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" 26c1bb86cdSEric Blake #include "qapi/error.h" 27c1bb86cdSEric Blake #include "qemu/cutils.h" 28c1bb86cdSEric Blake #include "qemu/error-report.h" 29c1bb86cdSEric Blake #include "block/block_int.h" 30c1bb86cdSEric Blake #include "qemu/module.h" 31922a01a0SMarkus Armbruster #include "qemu/option.h" 32c1bb86cdSEric Blake #include "trace.h" 33c1bb86cdSEric Blake #include "block/thread-pool.h" 34c1bb86cdSEric Blake #include "qemu/iov.h" 35c1bb86cdSEric Blake #include "block/raw-aio.h" 36452fcdbcSMarkus Armbruster #include "qapi/qmp/qdict.h" 37c1bb86cdSEric Blake #include "qapi/qmp/qstring.h" 38c1bb86cdSEric Blake 397c9e5276SPaolo Bonzini #include "scsi/pr-manager.h" 407c9e5276SPaolo Bonzini #include "scsi/constants.h" 417c9e5276SPaolo Bonzini 42c1bb86cdSEric Blake #if defined(__APPLE__) && (__MACH__) 43c1bb86cdSEric Blake #include <paths.h> 44c1bb86cdSEric Blake #include <sys/param.h> 45c1bb86cdSEric Blake #include <IOKit/IOKitLib.h> 46c1bb86cdSEric Blake #include <IOKit/IOBSD.h> 47c1bb86cdSEric Blake #include <IOKit/storage/IOMediaBSDClient.h> 48c1bb86cdSEric Blake #include <IOKit/storage/IOMedia.h> 49c1bb86cdSEric Blake #include <IOKit/storage/IOCDMedia.h> 50c1bb86cdSEric Blake //#include <IOKit/storage/IOCDTypes.h> 51c1bb86cdSEric Blake #include <IOKit/storage/IODVDMedia.h> 52c1bb86cdSEric Blake #include <CoreFoundation/CoreFoundation.h> 53c1bb86cdSEric Blake #endif 54c1bb86cdSEric Blake 55c1bb86cdSEric Blake #ifdef __sun__ 56c1bb86cdSEric Blake #define _POSIX_PTHREAD_SEMANTICS 1 57c1bb86cdSEric Blake #include <sys/dkio.h> 58c1bb86cdSEric Blake #endif 59c1bb86cdSEric Blake #ifdef __linux__ 60c1bb86cdSEric Blake #include <sys/ioctl.h> 61c1bb86cdSEric Blake #include <sys/param.h> 621efad060SFam Zheng #include <sys/syscall.h> 63c1bb86cdSEric Blake #include <linux/cdrom.h> 64c1bb86cdSEric Blake #include <linux/fd.h> 65c1bb86cdSEric Blake #include <linux/fs.h> 66c1bb86cdSEric Blake #include <linux/hdreg.h> 67c1bb86cdSEric Blake #include <scsi/sg.h> 68c1bb86cdSEric Blake #ifdef __s390__ 69c1bb86cdSEric Blake #include <asm/dasd.h> 70c1bb86cdSEric Blake #endif 71c1bb86cdSEric Blake #ifndef FS_NOCOW_FL 72c1bb86cdSEric Blake #define FS_NOCOW_FL 0x00800000 /* Do not cow file */ 73c1bb86cdSEric Blake #endif 74c1bb86cdSEric Blake #endif 75c1bb86cdSEric Blake #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE) 76c1bb86cdSEric Blake #include <linux/falloc.h> 77c1bb86cdSEric Blake #endif 78c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 79c1bb86cdSEric Blake #include <sys/disk.h> 80c1bb86cdSEric Blake #include <sys/cdio.h> 81c1bb86cdSEric Blake #endif 82c1bb86cdSEric Blake 83c1bb86cdSEric Blake #ifdef __OpenBSD__ 84c1bb86cdSEric Blake #include <sys/ioctl.h> 85c1bb86cdSEric Blake #include <sys/disklabel.h> 86c1bb86cdSEric Blake #include <sys/dkio.h> 87c1bb86cdSEric Blake #endif 88c1bb86cdSEric Blake 89c1bb86cdSEric Blake #ifdef __NetBSD__ 90c1bb86cdSEric Blake #include <sys/ioctl.h> 91c1bb86cdSEric Blake #include <sys/disklabel.h> 92c1bb86cdSEric Blake #include <sys/dkio.h> 93c1bb86cdSEric Blake #include <sys/disk.h> 94c1bb86cdSEric Blake #endif 95c1bb86cdSEric Blake 96c1bb86cdSEric Blake #ifdef __DragonFly__ 97c1bb86cdSEric Blake #include <sys/ioctl.h> 98c1bb86cdSEric Blake #include <sys/diskslice.h> 99c1bb86cdSEric Blake #endif 100c1bb86cdSEric Blake 101c1bb86cdSEric Blake #ifdef CONFIG_XFS 102c1bb86cdSEric Blake #include <xfs/xfs.h> 103c1bb86cdSEric Blake #endif 104c1bb86cdSEric Blake 105c1bb86cdSEric Blake //#define DEBUG_BLOCK 106c1bb86cdSEric Blake 107c1bb86cdSEric Blake #ifdef DEBUG_BLOCK 108c1bb86cdSEric Blake # define DEBUG_BLOCK_PRINT 1 109c1bb86cdSEric Blake #else 110c1bb86cdSEric Blake # define DEBUG_BLOCK_PRINT 0 111c1bb86cdSEric Blake #endif 112c1bb86cdSEric Blake #define DPRINTF(fmt, ...) \ 113c1bb86cdSEric Blake do { \ 114c1bb86cdSEric Blake if (DEBUG_BLOCK_PRINT) { \ 115c1bb86cdSEric Blake printf(fmt, ## __VA_ARGS__); \ 116c1bb86cdSEric Blake } \ 117c1bb86cdSEric Blake } while (0) 118c1bb86cdSEric Blake 119c1bb86cdSEric Blake /* OS X does not have O_DSYNC */ 120c1bb86cdSEric Blake #ifndef O_DSYNC 121c1bb86cdSEric Blake #ifdef O_SYNC 122c1bb86cdSEric Blake #define O_DSYNC O_SYNC 123c1bb86cdSEric Blake #elif defined(O_FSYNC) 124c1bb86cdSEric Blake #define O_DSYNC O_FSYNC 125c1bb86cdSEric Blake #endif 126c1bb86cdSEric Blake #endif 127c1bb86cdSEric Blake 128c1bb86cdSEric Blake /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ 129c1bb86cdSEric Blake #ifndef O_DIRECT 130c1bb86cdSEric Blake #define O_DIRECT O_DSYNC 131c1bb86cdSEric Blake #endif 132c1bb86cdSEric Blake 133c1bb86cdSEric Blake #define FTYPE_FILE 0 134c1bb86cdSEric Blake #define FTYPE_CD 1 135c1bb86cdSEric Blake 136c1bb86cdSEric Blake #define MAX_BLOCKSIZE 4096 137c1bb86cdSEric Blake 138244a5668SFam Zheng /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes, 139244a5668SFam Zheng * leaving a few more bytes for its future use. */ 140244a5668SFam Zheng #define RAW_LOCK_PERM_BASE 100 141244a5668SFam Zheng #define RAW_LOCK_SHARED_BASE 200 142244a5668SFam Zheng 143c1bb86cdSEric Blake typedef struct BDRVRawState { 144c1bb86cdSEric Blake int fd; 145244a5668SFam Zheng int lock_fd; 146244a5668SFam Zheng bool use_lock; 147c1bb86cdSEric Blake int type; 148c1bb86cdSEric Blake int open_flags; 149c1bb86cdSEric Blake size_t buf_align; 150c1bb86cdSEric Blake 151244a5668SFam Zheng /* The current permissions. */ 152244a5668SFam Zheng uint64_t perm; 153244a5668SFam Zheng uint64_t shared_perm; 154244a5668SFam Zheng 155c1bb86cdSEric Blake #ifdef CONFIG_XFS 156c1bb86cdSEric Blake bool is_xfs:1; 157c1bb86cdSEric Blake #endif 158c1bb86cdSEric Blake bool has_discard:1; 159c1bb86cdSEric Blake bool has_write_zeroes:1; 160c1bb86cdSEric Blake bool discard_zeroes:1; 161c1bb86cdSEric Blake bool use_linux_aio:1; 162e5bcf967SKevin Wolf bool page_cache_inconsistent:1; 163c1bb86cdSEric Blake bool has_fallocate; 164c1bb86cdSEric Blake bool needs_alignment; 16531be8a2aSStefan Hajnoczi bool check_cache_dropped; 1667c9e5276SPaolo Bonzini 1677c9e5276SPaolo Bonzini PRManager *pr_mgr; 168c1bb86cdSEric Blake } BDRVRawState; 169c1bb86cdSEric Blake 170c1bb86cdSEric Blake typedef struct BDRVRawReopenState { 171c1bb86cdSEric Blake int fd; 172c1bb86cdSEric Blake int open_flags; 17331be8a2aSStefan Hajnoczi bool check_cache_dropped; 174c1bb86cdSEric Blake } BDRVRawReopenState; 175c1bb86cdSEric Blake 176c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs); 177c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs); 178c1bb86cdSEric Blake 179c1bb86cdSEric Blake typedef struct RawPosixAIOData { 180c1bb86cdSEric Blake BlockDriverState *bs; 181c1bb86cdSEric Blake int aio_fildes; 182c1bb86cdSEric Blake union { 183c1bb86cdSEric Blake struct iovec *aio_iov; 184c1bb86cdSEric Blake void *aio_ioctl_buf; 185c1bb86cdSEric Blake }; 186c1bb86cdSEric Blake int aio_niov; 187c1bb86cdSEric Blake uint64_t aio_nbytes; 188c1bb86cdSEric Blake #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */ 189c1bb86cdSEric Blake off_t aio_offset; 190c1bb86cdSEric Blake int aio_type; 19193f4e2ffSKevin Wolf union { 19293f4e2ffSKevin Wolf struct { 1931efad060SFam Zheng int aio_fd2; 1941efad060SFam Zheng off_t aio_offset2; 19593f4e2ffSKevin Wolf }; 19693f4e2ffSKevin Wolf struct { 19793f4e2ffSKevin Wolf PreallocMode prealloc; 19893f4e2ffSKevin Wolf Error **errp; 19993f4e2ffSKevin Wolf }; 20093f4e2ffSKevin Wolf }; 201c1bb86cdSEric Blake } RawPosixAIOData; 202c1bb86cdSEric Blake 203c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 204c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs); 205c1bb86cdSEric Blake #endif 206c1bb86cdSEric Blake 207c1bb86cdSEric Blake #if defined(__NetBSD__) 208c1bb86cdSEric Blake static int raw_normalize_devicepath(const char **filename) 209c1bb86cdSEric Blake { 210c1bb86cdSEric Blake static char namebuf[PATH_MAX]; 211c1bb86cdSEric Blake const char *dp, *fname; 212c1bb86cdSEric Blake struct stat sb; 213c1bb86cdSEric Blake 214c1bb86cdSEric Blake fname = *filename; 215c1bb86cdSEric Blake dp = strrchr(fname, '/'); 216c1bb86cdSEric Blake if (lstat(fname, &sb) < 0) { 217c1bb86cdSEric Blake fprintf(stderr, "%s: stat failed: %s\n", 218c1bb86cdSEric Blake fname, strerror(errno)); 219c1bb86cdSEric Blake return -errno; 220c1bb86cdSEric Blake } 221c1bb86cdSEric Blake 222c1bb86cdSEric Blake if (!S_ISBLK(sb.st_mode)) { 223c1bb86cdSEric Blake return 0; 224c1bb86cdSEric Blake } 225c1bb86cdSEric Blake 226c1bb86cdSEric Blake if (dp == NULL) { 227c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "r%s", fname); 228c1bb86cdSEric Blake } else { 229c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "%.*s/r%s", 230c1bb86cdSEric Blake (int)(dp - fname), fname, dp + 1); 231c1bb86cdSEric Blake } 232c1bb86cdSEric Blake fprintf(stderr, "%s is a block device", fname); 233c1bb86cdSEric Blake *filename = namebuf; 234c1bb86cdSEric Blake fprintf(stderr, ", using %s\n", *filename); 235c1bb86cdSEric Blake 236c1bb86cdSEric Blake return 0; 237c1bb86cdSEric Blake } 238c1bb86cdSEric Blake #else 239c1bb86cdSEric Blake static int raw_normalize_devicepath(const char **filename) 240c1bb86cdSEric Blake { 241c1bb86cdSEric Blake return 0; 242c1bb86cdSEric Blake } 243c1bb86cdSEric Blake #endif 244c1bb86cdSEric Blake 245c1bb86cdSEric Blake /* 246c1bb86cdSEric Blake * Get logical block size via ioctl. On success store it in @sector_size_p. 247c1bb86cdSEric Blake */ 248c1bb86cdSEric Blake static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) 249c1bb86cdSEric Blake { 250c1bb86cdSEric Blake unsigned int sector_size; 251c1bb86cdSEric Blake bool success = false; 252700f9ce0SPeter Maydell int i; 253c1bb86cdSEric Blake 254c1bb86cdSEric Blake errno = ENOTSUP; 255700f9ce0SPeter Maydell static const unsigned long ioctl_list[] = { 256c1bb86cdSEric Blake #ifdef BLKSSZGET 257700f9ce0SPeter Maydell BLKSSZGET, 258c1bb86cdSEric Blake #endif 259c1bb86cdSEric Blake #ifdef DKIOCGETBLOCKSIZE 260700f9ce0SPeter Maydell DKIOCGETBLOCKSIZE, 261c1bb86cdSEric Blake #endif 262c1bb86cdSEric Blake #ifdef DIOCGSECTORSIZE 263700f9ce0SPeter Maydell DIOCGSECTORSIZE, 264700f9ce0SPeter Maydell #endif 265700f9ce0SPeter Maydell }; 266700f9ce0SPeter Maydell 267700f9ce0SPeter Maydell /* Try a few ioctls to get the right size */ 268700f9ce0SPeter Maydell for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { 269700f9ce0SPeter Maydell if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { 270c1bb86cdSEric Blake *sector_size_p = sector_size; 271c1bb86cdSEric Blake success = true; 272c1bb86cdSEric Blake } 273700f9ce0SPeter Maydell } 274c1bb86cdSEric Blake 275c1bb86cdSEric Blake return success ? 0 : -errno; 276c1bb86cdSEric Blake } 277c1bb86cdSEric Blake 278c1bb86cdSEric Blake /** 279c1bb86cdSEric Blake * Get physical block size of @fd. 280c1bb86cdSEric Blake * On success, store it in @blk_size and return 0. 281c1bb86cdSEric Blake * On failure, return -errno. 282c1bb86cdSEric Blake */ 283c1bb86cdSEric Blake static int probe_physical_blocksize(int fd, unsigned int *blk_size) 284c1bb86cdSEric Blake { 285c1bb86cdSEric Blake #ifdef BLKPBSZGET 286c1bb86cdSEric Blake if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { 287c1bb86cdSEric Blake return -errno; 288c1bb86cdSEric Blake } 289c1bb86cdSEric Blake return 0; 290c1bb86cdSEric Blake #else 291c1bb86cdSEric Blake return -ENOTSUP; 292c1bb86cdSEric Blake #endif 293c1bb86cdSEric Blake } 294c1bb86cdSEric Blake 295c1bb86cdSEric Blake /* Check if read is allowed with given memory buffer and length. 296c1bb86cdSEric Blake * 297c1bb86cdSEric Blake * This function is used to check O_DIRECT memory buffer and request alignment. 298c1bb86cdSEric Blake */ 299c1bb86cdSEric Blake static bool raw_is_io_aligned(int fd, void *buf, size_t len) 300c1bb86cdSEric Blake { 301c1bb86cdSEric Blake ssize_t ret = pread(fd, buf, len, 0); 302c1bb86cdSEric Blake 303c1bb86cdSEric Blake if (ret >= 0) { 304c1bb86cdSEric Blake return true; 305c1bb86cdSEric Blake } 306c1bb86cdSEric Blake 307c1bb86cdSEric Blake #ifdef __linux__ 308c1bb86cdSEric Blake /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore 309c1bb86cdSEric Blake * other errors (e.g. real I/O error), which could happen on a failed 310c1bb86cdSEric Blake * drive, since we only care about probing alignment. 311c1bb86cdSEric Blake */ 312c1bb86cdSEric Blake if (errno != EINVAL) { 313c1bb86cdSEric Blake return true; 314c1bb86cdSEric Blake } 315c1bb86cdSEric Blake #endif 316c1bb86cdSEric Blake 317c1bb86cdSEric Blake return false; 318c1bb86cdSEric Blake } 319c1bb86cdSEric Blake 320c1bb86cdSEric Blake static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) 321c1bb86cdSEric Blake { 322c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 323c1bb86cdSEric Blake char *buf; 324c1bb86cdSEric Blake size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize()); 325c1bb86cdSEric Blake 326c1bb86cdSEric Blake /* For SCSI generic devices the alignment is not really used. 327c1bb86cdSEric Blake With buffered I/O, we don't have any restrictions. */ 328c1bb86cdSEric Blake if (bdrv_is_sg(bs) || !s->needs_alignment) { 329c1bb86cdSEric Blake bs->bl.request_alignment = 1; 330c1bb86cdSEric Blake s->buf_align = 1; 331c1bb86cdSEric Blake return; 332c1bb86cdSEric Blake } 333c1bb86cdSEric Blake 334c1bb86cdSEric Blake bs->bl.request_alignment = 0; 335c1bb86cdSEric Blake s->buf_align = 0; 336c1bb86cdSEric Blake /* Let's try to use the logical blocksize for the alignment. */ 337c1bb86cdSEric Blake if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { 338c1bb86cdSEric Blake bs->bl.request_alignment = 0; 339c1bb86cdSEric Blake } 340c1bb86cdSEric Blake #ifdef CONFIG_XFS 341c1bb86cdSEric Blake if (s->is_xfs) { 342c1bb86cdSEric Blake struct dioattr da; 343c1bb86cdSEric Blake if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { 344c1bb86cdSEric Blake bs->bl.request_alignment = da.d_miniosz; 345c1bb86cdSEric Blake /* The kernel returns wrong information for d_mem */ 346c1bb86cdSEric Blake /* s->buf_align = da.d_mem; */ 347c1bb86cdSEric Blake } 348c1bb86cdSEric Blake } 349c1bb86cdSEric Blake #endif 350c1bb86cdSEric Blake 351c1bb86cdSEric Blake /* If we could not get the sizes so far, we can only guess them */ 352c1bb86cdSEric Blake if (!s->buf_align) { 353c1bb86cdSEric Blake size_t align; 354c1bb86cdSEric Blake buf = qemu_memalign(max_align, 2 * max_align); 355c1bb86cdSEric Blake for (align = 512; align <= max_align; align <<= 1) { 356c1bb86cdSEric Blake if (raw_is_io_aligned(fd, buf + align, max_align)) { 357c1bb86cdSEric Blake s->buf_align = align; 358c1bb86cdSEric Blake break; 359c1bb86cdSEric Blake } 360c1bb86cdSEric Blake } 361c1bb86cdSEric Blake qemu_vfree(buf); 362c1bb86cdSEric Blake } 363c1bb86cdSEric Blake 364c1bb86cdSEric Blake if (!bs->bl.request_alignment) { 365c1bb86cdSEric Blake size_t align; 366c1bb86cdSEric Blake buf = qemu_memalign(s->buf_align, max_align); 367c1bb86cdSEric Blake for (align = 512; align <= max_align; align <<= 1) { 368c1bb86cdSEric Blake if (raw_is_io_aligned(fd, buf, align)) { 369c1bb86cdSEric Blake bs->bl.request_alignment = align; 370c1bb86cdSEric Blake break; 371c1bb86cdSEric Blake } 372c1bb86cdSEric Blake } 373c1bb86cdSEric Blake qemu_vfree(buf); 374c1bb86cdSEric Blake } 375c1bb86cdSEric Blake 376c1bb86cdSEric Blake if (!s->buf_align || !bs->bl.request_alignment) { 377c1bb86cdSEric Blake error_setg(errp, "Could not find working O_DIRECT alignment"); 378c1bb86cdSEric Blake error_append_hint(errp, "Try cache.direct=off\n"); 379c1bb86cdSEric Blake } 380c1bb86cdSEric Blake } 381c1bb86cdSEric Blake 382c1bb86cdSEric Blake static void raw_parse_flags(int bdrv_flags, int *open_flags) 383c1bb86cdSEric Blake { 384c1bb86cdSEric Blake assert(open_flags != NULL); 385c1bb86cdSEric Blake 386c1bb86cdSEric Blake *open_flags |= O_BINARY; 387c1bb86cdSEric Blake *open_flags &= ~O_ACCMODE; 388c1bb86cdSEric Blake if (bdrv_flags & BDRV_O_RDWR) { 389c1bb86cdSEric Blake *open_flags |= O_RDWR; 390c1bb86cdSEric Blake } else { 391c1bb86cdSEric Blake *open_flags |= O_RDONLY; 392c1bb86cdSEric Blake } 393c1bb86cdSEric Blake 394c1bb86cdSEric Blake /* Use O_DSYNC for write-through caching, no flags for write-back caching, 395c1bb86cdSEric Blake * and O_DIRECT for no caching. */ 396c1bb86cdSEric Blake if ((bdrv_flags & BDRV_O_NOCACHE)) { 397c1bb86cdSEric Blake *open_flags |= O_DIRECT; 398c1bb86cdSEric Blake } 399c1bb86cdSEric Blake } 400c1bb86cdSEric Blake 401c1bb86cdSEric Blake static void raw_parse_filename(const char *filename, QDict *options, 402c1bb86cdSEric Blake Error **errp) 403c1bb86cdSEric Blake { 40403c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "file:", options); 405c1bb86cdSEric Blake } 406c1bb86cdSEric Blake 407c1bb86cdSEric Blake static QemuOptsList raw_runtime_opts = { 408c1bb86cdSEric Blake .name = "raw", 409c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), 410c1bb86cdSEric Blake .desc = { 411c1bb86cdSEric Blake { 412c1bb86cdSEric Blake .name = "filename", 413c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 414c1bb86cdSEric Blake .help = "File name of the image", 415c1bb86cdSEric Blake }, 416c1bb86cdSEric Blake { 417c1bb86cdSEric Blake .name = "aio", 418c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 419c1bb86cdSEric Blake .help = "host AIO implementation (threads, native)", 420c1bb86cdSEric Blake }, 42116b48d5dSFam Zheng { 42216b48d5dSFam Zheng .name = "locking", 42316b48d5dSFam Zheng .type = QEMU_OPT_STRING, 42416b48d5dSFam Zheng .help = "file locking mode (on/off/auto, default: auto)", 42516b48d5dSFam Zheng }, 4267c9e5276SPaolo Bonzini { 4277c9e5276SPaolo Bonzini .name = "pr-manager", 4287c9e5276SPaolo Bonzini .type = QEMU_OPT_STRING, 4297c9e5276SPaolo Bonzini .help = "id of persistent reservation manager object (default: none)", 4307c9e5276SPaolo Bonzini }, 43131be8a2aSStefan Hajnoczi { 43231be8a2aSStefan Hajnoczi .name = "x-check-cache-dropped", 43331be8a2aSStefan Hajnoczi .type = QEMU_OPT_BOOL, 43431be8a2aSStefan Hajnoczi .help = "check that page cache was dropped on live migration (default: off)" 43531be8a2aSStefan Hajnoczi }, 436c1bb86cdSEric Blake { /* end of list */ } 437c1bb86cdSEric Blake }, 438c1bb86cdSEric Blake }; 439c1bb86cdSEric Blake 440c1bb86cdSEric Blake static int raw_open_common(BlockDriverState *bs, QDict *options, 441230ff739SJohn Snow int bdrv_flags, int open_flags, 442230ff739SJohn Snow bool device, Error **errp) 443c1bb86cdSEric Blake { 444c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 445c1bb86cdSEric Blake QemuOpts *opts; 446c1bb86cdSEric Blake Error *local_err = NULL; 447c1bb86cdSEric Blake const char *filename = NULL; 4487c9e5276SPaolo Bonzini const char *str; 449c1bb86cdSEric Blake BlockdevAioOptions aio, aio_default; 450c1bb86cdSEric Blake int fd, ret; 451c1bb86cdSEric Blake struct stat st; 452244a5668SFam Zheng OnOffAuto locking; 453c1bb86cdSEric Blake 454c1bb86cdSEric Blake opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 455c1bb86cdSEric Blake qemu_opts_absorb_qdict(opts, options, &local_err); 456c1bb86cdSEric Blake if (local_err) { 457c1bb86cdSEric Blake error_propagate(errp, local_err); 458c1bb86cdSEric Blake ret = -EINVAL; 459c1bb86cdSEric Blake goto fail; 460c1bb86cdSEric Blake } 461c1bb86cdSEric Blake 462c1bb86cdSEric Blake filename = qemu_opt_get(opts, "filename"); 463c1bb86cdSEric Blake 464c1bb86cdSEric Blake ret = raw_normalize_devicepath(&filename); 465c1bb86cdSEric Blake if (ret != 0) { 466c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not normalize device path"); 467c1bb86cdSEric Blake goto fail; 468c1bb86cdSEric Blake } 469c1bb86cdSEric Blake 470c1bb86cdSEric Blake aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO) 471c1bb86cdSEric Blake ? BLOCKDEV_AIO_OPTIONS_NATIVE 472c1bb86cdSEric Blake : BLOCKDEV_AIO_OPTIONS_THREADS; 473f7abe0ecSMarc-André Lureau aio = qapi_enum_parse(&BlockdevAioOptions_lookup, 474f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "aio"), 47506c60b6cSMarkus Armbruster aio_default, &local_err); 476c1bb86cdSEric Blake if (local_err) { 477c1bb86cdSEric Blake error_propagate(errp, local_err); 478c1bb86cdSEric Blake ret = -EINVAL; 479c1bb86cdSEric Blake goto fail; 480c1bb86cdSEric Blake } 481c1bb86cdSEric Blake s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); 482c1bb86cdSEric Blake 483f7abe0ecSMarc-André Lureau locking = qapi_enum_parse(&OnOffAuto_lookup, 484f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "locking"), 48506c60b6cSMarkus Armbruster ON_OFF_AUTO_AUTO, &local_err); 486244a5668SFam Zheng if (local_err) { 487244a5668SFam Zheng error_propagate(errp, local_err); 488244a5668SFam Zheng ret = -EINVAL; 489244a5668SFam Zheng goto fail; 490244a5668SFam Zheng } 491244a5668SFam Zheng switch (locking) { 492244a5668SFam Zheng case ON_OFF_AUTO_ON: 493244a5668SFam Zheng s->use_lock = true; 4942b218f5dSFam Zheng if (!qemu_has_ofd_lock()) { 495244a5668SFam Zheng fprintf(stderr, 4962b218f5dSFam Zheng "File lock requested but OFD locking syscall is " 4972b218f5dSFam Zheng "unavailable, falling back to POSIX file locks.\n" 4982b218f5dSFam Zheng "Due to the implementation, locks can be lost " 4992b218f5dSFam Zheng "unexpectedly.\n"); 5002b218f5dSFam Zheng } 501244a5668SFam Zheng break; 502244a5668SFam Zheng case ON_OFF_AUTO_OFF: 503244a5668SFam Zheng s->use_lock = false; 504244a5668SFam Zheng break; 505244a5668SFam Zheng case ON_OFF_AUTO_AUTO: 5062b218f5dSFam Zheng s->use_lock = qemu_has_ofd_lock(); 507244a5668SFam Zheng break; 508244a5668SFam Zheng default: 509244a5668SFam Zheng abort(); 510244a5668SFam Zheng } 511244a5668SFam Zheng 5127c9e5276SPaolo Bonzini str = qemu_opt_get(opts, "pr-manager"); 5137c9e5276SPaolo Bonzini if (str) { 5147c9e5276SPaolo Bonzini s->pr_mgr = pr_manager_lookup(str, &local_err); 5157c9e5276SPaolo Bonzini if (local_err) { 5167c9e5276SPaolo Bonzini error_propagate(errp, local_err); 5177c9e5276SPaolo Bonzini ret = -EINVAL; 5187c9e5276SPaolo Bonzini goto fail; 5197c9e5276SPaolo Bonzini } 5207c9e5276SPaolo Bonzini } 5217c9e5276SPaolo Bonzini 52231be8a2aSStefan Hajnoczi s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", 52331be8a2aSStefan Hajnoczi false); 52431be8a2aSStefan Hajnoczi 525c1bb86cdSEric Blake s->open_flags = open_flags; 526c1bb86cdSEric Blake raw_parse_flags(bdrv_flags, &s->open_flags); 527c1bb86cdSEric Blake 528c1bb86cdSEric Blake s->fd = -1; 529c1bb86cdSEric Blake fd = qemu_open(filename, s->open_flags, 0644); 530c1bb86cdSEric Blake if (fd < 0) { 531c1bb86cdSEric Blake ret = -errno; 532c1bb86cdSEric Blake error_setg_errno(errp, errno, "Could not open '%s'", filename); 533c1bb86cdSEric Blake if (ret == -EROFS) { 534c1bb86cdSEric Blake ret = -EACCES; 535c1bb86cdSEric Blake } 536c1bb86cdSEric Blake goto fail; 537c1bb86cdSEric Blake } 538c1bb86cdSEric Blake s->fd = fd; 539c1bb86cdSEric Blake 540244a5668SFam Zheng s->lock_fd = -1; 541244a5668SFam Zheng if (s->use_lock) { 542244a5668SFam Zheng fd = qemu_open(filename, s->open_flags); 543244a5668SFam Zheng if (fd < 0) { 544244a5668SFam Zheng ret = -errno; 545244a5668SFam Zheng error_setg_errno(errp, errno, "Could not open '%s' for locking", 546244a5668SFam Zheng filename); 547244a5668SFam Zheng qemu_close(s->fd); 548244a5668SFam Zheng goto fail; 549244a5668SFam Zheng } 550244a5668SFam Zheng s->lock_fd = fd; 551244a5668SFam Zheng } 552244a5668SFam Zheng s->perm = 0; 553244a5668SFam Zheng s->shared_perm = BLK_PERM_ALL; 554244a5668SFam Zheng 555c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 556c1bb86cdSEric Blake /* Currently Linux does AIO only for files opened with O_DIRECT */ 557ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 558ed6e2161SNishanth Aravamudan if (!(s->open_flags & O_DIRECT)) { 559c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but it requires " 560c1bb86cdSEric Blake "cache.direct=on, which was not specified."); 561c1bb86cdSEric Blake ret = -EINVAL; 562c1bb86cdSEric Blake goto fail; 563c1bb86cdSEric Blake } 564ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { 565ed6e2161SNishanth Aravamudan error_prepend(errp, "Unable to use native AIO: "); 566ed6e2161SNishanth Aravamudan goto fail; 567ed6e2161SNishanth Aravamudan } 568ed6e2161SNishanth Aravamudan } 569c1bb86cdSEric Blake #else 570c1bb86cdSEric Blake if (s->use_linux_aio) { 571c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but is not supported " 572c1bb86cdSEric Blake "in this build."); 573c1bb86cdSEric Blake ret = -EINVAL; 574c1bb86cdSEric Blake goto fail; 575c1bb86cdSEric Blake } 576c1bb86cdSEric Blake #endif /* !defined(CONFIG_LINUX_AIO) */ 577c1bb86cdSEric Blake 578c1bb86cdSEric Blake s->has_discard = true; 579c1bb86cdSEric Blake s->has_write_zeroes = true; 580c1bb86cdSEric Blake if ((bs->open_flags & BDRV_O_NOCACHE) != 0) { 581c1bb86cdSEric Blake s->needs_alignment = true; 582c1bb86cdSEric Blake } 583c1bb86cdSEric Blake 584c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 585c1bb86cdSEric Blake ret = -errno; 586c1bb86cdSEric Blake error_setg_errno(errp, errno, "Could not stat file"); 587c1bb86cdSEric Blake goto fail; 588c1bb86cdSEric Blake } 589230ff739SJohn Snow 590230ff739SJohn Snow if (!device) { 591230ff739SJohn Snow if (S_ISBLK(st.st_mode)) { 592230ff739SJohn Snow warn_report("Opening a block device as a file using the '%s' " 593230ff739SJohn Snow "driver is deprecated", bs->drv->format_name); 594230ff739SJohn Snow } else if (S_ISCHR(st.st_mode)) { 595230ff739SJohn Snow warn_report("Opening a character device as a file using the '%s' " 596230ff739SJohn Snow "driver is deprecated", bs->drv->format_name); 597230ff739SJohn Snow } else if (!S_ISREG(st.st_mode)) { 598230ff739SJohn Snow error_setg(errp, "A regular file was expected by the '%s' driver, " 599230ff739SJohn Snow "but something else was given", bs->drv->format_name); 600230ff739SJohn Snow ret = -EINVAL; 601230ff739SJohn Snow goto fail; 602230ff739SJohn Snow } else { 603c1bb86cdSEric Blake s->discard_zeroes = true; 604c1bb86cdSEric Blake s->has_fallocate = true; 605c1bb86cdSEric Blake } 606230ff739SJohn Snow } else { 607230ff739SJohn Snow if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 608230ff739SJohn Snow error_setg(errp, "'%s' driver expects either " 609230ff739SJohn Snow "a character or block device", bs->drv->format_name); 610230ff739SJohn Snow ret = -EINVAL; 611230ff739SJohn Snow goto fail; 612230ff739SJohn Snow } 613230ff739SJohn Snow } 614230ff739SJohn Snow 615c1bb86cdSEric Blake if (S_ISBLK(st.st_mode)) { 616c1bb86cdSEric Blake #ifdef BLKDISCARDZEROES 617c1bb86cdSEric Blake unsigned int arg; 618c1bb86cdSEric Blake if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { 619c1bb86cdSEric Blake s->discard_zeroes = true; 620c1bb86cdSEric Blake } 621c1bb86cdSEric Blake #endif 622c1bb86cdSEric Blake #ifdef __linux__ 623c1bb86cdSEric Blake /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do 624c1bb86cdSEric Blake * not rely on the contents of discarded blocks unless using O_DIRECT. 625c1bb86cdSEric Blake * Same for BLKZEROOUT. 626c1bb86cdSEric Blake */ 627c1bb86cdSEric Blake if (!(bs->open_flags & BDRV_O_NOCACHE)) { 628c1bb86cdSEric Blake s->discard_zeroes = false; 629c1bb86cdSEric Blake s->has_write_zeroes = false; 630c1bb86cdSEric Blake } 631c1bb86cdSEric Blake #endif 632c1bb86cdSEric Blake } 633c1bb86cdSEric Blake #ifdef __FreeBSD__ 634c1bb86cdSEric Blake if (S_ISCHR(st.st_mode)) { 635c1bb86cdSEric Blake /* 636c1bb86cdSEric Blake * The file is a char device (disk), which on FreeBSD isn't behind 637c1bb86cdSEric Blake * a pager, so force all requests to be aligned. This is needed 638c1bb86cdSEric Blake * so QEMU makes sure all IO operations on the device are aligned 639c1bb86cdSEric Blake * to sector size, or else FreeBSD will reject them with EINVAL. 640c1bb86cdSEric Blake */ 641c1bb86cdSEric Blake s->needs_alignment = true; 642c1bb86cdSEric Blake } 643c1bb86cdSEric Blake #endif 644c1bb86cdSEric Blake 645c1bb86cdSEric Blake #ifdef CONFIG_XFS 646c1bb86cdSEric Blake if (platform_test_xfs_fd(s->fd)) { 647c1bb86cdSEric Blake s->is_xfs = true; 648c1bb86cdSEric Blake } 649c1bb86cdSEric Blake #endif 650c1bb86cdSEric Blake 651e24d813bSEric Blake bs->supported_zero_flags = s->discard_zeroes ? BDRV_REQ_MAY_UNMAP : 0; 652c1bb86cdSEric Blake ret = 0; 653c1bb86cdSEric Blake fail: 654c1bb86cdSEric Blake if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { 655c1bb86cdSEric Blake unlink(filename); 656c1bb86cdSEric Blake } 657c1bb86cdSEric Blake qemu_opts_del(opts); 658c1bb86cdSEric Blake return ret; 659c1bb86cdSEric Blake } 660c1bb86cdSEric Blake 661c1bb86cdSEric Blake static int raw_open(BlockDriverState *bs, QDict *options, int flags, 662c1bb86cdSEric Blake Error **errp) 663c1bb86cdSEric Blake { 664c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 665c1bb86cdSEric Blake 666c1bb86cdSEric Blake s->type = FTYPE_FILE; 667230ff739SJohn Snow return raw_open_common(bs, options, flags, 0, false, errp); 668c1bb86cdSEric Blake } 669c1bb86cdSEric Blake 670244a5668SFam Zheng typedef enum { 671244a5668SFam Zheng RAW_PL_PREPARE, 672244a5668SFam Zheng RAW_PL_COMMIT, 673244a5668SFam Zheng RAW_PL_ABORT, 674244a5668SFam Zheng } RawPermLockOp; 675244a5668SFam Zheng 676244a5668SFam Zheng #define PERM_FOREACH(i) \ 677244a5668SFam Zheng for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) 678244a5668SFam Zheng 679244a5668SFam Zheng /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the 680244a5668SFam Zheng * file; if @unlock == true, also unlock the unneeded bytes. 681244a5668SFam Zheng * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. 682244a5668SFam Zheng */ 683d0a96155SMax Reitz static int raw_apply_lock_bytes(int fd, 684244a5668SFam Zheng uint64_t perm_lock_bits, 685244a5668SFam Zheng uint64_t shared_perm_lock_bits, 686244a5668SFam Zheng bool unlock, Error **errp) 687244a5668SFam Zheng { 688244a5668SFam Zheng int ret; 689244a5668SFam Zheng int i; 690244a5668SFam Zheng 691244a5668SFam Zheng PERM_FOREACH(i) { 692244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 693244a5668SFam Zheng if (perm_lock_bits & (1ULL << i)) { 694d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 695244a5668SFam Zheng if (ret) { 696244a5668SFam Zheng error_setg(errp, "Failed to lock byte %d", off); 697244a5668SFam Zheng return ret; 698244a5668SFam Zheng } 699244a5668SFam Zheng } else if (unlock) { 700d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 701244a5668SFam Zheng if (ret) { 702244a5668SFam Zheng error_setg(errp, "Failed to unlock byte %d", off); 703244a5668SFam Zheng return ret; 704244a5668SFam Zheng } 705244a5668SFam Zheng } 706244a5668SFam Zheng } 707244a5668SFam Zheng PERM_FOREACH(i) { 708244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 709244a5668SFam Zheng if (shared_perm_lock_bits & (1ULL << i)) { 710d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 711244a5668SFam Zheng if (ret) { 712244a5668SFam Zheng error_setg(errp, "Failed to lock byte %d", off); 713244a5668SFam Zheng return ret; 714244a5668SFam Zheng } 715244a5668SFam Zheng } else if (unlock) { 716d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 717244a5668SFam Zheng if (ret) { 718244a5668SFam Zheng error_setg(errp, "Failed to unlock byte %d", off); 719244a5668SFam Zheng return ret; 720244a5668SFam Zheng } 721244a5668SFam Zheng } 722244a5668SFam Zheng } 723244a5668SFam Zheng return 0; 724244a5668SFam Zheng } 725244a5668SFam Zheng 726244a5668SFam Zheng /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ 727d0a96155SMax Reitz static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, 728244a5668SFam Zheng Error **errp) 729244a5668SFam Zheng { 730244a5668SFam Zheng int ret; 731244a5668SFam Zheng int i; 732244a5668SFam Zheng 733244a5668SFam Zheng PERM_FOREACH(i) { 734244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 735244a5668SFam Zheng uint64_t p = 1ULL << i; 736244a5668SFam Zheng if (perm & p) { 737d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 738244a5668SFam Zheng if (ret) { 739244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 740244a5668SFam Zheng error_setg(errp, 741244a5668SFam Zheng "Failed to get \"%s\" lock", 742244a5668SFam Zheng perm_name); 743244a5668SFam Zheng g_free(perm_name); 744244a5668SFam Zheng error_append_hint(errp, 745244a5668SFam Zheng "Is another process using the image?\n"); 746244a5668SFam Zheng return ret; 747244a5668SFam Zheng } 748244a5668SFam Zheng } 749244a5668SFam Zheng } 750244a5668SFam Zheng PERM_FOREACH(i) { 751244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 752244a5668SFam Zheng uint64_t p = 1ULL << i; 753244a5668SFam Zheng if (!(shared_perm & p)) { 754d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 755244a5668SFam Zheng if (ret) { 756244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 757244a5668SFam Zheng error_setg(errp, 758244a5668SFam Zheng "Failed to get shared \"%s\" lock", 759244a5668SFam Zheng perm_name); 760244a5668SFam Zheng g_free(perm_name); 761244a5668SFam Zheng error_append_hint(errp, 762244a5668SFam Zheng "Is another process using the image?\n"); 763244a5668SFam Zheng return ret; 764244a5668SFam Zheng } 765244a5668SFam Zheng } 766244a5668SFam Zheng } 767244a5668SFam Zheng return 0; 768244a5668SFam Zheng } 769244a5668SFam Zheng 770244a5668SFam Zheng static int raw_handle_perm_lock(BlockDriverState *bs, 771244a5668SFam Zheng RawPermLockOp op, 772244a5668SFam Zheng uint64_t new_perm, uint64_t new_shared, 773244a5668SFam Zheng Error **errp) 774244a5668SFam Zheng { 775244a5668SFam Zheng BDRVRawState *s = bs->opaque; 776244a5668SFam Zheng int ret = 0; 777244a5668SFam Zheng Error *local_err = NULL; 778244a5668SFam Zheng 779244a5668SFam Zheng if (!s->use_lock) { 780244a5668SFam Zheng return 0; 781244a5668SFam Zheng } 782244a5668SFam Zheng 783244a5668SFam Zheng if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { 784244a5668SFam Zheng return 0; 785244a5668SFam Zheng } 786244a5668SFam Zheng 787244a5668SFam Zheng assert(s->lock_fd > 0); 788244a5668SFam Zheng 789244a5668SFam Zheng switch (op) { 790244a5668SFam Zheng case RAW_PL_PREPARE: 791d0a96155SMax Reitz ret = raw_apply_lock_bytes(s->lock_fd, s->perm | new_perm, 792244a5668SFam Zheng ~s->shared_perm | ~new_shared, 793244a5668SFam Zheng false, errp); 794244a5668SFam Zheng if (!ret) { 795d0a96155SMax Reitz ret = raw_check_lock_bytes(s->lock_fd, new_perm, new_shared, errp); 796244a5668SFam Zheng if (!ret) { 797244a5668SFam Zheng return 0; 798244a5668SFam Zheng } 799244a5668SFam Zheng } 800244a5668SFam Zheng op = RAW_PL_ABORT; 801244a5668SFam Zheng /* fall through to unlock bytes. */ 802244a5668SFam Zheng case RAW_PL_ABORT: 803d0a96155SMax Reitz raw_apply_lock_bytes(s->lock_fd, s->perm, ~s->shared_perm, 804d0a96155SMax Reitz true, &local_err); 805244a5668SFam Zheng if (local_err) { 806244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 807244a5668SFam Zheng * fail. Something weird happened, report it. 808244a5668SFam Zheng */ 809244a5668SFam Zheng error_report_err(local_err); 810244a5668SFam Zheng } 811244a5668SFam Zheng break; 812244a5668SFam Zheng case RAW_PL_COMMIT: 813d0a96155SMax Reitz raw_apply_lock_bytes(s->lock_fd, new_perm, ~new_shared, 814d0a96155SMax Reitz true, &local_err); 815244a5668SFam Zheng if (local_err) { 816244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 817244a5668SFam Zheng * fail. Something weird happened, report it. 818244a5668SFam Zheng */ 819244a5668SFam Zheng error_report_err(local_err); 820244a5668SFam Zheng } 821244a5668SFam Zheng break; 822244a5668SFam Zheng } 823244a5668SFam Zheng return ret; 824244a5668SFam Zheng } 825244a5668SFam Zheng 826c1bb86cdSEric Blake static int raw_reopen_prepare(BDRVReopenState *state, 827c1bb86cdSEric Blake BlockReopenQueue *queue, Error **errp) 828c1bb86cdSEric Blake { 829c1bb86cdSEric Blake BDRVRawState *s; 830c1bb86cdSEric Blake BDRVRawReopenState *rs; 83131be8a2aSStefan Hajnoczi QemuOpts *opts; 832c1bb86cdSEric Blake int ret = 0; 833c1bb86cdSEric Blake Error *local_err = NULL; 834c1bb86cdSEric Blake 835c1bb86cdSEric Blake assert(state != NULL); 836c1bb86cdSEric Blake assert(state->bs != NULL); 837c1bb86cdSEric Blake 838c1bb86cdSEric Blake s = state->bs->opaque; 839c1bb86cdSEric Blake 840c1bb86cdSEric Blake state->opaque = g_new0(BDRVRawReopenState, 1); 841c1bb86cdSEric Blake rs = state->opaque; 84231be8a2aSStefan Hajnoczi rs->fd = -1; 84331be8a2aSStefan Hajnoczi 84431be8a2aSStefan Hajnoczi /* Handle options changes */ 84531be8a2aSStefan Hajnoczi opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 84631be8a2aSStefan Hajnoczi qemu_opts_absorb_qdict(opts, state->options, &local_err); 84731be8a2aSStefan Hajnoczi if (local_err) { 84831be8a2aSStefan Hajnoczi error_propagate(errp, local_err); 84931be8a2aSStefan Hajnoczi ret = -EINVAL; 85031be8a2aSStefan Hajnoczi goto out; 85131be8a2aSStefan Hajnoczi } 85231be8a2aSStefan Hajnoczi 85331be8a2aSStefan Hajnoczi rs->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", 85431be8a2aSStefan Hajnoczi s->check_cache_dropped); 855c1bb86cdSEric Blake 856c1bb86cdSEric Blake if (s->type == FTYPE_CD) { 857c1bb86cdSEric Blake rs->open_flags |= O_NONBLOCK; 858c1bb86cdSEric Blake } 859c1bb86cdSEric Blake 860c1bb86cdSEric Blake raw_parse_flags(state->flags, &rs->open_flags); 861c1bb86cdSEric Blake 862c1bb86cdSEric Blake int fcntl_flags = O_APPEND | O_NONBLOCK; 863c1bb86cdSEric Blake #ifdef O_NOATIME 864c1bb86cdSEric Blake fcntl_flags |= O_NOATIME; 865c1bb86cdSEric Blake #endif 866c1bb86cdSEric Blake 867c1bb86cdSEric Blake #ifdef O_ASYNC 868c1bb86cdSEric Blake /* Not all operating systems have O_ASYNC, and those that don't 869c1bb86cdSEric Blake * will not let us track the state into rs->open_flags (typically 870c1bb86cdSEric Blake * you achieve the same effect with an ioctl, for example I_SETSIG 871c1bb86cdSEric Blake * on Solaris). But we do not use O_ASYNC, so that's fine. 872c1bb86cdSEric Blake */ 873c1bb86cdSEric Blake assert((s->open_flags & O_ASYNC) == 0); 874c1bb86cdSEric Blake #endif 875c1bb86cdSEric Blake 876c1bb86cdSEric Blake if ((rs->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { 877c1bb86cdSEric Blake /* dup the original fd */ 878c1bb86cdSEric Blake rs->fd = qemu_dup(s->fd); 879c1bb86cdSEric Blake if (rs->fd >= 0) { 880c1bb86cdSEric Blake ret = fcntl_setfl(rs->fd, rs->open_flags); 881c1bb86cdSEric Blake if (ret) { 882c1bb86cdSEric Blake qemu_close(rs->fd); 883c1bb86cdSEric Blake rs->fd = -1; 884c1bb86cdSEric Blake } 885c1bb86cdSEric Blake } 886c1bb86cdSEric Blake } 887c1bb86cdSEric Blake 888c1bb86cdSEric Blake /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ 889c1bb86cdSEric Blake if (rs->fd == -1) { 890c1bb86cdSEric Blake const char *normalized_filename = state->bs->filename; 891c1bb86cdSEric Blake ret = raw_normalize_devicepath(&normalized_filename); 892c1bb86cdSEric Blake if (ret < 0) { 893c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not normalize device path"); 894c1bb86cdSEric Blake } else { 895c1bb86cdSEric Blake assert(!(rs->open_flags & O_CREAT)); 896c1bb86cdSEric Blake rs->fd = qemu_open(normalized_filename, rs->open_flags); 897c1bb86cdSEric Blake if (rs->fd == -1) { 898c1bb86cdSEric Blake error_setg_errno(errp, errno, "Could not reopen file"); 899c1bb86cdSEric Blake ret = -1; 900c1bb86cdSEric Blake } 901c1bb86cdSEric Blake } 902c1bb86cdSEric Blake } 903c1bb86cdSEric Blake 904c1bb86cdSEric Blake /* Fail already reopen_prepare() if we can't get a working O_DIRECT 905c1bb86cdSEric Blake * alignment with the new fd. */ 906c1bb86cdSEric Blake if (rs->fd != -1) { 907c1bb86cdSEric Blake raw_probe_alignment(state->bs, rs->fd, &local_err); 908c1bb86cdSEric Blake if (local_err) { 909c1bb86cdSEric Blake qemu_close(rs->fd); 910c1bb86cdSEric Blake rs->fd = -1; 911c1bb86cdSEric Blake error_propagate(errp, local_err); 912c1bb86cdSEric Blake ret = -EINVAL; 913c1bb86cdSEric Blake } 914c1bb86cdSEric Blake } 915c1bb86cdSEric Blake 91631be8a2aSStefan Hajnoczi out: 91731be8a2aSStefan Hajnoczi qemu_opts_del(opts); 918c1bb86cdSEric Blake return ret; 919c1bb86cdSEric Blake } 920c1bb86cdSEric Blake 921c1bb86cdSEric Blake static void raw_reopen_commit(BDRVReopenState *state) 922c1bb86cdSEric Blake { 923c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 924c1bb86cdSEric Blake BDRVRawState *s = state->bs->opaque; 925c1bb86cdSEric Blake 92631be8a2aSStefan Hajnoczi s->check_cache_dropped = rs->check_cache_dropped; 927c1bb86cdSEric Blake s->open_flags = rs->open_flags; 928c1bb86cdSEric Blake 929c1bb86cdSEric Blake qemu_close(s->fd); 930c1bb86cdSEric Blake s->fd = rs->fd; 931c1bb86cdSEric Blake 932c1bb86cdSEric Blake g_free(state->opaque); 933c1bb86cdSEric Blake state->opaque = NULL; 934c1bb86cdSEric Blake } 935c1bb86cdSEric Blake 936c1bb86cdSEric Blake 937c1bb86cdSEric Blake static void raw_reopen_abort(BDRVReopenState *state) 938c1bb86cdSEric Blake { 939c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 940c1bb86cdSEric Blake 941c1bb86cdSEric Blake /* nothing to do if NULL, we didn't get far enough */ 942c1bb86cdSEric Blake if (rs == NULL) { 943c1bb86cdSEric Blake return; 944c1bb86cdSEric Blake } 945c1bb86cdSEric Blake 946c1bb86cdSEric Blake if (rs->fd >= 0) { 947c1bb86cdSEric Blake qemu_close(rs->fd); 948c1bb86cdSEric Blake rs->fd = -1; 949c1bb86cdSEric Blake } 950c1bb86cdSEric Blake g_free(state->opaque); 951c1bb86cdSEric Blake state->opaque = NULL; 952c1bb86cdSEric Blake } 953c1bb86cdSEric Blake 95448265250SEric Farman static int hdev_get_max_transfer_length(BlockDriverState *bs, int fd) 955c1bb86cdSEric Blake { 956c1bb86cdSEric Blake #ifdef BLKSECTGET 95748265250SEric Farman int max_bytes = 0; 95848265250SEric Farman short max_sectors = 0; 95948265250SEric Farman if (bs->sg && ioctl(fd, BLKSECTGET, &max_bytes) == 0) { 96048265250SEric Farman return max_bytes; 96148265250SEric Farman } else if (!bs->sg && ioctl(fd, BLKSECTGET, &max_sectors) == 0) { 96248265250SEric Farman return max_sectors << BDRV_SECTOR_BITS; 963c1bb86cdSEric Blake } else { 964c1bb86cdSEric Blake return -errno; 965c1bb86cdSEric Blake } 966c1bb86cdSEric Blake #else 967c1bb86cdSEric Blake return -ENOSYS; 968c1bb86cdSEric Blake #endif 969c1bb86cdSEric Blake } 970c1bb86cdSEric Blake 9719103f1ceSFam Zheng static int hdev_get_max_segments(const struct stat *st) 9729103f1ceSFam Zheng { 9739103f1ceSFam Zheng #ifdef CONFIG_LINUX 9749103f1ceSFam Zheng char buf[32]; 9759103f1ceSFam Zheng const char *end; 9769103f1ceSFam Zheng char *sysfspath; 9779103f1ceSFam Zheng int ret; 9789103f1ceSFam Zheng int fd = -1; 9799103f1ceSFam Zheng long max_segments; 9809103f1ceSFam Zheng 9819103f1ceSFam Zheng sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", 9829103f1ceSFam Zheng major(st->st_rdev), minor(st->st_rdev)); 9839103f1ceSFam Zheng fd = open(sysfspath, O_RDONLY); 9849103f1ceSFam Zheng if (fd == -1) { 9859103f1ceSFam Zheng ret = -errno; 9869103f1ceSFam Zheng goto out; 9879103f1ceSFam Zheng } 9889103f1ceSFam Zheng do { 98969583490SStefan Hajnoczi ret = read(fd, buf, sizeof(buf) - 1); 9909103f1ceSFam Zheng } while (ret == -1 && errno == EINTR); 9919103f1ceSFam Zheng if (ret < 0) { 9929103f1ceSFam Zheng ret = -errno; 9939103f1ceSFam Zheng goto out; 9949103f1ceSFam Zheng } else if (ret == 0) { 9959103f1ceSFam Zheng ret = -EIO; 9969103f1ceSFam Zheng goto out; 9979103f1ceSFam Zheng } 9989103f1ceSFam Zheng buf[ret] = 0; 9999103f1ceSFam Zheng /* The file is ended with '\n', pass 'end' to accept that. */ 10009103f1ceSFam Zheng ret = qemu_strtol(buf, &end, 10, &max_segments); 10019103f1ceSFam Zheng if (ret == 0 && end && *end == '\n') { 10029103f1ceSFam Zheng ret = max_segments; 10039103f1ceSFam Zheng } 10049103f1ceSFam Zheng 10059103f1ceSFam Zheng out: 1006fed414dfSFam Zheng if (fd != -1) { 1007fed414dfSFam Zheng close(fd); 1008fed414dfSFam Zheng } 10099103f1ceSFam Zheng g_free(sysfspath); 10109103f1ceSFam Zheng return ret; 10119103f1ceSFam Zheng #else 10129103f1ceSFam Zheng return -ENOTSUP; 10139103f1ceSFam Zheng #endif 10149103f1ceSFam Zheng } 10159103f1ceSFam Zheng 1016c1bb86cdSEric Blake static void raw_refresh_limits(BlockDriverState *bs, Error **errp) 1017c1bb86cdSEric Blake { 1018c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1019c1bb86cdSEric Blake struct stat st; 1020c1bb86cdSEric Blake 1021c1bb86cdSEric Blake if (!fstat(s->fd, &st)) { 1022c4c41a0aSEric Farman if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode)) { 102348265250SEric Farman int ret = hdev_get_max_transfer_length(bs, s->fd); 102448265250SEric Farman if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { 102548265250SEric Farman bs->bl.max_transfer = pow2floor(ret); 1026c1bb86cdSEric Blake } 10279103f1ceSFam Zheng ret = hdev_get_max_segments(&st); 10289103f1ceSFam Zheng if (ret > 0) { 10299103f1ceSFam Zheng bs->bl.max_transfer = MIN(bs->bl.max_transfer, 10309103f1ceSFam Zheng ret * getpagesize()); 10319103f1ceSFam Zheng } 1032c1bb86cdSEric Blake } 1033c1bb86cdSEric Blake } 1034c1bb86cdSEric Blake 1035c1bb86cdSEric Blake raw_probe_alignment(bs, s->fd, errp); 1036c1bb86cdSEric Blake bs->bl.min_mem_alignment = s->buf_align; 1037c1bb86cdSEric Blake bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize()); 1038c1bb86cdSEric Blake } 1039c1bb86cdSEric Blake 1040c1bb86cdSEric Blake static int check_for_dasd(int fd) 1041c1bb86cdSEric Blake { 1042c1bb86cdSEric Blake #ifdef BIODASDINFO2 1043c1bb86cdSEric Blake struct dasd_information2_t info = {0}; 1044c1bb86cdSEric Blake 1045c1bb86cdSEric Blake return ioctl(fd, BIODASDINFO2, &info); 1046c1bb86cdSEric Blake #else 1047c1bb86cdSEric Blake return -1; 1048c1bb86cdSEric Blake #endif 1049c1bb86cdSEric Blake } 1050c1bb86cdSEric Blake 1051c1bb86cdSEric Blake /** 1052c1bb86cdSEric Blake * Try to get @bs's logical and physical block size. 1053c1bb86cdSEric Blake * On success, store them in @bsz and return zero. 1054c1bb86cdSEric Blake * On failure, return negative errno. 1055c1bb86cdSEric Blake */ 1056c1bb86cdSEric Blake static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) 1057c1bb86cdSEric Blake { 1058c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1059c1bb86cdSEric Blake int ret; 1060c1bb86cdSEric Blake 1061c1bb86cdSEric Blake /* If DASD, get blocksizes */ 1062c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1063c1bb86cdSEric Blake return -ENOTSUP; 1064c1bb86cdSEric Blake } 1065c1bb86cdSEric Blake ret = probe_logical_blocksize(s->fd, &bsz->log); 1066c1bb86cdSEric Blake if (ret < 0) { 1067c1bb86cdSEric Blake return ret; 1068c1bb86cdSEric Blake } 1069c1bb86cdSEric Blake return probe_physical_blocksize(s->fd, &bsz->phys); 1070c1bb86cdSEric Blake } 1071c1bb86cdSEric Blake 1072c1bb86cdSEric Blake /** 1073c1bb86cdSEric Blake * Try to get @bs's geometry: cyls, heads, sectors. 1074c1bb86cdSEric Blake * On success, store them in @geo and return 0. 1075c1bb86cdSEric Blake * On failure return -errno. 1076c1bb86cdSEric Blake * (Allows block driver to assign default geometry values that guest sees) 1077c1bb86cdSEric Blake */ 1078c1bb86cdSEric Blake #ifdef __linux__ 1079c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1080c1bb86cdSEric Blake { 1081c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1082c1bb86cdSEric Blake struct hd_geometry ioctl_geo = {0}; 1083c1bb86cdSEric Blake 1084c1bb86cdSEric Blake /* If DASD, get its geometry */ 1085c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1086c1bb86cdSEric Blake return -ENOTSUP; 1087c1bb86cdSEric Blake } 1088c1bb86cdSEric Blake if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { 1089c1bb86cdSEric Blake return -errno; 1090c1bb86cdSEric Blake } 1091c1bb86cdSEric Blake /* HDIO_GETGEO may return success even though geo contains zeros 1092c1bb86cdSEric Blake (e.g. certain multipath setups) */ 1093c1bb86cdSEric Blake if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { 1094c1bb86cdSEric Blake return -ENOTSUP; 1095c1bb86cdSEric Blake } 1096c1bb86cdSEric Blake /* Do not return a geometry for partition */ 1097c1bb86cdSEric Blake if (ioctl_geo.start != 0) { 1098c1bb86cdSEric Blake return -ENOTSUP; 1099c1bb86cdSEric Blake } 1100c1bb86cdSEric Blake geo->heads = ioctl_geo.heads; 1101c1bb86cdSEric Blake geo->sectors = ioctl_geo.sectors; 1102c1bb86cdSEric Blake geo->cylinders = ioctl_geo.cylinders; 1103c1bb86cdSEric Blake 1104c1bb86cdSEric Blake return 0; 1105c1bb86cdSEric Blake } 1106c1bb86cdSEric Blake #else /* __linux__ */ 1107c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1108c1bb86cdSEric Blake { 1109c1bb86cdSEric Blake return -ENOTSUP; 1110c1bb86cdSEric Blake } 1111c1bb86cdSEric Blake #endif 1112c1bb86cdSEric Blake 1113c1bb86cdSEric Blake static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb) 1114c1bb86cdSEric Blake { 1115c1bb86cdSEric Blake int ret; 1116c1bb86cdSEric Blake 1117c1bb86cdSEric Blake ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf); 1118c1bb86cdSEric Blake if (ret == -1) { 1119c1bb86cdSEric Blake return -errno; 1120c1bb86cdSEric Blake } 1121c1bb86cdSEric Blake 1122c1bb86cdSEric Blake return 0; 1123c1bb86cdSEric Blake } 1124c1bb86cdSEric Blake 1125c1bb86cdSEric Blake static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb) 1126c1bb86cdSEric Blake { 1127e5bcf967SKevin Wolf BDRVRawState *s = aiocb->bs->opaque; 1128c1bb86cdSEric Blake int ret; 1129c1bb86cdSEric Blake 1130e5bcf967SKevin Wolf if (s->page_cache_inconsistent) { 1131e5bcf967SKevin Wolf return -EIO; 1132e5bcf967SKevin Wolf } 1133e5bcf967SKevin Wolf 1134c1bb86cdSEric Blake ret = qemu_fdatasync(aiocb->aio_fildes); 1135c1bb86cdSEric Blake if (ret == -1) { 1136e5bcf967SKevin Wolf /* There is no clear definition of the semantics of a failing fsync(), 1137e5bcf967SKevin Wolf * so we may have to assume the worst. The sad truth is that this 1138e5bcf967SKevin Wolf * assumption is correct for Linux. Some pages are now probably marked 1139e5bcf967SKevin Wolf * clean in the page cache even though they are inconsistent with the 1140e5bcf967SKevin Wolf * on-disk contents. The next fdatasync() call would succeed, but no 1141e5bcf967SKevin Wolf * further writeback attempt will be made. We can't get back to a state 1142e5bcf967SKevin Wolf * in which we know what is on disk (we would have to rewrite 1143e5bcf967SKevin Wolf * everything that was touched since the last fdatasync() at least), so 1144e5bcf967SKevin Wolf * make bdrv_flush() fail permanently. Given that the behaviour isn't 1145e5bcf967SKevin Wolf * really defined, I have little hope that other OSes are doing better. 1146e5bcf967SKevin Wolf * 1147e5bcf967SKevin Wolf * Obviously, this doesn't affect O_DIRECT, which bypasses the page 1148e5bcf967SKevin Wolf * cache. */ 1149e5bcf967SKevin Wolf if ((s->open_flags & O_DIRECT) == 0) { 1150e5bcf967SKevin Wolf s->page_cache_inconsistent = true; 1151e5bcf967SKevin Wolf } 1152c1bb86cdSEric Blake return -errno; 1153c1bb86cdSEric Blake } 1154c1bb86cdSEric Blake return 0; 1155c1bb86cdSEric Blake } 1156c1bb86cdSEric Blake 1157c1bb86cdSEric Blake #ifdef CONFIG_PREADV 1158c1bb86cdSEric Blake 1159c1bb86cdSEric Blake static bool preadv_present = true; 1160c1bb86cdSEric Blake 1161c1bb86cdSEric Blake static ssize_t 1162c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1163c1bb86cdSEric Blake { 1164c1bb86cdSEric Blake return preadv(fd, iov, nr_iov, offset); 1165c1bb86cdSEric Blake } 1166c1bb86cdSEric Blake 1167c1bb86cdSEric Blake static ssize_t 1168c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1169c1bb86cdSEric Blake { 1170c1bb86cdSEric Blake return pwritev(fd, iov, nr_iov, offset); 1171c1bb86cdSEric Blake } 1172c1bb86cdSEric Blake 1173c1bb86cdSEric Blake #else 1174c1bb86cdSEric Blake 1175c1bb86cdSEric Blake static bool preadv_present = false; 1176c1bb86cdSEric Blake 1177c1bb86cdSEric Blake static ssize_t 1178c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1179c1bb86cdSEric Blake { 1180c1bb86cdSEric Blake return -ENOSYS; 1181c1bb86cdSEric Blake } 1182c1bb86cdSEric Blake 1183c1bb86cdSEric Blake static ssize_t 1184c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1185c1bb86cdSEric Blake { 1186c1bb86cdSEric Blake return -ENOSYS; 1187c1bb86cdSEric Blake } 1188c1bb86cdSEric Blake 1189c1bb86cdSEric Blake #endif 1190c1bb86cdSEric Blake 1191c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) 1192c1bb86cdSEric Blake { 1193c1bb86cdSEric Blake ssize_t len; 1194c1bb86cdSEric Blake 1195c1bb86cdSEric Blake do { 1196c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) 1197c1bb86cdSEric Blake len = qemu_pwritev(aiocb->aio_fildes, 1198c1bb86cdSEric Blake aiocb->aio_iov, 1199c1bb86cdSEric Blake aiocb->aio_niov, 1200c1bb86cdSEric Blake aiocb->aio_offset); 1201c1bb86cdSEric Blake else 1202c1bb86cdSEric Blake len = qemu_preadv(aiocb->aio_fildes, 1203c1bb86cdSEric Blake aiocb->aio_iov, 1204c1bb86cdSEric Blake aiocb->aio_niov, 1205c1bb86cdSEric Blake aiocb->aio_offset); 1206c1bb86cdSEric Blake } while (len == -1 && errno == EINTR); 1207c1bb86cdSEric Blake 1208c1bb86cdSEric Blake if (len == -1) { 1209c1bb86cdSEric Blake return -errno; 1210c1bb86cdSEric Blake } 1211c1bb86cdSEric Blake return len; 1212c1bb86cdSEric Blake } 1213c1bb86cdSEric Blake 1214c1bb86cdSEric Blake /* 1215c1bb86cdSEric Blake * Read/writes the data to/from a given linear buffer. 1216c1bb86cdSEric Blake * 1217c1bb86cdSEric Blake * Returns the number of bytes handles or -errno in case of an error. Short 1218c1bb86cdSEric Blake * reads are only returned if the end of the file is reached. 1219c1bb86cdSEric Blake */ 1220c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) 1221c1bb86cdSEric Blake { 1222c1bb86cdSEric Blake ssize_t offset = 0; 1223c1bb86cdSEric Blake ssize_t len; 1224c1bb86cdSEric Blake 1225c1bb86cdSEric Blake while (offset < aiocb->aio_nbytes) { 1226c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1227c1bb86cdSEric Blake len = pwrite(aiocb->aio_fildes, 1228c1bb86cdSEric Blake (const char *)buf + offset, 1229c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1230c1bb86cdSEric Blake aiocb->aio_offset + offset); 1231c1bb86cdSEric Blake } else { 1232c1bb86cdSEric Blake len = pread(aiocb->aio_fildes, 1233c1bb86cdSEric Blake buf + offset, 1234c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1235c1bb86cdSEric Blake aiocb->aio_offset + offset); 1236c1bb86cdSEric Blake } 1237c1bb86cdSEric Blake if (len == -1 && errno == EINTR) { 1238c1bb86cdSEric Blake continue; 1239c1bb86cdSEric Blake } else if (len == -1 && errno == EINVAL && 1240c1bb86cdSEric Blake (aiocb->bs->open_flags & BDRV_O_NOCACHE) && 1241c1bb86cdSEric Blake !(aiocb->aio_type & QEMU_AIO_WRITE) && 1242c1bb86cdSEric Blake offset > 0) { 1243c1bb86cdSEric Blake /* O_DIRECT pread() may fail with EINVAL when offset is unaligned 1244c1bb86cdSEric Blake * after a short read. Assume that O_DIRECT short reads only occur 1245c1bb86cdSEric Blake * at EOF. Therefore this is a short read, not an I/O error. 1246c1bb86cdSEric Blake */ 1247c1bb86cdSEric Blake break; 1248c1bb86cdSEric Blake } else if (len == -1) { 1249c1bb86cdSEric Blake offset = -errno; 1250c1bb86cdSEric Blake break; 1251c1bb86cdSEric Blake } else if (len == 0) { 1252c1bb86cdSEric Blake break; 1253c1bb86cdSEric Blake } 1254c1bb86cdSEric Blake offset += len; 1255c1bb86cdSEric Blake } 1256c1bb86cdSEric Blake 1257c1bb86cdSEric Blake return offset; 1258c1bb86cdSEric Blake } 1259c1bb86cdSEric Blake 1260c1bb86cdSEric Blake static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb) 1261c1bb86cdSEric Blake { 1262c1bb86cdSEric Blake ssize_t nbytes; 1263c1bb86cdSEric Blake char *buf; 1264c1bb86cdSEric Blake 1265c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { 1266c1bb86cdSEric Blake /* 1267c1bb86cdSEric Blake * If there is just a single buffer, and it is properly aligned 1268c1bb86cdSEric Blake * we can just use plain pread/pwrite without any problems. 1269c1bb86cdSEric Blake */ 1270c1bb86cdSEric Blake if (aiocb->aio_niov == 1) { 1271c1bb86cdSEric Blake return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base); 1272c1bb86cdSEric Blake } 1273c1bb86cdSEric Blake /* 1274c1bb86cdSEric Blake * We have more than one iovec, and all are properly aligned. 1275c1bb86cdSEric Blake * 1276c1bb86cdSEric Blake * Try preadv/pwritev first and fall back to linearizing the 1277c1bb86cdSEric Blake * buffer if it's not supported. 1278c1bb86cdSEric Blake */ 1279c1bb86cdSEric Blake if (preadv_present) { 1280c1bb86cdSEric Blake nbytes = handle_aiocb_rw_vector(aiocb); 1281c1bb86cdSEric Blake if (nbytes == aiocb->aio_nbytes || 1282c1bb86cdSEric Blake (nbytes < 0 && nbytes != -ENOSYS)) { 1283c1bb86cdSEric Blake return nbytes; 1284c1bb86cdSEric Blake } 1285c1bb86cdSEric Blake preadv_present = false; 1286c1bb86cdSEric Blake } 1287c1bb86cdSEric Blake 1288c1bb86cdSEric Blake /* 1289c1bb86cdSEric Blake * XXX(hch): short read/write. no easy way to handle the reminder 1290c1bb86cdSEric Blake * using these interfaces. For now retry using plain 1291c1bb86cdSEric Blake * pread/pwrite? 1292c1bb86cdSEric Blake */ 1293c1bb86cdSEric Blake } 1294c1bb86cdSEric Blake 1295c1bb86cdSEric Blake /* 1296c1bb86cdSEric Blake * Ok, we have to do it the hard way, copy all segments into 1297c1bb86cdSEric Blake * a single aligned buffer. 1298c1bb86cdSEric Blake */ 1299c1bb86cdSEric Blake buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); 1300c1bb86cdSEric Blake if (buf == NULL) { 1301c1bb86cdSEric Blake return -ENOMEM; 1302c1bb86cdSEric Blake } 1303c1bb86cdSEric Blake 1304c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1305c1bb86cdSEric Blake char *p = buf; 1306c1bb86cdSEric Blake int i; 1307c1bb86cdSEric Blake 1308c1bb86cdSEric Blake for (i = 0; i < aiocb->aio_niov; ++i) { 1309c1bb86cdSEric Blake memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len); 1310c1bb86cdSEric Blake p += aiocb->aio_iov[i].iov_len; 1311c1bb86cdSEric Blake } 1312c1bb86cdSEric Blake assert(p - buf == aiocb->aio_nbytes); 1313c1bb86cdSEric Blake } 1314c1bb86cdSEric Blake 1315c1bb86cdSEric Blake nbytes = handle_aiocb_rw_linear(aiocb, buf); 1316c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { 1317c1bb86cdSEric Blake char *p = buf; 1318c1bb86cdSEric Blake size_t count = aiocb->aio_nbytes, copy; 1319c1bb86cdSEric Blake int i; 1320c1bb86cdSEric Blake 1321c1bb86cdSEric Blake for (i = 0; i < aiocb->aio_niov && count; ++i) { 1322c1bb86cdSEric Blake copy = count; 1323c1bb86cdSEric Blake if (copy > aiocb->aio_iov[i].iov_len) { 1324c1bb86cdSEric Blake copy = aiocb->aio_iov[i].iov_len; 1325c1bb86cdSEric Blake } 1326c1bb86cdSEric Blake memcpy(aiocb->aio_iov[i].iov_base, p, copy); 1327c1bb86cdSEric Blake assert(count >= copy); 1328c1bb86cdSEric Blake p += copy; 1329c1bb86cdSEric Blake count -= copy; 1330c1bb86cdSEric Blake } 1331c1bb86cdSEric Blake assert(count == 0); 1332c1bb86cdSEric Blake } 1333c1bb86cdSEric Blake qemu_vfree(buf); 1334c1bb86cdSEric Blake 1335c1bb86cdSEric Blake return nbytes; 1336c1bb86cdSEric Blake } 1337c1bb86cdSEric Blake 1338c1bb86cdSEric Blake #ifdef CONFIG_XFS 1339c1bb86cdSEric Blake static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes) 1340c1bb86cdSEric Blake { 1341c1bb86cdSEric Blake struct xfs_flock64 fl; 1342c1bb86cdSEric Blake int err; 1343c1bb86cdSEric Blake 1344c1bb86cdSEric Blake memset(&fl, 0, sizeof(fl)); 1345c1bb86cdSEric Blake fl.l_whence = SEEK_SET; 1346c1bb86cdSEric Blake fl.l_start = offset; 1347c1bb86cdSEric Blake fl.l_len = bytes; 1348c1bb86cdSEric Blake 1349c1bb86cdSEric Blake if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) { 1350c1bb86cdSEric Blake err = errno; 1351c1bb86cdSEric Blake DPRINTF("cannot write zero range (%s)\n", strerror(errno)); 1352c1bb86cdSEric Blake return -err; 1353c1bb86cdSEric Blake } 1354c1bb86cdSEric Blake 1355c1bb86cdSEric Blake return 0; 1356c1bb86cdSEric Blake } 1357c1bb86cdSEric Blake 1358c1bb86cdSEric Blake static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes) 1359c1bb86cdSEric Blake { 1360c1bb86cdSEric Blake struct xfs_flock64 fl; 1361c1bb86cdSEric Blake int err; 1362c1bb86cdSEric Blake 1363c1bb86cdSEric Blake memset(&fl, 0, sizeof(fl)); 1364c1bb86cdSEric Blake fl.l_whence = SEEK_SET; 1365c1bb86cdSEric Blake fl.l_start = offset; 1366c1bb86cdSEric Blake fl.l_len = bytes; 1367c1bb86cdSEric Blake 1368c1bb86cdSEric Blake if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) { 1369c1bb86cdSEric Blake err = errno; 1370c1bb86cdSEric Blake DPRINTF("cannot punch hole (%s)\n", strerror(errno)); 1371c1bb86cdSEric Blake return -err; 1372c1bb86cdSEric Blake } 1373c1bb86cdSEric Blake 1374c1bb86cdSEric Blake return 0; 1375c1bb86cdSEric Blake } 1376c1bb86cdSEric Blake #endif 1377c1bb86cdSEric Blake 1378c1bb86cdSEric Blake static int translate_err(int err) 1379c1bb86cdSEric Blake { 1380c1bb86cdSEric Blake if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || 1381c1bb86cdSEric Blake err == -ENOTTY) { 1382c1bb86cdSEric Blake err = -ENOTSUP; 1383c1bb86cdSEric Blake } 1384c1bb86cdSEric Blake return err; 1385c1bb86cdSEric Blake } 1386c1bb86cdSEric Blake 1387c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 1388c1bb86cdSEric Blake static int do_fallocate(int fd, int mode, off_t offset, off_t len) 1389c1bb86cdSEric Blake { 1390c1bb86cdSEric Blake do { 1391c1bb86cdSEric Blake if (fallocate(fd, mode, offset, len) == 0) { 1392c1bb86cdSEric Blake return 0; 1393c1bb86cdSEric Blake } 1394c1bb86cdSEric Blake } while (errno == EINTR); 1395c1bb86cdSEric Blake return translate_err(-errno); 1396c1bb86cdSEric Blake } 1397c1bb86cdSEric Blake #endif 1398c1bb86cdSEric Blake 1399c1bb86cdSEric Blake static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) 1400c1bb86cdSEric Blake { 1401c1bb86cdSEric Blake int ret = -ENOTSUP; 1402c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1403c1bb86cdSEric Blake 1404c1bb86cdSEric Blake if (!s->has_write_zeroes) { 1405c1bb86cdSEric Blake return -ENOTSUP; 1406c1bb86cdSEric Blake } 1407c1bb86cdSEric Blake 1408c1bb86cdSEric Blake #ifdef BLKZEROOUT 1409c1bb86cdSEric Blake do { 1410c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1411c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { 1412c1bb86cdSEric Blake return 0; 1413c1bb86cdSEric Blake } 1414c1bb86cdSEric Blake } while (errno == EINTR); 1415c1bb86cdSEric Blake 1416c1bb86cdSEric Blake ret = translate_err(-errno); 1417c1bb86cdSEric Blake #endif 1418c1bb86cdSEric Blake 1419c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1420c1bb86cdSEric Blake s->has_write_zeroes = false; 1421c1bb86cdSEric Blake } 1422c1bb86cdSEric Blake return ret; 1423c1bb86cdSEric Blake } 1424c1bb86cdSEric Blake 1425c1bb86cdSEric Blake static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb) 1426c1bb86cdSEric Blake { 1427c1bb86cdSEric Blake #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS) 1428c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1429c1bb86cdSEric Blake #endif 143070d9110bSDenis V. Lunev #ifdef CONFIG_FALLOCATE 143170d9110bSDenis V. Lunev int64_t len; 143270d9110bSDenis V. Lunev #endif 1433c1bb86cdSEric Blake 1434c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1435c1bb86cdSEric Blake return handle_aiocb_write_zeroes_block(aiocb); 1436c1bb86cdSEric Blake } 1437c1bb86cdSEric Blake 1438c1bb86cdSEric Blake #ifdef CONFIG_XFS 1439c1bb86cdSEric Blake if (s->is_xfs) { 1440c1bb86cdSEric Blake return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes); 1441c1bb86cdSEric Blake } 1442c1bb86cdSEric Blake #endif 1443c1bb86cdSEric Blake 1444c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_ZERO_RANGE 1445c1bb86cdSEric Blake if (s->has_write_zeroes) { 1446c1bb86cdSEric Blake int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, 1447c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1448c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1449c1bb86cdSEric Blake return ret; 1450c1bb86cdSEric Blake } 1451c1bb86cdSEric Blake s->has_write_zeroes = false; 1452c1bb86cdSEric Blake } 1453c1bb86cdSEric Blake #endif 1454c1bb86cdSEric Blake 1455c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1456c1bb86cdSEric Blake if (s->has_discard && s->has_fallocate) { 1457c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 1458c1bb86cdSEric Blake FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1459c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1460c1bb86cdSEric Blake if (ret == 0) { 1461c1bb86cdSEric Blake ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1462c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1463c1bb86cdSEric Blake return ret; 1464c1bb86cdSEric Blake } 1465c1bb86cdSEric Blake s->has_fallocate = false; 1466c1bb86cdSEric Blake } else if (ret != -ENOTSUP) { 1467c1bb86cdSEric Blake return ret; 1468c1bb86cdSEric Blake } else { 1469c1bb86cdSEric Blake s->has_discard = false; 1470c1bb86cdSEric Blake } 1471c1bb86cdSEric Blake } 1472c1bb86cdSEric Blake #endif 1473c1bb86cdSEric Blake 1474c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 147570d9110bSDenis V. Lunev /* Last resort: we are trying to extend the file with zeroed data. This 147670d9110bSDenis V. Lunev * can be done via fallocate(fd, 0) */ 147770d9110bSDenis V. Lunev len = bdrv_getlength(aiocb->bs); 147870d9110bSDenis V. Lunev if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { 1479c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1480c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1481c1bb86cdSEric Blake return ret; 1482c1bb86cdSEric Blake } 1483c1bb86cdSEric Blake s->has_fallocate = false; 1484c1bb86cdSEric Blake } 1485c1bb86cdSEric Blake #endif 1486c1bb86cdSEric Blake 1487c1bb86cdSEric Blake return -ENOTSUP; 1488c1bb86cdSEric Blake } 1489c1bb86cdSEric Blake 14901efad060SFam Zheng #ifndef HAVE_COPY_FILE_RANGE 14911efad060SFam Zheng static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, 14921efad060SFam Zheng off_t *out_off, size_t len, unsigned int flags) 14931efad060SFam Zheng { 14941efad060SFam Zheng #ifdef __NR_copy_file_range 14951efad060SFam Zheng return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, 14961efad060SFam Zheng out_off, len, flags); 14971efad060SFam Zheng #else 14981efad060SFam Zheng errno = ENOSYS; 14991efad060SFam Zheng return -1; 15001efad060SFam Zheng #endif 15011efad060SFam Zheng } 15021efad060SFam Zheng #endif 15031efad060SFam Zheng 15041efad060SFam Zheng static ssize_t handle_aiocb_copy_range(RawPosixAIOData *aiocb) 15051efad060SFam Zheng { 15061efad060SFam Zheng uint64_t bytes = aiocb->aio_nbytes; 15071efad060SFam Zheng off_t in_off = aiocb->aio_offset; 15081efad060SFam Zheng off_t out_off = aiocb->aio_offset2; 15091efad060SFam Zheng 15101efad060SFam Zheng while (bytes) { 15111efad060SFam Zheng ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, 15121efad060SFam Zheng aiocb->aio_fd2, &out_off, 15131efad060SFam Zheng bytes, 0); 1514ecc983a5SFam Zheng trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, 1515ecc983a5SFam Zheng aiocb->aio_fd2, out_off, bytes, 0, ret); 1516c436e3d0SFam Zheng if (ret == 0) { 1517c436e3d0SFam Zheng /* No progress (e.g. when beyond EOF), let the caller fall back to 1518c436e3d0SFam Zheng * buffer I/O. */ 1519c436e3d0SFam Zheng return -ENOSPC; 15201efad060SFam Zheng } 15211efad060SFam Zheng if (ret < 0) { 1522c436e3d0SFam Zheng switch (errno) { 1523c436e3d0SFam Zheng case ENOSYS: 15241efad060SFam Zheng return -ENOTSUP; 1525c436e3d0SFam Zheng case EINTR: 1526c436e3d0SFam Zheng continue; 1527c436e3d0SFam Zheng default: 15281efad060SFam Zheng return -errno; 15291efad060SFam Zheng } 15301efad060SFam Zheng } 15311efad060SFam Zheng bytes -= ret; 15321efad060SFam Zheng } 15331efad060SFam Zheng return 0; 15341efad060SFam Zheng } 15351efad060SFam Zheng 1536c1bb86cdSEric Blake static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb) 1537c1bb86cdSEric Blake { 1538c1bb86cdSEric Blake int ret = -EOPNOTSUPP; 1539c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1540c1bb86cdSEric Blake 1541c1bb86cdSEric Blake if (!s->has_discard) { 1542c1bb86cdSEric Blake return -ENOTSUP; 1543c1bb86cdSEric Blake } 1544c1bb86cdSEric Blake 1545c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1546c1bb86cdSEric Blake #ifdef BLKDISCARD 1547c1bb86cdSEric Blake do { 1548c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1549c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { 1550c1bb86cdSEric Blake return 0; 1551c1bb86cdSEric Blake } 1552c1bb86cdSEric Blake } while (errno == EINTR); 1553c1bb86cdSEric Blake 1554c1bb86cdSEric Blake ret = -errno; 1555c1bb86cdSEric Blake #endif 1556c1bb86cdSEric Blake } else { 1557c1bb86cdSEric Blake #ifdef CONFIG_XFS 1558c1bb86cdSEric Blake if (s->is_xfs) { 1559c1bb86cdSEric Blake return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes); 1560c1bb86cdSEric Blake } 1561c1bb86cdSEric Blake #endif 1562c1bb86cdSEric Blake 1563c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1564c1bb86cdSEric Blake ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1565c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1566c1bb86cdSEric Blake #endif 1567c1bb86cdSEric Blake } 1568c1bb86cdSEric Blake 1569c1bb86cdSEric Blake ret = translate_err(ret); 1570c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1571c1bb86cdSEric Blake s->has_discard = false; 1572c1bb86cdSEric Blake } 1573c1bb86cdSEric Blake return ret; 1574c1bb86cdSEric Blake } 1575c1bb86cdSEric Blake 157693f4e2ffSKevin Wolf static int handle_aiocb_truncate(RawPosixAIOData *aiocb) 157793f4e2ffSKevin Wolf { 157893f4e2ffSKevin Wolf int result = 0; 157993f4e2ffSKevin Wolf int64_t current_length = 0; 158093f4e2ffSKevin Wolf char *buf = NULL; 158193f4e2ffSKevin Wolf struct stat st; 158293f4e2ffSKevin Wolf int fd = aiocb->aio_fildes; 158393f4e2ffSKevin Wolf int64_t offset = aiocb->aio_offset; 158493f4e2ffSKevin Wolf Error **errp = aiocb->errp; 158593f4e2ffSKevin Wolf 158693f4e2ffSKevin Wolf if (fstat(fd, &st) < 0) { 158793f4e2ffSKevin Wolf result = -errno; 158893f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not stat file"); 158993f4e2ffSKevin Wolf return result; 159093f4e2ffSKevin Wolf } 159193f4e2ffSKevin Wolf 159293f4e2ffSKevin Wolf current_length = st.st_size; 159393f4e2ffSKevin Wolf if (current_length > offset && aiocb->prealloc != PREALLOC_MODE_OFF) { 159493f4e2ffSKevin Wolf error_setg(errp, "Cannot use preallocation for shrinking files"); 159593f4e2ffSKevin Wolf return -ENOTSUP; 159693f4e2ffSKevin Wolf } 159793f4e2ffSKevin Wolf 159893f4e2ffSKevin Wolf switch (aiocb->prealloc) { 159993f4e2ffSKevin Wolf #ifdef CONFIG_POSIX_FALLOCATE 160093f4e2ffSKevin Wolf case PREALLOC_MODE_FALLOC: 160193f4e2ffSKevin Wolf /* 160293f4e2ffSKevin Wolf * Truncating before posix_fallocate() makes it about twice slower on 160393f4e2ffSKevin Wolf * file systems that do not support fallocate(), trying to check if a 160493f4e2ffSKevin Wolf * block is allocated before allocating it, so don't do that here. 160593f4e2ffSKevin Wolf */ 160693f4e2ffSKevin Wolf if (offset != current_length) { 160793f4e2ffSKevin Wolf result = -posix_fallocate(fd, current_length, 160893f4e2ffSKevin Wolf offset - current_length); 160993f4e2ffSKevin Wolf if (result != 0) { 161093f4e2ffSKevin Wolf /* posix_fallocate() doesn't set errno. */ 161193f4e2ffSKevin Wolf error_setg_errno(errp, -result, 161293f4e2ffSKevin Wolf "Could not preallocate new data"); 161393f4e2ffSKevin Wolf } 161493f4e2ffSKevin Wolf } else { 161593f4e2ffSKevin Wolf result = 0; 161693f4e2ffSKevin Wolf } 161793f4e2ffSKevin Wolf goto out; 161893f4e2ffSKevin Wolf #endif 161993f4e2ffSKevin Wolf case PREALLOC_MODE_FULL: 162093f4e2ffSKevin Wolf { 162193f4e2ffSKevin Wolf int64_t num = 0, left = offset - current_length; 162293f4e2ffSKevin Wolf off_t seek_result; 162393f4e2ffSKevin Wolf 162493f4e2ffSKevin Wolf /* 162593f4e2ffSKevin Wolf * Knowing the final size from the beginning could allow the file 162693f4e2ffSKevin Wolf * system driver to do less allocations and possibly avoid 162793f4e2ffSKevin Wolf * fragmentation of the file. 162893f4e2ffSKevin Wolf */ 162993f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 163093f4e2ffSKevin Wolf result = -errno; 163193f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 163293f4e2ffSKevin Wolf goto out; 163393f4e2ffSKevin Wolf } 163493f4e2ffSKevin Wolf 163593f4e2ffSKevin Wolf buf = g_malloc0(65536); 163693f4e2ffSKevin Wolf 163793f4e2ffSKevin Wolf seek_result = lseek(fd, current_length, SEEK_SET); 163893f4e2ffSKevin Wolf if (seek_result < 0) { 163993f4e2ffSKevin Wolf result = -errno; 164093f4e2ffSKevin Wolf error_setg_errno(errp, -result, 164193f4e2ffSKevin Wolf "Failed to seek to the old end of file"); 164293f4e2ffSKevin Wolf goto out; 164393f4e2ffSKevin Wolf } 164493f4e2ffSKevin Wolf 164593f4e2ffSKevin Wolf while (left > 0) { 164693f4e2ffSKevin Wolf num = MIN(left, 65536); 164793f4e2ffSKevin Wolf result = write(fd, buf, num); 164893f4e2ffSKevin Wolf if (result < 0) { 164993f4e2ffSKevin Wolf result = -errno; 165093f4e2ffSKevin Wolf error_setg_errno(errp, -result, 165193f4e2ffSKevin Wolf "Could not write zeros for preallocation"); 165293f4e2ffSKevin Wolf goto out; 165393f4e2ffSKevin Wolf } 165493f4e2ffSKevin Wolf left -= result; 165593f4e2ffSKevin Wolf } 165693f4e2ffSKevin Wolf if (result >= 0) { 165793f4e2ffSKevin Wolf result = fsync(fd); 165893f4e2ffSKevin Wolf if (result < 0) { 165993f4e2ffSKevin Wolf result = -errno; 166093f4e2ffSKevin Wolf error_setg_errno(errp, -result, 166193f4e2ffSKevin Wolf "Could not flush file to disk"); 166293f4e2ffSKevin Wolf goto out; 166393f4e2ffSKevin Wolf } 166493f4e2ffSKevin Wolf } 166593f4e2ffSKevin Wolf goto out; 166693f4e2ffSKevin Wolf } 166793f4e2ffSKevin Wolf case PREALLOC_MODE_OFF: 166893f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 166993f4e2ffSKevin Wolf result = -errno; 167093f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 167193f4e2ffSKevin Wolf } 167293f4e2ffSKevin Wolf return result; 167393f4e2ffSKevin Wolf default: 167493f4e2ffSKevin Wolf result = -ENOTSUP; 167593f4e2ffSKevin Wolf error_setg(errp, "Unsupported preallocation mode: %s", 167693f4e2ffSKevin Wolf PreallocMode_str(aiocb->prealloc)); 167793f4e2ffSKevin Wolf return result; 167893f4e2ffSKevin Wolf } 167993f4e2ffSKevin Wolf 168093f4e2ffSKevin Wolf out: 168193f4e2ffSKevin Wolf if (result < 0) { 168293f4e2ffSKevin Wolf if (ftruncate(fd, current_length) < 0) { 168393f4e2ffSKevin Wolf error_report("Failed to restore old file length: %s", 168493f4e2ffSKevin Wolf strerror(errno)); 168593f4e2ffSKevin Wolf } 168693f4e2ffSKevin Wolf } 168793f4e2ffSKevin Wolf 168893f4e2ffSKevin Wolf g_free(buf); 168993f4e2ffSKevin Wolf return result; 169093f4e2ffSKevin Wolf } 169193f4e2ffSKevin Wolf 1692c1bb86cdSEric Blake static int aio_worker(void *arg) 1693c1bb86cdSEric Blake { 1694c1bb86cdSEric Blake RawPosixAIOData *aiocb = arg; 1695c1bb86cdSEric Blake ssize_t ret = 0; 1696c1bb86cdSEric Blake 1697c1bb86cdSEric Blake switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) { 1698c1bb86cdSEric Blake case QEMU_AIO_READ: 1699c1bb86cdSEric Blake ret = handle_aiocb_rw(aiocb); 1700c1bb86cdSEric Blake if (ret >= 0 && ret < aiocb->aio_nbytes) { 1701c1bb86cdSEric Blake iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret, 1702c1bb86cdSEric Blake 0, aiocb->aio_nbytes - ret); 1703c1bb86cdSEric Blake 1704c1bb86cdSEric Blake ret = aiocb->aio_nbytes; 1705c1bb86cdSEric Blake } 1706c1bb86cdSEric Blake if (ret == aiocb->aio_nbytes) { 1707c1bb86cdSEric Blake ret = 0; 1708c1bb86cdSEric Blake } else if (ret >= 0 && ret < aiocb->aio_nbytes) { 1709c1bb86cdSEric Blake ret = -EINVAL; 1710c1bb86cdSEric Blake } 1711c1bb86cdSEric Blake break; 1712c1bb86cdSEric Blake case QEMU_AIO_WRITE: 1713c1bb86cdSEric Blake ret = handle_aiocb_rw(aiocb); 1714c1bb86cdSEric Blake if (ret == aiocb->aio_nbytes) { 1715c1bb86cdSEric Blake ret = 0; 1716c1bb86cdSEric Blake } else if (ret >= 0 && ret < aiocb->aio_nbytes) { 1717c1bb86cdSEric Blake ret = -EINVAL; 1718c1bb86cdSEric Blake } 1719c1bb86cdSEric Blake break; 1720c1bb86cdSEric Blake case QEMU_AIO_FLUSH: 1721c1bb86cdSEric Blake ret = handle_aiocb_flush(aiocb); 1722c1bb86cdSEric Blake break; 1723c1bb86cdSEric Blake case QEMU_AIO_IOCTL: 1724c1bb86cdSEric Blake ret = handle_aiocb_ioctl(aiocb); 1725c1bb86cdSEric Blake break; 1726c1bb86cdSEric Blake case QEMU_AIO_DISCARD: 1727c1bb86cdSEric Blake ret = handle_aiocb_discard(aiocb); 1728c1bb86cdSEric Blake break; 1729c1bb86cdSEric Blake case QEMU_AIO_WRITE_ZEROES: 1730c1bb86cdSEric Blake ret = handle_aiocb_write_zeroes(aiocb); 1731c1bb86cdSEric Blake break; 17321efad060SFam Zheng case QEMU_AIO_COPY_RANGE: 17331efad060SFam Zheng ret = handle_aiocb_copy_range(aiocb); 17341efad060SFam Zheng break; 173593f4e2ffSKevin Wolf case QEMU_AIO_TRUNCATE: 173693f4e2ffSKevin Wolf ret = handle_aiocb_truncate(aiocb); 173793f4e2ffSKevin Wolf break; 1738c1bb86cdSEric Blake default: 1739c1bb86cdSEric Blake fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type); 1740c1bb86cdSEric Blake ret = -EINVAL; 1741c1bb86cdSEric Blake break; 1742c1bb86cdSEric Blake } 1743c1bb86cdSEric Blake 1744c1bb86cdSEric Blake g_free(aiocb); 1745c1bb86cdSEric Blake return ret; 1746c1bb86cdSEric Blake } 1747c1bb86cdSEric Blake 17481efad060SFam Zheng static int paio_submit_co_full(BlockDriverState *bs, int fd, 17491efad060SFam Zheng int64_t offset, int fd2, int64_t offset2, 17501efad060SFam Zheng QEMUIOVector *qiov, 1751f5a5ca79SManos Pitsidianakis int bytes, int type) 1752c1bb86cdSEric Blake { 1753c1bb86cdSEric Blake RawPosixAIOData *acb = g_new(RawPosixAIOData, 1); 1754c1bb86cdSEric Blake ThreadPool *pool; 1755c1bb86cdSEric Blake 1756c1bb86cdSEric Blake acb->bs = bs; 1757c1bb86cdSEric Blake acb->aio_type = type; 1758c1bb86cdSEric Blake acb->aio_fildes = fd; 17591efad060SFam Zheng acb->aio_fd2 = fd2; 17601efad060SFam Zheng acb->aio_offset2 = offset2; 1761c1bb86cdSEric Blake 1762f5a5ca79SManos Pitsidianakis acb->aio_nbytes = bytes; 1763c1bb86cdSEric Blake acb->aio_offset = offset; 1764c1bb86cdSEric Blake 1765c1bb86cdSEric Blake if (qiov) { 1766c1bb86cdSEric Blake acb->aio_iov = qiov->iov; 1767c1bb86cdSEric Blake acb->aio_niov = qiov->niov; 1768f5a5ca79SManos Pitsidianakis assert(qiov->size == bytes); 1769c1bb86cdSEric Blake } 1770c1bb86cdSEric Blake 1771f8a30874SFam Zheng trace_file_paio_submit_co(offset, bytes, type); 1772c1bb86cdSEric Blake pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 1773c1bb86cdSEric Blake return thread_pool_submit_co(pool, aio_worker, acb); 1774c1bb86cdSEric Blake } 1775c1bb86cdSEric Blake 17761efad060SFam Zheng static inline int paio_submit_co(BlockDriverState *bs, int fd, 17771efad060SFam Zheng int64_t offset, QEMUIOVector *qiov, 17781efad060SFam Zheng int bytes, int type) 17791efad060SFam Zheng { 17801efad060SFam Zheng return paio_submit_co_full(bs, fd, offset, -1, 0, qiov, bytes, type); 17811efad060SFam Zheng } 17821efad060SFam Zheng 1783c1bb86cdSEric Blake static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, 1784c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, int type) 1785c1bb86cdSEric Blake { 1786c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1787c1bb86cdSEric Blake 1788c1bb86cdSEric Blake if (fd_open(bs) < 0) 1789c1bb86cdSEric Blake return -EIO; 1790c1bb86cdSEric Blake 1791c1bb86cdSEric Blake /* 1792c1bb86cdSEric Blake * Check if the underlying device requires requests to be aligned, 1793c1bb86cdSEric Blake * and if the request we are trying to submit is aligned or not. 1794c1bb86cdSEric Blake * If this is the case tell the low-level driver that it needs 1795c1bb86cdSEric Blake * to copy the buffer. 1796c1bb86cdSEric Blake */ 1797c1bb86cdSEric Blake if (s->needs_alignment) { 1798c1bb86cdSEric Blake if (!bdrv_qiov_is_aligned(bs, qiov)) { 1799c1bb86cdSEric Blake type |= QEMU_AIO_MISALIGNED; 1800c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1801c1bb86cdSEric Blake } else if (s->use_linux_aio) { 1802c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1803c1bb86cdSEric Blake assert(qiov->size == bytes); 1804c1bb86cdSEric Blake return laio_co_submit(bs, aio, s->fd, offset, qiov, type); 1805c1bb86cdSEric Blake #endif 1806c1bb86cdSEric Blake } 1807c1bb86cdSEric Blake } 1808c1bb86cdSEric Blake 1809c1bb86cdSEric Blake return paio_submit_co(bs, s->fd, offset, qiov, bytes, type); 1810c1bb86cdSEric Blake } 1811c1bb86cdSEric Blake 1812c1bb86cdSEric Blake static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset, 1813c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 1814c1bb86cdSEric Blake int flags) 1815c1bb86cdSEric Blake { 1816c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); 1817c1bb86cdSEric Blake } 1818c1bb86cdSEric Blake 1819c1bb86cdSEric Blake static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset, 1820c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 1821c1bb86cdSEric Blake int flags) 1822c1bb86cdSEric Blake { 1823c1bb86cdSEric Blake assert(flags == 0); 1824c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); 1825c1bb86cdSEric Blake } 1826c1bb86cdSEric Blake 1827c1bb86cdSEric Blake static void raw_aio_plug(BlockDriverState *bs) 1828c1bb86cdSEric Blake { 1829c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1830c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1831c1bb86cdSEric Blake if (s->use_linux_aio) { 1832c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1833c1bb86cdSEric Blake laio_io_plug(bs, aio); 1834c1bb86cdSEric Blake } 1835c1bb86cdSEric Blake #endif 1836c1bb86cdSEric Blake } 1837c1bb86cdSEric Blake 1838c1bb86cdSEric Blake static void raw_aio_unplug(BlockDriverState *bs) 1839c1bb86cdSEric Blake { 1840c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1841c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1842c1bb86cdSEric Blake if (s->use_linux_aio) { 1843c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1844c1bb86cdSEric Blake laio_io_unplug(bs, aio); 1845c1bb86cdSEric Blake } 1846c1bb86cdSEric Blake #endif 1847c1bb86cdSEric Blake } 1848c1bb86cdSEric Blake 184933d70fb6SKevin Wolf static int raw_co_flush_to_disk(BlockDriverState *bs) 1850c1bb86cdSEric Blake { 1851c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 185233d70fb6SKevin Wolf int ret; 1853c1bb86cdSEric Blake 185433d70fb6SKevin Wolf ret = fd_open(bs); 185533d70fb6SKevin Wolf if (ret < 0) { 185633d70fb6SKevin Wolf return ret; 185733d70fb6SKevin Wolf } 1858c1bb86cdSEric Blake 185933d70fb6SKevin Wolf return paio_submit_co(bs, s->fd, 0, NULL, 0, QEMU_AIO_FLUSH); 1860c1bb86cdSEric Blake } 1861c1bb86cdSEric Blake 1862ed6e2161SNishanth Aravamudan static void raw_aio_attach_aio_context(BlockDriverState *bs, 1863ed6e2161SNishanth Aravamudan AioContext *new_context) 1864ed6e2161SNishanth Aravamudan { 1865ed6e2161SNishanth Aravamudan #ifdef CONFIG_LINUX_AIO 1866ed6e2161SNishanth Aravamudan BDRVRawState *s = bs->opaque; 1867ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 1868ed6e2161SNishanth Aravamudan Error *local_err; 1869ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(new_context, &local_err)) { 1870ed6e2161SNishanth Aravamudan error_reportf_err(local_err, "Unable to use native AIO, " 1871ed6e2161SNishanth Aravamudan "falling back to thread pool: "); 1872ed6e2161SNishanth Aravamudan s->use_linux_aio = false; 1873ed6e2161SNishanth Aravamudan } 1874ed6e2161SNishanth Aravamudan } 1875ed6e2161SNishanth Aravamudan #endif 1876ed6e2161SNishanth Aravamudan } 1877ed6e2161SNishanth Aravamudan 1878c1bb86cdSEric Blake static void raw_close(BlockDriverState *bs) 1879c1bb86cdSEric Blake { 1880c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1881c1bb86cdSEric Blake 1882c1bb86cdSEric Blake if (s->fd >= 0) { 1883c1bb86cdSEric Blake qemu_close(s->fd); 1884c1bb86cdSEric Blake s->fd = -1; 1885c1bb86cdSEric Blake } 1886244a5668SFam Zheng if (s->lock_fd >= 0) { 1887244a5668SFam Zheng qemu_close(s->lock_fd); 1888244a5668SFam Zheng s->lock_fd = -1; 1889244a5668SFam Zheng } 1890c1bb86cdSEric Blake } 1891c1bb86cdSEric Blake 1892d0bc9e5dSMax Reitz /** 1893d0bc9e5dSMax Reitz * Truncates the given regular file @fd to @offset and, when growing, fills the 1894d0bc9e5dSMax Reitz * new space according to @prealloc. 1895d0bc9e5dSMax Reitz * 1896d0bc9e5dSMax Reitz * Returns: 0 on success, -errno on failure. 1897d0bc9e5dSMax Reitz */ 189893f4e2ffSKevin Wolf static int coroutine_fn 189993f4e2ffSKevin Wolf raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, 190093f4e2ffSKevin Wolf PreallocMode prealloc, Error **errp) 19019f63b07eSMax Reitz { 190293f4e2ffSKevin Wolf RawPosixAIOData *acb = g_new(RawPosixAIOData, 1); 190393f4e2ffSKevin Wolf ThreadPool *pool; 1904d0bc9e5dSMax Reitz 190593f4e2ffSKevin Wolf *acb = (RawPosixAIOData) { 190693f4e2ffSKevin Wolf .bs = bs, 190793f4e2ffSKevin Wolf .aio_fildes = fd, 190893f4e2ffSKevin Wolf .aio_type = QEMU_AIO_TRUNCATE, 190993f4e2ffSKevin Wolf .aio_offset = offset, 191093f4e2ffSKevin Wolf .prealloc = prealloc, 191193f4e2ffSKevin Wolf .errp = errp, 191293f4e2ffSKevin Wolf }; 1913d0bc9e5dSMax Reitz 191493f4e2ffSKevin Wolf /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ 191593f4e2ffSKevin Wolf pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 191693f4e2ffSKevin Wolf return thread_pool_submit_co(pool, aio_worker, acb); 19179f63b07eSMax Reitz } 19189f63b07eSMax Reitz 1919061ca8a3SKevin Wolf static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, 19208243ccb7SMax Reitz PreallocMode prealloc, Error **errp) 1921c1bb86cdSEric Blake { 1922c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1923c1bb86cdSEric Blake struct stat st; 1924f59adb32SMax Reitz int ret; 1925c1bb86cdSEric Blake 1926c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 1927f59adb32SMax Reitz ret = -errno; 1928f59adb32SMax Reitz error_setg_errno(errp, -ret, "Failed to fstat() the file"); 1929f59adb32SMax Reitz return ret; 1930c1bb86cdSEric Blake } 1931c1bb86cdSEric Blake 1932c1bb86cdSEric Blake if (S_ISREG(st.st_mode)) { 193393f4e2ffSKevin Wolf return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); 1934c1bb86cdSEric Blake } 193535d72602SMax Reitz 193635d72602SMax Reitz if (prealloc != PREALLOC_MODE_OFF) { 193735d72602SMax Reitz error_setg(errp, "Preallocation mode '%s' unsupported for this " 1938977c736fSMarkus Armbruster "non-regular file", PreallocMode_str(prealloc)); 193935d72602SMax Reitz return -ENOTSUP; 194035d72602SMax Reitz } 194135d72602SMax Reitz 194235d72602SMax Reitz if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 1943c1bb86cdSEric Blake if (offset > raw_getlength(bs)) { 1944f59adb32SMax Reitz error_setg(errp, "Cannot grow device files"); 1945c1bb86cdSEric Blake return -EINVAL; 1946c1bb86cdSEric Blake } 1947c1bb86cdSEric Blake } else { 1948f59adb32SMax Reitz error_setg(errp, "Resizing this file is not supported"); 1949c1bb86cdSEric Blake return -ENOTSUP; 1950c1bb86cdSEric Blake } 1951c1bb86cdSEric Blake 1952c1bb86cdSEric Blake return 0; 1953c1bb86cdSEric Blake } 1954c1bb86cdSEric Blake 1955c1bb86cdSEric Blake #ifdef __OpenBSD__ 1956c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 1957c1bb86cdSEric Blake { 1958c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1959c1bb86cdSEric Blake int fd = s->fd; 1960c1bb86cdSEric Blake struct stat st; 1961c1bb86cdSEric Blake 1962c1bb86cdSEric Blake if (fstat(fd, &st)) 1963c1bb86cdSEric Blake return -errno; 1964c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 1965c1bb86cdSEric Blake struct disklabel dl; 1966c1bb86cdSEric Blake 1967c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 1968c1bb86cdSEric Blake return -errno; 1969c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 1970c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 1971c1bb86cdSEric Blake } else 1972c1bb86cdSEric Blake return st.st_size; 1973c1bb86cdSEric Blake } 1974c1bb86cdSEric Blake #elif defined(__NetBSD__) 1975c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 1976c1bb86cdSEric Blake { 1977c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1978c1bb86cdSEric Blake int fd = s->fd; 1979c1bb86cdSEric Blake struct stat st; 1980c1bb86cdSEric Blake 1981c1bb86cdSEric Blake if (fstat(fd, &st)) 1982c1bb86cdSEric Blake return -errno; 1983c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 1984c1bb86cdSEric Blake struct dkwedge_info dkw; 1985c1bb86cdSEric Blake 1986c1bb86cdSEric Blake if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { 1987c1bb86cdSEric Blake return dkw.dkw_size * 512; 1988c1bb86cdSEric Blake } else { 1989c1bb86cdSEric Blake struct disklabel dl; 1990c1bb86cdSEric Blake 1991c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 1992c1bb86cdSEric Blake return -errno; 1993c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 1994c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 1995c1bb86cdSEric Blake } 1996c1bb86cdSEric Blake } else 1997c1bb86cdSEric Blake return st.st_size; 1998c1bb86cdSEric Blake } 1999c1bb86cdSEric Blake #elif defined(__sun__) 2000c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2001c1bb86cdSEric Blake { 2002c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2003c1bb86cdSEric Blake struct dk_minfo minfo; 2004c1bb86cdSEric Blake int ret; 2005c1bb86cdSEric Blake int64_t size; 2006c1bb86cdSEric Blake 2007c1bb86cdSEric Blake ret = fd_open(bs); 2008c1bb86cdSEric Blake if (ret < 0) { 2009c1bb86cdSEric Blake return ret; 2010c1bb86cdSEric Blake } 2011c1bb86cdSEric Blake 2012c1bb86cdSEric Blake /* 2013c1bb86cdSEric Blake * Use the DKIOCGMEDIAINFO ioctl to read the size. 2014c1bb86cdSEric Blake */ 2015c1bb86cdSEric Blake ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); 2016c1bb86cdSEric Blake if (ret != -1) { 2017c1bb86cdSEric Blake return minfo.dki_lbsize * minfo.dki_capacity; 2018c1bb86cdSEric Blake } 2019c1bb86cdSEric Blake 2020c1bb86cdSEric Blake /* 2021c1bb86cdSEric Blake * There are reports that lseek on some devices fails, but 2022c1bb86cdSEric Blake * irc discussion said that contingency on contingency was overkill. 2023c1bb86cdSEric Blake */ 2024c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2025c1bb86cdSEric Blake if (size < 0) { 2026c1bb86cdSEric Blake return -errno; 2027c1bb86cdSEric Blake } 2028c1bb86cdSEric Blake return size; 2029c1bb86cdSEric Blake } 2030c1bb86cdSEric Blake #elif defined(CONFIG_BSD) 2031c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2032c1bb86cdSEric Blake { 2033c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2034c1bb86cdSEric Blake int fd = s->fd; 2035c1bb86cdSEric Blake int64_t size; 2036c1bb86cdSEric Blake struct stat sb; 2037c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2038c1bb86cdSEric Blake int reopened = 0; 2039c1bb86cdSEric Blake #endif 2040c1bb86cdSEric Blake int ret; 2041c1bb86cdSEric Blake 2042c1bb86cdSEric Blake ret = fd_open(bs); 2043c1bb86cdSEric Blake if (ret < 0) 2044c1bb86cdSEric Blake return ret; 2045c1bb86cdSEric Blake 2046c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2047c1bb86cdSEric Blake again: 2048c1bb86cdSEric Blake #endif 2049c1bb86cdSEric Blake if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { 2050c1bb86cdSEric Blake #ifdef DIOCGMEDIASIZE 2051c1bb86cdSEric Blake if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) 2052c1bb86cdSEric Blake #elif defined(DIOCGPART) 2053c1bb86cdSEric Blake { 2054c1bb86cdSEric Blake struct partinfo pi; 2055c1bb86cdSEric Blake if (ioctl(fd, DIOCGPART, &pi) == 0) 2056c1bb86cdSEric Blake size = pi.media_size; 2057c1bb86cdSEric Blake else 2058c1bb86cdSEric Blake size = 0; 2059c1bb86cdSEric Blake } 2060c1bb86cdSEric Blake if (size == 0) 2061c1bb86cdSEric Blake #endif 2062c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2063c1bb86cdSEric Blake { 2064c1bb86cdSEric Blake uint64_t sectors = 0; 2065c1bb86cdSEric Blake uint32_t sector_size = 0; 2066c1bb86cdSEric Blake 2067c1bb86cdSEric Blake if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 2068c1bb86cdSEric Blake && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { 2069c1bb86cdSEric Blake size = sectors * sector_size; 2070c1bb86cdSEric Blake } else { 2071c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2072c1bb86cdSEric Blake if (size < 0) { 2073c1bb86cdSEric Blake return -errno; 2074c1bb86cdSEric Blake } 2075c1bb86cdSEric Blake } 2076c1bb86cdSEric Blake } 2077c1bb86cdSEric Blake #else 2078c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2079c1bb86cdSEric Blake if (size < 0) { 2080c1bb86cdSEric Blake return -errno; 2081c1bb86cdSEric Blake } 2082c1bb86cdSEric Blake #endif 2083c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 2084c1bb86cdSEric Blake switch(s->type) { 2085c1bb86cdSEric Blake case FTYPE_CD: 2086c1bb86cdSEric Blake /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ 2087c1bb86cdSEric Blake if (size == 2048LL * (unsigned)-1) 2088c1bb86cdSEric Blake size = 0; 2089c1bb86cdSEric Blake /* XXX no disc? maybe we need to reopen... */ 2090c1bb86cdSEric Blake if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { 2091c1bb86cdSEric Blake reopened = 1; 2092c1bb86cdSEric Blake goto again; 2093c1bb86cdSEric Blake } 2094c1bb86cdSEric Blake } 2095c1bb86cdSEric Blake #endif 2096c1bb86cdSEric Blake } else { 2097c1bb86cdSEric Blake size = lseek(fd, 0, SEEK_END); 2098c1bb86cdSEric Blake if (size < 0) { 2099c1bb86cdSEric Blake return -errno; 2100c1bb86cdSEric Blake } 2101c1bb86cdSEric Blake } 2102c1bb86cdSEric Blake return size; 2103c1bb86cdSEric Blake } 2104c1bb86cdSEric Blake #else 2105c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2106c1bb86cdSEric Blake { 2107c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2108c1bb86cdSEric Blake int ret; 2109c1bb86cdSEric Blake int64_t size; 2110c1bb86cdSEric Blake 2111c1bb86cdSEric Blake ret = fd_open(bs); 2112c1bb86cdSEric Blake if (ret < 0) { 2113c1bb86cdSEric Blake return ret; 2114c1bb86cdSEric Blake } 2115c1bb86cdSEric Blake 2116c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2117c1bb86cdSEric Blake if (size < 0) { 2118c1bb86cdSEric Blake return -errno; 2119c1bb86cdSEric Blake } 2120c1bb86cdSEric Blake return size; 2121c1bb86cdSEric Blake } 2122c1bb86cdSEric Blake #endif 2123c1bb86cdSEric Blake 2124c1bb86cdSEric Blake static int64_t raw_get_allocated_file_size(BlockDriverState *bs) 2125c1bb86cdSEric Blake { 2126c1bb86cdSEric Blake struct stat st; 2127c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2128c1bb86cdSEric Blake 2129c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 2130c1bb86cdSEric Blake return -errno; 2131c1bb86cdSEric Blake } 2132c1bb86cdSEric Blake return (int64_t)st.st_blocks * 512; 2133c1bb86cdSEric Blake } 2134c1bb86cdSEric Blake 213593f4e2ffSKevin Wolf static int coroutine_fn 213693f4e2ffSKevin Wolf raw_co_create(BlockdevCreateOptions *options, Error **errp) 2137c1bb86cdSEric Blake { 2138927f11e1SKevin Wolf BlockdevCreateOptionsFile *file_opts; 21397c20c808SMax Reitz Error *local_err = NULL; 2140c1bb86cdSEric Blake int fd; 2141d815efcaSMax Reitz uint64_t perm, shared; 2142c1bb86cdSEric Blake int result = 0; 2143c1bb86cdSEric Blake 2144927f11e1SKevin Wolf /* Validate options and set default values */ 2145927f11e1SKevin Wolf assert(options->driver == BLOCKDEV_DRIVER_FILE); 2146927f11e1SKevin Wolf file_opts = &options->u.file; 2147c1bb86cdSEric Blake 2148927f11e1SKevin Wolf if (!file_opts->has_nocow) { 2149927f11e1SKevin Wolf file_opts->nocow = false; 2150927f11e1SKevin Wolf } 2151927f11e1SKevin Wolf if (!file_opts->has_preallocation) { 2152927f11e1SKevin Wolf file_opts->preallocation = PREALLOC_MODE_OFF; 2153c1bb86cdSEric Blake } 2154c1bb86cdSEric Blake 2155927f11e1SKevin Wolf /* Create file */ 2156b8cf1913SMax Reitz fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_BINARY, 0644); 2157c1bb86cdSEric Blake if (fd < 0) { 2158c1bb86cdSEric Blake result = -errno; 2159c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not create file"); 2160c1bb86cdSEric Blake goto out; 2161c1bb86cdSEric Blake } 2162c1bb86cdSEric Blake 2163b8cf1913SMax Reitz /* Take permissions: We want to discard everything, so we need 2164b8cf1913SMax Reitz * BLK_PERM_WRITE; and truncation to the desired size requires 2165b8cf1913SMax Reitz * BLK_PERM_RESIZE. 2166b8cf1913SMax Reitz * On the other hand, we cannot share the RESIZE permission 2167b8cf1913SMax Reitz * because we promise that after this function, the file has the 2168b8cf1913SMax Reitz * size given in the options. If someone else were to resize it 2169b8cf1913SMax Reitz * concurrently, we could not guarantee that. 2170b8cf1913SMax Reitz * Note that after this function, we can no longer guarantee that 2171b8cf1913SMax Reitz * the file is not touched by a third party, so it may be resized 2172b8cf1913SMax Reitz * then. */ 2173b8cf1913SMax Reitz perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; 2174b8cf1913SMax Reitz shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; 2175b8cf1913SMax Reitz 2176b8cf1913SMax Reitz /* Step one: Take locks */ 2177d815efcaSMax Reitz result = raw_apply_lock_bytes(fd, perm, ~shared, false, errp); 2178b8cf1913SMax Reitz if (result < 0) { 2179b8cf1913SMax Reitz goto out_close; 2180b8cf1913SMax Reitz } 2181b8cf1913SMax Reitz 2182b8cf1913SMax Reitz /* Step two: Check that nobody else has taken conflicting locks */ 2183b8cf1913SMax Reitz result = raw_check_lock_bytes(fd, perm, shared, errp); 2184b8cf1913SMax Reitz if (result < 0) { 21857c20c808SMax Reitz goto out_unlock; 2186b8cf1913SMax Reitz } 2187b8cf1913SMax Reitz 2188b8cf1913SMax Reitz /* Clear the file by truncating it to 0 */ 218993f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); 2190b8cf1913SMax Reitz if (result < 0) { 21917c20c808SMax Reitz goto out_unlock; 2192b8cf1913SMax Reitz } 2193b8cf1913SMax Reitz 2194927f11e1SKevin Wolf if (file_opts->nocow) { 2195c1bb86cdSEric Blake #ifdef __linux__ 2196c1bb86cdSEric Blake /* Set NOCOW flag to solve performance issue on fs like btrfs. 2197c1bb86cdSEric Blake * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value 2198c1bb86cdSEric Blake * will be ignored since any failure of this operation should not 2199c1bb86cdSEric Blake * block the left work. 2200c1bb86cdSEric Blake */ 2201c1bb86cdSEric Blake int attr; 2202c1bb86cdSEric Blake if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { 2203c1bb86cdSEric Blake attr |= FS_NOCOW_FL; 2204c1bb86cdSEric Blake ioctl(fd, FS_IOC_SETFLAGS, &attr); 2205c1bb86cdSEric Blake } 2206c1bb86cdSEric Blake #endif 2207c1bb86cdSEric Blake } 2208c1bb86cdSEric Blake 2209b8cf1913SMax Reitz /* Resize and potentially preallocate the file to the desired 2210b8cf1913SMax Reitz * final size */ 221193f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, file_opts->size, 221293f4e2ffSKevin Wolf file_opts->preallocation, errp); 22139f63b07eSMax Reitz if (result < 0) { 22147c20c808SMax Reitz goto out_unlock; 22157c20c808SMax Reitz } 22167c20c808SMax Reitz 22177c20c808SMax Reitz out_unlock: 22187c20c808SMax Reitz raw_apply_lock_bytes(fd, 0, 0, true, &local_err); 22197c20c808SMax Reitz if (local_err) { 22207c20c808SMax Reitz /* The above call should not fail, and if it does, that does 22217c20c808SMax Reitz * not mean the whole creation operation has failed. So 22227c20c808SMax Reitz * report it the user for their convenience, but do not report 22237c20c808SMax Reitz * it to the caller. */ 22247c20c808SMax Reitz error_report_err(local_err); 22255a1dad9dSNir Soffer } 22265a1dad9dSNir Soffer 22275a1dad9dSNir Soffer out_close: 2228c1bb86cdSEric Blake if (qemu_close(fd) != 0 && result == 0) { 2229c1bb86cdSEric Blake result = -errno; 2230c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not close the new file"); 2231c1bb86cdSEric Blake } 2232c1bb86cdSEric Blake out: 2233c1bb86cdSEric Blake return result; 2234c1bb86cdSEric Blake } 2235c1bb86cdSEric Blake 2236927f11e1SKevin Wolf static int coroutine_fn raw_co_create_opts(const char *filename, QemuOpts *opts, 2237927f11e1SKevin Wolf Error **errp) 2238927f11e1SKevin Wolf { 2239927f11e1SKevin Wolf BlockdevCreateOptions options; 2240927f11e1SKevin Wolf int64_t total_size = 0; 2241927f11e1SKevin Wolf bool nocow = false; 2242927f11e1SKevin Wolf PreallocMode prealloc; 2243927f11e1SKevin Wolf char *buf = NULL; 2244927f11e1SKevin Wolf Error *local_err = NULL; 2245927f11e1SKevin Wolf 2246927f11e1SKevin Wolf /* Skip file: protocol prefix */ 2247927f11e1SKevin Wolf strstart(filename, "file:", &filename); 2248927f11e1SKevin Wolf 2249927f11e1SKevin Wolf /* Read out options */ 2250927f11e1SKevin Wolf total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2251927f11e1SKevin Wolf BDRV_SECTOR_SIZE); 2252927f11e1SKevin Wolf nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); 2253927f11e1SKevin Wolf buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2254927f11e1SKevin Wolf prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, 2255927f11e1SKevin Wolf PREALLOC_MODE_OFF, &local_err); 2256927f11e1SKevin Wolf g_free(buf); 2257927f11e1SKevin Wolf if (local_err) { 2258927f11e1SKevin Wolf error_propagate(errp, local_err); 2259927f11e1SKevin Wolf return -EINVAL; 2260927f11e1SKevin Wolf } 2261927f11e1SKevin Wolf 2262927f11e1SKevin Wolf options = (BlockdevCreateOptions) { 2263927f11e1SKevin Wolf .driver = BLOCKDEV_DRIVER_FILE, 2264927f11e1SKevin Wolf .u.file = { 2265927f11e1SKevin Wolf .filename = (char *) filename, 2266927f11e1SKevin Wolf .size = total_size, 2267927f11e1SKevin Wolf .has_preallocation = true, 2268927f11e1SKevin Wolf .preallocation = prealloc, 2269927f11e1SKevin Wolf .has_nocow = true, 2270927f11e1SKevin Wolf .nocow = nocow, 2271927f11e1SKevin Wolf }, 2272927f11e1SKevin Wolf }; 2273927f11e1SKevin Wolf return raw_co_create(&options, errp); 2274927f11e1SKevin Wolf } 2275927f11e1SKevin Wolf 2276c1bb86cdSEric Blake /* 2277c1bb86cdSEric Blake * Find allocation range in @bs around offset @start. 2278c1bb86cdSEric Blake * May change underlying file descriptor's file offset. 2279c1bb86cdSEric Blake * If @start is not in a hole, store @start in @data, and the 2280c1bb86cdSEric Blake * beginning of the next hole in @hole, and return 0. 2281c1bb86cdSEric Blake * If @start is in a non-trailing hole, store @start in @hole and the 2282c1bb86cdSEric Blake * beginning of the next non-hole in @data, and return 0. 2283c1bb86cdSEric Blake * If @start is in a trailing hole or beyond EOF, return -ENXIO. 2284c1bb86cdSEric Blake * If we can't find out, return a negative errno other than -ENXIO. 2285c1bb86cdSEric Blake */ 2286c1bb86cdSEric Blake static int find_allocation(BlockDriverState *bs, off_t start, 2287c1bb86cdSEric Blake off_t *data, off_t *hole) 2288c1bb86cdSEric Blake { 2289c1bb86cdSEric Blake #if defined SEEK_HOLE && defined SEEK_DATA 2290c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2291c1bb86cdSEric Blake off_t offs; 2292c1bb86cdSEric Blake 2293c1bb86cdSEric Blake /* 2294c1bb86cdSEric Blake * SEEK_DATA cases: 2295c1bb86cdSEric Blake * D1. offs == start: start is in data 2296c1bb86cdSEric Blake * D2. offs > start: start is in a hole, next data at offs 2297c1bb86cdSEric Blake * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 2298c1bb86cdSEric Blake * or start is beyond EOF 2299c1bb86cdSEric Blake * If the latter happens, the file has been truncated behind 2300c1bb86cdSEric Blake * our back since we opened it. All bets are off then. 2301c1bb86cdSEric Blake * Treating like a trailing hole is simplest. 2302c1bb86cdSEric Blake * D4. offs < 0, errno != ENXIO: we learned nothing 2303c1bb86cdSEric Blake */ 2304c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_DATA); 2305c1bb86cdSEric Blake if (offs < 0) { 2306c1bb86cdSEric Blake return -errno; /* D3 or D4 */ 2307c1bb86cdSEric Blake } 2308a03083a0SJeff Cody 2309a03083a0SJeff Cody if (offs < start) { 2310a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2311a03083a0SJeff Cody * -EIO in this case, and we'll treat it like D4. */ 2312a03083a0SJeff Cody return -EIO; 2313a03083a0SJeff Cody } 2314c1bb86cdSEric Blake 2315c1bb86cdSEric Blake if (offs > start) { 2316c1bb86cdSEric Blake /* D2: in hole, next data at offs */ 2317c1bb86cdSEric Blake *hole = start; 2318c1bb86cdSEric Blake *data = offs; 2319c1bb86cdSEric Blake return 0; 2320c1bb86cdSEric Blake } 2321c1bb86cdSEric Blake 2322c1bb86cdSEric Blake /* D1: in data, end not yet known */ 2323c1bb86cdSEric Blake 2324c1bb86cdSEric Blake /* 2325c1bb86cdSEric Blake * SEEK_HOLE cases: 2326c1bb86cdSEric Blake * H1. offs == start: start is in a hole 2327c1bb86cdSEric Blake * If this happens here, a hole has been dug behind our back 2328c1bb86cdSEric Blake * since the previous lseek(). 2329c1bb86cdSEric Blake * H2. offs > start: either start is in data, next hole at offs, 2330c1bb86cdSEric Blake * or start is in trailing hole, EOF at offs 2331c1bb86cdSEric Blake * Linux treats trailing holes like any other hole: offs == 2332c1bb86cdSEric Blake * start. Solaris seeks to EOF instead: offs > start (blech). 2333c1bb86cdSEric Blake * If that happens here, a hole has been dug behind our back 2334c1bb86cdSEric Blake * since the previous lseek(). 2335c1bb86cdSEric Blake * H3. offs < 0, errno = ENXIO: start is beyond EOF 2336c1bb86cdSEric Blake * If this happens, the file has been truncated behind our 2337c1bb86cdSEric Blake * back since we opened it. Treat it like a trailing hole. 2338c1bb86cdSEric Blake * H4. offs < 0, errno != ENXIO: we learned nothing 2339c1bb86cdSEric Blake * Pretend we know nothing at all, i.e. "forget" about D1. 2340c1bb86cdSEric Blake */ 2341c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_HOLE); 2342c1bb86cdSEric Blake if (offs < 0) { 2343c1bb86cdSEric Blake return -errno; /* D1 and (H3 or H4) */ 2344c1bb86cdSEric Blake } 2345a03083a0SJeff Cody 2346a03083a0SJeff Cody if (offs < start) { 2347a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2348a03083a0SJeff Cody * -EIO in this case, and we'll treat it like H4. */ 2349a03083a0SJeff Cody return -EIO; 2350a03083a0SJeff Cody } 2351c1bb86cdSEric Blake 2352c1bb86cdSEric Blake if (offs > start) { 2353c1bb86cdSEric Blake /* 2354c1bb86cdSEric Blake * D1 and H2: either in data, next hole at offs, or it was in 2355c1bb86cdSEric Blake * data but is now in a trailing hole. In the latter case, 2356c1bb86cdSEric Blake * all bets are off. Treating it as if it there was data all 2357c1bb86cdSEric Blake * the way to EOF is safe, so simply do that. 2358c1bb86cdSEric Blake */ 2359c1bb86cdSEric Blake *data = start; 2360c1bb86cdSEric Blake *hole = offs; 2361c1bb86cdSEric Blake return 0; 2362c1bb86cdSEric Blake } 2363c1bb86cdSEric Blake 2364c1bb86cdSEric Blake /* D1 and H1 */ 2365c1bb86cdSEric Blake return -EBUSY; 2366c1bb86cdSEric Blake #else 2367c1bb86cdSEric Blake return -ENOTSUP; 2368c1bb86cdSEric Blake #endif 2369c1bb86cdSEric Blake } 2370c1bb86cdSEric Blake 2371c1bb86cdSEric Blake /* 2372a290f085SEric Blake * Returns the allocation status of the specified offset. 2373c1bb86cdSEric Blake * 2374a290f085SEric Blake * The block layer guarantees 'offset' and 'bytes' are within bounds. 2375c1bb86cdSEric Blake * 2376a290f085SEric Blake * 'pnum' is set to the number of bytes (including and immediately following 2377a290f085SEric Blake * the specified offset) that are known to be in the same 2378c1bb86cdSEric Blake * allocated/unallocated state. 2379c1bb86cdSEric Blake * 2380a290f085SEric Blake * 'bytes' is the max value 'pnum' should be set to. 2381c1bb86cdSEric Blake */ 2382a290f085SEric Blake static int coroutine_fn raw_co_block_status(BlockDriverState *bs, 2383a290f085SEric Blake bool want_zero, 2384a290f085SEric Blake int64_t offset, 2385a290f085SEric Blake int64_t bytes, int64_t *pnum, 2386a290f085SEric Blake int64_t *map, 2387c1bb86cdSEric Blake BlockDriverState **file) 2388c1bb86cdSEric Blake { 2389a290f085SEric Blake off_t data = 0, hole = 0; 2390c1bb86cdSEric Blake int ret; 2391c1bb86cdSEric Blake 2392c1bb86cdSEric Blake ret = fd_open(bs); 2393c1bb86cdSEric Blake if (ret < 0) { 2394c1bb86cdSEric Blake return ret; 2395c1bb86cdSEric Blake } 2396c1bb86cdSEric Blake 2397a290f085SEric Blake if (!want_zero) { 2398a290f085SEric Blake *pnum = bytes; 2399a290f085SEric Blake *map = offset; 2400a290f085SEric Blake *file = bs; 2401a290f085SEric Blake return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 2402c1bb86cdSEric Blake } 2403c1bb86cdSEric Blake 2404a290f085SEric Blake ret = find_allocation(bs, offset, &data, &hole); 2405c1bb86cdSEric Blake if (ret == -ENXIO) { 2406c1bb86cdSEric Blake /* Trailing hole */ 2407a290f085SEric Blake *pnum = bytes; 2408c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2409c1bb86cdSEric Blake } else if (ret < 0) { 2410c1bb86cdSEric Blake /* No info available, so pretend there are no holes */ 2411a290f085SEric Blake *pnum = bytes; 2412c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2413a290f085SEric Blake } else if (data == offset) { 2414a290f085SEric Blake /* On a data extent, compute bytes to the end of the extent, 2415c1bb86cdSEric Blake * possibly including a partial sector at EOF. */ 2416a290f085SEric Blake *pnum = MIN(bytes, hole - offset); 2417c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2418c1bb86cdSEric Blake } else { 2419a290f085SEric Blake /* On a hole, compute bytes to the beginning of the next extent. */ 2420a290f085SEric Blake assert(hole == offset); 2421a290f085SEric Blake *pnum = MIN(bytes, data - offset); 2422c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2423c1bb86cdSEric Blake } 2424a290f085SEric Blake *map = offset; 2425c1bb86cdSEric Blake *file = bs; 2426a290f085SEric Blake return ret | BDRV_BLOCK_OFFSET_VALID; 2427c1bb86cdSEric Blake } 2428c1bb86cdSEric Blake 242931be8a2aSStefan Hajnoczi #if defined(__linux__) 243031be8a2aSStefan Hajnoczi /* Verify that the file is not in the page cache */ 243131be8a2aSStefan Hajnoczi static void check_cache_dropped(BlockDriverState *bs, Error **errp) 243231be8a2aSStefan Hajnoczi { 243331be8a2aSStefan Hajnoczi const size_t window_size = 128 * 1024 * 1024; 243431be8a2aSStefan Hajnoczi BDRVRawState *s = bs->opaque; 243531be8a2aSStefan Hajnoczi void *window = NULL; 243631be8a2aSStefan Hajnoczi size_t length = 0; 243731be8a2aSStefan Hajnoczi unsigned char *vec; 243831be8a2aSStefan Hajnoczi size_t page_size; 243931be8a2aSStefan Hajnoczi off_t offset; 244031be8a2aSStefan Hajnoczi off_t end; 244131be8a2aSStefan Hajnoczi 244231be8a2aSStefan Hajnoczi /* mincore(2) page status information requires 1 byte per page */ 244331be8a2aSStefan Hajnoczi page_size = sysconf(_SC_PAGESIZE); 244431be8a2aSStefan Hajnoczi vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); 244531be8a2aSStefan Hajnoczi 244631be8a2aSStefan Hajnoczi end = raw_getlength(bs); 244731be8a2aSStefan Hajnoczi 244831be8a2aSStefan Hajnoczi for (offset = 0; offset < end; offset += window_size) { 244931be8a2aSStefan Hajnoczi void *new_window; 245031be8a2aSStefan Hajnoczi size_t new_length; 245131be8a2aSStefan Hajnoczi size_t vec_end; 245231be8a2aSStefan Hajnoczi size_t i; 245331be8a2aSStefan Hajnoczi int ret; 245431be8a2aSStefan Hajnoczi 245531be8a2aSStefan Hajnoczi /* Unmap previous window if size has changed */ 245631be8a2aSStefan Hajnoczi new_length = MIN(end - offset, window_size); 245731be8a2aSStefan Hajnoczi if (new_length != length) { 245831be8a2aSStefan Hajnoczi munmap(window, length); 245931be8a2aSStefan Hajnoczi window = NULL; 246031be8a2aSStefan Hajnoczi length = 0; 246131be8a2aSStefan Hajnoczi } 246231be8a2aSStefan Hajnoczi 246331be8a2aSStefan Hajnoczi new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, 246431be8a2aSStefan Hajnoczi s->fd, offset); 246531be8a2aSStefan Hajnoczi if (new_window == MAP_FAILED) { 246631be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mmap failed"); 246731be8a2aSStefan Hajnoczi break; 246831be8a2aSStefan Hajnoczi } 246931be8a2aSStefan Hajnoczi 247031be8a2aSStefan Hajnoczi window = new_window; 247131be8a2aSStefan Hajnoczi length = new_length; 247231be8a2aSStefan Hajnoczi 247331be8a2aSStefan Hajnoczi ret = mincore(window, length, vec); 247431be8a2aSStefan Hajnoczi if (ret < 0) { 247531be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mincore failed"); 247631be8a2aSStefan Hajnoczi break; 247731be8a2aSStefan Hajnoczi } 247831be8a2aSStefan Hajnoczi 247931be8a2aSStefan Hajnoczi vec_end = DIV_ROUND_UP(length, page_size); 248031be8a2aSStefan Hajnoczi for (i = 0; i < vec_end; i++) { 248131be8a2aSStefan Hajnoczi if (vec[i] & 0x1) { 248231be8a2aSStefan Hajnoczi error_setg(errp, "page cache still in use!"); 248331be8a2aSStefan Hajnoczi break; 248431be8a2aSStefan Hajnoczi } 248531be8a2aSStefan Hajnoczi } 248631be8a2aSStefan Hajnoczi } 248731be8a2aSStefan Hajnoczi 248831be8a2aSStefan Hajnoczi if (window) { 248931be8a2aSStefan Hajnoczi munmap(window, length); 249031be8a2aSStefan Hajnoczi } 249131be8a2aSStefan Hajnoczi 249231be8a2aSStefan Hajnoczi g_free(vec); 249331be8a2aSStefan Hajnoczi } 249431be8a2aSStefan Hajnoczi #endif /* __linux__ */ 249531be8a2aSStefan Hajnoczi 2496dd577a26SStefan Hajnoczi static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, 2497dd577a26SStefan Hajnoczi Error **errp) 2498dd577a26SStefan Hajnoczi { 2499dd577a26SStefan Hajnoczi BDRVRawState *s = bs->opaque; 2500dd577a26SStefan Hajnoczi int ret; 2501dd577a26SStefan Hajnoczi 2502dd577a26SStefan Hajnoczi ret = fd_open(bs); 2503dd577a26SStefan Hajnoczi if (ret < 0) { 2504dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "The file descriptor is not open"); 2505dd577a26SStefan Hajnoczi return; 2506dd577a26SStefan Hajnoczi } 2507dd577a26SStefan Hajnoczi 2508dd577a26SStefan Hajnoczi if (s->open_flags & O_DIRECT) { 2509dd577a26SStefan Hajnoczi return; /* No host kernel page cache */ 2510dd577a26SStefan Hajnoczi } 2511dd577a26SStefan Hajnoczi 2512dd577a26SStefan Hajnoczi #if defined(__linux__) 2513dd577a26SStefan Hajnoczi /* This sets the scene for the next syscall... */ 2514dd577a26SStefan Hajnoczi ret = bdrv_co_flush(bs); 2515dd577a26SStefan Hajnoczi if (ret < 0) { 2516dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "flush failed"); 2517dd577a26SStefan Hajnoczi return; 2518dd577a26SStefan Hajnoczi } 2519dd577a26SStefan Hajnoczi 2520dd577a26SStefan Hajnoczi /* Linux does not invalidate pages that are dirty, locked, or mmapped by a 2521dd577a26SStefan Hajnoczi * process. These limitations are okay because we just fsynced the file, 2522dd577a26SStefan Hajnoczi * we don't use mmap, and the file should not be in use by other processes. 2523dd577a26SStefan Hajnoczi */ 2524dd577a26SStefan Hajnoczi ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); 2525dd577a26SStefan Hajnoczi if (ret != 0) { /* the return value is a positive errno */ 2526dd577a26SStefan Hajnoczi error_setg_errno(errp, ret, "fadvise failed"); 2527dd577a26SStefan Hajnoczi return; 2528dd577a26SStefan Hajnoczi } 252931be8a2aSStefan Hajnoczi 253031be8a2aSStefan Hajnoczi if (s->check_cache_dropped) { 253131be8a2aSStefan Hajnoczi check_cache_dropped(bs, errp); 253231be8a2aSStefan Hajnoczi } 2533dd577a26SStefan Hajnoczi #else /* __linux__ */ 2534dd577a26SStefan Hajnoczi /* Do nothing. Live migration to a remote host with cache.direct=off is 2535dd577a26SStefan Hajnoczi * unsupported on other host operating systems. Cache consistency issues 2536dd577a26SStefan Hajnoczi * may occur but no error is reported here, partly because that's the 2537dd577a26SStefan Hajnoczi * historical behavior and partly because it's hard to differentiate valid 2538dd577a26SStefan Hajnoczi * configurations that should not cause errors. 2539dd577a26SStefan Hajnoczi */ 2540dd577a26SStefan Hajnoczi #endif /* !__linux__ */ 2541dd577a26SStefan Hajnoczi } 2542dd577a26SStefan Hajnoczi 254333d70fb6SKevin Wolf static coroutine_fn int 254433d70fb6SKevin Wolf raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 2545c1bb86cdSEric Blake { 2546c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2547c1bb86cdSEric Blake 254833d70fb6SKevin Wolf return paio_submit_co(bs, s->fd, offset, NULL, bytes, QEMU_AIO_DISCARD); 2549c1bb86cdSEric Blake } 2550c1bb86cdSEric Blake 2551c1bb86cdSEric Blake static int coroutine_fn raw_co_pwrite_zeroes( 2552c1bb86cdSEric Blake BlockDriverState *bs, int64_t offset, 2553f5a5ca79SManos Pitsidianakis int bytes, BdrvRequestFlags flags) 2554c1bb86cdSEric Blake { 2555c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2556c1bb86cdSEric Blake 2557c1bb86cdSEric Blake if (!(flags & BDRV_REQ_MAY_UNMAP)) { 2558f5a5ca79SManos Pitsidianakis return paio_submit_co(bs, s->fd, offset, NULL, bytes, 2559c1bb86cdSEric Blake QEMU_AIO_WRITE_ZEROES); 2560c1bb86cdSEric Blake } else if (s->discard_zeroes) { 2561f5a5ca79SManos Pitsidianakis return paio_submit_co(bs, s->fd, offset, NULL, bytes, 2562c1bb86cdSEric Blake QEMU_AIO_DISCARD); 2563c1bb86cdSEric Blake } 2564c1bb86cdSEric Blake return -ENOTSUP; 2565c1bb86cdSEric Blake } 2566c1bb86cdSEric Blake 2567c1bb86cdSEric Blake static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2568c1bb86cdSEric Blake { 2569c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2570c1bb86cdSEric Blake 2571c1bb86cdSEric Blake bdi->unallocated_blocks_are_zero = s->discard_zeroes; 2572c1bb86cdSEric Blake return 0; 2573c1bb86cdSEric Blake } 2574c1bb86cdSEric Blake 2575c1bb86cdSEric Blake static QemuOptsList raw_create_opts = { 2576c1bb86cdSEric Blake .name = "raw-create-opts", 2577c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), 2578c1bb86cdSEric Blake .desc = { 2579c1bb86cdSEric Blake { 2580c1bb86cdSEric Blake .name = BLOCK_OPT_SIZE, 2581c1bb86cdSEric Blake .type = QEMU_OPT_SIZE, 2582c1bb86cdSEric Blake .help = "Virtual disk size" 2583c1bb86cdSEric Blake }, 2584c1bb86cdSEric Blake { 2585c1bb86cdSEric Blake .name = BLOCK_OPT_NOCOW, 2586c1bb86cdSEric Blake .type = QEMU_OPT_BOOL, 2587c1bb86cdSEric Blake .help = "Turn off copy-on-write (valid only on btrfs)" 2588c1bb86cdSEric Blake }, 2589c1bb86cdSEric Blake { 2590c1bb86cdSEric Blake .name = BLOCK_OPT_PREALLOC, 2591c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 2592c1bb86cdSEric Blake .help = "Preallocation mode (allowed values: off, falloc, full)" 2593c1bb86cdSEric Blake }, 2594c1bb86cdSEric Blake { /* end of list */ } 2595c1bb86cdSEric Blake } 2596c1bb86cdSEric Blake }; 2597c1bb86cdSEric Blake 2598244a5668SFam Zheng static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, 2599244a5668SFam Zheng Error **errp) 2600244a5668SFam Zheng { 2601244a5668SFam Zheng return raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); 2602244a5668SFam Zheng } 2603244a5668SFam Zheng 2604244a5668SFam Zheng static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) 2605244a5668SFam Zheng { 2606244a5668SFam Zheng BDRVRawState *s = bs->opaque; 2607244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); 2608244a5668SFam Zheng s->perm = perm; 2609244a5668SFam Zheng s->shared_perm = shared; 2610244a5668SFam Zheng } 2611244a5668SFam Zheng 2612244a5668SFam Zheng static void raw_abort_perm_update(BlockDriverState *bs) 2613244a5668SFam Zheng { 2614244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 2615244a5668SFam Zheng } 2616244a5668SFam Zheng 261767b51fb9SVladimir Sementsov-Ogievskiy static int coroutine_fn raw_co_copy_range_from( 261867b51fb9SVladimir Sementsov-Ogievskiy BlockDriverState *bs, BdrvChild *src, uint64_t src_offset, 261967b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, uint64_t dst_offset, uint64_t bytes, 262067b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) 26211efad060SFam Zheng { 262267b51fb9SVladimir Sementsov-Ogievskiy return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, 262367b51fb9SVladimir Sementsov-Ogievskiy read_flags, write_flags); 26241efad060SFam Zheng } 26251efad060SFam Zheng 26261efad060SFam Zheng static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, 262767b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *src, 262867b51fb9SVladimir Sementsov-Ogievskiy uint64_t src_offset, 262967b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, 263067b51fb9SVladimir Sementsov-Ogievskiy uint64_t dst_offset, 263167b51fb9SVladimir Sementsov-Ogievskiy uint64_t bytes, 263267b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, 263367b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags write_flags) 26341efad060SFam Zheng { 26351efad060SFam Zheng BDRVRawState *s = bs->opaque; 26361efad060SFam Zheng BDRVRawState *src_s; 26371efad060SFam Zheng 26381efad060SFam Zheng assert(dst->bs == bs); 26391efad060SFam Zheng if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { 26401efad060SFam Zheng return -ENOTSUP; 26411efad060SFam Zheng } 26421efad060SFam Zheng 26431efad060SFam Zheng src_s = src->bs->opaque; 26449f850f67SFam Zheng if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { 26451efad060SFam Zheng return -EIO; 26461efad060SFam Zheng } 26471efad060SFam Zheng return paio_submit_co_full(bs, src_s->fd, src_offset, s->fd, dst_offset, 26481efad060SFam Zheng NULL, bytes, QEMU_AIO_COPY_RANGE); 26491efad060SFam Zheng } 26501efad060SFam Zheng 2651c1bb86cdSEric Blake BlockDriver bdrv_file = { 2652c1bb86cdSEric Blake .format_name = "file", 2653c1bb86cdSEric Blake .protocol_name = "file", 2654c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 2655c1bb86cdSEric Blake .bdrv_needs_filename = true, 2656c1bb86cdSEric Blake .bdrv_probe = NULL, /* no probe for protocols */ 2657c1bb86cdSEric Blake .bdrv_parse_filename = raw_parse_filename, 2658c1bb86cdSEric Blake .bdrv_file_open = raw_open, 2659c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 2660c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 2661c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 2662c1bb86cdSEric Blake .bdrv_close = raw_close, 2663927f11e1SKevin Wolf .bdrv_co_create = raw_co_create, 2664efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = raw_co_create_opts, 2665c1bb86cdSEric Blake .bdrv_has_zero_init = bdrv_has_zero_init_1, 2666a290f085SEric Blake .bdrv_co_block_status = raw_co_block_status, 2667dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 2668c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, 2669c1bb86cdSEric Blake 2670c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 2671c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 267233d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 267333d70fb6SKevin Wolf .bdrv_co_pdiscard = raw_co_pdiscard, 26741efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 26751efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 2676c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 2677c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 2678c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 2679ed6e2161SNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 2680c1bb86cdSEric Blake 2681061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 2682c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 2683c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 2684c1bb86cdSEric Blake .bdrv_get_allocated_file_size 2685c1bb86cdSEric Blake = raw_get_allocated_file_size, 2686244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 2687244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 2688244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 2689c1bb86cdSEric Blake .create_opts = &raw_create_opts, 2690c1bb86cdSEric Blake }; 2691c1bb86cdSEric Blake 2692c1bb86cdSEric Blake /***********************************************/ 2693c1bb86cdSEric Blake /* host device */ 2694c1bb86cdSEric Blake 2695c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2696c1bb86cdSEric Blake static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 2697c1bb86cdSEric Blake CFIndex maxPathSize, int flags); 2698c1bb86cdSEric Blake static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) 2699c1bb86cdSEric Blake { 2700c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 2701c1bb86cdSEric Blake mach_port_t masterPort; 2702c1bb86cdSEric Blake CFMutableDictionaryRef classesToMatch; 2703c1bb86cdSEric Blake const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; 2704c1bb86cdSEric Blake char *mediaType = NULL; 2705c1bb86cdSEric Blake 2706c1bb86cdSEric Blake kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); 2707c1bb86cdSEric Blake if ( KERN_SUCCESS != kernResult ) { 2708c1bb86cdSEric Blake printf( "IOMasterPort returned %d\n", kernResult ); 2709c1bb86cdSEric Blake } 2710c1bb86cdSEric Blake 2711c1bb86cdSEric Blake int index; 2712c1bb86cdSEric Blake for (index = 0; index < ARRAY_SIZE(matching_array); index++) { 2713c1bb86cdSEric Blake classesToMatch = IOServiceMatching(matching_array[index]); 2714c1bb86cdSEric Blake if (classesToMatch == NULL) { 2715c1bb86cdSEric Blake error_report("IOServiceMatching returned NULL for %s", 2716c1bb86cdSEric Blake matching_array[index]); 2717c1bb86cdSEric Blake continue; 2718c1bb86cdSEric Blake } 2719c1bb86cdSEric Blake CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), 2720c1bb86cdSEric Blake kCFBooleanTrue); 2721c1bb86cdSEric Blake kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, 2722c1bb86cdSEric Blake mediaIterator); 2723c1bb86cdSEric Blake if (kernResult != KERN_SUCCESS) { 2724c1bb86cdSEric Blake error_report("Note: IOServiceGetMatchingServices returned %d", 2725c1bb86cdSEric Blake kernResult); 2726c1bb86cdSEric Blake continue; 2727c1bb86cdSEric Blake } 2728c1bb86cdSEric Blake 2729c1bb86cdSEric Blake /* If a match was found, leave the loop */ 2730c1bb86cdSEric Blake if (*mediaIterator != 0) { 2731c1bb86cdSEric Blake DPRINTF("Matching using %s\n", matching_array[index]); 2732c1bb86cdSEric Blake mediaType = g_strdup(matching_array[index]); 2733c1bb86cdSEric Blake break; 2734c1bb86cdSEric Blake } 2735c1bb86cdSEric Blake } 2736c1bb86cdSEric Blake return mediaType; 2737c1bb86cdSEric Blake } 2738c1bb86cdSEric Blake 2739c1bb86cdSEric Blake kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 2740c1bb86cdSEric Blake CFIndex maxPathSize, int flags) 2741c1bb86cdSEric Blake { 2742c1bb86cdSEric Blake io_object_t nextMedia; 2743c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 2744c1bb86cdSEric Blake *bsdPath = '\0'; 2745c1bb86cdSEric Blake nextMedia = IOIteratorNext( mediaIterator ); 2746c1bb86cdSEric Blake if ( nextMedia ) 2747c1bb86cdSEric Blake { 2748c1bb86cdSEric Blake CFTypeRef bsdPathAsCFString; 2749c1bb86cdSEric Blake bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); 2750c1bb86cdSEric Blake if ( bsdPathAsCFString ) { 2751c1bb86cdSEric Blake size_t devPathLength; 2752c1bb86cdSEric Blake strcpy( bsdPath, _PATH_DEV ); 2753c1bb86cdSEric Blake if (flags & BDRV_O_NOCACHE) { 2754c1bb86cdSEric Blake strcat(bsdPath, "r"); 2755c1bb86cdSEric Blake } 2756c1bb86cdSEric Blake devPathLength = strlen( bsdPath ); 2757c1bb86cdSEric Blake if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { 2758c1bb86cdSEric Blake kernResult = KERN_SUCCESS; 2759c1bb86cdSEric Blake } 2760c1bb86cdSEric Blake CFRelease( bsdPathAsCFString ); 2761c1bb86cdSEric Blake } 2762c1bb86cdSEric Blake IOObjectRelease( nextMedia ); 2763c1bb86cdSEric Blake } 2764c1bb86cdSEric Blake 2765c1bb86cdSEric Blake return kernResult; 2766c1bb86cdSEric Blake } 2767c1bb86cdSEric Blake 2768c1bb86cdSEric Blake /* Sets up a real cdrom for use in QEMU */ 2769c1bb86cdSEric Blake static bool setup_cdrom(char *bsd_path, Error **errp) 2770c1bb86cdSEric Blake { 2771c1bb86cdSEric Blake int index, num_of_test_partitions = 2, fd; 2772c1bb86cdSEric Blake char test_partition[MAXPATHLEN]; 2773c1bb86cdSEric Blake bool partition_found = false; 2774c1bb86cdSEric Blake 2775c1bb86cdSEric Blake /* look for a working partition */ 2776c1bb86cdSEric Blake for (index = 0; index < num_of_test_partitions; index++) { 2777c1bb86cdSEric Blake snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, 2778c1bb86cdSEric Blake index); 2779c1bb86cdSEric Blake fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE); 2780c1bb86cdSEric Blake if (fd >= 0) { 2781c1bb86cdSEric Blake partition_found = true; 2782c1bb86cdSEric Blake qemu_close(fd); 2783c1bb86cdSEric Blake break; 2784c1bb86cdSEric Blake } 2785c1bb86cdSEric Blake } 2786c1bb86cdSEric Blake 2787c1bb86cdSEric Blake /* if a working partition on the device was not found */ 2788c1bb86cdSEric Blake if (partition_found == false) { 2789c1bb86cdSEric Blake error_setg(errp, "Failed to find a working partition on disc"); 2790c1bb86cdSEric Blake } else { 2791c1bb86cdSEric Blake DPRINTF("Using %s as optical disc\n", test_partition); 2792c1bb86cdSEric Blake pstrcpy(bsd_path, MAXPATHLEN, test_partition); 2793c1bb86cdSEric Blake } 2794c1bb86cdSEric Blake return partition_found; 2795c1bb86cdSEric Blake } 2796c1bb86cdSEric Blake 2797c1bb86cdSEric Blake /* Prints directions on mounting and unmounting a device */ 2798c1bb86cdSEric Blake static void print_unmounting_directions(const char *file_name) 2799c1bb86cdSEric Blake { 2800c1bb86cdSEric Blake error_report("If device %s is mounted on the desktop, unmount" 2801c1bb86cdSEric Blake " it first before using it in QEMU", file_name); 2802c1bb86cdSEric Blake error_report("Command to unmount device: diskutil unmountDisk %s", 2803c1bb86cdSEric Blake file_name); 2804c1bb86cdSEric Blake error_report("Command to mount device: diskutil mountDisk %s", file_name); 2805c1bb86cdSEric Blake } 2806c1bb86cdSEric Blake 2807c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 2808c1bb86cdSEric Blake 2809c1bb86cdSEric Blake static int hdev_probe_device(const char *filename) 2810c1bb86cdSEric Blake { 2811c1bb86cdSEric Blake struct stat st; 2812c1bb86cdSEric Blake 2813c1bb86cdSEric Blake /* allow a dedicated CD-ROM driver to match with a higher priority */ 2814c1bb86cdSEric Blake if (strstart(filename, "/dev/cdrom", NULL)) 2815c1bb86cdSEric Blake return 50; 2816c1bb86cdSEric Blake 2817c1bb86cdSEric Blake if (stat(filename, &st) >= 0 && 2818c1bb86cdSEric Blake (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 2819c1bb86cdSEric Blake return 100; 2820c1bb86cdSEric Blake } 2821c1bb86cdSEric Blake 2822c1bb86cdSEric Blake return 0; 2823c1bb86cdSEric Blake } 2824c1bb86cdSEric Blake 2825c1bb86cdSEric Blake static int check_hdev_writable(BDRVRawState *s) 2826c1bb86cdSEric Blake { 2827c1bb86cdSEric Blake #if defined(BLKROGET) 2828c1bb86cdSEric Blake /* Linux block devices can be configured "read-only" using blockdev(8). 2829c1bb86cdSEric Blake * This is independent of device node permissions and therefore open(2) 2830c1bb86cdSEric Blake * with O_RDWR succeeds. Actual writes fail with EPERM. 2831c1bb86cdSEric Blake * 2832c1bb86cdSEric Blake * bdrv_open() is supposed to fail if the disk is read-only. Explicitly 2833c1bb86cdSEric Blake * check for read-only block devices so that Linux block devices behave 2834c1bb86cdSEric Blake * properly. 2835c1bb86cdSEric Blake */ 2836c1bb86cdSEric Blake struct stat st; 2837c1bb86cdSEric Blake int readonly = 0; 2838c1bb86cdSEric Blake 2839c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 2840c1bb86cdSEric Blake return -errno; 2841c1bb86cdSEric Blake } 2842c1bb86cdSEric Blake 2843c1bb86cdSEric Blake if (!S_ISBLK(st.st_mode)) { 2844c1bb86cdSEric Blake return 0; 2845c1bb86cdSEric Blake } 2846c1bb86cdSEric Blake 2847c1bb86cdSEric Blake if (ioctl(s->fd, BLKROGET, &readonly) < 0) { 2848c1bb86cdSEric Blake return -errno; 2849c1bb86cdSEric Blake } 2850c1bb86cdSEric Blake 2851c1bb86cdSEric Blake if (readonly) { 2852c1bb86cdSEric Blake return -EACCES; 2853c1bb86cdSEric Blake } 2854c1bb86cdSEric Blake #endif /* defined(BLKROGET) */ 2855c1bb86cdSEric Blake return 0; 2856c1bb86cdSEric Blake } 2857c1bb86cdSEric Blake 2858c1bb86cdSEric Blake static void hdev_parse_filename(const char *filename, QDict *options, 2859c1bb86cdSEric Blake Error **errp) 2860c1bb86cdSEric Blake { 286103c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_device:", options); 2862c1bb86cdSEric Blake } 2863c1bb86cdSEric Blake 2864c1bb86cdSEric Blake static bool hdev_is_sg(BlockDriverState *bs) 2865c1bb86cdSEric Blake { 2866c1bb86cdSEric Blake 2867c1bb86cdSEric Blake #if defined(__linux__) 2868c1bb86cdSEric Blake 2869c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2870c1bb86cdSEric Blake struct stat st; 2871c1bb86cdSEric Blake struct sg_scsi_id scsiid; 2872c1bb86cdSEric Blake int sg_version; 2873c1bb86cdSEric Blake int ret; 2874c1bb86cdSEric Blake 2875c1bb86cdSEric Blake if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { 2876c1bb86cdSEric Blake return false; 2877c1bb86cdSEric Blake } 2878c1bb86cdSEric Blake 2879c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); 2880c1bb86cdSEric Blake if (ret < 0) { 2881c1bb86cdSEric Blake return false; 2882c1bb86cdSEric Blake } 2883c1bb86cdSEric Blake 2884c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); 2885c1bb86cdSEric Blake if (ret >= 0) { 2886c1bb86cdSEric Blake DPRINTF("SG device found: type=%d, version=%d\n", 2887c1bb86cdSEric Blake scsiid.scsi_type, sg_version); 2888c1bb86cdSEric Blake return true; 2889c1bb86cdSEric Blake } 2890c1bb86cdSEric Blake 2891c1bb86cdSEric Blake #endif 2892c1bb86cdSEric Blake 2893c1bb86cdSEric Blake return false; 2894c1bb86cdSEric Blake } 2895c1bb86cdSEric Blake 2896c1bb86cdSEric Blake static int hdev_open(BlockDriverState *bs, QDict *options, int flags, 2897c1bb86cdSEric Blake Error **errp) 2898c1bb86cdSEric Blake { 2899c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2900c1bb86cdSEric Blake Error *local_err = NULL; 2901c1bb86cdSEric Blake int ret; 2902c1bb86cdSEric Blake 2903c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2904129c7d1cSMarkus Armbruster /* 2905129c7d1cSMarkus Armbruster * Caution: while qdict_get_str() is fine, getting non-string types 2906129c7d1cSMarkus Armbruster * would require more care. When @options come from -blockdev or 2907129c7d1cSMarkus Armbruster * blockdev_add, its members are typed according to the QAPI 2908129c7d1cSMarkus Armbruster * schema, but when they come from -drive, they're all QString. 2909129c7d1cSMarkus Armbruster */ 2910c1bb86cdSEric Blake const char *filename = qdict_get_str(options, "filename"); 2911c1bb86cdSEric Blake char bsd_path[MAXPATHLEN] = ""; 2912c1bb86cdSEric Blake bool error_occurred = false; 2913c1bb86cdSEric Blake 2914c1bb86cdSEric Blake /* If using a real cdrom */ 2915c1bb86cdSEric Blake if (strcmp(filename, "/dev/cdrom") == 0) { 2916c1bb86cdSEric Blake char *mediaType = NULL; 2917c1bb86cdSEric Blake kern_return_t ret_val; 2918c1bb86cdSEric Blake io_iterator_t mediaIterator = 0; 2919c1bb86cdSEric Blake 2920c1bb86cdSEric Blake mediaType = FindEjectableOpticalMedia(&mediaIterator); 2921c1bb86cdSEric Blake if (mediaType == NULL) { 2922c1bb86cdSEric Blake error_setg(errp, "Please make sure your CD/DVD is in the optical" 2923c1bb86cdSEric Blake " drive"); 2924c1bb86cdSEric Blake error_occurred = true; 2925c1bb86cdSEric Blake goto hdev_open_Mac_error; 2926c1bb86cdSEric Blake } 2927c1bb86cdSEric Blake 2928c1bb86cdSEric Blake ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); 2929c1bb86cdSEric Blake if (ret_val != KERN_SUCCESS) { 2930c1bb86cdSEric Blake error_setg(errp, "Could not get BSD path for optical drive"); 2931c1bb86cdSEric Blake error_occurred = true; 2932c1bb86cdSEric Blake goto hdev_open_Mac_error; 2933c1bb86cdSEric Blake } 2934c1bb86cdSEric Blake 2935c1bb86cdSEric Blake /* If a real optical drive was not found */ 2936c1bb86cdSEric Blake if (bsd_path[0] == '\0') { 2937c1bb86cdSEric Blake error_setg(errp, "Failed to obtain bsd path for optical drive"); 2938c1bb86cdSEric Blake error_occurred = true; 2939c1bb86cdSEric Blake goto hdev_open_Mac_error; 2940c1bb86cdSEric Blake } 2941c1bb86cdSEric Blake 2942c1bb86cdSEric Blake /* If using a cdrom disc and finding a partition on the disc failed */ 2943c1bb86cdSEric Blake if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && 2944c1bb86cdSEric Blake setup_cdrom(bsd_path, errp) == false) { 2945c1bb86cdSEric Blake print_unmounting_directions(bsd_path); 2946c1bb86cdSEric Blake error_occurred = true; 2947c1bb86cdSEric Blake goto hdev_open_Mac_error; 2948c1bb86cdSEric Blake } 2949c1bb86cdSEric Blake 295046f5ac20SEric Blake qdict_put_str(options, "filename", bsd_path); 2951c1bb86cdSEric Blake 2952c1bb86cdSEric Blake hdev_open_Mac_error: 2953c1bb86cdSEric Blake g_free(mediaType); 2954c1bb86cdSEric Blake if (mediaIterator) { 2955c1bb86cdSEric Blake IOObjectRelease(mediaIterator); 2956c1bb86cdSEric Blake } 2957c1bb86cdSEric Blake if (error_occurred) { 2958c1bb86cdSEric Blake return -ENOENT; 2959c1bb86cdSEric Blake } 2960c1bb86cdSEric Blake } 2961c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 2962c1bb86cdSEric Blake 2963c1bb86cdSEric Blake s->type = FTYPE_FILE; 2964c1bb86cdSEric Blake 2965230ff739SJohn Snow ret = raw_open_common(bs, options, flags, 0, true, &local_err); 2966c1bb86cdSEric Blake if (ret < 0) { 2967c1bb86cdSEric Blake error_propagate(errp, local_err); 2968c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2969c1bb86cdSEric Blake if (*bsd_path) { 2970c1bb86cdSEric Blake filename = bsd_path; 2971c1bb86cdSEric Blake } 2972c1bb86cdSEric Blake /* if a physical device experienced an error while being opened */ 2973c1bb86cdSEric Blake if (strncmp(filename, "/dev/", 5) == 0) { 2974c1bb86cdSEric Blake print_unmounting_directions(filename); 2975c1bb86cdSEric Blake } 2976c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 2977c1bb86cdSEric Blake return ret; 2978c1bb86cdSEric Blake } 2979c1bb86cdSEric Blake 2980c1bb86cdSEric Blake /* Since this does ioctl the device must be already opened */ 2981c1bb86cdSEric Blake bs->sg = hdev_is_sg(bs); 2982c1bb86cdSEric Blake 2983c1bb86cdSEric Blake if (flags & BDRV_O_RDWR) { 2984c1bb86cdSEric Blake ret = check_hdev_writable(s); 2985c1bb86cdSEric Blake if (ret < 0) { 2986c1bb86cdSEric Blake raw_close(bs); 2987c1bb86cdSEric Blake error_setg_errno(errp, -ret, "The device is not writable"); 2988c1bb86cdSEric Blake return ret; 2989c1bb86cdSEric Blake } 2990c1bb86cdSEric Blake } 2991c1bb86cdSEric Blake 2992c1bb86cdSEric Blake return ret; 2993c1bb86cdSEric Blake } 2994c1bb86cdSEric Blake 2995c1bb86cdSEric Blake #if defined(__linux__) 2996c1bb86cdSEric Blake 2997c1bb86cdSEric Blake static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs, 2998c1bb86cdSEric Blake unsigned long int req, void *buf, 2999c1bb86cdSEric Blake BlockCompletionFunc *cb, void *opaque) 3000c1bb86cdSEric Blake { 3001c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3002c1bb86cdSEric Blake RawPosixAIOData *acb; 3003c1bb86cdSEric Blake ThreadPool *pool; 3004c1bb86cdSEric Blake 3005c1bb86cdSEric Blake if (fd_open(bs) < 0) 3006c1bb86cdSEric Blake return NULL; 3007c1bb86cdSEric Blake 30087c9e5276SPaolo Bonzini if (req == SG_IO && s->pr_mgr) { 30097c9e5276SPaolo Bonzini struct sg_io_hdr *io_hdr = buf; 30107c9e5276SPaolo Bonzini if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || 30117c9e5276SPaolo Bonzini io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { 30127c9e5276SPaolo Bonzini return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), 30137c9e5276SPaolo Bonzini s->fd, io_hdr, cb, opaque); 30147c9e5276SPaolo Bonzini } 30157c9e5276SPaolo Bonzini } 30167c9e5276SPaolo Bonzini 3017c1bb86cdSEric Blake acb = g_new(RawPosixAIOData, 1); 3018c1bb86cdSEric Blake acb->bs = bs; 3019c1bb86cdSEric Blake acb->aio_type = QEMU_AIO_IOCTL; 3020c1bb86cdSEric Blake acb->aio_fildes = s->fd; 3021c1bb86cdSEric Blake acb->aio_offset = 0; 3022c1bb86cdSEric Blake acb->aio_ioctl_buf = buf; 3023c1bb86cdSEric Blake acb->aio_ioctl_cmd = req; 3024c1bb86cdSEric Blake pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 3025c1bb86cdSEric Blake return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque); 3026c1bb86cdSEric Blake } 3027c1bb86cdSEric Blake #endif /* linux */ 3028c1bb86cdSEric Blake 3029c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs) 3030c1bb86cdSEric Blake { 3031c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3032c1bb86cdSEric Blake 3033c1bb86cdSEric Blake /* this is just to ensure s->fd is sane (its called by io ops) */ 3034c1bb86cdSEric Blake if (s->fd >= 0) 3035c1bb86cdSEric Blake return 0; 3036c1bb86cdSEric Blake return -EIO; 3037c1bb86cdSEric Blake } 3038c1bb86cdSEric Blake 303933d70fb6SKevin Wolf static coroutine_fn int 304033d70fb6SKevin Wolf hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 3041c1bb86cdSEric Blake { 3042c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 304333d70fb6SKevin Wolf int ret; 3044c1bb86cdSEric Blake 304533d70fb6SKevin Wolf ret = fd_open(bs); 304633d70fb6SKevin Wolf if (ret < 0) { 304733d70fb6SKevin Wolf return ret; 3048c1bb86cdSEric Blake } 304933d70fb6SKevin Wolf return paio_submit_co(bs, s->fd, offset, NULL, bytes, 305033d70fb6SKevin Wolf QEMU_AIO_DISCARD | QEMU_AIO_BLKDEV); 3051c1bb86cdSEric Blake } 3052c1bb86cdSEric Blake 3053c1bb86cdSEric Blake static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, 3054f5a5ca79SManos Pitsidianakis int64_t offset, int bytes, BdrvRequestFlags flags) 3055c1bb86cdSEric Blake { 3056c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3057c1bb86cdSEric Blake int rc; 3058c1bb86cdSEric Blake 3059c1bb86cdSEric Blake rc = fd_open(bs); 3060c1bb86cdSEric Blake if (rc < 0) { 3061c1bb86cdSEric Blake return rc; 3062c1bb86cdSEric Blake } 3063c1bb86cdSEric Blake if (!(flags & BDRV_REQ_MAY_UNMAP)) { 3064f5a5ca79SManos Pitsidianakis return paio_submit_co(bs, s->fd, offset, NULL, bytes, 3065c1bb86cdSEric Blake QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV); 3066c1bb86cdSEric Blake } else if (s->discard_zeroes) { 3067f5a5ca79SManos Pitsidianakis return paio_submit_co(bs, s->fd, offset, NULL, bytes, 3068c1bb86cdSEric Blake QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV); 3069c1bb86cdSEric Blake } 3070c1bb86cdSEric Blake return -ENOTSUP; 3071c1bb86cdSEric Blake } 3072c1bb86cdSEric Blake 3073efc75e2aSStefan Hajnoczi static int coroutine_fn hdev_co_create_opts(const char *filename, QemuOpts *opts, 3074c1bb86cdSEric Blake Error **errp) 3075c1bb86cdSEric Blake { 3076c1bb86cdSEric Blake int fd; 3077c1bb86cdSEric Blake int ret = 0; 3078c1bb86cdSEric Blake struct stat stat_buf; 3079c1bb86cdSEric Blake int64_t total_size = 0; 3080c1bb86cdSEric Blake bool has_prefix; 3081c1bb86cdSEric Blake 3082c1bb86cdSEric Blake /* This function is used by both protocol block drivers and therefore either 3083c1bb86cdSEric Blake * of these prefixes may be given. 3084c1bb86cdSEric Blake * The return value has to be stored somewhere, otherwise this is an error 3085c1bb86cdSEric Blake * due to -Werror=unused-value. */ 3086c1bb86cdSEric Blake has_prefix = 3087c1bb86cdSEric Blake strstart(filename, "host_device:", &filename) || 3088c1bb86cdSEric Blake strstart(filename, "host_cdrom:" , &filename); 3089c1bb86cdSEric Blake 3090c1bb86cdSEric Blake (void)has_prefix; 3091c1bb86cdSEric Blake 3092c1bb86cdSEric Blake ret = raw_normalize_devicepath(&filename); 3093c1bb86cdSEric Blake if (ret < 0) { 3094c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not normalize device path"); 3095c1bb86cdSEric Blake return ret; 3096c1bb86cdSEric Blake } 3097c1bb86cdSEric Blake 3098c1bb86cdSEric Blake /* Read out options */ 3099c1bb86cdSEric Blake total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 3100c1bb86cdSEric Blake BDRV_SECTOR_SIZE); 3101c1bb86cdSEric Blake 3102c1bb86cdSEric Blake fd = qemu_open(filename, O_WRONLY | O_BINARY); 3103c1bb86cdSEric Blake if (fd < 0) { 3104c1bb86cdSEric Blake ret = -errno; 3105c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not open device"); 3106c1bb86cdSEric Blake return ret; 3107c1bb86cdSEric Blake } 3108c1bb86cdSEric Blake 3109c1bb86cdSEric Blake if (fstat(fd, &stat_buf) < 0) { 3110c1bb86cdSEric Blake ret = -errno; 3111c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not stat device"); 3112c1bb86cdSEric Blake } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) { 3113c1bb86cdSEric Blake error_setg(errp, 3114c1bb86cdSEric Blake "The given file is neither a block nor a character device"); 3115c1bb86cdSEric Blake ret = -ENODEV; 3116c1bb86cdSEric Blake } else if (lseek(fd, 0, SEEK_END) < total_size) { 3117c1bb86cdSEric Blake error_setg(errp, "Device is too small"); 3118c1bb86cdSEric Blake ret = -ENOSPC; 3119c1bb86cdSEric Blake } 3120c1bb86cdSEric Blake 312197ec9117SFam Zheng if (!ret && total_size) { 312297ec9117SFam Zheng uint8_t buf[BDRV_SECTOR_SIZE] = { 0 }; 312397ec9117SFam Zheng int64_t zero_size = MIN(BDRV_SECTOR_SIZE, total_size); 312497ec9117SFam Zheng if (lseek(fd, 0, SEEK_SET) == -1) { 312597ec9117SFam Zheng ret = -errno; 312697ec9117SFam Zheng } else { 312797ec9117SFam Zheng ret = qemu_write_full(fd, buf, zero_size); 312897ec9117SFam Zheng ret = ret == zero_size ? 0 : -errno; 312997ec9117SFam Zheng } 313097ec9117SFam Zheng } 3131c1bb86cdSEric Blake qemu_close(fd); 3132c1bb86cdSEric Blake return ret; 3133c1bb86cdSEric Blake } 3134c1bb86cdSEric Blake 3135c1bb86cdSEric Blake static BlockDriver bdrv_host_device = { 3136c1bb86cdSEric Blake .format_name = "host_device", 3137c1bb86cdSEric Blake .protocol_name = "host_device", 3138c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3139c1bb86cdSEric Blake .bdrv_needs_filename = true, 3140c1bb86cdSEric Blake .bdrv_probe_device = hdev_probe_device, 3141c1bb86cdSEric Blake .bdrv_parse_filename = hdev_parse_filename, 3142c1bb86cdSEric Blake .bdrv_file_open = hdev_open, 3143c1bb86cdSEric Blake .bdrv_close = raw_close, 3144c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3145c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3146c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3147efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = hdev_co_create_opts, 3148c1bb86cdSEric Blake .create_opts = &raw_create_opts, 3149dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3150c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, 3151c1bb86cdSEric Blake 3152c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3153c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 315433d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 315533d70fb6SKevin Wolf .bdrv_co_pdiscard = hdev_co_pdiscard, 31561efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 31571efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 3158c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3159c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3160c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3161*042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3162c1bb86cdSEric Blake 3163061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3164c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3165c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 3166c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3167c1bb86cdSEric Blake = raw_get_allocated_file_size, 3168244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 3169244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 3170244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 3171c1bb86cdSEric Blake .bdrv_probe_blocksizes = hdev_probe_blocksizes, 3172c1bb86cdSEric Blake .bdrv_probe_geometry = hdev_probe_geometry, 3173c1bb86cdSEric Blake 3174c1bb86cdSEric Blake /* generic scsi device */ 3175c1bb86cdSEric Blake #ifdef __linux__ 3176c1bb86cdSEric Blake .bdrv_aio_ioctl = hdev_aio_ioctl, 3177c1bb86cdSEric Blake #endif 3178c1bb86cdSEric Blake }; 3179c1bb86cdSEric Blake 3180c1bb86cdSEric Blake #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3181c1bb86cdSEric Blake static void cdrom_parse_filename(const char *filename, QDict *options, 3182c1bb86cdSEric Blake Error **errp) 3183c1bb86cdSEric Blake { 318403c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); 3185c1bb86cdSEric Blake } 3186c1bb86cdSEric Blake #endif 3187c1bb86cdSEric Blake 3188c1bb86cdSEric Blake #ifdef __linux__ 3189c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3190c1bb86cdSEric Blake Error **errp) 3191c1bb86cdSEric Blake { 3192c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3193c1bb86cdSEric Blake 3194c1bb86cdSEric Blake s->type = FTYPE_CD; 3195c1bb86cdSEric Blake 3196c1bb86cdSEric Blake /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ 3197230ff739SJohn Snow return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); 3198c1bb86cdSEric Blake } 3199c1bb86cdSEric Blake 3200c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3201c1bb86cdSEric Blake { 3202c1bb86cdSEric Blake int fd, ret; 3203c1bb86cdSEric Blake int prio = 0; 3204c1bb86cdSEric Blake struct stat st; 3205c1bb86cdSEric Blake 3206c1bb86cdSEric Blake fd = qemu_open(filename, O_RDONLY | O_NONBLOCK); 3207c1bb86cdSEric Blake if (fd < 0) { 3208c1bb86cdSEric Blake goto out; 3209c1bb86cdSEric Blake } 3210c1bb86cdSEric Blake ret = fstat(fd, &st); 3211c1bb86cdSEric Blake if (ret == -1 || !S_ISBLK(st.st_mode)) { 3212c1bb86cdSEric Blake goto outc; 3213c1bb86cdSEric Blake } 3214c1bb86cdSEric Blake 3215c1bb86cdSEric Blake /* Attempt to detect via a CDROM specific ioctl */ 3216c1bb86cdSEric Blake ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3217c1bb86cdSEric Blake if (ret >= 0) 3218c1bb86cdSEric Blake prio = 100; 3219c1bb86cdSEric Blake 3220c1bb86cdSEric Blake outc: 3221c1bb86cdSEric Blake qemu_close(fd); 3222c1bb86cdSEric Blake out: 3223c1bb86cdSEric Blake return prio; 3224c1bb86cdSEric Blake } 3225c1bb86cdSEric Blake 3226c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3227c1bb86cdSEric Blake { 3228c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3229c1bb86cdSEric Blake int ret; 3230c1bb86cdSEric Blake 3231c1bb86cdSEric Blake ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3232c1bb86cdSEric Blake return ret == CDS_DISC_OK; 3233c1bb86cdSEric Blake } 3234c1bb86cdSEric Blake 3235c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3236c1bb86cdSEric Blake { 3237c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3238c1bb86cdSEric Blake 3239c1bb86cdSEric Blake if (eject_flag) { 3240c1bb86cdSEric Blake if (ioctl(s->fd, CDROMEJECT, NULL) < 0) 3241c1bb86cdSEric Blake perror("CDROMEJECT"); 3242c1bb86cdSEric Blake } else { 3243c1bb86cdSEric Blake if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) 3244c1bb86cdSEric Blake perror("CDROMEJECT"); 3245c1bb86cdSEric Blake } 3246c1bb86cdSEric Blake } 3247c1bb86cdSEric Blake 3248c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3249c1bb86cdSEric Blake { 3250c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3251c1bb86cdSEric Blake 3252c1bb86cdSEric Blake if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { 3253c1bb86cdSEric Blake /* 3254c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3255c1bb86cdSEric Blake * mounts the CD-ROM 3256c1bb86cdSEric Blake */ 3257c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3258c1bb86cdSEric Blake } 3259c1bb86cdSEric Blake } 3260c1bb86cdSEric Blake 3261c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3262c1bb86cdSEric Blake .format_name = "host_cdrom", 3263c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3264c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3265c1bb86cdSEric Blake .bdrv_needs_filename = true, 3266c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3267c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3268c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3269c1bb86cdSEric Blake .bdrv_close = raw_close, 3270c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3271c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3272c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3273efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = hdev_co_create_opts, 3274c1bb86cdSEric Blake .create_opts = &raw_create_opts, 3275dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3276c1bb86cdSEric Blake 3277c1bb86cdSEric Blake 3278c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3279c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 328033d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3281c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3282c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3283c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3284*042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3285c1bb86cdSEric Blake 3286061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3287c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3288c1bb86cdSEric Blake .has_variable_length = true, 3289c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3290c1bb86cdSEric Blake = raw_get_allocated_file_size, 3291c1bb86cdSEric Blake 3292c1bb86cdSEric Blake /* removable device support */ 3293c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3294c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3295c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3296c1bb86cdSEric Blake 3297c1bb86cdSEric Blake /* generic scsi device */ 3298c1bb86cdSEric Blake .bdrv_aio_ioctl = hdev_aio_ioctl, 3299c1bb86cdSEric Blake }; 3300c1bb86cdSEric Blake #endif /* __linux__ */ 3301c1bb86cdSEric Blake 3302c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 3303c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3304c1bb86cdSEric Blake Error **errp) 3305c1bb86cdSEric Blake { 3306c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3307c1bb86cdSEric Blake Error *local_err = NULL; 3308c1bb86cdSEric Blake int ret; 3309c1bb86cdSEric Blake 3310c1bb86cdSEric Blake s->type = FTYPE_CD; 3311c1bb86cdSEric Blake 3312230ff739SJohn Snow ret = raw_open_common(bs, options, flags, 0, true, &local_err); 3313c1bb86cdSEric Blake if (ret) { 3314c1bb86cdSEric Blake error_propagate(errp, local_err); 3315c1bb86cdSEric Blake return ret; 3316c1bb86cdSEric Blake } 3317c1bb86cdSEric Blake 3318c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3319c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3320c1bb86cdSEric Blake return 0; 3321c1bb86cdSEric Blake } 3322c1bb86cdSEric Blake 3323c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3324c1bb86cdSEric Blake { 3325c1bb86cdSEric Blake if (strstart(filename, "/dev/cd", NULL) || 3326c1bb86cdSEric Blake strstart(filename, "/dev/acd", NULL)) 3327c1bb86cdSEric Blake return 100; 3328c1bb86cdSEric Blake return 0; 3329c1bb86cdSEric Blake } 3330c1bb86cdSEric Blake 3331c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs) 3332c1bb86cdSEric Blake { 3333c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3334c1bb86cdSEric Blake int fd; 3335c1bb86cdSEric Blake 3336c1bb86cdSEric Blake /* 3337c1bb86cdSEric Blake * Force reread of possibly changed/newly loaded disc, 3338c1bb86cdSEric Blake * FreeBSD seems to not notice sometimes... 3339c1bb86cdSEric Blake */ 3340c1bb86cdSEric Blake if (s->fd >= 0) 3341c1bb86cdSEric Blake qemu_close(s->fd); 3342c1bb86cdSEric Blake fd = qemu_open(bs->filename, s->open_flags, 0644); 3343c1bb86cdSEric Blake if (fd < 0) { 3344c1bb86cdSEric Blake s->fd = -1; 3345c1bb86cdSEric Blake return -EIO; 3346c1bb86cdSEric Blake } 3347c1bb86cdSEric Blake s->fd = fd; 3348c1bb86cdSEric Blake 3349c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3350c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3351c1bb86cdSEric Blake return 0; 3352c1bb86cdSEric Blake } 3353c1bb86cdSEric Blake 3354c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3355c1bb86cdSEric Blake { 3356c1bb86cdSEric Blake return raw_getlength(bs) > 0; 3357c1bb86cdSEric Blake } 3358c1bb86cdSEric Blake 3359c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3360c1bb86cdSEric Blake { 3361c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3362c1bb86cdSEric Blake 3363c1bb86cdSEric Blake if (s->fd < 0) 3364c1bb86cdSEric Blake return; 3365c1bb86cdSEric Blake 3366c1bb86cdSEric Blake (void) ioctl(s->fd, CDIOCALLOW); 3367c1bb86cdSEric Blake 3368c1bb86cdSEric Blake if (eject_flag) { 3369c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCEJECT) < 0) 3370c1bb86cdSEric Blake perror("CDIOCEJECT"); 3371c1bb86cdSEric Blake } else { 3372c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCCLOSE) < 0) 3373c1bb86cdSEric Blake perror("CDIOCCLOSE"); 3374c1bb86cdSEric Blake } 3375c1bb86cdSEric Blake 3376c1bb86cdSEric Blake cdrom_reopen(bs); 3377c1bb86cdSEric Blake } 3378c1bb86cdSEric Blake 3379c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3380c1bb86cdSEric Blake { 3381c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3382c1bb86cdSEric Blake 3383c1bb86cdSEric Blake if (s->fd < 0) 3384c1bb86cdSEric Blake return; 3385c1bb86cdSEric Blake if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { 3386c1bb86cdSEric Blake /* 3387c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3388c1bb86cdSEric Blake * mounts the CD-ROM 3389c1bb86cdSEric Blake */ 3390c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3391c1bb86cdSEric Blake } 3392c1bb86cdSEric Blake } 3393c1bb86cdSEric Blake 3394c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3395c1bb86cdSEric Blake .format_name = "host_cdrom", 3396c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3397c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3398c1bb86cdSEric Blake .bdrv_needs_filename = true, 3399c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3400c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3401c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3402c1bb86cdSEric Blake .bdrv_close = raw_close, 3403c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3404c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3405c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3406efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = hdev_co_create_opts, 3407c1bb86cdSEric Blake .create_opts = &raw_create_opts, 3408c1bb86cdSEric Blake 3409c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3410c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 341133d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3412c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3413c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3414c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3415*042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3416c1bb86cdSEric Blake 3417061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3418c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3419c1bb86cdSEric Blake .has_variable_length = true, 3420c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3421c1bb86cdSEric Blake = raw_get_allocated_file_size, 3422c1bb86cdSEric Blake 3423c1bb86cdSEric Blake /* removable device support */ 3424c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3425c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3426c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3427c1bb86cdSEric Blake }; 3428c1bb86cdSEric Blake #endif /* __FreeBSD__ */ 3429c1bb86cdSEric Blake 3430c1bb86cdSEric Blake static void bdrv_file_init(void) 3431c1bb86cdSEric Blake { 3432c1bb86cdSEric Blake /* 3433c1bb86cdSEric Blake * Register all the drivers. Note that order is important, the driver 3434c1bb86cdSEric Blake * registered last will get probed first. 3435c1bb86cdSEric Blake */ 3436c1bb86cdSEric Blake bdrv_register(&bdrv_file); 3437c1bb86cdSEric Blake bdrv_register(&bdrv_host_device); 3438c1bb86cdSEric Blake #ifdef __linux__ 3439c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3440c1bb86cdSEric Blake #endif 3441c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3442c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3443c1bb86cdSEric Blake #endif 3444c1bb86cdSEric Blake } 3445c1bb86cdSEric Blake 3446c1bb86cdSEric Blake block_init(bdrv_file_init); 3447