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; 159e5bcf967SKevin Wolf bool page_cache_inconsistent:1; 160c1bb86cdSEric Blake bool has_fallocate; 161c1bb86cdSEric Blake bool needs_alignment; 162f357fcd8SStefan Hajnoczi bool drop_cache; 16331be8a2aSStefan Hajnoczi bool check_cache_dropped; 1641c450366SAnton Nefedov struct { 1651c450366SAnton Nefedov uint64_t discard_nb_ok; 1661c450366SAnton Nefedov uint64_t discard_nb_failed; 1671c450366SAnton Nefedov uint64_t discard_bytes_ok; 1681c450366SAnton Nefedov } stats; 1697c9e5276SPaolo Bonzini 1707c9e5276SPaolo Bonzini PRManager *pr_mgr; 171c1bb86cdSEric Blake } BDRVRawState; 172c1bb86cdSEric Blake 173c1bb86cdSEric Blake typedef struct BDRVRawReopenState { 174c1bb86cdSEric Blake int fd; 175c1bb86cdSEric Blake int open_flags; 176f357fcd8SStefan Hajnoczi bool drop_cache; 17731be8a2aSStefan Hajnoczi bool check_cache_dropped; 178c1bb86cdSEric Blake } BDRVRawReopenState; 179c1bb86cdSEric Blake 180c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs); 181c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs); 182c1bb86cdSEric Blake 183c1bb86cdSEric Blake typedef struct RawPosixAIOData { 184c1bb86cdSEric Blake BlockDriverState *bs; 185c1bb86cdSEric Blake int aio_type; 186d57c44d0SKevin Wolf int aio_fildes; 187d57c44d0SKevin Wolf 188d57c44d0SKevin Wolf off_t aio_offset; 189d57c44d0SKevin Wolf uint64_t aio_nbytes; 190d57c44d0SKevin Wolf 19193f4e2ffSKevin Wolf union { 19293f4e2ffSKevin Wolf struct { 193d57c44d0SKevin Wolf struct iovec *iov; 194d57c44d0SKevin Wolf int niov; 195d57c44d0SKevin Wolf } io; 196d57c44d0SKevin Wolf struct { 197d57c44d0SKevin Wolf uint64_t cmd; 198d57c44d0SKevin Wolf void *buf; 199d57c44d0SKevin Wolf } ioctl; 200d57c44d0SKevin Wolf struct { 2011efad060SFam Zheng int aio_fd2; 2021efad060SFam Zheng off_t aio_offset2; 203d57c44d0SKevin Wolf } copy_range; 20493f4e2ffSKevin Wolf struct { 20593f4e2ffSKevin Wolf PreallocMode prealloc; 20693f4e2ffSKevin Wolf Error **errp; 207d57c44d0SKevin Wolf } truncate; 20893f4e2ffSKevin Wolf }; 209c1bb86cdSEric Blake } RawPosixAIOData; 210c1bb86cdSEric Blake 211c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 212c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs); 213c1bb86cdSEric Blake #endif 214c1bb86cdSEric Blake 215c1bb86cdSEric Blake #if defined(__NetBSD__) 216db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 217c1bb86cdSEric Blake { 218c1bb86cdSEric Blake static char namebuf[PATH_MAX]; 219c1bb86cdSEric Blake const char *dp, *fname; 220c1bb86cdSEric Blake struct stat sb; 221c1bb86cdSEric Blake 222c1bb86cdSEric Blake fname = *filename; 223c1bb86cdSEric Blake dp = strrchr(fname, '/'); 224c1bb86cdSEric Blake if (lstat(fname, &sb) < 0) { 225f6fc1e30SPaolo Bonzini error_setg_file_open(errp, errno, fname); 226c1bb86cdSEric Blake return -errno; 227c1bb86cdSEric Blake } 228c1bb86cdSEric Blake 229c1bb86cdSEric Blake if (!S_ISBLK(sb.st_mode)) { 230c1bb86cdSEric Blake return 0; 231c1bb86cdSEric Blake } 232c1bb86cdSEric Blake 233c1bb86cdSEric Blake if (dp == NULL) { 234c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "r%s", fname); 235c1bb86cdSEric Blake } else { 236c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "%.*s/r%s", 237c1bb86cdSEric Blake (int)(dp - fname), fname, dp + 1); 238c1bb86cdSEric Blake } 239c1bb86cdSEric Blake *filename = namebuf; 240db0754dfSFam Zheng warn_report("%s is a block device, using %s", fname, *filename); 241c1bb86cdSEric Blake 242c1bb86cdSEric Blake return 0; 243c1bb86cdSEric Blake } 244c1bb86cdSEric Blake #else 245db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 246c1bb86cdSEric Blake { 247c1bb86cdSEric Blake return 0; 248c1bb86cdSEric Blake } 249c1bb86cdSEric Blake #endif 250c1bb86cdSEric Blake 251c1bb86cdSEric Blake /* 252c1bb86cdSEric Blake * Get logical block size via ioctl. On success store it in @sector_size_p. 253c1bb86cdSEric Blake */ 254c1bb86cdSEric Blake static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) 255c1bb86cdSEric Blake { 256c1bb86cdSEric Blake unsigned int sector_size; 257c1bb86cdSEric Blake bool success = false; 258700f9ce0SPeter Maydell int i; 259c1bb86cdSEric Blake 260c1bb86cdSEric Blake errno = ENOTSUP; 261700f9ce0SPeter Maydell static const unsigned long ioctl_list[] = { 262c1bb86cdSEric Blake #ifdef BLKSSZGET 263700f9ce0SPeter Maydell BLKSSZGET, 264c1bb86cdSEric Blake #endif 265c1bb86cdSEric Blake #ifdef DKIOCGETBLOCKSIZE 266700f9ce0SPeter Maydell DKIOCGETBLOCKSIZE, 267c1bb86cdSEric Blake #endif 268c1bb86cdSEric Blake #ifdef DIOCGSECTORSIZE 269700f9ce0SPeter Maydell DIOCGSECTORSIZE, 270700f9ce0SPeter Maydell #endif 271700f9ce0SPeter Maydell }; 272700f9ce0SPeter Maydell 273700f9ce0SPeter Maydell /* Try a few ioctls to get the right size */ 274700f9ce0SPeter Maydell for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { 275700f9ce0SPeter Maydell if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { 276c1bb86cdSEric Blake *sector_size_p = sector_size; 277c1bb86cdSEric Blake success = true; 278c1bb86cdSEric Blake } 279700f9ce0SPeter Maydell } 280c1bb86cdSEric Blake 281c1bb86cdSEric Blake return success ? 0 : -errno; 282c1bb86cdSEric Blake } 283c1bb86cdSEric Blake 284c1bb86cdSEric Blake /** 285c1bb86cdSEric Blake * Get physical block size of @fd. 286c1bb86cdSEric Blake * On success, store it in @blk_size and return 0. 287c1bb86cdSEric Blake * On failure, return -errno. 288c1bb86cdSEric Blake */ 289c1bb86cdSEric Blake static int probe_physical_blocksize(int fd, unsigned int *blk_size) 290c1bb86cdSEric Blake { 291c1bb86cdSEric Blake #ifdef BLKPBSZGET 292c1bb86cdSEric Blake if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { 293c1bb86cdSEric Blake return -errno; 294c1bb86cdSEric Blake } 295c1bb86cdSEric Blake return 0; 296c1bb86cdSEric Blake #else 297c1bb86cdSEric Blake return -ENOTSUP; 298c1bb86cdSEric Blake #endif 299c1bb86cdSEric Blake } 300c1bb86cdSEric Blake 301c1bb86cdSEric Blake /* Check if read is allowed with given memory buffer and length. 302c1bb86cdSEric Blake * 303c1bb86cdSEric Blake * This function is used to check O_DIRECT memory buffer and request alignment. 304c1bb86cdSEric Blake */ 305c1bb86cdSEric Blake static bool raw_is_io_aligned(int fd, void *buf, size_t len) 306c1bb86cdSEric Blake { 307c1bb86cdSEric Blake ssize_t ret = pread(fd, buf, len, 0); 308c1bb86cdSEric Blake 309c1bb86cdSEric Blake if (ret >= 0) { 310c1bb86cdSEric Blake return true; 311c1bb86cdSEric Blake } 312c1bb86cdSEric Blake 313c1bb86cdSEric Blake #ifdef __linux__ 314c1bb86cdSEric Blake /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore 315c1bb86cdSEric Blake * other errors (e.g. real I/O error), which could happen on a failed 316c1bb86cdSEric Blake * drive, since we only care about probing alignment. 317c1bb86cdSEric Blake */ 318c1bb86cdSEric Blake if (errno != EINVAL) { 319c1bb86cdSEric Blake return true; 320c1bb86cdSEric Blake } 321c1bb86cdSEric Blake #endif 322c1bb86cdSEric Blake 323c1bb86cdSEric Blake return false; 324c1bb86cdSEric Blake } 325c1bb86cdSEric Blake 326c1bb86cdSEric Blake static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) 327c1bb86cdSEric Blake { 328c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 329c1bb86cdSEric Blake char *buf; 330*038adc2fSWei Yang size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); 331a6b257a0SNir Soffer size_t alignments[] = {1, 512, 1024, 2048, 4096}; 332c1bb86cdSEric Blake 333c1bb86cdSEric Blake /* For SCSI generic devices the alignment is not really used. 334c1bb86cdSEric Blake With buffered I/O, we don't have any restrictions. */ 335c1bb86cdSEric Blake if (bdrv_is_sg(bs) || !s->needs_alignment) { 336c1bb86cdSEric Blake bs->bl.request_alignment = 1; 337c1bb86cdSEric Blake s->buf_align = 1; 338c1bb86cdSEric Blake return; 339c1bb86cdSEric Blake } 340c1bb86cdSEric Blake 341c1bb86cdSEric Blake bs->bl.request_alignment = 0; 342c1bb86cdSEric Blake s->buf_align = 0; 343c1bb86cdSEric Blake /* Let's try to use the logical blocksize for the alignment. */ 344c1bb86cdSEric Blake if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { 345c1bb86cdSEric Blake bs->bl.request_alignment = 0; 346c1bb86cdSEric Blake } 347c1bb86cdSEric Blake #ifdef CONFIG_XFS 348c1bb86cdSEric Blake if (s->is_xfs) { 349c1bb86cdSEric Blake struct dioattr da; 350c1bb86cdSEric Blake if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { 351c1bb86cdSEric Blake bs->bl.request_alignment = da.d_miniosz; 352c1bb86cdSEric Blake /* The kernel returns wrong information for d_mem */ 353c1bb86cdSEric Blake /* s->buf_align = da.d_mem; */ 354c1bb86cdSEric Blake } 355c1bb86cdSEric Blake } 356c1bb86cdSEric Blake #endif 357c1bb86cdSEric Blake 358a6b257a0SNir Soffer /* 359a6b257a0SNir Soffer * If we could not get the sizes so far, we can only guess them. First try 360a6b257a0SNir Soffer * to detect request alignment, since it is more likely to succeed. Then 361a6b257a0SNir Soffer * try to detect buf_align, which cannot be detected in some cases (e.g. 362a6b257a0SNir Soffer * Gluster). If buf_align cannot be detected, we fallback to the value of 363a6b257a0SNir Soffer * request_alignment. 364a6b257a0SNir Soffer */ 365a6b257a0SNir Soffer 366a6b257a0SNir Soffer if (!bs->bl.request_alignment) { 367a6b257a0SNir Soffer int i; 368c1bb86cdSEric Blake size_t align; 369a6b257a0SNir Soffer buf = qemu_memalign(max_align, max_align); 370a6b257a0SNir Soffer for (i = 0; i < ARRAY_SIZE(alignments); i++) { 371a6b257a0SNir Soffer align = alignments[i]; 372a6b257a0SNir Soffer if (raw_is_io_aligned(fd, buf, align)) { 373a6b257a0SNir Soffer /* Fallback to safe value. */ 374a6b257a0SNir Soffer bs->bl.request_alignment = (align != 1) ? align : max_align; 375c1bb86cdSEric Blake break; 376c1bb86cdSEric Blake } 377c1bb86cdSEric Blake } 378c1bb86cdSEric Blake qemu_vfree(buf); 379c1bb86cdSEric Blake } 380c1bb86cdSEric Blake 381a6b257a0SNir Soffer if (!s->buf_align) { 382a6b257a0SNir Soffer int i; 383c1bb86cdSEric Blake size_t align; 384a6b257a0SNir Soffer buf = qemu_memalign(max_align, 2 * max_align); 385a6b257a0SNir Soffer for (i = 0; i < ARRAY_SIZE(alignments); i++) { 386a6b257a0SNir Soffer align = alignments[i]; 387a6b257a0SNir Soffer if (raw_is_io_aligned(fd, buf + align, max_align)) { 388236094c7SStefan Hajnoczi /* Fallback to request_alignment. */ 389a6b257a0SNir Soffer s->buf_align = (align != 1) ? align : bs->bl.request_alignment; 390c1bb86cdSEric Blake break; 391c1bb86cdSEric Blake } 392c1bb86cdSEric Blake } 393c1bb86cdSEric Blake qemu_vfree(buf); 394c1bb86cdSEric Blake } 395c1bb86cdSEric Blake 396c1bb86cdSEric Blake if (!s->buf_align || !bs->bl.request_alignment) { 397c1bb86cdSEric Blake error_setg(errp, "Could not find working O_DIRECT alignment"); 398c1bb86cdSEric Blake error_append_hint(errp, "Try cache.direct=off\n"); 399c1bb86cdSEric Blake } 400c1bb86cdSEric Blake } 401c1bb86cdSEric Blake 40223dece19SKevin Wolf static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) 403c1bb86cdSEric Blake { 40423dece19SKevin Wolf bool read_write = false; 405c1bb86cdSEric Blake assert(open_flags != NULL); 406c1bb86cdSEric Blake 407c1bb86cdSEric Blake *open_flags |= O_BINARY; 408c1bb86cdSEric Blake *open_flags &= ~O_ACCMODE; 40923dece19SKevin Wolf 41023dece19SKevin Wolf if (bdrv_flags & BDRV_O_AUTO_RDONLY) { 41123dece19SKevin Wolf read_write = has_writers; 41223dece19SKevin Wolf } else if (bdrv_flags & BDRV_O_RDWR) { 41323dece19SKevin Wolf read_write = true; 41423dece19SKevin Wolf } 41523dece19SKevin Wolf 41623dece19SKevin Wolf if (read_write) { 417c1bb86cdSEric Blake *open_flags |= O_RDWR; 418c1bb86cdSEric Blake } else { 419c1bb86cdSEric Blake *open_flags |= O_RDONLY; 420c1bb86cdSEric Blake } 421c1bb86cdSEric Blake 422c1bb86cdSEric Blake /* Use O_DSYNC for write-through caching, no flags for write-back caching, 423c1bb86cdSEric Blake * and O_DIRECT for no caching. */ 424c1bb86cdSEric Blake if ((bdrv_flags & BDRV_O_NOCACHE)) { 425c1bb86cdSEric Blake *open_flags |= O_DIRECT; 426c1bb86cdSEric Blake } 427c1bb86cdSEric Blake } 428c1bb86cdSEric Blake 429c1bb86cdSEric Blake static void raw_parse_filename(const char *filename, QDict *options, 430c1bb86cdSEric Blake Error **errp) 431c1bb86cdSEric Blake { 43203c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "file:", options); 433c1bb86cdSEric Blake } 434c1bb86cdSEric Blake 435c1bb86cdSEric Blake static QemuOptsList raw_runtime_opts = { 436c1bb86cdSEric Blake .name = "raw", 437c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), 438c1bb86cdSEric Blake .desc = { 439c1bb86cdSEric Blake { 440c1bb86cdSEric Blake .name = "filename", 441c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 442c1bb86cdSEric Blake .help = "File name of the image", 443c1bb86cdSEric Blake }, 444c1bb86cdSEric Blake { 445c1bb86cdSEric Blake .name = "aio", 446c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 447c1bb86cdSEric Blake .help = "host AIO implementation (threads, native)", 448c1bb86cdSEric Blake }, 44916b48d5dSFam Zheng { 45016b48d5dSFam Zheng .name = "locking", 45116b48d5dSFam Zheng .type = QEMU_OPT_STRING, 45216b48d5dSFam Zheng .help = "file locking mode (on/off/auto, default: auto)", 45316b48d5dSFam Zheng }, 4547c9e5276SPaolo Bonzini { 4557c9e5276SPaolo Bonzini .name = "pr-manager", 4567c9e5276SPaolo Bonzini .type = QEMU_OPT_STRING, 4577c9e5276SPaolo Bonzini .help = "id of persistent reservation manager object (default: none)", 4587c9e5276SPaolo Bonzini }, 459f357fcd8SStefan Hajnoczi #if defined(__linux__) 460f357fcd8SStefan Hajnoczi { 461f357fcd8SStefan Hajnoczi .name = "drop-cache", 462f357fcd8SStefan Hajnoczi .type = QEMU_OPT_BOOL, 463f357fcd8SStefan Hajnoczi .help = "invalidate page cache during live migration (default: on)", 464f357fcd8SStefan Hajnoczi }, 465f357fcd8SStefan Hajnoczi #endif 46631be8a2aSStefan Hajnoczi { 46731be8a2aSStefan Hajnoczi .name = "x-check-cache-dropped", 46831be8a2aSStefan Hajnoczi .type = QEMU_OPT_BOOL, 46931be8a2aSStefan Hajnoczi .help = "check that page cache was dropped on live migration (default: off)" 47031be8a2aSStefan Hajnoczi }, 471c1bb86cdSEric Blake { /* end of list */ } 472c1bb86cdSEric Blake }, 473c1bb86cdSEric Blake }; 474c1bb86cdSEric Blake 4758a2ce0bcSAlberto Garcia static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; 4768a2ce0bcSAlberto Garcia 477c1bb86cdSEric Blake static int raw_open_common(BlockDriverState *bs, QDict *options, 478230ff739SJohn Snow int bdrv_flags, int open_flags, 479230ff739SJohn Snow bool device, Error **errp) 480c1bb86cdSEric Blake { 481c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 482c1bb86cdSEric Blake QemuOpts *opts; 483c1bb86cdSEric Blake Error *local_err = NULL; 484c1bb86cdSEric Blake const char *filename = NULL; 4857c9e5276SPaolo Bonzini const char *str; 486c1bb86cdSEric Blake BlockdevAioOptions aio, aio_default; 487c1bb86cdSEric Blake int fd, ret; 488c1bb86cdSEric Blake struct stat st; 489244a5668SFam Zheng OnOffAuto locking; 490c1bb86cdSEric Blake 491c1bb86cdSEric Blake opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 492c1bb86cdSEric Blake qemu_opts_absorb_qdict(opts, options, &local_err); 493c1bb86cdSEric Blake if (local_err) { 494c1bb86cdSEric Blake error_propagate(errp, local_err); 495c1bb86cdSEric Blake ret = -EINVAL; 496c1bb86cdSEric Blake goto fail; 497c1bb86cdSEric Blake } 498c1bb86cdSEric Blake 499c1bb86cdSEric Blake filename = qemu_opt_get(opts, "filename"); 500c1bb86cdSEric Blake 501db0754dfSFam Zheng ret = raw_normalize_devicepath(&filename, errp); 502c1bb86cdSEric Blake if (ret != 0) { 503c1bb86cdSEric Blake goto fail; 504c1bb86cdSEric Blake } 505c1bb86cdSEric Blake 506c1bb86cdSEric Blake aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO) 507c1bb86cdSEric Blake ? BLOCKDEV_AIO_OPTIONS_NATIVE 508c1bb86cdSEric Blake : BLOCKDEV_AIO_OPTIONS_THREADS; 509f7abe0ecSMarc-André Lureau aio = qapi_enum_parse(&BlockdevAioOptions_lookup, 510f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "aio"), 51106c60b6cSMarkus Armbruster aio_default, &local_err); 512c1bb86cdSEric Blake if (local_err) { 513c1bb86cdSEric Blake error_propagate(errp, local_err); 514c1bb86cdSEric Blake ret = -EINVAL; 515c1bb86cdSEric Blake goto fail; 516c1bb86cdSEric Blake } 517c1bb86cdSEric Blake s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); 518c1bb86cdSEric Blake 519f7abe0ecSMarc-André Lureau locking = qapi_enum_parse(&OnOffAuto_lookup, 520f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "locking"), 52106c60b6cSMarkus Armbruster ON_OFF_AUTO_AUTO, &local_err); 522244a5668SFam Zheng if (local_err) { 523244a5668SFam Zheng error_propagate(errp, local_err); 524244a5668SFam Zheng ret = -EINVAL; 525244a5668SFam Zheng goto fail; 526244a5668SFam Zheng } 527244a5668SFam Zheng switch (locking) { 528244a5668SFam Zheng case ON_OFF_AUTO_ON: 529244a5668SFam Zheng s->use_lock = true; 5302b218f5dSFam Zheng if (!qemu_has_ofd_lock()) { 531db0754dfSFam Zheng warn_report("File lock requested but OFD locking syscall is " 532db0754dfSFam Zheng "unavailable, falling back to POSIX file locks"); 533db0754dfSFam Zheng error_printf("Due to the implementation, locks can be lost " 5342b218f5dSFam Zheng "unexpectedly.\n"); 5352b218f5dSFam Zheng } 536244a5668SFam Zheng break; 537244a5668SFam Zheng case ON_OFF_AUTO_OFF: 538244a5668SFam Zheng s->use_lock = false; 539244a5668SFam Zheng break; 540244a5668SFam Zheng case ON_OFF_AUTO_AUTO: 5412b218f5dSFam Zheng s->use_lock = qemu_has_ofd_lock(); 542244a5668SFam Zheng break; 543244a5668SFam Zheng default: 544244a5668SFam Zheng abort(); 545244a5668SFam Zheng } 546244a5668SFam Zheng 5477c9e5276SPaolo Bonzini str = qemu_opt_get(opts, "pr-manager"); 5487c9e5276SPaolo Bonzini if (str) { 5497c9e5276SPaolo Bonzini s->pr_mgr = pr_manager_lookup(str, &local_err); 5507c9e5276SPaolo Bonzini if (local_err) { 5517c9e5276SPaolo Bonzini error_propagate(errp, local_err); 5527c9e5276SPaolo Bonzini ret = -EINVAL; 5537c9e5276SPaolo Bonzini goto fail; 5547c9e5276SPaolo Bonzini } 5557c9e5276SPaolo Bonzini } 5567c9e5276SPaolo Bonzini 557f357fcd8SStefan Hajnoczi s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); 55831be8a2aSStefan Hajnoczi s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", 55931be8a2aSStefan Hajnoczi false); 56031be8a2aSStefan Hajnoczi 561c1bb86cdSEric Blake s->open_flags = open_flags; 56223dece19SKevin Wolf raw_parse_flags(bdrv_flags, &s->open_flags, false); 563c1bb86cdSEric Blake 564c1bb86cdSEric Blake s->fd = -1; 565c1bb86cdSEric Blake fd = qemu_open(filename, s->open_flags, 0644); 56664107dc0SKevin Wolf ret = fd < 0 ? -errno : 0; 56764107dc0SKevin Wolf 56864107dc0SKevin Wolf if (ret < 0) { 569f6fc1e30SPaolo Bonzini error_setg_file_open(errp, -ret, filename); 570c1bb86cdSEric Blake if (ret == -EROFS) { 571c1bb86cdSEric Blake ret = -EACCES; 572c1bb86cdSEric Blake } 573c1bb86cdSEric Blake goto fail; 574c1bb86cdSEric Blake } 575c1bb86cdSEric Blake s->fd = fd; 576c1bb86cdSEric Blake 577244a5668SFam Zheng s->perm = 0; 578244a5668SFam Zheng s->shared_perm = BLK_PERM_ALL; 579244a5668SFam Zheng 580c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 581c1bb86cdSEric Blake /* Currently Linux does AIO only for files opened with O_DIRECT */ 582ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 583ed6e2161SNishanth Aravamudan if (!(s->open_flags & O_DIRECT)) { 584c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but it requires " 585c1bb86cdSEric Blake "cache.direct=on, which was not specified."); 586c1bb86cdSEric Blake ret = -EINVAL; 587c1bb86cdSEric Blake goto fail; 588c1bb86cdSEric Blake } 589ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { 590ed6e2161SNishanth Aravamudan error_prepend(errp, "Unable to use native AIO: "); 591ed6e2161SNishanth Aravamudan goto fail; 592ed6e2161SNishanth Aravamudan } 593ed6e2161SNishanth Aravamudan } 594c1bb86cdSEric Blake #else 595c1bb86cdSEric Blake if (s->use_linux_aio) { 596c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but is not supported " 597c1bb86cdSEric Blake "in this build."); 598c1bb86cdSEric Blake ret = -EINVAL; 599c1bb86cdSEric Blake goto fail; 600c1bb86cdSEric Blake } 601c1bb86cdSEric Blake #endif /* !defined(CONFIG_LINUX_AIO) */ 602c1bb86cdSEric Blake 603c1bb86cdSEric Blake s->has_discard = true; 604c1bb86cdSEric Blake s->has_write_zeroes = true; 605c1bb86cdSEric Blake if ((bs->open_flags & BDRV_O_NOCACHE) != 0) { 606c1bb86cdSEric Blake s->needs_alignment = true; 607c1bb86cdSEric Blake } 608c1bb86cdSEric Blake 609c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 610c1bb86cdSEric Blake ret = -errno; 611c1bb86cdSEric Blake error_setg_errno(errp, errno, "Could not stat file"); 612c1bb86cdSEric Blake goto fail; 613c1bb86cdSEric Blake } 614230ff739SJohn Snow 615230ff739SJohn Snow if (!device) { 616230ff739SJohn Snow if (S_ISBLK(st.st_mode)) { 617230ff739SJohn Snow warn_report("Opening a block device as a file using the '%s' " 618230ff739SJohn Snow "driver is deprecated", bs->drv->format_name); 619230ff739SJohn Snow } else if (S_ISCHR(st.st_mode)) { 620230ff739SJohn Snow warn_report("Opening a character device as a file using the '%s' " 621230ff739SJohn Snow "driver is deprecated", bs->drv->format_name); 622230ff739SJohn Snow } else if (!S_ISREG(st.st_mode)) { 623230ff739SJohn Snow error_setg(errp, "A regular file was expected by the '%s' driver, " 624230ff739SJohn Snow "but something else was given", bs->drv->format_name); 625230ff739SJohn Snow ret = -EINVAL; 626230ff739SJohn Snow goto fail; 627230ff739SJohn Snow } else { 628c1bb86cdSEric Blake s->discard_zeroes = true; 629c1bb86cdSEric Blake s->has_fallocate = true; 630c1bb86cdSEric Blake } 631230ff739SJohn Snow } else { 632230ff739SJohn Snow if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 633230ff739SJohn Snow error_setg(errp, "'%s' driver expects either " 634230ff739SJohn Snow "a character or block device", bs->drv->format_name); 635230ff739SJohn Snow ret = -EINVAL; 636230ff739SJohn Snow goto fail; 637230ff739SJohn Snow } 638230ff739SJohn Snow } 639230ff739SJohn Snow 640c1bb86cdSEric Blake if (S_ISBLK(st.st_mode)) { 641c1bb86cdSEric Blake #ifdef BLKDISCARDZEROES 642c1bb86cdSEric Blake unsigned int arg; 643c1bb86cdSEric Blake if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { 644c1bb86cdSEric Blake s->discard_zeroes = true; 645c1bb86cdSEric Blake } 646c1bb86cdSEric Blake #endif 647c1bb86cdSEric Blake #ifdef __linux__ 648c1bb86cdSEric Blake /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do 649c1bb86cdSEric Blake * not rely on the contents of discarded blocks unless using O_DIRECT. 650c1bb86cdSEric Blake * Same for BLKZEROOUT. 651c1bb86cdSEric Blake */ 652c1bb86cdSEric Blake if (!(bs->open_flags & BDRV_O_NOCACHE)) { 653c1bb86cdSEric Blake s->discard_zeroes = false; 654c1bb86cdSEric Blake s->has_write_zeroes = false; 655c1bb86cdSEric Blake } 656c1bb86cdSEric Blake #endif 657c1bb86cdSEric Blake } 658c1bb86cdSEric Blake #ifdef __FreeBSD__ 659c1bb86cdSEric Blake if (S_ISCHR(st.st_mode)) { 660c1bb86cdSEric Blake /* 661c1bb86cdSEric Blake * The file is a char device (disk), which on FreeBSD isn't behind 662c1bb86cdSEric Blake * a pager, so force all requests to be aligned. This is needed 663c1bb86cdSEric Blake * so QEMU makes sure all IO operations on the device are aligned 664c1bb86cdSEric Blake * to sector size, or else FreeBSD will reject them with EINVAL. 665c1bb86cdSEric Blake */ 666c1bb86cdSEric Blake s->needs_alignment = true; 667c1bb86cdSEric Blake } 668c1bb86cdSEric Blake #endif 669c1bb86cdSEric Blake 670c1bb86cdSEric Blake #ifdef CONFIG_XFS 671c1bb86cdSEric Blake if (platform_test_xfs_fd(s->fd)) { 672c1bb86cdSEric Blake s->is_xfs = true; 673c1bb86cdSEric Blake } 674c1bb86cdSEric Blake #endif 675c1bb86cdSEric Blake 676738301e1SKevin Wolf bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; 677c1bb86cdSEric Blake ret = 0; 678c1bb86cdSEric Blake fail: 679c1bb86cdSEric Blake if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { 680c1bb86cdSEric Blake unlink(filename); 681c1bb86cdSEric Blake } 682c1bb86cdSEric Blake qemu_opts_del(opts); 683c1bb86cdSEric Blake return ret; 684c1bb86cdSEric Blake } 685c1bb86cdSEric Blake 686c1bb86cdSEric Blake static int raw_open(BlockDriverState *bs, QDict *options, int flags, 687c1bb86cdSEric Blake Error **errp) 688c1bb86cdSEric Blake { 689c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 690c1bb86cdSEric Blake 691c1bb86cdSEric Blake s->type = FTYPE_FILE; 692230ff739SJohn Snow return raw_open_common(bs, options, flags, 0, false, errp); 693c1bb86cdSEric Blake } 694c1bb86cdSEric Blake 695244a5668SFam Zheng typedef enum { 696244a5668SFam Zheng RAW_PL_PREPARE, 697244a5668SFam Zheng RAW_PL_COMMIT, 698244a5668SFam Zheng RAW_PL_ABORT, 699244a5668SFam Zheng } RawPermLockOp; 700244a5668SFam Zheng 701244a5668SFam Zheng #define PERM_FOREACH(i) \ 702244a5668SFam Zheng for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) 703244a5668SFam Zheng 704244a5668SFam Zheng /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the 705244a5668SFam Zheng * file; if @unlock == true, also unlock the unneeded bytes. 706244a5668SFam Zheng * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. 707244a5668SFam Zheng */ 7082996ffadSFam Zheng static int raw_apply_lock_bytes(BDRVRawState *s, int fd, 709244a5668SFam Zheng uint64_t perm_lock_bits, 710244a5668SFam Zheng uint64_t shared_perm_lock_bits, 711244a5668SFam Zheng bool unlock, Error **errp) 712244a5668SFam Zheng { 713244a5668SFam Zheng int ret; 714244a5668SFam Zheng int i; 7152996ffadSFam Zheng uint64_t locked_perm, locked_shared_perm; 7162996ffadSFam Zheng 7172996ffadSFam Zheng if (s) { 7182996ffadSFam Zheng locked_perm = s->locked_perm; 7192996ffadSFam Zheng locked_shared_perm = s->locked_shared_perm; 7202996ffadSFam Zheng } else { 7212996ffadSFam Zheng /* 7222996ffadSFam Zheng * We don't have the previous bits, just lock/unlock for each of the 7232996ffadSFam Zheng * requested bits. 7242996ffadSFam Zheng */ 7252996ffadSFam Zheng if (unlock) { 7262996ffadSFam Zheng locked_perm = BLK_PERM_ALL; 7272996ffadSFam Zheng locked_shared_perm = BLK_PERM_ALL; 7282996ffadSFam Zheng } else { 7292996ffadSFam Zheng locked_perm = 0; 7302996ffadSFam Zheng locked_shared_perm = 0; 7312996ffadSFam Zheng } 7322996ffadSFam Zheng } 733244a5668SFam Zheng 734244a5668SFam Zheng PERM_FOREACH(i) { 735244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 7362996ffadSFam Zheng uint64_t bit = (1ULL << i); 7372996ffadSFam Zheng if ((perm_lock_bits & bit) && !(locked_perm & bit)) { 738d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 739244a5668SFam Zheng if (ret) { 740244a5668SFam Zheng error_setg(errp, "Failed to lock byte %d", off); 741244a5668SFam Zheng return ret; 7422996ffadSFam Zheng } else if (s) { 7432996ffadSFam Zheng s->locked_perm |= bit; 744244a5668SFam Zheng } 7452996ffadSFam Zheng } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { 746d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 747244a5668SFam Zheng if (ret) { 748244a5668SFam Zheng error_setg(errp, "Failed to unlock byte %d", off); 749244a5668SFam Zheng return ret; 7502996ffadSFam Zheng } else if (s) { 7512996ffadSFam Zheng s->locked_perm &= ~bit; 752244a5668SFam Zheng } 753244a5668SFam Zheng } 754244a5668SFam Zheng } 755244a5668SFam Zheng PERM_FOREACH(i) { 756244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 7572996ffadSFam Zheng uint64_t bit = (1ULL << i); 7582996ffadSFam Zheng if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { 759d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 760244a5668SFam Zheng if (ret) { 761244a5668SFam Zheng error_setg(errp, "Failed to lock byte %d", off); 762244a5668SFam Zheng return ret; 7632996ffadSFam Zheng } else if (s) { 7642996ffadSFam Zheng s->locked_shared_perm |= bit; 765244a5668SFam Zheng } 7662996ffadSFam Zheng } else if (unlock && (locked_shared_perm & bit) && 7672996ffadSFam Zheng !(shared_perm_lock_bits & bit)) { 768d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 769244a5668SFam Zheng if (ret) { 770244a5668SFam Zheng error_setg(errp, "Failed to unlock byte %d", off); 771244a5668SFam Zheng return ret; 7722996ffadSFam Zheng } else if (s) { 7732996ffadSFam Zheng s->locked_shared_perm &= ~bit; 774244a5668SFam Zheng } 775244a5668SFam Zheng } 776244a5668SFam Zheng } 777244a5668SFam Zheng return 0; 778244a5668SFam Zheng } 779244a5668SFam Zheng 780244a5668SFam Zheng /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ 781d0a96155SMax Reitz static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, 782244a5668SFam Zheng Error **errp) 783244a5668SFam Zheng { 784244a5668SFam Zheng int ret; 785244a5668SFam Zheng int i; 786244a5668SFam Zheng 787244a5668SFam Zheng PERM_FOREACH(i) { 788244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 789244a5668SFam Zheng uint64_t p = 1ULL << i; 790244a5668SFam Zheng if (perm & p) { 791d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 792244a5668SFam Zheng if (ret) { 793244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 794244a5668SFam Zheng error_setg(errp, 795244a5668SFam Zheng "Failed to get \"%s\" lock", 796244a5668SFam Zheng perm_name); 797244a5668SFam Zheng g_free(perm_name); 798244a5668SFam Zheng return ret; 799244a5668SFam Zheng } 800244a5668SFam Zheng } 801244a5668SFam Zheng } 802244a5668SFam Zheng PERM_FOREACH(i) { 803244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 804244a5668SFam Zheng uint64_t p = 1ULL << i; 805244a5668SFam Zheng if (!(shared_perm & p)) { 806d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 807244a5668SFam Zheng if (ret) { 808244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 809244a5668SFam Zheng error_setg(errp, 810244a5668SFam Zheng "Failed to get shared \"%s\" lock", 811244a5668SFam Zheng perm_name); 812244a5668SFam Zheng g_free(perm_name); 813244a5668SFam Zheng return ret; 814244a5668SFam Zheng } 815244a5668SFam Zheng } 816244a5668SFam Zheng } 817244a5668SFam Zheng return 0; 818244a5668SFam Zheng } 819244a5668SFam Zheng 820244a5668SFam Zheng static int raw_handle_perm_lock(BlockDriverState *bs, 821244a5668SFam Zheng RawPermLockOp op, 822244a5668SFam Zheng uint64_t new_perm, uint64_t new_shared, 823244a5668SFam Zheng Error **errp) 824244a5668SFam Zheng { 825244a5668SFam Zheng BDRVRawState *s = bs->opaque; 826244a5668SFam Zheng int ret = 0; 827244a5668SFam Zheng Error *local_err = NULL; 828244a5668SFam Zheng 829244a5668SFam Zheng if (!s->use_lock) { 830244a5668SFam Zheng return 0; 831244a5668SFam Zheng } 832244a5668SFam Zheng 833244a5668SFam Zheng if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { 834244a5668SFam Zheng return 0; 835244a5668SFam Zheng } 836244a5668SFam Zheng 837244a5668SFam Zheng switch (op) { 838244a5668SFam Zheng case RAW_PL_PREPARE: 839696aaaedSVladimir Sementsov-Ogievskiy if ((s->perm | new_perm) == s->perm && 840696aaaedSVladimir Sementsov-Ogievskiy (s->shared_perm & new_shared) == s->shared_perm) 841696aaaedSVladimir Sementsov-Ogievskiy { 842696aaaedSVladimir Sementsov-Ogievskiy /* 843696aaaedSVladimir Sementsov-Ogievskiy * We are going to unlock bytes, it should not fail. If it fail due 844696aaaedSVladimir Sementsov-Ogievskiy * to some fs-dependent permission-unrelated reasons (which occurs 845696aaaedSVladimir Sementsov-Ogievskiy * sometimes on NFS and leads to abort in bdrv_replace_child) we 846696aaaedSVladimir Sementsov-Ogievskiy * can't prevent such errors by any check here. And we ignore them 847696aaaedSVladimir Sementsov-Ogievskiy * anyway in ABORT and COMMIT. 848696aaaedSVladimir Sementsov-Ogievskiy */ 849696aaaedSVladimir Sementsov-Ogievskiy return 0; 850696aaaedSVladimir Sementsov-Ogievskiy } 851f2e3af29SFam Zheng ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, 852244a5668SFam Zheng ~s->shared_perm | ~new_shared, 853244a5668SFam Zheng false, errp); 854244a5668SFam Zheng if (!ret) { 855f2e3af29SFam Zheng ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); 856244a5668SFam Zheng if (!ret) { 857244a5668SFam Zheng return 0; 858244a5668SFam Zheng } 859b857431dSFam Zheng error_append_hint(errp, 860b857431dSFam Zheng "Is another process using the image [%s]?\n", 861b857431dSFam Zheng bs->filename); 862244a5668SFam Zheng } 863244a5668SFam Zheng op = RAW_PL_ABORT; 864244a5668SFam Zheng /* fall through to unlock bytes. */ 865244a5668SFam Zheng case RAW_PL_ABORT: 866f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, 867d0a96155SMax Reitz true, &local_err); 868244a5668SFam Zheng if (local_err) { 869244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 870244a5668SFam Zheng * fail. Something weird happened, report it. 871244a5668SFam Zheng */ 872db0754dfSFam Zheng warn_report_err(local_err); 873244a5668SFam Zheng } 874244a5668SFam Zheng break; 875244a5668SFam Zheng case RAW_PL_COMMIT: 876f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, 877d0a96155SMax Reitz true, &local_err); 878244a5668SFam Zheng if (local_err) { 879244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 880244a5668SFam Zheng * fail. Something weird happened, report it. 881244a5668SFam Zheng */ 882db0754dfSFam Zheng warn_report_err(local_err); 883244a5668SFam Zheng } 884244a5668SFam Zheng break; 885244a5668SFam Zheng } 886244a5668SFam Zheng return ret; 887244a5668SFam Zheng } 888244a5668SFam Zheng 8895cec2870SKevin Wolf static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, 89023dece19SKevin Wolf int *open_flags, uint64_t perm, bool force_dup, 8916ceabe6fSKevin Wolf Error **errp) 8925cec2870SKevin Wolf { 8935cec2870SKevin Wolf BDRVRawState *s = bs->opaque; 8945cec2870SKevin Wolf int fd = -1; 8955cec2870SKevin Wolf int ret; 89623dece19SKevin Wolf bool has_writers = perm & 89723dece19SKevin Wolf (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); 8985cec2870SKevin Wolf int fcntl_flags = O_APPEND | O_NONBLOCK; 8995cec2870SKevin Wolf #ifdef O_NOATIME 9005cec2870SKevin Wolf fcntl_flags |= O_NOATIME; 9015cec2870SKevin Wolf #endif 9025cec2870SKevin Wolf 9035cec2870SKevin Wolf *open_flags = 0; 9045cec2870SKevin Wolf if (s->type == FTYPE_CD) { 9055cec2870SKevin Wolf *open_flags |= O_NONBLOCK; 9065cec2870SKevin Wolf } 9075cec2870SKevin Wolf 90823dece19SKevin Wolf raw_parse_flags(flags, open_flags, has_writers); 9095cec2870SKevin Wolf 9105cec2870SKevin Wolf #ifdef O_ASYNC 9115cec2870SKevin Wolf /* Not all operating systems have O_ASYNC, and those that don't 9125cec2870SKevin Wolf * will not let us track the state into rs->open_flags (typically 9135cec2870SKevin Wolf * you achieve the same effect with an ioctl, for example I_SETSIG 9145cec2870SKevin Wolf * on Solaris). But we do not use O_ASYNC, so that's fine. 9155cec2870SKevin Wolf */ 9165cec2870SKevin Wolf assert((s->open_flags & O_ASYNC) == 0); 9175cec2870SKevin Wolf #endif 9185cec2870SKevin Wolf 9196ceabe6fSKevin Wolf if (!force_dup && *open_flags == s->open_flags) { 9206ceabe6fSKevin Wolf /* We're lucky, the existing fd is fine */ 9216ceabe6fSKevin Wolf return s->fd; 9226ceabe6fSKevin Wolf } 9236ceabe6fSKevin Wolf 9245cec2870SKevin Wolf if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { 9255cec2870SKevin Wolf /* dup the original fd */ 9265cec2870SKevin Wolf fd = qemu_dup(s->fd); 9275cec2870SKevin Wolf if (fd >= 0) { 9285cec2870SKevin Wolf ret = fcntl_setfl(fd, *open_flags); 9295cec2870SKevin Wolf if (ret) { 9305cec2870SKevin Wolf qemu_close(fd); 9315cec2870SKevin Wolf fd = -1; 9325cec2870SKevin Wolf } 9335cec2870SKevin Wolf } 9345cec2870SKevin Wolf } 9355cec2870SKevin Wolf 9365cec2870SKevin Wolf /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ 9375cec2870SKevin Wolf if (fd == -1) { 9385cec2870SKevin Wolf const char *normalized_filename = bs->filename; 9395cec2870SKevin Wolf ret = raw_normalize_devicepath(&normalized_filename, errp); 9405cec2870SKevin Wolf if (ret >= 0) { 9415cec2870SKevin Wolf assert(!(*open_flags & O_CREAT)); 9425cec2870SKevin Wolf fd = qemu_open(normalized_filename, *open_flags); 9435cec2870SKevin Wolf if (fd == -1) { 9445cec2870SKevin Wolf error_setg_errno(errp, errno, "Could not reopen file"); 9455cec2870SKevin Wolf return -1; 9465cec2870SKevin Wolf } 9475cec2870SKevin Wolf } 9485cec2870SKevin Wolf } 9495cec2870SKevin Wolf 9505cec2870SKevin Wolf return fd; 9515cec2870SKevin Wolf } 9525cec2870SKevin Wolf 953c1bb86cdSEric Blake static int raw_reopen_prepare(BDRVReopenState *state, 954c1bb86cdSEric Blake BlockReopenQueue *queue, Error **errp) 955c1bb86cdSEric Blake { 956c1bb86cdSEric Blake BDRVRawState *s; 957c1bb86cdSEric Blake BDRVRawReopenState *rs; 95831be8a2aSStefan Hajnoczi QemuOpts *opts; 959a6aeca0cSKevin Wolf int ret; 960c1bb86cdSEric Blake Error *local_err = NULL; 961c1bb86cdSEric Blake 962c1bb86cdSEric Blake assert(state != NULL); 963c1bb86cdSEric Blake assert(state->bs != NULL); 964c1bb86cdSEric Blake 965c1bb86cdSEric Blake s = state->bs->opaque; 966c1bb86cdSEric Blake 967c1bb86cdSEric Blake state->opaque = g_new0(BDRVRawReopenState, 1); 968c1bb86cdSEric Blake rs = state->opaque; 96931be8a2aSStefan Hajnoczi 97031be8a2aSStefan Hajnoczi /* Handle options changes */ 97131be8a2aSStefan Hajnoczi opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 97231be8a2aSStefan Hajnoczi qemu_opts_absorb_qdict(opts, state->options, &local_err); 97331be8a2aSStefan Hajnoczi if (local_err) { 97431be8a2aSStefan Hajnoczi error_propagate(errp, local_err); 97531be8a2aSStefan Hajnoczi ret = -EINVAL; 97631be8a2aSStefan Hajnoczi goto out; 97731be8a2aSStefan Hajnoczi } 97831be8a2aSStefan Hajnoczi 979f357fcd8SStefan Hajnoczi rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); 9808d324575SAlberto Garcia rs->check_cache_dropped = 9818d324575SAlberto Garcia qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); 9828d324575SAlberto Garcia 9838d324575SAlberto Garcia /* This driver's reopen function doesn't currently allow changing 9848d324575SAlberto Garcia * other options, so let's put them back in the original QDict and 9858d324575SAlberto Garcia * bdrv_reopen_prepare() will detect changes and complain. */ 9868d324575SAlberto Garcia qemu_opts_to_qdict(opts, state->options); 987c1bb86cdSEric Blake 9885cec2870SKevin Wolf rs->fd = raw_reconfigure_getfd(state->bs, state->flags, &rs->open_flags, 98923dece19SKevin Wolf state->perm, true, &local_err); 9905cec2870SKevin Wolf if (local_err) { 9915cec2870SKevin Wolf error_propagate(errp, local_err); 992c1bb86cdSEric Blake ret = -1; 9935cec2870SKevin Wolf goto out; 994c1bb86cdSEric Blake } 995c1bb86cdSEric Blake 996c1bb86cdSEric Blake /* Fail already reopen_prepare() if we can't get a working O_DIRECT 997c1bb86cdSEric Blake * alignment with the new fd. */ 998c1bb86cdSEric Blake if (rs->fd != -1) { 999c1bb86cdSEric Blake raw_probe_alignment(state->bs, rs->fd, &local_err); 1000c1bb86cdSEric Blake if (local_err) { 1001c1bb86cdSEric Blake error_propagate(errp, local_err); 1002c1bb86cdSEric Blake ret = -EINVAL; 1003a6aeca0cSKevin Wolf goto out_fd; 1004a6aeca0cSKevin Wolf } 1005c1bb86cdSEric Blake } 1006c1bb86cdSEric Blake 1007e0c9cf3aSKevin Wolf s->reopen_state = state; 1008a6aeca0cSKevin Wolf ret = 0; 1009a6aeca0cSKevin Wolf out_fd: 1010a6aeca0cSKevin Wolf if (ret < 0) { 1011a6aeca0cSKevin Wolf qemu_close(rs->fd); 1012a6aeca0cSKevin Wolf rs->fd = -1; 1013a6aeca0cSKevin Wolf } 101431be8a2aSStefan Hajnoczi out: 101531be8a2aSStefan Hajnoczi qemu_opts_del(opts); 1016c1bb86cdSEric Blake return ret; 1017c1bb86cdSEric Blake } 1018c1bb86cdSEric Blake 1019c1bb86cdSEric Blake static void raw_reopen_commit(BDRVReopenState *state) 1020c1bb86cdSEric Blake { 1021c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1022c1bb86cdSEric Blake BDRVRawState *s = state->bs->opaque; 1023c1bb86cdSEric Blake 1024f357fcd8SStefan Hajnoczi s->drop_cache = rs->drop_cache; 102531be8a2aSStefan Hajnoczi s->check_cache_dropped = rs->check_cache_dropped; 1026c1bb86cdSEric Blake s->open_flags = rs->open_flags; 1027c1bb86cdSEric Blake 1028c1bb86cdSEric Blake qemu_close(s->fd); 1029c1bb86cdSEric Blake s->fd = rs->fd; 1030c1bb86cdSEric Blake 1031c1bb86cdSEric Blake g_free(state->opaque); 1032c1bb86cdSEric Blake state->opaque = NULL; 1033e0c9cf3aSKevin Wolf 1034e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1035e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1036c1bb86cdSEric Blake } 1037c1bb86cdSEric Blake 1038c1bb86cdSEric Blake 1039c1bb86cdSEric Blake static void raw_reopen_abort(BDRVReopenState *state) 1040c1bb86cdSEric Blake { 1041c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1042e0c9cf3aSKevin Wolf BDRVRawState *s = state->bs->opaque; 1043c1bb86cdSEric Blake 1044c1bb86cdSEric Blake /* nothing to do if NULL, we didn't get far enough */ 1045c1bb86cdSEric Blake if (rs == NULL) { 1046c1bb86cdSEric Blake return; 1047c1bb86cdSEric Blake } 1048c1bb86cdSEric Blake 1049c1bb86cdSEric Blake if (rs->fd >= 0) { 1050c1bb86cdSEric Blake qemu_close(rs->fd); 1051c1bb86cdSEric Blake rs->fd = -1; 1052c1bb86cdSEric Blake } 1053c1bb86cdSEric Blake g_free(state->opaque); 1054c1bb86cdSEric Blake state->opaque = NULL; 1055e0c9cf3aSKevin Wolf 1056e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1057e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1058c1bb86cdSEric Blake } 1059c1bb86cdSEric Blake 1060867eccfeSMaxim Levitsky static int sg_get_max_transfer_length(int fd) 1061c1bb86cdSEric Blake { 1062c1bb86cdSEric Blake #ifdef BLKSECTGET 106348265250SEric Farman int max_bytes = 0; 1064867eccfeSMaxim Levitsky 1065867eccfeSMaxim Levitsky if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) { 106648265250SEric Farman return max_bytes; 1067c1bb86cdSEric Blake } else { 1068c1bb86cdSEric Blake return -errno; 1069c1bb86cdSEric Blake } 1070c1bb86cdSEric Blake #else 1071c1bb86cdSEric Blake return -ENOSYS; 1072c1bb86cdSEric Blake #endif 1073c1bb86cdSEric Blake } 1074c1bb86cdSEric Blake 1075867eccfeSMaxim Levitsky static int sg_get_max_segments(int fd) 10769103f1ceSFam Zheng { 10779103f1ceSFam Zheng #ifdef CONFIG_LINUX 10789103f1ceSFam Zheng char buf[32]; 10799103f1ceSFam Zheng const char *end; 1080867eccfeSMaxim Levitsky char *sysfspath = NULL; 10819103f1ceSFam Zheng int ret; 1082867eccfeSMaxim Levitsky int sysfd = -1; 10839103f1ceSFam Zheng long max_segments; 1084867eccfeSMaxim Levitsky struct stat st; 1085867eccfeSMaxim Levitsky 1086867eccfeSMaxim Levitsky if (fstat(fd, &st)) { 1087867eccfeSMaxim Levitsky ret = -errno; 1088867eccfeSMaxim Levitsky goto out; 1089867eccfeSMaxim Levitsky } 10909103f1ceSFam Zheng 10919103f1ceSFam Zheng sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", 1092867eccfeSMaxim Levitsky major(st.st_rdev), minor(st.st_rdev)); 1093867eccfeSMaxim Levitsky sysfd = open(sysfspath, O_RDONLY); 1094867eccfeSMaxim Levitsky if (sysfd == -1) { 10959103f1ceSFam Zheng ret = -errno; 10969103f1ceSFam Zheng goto out; 10979103f1ceSFam Zheng } 10989103f1ceSFam Zheng do { 1099867eccfeSMaxim Levitsky ret = read(sysfd, buf, sizeof(buf) - 1); 11009103f1ceSFam Zheng } while (ret == -1 && errno == EINTR); 11019103f1ceSFam Zheng if (ret < 0) { 11029103f1ceSFam Zheng ret = -errno; 11039103f1ceSFam Zheng goto out; 11049103f1ceSFam Zheng } else if (ret == 0) { 11059103f1ceSFam Zheng ret = -EIO; 11069103f1ceSFam Zheng goto out; 11079103f1ceSFam Zheng } 11089103f1ceSFam Zheng buf[ret] = 0; 11099103f1ceSFam Zheng /* The file is ended with '\n', pass 'end' to accept that. */ 11109103f1ceSFam Zheng ret = qemu_strtol(buf, &end, 10, &max_segments); 11119103f1ceSFam Zheng if (ret == 0 && end && *end == '\n') { 11129103f1ceSFam Zheng ret = max_segments; 11139103f1ceSFam Zheng } 11149103f1ceSFam Zheng 11159103f1ceSFam Zheng out: 1116867eccfeSMaxim Levitsky if (sysfd != -1) { 1117867eccfeSMaxim Levitsky close(sysfd); 1118fed414dfSFam Zheng } 11199103f1ceSFam Zheng g_free(sysfspath); 11209103f1ceSFam Zheng return ret; 11219103f1ceSFam Zheng #else 11229103f1ceSFam Zheng return -ENOTSUP; 11239103f1ceSFam Zheng #endif 11249103f1ceSFam Zheng } 11259103f1ceSFam Zheng 1126c1bb86cdSEric Blake static void raw_refresh_limits(BlockDriverState *bs, Error **errp) 1127c1bb86cdSEric Blake { 1128c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1129c1bb86cdSEric Blake 1130867eccfeSMaxim Levitsky if (bs->sg) { 1131867eccfeSMaxim Levitsky int ret = sg_get_max_transfer_length(s->fd); 1132867eccfeSMaxim Levitsky 113348265250SEric Farman if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { 113448265250SEric Farman bs->bl.max_transfer = pow2floor(ret); 1135c1bb86cdSEric Blake } 1136867eccfeSMaxim Levitsky 1137867eccfeSMaxim Levitsky ret = sg_get_max_segments(s->fd); 11389103f1ceSFam Zheng if (ret > 0) { 1139*038adc2fSWei Yang bs->bl.max_transfer = MIN(bs->bl.max_transfer, 1140*038adc2fSWei Yang ret * qemu_real_host_page_size); 1141c1bb86cdSEric Blake } 1142c1bb86cdSEric Blake } 1143c1bb86cdSEric Blake 1144c1bb86cdSEric Blake raw_probe_alignment(bs, s->fd, errp); 1145c1bb86cdSEric Blake bs->bl.min_mem_alignment = s->buf_align; 1146*038adc2fSWei Yang bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size); 1147c1bb86cdSEric Blake } 1148c1bb86cdSEric Blake 1149c1bb86cdSEric Blake static int check_for_dasd(int fd) 1150c1bb86cdSEric Blake { 1151c1bb86cdSEric Blake #ifdef BIODASDINFO2 1152c1bb86cdSEric Blake struct dasd_information2_t info = {0}; 1153c1bb86cdSEric Blake 1154c1bb86cdSEric Blake return ioctl(fd, BIODASDINFO2, &info); 1155c1bb86cdSEric Blake #else 1156c1bb86cdSEric Blake return -1; 1157c1bb86cdSEric Blake #endif 1158c1bb86cdSEric Blake } 1159c1bb86cdSEric Blake 1160c1bb86cdSEric Blake /** 1161c1bb86cdSEric Blake * Try to get @bs's logical and physical block size. 1162c1bb86cdSEric Blake * On success, store them in @bsz and return zero. 1163c1bb86cdSEric Blake * On failure, return negative errno. 1164c1bb86cdSEric Blake */ 1165c1bb86cdSEric Blake static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) 1166c1bb86cdSEric Blake { 1167c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1168c1bb86cdSEric Blake int ret; 1169c1bb86cdSEric Blake 1170c1bb86cdSEric Blake /* If DASD, get blocksizes */ 1171c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1172c1bb86cdSEric Blake return -ENOTSUP; 1173c1bb86cdSEric Blake } 1174c1bb86cdSEric Blake ret = probe_logical_blocksize(s->fd, &bsz->log); 1175c1bb86cdSEric Blake if (ret < 0) { 1176c1bb86cdSEric Blake return ret; 1177c1bb86cdSEric Blake } 1178c1bb86cdSEric Blake return probe_physical_blocksize(s->fd, &bsz->phys); 1179c1bb86cdSEric Blake } 1180c1bb86cdSEric Blake 1181c1bb86cdSEric Blake /** 1182c1bb86cdSEric Blake * Try to get @bs's geometry: cyls, heads, sectors. 1183c1bb86cdSEric Blake * On success, store them in @geo and return 0. 1184c1bb86cdSEric Blake * On failure return -errno. 1185c1bb86cdSEric Blake * (Allows block driver to assign default geometry values that guest sees) 1186c1bb86cdSEric Blake */ 1187c1bb86cdSEric Blake #ifdef __linux__ 1188c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1189c1bb86cdSEric Blake { 1190c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1191c1bb86cdSEric Blake struct hd_geometry ioctl_geo = {0}; 1192c1bb86cdSEric Blake 1193c1bb86cdSEric Blake /* If DASD, get its geometry */ 1194c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1195c1bb86cdSEric Blake return -ENOTSUP; 1196c1bb86cdSEric Blake } 1197c1bb86cdSEric Blake if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { 1198c1bb86cdSEric Blake return -errno; 1199c1bb86cdSEric Blake } 1200c1bb86cdSEric Blake /* HDIO_GETGEO may return success even though geo contains zeros 1201c1bb86cdSEric Blake (e.g. certain multipath setups) */ 1202c1bb86cdSEric Blake if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { 1203c1bb86cdSEric Blake return -ENOTSUP; 1204c1bb86cdSEric Blake } 1205c1bb86cdSEric Blake /* Do not return a geometry for partition */ 1206c1bb86cdSEric Blake if (ioctl_geo.start != 0) { 1207c1bb86cdSEric Blake return -ENOTSUP; 1208c1bb86cdSEric Blake } 1209c1bb86cdSEric Blake geo->heads = ioctl_geo.heads; 1210c1bb86cdSEric Blake geo->sectors = ioctl_geo.sectors; 1211c1bb86cdSEric Blake geo->cylinders = ioctl_geo.cylinders; 1212c1bb86cdSEric Blake 1213c1bb86cdSEric Blake return 0; 1214c1bb86cdSEric Blake } 1215c1bb86cdSEric Blake #else /* __linux__ */ 1216c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1217c1bb86cdSEric Blake { 1218c1bb86cdSEric Blake return -ENOTSUP; 1219c1bb86cdSEric Blake } 1220c1bb86cdSEric Blake #endif 1221c1bb86cdSEric Blake 122203425671SKevin Wolf #if defined(__linux__) 122303425671SKevin Wolf static int handle_aiocb_ioctl(void *opaque) 1224c1bb86cdSEric Blake { 122503425671SKevin Wolf RawPosixAIOData *aiocb = opaque; 1226c1bb86cdSEric Blake int ret; 1227c1bb86cdSEric Blake 1228d57c44d0SKevin Wolf ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); 1229c1bb86cdSEric Blake if (ret == -1) { 1230c1bb86cdSEric Blake return -errno; 1231c1bb86cdSEric Blake } 1232c1bb86cdSEric Blake 1233c1bb86cdSEric Blake return 0; 1234c1bb86cdSEric Blake } 123503425671SKevin Wolf #endif /* linux */ 1236c1bb86cdSEric Blake 123706dc9bd5SKevin Wolf static int handle_aiocb_flush(void *opaque) 1238c1bb86cdSEric Blake { 123906dc9bd5SKevin Wolf RawPosixAIOData *aiocb = opaque; 1240e5bcf967SKevin Wolf BDRVRawState *s = aiocb->bs->opaque; 1241c1bb86cdSEric Blake int ret; 1242c1bb86cdSEric Blake 1243e5bcf967SKevin Wolf if (s->page_cache_inconsistent) { 1244e5bcf967SKevin Wolf return -EIO; 1245e5bcf967SKevin Wolf } 1246e5bcf967SKevin Wolf 1247c1bb86cdSEric Blake ret = qemu_fdatasync(aiocb->aio_fildes); 1248c1bb86cdSEric Blake if (ret == -1) { 1249e5bcf967SKevin Wolf /* There is no clear definition of the semantics of a failing fsync(), 1250e5bcf967SKevin Wolf * so we may have to assume the worst. The sad truth is that this 1251e5bcf967SKevin Wolf * assumption is correct for Linux. Some pages are now probably marked 1252e5bcf967SKevin Wolf * clean in the page cache even though they are inconsistent with the 1253e5bcf967SKevin Wolf * on-disk contents. The next fdatasync() call would succeed, but no 1254e5bcf967SKevin Wolf * further writeback attempt will be made. We can't get back to a state 1255e5bcf967SKevin Wolf * in which we know what is on disk (we would have to rewrite 1256e5bcf967SKevin Wolf * everything that was touched since the last fdatasync() at least), so 1257e5bcf967SKevin Wolf * make bdrv_flush() fail permanently. Given that the behaviour isn't 1258e5bcf967SKevin Wolf * really defined, I have little hope that other OSes are doing better. 1259e5bcf967SKevin Wolf * 1260e5bcf967SKevin Wolf * Obviously, this doesn't affect O_DIRECT, which bypasses the page 1261e5bcf967SKevin Wolf * cache. */ 1262e5bcf967SKevin Wolf if ((s->open_flags & O_DIRECT) == 0) { 1263e5bcf967SKevin Wolf s->page_cache_inconsistent = true; 1264e5bcf967SKevin Wolf } 1265c1bb86cdSEric Blake return -errno; 1266c1bb86cdSEric Blake } 1267c1bb86cdSEric Blake return 0; 1268c1bb86cdSEric Blake } 1269c1bb86cdSEric Blake 1270c1bb86cdSEric Blake #ifdef CONFIG_PREADV 1271c1bb86cdSEric Blake 1272c1bb86cdSEric Blake static bool preadv_present = true; 1273c1bb86cdSEric Blake 1274c1bb86cdSEric Blake static ssize_t 1275c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1276c1bb86cdSEric Blake { 1277c1bb86cdSEric Blake return preadv(fd, iov, nr_iov, offset); 1278c1bb86cdSEric Blake } 1279c1bb86cdSEric Blake 1280c1bb86cdSEric Blake static ssize_t 1281c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1282c1bb86cdSEric Blake { 1283c1bb86cdSEric Blake return pwritev(fd, iov, nr_iov, offset); 1284c1bb86cdSEric Blake } 1285c1bb86cdSEric Blake 1286c1bb86cdSEric Blake #else 1287c1bb86cdSEric Blake 1288c1bb86cdSEric Blake static bool preadv_present = false; 1289c1bb86cdSEric Blake 1290c1bb86cdSEric Blake static ssize_t 1291c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1292c1bb86cdSEric Blake { 1293c1bb86cdSEric Blake return -ENOSYS; 1294c1bb86cdSEric Blake } 1295c1bb86cdSEric Blake 1296c1bb86cdSEric Blake static ssize_t 1297c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1298c1bb86cdSEric Blake { 1299c1bb86cdSEric Blake return -ENOSYS; 1300c1bb86cdSEric Blake } 1301c1bb86cdSEric Blake 1302c1bb86cdSEric Blake #endif 1303c1bb86cdSEric Blake 1304c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) 1305c1bb86cdSEric Blake { 1306c1bb86cdSEric Blake ssize_t len; 1307c1bb86cdSEric Blake 1308c1bb86cdSEric Blake do { 1309c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) 1310c1bb86cdSEric Blake len = qemu_pwritev(aiocb->aio_fildes, 1311d57c44d0SKevin Wolf aiocb->io.iov, 1312d57c44d0SKevin Wolf aiocb->io.niov, 1313c1bb86cdSEric Blake aiocb->aio_offset); 1314c1bb86cdSEric Blake else 1315c1bb86cdSEric Blake len = qemu_preadv(aiocb->aio_fildes, 1316d57c44d0SKevin Wolf aiocb->io.iov, 1317d57c44d0SKevin Wolf aiocb->io.niov, 1318c1bb86cdSEric Blake aiocb->aio_offset); 1319c1bb86cdSEric Blake } while (len == -1 && errno == EINTR); 1320c1bb86cdSEric Blake 1321c1bb86cdSEric Blake if (len == -1) { 1322c1bb86cdSEric Blake return -errno; 1323c1bb86cdSEric Blake } 1324c1bb86cdSEric Blake return len; 1325c1bb86cdSEric Blake } 1326c1bb86cdSEric Blake 1327c1bb86cdSEric Blake /* 1328c1bb86cdSEric Blake * Read/writes the data to/from a given linear buffer. 1329c1bb86cdSEric Blake * 1330c1bb86cdSEric Blake * Returns the number of bytes handles or -errno in case of an error. Short 1331c1bb86cdSEric Blake * reads are only returned if the end of the file is reached. 1332c1bb86cdSEric Blake */ 1333c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) 1334c1bb86cdSEric Blake { 1335c1bb86cdSEric Blake ssize_t offset = 0; 1336c1bb86cdSEric Blake ssize_t len; 1337c1bb86cdSEric Blake 1338c1bb86cdSEric Blake while (offset < aiocb->aio_nbytes) { 1339c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1340c1bb86cdSEric Blake len = pwrite(aiocb->aio_fildes, 1341c1bb86cdSEric Blake (const char *)buf + offset, 1342c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1343c1bb86cdSEric Blake aiocb->aio_offset + offset); 1344c1bb86cdSEric Blake } else { 1345c1bb86cdSEric Blake len = pread(aiocb->aio_fildes, 1346c1bb86cdSEric Blake buf + offset, 1347c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1348c1bb86cdSEric Blake aiocb->aio_offset + offset); 1349c1bb86cdSEric Blake } 1350c1bb86cdSEric Blake if (len == -1 && errno == EINTR) { 1351c1bb86cdSEric Blake continue; 1352c1bb86cdSEric Blake } else if (len == -1 && errno == EINVAL && 1353c1bb86cdSEric Blake (aiocb->bs->open_flags & BDRV_O_NOCACHE) && 1354c1bb86cdSEric Blake !(aiocb->aio_type & QEMU_AIO_WRITE) && 1355c1bb86cdSEric Blake offset > 0) { 1356c1bb86cdSEric Blake /* O_DIRECT pread() may fail with EINVAL when offset is unaligned 1357c1bb86cdSEric Blake * after a short read. Assume that O_DIRECT short reads only occur 1358c1bb86cdSEric Blake * at EOF. Therefore this is a short read, not an I/O error. 1359c1bb86cdSEric Blake */ 1360c1bb86cdSEric Blake break; 1361c1bb86cdSEric Blake } else if (len == -1) { 1362c1bb86cdSEric Blake offset = -errno; 1363c1bb86cdSEric Blake break; 1364c1bb86cdSEric Blake } else if (len == 0) { 1365c1bb86cdSEric Blake break; 1366c1bb86cdSEric Blake } 1367c1bb86cdSEric Blake offset += len; 1368c1bb86cdSEric Blake } 1369c1bb86cdSEric Blake 1370c1bb86cdSEric Blake return offset; 1371c1bb86cdSEric Blake } 1372c1bb86cdSEric Blake 1373999e6b69SKevin Wolf static int handle_aiocb_rw(void *opaque) 1374c1bb86cdSEric Blake { 1375999e6b69SKevin Wolf RawPosixAIOData *aiocb = opaque; 1376c1bb86cdSEric Blake ssize_t nbytes; 1377c1bb86cdSEric Blake char *buf; 1378c1bb86cdSEric Blake 1379c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { 1380c1bb86cdSEric Blake /* 1381c1bb86cdSEric Blake * If there is just a single buffer, and it is properly aligned 1382c1bb86cdSEric Blake * we can just use plain pread/pwrite without any problems. 1383c1bb86cdSEric Blake */ 1384d57c44d0SKevin Wolf if (aiocb->io.niov == 1) { 138554c7ca1bSKevin Wolf nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); 138654c7ca1bSKevin Wolf goto out; 1387c1bb86cdSEric Blake } 1388c1bb86cdSEric Blake /* 1389c1bb86cdSEric Blake * We have more than one iovec, and all are properly aligned. 1390c1bb86cdSEric Blake * 1391c1bb86cdSEric Blake * Try preadv/pwritev first and fall back to linearizing the 1392c1bb86cdSEric Blake * buffer if it's not supported. 1393c1bb86cdSEric Blake */ 1394c1bb86cdSEric Blake if (preadv_present) { 1395c1bb86cdSEric Blake nbytes = handle_aiocb_rw_vector(aiocb); 1396c1bb86cdSEric Blake if (nbytes == aiocb->aio_nbytes || 1397c1bb86cdSEric Blake (nbytes < 0 && nbytes != -ENOSYS)) { 139854c7ca1bSKevin Wolf goto out; 1399c1bb86cdSEric Blake } 1400c1bb86cdSEric Blake preadv_present = false; 1401c1bb86cdSEric Blake } 1402c1bb86cdSEric Blake 1403c1bb86cdSEric Blake /* 1404c1bb86cdSEric Blake * XXX(hch): short read/write. no easy way to handle the reminder 1405c1bb86cdSEric Blake * using these interfaces. For now retry using plain 1406c1bb86cdSEric Blake * pread/pwrite? 1407c1bb86cdSEric Blake */ 1408c1bb86cdSEric Blake } 1409c1bb86cdSEric Blake 1410c1bb86cdSEric Blake /* 1411c1bb86cdSEric Blake * Ok, we have to do it the hard way, copy all segments into 1412c1bb86cdSEric Blake * a single aligned buffer. 1413c1bb86cdSEric Blake */ 1414c1bb86cdSEric Blake buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); 1415c1bb86cdSEric Blake if (buf == NULL) { 141654c7ca1bSKevin Wolf nbytes = -ENOMEM; 141754c7ca1bSKevin Wolf goto out; 1418c1bb86cdSEric Blake } 1419c1bb86cdSEric Blake 1420c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1421c1bb86cdSEric Blake char *p = buf; 1422c1bb86cdSEric Blake int i; 1423c1bb86cdSEric Blake 1424d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov; ++i) { 1425d57c44d0SKevin Wolf memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); 1426d57c44d0SKevin Wolf p += aiocb->io.iov[i].iov_len; 1427c1bb86cdSEric Blake } 1428c1bb86cdSEric Blake assert(p - buf == aiocb->aio_nbytes); 1429c1bb86cdSEric Blake } 1430c1bb86cdSEric Blake 1431c1bb86cdSEric Blake nbytes = handle_aiocb_rw_linear(aiocb, buf); 1432c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { 1433c1bb86cdSEric Blake char *p = buf; 1434c1bb86cdSEric Blake size_t count = aiocb->aio_nbytes, copy; 1435c1bb86cdSEric Blake int i; 1436c1bb86cdSEric Blake 1437d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov && count; ++i) { 1438c1bb86cdSEric Blake copy = count; 1439d57c44d0SKevin Wolf if (copy > aiocb->io.iov[i].iov_len) { 1440d57c44d0SKevin Wolf copy = aiocb->io.iov[i].iov_len; 1441c1bb86cdSEric Blake } 1442d57c44d0SKevin Wolf memcpy(aiocb->io.iov[i].iov_base, p, copy); 1443c1bb86cdSEric Blake assert(count >= copy); 1444c1bb86cdSEric Blake p += copy; 1445c1bb86cdSEric Blake count -= copy; 1446c1bb86cdSEric Blake } 1447c1bb86cdSEric Blake assert(count == 0); 1448c1bb86cdSEric Blake } 1449c1bb86cdSEric Blake qemu_vfree(buf); 1450c1bb86cdSEric Blake 145154c7ca1bSKevin Wolf out: 145254c7ca1bSKevin Wolf if (nbytes == aiocb->aio_nbytes) { 145354c7ca1bSKevin Wolf return 0; 145454c7ca1bSKevin Wolf } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { 145554c7ca1bSKevin Wolf if (aiocb->aio_type & QEMU_AIO_WRITE) { 145654c7ca1bSKevin Wolf return -EINVAL; 145754c7ca1bSKevin Wolf } else { 145854c7ca1bSKevin Wolf iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, 145954c7ca1bSKevin Wolf 0, aiocb->aio_nbytes - nbytes); 146054c7ca1bSKevin Wolf return 0; 146154c7ca1bSKevin Wolf } 146254c7ca1bSKevin Wolf } else { 146354c7ca1bSKevin Wolf assert(nbytes < 0); 1464c1bb86cdSEric Blake return nbytes; 1465c1bb86cdSEric Blake } 146654c7ca1bSKevin Wolf } 1467c1bb86cdSEric Blake 1468c1bb86cdSEric Blake static int translate_err(int err) 1469c1bb86cdSEric Blake { 1470c1bb86cdSEric Blake if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || 1471c1bb86cdSEric Blake err == -ENOTTY) { 1472c1bb86cdSEric Blake err = -ENOTSUP; 1473c1bb86cdSEric Blake } 1474c1bb86cdSEric Blake return err; 1475c1bb86cdSEric Blake } 1476c1bb86cdSEric Blake 1477c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 1478c1bb86cdSEric Blake static int do_fallocate(int fd, int mode, off_t offset, off_t len) 1479c1bb86cdSEric Blake { 1480c1bb86cdSEric Blake do { 1481c1bb86cdSEric Blake if (fallocate(fd, mode, offset, len) == 0) { 1482c1bb86cdSEric Blake return 0; 1483c1bb86cdSEric Blake } 1484c1bb86cdSEric Blake } while (errno == EINTR); 1485c1bb86cdSEric Blake return translate_err(-errno); 1486c1bb86cdSEric Blake } 1487c1bb86cdSEric Blake #endif 1488c1bb86cdSEric Blake 1489c1bb86cdSEric Blake static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) 1490c1bb86cdSEric Blake { 1491c1bb86cdSEric Blake int ret = -ENOTSUP; 1492c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1493c1bb86cdSEric Blake 1494c1bb86cdSEric Blake if (!s->has_write_zeroes) { 1495c1bb86cdSEric Blake return -ENOTSUP; 1496c1bb86cdSEric Blake } 1497c1bb86cdSEric Blake 1498c1bb86cdSEric Blake #ifdef BLKZEROOUT 1499738301e1SKevin Wolf /* The BLKZEROOUT implementation in the kernel doesn't set 1500738301e1SKevin Wolf * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow 1501738301e1SKevin Wolf * fallbacks. */ 1502738301e1SKevin Wolf if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { 1503c1bb86cdSEric Blake do { 1504c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1505c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { 1506c1bb86cdSEric Blake return 0; 1507c1bb86cdSEric Blake } 1508c1bb86cdSEric Blake } while (errno == EINTR); 1509c1bb86cdSEric Blake 1510c1bb86cdSEric Blake ret = translate_err(-errno); 1511c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1512c1bb86cdSEric Blake s->has_write_zeroes = false; 1513c1bb86cdSEric Blake } 1514effecce6SKevin Wolf } 1515effecce6SKevin Wolf #endif 1516effecce6SKevin Wolf 1517c1bb86cdSEric Blake return ret; 1518c1bb86cdSEric Blake } 1519c1bb86cdSEric Blake 15207154d8aeSKevin Wolf static int handle_aiocb_write_zeroes(void *opaque) 1521c1bb86cdSEric Blake { 15227154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 152370d9110bSDenis V. Lunev #ifdef CONFIG_FALLOCATE 1524b2c6f23fSMax Reitz BDRVRawState *s = aiocb->bs->opaque; 152570d9110bSDenis V. Lunev int64_t len; 152670d9110bSDenis V. Lunev #endif 1527c1bb86cdSEric Blake 1528c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1529c1bb86cdSEric Blake return handle_aiocb_write_zeroes_block(aiocb); 1530c1bb86cdSEric Blake } 1531c1bb86cdSEric Blake 1532c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_ZERO_RANGE 1533c1bb86cdSEric Blake if (s->has_write_zeroes) { 1534c1bb86cdSEric Blake int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, 1535c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1536294682ccSAndrey Shinkevich if (ret == -EINVAL) { 1537294682ccSAndrey Shinkevich /* 1538294682ccSAndrey Shinkevich * Allow falling back to pwrite for file systems that 1539294682ccSAndrey Shinkevich * do not support fallocate() for an unaligned byte range. 1540294682ccSAndrey Shinkevich */ 1541294682ccSAndrey Shinkevich return -ENOTSUP; 1542294682ccSAndrey Shinkevich } 1543c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1544c1bb86cdSEric Blake return ret; 1545c1bb86cdSEric Blake } 1546c1bb86cdSEric Blake s->has_write_zeroes = false; 1547c1bb86cdSEric Blake } 1548c1bb86cdSEric Blake #endif 1549c1bb86cdSEric Blake 1550c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1551c1bb86cdSEric Blake if (s->has_discard && s->has_fallocate) { 1552c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 1553c1bb86cdSEric Blake FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1554c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1555c1bb86cdSEric Blake if (ret == 0) { 1556c1bb86cdSEric Blake ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1557c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1558c1bb86cdSEric Blake return ret; 1559c1bb86cdSEric Blake } 1560c1bb86cdSEric Blake s->has_fallocate = false; 1561c1bb86cdSEric Blake } else if (ret != -ENOTSUP) { 1562c1bb86cdSEric Blake return ret; 1563c1bb86cdSEric Blake } else { 1564c1bb86cdSEric Blake s->has_discard = false; 1565c1bb86cdSEric Blake } 1566c1bb86cdSEric Blake } 1567c1bb86cdSEric Blake #endif 1568c1bb86cdSEric Blake 1569c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 157070d9110bSDenis V. Lunev /* Last resort: we are trying to extend the file with zeroed data. This 157170d9110bSDenis V. Lunev * can be done via fallocate(fd, 0) */ 157270d9110bSDenis V. Lunev len = bdrv_getlength(aiocb->bs); 157370d9110bSDenis V. Lunev if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { 1574c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1575c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1576c1bb86cdSEric Blake return ret; 1577c1bb86cdSEric Blake } 1578c1bb86cdSEric Blake s->has_fallocate = false; 1579c1bb86cdSEric Blake } 1580c1bb86cdSEric Blake #endif 1581c1bb86cdSEric Blake 1582c1bb86cdSEric Blake return -ENOTSUP; 1583c1bb86cdSEric Blake } 1584c1bb86cdSEric Blake 15857154d8aeSKevin Wolf static int handle_aiocb_write_zeroes_unmap(void *opaque) 158634fa110eSKevin Wolf { 15877154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 158834fa110eSKevin Wolf BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; 158934fa110eSKevin Wolf int ret; 159034fa110eSKevin Wolf 159134fa110eSKevin Wolf /* First try to write zeros and unmap at the same time */ 159234fa110eSKevin Wolf 159334fa110eSKevin Wolf #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 159434fa110eSKevin Wolf ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 159534fa110eSKevin Wolf aiocb->aio_offset, aiocb->aio_nbytes); 159634fa110eSKevin Wolf if (ret != -ENOTSUP) { 159734fa110eSKevin Wolf return ret; 159834fa110eSKevin Wolf } 159934fa110eSKevin Wolf #endif 160034fa110eSKevin Wolf 160134fa110eSKevin Wolf /* If we couldn't manage to unmap while guaranteed that the area reads as 160234fa110eSKevin Wolf * all-zero afterwards, just write zeroes without unmapping */ 160334fa110eSKevin Wolf ret = handle_aiocb_write_zeroes(aiocb); 160434fa110eSKevin Wolf return ret; 160534fa110eSKevin Wolf } 160634fa110eSKevin Wolf 16071efad060SFam Zheng #ifndef HAVE_COPY_FILE_RANGE 16081efad060SFam Zheng static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, 16091efad060SFam Zheng off_t *out_off, size_t len, unsigned int flags) 16101efad060SFam Zheng { 16111efad060SFam Zheng #ifdef __NR_copy_file_range 16121efad060SFam Zheng return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, 16131efad060SFam Zheng out_off, len, flags); 16141efad060SFam Zheng #else 16151efad060SFam Zheng errno = ENOSYS; 16161efad060SFam Zheng return -1; 16171efad060SFam Zheng #endif 16181efad060SFam Zheng } 16191efad060SFam Zheng #endif 16201efad060SFam Zheng 162158a209c4SKevin Wolf static int handle_aiocb_copy_range(void *opaque) 16221efad060SFam Zheng { 162358a209c4SKevin Wolf RawPosixAIOData *aiocb = opaque; 16241efad060SFam Zheng uint64_t bytes = aiocb->aio_nbytes; 16251efad060SFam Zheng off_t in_off = aiocb->aio_offset; 1626d57c44d0SKevin Wolf off_t out_off = aiocb->copy_range.aio_offset2; 16271efad060SFam Zheng 16281efad060SFam Zheng while (bytes) { 16291efad060SFam Zheng ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, 1630d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, &out_off, 16311efad060SFam Zheng bytes, 0); 1632ecc983a5SFam Zheng trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, 1633d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, out_off, bytes, 1634d57c44d0SKevin Wolf 0, ret); 1635c436e3d0SFam Zheng if (ret == 0) { 1636c436e3d0SFam Zheng /* No progress (e.g. when beyond EOF), let the caller fall back to 1637c436e3d0SFam Zheng * buffer I/O. */ 1638c436e3d0SFam Zheng return -ENOSPC; 16391efad060SFam Zheng } 16401efad060SFam Zheng if (ret < 0) { 1641c436e3d0SFam Zheng switch (errno) { 1642c436e3d0SFam Zheng case ENOSYS: 16431efad060SFam Zheng return -ENOTSUP; 1644c436e3d0SFam Zheng case EINTR: 1645c436e3d0SFam Zheng continue; 1646c436e3d0SFam Zheng default: 16471efad060SFam Zheng return -errno; 16481efad060SFam Zheng } 16491efad060SFam Zheng } 16501efad060SFam Zheng bytes -= ret; 16511efad060SFam Zheng } 16521efad060SFam Zheng return 0; 16531efad060SFam Zheng } 16541efad060SFam Zheng 165546ee0f46SKevin Wolf static int handle_aiocb_discard(void *opaque) 1656c1bb86cdSEric Blake { 165746ee0f46SKevin Wolf RawPosixAIOData *aiocb = opaque; 1658c1bb86cdSEric Blake int ret = -EOPNOTSUPP; 1659c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1660c1bb86cdSEric Blake 1661c1bb86cdSEric Blake if (!s->has_discard) { 1662c1bb86cdSEric Blake return -ENOTSUP; 1663c1bb86cdSEric Blake } 1664c1bb86cdSEric Blake 1665c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1666c1bb86cdSEric Blake #ifdef BLKDISCARD 1667c1bb86cdSEric Blake do { 1668c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1669c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { 1670c1bb86cdSEric Blake return 0; 1671c1bb86cdSEric Blake } 1672c1bb86cdSEric Blake } while (errno == EINTR); 1673c1bb86cdSEric Blake 1674c1bb86cdSEric Blake ret = -errno; 1675c1bb86cdSEric Blake #endif 1676c1bb86cdSEric Blake } else { 1677c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1678c1bb86cdSEric Blake ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1679c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1680c1bb86cdSEric Blake #endif 1681c1bb86cdSEric Blake } 1682c1bb86cdSEric Blake 1683c1bb86cdSEric Blake ret = translate_err(ret); 1684c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1685c1bb86cdSEric Blake s->has_discard = false; 1686c1bb86cdSEric Blake } 1687c1bb86cdSEric Blake return ret; 1688c1bb86cdSEric Blake } 1689c1bb86cdSEric Blake 16903a20013fSNir Soffer /* 16913a20013fSNir Soffer * Help alignment probing by allocating the first block. 16923a20013fSNir Soffer * 16933a20013fSNir Soffer * When reading with direct I/O from unallocated area on Gluster backed by XFS, 16943a20013fSNir Soffer * reading succeeds regardless of request length. In this case we fallback to 16953a20013fSNir Soffer * safe alignment which is not optimal. Allocating the first block avoids this 16963a20013fSNir Soffer * fallback. 16973a20013fSNir Soffer * 16983a20013fSNir Soffer * fd may be opened with O_DIRECT, but we don't know the buffer alignment or 16993a20013fSNir Soffer * request alignment, so we use safe values. 17003a20013fSNir Soffer * 17013a20013fSNir Soffer * Returns: 0 on success, -errno on failure. Since this is an optimization, 17023a20013fSNir Soffer * caller may ignore failures. 17033a20013fSNir Soffer */ 17043a20013fSNir Soffer static int allocate_first_block(int fd, size_t max_size) 17053a20013fSNir Soffer { 17063a20013fSNir Soffer size_t write_size = (max_size < MAX_BLOCKSIZE) 17073a20013fSNir Soffer ? BDRV_SECTOR_SIZE 17083a20013fSNir Soffer : MAX_BLOCKSIZE; 1709*038adc2fSWei Yang size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); 17103a20013fSNir Soffer void *buf; 17113a20013fSNir Soffer ssize_t n; 17123a20013fSNir Soffer int ret; 17133a20013fSNir Soffer 17143a20013fSNir Soffer buf = qemu_memalign(max_align, write_size); 17153a20013fSNir Soffer memset(buf, 0, write_size); 17163a20013fSNir Soffer 17173a20013fSNir Soffer do { 17183a20013fSNir Soffer n = pwrite(fd, buf, write_size, 0); 17193a20013fSNir Soffer } while (n == -1 && errno == EINTR); 17203a20013fSNir Soffer 17213a20013fSNir Soffer ret = (n == -1) ? -errno : 0; 17223a20013fSNir Soffer 17233a20013fSNir Soffer qemu_vfree(buf); 17243a20013fSNir Soffer return ret; 17253a20013fSNir Soffer } 17263a20013fSNir Soffer 172729cb4c01SKevin Wolf static int handle_aiocb_truncate(void *opaque) 172893f4e2ffSKevin Wolf { 172929cb4c01SKevin Wolf RawPosixAIOData *aiocb = opaque; 173093f4e2ffSKevin Wolf int result = 0; 173193f4e2ffSKevin Wolf int64_t current_length = 0; 173293f4e2ffSKevin Wolf char *buf = NULL; 173393f4e2ffSKevin Wolf struct stat st; 173493f4e2ffSKevin Wolf int fd = aiocb->aio_fildes; 173593f4e2ffSKevin Wolf int64_t offset = aiocb->aio_offset; 1736d57c44d0SKevin Wolf PreallocMode prealloc = aiocb->truncate.prealloc; 1737d57c44d0SKevin Wolf Error **errp = aiocb->truncate.errp; 173893f4e2ffSKevin Wolf 173993f4e2ffSKevin Wolf if (fstat(fd, &st) < 0) { 174093f4e2ffSKevin Wolf result = -errno; 174193f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not stat file"); 174293f4e2ffSKevin Wolf return result; 174393f4e2ffSKevin Wolf } 174493f4e2ffSKevin Wolf 174593f4e2ffSKevin Wolf current_length = st.st_size; 1746d57c44d0SKevin Wolf if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { 174793f4e2ffSKevin Wolf error_setg(errp, "Cannot use preallocation for shrinking files"); 174893f4e2ffSKevin Wolf return -ENOTSUP; 174993f4e2ffSKevin Wolf } 175093f4e2ffSKevin Wolf 1751d57c44d0SKevin Wolf switch (prealloc) { 175293f4e2ffSKevin Wolf #ifdef CONFIG_POSIX_FALLOCATE 175393f4e2ffSKevin Wolf case PREALLOC_MODE_FALLOC: 175493f4e2ffSKevin Wolf /* 175593f4e2ffSKevin Wolf * Truncating before posix_fallocate() makes it about twice slower on 175693f4e2ffSKevin Wolf * file systems that do not support fallocate(), trying to check if a 175793f4e2ffSKevin Wolf * block is allocated before allocating it, so don't do that here. 175893f4e2ffSKevin Wolf */ 175993f4e2ffSKevin Wolf if (offset != current_length) { 176093f4e2ffSKevin Wolf result = -posix_fallocate(fd, current_length, 176193f4e2ffSKevin Wolf offset - current_length); 176293f4e2ffSKevin Wolf if (result != 0) { 176393f4e2ffSKevin Wolf /* posix_fallocate() doesn't set errno. */ 176493f4e2ffSKevin Wolf error_setg_errno(errp, -result, 176593f4e2ffSKevin Wolf "Could not preallocate new data"); 17663a20013fSNir Soffer } else if (current_length == 0) { 17673a20013fSNir Soffer /* 17683a20013fSNir Soffer * posix_fallocate() uses fallocate() if the filesystem 17693a20013fSNir Soffer * supports it, or fallback to manually writing zeroes. If 17703a20013fSNir Soffer * fallocate() was used, unaligned reads from the fallocated 17713a20013fSNir Soffer * area in raw_probe_alignment() will succeed, hence we need to 17723a20013fSNir Soffer * allocate the first block. 17733a20013fSNir Soffer * 17743a20013fSNir Soffer * Optimize future alignment probing; ignore failures. 17753a20013fSNir Soffer */ 17763a20013fSNir Soffer allocate_first_block(fd, offset); 177793f4e2ffSKevin Wolf } 177893f4e2ffSKevin Wolf } else { 177993f4e2ffSKevin Wolf result = 0; 178093f4e2ffSKevin Wolf } 178193f4e2ffSKevin Wolf goto out; 178293f4e2ffSKevin Wolf #endif 178393f4e2ffSKevin Wolf case PREALLOC_MODE_FULL: 178493f4e2ffSKevin Wolf { 178593f4e2ffSKevin Wolf int64_t num = 0, left = offset - current_length; 178693f4e2ffSKevin Wolf off_t seek_result; 178793f4e2ffSKevin Wolf 178893f4e2ffSKevin Wolf /* 178993f4e2ffSKevin Wolf * Knowing the final size from the beginning could allow the file 179093f4e2ffSKevin Wolf * system driver to do less allocations and possibly avoid 179193f4e2ffSKevin Wolf * fragmentation of the file. 179293f4e2ffSKevin Wolf */ 179393f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 179493f4e2ffSKevin Wolf result = -errno; 179593f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 179693f4e2ffSKevin Wolf goto out; 179793f4e2ffSKevin Wolf } 179893f4e2ffSKevin Wolf 179993f4e2ffSKevin Wolf buf = g_malloc0(65536); 180093f4e2ffSKevin Wolf 180193f4e2ffSKevin Wolf seek_result = lseek(fd, current_length, SEEK_SET); 180293f4e2ffSKevin Wolf if (seek_result < 0) { 180393f4e2ffSKevin Wolf result = -errno; 180493f4e2ffSKevin Wolf error_setg_errno(errp, -result, 180593f4e2ffSKevin Wolf "Failed to seek to the old end of file"); 180693f4e2ffSKevin Wolf goto out; 180793f4e2ffSKevin Wolf } 180893f4e2ffSKevin Wolf 180993f4e2ffSKevin Wolf while (left > 0) { 181093f4e2ffSKevin Wolf num = MIN(left, 65536); 181193f4e2ffSKevin Wolf result = write(fd, buf, num); 181293f4e2ffSKevin Wolf if (result < 0) { 1813a1c81f4fSFam Zheng if (errno == EINTR) { 1814a1c81f4fSFam Zheng continue; 1815a1c81f4fSFam Zheng } 181693f4e2ffSKevin Wolf result = -errno; 181793f4e2ffSKevin Wolf error_setg_errno(errp, -result, 181893f4e2ffSKevin Wolf "Could not write zeros for preallocation"); 181993f4e2ffSKevin Wolf goto out; 182093f4e2ffSKevin Wolf } 182193f4e2ffSKevin Wolf left -= result; 182293f4e2ffSKevin Wolf } 182393f4e2ffSKevin Wolf if (result >= 0) { 182493f4e2ffSKevin Wolf result = fsync(fd); 182593f4e2ffSKevin Wolf if (result < 0) { 182693f4e2ffSKevin Wolf result = -errno; 182793f4e2ffSKevin Wolf error_setg_errno(errp, -result, 182893f4e2ffSKevin Wolf "Could not flush file to disk"); 182993f4e2ffSKevin Wolf goto out; 183093f4e2ffSKevin Wolf } 183193f4e2ffSKevin Wolf } 183293f4e2ffSKevin Wolf goto out; 183393f4e2ffSKevin Wolf } 183493f4e2ffSKevin Wolf case PREALLOC_MODE_OFF: 183593f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 183693f4e2ffSKevin Wolf result = -errno; 183793f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 18383a20013fSNir Soffer } else if (current_length == 0 && offset > current_length) { 18393a20013fSNir Soffer /* Optimize future alignment probing; ignore failures. */ 18403a20013fSNir Soffer allocate_first_block(fd, offset); 184193f4e2ffSKevin Wolf } 184293f4e2ffSKevin Wolf return result; 184393f4e2ffSKevin Wolf default: 184493f4e2ffSKevin Wolf result = -ENOTSUP; 184593f4e2ffSKevin Wolf error_setg(errp, "Unsupported preallocation mode: %s", 1846d57c44d0SKevin Wolf PreallocMode_str(prealloc)); 184793f4e2ffSKevin Wolf return result; 184893f4e2ffSKevin Wolf } 184993f4e2ffSKevin Wolf 185093f4e2ffSKevin Wolf out: 185193f4e2ffSKevin Wolf if (result < 0) { 185293f4e2ffSKevin Wolf if (ftruncate(fd, current_length) < 0) { 185393f4e2ffSKevin Wolf error_report("Failed to restore old file length: %s", 185493f4e2ffSKevin Wolf strerror(errno)); 185593f4e2ffSKevin Wolf } 185693f4e2ffSKevin Wolf } 185793f4e2ffSKevin Wolf 185893f4e2ffSKevin Wolf g_free(buf); 185993f4e2ffSKevin Wolf return result; 186093f4e2ffSKevin Wolf } 186193f4e2ffSKevin Wolf 18625d5de250SKevin Wolf static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs, 18635d5de250SKevin Wolf ThreadPoolFunc func, void *arg) 18645d5de250SKevin Wolf { 18655d5de250SKevin Wolf /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ 18665d5de250SKevin Wolf ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 18675d5de250SKevin Wolf return thread_pool_submit_co(pool, func, arg); 18685d5de250SKevin Wolf } 18695d5de250SKevin Wolf 1870c1bb86cdSEric Blake static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, 1871c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, int type) 1872c1bb86cdSEric Blake { 1873c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1874999e6b69SKevin Wolf RawPosixAIOData acb; 1875c1bb86cdSEric Blake 1876c1bb86cdSEric Blake if (fd_open(bs) < 0) 1877c1bb86cdSEric Blake return -EIO; 1878c1bb86cdSEric Blake 1879c1bb86cdSEric Blake /* 1880c1bb86cdSEric Blake * Check if the underlying device requires requests to be aligned, 1881c1bb86cdSEric Blake * and if the request we are trying to submit is aligned or not. 1882c1bb86cdSEric Blake * If this is the case tell the low-level driver that it needs 1883c1bb86cdSEric Blake * to copy the buffer. 1884c1bb86cdSEric Blake */ 1885c1bb86cdSEric Blake if (s->needs_alignment) { 1886c1bb86cdSEric Blake if (!bdrv_qiov_is_aligned(bs, qiov)) { 1887c1bb86cdSEric Blake type |= QEMU_AIO_MISALIGNED; 1888c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1889c1bb86cdSEric Blake } else if (s->use_linux_aio) { 1890c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1891c1bb86cdSEric Blake assert(qiov->size == bytes); 1892c1bb86cdSEric Blake return laio_co_submit(bs, aio, s->fd, offset, qiov, type); 1893c1bb86cdSEric Blake #endif 1894c1bb86cdSEric Blake } 1895c1bb86cdSEric Blake } 1896c1bb86cdSEric Blake 1897999e6b69SKevin Wolf acb = (RawPosixAIOData) { 1898999e6b69SKevin Wolf .bs = bs, 1899999e6b69SKevin Wolf .aio_fildes = s->fd, 1900999e6b69SKevin Wolf .aio_type = type, 1901999e6b69SKevin Wolf .aio_offset = offset, 1902999e6b69SKevin Wolf .aio_nbytes = bytes, 1903999e6b69SKevin Wolf .io = { 1904999e6b69SKevin Wolf .iov = qiov->iov, 1905999e6b69SKevin Wolf .niov = qiov->niov, 1906999e6b69SKevin Wolf }, 1907999e6b69SKevin Wolf }; 1908999e6b69SKevin Wolf 1909999e6b69SKevin Wolf assert(qiov->size == bytes); 1910999e6b69SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb); 1911c1bb86cdSEric Blake } 1912c1bb86cdSEric Blake 1913c1bb86cdSEric Blake static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset, 1914c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 1915c1bb86cdSEric Blake int flags) 1916c1bb86cdSEric Blake { 1917c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); 1918c1bb86cdSEric Blake } 1919c1bb86cdSEric Blake 1920c1bb86cdSEric Blake static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset, 1921c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 1922c1bb86cdSEric Blake int flags) 1923c1bb86cdSEric Blake { 1924c1bb86cdSEric Blake assert(flags == 0); 1925c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); 1926c1bb86cdSEric Blake } 1927c1bb86cdSEric Blake 1928c1bb86cdSEric Blake static void raw_aio_plug(BlockDriverState *bs) 1929c1bb86cdSEric Blake { 1930c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1931c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1932c1bb86cdSEric Blake if (s->use_linux_aio) { 1933c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1934c1bb86cdSEric Blake laio_io_plug(bs, aio); 1935c1bb86cdSEric Blake } 1936c1bb86cdSEric Blake #endif 1937c1bb86cdSEric Blake } 1938c1bb86cdSEric Blake 1939c1bb86cdSEric Blake static void raw_aio_unplug(BlockDriverState *bs) 1940c1bb86cdSEric Blake { 1941c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1942c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1943c1bb86cdSEric Blake if (s->use_linux_aio) { 1944c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1945c1bb86cdSEric Blake laio_io_unplug(bs, aio); 1946c1bb86cdSEric Blake } 1947c1bb86cdSEric Blake #endif 1948c1bb86cdSEric Blake } 1949c1bb86cdSEric Blake 195033d70fb6SKevin Wolf static int raw_co_flush_to_disk(BlockDriverState *bs) 1951c1bb86cdSEric Blake { 1952c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 195306dc9bd5SKevin Wolf RawPosixAIOData acb; 195433d70fb6SKevin Wolf int ret; 1955c1bb86cdSEric Blake 195633d70fb6SKevin Wolf ret = fd_open(bs); 195733d70fb6SKevin Wolf if (ret < 0) { 195833d70fb6SKevin Wolf return ret; 195933d70fb6SKevin Wolf } 1960c1bb86cdSEric Blake 196106dc9bd5SKevin Wolf acb = (RawPosixAIOData) { 196206dc9bd5SKevin Wolf .bs = bs, 196306dc9bd5SKevin Wolf .aio_fildes = s->fd, 196406dc9bd5SKevin Wolf .aio_type = QEMU_AIO_FLUSH, 196506dc9bd5SKevin Wolf }; 196606dc9bd5SKevin Wolf 196706dc9bd5SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb); 1968c1bb86cdSEric Blake } 1969c1bb86cdSEric Blake 1970ed6e2161SNishanth Aravamudan static void raw_aio_attach_aio_context(BlockDriverState *bs, 1971ed6e2161SNishanth Aravamudan AioContext *new_context) 1972ed6e2161SNishanth Aravamudan { 1973ed6e2161SNishanth Aravamudan #ifdef CONFIG_LINUX_AIO 1974ed6e2161SNishanth Aravamudan BDRVRawState *s = bs->opaque; 1975ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 1976ed6e2161SNishanth Aravamudan Error *local_err; 1977ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(new_context, &local_err)) { 1978ed6e2161SNishanth Aravamudan error_reportf_err(local_err, "Unable to use native AIO, " 1979ed6e2161SNishanth Aravamudan "falling back to thread pool: "); 1980ed6e2161SNishanth Aravamudan s->use_linux_aio = false; 1981ed6e2161SNishanth Aravamudan } 1982ed6e2161SNishanth Aravamudan } 1983ed6e2161SNishanth Aravamudan #endif 1984ed6e2161SNishanth Aravamudan } 1985ed6e2161SNishanth Aravamudan 1986c1bb86cdSEric Blake static void raw_close(BlockDriverState *bs) 1987c1bb86cdSEric Blake { 1988c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1989c1bb86cdSEric Blake 1990c1bb86cdSEric Blake if (s->fd >= 0) { 1991c1bb86cdSEric Blake qemu_close(s->fd); 1992c1bb86cdSEric Blake s->fd = -1; 1993c1bb86cdSEric Blake } 1994c1bb86cdSEric Blake } 1995c1bb86cdSEric Blake 1996d0bc9e5dSMax Reitz /** 1997d0bc9e5dSMax Reitz * Truncates the given regular file @fd to @offset and, when growing, fills the 1998d0bc9e5dSMax Reitz * new space according to @prealloc. 1999d0bc9e5dSMax Reitz * 2000d0bc9e5dSMax Reitz * Returns: 0 on success, -errno on failure. 2001d0bc9e5dSMax Reitz */ 200293f4e2ffSKevin Wolf static int coroutine_fn 200393f4e2ffSKevin Wolf raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, 200493f4e2ffSKevin Wolf PreallocMode prealloc, Error **errp) 20059f63b07eSMax Reitz { 200629cb4c01SKevin Wolf RawPosixAIOData acb; 2007d0bc9e5dSMax Reitz 200829cb4c01SKevin Wolf acb = (RawPosixAIOData) { 200993f4e2ffSKevin Wolf .bs = bs, 201093f4e2ffSKevin Wolf .aio_fildes = fd, 201193f4e2ffSKevin Wolf .aio_type = QEMU_AIO_TRUNCATE, 201293f4e2ffSKevin Wolf .aio_offset = offset, 2013d57c44d0SKevin Wolf .truncate = { 201493f4e2ffSKevin Wolf .prealloc = prealloc, 201593f4e2ffSKevin Wolf .errp = errp, 2016d57c44d0SKevin Wolf }, 201793f4e2ffSKevin Wolf }; 2018d0bc9e5dSMax Reitz 201929cb4c01SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb); 20209f63b07eSMax Reitz } 20219f63b07eSMax Reitz 2022061ca8a3SKevin Wolf static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, 20238243ccb7SMax Reitz PreallocMode prealloc, Error **errp) 2024c1bb86cdSEric Blake { 2025c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2026c1bb86cdSEric Blake struct stat st; 2027f59adb32SMax Reitz int ret; 2028c1bb86cdSEric Blake 2029c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 2030f59adb32SMax Reitz ret = -errno; 2031f59adb32SMax Reitz error_setg_errno(errp, -ret, "Failed to fstat() the file"); 2032f59adb32SMax Reitz return ret; 2033c1bb86cdSEric Blake } 2034c1bb86cdSEric Blake 2035c1bb86cdSEric Blake if (S_ISREG(st.st_mode)) { 203693f4e2ffSKevin Wolf return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); 2037c1bb86cdSEric Blake } 203835d72602SMax Reitz 203935d72602SMax Reitz if (prealloc != PREALLOC_MODE_OFF) { 204035d72602SMax Reitz error_setg(errp, "Preallocation mode '%s' unsupported for this " 2041977c736fSMarkus Armbruster "non-regular file", PreallocMode_str(prealloc)); 204235d72602SMax Reitz return -ENOTSUP; 204335d72602SMax Reitz } 204435d72602SMax Reitz 204535d72602SMax Reitz if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2046c1bb86cdSEric Blake if (offset > raw_getlength(bs)) { 2047f59adb32SMax Reitz error_setg(errp, "Cannot grow device files"); 2048c1bb86cdSEric Blake return -EINVAL; 2049c1bb86cdSEric Blake } 2050c1bb86cdSEric Blake } else { 2051f59adb32SMax Reitz error_setg(errp, "Resizing this file is not supported"); 2052c1bb86cdSEric Blake return -ENOTSUP; 2053c1bb86cdSEric Blake } 2054c1bb86cdSEric Blake 2055c1bb86cdSEric Blake return 0; 2056c1bb86cdSEric Blake } 2057c1bb86cdSEric Blake 2058c1bb86cdSEric Blake #ifdef __OpenBSD__ 2059c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2060c1bb86cdSEric Blake { 2061c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2062c1bb86cdSEric Blake int fd = s->fd; 2063c1bb86cdSEric Blake struct stat st; 2064c1bb86cdSEric Blake 2065c1bb86cdSEric Blake if (fstat(fd, &st)) 2066c1bb86cdSEric Blake return -errno; 2067c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2068c1bb86cdSEric Blake struct disklabel dl; 2069c1bb86cdSEric Blake 2070c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2071c1bb86cdSEric Blake return -errno; 2072c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2073c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2074c1bb86cdSEric Blake } else 2075c1bb86cdSEric Blake return st.st_size; 2076c1bb86cdSEric Blake } 2077c1bb86cdSEric Blake #elif defined(__NetBSD__) 2078c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2079c1bb86cdSEric Blake { 2080c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2081c1bb86cdSEric Blake int fd = s->fd; 2082c1bb86cdSEric Blake struct stat st; 2083c1bb86cdSEric Blake 2084c1bb86cdSEric Blake if (fstat(fd, &st)) 2085c1bb86cdSEric Blake return -errno; 2086c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2087c1bb86cdSEric Blake struct dkwedge_info dkw; 2088c1bb86cdSEric Blake 2089c1bb86cdSEric Blake if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { 2090c1bb86cdSEric Blake return dkw.dkw_size * 512; 2091c1bb86cdSEric Blake } else { 2092c1bb86cdSEric Blake struct disklabel dl; 2093c1bb86cdSEric Blake 2094c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2095c1bb86cdSEric Blake return -errno; 2096c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2097c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2098c1bb86cdSEric Blake } 2099c1bb86cdSEric Blake } else 2100c1bb86cdSEric Blake return st.st_size; 2101c1bb86cdSEric Blake } 2102c1bb86cdSEric Blake #elif defined(__sun__) 2103c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2104c1bb86cdSEric Blake { 2105c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2106c1bb86cdSEric Blake struct dk_minfo minfo; 2107c1bb86cdSEric Blake int ret; 2108c1bb86cdSEric Blake int64_t size; 2109c1bb86cdSEric Blake 2110c1bb86cdSEric Blake ret = fd_open(bs); 2111c1bb86cdSEric Blake if (ret < 0) { 2112c1bb86cdSEric Blake return ret; 2113c1bb86cdSEric Blake } 2114c1bb86cdSEric Blake 2115c1bb86cdSEric Blake /* 2116c1bb86cdSEric Blake * Use the DKIOCGMEDIAINFO ioctl to read the size. 2117c1bb86cdSEric Blake */ 2118c1bb86cdSEric Blake ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); 2119c1bb86cdSEric Blake if (ret != -1) { 2120c1bb86cdSEric Blake return minfo.dki_lbsize * minfo.dki_capacity; 2121c1bb86cdSEric Blake } 2122c1bb86cdSEric Blake 2123c1bb86cdSEric Blake /* 2124c1bb86cdSEric Blake * There are reports that lseek on some devices fails, but 2125c1bb86cdSEric Blake * irc discussion said that contingency on contingency was overkill. 2126c1bb86cdSEric Blake */ 2127c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2128c1bb86cdSEric Blake if (size < 0) { 2129c1bb86cdSEric Blake return -errno; 2130c1bb86cdSEric Blake } 2131c1bb86cdSEric Blake return size; 2132c1bb86cdSEric Blake } 2133c1bb86cdSEric Blake #elif defined(CONFIG_BSD) 2134c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2135c1bb86cdSEric Blake { 2136c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2137c1bb86cdSEric Blake int fd = s->fd; 2138c1bb86cdSEric Blake int64_t size; 2139c1bb86cdSEric Blake struct stat sb; 2140c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2141c1bb86cdSEric Blake int reopened = 0; 2142c1bb86cdSEric Blake #endif 2143c1bb86cdSEric Blake int ret; 2144c1bb86cdSEric Blake 2145c1bb86cdSEric Blake ret = fd_open(bs); 2146c1bb86cdSEric Blake if (ret < 0) 2147c1bb86cdSEric Blake return ret; 2148c1bb86cdSEric Blake 2149c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2150c1bb86cdSEric Blake again: 2151c1bb86cdSEric Blake #endif 2152c1bb86cdSEric Blake if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { 2153c1bb86cdSEric Blake #ifdef DIOCGMEDIASIZE 2154c1bb86cdSEric Blake if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) 2155c1bb86cdSEric Blake #elif defined(DIOCGPART) 2156c1bb86cdSEric Blake { 2157c1bb86cdSEric Blake struct partinfo pi; 2158c1bb86cdSEric Blake if (ioctl(fd, DIOCGPART, &pi) == 0) 2159c1bb86cdSEric Blake size = pi.media_size; 2160c1bb86cdSEric Blake else 2161c1bb86cdSEric Blake size = 0; 2162c1bb86cdSEric Blake } 2163c1bb86cdSEric Blake if (size == 0) 2164c1bb86cdSEric Blake #endif 2165c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2166c1bb86cdSEric Blake { 2167c1bb86cdSEric Blake uint64_t sectors = 0; 2168c1bb86cdSEric Blake uint32_t sector_size = 0; 2169c1bb86cdSEric Blake 2170c1bb86cdSEric Blake if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 2171c1bb86cdSEric Blake && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { 2172c1bb86cdSEric Blake size = sectors * sector_size; 2173c1bb86cdSEric Blake } else { 2174c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2175c1bb86cdSEric Blake if (size < 0) { 2176c1bb86cdSEric Blake return -errno; 2177c1bb86cdSEric Blake } 2178c1bb86cdSEric Blake } 2179c1bb86cdSEric Blake } 2180c1bb86cdSEric Blake #else 2181c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2182c1bb86cdSEric Blake if (size < 0) { 2183c1bb86cdSEric Blake return -errno; 2184c1bb86cdSEric Blake } 2185c1bb86cdSEric Blake #endif 2186c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 2187c1bb86cdSEric Blake switch(s->type) { 2188c1bb86cdSEric Blake case FTYPE_CD: 2189c1bb86cdSEric Blake /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ 2190c1bb86cdSEric Blake if (size == 2048LL * (unsigned)-1) 2191c1bb86cdSEric Blake size = 0; 2192c1bb86cdSEric Blake /* XXX no disc? maybe we need to reopen... */ 2193c1bb86cdSEric Blake if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { 2194c1bb86cdSEric Blake reopened = 1; 2195c1bb86cdSEric Blake goto again; 2196c1bb86cdSEric Blake } 2197c1bb86cdSEric Blake } 2198c1bb86cdSEric Blake #endif 2199c1bb86cdSEric Blake } else { 2200c1bb86cdSEric Blake size = lseek(fd, 0, SEEK_END); 2201c1bb86cdSEric Blake if (size < 0) { 2202c1bb86cdSEric Blake return -errno; 2203c1bb86cdSEric Blake } 2204c1bb86cdSEric Blake } 2205c1bb86cdSEric Blake return size; 2206c1bb86cdSEric Blake } 2207c1bb86cdSEric Blake #else 2208c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2209c1bb86cdSEric Blake { 2210c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2211c1bb86cdSEric Blake int ret; 2212c1bb86cdSEric Blake int64_t size; 2213c1bb86cdSEric Blake 2214c1bb86cdSEric Blake ret = fd_open(bs); 2215c1bb86cdSEric Blake if (ret < 0) { 2216c1bb86cdSEric Blake return ret; 2217c1bb86cdSEric Blake } 2218c1bb86cdSEric Blake 2219c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2220c1bb86cdSEric Blake if (size < 0) { 2221c1bb86cdSEric Blake return -errno; 2222c1bb86cdSEric Blake } 2223c1bb86cdSEric Blake return size; 2224c1bb86cdSEric Blake } 2225c1bb86cdSEric Blake #endif 2226c1bb86cdSEric Blake 2227c1bb86cdSEric Blake static int64_t raw_get_allocated_file_size(BlockDriverState *bs) 2228c1bb86cdSEric Blake { 2229c1bb86cdSEric Blake struct stat st; 2230c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2231c1bb86cdSEric Blake 2232c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 2233c1bb86cdSEric Blake return -errno; 2234c1bb86cdSEric Blake } 2235c1bb86cdSEric Blake return (int64_t)st.st_blocks * 512; 2236c1bb86cdSEric Blake } 2237c1bb86cdSEric Blake 223893f4e2ffSKevin Wolf static int coroutine_fn 223993f4e2ffSKevin Wolf raw_co_create(BlockdevCreateOptions *options, Error **errp) 2240c1bb86cdSEric Blake { 2241927f11e1SKevin Wolf BlockdevCreateOptionsFile *file_opts; 22427c20c808SMax Reitz Error *local_err = NULL; 2243c1bb86cdSEric Blake int fd; 2244d815efcaSMax Reitz uint64_t perm, shared; 2245c1bb86cdSEric Blake int result = 0; 2246c1bb86cdSEric Blake 2247927f11e1SKevin Wolf /* Validate options and set default values */ 2248927f11e1SKevin Wolf assert(options->driver == BLOCKDEV_DRIVER_FILE); 2249927f11e1SKevin Wolf file_opts = &options->u.file; 2250c1bb86cdSEric Blake 2251927f11e1SKevin Wolf if (!file_opts->has_nocow) { 2252927f11e1SKevin Wolf file_opts->nocow = false; 2253927f11e1SKevin Wolf } 2254927f11e1SKevin Wolf if (!file_opts->has_preallocation) { 2255927f11e1SKevin Wolf file_opts->preallocation = PREALLOC_MODE_OFF; 2256c1bb86cdSEric Blake } 2257c1bb86cdSEric Blake 2258927f11e1SKevin Wolf /* Create file */ 2259b8cf1913SMax Reitz fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_BINARY, 0644); 2260c1bb86cdSEric Blake if (fd < 0) { 2261c1bb86cdSEric Blake result = -errno; 2262c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not create file"); 2263c1bb86cdSEric Blake goto out; 2264c1bb86cdSEric Blake } 2265c1bb86cdSEric Blake 2266b8cf1913SMax Reitz /* Take permissions: We want to discard everything, so we need 2267b8cf1913SMax Reitz * BLK_PERM_WRITE; and truncation to the desired size requires 2268b8cf1913SMax Reitz * BLK_PERM_RESIZE. 2269b8cf1913SMax Reitz * On the other hand, we cannot share the RESIZE permission 2270b8cf1913SMax Reitz * because we promise that after this function, the file has the 2271b8cf1913SMax Reitz * size given in the options. If someone else were to resize it 2272b8cf1913SMax Reitz * concurrently, we could not guarantee that. 2273b8cf1913SMax Reitz * Note that after this function, we can no longer guarantee that 2274b8cf1913SMax Reitz * the file is not touched by a third party, so it may be resized 2275b8cf1913SMax Reitz * then. */ 2276b8cf1913SMax Reitz perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; 2277b8cf1913SMax Reitz shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; 2278b8cf1913SMax Reitz 2279b8cf1913SMax Reitz /* Step one: Take locks */ 22802996ffadSFam Zheng result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); 2281b8cf1913SMax Reitz if (result < 0) { 2282b8cf1913SMax Reitz goto out_close; 2283b8cf1913SMax Reitz } 2284b8cf1913SMax Reitz 2285b8cf1913SMax Reitz /* Step two: Check that nobody else has taken conflicting locks */ 2286b8cf1913SMax Reitz result = raw_check_lock_bytes(fd, perm, shared, errp); 2287b8cf1913SMax Reitz if (result < 0) { 2288b857431dSFam Zheng error_append_hint(errp, 2289b857431dSFam Zheng "Is another process using the image [%s]?\n", 2290b857431dSFam Zheng file_opts->filename); 22917c20c808SMax Reitz goto out_unlock; 2292b8cf1913SMax Reitz } 2293b8cf1913SMax Reitz 2294b8cf1913SMax Reitz /* Clear the file by truncating it to 0 */ 229593f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); 2296b8cf1913SMax Reitz if (result < 0) { 22977c20c808SMax Reitz goto out_unlock; 2298b8cf1913SMax Reitz } 2299b8cf1913SMax Reitz 2300927f11e1SKevin Wolf if (file_opts->nocow) { 2301c1bb86cdSEric Blake #ifdef __linux__ 2302c1bb86cdSEric Blake /* Set NOCOW flag to solve performance issue on fs like btrfs. 2303c1bb86cdSEric Blake * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value 2304c1bb86cdSEric Blake * will be ignored since any failure of this operation should not 2305c1bb86cdSEric Blake * block the left work. 2306c1bb86cdSEric Blake */ 2307c1bb86cdSEric Blake int attr; 2308c1bb86cdSEric Blake if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { 2309c1bb86cdSEric Blake attr |= FS_NOCOW_FL; 2310c1bb86cdSEric Blake ioctl(fd, FS_IOC_SETFLAGS, &attr); 2311c1bb86cdSEric Blake } 2312c1bb86cdSEric Blake #endif 2313c1bb86cdSEric Blake } 2314c1bb86cdSEric Blake 2315b8cf1913SMax Reitz /* Resize and potentially preallocate the file to the desired 2316b8cf1913SMax Reitz * final size */ 231793f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, file_opts->size, 231893f4e2ffSKevin Wolf file_opts->preallocation, errp); 23199f63b07eSMax Reitz if (result < 0) { 23207c20c808SMax Reitz goto out_unlock; 23217c20c808SMax Reitz } 23227c20c808SMax Reitz 23237c20c808SMax Reitz out_unlock: 23242996ffadSFam Zheng raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); 23257c20c808SMax Reitz if (local_err) { 23267c20c808SMax Reitz /* The above call should not fail, and if it does, that does 23277c20c808SMax Reitz * not mean the whole creation operation has failed. So 23287c20c808SMax Reitz * report it the user for their convenience, but do not report 23297c20c808SMax Reitz * it to the caller. */ 2330db0754dfSFam Zheng warn_report_err(local_err); 23315a1dad9dSNir Soffer } 23325a1dad9dSNir Soffer 23335a1dad9dSNir Soffer out_close: 2334c1bb86cdSEric Blake if (qemu_close(fd) != 0 && result == 0) { 2335c1bb86cdSEric Blake result = -errno; 2336c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not close the new file"); 2337c1bb86cdSEric Blake } 2338c1bb86cdSEric Blake out: 2339c1bb86cdSEric Blake return result; 2340c1bb86cdSEric Blake } 2341c1bb86cdSEric Blake 2342927f11e1SKevin Wolf static int coroutine_fn raw_co_create_opts(const char *filename, QemuOpts *opts, 2343927f11e1SKevin Wolf Error **errp) 2344927f11e1SKevin Wolf { 2345927f11e1SKevin Wolf BlockdevCreateOptions options; 2346927f11e1SKevin Wolf int64_t total_size = 0; 2347927f11e1SKevin Wolf bool nocow = false; 2348927f11e1SKevin Wolf PreallocMode prealloc; 2349927f11e1SKevin Wolf char *buf = NULL; 2350927f11e1SKevin Wolf Error *local_err = NULL; 2351927f11e1SKevin Wolf 2352927f11e1SKevin Wolf /* Skip file: protocol prefix */ 2353927f11e1SKevin Wolf strstart(filename, "file:", &filename); 2354927f11e1SKevin Wolf 2355927f11e1SKevin Wolf /* Read out options */ 2356927f11e1SKevin Wolf total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2357927f11e1SKevin Wolf BDRV_SECTOR_SIZE); 2358927f11e1SKevin Wolf nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); 2359927f11e1SKevin Wolf buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2360927f11e1SKevin Wolf prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, 2361927f11e1SKevin Wolf PREALLOC_MODE_OFF, &local_err); 2362927f11e1SKevin Wolf g_free(buf); 2363927f11e1SKevin Wolf if (local_err) { 2364927f11e1SKevin Wolf error_propagate(errp, local_err); 2365927f11e1SKevin Wolf return -EINVAL; 2366927f11e1SKevin Wolf } 2367927f11e1SKevin Wolf 2368927f11e1SKevin Wolf options = (BlockdevCreateOptions) { 2369927f11e1SKevin Wolf .driver = BLOCKDEV_DRIVER_FILE, 2370927f11e1SKevin Wolf .u.file = { 2371927f11e1SKevin Wolf .filename = (char *) filename, 2372927f11e1SKevin Wolf .size = total_size, 2373927f11e1SKevin Wolf .has_preallocation = true, 2374927f11e1SKevin Wolf .preallocation = prealloc, 2375927f11e1SKevin Wolf .has_nocow = true, 2376927f11e1SKevin Wolf .nocow = nocow, 2377927f11e1SKevin Wolf }, 2378927f11e1SKevin Wolf }; 2379927f11e1SKevin Wolf return raw_co_create(&options, errp); 2380927f11e1SKevin Wolf } 2381927f11e1SKevin Wolf 2382c1bb86cdSEric Blake /* 2383c1bb86cdSEric Blake * Find allocation range in @bs around offset @start. 2384c1bb86cdSEric Blake * May change underlying file descriptor's file offset. 2385c1bb86cdSEric Blake * If @start is not in a hole, store @start in @data, and the 2386c1bb86cdSEric Blake * beginning of the next hole in @hole, and return 0. 2387c1bb86cdSEric Blake * If @start is in a non-trailing hole, store @start in @hole and the 2388c1bb86cdSEric Blake * beginning of the next non-hole in @data, and return 0. 2389c1bb86cdSEric Blake * If @start is in a trailing hole or beyond EOF, return -ENXIO. 2390c1bb86cdSEric Blake * If we can't find out, return a negative errno other than -ENXIO. 2391c1bb86cdSEric Blake */ 2392c1bb86cdSEric Blake static int find_allocation(BlockDriverState *bs, off_t start, 2393c1bb86cdSEric Blake off_t *data, off_t *hole) 2394c1bb86cdSEric Blake { 2395c1bb86cdSEric Blake #if defined SEEK_HOLE && defined SEEK_DATA 2396c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2397c1bb86cdSEric Blake off_t offs; 2398c1bb86cdSEric Blake 2399c1bb86cdSEric Blake /* 2400c1bb86cdSEric Blake * SEEK_DATA cases: 2401c1bb86cdSEric Blake * D1. offs == start: start is in data 2402c1bb86cdSEric Blake * D2. offs > start: start is in a hole, next data at offs 2403c1bb86cdSEric Blake * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 2404c1bb86cdSEric Blake * or start is beyond EOF 2405c1bb86cdSEric Blake * If the latter happens, the file has been truncated behind 2406c1bb86cdSEric Blake * our back since we opened it. All bets are off then. 2407c1bb86cdSEric Blake * Treating like a trailing hole is simplest. 2408c1bb86cdSEric Blake * D4. offs < 0, errno != ENXIO: we learned nothing 2409c1bb86cdSEric Blake */ 2410c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_DATA); 2411c1bb86cdSEric Blake if (offs < 0) { 2412c1bb86cdSEric Blake return -errno; /* D3 or D4 */ 2413c1bb86cdSEric Blake } 2414a03083a0SJeff Cody 2415a03083a0SJeff Cody if (offs < start) { 2416a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2417a03083a0SJeff Cody * -EIO in this case, and we'll treat it like D4. */ 2418a03083a0SJeff Cody return -EIO; 2419a03083a0SJeff Cody } 2420c1bb86cdSEric Blake 2421c1bb86cdSEric Blake if (offs > start) { 2422c1bb86cdSEric Blake /* D2: in hole, next data at offs */ 2423c1bb86cdSEric Blake *hole = start; 2424c1bb86cdSEric Blake *data = offs; 2425c1bb86cdSEric Blake return 0; 2426c1bb86cdSEric Blake } 2427c1bb86cdSEric Blake 2428c1bb86cdSEric Blake /* D1: in data, end not yet known */ 2429c1bb86cdSEric Blake 2430c1bb86cdSEric Blake /* 2431c1bb86cdSEric Blake * SEEK_HOLE cases: 2432c1bb86cdSEric Blake * H1. offs == start: start is in a hole 2433c1bb86cdSEric Blake * If this happens here, a hole has been dug behind our back 2434c1bb86cdSEric Blake * since the previous lseek(). 2435c1bb86cdSEric Blake * H2. offs > start: either start is in data, next hole at offs, 2436c1bb86cdSEric Blake * or start is in trailing hole, EOF at offs 2437c1bb86cdSEric Blake * Linux treats trailing holes like any other hole: offs == 2438c1bb86cdSEric Blake * start. Solaris seeks to EOF instead: offs > start (blech). 2439c1bb86cdSEric Blake * If that happens here, a hole has been dug behind our back 2440c1bb86cdSEric Blake * since the previous lseek(). 2441c1bb86cdSEric Blake * H3. offs < 0, errno = ENXIO: start is beyond EOF 2442c1bb86cdSEric Blake * If this happens, the file has been truncated behind our 2443c1bb86cdSEric Blake * back since we opened it. Treat it like a trailing hole. 2444c1bb86cdSEric Blake * H4. offs < 0, errno != ENXIO: we learned nothing 2445c1bb86cdSEric Blake * Pretend we know nothing at all, i.e. "forget" about D1. 2446c1bb86cdSEric Blake */ 2447c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_HOLE); 2448c1bb86cdSEric Blake if (offs < 0) { 2449c1bb86cdSEric Blake return -errno; /* D1 and (H3 or H4) */ 2450c1bb86cdSEric Blake } 2451a03083a0SJeff Cody 2452a03083a0SJeff Cody if (offs < start) { 2453a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2454a03083a0SJeff Cody * -EIO in this case, and we'll treat it like H4. */ 2455a03083a0SJeff Cody return -EIO; 2456a03083a0SJeff Cody } 2457c1bb86cdSEric Blake 2458c1bb86cdSEric Blake if (offs > start) { 2459c1bb86cdSEric Blake /* 2460c1bb86cdSEric Blake * D1 and H2: either in data, next hole at offs, or it was in 2461c1bb86cdSEric Blake * data but is now in a trailing hole. In the latter case, 2462c1bb86cdSEric Blake * all bets are off. Treating it as if it there was data all 2463c1bb86cdSEric Blake * the way to EOF is safe, so simply do that. 2464c1bb86cdSEric Blake */ 2465c1bb86cdSEric Blake *data = start; 2466c1bb86cdSEric Blake *hole = offs; 2467c1bb86cdSEric Blake return 0; 2468c1bb86cdSEric Blake } 2469c1bb86cdSEric Blake 2470c1bb86cdSEric Blake /* D1 and H1 */ 2471c1bb86cdSEric Blake return -EBUSY; 2472c1bb86cdSEric Blake #else 2473c1bb86cdSEric Blake return -ENOTSUP; 2474c1bb86cdSEric Blake #endif 2475c1bb86cdSEric Blake } 2476c1bb86cdSEric Blake 2477c1bb86cdSEric Blake /* 2478a290f085SEric Blake * Returns the allocation status of the specified offset. 2479c1bb86cdSEric Blake * 2480a290f085SEric Blake * The block layer guarantees 'offset' and 'bytes' are within bounds. 2481c1bb86cdSEric Blake * 2482a290f085SEric Blake * 'pnum' is set to the number of bytes (including and immediately following 2483a290f085SEric Blake * the specified offset) that are known to be in the same 2484c1bb86cdSEric Blake * allocated/unallocated state. 2485c1bb86cdSEric Blake * 2486a290f085SEric Blake * 'bytes' is the max value 'pnum' should be set to. 2487c1bb86cdSEric Blake */ 2488a290f085SEric Blake static int coroutine_fn raw_co_block_status(BlockDriverState *bs, 2489a290f085SEric Blake bool want_zero, 2490a290f085SEric Blake int64_t offset, 2491a290f085SEric Blake int64_t bytes, int64_t *pnum, 2492a290f085SEric Blake int64_t *map, 2493c1bb86cdSEric Blake BlockDriverState **file) 2494c1bb86cdSEric Blake { 2495a290f085SEric Blake off_t data = 0, hole = 0; 2496c1bb86cdSEric Blake int ret; 2497c1bb86cdSEric Blake 24989c3db310SMax Reitz assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); 24999c3db310SMax Reitz 2500c1bb86cdSEric Blake ret = fd_open(bs); 2501c1bb86cdSEric Blake if (ret < 0) { 2502c1bb86cdSEric Blake return ret; 2503c1bb86cdSEric Blake } 2504c1bb86cdSEric Blake 2505a290f085SEric Blake if (!want_zero) { 2506a290f085SEric Blake *pnum = bytes; 2507a290f085SEric Blake *map = offset; 2508a290f085SEric Blake *file = bs; 2509a290f085SEric Blake return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 2510c1bb86cdSEric Blake } 2511c1bb86cdSEric Blake 2512a290f085SEric Blake ret = find_allocation(bs, offset, &data, &hole); 2513c1bb86cdSEric Blake if (ret == -ENXIO) { 2514c1bb86cdSEric Blake /* Trailing hole */ 2515a290f085SEric Blake *pnum = bytes; 2516c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2517c1bb86cdSEric Blake } else if (ret < 0) { 2518c1bb86cdSEric Blake /* No info available, so pretend there are no holes */ 2519a290f085SEric Blake *pnum = bytes; 2520c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2521a290f085SEric Blake } else if (data == offset) { 2522a290f085SEric Blake /* On a data extent, compute bytes to the end of the extent, 2523c1bb86cdSEric Blake * possibly including a partial sector at EOF. */ 2524a290f085SEric Blake *pnum = MIN(bytes, hole - offset); 25259c3db310SMax Reitz 25269c3db310SMax Reitz /* 25279c3db310SMax Reitz * We are not allowed to return partial sectors, though, so 25289c3db310SMax Reitz * round up if necessary. 25299c3db310SMax Reitz */ 25309c3db310SMax Reitz if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { 25319c3db310SMax Reitz int64_t file_length = raw_getlength(bs); 25329c3db310SMax Reitz if (file_length > 0) { 25339c3db310SMax Reitz /* Ignore errors, this is just a safeguard */ 25349c3db310SMax Reitz assert(hole == file_length); 25359c3db310SMax Reitz } 25369c3db310SMax Reitz *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); 25379c3db310SMax Reitz } 25389c3db310SMax Reitz 2539c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2540c1bb86cdSEric Blake } else { 2541a290f085SEric Blake /* On a hole, compute bytes to the beginning of the next extent. */ 2542a290f085SEric Blake assert(hole == offset); 2543a290f085SEric Blake *pnum = MIN(bytes, data - offset); 2544c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2545c1bb86cdSEric Blake } 2546a290f085SEric Blake *map = offset; 2547c1bb86cdSEric Blake *file = bs; 2548a290f085SEric Blake return ret | BDRV_BLOCK_OFFSET_VALID; 2549c1bb86cdSEric Blake } 2550c1bb86cdSEric Blake 255131be8a2aSStefan Hajnoczi #if defined(__linux__) 255231be8a2aSStefan Hajnoczi /* Verify that the file is not in the page cache */ 255331be8a2aSStefan Hajnoczi static void check_cache_dropped(BlockDriverState *bs, Error **errp) 255431be8a2aSStefan Hajnoczi { 255531be8a2aSStefan Hajnoczi const size_t window_size = 128 * 1024 * 1024; 255631be8a2aSStefan Hajnoczi BDRVRawState *s = bs->opaque; 255731be8a2aSStefan Hajnoczi void *window = NULL; 255831be8a2aSStefan Hajnoczi size_t length = 0; 255931be8a2aSStefan Hajnoczi unsigned char *vec; 256031be8a2aSStefan Hajnoczi size_t page_size; 256131be8a2aSStefan Hajnoczi off_t offset; 256231be8a2aSStefan Hajnoczi off_t end; 256331be8a2aSStefan Hajnoczi 256431be8a2aSStefan Hajnoczi /* mincore(2) page status information requires 1 byte per page */ 256531be8a2aSStefan Hajnoczi page_size = sysconf(_SC_PAGESIZE); 256631be8a2aSStefan Hajnoczi vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); 256731be8a2aSStefan Hajnoczi 256831be8a2aSStefan Hajnoczi end = raw_getlength(bs); 256931be8a2aSStefan Hajnoczi 257031be8a2aSStefan Hajnoczi for (offset = 0; offset < end; offset += window_size) { 257131be8a2aSStefan Hajnoczi void *new_window; 257231be8a2aSStefan Hajnoczi size_t new_length; 257331be8a2aSStefan Hajnoczi size_t vec_end; 257431be8a2aSStefan Hajnoczi size_t i; 257531be8a2aSStefan Hajnoczi int ret; 257631be8a2aSStefan Hajnoczi 257731be8a2aSStefan Hajnoczi /* Unmap previous window if size has changed */ 257831be8a2aSStefan Hajnoczi new_length = MIN(end - offset, window_size); 257931be8a2aSStefan Hajnoczi if (new_length != length) { 258031be8a2aSStefan Hajnoczi munmap(window, length); 258131be8a2aSStefan Hajnoczi window = NULL; 258231be8a2aSStefan Hajnoczi length = 0; 258331be8a2aSStefan Hajnoczi } 258431be8a2aSStefan Hajnoczi 258531be8a2aSStefan Hajnoczi new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, 258631be8a2aSStefan Hajnoczi s->fd, offset); 258731be8a2aSStefan Hajnoczi if (new_window == MAP_FAILED) { 258831be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mmap failed"); 258931be8a2aSStefan Hajnoczi break; 259031be8a2aSStefan Hajnoczi } 259131be8a2aSStefan Hajnoczi 259231be8a2aSStefan Hajnoczi window = new_window; 259331be8a2aSStefan Hajnoczi length = new_length; 259431be8a2aSStefan Hajnoczi 259531be8a2aSStefan Hajnoczi ret = mincore(window, length, vec); 259631be8a2aSStefan Hajnoczi if (ret < 0) { 259731be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mincore failed"); 259831be8a2aSStefan Hajnoczi break; 259931be8a2aSStefan Hajnoczi } 260031be8a2aSStefan Hajnoczi 260131be8a2aSStefan Hajnoczi vec_end = DIV_ROUND_UP(length, page_size); 260231be8a2aSStefan Hajnoczi for (i = 0; i < vec_end; i++) { 260331be8a2aSStefan Hajnoczi if (vec[i] & 0x1) { 260431be8a2aSStefan Hajnoczi error_setg(errp, "page cache still in use!"); 260531be8a2aSStefan Hajnoczi break; 260631be8a2aSStefan Hajnoczi } 260731be8a2aSStefan Hajnoczi } 260831be8a2aSStefan Hajnoczi } 260931be8a2aSStefan Hajnoczi 261031be8a2aSStefan Hajnoczi if (window) { 261131be8a2aSStefan Hajnoczi munmap(window, length); 261231be8a2aSStefan Hajnoczi } 261331be8a2aSStefan Hajnoczi 261431be8a2aSStefan Hajnoczi g_free(vec); 261531be8a2aSStefan Hajnoczi } 261631be8a2aSStefan Hajnoczi #endif /* __linux__ */ 261731be8a2aSStefan Hajnoczi 2618dd577a26SStefan Hajnoczi static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, 2619dd577a26SStefan Hajnoczi Error **errp) 2620dd577a26SStefan Hajnoczi { 2621dd577a26SStefan Hajnoczi BDRVRawState *s = bs->opaque; 2622dd577a26SStefan Hajnoczi int ret; 2623dd577a26SStefan Hajnoczi 2624dd577a26SStefan Hajnoczi ret = fd_open(bs); 2625dd577a26SStefan Hajnoczi if (ret < 0) { 2626dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "The file descriptor is not open"); 2627dd577a26SStefan Hajnoczi return; 2628dd577a26SStefan Hajnoczi } 2629dd577a26SStefan Hajnoczi 2630f357fcd8SStefan Hajnoczi if (!s->drop_cache) { 2631f357fcd8SStefan Hajnoczi return; 2632f357fcd8SStefan Hajnoczi } 2633f357fcd8SStefan Hajnoczi 2634dd577a26SStefan Hajnoczi if (s->open_flags & O_DIRECT) { 2635dd577a26SStefan Hajnoczi return; /* No host kernel page cache */ 2636dd577a26SStefan Hajnoczi } 2637dd577a26SStefan Hajnoczi 2638dd577a26SStefan Hajnoczi #if defined(__linux__) 2639dd577a26SStefan Hajnoczi /* This sets the scene for the next syscall... */ 2640dd577a26SStefan Hajnoczi ret = bdrv_co_flush(bs); 2641dd577a26SStefan Hajnoczi if (ret < 0) { 2642dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "flush failed"); 2643dd577a26SStefan Hajnoczi return; 2644dd577a26SStefan Hajnoczi } 2645dd577a26SStefan Hajnoczi 2646dd577a26SStefan Hajnoczi /* Linux does not invalidate pages that are dirty, locked, or mmapped by a 2647dd577a26SStefan Hajnoczi * process. These limitations are okay because we just fsynced the file, 2648dd577a26SStefan Hajnoczi * we don't use mmap, and the file should not be in use by other processes. 2649dd577a26SStefan Hajnoczi */ 2650dd577a26SStefan Hajnoczi ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); 2651dd577a26SStefan Hajnoczi if (ret != 0) { /* the return value is a positive errno */ 2652dd577a26SStefan Hajnoczi error_setg_errno(errp, ret, "fadvise failed"); 2653dd577a26SStefan Hajnoczi return; 2654dd577a26SStefan Hajnoczi } 265531be8a2aSStefan Hajnoczi 265631be8a2aSStefan Hajnoczi if (s->check_cache_dropped) { 265731be8a2aSStefan Hajnoczi check_cache_dropped(bs, errp); 265831be8a2aSStefan Hajnoczi } 2659dd577a26SStefan Hajnoczi #else /* __linux__ */ 2660dd577a26SStefan Hajnoczi /* Do nothing. Live migration to a remote host with cache.direct=off is 2661dd577a26SStefan Hajnoczi * unsupported on other host operating systems. Cache consistency issues 2662dd577a26SStefan Hajnoczi * may occur but no error is reported here, partly because that's the 2663dd577a26SStefan Hajnoczi * historical behavior and partly because it's hard to differentiate valid 2664dd577a26SStefan Hajnoczi * configurations that should not cause errors. 2665dd577a26SStefan Hajnoczi */ 2666dd577a26SStefan Hajnoczi #endif /* !__linux__ */ 2667dd577a26SStefan Hajnoczi } 2668dd577a26SStefan Hajnoczi 26691c450366SAnton Nefedov static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret) 26701c450366SAnton Nefedov { 26711c450366SAnton Nefedov if (ret) { 26721c450366SAnton Nefedov s->stats.discard_nb_failed++; 26731c450366SAnton Nefedov } else { 26741c450366SAnton Nefedov s->stats.discard_nb_ok++; 26751c450366SAnton Nefedov s->stats.discard_bytes_ok += nbytes; 26761c450366SAnton Nefedov } 26771c450366SAnton Nefedov } 26781c450366SAnton Nefedov 267933d70fb6SKevin Wolf static coroutine_fn int 268046ee0f46SKevin Wolf raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev) 2681c1bb86cdSEric Blake { 2682c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 268346ee0f46SKevin Wolf RawPosixAIOData acb; 26841c450366SAnton Nefedov int ret; 2685c1bb86cdSEric Blake 268646ee0f46SKevin Wolf acb = (RawPosixAIOData) { 268746ee0f46SKevin Wolf .bs = bs, 268846ee0f46SKevin Wolf .aio_fildes = s->fd, 268946ee0f46SKevin Wolf .aio_type = QEMU_AIO_DISCARD, 269046ee0f46SKevin Wolf .aio_offset = offset, 269146ee0f46SKevin Wolf .aio_nbytes = bytes, 269246ee0f46SKevin Wolf }; 269346ee0f46SKevin Wolf 269446ee0f46SKevin Wolf if (blkdev) { 269546ee0f46SKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 269646ee0f46SKevin Wolf } 269746ee0f46SKevin Wolf 26981c450366SAnton Nefedov ret = raw_thread_pool_submit(bs, handle_aiocb_discard, &acb); 26991c450366SAnton Nefedov raw_account_discard(s, bytes, ret); 27001c450366SAnton Nefedov return ret; 270146ee0f46SKevin Wolf } 270246ee0f46SKevin Wolf 270346ee0f46SKevin Wolf static coroutine_fn int 270446ee0f46SKevin Wolf raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 270546ee0f46SKevin Wolf { 270646ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, false); 2707c1bb86cdSEric Blake } 2708c1bb86cdSEric Blake 27097154d8aeSKevin Wolf static int coroutine_fn 27107154d8aeSKevin Wolf raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, 27117154d8aeSKevin Wolf BdrvRequestFlags flags, bool blkdev) 27127154d8aeSKevin Wolf { 27137154d8aeSKevin Wolf BDRVRawState *s = bs->opaque; 27147154d8aeSKevin Wolf RawPosixAIOData acb; 27157154d8aeSKevin Wolf ThreadPoolFunc *handler; 27167154d8aeSKevin Wolf 27177154d8aeSKevin Wolf acb = (RawPosixAIOData) { 27187154d8aeSKevin Wolf .bs = bs, 27197154d8aeSKevin Wolf .aio_fildes = s->fd, 27207154d8aeSKevin Wolf .aio_type = QEMU_AIO_WRITE_ZEROES, 27217154d8aeSKevin Wolf .aio_offset = offset, 27227154d8aeSKevin Wolf .aio_nbytes = bytes, 27237154d8aeSKevin Wolf }; 27247154d8aeSKevin Wolf 27257154d8aeSKevin Wolf if (blkdev) { 27267154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 27277154d8aeSKevin Wolf } 2728738301e1SKevin Wolf if (flags & BDRV_REQ_NO_FALLBACK) { 2729738301e1SKevin Wolf acb.aio_type |= QEMU_AIO_NO_FALLBACK; 2730738301e1SKevin Wolf } 27317154d8aeSKevin Wolf 27327154d8aeSKevin Wolf if (flags & BDRV_REQ_MAY_UNMAP) { 27337154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_DISCARD; 27347154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes_unmap; 27357154d8aeSKevin Wolf } else { 27367154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes; 27377154d8aeSKevin Wolf } 27387154d8aeSKevin Wolf 27397154d8aeSKevin Wolf return raw_thread_pool_submit(bs, handler, &acb); 27407154d8aeSKevin Wolf } 27417154d8aeSKevin Wolf 2742c1bb86cdSEric Blake static int coroutine_fn raw_co_pwrite_zeroes( 2743c1bb86cdSEric Blake BlockDriverState *bs, int64_t offset, 2744f5a5ca79SManos Pitsidianakis int bytes, BdrvRequestFlags flags) 2745c1bb86cdSEric Blake { 27467154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); 2747c1bb86cdSEric Blake } 2748c1bb86cdSEric Blake 2749c1bb86cdSEric Blake static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2750c1bb86cdSEric Blake { 2751c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2752c1bb86cdSEric Blake 2753c1bb86cdSEric Blake bdi->unallocated_blocks_are_zero = s->discard_zeroes; 2754c1bb86cdSEric Blake return 0; 2755c1bb86cdSEric Blake } 2756c1bb86cdSEric Blake 2757d9245599SAnton Nefedov static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs) 2758d9245599SAnton Nefedov { 2759d9245599SAnton Nefedov BDRVRawState *s = bs->opaque; 2760d9245599SAnton Nefedov return (BlockStatsSpecificFile) { 2761d9245599SAnton Nefedov .discard_nb_ok = s->stats.discard_nb_ok, 2762d9245599SAnton Nefedov .discard_nb_failed = s->stats.discard_nb_failed, 2763d9245599SAnton Nefedov .discard_bytes_ok = s->stats.discard_bytes_ok, 2764d9245599SAnton Nefedov }; 2765d9245599SAnton Nefedov } 2766d9245599SAnton Nefedov 2767d9245599SAnton Nefedov static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs) 2768d9245599SAnton Nefedov { 2769d9245599SAnton Nefedov BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 2770d9245599SAnton Nefedov 2771d9245599SAnton Nefedov stats->driver = BLOCKDEV_DRIVER_FILE; 2772d9245599SAnton Nefedov stats->u.file = get_blockstats_specific_file(bs); 2773d9245599SAnton Nefedov 2774d9245599SAnton Nefedov return stats; 2775d9245599SAnton Nefedov } 2776d9245599SAnton Nefedov 2777d9245599SAnton Nefedov static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs) 2778d9245599SAnton Nefedov { 2779d9245599SAnton Nefedov BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 2780d9245599SAnton Nefedov 2781d9245599SAnton Nefedov stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE; 2782d9245599SAnton Nefedov stats->u.host_device = get_blockstats_specific_file(bs); 2783d9245599SAnton Nefedov 2784d9245599SAnton Nefedov return stats; 2785d9245599SAnton Nefedov } 2786d9245599SAnton Nefedov 2787c1bb86cdSEric Blake static QemuOptsList raw_create_opts = { 2788c1bb86cdSEric Blake .name = "raw-create-opts", 2789c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), 2790c1bb86cdSEric Blake .desc = { 2791c1bb86cdSEric Blake { 2792c1bb86cdSEric Blake .name = BLOCK_OPT_SIZE, 2793c1bb86cdSEric Blake .type = QEMU_OPT_SIZE, 2794c1bb86cdSEric Blake .help = "Virtual disk size" 2795c1bb86cdSEric Blake }, 2796c1bb86cdSEric Blake { 2797c1bb86cdSEric Blake .name = BLOCK_OPT_NOCOW, 2798c1bb86cdSEric Blake .type = QEMU_OPT_BOOL, 2799c1bb86cdSEric Blake .help = "Turn off copy-on-write (valid only on btrfs)" 2800c1bb86cdSEric Blake }, 2801c1bb86cdSEric Blake { 2802c1bb86cdSEric Blake .name = BLOCK_OPT_PREALLOC, 2803c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 2804abea0053SStefano Garzarella .help = "Preallocation mode (allowed values: off" 2805abea0053SStefano Garzarella #ifdef CONFIG_POSIX_FALLOCATE 2806abea0053SStefano Garzarella ", falloc" 2807abea0053SStefano Garzarella #endif 2808abea0053SStefano Garzarella ", full)" 2809c1bb86cdSEric Blake }, 2810c1bb86cdSEric Blake { /* end of list */ } 2811c1bb86cdSEric Blake } 2812c1bb86cdSEric Blake }; 2813c1bb86cdSEric Blake 2814244a5668SFam Zheng static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, 2815244a5668SFam Zheng Error **errp) 2816244a5668SFam Zheng { 28176ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 28186ceabe6fSKevin Wolf BDRVRawReopenState *rs = NULL; 28196ceabe6fSKevin Wolf int open_flags; 28206ceabe6fSKevin Wolf int ret; 28216ceabe6fSKevin Wolf 28226ceabe6fSKevin Wolf if (s->perm_change_fd) { 28236ceabe6fSKevin Wolf /* 28246ceabe6fSKevin Wolf * In the context of reopen, this function may be called several times 28256ceabe6fSKevin Wolf * (directly and recursively while change permissions of the parent). 28266ceabe6fSKevin Wolf * This is even true for children that don't inherit from the original 28276ceabe6fSKevin Wolf * reopen node, so s->reopen_state is not set. 28286ceabe6fSKevin Wolf * 28296ceabe6fSKevin Wolf * Ignore all but the first call. 28306ceabe6fSKevin Wolf */ 28316ceabe6fSKevin Wolf return 0; 28326ceabe6fSKevin Wolf } 28336ceabe6fSKevin Wolf 28346ceabe6fSKevin Wolf if (s->reopen_state) { 28356ceabe6fSKevin Wolf /* We already have a new file descriptor to set permissions for */ 28366ceabe6fSKevin Wolf assert(s->reopen_state->perm == perm); 28376ceabe6fSKevin Wolf assert(s->reopen_state->shared_perm == shared); 28386ceabe6fSKevin Wolf rs = s->reopen_state->opaque; 28396ceabe6fSKevin Wolf s->perm_change_fd = rs->fd; 2840094e3639SMax Reitz s->perm_change_flags = rs->open_flags; 28416ceabe6fSKevin Wolf } else { 28426ceabe6fSKevin Wolf /* We may need a new fd if auto-read-only switches the mode */ 284323dece19SKevin Wolf ret = raw_reconfigure_getfd(bs, bs->open_flags, &open_flags, perm, 28446ceabe6fSKevin Wolf false, errp); 28456ceabe6fSKevin Wolf if (ret < 0) { 28466ceabe6fSKevin Wolf return ret; 28476ceabe6fSKevin Wolf } else if (ret != s->fd) { 28486ceabe6fSKevin Wolf s->perm_change_fd = ret; 2849094e3639SMax Reitz s->perm_change_flags = open_flags; 28506ceabe6fSKevin Wolf } 28516ceabe6fSKevin Wolf } 28526ceabe6fSKevin Wolf 28536ceabe6fSKevin Wolf /* Prepare permissions on old fd to avoid conflicts between old and new, 28546ceabe6fSKevin Wolf * but keep everything locked that new will need. */ 28556ceabe6fSKevin Wolf ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); 28566ceabe6fSKevin Wolf if (ret < 0) { 28576ceabe6fSKevin Wolf goto fail; 28586ceabe6fSKevin Wolf } 28596ceabe6fSKevin Wolf 28606ceabe6fSKevin Wolf /* Copy locks to the new fd */ 28616ceabe6fSKevin Wolf if (s->perm_change_fd) { 28626ceabe6fSKevin Wolf ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, 28636ceabe6fSKevin Wolf false, errp); 28646ceabe6fSKevin Wolf if (ret < 0) { 28656ceabe6fSKevin Wolf raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 28666ceabe6fSKevin Wolf goto fail; 28676ceabe6fSKevin Wolf } 28686ceabe6fSKevin Wolf } 28696ceabe6fSKevin Wolf return 0; 28706ceabe6fSKevin Wolf 28716ceabe6fSKevin Wolf fail: 28726ceabe6fSKevin Wolf if (s->perm_change_fd && !s->reopen_state) { 28736ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 28746ceabe6fSKevin Wolf } 28756ceabe6fSKevin Wolf s->perm_change_fd = 0; 28766ceabe6fSKevin Wolf return ret; 2877244a5668SFam Zheng } 2878244a5668SFam Zheng 2879244a5668SFam Zheng static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) 2880244a5668SFam Zheng { 2881244a5668SFam Zheng BDRVRawState *s = bs->opaque; 28826ceabe6fSKevin Wolf 28836ceabe6fSKevin Wolf /* For reopen, we have already switched to the new fd (.bdrv_set_perm is 28846ceabe6fSKevin Wolf * called after .bdrv_reopen_commit) */ 28856ceabe6fSKevin Wolf if (s->perm_change_fd && s->fd != s->perm_change_fd) { 28866ceabe6fSKevin Wolf qemu_close(s->fd); 28876ceabe6fSKevin Wolf s->fd = s->perm_change_fd; 2888094e3639SMax Reitz s->open_flags = s->perm_change_flags; 28896ceabe6fSKevin Wolf } 28906ceabe6fSKevin Wolf s->perm_change_fd = 0; 28916ceabe6fSKevin Wolf 2892244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); 2893244a5668SFam Zheng s->perm = perm; 2894244a5668SFam Zheng s->shared_perm = shared; 2895244a5668SFam Zheng } 2896244a5668SFam Zheng 2897244a5668SFam Zheng static void raw_abort_perm_update(BlockDriverState *bs) 2898244a5668SFam Zheng { 28996ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 29006ceabe6fSKevin Wolf 29016ceabe6fSKevin Wolf /* For reopen, .bdrv_reopen_abort is called afterwards and will close 29026ceabe6fSKevin Wolf * the file descriptor. */ 29036ceabe6fSKevin Wolf if (s->perm_change_fd && !s->reopen_state) { 29046ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 29056ceabe6fSKevin Wolf } 29066ceabe6fSKevin Wolf s->perm_change_fd = 0; 29076ceabe6fSKevin Wolf 2908244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 2909244a5668SFam Zheng } 2910244a5668SFam Zheng 291167b51fb9SVladimir Sementsov-Ogievskiy static int coroutine_fn raw_co_copy_range_from( 291267b51fb9SVladimir Sementsov-Ogievskiy BlockDriverState *bs, BdrvChild *src, uint64_t src_offset, 291367b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, uint64_t dst_offset, uint64_t bytes, 291467b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) 29151efad060SFam Zheng { 291667b51fb9SVladimir Sementsov-Ogievskiy return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, 291767b51fb9SVladimir Sementsov-Ogievskiy read_flags, write_flags); 29181efad060SFam Zheng } 29191efad060SFam Zheng 29201efad060SFam Zheng static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, 292167b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *src, 292267b51fb9SVladimir Sementsov-Ogievskiy uint64_t src_offset, 292367b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, 292467b51fb9SVladimir Sementsov-Ogievskiy uint64_t dst_offset, 292567b51fb9SVladimir Sementsov-Ogievskiy uint64_t bytes, 292667b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, 292767b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags write_flags) 29281efad060SFam Zheng { 292958a209c4SKevin Wolf RawPosixAIOData acb; 29301efad060SFam Zheng BDRVRawState *s = bs->opaque; 29311efad060SFam Zheng BDRVRawState *src_s; 29321efad060SFam Zheng 29331efad060SFam Zheng assert(dst->bs == bs); 29341efad060SFam Zheng if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { 29351efad060SFam Zheng return -ENOTSUP; 29361efad060SFam Zheng } 29371efad060SFam Zheng 29381efad060SFam Zheng src_s = src->bs->opaque; 29399f850f67SFam Zheng if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { 29401efad060SFam Zheng return -EIO; 29411efad060SFam Zheng } 294258a209c4SKevin Wolf 294358a209c4SKevin Wolf acb = (RawPosixAIOData) { 294458a209c4SKevin Wolf .bs = bs, 294558a209c4SKevin Wolf .aio_type = QEMU_AIO_COPY_RANGE, 294658a209c4SKevin Wolf .aio_fildes = src_s->fd, 294758a209c4SKevin Wolf .aio_offset = src_offset, 294858a209c4SKevin Wolf .aio_nbytes = bytes, 294958a209c4SKevin Wolf .copy_range = { 295058a209c4SKevin Wolf .aio_fd2 = s->fd, 295158a209c4SKevin Wolf .aio_offset2 = dst_offset, 295258a209c4SKevin Wolf }, 295358a209c4SKevin Wolf }; 295458a209c4SKevin Wolf 295558a209c4SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); 29561efad060SFam Zheng } 29571efad060SFam Zheng 2958c1bb86cdSEric Blake BlockDriver bdrv_file = { 2959c1bb86cdSEric Blake .format_name = "file", 2960c1bb86cdSEric Blake .protocol_name = "file", 2961c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 2962c1bb86cdSEric Blake .bdrv_needs_filename = true, 2963c1bb86cdSEric Blake .bdrv_probe = NULL, /* no probe for protocols */ 2964c1bb86cdSEric Blake .bdrv_parse_filename = raw_parse_filename, 2965c1bb86cdSEric Blake .bdrv_file_open = raw_open, 2966c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 2967c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 2968c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 2969c1bb86cdSEric Blake .bdrv_close = raw_close, 2970927f11e1SKevin Wolf .bdrv_co_create = raw_co_create, 2971efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = raw_co_create_opts, 2972c1bb86cdSEric Blake .bdrv_has_zero_init = bdrv_has_zero_init_1, 29731dcaf527SMax Reitz .bdrv_has_zero_init_truncate = bdrv_has_zero_init_1, 2974a290f085SEric Blake .bdrv_co_block_status = raw_co_block_status, 2975dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 2976c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, 2977c1bb86cdSEric Blake 2978c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 2979c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 298033d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 298133d70fb6SKevin Wolf .bdrv_co_pdiscard = raw_co_pdiscard, 29821efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 29831efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 2984c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 2985c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 2986c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 2987ed6e2161SNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 2988c1bb86cdSEric Blake 2989061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 2990c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 2991c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 2992c1bb86cdSEric Blake .bdrv_get_allocated_file_size 2993c1bb86cdSEric Blake = raw_get_allocated_file_size, 2994d9245599SAnton Nefedov .bdrv_get_specific_stats = raw_get_specific_stats, 2995244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 2996244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 2997244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 2998c1bb86cdSEric Blake .create_opts = &raw_create_opts, 29998a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3000c1bb86cdSEric Blake }; 3001c1bb86cdSEric Blake 3002c1bb86cdSEric Blake /***********************************************/ 3003c1bb86cdSEric Blake /* host device */ 3004c1bb86cdSEric Blake 3005c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3006c1bb86cdSEric Blake static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3007c1bb86cdSEric Blake CFIndex maxPathSize, int flags); 3008c1bb86cdSEric Blake static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) 3009c1bb86cdSEric Blake { 3010c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 3011c1bb86cdSEric Blake mach_port_t masterPort; 3012c1bb86cdSEric Blake CFMutableDictionaryRef classesToMatch; 3013c1bb86cdSEric Blake const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; 3014c1bb86cdSEric Blake char *mediaType = NULL; 3015c1bb86cdSEric Blake 3016c1bb86cdSEric Blake kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); 3017c1bb86cdSEric Blake if ( KERN_SUCCESS != kernResult ) { 3018c1bb86cdSEric Blake printf( "IOMasterPort returned %d\n", kernResult ); 3019c1bb86cdSEric Blake } 3020c1bb86cdSEric Blake 3021c1bb86cdSEric Blake int index; 3022c1bb86cdSEric Blake for (index = 0; index < ARRAY_SIZE(matching_array); index++) { 3023c1bb86cdSEric Blake classesToMatch = IOServiceMatching(matching_array[index]); 3024c1bb86cdSEric Blake if (classesToMatch == NULL) { 3025c1bb86cdSEric Blake error_report("IOServiceMatching returned NULL for %s", 3026c1bb86cdSEric Blake matching_array[index]); 3027c1bb86cdSEric Blake continue; 3028c1bb86cdSEric Blake } 3029c1bb86cdSEric Blake CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), 3030c1bb86cdSEric Blake kCFBooleanTrue); 3031c1bb86cdSEric Blake kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, 3032c1bb86cdSEric Blake mediaIterator); 3033c1bb86cdSEric Blake if (kernResult != KERN_SUCCESS) { 3034c1bb86cdSEric Blake error_report("Note: IOServiceGetMatchingServices returned %d", 3035c1bb86cdSEric Blake kernResult); 3036c1bb86cdSEric Blake continue; 3037c1bb86cdSEric Blake } 3038c1bb86cdSEric Blake 3039c1bb86cdSEric Blake /* If a match was found, leave the loop */ 3040c1bb86cdSEric Blake if (*mediaIterator != 0) { 30414f7d28d7SLaurent Vivier trace_file_FindEjectableOpticalMedia(matching_array[index]); 3042c1bb86cdSEric Blake mediaType = g_strdup(matching_array[index]); 3043c1bb86cdSEric Blake break; 3044c1bb86cdSEric Blake } 3045c1bb86cdSEric Blake } 3046c1bb86cdSEric Blake return mediaType; 3047c1bb86cdSEric Blake } 3048c1bb86cdSEric Blake 3049c1bb86cdSEric Blake kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3050c1bb86cdSEric Blake CFIndex maxPathSize, int flags) 3051c1bb86cdSEric Blake { 3052c1bb86cdSEric Blake io_object_t nextMedia; 3053c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 3054c1bb86cdSEric Blake *bsdPath = '\0'; 3055c1bb86cdSEric Blake nextMedia = IOIteratorNext( mediaIterator ); 3056c1bb86cdSEric Blake if ( nextMedia ) 3057c1bb86cdSEric Blake { 3058c1bb86cdSEric Blake CFTypeRef bsdPathAsCFString; 3059c1bb86cdSEric Blake bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); 3060c1bb86cdSEric Blake if ( bsdPathAsCFString ) { 3061c1bb86cdSEric Blake size_t devPathLength; 3062c1bb86cdSEric Blake strcpy( bsdPath, _PATH_DEV ); 3063c1bb86cdSEric Blake if (flags & BDRV_O_NOCACHE) { 3064c1bb86cdSEric Blake strcat(bsdPath, "r"); 3065c1bb86cdSEric Blake } 3066c1bb86cdSEric Blake devPathLength = strlen( bsdPath ); 3067c1bb86cdSEric Blake if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { 3068c1bb86cdSEric Blake kernResult = KERN_SUCCESS; 3069c1bb86cdSEric Blake } 3070c1bb86cdSEric Blake CFRelease( bsdPathAsCFString ); 3071c1bb86cdSEric Blake } 3072c1bb86cdSEric Blake IOObjectRelease( nextMedia ); 3073c1bb86cdSEric Blake } 3074c1bb86cdSEric Blake 3075c1bb86cdSEric Blake return kernResult; 3076c1bb86cdSEric Blake } 3077c1bb86cdSEric Blake 3078c1bb86cdSEric Blake /* Sets up a real cdrom for use in QEMU */ 3079c1bb86cdSEric Blake static bool setup_cdrom(char *bsd_path, Error **errp) 3080c1bb86cdSEric Blake { 3081c1bb86cdSEric Blake int index, num_of_test_partitions = 2, fd; 3082c1bb86cdSEric Blake char test_partition[MAXPATHLEN]; 3083c1bb86cdSEric Blake bool partition_found = false; 3084c1bb86cdSEric Blake 3085c1bb86cdSEric Blake /* look for a working partition */ 3086c1bb86cdSEric Blake for (index = 0; index < num_of_test_partitions; index++) { 3087c1bb86cdSEric Blake snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, 3088c1bb86cdSEric Blake index); 3089c1bb86cdSEric Blake fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE); 3090c1bb86cdSEric Blake if (fd >= 0) { 3091c1bb86cdSEric Blake partition_found = true; 3092c1bb86cdSEric Blake qemu_close(fd); 3093c1bb86cdSEric Blake break; 3094c1bb86cdSEric Blake } 3095c1bb86cdSEric Blake } 3096c1bb86cdSEric Blake 3097c1bb86cdSEric Blake /* if a working partition on the device was not found */ 3098c1bb86cdSEric Blake if (partition_found == false) { 3099c1bb86cdSEric Blake error_setg(errp, "Failed to find a working partition on disc"); 3100c1bb86cdSEric Blake } else { 31014f7d28d7SLaurent Vivier trace_file_setup_cdrom(test_partition); 3102c1bb86cdSEric Blake pstrcpy(bsd_path, MAXPATHLEN, test_partition); 3103c1bb86cdSEric Blake } 3104c1bb86cdSEric Blake return partition_found; 3105c1bb86cdSEric Blake } 3106c1bb86cdSEric Blake 3107c1bb86cdSEric Blake /* Prints directions on mounting and unmounting a device */ 3108c1bb86cdSEric Blake static void print_unmounting_directions(const char *file_name) 3109c1bb86cdSEric Blake { 3110c1bb86cdSEric Blake error_report("If device %s is mounted on the desktop, unmount" 3111c1bb86cdSEric Blake " it first before using it in QEMU", file_name); 3112c1bb86cdSEric Blake error_report("Command to unmount device: diskutil unmountDisk %s", 3113c1bb86cdSEric Blake file_name); 3114c1bb86cdSEric Blake error_report("Command to mount device: diskutil mountDisk %s", file_name); 3115c1bb86cdSEric Blake } 3116c1bb86cdSEric Blake 3117c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3118c1bb86cdSEric Blake 3119c1bb86cdSEric Blake static int hdev_probe_device(const char *filename) 3120c1bb86cdSEric Blake { 3121c1bb86cdSEric Blake struct stat st; 3122c1bb86cdSEric Blake 3123c1bb86cdSEric Blake /* allow a dedicated CD-ROM driver to match with a higher priority */ 3124c1bb86cdSEric Blake if (strstart(filename, "/dev/cdrom", NULL)) 3125c1bb86cdSEric Blake return 50; 3126c1bb86cdSEric Blake 3127c1bb86cdSEric Blake if (stat(filename, &st) >= 0 && 3128c1bb86cdSEric Blake (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 3129c1bb86cdSEric Blake return 100; 3130c1bb86cdSEric Blake } 3131c1bb86cdSEric Blake 3132c1bb86cdSEric Blake return 0; 3133c1bb86cdSEric Blake } 3134c1bb86cdSEric Blake 3135c1bb86cdSEric Blake static int check_hdev_writable(BDRVRawState *s) 3136c1bb86cdSEric Blake { 3137c1bb86cdSEric Blake #if defined(BLKROGET) 3138c1bb86cdSEric Blake /* Linux block devices can be configured "read-only" using blockdev(8). 3139c1bb86cdSEric Blake * This is independent of device node permissions and therefore open(2) 3140c1bb86cdSEric Blake * with O_RDWR succeeds. Actual writes fail with EPERM. 3141c1bb86cdSEric Blake * 3142c1bb86cdSEric Blake * bdrv_open() is supposed to fail if the disk is read-only. Explicitly 3143c1bb86cdSEric Blake * check for read-only block devices so that Linux block devices behave 3144c1bb86cdSEric Blake * properly. 3145c1bb86cdSEric Blake */ 3146c1bb86cdSEric Blake struct stat st; 3147c1bb86cdSEric Blake int readonly = 0; 3148c1bb86cdSEric Blake 3149c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 3150c1bb86cdSEric Blake return -errno; 3151c1bb86cdSEric Blake } 3152c1bb86cdSEric Blake 3153c1bb86cdSEric Blake if (!S_ISBLK(st.st_mode)) { 3154c1bb86cdSEric Blake return 0; 3155c1bb86cdSEric Blake } 3156c1bb86cdSEric Blake 3157c1bb86cdSEric Blake if (ioctl(s->fd, BLKROGET, &readonly) < 0) { 3158c1bb86cdSEric Blake return -errno; 3159c1bb86cdSEric Blake } 3160c1bb86cdSEric Blake 3161c1bb86cdSEric Blake if (readonly) { 3162c1bb86cdSEric Blake return -EACCES; 3163c1bb86cdSEric Blake } 3164c1bb86cdSEric Blake #endif /* defined(BLKROGET) */ 3165c1bb86cdSEric Blake return 0; 3166c1bb86cdSEric Blake } 3167c1bb86cdSEric Blake 3168c1bb86cdSEric Blake static void hdev_parse_filename(const char *filename, QDict *options, 3169c1bb86cdSEric Blake Error **errp) 3170c1bb86cdSEric Blake { 317103c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_device:", options); 3172c1bb86cdSEric Blake } 3173c1bb86cdSEric Blake 3174c1bb86cdSEric Blake static bool hdev_is_sg(BlockDriverState *bs) 3175c1bb86cdSEric Blake { 3176c1bb86cdSEric Blake 3177c1bb86cdSEric Blake #if defined(__linux__) 3178c1bb86cdSEric Blake 3179c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3180c1bb86cdSEric Blake struct stat st; 3181c1bb86cdSEric Blake struct sg_scsi_id scsiid; 3182c1bb86cdSEric Blake int sg_version; 3183c1bb86cdSEric Blake int ret; 3184c1bb86cdSEric Blake 3185c1bb86cdSEric Blake if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { 3186c1bb86cdSEric Blake return false; 3187c1bb86cdSEric Blake } 3188c1bb86cdSEric Blake 3189c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); 3190c1bb86cdSEric Blake if (ret < 0) { 3191c1bb86cdSEric Blake return false; 3192c1bb86cdSEric Blake } 3193c1bb86cdSEric Blake 3194c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); 3195c1bb86cdSEric Blake if (ret >= 0) { 31964f7d28d7SLaurent Vivier trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); 3197c1bb86cdSEric Blake return true; 3198c1bb86cdSEric Blake } 3199c1bb86cdSEric Blake 3200c1bb86cdSEric Blake #endif 3201c1bb86cdSEric Blake 3202c1bb86cdSEric Blake return false; 3203c1bb86cdSEric Blake } 3204c1bb86cdSEric Blake 3205c1bb86cdSEric Blake static int hdev_open(BlockDriverState *bs, QDict *options, int flags, 3206c1bb86cdSEric Blake Error **errp) 3207c1bb86cdSEric Blake { 3208c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3209c1bb86cdSEric Blake Error *local_err = NULL; 3210c1bb86cdSEric Blake int ret; 3211c1bb86cdSEric Blake 3212c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3213129c7d1cSMarkus Armbruster /* 3214129c7d1cSMarkus Armbruster * Caution: while qdict_get_str() is fine, getting non-string types 3215129c7d1cSMarkus Armbruster * would require more care. When @options come from -blockdev or 3216129c7d1cSMarkus Armbruster * blockdev_add, its members are typed according to the QAPI 3217129c7d1cSMarkus Armbruster * schema, but when they come from -drive, they're all QString. 3218129c7d1cSMarkus Armbruster */ 3219c1bb86cdSEric Blake const char *filename = qdict_get_str(options, "filename"); 3220c1bb86cdSEric Blake char bsd_path[MAXPATHLEN] = ""; 3221c1bb86cdSEric Blake bool error_occurred = false; 3222c1bb86cdSEric Blake 3223c1bb86cdSEric Blake /* If using a real cdrom */ 3224c1bb86cdSEric Blake if (strcmp(filename, "/dev/cdrom") == 0) { 3225c1bb86cdSEric Blake char *mediaType = NULL; 3226c1bb86cdSEric Blake kern_return_t ret_val; 3227c1bb86cdSEric Blake io_iterator_t mediaIterator = 0; 3228c1bb86cdSEric Blake 3229c1bb86cdSEric Blake mediaType = FindEjectableOpticalMedia(&mediaIterator); 3230c1bb86cdSEric Blake if (mediaType == NULL) { 3231c1bb86cdSEric Blake error_setg(errp, "Please make sure your CD/DVD is in the optical" 3232c1bb86cdSEric Blake " drive"); 3233c1bb86cdSEric Blake error_occurred = true; 3234c1bb86cdSEric Blake goto hdev_open_Mac_error; 3235c1bb86cdSEric Blake } 3236c1bb86cdSEric Blake 3237c1bb86cdSEric Blake ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); 3238c1bb86cdSEric Blake if (ret_val != KERN_SUCCESS) { 3239c1bb86cdSEric Blake error_setg(errp, "Could not get BSD path for optical drive"); 3240c1bb86cdSEric Blake error_occurred = true; 3241c1bb86cdSEric Blake goto hdev_open_Mac_error; 3242c1bb86cdSEric Blake } 3243c1bb86cdSEric Blake 3244c1bb86cdSEric Blake /* If a real optical drive was not found */ 3245c1bb86cdSEric Blake if (bsd_path[0] == '\0') { 3246c1bb86cdSEric Blake error_setg(errp, "Failed to obtain bsd path for optical drive"); 3247c1bb86cdSEric Blake error_occurred = true; 3248c1bb86cdSEric Blake goto hdev_open_Mac_error; 3249c1bb86cdSEric Blake } 3250c1bb86cdSEric Blake 3251c1bb86cdSEric Blake /* If using a cdrom disc and finding a partition on the disc failed */ 3252c1bb86cdSEric Blake if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && 3253c1bb86cdSEric Blake setup_cdrom(bsd_path, errp) == false) { 3254c1bb86cdSEric Blake print_unmounting_directions(bsd_path); 3255c1bb86cdSEric Blake error_occurred = true; 3256c1bb86cdSEric Blake goto hdev_open_Mac_error; 3257c1bb86cdSEric Blake } 3258c1bb86cdSEric Blake 325946f5ac20SEric Blake qdict_put_str(options, "filename", bsd_path); 3260c1bb86cdSEric Blake 3261c1bb86cdSEric Blake hdev_open_Mac_error: 3262c1bb86cdSEric Blake g_free(mediaType); 3263c1bb86cdSEric Blake if (mediaIterator) { 3264c1bb86cdSEric Blake IOObjectRelease(mediaIterator); 3265c1bb86cdSEric Blake } 3266c1bb86cdSEric Blake if (error_occurred) { 3267c1bb86cdSEric Blake return -ENOENT; 3268c1bb86cdSEric Blake } 3269c1bb86cdSEric Blake } 3270c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3271c1bb86cdSEric Blake 3272c1bb86cdSEric Blake s->type = FTYPE_FILE; 3273c1bb86cdSEric Blake 3274230ff739SJohn Snow ret = raw_open_common(bs, options, flags, 0, true, &local_err); 3275c1bb86cdSEric Blake if (ret < 0) { 3276c1bb86cdSEric Blake error_propagate(errp, local_err); 3277c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3278c1bb86cdSEric Blake if (*bsd_path) { 3279c1bb86cdSEric Blake filename = bsd_path; 3280c1bb86cdSEric Blake } 3281c1bb86cdSEric Blake /* if a physical device experienced an error while being opened */ 3282c1bb86cdSEric Blake if (strncmp(filename, "/dev/", 5) == 0) { 3283c1bb86cdSEric Blake print_unmounting_directions(filename); 3284c1bb86cdSEric Blake } 3285c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3286c1bb86cdSEric Blake return ret; 3287c1bb86cdSEric Blake } 3288c1bb86cdSEric Blake 3289c1bb86cdSEric Blake /* Since this does ioctl the device must be already opened */ 3290c1bb86cdSEric Blake bs->sg = hdev_is_sg(bs); 3291c1bb86cdSEric Blake 3292c1bb86cdSEric Blake if (flags & BDRV_O_RDWR) { 3293c1bb86cdSEric Blake ret = check_hdev_writable(s); 3294c1bb86cdSEric Blake if (ret < 0) { 3295c1bb86cdSEric Blake raw_close(bs); 3296c1bb86cdSEric Blake error_setg_errno(errp, -ret, "The device is not writable"); 3297c1bb86cdSEric Blake return ret; 3298c1bb86cdSEric Blake } 3299c1bb86cdSEric Blake } 3300c1bb86cdSEric Blake 3301c1bb86cdSEric Blake return ret; 3302c1bb86cdSEric Blake } 3303c1bb86cdSEric Blake 3304c1bb86cdSEric Blake #if defined(__linux__) 33052f3a7ab3SKevin Wolf static int coroutine_fn 33062f3a7ab3SKevin Wolf hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) 3307c1bb86cdSEric Blake { 3308c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 330903425671SKevin Wolf RawPosixAIOData acb; 33102f3a7ab3SKevin Wolf int ret; 3311c1bb86cdSEric Blake 33122f3a7ab3SKevin Wolf ret = fd_open(bs); 33132f3a7ab3SKevin Wolf if (ret < 0) { 33142f3a7ab3SKevin Wolf return ret; 33152f3a7ab3SKevin Wolf } 3316c1bb86cdSEric Blake 33177c9e5276SPaolo Bonzini if (req == SG_IO && s->pr_mgr) { 33187c9e5276SPaolo Bonzini struct sg_io_hdr *io_hdr = buf; 33197c9e5276SPaolo Bonzini if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || 33207c9e5276SPaolo Bonzini io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { 33217c9e5276SPaolo Bonzini return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), 33222f3a7ab3SKevin Wolf s->fd, io_hdr); 33237c9e5276SPaolo Bonzini } 33247c9e5276SPaolo Bonzini } 33257c9e5276SPaolo Bonzini 332603425671SKevin Wolf acb = (RawPosixAIOData) { 332703425671SKevin Wolf .bs = bs, 332803425671SKevin Wolf .aio_type = QEMU_AIO_IOCTL, 332903425671SKevin Wolf .aio_fildes = s->fd, 333003425671SKevin Wolf .aio_offset = 0, 333103425671SKevin Wolf .ioctl = { 333203425671SKevin Wolf .buf = buf, 333303425671SKevin Wolf .cmd = req, 333403425671SKevin Wolf }, 333503425671SKevin Wolf }; 333603425671SKevin Wolf 333703425671SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); 3338c1bb86cdSEric Blake } 3339c1bb86cdSEric Blake #endif /* linux */ 3340c1bb86cdSEric Blake 3341c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs) 3342c1bb86cdSEric Blake { 3343c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3344c1bb86cdSEric Blake 3345c1bb86cdSEric Blake /* this is just to ensure s->fd is sane (its called by io ops) */ 3346c1bb86cdSEric Blake if (s->fd >= 0) 3347c1bb86cdSEric Blake return 0; 3348c1bb86cdSEric Blake return -EIO; 3349c1bb86cdSEric Blake } 3350c1bb86cdSEric Blake 335133d70fb6SKevin Wolf static coroutine_fn int 335233d70fb6SKevin Wolf hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 3353c1bb86cdSEric Blake { 33541c450366SAnton Nefedov BDRVRawState *s = bs->opaque; 335533d70fb6SKevin Wolf int ret; 3356c1bb86cdSEric Blake 335733d70fb6SKevin Wolf ret = fd_open(bs); 335833d70fb6SKevin Wolf if (ret < 0) { 33591c450366SAnton Nefedov raw_account_discard(s, bytes, ret); 336033d70fb6SKevin Wolf return ret; 3361c1bb86cdSEric Blake } 336246ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, true); 3363c1bb86cdSEric Blake } 3364c1bb86cdSEric Blake 3365c1bb86cdSEric Blake static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, 3366f5a5ca79SManos Pitsidianakis int64_t offset, int bytes, BdrvRequestFlags flags) 3367c1bb86cdSEric Blake { 3368c1bb86cdSEric Blake int rc; 3369c1bb86cdSEric Blake 3370c1bb86cdSEric Blake rc = fd_open(bs); 3371c1bb86cdSEric Blake if (rc < 0) { 3372c1bb86cdSEric Blake return rc; 3373c1bb86cdSEric Blake } 337434fa110eSKevin Wolf 33757154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); 3376c1bb86cdSEric Blake } 3377c1bb86cdSEric Blake 3378efc75e2aSStefan Hajnoczi static int coroutine_fn hdev_co_create_opts(const char *filename, QemuOpts *opts, 3379c1bb86cdSEric Blake Error **errp) 3380c1bb86cdSEric Blake { 3381c1bb86cdSEric Blake int fd; 3382c1bb86cdSEric Blake int ret = 0; 3383c1bb86cdSEric Blake struct stat stat_buf; 3384c1bb86cdSEric Blake int64_t total_size = 0; 3385c1bb86cdSEric Blake bool has_prefix; 3386c1bb86cdSEric Blake 3387c1bb86cdSEric Blake /* This function is used by both protocol block drivers and therefore either 3388c1bb86cdSEric Blake * of these prefixes may be given. 3389c1bb86cdSEric Blake * The return value has to be stored somewhere, otherwise this is an error 3390c1bb86cdSEric Blake * due to -Werror=unused-value. */ 3391c1bb86cdSEric Blake has_prefix = 3392c1bb86cdSEric Blake strstart(filename, "host_device:", &filename) || 3393c1bb86cdSEric Blake strstart(filename, "host_cdrom:" , &filename); 3394c1bb86cdSEric Blake 3395c1bb86cdSEric Blake (void)has_prefix; 3396c1bb86cdSEric Blake 3397db0754dfSFam Zheng ret = raw_normalize_devicepath(&filename, errp); 3398c1bb86cdSEric Blake if (ret < 0) { 3399c1bb86cdSEric Blake return ret; 3400c1bb86cdSEric Blake } 3401c1bb86cdSEric Blake 3402c1bb86cdSEric Blake /* Read out options */ 3403c1bb86cdSEric Blake total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 3404c1bb86cdSEric Blake BDRV_SECTOR_SIZE); 3405c1bb86cdSEric Blake 3406c1bb86cdSEric Blake fd = qemu_open(filename, O_WRONLY | O_BINARY); 3407c1bb86cdSEric Blake if (fd < 0) { 3408c1bb86cdSEric Blake ret = -errno; 3409c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not open device"); 3410c1bb86cdSEric Blake return ret; 3411c1bb86cdSEric Blake } 3412c1bb86cdSEric Blake 3413c1bb86cdSEric Blake if (fstat(fd, &stat_buf) < 0) { 3414c1bb86cdSEric Blake ret = -errno; 3415c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not stat device"); 3416c1bb86cdSEric Blake } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) { 3417c1bb86cdSEric Blake error_setg(errp, 3418c1bb86cdSEric Blake "The given file is neither a block nor a character device"); 3419c1bb86cdSEric Blake ret = -ENODEV; 3420c1bb86cdSEric Blake } else if (lseek(fd, 0, SEEK_END) < total_size) { 3421c1bb86cdSEric Blake error_setg(errp, "Device is too small"); 3422c1bb86cdSEric Blake ret = -ENOSPC; 3423c1bb86cdSEric Blake } 3424c1bb86cdSEric Blake 342597ec9117SFam Zheng if (!ret && total_size) { 342697ec9117SFam Zheng uint8_t buf[BDRV_SECTOR_SIZE] = { 0 }; 342797ec9117SFam Zheng int64_t zero_size = MIN(BDRV_SECTOR_SIZE, total_size); 342897ec9117SFam Zheng if (lseek(fd, 0, SEEK_SET) == -1) { 342997ec9117SFam Zheng ret = -errno; 343097ec9117SFam Zheng } else { 343197ec9117SFam Zheng ret = qemu_write_full(fd, buf, zero_size); 343297ec9117SFam Zheng ret = ret == zero_size ? 0 : -errno; 343397ec9117SFam Zheng } 343497ec9117SFam Zheng } 3435c1bb86cdSEric Blake qemu_close(fd); 3436c1bb86cdSEric Blake return ret; 3437c1bb86cdSEric Blake } 3438c1bb86cdSEric Blake 3439c1bb86cdSEric Blake static BlockDriver bdrv_host_device = { 3440c1bb86cdSEric Blake .format_name = "host_device", 3441c1bb86cdSEric Blake .protocol_name = "host_device", 3442c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3443c1bb86cdSEric Blake .bdrv_needs_filename = true, 3444c1bb86cdSEric Blake .bdrv_probe_device = hdev_probe_device, 3445c1bb86cdSEric Blake .bdrv_parse_filename = hdev_parse_filename, 3446c1bb86cdSEric Blake .bdrv_file_open = hdev_open, 3447c1bb86cdSEric Blake .bdrv_close = raw_close, 3448c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3449c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3450c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3451efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = hdev_co_create_opts, 3452c1bb86cdSEric Blake .create_opts = &raw_create_opts, 34538a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3454dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3455c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, 3456c1bb86cdSEric Blake 3457c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3458c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 345933d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 346033d70fb6SKevin Wolf .bdrv_co_pdiscard = hdev_co_pdiscard, 34611efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 34621efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 3463c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3464c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3465c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3466042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3467c1bb86cdSEric Blake 3468061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3469c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3470c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 3471c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3472c1bb86cdSEric Blake = raw_get_allocated_file_size, 3473d9245599SAnton Nefedov .bdrv_get_specific_stats = hdev_get_specific_stats, 3474244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 3475244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 3476244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 3477c1bb86cdSEric Blake .bdrv_probe_blocksizes = hdev_probe_blocksizes, 3478c1bb86cdSEric Blake .bdrv_probe_geometry = hdev_probe_geometry, 3479c1bb86cdSEric Blake 3480c1bb86cdSEric Blake /* generic scsi device */ 3481c1bb86cdSEric Blake #ifdef __linux__ 34822f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3483c1bb86cdSEric Blake #endif 3484c1bb86cdSEric Blake }; 3485c1bb86cdSEric Blake 3486c1bb86cdSEric Blake #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3487c1bb86cdSEric Blake static void cdrom_parse_filename(const char *filename, QDict *options, 3488c1bb86cdSEric Blake Error **errp) 3489c1bb86cdSEric Blake { 349003c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); 3491c1bb86cdSEric Blake } 3492c1bb86cdSEric Blake #endif 3493c1bb86cdSEric Blake 3494c1bb86cdSEric Blake #ifdef __linux__ 3495c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3496c1bb86cdSEric Blake Error **errp) 3497c1bb86cdSEric Blake { 3498c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3499c1bb86cdSEric Blake 3500c1bb86cdSEric Blake s->type = FTYPE_CD; 3501c1bb86cdSEric Blake 3502c1bb86cdSEric Blake /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ 3503230ff739SJohn Snow return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); 3504c1bb86cdSEric Blake } 3505c1bb86cdSEric Blake 3506c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3507c1bb86cdSEric Blake { 3508c1bb86cdSEric Blake int fd, ret; 3509c1bb86cdSEric Blake int prio = 0; 3510c1bb86cdSEric Blake struct stat st; 3511c1bb86cdSEric Blake 3512c1bb86cdSEric Blake fd = qemu_open(filename, O_RDONLY | O_NONBLOCK); 3513c1bb86cdSEric Blake if (fd < 0) { 3514c1bb86cdSEric Blake goto out; 3515c1bb86cdSEric Blake } 3516c1bb86cdSEric Blake ret = fstat(fd, &st); 3517c1bb86cdSEric Blake if (ret == -1 || !S_ISBLK(st.st_mode)) { 3518c1bb86cdSEric Blake goto outc; 3519c1bb86cdSEric Blake } 3520c1bb86cdSEric Blake 3521c1bb86cdSEric Blake /* Attempt to detect via a CDROM specific ioctl */ 3522c1bb86cdSEric Blake ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3523c1bb86cdSEric Blake if (ret >= 0) 3524c1bb86cdSEric Blake prio = 100; 3525c1bb86cdSEric Blake 3526c1bb86cdSEric Blake outc: 3527c1bb86cdSEric Blake qemu_close(fd); 3528c1bb86cdSEric Blake out: 3529c1bb86cdSEric Blake return prio; 3530c1bb86cdSEric Blake } 3531c1bb86cdSEric Blake 3532c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3533c1bb86cdSEric Blake { 3534c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3535c1bb86cdSEric Blake int ret; 3536c1bb86cdSEric Blake 3537c1bb86cdSEric Blake ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3538c1bb86cdSEric Blake return ret == CDS_DISC_OK; 3539c1bb86cdSEric Blake } 3540c1bb86cdSEric Blake 3541c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3542c1bb86cdSEric Blake { 3543c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3544c1bb86cdSEric Blake 3545c1bb86cdSEric Blake if (eject_flag) { 3546c1bb86cdSEric Blake if (ioctl(s->fd, CDROMEJECT, NULL) < 0) 3547c1bb86cdSEric Blake perror("CDROMEJECT"); 3548c1bb86cdSEric Blake } else { 3549c1bb86cdSEric Blake if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) 3550c1bb86cdSEric Blake perror("CDROMEJECT"); 3551c1bb86cdSEric Blake } 3552c1bb86cdSEric Blake } 3553c1bb86cdSEric Blake 3554c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3555c1bb86cdSEric Blake { 3556c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3557c1bb86cdSEric Blake 3558c1bb86cdSEric Blake if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { 3559c1bb86cdSEric Blake /* 3560c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3561c1bb86cdSEric Blake * mounts the CD-ROM 3562c1bb86cdSEric Blake */ 3563c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3564c1bb86cdSEric Blake } 3565c1bb86cdSEric Blake } 3566c1bb86cdSEric Blake 3567c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3568c1bb86cdSEric Blake .format_name = "host_cdrom", 3569c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3570c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3571c1bb86cdSEric Blake .bdrv_needs_filename = true, 3572c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3573c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3574c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3575c1bb86cdSEric Blake .bdrv_close = raw_close, 3576c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3577c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3578c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3579efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = hdev_co_create_opts, 3580c1bb86cdSEric Blake .create_opts = &raw_create_opts, 35818a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3582dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3583c1bb86cdSEric Blake 3584c1bb86cdSEric Blake 3585c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3586c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 358733d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3588c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3589c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3590c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3591042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3592c1bb86cdSEric Blake 3593061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3594c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3595c1bb86cdSEric Blake .has_variable_length = true, 3596c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3597c1bb86cdSEric Blake = raw_get_allocated_file_size, 3598c1bb86cdSEric Blake 3599c1bb86cdSEric Blake /* removable device support */ 3600c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3601c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3602c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3603c1bb86cdSEric Blake 3604c1bb86cdSEric Blake /* generic scsi device */ 36052f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3606c1bb86cdSEric Blake }; 3607c1bb86cdSEric Blake #endif /* __linux__ */ 3608c1bb86cdSEric Blake 3609c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 3610c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3611c1bb86cdSEric Blake Error **errp) 3612c1bb86cdSEric Blake { 3613c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3614c1bb86cdSEric Blake Error *local_err = NULL; 3615c1bb86cdSEric Blake int ret; 3616c1bb86cdSEric Blake 3617c1bb86cdSEric Blake s->type = FTYPE_CD; 3618c1bb86cdSEric Blake 3619230ff739SJohn Snow ret = raw_open_common(bs, options, flags, 0, true, &local_err); 3620c1bb86cdSEric Blake if (ret) { 3621c1bb86cdSEric Blake error_propagate(errp, local_err); 3622c1bb86cdSEric Blake return ret; 3623c1bb86cdSEric Blake } 3624c1bb86cdSEric Blake 3625c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3626c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3627c1bb86cdSEric Blake return 0; 3628c1bb86cdSEric Blake } 3629c1bb86cdSEric Blake 3630c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3631c1bb86cdSEric Blake { 3632c1bb86cdSEric Blake if (strstart(filename, "/dev/cd", NULL) || 3633c1bb86cdSEric Blake strstart(filename, "/dev/acd", NULL)) 3634c1bb86cdSEric Blake return 100; 3635c1bb86cdSEric Blake return 0; 3636c1bb86cdSEric Blake } 3637c1bb86cdSEric Blake 3638c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs) 3639c1bb86cdSEric Blake { 3640c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3641c1bb86cdSEric Blake int fd; 3642c1bb86cdSEric Blake 3643c1bb86cdSEric Blake /* 3644c1bb86cdSEric Blake * Force reread of possibly changed/newly loaded disc, 3645c1bb86cdSEric Blake * FreeBSD seems to not notice sometimes... 3646c1bb86cdSEric Blake */ 3647c1bb86cdSEric Blake if (s->fd >= 0) 3648c1bb86cdSEric Blake qemu_close(s->fd); 3649c1bb86cdSEric Blake fd = qemu_open(bs->filename, s->open_flags, 0644); 3650c1bb86cdSEric Blake if (fd < 0) { 3651c1bb86cdSEric Blake s->fd = -1; 3652c1bb86cdSEric Blake return -EIO; 3653c1bb86cdSEric Blake } 3654c1bb86cdSEric Blake s->fd = fd; 3655c1bb86cdSEric Blake 3656c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3657c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3658c1bb86cdSEric Blake return 0; 3659c1bb86cdSEric Blake } 3660c1bb86cdSEric Blake 3661c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3662c1bb86cdSEric Blake { 3663c1bb86cdSEric Blake return raw_getlength(bs) > 0; 3664c1bb86cdSEric Blake } 3665c1bb86cdSEric Blake 3666c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3667c1bb86cdSEric Blake { 3668c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3669c1bb86cdSEric Blake 3670c1bb86cdSEric Blake if (s->fd < 0) 3671c1bb86cdSEric Blake return; 3672c1bb86cdSEric Blake 3673c1bb86cdSEric Blake (void) ioctl(s->fd, CDIOCALLOW); 3674c1bb86cdSEric Blake 3675c1bb86cdSEric Blake if (eject_flag) { 3676c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCEJECT) < 0) 3677c1bb86cdSEric Blake perror("CDIOCEJECT"); 3678c1bb86cdSEric Blake } else { 3679c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCCLOSE) < 0) 3680c1bb86cdSEric Blake perror("CDIOCCLOSE"); 3681c1bb86cdSEric Blake } 3682c1bb86cdSEric Blake 3683c1bb86cdSEric Blake cdrom_reopen(bs); 3684c1bb86cdSEric Blake } 3685c1bb86cdSEric Blake 3686c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3687c1bb86cdSEric Blake { 3688c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3689c1bb86cdSEric Blake 3690c1bb86cdSEric Blake if (s->fd < 0) 3691c1bb86cdSEric Blake return; 3692c1bb86cdSEric Blake if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { 3693c1bb86cdSEric Blake /* 3694c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3695c1bb86cdSEric Blake * mounts the CD-ROM 3696c1bb86cdSEric Blake */ 3697c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3698c1bb86cdSEric Blake } 3699c1bb86cdSEric Blake } 3700c1bb86cdSEric Blake 3701c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3702c1bb86cdSEric Blake .format_name = "host_cdrom", 3703c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3704c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3705c1bb86cdSEric Blake .bdrv_needs_filename = true, 3706c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3707c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3708c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3709c1bb86cdSEric Blake .bdrv_close = raw_close, 3710c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3711c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3712c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3713efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = hdev_co_create_opts, 3714c1bb86cdSEric Blake .create_opts = &raw_create_opts, 37158a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3716c1bb86cdSEric Blake 3717c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3718c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 371933d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3720c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3721c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3722c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3723042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3724c1bb86cdSEric Blake 3725061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3726c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3727c1bb86cdSEric Blake .has_variable_length = true, 3728c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3729c1bb86cdSEric Blake = raw_get_allocated_file_size, 3730c1bb86cdSEric Blake 3731c1bb86cdSEric Blake /* removable device support */ 3732c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3733c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3734c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3735c1bb86cdSEric Blake }; 3736c1bb86cdSEric Blake #endif /* __FreeBSD__ */ 3737c1bb86cdSEric Blake 3738c1bb86cdSEric Blake static void bdrv_file_init(void) 3739c1bb86cdSEric Blake { 3740c1bb86cdSEric Blake /* 3741c1bb86cdSEric Blake * Register all the drivers. Note that order is important, the driver 3742c1bb86cdSEric Blake * registered last will get probed first. 3743c1bb86cdSEric Blake */ 3744c1bb86cdSEric Blake bdrv_register(&bdrv_file); 3745c1bb86cdSEric Blake bdrv_register(&bdrv_host_device); 3746c1bb86cdSEric Blake #ifdef __linux__ 3747c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3748c1bb86cdSEric Blake #endif 3749c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3750c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3751c1bb86cdSEric Blake #endif 3752c1bb86cdSEric Blake } 3753c1bb86cdSEric Blake 3754c1bb86cdSEric Blake block_init(bdrv_file_init); 3755