Cole Robinson3baf7202010-01-14 11:19:40 -05003289
Corey Bryant6165f4d2012-08-14 16:43:45 -04003290 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
Cole Robinson3baf7202010-01-14 11:19:40 -05003291 if (fd < 0) {
3292 goto out;
3293 }
Johannes Stezenbach343f8562011-06-29 16:25:17 +02003294 ret = fstat(fd, &st);
3295 if (ret == -1 || !S_ISBLK(st.st_mode)) {
3296 goto outc;
3297 }
Cole Robinson3baf7202010-01-14 11:19:40 -05003298
3299 /* Attempt to detect via a CDROM specific ioctl */
3300 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
3301 if (ret >= 0)
3302 prio = 100;
3303
Johannes Stezenbach343f8562011-06-29 16:25:17 +02003304outc:
Corey Bryant2e1e79d2012-08-14 16:43:46 -04003305 qemu_close(fd);
Cole Robinson3baf7202010-01-14 11:19:40 -05003306out:
3307 return prio;
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02003308}
3309
Max Reitze031f752015-10-19 17:53:11 +02003310static bool cdrom_is_inserted(BlockDriverState *bs)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003311{
3312 BDRVRawState *s = bs->opaque;
3313 int ret;
3314
3315 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
Max Reitze031f752015-10-19 17:53:11 +02003316 return ret == CDS_DISC_OK;
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003317}
3318
Luiz Capitulinof36f3942012-02-03 16:24:53 -02003319static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003320{
3321 BDRVRawState *s = bs->opaque;
3322
3323 if (eject_flag) {
3324 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
3325 perror("CDROMEJECT");
3326 } else {
3327 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
3328 perror("CDROMEJECT");
3329 }
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003330}
3331
Markus Armbruster025e8492011-09-06 18:58:47 +02003332static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003333{
3334 BDRVRawState *s = bs->opaque;
3335
3336 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
3337 /*
3338 * Note: an error can happen if the distribution automatically
3339 * mounts the CD-ROM
3340 */
3341 /* perror("CDROM_LOCKDOOR"); */
3342 }
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003343}
3344
3345static BlockDriver bdrv_host_cdrom = {
3346 .format_name = "host_cdrom",
Christoph Hellwig84a12e62010-04-07 22:30:24 +02003347 .protocol_name = "host_cdrom",
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003348 .instance_size = sizeof(BDRVRawState),
BenoƮt Canet030be322013-09-24 17:07:04 +02003349 .bdrv_needs_filename = true,
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02003350 .bdrv_probe_device = cdrom_probe_device,
Max Reitz18fa1c42014-03-08 00:39:43 +01003351 .bdrv_parse_filename = cdrom_parse_filename,
Kevin Wolf66f82ce2010-04-14 14:17:38 +02003352 .bdrv_file_open = cdrom_open,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003353 .bdrv_close = raw_close,
Jeff Cody1bc6b702012-11-20 10:21:10 -05003354 .bdrv_reopen_prepare = raw_reopen_prepare,
3355 .bdrv_reopen_commit = raw_reopen_commit,
3356 .bdrv_reopen_abort = raw_reopen_abort,
Stefan Hajnocziefc75e22018-01-18 13:43:45 +01003357 .bdrv_co_create_opts = hdev_co_create_opts,
Chunyan Liu6f482f72014-06-05 17:21:01 +08003358 .create_opts = &raw_create_opts,
Stefan Hajnoczidd577a22018-04-27 17:23:11 +01003359 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003360
Kevin Wolf9d52aa32016-06-03 17:36:27 +02003361
3362 .bdrv_co_preadv = raw_co_preadv,
3363 .bdrv_co_pwritev = raw_co_pwritev,
Kevin Wolf33d70fb2018-06-21 19:07:32 +02003364 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
Paolo Bonzinic25f53b2011-11-29 12:42:20 +01003365 .bdrv_refresh_limits = raw_refresh_limits,
Ming Lei1b3abdc2014-07-04 18:04:34 +08003366 .bdrv_io_plug = raw_aio_plug,
3367 .bdrv_io_unplug = raw_aio_unplug,
Nishanth Aravamudan042b7572018-07-18 14:12:56 -07003368 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003369
Kevin Wolf061ca8a2018-06-21 17:54:35 +02003370 .bdrv_co_truncate = raw_co_truncate,
Kevin Wolfb94a2612013-10-29 12:18:58 +01003371 .bdrv_getlength = raw_getlength,
3372 .has_variable_length = true,
Fam Zheng4a1d5e12011-07-12 19:56:39 +08003373 .bdrv_get_allocated_file_size
3374 = raw_get_allocated_file_size,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003375
3376 /* removable device support */
3377 .bdrv_is_inserted = cdrom_is_inserted,
3378 .bdrv_eject = cdrom_eject,
Markus Armbruster025e8492011-09-06 18:58:47 +02003379 .bdrv_lock_medium = cdrom_lock_medium,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003380
3381 /* generic scsi device */
Kevin Wolf2f3a7ab2018-10-31 11:25:18 +01003382 .bdrv_co_ioctl = hdev_co_ioctl,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003383};
3384#endif /* __linux__ */
3385
Aurelien Jarnoa167ba52009-11-29 18:00:41 +01003386#if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
Andreas Tobler511018e2013-10-31 22:41:46 +01003387static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
3388 Error **errp)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003389{
3390 BDRVRawState *s = bs->opaque;
Max Reitze428e432013-10-11 11:37:01 +02003391 Error *local_err = NULL;
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003392 int ret;
3393
3394 s->type = FTYPE_CD;
3395
John Snow230ff732018-07-10 13:00:40 -04003396 ret = raw_open_common(bs, options, flags, 0, true, &local_err);
Max Reitze428e432013-10-11 11:37:01 +02003397 if (ret) {
Eduardo Habkost621ff942016-06-13 18:57:56 -03003398 error_propagate(errp, local_err);
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003399 return ret;
Max Reitze428e432013-10-11 11:37:01 +02003400 }
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003401
Dong Xu Wang9b2260c2011-11-22 18:06:25 +08003402 /* make sure the door isn't locked at this time */
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003403 ioctl(s->fd, CDIOCALLOW);
3404 return 0;
3405}
3406
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02003407static int cdrom_probe_device(const char *filename)
3408{
3409 if (strstart(filename, "/dev/cd", NULL) ||
3410 strstart(filename, "/dev/acd", NULL))
3411 return 100;
3412 return 0;
3413}
3414
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003415static int cdrom_reopen(BlockDriverState *bs)
3416{
3417 BDRVRawState *s = bs->opaque;
3418 int fd;
3419
3420 /*
3421 * Force reread of possibly changed/newly loaded disc,
3422 * FreeBSD seems to not notice sometimes...
3423 */
3424 if (s->fd >= 0)
Corey Bryant2e1e79d2012-08-14 16:43:46 -04003425 qemu_close(s->fd);
Corey Bryant6165f4d2012-08-14 16:43:45 -04003426 fd = qemu_open(bs->filename, s->open_flags, 0644);
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003427 if (fd < 0) {
3428 s->fd = -1;
3429 return -EIO;
3430 }
3431 s->fd = fd;
3432
Dong Xu Wang9b2260c2011-11-22 18:06:25 +08003433 /* make sure the door isn't locked at this time */
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003434 ioctl(s->fd, CDIOCALLOW);
3435 return 0;
3436}
3437
Max Reitze031f752015-10-19 17:53:11 +02003438static bool cdrom_is_inserted(BlockDriverState *bs)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003439{
3440 return raw_getlength(bs) > 0;
3441}
3442
Luiz Capitulinof36f3942012-02-03 16:24:53 -02003443static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003444{
3445 BDRVRawState *s = bs->opaque;
3446
3447 if (s->fd < 0)
Markus Armbruster822e1cd2011-07-20 18:23:42 +02003448 return;
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003449
3450 (void) ioctl(s->fd, CDIOCALLOW);
3451
3452 if (eject_flag) {
3453 if (ioctl(s->fd, CDIOCEJECT) < 0)
3454 perror("CDIOCEJECT");
3455 } else {
3456 if (ioctl(s->fd, CDIOCCLOSE) < 0)
3457 perror("CDIOCCLOSE");
3458 }
3459
Markus Armbruster822e1cd2011-07-20 18:23:42 +02003460 cdrom_reopen(bs);
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003461}
3462
Markus Armbruster025e8492011-09-06 18:58:47 +02003463static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003464{
3465 BDRVRawState *s = bs->opaque;
3466
3467 if (s->fd < 0)
Markus Armbruster7bf37fe2011-07-20 18:23:41 +02003468 return;
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003469 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
3470 /*
3471 * Note: an error can happen if the distribution automatically
3472 * mounts the CD-ROM
3473 */
3474 /* perror("CDROM_LOCKDOOR"); */
3475 }
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003476}
3477
3478static BlockDriver bdrv_host_cdrom = {
3479 .format_name = "host_cdrom",
Christoph Hellwig84a12e62010-04-07 22:30:24 +02003480 .protocol_name = "host_cdrom",
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003481 .instance_size = sizeof(BDRVRawState),
BenoƮt Canet030be322013-09-24 17:07:04 +02003482 .bdrv_needs_filename = true,
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02003483 .bdrv_probe_device = cdrom_probe_device,
Max Reitz18fa1c42014-03-08 00:39:43 +01003484 .bdrv_parse_filename = cdrom_parse_filename,
Kevin Wolf66f82ce2010-04-14 14:17:38 +02003485 .bdrv_file_open = cdrom_open,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003486 .bdrv_close = raw_close,
Jeff Cody1bc6b702012-11-20 10:21:10 -05003487 .bdrv_reopen_prepare = raw_reopen_prepare,
3488 .bdrv_reopen_commit = raw_reopen_commit,
3489 .bdrv_reopen_abort = raw_reopen_abort,
Stefan Hajnocziefc75e22018-01-18 13:43:45 +01003490 .bdrv_co_create_opts = hdev_co_create_opts,
Chunyan Liu6f482f72014-06-05 17:21:01 +08003491 .create_opts = &raw_create_opts,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003492
Kevin Wolf9d52aa32016-06-03 17:36:27 +02003493 .bdrv_co_preadv = raw_co_preadv,
3494 .bdrv_co_pwritev = raw_co_pwritev,
Kevin Wolf33d70fb2018-06-21 19:07:32 +02003495 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
Paolo Bonzinic25f53b2011-11-29 12:42:20 +01003496 .bdrv_refresh_limits = raw_refresh_limits,
Ming Lei1b3abdc2014-07-04 18:04:34 +08003497 .bdrv_io_plug = raw_aio_plug,
3498 .bdrv_io_unplug = raw_aio_unplug,
Nishanth Aravamudan042b7572018-07-18 14:12:56 -07003499 .bdrv_attach_aio_context = raw_aio_attach_aio_context,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003500
Kevin Wolf061ca8a2018-06-21 17:54:35 +02003501 .bdrv_co_truncate = raw_co_truncate,
Kevin Wolfb94a2612013-10-29 12:18:58 +01003502 .bdrv_getlength = raw_getlength,
3503 .has_variable_length = true,
Fam Zheng4a1d5e12011-07-12 19:56:39 +08003504 .bdrv_get_allocated_file_size
3505 = raw_get_allocated_file_size,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003506
3507 /* removable device support */
3508 .bdrv_is_inserted = cdrom_is_inserted,
3509 .bdrv_eject = cdrom_eject,
Markus Armbruster025e8492011-09-06 18:58:47 +02003510 .bdrv_lock_medium = cdrom_lock_medium,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003511};
3512#endif /* __FreeBSD__ */
3513
Christoph Hellwig84a12e62010-04-07 22:30:24 +02003514static void bdrv_file_init(void)
Anthony Liguori5efa9d52009-05-09 17:03:42 -05003515{
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02003516 /*
3517 * Register all the drivers. Note that order is important, the driver
3518 * registered last will get probed first.
3519 */
Christoph Hellwig84a12e62010-04-07 22:30:24 +02003520 bdrv_register(&bdrv_file);
Anthony Liguori5efa9d52009-05-09 17:03:42 -05003521 bdrv_register(&bdrv_host_device);
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003522#ifdef __linux__
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003523 bdrv_register(&bdrv_host_cdrom);
3524#endif
Aurelien Jarnoa167ba52009-11-29 18:00:41 +01003525#if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02003526 bdrv_register(&bdrv_host_cdrom);
3527#endif
Anthony Liguori5efa9d52009-05-09 17:03:42 -05003528}
3529
Christoph Hellwig84a12e62010-04-07 22:30:24 +02003530block_init(bdrv_file_init);