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 */ 24c1bb86cdSEric Blake #include "qemu/osdep.h" 25c1bb86cdSEric Blake #include "qapi/error.h" 26c1bb86cdSEric Blake #include "qemu/cutils.h" 27c1bb86cdSEric Blake #include "qemu/error-report.h" 28c1bb86cdSEric Blake #include "block/block_int.h" 29c1bb86cdSEric Blake #include "qemu/module.h" 30c1bb86cdSEric Blake #include "trace.h" 31c1bb86cdSEric Blake #include "block/thread-pool.h" 32c1bb86cdSEric Blake #include "qemu/iov.h" 33c1bb86cdSEric Blake #include "block/raw-aio.h" 34c1bb86cdSEric Blake #include "qapi/util.h" 35c1bb86cdSEric Blake #include "qapi/qmp/qstring.h" 36c1bb86cdSEric Blake 37c1bb86cdSEric Blake #if defined(__APPLE__) && (__MACH__) 38c1bb86cdSEric Blake #include <paths.h> 39c1bb86cdSEric Blake #include <sys/param.h> 40c1bb86cdSEric Blake #include <IOKit/IOKitLib.h> 41c1bb86cdSEric Blake #include <IOKit/IOBSD.h> 42c1bb86cdSEric Blake #include <IOKit/storage/IOMediaBSDClient.h> 43c1bb86cdSEric Blake #include <IOKit/storage/IOMedia.h> 44c1bb86cdSEric Blake #include <IOKit/storage/IOCDMedia.h> 45c1bb86cdSEric Blake //#include <IOKit/storage/IOCDTypes.h> 46c1bb86cdSEric Blake #include <IOKit/storage/IODVDMedia.h> 47c1bb86cdSEric Blake #include <CoreFoundation/CoreFoundation.h> 48c1bb86cdSEric Blake #endif 49c1bb86cdSEric Blake 50c1bb86cdSEric Blake #ifdef __sun__ 51c1bb86cdSEric Blake #define _POSIX_PTHREAD_SEMANTICS 1 52c1bb86cdSEric Blake #include <sys/dkio.h> 53c1bb86cdSEric Blake #endif 54c1bb86cdSEric Blake #ifdef __linux__ 55c1bb86cdSEric Blake #include <sys/ioctl.h> 56c1bb86cdSEric Blake #include <sys/param.h> 57c1bb86cdSEric Blake #include <linux/cdrom.h> 58c1bb86cdSEric Blake #include <linux/fd.h> 59c1bb86cdSEric Blake #include <linux/fs.h> 60c1bb86cdSEric Blake #include <linux/hdreg.h> 61c1bb86cdSEric Blake #include <scsi/sg.h> 62c1bb86cdSEric Blake #ifdef __s390__ 63c1bb86cdSEric Blake #include <asm/dasd.h> 64c1bb86cdSEric Blake #endif 65c1bb86cdSEric Blake #ifndef FS_NOCOW_FL 66c1bb86cdSEric Blake #define FS_NOCOW_FL 0x00800000 /* Do not cow file */ 67c1bb86cdSEric Blake #endif 68c1bb86cdSEric Blake #endif 69c1bb86cdSEric Blake #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE) 70c1bb86cdSEric Blake #include <linux/falloc.h> 71c1bb86cdSEric Blake #endif 72c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 73c1bb86cdSEric Blake #include <sys/disk.h> 74c1bb86cdSEric Blake #include <sys/cdio.h> 75c1bb86cdSEric Blake #endif 76c1bb86cdSEric Blake 77c1bb86cdSEric Blake #ifdef __OpenBSD__ 78c1bb86cdSEric Blake #include <sys/ioctl.h> 79c1bb86cdSEric Blake #include <sys/disklabel.h> 80c1bb86cdSEric Blake #include <sys/dkio.h> 81c1bb86cdSEric Blake #endif 82c1bb86cdSEric Blake 83c1bb86cdSEric Blake #ifdef __NetBSD__ 84c1bb86cdSEric Blake #include <sys/ioctl.h> 85c1bb86cdSEric Blake #include <sys/disklabel.h> 86c1bb86cdSEric Blake #include <sys/dkio.h> 87c1bb86cdSEric Blake #include <sys/disk.h> 88c1bb86cdSEric Blake #endif 89c1bb86cdSEric Blake 90c1bb86cdSEric Blake #ifdef __DragonFly__ 91c1bb86cdSEric Blake #include <sys/ioctl.h> 92c1bb86cdSEric Blake #include <sys/diskslice.h> 93c1bb86cdSEric Blake #endif 94c1bb86cdSEric Blake 95c1bb86cdSEric Blake #ifdef CONFIG_XFS 96c1bb86cdSEric Blake #include <xfs/xfs.h> 97c1bb86cdSEric Blake #endif 98c1bb86cdSEric Blake 99c1bb86cdSEric Blake //#define DEBUG_BLOCK 100c1bb86cdSEric Blake 101c1bb86cdSEric Blake #ifdef DEBUG_BLOCK 102c1bb86cdSEric Blake # define DEBUG_BLOCK_PRINT 1 103c1bb86cdSEric Blake #else 104c1bb86cdSEric Blake # define DEBUG_BLOCK_PRINT 0 105c1bb86cdSEric Blake #endif 106c1bb86cdSEric Blake #define DPRINTF(fmt, ...) \ 107c1bb86cdSEric Blake do { \ 108c1bb86cdSEric Blake if (DEBUG_BLOCK_PRINT) { \ 109c1bb86cdSEric Blake printf(fmt, ## __VA_ARGS__); \ 110c1bb86cdSEric Blake } \ 111c1bb86cdSEric Blake } while (0) 112c1bb86cdSEric Blake 113c1bb86cdSEric Blake /* OS X does not have O_DSYNC */ 114c1bb86cdSEric Blake #ifndef O_DSYNC 115c1bb86cdSEric Blake #ifdef O_SYNC 116c1bb86cdSEric Blake #define O_DSYNC O_SYNC 117c1bb86cdSEric Blake #elif defined(O_FSYNC) 118c1bb86cdSEric Blake #define O_DSYNC O_FSYNC 119c1bb86cdSEric Blake #endif 120c1bb86cdSEric Blake #endif 121c1bb86cdSEric Blake 122c1bb86cdSEric Blake /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ 123c1bb86cdSEric Blake #ifndef O_DIRECT 124c1bb86cdSEric Blake #define O_DIRECT O_DSYNC 125c1bb86cdSEric Blake #endif 126c1bb86cdSEric Blake 127c1bb86cdSEric Blake #define FTYPE_FILE 0 128c1bb86cdSEric Blake #define FTYPE_CD 1 129c1bb86cdSEric Blake 130c1bb86cdSEric Blake #define MAX_BLOCKSIZE 4096 131c1bb86cdSEric Blake 132*244a5668SFam Zheng /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes, 133*244a5668SFam Zheng * leaving a few more bytes for its future use. */ 134*244a5668SFam Zheng #define RAW_LOCK_PERM_BASE 100 135*244a5668SFam Zheng #define RAW_LOCK_SHARED_BASE 200 136*244a5668SFam Zheng 137c1bb86cdSEric Blake typedef struct BDRVRawState { 138c1bb86cdSEric Blake int fd; 139*244a5668SFam Zheng int lock_fd; 140*244a5668SFam Zheng bool use_lock; 141c1bb86cdSEric Blake int type; 142c1bb86cdSEric Blake int open_flags; 143c1bb86cdSEric Blake size_t buf_align; 144c1bb86cdSEric Blake 145*244a5668SFam Zheng /* The current permissions. */ 146*244a5668SFam Zheng uint64_t perm; 147*244a5668SFam Zheng uint64_t shared_perm; 148*244a5668SFam Zheng 149c1bb86cdSEric Blake #ifdef CONFIG_XFS 150c1bb86cdSEric Blake bool is_xfs:1; 151c1bb86cdSEric Blake #endif 152c1bb86cdSEric Blake bool has_discard:1; 153c1bb86cdSEric Blake bool has_write_zeroes:1; 154c1bb86cdSEric Blake bool discard_zeroes:1; 155c1bb86cdSEric Blake bool use_linux_aio:1; 156e5bcf967SKevin Wolf bool page_cache_inconsistent:1; 157c1bb86cdSEric Blake bool has_fallocate; 158c1bb86cdSEric Blake bool needs_alignment; 159c1bb86cdSEric Blake } BDRVRawState; 160c1bb86cdSEric Blake 161c1bb86cdSEric Blake typedef struct BDRVRawReopenState { 162c1bb86cdSEric Blake int fd; 163c1bb86cdSEric Blake int open_flags; 164c1bb86cdSEric Blake } BDRVRawReopenState; 165c1bb86cdSEric Blake 166c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs); 167c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs); 168c1bb86cdSEric Blake 169c1bb86cdSEric Blake typedef struct RawPosixAIOData { 170c1bb86cdSEric Blake BlockDriverState *bs; 171c1bb86cdSEric Blake int aio_fildes; 172c1bb86cdSEric Blake union { 173c1bb86cdSEric Blake struct iovec *aio_iov; 174c1bb86cdSEric Blake void *aio_ioctl_buf; 175c1bb86cdSEric Blake }; 176c1bb86cdSEric Blake int aio_niov; 177c1bb86cdSEric Blake uint64_t aio_nbytes; 178c1bb86cdSEric Blake #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */ 179c1bb86cdSEric Blake off_t aio_offset; 180c1bb86cdSEric Blake int aio_type; 181c1bb86cdSEric Blake } RawPosixAIOData; 182c1bb86cdSEric Blake 183c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 184c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs); 185c1bb86cdSEric Blake #endif 186c1bb86cdSEric Blake 187c1bb86cdSEric Blake #if defined(__NetBSD__) 188c1bb86cdSEric Blake static int raw_normalize_devicepath(const char **filename) 189c1bb86cdSEric Blake { 190c1bb86cdSEric Blake static char namebuf[PATH_MAX]; 191c1bb86cdSEric Blake const char *dp, *fname; 192c1bb86cdSEric Blake struct stat sb; 193c1bb86cdSEric Blake 194c1bb86cdSEric Blake fname = *filename; 195c1bb86cdSEric Blake dp = strrchr(fname, '/'); 196c1bb86cdSEric Blake if (lstat(fname, &sb) < 0) { 197c1bb86cdSEric Blake fprintf(stderr, "%s: stat failed: %s\n", 198c1bb86cdSEric Blake fname, strerror(errno)); 199c1bb86cdSEric Blake return -errno; 200c1bb86cdSEric Blake } 201c1bb86cdSEric Blake 202c1bb86cdSEric Blake if (!S_ISBLK(sb.st_mode)) { 203c1bb86cdSEric Blake return 0; 204c1bb86cdSEric Blake } 205c1bb86cdSEric Blake 206c1bb86cdSEric Blake if (dp == NULL) { 207c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "r%s", fname); 208c1bb86cdSEric Blake } else { 209c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "%.*s/r%s", 210c1bb86cdSEric Blake (int)(dp - fname), fname, dp + 1); 211c1bb86cdSEric Blake } 212c1bb86cdSEric Blake fprintf(stderr, "%s is a block device", fname); 213c1bb86cdSEric Blake *filename = namebuf; 214c1bb86cdSEric Blake fprintf(stderr, ", using %s\n", *filename); 215c1bb86cdSEric Blake 216c1bb86cdSEric Blake return 0; 217c1bb86cdSEric Blake } 218c1bb86cdSEric Blake #else 219c1bb86cdSEric Blake static int raw_normalize_devicepath(const char **filename) 220c1bb86cdSEric Blake { 221c1bb86cdSEric Blake return 0; 222c1bb86cdSEric Blake } 223c1bb86cdSEric Blake #endif 224c1bb86cdSEric Blake 225c1bb86cdSEric Blake /* 226c1bb86cdSEric Blake * Get logical block size via ioctl. On success store it in @sector_size_p. 227c1bb86cdSEric Blake */ 228c1bb86cdSEric Blake static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) 229c1bb86cdSEric Blake { 230c1bb86cdSEric Blake unsigned int sector_size; 231c1bb86cdSEric Blake bool success = false; 232700f9ce0SPeter Maydell int i; 233c1bb86cdSEric Blake 234c1bb86cdSEric Blake errno = ENOTSUP; 235700f9ce0SPeter Maydell static const unsigned long ioctl_list[] = { 236c1bb86cdSEric Blake #ifdef BLKSSZGET 237700f9ce0SPeter Maydell BLKSSZGET, 238c1bb86cdSEric Blake #endif 239c1bb86cdSEric Blake #ifdef DKIOCGETBLOCKSIZE 240700f9ce0SPeter Maydell DKIOCGETBLOCKSIZE, 241c1bb86cdSEric Blake #endif 242c1bb86cdSEric Blake #ifdef DIOCGSECTORSIZE 243700f9ce0SPeter Maydell DIOCGSECTORSIZE, 244700f9ce0SPeter Maydell #endif 245700f9ce0SPeter Maydell }; 246700f9ce0SPeter Maydell 247700f9ce0SPeter Maydell /* Try a few ioctls to get the right size */ 248700f9ce0SPeter Maydell for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { 249700f9ce0SPeter Maydell if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { 250c1bb86cdSEric Blake *sector_size_p = sector_size; 251c1bb86cdSEric Blake success = true; 252c1bb86cdSEric Blake } 253700f9ce0SPeter Maydell } 254c1bb86cdSEric Blake 255c1bb86cdSEric Blake return success ? 0 : -errno; 256c1bb86cdSEric Blake } 257c1bb86cdSEric Blake 258c1bb86cdSEric Blake /** 259c1bb86cdSEric Blake * Get physical block size of @fd. 260c1bb86cdSEric Blake * On success, store it in @blk_size and return 0. 261c1bb86cdSEric Blake * On failure, return -errno. 262c1bb86cdSEric Blake */ 263c1bb86cdSEric Blake static int probe_physical_blocksize(int fd, unsigned int *blk_size) 264c1bb86cdSEric Blake { 265c1bb86cdSEric Blake #ifdef BLKPBSZGET 266c1bb86cdSEric Blake if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { 267c1bb86cdSEric Blake return -errno; 268c1bb86cdSEric Blake } 269c1bb86cdSEric Blake return 0; 270c1bb86cdSEric Blake #else 271c1bb86cdSEric Blake return -ENOTSUP; 272c1bb86cdSEric Blake #endif 273c1bb86cdSEric Blake } 274c1bb86cdSEric Blake 275c1bb86cdSEric Blake /* Check if read is allowed with given memory buffer and length. 276c1bb86cdSEric Blake * 277c1bb86cdSEric Blake * This function is used to check O_DIRECT memory buffer and request alignment. 278c1bb86cdSEric Blake */ 279c1bb86cdSEric Blake static bool raw_is_io_aligned(int fd, void *buf, size_t len) 280c1bb86cdSEric Blake { 281c1bb86cdSEric Blake ssize_t ret = pread(fd, buf, len, 0); 282c1bb86cdSEric Blake 283c1bb86cdSEric Blake if (ret >= 0) { 284c1bb86cdSEric Blake return true; 285c1bb86cdSEric Blake } 286c1bb86cdSEric Blake 287c1bb86cdSEric Blake #ifdef __linux__ 288c1bb86cdSEric Blake /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore 289c1bb86cdSEric Blake * other errors (e.g. real I/O error), which could happen on a failed 290c1bb86cdSEric Blake * drive, since we only care about probing alignment. 291c1bb86cdSEric Blake */ 292c1bb86cdSEric Blake if (errno != EINVAL) { 293c1bb86cdSEric Blake return true; 294c1bb86cdSEric Blake } 295c1bb86cdSEric Blake #endif 296c1bb86cdSEric Blake 297c1bb86cdSEric Blake return false; 298c1bb86cdSEric Blake } 299c1bb86cdSEric Blake 300c1bb86cdSEric Blake static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) 301c1bb86cdSEric Blake { 302c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 303c1bb86cdSEric Blake char *buf; 304c1bb86cdSEric Blake size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize()); 305c1bb86cdSEric Blake 306c1bb86cdSEric Blake /* For SCSI generic devices the alignment is not really used. 307c1bb86cdSEric Blake With buffered I/O, we don't have any restrictions. */ 308c1bb86cdSEric Blake if (bdrv_is_sg(bs) || !s->needs_alignment) { 309c1bb86cdSEric Blake bs->bl.request_alignment = 1; 310c1bb86cdSEric Blake s->buf_align = 1; 311c1bb86cdSEric Blake return; 312c1bb86cdSEric Blake } 313c1bb86cdSEric Blake 314c1bb86cdSEric Blake bs->bl.request_alignment = 0; 315c1bb86cdSEric Blake s->buf_align = 0; 316c1bb86cdSEric Blake /* Let's try to use the logical blocksize for the alignment. */ 317c1bb86cdSEric Blake if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { 318c1bb86cdSEric Blake bs->bl.request_alignment = 0; 319c1bb86cdSEric Blake } 320c1bb86cdSEric Blake #ifdef CONFIG_XFS 321c1bb86cdSEric Blake if (s->is_xfs) { 322c1bb86cdSEric Blake struct dioattr da; 323c1bb86cdSEric Blake if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { 324c1bb86cdSEric Blake bs->bl.request_alignment = da.d_miniosz; 325c1bb86cdSEric Blake /* The kernel returns wrong information for d_mem */ 326c1bb86cdSEric Blake /* s->buf_align = da.d_mem; */ 327c1bb86cdSEric Blake } 328c1bb86cdSEric Blake } 329c1bb86cdSEric Blake #endif 330c1bb86cdSEric Blake 331c1bb86cdSEric Blake /* If we could not get the sizes so far, we can only guess them */ 332c1bb86cdSEric Blake if (!s->buf_align) { 333c1bb86cdSEric Blake size_t align; 334c1bb86cdSEric Blake buf = qemu_memalign(max_align, 2 * max_align); 335c1bb86cdSEric Blake for (align = 512; align <= max_align; align <<= 1) { 336c1bb86cdSEric Blake if (raw_is_io_aligned(fd, buf + align, max_align)) { 337c1bb86cdSEric Blake s->buf_align = align; 338c1bb86cdSEric Blake break; 339c1bb86cdSEric Blake } 340c1bb86cdSEric Blake } 341c1bb86cdSEric Blake qemu_vfree(buf); 342c1bb86cdSEric Blake } 343c1bb86cdSEric Blake 344c1bb86cdSEric Blake if (!bs->bl.request_alignment) { 345c1bb86cdSEric Blake size_t align; 346c1bb86cdSEric Blake buf = qemu_memalign(s->buf_align, max_align); 347c1bb86cdSEric Blake for (align = 512; align <= max_align; align <<= 1) { 348c1bb86cdSEric Blake if (raw_is_io_aligned(fd, buf, align)) { 349c1bb86cdSEric Blake bs->bl.request_alignment = align; 350c1bb86cdSEric Blake break; 351c1bb86cdSEric Blake } 352c1bb86cdSEric Blake } 353c1bb86cdSEric Blake qemu_vfree(buf); 354c1bb86cdSEric Blake } 355c1bb86cdSEric Blake 356c1bb86cdSEric Blake if (!s->buf_align || !bs->bl.request_alignment) { 357c1bb86cdSEric Blake error_setg(errp, "Could not find working O_DIRECT alignment"); 358c1bb86cdSEric Blake error_append_hint(errp, "Try cache.direct=off\n"); 359c1bb86cdSEric Blake } 360c1bb86cdSEric Blake } 361c1bb86cdSEric Blake 362c1bb86cdSEric Blake static void raw_parse_flags(int bdrv_flags, int *open_flags) 363c1bb86cdSEric Blake { 364c1bb86cdSEric Blake assert(open_flags != NULL); 365c1bb86cdSEric Blake 366c1bb86cdSEric Blake *open_flags |= O_BINARY; 367c1bb86cdSEric Blake *open_flags &= ~O_ACCMODE; 368c1bb86cdSEric Blake if (bdrv_flags & BDRV_O_RDWR) { 369c1bb86cdSEric Blake *open_flags |= O_RDWR; 370c1bb86cdSEric Blake } else { 371c1bb86cdSEric Blake *open_flags |= O_RDONLY; 372c1bb86cdSEric Blake } 373c1bb86cdSEric Blake 374c1bb86cdSEric Blake /* Use O_DSYNC for write-through caching, no flags for write-back caching, 375c1bb86cdSEric Blake * and O_DIRECT for no caching. */ 376c1bb86cdSEric Blake if ((bdrv_flags & BDRV_O_NOCACHE)) { 377c1bb86cdSEric Blake *open_flags |= O_DIRECT; 378c1bb86cdSEric Blake } 379c1bb86cdSEric Blake } 380c1bb86cdSEric Blake 381c1bb86cdSEric Blake static void raw_parse_filename(const char *filename, QDict *options, 382c1bb86cdSEric Blake Error **errp) 383c1bb86cdSEric Blake { 384c1bb86cdSEric Blake /* The filename does not have to be prefixed by the protocol name, since 385c1bb86cdSEric Blake * "file" is the default protocol; therefore, the return value of this 386c1bb86cdSEric Blake * function call can be ignored. */ 387c1bb86cdSEric Blake strstart(filename, "file:", &filename); 388c1bb86cdSEric Blake 38946f5ac20SEric Blake qdict_put_str(options, "filename", filename); 390c1bb86cdSEric Blake } 391c1bb86cdSEric Blake 392c1bb86cdSEric Blake static QemuOptsList raw_runtime_opts = { 393c1bb86cdSEric Blake .name = "raw", 394c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), 395c1bb86cdSEric Blake .desc = { 396c1bb86cdSEric Blake { 397c1bb86cdSEric Blake .name = "filename", 398c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 399c1bb86cdSEric Blake .help = "File name of the image", 400c1bb86cdSEric Blake }, 401c1bb86cdSEric Blake { 402c1bb86cdSEric Blake .name = "aio", 403c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 404c1bb86cdSEric Blake .help = "host AIO implementation (threads, native)", 405c1bb86cdSEric Blake }, 40616b48d5dSFam Zheng { 40716b48d5dSFam Zheng .name = "locking", 40816b48d5dSFam Zheng .type = QEMU_OPT_STRING, 40916b48d5dSFam Zheng .help = "file locking mode (on/off/auto, default: auto)", 41016b48d5dSFam Zheng }, 411c1bb86cdSEric Blake { /* end of list */ } 412c1bb86cdSEric Blake }, 413c1bb86cdSEric Blake }; 414c1bb86cdSEric Blake 415c1bb86cdSEric Blake static int raw_open_common(BlockDriverState *bs, QDict *options, 416c1bb86cdSEric Blake int bdrv_flags, int open_flags, Error **errp) 417c1bb86cdSEric Blake { 418c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 419c1bb86cdSEric Blake QemuOpts *opts; 420c1bb86cdSEric Blake Error *local_err = NULL; 421c1bb86cdSEric Blake const char *filename = NULL; 422c1bb86cdSEric Blake BlockdevAioOptions aio, aio_default; 423c1bb86cdSEric Blake int fd, ret; 424c1bb86cdSEric Blake struct stat st; 425*244a5668SFam Zheng OnOffAuto locking; 426c1bb86cdSEric Blake 427c1bb86cdSEric Blake opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 428c1bb86cdSEric Blake qemu_opts_absorb_qdict(opts, options, &local_err); 429c1bb86cdSEric Blake if (local_err) { 430c1bb86cdSEric Blake error_propagate(errp, local_err); 431c1bb86cdSEric Blake ret = -EINVAL; 432c1bb86cdSEric Blake goto fail; 433c1bb86cdSEric Blake } 434c1bb86cdSEric Blake 435c1bb86cdSEric Blake filename = qemu_opt_get(opts, "filename"); 436c1bb86cdSEric Blake 437c1bb86cdSEric Blake ret = raw_normalize_devicepath(&filename); 438c1bb86cdSEric Blake if (ret != 0) { 439c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not normalize device path"); 440c1bb86cdSEric Blake goto fail; 441c1bb86cdSEric Blake } 442c1bb86cdSEric Blake 443c1bb86cdSEric Blake aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO) 444c1bb86cdSEric Blake ? BLOCKDEV_AIO_OPTIONS_NATIVE 445c1bb86cdSEric Blake : BLOCKDEV_AIO_OPTIONS_THREADS; 446c1bb86cdSEric Blake aio = qapi_enum_parse(BlockdevAioOptions_lookup, qemu_opt_get(opts, "aio"), 447c1bb86cdSEric Blake BLOCKDEV_AIO_OPTIONS__MAX, aio_default, &local_err); 448c1bb86cdSEric Blake if (local_err) { 449c1bb86cdSEric Blake error_propagate(errp, local_err); 450c1bb86cdSEric Blake ret = -EINVAL; 451c1bb86cdSEric Blake goto fail; 452c1bb86cdSEric Blake } 453c1bb86cdSEric Blake s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); 454c1bb86cdSEric Blake 455*244a5668SFam Zheng locking = qapi_enum_parse(OnOffAuto_lookup, qemu_opt_get(opts, "locking"), 456*244a5668SFam Zheng ON_OFF_AUTO__MAX, ON_OFF_AUTO_AUTO, &local_err); 457*244a5668SFam Zheng if (local_err) { 458*244a5668SFam Zheng error_propagate(errp, local_err); 459*244a5668SFam Zheng ret = -EINVAL; 460*244a5668SFam Zheng goto fail; 461*244a5668SFam Zheng } 462*244a5668SFam Zheng switch (locking) { 463*244a5668SFam Zheng case ON_OFF_AUTO_ON: 464*244a5668SFam Zheng s->use_lock = true; 465*244a5668SFam Zheng #ifndef F_OFD_SETLK 466*244a5668SFam Zheng fprintf(stderr, 467*244a5668SFam Zheng "File lock requested but OFD locking syscall is unavailable, " 468*244a5668SFam Zheng "falling back to POSIX file locks.\n" 469*244a5668SFam Zheng "Due to the implementation, locks can be lost unexpectedly.\n"); 470*244a5668SFam Zheng #endif 471*244a5668SFam Zheng break; 472*244a5668SFam Zheng case ON_OFF_AUTO_OFF: 473*244a5668SFam Zheng s->use_lock = false; 474*244a5668SFam Zheng break; 475*244a5668SFam Zheng case ON_OFF_AUTO_AUTO: 476*244a5668SFam Zheng #ifdef F_OFD_SETLK 477*244a5668SFam Zheng s->use_lock = true; 478*244a5668SFam Zheng #else 479*244a5668SFam Zheng s->use_lock = false; 480*244a5668SFam Zheng #endif 481*244a5668SFam Zheng break; 482*244a5668SFam Zheng default: 483*244a5668SFam Zheng abort(); 484*244a5668SFam Zheng } 485*244a5668SFam Zheng 486c1bb86cdSEric Blake s->open_flags = open_flags; 487c1bb86cdSEric Blake raw_parse_flags(bdrv_flags, &s->open_flags); 488c1bb86cdSEric Blake 489c1bb86cdSEric Blake s->fd = -1; 490c1bb86cdSEric Blake fd = qemu_open(filename, s->open_flags, 0644); 491c1bb86cdSEric Blake if (fd < 0) { 492c1bb86cdSEric Blake ret = -errno; 493c1bb86cdSEric Blake error_setg_errno(errp, errno, "Could not open '%s'", filename); 494c1bb86cdSEric Blake if (ret == -EROFS) { 495c1bb86cdSEric Blake ret = -EACCES; 496c1bb86cdSEric Blake } 497c1bb86cdSEric Blake goto fail; 498c1bb86cdSEric Blake } 499c1bb86cdSEric Blake s->fd = fd; 500c1bb86cdSEric Blake 501*244a5668SFam Zheng s->lock_fd = -1; 502*244a5668SFam Zheng if (s->use_lock) { 503*244a5668SFam Zheng fd = qemu_open(filename, s->open_flags); 504*244a5668SFam Zheng if (fd < 0) { 505*244a5668SFam Zheng ret = -errno; 506*244a5668SFam Zheng error_setg_errno(errp, errno, "Could not open '%s' for locking", 507*244a5668SFam Zheng filename); 508*244a5668SFam Zheng qemu_close(s->fd); 509*244a5668SFam Zheng goto fail; 510*244a5668SFam Zheng } 511*244a5668SFam Zheng s->lock_fd = fd; 512*244a5668SFam Zheng } 513*244a5668SFam Zheng s->perm = 0; 514*244a5668SFam Zheng s->shared_perm = BLK_PERM_ALL; 515*244a5668SFam Zheng 516c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 517c1bb86cdSEric Blake /* Currently Linux does AIO only for files opened with O_DIRECT */ 518c1bb86cdSEric Blake if (s->use_linux_aio && !(s->open_flags & O_DIRECT)) { 519c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but it requires " 520c1bb86cdSEric Blake "cache.direct=on, which was not specified."); 521c1bb86cdSEric Blake ret = -EINVAL; 522c1bb86cdSEric Blake goto fail; 523c1bb86cdSEric Blake } 524c1bb86cdSEric Blake #else 525c1bb86cdSEric Blake if (s->use_linux_aio) { 526c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but is not supported " 527c1bb86cdSEric Blake "in this build."); 528c1bb86cdSEric Blake ret = -EINVAL; 529c1bb86cdSEric Blake goto fail; 530c1bb86cdSEric Blake } 531c1bb86cdSEric Blake #endif /* !defined(CONFIG_LINUX_AIO) */ 532c1bb86cdSEric Blake 533c1bb86cdSEric Blake s->has_discard = true; 534c1bb86cdSEric Blake s->has_write_zeroes = true; 535c1bb86cdSEric Blake bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP; 536c1bb86cdSEric Blake if ((bs->open_flags & BDRV_O_NOCACHE) != 0) { 537c1bb86cdSEric Blake s->needs_alignment = true; 538c1bb86cdSEric Blake } 539c1bb86cdSEric Blake 540c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 541c1bb86cdSEric Blake ret = -errno; 542c1bb86cdSEric Blake error_setg_errno(errp, errno, "Could not stat file"); 543c1bb86cdSEric Blake goto fail; 544c1bb86cdSEric Blake } 545c1bb86cdSEric Blake if (S_ISREG(st.st_mode)) { 546c1bb86cdSEric Blake s->discard_zeroes = true; 547c1bb86cdSEric Blake s->has_fallocate = true; 548c1bb86cdSEric Blake } 549c1bb86cdSEric Blake if (S_ISBLK(st.st_mode)) { 550c1bb86cdSEric Blake #ifdef BLKDISCARDZEROES 551c1bb86cdSEric Blake unsigned int arg; 552c1bb86cdSEric Blake if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { 553c1bb86cdSEric Blake s->discard_zeroes = true; 554c1bb86cdSEric Blake } 555c1bb86cdSEric Blake #endif 556c1bb86cdSEric Blake #ifdef __linux__ 557c1bb86cdSEric Blake /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do 558c1bb86cdSEric Blake * not rely on the contents of discarded blocks unless using O_DIRECT. 559c1bb86cdSEric Blake * Same for BLKZEROOUT. 560c1bb86cdSEric Blake */ 561c1bb86cdSEric Blake if (!(bs->open_flags & BDRV_O_NOCACHE)) { 562c1bb86cdSEric Blake s->discard_zeroes = false; 563c1bb86cdSEric Blake s->has_write_zeroes = false; 564c1bb86cdSEric Blake } 565c1bb86cdSEric Blake #endif 566c1bb86cdSEric Blake } 567c1bb86cdSEric Blake #ifdef __FreeBSD__ 568c1bb86cdSEric Blake if (S_ISCHR(st.st_mode)) { 569c1bb86cdSEric Blake /* 570c1bb86cdSEric Blake * The file is a char device (disk), which on FreeBSD isn't behind 571c1bb86cdSEric Blake * a pager, so force all requests to be aligned. This is needed 572c1bb86cdSEric Blake * so QEMU makes sure all IO operations on the device are aligned 573c1bb86cdSEric Blake * to sector size, or else FreeBSD will reject them with EINVAL. 574c1bb86cdSEric Blake */ 575c1bb86cdSEric Blake s->needs_alignment = true; 576c1bb86cdSEric Blake } 577c1bb86cdSEric Blake #endif 578c1bb86cdSEric Blake 579c1bb86cdSEric Blake #ifdef CONFIG_XFS 580c1bb86cdSEric Blake if (platform_test_xfs_fd(s->fd)) { 581c1bb86cdSEric Blake s->is_xfs = true; 582c1bb86cdSEric Blake } 583c1bb86cdSEric Blake #endif 584c1bb86cdSEric Blake 585c1bb86cdSEric Blake ret = 0; 586c1bb86cdSEric Blake fail: 587c1bb86cdSEric Blake if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { 588c1bb86cdSEric Blake unlink(filename); 589c1bb86cdSEric Blake } 590c1bb86cdSEric Blake qemu_opts_del(opts); 591c1bb86cdSEric Blake return ret; 592c1bb86cdSEric Blake } 593c1bb86cdSEric Blake 594c1bb86cdSEric Blake static int raw_open(BlockDriverState *bs, QDict *options, int flags, 595c1bb86cdSEric Blake Error **errp) 596c1bb86cdSEric Blake { 597c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 598c1bb86cdSEric Blake 599c1bb86cdSEric Blake s->type = FTYPE_FILE; 600c1bb86cdSEric Blake return raw_open_common(bs, options, flags, 0, errp); 601c1bb86cdSEric Blake } 602c1bb86cdSEric Blake 603*244a5668SFam Zheng typedef enum { 604*244a5668SFam Zheng RAW_PL_PREPARE, 605*244a5668SFam Zheng RAW_PL_COMMIT, 606*244a5668SFam Zheng RAW_PL_ABORT, 607*244a5668SFam Zheng } RawPermLockOp; 608*244a5668SFam Zheng 609*244a5668SFam Zheng #define PERM_FOREACH(i) \ 610*244a5668SFam Zheng for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) 611*244a5668SFam Zheng 612*244a5668SFam Zheng /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the 613*244a5668SFam Zheng * file; if @unlock == true, also unlock the unneeded bytes. 614*244a5668SFam Zheng * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. 615*244a5668SFam Zheng */ 616*244a5668SFam Zheng static int raw_apply_lock_bytes(BDRVRawState *s, 617*244a5668SFam Zheng uint64_t perm_lock_bits, 618*244a5668SFam Zheng uint64_t shared_perm_lock_bits, 619*244a5668SFam Zheng bool unlock, Error **errp) 620*244a5668SFam Zheng { 621*244a5668SFam Zheng int ret; 622*244a5668SFam Zheng int i; 623*244a5668SFam Zheng 624*244a5668SFam Zheng PERM_FOREACH(i) { 625*244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 626*244a5668SFam Zheng if (perm_lock_bits & (1ULL << i)) { 627*244a5668SFam Zheng ret = qemu_lock_fd(s->lock_fd, off, 1, false); 628*244a5668SFam Zheng if (ret) { 629*244a5668SFam Zheng error_setg(errp, "Failed to lock byte %d", off); 630*244a5668SFam Zheng return ret; 631*244a5668SFam Zheng } 632*244a5668SFam Zheng } else if (unlock) { 633*244a5668SFam Zheng ret = qemu_unlock_fd(s->lock_fd, off, 1); 634*244a5668SFam Zheng if (ret) { 635*244a5668SFam Zheng error_setg(errp, "Failed to unlock byte %d", off); 636*244a5668SFam Zheng return ret; 637*244a5668SFam Zheng } 638*244a5668SFam Zheng } 639*244a5668SFam Zheng } 640*244a5668SFam Zheng PERM_FOREACH(i) { 641*244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 642*244a5668SFam Zheng if (shared_perm_lock_bits & (1ULL << i)) { 643*244a5668SFam Zheng ret = qemu_lock_fd(s->lock_fd, off, 1, false); 644*244a5668SFam Zheng if (ret) { 645*244a5668SFam Zheng error_setg(errp, "Failed to lock byte %d", off); 646*244a5668SFam Zheng return ret; 647*244a5668SFam Zheng } 648*244a5668SFam Zheng } else if (unlock) { 649*244a5668SFam Zheng ret = qemu_unlock_fd(s->lock_fd, off, 1); 650*244a5668SFam Zheng if (ret) { 651*244a5668SFam Zheng error_setg(errp, "Failed to unlock byte %d", off); 652*244a5668SFam Zheng return ret; 653*244a5668SFam Zheng } 654*244a5668SFam Zheng } 655*244a5668SFam Zheng } 656*244a5668SFam Zheng return 0; 657*244a5668SFam Zheng } 658*244a5668SFam Zheng 659*244a5668SFam Zheng /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ 660*244a5668SFam Zheng static int raw_check_lock_bytes(BDRVRawState *s, 661*244a5668SFam Zheng uint64_t perm, uint64_t shared_perm, 662*244a5668SFam Zheng Error **errp) 663*244a5668SFam Zheng { 664*244a5668SFam Zheng int ret; 665*244a5668SFam Zheng int i; 666*244a5668SFam Zheng 667*244a5668SFam Zheng PERM_FOREACH(i) { 668*244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 669*244a5668SFam Zheng uint64_t p = 1ULL << i; 670*244a5668SFam Zheng if (perm & p) { 671*244a5668SFam Zheng ret = qemu_lock_fd_test(s->lock_fd, off, 1, true); 672*244a5668SFam Zheng if (ret) { 673*244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 674*244a5668SFam Zheng error_setg(errp, 675*244a5668SFam Zheng "Failed to get \"%s\" lock", 676*244a5668SFam Zheng perm_name); 677*244a5668SFam Zheng g_free(perm_name); 678*244a5668SFam Zheng error_append_hint(errp, 679*244a5668SFam Zheng "Is another process using the image?\n"); 680*244a5668SFam Zheng return ret; 681*244a5668SFam Zheng } 682*244a5668SFam Zheng } 683*244a5668SFam Zheng } 684*244a5668SFam Zheng PERM_FOREACH(i) { 685*244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 686*244a5668SFam Zheng uint64_t p = 1ULL << i; 687*244a5668SFam Zheng if (!(shared_perm & p)) { 688*244a5668SFam Zheng ret = qemu_lock_fd_test(s->lock_fd, off, 1, true); 689*244a5668SFam Zheng if (ret) { 690*244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 691*244a5668SFam Zheng error_setg(errp, 692*244a5668SFam Zheng "Failed to get shared \"%s\" lock", 693*244a5668SFam Zheng perm_name); 694*244a5668SFam Zheng g_free(perm_name); 695*244a5668SFam Zheng error_append_hint(errp, 696*244a5668SFam Zheng "Is another process using the image?\n"); 697*244a5668SFam Zheng return ret; 698*244a5668SFam Zheng } 699*244a5668SFam Zheng } 700*244a5668SFam Zheng } 701*244a5668SFam Zheng return 0; 702*244a5668SFam Zheng } 703*244a5668SFam Zheng 704*244a5668SFam Zheng static int raw_handle_perm_lock(BlockDriverState *bs, 705*244a5668SFam Zheng RawPermLockOp op, 706*244a5668SFam Zheng uint64_t new_perm, uint64_t new_shared, 707*244a5668SFam Zheng Error **errp) 708*244a5668SFam Zheng { 709*244a5668SFam Zheng BDRVRawState *s = bs->opaque; 710*244a5668SFam Zheng int ret = 0; 711*244a5668SFam Zheng Error *local_err = NULL; 712*244a5668SFam Zheng 713*244a5668SFam Zheng if (!s->use_lock) { 714*244a5668SFam Zheng return 0; 715*244a5668SFam Zheng } 716*244a5668SFam Zheng 717*244a5668SFam Zheng if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { 718*244a5668SFam Zheng return 0; 719*244a5668SFam Zheng } 720*244a5668SFam Zheng 721*244a5668SFam Zheng assert(s->lock_fd > 0); 722*244a5668SFam Zheng 723*244a5668SFam Zheng switch (op) { 724*244a5668SFam Zheng case RAW_PL_PREPARE: 725*244a5668SFam Zheng ret = raw_apply_lock_bytes(s, s->perm | new_perm, 726*244a5668SFam Zheng ~s->shared_perm | ~new_shared, 727*244a5668SFam Zheng false, errp); 728*244a5668SFam Zheng if (!ret) { 729*244a5668SFam Zheng ret = raw_check_lock_bytes(s, new_perm, new_shared, errp); 730*244a5668SFam Zheng if (!ret) { 731*244a5668SFam Zheng return 0; 732*244a5668SFam Zheng } 733*244a5668SFam Zheng } 734*244a5668SFam Zheng op = RAW_PL_ABORT; 735*244a5668SFam Zheng /* fall through to unlock bytes. */ 736*244a5668SFam Zheng case RAW_PL_ABORT: 737*244a5668SFam Zheng raw_apply_lock_bytes(s, s->perm, ~s->shared_perm, true, &local_err); 738*244a5668SFam Zheng if (local_err) { 739*244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 740*244a5668SFam Zheng * fail. Something weird happened, report it. 741*244a5668SFam Zheng */ 742*244a5668SFam Zheng error_report_err(local_err); 743*244a5668SFam Zheng } 744*244a5668SFam Zheng break; 745*244a5668SFam Zheng case RAW_PL_COMMIT: 746*244a5668SFam Zheng raw_apply_lock_bytes(s, new_perm, ~new_shared, true, &local_err); 747*244a5668SFam Zheng if (local_err) { 748*244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 749*244a5668SFam Zheng * fail. Something weird happened, report it. 750*244a5668SFam Zheng */ 751*244a5668SFam Zheng error_report_err(local_err); 752*244a5668SFam Zheng } 753*244a5668SFam Zheng break; 754*244a5668SFam Zheng } 755*244a5668SFam Zheng return ret; 756*244a5668SFam Zheng } 757*244a5668SFam Zheng 758c1bb86cdSEric Blake static int raw_reopen_prepare(BDRVReopenState *state, 759c1bb86cdSEric Blake BlockReopenQueue *queue, Error **errp) 760c1bb86cdSEric Blake { 761c1bb86cdSEric Blake BDRVRawState *s; 762c1bb86cdSEric Blake BDRVRawReopenState *rs; 763c1bb86cdSEric Blake int ret = 0; 764c1bb86cdSEric Blake Error *local_err = NULL; 765c1bb86cdSEric Blake 766c1bb86cdSEric Blake assert(state != NULL); 767c1bb86cdSEric Blake assert(state->bs != NULL); 768c1bb86cdSEric Blake 769c1bb86cdSEric Blake s = state->bs->opaque; 770c1bb86cdSEric Blake 771c1bb86cdSEric Blake state->opaque = g_new0(BDRVRawReopenState, 1); 772c1bb86cdSEric Blake rs = state->opaque; 773c1bb86cdSEric Blake 774c1bb86cdSEric Blake if (s->type == FTYPE_CD) { 775c1bb86cdSEric Blake rs->open_flags |= O_NONBLOCK; 776c1bb86cdSEric Blake } 777c1bb86cdSEric Blake 778c1bb86cdSEric Blake raw_parse_flags(state->flags, &rs->open_flags); 779c1bb86cdSEric Blake 780c1bb86cdSEric Blake rs->fd = -1; 781c1bb86cdSEric Blake 782c1bb86cdSEric Blake int fcntl_flags = O_APPEND | O_NONBLOCK; 783c1bb86cdSEric Blake #ifdef O_NOATIME 784c1bb86cdSEric Blake fcntl_flags |= O_NOATIME; 785c1bb86cdSEric Blake #endif 786c1bb86cdSEric Blake 787c1bb86cdSEric Blake #ifdef O_ASYNC 788c1bb86cdSEric Blake /* Not all operating systems have O_ASYNC, and those that don't 789c1bb86cdSEric Blake * will not let us track the state into rs->open_flags (typically 790c1bb86cdSEric Blake * you achieve the same effect with an ioctl, for example I_SETSIG 791c1bb86cdSEric Blake * on Solaris). But we do not use O_ASYNC, so that's fine. 792c1bb86cdSEric Blake */ 793c1bb86cdSEric Blake assert((s->open_flags & O_ASYNC) == 0); 794c1bb86cdSEric Blake #endif 795c1bb86cdSEric Blake 796c1bb86cdSEric Blake if ((rs->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { 797c1bb86cdSEric Blake /* dup the original fd */ 798c1bb86cdSEric Blake rs->fd = qemu_dup(s->fd); 799c1bb86cdSEric Blake if (rs->fd >= 0) { 800c1bb86cdSEric Blake ret = fcntl_setfl(rs->fd, rs->open_flags); 801c1bb86cdSEric Blake if (ret) { 802c1bb86cdSEric Blake qemu_close(rs->fd); 803c1bb86cdSEric Blake rs->fd = -1; 804c1bb86cdSEric Blake } 805c1bb86cdSEric Blake } 806c1bb86cdSEric Blake } 807c1bb86cdSEric Blake 808c1bb86cdSEric Blake /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ 809c1bb86cdSEric Blake if (rs->fd == -1) { 810c1bb86cdSEric Blake const char *normalized_filename = state->bs->filename; 811c1bb86cdSEric Blake ret = raw_normalize_devicepath(&normalized_filename); 812c1bb86cdSEric Blake if (ret < 0) { 813c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not normalize device path"); 814c1bb86cdSEric Blake } else { 815c1bb86cdSEric Blake assert(!(rs->open_flags & O_CREAT)); 816c1bb86cdSEric Blake rs->fd = qemu_open(normalized_filename, rs->open_flags); 817c1bb86cdSEric Blake if (rs->fd == -1) { 818c1bb86cdSEric Blake error_setg_errno(errp, errno, "Could not reopen file"); 819c1bb86cdSEric Blake ret = -1; 820c1bb86cdSEric Blake } 821c1bb86cdSEric Blake } 822c1bb86cdSEric Blake } 823c1bb86cdSEric Blake 824c1bb86cdSEric Blake /* Fail already reopen_prepare() if we can't get a working O_DIRECT 825c1bb86cdSEric Blake * alignment with the new fd. */ 826c1bb86cdSEric Blake if (rs->fd != -1) { 827c1bb86cdSEric Blake raw_probe_alignment(state->bs, rs->fd, &local_err); 828c1bb86cdSEric Blake if (local_err) { 829c1bb86cdSEric Blake qemu_close(rs->fd); 830c1bb86cdSEric Blake rs->fd = -1; 831c1bb86cdSEric Blake error_propagate(errp, local_err); 832c1bb86cdSEric Blake ret = -EINVAL; 833c1bb86cdSEric Blake } 834c1bb86cdSEric Blake } 835c1bb86cdSEric Blake 836c1bb86cdSEric Blake return ret; 837c1bb86cdSEric Blake } 838c1bb86cdSEric Blake 839c1bb86cdSEric Blake static void raw_reopen_commit(BDRVReopenState *state) 840c1bb86cdSEric Blake { 841c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 842c1bb86cdSEric Blake BDRVRawState *s = state->bs->opaque; 843c1bb86cdSEric Blake 844c1bb86cdSEric Blake s->open_flags = rs->open_flags; 845c1bb86cdSEric Blake 846c1bb86cdSEric Blake qemu_close(s->fd); 847c1bb86cdSEric Blake s->fd = rs->fd; 848c1bb86cdSEric Blake 849c1bb86cdSEric Blake g_free(state->opaque); 850c1bb86cdSEric Blake state->opaque = NULL; 851c1bb86cdSEric Blake } 852c1bb86cdSEric Blake 853c1bb86cdSEric Blake 854c1bb86cdSEric Blake static void raw_reopen_abort(BDRVReopenState *state) 855c1bb86cdSEric Blake { 856c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 857c1bb86cdSEric Blake 858c1bb86cdSEric Blake /* nothing to do if NULL, we didn't get far enough */ 859c1bb86cdSEric Blake if (rs == NULL) { 860c1bb86cdSEric Blake return; 861c1bb86cdSEric Blake } 862c1bb86cdSEric Blake 863c1bb86cdSEric Blake if (rs->fd >= 0) { 864c1bb86cdSEric Blake qemu_close(rs->fd); 865c1bb86cdSEric Blake rs->fd = -1; 866c1bb86cdSEric Blake } 867c1bb86cdSEric Blake g_free(state->opaque); 868c1bb86cdSEric Blake state->opaque = NULL; 869c1bb86cdSEric Blake } 870c1bb86cdSEric Blake 87148265250SEric Farman static int hdev_get_max_transfer_length(BlockDriverState *bs, int fd) 872c1bb86cdSEric Blake { 873c1bb86cdSEric Blake #ifdef BLKSECTGET 87448265250SEric Farman int max_bytes = 0; 87548265250SEric Farman short max_sectors = 0; 87648265250SEric Farman if (bs->sg && ioctl(fd, BLKSECTGET, &max_bytes) == 0) { 87748265250SEric Farman return max_bytes; 87848265250SEric Farman } else if (!bs->sg && ioctl(fd, BLKSECTGET, &max_sectors) == 0) { 87948265250SEric Farman return max_sectors << BDRV_SECTOR_BITS; 880c1bb86cdSEric Blake } else { 881c1bb86cdSEric Blake return -errno; 882c1bb86cdSEric Blake } 883c1bb86cdSEric Blake #else 884c1bb86cdSEric Blake return -ENOSYS; 885c1bb86cdSEric Blake #endif 886c1bb86cdSEric Blake } 887c1bb86cdSEric Blake 8889103f1ceSFam Zheng static int hdev_get_max_segments(const struct stat *st) 8899103f1ceSFam Zheng { 8909103f1ceSFam Zheng #ifdef CONFIG_LINUX 8919103f1ceSFam Zheng char buf[32]; 8929103f1ceSFam Zheng const char *end; 8939103f1ceSFam Zheng char *sysfspath; 8949103f1ceSFam Zheng int ret; 8959103f1ceSFam Zheng int fd = -1; 8969103f1ceSFam Zheng long max_segments; 8979103f1ceSFam Zheng 8989103f1ceSFam Zheng sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", 8999103f1ceSFam Zheng major(st->st_rdev), minor(st->st_rdev)); 9009103f1ceSFam Zheng fd = open(sysfspath, O_RDONLY); 9019103f1ceSFam Zheng if (fd == -1) { 9029103f1ceSFam Zheng ret = -errno; 9039103f1ceSFam Zheng goto out; 9049103f1ceSFam Zheng } 9059103f1ceSFam Zheng do { 90669583490SStefan Hajnoczi ret = read(fd, buf, sizeof(buf) - 1); 9079103f1ceSFam Zheng } while (ret == -1 && errno == EINTR); 9089103f1ceSFam Zheng if (ret < 0) { 9099103f1ceSFam Zheng ret = -errno; 9109103f1ceSFam Zheng goto out; 9119103f1ceSFam Zheng } else if (ret == 0) { 9129103f1ceSFam Zheng ret = -EIO; 9139103f1ceSFam Zheng goto out; 9149103f1ceSFam Zheng } 9159103f1ceSFam Zheng buf[ret] = 0; 9169103f1ceSFam Zheng /* The file is ended with '\n', pass 'end' to accept that. */ 9179103f1ceSFam Zheng ret = qemu_strtol(buf, &end, 10, &max_segments); 9189103f1ceSFam Zheng if (ret == 0 && end && *end == '\n') { 9199103f1ceSFam Zheng ret = max_segments; 9209103f1ceSFam Zheng } 9219103f1ceSFam Zheng 9229103f1ceSFam Zheng out: 923fed414dfSFam Zheng if (fd != -1) { 924fed414dfSFam Zheng close(fd); 925fed414dfSFam Zheng } 9269103f1ceSFam Zheng g_free(sysfspath); 9279103f1ceSFam Zheng return ret; 9289103f1ceSFam Zheng #else 9299103f1ceSFam Zheng return -ENOTSUP; 9309103f1ceSFam Zheng #endif 9319103f1ceSFam Zheng } 9329103f1ceSFam Zheng 933c1bb86cdSEric Blake static void raw_refresh_limits(BlockDriverState *bs, Error **errp) 934c1bb86cdSEric Blake { 935c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 936c1bb86cdSEric Blake struct stat st; 937c1bb86cdSEric Blake 938c1bb86cdSEric Blake if (!fstat(s->fd, &st)) { 939c4c41a0aSEric Farman if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode)) { 94048265250SEric Farman int ret = hdev_get_max_transfer_length(bs, s->fd); 94148265250SEric Farman if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { 94248265250SEric Farman bs->bl.max_transfer = pow2floor(ret); 943c1bb86cdSEric Blake } 9449103f1ceSFam Zheng ret = hdev_get_max_segments(&st); 9459103f1ceSFam Zheng if (ret > 0) { 9469103f1ceSFam Zheng bs->bl.max_transfer = MIN(bs->bl.max_transfer, 9479103f1ceSFam Zheng ret * getpagesize()); 9489103f1ceSFam Zheng } 949c1bb86cdSEric Blake } 950c1bb86cdSEric Blake } 951c1bb86cdSEric Blake 952c1bb86cdSEric Blake raw_probe_alignment(bs, s->fd, errp); 953c1bb86cdSEric Blake bs->bl.min_mem_alignment = s->buf_align; 954c1bb86cdSEric Blake bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize()); 955c1bb86cdSEric Blake } 956c1bb86cdSEric Blake 957c1bb86cdSEric Blake static int check_for_dasd(int fd) 958c1bb86cdSEric Blake { 959c1bb86cdSEric Blake #ifdef BIODASDINFO2 960c1bb86cdSEric Blake struct dasd_information2_t info = {0}; 961c1bb86cdSEric Blake 962c1bb86cdSEric Blake return ioctl(fd, BIODASDINFO2, &info); 963c1bb86cdSEric Blake #else 964c1bb86cdSEric Blake return -1; 965c1bb86cdSEric Blake #endif 966c1bb86cdSEric Blake } 967c1bb86cdSEric Blake 968c1bb86cdSEric Blake /** 969c1bb86cdSEric Blake * Try to get @bs's logical and physical block size. 970c1bb86cdSEric Blake * On success, store them in @bsz and return zero. 971c1bb86cdSEric Blake * On failure, return negative errno. 972c1bb86cdSEric Blake */ 973c1bb86cdSEric Blake static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) 974c1bb86cdSEric Blake { 975c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 976c1bb86cdSEric Blake int ret; 977c1bb86cdSEric Blake 978c1bb86cdSEric Blake /* If DASD, get blocksizes */ 979c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 980c1bb86cdSEric Blake return -ENOTSUP; 981c1bb86cdSEric Blake } 982c1bb86cdSEric Blake ret = probe_logical_blocksize(s->fd, &bsz->log); 983c1bb86cdSEric Blake if (ret < 0) { 984c1bb86cdSEric Blake return ret; 985c1bb86cdSEric Blake } 986c1bb86cdSEric Blake return probe_physical_blocksize(s->fd, &bsz->phys); 987c1bb86cdSEric Blake } 988c1bb86cdSEric Blake 989c1bb86cdSEric Blake /** 990c1bb86cdSEric Blake * Try to get @bs's geometry: cyls, heads, sectors. 991c1bb86cdSEric Blake * On success, store them in @geo and return 0. 992c1bb86cdSEric Blake * On failure return -errno. 993c1bb86cdSEric Blake * (Allows block driver to assign default geometry values that guest sees) 994c1bb86cdSEric Blake */ 995c1bb86cdSEric Blake #ifdef __linux__ 996c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 997c1bb86cdSEric Blake { 998c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 999c1bb86cdSEric Blake struct hd_geometry ioctl_geo = {0}; 1000c1bb86cdSEric Blake 1001c1bb86cdSEric Blake /* If DASD, get its geometry */ 1002c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1003c1bb86cdSEric Blake return -ENOTSUP; 1004c1bb86cdSEric Blake } 1005c1bb86cdSEric Blake if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { 1006c1bb86cdSEric Blake return -errno; 1007c1bb86cdSEric Blake } 1008c1bb86cdSEric Blake /* HDIO_GETGEO may return success even though geo contains zeros 1009c1bb86cdSEric Blake (e.g. certain multipath setups) */ 1010c1bb86cdSEric Blake if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { 1011c1bb86cdSEric Blake return -ENOTSUP; 1012c1bb86cdSEric Blake } 1013c1bb86cdSEric Blake /* Do not return a geometry for partition */ 1014c1bb86cdSEric Blake if (ioctl_geo.start != 0) { 1015c1bb86cdSEric Blake return -ENOTSUP; 1016c1bb86cdSEric Blake } 1017c1bb86cdSEric Blake geo->heads = ioctl_geo.heads; 1018c1bb86cdSEric Blake geo->sectors = ioctl_geo.sectors; 1019c1bb86cdSEric Blake geo->cylinders = ioctl_geo.cylinders; 1020c1bb86cdSEric Blake 1021c1bb86cdSEric Blake return 0; 1022c1bb86cdSEric Blake } 1023c1bb86cdSEric Blake #else /* __linux__ */ 1024c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1025c1bb86cdSEric Blake { 1026c1bb86cdSEric Blake return -ENOTSUP; 1027c1bb86cdSEric Blake } 1028c1bb86cdSEric Blake #endif 1029c1bb86cdSEric Blake 1030c1bb86cdSEric Blake static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb) 1031c1bb86cdSEric Blake { 1032c1bb86cdSEric Blake int ret; 1033c1bb86cdSEric Blake 1034c1bb86cdSEric Blake ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf); 1035c1bb86cdSEric Blake if (ret == -1) { 1036c1bb86cdSEric Blake return -errno; 1037c1bb86cdSEric Blake } 1038c1bb86cdSEric Blake 1039c1bb86cdSEric Blake return 0; 1040c1bb86cdSEric Blake } 1041c1bb86cdSEric Blake 1042c1bb86cdSEric Blake static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb) 1043c1bb86cdSEric Blake { 1044e5bcf967SKevin Wolf BDRVRawState *s = aiocb->bs->opaque; 1045c1bb86cdSEric Blake int ret; 1046c1bb86cdSEric Blake 1047e5bcf967SKevin Wolf if (s->page_cache_inconsistent) { 1048e5bcf967SKevin Wolf return -EIO; 1049e5bcf967SKevin Wolf } 1050e5bcf967SKevin Wolf 1051c1bb86cdSEric Blake ret = qemu_fdatasync(aiocb->aio_fildes); 1052c1bb86cdSEric Blake if (ret == -1) { 1053e5bcf967SKevin Wolf /* There is no clear definition of the semantics of a failing fsync(), 1054e5bcf967SKevin Wolf * so we may have to assume the worst. The sad truth is that this 1055e5bcf967SKevin Wolf * assumption is correct for Linux. Some pages are now probably marked 1056e5bcf967SKevin Wolf * clean in the page cache even though they are inconsistent with the 1057e5bcf967SKevin Wolf * on-disk contents. The next fdatasync() call would succeed, but no 1058e5bcf967SKevin Wolf * further writeback attempt will be made. We can't get back to a state 1059e5bcf967SKevin Wolf * in which we know what is on disk (we would have to rewrite 1060e5bcf967SKevin Wolf * everything that was touched since the last fdatasync() at least), so 1061e5bcf967SKevin Wolf * make bdrv_flush() fail permanently. Given that the behaviour isn't 1062e5bcf967SKevin Wolf * really defined, I have little hope that other OSes are doing better. 1063e5bcf967SKevin Wolf * 1064e5bcf967SKevin Wolf * Obviously, this doesn't affect O_DIRECT, which bypasses the page 1065e5bcf967SKevin Wolf * cache. */ 1066e5bcf967SKevin Wolf if ((s->open_flags & O_DIRECT) == 0) { 1067e5bcf967SKevin Wolf s->page_cache_inconsistent = true; 1068e5bcf967SKevin Wolf } 1069c1bb86cdSEric Blake return -errno; 1070c1bb86cdSEric Blake } 1071c1bb86cdSEric Blake return 0; 1072c1bb86cdSEric Blake } 1073c1bb86cdSEric Blake 1074c1bb86cdSEric Blake #ifdef CONFIG_PREADV 1075c1bb86cdSEric Blake 1076c1bb86cdSEric Blake static bool preadv_present = true; 1077c1bb86cdSEric Blake 1078c1bb86cdSEric Blake static ssize_t 1079c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1080c1bb86cdSEric Blake { 1081c1bb86cdSEric Blake return preadv(fd, iov, nr_iov, offset); 1082c1bb86cdSEric Blake } 1083c1bb86cdSEric Blake 1084c1bb86cdSEric Blake static ssize_t 1085c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1086c1bb86cdSEric Blake { 1087c1bb86cdSEric Blake return pwritev(fd, iov, nr_iov, offset); 1088c1bb86cdSEric Blake } 1089c1bb86cdSEric Blake 1090c1bb86cdSEric Blake #else 1091c1bb86cdSEric Blake 1092c1bb86cdSEric Blake static bool preadv_present = false; 1093c1bb86cdSEric Blake 1094c1bb86cdSEric Blake static ssize_t 1095c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1096c1bb86cdSEric Blake { 1097c1bb86cdSEric Blake return -ENOSYS; 1098c1bb86cdSEric Blake } 1099c1bb86cdSEric Blake 1100c1bb86cdSEric Blake static ssize_t 1101c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1102c1bb86cdSEric Blake { 1103c1bb86cdSEric Blake return -ENOSYS; 1104c1bb86cdSEric Blake } 1105c1bb86cdSEric Blake 1106c1bb86cdSEric Blake #endif 1107c1bb86cdSEric Blake 1108c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) 1109c1bb86cdSEric Blake { 1110c1bb86cdSEric Blake ssize_t len; 1111c1bb86cdSEric Blake 1112c1bb86cdSEric Blake do { 1113c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) 1114c1bb86cdSEric Blake len = qemu_pwritev(aiocb->aio_fildes, 1115c1bb86cdSEric Blake aiocb->aio_iov, 1116c1bb86cdSEric Blake aiocb->aio_niov, 1117c1bb86cdSEric Blake aiocb->aio_offset); 1118c1bb86cdSEric Blake else 1119c1bb86cdSEric Blake len = qemu_preadv(aiocb->aio_fildes, 1120c1bb86cdSEric Blake aiocb->aio_iov, 1121c1bb86cdSEric Blake aiocb->aio_niov, 1122c1bb86cdSEric Blake aiocb->aio_offset); 1123c1bb86cdSEric Blake } while (len == -1 && errno == EINTR); 1124c1bb86cdSEric Blake 1125c1bb86cdSEric Blake if (len == -1) { 1126c1bb86cdSEric Blake return -errno; 1127c1bb86cdSEric Blake } 1128c1bb86cdSEric Blake return len; 1129c1bb86cdSEric Blake } 1130c1bb86cdSEric Blake 1131c1bb86cdSEric Blake /* 1132c1bb86cdSEric Blake * Read/writes the data to/from a given linear buffer. 1133c1bb86cdSEric Blake * 1134c1bb86cdSEric Blake * Returns the number of bytes handles or -errno in case of an error. Short 1135c1bb86cdSEric Blake * reads are only returned if the end of the file is reached. 1136c1bb86cdSEric Blake */ 1137c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) 1138c1bb86cdSEric Blake { 1139c1bb86cdSEric Blake ssize_t offset = 0; 1140c1bb86cdSEric Blake ssize_t len; 1141c1bb86cdSEric Blake 1142c1bb86cdSEric Blake while (offset < aiocb->aio_nbytes) { 1143c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1144c1bb86cdSEric Blake len = pwrite(aiocb->aio_fildes, 1145c1bb86cdSEric Blake (const char *)buf + offset, 1146c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1147c1bb86cdSEric Blake aiocb->aio_offset + offset); 1148c1bb86cdSEric Blake } else { 1149c1bb86cdSEric Blake len = pread(aiocb->aio_fildes, 1150c1bb86cdSEric Blake buf + offset, 1151c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1152c1bb86cdSEric Blake aiocb->aio_offset + offset); 1153c1bb86cdSEric Blake } 1154c1bb86cdSEric Blake if (len == -1 && errno == EINTR) { 1155c1bb86cdSEric Blake continue; 1156c1bb86cdSEric Blake } else if (len == -1 && errno == EINVAL && 1157c1bb86cdSEric Blake (aiocb->bs->open_flags & BDRV_O_NOCACHE) && 1158c1bb86cdSEric Blake !(aiocb->aio_type & QEMU_AIO_WRITE) && 1159c1bb86cdSEric Blake offset > 0) { 1160c1bb86cdSEric Blake /* O_DIRECT pread() may fail with EINVAL when offset is unaligned 1161c1bb86cdSEric Blake * after a short read. Assume that O_DIRECT short reads only occur 1162c1bb86cdSEric Blake * at EOF. Therefore this is a short read, not an I/O error. 1163c1bb86cdSEric Blake */ 1164c1bb86cdSEric Blake break; 1165c1bb86cdSEric Blake } else if (len == -1) { 1166c1bb86cdSEric Blake offset = -errno; 1167c1bb86cdSEric Blake break; 1168c1bb86cdSEric Blake } else if (len == 0) { 1169c1bb86cdSEric Blake break; 1170c1bb86cdSEric Blake } 1171c1bb86cdSEric Blake offset += len; 1172c1bb86cdSEric Blake } 1173c1bb86cdSEric Blake 1174c1bb86cdSEric Blake return offset; 1175c1bb86cdSEric Blake } 1176c1bb86cdSEric Blake 1177c1bb86cdSEric Blake static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb) 1178c1bb86cdSEric Blake { 1179c1bb86cdSEric Blake ssize_t nbytes; 1180c1bb86cdSEric Blake char *buf; 1181c1bb86cdSEric Blake 1182c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { 1183c1bb86cdSEric Blake /* 1184c1bb86cdSEric Blake * If there is just a single buffer, and it is properly aligned 1185c1bb86cdSEric Blake * we can just use plain pread/pwrite without any problems. 1186c1bb86cdSEric Blake */ 1187c1bb86cdSEric Blake if (aiocb->aio_niov == 1) { 1188c1bb86cdSEric Blake return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base); 1189c1bb86cdSEric Blake } 1190c1bb86cdSEric Blake /* 1191c1bb86cdSEric Blake * We have more than one iovec, and all are properly aligned. 1192c1bb86cdSEric Blake * 1193c1bb86cdSEric Blake * Try preadv/pwritev first and fall back to linearizing the 1194c1bb86cdSEric Blake * buffer if it's not supported. 1195c1bb86cdSEric Blake */ 1196c1bb86cdSEric Blake if (preadv_present) { 1197c1bb86cdSEric Blake nbytes = handle_aiocb_rw_vector(aiocb); 1198c1bb86cdSEric Blake if (nbytes == aiocb->aio_nbytes || 1199c1bb86cdSEric Blake (nbytes < 0 && nbytes != -ENOSYS)) { 1200c1bb86cdSEric Blake return nbytes; 1201c1bb86cdSEric Blake } 1202c1bb86cdSEric Blake preadv_present = false; 1203c1bb86cdSEric Blake } 1204c1bb86cdSEric Blake 1205c1bb86cdSEric Blake /* 1206c1bb86cdSEric Blake * XXX(hch): short read/write. no easy way to handle the reminder 1207c1bb86cdSEric Blake * using these interfaces. For now retry using plain 1208c1bb86cdSEric Blake * pread/pwrite? 1209c1bb86cdSEric Blake */ 1210c1bb86cdSEric Blake } 1211c1bb86cdSEric Blake 1212c1bb86cdSEric Blake /* 1213c1bb86cdSEric Blake * Ok, we have to do it the hard way, copy all segments into 1214c1bb86cdSEric Blake * a single aligned buffer. 1215c1bb86cdSEric Blake */ 1216c1bb86cdSEric Blake buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); 1217c1bb86cdSEric Blake if (buf == NULL) { 1218c1bb86cdSEric Blake return -ENOMEM; 1219c1bb86cdSEric Blake } 1220c1bb86cdSEric Blake 1221c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1222c1bb86cdSEric Blake char *p = buf; 1223c1bb86cdSEric Blake int i; 1224c1bb86cdSEric Blake 1225c1bb86cdSEric Blake for (i = 0; i < aiocb->aio_niov; ++i) { 1226c1bb86cdSEric Blake memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len); 1227c1bb86cdSEric Blake p += aiocb->aio_iov[i].iov_len; 1228c1bb86cdSEric Blake } 1229c1bb86cdSEric Blake assert(p - buf == aiocb->aio_nbytes); 1230c1bb86cdSEric Blake } 1231c1bb86cdSEric Blake 1232c1bb86cdSEric Blake nbytes = handle_aiocb_rw_linear(aiocb, buf); 1233c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { 1234c1bb86cdSEric Blake char *p = buf; 1235c1bb86cdSEric Blake size_t count = aiocb->aio_nbytes, copy; 1236c1bb86cdSEric Blake int i; 1237c1bb86cdSEric Blake 1238c1bb86cdSEric Blake for (i = 0; i < aiocb->aio_niov && count; ++i) { 1239c1bb86cdSEric Blake copy = count; 1240c1bb86cdSEric Blake if (copy > aiocb->aio_iov[i].iov_len) { 1241c1bb86cdSEric Blake copy = aiocb->aio_iov[i].iov_len; 1242c1bb86cdSEric Blake } 1243c1bb86cdSEric Blake memcpy(aiocb->aio_iov[i].iov_base, p, copy); 1244c1bb86cdSEric Blake assert(count >= copy); 1245c1bb86cdSEric Blake p += copy; 1246c1bb86cdSEric Blake count -= copy; 1247c1bb86cdSEric Blake } 1248c1bb86cdSEric Blake assert(count == 0); 1249c1bb86cdSEric Blake } 1250c1bb86cdSEric Blake qemu_vfree(buf); 1251c1bb86cdSEric Blake 1252c1bb86cdSEric Blake return nbytes; 1253c1bb86cdSEric Blake } 1254c1bb86cdSEric Blake 1255c1bb86cdSEric Blake #ifdef CONFIG_XFS 1256c1bb86cdSEric Blake static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes) 1257c1bb86cdSEric Blake { 1258c1bb86cdSEric Blake struct xfs_flock64 fl; 1259c1bb86cdSEric Blake int err; 1260c1bb86cdSEric Blake 1261c1bb86cdSEric Blake memset(&fl, 0, sizeof(fl)); 1262c1bb86cdSEric Blake fl.l_whence = SEEK_SET; 1263c1bb86cdSEric Blake fl.l_start = offset; 1264c1bb86cdSEric Blake fl.l_len = bytes; 1265c1bb86cdSEric Blake 1266c1bb86cdSEric Blake if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) { 1267c1bb86cdSEric Blake err = errno; 1268c1bb86cdSEric Blake DPRINTF("cannot write zero range (%s)\n", strerror(errno)); 1269c1bb86cdSEric Blake return -err; 1270c1bb86cdSEric Blake } 1271c1bb86cdSEric Blake 1272c1bb86cdSEric Blake return 0; 1273c1bb86cdSEric Blake } 1274c1bb86cdSEric Blake 1275c1bb86cdSEric Blake static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes) 1276c1bb86cdSEric Blake { 1277c1bb86cdSEric Blake struct xfs_flock64 fl; 1278c1bb86cdSEric Blake int err; 1279c1bb86cdSEric Blake 1280c1bb86cdSEric Blake memset(&fl, 0, sizeof(fl)); 1281c1bb86cdSEric Blake fl.l_whence = SEEK_SET; 1282c1bb86cdSEric Blake fl.l_start = offset; 1283c1bb86cdSEric Blake fl.l_len = bytes; 1284c1bb86cdSEric Blake 1285c1bb86cdSEric Blake if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) { 1286c1bb86cdSEric Blake err = errno; 1287c1bb86cdSEric Blake DPRINTF("cannot punch hole (%s)\n", strerror(errno)); 1288c1bb86cdSEric Blake return -err; 1289c1bb86cdSEric Blake } 1290c1bb86cdSEric Blake 1291c1bb86cdSEric Blake return 0; 1292c1bb86cdSEric Blake } 1293c1bb86cdSEric Blake #endif 1294c1bb86cdSEric Blake 1295c1bb86cdSEric Blake static int translate_err(int err) 1296c1bb86cdSEric Blake { 1297c1bb86cdSEric Blake if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || 1298c1bb86cdSEric Blake err == -ENOTTY) { 1299c1bb86cdSEric Blake err = -ENOTSUP; 1300c1bb86cdSEric Blake } 1301c1bb86cdSEric Blake return err; 1302c1bb86cdSEric Blake } 1303c1bb86cdSEric Blake 1304c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 1305c1bb86cdSEric Blake static int do_fallocate(int fd, int mode, off_t offset, off_t len) 1306c1bb86cdSEric Blake { 1307c1bb86cdSEric Blake do { 1308c1bb86cdSEric Blake if (fallocate(fd, mode, offset, len) == 0) { 1309c1bb86cdSEric Blake return 0; 1310c1bb86cdSEric Blake } 1311c1bb86cdSEric Blake } while (errno == EINTR); 1312c1bb86cdSEric Blake return translate_err(-errno); 1313c1bb86cdSEric Blake } 1314c1bb86cdSEric Blake #endif 1315c1bb86cdSEric Blake 1316c1bb86cdSEric Blake static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) 1317c1bb86cdSEric Blake { 1318c1bb86cdSEric Blake int ret = -ENOTSUP; 1319c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1320c1bb86cdSEric Blake 1321c1bb86cdSEric Blake if (!s->has_write_zeroes) { 1322c1bb86cdSEric Blake return -ENOTSUP; 1323c1bb86cdSEric Blake } 1324c1bb86cdSEric Blake 1325c1bb86cdSEric Blake #ifdef BLKZEROOUT 1326c1bb86cdSEric Blake do { 1327c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1328c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { 1329c1bb86cdSEric Blake return 0; 1330c1bb86cdSEric Blake } 1331c1bb86cdSEric Blake } while (errno == EINTR); 1332c1bb86cdSEric Blake 1333c1bb86cdSEric Blake ret = translate_err(-errno); 1334c1bb86cdSEric Blake #endif 1335c1bb86cdSEric Blake 1336c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1337c1bb86cdSEric Blake s->has_write_zeroes = false; 1338c1bb86cdSEric Blake } 1339c1bb86cdSEric Blake return ret; 1340c1bb86cdSEric Blake } 1341c1bb86cdSEric Blake 1342c1bb86cdSEric Blake static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb) 1343c1bb86cdSEric Blake { 1344c1bb86cdSEric Blake #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS) 1345c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1346c1bb86cdSEric Blake #endif 1347c1bb86cdSEric Blake 1348c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1349c1bb86cdSEric Blake return handle_aiocb_write_zeroes_block(aiocb); 1350c1bb86cdSEric Blake } 1351c1bb86cdSEric Blake 1352c1bb86cdSEric Blake #ifdef CONFIG_XFS 1353c1bb86cdSEric Blake if (s->is_xfs) { 1354c1bb86cdSEric Blake return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes); 1355c1bb86cdSEric Blake } 1356c1bb86cdSEric Blake #endif 1357c1bb86cdSEric Blake 1358c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_ZERO_RANGE 1359c1bb86cdSEric Blake if (s->has_write_zeroes) { 1360c1bb86cdSEric Blake int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, 1361c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1362c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1363c1bb86cdSEric Blake return ret; 1364c1bb86cdSEric Blake } 1365c1bb86cdSEric Blake s->has_write_zeroes = false; 1366c1bb86cdSEric Blake } 1367c1bb86cdSEric Blake #endif 1368c1bb86cdSEric Blake 1369c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1370c1bb86cdSEric Blake if (s->has_discard && s->has_fallocate) { 1371c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 1372c1bb86cdSEric Blake FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1373c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1374c1bb86cdSEric Blake if (ret == 0) { 1375c1bb86cdSEric Blake ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1376c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1377c1bb86cdSEric Blake return ret; 1378c1bb86cdSEric Blake } 1379c1bb86cdSEric Blake s->has_fallocate = false; 1380c1bb86cdSEric Blake } else if (ret != -ENOTSUP) { 1381c1bb86cdSEric Blake return ret; 1382c1bb86cdSEric Blake } else { 1383c1bb86cdSEric Blake s->has_discard = false; 1384c1bb86cdSEric Blake } 1385c1bb86cdSEric Blake } 1386c1bb86cdSEric Blake #endif 1387c1bb86cdSEric Blake 1388c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 1389c1bb86cdSEric Blake if (s->has_fallocate && aiocb->aio_offset >= bdrv_getlength(aiocb->bs)) { 1390c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1391c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1392c1bb86cdSEric Blake return ret; 1393c1bb86cdSEric Blake } 1394c1bb86cdSEric Blake s->has_fallocate = false; 1395c1bb86cdSEric Blake } 1396c1bb86cdSEric Blake #endif 1397c1bb86cdSEric Blake 1398c1bb86cdSEric Blake return -ENOTSUP; 1399c1bb86cdSEric Blake } 1400c1bb86cdSEric Blake 1401c1bb86cdSEric Blake static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb) 1402c1bb86cdSEric Blake { 1403c1bb86cdSEric Blake int ret = -EOPNOTSUPP; 1404c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1405c1bb86cdSEric Blake 1406c1bb86cdSEric Blake if (!s->has_discard) { 1407c1bb86cdSEric Blake return -ENOTSUP; 1408c1bb86cdSEric Blake } 1409c1bb86cdSEric Blake 1410c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1411c1bb86cdSEric Blake #ifdef BLKDISCARD 1412c1bb86cdSEric Blake do { 1413c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1414c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { 1415c1bb86cdSEric Blake return 0; 1416c1bb86cdSEric Blake } 1417c1bb86cdSEric Blake } while (errno == EINTR); 1418c1bb86cdSEric Blake 1419c1bb86cdSEric Blake ret = -errno; 1420c1bb86cdSEric Blake #endif 1421c1bb86cdSEric Blake } else { 1422c1bb86cdSEric Blake #ifdef CONFIG_XFS 1423c1bb86cdSEric Blake if (s->is_xfs) { 1424c1bb86cdSEric Blake return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes); 1425c1bb86cdSEric Blake } 1426c1bb86cdSEric Blake #endif 1427c1bb86cdSEric Blake 1428c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1429c1bb86cdSEric Blake ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1430c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1431c1bb86cdSEric Blake #endif 1432c1bb86cdSEric Blake } 1433c1bb86cdSEric Blake 1434c1bb86cdSEric Blake ret = translate_err(ret); 1435c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1436c1bb86cdSEric Blake s->has_discard = false; 1437c1bb86cdSEric Blake } 1438c1bb86cdSEric Blake return ret; 1439c1bb86cdSEric Blake } 1440c1bb86cdSEric Blake 1441c1bb86cdSEric Blake static int aio_worker(void *arg) 1442c1bb86cdSEric Blake { 1443c1bb86cdSEric Blake RawPosixAIOData *aiocb = arg; 1444c1bb86cdSEric Blake ssize_t ret = 0; 1445c1bb86cdSEric Blake 1446c1bb86cdSEric Blake switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) { 1447c1bb86cdSEric Blake case QEMU_AIO_READ: 1448c1bb86cdSEric Blake ret = handle_aiocb_rw(aiocb); 1449c1bb86cdSEric Blake if (ret >= 0 && ret < aiocb->aio_nbytes) { 1450c1bb86cdSEric Blake iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret, 1451c1bb86cdSEric Blake 0, aiocb->aio_nbytes - ret); 1452c1bb86cdSEric Blake 1453c1bb86cdSEric Blake ret = aiocb->aio_nbytes; 1454c1bb86cdSEric Blake } 1455c1bb86cdSEric Blake if (ret == aiocb->aio_nbytes) { 1456c1bb86cdSEric Blake ret = 0; 1457c1bb86cdSEric Blake } else if (ret >= 0 && ret < aiocb->aio_nbytes) { 1458c1bb86cdSEric Blake ret = -EINVAL; 1459c1bb86cdSEric Blake } 1460c1bb86cdSEric Blake break; 1461c1bb86cdSEric Blake case QEMU_AIO_WRITE: 1462c1bb86cdSEric Blake ret = handle_aiocb_rw(aiocb); 1463c1bb86cdSEric Blake if (ret == aiocb->aio_nbytes) { 1464c1bb86cdSEric Blake ret = 0; 1465c1bb86cdSEric Blake } else if (ret >= 0 && ret < aiocb->aio_nbytes) { 1466c1bb86cdSEric Blake ret = -EINVAL; 1467c1bb86cdSEric Blake } 1468c1bb86cdSEric Blake break; 1469c1bb86cdSEric Blake case QEMU_AIO_FLUSH: 1470c1bb86cdSEric Blake ret = handle_aiocb_flush(aiocb); 1471c1bb86cdSEric Blake break; 1472c1bb86cdSEric Blake case QEMU_AIO_IOCTL: 1473c1bb86cdSEric Blake ret = handle_aiocb_ioctl(aiocb); 1474c1bb86cdSEric Blake break; 1475c1bb86cdSEric Blake case QEMU_AIO_DISCARD: 1476c1bb86cdSEric Blake ret = handle_aiocb_discard(aiocb); 1477c1bb86cdSEric Blake break; 1478c1bb86cdSEric Blake case QEMU_AIO_WRITE_ZEROES: 1479c1bb86cdSEric Blake ret = handle_aiocb_write_zeroes(aiocb); 1480c1bb86cdSEric Blake break; 1481c1bb86cdSEric Blake default: 1482c1bb86cdSEric Blake fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type); 1483c1bb86cdSEric Blake ret = -EINVAL; 1484c1bb86cdSEric Blake break; 1485c1bb86cdSEric Blake } 1486c1bb86cdSEric Blake 1487c1bb86cdSEric Blake g_free(aiocb); 1488c1bb86cdSEric Blake return ret; 1489c1bb86cdSEric Blake } 1490c1bb86cdSEric Blake 1491c1bb86cdSEric Blake static int paio_submit_co(BlockDriverState *bs, int fd, 1492c1bb86cdSEric Blake int64_t offset, QEMUIOVector *qiov, 1493c1bb86cdSEric Blake int count, int type) 1494c1bb86cdSEric Blake { 1495c1bb86cdSEric Blake RawPosixAIOData *acb = g_new(RawPosixAIOData, 1); 1496c1bb86cdSEric Blake ThreadPool *pool; 1497c1bb86cdSEric Blake 1498c1bb86cdSEric Blake acb->bs = bs; 1499c1bb86cdSEric Blake acb->aio_type = type; 1500c1bb86cdSEric Blake acb->aio_fildes = fd; 1501c1bb86cdSEric Blake 1502c1bb86cdSEric Blake acb->aio_nbytes = count; 1503c1bb86cdSEric Blake acb->aio_offset = offset; 1504c1bb86cdSEric Blake 1505c1bb86cdSEric Blake if (qiov) { 1506c1bb86cdSEric Blake acb->aio_iov = qiov->iov; 1507c1bb86cdSEric Blake acb->aio_niov = qiov->niov; 1508c1bb86cdSEric Blake assert(qiov->size == count); 1509c1bb86cdSEric Blake } 1510c1bb86cdSEric Blake 1511c1bb86cdSEric Blake trace_paio_submit_co(offset, count, type); 1512c1bb86cdSEric Blake pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 1513c1bb86cdSEric Blake return thread_pool_submit_co(pool, aio_worker, acb); 1514c1bb86cdSEric Blake } 1515c1bb86cdSEric Blake 1516c1bb86cdSEric Blake static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd, 1517c1bb86cdSEric Blake int64_t offset, QEMUIOVector *qiov, int count, 1518c1bb86cdSEric Blake BlockCompletionFunc *cb, void *opaque, int type) 1519c1bb86cdSEric Blake { 1520c1bb86cdSEric Blake RawPosixAIOData *acb = g_new(RawPosixAIOData, 1); 1521c1bb86cdSEric Blake ThreadPool *pool; 1522c1bb86cdSEric Blake 1523c1bb86cdSEric Blake acb->bs = bs; 1524c1bb86cdSEric Blake acb->aio_type = type; 1525c1bb86cdSEric Blake acb->aio_fildes = fd; 1526c1bb86cdSEric Blake 1527c1bb86cdSEric Blake acb->aio_nbytes = count; 1528c1bb86cdSEric Blake acb->aio_offset = offset; 1529c1bb86cdSEric Blake 1530c1bb86cdSEric Blake if (qiov) { 1531c1bb86cdSEric Blake acb->aio_iov = qiov->iov; 1532c1bb86cdSEric Blake acb->aio_niov = qiov->niov; 1533c1bb86cdSEric Blake assert(qiov->size == acb->aio_nbytes); 1534c1bb86cdSEric Blake } 1535c1bb86cdSEric Blake 1536c1bb86cdSEric Blake trace_paio_submit(acb, opaque, offset, count, type); 1537c1bb86cdSEric Blake pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 1538c1bb86cdSEric Blake return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque); 1539c1bb86cdSEric Blake } 1540c1bb86cdSEric Blake 1541c1bb86cdSEric Blake static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, 1542c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, int type) 1543c1bb86cdSEric Blake { 1544c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1545c1bb86cdSEric Blake 1546c1bb86cdSEric Blake if (fd_open(bs) < 0) 1547c1bb86cdSEric Blake return -EIO; 1548c1bb86cdSEric Blake 1549c1bb86cdSEric Blake /* 1550c1bb86cdSEric Blake * Check if the underlying device requires requests to be aligned, 1551c1bb86cdSEric Blake * and if the request we are trying to submit is aligned or not. 1552c1bb86cdSEric Blake * If this is the case tell the low-level driver that it needs 1553c1bb86cdSEric Blake * to copy the buffer. 1554c1bb86cdSEric Blake */ 1555c1bb86cdSEric Blake if (s->needs_alignment) { 1556c1bb86cdSEric Blake if (!bdrv_qiov_is_aligned(bs, qiov)) { 1557c1bb86cdSEric Blake type |= QEMU_AIO_MISALIGNED; 1558c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1559c1bb86cdSEric Blake } else if (s->use_linux_aio) { 1560c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1561c1bb86cdSEric Blake assert(qiov->size == bytes); 1562c1bb86cdSEric Blake return laio_co_submit(bs, aio, s->fd, offset, qiov, type); 1563c1bb86cdSEric Blake #endif 1564c1bb86cdSEric Blake } 1565c1bb86cdSEric Blake } 1566c1bb86cdSEric Blake 1567c1bb86cdSEric Blake return paio_submit_co(bs, s->fd, offset, qiov, bytes, type); 1568c1bb86cdSEric Blake } 1569c1bb86cdSEric Blake 1570c1bb86cdSEric Blake static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset, 1571c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 1572c1bb86cdSEric Blake int flags) 1573c1bb86cdSEric Blake { 1574c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); 1575c1bb86cdSEric Blake } 1576c1bb86cdSEric Blake 1577c1bb86cdSEric Blake static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset, 1578c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 1579c1bb86cdSEric Blake int flags) 1580c1bb86cdSEric Blake { 1581c1bb86cdSEric Blake assert(flags == 0); 1582c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); 1583c1bb86cdSEric Blake } 1584c1bb86cdSEric Blake 1585c1bb86cdSEric Blake static void raw_aio_plug(BlockDriverState *bs) 1586c1bb86cdSEric Blake { 1587c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1588c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1589c1bb86cdSEric Blake if (s->use_linux_aio) { 1590c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1591c1bb86cdSEric Blake laio_io_plug(bs, aio); 1592c1bb86cdSEric Blake } 1593c1bb86cdSEric Blake #endif 1594c1bb86cdSEric Blake } 1595c1bb86cdSEric Blake 1596c1bb86cdSEric Blake static void raw_aio_unplug(BlockDriverState *bs) 1597c1bb86cdSEric Blake { 1598c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1599c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1600c1bb86cdSEric Blake if (s->use_linux_aio) { 1601c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1602c1bb86cdSEric Blake laio_io_unplug(bs, aio); 1603c1bb86cdSEric Blake } 1604c1bb86cdSEric Blake #endif 1605c1bb86cdSEric Blake } 1606c1bb86cdSEric Blake 1607c1bb86cdSEric Blake static BlockAIOCB *raw_aio_flush(BlockDriverState *bs, 1608c1bb86cdSEric Blake BlockCompletionFunc *cb, void *opaque) 1609c1bb86cdSEric Blake { 1610c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1611c1bb86cdSEric Blake 1612c1bb86cdSEric Blake if (fd_open(bs) < 0) 1613c1bb86cdSEric Blake return NULL; 1614c1bb86cdSEric Blake 1615c1bb86cdSEric Blake return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH); 1616c1bb86cdSEric Blake } 1617c1bb86cdSEric Blake 1618c1bb86cdSEric Blake static void raw_close(BlockDriverState *bs) 1619c1bb86cdSEric Blake { 1620c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1621c1bb86cdSEric Blake 1622c1bb86cdSEric Blake if (s->fd >= 0) { 1623c1bb86cdSEric Blake qemu_close(s->fd); 1624c1bb86cdSEric Blake s->fd = -1; 1625c1bb86cdSEric Blake } 1626*244a5668SFam Zheng if (s->lock_fd >= 0) { 1627*244a5668SFam Zheng qemu_close(s->lock_fd); 1628*244a5668SFam Zheng s->lock_fd = -1; 1629*244a5668SFam Zheng } 1630c1bb86cdSEric Blake } 1631c1bb86cdSEric Blake 16324bff28b8SMax Reitz static int raw_truncate(BlockDriverState *bs, int64_t offset, Error **errp) 1633c1bb86cdSEric Blake { 1634c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1635c1bb86cdSEric Blake struct stat st; 1636f59adb32SMax Reitz int ret; 1637c1bb86cdSEric Blake 1638c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 1639f59adb32SMax Reitz ret = -errno; 1640f59adb32SMax Reitz error_setg_errno(errp, -ret, "Failed to fstat() the file"); 1641f59adb32SMax Reitz return ret; 1642c1bb86cdSEric Blake } 1643c1bb86cdSEric Blake 1644c1bb86cdSEric Blake if (S_ISREG(st.st_mode)) { 1645c1bb86cdSEric Blake if (ftruncate(s->fd, offset) < 0) { 1646f59adb32SMax Reitz ret = -errno; 1647f59adb32SMax Reitz error_setg_errno(errp, -ret, "Failed to resize the file"); 1648f59adb32SMax Reitz return ret; 1649c1bb86cdSEric Blake } 1650c1bb86cdSEric Blake } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 1651c1bb86cdSEric Blake if (offset > raw_getlength(bs)) { 1652f59adb32SMax Reitz error_setg(errp, "Cannot grow device files"); 1653c1bb86cdSEric Blake return -EINVAL; 1654c1bb86cdSEric Blake } 1655c1bb86cdSEric Blake } else { 1656f59adb32SMax Reitz error_setg(errp, "Resizing this file is not supported"); 1657c1bb86cdSEric Blake return -ENOTSUP; 1658c1bb86cdSEric Blake } 1659c1bb86cdSEric Blake 1660c1bb86cdSEric Blake return 0; 1661c1bb86cdSEric Blake } 1662c1bb86cdSEric Blake 1663c1bb86cdSEric Blake #ifdef __OpenBSD__ 1664c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 1665c1bb86cdSEric Blake { 1666c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1667c1bb86cdSEric Blake int fd = s->fd; 1668c1bb86cdSEric Blake struct stat st; 1669c1bb86cdSEric Blake 1670c1bb86cdSEric Blake if (fstat(fd, &st)) 1671c1bb86cdSEric Blake return -errno; 1672c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 1673c1bb86cdSEric Blake struct disklabel dl; 1674c1bb86cdSEric Blake 1675c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 1676c1bb86cdSEric Blake return -errno; 1677c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 1678c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 1679c1bb86cdSEric Blake } else 1680c1bb86cdSEric Blake return st.st_size; 1681c1bb86cdSEric Blake } 1682c1bb86cdSEric Blake #elif defined(__NetBSD__) 1683c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 1684c1bb86cdSEric Blake { 1685c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1686c1bb86cdSEric Blake int fd = s->fd; 1687c1bb86cdSEric Blake struct stat st; 1688c1bb86cdSEric Blake 1689c1bb86cdSEric Blake if (fstat(fd, &st)) 1690c1bb86cdSEric Blake return -errno; 1691c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 1692c1bb86cdSEric Blake struct dkwedge_info dkw; 1693c1bb86cdSEric Blake 1694c1bb86cdSEric Blake if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { 1695c1bb86cdSEric Blake return dkw.dkw_size * 512; 1696c1bb86cdSEric Blake } else { 1697c1bb86cdSEric Blake struct disklabel dl; 1698c1bb86cdSEric Blake 1699c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 1700c1bb86cdSEric Blake return -errno; 1701c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 1702c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 1703c1bb86cdSEric Blake } 1704c1bb86cdSEric Blake } else 1705c1bb86cdSEric Blake return st.st_size; 1706c1bb86cdSEric Blake } 1707c1bb86cdSEric Blake #elif defined(__sun__) 1708c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 1709c1bb86cdSEric Blake { 1710c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1711c1bb86cdSEric Blake struct dk_minfo minfo; 1712c1bb86cdSEric Blake int ret; 1713c1bb86cdSEric Blake int64_t size; 1714c1bb86cdSEric Blake 1715c1bb86cdSEric Blake ret = fd_open(bs); 1716c1bb86cdSEric Blake if (ret < 0) { 1717c1bb86cdSEric Blake return ret; 1718c1bb86cdSEric Blake } 1719c1bb86cdSEric Blake 1720c1bb86cdSEric Blake /* 1721c1bb86cdSEric Blake * Use the DKIOCGMEDIAINFO ioctl to read the size. 1722c1bb86cdSEric Blake */ 1723c1bb86cdSEric Blake ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); 1724c1bb86cdSEric Blake if (ret != -1) { 1725c1bb86cdSEric Blake return minfo.dki_lbsize * minfo.dki_capacity; 1726c1bb86cdSEric Blake } 1727c1bb86cdSEric Blake 1728c1bb86cdSEric Blake /* 1729c1bb86cdSEric Blake * There are reports that lseek on some devices fails, but 1730c1bb86cdSEric Blake * irc discussion said that contingency on contingency was overkill. 1731c1bb86cdSEric Blake */ 1732c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 1733c1bb86cdSEric Blake if (size < 0) { 1734c1bb86cdSEric Blake return -errno; 1735c1bb86cdSEric Blake } 1736c1bb86cdSEric Blake return size; 1737c1bb86cdSEric Blake } 1738c1bb86cdSEric Blake #elif defined(CONFIG_BSD) 1739c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 1740c1bb86cdSEric Blake { 1741c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1742c1bb86cdSEric Blake int fd = s->fd; 1743c1bb86cdSEric Blake int64_t size; 1744c1bb86cdSEric Blake struct stat sb; 1745c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 1746c1bb86cdSEric Blake int reopened = 0; 1747c1bb86cdSEric Blake #endif 1748c1bb86cdSEric Blake int ret; 1749c1bb86cdSEric Blake 1750c1bb86cdSEric Blake ret = fd_open(bs); 1751c1bb86cdSEric Blake if (ret < 0) 1752c1bb86cdSEric Blake return ret; 1753c1bb86cdSEric Blake 1754c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 1755c1bb86cdSEric Blake again: 1756c1bb86cdSEric Blake #endif 1757c1bb86cdSEric Blake if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { 1758c1bb86cdSEric Blake #ifdef DIOCGMEDIASIZE 1759c1bb86cdSEric Blake if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) 1760c1bb86cdSEric Blake #elif defined(DIOCGPART) 1761c1bb86cdSEric Blake { 1762c1bb86cdSEric Blake struct partinfo pi; 1763c1bb86cdSEric Blake if (ioctl(fd, DIOCGPART, &pi) == 0) 1764c1bb86cdSEric Blake size = pi.media_size; 1765c1bb86cdSEric Blake else 1766c1bb86cdSEric Blake size = 0; 1767c1bb86cdSEric Blake } 1768c1bb86cdSEric Blake if (size == 0) 1769c1bb86cdSEric Blake #endif 1770c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 1771c1bb86cdSEric Blake { 1772c1bb86cdSEric Blake uint64_t sectors = 0; 1773c1bb86cdSEric Blake uint32_t sector_size = 0; 1774c1bb86cdSEric Blake 1775c1bb86cdSEric Blake if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 1776c1bb86cdSEric Blake && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { 1777c1bb86cdSEric Blake size = sectors * sector_size; 1778c1bb86cdSEric Blake } else { 1779c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 1780c1bb86cdSEric Blake if (size < 0) { 1781c1bb86cdSEric Blake return -errno; 1782c1bb86cdSEric Blake } 1783c1bb86cdSEric Blake } 1784c1bb86cdSEric Blake } 1785c1bb86cdSEric Blake #else 1786c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 1787c1bb86cdSEric Blake if (size < 0) { 1788c1bb86cdSEric Blake return -errno; 1789c1bb86cdSEric Blake } 1790c1bb86cdSEric Blake #endif 1791c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 1792c1bb86cdSEric Blake switch(s->type) { 1793c1bb86cdSEric Blake case FTYPE_CD: 1794c1bb86cdSEric Blake /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ 1795c1bb86cdSEric Blake if (size == 2048LL * (unsigned)-1) 1796c1bb86cdSEric Blake size = 0; 1797c1bb86cdSEric Blake /* XXX no disc? maybe we need to reopen... */ 1798c1bb86cdSEric Blake if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { 1799c1bb86cdSEric Blake reopened = 1; 1800c1bb86cdSEric Blake goto again; 1801c1bb86cdSEric Blake } 1802c1bb86cdSEric Blake } 1803c1bb86cdSEric Blake #endif 1804c1bb86cdSEric Blake } else { 1805c1bb86cdSEric Blake size = lseek(fd, 0, SEEK_END); 1806c1bb86cdSEric Blake if (size < 0) { 1807c1bb86cdSEric Blake return -errno; 1808c1bb86cdSEric Blake } 1809c1bb86cdSEric Blake } 1810c1bb86cdSEric Blake return size; 1811c1bb86cdSEric Blake } 1812c1bb86cdSEric Blake #else 1813c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 1814c1bb86cdSEric Blake { 1815c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1816c1bb86cdSEric Blake int ret; 1817c1bb86cdSEric Blake int64_t size; 1818c1bb86cdSEric Blake 1819c1bb86cdSEric Blake ret = fd_open(bs); 1820c1bb86cdSEric Blake if (ret < 0) { 1821c1bb86cdSEric Blake return ret; 1822c1bb86cdSEric Blake } 1823c1bb86cdSEric Blake 1824c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 1825c1bb86cdSEric Blake if (size < 0) { 1826c1bb86cdSEric Blake return -errno; 1827c1bb86cdSEric Blake } 1828c1bb86cdSEric Blake return size; 1829c1bb86cdSEric Blake } 1830c1bb86cdSEric Blake #endif 1831c1bb86cdSEric Blake 1832c1bb86cdSEric Blake static int64_t raw_get_allocated_file_size(BlockDriverState *bs) 1833c1bb86cdSEric Blake { 1834c1bb86cdSEric Blake struct stat st; 1835c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1836c1bb86cdSEric Blake 1837c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 1838c1bb86cdSEric Blake return -errno; 1839c1bb86cdSEric Blake } 1840c1bb86cdSEric Blake return (int64_t)st.st_blocks * 512; 1841c1bb86cdSEric Blake } 1842c1bb86cdSEric Blake 1843c1bb86cdSEric Blake static int raw_create(const char *filename, QemuOpts *opts, Error **errp) 1844c1bb86cdSEric Blake { 1845c1bb86cdSEric Blake int fd; 1846c1bb86cdSEric Blake int result = 0; 1847c1bb86cdSEric Blake int64_t total_size = 0; 1848c1bb86cdSEric Blake bool nocow = false; 1849c1bb86cdSEric Blake PreallocMode prealloc; 1850c1bb86cdSEric Blake char *buf = NULL; 1851c1bb86cdSEric Blake Error *local_err = NULL; 1852c1bb86cdSEric Blake 1853c1bb86cdSEric Blake strstart(filename, "file:", &filename); 1854c1bb86cdSEric Blake 1855c1bb86cdSEric Blake /* Read out options */ 1856c1bb86cdSEric Blake total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 1857c1bb86cdSEric Blake BDRV_SECTOR_SIZE); 1858c1bb86cdSEric Blake nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); 1859c1bb86cdSEric Blake buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 1860c1bb86cdSEric Blake prealloc = qapi_enum_parse(PreallocMode_lookup, buf, 1861c1bb86cdSEric Blake PREALLOC_MODE__MAX, PREALLOC_MODE_OFF, 1862c1bb86cdSEric Blake &local_err); 1863c1bb86cdSEric Blake g_free(buf); 1864c1bb86cdSEric Blake if (local_err) { 1865c1bb86cdSEric Blake error_propagate(errp, local_err); 1866c1bb86cdSEric Blake result = -EINVAL; 1867c1bb86cdSEric Blake goto out; 1868c1bb86cdSEric Blake } 1869c1bb86cdSEric Blake 1870c1bb86cdSEric Blake fd = qemu_open(filename, O_RDWR | O_CREAT | O_TRUNC | O_BINARY, 1871c1bb86cdSEric Blake 0644); 1872c1bb86cdSEric Blake if (fd < 0) { 1873c1bb86cdSEric Blake result = -errno; 1874c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not create file"); 1875c1bb86cdSEric Blake goto out; 1876c1bb86cdSEric Blake } 1877c1bb86cdSEric Blake 1878c1bb86cdSEric Blake if (nocow) { 1879c1bb86cdSEric Blake #ifdef __linux__ 1880c1bb86cdSEric Blake /* Set NOCOW flag to solve performance issue on fs like btrfs. 1881c1bb86cdSEric Blake * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value 1882c1bb86cdSEric Blake * will be ignored since any failure of this operation should not 1883c1bb86cdSEric Blake * block the left work. 1884c1bb86cdSEric Blake */ 1885c1bb86cdSEric Blake int attr; 1886c1bb86cdSEric Blake if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { 1887c1bb86cdSEric Blake attr |= FS_NOCOW_FL; 1888c1bb86cdSEric Blake ioctl(fd, FS_IOC_SETFLAGS, &attr); 1889c1bb86cdSEric Blake } 1890c1bb86cdSEric Blake #endif 1891c1bb86cdSEric Blake } 1892c1bb86cdSEric Blake 1893c1bb86cdSEric Blake switch (prealloc) { 1894c1bb86cdSEric Blake #ifdef CONFIG_POSIX_FALLOCATE 1895c1bb86cdSEric Blake case PREALLOC_MODE_FALLOC: 1896c6ccc2c5SNir Soffer /* 1897c6ccc2c5SNir Soffer * Truncating before posix_fallocate() makes it about twice slower on 1898c6ccc2c5SNir Soffer * file systems that do not support fallocate(), trying to check if a 1899c6ccc2c5SNir Soffer * block is allocated before allocating it, so don't do that here. 1900c6ccc2c5SNir Soffer */ 1901c1bb86cdSEric Blake result = -posix_fallocate(fd, 0, total_size); 1902c1bb86cdSEric Blake if (result != 0) { 1903c6ccc2c5SNir Soffer /* posix_fallocate() doesn't set errno. */ 1904c1bb86cdSEric Blake error_setg_errno(errp, -result, 1905c1bb86cdSEric Blake "Could not preallocate data for the new file"); 1906c1bb86cdSEric Blake } 1907c1bb86cdSEric Blake break; 1908c1bb86cdSEric Blake #endif 1909c1bb86cdSEric Blake case PREALLOC_MODE_FULL: 1910c1bb86cdSEric Blake { 19115a1dad9dSNir Soffer /* 19125a1dad9dSNir Soffer * Knowing the final size from the beginning could allow the file 19135a1dad9dSNir Soffer * system driver to do less allocations and possibly avoid 19145a1dad9dSNir Soffer * fragmentation of the file. 19155a1dad9dSNir Soffer */ 19165a1dad9dSNir Soffer if (ftruncate(fd, total_size) != 0) { 19175a1dad9dSNir Soffer result = -errno; 19185a1dad9dSNir Soffer error_setg_errno(errp, -result, "Could not resize file"); 19195a1dad9dSNir Soffer goto out_close; 19205a1dad9dSNir Soffer } 19215a1dad9dSNir Soffer 1922c1bb86cdSEric Blake int64_t num = 0, left = total_size; 1923c1bb86cdSEric Blake buf = g_malloc0(65536); 1924c1bb86cdSEric Blake 1925c1bb86cdSEric Blake while (left > 0) { 1926c1bb86cdSEric Blake num = MIN(left, 65536); 1927c1bb86cdSEric Blake result = write(fd, buf, num); 1928c1bb86cdSEric Blake if (result < 0) { 1929c1bb86cdSEric Blake result = -errno; 1930c1bb86cdSEric Blake error_setg_errno(errp, -result, 1931c1bb86cdSEric Blake "Could not write to the new file"); 1932c1bb86cdSEric Blake break; 1933c1bb86cdSEric Blake } 1934c1bb86cdSEric Blake left -= result; 1935c1bb86cdSEric Blake } 1936c1bb86cdSEric Blake if (result >= 0) { 1937c1bb86cdSEric Blake result = fsync(fd); 1938c1bb86cdSEric Blake if (result < 0) { 1939c1bb86cdSEric Blake result = -errno; 1940c1bb86cdSEric Blake error_setg_errno(errp, -result, 1941c1bb86cdSEric Blake "Could not flush new file to disk"); 1942c1bb86cdSEric Blake } 1943c1bb86cdSEric Blake } 1944c1bb86cdSEric Blake g_free(buf); 1945c1bb86cdSEric Blake break; 1946c1bb86cdSEric Blake } 1947c1bb86cdSEric Blake case PREALLOC_MODE_OFF: 1948f6a72404SNir Soffer if (ftruncate(fd, total_size) != 0) { 1949f6a72404SNir Soffer result = -errno; 1950f6a72404SNir Soffer error_setg_errno(errp, -result, "Could not resize file"); 1951f6a72404SNir Soffer } 1952c1bb86cdSEric Blake break; 1953c1bb86cdSEric Blake default: 1954c1bb86cdSEric Blake result = -EINVAL; 1955c1bb86cdSEric Blake error_setg(errp, "Unsupported preallocation mode: %s", 1956c1bb86cdSEric Blake PreallocMode_lookup[prealloc]); 1957c1bb86cdSEric Blake break; 1958c1bb86cdSEric Blake } 1959c1bb86cdSEric Blake 19605a1dad9dSNir Soffer out_close: 1961c1bb86cdSEric Blake if (qemu_close(fd) != 0 && result == 0) { 1962c1bb86cdSEric Blake result = -errno; 1963c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not close the new file"); 1964c1bb86cdSEric Blake } 1965c1bb86cdSEric Blake out: 1966c1bb86cdSEric Blake return result; 1967c1bb86cdSEric Blake } 1968c1bb86cdSEric Blake 1969c1bb86cdSEric Blake /* 1970c1bb86cdSEric Blake * Find allocation range in @bs around offset @start. 1971c1bb86cdSEric Blake * May change underlying file descriptor's file offset. 1972c1bb86cdSEric Blake * If @start is not in a hole, store @start in @data, and the 1973c1bb86cdSEric Blake * beginning of the next hole in @hole, and return 0. 1974c1bb86cdSEric Blake * If @start is in a non-trailing hole, store @start in @hole and the 1975c1bb86cdSEric Blake * beginning of the next non-hole in @data, and return 0. 1976c1bb86cdSEric Blake * If @start is in a trailing hole or beyond EOF, return -ENXIO. 1977c1bb86cdSEric Blake * If we can't find out, return a negative errno other than -ENXIO. 1978c1bb86cdSEric Blake */ 1979c1bb86cdSEric Blake static int find_allocation(BlockDriverState *bs, off_t start, 1980c1bb86cdSEric Blake off_t *data, off_t *hole) 1981c1bb86cdSEric Blake { 1982c1bb86cdSEric Blake #if defined SEEK_HOLE && defined SEEK_DATA 1983c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1984c1bb86cdSEric Blake off_t offs; 1985c1bb86cdSEric Blake 1986c1bb86cdSEric Blake /* 1987c1bb86cdSEric Blake * SEEK_DATA cases: 1988c1bb86cdSEric Blake * D1. offs == start: start is in data 1989c1bb86cdSEric Blake * D2. offs > start: start is in a hole, next data at offs 1990c1bb86cdSEric Blake * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 1991c1bb86cdSEric Blake * or start is beyond EOF 1992c1bb86cdSEric Blake * If the latter happens, the file has been truncated behind 1993c1bb86cdSEric Blake * our back since we opened it. All bets are off then. 1994c1bb86cdSEric Blake * Treating like a trailing hole is simplest. 1995c1bb86cdSEric Blake * D4. offs < 0, errno != ENXIO: we learned nothing 1996c1bb86cdSEric Blake */ 1997c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_DATA); 1998c1bb86cdSEric Blake if (offs < 0) { 1999c1bb86cdSEric Blake return -errno; /* D3 or D4 */ 2000c1bb86cdSEric Blake } 2001c1bb86cdSEric Blake assert(offs >= start); 2002c1bb86cdSEric Blake 2003c1bb86cdSEric Blake if (offs > start) { 2004c1bb86cdSEric Blake /* D2: in hole, next data at offs */ 2005c1bb86cdSEric Blake *hole = start; 2006c1bb86cdSEric Blake *data = offs; 2007c1bb86cdSEric Blake return 0; 2008c1bb86cdSEric Blake } 2009c1bb86cdSEric Blake 2010c1bb86cdSEric Blake /* D1: in data, end not yet known */ 2011c1bb86cdSEric Blake 2012c1bb86cdSEric Blake /* 2013c1bb86cdSEric Blake * SEEK_HOLE cases: 2014c1bb86cdSEric Blake * H1. offs == start: start is in a hole 2015c1bb86cdSEric Blake * If this happens here, a hole has been dug behind our back 2016c1bb86cdSEric Blake * since the previous lseek(). 2017c1bb86cdSEric Blake * H2. offs > start: either start is in data, next hole at offs, 2018c1bb86cdSEric Blake * or start is in trailing hole, EOF at offs 2019c1bb86cdSEric Blake * Linux treats trailing holes like any other hole: offs == 2020c1bb86cdSEric Blake * start. Solaris seeks to EOF instead: offs > start (blech). 2021c1bb86cdSEric Blake * If that happens here, a hole has been dug behind our back 2022c1bb86cdSEric Blake * since the previous lseek(). 2023c1bb86cdSEric Blake * H3. offs < 0, errno = ENXIO: start is beyond EOF 2024c1bb86cdSEric Blake * If this happens, the file has been truncated behind our 2025c1bb86cdSEric Blake * back since we opened it. Treat it like a trailing hole. 2026c1bb86cdSEric Blake * H4. offs < 0, errno != ENXIO: we learned nothing 2027c1bb86cdSEric Blake * Pretend we know nothing at all, i.e. "forget" about D1. 2028c1bb86cdSEric Blake */ 2029c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_HOLE); 2030c1bb86cdSEric Blake if (offs < 0) { 2031c1bb86cdSEric Blake return -errno; /* D1 and (H3 or H4) */ 2032c1bb86cdSEric Blake } 2033c1bb86cdSEric Blake assert(offs >= start); 2034c1bb86cdSEric Blake 2035c1bb86cdSEric Blake if (offs > start) { 2036c1bb86cdSEric Blake /* 2037c1bb86cdSEric Blake * D1 and H2: either in data, next hole at offs, or it was in 2038c1bb86cdSEric Blake * data but is now in a trailing hole. In the latter case, 2039c1bb86cdSEric Blake * all bets are off. Treating it as if it there was data all 2040c1bb86cdSEric Blake * the way to EOF is safe, so simply do that. 2041c1bb86cdSEric Blake */ 2042c1bb86cdSEric Blake *data = start; 2043c1bb86cdSEric Blake *hole = offs; 2044c1bb86cdSEric Blake return 0; 2045c1bb86cdSEric Blake } 2046c1bb86cdSEric Blake 2047c1bb86cdSEric Blake /* D1 and H1 */ 2048c1bb86cdSEric Blake return -EBUSY; 2049c1bb86cdSEric Blake #else 2050c1bb86cdSEric Blake return -ENOTSUP; 2051c1bb86cdSEric Blake #endif 2052c1bb86cdSEric Blake } 2053c1bb86cdSEric Blake 2054c1bb86cdSEric Blake /* 2055c1bb86cdSEric Blake * Returns the allocation status of the specified sectors. 2056c1bb86cdSEric Blake * 2057c1bb86cdSEric Blake * If 'sector_num' is beyond the end of the disk image the return value is 0 2058c1bb86cdSEric Blake * and 'pnum' is set to 0. 2059c1bb86cdSEric Blake * 2060c1bb86cdSEric Blake * 'pnum' is set to the number of sectors (including and immediately following 2061c1bb86cdSEric Blake * the specified sector) that are known to be in the same 2062c1bb86cdSEric Blake * allocated/unallocated state. 2063c1bb86cdSEric Blake * 2064c1bb86cdSEric Blake * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes 2065c1bb86cdSEric Blake * beyond the end of the disk image it will be clamped. 2066c1bb86cdSEric Blake */ 2067c1bb86cdSEric Blake static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs, 2068c1bb86cdSEric Blake int64_t sector_num, 2069c1bb86cdSEric Blake int nb_sectors, int *pnum, 2070c1bb86cdSEric Blake BlockDriverState **file) 2071c1bb86cdSEric Blake { 2072c1bb86cdSEric Blake off_t start, data = 0, hole = 0; 2073c1bb86cdSEric Blake int64_t total_size; 2074c1bb86cdSEric Blake int ret; 2075c1bb86cdSEric Blake 2076c1bb86cdSEric Blake ret = fd_open(bs); 2077c1bb86cdSEric Blake if (ret < 0) { 2078c1bb86cdSEric Blake return ret; 2079c1bb86cdSEric Blake } 2080c1bb86cdSEric Blake 2081c1bb86cdSEric Blake start = sector_num * BDRV_SECTOR_SIZE; 2082c1bb86cdSEric Blake total_size = bdrv_getlength(bs); 2083c1bb86cdSEric Blake if (total_size < 0) { 2084c1bb86cdSEric Blake return total_size; 2085c1bb86cdSEric Blake } else if (start >= total_size) { 2086c1bb86cdSEric Blake *pnum = 0; 2087c1bb86cdSEric Blake return 0; 2088c1bb86cdSEric Blake } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) { 2089c1bb86cdSEric Blake nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE); 2090c1bb86cdSEric Blake } 2091c1bb86cdSEric Blake 2092c1bb86cdSEric Blake ret = find_allocation(bs, start, &data, &hole); 2093c1bb86cdSEric Blake if (ret == -ENXIO) { 2094c1bb86cdSEric Blake /* Trailing hole */ 2095c1bb86cdSEric Blake *pnum = nb_sectors; 2096c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2097c1bb86cdSEric Blake } else if (ret < 0) { 2098c1bb86cdSEric Blake /* No info available, so pretend there are no holes */ 2099c1bb86cdSEric Blake *pnum = nb_sectors; 2100c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2101c1bb86cdSEric Blake } else if (data == start) { 2102c1bb86cdSEric Blake /* On a data extent, compute sectors to the end of the extent, 2103c1bb86cdSEric Blake * possibly including a partial sector at EOF. */ 2104c1bb86cdSEric Blake *pnum = MIN(nb_sectors, DIV_ROUND_UP(hole - start, BDRV_SECTOR_SIZE)); 2105c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2106c1bb86cdSEric Blake } else { 2107c1bb86cdSEric Blake /* On a hole, compute sectors to the beginning of the next extent. */ 2108c1bb86cdSEric Blake assert(hole == start); 2109c1bb86cdSEric Blake *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE); 2110c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2111c1bb86cdSEric Blake } 2112c1bb86cdSEric Blake *file = bs; 2113c1bb86cdSEric Blake return ret | BDRV_BLOCK_OFFSET_VALID | start; 2114c1bb86cdSEric Blake } 2115c1bb86cdSEric Blake 2116c1bb86cdSEric Blake static coroutine_fn BlockAIOCB *raw_aio_pdiscard(BlockDriverState *bs, 2117c1bb86cdSEric Blake int64_t offset, int count, 2118c1bb86cdSEric Blake BlockCompletionFunc *cb, void *opaque) 2119c1bb86cdSEric Blake { 2120c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2121c1bb86cdSEric Blake 2122c1bb86cdSEric Blake return paio_submit(bs, s->fd, offset, NULL, count, 2123c1bb86cdSEric Blake cb, opaque, QEMU_AIO_DISCARD); 2124c1bb86cdSEric Blake } 2125c1bb86cdSEric Blake 2126c1bb86cdSEric Blake static int coroutine_fn raw_co_pwrite_zeroes( 2127c1bb86cdSEric Blake BlockDriverState *bs, int64_t offset, 2128c1bb86cdSEric Blake int count, BdrvRequestFlags flags) 2129c1bb86cdSEric Blake { 2130c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2131c1bb86cdSEric Blake 2132c1bb86cdSEric Blake if (!(flags & BDRV_REQ_MAY_UNMAP)) { 2133c1bb86cdSEric Blake return paio_submit_co(bs, s->fd, offset, NULL, count, 2134c1bb86cdSEric Blake QEMU_AIO_WRITE_ZEROES); 2135c1bb86cdSEric Blake } else if (s->discard_zeroes) { 2136c1bb86cdSEric Blake return paio_submit_co(bs, s->fd, offset, NULL, count, 2137c1bb86cdSEric Blake QEMU_AIO_DISCARD); 2138c1bb86cdSEric Blake } 2139c1bb86cdSEric Blake return -ENOTSUP; 2140c1bb86cdSEric Blake } 2141c1bb86cdSEric Blake 2142c1bb86cdSEric Blake static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2143c1bb86cdSEric Blake { 2144c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2145c1bb86cdSEric Blake 2146c1bb86cdSEric Blake bdi->unallocated_blocks_are_zero = s->discard_zeroes; 2147c1bb86cdSEric Blake bdi->can_write_zeroes_with_unmap = s->discard_zeroes; 2148c1bb86cdSEric Blake return 0; 2149c1bb86cdSEric Blake } 2150c1bb86cdSEric Blake 2151c1bb86cdSEric Blake static QemuOptsList raw_create_opts = { 2152c1bb86cdSEric Blake .name = "raw-create-opts", 2153c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), 2154c1bb86cdSEric Blake .desc = { 2155c1bb86cdSEric Blake { 2156c1bb86cdSEric Blake .name = BLOCK_OPT_SIZE, 2157c1bb86cdSEric Blake .type = QEMU_OPT_SIZE, 2158c1bb86cdSEric Blake .help = "Virtual disk size" 2159c1bb86cdSEric Blake }, 2160c1bb86cdSEric Blake { 2161c1bb86cdSEric Blake .name = BLOCK_OPT_NOCOW, 2162c1bb86cdSEric Blake .type = QEMU_OPT_BOOL, 2163c1bb86cdSEric Blake .help = "Turn off copy-on-write (valid only on btrfs)" 2164c1bb86cdSEric Blake }, 2165c1bb86cdSEric Blake { 2166c1bb86cdSEric Blake .name = BLOCK_OPT_PREALLOC, 2167c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 2168c1bb86cdSEric Blake .help = "Preallocation mode (allowed values: off, falloc, full)" 2169c1bb86cdSEric Blake }, 2170c1bb86cdSEric Blake { /* end of list */ } 2171c1bb86cdSEric Blake } 2172c1bb86cdSEric Blake }; 2173c1bb86cdSEric Blake 2174*244a5668SFam Zheng static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, 2175*244a5668SFam Zheng Error **errp) 2176*244a5668SFam Zheng { 2177*244a5668SFam Zheng return raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); 2178*244a5668SFam Zheng } 2179*244a5668SFam Zheng 2180*244a5668SFam Zheng static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) 2181*244a5668SFam Zheng { 2182*244a5668SFam Zheng BDRVRawState *s = bs->opaque; 2183*244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); 2184*244a5668SFam Zheng s->perm = perm; 2185*244a5668SFam Zheng s->shared_perm = shared; 2186*244a5668SFam Zheng } 2187*244a5668SFam Zheng 2188*244a5668SFam Zheng static void raw_abort_perm_update(BlockDriverState *bs) 2189*244a5668SFam Zheng { 2190*244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 2191*244a5668SFam Zheng } 2192*244a5668SFam Zheng 2193*244a5668SFam Zheng static int raw_inactivate(BlockDriverState *bs) 2194*244a5668SFam Zheng { 2195*244a5668SFam Zheng int ret; 2196*244a5668SFam Zheng uint64_t perm = 0; 2197*244a5668SFam Zheng uint64_t shared = BLK_PERM_ALL; 2198*244a5668SFam Zheng 2199*244a5668SFam Zheng ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, NULL); 2200*244a5668SFam Zheng if (ret) { 2201*244a5668SFam Zheng return ret; 2202*244a5668SFam Zheng } 2203*244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); 2204*244a5668SFam Zheng return 0; 2205*244a5668SFam Zheng } 2206*244a5668SFam Zheng 2207*244a5668SFam Zheng 2208*244a5668SFam Zheng static void raw_invalidate_cache(BlockDriverState *bs, Error **errp) 2209*244a5668SFam Zheng { 2210*244a5668SFam Zheng BDRVRawState *s = bs->opaque; 2211*244a5668SFam Zheng int ret; 2212*244a5668SFam Zheng 2213*244a5668SFam Zheng assert(!(bdrv_get_flags(bs) & BDRV_O_INACTIVE)); 2214*244a5668SFam Zheng ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, s->perm, s->shared_perm, 2215*244a5668SFam Zheng errp); 2216*244a5668SFam Zheng if (ret) { 2217*244a5668SFam Zheng return; 2218*244a5668SFam Zheng } 2219*244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_COMMIT, s->perm, s->shared_perm, NULL); 2220*244a5668SFam Zheng } 2221*244a5668SFam Zheng 2222c1bb86cdSEric Blake BlockDriver bdrv_file = { 2223c1bb86cdSEric Blake .format_name = "file", 2224c1bb86cdSEric Blake .protocol_name = "file", 2225c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 2226c1bb86cdSEric Blake .bdrv_needs_filename = true, 2227c1bb86cdSEric Blake .bdrv_probe = NULL, /* no probe for protocols */ 2228c1bb86cdSEric Blake .bdrv_parse_filename = raw_parse_filename, 2229c1bb86cdSEric Blake .bdrv_file_open = raw_open, 2230c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 2231c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 2232c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 2233c1bb86cdSEric Blake .bdrv_close = raw_close, 2234c1bb86cdSEric Blake .bdrv_create = raw_create, 2235c1bb86cdSEric Blake .bdrv_has_zero_init = bdrv_has_zero_init_1, 2236c1bb86cdSEric Blake .bdrv_co_get_block_status = raw_co_get_block_status, 2237c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, 2238c1bb86cdSEric Blake 2239c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 2240c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 2241c1bb86cdSEric Blake .bdrv_aio_flush = raw_aio_flush, 2242c1bb86cdSEric Blake .bdrv_aio_pdiscard = raw_aio_pdiscard, 2243c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 2244c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 2245c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 2246c1bb86cdSEric Blake 2247c1bb86cdSEric Blake .bdrv_truncate = raw_truncate, 2248c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 2249c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 2250c1bb86cdSEric Blake .bdrv_get_allocated_file_size 2251c1bb86cdSEric Blake = raw_get_allocated_file_size, 2252*244a5668SFam Zheng .bdrv_inactivate = raw_inactivate, 2253*244a5668SFam Zheng .bdrv_invalidate_cache = raw_invalidate_cache, 2254*244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 2255*244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 2256*244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 2257c1bb86cdSEric Blake .create_opts = &raw_create_opts, 2258c1bb86cdSEric Blake }; 2259c1bb86cdSEric Blake 2260c1bb86cdSEric Blake /***********************************************/ 2261c1bb86cdSEric Blake /* host device */ 2262c1bb86cdSEric Blake 2263c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2264c1bb86cdSEric Blake static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 2265c1bb86cdSEric Blake CFIndex maxPathSize, int flags); 2266c1bb86cdSEric Blake static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) 2267c1bb86cdSEric Blake { 2268c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 2269c1bb86cdSEric Blake mach_port_t masterPort; 2270c1bb86cdSEric Blake CFMutableDictionaryRef classesToMatch; 2271c1bb86cdSEric Blake const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; 2272c1bb86cdSEric Blake char *mediaType = NULL; 2273c1bb86cdSEric Blake 2274c1bb86cdSEric Blake kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); 2275c1bb86cdSEric Blake if ( KERN_SUCCESS != kernResult ) { 2276c1bb86cdSEric Blake printf( "IOMasterPort returned %d\n", kernResult ); 2277c1bb86cdSEric Blake } 2278c1bb86cdSEric Blake 2279c1bb86cdSEric Blake int index; 2280c1bb86cdSEric Blake for (index = 0; index < ARRAY_SIZE(matching_array); index++) { 2281c1bb86cdSEric Blake classesToMatch = IOServiceMatching(matching_array[index]); 2282c1bb86cdSEric Blake if (classesToMatch == NULL) { 2283c1bb86cdSEric Blake error_report("IOServiceMatching returned NULL for %s", 2284c1bb86cdSEric Blake matching_array[index]); 2285c1bb86cdSEric Blake continue; 2286c1bb86cdSEric Blake } 2287c1bb86cdSEric Blake CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), 2288c1bb86cdSEric Blake kCFBooleanTrue); 2289c1bb86cdSEric Blake kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, 2290c1bb86cdSEric Blake mediaIterator); 2291c1bb86cdSEric Blake if (kernResult != KERN_SUCCESS) { 2292c1bb86cdSEric Blake error_report("Note: IOServiceGetMatchingServices returned %d", 2293c1bb86cdSEric Blake kernResult); 2294c1bb86cdSEric Blake continue; 2295c1bb86cdSEric Blake } 2296c1bb86cdSEric Blake 2297c1bb86cdSEric Blake /* If a match was found, leave the loop */ 2298c1bb86cdSEric Blake if (*mediaIterator != 0) { 2299c1bb86cdSEric Blake DPRINTF("Matching using %s\n", matching_array[index]); 2300c1bb86cdSEric Blake mediaType = g_strdup(matching_array[index]); 2301c1bb86cdSEric Blake break; 2302c1bb86cdSEric Blake } 2303c1bb86cdSEric Blake } 2304c1bb86cdSEric Blake return mediaType; 2305c1bb86cdSEric Blake } 2306c1bb86cdSEric Blake 2307c1bb86cdSEric Blake kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 2308c1bb86cdSEric Blake CFIndex maxPathSize, int flags) 2309c1bb86cdSEric Blake { 2310c1bb86cdSEric Blake io_object_t nextMedia; 2311c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 2312c1bb86cdSEric Blake *bsdPath = '\0'; 2313c1bb86cdSEric Blake nextMedia = IOIteratorNext( mediaIterator ); 2314c1bb86cdSEric Blake if ( nextMedia ) 2315c1bb86cdSEric Blake { 2316c1bb86cdSEric Blake CFTypeRef bsdPathAsCFString; 2317c1bb86cdSEric Blake bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); 2318c1bb86cdSEric Blake if ( bsdPathAsCFString ) { 2319c1bb86cdSEric Blake size_t devPathLength; 2320c1bb86cdSEric Blake strcpy( bsdPath, _PATH_DEV ); 2321c1bb86cdSEric Blake if (flags & BDRV_O_NOCACHE) { 2322c1bb86cdSEric Blake strcat(bsdPath, "r"); 2323c1bb86cdSEric Blake } 2324c1bb86cdSEric Blake devPathLength = strlen( bsdPath ); 2325c1bb86cdSEric Blake if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { 2326c1bb86cdSEric Blake kernResult = KERN_SUCCESS; 2327c1bb86cdSEric Blake } 2328c1bb86cdSEric Blake CFRelease( bsdPathAsCFString ); 2329c1bb86cdSEric Blake } 2330c1bb86cdSEric Blake IOObjectRelease( nextMedia ); 2331c1bb86cdSEric Blake } 2332c1bb86cdSEric Blake 2333c1bb86cdSEric Blake return kernResult; 2334c1bb86cdSEric Blake } 2335c1bb86cdSEric Blake 2336c1bb86cdSEric Blake /* Sets up a real cdrom for use in QEMU */ 2337c1bb86cdSEric Blake static bool setup_cdrom(char *bsd_path, Error **errp) 2338c1bb86cdSEric Blake { 2339c1bb86cdSEric Blake int index, num_of_test_partitions = 2, fd; 2340c1bb86cdSEric Blake char test_partition[MAXPATHLEN]; 2341c1bb86cdSEric Blake bool partition_found = false; 2342c1bb86cdSEric Blake 2343c1bb86cdSEric Blake /* look for a working partition */ 2344c1bb86cdSEric Blake for (index = 0; index < num_of_test_partitions; index++) { 2345c1bb86cdSEric Blake snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, 2346c1bb86cdSEric Blake index); 2347c1bb86cdSEric Blake fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE); 2348c1bb86cdSEric Blake if (fd >= 0) { 2349c1bb86cdSEric Blake partition_found = true; 2350c1bb86cdSEric Blake qemu_close(fd); 2351c1bb86cdSEric Blake break; 2352c1bb86cdSEric Blake } 2353c1bb86cdSEric Blake } 2354c1bb86cdSEric Blake 2355c1bb86cdSEric Blake /* if a working partition on the device was not found */ 2356c1bb86cdSEric Blake if (partition_found == false) { 2357c1bb86cdSEric Blake error_setg(errp, "Failed to find a working partition on disc"); 2358c1bb86cdSEric Blake } else { 2359c1bb86cdSEric Blake DPRINTF("Using %s as optical disc\n", test_partition); 2360c1bb86cdSEric Blake pstrcpy(bsd_path, MAXPATHLEN, test_partition); 2361c1bb86cdSEric Blake } 2362c1bb86cdSEric Blake return partition_found; 2363c1bb86cdSEric Blake } 2364c1bb86cdSEric Blake 2365c1bb86cdSEric Blake /* Prints directions on mounting and unmounting a device */ 2366c1bb86cdSEric Blake static void print_unmounting_directions(const char *file_name) 2367c1bb86cdSEric Blake { 2368c1bb86cdSEric Blake error_report("If device %s is mounted on the desktop, unmount" 2369c1bb86cdSEric Blake " it first before using it in QEMU", file_name); 2370c1bb86cdSEric Blake error_report("Command to unmount device: diskutil unmountDisk %s", 2371c1bb86cdSEric Blake file_name); 2372c1bb86cdSEric Blake error_report("Command to mount device: diskutil mountDisk %s", file_name); 2373c1bb86cdSEric Blake } 2374c1bb86cdSEric Blake 2375c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 2376c1bb86cdSEric Blake 2377c1bb86cdSEric Blake static int hdev_probe_device(const char *filename) 2378c1bb86cdSEric Blake { 2379c1bb86cdSEric Blake struct stat st; 2380c1bb86cdSEric Blake 2381c1bb86cdSEric Blake /* allow a dedicated CD-ROM driver to match with a higher priority */ 2382c1bb86cdSEric Blake if (strstart(filename, "/dev/cdrom", NULL)) 2383c1bb86cdSEric Blake return 50; 2384c1bb86cdSEric Blake 2385c1bb86cdSEric Blake if (stat(filename, &st) >= 0 && 2386c1bb86cdSEric Blake (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 2387c1bb86cdSEric Blake return 100; 2388c1bb86cdSEric Blake } 2389c1bb86cdSEric Blake 2390c1bb86cdSEric Blake return 0; 2391c1bb86cdSEric Blake } 2392c1bb86cdSEric Blake 2393c1bb86cdSEric Blake static int check_hdev_writable(BDRVRawState *s) 2394c1bb86cdSEric Blake { 2395c1bb86cdSEric Blake #if defined(BLKROGET) 2396c1bb86cdSEric Blake /* Linux block devices can be configured "read-only" using blockdev(8). 2397c1bb86cdSEric Blake * This is independent of device node permissions and therefore open(2) 2398c1bb86cdSEric Blake * with O_RDWR succeeds. Actual writes fail with EPERM. 2399c1bb86cdSEric Blake * 2400c1bb86cdSEric Blake * bdrv_open() is supposed to fail if the disk is read-only. Explicitly 2401c1bb86cdSEric Blake * check for read-only block devices so that Linux block devices behave 2402c1bb86cdSEric Blake * properly. 2403c1bb86cdSEric Blake */ 2404c1bb86cdSEric Blake struct stat st; 2405c1bb86cdSEric Blake int readonly = 0; 2406c1bb86cdSEric Blake 2407c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 2408c1bb86cdSEric Blake return -errno; 2409c1bb86cdSEric Blake } 2410c1bb86cdSEric Blake 2411c1bb86cdSEric Blake if (!S_ISBLK(st.st_mode)) { 2412c1bb86cdSEric Blake return 0; 2413c1bb86cdSEric Blake } 2414c1bb86cdSEric Blake 2415c1bb86cdSEric Blake if (ioctl(s->fd, BLKROGET, &readonly) < 0) { 2416c1bb86cdSEric Blake return -errno; 2417c1bb86cdSEric Blake } 2418c1bb86cdSEric Blake 2419c1bb86cdSEric Blake if (readonly) { 2420c1bb86cdSEric Blake return -EACCES; 2421c1bb86cdSEric Blake } 2422c1bb86cdSEric Blake #endif /* defined(BLKROGET) */ 2423c1bb86cdSEric Blake return 0; 2424c1bb86cdSEric Blake } 2425c1bb86cdSEric Blake 2426c1bb86cdSEric Blake static void hdev_parse_filename(const char *filename, QDict *options, 2427c1bb86cdSEric Blake Error **errp) 2428c1bb86cdSEric Blake { 2429c1bb86cdSEric Blake /* The prefix is optional, just as for "file". */ 2430c1bb86cdSEric Blake strstart(filename, "host_device:", &filename); 2431c1bb86cdSEric Blake 243246f5ac20SEric Blake qdict_put_str(options, "filename", filename); 2433c1bb86cdSEric Blake } 2434c1bb86cdSEric Blake 2435c1bb86cdSEric Blake static bool hdev_is_sg(BlockDriverState *bs) 2436c1bb86cdSEric Blake { 2437c1bb86cdSEric Blake 2438c1bb86cdSEric Blake #if defined(__linux__) 2439c1bb86cdSEric Blake 2440c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2441c1bb86cdSEric Blake struct stat st; 2442c1bb86cdSEric Blake struct sg_scsi_id scsiid; 2443c1bb86cdSEric Blake int sg_version; 2444c1bb86cdSEric Blake int ret; 2445c1bb86cdSEric Blake 2446c1bb86cdSEric Blake if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { 2447c1bb86cdSEric Blake return false; 2448c1bb86cdSEric Blake } 2449c1bb86cdSEric Blake 2450c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); 2451c1bb86cdSEric Blake if (ret < 0) { 2452c1bb86cdSEric Blake return false; 2453c1bb86cdSEric Blake } 2454c1bb86cdSEric Blake 2455c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); 2456c1bb86cdSEric Blake if (ret >= 0) { 2457c1bb86cdSEric Blake DPRINTF("SG device found: type=%d, version=%d\n", 2458c1bb86cdSEric Blake scsiid.scsi_type, sg_version); 2459c1bb86cdSEric Blake return true; 2460c1bb86cdSEric Blake } 2461c1bb86cdSEric Blake 2462c1bb86cdSEric Blake #endif 2463c1bb86cdSEric Blake 2464c1bb86cdSEric Blake return false; 2465c1bb86cdSEric Blake } 2466c1bb86cdSEric Blake 2467c1bb86cdSEric Blake static int hdev_open(BlockDriverState *bs, QDict *options, int flags, 2468c1bb86cdSEric Blake Error **errp) 2469c1bb86cdSEric Blake { 2470c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2471c1bb86cdSEric Blake Error *local_err = NULL; 2472c1bb86cdSEric Blake int ret; 2473c1bb86cdSEric Blake 2474c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2475129c7d1cSMarkus Armbruster /* 2476129c7d1cSMarkus Armbruster * Caution: while qdict_get_str() is fine, getting non-string types 2477129c7d1cSMarkus Armbruster * would require more care. When @options come from -blockdev or 2478129c7d1cSMarkus Armbruster * blockdev_add, its members are typed according to the QAPI 2479129c7d1cSMarkus Armbruster * schema, but when they come from -drive, they're all QString. 2480129c7d1cSMarkus Armbruster */ 2481c1bb86cdSEric Blake const char *filename = qdict_get_str(options, "filename"); 2482c1bb86cdSEric Blake char bsd_path[MAXPATHLEN] = ""; 2483c1bb86cdSEric Blake bool error_occurred = false; 2484c1bb86cdSEric Blake 2485c1bb86cdSEric Blake /* If using a real cdrom */ 2486c1bb86cdSEric Blake if (strcmp(filename, "/dev/cdrom") == 0) { 2487c1bb86cdSEric Blake char *mediaType = NULL; 2488c1bb86cdSEric Blake kern_return_t ret_val; 2489c1bb86cdSEric Blake io_iterator_t mediaIterator = 0; 2490c1bb86cdSEric Blake 2491c1bb86cdSEric Blake mediaType = FindEjectableOpticalMedia(&mediaIterator); 2492c1bb86cdSEric Blake if (mediaType == NULL) { 2493c1bb86cdSEric Blake error_setg(errp, "Please make sure your CD/DVD is in the optical" 2494c1bb86cdSEric Blake " drive"); 2495c1bb86cdSEric Blake error_occurred = true; 2496c1bb86cdSEric Blake goto hdev_open_Mac_error; 2497c1bb86cdSEric Blake } 2498c1bb86cdSEric Blake 2499c1bb86cdSEric Blake ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); 2500c1bb86cdSEric Blake if (ret_val != KERN_SUCCESS) { 2501c1bb86cdSEric Blake error_setg(errp, "Could not get BSD path for optical drive"); 2502c1bb86cdSEric Blake error_occurred = true; 2503c1bb86cdSEric Blake goto hdev_open_Mac_error; 2504c1bb86cdSEric Blake } 2505c1bb86cdSEric Blake 2506c1bb86cdSEric Blake /* If a real optical drive was not found */ 2507c1bb86cdSEric Blake if (bsd_path[0] == '\0') { 2508c1bb86cdSEric Blake error_setg(errp, "Failed to obtain bsd path for optical drive"); 2509c1bb86cdSEric Blake error_occurred = true; 2510c1bb86cdSEric Blake goto hdev_open_Mac_error; 2511c1bb86cdSEric Blake } 2512c1bb86cdSEric Blake 2513c1bb86cdSEric Blake /* If using a cdrom disc and finding a partition on the disc failed */ 2514c1bb86cdSEric Blake if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && 2515c1bb86cdSEric Blake setup_cdrom(bsd_path, errp) == false) { 2516c1bb86cdSEric Blake print_unmounting_directions(bsd_path); 2517c1bb86cdSEric Blake error_occurred = true; 2518c1bb86cdSEric Blake goto hdev_open_Mac_error; 2519c1bb86cdSEric Blake } 2520c1bb86cdSEric Blake 252146f5ac20SEric Blake qdict_put_str(options, "filename", bsd_path); 2522c1bb86cdSEric Blake 2523c1bb86cdSEric Blake hdev_open_Mac_error: 2524c1bb86cdSEric Blake g_free(mediaType); 2525c1bb86cdSEric Blake if (mediaIterator) { 2526c1bb86cdSEric Blake IOObjectRelease(mediaIterator); 2527c1bb86cdSEric Blake } 2528c1bb86cdSEric Blake if (error_occurred) { 2529c1bb86cdSEric Blake return -ENOENT; 2530c1bb86cdSEric Blake } 2531c1bb86cdSEric Blake } 2532c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 2533c1bb86cdSEric Blake 2534c1bb86cdSEric Blake s->type = FTYPE_FILE; 2535c1bb86cdSEric Blake 2536c1bb86cdSEric Blake ret = raw_open_common(bs, options, flags, 0, &local_err); 2537c1bb86cdSEric Blake if (ret < 0) { 2538c1bb86cdSEric Blake error_propagate(errp, local_err); 2539c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2540c1bb86cdSEric Blake if (*bsd_path) { 2541c1bb86cdSEric Blake filename = bsd_path; 2542c1bb86cdSEric Blake } 2543c1bb86cdSEric Blake /* if a physical device experienced an error while being opened */ 2544c1bb86cdSEric Blake if (strncmp(filename, "/dev/", 5) == 0) { 2545c1bb86cdSEric Blake print_unmounting_directions(filename); 2546c1bb86cdSEric Blake } 2547c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 2548c1bb86cdSEric Blake return ret; 2549c1bb86cdSEric Blake } 2550c1bb86cdSEric Blake 2551c1bb86cdSEric Blake /* Since this does ioctl the device must be already opened */ 2552c1bb86cdSEric Blake bs->sg = hdev_is_sg(bs); 2553c1bb86cdSEric Blake 2554c1bb86cdSEric Blake if (flags & BDRV_O_RDWR) { 2555c1bb86cdSEric Blake ret = check_hdev_writable(s); 2556c1bb86cdSEric Blake if (ret < 0) { 2557c1bb86cdSEric Blake raw_close(bs); 2558c1bb86cdSEric Blake error_setg_errno(errp, -ret, "The device is not writable"); 2559c1bb86cdSEric Blake return ret; 2560c1bb86cdSEric Blake } 2561c1bb86cdSEric Blake } 2562c1bb86cdSEric Blake 2563c1bb86cdSEric Blake return ret; 2564c1bb86cdSEric Blake } 2565c1bb86cdSEric Blake 2566c1bb86cdSEric Blake #if defined(__linux__) 2567c1bb86cdSEric Blake 2568c1bb86cdSEric Blake static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs, 2569c1bb86cdSEric Blake unsigned long int req, void *buf, 2570c1bb86cdSEric Blake BlockCompletionFunc *cb, void *opaque) 2571c1bb86cdSEric Blake { 2572c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2573c1bb86cdSEric Blake RawPosixAIOData *acb; 2574c1bb86cdSEric Blake ThreadPool *pool; 2575c1bb86cdSEric Blake 2576c1bb86cdSEric Blake if (fd_open(bs) < 0) 2577c1bb86cdSEric Blake return NULL; 2578c1bb86cdSEric Blake 2579c1bb86cdSEric Blake acb = g_new(RawPosixAIOData, 1); 2580c1bb86cdSEric Blake acb->bs = bs; 2581c1bb86cdSEric Blake acb->aio_type = QEMU_AIO_IOCTL; 2582c1bb86cdSEric Blake acb->aio_fildes = s->fd; 2583c1bb86cdSEric Blake acb->aio_offset = 0; 2584c1bb86cdSEric Blake acb->aio_ioctl_buf = buf; 2585c1bb86cdSEric Blake acb->aio_ioctl_cmd = req; 2586c1bb86cdSEric Blake pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 2587c1bb86cdSEric Blake return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque); 2588c1bb86cdSEric Blake } 2589c1bb86cdSEric Blake #endif /* linux */ 2590c1bb86cdSEric Blake 2591c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs) 2592c1bb86cdSEric Blake { 2593c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2594c1bb86cdSEric Blake 2595c1bb86cdSEric Blake /* this is just to ensure s->fd is sane (its called by io ops) */ 2596c1bb86cdSEric Blake if (s->fd >= 0) 2597c1bb86cdSEric Blake return 0; 2598c1bb86cdSEric Blake return -EIO; 2599c1bb86cdSEric Blake } 2600c1bb86cdSEric Blake 2601c1bb86cdSEric Blake static coroutine_fn BlockAIOCB *hdev_aio_pdiscard(BlockDriverState *bs, 2602c1bb86cdSEric Blake int64_t offset, int count, 2603c1bb86cdSEric Blake BlockCompletionFunc *cb, void *opaque) 2604c1bb86cdSEric Blake { 2605c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2606c1bb86cdSEric Blake 2607c1bb86cdSEric Blake if (fd_open(bs) < 0) { 2608c1bb86cdSEric Blake return NULL; 2609c1bb86cdSEric Blake } 2610c1bb86cdSEric Blake return paio_submit(bs, s->fd, offset, NULL, count, 2611c1bb86cdSEric Blake cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV); 2612c1bb86cdSEric Blake } 2613c1bb86cdSEric Blake 2614c1bb86cdSEric Blake static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, 2615c1bb86cdSEric Blake int64_t offset, int count, BdrvRequestFlags flags) 2616c1bb86cdSEric Blake { 2617c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2618c1bb86cdSEric Blake int rc; 2619c1bb86cdSEric Blake 2620c1bb86cdSEric Blake rc = fd_open(bs); 2621c1bb86cdSEric Blake if (rc < 0) { 2622c1bb86cdSEric Blake return rc; 2623c1bb86cdSEric Blake } 2624c1bb86cdSEric Blake if (!(flags & BDRV_REQ_MAY_UNMAP)) { 2625c1bb86cdSEric Blake return paio_submit_co(bs, s->fd, offset, NULL, count, 2626c1bb86cdSEric Blake QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV); 2627c1bb86cdSEric Blake } else if (s->discard_zeroes) { 2628c1bb86cdSEric Blake return paio_submit_co(bs, s->fd, offset, NULL, count, 2629c1bb86cdSEric Blake QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV); 2630c1bb86cdSEric Blake } 2631c1bb86cdSEric Blake return -ENOTSUP; 2632c1bb86cdSEric Blake } 2633c1bb86cdSEric Blake 2634c1bb86cdSEric Blake static int hdev_create(const char *filename, QemuOpts *opts, 2635c1bb86cdSEric Blake Error **errp) 2636c1bb86cdSEric Blake { 2637c1bb86cdSEric Blake int fd; 2638c1bb86cdSEric Blake int ret = 0; 2639c1bb86cdSEric Blake struct stat stat_buf; 2640c1bb86cdSEric Blake int64_t total_size = 0; 2641c1bb86cdSEric Blake bool has_prefix; 2642c1bb86cdSEric Blake 2643c1bb86cdSEric Blake /* This function is used by both protocol block drivers and therefore either 2644c1bb86cdSEric Blake * of these prefixes may be given. 2645c1bb86cdSEric Blake * The return value has to be stored somewhere, otherwise this is an error 2646c1bb86cdSEric Blake * due to -Werror=unused-value. */ 2647c1bb86cdSEric Blake has_prefix = 2648c1bb86cdSEric Blake strstart(filename, "host_device:", &filename) || 2649c1bb86cdSEric Blake strstart(filename, "host_cdrom:" , &filename); 2650c1bb86cdSEric Blake 2651c1bb86cdSEric Blake (void)has_prefix; 2652c1bb86cdSEric Blake 2653c1bb86cdSEric Blake ret = raw_normalize_devicepath(&filename); 2654c1bb86cdSEric Blake if (ret < 0) { 2655c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not normalize device path"); 2656c1bb86cdSEric Blake return ret; 2657c1bb86cdSEric Blake } 2658c1bb86cdSEric Blake 2659c1bb86cdSEric Blake /* Read out options */ 2660c1bb86cdSEric Blake total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2661c1bb86cdSEric Blake BDRV_SECTOR_SIZE); 2662c1bb86cdSEric Blake 2663c1bb86cdSEric Blake fd = qemu_open(filename, O_WRONLY | O_BINARY); 2664c1bb86cdSEric Blake if (fd < 0) { 2665c1bb86cdSEric Blake ret = -errno; 2666c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not open device"); 2667c1bb86cdSEric Blake return ret; 2668c1bb86cdSEric Blake } 2669c1bb86cdSEric Blake 2670c1bb86cdSEric Blake if (fstat(fd, &stat_buf) < 0) { 2671c1bb86cdSEric Blake ret = -errno; 2672c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not stat device"); 2673c1bb86cdSEric Blake } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) { 2674c1bb86cdSEric Blake error_setg(errp, 2675c1bb86cdSEric Blake "The given file is neither a block nor a character device"); 2676c1bb86cdSEric Blake ret = -ENODEV; 2677c1bb86cdSEric Blake } else if (lseek(fd, 0, SEEK_END) < total_size) { 2678c1bb86cdSEric Blake error_setg(errp, "Device is too small"); 2679c1bb86cdSEric Blake ret = -ENOSPC; 2680c1bb86cdSEric Blake } 2681c1bb86cdSEric Blake 2682c1bb86cdSEric Blake qemu_close(fd); 2683c1bb86cdSEric Blake return ret; 2684c1bb86cdSEric Blake } 2685c1bb86cdSEric Blake 2686c1bb86cdSEric Blake static BlockDriver bdrv_host_device = { 2687c1bb86cdSEric Blake .format_name = "host_device", 2688c1bb86cdSEric Blake .protocol_name = "host_device", 2689c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 2690c1bb86cdSEric Blake .bdrv_needs_filename = true, 2691c1bb86cdSEric Blake .bdrv_probe_device = hdev_probe_device, 2692c1bb86cdSEric Blake .bdrv_parse_filename = hdev_parse_filename, 2693c1bb86cdSEric Blake .bdrv_file_open = hdev_open, 2694c1bb86cdSEric Blake .bdrv_close = raw_close, 2695c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 2696c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 2697c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 2698c1bb86cdSEric Blake .bdrv_create = hdev_create, 2699c1bb86cdSEric Blake .create_opts = &raw_create_opts, 2700c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, 2701c1bb86cdSEric Blake 2702c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 2703c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 2704c1bb86cdSEric Blake .bdrv_aio_flush = raw_aio_flush, 2705c1bb86cdSEric Blake .bdrv_aio_pdiscard = hdev_aio_pdiscard, 2706c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 2707c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 2708c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 2709c1bb86cdSEric Blake 2710c1bb86cdSEric Blake .bdrv_truncate = raw_truncate, 2711c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 2712c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 2713c1bb86cdSEric Blake .bdrv_get_allocated_file_size 2714c1bb86cdSEric Blake = raw_get_allocated_file_size, 2715*244a5668SFam Zheng .bdrv_inactivate = raw_inactivate, 2716*244a5668SFam Zheng .bdrv_invalidate_cache = raw_invalidate_cache, 2717*244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 2718*244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 2719*244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 2720c1bb86cdSEric Blake .bdrv_probe_blocksizes = hdev_probe_blocksizes, 2721c1bb86cdSEric Blake .bdrv_probe_geometry = hdev_probe_geometry, 2722c1bb86cdSEric Blake 2723c1bb86cdSEric Blake /* generic scsi device */ 2724c1bb86cdSEric Blake #ifdef __linux__ 2725c1bb86cdSEric Blake .bdrv_aio_ioctl = hdev_aio_ioctl, 2726c1bb86cdSEric Blake #endif 2727c1bb86cdSEric Blake }; 2728c1bb86cdSEric Blake 2729c1bb86cdSEric Blake #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 2730c1bb86cdSEric Blake static void cdrom_parse_filename(const char *filename, QDict *options, 2731c1bb86cdSEric Blake Error **errp) 2732c1bb86cdSEric Blake { 2733c1bb86cdSEric Blake /* The prefix is optional, just as for "file". */ 2734c1bb86cdSEric Blake strstart(filename, "host_cdrom:", &filename); 2735c1bb86cdSEric Blake 273646f5ac20SEric Blake qdict_put_str(options, "filename", filename); 2737c1bb86cdSEric Blake } 2738c1bb86cdSEric Blake #endif 2739c1bb86cdSEric Blake 2740c1bb86cdSEric Blake #ifdef __linux__ 2741c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 2742c1bb86cdSEric Blake Error **errp) 2743c1bb86cdSEric Blake { 2744c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2745c1bb86cdSEric Blake 2746c1bb86cdSEric Blake s->type = FTYPE_CD; 2747c1bb86cdSEric Blake 2748c1bb86cdSEric Blake /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ 2749c1bb86cdSEric Blake return raw_open_common(bs, options, flags, O_NONBLOCK, errp); 2750c1bb86cdSEric Blake } 2751c1bb86cdSEric Blake 2752c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 2753c1bb86cdSEric Blake { 2754c1bb86cdSEric Blake int fd, ret; 2755c1bb86cdSEric Blake int prio = 0; 2756c1bb86cdSEric Blake struct stat st; 2757c1bb86cdSEric Blake 2758c1bb86cdSEric Blake fd = qemu_open(filename, O_RDONLY | O_NONBLOCK); 2759c1bb86cdSEric Blake if (fd < 0) { 2760c1bb86cdSEric Blake goto out; 2761c1bb86cdSEric Blake } 2762c1bb86cdSEric Blake ret = fstat(fd, &st); 2763c1bb86cdSEric Blake if (ret == -1 || !S_ISBLK(st.st_mode)) { 2764c1bb86cdSEric Blake goto outc; 2765c1bb86cdSEric Blake } 2766c1bb86cdSEric Blake 2767c1bb86cdSEric Blake /* Attempt to detect via a CDROM specific ioctl */ 2768c1bb86cdSEric Blake ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 2769c1bb86cdSEric Blake if (ret >= 0) 2770c1bb86cdSEric Blake prio = 100; 2771c1bb86cdSEric Blake 2772c1bb86cdSEric Blake outc: 2773c1bb86cdSEric Blake qemu_close(fd); 2774c1bb86cdSEric Blake out: 2775c1bb86cdSEric Blake return prio; 2776c1bb86cdSEric Blake } 2777c1bb86cdSEric Blake 2778c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 2779c1bb86cdSEric Blake { 2780c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2781c1bb86cdSEric Blake int ret; 2782c1bb86cdSEric Blake 2783c1bb86cdSEric Blake ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 2784c1bb86cdSEric Blake return ret == CDS_DISC_OK; 2785c1bb86cdSEric Blake } 2786c1bb86cdSEric Blake 2787c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 2788c1bb86cdSEric Blake { 2789c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2790c1bb86cdSEric Blake 2791c1bb86cdSEric Blake if (eject_flag) { 2792c1bb86cdSEric Blake if (ioctl(s->fd, CDROMEJECT, NULL) < 0) 2793c1bb86cdSEric Blake perror("CDROMEJECT"); 2794c1bb86cdSEric Blake } else { 2795c1bb86cdSEric Blake if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) 2796c1bb86cdSEric Blake perror("CDROMEJECT"); 2797c1bb86cdSEric Blake } 2798c1bb86cdSEric Blake } 2799c1bb86cdSEric Blake 2800c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 2801c1bb86cdSEric Blake { 2802c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2803c1bb86cdSEric Blake 2804c1bb86cdSEric Blake if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { 2805c1bb86cdSEric Blake /* 2806c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 2807c1bb86cdSEric Blake * mounts the CD-ROM 2808c1bb86cdSEric Blake */ 2809c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 2810c1bb86cdSEric Blake } 2811c1bb86cdSEric Blake } 2812c1bb86cdSEric Blake 2813c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 2814c1bb86cdSEric Blake .format_name = "host_cdrom", 2815c1bb86cdSEric Blake .protocol_name = "host_cdrom", 2816c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 2817c1bb86cdSEric Blake .bdrv_needs_filename = true, 2818c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 2819c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 2820c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 2821c1bb86cdSEric Blake .bdrv_close = raw_close, 2822c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 2823c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 2824c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 2825c1bb86cdSEric Blake .bdrv_create = hdev_create, 2826c1bb86cdSEric Blake .create_opts = &raw_create_opts, 2827c1bb86cdSEric Blake 2828c1bb86cdSEric Blake 2829c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 2830c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 2831c1bb86cdSEric Blake .bdrv_aio_flush = raw_aio_flush, 2832c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 2833c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 2834c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 2835c1bb86cdSEric Blake 2836c1bb86cdSEric Blake .bdrv_truncate = raw_truncate, 2837c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 2838c1bb86cdSEric Blake .has_variable_length = true, 2839c1bb86cdSEric Blake .bdrv_get_allocated_file_size 2840c1bb86cdSEric Blake = raw_get_allocated_file_size, 2841c1bb86cdSEric Blake 2842c1bb86cdSEric Blake /* removable device support */ 2843c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 2844c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 2845c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 2846c1bb86cdSEric Blake 2847c1bb86cdSEric Blake /* generic scsi device */ 2848c1bb86cdSEric Blake .bdrv_aio_ioctl = hdev_aio_ioctl, 2849c1bb86cdSEric Blake }; 2850c1bb86cdSEric Blake #endif /* __linux__ */ 2851c1bb86cdSEric Blake 2852c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2853c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 2854c1bb86cdSEric Blake Error **errp) 2855c1bb86cdSEric Blake { 2856c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2857c1bb86cdSEric Blake Error *local_err = NULL; 2858c1bb86cdSEric Blake int ret; 2859c1bb86cdSEric Blake 2860c1bb86cdSEric Blake s->type = FTYPE_CD; 2861c1bb86cdSEric Blake 2862c1bb86cdSEric Blake ret = raw_open_common(bs, options, flags, 0, &local_err); 2863c1bb86cdSEric Blake if (ret) { 2864c1bb86cdSEric Blake error_propagate(errp, local_err); 2865c1bb86cdSEric Blake return ret; 2866c1bb86cdSEric Blake } 2867c1bb86cdSEric Blake 2868c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 2869c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 2870c1bb86cdSEric Blake return 0; 2871c1bb86cdSEric Blake } 2872c1bb86cdSEric Blake 2873c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 2874c1bb86cdSEric Blake { 2875c1bb86cdSEric Blake if (strstart(filename, "/dev/cd", NULL) || 2876c1bb86cdSEric Blake strstart(filename, "/dev/acd", NULL)) 2877c1bb86cdSEric Blake return 100; 2878c1bb86cdSEric Blake return 0; 2879c1bb86cdSEric Blake } 2880c1bb86cdSEric Blake 2881c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs) 2882c1bb86cdSEric Blake { 2883c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2884c1bb86cdSEric Blake int fd; 2885c1bb86cdSEric Blake 2886c1bb86cdSEric Blake /* 2887c1bb86cdSEric Blake * Force reread of possibly changed/newly loaded disc, 2888c1bb86cdSEric Blake * FreeBSD seems to not notice sometimes... 2889c1bb86cdSEric Blake */ 2890c1bb86cdSEric Blake if (s->fd >= 0) 2891c1bb86cdSEric Blake qemu_close(s->fd); 2892c1bb86cdSEric Blake fd = qemu_open(bs->filename, s->open_flags, 0644); 2893c1bb86cdSEric Blake if (fd < 0) { 2894c1bb86cdSEric Blake s->fd = -1; 2895c1bb86cdSEric Blake return -EIO; 2896c1bb86cdSEric Blake } 2897c1bb86cdSEric Blake s->fd = fd; 2898c1bb86cdSEric Blake 2899c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 2900c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 2901c1bb86cdSEric Blake return 0; 2902c1bb86cdSEric Blake } 2903c1bb86cdSEric Blake 2904c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 2905c1bb86cdSEric Blake { 2906c1bb86cdSEric Blake return raw_getlength(bs) > 0; 2907c1bb86cdSEric Blake } 2908c1bb86cdSEric Blake 2909c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 2910c1bb86cdSEric Blake { 2911c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2912c1bb86cdSEric Blake 2913c1bb86cdSEric Blake if (s->fd < 0) 2914c1bb86cdSEric Blake return; 2915c1bb86cdSEric Blake 2916c1bb86cdSEric Blake (void) ioctl(s->fd, CDIOCALLOW); 2917c1bb86cdSEric Blake 2918c1bb86cdSEric Blake if (eject_flag) { 2919c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCEJECT) < 0) 2920c1bb86cdSEric Blake perror("CDIOCEJECT"); 2921c1bb86cdSEric Blake } else { 2922c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCCLOSE) < 0) 2923c1bb86cdSEric Blake perror("CDIOCCLOSE"); 2924c1bb86cdSEric Blake } 2925c1bb86cdSEric Blake 2926c1bb86cdSEric Blake cdrom_reopen(bs); 2927c1bb86cdSEric Blake } 2928c1bb86cdSEric Blake 2929c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 2930c1bb86cdSEric Blake { 2931c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2932c1bb86cdSEric Blake 2933c1bb86cdSEric Blake if (s->fd < 0) 2934c1bb86cdSEric Blake return; 2935c1bb86cdSEric Blake if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { 2936c1bb86cdSEric Blake /* 2937c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 2938c1bb86cdSEric Blake * mounts the CD-ROM 2939c1bb86cdSEric Blake */ 2940c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 2941c1bb86cdSEric Blake } 2942c1bb86cdSEric Blake } 2943c1bb86cdSEric Blake 2944c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 2945c1bb86cdSEric Blake .format_name = "host_cdrom", 2946c1bb86cdSEric Blake .protocol_name = "host_cdrom", 2947c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 2948c1bb86cdSEric Blake .bdrv_needs_filename = true, 2949c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 2950c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 2951c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 2952c1bb86cdSEric Blake .bdrv_close = raw_close, 2953c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 2954c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 2955c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 2956c1bb86cdSEric Blake .bdrv_create = hdev_create, 2957c1bb86cdSEric Blake .create_opts = &raw_create_opts, 2958c1bb86cdSEric Blake 2959c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 2960c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 2961c1bb86cdSEric Blake .bdrv_aio_flush = raw_aio_flush, 2962c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 2963c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 2964c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 2965c1bb86cdSEric Blake 2966c1bb86cdSEric Blake .bdrv_truncate = raw_truncate, 2967c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 2968c1bb86cdSEric Blake .has_variable_length = true, 2969c1bb86cdSEric Blake .bdrv_get_allocated_file_size 2970c1bb86cdSEric Blake = raw_get_allocated_file_size, 2971c1bb86cdSEric Blake 2972c1bb86cdSEric Blake /* removable device support */ 2973c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 2974c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 2975c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 2976c1bb86cdSEric Blake }; 2977c1bb86cdSEric Blake #endif /* __FreeBSD__ */ 2978c1bb86cdSEric Blake 2979c1bb86cdSEric Blake static void bdrv_file_init(void) 2980c1bb86cdSEric Blake { 2981c1bb86cdSEric Blake /* 2982c1bb86cdSEric Blake * Register all the drivers. Note that order is important, the driver 2983c1bb86cdSEric Blake * registered last will get probed first. 2984c1bb86cdSEric Blake */ 2985c1bb86cdSEric Blake bdrv_register(&bdrv_file); 2986c1bb86cdSEric Blake bdrv_register(&bdrv_host_device); 2987c1bb86cdSEric Blake #ifdef __linux__ 2988c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 2989c1bb86cdSEric Blake #endif 2990c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 2991c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 2992c1bb86cdSEric Blake #endif 2993c1bb86cdSEric Blake } 2994c1bb86cdSEric Blake 2995c1bb86cdSEric Blake block_init(bdrv_file_init); 2996