1c1bb86cdSEric Blake /* 2c1bb86cdSEric Blake * Block driver for RAW files (posix) 3c1bb86cdSEric Blake * 4c1bb86cdSEric Blake * Copyright (c) 2006 Fabrice Bellard 5c1bb86cdSEric Blake * 6c1bb86cdSEric Blake * Permission is hereby granted, free of charge, to any person obtaining a copy 7c1bb86cdSEric Blake * of this software and associated documentation files (the "Software"), to deal 8c1bb86cdSEric Blake * in the Software without restriction, including without limitation the rights 9c1bb86cdSEric Blake * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10c1bb86cdSEric Blake * copies of the Software, and to permit persons to whom the Software is 11c1bb86cdSEric Blake * furnished to do so, subject to the following conditions: 12c1bb86cdSEric Blake * 13c1bb86cdSEric Blake * The above copyright notice and this permission notice shall be included in 14c1bb86cdSEric Blake * all copies or substantial portions of the Software. 15c1bb86cdSEric Blake * 16c1bb86cdSEric Blake * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17c1bb86cdSEric Blake * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18c1bb86cdSEric Blake * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19c1bb86cdSEric Blake * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20c1bb86cdSEric Blake * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21c1bb86cdSEric Blake * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22c1bb86cdSEric Blake * THE SOFTWARE. 23c1bb86cdSEric Blake */ 24922a01a0SMarkus Armbruster 25c1bb86cdSEric Blake #include "qemu/osdep.h" 26a8d25326SMarkus Armbruster #include "qemu-common.h" 27c1bb86cdSEric Blake #include "qapi/error.h" 28c1bb86cdSEric Blake #include "qemu/cutils.h" 29c1bb86cdSEric Blake #include "qemu/error-report.h" 30c1bb86cdSEric Blake #include "block/block_int.h" 31c1bb86cdSEric Blake #include "qemu/module.h" 32922a01a0SMarkus Armbruster #include "qemu/option.h" 33ffa244c8SKevin Wolf #include "qemu/units.h" 345df022cfSPeter Maydell #include "qemu/memalign.h" 35c1bb86cdSEric Blake #include "trace.h" 36c1bb86cdSEric Blake #include "block/thread-pool.h" 37c1bb86cdSEric Blake #include "qemu/iov.h" 38c1bb86cdSEric Blake #include "block/raw-aio.h" 39452fcdbcSMarkus Armbruster #include "qapi/qmp/qdict.h" 40c1bb86cdSEric Blake #include "qapi/qmp/qstring.h" 41c1bb86cdSEric Blake 427c9e5276SPaolo Bonzini #include "scsi/pr-manager.h" 437c9e5276SPaolo Bonzini #include "scsi/constants.h" 447c9e5276SPaolo Bonzini 45c1bb86cdSEric Blake #if defined(__APPLE__) && (__MACH__) 4614176c8dSJoelle van Dyne #include <sys/ioctl.h> 4714176c8dSJoelle van Dyne #if defined(HAVE_HOST_BLOCK_DEVICE) 48c1bb86cdSEric Blake #include <paths.h> 49c1bb86cdSEric Blake #include <sys/param.h> 500dfc7af2SAkihiko Odaki #include <sys/mount.h> 51c1bb86cdSEric Blake #include <IOKit/IOKitLib.h> 52c1bb86cdSEric Blake #include <IOKit/IOBSD.h> 53c1bb86cdSEric Blake #include <IOKit/storage/IOMediaBSDClient.h> 54c1bb86cdSEric Blake #include <IOKit/storage/IOMedia.h> 55c1bb86cdSEric Blake #include <IOKit/storage/IOCDMedia.h> 56c1bb86cdSEric Blake //#include <IOKit/storage/IOCDTypes.h> 57c1bb86cdSEric Blake #include <IOKit/storage/IODVDMedia.h> 58c1bb86cdSEric Blake #include <CoreFoundation/CoreFoundation.h> 5914176c8dSJoelle van Dyne #endif /* defined(HAVE_HOST_BLOCK_DEVICE) */ 60c1bb86cdSEric Blake #endif 61c1bb86cdSEric Blake 62c1bb86cdSEric Blake #ifdef __sun__ 63c1bb86cdSEric Blake #define _POSIX_PTHREAD_SEMANTICS 1 64c1bb86cdSEric Blake #include <sys/dkio.h> 65c1bb86cdSEric Blake #endif 66c1bb86cdSEric Blake #ifdef __linux__ 67c1bb86cdSEric Blake #include <sys/ioctl.h> 68c1bb86cdSEric Blake #include <sys/param.h> 691efad060SFam Zheng #include <sys/syscall.h> 705edc8557SKevin Wolf #include <sys/vfs.h> 71c1bb86cdSEric Blake #include <linux/cdrom.h> 72c1bb86cdSEric Blake #include <linux/fd.h> 73c1bb86cdSEric Blake #include <linux/fs.h> 74c1bb86cdSEric Blake #include <linux/hdreg.h> 755edc8557SKevin Wolf #include <linux/magic.h> 76c1bb86cdSEric Blake #include <scsi/sg.h> 77c1bb86cdSEric Blake #ifdef __s390__ 78c1bb86cdSEric Blake #include <asm/dasd.h> 79c1bb86cdSEric Blake #endif 80c1bb86cdSEric Blake #ifndef FS_NOCOW_FL 81c1bb86cdSEric Blake #define FS_NOCOW_FL 0x00800000 /* Do not cow file */ 82c1bb86cdSEric Blake #endif 83c1bb86cdSEric Blake #endif 84c1bb86cdSEric Blake #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE) 85c1bb86cdSEric Blake #include <linux/falloc.h> 86c1bb86cdSEric Blake #endif 87c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 88c1bb86cdSEric Blake #include <sys/disk.h> 89c1bb86cdSEric Blake #include <sys/cdio.h> 90c1bb86cdSEric Blake #endif 91c1bb86cdSEric Blake 92c1bb86cdSEric Blake #ifdef __OpenBSD__ 93c1bb86cdSEric Blake #include <sys/ioctl.h> 94c1bb86cdSEric Blake #include <sys/disklabel.h> 95c1bb86cdSEric Blake #include <sys/dkio.h> 96c1bb86cdSEric Blake #endif 97c1bb86cdSEric Blake 98c1bb86cdSEric Blake #ifdef __NetBSD__ 99c1bb86cdSEric Blake #include <sys/ioctl.h> 100c1bb86cdSEric Blake #include <sys/disklabel.h> 101c1bb86cdSEric Blake #include <sys/dkio.h> 102c1bb86cdSEric Blake #include <sys/disk.h> 103c1bb86cdSEric Blake #endif 104c1bb86cdSEric Blake 105c1bb86cdSEric Blake #ifdef __DragonFly__ 106c1bb86cdSEric Blake #include <sys/ioctl.h> 107c1bb86cdSEric Blake #include <sys/diskslice.h> 108c1bb86cdSEric Blake #endif 109c1bb86cdSEric Blake 110c1bb86cdSEric Blake /* OS X does not have O_DSYNC */ 111c1bb86cdSEric Blake #ifndef O_DSYNC 112c1bb86cdSEric Blake #ifdef O_SYNC 113c1bb86cdSEric Blake #define O_DSYNC O_SYNC 114c1bb86cdSEric Blake #elif defined(O_FSYNC) 115c1bb86cdSEric Blake #define O_DSYNC O_FSYNC 116c1bb86cdSEric Blake #endif 117c1bb86cdSEric Blake #endif 118c1bb86cdSEric Blake 119c1bb86cdSEric Blake /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ 120c1bb86cdSEric Blake #ifndef O_DIRECT 121c1bb86cdSEric Blake #define O_DIRECT O_DSYNC 122c1bb86cdSEric Blake #endif 123c1bb86cdSEric Blake 124c1bb86cdSEric Blake #define FTYPE_FILE 0 125c1bb86cdSEric Blake #define FTYPE_CD 1 126c1bb86cdSEric Blake 127c1bb86cdSEric Blake #define MAX_BLOCKSIZE 4096 128c1bb86cdSEric Blake 129244a5668SFam Zheng /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes, 130244a5668SFam Zheng * leaving a few more bytes for its future use. */ 131244a5668SFam Zheng #define RAW_LOCK_PERM_BASE 100 132244a5668SFam Zheng #define RAW_LOCK_SHARED_BASE 200 133244a5668SFam Zheng 134c1bb86cdSEric Blake typedef struct BDRVRawState { 135c1bb86cdSEric Blake int fd; 136244a5668SFam Zheng bool use_lock; 137c1bb86cdSEric Blake int type; 138c1bb86cdSEric Blake int open_flags; 139c1bb86cdSEric Blake size_t buf_align; 140c1bb86cdSEric Blake 141244a5668SFam Zheng /* The current permissions. */ 142244a5668SFam Zheng uint64_t perm; 143244a5668SFam Zheng uint64_t shared_perm; 144244a5668SFam Zheng 1452996ffadSFam Zheng /* The perms bits whose corresponding bytes are already locked in 146f2e3af29SFam Zheng * s->fd. */ 1472996ffadSFam Zheng uint64_t locked_perm; 1482996ffadSFam Zheng uint64_t locked_shared_perm; 1492996ffadSFam Zheng 150684960d4SStefano Garzarella uint64_t aio_max_batch; 151684960d4SStefano Garzarella 1526ceabe6fSKevin Wolf int perm_change_fd; 153094e3639SMax Reitz int perm_change_flags; 154e0c9cf3aSKevin Wolf BDRVReopenState *reopen_state; 155e0c9cf3aSKevin Wolf 156c1bb86cdSEric Blake bool has_discard:1; 157c1bb86cdSEric Blake bool has_write_zeroes:1; 158c1bb86cdSEric Blake bool discard_zeroes:1; 159c1bb86cdSEric Blake bool use_linux_aio:1; 160c6447510SAarushi Mehta bool use_linux_io_uring:1; 161c7ddc882SDaniel P. Berrangé int page_cache_inconsistent; /* errno from fdatasync failure */ 162c1bb86cdSEric Blake bool has_fallocate; 163c1bb86cdSEric Blake bool needs_alignment; 1645dbd0ce1SKevin Wolf bool force_alignment; 165f357fcd8SStefan Hajnoczi bool drop_cache; 16631be8a2aSStefan Hajnoczi bool check_cache_dropped; 1671c450366SAnton Nefedov struct { 1681c450366SAnton Nefedov uint64_t discard_nb_ok; 1691c450366SAnton Nefedov uint64_t discard_nb_failed; 1701c450366SAnton Nefedov uint64_t discard_bytes_ok; 1711c450366SAnton Nefedov } stats; 1727c9e5276SPaolo Bonzini 1737c9e5276SPaolo Bonzini PRManager *pr_mgr; 174c1bb86cdSEric Blake } BDRVRawState; 175c1bb86cdSEric Blake 176c1bb86cdSEric Blake typedef struct BDRVRawReopenState { 177c1bb86cdSEric Blake int open_flags; 178f357fcd8SStefan Hajnoczi bool drop_cache; 17931be8a2aSStefan Hajnoczi bool check_cache_dropped; 180c1bb86cdSEric Blake } BDRVRawReopenState; 181c1bb86cdSEric Blake 18214176c8dSJoelle van Dyne static int fd_open(BlockDriverState *bs) 18314176c8dSJoelle van Dyne { 18414176c8dSJoelle van Dyne BDRVRawState *s = bs->opaque; 18514176c8dSJoelle van Dyne 18614176c8dSJoelle van Dyne /* this is just to ensure s->fd is sane (its called by io ops) */ 18714176c8dSJoelle van Dyne if (s->fd >= 0) { 18814176c8dSJoelle van Dyne return 0; 18914176c8dSJoelle van Dyne } 19014176c8dSJoelle van Dyne return -EIO; 19114176c8dSJoelle van Dyne } 19214176c8dSJoelle van Dyne 193c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs); 194c1bb86cdSEric Blake 195c1bb86cdSEric Blake typedef struct RawPosixAIOData { 196c1bb86cdSEric Blake BlockDriverState *bs; 197c1bb86cdSEric Blake int aio_type; 198d57c44d0SKevin Wolf int aio_fildes; 199d57c44d0SKevin Wolf 200d57c44d0SKevin Wolf off_t aio_offset; 201d57c44d0SKevin Wolf uint64_t aio_nbytes; 202d57c44d0SKevin Wolf 20393f4e2ffSKevin Wolf union { 20493f4e2ffSKevin Wolf struct { 205d57c44d0SKevin Wolf struct iovec *iov; 206d57c44d0SKevin Wolf int niov; 207d57c44d0SKevin Wolf } io; 208d57c44d0SKevin Wolf struct { 209d57c44d0SKevin Wolf uint64_t cmd; 210d57c44d0SKevin Wolf void *buf; 211d57c44d0SKevin Wolf } ioctl; 212d57c44d0SKevin Wolf struct { 2131efad060SFam Zheng int aio_fd2; 2141efad060SFam Zheng off_t aio_offset2; 215d57c44d0SKevin Wolf } copy_range; 21693f4e2ffSKevin Wolf struct { 21793f4e2ffSKevin Wolf PreallocMode prealloc; 21893f4e2ffSKevin Wolf Error **errp; 219d57c44d0SKevin Wolf } truncate; 22093f4e2ffSKevin Wolf }; 221c1bb86cdSEric Blake } RawPosixAIOData; 222c1bb86cdSEric Blake 223c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 224c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs); 225c1bb86cdSEric Blake #endif 226c1bb86cdSEric Blake 227797e3e38SDavid Edmondson /* 228797e3e38SDavid Edmondson * Elide EAGAIN and EACCES details when failing to lock, as this 229797e3e38SDavid Edmondson * indicates that the specified file region is already locked by 230797e3e38SDavid Edmondson * another process, which is considered a common scenario. 231797e3e38SDavid Edmondson */ 232797e3e38SDavid Edmondson #define raw_lock_error_setg_errno(errp, err, fmt, ...) \ 233797e3e38SDavid Edmondson do { \ 234797e3e38SDavid Edmondson if ((err) == EAGAIN || (err) == EACCES) { \ 235797e3e38SDavid Edmondson error_setg((errp), (fmt), ## __VA_ARGS__); \ 236797e3e38SDavid Edmondson } else { \ 237797e3e38SDavid Edmondson error_setg_errno((errp), (err), (fmt), ## __VA_ARGS__); \ 238797e3e38SDavid Edmondson } \ 239797e3e38SDavid Edmondson } while (0) 240797e3e38SDavid Edmondson 241c1bb86cdSEric Blake #if defined(__NetBSD__) 242db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 243c1bb86cdSEric Blake { 244c1bb86cdSEric Blake static char namebuf[PATH_MAX]; 245c1bb86cdSEric Blake const char *dp, *fname; 246c1bb86cdSEric Blake struct stat sb; 247c1bb86cdSEric Blake 248c1bb86cdSEric Blake fname = *filename; 249c1bb86cdSEric Blake dp = strrchr(fname, '/'); 250c1bb86cdSEric Blake if (lstat(fname, &sb) < 0) { 251f6fc1e30SPaolo Bonzini error_setg_file_open(errp, errno, fname); 252c1bb86cdSEric Blake return -errno; 253c1bb86cdSEric Blake } 254c1bb86cdSEric Blake 255c1bb86cdSEric Blake if (!S_ISBLK(sb.st_mode)) { 256c1bb86cdSEric Blake return 0; 257c1bb86cdSEric Blake } 258c1bb86cdSEric Blake 259c1bb86cdSEric Blake if (dp == NULL) { 260c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "r%s", fname); 261c1bb86cdSEric Blake } else { 262c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "%.*s/r%s", 263c1bb86cdSEric Blake (int)(dp - fname), fname, dp + 1); 264c1bb86cdSEric Blake } 265c1bb86cdSEric Blake *filename = namebuf; 266db0754dfSFam Zheng warn_report("%s is a block device, using %s", fname, *filename); 267c1bb86cdSEric Blake 268c1bb86cdSEric Blake return 0; 269c1bb86cdSEric Blake } 270c1bb86cdSEric Blake #else 271db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 272c1bb86cdSEric Blake { 273c1bb86cdSEric Blake return 0; 274c1bb86cdSEric Blake } 275c1bb86cdSEric Blake #endif 276c1bb86cdSEric Blake 277c1bb86cdSEric Blake /* 278c1bb86cdSEric Blake * Get logical block size via ioctl. On success store it in @sector_size_p. 279c1bb86cdSEric Blake */ 280c1bb86cdSEric Blake static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) 281c1bb86cdSEric Blake { 282c1bb86cdSEric Blake unsigned int sector_size; 283c1bb86cdSEric Blake bool success = false; 284700f9ce0SPeter Maydell int i; 285c1bb86cdSEric Blake 286c1bb86cdSEric Blake errno = ENOTSUP; 287700f9ce0SPeter Maydell static const unsigned long ioctl_list[] = { 288c1bb86cdSEric Blake #ifdef BLKSSZGET 289700f9ce0SPeter Maydell BLKSSZGET, 290c1bb86cdSEric Blake #endif 291c1bb86cdSEric Blake #ifdef DKIOCGETBLOCKSIZE 292700f9ce0SPeter Maydell DKIOCGETBLOCKSIZE, 293c1bb86cdSEric Blake #endif 294c1bb86cdSEric Blake #ifdef DIOCGSECTORSIZE 295700f9ce0SPeter Maydell DIOCGSECTORSIZE, 296700f9ce0SPeter Maydell #endif 297700f9ce0SPeter Maydell }; 298700f9ce0SPeter Maydell 299700f9ce0SPeter Maydell /* Try a few ioctls to get the right size */ 300700f9ce0SPeter Maydell for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { 301700f9ce0SPeter Maydell if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { 302c1bb86cdSEric Blake *sector_size_p = sector_size; 303c1bb86cdSEric Blake success = true; 304c1bb86cdSEric Blake } 305700f9ce0SPeter Maydell } 306c1bb86cdSEric Blake 307c1bb86cdSEric Blake return success ? 0 : -errno; 308c1bb86cdSEric Blake } 309c1bb86cdSEric Blake 310c1bb86cdSEric Blake /** 311c1bb86cdSEric Blake * Get physical block size of @fd. 312c1bb86cdSEric Blake * On success, store it in @blk_size and return 0. 313c1bb86cdSEric Blake * On failure, return -errno. 314c1bb86cdSEric Blake */ 315c1bb86cdSEric Blake static int probe_physical_blocksize(int fd, unsigned int *blk_size) 316c1bb86cdSEric Blake { 317c1bb86cdSEric Blake #ifdef BLKPBSZGET 318c1bb86cdSEric Blake if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { 319c1bb86cdSEric Blake return -errno; 320c1bb86cdSEric Blake } 321c1bb86cdSEric Blake return 0; 322c1bb86cdSEric Blake #else 323c1bb86cdSEric Blake return -ENOTSUP; 324c1bb86cdSEric Blake #endif 325c1bb86cdSEric Blake } 326c1bb86cdSEric Blake 3275edc8557SKevin Wolf /* 3285edc8557SKevin Wolf * Returns true if no alignment restrictions are necessary even for files 3295edc8557SKevin Wolf * opened with O_DIRECT. 3305edc8557SKevin Wolf * 3315edc8557SKevin Wolf * raw_probe_alignment() probes the required alignment and assume that 1 means 3325edc8557SKevin Wolf * the probing failed, so it falls back to a safe default of 4k. This can be 3335edc8557SKevin Wolf * avoided if we know that byte alignment is okay for the file. 3345edc8557SKevin Wolf */ 3355edc8557SKevin Wolf static bool dio_byte_aligned(int fd) 3365edc8557SKevin Wolf { 3375edc8557SKevin Wolf #ifdef __linux__ 3385edc8557SKevin Wolf struct statfs buf; 3395edc8557SKevin Wolf int ret; 3405edc8557SKevin Wolf 3415edc8557SKevin Wolf ret = fstatfs(fd, &buf); 3425edc8557SKevin Wolf if (ret == 0 && buf.f_type == NFS_SUPER_MAGIC) { 3435edc8557SKevin Wolf return true; 3445edc8557SKevin Wolf } 3455edc8557SKevin Wolf #endif 3465edc8557SKevin Wolf return false; 3475edc8557SKevin Wolf } 3485edc8557SKevin Wolf 3495dbd0ce1SKevin Wolf static bool raw_needs_alignment(BlockDriverState *bs) 3505dbd0ce1SKevin Wolf { 3515dbd0ce1SKevin Wolf BDRVRawState *s = bs->opaque; 3525dbd0ce1SKevin Wolf 3535dbd0ce1SKevin Wolf if ((bs->open_flags & BDRV_O_NOCACHE) != 0 && !dio_byte_aligned(s->fd)) { 3545dbd0ce1SKevin Wolf return true; 3555dbd0ce1SKevin Wolf } 3565dbd0ce1SKevin Wolf 3575dbd0ce1SKevin Wolf return s->force_alignment; 3585dbd0ce1SKevin Wolf } 3595dbd0ce1SKevin Wolf 360c1bb86cdSEric Blake /* Check if read is allowed with given memory buffer and length. 361c1bb86cdSEric Blake * 362c1bb86cdSEric Blake * This function is used to check O_DIRECT memory buffer and request alignment. 363c1bb86cdSEric Blake */ 364c1bb86cdSEric Blake static bool raw_is_io_aligned(int fd, void *buf, size_t len) 365c1bb86cdSEric Blake { 366c1bb86cdSEric Blake ssize_t ret = pread(fd, buf, len, 0); 367c1bb86cdSEric Blake 368c1bb86cdSEric Blake if (ret >= 0) { 369c1bb86cdSEric Blake return true; 370c1bb86cdSEric Blake } 371c1bb86cdSEric Blake 372c1bb86cdSEric Blake #ifdef __linux__ 373c1bb86cdSEric Blake /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore 374c1bb86cdSEric Blake * other errors (e.g. real I/O error), which could happen on a failed 375c1bb86cdSEric Blake * drive, since we only care about probing alignment. 376c1bb86cdSEric Blake */ 377c1bb86cdSEric Blake if (errno != EINVAL) { 378c1bb86cdSEric Blake return true; 379c1bb86cdSEric Blake } 380c1bb86cdSEric Blake #endif 381c1bb86cdSEric Blake 382c1bb86cdSEric Blake return false; 383c1bb86cdSEric Blake } 384c1bb86cdSEric Blake 385c1bb86cdSEric Blake static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) 386c1bb86cdSEric Blake { 387c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 388c1bb86cdSEric Blake char *buf; 389038adc2fSWei Yang size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); 390a6b257a0SNir Soffer size_t alignments[] = {1, 512, 1024, 2048, 4096}; 391c1bb86cdSEric Blake 392c1bb86cdSEric Blake /* For SCSI generic devices the alignment is not really used. 393c1bb86cdSEric Blake With buffered I/O, we don't have any restrictions. */ 394c1bb86cdSEric Blake if (bdrv_is_sg(bs) || !s->needs_alignment) { 395c1bb86cdSEric Blake bs->bl.request_alignment = 1; 396c1bb86cdSEric Blake s->buf_align = 1; 397c1bb86cdSEric Blake return; 398c1bb86cdSEric Blake } 399c1bb86cdSEric Blake 400c1bb86cdSEric Blake bs->bl.request_alignment = 0; 401c1bb86cdSEric Blake s->buf_align = 0; 402c1bb86cdSEric Blake /* Let's try to use the logical blocksize for the alignment. */ 403c1bb86cdSEric Blake if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { 404c1bb86cdSEric Blake bs->bl.request_alignment = 0; 405c1bb86cdSEric Blake } 406a5730b8bSThomas Huth 407a5730b8bSThomas Huth #ifdef __linux__ 408a5730b8bSThomas Huth /* 409a5730b8bSThomas Huth * The XFS ioctl definitions are shipped in extra packages that might 410a5730b8bSThomas Huth * not always be available. Since we just need the XFS_IOC_DIOINFO ioctl 411a5730b8bSThomas Huth * here, we simply use our own definition instead: 412a5730b8bSThomas Huth */ 413a5730b8bSThomas Huth struct xfs_dioattr { 414a5730b8bSThomas Huth uint32_t d_mem; 415a5730b8bSThomas Huth uint32_t d_miniosz; 416a5730b8bSThomas Huth uint32_t d_maxiosz; 417a5730b8bSThomas Huth } da; 418a5730b8bSThomas Huth if (ioctl(fd, _IOR('X', 30, struct xfs_dioattr), &da) >= 0) { 419c1bb86cdSEric Blake bs->bl.request_alignment = da.d_miniosz; 420c1bb86cdSEric Blake /* The kernel returns wrong information for d_mem */ 421c1bb86cdSEric Blake /* s->buf_align = da.d_mem; */ 422c1bb86cdSEric Blake } 423c1bb86cdSEric Blake #endif 424c1bb86cdSEric Blake 425a6b257a0SNir Soffer /* 426a6b257a0SNir Soffer * If we could not get the sizes so far, we can only guess them. First try 427a6b257a0SNir Soffer * to detect request alignment, since it is more likely to succeed. Then 428a6b257a0SNir Soffer * try to detect buf_align, which cannot be detected in some cases (e.g. 429a6b257a0SNir Soffer * Gluster). If buf_align cannot be detected, we fallback to the value of 430a6b257a0SNir Soffer * request_alignment. 431a6b257a0SNir Soffer */ 432a6b257a0SNir Soffer 433a6b257a0SNir Soffer if (!bs->bl.request_alignment) { 434a6b257a0SNir Soffer int i; 435c1bb86cdSEric Blake size_t align; 436a6b257a0SNir Soffer buf = qemu_memalign(max_align, max_align); 437a6b257a0SNir Soffer for (i = 0; i < ARRAY_SIZE(alignments); i++) { 438a6b257a0SNir Soffer align = alignments[i]; 439a6b257a0SNir Soffer if (raw_is_io_aligned(fd, buf, align)) { 440a6b257a0SNir Soffer /* Fallback to safe value. */ 441a6b257a0SNir Soffer bs->bl.request_alignment = (align != 1) ? align : max_align; 442c1bb86cdSEric Blake break; 443c1bb86cdSEric Blake } 444c1bb86cdSEric Blake } 445c1bb86cdSEric Blake qemu_vfree(buf); 446c1bb86cdSEric Blake } 447c1bb86cdSEric Blake 448a6b257a0SNir Soffer if (!s->buf_align) { 449a6b257a0SNir Soffer int i; 450c1bb86cdSEric Blake size_t align; 451a6b257a0SNir Soffer buf = qemu_memalign(max_align, 2 * max_align); 452a6b257a0SNir Soffer for (i = 0; i < ARRAY_SIZE(alignments); i++) { 453a6b257a0SNir Soffer align = alignments[i]; 454a6b257a0SNir Soffer if (raw_is_io_aligned(fd, buf + align, max_align)) { 455236094c7SStefan Hajnoczi /* Fallback to request_alignment. */ 456a6b257a0SNir Soffer s->buf_align = (align != 1) ? align : bs->bl.request_alignment; 457c1bb86cdSEric Blake break; 458c1bb86cdSEric Blake } 459c1bb86cdSEric Blake } 460c1bb86cdSEric Blake qemu_vfree(buf); 461c1bb86cdSEric Blake } 462c1bb86cdSEric Blake 463c1bb86cdSEric Blake if (!s->buf_align || !bs->bl.request_alignment) { 464c1bb86cdSEric Blake error_setg(errp, "Could not find working O_DIRECT alignment"); 465c1bb86cdSEric Blake error_append_hint(errp, "Try cache.direct=off\n"); 466c1bb86cdSEric Blake } 467c1bb86cdSEric Blake } 468c1bb86cdSEric Blake 469bca5283bSKevin Wolf static int check_hdev_writable(int fd) 47020eaf1bfSKevin Wolf { 47120eaf1bfSKevin Wolf #if defined(BLKROGET) 47220eaf1bfSKevin Wolf /* Linux block devices can be configured "read-only" using blockdev(8). 47320eaf1bfSKevin Wolf * This is independent of device node permissions and therefore open(2) 47420eaf1bfSKevin Wolf * with O_RDWR succeeds. Actual writes fail with EPERM. 47520eaf1bfSKevin Wolf * 47620eaf1bfSKevin Wolf * bdrv_open() is supposed to fail if the disk is read-only. Explicitly 47720eaf1bfSKevin Wolf * check for read-only block devices so that Linux block devices behave 47820eaf1bfSKevin Wolf * properly. 47920eaf1bfSKevin Wolf */ 48020eaf1bfSKevin Wolf struct stat st; 48120eaf1bfSKevin Wolf int readonly = 0; 48220eaf1bfSKevin Wolf 483bca5283bSKevin Wolf if (fstat(fd, &st)) { 48420eaf1bfSKevin Wolf return -errno; 48520eaf1bfSKevin Wolf } 48620eaf1bfSKevin Wolf 48720eaf1bfSKevin Wolf if (!S_ISBLK(st.st_mode)) { 48820eaf1bfSKevin Wolf return 0; 48920eaf1bfSKevin Wolf } 49020eaf1bfSKevin Wolf 491bca5283bSKevin Wolf if (ioctl(fd, BLKROGET, &readonly) < 0) { 49220eaf1bfSKevin Wolf return -errno; 49320eaf1bfSKevin Wolf } 49420eaf1bfSKevin Wolf 49520eaf1bfSKevin Wolf if (readonly) { 49620eaf1bfSKevin Wolf return -EACCES; 49720eaf1bfSKevin Wolf } 49820eaf1bfSKevin Wolf #endif /* defined(BLKROGET) */ 49920eaf1bfSKevin Wolf return 0; 50020eaf1bfSKevin Wolf } 50120eaf1bfSKevin Wolf 50223dece19SKevin Wolf static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) 503c1bb86cdSEric Blake { 50423dece19SKevin Wolf bool read_write = false; 505c1bb86cdSEric Blake assert(open_flags != NULL); 506c1bb86cdSEric Blake 507c1bb86cdSEric Blake *open_flags |= O_BINARY; 508c1bb86cdSEric Blake *open_flags &= ~O_ACCMODE; 50923dece19SKevin Wolf 51023dece19SKevin Wolf if (bdrv_flags & BDRV_O_AUTO_RDONLY) { 51123dece19SKevin Wolf read_write = has_writers; 51223dece19SKevin Wolf } else if (bdrv_flags & BDRV_O_RDWR) { 51323dece19SKevin Wolf read_write = true; 51423dece19SKevin Wolf } 51523dece19SKevin Wolf 51623dece19SKevin Wolf if (read_write) { 517c1bb86cdSEric Blake *open_flags |= O_RDWR; 518c1bb86cdSEric Blake } else { 519c1bb86cdSEric Blake *open_flags |= O_RDONLY; 520c1bb86cdSEric Blake } 521c1bb86cdSEric Blake 522c1bb86cdSEric Blake /* Use O_DSYNC for write-through caching, no flags for write-back caching, 523c1bb86cdSEric Blake * and O_DIRECT for no caching. */ 524c1bb86cdSEric Blake if ((bdrv_flags & BDRV_O_NOCACHE)) { 525c1bb86cdSEric Blake *open_flags |= O_DIRECT; 526c1bb86cdSEric Blake } 527c1bb86cdSEric Blake } 528c1bb86cdSEric Blake 529c1bb86cdSEric Blake static void raw_parse_filename(const char *filename, QDict *options, 530c1bb86cdSEric Blake Error **errp) 531c1bb86cdSEric Blake { 53203c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "file:", options); 533c1bb86cdSEric Blake } 534c1bb86cdSEric Blake 535c1bb86cdSEric Blake static QemuOptsList raw_runtime_opts = { 536c1bb86cdSEric Blake .name = "raw", 537c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), 538c1bb86cdSEric Blake .desc = { 539c1bb86cdSEric Blake { 540c1bb86cdSEric Blake .name = "filename", 541c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 542c1bb86cdSEric Blake .help = "File name of the image", 543c1bb86cdSEric Blake }, 544c1bb86cdSEric Blake { 545c1bb86cdSEric Blake .name = "aio", 546c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 547c6447510SAarushi Mehta .help = "host AIO implementation (threads, native, io_uring)", 548c1bb86cdSEric Blake }, 54916b48d5dSFam Zheng { 550684960d4SStefano Garzarella .name = "aio-max-batch", 551684960d4SStefano Garzarella .type = QEMU_OPT_NUMBER, 552684960d4SStefano Garzarella .help = "AIO max batch size (0 = auto handled by AIO backend, default: 0)", 553684960d4SStefano Garzarella }, 554684960d4SStefano Garzarella { 55516b48d5dSFam Zheng .name = "locking", 55616b48d5dSFam Zheng .type = QEMU_OPT_STRING, 55716b48d5dSFam Zheng .help = "file locking mode (on/off/auto, default: auto)", 55816b48d5dSFam Zheng }, 5597c9e5276SPaolo Bonzini { 5607c9e5276SPaolo Bonzini .name = "pr-manager", 5617c9e5276SPaolo Bonzini .type = QEMU_OPT_STRING, 5627c9e5276SPaolo Bonzini .help = "id of persistent reservation manager object (default: none)", 5637c9e5276SPaolo Bonzini }, 564f357fcd8SStefan Hajnoczi #if defined(__linux__) 565f357fcd8SStefan Hajnoczi { 566f357fcd8SStefan Hajnoczi .name = "drop-cache", 567f357fcd8SStefan Hajnoczi .type = QEMU_OPT_BOOL, 568f357fcd8SStefan Hajnoczi .help = "invalidate page cache during live migration (default: on)", 569f357fcd8SStefan Hajnoczi }, 570f357fcd8SStefan Hajnoczi #endif 57131be8a2aSStefan Hajnoczi { 57231be8a2aSStefan Hajnoczi .name = "x-check-cache-dropped", 57331be8a2aSStefan Hajnoczi .type = QEMU_OPT_BOOL, 57431be8a2aSStefan Hajnoczi .help = "check that page cache was dropped on live migration (default: off)" 57531be8a2aSStefan Hajnoczi }, 576c1bb86cdSEric Blake { /* end of list */ } 577c1bb86cdSEric Blake }, 578c1bb86cdSEric Blake }; 579c1bb86cdSEric Blake 5808a2ce0bcSAlberto Garcia static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; 5818a2ce0bcSAlberto Garcia 582c1bb86cdSEric Blake static int raw_open_common(BlockDriverState *bs, QDict *options, 583230ff739SJohn Snow int bdrv_flags, int open_flags, 584230ff739SJohn Snow bool device, Error **errp) 585c1bb86cdSEric Blake { 586c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 587c1bb86cdSEric Blake QemuOpts *opts; 588c1bb86cdSEric Blake Error *local_err = NULL; 589c1bb86cdSEric Blake const char *filename = NULL; 5907c9e5276SPaolo Bonzini const char *str; 591c1bb86cdSEric Blake BlockdevAioOptions aio, aio_default; 592c1bb86cdSEric Blake int fd, ret; 593c1bb86cdSEric Blake struct stat st; 594244a5668SFam Zheng OnOffAuto locking; 595c1bb86cdSEric Blake 596c1bb86cdSEric Blake opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 597af175e85SMarkus Armbruster if (!qemu_opts_absorb_qdict(opts, options, errp)) { 598c1bb86cdSEric Blake ret = -EINVAL; 599c1bb86cdSEric Blake goto fail; 600c1bb86cdSEric Blake } 601c1bb86cdSEric Blake 602c1bb86cdSEric Blake filename = qemu_opt_get(opts, "filename"); 603c1bb86cdSEric Blake 604db0754dfSFam Zheng ret = raw_normalize_devicepath(&filename, errp); 605c1bb86cdSEric Blake if (ret != 0) { 606c1bb86cdSEric Blake goto fail; 607c1bb86cdSEric Blake } 608c1bb86cdSEric Blake 609c6447510SAarushi Mehta if (bdrv_flags & BDRV_O_NATIVE_AIO) { 610c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE; 611c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 612c6447510SAarushi Mehta } else if (bdrv_flags & BDRV_O_IO_URING) { 613c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING; 614c6447510SAarushi Mehta #endif 615c6447510SAarushi Mehta } else { 616c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_THREADS; 617c6447510SAarushi Mehta } 618c6447510SAarushi Mehta 619f7abe0ecSMarc-André Lureau aio = qapi_enum_parse(&BlockdevAioOptions_lookup, 620f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "aio"), 62106c60b6cSMarkus Armbruster aio_default, &local_err); 622c1bb86cdSEric Blake if (local_err) { 623c1bb86cdSEric Blake error_propagate(errp, local_err); 624c1bb86cdSEric Blake ret = -EINVAL; 625c1bb86cdSEric Blake goto fail; 626c1bb86cdSEric Blake } 627c6447510SAarushi Mehta 628c1bb86cdSEric Blake s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); 629c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 630c6447510SAarushi Mehta s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING); 631c6447510SAarushi Mehta #endif 632c1bb86cdSEric Blake 633684960d4SStefano Garzarella s->aio_max_batch = qemu_opt_get_number(opts, "aio-max-batch", 0); 634684960d4SStefano Garzarella 635f7abe0ecSMarc-André Lureau locking = qapi_enum_parse(&OnOffAuto_lookup, 636f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "locking"), 63706c60b6cSMarkus Armbruster ON_OFF_AUTO_AUTO, &local_err); 638244a5668SFam Zheng if (local_err) { 639244a5668SFam Zheng error_propagate(errp, local_err); 640244a5668SFam Zheng ret = -EINVAL; 641244a5668SFam Zheng goto fail; 642244a5668SFam Zheng } 643244a5668SFam Zheng switch (locking) { 644244a5668SFam Zheng case ON_OFF_AUTO_ON: 645244a5668SFam Zheng s->use_lock = true; 6462b218f5dSFam Zheng if (!qemu_has_ofd_lock()) { 647db0754dfSFam Zheng warn_report("File lock requested but OFD locking syscall is " 648db0754dfSFam Zheng "unavailable, falling back to POSIX file locks"); 649db0754dfSFam Zheng error_printf("Due to the implementation, locks can be lost " 6502b218f5dSFam Zheng "unexpectedly.\n"); 6512b218f5dSFam Zheng } 652244a5668SFam Zheng break; 653244a5668SFam Zheng case ON_OFF_AUTO_OFF: 654244a5668SFam Zheng s->use_lock = false; 655244a5668SFam Zheng break; 656244a5668SFam Zheng case ON_OFF_AUTO_AUTO: 6572b218f5dSFam Zheng s->use_lock = qemu_has_ofd_lock(); 658244a5668SFam Zheng break; 659244a5668SFam Zheng default: 660244a5668SFam Zheng abort(); 661244a5668SFam Zheng } 662244a5668SFam Zheng 6637c9e5276SPaolo Bonzini str = qemu_opt_get(opts, "pr-manager"); 6647c9e5276SPaolo Bonzini if (str) { 6657c9e5276SPaolo Bonzini s->pr_mgr = pr_manager_lookup(str, &local_err); 6667c9e5276SPaolo Bonzini if (local_err) { 6677c9e5276SPaolo Bonzini error_propagate(errp, local_err); 6687c9e5276SPaolo Bonzini ret = -EINVAL; 6697c9e5276SPaolo Bonzini goto fail; 6707c9e5276SPaolo Bonzini } 6717c9e5276SPaolo Bonzini } 6727c9e5276SPaolo Bonzini 673f357fcd8SStefan Hajnoczi s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); 67431be8a2aSStefan Hajnoczi s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", 67531be8a2aSStefan Hajnoczi false); 67631be8a2aSStefan Hajnoczi 677c1bb86cdSEric Blake s->open_flags = open_flags; 67823dece19SKevin Wolf raw_parse_flags(bdrv_flags, &s->open_flags, false); 679c1bb86cdSEric Blake 680c1bb86cdSEric Blake s->fd = -1; 681b18a24a9SDaniel P. Berrangé fd = qemu_open(filename, s->open_flags, errp); 68264107dc0SKevin Wolf ret = fd < 0 ? -errno : 0; 68364107dc0SKevin Wolf 68464107dc0SKevin Wolf if (ret < 0) { 685c1bb86cdSEric Blake if (ret == -EROFS) { 686c1bb86cdSEric Blake ret = -EACCES; 687c1bb86cdSEric Blake } 688c1bb86cdSEric Blake goto fail; 689c1bb86cdSEric Blake } 690c1bb86cdSEric Blake s->fd = fd; 691c1bb86cdSEric Blake 692bca5283bSKevin Wolf /* Check s->open_flags rather than bdrv_flags due to auto-read-only */ 693bca5283bSKevin Wolf if (s->open_flags & O_RDWR) { 694bca5283bSKevin Wolf ret = check_hdev_writable(s->fd); 695bca5283bSKevin Wolf if (ret < 0) { 696bca5283bSKevin Wolf error_setg_errno(errp, -ret, "The device is not writable"); 697bca5283bSKevin Wolf goto fail; 698bca5283bSKevin Wolf } 699bca5283bSKevin Wolf } 700bca5283bSKevin Wolf 701244a5668SFam Zheng s->perm = 0; 702244a5668SFam Zheng s->shared_perm = BLK_PERM_ALL; 703244a5668SFam Zheng 704c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 705c1bb86cdSEric Blake /* Currently Linux does AIO only for files opened with O_DIRECT */ 706ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 707ed6e2161SNishanth Aravamudan if (!(s->open_flags & O_DIRECT)) { 708c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but it requires " 709c1bb86cdSEric Blake "cache.direct=on, which was not specified."); 710c1bb86cdSEric Blake ret = -EINVAL; 711c1bb86cdSEric Blake goto fail; 712c1bb86cdSEric Blake } 713ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { 714ed6e2161SNishanth Aravamudan error_prepend(errp, "Unable to use native AIO: "); 715ed6e2161SNishanth Aravamudan goto fail; 716ed6e2161SNishanth Aravamudan } 717ed6e2161SNishanth Aravamudan } 718c1bb86cdSEric Blake #else 719c1bb86cdSEric Blake if (s->use_linux_aio) { 720c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but is not supported " 721c1bb86cdSEric Blake "in this build."); 722c1bb86cdSEric Blake ret = -EINVAL; 723c1bb86cdSEric Blake goto fail; 724c1bb86cdSEric Blake } 725c1bb86cdSEric Blake #endif /* !defined(CONFIG_LINUX_AIO) */ 726c1bb86cdSEric Blake 727c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 728c6447510SAarushi Mehta if (s->use_linux_io_uring) { 729c6447510SAarushi Mehta if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs), errp)) { 730c6447510SAarushi Mehta error_prepend(errp, "Unable to use io_uring: "); 731c6447510SAarushi Mehta goto fail; 732c6447510SAarushi Mehta } 733c6447510SAarushi Mehta } 734c6447510SAarushi Mehta #else 735c6447510SAarushi Mehta if (s->use_linux_io_uring) { 736c6447510SAarushi Mehta error_setg(errp, "aio=io_uring was specified, but is not supported " 737c6447510SAarushi Mehta "in this build."); 738c6447510SAarushi Mehta ret = -EINVAL; 739c6447510SAarushi Mehta goto fail; 740c6447510SAarushi Mehta } 741c6447510SAarushi Mehta #endif /* !defined(CONFIG_LINUX_IO_URING) */ 742c6447510SAarushi Mehta 743c1bb86cdSEric Blake s->has_discard = true; 744c1bb86cdSEric Blake s->has_write_zeroes = true; 745c1bb86cdSEric Blake 746c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 747c1bb86cdSEric Blake ret = -errno; 748c1bb86cdSEric Blake error_setg_errno(errp, errno, "Could not stat file"); 749c1bb86cdSEric Blake goto fail; 750c1bb86cdSEric Blake } 751230ff739SJohn Snow 752230ff739SJohn Snow if (!device) { 7538d17adf3SDaniel P. Berrangé if (!S_ISREG(st.st_mode)) { 7548d17adf3SDaniel P. Berrangé error_setg(errp, "'%s' driver requires '%s' to be a regular file", 7558d17adf3SDaniel P. Berrangé bs->drv->format_name, bs->filename); 756230ff739SJohn Snow ret = -EINVAL; 757230ff739SJohn Snow goto fail; 758230ff739SJohn Snow } else { 759c1bb86cdSEric Blake s->discard_zeroes = true; 760c1bb86cdSEric Blake s->has_fallocate = true; 761c1bb86cdSEric Blake } 762230ff739SJohn Snow } else { 763230ff739SJohn Snow if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 7648d17adf3SDaniel P. Berrangé error_setg(errp, "'%s' driver requires '%s' to be either " 7658d17adf3SDaniel P. Berrangé "a character or block device", 7668d17adf3SDaniel P. Berrangé bs->drv->format_name, bs->filename); 767230ff739SJohn Snow ret = -EINVAL; 768230ff739SJohn Snow goto fail; 769230ff739SJohn Snow } 770230ff739SJohn Snow } 771230ff739SJohn Snow 772c1bb86cdSEric Blake if (S_ISBLK(st.st_mode)) { 773c1bb86cdSEric Blake #ifdef BLKDISCARDZEROES 774c1bb86cdSEric Blake unsigned int arg; 775c1bb86cdSEric Blake if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { 776c1bb86cdSEric Blake s->discard_zeroes = true; 777c1bb86cdSEric Blake } 778c1bb86cdSEric Blake #endif 779c1bb86cdSEric Blake #ifdef __linux__ 780c1bb86cdSEric Blake /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do 781c1bb86cdSEric Blake * not rely on the contents of discarded blocks unless using O_DIRECT. 782c1bb86cdSEric Blake * Same for BLKZEROOUT. 783c1bb86cdSEric Blake */ 784c1bb86cdSEric Blake if (!(bs->open_flags & BDRV_O_NOCACHE)) { 785c1bb86cdSEric Blake s->discard_zeroes = false; 786c1bb86cdSEric Blake s->has_write_zeroes = false; 787c1bb86cdSEric Blake } 788c1bb86cdSEric Blake #endif 789c1bb86cdSEric Blake } 790c1bb86cdSEric Blake #ifdef __FreeBSD__ 791c1bb86cdSEric Blake if (S_ISCHR(st.st_mode)) { 792c1bb86cdSEric Blake /* 793c1bb86cdSEric Blake * The file is a char device (disk), which on FreeBSD isn't behind 794c1bb86cdSEric Blake * a pager, so force all requests to be aligned. This is needed 795c1bb86cdSEric Blake * so QEMU makes sure all IO operations on the device are aligned 796c1bb86cdSEric Blake * to sector size, or else FreeBSD will reject them with EINVAL. 797c1bb86cdSEric Blake */ 7985dbd0ce1SKevin Wolf s->force_alignment = true; 799c1bb86cdSEric Blake } 800c1bb86cdSEric Blake #endif 8015dbd0ce1SKevin Wolf s->needs_alignment = raw_needs_alignment(bs); 802c1bb86cdSEric Blake 803738301e1SKevin Wolf bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; 8042f0c6e7aSKevin Wolf if (S_ISREG(st.st_mode)) { 8052f0c6e7aSKevin Wolf /* When extending regular files, we get zeros from the OS */ 8062f0c6e7aSKevin Wolf bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; 8072f0c6e7aSKevin Wolf } 808c1bb86cdSEric Blake ret = 0; 809c1bb86cdSEric Blake fail: 810a8c5cf27SKevin Wolf if (ret < 0 && s->fd != -1) { 811a8c5cf27SKevin Wolf qemu_close(s->fd); 812a8c5cf27SKevin Wolf } 813c1bb86cdSEric Blake if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { 814c1bb86cdSEric Blake unlink(filename); 815c1bb86cdSEric Blake } 816c1bb86cdSEric Blake qemu_opts_del(opts); 817c1bb86cdSEric Blake return ret; 818c1bb86cdSEric Blake } 819c1bb86cdSEric Blake 820c1bb86cdSEric Blake static int raw_open(BlockDriverState *bs, QDict *options, int flags, 821c1bb86cdSEric Blake Error **errp) 822c1bb86cdSEric Blake { 823c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 824c1bb86cdSEric Blake 825c1bb86cdSEric Blake s->type = FTYPE_FILE; 826230ff739SJohn Snow return raw_open_common(bs, options, flags, 0, false, errp); 827c1bb86cdSEric Blake } 828c1bb86cdSEric Blake 829244a5668SFam Zheng typedef enum { 830244a5668SFam Zheng RAW_PL_PREPARE, 831244a5668SFam Zheng RAW_PL_COMMIT, 832244a5668SFam Zheng RAW_PL_ABORT, 833244a5668SFam Zheng } RawPermLockOp; 834244a5668SFam Zheng 835244a5668SFam Zheng #define PERM_FOREACH(i) \ 836244a5668SFam Zheng for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) 837244a5668SFam Zheng 838244a5668SFam Zheng /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the 839244a5668SFam Zheng * file; if @unlock == true, also unlock the unneeded bytes. 840244a5668SFam Zheng * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. 841244a5668SFam Zheng */ 8422996ffadSFam Zheng static int raw_apply_lock_bytes(BDRVRawState *s, int fd, 843244a5668SFam Zheng uint64_t perm_lock_bits, 844244a5668SFam Zheng uint64_t shared_perm_lock_bits, 845244a5668SFam Zheng bool unlock, Error **errp) 846244a5668SFam Zheng { 847244a5668SFam Zheng int ret; 848244a5668SFam Zheng int i; 8492996ffadSFam Zheng uint64_t locked_perm, locked_shared_perm; 8502996ffadSFam Zheng 8512996ffadSFam Zheng if (s) { 8522996ffadSFam Zheng locked_perm = s->locked_perm; 8532996ffadSFam Zheng locked_shared_perm = s->locked_shared_perm; 8542996ffadSFam Zheng } else { 8552996ffadSFam Zheng /* 8562996ffadSFam Zheng * We don't have the previous bits, just lock/unlock for each of the 8572996ffadSFam Zheng * requested bits. 8582996ffadSFam Zheng */ 8592996ffadSFam Zheng if (unlock) { 8602996ffadSFam Zheng locked_perm = BLK_PERM_ALL; 8612996ffadSFam Zheng locked_shared_perm = BLK_PERM_ALL; 8622996ffadSFam Zheng } else { 8632996ffadSFam Zheng locked_perm = 0; 8642996ffadSFam Zheng locked_shared_perm = 0; 8652996ffadSFam Zheng } 8662996ffadSFam Zheng } 867244a5668SFam Zheng 868244a5668SFam Zheng PERM_FOREACH(i) { 869244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 8702996ffadSFam Zheng uint64_t bit = (1ULL << i); 8712996ffadSFam Zheng if ((perm_lock_bits & bit) && !(locked_perm & bit)) { 872d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 873244a5668SFam Zheng if (ret) { 874797e3e38SDavid Edmondson raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", 875797e3e38SDavid Edmondson off); 876244a5668SFam Zheng return ret; 8772996ffadSFam Zheng } else if (s) { 8782996ffadSFam Zheng s->locked_perm |= bit; 879244a5668SFam Zheng } 8802996ffadSFam Zheng } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { 881d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 882244a5668SFam Zheng if (ret) { 883797e3e38SDavid Edmondson error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); 884244a5668SFam Zheng return ret; 8852996ffadSFam Zheng } else if (s) { 8862996ffadSFam Zheng s->locked_perm &= ~bit; 887244a5668SFam Zheng } 888244a5668SFam Zheng } 889244a5668SFam Zheng } 890244a5668SFam Zheng PERM_FOREACH(i) { 891244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 8922996ffadSFam Zheng uint64_t bit = (1ULL << i); 8932996ffadSFam Zheng if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { 894d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 895244a5668SFam Zheng if (ret) { 896797e3e38SDavid Edmondson raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", 897797e3e38SDavid Edmondson off); 898244a5668SFam Zheng return ret; 8992996ffadSFam Zheng } else if (s) { 9002996ffadSFam Zheng s->locked_shared_perm |= bit; 901244a5668SFam Zheng } 9022996ffadSFam Zheng } else if (unlock && (locked_shared_perm & bit) && 9032996ffadSFam Zheng !(shared_perm_lock_bits & bit)) { 904d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 905244a5668SFam Zheng if (ret) { 906797e3e38SDavid Edmondson error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); 907244a5668SFam Zheng return ret; 9082996ffadSFam Zheng } else if (s) { 9092996ffadSFam Zheng s->locked_shared_perm &= ~bit; 910244a5668SFam Zheng } 911244a5668SFam Zheng } 912244a5668SFam Zheng } 913244a5668SFam Zheng return 0; 914244a5668SFam Zheng } 915244a5668SFam Zheng 916244a5668SFam Zheng /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ 917d0a96155SMax Reitz static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, 918244a5668SFam Zheng Error **errp) 919244a5668SFam Zheng { 920244a5668SFam Zheng int ret; 921244a5668SFam Zheng int i; 922244a5668SFam Zheng 923244a5668SFam Zheng PERM_FOREACH(i) { 924244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 925244a5668SFam Zheng uint64_t p = 1ULL << i; 926244a5668SFam Zheng if (perm & p) { 927d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 928244a5668SFam Zheng if (ret) { 929244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 930797e3e38SDavid Edmondson 931797e3e38SDavid Edmondson raw_lock_error_setg_errno(errp, -ret, 932244a5668SFam Zheng "Failed to get \"%s\" lock", 933244a5668SFam Zheng perm_name); 934244a5668SFam Zheng g_free(perm_name); 935244a5668SFam Zheng return ret; 936244a5668SFam Zheng } 937244a5668SFam Zheng } 938244a5668SFam Zheng } 939244a5668SFam Zheng PERM_FOREACH(i) { 940244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 941244a5668SFam Zheng uint64_t p = 1ULL << i; 942244a5668SFam Zheng if (!(shared_perm & p)) { 943d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 944244a5668SFam Zheng if (ret) { 945244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 946797e3e38SDavid Edmondson 947797e3e38SDavid Edmondson raw_lock_error_setg_errno(errp, -ret, 948244a5668SFam Zheng "Failed to get shared \"%s\" lock", 949244a5668SFam Zheng perm_name); 950244a5668SFam Zheng g_free(perm_name); 951244a5668SFam Zheng return ret; 952244a5668SFam Zheng } 953244a5668SFam Zheng } 954244a5668SFam Zheng } 955244a5668SFam Zheng return 0; 956244a5668SFam Zheng } 957244a5668SFam Zheng 958244a5668SFam Zheng static int raw_handle_perm_lock(BlockDriverState *bs, 959244a5668SFam Zheng RawPermLockOp op, 960244a5668SFam Zheng uint64_t new_perm, uint64_t new_shared, 961244a5668SFam Zheng Error **errp) 962244a5668SFam Zheng { 963244a5668SFam Zheng BDRVRawState *s = bs->opaque; 964244a5668SFam Zheng int ret = 0; 965244a5668SFam Zheng Error *local_err = NULL; 966244a5668SFam Zheng 967244a5668SFam Zheng if (!s->use_lock) { 968244a5668SFam Zheng return 0; 969244a5668SFam Zheng } 970244a5668SFam Zheng 971244a5668SFam Zheng if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { 972244a5668SFam Zheng return 0; 973244a5668SFam Zheng } 974244a5668SFam Zheng 975244a5668SFam Zheng switch (op) { 976244a5668SFam Zheng case RAW_PL_PREPARE: 977696aaaedSVladimir Sementsov-Ogievskiy if ((s->perm | new_perm) == s->perm && 978696aaaedSVladimir Sementsov-Ogievskiy (s->shared_perm & new_shared) == s->shared_perm) 979696aaaedSVladimir Sementsov-Ogievskiy { 980696aaaedSVladimir Sementsov-Ogievskiy /* 981696aaaedSVladimir Sementsov-Ogievskiy * We are going to unlock bytes, it should not fail. If it fail due 982696aaaedSVladimir Sementsov-Ogievskiy * to some fs-dependent permission-unrelated reasons (which occurs 983696aaaedSVladimir Sementsov-Ogievskiy * sometimes on NFS and leads to abort in bdrv_replace_child) we 984696aaaedSVladimir Sementsov-Ogievskiy * can't prevent such errors by any check here. And we ignore them 985696aaaedSVladimir Sementsov-Ogievskiy * anyway in ABORT and COMMIT. 986696aaaedSVladimir Sementsov-Ogievskiy */ 987696aaaedSVladimir Sementsov-Ogievskiy return 0; 988696aaaedSVladimir Sementsov-Ogievskiy } 989f2e3af29SFam Zheng ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, 990244a5668SFam Zheng ~s->shared_perm | ~new_shared, 991244a5668SFam Zheng false, errp); 992244a5668SFam Zheng if (!ret) { 993f2e3af29SFam Zheng ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); 994244a5668SFam Zheng if (!ret) { 995244a5668SFam Zheng return 0; 996244a5668SFam Zheng } 997b857431dSFam Zheng error_append_hint(errp, 998b857431dSFam Zheng "Is another process using the image [%s]?\n", 999b857431dSFam Zheng bs->filename); 1000244a5668SFam Zheng } 1001244a5668SFam Zheng /* fall through to unlock bytes. */ 1002244a5668SFam Zheng case RAW_PL_ABORT: 1003f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, 1004d0a96155SMax Reitz true, &local_err); 1005244a5668SFam Zheng if (local_err) { 1006244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 1007244a5668SFam Zheng * fail. Something weird happened, report it. 1008244a5668SFam Zheng */ 1009db0754dfSFam Zheng warn_report_err(local_err); 1010244a5668SFam Zheng } 1011244a5668SFam Zheng break; 1012244a5668SFam Zheng case RAW_PL_COMMIT: 1013f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, 1014d0a96155SMax Reitz true, &local_err); 1015244a5668SFam Zheng if (local_err) { 1016244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 1017244a5668SFam Zheng * fail. Something weird happened, report it. 1018244a5668SFam Zheng */ 1019db0754dfSFam Zheng warn_report_err(local_err); 1020244a5668SFam Zheng } 1021244a5668SFam Zheng break; 1022244a5668SFam Zheng } 1023244a5668SFam Zheng return ret; 1024244a5668SFam Zheng } 1025244a5668SFam Zheng 10265cec2870SKevin Wolf static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, 102723dece19SKevin Wolf int *open_flags, uint64_t perm, bool force_dup, 10286ceabe6fSKevin Wolf Error **errp) 10295cec2870SKevin Wolf { 10305cec2870SKevin Wolf BDRVRawState *s = bs->opaque; 10315cec2870SKevin Wolf int fd = -1; 10325cec2870SKevin Wolf int ret; 103323dece19SKevin Wolf bool has_writers = perm & 103423dece19SKevin Wolf (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); 10355cec2870SKevin Wolf int fcntl_flags = O_APPEND | O_NONBLOCK; 10365cec2870SKevin Wolf #ifdef O_NOATIME 10375cec2870SKevin Wolf fcntl_flags |= O_NOATIME; 10385cec2870SKevin Wolf #endif 10395cec2870SKevin Wolf 10405cec2870SKevin Wolf *open_flags = 0; 10415cec2870SKevin Wolf if (s->type == FTYPE_CD) { 10425cec2870SKevin Wolf *open_flags |= O_NONBLOCK; 10435cec2870SKevin Wolf } 10445cec2870SKevin Wolf 104523dece19SKevin Wolf raw_parse_flags(flags, open_flags, has_writers); 10465cec2870SKevin Wolf 10475cec2870SKevin Wolf #ifdef O_ASYNC 10485cec2870SKevin Wolf /* Not all operating systems have O_ASYNC, and those that don't 10495cec2870SKevin Wolf * will not let us track the state into rs->open_flags (typically 10505cec2870SKevin Wolf * you achieve the same effect with an ioctl, for example I_SETSIG 10515cec2870SKevin Wolf * on Solaris). But we do not use O_ASYNC, so that's fine. 10525cec2870SKevin Wolf */ 10535cec2870SKevin Wolf assert((s->open_flags & O_ASYNC) == 0); 10545cec2870SKevin Wolf #endif 10555cec2870SKevin Wolf 10566ceabe6fSKevin Wolf if (!force_dup && *open_flags == s->open_flags) { 10576ceabe6fSKevin Wolf /* We're lucky, the existing fd is fine */ 10586ceabe6fSKevin Wolf return s->fd; 10596ceabe6fSKevin Wolf } 10606ceabe6fSKevin Wolf 10615cec2870SKevin Wolf if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { 10625cec2870SKevin Wolf /* dup the original fd */ 10635cec2870SKevin Wolf fd = qemu_dup(s->fd); 10645cec2870SKevin Wolf if (fd >= 0) { 10655cec2870SKevin Wolf ret = fcntl_setfl(fd, *open_flags); 10665cec2870SKevin Wolf if (ret) { 10675cec2870SKevin Wolf qemu_close(fd); 10685cec2870SKevin Wolf fd = -1; 10695cec2870SKevin Wolf } 10705cec2870SKevin Wolf } 10715cec2870SKevin Wolf } 10725cec2870SKevin Wolf 1073b18a24a9SDaniel P. Berrangé /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ 10745cec2870SKevin Wolf if (fd == -1) { 10755cec2870SKevin Wolf const char *normalized_filename = bs->filename; 10765cec2870SKevin Wolf ret = raw_normalize_devicepath(&normalized_filename, errp); 10775cec2870SKevin Wolf if (ret >= 0) { 1078b18a24a9SDaniel P. Berrangé fd = qemu_open(normalized_filename, *open_flags, errp); 10795cec2870SKevin Wolf if (fd == -1) { 10805cec2870SKevin Wolf return -1; 10815cec2870SKevin Wolf } 10825cec2870SKevin Wolf } 10835cec2870SKevin Wolf } 10845cec2870SKevin Wolf 1085bca5283bSKevin Wolf if (fd != -1 && (*open_flags & O_RDWR)) { 1086bca5283bSKevin Wolf ret = check_hdev_writable(fd); 1087bca5283bSKevin Wolf if (ret < 0) { 1088bca5283bSKevin Wolf qemu_close(fd); 1089bca5283bSKevin Wolf error_setg_errno(errp, -ret, "The device is not writable"); 1090bca5283bSKevin Wolf return -1; 1091bca5283bSKevin Wolf } 1092bca5283bSKevin Wolf } 1093bca5283bSKevin Wolf 10945cec2870SKevin Wolf return fd; 10955cec2870SKevin Wolf } 10965cec2870SKevin Wolf 1097c1bb86cdSEric Blake static int raw_reopen_prepare(BDRVReopenState *state, 1098c1bb86cdSEric Blake BlockReopenQueue *queue, Error **errp) 1099c1bb86cdSEric Blake { 1100c1bb86cdSEric Blake BDRVRawState *s; 1101c1bb86cdSEric Blake BDRVRawReopenState *rs; 110231be8a2aSStefan Hajnoczi QemuOpts *opts; 1103a6aeca0cSKevin Wolf int ret; 1104c1bb86cdSEric Blake 1105c1bb86cdSEric Blake assert(state != NULL); 1106c1bb86cdSEric Blake assert(state->bs != NULL); 1107c1bb86cdSEric Blake 1108c1bb86cdSEric Blake s = state->bs->opaque; 1109c1bb86cdSEric Blake 1110c1bb86cdSEric Blake state->opaque = g_new0(BDRVRawReopenState, 1); 1111c1bb86cdSEric Blake rs = state->opaque; 111231be8a2aSStefan Hajnoczi 111331be8a2aSStefan Hajnoczi /* Handle options changes */ 111431be8a2aSStefan Hajnoczi opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 1115af175e85SMarkus Armbruster if (!qemu_opts_absorb_qdict(opts, state->options, errp)) { 111631be8a2aSStefan Hajnoczi ret = -EINVAL; 111731be8a2aSStefan Hajnoczi goto out; 111831be8a2aSStefan Hajnoczi } 111931be8a2aSStefan Hajnoczi 1120f357fcd8SStefan Hajnoczi rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); 11218d324575SAlberto Garcia rs->check_cache_dropped = 11228d324575SAlberto Garcia qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); 11238d324575SAlberto Garcia 11248d324575SAlberto Garcia /* This driver's reopen function doesn't currently allow changing 11258d324575SAlberto Garcia * other options, so let's put them back in the original QDict and 11268d324575SAlberto Garcia * bdrv_reopen_prepare() will detect changes and complain. */ 11278d324575SAlberto Garcia qemu_opts_to_qdict(opts, state->options); 1128c1bb86cdSEric Blake 112972373e40SVladimir Sementsov-Ogievskiy /* 113072373e40SVladimir Sementsov-Ogievskiy * As part of reopen prepare we also want to create new fd by 113172373e40SVladimir Sementsov-Ogievskiy * raw_reconfigure_getfd(). But it wants updated "perm", when in 113272373e40SVladimir Sementsov-Ogievskiy * bdrv_reopen_multiple() .bdrv_reopen_prepare() callback called prior to 113372373e40SVladimir Sementsov-Ogievskiy * permission update. Happily, permission update is always a part (a seprate 113472373e40SVladimir Sementsov-Ogievskiy * stage) of bdrv_reopen_multiple() so we can rely on this fact and 113572373e40SVladimir Sementsov-Ogievskiy * reconfigure fd in raw_check_perm(). 113672373e40SVladimir Sementsov-Ogievskiy */ 1137c1bb86cdSEric Blake 1138e0c9cf3aSKevin Wolf s->reopen_state = state; 1139a6aeca0cSKevin Wolf ret = 0; 114072373e40SVladimir Sementsov-Ogievskiy 114131be8a2aSStefan Hajnoczi out: 114231be8a2aSStefan Hajnoczi qemu_opts_del(opts); 1143c1bb86cdSEric Blake return ret; 1144c1bb86cdSEric Blake } 1145c1bb86cdSEric Blake 1146c1bb86cdSEric Blake static void raw_reopen_commit(BDRVReopenState *state) 1147c1bb86cdSEric Blake { 1148c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1149c1bb86cdSEric Blake BDRVRawState *s = state->bs->opaque; 1150c1bb86cdSEric Blake 1151f357fcd8SStefan Hajnoczi s->drop_cache = rs->drop_cache; 115231be8a2aSStefan Hajnoczi s->check_cache_dropped = rs->check_cache_dropped; 1153c1bb86cdSEric Blake s->open_flags = rs->open_flags; 1154c1bb86cdSEric Blake g_free(state->opaque); 1155c1bb86cdSEric Blake state->opaque = NULL; 1156e0c9cf3aSKevin Wolf 1157e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1158e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1159c1bb86cdSEric Blake } 1160c1bb86cdSEric Blake 1161c1bb86cdSEric Blake 1162c1bb86cdSEric Blake static void raw_reopen_abort(BDRVReopenState *state) 1163c1bb86cdSEric Blake { 1164c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1165e0c9cf3aSKevin Wolf BDRVRawState *s = state->bs->opaque; 1166c1bb86cdSEric Blake 1167c1bb86cdSEric Blake /* nothing to do if NULL, we didn't get far enough */ 1168c1bb86cdSEric Blake if (rs == NULL) { 1169c1bb86cdSEric Blake return; 1170c1bb86cdSEric Blake } 1171c1bb86cdSEric Blake 1172c1bb86cdSEric Blake g_free(state->opaque); 1173c1bb86cdSEric Blake state->opaque = NULL; 1174e0c9cf3aSKevin Wolf 1175e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1176e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1177c1bb86cdSEric Blake } 1178c1bb86cdSEric Blake 117918473467SPaolo Bonzini static int hdev_get_max_hw_transfer(int fd, struct stat *st) 1180c1bb86cdSEric Blake { 1181c1bb86cdSEric Blake #ifdef BLKSECTGET 118218473467SPaolo Bonzini if (S_ISBLK(st->st_mode)) { 118318473467SPaolo Bonzini unsigned short max_sectors = 0; 118418473467SPaolo Bonzini if (ioctl(fd, BLKSECTGET, &max_sectors) == 0) { 118518473467SPaolo Bonzini return max_sectors * 512; 118618473467SPaolo Bonzini } 118718473467SPaolo Bonzini } else { 118848265250SEric Farman int max_bytes = 0; 1189867eccfeSMaxim Levitsky if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) { 119048265250SEric Farman return max_bytes; 1191c1bb86cdSEric Blake } 119218473467SPaolo Bonzini } 119318473467SPaolo Bonzini return -errno; 1194c1bb86cdSEric Blake #else 1195c1bb86cdSEric Blake return -ENOSYS; 1196c1bb86cdSEric Blake #endif 1197c1bb86cdSEric Blake } 1198c1bb86cdSEric Blake 119918473467SPaolo Bonzini static int hdev_get_max_segments(int fd, struct stat *st) 12009103f1ceSFam Zheng { 12019103f1ceSFam Zheng #ifdef CONFIG_LINUX 12029103f1ceSFam Zheng char buf[32]; 12039103f1ceSFam Zheng const char *end; 1204867eccfeSMaxim Levitsky char *sysfspath = NULL; 12059103f1ceSFam Zheng int ret; 1206867eccfeSMaxim Levitsky int sysfd = -1; 12079103f1ceSFam Zheng long max_segments; 1208867eccfeSMaxim Levitsky 120918473467SPaolo Bonzini if (S_ISCHR(st->st_mode)) { 12108ad5ab61SPaolo Bonzini if (ioctl(fd, SG_GET_SG_TABLESIZE, &ret) == 0) { 12118ad5ab61SPaolo Bonzini return ret; 12128ad5ab61SPaolo Bonzini } 12138ad5ab61SPaolo Bonzini return -ENOTSUP; 12148ad5ab61SPaolo Bonzini } 12158ad5ab61SPaolo Bonzini 121618473467SPaolo Bonzini if (!S_ISBLK(st->st_mode)) { 12178ad5ab61SPaolo Bonzini return -ENOTSUP; 12188ad5ab61SPaolo Bonzini } 12198ad5ab61SPaolo Bonzini 12209103f1ceSFam Zheng sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", 122118473467SPaolo Bonzini major(st->st_rdev), minor(st->st_rdev)); 1222867eccfeSMaxim Levitsky sysfd = open(sysfspath, O_RDONLY); 1223867eccfeSMaxim Levitsky if (sysfd == -1) { 12249103f1ceSFam Zheng ret = -errno; 12259103f1ceSFam Zheng goto out; 12269103f1ceSFam Zheng } 12279103f1ceSFam Zheng do { 1228867eccfeSMaxim Levitsky ret = read(sysfd, buf, sizeof(buf) - 1); 12299103f1ceSFam Zheng } while (ret == -1 && errno == EINTR); 12309103f1ceSFam Zheng if (ret < 0) { 12319103f1ceSFam Zheng ret = -errno; 12329103f1ceSFam Zheng goto out; 12339103f1ceSFam Zheng } else if (ret == 0) { 12349103f1ceSFam Zheng ret = -EIO; 12359103f1ceSFam Zheng goto out; 12369103f1ceSFam Zheng } 12379103f1ceSFam Zheng buf[ret] = 0; 12389103f1ceSFam Zheng /* The file is ended with '\n', pass 'end' to accept that. */ 12399103f1ceSFam Zheng ret = qemu_strtol(buf, &end, 10, &max_segments); 12409103f1ceSFam Zheng if (ret == 0 && end && *end == '\n') { 12419103f1ceSFam Zheng ret = max_segments; 12429103f1ceSFam Zheng } 12439103f1ceSFam Zheng 12449103f1ceSFam Zheng out: 1245867eccfeSMaxim Levitsky if (sysfd != -1) { 1246867eccfeSMaxim Levitsky close(sysfd); 1247fed414dfSFam Zheng } 12489103f1ceSFam Zheng g_free(sysfspath); 12499103f1ceSFam Zheng return ret; 12509103f1ceSFam Zheng #else 12519103f1ceSFam Zheng return -ENOTSUP; 12529103f1ceSFam Zheng #endif 12539103f1ceSFam Zheng } 12549103f1ceSFam Zheng 1255c1bb86cdSEric Blake static void raw_refresh_limits(BlockDriverState *bs, Error **errp) 1256c1bb86cdSEric Blake { 1257c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 125818473467SPaolo Bonzini struct stat st; 1259c1bb86cdSEric Blake 12605dbd0ce1SKevin Wolf s->needs_alignment = raw_needs_alignment(bs); 1261c1bb86cdSEric Blake raw_probe_alignment(bs, s->fd, errp); 12625dbd0ce1SKevin Wolf 1263c1bb86cdSEric Blake bs->bl.min_mem_alignment = s->buf_align; 1264038adc2fSWei Yang bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size); 126518473467SPaolo Bonzini 126618473467SPaolo Bonzini /* 126718473467SPaolo Bonzini * Maximum transfers are best effort, so it is okay to ignore any 126818473467SPaolo Bonzini * errors. That said, based on the man page errors in fstat would be 126918473467SPaolo Bonzini * very much unexpected; the only possible case seems to be ENOMEM. 127018473467SPaolo Bonzini */ 127118473467SPaolo Bonzini if (fstat(s->fd, &st)) { 127218473467SPaolo Bonzini return; 127318473467SPaolo Bonzini } 127418473467SPaolo Bonzini 12750dfc7af2SAkihiko Odaki #if defined(__APPLE__) && (__MACH__) 12760dfc7af2SAkihiko Odaki struct statfs buf; 12770dfc7af2SAkihiko Odaki 12780dfc7af2SAkihiko Odaki if (!fstatfs(s->fd, &buf)) { 12790dfc7af2SAkihiko Odaki bs->bl.opt_transfer = buf.f_iosize; 12800dfc7af2SAkihiko Odaki bs->bl.pdiscard_alignment = buf.f_bsize; 12810dfc7af2SAkihiko Odaki } 12820dfc7af2SAkihiko Odaki #endif 12830dfc7af2SAkihiko Odaki 128418473467SPaolo Bonzini if (bs->sg || S_ISBLK(st.st_mode)) { 128518473467SPaolo Bonzini int ret = hdev_get_max_hw_transfer(s->fd, &st); 128618473467SPaolo Bonzini 128718473467SPaolo Bonzini if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { 128818473467SPaolo Bonzini bs->bl.max_hw_transfer = ret; 128918473467SPaolo Bonzini } 129018473467SPaolo Bonzini 129118473467SPaolo Bonzini ret = hdev_get_max_segments(s->fd, &st); 129218473467SPaolo Bonzini if (ret > 0) { 1293cc071629SPaolo Bonzini bs->bl.max_hw_iov = ret; 129418473467SPaolo Bonzini } 129518473467SPaolo Bonzini } 1296c1bb86cdSEric Blake } 1297c1bb86cdSEric Blake 1298c1bb86cdSEric Blake static int check_for_dasd(int fd) 1299c1bb86cdSEric Blake { 1300c1bb86cdSEric Blake #ifdef BIODASDINFO2 1301c1bb86cdSEric Blake struct dasd_information2_t info = {0}; 1302c1bb86cdSEric Blake 1303c1bb86cdSEric Blake return ioctl(fd, BIODASDINFO2, &info); 1304c1bb86cdSEric Blake #else 1305c1bb86cdSEric Blake return -1; 1306c1bb86cdSEric Blake #endif 1307c1bb86cdSEric Blake } 1308c1bb86cdSEric Blake 1309c1bb86cdSEric Blake /** 1310c1bb86cdSEric Blake * Try to get @bs's logical and physical block size. 1311c1bb86cdSEric Blake * On success, store them in @bsz and return zero. 1312c1bb86cdSEric Blake * On failure, return negative errno. 1313c1bb86cdSEric Blake */ 1314c1bb86cdSEric Blake static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) 1315c1bb86cdSEric Blake { 1316c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1317c1bb86cdSEric Blake int ret; 1318c1bb86cdSEric Blake 1319c1bb86cdSEric Blake /* If DASD, get blocksizes */ 1320c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1321c1bb86cdSEric Blake return -ENOTSUP; 1322c1bb86cdSEric Blake } 1323c1bb86cdSEric Blake ret = probe_logical_blocksize(s->fd, &bsz->log); 1324c1bb86cdSEric Blake if (ret < 0) { 1325c1bb86cdSEric Blake return ret; 1326c1bb86cdSEric Blake } 1327c1bb86cdSEric Blake return probe_physical_blocksize(s->fd, &bsz->phys); 1328c1bb86cdSEric Blake } 1329c1bb86cdSEric Blake 1330c1bb86cdSEric Blake /** 1331c1bb86cdSEric Blake * Try to get @bs's geometry: cyls, heads, sectors. 1332c1bb86cdSEric Blake * On success, store them in @geo and return 0. 1333c1bb86cdSEric Blake * On failure return -errno. 1334c1bb86cdSEric Blake * (Allows block driver to assign default geometry values that guest sees) 1335c1bb86cdSEric Blake */ 1336c1bb86cdSEric Blake #ifdef __linux__ 1337c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1338c1bb86cdSEric Blake { 1339c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1340c1bb86cdSEric Blake struct hd_geometry ioctl_geo = {0}; 1341c1bb86cdSEric Blake 1342c1bb86cdSEric Blake /* If DASD, get its geometry */ 1343c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1344c1bb86cdSEric Blake return -ENOTSUP; 1345c1bb86cdSEric Blake } 1346c1bb86cdSEric Blake if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { 1347c1bb86cdSEric Blake return -errno; 1348c1bb86cdSEric Blake } 1349c1bb86cdSEric Blake /* HDIO_GETGEO may return success even though geo contains zeros 1350c1bb86cdSEric Blake (e.g. certain multipath setups) */ 1351c1bb86cdSEric Blake if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { 1352c1bb86cdSEric Blake return -ENOTSUP; 1353c1bb86cdSEric Blake } 1354c1bb86cdSEric Blake /* Do not return a geometry for partition */ 1355c1bb86cdSEric Blake if (ioctl_geo.start != 0) { 1356c1bb86cdSEric Blake return -ENOTSUP; 1357c1bb86cdSEric Blake } 1358c1bb86cdSEric Blake geo->heads = ioctl_geo.heads; 1359c1bb86cdSEric Blake geo->sectors = ioctl_geo.sectors; 1360c1bb86cdSEric Blake geo->cylinders = ioctl_geo.cylinders; 1361c1bb86cdSEric Blake 1362c1bb86cdSEric Blake return 0; 1363c1bb86cdSEric Blake } 1364c1bb86cdSEric Blake #else /* __linux__ */ 1365c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1366c1bb86cdSEric Blake { 1367c1bb86cdSEric Blake return -ENOTSUP; 1368c1bb86cdSEric Blake } 1369c1bb86cdSEric Blake #endif 1370c1bb86cdSEric Blake 137103425671SKevin Wolf #if defined(__linux__) 137203425671SKevin Wolf static int handle_aiocb_ioctl(void *opaque) 1373c1bb86cdSEric Blake { 137403425671SKevin Wolf RawPosixAIOData *aiocb = opaque; 1375c1bb86cdSEric Blake int ret; 1376c1bb86cdSEric Blake 1377bd80936aSPaolo Bonzini do { 1378d57c44d0SKevin Wolf ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); 1379bd80936aSPaolo Bonzini } while (ret == -1 && errno == EINTR); 1380c1bb86cdSEric Blake if (ret == -1) { 1381c1bb86cdSEric Blake return -errno; 1382c1bb86cdSEric Blake } 1383c1bb86cdSEric Blake 1384c1bb86cdSEric Blake return 0; 1385c1bb86cdSEric Blake } 138603425671SKevin Wolf #endif /* linux */ 1387c1bb86cdSEric Blake 138806dc9bd5SKevin Wolf static int handle_aiocb_flush(void *opaque) 1389c1bb86cdSEric Blake { 139006dc9bd5SKevin Wolf RawPosixAIOData *aiocb = opaque; 1391e5bcf967SKevin Wolf BDRVRawState *s = aiocb->bs->opaque; 1392c1bb86cdSEric Blake int ret; 1393c1bb86cdSEric Blake 1394e5bcf967SKevin Wolf if (s->page_cache_inconsistent) { 1395c7ddc882SDaniel P. Berrangé return -s->page_cache_inconsistent; 1396e5bcf967SKevin Wolf } 1397e5bcf967SKevin Wolf 1398c1bb86cdSEric Blake ret = qemu_fdatasync(aiocb->aio_fildes); 1399c1bb86cdSEric Blake if (ret == -1) { 140060ff2ae2SDaniel P. Berrangé trace_file_flush_fdatasync_failed(errno); 140160ff2ae2SDaniel P. Berrangé 1402e5bcf967SKevin Wolf /* There is no clear definition of the semantics of a failing fsync(), 1403e5bcf967SKevin Wolf * so we may have to assume the worst. The sad truth is that this 1404e5bcf967SKevin Wolf * assumption is correct for Linux. Some pages are now probably marked 1405e5bcf967SKevin Wolf * clean in the page cache even though they are inconsistent with the 1406e5bcf967SKevin Wolf * on-disk contents. The next fdatasync() call would succeed, but no 1407e5bcf967SKevin Wolf * further writeback attempt will be made. We can't get back to a state 1408e5bcf967SKevin Wolf * in which we know what is on disk (we would have to rewrite 1409e5bcf967SKevin Wolf * everything that was touched since the last fdatasync() at least), so 1410e5bcf967SKevin Wolf * make bdrv_flush() fail permanently. Given that the behaviour isn't 1411e5bcf967SKevin Wolf * really defined, I have little hope that other OSes are doing better. 1412e5bcf967SKevin Wolf * 1413e5bcf967SKevin Wolf * Obviously, this doesn't affect O_DIRECT, which bypasses the page 1414e5bcf967SKevin Wolf * cache. */ 1415e5bcf967SKevin Wolf if ((s->open_flags & O_DIRECT) == 0) { 1416c7ddc882SDaniel P. Berrangé s->page_cache_inconsistent = errno; 1417e5bcf967SKevin Wolf } 1418c1bb86cdSEric Blake return -errno; 1419c1bb86cdSEric Blake } 1420c1bb86cdSEric Blake return 0; 1421c1bb86cdSEric Blake } 1422c1bb86cdSEric Blake 1423c1bb86cdSEric Blake #ifdef CONFIG_PREADV 1424c1bb86cdSEric Blake 1425c1bb86cdSEric Blake static bool preadv_present = true; 1426c1bb86cdSEric Blake 1427c1bb86cdSEric Blake static ssize_t 1428c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1429c1bb86cdSEric Blake { 1430c1bb86cdSEric Blake return preadv(fd, iov, nr_iov, offset); 1431c1bb86cdSEric Blake } 1432c1bb86cdSEric Blake 1433c1bb86cdSEric Blake static ssize_t 1434c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1435c1bb86cdSEric Blake { 1436c1bb86cdSEric Blake return pwritev(fd, iov, nr_iov, offset); 1437c1bb86cdSEric Blake } 1438c1bb86cdSEric Blake 1439c1bb86cdSEric Blake #else 1440c1bb86cdSEric Blake 1441c1bb86cdSEric Blake static bool preadv_present = false; 1442c1bb86cdSEric Blake 1443c1bb86cdSEric Blake static ssize_t 1444c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1445c1bb86cdSEric Blake { 1446c1bb86cdSEric Blake return -ENOSYS; 1447c1bb86cdSEric Blake } 1448c1bb86cdSEric Blake 1449c1bb86cdSEric Blake static ssize_t 1450c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1451c1bb86cdSEric Blake { 1452c1bb86cdSEric Blake return -ENOSYS; 1453c1bb86cdSEric Blake } 1454c1bb86cdSEric Blake 1455c1bb86cdSEric Blake #endif 1456c1bb86cdSEric Blake 1457c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) 1458c1bb86cdSEric Blake { 1459c1bb86cdSEric Blake ssize_t len; 1460c1bb86cdSEric Blake 1461c1bb86cdSEric Blake do { 1462c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) 1463c1bb86cdSEric Blake len = qemu_pwritev(aiocb->aio_fildes, 1464d57c44d0SKevin Wolf aiocb->io.iov, 1465d57c44d0SKevin Wolf aiocb->io.niov, 1466c1bb86cdSEric Blake aiocb->aio_offset); 1467c1bb86cdSEric Blake else 1468c1bb86cdSEric Blake len = qemu_preadv(aiocb->aio_fildes, 1469d57c44d0SKevin Wolf aiocb->io.iov, 1470d57c44d0SKevin Wolf aiocb->io.niov, 1471c1bb86cdSEric Blake aiocb->aio_offset); 1472c1bb86cdSEric Blake } while (len == -1 && errno == EINTR); 1473c1bb86cdSEric Blake 1474c1bb86cdSEric Blake if (len == -1) { 1475c1bb86cdSEric Blake return -errno; 1476c1bb86cdSEric Blake } 1477c1bb86cdSEric Blake return len; 1478c1bb86cdSEric Blake } 1479c1bb86cdSEric Blake 1480c1bb86cdSEric Blake /* 1481c1bb86cdSEric Blake * Read/writes the data to/from a given linear buffer. 1482c1bb86cdSEric Blake * 1483c1bb86cdSEric Blake * Returns the number of bytes handles or -errno in case of an error. Short 1484c1bb86cdSEric Blake * reads are only returned if the end of the file is reached. 1485c1bb86cdSEric Blake */ 1486c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) 1487c1bb86cdSEric Blake { 1488c1bb86cdSEric Blake ssize_t offset = 0; 1489c1bb86cdSEric Blake ssize_t len; 1490c1bb86cdSEric Blake 1491c1bb86cdSEric Blake while (offset < aiocb->aio_nbytes) { 1492c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1493c1bb86cdSEric Blake len = pwrite(aiocb->aio_fildes, 1494c1bb86cdSEric Blake (const char *)buf + offset, 1495c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1496c1bb86cdSEric Blake aiocb->aio_offset + offset); 1497c1bb86cdSEric Blake } else { 1498c1bb86cdSEric Blake len = pread(aiocb->aio_fildes, 1499c1bb86cdSEric Blake buf + offset, 1500c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1501c1bb86cdSEric Blake aiocb->aio_offset + offset); 1502c1bb86cdSEric Blake } 1503c1bb86cdSEric Blake if (len == -1 && errno == EINTR) { 1504c1bb86cdSEric Blake continue; 1505c1bb86cdSEric Blake } else if (len == -1 && errno == EINVAL && 1506c1bb86cdSEric Blake (aiocb->bs->open_flags & BDRV_O_NOCACHE) && 1507c1bb86cdSEric Blake !(aiocb->aio_type & QEMU_AIO_WRITE) && 1508c1bb86cdSEric Blake offset > 0) { 1509c1bb86cdSEric Blake /* O_DIRECT pread() may fail with EINVAL when offset is unaligned 1510c1bb86cdSEric Blake * after a short read. Assume that O_DIRECT short reads only occur 1511c1bb86cdSEric Blake * at EOF. Therefore this is a short read, not an I/O error. 1512c1bb86cdSEric Blake */ 1513c1bb86cdSEric Blake break; 1514c1bb86cdSEric Blake } else if (len == -1) { 1515c1bb86cdSEric Blake offset = -errno; 1516c1bb86cdSEric Blake break; 1517c1bb86cdSEric Blake } else if (len == 0) { 1518c1bb86cdSEric Blake break; 1519c1bb86cdSEric Blake } 1520c1bb86cdSEric Blake offset += len; 1521c1bb86cdSEric Blake } 1522c1bb86cdSEric Blake 1523c1bb86cdSEric Blake return offset; 1524c1bb86cdSEric Blake } 1525c1bb86cdSEric Blake 1526999e6b69SKevin Wolf static int handle_aiocb_rw(void *opaque) 1527c1bb86cdSEric Blake { 1528999e6b69SKevin Wolf RawPosixAIOData *aiocb = opaque; 1529c1bb86cdSEric Blake ssize_t nbytes; 1530c1bb86cdSEric Blake char *buf; 1531c1bb86cdSEric Blake 1532c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { 1533c1bb86cdSEric Blake /* 1534c1bb86cdSEric Blake * If there is just a single buffer, and it is properly aligned 1535c1bb86cdSEric Blake * we can just use plain pread/pwrite without any problems. 1536c1bb86cdSEric Blake */ 1537d57c44d0SKevin Wolf if (aiocb->io.niov == 1) { 153854c7ca1bSKevin Wolf nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); 153954c7ca1bSKevin Wolf goto out; 1540c1bb86cdSEric Blake } 1541c1bb86cdSEric Blake /* 1542c1bb86cdSEric Blake * We have more than one iovec, and all are properly aligned. 1543c1bb86cdSEric Blake * 1544c1bb86cdSEric Blake * Try preadv/pwritev first and fall back to linearizing the 1545c1bb86cdSEric Blake * buffer if it's not supported. 1546c1bb86cdSEric Blake */ 1547c1bb86cdSEric Blake if (preadv_present) { 1548c1bb86cdSEric Blake nbytes = handle_aiocb_rw_vector(aiocb); 1549c1bb86cdSEric Blake if (nbytes == aiocb->aio_nbytes || 1550c1bb86cdSEric Blake (nbytes < 0 && nbytes != -ENOSYS)) { 155154c7ca1bSKevin Wolf goto out; 1552c1bb86cdSEric Blake } 1553c1bb86cdSEric Blake preadv_present = false; 1554c1bb86cdSEric Blake } 1555c1bb86cdSEric Blake 1556c1bb86cdSEric Blake /* 1557c1bb86cdSEric Blake * XXX(hch): short read/write. no easy way to handle the reminder 1558c1bb86cdSEric Blake * using these interfaces. For now retry using plain 1559c1bb86cdSEric Blake * pread/pwrite? 1560c1bb86cdSEric Blake */ 1561c1bb86cdSEric Blake } 1562c1bb86cdSEric Blake 1563c1bb86cdSEric Blake /* 1564c1bb86cdSEric Blake * Ok, we have to do it the hard way, copy all segments into 1565c1bb86cdSEric Blake * a single aligned buffer. 1566c1bb86cdSEric Blake */ 1567c1bb86cdSEric Blake buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); 1568c1bb86cdSEric Blake if (buf == NULL) { 156954c7ca1bSKevin Wolf nbytes = -ENOMEM; 157054c7ca1bSKevin Wolf goto out; 1571c1bb86cdSEric Blake } 1572c1bb86cdSEric Blake 1573c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1574c1bb86cdSEric Blake char *p = buf; 1575c1bb86cdSEric Blake int i; 1576c1bb86cdSEric Blake 1577d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov; ++i) { 1578d57c44d0SKevin Wolf memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); 1579d57c44d0SKevin Wolf p += aiocb->io.iov[i].iov_len; 1580c1bb86cdSEric Blake } 1581c1bb86cdSEric Blake assert(p - buf == aiocb->aio_nbytes); 1582c1bb86cdSEric Blake } 1583c1bb86cdSEric Blake 1584c1bb86cdSEric Blake nbytes = handle_aiocb_rw_linear(aiocb, buf); 1585c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { 1586c1bb86cdSEric Blake char *p = buf; 1587c1bb86cdSEric Blake size_t count = aiocb->aio_nbytes, copy; 1588c1bb86cdSEric Blake int i; 1589c1bb86cdSEric Blake 1590d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov && count; ++i) { 1591c1bb86cdSEric Blake copy = count; 1592d57c44d0SKevin Wolf if (copy > aiocb->io.iov[i].iov_len) { 1593d57c44d0SKevin Wolf copy = aiocb->io.iov[i].iov_len; 1594c1bb86cdSEric Blake } 1595d57c44d0SKevin Wolf memcpy(aiocb->io.iov[i].iov_base, p, copy); 1596c1bb86cdSEric Blake assert(count >= copy); 1597c1bb86cdSEric Blake p += copy; 1598c1bb86cdSEric Blake count -= copy; 1599c1bb86cdSEric Blake } 1600c1bb86cdSEric Blake assert(count == 0); 1601c1bb86cdSEric Blake } 1602c1bb86cdSEric Blake qemu_vfree(buf); 1603c1bb86cdSEric Blake 160454c7ca1bSKevin Wolf out: 160554c7ca1bSKevin Wolf if (nbytes == aiocb->aio_nbytes) { 160654c7ca1bSKevin Wolf return 0; 160754c7ca1bSKevin Wolf } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { 160854c7ca1bSKevin Wolf if (aiocb->aio_type & QEMU_AIO_WRITE) { 160954c7ca1bSKevin Wolf return -EINVAL; 161054c7ca1bSKevin Wolf } else { 161154c7ca1bSKevin Wolf iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, 161254c7ca1bSKevin Wolf 0, aiocb->aio_nbytes - nbytes); 161354c7ca1bSKevin Wolf return 0; 161454c7ca1bSKevin Wolf } 161554c7ca1bSKevin Wolf } else { 161654c7ca1bSKevin Wolf assert(nbytes < 0); 1617c1bb86cdSEric Blake return nbytes; 1618c1bb86cdSEric Blake } 161954c7ca1bSKevin Wolf } 1620c1bb86cdSEric Blake 16210dfc7af2SAkihiko Odaki #if defined(CONFIG_FALLOCATE) || defined(BLKZEROOUT) || defined(BLKDISCARD) 1622c1bb86cdSEric Blake static int translate_err(int err) 1623c1bb86cdSEric Blake { 1624c1bb86cdSEric Blake if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || 1625c1bb86cdSEric Blake err == -ENOTTY) { 1626c1bb86cdSEric Blake err = -ENOTSUP; 1627c1bb86cdSEric Blake } 1628c1bb86cdSEric Blake return err; 1629c1bb86cdSEric Blake } 16300dfc7af2SAkihiko Odaki #endif 1631c1bb86cdSEric Blake 1632c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 1633c1bb86cdSEric Blake static int do_fallocate(int fd, int mode, off_t offset, off_t len) 1634c1bb86cdSEric Blake { 1635c1bb86cdSEric Blake do { 1636c1bb86cdSEric Blake if (fallocate(fd, mode, offset, len) == 0) { 1637c1bb86cdSEric Blake return 0; 1638c1bb86cdSEric Blake } 1639c1bb86cdSEric Blake } while (errno == EINTR); 1640c1bb86cdSEric Blake return translate_err(-errno); 1641c1bb86cdSEric Blake } 1642c1bb86cdSEric Blake #endif 1643c1bb86cdSEric Blake 1644c1bb86cdSEric Blake static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) 1645c1bb86cdSEric Blake { 1646c1bb86cdSEric Blake int ret = -ENOTSUP; 1647c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1648c1bb86cdSEric Blake 1649c1bb86cdSEric Blake if (!s->has_write_zeroes) { 1650c1bb86cdSEric Blake return -ENOTSUP; 1651c1bb86cdSEric Blake } 1652c1bb86cdSEric Blake 1653c1bb86cdSEric Blake #ifdef BLKZEROOUT 1654738301e1SKevin Wolf /* The BLKZEROOUT implementation in the kernel doesn't set 1655738301e1SKevin Wolf * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow 1656738301e1SKevin Wolf * fallbacks. */ 1657738301e1SKevin Wolf if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { 1658c1bb86cdSEric Blake do { 1659c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1660c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { 1661c1bb86cdSEric Blake return 0; 1662c1bb86cdSEric Blake } 1663c1bb86cdSEric Blake } while (errno == EINTR); 1664c1bb86cdSEric Blake 1665c1bb86cdSEric Blake ret = translate_err(-errno); 1666c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1667c1bb86cdSEric Blake s->has_write_zeroes = false; 1668c1bb86cdSEric Blake } 1669effecce6SKevin Wolf } 1670effecce6SKevin Wolf #endif 1671effecce6SKevin Wolf 1672c1bb86cdSEric Blake return ret; 1673c1bb86cdSEric Blake } 1674c1bb86cdSEric Blake 16757154d8aeSKevin Wolf static int handle_aiocb_write_zeroes(void *opaque) 1676c1bb86cdSEric Blake { 16777154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 167870d9110bSDenis V. Lunev #ifdef CONFIG_FALLOCATE 1679b2c6f23fSMax Reitz BDRVRawState *s = aiocb->bs->opaque; 168070d9110bSDenis V. Lunev int64_t len; 168170d9110bSDenis V. Lunev #endif 1682c1bb86cdSEric Blake 1683c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1684c1bb86cdSEric Blake return handle_aiocb_write_zeroes_block(aiocb); 1685c1bb86cdSEric Blake } 1686c1bb86cdSEric Blake 1687c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_ZERO_RANGE 1688c1bb86cdSEric Blake if (s->has_write_zeroes) { 1689c1bb86cdSEric Blake int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, 1690c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1691fa95e9fbSThomas Huth if (ret == -ENOTSUP) { 1692fa95e9fbSThomas Huth s->has_write_zeroes = false; 1693fa95e9fbSThomas Huth } else if (ret == 0 || ret != -EINVAL) { 1694c1bb86cdSEric Blake return ret; 1695c1bb86cdSEric Blake } 1696fa95e9fbSThomas Huth /* 1697fa95e9fbSThomas Huth * Note: Some file systems do not like unaligned byte ranges, and 1698fa95e9fbSThomas Huth * return EINVAL in such a case, though they should not do it according 1699fa95e9fbSThomas Huth * to the man-page of fallocate(). Thus we simply ignore this return 1700fa95e9fbSThomas Huth * value and try the other fallbacks instead. 1701fa95e9fbSThomas Huth */ 1702c1bb86cdSEric Blake } 1703c1bb86cdSEric Blake #endif 1704c1bb86cdSEric Blake 1705c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1706c1bb86cdSEric Blake if (s->has_discard && s->has_fallocate) { 1707c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 1708c1bb86cdSEric Blake FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1709c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1710c1bb86cdSEric Blake if (ret == 0) { 1711c1bb86cdSEric Blake ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1712c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1713c1bb86cdSEric Blake return ret; 1714c1bb86cdSEric Blake } 1715c1bb86cdSEric Blake s->has_fallocate = false; 171673ebf297SThomas Huth } else if (ret == -EINVAL) { 171773ebf297SThomas Huth /* 171873ebf297SThomas Huth * Some file systems like older versions of GPFS do not like un- 171973ebf297SThomas Huth * aligned byte ranges, and return EINVAL in such a case, though 172073ebf297SThomas Huth * they should not do it according to the man-page of fallocate(). 172173ebf297SThomas Huth * Warn about the bad filesystem and try the final fallback instead. 172273ebf297SThomas Huth */ 172373ebf297SThomas Huth warn_report_once("Your file system is misbehaving: " 172473ebf297SThomas Huth "fallocate(FALLOC_FL_PUNCH_HOLE) returned EINVAL. " 172568857f13SMichael Tokarev "Please report this bug to your file system " 172673ebf297SThomas Huth "vendor."); 1727c1bb86cdSEric Blake } else if (ret != -ENOTSUP) { 1728c1bb86cdSEric Blake return ret; 1729c1bb86cdSEric Blake } else { 1730c1bb86cdSEric Blake s->has_discard = false; 1731c1bb86cdSEric Blake } 1732c1bb86cdSEric Blake } 1733c1bb86cdSEric Blake #endif 1734c1bb86cdSEric Blake 1735c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 173670d9110bSDenis V. Lunev /* Last resort: we are trying to extend the file with zeroed data. This 173770d9110bSDenis V. Lunev * can be done via fallocate(fd, 0) */ 173870d9110bSDenis V. Lunev len = bdrv_getlength(aiocb->bs); 173970d9110bSDenis V. Lunev if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { 1740c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1741c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1742c1bb86cdSEric Blake return ret; 1743c1bb86cdSEric Blake } 1744c1bb86cdSEric Blake s->has_fallocate = false; 1745c1bb86cdSEric Blake } 1746c1bb86cdSEric Blake #endif 1747c1bb86cdSEric Blake 1748c1bb86cdSEric Blake return -ENOTSUP; 1749c1bb86cdSEric Blake } 1750c1bb86cdSEric Blake 17517154d8aeSKevin Wolf static int handle_aiocb_write_zeroes_unmap(void *opaque) 175234fa110eSKevin Wolf { 17537154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 175434fa110eSKevin Wolf BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; 175534fa110eSKevin Wolf 175634fa110eSKevin Wolf /* First try to write zeros and unmap at the same time */ 175734fa110eSKevin Wolf 175834fa110eSKevin Wolf #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1759b3ac2b94SSimran Singhal int ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 176034fa110eSKevin Wolf aiocb->aio_offset, aiocb->aio_nbytes); 1761bae127d4SAntoine Damhet switch (ret) { 1762bae127d4SAntoine Damhet case -ENOTSUP: 1763bae127d4SAntoine Damhet case -EINVAL: 1764ece4fa91SMaxim Levitsky case -EBUSY: 1765bae127d4SAntoine Damhet break; 1766bae127d4SAntoine Damhet default: 176734fa110eSKevin Wolf return ret; 176834fa110eSKevin Wolf } 176934fa110eSKevin Wolf #endif 177034fa110eSKevin Wolf 177134fa110eSKevin Wolf /* If we couldn't manage to unmap while guaranteed that the area reads as 177234fa110eSKevin Wolf * all-zero afterwards, just write zeroes without unmapping */ 1773b3ac2b94SSimran Singhal return handle_aiocb_write_zeroes(aiocb); 177434fa110eSKevin Wolf } 177534fa110eSKevin Wolf 17761efad060SFam Zheng #ifndef HAVE_COPY_FILE_RANGE 17771efad060SFam Zheng static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, 17781efad060SFam Zheng off_t *out_off, size_t len, unsigned int flags) 17791efad060SFam Zheng { 17801efad060SFam Zheng #ifdef __NR_copy_file_range 17811efad060SFam Zheng return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, 17821efad060SFam Zheng out_off, len, flags); 17831efad060SFam Zheng #else 17841efad060SFam Zheng errno = ENOSYS; 17851efad060SFam Zheng return -1; 17861efad060SFam Zheng #endif 17871efad060SFam Zheng } 17881efad060SFam Zheng #endif 17891efad060SFam Zheng 179058a209c4SKevin Wolf static int handle_aiocb_copy_range(void *opaque) 17911efad060SFam Zheng { 179258a209c4SKevin Wolf RawPosixAIOData *aiocb = opaque; 17931efad060SFam Zheng uint64_t bytes = aiocb->aio_nbytes; 17941efad060SFam Zheng off_t in_off = aiocb->aio_offset; 1795d57c44d0SKevin Wolf off_t out_off = aiocb->copy_range.aio_offset2; 17961efad060SFam Zheng 17971efad060SFam Zheng while (bytes) { 17981efad060SFam Zheng ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, 1799d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, &out_off, 18001efad060SFam Zheng bytes, 0); 1801ecc983a5SFam Zheng trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, 1802d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, out_off, bytes, 1803d57c44d0SKevin Wolf 0, ret); 1804c436e3d0SFam Zheng if (ret == 0) { 1805c436e3d0SFam Zheng /* No progress (e.g. when beyond EOF), let the caller fall back to 1806c436e3d0SFam Zheng * buffer I/O. */ 1807c436e3d0SFam Zheng return -ENOSPC; 18081efad060SFam Zheng } 18091efad060SFam Zheng if (ret < 0) { 1810c436e3d0SFam Zheng switch (errno) { 1811c436e3d0SFam Zheng case ENOSYS: 18121efad060SFam Zheng return -ENOTSUP; 1813c436e3d0SFam Zheng case EINTR: 1814c436e3d0SFam Zheng continue; 1815c436e3d0SFam Zheng default: 18161efad060SFam Zheng return -errno; 18171efad060SFam Zheng } 18181efad060SFam Zheng } 18191efad060SFam Zheng bytes -= ret; 18201efad060SFam Zheng } 18211efad060SFam Zheng return 0; 18221efad060SFam Zheng } 18231efad060SFam Zheng 182446ee0f46SKevin Wolf static int handle_aiocb_discard(void *opaque) 1825c1bb86cdSEric Blake { 182646ee0f46SKevin Wolf RawPosixAIOData *aiocb = opaque; 182713a02833SAri Sundholm int ret = -ENOTSUP; 1828c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1829c1bb86cdSEric Blake 1830c1bb86cdSEric Blake if (!s->has_discard) { 1831c1bb86cdSEric Blake return -ENOTSUP; 1832c1bb86cdSEric Blake } 1833c1bb86cdSEric Blake 1834c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1835c1bb86cdSEric Blake #ifdef BLKDISCARD 1836c1bb86cdSEric Blake do { 1837c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1838c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { 1839c1bb86cdSEric Blake return 0; 1840c1bb86cdSEric Blake } 1841c1bb86cdSEric Blake } while (errno == EINTR); 1842c1bb86cdSEric Blake 18430dfc7af2SAkihiko Odaki ret = translate_err(-errno); 1844c1bb86cdSEric Blake #endif 1845c1bb86cdSEric Blake } else { 1846c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1847c1bb86cdSEric Blake ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1848c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 184913a02833SAri Sundholm ret = translate_err(ret); 18500dfc7af2SAkihiko Odaki #elif defined(__APPLE__) && (__MACH__) 18510dfc7af2SAkihiko Odaki fpunchhole_t fpunchhole; 18520dfc7af2SAkihiko Odaki fpunchhole.fp_flags = 0; 18530dfc7af2SAkihiko Odaki fpunchhole.reserved = 0; 18540dfc7af2SAkihiko Odaki fpunchhole.fp_offset = aiocb->aio_offset; 18550dfc7af2SAkihiko Odaki fpunchhole.fp_length = aiocb->aio_nbytes; 18560dfc7af2SAkihiko Odaki if (fcntl(s->fd, F_PUNCHHOLE, &fpunchhole) == -1) { 18570dfc7af2SAkihiko Odaki ret = errno == ENODEV ? -ENOTSUP : -errno; 18580dfc7af2SAkihiko Odaki } else { 18590dfc7af2SAkihiko Odaki ret = 0; 18600dfc7af2SAkihiko Odaki } 1861c1bb86cdSEric Blake #endif 1862c1bb86cdSEric Blake } 1863c1bb86cdSEric Blake 1864c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1865c1bb86cdSEric Blake s->has_discard = false; 1866c1bb86cdSEric Blake } 1867c1bb86cdSEric Blake return ret; 1868c1bb86cdSEric Blake } 1869c1bb86cdSEric Blake 18703a20013fSNir Soffer /* 18713a20013fSNir Soffer * Help alignment probing by allocating the first block. 18723a20013fSNir Soffer * 18733a20013fSNir Soffer * When reading with direct I/O from unallocated area on Gluster backed by XFS, 18743a20013fSNir Soffer * reading succeeds regardless of request length. In this case we fallback to 18753a20013fSNir Soffer * safe alignment which is not optimal. Allocating the first block avoids this 18763a20013fSNir Soffer * fallback. 18773a20013fSNir Soffer * 18783a20013fSNir Soffer * fd may be opened with O_DIRECT, but we don't know the buffer alignment or 18793a20013fSNir Soffer * request alignment, so we use safe values. 18803a20013fSNir Soffer * 18813a20013fSNir Soffer * Returns: 0 on success, -errno on failure. Since this is an optimization, 18823a20013fSNir Soffer * caller may ignore failures. 18833a20013fSNir Soffer */ 18843a20013fSNir Soffer static int allocate_first_block(int fd, size_t max_size) 18853a20013fSNir Soffer { 18863a20013fSNir Soffer size_t write_size = (max_size < MAX_BLOCKSIZE) 18873a20013fSNir Soffer ? BDRV_SECTOR_SIZE 18883a20013fSNir Soffer : MAX_BLOCKSIZE; 1889038adc2fSWei Yang size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); 18903a20013fSNir Soffer void *buf; 18913a20013fSNir Soffer ssize_t n; 18923a20013fSNir Soffer int ret; 18933a20013fSNir Soffer 18943a20013fSNir Soffer buf = qemu_memalign(max_align, write_size); 18953a20013fSNir Soffer memset(buf, 0, write_size); 18963a20013fSNir Soffer 18973a20013fSNir Soffer do { 18983a20013fSNir Soffer n = pwrite(fd, buf, write_size, 0); 18993a20013fSNir Soffer } while (n == -1 && errno == EINTR); 19003a20013fSNir Soffer 19013a20013fSNir Soffer ret = (n == -1) ? -errno : 0; 19023a20013fSNir Soffer 19033a20013fSNir Soffer qemu_vfree(buf); 19043a20013fSNir Soffer return ret; 19053a20013fSNir Soffer } 19063a20013fSNir Soffer 190729cb4c01SKevin Wolf static int handle_aiocb_truncate(void *opaque) 190893f4e2ffSKevin Wolf { 190929cb4c01SKevin Wolf RawPosixAIOData *aiocb = opaque; 191093f4e2ffSKevin Wolf int result = 0; 191193f4e2ffSKevin Wolf int64_t current_length = 0; 191293f4e2ffSKevin Wolf char *buf = NULL; 191393f4e2ffSKevin Wolf struct stat st; 191493f4e2ffSKevin Wolf int fd = aiocb->aio_fildes; 191593f4e2ffSKevin Wolf int64_t offset = aiocb->aio_offset; 1916d57c44d0SKevin Wolf PreallocMode prealloc = aiocb->truncate.prealloc; 1917d57c44d0SKevin Wolf Error **errp = aiocb->truncate.errp; 191893f4e2ffSKevin Wolf 191993f4e2ffSKevin Wolf if (fstat(fd, &st) < 0) { 192093f4e2ffSKevin Wolf result = -errno; 192193f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not stat file"); 192293f4e2ffSKevin Wolf return result; 192393f4e2ffSKevin Wolf } 192493f4e2ffSKevin Wolf 192593f4e2ffSKevin Wolf current_length = st.st_size; 1926d57c44d0SKevin Wolf if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { 192793f4e2ffSKevin Wolf error_setg(errp, "Cannot use preallocation for shrinking files"); 192893f4e2ffSKevin Wolf return -ENOTSUP; 192993f4e2ffSKevin Wolf } 193093f4e2ffSKevin Wolf 1931d57c44d0SKevin Wolf switch (prealloc) { 193293f4e2ffSKevin Wolf #ifdef CONFIG_POSIX_FALLOCATE 193393f4e2ffSKevin Wolf case PREALLOC_MODE_FALLOC: 193493f4e2ffSKevin Wolf /* 193593f4e2ffSKevin Wolf * Truncating before posix_fallocate() makes it about twice slower on 193693f4e2ffSKevin Wolf * file systems that do not support fallocate(), trying to check if a 193793f4e2ffSKevin Wolf * block is allocated before allocating it, so don't do that here. 193893f4e2ffSKevin Wolf */ 193993f4e2ffSKevin Wolf if (offset != current_length) { 194093f4e2ffSKevin Wolf result = -posix_fallocate(fd, current_length, 194193f4e2ffSKevin Wolf offset - current_length); 194293f4e2ffSKevin Wolf if (result != 0) { 194393f4e2ffSKevin Wolf /* posix_fallocate() doesn't set errno. */ 194493f4e2ffSKevin Wolf error_setg_errno(errp, -result, 194593f4e2ffSKevin Wolf "Could not preallocate new data"); 19463a20013fSNir Soffer } else if (current_length == 0) { 19473a20013fSNir Soffer /* 19483a20013fSNir Soffer * posix_fallocate() uses fallocate() if the filesystem 19493a20013fSNir Soffer * supports it, or fallback to manually writing zeroes. If 19503a20013fSNir Soffer * fallocate() was used, unaligned reads from the fallocated 19513a20013fSNir Soffer * area in raw_probe_alignment() will succeed, hence we need to 19523a20013fSNir Soffer * allocate the first block. 19533a20013fSNir Soffer * 19543a20013fSNir Soffer * Optimize future alignment probing; ignore failures. 19553a20013fSNir Soffer */ 19563a20013fSNir Soffer allocate_first_block(fd, offset); 195793f4e2ffSKevin Wolf } 195893f4e2ffSKevin Wolf } else { 195993f4e2ffSKevin Wolf result = 0; 196093f4e2ffSKevin Wolf } 196193f4e2ffSKevin Wolf goto out; 196293f4e2ffSKevin Wolf #endif 196393f4e2ffSKevin Wolf case PREALLOC_MODE_FULL: 196493f4e2ffSKevin Wolf { 196593f4e2ffSKevin Wolf int64_t num = 0, left = offset - current_length; 196693f4e2ffSKevin Wolf off_t seek_result; 196793f4e2ffSKevin Wolf 196893f4e2ffSKevin Wolf /* 196993f4e2ffSKevin Wolf * Knowing the final size from the beginning could allow the file 197093f4e2ffSKevin Wolf * system driver to do less allocations and possibly avoid 197193f4e2ffSKevin Wolf * fragmentation of the file. 197293f4e2ffSKevin Wolf */ 197393f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 197493f4e2ffSKevin Wolf result = -errno; 197593f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 197693f4e2ffSKevin Wolf goto out; 197793f4e2ffSKevin Wolf } 197893f4e2ffSKevin Wolf 197993f4e2ffSKevin Wolf buf = g_malloc0(65536); 198093f4e2ffSKevin Wolf 198193f4e2ffSKevin Wolf seek_result = lseek(fd, current_length, SEEK_SET); 198293f4e2ffSKevin Wolf if (seek_result < 0) { 198393f4e2ffSKevin Wolf result = -errno; 198493f4e2ffSKevin Wolf error_setg_errno(errp, -result, 198593f4e2ffSKevin Wolf "Failed to seek to the old end of file"); 198693f4e2ffSKevin Wolf goto out; 198793f4e2ffSKevin Wolf } 198893f4e2ffSKevin Wolf 198993f4e2ffSKevin Wolf while (left > 0) { 199093f4e2ffSKevin Wolf num = MIN(left, 65536); 199193f4e2ffSKevin Wolf result = write(fd, buf, num); 199293f4e2ffSKevin Wolf if (result < 0) { 1993a1c81f4fSFam Zheng if (errno == EINTR) { 1994a1c81f4fSFam Zheng continue; 1995a1c81f4fSFam Zheng } 199693f4e2ffSKevin Wolf result = -errno; 199793f4e2ffSKevin Wolf error_setg_errno(errp, -result, 199893f4e2ffSKevin Wolf "Could not write zeros for preallocation"); 199993f4e2ffSKevin Wolf goto out; 200093f4e2ffSKevin Wolf } 200193f4e2ffSKevin Wolf left -= result; 200293f4e2ffSKevin Wolf } 200393f4e2ffSKevin Wolf if (result >= 0) { 200493f4e2ffSKevin Wolf result = fsync(fd); 200593f4e2ffSKevin Wolf if (result < 0) { 200693f4e2ffSKevin Wolf result = -errno; 200793f4e2ffSKevin Wolf error_setg_errno(errp, -result, 200893f4e2ffSKevin Wolf "Could not flush file to disk"); 200993f4e2ffSKevin Wolf goto out; 201093f4e2ffSKevin Wolf } 201193f4e2ffSKevin Wolf } 201293f4e2ffSKevin Wolf goto out; 201393f4e2ffSKevin Wolf } 201493f4e2ffSKevin Wolf case PREALLOC_MODE_OFF: 201593f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 201693f4e2ffSKevin Wolf result = -errno; 201793f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 20183a20013fSNir Soffer } else if (current_length == 0 && offset > current_length) { 20193a20013fSNir Soffer /* Optimize future alignment probing; ignore failures. */ 20203a20013fSNir Soffer allocate_first_block(fd, offset); 202193f4e2ffSKevin Wolf } 202293f4e2ffSKevin Wolf return result; 202393f4e2ffSKevin Wolf default: 202493f4e2ffSKevin Wolf result = -ENOTSUP; 202593f4e2ffSKevin Wolf error_setg(errp, "Unsupported preallocation mode: %s", 2026d57c44d0SKevin Wolf PreallocMode_str(prealloc)); 202793f4e2ffSKevin Wolf return result; 202893f4e2ffSKevin Wolf } 202993f4e2ffSKevin Wolf 203093f4e2ffSKevin Wolf out: 203193f4e2ffSKevin Wolf if (result < 0) { 203293f4e2ffSKevin Wolf if (ftruncate(fd, current_length) < 0) { 203393f4e2ffSKevin Wolf error_report("Failed to restore old file length: %s", 203493f4e2ffSKevin Wolf strerror(errno)); 203593f4e2ffSKevin Wolf } 203693f4e2ffSKevin Wolf } 203793f4e2ffSKevin Wolf 203893f4e2ffSKevin Wolf g_free(buf); 203993f4e2ffSKevin Wolf return result; 204093f4e2ffSKevin Wolf } 204193f4e2ffSKevin Wolf 20425d5de250SKevin Wolf static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs, 20435d5de250SKevin Wolf ThreadPoolFunc func, void *arg) 20445d5de250SKevin Wolf { 20455d5de250SKevin Wolf /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ 20465d5de250SKevin Wolf ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 20475d5de250SKevin Wolf return thread_pool_submit_co(pool, func, arg); 20485d5de250SKevin Wolf } 20495d5de250SKevin Wolf 2050c1bb86cdSEric Blake static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, 2051c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, int type) 2052c1bb86cdSEric Blake { 2053c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2054999e6b69SKevin Wolf RawPosixAIOData acb; 2055c1bb86cdSEric Blake 2056c1bb86cdSEric Blake if (fd_open(bs) < 0) 2057c1bb86cdSEric Blake return -EIO; 2058c1bb86cdSEric Blake 2059c1bb86cdSEric Blake /* 2060c6447510SAarushi Mehta * When using O_DIRECT, the request must be aligned to be able to use 2061c6447510SAarushi Mehta * either libaio or io_uring interface. If not fail back to regular thread 2062c6447510SAarushi Mehta * pool read/write code which emulates this for us if we 2063c6447510SAarushi Mehta * set QEMU_AIO_MISALIGNED. 2064c1bb86cdSEric Blake */ 2065c6447510SAarushi Mehta if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) { 2066c1bb86cdSEric Blake type |= QEMU_AIO_MISALIGNED; 2067c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2068c6447510SAarushi Mehta } else if (s->use_linux_io_uring) { 2069c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2070c6447510SAarushi Mehta assert(qiov->size == bytes); 2071c6447510SAarushi Mehta return luring_co_submit(bs, aio, s->fd, offset, qiov, type); 2072c6447510SAarushi Mehta #endif 2073c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 2074c1bb86cdSEric Blake } else if (s->use_linux_aio) { 2075c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 2076c1bb86cdSEric Blake assert(qiov->size == bytes); 2077512da211SStefano Garzarella return laio_co_submit(bs, aio, s->fd, offset, qiov, type, 2078512da211SStefano Garzarella s->aio_max_batch); 2079c1bb86cdSEric Blake #endif 2080c1bb86cdSEric Blake } 2081c1bb86cdSEric Blake 2082999e6b69SKevin Wolf acb = (RawPosixAIOData) { 2083999e6b69SKevin Wolf .bs = bs, 2084999e6b69SKevin Wolf .aio_fildes = s->fd, 2085999e6b69SKevin Wolf .aio_type = type, 2086999e6b69SKevin Wolf .aio_offset = offset, 2087999e6b69SKevin Wolf .aio_nbytes = bytes, 2088999e6b69SKevin Wolf .io = { 2089999e6b69SKevin Wolf .iov = qiov->iov, 2090999e6b69SKevin Wolf .niov = qiov->niov, 2091999e6b69SKevin Wolf }, 2092999e6b69SKevin Wolf }; 2093999e6b69SKevin Wolf 2094999e6b69SKevin Wolf assert(qiov->size == bytes); 2095999e6b69SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb); 2096c1bb86cdSEric Blake } 2097c1bb86cdSEric Blake 2098f7ef38ddSVladimir Sementsov-Ogievskiy static int coroutine_fn raw_co_preadv(BlockDriverState *bs, int64_t offset, 2099f7ef38ddSVladimir Sementsov-Ogievskiy int64_t bytes, QEMUIOVector *qiov, 2100f7ef38ddSVladimir Sementsov-Ogievskiy BdrvRequestFlags flags) 2101c1bb86cdSEric Blake { 2102c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); 2103c1bb86cdSEric Blake } 2104c1bb86cdSEric Blake 2105e75abedaSVladimir Sementsov-Ogievskiy static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, int64_t offset, 2106e75abedaSVladimir Sementsov-Ogievskiy int64_t bytes, QEMUIOVector *qiov, 2107e75abedaSVladimir Sementsov-Ogievskiy BdrvRequestFlags flags) 2108c1bb86cdSEric Blake { 2109c1bb86cdSEric Blake assert(flags == 0); 2110c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); 2111c1bb86cdSEric Blake } 2112c1bb86cdSEric Blake 2113c1bb86cdSEric Blake static void raw_aio_plug(BlockDriverState *bs) 2114c1bb86cdSEric Blake { 2115c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 2116c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 2117c1bb86cdSEric Blake if (s->use_linux_aio) { 2118c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 2119c1bb86cdSEric Blake laio_io_plug(bs, aio); 2120c1bb86cdSEric Blake } 2121c1bb86cdSEric Blake #endif 2122c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2123c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2124c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2125c6447510SAarushi Mehta luring_io_plug(bs, aio); 2126c6447510SAarushi Mehta } 2127c6447510SAarushi Mehta #endif 2128c1bb86cdSEric Blake } 2129c1bb86cdSEric Blake 2130c1bb86cdSEric Blake static void raw_aio_unplug(BlockDriverState *bs) 2131c1bb86cdSEric Blake { 2132c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 2133c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 2134c1bb86cdSEric Blake if (s->use_linux_aio) { 2135c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 213668d79466SStefano Garzarella laio_io_unplug(bs, aio, s->aio_max_batch); 2137c1bb86cdSEric Blake } 2138c1bb86cdSEric Blake #endif 2139c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2140c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2141c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2142c6447510SAarushi Mehta luring_io_unplug(bs, aio); 2143c6447510SAarushi Mehta } 2144c6447510SAarushi Mehta #endif 2145c1bb86cdSEric Blake } 2146c1bb86cdSEric Blake 214733d70fb6SKevin Wolf static int raw_co_flush_to_disk(BlockDriverState *bs) 2148c1bb86cdSEric Blake { 2149c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 215006dc9bd5SKevin Wolf RawPosixAIOData acb; 215133d70fb6SKevin Wolf int ret; 2152c1bb86cdSEric Blake 215333d70fb6SKevin Wolf ret = fd_open(bs); 215433d70fb6SKevin Wolf if (ret < 0) { 215533d70fb6SKevin Wolf return ret; 215633d70fb6SKevin Wolf } 2157c1bb86cdSEric Blake 215806dc9bd5SKevin Wolf acb = (RawPosixAIOData) { 215906dc9bd5SKevin Wolf .bs = bs, 216006dc9bd5SKevin Wolf .aio_fildes = s->fd, 216106dc9bd5SKevin Wolf .aio_type = QEMU_AIO_FLUSH, 216206dc9bd5SKevin Wolf }; 216306dc9bd5SKevin Wolf 2164c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2165c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2166c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2167c6447510SAarushi Mehta return luring_co_submit(bs, aio, s->fd, 0, NULL, QEMU_AIO_FLUSH); 2168c6447510SAarushi Mehta } 2169c6447510SAarushi Mehta #endif 217006dc9bd5SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb); 2171c1bb86cdSEric Blake } 2172c1bb86cdSEric Blake 2173ed6e2161SNishanth Aravamudan static void raw_aio_attach_aio_context(BlockDriverState *bs, 2174ed6e2161SNishanth Aravamudan AioContext *new_context) 2175ed6e2161SNishanth Aravamudan { 2176c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 2177ed6e2161SNishanth Aravamudan #ifdef CONFIG_LINUX_AIO 2178ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 2179cb09104eSMarkus Armbruster Error *local_err = NULL; 2180ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(new_context, &local_err)) { 2181ed6e2161SNishanth Aravamudan error_reportf_err(local_err, "Unable to use native AIO, " 2182ed6e2161SNishanth Aravamudan "falling back to thread pool: "); 2183ed6e2161SNishanth Aravamudan s->use_linux_aio = false; 2184ed6e2161SNishanth Aravamudan } 2185ed6e2161SNishanth Aravamudan } 2186ed6e2161SNishanth Aravamudan #endif 2187c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2188c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2189cb8d0851SPan Nengyuan Error *local_err = NULL; 2190c6447510SAarushi Mehta if (!aio_setup_linux_io_uring(new_context, &local_err)) { 2191c6447510SAarushi Mehta error_reportf_err(local_err, "Unable to use linux io_uring, " 2192c6447510SAarushi Mehta "falling back to thread pool: "); 2193c6447510SAarushi Mehta s->use_linux_io_uring = false; 2194c6447510SAarushi Mehta } 2195c6447510SAarushi Mehta } 2196c6447510SAarushi Mehta #endif 2197ed6e2161SNishanth Aravamudan } 2198ed6e2161SNishanth Aravamudan 2199c1bb86cdSEric Blake static void raw_close(BlockDriverState *bs) 2200c1bb86cdSEric Blake { 2201c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2202c1bb86cdSEric Blake 2203c1bb86cdSEric Blake if (s->fd >= 0) { 2204c1bb86cdSEric Blake qemu_close(s->fd); 2205c1bb86cdSEric Blake s->fd = -1; 2206c1bb86cdSEric Blake } 2207c1bb86cdSEric Blake } 2208c1bb86cdSEric Blake 2209d0bc9e5dSMax Reitz /** 2210d0bc9e5dSMax Reitz * Truncates the given regular file @fd to @offset and, when growing, fills the 2211d0bc9e5dSMax Reitz * new space according to @prealloc. 2212d0bc9e5dSMax Reitz * 2213d0bc9e5dSMax Reitz * Returns: 0 on success, -errno on failure. 2214d0bc9e5dSMax Reitz */ 221593f4e2ffSKevin Wolf static int coroutine_fn 221693f4e2ffSKevin Wolf raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, 221793f4e2ffSKevin Wolf PreallocMode prealloc, Error **errp) 22189f63b07eSMax Reitz { 221929cb4c01SKevin Wolf RawPosixAIOData acb; 2220d0bc9e5dSMax Reitz 222129cb4c01SKevin Wolf acb = (RawPosixAIOData) { 222293f4e2ffSKevin Wolf .bs = bs, 222393f4e2ffSKevin Wolf .aio_fildes = fd, 222493f4e2ffSKevin Wolf .aio_type = QEMU_AIO_TRUNCATE, 222593f4e2ffSKevin Wolf .aio_offset = offset, 2226d57c44d0SKevin Wolf .truncate = { 222793f4e2ffSKevin Wolf .prealloc = prealloc, 222893f4e2ffSKevin Wolf .errp = errp, 2229d57c44d0SKevin Wolf }, 223093f4e2ffSKevin Wolf }; 2231d0bc9e5dSMax Reitz 223229cb4c01SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb); 22339f63b07eSMax Reitz } 22349f63b07eSMax Reitz 2235061ca8a3SKevin Wolf static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, 2236c80d8b06SMax Reitz bool exact, PreallocMode prealloc, 223792b92799SKevin Wolf BdrvRequestFlags flags, Error **errp) 2238c1bb86cdSEric Blake { 2239c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2240c1bb86cdSEric Blake struct stat st; 2241f59adb32SMax Reitz int ret; 2242c1bb86cdSEric Blake 2243c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 2244f59adb32SMax Reitz ret = -errno; 2245f59adb32SMax Reitz error_setg_errno(errp, -ret, "Failed to fstat() the file"); 2246f59adb32SMax Reitz return ret; 2247c1bb86cdSEric Blake } 2248c1bb86cdSEric Blake 2249c1bb86cdSEric Blake if (S_ISREG(st.st_mode)) { 225082325ae5SMax Reitz /* Always resizes to the exact @offset */ 225193f4e2ffSKevin Wolf return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); 2252c1bb86cdSEric Blake } 225335d72602SMax Reitz 225435d72602SMax Reitz if (prealloc != PREALLOC_MODE_OFF) { 225535d72602SMax Reitz error_setg(errp, "Preallocation mode '%s' unsupported for this " 2256977c736fSMarkus Armbruster "non-regular file", PreallocMode_str(prealloc)); 225735d72602SMax Reitz return -ENOTSUP; 225835d72602SMax Reitz } 225935d72602SMax Reitz 226035d72602SMax Reitz if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 226182325ae5SMax Reitz int64_t cur_length = raw_getlength(bs); 226282325ae5SMax Reitz 226382325ae5SMax Reitz if (offset != cur_length && exact) { 226482325ae5SMax Reitz error_setg(errp, "Cannot resize device files"); 226582325ae5SMax Reitz return -ENOTSUP; 226682325ae5SMax Reitz } else if (offset > cur_length) { 2267f59adb32SMax Reitz error_setg(errp, "Cannot grow device files"); 2268c1bb86cdSEric Blake return -EINVAL; 2269c1bb86cdSEric Blake } 2270c1bb86cdSEric Blake } else { 2271f59adb32SMax Reitz error_setg(errp, "Resizing this file is not supported"); 2272c1bb86cdSEric Blake return -ENOTSUP; 2273c1bb86cdSEric Blake } 2274c1bb86cdSEric Blake 2275c1bb86cdSEric Blake return 0; 2276c1bb86cdSEric Blake } 2277c1bb86cdSEric Blake 2278c1bb86cdSEric Blake #ifdef __OpenBSD__ 2279c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2280c1bb86cdSEric Blake { 2281c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2282c1bb86cdSEric Blake int fd = s->fd; 2283c1bb86cdSEric Blake struct stat st; 2284c1bb86cdSEric Blake 2285c1bb86cdSEric Blake if (fstat(fd, &st)) 2286c1bb86cdSEric Blake return -errno; 2287c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2288c1bb86cdSEric Blake struct disklabel dl; 2289c1bb86cdSEric Blake 2290c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2291c1bb86cdSEric Blake return -errno; 2292c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2293c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2294c1bb86cdSEric Blake } else 2295c1bb86cdSEric Blake return st.st_size; 2296c1bb86cdSEric Blake } 2297c1bb86cdSEric Blake #elif defined(__NetBSD__) 2298c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2299c1bb86cdSEric Blake { 2300c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2301c1bb86cdSEric Blake int fd = s->fd; 2302c1bb86cdSEric Blake struct stat st; 2303c1bb86cdSEric Blake 2304c1bb86cdSEric Blake if (fstat(fd, &st)) 2305c1bb86cdSEric Blake return -errno; 2306c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2307c1bb86cdSEric Blake struct dkwedge_info dkw; 2308c1bb86cdSEric Blake 2309c1bb86cdSEric Blake if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { 2310c1bb86cdSEric Blake return dkw.dkw_size * 512; 2311c1bb86cdSEric Blake } else { 2312c1bb86cdSEric Blake struct disklabel dl; 2313c1bb86cdSEric Blake 2314c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2315c1bb86cdSEric Blake return -errno; 2316c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2317c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2318c1bb86cdSEric Blake } 2319c1bb86cdSEric Blake } else 2320c1bb86cdSEric Blake return st.st_size; 2321c1bb86cdSEric Blake } 2322c1bb86cdSEric Blake #elif defined(__sun__) 2323c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2324c1bb86cdSEric Blake { 2325c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2326c1bb86cdSEric Blake struct dk_minfo minfo; 2327c1bb86cdSEric Blake int ret; 2328c1bb86cdSEric Blake int64_t size; 2329c1bb86cdSEric Blake 2330c1bb86cdSEric Blake ret = fd_open(bs); 2331c1bb86cdSEric Blake if (ret < 0) { 2332c1bb86cdSEric Blake return ret; 2333c1bb86cdSEric Blake } 2334c1bb86cdSEric Blake 2335c1bb86cdSEric Blake /* 2336c1bb86cdSEric Blake * Use the DKIOCGMEDIAINFO ioctl to read the size. 2337c1bb86cdSEric Blake */ 2338c1bb86cdSEric Blake ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); 2339c1bb86cdSEric Blake if (ret != -1) { 2340c1bb86cdSEric Blake return minfo.dki_lbsize * minfo.dki_capacity; 2341c1bb86cdSEric Blake } 2342c1bb86cdSEric Blake 2343c1bb86cdSEric Blake /* 2344c1bb86cdSEric Blake * There are reports that lseek on some devices fails, but 2345c1bb86cdSEric Blake * irc discussion said that contingency on contingency was overkill. 2346c1bb86cdSEric Blake */ 2347c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2348c1bb86cdSEric Blake if (size < 0) { 2349c1bb86cdSEric Blake return -errno; 2350c1bb86cdSEric Blake } 2351c1bb86cdSEric Blake return size; 2352c1bb86cdSEric Blake } 2353c1bb86cdSEric Blake #elif defined(CONFIG_BSD) 2354c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2355c1bb86cdSEric Blake { 2356c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2357c1bb86cdSEric Blake int fd = s->fd; 2358c1bb86cdSEric Blake int64_t size; 2359c1bb86cdSEric Blake struct stat sb; 2360c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2361c1bb86cdSEric Blake int reopened = 0; 2362c1bb86cdSEric Blake #endif 2363c1bb86cdSEric Blake int ret; 2364c1bb86cdSEric Blake 2365c1bb86cdSEric Blake ret = fd_open(bs); 2366c1bb86cdSEric Blake if (ret < 0) 2367c1bb86cdSEric Blake return ret; 2368c1bb86cdSEric Blake 2369c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2370c1bb86cdSEric Blake again: 2371c1bb86cdSEric Blake #endif 2372c1bb86cdSEric Blake if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { 2373267cd53fSPaolo Bonzini size = 0; 2374c1bb86cdSEric Blake #ifdef DIOCGMEDIASIZE 2375267cd53fSPaolo Bonzini if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) { 2376c1bb86cdSEric Blake size = 0; 2377c1bb86cdSEric Blake } 2378267cd53fSPaolo Bonzini #endif 2379267cd53fSPaolo Bonzini #ifdef DIOCGPART 2380267cd53fSPaolo Bonzini if (size == 0) { 2381267cd53fSPaolo Bonzini struct partinfo pi; 2382267cd53fSPaolo Bonzini if (ioctl(fd, DIOCGPART, &pi) == 0) { 2383267cd53fSPaolo Bonzini size = pi.media_size; 2384267cd53fSPaolo Bonzini } 2385267cd53fSPaolo Bonzini } 2386c1bb86cdSEric Blake #endif 238709e20abdSJoelle van Dyne #if defined(DKIOCGETBLOCKCOUNT) && defined(DKIOCGETBLOCKSIZE) 2388267cd53fSPaolo Bonzini if (size == 0) { 2389c1bb86cdSEric Blake uint64_t sectors = 0; 2390c1bb86cdSEric Blake uint32_t sector_size = 0; 2391c1bb86cdSEric Blake 2392c1bb86cdSEric Blake if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 2393c1bb86cdSEric Blake && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { 2394c1bb86cdSEric Blake size = sectors * sector_size; 2395c1bb86cdSEric Blake } 2396c1bb86cdSEric Blake } 2397c1bb86cdSEric Blake #endif 2398267cd53fSPaolo Bonzini if (size == 0) { 2399267cd53fSPaolo Bonzini size = lseek(fd, 0LL, SEEK_END); 2400267cd53fSPaolo Bonzini } 2401267cd53fSPaolo Bonzini if (size < 0) { 2402267cd53fSPaolo Bonzini return -errno; 2403267cd53fSPaolo Bonzini } 2404c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 2405c1bb86cdSEric Blake switch(s->type) { 2406c1bb86cdSEric Blake case FTYPE_CD: 2407c1bb86cdSEric Blake /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ 2408c1bb86cdSEric Blake if (size == 2048LL * (unsigned)-1) 2409c1bb86cdSEric Blake size = 0; 2410c1bb86cdSEric Blake /* XXX no disc? maybe we need to reopen... */ 2411c1bb86cdSEric Blake if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { 2412c1bb86cdSEric Blake reopened = 1; 2413c1bb86cdSEric Blake goto again; 2414c1bb86cdSEric Blake } 2415c1bb86cdSEric Blake } 2416c1bb86cdSEric Blake #endif 2417c1bb86cdSEric Blake } else { 2418c1bb86cdSEric Blake size = lseek(fd, 0, SEEK_END); 2419c1bb86cdSEric Blake if (size < 0) { 2420c1bb86cdSEric Blake return -errno; 2421c1bb86cdSEric Blake } 2422c1bb86cdSEric Blake } 2423c1bb86cdSEric Blake return size; 2424c1bb86cdSEric Blake } 2425c1bb86cdSEric Blake #else 2426c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2427c1bb86cdSEric Blake { 2428c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2429c1bb86cdSEric Blake int ret; 2430c1bb86cdSEric Blake int64_t size; 2431c1bb86cdSEric Blake 2432c1bb86cdSEric Blake ret = fd_open(bs); 2433c1bb86cdSEric Blake if (ret < 0) { 2434c1bb86cdSEric Blake return ret; 2435c1bb86cdSEric Blake } 2436c1bb86cdSEric Blake 2437c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2438c1bb86cdSEric Blake if (size < 0) { 2439c1bb86cdSEric Blake return -errno; 2440c1bb86cdSEric Blake } 2441c1bb86cdSEric Blake return size; 2442c1bb86cdSEric Blake } 2443c1bb86cdSEric Blake #endif 2444c1bb86cdSEric Blake 2445c1bb86cdSEric Blake static int64_t raw_get_allocated_file_size(BlockDriverState *bs) 2446c1bb86cdSEric Blake { 2447c1bb86cdSEric Blake struct stat st; 2448c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2449c1bb86cdSEric Blake 2450c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 2451c1bb86cdSEric Blake return -errno; 2452c1bb86cdSEric Blake } 2453c1bb86cdSEric Blake return (int64_t)st.st_blocks * 512; 2454c1bb86cdSEric Blake } 2455c1bb86cdSEric Blake 245693f4e2ffSKevin Wolf static int coroutine_fn 245793f4e2ffSKevin Wolf raw_co_create(BlockdevCreateOptions *options, Error **errp) 2458c1bb86cdSEric Blake { 2459927f11e1SKevin Wolf BlockdevCreateOptionsFile *file_opts; 24607c20c808SMax Reitz Error *local_err = NULL; 2461c1bb86cdSEric Blake int fd; 2462d815efcaSMax Reitz uint64_t perm, shared; 2463c1bb86cdSEric Blake int result = 0; 2464c1bb86cdSEric Blake 2465927f11e1SKevin Wolf /* Validate options and set default values */ 2466927f11e1SKevin Wolf assert(options->driver == BLOCKDEV_DRIVER_FILE); 2467927f11e1SKevin Wolf file_opts = &options->u.file; 2468c1bb86cdSEric Blake 2469927f11e1SKevin Wolf if (!file_opts->has_nocow) { 2470927f11e1SKevin Wolf file_opts->nocow = false; 2471927f11e1SKevin Wolf } 2472927f11e1SKevin Wolf if (!file_opts->has_preallocation) { 2473927f11e1SKevin Wolf file_opts->preallocation = PREALLOC_MODE_OFF; 2474c1bb86cdSEric Blake } 2475ffa244c8SKevin Wolf if (!file_opts->has_extent_size_hint) { 2476ffa244c8SKevin Wolf file_opts->extent_size_hint = 1 * MiB; 2477ffa244c8SKevin Wolf } 2478ffa244c8SKevin Wolf if (file_opts->extent_size_hint > UINT32_MAX) { 2479ffa244c8SKevin Wolf result = -EINVAL; 2480ffa244c8SKevin Wolf error_setg(errp, "Extent size hint is too large"); 2481ffa244c8SKevin Wolf goto out; 2482ffa244c8SKevin Wolf } 2483c1bb86cdSEric Blake 2484927f11e1SKevin Wolf /* Create file */ 2485b18a24a9SDaniel P. Berrangé fd = qemu_create(file_opts->filename, O_RDWR | O_BINARY, 0644, errp); 2486c1bb86cdSEric Blake if (fd < 0) { 2487c1bb86cdSEric Blake result = -errno; 2488c1bb86cdSEric Blake goto out; 2489c1bb86cdSEric Blake } 2490c1bb86cdSEric Blake 2491b8cf1913SMax Reitz /* Take permissions: We want to discard everything, so we need 2492b8cf1913SMax Reitz * BLK_PERM_WRITE; and truncation to the desired size requires 2493b8cf1913SMax Reitz * BLK_PERM_RESIZE. 2494b8cf1913SMax Reitz * On the other hand, we cannot share the RESIZE permission 2495b8cf1913SMax Reitz * because we promise that after this function, the file has the 2496b8cf1913SMax Reitz * size given in the options. If someone else were to resize it 2497b8cf1913SMax Reitz * concurrently, we could not guarantee that. 2498b8cf1913SMax Reitz * Note that after this function, we can no longer guarantee that 2499b8cf1913SMax Reitz * the file is not touched by a third party, so it may be resized 2500b8cf1913SMax Reitz * then. */ 2501b8cf1913SMax Reitz perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; 2502b8cf1913SMax Reitz shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; 2503b8cf1913SMax Reitz 2504b8cf1913SMax Reitz /* Step one: Take locks */ 25052996ffadSFam Zheng result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); 2506b8cf1913SMax Reitz if (result < 0) { 2507b8cf1913SMax Reitz goto out_close; 2508b8cf1913SMax Reitz } 2509b8cf1913SMax Reitz 2510b8cf1913SMax Reitz /* Step two: Check that nobody else has taken conflicting locks */ 2511b8cf1913SMax Reitz result = raw_check_lock_bytes(fd, perm, shared, errp); 2512b8cf1913SMax Reitz if (result < 0) { 2513b857431dSFam Zheng error_append_hint(errp, 2514b857431dSFam Zheng "Is another process using the image [%s]?\n", 2515b857431dSFam Zheng file_opts->filename); 25167c20c808SMax Reitz goto out_unlock; 2517b8cf1913SMax Reitz } 2518b8cf1913SMax Reitz 2519b8cf1913SMax Reitz /* Clear the file by truncating it to 0 */ 252093f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); 2521b8cf1913SMax Reitz if (result < 0) { 25227c20c808SMax Reitz goto out_unlock; 2523b8cf1913SMax Reitz } 2524b8cf1913SMax Reitz 2525927f11e1SKevin Wolf if (file_opts->nocow) { 2526c1bb86cdSEric Blake #ifdef __linux__ 2527c1bb86cdSEric Blake /* Set NOCOW flag to solve performance issue on fs like btrfs. 2528c1bb86cdSEric Blake * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value 2529c1bb86cdSEric Blake * will be ignored since any failure of this operation should not 2530c1bb86cdSEric Blake * block the left work. 2531c1bb86cdSEric Blake */ 2532c1bb86cdSEric Blake int attr; 2533c1bb86cdSEric Blake if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { 2534c1bb86cdSEric Blake attr |= FS_NOCOW_FL; 2535c1bb86cdSEric Blake ioctl(fd, FS_IOC_SETFLAGS, &attr); 2536c1bb86cdSEric Blake } 2537c1bb86cdSEric Blake #endif 2538c1bb86cdSEric Blake } 2539ffa244c8SKevin Wolf #ifdef FS_IOC_FSSETXATTR 2540ffa244c8SKevin Wolf /* 2541ffa244c8SKevin Wolf * Try to set the extent size hint. Failure is not fatal, and a warning is 2542ffa244c8SKevin Wolf * only printed if the option was explicitly specified. 2543ffa244c8SKevin Wolf */ 2544ffa244c8SKevin Wolf { 2545ffa244c8SKevin Wolf struct fsxattr attr; 2546ffa244c8SKevin Wolf result = ioctl(fd, FS_IOC_FSGETXATTR, &attr); 2547ffa244c8SKevin Wolf if (result == 0) { 2548ffa244c8SKevin Wolf attr.fsx_xflags |= FS_XFLAG_EXTSIZE; 2549ffa244c8SKevin Wolf attr.fsx_extsize = file_opts->extent_size_hint; 2550ffa244c8SKevin Wolf result = ioctl(fd, FS_IOC_FSSETXATTR, &attr); 2551ffa244c8SKevin Wolf } 2552ffa244c8SKevin Wolf if (result < 0 && file_opts->has_extent_size_hint && 2553ffa244c8SKevin Wolf file_opts->extent_size_hint) 2554ffa244c8SKevin Wolf { 2555ffa244c8SKevin Wolf warn_report("Failed to set extent size hint: %s", 2556ffa244c8SKevin Wolf strerror(errno)); 2557ffa244c8SKevin Wolf } 2558ffa244c8SKevin Wolf } 2559ffa244c8SKevin Wolf #endif 2560c1bb86cdSEric Blake 2561b8cf1913SMax Reitz /* Resize and potentially preallocate the file to the desired 2562b8cf1913SMax Reitz * final size */ 256393f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, file_opts->size, 256493f4e2ffSKevin Wolf file_opts->preallocation, errp); 25659f63b07eSMax Reitz if (result < 0) { 25667c20c808SMax Reitz goto out_unlock; 25677c20c808SMax Reitz } 25687c20c808SMax Reitz 25697c20c808SMax Reitz out_unlock: 25702996ffadSFam Zheng raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); 25717c20c808SMax Reitz if (local_err) { 25727c20c808SMax Reitz /* The above call should not fail, and if it does, that does 25737c20c808SMax Reitz * not mean the whole creation operation has failed. So 25747c20c808SMax Reitz * report it the user for their convenience, but do not report 25757c20c808SMax Reitz * it to the caller. */ 2576db0754dfSFam Zheng warn_report_err(local_err); 25775a1dad9dSNir Soffer } 25785a1dad9dSNir Soffer 25795a1dad9dSNir Soffer out_close: 2580c1bb86cdSEric Blake if (qemu_close(fd) != 0 && result == 0) { 2581c1bb86cdSEric Blake result = -errno; 2582c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not close the new file"); 2583c1bb86cdSEric Blake } 2584c1bb86cdSEric Blake out: 2585c1bb86cdSEric Blake return result; 2586c1bb86cdSEric Blake } 2587c1bb86cdSEric Blake 2588b92902dfSMaxim Levitsky static int coroutine_fn raw_co_create_opts(BlockDriver *drv, 2589b92902dfSMaxim Levitsky const char *filename, 2590b92902dfSMaxim Levitsky QemuOpts *opts, 2591927f11e1SKevin Wolf Error **errp) 2592927f11e1SKevin Wolf { 2593927f11e1SKevin Wolf BlockdevCreateOptions options; 2594927f11e1SKevin Wolf int64_t total_size = 0; 2595ffa244c8SKevin Wolf int64_t extent_size_hint = 0; 2596ffa244c8SKevin Wolf bool has_extent_size_hint = false; 2597927f11e1SKevin Wolf bool nocow = false; 2598927f11e1SKevin Wolf PreallocMode prealloc; 2599927f11e1SKevin Wolf char *buf = NULL; 2600927f11e1SKevin Wolf Error *local_err = NULL; 2601927f11e1SKevin Wolf 2602927f11e1SKevin Wolf /* Skip file: protocol prefix */ 2603927f11e1SKevin Wolf strstart(filename, "file:", &filename); 2604927f11e1SKevin Wolf 2605927f11e1SKevin Wolf /* Read out options */ 2606927f11e1SKevin Wolf total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2607927f11e1SKevin Wolf BDRV_SECTOR_SIZE); 2608ffa244c8SKevin Wolf if (qemu_opt_get(opts, BLOCK_OPT_EXTENT_SIZE_HINT)) { 2609ffa244c8SKevin Wolf has_extent_size_hint = true; 2610ffa244c8SKevin Wolf extent_size_hint = 2611ffa244c8SKevin Wolf qemu_opt_get_size_del(opts, BLOCK_OPT_EXTENT_SIZE_HINT, -1); 2612ffa244c8SKevin Wolf } 2613927f11e1SKevin Wolf nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); 2614927f11e1SKevin Wolf buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2615927f11e1SKevin Wolf prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, 2616927f11e1SKevin Wolf PREALLOC_MODE_OFF, &local_err); 2617927f11e1SKevin Wolf g_free(buf); 2618927f11e1SKevin Wolf if (local_err) { 2619927f11e1SKevin Wolf error_propagate(errp, local_err); 2620927f11e1SKevin Wolf return -EINVAL; 2621927f11e1SKevin Wolf } 2622927f11e1SKevin Wolf 2623927f11e1SKevin Wolf options = (BlockdevCreateOptions) { 2624927f11e1SKevin Wolf .driver = BLOCKDEV_DRIVER_FILE, 2625927f11e1SKevin Wolf .u.file = { 2626927f11e1SKevin Wolf .filename = (char *) filename, 2627927f11e1SKevin Wolf .size = total_size, 2628927f11e1SKevin Wolf .has_preallocation = true, 2629927f11e1SKevin Wolf .preallocation = prealloc, 2630927f11e1SKevin Wolf .has_nocow = true, 2631927f11e1SKevin Wolf .nocow = nocow, 2632ffa244c8SKevin Wolf .has_extent_size_hint = has_extent_size_hint, 2633ffa244c8SKevin Wolf .extent_size_hint = extent_size_hint, 2634927f11e1SKevin Wolf }, 2635927f11e1SKevin Wolf }; 2636927f11e1SKevin Wolf return raw_co_create(&options, errp); 2637927f11e1SKevin Wolf } 2638927f11e1SKevin Wolf 26399bffae14SDaniel Henrique Barboza static int coroutine_fn raw_co_delete_file(BlockDriverState *bs, 26409bffae14SDaniel Henrique Barboza Error **errp) 26419bffae14SDaniel Henrique Barboza { 26429bffae14SDaniel Henrique Barboza struct stat st; 26439bffae14SDaniel Henrique Barboza int ret; 26449bffae14SDaniel Henrique Barboza 26459bffae14SDaniel Henrique Barboza if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) { 26469bffae14SDaniel Henrique Barboza error_setg_errno(errp, ENOENT, "%s is not a regular file", 26479bffae14SDaniel Henrique Barboza bs->filename); 26489bffae14SDaniel Henrique Barboza return -ENOENT; 26499bffae14SDaniel Henrique Barboza } 26509bffae14SDaniel Henrique Barboza 26519bffae14SDaniel Henrique Barboza ret = unlink(bs->filename); 26529bffae14SDaniel Henrique Barboza if (ret < 0) { 26539bffae14SDaniel Henrique Barboza ret = -errno; 26549bffae14SDaniel Henrique Barboza error_setg_errno(errp, -ret, "Error when deleting file %s", 26559bffae14SDaniel Henrique Barboza bs->filename); 26569bffae14SDaniel Henrique Barboza } 26579bffae14SDaniel Henrique Barboza 26589bffae14SDaniel Henrique Barboza return ret; 26599bffae14SDaniel Henrique Barboza } 26609bffae14SDaniel Henrique Barboza 2661c1bb86cdSEric Blake /* 2662c1bb86cdSEric Blake * Find allocation range in @bs around offset @start. 2663c1bb86cdSEric Blake * May change underlying file descriptor's file offset. 2664c1bb86cdSEric Blake * If @start is not in a hole, store @start in @data, and the 2665c1bb86cdSEric Blake * beginning of the next hole in @hole, and return 0. 2666c1bb86cdSEric Blake * If @start is in a non-trailing hole, store @start in @hole and the 2667c1bb86cdSEric Blake * beginning of the next non-hole in @data, and return 0. 2668c1bb86cdSEric Blake * If @start is in a trailing hole or beyond EOF, return -ENXIO. 2669c1bb86cdSEric Blake * If we can't find out, return a negative errno other than -ENXIO. 2670c1bb86cdSEric Blake */ 2671c1bb86cdSEric Blake static int find_allocation(BlockDriverState *bs, off_t start, 2672c1bb86cdSEric Blake off_t *data, off_t *hole) 2673c1bb86cdSEric Blake { 2674c1bb86cdSEric Blake #if defined SEEK_HOLE && defined SEEK_DATA 2675c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2676c1bb86cdSEric Blake off_t offs; 2677c1bb86cdSEric Blake 2678c1bb86cdSEric Blake /* 2679c1bb86cdSEric Blake * SEEK_DATA cases: 2680c1bb86cdSEric Blake * D1. offs == start: start is in data 2681c1bb86cdSEric Blake * D2. offs > start: start is in a hole, next data at offs 2682c1bb86cdSEric Blake * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 2683c1bb86cdSEric Blake * or start is beyond EOF 2684c1bb86cdSEric Blake * If the latter happens, the file has been truncated behind 2685c1bb86cdSEric Blake * our back since we opened it. All bets are off then. 2686c1bb86cdSEric Blake * Treating like a trailing hole is simplest. 2687c1bb86cdSEric Blake * D4. offs < 0, errno != ENXIO: we learned nothing 2688c1bb86cdSEric Blake */ 2689c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_DATA); 2690c1bb86cdSEric Blake if (offs < 0) { 2691c1bb86cdSEric Blake return -errno; /* D3 or D4 */ 2692c1bb86cdSEric Blake } 2693a03083a0SJeff Cody 2694a03083a0SJeff Cody if (offs < start) { 2695a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2696a03083a0SJeff Cody * -EIO in this case, and we'll treat it like D4. */ 2697a03083a0SJeff Cody return -EIO; 2698a03083a0SJeff Cody } 2699c1bb86cdSEric Blake 2700c1bb86cdSEric Blake if (offs > start) { 2701c1bb86cdSEric Blake /* D2: in hole, next data at offs */ 2702c1bb86cdSEric Blake *hole = start; 2703c1bb86cdSEric Blake *data = offs; 2704c1bb86cdSEric Blake return 0; 2705c1bb86cdSEric Blake } 2706c1bb86cdSEric Blake 2707c1bb86cdSEric Blake /* D1: in data, end not yet known */ 2708c1bb86cdSEric Blake 2709c1bb86cdSEric Blake /* 2710c1bb86cdSEric Blake * SEEK_HOLE cases: 2711c1bb86cdSEric Blake * H1. offs == start: start is in a hole 2712c1bb86cdSEric Blake * If this happens here, a hole has been dug behind our back 2713c1bb86cdSEric Blake * since the previous lseek(). 2714c1bb86cdSEric Blake * H2. offs > start: either start is in data, next hole at offs, 2715c1bb86cdSEric Blake * or start is in trailing hole, EOF at offs 2716c1bb86cdSEric Blake * Linux treats trailing holes like any other hole: offs == 2717c1bb86cdSEric Blake * start. Solaris seeks to EOF instead: offs > start (blech). 2718c1bb86cdSEric Blake * If that happens here, a hole has been dug behind our back 2719c1bb86cdSEric Blake * since the previous lseek(). 2720c1bb86cdSEric Blake * H3. offs < 0, errno = ENXIO: start is beyond EOF 2721c1bb86cdSEric Blake * If this happens, the file has been truncated behind our 2722c1bb86cdSEric Blake * back since we opened it. Treat it like a trailing hole. 2723c1bb86cdSEric Blake * H4. offs < 0, errno != ENXIO: we learned nothing 2724c1bb86cdSEric Blake * Pretend we know nothing at all, i.e. "forget" about D1. 2725c1bb86cdSEric Blake */ 2726c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_HOLE); 2727c1bb86cdSEric Blake if (offs < 0) { 2728c1bb86cdSEric Blake return -errno; /* D1 and (H3 or H4) */ 2729c1bb86cdSEric Blake } 2730a03083a0SJeff Cody 2731a03083a0SJeff Cody if (offs < start) { 2732a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2733a03083a0SJeff Cody * -EIO in this case, and we'll treat it like H4. */ 2734a03083a0SJeff Cody return -EIO; 2735a03083a0SJeff Cody } 2736c1bb86cdSEric Blake 2737c1bb86cdSEric Blake if (offs > start) { 2738c1bb86cdSEric Blake /* 2739c1bb86cdSEric Blake * D1 and H2: either in data, next hole at offs, or it was in 2740c1bb86cdSEric Blake * data but is now in a trailing hole. In the latter case, 2741c1bb86cdSEric Blake * all bets are off. Treating it as if it there was data all 2742c1bb86cdSEric Blake * the way to EOF is safe, so simply do that. 2743c1bb86cdSEric Blake */ 2744c1bb86cdSEric Blake *data = start; 2745c1bb86cdSEric Blake *hole = offs; 2746c1bb86cdSEric Blake return 0; 2747c1bb86cdSEric Blake } 2748c1bb86cdSEric Blake 2749c1bb86cdSEric Blake /* D1 and H1 */ 2750c1bb86cdSEric Blake return -EBUSY; 2751c1bb86cdSEric Blake #else 2752c1bb86cdSEric Blake return -ENOTSUP; 2753c1bb86cdSEric Blake #endif 2754c1bb86cdSEric Blake } 2755c1bb86cdSEric Blake 2756c1bb86cdSEric Blake /* 2757a290f085SEric Blake * Returns the allocation status of the specified offset. 2758c1bb86cdSEric Blake * 2759a290f085SEric Blake * The block layer guarantees 'offset' and 'bytes' are within bounds. 2760c1bb86cdSEric Blake * 2761a290f085SEric Blake * 'pnum' is set to the number of bytes (including and immediately following 2762a290f085SEric Blake * the specified offset) that are known to be in the same 2763c1bb86cdSEric Blake * allocated/unallocated state. 2764c1bb86cdSEric Blake * 2765869e7ee8SHanna Reitz * 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may 2766869e7ee8SHanna Reitz * well exceed it. 2767c1bb86cdSEric Blake */ 2768a290f085SEric Blake static int coroutine_fn raw_co_block_status(BlockDriverState *bs, 2769a290f085SEric Blake bool want_zero, 2770a290f085SEric Blake int64_t offset, 2771a290f085SEric Blake int64_t bytes, int64_t *pnum, 2772a290f085SEric Blake int64_t *map, 2773c1bb86cdSEric Blake BlockDriverState **file) 2774c1bb86cdSEric Blake { 2775a290f085SEric Blake off_t data = 0, hole = 0; 2776c1bb86cdSEric Blake int ret; 2777c1bb86cdSEric Blake 27789c3db310SMax Reitz assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); 27799c3db310SMax Reitz 2780c1bb86cdSEric Blake ret = fd_open(bs); 2781c1bb86cdSEric Blake if (ret < 0) { 2782c1bb86cdSEric Blake return ret; 2783c1bb86cdSEric Blake } 2784c1bb86cdSEric Blake 2785a290f085SEric Blake if (!want_zero) { 2786a290f085SEric Blake *pnum = bytes; 2787a290f085SEric Blake *map = offset; 2788a290f085SEric Blake *file = bs; 2789a290f085SEric Blake return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 2790c1bb86cdSEric Blake } 2791c1bb86cdSEric Blake 2792a290f085SEric Blake ret = find_allocation(bs, offset, &data, &hole); 2793c1bb86cdSEric Blake if (ret == -ENXIO) { 2794c1bb86cdSEric Blake /* Trailing hole */ 2795a290f085SEric Blake *pnum = bytes; 2796c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2797c1bb86cdSEric Blake } else if (ret < 0) { 2798c1bb86cdSEric Blake /* No info available, so pretend there are no holes */ 2799a290f085SEric Blake *pnum = bytes; 2800c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2801a290f085SEric Blake } else if (data == offset) { 2802a290f085SEric Blake /* On a data extent, compute bytes to the end of the extent, 2803c1bb86cdSEric Blake * possibly including a partial sector at EOF. */ 2804869e7ee8SHanna Reitz *pnum = hole - offset; 28059c3db310SMax Reitz 28069c3db310SMax Reitz /* 28079c3db310SMax Reitz * We are not allowed to return partial sectors, though, so 28089c3db310SMax Reitz * round up if necessary. 28099c3db310SMax Reitz */ 28109c3db310SMax Reitz if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { 28119c3db310SMax Reitz int64_t file_length = raw_getlength(bs); 28129c3db310SMax Reitz if (file_length > 0) { 28139c3db310SMax Reitz /* Ignore errors, this is just a safeguard */ 28149c3db310SMax Reitz assert(hole == file_length); 28159c3db310SMax Reitz } 28169c3db310SMax Reitz *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); 28179c3db310SMax Reitz } 28189c3db310SMax Reitz 2819c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2820c1bb86cdSEric Blake } else { 2821a290f085SEric Blake /* On a hole, compute bytes to the beginning of the next extent. */ 2822a290f085SEric Blake assert(hole == offset); 2823869e7ee8SHanna Reitz *pnum = data - offset; 2824c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2825c1bb86cdSEric Blake } 2826a290f085SEric Blake *map = offset; 2827c1bb86cdSEric Blake *file = bs; 2828a290f085SEric Blake return ret | BDRV_BLOCK_OFFSET_VALID; 2829c1bb86cdSEric Blake } 2830c1bb86cdSEric Blake 283131be8a2aSStefan Hajnoczi #if defined(__linux__) 283231be8a2aSStefan Hajnoczi /* Verify that the file is not in the page cache */ 283331be8a2aSStefan Hajnoczi static void check_cache_dropped(BlockDriverState *bs, Error **errp) 283431be8a2aSStefan Hajnoczi { 283531be8a2aSStefan Hajnoczi const size_t window_size = 128 * 1024 * 1024; 283631be8a2aSStefan Hajnoczi BDRVRawState *s = bs->opaque; 283731be8a2aSStefan Hajnoczi void *window = NULL; 283831be8a2aSStefan Hajnoczi size_t length = 0; 283931be8a2aSStefan Hajnoczi unsigned char *vec; 284031be8a2aSStefan Hajnoczi size_t page_size; 284131be8a2aSStefan Hajnoczi off_t offset; 284231be8a2aSStefan Hajnoczi off_t end; 284331be8a2aSStefan Hajnoczi 284431be8a2aSStefan Hajnoczi /* mincore(2) page status information requires 1 byte per page */ 284531be8a2aSStefan Hajnoczi page_size = sysconf(_SC_PAGESIZE); 284631be8a2aSStefan Hajnoczi vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); 284731be8a2aSStefan Hajnoczi 284831be8a2aSStefan Hajnoczi end = raw_getlength(bs); 284931be8a2aSStefan Hajnoczi 285031be8a2aSStefan Hajnoczi for (offset = 0; offset < end; offset += window_size) { 285131be8a2aSStefan Hajnoczi void *new_window; 285231be8a2aSStefan Hajnoczi size_t new_length; 285331be8a2aSStefan Hajnoczi size_t vec_end; 285431be8a2aSStefan Hajnoczi size_t i; 285531be8a2aSStefan Hajnoczi int ret; 285631be8a2aSStefan Hajnoczi 285731be8a2aSStefan Hajnoczi /* Unmap previous window if size has changed */ 285831be8a2aSStefan Hajnoczi new_length = MIN(end - offset, window_size); 285931be8a2aSStefan Hajnoczi if (new_length != length) { 286031be8a2aSStefan Hajnoczi munmap(window, length); 286131be8a2aSStefan Hajnoczi window = NULL; 286231be8a2aSStefan Hajnoczi length = 0; 286331be8a2aSStefan Hajnoczi } 286431be8a2aSStefan Hajnoczi 286531be8a2aSStefan Hajnoczi new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, 286631be8a2aSStefan Hajnoczi s->fd, offset); 286731be8a2aSStefan Hajnoczi if (new_window == MAP_FAILED) { 286831be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mmap failed"); 286931be8a2aSStefan Hajnoczi break; 287031be8a2aSStefan Hajnoczi } 287131be8a2aSStefan Hajnoczi 287231be8a2aSStefan Hajnoczi window = new_window; 287331be8a2aSStefan Hajnoczi length = new_length; 287431be8a2aSStefan Hajnoczi 287531be8a2aSStefan Hajnoczi ret = mincore(window, length, vec); 287631be8a2aSStefan Hajnoczi if (ret < 0) { 287731be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mincore failed"); 287831be8a2aSStefan Hajnoczi break; 287931be8a2aSStefan Hajnoczi } 288031be8a2aSStefan Hajnoczi 288131be8a2aSStefan Hajnoczi vec_end = DIV_ROUND_UP(length, page_size); 288231be8a2aSStefan Hajnoczi for (i = 0; i < vec_end; i++) { 288331be8a2aSStefan Hajnoczi if (vec[i] & 0x1) { 288431be8a2aSStefan Hajnoczi break; 288531be8a2aSStefan Hajnoczi } 288631be8a2aSStefan Hajnoczi } 288777ed971bSMarkus Armbruster if (i < vec_end) { 288877ed971bSMarkus Armbruster error_setg(errp, "page cache still in use!"); 288977ed971bSMarkus Armbruster break; 289077ed971bSMarkus Armbruster } 289131be8a2aSStefan Hajnoczi } 289231be8a2aSStefan Hajnoczi 289331be8a2aSStefan Hajnoczi if (window) { 289431be8a2aSStefan Hajnoczi munmap(window, length); 289531be8a2aSStefan Hajnoczi } 289631be8a2aSStefan Hajnoczi 289731be8a2aSStefan Hajnoczi g_free(vec); 289831be8a2aSStefan Hajnoczi } 289931be8a2aSStefan Hajnoczi #endif /* __linux__ */ 290031be8a2aSStefan Hajnoczi 2901dd577a26SStefan Hajnoczi static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, 2902dd577a26SStefan Hajnoczi Error **errp) 2903dd577a26SStefan Hajnoczi { 2904dd577a26SStefan Hajnoczi BDRVRawState *s = bs->opaque; 2905dd577a26SStefan Hajnoczi int ret; 2906dd577a26SStefan Hajnoczi 2907dd577a26SStefan Hajnoczi ret = fd_open(bs); 2908dd577a26SStefan Hajnoczi if (ret < 0) { 2909dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "The file descriptor is not open"); 2910dd577a26SStefan Hajnoczi return; 2911dd577a26SStefan Hajnoczi } 2912dd577a26SStefan Hajnoczi 2913f357fcd8SStefan Hajnoczi if (!s->drop_cache) { 2914f357fcd8SStefan Hajnoczi return; 2915f357fcd8SStefan Hajnoczi } 2916f357fcd8SStefan Hajnoczi 2917dd577a26SStefan Hajnoczi if (s->open_flags & O_DIRECT) { 2918dd577a26SStefan Hajnoczi return; /* No host kernel page cache */ 2919dd577a26SStefan Hajnoczi } 2920dd577a26SStefan Hajnoczi 2921dd577a26SStefan Hajnoczi #if defined(__linux__) 2922dd577a26SStefan Hajnoczi /* This sets the scene for the next syscall... */ 2923dd577a26SStefan Hajnoczi ret = bdrv_co_flush(bs); 2924dd577a26SStefan Hajnoczi if (ret < 0) { 2925dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "flush failed"); 2926dd577a26SStefan Hajnoczi return; 2927dd577a26SStefan Hajnoczi } 2928dd577a26SStefan Hajnoczi 2929dd577a26SStefan Hajnoczi /* Linux does not invalidate pages that are dirty, locked, or mmapped by a 2930dd577a26SStefan Hajnoczi * process. These limitations are okay because we just fsynced the file, 2931dd577a26SStefan Hajnoczi * we don't use mmap, and the file should not be in use by other processes. 2932dd577a26SStefan Hajnoczi */ 2933dd577a26SStefan Hajnoczi ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); 2934dd577a26SStefan Hajnoczi if (ret != 0) { /* the return value is a positive errno */ 2935dd577a26SStefan Hajnoczi error_setg_errno(errp, ret, "fadvise failed"); 2936dd577a26SStefan Hajnoczi return; 2937dd577a26SStefan Hajnoczi } 293831be8a2aSStefan Hajnoczi 293931be8a2aSStefan Hajnoczi if (s->check_cache_dropped) { 294031be8a2aSStefan Hajnoczi check_cache_dropped(bs, errp); 294131be8a2aSStefan Hajnoczi } 2942dd577a26SStefan Hajnoczi #else /* __linux__ */ 2943dd577a26SStefan Hajnoczi /* Do nothing. Live migration to a remote host with cache.direct=off is 2944dd577a26SStefan Hajnoczi * unsupported on other host operating systems. Cache consistency issues 2945dd577a26SStefan Hajnoczi * may occur but no error is reported here, partly because that's the 2946dd577a26SStefan Hajnoczi * historical behavior and partly because it's hard to differentiate valid 2947dd577a26SStefan Hajnoczi * configurations that should not cause errors. 2948dd577a26SStefan Hajnoczi */ 2949dd577a26SStefan Hajnoczi #endif /* !__linux__ */ 2950dd577a26SStefan Hajnoczi } 2951dd577a26SStefan Hajnoczi 29521c450366SAnton Nefedov static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret) 29531c450366SAnton Nefedov { 29541c450366SAnton Nefedov if (ret) { 29551c450366SAnton Nefedov s->stats.discard_nb_failed++; 29561c450366SAnton Nefedov } else { 29571c450366SAnton Nefedov s->stats.discard_nb_ok++; 29581c450366SAnton Nefedov s->stats.discard_bytes_ok += nbytes; 29591c450366SAnton Nefedov } 29601c450366SAnton Nefedov } 29611c450366SAnton Nefedov 296233d70fb6SKevin Wolf static coroutine_fn int 29630c802287SVladimir Sementsov-Ogievskiy raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes, 29640c802287SVladimir Sementsov-Ogievskiy bool blkdev) 2965c1bb86cdSEric Blake { 2966c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 296746ee0f46SKevin Wolf RawPosixAIOData acb; 29681c450366SAnton Nefedov int ret; 2969c1bb86cdSEric Blake 297046ee0f46SKevin Wolf acb = (RawPosixAIOData) { 297146ee0f46SKevin Wolf .bs = bs, 297246ee0f46SKevin Wolf .aio_fildes = s->fd, 297346ee0f46SKevin Wolf .aio_type = QEMU_AIO_DISCARD, 297446ee0f46SKevin Wolf .aio_offset = offset, 297546ee0f46SKevin Wolf .aio_nbytes = bytes, 297646ee0f46SKevin Wolf }; 297746ee0f46SKevin Wolf 297846ee0f46SKevin Wolf if (blkdev) { 297946ee0f46SKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 298046ee0f46SKevin Wolf } 298146ee0f46SKevin Wolf 29821c450366SAnton Nefedov ret = raw_thread_pool_submit(bs, handle_aiocb_discard, &acb); 29831c450366SAnton Nefedov raw_account_discard(s, bytes, ret); 29841c450366SAnton Nefedov return ret; 298546ee0f46SKevin Wolf } 298646ee0f46SKevin Wolf 298746ee0f46SKevin Wolf static coroutine_fn int 29880c802287SVladimir Sementsov-Ogievskiy raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) 298946ee0f46SKevin Wolf { 299046ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, false); 2991c1bb86cdSEric Blake } 2992c1bb86cdSEric Blake 29937154d8aeSKevin Wolf static int coroutine_fn 2994f34b2bcfSVladimir Sementsov-Ogievskiy raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, 29957154d8aeSKevin Wolf BdrvRequestFlags flags, bool blkdev) 29967154d8aeSKevin Wolf { 29977154d8aeSKevin Wolf BDRVRawState *s = bs->opaque; 29987154d8aeSKevin Wolf RawPosixAIOData acb; 29997154d8aeSKevin Wolf ThreadPoolFunc *handler; 30007154d8aeSKevin Wolf 3001292d06b9SMax Reitz #ifdef CONFIG_FALLOCATE 3002292d06b9SMax Reitz if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { 3003292d06b9SMax Reitz BdrvTrackedRequest *req; 3004292d06b9SMax Reitz 3005292d06b9SMax Reitz /* 3006292d06b9SMax Reitz * This is a workaround for a bug in the Linux XFS driver, 3007292d06b9SMax Reitz * where writes submitted through the AIO interface will be 3008292d06b9SMax Reitz * discarded if they happen beyond a concurrently running 3009292d06b9SMax Reitz * fallocate() that increases the file length (i.e., both the 3010292d06b9SMax Reitz * write and the fallocate() happen beyond the EOF). 3011292d06b9SMax Reitz * 3012292d06b9SMax Reitz * To work around it, we extend the tracked request for this 3013292d06b9SMax Reitz * zero write until INT64_MAX (effectively infinity), and mark 3014292d06b9SMax Reitz * it as serializing. 3015292d06b9SMax Reitz * 3016292d06b9SMax Reitz * We have to enable this workaround for all filesystems and 3017292d06b9SMax Reitz * AIO modes (not just XFS with aio=native), because for 3018292d06b9SMax Reitz * remote filesystems we do not know the host configuration. 3019292d06b9SMax Reitz */ 3020292d06b9SMax Reitz 3021292d06b9SMax Reitz req = bdrv_co_get_self_request(bs); 3022292d06b9SMax Reitz assert(req); 3023292d06b9SMax Reitz assert(req->type == BDRV_TRACKED_WRITE); 3024292d06b9SMax Reitz assert(req->offset <= offset); 3025292d06b9SMax Reitz assert(req->offset + req->bytes >= offset + bytes); 3026292d06b9SMax Reitz 30278b117001SVladimir Sementsov-Ogievskiy req->bytes = BDRV_MAX_LENGTH - req->offset; 30288b117001SVladimir Sementsov-Ogievskiy 302969b55e03SVladimir Sementsov-Ogievskiy bdrv_check_request(req->offset, req->bytes, &error_abort); 3030292d06b9SMax Reitz 30318ac5aab2SVladimir Sementsov-Ogievskiy bdrv_make_request_serialising(req, bs->bl.request_alignment); 3032292d06b9SMax Reitz } 3033292d06b9SMax Reitz #endif 3034292d06b9SMax Reitz 30357154d8aeSKevin Wolf acb = (RawPosixAIOData) { 30367154d8aeSKevin Wolf .bs = bs, 30377154d8aeSKevin Wolf .aio_fildes = s->fd, 30387154d8aeSKevin Wolf .aio_type = QEMU_AIO_WRITE_ZEROES, 30397154d8aeSKevin Wolf .aio_offset = offset, 30407154d8aeSKevin Wolf .aio_nbytes = bytes, 30417154d8aeSKevin Wolf }; 30427154d8aeSKevin Wolf 30437154d8aeSKevin Wolf if (blkdev) { 30447154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 30457154d8aeSKevin Wolf } 3046738301e1SKevin Wolf if (flags & BDRV_REQ_NO_FALLBACK) { 3047738301e1SKevin Wolf acb.aio_type |= QEMU_AIO_NO_FALLBACK; 3048738301e1SKevin Wolf } 30497154d8aeSKevin Wolf 30507154d8aeSKevin Wolf if (flags & BDRV_REQ_MAY_UNMAP) { 30517154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_DISCARD; 30527154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes_unmap; 30537154d8aeSKevin Wolf } else { 30547154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes; 30557154d8aeSKevin Wolf } 30567154d8aeSKevin Wolf 30577154d8aeSKevin Wolf return raw_thread_pool_submit(bs, handler, &acb); 30587154d8aeSKevin Wolf } 30597154d8aeSKevin Wolf 3060c1bb86cdSEric Blake static int coroutine_fn raw_co_pwrite_zeroes( 3061c1bb86cdSEric Blake BlockDriverState *bs, int64_t offset, 3062f34b2bcfSVladimir Sementsov-Ogievskiy int64_t bytes, BdrvRequestFlags flags) 3063c1bb86cdSEric Blake { 30647154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); 3065c1bb86cdSEric Blake } 3066c1bb86cdSEric Blake 3067c1bb86cdSEric Blake static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 3068c1bb86cdSEric Blake { 3069c1bb86cdSEric Blake return 0; 3070c1bb86cdSEric Blake } 3071c1bb86cdSEric Blake 3072d9245599SAnton Nefedov static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs) 3073d9245599SAnton Nefedov { 3074d9245599SAnton Nefedov BDRVRawState *s = bs->opaque; 3075d9245599SAnton Nefedov return (BlockStatsSpecificFile) { 3076d9245599SAnton Nefedov .discard_nb_ok = s->stats.discard_nb_ok, 3077d9245599SAnton Nefedov .discard_nb_failed = s->stats.discard_nb_failed, 3078d9245599SAnton Nefedov .discard_bytes_ok = s->stats.discard_bytes_ok, 3079d9245599SAnton Nefedov }; 3080d9245599SAnton Nefedov } 3081d9245599SAnton Nefedov 3082d9245599SAnton Nefedov static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs) 3083d9245599SAnton Nefedov { 3084d9245599SAnton Nefedov BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 3085d9245599SAnton Nefedov 3086d9245599SAnton Nefedov stats->driver = BLOCKDEV_DRIVER_FILE; 3087d9245599SAnton Nefedov stats->u.file = get_blockstats_specific_file(bs); 3088d9245599SAnton Nefedov 3089d9245599SAnton Nefedov return stats; 3090d9245599SAnton Nefedov } 3091d9245599SAnton Nefedov 309214176c8dSJoelle van Dyne #if defined(HAVE_HOST_BLOCK_DEVICE) 3093d9245599SAnton Nefedov static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs) 3094d9245599SAnton Nefedov { 3095d9245599SAnton Nefedov BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 3096d9245599SAnton Nefedov 3097d9245599SAnton Nefedov stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE; 3098d9245599SAnton Nefedov stats->u.host_device = get_blockstats_specific_file(bs); 3099d9245599SAnton Nefedov 3100d9245599SAnton Nefedov return stats; 3101d9245599SAnton Nefedov } 310214176c8dSJoelle van Dyne #endif /* HAVE_HOST_BLOCK_DEVICE */ 3103d9245599SAnton Nefedov 3104c1bb86cdSEric Blake static QemuOptsList raw_create_opts = { 3105c1bb86cdSEric Blake .name = "raw-create-opts", 3106c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), 3107c1bb86cdSEric Blake .desc = { 3108c1bb86cdSEric Blake { 3109c1bb86cdSEric Blake .name = BLOCK_OPT_SIZE, 3110c1bb86cdSEric Blake .type = QEMU_OPT_SIZE, 3111c1bb86cdSEric Blake .help = "Virtual disk size" 3112c1bb86cdSEric Blake }, 3113c1bb86cdSEric Blake { 3114c1bb86cdSEric Blake .name = BLOCK_OPT_NOCOW, 3115c1bb86cdSEric Blake .type = QEMU_OPT_BOOL, 3116c1bb86cdSEric Blake .help = "Turn off copy-on-write (valid only on btrfs)" 3117c1bb86cdSEric Blake }, 3118c1bb86cdSEric Blake { 3119c1bb86cdSEric Blake .name = BLOCK_OPT_PREALLOC, 3120c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 3121abea0053SStefano Garzarella .help = "Preallocation mode (allowed values: off" 3122abea0053SStefano Garzarella #ifdef CONFIG_POSIX_FALLOCATE 3123abea0053SStefano Garzarella ", falloc" 3124abea0053SStefano Garzarella #endif 3125abea0053SStefano Garzarella ", full)" 3126c1bb86cdSEric Blake }, 3127ffa244c8SKevin Wolf { 3128ffa244c8SKevin Wolf .name = BLOCK_OPT_EXTENT_SIZE_HINT, 3129ffa244c8SKevin Wolf .type = QEMU_OPT_SIZE, 3130ffa244c8SKevin Wolf .help = "Extent size hint for the image file, 0 to disable" 3131ffa244c8SKevin Wolf }, 3132c1bb86cdSEric Blake { /* end of list */ } 3133c1bb86cdSEric Blake } 3134c1bb86cdSEric Blake }; 3135c1bb86cdSEric Blake 3136244a5668SFam Zheng static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, 3137244a5668SFam Zheng Error **errp) 3138244a5668SFam Zheng { 31396ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 314072373e40SVladimir Sementsov-Ogievskiy int input_flags = s->reopen_state ? s->reopen_state->flags : bs->open_flags; 31416ceabe6fSKevin Wolf int open_flags; 31426ceabe6fSKevin Wolf int ret; 31436ceabe6fSKevin Wolf 31446ceabe6fSKevin Wolf /* We may need a new fd if auto-read-only switches the mode */ 314572373e40SVladimir Sementsov-Ogievskiy ret = raw_reconfigure_getfd(bs, input_flags, &open_flags, perm, 31466ceabe6fSKevin Wolf false, errp); 31476ceabe6fSKevin Wolf if (ret < 0) { 31486ceabe6fSKevin Wolf return ret; 31496ceabe6fSKevin Wolf } else if (ret != s->fd) { 315072373e40SVladimir Sementsov-Ogievskiy Error *local_err = NULL; 315172373e40SVladimir Sementsov-Ogievskiy 315272373e40SVladimir Sementsov-Ogievskiy /* 315372373e40SVladimir Sementsov-Ogievskiy * Fail already check_perm() if we can't get a working O_DIRECT 315472373e40SVladimir Sementsov-Ogievskiy * alignment with the new fd. 315572373e40SVladimir Sementsov-Ogievskiy */ 315672373e40SVladimir Sementsov-Ogievskiy raw_probe_alignment(bs, ret, &local_err); 315772373e40SVladimir Sementsov-Ogievskiy if (local_err) { 315872373e40SVladimir Sementsov-Ogievskiy error_propagate(errp, local_err); 315972373e40SVladimir Sementsov-Ogievskiy return -EINVAL; 316072373e40SVladimir Sementsov-Ogievskiy } 316172373e40SVladimir Sementsov-Ogievskiy 31626ceabe6fSKevin Wolf s->perm_change_fd = ret; 3163094e3639SMax Reitz s->perm_change_flags = open_flags; 31646ceabe6fSKevin Wolf } 31656ceabe6fSKevin Wolf 31666ceabe6fSKevin Wolf /* Prepare permissions on old fd to avoid conflicts between old and new, 31676ceabe6fSKevin Wolf * but keep everything locked that new will need. */ 31686ceabe6fSKevin Wolf ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); 31696ceabe6fSKevin Wolf if (ret < 0) { 31706ceabe6fSKevin Wolf goto fail; 31716ceabe6fSKevin Wolf } 31726ceabe6fSKevin Wolf 31736ceabe6fSKevin Wolf /* Copy locks to the new fd */ 3174eb43ea16SLi Feng if (s->perm_change_fd && s->use_lock) { 31756ceabe6fSKevin Wolf ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, 31766ceabe6fSKevin Wolf false, errp); 31776ceabe6fSKevin Wolf if (ret < 0) { 31786ceabe6fSKevin Wolf raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 31796ceabe6fSKevin Wolf goto fail; 31806ceabe6fSKevin Wolf } 31816ceabe6fSKevin Wolf } 31826ceabe6fSKevin Wolf return 0; 31836ceabe6fSKevin Wolf 31846ceabe6fSKevin Wolf fail: 318572373e40SVladimir Sementsov-Ogievskiy if (s->perm_change_fd) { 31866ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 31876ceabe6fSKevin Wolf } 31886ceabe6fSKevin Wolf s->perm_change_fd = 0; 31896ceabe6fSKevin Wolf return ret; 3190244a5668SFam Zheng } 3191244a5668SFam Zheng 3192244a5668SFam Zheng static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) 3193244a5668SFam Zheng { 3194244a5668SFam Zheng BDRVRawState *s = bs->opaque; 31956ceabe6fSKevin Wolf 31966ceabe6fSKevin Wolf /* For reopen, we have already switched to the new fd (.bdrv_set_perm is 31976ceabe6fSKevin Wolf * called after .bdrv_reopen_commit) */ 31986ceabe6fSKevin Wolf if (s->perm_change_fd && s->fd != s->perm_change_fd) { 31996ceabe6fSKevin Wolf qemu_close(s->fd); 32006ceabe6fSKevin Wolf s->fd = s->perm_change_fd; 3201094e3639SMax Reitz s->open_flags = s->perm_change_flags; 32026ceabe6fSKevin Wolf } 32036ceabe6fSKevin Wolf s->perm_change_fd = 0; 32046ceabe6fSKevin Wolf 3205244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); 3206244a5668SFam Zheng s->perm = perm; 3207244a5668SFam Zheng s->shared_perm = shared; 3208244a5668SFam Zheng } 3209244a5668SFam Zheng 3210244a5668SFam Zheng static void raw_abort_perm_update(BlockDriverState *bs) 3211244a5668SFam Zheng { 32126ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 32136ceabe6fSKevin Wolf 32146ceabe6fSKevin Wolf /* For reopen, .bdrv_reopen_abort is called afterwards and will close 32156ceabe6fSKevin Wolf * the file descriptor. */ 321672373e40SVladimir Sementsov-Ogievskiy if (s->perm_change_fd) { 32176ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 32186ceabe6fSKevin Wolf } 32196ceabe6fSKevin Wolf s->perm_change_fd = 0; 32206ceabe6fSKevin Wolf 3221244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 3222244a5668SFam Zheng } 3223244a5668SFam Zheng 322467b51fb9SVladimir Sementsov-Ogievskiy static int coroutine_fn raw_co_copy_range_from( 322548535049SVladimir Sementsov-Ogievskiy BlockDriverState *bs, BdrvChild *src, int64_t src_offset, 322648535049SVladimir Sementsov-Ogievskiy BdrvChild *dst, int64_t dst_offset, int64_t bytes, 322767b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) 32281efad060SFam Zheng { 322967b51fb9SVladimir Sementsov-Ogievskiy return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, 323067b51fb9SVladimir Sementsov-Ogievskiy read_flags, write_flags); 32311efad060SFam Zheng } 32321efad060SFam Zheng 32331efad060SFam Zheng static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, 323467b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *src, 323548535049SVladimir Sementsov-Ogievskiy int64_t src_offset, 323667b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, 323748535049SVladimir Sementsov-Ogievskiy int64_t dst_offset, 323848535049SVladimir Sementsov-Ogievskiy int64_t bytes, 323967b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, 324067b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags write_flags) 32411efad060SFam Zheng { 324258a209c4SKevin Wolf RawPosixAIOData acb; 32431efad060SFam Zheng BDRVRawState *s = bs->opaque; 32441efad060SFam Zheng BDRVRawState *src_s; 32451efad060SFam Zheng 32461efad060SFam Zheng assert(dst->bs == bs); 32471efad060SFam Zheng if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { 32481efad060SFam Zheng return -ENOTSUP; 32491efad060SFam Zheng } 32501efad060SFam Zheng 32511efad060SFam Zheng src_s = src->bs->opaque; 32529f850f67SFam Zheng if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { 32531efad060SFam Zheng return -EIO; 32541efad060SFam Zheng } 325558a209c4SKevin Wolf 325658a209c4SKevin Wolf acb = (RawPosixAIOData) { 325758a209c4SKevin Wolf .bs = bs, 325858a209c4SKevin Wolf .aio_type = QEMU_AIO_COPY_RANGE, 325958a209c4SKevin Wolf .aio_fildes = src_s->fd, 326058a209c4SKevin Wolf .aio_offset = src_offset, 326158a209c4SKevin Wolf .aio_nbytes = bytes, 326258a209c4SKevin Wolf .copy_range = { 326358a209c4SKevin Wolf .aio_fd2 = s->fd, 326458a209c4SKevin Wolf .aio_offset2 = dst_offset, 326558a209c4SKevin Wolf }, 326658a209c4SKevin Wolf }; 326758a209c4SKevin Wolf 326858a209c4SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); 32691efad060SFam Zheng } 32701efad060SFam Zheng 3271c1bb86cdSEric Blake BlockDriver bdrv_file = { 3272c1bb86cdSEric Blake .format_name = "file", 3273c1bb86cdSEric Blake .protocol_name = "file", 3274c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3275c1bb86cdSEric Blake .bdrv_needs_filename = true, 3276c1bb86cdSEric Blake .bdrv_probe = NULL, /* no probe for protocols */ 3277c1bb86cdSEric Blake .bdrv_parse_filename = raw_parse_filename, 3278c1bb86cdSEric Blake .bdrv_file_open = raw_open, 3279c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3280c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3281c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3282c1bb86cdSEric Blake .bdrv_close = raw_close, 3283927f11e1SKevin Wolf .bdrv_co_create = raw_co_create, 3284efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = raw_co_create_opts, 3285c1bb86cdSEric Blake .bdrv_has_zero_init = bdrv_has_zero_init_1, 3286a290f085SEric Blake .bdrv_co_block_status = raw_co_block_status, 3287dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3288c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, 32899bffae14SDaniel Henrique Barboza .bdrv_co_delete_file = raw_co_delete_file, 3290c1bb86cdSEric Blake 3291c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3292c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 329333d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 329433d70fb6SKevin Wolf .bdrv_co_pdiscard = raw_co_pdiscard, 32951efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 32961efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 3297c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3298c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3299c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3300ed6e2161SNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3301c1bb86cdSEric Blake 3302061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3303c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3304c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 3305c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3306c1bb86cdSEric Blake = raw_get_allocated_file_size, 3307d9245599SAnton Nefedov .bdrv_get_specific_stats = raw_get_specific_stats, 3308244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 3309244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 3310244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 3311c1bb86cdSEric Blake .create_opts = &raw_create_opts, 33128a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3313c1bb86cdSEric Blake }; 3314c1bb86cdSEric Blake 3315c1bb86cdSEric Blake /***********************************************/ 3316c1bb86cdSEric Blake /* host device */ 3317c1bb86cdSEric Blake 331814176c8dSJoelle van Dyne #if defined(HAVE_HOST_BLOCK_DEVICE) 331914176c8dSJoelle van Dyne 3320c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3321c1bb86cdSEric Blake static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3322c1bb86cdSEric Blake CFIndex maxPathSize, int flags); 3323*aa44d3f6SPhilippe Mathieu-Daudé 3324*aa44d3f6SPhilippe Mathieu-Daudé #if !defined(MAC_OS_VERSION_12_0) \ 3325*aa44d3f6SPhilippe Mathieu-Daudé || (MAC_OS_X_VERSION_MIN_REQUIRED < MAC_OS_VERSION_12_0) 3326*aa44d3f6SPhilippe Mathieu-Daudé #define IOMainPort IOMasterPort 3327*aa44d3f6SPhilippe Mathieu-Daudé #endif 3328*aa44d3f6SPhilippe Mathieu-Daudé 3329c1bb86cdSEric Blake static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) 3330c1bb86cdSEric Blake { 3331c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 3332*aa44d3f6SPhilippe Mathieu-Daudé mach_port_t mainPort; 3333c1bb86cdSEric Blake CFMutableDictionaryRef classesToMatch; 3334c1bb86cdSEric Blake const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; 3335c1bb86cdSEric Blake char *mediaType = NULL; 3336c1bb86cdSEric Blake 3337*aa44d3f6SPhilippe Mathieu-Daudé kernResult = IOMainPort(MACH_PORT_NULL, &mainPort); 3338c1bb86cdSEric Blake if ( KERN_SUCCESS != kernResult ) { 3339*aa44d3f6SPhilippe Mathieu-Daudé printf("IOMainPort returned %d\n", kernResult); 3340c1bb86cdSEric Blake } 3341c1bb86cdSEric Blake 3342c1bb86cdSEric Blake int index; 3343c1bb86cdSEric Blake for (index = 0; index < ARRAY_SIZE(matching_array); index++) { 3344c1bb86cdSEric Blake classesToMatch = IOServiceMatching(matching_array[index]); 3345c1bb86cdSEric Blake if (classesToMatch == NULL) { 3346c1bb86cdSEric Blake error_report("IOServiceMatching returned NULL for %s", 3347c1bb86cdSEric Blake matching_array[index]); 3348c1bb86cdSEric Blake continue; 3349c1bb86cdSEric Blake } 3350c1bb86cdSEric Blake CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), 3351c1bb86cdSEric Blake kCFBooleanTrue); 3352*aa44d3f6SPhilippe Mathieu-Daudé kernResult = IOServiceGetMatchingServices(mainPort, classesToMatch, 3353c1bb86cdSEric Blake mediaIterator); 3354c1bb86cdSEric Blake if (kernResult != KERN_SUCCESS) { 3355c1bb86cdSEric Blake error_report("Note: IOServiceGetMatchingServices returned %d", 3356c1bb86cdSEric Blake kernResult); 3357c1bb86cdSEric Blake continue; 3358c1bb86cdSEric Blake } 3359c1bb86cdSEric Blake 3360c1bb86cdSEric Blake /* If a match was found, leave the loop */ 3361c1bb86cdSEric Blake if (*mediaIterator != 0) { 33624f7d28d7SLaurent Vivier trace_file_FindEjectableOpticalMedia(matching_array[index]); 3363c1bb86cdSEric Blake mediaType = g_strdup(matching_array[index]); 3364c1bb86cdSEric Blake break; 3365c1bb86cdSEric Blake } 3366c1bb86cdSEric Blake } 3367c1bb86cdSEric Blake return mediaType; 3368c1bb86cdSEric Blake } 3369c1bb86cdSEric Blake 3370c1bb86cdSEric Blake kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3371c1bb86cdSEric Blake CFIndex maxPathSize, int flags) 3372c1bb86cdSEric Blake { 3373c1bb86cdSEric Blake io_object_t nextMedia; 3374c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 3375c1bb86cdSEric Blake *bsdPath = '\0'; 3376c1bb86cdSEric Blake nextMedia = IOIteratorNext( mediaIterator ); 3377c1bb86cdSEric Blake if ( nextMedia ) 3378c1bb86cdSEric Blake { 3379c1bb86cdSEric Blake CFTypeRef bsdPathAsCFString; 3380c1bb86cdSEric Blake bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); 3381c1bb86cdSEric Blake if ( bsdPathAsCFString ) { 3382c1bb86cdSEric Blake size_t devPathLength; 3383c1bb86cdSEric Blake strcpy( bsdPath, _PATH_DEV ); 3384c1bb86cdSEric Blake if (flags & BDRV_O_NOCACHE) { 3385c1bb86cdSEric Blake strcat(bsdPath, "r"); 3386c1bb86cdSEric Blake } 3387c1bb86cdSEric Blake devPathLength = strlen( bsdPath ); 3388c1bb86cdSEric Blake if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { 3389c1bb86cdSEric Blake kernResult = KERN_SUCCESS; 3390c1bb86cdSEric Blake } 3391c1bb86cdSEric Blake CFRelease( bsdPathAsCFString ); 3392c1bb86cdSEric Blake } 3393c1bb86cdSEric Blake IOObjectRelease( nextMedia ); 3394c1bb86cdSEric Blake } 3395c1bb86cdSEric Blake 3396c1bb86cdSEric Blake return kernResult; 3397c1bb86cdSEric Blake } 3398c1bb86cdSEric Blake 3399c1bb86cdSEric Blake /* Sets up a real cdrom for use in QEMU */ 3400c1bb86cdSEric Blake static bool setup_cdrom(char *bsd_path, Error **errp) 3401c1bb86cdSEric Blake { 3402c1bb86cdSEric Blake int index, num_of_test_partitions = 2, fd; 3403c1bb86cdSEric Blake char test_partition[MAXPATHLEN]; 3404c1bb86cdSEric Blake bool partition_found = false; 3405c1bb86cdSEric Blake 3406c1bb86cdSEric Blake /* look for a working partition */ 3407c1bb86cdSEric Blake for (index = 0; index < num_of_test_partitions; index++) { 3408c1bb86cdSEric Blake snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, 3409c1bb86cdSEric Blake index); 3410b18a24a9SDaniel P. Berrangé fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE, NULL); 3411c1bb86cdSEric Blake if (fd >= 0) { 3412c1bb86cdSEric Blake partition_found = true; 3413c1bb86cdSEric Blake qemu_close(fd); 3414c1bb86cdSEric Blake break; 3415c1bb86cdSEric Blake } 3416c1bb86cdSEric Blake } 3417c1bb86cdSEric Blake 3418c1bb86cdSEric Blake /* if a working partition on the device was not found */ 3419c1bb86cdSEric Blake if (partition_found == false) { 3420c1bb86cdSEric Blake error_setg(errp, "Failed to find a working partition on disc"); 3421c1bb86cdSEric Blake } else { 34224f7d28d7SLaurent Vivier trace_file_setup_cdrom(test_partition); 3423c1bb86cdSEric Blake pstrcpy(bsd_path, MAXPATHLEN, test_partition); 3424c1bb86cdSEric Blake } 3425c1bb86cdSEric Blake return partition_found; 3426c1bb86cdSEric Blake } 3427c1bb86cdSEric Blake 3428c1bb86cdSEric Blake /* Prints directions on mounting and unmounting a device */ 3429c1bb86cdSEric Blake static void print_unmounting_directions(const char *file_name) 3430c1bb86cdSEric Blake { 3431c1bb86cdSEric Blake error_report("If device %s is mounted on the desktop, unmount" 3432c1bb86cdSEric Blake " it first before using it in QEMU", file_name); 3433c1bb86cdSEric Blake error_report("Command to unmount device: diskutil unmountDisk %s", 3434c1bb86cdSEric Blake file_name); 3435c1bb86cdSEric Blake error_report("Command to mount device: diskutil mountDisk %s", file_name); 3436c1bb86cdSEric Blake } 3437c1bb86cdSEric Blake 3438c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3439c1bb86cdSEric Blake 3440c1bb86cdSEric Blake static int hdev_probe_device(const char *filename) 3441c1bb86cdSEric Blake { 3442c1bb86cdSEric Blake struct stat st; 3443c1bb86cdSEric Blake 3444c1bb86cdSEric Blake /* allow a dedicated CD-ROM driver to match with a higher priority */ 3445c1bb86cdSEric Blake if (strstart(filename, "/dev/cdrom", NULL)) 3446c1bb86cdSEric Blake return 50; 3447c1bb86cdSEric Blake 3448c1bb86cdSEric Blake if (stat(filename, &st) >= 0 && 3449c1bb86cdSEric Blake (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 3450c1bb86cdSEric Blake return 100; 3451c1bb86cdSEric Blake } 3452c1bb86cdSEric Blake 3453c1bb86cdSEric Blake return 0; 3454c1bb86cdSEric Blake } 3455c1bb86cdSEric Blake 3456c1bb86cdSEric Blake static void hdev_parse_filename(const char *filename, QDict *options, 3457c1bb86cdSEric Blake Error **errp) 3458c1bb86cdSEric Blake { 345903c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_device:", options); 3460c1bb86cdSEric Blake } 3461c1bb86cdSEric Blake 3462c1bb86cdSEric Blake static bool hdev_is_sg(BlockDriverState *bs) 3463c1bb86cdSEric Blake { 3464c1bb86cdSEric Blake 3465c1bb86cdSEric Blake #if defined(__linux__) 3466c1bb86cdSEric Blake 3467c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3468c1bb86cdSEric Blake struct stat st; 3469c1bb86cdSEric Blake struct sg_scsi_id scsiid; 3470c1bb86cdSEric Blake int sg_version; 3471c1bb86cdSEric Blake int ret; 3472c1bb86cdSEric Blake 3473c1bb86cdSEric Blake if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { 3474c1bb86cdSEric Blake return false; 3475c1bb86cdSEric Blake } 3476c1bb86cdSEric Blake 3477c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); 3478c1bb86cdSEric Blake if (ret < 0) { 3479c1bb86cdSEric Blake return false; 3480c1bb86cdSEric Blake } 3481c1bb86cdSEric Blake 3482c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); 3483c1bb86cdSEric Blake if (ret >= 0) { 34844f7d28d7SLaurent Vivier trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); 3485c1bb86cdSEric Blake return true; 3486c1bb86cdSEric Blake } 3487c1bb86cdSEric Blake 3488c1bb86cdSEric Blake #endif 3489c1bb86cdSEric Blake 3490c1bb86cdSEric Blake return false; 3491c1bb86cdSEric Blake } 3492c1bb86cdSEric Blake 3493c1bb86cdSEric Blake static int hdev_open(BlockDriverState *bs, QDict *options, int flags, 3494c1bb86cdSEric Blake Error **errp) 3495c1bb86cdSEric Blake { 3496c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3497c1bb86cdSEric Blake int ret; 3498c1bb86cdSEric Blake 3499c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3500129c7d1cSMarkus Armbruster /* 3501129c7d1cSMarkus Armbruster * Caution: while qdict_get_str() is fine, getting non-string types 3502129c7d1cSMarkus Armbruster * would require more care. When @options come from -blockdev or 3503129c7d1cSMarkus Armbruster * blockdev_add, its members are typed according to the QAPI 3504129c7d1cSMarkus Armbruster * schema, but when they come from -drive, they're all QString. 3505129c7d1cSMarkus Armbruster */ 3506c1bb86cdSEric Blake const char *filename = qdict_get_str(options, "filename"); 3507c1bb86cdSEric Blake char bsd_path[MAXPATHLEN] = ""; 3508c1bb86cdSEric Blake bool error_occurred = false; 3509c1bb86cdSEric Blake 3510c1bb86cdSEric Blake /* If using a real cdrom */ 3511c1bb86cdSEric Blake if (strcmp(filename, "/dev/cdrom") == 0) { 3512c1bb86cdSEric Blake char *mediaType = NULL; 3513c1bb86cdSEric Blake kern_return_t ret_val; 3514c1bb86cdSEric Blake io_iterator_t mediaIterator = 0; 3515c1bb86cdSEric Blake 3516c1bb86cdSEric Blake mediaType = FindEjectableOpticalMedia(&mediaIterator); 3517c1bb86cdSEric Blake if (mediaType == NULL) { 3518c1bb86cdSEric Blake error_setg(errp, "Please make sure your CD/DVD is in the optical" 3519c1bb86cdSEric Blake " drive"); 3520c1bb86cdSEric Blake error_occurred = true; 3521c1bb86cdSEric Blake goto hdev_open_Mac_error; 3522c1bb86cdSEric Blake } 3523c1bb86cdSEric Blake 3524c1bb86cdSEric Blake ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); 3525c1bb86cdSEric Blake if (ret_val != KERN_SUCCESS) { 3526c1bb86cdSEric Blake error_setg(errp, "Could not get BSD path for optical drive"); 3527c1bb86cdSEric Blake error_occurred = true; 3528c1bb86cdSEric Blake goto hdev_open_Mac_error; 3529c1bb86cdSEric Blake } 3530c1bb86cdSEric Blake 3531c1bb86cdSEric Blake /* If a real optical drive was not found */ 3532c1bb86cdSEric Blake if (bsd_path[0] == '\0') { 3533c1bb86cdSEric Blake error_setg(errp, "Failed to obtain bsd path for optical drive"); 3534c1bb86cdSEric Blake error_occurred = true; 3535c1bb86cdSEric Blake goto hdev_open_Mac_error; 3536c1bb86cdSEric Blake } 3537c1bb86cdSEric Blake 3538c1bb86cdSEric Blake /* If using a cdrom disc and finding a partition on the disc failed */ 3539c1bb86cdSEric Blake if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && 3540c1bb86cdSEric Blake setup_cdrom(bsd_path, errp) == false) { 3541c1bb86cdSEric Blake print_unmounting_directions(bsd_path); 3542c1bb86cdSEric Blake error_occurred = true; 3543c1bb86cdSEric Blake goto hdev_open_Mac_error; 3544c1bb86cdSEric Blake } 3545c1bb86cdSEric Blake 354646f5ac20SEric Blake qdict_put_str(options, "filename", bsd_path); 3547c1bb86cdSEric Blake 3548c1bb86cdSEric Blake hdev_open_Mac_error: 3549c1bb86cdSEric Blake g_free(mediaType); 3550c1bb86cdSEric Blake if (mediaIterator) { 3551c1bb86cdSEric Blake IOObjectRelease(mediaIterator); 3552c1bb86cdSEric Blake } 3553c1bb86cdSEric Blake if (error_occurred) { 3554c1bb86cdSEric Blake return -ENOENT; 3555c1bb86cdSEric Blake } 3556c1bb86cdSEric Blake } 3557c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3558c1bb86cdSEric Blake 3559c1bb86cdSEric Blake s->type = FTYPE_FILE; 3560c1bb86cdSEric Blake 3561668f62ecSMarkus Armbruster ret = raw_open_common(bs, options, flags, 0, true, errp); 3562c1bb86cdSEric Blake if (ret < 0) { 3563c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3564c1bb86cdSEric Blake if (*bsd_path) { 3565c1bb86cdSEric Blake filename = bsd_path; 3566c1bb86cdSEric Blake } 3567c1bb86cdSEric Blake /* if a physical device experienced an error while being opened */ 3568c1bb86cdSEric Blake if (strncmp(filename, "/dev/", 5) == 0) { 3569c1bb86cdSEric Blake print_unmounting_directions(filename); 3570c1bb86cdSEric Blake } 3571c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3572c1bb86cdSEric Blake return ret; 3573c1bb86cdSEric Blake } 3574c1bb86cdSEric Blake 3575c1bb86cdSEric Blake /* Since this does ioctl the device must be already opened */ 3576c1bb86cdSEric Blake bs->sg = hdev_is_sg(bs); 3577c1bb86cdSEric Blake 3578c1bb86cdSEric Blake return ret; 3579c1bb86cdSEric Blake } 3580c1bb86cdSEric Blake 3581c1bb86cdSEric Blake #if defined(__linux__) 35822f3a7ab3SKevin Wolf static int coroutine_fn 35832f3a7ab3SKevin Wolf hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) 3584c1bb86cdSEric Blake { 3585c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 358603425671SKevin Wolf RawPosixAIOData acb; 35872f3a7ab3SKevin Wolf int ret; 3588c1bb86cdSEric Blake 35892f3a7ab3SKevin Wolf ret = fd_open(bs); 35902f3a7ab3SKevin Wolf if (ret < 0) { 35912f3a7ab3SKevin Wolf return ret; 35922f3a7ab3SKevin Wolf } 3593c1bb86cdSEric Blake 35947c9e5276SPaolo Bonzini if (req == SG_IO && s->pr_mgr) { 35957c9e5276SPaolo Bonzini struct sg_io_hdr *io_hdr = buf; 35967c9e5276SPaolo Bonzini if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || 35977c9e5276SPaolo Bonzini io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { 35987c9e5276SPaolo Bonzini return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), 35992f3a7ab3SKevin Wolf s->fd, io_hdr); 36007c9e5276SPaolo Bonzini } 36017c9e5276SPaolo Bonzini } 36027c9e5276SPaolo Bonzini 360303425671SKevin Wolf acb = (RawPosixAIOData) { 360403425671SKevin Wolf .bs = bs, 360503425671SKevin Wolf .aio_type = QEMU_AIO_IOCTL, 360603425671SKevin Wolf .aio_fildes = s->fd, 360703425671SKevin Wolf .aio_offset = 0, 360803425671SKevin Wolf .ioctl = { 360903425671SKevin Wolf .buf = buf, 361003425671SKevin Wolf .cmd = req, 361103425671SKevin Wolf }, 361203425671SKevin Wolf }; 361303425671SKevin Wolf 361403425671SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); 3615c1bb86cdSEric Blake } 3616c1bb86cdSEric Blake #endif /* linux */ 3617c1bb86cdSEric Blake 361833d70fb6SKevin Wolf static coroutine_fn int 36190c802287SVladimir Sementsov-Ogievskiy hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) 3620c1bb86cdSEric Blake { 36211c450366SAnton Nefedov BDRVRawState *s = bs->opaque; 362233d70fb6SKevin Wolf int ret; 3623c1bb86cdSEric Blake 362433d70fb6SKevin Wolf ret = fd_open(bs); 362533d70fb6SKevin Wolf if (ret < 0) { 36261c450366SAnton Nefedov raw_account_discard(s, bytes, ret); 362733d70fb6SKevin Wolf return ret; 3628c1bb86cdSEric Blake } 362946ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, true); 3630c1bb86cdSEric Blake } 3631c1bb86cdSEric Blake 3632c1bb86cdSEric Blake static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, 3633f34b2bcfSVladimir Sementsov-Ogievskiy int64_t offset, int64_t bytes, BdrvRequestFlags flags) 3634c1bb86cdSEric Blake { 3635c1bb86cdSEric Blake int rc; 3636c1bb86cdSEric Blake 3637c1bb86cdSEric Blake rc = fd_open(bs); 3638c1bb86cdSEric Blake if (rc < 0) { 3639c1bb86cdSEric Blake return rc; 3640c1bb86cdSEric Blake } 364134fa110eSKevin Wolf 36427154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); 3643c1bb86cdSEric Blake } 3644c1bb86cdSEric Blake 3645c1bb86cdSEric Blake static BlockDriver bdrv_host_device = { 3646c1bb86cdSEric Blake .format_name = "host_device", 3647c1bb86cdSEric Blake .protocol_name = "host_device", 3648c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3649c1bb86cdSEric Blake .bdrv_needs_filename = true, 3650c1bb86cdSEric Blake .bdrv_probe_device = hdev_probe_device, 3651c1bb86cdSEric Blake .bdrv_parse_filename = hdev_parse_filename, 3652c1bb86cdSEric Blake .bdrv_file_open = hdev_open, 3653c1bb86cdSEric Blake .bdrv_close = raw_close, 3654c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3655c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3656c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 36575a5e7f8cSMaxim Levitsky .bdrv_co_create_opts = bdrv_co_create_opts_simple, 36585a5e7f8cSMaxim Levitsky .create_opts = &bdrv_create_opts_simple, 36598a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3660dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3661c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, 3662c1bb86cdSEric Blake 3663c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3664c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 366533d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 366633d70fb6SKevin Wolf .bdrv_co_pdiscard = hdev_co_pdiscard, 36671efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 36681efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 3669c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3670c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3671c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3672042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3673c1bb86cdSEric Blake 3674061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3675c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3676c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 3677c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3678c1bb86cdSEric Blake = raw_get_allocated_file_size, 3679d9245599SAnton Nefedov .bdrv_get_specific_stats = hdev_get_specific_stats, 3680244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 3681244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 3682244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 3683c1bb86cdSEric Blake .bdrv_probe_blocksizes = hdev_probe_blocksizes, 3684c1bb86cdSEric Blake .bdrv_probe_geometry = hdev_probe_geometry, 3685c1bb86cdSEric Blake 3686c1bb86cdSEric Blake /* generic scsi device */ 3687c1bb86cdSEric Blake #ifdef __linux__ 36882f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3689c1bb86cdSEric Blake #endif 3690c1bb86cdSEric Blake }; 3691c1bb86cdSEric Blake 3692c1bb86cdSEric Blake #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3693c1bb86cdSEric Blake static void cdrom_parse_filename(const char *filename, QDict *options, 3694c1bb86cdSEric Blake Error **errp) 3695c1bb86cdSEric Blake { 369603c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); 3697c1bb86cdSEric Blake } 3698c1bb86cdSEric Blake #endif 3699c1bb86cdSEric Blake 3700c1bb86cdSEric Blake #ifdef __linux__ 3701c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3702c1bb86cdSEric Blake Error **errp) 3703c1bb86cdSEric Blake { 3704c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3705c1bb86cdSEric Blake 3706c1bb86cdSEric Blake s->type = FTYPE_CD; 3707c1bb86cdSEric Blake 3708c1bb86cdSEric Blake /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ 3709230ff739SJohn Snow return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); 3710c1bb86cdSEric Blake } 3711c1bb86cdSEric Blake 3712c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3713c1bb86cdSEric Blake { 3714c1bb86cdSEric Blake int fd, ret; 3715c1bb86cdSEric Blake int prio = 0; 3716c1bb86cdSEric Blake struct stat st; 3717c1bb86cdSEric Blake 3718b18a24a9SDaniel P. Berrangé fd = qemu_open(filename, O_RDONLY | O_NONBLOCK, NULL); 3719c1bb86cdSEric Blake if (fd < 0) { 3720c1bb86cdSEric Blake goto out; 3721c1bb86cdSEric Blake } 3722c1bb86cdSEric Blake ret = fstat(fd, &st); 3723c1bb86cdSEric Blake if (ret == -1 || !S_ISBLK(st.st_mode)) { 3724c1bb86cdSEric Blake goto outc; 3725c1bb86cdSEric Blake } 3726c1bb86cdSEric Blake 3727c1bb86cdSEric Blake /* Attempt to detect via a CDROM specific ioctl */ 3728c1bb86cdSEric Blake ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3729c1bb86cdSEric Blake if (ret >= 0) 3730c1bb86cdSEric Blake prio = 100; 3731c1bb86cdSEric Blake 3732c1bb86cdSEric Blake outc: 3733c1bb86cdSEric Blake qemu_close(fd); 3734c1bb86cdSEric Blake out: 3735c1bb86cdSEric Blake return prio; 3736c1bb86cdSEric Blake } 3737c1bb86cdSEric Blake 3738c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3739c1bb86cdSEric Blake { 3740c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3741c1bb86cdSEric Blake int ret; 3742c1bb86cdSEric Blake 3743c1bb86cdSEric Blake ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3744c1bb86cdSEric Blake return ret == CDS_DISC_OK; 3745c1bb86cdSEric Blake } 3746c1bb86cdSEric Blake 3747c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3748c1bb86cdSEric Blake { 3749c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3750c1bb86cdSEric Blake 3751c1bb86cdSEric Blake if (eject_flag) { 3752c1bb86cdSEric Blake if (ioctl(s->fd, CDROMEJECT, NULL) < 0) 3753c1bb86cdSEric Blake perror("CDROMEJECT"); 3754c1bb86cdSEric Blake } else { 3755c1bb86cdSEric Blake if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) 3756c1bb86cdSEric Blake perror("CDROMEJECT"); 3757c1bb86cdSEric Blake } 3758c1bb86cdSEric Blake } 3759c1bb86cdSEric Blake 3760c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3761c1bb86cdSEric Blake { 3762c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3763c1bb86cdSEric Blake 3764c1bb86cdSEric Blake if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { 3765c1bb86cdSEric Blake /* 3766c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3767c1bb86cdSEric Blake * mounts the CD-ROM 3768c1bb86cdSEric Blake */ 3769c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3770c1bb86cdSEric Blake } 3771c1bb86cdSEric Blake } 3772c1bb86cdSEric Blake 3773c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3774c1bb86cdSEric Blake .format_name = "host_cdrom", 3775c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3776c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3777c1bb86cdSEric Blake .bdrv_needs_filename = true, 3778c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3779c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3780c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3781c1bb86cdSEric Blake .bdrv_close = raw_close, 3782c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3783c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3784c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 37855a5e7f8cSMaxim Levitsky .bdrv_co_create_opts = bdrv_co_create_opts_simple, 37865a5e7f8cSMaxim Levitsky .create_opts = &bdrv_create_opts_simple, 37878a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3788dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3789c1bb86cdSEric Blake 3790c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3791c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 379233d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3793c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3794c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3795c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3796042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3797c1bb86cdSEric Blake 3798061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3799c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3800c1bb86cdSEric Blake .has_variable_length = true, 3801c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3802c1bb86cdSEric Blake = raw_get_allocated_file_size, 3803c1bb86cdSEric Blake 3804c1bb86cdSEric Blake /* removable device support */ 3805c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3806c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3807c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3808c1bb86cdSEric Blake 3809c1bb86cdSEric Blake /* generic scsi device */ 38102f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3811c1bb86cdSEric Blake }; 3812c1bb86cdSEric Blake #endif /* __linux__ */ 3813c1bb86cdSEric Blake 3814c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 3815c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3816c1bb86cdSEric Blake Error **errp) 3817c1bb86cdSEric Blake { 3818c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3819c1bb86cdSEric Blake int ret; 3820c1bb86cdSEric Blake 3821c1bb86cdSEric Blake s->type = FTYPE_CD; 3822c1bb86cdSEric Blake 3823668f62ecSMarkus Armbruster ret = raw_open_common(bs, options, flags, 0, true, errp); 3824c1bb86cdSEric Blake if (ret) { 3825c1bb86cdSEric Blake return ret; 3826c1bb86cdSEric Blake } 3827c1bb86cdSEric Blake 3828c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3829c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3830c1bb86cdSEric Blake return 0; 3831c1bb86cdSEric Blake } 3832c1bb86cdSEric Blake 3833c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3834c1bb86cdSEric Blake { 3835c1bb86cdSEric Blake if (strstart(filename, "/dev/cd", NULL) || 3836c1bb86cdSEric Blake strstart(filename, "/dev/acd", NULL)) 3837c1bb86cdSEric Blake return 100; 3838c1bb86cdSEric Blake return 0; 3839c1bb86cdSEric Blake } 3840c1bb86cdSEric Blake 3841c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs) 3842c1bb86cdSEric Blake { 3843c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3844c1bb86cdSEric Blake int fd; 3845c1bb86cdSEric Blake 3846c1bb86cdSEric Blake /* 3847c1bb86cdSEric Blake * Force reread of possibly changed/newly loaded disc, 3848c1bb86cdSEric Blake * FreeBSD seems to not notice sometimes... 3849c1bb86cdSEric Blake */ 3850c1bb86cdSEric Blake if (s->fd >= 0) 3851c1bb86cdSEric Blake qemu_close(s->fd); 3852b18a24a9SDaniel P. Berrangé fd = qemu_open(bs->filename, s->open_flags, NULL); 3853c1bb86cdSEric Blake if (fd < 0) { 3854c1bb86cdSEric Blake s->fd = -1; 3855c1bb86cdSEric Blake return -EIO; 3856c1bb86cdSEric Blake } 3857c1bb86cdSEric Blake s->fd = fd; 3858c1bb86cdSEric Blake 3859c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3860c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3861c1bb86cdSEric Blake return 0; 3862c1bb86cdSEric Blake } 3863c1bb86cdSEric Blake 3864c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3865c1bb86cdSEric Blake { 3866c1bb86cdSEric Blake return raw_getlength(bs) > 0; 3867c1bb86cdSEric Blake } 3868c1bb86cdSEric Blake 3869c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3870c1bb86cdSEric Blake { 3871c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3872c1bb86cdSEric Blake 3873c1bb86cdSEric Blake if (s->fd < 0) 3874c1bb86cdSEric Blake return; 3875c1bb86cdSEric Blake 3876c1bb86cdSEric Blake (void) ioctl(s->fd, CDIOCALLOW); 3877c1bb86cdSEric Blake 3878c1bb86cdSEric Blake if (eject_flag) { 3879c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCEJECT) < 0) 3880c1bb86cdSEric Blake perror("CDIOCEJECT"); 3881c1bb86cdSEric Blake } else { 3882c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCCLOSE) < 0) 3883c1bb86cdSEric Blake perror("CDIOCCLOSE"); 3884c1bb86cdSEric Blake } 3885c1bb86cdSEric Blake 3886c1bb86cdSEric Blake cdrom_reopen(bs); 3887c1bb86cdSEric Blake } 3888c1bb86cdSEric Blake 3889c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3890c1bb86cdSEric Blake { 3891c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3892c1bb86cdSEric Blake 3893c1bb86cdSEric Blake if (s->fd < 0) 3894c1bb86cdSEric Blake return; 3895c1bb86cdSEric Blake if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { 3896c1bb86cdSEric Blake /* 3897c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3898c1bb86cdSEric Blake * mounts the CD-ROM 3899c1bb86cdSEric Blake */ 3900c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3901c1bb86cdSEric Blake } 3902c1bb86cdSEric Blake } 3903c1bb86cdSEric Blake 3904c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3905c1bb86cdSEric Blake .format_name = "host_cdrom", 3906c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3907c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3908c1bb86cdSEric Blake .bdrv_needs_filename = true, 3909c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3910c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3911c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3912c1bb86cdSEric Blake .bdrv_close = raw_close, 3913c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3914c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3915c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 39165a5e7f8cSMaxim Levitsky .bdrv_co_create_opts = bdrv_co_create_opts_simple, 39175a5e7f8cSMaxim Levitsky .create_opts = &bdrv_create_opts_simple, 39188a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3919c1bb86cdSEric Blake 3920c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3921c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 392233d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3923c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3924c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3925c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3926042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3927c1bb86cdSEric Blake 3928061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3929c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3930c1bb86cdSEric Blake .has_variable_length = true, 3931c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3932c1bb86cdSEric Blake = raw_get_allocated_file_size, 3933c1bb86cdSEric Blake 3934c1bb86cdSEric Blake /* removable device support */ 3935c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3936c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3937c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3938c1bb86cdSEric Blake }; 3939c1bb86cdSEric Blake #endif /* __FreeBSD__ */ 3940c1bb86cdSEric Blake 394114176c8dSJoelle van Dyne #endif /* HAVE_HOST_BLOCK_DEVICE */ 394214176c8dSJoelle van Dyne 3943c1bb86cdSEric Blake static void bdrv_file_init(void) 3944c1bb86cdSEric Blake { 3945c1bb86cdSEric Blake /* 3946c1bb86cdSEric Blake * Register all the drivers. Note that order is important, the driver 3947c1bb86cdSEric Blake * registered last will get probed first. 3948c1bb86cdSEric Blake */ 3949c1bb86cdSEric Blake bdrv_register(&bdrv_file); 395014176c8dSJoelle van Dyne #if defined(HAVE_HOST_BLOCK_DEVICE) 3951c1bb86cdSEric Blake bdrv_register(&bdrv_host_device); 3952c1bb86cdSEric Blake #ifdef __linux__ 3953c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3954c1bb86cdSEric Blake #endif 3955c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3956c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3957c1bb86cdSEric Blake #endif 395814176c8dSJoelle van Dyne #endif /* HAVE_HOST_BLOCK_DEVICE */ 3959c1bb86cdSEric Blake } 3960c1bb86cdSEric Blake 3961c1bb86cdSEric Blake block_init(bdrv_file_init); 3962