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 open_flags; 179f357fcd8SStefan Hajnoczi bool drop_cache; 18031be8a2aSStefan Hajnoczi bool check_cache_dropped; 181c1bb86cdSEric Blake } BDRVRawReopenState; 182c1bb86cdSEric Blake 183c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs); 184c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs); 185c1bb86cdSEric Blake 186c1bb86cdSEric Blake typedef struct RawPosixAIOData { 187c1bb86cdSEric Blake BlockDriverState *bs; 188c1bb86cdSEric Blake int aio_type; 189d57c44d0SKevin Wolf int aio_fildes; 190d57c44d0SKevin Wolf 191d57c44d0SKevin Wolf off_t aio_offset; 192d57c44d0SKevin Wolf uint64_t aio_nbytes; 193d57c44d0SKevin Wolf 19493f4e2ffSKevin Wolf union { 19593f4e2ffSKevin Wolf struct { 196d57c44d0SKevin Wolf struct iovec *iov; 197d57c44d0SKevin Wolf int niov; 198d57c44d0SKevin Wolf } io; 199d57c44d0SKevin Wolf struct { 200d57c44d0SKevin Wolf uint64_t cmd; 201d57c44d0SKevin Wolf void *buf; 202d57c44d0SKevin Wolf } ioctl; 203d57c44d0SKevin Wolf struct { 2041efad060SFam Zheng int aio_fd2; 2051efad060SFam Zheng off_t aio_offset2; 206d57c44d0SKevin Wolf } copy_range; 20793f4e2ffSKevin Wolf struct { 20893f4e2ffSKevin Wolf PreallocMode prealloc; 20993f4e2ffSKevin Wolf Error **errp; 210d57c44d0SKevin Wolf } truncate; 21193f4e2ffSKevin Wolf }; 212c1bb86cdSEric Blake } RawPosixAIOData; 213c1bb86cdSEric Blake 214c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 215c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs); 216c1bb86cdSEric Blake #endif 217c1bb86cdSEric Blake 218797e3e38SDavid Edmondson /* 219797e3e38SDavid Edmondson * Elide EAGAIN and EACCES details when failing to lock, as this 220797e3e38SDavid Edmondson * indicates that the specified file region is already locked by 221797e3e38SDavid Edmondson * another process, which is considered a common scenario. 222797e3e38SDavid Edmondson */ 223797e3e38SDavid Edmondson #define raw_lock_error_setg_errno(errp, err, fmt, ...) \ 224797e3e38SDavid Edmondson do { \ 225797e3e38SDavid Edmondson if ((err) == EAGAIN || (err) == EACCES) { \ 226797e3e38SDavid Edmondson error_setg((errp), (fmt), ## __VA_ARGS__); \ 227797e3e38SDavid Edmondson } else { \ 228797e3e38SDavid Edmondson error_setg_errno((errp), (err), (fmt), ## __VA_ARGS__); \ 229797e3e38SDavid Edmondson } \ 230797e3e38SDavid Edmondson } while (0) 231797e3e38SDavid Edmondson 232c1bb86cdSEric Blake #if defined(__NetBSD__) 233db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 234c1bb86cdSEric Blake { 235c1bb86cdSEric Blake static char namebuf[PATH_MAX]; 236c1bb86cdSEric Blake const char *dp, *fname; 237c1bb86cdSEric Blake struct stat sb; 238c1bb86cdSEric Blake 239c1bb86cdSEric Blake fname = *filename; 240c1bb86cdSEric Blake dp = strrchr(fname, '/'); 241c1bb86cdSEric Blake if (lstat(fname, &sb) < 0) { 242f6fc1e30SPaolo Bonzini error_setg_file_open(errp, errno, fname); 243c1bb86cdSEric Blake return -errno; 244c1bb86cdSEric Blake } 245c1bb86cdSEric Blake 246c1bb86cdSEric Blake if (!S_ISBLK(sb.st_mode)) { 247c1bb86cdSEric Blake return 0; 248c1bb86cdSEric Blake } 249c1bb86cdSEric Blake 250c1bb86cdSEric Blake if (dp == NULL) { 251c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "r%s", fname); 252c1bb86cdSEric Blake } else { 253c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "%.*s/r%s", 254c1bb86cdSEric Blake (int)(dp - fname), fname, dp + 1); 255c1bb86cdSEric Blake } 256c1bb86cdSEric Blake *filename = namebuf; 257db0754dfSFam Zheng warn_report("%s is a block device, using %s", fname, *filename); 258c1bb86cdSEric Blake 259c1bb86cdSEric Blake return 0; 260c1bb86cdSEric Blake } 261c1bb86cdSEric Blake #else 262db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 263c1bb86cdSEric Blake { 264c1bb86cdSEric Blake return 0; 265c1bb86cdSEric Blake } 266c1bb86cdSEric Blake #endif 267c1bb86cdSEric Blake 268c1bb86cdSEric Blake /* 269c1bb86cdSEric Blake * Get logical block size via ioctl. On success store it in @sector_size_p. 270c1bb86cdSEric Blake */ 271c1bb86cdSEric Blake static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) 272c1bb86cdSEric Blake { 273c1bb86cdSEric Blake unsigned int sector_size; 274c1bb86cdSEric Blake bool success = false; 275700f9ce0SPeter Maydell int i; 276c1bb86cdSEric Blake 277c1bb86cdSEric Blake errno = ENOTSUP; 278700f9ce0SPeter Maydell static const unsigned long ioctl_list[] = { 279c1bb86cdSEric Blake #ifdef BLKSSZGET 280700f9ce0SPeter Maydell BLKSSZGET, 281c1bb86cdSEric Blake #endif 282c1bb86cdSEric Blake #ifdef DKIOCGETBLOCKSIZE 283700f9ce0SPeter Maydell DKIOCGETBLOCKSIZE, 284c1bb86cdSEric Blake #endif 285c1bb86cdSEric Blake #ifdef DIOCGSECTORSIZE 286700f9ce0SPeter Maydell DIOCGSECTORSIZE, 287700f9ce0SPeter Maydell #endif 288700f9ce0SPeter Maydell }; 289700f9ce0SPeter Maydell 290700f9ce0SPeter Maydell /* Try a few ioctls to get the right size */ 291700f9ce0SPeter Maydell for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { 292700f9ce0SPeter Maydell if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { 293c1bb86cdSEric Blake *sector_size_p = sector_size; 294c1bb86cdSEric Blake success = true; 295c1bb86cdSEric Blake } 296700f9ce0SPeter Maydell } 297c1bb86cdSEric Blake 298c1bb86cdSEric Blake return success ? 0 : -errno; 299c1bb86cdSEric Blake } 300c1bb86cdSEric Blake 301c1bb86cdSEric Blake /** 302c1bb86cdSEric Blake * Get physical block size of @fd. 303c1bb86cdSEric Blake * On success, store it in @blk_size and return 0. 304c1bb86cdSEric Blake * On failure, return -errno. 305c1bb86cdSEric Blake */ 306c1bb86cdSEric Blake static int probe_physical_blocksize(int fd, unsigned int *blk_size) 307c1bb86cdSEric Blake { 308c1bb86cdSEric Blake #ifdef BLKPBSZGET 309c1bb86cdSEric Blake if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { 310c1bb86cdSEric Blake return -errno; 311c1bb86cdSEric Blake } 312c1bb86cdSEric Blake return 0; 313c1bb86cdSEric Blake #else 314c1bb86cdSEric Blake return -ENOTSUP; 315c1bb86cdSEric Blake #endif 316c1bb86cdSEric Blake } 317c1bb86cdSEric Blake 3185edc8557SKevin Wolf /* 3195edc8557SKevin Wolf * Returns true if no alignment restrictions are necessary even for files 3205edc8557SKevin Wolf * opened with O_DIRECT. 3215edc8557SKevin Wolf * 3225edc8557SKevin Wolf * raw_probe_alignment() probes the required alignment and assume that 1 means 3235edc8557SKevin Wolf * the probing failed, so it falls back to a safe default of 4k. This can be 3245edc8557SKevin Wolf * avoided if we know that byte alignment is okay for the file. 3255edc8557SKevin Wolf */ 3265edc8557SKevin Wolf static bool dio_byte_aligned(int fd) 3275edc8557SKevin Wolf { 3285edc8557SKevin Wolf #ifdef __linux__ 3295edc8557SKevin Wolf struct statfs buf; 3305edc8557SKevin Wolf int ret; 3315edc8557SKevin Wolf 3325edc8557SKevin Wolf ret = fstatfs(fd, &buf); 3335edc8557SKevin Wolf if (ret == 0 && buf.f_type == NFS_SUPER_MAGIC) { 3345edc8557SKevin Wolf return true; 3355edc8557SKevin Wolf } 3365edc8557SKevin Wolf #endif 3375edc8557SKevin Wolf return false; 3385edc8557SKevin Wolf } 3395edc8557SKevin Wolf 340c1bb86cdSEric Blake /* Check if read is allowed with given memory buffer and length. 341c1bb86cdSEric Blake * 342c1bb86cdSEric Blake * This function is used to check O_DIRECT memory buffer and request alignment. 343c1bb86cdSEric Blake */ 344c1bb86cdSEric Blake static bool raw_is_io_aligned(int fd, void *buf, size_t len) 345c1bb86cdSEric Blake { 346c1bb86cdSEric Blake ssize_t ret = pread(fd, buf, len, 0); 347c1bb86cdSEric Blake 348c1bb86cdSEric Blake if (ret >= 0) { 349c1bb86cdSEric Blake return true; 350c1bb86cdSEric Blake } 351c1bb86cdSEric Blake 352c1bb86cdSEric Blake #ifdef __linux__ 353c1bb86cdSEric Blake /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore 354c1bb86cdSEric Blake * other errors (e.g. real I/O error), which could happen on a failed 355c1bb86cdSEric Blake * drive, since we only care about probing alignment. 356c1bb86cdSEric Blake */ 357c1bb86cdSEric Blake if (errno != EINVAL) { 358c1bb86cdSEric Blake return true; 359c1bb86cdSEric Blake } 360c1bb86cdSEric Blake #endif 361c1bb86cdSEric Blake 362c1bb86cdSEric Blake return false; 363c1bb86cdSEric Blake } 364c1bb86cdSEric Blake 365c1bb86cdSEric Blake static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) 366c1bb86cdSEric Blake { 367c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 368c1bb86cdSEric Blake char *buf; 369038adc2fSWei Yang size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); 370a6b257a0SNir Soffer size_t alignments[] = {1, 512, 1024, 2048, 4096}; 371c1bb86cdSEric Blake 372c1bb86cdSEric Blake /* For SCSI generic devices the alignment is not really used. 373c1bb86cdSEric Blake With buffered I/O, we don't have any restrictions. */ 374c1bb86cdSEric Blake if (bdrv_is_sg(bs) || !s->needs_alignment) { 375c1bb86cdSEric Blake bs->bl.request_alignment = 1; 376c1bb86cdSEric Blake s->buf_align = 1; 377c1bb86cdSEric Blake return; 378c1bb86cdSEric Blake } 379c1bb86cdSEric Blake 380c1bb86cdSEric Blake bs->bl.request_alignment = 0; 381c1bb86cdSEric Blake s->buf_align = 0; 382c1bb86cdSEric Blake /* Let's try to use the logical blocksize for the alignment. */ 383c1bb86cdSEric Blake if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { 384c1bb86cdSEric Blake bs->bl.request_alignment = 0; 385c1bb86cdSEric Blake } 386c1bb86cdSEric Blake #ifdef CONFIG_XFS 387c1bb86cdSEric Blake if (s->is_xfs) { 388c1bb86cdSEric Blake struct dioattr da; 389c1bb86cdSEric Blake if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { 390c1bb86cdSEric Blake bs->bl.request_alignment = da.d_miniosz; 391c1bb86cdSEric Blake /* The kernel returns wrong information for d_mem */ 392c1bb86cdSEric Blake /* s->buf_align = da.d_mem; */ 393c1bb86cdSEric Blake } 394c1bb86cdSEric Blake } 395c1bb86cdSEric Blake #endif 396c1bb86cdSEric Blake 397a6b257a0SNir Soffer /* 398a6b257a0SNir Soffer * If we could not get the sizes so far, we can only guess them. First try 399a6b257a0SNir Soffer * to detect request alignment, since it is more likely to succeed. Then 400a6b257a0SNir Soffer * try to detect buf_align, which cannot be detected in some cases (e.g. 401a6b257a0SNir Soffer * Gluster). If buf_align cannot be detected, we fallback to the value of 402a6b257a0SNir Soffer * request_alignment. 403a6b257a0SNir Soffer */ 404a6b257a0SNir Soffer 405a6b257a0SNir Soffer if (!bs->bl.request_alignment) { 406a6b257a0SNir Soffer int i; 407c1bb86cdSEric Blake size_t align; 408a6b257a0SNir Soffer buf = qemu_memalign(max_align, max_align); 409a6b257a0SNir Soffer for (i = 0; i < ARRAY_SIZE(alignments); i++) { 410a6b257a0SNir Soffer align = alignments[i]; 411a6b257a0SNir Soffer if (raw_is_io_aligned(fd, buf, align)) { 412a6b257a0SNir Soffer /* Fallback to safe value. */ 413a6b257a0SNir Soffer bs->bl.request_alignment = (align != 1) ? align : max_align; 414c1bb86cdSEric Blake break; 415c1bb86cdSEric Blake } 416c1bb86cdSEric Blake } 417c1bb86cdSEric Blake qemu_vfree(buf); 418c1bb86cdSEric Blake } 419c1bb86cdSEric Blake 420a6b257a0SNir Soffer if (!s->buf_align) { 421a6b257a0SNir Soffer int i; 422c1bb86cdSEric Blake size_t align; 423a6b257a0SNir Soffer buf = qemu_memalign(max_align, 2 * max_align); 424a6b257a0SNir Soffer for (i = 0; i < ARRAY_SIZE(alignments); i++) { 425a6b257a0SNir Soffer align = alignments[i]; 426a6b257a0SNir Soffer if (raw_is_io_aligned(fd, buf + align, max_align)) { 427236094c7SStefan Hajnoczi /* Fallback to request_alignment. */ 428a6b257a0SNir Soffer s->buf_align = (align != 1) ? align : bs->bl.request_alignment; 429c1bb86cdSEric Blake break; 430c1bb86cdSEric Blake } 431c1bb86cdSEric Blake } 432c1bb86cdSEric Blake qemu_vfree(buf); 433c1bb86cdSEric Blake } 434c1bb86cdSEric Blake 435c1bb86cdSEric Blake if (!s->buf_align || !bs->bl.request_alignment) { 436c1bb86cdSEric Blake error_setg(errp, "Could not find working O_DIRECT alignment"); 437c1bb86cdSEric Blake error_append_hint(errp, "Try cache.direct=off\n"); 438c1bb86cdSEric Blake } 439c1bb86cdSEric Blake } 440c1bb86cdSEric Blake 441bca5283bSKevin Wolf static int check_hdev_writable(int fd) 44220eaf1bfSKevin Wolf { 44320eaf1bfSKevin Wolf #if defined(BLKROGET) 44420eaf1bfSKevin Wolf /* Linux block devices can be configured "read-only" using blockdev(8). 44520eaf1bfSKevin Wolf * This is independent of device node permissions and therefore open(2) 44620eaf1bfSKevin Wolf * with O_RDWR succeeds. Actual writes fail with EPERM. 44720eaf1bfSKevin Wolf * 44820eaf1bfSKevin Wolf * bdrv_open() is supposed to fail if the disk is read-only. Explicitly 44920eaf1bfSKevin Wolf * check for read-only block devices so that Linux block devices behave 45020eaf1bfSKevin Wolf * properly. 45120eaf1bfSKevin Wolf */ 45220eaf1bfSKevin Wolf struct stat st; 45320eaf1bfSKevin Wolf int readonly = 0; 45420eaf1bfSKevin Wolf 455bca5283bSKevin Wolf if (fstat(fd, &st)) { 45620eaf1bfSKevin Wolf return -errno; 45720eaf1bfSKevin Wolf } 45820eaf1bfSKevin Wolf 45920eaf1bfSKevin Wolf if (!S_ISBLK(st.st_mode)) { 46020eaf1bfSKevin Wolf return 0; 46120eaf1bfSKevin Wolf } 46220eaf1bfSKevin Wolf 463bca5283bSKevin Wolf if (ioctl(fd, BLKROGET, &readonly) < 0) { 46420eaf1bfSKevin Wolf return -errno; 46520eaf1bfSKevin Wolf } 46620eaf1bfSKevin Wolf 46720eaf1bfSKevin Wolf if (readonly) { 46820eaf1bfSKevin Wolf return -EACCES; 46920eaf1bfSKevin Wolf } 47020eaf1bfSKevin Wolf #endif /* defined(BLKROGET) */ 47120eaf1bfSKevin Wolf return 0; 47220eaf1bfSKevin Wolf } 47320eaf1bfSKevin Wolf 47423dece19SKevin Wolf static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) 475c1bb86cdSEric Blake { 47623dece19SKevin Wolf bool read_write = false; 477c1bb86cdSEric Blake assert(open_flags != NULL); 478c1bb86cdSEric Blake 479c1bb86cdSEric Blake *open_flags |= O_BINARY; 480c1bb86cdSEric Blake *open_flags &= ~O_ACCMODE; 48123dece19SKevin Wolf 48223dece19SKevin Wolf if (bdrv_flags & BDRV_O_AUTO_RDONLY) { 48323dece19SKevin Wolf read_write = has_writers; 48423dece19SKevin Wolf } else if (bdrv_flags & BDRV_O_RDWR) { 48523dece19SKevin Wolf read_write = true; 48623dece19SKevin Wolf } 48723dece19SKevin Wolf 48823dece19SKevin Wolf if (read_write) { 489c1bb86cdSEric Blake *open_flags |= O_RDWR; 490c1bb86cdSEric Blake } else { 491c1bb86cdSEric Blake *open_flags |= O_RDONLY; 492c1bb86cdSEric Blake } 493c1bb86cdSEric Blake 494c1bb86cdSEric Blake /* Use O_DSYNC for write-through caching, no flags for write-back caching, 495c1bb86cdSEric Blake * and O_DIRECT for no caching. */ 496c1bb86cdSEric Blake if ((bdrv_flags & BDRV_O_NOCACHE)) { 497c1bb86cdSEric Blake *open_flags |= O_DIRECT; 498c1bb86cdSEric Blake } 499c1bb86cdSEric Blake } 500c1bb86cdSEric Blake 501c1bb86cdSEric Blake static void raw_parse_filename(const char *filename, QDict *options, 502c1bb86cdSEric Blake Error **errp) 503c1bb86cdSEric Blake { 50403c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "file:", options); 505c1bb86cdSEric Blake } 506c1bb86cdSEric Blake 507c1bb86cdSEric Blake static QemuOptsList raw_runtime_opts = { 508c1bb86cdSEric Blake .name = "raw", 509c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), 510c1bb86cdSEric Blake .desc = { 511c1bb86cdSEric Blake { 512c1bb86cdSEric Blake .name = "filename", 513c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 514c1bb86cdSEric Blake .help = "File name of the image", 515c1bb86cdSEric Blake }, 516c1bb86cdSEric Blake { 517c1bb86cdSEric Blake .name = "aio", 518c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 519c6447510SAarushi Mehta .help = "host AIO implementation (threads, native, io_uring)", 520c1bb86cdSEric Blake }, 52116b48d5dSFam Zheng { 52216b48d5dSFam Zheng .name = "locking", 52316b48d5dSFam Zheng .type = QEMU_OPT_STRING, 52416b48d5dSFam Zheng .help = "file locking mode (on/off/auto, default: auto)", 52516b48d5dSFam Zheng }, 5267c9e5276SPaolo Bonzini { 5277c9e5276SPaolo Bonzini .name = "pr-manager", 5287c9e5276SPaolo Bonzini .type = QEMU_OPT_STRING, 5297c9e5276SPaolo Bonzini .help = "id of persistent reservation manager object (default: none)", 5307c9e5276SPaolo Bonzini }, 531f357fcd8SStefan Hajnoczi #if defined(__linux__) 532f357fcd8SStefan Hajnoczi { 533f357fcd8SStefan Hajnoczi .name = "drop-cache", 534f357fcd8SStefan Hajnoczi .type = QEMU_OPT_BOOL, 535f357fcd8SStefan Hajnoczi .help = "invalidate page cache during live migration (default: on)", 536f357fcd8SStefan Hajnoczi }, 537f357fcd8SStefan Hajnoczi #endif 53831be8a2aSStefan Hajnoczi { 53931be8a2aSStefan Hajnoczi .name = "x-check-cache-dropped", 54031be8a2aSStefan Hajnoczi .type = QEMU_OPT_BOOL, 54131be8a2aSStefan Hajnoczi .help = "check that page cache was dropped on live migration (default: off)" 54231be8a2aSStefan Hajnoczi }, 543c1bb86cdSEric Blake { /* end of list */ } 544c1bb86cdSEric Blake }, 545c1bb86cdSEric Blake }; 546c1bb86cdSEric Blake 5478a2ce0bcSAlberto Garcia static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; 5488a2ce0bcSAlberto Garcia 549c1bb86cdSEric Blake static int raw_open_common(BlockDriverState *bs, QDict *options, 550230ff739SJohn Snow int bdrv_flags, int open_flags, 551230ff739SJohn Snow bool device, Error **errp) 552c1bb86cdSEric Blake { 553c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 554c1bb86cdSEric Blake QemuOpts *opts; 555c1bb86cdSEric Blake Error *local_err = NULL; 556c1bb86cdSEric Blake const char *filename = NULL; 5577c9e5276SPaolo Bonzini const char *str; 558c1bb86cdSEric Blake BlockdevAioOptions aio, aio_default; 559c1bb86cdSEric Blake int fd, ret; 560c1bb86cdSEric Blake struct stat st; 561244a5668SFam Zheng OnOffAuto locking; 562c1bb86cdSEric Blake 563c1bb86cdSEric Blake opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 564af175e85SMarkus Armbruster if (!qemu_opts_absorb_qdict(opts, options, errp)) { 565c1bb86cdSEric Blake ret = -EINVAL; 566c1bb86cdSEric Blake goto fail; 567c1bb86cdSEric Blake } 568c1bb86cdSEric Blake 569c1bb86cdSEric Blake filename = qemu_opt_get(opts, "filename"); 570c1bb86cdSEric Blake 571db0754dfSFam Zheng ret = raw_normalize_devicepath(&filename, errp); 572c1bb86cdSEric Blake if (ret != 0) { 573c1bb86cdSEric Blake goto fail; 574c1bb86cdSEric Blake } 575c1bb86cdSEric Blake 576c6447510SAarushi Mehta if (bdrv_flags & BDRV_O_NATIVE_AIO) { 577c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE; 578c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 579c6447510SAarushi Mehta } else if (bdrv_flags & BDRV_O_IO_URING) { 580c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING; 581c6447510SAarushi Mehta #endif 582c6447510SAarushi Mehta } else { 583c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_THREADS; 584c6447510SAarushi Mehta } 585c6447510SAarushi Mehta 586f7abe0ecSMarc-André Lureau aio = qapi_enum_parse(&BlockdevAioOptions_lookup, 587f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "aio"), 58806c60b6cSMarkus Armbruster aio_default, &local_err); 589c1bb86cdSEric Blake if (local_err) { 590c1bb86cdSEric Blake error_propagate(errp, local_err); 591c1bb86cdSEric Blake ret = -EINVAL; 592c1bb86cdSEric Blake goto fail; 593c1bb86cdSEric Blake } 594c6447510SAarushi Mehta 595c1bb86cdSEric Blake s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); 596c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 597c6447510SAarushi Mehta s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING); 598c6447510SAarushi Mehta #endif 599c1bb86cdSEric Blake 600f7abe0ecSMarc-André Lureau locking = qapi_enum_parse(&OnOffAuto_lookup, 601f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "locking"), 60206c60b6cSMarkus Armbruster ON_OFF_AUTO_AUTO, &local_err); 603244a5668SFam Zheng if (local_err) { 604244a5668SFam Zheng error_propagate(errp, local_err); 605244a5668SFam Zheng ret = -EINVAL; 606244a5668SFam Zheng goto fail; 607244a5668SFam Zheng } 608244a5668SFam Zheng switch (locking) { 609244a5668SFam Zheng case ON_OFF_AUTO_ON: 610244a5668SFam Zheng s->use_lock = true; 6112b218f5dSFam Zheng if (!qemu_has_ofd_lock()) { 612db0754dfSFam Zheng warn_report("File lock requested but OFD locking syscall is " 613db0754dfSFam Zheng "unavailable, falling back to POSIX file locks"); 614db0754dfSFam Zheng error_printf("Due to the implementation, locks can be lost " 6152b218f5dSFam Zheng "unexpectedly.\n"); 6162b218f5dSFam Zheng } 617244a5668SFam Zheng break; 618244a5668SFam Zheng case ON_OFF_AUTO_OFF: 619244a5668SFam Zheng s->use_lock = false; 620244a5668SFam Zheng break; 621244a5668SFam Zheng case ON_OFF_AUTO_AUTO: 6222b218f5dSFam Zheng s->use_lock = qemu_has_ofd_lock(); 623244a5668SFam Zheng break; 624244a5668SFam Zheng default: 625244a5668SFam Zheng abort(); 626244a5668SFam Zheng } 627244a5668SFam Zheng 6287c9e5276SPaolo Bonzini str = qemu_opt_get(opts, "pr-manager"); 6297c9e5276SPaolo Bonzini if (str) { 6307c9e5276SPaolo Bonzini s->pr_mgr = pr_manager_lookup(str, &local_err); 6317c9e5276SPaolo Bonzini if (local_err) { 6327c9e5276SPaolo Bonzini error_propagate(errp, local_err); 6337c9e5276SPaolo Bonzini ret = -EINVAL; 6347c9e5276SPaolo Bonzini goto fail; 6357c9e5276SPaolo Bonzini } 6367c9e5276SPaolo Bonzini } 6377c9e5276SPaolo Bonzini 638f357fcd8SStefan Hajnoczi s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); 63931be8a2aSStefan Hajnoczi s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", 64031be8a2aSStefan Hajnoczi false); 64131be8a2aSStefan Hajnoczi 642c1bb86cdSEric Blake s->open_flags = open_flags; 64323dece19SKevin Wolf raw_parse_flags(bdrv_flags, &s->open_flags, false); 644c1bb86cdSEric Blake 645c1bb86cdSEric Blake s->fd = -1; 646b18a24a9SDaniel P. Berrangé fd = qemu_open(filename, s->open_flags, errp); 64764107dc0SKevin Wolf ret = fd < 0 ? -errno : 0; 64864107dc0SKevin Wolf 64964107dc0SKevin Wolf if (ret < 0) { 650c1bb86cdSEric Blake if (ret == -EROFS) { 651c1bb86cdSEric Blake ret = -EACCES; 652c1bb86cdSEric Blake } 653c1bb86cdSEric Blake goto fail; 654c1bb86cdSEric Blake } 655c1bb86cdSEric Blake s->fd = fd; 656c1bb86cdSEric Blake 657bca5283bSKevin Wolf /* Check s->open_flags rather than bdrv_flags due to auto-read-only */ 658bca5283bSKevin Wolf if (s->open_flags & O_RDWR) { 659bca5283bSKevin Wolf ret = check_hdev_writable(s->fd); 660bca5283bSKevin Wolf if (ret < 0) { 661bca5283bSKevin Wolf error_setg_errno(errp, -ret, "The device is not writable"); 662bca5283bSKevin Wolf goto fail; 663bca5283bSKevin Wolf } 664bca5283bSKevin Wolf } 665bca5283bSKevin Wolf 666244a5668SFam Zheng s->perm = 0; 667244a5668SFam Zheng s->shared_perm = BLK_PERM_ALL; 668244a5668SFam Zheng 669c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 670c1bb86cdSEric Blake /* Currently Linux does AIO only for files opened with O_DIRECT */ 671ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 672ed6e2161SNishanth Aravamudan if (!(s->open_flags & O_DIRECT)) { 673c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but it requires " 674c1bb86cdSEric Blake "cache.direct=on, which was not specified."); 675c1bb86cdSEric Blake ret = -EINVAL; 676c1bb86cdSEric Blake goto fail; 677c1bb86cdSEric Blake } 678ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { 679ed6e2161SNishanth Aravamudan error_prepend(errp, "Unable to use native AIO: "); 680ed6e2161SNishanth Aravamudan goto fail; 681ed6e2161SNishanth Aravamudan } 682ed6e2161SNishanth Aravamudan } 683c1bb86cdSEric Blake #else 684c1bb86cdSEric Blake if (s->use_linux_aio) { 685c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but is not supported " 686c1bb86cdSEric Blake "in this build."); 687c1bb86cdSEric Blake ret = -EINVAL; 688c1bb86cdSEric Blake goto fail; 689c1bb86cdSEric Blake } 690c1bb86cdSEric Blake #endif /* !defined(CONFIG_LINUX_AIO) */ 691c1bb86cdSEric Blake 692c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 693c6447510SAarushi Mehta if (s->use_linux_io_uring) { 694c6447510SAarushi Mehta if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs), errp)) { 695c6447510SAarushi Mehta error_prepend(errp, "Unable to use io_uring: "); 696c6447510SAarushi Mehta goto fail; 697c6447510SAarushi Mehta } 698c6447510SAarushi Mehta } 699c6447510SAarushi Mehta #else 700c6447510SAarushi Mehta if (s->use_linux_io_uring) { 701c6447510SAarushi Mehta error_setg(errp, "aio=io_uring was specified, but is not supported " 702c6447510SAarushi Mehta "in this build."); 703c6447510SAarushi Mehta ret = -EINVAL; 704c6447510SAarushi Mehta goto fail; 705c6447510SAarushi Mehta } 706c6447510SAarushi Mehta #endif /* !defined(CONFIG_LINUX_IO_URING) */ 707c6447510SAarushi Mehta 708c1bb86cdSEric Blake s->has_discard = true; 709c1bb86cdSEric Blake s->has_write_zeroes = true; 7105edc8557SKevin Wolf if ((bs->open_flags & BDRV_O_NOCACHE) != 0 && !dio_byte_aligned(s->fd)) { 711c1bb86cdSEric Blake s->needs_alignment = true; 712c1bb86cdSEric Blake } 713c1bb86cdSEric Blake 714c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 715c1bb86cdSEric Blake ret = -errno; 716c1bb86cdSEric Blake error_setg_errno(errp, errno, "Could not stat file"); 717c1bb86cdSEric Blake goto fail; 718c1bb86cdSEric Blake } 719230ff739SJohn Snow 720230ff739SJohn Snow if (!device) { 7218d17adf3SDaniel P. Berrangé if (!S_ISREG(st.st_mode)) { 7228d17adf3SDaniel P. Berrangé error_setg(errp, "'%s' driver requires '%s' to be a regular file", 7238d17adf3SDaniel P. Berrangé bs->drv->format_name, bs->filename); 724230ff739SJohn Snow ret = -EINVAL; 725230ff739SJohn Snow goto fail; 726230ff739SJohn Snow } else { 727c1bb86cdSEric Blake s->discard_zeroes = true; 728c1bb86cdSEric Blake s->has_fallocate = true; 729c1bb86cdSEric Blake } 730230ff739SJohn Snow } else { 731230ff739SJohn Snow if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 7328d17adf3SDaniel P. Berrangé error_setg(errp, "'%s' driver requires '%s' to be either " 7338d17adf3SDaniel P. Berrangé "a character or block device", 7348d17adf3SDaniel P. Berrangé bs->drv->format_name, bs->filename); 735230ff739SJohn Snow ret = -EINVAL; 736230ff739SJohn Snow goto fail; 737230ff739SJohn Snow } 738230ff739SJohn Snow } 739230ff739SJohn Snow 740c1bb86cdSEric Blake if (S_ISBLK(st.st_mode)) { 741c1bb86cdSEric Blake #ifdef BLKDISCARDZEROES 742c1bb86cdSEric Blake unsigned int arg; 743c1bb86cdSEric Blake if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { 744c1bb86cdSEric Blake s->discard_zeroes = true; 745c1bb86cdSEric Blake } 746c1bb86cdSEric Blake #endif 747c1bb86cdSEric Blake #ifdef __linux__ 748c1bb86cdSEric Blake /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do 749c1bb86cdSEric Blake * not rely on the contents of discarded blocks unless using O_DIRECT. 750c1bb86cdSEric Blake * Same for BLKZEROOUT. 751c1bb86cdSEric Blake */ 752c1bb86cdSEric Blake if (!(bs->open_flags & BDRV_O_NOCACHE)) { 753c1bb86cdSEric Blake s->discard_zeroes = false; 754c1bb86cdSEric Blake s->has_write_zeroes = false; 755c1bb86cdSEric Blake } 756c1bb86cdSEric Blake #endif 757c1bb86cdSEric Blake } 758c1bb86cdSEric Blake #ifdef __FreeBSD__ 759c1bb86cdSEric Blake if (S_ISCHR(st.st_mode)) { 760c1bb86cdSEric Blake /* 761c1bb86cdSEric Blake * The file is a char device (disk), which on FreeBSD isn't behind 762c1bb86cdSEric Blake * a pager, so force all requests to be aligned. This is needed 763c1bb86cdSEric Blake * so QEMU makes sure all IO operations on the device are aligned 764c1bb86cdSEric Blake * to sector size, or else FreeBSD will reject them with EINVAL. 765c1bb86cdSEric Blake */ 766c1bb86cdSEric Blake s->needs_alignment = true; 767c1bb86cdSEric Blake } 768c1bb86cdSEric Blake #endif 769c1bb86cdSEric Blake 770c1bb86cdSEric Blake #ifdef CONFIG_XFS 771c1bb86cdSEric Blake if (platform_test_xfs_fd(s->fd)) { 772c1bb86cdSEric Blake s->is_xfs = true; 773c1bb86cdSEric Blake } 774c1bb86cdSEric Blake #endif 775c1bb86cdSEric Blake 776738301e1SKevin Wolf bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; 7772f0c6e7aSKevin Wolf if (S_ISREG(st.st_mode)) { 7782f0c6e7aSKevin Wolf /* When extending regular files, we get zeros from the OS */ 7792f0c6e7aSKevin Wolf bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; 7802f0c6e7aSKevin Wolf } 781c1bb86cdSEric Blake ret = 0; 782c1bb86cdSEric Blake fail: 783a8c5cf27SKevin Wolf if (ret < 0 && s->fd != -1) { 784a8c5cf27SKevin Wolf qemu_close(s->fd); 785a8c5cf27SKevin Wolf } 786c1bb86cdSEric Blake if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { 787c1bb86cdSEric Blake unlink(filename); 788c1bb86cdSEric Blake } 789c1bb86cdSEric Blake qemu_opts_del(opts); 790c1bb86cdSEric Blake return ret; 791c1bb86cdSEric Blake } 792c1bb86cdSEric Blake 793c1bb86cdSEric Blake static int raw_open(BlockDriverState *bs, QDict *options, int flags, 794c1bb86cdSEric Blake Error **errp) 795c1bb86cdSEric Blake { 796c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 797c1bb86cdSEric Blake 798c1bb86cdSEric Blake s->type = FTYPE_FILE; 799230ff739SJohn Snow return raw_open_common(bs, options, flags, 0, false, errp); 800c1bb86cdSEric Blake } 801c1bb86cdSEric Blake 802244a5668SFam Zheng typedef enum { 803244a5668SFam Zheng RAW_PL_PREPARE, 804244a5668SFam Zheng RAW_PL_COMMIT, 805244a5668SFam Zheng RAW_PL_ABORT, 806244a5668SFam Zheng } RawPermLockOp; 807244a5668SFam Zheng 808244a5668SFam Zheng #define PERM_FOREACH(i) \ 809244a5668SFam Zheng for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) 810244a5668SFam Zheng 811244a5668SFam Zheng /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the 812244a5668SFam Zheng * file; if @unlock == true, also unlock the unneeded bytes. 813244a5668SFam Zheng * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. 814244a5668SFam Zheng */ 8152996ffadSFam Zheng static int raw_apply_lock_bytes(BDRVRawState *s, int fd, 816244a5668SFam Zheng uint64_t perm_lock_bits, 817244a5668SFam Zheng uint64_t shared_perm_lock_bits, 818244a5668SFam Zheng bool unlock, Error **errp) 819244a5668SFam Zheng { 820244a5668SFam Zheng int ret; 821244a5668SFam Zheng int i; 8222996ffadSFam Zheng uint64_t locked_perm, locked_shared_perm; 8232996ffadSFam Zheng 8242996ffadSFam Zheng if (s) { 8252996ffadSFam Zheng locked_perm = s->locked_perm; 8262996ffadSFam Zheng locked_shared_perm = s->locked_shared_perm; 8272996ffadSFam Zheng } else { 8282996ffadSFam Zheng /* 8292996ffadSFam Zheng * We don't have the previous bits, just lock/unlock for each of the 8302996ffadSFam Zheng * requested bits. 8312996ffadSFam Zheng */ 8322996ffadSFam Zheng if (unlock) { 8332996ffadSFam Zheng locked_perm = BLK_PERM_ALL; 8342996ffadSFam Zheng locked_shared_perm = BLK_PERM_ALL; 8352996ffadSFam Zheng } else { 8362996ffadSFam Zheng locked_perm = 0; 8372996ffadSFam Zheng locked_shared_perm = 0; 8382996ffadSFam Zheng } 8392996ffadSFam Zheng } 840244a5668SFam Zheng 841244a5668SFam Zheng PERM_FOREACH(i) { 842244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 8432996ffadSFam Zheng uint64_t bit = (1ULL << i); 8442996ffadSFam Zheng if ((perm_lock_bits & bit) && !(locked_perm & bit)) { 845d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 846244a5668SFam Zheng if (ret) { 847797e3e38SDavid Edmondson raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", 848797e3e38SDavid Edmondson off); 849244a5668SFam Zheng return ret; 8502996ffadSFam Zheng } else if (s) { 8512996ffadSFam Zheng s->locked_perm |= bit; 852244a5668SFam Zheng } 8532996ffadSFam Zheng } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { 854d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 855244a5668SFam Zheng if (ret) { 856797e3e38SDavid Edmondson error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); 857244a5668SFam Zheng return ret; 8582996ffadSFam Zheng } else if (s) { 8592996ffadSFam Zheng s->locked_perm &= ~bit; 860244a5668SFam Zheng } 861244a5668SFam Zheng } 862244a5668SFam Zheng } 863244a5668SFam Zheng PERM_FOREACH(i) { 864244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 8652996ffadSFam Zheng uint64_t bit = (1ULL << i); 8662996ffadSFam Zheng if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { 867d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 868244a5668SFam Zheng if (ret) { 869797e3e38SDavid Edmondson raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", 870797e3e38SDavid Edmondson off); 871244a5668SFam Zheng return ret; 8722996ffadSFam Zheng } else if (s) { 8732996ffadSFam Zheng s->locked_shared_perm |= bit; 874244a5668SFam Zheng } 8752996ffadSFam Zheng } else if (unlock && (locked_shared_perm & bit) && 8762996ffadSFam Zheng !(shared_perm_lock_bits & bit)) { 877d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 878244a5668SFam Zheng if (ret) { 879797e3e38SDavid Edmondson error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); 880244a5668SFam Zheng return ret; 8812996ffadSFam Zheng } else if (s) { 8822996ffadSFam Zheng s->locked_shared_perm &= ~bit; 883244a5668SFam Zheng } 884244a5668SFam Zheng } 885244a5668SFam Zheng } 886244a5668SFam Zheng return 0; 887244a5668SFam Zheng } 888244a5668SFam Zheng 889244a5668SFam Zheng /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ 890d0a96155SMax Reitz static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, 891244a5668SFam Zheng Error **errp) 892244a5668SFam Zheng { 893244a5668SFam Zheng int ret; 894244a5668SFam Zheng int i; 895244a5668SFam Zheng 896244a5668SFam Zheng PERM_FOREACH(i) { 897244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 898244a5668SFam Zheng uint64_t p = 1ULL << i; 899244a5668SFam Zheng if (perm & p) { 900d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 901244a5668SFam Zheng if (ret) { 902244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 903797e3e38SDavid Edmondson 904797e3e38SDavid Edmondson raw_lock_error_setg_errno(errp, -ret, 905244a5668SFam Zheng "Failed to get \"%s\" lock", 906244a5668SFam Zheng perm_name); 907244a5668SFam Zheng g_free(perm_name); 908244a5668SFam Zheng return ret; 909244a5668SFam Zheng } 910244a5668SFam Zheng } 911244a5668SFam Zheng } 912244a5668SFam Zheng PERM_FOREACH(i) { 913244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 914244a5668SFam Zheng uint64_t p = 1ULL << i; 915244a5668SFam Zheng if (!(shared_perm & p)) { 916d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 917244a5668SFam Zheng if (ret) { 918244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 919797e3e38SDavid Edmondson 920797e3e38SDavid Edmondson raw_lock_error_setg_errno(errp, -ret, 921244a5668SFam Zheng "Failed to get shared \"%s\" lock", 922244a5668SFam Zheng perm_name); 923244a5668SFam Zheng g_free(perm_name); 924244a5668SFam Zheng return ret; 925244a5668SFam Zheng } 926244a5668SFam Zheng } 927244a5668SFam Zheng } 928244a5668SFam Zheng return 0; 929244a5668SFam Zheng } 930244a5668SFam Zheng 931244a5668SFam Zheng static int raw_handle_perm_lock(BlockDriverState *bs, 932244a5668SFam Zheng RawPermLockOp op, 933244a5668SFam Zheng uint64_t new_perm, uint64_t new_shared, 934244a5668SFam Zheng Error **errp) 935244a5668SFam Zheng { 936244a5668SFam Zheng BDRVRawState *s = bs->opaque; 937244a5668SFam Zheng int ret = 0; 938244a5668SFam Zheng Error *local_err = NULL; 939244a5668SFam Zheng 940244a5668SFam Zheng if (!s->use_lock) { 941244a5668SFam Zheng return 0; 942244a5668SFam Zheng } 943244a5668SFam Zheng 944244a5668SFam Zheng if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { 945244a5668SFam Zheng return 0; 946244a5668SFam Zheng } 947244a5668SFam Zheng 948244a5668SFam Zheng switch (op) { 949244a5668SFam Zheng case RAW_PL_PREPARE: 950696aaaedSVladimir Sementsov-Ogievskiy if ((s->perm | new_perm) == s->perm && 951696aaaedSVladimir Sementsov-Ogievskiy (s->shared_perm & new_shared) == s->shared_perm) 952696aaaedSVladimir Sementsov-Ogievskiy { 953696aaaedSVladimir Sementsov-Ogievskiy /* 954696aaaedSVladimir Sementsov-Ogievskiy * We are going to unlock bytes, it should not fail. If it fail due 955696aaaedSVladimir Sementsov-Ogievskiy * to some fs-dependent permission-unrelated reasons (which occurs 956696aaaedSVladimir Sementsov-Ogievskiy * sometimes on NFS and leads to abort in bdrv_replace_child) we 957696aaaedSVladimir Sementsov-Ogievskiy * can't prevent such errors by any check here. And we ignore them 958696aaaedSVladimir Sementsov-Ogievskiy * anyway in ABORT and COMMIT. 959696aaaedSVladimir Sementsov-Ogievskiy */ 960696aaaedSVladimir Sementsov-Ogievskiy return 0; 961696aaaedSVladimir Sementsov-Ogievskiy } 962f2e3af29SFam Zheng ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, 963244a5668SFam Zheng ~s->shared_perm | ~new_shared, 964244a5668SFam Zheng false, errp); 965244a5668SFam Zheng if (!ret) { 966f2e3af29SFam Zheng ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); 967244a5668SFam Zheng if (!ret) { 968244a5668SFam Zheng return 0; 969244a5668SFam Zheng } 970b857431dSFam Zheng error_append_hint(errp, 971b857431dSFam Zheng "Is another process using the image [%s]?\n", 972b857431dSFam Zheng bs->filename); 973244a5668SFam Zheng } 974244a5668SFam Zheng /* fall through to unlock bytes. */ 975244a5668SFam Zheng case RAW_PL_ABORT: 976f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, 977d0a96155SMax Reitz true, &local_err); 978244a5668SFam Zheng if (local_err) { 979244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 980244a5668SFam Zheng * fail. Something weird happened, report it. 981244a5668SFam Zheng */ 982db0754dfSFam Zheng warn_report_err(local_err); 983244a5668SFam Zheng } 984244a5668SFam Zheng break; 985244a5668SFam Zheng case RAW_PL_COMMIT: 986f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, 987d0a96155SMax Reitz true, &local_err); 988244a5668SFam Zheng if (local_err) { 989244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 990244a5668SFam Zheng * fail. Something weird happened, report it. 991244a5668SFam Zheng */ 992db0754dfSFam Zheng warn_report_err(local_err); 993244a5668SFam Zheng } 994244a5668SFam Zheng break; 995244a5668SFam Zheng } 996244a5668SFam Zheng return ret; 997244a5668SFam Zheng } 998244a5668SFam Zheng 9995cec2870SKevin Wolf static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, 100023dece19SKevin Wolf int *open_flags, uint64_t perm, bool force_dup, 10016ceabe6fSKevin Wolf Error **errp) 10025cec2870SKevin Wolf { 10035cec2870SKevin Wolf BDRVRawState *s = bs->opaque; 10045cec2870SKevin Wolf int fd = -1; 10055cec2870SKevin Wolf int ret; 100623dece19SKevin Wolf bool has_writers = perm & 100723dece19SKevin Wolf (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); 10085cec2870SKevin Wolf int fcntl_flags = O_APPEND | O_NONBLOCK; 10095cec2870SKevin Wolf #ifdef O_NOATIME 10105cec2870SKevin Wolf fcntl_flags |= O_NOATIME; 10115cec2870SKevin Wolf #endif 10125cec2870SKevin Wolf 10135cec2870SKevin Wolf *open_flags = 0; 10145cec2870SKevin Wolf if (s->type == FTYPE_CD) { 10155cec2870SKevin Wolf *open_flags |= O_NONBLOCK; 10165cec2870SKevin Wolf } 10175cec2870SKevin Wolf 101823dece19SKevin Wolf raw_parse_flags(flags, open_flags, has_writers); 10195cec2870SKevin Wolf 10205cec2870SKevin Wolf #ifdef O_ASYNC 10215cec2870SKevin Wolf /* Not all operating systems have O_ASYNC, and those that don't 10225cec2870SKevin Wolf * will not let us track the state into rs->open_flags (typically 10235cec2870SKevin Wolf * you achieve the same effect with an ioctl, for example I_SETSIG 10245cec2870SKevin Wolf * on Solaris). But we do not use O_ASYNC, so that's fine. 10255cec2870SKevin Wolf */ 10265cec2870SKevin Wolf assert((s->open_flags & O_ASYNC) == 0); 10275cec2870SKevin Wolf #endif 10285cec2870SKevin Wolf 10296ceabe6fSKevin Wolf if (!force_dup && *open_flags == s->open_flags) { 10306ceabe6fSKevin Wolf /* We're lucky, the existing fd is fine */ 10316ceabe6fSKevin Wolf return s->fd; 10326ceabe6fSKevin Wolf } 10336ceabe6fSKevin Wolf 10345cec2870SKevin Wolf if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { 10355cec2870SKevin Wolf /* dup the original fd */ 10365cec2870SKevin Wolf fd = qemu_dup(s->fd); 10375cec2870SKevin Wolf if (fd >= 0) { 10385cec2870SKevin Wolf ret = fcntl_setfl(fd, *open_flags); 10395cec2870SKevin Wolf if (ret) { 10405cec2870SKevin Wolf qemu_close(fd); 10415cec2870SKevin Wolf fd = -1; 10425cec2870SKevin Wolf } 10435cec2870SKevin Wolf } 10445cec2870SKevin Wolf } 10455cec2870SKevin Wolf 1046b18a24a9SDaniel P. Berrangé /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ 10475cec2870SKevin Wolf if (fd == -1) { 10485cec2870SKevin Wolf const char *normalized_filename = bs->filename; 10495cec2870SKevin Wolf ret = raw_normalize_devicepath(&normalized_filename, errp); 10505cec2870SKevin Wolf if (ret >= 0) { 1051b18a24a9SDaniel P. Berrangé fd = qemu_open(normalized_filename, *open_flags, errp); 10525cec2870SKevin Wolf if (fd == -1) { 10535cec2870SKevin Wolf return -1; 10545cec2870SKevin Wolf } 10555cec2870SKevin Wolf } 10565cec2870SKevin Wolf } 10575cec2870SKevin Wolf 1058bca5283bSKevin Wolf if (fd != -1 && (*open_flags & O_RDWR)) { 1059bca5283bSKevin Wolf ret = check_hdev_writable(fd); 1060bca5283bSKevin Wolf if (ret < 0) { 1061bca5283bSKevin Wolf qemu_close(fd); 1062bca5283bSKevin Wolf error_setg_errno(errp, -ret, "The device is not writable"); 1063bca5283bSKevin Wolf return -1; 1064bca5283bSKevin Wolf } 1065bca5283bSKevin Wolf } 1066bca5283bSKevin Wolf 10675cec2870SKevin Wolf return fd; 10685cec2870SKevin Wolf } 10695cec2870SKevin Wolf 1070c1bb86cdSEric Blake static int raw_reopen_prepare(BDRVReopenState *state, 1071c1bb86cdSEric Blake BlockReopenQueue *queue, Error **errp) 1072c1bb86cdSEric Blake { 1073c1bb86cdSEric Blake BDRVRawState *s; 1074c1bb86cdSEric Blake BDRVRawReopenState *rs; 107531be8a2aSStefan Hajnoczi QemuOpts *opts; 1076a6aeca0cSKevin Wolf int ret; 1077c1bb86cdSEric Blake 1078c1bb86cdSEric Blake assert(state != NULL); 1079c1bb86cdSEric Blake assert(state->bs != NULL); 1080c1bb86cdSEric Blake 1081c1bb86cdSEric Blake s = state->bs->opaque; 1082c1bb86cdSEric Blake 1083c1bb86cdSEric Blake state->opaque = g_new0(BDRVRawReopenState, 1); 1084c1bb86cdSEric Blake rs = state->opaque; 108531be8a2aSStefan Hajnoczi 108631be8a2aSStefan Hajnoczi /* Handle options changes */ 108731be8a2aSStefan Hajnoczi opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 1088af175e85SMarkus Armbruster if (!qemu_opts_absorb_qdict(opts, state->options, errp)) { 108931be8a2aSStefan Hajnoczi ret = -EINVAL; 109031be8a2aSStefan Hajnoczi goto out; 109131be8a2aSStefan Hajnoczi } 109231be8a2aSStefan Hajnoczi 1093f357fcd8SStefan Hajnoczi rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); 10948d324575SAlberto Garcia rs->check_cache_dropped = 10958d324575SAlberto Garcia qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); 10968d324575SAlberto Garcia 10978d324575SAlberto Garcia /* This driver's reopen function doesn't currently allow changing 10988d324575SAlberto Garcia * other options, so let's put them back in the original QDict and 10998d324575SAlberto Garcia * bdrv_reopen_prepare() will detect changes and complain. */ 11008d324575SAlberto Garcia qemu_opts_to_qdict(opts, state->options); 1101c1bb86cdSEric Blake 1102*72373e40SVladimir Sementsov-Ogievskiy /* 1103*72373e40SVladimir Sementsov-Ogievskiy * As part of reopen prepare we also want to create new fd by 1104*72373e40SVladimir Sementsov-Ogievskiy * raw_reconfigure_getfd(). But it wants updated "perm", when in 1105*72373e40SVladimir Sementsov-Ogievskiy * bdrv_reopen_multiple() .bdrv_reopen_prepare() callback called prior to 1106*72373e40SVladimir Sementsov-Ogievskiy * permission update. Happily, permission update is always a part (a seprate 1107*72373e40SVladimir Sementsov-Ogievskiy * stage) of bdrv_reopen_multiple() so we can rely on this fact and 1108*72373e40SVladimir Sementsov-Ogievskiy * reconfigure fd in raw_check_perm(). 1109*72373e40SVladimir Sementsov-Ogievskiy */ 1110c1bb86cdSEric Blake 1111e0c9cf3aSKevin Wolf s->reopen_state = state; 1112a6aeca0cSKevin Wolf ret = 0; 1113*72373e40SVladimir Sementsov-Ogievskiy 111431be8a2aSStefan Hajnoczi out: 111531be8a2aSStefan Hajnoczi qemu_opts_del(opts); 1116c1bb86cdSEric Blake return ret; 1117c1bb86cdSEric Blake } 1118c1bb86cdSEric Blake 1119c1bb86cdSEric Blake static void raw_reopen_commit(BDRVReopenState *state) 1120c1bb86cdSEric Blake { 1121c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1122c1bb86cdSEric Blake BDRVRawState *s = state->bs->opaque; 1123c1bb86cdSEric Blake 1124f357fcd8SStefan Hajnoczi s->drop_cache = rs->drop_cache; 112531be8a2aSStefan Hajnoczi s->check_cache_dropped = rs->check_cache_dropped; 1126c1bb86cdSEric Blake s->open_flags = rs->open_flags; 1127c1bb86cdSEric Blake g_free(state->opaque); 1128c1bb86cdSEric Blake state->opaque = NULL; 1129e0c9cf3aSKevin Wolf 1130e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1131e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1132c1bb86cdSEric Blake } 1133c1bb86cdSEric Blake 1134c1bb86cdSEric Blake 1135c1bb86cdSEric Blake static void raw_reopen_abort(BDRVReopenState *state) 1136c1bb86cdSEric Blake { 1137c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1138e0c9cf3aSKevin Wolf BDRVRawState *s = state->bs->opaque; 1139c1bb86cdSEric Blake 1140c1bb86cdSEric Blake /* nothing to do if NULL, we didn't get far enough */ 1141c1bb86cdSEric Blake if (rs == NULL) { 1142c1bb86cdSEric Blake return; 1143c1bb86cdSEric Blake } 1144c1bb86cdSEric Blake 1145c1bb86cdSEric Blake g_free(state->opaque); 1146c1bb86cdSEric Blake state->opaque = NULL; 1147e0c9cf3aSKevin Wolf 1148e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1149e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1150c1bb86cdSEric Blake } 1151c1bb86cdSEric Blake 1152867eccfeSMaxim Levitsky static int sg_get_max_transfer_length(int fd) 1153c1bb86cdSEric Blake { 1154c1bb86cdSEric Blake #ifdef BLKSECTGET 115548265250SEric Farman int max_bytes = 0; 1156867eccfeSMaxim Levitsky 1157867eccfeSMaxim Levitsky if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) { 115848265250SEric Farman return max_bytes; 1159c1bb86cdSEric Blake } else { 1160c1bb86cdSEric Blake return -errno; 1161c1bb86cdSEric Blake } 1162c1bb86cdSEric Blake #else 1163c1bb86cdSEric Blake return -ENOSYS; 1164c1bb86cdSEric Blake #endif 1165c1bb86cdSEric Blake } 1166c1bb86cdSEric Blake 1167867eccfeSMaxim Levitsky static int sg_get_max_segments(int fd) 11689103f1ceSFam Zheng { 11699103f1ceSFam Zheng #ifdef CONFIG_LINUX 11709103f1ceSFam Zheng char buf[32]; 11719103f1ceSFam Zheng const char *end; 1172867eccfeSMaxim Levitsky char *sysfspath = NULL; 11739103f1ceSFam Zheng int ret; 1174867eccfeSMaxim Levitsky int sysfd = -1; 11759103f1ceSFam Zheng long max_segments; 1176867eccfeSMaxim Levitsky struct stat st; 1177867eccfeSMaxim Levitsky 1178867eccfeSMaxim Levitsky if (fstat(fd, &st)) { 1179867eccfeSMaxim Levitsky ret = -errno; 1180867eccfeSMaxim Levitsky goto out; 1181867eccfeSMaxim Levitsky } 11829103f1ceSFam Zheng 11839103f1ceSFam Zheng sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", 1184867eccfeSMaxim Levitsky major(st.st_rdev), minor(st.st_rdev)); 1185867eccfeSMaxim Levitsky sysfd = open(sysfspath, O_RDONLY); 1186867eccfeSMaxim Levitsky if (sysfd == -1) { 11879103f1ceSFam Zheng ret = -errno; 11889103f1ceSFam Zheng goto out; 11899103f1ceSFam Zheng } 11909103f1ceSFam Zheng do { 1191867eccfeSMaxim Levitsky ret = read(sysfd, buf, sizeof(buf) - 1); 11929103f1ceSFam Zheng } while (ret == -1 && errno == EINTR); 11939103f1ceSFam Zheng if (ret < 0) { 11949103f1ceSFam Zheng ret = -errno; 11959103f1ceSFam Zheng goto out; 11969103f1ceSFam Zheng } else if (ret == 0) { 11979103f1ceSFam Zheng ret = -EIO; 11989103f1ceSFam Zheng goto out; 11999103f1ceSFam Zheng } 12009103f1ceSFam Zheng buf[ret] = 0; 12019103f1ceSFam Zheng /* The file is ended with '\n', pass 'end' to accept that. */ 12029103f1ceSFam Zheng ret = qemu_strtol(buf, &end, 10, &max_segments); 12039103f1ceSFam Zheng if (ret == 0 && end && *end == '\n') { 12049103f1ceSFam Zheng ret = max_segments; 12059103f1ceSFam Zheng } 12069103f1ceSFam Zheng 12079103f1ceSFam Zheng out: 1208867eccfeSMaxim Levitsky if (sysfd != -1) { 1209867eccfeSMaxim Levitsky close(sysfd); 1210fed414dfSFam Zheng } 12119103f1ceSFam Zheng g_free(sysfspath); 12129103f1ceSFam Zheng return ret; 12139103f1ceSFam Zheng #else 12149103f1ceSFam Zheng return -ENOTSUP; 12159103f1ceSFam Zheng #endif 12169103f1ceSFam Zheng } 12179103f1ceSFam Zheng 1218c1bb86cdSEric Blake static void raw_refresh_limits(BlockDriverState *bs, Error **errp) 1219c1bb86cdSEric Blake { 1220c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1221c1bb86cdSEric Blake 1222867eccfeSMaxim Levitsky if (bs->sg) { 1223867eccfeSMaxim Levitsky int ret = sg_get_max_transfer_length(s->fd); 1224867eccfeSMaxim Levitsky 122548265250SEric Farman if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { 122648265250SEric Farman bs->bl.max_transfer = pow2floor(ret); 1227c1bb86cdSEric Blake } 1228867eccfeSMaxim Levitsky 1229867eccfeSMaxim Levitsky ret = sg_get_max_segments(s->fd); 12309103f1ceSFam Zheng if (ret > 0) { 1231038adc2fSWei Yang bs->bl.max_transfer = MIN(bs->bl.max_transfer, 1232038adc2fSWei Yang ret * qemu_real_host_page_size); 1233c1bb86cdSEric Blake } 1234c1bb86cdSEric Blake } 1235c1bb86cdSEric Blake 1236c1bb86cdSEric Blake raw_probe_alignment(bs, s->fd, errp); 1237c1bb86cdSEric Blake bs->bl.min_mem_alignment = s->buf_align; 1238038adc2fSWei Yang bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size); 1239c1bb86cdSEric Blake } 1240c1bb86cdSEric Blake 1241c1bb86cdSEric Blake static int check_for_dasd(int fd) 1242c1bb86cdSEric Blake { 1243c1bb86cdSEric Blake #ifdef BIODASDINFO2 1244c1bb86cdSEric Blake struct dasd_information2_t info = {0}; 1245c1bb86cdSEric Blake 1246c1bb86cdSEric Blake return ioctl(fd, BIODASDINFO2, &info); 1247c1bb86cdSEric Blake #else 1248c1bb86cdSEric Blake return -1; 1249c1bb86cdSEric Blake #endif 1250c1bb86cdSEric Blake } 1251c1bb86cdSEric Blake 1252c1bb86cdSEric Blake /** 1253c1bb86cdSEric Blake * Try to get @bs's logical and physical block size. 1254c1bb86cdSEric Blake * On success, store them in @bsz and return zero. 1255c1bb86cdSEric Blake * On failure, return negative errno. 1256c1bb86cdSEric Blake */ 1257c1bb86cdSEric Blake static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) 1258c1bb86cdSEric Blake { 1259c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1260c1bb86cdSEric Blake int ret; 1261c1bb86cdSEric Blake 1262c1bb86cdSEric Blake /* If DASD, get blocksizes */ 1263c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1264c1bb86cdSEric Blake return -ENOTSUP; 1265c1bb86cdSEric Blake } 1266c1bb86cdSEric Blake ret = probe_logical_blocksize(s->fd, &bsz->log); 1267c1bb86cdSEric Blake if (ret < 0) { 1268c1bb86cdSEric Blake return ret; 1269c1bb86cdSEric Blake } 1270c1bb86cdSEric Blake return probe_physical_blocksize(s->fd, &bsz->phys); 1271c1bb86cdSEric Blake } 1272c1bb86cdSEric Blake 1273c1bb86cdSEric Blake /** 1274c1bb86cdSEric Blake * Try to get @bs's geometry: cyls, heads, sectors. 1275c1bb86cdSEric Blake * On success, store them in @geo and return 0. 1276c1bb86cdSEric Blake * On failure return -errno. 1277c1bb86cdSEric Blake * (Allows block driver to assign default geometry values that guest sees) 1278c1bb86cdSEric Blake */ 1279c1bb86cdSEric Blake #ifdef __linux__ 1280c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1281c1bb86cdSEric Blake { 1282c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1283c1bb86cdSEric Blake struct hd_geometry ioctl_geo = {0}; 1284c1bb86cdSEric Blake 1285c1bb86cdSEric Blake /* If DASD, get its geometry */ 1286c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1287c1bb86cdSEric Blake return -ENOTSUP; 1288c1bb86cdSEric Blake } 1289c1bb86cdSEric Blake if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { 1290c1bb86cdSEric Blake return -errno; 1291c1bb86cdSEric Blake } 1292c1bb86cdSEric Blake /* HDIO_GETGEO may return success even though geo contains zeros 1293c1bb86cdSEric Blake (e.g. certain multipath setups) */ 1294c1bb86cdSEric Blake if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { 1295c1bb86cdSEric Blake return -ENOTSUP; 1296c1bb86cdSEric Blake } 1297c1bb86cdSEric Blake /* Do not return a geometry for partition */ 1298c1bb86cdSEric Blake if (ioctl_geo.start != 0) { 1299c1bb86cdSEric Blake return -ENOTSUP; 1300c1bb86cdSEric Blake } 1301c1bb86cdSEric Blake geo->heads = ioctl_geo.heads; 1302c1bb86cdSEric Blake geo->sectors = ioctl_geo.sectors; 1303c1bb86cdSEric Blake geo->cylinders = ioctl_geo.cylinders; 1304c1bb86cdSEric Blake 1305c1bb86cdSEric Blake return 0; 1306c1bb86cdSEric Blake } 1307c1bb86cdSEric Blake #else /* __linux__ */ 1308c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1309c1bb86cdSEric Blake { 1310c1bb86cdSEric Blake return -ENOTSUP; 1311c1bb86cdSEric Blake } 1312c1bb86cdSEric Blake #endif 1313c1bb86cdSEric Blake 131403425671SKevin Wolf #if defined(__linux__) 131503425671SKevin Wolf static int handle_aiocb_ioctl(void *opaque) 1316c1bb86cdSEric Blake { 131703425671SKevin Wolf RawPosixAIOData *aiocb = opaque; 1318c1bb86cdSEric Blake int ret; 1319c1bb86cdSEric Blake 1320d57c44d0SKevin Wolf ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); 1321c1bb86cdSEric Blake if (ret == -1) { 1322c1bb86cdSEric Blake return -errno; 1323c1bb86cdSEric Blake } 1324c1bb86cdSEric Blake 1325c1bb86cdSEric Blake return 0; 1326c1bb86cdSEric Blake } 132703425671SKevin Wolf #endif /* linux */ 1328c1bb86cdSEric Blake 132906dc9bd5SKevin Wolf static int handle_aiocb_flush(void *opaque) 1330c1bb86cdSEric Blake { 133106dc9bd5SKevin Wolf RawPosixAIOData *aiocb = opaque; 1332e5bcf967SKevin Wolf BDRVRawState *s = aiocb->bs->opaque; 1333c1bb86cdSEric Blake int ret; 1334c1bb86cdSEric Blake 1335e5bcf967SKevin Wolf if (s->page_cache_inconsistent) { 1336e5bcf967SKevin Wolf return -EIO; 1337e5bcf967SKevin Wolf } 1338e5bcf967SKevin Wolf 1339c1bb86cdSEric Blake ret = qemu_fdatasync(aiocb->aio_fildes); 1340c1bb86cdSEric Blake if (ret == -1) { 1341e5bcf967SKevin Wolf /* There is no clear definition of the semantics of a failing fsync(), 1342e5bcf967SKevin Wolf * so we may have to assume the worst. The sad truth is that this 1343e5bcf967SKevin Wolf * assumption is correct for Linux. Some pages are now probably marked 1344e5bcf967SKevin Wolf * clean in the page cache even though they are inconsistent with the 1345e5bcf967SKevin Wolf * on-disk contents. The next fdatasync() call would succeed, but no 1346e5bcf967SKevin Wolf * further writeback attempt will be made. We can't get back to a state 1347e5bcf967SKevin Wolf * in which we know what is on disk (we would have to rewrite 1348e5bcf967SKevin Wolf * everything that was touched since the last fdatasync() at least), so 1349e5bcf967SKevin Wolf * make bdrv_flush() fail permanently. Given that the behaviour isn't 1350e5bcf967SKevin Wolf * really defined, I have little hope that other OSes are doing better. 1351e5bcf967SKevin Wolf * 1352e5bcf967SKevin Wolf * Obviously, this doesn't affect O_DIRECT, which bypasses the page 1353e5bcf967SKevin Wolf * cache. */ 1354e5bcf967SKevin Wolf if ((s->open_flags & O_DIRECT) == 0) { 1355e5bcf967SKevin Wolf s->page_cache_inconsistent = true; 1356e5bcf967SKevin Wolf } 1357c1bb86cdSEric Blake return -errno; 1358c1bb86cdSEric Blake } 1359c1bb86cdSEric Blake return 0; 1360c1bb86cdSEric Blake } 1361c1bb86cdSEric Blake 1362c1bb86cdSEric Blake #ifdef CONFIG_PREADV 1363c1bb86cdSEric Blake 1364c1bb86cdSEric Blake static bool preadv_present = true; 1365c1bb86cdSEric Blake 1366c1bb86cdSEric Blake static ssize_t 1367c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1368c1bb86cdSEric Blake { 1369c1bb86cdSEric Blake return preadv(fd, iov, nr_iov, offset); 1370c1bb86cdSEric Blake } 1371c1bb86cdSEric Blake 1372c1bb86cdSEric Blake static ssize_t 1373c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1374c1bb86cdSEric Blake { 1375c1bb86cdSEric Blake return pwritev(fd, iov, nr_iov, offset); 1376c1bb86cdSEric Blake } 1377c1bb86cdSEric Blake 1378c1bb86cdSEric Blake #else 1379c1bb86cdSEric Blake 1380c1bb86cdSEric Blake static bool preadv_present = false; 1381c1bb86cdSEric Blake 1382c1bb86cdSEric Blake static ssize_t 1383c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1384c1bb86cdSEric Blake { 1385c1bb86cdSEric Blake return -ENOSYS; 1386c1bb86cdSEric Blake } 1387c1bb86cdSEric Blake 1388c1bb86cdSEric Blake static ssize_t 1389c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1390c1bb86cdSEric Blake { 1391c1bb86cdSEric Blake return -ENOSYS; 1392c1bb86cdSEric Blake } 1393c1bb86cdSEric Blake 1394c1bb86cdSEric Blake #endif 1395c1bb86cdSEric Blake 1396c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) 1397c1bb86cdSEric Blake { 1398c1bb86cdSEric Blake ssize_t len; 1399c1bb86cdSEric Blake 1400c1bb86cdSEric Blake do { 1401c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) 1402c1bb86cdSEric Blake len = qemu_pwritev(aiocb->aio_fildes, 1403d57c44d0SKevin Wolf aiocb->io.iov, 1404d57c44d0SKevin Wolf aiocb->io.niov, 1405c1bb86cdSEric Blake aiocb->aio_offset); 1406c1bb86cdSEric Blake else 1407c1bb86cdSEric Blake len = qemu_preadv(aiocb->aio_fildes, 1408d57c44d0SKevin Wolf aiocb->io.iov, 1409d57c44d0SKevin Wolf aiocb->io.niov, 1410c1bb86cdSEric Blake aiocb->aio_offset); 1411c1bb86cdSEric Blake } while (len == -1 && errno == EINTR); 1412c1bb86cdSEric Blake 1413c1bb86cdSEric Blake if (len == -1) { 1414c1bb86cdSEric Blake return -errno; 1415c1bb86cdSEric Blake } 1416c1bb86cdSEric Blake return len; 1417c1bb86cdSEric Blake } 1418c1bb86cdSEric Blake 1419c1bb86cdSEric Blake /* 1420c1bb86cdSEric Blake * Read/writes the data to/from a given linear buffer. 1421c1bb86cdSEric Blake * 1422c1bb86cdSEric Blake * Returns the number of bytes handles or -errno in case of an error. Short 1423c1bb86cdSEric Blake * reads are only returned if the end of the file is reached. 1424c1bb86cdSEric Blake */ 1425c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) 1426c1bb86cdSEric Blake { 1427c1bb86cdSEric Blake ssize_t offset = 0; 1428c1bb86cdSEric Blake ssize_t len; 1429c1bb86cdSEric Blake 1430c1bb86cdSEric Blake while (offset < aiocb->aio_nbytes) { 1431c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1432c1bb86cdSEric Blake len = pwrite(aiocb->aio_fildes, 1433c1bb86cdSEric Blake (const char *)buf + offset, 1434c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1435c1bb86cdSEric Blake aiocb->aio_offset + offset); 1436c1bb86cdSEric Blake } else { 1437c1bb86cdSEric Blake len = pread(aiocb->aio_fildes, 1438c1bb86cdSEric Blake buf + offset, 1439c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1440c1bb86cdSEric Blake aiocb->aio_offset + offset); 1441c1bb86cdSEric Blake } 1442c1bb86cdSEric Blake if (len == -1 && errno == EINTR) { 1443c1bb86cdSEric Blake continue; 1444c1bb86cdSEric Blake } else if (len == -1 && errno == EINVAL && 1445c1bb86cdSEric Blake (aiocb->bs->open_flags & BDRV_O_NOCACHE) && 1446c1bb86cdSEric Blake !(aiocb->aio_type & QEMU_AIO_WRITE) && 1447c1bb86cdSEric Blake offset > 0) { 1448c1bb86cdSEric Blake /* O_DIRECT pread() may fail with EINVAL when offset is unaligned 1449c1bb86cdSEric Blake * after a short read. Assume that O_DIRECT short reads only occur 1450c1bb86cdSEric Blake * at EOF. Therefore this is a short read, not an I/O error. 1451c1bb86cdSEric Blake */ 1452c1bb86cdSEric Blake break; 1453c1bb86cdSEric Blake } else if (len == -1) { 1454c1bb86cdSEric Blake offset = -errno; 1455c1bb86cdSEric Blake break; 1456c1bb86cdSEric Blake } else if (len == 0) { 1457c1bb86cdSEric Blake break; 1458c1bb86cdSEric Blake } 1459c1bb86cdSEric Blake offset += len; 1460c1bb86cdSEric Blake } 1461c1bb86cdSEric Blake 1462c1bb86cdSEric Blake return offset; 1463c1bb86cdSEric Blake } 1464c1bb86cdSEric Blake 1465999e6b69SKevin Wolf static int handle_aiocb_rw(void *opaque) 1466c1bb86cdSEric Blake { 1467999e6b69SKevin Wolf RawPosixAIOData *aiocb = opaque; 1468c1bb86cdSEric Blake ssize_t nbytes; 1469c1bb86cdSEric Blake char *buf; 1470c1bb86cdSEric Blake 1471c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { 1472c1bb86cdSEric Blake /* 1473c1bb86cdSEric Blake * If there is just a single buffer, and it is properly aligned 1474c1bb86cdSEric Blake * we can just use plain pread/pwrite without any problems. 1475c1bb86cdSEric Blake */ 1476d57c44d0SKevin Wolf if (aiocb->io.niov == 1) { 147754c7ca1bSKevin Wolf nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); 147854c7ca1bSKevin Wolf goto out; 1479c1bb86cdSEric Blake } 1480c1bb86cdSEric Blake /* 1481c1bb86cdSEric Blake * We have more than one iovec, and all are properly aligned. 1482c1bb86cdSEric Blake * 1483c1bb86cdSEric Blake * Try preadv/pwritev first and fall back to linearizing the 1484c1bb86cdSEric Blake * buffer if it's not supported. 1485c1bb86cdSEric Blake */ 1486c1bb86cdSEric Blake if (preadv_present) { 1487c1bb86cdSEric Blake nbytes = handle_aiocb_rw_vector(aiocb); 1488c1bb86cdSEric Blake if (nbytes == aiocb->aio_nbytes || 1489c1bb86cdSEric Blake (nbytes < 0 && nbytes != -ENOSYS)) { 149054c7ca1bSKevin Wolf goto out; 1491c1bb86cdSEric Blake } 1492c1bb86cdSEric Blake preadv_present = false; 1493c1bb86cdSEric Blake } 1494c1bb86cdSEric Blake 1495c1bb86cdSEric Blake /* 1496c1bb86cdSEric Blake * XXX(hch): short read/write. no easy way to handle the reminder 1497c1bb86cdSEric Blake * using these interfaces. For now retry using plain 1498c1bb86cdSEric Blake * pread/pwrite? 1499c1bb86cdSEric Blake */ 1500c1bb86cdSEric Blake } 1501c1bb86cdSEric Blake 1502c1bb86cdSEric Blake /* 1503c1bb86cdSEric Blake * Ok, we have to do it the hard way, copy all segments into 1504c1bb86cdSEric Blake * a single aligned buffer. 1505c1bb86cdSEric Blake */ 1506c1bb86cdSEric Blake buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); 1507c1bb86cdSEric Blake if (buf == NULL) { 150854c7ca1bSKevin Wolf nbytes = -ENOMEM; 150954c7ca1bSKevin Wolf goto out; 1510c1bb86cdSEric Blake } 1511c1bb86cdSEric Blake 1512c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1513c1bb86cdSEric Blake char *p = buf; 1514c1bb86cdSEric Blake int i; 1515c1bb86cdSEric Blake 1516d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov; ++i) { 1517d57c44d0SKevin Wolf memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); 1518d57c44d0SKevin Wolf p += aiocb->io.iov[i].iov_len; 1519c1bb86cdSEric Blake } 1520c1bb86cdSEric Blake assert(p - buf == aiocb->aio_nbytes); 1521c1bb86cdSEric Blake } 1522c1bb86cdSEric Blake 1523c1bb86cdSEric Blake nbytes = handle_aiocb_rw_linear(aiocb, buf); 1524c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { 1525c1bb86cdSEric Blake char *p = buf; 1526c1bb86cdSEric Blake size_t count = aiocb->aio_nbytes, copy; 1527c1bb86cdSEric Blake int i; 1528c1bb86cdSEric Blake 1529d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov && count; ++i) { 1530c1bb86cdSEric Blake copy = count; 1531d57c44d0SKevin Wolf if (copy > aiocb->io.iov[i].iov_len) { 1532d57c44d0SKevin Wolf copy = aiocb->io.iov[i].iov_len; 1533c1bb86cdSEric Blake } 1534d57c44d0SKevin Wolf memcpy(aiocb->io.iov[i].iov_base, p, copy); 1535c1bb86cdSEric Blake assert(count >= copy); 1536c1bb86cdSEric Blake p += copy; 1537c1bb86cdSEric Blake count -= copy; 1538c1bb86cdSEric Blake } 1539c1bb86cdSEric Blake assert(count == 0); 1540c1bb86cdSEric Blake } 1541c1bb86cdSEric Blake qemu_vfree(buf); 1542c1bb86cdSEric Blake 154354c7ca1bSKevin Wolf out: 154454c7ca1bSKevin Wolf if (nbytes == aiocb->aio_nbytes) { 154554c7ca1bSKevin Wolf return 0; 154654c7ca1bSKevin Wolf } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { 154754c7ca1bSKevin Wolf if (aiocb->aio_type & QEMU_AIO_WRITE) { 154854c7ca1bSKevin Wolf return -EINVAL; 154954c7ca1bSKevin Wolf } else { 155054c7ca1bSKevin Wolf iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, 155154c7ca1bSKevin Wolf 0, aiocb->aio_nbytes - nbytes); 155254c7ca1bSKevin Wolf return 0; 155354c7ca1bSKevin Wolf } 155454c7ca1bSKevin Wolf } else { 155554c7ca1bSKevin Wolf assert(nbytes < 0); 1556c1bb86cdSEric Blake return nbytes; 1557c1bb86cdSEric Blake } 155854c7ca1bSKevin Wolf } 1559c1bb86cdSEric Blake 1560c1bb86cdSEric Blake static int translate_err(int err) 1561c1bb86cdSEric Blake { 1562c1bb86cdSEric Blake if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || 1563c1bb86cdSEric Blake err == -ENOTTY) { 1564c1bb86cdSEric Blake err = -ENOTSUP; 1565c1bb86cdSEric Blake } 1566c1bb86cdSEric Blake return err; 1567c1bb86cdSEric Blake } 1568c1bb86cdSEric Blake 1569c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 1570c1bb86cdSEric Blake static int do_fallocate(int fd, int mode, off_t offset, off_t len) 1571c1bb86cdSEric Blake { 1572c1bb86cdSEric Blake do { 1573c1bb86cdSEric Blake if (fallocate(fd, mode, offset, len) == 0) { 1574c1bb86cdSEric Blake return 0; 1575c1bb86cdSEric Blake } 1576c1bb86cdSEric Blake } while (errno == EINTR); 1577c1bb86cdSEric Blake return translate_err(-errno); 1578c1bb86cdSEric Blake } 1579c1bb86cdSEric Blake #endif 1580c1bb86cdSEric Blake 1581c1bb86cdSEric Blake static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) 1582c1bb86cdSEric Blake { 1583c1bb86cdSEric Blake int ret = -ENOTSUP; 1584c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1585c1bb86cdSEric Blake 1586c1bb86cdSEric Blake if (!s->has_write_zeroes) { 1587c1bb86cdSEric Blake return -ENOTSUP; 1588c1bb86cdSEric Blake } 1589c1bb86cdSEric Blake 1590c1bb86cdSEric Blake #ifdef BLKZEROOUT 1591738301e1SKevin Wolf /* The BLKZEROOUT implementation in the kernel doesn't set 1592738301e1SKevin Wolf * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow 1593738301e1SKevin Wolf * fallbacks. */ 1594738301e1SKevin Wolf if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { 1595c1bb86cdSEric Blake do { 1596c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1597c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { 1598c1bb86cdSEric Blake return 0; 1599c1bb86cdSEric Blake } 1600c1bb86cdSEric Blake } while (errno == EINTR); 1601c1bb86cdSEric Blake 1602c1bb86cdSEric Blake ret = translate_err(-errno); 1603c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1604c1bb86cdSEric Blake s->has_write_zeroes = false; 1605c1bb86cdSEric Blake } 1606effecce6SKevin Wolf } 1607effecce6SKevin Wolf #endif 1608effecce6SKevin Wolf 1609c1bb86cdSEric Blake return ret; 1610c1bb86cdSEric Blake } 1611c1bb86cdSEric Blake 16127154d8aeSKevin Wolf static int handle_aiocb_write_zeroes(void *opaque) 1613c1bb86cdSEric Blake { 16147154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 161570d9110bSDenis V. Lunev #ifdef CONFIG_FALLOCATE 1616b2c6f23fSMax Reitz BDRVRawState *s = aiocb->bs->opaque; 161770d9110bSDenis V. Lunev int64_t len; 161870d9110bSDenis V. Lunev #endif 1619c1bb86cdSEric Blake 1620c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1621c1bb86cdSEric Blake return handle_aiocb_write_zeroes_block(aiocb); 1622c1bb86cdSEric Blake } 1623c1bb86cdSEric Blake 1624c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_ZERO_RANGE 1625c1bb86cdSEric Blake if (s->has_write_zeroes) { 1626c1bb86cdSEric Blake int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, 1627c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1628294682ccSAndrey Shinkevich if (ret == -EINVAL) { 1629294682ccSAndrey Shinkevich /* 1630294682ccSAndrey Shinkevich * Allow falling back to pwrite for file systems that 1631294682ccSAndrey Shinkevich * do not support fallocate() for an unaligned byte range. 1632294682ccSAndrey Shinkevich */ 1633294682ccSAndrey Shinkevich return -ENOTSUP; 1634294682ccSAndrey Shinkevich } 1635c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1636c1bb86cdSEric Blake return ret; 1637c1bb86cdSEric Blake } 1638c1bb86cdSEric Blake s->has_write_zeroes = false; 1639c1bb86cdSEric Blake } 1640c1bb86cdSEric Blake #endif 1641c1bb86cdSEric Blake 1642c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1643c1bb86cdSEric Blake if (s->has_discard && s->has_fallocate) { 1644c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 1645c1bb86cdSEric Blake FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1646c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1647c1bb86cdSEric Blake if (ret == 0) { 1648c1bb86cdSEric Blake ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1649c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1650c1bb86cdSEric Blake return ret; 1651c1bb86cdSEric Blake } 1652c1bb86cdSEric Blake s->has_fallocate = false; 1653c1bb86cdSEric Blake } else if (ret != -ENOTSUP) { 1654c1bb86cdSEric Blake return ret; 1655c1bb86cdSEric Blake } else { 1656c1bb86cdSEric Blake s->has_discard = false; 1657c1bb86cdSEric Blake } 1658c1bb86cdSEric Blake } 1659c1bb86cdSEric Blake #endif 1660c1bb86cdSEric Blake 1661c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 166270d9110bSDenis V. Lunev /* Last resort: we are trying to extend the file with zeroed data. This 166370d9110bSDenis V. Lunev * can be done via fallocate(fd, 0) */ 166470d9110bSDenis V. Lunev len = bdrv_getlength(aiocb->bs); 166570d9110bSDenis V. Lunev if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { 1666c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1667c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1668c1bb86cdSEric Blake return ret; 1669c1bb86cdSEric Blake } 1670c1bb86cdSEric Blake s->has_fallocate = false; 1671c1bb86cdSEric Blake } 1672c1bb86cdSEric Blake #endif 1673c1bb86cdSEric Blake 1674c1bb86cdSEric Blake return -ENOTSUP; 1675c1bb86cdSEric Blake } 1676c1bb86cdSEric Blake 16777154d8aeSKevin Wolf static int handle_aiocb_write_zeroes_unmap(void *opaque) 167834fa110eSKevin Wolf { 16797154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 168034fa110eSKevin Wolf BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; 168134fa110eSKevin Wolf 168234fa110eSKevin Wolf /* First try to write zeros and unmap at the same time */ 168334fa110eSKevin Wolf 168434fa110eSKevin Wolf #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1685b3ac2b94SSimran Singhal int ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 168634fa110eSKevin Wolf aiocb->aio_offset, aiocb->aio_nbytes); 1687bae127d4SAntoine Damhet switch (ret) { 1688bae127d4SAntoine Damhet case -ENOTSUP: 1689bae127d4SAntoine Damhet case -EINVAL: 1690ece4fa91SMaxim Levitsky case -EBUSY: 1691bae127d4SAntoine Damhet break; 1692bae127d4SAntoine Damhet default: 169334fa110eSKevin Wolf return ret; 169434fa110eSKevin Wolf } 169534fa110eSKevin Wolf #endif 169634fa110eSKevin Wolf 169734fa110eSKevin Wolf /* If we couldn't manage to unmap while guaranteed that the area reads as 169834fa110eSKevin Wolf * all-zero afterwards, just write zeroes without unmapping */ 1699b3ac2b94SSimran Singhal return handle_aiocb_write_zeroes(aiocb); 170034fa110eSKevin Wolf } 170134fa110eSKevin Wolf 17021efad060SFam Zheng #ifndef HAVE_COPY_FILE_RANGE 17031efad060SFam Zheng static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, 17041efad060SFam Zheng off_t *out_off, size_t len, unsigned int flags) 17051efad060SFam Zheng { 17061efad060SFam Zheng #ifdef __NR_copy_file_range 17071efad060SFam Zheng return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, 17081efad060SFam Zheng out_off, len, flags); 17091efad060SFam Zheng #else 17101efad060SFam Zheng errno = ENOSYS; 17111efad060SFam Zheng return -1; 17121efad060SFam Zheng #endif 17131efad060SFam Zheng } 17141efad060SFam Zheng #endif 17151efad060SFam Zheng 171658a209c4SKevin Wolf static int handle_aiocb_copy_range(void *opaque) 17171efad060SFam Zheng { 171858a209c4SKevin Wolf RawPosixAIOData *aiocb = opaque; 17191efad060SFam Zheng uint64_t bytes = aiocb->aio_nbytes; 17201efad060SFam Zheng off_t in_off = aiocb->aio_offset; 1721d57c44d0SKevin Wolf off_t out_off = aiocb->copy_range.aio_offset2; 17221efad060SFam Zheng 17231efad060SFam Zheng while (bytes) { 17241efad060SFam Zheng ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, 1725d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, &out_off, 17261efad060SFam Zheng bytes, 0); 1727ecc983a5SFam Zheng trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, 1728d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, out_off, bytes, 1729d57c44d0SKevin Wolf 0, ret); 1730c436e3d0SFam Zheng if (ret == 0) { 1731c436e3d0SFam Zheng /* No progress (e.g. when beyond EOF), let the caller fall back to 1732c436e3d0SFam Zheng * buffer I/O. */ 1733c436e3d0SFam Zheng return -ENOSPC; 17341efad060SFam Zheng } 17351efad060SFam Zheng if (ret < 0) { 1736c436e3d0SFam Zheng switch (errno) { 1737c436e3d0SFam Zheng case ENOSYS: 17381efad060SFam Zheng return -ENOTSUP; 1739c436e3d0SFam Zheng case EINTR: 1740c436e3d0SFam Zheng continue; 1741c436e3d0SFam Zheng default: 17421efad060SFam Zheng return -errno; 17431efad060SFam Zheng } 17441efad060SFam Zheng } 17451efad060SFam Zheng bytes -= ret; 17461efad060SFam Zheng } 17471efad060SFam Zheng return 0; 17481efad060SFam Zheng } 17491efad060SFam Zheng 175046ee0f46SKevin Wolf static int handle_aiocb_discard(void *opaque) 1751c1bb86cdSEric Blake { 175246ee0f46SKevin Wolf RawPosixAIOData *aiocb = opaque; 1753c1bb86cdSEric Blake int ret = -EOPNOTSUPP; 1754c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1755c1bb86cdSEric Blake 1756c1bb86cdSEric Blake if (!s->has_discard) { 1757c1bb86cdSEric Blake return -ENOTSUP; 1758c1bb86cdSEric Blake } 1759c1bb86cdSEric Blake 1760c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1761c1bb86cdSEric Blake #ifdef BLKDISCARD 1762c1bb86cdSEric Blake do { 1763c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1764c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { 1765c1bb86cdSEric Blake return 0; 1766c1bb86cdSEric Blake } 1767c1bb86cdSEric Blake } while (errno == EINTR); 1768c1bb86cdSEric Blake 1769c1bb86cdSEric Blake ret = -errno; 1770c1bb86cdSEric Blake #endif 1771c1bb86cdSEric Blake } else { 1772c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1773c1bb86cdSEric Blake ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1774c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1775c1bb86cdSEric Blake #endif 1776c1bb86cdSEric Blake } 1777c1bb86cdSEric Blake 1778c1bb86cdSEric Blake ret = translate_err(ret); 1779c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1780c1bb86cdSEric Blake s->has_discard = false; 1781c1bb86cdSEric Blake } 1782c1bb86cdSEric Blake return ret; 1783c1bb86cdSEric Blake } 1784c1bb86cdSEric Blake 17853a20013fSNir Soffer /* 17863a20013fSNir Soffer * Help alignment probing by allocating the first block. 17873a20013fSNir Soffer * 17883a20013fSNir Soffer * When reading with direct I/O from unallocated area on Gluster backed by XFS, 17893a20013fSNir Soffer * reading succeeds regardless of request length. In this case we fallback to 17903a20013fSNir Soffer * safe alignment which is not optimal. Allocating the first block avoids this 17913a20013fSNir Soffer * fallback. 17923a20013fSNir Soffer * 17933a20013fSNir Soffer * fd may be opened with O_DIRECT, but we don't know the buffer alignment or 17943a20013fSNir Soffer * request alignment, so we use safe values. 17953a20013fSNir Soffer * 17963a20013fSNir Soffer * Returns: 0 on success, -errno on failure. Since this is an optimization, 17973a20013fSNir Soffer * caller may ignore failures. 17983a20013fSNir Soffer */ 17993a20013fSNir Soffer static int allocate_first_block(int fd, size_t max_size) 18003a20013fSNir Soffer { 18013a20013fSNir Soffer size_t write_size = (max_size < MAX_BLOCKSIZE) 18023a20013fSNir Soffer ? BDRV_SECTOR_SIZE 18033a20013fSNir Soffer : MAX_BLOCKSIZE; 1804038adc2fSWei Yang size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); 18053a20013fSNir Soffer void *buf; 18063a20013fSNir Soffer ssize_t n; 18073a20013fSNir Soffer int ret; 18083a20013fSNir Soffer 18093a20013fSNir Soffer buf = qemu_memalign(max_align, write_size); 18103a20013fSNir Soffer memset(buf, 0, write_size); 18113a20013fSNir Soffer 18123a20013fSNir Soffer do { 18133a20013fSNir Soffer n = pwrite(fd, buf, write_size, 0); 18143a20013fSNir Soffer } while (n == -1 && errno == EINTR); 18153a20013fSNir Soffer 18163a20013fSNir Soffer ret = (n == -1) ? -errno : 0; 18173a20013fSNir Soffer 18183a20013fSNir Soffer qemu_vfree(buf); 18193a20013fSNir Soffer return ret; 18203a20013fSNir Soffer } 18213a20013fSNir Soffer 182229cb4c01SKevin Wolf static int handle_aiocb_truncate(void *opaque) 182393f4e2ffSKevin Wolf { 182429cb4c01SKevin Wolf RawPosixAIOData *aiocb = opaque; 182593f4e2ffSKevin Wolf int result = 0; 182693f4e2ffSKevin Wolf int64_t current_length = 0; 182793f4e2ffSKevin Wolf char *buf = NULL; 182893f4e2ffSKevin Wolf struct stat st; 182993f4e2ffSKevin Wolf int fd = aiocb->aio_fildes; 183093f4e2ffSKevin Wolf int64_t offset = aiocb->aio_offset; 1831d57c44d0SKevin Wolf PreallocMode prealloc = aiocb->truncate.prealloc; 1832d57c44d0SKevin Wolf Error **errp = aiocb->truncate.errp; 183393f4e2ffSKevin Wolf 183493f4e2ffSKevin Wolf if (fstat(fd, &st) < 0) { 183593f4e2ffSKevin Wolf result = -errno; 183693f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not stat file"); 183793f4e2ffSKevin Wolf return result; 183893f4e2ffSKevin Wolf } 183993f4e2ffSKevin Wolf 184093f4e2ffSKevin Wolf current_length = st.st_size; 1841d57c44d0SKevin Wolf if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { 184293f4e2ffSKevin Wolf error_setg(errp, "Cannot use preallocation for shrinking files"); 184393f4e2ffSKevin Wolf return -ENOTSUP; 184493f4e2ffSKevin Wolf } 184593f4e2ffSKevin Wolf 1846d57c44d0SKevin Wolf switch (prealloc) { 184793f4e2ffSKevin Wolf #ifdef CONFIG_POSIX_FALLOCATE 184893f4e2ffSKevin Wolf case PREALLOC_MODE_FALLOC: 184993f4e2ffSKevin Wolf /* 185093f4e2ffSKevin Wolf * Truncating before posix_fallocate() makes it about twice slower on 185193f4e2ffSKevin Wolf * file systems that do not support fallocate(), trying to check if a 185293f4e2ffSKevin Wolf * block is allocated before allocating it, so don't do that here. 185393f4e2ffSKevin Wolf */ 185493f4e2ffSKevin Wolf if (offset != current_length) { 185593f4e2ffSKevin Wolf result = -posix_fallocate(fd, current_length, 185693f4e2ffSKevin Wolf offset - current_length); 185793f4e2ffSKevin Wolf if (result != 0) { 185893f4e2ffSKevin Wolf /* posix_fallocate() doesn't set errno. */ 185993f4e2ffSKevin Wolf error_setg_errno(errp, -result, 186093f4e2ffSKevin Wolf "Could not preallocate new data"); 18613a20013fSNir Soffer } else if (current_length == 0) { 18623a20013fSNir Soffer /* 18633a20013fSNir Soffer * posix_fallocate() uses fallocate() if the filesystem 18643a20013fSNir Soffer * supports it, or fallback to manually writing zeroes. If 18653a20013fSNir Soffer * fallocate() was used, unaligned reads from the fallocated 18663a20013fSNir Soffer * area in raw_probe_alignment() will succeed, hence we need to 18673a20013fSNir Soffer * allocate the first block. 18683a20013fSNir Soffer * 18693a20013fSNir Soffer * Optimize future alignment probing; ignore failures. 18703a20013fSNir Soffer */ 18713a20013fSNir Soffer allocate_first_block(fd, offset); 187293f4e2ffSKevin Wolf } 187393f4e2ffSKevin Wolf } else { 187493f4e2ffSKevin Wolf result = 0; 187593f4e2ffSKevin Wolf } 187693f4e2ffSKevin Wolf goto out; 187793f4e2ffSKevin Wolf #endif 187893f4e2ffSKevin Wolf case PREALLOC_MODE_FULL: 187993f4e2ffSKevin Wolf { 188093f4e2ffSKevin Wolf int64_t num = 0, left = offset - current_length; 188193f4e2ffSKevin Wolf off_t seek_result; 188293f4e2ffSKevin Wolf 188393f4e2ffSKevin Wolf /* 188493f4e2ffSKevin Wolf * Knowing the final size from the beginning could allow the file 188593f4e2ffSKevin Wolf * system driver to do less allocations and possibly avoid 188693f4e2ffSKevin Wolf * fragmentation of the file. 188793f4e2ffSKevin Wolf */ 188893f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 188993f4e2ffSKevin Wolf result = -errno; 189093f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 189193f4e2ffSKevin Wolf goto out; 189293f4e2ffSKevin Wolf } 189393f4e2ffSKevin Wolf 189493f4e2ffSKevin Wolf buf = g_malloc0(65536); 189593f4e2ffSKevin Wolf 189693f4e2ffSKevin Wolf seek_result = lseek(fd, current_length, SEEK_SET); 189793f4e2ffSKevin Wolf if (seek_result < 0) { 189893f4e2ffSKevin Wolf result = -errno; 189993f4e2ffSKevin Wolf error_setg_errno(errp, -result, 190093f4e2ffSKevin Wolf "Failed to seek to the old end of file"); 190193f4e2ffSKevin Wolf goto out; 190293f4e2ffSKevin Wolf } 190393f4e2ffSKevin Wolf 190493f4e2ffSKevin Wolf while (left > 0) { 190593f4e2ffSKevin Wolf num = MIN(left, 65536); 190693f4e2ffSKevin Wolf result = write(fd, buf, num); 190793f4e2ffSKevin Wolf if (result < 0) { 1908a1c81f4fSFam Zheng if (errno == EINTR) { 1909a1c81f4fSFam Zheng continue; 1910a1c81f4fSFam Zheng } 191193f4e2ffSKevin Wolf result = -errno; 191293f4e2ffSKevin Wolf error_setg_errno(errp, -result, 191393f4e2ffSKevin Wolf "Could not write zeros for preallocation"); 191493f4e2ffSKevin Wolf goto out; 191593f4e2ffSKevin Wolf } 191693f4e2ffSKevin Wolf left -= result; 191793f4e2ffSKevin Wolf } 191893f4e2ffSKevin Wolf if (result >= 0) { 191993f4e2ffSKevin Wolf result = fsync(fd); 192093f4e2ffSKevin Wolf if (result < 0) { 192193f4e2ffSKevin Wolf result = -errno; 192293f4e2ffSKevin Wolf error_setg_errno(errp, -result, 192393f4e2ffSKevin Wolf "Could not flush file to disk"); 192493f4e2ffSKevin Wolf goto out; 192593f4e2ffSKevin Wolf } 192693f4e2ffSKevin Wolf } 192793f4e2ffSKevin Wolf goto out; 192893f4e2ffSKevin Wolf } 192993f4e2ffSKevin Wolf case PREALLOC_MODE_OFF: 193093f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 193193f4e2ffSKevin Wolf result = -errno; 193293f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 19333a20013fSNir Soffer } else if (current_length == 0 && offset > current_length) { 19343a20013fSNir Soffer /* Optimize future alignment probing; ignore failures. */ 19353a20013fSNir Soffer allocate_first_block(fd, offset); 193693f4e2ffSKevin Wolf } 193793f4e2ffSKevin Wolf return result; 193893f4e2ffSKevin Wolf default: 193993f4e2ffSKevin Wolf result = -ENOTSUP; 194093f4e2ffSKevin Wolf error_setg(errp, "Unsupported preallocation mode: %s", 1941d57c44d0SKevin Wolf PreallocMode_str(prealloc)); 194293f4e2ffSKevin Wolf return result; 194393f4e2ffSKevin Wolf } 194493f4e2ffSKevin Wolf 194593f4e2ffSKevin Wolf out: 194693f4e2ffSKevin Wolf if (result < 0) { 194793f4e2ffSKevin Wolf if (ftruncate(fd, current_length) < 0) { 194893f4e2ffSKevin Wolf error_report("Failed to restore old file length: %s", 194993f4e2ffSKevin Wolf strerror(errno)); 195093f4e2ffSKevin Wolf } 195193f4e2ffSKevin Wolf } 195293f4e2ffSKevin Wolf 195393f4e2ffSKevin Wolf g_free(buf); 195493f4e2ffSKevin Wolf return result; 195593f4e2ffSKevin Wolf } 195693f4e2ffSKevin Wolf 19575d5de250SKevin Wolf static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs, 19585d5de250SKevin Wolf ThreadPoolFunc func, void *arg) 19595d5de250SKevin Wolf { 19605d5de250SKevin Wolf /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ 19615d5de250SKevin Wolf ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 19625d5de250SKevin Wolf return thread_pool_submit_co(pool, func, arg); 19635d5de250SKevin Wolf } 19645d5de250SKevin Wolf 1965c1bb86cdSEric Blake static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, 1966c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, int type) 1967c1bb86cdSEric Blake { 1968c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1969999e6b69SKevin Wolf RawPosixAIOData acb; 1970c1bb86cdSEric Blake 1971c1bb86cdSEric Blake if (fd_open(bs) < 0) 1972c1bb86cdSEric Blake return -EIO; 1973c1bb86cdSEric Blake 1974c1bb86cdSEric Blake /* 1975c6447510SAarushi Mehta * When using O_DIRECT, the request must be aligned to be able to use 1976c6447510SAarushi Mehta * either libaio or io_uring interface. If not fail back to regular thread 1977c6447510SAarushi Mehta * pool read/write code which emulates this for us if we 1978c6447510SAarushi Mehta * set QEMU_AIO_MISALIGNED. 1979c1bb86cdSEric Blake */ 1980c6447510SAarushi Mehta if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) { 1981c1bb86cdSEric Blake type |= QEMU_AIO_MISALIGNED; 1982c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 1983c6447510SAarushi Mehta } else if (s->use_linux_io_uring) { 1984c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 1985c6447510SAarushi Mehta assert(qiov->size == bytes); 1986c6447510SAarushi Mehta return luring_co_submit(bs, aio, s->fd, offset, qiov, type); 1987c6447510SAarushi Mehta #endif 1988c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1989c1bb86cdSEric Blake } else if (s->use_linux_aio) { 1990c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1991c1bb86cdSEric Blake assert(qiov->size == bytes); 1992c1bb86cdSEric Blake return laio_co_submit(bs, aio, s->fd, offset, qiov, type); 1993c1bb86cdSEric Blake #endif 1994c1bb86cdSEric Blake } 1995c1bb86cdSEric Blake 1996999e6b69SKevin Wolf acb = (RawPosixAIOData) { 1997999e6b69SKevin Wolf .bs = bs, 1998999e6b69SKevin Wolf .aio_fildes = s->fd, 1999999e6b69SKevin Wolf .aio_type = type, 2000999e6b69SKevin Wolf .aio_offset = offset, 2001999e6b69SKevin Wolf .aio_nbytes = bytes, 2002999e6b69SKevin Wolf .io = { 2003999e6b69SKevin Wolf .iov = qiov->iov, 2004999e6b69SKevin Wolf .niov = qiov->niov, 2005999e6b69SKevin Wolf }, 2006999e6b69SKevin Wolf }; 2007999e6b69SKevin Wolf 2008999e6b69SKevin Wolf assert(qiov->size == bytes); 2009999e6b69SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb); 2010c1bb86cdSEric Blake } 2011c1bb86cdSEric Blake 2012c1bb86cdSEric Blake static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset, 2013c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 2014c1bb86cdSEric Blake int flags) 2015c1bb86cdSEric Blake { 2016c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); 2017c1bb86cdSEric Blake } 2018c1bb86cdSEric Blake 2019c1bb86cdSEric Blake static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset, 2020c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 2021c1bb86cdSEric Blake int flags) 2022c1bb86cdSEric Blake { 2023c1bb86cdSEric Blake assert(flags == 0); 2024c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); 2025c1bb86cdSEric Blake } 2026c1bb86cdSEric Blake 2027c1bb86cdSEric Blake static void raw_aio_plug(BlockDriverState *bs) 2028c1bb86cdSEric Blake { 2029c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 2030c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 2031c1bb86cdSEric Blake if (s->use_linux_aio) { 2032c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 2033c1bb86cdSEric Blake laio_io_plug(bs, aio); 2034c1bb86cdSEric Blake } 2035c1bb86cdSEric Blake #endif 2036c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2037c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2038c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2039c6447510SAarushi Mehta luring_io_plug(bs, aio); 2040c6447510SAarushi Mehta } 2041c6447510SAarushi Mehta #endif 2042c1bb86cdSEric Blake } 2043c1bb86cdSEric Blake 2044c1bb86cdSEric Blake static void raw_aio_unplug(BlockDriverState *bs) 2045c1bb86cdSEric Blake { 2046c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 2047c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 2048c1bb86cdSEric Blake if (s->use_linux_aio) { 2049c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 2050c1bb86cdSEric Blake laio_io_unplug(bs, aio); 2051c1bb86cdSEric Blake } 2052c1bb86cdSEric Blake #endif 2053c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2054c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2055c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2056c6447510SAarushi Mehta luring_io_unplug(bs, aio); 2057c6447510SAarushi Mehta } 2058c6447510SAarushi Mehta #endif 2059c1bb86cdSEric Blake } 2060c1bb86cdSEric Blake 206133d70fb6SKevin Wolf static int raw_co_flush_to_disk(BlockDriverState *bs) 2062c1bb86cdSEric Blake { 2063c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 206406dc9bd5SKevin Wolf RawPosixAIOData acb; 206533d70fb6SKevin Wolf int ret; 2066c1bb86cdSEric Blake 206733d70fb6SKevin Wolf ret = fd_open(bs); 206833d70fb6SKevin Wolf if (ret < 0) { 206933d70fb6SKevin Wolf return ret; 207033d70fb6SKevin Wolf } 2071c1bb86cdSEric Blake 207206dc9bd5SKevin Wolf acb = (RawPosixAIOData) { 207306dc9bd5SKevin Wolf .bs = bs, 207406dc9bd5SKevin Wolf .aio_fildes = s->fd, 207506dc9bd5SKevin Wolf .aio_type = QEMU_AIO_FLUSH, 207606dc9bd5SKevin Wolf }; 207706dc9bd5SKevin Wolf 2078c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2079c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2080c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2081c6447510SAarushi Mehta return luring_co_submit(bs, aio, s->fd, 0, NULL, QEMU_AIO_FLUSH); 2082c6447510SAarushi Mehta } 2083c6447510SAarushi Mehta #endif 208406dc9bd5SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb); 2085c1bb86cdSEric Blake } 2086c1bb86cdSEric Blake 2087ed6e2161SNishanth Aravamudan static void raw_aio_attach_aio_context(BlockDriverState *bs, 2088ed6e2161SNishanth Aravamudan AioContext *new_context) 2089ed6e2161SNishanth Aravamudan { 2090c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 2091ed6e2161SNishanth Aravamudan #ifdef CONFIG_LINUX_AIO 2092ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 2093cb09104eSMarkus Armbruster Error *local_err = NULL; 2094ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(new_context, &local_err)) { 2095ed6e2161SNishanth Aravamudan error_reportf_err(local_err, "Unable to use native AIO, " 2096ed6e2161SNishanth Aravamudan "falling back to thread pool: "); 2097ed6e2161SNishanth Aravamudan s->use_linux_aio = false; 2098ed6e2161SNishanth Aravamudan } 2099ed6e2161SNishanth Aravamudan } 2100ed6e2161SNishanth Aravamudan #endif 2101c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2102c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2103cb8d0851SPan Nengyuan Error *local_err = NULL; 2104c6447510SAarushi Mehta if (!aio_setup_linux_io_uring(new_context, &local_err)) { 2105c6447510SAarushi Mehta error_reportf_err(local_err, "Unable to use linux io_uring, " 2106c6447510SAarushi Mehta "falling back to thread pool: "); 2107c6447510SAarushi Mehta s->use_linux_io_uring = false; 2108c6447510SAarushi Mehta } 2109c6447510SAarushi Mehta } 2110c6447510SAarushi Mehta #endif 2111ed6e2161SNishanth Aravamudan } 2112ed6e2161SNishanth Aravamudan 2113c1bb86cdSEric Blake static void raw_close(BlockDriverState *bs) 2114c1bb86cdSEric Blake { 2115c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2116c1bb86cdSEric Blake 2117c1bb86cdSEric Blake if (s->fd >= 0) { 2118c1bb86cdSEric Blake qemu_close(s->fd); 2119c1bb86cdSEric Blake s->fd = -1; 2120c1bb86cdSEric Blake } 2121c1bb86cdSEric Blake } 2122c1bb86cdSEric Blake 2123d0bc9e5dSMax Reitz /** 2124d0bc9e5dSMax Reitz * Truncates the given regular file @fd to @offset and, when growing, fills the 2125d0bc9e5dSMax Reitz * new space according to @prealloc. 2126d0bc9e5dSMax Reitz * 2127d0bc9e5dSMax Reitz * Returns: 0 on success, -errno on failure. 2128d0bc9e5dSMax Reitz */ 212993f4e2ffSKevin Wolf static int coroutine_fn 213093f4e2ffSKevin Wolf raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, 213193f4e2ffSKevin Wolf PreallocMode prealloc, Error **errp) 21329f63b07eSMax Reitz { 213329cb4c01SKevin Wolf RawPosixAIOData acb; 2134d0bc9e5dSMax Reitz 213529cb4c01SKevin Wolf acb = (RawPosixAIOData) { 213693f4e2ffSKevin Wolf .bs = bs, 213793f4e2ffSKevin Wolf .aio_fildes = fd, 213893f4e2ffSKevin Wolf .aio_type = QEMU_AIO_TRUNCATE, 213993f4e2ffSKevin Wolf .aio_offset = offset, 2140d57c44d0SKevin Wolf .truncate = { 214193f4e2ffSKevin Wolf .prealloc = prealloc, 214293f4e2ffSKevin Wolf .errp = errp, 2143d57c44d0SKevin Wolf }, 214493f4e2ffSKevin Wolf }; 2145d0bc9e5dSMax Reitz 214629cb4c01SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb); 21479f63b07eSMax Reitz } 21489f63b07eSMax Reitz 2149061ca8a3SKevin Wolf static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, 2150c80d8b06SMax Reitz bool exact, PreallocMode prealloc, 215192b92799SKevin Wolf BdrvRequestFlags flags, Error **errp) 2152c1bb86cdSEric Blake { 2153c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2154c1bb86cdSEric Blake struct stat st; 2155f59adb32SMax Reitz int ret; 2156c1bb86cdSEric Blake 2157c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 2158f59adb32SMax Reitz ret = -errno; 2159f59adb32SMax Reitz error_setg_errno(errp, -ret, "Failed to fstat() the file"); 2160f59adb32SMax Reitz return ret; 2161c1bb86cdSEric Blake } 2162c1bb86cdSEric Blake 2163c1bb86cdSEric Blake if (S_ISREG(st.st_mode)) { 216482325ae5SMax Reitz /* Always resizes to the exact @offset */ 216593f4e2ffSKevin Wolf return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); 2166c1bb86cdSEric Blake } 216735d72602SMax Reitz 216835d72602SMax Reitz if (prealloc != PREALLOC_MODE_OFF) { 216935d72602SMax Reitz error_setg(errp, "Preallocation mode '%s' unsupported for this " 2170977c736fSMarkus Armbruster "non-regular file", PreallocMode_str(prealloc)); 217135d72602SMax Reitz return -ENOTSUP; 217235d72602SMax Reitz } 217335d72602SMax Reitz 217435d72602SMax Reitz if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 217582325ae5SMax Reitz int64_t cur_length = raw_getlength(bs); 217682325ae5SMax Reitz 217782325ae5SMax Reitz if (offset != cur_length && exact) { 217882325ae5SMax Reitz error_setg(errp, "Cannot resize device files"); 217982325ae5SMax Reitz return -ENOTSUP; 218082325ae5SMax Reitz } else if (offset > cur_length) { 2181f59adb32SMax Reitz error_setg(errp, "Cannot grow device files"); 2182c1bb86cdSEric Blake return -EINVAL; 2183c1bb86cdSEric Blake } 2184c1bb86cdSEric Blake } else { 2185f59adb32SMax Reitz error_setg(errp, "Resizing this file is not supported"); 2186c1bb86cdSEric Blake return -ENOTSUP; 2187c1bb86cdSEric Blake } 2188c1bb86cdSEric Blake 2189c1bb86cdSEric Blake return 0; 2190c1bb86cdSEric Blake } 2191c1bb86cdSEric Blake 2192c1bb86cdSEric Blake #ifdef __OpenBSD__ 2193c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2194c1bb86cdSEric Blake { 2195c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2196c1bb86cdSEric Blake int fd = s->fd; 2197c1bb86cdSEric Blake struct stat st; 2198c1bb86cdSEric Blake 2199c1bb86cdSEric Blake if (fstat(fd, &st)) 2200c1bb86cdSEric Blake return -errno; 2201c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2202c1bb86cdSEric Blake struct disklabel dl; 2203c1bb86cdSEric Blake 2204c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2205c1bb86cdSEric Blake return -errno; 2206c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2207c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2208c1bb86cdSEric Blake } else 2209c1bb86cdSEric Blake return st.st_size; 2210c1bb86cdSEric Blake } 2211c1bb86cdSEric Blake #elif defined(__NetBSD__) 2212c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2213c1bb86cdSEric Blake { 2214c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2215c1bb86cdSEric Blake int fd = s->fd; 2216c1bb86cdSEric Blake struct stat st; 2217c1bb86cdSEric Blake 2218c1bb86cdSEric Blake if (fstat(fd, &st)) 2219c1bb86cdSEric Blake return -errno; 2220c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2221c1bb86cdSEric Blake struct dkwedge_info dkw; 2222c1bb86cdSEric Blake 2223c1bb86cdSEric Blake if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { 2224c1bb86cdSEric Blake return dkw.dkw_size * 512; 2225c1bb86cdSEric Blake } else { 2226c1bb86cdSEric Blake struct disklabel dl; 2227c1bb86cdSEric Blake 2228c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2229c1bb86cdSEric Blake return -errno; 2230c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2231c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2232c1bb86cdSEric Blake } 2233c1bb86cdSEric Blake } else 2234c1bb86cdSEric Blake return st.st_size; 2235c1bb86cdSEric Blake } 2236c1bb86cdSEric Blake #elif defined(__sun__) 2237c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2238c1bb86cdSEric Blake { 2239c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2240c1bb86cdSEric Blake struct dk_minfo minfo; 2241c1bb86cdSEric Blake int ret; 2242c1bb86cdSEric Blake int64_t size; 2243c1bb86cdSEric Blake 2244c1bb86cdSEric Blake ret = fd_open(bs); 2245c1bb86cdSEric Blake if (ret < 0) { 2246c1bb86cdSEric Blake return ret; 2247c1bb86cdSEric Blake } 2248c1bb86cdSEric Blake 2249c1bb86cdSEric Blake /* 2250c1bb86cdSEric Blake * Use the DKIOCGMEDIAINFO ioctl to read the size. 2251c1bb86cdSEric Blake */ 2252c1bb86cdSEric Blake ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); 2253c1bb86cdSEric Blake if (ret != -1) { 2254c1bb86cdSEric Blake return minfo.dki_lbsize * minfo.dki_capacity; 2255c1bb86cdSEric Blake } 2256c1bb86cdSEric Blake 2257c1bb86cdSEric Blake /* 2258c1bb86cdSEric Blake * There are reports that lseek on some devices fails, but 2259c1bb86cdSEric Blake * irc discussion said that contingency on contingency was overkill. 2260c1bb86cdSEric Blake */ 2261c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2262c1bb86cdSEric Blake if (size < 0) { 2263c1bb86cdSEric Blake return -errno; 2264c1bb86cdSEric Blake } 2265c1bb86cdSEric Blake return size; 2266c1bb86cdSEric Blake } 2267c1bb86cdSEric Blake #elif defined(CONFIG_BSD) 2268c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2269c1bb86cdSEric Blake { 2270c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2271c1bb86cdSEric Blake int fd = s->fd; 2272c1bb86cdSEric Blake int64_t size; 2273c1bb86cdSEric Blake struct stat sb; 2274c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2275c1bb86cdSEric Blake int reopened = 0; 2276c1bb86cdSEric Blake #endif 2277c1bb86cdSEric Blake int ret; 2278c1bb86cdSEric Blake 2279c1bb86cdSEric Blake ret = fd_open(bs); 2280c1bb86cdSEric Blake if (ret < 0) 2281c1bb86cdSEric Blake return ret; 2282c1bb86cdSEric Blake 2283c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2284c1bb86cdSEric Blake again: 2285c1bb86cdSEric Blake #endif 2286c1bb86cdSEric Blake if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { 2287c1bb86cdSEric Blake #ifdef DIOCGMEDIASIZE 2288c1bb86cdSEric Blake if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) 2289c1bb86cdSEric Blake #elif defined(DIOCGPART) 2290c1bb86cdSEric Blake { 2291c1bb86cdSEric Blake struct partinfo pi; 2292c1bb86cdSEric Blake if (ioctl(fd, DIOCGPART, &pi) == 0) 2293c1bb86cdSEric Blake size = pi.media_size; 2294c1bb86cdSEric Blake else 2295c1bb86cdSEric Blake size = 0; 2296c1bb86cdSEric Blake } 2297c1bb86cdSEric Blake if (size == 0) 2298c1bb86cdSEric Blake #endif 2299c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2300c1bb86cdSEric Blake { 2301c1bb86cdSEric Blake uint64_t sectors = 0; 2302c1bb86cdSEric Blake uint32_t sector_size = 0; 2303c1bb86cdSEric Blake 2304c1bb86cdSEric Blake if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 2305c1bb86cdSEric Blake && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { 2306c1bb86cdSEric Blake size = sectors * sector_size; 2307c1bb86cdSEric Blake } else { 2308c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2309c1bb86cdSEric Blake if (size < 0) { 2310c1bb86cdSEric Blake return -errno; 2311c1bb86cdSEric Blake } 2312c1bb86cdSEric Blake } 2313c1bb86cdSEric Blake } 2314c1bb86cdSEric Blake #else 2315c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2316c1bb86cdSEric Blake if (size < 0) { 2317c1bb86cdSEric Blake return -errno; 2318c1bb86cdSEric Blake } 2319c1bb86cdSEric Blake #endif 2320c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 2321c1bb86cdSEric Blake switch(s->type) { 2322c1bb86cdSEric Blake case FTYPE_CD: 2323c1bb86cdSEric Blake /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ 2324c1bb86cdSEric Blake if (size == 2048LL * (unsigned)-1) 2325c1bb86cdSEric Blake size = 0; 2326c1bb86cdSEric Blake /* XXX no disc? maybe we need to reopen... */ 2327c1bb86cdSEric Blake if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { 2328c1bb86cdSEric Blake reopened = 1; 2329c1bb86cdSEric Blake goto again; 2330c1bb86cdSEric Blake } 2331c1bb86cdSEric Blake } 2332c1bb86cdSEric Blake #endif 2333c1bb86cdSEric Blake } else { 2334c1bb86cdSEric Blake size = lseek(fd, 0, SEEK_END); 2335c1bb86cdSEric Blake if (size < 0) { 2336c1bb86cdSEric Blake return -errno; 2337c1bb86cdSEric Blake } 2338c1bb86cdSEric Blake } 2339c1bb86cdSEric Blake return size; 2340c1bb86cdSEric Blake } 2341c1bb86cdSEric Blake #else 2342c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2343c1bb86cdSEric Blake { 2344c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2345c1bb86cdSEric Blake int ret; 2346c1bb86cdSEric Blake int64_t size; 2347c1bb86cdSEric Blake 2348c1bb86cdSEric Blake ret = fd_open(bs); 2349c1bb86cdSEric Blake if (ret < 0) { 2350c1bb86cdSEric Blake return ret; 2351c1bb86cdSEric Blake } 2352c1bb86cdSEric Blake 2353c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2354c1bb86cdSEric Blake if (size < 0) { 2355c1bb86cdSEric Blake return -errno; 2356c1bb86cdSEric Blake } 2357c1bb86cdSEric Blake return size; 2358c1bb86cdSEric Blake } 2359c1bb86cdSEric Blake #endif 2360c1bb86cdSEric Blake 2361c1bb86cdSEric Blake static int64_t raw_get_allocated_file_size(BlockDriverState *bs) 2362c1bb86cdSEric Blake { 2363c1bb86cdSEric Blake struct stat st; 2364c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2365c1bb86cdSEric Blake 2366c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 2367c1bb86cdSEric Blake return -errno; 2368c1bb86cdSEric Blake } 2369c1bb86cdSEric Blake return (int64_t)st.st_blocks * 512; 2370c1bb86cdSEric Blake } 2371c1bb86cdSEric Blake 237293f4e2ffSKevin Wolf static int coroutine_fn 237393f4e2ffSKevin Wolf raw_co_create(BlockdevCreateOptions *options, Error **errp) 2374c1bb86cdSEric Blake { 2375927f11e1SKevin Wolf BlockdevCreateOptionsFile *file_opts; 23767c20c808SMax Reitz Error *local_err = NULL; 2377c1bb86cdSEric Blake int fd; 2378d815efcaSMax Reitz uint64_t perm, shared; 2379c1bb86cdSEric Blake int result = 0; 2380c1bb86cdSEric Blake 2381927f11e1SKevin Wolf /* Validate options and set default values */ 2382927f11e1SKevin Wolf assert(options->driver == BLOCKDEV_DRIVER_FILE); 2383927f11e1SKevin Wolf file_opts = &options->u.file; 2384c1bb86cdSEric Blake 2385927f11e1SKevin Wolf if (!file_opts->has_nocow) { 2386927f11e1SKevin Wolf file_opts->nocow = false; 2387927f11e1SKevin Wolf } 2388927f11e1SKevin Wolf if (!file_opts->has_preallocation) { 2389927f11e1SKevin Wolf file_opts->preallocation = PREALLOC_MODE_OFF; 2390c1bb86cdSEric Blake } 2391ffa244c8SKevin Wolf if (!file_opts->has_extent_size_hint) { 2392ffa244c8SKevin Wolf file_opts->extent_size_hint = 1 * MiB; 2393ffa244c8SKevin Wolf } 2394ffa244c8SKevin Wolf if (file_opts->extent_size_hint > UINT32_MAX) { 2395ffa244c8SKevin Wolf result = -EINVAL; 2396ffa244c8SKevin Wolf error_setg(errp, "Extent size hint is too large"); 2397ffa244c8SKevin Wolf goto out; 2398ffa244c8SKevin Wolf } 2399c1bb86cdSEric Blake 2400927f11e1SKevin Wolf /* Create file */ 2401b18a24a9SDaniel P. Berrangé fd = qemu_create(file_opts->filename, O_RDWR | O_BINARY, 0644, errp); 2402c1bb86cdSEric Blake if (fd < 0) { 2403c1bb86cdSEric Blake result = -errno; 2404c1bb86cdSEric Blake goto out; 2405c1bb86cdSEric Blake } 2406c1bb86cdSEric Blake 2407b8cf1913SMax Reitz /* Take permissions: We want to discard everything, so we need 2408b8cf1913SMax Reitz * BLK_PERM_WRITE; and truncation to the desired size requires 2409b8cf1913SMax Reitz * BLK_PERM_RESIZE. 2410b8cf1913SMax Reitz * On the other hand, we cannot share the RESIZE permission 2411b8cf1913SMax Reitz * because we promise that after this function, the file has the 2412b8cf1913SMax Reitz * size given in the options. If someone else were to resize it 2413b8cf1913SMax Reitz * concurrently, we could not guarantee that. 2414b8cf1913SMax Reitz * Note that after this function, we can no longer guarantee that 2415b8cf1913SMax Reitz * the file is not touched by a third party, so it may be resized 2416b8cf1913SMax Reitz * then. */ 2417b8cf1913SMax Reitz perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; 2418b8cf1913SMax Reitz shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; 2419b8cf1913SMax Reitz 2420b8cf1913SMax Reitz /* Step one: Take locks */ 24212996ffadSFam Zheng result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); 2422b8cf1913SMax Reitz if (result < 0) { 2423b8cf1913SMax Reitz goto out_close; 2424b8cf1913SMax Reitz } 2425b8cf1913SMax Reitz 2426b8cf1913SMax Reitz /* Step two: Check that nobody else has taken conflicting locks */ 2427b8cf1913SMax Reitz result = raw_check_lock_bytes(fd, perm, shared, errp); 2428b8cf1913SMax Reitz if (result < 0) { 2429b857431dSFam Zheng error_append_hint(errp, 2430b857431dSFam Zheng "Is another process using the image [%s]?\n", 2431b857431dSFam Zheng file_opts->filename); 24327c20c808SMax Reitz goto out_unlock; 2433b8cf1913SMax Reitz } 2434b8cf1913SMax Reitz 2435b8cf1913SMax Reitz /* Clear the file by truncating it to 0 */ 243693f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); 2437b8cf1913SMax Reitz if (result < 0) { 24387c20c808SMax Reitz goto out_unlock; 2439b8cf1913SMax Reitz } 2440b8cf1913SMax Reitz 2441927f11e1SKevin Wolf if (file_opts->nocow) { 2442c1bb86cdSEric Blake #ifdef __linux__ 2443c1bb86cdSEric Blake /* Set NOCOW flag to solve performance issue on fs like btrfs. 2444c1bb86cdSEric Blake * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value 2445c1bb86cdSEric Blake * will be ignored since any failure of this operation should not 2446c1bb86cdSEric Blake * block the left work. 2447c1bb86cdSEric Blake */ 2448c1bb86cdSEric Blake int attr; 2449c1bb86cdSEric Blake if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { 2450c1bb86cdSEric Blake attr |= FS_NOCOW_FL; 2451c1bb86cdSEric Blake ioctl(fd, FS_IOC_SETFLAGS, &attr); 2452c1bb86cdSEric Blake } 2453c1bb86cdSEric Blake #endif 2454c1bb86cdSEric Blake } 2455ffa244c8SKevin Wolf #ifdef FS_IOC_FSSETXATTR 2456ffa244c8SKevin Wolf /* 2457ffa244c8SKevin Wolf * Try to set the extent size hint. Failure is not fatal, and a warning is 2458ffa244c8SKevin Wolf * only printed if the option was explicitly specified. 2459ffa244c8SKevin Wolf */ 2460ffa244c8SKevin Wolf { 2461ffa244c8SKevin Wolf struct fsxattr attr; 2462ffa244c8SKevin Wolf result = ioctl(fd, FS_IOC_FSGETXATTR, &attr); 2463ffa244c8SKevin Wolf if (result == 0) { 2464ffa244c8SKevin Wolf attr.fsx_xflags |= FS_XFLAG_EXTSIZE; 2465ffa244c8SKevin Wolf attr.fsx_extsize = file_opts->extent_size_hint; 2466ffa244c8SKevin Wolf result = ioctl(fd, FS_IOC_FSSETXATTR, &attr); 2467ffa244c8SKevin Wolf } 2468ffa244c8SKevin Wolf if (result < 0 && file_opts->has_extent_size_hint && 2469ffa244c8SKevin Wolf file_opts->extent_size_hint) 2470ffa244c8SKevin Wolf { 2471ffa244c8SKevin Wolf warn_report("Failed to set extent size hint: %s", 2472ffa244c8SKevin Wolf strerror(errno)); 2473ffa244c8SKevin Wolf } 2474ffa244c8SKevin Wolf } 2475ffa244c8SKevin Wolf #endif 2476c1bb86cdSEric Blake 2477b8cf1913SMax Reitz /* Resize and potentially preallocate the file to the desired 2478b8cf1913SMax Reitz * final size */ 247993f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, file_opts->size, 248093f4e2ffSKevin Wolf file_opts->preallocation, errp); 24819f63b07eSMax Reitz if (result < 0) { 24827c20c808SMax Reitz goto out_unlock; 24837c20c808SMax Reitz } 24847c20c808SMax Reitz 24857c20c808SMax Reitz out_unlock: 24862996ffadSFam Zheng raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); 24877c20c808SMax Reitz if (local_err) { 24887c20c808SMax Reitz /* The above call should not fail, and if it does, that does 24897c20c808SMax Reitz * not mean the whole creation operation has failed. So 24907c20c808SMax Reitz * report it the user for their convenience, but do not report 24917c20c808SMax Reitz * it to the caller. */ 2492db0754dfSFam Zheng warn_report_err(local_err); 24935a1dad9dSNir Soffer } 24945a1dad9dSNir Soffer 24955a1dad9dSNir Soffer out_close: 2496c1bb86cdSEric Blake if (qemu_close(fd) != 0 && result == 0) { 2497c1bb86cdSEric Blake result = -errno; 2498c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not close the new file"); 2499c1bb86cdSEric Blake } 2500c1bb86cdSEric Blake out: 2501c1bb86cdSEric Blake return result; 2502c1bb86cdSEric Blake } 2503c1bb86cdSEric Blake 2504b92902dfSMaxim Levitsky static int coroutine_fn raw_co_create_opts(BlockDriver *drv, 2505b92902dfSMaxim Levitsky const char *filename, 2506b92902dfSMaxim Levitsky QemuOpts *opts, 2507927f11e1SKevin Wolf Error **errp) 2508927f11e1SKevin Wolf { 2509927f11e1SKevin Wolf BlockdevCreateOptions options; 2510927f11e1SKevin Wolf int64_t total_size = 0; 2511ffa244c8SKevin Wolf int64_t extent_size_hint = 0; 2512ffa244c8SKevin Wolf bool has_extent_size_hint = false; 2513927f11e1SKevin Wolf bool nocow = false; 2514927f11e1SKevin Wolf PreallocMode prealloc; 2515927f11e1SKevin Wolf char *buf = NULL; 2516927f11e1SKevin Wolf Error *local_err = NULL; 2517927f11e1SKevin Wolf 2518927f11e1SKevin Wolf /* Skip file: protocol prefix */ 2519927f11e1SKevin Wolf strstart(filename, "file:", &filename); 2520927f11e1SKevin Wolf 2521927f11e1SKevin Wolf /* Read out options */ 2522927f11e1SKevin Wolf total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2523927f11e1SKevin Wolf BDRV_SECTOR_SIZE); 2524ffa244c8SKevin Wolf if (qemu_opt_get(opts, BLOCK_OPT_EXTENT_SIZE_HINT)) { 2525ffa244c8SKevin Wolf has_extent_size_hint = true; 2526ffa244c8SKevin Wolf extent_size_hint = 2527ffa244c8SKevin Wolf qemu_opt_get_size_del(opts, BLOCK_OPT_EXTENT_SIZE_HINT, -1); 2528ffa244c8SKevin Wolf } 2529927f11e1SKevin Wolf nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); 2530927f11e1SKevin Wolf buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2531927f11e1SKevin Wolf prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, 2532927f11e1SKevin Wolf PREALLOC_MODE_OFF, &local_err); 2533927f11e1SKevin Wolf g_free(buf); 2534927f11e1SKevin Wolf if (local_err) { 2535927f11e1SKevin Wolf error_propagate(errp, local_err); 2536927f11e1SKevin Wolf return -EINVAL; 2537927f11e1SKevin Wolf } 2538927f11e1SKevin Wolf 2539927f11e1SKevin Wolf options = (BlockdevCreateOptions) { 2540927f11e1SKevin Wolf .driver = BLOCKDEV_DRIVER_FILE, 2541927f11e1SKevin Wolf .u.file = { 2542927f11e1SKevin Wolf .filename = (char *) filename, 2543927f11e1SKevin Wolf .size = total_size, 2544927f11e1SKevin Wolf .has_preallocation = true, 2545927f11e1SKevin Wolf .preallocation = prealloc, 2546927f11e1SKevin Wolf .has_nocow = true, 2547927f11e1SKevin Wolf .nocow = nocow, 2548ffa244c8SKevin Wolf .has_extent_size_hint = has_extent_size_hint, 2549ffa244c8SKevin Wolf .extent_size_hint = extent_size_hint, 2550927f11e1SKevin Wolf }, 2551927f11e1SKevin Wolf }; 2552927f11e1SKevin Wolf return raw_co_create(&options, errp); 2553927f11e1SKevin Wolf } 2554927f11e1SKevin Wolf 25559bffae14SDaniel Henrique Barboza static int coroutine_fn raw_co_delete_file(BlockDriverState *bs, 25569bffae14SDaniel Henrique Barboza Error **errp) 25579bffae14SDaniel Henrique Barboza { 25589bffae14SDaniel Henrique Barboza struct stat st; 25599bffae14SDaniel Henrique Barboza int ret; 25609bffae14SDaniel Henrique Barboza 25619bffae14SDaniel Henrique Barboza if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) { 25629bffae14SDaniel Henrique Barboza error_setg_errno(errp, ENOENT, "%s is not a regular file", 25639bffae14SDaniel Henrique Barboza bs->filename); 25649bffae14SDaniel Henrique Barboza return -ENOENT; 25659bffae14SDaniel Henrique Barboza } 25669bffae14SDaniel Henrique Barboza 25679bffae14SDaniel Henrique Barboza ret = unlink(bs->filename); 25689bffae14SDaniel Henrique Barboza if (ret < 0) { 25699bffae14SDaniel Henrique Barboza ret = -errno; 25709bffae14SDaniel Henrique Barboza error_setg_errno(errp, -ret, "Error when deleting file %s", 25719bffae14SDaniel Henrique Barboza bs->filename); 25729bffae14SDaniel Henrique Barboza } 25739bffae14SDaniel Henrique Barboza 25749bffae14SDaniel Henrique Barboza return ret; 25759bffae14SDaniel Henrique Barboza } 25769bffae14SDaniel Henrique Barboza 2577c1bb86cdSEric Blake /* 2578c1bb86cdSEric Blake * Find allocation range in @bs around offset @start. 2579c1bb86cdSEric Blake * May change underlying file descriptor's file offset. 2580c1bb86cdSEric Blake * If @start is not in a hole, store @start in @data, and the 2581c1bb86cdSEric Blake * beginning of the next hole in @hole, and return 0. 2582c1bb86cdSEric Blake * If @start is in a non-trailing hole, store @start in @hole and the 2583c1bb86cdSEric Blake * beginning of the next non-hole in @data, and return 0. 2584c1bb86cdSEric Blake * If @start is in a trailing hole or beyond EOF, return -ENXIO. 2585c1bb86cdSEric Blake * If we can't find out, return a negative errno other than -ENXIO. 2586c1bb86cdSEric Blake */ 2587c1bb86cdSEric Blake static int find_allocation(BlockDriverState *bs, off_t start, 2588c1bb86cdSEric Blake off_t *data, off_t *hole) 2589c1bb86cdSEric Blake { 2590c1bb86cdSEric Blake #if defined SEEK_HOLE && defined SEEK_DATA 2591c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2592c1bb86cdSEric Blake off_t offs; 2593c1bb86cdSEric Blake 2594c1bb86cdSEric Blake /* 2595c1bb86cdSEric Blake * SEEK_DATA cases: 2596c1bb86cdSEric Blake * D1. offs == start: start is in data 2597c1bb86cdSEric Blake * D2. offs > start: start is in a hole, next data at offs 2598c1bb86cdSEric Blake * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 2599c1bb86cdSEric Blake * or start is beyond EOF 2600c1bb86cdSEric Blake * If the latter happens, the file has been truncated behind 2601c1bb86cdSEric Blake * our back since we opened it. All bets are off then. 2602c1bb86cdSEric Blake * Treating like a trailing hole is simplest. 2603c1bb86cdSEric Blake * D4. offs < 0, errno != ENXIO: we learned nothing 2604c1bb86cdSEric Blake */ 2605c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_DATA); 2606c1bb86cdSEric Blake if (offs < 0) { 2607c1bb86cdSEric Blake return -errno; /* D3 or D4 */ 2608c1bb86cdSEric Blake } 2609a03083a0SJeff Cody 2610a03083a0SJeff Cody if (offs < start) { 2611a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2612a03083a0SJeff Cody * -EIO in this case, and we'll treat it like D4. */ 2613a03083a0SJeff Cody return -EIO; 2614a03083a0SJeff Cody } 2615c1bb86cdSEric Blake 2616c1bb86cdSEric Blake if (offs > start) { 2617c1bb86cdSEric Blake /* D2: in hole, next data at offs */ 2618c1bb86cdSEric Blake *hole = start; 2619c1bb86cdSEric Blake *data = offs; 2620c1bb86cdSEric Blake return 0; 2621c1bb86cdSEric Blake } 2622c1bb86cdSEric Blake 2623c1bb86cdSEric Blake /* D1: in data, end not yet known */ 2624c1bb86cdSEric Blake 2625c1bb86cdSEric Blake /* 2626c1bb86cdSEric Blake * SEEK_HOLE cases: 2627c1bb86cdSEric Blake * H1. offs == start: start is in a hole 2628c1bb86cdSEric Blake * If this happens here, a hole has been dug behind our back 2629c1bb86cdSEric Blake * since the previous lseek(). 2630c1bb86cdSEric Blake * H2. offs > start: either start is in data, next hole at offs, 2631c1bb86cdSEric Blake * or start is in trailing hole, EOF at offs 2632c1bb86cdSEric Blake * Linux treats trailing holes like any other hole: offs == 2633c1bb86cdSEric Blake * start. Solaris seeks to EOF instead: offs > start (blech). 2634c1bb86cdSEric Blake * If that happens here, a hole has been dug behind our back 2635c1bb86cdSEric Blake * since the previous lseek(). 2636c1bb86cdSEric Blake * H3. offs < 0, errno = ENXIO: start is beyond EOF 2637c1bb86cdSEric Blake * If this happens, the file has been truncated behind our 2638c1bb86cdSEric Blake * back since we opened it. Treat it like a trailing hole. 2639c1bb86cdSEric Blake * H4. offs < 0, errno != ENXIO: we learned nothing 2640c1bb86cdSEric Blake * Pretend we know nothing at all, i.e. "forget" about D1. 2641c1bb86cdSEric Blake */ 2642c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_HOLE); 2643c1bb86cdSEric Blake if (offs < 0) { 2644c1bb86cdSEric Blake return -errno; /* D1 and (H3 or H4) */ 2645c1bb86cdSEric Blake } 2646a03083a0SJeff Cody 2647a03083a0SJeff Cody if (offs < start) { 2648a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2649a03083a0SJeff Cody * -EIO in this case, and we'll treat it like H4. */ 2650a03083a0SJeff Cody return -EIO; 2651a03083a0SJeff Cody } 2652c1bb86cdSEric Blake 2653c1bb86cdSEric Blake if (offs > start) { 2654c1bb86cdSEric Blake /* 2655c1bb86cdSEric Blake * D1 and H2: either in data, next hole at offs, or it was in 2656c1bb86cdSEric Blake * data but is now in a trailing hole. In the latter case, 2657c1bb86cdSEric Blake * all bets are off. Treating it as if it there was data all 2658c1bb86cdSEric Blake * the way to EOF is safe, so simply do that. 2659c1bb86cdSEric Blake */ 2660c1bb86cdSEric Blake *data = start; 2661c1bb86cdSEric Blake *hole = offs; 2662c1bb86cdSEric Blake return 0; 2663c1bb86cdSEric Blake } 2664c1bb86cdSEric Blake 2665c1bb86cdSEric Blake /* D1 and H1 */ 2666c1bb86cdSEric Blake return -EBUSY; 2667c1bb86cdSEric Blake #else 2668c1bb86cdSEric Blake return -ENOTSUP; 2669c1bb86cdSEric Blake #endif 2670c1bb86cdSEric Blake } 2671c1bb86cdSEric Blake 2672c1bb86cdSEric Blake /* 2673a290f085SEric Blake * Returns the allocation status of the specified offset. 2674c1bb86cdSEric Blake * 2675a290f085SEric Blake * The block layer guarantees 'offset' and 'bytes' are within bounds. 2676c1bb86cdSEric Blake * 2677a290f085SEric Blake * 'pnum' is set to the number of bytes (including and immediately following 2678a290f085SEric Blake * the specified offset) that are known to be in the same 2679c1bb86cdSEric Blake * allocated/unallocated state. 2680c1bb86cdSEric Blake * 2681a290f085SEric Blake * 'bytes' is the max value 'pnum' should be set to. 2682c1bb86cdSEric Blake */ 2683a290f085SEric Blake static int coroutine_fn raw_co_block_status(BlockDriverState *bs, 2684a290f085SEric Blake bool want_zero, 2685a290f085SEric Blake int64_t offset, 2686a290f085SEric Blake int64_t bytes, int64_t *pnum, 2687a290f085SEric Blake int64_t *map, 2688c1bb86cdSEric Blake BlockDriverState **file) 2689c1bb86cdSEric Blake { 2690a290f085SEric Blake off_t data = 0, hole = 0; 2691c1bb86cdSEric Blake int ret; 2692c1bb86cdSEric Blake 26939c3db310SMax Reitz assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); 26949c3db310SMax Reitz 2695c1bb86cdSEric Blake ret = fd_open(bs); 2696c1bb86cdSEric Blake if (ret < 0) { 2697c1bb86cdSEric Blake return ret; 2698c1bb86cdSEric Blake } 2699c1bb86cdSEric Blake 2700a290f085SEric Blake if (!want_zero) { 2701a290f085SEric Blake *pnum = bytes; 2702a290f085SEric Blake *map = offset; 2703a290f085SEric Blake *file = bs; 2704a290f085SEric Blake return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 2705c1bb86cdSEric Blake } 2706c1bb86cdSEric Blake 2707a290f085SEric Blake ret = find_allocation(bs, offset, &data, &hole); 2708c1bb86cdSEric Blake if (ret == -ENXIO) { 2709c1bb86cdSEric Blake /* Trailing hole */ 2710a290f085SEric Blake *pnum = bytes; 2711c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2712c1bb86cdSEric Blake } else if (ret < 0) { 2713c1bb86cdSEric Blake /* No info available, so pretend there are no holes */ 2714a290f085SEric Blake *pnum = bytes; 2715c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2716a290f085SEric Blake } else if (data == offset) { 2717a290f085SEric Blake /* On a data extent, compute bytes to the end of the extent, 2718c1bb86cdSEric Blake * possibly including a partial sector at EOF. */ 2719a290f085SEric Blake *pnum = MIN(bytes, hole - offset); 27209c3db310SMax Reitz 27219c3db310SMax Reitz /* 27229c3db310SMax Reitz * We are not allowed to return partial sectors, though, so 27239c3db310SMax Reitz * round up if necessary. 27249c3db310SMax Reitz */ 27259c3db310SMax Reitz if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { 27269c3db310SMax Reitz int64_t file_length = raw_getlength(bs); 27279c3db310SMax Reitz if (file_length > 0) { 27289c3db310SMax Reitz /* Ignore errors, this is just a safeguard */ 27299c3db310SMax Reitz assert(hole == file_length); 27309c3db310SMax Reitz } 27319c3db310SMax Reitz *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); 27329c3db310SMax Reitz } 27339c3db310SMax Reitz 2734c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2735c1bb86cdSEric Blake } else { 2736a290f085SEric Blake /* On a hole, compute bytes to the beginning of the next extent. */ 2737a290f085SEric Blake assert(hole == offset); 2738a290f085SEric Blake *pnum = MIN(bytes, data - offset); 2739c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2740c1bb86cdSEric Blake } 2741a290f085SEric Blake *map = offset; 2742c1bb86cdSEric Blake *file = bs; 2743a290f085SEric Blake return ret | BDRV_BLOCK_OFFSET_VALID; 2744c1bb86cdSEric Blake } 2745c1bb86cdSEric Blake 274631be8a2aSStefan Hajnoczi #if defined(__linux__) 274731be8a2aSStefan Hajnoczi /* Verify that the file is not in the page cache */ 274831be8a2aSStefan Hajnoczi static void check_cache_dropped(BlockDriverState *bs, Error **errp) 274931be8a2aSStefan Hajnoczi { 275031be8a2aSStefan Hajnoczi const size_t window_size = 128 * 1024 * 1024; 275131be8a2aSStefan Hajnoczi BDRVRawState *s = bs->opaque; 275231be8a2aSStefan Hajnoczi void *window = NULL; 275331be8a2aSStefan Hajnoczi size_t length = 0; 275431be8a2aSStefan Hajnoczi unsigned char *vec; 275531be8a2aSStefan Hajnoczi size_t page_size; 275631be8a2aSStefan Hajnoczi off_t offset; 275731be8a2aSStefan Hajnoczi off_t end; 275831be8a2aSStefan Hajnoczi 275931be8a2aSStefan Hajnoczi /* mincore(2) page status information requires 1 byte per page */ 276031be8a2aSStefan Hajnoczi page_size = sysconf(_SC_PAGESIZE); 276131be8a2aSStefan Hajnoczi vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); 276231be8a2aSStefan Hajnoczi 276331be8a2aSStefan Hajnoczi end = raw_getlength(bs); 276431be8a2aSStefan Hajnoczi 276531be8a2aSStefan Hajnoczi for (offset = 0; offset < end; offset += window_size) { 276631be8a2aSStefan Hajnoczi void *new_window; 276731be8a2aSStefan Hajnoczi size_t new_length; 276831be8a2aSStefan Hajnoczi size_t vec_end; 276931be8a2aSStefan Hajnoczi size_t i; 277031be8a2aSStefan Hajnoczi int ret; 277131be8a2aSStefan Hajnoczi 277231be8a2aSStefan Hajnoczi /* Unmap previous window if size has changed */ 277331be8a2aSStefan Hajnoczi new_length = MIN(end - offset, window_size); 277431be8a2aSStefan Hajnoczi if (new_length != length) { 277531be8a2aSStefan Hajnoczi munmap(window, length); 277631be8a2aSStefan Hajnoczi window = NULL; 277731be8a2aSStefan Hajnoczi length = 0; 277831be8a2aSStefan Hajnoczi } 277931be8a2aSStefan Hajnoczi 278031be8a2aSStefan Hajnoczi new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, 278131be8a2aSStefan Hajnoczi s->fd, offset); 278231be8a2aSStefan Hajnoczi if (new_window == MAP_FAILED) { 278331be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mmap failed"); 278431be8a2aSStefan Hajnoczi break; 278531be8a2aSStefan Hajnoczi } 278631be8a2aSStefan Hajnoczi 278731be8a2aSStefan Hajnoczi window = new_window; 278831be8a2aSStefan Hajnoczi length = new_length; 278931be8a2aSStefan Hajnoczi 279031be8a2aSStefan Hajnoczi ret = mincore(window, length, vec); 279131be8a2aSStefan Hajnoczi if (ret < 0) { 279231be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mincore failed"); 279331be8a2aSStefan Hajnoczi break; 279431be8a2aSStefan Hajnoczi } 279531be8a2aSStefan Hajnoczi 279631be8a2aSStefan Hajnoczi vec_end = DIV_ROUND_UP(length, page_size); 279731be8a2aSStefan Hajnoczi for (i = 0; i < vec_end; i++) { 279831be8a2aSStefan Hajnoczi if (vec[i] & 0x1) { 279931be8a2aSStefan Hajnoczi break; 280031be8a2aSStefan Hajnoczi } 280131be8a2aSStefan Hajnoczi } 280277ed971bSMarkus Armbruster if (i < vec_end) { 280377ed971bSMarkus Armbruster error_setg(errp, "page cache still in use!"); 280477ed971bSMarkus Armbruster break; 280577ed971bSMarkus Armbruster } 280631be8a2aSStefan Hajnoczi } 280731be8a2aSStefan Hajnoczi 280831be8a2aSStefan Hajnoczi if (window) { 280931be8a2aSStefan Hajnoczi munmap(window, length); 281031be8a2aSStefan Hajnoczi } 281131be8a2aSStefan Hajnoczi 281231be8a2aSStefan Hajnoczi g_free(vec); 281331be8a2aSStefan Hajnoczi } 281431be8a2aSStefan Hajnoczi #endif /* __linux__ */ 281531be8a2aSStefan Hajnoczi 2816dd577a26SStefan Hajnoczi static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, 2817dd577a26SStefan Hajnoczi Error **errp) 2818dd577a26SStefan Hajnoczi { 2819dd577a26SStefan Hajnoczi BDRVRawState *s = bs->opaque; 2820dd577a26SStefan Hajnoczi int ret; 2821dd577a26SStefan Hajnoczi 2822dd577a26SStefan Hajnoczi ret = fd_open(bs); 2823dd577a26SStefan Hajnoczi if (ret < 0) { 2824dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "The file descriptor is not open"); 2825dd577a26SStefan Hajnoczi return; 2826dd577a26SStefan Hajnoczi } 2827dd577a26SStefan Hajnoczi 2828f357fcd8SStefan Hajnoczi if (!s->drop_cache) { 2829f357fcd8SStefan Hajnoczi return; 2830f357fcd8SStefan Hajnoczi } 2831f357fcd8SStefan Hajnoczi 2832dd577a26SStefan Hajnoczi if (s->open_flags & O_DIRECT) { 2833dd577a26SStefan Hajnoczi return; /* No host kernel page cache */ 2834dd577a26SStefan Hajnoczi } 2835dd577a26SStefan Hajnoczi 2836dd577a26SStefan Hajnoczi #if defined(__linux__) 2837dd577a26SStefan Hajnoczi /* This sets the scene for the next syscall... */ 2838dd577a26SStefan Hajnoczi ret = bdrv_co_flush(bs); 2839dd577a26SStefan Hajnoczi if (ret < 0) { 2840dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "flush failed"); 2841dd577a26SStefan Hajnoczi return; 2842dd577a26SStefan Hajnoczi } 2843dd577a26SStefan Hajnoczi 2844dd577a26SStefan Hajnoczi /* Linux does not invalidate pages that are dirty, locked, or mmapped by a 2845dd577a26SStefan Hajnoczi * process. These limitations are okay because we just fsynced the file, 2846dd577a26SStefan Hajnoczi * we don't use mmap, and the file should not be in use by other processes. 2847dd577a26SStefan Hajnoczi */ 2848dd577a26SStefan Hajnoczi ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); 2849dd577a26SStefan Hajnoczi if (ret != 0) { /* the return value is a positive errno */ 2850dd577a26SStefan Hajnoczi error_setg_errno(errp, ret, "fadvise failed"); 2851dd577a26SStefan Hajnoczi return; 2852dd577a26SStefan Hajnoczi } 285331be8a2aSStefan Hajnoczi 285431be8a2aSStefan Hajnoczi if (s->check_cache_dropped) { 285531be8a2aSStefan Hajnoczi check_cache_dropped(bs, errp); 285631be8a2aSStefan Hajnoczi } 2857dd577a26SStefan Hajnoczi #else /* __linux__ */ 2858dd577a26SStefan Hajnoczi /* Do nothing. Live migration to a remote host with cache.direct=off is 2859dd577a26SStefan Hajnoczi * unsupported on other host operating systems. Cache consistency issues 2860dd577a26SStefan Hajnoczi * may occur but no error is reported here, partly because that's the 2861dd577a26SStefan Hajnoczi * historical behavior and partly because it's hard to differentiate valid 2862dd577a26SStefan Hajnoczi * configurations that should not cause errors. 2863dd577a26SStefan Hajnoczi */ 2864dd577a26SStefan Hajnoczi #endif /* !__linux__ */ 2865dd577a26SStefan Hajnoczi } 2866dd577a26SStefan Hajnoczi 28671c450366SAnton Nefedov static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret) 28681c450366SAnton Nefedov { 28691c450366SAnton Nefedov if (ret) { 28701c450366SAnton Nefedov s->stats.discard_nb_failed++; 28711c450366SAnton Nefedov } else { 28721c450366SAnton Nefedov s->stats.discard_nb_ok++; 28731c450366SAnton Nefedov s->stats.discard_bytes_ok += nbytes; 28741c450366SAnton Nefedov } 28751c450366SAnton Nefedov } 28761c450366SAnton Nefedov 287733d70fb6SKevin Wolf static coroutine_fn int 287846ee0f46SKevin Wolf raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev) 2879c1bb86cdSEric Blake { 2880c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 288146ee0f46SKevin Wolf RawPosixAIOData acb; 28821c450366SAnton Nefedov int ret; 2883c1bb86cdSEric Blake 288446ee0f46SKevin Wolf acb = (RawPosixAIOData) { 288546ee0f46SKevin Wolf .bs = bs, 288646ee0f46SKevin Wolf .aio_fildes = s->fd, 288746ee0f46SKevin Wolf .aio_type = QEMU_AIO_DISCARD, 288846ee0f46SKevin Wolf .aio_offset = offset, 288946ee0f46SKevin Wolf .aio_nbytes = bytes, 289046ee0f46SKevin Wolf }; 289146ee0f46SKevin Wolf 289246ee0f46SKevin Wolf if (blkdev) { 289346ee0f46SKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 289446ee0f46SKevin Wolf } 289546ee0f46SKevin Wolf 28961c450366SAnton Nefedov ret = raw_thread_pool_submit(bs, handle_aiocb_discard, &acb); 28971c450366SAnton Nefedov raw_account_discard(s, bytes, ret); 28981c450366SAnton Nefedov return ret; 289946ee0f46SKevin Wolf } 290046ee0f46SKevin Wolf 290146ee0f46SKevin Wolf static coroutine_fn int 290246ee0f46SKevin Wolf raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 290346ee0f46SKevin Wolf { 290446ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, false); 2905c1bb86cdSEric Blake } 2906c1bb86cdSEric Blake 29077154d8aeSKevin Wolf static int coroutine_fn 29087154d8aeSKevin Wolf raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, 29097154d8aeSKevin Wolf BdrvRequestFlags flags, bool blkdev) 29107154d8aeSKevin Wolf { 29117154d8aeSKevin Wolf BDRVRawState *s = bs->opaque; 29127154d8aeSKevin Wolf RawPosixAIOData acb; 29137154d8aeSKevin Wolf ThreadPoolFunc *handler; 29147154d8aeSKevin Wolf 2915292d06b9SMax Reitz #ifdef CONFIG_FALLOCATE 2916292d06b9SMax Reitz if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { 2917292d06b9SMax Reitz BdrvTrackedRequest *req; 2918292d06b9SMax Reitz 2919292d06b9SMax Reitz /* 2920292d06b9SMax Reitz * This is a workaround for a bug in the Linux XFS driver, 2921292d06b9SMax Reitz * where writes submitted through the AIO interface will be 2922292d06b9SMax Reitz * discarded if they happen beyond a concurrently running 2923292d06b9SMax Reitz * fallocate() that increases the file length (i.e., both the 2924292d06b9SMax Reitz * write and the fallocate() happen beyond the EOF). 2925292d06b9SMax Reitz * 2926292d06b9SMax Reitz * To work around it, we extend the tracked request for this 2927292d06b9SMax Reitz * zero write until INT64_MAX (effectively infinity), and mark 2928292d06b9SMax Reitz * it as serializing. 2929292d06b9SMax Reitz * 2930292d06b9SMax Reitz * We have to enable this workaround for all filesystems and 2931292d06b9SMax Reitz * AIO modes (not just XFS with aio=native), because for 2932292d06b9SMax Reitz * remote filesystems we do not know the host configuration. 2933292d06b9SMax Reitz */ 2934292d06b9SMax Reitz 2935292d06b9SMax Reitz req = bdrv_co_get_self_request(bs); 2936292d06b9SMax Reitz assert(req); 2937292d06b9SMax Reitz assert(req->type == BDRV_TRACKED_WRITE); 2938292d06b9SMax Reitz assert(req->offset <= offset); 2939292d06b9SMax Reitz assert(req->offset + req->bytes >= offset + bytes); 2940292d06b9SMax Reitz 29418b117001SVladimir Sementsov-Ogievskiy req->bytes = BDRV_MAX_LENGTH - req->offset; 29428b117001SVladimir Sementsov-Ogievskiy 294369b55e03SVladimir Sementsov-Ogievskiy bdrv_check_request(req->offset, req->bytes, &error_abort); 2944292d06b9SMax Reitz 29458ac5aab2SVladimir Sementsov-Ogievskiy bdrv_make_request_serialising(req, bs->bl.request_alignment); 2946292d06b9SMax Reitz } 2947292d06b9SMax Reitz #endif 2948292d06b9SMax Reitz 29497154d8aeSKevin Wolf acb = (RawPosixAIOData) { 29507154d8aeSKevin Wolf .bs = bs, 29517154d8aeSKevin Wolf .aio_fildes = s->fd, 29527154d8aeSKevin Wolf .aio_type = QEMU_AIO_WRITE_ZEROES, 29537154d8aeSKevin Wolf .aio_offset = offset, 29547154d8aeSKevin Wolf .aio_nbytes = bytes, 29557154d8aeSKevin Wolf }; 29567154d8aeSKevin Wolf 29577154d8aeSKevin Wolf if (blkdev) { 29587154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 29597154d8aeSKevin Wolf } 2960738301e1SKevin Wolf if (flags & BDRV_REQ_NO_FALLBACK) { 2961738301e1SKevin Wolf acb.aio_type |= QEMU_AIO_NO_FALLBACK; 2962738301e1SKevin Wolf } 29637154d8aeSKevin Wolf 29647154d8aeSKevin Wolf if (flags & BDRV_REQ_MAY_UNMAP) { 29657154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_DISCARD; 29667154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes_unmap; 29677154d8aeSKevin Wolf } else { 29687154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes; 29697154d8aeSKevin Wolf } 29707154d8aeSKevin Wolf 29717154d8aeSKevin Wolf return raw_thread_pool_submit(bs, handler, &acb); 29727154d8aeSKevin Wolf } 29737154d8aeSKevin Wolf 2974c1bb86cdSEric Blake static int coroutine_fn raw_co_pwrite_zeroes( 2975c1bb86cdSEric Blake BlockDriverState *bs, int64_t offset, 2976f5a5ca79SManos Pitsidianakis int bytes, BdrvRequestFlags flags) 2977c1bb86cdSEric Blake { 29787154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); 2979c1bb86cdSEric Blake } 2980c1bb86cdSEric Blake 2981c1bb86cdSEric Blake static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2982c1bb86cdSEric Blake { 2983c1bb86cdSEric Blake return 0; 2984c1bb86cdSEric Blake } 2985c1bb86cdSEric Blake 2986d9245599SAnton Nefedov static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs) 2987d9245599SAnton Nefedov { 2988d9245599SAnton Nefedov BDRVRawState *s = bs->opaque; 2989d9245599SAnton Nefedov return (BlockStatsSpecificFile) { 2990d9245599SAnton Nefedov .discard_nb_ok = s->stats.discard_nb_ok, 2991d9245599SAnton Nefedov .discard_nb_failed = s->stats.discard_nb_failed, 2992d9245599SAnton Nefedov .discard_bytes_ok = s->stats.discard_bytes_ok, 2993d9245599SAnton Nefedov }; 2994d9245599SAnton Nefedov } 2995d9245599SAnton Nefedov 2996d9245599SAnton Nefedov static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs) 2997d9245599SAnton Nefedov { 2998d9245599SAnton Nefedov BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 2999d9245599SAnton Nefedov 3000d9245599SAnton Nefedov stats->driver = BLOCKDEV_DRIVER_FILE; 3001d9245599SAnton Nefedov stats->u.file = get_blockstats_specific_file(bs); 3002d9245599SAnton Nefedov 3003d9245599SAnton Nefedov return stats; 3004d9245599SAnton Nefedov } 3005d9245599SAnton Nefedov 3006d9245599SAnton Nefedov static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs) 3007d9245599SAnton Nefedov { 3008d9245599SAnton Nefedov BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 3009d9245599SAnton Nefedov 3010d9245599SAnton Nefedov stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE; 3011d9245599SAnton Nefedov stats->u.host_device = get_blockstats_specific_file(bs); 3012d9245599SAnton Nefedov 3013d9245599SAnton Nefedov return stats; 3014d9245599SAnton Nefedov } 3015d9245599SAnton Nefedov 3016c1bb86cdSEric Blake static QemuOptsList raw_create_opts = { 3017c1bb86cdSEric Blake .name = "raw-create-opts", 3018c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), 3019c1bb86cdSEric Blake .desc = { 3020c1bb86cdSEric Blake { 3021c1bb86cdSEric Blake .name = BLOCK_OPT_SIZE, 3022c1bb86cdSEric Blake .type = QEMU_OPT_SIZE, 3023c1bb86cdSEric Blake .help = "Virtual disk size" 3024c1bb86cdSEric Blake }, 3025c1bb86cdSEric Blake { 3026c1bb86cdSEric Blake .name = BLOCK_OPT_NOCOW, 3027c1bb86cdSEric Blake .type = QEMU_OPT_BOOL, 3028c1bb86cdSEric Blake .help = "Turn off copy-on-write (valid only on btrfs)" 3029c1bb86cdSEric Blake }, 3030c1bb86cdSEric Blake { 3031c1bb86cdSEric Blake .name = BLOCK_OPT_PREALLOC, 3032c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 3033abea0053SStefano Garzarella .help = "Preallocation mode (allowed values: off" 3034abea0053SStefano Garzarella #ifdef CONFIG_POSIX_FALLOCATE 3035abea0053SStefano Garzarella ", falloc" 3036abea0053SStefano Garzarella #endif 3037abea0053SStefano Garzarella ", full)" 3038c1bb86cdSEric Blake }, 3039ffa244c8SKevin Wolf { 3040ffa244c8SKevin Wolf .name = BLOCK_OPT_EXTENT_SIZE_HINT, 3041ffa244c8SKevin Wolf .type = QEMU_OPT_SIZE, 3042ffa244c8SKevin Wolf .help = "Extent size hint for the image file, 0 to disable" 3043ffa244c8SKevin Wolf }, 3044c1bb86cdSEric Blake { /* end of list */ } 3045c1bb86cdSEric Blake } 3046c1bb86cdSEric Blake }; 3047c1bb86cdSEric Blake 3048244a5668SFam Zheng static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, 3049244a5668SFam Zheng Error **errp) 3050244a5668SFam Zheng { 30516ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 3052*72373e40SVladimir Sementsov-Ogievskiy int input_flags = s->reopen_state ? s->reopen_state->flags : bs->open_flags; 30536ceabe6fSKevin Wolf int open_flags; 30546ceabe6fSKevin Wolf int ret; 30556ceabe6fSKevin Wolf 30566ceabe6fSKevin Wolf /* We may need a new fd if auto-read-only switches the mode */ 3057*72373e40SVladimir Sementsov-Ogievskiy ret = raw_reconfigure_getfd(bs, input_flags, &open_flags, perm, 30586ceabe6fSKevin Wolf false, errp); 30596ceabe6fSKevin Wolf if (ret < 0) { 30606ceabe6fSKevin Wolf return ret; 30616ceabe6fSKevin Wolf } else if (ret != s->fd) { 3062*72373e40SVladimir Sementsov-Ogievskiy Error *local_err = NULL; 3063*72373e40SVladimir Sementsov-Ogievskiy 3064*72373e40SVladimir Sementsov-Ogievskiy /* 3065*72373e40SVladimir Sementsov-Ogievskiy * Fail already check_perm() if we can't get a working O_DIRECT 3066*72373e40SVladimir Sementsov-Ogievskiy * alignment with the new fd. 3067*72373e40SVladimir Sementsov-Ogievskiy */ 3068*72373e40SVladimir Sementsov-Ogievskiy raw_probe_alignment(bs, ret, &local_err); 3069*72373e40SVladimir Sementsov-Ogievskiy if (local_err) { 3070*72373e40SVladimir Sementsov-Ogievskiy error_propagate(errp, local_err); 3071*72373e40SVladimir Sementsov-Ogievskiy return -EINVAL; 3072*72373e40SVladimir Sementsov-Ogievskiy } 3073*72373e40SVladimir Sementsov-Ogievskiy 30746ceabe6fSKevin Wolf s->perm_change_fd = ret; 3075094e3639SMax Reitz s->perm_change_flags = open_flags; 30766ceabe6fSKevin Wolf } 30776ceabe6fSKevin Wolf 30786ceabe6fSKevin Wolf /* Prepare permissions on old fd to avoid conflicts between old and new, 30796ceabe6fSKevin Wolf * but keep everything locked that new will need. */ 30806ceabe6fSKevin Wolf ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); 30816ceabe6fSKevin Wolf if (ret < 0) { 30826ceabe6fSKevin Wolf goto fail; 30836ceabe6fSKevin Wolf } 30846ceabe6fSKevin Wolf 30856ceabe6fSKevin Wolf /* Copy locks to the new fd */ 3086eb43ea16SLi Feng if (s->perm_change_fd && s->use_lock) { 30876ceabe6fSKevin Wolf ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, 30886ceabe6fSKevin Wolf false, errp); 30896ceabe6fSKevin Wolf if (ret < 0) { 30906ceabe6fSKevin Wolf raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 30916ceabe6fSKevin Wolf goto fail; 30926ceabe6fSKevin Wolf } 30936ceabe6fSKevin Wolf } 30946ceabe6fSKevin Wolf return 0; 30956ceabe6fSKevin Wolf 30966ceabe6fSKevin Wolf fail: 3097*72373e40SVladimir Sementsov-Ogievskiy if (s->perm_change_fd) { 30986ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 30996ceabe6fSKevin Wolf } 31006ceabe6fSKevin Wolf s->perm_change_fd = 0; 31016ceabe6fSKevin Wolf return ret; 3102244a5668SFam Zheng } 3103244a5668SFam Zheng 3104244a5668SFam Zheng static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) 3105244a5668SFam Zheng { 3106244a5668SFam Zheng BDRVRawState *s = bs->opaque; 31076ceabe6fSKevin Wolf 31086ceabe6fSKevin Wolf /* For reopen, we have already switched to the new fd (.bdrv_set_perm is 31096ceabe6fSKevin Wolf * called after .bdrv_reopen_commit) */ 31106ceabe6fSKevin Wolf if (s->perm_change_fd && s->fd != s->perm_change_fd) { 31116ceabe6fSKevin Wolf qemu_close(s->fd); 31126ceabe6fSKevin Wolf s->fd = s->perm_change_fd; 3113094e3639SMax Reitz s->open_flags = s->perm_change_flags; 31146ceabe6fSKevin Wolf } 31156ceabe6fSKevin Wolf s->perm_change_fd = 0; 31166ceabe6fSKevin Wolf 3117244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); 3118244a5668SFam Zheng s->perm = perm; 3119244a5668SFam Zheng s->shared_perm = shared; 3120244a5668SFam Zheng } 3121244a5668SFam Zheng 3122244a5668SFam Zheng static void raw_abort_perm_update(BlockDriverState *bs) 3123244a5668SFam Zheng { 31246ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 31256ceabe6fSKevin Wolf 31266ceabe6fSKevin Wolf /* For reopen, .bdrv_reopen_abort is called afterwards and will close 31276ceabe6fSKevin Wolf * the file descriptor. */ 3128*72373e40SVladimir Sementsov-Ogievskiy if (s->perm_change_fd) { 31296ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 31306ceabe6fSKevin Wolf } 31316ceabe6fSKevin Wolf s->perm_change_fd = 0; 31326ceabe6fSKevin Wolf 3133244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 3134244a5668SFam Zheng } 3135244a5668SFam Zheng 313667b51fb9SVladimir Sementsov-Ogievskiy static int coroutine_fn raw_co_copy_range_from( 313767b51fb9SVladimir Sementsov-Ogievskiy BlockDriverState *bs, BdrvChild *src, uint64_t src_offset, 313867b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, uint64_t dst_offset, uint64_t bytes, 313967b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) 31401efad060SFam Zheng { 314167b51fb9SVladimir Sementsov-Ogievskiy return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, 314267b51fb9SVladimir Sementsov-Ogievskiy read_flags, write_flags); 31431efad060SFam Zheng } 31441efad060SFam Zheng 31451efad060SFam Zheng static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, 314667b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *src, 314767b51fb9SVladimir Sementsov-Ogievskiy uint64_t src_offset, 314867b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, 314967b51fb9SVladimir Sementsov-Ogievskiy uint64_t dst_offset, 315067b51fb9SVladimir Sementsov-Ogievskiy uint64_t bytes, 315167b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, 315267b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags write_flags) 31531efad060SFam Zheng { 315458a209c4SKevin Wolf RawPosixAIOData acb; 31551efad060SFam Zheng BDRVRawState *s = bs->opaque; 31561efad060SFam Zheng BDRVRawState *src_s; 31571efad060SFam Zheng 31581efad060SFam Zheng assert(dst->bs == bs); 31591efad060SFam Zheng if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { 31601efad060SFam Zheng return -ENOTSUP; 31611efad060SFam Zheng } 31621efad060SFam Zheng 31631efad060SFam Zheng src_s = src->bs->opaque; 31649f850f67SFam Zheng if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { 31651efad060SFam Zheng return -EIO; 31661efad060SFam Zheng } 316758a209c4SKevin Wolf 316858a209c4SKevin Wolf acb = (RawPosixAIOData) { 316958a209c4SKevin Wolf .bs = bs, 317058a209c4SKevin Wolf .aio_type = QEMU_AIO_COPY_RANGE, 317158a209c4SKevin Wolf .aio_fildes = src_s->fd, 317258a209c4SKevin Wolf .aio_offset = src_offset, 317358a209c4SKevin Wolf .aio_nbytes = bytes, 317458a209c4SKevin Wolf .copy_range = { 317558a209c4SKevin Wolf .aio_fd2 = s->fd, 317658a209c4SKevin Wolf .aio_offset2 = dst_offset, 317758a209c4SKevin Wolf }, 317858a209c4SKevin Wolf }; 317958a209c4SKevin Wolf 318058a209c4SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); 31811efad060SFam Zheng } 31821efad060SFam Zheng 3183c1bb86cdSEric Blake BlockDriver bdrv_file = { 3184c1bb86cdSEric Blake .format_name = "file", 3185c1bb86cdSEric Blake .protocol_name = "file", 3186c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3187c1bb86cdSEric Blake .bdrv_needs_filename = true, 3188c1bb86cdSEric Blake .bdrv_probe = NULL, /* no probe for protocols */ 3189c1bb86cdSEric Blake .bdrv_parse_filename = raw_parse_filename, 3190c1bb86cdSEric Blake .bdrv_file_open = raw_open, 3191c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3192c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3193c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3194c1bb86cdSEric Blake .bdrv_close = raw_close, 3195927f11e1SKevin Wolf .bdrv_co_create = raw_co_create, 3196efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = raw_co_create_opts, 3197c1bb86cdSEric Blake .bdrv_has_zero_init = bdrv_has_zero_init_1, 3198a290f085SEric Blake .bdrv_co_block_status = raw_co_block_status, 3199dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3200c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, 32019bffae14SDaniel Henrique Barboza .bdrv_co_delete_file = raw_co_delete_file, 3202c1bb86cdSEric Blake 3203c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3204c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 320533d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 320633d70fb6SKevin Wolf .bdrv_co_pdiscard = raw_co_pdiscard, 32071efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 32081efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 3209c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3210c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3211c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3212ed6e2161SNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3213c1bb86cdSEric Blake 3214061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3215c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3216c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 3217c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3218c1bb86cdSEric Blake = raw_get_allocated_file_size, 3219d9245599SAnton Nefedov .bdrv_get_specific_stats = raw_get_specific_stats, 3220244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 3221244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 3222244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 3223c1bb86cdSEric Blake .create_opts = &raw_create_opts, 32248a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3225c1bb86cdSEric Blake }; 3226c1bb86cdSEric Blake 3227c1bb86cdSEric Blake /***********************************************/ 3228c1bb86cdSEric Blake /* host device */ 3229c1bb86cdSEric Blake 3230c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3231c1bb86cdSEric Blake static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3232c1bb86cdSEric Blake CFIndex maxPathSize, int flags); 3233c1bb86cdSEric Blake static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) 3234c1bb86cdSEric Blake { 3235c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 3236c1bb86cdSEric Blake mach_port_t masterPort; 3237c1bb86cdSEric Blake CFMutableDictionaryRef classesToMatch; 3238c1bb86cdSEric Blake const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; 3239c1bb86cdSEric Blake char *mediaType = NULL; 3240c1bb86cdSEric Blake 3241c1bb86cdSEric Blake kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); 3242c1bb86cdSEric Blake if ( KERN_SUCCESS != kernResult ) { 3243c1bb86cdSEric Blake printf( "IOMasterPort returned %d\n", kernResult ); 3244c1bb86cdSEric Blake } 3245c1bb86cdSEric Blake 3246c1bb86cdSEric Blake int index; 3247c1bb86cdSEric Blake for (index = 0; index < ARRAY_SIZE(matching_array); index++) { 3248c1bb86cdSEric Blake classesToMatch = IOServiceMatching(matching_array[index]); 3249c1bb86cdSEric Blake if (classesToMatch == NULL) { 3250c1bb86cdSEric Blake error_report("IOServiceMatching returned NULL for %s", 3251c1bb86cdSEric Blake matching_array[index]); 3252c1bb86cdSEric Blake continue; 3253c1bb86cdSEric Blake } 3254c1bb86cdSEric Blake CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), 3255c1bb86cdSEric Blake kCFBooleanTrue); 3256c1bb86cdSEric Blake kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, 3257c1bb86cdSEric Blake mediaIterator); 3258c1bb86cdSEric Blake if (kernResult != KERN_SUCCESS) { 3259c1bb86cdSEric Blake error_report("Note: IOServiceGetMatchingServices returned %d", 3260c1bb86cdSEric Blake kernResult); 3261c1bb86cdSEric Blake continue; 3262c1bb86cdSEric Blake } 3263c1bb86cdSEric Blake 3264c1bb86cdSEric Blake /* If a match was found, leave the loop */ 3265c1bb86cdSEric Blake if (*mediaIterator != 0) { 32664f7d28d7SLaurent Vivier trace_file_FindEjectableOpticalMedia(matching_array[index]); 3267c1bb86cdSEric Blake mediaType = g_strdup(matching_array[index]); 3268c1bb86cdSEric Blake break; 3269c1bb86cdSEric Blake } 3270c1bb86cdSEric Blake } 3271c1bb86cdSEric Blake return mediaType; 3272c1bb86cdSEric Blake } 3273c1bb86cdSEric Blake 3274c1bb86cdSEric Blake kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3275c1bb86cdSEric Blake CFIndex maxPathSize, int flags) 3276c1bb86cdSEric Blake { 3277c1bb86cdSEric Blake io_object_t nextMedia; 3278c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 3279c1bb86cdSEric Blake *bsdPath = '\0'; 3280c1bb86cdSEric Blake nextMedia = IOIteratorNext( mediaIterator ); 3281c1bb86cdSEric Blake if ( nextMedia ) 3282c1bb86cdSEric Blake { 3283c1bb86cdSEric Blake CFTypeRef bsdPathAsCFString; 3284c1bb86cdSEric Blake bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); 3285c1bb86cdSEric Blake if ( bsdPathAsCFString ) { 3286c1bb86cdSEric Blake size_t devPathLength; 3287c1bb86cdSEric Blake strcpy( bsdPath, _PATH_DEV ); 3288c1bb86cdSEric Blake if (flags & BDRV_O_NOCACHE) { 3289c1bb86cdSEric Blake strcat(bsdPath, "r"); 3290c1bb86cdSEric Blake } 3291c1bb86cdSEric Blake devPathLength = strlen( bsdPath ); 3292c1bb86cdSEric Blake if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { 3293c1bb86cdSEric Blake kernResult = KERN_SUCCESS; 3294c1bb86cdSEric Blake } 3295c1bb86cdSEric Blake CFRelease( bsdPathAsCFString ); 3296c1bb86cdSEric Blake } 3297c1bb86cdSEric Blake IOObjectRelease( nextMedia ); 3298c1bb86cdSEric Blake } 3299c1bb86cdSEric Blake 3300c1bb86cdSEric Blake return kernResult; 3301c1bb86cdSEric Blake } 3302c1bb86cdSEric Blake 3303c1bb86cdSEric Blake /* Sets up a real cdrom for use in QEMU */ 3304c1bb86cdSEric Blake static bool setup_cdrom(char *bsd_path, Error **errp) 3305c1bb86cdSEric Blake { 3306c1bb86cdSEric Blake int index, num_of_test_partitions = 2, fd; 3307c1bb86cdSEric Blake char test_partition[MAXPATHLEN]; 3308c1bb86cdSEric Blake bool partition_found = false; 3309c1bb86cdSEric Blake 3310c1bb86cdSEric Blake /* look for a working partition */ 3311c1bb86cdSEric Blake for (index = 0; index < num_of_test_partitions; index++) { 3312c1bb86cdSEric Blake snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, 3313c1bb86cdSEric Blake index); 3314b18a24a9SDaniel P. Berrangé fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE, NULL); 3315c1bb86cdSEric Blake if (fd >= 0) { 3316c1bb86cdSEric Blake partition_found = true; 3317c1bb86cdSEric Blake qemu_close(fd); 3318c1bb86cdSEric Blake break; 3319c1bb86cdSEric Blake } 3320c1bb86cdSEric Blake } 3321c1bb86cdSEric Blake 3322c1bb86cdSEric Blake /* if a working partition on the device was not found */ 3323c1bb86cdSEric Blake if (partition_found == false) { 3324c1bb86cdSEric Blake error_setg(errp, "Failed to find a working partition on disc"); 3325c1bb86cdSEric Blake } else { 33264f7d28d7SLaurent Vivier trace_file_setup_cdrom(test_partition); 3327c1bb86cdSEric Blake pstrcpy(bsd_path, MAXPATHLEN, test_partition); 3328c1bb86cdSEric Blake } 3329c1bb86cdSEric Blake return partition_found; 3330c1bb86cdSEric Blake } 3331c1bb86cdSEric Blake 3332c1bb86cdSEric Blake /* Prints directions on mounting and unmounting a device */ 3333c1bb86cdSEric Blake static void print_unmounting_directions(const char *file_name) 3334c1bb86cdSEric Blake { 3335c1bb86cdSEric Blake error_report("If device %s is mounted on the desktop, unmount" 3336c1bb86cdSEric Blake " it first before using it in QEMU", file_name); 3337c1bb86cdSEric Blake error_report("Command to unmount device: diskutil unmountDisk %s", 3338c1bb86cdSEric Blake file_name); 3339c1bb86cdSEric Blake error_report("Command to mount device: diskutil mountDisk %s", file_name); 3340c1bb86cdSEric Blake } 3341c1bb86cdSEric Blake 3342c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3343c1bb86cdSEric Blake 3344c1bb86cdSEric Blake static int hdev_probe_device(const char *filename) 3345c1bb86cdSEric Blake { 3346c1bb86cdSEric Blake struct stat st; 3347c1bb86cdSEric Blake 3348c1bb86cdSEric Blake /* allow a dedicated CD-ROM driver to match with a higher priority */ 3349c1bb86cdSEric Blake if (strstart(filename, "/dev/cdrom", NULL)) 3350c1bb86cdSEric Blake return 50; 3351c1bb86cdSEric Blake 3352c1bb86cdSEric Blake if (stat(filename, &st) >= 0 && 3353c1bb86cdSEric Blake (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 3354c1bb86cdSEric Blake return 100; 3355c1bb86cdSEric Blake } 3356c1bb86cdSEric Blake 3357c1bb86cdSEric Blake return 0; 3358c1bb86cdSEric Blake } 3359c1bb86cdSEric Blake 3360c1bb86cdSEric Blake static void hdev_parse_filename(const char *filename, QDict *options, 3361c1bb86cdSEric Blake Error **errp) 3362c1bb86cdSEric Blake { 336303c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_device:", options); 3364c1bb86cdSEric Blake } 3365c1bb86cdSEric Blake 3366c1bb86cdSEric Blake static bool hdev_is_sg(BlockDriverState *bs) 3367c1bb86cdSEric Blake { 3368c1bb86cdSEric Blake 3369c1bb86cdSEric Blake #if defined(__linux__) 3370c1bb86cdSEric Blake 3371c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3372c1bb86cdSEric Blake struct stat st; 3373c1bb86cdSEric Blake struct sg_scsi_id scsiid; 3374c1bb86cdSEric Blake int sg_version; 3375c1bb86cdSEric Blake int ret; 3376c1bb86cdSEric Blake 3377c1bb86cdSEric Blake if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { 3378c1bb86cdSEric Blake return false; 3379c1bb86cdSEric Blake } 3380c1bb86cdSEric Blake 3381c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); 3382c1bb86cdSEric Blake if (ret < 0) { 3383c1bb86cdSEric Blake return false; 3384c1bb86cdSEric Blake } 3385c1bb86cdSEric Blake 3386c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); 3387c1bb86cdSEric Blake if (ret >= 0) { 33884f7d28d7SLaurent Vivier trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); 3389c1bb86cdSEric Blake return true; 3390c1bb86cdSEric Blake } 3391c1bb86cdSEric Blake 3392c1bb86cdSEric Blake #endif 3393c1bb86cdSEric Blake 3394c1bb86cdSEric Blake return false; 3395c1bb86cdSEric Blake } 3396c1bb86cdSEric Blake 3397c1bb86cdSEric Blake static int hdev_open(BlockDriverState *bs, QDict *options, int flags, 3398c1bb86cdSEric Blake Error **errp) 3399c1bb86cdSEric Blake { 3400c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3401c1bb86cdSEric Blake int ret; 3402c1bb86cdSEric Blake 3403c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3404129c7d1cSMarkus Armbruster /* 3405129c7d1cSMarkus Armbruster * Caution: while qdict_get_str() is fine, getting non-string types 3406129c7d1cSMarkus Armbruster * would require more care. When @options come from -blockdev or 3407129c7d1cSMarkus Armbruster * blockdev_add, its members are typed according to the QAPI 3408129c7d1cSMarkus Armbruster * schema, but when they come from -drive, they're all QString. 3409129c7d1cSMarkus Armbruster */ 3410c1bb86cdSEric Blake const char *filename = qdict_get_str(options, "filename"); 3411c1bb86cdSEric Blake char bsd_path[MAXPATHLEN] = ""; 3412c1bb86cdSEric Blake bool error_occurred = false; 3413c1bb86cdSEric Blake 3414c1bb86cdSEric Blake /* If using a real cdrom */ 3415c1bb86cdSEric Blake if (strcmp(filename, "/dev/cdrom") == 0) { 3416c1bb86cdSEric Blake char *mediaType = NULL; 3417c1bb86cdSEric Blake kern_return_t ret_val; 3418c1bb86cdSEric Blake io_iterator_t mediaIterator = 0; 3419c1bb86cdSEric Blake 3420c1bb86cdSEric Blake mediaType = FindEjectableOpticalMedia(&mediaIterator); 3421c1bb86cdSEric Blake if (mediaType == NULL) { 3422c1bb86cdSEric Blake error_setg(errp, "Please make sure your CD/DVD is in the optical" 3423c1bb86cdSEric Blake " drive"); 3424c1bb86cdSEric Blake error_occurred = true; 3425c1bb86cdSEric Blake goto hdev_open_Mac_error; 3426c1bb86cdSEric Blake } 3427c1bb86cdSEric Blake 3428c1bb86cdSEric Blake ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); 3429c1bb86cdSEric Blake if (ret_val != KERN_SUCCESS) { 3430c1bb86cdSEric Blake error_setg(errp, "Could not get BSD path for optical drive"); 3431c1bb86cdSEric Blake error_occurred = true; 3432c1bb86cdSEric Blake goto hdev_open_Mac_error; 3433c1bb86cdSEric Blake } 3434c1bb86cdSEric Blake 3435c1bb86cdSEric Blake /* If a real optical drive was not found */ 3436c1bb86cdSEric Blake if (bsd_path[0] == '\0') { 3437c1bb86cdSEric Blake error_setg(errp, "Failed to obtain bsd path for optical drive"); 3438c1bb86cdSEric Blake error_occurred = true; 3439c1bb86cdSEric Blake goto hdev_open_Mac_error; 3440c1bb86cdSEric Blake } 3441c1bb86cdSEric Blake 3442c1bb86cdSEric Blake /* If using a cdrom disc and finding a partition on the disc failed */ 3443c1bb86cdSEric Blake if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && 3444c1bb86cdSEric Blake setup_cdrom(bsd_path, errp) == false) { 3445c1bb86cdSEric Blake print_unmounting_directions(bsd_path); 3446c1bb86cdSEric Blake error_occurred = true; 3447c1bb86cdSEric Blake goto hdev_open_Mac_error; 3448c1bb86cdSEric Blake } 3449c1bb86cdSEric Blake 345046f5ac20SEric Blake qdict_put_str(options, "filename", bsd_path); 3451c1bb86cdSEric Blake 3452c1bb86cdSEric Blake hdev_open_Mac_error: 3453c1bb86cdSEric Blake g_free(mediaType); 3454c1bb86cdSEric Blake if (mediaIterator) { 3455c1bb86cdSEric Blake IOObjectRelease(mediaIterator); 3456c1bb86cdSEric Blake } 3457c1bb86cdSEric Blake if (error_occurred) { 3458c1bb86cdSEric Blake return -ENOENT; 3459c1bb86cdSEric Blake } 3460c1bb86cdSEric Blake } 3461c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3462c1bb86cdSEric Blake 3463c1bb86cdSEric Blake s->type = FTYPE_FILE; 3464c1bb86cdSEric Blake 3465668f62ecSMarkus Armbruster ret = raw_open_common(bs, options, flags, 0, true, errp); 3466c1bb86cdSEric Blake if (ret < 0) { 3467c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3468c1bb86cdSEric Blake if (*bsd_path) { 3469c1bb86cdSEric Blake filename = bsd_path; 3470c1bb86cdSEric Blake } 3471c1bb86cdSEric Blake /* if a physical device experienced an error while being opened */ 3472c1bb86cdSEric Blake if (strncmp(filename, "/dev/", 5) == 0) { 3473c1bb86cdSEric Blake print_unmounting_directions(filename); 3474c1bb86cdSEric Blake } 3475c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3476c1bb86cdSEric Blake return ret; 3477c1bb86cdSEric Blake } 3478c1bb86cdSEric Blake 3479c1bb86cdSEric Blake /* Since this does ioctl the device must be already opened */ 3480c1bb86cdSEric Blake bs->sg = hdev_is_sg(bs); 3481c1bb86cdSEric Blake 3482c1bb86cdSEric Blake return ret; 3483c1bb86cdSEric Blake } 3484c1bb86cdSEric Blake 3485c1bb86cdSEric Blake #if defined(__linux__) 34862f3a7ab3SKevin Wolf static int coroutine_fn 34872f3a7ab3SKevin Wolf hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) 3488c1bb86cdSEric Blake { 3489c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 349003425671SKevin Wolf RawPosixAIOData acb; 34912f3a7ab3SKevin Wolf int ret; 3492c1bb86cdSEric Blake 34932f3a7ab3SKevin Wolf ret = fd_open(bs); 34942f3a7ab3SKevin Wolf if (ret < 0) { 34952f3a7ab3SKevin Wolf return ret; 34962f3a7ab3SKevin Wolf } 3497c1bb86cdSEric Blake 34987c9e5276SPaolo Bonzini if (req == SG_IO && s->pr_mgr) { 34997c9e5276SPaolo Bonzini struct sg_io_hdr *io_hdr = buf; 35007c9e5276SPaolo Bonzini if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || 35017c9e5276SPaolo Bonzini io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { 35027c9e5276SPaolo Bonzini return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), 35032f3a7ab3SKevin Wolf s->fd, io_hdr); 35047c9e5276SPaolo Bonzini } 35057c9e5276SPaolo Bonzini } 35067c9e5276SPaolo Bonzini 350703425671SKevin Wolf acb = (RawPosixAIOData) { 350803425671SKevin Wolf .bs = bs, 350903425671SKevin Wolf .aio_type = QEMU_AIO_IOCTL, 351003425671SKevin Wolf .aio_fildes = s->fd, 351103425671SKevin Wolf .aio_offset = 0, 351203425671SKevin Wolf .ioctl = { 351303425671SKevin Wolf .buf = buf, 351403425671SKevin Wolf .cmd = req, 351503425671SKevin Wolf }, 351603425671SKevin Wolf }; 351703425671SKevin Wolf 351803425671SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); 3519c1bb86cdSEric Blake } 3520c1bb86cdSEric Blake #endif /* linux */ 3521c1bb86cdSEric Blake 3522c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs) 3523c1bb86cdSEric Blake { 3524c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3525c1bb86cdSEric Blake 3526c1bb86cdSEric Blake /* this is just to ensure s->fd is sane (its called by io ops) */ 3527c1bb86cdSEric Blake if (s->fd >= 0) 3528c1bb86cdSEric Blake return 0; 3529c1bb86cdSEric Blake return -EIO; 3530c1bb86cdSEric Blake } 3531c1bb86cdSEric Blake 353233d70fb6SKevin Wolf static coroutine_fn int 353333d70fb6SKevin Wolf hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 3534c1bb86cdSEric Blake { 35351c450366SAnton Nefedov BDRVRawState *s = bs->opaque; 353633d70fb6SKevin Wolf int ret; 3537c1bb86cdSEric Blake 353833d70fb6SKevin Wolf ret = fd_open(bs); 353933d70fb6SKevin Wolf if (ret < 0) { 35401c450366SAnton Nefedov raw_account_discard(s, bytes, ret); 354133d70fb6SKevin Wolf return ret; 3542c1bb86cdSEric Blake } 354346ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, true); 3544c1bb86cdSEric Blake } 3545c1bb86cdSEric Blake 3546c1bb86cdSEric Blake static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, 3547f5a5ca79SManos Pitsidianakis int64_t offset, int bytes, BdrvRequestFlags flags) 3548c1bb86cdSEric Blake { 3549c1bb86cdSEric Blake int rc; 3550c1bb86cdSEric Blake 3551c1bb86cdSEric Blake rc = fd_open(bs); 3552c1bb86cdSEric Blake if (rc < 0) { 3553c1bb86cdSEric Blake return rc; 3554c1bb86cdSEric Blake } 355534fa110eSKevin Wolf 35567154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); 3557c1bb86cdSEric Blake } 3558c1bb86cdSEric Blake 3559c1bb86cdSEric Blake static BlockDriver bdrv_host_device = { 3560c1bb86cdSEric Blake .format_name = "host_device", 3561c1bb86cdSEric Blake .protocol_name = "host_device", 3562c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3563c1bb86cdSEric Blake .bdrv_needs_filename = true, 3564c1bb86cdSEric Blake .bdrv_probe_device = hdev_probe_device, 3565c1bb86cdSEric Blake .bdrv_parse_filename = hdev_parse_filename, 3566c1bb86cdSEric Blake .bdrv_file_open = hdev_open, 3567c1bb86cdSEric Blake .bdrv_close = raw_close, 3568c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3569c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3570c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 35715a5e7f8cSMaxim Levitsky .bdrv_co_create_opts = bdrv_co_create_opts_simple, 35725a5e7f8cSMaxim Levitsky .create_opts = &bdrv_create_opts_simple, 35738a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3574dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3575c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, 3576c1bb86cdSEric Blake 3577c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3578c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 357933d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 358033d70fb6SKevin Wolf .bdrv_co_pdiscard = hdev_co_pdiscard, 35811efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 35821efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 3583c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3584c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3585c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3586042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3587c1bb86cdSEric Blake 3588061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3589c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3590c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 3591c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3592c1bb86cdSEric Blake = raw_get_allocated_file_size, 3593d9245599SAnton Nefedov .bdrv_get_specific_stats = hdev_get_specific_stats, 3594244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 3595244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 3596244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 3597c1bb86cdSEric Blake .bdrv_probe_blocksizes = hdev_probe_blocksizes, 3598c1bb86cdSEric Blake .bdrv_probe_geometry = hdev_probe_geometry, 3599c1bb86cdSEric Blake 3600c1bb86cdSEric Blake /* generic scsi device */ 3601c1bb86cdSEric Blake #ifdef __linux__ 36022f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3603c1bb86cdSEric Blake #endif 3604c1bb86cdSEric Blake }; 3605c1bb86cdSEric Blake 3606c1bb86cdSEric Blake #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3607c1bb86cdSEric Blake static void cdrom_parse_filename(const char *filename, QDict *options, 3608c1bb86cdSEric Blake Error **errp) 3609c1bb86cdSEric Blake { 361003c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); 3611c1bb86cdSEric Blake } 3612c1bb86cdSEric Blake #endif 3613c1bb86cdSEric Blake 3614c1bb86cdSEric Blake #ifdef __linux__ 3615c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3616c1bb86cdSEric Blake Error **errp) 3617c1bb86cdSEric Blake { 3618c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3619c1bb86cdSEric Blake 3620c1bb86cdSEric Blake s->type = FTYPE_CD; 3621c1bb86cdSEric Blake 3622c1bb86cdSEric Blake /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ 3623230ff739SJohn Snow return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); 3624c1bb86cdSEric Blake } 3625c1bb86cdSEric Blake 3626c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3627c1bb86cdSEric Blake { 3628c1bb86cdSEric Blake int fd, ret; 3629c1bb86cdSEric Blake int prio = 0; 3630c1bb86cdSEric Blake struct stat st; 3631c1bb86cdSEric Blake 3632b18a24a9SDaniel P. Berrangé fd = qemu_open(filename, O_RDONLY | O_NONBLOCK, NULL); 3633c1bb86cdSEric Blake if (fd < 0) { 3634c1bb86cdSEric Blake goto out; 3635c1bb86cdSEric Blake } 3636c1bb86cdSEric Blake ret = fstat(fd, &st); 3637c1bb86cdSEric Blake if (ret == -1 || !S_ISBLK(st.st_mode)) { 3638c1bb86cdSEric Blake goto outc; 3639c1bb86cdSEric Blake } 3640c1bb86cdSEric Blake 3641c1bb86cdSEric Blake /* Attempt to detect via a CDROM specific ioctl */ 3642c1bb86cdSEric Blake ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3643c1bb86cdSEric Blake if (ret >= 0) 3644c1bb86cdSEric Blake prio = 100; 3645c1bb86cdSEric Blake 3646c1bb86cdSEric Blake outc: 3647c1bb86cdSEric Blake qemu_close(fd); 3648c1bb86cdSEric Blake out: 3649c1bb86cdSEric Blake return prio; 3650c1bb86cdSEric Blake } 3651c1bb86cdSEric Blake 3652c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3653c1bb86cdSEric Blake { 3654c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3655c1bb86cdSEric Blake int ret; 3656c1bb86cdSEric Blake 3657c1bb86cdSEric Blake ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3658c1bb86cdSEric Blake return ret == CDS_DISC_OK; 3659c1bb86cdSEric Blake } 3660c1bb86cdSEric Blake 3661c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3662c1bb86cdSEric Blake { 3663c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3664c1bb86cdSEric Blake 3665c1bb86cdSEric Blake if (eject_flag) { 3666c1bb86cdSEric Blake if (ioctl(s->fd, CDROMEJECT, NULL) < 0) 3667c1bb86cdSEric Blake perror("CDROMEJECT"); 3668c1bb86cdSEric Blake } else { 3669c1bb86cdSEric Blake if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) 3670c1bb86cdSEric Blake perror("CDROMEJECT"); 3671c1bb86cdSEric Blake } 3672c1bb86cdSEric Blake } 3673c1bb86cdSEric Blake 3674c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3675c1bb86cdSEric Blake { 3676c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3677c1bb86cdSEric Blake 3678c1bb86cdSEric Blake if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { 3679c1bb86cdSEric Blake /* 3680c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3681c1bb86cdSEric Blake * mounts the CD-ROM 3682c1bb86cdSEric Blake */ 3683c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3684c1bb86cdSEric Blake } 3685c1bb86cdSEric Blake } 3686c1bb86cdSEric Blake 3687c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3688c1bb86cdSEric Blake .format_name = "host_cdrom", 3689c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3690c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3691c1bb86cdSEric Blake .bdrv_needs_filename = true, 3692c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3693c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3694c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3695c1bb86cdSEric Blake .bdrv_close = raw_close, 3696c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3697c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3698c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 36995a5e7f8cSMaxim Levitsky .bdrv_co_create_opts = bdrv_co_create_opts_simple, 37005a5e7f8cSMaxim Levitsky .create_opts = &bdrv_create_opts_simple, 37018a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3702dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3703c1bb86cdSEric Blake 3704c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3705c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 370633d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3707c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3708c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3709c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3710042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3711c1bb86cdSEric Blake 3712061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3713c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3714c1bb86cdSEric Blake .has_variable_length = true, 3715c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3716c1bb86cdSEric Blake = raw_get_allocated_file_size, 3717c1bb86cdSEric Blake 3718c1bb86cdSEric Blake /* removable device support */ 3719c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3720c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3721c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3722c1bb86cdSEric Blake 3723c1bb86cdSEric Blake /* generic scsi device */ 37242f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3725c1bb86cdSEric Blake }; 3726c1bb86cdSEric Blake #endif /* __linux__ */ 3727c1bb86cdSEric Blake 3728c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 3729c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3730c1bb86cdSEric Blake Error **errp) 3731c1bb86cdSEric Blake { 3732c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3733c1bb86cdSEric Blake int ret; 3734c1bb86cdSEric Blake 3735c1bb86cdSEric Blake s->type = FTYPE_CD; 3736c1bb86cdSEric Blake 3737668f62ecSMarkus Armbruster ret = raw_open_common(bs, options, flags, 0, true, errp); 3738c1bb86cdSEric Blake if (ret) { 3739c1bb86cdSEric Blake return ret; 3740c1bb86cdSEric Blake } 3741c1bb86cdSEric Blake 3742c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3743c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3744c1bb86cdSEric Blake return 0; 3745c1bb86cdSEric Blake } 3746c1bb86cdSEric Blake 3747c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3748c1bb86cdSEric Blake { 3749c1bb86cdSEric Blake if (strstart(filename, "/dev/cd", NULL) || 3750c1bb86cdSEric Blake strstart(filename, "/dev/acd", NULL)) 3751c1bb86cdSEric Blake return 100; 3752c1bb86cdSEric Blake return 0; 3753c1bb86cdSEric Blake } 3754c1bb86cdSEric Blake 3755c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs) 3756c1bb86cdSEric Blake { 3757c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3758c1bb86cdSEric Blake int fd; 3759c1bb86cdSEric Blake 3760c1bb86cdSEric Blake /* 3761c1bb86cdSEric Blake * Force reread of possibly changed/newly loaded disc, 3762c1bb86cdSEric Blake * FreeBSD seems to not notice sometimes... 3763c1bb86cdSEric Blake */ 3764c1bb86cdSEric Blake if (s->fd >= 0) 3765c1bb86cdSEric Blake qemu_close(s->fd); 3766b18a24a9SDaniel P. Berrangé fd = qemu_open(bs->filename, s->open_flags, NULL); 3767c1bb86cdSEric Blake if (fd < 0) { 3768c1bb86cdSEric Blake s->fd = -1; 3769c1bb86cdSEric Blake return -EIO; 3770c1bb86cdSEric Blake } 3771c1bb86cdSEric Blake s->fd = fd; 3772c1bb86cdSEric Blake 3773c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3774c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3775c1bb86cdSEric Blake return 0; 3776c1bb86cdSEric Blake } 3777c1bb86cdSEric Blake 3778c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3779c1bb86cdSEric Blake { 3780c1bb86cdSEric Blake return raw_getlength(bs) > 0; 3781c1bb86cdSEric Blake } 3782c1bb86cdSEric Blake 3783c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3784c1bb86cdSEric Blake { 3785c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3786c1bb86cdSEric Blake 3787c1bb86cdSEric Blake if (s->fd < 0) 3788c1bb86cdSEric Blake return; 3789c1bb86cdSEric Blake 3790c1bb86cdSEric Blake (void) ioctl(s->fd, CDIOCALLOW); 3791c1bb86cdSEric Blake 3792c1bb86cdSEric Blake if (eject_flag) { 3793c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCEJECT) < 0) 3794c1bb86cdSEric Blake perror("CDIOCEJECT"); 3795c1bb86cdSEric Blake } else { 3796c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCCLOSE) < 0) 3797c1bb86cdSEric Blake perror("CDIOCCLOSE"); 3798c1bb86cdSEric Blake } 3799c1bb86cdSEric Blake 3800c1bb86cdSEric Blake cdrom_reopen(bs); 3801c1bb86cdSEric Blake } 3802c1bb86cdSEric Blake 3803c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3804c1bb86cdSEric Blake { 3805c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3806c1bb86cdSEric Blake 3807c1bb86cdSEric Blake if (s->fd < 0) 3808c1bb86cdSEric Blake return; 3809c1bb86cdSEric Blake if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { 3810c1bb86cdSEric Blake /* 3811c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3812c1bb86cdSEric Blake * mounts the CD-ROM 3813c1bb86cdSEric Blake */ 3814c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3815c1bb86cdSEric Blake } 3816c1bb86cdSEric Blake } 3817c1bb86cdSEric Blake 3818c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3819c1bb86cdSEric Blake .format_name = "host_cdrom", 3820c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3821c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3822c1bb86cdSEric Blake .bdrv_needs_filename = true, 3823c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3824c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3825c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3826c1bb86cdSEric Blake .bdrv_close = raw_close, 3827c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3828c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3829c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 38305a5e7f8cSMaxim Levitsky .bdrv_co_create_opts = bdrv_co_create_opts_simple, 38315a5e7f8cSMaxim Levitsky .create_opts = &bdrv_create_opts_simple, 38328a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3833c1bb86cdSEric Blake 3834c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3835c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 383633d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3837c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3838c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3839c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3840042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3841c1bb86cdSEric Blake 3842061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3843c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3844c1bb86cdSEric Blake .has_variable_length = true, 3845c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3846c1bb86cdSEric Blake = raw_get_allocated_file_size, 3847c1bb86cdSEric Blake 3848c1bb86cdSEric Blake /* removable device support */ 3849c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3850c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3851c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3852c1bb86cdSEric Blake }; 3853c1bb86cdSEric Blake #endif /* __FreeBSD__ */ 3854c1bb86cdSEric Blake 3855c1bb86cdSEric Blake static void bdrv_file_init(void) 3856c1bb86cdSEric Blake { 3857c1bb86cdSEric Blake /* 3858c1bb86cdSEric Blake * Register all the drivers. Note that order is important, the driver 3859c1bb86cdSEric Blake * registered last will get probed first. 3860c1bb86cdSEric Blake */ 3861c1bb86cdSEric Blake bdrv_register(&bdrv_file); 3862c1bb86cdSEric Blake bdrv_register(&bdrv_host_device); 3863c1bb86cdSEric Blake #ifdef __linux__ 3864c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3865c1bb86cdSEric Blake #endif 3866c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3867c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3868c1bb86cdSEric Blake #endif 3869c1bb86cdSEric Blake } 3870c1bb86cdSEric Blake 3871c1bb86cdSEric Blake block_init(bdrv_file_init); 3872