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" 34c1bb86cdSEric Blake #include "trace.h" 35c1bb86cdSEric Blake #include "block/thread-pool.h" 36c1bb86cdSEric Blake #include "qemu/iov.h" 37c1bb86cdSEric Blake #include "block/raw-aio.h" 38452fcdbcSMarkus Armbruster #include "qapi/qmp/qdict.h" 39c1bb86cdSEric Blake #include "qapi/qmp/qstring.h" 40c1bb86cdSEric Blake 417c9e5276SPaolo Bonzini #include "scsi/pr-manager.h" 427c9e5276SPaolo Bonzini #include "scsi/constants.h" 437c9e5276SPaolo Bonzini 44c1bb86cdSEric Blake #if defined(__APPLE__) && (__MACH__) 45c1bb86cdSEric Blake #include <paths.h> 46c1bb86cdSEric Blake #include <sys/param.h> 47c1bb86cdSEric Blake #include <IOKit/IOKitLib.h> 48c1bb86cdSEric Blake #include <IOKit/IOBSD.h> 49c1bb86cdSEric Blake #include <IOKit/storage/IOMediaBSDClient.h> 50c1bb86cdSEric Blake #include <IOKit/storage/IOMedia.h> 51c1bb86cdSEric Blake #include <IOKit/storage/IOCDMedia.h> 52c1bb86cdSEric Blake //#include <IOKit/storage/IOCDTypes.h> 53c1bb86cdSEric Blake #include <IOKit/storage/IODVDMedia.h> 54c1bb86cdSEric Blake #include <CoreFoundation/CoreFoundation.h> 55c1bb86cdSEric Blake #endif 56c1bb86cdSEric Blake 57c1bb86cdSEric Blake #ifdef __sun__ 58c1bb86cdSEric Blake #define _POSIX_PTHREAD_SEMANTICS 1 59c1bb86cdSEric Blake #include <sys/dkio.h> 60c1bb86cdSEric Blake #endif 61c1bb86cdSEric Blake #ifdef __linux__ 62c1bb86cdSEric Blake #include <sys/ioctl.h> 63c1bb86cdSEric Blake #include <sys/param.h> 641efad060SFam Zheng #include <sys/syscall.h> 655edc8557SKevin Wolf #include <sys/vfs.h> 66c1bb86cdSEric Blake #include <linux/cdrom.h> 67c1bb86cdSEric Blake #include <linux/fd.h> 68c1bb86cdSEric Blake #include <linux/fs.h> 69c1bb86cdSEric Blake #include <linux/hdreg.h> 705edc8557SKevin Wolf #include <linux/magic.h> 71c1bb86cdSEric Blake #include <scsi/sg.h> 72c1bb86cdSEric Blake #ifdef __s390__ 73c1bb86cdSEric Blake #include <asm/dasd.h> 74c1bb86cdSEric Blake #endif 75c1bb86cdSEric Blake #ifndef FS_NOCOW_FL 76c1bb86cdSEric Blake #define FS_NOCOW_FL 0x00800000 /* Do not cow file */ 77c1bb86cdSEric Blake #endif 78c1bb86cdSEric Blake #endif 79c1bb86cdSEric Blake #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE) 80c1bb86cdSEric Blake #include <linux/falloc.h> 81c1bb86cdSEric Blake #endif 82c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 83c1bb86cdSEric Blake #include <sys/disk.h> 84c1bb86cdSEric Blake #include <sys/cdio.h> 85c1bb86cdSEric Blake #endif 86c1bb86cdSEric Blake 87c1bb86cdSEric Blake #ifdef __OpenBSD__ 88c1bb86cdSEric Blake #include <sys/ioctl.h> 89c1bb86cdSEric Blake #include <sys/disklabel.h> 90c1bb86cdSEric Blake #include <sys/dkio.h> 91c1bb86cdSEric Blake #endif 92c1bb86cdSEric Blake 93c1bb86cdSEric Blake #ifdef __NetBSD__ 94c1bb86cdSEric Blake #include <sys/ioctl.h> 95c1bb86cdSEric Blake #include <sys/disklabel.h> 96c1bb86cdSEric Blake #include <sys/dkio.h> 97c1bb86cdSEric Blake #include <sys/disk.h> 98c1bb86cdSEric Blake #endif 99c1bb86cdSEric Blake 100c1bb86cdSEric Blake #ifdef __DragonFly__ 101c1bb86cdSEric Blake #include <sys/ioctl.h> 102c1bb86cdSEric Blake #include <sys/diskslice.h> 103c1bb86cdSEric Blake #endif 104c1bb86cdSEric Blake 105c1bb86cdSEric Blake #ifdef CONFIG_XFS 106c1bb86cdSEric Blake #include <xfs/xfs.h> 107c1bb86cdSEric Blake #endif 108c1bb86cdSEric Blake 1094f7d28d7SLaurent Vivier #include "trace.h" 110c1bb86cdSEric Blake 111c1bb86cdSEric Blake /* OS X does not have O_DSYNC */ 112c1bb86cdSEric Blake #ifndef O_DSYNC 113c1bb86cdSEric Blake #ifdef O_SYNC 114c1bb86cdSEric Blake #define O_DSYNC O_SYNC 115c1bb86cdSEric Blake #elif defined(O_FSYNC) 116c1bb86cdSEric Blake #define O_DSYNC O_FSYNC 117c1bb86cdSEric Blake #endif 118c1bb86cdSEric Blake #endif 119c1bb86cdSEric Blake 120c1bb86cdSEric Blake /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ 121c1bb86cdSEric Blake #ifndef O_DIRECT 122c1bb86cdSEric Blake #define O_DIRECT O_DSYNC 123c1bb86cdSEric Blake #endif 124c1bb86cdSEric Blake 125c1bb86cdSEric Blake #define FTYPE_FILE 0 126c1bb86cdSEric Blake #define FTYPE_CD 1 127c1bb86cdSEric Blake 128c1bb86cdSEric Blake #define MAX_BLOCKSIZE 4096 129c1bb86cdSEric Blake 130244a5668SFam Zheng /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes, 131244a5668SFam Zheng * leaving a few more bytes for its future use. */ 132244a5668SFam Zheng #define RAW_LOCK_PERM_BASE 100 133244a5668SFam Zheng #define RAW_LOCK_SHARED_BASE 200 134244a5668SFam Zheng 135c1bb86cdSEric Blake typedef struct BDRVRawState { 136c1bb86cdSEric Blake int fd; 137244a5668SFam Zheng bool use_lock; 138c1bb86cdSEric Blake int type; 139c1bb86cdSEric Blake int open_flags; 140c1bb86cdSEric Blake size_t buf_align; 141c1bb86cdSEric Blake 142244a5668SFam Zheng /* The current permissions. */ 143244a5668SFam Zheng uint64_t perm; 144244a5668SFam Zheng uint64_t shared_perm; 145244a5668SFam Zheng 1462996ffadSFam Zheng /* The perms bits whose corresponding bytes are already locked in 147f2e3af29SFam Zheng * s->fd. */ 1482996ffadSFam Zheng uint64_t locked_perm; 1492996ffadSFam Zheng uint64_t locked_shared_perm; 1502996ffadSFam Zheng 1516ceabe6fSKevin Wolf int perm_change_fd; 152094e3639SMax Reitz int perm_change_flags; 153e0c9cf3aSKevin Wolf BDRVReopenState *reopen_state; 154e0c9cf3aSKevin Wolf 155c1bb86cdSEric Blake #ifdef CONFIG_XFS 156c1bb86cdSEric Blake bool is_xfs:1; 157c1bb86cdSEric Blake #endif 158c1bb86cdSEric Blake bool has_discard:1; 159c1bb86cdSEric Blake bool has_write_zeroes:1; 160c1bb86cdSEric Blake bool discard_zeroes:1; 161c1bb86cdSEric Blake bool use_linux_aio:1; 162c6447510SAarushi Mehta bool use_linux_io_uring:1; 163e5bcf967SKevin Wolf bool page_cache_inconsistent:1; 164c1bb86cdSEric Blake bool has_fallocate; 165c1bb86cdSEric Blake bool needs_alignment; 166f357fcd8SStefan Hajnoczi bool drop_cache; 16731be8a2aSStefan Hajnoczi bool check_cache_dropped; 1681c450366SAnton Nefedov struct { 1691c450366SAnton Nefedov uint64_t discard_nb_ok; 1701c450366SAnton Nefedov uint64_t discard_nb_failed; 1711c450366SAnton Nefedov uint64_t discard_bytes_ok; 1721c450366SAnton Nefedov } stats; 1737c9e5276SPaolo Bonzini 1747c9e5276SPaolo Bonzini PRManager *pr_mgr; 175c1bb86cdSEric Blake } BDRVRawState; 176c1bb86cdSEric Blake 177c1bb86cdSEric Blake typedef struct BDRVRawReopenState { 178c1bb86cdSEric Blake int fd; 179c1bb86cdSEric Blake int open_flags; 180f357fcd8SStefan Hajnoczi bool drop_cache; 18131be8a2aSStefan Hajnoczi bool check_cache_dropped; 182c1bb86cdSEric Blake } BDRVRawReopenState; 183c1bb86cdSEric Blake 184c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs); 185c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs); 186c1bb86cdSEric Blake 187c1bb86cdSEric Blake typedef struct RawPosixAIOData { 188c1bb86cdSEric Blake BlockDriverState *bs; 189c1bb86cdSEric Blake int aio_type; 190d57c44d0SKevin Wolf int aio_fildes; 191d57c44d0SKevin Wolf 192d57c44d0SKevin Wolf off_t aio_offset; 193d57c44d0SKevin Wolf uint64_t aio_nbytes; 194d57c44d0SKevin Wolf 19593f4e2ffSKevin Wolf union { 19693f4e2ffSKevin Wolf struct { 197d57c44d0SKevin Wolf struct iovec *iov; 198d57c44d0SKevin Wolf int niov; 199d57c44d0SKevin Wolf } io; 200d57c44d0SKevin Wolf struct { 201d57c44d0SKevin Wolf uint64_t cmd; 202d57c44d0SKevin Wolf void *buf; 203d57c44d0SKevin Wolf } ioctl; 204d57c44d0SKevin Wolf struct { 2051efad060SFam Zheng int aio_fd2; 2061efad060SFam Zheng off_t aio_offset2; 207d57c44d0SKevin Wolf } copy_range; 20893f4e2ffSKevin Wolf struct { 20993f4e2ffSKevin Wolf PreallocMode prealloc; 21093f4e2ffSKevin Wolf Error **errp; 211d57c44d0SKevin Wolf } truncate; 21293f4e2ffSKevin Wolf }; 213c1bb86cdSEric Blake } RawPosixAIOData; 214c1bb86cdSEric Blake 215c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 216c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs); 217c1bb86cdSEric Blake #endif 218c1bb86cdSEric Blake 219*797e3e38SDavid Edmondson /* 220*797e3e38SDavid Edmondson * Elide EAGAIN and EACCES details when failing to lock, as this 221*797e3e38SDavid Edmondson * indicates that the specified file region is already locked by 222*797e3e38SDavid Edmondson * another process, which is considered a common scenario. 223*797e3e38SDavid Edmondson */ 224*797e3e38SDavid Edmondson #define raw_lock_error_setg_errno(errp, err, fmt, ...) \ 225*797e3e38SDavid Edmondson do { \ 226*797e3e38SDavid Edmondson if ((err) == EAGAIN || (err) == EACCES) { \ 227*797e3e38SDavid Edmondson error_setg((errp), (fmt), ## __VA_ARGS__); \ 228*797e3e38SDavid Edmondson } else { \ 229*797e3e38SDavid Edmondson error_setg_errno((errp), (err), (fmt), ## __VA_ARGS__); \ 230*797e3e38SDavid Edmondson } \ 231*797e3e38SDavid Edmondson } while (0) 232*797e3e38SDavid Edmondson 233c1bb86cdSEric Blake #if defined(__NetBSD__) 234db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 235c1bb86cdSEric Blake { 236c1bb86cdSEric Blake static char namebuf[PATH_MAX]; 237c1bb86cdSEric Blake const char *dp, *fname; 238c1bb86cdSEric Blake struct stat sb; 239c1bb86cdSEric Blake 240c1bb86cdSEric Blake fname = *filename; 241c1bb86cdSEric Blake dp = strrchr(fname, '/'); 242c1bb86cdSEric Blake if (lstat(fname, &sb) < 0) { 243f6fc1e30SPaolo Bonzini error_setg_file_open(errp, errno, fname); 244c1bb86cdSEric Blake return -errno; 245c1bb86cdSEric Blake } 246c1bb86cdSEric Blake 247c1bb86cdSEric Blake if (!S_ISBLK(sb.st_mode)) { 248c1bb86cdSEric Blake return 0; 249c1bb86cdSEric Blake } 250c1bb86cdSEric Blake 251c1bb86cdSEric Blake if (dp == NULL) { 252c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "r%s", fname); 253c1bb86cdSEric Blake } else { 254c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "%.*s/r%s", 255c1bb86cdSEric Blake (int)(dp - fname), fname, dp + 1); 256c1bb86cdSEric Blake } 257c1bb86cdSEric Blake *filename = namebuf; 258db0754dfSFam Zheng warn_report("%s is a block device, using %s", fname, *filename); 259c1bb86cdSEric Blake 260c1bb86cdSEric Blake return 0; 261c1bb86cdSEric Blake } 262c1bb86cdSEric Blake #else 263db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 264c1bb86cdSEric Blake { 265c1bb86cdSEric Blake return 0; 266c1bb86cdSEric Blake } 267c1bb86cdSEric Blake #endif 268c1bb86cdSEric Blake 269c1bb86cdSEric Blake /* 270c1bb86cdSEric Blake * Get logical block size via ioctl. On success store it in @sector_size_p. 271c1bb86cdSEric Blake */ 272c1bb86cdSEric Blake static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) 273c1bb86cdSEric Blake { 274c1bb86cdSEric Blake unsigned int sector_size; 275c1bb86cdSEric Blake bool success = false; 276700f9ce0SPeter Maydell int i; 277c1bb86cdSEric Blake 278c1bb86cdSEric Blake errno = ENOTSUP; 279700f9ce0SPeter Maydell static const unsigned long ioctl_list[] = { 280c1bb86cdSEric Blake #ifdef BLKSSZGET 281700f9ce0SPeter Maydell BLKSSZGET, 282c1bb86cdSEric Blake #endif 283c1bb86cdSEric Blake #ifdef DKIOCGETBLOCKSIZE 284700f9ce0SPeter Maydell DKIOCGETBLOCKSIZE, 285c1bb86cdSEric Blake #endif 286c1bb86cdSEric Blake #ifdef DIOCGSECTORSIZE 287700f9ce0SPeter Maydell DIOCGSECTORSIZE, 288700f9ce0SPeter Maydell #endif 289700f9ce0SPeter Maydell }; 290700f9ce0SPeter Maydell 291700f9ce0SPeter Maydell /* Try a few ioctls to get the right size */ 292700f9ce0SPeter Maydell for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { 293700f9ce0SPeter Maydell if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { 294c1bb86cdSEric Blake *sector_size_p = sector_size; 295c1bb86cdSEric Blake success = true; 296c1bb86cdSEric Blake } 297700f9ce0SPeter Maydell } 298c1bb86cdSEric Blake 299c1bb86cdSEric Blake return success ? 0 : -errno; 300c1bb86cdSEric Blake } 301c1bb86cdSEric Blake 302c1bb86cdSEric Blake /** 303c1bb86cdSEric Blake * Get physical block size of @fd. 304c1bb86cdSEric Blake * On success, store it in @blk_size and return 0. 305c1bb86cdSEric Blake * On failure, return -errno. 306c1bb86cdSEric Blake */ 307c1bb86cdSEric Blake static int probe_physical_blocksize(int fd, unsigned int *blk_size) 308c1bb86cdSEric Blake { 309c1bb86cdSEric Blake #ifdef BLKPBSZGET 310c1bb86cdSEric Blake if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { 311c1bb86cdSEric Blake return -errno; 312c1bb86cdSEric Blake } 313c1bb86cdSEric Blake return 0; 314c1bb86cdSEric Blake #else 315c1bb86cdSEric Blake return -ENOTSUP; 316c1bb86cdSEric Blake #endif 317c1bb86cdSEric Blake } 318c1bb86cdSEric Blake 3195edc8557SKevin Wolf /* 3205edc8557SKevin Wolf * Returns true if no alignment restrictions are necessary even for files 3215edc8557SKevin Wolf * opened with O_DIRECT. 3225edc8557SKevin Wolf * 3235edc8557SKevin Wolf * raw_probe_alignment() probes the required alignment and assume that 1 means 3245edc8557SKevin Wolf * the probing failed, so it falls back to a safe default of 4k. This can be 3255edc8557SKevin Wolf * avoided if we know that byte alignment is okay for the file. 3265edc8557SKevin Wolf */ 3275edc8557SKevin Wolf static bool dio_byte_aligned(int fd) 3285edc8557SKevin Wolf { 3295edc8557SKevin Wolf #ifdef __linux__ 3305edc8557SKevin Wolf struct statfs buf; 3315edc8557SKevin Wolf int ret; 3325edc8557SKevin Wolf 3335edc8557SKevin Wolf ret = fstatfs(fd, &buf); 3345edc8557SKevin Wolf if (ret == 0 && buf.f_type == NFS_SUPER_MAGIC) { 3355edc8557SKevin Wolf return true; 3365edc8557SKevin Wolf } 3375edc8557SKevin Wolf #endif 3385edc8557SKevin Wolf return false; 3395edc8557SKevin Wolf } 3405edc8557SKevin Wolf 341c1bb86cdSEric Blake /* Check if read is allowed with given memory buffer and length. 342c1bb86cdSEric Blake * 343c1bb86cdSEric Blake * This function is used to check O_DIRECT memory buffer and request alignment. 344c1bb86cdSEric Blake */ 345c1bb86cdSEric Blake static bool raw_is_io_aligned(int fd, void *buf, size_t len) 346c1bb86cdSEric Blake { 347c1bb86cdSEric Blake ssize_t ret = pread(fd, buf, len, 0); 348c1bb86cdSEric Blake 349c1bb86cdSEric Blake if (ret >= 0) { 350c1bb86cdSEric Blake return true; 351c1bb86cdSEric Blake } 352c1bb86cdSEric Blake 353c1bb86cdSEric Blake #ifdef __linux__ 354c1bb86cdSEric Blake /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore 355c1bb86cdSEric Blake * other errors (e.g. real I/O error), which could happen on a failed 356c1bb86cdSEric Blake * drive, since we only care about probing alignment. 357c1bb86cdSEric Blake */ 358c1bb86cdSEric Blake if (errno != EINVAL) { 359c1bb86cdSEric Blake return true; 360c1bb86cdSEric Blake } 361c1bb86cdSEric Blake #endif 362c1bb86cdSEric Blake 363c1bb86cdSEric Blake return false; 364c1bb86cdSEric Blake } 365c1bb86cdSEric Blake 366c1bb86cdSEric Blake static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) 367c1bb86cdSEric Blake { 368c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 369c1bb86cdSEric Blake char *buf; 370038adc2fSWei Yang size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); 371a6b257a0SNir Soffer size_t alignments[] = {1, 512, 1024, 2048, 4096}; 372c1bb86cdSEric Blake 373c1bb86cdSEric Blake /* For SCSI generic devices the alignment is not really used. 374c1bb86cdSEric Blake With buffered I/O, we don't have any restrictions. */ 375c1bb86cdSEric Blake if (bdrv_is_sg(bs) || !s->needs_alignment) { 376c1bb86cdSEric Blake bs->bl.request_alignment = 1; 377c1bb86cdSEric Blake s->buf_align = 1; 378c1bb86cdSEric Blake return; 379c1bb86cdSEric Blake } 380c1bb86cdSEric Blake 381c1bb86cdSEric Blake bs->bl.request_alignment = 0; 382c1bb86cdSEric Blake s->buf_align = 0; 383c1bb86cdSEric Blake /* Let's try to use the logical blocksize for the alignment. */ 384c1bb86cdSEric Blake if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { 385c1bb86cdSEric Blake bs->bl.request_alignment = 0; 386c1bb86cdSEric Blake } 387c1bb86cdSEric Blake #ifdef CONFIG_XFS 388c1bb86cdSEric Blake if (s->is_xfs) { 389c1bb86cdSEric Blake struct dioattr da; 390c1bb86cdSEric Blake if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { 391c1bb86cdSEric Blake bs->bl.request_alignment = da.d_miniosz; 392c1bb86cdSEric Blake /* The kernel returns wrong information for d_mem */ 393c1bb86cdSEric Blake /* s->buf_align = da.d_mem; */ 394c1bb86cdSEric Blake } 395c1bb86cdSEric Blake } 396c1bb86cdSEric Blake #endif 397c1bb86cdSEric Blake 398a6b257a0SNir Soffer /* 399a6b257a0SNir Soffer * If we could not get the sizes so far, we can only guess them. First try 400a6b257a0SNir Soffer * to detect request alignment, since it is more likely to succeed. Then 401a6b257a0SNir Soffer * try to detect buf_align, which cannot be detected in some cases (e.g. 402a6b257a0SNir Soffer * Gluster). If buf_align cannot be detected, we fallback to the value of 403a6b257a0SNir Soffer * request_alignment. 404a6b257a0SNir Soffer */ 405a6b257a0SNir Soffer 406a6b257a0SNir Soffer if (!bs->bl.request_alignment) { 407a6b257a0SNir Soffer int i; 408c1bb86cdSEric Blake size_t align; 409a6b257a0SNir Soffer buf = qemu_memalign(max_align, max_align); 410a6b257a0SNir Soffer for (i = 0; i < ARRAY_SIZE(alignments); i++) { 411a6b257a0SNir Soffer align = alignments[i]; 412a6b257a0SNir Soffer if (raw_is_io_aligned(fd, buf, align)) { 413a6b257a0SNir Soffer /* Fallback to safe value. */ 414a6b257a0SNir Soffer bs->bl.request_alignment = (align != 1) ? align : max_align; 415c1bb86cdSEric Blake break; 416c1bb86cdSEric Blake } 417c1bb86cdSEric Blake } 418c1bb86cdSEric Blake qemu_vfree(buf); 419c1bb86cdSEric Blake } 420c1bb86cdSEric Blake 421a6b257a0SNir Soffer if (!s->buf_align) { 422a6b257a0SNir Soffer int i; 423c1bb86cdSEric Blake size_t align; 424a6b257a0SNir Soffer buf = qemu_memalign(max_align, 2 * max_align); 425a6b257a0SNir Soffer for (i = 0; i < ARRAY_SIZE(alignments); i++) { 426a6b257a0SNir Soffer align = alignments[i]; 427a6b257a0SNir Soffer if (raw_is_io_aligned(fd, buf + align, max_align)) { 428236094c7SStefan Hajnoczi /* Fallback to request_alignment. */ 429a6b257a0SNir Soffer s->buf_align = (align != 1) ? align : bs->bl.request_alignment; 430c1bb86cdSEric Blake break; 431c1bb86cdSEric Blake } 432c1bb86cdSEric Blake } 433c1bb86cdSEric Blake qemu_vfree(buf); 434c1bb86cdSEric Blake } 435c1bb86cdSEric Blake 436c1bb86cdSEric Blake if (!s->buf_align || !bs->bl.request_alignment) { 437c1bb86cdSEric Blake error_setg(errp, "Could not find working O_DIRECT alignment"); 438c1bb86cdSEric Blake error_append_hint(errp, "Try cache.direct=off\n"); 439c1bb86cdSEric Blake } 440c1bb86cdSEric Blake } 441c1bb86cdSEric Blake 442bca5283bSKevin Wolf static int check_hdev_writable(int fd) 44320eaf1bfSKevin Wolf { 44420eaf1bfSKevin Wolf #if defined(BLKROGET) 44520eaf1bfSKevin Wolf /* Linux block devices can be configured "read-only" using blockdev(8). 44620eaf1bfSKevin Wolf * This is independent of device node permissions and therefore open(2) 44720eaf1bfSKevin Wolf * with O_RDWR succeeds. Actual writes fail with EPERM. 44820eaf1bfSKevin Wolf * 44920eaf1bfSKevin Wolf * bdrv_open() is supposed to fail if the disk is read-only. Explicitly 45020eaf1bfSKevin Wolf * check for read-only block devices so that Linux block devices behave 45120eaf1bfSKevin Wolf * properly. 45220eaf1bfSKevin Wolf */ 45320eaf1bfSKevin Wolf struct stat st; 45420eaf1bfSKevin Wolf int readonly = 0; 45520eaf1bfSKevin Wolf 456bca5283bSKevin Wolf if (fstat(fd, &st)) { 45720eaf1bfSKevin Wolf return -errno; 45820eaf1bfSKevin Wolf } 45920eaf1bfSKevin Wolf 46020eaf1bfSKevin Wolf if (!S_ISBLK(st.st_mode)) { 46120eaf1bfSKevin Wolf return 0; 46220eaf1bfSKevin Wolf } 46320eaf1bfSKevin Wolf 464bca5283bSKevin Wolf if (ioctl(fd, BLKROGET, &readonly) < 0) { 46520eaf1bfSKevin Wolf return -errno; 46620eaf1bfSKevin Wolf } 46720eaf1bfSKevin Wolf 46820eaf1bfSKevin Wolf if (readonly) { 46920eaf1bfSKevin Wolf return -EACCES; 47020eaf1bfSKevin Wolf } 47120eaf1bfSKevin Wolf #endif /* defined(BLKROGET) */ 47220eaf1bfSKevin Wolf return 0; 47320eaf1bfSKevin Wolf } 47420eaf1bfSKevin Wolf 47523dece19SKevin Wolf static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) 476c1bb86cdSEric Blake { 47723dece19SKevin Wolf bool read_write = false; 478c1bb86cdSEric Blake assert(open_flags != NULL); 479c1bb86cdSEric Blake 480c1bb86cdSEric Blake *open_flags |= O_BINARY; 481c1bb86cdSEric Blake *open_flags &= ~O_ACCMODE; 48223dece19SKevin Wolf 48323dece19SKevin Wolf if (bdrv_flags & BDRV_O_AUTO_RDONLY) { 48423dece19SKevin Wolf read_write = has_writers; 48523dece19SKevin Wolf } else if (bdrv_flags & BDRV_O_RDWR) { 48623dece19SKevin Wolf read_write = true; 48723dece19SKevin Wolf } 48823dece19SKevin Wolf 48923dece19SKevin Wolf if (read_write) { 490c1bb86cdSEric Blake *open_flags |= O_RDWR; 491c1bb86cdSEric Blake } else { 492c1bb86cdSEric Blake *open_flags |= O_RDONLY; 493c1bb86cdSEric Blake } 494c1bb86cdSEric Blake 495c1bb86cdSEric Blake /* Use O_DSYNC for write-through caching, no flags for write-back caching, 496c1bb86cdSEric Blake * and O_DIRECT for no caching. */ 497c1bb86cdSEric Blake if ((bdrv_flags & BDRV_O_NOCACHE)) { 498c1bb86cdSEric Blake *open_flags |= O_DIRECT; 499c1bb86cdSEric Blake } 500c1bb86cdSEric Blake } 501c1bb86cdSEric Blake 502c1bb86cdSEric Blake static void raw_parse_filename(const char *filename, QDict *options, 503c1bb86cdSEric Blake Error **errp) 504c1bb86cdSEric Blake { 50503c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "file:", options); 506c1bb86cdSEric Blake } 507c1bb86cdSEric Blake 508c1bb86cdSEric Blake static QemuOptsList raw_runtime_opts = { 509c1bb86cdSEric Blake .name = "raw", 510c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), 511c1bb86cdSEric Blake .desc = { 512c1bb86cdSEric Blake { 513c1bb86cdSEric Blake .name = "filename", 514c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 515c1bb86cdSEric Blake .help = "File name of the image", 516c1bb86cdSEric Blake }, 517c1bb86cdSEric Blake { 518c1bb86cdSEric Blake .name = "aio", 519c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 520c6447510SAarushi Mehta .help = "host AIO implementation (threads, native, io_uring)", 521c1bb86cdSEric Blake }, 52216b48d5dSFam Zheng { 52316b48d5dSFam Zheng .name = "locking", 52416b48d5dSFam Zheng .type = QEMU_OPT_STRING, 52516b48d5dSFam Zheng .help = "file locking mode (on/off/auto, default: auto)", 52616b48d5dSFam Zheng }, 5277c9e5276SPaolo Bonzini { 5287c9e5276SPaolo Bonzini .name = "pr-manager", 5297c9e5276SPaolo Bonzini .type = QEMU_OPT_STRING, 5307c9e5276SPaolo Bonzini .help = "id of persistent reservation manager object (default: none)", 5317c9e5276SPaolo Bonzini }, 532f357fcd8SStefan Hajnoczi #if defined(__linux__) 533f357fcd8SStefan Hajnoczi { 534f357fcd8SStefan Hajnoczi .name = "drop-cache", 535f357fcd8SStefan Hajnoczi .type = QEMU_OPT_BOOL, 536f357fcd8SStefan Hajnoczi .help = "invalidate page cache during live migration (default: on)", 537f357fcd8SStefan Hajnoczi }, 538f357fcd8SStefan Hajnoczi #endif 53931be8a2aSStefan Hajnoczi { 54031be8a2aSStefan Hajnoczi .name = "x-check-cache-dropped", 54131be8a2aSStefan Hajnoczi .type = QEMU_OPT_BOOL, 54231be8a2aSStefan Hajnoczi .help = "check that page cache was dropped on live migration (default: off)" 54331be8a2aSStefan Hajnoczi }, 544c1bb86cdSEric Blake { /* end of list */ } 545c1bb86cdSEric Blake }, 546c1bb86cdSEric Blake }; 547c1bb86cdSEric Blake 5488a2ce0bcSAlberto Garcia static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; 5498a2ce0bcSAlberto Garcia 550c1bb86cdSEric Blake static int raw_open_common(BlockDriverState *bs, QDict *options, 551230ff739SJohn Snow int bdrv_flags, int open_flags, 552230ff739SJohn Snow bool device, Error **errp) 553c1bb86cdSEric Blake { 554c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 555c1bb86cdSEric Blake QemuOpts *opts; 556c1bb86cdSEric Blake Error *local_err = NULL; 557c1bb86cdSEric Blake const char *filename = NULL; 5587c9e5276SPaolo Bonzini const char *str; 559c1bb86cdSEric Blake BlockdevAioOptions aio, aio_default; 560c1bb86cdSEric Blake int fd, ret; 561c1bb86cdSEric Blake struct stat st; 562244a5668SFam Zheng OnOffAuto locking; 563c1bb86cdSEric Blake 564c1bb86cdSEric Blake opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 565af175e85SMarkus Armbruster if (!qemu_opts_absorb_qdict(opts, options, errp)) { 566c1bb86cdSEric Blake ret = -EINVAL; 567c1bb86cdSEric Blake goto fail; 568c1bb86cdSEric Blake } 569c1bb86cdSEric Blake 570c1bb86cdSEric Blake filename = qemu_opt_get(opts, "filename"); 571c1bb86cdSEric Blake 572db0754dfSFam Zheng ret = raw_normalize_devicepath(&filename, errp); 573c1bb86cdSEric Blake if (ret != 0) { 574c1bb86cdSEric Blake goto fail; 575c1bb86cdSEric Blake } 576c1bb86cdSEric Blake 577c6447510SAarushi Mehta if (bdrv_flags & BDRV_O_NATIVE_AIO) { 578c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE; 579c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 580c6447510SAarushi Mehta } else if (bdrv_flags & BDRV_O_IO_URING) { 581c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING; 582c6447510SAarushi Mehta #endif 583c6447510SAarushi Mehta } else { 584c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_THREADS; 585c6447510SAarushi Mehta } 586c6447510SAarushi Mehta 587f7abe0ecSMarc-André Lureau aio = qapi_enum_parse(&BlockdevAioOptions_lookup, 588f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "aio"), 58906c60b6cSMarkus Armbruster aio_default, &local_err); 590c1bb86cdSEric Blake if (local_err) { 591c1bb86cdSEric Blake error_propagate(errp, local_err); 592c1bb86cdSEric Blake ret = -EINVAL; 593c1bb86cdSEric Blake goto fail; 594c1bb86cdSEric Blake } 595c6447510SAarushi Mehta 596c1bb86cdSEric Blake s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); 597c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 598c6447510SAarushi Mehta s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING); 599c6447510SAarushi Mehta #endif 600c1bb86cdSEric Blake 601f7abe0ecSMarc-André Lureau locking = qapi_enum_parse(&OnOffAuto_lookup, 602f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "locking"), 60306c60b6cSMarkus Armbruster ON_OFF_AUTO_AUTO, &local_err); 604244a5668SFam Zheng if (local_err) { 605244a5668SFam Zheng error_propagate(errp, local_err); 606244a5668SFam Zheng ret = -EINVAL; 607244a5668SFam Zheng goto fail; 608244a5668SFam Zheng } 609244a5668SFam Zheng switch (locking) { 610244a5668SFam Zheng case ON_OFF_AUTO_ON: 611244a5668SFam Zheng s->use_lock = true; 6122b218f5dSFam Zheng if (!qemu_has_ofd_lock()) { 613db0754dfSFam Zheng warn_report("File lock requested but OFD locking syscall is " 614db0754dfSFam Zheng "unavailable, falling back to POSIX file locks"); 615db0754dfSFam Zheng error_printf("Due to the implementation, locks can be lost " 6162b218f5dSFam Zheng "unexpectedly.\n"); 6172b218f5dSFam Zheng } 618244a5668SFam Zheng break; 619244a5668SFam Zheng case ON_OFF_AUTO_OFF: 620244a5668SFam Zheng s->use_lock = false; 621244a5668SFam Zheng break; 622244a5668SFam Zheng case ON_OFF_AUTO_AUTO: 6232b218f5dSFam Zheng s->use_lock = qemu_has_ofd_lock(); 624244a5668SFam Zheng break; 625244a5668SFam Zheng default: 626244a5668SFam Zheng abort(); 627244a5668SFam Zheng } 628244a5668SFam Zheng 6297c9e5276SPaolo Bonzini str = qemu_opt_get(opts, "pr-manager"); 6307c9e5276SPaolo Bonzini if (str) { 6317c9e5276SPaolo Bonzini s->pr_mgr = pr_manager_lookup(str, &local_err); 6327c9e5276SPaolo Bonzini if (local_err) { 6337c9e5276SPaolo Bonzini error_propagate(errp, local_err); 6347c9e5276SPaolo Bonzini ret = -EINVAL; 6357c9e5276SPaolo Bonzini goto fail; 6367c9e5276SPaolo Bonzini } 6377c9e5276SPaolo Bonzini } 6387c9e5276SPaolo Bonzini 639f357fcd8SStefan Hajnoczi s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); 64031be8a2aSStefan Hajnoczi s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", 64131be8a2aSStefan Hajnoczi false); 64231be8a2aSStefan Hajnoczi 643c1bb86cdSEric Blake s->open_flags = open_flags; 64423dece19SKevin Wolf raw_parse_flags(bdrv_flags, &s->open_flags, false); 645c1bb86cdSEric Blake 646c1bb86cdSEric Blake s->fd = -1; 647b18a24a9SDaniel P. Berrangé fd = qemu_open(filename, s->open_flags, errp); 64864107dc0SKevin Wolf ret = fd < 0 ? -errno : 0; 64964107dc0SKevin Wolf 65064107dc0SKevin Wolf if (ret < 0) { 651c1bb86cdSEric Blake if (ret == -EROFS) { 652c1bb86cdSEric Blake ret = -EACCES; 653c1bb86cdSEric Blake } 654c1bb86cdSEric Blake goto fail; 655c1bb86cdSEric Blake } 656c1bb86cdSEric Blake s->fd = fd; 657c1bb86cdSEric Blake 658bca5283bSKevin Wolf /* Check s->open_flags rather than bdrv_flags due to auto-read-only */ 659bca5283bSKevin Wolf if (s->open_flags & O_RDWR) { 660bca5283bSKevin Wolf ret = check_hdev_writable(s->fd); 661bca5283bSKevin Wolf if (ret < 0) { 662bca5283bSKevin Wolf error_setg_errno(errp, -ret, "The device is not writable"); 663bca5283bSKevin Wolf goto fail; 664bca5283bSKevin Wolf } 665bca5283bSKevin Wolf } 666bca5283bSKevin Wolf 667244a5668SFam Zheng s->perm = 0; 668244a5668SFam Zheng s->shared_perm = BLK_PERM_ALL; 669244a5668SFam Zheng 670c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 671c1bb86cdSEric Blake /* Currently Linux does AIO only for files opened with O_DIRECT */ 672ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 673ed6e2161SNishanth Aravamudan if (!(s->open_flags & O_DIRECT)) { 674c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but it requires " 675c1bb86cdSEric Blake "cache.direct=on, which was not specified."); 676c1bb86cdSEric Blake ret = -EINVAL; 677c1bb86cdSEric Blake goto fail; 678c1bb86cdSEric Blake } 679ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { 680ed6e2161SNishanth Aravamudan error_prepend(errp, "Unable to use native AIO: "); 681ed6e2161SNishanth Aravamudan goto fail; 682ed6e2161SNishanth Aravamudan } 683ed6e2161SNishanth Aravamudan } 684c1bb86cdSEric Blake #else 685c1bb86cdSEric Blake if (s->use_linux_aio) { 686c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but is not supported " 687c1bb86cdSEric Blake "in this build."); 688c1bb86cdSEric Blake ret = -EINVAL; 689c1bb86cdSEric Blake goto fail; 690c1bb86cdSEric Blake } 691c1bb86cdSEric Blake #endif /* !defined(CONFIG_LINUX_AIO) */ 692c1bb86cdSEric Blake 693c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 694c6447510SAarushi Mehta if (s->use_linux_io_uring) { 695c6447510SAarushi Mehta if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs), errp)) { 696c6447510SAarushi Mehta error_prepend(errp, "Unable to use io_uring: "); 697c6447510SAarushi Mehta goto fail; 698c6447510SAarushi Mehta } 699c6447510SAarushi Mehta } 700c6447510SAarushi Mehta #else 701c6447510SAarushi Mehta if (s->use_linux_io_uring) { 702c6447510SAarushi Mehta error_setg(errp, "aio=io_uring was specified, but is not supported " 703c6447510SAarushi Mehta "in this build."); 704c6447510SAarushi Mehta ret = -EINVAL; 705c6447510SAarushi Mehta goto fail; 706c6447510SAarushi Mehta } 707c6447510SAarushi Mehta #endif /* !defined(CONFIG_LINUX_IO_URING) */ 708c6447510SAarushi Mehta 709c1bb86cdSEric Blake s->has_discard = true; 710c1bb86cdSEric Blake s->has_write_zeroes = true; 7115edc8557SKevin Wolf if ((bs->open_flags & BDRV_O_NOCACHE) != 0 && !dio_byte_aligned(s->fd)) { 712c1bb86cdSEric Blake s->needs_alignment = true; 713c1bb86cdSEric Blake } 714c1bb86cdSEric Blake 715c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 716c1bb86cdSEric Blake ret = -errno; 717c1bb86cdSEric Blake error_setg_errno(errp, errno, "Could not stat file"); 718c1bb86cdSEric Blake goto fail; 719c1bb86cdSEric Blake } 720230ff739SJohn Snow 721230ff739SJohn Snow if (!device) { 722230ff739SJohn Snow if (S_ISBLK(st.st_mode)) { 723230ff739SJohn Snow warn_report("Opening a block device as a file using the '%s' " 724230ff739SJohn Snow "driver is deprecated", bs->drv->format_name); 725230ff739SJohn Snow } else if (S_ISCHR(st.st_mode)) { 726230ff739SJohn Snow warn_report("Opening a character device as a file using the '%s' " 727230ff739SJohn Snow "driver is deprecated", bs->drv->format_name); 728230ff739SJohn Snow } else if (!S_ISREG(st.st_mode)) { 729230ff739SJohn Snow error_setg(errp, "A regular file was expected by the '%s' driver, " 730230ff739SJohn Snow "but something else was given", bs->drv->format_name); 731230ff739SJohn Snow ret = -EINVAL; 732230ff739SJohn Snow goto fail; 733230ff739SJohn Snow } else { 734c1bb86cdSEric Blake s->discard_zeroes = true; 735c1bb86cdSEric Blake s->has_fallocate = true; 736c1bb86cdSEric Blake } 737230ff739SJohn Snow } else { 738230ff739SJohn Snow if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 739230ff739SJohn Snow error_setg(errp, "'%s' driver expects either " 740230ff739SJohn Snow "a character or block device", bs->drv->format_name); 741230ff739SJohn Snow ret = -EINVAL; 742230ff739SJohn Snow goto fail; 743230ff739SJohn Snow } 744230ff739SJohn Snow } 745230ff739SJohn Snow 746c1bb86cdSEric Blake if (S_ISBLK(st.st_mode)) { 747c1bb86cdSEric Blake #ifdef BLKDISCARDZEROES 748c1bb86cdSEric Blake unsigned int arg; 749c1bb86cdSEric Blake if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { 750c1bb86cdSEric Blake s->discard_zeroes = true; 751c1bb86cdSEric Blake } 752c1bb86cdSEric Blake #endif 753c1bb86cdSEric Blake #ifdef __linux__ 754c1bb86cdSEric Blake /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do 755c1bb86cdSEric Blake * not rely on the contents of discarded blocks unless using O_DIRECT. 756c1bb86cdSEric Blake * Same for BLKZEROOUT. 757c1bb86cdSEric Blake */ 758c1bb86cdSEric Blake if (!(bs->open_flags & BDRV_O_NOCACHE)) { 759c1bb86cdSEric Blake s->discard_zeroes = false; 760c1bb86cdSEric Blake s->has_write_zeroes = false; 761c1bb86cdSEric Blake } 762c1bb86cdSEric Blake #endif 763c1bb86cdSEric Blake } 764c1bb86cdSEric Blake #ifdef __FreeBSD__ 765c1bb86cdSEric Blake if (S_ISCHR(st.st_mode)) { 766c1bb86cdSEric Blake /* 767c1bb86cdSEric Blake * The file is a char device (disk), which on FreeBSD isn't behind 768c1bb86cdSEric Blake * a pager, so force all requests to be aligned. This is needed 769c1bb86cdSEric Blake * so QEMU makes sure all IO operations on the device are aligned 770c1bb86cdSEric Blake * to sector size, or else FreeBSD will reject them with EINVAL. 771c1bb86cdSEric Blake */ 772c1bb86cdSEric Blake s->needs_alignment = true; 773c1bb86cdSEric Blake } 774c1bb86cdSEric Blake #endif 775c1bb86cdSEric Blake 776c1bb86cdSEric Blake #ifdef CONFIG_XFS 777c1bb86cdSEric Blake if (platform_test_xfs_fd(s->fd)) { 778c1bb86cdSEric Blake s->is_xfs = true; 779c1bb86cdSEric Blake } 780c1bb86cdSEric Blake #endif 781c1bb86cdSEric Blake 782738301e1SKevin Wolf bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; 7832f0c6e7aSKevin Wolf if (S_ISREG(st.st_mode)) { 7842f0c6e7aSKevin Wolf /* When extending regular files, we get zeros from the OS */ 7852f0c6e7aSKevin Wolf bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; 7862f0c6e7aSKevin Wolf } 787c1bb86cdSEric Blake ret = 0; 788c1bb86cdSEric Blake fail: 789a8c5cf27SKevin Wolf if (ret < 0 && s->fd != -1) { 790a8c5cf27SKevin Wolf qemu_close(s->fd); 791a8c5cf27SKevin Wolf } 792c1bb86cdSEric Blake if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { 793c1bb86cdSEric Blake unlink(filename); 794c1bb86cdSEric Blake } 795c1bb86cdSEric Blake qemu_opts_del(opts); 796c1bb86cdSEric Blake return ret; 797c1bb86cdSEric Blake } 798c1bb86cdSEric Blake 799c1bb86cdSEric Blake static int raw_open(BlockDriverState *bs, QDict *options, int flags, 800c1bb86cdSEric Blake Error **errp) 801c1bb86cdSEric Blake { 802c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 803c1bb86cdSEric Blake 804c1bb86cdSEric Blake s->type = FTYPE_FILE; 805230ff739SJohn Snow return raw_open_common(bs, options, flags, 0, false, errp); 806c1bb86cdSEric Blake } 807c1bb86cdSEric Blake 808244a5668SFam Zheng typedef enum { 809244a5668SFam Zheng RAW_PL_PREPARE, 810244a5668SFam Zheng RAW_PL_COMMIT, 811244a5668SFam Zheng RAW_PL_ABORT, 812244a5668SFam Zheng } RawPermLockOp; 813244a5668SFam Zheng 814244a5668SFam Zheng #define PERM_FOREACH(i) \ 815244a5668SFam Zheng for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) 816244a5668SFam Zheng 817244a5668SFam Zheng /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the 818244a5668SFam Zheng * file; if @unlock == true, also unlock the unneeded bytes. 819244a5668SFam Zheng * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. 820244a5668SFam Zheng */ 8212996ffadSFam Zheng static int raw_apply_lock_bytes(BDRVRawState *s, int fd, 822244a5668SFam Zheng uint64_t perm_lock_bits, 823244a5668SFam Zheng uint64_t shared_perm_lock_bits, 824244a5668SFam Zheng bool unlock, Error **errp) 825244a5668SFam Zheng { 826244a5668SFam Zheng int ret; 827244a5668SFam Zheng int i; 8282996ffadSFam Zheng uint64_t locked_perm, locked_shared_perm; 8292996ffadSFam Zheng 8302996ffadSFam Zheng if (s) { 8312996ffadSFam Zheng locked_perm = s->locked_perm; 8322996ffadSFam Zheng locked_shared_perm = s->locked_shared_perm; 8332996ffadSFam Zheng } else { 8342996ffadSFam Zheng /* 8352996ffadSFam Zheng * We don't have the previous bits, just lock/unlock for each of the 8362996ffadSFam Zheng * requested bits. 8372996ffadSFam Zheng */ 8382996ffadSFam Zheng if (unlock) { 8392996ffadSFam Zheng locked_perm = BLK_PERM_ALL; 8402996ffadSFam Zheng locked_shared_perm = BLK_PERM_ALL; 8412996ffadSFam Zheng } else { 8422996ffadSFam Zheng locked_perm = 0; 8432996ffadSFam Zheng locked_shared_perm = 0; 8442996ffadSFam Zheng } 8452996ffadSFam Zheng } 846244a5668SFam Zheng 847244a5668SFam Zheng PERM_FOREACH(i) { 848244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 8492996ffadSFam Zheng uint64_t bit = (1ULL << i); 8502996ffadSFam Zheng if ((perm_lock_bits & bit) && !(locked_perm & bit)) { 851d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 852244a5668SFam Zheng if (ret) { 853*797e3e38SDavid Edmondson raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", 854*797e3e38SDavid Edmondson off); 855244a5668SFam Zheng return ret; 8562996ffadSFam Zheng } else if (s) { 8572996ffadSFam Zheng s->locked_perm |= bit; 858244a5668SFam Zheng } 8592996ffadSFam Zheng } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { 860d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 861244a5668SFam Zheng if (ret) { 862*797e3e38SDavid Edmondson error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); 863244a5668SFam Zheng return ret; 8642996ffadSFam Zheng } else if (s) { 8652996ffadSFam Zheng s->locked_perm &= ~bit; 866244a5668SFam Zheng } 867244a5668SFam Zheng } 868244a5668SFam Zheng } 869244a5668SFam Zheng PERM_FOREACH(i) { 870244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 8712996ffadSFam Zheng uint64_t bit = (1ULL << i); 8722996ffadSFam Zheng if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { 873d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 874244a5668SFam Zheng if (ret) { 875*797e3e38SDavid Edmondson raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", 876*797e3e38SDavid Edmondson off); 877244a5668SFam Zheng return ret; 8782996ffadSFam Zheng } else if (s) { 8792996ffadSFam Zheng s->locked_shared_perm |= bit; 880244a5668SFam Zheng } 8812996ffadSFam Zheng } else if (unlock && (locked_shared_perm & bit) && 8822996ffadSFam Zheng !(shared_perm_lock_bits & bit)) { 883d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 884244a5668SFam Zheng if (ret) { 885*797e3e38SDavid Edmondson error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); 886244a5668SFam Zheng return ret; 8872996ffadSFam Zheng } else if (s) { 8882996ffadSFam Zheng s->locked_shared_perm &= ~bit; 889244a5668SFam Zheng } 890244a5668SFam Zheng } 891244a5668SFam Zheng } 892244a5668SFam Zheng return 0; 893244a5668SFam Zheng } 894244a5668SFam Zheng 895244a5668SFam Zheng /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ 896d0a96155SMax Reitz static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, 897244a5668SFam Zheng Error **errp) 898244a5668SFam Zheng { 899244a5668SFam Zheng int ret; 900244a5668SFam Zheng int i; 901244a5668SFam Zheng 902244a5668SFam Zheng PERM_FOREACH(i) { 903244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 904244a5668SFam Zheng uint64_t p = 1ULL << i; 905244a5668SFam Zheng if (perm & p) { 906d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 907244a5668SFam Zheng if (ret) { 908244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 909*797e3e38SDavid Edmondson 910*797e3e38SDavid Edmondson raw_lock_error_setg_errno(errp, -ret, 911244a5668SFam Zheng "Failed to get \"%s\" lock", 912244a5668SFam Zheng perm_name); 913244a5668SFam Zheng g_free(perm_name); 914244a5668SFam Zheng return ret; 915244a5668SFam Zheng } 916244a5668SFam Zheng } 917244a5668SFam Zheng } 918244a5668SFam Zheng PERM_FOREACH(i) { 919244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 920244a5668SFam Zheng uint64_t p = 1ULL << i; 921244a5668SFam Zheng if (!(shared_perm & p)) { 922d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 923244a5668SFam Zheng if (ret) { 924244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 925*797e3e38SDavid Edmondson 926*797e3e38SDavid Edmondson raw_lock_error_setg_errno(errp, -ret, 927244a5668SFam Zheng "Failed to get shared \"%s\" lock", 928244a5668SFam Zheng perm_name); 929244a5668SFam Zheng g_free(perm_name); 930244a5668SFam Zheng return ret; 931244a5668SFam Zheng } 932244a5668SFam Zheng } 933244a5668SFam Zheng } 934244a5668SFam Zheng return 0; 935244a5668SFam Zheng } 936244a5668SFam Zheng 937244a5668SFam Zheng static int raw_handle_perm_lock(BlockDriverState *bs, 938244a5668SFam Zheng RawPermLockOp op, 939244a5668SFam Zheng uint64_t new_perm, uint64_t new_shared, 940244a5668SFam Zheng Error **errp) 941244a5668SFam Zheng { 942244a5668SFam Zheng BDRVRawState *s = bs->opaque; 943244a5668SFam Zheng int ret = 0; 944244a5668SFam Zheng Error *local_err = NULL; 945244a5668SFam Zheng 946244a5668SFam Zheng if (!s->use_lock) { 947244a5668SFam Zheng return 0; 948244a5668SFam Zheng } 949244a5668SFam Zheng 950244a5668SFam Zheng if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { 951244a5668SFam Zheng return 0; 952244a5668SFam Zheng } 953244a5668SFam Zheng 954244a5668SFam Zheng switch (op) { 955244a5668SFam Zheng case RAW_PL_PREPARE: 956696aaaedSVladimir Sementsov-Ogievskiy if ((s->perm | new_perm) == s->perm && 957696aaaedSVladimir Sementsov-Ogievskiy (s->shared_perm & new_shared) == s->shared_perm) 958696aaaedSVladimir Sementsov-Ogievskiy { 959696aaaedSVladimir Sementsov-Ogievskiy /* 960696aaaedSVladimir Sementsov-Ogievskiy * We are going to unlock bytes, it should not fail. If it fail due 961696aaaedSVladimir Sementsov-Ogievskiy * to some fs-dependent permission-unrelated reasons (which occurs 962696aaaedSVladimir Sementsov-Ogievskiy * sometimes on NFS and leads to abort in bdrv_replace_child) we 963696aaaedSVladimir Sementsov-Ogievskiy * can't prevent such errors by any check here. And we ignore them 964696aaaedSVladimir Sementsov-Ogievskiy * anyway in ABORT and COMMIT. 965696aaaedSVladimir Sementsov-Ogievskiy */ 966696aaaedSVladimir Sementsov-Ogievskiy return 0; 967696aaaedSVladimir Sementsov-Ogievskiy } 968f2e3af29SFam Zheng ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, 969244a5668SFam Zheng ~s->shared_perm | ~new_shared, 970244a5668SFam Zheng false, errp); 971244a5668SFam Zheng if (!ret) { 972f2e3af29SFam Zheng ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); 973244a5668SFam Zheng if (!ret) { 974244a5668SFam Zheng return 0; 975244a5668SFam Zheng } 976b857431dSFam Zheng error_append_hint(errp, 977b857431dSFam Zheng "Is another process using the image [%s]?\n", 978b857431dSFam Zheng bs->filename); 979244a5668SFam Zheng } 980244a5668SFam Zheng /* fall through to unlock bytes. */ 981244a5668SFam Zheng case RAW_PL_ABORT: 982f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, 983d0a96155SMax Reitz true, &local_err); 984244a5668SFam Zheng if (local_err) { 985244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 986244a5668SFam Zheng * fail. Something weird happened, report it. 987244a5668SFam Zheng */ 988db0754dfSFam Zheng warn_report_err(local_err); 989244a5668SFam Zheng } 990244a5668SFam Zheng break; 991244a5668SFam Zheng case RAW_PL_COMMIT: 992f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, 993d0a96155SMax Reitz true, &local_err); 994244a5668SFam Zheng if (local_err) { 995244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 996244a5668SFam Zheng * fail. Something weird happened, report it. 997244a5668SFam Zheng */ 998db0754dfSFam Zheng warn_report_err(local_err); 999244a5668SFam Zheng } 1000244a5668SFam Zheng break; 1001244a5668SFam Zheng } 1002244a5668SFam Zheng return ret; 1003244a5668SFam Zheng } 1004244a5668SFam Zheng 10055cec2870SKevin Wolf static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, 100623dece19SKevin Wolf int *open_flags, uint64_t perm, bool force_dup, 10076ceabe6fSKevin Wolf Error **errp) 10085cec2870SKevin Wolf { 10095cec2870SKevin Wolf BDRVRawState *s = bs->opaque; 10105cec2870SKevin Wolf int fd = -1; 10115cec2870SKevin Wolf int ret; 101223dece19SKevin Wolf bool has_writers = perm & 101323dece19SKevin Wolf (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); 10145cec2870SKevin Wolf int fcntl_flags = O_APPEND | O_NONBLOCK; 10155cec2870SKevin Wolf #ifdef O_NOATIME 10165cec2870SKevin Wolf fcntl_flags |= O_NOATIME; 10175cec2870SKevin Wolf #endif 10185cec2870SKevin Wolf 10195cec2870SKevin Wolf *open_flags = 0; 10205cec2870SKevin Wolf if (s->type == FTYPE_CD) { 10215cec2870SKevin Wolf *open_flags |= O_NONBLOCK; 10225cec2870SKevin Wolf } 10235cec2870SKevin Wolf 102423dece19SKevin Wolf raw_parse_flags(flags, open_flags, has_writers); 10255cec2870SKevin Wolf 10265cec2870SKevin Wolf #ifdef O_ASYNC 10275cec2870SKevin Wolf /* Not all operating systems have O_ASYNC, and those that don't 10285cec2870SKevin Wolf * will not let us track the state into rs->open_flags (typically 10295cec2870SKevin Wolf * you achieve the same effect with an ioctl, for example I_SETSIG 10305cec2870SKevin Wolf * on Solaris). But we do not use O_ASYNC, so that's fine. 10315cec2870SKevin Wolf */ 10325cec2870SKevin Wolf assert((s->open_flags & O_ASYNC) == 0); 10335cec2870SKevin Wolf #endif 10345cec2870SKevin Wolf 10356ceabe6fSKevin Wolf if (!force_dup && *open_flags == s->open_flags) { 10366ceabe6fSKevin Wolf /* We're lucky, the existing fd is fine */ 10376ceabe6fSKevin Wolf return s->fd; 10386ceabe6fSKevin Wolf } 10396ceabe6fSKevin Wolf 10405cec2870SKevin Wolf if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { 10415cec2870SKevin Wolf /* dup the original fd */ 10425cec2870SKevin Wolf fd = qemu_dup(s->fd); 10435cec2870SKevin Wolf if (fd >= 0) { 10445cec2870SKevin Wolf ret = fcntl_setfl(fd, *open_flags); 10455cec2870SKevin Wolf if (ret) { 10465cec2870SKevin Wolf qemu_close(fd); 10475cec2870SKevin Wolf fd = -1; 10485cec2870SKevin Wolf } 10495cec2870SKevin Wolf } 10505cec2870SKevin Wolf } 10515cec2870SKevin Wolf 1052b18a24a9SDaniel P. Berrangé /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ 10535cec2870SKevin Wolf if (fd == -1) { 10545cec2870SKevin Wolf const char *normalized_filename = bs->filename; 10555cec2870SKevin Wolf ret = raw_normalize_devicepath(&normalized_filename, errp); 10565cec2870SKevin Wolf if (ret >= 0) { 1057b18a24a9SDaniel P. Berrangé fd = qemu_open(normalized_filename, *open_flags, errp); 10585cec2870SKevin Wolf if (fd == -1) { 10595cec2870SKevin Wolf return -1; 10605cec2870SKevin Wolf } 10615cec2870SKevin Wolf } 10625cec2870SKevin Wolf } 10635cec2870SKevin Wolf 1064bca5283bSKevin Wolf if (fd != -1 && (*open_flags & O_RDWR)) { 1065bca5283bSKevin Wolf ret = check_hdev_writable(fd); 1066bca5283bSKevin Wolf if (ret < 0) { 1067bca5283bSKevin Wolf qemu_close(fd); 1068bca5283bSKevin Wolf error_setg_errno(errp, -ret, "The device is not writable"); 1069bca5283bSKevin Wolf return -1; 1070bca5283bSKevin Wolf } 1071bca5283bSKevin Wolf } 1072bca5283bSKevin Wolf 10735cec2870SKevin Wolf return fd; 10745cec2870SKevin Wolf } 10755cec2870SKevin Wolf 1076c1bb86cdSEric Blake static int raw_reopen_prepare(BDRVReopenState *state, 1077c1bb86cdSEric Blake BlockReopenQueue *queue, Error **errp) 1078c1bb86cdSEric Blake { 1079c1bb86cdSEric Blake BDRVRawState *s; 1080c1bb86cdSEric Blake BDRVRawReopenState *rs; 108131be8a2aSStefan Hajnoczi QemuOpts *opts; 1082a6aeca0cSKevin Wolf int ret; 1083c1bb86cdSEric Blake Error *local_err = NULL; 1084c1bb86cdSEric Blake 1085c1bb86cdSEric Blake assert(state != NULL); 1086c1bb86cdSEric Blake assert(state->bs != NULL); 1087c1bb86cdSEric Blake 1088c1bb86cdSEric Blake s = state->bs->opaque; 1089c1bb86cdSEric Blake 1090c1bb86cdSEric Blake state->opaque = g_new0(BDRVRawReopenState, 1); 1091c1bb86cdSEric Blake rs = state->opaque; 109231be8a2aSStefan Hajnoczi 109331be8a2aSStefan Hajnoczi /* Handle options changes */ 109431be8a2aSStefan Hajnoczi opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 1095af175e85SMarkus Armbruster if (!qemu_opts_absorb_qdict(opts, state->options, errp)) { 109631be8a2aSStefan Hajnoczi ret = -EINVAL; 109731be8a2aSStefan Hajnoczi goto out; 109831be8a2aSStefan Hajnoczi } 109931be8a2aSStefan Hajnoczi 1100f357fcd8SStefan Hajnoczi rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); 11018d324575SAlberto Garcia rs->check_cache_dropped = 11028d324575SAlberto Garcia qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); 11038d324575SAlberto Garcia 11048d324575SAlberto Garcia /* This driver's reopen function doesn't currently allow changing 11058d324575SAlberto Garcia * other options, so let's put them back in the original QDict and 11068d324575SAlberto Garcia * bdrv_reopen_prepare() will detect changes and complain. */ 11078d324575SAlberto Garcia qemu_opts_to_qdict(opts, state->options); 1108c1bb86cdSEric Blake 11095cec2870SKevin Wolf rs->fd = raw_reconfigure_getfd(state->bs, state->flags, &rs->open_flags, 111023dece19SKevin Wolf state->perm, true, &local_err); 11115cec2870SKevin Wolf if (local_err) { 11125cec2870SKevin Wolf error_propagate(errp, local_err); 1113c1bb86cdSEric Blake ret = -1; 11145cec2870SKevin Wolf goto out; 1115c1bb86cdSEric Blake } 1116c1bb86cdSEric Blake 1117c1bb86cdSEric Blake /* Fail already reopen_prepare() if we can't get a working O_DIRECT 1118c1bb86cdSEric Blake * alignment with the new fd. */ 1119c1bb86cdSEric Blake if (rs->fd != -1) { 1120c1bb86cdSEric Blake raw_probe_alignment(state->bs, rs->fd, &local_err); 1121c1bb86cdSEric Blake if (local_err) { 1122c1bb86cdSEric Blake error_propagate(errp, local_err); 1123c1bb86cdSEric Blake ret = -EINVAL; 1124a6aeca0cSKevin Wolf goto out_fd; 1125a6aeca0cSKevin Wolf } 1126c1bb86cdSEric Blake } 1127c1bb86cdSEric Blake 1128e0c9cf3aSKevin Wolf s->reopen_state = state; 1129a6aeca0cSKevin Wolf ret = 0; 1130a6aeca0cSKevin Wolf out_fd: 1131a6aeca0cSKevin Wolf if (ret < 0) { 1132a6aeca0cSKevin Wolf qemu_close(rs->fd); 1133a6aeca0cSKevin Wolf rs->fd = -1; 1134a6aeca0cSKevin Wolf } 113531be8a2aSStefan Hajnoczi out: 113631be8a2aSStefan Hajnoczi qemu_opts_del(opts); 1137c1bb86cdSEric Blake return ret; 1138c1bb86cdSEric Blake } 1139c1bb86cdSEric Blake 1140c1bb86cdSEric Blake static void raw_reopen_commit(BDRVReopenState *state) 1141c1bb86cdSEric Blake { 1142c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1143c1bb86cdSEric Blake BDRVRawState *s = state->bs->opaque; 1144c1bb86cdSEric Blake 1145f357fcd8SStefan Hajnoczi s->drop_cache = rs->drop_cache; 114631be8a2aSStefan Hajnoczi s->check_cache_dropped = rs->check_cache_dropped; 1147c1bb86cdSEric Blake s->open_flags = rs->open_flags; 1148c1bb86cdSEric Blake 1149c1bb86cdSEric Blake qemu_close(s->fd); 1150c1bb86cdSEric Blake s->fd = rs->fd; 1151c1bb86cdSEric Blake 1152c1bb86cdSEric Blake g_free(state->opaque); 1153c1bb86cdSEric Blake state->opaque = NULL; 1154e0c9cf3aSKevin Wolf 1155e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1156e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1157c1bb86cdSEric Blake } 1158c1bb86cdSEric Blake 1159c1bb86cdSEric Blake 1160c1bb86cdSEric Blake static void raw_reopen_abort(BDRVReopenState *state) 1161c1bb86cdSEric Blake { 1162c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1163e0c9cf3aSKevin Wolf BDRVRawState *s = state->bs->opaque; 1164c1bb86cdSEric Blake 1165c1bb86cdSEric Blake /* nothing to do if NULL, we didn't get far enough */ 1166c1bb86cdSEric Blake if (rs == NULL) { 1167c1bb86cdSEric Blake return; 1168c1bb86cdSEric Blake } 1169c1bb86cdSEric Blake 1170c1bb86cdSEric Blake if (rs->fd >= 0) { 1171c1bb86cdSEric Blake qemu_close(rs->fd); 1172c1bb86cdSEric Blake rs->fd = -1; 1173c1bb86cdSEric Blake } 1174c1bb86cdSEric Blake g_free(state->opaque); 1175c1bb86cdSEric Blake state->opaque = NULL; 1176e0c9cf3aSKevin Wolf 1177e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1178e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1179c1bb86cdSEric Blake } 1180c1bb86cdSEric Blake 1181867eccfeSMaxim Levitsky static int sg_get_max_transfer_length(int fd) 1182c1bb86cdSEric Blake { 1183c1bb86cdSEric Blake #ifdef BLKSECTGET 118448265250SEric Farman int max_bytes = 0; 1185867eccfeSMaxim Levitsky 1186867eccfeSMaxim Levitsky if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) { 118748265250SEric Farman return max_bytes; 1188c1bb86cdSEric Blake } else { 1189c1bb86cdSEric Blake return -errno; 1190c1bb86cdSEric Blake } 1191c1bb86cdSEric Blake #else 1192c1bb86cdSEric Blake return -ENOSYS; 1193c1bb86cdSEric Blake #endif 1194c1bb86cdSEric Blake } 1195c1bb86cdSEric Blake 1196867eccfeSMaxim Levitsky static int sg_get_max_segments(int fd) 11979103f1ceSFam Zheng { 11989103f1ceSFam Zheng #ifdef CONFIG_LINUX 11999103f1ceSFam Zheng char buf[32]; 12009103f1ceSFam Zheng const char *end; 1201867eccfeSMaxim Levitsky char *sysfspath = NULL; 12029103f1ceSFam Zheng int ret; 1203867eccfeSMaxim Levitsky int sysfd = -1; 12049103f1ceSFam Zheng long max_segments; 1205867eccfeSMaxim Levitsky struct stat st; 1206867eccfeSMaxim Levitsky 1207867eccfeSMaxim Levitsky if (fstat(fd, &st)) { 1208867eccfeSMaxim Levitsky ret = -errno; 1209867eccfeSMaxim Levitsky goto out; 1210867eccfeSMaxim Levitsky } 12119103f1ceSFam Zheng 12129103f1ceSFam Zheng sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", 1213867eccfeSMaxim Levitsky major(st.st_rdev), minor(st.st_rdev)); 1214867eccfeSMaxim Levitsky sysfd = open(sysfspath, O_RDONLY); 1215867eccfeSMaxim Levitsky if (sysfd == -1) { 12169103f1ceSFam Zheng ret = -errno; 12179103f1ceSFam Zheng goto out; 12189103f1ceSFam Zheng } 12199103f1ceSFam Zheng do { 1220867eccfeSMaxim Levitsky ret = read(sysfd, buf, sizeof(buf) - 1); 12219103f1ceSFam Zheng } while (ret == -1 && errno == EINTR); 12229103f1ceSFam Zheng if (ret < 0) { 12239103f1ceSFam Zheng ret = -errno; 12249103f1ceSFam Zheng goto out; 12259103f1ceSFam Zheng } else if (ret == 0) { 12269103f1ceSFam Zheng ret = -EIO; 12279103f1ceSFam Zheng goto out; 12289103f1ceSFam Zheng } 12299103f1ceSFam Zheng buf[ret] = 0; 12309103f1ceSFam Zheng /* The file is ended with '\n', pass 'end' to accept that. */ 12319103f1ceSFam Zheng ret = qemu_strtol(buf, &end, 10, &max_segments); 12329103f1ceSFam Zheng if (ret == 0 && end && *end == '\n') { 12339103f1ceSFam Zheng ret = max_segments; 12349103f1ceSFam Zheng } 12359103f1ceSFam Zheng 12369103f1ceSFam Zheng out: 1237867eccfeSMaxim Levitsky if (sysfd != -1) { 1238867eccfeSMaxim Levitsky close(sysfd); 1239fed414dfSFam Zheng } 12409103f1ceSFam Zheng g_free(sysfspath); 12419103f1ceSFam Zheng return ret; 12429103f1ceSFam Zheng #else 12439103f1ceSFam Zheng return -ENOTSUP; 12449103f1ceSFam Zheng #endif 12459103f1ceSFam Zheng } 12469103f1ceSFam Zheng 1247c1bb86cdSEric Blake static void raw_refresh_limits(BlockDriverState *bs, Error **errp) 1248c1bb86cdSEric Blake { 1249c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1250c1bb86cdSEric Blake 1251867eccfeSMaxim Levitsky if (bs->sg) { 1252867eccfeSMaxim Levitsky int ret = sg_get_max_transfer_length(s->fd); 1253867eccfeSMaxim Levitsky 125448265250SEric Farman if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { 125548265250SEric Farman bs->bl.max_transfer = pow2floor(ret); 1256c1bb86cdSEric Blake } 1257867eccfeSMaxim Levitsky 1258867eccfeSMaxim Levitsky ret = sg_get_max_segments(s->fd); 12599103f1ceSFam Zheng if (ret > 0) { 1260038adc2fSWei Yang bs->bl.max_transfer = MIN(bs->bl.max_transfer, 1261038adc2fSWei Yang ret * qemu_real_host_page_size); 1262c1bb86cdSEric Blake } 1263c1bb86cdSEric Blake } 1264c1bb86cdSEric Blake 1265c1bb86cdSEric Blake raw_probe_alignment(bs, s->fd, errp); 1266c1bb86cdSEric Blake bs->bl.min_mem_alignment = s->buf_align; 1267038adc2fSWei Yang bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size); 1268c1bb86cdSEric Blake } 1269c1bb86cdSEric Blake 1270c1bb86cdSEric Blake static int check_for_dasd(int fd) 1271c1bb86cdSEric Blake { 1272c1bb86cdSEric Blake #ifdef BIODASDINFO2 1273c1bb86cdSEric Blake struct dasd_information2_t info = {0}; 1274c1bb86cdSEric Blake 1275c1bb86cdSEric Blake return ioctl(fd, BIODASDINFO2, &info); 1276c1bb86cdSEric Blake #else 1277c1bb86cdSEric Blake return -1; 1278c1bb86cdSEric Blake #endif 1279c1bb86cdSEric Blake } 1280c1bb86cdSEric Blake 1281c1bb86cdSEric Blake /** 1282c1bb86cdSEric Blake * Try to get @bs's logical and physical block size. 1283c1bb86cdSEric Blake * On success, store them in @bsz and return zero. 1284c1bb86cdSEric Blake * On failure, return negative errno. 1285c1bb86cdSEric Blake */ 1286c1bb86cdSEric Blake static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) 1287c1bb86cdSEric Blake { 1288c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1289c1bb86cdSEric Blake int ret; 1290c1bb86cdSEric Blake 1291c1bb86cdSEric Blake /* If DASD, get blocksizes */ 1292c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1293c1bb86cdSEric Blake return -ENOTSUP; 1294c1bb86cdSEric Blake } 1295c1bb86cdSEric Blake ret = probe_logical_blocksize(s->fd, &bsz->log); 1296c1bb86cdSEric Blake if (ret < 0) { 1297c1bb86cdSEric Blake return ret; 1298c1bb86cdSEric Blake } 1299c1bb86cdSEric Blake return probe_physical_blocksize(s->fd, &bsz->phys); 1300c1bb86cdSEric Blake } 1301c1bb86cdSEric Blake 1302c1bb86cdSEric Blake /** 1303c1bb86cdSEric Blake * Try to get @bs's geometry: cyls, heads, sectors. 1304c1bb86cdSEric Blake * On success, store them in @geo and return 0. 1305c1bb86cdSEric Blake * On failure return -errno. 1306c1bb86cdSEric Blake * (Allows block driver to assign default geometry values that guest sees) 1307c1bb86cdSEric Blake */ 1308c1bb86cdSEric Blake #ifdef __linux__ 1309c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1310c1bb86cdSEric Blake { 1311c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1312c1bb86cdSEric Blake struct hd_geometry ioctl_geo = {0}; 1313c1bb86cdSEric Blake 1314c1bb86cdSEric Blake /* If DASD, get its geometry */ 1315c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1316c1bb86cdSEric Blake return -ENOTSUP; 1317c1bb86cdSEric Blake } 1318c1bb86cdSEric Blake if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { 1319c1bb86cdSEric Blake return -errno; 1320c1bb86cdSEric Blake } 1321c1bb86cdSEric Blake /* HDIO_GETGEO may return success even though geo contains zeros 1322c1bb86cdSEric Blake (e.g. certain multipath setups) */ 1323c1bb86cdSEric Blake if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { 1324c1bb86cdSEric Blake return -ENOTSUP; 1325c1bb86cdSEric Blake } 1326c1bb86cdSEric Blake /* Do not return a geometry for partition */ 1327c1bb86cdSEric Blake if (ioctl_geo.start != 0) { 1328c1bb86cdSEric Blake return -ENOTSUP; 1329c1bb86cdSEric Blake } 1330c1bb86cdSEric Blake geo->heads = ioctl_geo.heads; 1331c1bb86cdSEric Blake geo->sectors = ioctl_geo.sectors; 1332c1bb86cdSEric Blake geo->cylinders = ioctl_geo.cylinders; 1333c1bb86cdSEric Blake 1334c1bb86cdSEric Blake return 0; 1335c1bb86cdSEric Blake } 1336c1bb86cdSEric Blake #else /* __linux__ */ 1337c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1338c1bb86cdSEric Blake { 1339c1bb86cdSEric Blake return -ENOTSUP; 1340c1bb86cdSEric Blake } 1341c1bb86cdSEric Blake #endif 1342c1bb86cdSEric Blake 134303425671SKevin Wolf #if defined(__linux__) 134403425671SKevin Wolf static int handle_aiocb_ioctl(void *opaque) 1345c1bb86cdSEric Blake { 134603425671SKevin Wolf RawPosixAIOData *aiocb = opaque; 1347c1bb86cdSEric Blake int ret; 1348c1bb86cdSEric Blake 1349d57c44d0SKevin Wolf ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); 1350c1bb86cdSEric Blake if (ret == -1) { 1351c1bb86cdSEric Blake return -errno; 1352c1bb86cdSEric Blake } 1353c1bb86cdSEric Blake 1354c1bb86cdSEric Blake return 0; 1355c1bb86cdSEric Blake } 135603425671SKevin Wolf #endif /* linux */ 1357c1bb86cdSEric Blake 135806dc9bd5SKevin Wolf static int handle_aiocb_flush(void *opaque) 1359c1bb86cdSEric Blake { 136006dc9bd5SKevin Wolf RawPosixAIOData *aiocb = opaque; 1361e5bcf967SKevin Wolf BDRVRawState *s = aiocb->bs->opaque; 1362c1bb86cdSEric Blake int ret; 1363c1bb86cdSEric Blake 1364e5bcf967SKevin Wolf if (s->page_cache_inconsistent) { 1365e5bcf967SKevin Wolf return -EIO; 1366e5bcf967SKevin Wolf } 1367e5bcf967SKevin Wolf 1368c1bb86cdSEric Blake ret = qemu_fdatasync(aiocb->aio_fildes); 1369c1bb86cdSEric Blake if (ret == -1) { 1370e5bcf967SKevin Wolf /* There is no clear definition of the semantics of a failing fsync(), 1371e5bcf967SKevin Wolf * so we may have to assume the worst. The sad truth is that this 1372e5bcf967SKevin Wolf * assumption is correct for Linux. Some pages are now probably marked 1373e5bcf967SKevin Wolf * clean in the page cache even though they are inconsistent with the 1374e5bcf967SKevin Wolf * on-disk contents. The next fdatasync() call would succeed, but no 1375e5bcf967SKevin Wolf * further writeback attempt will be made. We can't get back to a state 1376e5bcf967SKevin Wolf * in which we know what is on disk (we would have to rewrite 1377e5bcf967SKevin Wolf * everything that was touched since the last fdatasync() at least), so 1378e5bcf967SKevin Wolf * make bdrv_flush() fail permanently. Given that the behaviour isn't 1379e5bcf967SKevin Wolf * really defined, I have little hope that other OSes are doing better. 1380e5bcf967SKevin Wolf * 1381e5bcf967SKevin Wolf * Obviously, this doesn't affect O_DIRECT, which bypasses the page 1382e5bcf967SKevin Wolf * cache. */ 1383e5bcf967SKevin Wolf if ((s->open_flags & O_DIRECT) == 0) { 1384e5bcf967SKevin Wolf s->page_cache_inconsistent = true; 1385e5bcf967SKevin Wolf } 1386c1bb86cdSEric Blake return -errno; 1387c1bb86cdSEric Blake } 1388c1bb86cdSEric Blake return 0; 1389c1bb86cdSEric Blake } 1390c1bb86cdSEric Blake 1391c1bb86cdSEric Blake #ifdef CONFIG_PREADV 1392c1bb86cdSEric Blake 1393c1bb86cdSEric Blake static bool preadv_present = true; 1394c1bb86cdSEric Blake 1395c1bb86cdSEric Blake static ssize_t 1396c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1397c1bb86cdSEric Blake { 1398c1bb86cdSEric Blake return preadv(fd, iov, nr_iov, offset); 1399c1bb86cdSEric Blake } 1400c1bb86cdSEric Blake 1401c1bb86cdSEric Blake static ssize_t 1402c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1403c1bb86cdSEric Blake { 1404c1bb86cdSEric Blake return pwritev(fd, iov, nr_iov, offset); 1405c1bb86cdSEric Blake } 1406c1bb86cdSEric Blake 1407c1bb86cdSEric Blake #else 1408c1bb86cdSEric Blake 1409c1bb86cdSEric Blake static bool preadv_present = false; 1410c1bb86cdSEric Blake 1411c1bb86cdSEric Blake static ssize_t 1412c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1413c1bb86cdSEric Blake { 1414c1bb86cdSEric Blake return -ENOSYS; 1415c1bb86cdSEric Blake } 1416c1bb86cdSEric Blake 1417c1bb86cdSEric Blake static ssize_t 1418c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1419c1bb86cdSEric Blake { 1420c1bb86cdSEric Blake return -ENOSYS; 1421c1bb86cdSEric Blake } 1422c1bb86cdSEric Blake 1423c1bb86cdSEric Blake #endif 1424c1bb86cdSEric Blake 1425c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) 1426c1bb86cdSEric Blake { 1427c1bb86cdSEric Blake ssize_t len; 1428c1bb86cdSEric Blake 1429c1bb86cdSEric Blake do { 1430c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) 1431c1bb86cdSEric Blake len = qemu_pwritev(aiocb->aio_fildes, 1432d57c44d0SKevin Wolf aiocb->io.iov, 1433d57c44d0SKevin Wolf aiocb->io.niov, 1434c1bb86cdSEric Blake aiocb->aio_offset); 1435c1bb86cdSEric Blake else 1436c1bb86cdSEric Blake len = qemu_preadv(aiocb->aio_fildes, 1437d57c44d0SKevin Wolf aiocb->io.iov, 1438d57c44d0SKevin Wolf aiocb->io.niov, 1439c1bb86cdSEric Blake aiocb->aio_offset); 1440c1bb86cdSEric Blake } while (len == -1 && errno == EINTR); 1441c1bb86cdSEric Blake 1442c1bb86cdSEric Blake if (len == -1) { 1443c1bb86cdSEric Blake return -errno; 1444c1bb86cdSEric Blake } 1445c1bb86cdSEric Blake return len; 1446c1bb86cdSEric Blake } 1447c1bb86cdSEric Blake 1448c1bb86cdSEric Blake /* 1449c1bb86cdSEric Blake * Read/writes the data to/from a given linear buffer. 1450c1bb86cdSEric Blake * 1451c1bb86cdSEric Blake * Returns the number of bytes handles or -errno in case of an error. Short 1452c1bb86cdSEric Blake * reads are only returned if the end of the file is reached. 1453c1bb86cdSEric Blake */ 1454c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) 1455c1bb86cdSEric Blake { 1456c1bb86cdSEric Blake ssize_t offset = 0; 1457c1bb86cdSEric Blake ssize_t len; 1458c1bb86cdSEric Blake 1459c1bb86cdSEric Blake while (offset < aiocb->aio_nbytes) { 1460c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1461c1bb86cdSEric Blake len = pwrite(aiocb->aio_fildes, 1462c1bb86cdSEric Blake (const char *)buf + offset, 1463c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1464c1bb86cdSEric Blake aiocb->aio_offset + offset); 1465c1bb86cdSEric Blake } else { 1466c1bb86cdSEric Blake len = pread(aiocb->aio_fildes, 1467c1bb86cdSEric Blake buf + offset, 1468c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1469c1bb86cdSEric Blake aiocb->aio_offset + offset); 1470c1bb86cdSEric Blake } 1471c1bb86cdSEric Blake if (len == -1 && errno == EINTR) { 1472c1bb86cdSEric Blake continue; 1473c1bb86cdSEric Blake } else if (len == -1 && errno == EINVAL && 1474c1bb86cdSEric Blake (aiocb->bs->open_flags & BDRV_O_NOCACHE) && 1475c1bb86cdSEric Blake !(aiocb->aio_type & QEMU_AIO_WRITE) && 1476c1bb86cdSEric Blake offset > 0) { 1477c1bb86cdSEric Blake /* O_DIRECT pread() may fail with EINVAL when offset is unaligned 1478c1bb86cdSEric Blake * after a short read. Assume that O_DIRECT short reads only occur 1479c1bb86cdSEric Blake * at EOF. Therefore this is a short read, not an I/O error. 1480c1bb86cdSEric Blake */ 1481c1bb86cdSEric Blake break; 1482c1bb86cdSEric Blake } else if (len == -1) { 1483c1bb86cdSEric Blake offset = -errno; 1484c1bb86cdSEric Blake break; 1485c1bb86cdSEric Blake } else if (len == 0) { 1486c1bb86cdSEric Blake break; 1487c1bb86cdSEric Blake } 1488c1bb86cdSEric Blake offset += len; 1489c1bb86cdSEric Blake } 1490c1bb86cdSEric Blake 1491c1bb86cdSEric Blake return offset; 1492c1bb86cdSEric Blake } 1493c1bb86cdSEric Blake 1494999e6b69SKevin Wolf static int handle_aiocb_rw(void *opaque) 1495c1bb86cdSEric Blake { 1496999e6b69SKevin Wolf RawPosixAIOData *aiocb = opaque; 1497c1bb86cdSEric Blake ssize_t nbytes; 1498c1bb86cdSEric Blake char *buf; 1499c1bb86cdSEric Blake 1500c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { 1501c1bb86cdSEric Blake /* 1502c1bb86cdSEric Blake * If there is just a single buffer, and it is properly aligned 1503c1bb86cdSEric Blake * we can just use plain pread/pwrite without any problems. 1504c1bb86cdSEric Blake */ 1505d57c44d0SKevin Wolf if (aiocb->io.niov == 1) { 150654c7ca1bSKevin Wolf nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); 150754c7ca1bSKevin Wolf goto out; 1508c1bb86cdSEric Blake } 1509c1bb86cdSEric Blake /* 1510c1bb86cdSEric Blake * We have more than one iovec, and all are properly aligned. 1511c1bb86cdSEric Blake * 1512c1bb86cdSEric Blake * Try preadv/pwritev first and fall back to linearizing the 1513c1bb86cdSEric Blake * buffer if it's not supported. 1514c1bb86cdSEric Blake */ 1515c1bb86cdSEric Blake if (preadv_present) { 1516c1bb86cdSEric Blake nbytes = handle_aiocb_rw_vector(aiocb); 1517c1bb86cdSEric Blake if (nbytes == aiocb->aio_nbytes || 1518c1bb86cdSEric Blake (nbytes < 0 && nbytes != -ENOSYS)) { 151954c7ca1bSKevin Wolf goto out; 1520c1bb86cdSEric Blake } 1521c1bb86cdSEric Blake preadv_present = false; 1522c1bb86cdSEric Blake } 1523c1bb86cdSEric Blake 1524c1bb86cdSEric Blake /* 1525c1bb86cdSEric Blake * XXX(hch): short read/write. no easy way to handle the reminder 1526c1bb86cdSEric Blake * using these interfaces. For now retry using plain 1527c1bb86cdSEric Blake * pread/pwrite? 1528c1bb86cdSEric Blake */ 1529c1bb86cdSEric Blake } 1530c1bb86cdSEric Blake 1531c1bb86cdSEric Blake /* 1532c1bb86cdSEric Blake * Ok, we have to do it the hard way, copy all segments into 1533c1bb86cdSEric Blake * a single aligned buffer. 1534c1bb86cdSEric Blake */ 1535c1bb86cdSEric Blake buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); 1536c1bb86cdSEric Blake if (buf == NULL) { 153754c7ca1bSKevin Wolf nbytes = -ENOMEM; 153854c7ca1bSKevin Wolf goto out; 1539c1bb86cdSEric Blake } 1540c1bb86cdSEric Blake 1541c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1542c1bb86cdSEric Blake char *p = buf; 1543c1bb86cdSEric Blake int i; 1544c1bb86cdSEric Blake 1545d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov; ++i) { 1546d57c44d0SKevin Wolf memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); 1547d57c44d0SKevin Wolf p += aiocb->io.iov[i].iov_len; 1548c1bb86cdSEric Blake } 1549c1bb86cdSEric Blake assert(p - buf == aiocb->aio_nbytes); 1550c1bb86cdSEric Blake } 1551c1bb86cdSEric Blake 1552c1bb86cdSEric Blake nbytes = handle_aiocb_rw_linear(aiocb, buf); 1553c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { 1554c1bb86cdSEric Blake char *p = buf; 1555c1bb86cdSEric Blake size_t count = aiocb->aio_nbytes, copy; 1556c1bb86cdSEric Blake int i; 1557c1bb86cdSEric Blake 1558d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov && count; ++i) { 1559c1bb86cdSEric Blake copy = count; 1560d57c44d0SKevin Wolf if (copy > aiocb->io.iov[i].iov_len) { 1561d57c44d0SKevin Wolf copy = aiocb->io.iov[i].iov_len; 1562c1bb86cdSEric Blake } 1563d57c44d0SKevin Wolf memcpy(aiocb->io.iov[i].iov_base, p, copy); 1564c1bb86cdSEric Blake assert(count >= copy); 1565c1bb86cdSEric Blake p += copy; 1566c1bb86cdSEric Blake count -= copy; 1567c1bb86cdSEric Blake } 1568c1bb86cdSEric Blake assert(count == 0); 1569c1bb86cdSEric Blake } 1570c1bb86cdSEric Blake qemu_vfree(buf); 1571c1bb86cdSEric Blake 157254c7ca1bSKevin Wolf out: 157354c7ca1bSKevin Wolf if (nbytes == aiocb->aio_nbytes) { 157454c7ca1bSKevin Wolf return 0; 157554c7ca1bSKevin Wolf } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { 157654c7ca1bSKevin Wolf if (aiocb->aio_type & QEMU_AIO_WRITE) { 157754c7ca1bSKevin Wolf return -EINVAL; 157854c7ca1bSKevin Wolf } else { 157954c7ca1bSKevin Wolf iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, 158054c7ca1bSKevin Wolf 0, aiocb->aio_nbytes - nbytes); 158154c7ca1bSKevin Wolf return 0; 158254c7ca1bSKevin Wolf } 158354c7ca1bSKevin Wolf } else { 158454c7ca1bSKevin Wolf assert(nbytes < 0); 1585c1bb86cdSEric Blake return nbytes; 1586c1bb86cdSEric Blake } 158754c7ca1bSKevin Wolf } 1588c1bb86cdSEric Blake 1589c1bb86cdSEric Blake static int translate_err(int err) 1590c1bb86cdSEric Blake { 1591c1bb86cdSEric Blake if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || 1592c1bb86cdSEric Blake err == -ENOTTY) { 1593c1bb86cdSEric Blake err = -ENOTSUP; 1594c1bb86cdSEric Blake } 1595c1bb86cdSEric Blake return err; 1596c1bb86cdSEric Blake } 1597c1bb86cdSEric Blake 1598c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 1599c1bb86cdSEric Blake static int do_fallocate(int fd, int mode, off_t offset, off_t len) 1600c1bb86cdSEric Blake { 1601c1bb86cdSEric Blake do { 1602c1bb86cdSEric Blake if (fallocate(fd, mode, offset, len) == 0) { 1603c1bb86cdSEric Blake return 0; 1604c1bb86cdSEric Blake } 1605c1bb86cdSEric Blake } while (errno == EINTR); 1606c1bb86cdSEric Blake return translate_err(-errno); 1607c1bb86cdSEric Blake } 1608c1bb86cdSEric Blake #endif 1609c1bb86cdSEric Blake 1610c1bb86cdSEric Blake static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) 1611c1bb86cdSEric Blake { 1612c1bb86cdSEric Blake int ret = -ENOTSUP; 1613c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1614c1bb86cdSEric Blake 1615c1bb86cdSEric Blake if (!s->has_write_zeroes) { 1616c1bb86cdSEric Blake return -ENOTSUP; 1617c1bb86cdSEric Blake } 1618c1bb86cdSEric Blake 1619c1bb86cdSEric Blake #ifdef BLKZEROOUT 1620738301e1SKevin Wolf /* The BLKZEROOUT implementation in the kernel doesn't set 1621738301e1SKevin Wolf * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow 1622738301e1SKevin Wolf * fallbacks. */ 1623738301e1SKevin Wolf if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { 1624c1bb86cdSEric Blake do { 1625c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1626c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { 1627c1bb86cdSEric Blake return 0; 1628c1bb86cdSEric Blake } 1629c1bb86cdSEric Blake } while (errno == EINTR); 1630c1bb86cdSEric Blake 1631c1bb86cdSEric Blake ret = translate_err(-errno); 1632c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1633c1bb86cdSEric Blake s->has_write_zeroes = false; 1634c1bb86cdSEric Blake } 1635effecce6SKevin Wolf } 1636effecce6SKevin Wolf #endif 1637effecce6SKevin Wolf 1638c1bb86cdSEric Blake return ret; 1639c1bb86cdSEric Blake } 1640c1bb86cdSEric Blake 16417154d8aeSKevin Wolf static int handle_aiocb_write_zeroes(void *opaque) 1642c1bb86cdSEric Blake { 16437154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 164470d9110bSDenis V. Lunev #ifdef CONFIG_FALLOCATE 1645b2c6f23fSMax Reitz BDRVRawState *s = aiocb->bs->opaque; 164670d9110bSDenis V. Lunev int64_t len; 164770d9110bSDenis V. Lunev #endif 1648c1bb86cdSEric Blake 1649c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1650c1bb86cdSEric Blake return handle_aiocb_write_zeroes_block(aiocb); 1651c1bb86cdSEric Blake } 1652c1bb86cdSEric Blake 1653c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_ZERO_RANGE 1654c1bb86cdSEric Blake if (s->has_write_zeroes) { 1655c1bb86cdSEric Blake int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, 1656c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1657294682ccSAndrey Shinkevich if (ret == -EINVAL) { 1658294682ccSAndrey Shinkevich /* 1659294682ccSAndrey Shinkevich * Allow falling back to pwrite for file systems that 1660294682ccSAndrey Shinkevich * do not support fallocate() for an unaligned byte range. 1661294682ccSAndrey Shinkevich */ 1662294682ccSAndrey Shinkevich return -ENOTSUP; 1663294682ccSAndrey Shinkevich } 1664c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1665c1bb86cdSEric Blake return ret; 1666c1bb86cdSEric Blake } 1667c1bb86cdSEric Blake s->has_write_zeroes = false; 1668c1bb86cdSEric Blake } 1669c1bb86cdSEric Blake #endif 1670c1bb86cdSEric Blake 1671c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1672c1bb86cdSEric Blake if (s->has_discard && s->has_fallocate) { 1673c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 1674c1bb86cdSEric Blake FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1675c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1676c1bb86cdSEric Blake if (ret == 0) { 1677c1bb86cdSEric Blake ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1678c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1679c1bb86cdSEric Blake return ret; 1680c1bb86cdSEric Blake } 1681c1bb86cdSEric Blake s->has_fallocate = false; 1682c1bb86cdSEric Blake } else if (ret != -ENOTSUP) { 1683c1bb86cdSEric Blake return ret; 1684c1bb86cdSEric Blake } else { 1685c1bb86cdSEric Blake s->has_discard = false; 1686c1bb86cdSEric Blake } 1687c1bb86cdSEric Blake } 1688c1bb86cdSEric Blake #endif 1689c1bb86cdSEric Blake 1690c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 169170d9110bSDenis V. Lunev /* Last resort: we are trying to extend the file with zeroed data. This 169270d9110bSDenis V. Lunev * can be done via fallocate(fd, 0) */ 169370d9110bSDenis V. Lunev len = bdrv_getlength(aiocb->bs); 169470d9110bSDenis V. Lunev if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { 1695c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1696c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1697c1bb86cdSEric Blake return ret; 1698c1bb86cdSEric Blake } 1699c1bb86cdSEric Blake s->has_fallocate = false; 1700c1bb86cdSEric Blake } 1701c1bb86cdSEric Blake #endif 1702c1bb86cdSEric Blake 1703c1bb86cdSEric Blake return -ENOTSUP; 1704c1bb86cdSEric Blake } 1705c1bb86cdSEric Blake 17067154d8aeSKevin Wolf static int handle_aiocb_write_zeroes_unmap(void *opaque) 170734fa110eSKevin Wolf { 17087154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 170934fa110eSKevin Wolf BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; 171034fa110eSKevin Wolf 171134fa110eSKevin Wolf /* First try to write zeros and unmap at the same time */ 171234fa110eSKevin Wolf 171334fa110eSKevin Wolf #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1714b3ac2b94SSimran Singhal int ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 171534fa110eSKevin Wolf aiocb->aio_offset, aiocb->aio_nbytes); 1716bae127d4SAntoine Damhet switch (ret) { 1717bae127d4SAntoine Damhet case -ENOTSUP: 1718bae127d4SAntoine Damhet case -EINVAL: 1719ece4fa91SMaxim Levitsky case -EBUSY: 1720bae127d4SAntoine Damhet break; 1721bae127d4SAntoine Damhet default: 172234fa110eSKevin Wolf return ret; 172334fa110eSKevin Wolf } 172434fa110eSKevin Wolf #endif 172534fa110eSKevin Wolf 172634fa110eSKevin Wolf /* If we couldn't manage to unmap while guaranteed that the area reads as 172734fa110eSKevin Wolf * all-zero afterwards, just write zeroes without unmapping */ 1728b3ac2b94SSimran Singhal return handle_aiocb_write_zeroes(aiocb); 172934fa110eSKevin Wolf } 173034fa110eSKevin Wolf 17311efad060SFam Zheng #ifndef HAVE_COPY_FILE_RANGE 17321efad060SFam Zheng static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, 17331efad060SFam Zheng off_t *out_off, size_t len, unsigned int flags) 17341efad060SFam Zheng { 17351efad060SFam Zheng #ifdef __NR_copy_file_range 17361efad060SFam Zheng return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, 17371efad060SFam Zheng out_off, len, flags); 17381efad060SFam Zheng #else 17391efad060SFam Zheng errno = ENOSYS; 17401efad060SFam Zheng return -1; 17411efad060SFam Zheng #endif 17421efad060SFam Zheng } 17431efad060SFam Zheng #endif 17441efad060SFam Zheng 174558a209c4SKevin Wolf static int handle_aiocb_copy_range(void *opaque) 17461efad060SFam Zheng { 174758a209c4SKevin Wolf RawPosixAIOData *aiocb = opaque; 17481efad060SFam Zheng uint64_t bytes = aiocb->aio_nbytes; 17491efad060SFam Zheng off_t in_off = aiocb->aio_offset; 1750d57c44d0SKevin Wolf off_t out_off = aiocb->copy_range.aio_offset2; 17511efad060SFam Zheng 17521efad060SFam Zheng while (bytes) { 17531efad060SFam Zheng ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, 1754d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, &out_off, 17551efad060SFam Zheng bytes, 0); 1756ecc983a5SFam Zheng trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, 1757d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, out_off, bytes, 1758d57c44d0SKevin Wolf 0, ret); 1759c436e3d0SFam Zheng if (ret == 0) { 1760c436e3d0SFam Zheng /* No progress (e.g. when beyond EOF), let the caller fall back to 1761c436e3d0SFam Zheng * buffer I/O. */ 1762c436e3d0SFam Zheng return -ENOSPC; 17631efad060SFam Zheng } 17641efad060SFam Zheng if (ret < 0) { 1765c436e3d0SFam Zheng switch (errno) { 1766c436e3d0SFam Zheng case ENOSYS: 17671efad060SFam Zheng return -ENOTSUP; 1768c436e3d0SFam Zheng case EINTR: 1769c436e3d0SFam Zheng continue; 1770c436e3d0SFam Zheng default: 17711efad060SFam Zheng return -errno; 17721efad060SFam Zheng } 17731efad060SFam Zheng } 17741efad060SFam Zheng bytes -= ret; 17751efad060SFam Zheng } 17761efad060SFam Zheng return 0; 17771efad060SFam Zheng } 17781efad060SFam Zheng 177946ee0f46SKevin Wolf static int handle_aiocb_discard(void *opaque) 1780c1bb86cdSEric Blake { 178146ee0f46SKevin Wolf RawPosixAIOData *aiocb = opaque; 1782c1bb86cdSEric Blake int ret = -EOPNOTSUPP; 1783c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1784c1bb86cdSEric Blake 1785c1bb86cdSEric Blake if (!s->has_discard) { 1786c1bb86cdSEric Blake return -ENOTSUP; 1787c1bb86cdSEric Blake } 1788c1bb86cdSEric Blake 1789c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1790c1bb86cdSEric Blake #ifdef BLKDISCARD 1791c1bb86cdSEric Blake do { 1792c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1793c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { 1794c1bb86cdSEric Blake return 0; 1795c1bb86cdSEric Blake } 1796c1bb86cdSEric Blake } while (errno == EINTR); 1797c1bb86cdSEric Blake 1798c1bb86cdSEric Blake ret = -errno; 1799c1bb86cdSEric Blake #endif 1800c1bb86cdSEric Blake } else { 1801c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1802c1bb86cdSEric Blake ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1803c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1804c1bb86cdSEric Blake #endif 1805c1bb86cdSEric Blake } 1806c1bb86cdSEric Blake 1807c1bb86cdSEric Blake ret = translate_err(ret); 1808c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1809c1bb86cdSEric Blake s->has_discard = false; 1810c1bb86cdSEric Blake } 1811c1bb86cdSEric Blake return ret; 1812c1bb86cdSEric Blake } 1813c1bb86cdSEric Blake 18143a20013fSNir Soffer /* 18153a20013fSNir Soffer * Help alignment probing by allocating the first block. 18163a20013fSNir Soffer * 18173a20013fSNir Soffer * When reading with direct I/O from unallocated area on Gluster backed by XFS, 18183a20013fSNir Soffer * reading succeeds regardless of request length. In this case we fallback to 18193a20013fSNir Soffer * safe alignment which is not optimal. Allocating the first block avoids this 18203a20013fSNir Soffer * fallback. 18213a20013fSNir Soffer * 18223a20013fSNir Soffer * fd may be opened with O_DIRECT, but we don't know the buffer alignment or 18233a20013fSNir Soffer * request alignment, so we use safe values. 18243a20013fSNir Soffer * 18253a20013fSNir Soffer * Returns: 0 on success, -errno on failure. Since this is an optimization, 18263a20013fSNir Soffer * caller may ignore failures. 18273a20013fSNir Soffer */ 18283a20013fSNir Soffer static int allocate_first_block(int fd, size_t max_size) 18293a20013fSNir Soffer { 18303a20013fSNir Soffer size_t write_size = (max_size < MAX_BLOCKSIZE) 18313a20013fSNir Soffer ? BDRV_SECTOR_SIZE 18323a20013fSNir Soffer : MAX_BLOCKSIZE; 1833038adc2fSWei Yang size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); 18343a20013fSNir Soffer void *buf; 18353a20013fSNir Soffer ssize_t n; 18363a20013fSNir Soffer int ret; 18373a20013fSNir Soffer 18383a20013fSNir Soffer buf = qemu_memalign(max_align, write_size); 18393a20013fSNir Soffer memset(buf, 0, write_size); 18403a20013fSNir Soffer 18413a20013fSNir Soffer do { 18423a20013fSNir Soffer n = pwrite(fd, buf, write_size, 0); 18433a20013fSNir Soffer } while (n == -1 && errno == EINTR); 18443a20013fSNir Soffer 18453a20013fSNir Soffer ret = (n == -1) ? -errno : 0; 18463a20013fSNir Soffer 18473a20013fSNir Soffer qemu_vfree(buf); 18483a20013fSNir Soffer return ret; 18493a20013fSNir Soffer } 18503a20013fSNir Soffer 185129cb4c01SKevin Wolf static int handle_aiocb_truncate(void *opaque) 185293f4e2ffSKevin Wolf { 185329cb4c01SKevin Wolf RawPosixAIOData *aiocb = opaque; 185493f4e2ffSKevin Wolf int result = 0; 185593f4e2ffSKevin Wolf int64_t current_length = 0; 185693f4e2ffSKevin Wolf char *buf = NULL; 185793f4e2ffSKevin Wolf struct stat st; 185893f4e2ffSKevin Wolf int fd = aiocb->aio_fildes; 185993f4e2ffSKevin Wolf int64_t offset = aiocb->aio_offset; 1860d57c44d0SKevin Wolf PreallocMode prealloc = aiocb->truncate.prealloc; 1861d57c44d0SKevin Wolf Error **errp = aiocb->truncate.errp; 186293f4e2ffSKevin Wolf 186393f4e2ffSKevin Wolf if (fstat(fd, &st) < 0) { 186493f4e2ffSKevin Wolf result = -errno; 186593f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not stat file"); 186693f4e2ffSKevin Wolf return result; 186793f4e2ffSKevin Wolf } 186893f4e2ffSKevin Wolf 186993f4e2ffSKevin Wolf current_length = st.st_size; 1870d57c44d0SKevin Wolf if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { 187193f4e2ffSKevin Wolf error_setg(errp, "Cannot use preallocation for shrinking files"); 187293f4e2ffSKevin Wolf return -ENOTSUP; 187393f4e2ffSKevin Wolf } 187493f4e2ffSKevin Wolf 1875d57c44d0SKevin Wolf switch (prealloc) { 187693f4e2ffSKevin Wolf #ifdef CONFIG_POSIX_FALLOCATE 187793f4e2ffSKevin Wolf case PREALLOC_MODE_FALLOC: 187893f4e2ffSKevin Wolf /* 187993f4e2ffSKevin Wolf * Truncating before posix_fallocate() makes it about twice slower on 188093f4e2ffSKevin Wolf * file systems that do not support fallocate(), trying to check if a 188193f4e2ffSKevin Wolf * block is allocated before allocating it, so don't do that here. 188293f4e2ffSKevin Wolf */ 188393f4e2ffSKevin Wolf if (offset != current_length) { 188493f4e2ffSKevin Wolf result = -posix_fallocate(fd, current_length, 188593f4e2ffSKevin Wolf offset - current_length); 188693f4e2ffSKevin Wolf if (result != 0) { 188793f4e2ffSKevin Wolf /* posix_fallocate() doesn't set errno. */ 188893f4e2ffSKevin Wolf error_setg_errno(errp, -result, 188993f4e2ffSKevin Wolf "Could not preallocate new data"); 18903a20013fSNir Soffer } else if (current_length == 0) { 18913a20013fSNir Soffer /* 18923a20013fSNir Soffer * posix_fallocate() uses fallocate() if the filesystem 18933a20013fSNir Soffer * supports it, or fallback to manually writing zeroes. If 18943a20013fSNir Soffer * fallocate() was used, unaligned reads from the fallocated 18953a20013fSNir Soffer * area in raw_probe_alignment() will succeed, hence we need to 18963a20013fSNir Soffer * allocate the first block. 18973a20013fSNir Soffer * 18983a20013fSNir Soffer * Optimize future alignment probing; ignore failures. 18993a20013fSNir Soffer */ 19003a20013fSNir Soffer allocate_first_block(fd, offset); 190193f4e2ffSKevin Wolf } 190293f4e2ffSKevin Wolf } else { 190393f4e2ffSKevin Wolf result = 0; 190493f4e2ffSKevin Wolf } 190593f4e2ffSKevin Wolf goto out; 190693f4e2ffSKevin Wolf #endif 190793f4e2ffSKevin Wolf case PREALLOC_MODE_FULL: 190893f4e2ffSKevin Wolf { 190993f4e2ffSKevin Wolf int64_t num = 0, left = offset - current_length; 191093f4e2ffSKevin Wolf off_t seek_result; 191193f4e2ffSKevin Wolf 191293f4e2ffSKevin Wolf /* 191393f4e2ffSKevin Wolf * Knowing the final size from the beginning could allow the file 191493f4e2ffSKevin Wolf * system driver to do less allocations and possibly avoid 191593f4e2ffSKevin Wolf * fragmentation of the file. 191693f4e2ffSKevin Wolf */ 191793f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 191893f4e2ffSKevin Wolf result = -errno; 191993f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 192093f4e2ffSKevin Wolf goto out; 192193f4e2ffSKevin Wolf } 192293f4e2ffSKevin Wolf 192393f4e2ffSKevin Wolf buf = g_malloc0(65536); 192493f4e2ffSKevin Wolf 192593f4e2ffSKevin Wolf seek_result = lseek(fd, current_length, SEEK_SET); 192693f4e2ffSKevin Wolf if (seek_result < 0) { 192793f4e2ffSKevin Wolf result = -errno; 192893f4e2ffSKevin Wolf error_setg_errno(errp, -result, 192993f4e2ffSKevin Wolf "Failed to seek to the old end of file"); 193093f4e2ffSKevin Wolf goto out; 193193f4e2ffSKevin Wolf } 193293f4e2ffSKevin Wolf 193393f4e2ffSKevin Wolf while (left > 0) { 193493f4e2ffSKevin Wolf num = MIN(left, 65536); 193593f4e2ffSKevin Wolf result = write(fd, buf, num); 193693f4e2ffSKevin Wolf if (result < 0) { 1937a1c81f4fSFam Zheng if (errno == EINTR) { 1938a1c81f4fSFam Zheng continue; 1939a1c81f4fSFam Zheng } 194093f4e2ffSKevin Wolf result = -errno; 194193f4e2ffSKevin Wolf error_setg_errno(errp, -result, 194293f4e2ffSKevin Wolf "Could not write zeros for preallocation"); 194393f4e2ffSKevin Wolf goto out; 194493f4e2ffSKevin Wolf } 194593f4e2ffSKevin Wolf left -= result; 194693f4e2ffSKevin Wolf } 194793f4e2ffSKevin Wolf if (result >= 0) { 194893f4e2ffSKevin Wolf result = fsync(fd); 194993f4e2ffSKevin Wolf if (result < 0) { 195093f4e2ffSKevin Wolf result = -errno; 195193f4e2ffSKevin Wolf error_setg_errno(errp, -result, 195293f4e2ffSKevin Wolf "Could not flush file to disk"); 195393f4e2ffSKevin Wolf goto out; 195493f4e2ffSKevin Wolf } 195593f4e2ffSKevin Wolf } 195693f4e2ffSKevin Wolf goto out; 195793f4e2ffSKevin Wolf } 195893f4e2ffSKevin Wolf case PREALLOC_MODE_OFF: 195993f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 196093f4e2ffSKevin Wolf result = -errno; 196193f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 19623a20013fSNir Soffer } else if (current_length == 0 && offset > current_length) { 19633a20013fSNir Soffer /* Optimize future alignment probing; ignore failures. */ 19643a20013fSNir Soffer allocate_first_block(fd, offset); 196593f4e2ffSKevin Wolf } 196693f4e2ffSKevin Wolf return result; 196793f4e2ffSKevin Wolf default: 196893f4e2ffSKevin Wolf result = -ENOTSUP; 196993f4e2ffSKevin Wolf error_setg(errp, "Unsupported preallocation mode: %s", 1970d57c44d0SKevin Wolf PreallocMode_str(prealloc)); 197193f4e2ffSKevin Wolf return result; 197293f4e2ffSKevin Wolf } 197393f4e2ffSKevin Wolf 197493f4e2ffSKevin Wolf out: 197593f4e2ffSKevin Wolf if (result < 0) { 197693f4e2ffSKevin Wolf if (ftruncate(fd, current_length) < 0) { 197793f4e2ffSKevin Wolf error_report("Failed to restore old file length: %s", 197893f4e2ffSKevin Wolf strerror(errno)); 197993f4e2ffSKevin Wolf } 198093f4e2ffSKevin Wolf } 198193f4e2ffSKevin Wolf 198293f4e2ffSKevin Wolf g_free(buf); 198393f4e2ffSKevin Wolf return result; 198493f4e2ffSKevin Wolf } 198593f4e2ffSKevin Wolf 19865d5de250SKevin Wolf static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs, 19875d5de250SKevin Wolf ThreadPoolFunc func, void *arg) 19885d5de250SKevin Wolf { 19895d5de250SKevin Wolf /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ 19905d5de250SKevin Wolf ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 19915d5de250SKevin Wolf return thread_pool_submit_co(pool, func, arg); 19925d5de250SKevin Wolf } 19935d5de250SKevin Wolf 1994c1bb86cdSEric Blake static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, 1995c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, int type) 1996c1bb86cdSEric Blake { 1997c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1998999e6b69SKevin Wolf RawPosixAIOData acb; 1999c1bb86cdSEric Blake 2000c1bb86cdSEric Blake if (fd_open(bs) < 0) 2001c1bb86cdSEric Blake return -EIO; 2002c1bb86cdSEric Blake 2003c1bb86cdSEric Blake /* 2004c6447510SAarushi Mehta * When using O_DIRECT, the request must be aligned to be able to use 2005c6447510SAarushi Mehta * either libaio or io_uring interface. If not fail back to regular thread 2006c6447510SAarushi Mehta * pool read/write code which emulates this for us if we 2007c6447510SAarushi Mehta * set QEMU_AIO_MISALIGNED. 2008c1bb86cdSEric Blake */ 2009c6447510SAarushi Mehta if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) { 2010c1bb86cdSEric Blake type |= QEMU_AIO_MISALIGNED; 2011c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2012c6447510SAarushi Mehta } else if (s->use_linux_io_uring) { 2013c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2014c6447510SAarushi Mehta assert(qiov->size == bytes); 2015c6447510SAarushi Mehta return luring_co_submit(bs, aio, s->fd, offset, qiov, type); 2016c6447510SAarushi Mehta #endif 2017c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 2018c1bb86cdSEric Blake } else if (s->use_linux_aio) { 2019c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 2020c1bb86cdSEric Blake assert(qiov->size == bytes); 2021c1bb86cdSEric Blake return laio_co_submit(bs, aio, s->fd, offset, qiov, type); 2022c1bb86cdSEric Blake #endif 2023c1bb86cdSEric Blake } 2024c1bb86cdSEric Blake 2025999e6b69SKevin Wolf acb = (RawPosixAIOData) { 2026999e6b69SKevin Wolf .bs = bs, 2027999e6b69SKevin Wolf .aio_fildes = s->fd, 2028999e6b69SKevin Wolf .aio_type = type, 2029999e6b69SKevin Wolf .aio_offset = offset, 2030999e6b69SKevin Wolf .aio_nbytes = bytes, 2031999e6b69SKevin Wolf .io = { 2032999e6b69SKevin Wolf .iov = qiov->iov, 2033999e6b69SKevin Wolf .niov = qiov->niov, 2034999e6b69SKevin Wolf }, 2035999e6b69SKevin Wolf }; 2036999e6b69SKevin Wolf 2037999e6b69SKevin Wolf assert(qiov->size == bytes); 2038999e6b69SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb); 2039c1bb86cdSEric Blake } 2040c1bb86cdSEric Blake 2041c1bb86cdSEric Blake static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset, 2042c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 2043c1bb86cdSEric Blake int flags) 2044c1bb86cdSEric Blake { 2045c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); 2046c1bb86cdSEric Blake } 2047c1bb86cdSEric Blake 2048c1bb86cdSEric Blake static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset, 2049c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 2050c1bb86cdSEric Blake int flags) 2051c1bb86cdSEric Blake { 2052c1bb86cdSEric Blake assert(flags == 0); 2053c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); 2054c1bb86cdSEric Blake } 2055c1bb86cdSEric Blake 2056c1bb86cdSEric Blake static void raw_aio_plug(BlockDriverState *bs) 2057c1bb86cdSEric Blake { 2058c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 2059c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 2060c1bb86cdSEric Blake if (s->use_linux_aio) { 2061c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 2062c1bb86cdSEric Blake laio_io_plug(bs, aio); 2063c1bb86cdSEric Blake } 2064c1bb86cdSEric Blake #endif 2065c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2066c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2067c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2068c6447510SAarushi Mehta luring_io_plug(bs, aio); 2069c6447510SAarushi Mehta } 2070c6447510SAarushi Mehta #endif 2071c1bb86cdSEric Blake } 2072c1bb86cdSEric Blake 2073c1bb86cdSEric Blake static void raw_aio_unplug(BlockDriverState *bs) 2074c1bb86cdSEric Blake { 2075c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 2076c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 2077c1bb86cdSEric Blake if (s->use_linux_aio) { 2078c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 2079c1bb86cdSEric Blake laio_io_unplug(bs, aio); 2080c1bb86cdSEric Blake } 2081c1bb86cdSEric Blake #endif 2082c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2083c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2084c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2085c6447510SAarushi Mehta luring_io_unplug(bs, aio); 2086c6447510SAarushi Mehta } 2087c6447510SAarushi Mehta #endif 2088c1bb86cdSEric Blake } 2089c1bb86cdSEric Blake 209033d70fb6SKevin Wolf static int raw_co_flush_to_disk(BlockDriverState *bs) 2091c1bb86cdSEric Blake { 2092c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 209306dc9bd5SKevin Wolf RawPosixAIOData acb; 209433d70fb6SKevin Wolf int ret; 2095c1bb86cdSEric Blake 209633d70fb6SKevin Wolf ret = fd_open(bs); 209733d70fb6SKevin Wolf if (ret < 0) { 209833d70fb6SKevin Wolf return ret; 209933d70fb6SKevin Wolf } 2100c1bb86cdSEric Blake 210106dc9bd5SKevin Wolf acb = (RawPosixAIOData) { 210206dc9bd5SKevin Wolf .bs = bs, 210306dc9bd5SKevin Wolf .aio_fildes = s->fd, 210406dc9bd5SKevin Wolf .aio_type = QEMU_AIO_FLUSH, 210506dc9bd5SKevin Wolf }; 210606dc9bd5SKevin Wolf 2107c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2108c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2109c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2110c6447510SAarushi Mehta return luring_co_submit(bs, aio, s->fd, 0, NULL, QEMU_AIO_FLUSH); 2111c6447510SAarushi Mehta } 2112c6447510SAarushi Mehta #endif 211306dc9bd5SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb); 2114c1bb86cdSEric Blake } 2115c1bb86cdSEric Blake 2116ed6e2161SNishanth Aravamudan static void raw_aio_attach_aio_context(BlockDriverState *bs, 2117ed6e2161SNishanth Aravamudan AioContext *new_context) 2118ed6e2161SNishanth Aravamudan { 2119c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 2120ed6e2161SNishanth Aravamudan #ifdef CONFIG_LINUX_AIO 2121ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 2122cb09104eSMarkus Armbruster Error *local_err = NULL; 2123ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(new_context, &local_err)) { 2124ed6e2161SNishanth Aravamudan error_reportf_err(local_err, "Unable to use native AIO, " 2125ed6e2161SNishanth Aravamudan "falling back to thread pool: "); 2126ed6e2161SNishanth Aravamudan s->use_linux_aio = false; 2127ed6e2161SNishanth Aravamudan } 2128ed6e2161SNishanth Aravamudan } 2129ed6e2161SNishanth Aravamudan #endif 2130c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2131c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2132cb8d0851SPan Nengyuan Error *local_err = NULL; 2133c6447510SAarushi Mehta if (!aio_setup_linux_io_uring(new_context, &local_err)) { 2134c6447510SAarushi Mehta error_reportf_err(local_err, "Unable to use linux io_uring, " 2135c6447510SAarushi Mehta "falling back to thread pool: "); 2136c6447510SAarushi Mehta s->use_linux_io_uring = false; 2137c6447510SAarushi Mehta } 2138c6447510SAarushi Mehta } 2139c6447510SAarushi Mehta #endif 2140ed6e2161SNishanth Aravamudan } 2141ed6e2161SNishanth Aravamudan 2142c1bb86cdSEric Blake static void raw_close(BlockDriverState *bs) 2143c1bb86cdSEric Blake { 2144c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2145c1bb86cdSEric Blake 2146c1bb86cdSEric Blake if (s->fd >= 0) { 2147c1bb86cdSEric Blake qemu_close(s->fd); 2148c1bb86cdSEric Blake s->fd = -1; 2149c1bb86cdSEric Blake } 2150c1bb86cdSEric Blake } 2151c1bb86cdSEric Blake 2152d0bc9e5dSMax Reitz /** 2153d0bc9e5dSMax Reitz * Truncates the given regular file @fd to @offset and, when growing, fills the 2154d0bc9e5dSMax Reitz * new space according to @prealloc. 2155d0bc9e5dSMax Reitz * 2156d0bc9e5dSMax Reitz * Returns: 0 on success, -errno on failure. 2157d0bc9e5dSMax Reitz */ 215893f4e2ffSKevin Wolf static int coroutine_fn 215993f4e2ffSKevin Wolf raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, 216093f4e2ffSKevin Wolf PreallocMode prealloc, Error **errp) 21619f63b07eSMax Reitz { 216229cb4c01SKevin Wolf RawPosixAIOData acb; 2163d0bc9e5dSMax Reitz 216429cb4c01SKevin Wolf acb = (RawPosixAIOData) { 216593f4e2ffSKevin Wolf .bs = bs, 216693f4e2ffSKevin Wolf .aio_fildes = fd, 216793f4e2ffSKevin Wolf .aio_type = QEMU_AIO_TRUNCATE, 216893f4e2ffSKevin Wolf .aio_offset = offset, 2169d57c44d0SKevin Wolf .truncate = { 217093f4e2ffSKevin Wolf .prealloc = prealloc, 217193f4e2ffSKevin Wolf .errp = errp, 2172d57c44d0SKevin Wolf }, 217393f4e2ffSKevin Wolf }; 2174d0bc9e5dSMax Reitz 217529cb4c01SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb); 21769f63b07eSMax Reitz } 21779f63b07eSMax Reitz 2178061ca8a3SKevin Wolf static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, 2179c80d8b06SMax Reitz bool exact, PreallocMode prealloc, 218092b92799SKevin Wolf BdrvRequestFlags flags, Error **errp) 2181c1bb86cdSEric Blake { 2182c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2183c1bb86cdSEric Blake struct stat st; 2184f59adb32SMax Reitz int ret; 2185c1bb86cdSEric Blake 2186c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 2187f59adb32SMax Reitz ret = -errno; 2188f59adb32SMax Reitz error_setg_errno(errp, -ret, "Failed to fstat() the file"); 2189f59adb32SMax Reitz return ret; 2190c1bb86cdSEric Blake } 2191c1bb86cdSEric Blake 2192c1bb86cdSEric Blake if (S_ISREG(st.st_mode)) { 219382325ae5SMax Reitz /* Always resizes to the exact @offset */ 219493f4e2ffSKevin Wolf return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); 2195c1bb86cdSEric Blake } 219635d72602SMax Reitz 219735d72602SMax Reitz if (prealloc != PREALLOC_MODE_OFF) { 219835d72602SMax Reitz error_setg(errp, "Preallocation mode '%s' unsupported for this " 2199977c736fSMarkus Armbruster "non-regular file", PreallocMode_str(prealloc)); 220035d72602SMax Reitz return -ENOTSUP; 220135d72602SMax Reitz } 220235d72602SMax Reitz 220335d72602SMax Reitz if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 220482325ae5SMax Reitz int64_t cur_length = raw_getlength(bs); 220582325ae5SMax Reitz 220682325ae5SMax Reitz if (offset != cur_length && exact) { 220782325ae5SMax Reitz error_setg(errp, "Cannot resize device files"); 220882325ae5SMax Reitz return -ENOTSUP; 220982325ae5SMax Reitz } else if (offset > cur_length) { 2210f59adb32SMax Reitz error_setg(errp, "Cannot grow device files"); 2211c1bb86cdSEric Blake return -EINVAL; 2212c1bb86cdSEric Blake } 2213c1bb86cdSEric Blake } else { 2214f59adb32SMax Reitz error_setg(errp, "Resizing this file is not supported"); 2215c1bb86cdSEric Blake return -ENOTSUP; 2216c1bb86cdSEric Blake } 2217c1bb86cdSEric Blake 2218c1bb86cdSEric Blake return 0; 2219c1bb86cdSEric Blake } 2220c1bb86cdSEric Blake 2221c1bb86cdSEric Blake #ifdef __OpenBSD__ 2222c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2223c1bb86cdSEric Blake { 2224c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2225c1bb86cdSEric Blake int fd = s->fd; 2226c1bb86cdSEric Blake struct stat st; 2227c1bb86cdSEric Blake 2228c1bb86cdSEric Blake if (fstat(fd, &st)) 2229c1bb86cdSEric Blake return -errno; 2230c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2231c1bb86cdSEric Blake struct disklabel dl; 2232c1bb86cdSEric Blake 2233c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2234c1bb86cdSEric Blake return -errno; 2235c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2236c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2237c1bb86cdSEric Blake } else 2238c1bb86cdSEric Blake return st.st_size; 2239c1bb86cdSEric Blake } 2240c1bb86cdSEric Blake #elif defined(__NetBSD__) 2241c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2242c1bb86cdSEric Blake { 2243c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2244c1bb86cdSEric Blake int fd = s->fd; 2245c1bb86cdSEric Blake struct stat st; 2246c1bb86cdSEric Blake 2247c1bb86cdSEric Blake if (fstat(fd, &st)) 2248c1bb86cdSEric Blake return -errno; 2249c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2250c1bb86cdSEric Blake struct dkwedge_info dkw; 2251c1bb86cdSEric Blake 2252c1bb86cdSEric Blake if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { 2253c1bb86cdSEric Blake return dkw.dkw_size * 512; 2254c1bb86cdSEric Blake } else { 2255c1bb86cdSEric Blake struct disklabel dl; 2256c1bb86cdSEric Blake 2257c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2258c1bb86cdSEric Blake return -errno; 2259c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2260c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2261c1bb86cdSEric Blake } 2262c1bb86cdSEric Blake } else 2263c1bb86cdSEric Blake return st.st_size; 2264c1bb86cdSEric Blake } 2265c1bb86cdSEric Blake #elif defined(__sun__) 2266c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2267c1bb86cdSEric Blake { 2268c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2269c1bb86cdSEric Blake struct dk_minfo minfo; 2270c1bb86cdSEric Blake int ret; 2271c1bb86cdSEric Blake int64_t size; 2272c1bb86cdSEric Blake 2273c1bb86cdSEric Blake ret = fd_open(bs); 2274c1bb86cdSEric Blake if (ret < 0) { 2275c1bb86cdSEric Blake return ret; 2276c1bb86cdSEric Blake } 2277c1bb86cdSEric Blake 2278c1bb86cdSEric Blake /* 2279c1bb86cdSEric Blake * Use the DKIOCGMEDIAINFO ioctl to read the size. 2280c1bb86cdSEric Blake */ 2281c1bb86cdSEric Blake ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); 2282c1bb86cdSEric Blake if (ret != -1) { 2283c1bb86cdSEric Blake return minfo.dki_lbsize * minfo.dki_capacity; 2284c1bb86cdSEric Blake } 2285c1bb86cdSEric Blake 2286c1bb86cdSEric Blake /* 2287c1bb86cdSEric Blake * There are reports that lseek on some devices fails, but 2288c1bb86cdSEric Blake * irc discussion said that contingency on contingency was overkill. 2289c1bb86cdSEric Blake */ 2290c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2291c1bb86cdSEric Blake if (size < 0) { 2292c1bb86cdSEric Blake return -errno; 2293c1bb86cdSEric Blake } 2294c1bb86cdSEric Blake return size; 2295c1bb86cdSEric Blake } 2296c1bb86cdSEric Blake #elif defined(CONFIG_BSD) 2297c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2298c1bb86cdSEric Blake { 2299c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2300c1bb86cdSEric Blake int fd = s->fd; 2301c1bb86cdSEric Blake int64_t size; 2302c1bb86cdSEric Blake struct stat sb; 2303c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2304c1bb86cdSEric Blake int reopened = 0; 2305c1bb86cdSEric Blake #endif 2306c1bb86cdSEric Blake int ret; 2307c1bb86cdSEric Blake 2308c1bb86cdSEric Blake ret = fd_open(bs); 2309c1bb86cdSEric Blake if (ret < 0) 2310c1bb86cdSEric Blake return ret; 2311c1bb86cdSEric Blake 2312c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2313c1bb86cdSEric Blake again: 2314c1bb86cdSEric Blake #endif 2315c1bb86cdSEric Blake if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { 2316c1bb86cdSEric Blake #ifdef DIOCGMEDIASIZE 2317c1bb86cdSEric Blake if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) 2318c1bb86cdSEric Blake #elif defined(DIOCGPART) 2319c1bb86cdSEric Blake { 2320c1bb86cdSEric Blake struct partinfo pi; 2321c1bb86cdSEric Blake if (ioctl(fd, DIOCGPART, &pi) == 0) 2322c1bb86cdSEric Blake size = pi.media_size; 2323c1bb86cdSEric Blake else 2324c1bb86cdSEric Blake size = 0; 2325c1bb86cdSEric Blake } 2326c1bb86cdSEric Blake if (size == 0) 2327c1bb86cdSEric Blake #endif 2328c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2329c1bb86cdSEric Blake { 2330c1bb86cdSEric Blake uint64_t sectors = 0; 2331c1bb86cdSEric Blake uint32_t sector_size = 0; 2332c1bb86cdSEric Blake 2333c1bb86cdSEric Blake if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 2334c1bb86cdSEric Blake && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { 2335c1bb86cdSEric Blake size = sectors * sector_size; 2336c1bb86cdSEric Blake } else { 2337c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2338c1bb86cdSEric Blake if (size < 0) { 2339c1bb86cdSEric Blake return -errno; 2340c1bb86cdSEric Blake } 2341c1bb86cdSEric Blake } 2342c1bb86cdSEric Blake } 2343c1bb86cdSEric Blake #else 2344c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2345c1bb86cdSEric Blake if (size < 0) { 2346c1bb86cdSEric Blake return -errno; 2347c1bb86cdSEric Blake } 2348c1bb86cdSEric Blake #endif 2349c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 2350c1bb86cdSEric Blake switch(s->type) { 2351c1bb86cdSEric Blake case FTYPE_CD: 2352c1bb86cdSEric Blake /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ 2353c1bb86cdSEric Blake if (size == 2048LL * (unsigned)-1) 2354c1bb86cdSEric Blake size = 0; 2355c1bb86cdSEric Blake /* XXX no disc? maybe we need to reopen... */ 2356c1bb86cdSEric Blake if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { 2357c1bb86cdSEric Blake reopened = 1; 2358c1bb86cdSEric Blake goto again; 2359c1bb86cdSEric Blake } 2360c1bb86cdSEric Blake } 2361c1bb86cdSEric Blake #endif 2362c1bb86cdSEric Blake } else { 2363c1bb86cdSEric Blake size = lseek(fd, 0, SEEK_END); 2364c1bb86cdSEric Blake if (size < 0) { 2365c1bb86cdSEric Blake return -errno; 2366c1bb86cdSEric Blake } 2367c1bb86cdSEric Blake } 2368c1bb86cdSEric Blake return size; 2369c1bb86cdSEric Blake } 2370c1bb86cdSEric Blake #else 2371c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2372c1bb86cdSEric Blake { 2373c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2374c1bb86cdSEric Blake int ret; 2375c1bb86cdSEric Blake int64_t size; 2376c1bb86cdSEric Blake 2377c1bb86cdSEric Blake ret = fd_open(bs); 2378c1bb86cdSEric Blake if (ret < 0) { 2379c1bb86cdSEric Blake return ret; 2380c1bb86cdSEric Blake } 2381c1bb86cdSEric Blake 2382c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2383c1bb86cdSEric Blake if (size < 0) { 2384c1bb86cdSEric Blake return -errno; 2385c1bb86cdSEric Blake } 2386c1bb86cdSEric Blake return size; 2387c1bb86cdSEric Blake } 2388c1bb86cdSEric Blake #endif 2389c1bb86cdSEric Blake 2390c1bb86cdSEric Blake static int64_t raw_get_allocated_file_size(BlockDriverState *bs) 2391c1bb86cdSEric Blake { 2392c1bb86cdSEric Blake struct stat st; 2393c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2394c1bb86cdSEric Blake 2395c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 2396c1bb86cdSEric Blake return -errno; 2397c1bb86cdSEric Blake } 2398c1bb86cdSEric Blake return (int64_t)st.st_blocks * 512; 2399c1bb86cdSEric Blake } 2400c1bb86cdSEric Blake 240193f4e2ffSKevin Wolf static int coroutine_fn 240293f4e2ffSKevin Wolf raw_co_create(BlockdevCreateOptions *options, Error **errp) 2403c1bb86cdSEric Blake { 2404927f11e1SKevin Wolf BlockdevCreateOptionsFile *file_opts; 24057c20c808SMax Reitz Error *local_err = NULL; 2406c1bb86cdSEric Blake int fd; 2407d815efcaSMax Reitz uint64_t perm, shared; 2408c1bb86cdSEric Blake int result = 0; 2409c1bb86cdSEric Blake 2410927f11e1SKevin Wolf /* Validate options and set default values */ 2411927f11e1SKevin Wolf assert(options->driver == BLOCKDEV_DRIVER_FILE); 2412927f11e1SKevin Wolf file_opts = &options->u.file; 2413c1bb86cdSEric Blake 2414927f11e1SKevin Wolf if (!file_opts->has_nocow) { 2415927f11e1SKevin Wolf file_opts->nocow = false; 2416927f11e1SKevin Wolf } 2417927f11e1SKevin Wolf if (!file_opts->has_preallocation) { 2418927f11e1SKevin Wolf file_opts->preallocation = PREALLOC_MODE_OFF; 2419c1bb86cdSEric Blake } 2420ffa244c8SKevin Wolf if (!file_opts->has_extent_size_hint) { 2421ffa244c8SKevin Wolf file_opts->extent_size_hint = 1 * MiB; 2422ffa244c8SKevin Wolf } 2423ffa244c8SKevin Wolf if (file_opts->extent_size_hint > UINT32_MAX) { 2424ffa244c8SKevin Wolf result = -EINVAL; 2425ffa244c8SKevin Wolf error_setg(errp, "Extent size hint is too large"); 2426ffa244c8SKevin Wolf goto out; 2427ffa244c8SKevin Wolf } 2428c1bb86cdSEric Blake 2429927f11e1SKevin Wolf /* Create file */ 2430b18a24a9SDaniel P. Berrangé fd = qemu_create(file_opts->filename, O_RDWR | O_BINARY, 0644, errp); 2431c1bb86cdSEric Blake if (fd < 0) { 2432c1bb86cdSEric Blake result = -errno; 2433c1bb86cdSEric Blake goto out; 2434c1bb86cdSEric Blake } 2435c1bb86cdSEric Blake 2436b8cf1913SMax Reitz /* Take permissions: We want to discard everything, so we need 2437b8cf1913SMax Reitz * BLK_PERM_WRITE; and truncation to the desired size requires 2438b8cf1913SMax Reitz * BLK_PERM_RESIZE. 2439b8cf1913SMax Reitz * On the other hand, we cannot share the RESIZE permission 2440b8cf1913SMax Reitz * because we promise that after this function, the file has the 2441b8cf1913SMax Reitz * size given in the options. If someone else were to resize it 2442b8cf1913SMax Reitz * concurrently, we could not guarantee that. 2443b8cf1913SMax Reitz * Note that after this function, we can no longer guarantee that 2444b8cf1913SMax Reitz * the file is not touched by a third party, so it may be resized 2445b8cf1913SMax Reitz * then. */ 2446b8cf1913SMax Reitz perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; 2447b8cf1913SMax Reitz shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; 2448b8cf1913SMax Reitz 2449b8cf1913SMax Reitz /* Step one: Take locks */ 24502996ffadSFam Zheng result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); 2451b8cf1913SMax Reitz if (result < 0) { 2452b8cf1913SMax Reitz goto out_close; 2453b8cf1913SMax Reitz } 2454b8cf1913SMax Reitz 2455b8cf1913SMax Reitz /* Step two: Check that nobody else has taken conflicting locks */ 2456b8cf1913SMax Reitz result = raw_check_lock_bytes(fd, perm, shared, errp); 2457b8cf1913SMax Reitz if (result < 0) { 2458b857431dSFam Zheng error_append_hint(errp, 2459b857431dSFam Zheng "Is another process using the image [%s]?\n", 2460b857431dSFam Zheng file_opts->filename); 24617c20c808SMax Reitz goto out_unlock; 2462b8cf1913SMax Reitz } 2463b8cf1913SMax Reitz 2464b8cf1913SMax Reitz /* Clear the file by truncating it to 0 */ 246593f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); 2466b8cf1913SMax Reitz if (result < 0) { 24677c20c808SMax Reitz goto out_unlock; 2468b8cf1913SMax Reitz } 2469b8cf1913SMax Reitz 2470927f11e1SKevin Wolf if (file_opts->nocow) { 2471c1bb86cdSEric Blake #ifdef __linux__ 2472c1bb86cdSEric Blake /* Set NOCOW flag to solve performance issue on fs like btrfs. 2473c1bb86cdSEric Blake * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value 2474c1bb86cdSEric Blake * will be ignored since any failure of this operation should not 2475c1bb86cdSEric Blake * block the left work. 2476c1bb86cdSEric Blake */ 2477c1bb86cdSEric Blake int attr; 2478c1bb86cdSEric Blake if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { 2479c1bb86cdSEric Blake attr |= FS_NOCOW_FL; 2480c1bb86cdSEric Blake ioctl(fd, FS_IOC_SETFLAGS, &attr); 2481c1bb86cdSEric Blake } 2482c1bb86cdSEric Blake #endif 2483c1bb86cdSEric Blake } 2484ffa244c8SKevin Wolf #ifdef FS_IOC_FSSETXATTR 2485ffa244c8SKevin Wolf /* 2486ffa244c8SKevin Wolf * Try to set the extent size hint. Failure is not fatal, and a warning is 2487ffa244c8SKevin Wolf * only printed if the option was explicitly specified. 2488ffa244c8SKevin Wolf */ 2489ffa244c8SKevin Wolf { 2490ffa244c8SKevin Wolf struct fsxattr attr; 2491ffa244c8SKevin Wolf result = ioctl(fd, FS_IOC_FSGETXATTR, &attr); 2492ffa244c8SKevin Wolf if (result == 0) { 2493ffa244c8SKevin Wolf attr.fsx_xflags |= FS_XFLAG_EXTSIZE; 2494ffa244c8SKevin Wolf attr.fsx_extsize = file_opts->extent_size_hint; 2495ffa244c8SKevin Wolf result = ioctl(fd, FS_IOC_FSSETXATTR, &attr); 2496ffa244c8SKevin Wolf } 2497ffa244c8SKevin Wolf if (result < 0 && file_opts->has_extent_size_hint && 2498ffa244c8SKevin Wolf file_opts->extent_size_hint) 2499ffa244c8SKevin Wolf { 2500ffa244c8SKevin Wolf warn_report("Failed to set extent size hint: %s", 2501ffa244c8SKevin Wolf strerror(errno)); 2502ffa244c8SKevin Wolf } 2503ffa244c8SKevin Wolf } 2504ffa244c8SKevin Wolf #endif 2505c1bb86cdSEric Blake 2506b8cf1913SMax Reitz /* Resize and potentially preallocate the file to the desired 2507b8cf1913SMax Reitz * final size */ 250893f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, file_opts->size, 250993f4e2ffSKevin Wolf file_opts->preallocation, errp); 25109f63b07eSMax Reitz if (result < 0) { 25117c20c808SMax Reitz goto out_unlock; 25127c20c808SMax Reitz } 25137c20c808SMax Reitz 25147c20c808SMax Reitz out_unlock: 25152996ffadSFam Zheng raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); 25167c20c808SMax Reitz if (local_err) { 25177c20c808SMax Reitz /* The above call should not fail, and if it does, that does 25187c20c808SMax Reitz * not mean the whole creation operation has failed. So 25197c20c808SMax Reitz * report it the user for their convenience, but do not report 25207c20c808SMax Reitz * it to the caller. */ 2521db0754dfSFam Zheng warn_report_err(local_err); 25225a1dad9dSNir Soffer } 25235a1dad9dSNir Soffer 25245a1dad9dSNir Soffer out_close: 2525c1bb86cdSEric Blake if (qemu_close(fd) != 0 && result == 0) { 2526c1bb86cdSEric Blake result = -errno; 2527c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not close the new file"); 2528c1bb86cdSEric Blake } 2529c1bb86cdSEric Blake out: 2530c1bb86cdSEric Blake return result; 2531c1bb86cdSEric Blake } 2532c1bb86cdSEric Blake 2533b92902dfSMaxim Levitsky static int coroutine_fn raw_co_create_opts(BlockDriver *drv, 2534b92902dfSMaxim Levitsky const char *filename, 2535b92902dfSMaxim Levitsky QemuOpts *opts, 2536927f11e1SKevin Wolf Error **errp) 2537927f11e1SKevin Wolf { 2538927f11e1SKevin Wolf BlockdevCreateOptions options; 2539927f11e1SKevin Wolf int64_t total_size = 0; 2540ffa244c8SKevin Wolf int64_t extent_size_hint = 0; 2541ffa244c8SKevin Wolf bool has_extent_size_hint = false; 2542927f11e1SKevin Wolf bool nocow = false; 2543927f11e1SKevin Wolf PreallocMode prealloc; 2544927f11e1SKevin Wolf char *buf = NULL; 2545927f11e1SKevin Wolf Error *local_err = NULL; 2546927f11e1SKevin Wolf 2547927f11e1SKevin Wolf /* Skip file: protocol prefix */ 2548927f11e1SKevin Wolf strstart(filename, "file:", &filename); 2549927f11e1SKevin Wolf 2550927f11e1SKevin Wolf /* Read out options */ 2551927f11e1SKevin Wolf total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2552927f11e1SKevin Wolf BDRV_SECTOR_SIZE); 2553ffa244c8SKevin Wolf if (qemu_opt_get(opts, BLOCK_OPT_EXTENT_SIZE_HINT)) { 2554ffa244c8SKevin Wolf has_extent_size_hint = true; 2555ffa244c8SKevin Wolf extent_size_hint = 2556ffa244c8SKevin Wolf qemu_opt_get_size_del(opts, BLOCK_OPT_EXTENT_SIZE_HINT, -1); 2557ffa244c8SKevin Wolf } 2558927f11e1SKevin Wolf nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); 2559927f11e1SKevin Wolf buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2560927f11e1SKevin Wolf prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, 2561927f11e1SKevin Wolf PREALLOC_MODE_OFF, &local_err); 2562927f11e1SKevin Wolf g_free(buf); 2563927f11e1SKevin Wolf if (local_err) { 2564927f11e1SKevin Wolf error_propagate(errp, local_err); 2565927f11e1SKevin Wolf return -EINVAL; 2566927f11e1SKevin Wolf } 2567927f11e1SKevin Wolf 2568927f11e1SKevin Wolf options = (BlockdevCreateOptions) { 2569927f11e1SKevin Wolf .driver = BLOCKDEV_DRIVER_FILE, 2570927f11e1SKevin Wolf .u.file = { 2571927f11e1SKevin Wolf .filename = (char *) filename, 2572927f11e1SKevin Wolf .size = total_size, 2573927f11e1SKevin Wolf .has_preallocation = true, 2574927f11e1SKevin Wolf .preallocation = prealloc, 2575927f11e1SKevin Wolf .has_nocow = true, 2576927f11e1SKevin Wolf .nocow = nocow, 2577ffa244c8SKevin Wolf .has_extent_size_hint = has_extent_size_hint, 2578ffa244c8SKevin Wolf .extent_size_hint = extent_size_hint, 2579927f11e1SKevin Wolf }, 2580927f11e1SKevin Wolf }; 2581927f11e1SKevin Wolf return raw_co_create(&options, errp); 2582927f11e1SKevin Wolf } 2583927f11e1SKevin Wolf 25849bffae14SDaniel Henrique Barboza static int coroutine_fn raw_co_delete_file(BlockDriverState *bs, 25859bffae14SDaniel Henrique Barboza Error **errp) 25869bffae14SDaniel Henrique Barboza { 25879bffae14SDaniel Henrique Barboza struct stat st; 25889bffae14SDaniel Henrique Barboza int ret; 25899bffae14SDaniel Henrique Barboza 25909bffae14SDaniel Henrique Barboza if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) { 25919bffae14SDaniel Henrique Barboza error_setg_errno(errp, ENOENT, "%s is not a regular file", 25929bffae14SDaniel Henrique Barboza bs->filename); 25939bffae14SDaniel Henrique Barboza return -ENOENT; 25949bffae14SDaniel Henrique Barboza } 25959bffae14SDaniel Henrique Barboza 25969bffae14SDaniel Henrique Barboza ret = unlink(bs->filename); 25979bffae14SDaniel Henrique Barboza if (ret < 0) { 25989bffae14SDaniel Henrique Barboza ret = -errno; 25999bffae14SDaniel Henrique Barboza error_setg_errno(errp, -ret, "Error when deleting file %s", 26009bffae14SDaniel Henrique Barboza bs->filename); 26019bffae14SDaniel Henrique Barboza } 26029bffae14SDaniel Henrique Barboza 26039bffae14SDaniel Henrique Barboza return ret; 26049bffae14SDaniel Henrique Barboza } 26059bffae14SDaniel Henrique Barboza 2606c1bb86cdSEric Blake /* 2607c1bb86cdSEric Blake * Find allocation range in @bs around offset @start. 2608c1bb86cdSEric Blake * May change underlying file descriptor's file offset. 2609c1bb86cdSEric Blake * If @start is not in a hole, store @start in @data, and the 2610c1bb86cdSEric Blake * beginning of the next hole in @hole, and return 0. 2611c1bb86cdSEric Blake * If @start is in a non-trailing hole, store @start in @hole and the 2612c1bb86cdSEric Blake * beginning of the next non-hole in @data, and return 0. 2613c1bb86cdSEric Blake * If @start is in a trailing hole or beyond EOF, return -ENXIO. 2614c1bb86cdSEric Blake * If we can't find out, return a negative errno other than -ENXIO. 2615c1bb86cdSEric Blake */ 2616c1bb86cdSEric Blake static int find_allocation(BlockDriverState *bs, off_t start, 2617c1bb86cdSEric Blake off_t *data, off_t *hole) 2618c1bb86cdSEric Blake { 2619c1bb86cdSEric Blake #if defined SEEK_HOLE && defined SEEK_DATA 2620c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2621c1bb86cdSEric Blake off_t offs; 2622c1bb86cdSEric Blake 2623c1bb86cdSEric Blake /* 2624c1bb86cdSEric Blake * SEEK_DATA cases: 2625c1bb86cdSEric Blake * D1. offs == start: start is in data 2626c1bb86cdSEric Blake * D2. offs > start: start is in a hole, next data at offs 2627c1bb86cdSEric Blake * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 2628c1bb86cdSEric Blake * or start is beyond EOF 2629c1bb86cdSEric Blake * If the latter happens, the file has been truncated behind 2630c1bb86cdSEric Blake * our back since we opened it. All bets are off then. 2631c1bb86cdSEric Blake * Treating like a trailing hole is simplest. 2632c1bb86cdSEric Blake * D4. offs < 0, errno != ENXIO: we learned nothing 2633c1bb86cdSEric Blake */ 2634c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_DATA); 2635c1bb86cdSEric Blake if (offs < 0) { 2636c1bb86cdSEric Blake return -errno; /* D3 or D4 */ 2637c1bb86cdSEric Blake } 2638a03083a0SJeff Cody 2639a03083a0SJeff Cody if (offs < start) { 2640a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2641a03083a0SJeff Cody * -EIO in this case, and we'll treat it like D4. */ 2642a03083a0SJeff Cody return -EIO; 2643a03083a0SJeff Cody } 2644c1bb86cdSEric Blake 2645c1bb86cdSEric Blake if (offs > start) { 2646c1bb86cdSEric Blake /* D2: in hole, next data at offs */ 2647c1bb86cdSEric Blake *hole = start; 2648c1bb86cdSEric Blake *data = offs; 2649c1bb86cdSEric Blake return 0; 2650c1bb86cdSEric Blake } 2651c1bb86cdSEric Blake 2652c1bb86cdSEric Blake /* D1: in data, end not yet known */ 2653c1bb86cdSEric Blake 2654c1bb86cdSEric Blake /* 2655c1bb86cdSEric Blake * SEEK_HOLE cases: 2656c1bb86cdSEric Blake * H1. offs == start: start is in a hole 2657c1bb86cdSEric Blake * If this happens here, a hole has been dug behind our back 2658c1bb86cdSEric Blake * since the previous lseek(). 2659c1bb86cdSEric Blake * H2. offs > start: either start is in data, next hole at offs, 2660c1bb86cdSEric Blake * or start is in trailing hole, EOF at offs 2661c1bb86cdSEric Blake * Linux treats trailing holes like any other hole: offs == 2662c1bb86cdSEric Blake * start. Solaris seeks to EOF instead: offs > start (blech). 2663c1bb86cdSEric Blake * If that happens here, a hole has been dug behind our back 2664c1bb86cdSEric Blake * since the previous lseek(). 2665c1bb86cdSEric Blake * H3. offs < 0, errno = ENXIO: start is beyond EOF 2666c1bb86cdSEric Blake * If this happens, the file has been truncated behind our 2667c1bb86cdSEric Blake * back since we opened it. Treat it like a trailing hole. 2668c1bb86cdSEric Blake * H4. offs < 0, errno != ENXIO: we learned nothing 2669c1bb86cdSEric Blake * Pretend we know nothing at all, i.e. "forget" about D1. 2670c1bb86cdSEric Blake */ 2671c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_HOLE); 2672c1bb86cdSEric Blake if (offs < 0) { 2673c1bb86cdSEric Blake return -errno; /* D1 and (H3 or H4) */ 2674c1bb86cdSEric Blake } 2675a03083a0SJeff Cody 2676a03083a0SJeff Cody if (offs < start) { 2677a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2678a03083a0SJeff Cody * -EIO in this case, and we'll treat it like H4. */ 2679a03083a0SJeff Cody return -EIO; 2680a03083a0SJeff Cody } 2681c1bb86cdSEric Blake 2682c1bb86cdSEric Blake if (offs > start) { 2683c1bb86cdSEric Blake /* 2684c1bb86cdSEric Blake * D1 and H2: either in data, next hole at offs, or it was in 2685c1bb86cdSEric Blake * data but is now in a trailing hole. In the latter case, 2686c1bb86cdSEric Blake * all bets are off. Treating it as if it there was data all 2687c1bb86cdSEric Blake * the way to EOF is safe, so simply do that. 2688c1bb86cdSEric Blake */ 2689c1bb86cdSEric Blake *data = start; 2690c1bb86cdSEric Blake *hole = offs; 2691c1bb86cdSEric Blake return 0; 2692c1bb86cdSEric Blake } 2693c1bb86cdSEric Blake 2694c1bb86cdSEric Blake /* D1 and H1 */ 2695c1bb86cdSEric Blake return -EBUSY; 2696c1bb86cdSEric Blake #else 2697c1bb86cdSEric Blake return -ENOTSUP; 2698c1bb86cdSEric Blake #endif 2699c1bb86cdSEric Blake } 2700c1bb86cdSEric Blake 2701c1bb86cdSEric Blake /* 2702a290f085SEric Blake * Returns the allocation status of the specified offset. 2703c1bb86cdSEric Blake * 2704a290f085SEric Blake * The block layer guarantees 'offset' and 'bytes' are within bounds. 2705c1bb86cdSEric Blake * 2706a290f085SEric Blake * 'pnum' is set to the number of bytes (including and immediately following 2707a290f085SEric Blake * the specified offset) that are known to be in the same 2708c1bb86cdSEric Blake * allocated/unallocated state. 2709c1bb86cdSEric Blake * 2710a290f085SEric Blake * 'bytes' is the max value 'pnum' should be set to. 2711c1bb86cdSEric Blake */ 2712a290f085SEric Blake static int coroutine_fn raw_co_block_status(BlockDriverState *bs, 2713a290f085SEric Blake bool want_zero, 2714a290f085SEric Blake int64_t offset, 2715a290f085SEric Blake int64_t bytes, int64_t *pnum, 2716a290f085SEric Blake int64_t *map, 2717c1bb86cdSEric Blake BlockDriverState **file) 2718c1bb86cdSEric Blake { 2719a290f085SEric Blake off_t data = 0, hole = 0; 2720c1bb86cdSEric Blake int ret; 2721c1bb86cdSEric Blake 27229c3db310SMax Reitz assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); 27239c3db310SMax Reitz 2724c1bb86cdSEric Blake ret = fd_open(bs); 2725c1bb86cdSEric Blake if (ret < 0) { 2726c1bb86cdSEric Blake return ret; 2727c1bb86cdSEric Blake } 2728c1bb86cdSEric Blake 2729a290f085SEric Blake if (!want_zero) { 2730a290f085SEric Blake *pnum = bytes; 2731a290f085SEric Blake *map = offset; 2732a290f085SEric Blake *file = bs; 2733a290f085SEric Blake return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 2734c1bb86cdSEric Blake } 2735c1bb86cdSEric Blake 2736a290f085SEric Blake ret = find_allocation(bs, offset, &data, &hole); 2737c1bb86cdSEric Blake if (ret == -ENXIO) { 2738c1bb86cdSEric Blake /* Trailing hole */ 2739a290f085SEric Blake *pnum = bytes; 2740c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2741c1bb86cdSEric Blake } else if (ret < 0) { 2742c1bb86cdSEric Blake /* No info available, so pretend there are no holes */ 2743a290f085SEric Blake *pnum = bytes; 2744c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2745a290f085SEric Blake } else if (data == offset) { 2746a290f085SEric Blake /* On a data extent, compute bytes to the end of the extent, 2747c1bb86cdSEric Blake * possibly including a partial sector at EOF. */ 2748a290f085SEric Blake *pnum = MIN(bytes, hole - offset); 27499c3db310SMax Reitz 27509c3db310SMax Reitz /* 27519c3db310SMax Reitz * We are not allowed to return partial sectors, though, so 27529c3db310SMax Reitz * round up if necessary. 27539c3db310SMax Reitz */ 27549c3db310SMax Reitz if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { 27559c3db310SMax Reitz int64_t file_length = raw_getlength(bs); 27569c3db310SMax Reitz if (file_length > 0) { 27579c3db310SMax Reitz /* Ignore errors, this is just a safeguard */ 27589c3db310SMax Reitz assert(hole == file_length); 27599c3db310SMax Reitz } 27609c3db310SMax Reitz *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); 27619c3db310SMax Reitz } 27629c3db310SMax Reitz 2763c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2764c1bb86cdSEric Blake } else { 2765a290f085SEric Blake /* On a hole, compute bytes to the beginning of the next extent. */ 2766a290f085SEric Blake assert(hole == offset); 2767a290f085SEric Blake *pnum = MIN(bytes, data - offset); 2768c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2769c1bb86cdSEric Blake } 2770a290f085SEric Blake *map = offset; 2771c1bb86cdSEric Blake *file = bs; 2772a290f085SEric Blake return ret | BDRV_BLOCK_OFFSET_VALID; 2773c1bb86cdSEric Blake } 2774c1bb86cdSEric Blake 277531be8a2aSStefan Hajnoczi #if defined(__linux__) 277631be8a2aSStefan Hajnoczi /* Verify that the file is not in the page cache */ 277731be8a2aSStefan Hajnoczi static void check_cache_dropped(BlockDriverState *bs, Error **errp) 277831be8a2aSStefan Hajnoczi { 277931be8a2aSStefan Hajnoczi const size_t window_size = 128 * 1024 * 1024; 278031be8a2aSStefan Hajnoczi BDRVRawState *s = bs->opaque; 278131be8a2aSStefan Hajnoczi void *window = NULL; 278231be8a2aSStefan Hajnoczi size_t length = 0; 278331be8a2aSStefan Hajnoczi unsigned char *vec; 278431be8a2aSStefan Hajnoczi size_t page_size; 278531be8a2aSStefan Hajnoczi off_t offset; 278631be8a2aSStefan Hajnoczi off_t end; 278731be8a2aSStefan Hajnoczi 278831be8a2aSStefan Hajnoczi /* mincore(2) page status information requires 1 byte per page */ 278931be8a2aSStefan Hajnoczi page_size = sysconf(_SC_PAGESIZE); 279031be8a2aSStefan Hajnoczi vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); 279131be8a2aSStefan Hajnoczi 279231be8a2aSStefan Hajnoczi end = raw_getlength(bs); 279331be8a2aSStefan Hajnoczi 279431be8a2aSStefan Hajnoczi for (offset = 0; offset < end; offset += window_size) { 279531be8a2aSStefan Hajnoczi void *new_window; 279631be8a2aSStefan Hajnoczi size_t new_length; 279731be8a2aSStefan Hajnoczi size_t vec_end; 279831be8a2aSStefan Hajnoczi size_t i; 279931be8a2aSStefan Hajnoczi int ret; 280031be8a2aSStefan Hajnoczi 280131be8a2aSStefan Hajnoczi /* Unmap previous window if size has changed */ 280231be8a2aSStefan Hajnoczi new_length = MIN(end - offset, window_size); 280331be8a2aSStefan Hajnoczi if (new_length != length) { 280431be8a2aSStefan Hajnoczi munmap(window, length); 280531be8a2aSStefan Hajnoczi window = NULL; 280631be8a2aSStefan Hajnoczi length = 0; 280731be8a2aSStefan Hajnoczi } 280831be8a2aSStefan Hajnoczi 280931be8a2aSStefan Hajnoczi new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, 281031be8a2aSStefan Hajnoczi s->fd, offset); 281131be8a2aSStefan Hajnoczi if (new_window == MAP_FAILED) { 281231be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mmap failed"); 281331be8a2aSStefan Hajnoczi break; 281431be8a2aSStefan Hajnoczi } 281531be8a2aSStefan Hajnoczi 281631be8a2aSStefan Hajnoczi window = new_window; 281731be8a2aSStefan Hajnoczi length = new_length; 281831be8a2aSStefan Hajnoczi 281931be8a2aSStefan Hajnoczi ret = mincore(window, length, vec); 282031be8a2aSStefan Hajnoczi if (ret < 0) { 282131be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mincore failed"); 282231be8a2aSStefan Hajnoczi break; 282331be8a2aSStefan Hajnoczi } 282431be8a2aSStefan Hajnoczi 282531be8a2aSStefan Hajnoczi vec_end = DIV_ROUND_UP(length, page_size); 282631be8a2aSStefan Hajnoczi for (i = 0; i < vec_end; i++) { 282731be8a2aSStefan Hajnoczi if (vec[i] & 0x1) { 282831be8a2aSStefan Hajnoczi break; 282931be8a2aSStefan Hajnoczi } 283031be8a2aSStefan Hajnoczi } 283177ed971bSMarkus Armbruster if (i < vec_end) { 283277ed971bSMarkus Armbruster error_setg(errp, "page cache still in use!"); 283377ed971bSMarkus Armbruster break; 283477ed971bSMarkus Armbruster } 283531be8a2aSStefan Hajnoczi } 283631be8a2aSStefan Hajnoczi 283731be8a2aSStefan Hajnoczi if (window) { 283831be8a2aSStefan Hajnoczi munmap(window, length); 283931be8a2aSStefan Hajnoczi } 284031be8a2aSStefan Hajnoczi 284131be8a2aSStefan Hajnoczi g_free(vec); 284231be8a2aSStefan Hajnoczi } 284331be8a2aSStefan Hajnoczi #endif /* __linux__ */ 284431be8a2aSStefan Hajnoczi 2845dd577a26SStefan Hajnoczi static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, 2846dd577a26SStefan Hajnoczi Error **errp) 2847dd577a26SStefan Hajnoczi { 2848dd577a26SStefan Hajnoczi BDRVRawState *s = bs->opaque; 2849dd577a26SStefan Hajnoczi int ret; 2850dd577a26SStefan Hajnoczi 2851dd577a26SStefan Hajnoczi ret = fd_open(bs); 2852dd577a26SStefan Hajnoczi if (ret < 0) { 2853dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "The file descriptor is not open"); 2854dd577a26SStefan Hajnoczi return; 2855dd577a26SStefan Hajnoczi } 2856dd577a26SStefan Hajnoczi 2857f357fcd8SStefan Hajnoczi if (!s->drop_cache) { 2858f357fcd8SStefan Hajnoczi return; 2859f357fcd8SStefan Hajnoczi } 2860f357fcd8SStefan Hajnoczi 2861dd577a26SStefan Hajnoczi if (s->open_flags & O_DIRECT) { 2862dd577a26SStefan Hajnoczi return; /* No host kernel page cache */ 2863dd577a26SStefan Hajnoczi } 2864dd577a26SStefan Hajnoczi 2865dd577a26SStefan Hajnoczi #if defined(__linux__) 2866dd577a26SStefan Hajnoczi /* This sets the scene for the next syscall... */ 2867dd577a26SStefan Hajnoczi ret = bdrv_co_flush(bs); 2868dd577a26SStefan Hajnoczi if (ret < 0) { 2869dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "flush failed"); 2870dd577a26SStefan Hajnoczi return; 2871dd577a26SStefan Hajnoczi } 2872dd577a26SStefan Hajnoczi 2873dd577a26SStefan Hajnoczi /* Linux does not invalidate pages that are dirty, locked, or mmapped by a 2874dd577a26SStefan Hajnoczi * process. These limitations are okay because we just fsynced the file, 2875dd577a26SStefan Hajnoczi * we don't use mmap, and the file should not be in use by other processes. 2876dd577a26SStefan Hajnoczi */ 2877dd577a26SStefan Hajnoczi ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); 2878dd577a26SStefan Hajnoczi if (ret != 0) { /* the return value is a positive errno */ 2879dd577a26SStefan Hajnoczi error_setg_errno(errp, ret, "fadvise failed"); 2880dd577a26SStefan Hajnoczi return; 2881dd577a26SStefan Hajnoczi } 288231be8a2aSStefan Hajnoczi 288331be8a2aSStefan Hajnoczi if (s->check_cache_dropped) { 288431be8a2aSStefan Hajnoczi check_cache_dropped(bs, errp); 288531be8a2aSStefan Hajnoczi } 2886dd577a26SStefan Hajnoczi #else /* __linux__ */ 2887dd577a26SStefan Hajnoczi /* Do nothing. Live migration to a remote host with cache.direct=off is 2888dd577a26SStefan Hajnoczi * unsupported on other host operating systems. Cache consistency issues 2889dd577a26SStefan Hajnoczi * may occur but no error is reported here, partly because that's the 2890dd577a26SStefan Hajnoczi * historical behavior and partly because it's hard to differentiate valid 2891dd577a26SStefan Hajnoczi * configurations that should not cause errors. 2892dd577a26SStefan Hajnoczi */ 2893dd577a26SStefan Hajnoczi #endif /* !__linux__ */ 2894dd577a26SStefan Hajnoczi } 2895dd577a26SStefan Hajnoczi 28961c450366SAnton Nefedov static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret) 28971c450366SAnton Nefedov { 28981c450366SAnton Nefedov if (ret) { 28991c450366SAnton Nefedov s->stats.discard_nb_failed++; 29001c450366SAnton Nefedov } else { 29011c450366SAnton Nefedov s->stats.discard_nb_ok++; 29021c450366SAnton Nefedov s->stats.discard_bytes_ok += nbytes; 29031c450366SAnton Nefedov } 29041c450366SAnton Nefedov } 29051c450366SAnton Nefedov 290633d70fb6SKevin Wolf static coroutine_fn int 290746ee0f46SKevin Wolf raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev) 2908c1bb86cdSEric Blake { 2909c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 291046ee0f46SKevin Wolf RawPosixAIOData acb; 29111c450366SAnton Nefedov int ret; 2912c1bb86cdSEric Blake 291346ee0f46SKevin Wolf acb = (RawPosixAIOData) { 291446ee0f46SKevin Wolf .bs = bs, 291546ee0f46SKevin Wolf .aio_fildes = s->fd, 291646ee0f46SKevin Wolf .aio_type = QEMU_AIO_DISCARD, 291746ee0f46SKevin Wolf .aio_offset = offset, 291846ee0f46SKevin Wolf .aio_nbytes = bytes, 291946ee0f46SKevin Wolf }; 292046ee0f46SKevin Wolf 292146ee0f46SKevin Wolf if (blkdev) { 292246ee0f46SKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 292346ee0f46SKevin Wolf } 292446ee0f46SKevin Wolf 29251c450366SAnton Nefedov ret = raw_thread_pool_submit(bs, handle_aiocb_discard, &acb); 29261c450366SAnton Nefedov raw_account_discard(s, bytes, ret); 29271c450366SAnton Nefedov return ret; 292846ee0f46SKevin Wolf } 292946ee0f46SKevin Wolf 293046ee0f46SKevin Wolf static coroutine_fn int 293146ee0f46SKevin Wolf raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 293246ee0f46SKevin Wolf { 293346ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, false); 2934c1bb86cdSEric Blake } 2935c1bb86cdSEric Blake 29367154d8aeSKevin Wolf static int coroutine_fn 29377154d8aeSKevin Wolf raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, 29387154d8aeSKevin Wolf BdrvRequestFlags flags, bool blkdev) 29397154d8aeSKevin Wolf { 29407154d8aeSKevin Wolf BDRVRawState *s = bs->opaque; 29417154d8aeSKevin Wolf RawPosixAIOData acb; 29427154d8aeSKevin Wolf ThreadPoolFunc *handler; 29437154d8aeSKevin Wolf 2944292d06b9SMax Reitz #ifdef CONFIG_FALLOCATE 2945292d06b9SMax Reitz if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { 2946292d06b9SMax Reitz BdrvTrackedRequest *req; 2947292d06b9SMax Reitz 2948292d06b9SMax Reitz /* 2949292d06b9SMax Reitz * This is a workaround for a bug in the Linux XFS driver, 2950292d06b9SMax Reitz * where writes submitted through the AIO interface will be 2951292d06b9SMax Reitz * discarded if they happen beyond a concurrently running 2952292d06b9SMax Reitz * fallocate() that increases the file length (i.e., both the 2953292d06b9SMax Reitz * write and the fallocate() happen beyond the EOF). 2954292d06b9SMax Reitz * 2955292d06b9SMax Reitz * To work around it, we extend the tracked request for this 2956292d06b9SMax Reitz * zero write until INT64_MAX (effectively infinity), and mark 2957292d06b9SMax Reitz * it as serializing. 2958292d06b9SMax Reitz * 2959292d06b9SMax Reitz * We have to enable this workaround for all filesystems and 2960292d06b9SMax Reitz * AIO modes (not just XFS with aio=native), because for 2961292d06b9SMax Reitz * remote filesystems we do not know the host configuration. 2962292d06b9SMax Reitz */ 2963292d06b9SMax Reitz 2964292d06b9SMax Reitz req = bdrv_co_get_self_request(bs); 2965292d06b9SMax Reitz assert(req); 2966292d06b9SMax Reitz assert(req->type == BDRV_TRACKED_WRITE); 2967292d06b9SMax Reitz assert(req->offset <= offset); 2968292d06b9SMax Reitz assert(req->offset + req->bytes >= offset + bytes); 2969292d06b9SMax Reitz 29708b117001SVladimir Sementsov-Ogievskiy req->bytes = BDRV_MAX_LENGTH - req->offset; 29718b117001SVladimir Sementsov-Ogievskiy 29728b117001SVladimir Sementsov-Ogievskiy assert(bdrv_check_request(req->offset, req->bytes) == 0); 2973292d06b9SMax Reitz 29748ac5aab2SVladimir Sementsov-Ogievskiy bdrv_make_request_serialising(req, bs->bl.request_alignment); 2975292d06b9SMax Reitz } 2976292d06b9SMax Reitz #endif 2977292d06b9SMax Reitz 29787154d8aeSKevin Wolf acb = (RawPosixAIOData) { 29797154d8aeSKevin Wolf .bs = bs, 29807154d8aeSKevin Wolf .aio_fildes = s->fd, 29817154d8aeSKevin Wolf .aio_type = QEMU_AIO_WRITE_ZEROES, 29827154d8aeSKevin Wolf .aio_offset = offset, 29837154d8aeSKevin Wolf .aio_nbytes = bytes, 29847154d8aeSKevin Wolf }; 29857154d8aeSKevin Wolf 29867154d8aeSKevin Wolf if (blkdev) { 29877154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 29887154d8aeSKevin Wolf } 2989738301e1SKevin Wolf if (flags & BDRV_REQ_NO_FALLBACK) { 2990738301e1SKevin Wolf acb.aio_type |= QEMU_AIO_NO_FALLBACK; 2991738301e1SKevin Wolf } 29927154d8aeSKevin Wolf 29937154d8aeSKevin Wolf if (flags & BDRV_REQ_MAY_UNMAP) { 29947154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_DISCARD; 29957154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes_unmap; 29967154d8aeSKevin Wolf } else { 29977154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes; 29987154d8aeSKevin Wolf } 29997154d8aeSKevin Wolf 30007154d8aeSKevin Wolf return raw_thread_pool_submit(bs, handler, &acb); 30017154d8aeSKevin Wolf } 30027154d8aeSKevin Wolf 3003c1bb86cdSEric Blake static int coroutine_fn raw_co_pwrite_zeroes( 3004c1bb86cdSEric Blake BlockDriverState *bs, int64_t offset, 3005f5a5ca79SManos Pitsidianakis int bytes, BdrvRequestFlags flags) 3006c1bb86cdSEric Blake { 30077154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); 3008c1bb86cdSEric Blake } 3009c1bb86cdSEric Blake 3010c1bb86cdSEric Blake static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 3011c1bb86cdSEric Blake { 3012c1bb86cdSEric Blake return 0; 3013c1bb86cdSEric Blake } 3014c1bb86cdSEric Blake 3015d9245599SAnton Nefedov static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs) 3016d9245599SAnton Nefedov { 3017d9245599SAnton Nefedov BDRVRawState *s = bs->opaque; 3018d9245599SAnton Nefedov return (BlockStatsSpecificFile) { 3019d9245599SAnton Nefedov .discard_nb_ok = s->stats.discard_nb_ok, 3020d9245599SAnton Nefedov .discard_nb_failed = s->stats.discard_nb_failed, 3021d9245599SAnton Nefedov .discard_bytes_ok = s->stats.discard_bytes_ok, 3022d9245599SAnton Nefedov }; 3023d9245599SAnton Nefedov } 3024d9245599SAnton Nefedov 3025d9245599SAnton Nefedov static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs) 3026d9245599SAnton Nefedov { 3027d9245599SAnton Nefedov BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 3028d9245599SAnton Nefedov 3029d9245599SAnton Nefedov stats->driver = BLOCKDEV_DRIVER_FILE; 3030d9245599SAnton Nefedov stats->u.file = get_blockstats_specific_file(bs); 3031d9245599SAnton Nefedov 3032d9245599SAnton Nefedov return stats; 3033d9245599SAnton Nefedov } 3034d9245599SAnton Nefedov 3035d9245599SAnton Nefedov static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs) 3036d9245599SAnton Nefedov { 3037d9245599SAnton Nefedov BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 3038d9245599SAnton Nefedov 3039d9245599SAnton Nefedov stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE; 3040d9245599SAnton Nefedov stats->u.host_device = get_blockstats_specific_file(bs); 3041d9245599SAnton Nefedov 3042d9245599SAnton Nefedov return stats; 3043d9245599SAnton Nefedov } 3044d9245599SAnton Nefedov 3045c1bb86cdSEric Blake static QemuOptsList raw_create_opts = { 3046c1bb86cdSEric Blake .name = "raw-create-opts", 3047c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), 3048c1bb86cdSEric Blake .desc = { 3049c1bb86cdSEric Blake { 3050c1bb86cdSEric Blake .name = BLOCK_OPT_SIZE, 3051c1bb86cdSEric Blake .type = QEMU_OPT_SIZE, 3052c1bb86cdSEric Blake .help = "Virtual disk size" 3053c1bb86cdSEric Blake }, 3054c1bb86cdSEric Blake { 3055c1bb86cdSEric Blake .name = BLOCK_OPT_NOCOW, 3056c1bb86cdSEric Blake .type = QEMU_OPT_BOOL, 3057c1bb86cdSEric Blake .help = "Turn off copy-on-write (valid only on btrfs)" 3058c1bb86cdSEric Blake }, 3059c1bb86cdSEric Blake { 3060c1bb86cdSEric Blake .name = BLOCK_OPT_PREALLOC, 3061c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 3062abea0053SStefano Garzarella .help = "Preallocation mode (allowed values: off" 3063abea0053SStefano Garzarella #ifdef CONFIG_POSIX_FALLOCATE 3064abea0053SStefano Garzarella ", falloc" 3065abea0053SStefano Garzarella #endif 3066abea0053SStefano Garzarella ", full)" 3067c1bb86cdSEric Blake }, 3068ffa244c8SKevin Wolf { 3069ffa244c8SKevin Wolf .name = BLOCK_OPT_EXTENT_SIZE_HINT, 3070ffa244c8SKevin Wolf .type = QEMU_OPT_SIZE, 3071ffa244c8SKevin Wolf .help = "Extent size hint for the image file, 0 to disable" 3072ffa244c8SKevin Wolf }, 3073c1bb86cdSEric Blake { /* end of list */ } 3074c1bb86cdSEric Blake } 3075c1bb86cdSEric Blake }; 3076c1bb86cdSEric Blake 3077244a5668SFam Zheng static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, 3078244a5668SFam Zheng Error **errp) 3079244a5668SFam Zheng { 30806ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 30816ceabe6fSKevin Wolf BDRVRawReopenState *rs = NULL; 30826ceabe6fSKevin Wolf int open_flags; 30836ceabe6fSKevin Wolf int ret; 30846ceabe6fSKevin Wolf 30856ceabe6fSKevin Wolf if (s->perm_change_fd) { 30866ceabe6fSKevin Wolf /* 30876ceabe6fSKevin Wolf * In the context of reopen, this function may be called several times 30886ceabe6fSKevin Wolf * (directly and recursively while change permissions of the parent). 30896ceabe6fSKevin Wolf * This is even true for children that don't inherit from the original 30906ceabe6fSKevin Wolf * reopen node, so s->reopen_state is not set. 30916ceabe6fSKevin Wolf * 30926ceabe6fSKevin Wolf * Ignore all but the first call. 30936ceabe6fSKevin Wolf */ 30946ceabe6fSKevin Wolf return 0; 30956ceabe6fSKevin Wolf } 30966ceabe6fSKevin Wolf 30976ceabe6fSKevin Wolf if (s->reopen_state) { 30986ceabe6fSKevin Wolf /* We already have a new file descriptor to set permissions for */ 30996ceabe6fSKevin Wolf assert(s->reopen_state->perm == perm); 31006ceabe6fSKevin Wolf assert(s->reopen_state->shared_perm == shared); 31016ceabe6fSKevin Wolf rs = s->reopen_state->opaque; 31026ceabe6fSKevin Wolf s->perm_change_fd = rs->fd; 3103094e3639SMax Reitz s->perm_change_flags = rs->open_flags; 31046ceabe6fSKevin Wolf } else { 31056ceabe6fSKevin Wolf /* We may need a new fd if auto-read-only switches the mode */ 310623dece19SKevin Wolf ret = raw_reconfigure_getfd(bs, bs->open_flags, &open_flags, perm, 31076ceabe6fSKevin Wolf false, errp); 31086ceabe6fSKevin Wolf if (ret < 0) { 31096ceabe6fSKevin Wolf return ret; 31106ceabe6fSKevin Wolf } else if (ret != s->fd) { 31116ceabe6fSKevin Wolf s->perm_change_fd = ret; 3112094e3639SMax Reitz s->perm_change_flags = open_flags; 31136ceabe6fSKevin Wolf } 31146ceabe6fSKevin Wolf } 31156ceabe6fSKevin Wolf 31166ceabe6fSKevin Wolf /* Prepare permissions on old fd to avoid conflicts between old and new, 31176ceabe6fSKevin Wolf * but keep everything locked that new will need. */ 31186ceabe6fSKevin Wolf ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); 31196ceabe6fSKevin Wolf if (ret < 0) { 31206ceabe6fSKevin Wolf goto fail; 31216ceabe6fSKevin Wolf } 31226ceabe6fSKevin Wolf 31236ceabe6fSKevin Wolf /* Copy locks to the new fd */ 3124eb43ea16SLi Feng if (s->perm_change_fd && s->use_lock) { 31256ceabe6fSKevin Wolf ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, 31266ceabe6fSKevin Wolf false, errp); 31276ceabe6fSKevin Wolf if (ret < 0) { 31286ceabe6fSKevin Wolf raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 31296ceabe6fSKevin Wolf goto fail; 31306ceabe6fSKevin Wolf } 31316ceabe6fSKevin Wolf } 31326ceabe6fSKevin Wolf return 0; 31336ceabe6fSKevin Wolf 31346ceabe6fSKevin Wolf fail: 31356ceabe6fSKevin Wolf if (s->perm_change_fd && !s->reopen_state) { 31366ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 31376ceabe6fSKevin Wolf } 31386ceabe6fSKevin Wolf s->perm_change_fd = 0; 31396ceabe6fSKevin Wolf return ret; 3140244a5668SFam Zheng } 3141244a5668SFam Zheng 3142244a5668SFam Zheng static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) 3143244a5668SFam Zheng { 3144244a5668SFam Zheng BDRVRawState *s = bs->opaque; 31456ceabe6fSKevin Wolf 31466ceabe6fSKevin Wolf /* For reopen, we have already switched to the new fd (.bdrv_set_perm is 31476ceabe6fSKevin Wolf * called after .bdrv_reopen_commit) */ 31486ceabe6fSKevin Wolf if (s->perm_change_fd && s->fd != s->perm_change_fd) { 31496ceabe6fSKevin Wolf qemu_close(s->fd); 31506ceabe6fSKevin Wolf s->fd = s->perm_change_fd; 3151094e3639SMax Reitz s->open_flags = s->perm_change_flags; 31526ceabe6fSKevin Wolf } 31536ceabe6fSKevin Wolf s->perm_change_fd = 0; 31546ceabe6fSKevin Wolf 3155244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); 3156244a5668SFam Zheng s->perm = perm; 3157244a5668SFam Zheng s->shared_perm = shared; 3158244a5668SFam Zheng } 3159244a5668SFam Zheng 3160244a5668SFam Zheng static void raw_abort_perm_update(BlockDriverState *bs) 3161244a5668SFam Zheng { 31626ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 31636ceabe6fSKevin Wolf 31646ceabe6fSKevin Wolf /* For reopen, .bdrv_reopen_abort is called afterwards and will close 31656ceabe6fSKevin Wolf * the file descriptor. */ 31666ceabe6fSKevin Wolf if (s->perm_change_fd && !s->reopen_state) { 31676ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 31686ceabe6fSKevin Wolf } 31696ceabe6fSKevin Wolf s->perm_change_fd = 0; 31706ceabe6fSKevin Wolf 3171244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 3172244a5668SFam Zheng } 3173244a5668SFam Zheng 317467b51fb9SVladimir Sementsov-Ogievskiy static int coroutine_fn raw_co_copy_range_from( 317567b51fb9SVladimir Sementsov-Ogievskiy BlockDriverState *bs, BdrvChild *src, uint64_t src_offset, 317667b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, uint64_t dst_offset, uint64_t bytes, 317767b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) 31781efad060SFam Zheng { 317967b51fb9SVladimir Sementsov-Ogievskiy return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, 318067b51fb9SVladimir Sementsov-Ogievskiy read_flags, write_flags); 31811efad060SFam Zheng } 31821efad060SFam Zheng 31831efad060SFam Zheng static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, 318467b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *src, 318567b51fb9SVladimir Sementsov-Ogievskiy uint64_t src_offset, 318667b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, 318767b51fb9SVladimir Sementsov-Ogievskiy uint64_t dst_offset, 318867b51fb9SVladimir Sementsov-Ogievskiy uint64_t bytes, 318967b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, 319067b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags write_flags) 31911efad060SFam Zheng { 319258a209c4SKevin Wolf RawPosixAIOData acb; 31931efad060SFam Zheng BDRVRawState *s = bs->opaque; 31941efad060SFam Zheng BDRVRawState *src_s; 31951efad060SFam Zheng 31961efad060SFam Zheng assert(dst->bs == bs); 31971efad060SFam Zheng if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { 31981efad060SFam Zheng return -ENOTSUP; 31991efad060SFam Zheng } 32001efad060SFam Zheng 32011efad060SFam Zheng src_s = src->bs->opaque; 32029f850f67SFam Zheng if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { 32031efad060SFam Zheng return -EIO; 32041efad060SFam Zheng } 320558a209c4SKevin Wolf 320658a209c4SKevin Wolf acb = (RawPosixAIOData) { 320758a209c4SKevin Wolf .bs = bs, 320858a209c4SKevin Wolf .aio_type = QEMU_AIO_COPY_RANGE, 320958a209c4SKevin Wolf .aio_fildes = src_s->fd, 321058a209c4SKevin Wolf .aio_offset = src_offset, 321158a209c4SKevin Wolf .aio_nbytes = bytes, 321258a209c4SKevin Wolf .copy_range = { 321358a209c4SKevin Wolf .aio_fd2 = s->fd, 321458a209c4SKevin Wolf .aio_offset2 = dst_offset, 321558a209c4SKevin Wolf }, 321658a209c4SKevin Wolf }; 321758a209c4SKevin Wolf 321858a209c4SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); 32191efad060SFam Zheng } 32201efad060SFam Zheng 3221c1bb86cdSEric Blake BlockDriver bdrv_file = { 3222c1bb86cdSEric Blake .format_name = "file", 3223c1bb86cdSEric Blake .protocol_name = "file", 3224c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3225c1bb86cdSEric Blake .bdrv_needs_filename = true, 3226c1bb86cdSEric Blake .bdrv_probe = NULL, /* no probe for protocols */ 3227c1bb86cdSEric Blake .bdrv_parse_filename = raw_parse_filename, 3228c1bb86cdSEric Blake .bdrv_file_open = raw_open, 3229c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3230c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3231c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3232c1bb86cdSEric Blake .bdrv_close = raw_close, 3233927f11e1SKevin Wolf .bdrv_co_create = raw_co_create, 3234efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = raw_co_create_opts, 3235c1bb86cdSEric Blake .bdrv_has_zero_init = bdrv_has_zero_init_1, 3236a290f085SEric Blake .bdrv_co_block_status = raw_co_block_status, 3237dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3238c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, 32399bffae14SDaniel Henrique Barboza .bdrv_co_delete_file = raw_co_delete_file, 3240c1bb86cdSEric Blake 3241c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3242c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 324333d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 324433d70fb6SKevin Wolf .bdrv_co_pdiscard = raw_co_pdiscard, 32451efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 32461efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 3247c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3248c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3249c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3250ed6e2161SNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3251c1bb86cdSEric Blake 3252061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3253c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3254c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 3255c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3256c1bb86cdSEric Blake = raw_get_allocated_file_size, 3257d9245599SAnton Nefedov .bdrv_get_specific_stats = raw_get_specific_stats, 3258244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 3259244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 3260244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 3261c1bb86cdSEric Blake .create_opts = &raw_create_opts, 32628a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3263c1bb86cdSEric Blake }; 3264c1bb86cdSEric Blake 3265c1bb86cdSEric Blake /***********************************************/ 3266c1bb86cdSEric Blake /* host device */ 3267c1bb86cdSEric Blake 3268c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3269c1bb86cdSEric Blake static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3270c1bb86cdSEric Blake CFIndex maxPathSize, int flags); 3271c1bb86cdSEric Blake static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) 3272c1bb86cdSEric Blake { 3273c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 3274c1bb86cdSEric Blake mach_port_t masterPort; 3275c1bb86cdSEric Blake CFMutableDictionaryRef classesToMatch; 3276c1bb86cdSEric Blake const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; 3277c1bb86cdSEric Blake char *mediaType = NULL; 3278c1bb86cdSEric Blake 3279c1bb86cdSEric Blake kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); 3280c1bb86cdSEric Blake if ( KERN_SUCCESS != kernResult ) { 3281c1bb86cdSEric Blake printf( "IOMasterPort returned %d\n", kernResult ); 3282c1bb86cdSEric Blake } 3283c1bb86cdSEric Blake 3284c1bb86cdSEric Blake int index; 3285c1bb86cdSEric Blake for (index = 0; index < ARRAY_SIZE(matching_array); index++) { 3286c1bb86cdSEric Blake classesToMatch = IOServiceMatching(matching_array[index]); 3287c1bb86cdSEric Blake if (classesToMatch == NULL) { 3288c1bb86cdSEric Blake error_report("IOServiceMatching returned NULL for %s", 3289c1bb86cdSEric Blake matching_array[index]); 3290c1bb86cdSEric Blake continue; 3291c1bb86cdSEric Blake } 3292c1bb86cdSEric Blake CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), 3293c1bb86cdSEric Blake kCFBooleanTrue); 3294c1bb86cdSEric Blake kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, 3295c1bb86cdSEric Blake mediaIterator); 3296c1bb86cdSEric Blake if (kernResult != KERN_SUCCESS) { 3297c1bb86cdSEric Blake error_report("Note: IOServiceGetMatchingServices returned %d", 3298c1bb86cdSEric Blake kernResult); 3299c1bb86cdSEric Blake continue; 3300c1bb86cdSEric Blake } 3301c1bb86cdSEric Blake 3302c1bb86cdSEric Blake /* If a match was found, leave the loop */ 3303c1bb86cdSEric Blake if (*mediaIterator != 0) { 33044f7d28d7SLaurent Vivier trace_file_FindEjectableOpticalMedia(matching_array[index]); 3305c1bb86cdSEric Blake mediaType = g_strdup(matching_array[index]); 3306c1bb86cdSEric Blake break; 3307c1bb86cdSEric Blake } 3308c1bb86cdSEric Blake } 3309c1bb86cdSEric Blake return mediaType; 3310c1bb86cdSEric Blake } 3311c1bb86cdSEric Blake 3312c1bb86cdSEric Blake kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3313c1bb86cdSEric Blake CFIndex maxPathSize, int flags) 3314c1bb86cdSEric Blake { 3315c1bb86cdSEric Blake io_object_t nextMedia; 3316c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 3317c1bb86cdSEric Blake *bsdPath = '\0'; 3318c1bb86cdSEric Blake nextMedia = IOIteratorNext( mediaIterator ); 3319c1bb86cdSEric Blake if ( nextMedia ) 3320c1bb86cdSEric Blake { 3321c1bb86cdSEric Blake CFTypeRef bsdPathAsCFString; 3322c1bb86cdSEric Blake bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); 3323c1bb86cdSEric Blake if ( bsdPathAsCFString ) { 3324c1bb86cdSEric Blake size_t devPathLength; 3325c1bb86cdSEric Blake strcpy( bsdPath, _PATH_DEV ); 3326c1bb86cdSEric Blake if (flags & BDRV_O_NOCACHE) { 3327c1bb86cdSEric Blake strcat(bsdPath, "r"); 3328c1bb86cdSEric Blake } 3329c1bb86cdSEric Blake devPathLength = strlen( bsdPath ); 3330c1bb86cdSEric Blake if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { 3331c1bb86cdSEric Blake kernResult = KERN_SUCCESS; 3332c1bb86cdSEric Blake } 3333c1bb86cdSEric Blake CFRelease( bsdPathAsCFString ); 3334c1bb86cdSEric Blake } 3335c1bb86cdSEric Blake IOObjectRelease( nextMedia ); 3336c1bb86cdSEric Blake } 3337c1bb86cdSEric Blake 3338c1bb86cdSEric Blake return kernResult; 3339c1bb86cdSEric Blake } 3340c1bb86cdSEric Blake 3341c1bb86cdSEric Blake /* Sets up a real cdrom for use in QEMU */ 3342c1bb86cdSEric Blake static bool setup_cdrom(char *bsd_path, Error **errp) 3343c1bb86cdSEric Blake { 3344c1bb86cdSEric Blake int index, num_of_test_partitions = 2, fd; 3345c1bb86cdSEric Blake char test_partition[MAXPATHLEN]; 3346c1bb86cdSEric Blake bool partition_found = false; 3347c1bb86cdSEric Blake 3348c1bb86cdSEric Blake /* look for a working partition */ 3349c1bb86cdSEric Blake for (index = 0; index < num_of_test_partitions; index++) { 3350c1bb86cdSEric Blake snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, 3351c1bb86cdSEric Blake index); 3352b18a24a9SDaniel P. Berrangé fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE, NULL); 3353c1bb86cdSEric Blake if (fd >= 0) { 3354c1bb86cdSEric Blake partition_found = true; 3355c1bb86cdSEric Blake qemu_close(fd); 3356c1bb86cdSEric Blake break; 3357c1bb86cdSEric Blake } 3358c1bb86cdSEric Blake } 3359c1bb86cdSEric Blake 3360c1bb86cdSEric Blake /* if a working partition on the device was not found */ 3361c1bb86cdSEric Blake if (partition_found == false) { 3362c1bb86cdSEric Blake error_setg(errp, "Failed to find a working partition on disc"); 3363c1bb86cdSEric Blake } else { 33644f7d28d7SLaurent Vivier trace_file_setup_cdrom(test_partition); 3365c1bb86cdSEric Blake pstrcpy(bsd_path, MAXPATHLEN, test_partition); 3366c1bb86cdSEric Blake } 3367c1bb86cdSEric Blake return partition_found; 3368c1bb86cdSEric Blake } 3369c1bb86cdSEric Blake 3370c1bb86cdSEric Blake /* Prints directions on mounting and unmounting a device */ 3371c1bb86cdSEric Blake static void print_unmounting_directions(const char *file_name) 3372c1bb86cdSEric Blake { 3373c1bb86cdSEric Blake error_report("If device %s is mounted on the desktop, unmount" 3374c1bb86cdSEric Blake " it first before using it in QEMU", file_name); 3375c1bb86cdSEric Blake error_report("Command to unmount device: diskutil unmountDisk %s", 3376c1bb86cdSEric Blake file_name); 3377c1bb86cdSEric Blake error_report("Command to mount device: diskutil mountDisk %s", file_name); 3378c1bb86cdSEric Blake } 3379c1bb86cdSEric Blake 3380c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3381c1bb86cdSEric Blake 3382c1bb86cdSEric Blake static int hdev_probe_device(const char *filename) 3383c1bb86cdSEric Blake { 3384c1bb86cdSEric Blake struct stat st; 3385c1bb86cdSEric Blake 3386c1bb86cdSEric Blake /* allow a dedicated CD-ROM driver to match with a higher priority */ 3387c1bb86cdSEric Blake if (strstart(filename, "/dev/cdrom", NULL)) 3388c1bb86cdSEric Blake return 50; 3389c1bb86cdSEric Blake 3390c1bb86cdSEric Blake if (stat(filename, &st) >= 0 && 3391c1bb86cdSEric Blake (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 3392c1bb86cdSEric Blake return 100; 3393c1bb86cdSEric Blake } 3394c1bb86cdSEric Blake 3395c1bb86cdSEric Blake return 0; 3396c1bb86cdSEric Blake } 3397c1bb86cdSEric Blake 3398c1bb86cdSEric Blake static void hdev_parse_filename(const char *filename, QDict *options, 3399c1bb86cdSEric Blake Error **errp) 3400c1bb86cdSEric Blake { 340103c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_device:", options); 3402c1bb86cdSEric Blake } 3403c1bb86cdSEric Blake 3404c1bb86cdSEric Blake static bool hdev_is_sg(BlockDriverState *bs) 3405c1bb86cdSEric Blake { 3406c1bb86cdSEric Blake 3407c1bb86cdSEric Blake #if defined(__linux__) 3408c1bb86cdSEric Blake 3409c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3410c1bb86cdSEric Blake struct stat st; 3411c1bb86cdSEric Blake struct sg_scsi_id scsiid; 3412c1bb86cdSEric Blake int sg_version; 3413c1bb86cdSEric Blake int ret; 3414c1bb86cdSEric Blake 3415c1bb86cdSEric Blake if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { 3416c1bb86cdSEric Blake return false; 3417c1bb86cdSEric Blake } 3418c1bb86cdSEric Blake 3419c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); 3420c1bb86cdSEric Blake if (ret < 0) { 3421c1bb86cdSEric Blake return false; 3422c1bb86cdSEric Blake } 3423c1bb86cdSEric Blake 3424c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); 3425c1bb86cdSEric Blake if (ret >= 0) { 34264f7d28d7SLaurent Vivier trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); 3427c1bb86cdSEric Blake return true; 3428c1bb86cdSEric Blake } 3429c1bb86cdSEric Blake 3430c1bb86cdSEric Blake #endif 3431c1bb86cdSEric Blake 3432c1bb86cdSEric Blake return false; 3433c1bb86cdSEric Blake } 3434c1bb86cdSEric Blake 3435c1bb86cdSEric Blake static int hdev_open(BlockDriverState *bs, QDict *options, int flags, 3436c1bb86cdSEric Blake Error **errp) 3437c1bb86cdSEric Blake { 3438c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3439c1bb86cdSEric Blake int ret; 3440c1bb86cdSEric Blake 3441c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3442129c7d1cSMarkus Armbruster /* 3443129c7d1cSMarkus Armbruster * Caution: while qdict_get_str() is fine, getting non-string types 3444129c7d1cSMarkus Armbruster * would require more care. When @options come from -blockdev or 3445129c7d1cSMarkus Armbruster * blockdev_add, its members are typed according to the QAPI 3446129c7d1cSMarkus Armbruster * schema, but when they come from -drive, they're all QString. 3447129c7d1cSMarkus Armbruster */ 3448c1bb86cdSEric Blake const char *filename = qdict_get_str(options, "filename"); 3449c1bb86cdSEric Blake char bsd_path[MAXPATHLEN] = ""; 3450c1bb86cdSEric Blake bool error_occurred = false; 3451c1bb86cdSEric Blake 3452c1bb86cdSEric Blake /* If using a real cdrom */ 3453c1bb86cdSEric Blake if (strcmp(filename, "/dev/cdrom") == 0) { 3454c1bb86cdSEric Blake char *mediaType = NULL; 3455c1bb86cdSEric Blake kern_return_t ret_val; 3456c1bb86cdSEric Blake io_iterator_t mediaIterator = 0; 3457c1bb86cdSEric Blake 3458c1bb86cdSEric Blake mediaType = FindEjectableOpticalMedia(&mediaIterator); 3459c1bb86cdSEric Blake if (mediaType == NULL) { 3460c1bb86cdSEric Blake error_setg(errp, "Please make sure your CD/DVD is in the optical" 3461c1bb86cdSEric Blake " drive"); 3462c1bb86cdSEric Blake error_occurred = true; 3463c1bb86cdSEric Blake goto hdev_open_Mac_error; 3464c1bb86cdSEric Blake } 3465c1bb86cdSEric Blake 3466c1bb86cdSEric Blake ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); 3467c1bb86cdSEric Blake if (ret_val != KERN_SUCCESS) { 3468c1bb86cdSEric Blake error_setg(errp, "Could not get BSD path for optical drive"); 3469c1bb86cdSEric Blake error_occurred = true; 3470c1bb86cdSEric Blake goto hdev_open_Mac_error; 3471c1bb86cdSEric Blake } 3472c1bb86cdSEric Blake 3473c1bb86cdSEric Blake /* If a real optical drive was not found */ 3474c1bb86cdSEric Blake if (bsd_path[0] == '\0') { 3475c1bb86cdSEric Blake error_setg(errp, "Failed to obtain bsd path for optical drive"); 3476c1bb86cdSEric Blake error_occurred = true; 3477c1bb86cdSEric Blake goto hdev_open_Mac_error; 3478c1bb86cdSEric Blake } 3479c1bb86cdSEric Blake 3480c1bb86cdSEric Blake /* If using a cdrom disc and finding a partition on the disc failed */ 3481c1bb86cdSEric Blake if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && 3482c1bb86cdSEric Blake setup_cdrom(bsd_path, errp) == false) { 3483c1bb86cdSEric Blake print_unmounting_directions(bsd_path); 3484c1bb86cdSEric Blake error_occurred = true; 3485c1bb86cdSEric Blake goto hdev_open_Mac_error; 3486c1bb86cdSEric Blake } 3487c1bb86cdSEric Blake 348846f5ac20SEric Blake qdict_put_str(options, "filename", bsd_path); 3489c1bb86cdSEric Blake 3490c1bb86cdSEric Blake hdev_open_Mac_error: 3491c1bb86cdSEric Blake g_free(mediaType); 3492c1bb86cdSEric Blake if (mediaIterator) { 3493c1bb86cdSEric Blake IOObjectRelease(mediaIterator); 3494c1bb86cdSEric Blake } 3495c1bb86cdSEric Blake if (error_occurred) { 3496c1bb86cdSEric Blake return -ENOENT; 3497c1bb86cdSEric Blake } 3498c1bb86cdSEric Blake } 3499c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3500c1bb86cdSEric Blake 3501c1bb86cdSEric Blake s->type = FTYPE_FILE; 3502c1bb86cdSEric Blake 3503668f62ecSMarkus Armbruster ret = raw_open_common(bs, options, flags, 0, true, errp); 3504c1bb86cdSEric Blake if (ret < 0) { 3505c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3506c1bb86cdSEric Blake if (*bsd_path) { 3507c1bb86cdSEric Blake filename = bsd_path; 3508c1bb86cdSEric Blake } 3509c1bb86cdSEric Blake /* if a physical device experienced an error while being opened */ 3510c1bb86cdSEric Blake if (strncmp(filename, "/dev/", 5) == 0) { 3511c1bb86cdSEric Blake print_unmounting_directions(filename); 3512c1bb86cdSEric Blake } 3513c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3514c1bb86cdSEric Blake return ret; 3515c1bb86cdSEric Blake } 3516c1bb86cdSEric Blake 3517c1bb86cdSEric Blake /* Since this does ioctl the device must be already opened */ 3518c1bb86cdSEric Blake bs->sg = hdev_is_sg(bs); 3519c1bb86cdSEric Blake 3520c1bb86cdSEric Blake return ret; 3521c1bb86cdSEric Blake } 3522c1bb86cdSEric Blake 3523c1bb86cdSEric Blake #if defined(__linux__) 35242f3a7ab3SKevin Wolf static int coroutine_fn 35252f3a7ab3SKevin Wolf hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) 3526c1bb86cdSEric Blake { 3527c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 352803425671SKevin Wolf RawPosixAIOData acb; 35292f3a7ab3SKevin Wolf int ret; 3530c1bb86cdSEric Blake 35312f3a7ab3SKevin Wolf ret = fd_open(bs); 35322f3a7ab3SKevin Wolf if (ret < 0) { 35332f3a7ab3SKevin Wolf return ret; 35342f3a7ab3SKevin Wolf } 3535c1bb86cdSEric Blake 35367c9e5276SPaolo Bonzini if (req == SG_IO && s->pr_mgr) { 35377c9e5276SPaolo Bonzini struct sg_io_hdr *io_hdr = buf; 35387c9e5276SPaolo Bonzini if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || 35397c9e5276SPaolo Bonzini io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { 35407c9e5276SPaolo Bonzini return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), 35412f3a7ab3SKevin Wolf s->fd, io_hdr); 35427c9e5276SPaolo Bonzini } 35437c9e5276SPaolo Bonzini } 35447c9e5276SPaolo Bonzini 354503425671SKevin Wolf acb = (RawPosixAIOData) { 354603425671SKevin Wolf .bs = bs, 354703425671SKevin Wolf .aio_type = QEMU_AIO_IOCTL, 354803425671SKevin Wolf .aio_fildes = s->fd, 354903425671SKevin Wolf .aio_offset = 0, 355003425671SKevin Wolf .ioctl = { 355103425671SKevin Wolf .buf = buf, 355203425671SKevin Wolf .cmd = req, 355303425671SKevin Wolf }, 355403425671SKevin Wolf }; 355503425671SKevin Wolf 355603425671SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); 3557c1bb86cdSEric Blake } 3558c1bb86cdSEric Blake #endif /* linux */ 3559c1bb86cdSEric Blake 3560c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs) 3561c1bb86cdSEric Blake { 3562c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3563c1bb86cdSEric Blake 3564c1bb86cdSEric Blake /* this is just to ensure s->fd is sane (its called by io ops) */ 3565c1bb86cdSEric Blake if (s->fd >= 0) 3566c1bb86cdSEric Blake return 0; 3567c1bb86cdSEric Blake return -EIO; 3568c1bb86cdSEric Blake } 3569c1bb86cdSEric Blake 357033d70fb6SKevin Wolf static coroutine_fn int 357133d70fb6SKevin Wolf hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 3572c1bb86cdSEric Blake { 35731c450366SAnton Nefedov BDRVRawState *s = bs->opaque; 357433d70fb6SKevin Wolf int ret; 3575c1bb86cdSEric Blake 357633d70fb6SKevin Wolf ret = fd_open(bs); 357733d70fb6SKevin Wolf if (ret < 0) { 35781c450366SAnton Nefedov raw_account_discard(s, bytes, ret); 357933d70fb6SKevin Wolf return ret; 3580c1bb86cdSEric Blake } 358146ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, true); 3582c1bb86cdSEric Blake } 3583c1bb86cdSEric Blake 3584c1bb86cdSEric Blake static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, 3585f5a5ca79SManos Pitsidianakis int64_t offset, int bytes, BdrvRequestFlags flags) 3586c1bb86cdSEric Blake { 3587c1bb86cdSEric Blake int rc; 3588c1bb86cdSEric Blake 3589c1bb86cdSEric Blake rc = fd_open(bs); 3590c1bb86cdSEric Blake if (rc < 0) { 3591c1bb86cdSEric Blake return rc; 3592c1bb86cdSEric Blake } 359334fa110eSKevin Wolf 35947154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); 3595c1bb86cdSEric Blake } 3596c1bb86cdSEric Blake 3597c1bb86cdSEric Blake static BlockDriver bdrv_host_device = { 3598c1bb86cdSEric Blake .format_name = "host_device", 3599c1bb86cdSEric Blake .protocol_name = "host_device", 3600c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3601c1bb86cdSEric Blake .bdrv_needs_filename = true, 3602c1bb86cdSEric Blake .bdrv_probe_device = hdev_probe_device, 3603c1bb86cdSEric Blake .bdrv_parse_filename = hdev_parse_filename, 3604c1bb86cdSEric Blake .bdrv_file_open = hdev_open, 3605c1bb86cdSEric Blake .bdrv_close = raw_close, 3606c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3607c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3608c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 36095a5e7f8cSMaxim Levitsky .bdrv_co_create_opts = bdrv_co_create_opts_simple, 36105a5e7f8cSMaxim Levitsky .create_opts = &bdrv_create_opts_simple, 36118a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3612dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3613c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, 3614c1bb86cdSEric Blake 3615c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3616c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 361733d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 361833d70fb6SKevin Wolf .bdrv_co_pdiscard = hdev_co_pdiscard, 36191efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 36201efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 3621c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3622c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3623c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3624042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3625c1bb86cdSEric Blake 3626061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3627c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3628c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 3629c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3630c1bb86cdSEric Blake = raw_get_allocated_file_size, 3631d9245599SAnton Nefedov .bdrv_get_specific_stats = hdev_get_specific_stats, 3632244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 3633244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 3634244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 3635c1bb86cdSEric Blake .bdrv_probe_blocksizes = hdev_probe_blocksizes, 3636c1bb86cdSEric Blake .bdrv_probe_geometry = hdev_probe_geometry, 3637c1bb86cdSEric Blake 3638c1bb86cdSEric Blake /* generic scsi device */ 3639c1bb86cdSEric Blake #ifdef __linux__ 36402f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3641c1bb86cdSEric Blake #endif 3642c1bb86cdSEric Blake }; 3643c1bb86cdSEric Blake 3644c1bb86cdSEric Blake #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3645c1bb86cdSEric Blake static void cdrom_parse_filename(const char *filename, QDict *options, 3646c1bb86cdSEric Blake Error **errp) 3647c1bb86cdSEric Blake { 364803c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); 3649c1bb86cdSEric Blake } 3650c1bb86cdSEric Blake #endif 3651c1bb86cdSEric Blake 3652c1bb86cdSEric Blake #ifdef __linux__ 3653c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3654c1bb86cdSEric Blake Error **errp) 3655c1bb86cdSEric Blake { 3656c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3657c1bb86cdSEric Blake 3658c1bb86cdSEric Blake s->type = FTYPE_CD; 3659c1bb86cdSEric Blake 3660c1bb86cdSEric Blake /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ 3661230ff739SJohn Snow return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); 3662c1bb86cdSEric Blake } 3663c1bb86cdSEric Blake 3664c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3665c1bb86cdSEric Blake { 3666c1bb86cdSEric Blake int fd, ret; 3667c1bb86cdSEric Blake int prio = 0; 3668c1bb86cdSEric Blake struct stat st; 3669c1bb86cdSEric Blake 3670b18a24a9SDaniel P. Berrangé fd = qemu_open(filename, O_RDONLY | O_NONBLOCK, NULL); 3671c1bb86cdSEric Blake if (fd < 0) { 3672c1bb86cdSEric Blake goto out; 3673c1bb86cdSEric Blake } 3674c1bb86cdSEric Blake ret = fstat(fd, &st); 3675c1bb86cdSEric Blake if (ret == -1 || !S_ISBLK(st.st_mode)) { 3676c1bb86cdSEric Blake goto outc; 3677c1bb86cdSEric Blake } 3678c1bb86cdSEric Blake 3679c1bb86cdSEric Blake /* Attempt to detect via a CDROM specific ioctl */ 3680c1bb86cdSEric Blake ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3681c1bb86cdSEric Blake if (ret >= 0) 3682c1bb86cdSEric Blake prio = 100; 3683c1bb86cdSEric Blake 3684c1bb86cdSEric Blake outc: 3685c1bb86cdSEric Blake qemu_close(fd); 3686c1bb86cdSEric Blake out: 3687c1bb86cdSEric Blake return prio; 3688c1bb86cdSEric Blake } 3689c1bb86cdSEric Blake 3690c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3691c1bb86cdSEric Blake { 3692c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3693c1bb86cdSEric Blake int ret; 3694c1bb86cdSEric Blake 3695c1bb86cdSEric Blake ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3696c1bb86cdSEric Blake return ret == CDS_DISC_OK; 3697c1bb86cdSEric Blake } 3698c1bb86cdSEric Blake 3699c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3700c1bb86cdSEric Blake { 3701c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3702c1bb86cdSEric Blake 3703c1bb86cdSEric Blake if (eject_flag) { 3704c1bb86cdSEric Blake if (ioctl(s->fd, CDROMEJECT, NULL) < 0) 3705c1bb86cdSEric Blake perror("CDROMEJECT"); 3706c1bb86cdSEric Blake } else { 3707c1bb86cdSEric Blake if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) 3708c1bb86cdSEric Blake perror("CDROMEJECT"); 3709c1bb86cdSEric Blake } 3710c1bb86cdSEric Blake } 3711c1bb86cdSEric Blake 3712c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3713c1bb86cdSEric Blake { 3714c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3715c1bb86cdSEric Blake 3716c1bb86cdSEric Blake if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { 3717c1bb86cdSEric Blake /* 3718c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3719c1bb86cdSEric Blake * mounts the CD-ROM 3720c1bb86cdSEric Blake */ 3721c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3722c1bb86cdSEric Blake } 3723c1bb86cdSEric Blake } 3724c1bb86cdSEric Blake 3725c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3726c1bb86cdSEric Blake .format_name = "host_cdrom", 3727c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3728c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3729c1bb86cdSEric Blake .bdrv_needs_filename = true, 3730c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3731c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3732c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3733c1bb86cdSEric Blake .bdrv_close = raw_close, 3734c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3735c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3736c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 37375a5e7f8cSMaxim Levitsky .bdrv_co_create_opts = bdrv_co_create_opts_simple, 37385a5e7f8cSMaxim Levitsky .create_opts = &bdrv_create_opts_simple, 37398a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3740dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3741c1bb86cdSEric Blake 3742c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3743c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 374433d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3745c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3746c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3747c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3748042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3749c1bb86cdSEric Blake 3750061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3751c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3752c1bb86cdSEric Blake .has_variable_length = true, 3753c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3754c1bb86cdSEric Blake = raw_get_allocated_file_size, 3755c1bb86cdSEric Blake 3756c1bb86cdSEric Blake /* removable device support */ 3757c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3758c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3759c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3760c1bb86cdSEric Blake 3761c1bb86cdSEric Blake /* generic scsi device */ 37622f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3763c1bb86cdSEric Blake }; 3764c1bb86cdSEric Blake #endif /* __linux__ */ 3765c1bb86cdSEric Blake 3766c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 3767c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3768c1bb86cdSEric Blake Error **errp) 3769c1bb86cdSEric Blake { 3770c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3771c1bb86cdSEric Blake int ret; 3772c1bb86cdSEric Blake 3773c1bb86cdSEric Blake s->type = FTYPE_CD; 3774c1bb86cdSEric Blake 3775668f62ecSMarkus Armbruster ret = raw_open_common(bs, options, flags, 0, true, errp); 3776c1bb86cdSEric Blake if (ret) { 3777c1bb86cdSEric Blake return ret; 3778c1bb86cdSEric Blake } 3779c1bb86cdSEric Blake 3780c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3781c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3782c1bb86cdSEric Blake return 0; 3783c1bb86cdSEric Blake } 3784c1bb86cdSEric Blake 3785c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3786c1bb86cdSEric Blake { 3787c1bb86cdSEric Blake if (strstart(filename, "/dev/cd", NULL) || 3788c1bb86cdSEric Blake strstart(filename, "/dev/acd", NULL)) 3789c1bb86cdSEric Blake return 100; 3790c1bb86cdSEric Blake return 0; 3791c1bb86cdSEric Blake } 3792c1bb86cdSEric Blake 3793c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs) 3794c1bb86cdSEric Blake { 3795c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3796c1bb86cdSEric Blake int fd; 3797c1bb86cdSEric Blake 3798c1bb86cdSEric Blake /* 3799c1bb86cdSEric Blake * Force reread of possibly changed/newly loaded disc, 3800c1bb86cdSEric Blake * FreeBSD seems to not notice sometimes... 3801c1bb86cdSEric Blake */ 3802c1bb86cdSEric Blake if (s->fd >= 0) 3803c1bb86cdSEric Blake qemu_close(s->fd); 3804b18a24a9SDaniel P. Berrangé fd = qemu_open(bs->filename, s->open_flags, NULL); 3805c1bb86cdSEric Blake if (fd < 0) { 3806c1bb86cdSEric Blake s->fd = -1; 3807c1bb86cdSEric Blake return -EIO; 3808c1bb86cdSEric Blake } 3809c1bb86cdSEric Blake s->fd = fd; 3810c1bb86cdSEric Blake 3811c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3812c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3813c1bb86cdSEric Blake return 0; 3814c1bb86cdSEric Blake } 3815c1bb86cdSEric Blake 3816c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3817c1bb86cdSEric Blake { 3818c1bb86cdSEric Blake return raw_getlength(bs) > 0; 3819c1bb86cdSEric Blake } 3820c1bb86cdSEric Blake 3821c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3822c1bb86cdSEric Blake { 3823c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3824c1bb86cdSEric Blake 3825c1bb86cdSEric Blake if (s->fd < 0) 3826c1bb86cdSEric Blake return; 3827c1bb86cdSEric Blake 3828c1bb86cdSEric Blake (void) ioctl(s->fd, CDIOCALLOW); 3829c1bb86cdSEric Blake 3830c1bb86cdSEric Blake if (eject_flag) { 3831c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCEJECT) < 0) 3832c1bb86cdSEric Blake perror("CDIOCEJECT"); 3833c1bb86cdSEric Blake } else { 3834c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCCLOSE) < 0) 3835c1bb86cdSEric Blake perror("CDIOCCLOSE"); 3836c1bb86cdSEric Blake } 3837c1bb86cdSEric Blake 3838c1bb86cdSEric Blake cdrom_reopen(bs); 3839c1bb86cdSEric Blake } 3840c1bb86cdSEric Blake 3841c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3842c1bb86cdSEric Blake { 3843c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3844c1bb86cdSEric Blake 3845c1bb86cdSEric Blake if (s->fd < 0) 3846c1bb86cdSEric Blake return; 3847c1bb86cdSEric Blake if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { 3848c1bb86cdSEric Blake /* 3849c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3850c1bb86cdSEric Blake * mounts the CD-ROM 3851c1bb86cdSEric Blake */ 3852c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3853c1bb86cdSEric Blake } 3854c1bb86cdSEric Blake } 3855c1bb86cdSEric Blake 3856c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3857c1bb86cdSEric Blake .format_name = "host_cdrom", 3858c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3859c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3860c1bb86cdSEric Blake .bdrv_needs_filename = true, 3861c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3862c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3863c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3864c1bb86cdSEric Blake .bdrv_close = raw_close, 3865c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3866c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3867c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 38685a5e7f8cSMaxim Levitsky .bdrv_co_create_opts = bdrv_co_create_opts_simple, 38695a5e7f8cSMaxim Levitsky .create_opts = &bdrv_create_opts_simple, 38708a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3871c1bb86cdSEric Blake 3872c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3873c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 387433d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3875c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3876c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3877c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3878042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3879c1bb86cdSEric Blake 3880061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3881c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3882c1bb86cdSEric Blake .has_variable_length = true, 3883c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3884c1bb86cdSEric Blake = raw_get_allocated_file_size, 3885c1bb86cdSEric Blake 3886c1bb86cdSEric Blake /* removable device support */ 3887c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3888c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3889c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3890c1bb86cdSEric Blake }; 3891c1bb86cdSEric Blake #endif /* __FreeBSD__ */ 3892c1bb86cdSEric Blake 3893c1bb86cdSEric Blake static void bdrv_file_init(void) 3894c1bb86cdSEric Blake { 3895c1bb86cdSEric Blake /* 3896c1bb86cdSEric Blake * Register all the drivers. Note that order is important, the driver 3897c1bb86cdSEric Blake * registered last will get probed first. 3898c1bb86cdSEric Blake */ 3899c1bb86cdSEric Blake bdrv_register(&bdrv_file); 3900c1bb86cdSEric Blake bdrv_register(&bdrv_host_device); 3901c1bb86cdSEric Blake #ifdef __linux__ 3902c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3903c1bb86cdSEric Blake #endif 3904c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3905c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3906c1bb86cdSEric Blake #endif 3907c1bb86cdSEric Blake } 3908c1bb86cdSEric Blake 3909c1bb86cdSEric Blake block_init(bdrv_file_init); 3910