blob: 90fa54352c249a5fd51595e5b1f594f164fe4d33 [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"
Markus Armbrustere2c1c342022-12-21 14:35:49 +010029#include "block/block-io.h"
Paolo Bonzini737e1502012-12-17 18:19:44 +010030#include "block/block_int.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010031#include "qemu/module.h"
Markus Armbruster922a01a2018-02-01 12:18:46 +010032#include "qemu/option.h"
Kevin Wolfffa244c2020-07-07 16:23:29 +020033#include "qemu/units.h"
Peter Maydell5df022c2022-02-26 18:07:23 +000034#include "qemu/memalign.h"
Paolo Bonzinide81a162012-05-25 11:46:27 +020035#include "trace.h"
Paolo Bonzini737e1502012-12-17 18:19:44 +010036#include "block/thread-pool.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010037#include "qemu/iov.h"
Paolo Bonzini0187f5c2016-07-04 18:33:20 +020038#include "block/raw-aio.h"
Markus Armbruster452fcdb2018-02-01 12:18:39 +010039#include "qapi/qmp/qdict.h"
Markus Armbrusterd49b6832015-03-17 18:29:20 +010040#include "qapi/qmp/qstring.h"
bellard83f64092006-08-01 16:21:11 +000041
Paolo Bonzini7c9e5272017-08-21 18:58:56 +020042#include "scsi/pr-manager.h"
43#include "scsi/constants.h"
44
Pavel Borzenkov83affaa2011-11-10 22:40:06 +040045#if defined(__APPLE__) && (__MACH__)
Joelle van Dyne14176c82021-03-15 11:03:38 -070046#include <sys/ioctl.h>
47#if defined(HAVE_HOST_BLOCK_DEVICE)
bellard83f64092006-08-01 16:21:11 +000048#include <paths.h>
49#include <sys/param.h>
Akihiko Odaki0dfc7af2021-07-05 22:04:56 +090050#include <sys/mount.h>
bellard83f64092006-08-01 16:21:11 +000051#include <IOKit/IOKitLib.h>
52#include <IOKit/IOBSD.h>
53#include <IOKit/storage/IOMediaBSDClient.h>
54#include <IOKit/storage/IOMedia.h>
55#include <IOKit/storage/IOCDMedia.h>
56//#include <IOKit/storage/IOCDTypes.h>
Programmingkidd0855f12016-03-21 11:41:28 -040057#include <IOKit/storage/IODVDMedia.h>
bellard83f64092006-08-01 16:21:11 +000058#include <CoreFoundation/CoreFoundation.h>
Joelle van Dyne14176c82021-03-15 11:03:38 -070059#endif /* defined(HAVE_HOST_BLOCK_DEVICE) */
bellard83f64092006-08-01 16:21:11 +000060#endif
61
62#ifdef __sun__
ths2e9671d2007-01-05 17:44:41 +000063#define _POSIX_PTHREAD_SEMANTICS 1
bellard83f64092006-08-01 16:21:11 +000064#include <sys/dkio.h>
65#endif
bellard19cb3732006-08-19 11:45:59 +000066#ifdef __linux__
67#include <sys/ioctl.h>
Bernhard Kohl05acda42010-09-06 17:06:02 +020068#include <sys/param.h>
Fam Zheng1efad062018-06-01 17:26:43 +080069#include <sys/syscall.h>
Kevin Wolf5edc8552020-07-16 16:26:01 +020070#include <sys/vfs.h>
Sam Li6d43eaa2023-05-08 12:55:28 +080071#if defined(CONFIG_BLKZONED)
72#include <linux/blkzoned.h>
73#endif
bellard19cb3732006-08-19 11:45:59 +000074#include <linux/cdrom.h>
75#include <linux/fd.h>
Paolo Bonzini55003162012-05-09 16:49:58 +020076#include <linux/fs.h>
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +010077#include <linux/hdreg.h>
Kevin Wolf5edc8552020-07-16 16:26:01 +020078#include <linux/magic.h>
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +030079#include <scsi/sg.h>
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +010080#ifdef __s390__
81#include <asm/dasd.h>
82#endif
Chunyan Liu4ab15592014-06-30 14:29:58 +080083#ifndef FS_NOCOW_FL
84#define FS_NOCOW_FL 0x00800000 /* Do not cow file */
85#endif
Paolo Bonzini55003162012-05-09 16:49:58 +020086#endif
Denis V. Lunevb953f072015-01-30 11:42:14 +030087#if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
Kusanagi Kouichi3d4fa432013-01-14 16:26:52 +010088#include <linux/falloc.h>
89#endif
Aurelien Jarnoa167ba52009-11-29 18:00:41 +010090#if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
ths1cb6c3f2006-12-21 19:14:11 +000091#include <sys/disk.h>
blueswir19f230112009-03-28 08:37:13 +000092#include <sys/cdio.h>
ths1cb6c3f2006-12-21 19:14:11 +000093#endif
bellard19cb3732006-08-19 11:45:59 +000094
blueswir1128ab2f2008-08-15 18:33:42 +000095#ifdef __OpenBSD__
96#include <sys/ioctl.h>
97#include <sys/disklabel.h>
98#include <sys/dkio.h>
99#endif
100
Christoph Eggerd1f6fd82011-05-23 14:31:17 +0200101#ifdef __NetBSD__
102#include <sys/ioctl.h>
103#include <sys/disklabel.h>
104#include <sys/dkio.h>
105#include <sys/disk.h>
106#endif
107
blueswir1c5e97232009-03-07 20:06:23 +0000108#ifdef __DragonFly__
109#include <sys/ioctl.h>
110#include <sys/diskslice.h>
111#endif
112
aliguorif6465572008-10-14 18:00:38 +0000113/* OS X does not have O_DSYNC */
114#ifndef O_DSYNC
G 31c27a8b2009-07-01 13:28:32 -0400115#ifdef O_SYNC
aliguori7ab064d2008-10-14 18:14:47 +0000116#define O_DSYNC O_SYNC
G 31c27a8b2009-07-01 13:28:32 -0400117#elif defined(O_FSYNC)
118#define O_DSYNC O_FSYNC
119#endif
aliguorif6465572008-10-14 18:00:38 +0000120#endif
121
aliguori9f7965c2008-10-14 14:42:54 +0000122/* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
123#ifndef O_DIRECT
124#define O_DIRECT O_DSYNC
125#endif
126
bellard19cb3732006-08-19 11:45:59 +0000127#define FTYPE_FILE 0
128#define FTYPE_CD 1
bellard83f64092006-08-01 16:21:11 +0000129
Christoph Hellwig581b9e22010-09-12 23:43:21 +0200130#define MAX_BLOCKSIZE 4096
131
Fam Zheng244a5662017-05-03 00:35:56 +0800132/* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
133 * leaving a few more bytes for its future use. */
134#define RAW_LOCK_PERM_BASE 100
135#define RAW_LOCK_SHARED_BASE 200
136
bellard83f64092006-08-01 16:21:11 +0000137typedef struct BDRVRawState {
138 int fd;
Fam Zheng244a5662017-05-03 00:35:56 +0800139 bool use_lock;
bellard19cb3732006-08-19 11:45:59 +0000140 int type;
Christoph Hellwig0e1d8f42009-06-15 13:53:26 +0200141 int open_flags;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100142 size_t buf_align;
143
Fam Zheng244a5662017-05-03 00:35:56 +0800144 /* The current permissions. */
145 uint64_t perm;
146 uint64_t shared_perm;
147
Fam Zheng2996ffa2018-10-11 15:21:33 +0800148 /* The perms bits whose corresponding bytes are already locked in
Fam Zhengf2e3af22018-10-11 15:21:34 +0800149 * s->fd. */
Fam Zheng2996ffa2018-10-11 15:21:33 +0800150 uint64_t locked_perm;
151 uint64_t locked_shared_perm;
152
Stefano Garzarella684960d2021-10-26 18:23:44 +0200153 uint64_t aio_max_batch;
154
Kevin Wolf6ceabe62019-03-08 15:40:40 +0100155 int perm_change_fd;
Max Reitz094e3632019-05-22 19:03:45 +0200156 int perm_change_flags;
Kevin Wolfe0c9cf32019-03-07 19:07:35 +0100157 BDRVReopenState *reopen_state;
158
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100159 bool has_discard:1;
Paolo Bonzini97a2ae32013-11-22 13:39:57 +0100160 bool has_write_zeroes:1;
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200161 bool use_linux_aio:1;
Prasad Pandit24687ab2024-04-25 12:34:12 +0530162 bool has_laio_fdsync:1;
Aarushi Mehtac6447512020-01-20 14:18:51 +0000163 bool use_linux_io_uring:1;
Daniel P. Berrangéc7ddc882021-04-15 14:28:16 +0100164 int page_cache_inconsistent; /* errno from fdatasync failure */
Denis V. Lunevd50d8222015-01-30 11:42:15 +0300165 bool has_fallocate;
Roger Pau Monne3cad8302014-10-21 16:03:03 +0200166 bool needs_alignment;
Kevin Wolf5dbd0ce2021-11-16 11:14:31 +0100167 bool force_alignment;
Stefan Hajnoczif357fcd2019-03-07 16:49:41 +0000168 bool drop_cache;
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +0100169 bool check_cache_dropped;
Anton Nefedov1c450362019-09-23 15:17:36 +0300170 struct {
171 uint64_t discard_nb_ok;
172 uint64_t discard_nb_failed;
173 uint64_t discard_bytes_ok;
174 } stats;
Paolo Bonzini7c9e5272017-08-21 18:58:56 +0200175
176 PRManager *pr_mgr;
bellard83f64092006-08-01 16:21:11 +0000177} BDRVRawState;
178
Jeff Codyeeb6b452012-09-20 15:13:25 -0400179typedef struct BDRVRawReopenState {
Jeff Codyeeb6b452012-09-20 15:13:25 -0400180 int open_flags;
Stefan Hajnoczif357fcd2019-03-07 16:49:41 +0000181 bool drop_cache;
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +0100182 bool check_cache_dropped;
Jeff Codyeeb6b452012-09-20 15:13:25 -0400183} BDRVRawReopenState;
184
Joelle van Dyne14176c82021-03-15 11:03:38 -0700185static int fd_open(BlockDriverState *bs)
186{
187 BDRVRawState *s = bs->opaque;
188
189 /* this is just to ensure s->fd is sane (its called by io ops) */
190 if (s->fd >= 0) {
191 return 0;
192 }
193 return -EIO;
194}
195
Paolo Bonzini36c6c872023-06-01 13:51:34 +0200196static int64_t raw_getlength(BlockDriverState *bs);
bellard83f64092006-08-01 16:21:11 +0000197
Paolo Bonzinide81a162012-05-25 11:46:27 +0200198typedef struct RawPosixAIOData {
199 BlockDriverState *bs;
Paolo Bonzinide81a162012-05-25 11:46:27 +0200200 int aio_type;
Kevin Wolfd57c44d2018-10-25 16:21:14 +0100201 int aio_fildes;
202
203 off_t aio_offset;
204 uint64_t aio_nbytes;
205
Kevin Wolf93f4e2f2018-06-21 18:23:16 +0200206 union {
207 struct {
Kevin Wolfd57c44d2018-10-25 16:21:14 +0100208 struct iovec *iov;
209 int niov;
210 } io;
211 struct {
212 uint64_t cmd;
213 void *buf;
214 } ioctl;
215 struct {
Kevin Wolf93f4e2f2018-06-21 18:23:16 +0200216 int aio_fd2;
217 off_t aio_offset2;
Kevin Wolfd57c44d2018-10-25 16:21:14 +0100218 } copy_range;
Kevin Wolf93f4e2f2018-06-21 18:23:16 +0200219 struct {
220 PreallocMode prealloc;
221 Error **errp;
Kevin Wolfd57c44d2018-10-25 16:21:14 +0100222 } truncate;
Sam Li6d43eaa2023-05-08 12:55:28 +0800223 struct {
224 unsigned int *nr_zones;
225 BlockZoneDescriptor *zones;
226 } zone_report;
227 struct {
228 unsigned long op;
229 } zone_mgmt;
Kevin Wolf93f4e2f2018-06-21 18:23:16 +0200230 };
Paolo Bonzinide81a162012-05-25 11:46:27 +0200231} RawPosixAIOData;
232
Aurelien Jarnoa167ba52009-11-29 18:00:41 +0100233#if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +0200234static int cdrom_reopen(BlockDriverState *bs);
blueswir19f230112009-03-28 08:37:13 +0000235#endif
236
David Edmondson797e3e32021-01-13 16:44:47 +0000237/*
238 * Elide EAGAIN and EACCES details when failing to lock, as this
239 * indicates that the specified file region is already locked by
240 * another process, which is considered a common scenario.
241 */
242#define raw_lock_error_setg_errno(errp, err, fmt, ...) \
243 do { \
244 if ((err) == EAGAIN || (err) == EACCES) { \
245 error_setg((errp), (fmt), ## __VA_ARGS__); \
246 } else { \
247 error_setg_errno((errp), (err), (fmt), ## __VA_ARGS__); \
248 } \
249 } while (0)
250
Christoph Egger1de1ae02011-05-24 11:30:29 +0200251#if defined(__NetBSD__)
Fam Zhengdb0754d2018-11-01 14:29:09 +0800252static int raw_normalize_devicepath(const char **filename, Error **errp)
Christoph Egger1de1ae02011-05-24 11:30:29 +0200253{
254 static char namebuf[PATH_MAX];
255 const char *dp, *fname;
256 struct stat sb;
257
258 fname = *filename;
259 dp = strrchr(fname, '/');
260 if (lstat(fname, &sb) < 0) {
Paolo Bonzinif6fc1e32019-07-25 11:59:20 +0200261 error_setg_file_open(errp, errno, fname);
Christoph Egger1de1ae02011-05-24 11:30:29 +0200262 return -errno;
263 }
264
265 if (!S_ISBLK(sb.st_mode)) {
266 return 0;
267 }
268
269 if (dp == NULL) {
270 snprintf(namebuf, PATH_MAX, "r%s", fname);
271 } else {
272 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
273 (int)(dp - fname), fname, dp + 1);
274 }
Christoph Egger1de1ae02011-05-24 11:30:29 +0200275 *filename = namebuf;
Fam Zhengdb0754d2018-11-01 14:29:09 +0800276 warn_report("%s is a block device, using %s", fname, *filename);
Christoph Egger1de1ae02011-05-24 11:30:29 +0200277
278 return 0;
279}
280#else
Fam Zhengdb0754d2018-11-01 14:29:09 +0800281static int raw_normalize_devicepath(const char **filename, Error **errp)
Christoph Egger1de1ae02011-05-24 11:30:29 +0200282{
283 return 0;
284}
285#endif
286
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100287/*
288 * Get logical block size via ioctl. On success store it in @sector_size_p.
289 */
290static int probe_logical_blocksize(int fd, unsigned int *sector_size_p)
291{
292 unsigned int sector_size;
293 bool success = false;
Peter Maydell700f9ce2017-03-23 14:36:28 +0000294 int i;
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100295
296 errno = ENOTSUP;
Peter Maydell700f9ce2017-03-23 14:36:28 +0000297 static const unsigned long ioctl_list[] = {
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100298#ifdef BLKSSZGET
Peter Maydell700f9ce2017-03-23 14:36:28 +0000299 BLKSSZGET,
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100300#endif
301#ifdef DKIOCGETBLOCKSIZE
Peter Maydell700f9ce2017-03-23 14:36:28 +0000302 DKIOCGETBLOCKSIZE,
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100303#endif
304#ifdef DIOCGSECTORSIZE
Peter Maydell700f9ce2017-03-23 14:36:28 +0000305 DIOCGSECTORSIZE,
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100306#endif
Peter Maydell700f9ce2017-03-23 14:36:28 +0000307 };
308
309 /* Try a few ioctls to get the right size */
310 for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) {
311 if (ioctl(fd, ioctl_list[i], &sector_size) >= 0) {
312 *sector_size_p = sector_size;
313 success = true;
314 }
315 }
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100316
317 return success ? 0 : -errno;
318}
319
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +0100320/**
321 * Get physical block size of @fd.
322 * On success, store it in @blk_size and return 0.
323 * On failure, return -errno.
324 */
325static int probe_physical_blocksize(int fd, unsigned int *blk_size)
326{
327#ifdef BLKPBSZGET
328 if (ioctl(fd, BLKPBSZGET, blk_size) < 0) {
329 return -errno;
330 }
331 return 0;
332#else
333 return -ENOTSUP;
334#endif
335}
336
Kevin Wolf5edc8552020-07-16 16:26:01 +0200337/*
338 * Returns true if no alignment restrictions are necessary even for files
339 * opened with O_DIRECT.
340 *
341 * raw_probe_alignment() probes the required alignment and assume that 1 means
342 * the probing failed, so it falls back to a safe default of 4k. This can be
343 * avoided if we know that byte alignment is okay for the file.
344 */
345static bool dio_byte_aligned(int fd)
346{
347#ifdef __linux__
348 struct statfs buf;
349 int ret;
350
351 ret = fstatfs(fd, &buf);
352 if (ret == 0 && buf.f_type == NFS_SUPER_MAGIC) {
353 return true;
354 }
355#endif
356 return false;
357}
358
Kevin Wolf5dbd0ce2021-11-16 11:14:31 +0100359static bool raw_needs_alignment(BlockDriverState *bs)
360{
361 BDRVRawState *s = bs->opaque;
362
363 if ((bs->open_flags & BDRV_O_NOCACHE) != 0 && !dio_byte_aligned(s->fd)) {
364 return true;
365 }
366
367 return s->force_alignment;
368}
369
Stefan Hajnoczi22d182e2015-03-05 15:38:17 -0600370/* Check if read is allowed with given memory buffer and length.
371 *
372 * This function is used to check O_DIRECT memory buffer and request alignment.
373 */
374static bool raw_is_io_aligned(int fd, void *buf, size_t len)
375{
376 ssize_t ret = pread(fd, buf, len, 0);
377
378 if (ret >= 0) {
379 return true;
380 }
381
382#ifdef __linux__
383 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
384 * other errors (e.g. real I/O error), which could happen on a failed
385 * drive, since we only care about probing alignment.
386 */
387 if (errno != EINVAL) {
388 return true;
389 }
390#endif
391
392 return false;
393}
394
Kevin Wolfdf26a352014-07-16 17:48:17 +0200395static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100396{
397 BDRVRawState *s = bs->opaque;
398 char *buf;
Marc-André Lureau8e3b0cb2022-03-23 19:57:22 +0400399 size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size());
Nir Soffera6b257a2019-08-13 21:21:03 +0300400 size_t alignments[] = {1, 512, 1024, 2048, 4096};
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100401
Dimitris Aragiorgisb192af82015-06-23 13:44:56 +0300402 /* For SCSI generic devices the alignment is not really used.
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100403 With buffered I/O, we don't have any restrictions. */
Dimitris Aragiorgisb192af82015-06-23 13:44:56 +0300404 if (bdrv_is_sg(bs) || !s->needs_alignment) {
Eric Blakea5b8dd22016-06-23 16:37:24 -0600405 bs->bl.request_alignment = 1;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100406 s->buf_align = 1;
407 return;
408 }
409
Eric Blakea5b8dd22016-06-23 16:37:24 -0600410 bs->bl.request_alignment = 0;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100411 s->buf_align = 0;
Ekaterina Tumanova8a4ed0d2015-02-16 12:47:55 +0100412 /* Let's try to use the logical blocksize for the alignment. */
Eric Blakea5b8dd22016-06-23 16:37:24 -0600413 if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) {
414 bs->bl.request_alignment = 0;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100415 }
Thomas Hutha5730b8b2021-12-15 13:58:24 +0100416
417#ifdef __linux__
418 /*
419 * The XFS ioctl definitions are shipped in extra packages that might
420 * not always be available. Since we just need the XFS_IOC_DIOINFO ioctl
421 * here, we simply use our own definition instead:
422 */
423 struct xfs_dioattr {
424 uint32_t d_mem;
425 uint32_t d_miniosz;
426 uint32_t d_maxiosz;
427 } da;
428 if (ioctl(fd, _IOR('X', 30, struct xfs_dioattr), &da) >= 0) {
429 bs->bl.request_alignment = da.d_miniosz;
430 /* The kernel returns wrong information for d_mem */
431 /* s->buf_align = da.d_mem; */
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100432 }
433#endif
434
Nir Soffera6b257a2019-08-13 21:21:03 +0300435 /*
436 * If we could not get the sizes so far, we can only guess them. First try
437 * to detect request alignment, since it is more likely to succeed. Then
438 * try to detect buf_align, which cannot be detected in some cases (e.g.
439 * Gluster). If buf_align cannot be detected, we fallback to the value of
440 * request_alignment.
441 */
442
443 if (!bs->bl.request_alignment) {
444 int i;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100445 size_t align;
Nir Soffera6b257a2019-08-13 21:21:03 +0300446 buf = qemu_memalign(max_align, max_align);
447 for (i = 0; i < ARRAY_SIZE(alignments); i++) {
448 align = alignments[i];
449 if (raw_is_io_aligned(fd, buf, align)) {
450 /* Fallback to safe value. */
451 bs->bl.request_alignment = (align != 1) ? align : max_align;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100452 break;
453 }
454 }
455 qemu_vfree(buf);
456 }
457
Nir Soffera6b257a2019-08-13 21:21:03 +0300458 if (!s->buf_align) {
459 int i;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100460 size_t align;
Nir Soffera6b257a2019-08-13 21:21:03 +0300461 buf = qemu_memalign(max_align, 2 * max_align);
462 for (i = 0; i < ARRAY_SIZE(alignments); i++) {
463 align = alignments[i];
464 if (raw_is_io_aligned(fd, buf + align, max_align)) {
Stefan Hajnoczi236094c2019-08-27 11:13:28 +0100465 /* Fallback to request_alignment. */
Nir Soffera6b257a2019-08-13 21:21:03 +0300466 s->buf_align = (align != 1) ? align : bs->bl.request_alignment;
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100467 break;
468 }
469 }
470 qemu_vfree(buf);
471 }
Kevin Wolfdf26a352014-07-16 17:48:17 +0200472
Eric Blakea5b8dd22016-06-23 16:37:24 -0600473 if (!s->buf_align || !bs->bl.request_alignment) {
Eric Blake8cc9c6e2016-06-23 16:37:25 -0600474 error_setg(errp, "Could not find working O_DIRECT alignment");
475 error_append_hint(errp, "Try cache.direct=off\n");
Kevin Wolfdf26a352014-07-16 17:48:17 +0200476 }
Paolo Bonzinic25f53b2011-11-29 12:42:20 +0100477}
478
Kevin Wolfbca52832020-07-17 12:54:25 +0200479static int check_hdev_writable(int fd)
Kevin Wolf20eaf1b2020-07-17 12:54:24 +0200480{
481#if defined(BLKROGET)
482 /* Linux block devices can be configured "read-only" using blockdev(8).
483 * This is independent of device node permissions and therefore open(2)
484 * with O_RDWR succeeds. Actual writes fail with EPERM.
485 *
486 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
487 * check for read-only block devices so that Linux block devices behave
488 * properly.
489 */
490 struct stat st;
491 int readonly = 0;
492
Kevin Wolfbca52832020-07-17 12:54:25 +0200493 if (fstat(fd, &st)) {
Kevin Wolf20eaf1b2020-07-17 12:54:24 +0200494 return -errno;
495 }
496
497 if (!S_ISBLK(st.st_mode)) {
498 return 0;
499 }
500
Kevin Wolfbca52832020-07-17 12:54:25 +0200501 if (ioctl(fd, BLKROGET, &readonly) < 0) {
Kevin Wolf20eaf1b2020-07-17 12:54:24 +0200502 return -errno;
503 }
504
505 if (readonly) {
506 return -EACCES;
507 }
508#endif /* defined(BLKROGET) */
509 return 0;
510}
511
Kevin Wolf23dece12019-03-01 22:15:11 +0100512static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers)
Jeff Cody6a8dc042012-09-20 15:13:21 -0400513{
Kevin Wolf23dece12019-03-01 22:15:11 +0100514 bool read_write = false;
Jeff Cody6a8dc042012-09-20 15:13:21 -0400515 assert(open_flags != NULL);
516
517 *open_flags |= O_BINARY;
518 *open_flags &= ~O_ACCMODE;
Kevin Wolf23dece12019-03-01 22:15:11 +0100519
520 if (bdrv_flags & BDRV_O_AUTO_RDONLY) {
521 read_write = has_writers;
522 } else if (bdrv_flags & BDRV_O_RDWR) {
523 read_write = true;
524 }
525
526 if (read_write) {
Jeff Cody6a8dc042012-09-20 15:13:21 -0400527 *open_flags |= O_RDWR;
528 } else {
529 *open_flags |= O_RDONLY;
530 }
531
532 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
533 * and O_DIRECT for no caching. */
534 if ((bdrv_flags & BDRV_O_NOCACHE)) {
535 *open_flags |= O_DIRECT;
536 }
Jeff Cody6a8dc042012-09-20 15:13:21 -0400537}
538
Max Reitz078896a2014-03-05 22:41:37 +0100539static void raw_parse_filename(const char *filename, QDict *options,
540 Error **errp)
541{
Max Reitz03c320d2017-05-22 21:52:16 +0200542 bdrv_parse_filename_strip_prefix(filename, "file:", options);
Max Reitz078896a2014-03-05 22:41:37 +0100543}
544
Kevin Wolfc66a6152013-04-02 10:47:40 +0200545static QemuOptsList raw_runtime_opts = {
546 .name = "raw",
547 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
548 .desc = {
549 {
550 .name = "filename",
551 .type = QEMU_OPT_STRING,
552 .help = "File name of the image",
553 },
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200554 {
555 .name = "aio",
556 .type = QEMU_OPT_STRING,
Aarushi Mehtac6447512020-01-20 14:18:51 +0000557 .help = "host AIO implementation (threads, native, io_uring)",
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200558 },
Fam Zheng16b48d52017-05-03 00:35:50 +0800559 {
Stefano Garzarella684960d2021-10-26 18:23:44 +0200560 .name = "aio-max-batch",
561 .type = QEMU_OPT_NUMBER,
562 .help = "AIO max batch size (0 = auto handled by AIO backend, default: 0)",
563 },
564 {
Fam Zheng16b48d52017-05-03 00:35:50 +0800565 .name = "locking",
566 .type = QEMU_OPT_STRING,
567 .help = "file locking mode (on/off/auto, default: auto)",
568 },
Paolo Bonzini7c9e5272017-08-21 18:58:56 +0200569 {
570 .name = "pr-manager",
571 .type = QEMU_OPT_STRING,
572 .help = "id of persistent reservation manager object (default: none)",
573 },
Stefan Hajnoczif357fcd2019-03-07 16:49:41 +0000574#if defined(__linux__)
575 {
576 .name = "drop-cache",
577 .type = QEMU_OPT_BOOL,
578 .help = "invalidate page cache during live migration (default: on)",
579 },
580#endif
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +0100581 {
582 .name = "x-check-cache-dropped",
583 .type = QEMU_OPT_BOOL,
584 .help = "check that page cache was dropped on live migration (default: off)"
585 },
Kevin Wolfc66a6152013-04-02 10:47:40 +0200586 { /* end of list */ }
587 },
588};
589
Alberto Garcia8a2ce0b2019-03-12 18:48:48 +0200590static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL };
591
Kevin Wolfc66a6152013-04-02 10:47:40 +0200592static int raw_open_common(BlockDriverState *bs, QDict *options,
John Snow230ff732018-07-10 13:00:40 -0400593 int bdrv_flags, int open_flags,
594 bool device, Error **errp)
bellard83f64092006-08-01 16:21:11 +0000595{
596 BDRVRawState *s = bs->opaque;
Kevin Wolfc66a6152013-04-02 10:47:40 +0200597 QemuOpts *opts;
598 Error *local_err = NULL;
Kevin Wolf8bfea152014-04-11 19:16:36 +0200599 const char *filename = NULL;
Paolo Bonzini7c9e5272017-08-21 18:58:56 +0200600 const char *str;
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200601 BlockdevAioOptions aio, aio_default;
Christoph Hellwig0e1d8f42009-06-15 13:53:26 +0200602 int fd, ret;
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100603 struct stat st;
Fam Zheng244a5662017-05-03 00:35:56 +0800604 OnOffAuto locking;
bellard83f64092006-08-01 16:21:11 +0000605
Peter Crosthwaite87ea75d2014-01-01 18:49:17 -0800606 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
Markus Armbrusteraf175e82020-07-07 18:06:03 +0200607 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
Kevin Wolfc66a6152013-04-02 10:47:40 +0200608 ret = -EINVAL;
609 goto fail;
610 }
611
612 filename = qemu_opt_get(opts, "filename");
613
Fam Zhengdb0754d2018-11-01 14:29:09 +0800614 ret = raw_normalize_devicepath(&filename, errp);
Christoph Egger1de1ae02011-05-24 11:30:29 +0200615 if (ret != 0) {
Kevin Wolfc66a6152013-04-02 10:47:40 +0200616 goto fail;
Christoph Egger1de1ae02011-05-24 11:30:29 +0200617 }
618
Aarushi Mehtac6447512020-01-20 14:18:51 +0000619 if (bdrv_flags & BDRV_O_NATIVE_AIO) {
620 aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE;
621#ifdef CONFIG_LINUX_IO_URING
622 } else if (bdrv_flags & BDRV_O_IO_URING) {
623 aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING;
624#endif
625 } else {
626 aio_default = BLOCKDEV_AIO_OPTIONS_THREADS;
627 }
628
Marc-André Lureauf7abe0e2017-08-24 10:46:10 +0200629 aio = qapi_enum_parse(&BlockdevAioOptions_lookup,
630 qemu_opt_get(opts, "aio"),
Markus Armbruster06c60b62017-08-24 10:45:57 +0200631 aio_default, &local_err);
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200632 if (local_err) {
633 error_propagate(errp, local_err);
634 ret = -EINVAL;
635 goto fail;
636 }
Aarushi Mehtac6447512020-01-20 14:18:51 +0000637
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200638 s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE);
Aarushi Mehtac6447512020-01-20 14:18:51 +0000639#ifdef CONFIG_LINUX_IO_URING
640 s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING);
641#endif
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200642
Stefano Garzarella684960d2021-10-26 18:23:44 +0200643 s->aio_max_batch = qemu_opt_get_number(opts, "aio-max-batch", 0);
644
Marc-André Lureauf7abe0e2017-08-24 10:46:10 +0200645 locking = qapi_enum_parse(&OnOffAuto_lookup,
646 qemu_opt_get(opts, "locking"),
Markus Armbruster06c60b62017-08-24 10:45:57 +0200647 ON_OFF_AUTO_AUTO, &local_err);
Fam Zheng244a5662017-05-03 00:35:56 +0800648 if (local_err) {
649 error_propagate(errp, local_err);
650 ret = -EINVAL;
651 goto fail;
652 }
653 switch (locking) {
654 case ON_OFF_AUTO_ON:
655 s->use_lock = true;
Fam Zheng2b218f52017-08-11 19:44:47 +0800656 if (!qemu_has_ofd_lock()) {
Fam Zhengdb0754d2018-11-01 14:29:09 +0800657 warn_report("File lock requested but OFD locking syscall is "
658 "unavailable, falling back to POSIX file locks");
659 error_printf("Due to the implementation, locks can be lost "
660 "unexpectedly.\n");
Fam Zheng2b218f52017-08-11 19:44:47 +0800661 }
Fam Zheng244a5662017-05-03 00:35:56 +0800662 break;
663 case ON_OFF_AUTO_OFF:
664 s->use_lock = false;
665 break;
666 case ON_OFF_AUTO_AUTO:
Fam Zheng2b218f52017-08-11 19:44:47 +0800667 s->use_lock = qemu_has_ofd_lock();
Fam Zheng244a5662017-05-03 00:35:56 +0800668 break;
669 default:
670 abort();
671 }
672
Paolo Bonzini7c9e5272017-08-21 18:58:56 +0200673 str = qemu_opt_get(opts, "pr-manager");
674 if (str) {
675 s->pr_mgr = pr_manager_lookup(str, &local_err);
676 if (local_err) {
677 error_propagate(errp, local_err);
678 ret = -EINVAL;
679 goto fail;
680 }
681 }
682
Stefan Hajnoczif357fcd2019-03-07 16:49:41 +0000683 s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true);
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +0100684 s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped",
685 false);
686
Jeff Cody6a8dc042012-09-20 15:13:21 -0400687 s->open_flags = open_flags;
Kevin Wolf23dece12019-03-01 22:15:11 +0100688 raw_parse_flags(bdrv_flags, &s->open_flags, false);
bellard83f64092006-08-01 16:21:11 +0000689
Christoph Hellwig90babde2009-06-15 13:53:38 +0200690 s->fd = -1;
Daniel P. Berrangéb18a24a2020-07-01 15:22:43 +0100691 fd = qemu_open(filename, s->open_flags, errp);
Kevin Wolf64107dc2018-10-08 17:27:18 +0200692 ret = fd < 0 ? -errno : 0;
693
Kevin Wolf64107dc2018-10-08 17:27:18 +0200694 if (ret < 0) {
Kevin Wolfc66a6152013-04-02 10:47:40 +0200695 if (ret == -EROFS) {
bellard19cb3732006-08-19 11:45:59 +0000696 ret = -EACCES;
Kevin Wolfc66a6152013-04-02 10:47:40 +0200697 }
698 goto fail;
bellard19cb3732006-08-19 11:45:59 +0000699 }
bellard83f64092006-08-01 16:21:11 +0000700 s->fd = fd;
Christoph Hellwig9ef91a62009-08-20 16:58:19 +0200701
Kevin Wolfbca52832020-07-17 12:54:25 +0200702 /* Check s->open_flags rather than bdrv_flags due to auto-read-only */
703 if (s->open_flags & O_RDWR) {
704 ret = check_hdev_writable(s->fd);
705 if (ret < 0) {
706 error_setg_errno(errp, -ret, "The device is not writable");
707 goto fail;
708 }
709 }
710
Fam Zheng244a5662017-05-03 00:35:56 +0800711 s->perm = 0;
712 s->shared_perm = BLK_PERM_ALL;
713
Christoph Hellwig5c6c3a62009-08-20 16:58:35 +0200714#ifdef CONFIG_LINUX_AIO
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200715 /* Currently Linux does AIO only for files opened with O_DIRECT */
Stefan Hajnoczicd0c0db2023-09-14 10:00:58 -0400716 if (s->use_linux_aio && !(s->open_flags & O_DIRECT)) {
717 error_setg(errp, "aio=native was specified, but it requires "
718 "cache.direct=on, which was not specified.");
719 ret = -EINVAL;
720 goto fail;
Kevin Wolf96518252015-03-17 13:45:21 +0100721 }
Prasad Pandit24687ab2024-04-25 12:34:12 +0530722 if (s->use_linux_aio) {
723 s->has_laio_fdsync = laio_has_fdsync(s->fd);
724 }
Stefan Hajnoczi1501ecc2015-07-23 13:48:34 +0100725#else
Kevin Wolf0a4279d2016-09-08 15:09:01 +0200726 if (s->use_linux_aio) {
Kevin Wolfd657c0c2015-12-15 11:35:36 +0100727 error_setg(errp, "aio=native was specified, but is not supported "
728 "in this build.");
729 ret = -EINVAL;
730 goto fail;
Stefan Hajnoczi1501ecc2015-07-23 13:48:34 +0100731 }
732#endif /* !defined(CONFIG_LINUX_AIO) */
Christoph Hellwig9ef91a62009-08-20 16:58:19 +0200733
Stefan Hajnoczicd0c0db2023-09-14 10:00:58 -0400734#ifndef CONFIG_LINUX_IO_URING
Aarushi Mehtac6447512020-01-20 14:18:51 +0000735 if (s->use_linux_io_uring) {
736 error_setg(errp, "aio=io_uring was specified, but is not supported "
737 "in this build.");
738 ret = -EINVAL;
739 goto fail;
740 }
741#endif /* !defined(CONFIG_LINUX_IO_URING) */
742
Paolo Bonzini7ce21012013-11-22 13:39:47 +0100743 s->has_discard = true;
Paolo Bonzini97a2ae32013-11-22 13:39:57 +0100744 s->has_write_zeroes = true;
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100745
746 if (fstat(s->fd, &st) < 0) {
Max Reitz01212d42014-12-02 18:32:53 +0100747 ret = -errno;
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100748 error_setg_errno(errp, errno, "Could not stat file");
749 goto fail;
750 }
John Snow230ff732018-07-10 13:00:40 -0400751
752 if (!device) {
Daniel P. Berrangé8d17adf2021-02-22 11:16:32 +0000753 if (!S_ISREG(st.st_mode)) {
754 error_setg(errp, "'%s' driver requires '%s' to be a regular file",
755 bs->drv->format_name, bs->filename);
John Snow230ff732018-07-10 13:00:40 -0400756 ret = -EINVAL;
757 goto fail;
758 } else {
John Snow230ff732018-07-10 13:00:40 -0400759 s->has_fallocate = true;
760 }
761 } else {
762 if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
Daniel P. Berrangé8d17adf2021-02-22 11:16:32 +0000763 error_setg(errp, "'%s' driver requires '%s' to be either "
764 "a character or block device",
765 bs->drv->format_name, bs->filename);
John Snow230ff732018-07-10 13:00:40 -0400766 ret = -EINVAL;
767 goto fail;
768 }
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100769 }
Sam Li774c7262023-05-08 12:55:30 +0800770#ifdef CONFIG_BLKZONED
771 /*
772 * The kernel page cache does not reliably work for writes to SWR zones
773 * of zoned block device because it can not guarantee the order of writes.
774 */
775 if ((bs->bl.zoned != BLK_Z_NONE) &&
776 (!(s->open_flags & O_DIRECT))) {
777 error_setg(errp, "The driver supports zoned devices, and it requires "
778 "cache.direct=on, which was not specified.");
779 return -EINVAL; /* No host kernel page cache */
780 }
781#endif
John Snow230ff732018-07-10 13:00:40 -0400782
Paolo Bonzinid0b45032013-11-22 13:39:56 +0100783 if (S_ISBLK(st.st_mode)) {
Paolo Bonzinid0b45032013-11-22 13:39:56 +0100784#ifdef __linux__
785 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
786 * not rely on the contents of discarded blocks unless using O_DIRECT.
Paolo Bonzini97a2ae32013-11-22 13:39:57 +0100787 * Same for BLKZEROOUT.
Paolo Bonzinid0b45032013-11-22 13:39:56 +0100788 */
789 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
Paolo Bonzini97a2ae32013-11-22 13:39:57 +0100790 s->has_write_zeroes = false;
Paolo Bonzinid0b45032013-11-22 13:39:56 +0100791 }
792#endif
793 }
Roger Pau Monne3cad8302014-10-21 16:03:03 +0200794#ifdef __FreeBSD__
795 if (S_ISCHR(st.st_mode)) {
796 /*
797 * The file is a char device (disk), which on FreeBSD isn't behind
798 * a pager, so force all requests to be aligned. This is needed
799 * so QEMU makes sure all IO operations on the device are aligned
800 * to sector size, or else FreeBSD will reject them with EINVAL.
801 */
Kevin Wolf5dbd0ce2021-11-16 11:14:31 +0100802 s->force_alignment = true;
Roger Pau Monne3cad8302014-10-21 16:03:03 +0200803 }
804#endif
Kevin Wolf5dbd0ce2021-11-16 11:14:31 +0100805 s->needs_alignment = raw_needs_alignment(bs);
Paolo Bonzini260a82e2013-11-22 13:39:55 +0100806
Kevin Wolf738301e2019-03-22 13:45:23 +0100807 bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK;
Kevin Wolf2f0c6e72020-04-24 14:54:44 +0200808 if (S_ISREG(st.st_mode)) {
809 /* When extending regular files, we get zeros from the OS */
810 bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE;
811 }
Kevin Wolfc66a6152013-04-02 10:47:40 +0200812 ret = 0;
813fail:
Kevin Wolfa8c5cf22020-07-17 12:54:26 +0200814 if (ret < 0 && s->fd != -1) {
815 qemu_close(s->fd);
816 }
Kevin Wolf8bfea152014-04-11 19:16:36 +0200817 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
818 unlink(filename);
819 }
Kevin Wolfc66a6152013-04-02 10:47:40 +0200820 qemu_opts_del(opts);
821 return ret;
bellard83f64092006-08-01 16:21:11 +0000822}
823
Max Reitz015a1032013-09-05 14:22:29 +0200824static int raw_open(BlockDriverState *bs, QDict *options, int flags,
825 Error **errp)
Christoph Hellwig90babde2009-06-15 13:53:38 +0200826{
827 BDRVRawState *s = bs->opaque;
828
829 s->type = FTYPE_FILE;
John Snow230ff732018-07-10 13:00:40 -0400830 return raw_open_common(bs, options, flags, 0, false, errp);
Christoph Hellwig90babde2009-06-15 13:53:38 +0200831}
832
Fam Zheng244a5662017-05-03 00:35:56 +0800833typedef enum {
834 RAW_PL_PREPARE,
835 RAW_PL_COMMIT,
836 RAW_PL_ABORT,
837} RawPermLockOp;
838
839#define PERM_FOREACH(i) \
840 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
841
842/* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
843 * file; if @unlock == true, also unlock the unneeded bytes.
844 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
845 */
Fam Zheng2996ffa2018-10-11 15:21:33 +0800846static int raw_apply_lock_bytes(BDRVRawState *s, int fd,
Fam Zheng244a5662017-05-03 00:35:56 +0800847 uint64_t perm_lock_bits,
848 uint64_t shared_perm_lock_bits,
849 bool unlock, Error **errp)
850{
851 int ret;
852 int i;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800853 uint64_t locked_perm, locked_shared_perm;
854
855 if (s) {
856 locked_perm = s->locked_perm;
857 locked_shared_perm = s->locked_shared_perm;
858 } else {
859 /*
860 * We don't have the previous bits, just lock/unlock for each of the
861 * requested bits.
862 */
863 if (unlock) {
864 locked_perm = BLK_PERM_ALL;
865 locked_shared_perm = BLK_PERM_ALL;
866 } else {
867 locked_perm = 0;
868 locked_shared_perm = 0;
869 }
870 }
Fam Zheng244a5662017-05-03 00:35:56 +0800871
872 PERM_FOREACH(i) {
873 int off = RAW_LOCK_PERM_BASE + i;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800874 uint64_t bit = (1ULL << i);
875 if ((perm_lock_bits & bit) && !(locked_perm & bit)) {
Max Reitzd0a96152018-05-09 23:53:34 +0200876 ret = qemu_lock_fd(fd, off, 1, false);
Fam Zheng244a5662017-05-03 00:35:56 +0800877 if (ret) {
David Edmondson797e3e32021-01-13 16:44:47 +0000878 raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d",
879 off);
Fam Zheng244a5662017-05-03 00:35:56 +0800880 return ret;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800881 } else if (s) {
882 s->locked_perm |= bit;
Fam Zheng244a5662017-05-03 00:35:56 +0800883 }
Fam Zheng2996ffa2018-10-11 15:21:33 +0800884 } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) {
Max Reitzd0a96152018-05-09 23:53:34 +0200885 ret = qemu_unlock_fd(fd, off, 1);
Fam Zheng244a5662017-05-03 00:35:56 +0800886 if (ret) {
David Edmondson797e3e32021-01-13 16:44:47 +0000887 error_setg_errno(errp, -ret, "Failed to unlock byte %d", off);
Fam Zheng244a5662017-05-03 00:35:56 +0800888 return ret;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800889 } else if (s) {
890 s->locked_perm &= ~bit;
Fam Zheng244a5662017-05-03 00:35:56 +0800891 }
892 }
893 }
894 PERM_FOREACH(i) {
895 int off = RAW_LOCK_SHARED_BASE + i;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800896 uint64_t bit = (1ULL << i);
897 if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) {
Max Reitzd0a96152018-05-09 23:53:34 +0200898 ret = qemu_lock_fd(fd, off, 1, false);
Fam Zheng244a5662017-05-03 00:35:56 +0800899 if (ret) {
David Edmondson797e3e32021-01-13 16:44:47 +0000900 raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d",
901 off);
Fam Zheng244a5662017-05-03 00:35:56 +0800902 return ret;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800903 } else if (s) {
904 s->locked_shared_perm |= bit;
Fam Zheng244a5662017-05-03 00:35:56 +0800905 }
Fam Zheng2996ffa2018-10-11 15:21:33 +0800906 } else if (unlock && (locked_shared_perm & bit) &&
907 !(shared_perm_lock_bits & bit)) {
Max Reitzd0a96152018-05-09 23:53:34 +0200908 ret = qemu_unlock_fd(fd, off, 1);
Fam Zheng244a5662017-05-03 00:35:56 +0800909 if (ret) {
David Edmondson797e3e32021-01-13 16:44:47 +0000910 error_setg_errno(errp, -ret, "Failed to unlock byte %d", off);
Fam Zheng244a5662017-05-03 00:35:56 +0800911 return ret;
Fam Zheng2996ffa2018-10-11 15:21:33 +0800912 } else if (s) {
913 s->locked_shared_perm &= ~bit;
Fam Zheng244a5662017-05-03 00:35:56 +0800914 }
915 }
916 }
917 return 0;
918}
919
920/* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
Max Reitzd0a96152018-05-09 23:53:34 +0200921static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm,
Fam Zheng244a5662017-05-03 00:35:56 +0800922 Error **errp)
923{
924 int ret;
925 int i;
926
927 PERM_FOREACH(i) {
928 int off = RAW_LOCK_SHARED_BASE + i;
929 uint64_t p = 1ULL << i;
930 if (perm & p) {
Max Reitzd0a96152018-05-09 23:53:34 +0200931 ret = qemu_lock_fd_test(fd, off, 1, true);
Fam Zheng244a5662017-05-03 00:35:56 +0800932 if (ret) {
933 char *perm_name = bdrv_perm_names(p);
David Edmondson797e3e32021-01-13 16:44:47 +0000934
935 raw_lock_error_setg_errno(errp, -ret,
936 "Failed to get \"%s\" lock",
937 perm_name);
Fam Zheng244a5662017-05-03 00:35:56 +0800938 g_free(perm_name);
Fam Zheng244a5662017-05-03 00:35:56 +0800939 return ret;
940 }
941 }
942 }
943 PERM_FOREACH(i) {
944 int off = RAW_LOCK_PERM_BASE + i;
945 uint64_t p = 1ULL << i;
946 if (!(shared_perm & p)) {
Max Reitzd0a96152018-05-09 23:53:34 +0200947 ret = qemu_lock_fd_test(fd, off, 1, true);
Fam Zheng244a5662017-05-03 00:35:56 +0800948 if (ret) {
949 char *perm_name = bdrv_perm_names(p);
David Edmondson797e3e32021-01-13 16:44:47 +0000950
951 raw_lock_error_setg_errno(errp, -ret,
952 "Failed to get shared \"%s\" lock",
953 perm_name);
Fam Zheng244a5662017-05-03 00:35:56 +0800954 g_free(perm_name);
Fam Zheng244a5662017-05-03 00:35:56 +0800955 return ret;
956 }
957 }
958 }
959 return 0;
960}
961
962static int raw_handle_perm_lock(BlockDriverState *bs,
963 RawPermLockOp op,
964 uint64_t new_perm, uint64_t new_shared,
965 Error **errp)
966{
967 BDRVRawState *s = bs->opaque;
968 int ret = 0;
969 Error *local_err = NULL;
970
971 if (!s->use_lock) {
972 return 0;
973 }
974
975 if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) {
976 return 0;
977 }
978
Fam Zheng244a5662017-05-03 00:35:56 +0800979 switch (op) {
980 case RAW_PL_PREPARE:
Vladimir Sementsov-Ogievskiy696aaae2019-03-29 14:04:54 +0300981 if ((s->perm | new_perm) == s->perm &&
982 (s->shared_perm & new_shared) == s->shared_perm)
983 {
984 /*
985 * We are going to unlock bytes, it should not fail. If it fail due
986 * to some fs-dependent permission-unrelated reasons (which occurs
987 * sometimes on NFS and leads to abort in bdrv_replace_child) we
988 * can't prevent such errors by any check here. And we ignore them
989 * anyway in ABORT and COMMIT.
990 */
991 return 0;
992 }
Fam Zhengf2e3af22018-10-11 15:21:34 +0800993 ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm,
Fam Zheng244a5662017-05-03 00:35:56 +0800994 ~s->shared_perm | ~new_shared,
995 false, errp);
996 if (!ret) {
Fam Zhengf2e3af22018-10-11 15:21:34 +0800997 ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp);
Fam Zheng244a5662017-05-03 00:35:56 +0800998 if (!ret) {
999 return 0;
1000 }
Fam Zhengb8574312018-09-25 13:05:01 +08001001 error_append_hint(errp,
1002 "Is another process using the image [%s]?\n",
1003 bs->filename);
Fam Zheng244a5662017-05-03 00:35:56 +08001004 }
Fam Zheng244a5662017-05-03 00:35:56 +08001005 /* fall through to unlock bytes. */
1006 case RAW_PL_ABORT:
Fam Zhengf2e3af22018-10-11 15:21:34 +08001007 raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm,
Max Reitzd0a96152018-05-09 23:53:34 +02001008 true, &local_err);
Fam Zheng244a5662017-05-03 00:35:56 +08001009 if (local_err) {
1010 /* Theoretically the above call only unlocks bytes and it cannot
1011 * fail. Something weird happened, report it.
1012 */
Fam Zhengdb0754d2018-11-01 14:29:09 +08001013 warn_report_err(local_err);
Fam Zheng244a5662017-05-03 00:35:56 +08001014 }
1015 break;
1016 case RAW_PL_COMMIT:
Fam Zhengf2e3af22018-10-11 15:21:34 +08001017 raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared,
Max Reitzd0a96152018-05-09 23:53:34 +02001018 true, &local_err);
Fam Zheng244a5662017-05-03 00:35:56 +08001019 if (local_err) {
1020 /* Theoretically the above call only unlocks bytes and it cannot
1021 * fail. Something weird happened, report it.
1022 */
Fam Zhengdb0754d2018-11-01 14:29:09 +08001023 warn_report_err(local_err);
Fam Zheng244a5662017-05-03 00:35:56 +08001024 }
1025 break;
1026 }
1027 return ret;
1028}
1029
Marc-André Lureauad24b672022-03-29 15:27:12 +04001030/* Sets a specific flag */
1031static int fcntl_setfl(int fd, int flag)
1032{
1033 int flags;
1034
1035 flags = fcntl(fd, F_GETFL);
1036 if (flags == -1) {
1037 return -errno;
1038 }
1039 if (fcntl(fd, F_SETFL, flags | flag) == -1) {
1040 return -errno;
1041 }
1042 return 0;
1043}
1044
Kevin Wolf5cec2872019-03-02 00:26:30 +01001045static int raw_reconfigure_getfd(BlockDriverState *bs, int flags,
Denis V. Lunev viab67e3532024-04-30 19:02:13 +02001046 int *open_flags, uint64_t perm, Error **errp)
Kevin Wolf5cec2872019-03-02 00:26:30 +01001047{
1048 BDRVRawState *s = bs->opaque;
1049 int fd = -1;
1050 int ret;
Kevin Wolf23dece12019-03-01 22:15:11 +01001051 bool has_writers = perm &
1052 (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE);
Kevin Wolf5cec2872019-03-02 00:26:30 +01001053 int fcntl_flags = O_APPEND | O_NONBLOCK;
1054#ifdef O_NOATIME
1055 fcntl_flags |= O_NOATIME;
1056#endif
1057
1058 *open_flags = 0;
1059 if (s->type == FTYPE_CD) {
1060 *open_flags |= O_NONBLOCK;
1061 }
1062
Kevin Wolf23dece12019-03-01 22:15:11 +01001063 raw_parse_flags(flags, open_flags, has_writers);
Kevin Wolf5cec2872019-03-02 00:26:30 +01001064
1065#ifdef O_ASYNC
1066 /* Not all operating systems have O_ASYNC, and those that don't
1067 * will not let us track the state into rs->open_flags (typically
1068 * you achieve the same effect with an ioctl, for example I_SETSIG
1069 * on Solaris). But we do not use O_ASYNC, so that's fine.
1070 */
1071 assert((s->open_flags & O_ASYNC) == 0);
1072#endif
1073
Denis V. Lunev viab67e3532024-04-30 19:02:13 +02001074 if (*open_flags == s->open_flags) {
Kevin Wolf6ceabe62019-03-08 15:40:40 +01001075 /* We're lucky, the existing fd is fine */
1076 return s->fd;
1077 }
1078
Kevin Wolf5cec2872019-03-02 00:26:30 +01001079 if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
1080 /* dup the original fd */
1081 fd = qemu_dup(s->fd);
1082 if (fd >= 0) {
1083 ret = fcntl_setfl(fd, *open_flags);
1084 if (ret) {
1085 qemu_close(fd);
1086 fd = -1;
1087 }
1088 }
1089 }
1090
Daniel P. Berrangéb18a24a2020-07-01 15:22:43 +01001091 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
Kevin Wolf5cec2872019-03-02 00:26:30 +01001092 if (fd == -1) {
1093 const char *normalized_filename = bs->filename;
1094 ret = raw_normalize_devicepath(&normalized_filename, errp);
1095 if (ret >= 0) {
Daniel P. Berrangéb18a24a2020-07-01 15:22:43 +01001096 fd = qemu_open(normalized_filename, *open_flags, errp);
Kevin Wolf5cec2872019-03-02 00:26:30 +01001097 if (fd == -1) {
Kevin Wolf5cec2872019-03-02 00:26:30 +01001098 return -1;
1099 }
1100 }
1101 }
1102
Kevin Wolfbca52832020-07-17 12:54:25 +02001103 if (fd != -1 && (*open_flags & O_RDWR)) {
1104 ret = check_hdev_writable(fd);
1105 if (ret < 0) {
1106 qemu_close(fd);
1107 error_setg_errno(errp, -ret, "The device is not writable");
1108 return -1;
1109 }
1110 }
1111
Kevin Wolf5cec2872019-03-02 00:26:30 +01001112 return fd;
1113}
1114
Jeff Codyeeb6b452012-09-20 15:13:25 -04001115static int raw_reopen_prepare(BDRVReopenState *state,
1116 BlockReopenQueue *queue, Error **errp)
1117{
1118 BDRVRawState *s;
Fam Zheng4e6d13c2016-10-27 18:45:17 +08001119 BDRVRawReopenState *rs;
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +01001120 QemuOpts *opts;
Kevin Wolfa6aeca02019-03-01 22:48:45 +01001121 int ret;
Jeff Codyeeb6b452012-09-20 15:13:25 -04001122
1123 assert(state != NULL);
1124 assert(state->bs != NULL);
1125
1126 s = state->bs->opaque;
1127
Markus Armbruster5839e532014-08-19 10:31:08 +02001128 state->opaque = g_new0(BDRVRawReopenState, 1);
Fam Zheng4e6d13c2016-10-27 18:45:17 +08001129 rs = state->opaque;
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +01001130
1131 /* Handle options changes */
1132 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
Markus Armbrusteraf175e82020-07-07 18:06:03 +02001133 if (!qemu_opts_absorb_qdict(opts, state->options, errp)) {
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +01001134 ret = -EINVAL;
1135 goto out;
1136 }
1137
Stefan Hajnoczif357fcd2019-03-07 16:49:41 +00001138 rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true);
Alberto Garcia8d324572018-09-06 12:37:07 +03001139 rs->check_cache_dropped =
1140 qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false);
1141
1142 /* This driver's reopen function doesn't currently allow changing
1143 * other options, so let's put them back in the original QDict and
1144 * bdrv_reopen_prepare() will detect changes and complain. */
1145 qemu_opts_to_qdict(opts, state->options);
Jeff Codyeeb6b452012-09-20 15:13:25 -04001146
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03001147 /*
1148 * As part of reopen prepare we also want to create new fd by
1149 * raw_reconfigure_getfd(). But it wants updated "perm", when in
1150 * bdrv_reopen_multiple() .bdrv_reopen_prepare() callback called prior to
Michael Tokarev3202d8e2023-07-14 14:06:05 +03001151 * permission update. Happily, permission update is always a part
1152 * (a separate stage) of bdrv_reopen_multiple() so we can rely on this
1153 * fact and reconfigure fd in raw_check_perm().
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03001154 */
Kevin Wolfdf26a352014-07-16 17:48:17 +02001155
Kevin Wolfe0c9cf32019-03-07 19:07:35 +01001156 s->reopen_state = state;
Kevin Wolfa6aeca02019-03-01 22:48:45 +01001157 ret = 0;
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03001158
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +01001159out:
1160 qemu_opts_del(opts);
Jeff Codyeeb6b452012-09-20 15:13:25 -04001161 return ret;
1162}
1163
Jeff Codyeeb6b452012-09-20 15:13:25 -04001164static void raw_reopen_commit(BDRVReopenState *state)
1165{
Fam Zheng4e6d13c2016-10-27 18:45:17 +08001166 BDRVRawReopenState *rs = state->opaque;
Jeff Codyeeb6b452012-09-20 15:13:25 -04001167 BDRVRawState *s = state->bs->opaque;
1168
Stefan Hajnoczif357fcd2019-03-07 16:49:41 +00001169 s->drop_cache = rs->drop_cache;
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +01001170 s->check_cache_dropped = rs->check_cache_dropped;
Fam Zheng4e6d13c2016-10-27 18:45:17 +08001171 s->open_flags = rs->open_flags;
Jeff Codyeeb6b452012-09-20 15:13:25 -04001172 g_free(state->opaque);
1173 state->opaque = NULL;
Kevin Wolfe0c9cf32019-03-07 19:07:35 +01001174
1175 assert(s->reopen_state == state);
1176 s->reopen_state = NULL;
Jeff Codyeeb6b452012-09-20 15:13:25 -04001177}
1178
1179
1180static void raw_reopen_abort(BDRVReopenState *state)
1181{
Fam Zheng4e6d13c2016-10-27 18:45:17 +08001182 BDRVRawReopenState *rs = state->opaque;
Kevin Wolfe0c9cf32019-03-07 19:07:35 +01001183 BDRVRawState *s = state->bs->opaque;
Jeff Codyeeb6b452012-09-20 15:13:25 -04001184
1185 /* nothing to do if NULL, we didn't get far enough */
Fam Zheng4e6d13c2016-10-27 18:45:17 +08001186 if (rs == NULL) {
Jeff Codyeeb6b452012-09-20 15:13:25 -04001187 return;
1188 }
1189
Jeff Codyeeb6b452012-09-20 15:13:25 -04001190 g_free(state->opaque);
1191 state->opaque = NULL;
Kevin Wolfe0c9cf32019-03-07 19:07:35 +01001192
1193 assert(s->reopen_state == state);
1194 s->reopen_state = NULL;
Jeff Codyeeb6b452012-09-20 15:13:25 -04001195}
1196
Paolo Bonzini18473462021-04-14 19:52:26 +02001197static int hdev_get_max_hw_transfer(int fd, struct stat *st)
Fam Zheng6f607172016-06-03 10:07:02 +08001198{
1199#ifdef BLKSECTGET
Paolo Bonzini18473462021-04-14 19:52:26 +02001200 if (S_ISBLK(st->st_mode)) {
1201 unsigned short max_sectors = 0;
1202 if (ioctl(fd, BLKSECTGET, &max_sectors) == 0) {
1203 return max_sectors * 512;
1204 }
Fam Zheng6f607172016-06-03 10:07:02 +08001205 } else {
Paolo Bonzini18473462021-04-14 19:52:26 +02001206 int max_bytes = 0;
1207 if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) {
1208 return max_bytes;
1209 }
Fam Zheng6f607172016-06-03 10:07:02 +08001210 }
Paolo Bonzini18473462021-04-14 19:52:26 +02001211 return -errno;
Fam Zheng6f607172016-06-03 10:07:02 +08001212#else
1213 return -ENOSYS;
1214#endif
1215}
1216
Sam Lia735b562023-05-08 12:55:27 +08001217/*
1218 * Get a sysfs attribute value as character string.
1219 */
1220#ifdef CONFIG_LINUX
1221static int get_sysfs_str_val(struct stat *st, const char *attribute,
1222 char **val) {
1223 g_autofree char *sysfspath = NULL;
Sam Lia735b562023-05-08 12:55:27 +08001224 size_t len;
1225
1226 if (!S_ISBLK(st->st_mode)) {
1227 return -ENOTSUP;
1228 }
1229
1230 sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/%s",
1231 major(st->st_rdev), minor(st->st_rdev),
1232 attribute);
Sam Li29a242e2023-07-27 19:58:44 +08001233 if (!g_file_get_contents(sysfspath, val, &len, NULL)) {
Sam Lia735b562023-05-08 12:55:27 +08001234 return -ENOENT;
1235 }
1236
1237 /* The file is ended with '\n' */
1238 char *p;
1239 p = *val;
1240 if (*(p + len - 1) == '\n') {
1241 *(p + len - 1) = '\0';
1242 }
Sam Li29a242e2023-07-27 19:58:44 +08001243 return 0;
Sam Lia735b562023-05-08 12:55:27 +08001244}
1245#endif
1246
Sam Li6d43eaa2023-05-08 12:55:28 +08001247#if defined(CONFIG_BLKZONED)
Sam Lia735b562023-05-08 12:55:27 +08001248static int get_sysfs_zoned_model(struct stat *st, BlockZoneModel *zoned)
1249{
1250 g_autofree char *val = NULL;
1251 int ret;
1252
1253 ret = get_sysfs_str_val(st, "zoned", &val);
1254 if (ret < 0) {
1255 return ret;
1256 }
1257
1258 if (strcmp(val, "host-managed") == 0) {
1259 *zoned = BLK_Z_HM;
1260 } else if (strcmp(val, "host-aware") == 0) {
1261 *zoned = BLK_Z_HA;
1262 } else if (strcmp(val, "none") == 0) {
1263 *zoned = BLK_Z_NONE;
1264 } else {
1265 return -ENOTSUP;
1266 }
1267 return 0;
1268}
Sam Li6d43eaa2023-05-08 12:55:28 +08001269#endif /* defined(CONFIG_BLKZONED) */
Sam Lia735b562023-05-08 12:55:27 +08001270
1271/*
1272 * Get a sysfs attribute value as a long integer.
1273 */
1274#ifdef CONFIG_LINUX
1275static long get_sysfs_long_val(struct stat *st, const char *attribute)
1276{
1277 g_autofree char *str = NULL;
1278 const char *end;
1279 long val;
1280 int ret;
1281
1282 ret = get_sysfs_str_val(st, attribute, &str);
1283 if (ret < 0) {
1284 return ret;
1285 }
1286
1287 /* The file is ended with '\n', pass 'end' to accept that. */
1288 ret = qemu_strtol(str, &end, 10, &val);
1289 if (ret == 0 && end && *end == '\0') {
1290 ret = val;
1291 }
1292 return ret;
1293}
1294#endif
1295
Paolo Bonzini18473462021-04-14 19:52:26 +02001296static int hdev_get_max_segments(int fd, struct stat *st)
Fam Zheng9103f1c2017-03-08 20:08:14 +08001297{
1298#ifdef CONFIG_LINUX
Fam Zheng9103f1c2017-03-08 20:08:14 +08001299 int ret;
Maxim Levitsky867eccf2019-07-04 15:43:42 +03001300
Paolo Bonzini18473462021-04-14 19:52:26 +02001301 if (S_ISCHR(st->st_mode)) {
Paolo Bonzini8ad5ab62021-04-15 12:41:31 +02001302 if (ioctl(fd, SG_GET_SG_TABLESIZE, &ret) == 0) {
1303 return ret;
1304 }
1305 return -ENOTSUP;
1306 }
Sam Lia735b562023-05-08 12:55:27 +08001307 return get_sysfs_long_val(st, "max_segments");
Fam Zheng9103f1c2017-03-08 20:08:14 +08001308#else
1309 return -ENOTSUP;
1310#endif
1311}
1312
Sam Li6d43eaa2023-05-08 12:55:28 +08001313#if defined(CONFIG_BLKZONED)
Sam Lia3c41f02023-05-08 13:15:07 +08001314/*
1315 * If the reset_all flag is true, then the wps of zone whose state is
1316 * not readonly or offline should be all reset to the start sector.
1317 * Else, take the real wp of the device.
1318 */
1319static int get_zones_wp(BlockDriverState *bs, int fd, int64_t offset,
1320 unsigned int nrz, bool reset_all)
1321{
1322 struct blk_zone *blkz;
1323 size_t rep_size;
1324 uint64_t sector = offset >> BDRV_SECTOR_BITS;
1325 BlockZoneWps *wps = bs->wps;
1326 unsigned int j = offset / bs->bl.zone_size;
1327 unsigned int n = 0, i = 0;
1328 int ret;
1329 rep_size = sizeof(struct blk_zone_report) + nrz * sizeof(struct blk_zone);
1330 g_autofree struct blk_zone_report *rep = NULL;
1331
1332 rep = g_malloc(rep_size);
1333 blkz = (struct blk_zone *)(rep + 1);
1334 while (n < nrz) {
1335 memset(rep, 0, rep_size);
1336 rep->sector = sector;
1337 rep->nr_zones = nrz - n;
1338
1339 do {
1340 ret = ioctl(fd, BLKREPORTZONE, rep);
1341 } while (ret != 0 && errno == EINTR);
1342 if (ret != 0) {
1343 error_report("%d: ioctl BLKREPORTZONE at %" PRId64 " failed %d",
1344 fd, offset, errno);
1345 return -errno;
1346 }
1347
1348 if (!rep->nr_zones) {
1349 break;
1350 }
1351
1352 for (i = 0; i < rep->nr_zones; ++i, ++n, ++j) {
1353 /*
1354 * The wp tracking cares only about sequential writes required and
1355 * sequential write preferred zones so that the wp can advance to
1356 * the right location.
1357 * Use the most significant bit of the wp location to indicate the
1358 * zone type: 0 for SWR/SWP zones and 1 for conventional zones.
1359 */
1360 if (blkz[i].type == BLK_ZONE_TYPE_CONVENTIONAL) {
1361 wps->wp[j] |= 1ULL << 63;
1362 } else {
1363 switch(blkz[i].cond) {
1364 case BLK_ZONE_COND_FULL:
1365 case BLK_ZONE_COND_READONLY:
1366 /* Zone not writable */
1367 wps->wp[j] = (blkz[i].start + blkz[i].len) << BDRV_SECTOR_BITS;
1368 break;
1369 case BLK_ZONE_COND_OFFLINE:
1370 /* Zone not writable nor readable */
1371 wps->wp[j] = (blkz[i].start) << BDRV_SECTOR_BITS;
1372 break;
1373 default:
1374 if (reset_all) {
1375 wps->wp[j] = blkz[i].start << BDRV_SECTOR_BITS;
1376 } else {
1377 wps->wp[j] = blkz[i].wp << BDRV_SECTOR_BITS;
1378 }
1379 break;
1380 }
1381 }
1382 }
1383 sector = blkz[i - 1].start + blkz[i - 1].len;
1384 }
1385
1386 return 0;
1387}
1388
1389static void update_zones_wp(BlockDriverState *bs, int fd, int64_t offset,
1390 unsigned int nrz)
1391{
1392 if (get_zones_wp(bs, fd, offset, nrz, 0) < 0) {
1393 error_report("update zone wp failed");
1394 }
1395}
1396
Sam Lia735b562023-05-08 12:55:27 +08001397static void raw_refresh_zoned_limits(BlockDriverState *bs, struct stat *st,
1398 Error **errp)
1399{
Sam Lia3c41f02023-05-08 13:15:07 +08001400 BDRVRawState *s = bs->opaque;
Marc-André Lureaueb5d28c2024-09-24 16:45:48 +04001401 BlockZoneModel zoned = BLK_Z_NONE;
Sam Lia735b562023-05-08 12:55:27 +08001402 int ret;
1403
Sam Lia735b562023-05-08 12:55:27 +08001404 ret = get_sysfs_zoned_model(st, &zoned);
1405 if (ret < 0 || zoned == BLK_Z_NONE) {
Hanna Czenczek56d1a022023-08-24 17:53:40 +02001406 goto no_zoned;
Sam Lia735b562023-05-08 12:55:27 +08001407 }
1408 bs->bl.zoned = zoned;
Sam Li6d43eaa2023-05-08 12:55:28 +08001409
1410 ret = get_sysfs_long_val(st, "max_open_zones");
1411 if (ret >= 0) {
1412 bs->bl.max_open_zones = ret;
1413 }
1414
1415 ret = get_sysfs_long_val(st, "max_active_zones");
1416 if (ret >= 0) {
1417 bs->bl.max_active_zones = ret;
1418 }
1419
1420 /*
1421 * The zoned device must at least have zone size and nr_zones fields.
1422 */
1423 ret = get_sysfs_long_val(st, "chunk_sectors");
1424 if (ret < 0) {
1425 error_setg_errno(errp, -ret, "Unable to read chunk_sectors "
1426 "sysfs attribute");
Hanna Czenczek56d1a022023-08-24 17:53:40 +02001427 goto no_zoned;
Sam Li6d43eaa2023-05-08 12:55:28 +08001428 } else if (!ret) {
1429 error_setg(errp, "Read 0 from chunk_sectors sysfs attribute");
Hanna Czenczek56d1a022023-08-24 17:53:40 +02001430 goto no_zoned;
Sam Li6d43eaa2023-05-08 12:55:28 +08001431 }
1432 bs->bl.zone_size = ret << BDRV_SECTOR_BITS;
1433
1434 ret = get_sysfs_long_val(st, "nr_zones");
1435 if (ret < 0) {
1436 error_setg_errno(errp, -ret, "Unable to read nr_zones "
1437 "sysfs attribute");
Hanna Czenczek56d1a022023-08-24 17:53:40 +02001438 goto no_zoned;
Sam Li6d43eaa2023-05-08 12:55:28 +08001439 } else if (!ret) {
1440 error_setg(errp, "Read 0 from nr_zones sysfs attribute");
Hanna Czenczek56d1a022023-08-24 17:53:40 +02001441 goto no_zoned;
Sam Li6d43eaa2023-05-08 12:55:28 +08001442 }
1443 bs->bl.nr_zones = ret;
1444
1445 ret = get_sysfs_long_val(st, "zone_append_max_bytes");
1446 if (ret > 0) {
1447 bs->bl.max_append_sectors = ret >> BDRV_SECTOR_BITS;
1448 }
Sam Lia3c41f02023-05-08 13:15:07 +08001449
1450 ret = get_sysfs_long_val(st, "physical_block_size");
1451 if (ret >= 0) {
1452 bs->bl.write_granularity = ret;
1453 }
1454
1455 /* The refresh_limits() function can be called multiple times. */
1456 g_free(bs->wps);
1457 bs->wps = g_malloc(sizeof(BlockZoneWps) +
1458 sizeof(int64_t) * bs->bl.nr_zones);
1459 ret = get_zones_wp(bs, s->fd, 0, bs->bl.nr_zones, 0);
1460 if (ret < 0) {
1461 error_setg_errno(errp, -ret, "report wps failed");
Hanna Czenczek56d1a022023-08-24 17:53:40 +02001462 goto no_zoned;
Sam Lia3c41f02023-05-08 13:15:07 +08001463 }
1464 qemu_co_mutex_init(&bs->wps->colock);
Hanna Czenczek56d1a022023-08-24 17:53:40 +02001465 return;
1466
1467no_zoned:
1468 bs->bl.zoned = BLK_Z_NONE;
1469 g_free(bs->wps);
1470 bs->wps = NULL;
Sam Lia735b562023-05-08 12:55:27 +08001471}
Sam Li6d43eaa2023-05-08 12:55:28 +08001472#else /* !defined(CONFIG_BLKZONED) */
1473static void raw_refresh_zoned_limits(BlockDriverState *bs, struct stat *st,
1474 Error **errp)
1475{
1476 bs->bl.zoned = BLK_Z_NONE;
1477}
1478#endif /* !defined(CONFIG_BLKZONED) */
Sam Lia735b562023-05-08 12:55:27 +08001479
Kevin Wolf3baca892014-07-16 17:48:16 +02001480static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
Paolo Bonzinic25f53b2011-11-29 12:42:20 +01001481{
1482 BDRVRawState *s = bs->opaque;
Paolo Bonzini18473462021-04-14 19:52:26 +02001483 struct stat st;
Jeff Codyeeb6b452012-09-20 15:13:25 -04001484
Kevin Wolf5dbd0ce2021-11-16 11:14:31 +01001485 s->needs_alignment = raw_needs_alignment(bs);
Kevin Wolfdf26a352014-07-16 17:48:17 +02001486 raw_probe_alignment(bs, s->fd, errp);
Kevin Wolf5dbd0ce2021-11-16 11:14:31 +01001487
Denis V. Lunev4196d2f2015-05-12 17:30:55 +03001488 bs->bl.min_mem_alignment = s->buf_align;
Marc-André Lureau8e3b0cb2022-03-23 19:57:22 +04001489 bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size());
Paolo Bonzini18473462021-04-14 19:52:26 +02001490
1491 /*
1492 * Maximum transfers are best effort, so it is okay to ignore any
1493 * errors. That said, based on the man page errors in fstat would be
1494 * very much unexpected; the only possible case seems to be ENOMEM.
1495 */
1496 if (fstat(s->fd, &st)) {
1497 return;
1498 }
1499
Akihiko Odaki0dfc7af2021-07-05 22:04:56 +09001500#if defined(__APPLE__) && (__MACH__)
1501 struct statfs buf;
1502
1503 if (!fstatfs(s->fd, &buf)) {
1504 bs->bl.opt_transfer = buf.f_iosize;
1505 bs->bl.pdiscard_alignment = buf.f_bsize;
1506 }
1507#endif
1508
Denis V. Lunev006e1962022-08-17 10:37:35 +02001509 if (bdrv_is_sg(bs) || S_ISBLK(st.st_mode)) {
Paolo Bonzini18473462021-04-14 19:52:26 +02001510 int ret = hdev_get_max_hw_transfer(s->fd, &st);
1511
1512 if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) {
1513 bs->bl.max_hw_transfer = ret;
1514 }
1515
1516 ret = hdev_get_max_segments(s->fd, &st);
1517 if (ret > 0) {
Paolo Bonzinicc071622021-09-23 09:04:36 -04001518 bs->bl.max_hw_iov = ret;
Paolo Bonzini18473462021-04-14 19:52:26 +02001519 }
1520 }
Sam Lia735b562023-05-08 12:55:27 +08001521
1522 raw_refresh_zoned_limits(bs, &st, errp);
Paolo Bonzinic25f53b2011-11-29 12:42:20 +01001523}
bellard83f64092006-08-01 16:21:11 +00001524
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +01001525static int check_for_dasd(int fd)
1526{
1527#ifdef BIODASDINFO2
1528 struct dasd_information2_t info = {0};
1529
1530 return ioctl(fd, BIODASDINFO2, &info);
1531#else
1532 return -1;
1533#endif
1534}
1535
1536/**
1537 * Try to get @bs's logical and physical block size.
1538 * On success, store them in @bsz and return zero.
1539 * On failure, return negative errno.
1540 */
1541static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
1542{
1543 BDRVRawState *s = bs->opaque;
1544 int ret;
1545
Sam Li6d43eaa2023-05-08 12:55:28 +08001546 /* If DASD or zoned devices, get blocksizes */
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +01001547 if (check_for_dasd(s->fd) < 0) {
Sam Li6d43eaa2023-05-08 12:55:28 +08001548 /* zoned devices are not DASD */
1549 if (bs->bl.zoned == BLK_Z_NONE) {
1550 return -ENOTSUP;
1551 }
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +01001552 }
1553 ret = probe_logical_blocksize(s->fd, &bsz->log);
1554 if (ret < 0) {
1555 return ret;
1556 }
1557 return probe_physical_blocksize(s->fd, &bsz->phys);
1558}
1559
1560/**
1561 * Try to get @bs's geometry: cyls, heads, sectors.
1562 * On success, store them in @geo and return 0.
1563 * On failure return -errno.
1564 * (Allows block driver to assign default geometry values that guest sees)
1565 */
1566#ifdef __linux__
1567static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1568{
1569 BDRVRawState *s = bs->opaque;
1570 struct hd_geometry ioctl_geo = {0};
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +01001571
1572 /* If DASD, get its geometry */
1573 if (check_for_dasd(s->fd) < 0) {
1574 return -ENOTSUP;
1575 }
1576 if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) {
1577 return -errno;
1578 }
1579 /* HDIO_GETGEO may return success even though geo contains zeros
1580 (e.g. certain multipath setups) */
1581 if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) {
1582 return -ENOTSUP;
1583 }
1584 /* Do not return a geometry for partition */
1585 if (ioctl_geo.start != 0) {
1586 return -ENOTSUP;
1587 }
1588 geo->heads = ioctl_geo.heads;
1589 geo->sectors = ioctl_geo.sectors;
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +01001590 geo->cylinders = ioctl_geo.cylinders;
1591
1592 return 0;
1593}
1594#else /* __linux__ */
1595static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo)
1596{
1597 return -ENOTSUP;
1598}
1599#endif
1600
Kevin Wolf03425672018-10-31 11:30:42 +01001601#if defined(__linux__)
1602static int handle_aiocb_ioctl(void *opaque)
Paolo Bonzinide81a162012-05-25 11:46:27 +02001603{
Kevin Wolf03425672018-10-31 11:30:42 +01001604 RawPosixAIOData *aiocb = opaque;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001605 int ret;
1606
Nikita Ivanov37b0b242022-10-23 12:04:22 +03001607 ret = RETRY_ON_EINTR(
1608 ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf)
1609 );
Paolo Bonzinide81a162012-05-25 11:46:27 +02001610 if (ret == -1) {
1611 return -errno;
1612 }
1613
Paolo Bonzinib608c8d2013-01-10 15:28:35 +01001614 return 0;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001615}
Kevin Wolf03425672018-10-31 11:30:42 +01001616#endif /* linux */
Paolo Bonzinide81a162012-05-25 11:46:27 +02001617
Kevin Wolf06dc9bd2018-10-25 14:18:58 +01001618static int handle_aiocb_flush(void *opaque)
Paolo Bonzinide81a162012-05-25 11:46:27 +02001619{
Kevin Wolf06dc9bd2018-10-25 14:18:58 +01001620 RawPosixAIOData *aiocb = opaque;
Kevin Wolfe5bcf962017-03-22 22:00:05 +01001621 BDRVRawState *s = aiocb->bs->opaque;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001622 int ret;
1623
Kevin Wolfe5bcf962017-03-22 22:00:05 +01001624 if (s->page_cache_inconsistent) {
Daniel P. Berrangéc7ddc882021-04-15 14:28:16 +01001625 return -s->page_cache_inconsistent;
Kevin Wolfe5bcf962017-03-22 22:00:05 +01001626 }
1627
Paolo Bonzinide81a162012-05-25 11:46:27 +02001628 ret = qemu_fdatasync(aiocb->aio_fildes);
1629 if (ret == -1) {
Daniel P. Berrangé60ff2ae2021-04-15 14:28:16 +01001630 trace_file_flush_fdatasync_failed(errno);
1631
Kevin Wolfe5bcf962017-03-22 22:00:05 +01001632 /* There is no clear definition of the semantics of a failing fsync(),
1633 * so we may have to assume the worst. The sad truth is that this
1634 * assumption is correct for Linux. Some pages are now probably marked
1635 * clean in the page cache even though they are inconsistent with the
1636 * on-disk contents. The next fdatasync() call would succeed, but no
1637 * further writeback attempt will be made. We can't get back to a state
1638 * in which we know what is on disk (we would have to rewrite
1639 * everything that was touched since the last fdatasync() at least), so
1640 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1641 * really defined, I have little hope that other OSes are doing better.
1642 *
1643 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1644 * cache. */
1645 if ((s->open_flags & O_DIRECT) == 0) {
Daniel P. Berrangéc7ddc882021-04-15 14:28:16 +01001646 s->page_cache_inconsistent = errno;
Kevin Wolfe5bcf962017-03-22 22:00:05 +01001647 }
Paolo Bonzinide81a162012-05-25 11:46:27 +02001648 return -errno;
1649 }
1650 return 0;
1651}
1652
1653#ifdef CONFIG_PREADV
1654
1655static bool preadv_present = true;
1656
1657static ssize_t
1658qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1659{
1660 return preadv(fd, iov, nr_iov, offset);
1661}
1662
1663static ssize_t
1664qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1665{
1666 return pwritev(fd, iov, nr_iov, offset);
1667}
1668
1669#else
1670
1671static bool preadv_present = false;
1672
1673static ssize_t
1674qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1675{
1676 return -ENOSYS;
1677}
1678
1679static ssize_t
1680qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
1681{
1682 return -ENOSYS;
1683}
1684
1685#endif
1686
1687static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
1688{
1689 ssize_t len;
1690
Nikita Ivanov37b0b242022-10-23 12:04:22 +03001691 len = RETRY_ON_EINTR(
Sam Li4751d092023-05-08 13:15:08 +08001692 (aiocb->aio_type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) ?
Nikita Ivanov37b0b242022-10-23 12:04:22 +03001693 qemu_pwritev(aiocb->aio_fildes,
1694 aiocb->io.iov,
1695 aiocb->io.niov,
1696 aiocb->aio_offset) :
1697 qemu_preadv(aiocb->aio_fildes,
1698 aiocb->io.iov,
1699 aiocb->io.niov,
1700 aiocb->aio_offset)
1701 );
Paolo Bonzinide81a162012-05-25 11:46:27 +02001702
1703 if (len == -1) {
1704 return -errno;
1705 }
1706 return len;
1707}
1708
1709/*
1710 * Read/writes the data to/from a given linear buffer.
1711 *
1712 * Returns the number of bytes handles or -errno in case of an error. Short
1713 * reads are only returned if the end of the file is reached.
1714 */
1715static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
1716{
1717 ssize_t offset = 0;
1718 ssize_t len;
1719
1720 while (offset < aiocb->aio_nbytes) {
Sam Li4751d092023-05-08 13:15:08 +08001721 if (aiocb->aio_type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) {
Paolo Bonzinide81a162012-05-25 11:46:27 +02001722 len = pwrite(aiocb->aio_fildes,
1723 (const char *)buf + offset,
1724 aiocb->aio_nbytes - offset,
1725 aiocb->aio_offset + offset);
1726 } else {
1727 len = pread(aiocb->aio_fildes,
1728 buf + offset,
1729 aiocb->aio_nbytes - offset,
1730 aiocb->aio_offset + offset);
1731 }
1732 if (len == -1 && errno == EINTR) {
1733 continue;
Stefan Hajnoczi61ed73c2014-08-21 13:44:07 +01001734 } else if (len == -1 && errno == EINVAL &&
1735 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
1736 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
1737 offset > 0) {
1738 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1739 * after a short read. Assume that O_DIRECT short reads only occur
1740 * at EOF. Therefore this is a short read, not an I/O error.
1741 */
1742 break;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001743 } else if (len == -1) {
1744 offset = -errno;
1745 break;
1746 } else if (len == 0) {
1747 break;
1748 }
1749 offset += len;
1750 }
1751
1752 return offset;
1753}
1754
Kevin Wolf999e6b62018-10-25 14:18:58 +01001755static int handle_aiocb_rw(void *opaque)
Paolo Bonzinide81a162012-05-25 11:46:27 +02001756{
Kevin Wolf999e6b62018-10-25 14:18:58 +01001757 RawPosixAIOData *aiocb = opaque;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001758 ssize_t nbytes;
1759 char *buf;
1760
1761 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
1762 /*
1763 * If there is just a single buffer, and it is properly aligned
1764 * we can just use plain pread/pwrite without any problems.
1765 */
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001766 if (aiocb->io.niov == 1) {
Kevin Wolf54c7ca12018-10-31 09:43:17 +01001767 nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base);
1768 goto out;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001769 }
1770 /*
1771 * We have more than one iovec, and all are properly aligned.
1772 *
1773 * Try preadv/pwritev first and fall back to linearizing the
1774 * buffer if it's not supported.
1775 */
1776 if (preadv_present) {
1777 nbytes = handle_aiocb_rw_vector(aiocb);
1778 if (nbytes == aiocb->aio_nbytes ||
1779 (nbytes < 0 && nbytes != -ENOSYS)) {
Kevin Wolf54c7ca12018-10-31 09:43:17 +01001780 goto out;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001781 }
1782 preadv_present = false;
1783 }
1784
1785 /*
1786 * XXX(hch): short read/write. no easy way to handle the reminder
1787 * using these interfaces. For now retry using plain
1788 * pread/pwrite?
1789 */
1790 }
1791
1792 /*
1793 * Ok, we have to do it the hard way, copy all segments into
1794 * a single aligned buffer.
1795 */
Kevin Wolf50d4a852014-05-21 18:02:42 +02001796 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
1797 if (buf == NULL) {
Kevin Wolf54c7ca12018-10-31 09:43:17 +01001798 nbytes = -ENOMEM;
1799 goto out;
Kevin Wolf50d4a852014-05-21 18:02:42 +02001800 }
1801
Paolo Bonzinide81a162012-05-25 11:46:27 +02001802 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1803 char *p = buf;
1804 int i;
1805
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001806 for (i = 0; i < aiocb->io.niov; ++i) {
1807 memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len);
1808 p += aiocb->io.iov[i].iov_len;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001809 }
Kevin Wolf8eb029c2014-07-01 16:09:54 +02001810 assert(p - buf == aiocb->aio_nbytes);
Paolo Bonzinide81a162012-05-25 11:46:27 +02001811 }
1812
1813 nbytes = handle_aiocb_rw_linear(aiocb, buf);
Sam Li4751d092023-05-08 13:15:08 +08001814 if (!(aiocb->aio_type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND))) {
Paolo Bonzinide81a162012-05-25 11:46:27 +02001815 char *p = buf;
1816 size_t count = aiocb->aio_nbytes, copy;
1817 int i;
1818
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001819 for (i = 0; i < aiocb->io.niov && count; ++i) {
Paolo Bonzinide81a162012-05-25 11:46:27 +02001820 copy = count;
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001821 if (copy > aiocb->io.iov[i].iov_len) {
1822 copy = aiocb->io.iov[i].iov_len;
Paolo Bonzinide81a162012-05-25 11:46:27 +02001823 }
Kevin Wolfd57c44d2018-10-25 16:21:14 +01001824 memcpy(aiocb->io.iov[i].iov_base, p, copy);
Kevin Wolf8eb029c2014-07-01 16:09:54 +02001825 assert(count >= copy);
Paolo Bonzinide81a162012-05-25 11:46:27 +02001826 p += copy;
1827 count -= copy;
1828 }
Kevin Wolf8eb029c2014-07-01 16:09:54 +02001829 assert(count == 0);
Paolo Bonzinide81a162012-05-25 11:46:27 +02001830 }
1831 qemu_vfree(buf);
1832
Kevin Wolf54c7ca12018-10-31 09:43:17 +01001833out:
1834 if (nbytes == aiocb->aio_nbytes) {
1835 return 0;
1836 } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) {
1837 if (aiocb->aio_type & QEMU_AIO_WRITE) {
1838 return -EINVAL;
1839 } else {
1840 iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes,
1841 0, aiocb->aio_nbytes - nbytes);
1842 return 0;
1843 }
1844 } else {
1845 assert(nbytes < 0);
1846 return nbytes;
1847 }
Paolo Bonzinide81a162012-05-25 11:46:27 +02001848}
1849
Akihiko Odaki0dfc7af2021-07-05 22:04:56 +09001850#if defined(CONFIG_FALLOCATE) || defined(BLKZEROOUT) || defined(BLKDISCARD)
Denis V. Lunev1486df02015-01-30 11:42:11 +03001851static int translate_err(int err)
1852{
1853 if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP ||
1854 err == -ENOTTY) {
1855 err = -ENOTSUP;
1856 }
1857 return err;
1858}
Akihiko Odaki0dfc7af2021-07-05 22:04:56 +09001859#endif
Denis V. Lunev1486df02015-01-30 11:42:11 +03001860
Denis V. Lunevd50d8222015-01-30 11:42:15 +03001861#ifdef CONFIG_FALLOCATE
Denis V. Lunev0b991712015-01-30 11:42:12 +03001862static int do_fallocate(int fd, int mode, off_t offset, off_t len)
1863{
1864 do {
1865 if (fallocate(fd, mode, offset, len) == 0) {
1866 return 0;
1867 }
1868 } while (errno == EINTR);
1869 return translate_err(-errno);
1870}
1871#endif
1872
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001873static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb)
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001874{
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001875 int ret = -ENOTSUP;
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001876 BDRVRawState *s = aiocb->bs->opaque;
1877
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001878 if (!s->has_write_zeroes) {
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001879 return -ENOTSUP;
1880 }
1881
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001882#ifdef BLKZEROOUT
Kevin Wolf738301e2019-03-22 13:45:23 +01001883 /* The BLKZEROOUT implementation in the kernel doesn't set
1884 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1885 * fallbacks. */
1886 if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) {
1887 do {
1888 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
1889 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
1890 return 0;
1891 }
1892 } while (errno == EINTR);
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001893
Kevin Wolf738301e2019-03-22 13:45:23 +01001894 ret = translate_err(-errno);
Kevin Wolfeffecce2019-08-16 11:48:17 +02001895 if (ret == -ENOTSUP) {
1896 s->has_write_zeroes = false;
1897 }
Kevin Wolf738301e2019-03-22 13:45:23 +01001898 }
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001899#endif
1900
Paolo Bonzini97a2ae32013-11-22 13:39:57 +01001901 return ret;
1902}
1903
Kevin Wolf7154d8a2018-10-25 14:18:58 +01001904static int handle_aiocb_write_zeroes(void *opaque)
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001905{
Kevin Wolf7154d8a2018-10-25 14:18:58 +01001906 RawPosixAIOData *aiocb = opaque;
Denis V. Lunev70d91102017-08-04 18:10:11 +03001907#ifdef CONFIG_FALLOCATE
Max Reitzb2c6f232019-08-23 15:03:40 +02001908 BDRVRawState *s = aiocb->bs->opaque;
Denis V. Lunev70d91102017-08-04 18:10:11 +03001909 int64_t len;
1910#endif
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001911
1912 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
1913 return handle_aiocb_write_zeroes_block(aiocb);
1914 }
1915
Denis V. Lunevb953f072015-01-30 11:42:14 +03001916#ifdef CONFIG_FALLOCATE_ZERO_RANGE
1917 if (s->has_write_zeroes) {
1918 int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE,
1919 aiocb->aio_offset, aiocb->aio_nbytes);
Thomas Huthfa95e9f2021-05-27 19:20:20 +02001920 if (ret == -ENOTSUP) {
1921 s->has_write_zeroes = false;
1922 } else if (ret == 0 || ret != -EINVAL) {
Denis V. Lunevb953f072015-01-30 11:42:14 +03001923 return ret;
1924 }
Thomas Huthfa95e9f2021-05-27 19:20:20 +02001925 /*
1926 * Note: Some file systems do not like unaligned byte ranges, and
1927 * return EINVAL in such a case, though they should not do it according
1928 * to the man-page of fallocate(). Thus we simply ignore this return
1929 * value and try the other fallbacks instead.
1930 */
Denis V. Lunevb953f072015-01-30 11:42:14 +03001931 }
1932#endif
1933
Denis V. Lunev1cdc3232015-01-30 11:42:16 +03001934#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1935 if (s->has_discard && s->has_fallocate) {
1936 int ret = do_fallocate(s->fd,
1937 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1938 aiocb->aio_offset, aiocb->aio_nbytes);
1939 if (ret == 0) {
1940 ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1941 if (ret == 0 || ret != -ENOTSUP) {
1942 return ret;
1943 }
1944 s->has_fallocate = false;
Thomas Huth73ebf292021-05-27 19:20:19 +02001945 } else if (ret == -EINVAL) {
1946 /*
1947 * Some file systems like older versions of GPFS do not like un-
1948 * aligned byte ranges, and return EINVAL in such a case, though
1949 * they should not do it according to the man-page of fallocate().
1950 * Warn about the bad filesystem and try the final fallback instead.
1951 */
1952 warn_report_once("Your file system is misbehaving: "
1953 "fallocate(FALLOC_FL_PUNCH_HOLE) returned EINVAL. "
Michael Tokarev68857f12021-08-18 17:06:53 +03001954 "Please report this bug to your file system "
Thomas Huth73ebf292021-05-27 19:20:19 +02001955 "vendor.");
Denis V. Lunev1cdc3232015-01-30 11:42:16 +03001956 } else if (ret != -ENOTSUP) {
1957 return ret;
1958 } else {
1959 s->has_discard = false;
1960 }
1961 }
1962#endif
1963
Denis V. Lunevd50d8222015-01-30 11:42:15 +03001964#ifdef CONFIG_FALLOCATE
Denis V. Lunev70d91102017-08-04 18:10:11 +03001965 /* Last resort: we are trying to extend the file with zeroed data. This
1966 * can be done via fallocate(fd, 0) */
Paolo Bonzini36c6c872023-06-01 13:51:34 +02001967 len = raw_getlength(aiocb->bs);
Denis V. Lunev70d91102017-08-04 18:10:11 +03001968 if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) {
Denis V. Lunevd50d8222015-01-30 11:42:15 +03001969 int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes);
1970 if (ret == 0 || ret != -ENOTSUP) {
1971 return ret;
1972 }
1973 s->has_fallocate = false;
1974 }
1975#endif
1976
Denis V. Lunev37cc9f72015-01-30 11:42:13 +03001977 return -ENOTSUP;
1978}
1979
Kevin Wolf7154d8a2018-10-25 14:18:58 +01001980static int handle_aiocb_write_zeroes_unmap(void *opaque)
Kevin Wolf34fa1102018-07-26 11:28:30 +02001981{
Kevin Wolf7154d8a2018-10-25 14:18:58 +01001982 RawPosixAIOData *aiocb = opaque;
Kevin Wolf34fa1102018-07-26 11:28:30 +02001983 BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque;
Kevin Wolf34fa1102018-07-26 11:28:30 +02001984
1985 /* First try to write zeros and unmap at the same time */
1986
1987#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
Simran Singhalb3ac2b92020-04-01 22:23:14 +05301988 int ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1989 aiocb->aio_offset, aiocb->aio_nbytes);
Antoine Damhetbae127d2020-07-17 15:56:04 +02001990 switch (ret) {
1991 case -ENOTSUP:
1992 case -EINVAL:
Maxim Levitskyece4fa92020-11-11 17:39:12 +02001993 case -EBUSY:
Antoine Damhetbae127d2020-07-17 15:56:04 +02001994 break;
1995 default:
Kevin Wolf34fa1102018-07-26 11:28:30 +02001996 return ret;
1997 }
1998#endif
1999
Kevin Wolf34fa1102018-07-26 11:28:30 +02002000 /* If we couldn't manage to unmap while guaranteed that the area reads as
2001 * all-zero afterwards, just write zeroes without unmapping */
Simran Singhalb3ac2b92020-04-01 22:23:14 +05302002 return handle_aiocb_write_zeroes(aiocb);
Kevin Wolf34fa1102018-07-26 11:28:30 +02002003}
2004
Fam Zheng1efad062018-06-01 17:26:43 +08002005#ifndef HAVE_COPY_FILE_RANGE
2006static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd,
2007 off_t *out_off, size_t len, unsigned int flags)
2008{
2009#ifdef __NR_copy_file_range
2010 return syscall(__NR_copy_file_range, in_fd, in_off, out_fd,
2011 out_off, len, flags);
2012#else
2013 errno = ENOSYS;
2014 return -1;
2015#endif
2016}
2017#endif
2018
Sam Li6d43eaa2023-05-08 12:55:28 +08002019/*
2020 * parse_zone - Fill a zone descriptor
2021 */
2022#if defined(CONFIG_BLKZONED)
2023static inline int parse_zone(struct BlockZoneDescriptor *zone,
2024 const struct blk_zone *blkz) {
2025 zone->start = blkz->start << BDRV_SECTOR_BITS;
2026 zone->length = blkz->len << BDRV_SECTOR_BITS;
2027 zone->wp = blkz->wp << BDRV_SECTOR_BITS;
2028
2029#ifdef HAVE_BLK_ZONE_REP_CAPACITY
2030 zone->cap = blkz->capacity << BDRV_SECTOR_BITS;
2031#else
2032 zone->cap = blkz->len << BDRV_SECTOR_BITS;
2033#endif
2034
2035 switch (blkz->type) {
2036 case BLK_ZONE_TYPE_SEQWRITE_REQ:
2037 zone->type = BLK_ZT_SWR;
2038 break;
2039 case BLK_ZONE_TYPE_SEQWRITE_PREF:
2040 zone->type = BLK_ZT_SWP;
2041 break;
2042 case BLK_ZONE_TYPE_CONVENTIONAL:
2043 zone->type = BLK_ZT_CONV;
2044 break;
2045 default:
2046 error_report("Unsupported zone type: 0x%x", blkz->type);
2047 return -ENOTSUP;
2048 }
2049
2050 switch (blkz->cond) {
2051 case BLK_ZONE_COND_NOT_WP:
2052 zone->state = BLK_ZS_NOT_WP;
2053 break;
2054 case BLK_ZONE_COND_EMPTY:
2055 zone->state = BLK_ZS_EMPTY;
2056 break;
2057 case BLK_ZONE_COND_IMP_OPEN:
2058 zone->state = BLK_ZS_IOPEN;
2059 break;
2060 case BLK_ZONE_COND_EXP_OPEN:
2061 zone->state = BLK_ZS_EOPEN;
2062 break;
2063 case BLK_ZONE_COND_CLOSED:
2064 zone->state = BLK_ZS_CLOSED;
2065 break;
2066 case BLK_ZONE_COND_READONLY:
2067 zone->state = BLK_ZS_RDONLY;
2068 break;
2069 case BLK_ZONE_COND_FULL:
2070 zone->state = BLK_ZS_FULL;
2071 break;
2072 case BLK_ZONE_COND_OFFLINE:
2073 zone->state = BLK_ZS_OFFLINE;
2074 break;
2075 default:
2076 error_report("Unsupported zone state: 0x%x", blkz->cond);
2077 return -ENOTSUP;
2078 }
2079 return 0;
2080}
2081#endif
2082
2083#if defined(CONFIG_BLKZONED)
2084static int handle_aiocb_zone_report(void *opaque)
2085{
2086 RawPosixAIOData *aiocb = opaque;
2087 int fd = aiocb->aio_fildes;
2088 unsigned int *nr_zones = aiocb->zone_report.nr_zones;
2089 BlockZoneDescriptor *zones = aiocb->zone_report.zones;
2090 /* zoned block devices use 512-byte sectors */
2091 uint64_t sector = aiocb->aio_offset / 512;
2092
2093 struct blk_zone *blkz;
2094 size_t rep_size;
2095 unsigned int nrz;
2096 int ret;
2097 unsigned int n = 0, i = 0;
2098
2099 nrz = *nr_zones;
2100 rep_size = sizeof(struct blk_zone_report) + nrz * sizeof(struct blk_zone);
2101 g_autofree struct blk_zone_report *rep = NULL;
2102 rep = g_malloc(rep_size);
2103
2104 blkz = (struct blk_zone *)(rep + 1);
2105 while (n < nrz) {
2106 memset(rep, 0, rep_size);
2107 rep->sector = sector;
2108 rep->nr_zones = nrz - n;
2109
2110 do {
2111 ret = ioctl(fd, BLKREPORTZONE, rep);
2112 } while (ret != 0 && errno == EINTR);
2113 if (ret != 0) {
2114 error_report("%d: ioctl BLKREPORTZONE at %" PRId64 " failed %d",
2115 fd, sector, errno);
2116 return -errno;
2117 }
2118
2119 if (!rep->nr_zones) {
2120 break;
2121 }
2122
2123 for (i = 0; i < rep->nr_zones; i++, n++) {
2124 ret = parse_zone(&zones[n], &blkz[i]);
2125 if (ret != 0) {
2126 return ret;
2127 }
2128
2129 /* The next report should start after the last zone reported */
2130 sector = blkz[i].start + blkz[i].len;
2131 }
2132 }
2133
2134 *nr_zones = n;
2135 return 0;
2136}
2137#endif
2138
2139#if defined(CONFIG_BLKZONED)
2140static int handle_aiocb_zone_mgmt(void *opaque)
2141{
2142 RawPosixAIOData *aiocb = opaque;
2143 int fd = aiocb->aio_fildes;
2144 uint64_t sector = aiocb->aio_offset / 512;
2145 int64_t nr_sectors = aiocb->aio_nbytes / 512;
2146 struct blk_zone_range range;
2147 int ret;
2148
2149 /* Execute the operation */
2150 range.sector = sector;
2151 range.nr_sectors = nr_sectors;
2152 do {
2153 ret = ioctl(fd, aiocb->zone_mgmt.op, &range);
2154 } while (ret != 0 && errno == EINTR);
2155
Sam Lia3c41f02023-05-08 13:15:07 +08002156 return ret < 0 ? -errno : ret;
Sam Li6d43eaa2023-05-08 12:55:28 +08002157}
2158#endif
2159
Kevin Wolf58a209c2018-10-25 14:18:58 +01002160static int handle_aiocb_copy_range(void *opaque)
Fam Zheng1efad062018-06-01 17:26:43 +08002161{
Kevin Wolf58a209c2018-10-25 14:18:58 +01002162 RawPosixAIOData *aiocb = opaque;
Fam Zheng1efad062018-06-01 17:26:43 +08002163 uint64_t bytes = aiocb->aio_nbytes;
2164 off_t in_off = aiocb->aio_offset;
Kevin Wolfd57c44d2018-10-25 16:21:14 +01002165 off_t out_off = aiocb->copy_range.aio_offset2;
Fam Zheng1efad062018-06-01 17:26:43 +08002166
2167 while (bytes) {
2168 ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off,
Kevin Wolfd57c44d2018-10-25 16:21:14 +01002169 aiocb->copy_range.aio_fd2, &out_off,
Fam Zheng1efad062018-06-01 17:26:43 +08002170 bytes, 0);
Fam Zhengecc983a2018-07-10 14:31:16 +08002171 trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off,
Kevin Wolfd57c44d2018-10-25 16:21:14 +01002172 aiocb->copy_range.aio_fd2, out_off, bytes,
2173 0, ret);
Fam Zhengc436e3d2018-06-29 14:03:28 +08002174 if (ret == 0) {
2175 /* No progress (e.g. when beyond EOF), let the caller fall back to
2176 * buffer I/O. */
2177 return -ENOSPC;
Fam Zheng1efad062018-06-01 17:26:43 +08002178 }
2179 if (ret < 0) {
Fam Zhengc436e3d2018-06-29 14:03:28 +08002180 switch (errno) {
2181 case ENOSYS:
Fam Zheng1efad062018-06-01 17:26:43 +08002182 return -ENOTSUP;
Fam Zhengc436e3d2018-06-29 14:03:28 +08002183 case EINTR:
2184 continue;
2185 default:
Fam Zheng1efad062018-06-01 17:26:43 +08002186 return -errno;
2187 }
2188 }
Fam Zheng1efad062018-06-01 17:26:43 +08002189 bytes -= ret;
2190 }
2191 return 0;
2192}
2193
Kevin Wolf46ee0f42018-10-25 14:18:58 +01002194static int handle_aiocb_discard(void *opaque)
Paolo Bonzini82380102013-01-14 16:26:55 +01002195{
Kevin Wolf46ee0f42018-10-25 14:18:58 +01002196 RawPosixAIOData *aiocb = opaque;
Ari Sundholm13a02832021-10-19 14:09:55 +03002197 int ret = -ENOTSUP;
Paolo Bonzini82380102013-01-14 16:26:55 +01002198 BDRVRawState *s = aiocb->bs->opaque;
2199
Paolo Bonzini7ce21012013-11-22 13:39:47 +01002200 if (!s->has_discard) {
2201 return -ENOTSUP;
Paolo Bonzini82380102013-01-14 16:26:55 +01002202 }
2203
2204 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
2205#ifdef BLKDISCARD
2206 do {
2207 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
2208 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
2209 return 0;
2210 }
2211 } while (errno == EINTR);
2212
Akihiko Odaki0dfc7af2021-07-05 22:04:56 +09002213 ret = translate_err(-errno);
Paolo Bonzini82380102013-01-14 16:26:55 +01002214#endif
2215 } else {
Paolo Bonzini82380102013-01-14 16:26:55 +01002216#ifdef CONFIG_FALLOCATE_PUNCH_HOLE
Denis V. Lunev0b991712015-01-30 11:42:12 +03002217 ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
2218 aiocb->aio_offset, aiocb->aio_nbytes);
Ari Sundholm13a02832021-10-19 14:09:55 +03002219 ret = translate_err(ret);
Akihiko Odaki0dfc7af2021-07-05 22:04:56 +09002220#elif defined(__APPLE__) && (__MACH__)
2221 fpunchhole_t fpunchhole;
2222 fpunchhole.fp_flags = 0;
2223 fpunchhole.reserved = 0;
2224 fpunchhole.fp_offset = aiocb->aio_offset;
2225 fpunchhole.fp_length = aiocb->aio_nbytes;
2226 if (fcntl(s->fd, F_PUNCHHOLE, &fpunchhole) == -1) {
2227 ret = errno == ENODEV ? -ENOTSUP : -errno;
2228 } else {
2229 ret = 0;
2230 }
Paolo Bonzini82380102013-01-14 16:26:55 +01002231#endif
2232 }
2233
Denis V. Lunev1486df02015-01-30 11:42:11 +03002234 if (ret == -ENOTSUP) {
Paolo Bonzini7ce21012013-11-22 13:39:47 +01002235 s->has_discard = false;
Paolo Bonzini82380102013-01-14 16:26:55 +01002236 }
2237 return ret;
2238}
2239
Nir Soffer3a200132019-08-27 04:05:27 +03002240/*
2241 * Help alignment probing by allocating the first block.
2242 *
2243 * When reading with direct I/O from unallocated area on Gluster backed by XFS,
2244 * reading succeeds regardless of request length. In this case we fallback to
2245 * safe alignment which is not optimal. Allocating the first block avoids this
2246 * fallback.
2247 *
2248 * fd may be opened with O_DIRECT, but we don't know the buffer alignment or
2249 * request alignment, so we use safe values.
2250 *
2251 * Returns: 0 on success, -errno on failure. Since this is an optimization,
2252 * caller may ignore failures.
2253 */
2254static int allocate_first_block(int fd, size_t max_size)
2255{
2256 size_t write_size = (max_size < MAX_BLOCKSIZE)
2257 ? BDRV_SECTOR_SIZE
2258 : MAX_BLOCKSIZE;
Marc-André Lureau8e3b0cb2022-03-23 19:57:22 +04002259 size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size());
Nir Soffer3a200132019-08-27 04:05:27 +03002260 void *buf;
2261 ssize_t n;
2262 int ret;
2263
2264 buf = qemu_memalign(max_align, write_size);
2265 memset(buf, 0, write_size);
2266
Nikita Ivanov37b0b242022-10-23 12:04:22 +03002267 n = RETRY_ON_EINTR(pwrite(fd, buf, write_size, 0));
Nir Soffer3a200132019-08-27 04:05:27 +03002268
2269 ret = (n == -1) ? -errno : 0;
2270
2271 qemu_vfree(buf);
2272 return ret;
2273}
2274
Kevin Wolf29cb4c02018-10-25 14:18:58 +01002275static int handle_aiocb_truncate(void *opaque)
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002276{
Kevin Wolf29cb4c02018-10-25 14:18:58 +01002277 RawPosixAIOData *aiocb = opaque;
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002278 int result = 0;
2279 int64_t current_length = 0;
2280 char *buf = NULL;
2281 struct stat st;
2282 int fd = aiocb->aio_fildes;
2283 int64_t offset = aiocb->aio_offset;
Kevin Wolfd57c44d2018-10-25 16:21:14 +01002284 PreallocMode prealloc = aiocb->truncate.prealloc;
2285 Error **errp = aiocb->truncate.errp;
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002286
2287 if (fstat(fd, &st) < 0) {
2288 result = -errno;
2289 error_setg_errno(errp, -result, "Could not stat file");
2290 return result;
2291 }
2292
2293 current_length = st.st_size;
Kevin Wolfd57c44d2018-10-25 16:21:14 +01002294 if (current_length > offset && prealloc != PREALLOC_MODE_OFF) {
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002295 error_setg(errp, "Cannot use preallocation for shrinking files");
2296 return -ENOTSUP;
2297 }
2298
Kevin Wolfd57c44d2018-10-25 16:21:14 +01002299 switch (prealloc) {
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002300#ifdef CONFIG_POSIX_FALLOCATE
2301 case PREALLOC_MODE_FALLOC:
2302 /*
2303 * Truncating before posix_fallocate() makes it about twice slower on
2304 * file systems that do not support fallocate(), trying to check if a
2305 * block is allocated before allocating it, so don't do that here.
2306 */
2307 if (offset != current_length) {
2308 result = -posix_fallocate(fd, current_length,
2309 offset - current_length);
2310 if (result != 0) {
2311 /* posix_fallocate() doesn't set errno. */
2312 error_setg_errno(errp, -result,
2313 "Could not preallocate new data");
Nir Soffer3a200132019-08-27 04:05:27 +03002314 } else if (current_length == 0) {
2315 /*
2316 * posix_fallocate() uses fallocate() if the filesystem
2317 * supports it, or fallback to manually writing zeroes. If
2318 * fallocate() was used, unaligned reads from the fallocated
2319 * area in raw_probe_alignment() will succeed, hence we need to
2320 * allocate the first block.
2321 *
2322 * Optimize future alignment probing; ignore failures.
2323 */
2324 allocate_first_block(fd, offset);
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002325 }
2326 } else {
2327 result = 0;
2328 }
2329 goto out;
2330#endif
2331 case PREALLOC_MODE_FULL:
2332 {
2333 int64_t num = 0, left = offset - current_length;
2334 off_t seek_result;
2335
2336 /*
2337 * Knowing the final size from the beginning could allow the file
2338 * system driver to do less allocations and possibly avoid
2339 * fragmentation of the file.
2340 */
2341 if (ftruncate(fd, offset) != 0) {
2342 result = -errno;
2343 error_setg_errno(errp, -result, "Could not resize file");
2344 goto out;
2345 }
2346
2347 buf = g_malloc0(65536);
2348
2349 seek_result = lseek(fd, current_length, SEEK_SET);
2350 if (seek_result < 0) {
2351 result = -errno;
2352 error_setg_errno(errp, -result,
2353 "Failed to seek to the old end of file");
2354 goto out;
2355 }
2356
2357 while (left > 0) {
2358 num = MIN(left, 65536);
2359 result = write(fd, buf, num);
2360 if (result < 0) {
Fam Zhenga1c81f42018-07-27 14:53:14 +08002361 if (errno == EINTR) {
2362 continue;
2363 }
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002364 result = -errno;
2365 error_setg_errno(errp, -result,
2366 "Could not write zeros for preallocation");
2367 goto out;
2368 }
2369 left -= result;
2370 }
2371 if (result >= 0) {
2372 result = fsync(fd);
2373 if (result < 0) {
2374 result = -errno;
2375 error_setg_errno(errp, -result,
2376 "Could not flush file to disk");
2377 goto out;
2378 }
2379 }
2380 goto out;
2381 }
2382 case PREALLOC_MODE_OFF:
2383 if (ftruncate(fd, offset) != 0) {
2384 result = -errno;
2385 error_setg_errno(errp, -result, "Could not resize file");
Nir Soffer3a200132019-08-27 04:05:27 +03002386 } else if (current_length == 0 && offset > current_length) {
2387 /* Optimize future alignment probing; ignore failures. */
2388 allocate_first_block(fd, offset);
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002389 }
2390 return result;
2391 default:
2392 result = -ENOTSUP;
2393 error_setg(errp, "Unsupported preallocation mode: %s",
Kevin Wolfd57c44d2018-10-25 16:21:14 +01002394 PreallocMode_str(prealloc));
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002395 return result;
2396 }
2397
2398out:
2399 if (result < 0) {
2400 if (ftruncate(fd, current_length) < 0) {
2401 error_report("Failed to restore old file length: %s",
2402 strerror(errno));
2403 }
2404 }
2405
2406 g_free(buf);
2407 return result;
2408}
2409
Emanuele Giuseppe Esposito0fdb7312023-02-03 08:17:30 -05002410static int coroutine_fn raw_thread_pool_submit(ThreadPoolFunc func, void *arg)
Kevin Wolf5d5de252018-10-26 16:53:47 +01002411{
Emanuele Giuseppe Espositoaef04fc2023-02-03 08:17:31 -05002412 return thread_pool_submit_co(func, arg);
Kevin Wolf5d5de252018-10-26 16:53:47 +01002413}
2414
Keith Buscha7c5f672022-09-29 13:05:22 -07002415/*
2416 * Check if all memory in this vector is sector aligned.
2417 */
2418static bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
2419{
2420 int i;
2421 size_t alignment = bdrv_min_mem_align(bs);
Keith Busch25474d92022-09-29 13:05:23 -07002422 size_t len = bs->bl.request_alignment;
Keith Buscha7c5f672022-09-29 13:05:22 -07002423 IO_CODE();
2424
2425 for (i = 0; i < qiov->niov; i++) {
2426 if ((uintptr_t) qiov->iov[i].iov_base % alignment) {
2427 return false;
2428 }
Keith Busch25474d92022-09-29 13:05:23 -07002429 if (qiov->iov[i].iov_len % len) {
Keith Buscha7c5f672022-09-29 13:05:22 -07002430 return false;
2431 }
2432 }
2433
2434 return true;
2435}
2436
Stefan Hajnoczicd0c0db2023-09-14 10:00:58 -04002437#ifdef CONFIG_LINUX_IO_URING
2438static inline bool raw_check_linux_io_uring(BDRVRawState *s)
2439{
2440 Error *local_err = NULL;
2441 AioContext *ctx;
2442
2443 if (!s->use_linux_io_uring) {
2444 return false;
2445 }
2446
2447 ctx = qemu_get_current_aio_context();
2448 if (unlikely(!aio_setup_linux_io_uring(ctx, &local_err))) {
2449 error_reportf_err(local_err, "Unable to use linux io_uring, "
2450 "falling back to thread pool: ");
2451 s->use_linux_io_uring = false;
2452 return false;
2453 }
2454 return true;
2455}
2456#endif
2457
2458#ifdef CONFIG_LINUX_AIO
2459static inline bool raw_check_linux_aio(BDRVRawState *s)
2460{
2461 Error *local_err = NULL;
2462 AioContext *ctx;
2463
2464 if (!s->use_linux_aio) {
2465 return false;
2466 }
2467
2468 ctx = qemu_get_current_aio_context();
2469 if (unlikely(!aio_setup_linux_aio(ctx, &local_err))) {
2470 error_reportf_err(local_err, "Unable to use Linux AIO, "
2471 "falling back to thread pool: ");
2472 s->use_linux_aio = false;
2473 return false;
2474 }
2475 return true;
2476}
2477#endif
2478
Naohiro Aotaad4feac2023-10-30 16:38:53 +09002479static int coroutine_fn raw_co_prw(BlockDriverState *bs, int64_t *offset_ptr,
Kevin Wolf9d52aa32016-06-03 17:36:27 +02002480 uint64_t bytes, QEMUIOVector *qiov, int type)
bellard83f64092006-08-01 16:21:11 +00002481{
pbrookce1a14d2006-08-07 02:38:06 +00002482 BDRVRawState *s = bs->opaque;
Kevin Wolf999e6b62018-10-25 14:18:58 +01002483 RawPosixAIOData acb;
Sam Lia3c41f02023-05-08 13:15:07 +08002484 int ret;
Naohiro Aotaad4feac2023-10-30 16:38:53 +09002485 uint64_t offset = *offset_ptr;
bellard83f64092006-08-01 16:21:11 +00002486
bellard19cb3732006-08-19 11:45:59 +00002487 if (fd_open(bs) < 0)
Kevin Wolf2174f122014-08-06 17:18:07 +02002488 return -EIO;
Sam Lia3c41f02023-05-08 13:15:07 +08002489#if defined(CONFIG_BLKZONED)
Hanna Czenczek4b5d80f2023-08-24 17:53:41 +02002490 if ((type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) &&
2491 bs->bl.zoned != BLK_Z_NONE) {
Sam Lia3c41f02023-05-08 13:15:07 +08002492 qemu_co_mutex_lock(&bs->wps->colock);
Hanna Czenczek4b5d80f2023-08-24 17:53:41 +02002493 if (type & QEMU_AIO_ZONE_APPEND) {
Sam Li4751d092023-05-08 13:15:08 +08002494 int index = offset / bs->bl.zone_size;
2495 offset = bs->wps->wp[index];
2496 }
Sam Lia3c41f02023-05-08 13:15:07 +08002497 }
2498#endif
bellard19cb3732006-08-19 11:45:59 +00002499
aliguorif141eaf2009-04-07 18:43:24 +00002500 /*
Aarushi Mehtac6447512020-01-20 14:18:51 +00002501 * When using O_DIRECT, the request must be aligned to be able to use
2502 * either libaio or io_uring interface. If not fail back to regular thread
2503 * pool read/write code which emulates this for us if we
2504 * set QEMU_AIO_MISALIGNED.
aliguorif141eaf2009-04-07 18:43:24 +00002505 */
Aarushi Mehtac6447512020-01-20 14:18:51 +00002506 if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) {
2507 type |= QEMU_AIO_MISALIGNED;
2508#ifdef CONFIG_LINUX_IO_URING
Stefan Hajnoczicd0c0db2023-09-14 10:00:58 -04002509 } else if (raw_check_linux_io_uring(s)) {
Aarushi Mehtac6447512020-01-20 14:18:51 +00002510 assert(qiov->size == bytes);
Sam Lia3c41f02023-05-08 13:15:07 +08002511 ret = luring_co_submit(bs, s->fd, offset, qiov, type);
2512 goto out;
Stefan Weile44bd6f2009-08-28 14:39:31 +02002513#endif
Aarushi Mehtac6447512020-01-20 14:18:51 +00002514#ifdef CONFIG_LINUX_AIO
Stefan Hajnoczicd0c0db2023-09-14 10:00:58 -04002515 } else if (raw_check_linux_aio(s)) {
Aarushi Mehtac6447512020-01-20 14:18:51 +00002516 assert(qiov->size == bytes);
Sam Lia3c41f02023-05-08 13:15:07 +08002517 ret = laio_co_submit(s->fd, offset, qiov, type,
2518 s->aio_max_batch);
2519 goto out;
Aarushi Mehtac6447512020-01-20 14:18:51 +00002520#endif
Christoph Hellwig9ef91a62009-08-20 16:58:19 +02002521 }
aliguorif141eaf2009-04-07 18:43:24 +00002522
Kevin Wolf999e6b62018-10-25 14:18:58 +01002523 acb = (RawPosixAIOData) {
2524 .bs = bs,
2525 .aio_fildes = s->fd,
2526 .aio_type = type,
2527 .aio_offset = offset,
2528 .aio_nbytes = bytes,
2529 .io = {
2530 .iov = qiov->iov,
2531 .niov = qiov->niov,
2532 },
2533 };
2534
2535 assert(qiov->size == bytes);
Sam Lia3c41f02023-05-08 13:15:07 +08002536 ret = raw_thread_pool_submit(handle_aiocb_rw, &acb);
2537 goto out; /* Avoid the compiler err of unused label */
2538
2539out:
2540#if defined(CONFIG_BLKZONED)
Hanna Czenczekd31b50a2023-08-24 17:53:43 +02002541 if ((type & (QEMU_AIO_WRITE | QEMU_AIO_ZONE_APPEND)) &&
2542 bs->bl.zoned != BLK_Z_NONE) {
2543 BlockZoneWps *wps = bs->wps;
2544 if (ret == 0) {
Sam Lia3c41f02023-05-08 13:15:07 +08002545 uint64_t *wp = &wps->wp[offset / bs->bl.zone_size];
2546 if (!BDRV_ZT_IS_CONV(*wp)) {
Sam Li4751d092023-05-08 13:15:08 +08002547 if (type & QEMU_AIO_ZONE_APPEND) {
Naohiro Aotaad4feac2023-10-30 16:38:53 +09002548 *offset_ptr = *wp;
2549 trace_zbd_zone_append_complete(bs, *offset_ptr
Sam Li6c811e12023-05-08 13:15:10 +08002550 >> BDRV_SECTOR_BITS);
Sam Li4751d092023-05-08 13:15:08 +08002551 }
Sam Lia3c41f02023-05-08 13:15:07 +08002552 /* Advance the wp if needed */
2553 if (offset + bytes > *wp) {
2554 *wp = offset + bytes;
2555 }
2556 }
Hanna Czenczekd31b50a2023-08-24 17:53:43 +02002557 } else {
Sam Li10b9e082023-08-25 12:05:56 +08002558 /*
2559 * write and append write are not allowed to cross zone boundaries
2560 */
2561 update_zones_wp(bs, s->fd, offset, 1);
Sam Lia3c41f02023-05-08 13:15:07 +08002562 }
Sam Lia3c41f02023-05-08 13:15:07 +08002563
Sam Lia3c41f02023-05-08 13:15:07 +08002564 qemu_co_mutex_unlock(&wps->colock);
2565 }
Sam Lia3c41f02023-05-08 13:15:07 +08002566#endif
2567 return ret;
Kevin Wolf2174f122014-08-06 17:18:07 +02002568}
2569
Vladimir Sementsov-Ogievskiyf7ef38d2021-09-03 13:27:59 +03002570static int coroutine_fn raw_co_preadv(BlockDriverState *bs, int64_t offset,
2571 int64_t bytes, QEMUIOVector *qiov,
2572 BdrvRequestFlags flags)
Kevin Wolf2174f122014-08-06 17:18:07 +02002573{
Naohiro Aotaad4feac2023-10-30 16:38:53 +09002574 return raw_co_prw(bs, &offset, bytes, qiov, QEMU_AIO_READ);
Kevin Wolf2174f122014-08-06 17:18:07 +02002575}
2576
Vladimir Sementsov-Ogievskiye75abed2021-09-03 13:28:00 +03002577static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, int64_t offset,
2578 int64_t bytes, QEMUIOVector *qiov,
2579 BdrvRequestFlags flags)
Kevin Wolf2174f122014-08-06 17:18:07 +02002580{
Naohiro Aotaad4feac2023-10-30 16:38:53 +09002581 return raw_co_prw(bs, &offset, bytes, qiov, QEMU_AIO_WRITE);
bellard83f64092006-08-01 16:21:11 +00002582}
2583
Paolo Bonzinidda56b72022-09-22 10:49:06 +02002584static int coroutine_fn raw_co_flush_to_disk(BlockDriverState *bs)
Christoph Hellwigb2e12bc2009-09-04 19:01:49 +02002585{
2586 BDRVRawState *s = bs->opaque;
Kevin Wolf06dc9bd2018-10-25 14:18:58 +01002587 RawPosixAIOData acb;
Kevin Wolf33d70fb2018-06-21 19:07:32 +02002588 int ret;
Christoph Hellwigb2e12bc2009-09-04 19:01:49 +02002589
Kevin Wolf33d70fb2018-06-21 19:07:32 +02002590 ret = fd_open(bs);
2591 if (ret < 0) {
2592 return ret;
2593 }
Christoph Hellwigb2e12bc2009-09-04 19:01:49 +02002594
Kevin Wolf06dc9bd2018-10-25 14:18:58 +01002595 acb = (RawPosixAIOData) {
2596 .bs = bs,
2597 .aio_fildes = s->fd,
2598 .aio_type = QEMU_AIO_FLUSH,
2599 };
2600
Aarushi Mehtac6447512020-01-20 14:18:51 +00002601#ifdef CONFIG_LINUX_IO_URING
Stefan Hajnoczicd0c0db2023-09-14 10:00:58 -04002602 if (raw_check_linux_io_uring(s)) {
Emanuele Giuseppe Espositoa75e4e42023-02-03 08:17:29 -05002603 return luring_co_submit(bs, s->fd, 0, NULL, QEMU_AIO_FLUSH);
Aarushi Mehtac6447512020-01-20 14:18:51 +00002604 }
2605#endif
Prasad Pandit24687ab2024-04-25 12:34:12 +05302606#ifdef CONFIG_LINUX_AIO
2607 if (s->has_laio_fdsync && raw_check_linux_aio(s)) {
2608 return laio_co_submit(s->fd, 0, NULL, QEMU_AIO_FLUSH, 0);
2609 }
2610#endif
Emanuele Giuseppe Esposito0fdb7312023-02-03 08:17:30 -05002611 return raw_thread_pool_submit(handle_aiocb_flush, &acb);
Christoph Hellwigb2e12bc2009-09-04 19:01:49 +02002612}
2613
bellard83f64092006-08-01 16:21:11 +00002614static void raw_close(BlockDriverState *bs)
2615{
2616 BDRVRawState *s = bs->opaque;
Stefan Hajnoczic2f34262014-05-08 16:34:47 +02002617
bellard19cb3732006-08-19 11:45:59 +00002618 if (s->fd >= 0) {
Sam Lia3c41f02023-05-08 13:15:07 +08002619#if defined(CONFIG_BLKZONED)
2620 g_free(bs->wps);
2621#endif
Corey Bryant2e1e79d2012-08-14 16:43:46 -04002622 qemu_close(s->fd);
bellard19cb3732006-08-19 11:45:59 +00002623 s->fd = -1;
2624 }
bellard83f64092006-08-01 16:21:11 +00002625}
2626
Max Reitzd0bc9e52017-06-13 22:20:58 +02002627/**
2628 * Truncates the given regular file @fd to @offset and, when growing, fills the
2629 * new space according to @prealloc.
2630 *
2631 * Returns: 0 on success, -errno on failure.
2632 */
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002633static int coroutine_fn
2634raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset,
2635 PreallocMode prealloc, Error **errp)
Max Reitz9f63b072017-06-13 22:20:57 +02002636{
Kevin Wolf29cb4c02018-10-25 14:18:58 +01002637 RawPosixAIOData acb;
Max Reitzd0bc9e52017-06-13 22:20:58 +02002638
Kevin Wolf29cb4c02018-10-25 14:18:58 +01002639 acb = (RawPosixAIOData) {
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002640 .bs = bs,
2641 .aio_fildes = fd,
2642 .aio_type = QEMU_AIO_TRUNCATE,
2643 .aio_offset = offset,
Kevin Wolfd57c44d2018-10-25 16:21:14 +01002644 .truncate = {
2645 .prealloc = prealloc,
2646 .errp = errp,
2647 },
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002648 };
Max Reitzd0bc9e52017-06-13 22:20:58 +02002649
Emanuele Giuseppe Esposito0fdb7312023-02-03 08:17:30 -05002650 return raw_thread_pool_submit(handle_aiocb_truncate, &acb);
Max Reitz9f63b072017-06-13 22:20:57 +02002651}
2652
Kevin Wolf061ca8a2018-06-21 17:54:35 +02002653static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset,
Max Reitzc80d8b02019-09-18 11:51:40 +02002654 bool exact, PreallocMode prealloc,
Kevin Wolf92b92792020-04-24 14:54:39 +02002655 BdrvRequestFlags flags, Error **errp)
bellard83f64092006-08-01 16:21:11 +00002656{
2657 BDRVRawState *s = bs->opaque;
Christoph Hellwig55b949c2011-09-21 01:10:37 +02002658 struct stat st;
Max Reitzf59adb32017-03-28 22:51:29 +02002659 int ret;
Christoph Hellwig55b949c2011-09-21 01:10:37 +02002660
2661 if (fstat(s->fd, &st)) {
Max Reitzf59adb32017-03-28 22:51:29 +02002662 ret = -errno;
2663 error_setg_errno(errp, -ret, "Failed to fstat() the file");
2664 return ret;
Christoph Hellwig55b949c2011-09-21 01:10:37 +02002665 }
2666
2667 if (S_ISREG(st.st_mode)) {
Max Reitz82325ae2019-09-18 11:51:41 +02002668 /* Always resizes to the exact @offset */
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002669 return raw_regular_truncate(bs, s->fd, offset, prealloc, errp);
Max Reitz35d72602017-06-13 22:20:59 +02002670 }
2671
2672 if (prealloc != PREALLOC_MODE_OFF) {
2673 error_setg(errp, "Preallocation mode '%s' unsupported for this "
Markus Armbruster977c7362017-08-24 10:46:08 +02002674 "non-regular file", PreallocMode_str(prealloc));
Max Reitz35d72602017-06-13 22:20:59 +02002675 return -ENOTSUP;
2676 }
2677
2678 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
Paolo Bonzini36c6c872023-06-01 13:51:34 +02002679 int64_t cur_length = raw_getlength(bs);
Max Reitz82325ae2019-09-18 11:51:41 +02002680
2681 if (offset != cur_length && exact) {
2682 error_setg(errp, "Cannot resize device files");
2683 return -ENOTSUP;
2684 } else if (offset > cur_length) {
Max Reitzf59adb32017-03-28 22:51:29 +02002685 error_setg(errp, "Cannot grow device files");
2686 return -EINVAL;
2687 }
Christoph Hellwig55b949c2011-09-21 01:10:37 +02002688 } else {
Max Reitzf59adb32017-03-28 22:51:29 +02002689 error_setg(errp, "Resizing this file is not supported");
Christoph Hellwig55b949c2011-09-21 01:10:37 +02002690 return -ENOTSUP;
2691 }
2692
bellard83f64092006-08-01 16:21:11 +00002693 return 0;
2694}
2695
blueswir1128ab2f2008-08-15 18:33:42 +00002696#ifdef __OpenBSD__
Paolo Bonzini36c6c872023-06-01 13:51:34 +02002697static int64_t raw_getlength(BlockDriverState *bs)
blueswir1128ab2f2008-08-15 18:33:42 +00002698{
2699 BDRVRawState *s = bs->opaque;
2700 int fd = s->fd;
2701 struct stat st;
2702
2703 if (fstat(fd, &st))
Markus Armbrusteraa729702014-06-26 13:23:16 +02002704 return -errno;
blueswir1128ab2f2008-08-15 18:33:42 +00002705 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2706 struct disklabel dl;
2707
2708 if (ioctl(fd, DIOCGDINFO, &dl))
Markus Armbrusteraa729702014-06-26 13:23:16 +02002709 return -errno;
blueswir1128ab2f2008-08-15 18:33:42 +00002710 return (uint64_t)dl.d_secsize *
2711 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
2712 } else
2713 return st.st_size;
2714}
Christoph Eggerd1f6fd82011-05-23 14:31:17 +02002715#elif defined(__NetBSD__)
Paolo Bonzini36c6c872023-06-01 13:51:34 +02002716static int64_t raw_getlength(BlockDriverState *bs)
Christoph Eggerd1f6fd82011-05-23 14:31:17 +02002717{
2718 BDRVRawState *s = bs->opaque;
2719 int fd = s->fd;
2720 struct stat st;
2721
2722 if (fstat(fd, &st))
Markus Armbrusteraa729702014-06-26 13:23:16 +02002723 return -errno;
Christoph Eggerd1f6fd82011-05-23 14:31:17 +02002724 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
2725 struct dkwedge_info dkw;
2726
2727 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
2728 return dkw.dkw_size * 512;
2729 } else {
2730 struct disklabel dl;
2731
2732 if (ioctl(fd, DIOCGDINFO, &dl))
Markus Armbrusteraa729702014-06-26 13:23:16 +02002733 return -errno;
Christoph Eggerd1f6fd82011-05-23 14:31:17 +02002734 return (uint64_t)dl.d_secsize *
2735 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
2736 }
2737 } else
2738 return st.st_size;
2739}
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002740#elif defined(__sun__)
Paolo Bonzini36c6c872023-06-01 13:51:34 +02002741static int64_t raw_getlength(BlockDriverState *bs)
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002742{
2743 BDRVRawState *s = bs->opaque;
2744 struct dk_minfo minfo;
2745 int ret;
Markus Armbrusteraa729702014-06-26 13:23:16 +02002746 int64_t size;
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002747
2748 ret = fd_open(bs);
2749 if (ret < 0) {
2750 return ret;
2751 }
2752
2753 /*
2754 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2755 */
2756 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
2757 if (ret != -1) {
2758 return minfo.dki_lbsize * minfo.dki_capacity;
2759 }
2760
2761 /*
2762 * There are reports that lseek on some devices fails, but
2763 * irc discussion said that contingency on contingency was overkill.
2764 */
Markus Armbrusteraa729702014-06-26 13:23:16 +02002765 size = lseek(s->fd, 0, SEEK_END);
2766 if (size < 0) {
2767 return -errno;
2768 }
2769 return size;
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002770}
2771#elif defined(CONFIG_BSD)
Paolo Bonzini36c6c872023-06-01 13:51:34 +02002772static int64_t raw_getlength(BlockDriverState *bs)
bellard83f64092006-08-01 16:21:11 +00002773{
2774 BDRVRawState *s = bs->opaque;
2775 int fd = s->fd;
2776 int64_t size;
bellard83f64092006-08-01 16:21:11 +00002777 struct stat sb;
Aurelien Jarnoa167ba52009-11-29 18:00:41 +01002778#if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
blueswir19f230112009-03-28 08:37:13 +00002779 int reopened = 0;
2780#endif
bellard19cb3732006-08-19 11:45:59 +00002781 int ret;
2782
2783 ret = fd_open(bs);
2784 if (ret < 0)
2785 return ret;
bellard83f64092006-08-01 16:21:11 +00002786
Aurelien Jarnoa167ba52009-11-29 18:00:41 +01002787#if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
blueswir19f230112009-03-28 08:37:13 +00002788again:
2789#endif
bellard83f64092006-08-01 16:21:11 +00002790 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
Paolo Bonzini267cd532021-06-16 15:32:04 +02002791 size = 0;
bellard83f64092006-08-01 16:21:11 +00002792#ifdef DIOCGMEDIASIZE
Paolo Bonzini267cd532021-06-16 15:32:04 +02002793 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) {
2794 size = 0;
blueswir1c5e97232009-03-07 20:06:23 +00002795 }
Paolo Bonzini267cd532021-06-16 15:32:04 +02002796#endif
2797#ifdef DIOCGPART
2798 if (size == 0) {
2799 struct partinfo pi;
2800 if (ioctl(fd, DIOCGPART, &pi) == 0) {
2801 size = pi.media_size;
2802 }
2803 }
bellard83f64092006-08-01 16:21:11 +00002804#endif
Joelle van Dyne09e20ab2021-03-15 11:03:40 -07002805#if defined(DKIOCGETBLOCKCOUNT) && defined(DKIOCGETBLOCKSIZE)
Paolo Bonzini267cd532021-06-16 15:32:04 +02002806 if (size == 0) {
Programmingkid728dacb2015-01-19 17:12:55 -05002807 uint64_t sectors = 0;
2808 uint32_t sector_size = 0;
2809
2810 if (ioctl(fd, DKIOCGETBLOCKCOUNT, &sectors) == 0
2811 && ioctl(fd, DKIOCGETBLOCKSIZE, &sector_size) == 0) {
2812 size = sectors * sector_size;
Programmingkid728dacb2015-01-19 17:12:55 -05002813 }
2814 }
Paolo Bonzini267cd532021-06-16 15:32:04 +02002815#endif
2816 if (size == 0) {
2817 size = lseek(fd, 0LL, SEEK_END);
2818 }
Markus Armbrusteraa729702014-06-26 13:23:16 +02002819 if (size < 0) {
2820 return -errno;
2821 }
Aurelien Jarnoa167ba52009-11-29 18:00:41 +01002822#if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
blueswir19f230112009-03-28 08:37:13 +00002823 switch(s->type) {
2824 case FTYPE_CD:
2825 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2826 if (size == 2048LL * (unsigned)-1)
2827 size = 0;
2828 /* XXX no disc? maybe we need to reopen... */
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02002829 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
blueswir19f230112009-03-28 08:37:13 +00002830 reopened = 1;
2831 goto again;
2832 }
2833 }
2834#endif
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002835 } else {
bellard83f64092006-08-01 16:21:11 +00002836 size = lseek(fd, 0, SEEK_END);
Markus Armbrusteraa729702014-06-26 13:23:16 +02002837 if (size < 0) {
2838 return -errno;
2839 }
bellard83f64092006-08-01 16:21:11 +00002840 }
bellard83f64092006-08-01 16:21:11 +00002841 return size;
2842}
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002843#else
Paolo Bonzini36c6c872023-06-01 13:51:34 +02002844static int64_t raw_getlength(BlockDriverState *bs)
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002845{
2846 BDRVRawState *s = bs->opaque;
2847 int ret;
Markus Armbrusteraa729702014-06-26 13:23:16 +02002848 int64_t size;
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002849
2850 ret = fd_open(bs);
2851 if (ret < 0) {
2852 return ret;
2853 }
2854
Markus Armbrusteraa729702014-06-26 13:23:16 +02002855 size = lseek(s->fd, 0, SEEK_END);
2856 if (size < 0) {
2857 return -errno;
2858 }
2859 return size;
Christoph Hellwig50779cc2010-04-06 19:13:44 +02002860}
blueswir1128ab2f2008-08-15 18:33:42 +00002861#endif
bellard83f64092006-08-01 16:21:11 +00002862
Paolo Bonzini36c6c872023-06-01 13:51:34 +02002863static int64_t coroutine_fn raw_co_getlength(BlockDriverState *bs)
2864{
2865 return raw_getlength(bs);
2866}
2867
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01002868static int64_t coroutine_fn raw_co_get_allocated_file_size(BlockDriverState *bs)
Fam Zheng4a1d5e12011-07-12 19:56:39 +08002869{
2870 struct stat st;
2871 BDRVRawState *s = bs->opaque;
2872
2873 if (fstat(s->fd, &st) < 0) {
2874 return -errno;
2875 }
2876 return (int64_t)st.st_blocks * 512;
2877}
2878
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002879static int coroutine_fn
2880raw_co_create(BlockdevCreateOptions *options, Error **errp)
bellard83f64092006-08-01 16:21:11 +00002881{
Kevin Wolf927f11e2018-01-16 16:04:21 +01002882 BlockdevCreateOptionsFile *file_opts;
Max Reitz7c20c802018-07-04 16:47:51 +02002883 Error *local_err = NULL;
bellard83f64092006-08-01 16:21:11 +00002884 int fd;
Max Reitzd815efc2018-07-04 16:47:50 +02002885 uint64_t perm, shared;
Stefan Weil1e37d052009-07-11 16:43:37 +02002886 int result = 0;
bellard83f64092006-08-01 16:21:11 +00002887
Kevin Wolf927f11e2018-01-16 16:04:21 +01002888 /* Validate options and set default values */
2889 assert(options->driver == BLOCKDEV_DRIVER_FILE);
2890 file_opts = &options->u.file;
Max Reitz464d9f62014-03-05 22:41:38 +01002891
Kevin Wolf927f11e2018-01-16 16:04:21 +01002892 if (!file_opts->has_nocow) {
2893 file_opts->nocow = false;
2894 }
2895 if (!file_opts->has_preallocation) {
2896 file_opts->preallocation = PREALLOC_MODE_OFF;
Hu Tao06247422014-09-10 17:05:48 +08002897 }
Kevin Wolfffa244c2020-07-07 16:23:29 +02002898 if (!file_opts->has_extent_size_hint) {
2899 file_opts->extent_size_hint = 1 * MiB;
2900 }
2901 if (file_opts->extent_size_hint > UINT32_MAX) {
2902 result = -EINVAL;
2903 error_setg(errp, "Extent size hint is too large");
2904 goto out;
2905 }
bellard83f64092006-08-01 16:21:11 +00002906
Kevin Wolf927f11e2018-01-16 16:04:21 +01002907 /* Create file */
Daniel P. Berrangéb18a24a2020-07-01 15:22:43 +01002908 fd = qemu_create(file_opts->filename, O_RDWR | O_BINARY, 0644, errp);
Stefan Weil1e37d052009-07-11 16:43:37 +02002909 if (fd < 0) {
2910 result = -errno;
Hu Tao06247422014-09-10 17:05:48 +08002911 goto out;
Stefan Weil1e37d052009-07-11 16:43:37 +02002912 }
Hu Tao06247422014-09-10 17:05:48 +08002913
Max Reitzb8cf1912018-05-09 23:53:35 +02002914 /* Take permissions: We want to discard everything, so we need
2915 * BLK_PERM_WRITE; and truncation to the desired size requires
2916 * BLK_PERM_RESIZE.
2917 * On the other hand, we cannot share the RESIZE permission
2918 * because we promise that after this function, the file has the
2919 * size given in the options. If someone else were to resize it
2920 * concurrently, we could not guarantee that.
2921 * Note that after this function, we can no longer guarantee that
2922 * the file is not touched by a third party, so it may be resized
2923 * then. */
2924 perm = BLK_PERM_WRITE | BLK_PERM_RESIZE;
2925 shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE;
2926
2927 /* Step one: Take locks */
Fam Zheng2996ffa2018-10-11 15:21:33 +08002928 result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp);
Max Reitzb8cf1912018-05-09 23:53:35 +02002929 if (result < 0) {
2930 goto out_close;
2931 }
2932
2933 /* Step two: Check that nobody else has taken conflicting locks */
2934 result = raw_check_lock_bytes(fd, perm, shared, errp);
2935 if (result < 0) {
Fam Zhengb8574312018-09-25 13:05:01 +08002936 error_append_hint(errp,
2937 "Is another process using the image [%s]?\n",
2938 file_opts->filename);
Max Reitz7c20c802018-07-04 16:47:51 +02002939 goto out_unlock;
Max Reitzb8cf1912018-05-09 23:53:35 +02002940 }
2941
2942 /* Clear the file by truncating it to 0 */
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002943 result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp);
Max Reitzb8cf1912018-05-09 23:53:35 +02002944 if (result < 0) {
Max Reitz7c20c802018-07-04 16:47:51 +02002945 goto out_unlock;
Max Reitzb8cf1912018-05-09 23:53:35 +02002946 }
2947
Kevin Wolf927f11e2018-01-16 16:04:21 +01002948 if (file_opts->nocow) {
Hu Tao06247422014-09-10 17:05:48 +08002949#ifdef __linux__
2950 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2951 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2952 * will be ignored since any failure of this operation should not
2953 * block the left work.
2954 */
2955 int attr;
2956 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
2957 attr |= FS_NOCOW_FL;
2958 ioctl(fd, FS_IOC_SETFLAGS, &attr);
2959 }
2960#endif
2961 }
Kevin Wolfffa244c2020-07-07 16:23:29 +02002962#ifdef FS_IOC_FSSETXATTR
2963 /*
2964 * Try to set the extent size hint. Failure is not fatal, and a warning is
2965 * only printed if the option was explicitly specified.
2966 */
2967 {
2968 struct fsxattr attr;
2969 result = ioctl(fd, FS_IOC_FSGETXATTR, &attr);
2970 if (result == 0) {
2971 attr.fsx_xflags |= FS_XFLAG_EXTSIZE;
2972 attr.fsx_extsize = file_opts->extent_size_hint;
2973 result = ioctl(fd, FS_IOC_FSSETXATTR, &attr);
2974 }
2975 if (result < 0 && file_opts->has_extent_size_hint &&
2976 file_opts->extent_size_hint)
2977 {
2978 warn_report("Failed to set extent size hint: %s",
2979 strerror(errno));
2980 }
2981 }
2982#endif
Hu Tao06247422014-09-10 17:05:48 +08002983
Max Reitzb8cf1912018-05-09 23:53:35 +02002984 /* Resize and potentially preallocate the file to the desired
2985 * final size */
Kevin Wolf93f4e2f2018-06-21 18:23:16 +02002986 result = raw_regular_truncate(NULL, fd, file_opts->size,
2987 file_opts->preallocation, errp);
Max Reitz9f63b072017-06-13 22:20:57 +02002988 if (result < 0) {
Max Reitz7c20c802018-07-04 16:47:51 +02002989 goto out_unlock;
2990 }
2991
2992out_unlock:
Fam Zheng2996ffa2018-10-11 15:21:33 +08002993 raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err);
Max Reitz7c20c802018-07-04 16:47:51 +02002994 if (local_err) {
2995 /* The above call should not fail, and if it does, that does
2996 * not mean the whole creation operation has failed. So
2997 * report it the user for their convenience, but do not report
2998 * it to the caller. */
Fam Zhengdb0754d2018-11-01 14:29:09 +08002999 warn_report_err(local_err);
Hu Tao06247422014-09-10 17:05:48 +08003000 }
3001
Nir Soffer5a1dad92017-02-17 02:51:26 +02003002out_close:
Hu Tao06247422014-09-10 17:05:48 +08003003 if (qemu_close(fd) != 0 && result == 0) {
3004 result = -errno;
3005 error_setg_errno(errp, -result, "Could not close the new file");
3006 }
3007out:
Stefan Weil1e37d052009-07-11 16:43:37 +02003008 return result;
bellard83f64092006-08-01 16:21:11 +00003009}
3010
Kevin Wolf4ec8df02023-02-03 16:21:55 +01003011static int coroutine_fn GRAPH_RDLOCK
3012raw_co_create_opts(BlockDriver *drv, const char *filename,
3013 QemuOpts *opts, Error **errp)
Kevin Wolf927f11e2018-01-16 16:04:21 +01003014{
3015 BlockdevCreateOptions options;
3016 int64_t total_size = 0;
Kevin Wolfffa244c2020-07-07 16:23:29 +02003017 int64_t extent_size_hint = 0;
3018 bool has_extent_size_hint = false;
Kevin Wolf927f11e2018-01-16 16:04:21 +01003019 bool nocow = false;
3020 PreallocMode prealloc;
3021 char *buf = NULL;
3022 Error *local_err = NULL;
3023
3024 /* Skip file: protocol prefix */
3025 strstart(filename, "file:", &filename);
3026
3027 /* Read out options */
3028 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
3029 BDRV_SECTOR_SIZE);
Kevin Wolfffa244c2020-07-07 16:23:29 +02003030 if (qemu_opt_get(opts, BLOCK_OPT_EXTENT_SIZE_HINT)) {
3031 has_extent_size_hint = true;
3032 extent_size_hint =
3033 qemu_opt_get_size_del(opts, BLOCK_OPT_EXTENT_SIZE_HINT, -1);
3034 }
Kevin Wolf927f11e2018-01-16 16:04:21 +01003035 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
3036 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
3037 prealloc = qapi_enum_parse(&PreallocMode_lookup, buf,
3038 PREALLOC_MODE_OFF, &local_err);
3039 g_free(buf);
3040 if (local_err) {
3041 error_propagate(errp, local_err);
3042 return -EINVAL;
3043 }
3044
3045 options = (BlockdevCreateOptions) {
3046 .driver = BLOCKDEV_DRIVER_FILE,
3047 .u.file = {
3048 .filename = (char *) filename,
3049 .size = total_size,
3050 .has_preallocation = true,
3051 .preallocation = prealloc,
3052 .has_nocow = true,
3053 .nocow = nocow,
Kevin Wolfffa244c2020-07-07 16:23:29 +02003054 .has_extent_size_hint = has_extent_size_hint,
3055 .extent_size_hint = extent_size_hint,
Kevin Wolf927f11e2018-01-16 16:04:21 +01003056 },
3057 };
3058 return raw_co_create(&options, errp);
3059}
3060
Daniel Henrique Barboza9bffae12020-01-30 18:39:04 -03003061static int coroutine_fn raw_co_delete_file(BlockDriverState *bs,
3062 Error **errp)
3063{
3064 struct stat st;
3065 int ret;
3066
3067 if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) {
3068 error_setg_errno(errp, ENOENT, "%s is not a regular file",
3069 bs->filename);
3070 return -ENOENT;
3071 }
3072
3073 ret = unlink(bs->filename);
3074 if (ret < 0) {
3075 ret = -errno;
3076 error_setg_errno(errp, -ret, "Error when deleting file %s",
3077 bs->filename);
3078 }
3079
3080 return ret;
3081}
3082
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01003083/*
3084 * Find allocation range in @bs around offset @start.
3085 * May change underlying file descriptor's file offset.
3086 * If @start is not in a hole, store @start in @data, and the
3087 * beginning of the next hole in @hole, and return 0.
3088 * If @start is in a non-trailing hole, store @start in @hole and the
3089 * beginning of the next non-hole in @data, and return 0.
3090 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
3091 * If we can't find out, return a negative errno other than -ENXIO.
3092 */
3093static int find_allocation(BlockDriverState *bs, off_t start,
3094 off_t *data, off_t *hole)
Max Reitz4f11aa82014-05-08 20:57:55 +02003095{
3096#if defined SEEK_HOLE && defined SEEK_DATA
3097 BDRVRawState *s = bs->opaque;
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01003098 off_t offs;
Max Reitz4f11aa82014-05-08 20:57:55 +02003099
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01003100 /*
3101 * SEEK_DATA cases:
3102 * D1. offs == start: start is in data
3103 * D2. offs > start: start is in a hole, next data at offs
3104 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
3105 * or start is beyond EOF
3106 * If the latter happens, the file has been truncated behind
3107 * our back since we opened it. All bets are off then.
3108 * Treating like a trailing hole is simplest.
3109 * D4. offs < 0, errno != ENXIO: we learned nothing
3110 */
3111 offs = lseek(s->fd, start, SEEK_DATA);
3112 if (offs < 0) {
3113 return -errno; /* D3 or D4 */
3114 }
Jeff Codya03083a2018-04-03 00:37:26 -04003115
3116 if (offs < start) {
3117 /* This is not a valid return by lseek(). We are safe to just return
3118 * -EIO in this case, and we'll treat it like D4. */
3119 return -EIO;
3120 }
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01003121
3122 if (offs > start) {
3123 /* D2: in hole, next data at offs */
3124 *hole = start;
3125 *data = offs;
3126 return 0;
Max Reitz4f11aa82014-05-08 20:57:55 +02003127 }
3128
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01003129 /* D1: in data, end not yet known */
3130
3131 /*
3132 * SEEK_HOLE cases:
3133 * H1. offs == start: start is in a hole
3134 * If this happens here, a hole has been dug behind our back
3135 * since the previous lseek().
3136 * H2. offs > start: either start is in data, next hole at offs,
3137 * or start is in trailing hole, EOF at offs
3138 * Linux treats trailing holes like any other hole: offs ==
3139 * start. Solaris seeks to EOF instead: offs > start (blech).
3140 * If that happens here, a hole has been dug behind our back
3141 * since the previous lseek().
3142 * H3. offs < 0, errno = ENXIO: start is beyond EOF
3143 * If this happens, the file has been truncated behind our
3144 * back since we opened it. Treat it like a trailing hole.
3145 * H4. offs < 0, errno != ENXIO: we learned nothing
3146 * Pretend we know nothing at all, i.e. "forget" about D1.
3147 */
3148 offs = lseek(s->fd, start, SEEK_HOLE);
3149 if (offs < 0) {
3150 return -errno; /* D1 and (H3 or H4) */
3151 }
Jeff Codya03083a2018-04-03 00:37:26 -04003152
3153 if (offs < start) {
3154 /* This is not a valid return by lseek(). We are safe to just return
3155 * -EIO in this case, and we'll treat it like H4. */
3156 return -EIO;
3157 }
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01003158
3159 if (offs > start) {
3160 /*
3161 * D1 and H2: either in data, next hole at offs, or it was in
3162 * data but is now in a trailing hole. In the latter case,
3163 * all bets are off. Treating it as if it there was data all
3164 * the way to EOF is safe, so simply do that.
3165 */
Max Reitz4f11aa82014-05-08 20:57:55 +02003166 *data = start;
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01003167 *hole = offs;
3168 return 0;
Max Reitz4f11aa82014-05-08 20:57:55 +02003169 }
3170
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01003171 /* D1 and H1 */
3172 return -EBUSY;
Max Reitz4f11aa82014-05-08 20:57:55 +02003173#else
3174 return -ENOTSUP;
3175#endif
3176}
3177
Paolo Bonzini55003162012-05-09 16:49:58 +02003178/*
Eric Blakea290f082018-02-13 14:26:44 -06003179 * Returns the allocation status of the specified offset.
Paolo Bonzini55003162012-05-09 16:49:58 +02003180 *
Eric Blakea290f082018-02-13 14:26:44 -06003181 * The block layer guarantees 'offset' and 'bytes' are within bounds.
Paolo Bonzini55003162012-05-09 16:49:58 +02003182 *
Eric Blakea290f082018-02-13 14:26:44 -06003183 * 'pnum' is set to the number of bytes (including and immediately following
3184 * the specified offset) that are known to be in the same
Paolo Bonzini55003162012-05-09 16:49:58 +02003185 * allocated/unallocated state.
3186 *
Hanna Reitz869e7ee2021-08-12 10:41:46 +02003187 * 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may
3188 * well exceed it.
Paolo Bonzini55003162012-05-09 16:49:58 +02003189 */
Eric Blakea290f082018-02-13 14:26:44 -06003190static int coroutine_fn raw_co_block_status(BlockDriverState *bs,
3191 bool want_zero,
3192 int64_t offset,
3193 int64_t bytes, int64_t *pnum,
3194 int64_t *map,
3195 BlockDriverState **file)
Paolo Bonzini55003162012-05-09 16:49:58 +02003196{
Eric Blakea290f082018-02-13 14:26:44 -06003197 off_t data = 0, hole = 0;
Max Reitzd7f62752014-10-24 12:57:59 +02003198 int ret;
Paolo Bonzini55003162012-05-09 16:49:58 +02003199
Max Reitz9c3db312019-05-15 06:15:40 +02003200 assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment));
3201
Paolo Bonzini55003162012-05-09 16:49:58 +02003202 ret = fd_open(bs);
3203 if (ret < 0) {
3204 return ret;
3205 }
3206
Eric Blakea290f082018-02-13 14:26:44 -06003207 if (!want_zero) {
3208 *pnum = bytes;
3209 *map = offset;
3210 *file = bs;
3211 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
Max Reitze6d7ec32014-10-24 12:57:58 +02003212 }
Kevin Wolf94282e72012-06-19 22:02:51 +00003213
Eric Blakea290f082018-02-13 14:26:44 -06003214 ret = find_allocation(bs, offset, &data, &hole);
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01003215 if (ret == -ENXIO) {
3216 /* Trailing hole */
Eric Blakea290f082018-02-13 14:26:44 -06003217 *pnum = bytes;
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01003218 ret = BDRV_BLOCK_ZERO;
3219 } else if (ret < 0) {
3220 /* No info available, so pretend there are no holes */
Eric Blakea290f082018-02-13 14:26:44 -06003221 *pnum = bytes;
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01003222 ret = BDRV_BLOCK_DATA;
Eric Blakea290f082018-02-13 14:26:44 -06003223 } else if (data == offset) {
3224 /* On a data extent, compute bytes to the end of the extent,
Kevin Wolff4a769a2015-06-09 10:55:08 +02003225 * possibly including a partial sector at EOF. */
Hanna Reitz869e7ee2021-08-12 10:41:46 +02003226 *pnum = hole - offset;
Max Reitz9c3db312019-05-15 06:15:40 +02003227
3228 /*
3229 * We are not allowed to return partial sectors, though, so
3230 * round up if necessary.
3231 */
3232 if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) {
Paolo Bonzini36c6c872023-06-01 13:51:34 +02003233 int64_t file_length = raw_getlength(bs);
Max Reitz9c3db312019-05-15 06:15:40 +02003234 if (file_length > 0) {
3235 /* Ignore errors, this is just a safeguard */
3236 assert(hole == file_length);
3237 }
3238 *pnum = ROUND_UP(*pnum, bs->bl.request_alignment);
3239 }
3240
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01003241 ret = BDRV_BLOCK_DATA;
Paolo Bonzini55003162012-05-09 16:49:58 +02003242 } else {
Eric Blakea290f082018-02-13 14:26:44 -06003243 /* On a hole, compute bytes to the beginning of the next extent. */
3244 assert(hole == offset);
Hanna Reitz869e7ee2021-08-12 10:41:46 +02003245 *pnum = data - offset;
Markus Armbrusterd1f06fe2014-11-17 11:18:34 +01003246 ret = BDRV_BLOCK_ZERO;
Paolo Bonzini55003162012-05-09 16:49:58 +02003247 }
Eric Blakea290f082018-02-13 14:26:44 -06003248 *map = offset;
Fam Zheng02650ac2016-01-26 11:58:51 +08003249 *file = bs;
Eric Blakea290f082018-02-13 14:26:44 -06003250 return ret | BDRV_BLOCK_OFFSET_VALID;
Paolo Bonzini55003162012-05-09 16:49:58 +02003251}
3252
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +01003253#if defined(__linux__)
3254/* Verify that the file is not in the page cache */
Paolo Bonzini36c6c872023-06-01 13:51:34 +02003255static void check_cache_dropped(BlockDriverState *bs, Error **errp)
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +01003256{
3257 const size_t window_size = 128 * 1024 * 1024;
3258 BDRVRawState *s = bs->opaque;
3259 void *window = NULL;
3260 size_t length = 0;
3261 unsigned char *vec;
3262 size_t page_size;
3263 off_t offset;
3264 off_t end;
3265
3266 /* mincore(2) page status information requires 1 byte per page */
3267 page_size = sysconf(_SC_PAGESIZE);
3268 vec = g_malloc(DIV_ROUND_UP(window_size, page_size));
3269
Paolo Bonzini36c6c872023-06-01 13:51:34 +02003270 end = raw_getlength(bs);
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +01003271
3272 for (offset = 0; offset < end; offset += window_size) {
3273 void *new_window;
3274 size_t new_length;
3275 size_t vec_end;
3276 size_t i;
3277 int ret;
3278
3279 /* Unmap previous window if size has changed */
3280 new_length = MIN(end - offset, window_size);
3281 if (new_length != length) {
3282 munmap(window, length);
3283 window = NULL;
3284 length = 0;
3285 }
3286
3287 new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE,
3288 s->fd, offset);
3289 if (new_window == MAP_FAILED) {
3290 error_setg_errno(errp, errno, "mmap failed");
3291 break;
3292 }
3293
3294 window = new_window;
3295 length = new_length;
3296
3297 ret = mincore(window, length, vec);
3298 if (ret < 0) {
3299 error_setg_errno(errp, errno, "mincore failed");
3300 break;
3301 }
3302
3303 vec_end = DIV_ROUND_UP(length, page_size);
3304 for (i = 0; i < vec_end; i++) {
3305 if (vec[i] & 0x1) {
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +01003306 break;
3307 }
3308 }
Markus Armbruster77ed9712020-04-22 15:07:07 +02003309 if (i < vec_end) {
3310 error_setg(errp, "page cache still in use!");
3311 break;
3312 }
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +01003313 }
3314
3315 if (window) {
3316 munmap(window, length);
3317 }
3318
3319 g_free(vec);
3320}
3321#endif /* __linux__ */
3322
Emanuele Giuseppe Esposito88095342023-02-03 16:21:46 +01003323static void coroutine_fn GRAPH_RDLOCK
3324raw_co_invalidate_cache(BlockDriverState *bs, Error **errp)
Stefan Hajnoczidd577a22018-04-27 17:23:11 +01003325{
3326 BDRVRawState *s = bs->opaque;
3327 int ret;
3328
3329 ret = fd_open(bs);
3330 if (ret < 0) {
3331 error_setg_errno(errp, -ret, "The file descriptor is not open");
3332 return;
3333 }
3334
Stefan Hajnoczif357fcd2019-03-07 16:49:41 +00003335 if (!s->drop_cache) {
3336 return;
3337 }
3338
Stefan Hajnoczidd577a22018-04-27 17:23:11 +01003339 if (s->open_flags & O_DIRECT) {
3340 return; /* No host kernel page cache */
3341 }
3342
3343#if defined(__linux__)
3344 /* This sets the scene for the next syscall... */
3345 ret = bdrv_co_flush(bs);
3346 if (ret < 0) {
3347 error_setg_errno(errp, -ret, "flush failed");
3348 return;
3349 }
3350
3351 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
3352 * process. These limitations are okay because we just fsynced the file,
3353 * we don't use mmap, and the file should not be in use by other processes.
3354 */
3355 ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED);
3356 if (ret != 0) { /* the return value is a positive errno */
3357 error_setg_errno(errp, ret, "fadvise failed");
3358 return;
3359 }
Stefan Hajnoczi31be8a22018-04-27 17:23:12 +01003360
3361 if (s->check_cache_dropped) {
3362 check_cache_dropped(bs, errp);
3363 }
Stefan Hajnoczidd577a22018-04-27 17:23:11 +01003364#else /* __linux__ */
3365 /* Do nothing. Live migration to a remote host with cache.direct=off is
3366 * unsupported on other host operating systems. Cache consistency issues
3367 * may occur but no error is reported here, partly because that's the
3368 * historical behavior and partly because it's hard to differentiate valid
3369 * configurations that should not cause errors.
3370 */
3371#endif /* !__linux__ */
3372}
3373
Anton Nefedov1c450362019-09-23 15:17:36 +03003374static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret)
3375{
3376 if (ret) {
3377 s->stats.discard_nb_failed++;
3378 } else {
3379 s->stats.discard_nb_ok++;
3380 s->stats.discard_bytes_ok += nbytes;
3381 }
3382}
3383
Sam Li6d43eaa2023-05-08 12:55:28 +08003384/*
3385 * zone report - Get a zone block device's information in the form
3386 * of an array of zone descriptors.
3387 * zones is an array of zone descriptors to hold zone information on reply;
3388 * offset can be any byte within the entire size of the device;
Michael Tokarev3202d8e2023-07-14 14:06:05 +03003389 * nr_zones is the maximum number of sectors the command should operate on.
Sam Li6d43eaa2023-05-08 12:55:28 +08003390 */
3391#if defined(CONFIG_BLKZONED)
3392static int coroutine_fn raw_co_zone_report(BlockDriverState *bs, int64_t offset,
3393 unsigned int *nr_zones,
3394 BlockZoneDescriptor *zones) {
3395 BDRVRawState *s = bs->opaque;
3396 RawPosixAIOData acb = (RawPosixAIOData) {
3397 .bs = bs,
3398 .aio_fildes = s->fd,
3399 .aio_type = QEMU_AIO_ZONE_REPORT,
3400 .aio_offset = offset,
3401 .zone_report = {
3402 .nr_zones = nr_zones,
3403 .zones = zones,
3404 },
3405 };
3406
Sam Li142e3072023-05-08 12:55:32 +08003407 trace_zbd_zone_report(bs, *nr_zones, offset >> BDRV_SECTOR_BITS);
Sam Li6d43eaa2023-05-08 12:55:28 +08003408 return raw_thread_pool_submit(handle_aiocb_zone_report, &acb);
3409}
3410#endif
3411
3412/*
3413 * zone management operations - Execute an operation on a zone
3414 */
3415#if defined(CONFIG_BLKZONED)
3416static int coroutine_fn raw_co_zone_mgmt(BlockDriverState *bs, BlockZoneOp op,
3417 int64_t offset, int64_t len) {
3418 BDRVRawState *s = bs->opaque;
3419 RawPosixAIOData acb;
3420 int64_t zone_size, zone_size_mask;
3421 const char *op_name;
3422 unsigned long zo;
3423 int ret;
Sam Lia3c41f02023-05-08 13:15:07 +08003424 BlockZoneWps *wps = bs->wps;
Sam Li6d43eaa2023-05-08 12:55:28 +08003425 int64_t capacity = bs->total_sectors << BDRV_SECTOR_BITS;
3426
3427 zone_size = bs->bl.zone_size;
3428 zone_size_mask = zone_size - 1;
3429 if (offset & zone_size_mask) {
3430 error_report("sector offset %" PRId64 " is not aligned to zone size "
3431 "%" PRId64 "", offset / 512, zone_size / 512);
3432 return -EINVAL;
3433 }
3434
3435 if (((offset + len) < capacity && len & zone_size_mask) ||
3436 offset + len > capacity) {
3437 error_report("number of sectors %" PRId64 " is not aligned to zone size"
3438 " %" PRId64 "", len / 512, zone_size / 512);
3439 return -EINVAL;
3440 }
3441
Sam Lia3c41f02023-05-08 13:15:07 +08003442 uint32_t i = offset / bs->bl.zone_size;
3443 uint32_t nrz = len / bs->bl.zone_size;
3444 uint64_t *wp = &wps->wp[i];
3445 if (BDRV_ZT_IS_CONV(*wp) && len != capacity) {
3446 error_report("zone mgmt operations are not allowed for conventional zones");
3447 return -EIO;
3448 }
3449
Sam Li6d43eaa2023-05-08 12:55:28 +08003450 switch (op) {
3451 case BLK_ZO_OPEN:
3452 op_name = "BLKOPENZONE";
3453 zo = BLKOPENZONE;
3454 break;
3455 case BLK_ZO_CLOSE:
3456 op_name = "BLKCLOSEZONE";
3457 zo = BLKCLOSEZONE;
3458 break;
3459 case BLK_ZO_FINISH:
3460 op_name = "BLKFINISHZONE";
3461 zo = BLKFINISHZONE;
3462 break;
3463 case BLK_ZO_RESET:
3464 op_name = "BLKRESETZONE";
3465 zo = BLKRESETZONE;
3466 break;
3467 default:
3468 error_report("Unsupported zone op: 0x%x", op);
3469 return -ENOTSUP;
3470 }
3471
3472 acb = (RawPosixAIOData) {
3473 .bs = bs,
3474 .aio_fildes = s->fd,
3475 .aio_type = QEMU_AIO_ZONE_MGMT,
3476 .aio_offset = offset,
3477 .aio_nbytes = len,
3478 .zone_mgmt = {
3479 .op = zo,
3480 },
3481 };
3482
Sam Li142e3072023-05-08 12:55:32 +08003483 trace_zbd_zone_mgmt(bs, op_name, offset >> BDRV_SECTOR_BITS,
3484 len >> BDRV_SECTOR_BITS);
Sam Li6d43eaa2023-05-08 12:55:28 +08003485 ret = raw_thread_pool_submit(handle_aiocb_zone_mgmt, &acb);
3486 if (ret != 0) {
Sam Li10b9e082023-08-25 12:05:56 +08003487 update_zones_wp(bs, s->fd, offset, nrz);
Sam Li6d43eaa2023-05-08 12:55:28 +08003488 error_report("ioctl %s failed %d", op_name, ret);
Sam Lia3c41f02023-05-08 13:15:07 +08003489 return ret;
3490 }
3491
3492 if (zo == BLKRESETZONE && len == capacity) {
3493 ret = get_zones_wp(bs, s->fd, 0, bs->bl.nr_zones, 1);
3494 if (ret < 0) {
3495 error_report("reporting single wp failed");
3496 return ret;
3497 }
3498 } else if (zo == BLKRESETZONE) {
3499 for (unsigned int j = 0; j < nrz; ++j) {
3500 wp[j] = offset + j * zone_size;
3501 }
3502 } else if (zo == BLKFINISHZONE) {
3503 for (unsigned int j = 0; j < nrz; ++j) {
3504 /* The zoned device allows the last zone smaller that the
3505 * zone size. */
3506 wp[j] = MIN(offset + (j + 1) * zone_size, offset + len);
3507 }
Sam Li6d43eaa2023-05-08 12:55:28 +08003508 }
3509
3510 return ret;
3511}
3512#endif
3513
Sam Li4751d092023-05-08 13:15:08 +08003514#if defined(CONFIG_BLKZONED)
3515static int coroutine_fn raw_co_zone_append(BlockDriverState *bs,
3516 int64_t *offset,
3517 QEMUIOVector *qiov,
3518 BdrvRequestFlags flags) {
3519 assert(flags == 0);
3520 int64_t zone_size_mask = bs->bl.zone_size - 1;
3521 int64_t iov_len = 0;
3522 int64_t len = 0;
Sam Li4751d092023-05-08 13:15:08 +08003523
3524 if (*offset & zone_size_mask) {
3525 error_report("sector offset %" PRId64 " is not aligned to zone size "
3526 "%" PRId32 "", *offset / 512, bs->bl.zone_size / 512);
3527 return -EINVAL;
3528 }
3529
3530 int64_t wg = bs->bl.write_granularity;
3531 int64_t wg_mask = wg - 1;
3532 for (int i = 0; i < qiov->niov; i++) {
3533 iov_len = qiov->iov[i].iov_len;
3534 if (iov_len & wg_mask) {
3535 error_report("len of IOVector[%d] %" PRId64 " is not aligned to "
3536 "block size %" PRId64 "", i, iov_len, wg);
3537 return -EINVAL;
3538 }
3539 len += iov_len;
3540 }
3541
Sam Li6c811e12023-05-08 13:15:10 +08003542 trace_zbd_zone_append(bs, *offset >> BDRV_SECTOR_BITS);
Naohiro Aotaad4feac2023-10-30 16:38:53 +09003543 return raw_co_prw(bs, offset, len, qiov, QEMU_AIO_ZONE_APPEND);
Sam Li4751d092023-05-08 13:15:08 +08003544}
3545#endif
3546
Kevin Wolf33d70fb2018-06-21 19:07:32 +02003547static coroutine_fn int
Vladimir Sementsov-Ogievskiy0c802282021-09-03 13:28:06 +03003548raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes,
3549 bool blkdev)
Christoph Hellwigdce512d2010-12-17 11:41:15 +01003550{
Christoph Hellwigdce512d2010-12-17 11:41:15 +01003551 BDRVRawState *s = bs->opaque;
Kevin Wolf46ee0f42018-10-25 14:18:58 +01003552 RawPosixAIOData acb;
Anton Nefedov1c450362019-09-23 15:17:36 +03003553 int ret;
Christoph Hellwigdce512d2010-12-17 11:41:15 +01003554
Kevin Wolf46ee0f42018-10-25 14:18:58 +01003555 acb = (RawPosixAIOData) {
3556 .bs = bs,
3557 .aio_fildes = s->fd,
3558 .aio_type = QEMU_AIO_DISCARD,
3559 .aio_offset = offset,
3560 .aio_nbytes = bytes,
3561 };
3562
3563 if (blkdev) {
3564 acb.aio_type |= QEMU_AIO_BLKDEV;
3565 }
3566
Emanuele Giuseppe Esposito0fdb7312023-02-03 08:17:30 -05003567 ret = raw_thread_pool_submit(handle_aiocb_discard, &acb);
Anton Nefedov1c450362019-09-23 15:17:36 +03003568 raw_account_discard(s, bytes, ret);
3569 return ret;
Kevin Wolf46ee0f42018-10-25 14:18:58 +01003570}
3571
3572static coroutine_fn int
Vladimir Sementsov-Ogievskiy0c802282021-09-03 13:28:06 +03003573raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
Kevin Wolf46ee0f42018-10-25 14:18:58 +01003574{
3575 return raw_do_pdiscard(bs, offset, bytes, false);
Christoph Hellwigdce512d2010-12-17 11:41:15 +01003576}
Kevin Wolf0e7e1982009-05-18 16:42:10 +02003577
Kevin Wolf7154d8a2018-10-25 14:18:58 +01003578static int coroutine_fn
Vladimir Sementsov-Ogievskiyf34b2bc2021-09-03 13:28:03 +03003579raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
Kevin Wolf7154d8a2018-10-25 14:18:58 +01003580 BdrvRequestFlags flags, bool blkdev)
3581{
3582 BDRVRawState *s = bs->opaque;
3583 RawPosixAIOData acb;
3584 ThreadPoolFunc *handler;
3585
Max Reitz292d06b2019-11-01 16:25:10 +01003586#ifdef CONFIG_FALLOCATE
3587 if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) {
3588 BdrvTrackedRequest *req;
Max Reitz292d06b2019-11-01 16:25:10 +01003589
3590 /*
3591 * This is a workaround for a bug in the Linux XFS driver,
3592 * where writes submitted through the AIO interface will be
3593 * discarded if they happen beyond a concurrently running
3594 * fallocate() that increases the file length (i.e., both the
3595 * write and the fallocate() happen beyond the EOF).
3596 *
3597 * To work around it, we extend the tracked request for this
3598 * zero write until INT64_MAX (effectively infinity), and mark
3599 * it as serializing.
3600 *
3601 * We have to enable this workaround for all filesystems and
3602 * AIO modes (not just XFS with aio=native), because for
3603 * remote filesystems we do not know the host configuration.
3604 */
3605
3606 req = bdrv_co_get_self_request(bs);
3607 assert(req);
3608 assert(req->type == BDRV_TRACKED_WRITE);
3609 assert(req->offset <= offset);
3610 assert(req->offset + req->bytes >= offset + bytes);
3611
Vladimir Sementsov-Ogievskiy8b117002020-12-04 01:27:13 +03003612 req->bytes = BDRV_MAX_LENGTH - req->offset;
3613
Vladimir Sementsov-Ogievskiy69b55e02020-12-11 21:39:19 +03003614 bdrv_check_request(req->offset, req->bytes, &error_abort);
Max Reitz292d06b2019-11-01 16:25:10 +01003615
Vladimir Sementsov-Ogievskiy8ac5aab2020-10-21 17:58:43 +03003616 bdrv_make_request_serialising(req, bs->bl.request_alignment);
Max Reitz292d06b2019-11-01 16:25:10 +01003617 }
3618#endif
3619
Kevin Wolf7154d8a2018-10-25 14:18:58 +01003620 acb = (RawPosixAIOData) {
3621 .bs = bs,
3622 .aio_fildes = s->fd,
3623 .aio_type = QEMU_AIO_WRITE_ZEROES,
3624 .aio_offset = offset,
3625 .aio_nbytes = bytes,
3626 };
3627
3628 if (blkdev) {
3629 acb.aio_type |= QEMU_AIO_BLKDEV;
3630 }
Kevin Wolf738301e2019-03-22 13:45:23 +01003631 if (flags & BDRV_REQ_NO_FALLBACK) {
3632 acb.aio_type |= QEMU_AIO_NO_FALLBACK;
3633 }
Kevin Wolf7154d8a2018-10-25 14:18:58 +01003634
3635 if (flags & BDRV_REQ_MAY_UNMAP) {
3636 acb.aio_type |= QEMU_AIO_DISCARD;
3637 handler = handle_aiocb_write_zeroes_unmap;
3638 } else {
3639 handler = handle_aiocb_write_zeroes;
3640 }
3641
Emanuele Giuseppe Esposito0fdb7312023-02-03 08:17:30 -05003642 return raw_thread_pool_submit(handler, &acb);
Kevin Wolf7154d8a2018-10-25 14:18:58 +01003643}
3644
Eric Blake2ffa76c2016-06-01 15:10:10 -06003645static int coroutine_fn raw_co_pwrite_zeroes(
3646 BlockDriverState *bs, int64_t offset,
Vladimir Sementsov-Ogievskiyf34b2bc2021-09-03 13:28:03 +03003647 int64_t bytes, BdrvRequestFlags flags)
Paolo Bonzini260a82e2013-11-22 13:39:55 +01003648{
Kevin Wolf7154d8a2018-10-25 14:18:58 +01003649 return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false);
Paolo Bonzini260a82e2013-11-22 13:39:55 +01003650}
3651
Emanuele Giuseppe Esposito3d47eb02023-01-13 21:42:08 +01003652static int coroutine_fn
3653raw_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
Paolo Bonzini260a82e2013-11-22 13:39:55 +01003654{
Paolo Bonzini260a82e2013-11-22 13:39:55 +01003655 return 0;
3656}
3657
Hanna Reitz7f36a502022-06-20 18:26:54 +02003658static ImageInfoSpecific *raw_get_specific_info(BlockDriverState *bs,
3659 Error **errp)
3660{
3661 ImageInfoSpecificFile *file_info = g_new0(ImageInfoSpecificFile, 1);
3662 ImageInfoSpecific *spec_info = g_new(ImageInfoSpecific, 1);
3663
3664 *spec_info = (ImageInfoSpecific){
3665 .type = IMAGE_INFO_SPECIFIC_KIND_FILE,
3666 .u.file.data = file_info,
3667 };
3668
3669#ifdef FS_IOC_FSGETXATTR
3670 {
3671 BDRVRawState *s = bs->opaque;
3672 struct fsxattr attr;
3673 int ret;
3674
3675 ret = ioctl(s->fd, FS_IOC_FSGETXATTR, &attr);
3676 if (!ret && attr.fsx_extsize != 0) {
3677 file_info->has_extent_size_hint = true;
3678 file_info->extent_size_hint = attr.fsx_extsize;
3679 }
3680 }
3681#endif
3682
3683 return spec_info;
3684}
3685
Anton Nefedovd9245592019-09-23 15:17:37 +03003686static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs)
3687{
3688 BDRVRawState *s = bs->opaque;
3689 return (BlockStatsSpecificFile) {
3690 .discard_nb_ok = s->stats.discard_nb_ok,
3691 .discard_nb_failed = s->stats.discard_nb_failed,
3692 .discard_bytes_ok = s->stats.discard_bytes_ok,
3693 };
3694}
3695
3696static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs)
3697{
3698 BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1);
3699
3700 stats->driver = BLOCKDEV_DRIVER_FILE;
3701 stats->u.file = get_blockstats_specific_file(bs);
3702
3703 return stats;
3704}
3705
Joelle van Dyne14176c82021-03-15 11:03:38 -07003706#if defined(HAVE_HOST_BLOCK_DEVICE)
Anton Nefedovd9245592019-09-23 15:17:37 +03003707static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs)
3708{
3709 BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1);
3710
3711 stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE;
3712 stats->u.host_device = get_blockstats_specific_file(bs);
3713
3714 return stats;
3715}
Joelle van Dyne14176c82021-03-15 11:03:38 -07003716#endif /* HAVE_HOST_BLOCK_DEVICE */
Anton Nefedovd9245592019-09-23 15:17:37 +03003717
Chunyan Liu6f482f72014-06-05 17:21:01 +08003718static QemuOptsList raw_create_opts = {
3719 .name = "raw-create-opts",
3720 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
3721 .desc = {
3722 {
3723 .name = BLOCK_OPT_SIZE,
3724 .type = QEMU_OPT_SIZE,
3725 .help = "Virtual disk size"
3726 },
Chunyan Liu4ab15592014-06-30 14:29:58 +08003727 {
3728 .name = BLOCK_OPT_NOCOW,
3729 .type = QEMU_OPT_BOOL,
3730 .help = "Turn off copy-on-write (valid only on btrfs)"
3731 },
Hu Tao06247422014-09-10 17:05:48 +08003732 {
3733 .name = BLOCK_OPT_PREALLOC,
3734 .type = QEMU_OPT_STRING,
Stefano Garzarellaabea0052019-05-24 09:58:47 +02003735 .help = "Preallocation mode (allowed values: off"
3736#ifdef CONFIG_POSIX_FALLOCATE
3737 ", falloc"
3738#endif
3739 ", full)"
Hu Tao06247422014-09-10 17:05:48 +08003740 },
Kevin Wolfffa244c2020-07-07 16:23:29 +02003741 {
3742 .name = BLOCK_OPT_EXTENT_SIZE_HINT,
3743 .type = QEMU_OPT_SIZE,
3744 .help = "Extent size hint for the image file, 0 to disable"
3745 },
Chunyan Liu6f482f72014-06-05 17:21:01 +08003746 { /* end of list */ }
3747 }
Kevin Wolf0e7e1982009-05-18 16:42:10 +02003748};
3749
Fam Zheng244a5662017-05-03 00:35:56 +08003750static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared,
3751 Error **errp)
3752{
Kevin Wolf6ceabe62019-03-08 15:40:40 +01003753 BDRVRawState *s = bs->opaque;
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03003754 int input_flags = s->reopen_state ? s->reopen_state->flags : bs->open_flags;
Kevin Wolf6ceabe62019-03-08 15:40:40 +01003755 int open_flags;
3756 int ret;
3757
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03003758 /* We may need a new fd if auto-read-only switches the mode */
Denis V. Lunev viab67e3532024-04-30 19:02:13 +02003759 ret = raw_reconfigure_getfd(bs, input_flags, &open_flags, perm, errp);
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03003760 if (ret < 0) {
3761 return ret;
3762 } else if (ret != s->fd) {
3763 Error *local_err = NULL;
Kevin Wolf6ceabe62019-03-08 15:40:40 +01003764
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03003765 /*
3766 * Fail already check_perm() if we can't get a working O_DIRECT
3767 * alignment with the new fd.
3768 */
3769 raw_probe_alignment(bs, ret, &local_err);
3770 if (local_err) {
3771 error_propagate(errp, local_err);
3772 return -EINVAL;
Kevin Wolf6ceabe62019-03-08 15:40:40 +01003773 }
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03003774
3775 s->perm_change_fd = ret;
3776 s->perm_change_flags = open_flags;
Kevin Wolf6ceabe62019-03-08 15:40:40 +01003777 }
3778
3779 /* Prepare permissions on old fd to avoid conflicts between old and new,
3780 * but keep everything locked that new will need. */
3781 ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp);
3782 if (ret < 0) {
3783 goto fail;
3784 }
3785
3786 /* Copy locks to the new fd */
Li Fengeb43ea12020-12-07 19:44:06 +08003787 if (s->perm_change_fd && s->use_lock) {
Kevin Wolf6ceabe62019-03-08 15:40:40 +01003788 ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared,
3789 false, errp);
3790 if (ret < 0) {
3791 raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
3792 goto fail;
3793 }
3794 }
3795 return 0;
3796
3797fail:
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03003798 if (s->perm_change_fd) {
Kevin Wolf6ceabe62019-03-08 15:40:40 +01003799 qemu_close(s->perm_change_fd);
3800 }
3801 s->perm_change_fd = 0;
3802 return ret;
Fam Zheng244a5662017-05-03 00:35:56 +08003803}
3804
3805static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared)
3806{
3807 BDRVRawState *s = bs->opaque;
Kevin Wolf6ceabe62019-03-08 15:40:40 +01003808
3809 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
3810 * called after .bdrv_reopen_commit) */
3811 if (s->perm_change_fd && s->fd != s->perm_change_fd) {
3812 qemu_close(s->fd);
3813 s->fd = s->perm_change_fd;
Max Reitz094e3632019-05-22 19:03:45 +02003814 s->open_flags = s->perm_change_flags;
Kevin Wolf6ceabe62019-03-08 15:40:40 +01003815 }
3816 s->perm_change_fd = 0;
3817
Fam Zheng244a5662017-05-03 00:35:56 +08003818 raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL);
3819 s->perm = perm;
3820 s->shared_perm = shared;
3821}
3822
3823static void raw_abort_perm_update(BlockDriverState *bs)
3824{
Kevin Wolf6ceabe62019-03-08 15:40:40 +01003825 BDRVRawState *s = bs->opaque;
3826
3827 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
3828 * the file descriptor. */
Vladimir Sementsov-Ogievskiy72373e42021-04-28 18:17:58 +03003829 if (s->perm_change_fd) {
Kevin Wolf6ceabe62019-03-08 15:40:40 +01003830 qemu_close(s->perm_change_fd);
3831 }
3832 s->perm_change_fd = 0;
3833
Fam Zheng244a5662017-05-03 00:35:56 +08003834 raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL);
3835}
3836
Emanuele Giuseppe Esposito742bf092023-02-03 16:21:53 +01003837static int coroutine_fn GRAPH_RDLOCK raw_co_copy_range_from(
Vladimir Sementsov-Ogievskiy48535042021-09-03 13:28:01 +03003838 BlockDriverState *bs, BdrvChild *src, int64_t src_offset,
3839 BdrvChild *dst, int64_t dst_offset, int64_t bytes,
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03003840 BdrvRequestFlags read_flags, BdrvRequestFlags write_flags)
Fam Zheng1efad062018-06-01 17:26:43 +08003841{
Vladimir Sementsov-Ogievskiy67b51fb2018-07-09 19:37:17 +03003842 return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes,
3843 read_flags, write_flags);
Fam Zheng1efad062018-06-01 17:26:43 +08003844}
3845
Emanuele Giuseppe Esposito742bf092023-02-03 16:21:53 +01003846static int coroutine_fn GRAPH_RDLOCK
3847raw_co_copy_range_to(BlockDriverState *bs,
3848 BdrvChild *src, int64_t src_offset,
3849 BdrvChild *dst, int64_t dst_offset,
3850 int64_t bytes, BdrvRequestFlags read_flags,
3851 BdrvRequestFlags write_flags)
Fam Zheng1efad062018-06-01 17:26:43 +08003852{
Kevin Wolf58a209c2018-10-25 14:18:58 +01003853 RawPosixAIOData acb;
Fam Zheng1efad062018-06-01 17:26:43 +08003854 BDRVRawState *s = bs->opaque;
3855 BDRVRawState *src_s;
3856
3857 assert(dst->bs == bs);
3858 if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) {
3859 return -ENOTSUP;
3860 }
3861
3862 src_s = src->bs->opaque;
Fam Zheng9f850f62018-07-02 10:58:34 +08003863 if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) {
Fam Zheng1efad062018-06-01 17:26:43 +08003864 return -EIO;
3865 }
Kevin Wolf58a209c2018-10-25 14:18:58 +01003866
3867 acb = (RawPosixAIOData) {
3868 .bs = bs,
3869 .aio_type = QEMU_AIO_COPY_RANGE,
3870 .aio_fildes = src_s->fd,
3871 .aio_offset = src_offset,
3872 .aio_nbytes = bytes,
3873 .copy_range = {
3874 .aio_fd2 = s->fd,
3875 .aio_offset2 = dst_offset,
3876 },
3877 };
3878
Emanuele Giuseppe Esposito0fdb7312023-02-03 08:17:30 -05003879 return raw_thread_pool_submit(handle_aiocb_copy_range, &acb);
Fam Zheng1efad062018-06-01 17:26:43 +08003880}
3881
Max Reitz5f535a92014-12-02 18:32:41 +01003882BlockDriver bdrv_file = {
Christoph Hellwig84a12e62010-04-07 22:30:24 +02003883 .format_name = "file",
3884 .protocol_name = "file",
blueswir1856ae5c2009-04-07 17:57:09 +00003885 .instance_size = sizeof(BDRVRawState),
Benoît Canet030be322013-09-24 17:07:04 +02003886 .bdrv_needs_filename = true,
blueswir1856ae5c2009-04-07 17:57:09 +00003887 .bdrv_probe = NULL, /* no probe for protocols */
Max Reitz078896a2014-03-05 22:41:37 +01003888 .bdrv_parse_filename = raw_parse_filename,
Paolo Bonzini44b424d2022-11-24 16:22:22 +01003889 .bdrv_open = raw_open,
Jeff Codyeeb6b452012-09-20 15:13:25 -04003890 .bdrv_reopen_prepare = raw_reopen_prepare,
3891 .bdrv_reopen_commit = raw_reopen_commit,
3892 .bdrv_reopen_abort = raw_reopen_abort,
blueswir1856ae5c2009-04-07 17:57:09 +00003893 .bdrv_close = raw_close,
Kevin Wolf927f11e2018-01-16 16:04:21 +01003894 .bdrv_co_create = raw_co_create,
Stefan Hajnocziefc75e22018-01-18 13:43:45 +01003895 .bdrv_co_create_opts = raw_co_create_opts,
Peter Lieven3ac21622013-06-28 12:47:42 +02003896 .bdrv_has_zero_init = bdrv_has_zero_init_1,
Eric Blakea290f082018-02-13 14:26:44 -06003897 .bdrv_co_block_status = raw_co_block_status,
Stefan Hajnoczidd577a22018-04-27 17:23:11 +01003898 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
Eric Blake2ffa76c2016-06-01 15:10:10 -06003899 .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes,
Daniel Henrique Barboza9bffae12020-01-30 18:39:04 -03003900 .bdrv_co_delete_file = raw_co_delete_file,
ths3b46e622007-09-17 08:09:54 +00003901
Kevin Wolf9d52aa32016-06-03 17:36:27 +02003902 .bdrv_co_preadv = raw_co_preadv,
3903 .bdrv_co_pwritev = raw_co_pwritev,
Kevin Wolf33d70fb2018-06-21 19:07:32 +02003904 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
3905 .bdrv_co_pdiscard = raw_co_pdiscard,
Fam Zheng1efad062018-06-01 17:26:43 +08003906 .bdrv_co_copy_range_from = raw_co_copy_range_from,
3907 .bdrv_co_copy_range_to = raw_co_copy_range_to,
Paolo Bonzinic25f53b2011-11-29 12:42:20 +01003908 .bdrv_refresh_limits = raw_refresh_limits,
aliguori3c529d92008-12-12 16:41:40 +00003909
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01003910 .bdrv_co_truncate = raw_co_truncate,
3911 .bdrv_co_getlength = raw_co_getlength,
Emanuele Giuseppe Esposito3d47eb02023-01-13 21:42:08 +01003912 .bdrv_co_get_info = raw_co_get_info,
Hanna Reitz7f36a502022-06-20 18:26:54 +02003913 .bdrv_get_specific_info = raw_get_specific_info,
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01003914 .bdrv_co_get_allocated_file_size = raw_co_get_allocated_file_size,
Anton Nefedovd9245592019-09-23 15:17:37 +03003915 .bdrv_get_specific_stats = raw_get_specific_stats,
Fam Zheng244a5662017-05-03 00:35:56 +08003916 .bdrv_check_perm = raw_check_perm,
3917 .bdrv_set_perm = raw_set_perm,
3918 .bdrv_abort_perm_update = raw_abort_perm_update,
Chunyan Liu6f482f72014-06-05 17:21:01 +08003919 .create_opts = &raw_create_opts,
Alberto Garcia8a2ce0b2019-03-12 18:48:48 +02003920 .mutable_opts = mutable_opts,
bellard83f64092006-08-01 16:21:11 +00003921};
3922
bellard19cb3732006-08-19 11:45:59 +00003923/***********************************************/
3924/* host device */
3925
Joelle van Dyne14176c82021-03-15 11:03:38 -07003926#if defined(HAVE_HOST_BLOCK_DEVICE)
3927
Pavel Borzenkov83affaa2011-11-10 22:40:06 +04003928#if defined(__APPLE__) && defined(__MACH__)
Programmingkid98caa5b2015-11-20 19:17:48 -05003929static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
3930 CFIndex maxPathSize, int flags);
Philippe Mathieu-Daudéaa44d3f2022-01-06 00:46:01 +01003931
Programmingkidd0855f12016-03-21 11:41:28 -04003932static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator)
bellard19cb3732006-08-19 11:45:59 +00003933{
Programmingkidd0855f12016-03-21 11:41:28 -04003934 kern_return_t kernResult = KERN_FAILURE;
Philippe Mathieu-Daudéaa44d3f2022-01-06 00:46:01 +01003935 mach_port_t mainPort;
bellard19cb3732006-08-19 11:45:59 +00003936 CFMutableDictionaryRef classesToMatch;
Programmingkidd0855f12016-03-21 11:41:28 -04003937 const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass};
3938 char *mediaType = NULL;
bellard19cb3732006-08-19 11:45:59 +00003939
Philippe Mathieu-Daudéaa44d3f2022-01-06 00:46:01 +01003940 kernResult = IOMainPort(MACH_PORT_NULL, &mainPort);
bellard19cb3732006-08-19 11:45:59 +00003941 if ( KERN_SUCCESS != kernResult ) {
Philippe Mathieu-Daudéaa44d3f2022-01-06 00:46:01 +01003942 printf("IOMainPort returned %d\n", kernResult);
bellard19cb3732006-08-19 11:45:59 +00003943 }
ths3b46e622007-09-17 08:09:54 +00003944
Programmingkidd0855f12016-03-21 11:41:28 -04003945 int index;
3946 for (index = 0; index < ARRAY_SIZE(matching_array); index++) {
3947 classesToMatch = IOServiceMatching(matching_array[index]);
3948 if (classesToMatch == NULL) {
3949 error_report("IOServiceMatching returned NULL for %s",
3950 matching_array[index]);
3951 continue;
3952 }
3953 CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey),
3954 kCFBooleanTrue);
Philippe Mathieu-Daudéaa44d3f2022-01-06 00:46:01 +01003955 kernResult = IOServiceGetMatchingServices(mainPort, classesToMatch,
Programmingkidd0855f12016-03-21 11:41:28 -04003956 mediaIterator);
3957 if (kernResult != KERN_SUCCESS) {
3958 error_report("Note: IOServiceGetMatchingServices returned %d",
3959 kernResult);
3960 continue;
3961 }
ths3b46e622007-09-17 08:09:54 +00003962
Programmingkidd0855f12016-03-21 11:41:28 -04003963 /* If a match was found, leave the loop */
3964 if (*mediaIterator != 0) {
Laurent Vivier4f7d28d2018-12-13 17:27:26 +01003965 trace_file_FindEjectableOpticalMedia(matching_array[index]);
Programmingkidd0855f12016-03-21 11:41:28 -04003966 mediaType = g_strdup(matching_array[index]);
3967 break;
3968 }
3969 }
3970 return mediaType;
bellard19cb3732006-08-19 11:45:59 +00003971}
3972
Programmingkid98caa5b2015-11-20 19:17:48 -05003973kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath,
3974 CFIndex maxPathSize, int flags)
bellard19cb3732006-08-19 11:45:59 +00003975{
3976 io_object_t nextMedia;
3977 kern_return_t kernResult = KERN_FAILURE;
3978 *bsdPath = '\0';
3979 nextMedia = IOIteratorNext( mediaIterator );
3980 if ( nextMedia )
3981 {
3982 CFTypeRef bsdPathAsCFString;
3983 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
3984 if ( bsdPathAsCFString ) {
3985 size_t devPathLength;
3986 strcpy( bsdPath, _PATH_DEV );
Programmingkid98caa5b2015-11-20 19:17:48 -05003987 if (flags & BDRV_O_NOCACHE) {
3988 strcat(bsdPath, "r");
3989 }
bellard19cb3732006-08-19 11:45:59 +00003990 devPathLength = strlen( bsdPath );
3991 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
3992 kernResult = KERN_SUCCESS;
3993 }
3994 CFRelease( bsdPathAsCFString );
3995 }
3996 IOObjectRelease( nextMedia );
3997 }
ths3b46e622007-09-17 08:09:54 +00003998
bellard19cb3732006-08-19 11:45:59 +00003999 return kernResult;
4000}
4001
Programmingkidd0855f12016-03-21 11:41:28 -04004002/* Sets up a real cdrom for use in QEMU */
4003static bool setup_cdrom(char *bsd_path, Error **errp)
4004{
4005 int index, num_of_test_partitions = 2, fd;
4006 char test_partition[MAXPATHLEN];
4007 bool partition_found = false;
4008
4009 /* look for a working partition */
4010 for (index = 0; index < num_of_test_partitions; index++) {
4011 snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path,
4012 index);
Daniel P. Berrangéb18a24a2020-07-01 15:22:43 +01004013 fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE, NULL);
Programmingkidd0855f12016-03-21 11:41:28 -04004014 if (fd >= 0) {
4015 partition_found = true;
4016 qemu_close(fd);
4017 break;
4018 }
4019 }
4020
4021 /* if a working partition on the device was not found */
4022 if (partition_found == false) {
4023 error_setg(errp, "Failed to find a working partition on disc");
4024 } else {
Laurent Vivier4f7d28d2018-12-13 17:27:26 +01004025 trace_file_setup_cdrom(test_partition);
Programmingkidd0855f12016-03-21 11:41:28 -04004026 pstrcpy(bsd_path, MAXPATHLEN, test_partition);
4027 }
4028 return partition_found;
4029}
4030
4031/* Prints directions on mounting and unmounting a device */
4032static void print_unmounting_directions(const char *file_name)
4033{
4034 error_report("If device %s is mounted on the desktop, unmount"
4035 " it first before using it in QEMU", file_name);
4036 error_report("Command to unmount device: diskutil unmountDisk %s",
4037 file_name);
4038 error_report("Command to mount device: diskutil mountDisk %s", file_name);
4039}
4040
4041#endif /* defined(__APPLE__) && defined(__MACH__) */
bellard19cb3732006-08-19 11:45:59 +00004042
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02004043static int hdev_probe_device(const char *filename)
4044{
4045 struct stat st;
4046
4047 /* allow a dedicated CD-ROM driver to match with a higher priority */
4048 if (strstart(filename, "/dev/cdrom", NULL))
4049 return 50;
4050
4051 if (stat(filename, &st) >= 0 &&
4052 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
4053 return 100;
4054 }
4055
4056 return 0;
4057}
4058
Max Reitz7af803d2014-03-08 00:39:41 +01004059static void hdev_parse_filename(const char *filename, QDict *options,
4060 Error **errp)
4061{
Max Reitz03c320d2017-05-22 21:52:16 +02004062 bdrv_parse_filename_strip_prefix(filename, "host_device:", options);
Max Reitz7af803d2014-03-08 00:39:41 +01004063}
4064
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03004065static bool hdev_is_sg(BlockDriverState *bs)
4066{
4067
4068#if defined(__linux__)
4069
Kevin Wolf0d4377b2016-10-20 14:50:12 +02004070 BDRVRawState *s = bs->opaque;
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03004071 struct stat st;
4072 struct sg_scsi_id scsiid;
4073 int sg_version;
Kevin Wolf0d4377b2016-10-20 14:50:12 +02004074 int ret;
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03004075
Kevin Wolf0d4377b2016-10-20 14:50:12 +02004076 if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) {
4077 return false;
4078 }
4079
4080 ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version);
4081 if (ret < 0) {
4082 return false;
4083 }
4084
4085 ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid);
4086 if (ret >= 0) {
Laurent Vivier4f7d28d2018-12-13 17:27:26 +01004087 trace_file_hdev_is_sg(scsiid.scsi_type, sg_version);
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03004088 return true;
4089 }
4090
4091#endif
4092
4093 return false;
4094}
4095
Max Reitz015a1032013-09-05 14:22:29 +02004096static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
4097 Error **errp)
bellard19cb3732006-08-19 11:45:59 +00004098{
4099 BDRVRawState *s = bs->opaque;
Stefan Hajnoczida888d32013-02-05 12:28:33 +01004100 int ret;
aliguoria76bab42008-09-22 19:17:18 +00004101
Pavel Borzenkov83affaa2011-11-10 22:40:06 +04004102#if defined(__APPLE__) && defined(__MACH__)
Markus Armbruster129c7d12017-03-30 19:43:12 +02004103 /*
4104 * Caution: while qdict_get_str() is fine, getting non-string types
4105 * would require more care. When @options come from -blockdev or
4106 * blockdev_add, its members are typed according to the QAPI
4107 * schema, but when they come from -drive, they're all QString.
4108 */
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03004109 const char *filename = qdict_get_str(options, "filename");
Programmingkidd0855f12016-03-21 11:41:28 -04004110 char bsd_path[MAXPATHLEN] = "";
4111 bool error_occurred = false;
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03004112
Programmingkidd0855f12016-03-21 11:41:28 -04004113 /* If using a real cdrom */
4114 if (strcmp(filename, "/dev/cdrom") == 0) {
4115 char *mediaType = NULL;
4116 kern_return_t ret_val;
4117 io_iterator_t mediaIterator = 0;
ths5fafdf22007-09-16 21:08:06 +00004118
Programmingkidd0855f12016-03-21 11:41:28 -04004119 mediaType = FindEjectableOpticalMedia(&mediaIterator);
4120 if (mediaType == NULL) {
4121 error_setg(errp, "Please make sure your CD/DVD is in the optical"
4122 " drive");
4123 error_occurred = true;
4124 goto hdev_open_Mac_error;
bellard19cb3732006-08-19 11:45:59 +00004125 }
ths3b46e622007-09-17 08:09:54 +00004126
Programmingkidd0855f12016-03-21 11:41:28 -04004127 ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags);
4128 if (ret_val != KERN_SUCCESS) {
4129 error_setg(errp, "Could not get BSD path for optical drive");
4130 error_occurred = true;
4131 goto hdev_open_Mac_error;
4132 }
4133
4134 /* If a real optical drive was not found */
4135 if (bsd_path[0] == '\0') {
4136 error_setg(errp, "Failed to obtain bsd path for optical drive");
4137 error_occurred = true;
4138 goto hdev_open_Mac_error;
4139 }
4140
4141 /* If using a cdrom disc and finding a partition on the disc failed */
4142 if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 &&
4143 setup_cdrom(bsd_path, errp) == false) {
4144 print_unmounting_directions(bsd_path);
4145 error_occurred = true;
4146 goto hdev_open_Mac_error;
4147 }
4148
Eric Blake46f5ac22017-04-27 16:58:17 -05004149 qdict_put_str(options, "filename", bsd_path);
Programmingkidd0855f12016-03-21 11:41:28 -04004150
4151hdev_open_Mac_error:
4152 g_free(mediaType);
4153 if (mediaIterator) {
4154 IOObjectRelease(mediaIterator);
4155 }
4156 if (error_occurred) {
4157 return -ENOENT;
4158 }
bellard19cb3732006-08-19 11:45:59 +00004159 }
Programmingkidd0855f12016-03-21 11:41:28 -04004160#endif /* defined(__APPLE__) && defined(__MACH__) */
bellard19cb3732006-08-19 11:45:59 +00004161
4162 s->type = FTYPE_FILE;
Christoph Hellwig90babde2009-06-15 13:53:38 +02004163
Markus Armbruster668f62e2020-07-07 18:06:02 +02004164 ret = raw_open_common(bs, options, flags, 0, true, errp);
Stefan Hajnoczida888d32013-02-05 12:28:33 +01004165 if (ret < 0) {
Programmingkidd0855f12016-03-21 11:41:28 -04004166#if defined(__APPLE__) && defined(__MACH__)
4167 if (*bsd_path) {
4168 filename = bsd_path;
4169 }
4170 /* if a physical device experienced an error while being opened */
4171 if (strncmp(filename, "/dev/", 5) == 0) {
4172 print_unmounting_directions(filename);
4173 }
4174#endif /* defined(__APPLE__) && defined(__MACH__) */
Stefan Hajnoczida888d32013-02-05 12:28:33 +01004175 return ret;
4176 }
4177
Dimitris Aragiorgis3307ed72015-06-23 13:45:00 +03004178 /* Since this does ioctl the device must be already opened */
4179 bs->sg = hdev_is_sg(bs);
4180
Stefan Hajnoczida888d32013-02-05 12:28:33 +01004181 return ret;
bellard19cb3732006-08-19 11:45:59 +00004182}
4183
aliguori03ff3ca2008-09-15 15:51:35 +00004184#if defined(__linux__)
Kevin Wolf2f3a7ab2018-10-31 11:25:18 +01004185static int coroutine_fn
4186hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
aliguori221f7152009-03-28 17:28:41 +00004187{
aliguorif141eaf2009-04-07 18:43:24 +00004188 BDRVRawState *s = bs->opaque;
Kevin Wolf03425672018-10-31 11:30:42 +01004189 RawPosixAIOData acb;
Kevin Wolf2f3a7ab2018-10-31 11:25:18 +01004190 int ret;
aliguori221f7152009-03-28 17:28:41 +00004191
Kevin Wolf2f3a7ab2018-10-31 11:25:18 +01004192 ret = fd_open(bs);
4193 if (ret < 0) {
4194 return ret;
4195 }
Paolo Bonzinic208e8c2012-11-02 16:14:20 +01004196
Paolo Bonzini7c9e5272017-08-21 18:58:56 +02004197 if (req == SG_IO && s->pr_mgr) {
4198 struct sg_io_hdr *io_hdr = buf;
4199 if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT ||
4200 io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) {
Emanuele Giuseppe Esposito0fdb7312023-02-03 08:17:30 -05004201 return pr_manager_execute(s->pr_mgr, qemu_get_current_aio_context(),
Kevin Wolf2f3a7ab2018-10-31 11:25:18 +01004202 s->fd, io_hdr);
Paolo Bonzini7c9e5272017-08-21 18:58:56 +02004203 }
4204 }
4205
Kevin Wolf03425672018-10-31 11:30:42 +01004206 acb = (RawPosixAIOData) {
4207 .bs = bs,
4208 .aio_type = QEMU_AIO_IOCTL,
4209 .aio_fildes = s->fd,
4210 .aio_offset = 0,
4211 .ioctl = {
4212 .buf = buf,
4213 .cmd = req,
4214 },
4215 };
4216
Emanuele Giuseppe Esposito0fdb7312023-02-03 08:17:30 -05004217 return raw_thread_pool_submit(handle_aiocb_ioctl, &acb);
aliguori221f7152009-03-28 17:28:41 +00004218}
Max Reitzf7096232015-10-19 17:53:07 +02004219#endif /* linux */
aliguori221f7152009-03-28 17:28:41 +00004220
Kevin Wolf33d70fb2018-06-21 19:07:32 +02004221static coroutine_fn int
Vladimir Sementsov-Ogievskiy0c802282021-09-03 13:28:06 +03004222hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
Andreas Färberc36dd8a2013-01-18 16:43:35 +01004223{
Anton Nefedov1c450362019-09-23 15:17:36 +03004224 BDRVRawState *s = bs->opaque;
Kevin Wolf33d70fb2018-06-21 19:07:32 +02004225 int ret;
Andreas Färberc36dd8a2013-01-18 16:43:35 +01004226
Kevin Wolf33d70fb2018-06-21 19:07:32 +02004227 ret = fd_open(bs);
4228 if (ret < 0) {
Anton Nefedov1c450362019-09-23 15:17:36 +03004229 raw_account_discard(s, bytes, ret);
Kevin Wolf33d70fb2018-06-21 19:07:32 +02004230 return ret;
Andreas Färberc36dd8a2013-01-18 16:43:35 +01004231 }
Kevin Wolf46ee0f42018-10-25 14:18:58 +01004232 return raw_do_pdiscard(bs, offset, bytes, true);
Andreas Färberc36dd8a2013-01-18 16:43:35 +01004233}
4234
Eric Blake2ffa76c2016-06-01 15:10:10 -06004235static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs,
Vladimir Sementsov-Ogievskiyf34b2bc2021-09-03 13:28:03 +03004236 int64_t offset, int64_t bytes, BdrvRequestFlags flags)
Paolo Bonzinid0b45032013-11-22 13:39:56 +01004237{
Paolo Bonzinid0b45032013-11-22 13:39:56 +01004238 int rc;
4239
4240 rc = fd_open(bs);
4241 if (rc < 0) {
4242 return rc;
4243 }
Kevin Wolf34fa1102018-07-26 11:28:30 +02004244
Kevin Wolf7154d8a2018-10-25 14:18:58 +01004245 return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true);
Paolo Bonzinid0b45032013-11-22 13:39:56 +01004246}
4247
Anthony Liguori5efa9d52009-05-09 17:03:42 -05004248static BlockDriver bdrv_host_device = {
Kevin Wolf0b4ce022009-10-01 12:35:49 +02004249 .format_name = "host_device",
Christoph Hellwig84a12e62010-04-07 22:30:24 +02004250 .protocol_name = "host_device",
Kevin Wolf0b4ce022009-10-01 12:35:49 +02004251 .instance_size = sizeof(BDRVRawState),
Benoît Canet030be322013-09-24 17:07:04 +02004252 .bdrv_needs_filename = true,
Kevin Wolf0b4ce022009-10-01 12:35:49 +02004253 .bdrv_probe_device = hdev_probe_device,
Max Reitz7af803d2014-03-08 00:39:41 +01004254 .bdrv_parse_filename = hdev_parse_filename,
Paolo Bonzini44b424d2022-11-24 16:22:22 +01004255 .bdrv_open = hdev_open,
Kevin Wolf0b4ce022009-10-01 12:35:49 +02004256 .bdrv_close = raw_close,
Jeff Cody1bc6b702012-11-20 10:21:10 -05004257 .bdrv_reopen_prepare = raw_reopen_prepare,
4258 .bdrv_reopen_commit = raw_reopen_commit,
4259 .bdrv_reopen_abort = raw_reopen_abort,
Maxim Levitsky5a5e7f82020-03-26 03:12:18 +02004260 .bdrv_co_create_opts = bdrv_co_create_opts_simple,
4261 .create_opts = &bdrv_create_opts_simple,
Alberto Garcia8a2ce0b2019-03-12 18:48:48 +02004262 .mutable_opts = mutable_opts,
Stefan Hajnoczidd577a22018-04-27 17:23:11 +01004263 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
Eric Blake2ffa76c2016-06-01 15:10:10 -06004264 .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes,
ths3b46e622007-09-17 08:09:54 +00004265
Kevin Wolf9d52aa32016-06-03 17:36:27 +02004266 .bdrv_co_preadv = raw_co_preadv,
4267 .bdrv_co_pwritev = raw_co_pwritev,
Kevin Wolf33d70fb2018-06-21 19:07:32 +02004268 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
4269 .bdrv_co_pdiscard = hdev_co_pdiscard,
Fam Zheng1efad062018-06-01 17:26:43 +08004270 .bdrv_co_copy_range_from = raw_co_copy_range_from,
4271 .bdrv_co_copy_range_to = raw_co_copy_range_to,
Paolo Bonzinic25f53b2011-11-29 12:42:20 +01004272 .bdrv_refresh_limits = raw_refresh_limits,
aliguori3c529d92008-12-12 16:41:40 +00004273
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01004274 .bdrv_co_truncate = raw_co_truncate,
4275 .bdrv_co_getlength = raw_co_getlength,
Emanuele Giuseppe Esposito3d47eb02023-01-13 21:42:08 +01004276 .bdrv_co_get_info = raw_co_get_info,
Hanna Reitz7f36a502022-06-20 18:26:54 +02004277 .bdrv_get_specific_info = raw_get_specific_info,
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01004278 .bdrv_co_get_allocated_file_size = raw_co_get_allocated_file_size,
Anton Nefedovd9245592019-09-23 15:17:37 +03004279 .bdrv_get_specific_stats = hdev_get_specific_stats,
Fam Zheng244a5662017-05-03 00:35:56 +08004280 .bdrv_check_perm = raw_check_perm,
4281 .bdrv_set_perm = raw_set_perm,
4282 .bdrv_abort_perm_update = raw_abort_perm_update,
Ekaterina Tumanova1a9335e2015-02-16 12:47:56 +01004283 .bdrv_probe_blocksizes = hdev_probe_blocksizes,
4284 .bdrv_probe_geometry = hdev_probe_geometry,
bellard19cb3732006-08-19 11:45:59 +00004285
ths985a03b2007-12-24 16:10:43 +00004286 /* generic scsi device */
Christoph Hellwig63ec93d2009-06-15 14:04:34 +02004287#ifdef __linux__
Kevin Wolf2f3a7ab2018-10-31 11:25:18 +01004288 .bdrv_co_ioctl = hdev_co_ioctl,
Christoph Hellwig63ec93d2009-06-15 14:04:34 +02004289#endif
Sam Li6d43eaa2023-05-08 12:55:28 +08004290
4291 /* zoned device */
4292#if defined(CONFIG_BLKZONED)
4293 /* zone management operations */
4294 .bdrv_co_zone_report = raw_co_zone_report,
4295 .bdrv_co_zone_mgmt = raw_co_zone_mgmt,
Sam Li4751d092023-05-08 13:15:08 +08004296 .bdrv_co_zone_append = raw_co_zone_append,
Sam Li6d43eaa2023-05-08 12:55:28 +08004297#endif
bellard19cb3732006-08-19 11:45:59 +00004298};
Anthony Liguori5efa9d52009-05-09 17:03:42 -05004299
Max Reitz18fa1c42014-03-08 00:39:43 +01004300#if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
4301static void cdrom_parse_filename(const char *filename, QDict *options,
4302 Error **errp)
4303{
Max Reitz03c320d2017-05-22 21:52:16 +02004304 bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options);
Max Reitz18fa1c42014-03-08 00:39:43 +01004305}
Paolo Bonzini8c6f27e2023-04-07 17:32:59 +02004306
4307static void cdrom_refresh_limits(BlockDriverState *bs, Error **errp)
4308{
4309 bs->bl.has_variable_length = true;
4310 raw_refresh_limits(bs, errp);
4311}
Max Reitz18fa1c42014-03-08 00:39:43 +01004312#endif
4313
4314#ifdef __linux__
Max Reitz015a1032013-09-05 14:22:29 +02004315static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
4316 Error **errp)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004317{
4318 BDRVRawState *s = bs->opaque;
4319
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004320 s->type = FTYPE_CD;
4321
Blue Swirl19a3da72009-06-17 18:27:44 +03004322 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
John Snow230ff732018-07-10 13:00:40 -04004323 return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp);
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004324}
4325
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02004326static int cdrom_probe_device(const char *filename)
4327{
Cole Robinson3baf7202010-01-14 11:19:40 -05004328 int fd, ret;
4329 int prio = 0;
Johannes Stezenbach343f8562011-06-29 16:25:17 +02004330 struct stat st;
Cole Robinson3baf7202010-01-14 11:19:40 -05004331
Daniel P. Berrangéb18a24a2020-07-01 15:22:43 +01004332 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK, NULL);
Cole Robinson3baf7202010-01-14 11:19:40 -05004333 if (fd < 0) {
4334 goto out;
4335 }
Johannes Stezenbach343f8562011-06-29 16:25:17 +02004336 ret = fstat(fd, &st);
4337 if (ret == -1 || !S_ISBLK(st.st_mode)) {
4338 goto outc;
4339 }
Cole Robinson3baf7202010-01-14 11:19:40 -05004340
4341 /* Attempt to detect via a CDROM specific ioctl */
4342 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
4343 if (ret >= 0)
4344 prio = 100;
4345
Johannes Stezenbach343f8562011-06-29 16:25:17 +02004346outc:
Corey Bryant2e1e79d2012-08-14 16:43:46 -04004347 qemu_close(fd);
Cole Robinson3baf7202010-01-14 11:19:40 -05004348out:
4349 return prio;
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02004350}
4351
Emanuele Giuseppe Esposito1e97be92023-01-13 21:42:02 +01004352static bool coroutine_fn cdrom_co_is_inserted(BlockDriverState *bs)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004353{
4354 BDRVRawState *s = bs->opaque;
4355 int ret;
4356
4357 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
Max Reitze031f752015-10-19 17:53:11 +02004358 return ret == CDS_DISC_OK;
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004359}
4360
Emanuele Giuseppe Esposito2531b392023-01-13 21:42:09 +01004361static void coroutine_fn cdrom_co_eject(BlockDriverState *bs, bool eject_flag)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004362{
4363 BDRVRawState *s = bs->opaque;
4364
4365 if (eject_flag) {
4366 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
4367 perror("CDROMEJECT");
4368 } else {
4369 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
4370 perror("CDROMEJECT");
4371 }
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004372}
4373
Emanuele Giuseppe Esposito2c752612023-01-13 21:42:10 +01004374static void coroutine_fn cdrom_co_lock_medium(BlockDriverState *bs, bool locked)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004375{
4376 BDRVRawState *s = bs->opaque;
4377
4378 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
4379 /*
4380 * Note: an error can happen if the distribution automatically
4381 * mounts the CD-ROM
4382 */
4383 /* perror("CDROM_LOCKDOOR"); */
4384 }
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004385}
4386
4387static BlockDriver bdrv_host_cdrom = {
4388 .format_name = "host_cdrom",
Christoph Hellwig84a12e62010-04-07 22:30:24 +02004389 .protocol_name = "host_cdrom",
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004390 .instance_size = sizeof(BDRVRawState),
Benoît Canet030be322013-09-24 17:07:04 +02004391 .bdrv_needs_filename = true,
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02004392 .bdrv_probe_device = cdrom_probe_device,
Max Reitz18fa1c42014-03-08 00:39:43 +01004393 .bdrv_parse_filename = cdrom_parse_filename,
Paolo Bonzini44b424d2022-11-24 16:22:22 +01004394 .bdrv_open = cdrom_open,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004395 .bdrv_close = raw_close,
Jeff Cody1bc6b702012-11-20 10:21:10 -05004396 .bdrv_reopen_prepare = raw_reopen_prepare,
4397 .bdrv_reopen_commit = raw_reopen_commit,
4398 .bdrv_reopen_abort = raw_reopen_abort,
Maxim Levitsky5a5e7f82020-03-26 03:12:18 +02004399 .bdrv_co_create_opts = bdrv_co_create_opts_simple,
4400 .create_opts = &bdrv_create_opts_simple,
Alberto Garcia8a2ce0b2019-03-12 18:48:48 +02004401 .mutable_opts = mutable_opts,
Stefan Hajnoczidd577a22018-04-27 17:23:11 +01004402 .bdrv_co_invalidate_cache = raw_co_invalidate_cache,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004403
Kevin Wolf9d52aa32016-06-03 17:36:27 +02004404 .bdrv_co_preadv = raw_co_preadv,
4405 .bdrv_co_pwritev = raw_co_pwritev,
Kevin Wolf33d70fb2018-06-21 19:07:32 +02004406 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
Paolo Bonzini8c6f27e2023-04-07 17:32:59 +02004407 .bdrv_refresh_limits = cdrom_refresh_limits,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004408
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01004409 .bdrv_co_truncate = raw_co_truncate,
4410 .bdrv_co_getlength = raw_co_getlength,
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01004411 .bdrv_co_get_allocated_file_size = raw_co_get_allocated_file_size,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004412
4413 /* removable device support */
Emanuele Giuseppe Esposito1e97be92023-01-13 21:42:02 +01004414 .bdrv_co_is_inserted = cdrom_co_is_inserted,
Emanuele Giuseppe Esposito2531b392023-01-13 21:42:09 +01004415 .bdrv_co_eject = cdrom_co_eject,
Emanuele Giuseppe Esposito2c752612023-01-13 21:42:10 +01004416 .bdrv_co_lock_medium = cdrom_co_lock_medium,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004417
4418 /* generic scsi device */
Kevin Wolf2f3a7ab2018-10-31 11:25:18 +01004419 .bdrv_co_ioctl = hdev_co_ioctl,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004420};
4421#endif /* __linux__ */
4422
Aurelien Jarnoa167ba52009-11-29 18:00:41 +01004423#if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
Andreas Tobler511018e2013-10-31 22:41:46 +01004424static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
4425 Error **errp)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004426{
4427 BDRVRawState *s = bs->opaque;
4428 int ret;
4429
4430 s->type = FTYPE_CD;
4431
Markus Armbruster668f62e2020-07-07 18:06:02 +02004432 ret = raw_open_common(bs, options, flags, 0, true, errp);
Max Reitze428e432013-10-11 11:37:01 +02004433 if (ret) {
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004434 return ret;
Max Reitze428e432013-10-11 11:37:01 +02004435 }
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004436
Dong Xu Wang9b2260c2011-11-22 18:06:25 +08004437 /* make sure the door isn't locked at this time */
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004438 ioctl(s->fd, CDIOCALLOW);
4439 return 0;
4440}
4441
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02004442static int cdrom_probe_device(const char *filename)
4443{
4444 if (strstart(filename, "/dev/cd", NULL) ||
4445 strstart(filename, "/dev/acd", NULL))
4446 return 100;
4447 return 0;
4448}
4449
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004450static int cdrom_reopen(BlockDriverState *bs)
4451{
4452 BDRVRawState *s = bs->opaque;
4453 int fd;
4454
4455 /*
4456 * Force reread of possibly changed/newly loaded disc,
4457 * FreeBSD seems to not notice sometimes...
4458 */
4459 if (s->fd >= 0)
Corey Bryant2e1e79d2012-08-14 16:43:46 -04004460 qemu_close(s->fd);
Daniel P. Berrangéb18a24a2020-07-01 15:22:43 +01004461 fd = qemu_open(bs->filename, s->open_flags, NULL);
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004462 if (fd < 0) {
4463 s->fd = -1;
4464 return -EIO;
4465 }
4466 s->fd = fd;
4467
Dong Xu Wang9b2260c2011-11-22 18:06:25 +08004468 /* make sure the door isn't locked at this time */
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004469 ioctl(s->fd, CDIOCALLOW);
4470 return 0;
4471}
4472
Emanuele Giuseppe Esposito1e97be92023-01-13 21:42:02 +01004473static bool coroutine_fn cdrom_co_is_inserted(BlockDriverState *bs)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004474{
Paolo Bonzini36c6c872023-06-01 13:51:34 +02004475 return raw_getlength(bs) > 0;
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004476}
4477
Emanuele Giuseppe Esposito2531b392023-01-13 21:42:09 +01004478static void coroutine_fn cdrom_co_eject(BlockDriverState *bs, bool eject_flag)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004479{
4480 BDRVRawState *s = bs->opaque;
4481
4482 if (s->fd < 0)
Markus Armbruster822e1cd2011-07-20 18:23:42 +02004483 return;
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004484
4485 (void) ioctl(s->fd, CDIOCALLOW);
4486
4487 if (eject_flag) {
4488 if (ioctl(s->fd, CDIOCEJECT) < 0)
4489 perror("CDIOCEJECT");
4490 } else {
4491 if (ioctl(s->fd, CDIOCCLOSE) < 0)
4492 perror("CDIOCCLOSE");
4493 }
4494
Markus Armbruster822e1cd2011-07-20 18:23:42 +02004495 cdrom_reopen(bs);
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004496}
4497
Emanuele Giuseppe Esposito2c752612023-01-13 21:42:10 +01004498static void coroutine_fn cdrom_co_lock_medium(BlockDriverState *bs, bool locked)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004499{
4500 BDRVRawState *s = bs->opaque;
4501
4502 if (s->fd < 0)
Markus Armbruster7bf37fe2011-07-20 18:23:41 +02004503 return;
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004504 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
4505 /*
4506 * Note: an error can happen if the distribution automatically
4507 * mounts the CD-ROM
4508 */
4509 /* perror("CDROM_LOCKDOOR"); */
4510 }
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004511}
4512
4513static BlockDriver bdrv_host_cdrom = {
4514 .format_name = "host_cdrom",
Christoph Hellwig84a12e62010-04-07 22:30:24 +02004515 .protocol_name = "host_cdrom",
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004516 .instance_size = sizeof(BDRVRawState),
Benoît Canet030be322013-09-24 17:07:04 +02004517 .bdrv_needs_filename = true,
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02004518 .bdrv_probe_device = cdrom_probe_device,
Max Reitz18fa1c42014-03-08 00:39:43 +01004519 .bdrv_parse_filename = cdrom_parse_filename,
Paolo Bonzini44b424d2022-11-24 16:22:22 +01004520 .bdrv_open = cdrom_open,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004521 .bdrv_close = raw_close,
Jeff Cody1bc6b702012-11-20 10:21:10 -05004522 .bdrv_reopen_prepare = raw_reopen_prepare,
4523 .bdrv_reopen_commit = raw_reopen_commit,
4524 .bdrv_reopen_abort = raw_reopen_abort,
Maxim Levitsky5a5e7f82020-03-26 03:12:18 +02004525 .bdrv_co_create_opts = bdrv_co_create_opts_simple,
4526 .create_opts = &bdrv_create_opts_simple,
Alberto Garcia8a2ce0b2019-03-12 18:48:48 +02004527 .mutable_opts = mutable_opts,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004528
Kevin Wolf9d52aa32016-06-03 17:36:27 +02004529 .bdrv_co_preadv = raw_co_preadv,
4530 .bdrv_co_pwritev = raw_co_pwritev,
Kevin Wolf33d70fb2018-06-21 19:07:32 +02004531 .bdrv_co_flush_to_disk = raw_co_flush_to_disk,
Paolo Bonzini8c6f27e2023-04-07 17:32:59 +02004532 .bdrv_refresh_limits = cdrom_refresh_limits,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004533
Emanuele Giuseppe Espositoc86422c2023-01-13 21:42:04 +01004534 .bdrv_co_truncate = raw_co_truncate,
4535 .bdrv_co_getlength = raw_co_getlength,
Emanuele Giuseppe Esposito82618d72023-01-13 21:42:07 +01004536 .bdrv_co_get_allocated_file_size = raw_co_get_allocated_file_size,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004537
4538 /* removable device support */
Emanuele Giuseppe Esposito1e97be92023-01-13 21:42:02 +01004539 .bdrv_co_is_inserted = cdrom_co_is_inserted,
Emanuele Giuseppe Esposito2531b392023-01-13 21:42:09 +01004540 .bdrv_co_eject = cdrom_co_eject,
Emanuele Giuseppe Esposito2c752612023-01-13 21:42:10 +01004541 .bdrv_co_lock_medium = cdrom_co_lock_medium,
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004542};
4543#endif /* __FreeBSD__ */
4544
Joelle van Dyne14176c82021-03-15 11:03:38 -07004545#endif /* HAVE_HOST_BLOCK_DEVICE */
4546
Christoph Hellwig84a12e62010-04-07 22:30:24 +02004547static void bdrv_file_init(void)
Anthony Liguori5efa9d52009-05-09 17:03:42 -05004548{
Christoph Hellwig508c7cb2009-06-15 14:04:22 +02004549 /*
4550 * Register all the drivers. Note that order is important, the driver
4551 * registered last will get probed first.
4552 */
Christoph Hellwig84a12e62010-04-07 22:30:24 +02004553 bdrv_register(&bdrv_file);
Joelle van Dyne14176c82021-03-15 11:03:38 -07004554#if defined(HAVE_HOST_BLOCK_DEVICE)
Anthony Liguori5efa9d52009-05-09 17:03:42 -05004555 bdrv_register(&bdrv_host_device);
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004556#ifdef __linux__
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004557 bdrv_register(&bdrv_host_cdrom);
4558#endif
Aurelien Jarnoa167ba52009-11-29 18:00:41 +01004559#if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
Christoph Hellwigf3a5d3f2009-06-15 13:55:19 +02004560 bdrv_register(&bdrv_host_cdrom);
4561#endif
Joelle van Dyne14176c82021-03-15 11:03:38 -07004562#endif /* HAVE_HOST_BLOCK_DEVICE */
Anthony Liguori5efa9d52009-05-09 17:03:42 -05004563}
4564
Christoph Hellwig84a12e62010-04-07 22:30:24 +02004565block_init(bdrv_file_init);