1c1bb86cdSEric Blake /* 2c1bb86cdSEric Blake * Block driver for RAW files (posix) 3c1bb86cdSEric Blake * 4c1bb86cdSEric Blake * Copyright (c) 2006 Fabrice Bellard 5c1bb86cdSEric Blake * 6c1bb86cdSEric Blake * Permission is hereby granted, free of charge, to any person obtaining a copy 7c1bb86cdSEric Blake * of this software and associated documentation files (the "Software"), to deal 8c1bb86cdSEric Blake * in the Software without restriction, including without limitation the rights 9c1bb86cdSEric Blake * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10c1bb86cdSEric Blake * copies of the Software, and to permit persons to whom the Software is 11c1bb86cdSEric Blake * furnished to do so, subject to the following conditions: 12c1bb86cdSEric Blake * 13c1bb86cdSEric Blake * The above copyright notice and this permission notice shall be included in 14c1bb86cdSEric Blake * all copies or substantial portions of the Software. 15c1bb86cdSEric Blake * 16c1bb86cdSEric Blake * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17c1bb86cdSEric Blake * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18c1bb86cdSEric Blake * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19c1bb86cdSEric Blake * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20c1bb86cdSEric Blake * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21c1bb86cdSEric Blake * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22c1bb86cdSEric Blake * THE SOFTWARE. 23c1bb86cdSEric Blake */ 24922a01a0SMarkus Armbruster 25c1bb86cdSEric Blake #include "qemu/osdep.h" 26a8d25326SMarkus Armbruster #include "qemu-common.h" 27c1bb86cdSEric Blake #include "qapi/error.h" 28c1bb86cdSEric Blake #include "qemu/cutils.h" 29c1bb86cdSEric Blake #include "qemu/error-report.h" 30c1bb86cdSEric Blake #include "block/block_int.h" 31c1bb86cdSEric Blake #include "qemu/module.h" 32922a01a0SMarkus Armbruster #include "qemu/option.h" 33c1bb86cdSEric Blake #include "trace.h" 34c1bb86cdSEric Blake #include "block/thread-pool.h" 35c1bb86cdSEric Blake #include "qemu/iov.h" 36c1bb86cdSEric Blake #include "block/raw-aio.h" 37452fcdbcSMarkus Armbruster #include "qapi/qmp/qdict.h" 38c1bb86cdSEric Blake #include "qapi/qmp/qstring.h" 39c1bb86cdSEric Blake 407c9e5276SPaolo Bonzini #include "scsi/pr-manager.h" 417c9e5276SPaolo Bonzini #include "scsi/constants.h" 427c9e5276SPaolo Bonzini 43c1bb86cdSEric Blake #if defined(__APPLE__) && (__MACH__) 44c1bb86cdSEric Blake #include <paths.h> 45c1bb86cdSEric Blake #include <sys/param.h> 46c1bb86cdSEric Blake #include <IOKit/IOKitLib.h> 47c1bb86cdSEric Blake #include <IOKit/IOBSD.h> 48c1bb86cdSEric Blake #include <IOKit/storage/IOMediaBSDClient.h> 49c1bb86cdSEric Blake #include <IOKit/storage/IOMedia.h> 50c1bb86cdSEric Blake #include <IOKit/storage/IOCDMedia.h> 51c1bb86cdSEric Blake //#include <IOKit/storage/IOCDTypes.h> 52c1bb86cdSEric Blake #include <IOKit/storage/IODVDMedia.h> 53c1bb86cdSEric Blake #include <CoreFoundation/CoreFoundation.h> 54c1bb86cdSEric Blake #endif 55c1bb86cdSEric Blake 56c1bb86cdSEric Blake #ifdef __sun__ 57c1bb86cdSEric Blake #define _POSIX_PTHREAD_SEMANTICS 1 58c1bb86cdSEric Blake #include <sys/dkio.h> 59c1bb86cdSEric Blake #endif 60c1bb86cdSEric Blake #ifdef __linux__ 61c1bb86cdSEric Blake #include <sys/ioctl.h> 62c1bb86cdSEric Blake #include <sys/param.h> 631efad060SFam Zheng #include <sys/syscall.h> 64c1bb86cdSEric Blake #include <linux/cdrom.h> 65c1bb86cdSEric Blake #include <linux/fd.h> 66c1bb86cdSEric Blake #include <linux/fs.h> 67c1bb86cdSEric Blake #include <linux/hdreg.h> 68c1bb86cdSEric Blake #include <scsi/sg.h> 69c1bb86cdSEric Blake #ifdef __s390__ 70c1bb86cdSEric Blake #include <asm/dasd.h> 71c1bb86cdSEric Blake #endif 72c1bb86cdSEric Blake #ifndef FS_NOCOW_FL 73c1bb86cdSEric Blake #define FS_NOCOW_FL 0x00800000 /* Do not cow file */ 74c1bb86cdSEric Blake #endif 75c1bb86cdSEric Blake #endif 76c1bb86cdSEric Blake #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE) 77c1bb86cdSEric Blake #include <linux/falloc.h> 78c1bb86cdSEric Blake #endif 79c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 80c1bb86cdSEric Blake #include <sys/disk.h> 81c1bb86cdSEric Blake #include <sys/cdio.h> 82c1bb86cdSEric Blake #endif 83c1bb86cdSEric Blake 84c1bb86cdSEric Blake #ifdef __OpenBSD__ 85c1bb86cdSEric Blake #include <sys/ioctl.h> 86c1bb86cdSEric Blake #include <sys/disklabel.h> 87c1bb86cdSEric Blake #include <sys/dkio.h> 88c1bb86cdSEric Blake #endif 89c1bb86cdSEric Blake 90c1bb86cdSEric Blake #ifdef __NetBSD__ 91c1bb86cdSEric Blake #include <sys/ioctl.h> 92c1bb86cdSEric Blake #include <sys/disklabel.h> 93c1bb86cdSEric Blake #include <sys/dkio.h> 94c1bb86cdSEric Blake #include <sys/disk.h> 95c1bb86cdSEric Blake #endif 96c1bb86cdSEric Blake 97c1bb86cdSEric Blake #ifdef __DragonFly__ 98c1bb86cdSEric Blake #include <sys/ioctl.h> 99c1bb86cdSEric Blake #include <sys/diskslice.h> 100c1bb86cdSEric Blake #endif 101c1bb86cdSEric Blake 102c1bb86cdSEric Blake #ifdef CONFIG_XFS 103c1bb86cdSEric Blake #include <xfs/xfs.h> 104c1bb86cdSEric Blake #endif 105c1bb86cdSEric Blake 1064f7d28d7SLaurent Vivier #include "trace.h" 107c1bb86cdSEric Blake 108c1bb86cdSEric Blake /* OS X does not have O_DSYNC */ 109c1bb86cdSEric Blake #ifndef O_DSYNC 110c1bb86cdSEric Blake #ifdef O_SYNC 111c1bb86cdSEric Blake #define O_DSYNC O_SYNC 112c1bb86cdSEric Blake #elif defined(O_FSYNC) 113c1bb86cdSEric Blake #define O_DSYNC O_FSYNC 114c1bb86cdSEric Blake #endif 115c1bb86cdSEric Blake #endif 116c1bb86cdSEric Blake 117c1bb86cdSEric Blake /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ 118c1bb86cdSEric Blake #ifndef O_DIRECT 119c1bb86cdSEric Blake #define O_DIRECT O_DSYNC 120c1bb86cdSEric Blake #endif 121c1bb86cdSEric Blake 122c1bb86cdSEric Blake #define FTYPE_FILE 0 123c1bb86cdSEric Blake #define FTYPE_CD 1 124c1bb86cdSEric Blake 125c1bb86cdSEric Blake #define MAX_BLOCKSIZE 4096 126c1bb86cdSEric Blake 127244a5668SFam Zheng /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes, 128244a5668SFam Zheng * leaving a few more bytes for its future use. */ 129244a5668SFam Zheng #define RAW_LOCK_PERM_BASE 100 130244a5668SFam Zheng #define RAW_LOCK_SHARED_BASE 200 131244a5668SFam Zheng 132c1bb86cdSEric Blake typedef struct BDRVRawState { 133c1bb86cdSEric Blake int fd; 134244a5668SFam Zheng bool use_lock; 135c1bb86cdSEric Blake int type; 136c1bb86cdSEric Blake int open_flags; 137c1bb86cdSEric Blake size_t buf_align; 138c1bb86cdSEric Blake 139244a5668SFam Zheng /* The current permissions. */ 140244a5668SFam Zheng uint64_t perm; 141244a5668SFam Zheng uint64_t shared_perm; 142244a5668SFam Zheng 1432996ffadSFam Zheng /* The perms bits whose corresponding bytes are already locked in 144f2e3af29SFam Zheng * s->fd. */ 1452996ffadSFam Zheng uint64_t locked_perm; 1462996ffadSFam Zheng uint64_t locked_shared_perm; 1472996ffadSFam Zheng 1486ceabe6fSKevin Wolf int perm_change_fd; 149094e3639SMax Reitz int perm_change_flags; 150e0c9cf3aSKevin Wolf BDRVReopenState *reopen_state; 151e0c9cf3aSKevin Wolf 152c1bb86cdSEric Blake #ifdef CONFIG_XFS 153c1bb86cdSEric Blake bool is_xfs:1; 154c1bb86cdSEric Blake #endif 155c1bb86cdSEric Blake bool has_discard:1; 156c1bb86cdSEric Blake bool has_write_zeroes:1; 157c1bb86cdSEric Blake bool discard_zeroes:1; 158c1bb86cdSEric Blake bool use_linux_aio:1; 159c6447510SAarushi Mehta bool use_linux_io_uring:1; 160e5bcf967SKevin Wolf bool page_cache_inconsistent:1; 161c1bb86cdSEric Blake bool has_fallocate; 162c1bb86cdSEric Blake bool needs_alignment; 163f357fcd8SStefan Hajnoczi bool drop_cache; 16431be8a2aSStefan Hajnoczi bool check_cache_dropped; 1651c450366SAnton Nefedov struct { 1661c450366SAnton Nefedov uint64_t discard_nb_ok; 1671c450366SAnton Nefedov uint64_t discard_nb_failed; 1681c450366SAnton Nefedov uint64_t discard_bytes_ok; 1691c450366SAnton Nefedov } stats; 1707c9e5276SPaolo Bonzini 1717c9e5276SPaolo Bonzini PRManager *pr_mgr; 172c1bb86cdSEric Blake } BDRVRawState; 173c1bb86cdSEric Blake 174c1bb86cdSEric Blake typedef struct BDRVRawReopenState { 175c1bb86cdSEric Blake int fd; 176c1bb86cdSEric Blake int open_flags; 177f357fcd8SStefan Hajnoczi bool drop_cache; 17831be8a2aSStefan Hajnoczi bool check_cache_dropped; 179c1bb86cdSEric Blake } BDRVRawReopenState; 180c1bb86cdSEric Blake 181c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs); 182c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs); 183c1bb86cdSEric Blake 184c1bb86cdSEric Blake typedef struct RawPosixAIOData { 185c1bb86cdSEric Blake BlockDriverState *bs; 186c1bb86cdSEric Blake int aio_type; 187d57c44d0SKevin Wolf int aio_fildes; 188d57c44d0SKevin Wolf 189d57c44d0SKevin Wolf off_t aio_offset; 190d57c44d0SKevin Wolf uint64_t aio_nbytes; 191d57c44d0SKevin Wolf 19293f4e2ffSKevin Wolf union { 19393f4e2ffSKevin Wolf struct { 194d57c44d0SKevin Wolf struct iovec *iov; 195d57c44d0SKevin Wolf int niov; 196d57c44d0SKevin Wolf } io; 197d57c44d0SKevin Wolf struct { 198d57c44d0SKevin Wolf uint64_t cmd; 199d57c44d0SKevin Wolf void *buf; 200d57c44d0SKevin Wolf } ioctl; 201d57c44d0SKevin Wolf struct { 2021efad060SFam Zheng int aio_fd2; 2031efad060SFam Zheng off_t aio_offset2; 204d57c44d0SKevin Wolf } copy_range; 20593f4e2ffSKevin Wolf struct { 20693f4e2ffSKevin Wolf PreallocMode prealloc; 20793f4e2ffSKevin Wolf Error **errp; 208d57c44d0SKevin Wolf } truncate; 20993f4e2ffSKevin Wolf }; 210c1bb86cdSEric Blake } RawPosixAIOData; 211c1bb86cdSEric Blake 212c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 213c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs); 214c1bb86cdSEric Blake #endif 215c1bb86cdSEric Blake 216c1bb86cdSEric Blake #if defined(__NetBSD__) 217db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 218c1bb86cdSEric Blake { 219c1bb86cdSEric Blake static char namebuf[PATH_MAX]; 220c1bb86cdSEric Blake const char *dp, *fname; 221c1bb86cdSEric Blake struct stat sb; 222c1bb86cdSEric Blake 223c1bb86cdSEric Blake fname = *filename; 224c1bb86cdSEric Blake dp = strrchr(fname, '/'); 225c1bb86cdSEric Blake if (lstat(fname, &sb) < 0) { 226f6fc1e30SPaolo Bonzini error_setg_file_open(errp, errno, fname); 227c1bb86cdSEric Blake return -errno; 228c1bb86cdSEric Blake } 229c1bb86cdSEric Blake 230c1bb86cdSEric Blake if (!S_ISBLK(sb.st_mode)) { 231c1bb86cdSEric Blake return 0; 232c1bb86cdSEric Blake } 233c1bb86cdSEric Blake 234c1bb86cdSEric Blake if (dp == NULL) { 235c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "r%s", fname); 236c1bb86cdSEric Blake } else { 237c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "%.*s/r%s", 238c1bb86cdSEric Blake (int)(dp - fname), fname, dp + 1); 239c1bb86cdSEric Blake } 240c1bb86cdSEric Blake *filename = namebuf; 241db0754dfSFam Zheng warn_report("%s is a block device, using %s", fname, *filename); 242c1bb86cdSEric Blake 243c1bb86cdSEric Blake return 0; 244c1bb86cdSEric Blake } 245c1bb86cdSEric Blake #else 246db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 247c1bb86cdSEric Blake { 248c1bb86cdSEric Blake return 0; 249c1bb86cdSEric Blake } 250c1bb86cdSEric Blake #endif 251c1bb86cdSEric Blake 252c1bb86cdSEric Blake /* 253c1bb86cdSEric Blake * Get logical block size via ioctl. On success store it in @sector_size_p. 254c1bb86cdSEric Blake */ 255c1bb86cdSEric Blake static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) 256c1bb86cdSEric Blake { 257c1bb86cdSEric Blake unsigned int sector_size; 258c1bb86cdSEric Blake bool success = false; 259700f9ce0SPeter Maydell int i; 260c1bb86cdSEric Blake 261c1bb86cdSEric Blake errno = ENOTSUP; 262700f9ce0SPeter Maydell static const unsigned long ioctl_list[] = { 263c1bb86cdSEric Blake #ifdef BLKSSZGET 264700f9ce0SPeter Maydell BLKSSZGET, 265c1bb86cdSEric Blake #endif 266c1bb86cdSEric Blake #ifdef DKIOCGETBLOCKSIZE 267700f9ce0SPeter Maydell DKIOCGETBLOCKSIZE, 268c1bb86cdSEric Blake #endif 269c1bb86cdSEric Blake #ifdef DIOCGSECTORSIZE 270700f9ce0SPeter Maydell DIOCGSECTORSIZE, 271700f9ce0SPeter Maydell #endif 272700f9ce0SPeter Maydell }; 273700f9ce0SPeter Maydell 274700f9ce0SPeter Maydell /* Try a few ioctls to get the right size */ 275700f9ce0SPeter Maydell for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { 276700f9ce0SPeter Maydell if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { 277c1bb86cdSEric Blake *sector_size_p = sector_size; 278c1bb86cdSEric Blake success = true; 279c1bb86cdSEric Blake } 280700f9ce0SPeter Maydell } 281c1bb86cdSEric Blake 282c1bb86cdSEric Blake return success ? 0 : -errno; 283c1bb86cdSEric Blake } 284c1bb86cdSEric Blake 285c1bb86cdSEric Blake /** 286c1bb86cdSEric Blake * Get physical block size of @fd. 287c1bb86cdSEric Blake * On success, store it in @blk_size and return 0. 288c1bb86cdSEric Blake * On failure, return -errno. 289c1bb86cdSEric Blake */ 290c1bb86cdSEric Blake static int probe_physical_blocksize(int fd, unsigned int *blk_size) 291c1bb86cdSEric Blake { 292c1bb86cdSEric Blake #ifdef BLKPBSZGET 293c1bb86cdSEric Blake if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { 294c1bb86cdSEric Blake return -errno; 295c1bb86cdSEric Blake } 296c1bb86cdSEric Blake return 0; 297c1bb86cdSEric Blake #else 298c1bb86cdSEric Blake return -ENOTSUP; 299c1bb86cdSEric Blake #endif 300c1bb86cdSEric Blake } 301c1bb86cdSEric Blake 302c1bb86cdSEric Blake /* Check if read is allowed with given memory buffer and length. 303c1bb86cdSEric Blake * 304c1bb86cdSEric Blake * This function is used to check O_DIRECT memory buffer and request alignment. 305c1bb86cdSEric Blake */ 306c1bb86cdSEric Blake static bool raw_is_io_aligned(int fd, void *buf, size_t len) 307c1bb86cdSEric Blake { 308c1bb86cdSEric Blake ssize_t ret = pread(fd, buf, len, 0); 309c1bb86cdSEric Blake 310c1bb86cdSEric Blake if (ret >= 0) { 311c1bb86cdSEric Blake return true; 312c1bb86cdSEric Blake } 313c1bb86cdSEric Blake 314c1bb86cdSEric Blake #ifdef __linux__ 315c1bb86cdSEric Blake /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore 316c1bb86cdSEric Blake * other errors (e.g. real I/O error), which could happen on a failed 317c1bb86cdSEric Blake * drive, since we only care about probing alignment. 318c1bb86cdSEric Blake */ 319c1bb86cdSEric Blake if (errno != EINVAL) { 320c1bb86cdSEric Blake return true; 321c1bb86cdSEric Blake } 322c1bb86cdSEric Blake #endif 323c1bb86cdSEric Blake 324c1bb86cdSEric Blake return false; 325c1bb86cdSEric Blake } 326c1bb86cdSEric Blake 327c1bb86cdSEric Blake static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) 328c1bb86cdSEric Blake { 329c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 330c1bb86cdSEric Blake char *buf; 331038adc2fSWei Yang size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); 332a6b257a0SNir Soffer size_t alignments[] = {1, 512, 1024, 2048, 4096}; 333c1bb86cdSEric Blake 334c1bb86cdSEric Blake /* For SCSI generic devices the alignment is not really used. 335c1bb86cdSEric Blake With buffered I/O, we don't have any restrictions. */ 336c1bb86cdSEric Blake if (bdrv_is_sg(bs) || !s->needs_alignment) { 337c1bb86cdSEric Blake bs->bl.request_alignment = 1; 338c1bb86cdSEric Blake s->buf_align = 1; 339c1bb86cdSEric Blake return; 340c1bb86cdSEric Blake } 341c1bb86cdSEric Blake 342c1bb86cdSEric Blake bs->bl.request_alignment = 0; 343c1bb86cdSEric Blake s->buf_align = 0; 344c1bb86cdSEric Blake /* Let's try to use the logical blocksize for the alignment. */ 345c1bb86cdSEric Blake if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { 346c1bb86cdSEric Blake bs->bl.request_alignment = 0; 347c1bb86cdSEric Blake } 348c1bb86cdSEric Blake #ifdef CONFIG_XFS 349c1bb86cdSEric Blake if (s->is_xfs) { 350c1bb86cdSEric Blake struct dioattr da; 351c1bb86cdSEric Blake if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { 352c1bb86cdSEric Blake bs->bl.request_alignment = da.d_miniosz; 353c1bb86cdSEric Blake /* The kernel returns wrong information for d_mem */ 354c1bb86cdSEric Blake /* s->buf_align = da.d_mem; */ 355c1bb86cdSEric Blake } 356c1bb86cdSEric Blake } 357c1bb86cdSEric Blake #endif 358c1bb86cdSEric Blake 359a6b257a0SNir Soffer /* 360a6b257a0SNir Soffer * If we could not get the sizes so far, we can only guess them. First try 361a6b257a0SNir Soffer * to detect request alignment, since it is more likely to succeed. Then 362a6b257a0SNir Soffer * try to detect buf_align, which cannot be detected in some cases (e.g. 363a6b257a0SNir Soffer * Gluster). If buf_align cannot be detected, we fallback to the value of 364a6b257a0SNir Soffer * request_alignment. 365a6b257a0SNir Soffer */ 366a6b257a0SNir Soffer 367a6b257a0SNir Soffer if (!bs->bl.request_alignment) { 368a6b257a0SNir Soffer int i; 369c1bb86cdSEric Blake size_t align; 370a6b257a0SNir Soffer buf = qemu_memalign(max_align, max_align); 371a6b257a0SNir Soffer for (i = 0; i < ARRAY_SIZE(alignments); i++) { 372a6b257a0SNir Soffer align = alignments[i]; 373a6b257a0SNir Soffer if (raw_is_io_aligned(fd, buf, align)) { 374a6b257a0SNir Soffer /* Fallback to safe value. */ 375a6b257a0SNir Soffer bs->bl.request_alignment = (align != 1) ? align : max_align; 376c1bb86cdSEric Blake break; 377c1bb86cdSEric Blake } 378c1bb86cdSEric Blake } 379c1bb86cdSEric Blake qemu_vfree(buf); 380c1bb86cdSEric Blake } 381c1bb86cdSEric Blake 382a6b257a0SNir Soffer if (!s->buf_align) { 383a6b257a0SNir Soffer int i; 384c1bb86cdSEric Blake size_t align; 385a6b257a0SNir Soffer buf = qemu_memalign(max_align, 2 * max_align); 386a6b257a0SNir Soffer for (i = 0; i < ARRAY_SIZE(alignments); i++) { 387a6b257a0SNir Soffer align = alignments[i]; 388a6b257a0SNir Soffer if (raw_is_io_aligned(fd, buf + align, max_align)) { 389236094c7SStefan Hajnoczi /* Fallback to request_alignment. */ 390a6b257a0SNir Soffer s->buf_align = (align != 1) ? align : bs->bl.request_alignment; 391c1bb86cdSEric Blake break; 392c1bb86cdSEric Blake } 393c1bb86cdSEric Blake } 394c1bb86cdSEric Blake qemu_vfree(buf); 395c1bb86cdSEric Blake } 396c1bb86cdSEric Blake 397c1bb86cdSEric Blake if (!s->buf_align || !bs->bl.request_alignment) { 398c1bb86cdSEric Blake error_setg(errp, "Could not find working O_DIRECT alignment"); 399c1bb86cdSEric Blake error_append_hint(errp, "Try cache.direct=off\n"); 400c1bb86cdSEric Blake } 401c1bb86cdSEric Blake } 402c1bb86cdSEric Blake 40323dece19SKevin Wolf static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) 404c1bb86cdSEric Blake { 40523dece19SKevin Wolf bool read_write = false; 406c1bb86cdSEric Blake assert(open_flags != NULL); 407c1bb86cdSEric Blake 408c1bb86cdSEric Blake *open_flags |= O_BINARY; 409c1bb86cdSEric Blake *open_flags &= ~O_ACCMODE; 41023dece19SKevin Wolf 41123dece19SKevin Wolf if (bdrv_flags & BDRV_O_AUTO_RDONLY) { 41223dece19SKevin Wolf read_write = has_writers; 41323dece19SKevin Wolf } else if (bdrv_flags & BDRV_O_RDWR) { 41423dece19SKevin Wolf read_write = true; 41523dece19SKevin Wolf } 41623dece19SKevin Wolf 41723dece19SKevin Wolf if (read_write) { 418c1bb86cdSEric Blake *open_flags |= O_RDWR; 419c1bb86cdSEric Blake } else { 420c1bb86cdSEric Blake *open_flags |= O_RDONLY; 421c1bb86cdSEric Blake } 422c1bb86cdSEric Blake 423c1bb86cdSEric Blake /* Use O_DSYNC for write-through caching, no flags for write-back caching, 424c1bb86cdSEric Blake * and O_DIRECT for no caching. */ 425c1bb86cdSEric Blake if ((bdrv_flags & BDRV_O_NOCACHE)) { 426c1bb86cdSEric Blake *open_flags |= O_DIRECT; 427c1bb86cdSEric Blake } 428c1bb86cdSEric Blake } 429c1bb86cdSEric Blake 430c1bb86cdSEric Blake static void raw_parse_filename(const char *filename, QDict *options, 431c1bb86cdSEric Blake Error **errp) 432c1bb86cdSEric Blake { 43303c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "file:", options); 434c1bb86cdSEric Blake } 435c1bb86cdSEric Blake 436c1bb86cdSEric Blake static QemuOptsList raw_runtime_opts = { 437c1bb86cdSEric Blake .name = "raw", 438c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), 439c1bb86cdSEric Blake .desc = { 440c1bb86cdSEric Blake { 441c1bb86cdSEric Blake .name = "filename", 442c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 443c1bb86cdSEric Blake .help = "File name of the image", 444c1bb86cdSEric Blake }, 445c1bb86cdSEric Blake { 446c1bb86cdSEric Blake .name = "aio", 447c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 448c6447510SAarushi Mehta .help = "host AIO implementation (threads, native, io_uring)", 449c1bb86cdSEric Blake }, 45016b48d5dSFam Zheng { 45116b48d5dSFam Zheng .name = "locking", 45216b48d5dSFam Zheng .type = QEMU_OPT_STRING, 45316b48d5dSFam Zheng .help = "file locking mode (on/off/auto, default: auto)", 45416b48d5dSFam Zheng }, 4557c9e5276SPaolo Bonzini { 4567c9e5276SPaolo Bonzini .name = "pr-manager", 4577c9e5276SPaolo Bonzini .type = QEMU_OPT_STRING, 4587c9e5276SPaolo Bonzini .help = "id of persistent reservation manager object (default: none)", 4597c9e5276SPaolo Bonzini }, 460f357fcd8SStefan Hajnoczi #if defined(__linux__) 461f357fcd8SStefan Hajnoczi { 462f357fcd8SStefan Hajnoczi .name = "drop-cache", 463f357fcd8SStefan Hajnoczi .type = QEMU_OPT_BOOL, 464f357fcd8SStefan Hajnoczi .help = "invalidate page cache during live migration (default: on)", 465f357fcd8SStefan Hajnoczi }, 466f357fcd8SStefan Hajnoczi #endif 46731be8a2aSStefan Hajnoczi { 46831be8a2aSStefan Hajnoczi .name = "x-check-cache-dropped", 46931be8a2aSStefan Hajnoczi .type = QEMU_OPT_BOOL, 47031be8a2aSStefan Hajnoczi .help = "check that page cache was dropped on live migration (default: off)" 47131be8a2aSStefan Hajnoczi }, 472c1bb86cdSEric Blake { /* end of list */ } 473c1bb86cdSEric Blake }, 474c1bb86cdSEric Blake }; 475c1bb86cdSEric Blake 4768a2ce0bcSAlberto Garcia static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; 4778a2ce0bcSAlberto Garcia 478c1bb86cdSEric Blake static int raw_open_common(BlockDriverState *bs, QDict *options, 479230ff739SJohn Snow int bdrv_flags, int open_flags, 480230ff739SJohn Snow bool device, Error **errp) 481c1bb86cdSEric Blake { 482c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 483c1bb86cdSEric Blake QemuOpts *opts; 484c1bb86cdSEric Blake Error *local_err = NULL; 485c1bb86cdSEric Blake const char *filename = NULL; 4867c9e5276SPaolo Bonzini const char *str; 487c1bb86cdSEric Blake BlockdevAioOptions aio, aio_default; 488c1bb86cdSEric Blake int fd, ret; 489c1bb86cdSEric Blake struct stat st; 490244a5668SFam Zheng OnOffAuto locking; 491c1bb86cdSEric Blake 492c1bb86cdSEric Blake opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 493c1bb86cdSEric Blake qemu_opts_absorb_qdict(opts, options, &local_err); 494c1bb86cdSEric Blake if (local_err) { 495c1bb86cdSEric Blake error_propagate(errp, local_err); 496c1bb86cdSEric Blake ret = -EINVAL; 497c1bb86cdSEric Blake goto fail; 498c1bb86cdSEric Blake } 499c1bb86cdSEric Blake 500c1bb86cdSEric Blake filename = qemu_opt_get(opts, "filename"); 501c1bb86cdSEric Blake 502db0754dfSFam Zheng ret = raw_normalize_devicepath(&filename, errp); 503c1bb86cdSEric Blake if (ret != 0) { 504c1bb86cdSEric Blake goto fail; 505c1bb86cdSEric Blake } 506c1bb86cdSEric Blake 507c6447510SAarushi Mehta if (bdrv_flags & BDRV_O_NATIVE_AIO) { 508c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE; 509c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 510c6447510SAarushi Mehta } else if (bdrv_flags & BDRV_O_IO_URING) { 511c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING; 512c6447510SAarushi Mehta #endif 513c6447510SAarushi Mehta } else { 514c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_THREADS; 515c6447510SAarushi Mehta } 516c6447510SAarushi Mehta 517f7abe0ecSMarc-André Lureau aio = qapi_enum_parse(&BlockdevAioOptions_lookup, 518f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "aio"), 51906c60b6cSMarkus Armbruster aio_default, &local_err); 520c1bb86cdSEric Blake if (local_err) { 521c1bb86cdSEric Blake error_propagate(errp, local_err); 522c1bb86cdSEric Blake ret = -EINVAL; 523c1bb86cdSEric Blake goto fail; 524c1bb86cdSEric Blake } 525c6447510SAarushi Mehta 526c1bb86cdSEric Blake s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); 527c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 528c6447510SAarushi Mehta s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING); 529c6447510SAarushi Mehta #endif 530c1bb86cdSEric Blake 531f7abe0ecSMarc-André Lureau locking = qapi_enum_parse(&OnOffAuto_lookup, 532f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "locking"), 53306c60b6cSMarkus Armbruster ON_OFF_AUTO_AUTO, &local_err); 534244a5668SFam Zheng if (local_err) { 535244a5668SFam Zheng error_propagate(errp, local_err); 536244a5668SFam Zheng ret = -EINVAL; 537244a5668SFam Zheng goto fail; 538244a5668SFam Zheng } 539244a5668SFam Zheng switch (locking) { 540244a5668SFam Zheng case ON_OFF_AUTO_ON: 541244a5668SFam Zheng s->use_lock = true; 5422b218f5dSFam Zheng if (!qemu_has_ofd_lock()) { 543db0754dfSFam Zheng warn_report("File lock requested but OFD locking syscall is " 544db0754dfSFam Zheng "unavailable, falling back to POSIX file locks"); 545db0754dfSFam Zheng error_printf("Due to the implementation, locks can be lost " 5462b218f5dSFam Zheng "unexpectedly.\n"); 5472b218f5dSFam Zheng } 548244a5668SFam Zheng break; 549244a5668SFam Zheng case ON_OFF_AUTO_OFF: 550244a5668SFam Zheng s->use_lock = false; 551244a5668SFam Zheng break; 552244a5668SFam Zheng case ON_OFF_AUTO_AUTO: 5532b218f5dSFam Zheng s->use_lock = qemu_has_ofd_lock(); 554244a5668SFam Zheng break; 555244a5668SFam Zheng default: 556244a5668SFam Zheng abort(); 557244a5668SFam Zheng } 558244a5668SFam Zheng 5597c9e5276SPaolo Bonzini str = qemu_opt_get(opts, "pr-manager"); 5607c9e5276SPaolo Bonzini if (str) { 5617c9e5276SPaolo Bonzini s->pr_mgr = pr_manager_lookup(str, &local_err); 5627c9e5276SPaolo Bonzini if (local_err) { 5637c9e5276SPaolo Bonzini error_propagate(errp, local_err); 5647c9e5276SPaolo Bonzini ret = -EINVAL; 5657c9e5276SPaolo Bonzini goto fail; 5667c9e5276SPaolo Bonzini } 5677c9e5276SPaolo Bonzini } 5687c9e5276SPaolo Bonzini 569f357fcd8SStefan Hajnoczi s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); 57031be8a2aSStefan Hajnoczi s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", 57131be8a2aSStefan Hajnoczi false); 57231be8a2aSStefan Hajnoczi 573c1bb86cdSEric Blake s->open_flags = open_flags; 57423dece19SKevin Wolf raw_parse_flags(bdrv_flags, &s->open_flags, false); 575c1bb86cdSEric Blake 576c1bb86cdSEric Blake s->fd = -1; 577c1bb86cdSEric Blake fd = qemu_open(filename, s->open_flags, 0644); 57864107dc0SKevin Wolf ret = fd < 0 ? -errno : 0; 57964107dc0SKevin Wolf 58064107dc0SKevin Wolf if (ret < 0) { 581f6fc1e30SPaolo Bonzini error_setg_file_open(errp, -ret, filename); 582c1bb86cdSEric Blake if (ret == -EROFS) { 583c1bb86cdSEric Blake ret = -EACCES; 584c1bb86cdSEric Blake } 585c1bb86cdSEric Blake goto fail; 586c1bb86cdSEric Blake } 587c1bb86cdSEric Blake s->fd = fd; 588c1bb86cdSEric Blake 589244a5668SFam Zheng s->perm = 0; 590244a5668SFam Zheng s->shared_perm = BLK_PERM_ALL; 591244a5668SFam Zheng 592c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 593c1bb86cdSEric Blake /* Currently Linux does AIO only for files opened with O_DIRECT */ 594ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 595ed6e2161SNishanth Aravamudan if (!(s->open_flags & O_DIRECT)) { 596c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but it requires " 597c1bb86cdSEric Blake "cache.direct=on, which was not specified."); 598c1bb86cdSEric Blake ret = -EINVAL; 599c1bb86cdSEric Blake goto fail; 600c1bb86cdSEric Blake } 601ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { 602ed6e2161SNishanth Aravamudan error_prepend(errp, "Unable to use native AIO: "); 603ed6e2161SNishanth Aravamudan goto fail; 604ed6e2161SNishanth Aravamudan } 605ed6e2161SNishanth Aravamudan } 606c1bb86cdSEric Blake #else 607c1bb86cdSEric Blake if (s->use_linux_aio) { 608c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but is not supported " 609c1bb86cdSEric Blake "in this build."); 610c1bb86cdSEric Blake ret = -EINVAL; 611c1bb86cdSEric Blake goto fail; 612c1bb86cdSEric Blake } 613c1bb86cdSEric Blake #endif /* !defined(CONFIG_LINUX_AIO) */ 614c1bb86cdSEric Blake 615c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 616c6447510SAarushi Mehta if (s->use_linux_io_uring) { 617c6447510SAarushi Mehta if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs), errp)) { 618c6447510SAarushi Mehta error_prepend(errp, "Unable to use io_uring: "); 619c6447510SAarushi Mehta goto fail; 620c6447510SAarushi Mehta } 621c6447510SAarushi Mehta } 622c6447510SAarushi Mehta #else 623c6447510SAarushi Mehta if (s->use_linux_io_uring) { 624c6447510SAarushi Mehta error_setg(errp, "aio=io_uring was specified, but is not supported " 625c6447510SAarushi Mehta "in this build."); 626c6447510SAarushi Mehta ret = -EINVAL; 627c6447510SAarushi Mehta goto fail; 628c6447510SAarushi Mehta } 629c6447510SAarushi Mehta #endif /* !defined(CONFIG_LINUX_IO_URING) */ 630c6447510SAarushi Mehta 631c1bb86cdSEric Blake s->has_discard = true; 632c1bb86cdSEric Blake s->has_write_zeroes = true; 633c1bb86cdSEric Blake if ((bs->open_flags & BDRV_O_NOCACHE) != 0) { 634c1bb86cdSEric Blake s->needs_alignment = true; 635c1bb86cdSEric Blake } 636c1bb86cdSEric Blake 637c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 638c1bb86cdSEric Blake ret = -errno; 639c1bb86cdSEric Blake error_setg_errno(errp, errno, "Could not stat file"); 640c1bb86cdSEric Blake goto fail; 641c1bb86cdSEric Blake } 642230ff739SJohn Snow 643230ff739SJohn Snow if (!device) { 644230ff739SJohn Snow if (S_ISBLK(st.st_mode)) { 645230ff739SJohn Snow warn_report("Opening a block device as a file using the '%s' " 646230ff739SJohn Snow "driver is deprecated", bs->drv->format_name); 647230ff739SJohn Snow } else if (S_ISCHR(st.st_mode)) { 648230ff739SJohn Snow warn_report("Opening a character device as a file using the '%s' " 649230ff739SJohn Snow "driver is deprecated", bs->drv->format_name); 650230ff739SJohn Snow } else if (!S_ISREG(st.st_mode)) { 651230ff739SJohn Snow error_setg(errp, "A regular file was expected by the '%s' driver, " 652230ff739SJohn Snow "but something else was given", bs->drv->format_name); 653230ff739SJohn Snow ret = -EINVAL; 654230ff739SJohn Snow goto fail; 655230ff739SJohn Snow } else { 656c1bb86cdSEric Blake s->discard_zeroes = true; 657c1bb86cdSEric Blake s->has_fallocate = true; 658c1bb86cdSEric Blake } 659230ff739SJohn Snow } else { 660230ff739SJohn Snow if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 661230ff739SJohn Snow error_setg(errp, "'%s' driver expects either " 662230ff739SJohn Snow "a character or block device", bs->drv->format_name); 663230ff739SJohn Snow ret = -EINVAL; 664230ff739SJohn Snow goto fail; 665230ff739SJohn Snow } 666230ff739SJohn Snow } 667230ff739SJohn Snow 668c1bb86cdSEric Blake if (S_ISBLK(st.st_mode)) { 669c1bb86cdSEric Blake #ifdef BLKDISCARDZEROES 670c1bb86cdSEric Blake unsigned int arg; 671c1bb86cdSEric Blake if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { 672c1bb86cdSEric Blake s->discard_zeroes = true; 673c1bb86cdSEric Blake } 674c1bb86cdSEric Blake #endif 675c1bb86cdSEric Blake #ifdef __linux__ 676c1bb86cdSEric Blake /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do 677c1bb86cdSEric Blake * not rely on the contents of discarded blocks unless using O_DIRECT. 678c1bb86cdSEric Blake * Same for BLKZEROOUT. 679c1bb86cdSEric Blake */ 680c1bb86cdSEric Blake if (!(bs->open_flags & BDRV_O_NOCACHE)) { 681c1bb86cdSEric Blake s->discard_zeroes = false; 682c1bb86cdSEric Blake s->has_write_zeroes = false; 683c1bb86cdSEric Blake } 684c1bb86cdSEric Blake #endif 685c1bb86cdSEric Blake } 686c1bb86cdSEric Blake #ifdef __FreeBSD__ 687c1bb86cdSEric Blake if (S_ISCHR(st.st_mode)) { 688c1bb86cdSEric Blake /* 689c1bb86cdSEric Blake * The file is a char device (disk), which on FreeBSD isn't behind 690c1bb86cdSEric Blake * a pager, so force all requests to be aligned. This is needed 691c1bb86cdSEric Blake * so QEMU makes sure all IO operations on the device are aligned 692c1bb86cdSEric Blake * to sector size, or else FreeBSD will reject them with EINVAL. 693c1bb86cdSEric Blake */ 694c1bb86cdSEric Blake s->needs_alignment = true; 695c1bb86cdSEric Blake } 696c1bb86cdSEric Blake #endif 697c1bb86cdSEric Blake 698c1bb86cdSEric Blake #ifdef CONFIG_XFS 699c1bb86cdSEric Blake if (platform_test_xfs_fd(s->fd)) { 700c1bb86cdSEric Blake s->is_xfs = true; 701c1bb86cdSEric Blake } 702c1bb86cdSEric Blake #endif 703c1bb86cdSEric Blake 704738301e1SKevin Wolf bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; 705c1bb86cdSEric Blake ret = 0; 706c1bb86cdSEric Blake fail: 707c1bb86cdSEric Blake if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { 708c1bb86cdSEric Blake unlink(filename); 709c1bb86cdSEric Blake } 710c1bb86cdSEric Blake qemu_opts_del(opts); 711c1bb86cdSEric Blake return ret; 712c1bb86cdSEric Blake } 713c1bb86cdSEric Blake 714c1bb86cdSEric Blake static int raw_open(BlockDriverState *bs, QDict *options, int flags, 715c1bb86cdSEric Blake Error **errp) 716c1bb86cdSEric Blake { 717c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 718c1bb86cdSEric Blake 719c1bb86cdSEric Blake s->type = FTYPE_FILE; 720230ff739SJohn Snow return raw_open_common(bs, options, flags, 0, false, errp); 721c1bb86cdSEric Blake } 722c1bb86cdSEric Blake 723244a5668SFam Zheng typedef enum { 724244a5668SFam Zheng RAW_PL_PREPARE, 725244a5668SFam Zheng RAW_PL_COMMIT, 726244a5668SFam Zheng RAW_PL_ABORT, 727244a5668SFam Zheng } RawPermLockOp; 728244a5668SFam Zheng 729244a5668SFam Zheng #define PERM_FOREACH(i) \ 730244a5668SFam Zheng for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) 731244a5668SFam Zheng 732244a5668SFam Zheng /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the 733244a5668SFam Zheng * file; if @unlock == true, also unlock the unneeded bytes. 734244a5668SFam Zheng * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. 735244a5668SFam Zheng */ 7362996ffadSFam Zheng static int raw_apply_lock_bytes(BDRVRawState *s, int fd, 737244a5668SFam Zheng uint64_t perm_lock_bits, 738244a5668SFam Zheng uint64_t shared_perm_lock_bits, 739244a5668SFam Zheng bool unlock, Error **errp) 740244a5668SFam Zheng { 741244a5668SFam Zheng int ret; 742244a5668SFam Zheng int i; 7432996ffadSFam Zheng uint64_t locked_perm, locked_shared_perm; 7442996ffadSFam Zheng 7452996ffadSFam Zheng if (s) { 7462996ffadSFam Zheng locked_perm = s->locked_perm; 7472996ffadSFam Zheng locked_shared_perm = s->locked_shared_perm; 7482996ffadSFam Zheng } else { 7492996ffadSFam Zheng /* 7502996ffadSFam Zheng * We don't have the previous bits, just lock/unlock for each of the 7512996ffadSFam Zheng * requested bits. 7522996ffadSFam Zheng */ 7532996ffadSFam Zheng if (unlock) { 7542996ffadSFam Zheng locked_perm = BLK_PERM_ALL; 7552996ffadSFam Zheng locked_shared_perm = BLK_PERM_ALL; 7562996ffadSFam Zheng } else { 7572996ffadSFam Zheng locked_perm = 0; 7582996ffadSFam Zheng locked_shared_perm = 0; 7592996ffadSFam Zheng } 7602996ffadSFam Zheng } 761244a5668SFam Zheng 762244a5668SFam Zheng PERM_FOREACH(i) { 763244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 7642996ffadSFam Zheng uint64_t bit = (1ULL << i); 7652996ffadSFam Zheng if ((perm_lock_bits & bit) && !(locked_perm & bit)) { 766d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 767244a5668SFam Zheng if (ret) { 768244a5668SFam Zheng error_setg(errp, "Failed to lock byte %d", off); 769244a5668SFam Zheng return ret; 7702996ffadSFam Zheng } else if (s) { 7712996ffadSFam Zheng s->locked_perm |= bit; 772244a5668SFam Zheng } 7732996ffadSFam Zheng } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { 774d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 775244a5668SFam Zheng if (ret) { 776244a5668SFam Zheng error_setg(errp, "Failed to unlock byte %d", off); 777244a5668SFam Zheng return ret; 7782996ffadSFam Zheng } else if (s) { 7792996ffadSFam Zheng s->locked_perm &= ~bit; 780244a5668SFam Zheng } 781244a5668SFam Zheng } 782244a5668SFam Zheng } 783244a5668SFam Zheng PERM_FOREACH(i) { 784244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 7852996ffadSFam Zheng uint64_t bit = (1ULL << i); 7862996ffadSFam Zheng if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { 787d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 788244a5668SFam Zheng if (ret) { 789244a5668SFam Zheng error_setg(errp, "Failed to lock byte %d", off); 790244a5668SFam Zheng return ret; 7912996ffadSFam Zheng } else if (s) { 7922996ffadSFam Zheng s->locked_shared_perm |= bit; 793244a5668SFam Zheng } 7942996ffadSFam Zheng } else if (unlock && (locked_shared_perm & bit) && 7952996ffadSFam Zheng !(shared_perm_lock_bits & bit)) { 796d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 797244a5668SFam Zheng if (ret) { 798244a5668SFam Zheng error_setg(errp, "Failed to unlock byte %d", off); 799244a5668SFam Zheng return ret; 8002996ffadSFam Zheng } else if (s) { 8012996ffadSFam Zheng s->locked_shared_perm &= ~bit; 802244a5668SFam Zheng } 803244a5668SFam Zheng } 804244a5668SFam Zheng } 805244a5668SFam Zheng return 0; 806244a5668SFam Zheng } 807244a5668SFam Zheng 808244a5668SFam Zheng /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ 809d0a96155SMax Reitz static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, 810244a5668SFam Zheng Error **errp) 811244a5668SFam Zheng { 812244a5668SFam Zheng int ret; 813244a5668SFam Zheng int i; 814244a5668SFam Zheng 815244a5668SFam Zheng PERM_FOREACH(i) { 816244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 817244a5668SFam Zheng uint64_t p = 1ULL << i; 818244a5668SFam Zheng if (perm & p) { 819d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 820244a5668SFam Zheng if (ret) { 821244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 822244a5668SFam Zheng error_setg(errp, 823244a5668SFam Zheng "Failed to get \"%s\" lock", 824244a5668SFam Zheng perm_name); 825244a5668SFam Zheng g_free(perm_name); 826244a5668SFam Zheng return ret; 827244a5668SFam Zheng } 828244a5668SFam Zheng } 829244a5668SFam Zheng } 830244a5668SFam Zheng PERM_FOREACH(i) { 831244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 832244a5668SFam Zheng uint64_t p = 1ULL << i; 833244a5668SFam Zheng if (!(shared_perm & p)) { 834d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 835244a5668SFam Zheng if (ret) { 836244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 837244a5668SFam Zheng error_setg(errp, 838244a5668SFam Zheng "Failed to get shared \"%s\" lock", 839244a5668SFam Zheng perm_name); 840244a5668SFam Zheng g_free(perm_name); 841244a5668SFam Zheng return ret; 842244a5668SFam Zheng } 843244a5668SFam Zheng } 844244a5668SFam Zheng } 845244a5668SFam Zheng return 0; 846244a5668SFam Zheng } 847244a5668SFam Zheng 848244a5668SFam Zheng static int raw_handle_perm_lock(BlockDriverState *bs, 849244a5668SFam Zheng RawPermLockOp op, 850244a5668SFam Zheng uint64_t new_perm, uint64_t new_shared, 851244a5668SFam Zheng Error **errp) 852244a5668SFam Zheng { 853244a5668SFam Zheng BDRVRawState *s = bs->opaque; 854244a5668SFam Zheng int ret = 0; 855244a5668SFam Zheng Error *local_err = NULL; 856244a5668SFam Zheng 857244a5668SFam Zheng if (!s->use_lock) { 858244a5668SFam Zheng return 0; 859244a5668SFam Zheng } 860244a5668SFam Zheng 861244a5668SFam Zheng if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { 862244a5668SFam Zheng return 0; 863244a5668SFam Zheng } 864244a5668SFam Zheng 865244a5668SFam Zheng switch (op) { 866244a5668SFam Zheng case RAW_PL_PREPARE: 867696aaaedSVladimir Sementsov-Ogievskiy if ((s->perm | new_perm) == s->perm && 868696aaaedSVladimir Sementsov-Ogievskiy (s->shared_perm & new_shared) == s->shared_perm) 869696aaaedSVladimir Sementsov-Ogievskiy { 870696aaaedSVladimir Sementsov-Ogievskiy /* 871696aaaedSVladimir Sementsov-Ogievskiy * We are going to unlock bytes, it should not fail. If it fail due 872696aaaedSVladimir Sementsov-Ogievskiy * to some fs-dependent permission-unrelated reasons (which occurs 873696aaaedSVladimir Sementsov-Ogievskiy * sometimes on NFS and leads to abort in bdrv_replace_child) we 874696aaaedSVladimir Sementsov-Ogievskiy * can't prevent such errors by any check here. And we ignore them 875696aaaedSVladimir Sementsov-Ogievskiy * anyway in ABORT and COMMIT. 876696aaaedSVladimir Sementsov-Ogievskiy */ 877696aaaedSVladimir Sementsov-Ogievskiy return 0; 878696aaaedSVladimir Sementsov-Ogievskiy } 879f2e3af29SFam Zheng ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, 880244a5668SFam Zheng ~s->shared_perm | ~new_shared, 881244a5668SFam Zheng false, errp); 882244a5668SFam Zheng if (!ret) { 883f2e3af29SFam Zheng ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); 884244a5668SFam Zheng if (!ret) { 885244a5668SFam Zheng return 0; 886244a5668SFam Zheng } 887b857431dSFam Zheng error_append_hint(errp, 888b857431dSFam Zheng "Is another process using the image [%s]?\n", 889b857431dSFam Zheng bs->filename); 890244a5668SFam Zheng } 891244a5668SFam Zheng /* fall through to unlock bytes. */ 892244a5668SFam Zheng case RAW_PL_ABORT: 893f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, 894d0a96155SMax Reitz true, &local_err); 895244a5668SFam Zheng if (local_err) { 896244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 897244a5668SFam Zheng * fail. Something weird happened, report it. 898244a5668SFam Zheng */ 899db0754dfSFam Zheng warn_report_err(local_err); 900244a5668SFam Zheng } 901244a5668SFam Zheng break; 902244a5668SFam Zheng case RAW_PL_COMMIT: 903f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, 904d0a96155SMax Reitz true, &local_err); 905244a5668SFam Zheng if (local_err) { 906244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 907244a5668SFam Zheng * fail. Something weird happened, report it. 908244a5668SFam Zheng */ 909db0754dfSFam Zheng warn_report_err(local_err); 910244a5668SFam Zheng } 911244a5668SFam Zheng break; 912244a5668SFam Zheng } 913244a5668SFam Zheng return ret; 914244a5668SFam Zheng } 915244a5668SFam Zheng 9165cec2870SKevin Wolf static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, 91723dece19SKevin Wolf int *open_flags, uint64_t perm, bool force_dup, 9186ceabe6fSKevin Wolf Error **errp) 9195cec2870SKevin Wolf { 9205cec2870SKevin Wolf BDRVRawState *s = bs->opaque; 9215cec2870SKevin Wolf int fd = -1; 9225cec2870SKevin Wolf int ret; 92323dece19SKevin Wolf bool has_writers = perm & 92423dece19SKevin Wolf (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); 9255cec2870SKevin Wolf int fcntl_flags = O_APPEND | O_NONBLOCK; 9265cec2870SKevin Wolf #ifdef O_NOATIME 9275cec2870SKevin Wolf fcntl_flags |= O_NOATIME; 9285cec2870SKevin Wolf #endif 9295cec2870SKevin Wolf 9305cec2870SKevin Wolf *open_flags = 0; 9315cec2870SKevin Wolf if (s->type == FTYPE_CD) { 9325cec2870SKevin Wolf *open_flags |= O_NONBLOCK; 9335cec2870SKevin Wolf } 9345cec2870SKevin Wolf 93523dece19SKevin Wolf raw_parse_flags(flags, open_flags, has_writers); 9365cec2870SKevin Wolf 9375cec2870SKevin Wolf #ifdef O_ASYNC 9385cec2870SKevin Wolf /* Not all operating systems have O_ASYNC, and those that don't 9395cec2870SKevin Wolf * will not let us track the state into rs->open_flags (typically 9405cec2870SKevin Wolf * you achieve the same effect with an ioctl, for example I_SETSIG 9415cec2870SKevin Wolf * on Solaris). But we do not use O_ASYNC, so that's fine. 9425cec2870SKevin Wolf */ 9435cec2870SKevin Wolf assert((s->open_flags & O_ASYNC) == 0); 9445cec2870SKevin Wolf #endif 9455cec2870SKevin Wolf 9466ceabe6fSKevin Wolf if (!force_dup && *open_flags == s->open_flags) { 9476ceabe6fSKevin Wolf /* We're lucky, the existing fd is fine */ 9486ceabe6fSKevin Wolf return s->fd; 9496ceabe6fSKevin Wolf } 9506ceabe6fSKevin Wolf 9515cec2870SKevin Wolf if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { 9525cec2870SKevin Wolf /* dup the original fd */ 9535cec2870SKevin Wolf fd = qemu_dup(s->fd); 9545cec2870SKevin Wolf if (fd >= 0) { 9555cec2870SKevin Wolf ret = fcntl_setfl(fd, *open_flags); 9565cec2870SKevin Wolf if (ret) { 9575cec2870SKevin Wolf qemu_close(fd); 9585cec2870SKevin Wolf fd = -1; 9595cec2870SKevin Wolf } 9605cec2870SKevin Wolf } 9615cec2870SKevin Wolf } 9625cec2870SKevin Wolf 9635cec2870SKevin Wolf /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ 9645cec2870SKevin Wolf if (fd == -1) { 9655cec2870SKevin Wolf const char *normalized_filename = bs->filename; 9665cec2870SKevin Wolf ret = raw_normalize_devicepath(&normalized_filename, errp); 9675cec2870SKevin Wolf if (ret >= 0) { 9685cec2870SKevin Wolf assert(!(*open_flags & O_CREAT)); 9695cec2870SKevin Wolf fd = qemu_open(normalized_filename, *open_flags); 9705cec2870SKevin Wolf if (fd == -1) { 9715cec2870SKevin Wolf error_setg_errno(errp, errno, "Could not reopen file"); 9725cec2870SKevin Wolf return -1; 9735cec2870SKevin Wolf } 9745cec2870SKevin Wolf } 9755cec2870SKevin Wolf } 9765cec2870SKevin Wolf 9775cec2870SKevin Wolf return fd; 9785cec2870SKevin Wolf } 9795cec2870SKevin Wolf 980c1bb86cdSEric Blake static int raw_reopen_prepare(BDRVReopenState *state, 981c1bb86cdSEric Blake BlockReopenQueue *queue, Error **errp) 982c1bb86cdSEric Blake { 983c1bb86cdSEric Blake BDRVRawState *s; 984c1bb86cdSEric Blake BDRVRawReopenState *rs; 98531be8a2aSStefan Hajnoczi QemuOpts *opts; 986a6aeca0cSKevin Wolf int ret; 987c1bb86cdSEric Blake Error *local_err = NULL; 988c1bb86cdSEric Blake 989c1bb86cdSEric Blake assert(state != NULL); 990c1bb86cdSEric Blake assert(state->bs != NULL); 991c1bb86cdSEric Blake 992c1bb86cdSEric Blake s = state->bs->opaque; 993c1bb86cdSEric Blake 994c1bb86cdSEric Blake state->opaque = g_new0(BDRVRawReopenState, 1); 995c1bb86cdSEric Blake rs = state->opaque; 99631be8a2aSStefan Hajnoczi 99731be8a2aSStefan Hajnoczi /* Handle options changes */ 99831be8a2aSStefan Hajnoczi opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 99931be8a2aSStefan Hajnoczi qemu_opts_absorb_qdict(opts, state->options, &local_err); 100031be8a2aSStefan Hajnoczi if (local_err) { 100131be8a2aSStefan Hajnoczi error_propagate(errp, local_err); 100231be8a2aSStefan Hajnoczi ret = -EINVAL; 100331be8a2aSStefan Hajnoczi goto out; 100431be8a2aSStefan Hajnoczi } 100531be8a2aSStefan Hajnoczi 1006f357fcd8SStefan Hajnoczi rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); 10078d324575SAlberto Garcia rs->check_cache_dropped = 10088d324575SAlberto Garcia qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); 10098d324575SAlberto Garcia 10108d324575SAlberto Garcia /* This driver's reopen function doesn't currently allow changing 10118d324575SAlberto Garcia * other options, so let's put them back in the original QDict and 10128d324575SAlberto Garcia * bdrv_reopen_prepare() will detect changes and complain. */ 10138d324575SAlberto Garcia qemu_opts_to_qdict(opts, state->options); 1014c1bb86cdSEric Blake 10155cec2870SKevin Wolf rs->fd = raw_reconfigure_getfd(state->bs, state->flags, &rs->open_flags, 101623dece19SKevin Wolf state->perm, true, &local_err); 10175cec2870SKevin Wolf if (local_err) { 10185cec2870SKevin Wolf error_propagate(errp, local_err); 1019c1bb86cdSEric Blake ret = -1; 10205cec2870SKevin Wolf goto out; 1021c1bb86cdSEric Blake } 1022c1bb86cdSEric Blake 1023c1bb86cdSEric Blake /* Fail already reopen_prepare() if we can't get a working O_DIRECT 1024c1bb86cdSEric Blake * alignment with the new fd. */ 1025c1bb86cdSEric Blake if (rs->fd != -1) { 1026c1bb86cdSEric Blake raw_probe_alignment(state->bs, rs->fd, &local_err); 1027c1bb86cdSEric Blake if (local_err) { 1028c1bb86cdSEric Blake error_propagate(errp, local_err); 1029c1bb86cdSEric Blake ret = -EINVAL; 1030a6aeca0cSKevin Wolf goto out_fd; 1031a6aeca0cSKevin Wolf } 1032c1bb86cdSEric Blake } 1033c1bb86cdSEric Blake 1034e0c9cf3aSKevin Wolf s->reopen_state = state; 1035a6aeca0cSKevin Wolf ret = 0; 1036a6aeca0cSKevin Wolf out_fd: 1037a6aeca0cSKevin Wolf if (ret < 0) { 1038a6aeca0cSKevin Wolf qemu_close(rs->fd); 1039a6aeca0cSKevin Wolf rs->fd = -1; 1040a6aeca0cSKevin Wolf } 104131be8a2aSStefan Hajnoczi out: 104231be8a2aSStefan Hajnoczi qemu_opts_del(opts); 1043c1bb86cdSEric Blake return ret; 1044c1bb86cdSEric Blake } 1045c1bb86cdSEric Blake 1046c1bb86cdSEric Blake static void raw_reopen_commit(BDRVReopenState *state) 1047c1bb86cdSEric Blake { 1048c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1049c1bb86cdSEric Blake BDRVRawState *s = state->bs->opaque; 1050c1bb86cdSEric Blake 1051f357fcd8SStefan Hajnoczi s->drop_cache = rs->drop_cache; 105231be8a2aSStefan Hajnoczi s->check_cache_dropped = rs->check_cache_dropped; 1053c1bb86cdSEric Blake s->open_flags = rs->open_flags; 1054c1bb86cdSEric Blake 1055c1bb86cdSEric Blake qemu_close(s->fd); 1056c1bb86cdSEric Blake s->fd = rs->fd; 1057c1bb86cdSEric Blake 1058c1bb86cdSEric Blake g_free(state->opaque); 1059c1bb86cdSEric Blake state->opaque = NULL; 1060e0c9cf3aSKevin Wolf 1061e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1062e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1063c1bb86cdSEric Blake } 1064c1bb86cdSEric Blake 1065c1bb86cdSEric Blake 1066c1bb86cdSEric Blake static void raw_reopen_abort(BDRVReopenState *state) 1067c1bb86cdSEric Blake { 1068c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1069e0c9cf3aSKevin Wolf BDRVRawState *s = state->bs->opaque; 1070c1bb86cdSEric Blake 1071c1bb86cdSEric Blake /* nothing to do if NULL, we didn't get far enough */ 1072c1bb86cdSEric Blake if (rs == NULL) { 1073c1bb86cdSEric Blake return; 1074c1bb86cdSEric Blake } 1075c1bb86cdSEric Blake 1076c1bb86cdSEric Blake if (rs->fd >= 0) { 1077c1bb86cdSEric Blake qemu_close(rs->fd); 1078c1bb86cdSEric Blake rs->fd = -1; 1079c1bb86cdSEric Blake } 1080c1bb86cdSEric Blake g_free(state->opaque); 1081c1bb86cdSEric Blake state->opaque = NULL; 1082e0c9cf3aSKevin Wolf 1083e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1084e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1085c1bb86cdSEric Blake } 1086c1bb86cdSEric Blake 1087867eccfeSMaxim Levitsky static int sg_get_max_transfer_length(int fd) 1088c1bb86cdSEric Blake { 1089c1bb86cdSEric Blake #ifdef BLKSECTGET 109048265250SEric Farman int max_bytes = 0; 1091867eccfeSMaxim Levitsky 1092867eccfeSMaxim Levitsky if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) { 109348265250SEric Farman return max_bytes; 1094c1bb86cdSEric Blake } else { 1095c1bb86cdSEric Blake return -errno; 1096c1bb86cdSEric Blake } 1097c1bb86cdSEric Blake #else 1098c1bb86cdSEric Blake return -ENOSYS; 1099c1bb86cdSEric Blake #endif 1100c1bb86cdSEric Blake } 1101c1bb86cdSEric Blake 1102867eccfeSMaxim Levitsky static int sg_get_max_segments(int fd) 11039103f1ceSFam Zheng { 11049103f1ceSFam Zheng #ifdef CONFIG_LINUX 11059103f1ceSFam Zheng char buf[32]; 11069103f1ceSFam Zheng const char *end; 1107867eccfeSMaxim Levitsky char *sysfspath = NULL; 11089103f1ceSFam Zheng int ret; 1109867eccfeSMaxim Levitsky int sysfd = -1; 11109103f1ceSFam Zheng long max_segments; 1111867eccfeSMaxim Levitsky struct stat st; 1112867eccfeSMaxim Levitsky 1113867eccfeSMaxim Levitsky if (fstat(fd, &st)) { 1114867eccfeSMaxim Levitsky ret = -errno; 1115867eccfeSMaxim Levitsky goto out; 1116867eccfeSMaxim Levitsky } 11179103f1ceSFam Zheng 11189103f1ceSFam Zheng sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", 1119867eccfeSMaxim Levitsky major(st.st_rdev), minor(st.st_rdev)); 1120867eccfeSMaxim Levitsky sysfd = open(sysfspath, O_RDONLY); 1121867eccfeSMaxim Levitsky if (sysfd == -1) { 11229103f1ceSFam Zheng ret = -errno; 11239103f1ceSFam Zheng goto out; 11249103f1ceSFam Zheng } 11259103f1ceSFam Zheng do { 1126867eccfeSMaxim Levitsky ret = read(sysfd, buf, sizeof(buf) - 1); 11279103f1ceSFam Zheng } while (ret == -1 && errno == EINTR); 11289103f1ceSFam Zheng if (ret < 0) { 11299103f1ceSFam Zheng ret = -errno; 11309103f1ceSFam Zheng goto out; 11319103f1ceSFam Zheng } else if (ret == 0) { 11329103f1ceSFam Zheng ret = -EIO; 11339103f1ceSFam Zheng goto out; 11349103f1ceSFam Zheng } 11359103f1ceSFam Zheng buf[ret] = 0; 11369103f1ceSFam Zheng /* The file is ended with '\n', pass 'end' to accept that. */ 11379103f1ceSFam Zheng ret = qemu_strtol(buf, &end, 10, &max_segments); 11389103f1ceSFam Zheng if (ret == 0 && end && *end == '\n') { 11399103f1ceSFam Zheng ret = max_segments; 11409103f1ceSFam Zheng } 11419103f1ceSFam Zheng 11429103f1ceSFam Zheng out: 1143867eccfeSMaxim Levitsky if (sysfd != -1) { 1144867eccfeSMaxim Levitsky close(sysfd); 1145fed414dfSFam Zheng } 11469103f1ceSFam Zheng g_free(sysfspath); 11479103f1ceSFam Zheng return ret; 11489103f1ceSFam Zheng #else 11499103f1ceSFam Zheng return -ENOTSUP; 11509103f1ceSFam Zheng #endif 11519103f1ceSFam Zheng } 11529103f1ceSFam Zheng 1153c1bb86cdSEric Blake static void raw_refresh_limits(BlockDriverState *bs, Error **errp) 1154c1bb86cdSEric Blake { 1155c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1156c1bb86cdSEric Blake 1157867eccfeSMaxim Levitsky if (bs->sg) { 1158867eccfeSMaxim Levitsky int ret = sg_get_max_transfer_length(s->fd); 1159867eccfeSMaxim Levitsky 116048265250SEric Farman if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { 116148265250SEric Farman bs->bl.max_transfer = pow2floor(ret); 1162c1bb86cdSEric Blake } 1163867eccfeSMaxim Levitsky 1164867eccfeSMaxim Levitsky ret = sg_get_max_segments(s->fd); 11659103f1ceSFam Zheng if (ret > 0) { 1166038adc2fSWei Yang bs->bl.max_transfer = MIN(bs->bl.max_transfer, 1167038adc2fSWei Yang ret * qemu_real_host_page_size); 1168c1bb86cdSEric Blake } 1169c1bb86cdSEric Blake } 1170c1bb86cdSEric Blake 1171c1bb86cdSEric Blake raw_probe_alignment(bs, s->fd, errp); 1172c1bb86cdSEric Blake bs->bl.min_mem_alignment = s->buf_align; 1173038adc2fSWei Yang bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size); 1174c1bb86cdSEric Blake } 1175c1bb86cdSEric Blake 1176c1bb86cdSEric Blake static int check_for_dasd(int fd) 1177c1bb86cdSEric Blake { 1178c1bb86cdSEric Blake #ifdef BIODASDINFO2 1179c1bb86cdSEric Blake struct dasd_information2_t info = {0}; 1180c1bb86cdSEric Blake 1181c1bb86cdSEric Blake return ioctl(fd, BIODASDINFO2, &info); 1182c1bb86cdSEric Blake #else 1183c1bb86cdSEric Blake return -1; 1184c1bb86cdSEric Blake #endif 1185c1bb86cdSEric Blake } 1186c1bb86cdSEric Blake 1187c1bb86cdSEric Blake /** 1188c1bb86cdSEric Blake * Try to get @bs's logical and physical block size. 1189c1bb86cdSEric Blake * On success, store them in @bsz and return zero. 1190c1bb86cdSEric Blake * On failure, return negative errno. 1191c1bb86cdSEric Blake */ 1192c1bb86cdSEric Blake static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) 1193c1bb86cdSEric Blake { 1194c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1195c1bb86cdSEric Blake int ret; 1196c1bb86cdSEric Blake 1197c1bb86cdSEric Blake /* If DASD, get blocksizes */ 1198c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1199c1bb86cdSEric Blake return -ENOTSUP; 1200c1bb86cdSEric Blake } 1201c1bb86cdSEric Blake ret = probe_logical_blocksize(s->fd, &bsz->log); 1202c1bb86cdSEric Blake if (ret < 0) { 1203c1bb86cdSEric Blake return ret; 1204c1bb86cdSEric Blake } 1205c1bb86cdSEric Blake return probe_physical_blocksize(s->fd, &bsz->phys); 1206c1bb86cdSEric Blake } 1207c1bb86cdSEric Blake 1208c1bb86cdSEric Blake /** 1209c1bb86cdSEric Blake * Try to get @bs's geometry: cyls, heads, sectors. 1210c1bb86cdSEric Blake * On success, store them in @geo and return 0. 1211c1bb86cdSEric Blake * On failure return -errno. 1212c1bb86cdSEric Blake * (Allows block driver to assign default geometry values that guest sees) 1213c1bb86cdSEric Blake */ 1214c1bb86cdSEric Blake #ifdef __linux__ 1215c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1216c1bb86cdSEric Blake { 1217c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1218c1bb86cdSEric Blake struct hd_geometry ioctl_geo = {0}; 1219c1bb86cdSEric Blake 1220c1bb86cdSEric Blake /* If DASD, get its geometry */ 1221c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1222c1bb86cdSEric Blake return -ENOTSUP; 1223c1bb86cdSEric Blake } 1224c1bb86cdSEric Blake if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { 1225c1bb86cdSEric Blake return -errno; 1226c1bb86cdSEric Blake } 1227c1bb86cdSEric Blake /* HDIO_GETGEO may return success even though geo contains zeros 1228c1bb86cdSEric Blake (e.g. certain multipath setups) */ 1229c1bb86cdSEric Blake if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { 1230c1bb86cdSEric Blake return -ENOTSUP; 1231c1bb86cdSEric Blake } 1232c1bb86cdSEric Blake /* Do not return a geometry for partition */ 1233c1bb86cdSEric Blake if (ioctl_geo.start != 0) { 1234c1bb86cdSEric Blake return -ENOTSUP; 1235c1bb86cdSEric Blake } 1236c1bb86cdSEric Blake geo->heads = ioctl_geo.heads; 1237c1bb86cdSEric Blake geo->sectors = ioctl_geo.sectors; 1238c1bb86cdSEric Blake geo->cylinders = ioctl_geo.cylinders; 1239c1bb86cdSEric Blake 1240c1bb86cdSEric Blake return 0; 1241c1bb86cdSEric Blake } 1242c1bb86cdSEric Blake #else /* __linux__ */ 1243c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1244c1bb86cdSEric Blake { 1245c1bb86cdSEric Blake return -ENOTSUP; 1246c1bb86cdSEric Blake } 1247c1bb86cdSEric Blake #endif 1248c1bb86cdSEric Blake 124903425671SKevin Wolf #if defined(__linux__) 125003425671SKevin Wolf static int handle_aiocb_ioctl(void *opaque) 1251c1bb86cdSEric Blake { 125203425671SKevin Wolf RawPosixAIOData *aiocb = opaque; 1253c1bb86cdSEric Blake int ret; 1254c1bb86cdSEric Blake 1255d57c44d0SKevin Wolf ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); 1256c1bb86cdSEric Blake if (ret == -1) { 1257c1bb86cdSEric Blake return -errno; 1258c1bb86cdSEric Blake } 1259c1bb86cdSEric Blake 1260c1bb86cdSEric Blake return 0; 1261c1bb86cdSEric Blake } 126203425671SKevin Wolf #endif /* linux */ 1263c1bb86cdSEric Blake 126406dc9bd5SKevin Wolf static int handle_aiocb_flush(void *opaque) 1265c1bb86cdSEric Blake { 126606dc9bd5SKevin Wolf RawPosixAIOData *aiocb = opaque; 1267e5bcf967SKevin Wolf BDRVRawState *s = aiocb->bs->opaque; 1268c1bb86cdSEric Blake int ret; 1269c1bb86cdSEric Blake 1270e5bcf967SKevin Wolf if (s->page_cache_inconsistent) { 1271e5bcf967SKevin Wolf return -EIO; 1272e5bcf967SKevin Wolf } 1273e5bcf967SKevin Wolf 1274c1bb86cdSEric Blake ret = qemu_fdatasync(aiocb->aio_fildes); 1275c1bb86cdSEric Blake if (ret == -1) { 1276e5bcf967SKevin Wolf /* There is no clear definition of the semantics of a failing fsync(), 1277e5bcf967SKevin Wolf * so we may have to assume the worst. The sad truth is that this 1278e5bcf967SKevin Wolf * assumption is correct for Linux. Some pages are now probably marked 1279e5bcf967SKevin Wolf * clean in the page cache even though they are inconsistent with the 1280e5bcf967SKevin Wolf * on-disk contents. The next fdatasync() call would succeed, but no 1281e5bcf967SKevin Wolf * further writeback attempt will be made. We can't get back to a state 1282e5bcf967SKevin Wolf * in which we know what is on disk (we would have to rewrite 1283e5bcf967SKevin Wolf * everything that was touched since the last fdatasync() at least), so 1284e5bcf967SKevin Wolf * make bdrv_flush() fail permanently. Given that the behaviour isn't 1285e5bcf967SKevin Wolf * really defined, I have little hope that other OSes are doing better. 1286e5bcf967SKevin Wolf * 1287e5bcf967SKevin Wolf * Obviously, this doesn't affect O_DIRECT, which bypasses the page 1288e5bcf967SKevin Wolf * cache. */ 1289e5bcf967SKevin Wolf if ((s->open_flags & O_DIRECT) == 0) { 1290e5bcf967SKevin Wolf s->page_cache_inconsistent = true; 1291e5bcf967SKevin Wolf } 1292c1bb86cdSEric Blake return -errno; 1293c1bb86cdSEric Blake } 1294c1bb86cdSEric Blake return 0; 1295c1bb86cdSEric Blake } 1296c1bb86cdSEric Blake 1297c1bb86cdSEric Blake #ifdef CONFIG_PREADV 1298c1bb86cdSEric Blake 1299c1bb86cdSEric Blake static bool preadv_present = true; 1300c1bb86cdSEric Blake 1301c1bb86cdSEric Blake static ssize_t 1302c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1303c1bb86cdSEric Blake { 1304c1bb86cdSEric Blake return preadv(fd, iov, nr_iov, offset); 1305c1bb86cdSEric Blake } 1306c1bb86cdSEric Blake 1307c1bb86cdSEric Blake static ssize_t 1308c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1309c1bb86cdSEric Blake { 1310c1bb86cdSEric Blake return pwritev(fd, iov, nr_iov, offset); 1311c1bb86cdSEric Blake } 1312c1bb86cdSEric Blake 1313c1bb86cdSEric Blake #else 1314c1bb86cdSEric Blake 1315c1bb86cdSEric Blake static bool preadv_present = false; 1316c1bb86cdSEric Blake 1317c1bb86cdSEric Blake static ssize_t 1318c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1319c1bb86cdSEric Blake { 1320c1bb86cdSEric Blake return -ENOSYS; 1321c1bb86cdSEric Blake } 1322c1bb86cdSEric Blake 1323c1bb86cdSEric Blake static ssize_t 1324c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1325c1bb86cdSEric Blake { 1326c1bb86cdSEric Blake return -ENOSYS; 1327c1bb86cdSEric Blake } 1328c1bb86cdSEric Blake 1329c1bb86cdSEric Blake #endif 1330c1bb86cdSEric Blake 1331c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) 1332c1bb86cdSEric Blake { 1333c1bb86cdSEric Blake ssize_t len; 1334c1bb86cdSEric Blake 1335c1bb86cdSEric Blake do { 1336c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) 1337c1bb86cdSEric Blake len = qemu_pwritev(aiocb->aio_fildes, 1338d57c44d0SKevin Wolf aiocb->io.iov, 1339d57c44d0SKevin Wolf aiocb->io.niov, 1340c1bb86cdSEric Blake aiocb->aio_offset); 1341c1bb86cdSEric Blake else 1342c1bb86cdSEric Blake len = qemu_preadv(aiocb->aio_fildes, 1343d57c44d0SKevin Wolf aiocb->io.iov, 1344d57c44d0SKevin Wolf aiocb->io.niov, 1345c1bb86cdSEric Blake aiocb->aio_offset); 1346c1bb86cdSEric Blake } while (len == -1 && errno == EINTR); 1347c1bb86cdSEric Blake 1348c1bb86cdSEric Blake if (len == -1) { 1349c1bb86cdSEric Blake return -errno; 1350c1bb86cdSEric Blake } 1351c1bb86cdSEric Blake return len; 1352c1bb86cdSEric Blake } 1353c1bb86cdSEric Blake 1354c1bb86cdSEric Blake /* 1355c1bb86cdSEric Blake * Read/writes the data to/from a given linear buffer. 1356c1bb86cdSEric Blake * 1357c1bb86cdSEric Blake * Returns the number of bytes handles or -errno in case of an error. Short 1358c1bb86cdSEric Blake * reads are only returned if the end of the file is reached. 1359c1bb86cdSEric Blake */ 1360c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) 1361c1bb86cdSEric Blake { 1362c1bb86cdSEric Blake ssize_t offset = 0; 1363c1bb86cdSEric Blake ssize_t len; 1364c1bb86cdSEric Blake 1365c1bb86cdSEric Blake while (offset < aiocb->aio_nbytes) { 1366c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1367c1bb86cdSEric Blake len = pwrite(aiocb->aio_fildes, 1368c1bb86cdSEric Blake (const char *)buf + offset, 1369c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1370c1bb86cdSEric Blake aiocb->aio_offset + offset); 1371c1bb86cdSEric Blake } else { 1372c1bb86cdSEric Blake len = pread(aiocb->aio_fildes, 1373c1bb86cdSEric Blake buf + offset, 1374c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1375c1bb86cdSEric Blake aiocb->aio_offset + offset); 1376c1bb86cdSEric Blake } 1377c1bb86cdSEric Blake if (len == -1 && errno == EINTR) { 1378c1bb86cdSEric Blake continue; 1379c1bb86cdSEric Blake } else if (len == -1 && errno == EINVAL && 1380c1bb86cdSEric Blake (aiocb->bs->open_flags & BDRV_O_NOCACHE) && 1381c1bb86cdSEric Blake !(aiocb->aio_type & QEMU_AIO_WRITE) && 1382c1bb86cdSEric Blake offset > 0) { 1383c1bb86cdSEric Blake /* O_DIRECT pread() may fail with EINVAL when offset is unaligned 1384c1bb86cdSEric Blake * after a short read. Assume that O_DIRECT short reads only occur 1385c1bb86cdSEric Blake * at EOF. Therefore this is a short read, not an I/O error. 1386c1bb86cdSEric Blake */ 1387c1bb86cdSEric Blake break; 1388c1bb86cdSEric Blake } else if (len == -1) { 1389c1bb86cdSEric Blake offset = -errno; 1390c1bb86cdSEric Blake break; 1391c1bb86cdSEric Blake } else if (len == 0) { 1392c1bb86cdSEric Blake break; 1393c1bb86cdSEric Blake } 1394c1bb86cdSEric Blake offset += len; 1395c1bb86cdSEric Blake } 1396c1bb86cdSEric Blake 1397c1bb86cdSEric Blake return offset; 1398c1bb86cdSEric Blake } 1399c1bb86cdSEric Blake 1400999e6b69SKevin Wolf static int handle_aiocb_rw(void *opaque) 1401c1bb86cdSEric Blake { 1402999e6b69SKevin Wolf RawPosixAIOData *aiocb = opaque; 1403c1bb86cdSEric Blake ssize_t nbytes; 1404c1bb86cdSEric Blake char *buf; 1405c1bb86cdSEric Blake 1406c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { 1407c1bb86cdSEric Blake /* 1408c1bb86cdSEric Blake * If there is just a single buffer, and it is properly aligned 1409c1bb86cdSEric Blake * we can just use plain pread/pwrite without any problems. 1410c1bb86cdSEric Blake */ 1411d57c44d0SKevin Wolf if (aiocb->io.niov == 1) { 141254c7ca1bSKevin Wolf nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); 141354c7ca1bSKevin Wolf goto out; 1414c1bb86cdSEric Blake } 1415c1bb86cdSEric Blake /* 1416c1bb86cdSEric Blake * We have more than one iovec, and all are properly aligned. 1417c1bb86cdSEric Blake * 1418c1bb86cdSEric Blake * Try preadv/pwritev first and fall back to linearizing the 1419c1bb86cdSEric Blake * buffer if it's not supported. 1420c1bb86cdSEric Blake */ 1421c1bb86cdSEric Blake if (preadv_present) { 1422c1bb86cdSEric Blake nbytes = handle_aiocb_rw_vector(aiocb); 1423c1bb86cdSEric Blake if (nbytes == aiocb->aio_nbytes || 1424c1bb86cdSEric Blake (nbytes < 0 && nbytes != -ENOSYS)) { 142554c7ca1bSKevin Wolf goto out; 1426c1bb86cdSEric Blake } 1427c1bb86cdSEric Blake preadv_present = false; 1428c1bb86cdSEric Blake } 1429c1bb86cdSEric Blake 1430c1bb86cdSEric Blake /* 1431c1bb86cdSEric Blake * XXX(hch): short read/write. no easy way to handle the reminder 1432c1bb86cdSEric Blake * using these interfaces. For now retry using plain 1433c1bb86cdSEric Blake * pread/pwrite? 1434c1bb86cdSEric Blake */ 1435c1bb86cdSEric Blake } 1436c1bb86cdSEric Blake 1437c1bb86cdSEric Blake /* 1438c1bb86cdSEric Blake * Ok, we have to do it the hard way, copy all segments into 1439c1bb86cdSEric Blake * a single aligned buffer. 1440c1bb86cdSEric Blake */ 1441c1bb86cdSEric Blake buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); 1442c1bb86cdSEric Blake if (buf == NULL) { 144354c7ca1bSKevin Wolf nbytes = -ENOMEM; 144454c7ca1bSKevin Wolf goto out; 1445c1bb86cdSEric Blake } 1446c1bb86cdSEric Blake 1447c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1448c1bb86cdSEric Blake char *p = buf; 1449c1bb86cdSEric Blake int i; 1450c1bb86cdSEric Blake 1451d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov; ++i) { 1452d57c44d0SKevin Wolf memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); 1453d57c44d0SKevin Wolf p += aiocb->io.iov[i].iov_len; 1454c1bb86cdSEric Blake } 1455c1bb86cdSEric Blake assert(p - buf == aiocb->aio_nbytes); 1456c1bb86cdSEric Blake } 1457c1bb86cdSEric Blake 1458c1bb86cdSEric Blake nbytes = handle_aiocb_rw_linear(aiocb, buf); 1459c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { 1460c1bb86cdSEric Blake char *p = buf; 1461c1bb86cdSEric Blake size_t count = aiocb->aio_nbytes, copy; 1462c1bb86cdSEric Blake int i; 1463c1bb86cdSEric Blake 1464d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov && count; ++i) { 1465c1bb86cdSEric Blake copy = count; 1466d57c44d0SKevin Wolf if (copy > aiocb->io.iov[i].iov_len) { 1467d57c44d0SKevin Wolf copy = aiocb->io.iov[i].iov_len; 1468c1bb86cdSEric Blake } 1469d57c44d0SKevin Wolf memcpy(aiocb->io.iov[i].iov_base, p, copy); 1470c1bb86cdSEric Blake assert(count >= copy); 1471c1bb86cdSEric Blake p += copy; 1472c1bb86cdSEric Blake count -= copy; 1473c1bb86cdSEric Blake } 1474c1bb86cdSEric Blake assert(count == 0); 1475c1bb86cdSEric Blake } 1476c1bb86cdSEric Blake qemu_vfree(buf); 1477c1bb86cdSEric Blake 147854c7ca1bSKevin Wolf out: 147954c7ca1bSKevin Wolf if (nbytes == aiocb->aio_nbytes) { 148054c7ca1bSKevin Wolf return 0; 148154c7ca1bSKevin Wolf } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { 148254c7ca1bSKevin Wolf if (aiocb->aio_type & QEMU_AIO_WRITE) { 148354c7ca1bSKevin Wolf return -EINVAL; 148454c7ca1bSKevin Wolf } else { 148554c7ca1bSKevin Wolf iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, 148654c7ca1bSKevin Wolf 0, aiocb->aio_nbytes - nbytes); 148754c7ca1bSKevin Wolf return 0; 148854c7ca1bSKevin Wolf } 148954c7ca1bSKevin Wolf } else { 149054c7ca1bSKevin Wolf assert(nbytes < 0); 1491c1bb86cdSEric Blake return nbytes; 1492c1bb86cdSEric Blake } 149354c7ca1bSKevin Wolf } 1494c1bb86cdSEric Blake 1495c1bb86cdSEric Blake static int translate_err(int err) 1496c1bb86cdSEric Blake { 1497c1bb86cdSEric Blake if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || 1498c1bb86cdSEric Blake err == -ENOTTY) { 1499c1bb86cdSEric Blake err = -ENOTSUP; 1500c1bb86cdSEric Blake } 1501c1bb86cdSEric Blake return err; 1502c1bb86cdSEric Blake } 1503c1bb86cdSEric Blake 1504c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 1505c1bb86cdSEric Blake static int do_fallocate(int fd, int mode, off_t offset, off_t len) 1506c1bb86cdSEric Blake { 1507c1bb86cdSEric Blake do { 1508c1bb86cdSEric Blake if (fallocate(fd, mode, offset, len) == 0) { 1509c1bb86cdSEric Blake return 0; 1510c1bb86cdSEric Blake } 1511c1bb86cdSEric Blake } while (errno == EINTR); 1512c1bb86cdSEric Blake return translate_err(-errno); 1513c1bb86cdSEric Blake } 1514c1bb86cdSEric Blake #endif 1515c1bb86cdSEric Blake 1516c1bb86cdSEric Blake static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) 1517c1bb86cdSEric Blake { 1518c1bb86cdSEric Blake int ret = -ENOTSUP; 1519c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1520c1bb86cdSEric Blake 1521c1bb86cdSEric Blake if (!s->has_write_zeroes) { 1522c1bb86cdSEric Blake return -ENOTSUP; 1523c1bb86cdSEric Blake } 1524c1bb86cdSEric Blake 1525c1bb86cdSEric Blake #ifdef BLKZEROOUT 1526738301e1SKevin Wolf /* The BLKZEROOUT implementation in the kernel doesn't set 1527738301e1SKevin Wolf * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow 1528738301e1SKevin Wolf * fallbacks. */ 1529738301e1SKevin Wolf if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { 1530c1bb86cdSEric Blake do { 1531c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1532c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { 1533c1bb86cdSEric Blake return 0; 1534c1bb86cdSEric Blake } 1535c1bb86cdSEric Blake } while (errno == EINTR); 1536c1bb86cdSEric Blake 1537c1bb86cdSEric Blake ret = translate_err(-errno); 1538c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1539c1bb86cdSEric Blake s->has_write_zeroes = false; 1540c1bb86cdSEric Blake } 1541effecce6SKevin Wolf } 1542effecce6SKevin Wolf #endif 1543effecce6SKevin Wolf 1544c1bb86cdSEric Blake return ret; 1545c1bb86cdSEric Blake } 1546c1bb86cdSEric Blake 15477154d8aeSKevin Wolf static int handle_aiocb_write_zeroes(void *opaque) 1548c1bb86cdSEric Blake { 15497154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 155070d9110bSDenis V. Lunev #ifdef CONFIG_FALLOCATE 1551b2c6f23fSMax Reitz BDRVRawState *s = aiocb->bs->opaque; 155270d9110bSDenis V. Lunev int64_t len; 155370d9110bSDenis V. Lunev #endif 1554c1bb86cdSEric Blake 1555c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1556c1bb86cdSEric Blake return handle_aiocb_write_zeroes_block(aiocb); 1557c1bb86cdSEric Blake } 1558c1bb86cdSEric Blake 1559c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_ZERO_RANGE 1560c1bb86cdSEric Blake if (s->has_write_zeroes) { 1561c1bb86cdSEric Blake int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, 1562c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1563294682ccSAndrey Shinkevich if (ret == -EINVAL) { 1564294682ccSAndrey Shinkevich /* 1565294682ccSAndrey Shinkevich * Allow falling back to pwrite for file systems that 1566294682ccSAndrey Shinkevich * do not support fallocate() for an unaligned byte range. 1567294682ccSAndrey Shinkevich */ 1568294682ccSAndrey Shinkevich return -ENOTSUP; 1569294682ccSAndrey Shinkevich } 1570c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1571c1bb86cdSEric Blake return ret; 1572c1bb86cdSEric Blake } 1573c1bb86cdSEric Blake s->has_write_zeroes = false; 1574c1bb86cdSEric Blake } 1575c1bb86cdSEric Blake #endif 1576c1bb86cdSEric Blake 1577c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1578c1bb86cdSEric Blake if (s->has_discard && s->has_fallocate) { 1579c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 1580c1bb86cdSEric Blake FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1581c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1582c1bb86cdSEric Blake if (ret == 0) { 1583c1bb86cdSEric Blake ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1584c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1585c1bb86cdSEric Blake return ret; 1586c1bb86cdSEric Blake } 1587c1bb86cdSEric Blake s->has_fallocate = false; 1588c1bb86cdSEric Blake } else if (ret != -ENOTSUP) { 1589c1bb86cdSEric Blake return ret; 1590c1bb86cdSEric Blake } else { 1591c1bb86cdSEric Blake s->has_discard = false; 1592c1bb86cdSEric Blake } 1593c1bb86cdSEric Blake } 1594c1bb86cdSEric Blake #endif 1595c1bb86cdSEric Blake 1596c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 159770d9110bSDenis V. Lunev /* Last resort: we are trying to extend the file with zeroed data. This 159870d9110bSDenis V. Lunev * can be done via fallocate(fd, 0) */ 159970d9110bSDenis V. Lunev len = bdrv_getlength(aiocb->bs); 160070d9110bSDenis V. Lunev if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { 1601c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1602c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1603c1bb86cdSEric Blake return ret; 1604c1bb86cdSEric Blake } 1605c1bb86cdSEric Blake s->has_fallocate = false; 1606c1bb86cdSEric Blake } 1607c1bb86cdSEric Blake #endif 1608c1bb86cdSEric Blake 1609c1bb86cdSEric Blake return -ENOTSUP; 1610c1bb86cdSEric Blake } 1611c1bb86cdSEric Blake 16127154d8aeSKevin Wolf static int handle_aiocb_write_zeroes_unmap(void *opaque) 161334fa110eSKevin Wolf { 16147154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 161534fa110eSKevin Wolf BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; 161634fa110eSKevin Wolf int ret; 161734fa110eSKevin Wolf 161834fa110eSKevin Wolf /* First try to write zeros and unmap at the same time */ 161934fa110eSKevin Wolf 162034fa110eSKevin Wolf #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 162134fa110eSKevin Wolf ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 162234fa110eSKevin Wolf aiocb->aio_offset, aiocb->aio_nbytes); 162334fa110eSKevin Wolf if (ret != -ENOTSUP) { 162434fa110eSKevin Wolf return ret; 162534fa110eSKevin Wolf } 162634fa110eSKevin Wolf #endif 162734fa110eSKevin Wolf 162834fa110eSKevin Wolf /* If we couldn't manage to unmap while guaranteed that the area reads as 162934fa110eSKevin Wolf * all-zero afterwards, just write zeroes without unmapping */ 163034fa110eSKevin Wolf ret = handle_aiocb_write_zeroes(aiocb); 163134fa110eSKevin Wolf return ret; 163234fa110eSKevin Wolf } 163334fa110eSKevin Wolf 16341efad060SFam Zheng #ifndef HAVE_COPY_FILE_RANGE 16351efad060SFam Zheng static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, 16361efad060SFam Zheng off_t *out_off, size_t len, unsigned int flags) 16371efad060SFam Zheng { 16381efad060SFam Zheng #ifdef __NR_copy_file_range 16391efad060SFam Zheng return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, 16401efad060SFam Zheng out_off, len, flags); 16411efad060SFam Zheng #else 16421efad060SFam Zheng errno = ENOSYS; 16431efad060SFam Zheng return -1; 16441efad060SFam Zheng #endif 16451efad060SFam Zheng } 16461efad060SFam Zheng #endif 16471efad060SFam Zheng 164858a209c4SKevin Wolf static int handle_aiocb_copy_range(void *opaque) 16491efad060SFam Zheng { 165058a209c4SKevin Wolf RawPosixAIOData *aiocb = opaque; 16511efad060SFam Zheng uint64_t bytes = aiocb->aio_nbytes; 16521efad060SFam Zheng off_t in_off = aiocb->aio_offset; 1653d57c44d0SKevin Wolf off_t out_off = aiocb->copy_range.aio_offset2; 16541efad060SFam Zheng 16551efad060SFam Zheng while (bytes) { 16561efad060SFam Zheng ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, 1657d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, &out_off, 16581efad060SFam Zheng bytes, 0); 1659ecc983a5SFam Zheng trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, 1660d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, out_off, bytes, 1661d57c44d0SKevin Wolf 0, ret); 1662c436e3d0SFam Zheng if (ret == 0) { 1663c436e3d0SFam Zheng /* No progress (e.g. when beyond EOF), let the caller fall back to 1664c436e3d0SFam Zheng * buffer I/O. */ 1665c436e3d0SFam Zheng return -ENOSPC; 16661efad060SFam Zheng } 16671efad060SFam Zheng if (ret < 0) { 1668c436e3d0SFam Zheng switch (errno) { 1669c436e3d0SFam Zheng case ENOSYS: 16701efad060SFam Zheng return -ENOTSUP; 1671c436e3d0SFam Zheng case EINTR: 1672c436e3d0SFam Zheng continue; 1673c436e3d0SFam Zheng default: 16741efad060SFam Zheng return -errno; 16751efad060SFam Zheng } 16761efad060SFam Zheng } 16771efad060SFam Zheng bytes -= ret; 16781efad060SFam Zheng } 16791efad060SFam Zheng return 0; 16801efad060SFam Zheng } 16811efad060SFam Zheng 168246ee0f46SKevin Wolf static int handle_aiocb_discard(void *opaque) 1683c1bb86cdSEric Blake { 168446ee0f46SKevin Wolf RawPosixAIOData *aiocb = opaque; 1685c1bb86cdSEric Blake int ret = -EOPNOTSUPP; 1686c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1687c1bb86cdSEric Blake 1688c1bb86cdSEric Blake if (!s->has_discard) { 1689c1bb86cdSEric Blake return -ENOTSUP; 1690c1bb86cdSEric Blake } 1691c1bb86cdSEric Blake 1692c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1693c1bb86cdSEric Blake #ifdef BLKDISCARD 1694c1bb86cdSEric Blake do { 1695c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1696c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { 1697c1bb86cdSEric Blake return 0; 1698c1bb86cdSEric Blake } 1699c1bb86cdSEric Blake } while (errno == EINTR); 1700c1bb86cdSEric Blake 1701c1bb86cdSEric Blake ret = -errno; 1702c1bb86cdSEric Blake #endif 1703c1bb86cdSEric Blake } else { 1704c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1705c1bb86cdSEric Blake ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1706c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1707c1bb86cdSEric Blake #endif 1708c1bb86cdSEric Blake } 1709c1bb86cdSEric Blake 1710c1bb86cdSEric Blake ret = translate_err(ret); 1711c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1712c1bb86cdSEric Blake s->has_discard = false; 1713c1bb86cdSEric Blake } 1714c1bb86cdSEric Blake return ret; 1715c1bb86cdSEric Blake } 1716c1bb86cdSEric Blake 17173a20013fSNir Soffer /* 17183a20013fSNir Soffer * Help alignment probing by allocating the first block. 17193a20013fSNir Soffer * 17203a20013fSNir Soffer * When reading with direct I/O from unallocated area on Gluster backed by XFS, 17213a20013fSNir Soffer * reading succeeds regardless of request length. In this case we fallback to 17223a20013fSNir Soffer * safe alignment which is not optimal. Allocating the first block avoids this 17233a20013fSNir Soffer * fallback. 17243a20013fSNir Soffer * 17253a20013fSNir Soffer * fd may be opened with O_DIRECT, but we don't know the buffer alignment or 17263a20013fSNir Soffer * request alignment, so we use safe values. 17273a20013fSNir Soffer * 17283a20013fSNir Soffer * Returns: 0 on success, -errno on failure. Since this is an optimization, 17293a20013fSNir Soffer * caller may ignore failures. 17303a20013fSNir Soffer */ 17313a20013fSNir Soffer static int allocate_first_block(int fd, size_t max_size) 17323a20013fSNir Soffer { 17333a20013fSNir Soffer size_t write_size = (max_size < MAX_BLOCKSIZE) 17343a20013fSNir Soffer ? BDRV_SECTOR_SIZE 17353a20013fSNir Soffer : MAX_BLOCKSIZE; 1736038adc2fSWei Yang size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); 17373a20013fSNir Soffer void *buf; 17383a20013fSNir Soffer ssize_t n; 17393a20013fSNir Soffer int ret; 17403a20013fSNir Soffer 17413a20013fSNir Soffer buf = qemu_memalign(max_align, write_size); 17423a20013fSNir Soffer memset(buf, 0, write_size); 17433a20013fSNir Soffer 17443a20013fSNir Soffer do { 17453a20013fSNir Soffer n = pwrite(fd, buf, write_size, 0); 17463a20013fSNir Soffer } while (n == -1 && errno == EINTR); 17473a20013fSNir Soffer 17483a20013fSNir Soffer ret = (n == -1) ? -errno : 0; 17493a20013fSNir Soffer 17503a20013fSNir Soffer qemu_vfree(buf); 17513a20013fSNir Soffer return ret; 17523a20013fSNir Soffer } 17533a20013fSNir Soffer 175429cb4c01SKevin Wolf static int handle_aiocb_truncate(void *opaque) 175593f4e2ffSKevin Wolf { 175629cb4c01SKevin Wolf RawPosixAIOData *aiocb = opaque; 175793f4e2ffSKevin Wolf int result = 0; 175893f4e2ffSKevin Wolf int64_t current_length = 0; 175993f4e2ffSKevin Wolf char *buf = NULL; 176093f4e2ffSKevin Wolf struct stat st; 176193f4e2ffSKevin Wolf int fd = aiocb->aio_fildes; 176293f4e2ffSKevin Wolf int64_t offset = aiocb->aio_offset; 1763d57c44d0SKevin Wolf PreallocMode prealloc = aiocb->truncate.prealloc; 1764d57c44d0SKevin Wolf Error **errp = aiocb->truncate.errp; 176593f4e2ffSKevin Wolf 176693f4e2ffSKevin Wolf if (fstat(fd, &st) < 0) { 176793f4e2ffSKevin Wolf result = -errno; 176893f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not stat file"); 176993f4e2ffSKevin Wolf return result; 177093f4e2ffSKevin Wolf } 177193f4e2ffSKevin Wolf 177293f4e2ffSKevin Wolf current_length = st.st_size; 1773d57c44d0SKevin Wolf if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { 177493f4e2ffSKevin Wolf error_setg(errp, "Cannot use preallocation for shrinking files"); 177593f4e2ffSKevin Wolf return -ENOTSUP; 177693f4e2ffSKevin Wolf } 177793f4e2ffSKevin Wolf 1778d57c44d0SKevin Wolf switch (prealloc) { 177993f4e2ffSKevin Wolf #ifdef CONFIG_POSIX_FALLOCATE 178093f4e2ffSKevin Wolf case PREALLOC_MODE_FALLOC: 178193f4e2ffSKevin Wolf /* 178293f4e2ffSKevin Wolf * Truncating before posix_fallocate() makes it about twice slower on 178393f4e2ffSKevin Wolf * file systems that do not support fallocate(), trying to check if a 178493f4e2ffSKevin Wolf * block is allocated before allocating it, so don't do that here. 178593f4e2ffSKevin Wolf */ 178693f4e2ffSKevin Wolf if (offset != current_length) { 178793f4e2ffSKevin Wolf result = -posix_fallocate(fd, current_length, 178893f4e2ffSKevin Wolf offset - current_length); 178993f4e2ffSKevin Wolf if (result != 0) { 179093f4e2ffSKevin Wolf /* posix_fallocate() doesn't set errno. */ 179193f4e2ffSKevin Wolf error_setg_errno(errp, -result, 179293f4e2ffSKevin Wolf "Could not preallocate new data"); 17933a20013fSNir Soffer } else if (current_length == 0) { 17943a20013fSNir Soffer /* 17953a20013fSNir Soffer * posix_fallocate() uses fallocate() if the filesystem 17963a20013fSNir Soffer * supports it, or fallback to manually writing zeroes. If 17973a20013fSNir Soffer * fallocate() was used, unaligned reads from the fallocated 17983a20013fSNir Soffer * area in raw_probe_alignment() will succeed, hence we need to 17993a20013fSNir Soffer * allocate the first block. 18003a20013fSNir Soffer * 18013a20013fSNir Soffer * Optimize future alignment probing; ignore failures. 18023a20013fSNir Soffer */ 18033a20013fSNir Soffer allocate_first_block(fd, offset); 180493f4e2ffSKevin Wolf } 180593f4e2ffSKevin Wolf } else { 180693f4e2ffSKevin Wolf result = 0; 180793f4e2ffSKevin Wolf } 180893f4e2ffSKevin Wolf goto out; 180993f4e2ffSKevin Wolf #endif 181093f4e2ffSKevin Wolf case PREALLOC_MODE_FULL: 181193f4e2ffSKevin Wolf { 181293f4e2ffSKevin Wolf int64_t num = 0, left = offset - current_length; 181393f4e2ffSKevin Wolf off_t seek_result; 181493f4e2ffSKevin Wolf 181593f4e2ffSKevin Wolf /* 181693f4e2ffSKevin Wolf * Knowing the final size from the beginning could allow the file 181793f4e2ffSKevin Wolf * system driver to do less allocations and possibly avoid 181893f4e2ffSKevin Wolf * fragmentation of the file. 181993f4e2ffSKevin Wolf */ 182093f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 182193f4e2ffSKevin Wolf result = -errno; 182293f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 182393f4e2ffSKevin Wolf goto out; 182493f4e2ffSKevin Wolf } 182593f4e2ffSKevin Wolf 182693f4e2ffSKevin Wolf buf = g_malloc0(65536); 182793f4e2ffSKevin Wolf 182893f4e2ffSKevin Wolf seek_result = lseek(fd, current_length, SEEK_SET); 182993f4e2ffSKevin Wolf if (seek_result < 0) { 183093f4e2ffSKevin Wolf result = -errno; 183193f4e2ffSKevin Wolf error_setg_errno(errp, -result, 183293f4e2ffSKevin Wolf "Failed to seek to the old end of file"); 183393f4e2ffSKevin Wolf goto out; 183493f4e2ffSKevin Wolf } 183593f4e2ffSKevin Wolf 183693f4e2ffSKevin Wolf while (left > 0) { 183793f4e2ffSKevin Wolf num = MIN(left, 65536); 183893f4e2ffSKevin Wolf result = write(fd, buf, num); 183993f4e2ffSKevin Wolf if (result < 0) { 1840a1c81f4fSFam Zheng if (errno == EINTR) { 1841a1c81f4fSFam Zheng continue; 1842a1c81f4fSFam Zheng } 184393f4e2ffSKevin Wolf result = -errno; 184493f4e2ffSKevin Wolf error_setg_errno(errp, -result, 184593f4e2ffSKevin Wolf "Could not write zeros for preallocation"); 184693f4e2ffSKevin Wolf goto out; 184793f4e2ffSKevin Wolf } 184893f4e2ffSKevin Wolf left -= result; 184993f4e2ffSKevin Wolf } 185093f4e2ffSKevin Wolf if (result >= 0) { 185193f4e2ffSKevin Wolf result = fsync(fd); 185293f4e2ffSKevin Wolf if (result < 0) { 185393f4e2ffSKevin Wolf result = -errno; 185493f4e2ffSKevin Wolf error_setg_errno(errp, -result, 185593f4e2ffSKevin Wolf "Could not flush file to disk"); 185693f4e2ffSKevin Wolf goto out; 185793f4e2ffSKevin Wolf } 185893f4e2ffSKevin Wolf } 185993f4e2ffSKevin Wolf goto out; 186093f4e2ffSKevin Wolf } 186193f4e2ffSKevin Wolf case PREALLOC_MODE_OFF: 186293f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 186393f4e2ffSKevin Wolf result = -errno; 186493f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 18653a20013fSNir Soffer } else if (current_length == 0 && offset > current_length) { 18663a20013fSNir Soffer /* Optimize future alignment probing; ignore failures. */ 18673a20013fSNir Soffer allocate_first_block(fd, offset); 186893f4e2ffSKevin Wolf } 186993f4e2ffSKevin Wolf return result; 187093f4e2ffSKevin Wolf default: 187193f4e2ffSKevin Wolf result = -ENOTSUP; 187293f4e2ffSKevin Wolf error_setg(errp, "Unsupported preallocation mode: %s", 1873d57c44d0SKevin Wolf PreallocMode_str(prealloc)); 187493f4e2ffSKevin Wolf return result; 187593f4e2ffSKevin Wolf } 187693f4e2ffSKevin Wolf 187793f4e2ffSKevin Wolf out: 187893f4e2ffSKevin Wolf if (result < 0) { 187993f4e2ffSKevin Wolf if (ftruncate(fd, current_length) < 0) { 188093f4e2ffSKevin Wolf error_report("Failed to restore old file length: %s", 188193f4e2ffSKevin Wolf strerror(errno)); 188293f4e2ffSKevin Wolf } 188393f4e2ffSKevin Wolf } 188493f4e2ffSKevin Wolf 188593f4e2ffSKevin Wolf g_free(buf); 188693f4e2ffSKevin Wolf return result; 188793f4e2ffSKevin Wolf } 188893f4e2ffSKevin Wolf 18895d5de250SKevin Wolf static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs, 18905d5de250SKevin Wolf ThreadPoolFunc func, void *arg) 18915d5de250SKevin Wolf { 18925d5de250SKevin Wolf /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ 18935d5de250SKevin Wolf ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 18945d5de250SKevin Wolf return thread_pool_submit_co(pool, func, arg); 18955d5de250SKevin Wolf } 18965d5de250SKevin Wolf 1897c1bb86cdSEric Blake static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, 1898c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, int type) 1899c1bb86cdSEric Blake { 1900c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1901999e6b69SKevin Wolf RawPosixAIOData acb; 1902c1bb86cdSEric Blake 1903c1bb86cdSEric Blake if (fd_open(bs) < 0) 1904c1bb86cdSEric Blake return -EIO; 1905c1bb86cdSEric Blake 1906c1bb86cdSEric Blake /* 1907c6447510SAarushi Mehta * When using O_DIRECT, the request must be aligned to be able to use 1908c6447510SAarushi Mehta * either libaio or io_uring interface. If not fail back to regular thread 1909c6447510SAarushi Mehta * pool read/write code which emulates this for us if we 1910c6447510SAarushi Mehta * set QEMU_AIO_MISALIGNED. 1911c1bb86cdSEric Blake */ 1912c6447510SAarushi Mehta if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) { 1913c1bb86cdSEric Blake type |= QEMU_AIO_MISALIGNED; 1914c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 1915c6447510SAarushi Mehta } else if (s->use_linux_io_uring) { 1916c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 1917c6447510SAarushi Mehta assert(qiov->size == bytes); 1918c6447510SAarushi Mehta return luring_co_submit(bs, aio, s->fd, offset, qiov, type); 1919c6447510SAarushi Mehta #endif 1920c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1921c1bb86cdSEric Blake } else if (s->use_linux_aio) { 1922c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1923c1bb86cdSEric Blake assert(qiov->size == bytes); 1924c1bb86cdSEric Blake return laio_co_submit(bs, aio, s->fd, offset, qiov, type); 1925c1bb86cdSEric Blake #endif 1926c1bb86cdSEric Blake } 1927c1bb86cdSEric Blake 1928999e6b69SKevin Wolf acb = (RawPosixAIOData) { 1929999e6b69SKevin Wolf .bs = bs, 1930999e6b69SKevin Wolf .aio_fildes = s->fd, 1931999e6b69SKevin Wolf .aio_type = type, 1932999e6b69SKevin Wolf .aio_offset = offset, 1933999e6b69SKevin Wolf .aio_nbytes = bytes, 1934999e6b69SKevin Wolf .io = { 1935999e6b69SKevin Wolf .iov = qiov->iov, 1936999e6b69SKevin Wolf .niov = qiov->niov, 1937999e6b69SKevin Wolf }, 1938999e6b69SKevin Wolf }; 1939999e6b69SKevin Wolf 1940999e6b69SKevin Wolf assert(qiov->size == bytes); 1941999e6b69SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb); 1942c1bb86cdSEric Blake } 1943c1bb86cdSEric Blake 1944c1bb86cdSEric Blake static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset, 1945c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 1946c1bb86cdSEric Blake int flags) 1947c1bb86cdSEric Blake { 1948c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); 1949c1bb86cdSEric Blake } 1950c1bb86cdSEric Blake 1951c1bb86cdSEric Blake static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset, 1952c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 1953c1bb86cdSEric Blake int flags) 1954c1bb86cdSEric Blake { 1955c1bb86cdSEric Blake assert(flags == 0); 1956c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); 1957c1bb86cdSEric Blake } 1958c1bb86cdSEric Blake 1959c1bb86cdSEric Blake static void raw_aio_plug(BlockDriverState *bs) 1960c1bb86cdSEric Blake { 1961c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 1962c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1963c1bb86cdSEric Blake if (s->use_linux_aio) { 1964c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1965c1bb86cdSEric Blake laio_io_plug(bs, aio); 1966c1bb86cdSEric Blake } 1967c1bb86cdSEric Blake #endif 1968c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 1969c6447510SAarushi Mehta if (s->use_linux_io_uring) { 1970c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 1971c6447510SAarushi Mehta luring_io_plug(bs, aio); 1972c6447510SAarushi Mehta } 1973c6447510SAarushi Mehta #endif 1974c1bb86cdSEric Blake } 1975c1bb86cdSEric Blake 1976c1bb86cdSEric Blake static void raw_aio_unplug(BlockDriverState *bs) 1977c1bb86cdSEric Blake { 1978c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 1979c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1980c1bb86cdSEric Blake if (s->use_linux_aio) { 1981c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1982c1bb86cdSEric Blake laio_io_unplug(bs, aio); 1983c1bb86cdSEric Blake } 1984c1bb86cdSEric Blake #endif 1985c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 1986c6447510SAarushi Mehta if (s->use_linux_io_uring) { 1987c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 1988c6447510SAarushi Mehta luring_io_unplug(bs, aio); 1989c6447510SAarushi Mehta } 1990c6447510SAarushi Mehta #endif 1991c1bb86cdSEric Blake } 1992c1bb86cdSEric Blake 199333d70fb6SKevin Wolf static int raw_co_flush_to_disk(BlockDriverState *bs) 1994c1bb86cdSEric Blake { 1995c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 199606dc9bd5SKevin Wolf RawPosixAIOData acb; 199733d70fb6SKevin Wolf int ret; 1998c1bb86cdSEric Blake 199933d70fb6SKevin Wolf ret = fd_open(bs); 200033d70fb6SKevin Wolf if (ret < 0) { 200133d70fb6SKevin Wolf return ret; 200233d70fb6SKevin Wolf } 2003c1bb86cdSEric Blake 200406dc9bd5SKevin Wolf acb = (RawPosixAIOData) { 200506dc9bd5SKevin Wolf .bs = bs, 200606dc9bd5SKevin Wolf .aio_fildes = s->fd, 200706dc9bd5SKevin Wolf .aio_type = QEMU_AIO_FLUSH, 200806dc9bd5SKevin Wolf }; 200906dc9bd5SKevin Wolf 2010c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2011c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2012c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2013c6447510SAarushi Mehta return luring_co_submit(bs, aio, s->fd, 0, NULL, QEMU_AIO_FLUSH); 2014c6447510SAarushi Mehta } 2015c6447510SAarushi Mehta #endif 201606dc9bd5SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb); 2017c1bb86cdSEric Blake } 2018c1bb86cdSEric Blake 2019ed6e2161SNishanth Aravamudan static void raw_aio_attach_aio_context(BlockDriverState *bs, 2020ed6e2161SNishanth Aravamudan AioContext *new_context) 2021ed6e2161SNishanth Aravamudan { 2022c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 2023ed6e2161SNishanth Aravamudan #ifdef CONFIG_LINUX_AIO 2024ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 2025cb09104eSMarkus Armbruster Error *local_err = NULL; 2026ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(new_context, &local_err)) { 2027ed6e2161SNishanth Aravamudan error_reportf_err(local_err, "Unable to use native AIO, " 2028ed6e2161SNishanth Aravamudan "falling back to thread pool: "); 2029ed6e2161SNishanth Aravamudan s->use_linux_aio = false; 2030ed6e2161SNishanth Aravamudan } 2031ed6e2161SNishanth Aravamudan } 2032ed6e2161SNishanth Aravamudan #endif 2033c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2034c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2035c6447510SAarushi Mehta Error *local_err; 2036c6447510SAarushi Mehta if (!aio_setup_linux_io_uring(new_context, &local_err)) { 2037c6447510SAarushi Mehta error_reportf_err(local_err, "Unable to use linux io_uring, " 2038c6447510SAarushi Mehta "falling back to thread pool: "); 2039c6447510SAarushi Mehta s->use_linux_io_uring = false; 2040c6447510SAarushi Mehta } 2041c6447510SAarushi Mehta } 2042c6447510SAarushi Mehta #endif 2043ed6e2161SNishanth Aravamudan } 2044ed6e2161SNishanth Aravamudan 2045c1bb86cdSEric Blake static void raw_close(BlockDriverState *bs) 2046c1bb86cdSEric Blake { 2047c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2048c1bb86cdSEric Blake 2049c1bb86cdSEric Blake if (s->fd >= 0) { 2050c1bb86cdSEric Blake qemu_close(s->fd); 2051c1bb86cdSEric Blake s->fd = -1; 2052c1bb86cdSEric Blake } 2053c1bb86cdSEric Blake } 2054c1bb86cdSEric Blake 2055d0bc9e5dSMax Reitz /** 2056d0bc9e5dSMax Reitz * Truncates the given regular file @fd to @offset and, when growing, fills the 2057d0bc9e5dSMax Reitz * new space according to @prealloc. 2058d0bc9e5dSMax Reitz * 2059d0bc9e5dSMax Reitz * Returns: 0 on success, -errno on failure. 2060d0bc9e5dSMax Reitz */ 206193f4e2ffSKevin Wolf static int coroutine_fn 206293f4e2ffSKevin Wolf raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, 206393f4e2ffSKevin Wolf PreallocMode prealloc, Error **errp) 20649f63b07eSMax Reitz { 206529cb4c01SKevin Wolf RawPosixAIOData acb; 2066d0bc9e5dSMax Reitz 206729cb4c01SKevin Wolf acb = (RawPosixAIOData) { 206893f4e2ffSKevin Wolf .bs = bs, 206993f4e2ffSKevin Wolf .aio_fildes = fd, 207093f4e2ffSKevin Wolf .aio_type = QEMU_AIO_TRUNCATE, 207193f4e2ffSKevin Wolf .aio_offset = offset, 2072d57c44d0SKevin Wolf .truncate = { 207393f4e2ffSKevin Wolf .prealloc = prealloc, 207493f4e2ffSKevin Wolf .errp = errp, 2075d57c44d0SKevin Wolf }, 207693f4e2ffSKevin Wolf }; 2077d0bc9e5dSMax Reitz 207829cb4c01SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb); 20799f63b07eSMax Reitz } 20809f63b07eSMax Reitz 2081061ca8a3SKevin Wolf static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, 2082c80d8b06SMax Reitz bool exact, PreallocMode prealloc, 2083c80d8b06SMax Reitz Error **errp) 2084c1bb86cdSEric Blake { 2085c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2086c1bb86cdSEric Blake struct stat st; 2087f59adb32SMax Reitz int ret; 2088c1bb86cdSEric Blake 2089c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 2090f59adb32SMax Reitz ret = -errno; 2091f59adb32SMax Reitz error_setg_errno(errp, -ret, "Failed to fstat() the file"); 2092f59adb32SMax Reitz return ret; 2093c1bb86cdSEric Blake } 2094c1bb86cdSEric Blake 2095c1bb86cdSEric Blake if (S_ISREG(st.st_mode)) { 209682325ae5SMax Reitz /* Always resizes to the exact @offset */ 209793f4e2ffSKevin Wolf return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); 2098c1bb86cdSEric Blake } 209935d72602SMax Reitz 210035d72602SMax Reitz if (prealloc != PREALLOC_MODE_OFF) { 210135d72602SMax Reitz error_setg(errp, "Preallocation mode '%s' unsupported for this " 2102977c736fSMarkus Armbruster "non-regular file", PreallocMode_str(prealloc)); 210335d72602SMax Reitz return -ENOTSUP; 210435d72602SMax Reitz } 210535d72602SMax Reitz 210635d72602SMax Reitz if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 210782325ae5SMax Reitz int64_t cur_length = raw_getlength(bs); 210882325ae5SMax Reitz 210982325ae5SMax Reitz if (offset != cur_length && exact) { 211082325ae5SMax Reitz error_setg(errp, "Cannot resize device files"); 211182325ae5SMax Reitz return -ENOTSUP; 211282325ae5SMax Reitz } else if (offset > cur_length) { 2113f59adb32SMax Reitz error_setg(errp, "Cannot grow device files"); 2114c1bb86cdSEric Blake return -EINVAL; 2115c1bb86cdSEric Blake } 2116c1bb86cdSEric Blake } else { 2117f59adb32SMax Reitz error_setg(errp, "Resizing this file is not supported"); 2118c1bb86cdSEric Blake return -ENOTSUP; 2119c1bb86cdSEric Blake } 2120c1bb86cdSEric Blake 2121c1bb86cdSEric Blake return 0; 2122c1bb86cdSEric Blake } 2123c1bb86cdSEric Blake 2124c1bb86cdSEric Blake #ifdef __OpenBSD__ 2125c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2126c1bb86cdSEric Blake { 2127c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2128c1bb86cdSEric Blake int fd = s->fd; 2129c1bb86cdSEric Blake struct stat st; 2130c1bb86cdSEric Blake 2131c1bb86cdSEric Blake if (fstat(fd, &st)) 2132c1bb86cdSEric Blake return -errno; 2133c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2134c1bb86cdSEric Blake struct disklabel dl; 2135c1bb86cdSEric Blake 2136c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2137c1bb86cdSEric Blake return -errno; 2138c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2139c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2140c1bb86cdSEric Blake } else 2141c1bb86cdSEric Blake return st.st_size; 2142c1bb86cdSEric Blake } 2143c1bb86cdSEric Blake #elif defined(__NetBSD__) 2144c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2145c1bb86cdSEric Blake { 2146c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2147c1bb86cdSEric Blake int fd = s->fd; 2148c1bb86cdSEric Blake struct stat st; 2149c1bb86cdSEric Blake 2150c1bb86cdSEric Blake if (fstat(fd, &st)) 2151c1bb86cdSEric Blake return -errno; 2152c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2153c1bb86cdSEric Blake struct dkwedge_info dkw; 2154c1bb86cdSEric Blake 2155c1bb86cdSEric Blake if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { 2156c1bb86cdSEric Blake return dkw.dkw_size * 512; 2157c1bb86cdSEric Blake } else { 2158c1bb86cdSEric Blake struct disklabel dl; 2159c1bb86cdSEric Blake 2160c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2161c1bb86cdSEric Blake return -errno; 2162c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2163c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2164c1bb86cdSEric Blake } 2165c1bb86cdSEric Blake } else 2166c1bb86cdSEric Blake return st.st_size; 2167c1bb86cdSEric Blake } 2168c1bb86cdSEric Blake #elif defined(__sun__) 2169c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2170c1bb86cdSEric Blake { 2171c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2172c1bb86cdSEric Blake struct dk_minfo minfo; 2173c1bb86cdSEric Blake int ret; 2174c1bb86cdSEric Blake int64_t size; 2175c1bb86cdSEric Blake 2176c1bb86cdSEric Blake ret = fd_open(bs); 2177c1bb86cdSEric Blake if (ret < 0) { 2178c1bb86cdSEric Blake return ret; 2179c1bb86cdSEric Blake } 2180c1bb86cdSEric Blake 2181c1bb86cdSEric Blake /* 2182c1bb86cdSEric Blake * Use the DKIOCGMEDIAINFO ioctl to read the size. 2183c1bb86cdSEric Blake */ 2184c1bb86cdSEric Blake ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); 2185c1bb86cdSEric Blake if (ret != -1) { 2186c1bb86cdSEric Blake return minfo.dki_lbsize * minfo.dki_capacity; 2187c1bb86cdSEric Blake } 2188c1bb86cdSEric Blake 2189c1bb86cdSEric Blake /* 2190c1bb86cdSEric Blake * There are reports that lseek on some devices fails, but 2191c1bb86cdSEric Blake * irc discussion said that contingency on contingency was overkill. 2192c1bb86cdSEric Blake */ 2193c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2194c1bb86cdSEric Blake if (size < 0) { 2195c1bb86cdSEric Blake return -errno; 2196c1bb86cdSEric Blake } 2197c1bb86cdSEric Blake return size; 2198c1bb86cdSEric Blake } 2199c1bb86cdSEric Blake #elif defined(CONFIG_BSD) 2200c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2201c1bb86cdSEric Blake { 2202c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2203c1bb86cdSEric Blake int fd = s->fd; 2204c1bb86cdSEric Blake int64_t size; 2205c1bb86cdSEric Blake struct stat sb; 2206c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2207c1bb86cdSEric Blake int reopened = 0; 2208c1bb86cdSEric Blake #endif 2209c1bb86cdSEric Blake int ret; 2210c1bb86cdSEric Blake 2211c1bb86cdSEric Blake ret = fd_open(bs); 2212c1bb86cdSEric Blake if (ret < 0) 2213c1bb86cdSEric Blake return ret; 2214c1bb86cdSEric Blake 2215c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2216c1bb86cdSEric Blake again: 2217c1bb86cdSEric Blake #endif 2218c1bb86cdSEric Blake if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { 2219c1bb86cdSEric Blake #ifdef DIOCGMEDIASIZE 2220c1bb86cdSEric Blake if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) 2221c1bb86cdSEric Blake #elif defined(DIOCGPART) 2222c1bb86cdSEric Blake { 2223c1bb86cdSEric Blake struct partinfo pi; 2224c1bb86cdSEric Blake if (ioctl(fd, DIOCGPART, &pi) == 0) 2225c1bb86cdSEric Blake size = pi.media_size; 2226c1bb86cdSEric Blake else 2227c1bb86cdSEric Blake size = 0; 2228c1bb86cdSEric Blake } 2229c1bb86cdSEric Blake if (size == 0) 2230c1bb86cdSEric Blake #endif 2231c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2232c1bb86cdSEric Blake { 2233c1bb86cdSEric Blake uint64_t sectors = 0; 2234c1bb86cdSEric Blake uint32_t sector_size = 0; 2235c1bb86cdSEric Blake 2236c1bb86cdSEric Blake if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 2237c1bb86cdSEric Blake && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { 2238c1bb86cdSEric Blake size = sectors * sector_size; 2239c1bb86cdSEric Blake } else { 2240c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2241c1bb86cdSEric Blake if (size < 0) { 2242c1bb86cdSEric Blake return -errno; 2243c1bb86cdSEric Blake } 2244c1bb86cdSEric Blake } 2245c1bb86cdSEric Blake } 2246c1bb86cdSEric Blake #else 2247c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2248c1bb86cdSEric Blake if (size < 0) { 2249c1bb86cdSEric Blake return -errno; 2250c1bb86cdSEric Blake } 2251c1bb86cdSEric Blake #endif 2252c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 2253c1bb86cdSEric Blake switch(s->type) { 2254c1bb86cdSEric Blake case FTYPE_CD: 2255c1bb86cdSEric Blake /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ 2256c1bb86cdSEric Blake if (size == 2048LL * (unsigned)-1) 2257c1bb86cdSEric Blake size = 0; 2258c1bb86cdSEric Blake /* XXX no disc? maybe we need to reopen... */ 2259c1bb86cdSEric Blake if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { 2260c1bb86cdSEric Blake reopened = 1; 2261c1bb86cdSEric Blake goto again; 2262c1bb86cdSEric Blake } 2263c1bb86cdSEric Blake } 2264c1bb86cdSEric Blake #endif 2265c1bb86cdSEric Blake } else { 2266c1bb86cdSEric Blake size = lseek(fd, 0, SEEK_END); 2267c1bb86cdSEric Blake if (size < 0) { 2268c1bb86cdSEric Blake return -errno; 2269c1bb86cdSEric Blake } 2270c1bb86cdSEric Blake } 2271c1bb86cdSEric Blake return size; 2272c1bb86cdSEric Blake } 2273c1bb86cdSEric Blake #else 2274c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2275c1bb86cdSEric Blake { 2276c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2277c1bb86cdSEric Blake int ret; 2278c1bb86cdSEric Blake int64_t size; 2279c1bb86cdSEric Blake 2280c1bb86cdSEric Blake ret = fd_open(bs); 2281c1bb86cdSEric Blake if (ret < 0) { 2282c1bb86cdSEric Blake return ret; 2283c1bb86cdSEric Blake } 2284c1bb86cdSEric Blake 2285c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2286c1bb86cdSEric Blake if (size < 0) { 2287c1bb86cdSEric Blake return -errno; 2288c1bb86cdSEric Blake } 2289c1bb86cdSEric Blake return size; 2290c1bb86cdSEric Blake } 2291c1bb86cdSEric Blake #endif 2292c1bb86cdSEric Blake 2293c1bb86cdSEric Blake static int64_t raw_get_allocated_file_size(BlockDriverState *bs) 2294c1bb86cdSEric Blake { 2295c1bb86cdSEric Blake struct stat st; 2296c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2297c1bb86cdSEric Blake 2298c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 2299c1bb86cdSEric Blake return -errno; 2300c1bb86cdSEric Blake } 2301c1bb86cdSEric Blake return (int64_t)st.st_blocks * 512; 2302c1bb86cdSEric Blake } 2303c1bb86cdSEric Blake 230493f4e2ffSKevin Wolf static int coroutine_fn 230593f4e2ffSKevin Wolf raw_co_create(BlockdevCreateOptions *options, Error **errp) 2306c1bb86cdSEric Blake { 2307927f11e1SKevin Wolf BlockdevCreateOptionsFile *file_opts; 23087c20c808SMax Reitz Error *local_err = NULL; 2309c1bb86cdSEric Blake int fd; 2310d815efcaSMax Reitz uint64_t perm, shared; 2311c1bb86cdSEric Blake int result = 0; 2312c1bb86cdSEric Blake 2313927f11e1SKevin Wolf /* Validate options and set default values */ 2314927f11e1SKevin Wolf assert(options->driver == BLOCKDEV_DRIVER_FILE); 2315927f11e1SKevin Wolf file_opts = &options->u.file; 2316c1bb86cdSEric Blake 2317927f11e1SKevin Wolf if (!file_opts->has_nocow) { 2318927f11e1SKevin Wolf file_opts->nocow = false; 2319927f11e1SKevin Wolf } 2320927f11e1SKevin Wolf if (!file_opts->has_preallocation) { 2321927f11e1SKevin Wolf file_opts->preallocation = PREALLOC_MODE_OFF; 2322c1bb86cdSEric Blake } 2323c1bb86cdSEric Blake 2324927f11e1SKevin Wolf /* Create file */ 2325b8cf1913SMax Reitz fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_BINARY, 0644); 2326c1bb86cdSEric Blake if (fd < 0) { 2327c1bb86cdSEric Blake result = -errno; 2328c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not create file"); 2329c1bb86cdSEric Blake goto out; 2330c1bb86cdSEric Blake } 2331c1bb86cdSEric Blake 2332b8cf1913SMax Reitz /* Take permissions: We want to discard everything, so we need 2333b8cf1913SMax Reitz * BLK_PERM_WRITE; and truncation to the desired size requires 2334b8cf1913SMax Reitz * BLK_PERM_RESIZE. 2335b8cf1913SMax Reitz * On the other hand, we cannot share the RESIZE permission 2336b8cf1913SMax Reitz * because we promise that after this function, the file has the 2337b8cf1913SMax Reitz * size given in the options. If someone else were to resize it 2338b8cf1913SMax Reitz * concurrently, we could not guarantee that. 2339b8cf1913SMax Reitz * Note that after this function, we can no longer guarantee that 2340b8cf1913SMax Reitz * the file is not touched by a third party, so it may be resized 2341b8cf1913SMax Reitz * then. */ 2342b8cf1913SMax Reitz perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; 2343b8cf1913SMax Reitz shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; 2344b8cf1913SMax Reitz 2345b8cf1913SMax Reitz /* Step one: Take locks */ 23462996ffadSFam Zheng result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); 2347b8cf1913SMax Reitz if (result < 0) { 2348b8cf1913SMax Reitz goto out_close; 2349b8cf1913SMax Reitz } 2350b8cf1913SMax Reitz 2351b8cf1913SMax Reitz /* Step two: Check that nobody else has taken conflicting locks */ 2352b8cf1913SMax Reitz result = raw_check_lock_bytes(fd, perm, shared, errp); 2353b8cf1913SMax Reitz if (result < 0) { 2354b857431dSFam Zheng error_append_hint(errp, 2355b857431dSFam Zheng "Is another process using the image [%s]?\n", 2356b857431dSFam Zheng file_opts->filename); 23577c20c808SMax Reitz goto out_unlock; 2358b8cf1913SMax Reitz } 2359b8cf1913SMax Reitz 2360b8cf1913SMax Reitz /* Clear the file by truncating it to 0 */ 236193f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); 2362b8cf1913SMax Reitz if (result < 0) { 23637c20c808SMax Reitz goto out_unlock; 2364b8cf1913SMax Reitz } 2365b8cf1913SMax Reitz 2366927f11e1SKevin Wolf if (file_opts->nocow) { 2367c1bb86cdSEric Blake #ifdef __linux__ 2368c1bb86cdSEric Blake /* Set NOCOW flag to solve performance issue on fs like btrfs. 2369c1bb86cdSEric Blake * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value 2370c1bb86cdSEric Blake * will be ignored since any failure of this operation should not 2371c1bb86cdSEric Blake * block the left work. 2372c1bb86cdSEric Blake */ 2373c1bb86cdSEric Blake int attr; 2374c1bb86cdSEric Blake if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { 2375c1bb86cdSEric Blake attr |= FS_NOCOW_FL; 2376c1bb86cdSEric Blake ioctl(fd, FS_IOC_SETFLAGS, &attr); 2377c1bb86cdSEric Blake } 2378c1bb86cdSEric Blake #endif 2379c1bb86cdSEric Blake } 2380c1bb86cdSEric Blake 2381b8cf1913SMax Reitz /* Resize and potentially preallocate the file to the desired 2382b8cf1913SMax Reitz * final size */ 238393f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, file_opts->size, 238493f4e2ffSKevin Wolf file_opts->preallocation, errp); 23859f63b07eSMax Reitz if (result < 0) { 23867c20c808SMax Reitz goto out_unlock; 23877c20c808SMax Reitz } 23887c20c808SMax Reitz 23897c20c808SMax Reitz out_unlock: 23902996ffadSFam Zheng raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); 23917c20c808SMax Reitz if (local_err) { 23927c20c808SMax Reitz /* The above call should not fail, and if it does, that does 23937c20c808SMax Reitz * not mean the whole creation operation has failed. So 23947c20c808SMax Reitz * report it the user for their convenience, but do not report 23957c20c808SMax Reitz * it to the caller. */ 2396db0754dfSFam Zheng warn_report_err(local_err); 23975a1dad9dSNir Soffer } 23985a1dad9dSNir Soffer 23995a1dad9dSNir Soffer out_close: 2400c1bb86cdSEric Blake if (qemu_close(fd) != 0 && result == 0) { 2401c1bb86cdSEric Blake result = -errno; 2402c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not close the new file"); 2403c1bb86cdSEric Blake } 2404c1bb86cdSEric Blake out: 2405c1bb86cdSEric Blake return result; 2406c1bb86cdSEric Blake } 2407c1bb86cdSEric Blake 2408927f11e1SKevin Wolf static int coroutine_fn raw_co_create_opts(const char *filename, QemuOpts *opts, 2409927f11e1SKevin Wolf Error **errp) 2410927f11e1SKevin Wolf { 2411927f11e1SKevin Wolf BlockdevCreateOptions options; 2412927f11e1SKevin Wolf int64_t total_size = 0; 2413927f11e1SKevin Wolf bool nocow = false; 2414927f11e1SKevin Wolf PreallocMode prealloc; 2415927f11e1SKevin Wolf char *buf = NULL; 2416927f11e1SKevin Wolf Error *local_err = NULL; 2417927f11e1SKevin Wolf 2418927f11e1SKevin Wolf /* Skip file: protocol prefix */ 2419927f11e1SKevin Wolf strstart(filename, "file:", &filename); 2420927f11e1SKevin Wolf 2421927f11e1SKevin Wolf /* Read out options */ 2422927f11e1SKevin Wolf total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2423927f11e1SKevin Wolf BDRV_SECTOR_SIZE); 2424927f11e1SKevin Wolf nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); 2425927f11e1SKevin Wolf buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2426927f11e1SKevin Wolf prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, 2427927f11e1SKevin Wolf PREALLOC_MODE_OFF, &local_err); 2428927f11e1SKevin Wolf g_free(buf); 2429927f11e1SKevin Wolf if (local_err) { 2430927f11e1SKevin Wolf error_propagate(errp, local_err); 2431927f11e1SKevin Wolf return -EINVAL; 2432927f11e1SKevin Wolf } 2433927f11e1SKevin Wolf 2434927f11e1SKevin Wolf options = (BlockdevCreateOptions) { 2435927f11e1SKevin Wolf .driver = BLOCKDEV_DRIVER_FILE, 2436927f11e1SKevin Wolf .u.file = { 2437927f11e1SKevin Wolf .filename = (char *) filename, 2438927f11e1SKevin Wolf .size = total_size, 2439927f11e1SKevin Wolf .has_preallocation = true, 2440927f11e1SKevin Wolf .preallocation = prealloc, 2441927f11e1SKevin Wolf .has_nocow = true, 2442927f11e1SKevin Wolf .nocow = nocow, 2443927f11e1SKevin Wolf }, 2444927f11e1SKevin Wolf }; 2445927f11e1SKevin Wolf return raw_co_create(&options, errp); 2446927f11e1SKevin Wolf } 2447927f11e1SKevin Wolf 2448*9bffae14SDaniel Henrique Barboza static int coroutine_fn raw_co_delete_file(BlockDriverState *bs, 2449*9bffae14SDaniel Henrique Barboza Error **errp) 2450*9bffae14SDaniel Henrique Barboza { 2451*9bffae14SDaniel Henrique Barboza struct stat st; 2452*9bffae14SDaniel Henrique Barboza int ret; 2453*9bffae14SDaniel Henrique Barboza 2454*9bffae14SDaniel Henrique Barboza if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) { 2455*9bffae14SDaniel Henrique Barboza error_setg_errno(errp, ENOENT, "%s is not a regular file", 2456*9bffae14SDaniel Henrique Barboza bs->filename); 2457*9bffae14SDaniel Henrique Barboza return -ENOENT; 2458*9bffae14SDaniel Henrique Barboza } 2459*9bffae14SDaniel Henrique Barboza 2460*9bffae14SDaniel Henrique Barboza ret = unlink(bs->filename); 2461*9bffae14SDaniel Henrique Barboza if (ret < 0) { 2462*9bffae14SDaniel Henrique Barboza ret = -errno; 2463*9bffae14SDaniel Henrique Barboza error_setg_errno(errp, -ret, "Error when deleting file %s", 2464*9bffae14SDaniel Henrique Barboza bs->filename); 2465*9bffae14SDaniel Henrique Barboza } 2466*9bffae14SDaniel Henrique Barboza 2467*9bffae14SDaniel Henrique Barboza return ret; 2468*9bffae14SDaniel Henrique Barboza } 2469*9bffae14SDaniel Henrique Barboza 2470c1bb86cdSEric Blake /* 2471c1bb86cdSEric Blake * Find allocation range in @bs around offset @start. 2472c1bb86cdSEric Blake * May change underlying file descriptor's file offset. 2473c1bb86cdSEric Blake * If @start is not in a hole, store @start in @data, and the 2474c1bb86cdSEric Blake * beginning of the next hole in @hole, and return 0. 2475c1bb86cdSEric Blake * If @start is in a non-trailing hole, store @start in @hole and the 2476c1bb86cdSEric Blake * beginning of the next non-hole in @data, and return 0. 2477c1bb86cdSEric Blake * If @start is in a trailing hole or beyond EOF, return -ENXIO. 2478c1bb86cdSEric Blake * If we can't find out, return a negative errno other than -ENXIO. 2479c1bb86cdSEric Blake */ 2480c1bb86cdSEric Blake static int find_allocation(BlockDriverState *bs, off_t start, 2481c1bb86cdSEric Blake off_t *data, off_t *hole) 2482c1bb86cdSEric Blake { 2483c1bb86cdSEric Blake #if defined SEEK_HOLE && defined SEEK_DATA 2484c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2485c1bb86cdSEric Blake off_t offs; 2486c1bb86cdSEric Blake 2487c1bb86cdSEric Blake /* 2488c1bb86cdSEric Blake * SEEK_DATA cases: 2489c1bb86cdSEric Blake * D1. offs == start: start is in data 2490c1bb86cdSEric Blake * D2. offs > start: start is in a hole, next data at offs 2491c1bb86cdSEric Blake * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 2492c1bb86cdSEric Blake * or start is beyond EOF 2493c1bb86cdSEric Blake * If the latter happens, the file has been truncated behind 2494c1bb86cdSEric Blake * our back since we opened it. All bets are off then. 2495c1bb86cdSEric Blake * Treating like a trailing hole is simplest. 2496c1bb86cdSEric Blake * D4. offs < 0, errno != ENXIO: we learned nothing 2497c1bb86cdSEric Blake */ 2498c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_DATA); 2499c1bb86cdSEric Blake if (offs < 0) { 2500c1bb86cdSEric Blake return -errno; /* D3 or D4 */ 2501c1bb86cdSEric Blake } 2502a03083a0SJeff Cody 2503a03083a0SJeff Cody if (offs < start) { 2504a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2505a03083a0SJeff Cody * -EIO in this case, and we'll treat it like D4. */ 2506a03083a0SJeff Cody return -EIO; 2507a03083a0SJeff Cody } 2508c1bb86cdSEric Blake 2509c1bb86cdSEric Blake if (offs > start) { 2510c1bb86cdSEric Blake /* D2: in hole, next data at offs */ 2511c1bb86cdSEric Blake *hole = start; 2512c1bb86cdSEric Blake *data = offs; 2513c1bb86cdSEric Blake return 0; 2514c1bb86cdSEric Blake } 2515c1bb86cdSEric Blake 2516c1bb86cdSEric Blake /* D1: in data, end not yet known */ 2517c1bb86cdSEric Blake 2518c1bb86cdSEric Blake /* 2519c1bb86cdSEric Blake * SEEK_HOLE cases: 2520c1bb86cdSEric Blake * H1. offs == start: start is in a hole 2521c1bb86cdSEric Blake * If this happens here, a hole has been dug behind our back 2522c1bb86cdSEric Blake * since the previous lseek(). 2523c1bb86cdSEric Blake * H2. offs > start: either start is in data, next hole at offs, 2524c1bb86cdSEric Blake * or start is in trailing hole, EOF at offs 2525c1bb86cdSEric Blake * Linux treats trailing holes like any other hole: offs == 2526c1bb86cdSEric Blake * start. Solaris seeks to EOF instead: offs > start (blech). 2527c1bb86cdSEric Blake * If that happens here, a hole has been dug behind our back 2528c1bb86cdSEric Blake * since the previous lseek(). 2529c1bb86cdSEric Blake * H3. offs < 0, errno = ENXIO: start is beyond EOF 2530c1bb86cdSEric Blake * If this happens, the file has been truncated behind our 2531c1bb86cdSEric Blake * back since we opened it. Treat it like a trailing hole. 2532c1bb86cdSEric Blake * H4. offs < 0, errno != ENXIO: we learned nothing 2533c1bb86cdSEric Blake * Pretend we know nothing at all, i.e. "forget" about D1. 2534c1bb86cdSEric Blake */ 2535c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_HOLE); 2536c1bb86cdSEric Blake if (offs < 0) { 2537c1bb86cdSEric Blake return -errno; /* D1 and (H3 or H4) */ 2538c1bb86cdSEric Blake } 2539a03083a0SJeff Cody 2540a03083a0SJeff Cody if (offs < start) { 2541a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2542a03083a0SJeff Cody * -EIO in this case, and we'll treat it like H4. */ 2543a03083a0SJeff Cody return -EIO; 2544a03083a0SJeff Cody } 2545c1bb86cdSEric Blake 2546c1bb86cdSEric Blake if (offs > start) { 2547c1bb86cdSEric Blake /* 2548c1bb86cdSEric Blake * D1 and H2: either in data, next hole at offs, or it was in 2549c1bb86cdSEric Blake * data but is now in a trailing hole. In the latter case, 2550c1bb86cdSEric Blake * all bets are off. Treating it as if it there was data all 2551c1bb86cdSEric Blake * the way to EOF is safe, so simply do that. 2552c1bb86cdSEric Blake */ 2553c1bb86cdSEric Blake *data = start; 2554c1bb86cdSEric Blake *hole = offs; 2555c1bb86cdSEric Blake return 0; 2556c1bb86cdSEric Blake } 2557c1bb86cdSEric Blake 2558c1bb86cdSEric Blake /* D1 and H1 */ 2559c1bb86cdSEric Blake return -EBUSY; 2560c1bb86cdSEric Blake #else 2561c1bb86cdSEric Blake return -ENOTSUP; 2562c1bb86cdSEric Blake #endif 2563c1bb86cdSEric Blake } 2564c1bb86cdSEric Blake 2565c1bb86cdSEric Blake /* 2566a290f085SEric Blake * Returns the allocation status of the specified offset. 2567c1bb86cdSEric Blake * 2568a290f085SEric Blake * The block layer guarantees 'offset' and 'bytes' are within bounds. 2569c1bb86cdSEric Blake * 2570a290f085SEric Blake * 'pnum' is set to the number of bytes (including and immediately following 2571a290f085SEric Blake * the specified offset) that are known to be in the same 2572c1bb86cdSEric Blake * allocated/unallocated state. 2573c1bb86cdSEric Blake * 2574a290f085SEric Blake * 'bytes' is the max value 'pnum' should be set to. 2575c1bb86cdSEric Blake */ 2576a290f085SEric Blake static int coroutine_fn raw_co_block_status(BlockDriverState *bs, 2577a290f085SEric Blake bool want_zero, 2578a290f085SEric Blake int64_t offset, 2579a290f085SEric Blake int64_t bytes, int64_t *pnum, 2580a290f085SEric Blake int64_t *map, 2581c1bb86cdSEric Blake BlockDriverState **file) 2582c1bb86cdSEric Blake { 2583a290f085SEric Blake off_t data = 0, hole = 0; 2584c1bb86cdSEric Blake int ret; 2585c1bb86cdSEric Blake 25869c3db310SMax Reitz assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); 25879c3db310SMax Reitz 2588c1bb86cdSEric Blake ret = fd_open(bs); 2589c1bb86cdSEric Blake if (ret < 0) { 2590c1bb86cdSEric Blake return ret; 2591c1bb86cdSEric Blake } 2592c1bb86cdSEric Blake 2593a290f085SEric Blake if (!want_zero) { 2594a290f085SEric Blake *pnum = bytes; 2595a290f085SEric Blake *map = offset; 2596a290f085SEric Blake *file = bs; 2597a290f085SEric Blake return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 2598c1bb86cdSEric Blake } 2599c1bb86cdSEric Blake 2600a290f085SEric Blake ret = find_allocation(bs, offset, &data, &hole); 2601c1bb86cdSEric Blake if (ret == -ENXIO) { 2602c1bb86cdSEric Blake /* Trailing hole */ 2603a290f085SEric Blake *pnum = bytes; 2604c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2605c1bb86cdSEric Blake } else if (ret < 0) { 2606c1bb86cdSEric Blake /* No info available, so pretend there are no holes */ 2607a290f085SEric Blake *pnum = bytes; 2608c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2609a290f085SEric Blake } else if (data == offset) { 2610a290f085SEric Blake /* On a data extent, compute bytes to the end of the extent, 2611c1bb86cdSEric Blake * possibly including a partial sector at EOF. */ 2612a290f085SEric Blake *pnum = MIN(bytes, hole - offset); 26139c3db310SMax Reitz 26149c3db310SMax Reitz /* 26159c3db310SMax Reitz * We are not allowed to return partial sectors, though, so 26169c3db310SMax Reitz * round up if necessary. 26179c3db310SMax Reitz */ 26189c3db310SMax Reitz if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { 26199c3db310SMax Reitz int64_t file_length = raw_getlength(bs); 26209c3db310SMax Reitz if (file_length > 0) { 26219c3db310SMax Reitz /* Ignore errors, this is just a safeguard */ 26229c3db310SMax Reitz assert(hole == file_length); 26239c3db310SMax Reitz } 26249c3db310SMax Reitz *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); 26259c3db310SMax Reitz } 26269c3db310SMax Reitz 2627c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2628c1bb86cdSEric Blake } else { 2629a290f085SEric Blake /* On a hole, compute bytes to the beginning of the next extent. */ 2630a290f085SEric Blake assert(hole == offset); 2631a290f085SEric Blake *pnum = MIN(bytes, data - offset); 2632c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2633c1bb86cdSEric Blake } 2634a290f085SEric Blake *map = offset; 2635c1bb86cdSEric Blake *file = bs; 2636a290f085SEric Blake return ret | BDRV_BLOCK_OFFSET_VALID; 2637c1bb86cdSEric Blake } 2638c1bb86cdSEric Blake 263931be8a2aSStefan Hajnoczi #if defined(__linux__) 264031be8a2aSStefan Hajnoczi /* Verify that the file is not in the page cache */ 264131be8a2aSStefan Hajnoczi static void check_cache_dropped(BlockDriverState *bs, Error **errp) 264231be8a2aSStefan Hajnoczi { 264331be8a2aSStefan Hajnoczi const size_t window_size = 128 * 1024 * 1024; 264431be8a2aSStefan Hajnoczi BDRVRawState *s = bs->opaque; 264531be8a2aSStefan Hajnoczi void *window = NULL; 264631be8a2aSStefan Hajnoczi size_t length = 0; 264731be8a2aSStefan Hajnoczi unsigned char *vec; 264831be8a2aSStefan Hajnoczi size_t page_size; 264931be8a2aSStefan Hajnoczi off_t offset; 265031be8a2aSStefan Hajnoczi off_t end; 265131be8a2aSStefan Hajnoczi 265231be8a2aSStefan Hajnoczi /* mincore(2) page status information requires 1 byte per page */ 265331be8a2aSStefan Hajnoczi page_size = sysconf(_SC_PAGESIZE); 265431be8a2aSStefan Hajnoczi vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); 265531be8a2aSStefan Hajnoczi 265631be8a2aSStefan Hajnoczi end = raw_getlength(bs); 265731be8a2aSStefan Hajnoczi 265831be8a2aSStefan Hajnoczi for (offset = 0; offset < end; offset += window_size) { 265931be8a2aSStefan Hajnoczi void *new_window; 266031be8a2aSStefan Hajnoczi size_t new_length; 266131be8a2aSStefan Hajnoczi size_t vec_end; 266231be8a2aSStefan Hajnoczi size_t i; 266331be8a2aSStefan Hajnoczi int ret; 266431be8a2aSStefan Hajnoczi 266531be8a2aSStefan Hajnoczi /* Unmap previous window if size has changed */ 266631be8a2aSStefan Hajnoczi new_length = MIN(end - offset, window_size); 266731be8a2aSStefan Hajnoczi if (new_length != length) { 266831be8a2aSStefan Hajnoczi munmap(window, length); 266931be8a2aSStefan Hajnoczi window = NULL; 267031be8a2aSStefan Hajnoczi length = 0; 267131be8a2aSStefan Hajnoczi } 267231be8a2aSStefan Hajnoczi 267331be8a2aSStefan Hajnoczi new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, 267431be8a2aSStefan Hajnoczi s->fd, offset); 267531be8a2aSStefan Hajnoczi if (new_window == MAP_FAILED) { 267631be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mmap failed"); 267731be8a2aSStefan Hajnoczi break; 267831be8a2aSStefan Hajnoczi } 267931be8a2aSStefan Hajnoczi 268031be8a2aSStefan Hajnoczi window = new_window; 268131be8a2aSStefan Hajnoczi length = new_length; 268231be8a2aSStefan Hajnoczi 268331be8a2aSStefan Hajnoczi ret = mincore(window, length, vec); 268431be8a2aSStefan Hajnoczi if (ret < 0) { 268531be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mincore failed"); 268631be8a2aSStefan Hajnoczi break; 268731be8a2aSStefan Hajnoczi } 268831be8a2aSStefan Hajnoczi 268931be8a2aSStefan Hajnoczi vec_end = DIV_ROUND_UP(length, page_size); 269031be8a2aSStefan Hajnoczi for (i = 0; i < vec_end; i++) { 269131be8a2aSStefan Hajnoczi if (vec[i] & 0x1) { 269231be8a2aSStefan Hajnoczi error_setg(errp, "page cache still in use!"); 269331be8a2aSStefan Hajnoczi break; 269431be8a2aSStefan Hajnoczi } 269531be8a2aSStefan Hajnoczi } 269631be8a2aSStefan Hajnoczi } 269731be8a2aSStefan Hajnoczi 269831be8a2aSStefan Hajnoczi if (window) { 269931be8a2aSStefan Hajnoczi munmap(window, length); 270031be8a2aSStefan Hajnoczi } 270131be8a2aSStefan Hajnoczi 270231be8a2aSStefan Hajnoczi g_free(vec); 270331be8a2aSStefan Hajnoczi } 270431be8a2aSStefan Hajnoczi #endif /* __linux__ */ 270531be8a2aSStefan Hajnoczi 2706dd577a26SStefan Hajnoczi static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, 2707dd577a26SStefan Hajnoczi Error **errp) 2708dd577a26SStefan Hajnoczi { 2709dd577a26SStefan Hajnoczi BDRVRawState *s = bs->opaque; 2710dd577a26SStefan Hajnoczi int ret; 2711dd577a26SStefan Hajnoczi 2712dd577a26SStefan Hajnoczi ret = fd_open(bs); 2713dd577a26SStefan Hajnoczi if (ret < 0) { 2714dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "The file descriptor is not open"); 2715dd577a26SStefan Hajnoczi return; 2716dd577a26SStefan Hajnoczi } 2717dd577a26SStefan Hajnoczi 2718f357fcd8SStefan Hajnoczi if (!s->drop_cache) { 2719f357fcd8SStefan Hajnoczi return; 2720f357fcd8SStefan Hajnoczi } 2721f357fcd8SStefan Hajnoczi 2722dd577a26SStefan Hajnoczi if (s->open_flags & O_DIRECT) { 2723dd577a26SStefan Hajnoczi return; /* No host kernel page cache */ 2724dd577a26SStefan Hajnoczi } 2725dd577a26SStefan Hajnoczi 2726dd577a26SStefan Hajnoczi #if defined(__linux__) 2727dd577a26SStefan Hajnoczi /* This sets the scene for the next syscall... */ 2728dd577a26SStefan Hajnoczi ret = bdrv_co_flush(bs); 2729dd577a26SStefan Hajnoczi if (ret < 0) { 2730dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "flush failed"); 2731dd577a26SStefan Hajnoczi return; 2732dd577a26SStefan Hajnoczi } 2733dd577a26SStefan Hajnoczi 2734dd577a26SStefan Hajnoczi /* Linux does not invalidate pages that are dirty, locked, or mmapped by a 2735dd577a26SStefan Hajnoczi * process. These limitations are okay because we just fsynced the file, 2736dd577a26SStefan Hajnoczi * we don't use mmap, and the file should not be in use by other processes. 2737dd577a26SStefan Hajnoczi */ 2738dd577a26SStefan Hajnoczi ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); 2739dd577a26SStefan Hajnoczi if (ret != 0) { /* the return value is a positive errno */ 2740dd577a26SStefan Hajnoczi error_setg_errno(errp, ret, "fadvise failed"); 2741dd577a26SStefan Hajnoczi return; 2742dd577a26SStefan Hajnoczi } 274331be8a2aSStefan Hajnoczi 274431be8a2aSStefan Hajnoczi if (s->check_cache_dropped) { 274531be8a2aSStefan Hajnoczi check_cache_dropped(bs, errp); 274631be8a2aSStefan Hajnoczi } 2747dd577a26SStefan Hajnoczi #else /* __linux__ */ 2748dd577a26SStefan Hajnoczi /* Do nothing. Live migration to a remote host with cache.direct=off is 2749dd577a26SStefan Hajnoczi * unsupported on other host operating systems. Cache consistency issues 2750dd577a26SStefan Hajnoczi * may occur but no error is reported here, partly because that's the 2751dd577a26SStefan Hajnoczi * historical behavior and partly because it's hard to differentiate valid 2752dd577a26SStefan Hajnoczi * configurations that should not cause errors. 2753dd577a26SStefan Hajnoczi */ 2754dd577a26SStefan Hajnoczi #endif /* !__linux__ */ 2755dd577a26SStefan Hajnoczi } 2756dd577a26SStefan Hajnoczi 27571c450366SAnton Nefedov static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret) 27581c450366SAnton Nefedov { 27591c450366SAnton Nefedov if (ret) { 27601c450366SAnton Nefedov s->stats.discard_nb_failed++; 27611c450366SAnton Nefedov } else { 27621c450366SAnton Nefedov s->stats.discard_nb_ok++; 27631c450366SAnton Nefedov s->stats.discard_bytes_ok += nbytes; 27641c450366SAnton Nefedov } 27651c450366SAnton Nefedov } 27661c450366SAnton Nefedov 276733d70fb6SKevin Wolf static coroutine_fn int 276846ee0f46SKevin Wolf raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev) 2769c1bb86cdSEric Blake { 2770c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 277146ee0f46SKevin Wolf RawPosixAIOData acb; 27721c450366SAnton Nefedov int ret; 2773c1bb86cdSEric Blake 277446ee0f46SKevin Wolf acb = (RawPosixAIOData) { 277546ee0f46SKevin Wolf .bs = bs, 277646ee0f46SKevin Wolf .aio_fildes = s->fd, 277746ee0f46SKevin Wolf .aio_type = QEMU_AIO_DISCARD, 277846ee0f46SKevin Wolf .aio_offset = offset, 277946ee0f46SKevin Wolf .aio_nbytes = bytes, 278046ee0f46SKevin Wolf }; 278146ee0f46SKevin Wolf 278246ee0f46SKevin Wolf if (blkdev) { 278346ee0f46SKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 278446ee0f46SKevin Wolf } 278546ee0f46SKevin Wolf 27861c450366SAnton Nefedov ret = raw_thread_pool_submit(bs, handle_aiocb_discard, &acb); 27871c450366SAnton Nefedov raw_account_discard(s, bytes, ret); 27881c450366SAnton Nefedov return ret; 278946ee0f46SKevin Wolf } 279046ee0f46SKevin Wolf 279146ee0f46SKevin Wolf static coroutine_fn int 279246ee0f46SKevin Wolf raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 279346ee0f46SKevin Wolf { 279446ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, false); 2795c1bb86cdSEric Blake } 2796c1bb86cdSEric Blake 27977154d8aeSKevin Wolf static int coroutine_fn 27987154d8aeSKevin Wolf raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, 27997154d8aeSKevin Wolf BdrvRequestFlags flags, bool blkdev) 28007154d8aeSKevin Wolf { 28017154d8aeSKevin Wolf BDRVRawState *s = bs->opaque; 28027154d8aeSKevin Wolf RawPosixAIOData acb; 28037154d8aeSKevin Wolf ThreadPoolFunc *handler; 28047154d8aeSKevin Wolf 2805292d06b9SMax Reitz #ifdef CONFIG_FALLOCATE 2806292d06b9SMax Reitz if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { 2807292d06b9SMax Reitz BdrvTrackedRequest *req; 2808292d06b9SMax Reitz uint64_t end; 2809292d06b9SMax Reitz 2810292d06b9SMax Reitz /* 2811292d06b9SMax Reitz * This is a workaround for a bug in the Linux XFS driver, 2812292d06b9SMax Reitz * where writes submitted through the AIO interface will be 2813292d06b9SMax Reitz * discarded if they happen beyond a concurrently running 2814292d06b9SMax Reitz * fallocate() that increases the file length (i.e., both the 2815292d06b9SMax Reitz * write and the fallocate() happen beyond the EOF). 2816292d06b9SMax Reitz * 2817292d06b9SMax Reitz * To work around it, we extend the tracked request for this 2818292d06b9SMax Reitz * zero write until INT64_MAX (effectively infinity), and mark 2819292d06b9SMax Reitz * it as serializing. 2820292d06b9SMax Reitz * 2821292d06b9SMax Reitz * We have to enable this workaround for all filesystems and 2822292d06b9SMax Reitz * AIO modes (not just XFS with aio=native), because for 2823292d06b9SMax Reitz * remote filesystems we do not know the host configuration. 2824292d06b9SMax Reitz */ 2825292d06b9SMax Reitz 2826292d06b9SMax Reitz req = bdrv_co_get_self_request(bs); 2827292d06b9SMax Reitz assert(req); 2828292d06b9SMax Reitz assert(req->type == BDRV_TRACKED_WRITE); 2829292d06b9SMax Reitz assert(req->offset <= offset); 2830292d06b9SMax Reitz assert(req->offset + req->bytes >= offset + bytes); 2831292d06b9SMax Reitz 2832292d06b9SMax Reitz end = INT64_MAX & -(uint64_t)bs->bl.request_alignment; 2833292d06b9SMax Reitz req->bytes = end - req->offset; 2834292d06b9SMax Reitz req->overlap_bytes = req->bytes; 2835292d06b9SMax Reitz 2836292d06b9SMax Reitz bdrv_mark_request_serialising(req, bs->bl.request_alignment); 2837292d06b9SMax Reitz } 2838292d06b9SMax Reitz #endif 2839292d06b9SMax Reitz 28407154d8aeSKevin Wolf acb = (RawPosixAIOData) { 28417154d8aeSKevin Wolf .bs = bs, 28427154d8aeSKevin Wolf .aio_fildes = s->fd, 28437154d8aeSKevin Wolf .aio_type = QEMU_AIO_WRITE_ZEROES, 28447154d8aeSKevin Wolf .aio_offset = offset, 28457154d8aeSKevin Wolf .aio_nbytes = bytes, 28467154d8aeSKevin Wolf }; 28477154d8aeSKevin Wolf 28487154d8aeSKevin Wolf if (blkdev) { 28497154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 28507154d8aeSKevin Wolf } 2851738301e1SKevin Wolf if (flags & BDRV_REQ_NO_FALLBACK) { 2852738301e1SKevin Wolf acb.aio_type |= QEMU_AIO_NO_FALLBACK; 2853738301e1SKevin Wolf } 28547154d8aeSKevin Wolf 28557154d8aeSKevin Wolf if (flags & BDRV_REQ_MAY_UNMAP) { 28567154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_DISCARD; 28577154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes_unmap; 28587154d8aeSKevin Wolf } else { 28597154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes; 28607154d8aeSKevin Wolf } 28617154d8aeSKevin Wolf 28627154d8aeSKevin Wolf return raw_thread_pool_submit(bs, handler, &acb); 28637154d8aeSKevin Wolf } 28647154d8aeSKevin Wolf 2865c1bb86cdSEric Blake static int coroutine_fn raw_co_pwrite_zeroes( 2866c1bb86cdSEric Blake BlockDriverState *bs, int64_t offset, 2867f5a5ca79SManos Pitsidianakis int bytes, BdrvRequestFlags flags) 2868c1bb86cdSEric Blake { 28697154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); 2870c1bb86cdSEric Blake } 2871c1bb86cdSEric Blake 2872c1bb86cdSEric Blake static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2873c1bb86cdSEric Blake { 2874c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2875c1bb86cdSEric Blake 2876c1bb86cdSEric Blake bdi->unallocated_blocks_are_zero = s->discard_zeroes; 2877c1bb86cdSEric Blake return 0; 2878c1bb86cdSEric Blake } 2879c1bb86cdSEric Blake 2880d9245599SAnton Nefedov static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs) 2881d9245599SAnton Nefedov { 2882d9245599SAnton Nefedov BDRVRawState *s = bs->opaque; 2883d9245599SAnton Nefedov return (BlockStatsSpecificFile) { 2884d9245599SAnton Nefedov .discard_nb_ok = s->stats.discard_nb_ok, 2885d9245599SAnton Nefedov .discard_nb_failed = s->stats.discard_nb_failed, 2886d9245599SAnton Nefedov .discard_bytes_ok = s->stats.discard_bytes_ok, 2887d9245599SAnton Nefedov }; 2888d9245599SAnton Nefedov } 2889d9245599SAnton Nefedov 2890d9245599SAnton Nefedov static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs) 2891d9245599SAnton Nefedov { 2892d9245599SAnton Nefedov BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 2893d9245599SAnton Nefedov 2894d9245599SAnton Nefedov stats->driver = BLOCKDEV_DRIVER_FILE; 2895d9245599SAnton Nefedov stats->u.file = get_blockstats_specific_file(bs); 2896d9245599SAnton Nefedov 2897d9245599SAnton Nefedov return stats; 2898d9245599SAnton Nefedov } 2899d9245599SAnton Nefedov 2900d9245599SAnton Nefedov static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs) 2901d9245599SAnton Nefedov { 2902d9245599SAnton Nefedov BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 2903d9245599SAnton Nefedov 2904d9245599SAnton Nefedov stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE; 2905d9245599SAnton Nefedov stats->u.host_device = get_blockstats_specific_file(bs); 2906d9245599SAnton Nefedov 2907d9245599SAnton Nefedov return stats; 2908d9245599SAnton Nefedov } 2909d9245599SAnton Nefedov 2910c1bb86cdSEric Blake static QemuOptsList raw_create_opts = { 2911c1bb86cdSEric Blake .name = "raw-create-opts", 2912c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), 2913c1bb86cdSEric Blake .desc = { 2914c1bb86cdSEric Blake { 2915c1bb86cdSEric Blake .name = BLOCK_OPT_SIZE, 2916c1bb86cdSEric Blake .type = QEMU_OPT_SIZE, 2917c1bb86cdSEric Blake .help = "Virtual disk size" 2918c1bb86cdSEric Blake }, 2919c1bb86cdSEric Blake { 2920c1bb86cdSEric Blake .name = BLOCK_OPT_NOCOW, 2921c1bb86cdSEric Blake .type = QEMU_OPT_BOOL, 2922c1bb86cdSEric Blake .help = "Turn off copy-on-write (valid only on btrfs)" 2923c1bb86cdSEric Blake }, 2924c1bb86cdSEric Blake { 2925c1bb86cdSEric Blake .name = BLOCK_OPT_PREALLOC, 2926c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 2927abea0053SStefano Garzarella .help = "Preallocation mode (allowed values: off" 2928abea0053SStefano Garzarella #ifdef CONFIG_POSIX_FALLOCATE 2929abea0053SStefano Garzarella ", falloc" 2930abea0053SStefano Garzarella #endif 2931abea0053SStefano Garzarella ", full)" 2932c1bb86cdSEric Blake }, 2933c1bb86cdSEric Blake { /* end of list */ } 2934c1bb86cdSEric Blake } 2935c1bb86cdSEric Blake }; 2936c1bb86cdSEric Blake 2937244a5668SFam Zheng static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, 2938244a5668SFam Zheng Error **errp) 2939244a5668SFam Zheng { 29406ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 29416ceabe6fSKevin Wolf BDRVRawReopenState *rs = NULL; 29426ceabe6fSKevin Wolf int open_flags; 29436ceabe6fSKevin Wolf int ret; 29446ceabe6fSKevin Wolf 29456ceabe6fSKevin Wolf if (s->perm_change_fd) { 29466ceabe6fSKevin Wolf /* 29476ceabe6fSKevin Wolf * In the context of reopen, this function may be called several times 29486ceabe6fSKevin Wolf * (directly and recursively while change permissions of the parent). 29496ceabe6fSKevin Wolf * This is even true for children that don't inherit from the original 29506ceabe6fSKevin Wolf * reopen node, so s->reopen_state is not set. 29516ceabe6fSKevin Wolf * 29526ceabe6fSKevin Wolf * Ignore all but the first call. 29536ceabe6fSKevin Wolf */ 29546ceabe6fSKevin Wolf return 0; 29556ceabe6fSKevin Wolf } 29566ceabe6fSKevin Wolf 29576ceabe6fSKevin Wolf if (s->reopen_state) { 29586ceabe6fSKevin Wolf /* We already have a new file descriptor to set permissions for */ 29596ceabe6fSKevin Wolf assert(s->reopen_state->perm == perm); 29606ceabe6fSKevin Wolf assert(s->reopen_state->shared_perm == shared); 29616ceabe6fSKevin Wolf rs = s->reopen_state->opaque; 29626ceabe6fSKevin Wolf s->perm_change_fd = rs->fd; 2963094e3639SMax Reitz s->perm_change_flags = rs->open_flags; 29646ceabe6fSKevin Wolf } else { 29656ceabe6fSKevin Wolf /* We may need a new fd if auto-read-only switches the mode */ 296623dece19SKevin Wolf ret = raw_reconfigure_getfd(bs, bs->open_flags, &open_flags, perm, 29676ceabe6fSKevin Wolf false, errp); 29686ceabe6fSKevin Wolf if (ret < 0) { 29696ceabe6fSKevin Wolf return ret; 29706ceabe6fSKevin Wolf } else if (ret != s->fd) { 29716ceabe6fSKevin Wolf s->perm_change_fd = ret; 2972094e3639SMax Reitz s->perm_change_flags = open_flags; 29736ceabe6fSKevin Wolf } 29746ceabe6fSKevin Wolf } 29756ceabe6fSKevin Wolf 29766ceabe6fSKevin Wolf /* Prepare permissions on old fd to avoid conflicts between old and new, 29776ceabe6fSKevin Wolf * but keep everything locked that new will need. */ 29786ceabe6fSKevin Wolf ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); 29796ceabe6fSKevin Wolf if (ret < 0) { 29806ceabe6fSKevin Wolf goto fail; 29816ceabe6fSKevin Wolf } 29826ceabe6fSKevin Wolf 29836ceabe6fSKevin Wolf /* Copy locks to the new fd */ 29846ceabe6fSKevin Wolf if (s->perm_change_fd) { 29856ceabe6fSKevin Wolf ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, 29866ceabe6fSKevin Wolf false, errp); 29876ceabe6fSKevin Wolf if (ret < 0) { 29886ceabe6fSKevin Wolf raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 29896ceabe6fSKevin Wolf goto fail; 29906ceabe6fSKevin Wolf } 29916ceabe6fSKevin Wolf } 29926ceabe6fSKevin Wolf return 0; 29936ceabe6fSKevin Wolf 29946ceabe6fSKevin Wolf fail: 29956ceabe6fSKevin Wolf if (s->perm_change_fd && !s->reopen_state) { 29966ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 29976ceabe6fSKevin Wolf } 29986ceabe6fSKevin Wolf s->perm_change_fd = 0; 29996ceabe6fSKevin Wolf return ret; 3000244a5668SFam Zheng } 3001244a5668SFam Zheng 3002244a5668SFam Zheng static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) 3003244a5668SFam Zheng { 3004244a5668SFam Zheng BDRVRawState *s = bs->opaque; 30056ceabe6fSKevin Wolf 30066ceabe6fSKevin Wolf /* For reopen, we have already switched to the new fd (.bdrv_set_perm is 30076ceabe6fSKevin Wolf * called after .bdrv_reopen_commit) */ 30086ceabe6fSKevin Wolf if (s->perm_change_fd && s->fd != s->perm_change_fd) { 30096ceabe6fSKevin Wolf qemu_close(s->fd); 30106ceabe6fSKevin Wolf s->fd = s->perm_change_fd; 3011094e3639SMax Reitz s->open_flags = s->perm_change_flags; 30126ceabe6fSKevin Wolf } 30136ceabe6fSKevin Wolf s->perm_change_fd = 0; 30146ceabe6fSKevin Wolf 3015244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); 3016244a5668SFam Zheng s->perm = perm; 3017244a5668SFam Zheng s->shared_perm = shared; 3018244a5668SFam Zheng } 3019244a5668SFam Zheng 3020244a5668SFam Zheng static void raw_abort_perm_update(BlockDriverState *bs) 3021244a5668SFam Zheng { 30226ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 30236ceabe6fSKevin Wolf 30246ceabe6fSKevin Wolf /* For reopen, .bdrv_reopen_abort is called afterwards and will close 30256ceabe6fSKevin Wolf * the file descriptor. */ 30266ceabe6fSKevin Wolf if (s->perm_change_fd && !s->reopen_state) { 30276ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 30286ceabe6fSKevin Wolf } 30296ceabe6fSKevin Wolf s->perm_change_fd = 0; 30306ceabe6fSKevin Wolf 3031244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 3032244a5668SFam Zheng } 3033244a5668SFam Zheng 303467b51fb9SVladimir Sementsov-Ogievskiy static int coroutine_fn raw_co_copy_range_from( 303567b51fb9SVladimir Sementsov-Ogievskiy BlockDriverState *bs, BdrvChild *src, uint64_t src_offset, 303667b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, uint64_t dst_offset, uint64_t bytes, 303767b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) 30381efad060SFam Zheng { 303967b51fb9SVladimir Sementsov-Ogievskiy return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, 304067b51fb9SVladimir Sementsov-Ogievskiy read_flags, write_flags); 30411efad060SFam Zheng } 30421efad060SFam Zheng 30431efad060SFam Zheng static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, 304467b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *src, 304567b51fb9SVladimir Sementsov-Ogievskiy uint64_t src_offset, 304667b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, 304767b51fb9SVladimir Sementsov-Ogievskiy uint64_t dst_offset, 304867b51fb9SVladimir Sementsov-Ogievskiy uint64_t bytes, 304967b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, 305067b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags write_flags) 30511efad060SFam Zheng { 305258a209c4SKevin Wolf RawPosixAIOData acb; 30531efad060SFam Zheng BDRVRawState *s = bs->opaque; 30541efad060SFam Zheng BDRVRawState *src_s; 30551efad060SFam Zheng 30561efad060SFam Zheng assert(dst->bs == bs); 30571efad060SFam Zheng if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { 30581efad060SFam Zheng return -ENOTSUP; 30591efad060SFam Zheng } 30601efad060SFam Zheng 30611efad060SFam Zheng src_s = src->bs->opaque; 30629f850f67SFam Zheng if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { 30631efad060SFam Zheng return -EIO; 30641efad060SFam Zheng } 306558a209c4SKevin Wolf 306658a209c4SKevin Wolf acb = (RawPosixAIOData) { 306758a209c4SKevin Wolf .bs = bs, 306858a209c4SKevin Wolf .aio_type = QEMU_AIO_COPY_RANGE, 306958a209c4SKevin Wolf .aio_fildes = src_s->fd, 307058a209c4SKevin Wolf .aio_offset = src_offset, 307158a209c4SKevin Wolf .aio_nbytes = bytes, 307258a209c4SKevin Wolf .copy_range = { 307358a209c4SKevin Wolf .aio_fd2 = s->fd, 307458a209c4SKevin Wolf .aio_offset2 = dst_offset, 307558a209c4SKevin Wolf }, 307658a209c4SKevin Wolf }; 307758a209c4SKevin Wolf 307858a209c4SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); 30791efad060SFam Zheng } 30801efad060SFam Zheng 3081c1bb86cdSEric Blake BlockDriver bdrv_file = { 3082c1bb86cdSEric Blake .format_name = "file", 3083c1bb86cdSEric Blake .protocol_name = "file", 3084c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3085c1bb86cdSEric Blake .bdrv_needs_filename = true, 3086c1bb86cdSEric Blake .bdrv_probe = NULL, /* no probe for protocols */ 3087c1bb86cdSEric Blake .bdrv_parse_filename = raw_parse_filename, 3088c1bb86cdSEric Blake .bdrv_file_open = raw_open, 3089c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3090c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3091c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3092c1bb86cdSEric Blake .bdrv_close = raw_close, 3093927f11e1SKevin Wolf .bdrv_co_create = raw_co_create, 3094efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = raw_co_create_opts, 3095c1bb86cdSEric Blake .bdrv_has_zero_init = bdrv_has_zero_init_1, 30961dcaf527SMax Reitz .bdrv_has_zero_init_truncate = bdrv_has_zero_init_1, 3097a290f085SEric Blake .bdrv_co_block_status = raw_co_block_status, 3098dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3099c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, 3100*9bffae14SDaniel Henrique Barboza .bdrv_co_delete_file = raw_co_delete_file, 3101c1bb86cdSEric Blake 3102c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3103c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 310433d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 310533d70fb6SKevin Wolf .bdrv_co_pdiscard = raw_co_pdiscard, 31061efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 31071efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 3108c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3109c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3110c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3111ed6e2161SNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3112c1bb86cdSEric Blake 3113061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3114c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3115c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 3116c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3117c1bb86cdSEric Blake = raw_get_allocated_file_size, 3118d9245599SAnton Nefedov .bdrv_get_specific_stats = raw_get_specific_stats, 3119244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 3120244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 3121244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 3122c1bb86cdSEric Blake .create_opts = &raw_create_opts, 31238a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3124c1bb86cdSEric Blake }; 3125c1bb86cdSEric Blake 3126c1bb86cdSEric Blake /***********************************************/ 3127c1bb86cdSEric Blake /* host device */ 3128c1bb86cdSEric Blake 3129c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3130c1bb86cdSEric Blake static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3131c1bb86cdSEric Blake CFIndex maxPathSize, int flags); 3132c1bb86cdSEric Blake static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) 3133c1bb86cdSEric Blake { 3134c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 3135c1bb86cdSEric Blake mach_port_t masterPort; 3136c1bb86cdSEric Blake CFMutableDictionaryRef classesToMatch; 3137c1bb86cdSEric Blake const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; 3138c1bb86cdSEric Blake char *mediaType = NULL; 3139c1bb86cdSEric Blake 3140c1bb86cdSEric Blake kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); 3141c1bb86cdSEric Blake if ( KERN_SUCCESS != kernResult ) { 3142c1bb86cdSEric Blake printf( "IOMasterPort returned %d\n", kernResult ); 3143c1bb86cdSEric Blake } 3144c1bb86cdSEric Blake 3145c1bb86cdSEric Blake int index; 3146c1bb86cdSEric Blake for (index = 0; index < ARRAY_SIZE(matching_array); index++) { 3147c1bb86cdSEric Blake classesToMatch = IOServiceMatching(matching_array[index]); 3148c1bb86cdSEric Blake if (classesToMatch == NULL) { 3149c1bb86cdSEric Blake error_report("IOServiceMatching returned NULL for %s", 3150c1bb86cdSEric Blake matching_array[index]); 3151c1bb86cdSEric Blake continue; 3152c1bb86cdSEric Blake } 3153c1bb86cdSEric Blake CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), 3154c1bb86cdSEric Blake kCFBooleanTrue); 3155c1bb86cdSEric Blake kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, 3156c1bb86cdSEric Blake mediaIterator); 3157c1bb86cdSEric Blake if (kernResult != KERN_SUCCESS) { 3158c1bb86cdSEric Blake error_report("Note: IOServiceGetMatchingServices returned %d", 3159c1bb86cdSEric Blake kernResult); 3160c1bb86cdSEric Blake continue; 3161c1bb86cdSEric Blake } 3162c1bb86cdSEric Blake 3163c1bb86cdSEric Blake /* If a match was found, leave the loop */ 3164c1bb86cdSEric Blake if (*mediaIterator != 0) { 31654f7d28d7SLaurent Vivier trace_file_FindEjectableOpticalMedia(matching_array[index]); 3166c1bb86cdSEric Blake mediaType = g_strdup(matching_array[index]); 3167c1bb86cdSEric Blake break; 3168c1bb86cdSEric Blake } 3169c1bb86cdSEric Blake } 3170c1bb86cdSEric Blake return mediaType; 3171c1bb86cdSEric Blake } 3172c1bb86cdSEric Blake 3173c1bb86cdSEric Blake kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3174c1bb86cdSEric Blake CFIndex maxPathSize, int flags) 3175c1bb86cdSEric Blake { 3176c1bb86cdSEric Blake io_object_t nextMedia; 3177c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 3178c1bb86cdSEric Blake *bsdPath = '\0'; 3179c1bb86cdSEric Blake nextMedia = IOIteratorNext( mediaIterator ); 3180c1bb86cdSEric Blake if ( nextMedia ) 3181c1bb86cdSEric Blake { 3182c1bb86cdSEric Blake CFTypeRef bsdPathAsCFString; 3183c1bb86cdSEric Blake bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); 3184c1bb86cdSEric Blake if ( bsdPathAsCFString ) { 3185c1bb86cdSEric Blake size_t devPathLength; 3186c1bb86cdSEric Blake strcpy( bsdPath, _PATH_DEV ); 3187c1bb86cdSEric Blake if (flags & BDRV_O_NOCACHE) { 3188c1bb86cdSEric Blake strcat(bsdPath, "r"); 3189c1bb86cdSEric Blake } 3190c1bb86cdSEric Blake devPathLength = strlen( bsdPath ); 3191c1bb86cdSEric Blake if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { 3192c1bb86cdSEric Blake kernResult = KERN_SUCCESS; 3193c1bb86cdSEric Blake } 3194c1bb86cdSEric Blake CFRelease( bsdPathAsCFString ); 3195c1bb86cdSEric Blake } 3196c1bb86cdSEric Blake IOObjectRelease( nextMedia ); 3197c1bb86cdSEric Blake } 3198c1bb86cdSEric Blake 3199c1bb86cdSEric Blake return kernResult; 3200c1bb86cdSEric Blake } 3201c1bb86cdSEric Blake 3202c1bb86cdSEric Blake /* Sets up a real cdrom for use in QEMU */ 3203c1bb86cdSEric Blake static bool setup_cdrom(char *bsd_path, Error **errp) 3204c1bb86cdSEric Blake { 3205c1bb86cdSEric Blake int index, num_of_test_partitions = 2, fd; 3206c1bb86cdSEric Blake char test_partition[MAXPATHLEN]; 3207c1bb86cdSEric Blake bool partition_found = false; 3208c1bb86cdSEric Blake 3209c1bb86cdSEric Blake /* look for a working partition */ 3210c1bb86cdSEric Blake for (index = 0; index < num_of_test_partitions; index++) { 3211c1bb86cdSEric Blake snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, 3212c1bb86cdSEric Blake index); 3213c1bb86cdSEric Blake fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE); 3214c1bb86cdSEric Blake if (fd >= 0) { 3215c1bb86cdSEric Blake partition_found = true; 3216c1bb86cdSEric Blake qemu_close(fd); 3217c1bb86cdSEric Blake break; 3218c1bb86cdSEric Blake } 3219c1bb86cdSEric Blake } 3220c1bb86cdSEric Blake 3221c1bb86cdSEric Blake /* if a working partition on the device was not found */ 3222c1bb86cdSEric Blake if (partition_found == false) { 3223c1bb86cdSEric Blake error_setg(errp, "Failed to find a working partition on disc"); 3224c1bb86cdSEric Blake } else { 32254f7d28d7SLaurent Vivier trace_file_setup_cdrom(test_partition); 3226c1bb86cdSEric Blake pstrcpy(bsd_path, MAXPATHLEN, test_partition); 3227c1bb86cdSEric Blake } 3228c1bb86cdSEric Blake return partition_found; 3229c1bb86cdSEric Blake } 3230c1bb86cdSEric Blake 3231c1bb86cdSEric Blake /* Prints directions on mounting and unmounting a device */ 3232c1bb86cdSEric Blake static void print_unmounting_directions(const char *file_name) 3233c1bb86cdSEric Blake { 3234c1bb86cdSEric Blake error_report("If device %s is mounted on the desktop, unmount" 3235c1bb86cdSEric Blake " it first before using it in QEMU", file_name); 3236c1bb86cdSEric Blake error_report("Command to unmount device: diskutil unmountDisk %s", 3237c1bb86cdSEric Blake file_name); 3238c1bb86cdSEric Blake error_report("Command to mount device: diskutil mountDisk %s", file_name); 3239c1bb86cdSEric Blake } 3240c1bb86cdSEric Blake 3241c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3242c1bb86cdSEric Blake 3243c1bb86cdSEric Blake static int hdev_probe_device(const char *filename) 3244c1bb86cdSEric Blake { 3245c1bb86cdSEric Blake struct stat st; 3246c1bb86cdSEric Blake 3247c1bb86cdSEric Blake /* allow a dedicated CD-ROM driver to match with a higher priority */ 3248c1bb86cdSEric Blake if (strstart(filename, "/dev/cdrom", NULL)) 3249c1bb86cdSEric Blake return 50; 3250c1bb86cdSEric Blake 3251c1bb86cdSEric Blake if (stat(filename, &st) >= 0 && 3252c1bb86cdSEric Blake (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 3253c1bb86cdSEric Blake return 100; 3254c1bb86cdSEric Blake } 3255c1bb86cdSEric Blake 3256c1bb86cdSEric Blake return 0; 3257c1bb86cdSEric Blake } 3258c1bb86cdSEric Blake 3259c1bb86cdSEric Blake static int check_hdev_writable(BDRVRawState *s) 3260c1bb86cdSEric Blake { 3261c1bb86cdSEric Blake #if defined(BLKROGET) 3262c1bb86cdSEric Blake /* Linux block devices can be configured "read-only" using blockdev(8). 3263c1bb86cdSEric Blake * This is independent of device node permissions and therefore open(2) 3264c1bb86cdSEric Blake * with O_RDWR succeeds. Actual writes fail with EPERM. 3265c1bb86cdSEric Blake * 3266c1bb86cdSEric Blake * bdrv_open() is supposed to fail if the disk is read-only. Explicitly 3267c1bb86cdSEric Blake * check for read-only block devices so that Linux block devices behave 3268c1bb86cdSEric Blake * properly. 3269c1bb86cdSEric Blake */ 3270c1bb86cdSEric Blake struct stat st; 3271c1bb86cdSEric Blake int readonly = 0; 3272c1bb86cdSEric Blake 3273c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 3274c1bb86cdSEric Blake return -errno; 3275c1bb86cdSEric Blake } 3276c1bb86cdSEric Blake 3277c1bb86cdSEric Blake if (!S_ISBLK(st.st_mode)) { 3278c1bb86cdSEric Blake return 0; 3279c1bb86cdSEric Blake } 3280c1bb86cdSEric Blake 3281c1bb86cdSEric Blake if (ioctl(s->fd, BLKROGET, &readonly) < 0) { 3282c1bb86cdSEric Blake return -errno; 3283c1bb86cdSEric Blake } 3284c1bb86cdSEric Blake 3285c1bb86cdSEric Blake if (readonly) { 3286c1bb86cdSEric Blake return -EACCES; 3287c1bb86cdSEric Blake } 3288c1bb86cdSEric Blake #endif /* defined(BLKROGET) */ 3289c1bb86cdSEric Blake return 0; 3290c1bb86cdSEric Blake } 3291c1bb86cdSEric Blake 3292c1bb86cdSEric Blake static void hdev_parse_filename(const char *filename, QDict *options, 3293c1bb86cdSEric Blake Error **errp) 3294c1bb86cdSEric Blake { 329503c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_device:", options); 3296c1bb86cdSEric Blake } 3297c1bb86cdSEric Blake 3298c1bb86cdSEric Blake static bool hdev_is_sg(BlockDriverState *bs) 3299c1bb86cdSEric Blake { 3300c1bb86cdSEric Blake 3301c1bb86cdSEric Blake #if defined(__linux__) 3302c1bb86cdSEric Blake 3303c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3304c1bb86cdSEric Blake struct stat st; 3305c1bb86cdSEric Blake struct sg_scsi_id scsiid; 3306c1bb86cdSEric Blake int sg_version; 3307c1bb86cdSEric Blake int ret; 3308c1bb86cdSEric Blake 3309c1bb86cdSEric Blake if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { 3310c1bb86cdSEric Blake return false; 3311c1bb86cdSEric Blake } 3312c1bb86cdSEric Blake 3313c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); 3314c1bb86cdSEric Blake if (ret < 0) { 3315c1bb86cdSEric Blake return false; 3316c1bb86cdSEric Blake } 3317c1bb86cdSEric Blake 3318c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); 3319c1bb86cdSEric Blake if (ret >= 0) { 33204f7d28d7SLaurent Vivier trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); 3321c1bb86cdSEric Blake return true; 3322c1bb86cdSEric Blake } 3323c1bb86cdSEric Blake 3324c1bb86cdSEric Blake #endif 3325c1bb86cdSEric Blake 3326c1bb86cdSEric Blake return false; 3327c1bb86cdSEric Blake } 3328c1bb86cdSEric Blake 3329c1bb86cdSEric Blake static int hdev_open(BlockDriverState *bs, QDict *options, int flags, 3330c1bb86cdSEric Blake Error **errp) 3331c1bb86cdSEric Blake { 3332c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3333c1bb86cdSEric Blake Error *local_err = NULL; 3334c1bb86cdSEric Blake int ret; 3335c1bb86cdSEric Blake 3336c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3337129c7d1cSMarkus Armbruster /* 3338129c7d1cSMarkus Armbruster * Caution: while qdict_get_str() is fine, getting non-string types 3339129c7d1cSMarkus Armbruster * would require more care. When @options come from -blockdev or 3340129c7d1cSMarkus Armbruster * blockdev_add, its members are typed according to the QAPI 3341129c7d1cSMarkus Armbruster * schema, but when they come from -drive, they're all QString. 3342129c7d1cSMarkus Armbruster */ 3343c1bb86cdSEric Blake const char *filename = qdict_get_str(options, "filename"); 3344c1bb86cdSEric Blake char bsd_path[MAXPATHLEN] = ""; 3345c1bb86cdSEric Blake bool error_occurred = false; 3346c1bb86cdSEric Blake 3347c1bb86cdSEric Blake /* If using a real cdrom */ 3348c1bb86cdSEric Blake if (strcmp(filename, "/dev/cdrom") == 0) { 3349c1bb86cdSEric Blake char *mediaType = NULL; 3350c1bb86cdSEric Blake kern_return_t ret_val; 3351c1bb86cdSEric Blake io_iterator_t mediaIterator = 0; 3352c1bb86cdSEric Blake 3353c1bb86cdSEric Blake mediaType = FindEjectableOpticalMedia(&mediaIterator); 3354c1bb86cdSEric Blake if (mediaType == NULL) { 3355c1bb86cdSEric Blake error_setg(errp, "Please make sure your CD/DVD is in the optical" 3356c1bb86cdSEric Blake " drive"); 3357c1bb86cdSEric Blake error_occurred = true; 3358c1bb86cdSEric Blake goto hdev_open_Mac_error; 3359c1bb86cdSEric Blake } 3360c1bb86cdSEric Blake 3361c1bb86cdSEric Blake ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); 3362c1bb86cdSEric Blake if (ret_val != KERN_SUCCESS) { 3363c1bb86cdSEric Blake error_setg(errp, "Could not get BSD path for optical drive"); 3364c1bb86cdSEric Blake error_occurred = true; 3365c1bb86cdSEric Blake goto hdev_open_Mac_error; 3366c1bb86cdSEric Blake } 3367c1bb86cdSEric Blake 3368c1bb86cdSEric Blake /* If a real optical drive was not found */ 3369c1bb86cdSEric Blake if (bsd_path[0] == '\0') { 3370c1bb86cdSEric Blake error_setg(errp, "Failed to obtain bsd path for optical drive"); 3371c1bb86cdSEric Blake error_occurred = true; 3372c1bb86cdSEric Blake goto hdev_open_Mac_error; 3373c1bb86cdSEric Blake } 3374c1bb86cdSEric Blake 3375c1bb86cdSEric Blake /* If using a cdrom disc and finding a partition on the disc failed */ 3376c1bb86cdSEric Blake if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && 3377c1bb86cdSEric Blake setup_cdrom(bsd_path, errp) == false) { 3378c1bb86cdSEric Blake print_unmounting_directions(bsd_path); 3379c1bb86cdSEric Blake error_occurred = true; 3380c1bb86cdSEric Blake goto hdev_open_Mac_error; 3381c1bb86cdSEric Blake } 3382c1bb86cdSEric Blake 338346f5ac20SEric Blake qdict_put_str(options, "filename", bsd_path); 3384c1bb86cdSEric Blake 3385c1bb86cdSEric Blake hdev_open_Mac_error: 3386c1bb86cdSEric Blake g_free(mediaType); 3387c1bb86cdSEric Blake if (mediaIterator) { 3388c1bb86cdSEric Blake IOObjectRelease(mediaIterator); 3389c1bb86cdSEric Blake } 3390c1bb86cdSEric Blake if (error_occurred) { 3391c1bb86cdSEric Blake return -ENOENT; 3392c1bb86cdSEric Blake } 3393c1bb86cdSEric Blake } 3394c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3395c1bb86cdSEric Blake 3396c1bb86cdSEric Blake s->type = FTYPE_FILE; 3397c1bb86cdSEric Blake 3398230ff739SJohn Snow ret = raw_open_common(bs, options, flags, 0, true, &local_err); 3399c1bb86cdSEric Blake if (ret < 0) { 3400c1bb86cdSEric Blake error_propagate(errp, local_err); 3401c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3402c1bb86cdSEric Blake if (*bsd_path) { 3403c1bb86cdSEric Blake filename = bsd_path; 3404c1bb86cdSEric Blake } 3405c1bb86cdSEric Blake /* if a physical device experienced an error while being opened */ 3406c1bb86cdSEric Blake if (strncmp(filename, "/dev/", 5) == 0) { 3407c1bb86cdSEric Blake print_unmounting_directions(filename); 3408c1bb86cdSEric Blake } 3409c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3410c1bb86cdSEric Blake return ret; 3411c1bb86cdSEric Blake } 3412c1bb86cdSEric Blake 3413c1bb86cdSEric Blake /* Since this does ioctl the device must be already opened */ 3414c1bb86cdSEric Blake bs->sg = hdev_is_sg(bs); 3415c1bb86cdSEric Blake 3416c1bb86cdSEric Blake if (flags & BDRV_O_RDWR) { 3417c1bb86cdSEric Blake ret = check_hdev_writable(s); 3418c1bb86cdSEric Blake if (ret < 0) { 3419c1bb86cdSEric Blake raw_close(bs); 3420c1bb86cdSEric Blake error_setg_errno(errp, -ret, "The device is not writable"); 3421c1bb86cdSEric Blake return ret; 3422c1bb86cdSEric Blake } 3423c1bb86cdSEric Blake } 3424c1bb86cdSEric Blake 3425c1bb86cdSEric Blake return ret; 3426c1bb86cdSEric Blake } 3427c1bb86cdSEric Blake 3428c1bb86cdSEric Blake #if defined(__linux__) 34292f3a7ab3SKevin Wolf static int coroutine_fn 34302f3a7ab3SKevin Wolf hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) 3431c1bb86cdSEric Blake { 3432c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 343303425671SKevin Wolf RawPosixAIOData acb; 34342f3a7ab3SKevin Wolf int ret; 3435c1bb86cdSEric Blake 34362f3a7ab3SKevin Wolf ret = fd_open(bs); 34372f3a7ab3SKevin Wolf if (ret < 0) { 34382f3a7ab3SKevin Wolf return ret; 34392f3a7ab3SKevin Wolf } 3440c1bb86cdSEric Blake 34417c9e5276SPaolo Bonzini if (req == SG_IO && s->pr_mgr) { 34427c9e5276SPaolo Bonzini struct sg_io_hdr *io_hdr = buf; 34437c9e5276SPaolo Bonzini if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || 34447c9e5276SPaolo Bonzini io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { 34457c9e5276SPaolo Bonzini return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), 34462f3a7ab3SKevin Wolf s->fd, io_hdr); 34477c9e5276SPaolo Bonzini } 34487c9e5276SPaolo Bonzini } 34497c9e5276SPaolo Bonzini 345003425671SKevin Wolf acb = (RawPosixAIOData) { 345103425671SKevin Wolf .bs = bs, 345203425671SKevin Wolf .aio_type = QEMU_AIO_IOCTL, 345303425671SKevin Wolf .aio_fildes = s->fd, 345403425671SKevin Wolf .aio_offset = 0, 345503425671SKevin Wolf .ioctl = { 345603425671SKevin Wolf .buf = buf, 345703425671SKevin Wolf .cmd = req, 345803425671SKevin Wolf }, 345903425671SKevin Wolf }; 346003425671SKevin Wolf 346103425671SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); 3462c1bb86cdSEric Blake } 3463c1bb86cdSEric Blake #endif /* linux */ 3464c1bb86cdSEric Blake 3465c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs) 3466c1bb86cdSEric Blake { 3467c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3468c1bb86cdSEric Blake 3469c1bb86cdSEric Blake /* this is just to ensure s->fd is sane (its called by io ops) */ 3470c1bb86cdSEric Blake if (s->fd >= 0) 3471c1bb86cdSEric Blake return 0; 3472c1bb86cdSEric Blake return -EIO; 3473c1bb86cdSEric Blake } 3474c1bb86cdSEric Blake 347533d70fb6SKevin Wolf static coroutine_fn int 347633d70fb6SKevin Wolf hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 3477c1bb86cdSEric Blake { 34781c450366SAnton Nefedov BDRVRawState *s = bs->opaque; 347933d70fb6SKevin Wolf int ret; 3480c1bb86cdSEric Blake 348133d70fb6SKevin Wolf ret = fd_open(bs); 348233d70fb6SKevin Wolf if (ret < 0) { 34831c450366SAnton Nefedov raw_account_discard(s, bytes, ret); 348433d70fb6SKevin Wolf return ret; 3485c1bb86cdSEric Blake } 348646ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, true); 3487c1bb86cdSEric Blake } 3488c1bb86cdSEric Blake 3489c1bb86cdSEric Blake static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, 3490f5a5ca79SManos Pitsidianakis int64_t offset, int bytes, BdrvRequestFlags flags) 3491c1bb86cdSEric Blake { 3492c1bb86cdSEric Blake int rc; 3493c1bb86cdSEric Blake 3494c1bb86cdSEric Blake rc = fd_open(bs); 3495c1bb86cdSEric Blake if (rc < 0) { 3496c1bb86cdSEric Blake return rc; 3497c1bb86cdSEric Blake } 349834fa110eSKevin Wolf 34997154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); 3500c1bb86cdSEric Blake } 3501c1bb86cdSEric Blake 3502c1bb86cdSEric Blake static BlockDriver bdrv_host_device = { 3503c1bb86cdSEric Blake .format_name = "host_device", 3504c1bb86cdSEric Blake .protocol_name = "host_device", 3505c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3506c1bb86cdSEric Blake .bdrv_needs_filename = true, 3507c1bb86cdSEric Blake .bdrv_probe_device = hdev_probe_device, 3508c1bb86cdSEric Blake .bdrv_parse_filename = hdev_parse_filename, 3509c1bb86cdSEric Blake .bdrv_file_open = hdev_open, 3510c1bb86cdSEric Blake .bdrv_close = raw_close, 3511c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3512c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3513c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 35148a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3515dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3516c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, 3517c1bb86cdSEric Blake 3518c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3519c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 352033d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 352133d70fb6SKevin Wolf .bdrv_co_pdiscard = hdev_co_pdiscard, 35221efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 35231efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 3524c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3525c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3526c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3527042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3528c1bb86cdSEric Blake 3529061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3530c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3531c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 3532c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3533c1bb86cdSEric Blake = raw_get_allocated_file_size, 3534d9245599SAnton Nefedov .bdrv_get_specific_stats = hdev_get_specific_stats, 3535244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 3536244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 3537244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 3538c1bb86cdSEric Blake .bdrv_probe_blocksizes = hdev_probe_blocksizes, 3539c1bb86cdSEric Blake .bdrv_probe_geometry = hdev_probe_geometry, 3540c1bb86cdSEric Blake 3541c1bb86cdSEric Blake /* generic scsi device */ 3542c1bb86cdSEric Blake #ifdef __linux__ 35432f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3544c1bb86cdSEric Blake #endif 3545c1bb86cdSEric Blake }; 3546c1bb86cdSEric Blake 3547c1bb86cdSEric Blake #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3548c1bb86cdSEric Blake static void cdrom_parse_filename(const char *filename, QDict *options, 3549c1bb86cdSEric Blake Error **errp) 3550c1bb86cdSEric Blake { 355103c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); 3552c1bb86cdSEric Blake } 3553c1bb86cdSEric Blake #endif 3554c1bb86cdSEric Blake 3555c1bb86cdSEric Blake #ifdef __linux__ 3556c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3557c1bb86cdSEric Blake Error **errp) 3558c1bb86cdSEric Blake { 3559c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3560c1bb86cdSEric Blake 3561c1bb86cdSEric Blake s->type = FTYPE_CD; 3562c1bb86cdSEric Blake 3563c1bb86cdSEric Blake /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ 3564230ff739SJohn Snow return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); 3565c1bb86cdSEric Blake } 3566c1bb86cdSEric Blake 3567c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3568c1bb86cdSEric Blake { 3569c1bb86cdSEric Blake int fd, ret; 3570c1bb86cdSEric Blake int prio = 0; 3571c1bb86cdSEric Blake struct stat st; 3572c1bb86cdSEric Blake 3573c1bb86cdSEric Blake fd = qemu_open(filename, O_RDONLY | O_NONBLOCK); 3574c1bb86cdSEric Blake if (fd < 0) { 3575c1bb86cdSEric Blake goto out; 3576c1bb86cdSEric Blake } 3577c1bb86cdSEric Blake ret = fstat(fd, &st); 3578c1bb86cdSEric Blake if (ret == -1 || !S_ISBLK(st.st_mode)) { 3579c1bb86cdSEric Blake goto outc; 3580c1bb86cdSEric Blake } 3581c1bb86cdSEric Blake 3582c1bb86cdSEric Blake /* Attempt to detect via a CDROM specific ioctl */ 3583c1bb86cdSEric Blake ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3584c1bb86cdSEric Blake if (ret >= 0) 3585c1bb86cdSEric Blake prio = 100; 3586c1bb86cdSEric Blake 3587c1bb86cdSEric Blake outc: 3588c1bb86cdSEric Blake qemu_close(fd); 3589c1bb86cdSEric Blake out: 3590c1bb86cdSEric Blake return prio; 3591c1bb86cdSEric Blake } 3592c1bb86cdSEric Blake 3593c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3594c1bb86cdSEric Blake { 3595c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3596c1bb86cdSEric Blake int ret; 3597c1bb86cdSEric Blake 3598c1bb86cdSEric Blake ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3599c1bb86cdSEric Blake return ret == CDS_DISC_OK; 3600c1bb86cdSEric Blake } 3601c1bb86cdSEric Blake 3602c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3603c1bb86cdSEric Blake { 3604c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3605c1bb86cdSEric Blake 3606c1bb86cdSEric Blake if (eject_flag) { 3607c1bb86cdSEric Blake if (ioctl(s->fd, CDROMEJECT, NULL) < 0) 3608c1bb86cdSEric Blake perror("CDROMEJECT"); 3609c1bb86cdSEric Blake } else { 3610c1bb86cdSEric Blake if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) 3611c1bb86cdSEric Blake perror("CDROMEJECT"); 3612c1bb86cdSEric Blake } 3613c1bb86cdSEric Blake } 3614c1bb86cdSEric Blake 3615c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3616c1bb86cdSEric Blake { 3617c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3618c1bb86cdSEric Blake 3619c1bb86cdSEric Blake if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { 3620c1bb86cdSEric Blake /* 3621c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3622c1bb86cdSEric Blake * mounts the CD-ROM 3623c1bb86cdSEric Blake */ 3624c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3625c1bb86cdSEric Blake } 3626c1bb86cdSEric Blake } 3627c1bb86cdSEric Blake 3628c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3629c1bb86cdSEric Blake .format_name = "host_cdrom", 3630c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3631c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3632c1bb86cdSEric Blake .bdrv_needs_filename = true, 3633c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3634c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3635c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3636c1bb86cdSEric Blake .bdrv_close = raw_close, 3637c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3638c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3639c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 36408a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3641dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3642c1bb86cdSEric Blake 3643c1bb86cdSEric Blake 3644c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3645c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 364633d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3647c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3648c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3649c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3650042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3651c1bb86cdSEric Blake 3652061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3653c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3654c1bb86cdSEric Blake .has_variable_length = true, 3655c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3656c1bb86cdSEric Blake = raw_get_allocated_file_size, 3657c1bb86cdSEric Blake 3658c1bb86cdSEric Blake /* removable device support */ 3659c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3660c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3661c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3662c1bb86cdSEric Blake 3663c1bb86cdSEric Blake /* generic scsi device */ 36642f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3665c1bb86cdSEric Blake }; 3666c1bb86cdSEric Blake #endif /* __linux__ */ 3667c1bb86cdSEric Blake 3668c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 3669c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3670c1bb86cdSEric Blake Error **errp) 3671c1bb86cdSEric Blake { 3672c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3673c1bb86cdSEric Blake Error *local_err = NULL; 3674c1bb86cdSEric Blake int ret; 3675c1bb86cdSEric Blake 3676c1bb86cdSEric Blake s->type = FTYPE_CD; 3677c1bb86cdSEric Blake 3678230ff739SJohn Snow ret = raw_open_common(bs, options, flags, 0, true, &local_err); 3679c1bb86cdSEric Blake if (ret) { 3680c1bb86cdSEric Blake error_propagate(errp, local_err); 3681c1bb86cdSEric Blake return ret; 3682c1bb86cdSEric Blake } 3683c1bb86cdSEric Blake 3684c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3685c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3686c1bb86cdSEric Blake return 0; 3687c1bb86cdSEric Blake } 3688c1bb86cdSEric Blake 3689c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3690c1bb86cdSEric Blake { 3691c1bb86cdSEric Blake if (strstart(filename, "/dev/cd", NULL) || 3692c1bb86cdSEric Blake strstart(filename, "/dev/acd", NULL)) 3693c1bb86cdSEric Blake return 100; 3694c1bb86cdSEric Blake return 0; 3695c1bb86cdSEric Blake } 3696c1bb86cdSEric Blake 3697c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs) 3698c1bb86cdSEric Blake { 3699c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3700c1bb86cdSEric Blake int fd; 3701c1bb86cdSEric Blake 3702c1bb86cdSEric Blake /* 3703c1bb86cdSEric Blake * Force reread of possibly changed/newly loaded disc, 3704c1bb86cdSEric Blake * FreeBSD seems to not notice sometimes... 3705c1bb86cdSEric Blake */ 3706c1bb86cdSEric Blake if (s->fd >= 0) 3707c1bb86cdSEric Blake qemu_close(s->fd); 3708c1bb86cdSEric Blake fd = qemu_open(bs->filename, s->open_flags, 0644); 3709c1bb86cdSEric Blake if (fd < 0) { 3710c1bb86cdSEric Blake s->fd = -1; 3711c1bb86cdSEric Blake return -EIO; 3712c1bb86cdSEric Blake } 3713c1bb86cdSEric Blake s->fd = fd; 3714c1bb86cdSEric Blake 3715c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3716c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3717c1bb86cdSEric Blake return 0; 3718c1bb86cdSEric Blake } 3719c1bb86cdSEric Blake 3720c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3721c1bb86cdSEric Blake { 3722c1bb86cdSEric Blake return raw_getlength(bs) > 0; 3723c1bb86cdSEric Blake } 3724c1bb86cdSEric Blake 3725c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3726c1bb86cdSEric Blake { 3727c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3728c1bb86cdSEric Blake 3729c1bb86cdSEric Blake if (s->fd < 0) 3730c1bb86cdSEric Blake return; 3731c1bb86cdSEric Blake 3732c1bb86cdSEric Blake (void) ioctl(s->fd, CDIOCALLOW); 3733c1bb86cdSEric Blake 3734c1bb86cdSEric Blake if (eject_flag) { 3735c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCEJECT) < 0) 3736c1bb86cdSEric Blake perror("CDIOCEJECT"); 3737c1bb86cdSEric Blake } else { 3738c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCCLOSE) < 0) 3739c1bb86cdSEric Blake perror("CDIOCCLOSE"); 3740c1bb86cdSEric Blake } 3741c1bb86cdSEric Blake 3742c1bb86cdSEric Blake cdrom_reopen(bs); 3743c1bb86cdSEric Blake } 3744c1bb86cdSEric Blake 3745c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3746c1bb86cdSEric Blake { 3747c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3748c1bb86cdSEric Blake 3749c1bb86cdSEric Blake if (s->fd < 0) 3750c1bb86cdSEric Blake return; 3751c1bb86cdSEric Blake if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { 3752c1bb86cdSEric Blake /* 3753c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3754c1bb86cdSEric Blake * mounts the CD-ROM 3755c1bb86cdSEric Blake */ 3756c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3757c1bb86cdSEric Blake } 3758c1bb86cdSEric Blake } 3759c1bb86cdSEric Blake 3760c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3761c1bb86cdSEric Blake .format_name = "host_cdrom", 3762c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3763c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3764c1bb86cdSEric Blake .bdrv_needs_filename = true, 3765c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3766c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3767c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3768c1bb86cdSEric Blake .bdrv_close = raw_close, 3769c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3770c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3771c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 37728a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3773c1bb86cdSEric Blake 3774c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3775c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 377633d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3777c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3778c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3779c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3780042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3781c1bb86cdSEric Blake 3782061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3783c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3784c1bb86cdSEric Blake .has_variable_length = true, 3785c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3786c1bb86cdSEric Blake = raw_get_allocated_file_size, 3787c1bb86cdSEric Blake 3788c1bb86cdSEric Blake /* removable device support */ 3789c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3790c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3791c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3792c1bb86cdSEric Blake }; 3793c1bb86cdSEric Blake #endif /* __FreeBSD__ */ 3794c1bb86cdSEric Blake 3795c1bb86cdSEric Blake static void bdrv_file_init(void) 3796c1bb86cdSEric Blake { 3797c1bb86cdSEric Blake /* 3798c1bb86cdSEric Blake * Register all the drivers. Note that order is important, the driver 3799c1bb86cdSEric Blake * registered last will get probed first. 3800c1bb86cdSEric Blake */ 3801c1bb86cdSEric Blake bdrv_register(&bdrv_file); 3802c1bb86cdSEric Blake bdrv_register(&bdrv_host_device); 3803c1bb86cdSEric Blake #ifdef __linux__ 3804c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3805c1bb86cdSEric Blake #endif 3806c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3807c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3808c1bb86cdSEric Blake #endif 3809c1bb86cdSEric Blake } 3810c1bb86cdSEric Blake 3811c1bb86cdSEric Blake block_init(bdrv_file_init); 3812