1c1bb86cdSEric Blake /* 2c1bb86cdSEric Blake * Block driver for RAW files (posix) 3c1bb86cdSEric Blake * 4c1bb86cdSEric Blake * Copyright (c) 2006 Fabrice Bellard 5c1bb86cdSEric Blake * 6c1bb86cdSEric Blake * Permission is hereby granted, free of charge, to any person obtaining a copy 7c1bb86cdSEric Blake * of this software and associated documentation files (the "Software"), to deal 8c1bb86cdSEric Blake * in the Software without restriction, including without limitation the rights 9c1bb86cdSEric Blake * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10c1bb86cdSEric Blake * copies of the Software, and to permit persons to whom the Software is 11c1bb86cdSEric Blake * furnished to do so, subject to the following conditions: 12c1bb86cdSEric Blake * 13c1bb86cdSEric Blake * The above copyright notice and this permission notice shall be included in 14c1bb86cdSEric Blake * all copies or substantial portions of the Software. 15c1bb86cdSEric Blake * 16c1bb86cdSEric Blake * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17c1bb86cdSEric Blake * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18c1bb86cdSEric Blake * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19c1bb86cdSEric Blake * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20c1bb86cdSEric Blake * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21c1bb86cdSEric Blake * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22c1bb86cdSEric Blake * THE SOFTWARE. 23c1bb86cdSEric Blake */ 24922a01a0SMarkus Armbruster 25c1bb86cdSEric Blake #include "qemu/osdep.h" 26c1bb86cdSEric Blake #include "qapi/error.h" 27c1bb86cdSEric Blake #include "qemu/cutils.h" 28c1bb86cdSEric Blake #include "qemu/error-report.h" 29c1bb86cdSEric Blake #include "block/block_int.h" 30c1bb86cdSEric Blake #include "qemu/module.h" 31922a01a0SMarkus Armbruster #include "qemu/option.h" 32c1bb86cdSEric Blake #include "trace.h" 33c1bb86cdSEric Blake #include "block/thread-pool.h" 34c1bb86cdSEric Blake #include "qemu/iov.h" 35c1bb86cdSEric Blake #include "block/raw-aio.h" 36452fcdbcSMarkus Armbruster #include "qapi/qmp/qdict.h" 37c1bb86cdSEric Blake #include "qapi/qmp/qstring.h" 38c1bb86cdSEric Blake 397c9e5276SPaolo Bonzini #include "scsi/pr-manager.h" 407c9e5276SPaolo Bonzini #include "scsi/constants.h" 417c9e5276SPaolo Bonzini 42c1bb86cdSEric Blake #if defined(__APPLE__) && (__MACH__) 43c1bb86cdSEric Blake #include <paths.h> 44c1bb86cdSEric Blake #include <sys/param.h> 45c1bb86cdSEric Blake #include <IOKit/IOKitLib.h> 46c1bb86cdSEric Blake #include <IOKit/IOBSD.h> 47c1bb86cdSEric Blake #include <IOKit/storage/IOMediaBSDClient.h> 48c1bb86cdSEric Blake #include <IOKit/storage/IOMedia.h> 49c1bb86cdSEric Blake #include <IOKit/storage/IOCDMedia.h> 50c1bb86cdSEric Blake //#include <IOKit/storage/IOCDTypes.h> 51c1bb86cdSEric Blake #include <IOKit/storage/IODVDMedia.h> 52c1bb86cdSEric Blake #include <CoreFoundation/CoreFoundation.h> 53c1bb86cdSEric Blake #endif 54c1bb86cdSEric Blake 55c1bb86cdSEric Blake #ifdef __sun__ 56c1bb86cdSEric Blake #define _POSIX_PTHREAD_SEMANTICS 1 57c1bb86cdSEric Blake #include <sys/dkio.h> 58c1bb86cdSEric Blake #endif 59c1bb86cdSEric Blake #ifdef __linux__ 60c1bb86cdSEric Blake #include <sys/ioctl.h> 61c1bb86cdSEric Blake #include <sys/param.h> 621efad060SFam Zheng #include <sys/syscall.h> 63c1bb86cdSEric Blake #include <linux/cdrom.h> 64c1bb86cdSEric Blake #include <linux/fd.h> 65c1bb86cdSEric Blake #include <linux/fs.h> 66c1bb86cdSEric Blake #include <linux/hdreg.h> 67c1bb86cdSEric Blake #include <scsi/sg.h> 68c1bb86cdSEric Blake #ifdef __s390__ 69c1bb86cdSEric Blake #include <asm/dasd.h> 70c1bb86cdSEric Blake #endif 71c1bb86cdSEric Blake #ifndef FS_NOCOW_FL 72c1bb86cdSEric Blake #define FS_NOCOW_FL 0x00800000 /* Do not cow file */ 73c1bb86cdSEric Blake #endif 74c1bb86cdSEric Blake #endif 75c1bb86cdSEric Blake #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE) 76c1bb86cdSEric Blake #include <linux/falloc.h> 77c1bb86cdSEric Blake #endif 78c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 79c1bb86cdSEric Blake #include <sys/disk.h> 80c1bb86cdSEric Blake #include <sys/cdio.h> 81c1bb86cdSEric Blake #endif 82c1bb86cdSEric Blake 83c1bb86cdSEric Blake #ifdef __OpenBSD__ 84c1bb86cdSEric Blake #include <sys/ioctl.h> 85c1bb86cdSEric Blake #include <sys/disklabel.h> 86c1bb86cdSEric Blake #include <sys/dkio.h> 87c1bb86cdSEric Blake #endif 88c1bb86cdSEric Blake 89c1bb86cdSEric Blake #ifdef __NetBSD__ 90c1bb86cdSEric Blake #include <sys/ioctl.h> 91c1bb86cdSEric Blake #include <sys/disklabel.h> 92c1bb86cdSEric Blake #include <sys/dkio.h> 93c1bb86cdSEric Blake #include <sys/disk.h> 94c1bb86cdSEric Blake #endif 95c1bb86cdSEric Blake 96c1bb86cdSEric Blake #ifdef __DragonFly__ 97c1bb86cdSEric Blake #include <sys/ioctl.h> 98c1bb86cdSEric Blake #include <sys/diskslice.h> 99c1bb86cdSEric Blake #endif 100c1bb86cdSEric Blake 101c1bb86cdSEric Blake #ifdef CONFIG_XFS 102c1bb86cdSEric Blake #include <xfs/xfs.h> 103c1bb86cdSEric Blake #endif 104c1bb86cdSEric Blake 1054f7d28d7SLaurent Vivier #include "trace.h" 106c1bb86cdSEric Blake 107c1bb86cdSEric Blake /* OS X does not have O_DSYNC */ 108c1bb86cdSEric Blake #ifndef O_DSYNC 109c1bb86cdSEric Blake #ifdef O_SYNC 110c1bb86cdSEric Blake #define O_DSYNC O_SYNC 111c1bb86cdSEric Blake #elif defined(O_FSYNC) 112c1bb86cdSEric Blake #define O_DSYNC O_FSYNC 113c1bb86cdSEric Blake #endif 114c1bb86cdSEric Blake #endif 115c1bb86cdSEric Blake 116c1bb86cdSEric Blake /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ 117c1bb86cdSEric Blake #ifndef O_DIRECT 118c1bb86cdSEric Blake #define O_DIRECT O_DSYNC 119c1bb86cdSEric Blake #endif 120c1bb86cdSEric Blake 121c1bb86cdSEric Blake #define FTYPE_FILE 0 122c1bb86cdSEric Blake #define FTYPE_CD 1 123c1bb86cdSEric Blake 124c1bb86cdSEric Blake #define MAX_BLOCKSIZE 4096 125c1bb86cdSEric Blake 126244a5668SFam Zheng /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes, 127244a5668SFam Zheng * leaving a few more bytes for its future use. */ 128244a5668SFam Zheng #define RAW_LOCK_PERM_BASE 100 129244a5668SFam Zheng #define RAW_LOCK_SHARED_BASE 200 130244a5668SFam Zheng 131c1bb86cdSEric Blake typedef struct BDRVRawState { 132c1bb86cdSEric Blake int fd; 133244a5668SFam Zheng bool use_lock; 134c1bb86cdSEric Blake int type; 135c1bb86cdSEric Blake int open_flags; 136c1bb86cdSEric Blake size_t buf_align; 137c1bb86cdSEric Blake 138244a5668SFam Zheng /* The current permissions. */ 139244a5668SFam Zheng uint64_t perm; 140244a5668SFam Zheng uint64_t shared_perm; 141244a5668SFam Zheng 1422996ffadSFam Zheng /* The perms bits whose corresponding bytes are already locked in 143f2e3af29SFam Zheng * s->fd. */ 1442996ffadSFam Zheng uint64_t locked_perm; 1452996ffadSFam Zheng uint64_t locked_shared_perm; 1462996ffadSFam Zheng 1476ceabe6fSKevin Wolf int perm_change_fd; 148e0c9cf3aSKevin Wolf BDRVReopenState *reopen_state; 149e0c9cf3aSKevin Wolf 150c1bb86cdSEric Blake #ifdef CONFIG_XFS 151c1bb86cdSEric Blake bool is_xfs:1; 152c1bb86cdSEric Blake #endif 153c1bb86cdSEric Blake bool has_discard:1; 154c1bb86cdSEric Blake bool has_write_zeroes:1; 155c1bb86cdSEric Blake bool discard_zeroes:1; 156c1bb86cdSEric Blake bool use_linux_aio:1; 157e5bcf967SKevin Wolf bool page_cache_inconsistent:1; 158c1bb86cdSEric Blake bool has_fallocate; 159c1bb86cdSEric Blake bool needs_alignment; 160f357fcd8SStefan Hajnoczi bool drop_cache; 16131be8a2aSStefan Hajnoczi bool check_cache_dropped; 1627c9e5276SPaolo Bonzini 1637c9e5276SPaolo Bonzini PRManager *pr_mgr; 164c1bb86cdSEric Blake } BDRVRawState; 165c1bb86cdSEric Blake 166c1bb86cdSEric Blake typedef struct BDRVRawReopenState { 167c1bb86cdSEric Blake int fd; 168c1bb86cdSEric Blake int open_flags; 169f357fcd8SStefan Hajnoczi bool drop_cache; 17031be8a2aSStefan Hajnoczi bool check_cache_dropped; 171c1bb86cdSEric Blake } BDRVRawReopenState; 172c1bb86cdSEric Blake 173c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs); 174c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs); 175c1bb86cdSEric Blake 176c1bb86cdSEric Blake typedef struct RawPosixAIOData { 177c1bb86cdSEric Blake BlockDriverState *bs; 178c1bb86cdSEric Blake int aio_type; 179d57c44d0SKevin Wolf int aio_fildes; 180d57c44d0SKevin Wolf 181d57c44d0SKevin Wolf off_t aio_offset; 182d57c44d0SKevin Wolf uint64_t aio_nbytes; 183d57c44d0SKevin Wolf 18493f4e2ffSKevin Wolf union { 18593f4e2ffSKevin Wolf struct { 186d57c44d0SKevin Wolf struct iovec *iov; 187d57c44d0SKevin Wolf int niov; 188d57c44d0SKevin Wolf } io; 189d57c44d0SKevin Wolf struct { 190d57c44d0SKevin Wolf uint64_t cmd; 191d57c44d0SKevin Wolf void *buf; 192d57c44d0SKevin Wolf } ioctl; 193d57c44d0SKevin Wolf struct { 1941efad060SFam Zheng int aio_fd2; 1951efad060SFam Zheng off_t aio_offset2; 196d57c44d0SKevin Wolf } copy_range; 19793f4e2ffSKevin Wolf struct { 19893f4e2ffSKevin Wolf PreallocMode prealloc; 19993f4e2ffSKevin Wolf Error **errp; 200d57c44d0SKevin Wolf } truncate; 20193f4e2ffSKevin Wolf }; 202c1bb86cdSEric Blake } RawPosixAIOData; 203c1bb86cdSEric Blake 204c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 205c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs); 206c1bb86cdSEric Blake #endif 207c1bb86cdSEric Blake 208c1bb86cdSEric Blake #if defined(__NetBSD__) 209db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 210c1bb86cdSEric Blake { 211c1bb86cdSEric Blake static char namebuf[PATH_MAX]; 212c1bb86cdSEric Blake const char *dp, *fname; 213c1bb86cdSEric Blake struct stat sb; 214c1bb86cdSEric Blake 215c1bb86cdSEric Blake fname = *filename; 216c1bb86cdSEric Blake dp = strrchr(fname, '/'); 217c1bb86cdSEric Blake if (lstat(fname, &sb) < 0) { 218db0754dfSFam Zheng error_setg_errno(errp, errno, "%s: stat failed", fname); 219c1bb86cdSEric Blake return -errno; 220c1bb86cdSEric Blake } 221c1bb86cdSEric Blake 222c1bb86cdSEric Blake if (!S_ISBLK(sb.st_mode)) { 223c1bb86cdSEric Blake return 0; 224c1bb86cdSEric Blake } 225c1bb86cdSEric Blake 226c1bb86cdSEric Blake if (dp == NULL) { 227c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "r%s", fname); 228c1bb86cdSEric Blake } else { 229c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "%.*s/r%s", 230c1bb86cdSEric Blake (int)(dp - fname), fname, dp + 1); 231c1bb86cdSEric Blake } 232c1bb86cdSEric Blake *filename = namebuf; 233db0754dfSFam Zheng warn_report("%s is a block device, using %s", fname, *filename); 234c1bb86cdSEric Blake 235c1bb86cdSEric Blake return 0; 236c1bb86cdSEric Blake } 237c1bb86cdSEric Blake #else 238db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 239c1bb86cdSEric Blake { 240c1bb86cdSEric Blake return 0; 241c1bb86cdSEric Blake } 242c1bb86cdSEric Blake #endif 243c1bb86cdSEric Blake 244c1bb86cdSEric Blake /* 245c1bb86cdSEric Blake * Get logical block size via ioctl. On success store it in @sector_size_p. 246c1bb86cdSEric Blake */ 247c1bb86cdSEric Blake static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) 248c1bb86cdSEric Blake { 249c1bb86cdSEric Blake unsigned int sector_size; 250c1bb86cdSEric Blake bool success = false; 251700f9ce0SPeter Maydell int i; 252c1bb86cdSEric Blake 253c1bb86cdSEric Blake errno = ENOTSUP; 254700f9ce0SPeter Maydell static const unsigned long ioctl_list[] = { 255c1bb86cdSEric Blake #ifdef BLKSSZGET 256700f9ce0SPeter Maydell BLKSSZGET, 257c1bb86cdSEric Blake #endif 258c1bb86cdSEric Blake #ifdef DKIOCGETBLOCKSIZE 259700f9ce0SPeter Maydell DKIOCGETBLOCKSIZE, 260c1bb86cdSEric Blake #endif 261c1bb86cdSEric Blake #ifdef DIOCGSECTORSIZE 262700f9ce0SPeter Maydell DIOCGSECTORSIZE, 263700f9ce0SPeter Maydell #endif 264700f9ce0SPeter Maydell }; 265700f9ce0SPeter Maydell 266700f9ce0SPeter Maydell /* Try a few ioctls to get the right size */ 267700f9ce0SPeter Maydell for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { 268700f9ce0SPeter Maydell if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { 269c1bb86cdSEric Blake *sector_size_p = sector_size; 270c1bb86cdSEric Blake success = true; 271c1bb86cdSEric Blake } 272700f9ce0SPeter Maydell } 273c1bb86cdSEric Blake 274c1bb86cdSEric Blake return success ? 0 : -errno; 275c1bb86cdSEric Blake } 276c1bb86cdSEric Blake 277c1bb86cdSEric Blake /** 278c1bb86cdSEric Blake * Get physical block size of @fd. 279c1bb86cdSEric Blake * On success, store it in @blk_size and return 0. 280c1bb86cdSEric Blake * On failure, return -errno. 281c1bb86cdSEric Blake */ 282c1bb86cdSEric Blake static int probe_physical_blocksize(int fd, unsigned int *blk_size) 283c1bb86cdSEric Blake { 284c1bb86cdSEric Blake #ifdef BLKPBSZGET 285c1bb86cdSEric Blake if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { 286c1bb86cdSEric Blake return -errno; 287c1bb86cdSEric Blake } 288c1bb86cdSEric Blake return 0; 289c1bb86cdSEric Blake #else 290c1bb86cdSEric Blake return -ENOTSUP; 291c1bb86cdSEric Blake #endif 292c1bb86cdSEric Blake } 293c1bb86cdSEric Blake 294c1bb86cdSEric Blake /* Check if read is allowed with given memory buffer and length. 295c1bb86cdSEric Blake * 296c1bb86cdSEric Blake * This function is used to check O_DIRECT memory buffer and request alignment. 297c1bb86cdSEric Blake */ 298c1bb86cdSEric Blake static bool raw_is_io_aligned(int fd, void *buf, size_t len) 299c1bb86cdSEric Blake { 300c1bb86cdSEric Blake ssize_t ret = pread(fd, buf, len, 0); 301c1bb86cdSEric Blake 302c1bb86cdSEric Blake if (ret >= 0) { 303c1bb86cdSEric Blake return true; 304c1bb86cdSEric Blake } 305c1bb86cdSEric Blake 306c1bb86cdSEric Blake #ifdef __linux__ 307c1bb86cdSEric Blake /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore 308c1bb86cdSEric Blake * other errors (e.g. real I/O error), which could happen on a failed 309c1bb86cdSEric Blake * drive, since we only care about probing alignment. 310c1bb86cdSEric Blake */ 311c1bb86cdSEric Blake if (errno != EINVAL) { 312c1bb86cdSEric Blake return true; 313c1bb86cdSEric Blake } 314c1bb86cdSEric Blake #endif 315c1bb86cdSEric Blake 316c1bb86cdSEric Blake return false; 317c1bb86cdSEric Blake } 318c1bb86cdSEric Blake 319c1bb86cdSEric Blake static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) 320c1bb86cdSEric Blake { 321c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 322c1bb86cdSEric Blake char *buf; 323c1bb86cdSEric Blake size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize()); 324c1bb86cdSEric Blake 325c1bb86cdSEric Blake /* For SCSI generic devices the alignment is not really used. 326c1bb86cdSEric Blake With buffered I/O, we don't have any restrictions. */ 327c1bb86cdSEric Blake if (bdrv_is_sg(bs) || !s->needs_alignment) { 328c1bb86cdSEric Blake bs->bl.request_alignment = 1; 329c1bb86cdSEric Blake s->buf_align = 1; 330c1bb86cdSEric Blake return; 331c1bb86cdSEric Blake } 332c1bb86cdSEric Blake 333c1bb86cdSEric Blake bs->bl.request_alignment = 0; 334c1bb86cdSEric Blake s->buf_align = 0; 335c1bb86cdSEric Blake /* Let's try to use the logical blocksize for the alignment. */ 336c1bb86cdSEric Blake if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { 337c1bb86cdSEric Blake bs->bl.request_alignment = 0; 338c1bb86cdSEric Blake } 339c1bb86cdSEric Blake #ifdef CONFIG_XFS 340c1bb86cdSEric Blake if (s->is_xfs) { 341c1bb86cdSEric Blake struct dioattr da; 342c1bb86cdSEric Blake if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { 343c1bb86cdSEric Blake bs->bl.request_alignment = da.d_miniosz; 344c1bb86cdSEric Blake /* The kernel returns wrong information for d_mem */ 345c1bb86cdSEric Blake /* s->buf_align = da.d_mem; */ 346c1bb86cdSEric Blake } 347c1bb86cdSEric Blake } 348c1bb86cdSEric Blake #endif 349c1bb86cdSEric Blake 350c1bb86cdSEric Blake /* If we could not get the sizes so far, we can only guess them */ 351c1bb86cdSEric Blake if (!s->buf_align) { 352c1bb86cdSEric Blake size_t align; 353c1bb86cdSEric Blake buf = qemu_memalign(max_align, 2 * max_align); 354c1bb86cdSEric Blake for (align = 512; align <= max_align; align <<= 1) { 355c1bb86cdSEric Blake if (raw_is_io_aligned(fd, buf + align, max_align)) { 356c1bb86cdSEric Blake s->buf_align = align; 357c1bb86cdSEric Blake break; 358c1bb86cdSEric Blake } 359c1bb86cdSEric Blake } 360c1bb86cdSEric Blake qemu_vfree(buf); 361c1bb86cdSEric Blake } 362c1bb86cdSEric Blake 363c1bb86cdSEric Blake if (!bs->bl.request_alignment) { 364c1bb86cdSEric Blake size_t align; 365c1bb86cdSEric Blake buf = qemu_memalign(s->buf_align, max_align); 366c1bb86cdSEric Blake for (align = 512; align <= max_align; align <<= 1) { 367c1bb86cdSEric Blake if (raw_is_io_aligned(fd, buf, align)) { 368c1bb86cdSEric Blake bs->bl.request_alignment = align; 369c1bb86cdSEric Blake break; 370c1bb86cdSEric Blake } 371c1bb86cdSEric Blake } 372c1bb86cdSEric Blake qemu_vfree(buf); 373c1bb86cdSEric Blake } 374c1bb86cdSEric Blake 375c1bb86cdSEric Blake if (!s->buf_align || !bs->bl.request_alignment) { 376c1bb86cdSEric Blake error_setg(errp, "Could not find working O_DIRECT alignment"); 377c1bb86cdSEric Blake error_append_hint(errp, "Try cache.direct=off\n"); 378c1bb86cdSEric Blake } 379c1bb86cdSEric Blake } 380c1bb86cdSEric Blake 38123dece19SKevin Wolf static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) 382c1bb86cdSEric Blake { 38323dece19SKevin Wolf bool read_write = false; 384c1bb86cdSEric Blake assert(open_flags != NULL); 385c1bb86cdSEric Blake 386c1bb86cdSEric Blake *open_flags |= O_BINARY; 387c1bb86cdSEric Blake *open_flags &= ~O_ACCMODE; 38823dece19SKevin Wolf 38923dece19SKevin Wolf if (bdrv_flags & BDRV_O_AUTO_RDONLY) { 39023dece19SKevin Wolf read_write = has_writers; 39123dece19SKevin Wolf } else if (bdrv_flags & BDRV_O_RDWR) { 39223dece19SKevin Wolf read_write = true; 39323dece19SKevin Wolf } 39423dece19SKevin Wolf 39523dece19SKevin Wolf if (read_write) { 396c1bb86cdSEric Blake *open_flags |= O_RDWR; 397c1bb86cdSEric Blake } else { 398c1bb86cdSEric Blake *open_flags |= O_RDONLY; 399c1bb86cdSEric Blake } 400c1bb86cdSEric Blake 401c1bb86cdSEric Blake /* Use O_DSYNC for write-through caching, no flags for write-back caching, 402c1bb86cdSEric Blake * and O_DIRECT for no caching. */ 403c1bb86cdSEric Blake if ((bdrv_flags & BDRV_O_NOCACHE)) { 404c1bb86cdSEric Blake *open_flags |= O_DIRECT; 405c1bb86cdSEric Blake } 406c1bb86cdSEric Blake } 407c1bb86cdSEric Blake 408c1bb86cdSEric Blake static void raw_parse_filename(const char *filename, QDict *options, 409c1bb86cdSEric Blake Error **errp) 410c1bb86cdSEric Blake { 41103c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "file:", options); 412c1bb86cdSEric Blake } 413c1bb86cdSEric Blake 414c1bb86cdSEric Blake static QemuOptsList raw_runtime_opts = { 415c1bb86cdSEric Blake .name = "raw", 416c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), 417c1bb86cdSEric Blake .desc = { 418c1bb86cdSEric Blake { 419c1bb86cdSEric Blake .name = "filename", 420c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 421c1bb86cdSEric Blake .help = "File name of the image", 422c1bb86cdSEric Blake }, 423c1bb86cdSEric Blake { 424c1bb86cdSEric Blake .name = "aio", 425c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 426c1bb86cdSEric Blake .help = "host AIO implementation (threads, native)", 427c1bb86cdSEric Blake }, 42816b48d5dSFam Zheng { 42916b48d5dSFam Zheng .name = "locking", 43016b48d5dSFam Zheng .type = QEMU_OPT_STRING, 43116b48d5dSFam Zheng .help = "file locking mode (on/off/auto, default: auto)", 43216b48d5dSFam Zheng }, 4337c9e5276SPaolo Bonzini { 4347c9e5276SPaolo Bonzini .name = "pr-manager", 4357c9e5276SPaolo Bonzini .type = QEMU_OPT_STRING, 4367c9e5276SPaolo Bonzini .help = "id of persistent reservation manager object (default: none)", 4377c9e5276SPaolo Bonzini }, 438f357fcd8SStefan Hajnoczi #if defined(__linux__) 439f357fcd8SStefan Hajnoczi { 440f357fcd8SStefan Hajnoczi .name = "drop-cache", 441f357fcd8SStefan Hajnoczi .type = QEMU_OPT_BOOL, 442f357fcd8SStefan Hajnoczi .help = "invalidate page cache during live migration (default: on)", 443f357fcd8SStefan Hajnoczi }, 444f357fcd8SStefan Hajnoczi #endif 44531be8a2aSStefan Hajnoczi { 44631be8a2aSStefan Hajnoczi .name = "x-check-cache-dropped", 44731be8a2aSStefan Hajnoczi .type = QEMU_OPT_BOOL, 44831be8a2aSStefan Hajnoczi .help = "check that page cache was dropped on live migration (default: off)" 44931be8a2aSStefan Hajnoczi }, 450c1bb86cdSEric Blake { /* end of list */ } 451c1bb86cdSEric Blake }, 452c1bb86cdSEric Blake }; 453c1bb86cdSEric Blake 4548a2ce0bcSAlberto Garcia static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; 4558a2ce0bcSAlberto Garcia 456c1bb86cdSEric Blake static int raw_open_common(BlockDriverState *bs, QDict *options, 457230ff739SJohn Snow int bdrv_flags, int open_flags, 458230ff739SJohn Snow bool device, Error **errp) 459c1bb86cdSEric Blake { 460c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 461c1bb86cdSEric Blake QemuOpts *opts; 462c1bb86cdSEric Blake Error *local_err = NULL; 463c1bb86cdSEric Blake const char *filename = NULL; 4647c9e5276SPaolo Bonzini const char *str; 465c1bb86cdSEric Blake BlockdevAioOptions aio, aio_default; 466c1bb86cdSEric Blake int fd, ret; 467c1bb86cdSEric Blake struct stat st; 468244a5668SFam Zheng OnOffAuto locking; 469c1bb86cdSEric Blake 470c1bb86cdSEric Blake opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 471c1bb86cdSEric Blake qemu_opts_absorb_qdict(opts, options, &local_err); 472c1bb86cdSEric Blake if (local_err) { 473c1bb86cdSEric Blake error_propagate(errp, local_err); 474c1bb86cdSEric Blake ret = -EINVAL; 475c1bb86cdSEric Blake goto fail; 476c1bb86cdSEric Blake } 477c1bb86cdSEric Blake 478c1bb86cdSEric Blake filename = qemu_opt_get(opts, "filename"); 479c1bb86cdSEric Blake 480db0754dfSFam Zheng ret = raw_normalize_devicepath(&filename, errp); 481c1bb86cdSEric Blake if (ret != 0) { 482c1bb86cdSEric Blake goto fail; 483c1bb86cdSEric Blake } 484c1bb86cdSEric Blake 485c1bb86cdSEric Blake aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO) 486c1bb86cdSEric Blake ? BLOCKDEV_AIO_OPTIONS_NATIVE 487c1bb86cdSEric Blake : BLOCKDEV_AIO_OPTIONS_THREADS; 488f7abe0ecSMarc-André Lureau aio = qapi_enum_parse(&BlockdevAioOptions_lookup, 489f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "aio"), 49006c60b6cSMarkus Armbruster aio_default, &local_err); 491c1bb86cdSEric Blake if (local_err) { 492c1bb86cdSEric Blake error_propagate(errp, local_err); 493c1bb86cdSEric Blake ret = -EINVAL; 494c1bb86cdSEric Blake goto fail; 495c1bb86cdSEric Blake } 496c1bb86cdSEric Blake s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); 497c1bb86cdSEric Blake 498f7abe0ecSMarc-André Lureau locking = qapi_enum_parse(&OnOffAuto_lookup, 499f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "locking"), 50006c60b6cSMarkus Armbruster ON_OFF_AUTO_AUTO, &local_err); 501244a5668SFam Zheng if (local_err) { 502244a5668SFam Zheng error_propagate(errp, local_err); 503244a5668SFam Zheng ret = -EINVAL; 504244a5668SFam Zheng goto fail; 505244a5668SFam Zheng } 506244a5668SFam Zheng switch (locking) { 507244a5668SFam Zheng case ON_OFF_AUTO_ON: 508244a5668SFam Zheng s->use_lock = true; 5092b218f5dSFam Zheng if (!qemu_has_ofd_lock()) { 510db0754dfSFam Zheng warn_report("File lock requested but OFD locking syscall is " 511db0754dfSFam Zheng "unavailable, falling back to POSIX file locks"); 512db0754dfSFam Zheng error_printf("Due to the implementation, locks can be lost " 5132b218f5dSFam Zheng "unexpectedly.\n"); 5142b218f5dSFam Zheng } 515244a5668SFam Zheng break; 516244a5668SFam Zheng case ON_OFF_AUTO_OFF: 517244a5668SFam Zheng s->use_lock = false; 518244a5668SFam Zheng break; 519244a5668SFam Zheng case ON_OFF_AUTO_AUTO: 5202b218f5dSFam Zheng s->use_lock = qemu_has_ofd_lock(); 521244a5668SFam Zheng break; 522244a5668SFam Zheng default: 523244a5668SFam Zheng abort(); 524244a5668SFam Zheng } 525244a5668SFam Zheng 5267c9e5276SPaolo Bonzini str = qemu_opt_get(opts, "pr-manager"); 5277c9e5276SPaolo Bonzini if (str) { 5287c9e5276SPaolo Bonzini s->pr_mgr = pr_manager_lookup(str, &local_err); 5297c9e5276SPaolo Bonzini if (local_err) { 5307c9e5276SPaolo Bonzini error_propagate(errp, local_err); 5317c9e5276SPaolo Bonzini ret = -EINVAL; 5327c9e5276SPaolo Bonzini goto fail; 5337c9e5276SPaolo Bonzini } 5347c9e5276SPaolo Bonzini } 5357c9e5276SPaolo Bonzini 536f357fcd8SStefan Hajnoczi s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); 53731be8a2aSStefan Hajnoczi s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", 53831be8a2aSStefan Hajnoczi false); 53931be8a2aSStefan Hajnoczi 540c1bb86cdSEric Blake s->open_flags = open_flags; 54123dece19SKevin Wolf raw_parse_flags(bdrv_flags, &s->open_flags, false); 542c1bb86cdSEric Blake 543c1bb86cdSEric Blake s->fd = -1; 544c1bb86cdSEric Blake fd = qemu_open(filename, s->open_flags, 0644); 54564107dc0SKevin Wolf ret = fd < 0 ? -errno : 0; 54664107dc0SKevin Wolf 54764107dc0SKevin Wolf if (ret < 0) { 54864107dc0SKevin Wolf error_setg_errno(errp, -ret, "Could not open '%s'", filename); 549c1bb86cdSEric Blake if (ret == -EROFS) { 550c1bb86cdSEric Blake ret = -EACCES; 551c1bb86cdSEric Blake } 552c1bb86cdSEric Blake goto fail; 553c1bb86cdSEric Blake } 554c1bb86cdSEric Blake s->fd = fd; 555c1bb86cdSEric Blake 556244a5668SFam Zheng s->perm = 0; 557244a5668SFam Zheng s->shared_perm = BLK_PERM_ALL; 558244a5668SFam Zheng 559c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 560c1bb86cdSEric Blake /* Currently Linux does AIO only for files opened with O_DIRECT */ 561ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 562ed6e2161SNishanth Aravamudan if (!(s->open_flags & O_DIRECT)) { 563c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but it requires " 564c1bb86cdSEric Blake "cache.direct=on, which was not specified."); 565c1bb86cdSEric Blake ret = -EINVAL; 566c1bb86cdSEric Blake goto fail; 567c1bb86cdSEric Blake } 568ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { 569ed6e2161SNishanth Aravamudan error_prepend(errp, "Unable to use native AIO: "); 570ed6e2161SNishanth Aravamudan goto fail; 571ed6e2161SNishanth Aravamudan } 572ed6e2161SNishanth Aravamudan } 573c1bb86cdSEric Blake #else 574c1bb86cdSEric Blake if (s->use_linux_aio) { 575c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but is not supported " 576c1bb86cdSEric Blake "in this build."); 577c1bb86cdSEric Blake ret = -EINVAL; 578c1bb86cdSEric Blake goto fail; 579c1bb86cdSEric Blake } 580c1bb86cdSEric Blake #endif /* !defined(CONFIG_LINUX_AIO) */ 581c1bb86cdSEric Blake 582c1bb86cdSEric Blake s->has_discard = true; 583c1bb86cdSEric Blake s->has_write_zeroes = true; 584c1bb86cdSEric Blake if ((bs->open_flags & BDRV_O_NOCACHE) != 0) { 585c1bb86cdSEric Blake s->needs_alignment = true; 586c1bb86cdSEric Blake } 587c1bb86cdSEric Blake 588c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 589c1bb86cdSEric Blake ret = -errno; 590c1bb86cdSEric Blake error_setg_errno(errp, errno, "Could not stat file"); 591c1bb86cdSEric Blake goto fail; 592c1bb86cdSEric Blake } 593230ff739SJohn Snow 594230ff739SJohn Snow if (!device) { 595230ff739SJohn Snow if (S_ISBLK(st.st_mode)) { 596230ff739SJohn Snow warn_report("Opening a block device as a file using the '%s' " 597230ff739SJohn Snow "driver is deprecated", bs->drv->format_name); 598230ff739SJohn Snow } else if (S_ISCHR(st.st_mode)) { 599230ff739SJohn Snow warn_report("Opening a character device as a file using the '%s' " 600230ff739SJohn Snow "driver is deprecated", bs->drv->format_name); 601230ff739SJohn Snow } else if (!S_ISREG(st.st_mode)) { 602230ff739SJohn Snow error_setg(errp, "A regular file was expected by the '%s' driver, " 603230ff739SJohn Snow "but something else was given", bs->drv->format_name); 604230ff739SJohn Snow ret = -EINVAL; 605230ff739SJohn Snow goto fail; 606230ff739SJohn Snow } else { 607c1bb86cdSEric Blake s->discard_zeroes = true; 608c1bb86cdSEric Blake s->has_fallocate = true; 609c1bb86cdSEric Blake } 610230ff739SJohn Snow } else { 611230ff739SJohn Snow if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 612230ff739SJohn Snow error_setg(errp, "'%s' driver expects either " 613230ff739SJohn Snow "a character or block device", bs->drv->format_name); 614230ff739SJohn Snow ret = -EINVAL; 615230ff739SJohn Snow goto fail; 616230ff739SJohn Snow } 617230ff739SJohn Snow } 618230ff739SJohn Snow 619c1bb86cdSEric Blake if (S_ISBLK(st.st_mode)) { 620c1bb86cdSEric Blake #ifdef BLKDISCARDZEROES 621c1bb86cdSEric Blake unsigned int arg; 622c1bb86cdSEric Blake if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { 623c1bb86cdSEric Blake s->discard_zeroes = true; 624c1bb86cdSEric Blake } 625c1bb86cdSEric Blake #endif 626c1bb86cdSEric Blake #ifdef __linux__ 627c1bb86cdSEric Blake /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do 628c1bb86cdSEric Blake * not rely on the contents of discarded blocks unless using O_DIRECT. 629c1bb86cdSEric Blake * Same for BLKZEROOUT. 630c1bb86cdSEric Blake */ 631c1bb86cdSEric Blake if (!(bs->open_flags & BDRV_O_NOCACHE)) { 632c1bb86cdSEric Blake s->discard_zeroes = false; 633c1bb86cdSEric Blake s->has_write_zeroes = false; 634c1bb86cdSEric Blake } 635c1bb86cdSEric Blake #endif 636c1bb86cdSEric Blake } 637c1bb86cdSEric Blake #ifdef __FreeBSD__ 638c1bb86cdSEric Blake if (S_ISCHR(st.st_mode)) { 639c1bb86cdSEric Blake /* 640c1bb86cdSEric Blake * The file is a char device (disk), which on FreeBSD isn't behind 641c1bb86cdSEric Blake * a pager, so force all requests to be aligned. This is needed 642c1bb86cdSEric Blake * so QEMU makes sure all IO operations on the device are aligned 643c1bb86cdSEric Blake * to sector size, or else FreeBSD will reject them with EINVAL. 644c1bb86cdSEric Blake */ 645c1bb86cdSEric Blake s->needs_alignment = true; 646c1bb86cdSEric Blake } 647c1bb86cdSEric Blake #endif 648c1bb86cdSEric Blake 649c1bb86cdSEric Blake #ifdef CONFIG_XFS 650c1bb86cdSEric Blake if (platform_test_xfs_fd(s->fd)) { 651c1bb86cdSEric Blake s->is_xfs = true; 652c1bb86cdSEric Blake } 653c1bb86cdSEric Blake #endif 654c1bb86cdSEric Blake 655738301e1SKevin Wolf bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; 656c1bb86cdSEric Blake ret = 0; 657c1bb86cdSEric Blake fail: 658c1bb86cdSEric Blake if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { 659c1bb86cdSEric Blake unlink(filename); 660c1bb86cdSEric Blake } 661c1bb86cdSEric Blake qemu_opts_del(opts); 662c1bb86cdSEric Blake return ret; 663c1bb86cdSEric Blake } 664c1bb86cdSEric Blake 665c1bb86cdSEric Blake static int raw_open(BlockDriverState *bs, QDict *options, int flags, 666c1bb86cdSEric Blake Error **errp) 667c1bb86cdSEric Blake { 668c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 669c1bb86cdSEric Blake 670c1bb86cdSEric Blake s->type = FTYPE_FILE; 671230ff739SJohn Snow return raw_open_common(bs, options, flags, 0, false, errp); 672c1bb86cdSEric Blake } 673c1bb86cdSEric Blake 674244a5668SFam Zheng typedef enum { 675244a5668SFam Zheng RAW_PL_PREPARE, 676244a5668SFam Zheng RAW_PL_COMMIT, 677244a5668SFam Zheng RAW_PL_ABORT, 678244a5668SFam Zheng } RawPermLockOp; 679244a5668SFam Zheng 680244a5668SFam Zheng #define PERM_FOREACH(i) \ 681244a5668SFam Zheng for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) 682244a5668SFam Zheng 683244a5668SFam Zheng /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the 684244a5668SFam Zheng * file; if @unlock == true, also unlock the unneeded bytes. 685244a5668SFam Zheng * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. 686244a5668SFam Zheng */ 6872996ffadSFam Zheng static int raw_apply_lock_bytes(BDRVRawState *s, int fd, 688244a5668SFam Zheng uint64_t perm_lock_bits, 689244a5668SFam Zheng uint64_t shared_perm_lock_bits, 690244a5668SFam Zheng bool unlock, Error **errp) 691244a5668SFam Zheng { 692244a5668SFam Zheng int ret; 693244a5668SFam Zheng int i; 6942996ffadSFam Zheng uint64_t locked_perm, locked_shared_perm; 6952996ffadSFam Zheng 6962996ffadSFam Zheng if (s) { 6972996ffadSFam Zheng locked_perm = s->locked_perm; 6982996ffadSFam Zheng locked_shared_perm = s->locked_shared_perm; 6992996ffadSFam Zheng } else { 7002996ffadSFam Zheng /* 7012996ffadSFam Zheng * We don't have the previous bits, just lock/unlock for each of the 7022996ffadSFam Zheng * requested bits. 7032996ffadSFam Zheng */ 7042996ffadSFam Zheng if (unlock) { 7052996ffadSFam Zheng locked_perm = BLK_PERM_ALL; 7062996ffadSFam Zheng locked_shared_perm = BLK_PERM_ALL; 7072996ffadSFam Zheng } else { 7082996ffadSFam Zheng locked_perm = 0; 7092996ffadSFam Zheng locked_shared_perm = 0; 7102996ffadSFam Zheng } 7112996ffadSFam Zheng } 712244a5668SFam Zheng 713244a5668SFam Zheng PERM_FOREACH(i) { 714244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 7152996ffadSFam Zheng uint64_t bit = (1ULL << i); 7162996ffadSFam Zheng if ((perm_lock_bits & bit) && !(locked_perm & bit)) { 717d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 718244a5668SFam Zheng if (ret) { 719244a5668SFam Zheng error_setg(errp, "Failed to lock byte %d", off); 720244a5668SFam Zheng return ret; 7212996ffadSFam Zheng } else if (s) { 7222996ffadSFam Zheng s->locked_perm |= bit; 723244a5668SFam Zheng } 7242996ffadSFam Zheng } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { 725d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 726244a5668SFam Zheng if (ret) { 727244a5668SFam Zheng error_setg(errp, "Failed to unlock byte %d", off); 728244a5668SFam Zheng return ret; 7292996ffadSFam Zheng } else if (s) { 7302996ffadSFam Zheng s->locked_perm &= ~bit; 731244a5668SFam Zheng } 732244a5668SFam Zheng } 733244a5668SFam Zheng } 734244a5668SFam Zheng PERM_FOREACH(i) { 735244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 7362996ffadSFam Zheng uint64_t bit = (1ULL << i); 7372996ffadSFam Zheng if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { 738d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 739244a5668SFam Zheng if (ret) { 740244a5668SFam Zheng error_setg(errp, "Failed to lock byte %d", off); 741244a5668SFam Zheng return ret; 7422996ffadSFam Zheng } else if (s) { 7432996ffadSFam Zheng s->locked_shared_perm |= bit; 744244a5668SFam Zheng } 7452996ffadSFam Zheng } else if (unlock && (locked_shared_perm & bit) && 7462996ffadSFam Zheng !(shared_perm_lock_bits & bit)) { 747d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 748244a5668SFam Zheng if (ret) { 749244a5668SFam Zheng error_setg(errp, "Failed to unlock byte %d", off); 750244a5668SFam Zheng return ret; 7512996ffadSFam Zheng } else if (s) { 7522996ffadSFam Zheng s->locked_shared_perm &= ~bit; 753244a5668SFam Zheng } 754244a5668SFam Zheng } 755244a5668SFam Zheng } 756244a5668SFam Zheng return 0; 757244a5668SFam Zheng } 758244a5668SFam Zheng 759244a5668SFam Zheng /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ 760d0a96155SMax Reitz static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, 761244a5668SFam Zheng Error **errp) 762244a5668SFam Zheng { 763244a5668SFam Zheng int ret; 764244a5668SFam Zheng int i; 765244a5668SFam Zheng 766244a5668SFam Zheng PERM_FOREACH(i) { 767244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 768244a5668SFam Zheng uint64_t p = 1ULL << i; 769244a5668SFam Zheng if (perm & p) { 770d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 771244a5668SFam Zheng if (ret) { 772244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 773244a5668SFam Zheng error_setg(errp, 774244a5668SFam Zheng "Failed to get \"%s\" lock", 775244a5668SFam Zheng perm_name); 776244a5668SFam Zheng g_free(perm_name); 777244a5668SFam Zheng return ret; 778244a5668SFam Zheng } 779244a5668SFam Zheng } 780244a5668SFam Zheng } 781244a5668SFam Zheng PERM_FOREACH(i) { 782244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 783244a5668SFam Zheng uint64_t p = 1ULL << i; 784244a5668SFam Zheng if (!(shared_perm & p)) { 785d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 786244a5668SFam Zheng if (ret) { 787244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 788244a5668SFam Zheng error_setg(errp, 789244a5668SFam Zheng "Failed to get shared \"%s\" lock", 790244a5668SFam Zheng perm_name); 791244a5668SFam Zheng g_free(perm_name); 792244a5668SFam Zheng return ret; 793244a5668SFam Zheng } 794244a5668SFam Zheng } 795244a5668SFam Zheng } 796244a5668SFam Zheng return 0; 797244a5668SFam Zheng } 798244a5668SFam Zheng 799244a5668SFam Zheng static int raw_handle_perm_lock(BlockDriverState *bs, 800244a5668SFam Zheng RawPermLockOp op, 801244a5668SFam Zheng uint64_t new_perm, uint64_t new_shared, 802244a5668SFam Zheng Error **errp) 803244a5668SFam Zheng { 804244a5668SFam Zheng BDRVRawState *s = bs->opaque; 805244a5668SFam Zheng int ret = 0; 806244a5668SFam Zheng Error *local_err = NULL; 807244a5668SFam Zheng 808244a5668SFam Zheng if (!s->use_lock) { 809244a5668SFam Zheng return 0; 810244a5668SFam Zheng } 811244a5668SFam Zheng 812244a5668SFam Zheng if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { 813244a5668SFam Zheng return 0; 814244a5668SFam Zheng } 815244a5668SFam Zheng 816244a5668SFam Zheng switch (op) { 817244a5668SFam Zheng case RAW_PL_PREPARE: 818*696aaaedSVladimir Sementsov-Ogievskiy if ((s->perm | new_perm) == s->perm && 819*696aaaedSVladimir Sementsov-Ogievskiy (s->shared_perm & new_shared) == s->shared_perm) 820*696aaaedSVladimir Sementsov-Ogievskiy { 821*696aaaedSVladimir Sementsov-Ogievskiy /* 822*696aaaedSVladimir Sementsov-Ogievskiy * We are going to unlock bytes, it should not fail. If it fail due 823*696aaaedSVladimir Sementsov-Ogievskiy * to some fs-dependent permission-unrelated reasons (which occurs 824*696aaaedSVladimir Sementsov-Ogievskiy * sometimes on NFS and leads to abort in bdrv_replace_child) we 825*696aaaedSVladimir Sementsov-Ogievskiy * can't prevent such errors by any check here. And we ignore them 826*696aaaedSVladimir Sementsov-Ogievskiy * anyway in ABORT and COMMIT. 827*696aaaedSVladimir Sementsov-Ogievskiy */ 828*696aaaedSVladimir Sementsov-Ogievskiy return 0; 829*696aaaedSVladimir Sementsov-Ogievskiy } 830f2e3af29SFam Zheng ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, 831244a5668SFam Zheng ~s->shared_perm | ~new_shared, 832244a5668SFam Zheng false, errp); 833244a5668SFam Zheng if (!ret) { 834f2e3af29SFam Zheng ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); 835244a5668SFam Zheng if (!ret) { 836244a5668SFam Zheng return 0; 837244a5668SFam Zheng } 838b857431dSFam Zheng error_append_hint(errp, 839b857431dSFam Zheng "Is another process using the image [%s]?\n", 840b857431dSFam Zheng bs->filename); 841244a5668SFam Zheng } 842244a5668SFam Zheng op = RAW_PL_ABORT; 843244a5668SFam Zheng /* fall through to unlock bytes. */ 844244a5668SFam Zheng case RAW_PL_ABORT: 845f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, 846d0a96155SMax Reitz true, &local_err); 847244a5668SFam Zheng if (local_err) { 848244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 849244a5668SFam Zheng * fail. Something weird happened, report it. 850244a5668SFam Zheng */ 851db0754dfSFam Zheng warn_report_err(local_err); 852244a5668SFam Zheng } 853244a5668SFam Zheng break; 854244a5668SFam Zheng case RAW_PL_COMMIT: 855f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, 856d0a96155SMax Reitz true, &local_err); 857244a5668SFam Zheng if (local_err) { 858244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 859244a5668SFam Zheng * fail. Something weird happened, report it. 860244a5668SFam Zheng */ 861db0754dfSFam Zheng warn_report_err(local_err); 862244a5668SFam Zheng } 863244a5668SFam Zheng break; 864244a5668SFam Zheng } 865244a5668SFam Zheng return ret; 866244a5668SFam Zheng } 867244a5668SFam Zheng 8685cec2870SKevin Wolf static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, 86923dece19SKevin Wolf int *open_flags, uint64_t perm, bool force_dup, 8706ceabe6fSKevin Wolf Error **errp) 8715cec2870SKevin Wolf { 8725cec2870SKevin Wolf BDRVRawState *s = bs->opaque; 8735cec2870SKevin Wolf int fd = -1; 8745cec2870SKevin Wolf int ret; 87523dece19SKevin Wolf bool has_writers = perm & 87623dece19SKevin Wolf (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); 8775cec2870SKevin Wolf int fcntl_flags = O_APPEND | O_NONBLOCK; 8785cec2870SKevin Wolf #ifdef O_NOATIME 8795cec2870SKevin Wolf fcntl_flags |= O_NOATIME; 8805cec2870SKevin Wolf #endif 8815cec2870SKevin Wolf 8825cec2870SKevin Wolf *open_flags = 0; 8835cec2870SKevin Wolf if (s->type == FTYPE_CD) { 8845cec2870SKevin Wolf *open_flags |= O_NONBLOCK; 8855cec2870SKevin Wolf } 8865cec2870SKevin Wolf 88723dece19SKevin Wolf raw_parse_flags(flags, open_flags, has_writers); 8885cec2870SKevin Wolf 8895cec2870SKevin Wolf #ifdef O_ASYNC 8905cec2870SKevin Wolf /* Not all operating systems have O_ASYNC, and those that don't 8915cec2870SKevin Wolf * will not let us track the state into rs->open_flags (typically 8925cec2870SKevin Wolf * you achieve the same effect with an ioctl, for example I_SETSIG 8935cec2870SKevin Wolf * on Solaris). But we do not use O_ASYNC, so that's fine. 8945cec2870SKevin Wolf */ 8955cec2870SKevin Wolf assert((s->open_flags & O_ASYNC) == 0); 8965cec2870SKevin Wolf #endif 8975cec2870SKevin Wolf 8986ceabe6fSKevin Wolf if (!force_dup && *open_flags == s->open_flags) { 8996ceabe6fSKevin Wolf /* We're lucky, the existing fd is fine */ 9006ceabe6fSKevin Wolf return s->fd; 9016ceabe6fSKevin Wolf } 9026ceabe6fSKevin Wolf 9035cec2870SKevin Wolf if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { 9045cec2870SKevin Wolf /* dup the original fd */ 9055cec2870SKevin Wolf fd = qemu_dup(s->fd); 9065cec2870SKevin Wolf if (fd >= 0) { 9075cec2870SKevin Wolf ret = fcntl_setfl(fd, *open_flags); 9085cec2870SKevin Wolf if (ret) { 9095cec2870SKevin Wolf qemu_close(fd); 9105cec2870SKevin Wolf fd = -1; 9115cec2870SKevin Wolf } 9125cec2870SKevin Wolf } 9135cec2870SKevin Wolf } 9145cec2870SKevin Wolf 9155cec2870SKevin Wolf /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ 9165cec2870SKevin Wolf if (fd == -1) { 9175cec2870SKevin Wolf const char *normalized_filename = bs->filename; 9185cec2870SKevin Wolf ret = raw_normalize_devicepath(&normalized_filename, errp); 9195cec2870SKevin Wolf if (ret >= 0) { 9205cec2870SKevin Wolf assert(!(*open_flags & O_CREAT)); 9215cec2870SKevin Wolf fd = qemu_open(normalized_filename, *open_flags); 9225cec2870SKevin Wolf if (fd == -1) { 9235cec2870SKevin Wolf error_setg_errno(errp, errno, "Could not reopen file"); 9245cec2870SKevin Wolf return -1; 9255cec2870SKevin Wolf } 9265cec2870SKevin Wolf } 9275cec2870SKevin Wolf } 9285cec2870SKevin Wolf 9295cec2870SKevin Wolf return fd; 9305cec2870SKevin Wolf } 9315cec2870SKevin Wolf 932c1bb86cdSEric Blake static int raw_reopen_prepare(BDRVReopenState *state, 933c1bb86cdSEric Blake BlockReopenQueue *queue, Error **errp) 934c1bb86cdSEric Blake { 935c1bb86cdSEric Blake BDRVRawState *s; 936c1bb86cdSEric Blake BDRVRawReopenState *rs; 93731be8a2aSStefan Hajnoczi QemuOpts *opts; 938a6aeca0cSKevin Wolf int ret; 939c1bb86cdSEric Blake Error *local_err = NULL; 940c1bb86cdSEric Blake 941c1bb86cdSEric Blake assert(state != NULL); 942c1bb86cdSEric Blake assert(state->bs != NULL); 943c1bb86cdSEric Blake 944c1bb86cdSEric Blake s = state->bs->opaque; 945c1bb86cdSEric Blake 946c1bb86cdSEric Blake state->opaque = g_new0(BDRVRawReopenState, 1); 947c1bb86cdSEric Blake rs = state->opaque; 94831be8a2aSStefan Hajnoczi 94931be8a2aSStefan Hajnoczi /* Handle options changes */ 95031be8a2aSStefan Hajnoczi opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 95131be8a2aSStefan Hajnoczi qemu_opts_absorb_qdict(opts, state->options, &local_err); 95231be8a2aSStefan Hajnoczi if (local_err) { 95331be8a2aSStefan Hajnoczi error_propagate(errp, local_err); 95431be8a2aSStefan Hajnoczi ret = -EINVAL; 95531be8a2aSStefan Hajnoczi goto out; 95631be8a2aSStefan Hajnoczi } 95731be8a2aSStefan Hajnoczi 958f357fcd8SStefan Hajnoczi rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); 9598d324575SAlberto Garcia rs->check_cache_dropped = 9608d324575SAlberto Garcia qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); 9618d324575SAlberto Garcia 9628d324575SAlberto Garcia /* This driver's reopen function doesn't currently allow changing 9638d324575SAlberto Garcia * other options, so let's put them back in the original QDict and 9648d324575SAlberto Garcia * bdrv_reopen_prepare() will detect changes and complain. */ 9658d324575SAlberto Garcia qemu_opts_to_qdict(opts, state->options); 966c1bb86cdSEric Blake 9675cec2870SKevin Wolf rs->fd = raw_reconfigure_getfd(state->bs, state->flags, &rs->open_flags, 96823dece19SKevin Wolf state->perm, true, &local_err); 9695cec2870SKevin Wolf if (local_err) { 9705cec2870SKevin Wolf error_propagate(errp, local_err); 971c1bb86cdSEric Blake ret = -1; 9725cec2870SKevin Wolf goto out; 973c1bb86cdSEric Blake } 974c1bb86cdSEric Blake 975c1bb86cdSEric Blake /* Fail already reopen_prepare() if we can't get a working O_DIRECT 976c1bb86cdSEric Blake * alignment with the new fd. */ 977c1bb86cdSEric Blake if (rs->fd != -1) { 978c1bb86cdSEric Blake raw_probe_alignment(state->bs, rs->fd, &local_err); 979c1bb86cdSEric Blake if (local_err) { 980c1bb86cdSEric Blake error_propagate(errp, local_err); 981c1bb86cdSEric Blake ret = -EINVAL; 982a6aeca0cSKevin Wolf goto out_fd; 983a6aeca0cSKevin Wolf } 984c1bb86cdSEric Blake } 985c1bb86cdSEric Blake 986e0c9cf3aSKevin Wolf s->reopen_state = state; 987a6aeca0cSKevin Wolf ret = 0; 988a6aeca0cSKevin Wolf out_fd: 989a6aeca0cSKevin Wolf if (ret < 0) { 990a6aeca0cSKevin Wolf qemu_close(rs->fd); 991a6aeca0cSKevin Wolf rs->fd = -1; 992a6aeca0cSKevin Wolf } 99331be8a2aSStefan Hajnoczi out: 99431be8a2aSStefan Hajnoczi qemu_opts_del(opts); 995c1bb86cdSEric Blake return ret; 996c1bb86cdSEric Blake } 997c1bb86cdSEric Blake 998c1bb86cdSEric Blake static void raw_reopen_commit(BDRVReopenState *state) 999c1bb86cdSEric Blake { 1000c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1001c1bb86cdSEric Blake BDRVRawState *s = state->bs->opaque; 1002c1bb86cdSEric Blake 1003f357fcd8SStefan Hajnoczi s->drop_cache = rs->drop_cache; 100431be8a2aSStefan Hajnoczi s->check_cache_dropped = rs->check_cache_dropped; 1005c1bb86cdSEric Blake s->open_flags = rs->open_flags; 1006c1bb86cdSEric Blake 1007c1bb86cdSEric Blake qemu_close(s->fd); 1008c1bb86cdSEric Blake s->fd = rs->fd; 1009c1bb86cdSEric Blake 1010c1bb86cdSEric Blake g_free(state->opaque); 1011c1bb86cdSEric Blake state->opaque = NULL; 1012e0c9cf3aSKevin Wolf 1013e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1014e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1015c1bb86cdSEric Blake } 1016c1bb86cdSEric Blake 1017c1bb86cdSEric Blake 1018c1bb86cdSEric Blake static void raw_reopen_abort(BDRVReopenState *state) 1019c1bb86cdSEric Blake { 1020c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1021e0c9cf3aSKevin Wolf BDRVRawState *s = state->bs->opaque; 1022c1bb86cdSEric Blake 1023c1bb86cdSEric Blake /* nothing to do if NULL, we didn't get far enough */ 1024c1bb86cdSEric Blake if (rs == NULL) { 1025c1bb86cdSEric Blake return; 1026c1bb86cdSEric Blake } 1027c1bb86cdSEric Blake 1028c1bb86cdSEric Blake if (rs->fd >= 0) { 1029c1bb86cdSEric Blake qemu_close(rs->fd); 1030c1bb86cdSEric Blake rs->fd = -1; 1031c1bb86cdSEric Blake } 1032c1bb86cdSEric Blake g_free(state->opaque); 1033c1bb86cdSEric Blake state->opaque = NULL; 1034e0c9cf3aSKevin Wolf 1035e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1036e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1037c1bb86cdSEric Blake } 1038c1bb86cdSEric Blake 103948265250SEric Farman static int hdev_get_max_transfer_length(BlockDriverState *bs, int fd) 1040c1bb86cdSEric Blake { 1041c1bb86cdSEric Blake #ifdef BLKSECTGET 104248265250SEric Farman int max_bytes = 0; 104348265250SEric Farman short max_sectors = 0; 104448265250SEric Farman if (bs->sg && ioctl(fd, BLKSECTGET, &max_bytes) == 0) { 104548265250SEric Farman return max_bytes; 104648265250SEric Farman } else if (!bs->sg && ioctl(fd, BLKSECTGET, &max_sectors) == 0) { 104748265250SEric Farman return max_sectors << BDRV_SECTOR_BITS; 1048c1bb86cdSEric Blake } else { 1049c1bb86cdSEric Blake return -errno; 1050c1bb86cdSEric Blake } 1051c1bb86cdSEric Blake #else 1052c1bb86cdSEric Blake return -ENOSYS; 1053c1bb86cdSEric Blake #endif 1054c1bb86cdSEric Blake } 1055c1bb86cdSEric Blake 10569103f1ceSFam Zheng static int hdev_get_max_segments(const struct stat *st) 10579103f1ceSFam Zheng { 10589103f1ceSFam Zheng #ifdef CONFIG_LINUX 10599103f1ceSFam Zheng char buf[32]; 10609103f1ceSFam Zheng const char *end; 10619103f1ceSFam Zheng char *sysfspath; 10629103f1ceSFam Zheng int ret; 10639103f1ceSFam Zheng int fd = -1; 10649103f1ceSFam Zheng long max_segments; 10659103f1ceSFam Zheng 10669103f1ceSFam Zheng sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", 10679103f1ceSFam Zheng major(st->st_rdev), minor(st->st_rdev)); 10689103f1ceSFam Zheng fd = open(sysfspath, O_RDONLY); 10699103f1ceSFam Zheng if (fd == -1) { 10709103f1ceSFam Zheng ret = -errno; 10719103f1ceSFam Zheng goto out; 10729103f1ceSFam Zheng } 10739103f1ceSFam Zheng do { 107469583490SStefan Hajnoczi ret = read(fd, buf, sizeof(buf) - 1); 10759103f1ceSFam Zheng } while (ret == -1 && errno == EINTR); 10769103f1ceSFam Zheng if (ret < 0) { 10779103f1ceSFam Zheng ret = -errno; 10789103f1ceSFam Zheng goto out; 10799103f1ceSFam Zheng } else if (ret == 0) { 10809103f1ceSFam Zheng ret = -EIO; 10819103f1ceSFam Zheng goto out; 10829103f1ceSFam Zheng } 10839103f1ceSFam Zheng buf[ret] = 0; 10849103f1ceSFam Zheng /* The file is ended with '\n', pass 'end' to accept that. */ 10859103f1ceSFam Zheng ret = qemu_strtol(buf, &end, 10, &max_segments); 10869103f1ceSFam Zheng if (ret == 0 && end && *end == '\n') { 10879103f1ceSFam Zheng ret = max_segments; 10889103f1ceSFam Zheng } 10899103f1ceSFam Zheng 10909103f1ceSFam Zheng out: 1091fed414dfSFam Zheng if (fd != -1) { 1092fed414dfSFam Zheng close(fd); 1093fed414dfSFam Zheng } 10949103f1ceSFam Zheng g_free(sysfspath); 10959103f1ceSFam Zheng return ret; 10969103f1ceSFam Zheng #else 10979103f1ceSFam Zheng return -ENOTSUP; 10989103f1ceSFam Zheng #endif 10999103f1ceSFam Zheng } 11009103f1ceSFam Zheng 1101c1bb86cdSEric Blake static void raw_refresh_limits(BlockDriverState *bs, Error **errp) 1102c1bb86cdSEric Blake { 1103c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1104c1bb86cdSEric Blake struct stat st; 1105c1bb86cdSEric Blake 1106c1bb86cdSEric Blake if (!fstat(s->fd, &st)) { 1107c4c41a0aSEric Farman if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode)) { 110848265250SEric Farman int ret = hdev_get_max_transfer_length(bs, s->fd); 110948265250SEric Farman if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { 111048265250SEric Farman bs->bl.max_transfer = pow2floor(ret); 1111c1bb86cdSEric Blake } 11129103f1ceSFam Zheng ret = hdev_get_max_segments(&st); 11139103f1ceSFam Zheng if (ret > 0) { 11149103f1ceSFam Zheng bs->bl.max_transfer = MIN(bs->bl.max_transfer, 11159103f1ceSFam Zheng ret * getpagesize()); 11169103f1ceSFam Zheng } 1117c1bb86cdSEric Blake } 1118c1bb86cdSEric Blake } 1119c1bb86cdSEric Blake 1120c1bb86cdSEric Blake raw_probe_alignment(bs, s->fd, errp); 1121c1bb86cdSEric Blake bs->bl.min_mem_alignment = s->buf_align; 1122c1bb86cdSEric Blake bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize()); 1123c1bb86cdSEric Blake } 1124c1bb86cdSEric Blake 1125c1bb86cdSEric Blake static int check_for_dasd(int fd) 1126c1bb86cdSEric Blake { 1127c1bb86cdSEric Blake #ifdef BIODASDINFO2 1128c1bb86cdSEric Blake struct dasd_information2_t info = {0}; 1129c1bb86cdSEric Blake 1130c1bb86cdSEric Blake return ioctl(fd, BIODASDINFO2, &info); 1131c1bb86cdSEric Blake #else 1132c1bb86cdSEric Blake return -1; 1133c1bb86cdSEric Blake #endif 1134c1bb86cdSEric Blake } 1135c1bb86cdSEric Blake 1136c1bb86cdSEric Blake /** 1137c1bb86cdSEric Blake * Try to get @bs's logical and physical block size. 1138c1bb86cdSEric Blake * On success, store them in @bsz and return zero. 1139c1bb86cdSEric Blake * On failure, return negative errno. 1140c1bb86cdSEric Blake */ 1141c1bb86cdSEric Blake static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) 1142c1bb86cdSEric Blake { 1143c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1144c1bb86cdSEric Blake int ret; 1145c1bb86cdSEric Blake 1146c1bb86cdSEric Blake /* If DASD, get blocksizes */ 1147c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1148c1bb86cdSEric Blake return -ENOTSUP; 1149c1bb86cdSEric Blake } 1150c1bb86cdSEric Blake ret = probe_logical_blocksize(s->fd, &bsz->log); 1151c1bb86cdSEric Blake if (ret < 0) { 1152c1bb86cdSEric Blake return ret; 1153c1bb86cdSEric Blake } 1154c1bb86cdSEric Blake return probe_physical_blocksize(s->fd, &bsz->phys); 1155c1bb86cdSEric Blake } 1156c1bb86cdSEric Blake 1157c1bb86cdSEric Blake /** 1158c1bb86cdSEric Blake * Try to get @bs's geometry: cyls, heads, sectors. 1159c1bb86cdSEric Blake * On success, store them in @geo and return 0. 1160c1bb86cdSEric Blake * On failure return -errno. 1161c1bb86cdSEric Blake * (Allows block driver to assign default geometry values that guest sees) 1162c1bb86cdSEric Blake */ 1163c1bb86cdSEric Blake #ifdef __linux__ 1164c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1165c1bb86cdSEric Blake { 1166c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1167c1bb86cdSEric Blake struct hd_geometry ioctl_geo = {0}; 1168c1bb86cdSEric Blake 1169c1bb86cdSEric Blake /* If DASD, get its geometry */ 1170c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1171c1bb86cdSEric Blake return -ENOTSUP; 1172c1bb86cdSEric Blake } 1173c1bb86cdSEric Blake if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { 1174c1bb86cdSEric Blake return -errno; 1175c1bb86cdSEric Blake } 1176c1bb86cdSEric Blake /* HDIO_GETGEO may return success even though geo contains zeros 1177c1bb86cdSEric Blake (e.g. certain multipath setups) */ 1178c1bb86cdSEric Blake if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { 1179c1bb86cdSEric Blake return -ENOTSUP; 1180c1bb86cdSEric Blake } 1181c1bb86cdSEric Blake /* Do not return a geometry for partition */ 1182c1bb86cdSEric Blake if (ioctl_geo.start != 0) { 1183c1bb86cdSEric Blake return -ENOTSUP; 1184c1bb86cdSEric Blake } 1185c1bb86cdSEric Blake geo->heads = ioctl_geo.heads; 1186c1bb86cdSEric Blake geo->sectors = ioctl_geo.sectors; 1187c1bb86cdSEric Blake geo->cylinders = ioctl_geo.cylinders; 1188c1bb86cdSEric Blake 1189c1bb86cdSEric Blake return 0; 1190c1bb86cdSEric Blake } 1191c1bb86cdSEric Blake #else /* __linux__ */ 1192c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1193c1bb86cdSEric Blake { 1194c1bb86cdSEric Blake return -ENOTSUP; 1195c1bb86cdSEric Blake } 1196c1bb86cdSEric Blake #endif 1197c1bb86cdSEric Blake 119803425671SKevin Wolf #if defined(__linux__) 119903425671SKevin Wolf static int handle_aiocb_ioctl(void *opaque) 1200c1bb86cdSEric Blake { 120103425671SKevin Wolf RawPosixAIOData *aiocb = opaque; 1202c1bb86cdSEric Blake int ret; 1203c1bb86cdSEric Blake 1204d57c44d0SKevin Wolf ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); 1205c1bb86cdSEric Blake if (ret == -1) { 1206c1bb86cdSEric Blake return -errno; 1207c1bb86cdSEric Blake } 1208c1bb86cdSEric Blake 1209c1bb86cdSEric Blake return 0; 1210c1bb86cdSEric Blake } 121103425671SKevin Wolf #endif /* linux */ 1212c1bb86cdSEric Blake 121306dc9bd5SKevin Wolf static int handle_aiocb_flush(void *opaque) 1214c1bb86cdSEric Blake { 121506dc9bd5SKevin Wolf RawPosixAIOData *aiocb = opaque; 1216e5bcf967SKevin Wolf BDRVRawState *s = aiocb->bs->opaque; 1217c1bb86cdSEric Blake int ret; 1218c1bb86cdSEric Blake 1219e5bcf967SKevin Wolf if (s->page_cache_inconsistent) { 1220e5bcf967SKevin Wolf return -EIO; 1221e5bcf967SKevin Wolf } 1222e5bcf967SKevin Wolf 1223c1bb86cdSEric Blake ret = qemu_fdatasync(aiocb->aio_fildes); 1224c1bb86cdSEric Blake if (ret == -1) { 1225e5bcf967SKevin Wolf /* There is no clear definition of the semantics of a failing fsync(), 1226e5bcf967SKevin Wolf * so we may have to assume the worst. The sad truth is that this 1227e5bcf967SKevin Wolf * assumption is correct for Linux. Some pages are now probably marked 1228e5bcf967SKevin Wolf * clean in the page cache even though they are inconsistent with the 1229e5bcf967SKevin Wolf * on-disk contents. The next fdatasync() call would succeed, but no 1230e5bcf967SKevin Wolf * further writeback attempt will be made. We can't get back to a state 1231e5bcf967SKevin Wolf * in which we know what is on disk (we would have to rewrite 1232e5bcf967SKevin Wolf * everything that was touched since the last fdatasync() at least), so 1233e5bcf967SKevin Wolf * make bdrv_flush() fail permanently. Given that the behaviour isn't 1234e5bcf967SKevin Wolf * really defined, I have little hope that other OSes are doing better. 1235e5bcf967SKevin Wolf * 1236e5bcf967SKevin Wolf * Obviously, this doesn't affect O_DIRECT, which bypasses the page 1237e5bcf967SKevin Wolf * cache. */ 1238e5bcf967SKevin Wolf if ((s->open_flags & O_DIRECT) == 0) { 1239e5bcf967SKevin Wolf s->page_cache_inconsistent = true; 1240e5bcf967SKevin Wolf } 1241c1bb86cdSEric Blake return -errno; 1242c1bb86cdSEric Blake } 1243c1bb86cdSEric Blake return 0; 1244c1bb86cdSEric Blake } 1245c1bb86cdSEric Blake 1246c1bb86cdSEric Blake #ifdef CONFIG_PREADV 1247c1bb86cdSEric Blake 1248c1bb86cdSEric Blake static bool preadv_present = true; 1249c1bb86cdSEric Blake 1250c1bb86cdSEric Blake static ssize_t 1251c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1252c1bb86cdSEric Blake { 1253c1bb86cdSEric Blake return preadv(fd, iov, nr_iov, offset); 1254c1bb86cdSEric Blake } 1255c1bb86cdSEric Blake 1256c1bb86cdSEric Blake static ssize_t 1257c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1258c1bb86cdSEric Blake { 1259c1bb86cdSEric Blake return pwritev(fd, iov, nr_iov, offset); 1260c1bb86cdSEric Blake } 1261c1bb86cdSEric Blake 1262c1bb86cdSEric Blake #else 1263c1bb86cdSEric Blake 1264c1bb86cdSEric Blake static bool preadv_present = false; 1265c1bb86cdSEric Blake 1266c1bb86cdSEric Blake static ssize_t 1267c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1268c1bb86cdSEric Blake { 1269c1bb86cdSEric Blake return -ENOSYS; 1270c1bb86cdSEric Blake } 1271c1bb86cdSEric Blake 1272c1bb86cdSEric Blake static ssize_t 1273c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1274c1bb86cdSEric Blake { 1275c1bb86cdSEric Blake return -ENOSYS; 1276c1bb86cdSEric Blake } 1277c1bb86cdSEric Blake 1278c1bb86cdSEric Blake #endif 1279c1bb86cdSEric Blake 1280c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) 1281c1bb86cdSEric Blake { 1282c1bb86cdSEric Blake ssize_t len; 1283c1bb86cdSEric Blake 1284c1bb86cdSEric Blake do { 1285c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) 1286c1bb86cdSEric Blake len = qemu_pwritev(aiocb->aio_fildes, 1287d57c44d0SKevin Wolf aiocb->io.iov, 1288d57c44d0SKevin Wolf aiocb->io.niov, 1289c1bb86cdSEric Blake aiocb->aio_offset); 1290c1bb86cdSEric Blake else 1291c1bb86cdSEric Blake len = qemu_preadv(aiocb->aio_fildes, 1292d57c44d0SKevin Wolf aiocb->io.iov, 1293d57c44d0SKevin Wolf aiocb->io.niov, 1294c1bb86cdSEric Blake aiocb->aio_offset); 1295c1bb86cdSEric Blake } while (len == -1 && errno == EINTR); 1296c1bb86cdSEric Blake 1297c1bb86cdSEric Blake if (len == -1) { 1298c1bb86cdSEric Blake return -errno; 1299c1bb86cdSEric Blake } 1300c1bb86cdSEric Blake return len; 1301c1bb86cdSEric Blake } 1302c1bb86cdSEric Blake 1303c1bb86cdSEric Blake /* 1304c1bb86cdSEric Blake * Read/writes the data to/from a given linear buffer. 1305c1bb86cdSEric Blake * 1306c1bb86cdSEric Blake * Returns the number of bytes handles or -errno in case of an error. Short 1307c1bb86cdSEric Blake * reads are only returned if the end of the file is reached. 1308c1bb86cdSEric Blake */ 1309c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) 1310c1bb86cdSEric Blake { 1311c1bb86cdSEric Blake ssize_t offset = 0; 1312c1bb86cdSEric Blake ssize_t len; 1313c1bb86cdSEric Blake 1314c1bb86cdSEric Blake while (offset < aiocb->aio_nbytes) { 1315c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1316c1bb86cdSEric Blake len = pwrite(aiocb->aio_fildes, 1317c1bb86cdSEric Blake (const char *)buf + offset, 1318c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1319c1bb86cdSEric Blake aiocb->aio_offset + offset); 1320c1bb86cdSEric Blake } else { 1321c1bb86cdSEric Blake len = pread(aiocb->aio_fildes, 1322c1bb86cdSEric Blake buf + offset, 1323c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1324c1bb86cdSEric Blake aiocb->aio_offset + offset); 1325c1bb86cdSEric Blake } 1326c1bb86cdSEric Blake if (len == -1 && errno == EINTR) { 1327c1bb86cdSEric Blake continue; 1328c1bb86cdSEric Blake } else if (len == -1 && errno == EINVAL && 1329c1bb86cdSEric Blake (aiocb->bs->open_flags & BDRV_O_NOCACHE) && 1330c1bb86cdSEric Blake !(aiocb->aio_type & QEMU_AIO_WRITE) && 1331c1bb86cdSEric Blake offset > 0) { 1332c1bb86cdSEric Blake /* O_DIRECT pread() may fail with EINVAL when offset is unaligned 1333c1bb86cdSEric Blake * after a short read. Assume that O_DIRECT short reads only occur 1334c1bb86cdSEric Blake * at EOF. Therefore this is a short read, not an I/O error. 1335c1bb86cdSEric Blake */ 1336c1bb86cdSEric Blake break; 1337c1bb86cdSEric Blake } else if (len == -1) { 1338c1bb86cdSEric Blake offset = -errno; 1339c1bb86cdSEric Blake break; 1340c1bb86cdSEric Blake } else if (len == 0) { 1341c1bb86cdSEric Blake break; 1342c1bb86cdSEric Blake } 1343c1bb86cdSEric Blake offset += len; 1344c1bb86cdSEric Blake } 1345c1bb86cdSEric Blake 1346c1bb86cdSEric Blake return offset; 1347c1bb86cdSEric Blake } 1348c1bb86cdSEric Blake 1349999e6b69SKevin Wolf static int handle_aiocb_rw(void *opaque) 1350c1bb86cdSEric Blake { 1351999e6b69SKevin Wolf RawPosixAIOData *aiocb = opaque; 1352c1bb86cdSEric Blake ssize_t nbytes; 1353c1bb86cdSEric Blake char *buf; 1354c1bb86cdSEric Blake 1355c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { 1356c1bb86cdSEric Blake /* 1357c1bb86cdSEric Blake * If there is just a single buffer, and it is properly aligned 1358c1bb86cdSEric Blake * we can just use plain pread/pwrite without any problems. 1359c1bb86cdSEric Blake */ 1360d57c44d0SKevin Wolf if (aiocb->io.niov == 1) { 136154c7ca1bSKevin Wolf nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); 136254c7ca1bSKevin Wolf goto out; 1363c1bb86cdSEric Blake } 1364c1bb86cdSEric Blake /* 1365c1bb86cdSEric Blake * We have more than one iovec, and all are properly aligned. 1366c1bb86cdSEric Blake * 1367c1bb86cdSEric Blake * Try preadv/pwritev first and fall back to linearizing the 1368c1bb86cdSEric Blake * buffer if it's not supported. 1369c1bb86cdSEric Blake */ 1370c1bb86cdSEric Blake if (preadv_present) { 1371c1bb86cdSEric Blake nbytes = handle_aiocb_rw_vector(aiocb); 1372c1bb86cdSEric Blake if (nbytes == aiocb->aio_nbytes || 1373c1bb86cdSEric Blake (nbytes < 0 && nbytes != -ENOSYS)) { 137454c7ca1bSKevin Wolf goto out; 1375c1bb86cdSEric Blake } 1376c1bb86cdSEric Blake preadv_present = false; 1377c1bb86cdSEric Blake } 1378c1bb86cdSEric Blake 1379c1bb86cdSEric Blake /* 1380c1bb86cdSEric Blake * XXX(hch): short read/write. no easy way to handle the reminder 1381c1bb86cdSEric Blake * using these interfaces. For now retry using plain 1382c1bb86cdSEric Blake * pread/pwrite? 1383c1bb86cdSEric Blake */ 1384c1bb86cdSEric Blake } 1385c1bb86cdSEric Blake 1386c1bb86cdSEric Blake /* 1387c1bb86cdSEric Blake * Ok, we have to do it the hard way, copy all segments into 1388c1bb86cdSEric Blake * a single aligned buffer. 1389c1bb86cdSEric Blake */ 1390c1bb86cdSEric Blake buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); 1391c1bb86cdSEric Blake if (buf == NULL) { 139254c7ca1bSKevin Wolf nbytes = -ENOMEM; 139354c7ca1bSKevin Wolf goto out; 1394c1bb86cdSEric Blake } 1395c1bb86cdSEric Blake 1396c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1397c1bb86cdSEric Blake char *p = buf; 1398c1bb86cdSEric Blake int i; 1399c1bb86cdSEric Blake 1400d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov; ++i) { 1401d57c44d0SKevin Wolf memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); 1402d57c44d0SKevin Wolf p += aiocb->io.iov[i].iov_len; 1403c1bb86cdSEric Blake } 1404c1bb86cdSEric Blake assert(p - buf == aiocb->aio_nbytes); 1405c1bb86cdSEric Blake } 1406c1bb86cdSEric Blake 1407c1bb86cdSEric Blake nbytes = handle_aiocb_rw_linear(aiocb, buf); 1408c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { 1409c1bb86cdSEric Blake char *p = buf; 1410c1bb86cdSEric Blake size_t count = aiocb->aio_nbytes, copy; 1411c1bb86cdSEric Blake int i; 1412c1bb86cdSEric Blake 1413d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov && count; ++i) { 1414c1bb86cdSEric Blake copy = count; 1415d57c44d0SKevin Wolf if (copy > aiocb->io.iov[i].iov_len) { 1416d57c44d0SKevin Wolf copy = aiocb->io.iov[i].iov_len; 1417c1bb86cdSEric Blake } 1418d57c44d0SKevin Wolf memcpy(aiocb->io.iov[i].iov_base, p, copy); 1419c1bb86cdSEric Blake assert(count >= copy); 1420c1bb86cdSEric Blake p += copy; 1421c1bb86cdSEric Blake count -= copy; 1422c1bb86cdSEric Blake } 1423c1bb86cdSEric Blake assert(count == 0); 1424c1bb86cdSEric Blake } 1425c1bb86cdSEric Blake qemu_vfree(buf); 1426c1bb86cdSEric Blake 142754c7ca1bSKevin Wolf out: 142854c7ca1bSKevin Wolf if (nbytes == aiocb->aio_nbytes) { 142954c7ca1bSKevin Wolf return 0; 143054c7ca1bSKevin Wolf } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { 143154c7ca1bSKevin Wolf if (aiocb->aio_type & QEMU_AIO_WRITE) { 143254c7ca1bSKevin Wolf return -EINVAL; 143354c7ca1bSKevin Wolf } else { 143454c7ca1bSKevin Wolf iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, 143554c7ca1bSKevin Wolf 0, aiocb->aio_nbytes - nbytes); 143654c7ca1bSKevin Wolf return 0; 143754c7ca1bSKevin Wolf } 143854c7ca1bSKevin Wolf } else { 143954c7ca1bSKevin Wolf assert(nbytes < 0); 1440c1bb86cdSEric Blake return nbytes; 1441c1bb86cdSEric Blake } 144254c7ca1bSKevin Wolf } 1443c1bb86cdSEric Blake 1444c1bb86cdSEric Blake #ifdef CONFIG_XFS 1445c1bb86cdSEric Blake static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes) 1446c1bb86cdSEric Blake { 1447c1bb86cdSEric Blake struct xfs_flock64 fl; 1448c1bb86cdSEric Blake int err; 1449c1bb86cdSEric Blake 1450c1bb86cdSEric Blake memset(&fl, 0, sizeof(fl)); 1451c1bb86cdSEric Blake fl.l_whence = SEEK_SET; 1452c1bb86cdSEric Blake fl.l_start = offset; 1453c1bb86cdSEric Blake fl.l_len = bytes; 1454c1bb86cdSEric Blake 1455c1bb86cdSEric Blake if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) { 1456c1bb86cdSEric Blake err = errno; 14574f7d28d7SLaurent Vivier trace_file_xfs_write_zeroes(strerror(errno)); 1458c1bb86cdSEric Blake return -err; 1459c1bb86cdSEric Blake } 1460c1bb86cdSEric Blake 1461c1bb86cdSEric Blake return 0; 1462c1bb86cdSEric Blake } 1463c1bb86cdSEric Blake 1464c1bb86cdSEric Blake static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes) 1465c1bb86cdSEric Blake { 1466c1bb86cdSEric Blake struct xfs_flock64 fl; 1467c1bb86cdSEric Blake int err; 1468c1bb86cdSEric Blake 1469c1bb86cdSEric Blake memset(&fl, 0, sizeof(fl)); 1470c1bb86cdSEric Blake fl.l_whence = SEEK_SET; 1471c1bb86cdSEric Blake fl.l_start = offset; 1472c1bb86cdSEric Blake fl.l_len = bytes; 1473c1bb86cdSEric Blake 1474c1bb86cdSEric Blake if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) { 1475c1bb86cdSEric Blake err = errno; 14764f7d28d7SLaurent Vivier trace_file_xfs_discard(strerror(errno)); 1477c1bb86cdSEric Blake return -err; 1478c1bb86cdSEric Blake } 1479c1bb86cdSEric Blake 1480c1bb86cdSEric Blake return 0; 1481c1bb86cdSEric Blake } 1482c1bb86cdSEric Blake #endif 1483c1bb86cdSEric Blake 1484c1bb86cdSEric Blake static int translate_err(int err) 1485c1bb86cdSEric Blake { 1486c1bb86cdSEric Blake if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || 1487c1bb86cdSEric Blake err == -ENOTTY) { 1488c1bb86cdSEric Blake err = -ENOTSUP; 1489c1bb86cdSEric Blake } 1490c1bb86cdSEric Blake return err; 1491c1bb86cdSEric Blake } 1492c1bb86cdSEric Blake 1493c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 1494c1bb86cdSEric Blake static int do_fallocate(int fd, int mode, off_t offset, off_t len) 1495c1bb86cdSEric Blake { 1496c1bb86cdSEric Blake do { 1497c1bb86cdSEric Blake if (fallocate(fd, mode, offset, len) == 0) { 1498c1bb86cdSEric Blake return 0; 1499c1bb86cdSEric Blake } 1500c1bb86cdSEric Blake } while (errno == EINTR); 1501c1bb86cdSEric Blake return translate_err(-errno); 1502c1bb86cdSEric Blake } 1503c1bb86cdSEric Blake #endif 1504c1bb86cdSEric Blake 1505c1bb86cdSEric Blake static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) 1506c1bb86cdSEric Blake { 1507c1bb86cdSEric Blake int ret = -ENOTSUP; 1508c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1509c1bb86cdSEric Blake 1510c1bb86cdSEric Blake if (!s->has_write_zeroes) { 1511c1bb86cdSEric Blake return -ENOTSUP; 1512c1bb86cdSEric Blake } 1513c1bb86cdSEric Blake 1514c1bb86cdSEric Blake #ifdef BLKZEROOUT 1515738301e1SKevin Wolf /* The BLKZEROOUT implementation in the kernel doesn't set 1516738301e1SKevin Wolf * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow 1517738301e1SKevin Wolf * fallbacks. */ 1518738301e1SKevin Wolf if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { 1519c1bb86cdSEric Blake do { 1520c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1521c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { 1522c1bb86cdSEric Blake return 0; 1523c1bb86cdSEric Blake } 1524c1bb86cdSEric Blake } while (errno == EINTR); 1525c1bb86cdSEric Blake 1526c1bb86cdSEric Blake ret = translate_err(-errno); 1527738301e1SKevin Wolf } 1528c1bb86cdSEric Blake #endif 1529c1bb86cdSEric Blake 1530c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1531c1bb86cdSEric Blake s->has_write_zeroes = false; 1532c1bb86cdSEric Blake } 1533c1bb86cdSEric Blake return ret; 1534c1bb86cdSEric Blake } 1535c1bb86cdSEric Blake 15367154d8aeSKevin Wolf static int handle_aiocb_write_zeroes(void *opaque) 1537c1bb86cdSEric Blake { 15387154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 1539c1bb86cdSEric Blake #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS) 1540c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1541c1bb86cdSEric Blake #endif 154270d9110bSDenis V. Lunev #ifdef CONFIG_FALLOCATE 154370d9110bSDenis V. Lunev int64_t len; 154470d9110bSDenis V. Lunev #endif 1545c1bb86cdSEric Blake 1546c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1547c1bb86cdSEric Blake return handle_aiocb_write_zeroes_block(aiocb); 1548c1bb86cdSEric Blake } 1549c1bb86cdSEric Blake 1550c1bb86cdSEric Blake #ifdef CONFIG_XFS 1551c1bb86cdSEric Blake if (s->is_xfs) { 1552c1bb86cdSEric Blake return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes); 1553c1bb86cdSEric Blake } 1554c1bb86cdSEric Blake #endif 1555c1bb86cdSEric Blake 1556c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_ZERO_RANGE 1557c1bb86cdSEric Blake if (s->has_write_zeroes) { 1558c1bb86cdSEric Blake int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, 1559c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1560c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1561c1bb86cdSEric Blake return ret; 1562c1bb86cdSEric Blake } 1563c1bb86cdSEric Blake s->has_write_zeroes = false; 1564c1bb86cdSEric Blake } 1565c1bb86cdSEric Blake #endif 1566c1bb86cdSEric Blake 1567c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1568c1bb86cdSEric Blake if (s->has_discard && s->has_fallocate) { 1569c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 1570c1bb86cdSEric Blake FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1571c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1572c1bb86cdSEric Blake if (ret == 0) { 1573c1bb86cdSEric Blake ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1574c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1575c1bb86cdSEric Blake return ret; 1576c1bb86cdSEric Blake } 1577c1bb86cdSEric Blake s->has_fallocate = false; 1578c1bb86cdSEric Blake } else if (ret != -ENOTSUP) { 1579c1bb86cdSEric Blake return ret; 1580c1bb86cdSEric Blake } else { 1581c1bb86cdSEric Blake s->has_discard = false; 1582c1bb86cdSEric Blake } 1583c1bb86cdSEric Blake } 1584c1bb86cdSEric Blake #endif 1585c1bb86cdSEric Blake 1586c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 158770d9110bSDenis V. Lunev /* Last resort: we are trying to extend the file with zeroed data. This 158870d9110bSDenis V. Lunev * can be done via fallocate(fd, 0) */ 158970d9110bSDenis V. Lunev len = bdrv_getlength(aiocb->bs); 159070d9110bSDenis V. Lunev if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { 1591c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1592c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1593c1bb86cdSEric Blake return ret; 1594c1bb86cdSEric Blake } 1595c1bb86cdSEric Blake s->has_fallocate = false; 1596c1bb86cdSEric Blake } 1597c1bb86cdSEric Blake #endif 1598c1bb86cdSEric Blake 1599c1bb86cdSEric Blake return -ENOTSUP; 1600c1bb86cdSEric Blake } 1601c1bb86cdSEric Blake 16027154d8aeSKevin Wolf static int handle_aiocb_write_zeroes_unmap(void *opaque) 160334fa110eSKevin Wolf { 16047154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 160534fa110eSKevin Wolf BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; 160634fa110eSKevin Wolf int ret; 160734fa110eSKevin Wolf 160834fa110eSKevin Wolf /* First try to write zeros and unmap at the same time */ 160934fa110eSKevin Wolf 161034fa110eSKevin Wolf #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 161134fa110eSKevin Wolf ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 161234fa110eSKevin Wolf aiocb->aio_offset, aiocb->aio_nbytes); 161334fa110eSKevin Wolf if (ret != -ENOTSUP) { 161434fa110eSKevin Wolf return ret; 161534fa110eSKevin Wolf } 161634fa110eSKevin Wolf #endif 161734fa110eSKevin Wolf 161834fa110eSKevin Wolf #ifdef CONFIG_XFS 161934fa110eSKevin Wolf if (s->is_xfs) { 162034fa110eSKevin Wolf /* xfs_discard() guarantees that the discarded area reads as all-zero 162134fa110eSKevin Wolf * afterwards, so we can use it here. */ 162234fa110eSKevin Wolf return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes); 162334fa110eSKevin Wolf } 162434fa110eSKevin Wolf #endif 162534fa110eSKevin Wolf 162634fa110eSKevin Wolf /* If we couldn't manage to unmap while guaranteed that the area reads as 162734fa110eSKevin Wolf * all-zero afterwards, just write zeroes without unmapping */ 162834fa110eSKevin Wolf ret = handle_aiocb_write_zeroes(aiocb); 162934fa110eSKevin Wolf return ret; 163034fa110eSKevin Wolf } 163134fa110eSKevin Wolf 16321efad060SFam Zheng #ifndef HAVE_COPY_FILE_RANGE 16331efad060SFam Zheng static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, 16341efad060SFam Zheng off_t *out_off, size_t len, unsigned int flags) 16351efad060SFam Zheng { 16361efad060SFam Zheng #ifdef __NR_copy_file_range 16371efad060SFam Zheng return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, 16381efad060SFam Zheng out_off, len, flags); 16391efad060SFam Zheng #else 16401efad060SFam Zheng errno = ENOSYS; 16411efad060SFam Zheng return -1; 16421efad060SFam Zheng #endif 16431efad060SFam Zheng } 16441efad060SFam Zheng #endif 16451efad060SFam Zheng 164658a209c4SKevin Wolf static int handle_aiocb_copy_range(void *opaque) 16471efad060SFam Zheng { 164858a209c4SKevin Wolf RawPosixAIOData *aiocb = opaque; 16491efad060SFam Zheng uint64_t bytes = aiocb->aio_nbytes; 16501efad060SFam Zheng off_t in_off = aiocb->aio_offset; 1651d57c44d0SKevin Wolf off_t out_off = aiocb->copy_range.aio_offset2; 16521efad060SFam Zheng 16531efad060SFam Zheng while (bytes) { 16541efad060SFam Zheng ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, 1655d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, &out_off, 16561efad060SFam Zheng bytes, 0); 1657ecc983a5SFam Zheng trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, 1658d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, out_off, bytes, 1659d57c44d0SKevin Wolf 0, ret); 1660c436e3d0SFam Zheng if (ret == 0) { 1661c436e3d0SFam Zheng /* No progress (e.g. when beyond EOF), let the caller fall back to 1662c436e3d0SFam Zheng * buffer I/O. */ 1663c436e3d0SFam Zheng return -ENOSPC; 16641efad060SFam Zheng } 16651efad060SFam Zheng if (ret < 0) { 1666c436e3d0SFam Zheng switch (errno) { 1667c436e3d0SFam Zheng case ENOSYS: 16681efad060SFam Zheng return -ENOTSUP; 1669c436e3d0SFam Zheng case EINTR: 1670c436e3d0SFam Zheng continue; 1671c436e3d0SFam Zheng default: 16721efad060SFam Zheng return -errno; 16731efad060SFam Zheng } 16741efad060SFam Zheng } 16751efad060SFam Zheng bytes -= ret; 16761efad060SFam Zheng } 16771efad060SFam Zheng return 0; 16781efad060SFam Zheng } 16791efad060SFam Zheng 168046ee0f46SKevin Wolf static int handle_aiocb_discard(void *opaque) 1681c1bb86cdSEric Blake { 168246ee0f46SKevin Wolf RawPosixAIOData *aiocb = opaque; 1683c1bb86cdSEric Blake int ret = -EOPNOTSUPP; 1684c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1685c1bb86cdSEric Blake 1686c1bb86cdSEric Blake if (!s->has_discard) { 1687c1bb86cdSEric Blake return -ENOTSUP; 1688c1bb86cdSEric Blake } 1689c1bb86cdSEric Blake 1690c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1691c1bb86cdSEric Blake #ifdef BLKDISCARD 1692c1bb86cdSEric Blake do { 1693c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1694c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { 1695c1bb86cdSEric Blake return 0; 1696c1bb86cdSEric Blake } 1697c1bb86cdSEric Blake } while (errno == EINTR); 1698c1bb86cdSEric Blake 1699c1bb86cdSEric Blake ret = -errno; 1700c1bb86cdSEric Blake #endif 1701c1bb86cdSEric Blake } else { 1702c1bb86cdSEric Blake #ifdef CONFIG_XFS 1703c1bb86cdSEric Blake if (s->is_xfs) { 1704c1bb86cdSEric Blake return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes); 1705c1bb86cdSEric Blake } 1706c1bb86cdSEric Blake #endif 1707c1bb86cdSEric Blake 1708c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1709c1bb86cdSEric Blake ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1710c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1711c1bb86cdSEric Blake #endif 1712c1bb86cdSEric Blake } 1713c1bb86cdSEric Blake 1714c1bb86cdSEric Blake ret = translate_err(ret); 1715c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1716c1bb86cdSEric Blake s->has_discard = false; 1717c1bb86cdSEric Blake } 1718c1bb86cdSEric Blake return ret; 1719c1bb86cdSEric Blake } 1720c1bb86cdSEric Blake 172129cb4c01SKevin Wolf static int handle_aiocb_truncate(void *opaque) 172293f4e2ffSKevin Wolf { 172329cb4c01SKevin Wolf RawPosixAIOData *aiocb = opaque; 172493f4e2ffSKevin Wolf int result = 0; 172593f4e2ffSKevin Wolf int64_t current_length = 0; 172693f4e2ffSKevin Wolf char *buf = NULL; 172793f4e2ffSKevin Wolf struct stat st; 172893f4e2ffSKevin Wolf int fd = aiocb->aio_fildes; 172993f4e2ffSKevin Wolf int64_t offset = aiocb->aio_offset; 1730d57c44d0SKevin Wolf PreallocMode prealloc = aiocb->truncate.prealloc; 1731d57c44d0SKevin Wolf Error **errp = aiocb->truncate.errp; 173293f4e2ffSKevin Wolf 173393f4e2ffSKevin Wolf if (fstat(fd, &st) < 0) { 173493f4e2ffSKevin Wolf result = -errno; 173593f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not stat file"); 173693f4e2ffSKevin Wolf return result; 173793f4e2ffSKevin Wolf } 173893f4e2ffSKevin Wolf 173993f4e2ffSKevin Wolf current_length = st.st_size; 1740d57c44d0SKevin Wolf if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { 174193f4e2ffSKevin Wolf error_setg(errp, "Cannot use preallocation for shrinking files"); 174293f4e2ffSKevin Wolf return -ENOTSUP; 174393f4e2ffSKevin Wolf } 174493f4e2ffSKevin Wolf 1745d57c44d0SKevin Wolf switch (prealloc) { 174693f4e2ffSKevin Wolf #ifdef CONFIG_POSIX_FALLOCATE 174793f4e2ffSKevin Wolf case PREALLOC_MODE_FALLOC: 174893f4e2ffSKevin Wolf /* 174993f4e2ffSKevin Wolf * Truncating before posix_fallocate() makes it about twice slower on 175093f4e2ffSKevin Wolf * file systems that do not support fallocate(), trying to check if a 175193f4e2ffSKevin Wolf * block is allocated before allocating it, so don't do that here. 175293f4e2ffSKevin Wolf */ 175393f4e2ffSKevin Wolf if (offset != current_length) { 175493f4e2ffSKevin Wolf result = -posix_fallocate(fd, current_length, 175593f4e2ffSKevin Wolf offset - current_length); 175693f4e2ffSKevin Wolf if (result != 0) { 175793f4e2ffSKevin Wolf /* posix_fallocate() doesn't set errno. */ 175893f4e2ffSKevin Wolf error_setg_errno(errp, -result, 175993f4e2ffSKevin Wolf "Could not preallocate new data"); 176093f4e2ffSKevin Wolf } 176193f4e2ffSKevin Wolf } else { 176293f4e2ffSKevin Wolf result = 0; 176393f4e2ffSKevin Wolf } 176493f4e2ffSKevin Wolf goto out; 176593f4e2ffSKevin Wolf #endif 176693f4e2ffSKevin Wolf case PREALLOC_MODE_FULL: 176793f4e2ffSKevin Wolf { 176893f4e2ffSKevin Wolf int64_t num = 0, left = offset - current_length; 176993f4e2ffSKevin Wolf off_t seek_result; 177093f4e2ffSKevin Wolf 177193f4e2ffSKevin Wolf /* 177293f4e2ffSKevin Wolf * Knowing the final size from the beginning could allow the file 177393f4e2ffSKevin Wolf * system driver to do less allocations and possibly avoid 177493f4e2ffSKevin Wolf * fragmentation of the file. 177593f4e2ffSKevin Wolf */ 177693f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 177793f4e2ffSKevin Wolf result = -errno; 177893f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 177993f4e2ffSKevin Wolf goto out; 178093f4e2ffSKevin Wolf } 178193f4e2ffSKevin Wolf 178293f4e2ffSKevin Wolf buf = g_malloc0(65536); 178393f4e2ffSKevin Wolf 178493f4e2ffSKevin Wolf seek_result = lseek(fd, current_length, SEEK_SET); 178593f4e2ffSKevin Wolf if (seek_result < 0) { 178693f4e2ffSKevin Wolf result = -errno; 178793f4e2ffSKevin Wolf error_setg_errno(errp, -result, 178893f4e2ffSKevin Wolf "Failed to seek to the old end of file"); 178993f4e2ffSKevin Wolf goto out; 179093f4e2ffSKevin Wolf } 179193f4e2ffSKevin Wolf 179293f4e2ffSKevin Wolf while (left > 0) { 179393f4e2ffSKevin Wolf num = MIN(left, 65536); 179493f4e2ffSKevin Wolf result = write(fd, buf, num); 179593f4e2ffSKevin Wolf if (result < 0) { 1796a1c81f4fSFam Zheng if (errno == EINTR) { 1797a1c81f4fSFam Zheng continue; 1798a1c81f4fSFam Zheng } 179993f4e2ffSKevin Wolf result = -errno; 180093f4e2ffSKevin Wolf error_setg_errno(errp, -result, 180193f4e2ffSKevin Wolf "Could not write zeros for preallocation"); 180293f4e2ffSKevin Wolf goto out; 180393f4e2ffSKevin Wolf } 180493f4e2ffSKevin Wolf left -= result; 180593f4e2ffSKevin Wolf } 180693f4e2ffSKevin Wolf if (result >= 0) { 180793f4e2ffSKevin Wolf result = fsync(fd); 180893f4e2ffSKevin Wolf if (result < 0) { 180993f4e2ffSKevin Wolf result = -errno; 181093f4e2ffSKevin Wolf error_setg_errno(errp, -result, 181193f4e2ffSKevin Wolf "Could not flush file to disk"); 181293f4e2ffSKevin Wolf goto out; 181393f4e2ffSKevin Wolf } 181493f4e2ffSKevin Wolf } 181593f4e2ffSKevin Wolf goto out; 181693f4e2ffSKevin Wolf } 181793f4e2ffSKevin Wolf case PREALLOC_MODE_OFF: 181893f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 181993f4e2ffSKevin Wolf result = -errno; 182093f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 182193f4e2ffSKevin Wolf } 182293f4e2ffSKevin Wolf return result; 182393f4e2ffSKevin Wolf default: 182493f4e2ffSKevin Wolf result = -ENOTSUP; 182593f4e2ffSKevin Wolf error_setg(errp, "Unsupported preallocation mode: %s", 1826d57c44d0SKevin Wolf PreallocMode_str(prealloc)); 182793f4e2ffSKevin Wolf return result; 182893f4e2ffSKevin Wolf } 182993f4e2ffSKevin Wolf 183093f4e2ffSKevin Wolf out: 183193f4e2ffSKevin Wolf if (result < 0) { 183293f4e2ffSKevin Wolf if (ftruncate(fd, current_length) < 0) { 183393f4e2ffSKevin Wolf error_report("Failed to restore old file length: %s", 183493f4e2ffSKevin Wolf strerror(errno)); 183593f4e2ffSKevin Wolf } 183693f4e2ffSKevin Wolf } 183793f4e2ffSKevin Wolf 183893f4e2ffSKevin Wolf g_free(buf); 183993f4e2ffSKevin Wolf return result; 184093f4e2ffSKevin Wolf } 184193f4e2ffSKevin Wolf 18425d5de250SKevin Wolf static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs, 18435d5de250SKevin Wolf ThreadPoolFunc func, void *arg) 18445d5de250SKevin Wolf { 18455d5de250SKevin Wolf /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ 18465d5de250SKevin Wolf ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 18475d5de250SKevin Wolf return thread_pool_submit_co(pool, func, arg); 18485d5de250SKevin Wolf } 18495d5de250SKevin Wolf 1850c1bb86cdSEric Blake static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, 1851c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, int type) 1852c1bb86cdSEric Blake { 1853c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1854999e6b69SKevin Wolf RawPosixAIOData acb; 1855c1bb86cdSEric Blake 1856c1bb86cdSEric Blake if (fd_open(bs) < 0) 1857c1bb86cdSEric Blake return -EIO; 1858c1bb86cdSEric Blake 1859c1bb86cdSEric Blake /* 1860c1bb86cdSEric Blake * Check if the underlying device requires requests to be aligned, 1861c1bb86cdSEric Blake * and if the request we are trying to submit is aligned or not. 1862c1bb86cdSEric Blake * If this is the case tell the low-level driver that it needs 1863c1bb86cdSEric Blake * to copy the buffer. 1864c1bb86cdSEric Blake */ 1865c1bb86cdSEric Blake if (s->needs_alignment) { 1866c1bb86cdSEric Blake if (!bdrv_qiov_is_aligned(bs, qiov)) { 1867c1bb86cdSEric Blake type |= QEMU_AIO_MISALIGNED; 1868c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1869c1bb86cdSEric Blake } else if (s->use_linux_aio) { 1870c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1871c1bb86cdSEric Blake assert(qiov->size == bytes); 1872c1bb86cdSEric Blake return laio_co_submit(bs, aio, s->fd, offset, qiov, type); 1873c1bb86cdSEric Blake #endif 1874c1bb86cdSEric Blake } 1875c1bb86cdSEric Blake } 1876c1bb86cdSEric Blake 1877999e6b69SKevin Wolf acb = (RawPosixAIOData) { 1878999e6b69SKevin Wolf .bs = bs, 1879999e6b69SKevin Wolf .aio_fildes = s->fd, 1880999e6b69SKevin Wolf .aio_type = type, 1881999e6b69SKevin Wolf .aio_offset = offset, 1882999e6b69SKevin Wolf .aio_nbytes = bytes, 1883999e6b69SKevin Wolf .io = { 1884999e6b69SKevin Wolf .iov = qiov->iov, 1885999e6b69SKevin Wolf .niov = qiov->niov, 1886999e6b69SKevin Wolf }, 1887999e6b69SKevin Wolf }; 1888999e6b69SKevin Wolf 1889999e6b69SKevin Wolf assert(qiov->size == bytes); 1890999e6b69SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb); 1891c1bb86cdSEric Blake } 1892c1bb86cdSEric Blake 1893c1bb86cdSEric Blake static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset, 1894c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 1895c1bb86cdSEric Blake int flags) 1896c1bb86cdSEric Blake { 1897c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); 1898c1bb86cdSEric Blake } 1899c1bb86cdSEric Blake 1900c1bb86cdSEric Blake static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset, 1901c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 1902c1bb86cdSEric Blake int flags) 1903c1bb86cdSEric Blake { 1904c1bb86cdSEric Blake assert(flags == 0); 1905c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); 1906c1bb86cdSEric Blake } 1907c1bb86cdSEric Blake 1908c1bb86cdSEric Blake static void raw_aio_plug(BlockDriverState *bs) 1909c1bb86cdSEric Blake { 1910c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1911c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1912c1bb86cdSEric Blake if (s->use_linux_aio) { 1913c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1914c1bb86cdSEric Blake laio_io_plug(bs, aio); 1915c1bb86cdSEric Blake } 1916c1bb86cdSEric Blake #endif 1917c1bb86cdSEric Blake } 1918c1bb86cdSEric Blake 1919c1bb86cdSEric Blake static void raw_aio_unplug(BlockDriverState *bs) 1920c1bb86cdSEric Blake { 1921c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1922c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1923c1bb86cdSEric Blake if (s->use_linux_aio) { 1924c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1925c1bb86cdSEric Blake laio_io_unplug(bs, aio); 1926c1bb86cdSEric Blake } 1927c1bb86cdSEric Blake #endif 1928c1bb86cdSEric Blake } 1929c1bb86cdSEric Blake 193033d70fb6SKevin Wolf static int raw_co_flush_to_disk(BlockDriverState *bs) 1931c1bb86cdSEric Blake { 1932c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 193306dc9bd5SKevin Wolf RawPosixAIOData acb; 193433d70fb6SKevin Wolf int ret; 1935c1bb86cdSEric Blake 193633d70fb6SKevin Wolf ret = fd_open(bs); 193733d70fb6SKevin Wolf if (ret < 0) { 193833d70fb6SKevin Wolf return ret; 193933d70fb6SKevin Wolf } 1940c1bb86cdSEric Blake 194106dc9bd5SKevin Wolf acb = (RawPosixAIOData) { 194206dc9bd5SKevin Wolf .bs = bs, 194306dc9bd5SKevin Wolf .aio_fildes = s->fd, 194406dc9bd5SKevin Wolf .aio_type = QEMU_AIO_FLUSH, 194506dc9bd5SKevin Wolf }; 194606dc9bd5SKevin Wolf 194706dc9bd5SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb); 1948c1bb86cdSEric Blake } 1949c1bb86cdSEric Blake 1950ed6e2161SNishanth Aravamudan static void raw_aio_attach_aio_context(BlockDriverState *bs, 1951ed6e2161SNishanth Aravamudan AioContext *new_context) 1952ed6e2161SNishanth Aravamudan { 1953ed6e2161SNishanth Aravamudan #ifdef CONFIG_LINUX_AIO 1954ed6e2161SNishanth Aravamudan BDRVRawState *s = bs->opaque; 1955ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 1956ed6e2161SNishanth Aravamudan Error *local_err; 1957ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(new_context, &local_err)) { 1958ed6e2161SNishanth Aravamudan error_reportf_err(local_err, "Unable to use native AIO, " 1959ed6e2161SNishanth Aravamudan "falling back to thread pool: "); 1960ed6e2161SNishanth Aravamudan s->use_linux_aio = false; 1961ed6e2161SNishanth Aravamudan } 1962ed6e2161SNishanth Aravamudan } 1963ed6e2161SNishanth Aravamudan #endif 1964ed6e2161SNishanth Aravamudan } 1965ed6e2161SNishanth Aravamudan 1966c1bb86cdSEric Blake static void raw_close(BlockDriverState *bs) 1967c1bb86cdSEric Blake { 1968c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1969c1bb86cdSEric Blake 1970c1bb86cdSEric Blake if (s->fd >= 0) { 1971c1bb86cdSEric Blake qemu_close(s->fd); 1972c1bb86cdSEric Blake s->fd = -1; 1973c1bb86cdSEric Blake } 1974c1bb86cdSEric Blake } 1975c1bb86cdSEric Blake 1976d0bc9e5dSMax Reitz /** 1977d0bc9e5dSMax Reitz * Truncates the given regular file @fd to @offset and, when growing, fills the 1978d0bc9e5dSMax Reitz * new space according to @prealloc. 1979d0bc9e5dSMax Reitz * 1980d0bc9e5dSMax Reitz * Returns: 0 on success, -errno on failure. 1981d0bc9e5dSMax Reitz */ 198293f4e2ffSKevin Wolf static int coroutine_fn 198393f4e2ffSKevin Wolf raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, 198493f4e2ffSKevin Wolf PreallocMode prealloc, Error **errp) 19859f63b07eSMax Reitz { 198629cb4c01SKevin Wolf RawPosixAIOData acb; 1987d0bc9e5dSMax Reitz 198829cb4c01SKevin Wolf acb = (RawPosixAIOData) { 198993f4e2ffSKevin Wolf .bs = bs, 199093f4e2ffSKevin Wolf .aio_fildes = fd, 199193f4e2ffSKevin Wolf .aio_type = QEMU_AIO_TRUNCATE, 199293f4e2ffSKevin Wolf .aio_offset = offset, 1993d57c44d0SKevin Wolf .truncate = { 199493f4e2ffSKevin Wolf .prealloc = prealloc, 199593f4e2ffSKevin Wolf .errp = errp, 1996d57c44d0SKevin Wolf }, 199793f4e2ffSKevin Wolf }; 1998d0bc9e5dSMax Reitz 199929cb4c01SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb); 20009f63b07eSMax Reitz } 20019f63b07eSMax Reitz 2002061ca8a3SKevin Wolf static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, 20038243ccb7SMax Reitz PreallocMode prealloc, Error **errp) 2004c1bb86cdSEric Blake { 2005c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2006c1bb86cdSEric Blake struct stat st; 2007f59adb32SMax Reitz int ret; 2008c1bb86cdSEric Blake 2009c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 2010f59adb32SMax Reitz ret = -errno; 2011f59adb32SMax Reitz error_setg_errno(errp, -ret, "Failed to fstat() the file"); 2012f59adb32SMax Reitz return ret; 2013c1bb86cdSEric Blake } 2014c1bb86cdSEric Blake 2015c1bb86cdSEric Blake if (S_ISREG(st.st_mode)) { 201693f4e2ffSKevin Wolf return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); 2017c1bb86cdSEric Blake } 201835d72602SMax Reitz 201935d72602SMax Reitz if (prealloc != PREALLOC_MODE_OFF) { 202035d72602SMax Reitz error_setg(errp, "Preallocation mode '%s' unsupported for this " 2021977c736fSMarkus Armbruster "non-regular file", PreallocMode_str(prealloc)); 202235d72602SMax Reitz return -ENOTSUP; 202335d72602SMax Reitz } 202435d72602SMax Reitz 202535d72602SMax Reitz if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2026c1bb86cdSEric Blake if (offset > raw_getlength(bs)) { 2027f59adb32SMax Reitz error_setg(errp, "Cannot grow device files"); 2028c1bb86cdSEric Blake return -EINVAL; 2029c1bb86cdSEric Blake } 2030c1bb86cdSEric Blake } else { 2031f59adb32SMax Reitz error_setg(errp, "Resizing this file is not supported"); 2032c1bb86cdSEric Blake return -ENOTSUP; 2033c1bb86cdSEric Blake } 2034c1bb86cdSEric Blake 2035c1bb86cdSEric Blake return 0; 2036c1bb86cdSEric Blake } 2037c1bb86cdSEric Blake 2038c1bb86cdSEric Blake #ifdef __OpenBSD__ 2039c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2040c1bb86cdSEric Blake { 2041c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2042c1bb86cdSEric Blake int fd = s->fd; 2043c1bb86cdSEric Blake struct stat st; 2044c1bb86cdSEric Blake 2045c1bb86cdSEric Blake if (fstat(fd, &st)) 2046c1bb86cdSEric Blake return -errno; 2047c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2048c1bb86cdSEric Blake struct disklabel dl; 2049c1bb86cdSEric Blake 2050c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2051c1bb86cdSEric Blake return -errno; 2052c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2053c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2054c1bb86cdSEric Blake } else 2055c1bb86cdSEric Blake return st.st_size; 2056c1bb86cdSEric Blake } 2057c1bb86cdSEric Blake #elif defined(__NetBSD__) 2058c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2059c1bb86cdSEric Blake { 2060c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2061c1bb86cdSEric Blake int fd = s->fd; 2062c1bb86cdSEric Blake struct stat st; 2063c1bb86cdSEric Blake 2064c1bb86cdSEric Blake if (fstat(fd, &st)) 2065c1bb86cdSEric Blake return -errno; 2066c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2067c1bb86cdSEric Blake struct dkwedge_info dkw; 2068c1bb86cdSEric Blake 2069c1bb86cdSEric Blake if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { 2070c1bb86cdSEric Blake return dkw.dkw_size * 512; 2071c1bb86cdSEric Blake } else { 2072c1bb86cdSEric Blake struct disklabel dl; 2073c1bb86cdSEric Blake 2074c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2075c1bb86cdSEric Blake return -errno; 2076c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2077c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2078c1bb86cdSEric Blake } 2079c1bb86cdSEric Blake } else 2080c1bb86cdSEric Blake return st.st_size; 2081c1bb86cdSEric Blake } 2082c1bb86cdSEric Blake #elif defined(__sun__) 2083c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2084c1bb86cdSEric Blake { 2085c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2086c1bb86cdSEric Blake struct dk_minfo minfo; 2087c1bb86cdSEric Blake int ret; 2088c1bb86cdSEric Blake int64_t size; 2089c1bb86cdSEric Blake 2090c1bb86cdSEric Blake ret = fd_open(bs); 2091c1bb86cdSEric Blake if (ret < 0) { 2092c1bb86cdSEric Blake return ret; 2093c1bb86cdSEric Blake } 2094c1bb86cdSEric Blake 2095c1bb86cdSEric Blake /* 2096c1bb86cdSEric Blake * Use the DKIOCGMEDIAINFO ioctl to read the size. 2097c1bb86cdSEric Blake */ 2098c1bb86cdSEric Blake ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); 2099c1bb86cdSEric Blake if (ret != -1) { 2100c1bb86cdSEric Blake return minfo.dki_lbsize * minfo.dki_capacity; 2101c1bb86cdSEric Blake } 2102c1bb86cdSEric Blake 2103c1bb86cdSEric Blake /* 2104c1bb86cdSEric Blake * There are reports that lseek on some devices fails, but 2105c1bb86cdSEric Blake * irc discussion said that contingency on contingency was overkill. 2106c1bb86cdSEric Blake */ 2107c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2108c1bb86cdSEric Blake if (size < 0) { 2109c1bb86cdSEric Blake return -errno; 2110c1bb86cdSEric Blake } 2111c1bb86cdSEric Blake return size; 2112c1bb86cdSEric Blake } 2113c1bb86cdSEric Blake #elif defined(CONFIG_BSD) 2114c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2115c1bb86cdSEric Blake { 2116c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2117c1bb86cdSEric Blake int fd = s->fd; 2118c1bb86cdSEric Blake int64_t size; 2119c1bb86cdSEric Blake struct stat sb; 2120c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2121c1bb86cdSEric Blake int reopened = 0; 2122c1bb86cdSEric Blake #endif 2123c1bb86cdSEric Blake int ret; 2124c1bb86cdSEric Blake 2125c1bb86cdSEric Blake ret = fd_open(bs); 2126c1bb86cdSEric Blake if (ret < 0) 2127c1bb86cdSEric Blake return ret; 2128c1bb86cdSEric Blake 2129c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2130c1bb86cdSEric Blake again: 2131c1bb86cdSEric Blake #endif 2132c1bb86cdSEric Blake if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { 2133c1bb86cdSEric Blake #ifdef DIOCGMEDIASIZE 2134c1bb86cdSEric Blake if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) 2135c1bb86cdSEric Blake #elif defined(DIOCGPART) 2136c1bb86cdSEric Blake { 2137c1bb86cdSEric Blake struct partinfo pi; 2138c1bb86cdSEric Blake if (ioctl(fd, DIOCGPART, &pi) == 0) 2139c1bb86cdSEric Blake size = pi.media_size; 2140c1bb86cdSEric Blake else 2141c1bb86cdSEric Blake size = 0; 2142c1bb86cdSEric Blake } 2143c1bb86cdSEric Blake if (size == 0) 2144c1bb86cdSEric Blake #endif 2145c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2146c1bb86cdSEric Blake { 2147c1bb86cdSEric Blake uint64_t sectors = 0; 2148c1bb86cdSEric Blake uint32_t sector_size = 0; 2149c1bb86cdSEric Blake 2150c1bb86cdSEric Blake if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 2151c1bb86cdSEric Blake && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { 2152c1bb86cdSEric Blake size = sectors * sector_size; 2153c1bb86cdSEric Blake } else { 2154c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2155c1bb86cdSEric Blake if (size < 0) { 2156c1bb86cdSEric Blake return -errno; 2157c1bb86cdSEric Blake } 2158c1bb86cdSEric Blake } 2159c1bb86cdSEric Blake } 2160c1bb86cdSEric Blake #else 2161c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2162c1bb86cdSEric Blake if (size < 0) { 2163c1bb86cdSEric Blake return -errno; 2164c1bb86cdSEric Blake } 2165c1bb86cdSEric Blake #endif 2166c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 2167c1bb86cdSEric Blake switch(s->type) { 2168c1bb86cdSEric Blake case FTYPE_CD: 2169c1bb86cdSEric Blake /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ 2170c1bb86cdSEric Blake if (size == 2048LL * (unsigned)-1) 2171c1bb86cdSEric Blake size = 0; 2172c1bb86cdSEric Blake /* XXX no disc? maybe we need to reopen... */ 2173c1bb86cdSEric Blake if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { 2174c1bb86cdSEric Blake reopened = 1; 2175c1bb86cdSEric Blake goto again; 2176c1bb86cdSEric Blake } 2177c1bb86cdSEric Blake } 2178c1bb86cdSEric Blake #endif 2179c1bb86cdSEric Blake } else { 2180c1bb86cdSEric Blake size = lseek(fd, 0, SEEK_END); 2181c1bb86cdSEric Blake if (size < 0) { 2182c1bb86cdSEric Blake return -errno; 2183c1bb86cdSEric Blake } 2184c1bb86cdSEric Blake } 2185c1bb86cdSEric Blake return size; 2186c1bb86cdSEric Blake } 2187c1bb86cdSEric Blake #else 2188c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2189c1bb86cdSEric Blake { 2190c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2191c1bb86cdSEric Blake int ret; 2192c1bb86cdSEric Blake int64_t size; 2193c1bb86cdSEric Blake 2194c1bb86cdSEric Blake ret = fd_open(bs); 2195c1bb86cdSEric Blake if (ret < 0) { 2196c1bb86cdSEric Blake return ret; 2197c1bb86cdSEric Blake } 2198c1bb86cdSEric Blake 2199c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2200c1bb86cdSEric Blake if (size < 0) { 2201c1bb86cdSEric Blake return -errno; 2202c1bb86cdSEric Blake } 2203c1bb86cdSEric Blake return size; 2204c1bb86cdSEric Blake } 2205c1bb86cdSEric Blake #endif 2206c1bb86cdSEric Blake 2207c1bb86cdSEric Blake static int64_t raw_get_allocated_file_size(BlockDriverState *bs) 2208c1bb86cdSEric Blake { 2209c1bb86cdSEric Blake struct stat st; 2210c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2211c1bb86cdSEric Blake 2212c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 2213c1bb86cdSEric Blake return -errno; 2214c1bb86cdSEric Blake } 2215c1bb86cdSEric Blake return (int64_t)st.st_blocks * 512; 2216c1bb86cdSEric Blake } 2217c1bb86cdSEric Blake 221893f4e2ffSKevin Wolf static int coroutine_fn 221993f4e2ffSKevin Wolf raw_co_create(BlockdevCreateOptions *options, Error **errp) 2220c1bb86cdSEric Blake { 2221927f11e1SKevin Wolf BlockdevCreateOptionsFile *file_opts; 22227c20c808SMax Reitz Error *local_err = NULL; 2223c1bb86cdSEric Blake int fd; 2224d815efcaSMax Reitz uint64_t perm, shared; 2225c1bb86cdSEric Blake int result = 0; 2226c1bb86cdSEric Blake 2227927f11e1SKevin Wolf /* Validate options and set default values */ 2228927f11e1SKevin Wolf assert(options->driver == BLOCKDEV_DRIVER_FILE); 2229927f11e1SKevin Wolf file_opts = &options->u.file; 2230c1bb86cdSEric Blake 2231927f11e1SKevin Wolf if (!file_opts->has_nocow) { 2232927f11e1SKevin Wolf file_opts->nocow = false; 2233927f11e1SKevin Wolf } 2234927f11e1SKevin Wolf if (!file_opts->has_preallocation) { 2235927f11e1SKevin Wolf file_opts->preallocation = PREALLOC_MODE_OFF; 2236c1bb86cdSEric Blake } 2237c1bb86cdSEric Blake 2238927f11e1SKevin Wolf /* Create file */ 2239b8cf1913SMax Reitz fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_BINARY, 0644); 2240c1bb86cdSEric Blake if (fd < 0) { 2241c1bb86cdSEric Blake result = -errno; 2242c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not create file"); 2243c1bb86cdSEric Blake goto out; 2244c1bb86cdSEric Blake } 2245c1bb86cdSEric Blake 2246b8cf1913SMax Reitz /* Take permissions: We want to discard everything, so we need 2247b8cf1913SMax Reitz * BLK_PERM_WRITE; and truncation to the desired size requires 2248b8cf1913SMax Reitz * BLK_PERM_RESIZE. 2249b8cf1913SMax Reitz * On the other hand, we cannot share the RESIZE permission 2250b8cf1913SMax Reitz * because we promise that after this function, the file has the 2251b8cf1913SMax Reitz * size given in the options. If someone else were to resize it 2252b8cf1913SMax Reitz * concurrently, we could not guarantee that. 2253b8cf1913SMax Reitz * Note that after this function, we can no longer guarantee that 2254b8cf1913SMax Reitz * the file is not touched by a third party, so it may be resized 2255b8cf1913SMax Reitz * then. */ 2256b8cf1913SMax Reitz perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; 2257b8cf1913SMax Reitz shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; 2258b8cf1913SMax Reitz 2259b8cf1913SMax Reitz /* Step one: Take locks */ 22602996ffadSFam Zheng result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); 2261b8cf1913SMax Reitz if (result < 0) { 2262b8cf1913SMax Reitz goto out_close; 2263b8cf1913SMax Reitz } 2264b8cf1913SMax Reitz 2265b8cf1913SMax Reitz /* Step two: Check that nobody else has taken conflicting locks */ 2266b8cf1913SMax Reitz result = raw_check_lock_bytes(fd, perm, shared, errp); 2267b8cf1913SMax Reitz if (result < 0) { 2268b857431dSFam Zheng error_append_hint(errp, 2269b857431dSFam Zheng "Is another process using the image [%s]?\n", 2270b857431dSFam Zheng file_opts->filename); 22717c20c808SMax Reitz goto out_unlock; 2272b8cf1913SMax Reitz } 2273b8cf1913SMax Reitz 2274b8cf1913SMax Reitz /* Clear the file by truncating it to 0 */ 227593f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); 2276b8cf1913SMax Reitz if (result < 0) { 22777c20c808SMax Reitz goto out_unlock; 2278b8cf1913SMax Reitz } 2279b8cf1913SMax Reitz 2280927f11e1SKevin Wolf if (file_opts->nocow) { 2281c1bb86cdSEric Blake #ifdef __linux__ 2282c1bb86cdSEric Blake /* Set NOCOW flag to solve performance issue on fs like btrfs. 2283c1bb86cdSEric Blake * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value 2284c1bb86cdSEric Blake * will be ignored since any failure of this operation should not 2285c1bb86cdSEric Blake * block the left work. 2286c1bb86cdSEric Blake */ 2287c1bb86cdSEric Blake int attr; 2288c1bb86cdSEric Blake if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { 2289c1bb86cdSEric Blake attr |= FS_NOCOW_FL; 2290c1bb86cdSEric Blake ioctl(fd, FS_IOC_SETFLAGS, &attr); 2291c1bb86cdSEric Blake } 2292c1bb86cdSEric Blake #endif 2293c1bb86cdSEric Blake } 2294c1bb86cdSEric Blake 2295b8cf1913SMax Reitz /* Resize and potentially preallocate the file to the desired 2296b8cf1913SMax Reitz * final size */ 229793f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, file_opts->size, 229893f4e2ffSKevin Wolf file_opts->preallocation, errp); 22999f63b07eSMax Reitz if (result < 0) { 23007c20c808SMax Reitz goto out_unlock; 23017c20c808SMax Reitz } 23027c20c808SMax Reitz 23037c20c808SMax Reitz out_unlock: 23042996ffadSFam Zheng raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); 23057c20c808SMax Reitz if (local_err) { 23067c20c808SMax Reitz /* The above call should not fail, and if it does, that does 23077c20c808SMax Reitz * not mean the whole creation operation has failed. So 23087c20c808SMax Reitz * report it the user for their convenience, but do not report 23097c20c808SMax Reitz * it to the caller. */ 2310db0754dfSFam Zheng warn_report_err(local_err); 23115a1dad9dSNir Soffer } 23125a1dad9dSNir Soffer 23135a1dad9dSNir Soffer out_close: 2314c1bb86cdSEric Blake if (qemu_close(fd) != 0 && result == 0) { 2315c1bb86cdSEric Blake result = -errno; 2316c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not close the new file"); 2317c1bb86cdSEric Blake } 2318c1bb86cdSEric Blake out: 2319c1bb86cdSEric Blake return result; 2320c1bb86cdSEric Blake } 2321c1bb86cdSEric Blake 2322927f11e1SKevin Wolf static int coroutine_fn raw_co_create_opts(const char *filename, QemuOpts *opts, 2323927f11e1SKevin Wolf Error **errp) 2324927f11e1SKevin Wolf { 2325927f11e1SKevin Wolf BlockdevCreateOptions options; 2326927f11e1SKevin Wolf int64_t total_size = 0; 2327927f11e1SKevin Wolf bool nocow = false; 2328927f11e1SKevin Wolf PreallocMode prealloc; 2329927f11e1SKevin Wolf char *buf = NULL; 2330927f11e1SKevin Wolf Error *local_err = NULL; 2331927f11e1SKevin Wolf 2332927f11e1SKevin Wolf /* Skip file: protocol prefix */ 2333927f11e1SKevin Wolf strstart(filename, "file:", &filename); 2334927f11e1SKevin Wolf 2335927f11e1SKevin Wolf /* Read out options */ 2336927f11e1SKevin Wolf total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2337927f11e1SKevin Wolf BDRV_SECTOR_SIZE); 2338927f11e1SKevin Wolf nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); 2339927f11e1SKevin Wolf buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2340927f11e1SKevin Wolf prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, 2341927f11e1SKevin Wolf PREALLOC_MODE_OFF, &local_err); 2342927f11e1SKevin Wolf g_free(buf); 2343927f11e1SKevin Wolf if (local_err) { 2344927f11e1SKevin Wolf error_propagate(errp, local_err); 2345927f11e1SKevin Wolf return -EINVAL; 2346927f11e1SKevin Wolf } 2347927f11e1SKevin Wolf 2348927f11e1SKevin Wolf options = (BlockdevCreateOptions) { 2349927f11e1SKevin Wolf .driver = BLOCKDEV_DRIVER_FILE, 2350927f11e1SKevin Wolf .u.file = { 2351927f11e1SKevin Wolf .filename = (char *) filename, 2352927f11e1SKevin Wolf .size = total_size, 2353927f11e1SKevin Wolf .has_preallocation = true, 2354927f11e1SKevin Wolf .preallocation = prealloc, 2355927f11e1SKevin Wolf .has_nocow = true, 2356927f11e1SKevin Wolf .nocow = nocow, 2357927f11e1SKevin Wolf }, 2358927f11e1SKevin Wolf }; 2359927f11e1SKevin Wolf return raw_co_create(&options, errp); 2360927f11e1SKevin Wolf } 2361927f11e1SKevin Wolf 2362c1bb86cdSEric Blake /* 2363c1bb86cdSEric Blake * Find allocation range in @bs around offset @start. 2364c1bb86cdSEric Blake * May change underlying file descriptor's file offset. 2365c1bb86cdSEric Blake * If @start is not in a hole, store @start in @data, and the 2366c1bb86cdSEric Blake * beginning of the next hole in @hole, and return 0. 2367c1bb86cdSEric Blake * If @start is in a non-trailing hole, store @start in @hole and the 2368c1bb86cdSEric Blake * beginning of the next non-hole in @data, and return 0. 2369c1bb86cdSEric Blake * If @start is in a trailing hole or beyond EOF, return -ENXIO. 2370c1bb86cdSEric Blake * If we can't find out, return a negative errno other than -ENXIO. 2371c1bb86cdSEric Blake */ 2372c1bb86cdSEric Blake static int find_allocation(BlockDriverState *bs, off_t start, 2373c1bb86cdSEric Blake off_t *data, off_t *hole) 2374c1bb86cdSEric Blake { 2375c1bb86cdSEric Blake #if defined SEEK_HOLE && defined SEEK_DATA 2376c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2377c1bb86cdSEric Blake off_t offs; 2378c1bb86cdSEric Blake 2379c1bb86cdSEric Blake /* 2380c1bb86cdSEric Blake * SEEK_DATA cases: 2381c1bb86cdSEric Blake * D1. offs == start: start is in data 2382c1bb86cdSEric Blake * D2. offs > start: start is in a hole, next data at offs 2383c1bb86cdSEric Blake * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 2384c1bb86cdSEric Blake * or start is beyond EOF 2385c1bb86cdSEric Blake * If the latter happens, the file has been truncated behind 2386c1bb86cdSEric Blake * our back since we opened it. All bets are off then. 2387c1bb86cdSEric Blake * Treating like a trailing hole is simplest. 2388c1bb86cdSEric Blake * D4. offs < 0, errno != ENXIO: we learned nothing 2389c1bb86cdSEric Blake */ 2390c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_DATA); 2391c1bb86cdSEric Blake if (offs < 0) { 2392c1bb86cdSEric Blake return -errno; /* D3 or D4 */ 2393c1bb86cdSEric Blake } 2394a03083a0SJeff Cody 2395a03083a0SJeff Cody if (offs < start) { 2396a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2397a03083a0SJeff Cody * -EIO in this case, and we'll treat it like D4. */ 2398a03083a0SJeff Cody return -EIO; 2399a03083a0SJeff Cody } 2400c1bb86cdSEric Blake 2401c1bb86cdSEric Blake if (offs > start) { 2402c1bb86cdSEric Blake /* D2: in hole, next data at offs */ 2403c1bb86cdSEric Blake *hole = start; 2404c1bb86cdSEric Blake *data = offs; 2405c1bb86cdSEric Blake return 0; 2406c1bb86cdSEric Blake } 2407c1bb86cdSEric Blake 2408c1bb86cdSEric Blake /* D1: in data, end not yet known */ 2409c1bb86cdSEric Blake 2410c1bb86cdSEric Blake /* 2411c1bb86cdSEric Blake * SEEK_HOLE cases: 2412c1bb86cdSEric Blake * H1. offs == start: start is in a hole 2413c1bb86cdSEric Blake * If this happens here, a hole has been dug behind our back 2414c1bb86cdSEric Blake * since the previous lseek(). 2415c1bb86cdSEric Blake * H2. offs > start: either start is in data, next hole at offs, 2416c1bb86cdSEric Blake * or start is in trailing hole, EOF at offs 2417c1bb86cdSEric Blake * Linux treats trailing holes like any other hole: offs == 2418c1bb86cdSEric Blake * start. Solaris seeks to EOF instead: offs > start (blech). 2419c1bb86cdSEric Blake * If that happens here, a hole has been dug behind our back 2420c1bb86cdSEric Blake * since the previous lseek(). 2421c1bb86cdSEric Blake * H3. offs < 0, errno = ENXIO: start is beyond EOF 2422c1bb86cdSEric Blake * If this happens, the file has been truncated behind our 2423c1bb86cdSEric Blake * back since we opened it. Treat it like a trailing hole. 2424c1bb86cdSEric Blake * H4. offs < 0, errno != ENXIO: we learned nothing 2425c1bb86cdSEric Blake * Pretend we know nothing at all, i.e. "forget" about D1. 2426c1bb86cdSEric Blake */ 2427c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_HOLE); 2428c1bb86cdSEric Blake if (offs < 0) { 2429c1bb86cdSEric Blake return -errno; /* D1 and (H3 or H4) */ 2430c1bb86cdSEric Blake } 2431a03083a0SJeff Cody 2432a03083a0SJeff Cody if (offs < start) { 2433a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2434a03083a0SJeff Cody * -EIO in this case, and we'll treat it like H4. */ 2435a03083a0SJeff Cody return -EIO; 2436a03083a0SJeff Cody } 2437c1bb86cdSEric Blake 2438c1bb86cdSEric Blake if (offs > start) { 2439c1bb86cdSEric Blake /* 2440c1bb86cdSEric Blake * D1 and H2: either in data, next hole at offs, or it was in 2441c1bb86cdSEric Blake * data but is now in a trailing hole. In the latter case, 2442c1bb86cdSEric Blake * all bets are off. Treating it as if it there was data all 2443c1bb86cdSEric Blake * the way to EOF is safe, so simply do that. 2444c1bb86cdSEric Blake */ 2445c1bb86cdSEric Blake *data = start; 2446c1bb86cdSEric Blake *hole = offs; 2447c1bb86cdSEric Blake return 0; 2448c1bb86cdSEric Blake } 2449c1bb86cdSEric Blake 2450c1bb86cdSEric Blake /* D1 and H1 */ 2451c1bb86cdSEric Blake return -EBUSY; 2452c1bb86cdSEric Blake #else 2453c1bb86cdSEric Blake return -ENOTSUP; 2454c1bb86cdSEric Blake #endif 2455c1bb86cdSEric Blake } 2456c1bb86cdSEric Blake 2457c1bb86cdSEric Blake /* 2458a290f085SEric Blake * Returns the allocation status of the specified offset. 2459c1bb86cdSEric Blake * 2460a290f085SEric Blake * The block layer guarantees 'offset' and 'bytes' are within bounds. 2461c1bb86cdSEric Blake * 2462a290f085SEric Blake * 'pnum' is set to the number of bytes (including and immediately following 2463a290f085SEric Blake * the specified offset) that are known to be in the same 2464c1bb86cdSEric Blake * allocated/unallocated state. 2465c1bb86cdSEric Blake * 2466a290f085SEric Blake * 'bytes' is the max value 'pnum' should be set to. 2467c1bb86cdSEric Blake */ 2468a290f085SEric Blake static int coroutine_fn raw_co_block_status(BlockDriverState *bs, 2469a290f085SEric Blake bool want_zero, 2470a290f085SEric Blake int64_t offset, 2471a290f085SEric Blake int64_t bytes, int64_t *pnum, 2472a290f085SEric Blake int64_t *map, 2473c1bb86cdSEric Blake BlockDriverState **file) 2474c1bb86cdSEric Blake { 2475a290f085SEric Blake off_t data = 0, hole = 0; 2476c1bb86cdSEric Blake int ret; 2477c1bb86cdSEric Blake 2478c1bb86cdSEric Blake ret = fd_open(bs); 2479c1bb86cdSEric Blake if (ret < 0) { 2480c1bb86cdSEric Blake return ret; 2481c1bb86cdSEric Blake } 2482c1bb86cdSEric Blake 2483a290f085SEric Blake if (!want_zero) { 2484a290f085SEric Blake *pnum = bytes; 2485a290f085SEric Blake *map = offset; 2486a290f085SEric Blake *file = bs; 2487a290f085SEric Blake return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 2488c1bb86cdSEric Blake } 2489c1bb86cdSEric Blake 2490a290f085SEric Blake ret = find_allocation(bs, offset, &data, &hole); 2491c1bb86cdSEric Blake if (ret == -ENXIO) { 2492c1bb86cdSEric Blake /* Trailing hole */ 2493a290f085SEric Blake *pnum = bytes; 2494c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2495c1bb86cdSEric Blake } else if (ret < 0) { 2496c1bb86cdSEric Blake /* No info available, so pretend there are no holes */ 2497a290f085SEric Blake *pnum = bytes; 2498c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2499a290f085SEric Blake } else if (data == offset) { 2500a290f085SEric Blake /* On a data extent, compute bytes to the end of the extent, 2501c1bb86cdSEric Blake * possibly including a partial sector at EOF. */ 2502a290f085SEric Blake *pnum = MIN(bytes, hole - offset); 2503c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2504c1bb86cdSEric Blake } else { 2505a290f085SEric Blake /* On a hole, compute bytes to the beginning of the next extent. */ 2506a290f085SEric Blake assert(hole == offset); 2507a290f085SEric Blake *pnum = MIN(bytes, data - offset); 2508c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2509c1bb86cdSEric Blake } 2510a290f085SEric Blake *map = offset; 2511c1bb86cdSEric Blake *file = bs; 2512a290f085SEric Blake return ret | BDRV_BLOCK_OFFSET_VALID; 2513c1bb86cdSEric Blake } 2514c1bb86cdSEric Blake 251531be8a2aSStefan Hajnoczi #if defined(__linux__) 251631be8a2aSStefan Hajnoczi /* Verify that the file is not in the page cache */ 251731be8a2aSStefan Hajnoczi static void check_cache_dropped(BlockDriverState *bs, Error **errp) 251831be8a2aSStefan Hajnoczi { 251931be8a2aSStefan Hajnoczi const size_t window_size = 128 * 1024 * 1024; 252031be8a2aSStefan Hajnoczi BDRVRawState *s = bs->opaque; 252131be8a2aSStefan Hajnoczi void *window = NULL; 252231be8a2aSStefan Hajnoczi size_t length = 0; 252331be8a2aSStefan Hajnoczi unsigned char *vec; 252431be8a2aSStefan Hajnoczi size_t page_size; 252531be8a2aSStefan Hajnoczi off_t offset; 252631be8a2aSStefan Hajnoczi off_t end; 252731be8a2aSStefan Hajnoczi 252831be8a2aSStefan Hajnoczi /* mincore(2) page status information requires 1 byte per page */ 252931be8a2aSStefan Hajnoczi page_size = sysconf(_SC_PAGESIZE); 253031be8a2aSStefan Hajnoczi vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); 253131be8a2aSStefan Hajnoczi 253231be8a2aSStefan Hajnoczi end = raw_getlength(bs); 253331be8a2aSStefan Hajnoczi 253431be8a2aSStefan Hajnoczi for (offset = 0; offset < end; offset += window_size) { 253531be8a2aSStefan Hajnoczi void *new_window; 253631be8a2aSStefan Hajnoczi size_t new_length; 253731be8a2aSStefan Hajnoczi size_t vec_end; 253831be8a2aSStefan Hajnoczi size_t i; 253931be8a2aSStefan Hajnoczi int ret; 254031be8a2aSStefan Hajnoczi 254131be8a2aSStefan Hajnoczi /* Unmap previous window if size has changed */ 254231be8a2aSStefan Hajnoczi new_length = MIN(end - offset, window_size); 254331be8a2aSStefan Hajnoczi if (new_length != length) { 254431be8a2aSStefan Hajnoczi munmap(window, length); 254531be8a2aSStefan Hajnoczi window = NULL; 254631be8a2aSStefan Hajnoczi length = 0; 254731be8a2aSStefan Hajnoczi } 254831be8a2aSStefan Hajnoczi 254931be8a2aSStefan Hajnoczi new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, 255031be8a2aSStefan Hajnoczi s->fd, offset); 255131be8a2aSStefan Hajnoczi if (new_window == MAP_FAILED) { 255231be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mmap failed"); 255331be8a2aSStefan Hajnoczi break; 255431be8a2aSStefan Hajnoczi } 255531be8a2aSStefan Hajnoczi 255631be8a2aSStefan Hajnoczi window = new_window; 255731be8a2aSStefan Hajnoczi length = new_length; 255831be8a2aSStefan Hajnoczi 255931be8a2aSStefan Hajnoczi ret = mincore(window, length, vec); 256031be8a2aSStefan Hajnoczi if (ret < 0) { 256131be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mincore failed"); 256231be8a2aSStefan Hajnoczi break; 256331be8a2aSStefan Hajnoczi } 256431be8a2aSStefan Hajnoczi 256531be8a2aSStefan Hajnoczi vec_end = DIV_ROUND_UP(length, page_size); 256631be8a2aSStefan Hajnoczi for (i = 0; i < vec_end; i++) { 256731be8a2aSStefan Hajnoczi if (vec[i] & 0x1) { 256831be8a2aSStefan Hajnoczi error_setg(errp, "page cache still in use!"); 256931be8a2aSStefan Hajnoczi break; 257031be8a2aSStefan Hajnoczi } 257131be8a2aSStefan Hajnoczi } 257231be8a2aSStefan Hajnoczi } 257331be8a2aSStefan Hajnoczi 257431be8a2aSStefan Hajnoczi if (window) { 257531be8a2aSStefan Hajnoczi munmap(window, length); 257631be8a2aSStefan Hajnoczi } 257731be8a2aSStefan Hajnoczi 257831be8a2aSStefan Hajnoczi g_free(vec); 257931be8a2aSStefan Hajnoczi } 258031be8a2aSStefan Hajnoczi #endif /* __linux__ */ 258131be8a2aSStefan Hajnoczi 2582dd577a26SStefan Hajnoczi static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, 2583dd577a26SStefan Hajnoczi Error **errp) 2584dd577a26SStefan Hajnoczi { 2585dd577a26SStefan Hajnoczi BDRVRawState *s = bs->opaque; 2586dd577a26SStefan Hajnoczi int ret; 2587dd577a26SStefan Hajnoczi 2588dd577a26SStefan Hajnoczi ret = fd_open(bs); 2589dd577a26SStefan Hajnoczi if (ret < 0) { 2590dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "The file descriptor is not open"); 2591dd577a26SStefan Hajnoczi return; 2592dd577a26SStefan Hajnoczi } 2593dd577a26SStefan Hajnoczi 2594f357fcd8SStefan Hajnoczi if (!s->drop_cache) { 2595f357fcd8SStefan Hajnoczi return; 2596f357fcd8SStefan Hajnoczi } 2597f357fcd8SStefan Hajnoczi 2598dd577a26SStefan Hajnoczi if (s->open_flags & O_DIRECT) { 2599dd577a26SStefan Hajnoczi return; /* No host kernel page cache */ 2600dd577a26SStefan Hajnoczi } 2601dd577a26SStefan Hajnoczi 2602dd577a26SStefan Hajnoczi #if defined(__linux__) 2603dd577a26SStefan Hajnoczi /* This sets the scene for the next syscall... */ 2604dd577a26SStefan Hajnoczi ret = bdrv_co_flush(bs); 2605dd577a26SStefan Hajnoczi if (ret < 0) { 2606dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "flush failed"); 2607dd577a26SStefan Hajnoczi return; 2608dd577a26SStefan Hajnoczi } 2609dd577a26SStefan Hajnoczi 2610dd577a26SStefan Hajnoczi /* Linux does not invalidate pages that are dirty, locked, or mmapped by a 2611dd577a26SStefan Hajnoczi * process. These limitations are okay because we just fsynced the file, 2612dd577a26SStefan Hajnoczi * we don't use mmap, and the file should not be in use by other processes. 2613dd577a26SStefan Hajnoczi */ 2614dd577a26SStefan Hajnoczi ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); 2615dd577a26SStefan Hajnoczi if (ret != 0) { /* the return value is a positive errno */ 2616dd577a26SStefan Hajnoczi error_setg_errno(errp, ret, "fadvise failed"); 2617dd577a26SStefan Hajnoczi return; 2618dd577a26SStefan Hajnoczi } 261931be8a2aSStefan Hajnoczi 262031be8a2aSStefan Hajnoczi if (s->check_cache_dropped) { 262131be8a2aSStefan Hajnoczi check_cache_dropped(bs, errp); 262231be8a2aSStefan Hajnoczi } 2623dd577a26SStefan Hajnoczi #else /* __linux__ */ 2624dd577a26SStefan Hajnoczi /* Do nothing. Live migration to a remote host with cache.direct=off is 2625dd577a26SStefan Hajnoczi * unsupported on other host operating systems. Cache consistency issues 2626dd577a26SStefan Hajnoczi * may occur but no error is reported here, partly because that's the 2627dd577a26SStefan Hajnoczi * historical behavior and partly because it's hard to differentiate valid 2628dd577a26SStefan Hajnoczi * configurations that should not cause errors. 2629dd577a26SStefan Hajnoczi */ 2630dd577a26SStefan Hajnoczi #endif /* !__linux__ */ 2631dd577a26SStefan Hajnoczi } 2632dd577a26SStefan Hajnoczi 263333d70fb6SKevin Wolf static coroutine_fn int 263446ee0f46SKevin Wolf raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev) 2635c1bb86cdSEric Blake { 2636c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 263746ee0f46SKevin Wolf RawPosixAIOData acb; 2638c1bb86cdSEric Blake 263946ee0f46SKevin Wolf acb = (RawPosixAIOData) { 264046ee0f46SKevin Wolf .bs = bs, 264146ee0f46SKevin Wolf .aio_fildes = s->fd, 264246ee0f46SKevin Wolf .aio_type = QEMU_AIO_DISCARD, 264346ee0f46SKevin Wolf .aio_offset = offset, 264446ee0f46SKevin Wolf .aio_nbytes = bytes, 264546ee0f46SKevin Wolf }; 264646ee0f46SKevin Wolf 264746ee0f46SKevin Wolf if (blkdev) { 264846ee0f46SKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 264946ee0f46SKevin Wolf } 265046ee0f46SKevin Wolf 265146ee0f46SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_discard, &acb); 265246ee0f46SKevin Wolf } 265346ee0f46SKevin Wolf 265446ee0f46SKevin Wolf static coroutine_fn int 265546ee0f46SKevin Wolf raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 265646ee0f46SKevin Wolf { 265746ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, false); 2658c1bb86cdSEric Blake } 2659c1bb86cdSEric Blake 26607154d8aeSKevin Wolf static int coroutine_fn 26617154d8aeSKevin Wolf raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, 26627154d8aeSKevin Wolf BdrvRequestFlags flags, bool blkdev) 26637154d8aeSKevin Wolf { 26647154d8aeSKevin Wolf BDRVRawState *s = bs->opaque; 26657154d8aeSKevin Wolf RawPosixAIOData acb; 26667154d8aeSKevin Wolf ThreadPoolFunc *handler; 26677154d8aeSKevin Wolf 26687154d8aeSKevin Wolf acb = (RawPosixAIOData) { 26697154d8aeSKevin Wolf .bs = bs, 26707154d8aeSKevin Wolf .aio_fildes = s->fd, 26717154d8aeSKevin Wolf .aio_type = QEMU_AIO_WRITE_ZEROES, 26727154d8aeSKevin Wolf .aio_offset = offset, 26737154d8aeSKevin Wolf .aio_nbytes = bytes, 26747154d8aeSKevin Wolf }; 26757154d8aeSKevin Wolf 26767154d8aeSKevin Wolf if (blkdev) { 26777154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 26787154d8aeSKevin Wolf } 2679738301e1SKevin Wolf if (flags & BDRV_REQ_NO_FALLBACK) { 2680738301e1SKevin Wolf acb.aio_type |= QEMU_AIO_NO_FALLBACK; 2681738301e1SKevin Wolf } 26827154d8aeSKevin Wolf 26837154d8aeSKevin Wolf if (flags & BDRV_REQ_MAY_UNMAP) { 26847154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_DISCARD; 26857154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes_unmap; 26867154d8aeSKevin Wolf } else { 26877154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes; 26887154d8aeSKevin Wolf } 26897154d8aeSKevin Wolf 26907154d8aeSKevin Wolf return raw_thread_pool_submit(bs, handler, &acb); 26917154d8aeSKevin Wolf } 26927154d8aeSKevin Wolf 2693c1bb86cdSEric Blake static int coroutine_fn raw_co_pwrite_zeroes( 2694c1bb86cdSEric Blake BlockDriverState *bs, int64_t offset, 2695f5a5ca79SManos Pitsidianakis int bytes, BdrvRequestFlags flags) 2696c1bb86cdSEric Blake { 26977154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); 2698c1bb86cdSEric Blake } 2699c1bb86cdSEric Blake 2700c1bb86cdSEric Blake static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2701c1bb86cdSEric Blake { 2702c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2703c1bb86cdSEric Blake 2704c1bb86cdSEric Blake bdi->unallocated_blocks_are_zero = s->discard_zeroes; 2705c1bb86cdSEric Blake return 0; 2706c1bb86cdSEric Blake } 2707c1bb86cdSEric Blake 2708c1bb86cdSEric Blake static QemuOptsList raw_create_opts = { 2709c1bb86cdSEric Blake .name = "raw-create-opts", 2710c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), 2711c1bb86cdSEric Blake .desc = { 2712c1bb86cdSEric Blake { 2713c1bb86cdSEric Blake .name = BLOCK_OPT_SIZE, 2714c1bb86cdSEric Blake .type = QEMU_OPT_SIZE, 2715c1bb86cdSEric Blake .help = "Virtual disk size" 2716c1bb86cdSEric Blake }, 2717c1bb86cdSEric Blake { 2718c1bb86cdSEric Blake .name = BLOCK_OPT_NOCOW, 2719c1bb86cdSEric Blake .type = QEMU_OPT_BOOL, 2720c1bb86cdSEric Blake .help = "Turn off copy-on-write (valid only on btrfs)" 2721c1bb86cdSEric Blake }, 2722c1bb86cdSEric Blake { 2723c1bb86cdSEric Blake .name = BLOCK_OPT_PREALLOC, 2724c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 2725c1bb86cdSEric Blake .help = "Preallocation mode (allowed values: off, falloc, full)" 2726c1bb86cdSEric Blake }, 2727c1bb86cdSEric Blake { /* end of list */ } 2728c1bb86cdSEric Blake } 2729c1bb86cdSEric Blake }; 2730c1bb86cdSEric Blake 2731244a5668SFam Zheng static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, 2732244a5668SFam Zheng Error **errp) 2733244a5668SFam Zheng { 27346ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 27356ceabe6fSKevin Wolf BDRVRawReopenState *rs = NULL; 27366ceabe6fSKevin Wolf int open_flags; 27376ceabe6fSKevin Wolf int ret; 27386ceabe6fSKevin Wolf 27396ceabe6fSKevin Wolf if (s->perm_change_fd) { 27406ceabe6fSKevin Wolf /* 27416ceabe6fSKevin Wolf * In the context of reopen, this function may be called several times 27426ceabe6fSKevin Wolf * (directly and recursively while change permissions of the parent). 27436ceabe6fSKevin Wolf * This is even true for children that don't inherit from the original 27446ceabe6fSKevin Wolf * reopen node, so s->reopen_state is not set. 27456ceabe6fSKevin Wolf * 27466ceabe6fSKevin Wolf * Ignore all but the first call. 27476ceabe6fSKevin Wolf */ 27486ceabe6fSKevin Wolf return 0; 27496ceabe6fSKevin Wolf } 27506ceabe6fSKevin Wolf 27516ceabe6fSKevin Wolf if (s->reopen_state) { 27526ceabe6fSKevin Wolf /* We already have a new file descriptor to set permissions for */ 27536ceabe6fSKevin Wolf assert(s->reopen_state->perm == perm); 27546ceabe6fSKevin Wolf assert(s->reopen_state->shared_perm == shared); 27556ceabe6fSKevin Wolf rs = s->reopen_state->opaque; 27566ceabe6fSKevin Wolf s->perm_change_fd = rs->fd; 27576ceabe6fSKevin Wolf } else { 27586ceabe6fSKevin Wolf /* We may need a new fd if auto-read-only switches the mode */ 275923dece19SKevin Wolf ret = raw_reconfigure_getfd(bs, bs->open_flags, &open_flags, perm, 27606ceabe6fSKevin Wolf false, errp); 27616ceabe6fSKevin Wolf if (ret < 0) { 27626ceabe6fSKevin Wolf return ret; 27636ceabe6fSKevin Wolf } else if (ret != s->fd) { 27646ceabe6fSKevin Wolf s->perm_change_fd = ret; 27656ceabe6fSKevin Wolf } 27666ceabe6fSKevin Wolf } 27676ceabe6fSKevin Wolf 27686ceabe6fSKevin Wolf /* Prepare permissions on old fd to avoid conflicts between old and new, 27696ceabe6fSKevin Wolf * but keep everything locked that new will need. */ 27706ceabe6fSKevin Wolf ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); 27716ceabe6fSKevin Wolf if (ret < 0) { 27726ceabe6fSKevin Wolf goto fail; 27736ceabe6fSKevin Wolf } 27746ceabe6fSKevin Wolf 27756ceabe6fSKevin Wolf /* Copy locks to the new fd */ 27766ceabe6fSKevin Wolf if (s->perm_change_fd) { 27776ceabe6fSKevin Wolf ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, 27786ceabe6fSKevin Wolf false, errp); 27796ceabe6fSKevin Wolf if (ret < 0) { 27806ceabe6fSKevin Wolf raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 27816ceabe6fSKevin Wolf goto fail; 27826ceabe6fSKevin Wolf } 27836ceabe6fSKevin Wolf } 27846ceabe6fSKevin Wolf return 0; 27856ceabe6fSKevin Wolf 27866ceabe6fSKevin Wolf fail: 27876ceabe6fSKevin Wolf if (s->perm_change_fd && !s->reopen_state) { 27886ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 27896ceabe6fSKevin Wolf } 27906ceabe6fSKevin Wolf s->perm_change_fd = 0; 27916ceabe6fSKevin Wolf return ret; 2792244a5668SFam Zheng } 2793244a5668SFam Zheng 2794244a5668SFam Zheng static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) 2795244a5668SFam Zheng { 2796244a5668SFam Zheng BDRVRawState *s = bs->opaque; 27976ceabe6fSKevin Wolf 27986ceabe6fSKevin Wolf /* For reopen, we have already switched to the new fd (.bdrv_set_perm is 27996ceabe6fSKevin Wolf * called after .bdrv_reopen_commit) */ 28006ceabe6fSKevin Wolf if (s->perm_change_fd && s->fd != s->perm_change_fd) { 28016ceabe6fSKevin Wolf qemu_close(s->fd); 28026ceabe6fSKevin Wolf s->fd = s->perm_change_fd; 28036ceabe6fSKevin Wolf } 28046ceabe6fSKevin Wolf s->perm_change_fd = 0; 28056ceabe6fSKevin Wolf 2806244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); 2807244a5668SFam Zheng s->perm = perm; 2808244a5668SFam Zheng s->shared_perm = shared; 2809244a5668SFam Zheng } 2810244a5668SFam Zheng 2811244a5668SFam Zheng static void raw_abort_perm_update(BlockDriverState *bs) 2812244a5668SFam Zheng { 28136ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 28146ceabe6fSKevin Wolf 28156ceabe6fSKevin Wolf /* For reopen, .bdrv_reopen_abort is called afterwards and will close 28166ceabe6fSKevin Wolf * the file descriptor. */ 28176ceabe6fSKevin Wolf if (s->perm_change_fd && !s->reopen_state) { 28186ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 28196ceabe6fSKevin Wolf } 28206ceabe6fSKevin Wolf s->perm_change_fd = 0; 28216ceabe6fSKevin Wolf 2822244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 2823244a5668SFam Zheng } 2824244a5668SFam Zheng 282567b51fb9SVladimir Sementsov-Ogievskiy static int coroutine_fn raw_co_copy_range_from( 282667b51fb9SVladimir Sementsov-Ogievskiy BlockDriverState *bs, BdrvChild *src, uint64_t src_offset, 282767b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, uint64_t dst_offset, uint64_t bytes, 282867b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) 28291efad060SFam Zheng { 283067b51fb9SVladimir Sementsov-Ogievskiy return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, 283167b51fb9SVladimir Sementsov-Ogievskiy read_flags, write_flags); 28321efad060SFam Zheng } 28331efad060SFam Zheng 28341efad060SFam Zheng static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, 283567b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *src, 283667b51fb9SVladimir Sementsov-Ogievskiy uint64_t src_offset, 283767b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, 283867b51fb9SVladimir Sementsov-Ogievskiy uint64_t dst_offset, 283967b51fb9SVladimir Sementsov-Ogievskiy uint64_t bytes, 284067b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, 284167b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags write_flags) 28421efad060SFam Zheng { 284358a209c4SKevin Wolf RawPosixAIOData acb; 28441efad060SFam Zheng BDRVRawState *s = bs->opaque; 28451efad060SFam Zheng BDRVRawState *src_s; 28461efad060SFam Zheng 28471efad060SFam Zheng assert(dst->bs == bs); 28481efad060SFam Zheng if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { 28491efad060SFam Zheng return -ENOTSUP; 28501efad060SFam Zheng } 28511efad060SFam Zheng 28521efad060SFam Zheng src_s = src->bs->opaque; 28539f850f67SFam Zheng if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { 28541efad060SFam Zheng return -EIO; 28551efad060SFam Zheng } 285658a209c4SKevin Wolf 285758a209c4SKevin Wolf acb = (RawPosixAIOData) { 285858a209c4SKevin Wolf .bs = bs, 285958a209c4SKevin Wolf .aio_type = QEMU_AIO_COPY_RANGE, 286058a209c4SKevin Wolf .aio_fildes = src_s->fd, 286158a209c4SKevin Wolf .aio_offset = src_offset, 286258a209c4SKevin Wolf .aio_nbytes = bytes, 286358a209c4SKevin Wolf .copy_range = { 286458a209c4SKevin Wolf .aio_fd2 = s->fd, 286558a209c4SKevin Wolf .aio_offset2 = dst_offset, 286658a209c4SKevin Wolf }, 286758a209c4SKevin Wolf }; 286858a209c4SKevin Wolf 286958a209c4SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); 28701efad060SFam Zheng } 28711efad060SFam Zheng 2872c1bb86cdSEric Blake BlockDriver bdrv_file = { 2873c1bb86cdSEric Blake .format_name = "file", 2874c1bb86cdSEric Blake .protocol_name = "file", 2875c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 2876c1bb86cdSEric Blake .bdrv_needs_filename = true, 2877c1bb86cdSEric Blake .bdrv_probe = NULL, /* no probe for protocols */ 2878c1bb86cdSEric Blake .bdrv_parse_filename = raw_parse_filename, 2879c1bb86cdSEric Blake .bdrv_file_open = raw_open, 2880c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 2881c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 2882c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 2883c1bb86cdSEric Blake .bdrv_close = raw_close, 2884927f11e1SKevin Wolf .bdrv_co_create = raw_co_create, 2885efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = raw_co_create_opts, 2886c1bb86cdSEric Blake .bdrv_has_zero_init = bdrv_has_zero_init_1, 2887a290f085SEric Blake .bdrv_co_block_status = raw_co_block_status, 2888dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 2889c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, 2890c1bb86cdSEric Blake 2891c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 2892c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 289333d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 289433d70fb6SKevin Wolf .bdrv_co_pdiscard = raw_co_pdiscard, 28951efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 28961efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 2897c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 2898c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 2899c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 2900ed6e2161SNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 2901c1bb86cdSEric Blake 2902061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 2903c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 2904c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 2905c1bb86cdSEric Blake .bdrv_get_allocated_file_size 2906c1bb86cdSEric Blake = raw_get_allocated_file_size, 2907244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 2908244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 2909244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 2910c1bb86cdSEric Blake .create_opts = &raw_create_opts, 29118a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 2912c1bb86cdSEric Blake }; 2913c1bb86cdSEric Blake 2914c1bb86cdSEric Blake /***********************************************/ 2915c1bb86cdSEric Blake /* host device */ 2916c1bb86cdSEric Blake 2917c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2918c1bb86cdSEric Blake static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 2919c1bb86cdSEric Blake CFIndex maxPathSize, int flags); 2920c1bb86cdSEric Blake static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) 2921c1bb86cdSEric Blake { 2922c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 2923c1bb86cdSEric Blake mach_port_t masterPort; 2924c1bb86cdSEric Blake CFMutableDictionaryRef classesToMatch; 2925c1bb86cdSEric Blake const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; 2926c1bb86cdSEric Blake char *mediaType = NULL; 2927c1bb86cdSEric Blake 2928c1bb86cdSEric Blake kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); 2929c1bb86cdSEric Blake if ( KERN_SUCCESS != kernResult ) { 2930c1bb86cdSEric Blake printf( "IOMasterPort returned %d\n", kernResult ); 2931c1bb86cdSEric Blake } 2932c1bb86cdSEric Blake 2933c1bb86cdSEric Blake int index; 2934c1bb86cdSEric Blake for (index = 0; index < ARRAY_SIZE(matching_array); index++) { 2935c1bb86cdSEric Blake classesToMatch = IOServiceMatching(matching_array[index]); 2936c1bb86cdSEric Blake if (classesToMatch == NULL) { 2937c1bb86cdSEric Blake error_report("IOServiceMatching returned NULL for %s", 2938c1bb86cdSEric Blake matching_array[index]); 2939c1bb86cdSEric Blake continue; 2940c1bb86cdSEric Blake } 2941c1bb86cdSEric Blake CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), 2942c1bb86cdSEric Blake kCFBooleanTrue); 2943c1bb86cdSEric Blake kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, 2944c1bb86cdSEric Blake mediaIterator); 2945c1bb86cdSEric Blake if (kernResult != KERN_SUCCESS) { 2946c1bb86cdSEric Blake error_report("Note: IOServiceGetMatchingServices returned %d", 2947c1bb86cdSEric Blake kernResult); 2948c1bb86cdSEric Blake continue; 2949c1bb86cdSEric Blake } 2950c1bb86cdSEric Blake 2951c1bb86cdSEric Blake /* If a match was found, leave the loop */ 2952c1bb86cdSEric Blake if (*mediaIterator != 0) { 29534f7d28d7SLaurent Vivier trace_file_FindEjectableOpticalMedia(matching_array[index]); 2954c1bb86cdSEric Blake mediaType = g_strdup(matching_array[index]); 2955c1bb86cdSEric Blake break; 2956c1bb86cdSEric Blake } 2957c1bb86cdSEric Blake } 2958c1bb86cdSEric Blake return mediaType; 2959c1bb86cdSEric Blake } 2960c1bb86cdSEric Blake 2961c1bb86cdSEric Blake kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 2962c1bb86cdSEric Blake CFIndex maxPathSize, int flags) 2963c1bb86cdSEric Blake { 2964c1bb86cdSEric Blake io_object_t nextMedia; 2965c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 2966c1bb86cdSEric Blake *bsdPath = '\0'; 2967c1bb86cdSEric Blake nextMedia = IOIteratorNext( mediaIterator ); 2968c1bb86cdSEric Blake if ( nextMedia ) 2969c1bb86cdSEric Blake { 2970c1bb86cdSEric Blake CFTypeRef bsdPathAsCFString; 2971c1bb86cdSEric Blake bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); 2972c1bb86cdSEric Blake if ( bsdPathAsCFString ) { 2973c1bb86cdSEric Blake size_t devPathLength; 2974c1bb86cdSEric Blake strcpy( bsdPath, _PATH_DEV ); 2975c1bb86cdSEric Blake if (flags & BDRV_O_NOCACHE) { 2976c1bb86cdSEric Blake strcat(bsdPath, "r"); 2977c1bb86cdSEric Blake } 2978c1bb86cdSEric Blake devPathLength = strlen( bsdPath ); 2979c1bb86cdSEric Blake if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { 2980c1bb86cdSEric Blake kernResult = KERN_SUCCESS; 2981c1bb86cdSEric Blake } 2982c1bb86cdSEric Blake CFRelease( bsdPathAsCFString ); 2983c1bb86cdSEric Blake } 2984c1bb86cdSEric Blake IOObjectRelease( nextMedia ); 2985c1bb86cdSEric Blake } 2986c1bb86cdSEric Blake 2987c1bb86cdSEric Blake return kernResult; 2988c1bb86cdSEric Blake } 2989c1bb86cdSEric Blake 2990c1bb86cdSEric Blake /* Sets up a real cdrom for use in QEMU */ 2991c1bb86cdSEric Blake static bool setup_cdrom(char *bsd_path, Error **errp) 2992c1bb86cdSEric Blake { 2993c1bb86cdSEric Blake int index, num_of_test_partitions = 2, fd; 2994c1bb86cdSEric Blake char test_partition[MAXPATHLEN]; 2995c1bb86cdSEric Blake bool partition_found = false; 2996c1bb86cdSEric Blake 2997c1bb86cdSEric Blake /* look for a working partition */ 2998c1bb86cdSEric Blake for (index = 0; index < num_of_test_partitions; index++) { 2999c1bb86cdSEric Blake snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, 3000c1bb86cdSEric Blake index); 3001c1bb86cdSEric Blake fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE); 3002c1bb86cdSEric Blake if (fd >= 0) { 3003c1bb86cdSEric Blake partition_found = true; 3004c1bb86cdSEric Blake qemu_close(fd); 3005c1bb86cdSEric Blake break; 3006c1bb86cdSEric Blake } 3007c1bb86cdSEric Blake } 3008c1bb86cdSEric Blake 3009c1bb86cdSEric Blake /* if a working partition on the device was not found */ 3010c1bb86cdSEric Blake if (partition_found == false) { 3011c1bb86cdSEric Blake error_setg(errp, "Failed to find a working partition on disc"); 3012c1bb86cdSEric Blake } else { 30134f7d28d7SLaurent Vivier trace_file_setup_cdrom(test_partition); 3014c1bb86cdSEric Blake pstrcpy(bsd_path, MAXPATHLEN, test_partition); 3015c1bb86cdSEric Blake } 3016c1bb86cdSEric Blake return partition_found; 3017c1bb86cdSEric Blake } 3018c1bb86cdSEric Blake 3019c1bb86cdSEric Blake /* Prints directions on mounting and unmounting a device */ 3020c1bb86cdSEric Blake static void print_unmounting_directions(const char *file_name) 3021c1bb86cdSEric Blake { 3022c1bb86cdSEric Blake error_report("If device %s is mounted on the desktop, unmount" 3023c1bb86cdSEric Blake " it first before using it in QEMU", file_name); 3024c1bb86cdSEric Blake error_report("Command to unmount device: diskutil unmountDisk %s", 3025c1bb86cdSEric Blake file_name); 3026c1bb86cdSEric Blake error_report("Command to mount device: diskutil mountDisk %s", file_name); 3027c1bb86cdSEric Blake } 3028c1bb86cdSEric Blake 3029c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3030c1bb86cdSEric Blake 3031c1bb86cdSEric Blake static int hdev_probe_device(const char *filename) 3032c1bb86cdSEric Blake { 3033c1bb86cdSEric Blake struct stat st; 3034c1bb86cdSEric Blake 3035c1bb86cdSEric Blake /* allow a dedicated CD-ROM driver to match with a higher priority */ 3036c1bb86cdSEric Blake if (strstart(filename, "/dev/cdrom", NULL)) 3037c1bb86cdSEric Blake return 50; 3038c1bb86cdSEric Blake 3039c1bb86cdSEric Blake if (stat(filename, &st) >= 0 && 3040c1bb86cdSEric Blake (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 3041c1bb86cdSEric Blake return 100; 3042c1bb86cdSEric Blake } 3043c1bb86cdSEric Blake 3044c1bb86cdSEric Blake return 0; 3045c1bb86cdSEric Blake } 3046c1bb86cdSEric Blake 3047c1bb86cdSEric Blake static int check_hdev_writable(BDRVRawState *s) 3048c1bb86cdSEric Blake { 3049c1bb86cdSEric Blake #if defined(BLKROGET) 3050c1bb86cdSEric Blake /* Linux block devices can be configured "read-only" using blockdev(8). 3051c1bb86cdSEric Blake * This is independent of device node permissions and therefore open(2) 3052c1bb86cdSEric Blake * with O_RDWR succeeds. Actual writes fail with EPERM. 3053c1bb86cdSEric Blake * 3054c1bb86cdSEric Blake * bdrv_open() is supposed to fail if the disk is read-only. Explicitly 3055c1bb86cdSEric Blake * check for read-only block devices so that Linux block devices behave 3056c1bb86cdSEric Blake * properly. 3057c1bb86cdSEric Blake */ 3058c1bb86cdSEric Blake struct stat st; 3059c1bb86cdSEric Blake int readonly = 0; 3060c1bb86cdSEric Blake 3061c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 3062c1bb86cdSEric Blake return -errno; 3063c1bb86cdSEric Blake } 3064c1bb86cdSEric Blake 3065c1bb86cdSEric Blake if (!S_ISBLK(st.st_mode)) { 3066c1bb86cdSEric Blake return 0; 3067c1bb86cdSEric Blake } 3068c1bb86cdSEric Blake 3069c1bb86cdSEric Blake if (ioctl(s->fd, BLKROGET, &readonly) < 0) { 3070c1bb86cdSEric Blake return -errno; 3071c1bb86cdSEric Blake } 3072c1bb86cdSEric Blake 3073c1bb86cdSEric Blake if (readonly) { 3074c1bb86cdSEric Blake return -EACCES; 3075c1bb86cdSEric Blake } 3076c1bb86cdSEric Blake #endif /* defined(BLKROGET) */ 3077c1bb86cdSEric Blake return 0; 3078c1bb86cdSEric Blake } 3079c1bb86cdSEric Blake 3080c1bb86cdSEric Blake static void hdev_parse_filename(const char *filename, QDict *options, 3081c1bb86cdSEric Blake Error **errp) 3082c1bb86cdSEric Blake { 308303c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_device:", options); 3084c1bb86cdSEric Blake } 3085c1bb86cdSEric Blake 3086c1bb86cdSEric Blake static bool hdev_is_sg(BlockDriverState *bs) 3087c1bb86cdSEric Blake { 3088c1bb86cdSEric Blake 3089c1bb86cdSEric Blake #if defined(__linux__) 3090c1bb86cdSEric Blake 3091c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3092c1bb86cdSEric Blake struct stat st; 3093c1bb86cdSEric Blake struct sg_scsi_id scsiid; 3094c1bb86cdSEric Blake int sg_version; 3095c1bb86cdSEric Blake int ret; 3096c1bb86cdSEric Blake 3097c1bb86cdSEric Blake if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { 3098c1bb86cdSEric Blake return false; 3099c1bb86cdSEric Blake } 3100c1bb86cdSEric Blake 3101c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); 3102c1bb86cdSEric Blake if (ret < 0) { 3103c1bb86cdSEric Blake return false; 3104c1bb86cdSEric Blake } 3105c1bb86cdSEric Blake 3106c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); 3107c1bb86cdSEric Blake if (ret >= 0) { 31084f7d28d7SLaurent Vivier trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); 3109c1bb86cdSEric Blake return true; 3110c1bb86cdSEric Blake } 3111c1bb86cdSEric Blake 3112c1bb86cdSEric Blake #endif 3113c1bb86cdSEric Blake 3114c1bb86cdSEric Blake return false; 3115c1bb86cdSEric Blake } 3116c1bb86cdSEric Blake 3117c1bb86cdSEric Blake static int hdev_open(BlockDriverState *bs, QDict *options, int flags, 3118c1bb86cdSEric Blake Error **errp) 3119c1bb86cdSEric Blake { 3120c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3121c1bb86cdSEric Blake Error *local_err = NULL; 3122c1bb86cdSEric Blake int ret; 3123c1bb86cdSEric Blake 3124c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3125129c7d1cSMarkus Armbruster /* 3126129c7d1cSMarkus Armbruster * Caution: while qdict_get_str() is fine, getting non-string types 3127129c7d1cSMarkus Armbruster * would require more care. When @options come from -blockdev or 3128129c7d1cSMarkus Armbruster * blockdev_add, its members are typed according to the QAPI 3129129c7d1cSMarkus Armbruster * schema, but when they come from -drive, they're all QString. 3130129c7d1cSMarkus Armbruster */ 3131c1bb86cdSEric Blake const char *filename = qdict_get_str(options, "filename"); 3132c1bb86cdSEric Blake char bsd_path[MAXPATHLEN] = ""; 3133c1bb86cdSEric Blake bool error_occurred = false; 3134c1bb86cdSEric Blake 3135c1bb86cdSEric Blake /* If using a real cdrom */ 3136c1bb86cdSEric Blake if (strcmp(filename, "/dev/cdrom") == 0) { 3137c1bb86cdSEric Blake char *mediaType = NULL; 3138c1bb86cdSEric Blake kern_return_t ret_val; 3139c1bb86cdSEric Blake io_iterator_t mediaIterator = 0; 3140c1bb86cdSEric Blake 3141c1bb86cdSEric Blake mediaType = FindEjectableOpticalMedia(&mediaIterator); 3142c1bb86cdSEric Blake if (mediaType == NULL) { 3143c1bb86cdSEric Blake error_setg(errp, "Please make sure your CD/DVD is in the optical" 3144c1bb86cdSEric Blake " drive"); 3145c1bb86cdSEric Blake error_occurred = true; 3146c1bb86cdSEric Blake goto hdev_open_Mac_error; 3147c1bb86cdSEric Blake } 3148c1bb86cdSEric Blake 3149c1bb86cdSEric Blake ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); 3150c1bb86cdSEric Blake if (ret_val != KERN_SUCCESS) { 3151c1bb86cdSEric Blake error_setg(errp, "Could not get BSD path for optical drive"); 3152c1bb86cdSEric Blake error_occurred = true; 3153c1bb86cdSEric Blake goto hdev_open_Mac_error; 3154c1bb86cdSEric Blake } 3155c1bb86cdSEric Blake 3156c1bb86cdSEric Blake /* If a real optical drive was not found */ 3157c1bb86cdSEric Blake if (bsd_path[0] == '\0') { 3158c1bb86cdSEric Blake error_setg(errp, "Failed to obtain bsd path for optical drive"); 3159c1bb86cdSEric Blake error_occurred = true; 3160c1bb86cdSEric Blake goto hdev_open_Mac_error; 3161c1bb86cdSEric Blake } 3162c1bb86cdSEric Blake 3163c1bb86cdSEric Blake /* If using a cdrom disc and finding a partition on the disc failed */ 3164c1bb86cdSEric Blake if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && 3165c1bb86cdSEric Blake setup_cdrom(bsd_path, errp) == false) { 3166c1bb86cdSEric Blake print_unmounting_directions(bsd_path); 3167c1bb86cdSEric Blake error_occurred = true; 3168c1bb86cdSEric Blake goto hdev_open_Mac_error; 3169c1bb86cdSEric Blake } 3170c1bb86cdSEric Blake 317146f5ac20SEric Blake qdict_put_str(options, "filename", bsd_path); 3172c1bb86cdSEric Blake 3173c1bb86cdSEric Blake hdev_open_Mac_error: 3174c1bb86cdSEric Blake g_free(mediaType); 3175c1bb86cdSEric Blake if (mediaIterator) { 3176c1bb86cdSEric Blake IOObjectRelease(mediaIterator); 3177c1bb86cdSEric Blake } 3178c1bb86cdSEric Blake if (error_occurred) { 3179c1bb86cdSEric Blake return -ENOENT; 3180c1bb86cdSEric Blake } 3181c1bb86cdSEric Blake } 3182c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3183c1bb86cdSEric Blake 3184c1bb86cdSEric Blake s->type = FTYPE_FILE; 3185c1bb86cdSEric Blake 3186230ff739SJohn Snow ret = raw_open_common(bs, options, flags, 0, true, &local_err); 3187c1bb86cdSEric Blake if (ret < 0) { 3188c1bb86cdSEric Blake error_propagate(errp, local_err); 3189c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3190c1bb86cdSEric Blake if (*bsd_path) { 3191c1bb86cdSEric Blake filename = bsd_path; 3192c1bb86cdSEric Blake } 3193c1bb86cdSEric Blake /* if a physical device experienced an error while being opened */ 3194c1bb86cdSEric Blake if (strncmp(filename, "/dev/", 5) == 0) { 3195c1bb86cdSEric Blake print_unmounting_directions(filename); 3196c1bb86cdSEric Blake } 3197c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3198c1bb86cdSEric Blake return ret; 3199c1bb86cdSEric Blake } 3200c1bb86cdSEric Blake 3201c1bb86cdSEric Blake /* Since this does ioctl the device must be already opened */ 3202c1bb86cdSEric Blake bs->sg = hdev_is_sg(bs); 3203c1bb86cdSEric Blake 3204c1bb86cdSEric Blake if (flags & BDRV_O_RDWR) { 3205c1bb86cdSEric Blake ret = check_hdev_writable(s); 3206c1bb86cdSEric Blake if (ret < 0) { 3207c1bb86cdSEric Blake raw_close(bs); 3208c1bb86cdSEric Blake error_setg_errno(errp, -ret, "The device is not writable"); 3209c1bb86cdSEric Blake return ret; 3210c1bb86cdSEric Blake } 3211c1bb86cdSEric Blake } 3212c1bb86cdSEric Blake 3213c1bb86cdSEric Blake return ret; 3214c1bb86cdSEric Blake } 3215c1bb86cdSEric Blake 3216c1bb86cdSEric Blake #if defined(__linux__) 32172f3a7ab3SKevin Wolf static int coroutine_fn 32182f3a7ab3SKevin Wolf hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) 3219c1bb86cdSEric Blake { 3220c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 322103425671SKevin Wolf RawPosixAIOData acb; 32222f3a7ab3SKevin Wolf int ret; 3223c1bb86cdSEric Blake 32242f3a7ab3SKevin Wolf ret = fd_open(bs); 32252f3a7ab3SKevin Wolf if (ret < 0) { 32262f3a7ab3SKevin Wolf return ret; 32272f3a7ab3SKevin Wolf } 3228c1bb86cdSEric Blake 32297c9e5276SPaolo Bonzini if (req == SG_IO && s->pr_mgr) { 32307c9e5276SPaolo Bonzini struct sg_io_hdr *io_hdr = buf; 32317c9e5276SPaolo Bonzini if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || 32327c9e5276SPaolo Bonzini io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { 32337c9e5276SPaolo Bonzini return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), 32342f3a7ab3SKevin Wolf s->fd, io_hdr); 32357c9e5276SPaolo Bonzini } 32367c9e5276SPaolo Bonzini } 32377c9e5276SPaolo Bonzini 323803425671SKevin Wolf acb = (RawPosixAIOData) { 323903425671SKevin Wolf .bs = bs, 324003425671SKevin Wolf .aio_type = QEMU_AIO_IOCTL, 324103425671SKevin Wolf .aio_fildes = s->fd, 324203425671SKevin Wolf .aio_offset = 0, 324303425671SKevin Wolf .ioctl = { 324403425671SKevin Wolf .buf = buf, 324503425671SKevin Wolf .cmd = req, 324603425671SKevin Wolf }, 324703425671SKevin Wolf }; 324803425671SKevin Wolf 324903425671SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); 3250c1bb86cdSEric Blake } 3251c1bb86cdSEric Blake #endif /* linux */ 3252c1bb86cdSEric Blake 3253c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs) 3254c1bb86cdSEric Blake { 3255c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3256c1bb86cdSEric Blake 3257c1bb86cdSEric Blake /* this is just to ensure s->fd is sane (its called by io ops) */ 3258c1bb86cdSEric Blake if (s->fd >= 0) 3259c1bb86cdSEric Blake return 0; 3260c1bb86cdSEric Blake return -EIO; 3261c1bb86cdSEric Blake } 3262c1bb86cdSEric Blake 326333d70fb6SKevin Wolf static coroutine_fn int 326433d70fb6SKevin Wolf hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 3265c1bb86cdSEric Blake { 326633d70fb6SKevin Wolf int ret; 3267c1bb86cdSEric Blake 326833d70fb6SKevin Wolf ret = fd_open(bs); 326933d70fb6SKevin Wolf if (ret < 0) { 327033d70fb6SKevin Wolf return ret; 3271c1bb86cdSEric Blake } 327246ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, true); 3273c1bb86cdSEric Blake } 3274c1bb86cdSEric Blake 3275c1bb86cdSEric Blake static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, 3276f5a5ca79SManos Pitsidianakis int64_t offset, int bytes, BdrvRequestFlags flags) 3277c1bb86cdSEric Blake { 3278c1bb86cdSEric Blake int rc; 3279c1bb86cdSEric Blake 3280c1bb86cdSEric Blake rc = fd_open(bs); 3281c1bb86cdSEric Blake if (rc < 0) { 3282c1bb86cdSEric Blake return rc; 3283c1bb86cdSEric Blake } 328434fa110eSKevin Wolf 32857154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); 3286c1bb86cdSEric Blake } 3287c1bb86cdSEric Blake 3288efc75e2aSStefan Hajnoczi static int coroutine_fn hdev_co_create_opts(const char *filename, QemuOpts *opts, 3289c1bb86cdSEric Blake Error **errp) 3290c1bb86cdSEric Blake { 3291c1bb86cdSEric Blake int fd; 3292c1bb86cdSEric Blake int ret = 0; 3293c1bb86cdSEric Blake struct stat stat_buf; 3294c1bb86cdSEric Blake int64_t total_size = 0; 3295c1bb86cdSEric Blake bool has_prefix; 3296c1bb86cdSEric Blake 3297c1bb86cdSEric Blake /* This function is used by both protocol block drivers and therefore either 3298c1bb86cdSEric Blake * of these prefixes may be given. 3299c1bb86cdSEric Blake * The return value has to be stored somewhere, otherwise this is an error 3300c1bb86cdSEric Blake * due to -Werror=unused-value. */ 3301c1bb86cdSEric Blake has_prefix = 3302c1bb86cdSEric Blake strstart(filename, "host_device:", &filename) || 3303c1bb86cdSEric Blake strstart(filename, "host_cdrom:" , &filename); 3304c1bb86cdSEric Blake 3305c1bb86cdSEric Blake (void)has_prefix; 3306c1bb86cdSEric Blake 3307db0754dfSFam Zheng ret = raw_normalize_devicepath(&filename, errp); 3308c1bb86cdSEric Blake if (ret < 0) { 3309c1bb86cdSEric Blake return ret; 3310c1bb86cdSEric Blake } 3311c1bb86cdSEric Blake 3312c1bb86cdSEric Blake /* Read out options */ 3313c1bb86cdSEric Blake total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 3314c1bb86cdSEric Blake BDRV_SECTOR_SIZE); 3315c1bb86cdSEric Blake 3316c1bb86cdSEric Blake fd = qemu_open(filename, O_WRONLY | O_BINARY); 3317c1bb86cdSEric Blake if (fd < 0) { 3318c1bb86cdSEric Blake ret = -errno; 3319c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not open device"); 3320c1bb86cdSEric Blake return ret; 3321c1bb86cdSEric Blake } 3322c1bb86cdSEric Blake 3323c1bb86cdSEric Blake if (fstat(fd, &stat_buf) < 0) { 3324c1bb86cdSEric Blake ret = -errno; 3325c1bb86cdSEric Blake error_setg_errno(errp, -ret, "Could not stat device"); 3326c1bb86cdSEric Blake } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) { 3327c1bb86cdSEric Blake error_setg(errp, 3328c1bb86cdSEric Blake "The given file is neither a block nor a character device"); 3329c1bb86cdSEric Blake ret = -ENODEV; 3330c1bb86cdSEric Blake } else if (lseek(fd, 0, SEEK_END) < total_size) { 3331c1bb86cdSEric Blake error_setg(errp, "Device is too small"); 3332c1bb86cdSEric Blake ret = -ENOSPC; 3333c1bb86cdSEric Blake } 3334c1bb86cdSEric Blake 333597ec9117SFam Zheng if (!ret && total_size) { 333697ec9117SFam Zheng uint8_t buf[BDRV_SECTOR_SIZE] = { 0 }; 333797ec9117SFam Zheng int64_t zero_size = MIN(BDRV_SECTOR_SIZE, total_size); 333897ec9117SFam Zheng if (lseek(fd, 0, SEEK_SET) == -1) { 333997ec9117SFam Zheng ret = -errno; 334097ec9117SFam Zheng } else { 334197ec9117SFam Zheng ret = qemu_write_full(fd, buf, zero_size); 334297ec9117SFam Zheng ret = ret == zero_size ? 0 : -errno; 334397ec9117SFam Zheng } 334497ec9117SFam Zheng } 3345c1bb86cdSEric Blake qemu_close(fd); 3346c1bb86cdSEric Blake return ret; 3347c1bb86cdSEric Blake } 3348c1bb86cdSEric Blake 3349c1bb86cdSEric Blake static BlockDriver bdrv_host_device = { 3350c1bb86cdSEric Blake .format_name = "host_device", 3351c1bb86cdSEric Blake .protocol_name = "host_device", 3352c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3353c1bb86cdSEric Blake .bdrv_needs_filename = true, 3354c1bb86cdSEric Blake .bdrv_probe_device = hdev_probe_device, 3355c1bb86cdSEric Blake .bdrv_parse_filename = hdev_parse_filename, 3356c1bb86cdSEric Blake .bdrv_file_open = hdev_open, 3357c1bb86cdSEric Blake .bdrv_close = raw_close, 3358c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3359c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3360c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3361efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = hdev_co_create_opts, 3362c1bb86cdSEric Blake .create_opts = &raw_create_opts, 33638a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3364dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3365c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, 3366c1bb86cdSEric Blake 3367c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3368c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 336933d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 337033d70fb6SKevin Wolf .bdrv_co_pdiscard = hdev_co_pdiscard, 33711efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 33721efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 3373c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3374c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3375c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3376042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3377c1bb86cdSEric Blake 3378061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3379c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3380c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 3381c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3382c1bb86cdSEric Blake = raw_get_allocated_file_size, 3383244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 3384244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 3385244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 3386c1bb86cdSEric Blake .bdrv_probe_blocksizes = hdev_probe_blocksizes, 3387c1bb86cdSEric Blake .bdrv_probe_geometry = hdev_probe_geometry, 3388c1bb86cdSEric Blake 3389c1bb86cdSEric Blake /* generic scsi device */ 3390c1bb86cdSEric Blake #ifdef __linux__ 33912f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3392c1bb86cdSEric Blake #endif 3393c1bb86cdSEric Blake }; 3394c1bb86cdSEric Blake 3395c1bb86cdSEric Blake #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3396c1bb86cdSEric Blake static void cdrom_parse_filename(const char *filename, QDict *options, 3397c1bb86cdSEric Blake Error **errp) 3398c1bb86cdSEric Blake { 339903c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); 3400c1bb86cdSEric Blake } 3401c1bb86cdSEric Blake #endif 3402c1bb86cdSEric Blake 3403c1bb86cdSEric Blake #ifdef __linux__ 3404c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3405c1bb86cdSEric Blake Error **errp) 3406c1bb86cdSEric Blake { 3407c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3408c1bb86cdSEric Blake 3409c1bb86cdSEric Blake s->type = FTYPE_CD; 3410c1bb86cdSEric Blake 3411c1bb86cdSEric Blake /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ 3412230ff739SJohn Snow return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); 3413c1bb86cdSEric Blake } 3414c1bb86cdSEric Blake 3415c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3416c1bb86cdSEric Blake { 3417c1bb86cdSEric Blake int fd, ret; 3418c1bb86cdSEric Blake int prio = 0; 3419c1bb86cdSEric Blake struct stat st; 3420c1bb86cdSEric Blake 3421c1bb86cdSEric Blake fd = qemu_open(filename, O_RDONLY | O_NONBLOCK); 3422c1bb86cdSEric Blake if (fd < 0) { 3423c1bb86cdSEric Blake goto out; 3424c1bb86cdSEric Blake } 3425c1bb86cdSEric Blake ret = fstat(fd, &st); 3426c1bb86cdSEric Blake if (ret == -1 || !S_ISBLK(st.st_mode)) { 3427c1bb86cdSEric Blake goto outc; 3428c1bb86cdSEric Blake } 3429c1bb86cdSEric Blake 3430c1bb86cdSEric Blake /* Attempt to detect via a CDROM specific ioctl */ 3431c1bb86cdSEric Blake ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3432c1bb86cdSEric Blake if (ret >= 0) 3433c1bb86cdSEric Blake prio = 100; 3434c1bb86cdSEric Blake 3435c1bb86cdSEric Blake outc: 3436c1bb86cdSEric Blake qemu_close(fd); 3437c1bb86cdSEric Blake out: 3438c1bb86cdSEric Blake return prio; 3439c1bb86cdSEric Blake } 3440c1bb86cdSEric Blake 3441c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3442c1bb86cdSEric Blake { 3443c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3444c1bb86cdSEric Blake int ret; 3445c1bb86cdSEric Blake 3446c1bb86cdSEric Blake ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3447c1bb86cdSEric Blake return ret == CDS_DISC_OK; 3448c1bb86cdSEric Blake } 3449c1bb86cdSEric Blake 3450c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3451c1bb86cdSEric Blake { 3452c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3453c1bb86cdSEric Blake 3454c1bb86cdSEric Blake if (eject_flag) { 3455c1bb86cdSEric Blake if (ioctl(s->fd, CDROMEJECT, NULL) < 0) 3456c1bb86cdSEric Blake perror("CDROMEJECT"); 3457c1bb86cdSEric Blake } else { 3458c1bb86cdSEric Blake if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) 3459c1bb86cdSEric Blake perror("CDROMEJECT"); 3460c1bb86cdSEric Blake } 3461c1bb86cdSEric Blake } 3462c1bb86cdSEric Blake 3463c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3464c1bb86cdSEric Blake { 3465c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3466c1bb86cdSEric Blake 3467c1bb86cdSEric Blake if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { 3468c1bb86cdSEric Blake /* 3469c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3470c1bb86cdSEric Blake * mounts the CD-ROM 3471c1bb86cdSEric Blake */ 3472c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3473c1bb86cdSEric Blake } 3474c1bb86cdSEric Blake } 3475c1bb86cdSEric Blake 3476c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3477c1bb86cdSEric Blake .format_name = "host_cdrom", 3478c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3479c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3480c1bb86cdSEric Blake .bdrv_needs_filename = true, 3481c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3482c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3483c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3484c1bb86cdSEric Blake .bdrv_close = raw_close, 3485c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3486c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3487c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3488efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = hdev_co_create_opts, 3489c1bb86cdSEric Blake .create_opts = &raw_create_opts, 34908a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3491dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3492c1bb86cdSEric Blake 3493c1bb86cdSEric Blake 3494c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3495c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 349633d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3497c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3498c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3499c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3500042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3501c1bb86cdSEric Blake 3502061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3503c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3504c1bb86cdSEric Blake .has_variable_length = true, 3505c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3506c1bb86cdSEric Blake = raw_get_allocated_file_size, 3507c1bb86cdSEric Blake 3508c1bb86cdSEric Blake /* removable device support */ 3509c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3510c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3511c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3512c1bb86cdSEric Blake 3513c1bb86cdSEric Blake /* generic scsi device */ 35142f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3515c1bb86cdSEric Blake }; 3516c1bb86cdSEric Blake #endif /* __linux__ */ 3517c1bb86cdSEric Blake 3518c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 3519c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3520c1bb86cdSEric Blake Error **errp) 3521c1bb86cdSEric Blake { 3522c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3523c1bb86cdSEric Blake Error *local_err = NULL; 3524c1bb86cdSEric Blake int ret; 3525c1bb86cdSEric Blake 3526c1bb86cdSEric Blake s->type = FTYPE_CD; 3527c1bb86cdSEric Blake 3528230ff739SJohn Snow ret = raw_open_common(bs, options, flags, 0, true, &local_err); 3529c1bb86cdSEric Blake if (ret) { 3530c1bb86cdSEric Blake error_propagate(errp, local_err); 3531c1bb86cdSEric Blake return ret; 3532c1bb86cdSEric Blake } 3533c1bb86cdSEric Blake 3534c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3535c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3536c1bb86cdSEric Blake return 0; 3537c1bb86cdSEric Blake } 3538c1bb86cdSEric Blake 3539c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3540c1bb86cdSEric Blake { 3541c1bb86cdSEric Blake if (strstart(filename, "/dev/cd", NULL) || 3542c1bb86cdSEric Blake strstart(filename, "/dev/acd", NULL)) 3543c1bb86cdSEric Blake return 100; 3544c1bb86cdSEric Blake return 0; 3545c1bb86cdSEric Blake } 3546c1bb86cdSEric Blake 3547c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs) 3548c1bb86cdSEric Blake { 3549c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3550c1bb86cdSEric Blake int fd; 3551c1bb86cdSEric Blake 3552c1bb86cdSEric Blake /* 3553c1bb86cdSEric Blake * Force reread of possibly changed/newly loaded disc, 3554c1bb86cdSEric Blake * FreeBSD seems to not notice sometimes... 3555c1bb86cdSEric Blake */ 3556c1bb86cdSEric Blake if (s->fd >= 0) 3557c1bb86cdSEric Blake qemu_close(s->fd); 3558c1bb86cdSEric Blake fd = qemu_open(bs->filename, s->open_flags, 0644); 3559c1bb86cdSEric Blake if (fd < 0) { 3560c1bb86cdSEric Blake s->fd = -1; 3561c1bb86cdSEric Blake return -EIO; 3562c1bb86cdSEric Blake } 3563c1bb86cdSEric Blake s->fd = fd; 3564c1bb86cdSEric Blake 3565c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3566c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3567c1bb86cdSEric Blake return 0; 3568c1bb86cdSEric Blake } 3569c1bb86cdSEric Blake 3570c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3571c1bb86cdSEric Blake { 3572c1bb86cdSEric Blake return raw_getlength(bs) > 0; 3573c1bb86cdSEric Blake } 3574c1bb86cdSEric Blake 3575c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3576c1bb86cdSEric Blake { 3577c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3578c1bb86cdSEric Blake 3579c1bb86cdSEric Blake if (s->fd < 0) 3580c1bb86cdSEric Blake return; 3581c1bb86cdSEric Blake 3582c1bb86cdSEric Blake (void) ioctl(s->fd, CDIOCALLOW); 3583c1bb86cdSEric Blake 3584c1bb86cdSEric Blake if (eject_flag) { 3585c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCEJECT) < 0) 3586c1bb86cdSEric Blake perror("CDIOCEJECT"); 3587c1bb86cdSEric Blake } else { 3588c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCCLOSE) < 0) 3589c1bb86cdSEric Blake perror("CDIOCCLOSE"); 3590c1bb86cdSEric Blake } 3591c1bb86cdSEric Blake 3592c1bb86cdSEric Blake cdrom_reopen(bs); 3593c1bb86cdSEric Blake } 3594c1bb86cdSEric Blake 3595c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3596c1bb86cdSEric Blake { 3597c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3598c1bb86cdSEric Blake 3599c1bb86cdSEric Blake if (s->fd < 0) 3600c1bb86cdSEric Blake return; 3601c1bb86cdSEric Blake if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { 3602c1bb86cdSEric Blake /* 3603c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3604c1bb86cdSEric Blake * mounts the CD-ROM 3605c1bb86cdSEric Blake */ 3606c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3607c1bb86cdSEric Blake } 3608c1bb86cdSEric Blake } 3609c1bb86cdSEric Blake 3610c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3611c1bb86cdSEric Blake .format_name = "host_cdrom", 3612c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3613c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3614c1bb86cdSEric Blake .bdrv_needs_filename = true, 3615c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3616c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3617c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3618c1bb86cdSEric Blake .bdrv_close = raw_close, 3619c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3620c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3621c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3622efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = hdev_co_create_opts, 3623c1bb86cdSEric Blake .create_opts = &raw_create_opts, 36248a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3625c1bb86cdSEric Blake 3626c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3627c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 362833d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3629c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3630c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3631c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3632042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3633c1bb86cdSEric Blake 3634061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3635c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3636c1bb86cdSEric Blake .has_variable_length = true, 3637c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3638c1bb86cdSEric Blake = raw_get_allocated_file_size, 3639c1bb86cdSEric Blake 3640c1bb86cdSEric Blake /* removable device support */ 3641c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3642c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3643c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3644c1bb86cdSEric Blake }; 3645c1bb86cdSEric Blake #endif /* __FreeBSD__ */ 3646c1bb86cdSEric Blake 3647c1bb86cdSEric Blake static void bdrv_file_init(void) 3648c1bb86cdSEric Blake { 3649c1bb86cdSEric Blake /* 3650c1bb86cdSEric Blake * Register all the drivers. Note that order is important, the driver 3651c1bb86cdSEric Blake * registered last will get probed first. 3652c1bb86cdSEric Blake */ 3653c1bb86cdSEric Blake bdrv_register(&bdrv_file); 3654c1bb86cdSEric Blake bdrv_register(&bdrv_host_device); 3655c1bb86cdSEric Blake #ifdef __linux__ 3656c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3657c1bb86cdSEric Blake #endif 3658c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3659c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3660c1bb86cdSEric Blake #endif 3661c1bb86cdSEric Blake } 3662c1bb86cdSEric Blake 3663c1bb86cdSEric Blake block_init(bdrv_file_init); 3664