1c1bb86cdSEric Blake /* 2c1bb86cdSEric Blake * Block driver for RAW files (posix) 3c1bb86cdSEric Blake * 4c1bb86cdSEric Blake * Copyright (c) 2006 Fabrice Bellard 5c1bb86cdSEric Blake * 6c1bb86cdSEric Blake * Permission is hereby granted, free of charge, to any person obtaining a copy 7c1bb86cdSEric Blake * of this software and associated documentation files (the "Software"), to deal 8c1bb86cdSEric Blake * in the Software without restriction, including without limitation the rights 9c1bb86cdSEric Blake * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 10c1bb86cdSEric Blake * copies of the Software, and to permit persons to whom the Software is 11c1bb86cdSEric Blake * furnished to do so, subject to the following conditions: 12c1bb86cdSEric Blake * 13c1bb86cdSEric Blake * The above copyright notice and this permission notice shall be included in 14c1bb86cdSEric Blake * all copies or substantial portions of the Software. 15c1bb86cdSEric Blake * 16c1bb86cdSEric Blake * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17c1bb86cdSEric Blake * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18c1bb86cdSEric Blake * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19c1bb86cdSEric Blake * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 20c1bb86cdSEric Blake * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 21c1bb86cdSEric Blake * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 22c1bb86cdSEric Blake * THE SOFTWARE. 23c1bb86cdSEric Blake */ 24922a01a0SMarkus Armbruster 25c1bb86cdSEric Blake #include "qemu/osdep.h" 26a8d25326SMarkus Armbruster #include "qemu-common.h" 27c1bb86cdSEric Blake #include "qapi/error.h" 28c1bb86cdSEric Blake #include "qemu/cutils.h" 29c1bb86cdSEric Blake #include "qemu/error-report.h" 30c1bb86cdSEric Blake #include "block/block_int.h" 31c1bb86cdSEric Blake #include "qemu/module.h" 32922a01a0SMarkus Armbruster #include "qemu/option.h" 33ffa244c8SKevin Wolf #include "qemu/units.h" 34c1bb86cdSEric Blake #include "trace.h" 35c1bb86cdSEric Blake #include "block/thread-pool.h" 36c1bb86cdSEric Blake #include "qemu/iov.h" 37c1bb86cdSEric Blake #include "block/raw-aio.h" 38452fcdbcSMarkus Armbruster #include "qapi/qmp/qdict.h" 39c1bb86cdSEric Blake #include "qapi/qmp/qstring.h" 40c1bb86cdSEric Blake 417c9e5276SPaolo Bonzini #include "scsi/pr-manager.h" 427c9e5276SPaolo Bonzini #include "scsi/constants.h" 437c9e5276SPaolo Bonzini 44c1bb86cdSEric Blake #if defined(__APPLE__) && (__MACH__) 45c1bb86cdSEric Blake #include <paths.h> 46c1bb86cdSEric Blake #include <sys/param.h> 47c1bb86cdSEric Blake #include <IOKit/IOKitLib.h> 48c1bb86cdSEric Blake #include <IOKit/IOBSD.h> 49c1bb86cdSEric Blake #include <IOKit/storage/IOMediaBSDClient.h> 50c1bb86cdSEric Blake #include <IOKit/storage/IOMedia.h> 51c1bb86cdSEric Blake #include <IOKit/storage/IOCDMedia.h> 52c1bb86cdSEric Blake //#include <IOKit/storage/IOCDTypes.h> 53c1bb86cdSEric Blake #include <IOKit/storage/IODVDMedia.h> 54c1bb86cdSEric Blake #include <CoreFoundation/CoreFoundation.h> 55c1bb86cdSEric Blake #endif 56c1bb86cdSEric Blake 57c1bb86cdSEric Blake #ifdef __sun__ 58c1bb86cdSEric Blake #define _POSIX_PTHREAD_SEMANTICS 1 59c1bb86cdSEric Blake #include <sys/dkio.h> 60c1bb86cdSEric Blake #endif 61c1bb86cdSEric Blake #ifdef __linux__ 62c1bb86cdSEric Blake #include <sys/ioctl.h> 63c1bb86cdSEric Blake #include <sys/param.h> 641efad060SFam Zheng #include <sys/syscall.h> 655edc8557SKevin Wolf #include <sys/vfs.h> 66c1bb86cdSEric Blake #include <linux/cdrom.h> 67c1bb86cdSEric Blake #include <linux/fd.h> 68c1bb86cdSEric Blake #include <linux/fs.h> 69c1bb86cdSEric Blake #include <linux/hdreg.h> 705edc8557SKevin Wolf #include <linux/magic.h> 71c1bb86cdSEric Blake #include <scsi/sg.h> 72c1bb86cdSEric Blake #ifdef __s390__ 73c1bb86cdSEric Blake #include <asm/dasd.h> 74c1bb86cdSEric Blake #endif 75c1bb86cdSEric Blake #ifndef FS_NOCOW_FL 76c1bb86cdSEric Blake #define FS_NOCOW_FL 0x00800000 /* Do not cow file */ 77c1bb86cdSEric Blake #endif 78c1bb86cdSEric Blake #endif 79c1bb86cdSEric Blake #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE) 80c1bb86cdSEric Blake #include <linux/falloc.h> 81c1bb86cdSEric Blake #endif 82c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 83c1bb86cdSEric Blake #include <sys/disk.h> 84c1bb86cdSEric Blake #include <sys/cdio.h> 85c1bb86cdSEric Blake #endif 86c1bb86cdSEric Blake 87c1bb86cdSEric Blake #ifdef __OpenBSD__ 88c1bb86cdSEric Blake #include <sys/ioctl.h> 89c1bb86cdSEric Blake #include <sys/disklabel.h> 90c1bb86cdSEric Blake #include <sys/dkio.h> 91c1bb86cdSEric Blake #endif 92c1bb86cdSEric Blake 93c1bb86cdSEric Blake #ifdef __NetBSD__ 94c1bb86cdSEric Blake #include <sys/ioctl.h> 95c1bb86cdSEric Blake #include <sys/disklabel.h> 96c1bb86cdSEric Blake #include <sys/dkio.h> 97c1bb86cdSEric Blake #include <sys/disk.h> 98c1bb86cdSEric Blake #endif 99c1bb86cdSEric Blake 100c1bb86cdSEric Blake #ifdef __DragonFly__ 101c1bb86cdSEric Blake #include <sys/ioctl.h> 102c1bb86cdSEric Blake #include <sys/diskslice.h> 103c1bb86cdSEric Blake #endif 104c1bb86cdSEric Blake 105c1bb86cdSEric Blake #ifdef CONFIG_XFS 106c1bb86cdSEric Blake #include <xfs/xfs.h> 107c1bb86cdSEric Blake #endif 108c1bb86cdSEric Blake 1094f7d28d7SLaurent Vivier #include "trace.h" 110c1bb86cdSEric Blake 111c1bb86cdSEric Blake /* OS X does not have O_DSYNC */ 112c1bb86cdSEric Blake #ifndef O_DSYNC 113c1bb86cdSEric Blake #ifdef O_SYNC 114c1bb86cdSEric Blake #define O_DSYNC O_SYNC 115c1bb86cdSEric Blake #elif defined(O_FSYNC) 116c1bb86cdSEric Blake #define O_DSYNC O_FSYNC 117c1bb86cdSEric Blake #endif 118c1bb86cdSEric Blake #endif 119c1bb86cdSEric Blake 120c1bb86cdSEric Blake /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ 121c1bb86cdSEric Blake #ifndef O_DIRECT 122c1bb86cdSEric Blake #define O_DIRECT O_DSYNC 123c1bb86cdSEric Blake #endif 124c1bb86cdSEric Blake 125c1bb86cdSEric Blake #define FTYPE_FILE 0 126c1bb86cdSEric Blake #define FTYPE_CD 1 127c1bb86cdSEric Blake 128c1bb86cdSEric Blake #define MAX_BLOCKSIZE 4096 129c1bb86cdSEric Blake 130244a5668SFam Zheng /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes, 131244a5668SFam Zheng * leaving a few more bytes for its future use. */ 132244a5668SFam Zheng #define RAW_LOCK_PERM_BASE 100 133244a5668SFam Zheng #define RAW_LOCK_SHARED_BASE 200 134244a5668SFam Zheng 135c1bb86cdSEric Blake typedef struct BDRVRawState { 136c1bb86cdSEric Blake int fd; 137244a5668SFam Zheng bool use_lock; 138c1bb86cdSEric Blake int type; 139c1bb86cdSEric Blake int open_flags; 140c1bb86cdSEric Blake size_t buf_align; 141c1bb86cdSEric Blake 142244a5668SFam Zheng /* The current permissions. */ 143244a5668SFam Zheng uint64_t perm; 144244a5668SFam Zheng uint64_t shared_perm; 145244a5668SFam Zheng 1462996ffadSFam Zheng /* The perms bits whose corresponding bytes are already locked in 147f2e3af29SFam Zheng * s->fd. */ 1482996ffadSFam Zheng uint64_t locked_perm; 1492996ffadSFam Zheng uint64_t locked_shared_perm; 1502996ffadSFam Zheng 1516ceabe6fSKevin Wolf int perm_change_fd; 152094e3639SMax Reitz int perm_change_flags; 153e0c9cf3aSKevin Wolf BDRVReopenState *reopen_state; 154e0c9cf3aSKevin Wolf 155c1bb86cdSEric Blake #ifdef CONFIG_XFS 156c1bb86cdSEric Blake bool is_xfs:1; 157c1bb86cdSEric Blake #endif 158c1bb86cdSEric Blake bool has_discard:1; 159c1bb86cdSEric Blake bool has_write_zeroes:1; 160c1bb86cdSEric Blake bool discard_zeroes:1; 161c1bb86cdSEric Blake bool use_linux_aio:1; 162c6447510SAarushi Mehta bool use_linux_io_uring:1; 163e5bcf967SKevin Wolf bool page_cache_inconsistent:1; 164c1bb86cdSEric Blake bool has_fallocate; 165c1bb86cdSEric Blake bool needs_alignment; 166f357fcd8SStefan Hajnoczi bool drop_cache; 16731be8a2aSStefan Hajnoczi bool check_cache_dropped; 1681c450366SAnton Nefedov struct { 1691c450366SAnton Nefedov uint64_t discard_nb_ok; 1701c450366SAnton Nefedov uint64_t discard_nb_failed; 1711c450366SAnton Nefedov uint64_t discard_bytes_ok; 1721c450366SAnton Nefedov } stats; 1737c9e5276SPaolo Bonzini 1747c9e5276SPaolo Bonzini PRManager *pr_mgr; 175c1bb86cdSEric Blake } BDRVRawState; 176c1bb86cdSEric Blake 177c1bb86cdSEric Blake typedef struct BDRVRawReopenState { 178c1bb86cdSEric Blake int fd; 179c1bb86cdSEric Blake int open_flags; 180f357fcd8SStefan Hajnoczi bool drop_cache; 18131be8a2aSStefan Hajnoczi bool check_cache_dropped; 182c1bb86cdSEric Blake } BDRVRawReopenState; 183c1bb86cdSEric Blake 184c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs); 185c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs); 186c1bb86cdSEric Blake 187c1bb86cdSEric Blake typedef struct RawPosixAIOData { 188c1bb86cdSEric Blake BlockDriverState *bs; 189c1bb86cdSEric Blake int aio_type; 190d57c44d0SKevin Wolf int aio_fildes; 191d57c44d0SKevin Wolf 192d57c44d0SKevin Wolf off_t aio_offset; 193d57c44d0SKevin Wolf uint64_t aio_nbytes; 194d57c44d0SKevin Wolf 19593f4e2ffSKevin Wolf union { 19693f4e2ffSKevin Wolf struct { 197d57c44d0SKevin Wolf struct iovec *iov; 198d57c44d0SKevin Wolf int niov; 199d57c44d0SKevin Wolf } io; 200d57c44d0SKevin Wolf struct { 201d57c44d0SKevin Wolf uint64_t cmd; 202d57c44d0SKevin Wolf void *buf; 203d57c44d0SKevin Wolf } ioctl; 204d57c44d0SKevin Wolf struct { 2051efad060SFam Zheng int aio_fd2; 2061efad060SFam Zheng off_t aio_offset2; 207d57c44d0SKevin Wolf } copy_range; 20893f4e2ffSKevin Wolf struct { 20993f4e2ffSKevin Wolf PreallocMode prealloc; 21093f4e2ffSKevin Wolf Error **errp; 211d57c44d0SKevin Wolf } truncate; 21293f4e2ffSKevin Wolf }; 213c1bb86cdSEric Blake } RawPosixAIOData; 214c1bb86cdSEric Blake 215c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 216c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs); 217c1bb86cdSEric Blake #endif 218c1bb86cdSEric Blake 219c1bb86cdSEric Blake #if defined(__NetBSD__) 220db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 221c1bb86cdSEric Blake { 222c1bb86cdSEric Blake static char namebuf[PATH_MAX]; 223c1bb86cdSEric Blake const char *dp, *fname; 224c1bb86cdSEric Blake struct stat sb; 225c1bb86cdSEric Blake 226c1bb86cdSEric Blake fname = *filename; 227c1bb86cdSEric Blake dp = strrchr(fname, '/'); 228c1bb86cdSEric Blake if (lstat(fname, &sb) < 0) { 229f6fc1e30SPaolo Bonzini error_setg_file_open(errp, errno, fname); 230c1bb86cdSEric Blake return -errno; 231c1bb86cdSEric Blake } 232c1bb86cdSEric Blake 233c1bb86cdSEric Blake if (!S_ISBLK(sb.st_mode)) { 234c1bb86cdSEric Blake return 0; 235c1bb86cdSEric Blake } 236c1bb86cdSEric Blake 237c1bb86cdSEric Blake if (dp == NULL) { 238c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "r%s", fname); 239c1bb86cdSEric Blake } else { 240c1bb86cdSEric Blake snprintf(namebuf, PATH_MAX, "%.*s/r%s", 241c1bb86cdSEric Blake (int)(dp - fname), fname, dp + 1); 242c1bb86cdSEric Blake } 243c1bb86cdSEric Blake *filename = namebuf; 244db0754dfSFam Zheng warn_report("%s is a block device, using %s", fname, *filename); 245c1bb86cdSEric Blake 246c1bb86cdSEric Blake return 0; 247c1bb86cdSEric Blake } 248c1bb86cdSEric Blake #else 249db0754dfSFam Zheng static int raw_normalize_devicepath(const char **filename, Error **errp) 250c1bb86cdSEric Blake { 251c1bb86cdSEric Blake return 0; 252c1bb86cdSEric Blake } 253c1bb86cdSEric Blake #endif 254c1bb86cdSEric Blake 255c1bb86cdSEric Blake /* 256c1bb86cdSEric Blake * Get logical block size via ioctl. On success store it in @sector_size_p. 257c1bb86cdSEric Blake */ 258c1bb86cdSEric Blake static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) 259c1bb86cdSEric Blake { 260c1bb86cdSEric Blake unsigned int sector_size; 261c1bb86cdSEric Blake bool success = false; 262700f9ce0SPeter Maydell int i; 263c1bb86cdSEric Blake 264c1bb86cdSEric Blake errno = ENOTSUP; 265700f9ce0SPeter Maydell static const unsigned long ioctl_list[] = { 266c1bb86cdSEric Blake #ifdef BLKSSZGET 267700f9ce0SPeter Maydell BLKSSZGET, 268c1bb86cdSEric Blake #endif 269c1bb86cdSEric Blake #ifdef DKIOCGETBLOCKSIZE 270700f9ce0SPeter Maydell DKIOCGETBLOCKSIZE, 271c1bb86cdSEric Blake #endif 272c1bb86cdSEric Blake #ifdef DIOCGSECTORSIZE 273700f9ce0SPeter Maydell DIOCGSECTORSIZE, 274700f9ce0SPeter Maydell #endif 275700f9ce0SPeter Maydell }; 276700f9ce0SPeter Maydell 277700f9ce0SPeter Maydell /* Try a few ioctls to get the right size */ 278700f9ce0SPeter Maydell for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { 279700f9ce0SPeter Maydell if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { 280c1bb86cdSEric Blake *sector_size_p = sector_size; 281c1bb86cdSEric Blake success = true; 282c1bb86cdSEric Blake } 283700f9ce0SPeter Maydell } 284c1bb86cdSEric Blake 285c1bb86cdSEric Blake return success ? 0 : -errno; 286c1bb86cdSEric Blake } 287c1bb86cdSEric Blake 288c1bb86cdSEric Blake /** 289c1bb86cdSEric Blake * Get physical block size of @fd. 290c1bb86cdSEric Blake * On success, store it in @blk_size and return 0. 291c1bb86cdSEric Blake * On failure, return -errno. 292c1bb86cdSEric Blake */ 293c1bb86cdSEric Blake static int probe_physical_blocksize(int fd, unsigned int *blk_size) 294c1bb86cdSEric Blake { 295c1bb86cdSEric Blake #ifdef BLKPBSZGET 296c1bb86cdSEric Blake if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { 297c1bb86cdSEric Blake return -errno; 298c1bb86cdSEric Blake } 299c1bb86cdSEric Blake return 0; 300c1bb86cdSEric Blake #else 301c1bb86cdSEric Blake return -ENOTSUP; 302c1bb86cdSEric Blake #endif 303c1bb86cdSEric Blake } 304c1bb86cdSEric Blake 3055edc8557SKevin Wolf /* 3065edc8557SKevin Wolf * Returns true if no alignment restrictions are necessary even for files 3075edc8557SKevin Wolf * opened with O_DIRECT. 3085edc8557SKevin Wolf * 3095edc8557SKevin Wolf * raw_probe_alignment() probes the required alignment and assume that 1 means 3105edc8557SKevin Wolf * the probing failed, so it falls back to a safe default of 4k. This can be 3115edc8557SKevin Wolf * avoided if we know that byte alignment is okay for the file. 3125edc8557SKevin Wolf */ 3135edc8557SKevin Wolf static bool dio_byte_aligned(int fd) 3145edc8557SKevin Wolf { 3155edc8557SKevin Wolf #ifdef __linux__ 3165edc8557SKevin Wolf struct statfs buf; 3175edc8557SKevin Wolf int ret; 3185edc8557SKevin Wolf 3195edc8557SKevin Wolf ret = fstatfs(fd, &buf); 3205edc8557SKevin Wolf if (ret == 0 && buf.f_type == NFS_SUPER_MAGIC) { 3215edc8557SKevin Wolf return true; 3225edc8557SKevin Wolf } 3235edc8557SKevin Wolf #endif 3245edc8557SKevin Wolf return false; 3255edc8557SKevin Wolf } 3265edc8557SKevin Wolf 327c1bb86cdSEric Blake /* Check if read is allowed with given memory buffer and length. 328c1bb86cdSEric Blake * 329c1bb86cdSEric Blake * This function is used to check O_DIRECT memory buffer and request alignment. 330c1bb86cdSEric Blake */ 331c1bb86cdSEric Blake static bool raw_is_io_aligned(int fd, void *buf, size_t len) 332c1bb86cdSEric Blake { 333c1bb86cdSEric Blake ssize_t ret = pread(fd, buf, len, 0); 334c1bb86cdSEric Blake 335c1bb86cdSEric Blake if (ret >= 0) { 336c1bb86cdSEric Blake return true; 337c1bb86cdSEric Blake } 338c1bb86cdSEric Blake 339c1bb86cdSEric Blake #ifdef __linux__ 340c1bb86cdSEric Blake /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore 341c1bb86cdSEric Blake * other errors (e.g. real I/O error), which could happen on a failed 342c1bb86cdSEric Blake * drive, since we only care about probing alignment. 343c1bb86cdSEric Blake */ 344c1bb86cdSEric Blake if (errno != EINVAL) { 345c1bb86cdSEric Blake return true; 346c1bb86cdSEric Blake } 347c1bb86cdSEric Blake #endif 348c1bb86cdSEric Blake 349c1bb86cdSEric Blake return false; 350c1bb86cdSEric Blake } 351c1bb86cdSEric Blake 352c1bb86cdSEric Blake static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) 353c1bb86cdSEric Blake { 354c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 355c1bb86cdSEric Blake char *buf; 356038adc2fSWei Yang size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); 357a6b257a0SNir Soffer size_t alignments[] = {1, 512, 1024, 2048, 4096}; 358c1bb86cdSEric Blake 359c1bb86cdSEric Blake /* For SCSI generic devices the alignment is not really used. 360c1bb86cdSEric Blake With buffered I/O, we don't have any restrictions. */ 361c1bb86cdSEric Blake if (bdrv_is_sg(bs) || !s->needs_alignment) { 362c1bb86cdSEric Blake bs->bl.request_alignment = 1; 363c1bb86cdSEric Blake s->buf_align = 1; 364c1bb86cdSEric Blake return; 365c1bb86cdSEric Blake } 366c1bb86cdSEric Blake 367c1bb86cdSEric Blake bs->bl.request_alignment = 0; 368c1bb86cdSEric Blake s->buf_align = 0; 369c1bb86cdSEric Blake /* Let's try to use the logical blocksize for the alignment. */ 370c1bb86cdSEric Blake if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { 371c1bb86cdSEric Blake bs->bl.request_alignment = 0; 372c1bb86cdSEric Blake } 373c1bb86cdSEric Blake #ifdef CONFIG_XFS 374c1bb86cdSEric Blake if (s->is_xfs) { 375c1bb86cdSEric Blake struct dioattr da; 376c1bb86cdSEric Blake if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { 377c1bb86cdSEric Blake bs->bl.request_alignment = da.d_miniosz; 378c1bb86cdSEric Blake /* The kernel returns wrong information for d_mem */ 379c1bb86cdSEric Blake /* s->buf_align = da.d_mem; */ 380c1bb86cdSEric Blake } 381c1bb86cdSEric Blake } 382c1bb86cdSEric Blake #endif 383c1bb86cdSEric Blake 384a6b257a0SNir Soffer /* 385a6b257a0SNir Soffer * If we could not get the sizes so far, we can only guess them. First try 386a6b257a0SNir Soffer * to detect request alignment, since it is more likely to succeed. Then 387a6b257a0SNir Soffer * try to detect buf_align, which cannot be detected in some cases (e.g. 388a6b257a0SNir Soffer * Gluster). If buf_align cannot be detected, we fallback to the value of 389a6b257a0SNir Soffer * request_alignment. 390a6b257a0SNir Soffer */ 391a6b257a0SNir Soffer 392a6b257a0SNir Soffer if (!bs->bl.request_alignment) { 393a6b257a0SNir Soffer int i; 394c1bb86cdSEric Blake size_t align; 395a6b257a0SNir Soffer buf = qemu_memalign(max_align, max_align); 396a6b257a0SNir Soffer for (i = 0; i < ARRAY_SIZE(alignments); i++) { 397a6b257a0SNir Soffer align = alignments[i]; 398a6b257a0SNir Soffer if (raw_is_io_aligned(fd, buf, align)) { 399a6b257a0SNir Soffer /* Fallback to safe value. */ 400a6b257a0SNir Soffer bs->bl.request_alignment = (align != 1) ? align : max_align; 401c1bb86cdSEric Blake break; 402c1bb86cdSEric Blake } 403c1bb86cdSEric Blake } 404c1bb86cdSEric Blake qemu_vfree(buf); 405c1bb86cdSEric Blake } 406c1bb86cdSEric Blake 407a6b257a0SNir Soffer if (!s->buf_align) { 408a6b257a0SNir Soffer int i; 409c1bb86cdSEric Blake size_t align; 410a6b257a0SNir Soffer buf = qemu_memalign(max_align, 2 * max_align); 411a6b257a0SNir Soffer for (i = 0; i < ARRAY_SIZE(alignments); i++) { 412a6b257a0SNir Soffer align = alignments[i]; 413a6b257a0SNir Soffer if (raw_is_io_aligned(fd, buf + align, max_align)) { 414236094c7SStefan Hajnoczi /* Fallback to request_alignment. */ 415a6b257a0SNir Soffer s->buf_align = (align != 1) ? align : bs->bl.request_alignment; 416c1bb86cdSEric Blake break; 417c1bb86cdSEric Blake } 418c1bb86cdSEric Blake } 419c1bb86cdSEric Blake qemu_vfree(buf); 420c1bb86cdSEric Blake } 421c1bb86cdSEric Blake 422c1bb86cdSEric Blake if (!s->buf_align || !bs->bl.request_alignment) { 423c1bb86cdSEric Blake error_setg(errp, "Could not find working O_DIRECT alignment"); 424c1bb86cdSEric Blake error_append_hint(errp, "Try cache.direct=off\n"); 425c1bb86cdSEric Blake } 426c1bb86cdSEric Blake } 427c1bb86cdSEric Blake 428*bca5283bSKevin Wolf static int check_hdev_writable(int fd) 42920eaf1bfSKevin Wolf { 43020eaf1bfSKevin Wolf #if defined(BLKROGET) 43120eaf1bfSKevin Wolf /* Linux block devices can be configured "read-only" using blockdev(8). 43220eaf1bfSKevin Wolf * This is independent of device node permissions and therefore open(2) 43320eaf1bfSKevin Wolf * with O_RDWR succeeds. Actual writes fail with EPERM. 43420eaf1bfSKevin Wolf * 43520eaf1bfSKevin Wolf * bdrv_open() is supposed to fail if the disk is read-only. Explicitly 43620eaf1bfSKevin Wolf * check for read-only block devices so that Linux block devices behave 43720eaf1bfSKevin Wolf * properly. 43820eaf1bfSKevin Wolf */ 43920eaf1bfSKevin Wolf struct stat st; 44020eaf1bfSKevin Wolf int readonly = 0; 44120eaf1bfSKevin Wolf 442*bca5283bSKevin Wolf if (fstat(fd, &st)) { 44320eaf1bfSKevin Wolf return -errno; 44420eaf1bfSKevin Wolf } 44520eaf1bfSKevin Wolf 44620eaf1bfSKevin Wolf if (!S_ISBLK(st.st_mode)) { 44720eaf1bfSKevin Wolf return 0; 44820eaf1bfSKevin Wolf } 44920eaf1bfSKevin Wolf 450*bca5283bSKevin Wolf if (ioctl(fd, BLKROGET, &readonly) < 0) { 45120eaf1bfSKevin Wolf return -errno; 45220eaf1bfSKevin Wolf } 45320eaf1bfSKevin Wolf 45420eaf1bfSKevin Wolf if (readonly) { 45520eaf1bfSKevin Wolf return -EACCES; 45620eaf1bfSKevin Wolf } 45720eaf1bfSKevin Wolf #endif /* defined(BLKROGET) */ 45820eaf1bfSKevin Wolf return 0; 45920eaf1bfSKevin Wolf } 46020eaf1bfSKevin Wolf 46123dece19SKevin Wolf static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) 462c1bb86cdSEric Blake { 46323dece19SKevin Wolf bool read_write = false; 464c1bb86cdSEric Blake assert(open_flags != NULL); 465c1bb86cdSEric Blake 466c1bb86cdSEric Blake *open_flags |= O_BINARY; 467c1bb86cdSEric Blake *open_flags &= ~O_ACCMODE; 46823dece19SKevin Wolf 46923dece19SKevin Wolf if (bdrv_flags & BDRV_O_AUTO_RDONLY) { 47023dece19SKevin Wolf read_write = has_writers; 47123dece19SKevin Wolf } else if (bdrv_flags & BDRV_O_RDWR) { 47223dece19SKevin Wolf read_write = true; 47323dece19SKevin Wolf } 47423dece19SKevin Wolf 47523dece19SKevin Wolf if (read_write) { 476c1bb86cdSEric Blake *open_flags |= O_RDWR; 477c1bb86cdSEric Blake } else { 478c1bb86cdSEric Blake *open_flags |= O_RDONLY; 479c1bb86cdSEric Blake } 480c1bb86cdSEric Blake 481c1bb86cdSEric Blake /* Use O_DSYNC for write-through caching, no flags for write-back caching, 482c1bb86cdSEric Blake * and O_DIRECT for no caching. */ 483c1bb86cdSEric Blake if ((bdrv_flags & BDRV_O_NOCACHE)) { 484c1bb86cdSEric Blake *open_flags |= O_DIRECT; 485c1bb86cdSEric Blake } 486c1bb86cdSEric Blake } 487c1bb86cdSEric Blake 488c1bb86cdSEric Blake static void raw_parse_filename(const char *filename, QDict *options, 489c1bb86cdSEric Blake Error **errp) 490c1bb86cdSEric Blake { 49103c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "file:", options); 492c1bb86cdSEric Blake } 493c1bb86cdSEric Blake 494c1bb86cdSEric Blake static QemuOptsList raw_runtime_opts = { 495c1bb86cdSEric Blake .name = "raw", 496c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), 497c1bb86cdSEric Blake .desc = { 498c1bb86cdSEric Blake { 499c1bb86cdSEric Blake .name = "filename", 500c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 501c1bb86cdSEric Blake .help = "File name of the image", 502c1bb86cdSEric Blake }, 503c1bb86cdSEric Blake { 504c1bb86cdSEric Blake .name = "aio", 505c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 506c6447510SAarushi Mehta .help = "host AIO implementation (threads, native, io_uring)", 507c1bb86cdSEric Blake }, 50816b48d5dSFam Zheng { 50916b48d5dSFam Zheng .name = "locking", 51016b48d5dSFam Zheng .type = QEMU_OPT_STRING, 51116b48d5dSFam Zheng .help = "file locking mode (on/off/auto, default: auto)", 51216b48d5dSFam Zheng }, 5137c9e5276SPaolo Bonzini { 5147c9e5276SPaolo Bonzini .name = "pr-manager", 5157c9e5276SPaolo Bonzini .type = QEMU_OPT_STRING, 5167c9e5276SPaolo Bonzini .help = "id of persistent reservation manager object (default: none)", 5177c9e5276SPaolo Bonzini }, 518f357fcd8SStefan Hajnoczi #if defined(__linux__) 519f357fcd8SStefan Hajnoczi { 520f357fcd8SStefan Hajnoczi .name = "drop-cache", 521f357fcd8SStefan Hajnoczi .type = QEMU_OPT_BOOL, 522f357fcd8SStefan Hajnoczi .help = "invalidate page cache during live migration (default: on)", 523f357fcd8SStefan Hajnoczi }, 524f357fcd8SStefan Hajnoczi #endif 52531be8a2aSStefan Hajnoczi { 52631be8a2aSStefan Hajnoczi .name = "x-check-cache-dropped", 52731be8a2aSStefan Hajnoczi .type = QEMU_OPT_BOOL, 52831be8a2aSStefan Hajnoczi .help = "check that page cache was dropped on live migration (default: off)" 52931be8a2aSStefan Hajnoczi }, 530c1bb86cdSEric Blake { /* end of list */ } 531c1bb86cdSEric Blake }, 532c1bb86cdSEric Blake }; 533c1bb86cdSEric Blake 5348a2ce0bcSAlberto Garcia static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; 5358a2ce0bcSAlberto Garcia 536c1bb86cdSEric Blake static int raw_open_common(BlockDriverState *bs, QDict *options, 537230ff739SJohn Snow int bdrv_flags, int open_flags, 538230ff739SJohn Snow bool device, Error **errp) 539c1bb86cdSEric Blake { 540c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 541c1bb86cdSEric Blake QemuOpts *opts; 542c1bb86cdSEric Blake Error *local_err = NULL; 543c1bb86cdSEric Blake const char *filename = NULL; 5447c9e5276SPaolo Bonzini const char *str; 545c1bb86cdSEric Blake BlockdevAioOptions aio, aio_default; 546c1bb86cdSEric Blake int fd, ret; 547c1bb86cdSEric Blake struct stat st; 548244a5668SFam Zheng OnOffAuto locking; 549c1bb86cdSEric Blake 550c1bb86cdSEric Blake opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 551af175e85SMarkus Armbruster if (!qemu_opts_absorb_qdict(opts, options, errp)) { 552c1bb86cdSEric Blake ret = -EINVAL; 553c1bb86cdSEric Blake goto fail; 554c1bb86cdSEric Blake } 555c1bb86cdSEric Blake 556c1bb86cdSEric Blake filename = qemu_opt_get(opts, "filename"); 557c1bb86cdSEric Blake 558db0754dfSFam Zheng ret = raw_normalize_devicepath(&filename, errp); 559c1bb86cdSEric Blake if (ret != 0) { 560c1bb86cdSEric Blake goto fail; 561c1bb86cdSEric Blake } 562c1bb86cdSEric Blake 563c6447510SAarushi Mehta if (bdrv_flags & BDRV_O_NATIVE_AIO) { 564c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE; 565c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 566c6447510SAarushi Mehta } else if (bdrv_flags & BDRV_O_IO_URING) { 567c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING; 568c6447510SAarushi Mehta #endif 569c6447510SAarushi Mehta } else { 570c6447510SAarushi Mehta aio_default = BLOCKDEV_AIO_OPTIONS_THREADS; 571c6447510SAarushi Mehta } 572c6447510SAarushi Mehta 573f7abe0ecSMarc-André Lureau aio = qapi_enum_parse(&BlockdevAioOptions_lookup, 574f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "aio"), 57506c60b6cSMarkus Armbruster aio_default, &local_err); 576c1bb86cdSEric Blake if (local_err) { 577c1bb86cdSEric Blake error_propagate(errp, local_err); 578c1bb86cdSEric Blake ret = -EINVAL; 579c1bb86cdSEric Blake goto fail; 580c1bb86cdSEric Blake } 581c6447510SAarushi Mehta 582c1bb86cdSEric Blake s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); 583c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 584c6447510SAarushi Mehta s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING); 585c6447510SAarushi Mehta #endif 586c1bb86cdSEric Blake 587f7abe0ecSMarc-André Lureau locking = qapi_enum_parse(&OnOffAuto_lookup, 588f7abe0ecSMarc-André Lureau qemu_opt_get(opts, "locking"), 58906c60b6cSMarkus Armbruster ON_OFF_AUTO_AUTO, &local_err); 590244a5668SFam Zheng if (local_err) { 591244a5668SFam Zheng error_propagate(errp, local_err); 592244a5668SFam Zheng ret = -EINVAL; 593244a5668SFam Zheng goto fail; 594244a5668SFam Zheng } 595244a5668SFam Zheng switch (locking) { 596244a5668SFam Zheng case ON_OFF_AUTO_ON: 597244a5668SFam Zheng s->use_lock = true; 5982b218f5dSFam Zheng if (!qemu_has_ofd_lock()) { 599db0754dfSFam Zheng warn_report("File lock requested but OFD locking syscall is " 600db0754dfSFam Zheng "unavailable, falling back to POSIX file locks"); 601db0754dfSFam Zheng error_printf("Due to the implementation, locks can be lost " 6022b218f5dSFam Zheng "unexpectedly.\n"); 6032b218f5dSFam Zheng } 604244a5668SFam Zheng break; 605244a5668SFam Zheng case ON_OFF_AUTO_OFF: 606244a5668SFam Zheng s->use_lock = false; 607244a5668SFam Zheng break; 608244a5668SFam Zheng case ON_OFF_AUTO_AUTO: 6092b218f5dSFam Zheng s->use_lock = qemu_has_ofd_lock(); 610244a5668SFam Zheng break; 611244a5668SFam Zheng default: 612244a5668SFam Zheng abort(); 613244a5668SFam Zheng } 614244a5668SFam Zheng 6157c9e5276SPaolo Bonzini str = qemu_opt_get(opts, "pr-manager"); 6167c9e5276SPaolo Bonzini if (str) { 6177c9e5276SPaolo Bonzini s->pr_mgr = pr_manager_lookup(str, &local_err); 6187c9e5276SPaolo Bonzini if (local_err) { 6197c9e5276SPaolo Bonzini error_propagate(errp, local_err); 6207c9e5276SPaolo Bonzini ret = -EINVAL; 6217c9e5276SPaolo Bonzini goto fail; 6227c9e5276SPaolo Bonzini } 6237c9e5276SPaolo Bonzini } 6247c9e5276SPaolo Bonzini 625f357fcd8SStefan Hajnoczi s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); 62631be8a2aSStefan Hajnoczi s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", 62731be8a2aSStefan Hajnoczi false); 62831be8a2aSStefan Hajnoczi 629c1bb86cdSEric Blake s->open_flags = open_flags; 63023dece19SKevin Wolf raw_parse_flags(bdrv_flags, &s->open_flags, false); 631c1bb86cdSEric Blake 632c1bb86cdSEric Blake s->fd = -1; 633c1bb86cdSEric Blake fd = qemu_open(filename, s->open_flags, 0644); 63464107dc0SKevin Wolf ret = fd < 0 ? -errno : 0; 63564107dc0SKevin Wolf 63664107dc0SKevin Wolf if (ret < 0) { 637f6fc1e30SPaolo Bonzini error_setg_file_open(errp, -ret, filename); 638c1bb86cdSEric Blake if (ret == -EROFS) { 639c1bb86cdSEric Blake ret = -EACCES; 640c1bb86cdSEric Blake } 641c1bb86cdSEric Blake goto fail; 642c1bb86cdSEric Blake } 643c1bb86cdSEric Blake s->fd = fd; 644c1bb86cdSEric Blake 645*bca5283bSKevin Wolf /* Check s->open_flags rather than bdrv_flags due to auto-read-only */ 646*bca5283bSKevin Wolf if (s->open_flags & O_RDWR) { 647*bca5283bSKevin Wolf ret = check_hdev_writable(s->fd); 648*bca5283bSKevin Wolf if (ret < 0) { 649*bca5283bSKevin Wolf error_setg_errno(errp, -ret, "The device is not writable"); 650*bca5283bSKevin Wolf goto fail; 651*bca5283bSKevin Wolf } 652*bca5283bSKevin Wolf } 653*bca5283bSKevin Wolf 654244a5668SFam Zheng s->perm = 0; 655244a5668SFam Zheng s->shared_perm = BLK_PERM_ALL; 656244a5668SFam Zheng 657c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 658c1bb86cdSEric Blake /* Currently Linux does AIO only for files opened with O_DIRECT */ 659ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 660ed6e2161SNishanth Aravamudan if (!(s->open_flags & O_DIRECT)) { 661c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but it requires " 662c1bb86cdSEric Blake "cache.direct=on, which was not specified."); 663c1bb86cdSEric Blake ret = -EINVAL; 664c1bb86cdSEric Blake goto fail; 665c1bb86cdSEric Blake } 666ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { 667ed6e2161SNishanth Aravamudan error_prepend(errp, "Unable to use native AIO: "); 668ed6e2161SNishanth Aravamudan goto fail; 669ed6e2161SNishanth Aravamudan } 670ed6e2161SNishanth Aravamudan } 671c1bb86cdSEric Blake #else 672c1bb86cdSEric Blake if (s->use_linux_aio) { 673c1bb86cdSEric Blake error_setg(errp, "aio=native was specified, but is not supported " 674c1bb86cdSEric Blake "in this build."); 675c1bb86cdSEric Blake ret = -EINVAL; 676c1bb86cdSEric Blake goto fail; 677c1bb86cdSEric Blake } 678c1bb86cdSEric Blake #endif /* !defined(CONFIG_LINUX_AIO) */ 679c1bb86cdSEric Blake 680c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 681c6447510SAarushi Mehta if (s->use_linux_io_uring) { 682c6447510SAarushi Mehta if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs), errp)) { 683c6447510SAarushi Mehta error_prepend(errp, "Unable to use io_uring: "); 684c6447510SAarushi Mehta goto fail; 685c6447510SAarushi Mehta } 686c6447510SAarushi Mehta } 687c6447510SAarushi Mehta #else 688c6447510SAarushi Mehta if (s->use_linux_io_uring) { 689c6447510SAarushi Mehta error_setg(errp, "aio=io_uring was specified, but is not supported " 690c6447510SAarushi Mehta "in this build."); 691c6447510SAarushi Mehta ret = -EINVAL; 692c6447510SAarushi Mehta goto fail; 693c6447510SAarushi Mehta } 694c6447510SAarushi Mehta #endif /* !defined(CONFIG_LINUX_IO_URING) */ 695c6447510SAarushi Mehta 696c1bb86cdSEric Blake s->has_discard = true; 697c1bb86cdSEric Blake s->has_write_zeroes = true; 6985edc8557SKevin Wolf if ((bs->open_flags & BDRV_O_NOCACHE) != 0 && !dio_byte_aligned(s->fd)) { 699c1bb86cdSEric Blake s->needs_alignment = true; 700c1bb86cdSEric Blake } 701c1bb86cdSEric Blake 702c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 703c1bb86cdSEric Blake ret = -errno; 704c1bb86cdSEric Blake error_setg_errno(errp, errno, "Could not stat file"); 705c1bb86cdSEric Blake goto fail; 706c1bb86cdSEric Blake } 707230ff739SJohn Snow 708230ff739SJohn Snow if (!device) { 709230ff739SJohn Snow if (S_ISBLK(st.st_mode)) { 710230ff739SJohn Snow warn_report("Opening a block device as a file using the '%s' " 711230ff739SJohn Snow "driver is deprecated", bs->drv->format_name); 712230ff739SJohn Snow } else if (S_ISCHR(st.st_mode)) { 713230ff739SJohn Snow warn_report("Opening a character device as a file using the '%s' " 714230ff739SJohn Snow "driver is deprecated", bs->drv->format_name); 715230ff739SJohn Snow } else if (!S_ISREG(st.st_mode)) { 716230ff739SJohn Snow error_setg(errp, "A regular file was expected by the '%s' driver, " 717230ff739SJohn Snow "but something else was given", bs->drv->format_name); 718230ff739SJohn Snow ret = -EINVAL; 719230ff739SJohn Snow goto fail; 720230ff739SJohn Snow } else { 721c1bb86cdSEric Blake s->discard_zeroes = true; 722c1bb86cdSEric Blake s->has_fallocate = true; 723c1bb86cdSEric Blake } 724230ff739SJohn Snow } else { 725230ff739SJohn Snow if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 726230ff739SJohn Snow error_setg(errp, "'%s' driver expects either " 727230ff739SJohn Snow "a character or block device", bs->drv->format_name); 728230ff739SJohn Snow ret = -EINVAL; 729230ff739SJohn Snow goto fail; 730230ff739SJohn Snow } 731230ff739SJohn Snow } 732230ff739SJohn Snow 733c1bb86cdSEric Blake if (S_ISBLK(st.st_mode)) { 734c1bb86cdSEric Blake #ifdef BLKDISCARDZEROES 735c1bb86cdSEric Blake unsigned int arg; 736c1bb86cdSEric Blake if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { 737c1bb86cdSEric Blake s->discard_zeroes = true; 738c1bb86cdSEric Blake } 739c1bb86cdSEric Blake #endif 740c1bb86cdSEric Blake #ifdef __linux__ 741c1bb86cdSEric Blake /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do 742c1bb86cdSEric Blake * not rely on the contents of discarded blocks unless using O_DIRECT. 743c1bb86cdSEric Blake * Same for BLKZEROOUT. 744c1bb86cdSEric Blake */ 745c1bb86cdSEric Blake if (!(bs->open_flags & BDRV_O_NOCACHE)) { 746c1bb86cdSEric Blake s->discard_zeroes = false; 747c1bb86cdSEric Blake s->has_write_zeroes = false; 748c1bb86cdSEric Blake } 749c1bb86cdSEric Blake #endif 750c1bb86cdSEric Blake } 751c1bb86cdSEric Blake #ifdef __FreeBSD__ 752c1bb86cdSEric Blake if (S_ISCHR(st.st_mode)) { 753c1bb86cdSEric Blake /* 754c1bb86cdSEric Blake * The file is a char device (disk), which on FreeBSD isn't behind 755c1bb86cdSEric Blake * a pager, so force all requests to be aligned. This is needed 756c1bb86cdSEric Blake * so QEMU makes sure all IO operations on the device are aligned 757c1bb86cdSEric Blake * to sector size, or else FreeBSD will reject them with EINVAL. 758c1bb86cdSEric Blake */ 759c1bb86cdSEric Blake s->needs_alignment = true; 760c1bb86cdSEric Blake } 761c1bb86cdSEric Blake #endif 762c1bb86cdSEric Blake 763c1bb86cdSEric Blake #ifdef CONFIG_XFS 764c1bb86cdSEric Blake if (platform_test_xfs_fd(s->fd)) { 765c1bb86cdSEric Blake s->is_xfs = true; 766c1bb86cdSEric Blake } 767c1bb86cdSEric Blake #endif 768c1bb86cdSEric Blake 769738301e1SKevin Wolf bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; 7702f0c6e7aSKevin Wolf if (S_ISREG(st.st_mode)) { 7712f0c6e7aSKevin Wolf /* When extending regular files, we get zeros from the OS */ 7722f0c6e7aSKevin Wolf bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; 7732f0c6e7aSKevin Wolf } 774c1bb86cdSEric Blake ret = 0; 775c1bb86cdSEric Blake fail: 776c1bb86cdSEric Blake if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { 777c1bb86cdSEric Blake unlink(filename); 778c1bb86cdSEric Blake } 779c1bb86cdSEric Blake qemu_opts_del(opts); 780c1bb86cdSEric Blake return ret; 781c1bb86cdSEric Blake } 782c1bb86cdSEric Blake 783c1bb86cdSEric Blake static int raw_open(BlockDriverState *bs, QDict *options, int flags, 784c1bb86cdSEric Blake Error **errp) 785c1bb86cdSEric Blake { 786c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 787c1bb86cdSEric Blake 788c1bb86cdSEric Blake s->type = FTYPE_FILE; 789230ff739SJohn Snow return raw_open_common(bs, options, flags, 0, false, errp); 790c1bb86cdSEric Blake } 791c1bb86cdSEric Blake 792244a5668SFam Zheng typedef enum { 793244a5668SFam Zheng RAW_PL_PREPARE, 794244a5668SFam Zheng RAW_PL_COMMIT, 795244a5668SFam Zheng RAW_PL_ABORT, 796244a5668SFam Zheng } RawPermLockOp; 797244a5668SFam Zheng 798244a5668SFam Zheng #define PERM_FOREACH(i) \ 799244a5668SFam Zheng for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) 800244a5668SFam Zheng 801244a5668SFam Zheng /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the 802244a5668SFam Zheng * file; if @unlock == true, also unlock the unneeded bytes. 803244a5668SFam Zheng * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. 804244a5668SFam Zheng */ 8052996ffadSFam Zheng static int raw_apply_lock_bytes(BDRVRawState *s, int fd, 806244a5668SFam Zheng uint64_t perm_lock_bits, 807244a5668SFam Zheng uint64_t shared_perm_lock_bits, 808244a5668SFam Zheng bool unlock, Error **errp) 809244a5668SFam Zheng { 810244a5668SFam Zheng int ret; 811244a5668SFam Zheng int i; 8122996ffadSFam Zheng uint64_t locked_perm, locked_shared_perm; 8132996ffadSFam Zheng 8142996ffadSFam Zheng if (s) { 8152996ffadSFam Zheng locked_perm = s->locked_perm; 8162996ffadSFam Zheng locked_shared_perm = s->locked_shared_perm; 8172996ffadSFam Zheng } else { 8182996ffadSFam Zheng /* 8192996ffadSFam Zheng * We don't have the previous bits, just lock/unlock for each of the 8202996ffadSFam Zheng * requested bits. 8212996ffadSFam Zheng */ 8222996ffadSFam Zheng if (unlock) { 8232996ffadSFam Zheng locked_perm = BLK_PERM_ALL; 8242996ffadSFam Zheng locked_shared_perm = BLK_PERM_ALL; 8252996ffadSFam Zheng } else { 8262996ffadSFam Zheng locked_perm = 0; 8272996ffadSFam Zheng locked_shared_perm = 0; 8282996ffadSFam Zheng } 8292996ffadSFam Zheng } 830244a5668SFam Zheng 831244a5668SFam Zheng PERM_FOREACH(i) { 832244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 8332996ffadSFam Zheng uint64_t bit = (1ULL << i); 8342996ffadSFam Zheng if ((perm_lock_bits & bit) && !(locked_perm & bit)) { 835d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 836244a5668SFam Zheng if (ret) { 837244a5668SFam Zheng error_setg(errp, "Failed to lock byte %d", off); 838244a5668SFam Zheng return ret; 8392996ffadSFam Zheng } else if (s) { 8402996ffadSFam Zheng s->locked_perm |= bit; 841244a5668SFam Zheng } 8422996ffadSFam Zheng } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { 843d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 844244a5668SFam Zheng if (ret) { 845244a5668SFam Zheng error_setg(errp, "Failed to unlock byte %d", off); 846244a5668SFam Zheng return ret; 8472996ffadSFam Zheng } else if (s) { 8482996ffadSFam Zheng s->locked_perm &= ~bit; 849244a5668SFam Zheng } 850244a5668SFam Zheng } 851244a5668SFam Zheng } 852244a5668SFam Zheng PERM_FOREACH(i) { 853244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 8542996ffadSFam Zheng uint64_t bit = (1ULL << i); 8552996ffadSFam Zheng if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { 856d0a96155SMax Reitz ret = qemu_lock_fd(fd, off, 1, false); 857244a5668SFam Zheng if (ret) { 858244a5668SFam Zheng error_setg(errp, "Failed to lock byte %d", off); 859244a5668SFam Zheng return ret; 8602996ffadSFam Zheng } else if (s) { 8612996ffadSFam Zheng s->locked_shared_perm |= bit; 862244a5668SFam Zheng } 8632996ffadSFam Zheng } else if (unlock && (locked_shared_perm & bit) && 8642996ffadSFam Zheng !(shared_perm_lock_bits & bit)) { 865d0a96155SMax Reitz ret = qemu_unlock_fd(fd, off, 1); 866244a5668SFam Zheng if (ret) { 867244a5668SFam Zheng error_setg(errp, "Failed to unlock byte %d", off); 868244a5668SFam Zheng return ret; 8692996ffadSFam Zheng } else if (s) { 8702996ffadSFam Zheng s->locked_shared_perm &= ~bit; 871244a5668SFam Zheng } 872244a5668SFam Zheng } 873244a5668SFam Zheng } 874244a5668SFam Zheng return 0; 875244a5668SFam Zheng } 876244a5668SFam Zheng 877244a5668SFam Zheng /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ 878d0a96155SMax Reitz static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, 879244a5668SFam Zheng Error **errp) 880244a5668SFam Zheng { 881244a5668SFam Zheng int ret; 882244a5668SFam Zheng int i; 883244a5668SFam Zheng 884244a5668SFam Zheng PERM_FOREACH(i) { 885244a5668SFam Zheng int off = RAW_LOCK_SHARED_BASE + i; 886244a5668SFam Zheng uint64_t p = 1ULL << i; 887244a5668SFam Zheng if (perm & p) { 888d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 889244a5668SFam Zheng if (ret) { 890244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 891244a5668SFam Zheng error_setg(errp, 892244a5668SFam Zheng "Failed to get \"%s\" lock", 893244a5668SFam Zheng perm_name); 894244a5668SFam Zheng g_free(perm_name); 895244a5668SFam Zheng return ret; 896244a5668SFam Zheng } 897244a5668SFam Zheng } 898244a5668SFam Zheng } 899244a5668SFam Zheng PERM_FOREACH(i) { 900244a5668SFam Zheng int off = RAW_LOCK_PERM_BASE + i; 901244a5668SFam Zheng uint64_t p = 1ULL << i; 902244a5668SFam Zheng if (!(shared_perm & p)) { 903d0a96155SMax Reitz ret = qemu_lock_fd_test(fd, off, 1, true); 904244a5668SFam Zheng if (ret) { 905244a5668SFam Zheng char *perm_name = bdrv_perm_names(p); 906244a5668SFam Zheng error_setg(errp, 907244a5668SFam Zheng "Failed to get shared \"%s\" lock", 908244a5668SFam Zheng perm_name); 909244a5668SFam Zheng g_free(perm_name); 910244a5668SFam Zheng return ret; 911244a5668SFam Zheng } 912244a5668SFam Zheng } 913244a5668SFam Zheng } 914244a5668SFam Zheng return 0; 915244a5668SFam Zheng } 916244a5668SFam Zheng 917244a5668SFam Zheng static int raw_handle_perm_lock(BlockDriverState *bs, 918244a5668SFam Zheng RawPermLockOp op, 919244a5668SFam Zheng uint64_t new_perm, uint64_t new_shared, 920244a5668SFam Zheng Error **errp) 921244a5668SFam Zheng { 922244a5668SFam Zheng BDRVRawState *s = bs->opaque; 923244a5668SFam Zheng int ret = 0; 924244a5668SFam Zheng Error *local_err = NULL; 925244a5668SFam Zheng 926244a5668SFam Zheng if (!s->use_lock) { 927244a5668SFam Zheng return 0; 928244a5668SFam Zheng } 929244a5668SFam Zheng 930244a5668SFam Zheng if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { 931244a5668SFam Zheng return 0; 932244a5668SFam Zheng } 933244a5668SFam Zheng 934244a5668SFam Zheng switch (op) { 935244a5668SFam Zheng case RAW_PL_PREPARE: 936696aaaedSVladimir Sementsov-Ogievskiy if ((s->perm | new_perm) == s->perm && 937696aaaedSVladimir Sementsov-Ogievskiy (s->shared_perm & new_shared) == s->shared_perm) 938696aaaedSVladimir Sementsov-Ogievskiy { 939696aaaedSVladimir Sementsov-Ogievskiy /* 940696aaaedSVladimir Sementsov-Ogievskiy * We are going to unlock bytes, it should not fail. If it fail due 941696aaaedSVladimir Sementsov-Ogievskiy * to some fs-dependent permission-unrelated reasons (which occurs 942696aaaedSVladimir Sementsov-Ogievskiy * sometimes on NFS and leads to abort in bdrv_replace_child) we 943696aaaedSVladimir Sementsov-Ogievskiy * can't prevent such errors by any check here. And we ignore them 944696aaaedSVladimir Sementsov-Ogievskiy * anyway in ABORT and COMMIT. 945696aaaedSVladimir Sementsov-Ogievskiy */ 946696aaaedSVladimir Sementsov-Ogievskiy return 0; 947696aaaedSVladimir Sementsov-Ogievskiy } 948f2e3af29SFam Zheng ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, 949244a5668SFam Zheng ~s->shared_perm | ~new_shared, 950244a5668SFam Zheng false, errp); 951244a5668SFam Zheng if (!ret) { 952f2e3af29SFam Zheng ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); 953244a5668SFam Zheng if (!ret) { 954244a5668SFam Zheng return 0; 955244a5668SFam Zheng } 956b857431dSFam Zheng error_append_hint(errp, 957b857431dSFam Zheng "Is another process using the image [%s]?\n", 958b857431dSFam Zheng bs->filename); 959244a5668SFam Zheng } 960244a5668SFam Zheng /* fall through to unlock bytes. */ 961244a5668SFam Zheng case RAW_PL_ABORT: 962f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, 963d0a96155SMax Reitz true, &local_err); 964244a5668SFam Zheng if (local_err) { 965244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 966244a5668SFam Zheng * fail. Something weird happened, report it. 967244a5668SFam Zheng */ 968db0754dfSFam Zheng warn_report_err(local_err); 969244a5668SFam Zheng } 970244a5668SFam Zheng break; 971244a5668SFam Zheng case RAW_PL_COMMIT: 972f2e3af29SFam Zheng raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, 973d0a96155SMax Reitz true, &local_err); 974244a5668SFam Zheng if (local_err) { 975244a5668SFam Zheng /* Theoretically the above call only unlocks bytes and it cannot 976244a5668SFam Zheng * fail. Something weird happened, report it. 977244a5668SFam Zheng */ 978db0754dfSFam Zheng warn_report_err(local_err); 979244a5668SFam Zheng } 980244a5668SFam Zheng break; 981244a5668SFam Zheng } 982244a5668SFam Zheng return ret; 983244a5668SFam Zheng } 984244a5668SFam Zheng 9855cec2870SKevin Wolf static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, 98623dece19SKevin Wolf int *open_flags, uint64_t perm, bool force_dup, 9876ceabe6fSKevin Wolf Error **errp) 9885cec2870SKevin Wolf { 9895cec2870SKevin Wolf BDRVRawState *s = bs->opaque; 9905cec2870SKevin Wolf int fd = -1; 9915cec2870SKevin Wolf int ret; 99223dece19SKevin Wolf bool has_writers = perm & 99323dece19SKevin Wolf (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); 9945cec2870SKevin Wolf int fcntl_flags = O_APPEND | O_NONBLOCK; 9955cec2870SKevin Wolf #ifdef O_NOATIME 9965cec2870SKevin Wolf fcntl_flags |= O_NOATIME; 9975cec2870SKevin Wolf #endif 9985cec2870SKevin Wolf 9995cec2870SKevin Wolf *open_flags = 0; 10005cec2870SKevin Wolf if (s->type == FTYPE_CD) { 10015cec2870SKevin Wolf *open_flags |= O_NONBLOCK; 10025cec2870SKevin Wolf } 10035cec2870SKevin Wolf 100423dece19SKevin Wolf raw_parse_flags(flags, open_flags, has_writers); 10055cec2870SKevin Wolf 10065cec2870SKevin Wolf #ifdef O_ASYNC 10075cec2870SKevin Wolf /* Not all operating systems have O_ASYNC, and those that don't 10085cec2870SKevin Wolf * will not let us track the state into rs->open_flags (typically 10095cec2870SKevin Wolf * you achieve the same effect with an ioctl, for example I_SETSIG 10105cec2870SKevin Wolf * on Solaris). But we do not use O_ASYNC, so that's fine. 10115cec2870SKevin Wolf */ 10125cec2870SKevin Wolf assert((s->open_flags & O_ASYNC) == 0); 10135cec2870SKevin Wolf #endif 10145cec2870SKevin Wolf 10156ceabe6fSKevin Wolf if (!force_dup && *open_flags == s->open_flags) { 10166ceabe6fSKevin Wolf /* We're lucky, the existing fd is fine */ 10176ceabe6fSKevin Wolf return s->fd; 10186ceabe6fSKevin Wolf } 10196ceabe6fSKevin Wolf 10205cec2870SKevin Wolf if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { 10215cec2870SKevin Wolf /* dup the original fd */ 10225cec2870SKevin Wolf fd = qemu_dup(s->fd); 10235cec2870SKevin Wolf if (fd >= 0) { 10245cec2870SKevin Wolf ret = fcntl_setfl(fd, *open_flags); 10255cec2870SKevin Wolf if (ret) { 10265cec2870SKevin Wolf qemu_close(fd); 10275cec2870SKevin Wolf fd = -1; 10285cec2870SKevin Wolf } 10295cec2870SKevin Wolf } 10305cec2870SKevin Wolf } 10315cec2870SKevin Wolf 10325cec2870SKevin Wolf /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ 10335cec2870SKevin Wolf if (fd == -1) { 10345cec2870SKevin Wolf const char *normalized_filename = bs->filename; 10355cec2870SKevin Wolf ret = raw_normalize_devicepath(&normalized_filename, errp); 10365cec2870SKevin Wolf if (ret >= 0) { 10375cec2870SKevin Wolf assert(!(*open_flags & O_CREAT)); 10385cec2870SKevin Wolf fd = qemu_open(normalized_filename, *open_flags); 10395cec2870SKevin Wolf if (fd == -1) { 10405cec2870SKevin Wolf error_setg_errno(errp, errno, "Could not reopen file"); 10415cec2870SKevin Wolf return -1; 10425cec2870SKevin Wolf } 10435cec2870SKevin Wolf } 10445cec2870SKevin Wolf } 10455cec2870SKevin Wolf 1046*bca5283bSKevin Wolf if (fd != -1 && (*open_flags & O_RDWR)) { 1047*bca5283bSKevin Wolf ret = check_hdev_writable(fd); 1048*bca5283bSKevin Wolf if (ret < 0) { 1049*bca5283bSKevin Wolf qemu_close(fd); 1050*bca5283bSKevin Wolf error_setg_errno(errp, -ret, "The device is not writable"); 1051*bca5283bSKevin Wolf return -1; 1052*bca5283bSKevin Wolf } 1053*bca5283bSKevin Wolf } 1054*bca5283bSKevin Wolf 10555cec2870SKevin Wolf return fd; 10565cec2870SKevin Wolf } 10575cec2870SKevin Wolf 1058c1bb86cdSEric Blake static int raw_reopen_prepare(BDRVReopenState *state, 1059c1bb86cdSEric Blake BlockReopenQueue *queue, Error **errp) 1060c1bb86cdSEric Blake { 1061c1bb86cdSEric Blake BDRVRawState *s; 1062c1bb86cdSEric Blake BDRVRawReopenState *rs; 106331be8a2aSStefan Hajnoczi QemuOpts *opts; 1064a6aeca0cSKevin Wolf int ret; 1065c1bb86cdSEric Blake Error *local_err = NULL; 1066c1bb86cdSEric Blake 1067c1bb86cdSEric Blake assert(state != NULL); 1068c1bb86cdSEric Blake assert(state->bs != NULL); 1069c1bb86cdSEric Blake 1070c1bb86cdSEric Blake s = state->bs->opaque; 1071c1bb86cdSEric Blake 1072c1bb86cdSEric Blake state->opaque = g_new0(BDRVRawReopenState, 1); 1073c1bb86cdSEric Blake rs = state->opaque; 107431be8a2aSStefan Hajnoczi 107531be8a2aSStefan Hajnoczi /* Handle options changes */ 107631be8a2aSStefan Hajnoczi opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); 1077af175e85SMarkus Armbruster if (!qemu_opts_absorb_qdict(opts, state->options, errp)) { 107831be8a2aSStefan Hajnoczi ret = -EINVAL; 107931be8a2aSStefan Hajnoczi goto out; 108031be8a2aSStefan Hajnoczi } 108131be8a2aSStefan Hajnoczi 1082f357fcd8SStefan Hajnoczi rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); 10838d324575SAlberto Garcia rs->check_cache_dropped = 10848d324575SAlberto Garcia qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); 10858d324575SAlberto Garcia 10868d324575SAlberto Garcia /* This driver's reopen function doesn't currently allow changing 10878d324575SAlberto Garcia * other options, so let's put them back in the original QDict and 10888d324575SAlberto Garcia * bdrv_reopen_prepare() will detect changes and complain. */ 10898d324575SAlberto Garcia qemu_opts_to_qdict(opts, state->options); 1090c1bb86cdSEric Blake 10915cec2870SKevin Wolf rs->fd = raw_reconfigure_getfd(state->bs, state->flags, &rs->open_flags, 109223dece19SKevin Wolf state->perm, true, &local_err); 10935cec2870SKevin Wolf if (local_err) { 10945cec2870SKevin Wolf error_propagate(errp, local_err); 1095c1bb86cdSEric Blake ret = -1; 10965cec2870SKevin Wolf goto out; 1097c1bb86cdSEric Blake } 1098c1bb86cdSEric Blake 1099c1bb86cdSEric Blake /* Fail already reopen_prepare() if we can't get a working O_DIRECT 1100c1bb86cdSEric Blake * alignment with the new fd. */ 1101c1bb86cdSEric Blake if (rs->fd != -1) { 1102c1bb86cdSEric Blake raw_probe_alignment(state->bs, rs->fd, &local_err); 1103c1bb86cdSEric Blake if (local_err) { 1104c1bb86cdSEric Blake error_propagate(errp, local_err); 1105c1bb86cdSEric Blake ret = -EINVAL; 1106a6aeca0cSKevin Wolf goto out_fd; 1107a6aeca0cSKevin Wolf } 1108c1bb86cdSEric Blake } 1109c1bb86cdSEric Blake 1110e0c9cf3aSKevin Wolf s->reopen_state = state; 1111a6aeca0cSKevin Wolf ret = 0; 1112a6aeca0cSKevin Wolf out_fd: 1113a6aeca0cSKevin Wolf if (ret < 0) { 1114a6aeca0cSKevin Wolf qemu_close(rs->fd); 1115a6aeca0cSKevin Wolf rs->fd = -1; 1116a6aeca0cSKevin Wolf } 111731be8a2aSStefan Hajnoczi out: 111831be8a2aSStefan Hajnoczi qemu_opts_del(opts); 1119c1bb86cdSEric Blake return ret; 1120c1bb86cdSEric Blake } 1121c1bb86cdSEric Blake 1122c1bb86cdSEric Blake static void raw_reopen_commit(BDRVReopenState *state) 1123c1bb86cdSEric Blake { 1124c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1125c1bb86cdSEric Blake BDRVRawState *s = state->bs->opaque; 1126c1bb86cdSEric Blake 1127f357fcd8SStefan Hajnoczi s->drop_cache = rs->drop_cache; 112831be8a2aSStefan Hajnoczi s->check_cache_dropped = rs->check_cache_dropped; 1129c1bb86cdSEric Blake s->open_flags = rs->open_flags; 1130c1bb86cdSEric Blake 1131c1bb86cdSEric Blake qemu_close(s->fd); 1132c1bb86cdSEric Blake s->fd = rs->fd; 1133c1bb86cdSEric Blake 1134c1bb86cdSEric Blake g_free(state->opaque); 1135c1bb86cdSEric Blake state->opaque = NULL; 1136e0c9cf3aSKevin Wolf 1137e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1138e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1139c1bb86cdSEric Blake } 1140c1bb86cdSEric Blake 1141c1bb86cdSEric Blake 1142c1bb86cdSEric Blake static void raw_reopen_abort(BDRVReopenState *state) 1143c1bb86cdSEric Blake { 1144c1bb86cdSEric Blake BDRVRawReopenState *rs = state->opaque; 1145e0c9cf3aSKevin Wolf BDRVRawState *s = state->bs->opaque; 1146c1bb86cdSEric Blake 1147c1bb86cdSEric Blake /* nothing to do if NULL, we didn't get far enough */ 1148c1bb86cdSEric Blake if (rs == NULL) { 1149c1bb86cdSEric Blake return; 1150c1bb86cdSEric Blake } 1151c1bb86cdSEric Blake 1152c1bb86cdSEric Blake if (rs->fd >= 0) { 1153c1bb86cdSEric Blake qemu_close(rs->fd); 1154c1bb86cdSEric Blake rs->fd = -1; 1155c1bb86cdSEric Blake } 1156c1bb86cdSEric Blake g_free(state->opaque); 1157c1bb86cdSEric Blake state->opaque = NULL; 1158e0c9cf3aSKevin Wolf 1159e0c9cf3aSKevin Wolf assert(s->reopen_state == state); 1160e0c9cf3aSKevin Wolf s->reopen_state = NULL; 1161c1bb86cdSEric Blake } 1162c1bb86cdSEric Blake 1163867eccfeSMaxim Levitsky static int sg_get_max_transfer_length(int fd) 1164c1bb86cdSEric Blake { 1165c1bb86cdSEric Blake #ifdef BLKSECTGET 116648265250SEric Farman int max_bytes = 0; 1167867eccfeSMaxim Levitsky 1168867eccfeSMaxim Levitsky if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) { 116948265250SEric Farman return max_bytes; 1170c1bb86cdSEric Blake } else { 1171c1bb86cdSEric Blake return -errno; 1172c1bb86cdSEric Blake } 1173c1bb86cdSEric Blake #else 1174c1bb86cdSEric Blake return -ENOSYS; 1175c1bb86cdSEric Blake #endif 1176c1bb86cdSEric Blake } 1177c1bb86cdSEric Blake 1178867eccfeSMaxim Levitsky static int sg_get_max_segments(int fd) 11799103f1ceSFam Zheng { 11809103f1ceSFam Zheng #ifdef CONFIG_LINUX 11819103f1ceSFam Zheng char buf[32]; 11829103f1ceSFam Zheng const char *end; 1183867eccfeSMaxim Levitsky char *sysfspath = NULL; 11849103f1ceSFam Zheng int ret; 1185867eccfeSMaxim Levitsky int sysfd = -1; 11869103f1ceSFam Zheng long max_segments; 1187867eccfeSMaxim Levitsky struct stat st; 1188867eccfeSMaxim Levitsky 1189867eccfeSMaxim Levitsky if (fstat(fd, &st)) { 1190867eccfeSMaxim Levitsky ret = -errno; 1191867eccfeSMaxim Levitsky goto out; 1192867eccfeSMaxim Levitsky } 11939103f1ceSFam Zheng 11949103f1ceSFam Zheng sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", 1195867eccfeSMaxim Levitsky major(st.st_rdev), minor(st.st_rdev)); 1196867eccfeSMaxim Levitsky sysfd = open(sysfspath, O_RDONLY); 1197867eccfeSMaxim Levitsky if (sysfd == -1) { 11989103f1ceSFam Zheng ret = -errno; 11999103f1ceSFam Zheng goto out; 12009103f1ceSFam Zheng } 12019103f1ceSFam Zheng do { 1202867eccfeSMaxim Levitsky ret = read(sysfd, buf, sizeof(buf) - 1); 12039103f1ceSFam Zheng } while (ret == -1 && errno == EINTR); 12049103f1ceSFam Zheng if (ret < 0) { 12059103f1ceSFam Zheng ret = -errno; 12069103f1ceSFam Zheng goto out; 12079103f1ceSFam Zheng } else if (ret == 0) { 12089103f1ceSFam Zheng ret = -EIO; 12099103f1ceSFam Zheng goto out; 12109103f1ceSFam Zheng } 12119103f1ceSFam Zheng buf[ret] = 0; 12129103f1ceSFam Zheng /* The file is ended with '\n', pass 'end' to accept that. */ 12139103f1ceSFam Zheng ret = qemu_strtol(buf, &end, 10, &max_segments); 12149103f1ceSFam Zheng if (ret == 0 && end && *end == '\n') { 12159103f1ceSFam Zheng ret = max_segments; 12169103f1ceSFam Zheng } 12179103f1ceSFam Zheng 12189103f1ceSFam Zheng out: 1219867eccfeSMaxim Levitsky if (sysfd != -1) { 1220867eccfeSMaxim Levitsky close(sysfd); 1221fed414dfSFam Zheng } 12229103f1ceSFam Zheng g_free(sysfspath); 12239103f1ceSFam Zheng return ret; 12249103f1ceSFam Zheng #else 12259103f1ceSFam Zheng return -ENOTSUP; 12269103f1ceSFam Zheng #endif 12279103f1ceSFam Zheng } 12289103f1ceSFam Zheng 1229c1bb86cdSEric Blake static void raw_refresh_limits(BlockDriverState *bs, Error **errp) 1230c1bb86cdSEric Blake { 1231c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1232c1bb86cdSEric Blake 1233867eccfeSMaxim Levitsky if (bs->sg) { 1234867eccfeSMaxim Levitsky int ret = sg_get_max_transfer_length(s->fd); 1235867eccfeSMaxim Levitsky 123648265250SEric Farman if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { 123748265250SEric Farman bs->bl.max_transfer = pow2floor(ret); 1238c1bb86cdSEric Blake } 1239867eccfeSMaxim Levitsky 1240867eccfeSMaxim Levitsky ret = sg_get_max_segments(s->fd); 12419103f1ceSFam Zheng if (ret > 0) { 1242038adc2fSWei Yang bs->bl.max_transfer = MIN(bs->bl.max_transfer, 1243038adc2fSWei Yang ret * qemu_real_host_page_size); 1244c1bb86cdSEric Blake } 1245c1bb86cdSEric Blake } 1246c1bb86cdSEric Blake 1247c1bb86cdSEric Blake raw_probe_alignment(bs, s->fd, errp); 1248c1bb86cdSEric Blake bs->bl.min_mem_alignment = s->buf_align; 1249038adc2fSWei Yang bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size); 1250c1bb86cdSEric Blake } 1251c1bb86cdSEric Blake 1252c1bb86cdSEric Blake static int check_for_dasd(int fd) 1253c1bb86cdSEric Blake { 1254c1bb86cdSEric Blake #ifdef BIODASDINFO2 1255c1bb86cdSEric Blake struct dasd_information2_t info = {0}; 1256c1bb86cdSEric Blake 1257c1bb86cdSEric Blake return ioctl(fd, BIODASDINFO2, &info); 1258c1bb86cdSEric Blake #else 1259c1bb86cdSEric Blake return -1; 1260c1bb86cdSEric Blake #endif 1261c1bb86cdSEric Blake } 1262c1bb86cdSEric Blake 1263c1bb86cdSEric Blake /** 1264c1bb86cdSEric Blake * Try to get @bs's logical and physical block size. 1265c1bb86cdSEric Blake * On success, store them in @bsz and return zero. 1266c1bb86cdSEric Blake * On failure, return negative errno. 1267c1bb86cdSEric Blake */ 1268c1bb86cdSEric Blake static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) 1269c1bb86cdSEric Blake { 1270c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1271c1bb86cdSEric Blake int ret; 1272c1bb86cdSEric Blake 1273c1bb86cdSEric Blake /* If DASD, get blocksizes */ 1274c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1275c1bb86cdSEric Blake return -ENOTSUP; 1276c1bb86cdSEric Blake } 1277c1bb86cdSEric Blake ret = probe_logical_blocksize(s->fd, &bsz->log); 1278c1bb86cdSEric Blake if (ret < 0) { 1279c1bb86cdSEric Blake return ret; 1280c1bb86cdSEric Blake } 1281c1bb86cdSEric Blake return probe_physical_blocksize(s->fd, &bsz->phys); 1282c1bb86cdSEric Blake } 1283c1bb86cdSEric Blake 1284c1bb86cdSEric Blake /** 1285c1bb86cdSEric Blake * Try to get @bs's geometry: cyls, heads, sectors. 1286c1bb86cdSEric Blake * On success, store them in @geo and return 0. 1287c1bb86cdSEric Blake * On failure return -errno. 1288c1bb86cdSEric Blake * (Allows block driver to assign default geometry values that guest sees) 1289c1bb86cdSEric Blake */ 1290c1bb86cdSEric Blake #ifdef __linux__ 1291c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1292c1bb86cdSEric Blake { 1293c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1294c1bb86cdSEric Blake struct hd_geometry ioctl_geo = {0}; 1295c1bb86cdSEric Blake 1296c1bb86cdSEric Blake /* If DASD, get its geometry */ 1297c1bb86cdSEric Blake if (check_for_dasd(s->fd) < 0) { 1298c1bb86cdSEric Blake return -ENOTSUP; 1299c1bb86cdSEric Blake } 1300c1bb86cdSEric Blake if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { 1301c1bb86cdSEric Blake return -errno; 1302c1bb86cdSEric Blake } 1303c1bb86cdSEric Blake /* HDIO_GETGEO may return success even though geo contains zeros 1304c1bb86cdSEric Blake (e.g. certain multipath setups) */ 1305c1bb86cdSEric Blake if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { 1306c1bb86cdSEric Blake return -ENOTSUP; 1307c1bb86cdSEric Blake } 1308c1bb86cdSEric Blake /* Do not return a geometry for partition */ 1309c1bb86cdSEric Blake if (ioctl_geo.start != 0) { 1310c1bb86cdSEric Blake return -ENOTSUP; 1311c1bb86cdSEric Blake } 1312c1bb86cdSEric Blake geo->heads = ioctl_geo.heads; 1313c1bb86cdSEric Blake geo->sectors = ioctl_geo.sectors; 1314c1bb86cdSEric Blake geo->cylinders = ioctl_geo.cylinders; 1315c1bb86cdSEric Blake 1316c1bb86cdSEric Blake return 0; 1317c1bb86cdSEric Blake } 1318c1bb86cdSEric Blake #else /* __linux__ */ 1319c1bb86cdSEric Blake static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) 1320c1bb86cdSEric Blake { 1321c1bb86cdSEric Blake return -ENOTSUP; 1322c1bb86cdSEric Blake } 1323c1bb86cdSEric Blake #endif 1324c1bb86cdSEric Blake 132503425671SKevin Wolf #if defined(__linux__) 132603425671SKevin Wolf static int handle_aiocb_ioctl(void *opaque) 1327c1bb86cdSEric Blake { 132803425671SKevin Wolf RawPosixAIOData *aiocb = opaque; 1329c1bb86cdSEric Blake int ret; 1330c1bb86cdSEric Blake 1331d57c44d0SKevin Wolf ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); 1332c1bb86cdSEric Blake if (ret == -1) { 1333c1bb86cdSEric Blake return -errno; 1334c1bb86cdSEric Blake } 1335c1bb86cdSEric Blake 1336c1bb86cdSEric Blake return 0; 1337c1bb86cdSEric Blake } 133803425671SKevin Wolf #endif /* linux */ 1339c1bb86cdSEric Blake 134006dc9bd5SKevin Wolf static int handle_aiocb_flush(void *opaque) 1341c1bb86cdSEric Blake { 134206dc9bd5SKevin Wolf RawPosixAIOData *aiocb = opaque; 1343e5bcf967SKevin Wolf BDRVRawState *s = aiocb->bs->opaque; 1344c1bb86cdSEric Blake int ret; 1345c1bb86cdSEric Blake 1346e5bcf967SKevin Wolf if (s->page_cache_inconsistent) { 1347e5bcf967SKevin Wolf return -EIO; 1348e5bcf967SKevin Wolf } 1349e5bcf967SKevin Wolf 1350c1bb86cdSEric Blake ret = qemu_fdatasync(aiocb->aio_fildes); 1351c1bb86cdSEric Blake if (ret == -1) { 1352e5bcf967SKevin Wolf /* There is no clear definition of the semantics of a failing fsync(), 1353e5bcf967SKevin Wolf * so we may have to assume the worst. The sad truth is that this 1354e5bcf967SKevin Wolf * assumption is correct for Linux. Some pages are now probably marked 1355e5bcf967SKevin Wolf * clean in the page cache even though they are inconsistent with the 1356e5bcf967SKevin Wolf * on-disk contents. The next fdatasync() call would succeed, but no 1357e5bcf967SKevin Wolf * further writeback attempt will be made. We can't get back to a state 1358e5bcf967SKevin Wolf * in which we know what is on disk (we would have to rewrite 1359e5bcf967SKevin Wolf * everything that was touched since the last fdatasync() at least), so 1360e5bcf967SKevin Wolf * make bdrv_flush() fail permanently. Given that the behaviour isn't 1361e5bcf967SKevin Wolf * really defined, I have little hope that other OSes are doing better. 1362e5bcf967SKevin Wolf * 1363e5bcf967SKevin Wolf * Obviously, this doesn't affect O_DIRECT, which bypasses the page 1364e5bcf967SKevin Wolf * cache. */ 1365e5bcf967SKevin Wolf if ((s->open_flags & O_DIRECT) == 0) { 1366e5bcf967SKevin Wolf s->page_cache_inconsistent = true; 1367e5bcf967SKevin Wolf } 1368c1bb86cdSEric Blake return -errno; 1369c1bb86cdSEric Blake } 1370c1bb86cdSEric Blake return 0; 1371c1bb86cdSEric Blake } 1372c1bb86cdSEric Blake 1373c1bb86cdSEric Blake #ifdef CONFIG_PREADV 1374c1bb86cdSEric Blake 1375c1bb86cdSEric Blake static bool preadv_present = true; 1376c1bb86cdSEric Blake 1377c1bb86cdSEric Blake static ssize_t 1378c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1379c1bb86cdSEric Blake { 1380c1bb86cdSEric Blake return preadv(fd, iov, nr_iov, offset); 1381c1bb86cdSEric Blake } 1382c1bb86cdSEric Blake 1383c1bb86cdSEric Blake static ssize_t 1384c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1385c1bb86cdSEric Blake { 1386c1bb86cdSEric Blake return pwritev(fd, iov, nr_iov, offset); 1387c1bb86cdSEric Blake } 1388c1bb86cdSEric Blake 1389c1bb86cdSEric Blake #else 1390c1bb86cdSEric Blake 1391c1bb86cdSEric Blake static bool preadv_present = false; 1392c1bb86cdSEric Blake 1393c1bb86cdSEric Blake static ssize_t 1394c1bb86cdSEric Blake qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1395c1bb86cdSEric Blake { 1396c1bb86cdSEric Blake return -ENOSYS; 1397c1bb86cdSEric Blake } 1398c1bb86cdSEric Blake 1399c1bb86cdSEric Blake static ssize_t 1400c1bb86cdSEric Blake qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) 1401c1bb86cdSEric Blake { 1402c1bb86cdSEric Blake return -ENOSYS; 1403c1bb86cdSEric Blake } 1404c1bb86cdSEric Blake 1405c1bb86cdSEric Blake #endif 1406c1bb86cdSEric Blake 1407c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) 1408c1bb86cdSEric Blake { 1409c1bb86cdSEric Blake ssize_t len; 1410c1bb86cdSEric Blake 1411c1bb86cdSEric Blake do { 1412c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) 1413c1bb86cdSEric Blake len = qemu_pwritev(aiocb->aio_fildes, 1414d57c44d0SKevin Wolf aiocb->io.iov, 1415d57c44d0SKevin Wolf aiocb->io.niov, 1416c1bb86cdSEric Blake aiocb->aio_offset); 1417c1bb86cdSEric Blake else 1418c1bb86cdSEric Blake len = qemu_preadv(aiocb->aio_fildes, 1419d57c44d0SKevin Wolf aiocb->io.iov, 1420d57c44d0SKevin Wolf aiocb->io.niov, 1421c1bb86cdSEric Blake aiocb->aio_offset); 1422c1bb86cdSEric Blake } while (len == -1 && errno == EINTR); 1423c1bb86cdSEric Blake 1424c1bb86cdSEric Blake if (len == -1) { 1425c1bb86cdSEric Blake return -errno; 1426c1bb86cdSEric Blake } 1427c1bb86cdSEric Blake return len; 1428c1bb86cdSEric Blake } 1429c1bb86cdSEric Blake 1430c1bb86cdSEric Blake /* 1431c1bb86cdSEric Blake * Read/writes the data to/from a given linear buffer. 1432c1bb86cdSEric Blake * 1433c1bb86cdSEric Blake * Returns the number of bytes handles or -errno in case of an error. Short 1434c1bb86cdSEric Blake * reads are only returned if the end of the file is reached. 1435c1bb86cdSEric Blake */ 1436c1bb86cdSEric Blake static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) 1437c1bb86cdSEric Blake { 1438c1bb86cdSEric Blake ssize_t offset = 0; 1439c1bb86cdSEric Blake ssize_t len; 1440c1bb86cdSEric Blake 1441c1bb86cdSEric Blake while (offset < aiocb->aio_nbytes) { 1442c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1443c1bb86cdSEric Blake len = pwrite(aiocb->aio_fildes, 1444c1bb86cdSEric Blake (const char *)buf + offset, 1445c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1446c1bb86cdSEric Blake aiocb->aio_offset + offset); 1447c1bb86cdSEric Blake } else { 1448c1bb86cdSEric Blake len = pread(aiocb->aio_fildes, 1449c1bb86cdSEric Blake buf + offset, 1450c1bb86cdSEric Blake aiocb->aio_nbytes - offset, 1451c1bb86cdSEric Blake aiocb->aio_offset + offset); 1452c1bb86cdSEric Blake } 1453c1bb86cdSEric Blake if (len == -1 && errno == EINTR) { 1454c1bb86cdSEric Blake continue; 1455c1bb86cdSEric Blake } else if (len == -1 && errno == EINVAL && 1456c1bb86cdSEric Blake (aiocb->bs->open_flags & BDRV_O_NOCACHE) && 1457c1bb86cdSEric Blake !(aiocb->aio_type & QEMU_AIO_WRITE) && 1458c1bb86cdSEric Blake offset > 0) { 1459c1bb86cdSEric Blake /* O_DIRECT pread() may fail with EINVAL when offset is unaligned 1460c1bb86cdSEric Blake * after a short read. Assume that O_DIRECT short reads only occur 1461c1bb86cdSEric Blake * at EOF. Therefore this is a short read, not an I/O error. 1462c1bb86cdSEric Blake */ 1463c1bb86cdSEric Blake break; 1464c1bb86cdSEric Blake } else if (len == -1) { 1465c1bb86cdSEric Blake offset = -errno; 1466c1bb86cdSEric Blake break; 1467c1bb86cdSEric Blake } else if (len == 0) { 1468c1bb86cdSEric Blake break; 1469c1bb86cdSEric Blake } 1470c1bb86cdSEric Blake offset += len; 1471c1bb86cdSEric Blake } 1472c1bb86cdSEric Blake 1473c1bb86cdSEric Blake return offset; 1474c1bb86cdSEric Blake } 1475c1bb86cdSEric Blake 1476999e6b69SKevin Wolf static int handle_aiocb_rw(void *opaque) 1477c1bb86cdSEric Blake { 1478999e6b69SKevin Wolf RawPosixAIOData *aiocb = opaque; 1479c1bb86cdSEric Blake ssize_t nbytes; 1480c1bb86cdSEric Blake char *buf; 1481c1bb86cdSEric Blake 1482c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { 1483c1bb86cdSEric Blake /* 1484c1bb86cdSEric Blake * If there is just a single buffer, and it is properly aligned 1485c1bb86cdSEric Blake * we can just use plain pread/pwrite without any problems. 1486c1bb86cdSEric Blake */ 1487d57c44d0SKevin Wolf if (aiocb->io.niov == 1) { 148854c7ca1bSKevin Wolf nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); 148954c7ca1bSKevin Wolf goto out; 1490c1bb86cdSEric Blake } 1491c1bb86cdSEric Blake /* 1492c1bb86cdSEric Blake * We have more than one iovec, and all are properly aligned. 1493c1bb86cdSEric Blake * 1494c1bb86cdSEric Blake * Try preadv/pwritev first and fall back to linearizing the 1495c1bb86cdSEric Blake * buffer if it's not supported. 1496c1bb86cdSEric Blake */ 1497c1bb86cdSEric Blake if (preadv_present) { 1498c1bb86cdSEric Blake nbytes = handle_aiocb_rw_vector(aiocb); 1499c1bb86cdSEric Blake if (nbytes == aiocb->aio_nbytes || 1500c1bb86cdSEric Blake (nbytes < 0 && nbytes != -ENOSYS)) { 150154c7ca1bSKevin Wolf goto out; 1502c1bb86cdSEric Blake } 1503c1bb86cdSEric Blake preadv_present = false; 1504c1bb86cdSEric Blake } 1505c1bb86cdSEric Blake 1506c1bb86cdSEric Blake /* 1507c1bb86cdSEric Blake * XXX(hch): short read/write. no easy way to handle the reminder 1508c1bb86cdSEric Blake * using these interfaces. For now retry using plain 1509c1bb86cdSEric Blake * pread/pwrite? 1510c1bb86cdSEric Blake */ 1511c1bb86cdSEric Blake } 1512c1bb86cdSEric Blake 1513c1bb86cdSEric Blake /* 1514c1bb86cdSEric Blake * Ok, we have to do it the hard way, copy all segments into 1515c1bb86cdSEric Blake * a single aligned buffer. 1516c1bb86cdSEric Blake */ 1517c1bb86cdSEric Blake buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); 1518c1bb86cdSEric Blake if (buf == NULL) { 151954c7ca1bSKevin Wolf nbytes = -ENOMEM; 152054c7ca1bSKevin Wolf goto out; 1521c1bb86cdSEric Blake } 1522c1bb86cdSEric Blake 1523c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_WRITE) { 1524c1bb86cdSEric Blake char *p = buf; 1525c1bb86cdSEric Blake int i; 1526c1bb86cdSEric Blake 1527d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov; ++i) { 1528d57c44d0SKevin Wolf memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); 1529d57c44d0SKevin Wolf p += aiocb->io.iov[i].iov_len; 1530c1bb86cdSEric Blake } 1531c1bb86cdSEric Blake assert(p - buf == aiocb->aio_nbytes); 1532c1bb86cdSEric Blake } 1533c1bb86cdSEric Blake 1534c1bb86cdSEric Blake nbytes = handle_aiocb_rw_linear(aiocb, buf); 1535c1bb86cdSEric Blake if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { 1536c1bb86cdSEric Blake char *p = buf; 1537c1bb86cdSEric Blake size_t count = aiocb->aio_nbytes, copy; 1538c1bb86cdSEric Blake int i; 1539c1bb86cdSEric Blake 1540d57c44d0SKevin Wolf for (i = 0; i < aiocb->io.niov && count; ++i) { 1541c1bb86cdSEric Blake copy = count; 1542d57c44d0SKevin Wolf if (copy > aiocb->io.iov[i].iov_len) { 1543d57c44d0SKevin Wolf copy = aiocb->io.iov[i].iov_len; 1544c1bb86cdSEric Blake } 1545d57c44d0SKevin Wolf memcpy(aiocb->io.iov[i].iov_base, p, copy); 1546c1bb86cdSEric Blake assert(count >= copy); 1547c1bb86cdSEric Blake p += copy; 1548c1bb86cdSEric Blake count -= copy; 1549c1bb86cdSEric Blake } 1550c1bb86cdSEric Blake assert(count == 0); 1551c1bb86cdSEric Blake } 1552c1bb86cdSEric Blake qemu_vfree(buf); 1553c1bb86cdSEric Blake 155454c7ca1bSKevin Wolf out: 155554c7ca1bSKevin Wolf if (nbytes == aiocb->aio_nbytes) { 155654c7ca1bSKevin Wolf return 0; 155754c7ca1bSKevin Wolf } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { 155854c7ca1bSKevin Wolf if (aiocb->aio_type & QEMU_AIO_WRITE) { 155954c7ca1bSKevin Wolf return -EINVAL; 156054c7ca1bSKevin Wolf } else { 156154c7ca1bSKevin Wolf iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, 156254c7ca1bSKevin Wolf 0, aiocb->aio_nbytes - nbytes); 156354c7ca1bSKevin Wolf return 0; 156454c7ca1bSKevin Wolf } 156554c7ca1bSKevin Wolf } else { 156654c7ca1bSKevin Wolf assert(nbytes < 0); 1567c1bb86cdSEric Blake return nbytes; 1568c1bb86cdSEric Blake } 156954c7ca1bSKevin Wolf } 1570c1bb86cdSEric Blake 1571c1bb86cdSEric Blake static int translate_err(int err) 1572c1bb86cdSEric Blake { 1573c1bb86cdSEric Blake if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || 1574c1bb86cdSEric Blake err == -ENOTTY) { 1575c1bb86cdSEric Blake err = -ENOTSUP; 1576c1bb86cdSEric Blake } 1577c1bb86cdSEric Blake return err; 1578c1bb86cdSEric Blake } 1579c1bb86cdSEric Blake 1580c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 1581c1bb86cdSEric Blake static int do_fallocate(int fd, int mode, off_t offset, off_t len) 1582c1bb86cdSEric Blake { 1583c1bb86cdSEric Blake do { 1584c1bb86cdSEric Blake if (fallocate(fd, mode, offset, len) == 0) { 1585c1bb86cdSEric Blake return 0; 1586c1bb86cdSEric Blake } 1587c1bb86cdSEric Blake } while (errno == EINTR); 1588c1bb86cdSEric Blake return translate_err(-errno); 1589c1bb86cdSEric Blake } 1590c1bb86cdSEric Blake #endif 1591c1bb86cdSEric Blake 1592c1bb86cdSEric Blake static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) 1593c1bb86cdSEric Blake { 1594c1bb86cdSEric Blake int ret = -ENOTSUP; 1595c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1596c1bb86cdSEric Blake 1597c1bb86cdSEric Blake if (!s->has_write_zeroes) { 1598c1bb86cdSEric Blake return -ENOTSUP; 1599c1bb86cdSEric Blake } 1600c1bb86cdSEric Blake 1601c1bb86cdSEric Blake #ifdef BLKZEROOUT 1602738301e1SKevin Wolf /* The BLKZEROOUT implementation in the kernel doesn't set 1603738301e1SKevin Wolf * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow 1604738301e1SKevin Wolf * fallbacks. */ 1605738301e1SKevin Wolf if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { 1606c1bb86cdSEric Blake do { 1607c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1608c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { 1609c1bb86cdSEric Blake return 0; 1610c1bb86cdSEric Blake } 1611c1bb86cdSEric Blake } while (errno == EINTR); 1612c1bb86cdSEric Blake 1613c1bb86cdSEric Blake ret = translate_err(-errno); 1614c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1615c1bb86cdSEric Blake s->has_write_zeroes = false; 1616c1bb86cdSEric Blake } 1617effecce6SKevin Wolf } 1618effecce6SKevin Wolf #endif 1619effecce6SKevin Wolf 1620c1bb86cdSEric Blake return ret; 1621c1bb86cdSEric Blake } 1622c1bb86cdSEric Blake 16237154d8aeSKevin Wolf static int handle_aiocb_write_zeroes(void *opaque) 1624c1bb86cdSEric Blake { 16257154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 162670d9110bSDenis V. Lunev #ifdef CONFIG_FALLOCATE 1627b2c6f23fSMax Reitz BDRVRawState *s = aiocb->bs->opaque; 162870d9110bSDenis V. Lunev int64_t len; 162970d9110bSDenis V. Lunev #endif 1630c1bb86cdSEric Blake 1631c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1632c1bb86cdSEric Blake return handle_aiocb_write_zeroes_block(aiocb); 1633c1bb86cdSEric Blake } 1634c1bb86cdSEric Blake 1635c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_ZERO_RANGE 1636c1bb86cdSEric Blake if (s->has_write_zeroes) { 1637c1bb86cdSEric Blake int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, 1638c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1639294682ccSAndrey Shinkevich if (ret == -EINVAL) { 1640294682ccSAndrey Shinkevich /* 1641294682ccSAndrey Shinkevich * Allow falling back to pwrite for file systems that 1642294682ccSAndrey Shinkevich * do not support fallocate() for an unaligned byte range. 1643294682ccSAndrey Shinkevich */ 1644294682ccSAndrey Shinkevich return -ENOTSUP; 1645294682ccSAndrey Shinkevich } 1646c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1647c1bb86cdSEric Blake return ret; 1648c1bb86cdSEric Blake } 1649c1bb86cdSEric Blake s->has_write_zeroes = false; 1650c1bb86cdSEric Blake } 1651c1bb86cdSEric Blake #endif 1652c1bb86cdSEric Blake 1653c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1654c1bb86cdSEric Blake if (s->has_discard && s->has_fallocate) { 1655c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 1656c1bb86cdSEric Blake FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1657c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1658c1bb86cdSEric Blake if (ret == 0) { 1659c1bb86cdSEric Blake ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1660c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1661c1bb86cdSEric Blake return ret; 1662c1bb86cdSEric Blake } 1663c1bb86cdSEric Blake s->has_fallocate = false; 1664c1bb86cdSEric Blake } else if (ret != -ENOTSUP) { 1665c1bb86cdSEric Blake return ret; 1666c1bb86cdSEric Blake } else { 1667c1bb86cdSEric Blake s->has_discard = false; 1668c1bb86cdSEric Blake } 1669c1bb86cdSEric Blake } 1670c1bb86cdSEric Blake #endif 1671c1bb86cdSEric Blake 1672c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE 167370d9110bSDenis V. Lunev /* Last resort: we are trying to extend the file with zeroed data. This 167470d9110bSDenis V. Lunev * can be done via fallocate(fd, 0) */ 167570d9110bSDenis V. Lunev len = bdrv_getlength(aiocb->bs); 167670d9110bSDenis V. Lunev if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { 1677c1bb86cdSEric Blake int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); 1678c1bb86cdSEric Blake if (ret == 0 || ret != -ENOTSUP) { 1679c1bb86cdSEric Blake return ret; 1680c1bb86cdSEric Blake } 1681c1bb86cdSEric Blake s->has_fallocate = false; 1682c1bb86cdSEric Blake } 1683c1bb86cdSEric Blake #endif 1684c1bb86cdSEric Blake 1685c1bb86cdSEric Blake return -ENOTSUP; 1686c1bb86cdSEric Blake } 1687c1bb86cdSEric Blake 16887154d8aeSKevin Wolf static int handle_aiocb_write_zeroes_unmap(void *opaque) 168934fa110eSKevin Wolf { 16907154d8aeSKevin Wolf RawPosixAIOData *aiocb = opaque; 169134fa110eSKevin Wolf BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; 169234fa110eSKevin Wolf 169334fa110eSKevin Wolf /* First try to write zeros and unmap at the same time */ 169434fa110eSKevin Wolf 169534fa110eSKevin Wolf #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1696b3ac2b94SSimran Singhal int ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 169734fa110eSKevin Wolf aiocb->aio_offset, aiocb->aio_nbytes); 169834fa110eSKevin Wolf if (ret != -ENOTSUP) { 169934fa110eSKevin Wolf return ret; 170034fa110eSKevin Wolf } 170134fa110eSKevin Wolf #endif 170234fa110eSKevin Wolf 170334fa110eSKevin Wolf /* If we couldn't manage to unmap while guaranteed that the area reads as 170434fa110eSKevin Wolf * all-zero afterwards, just write zeroes without unmapping */ 1705b3ac2b94SSimran Singhal return handle_aiocb_write_zeroes(aiocb); 170634fa110eSKevin Wolf } 170734fa110eSKevin Wolf 17081efad060SFam Zheng #ifndef HAVE_COPY_FILE_RANGE 17091efad060SFam Zheng static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, 17101efad060SFam Zheng off_t *out_off, size_t len, unsigned int flags) 17111efad060SFam Zheng { 17121efad060SFam Zheng #ifdef __NR_copy_file_range 17131efad060SFam Zheng return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, 17141efad060SFam Zheng out_off, len, flags); 17151efad060SFam Zheng #else 17161efad060SFam Zheng errno = ENOSYS; 17171efad060SFam Zheng return -1; 17181efad060SFam Zheng #endif 17191efad060SFam Zheng } 17201efad060SFam Zheng #endif 17211efad060SFam Zheng 172258a209c4SKevin Wolf static int handle_aiocb_copy_range(void *opaque) 17231efad060SFam Zheng { 172458a209c4SKevin Wolf RawPosixAIOData *aiocb = opaque; 17251efad060SFam Zheng uint64_t bytes = aiocb->aio_nbytes; 17261efad060SFam Zheng off_t in_off = aiocb->aio_offset; 1727d57c44d0SKevin Wolf off_t out_off = aiocb->copy_range.aio_offset2; 17281efad060SFam Zheng 17291efad060SFam Zheng while (bytes) { 17301efad060SFam Zheng ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, 1731d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, &out_off, 17321efad060SFam Zheng bytes, 0); 1733ecc983a5SFam Zheng trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, 1734d57c44d0SKevin Wolf aiocb->copy_range.aio_fd2, out_off, bytes, 1735d57c44d0SKevin Wolf 0, ret); 1736c436e3d0SFam Zheng if (ret == 0) { 1737c436e3d0SFam Zheng /* No progress (e.g. when beyond EOF), let the caller fall back to 1738c436e3d0SFam Zheng * buffer I/O. */ 1739c436e3d0SFam Zheng return -ENOSPC; 17401efad060SFam Zheng } 17411efad060SFam Zheng if (ret < 0) { 1742c436e3d0SFam Zheng switch (errno) { 1743c436e3d0SFam Zheng case ENOSYS: 17441efad060SFam Zheng return -ENOTSUP; 1745c436e3d0SFam Zheng case EINTR: 1746c436e3d0SFam Zheng continue; 1747c436e3d0SFam Zheng default: 17481efad060SFam Zheng return -errno; 17491efad060SFam Zheng } 17501efad060SFam Zheng } 17511efad060SFam Zheng bytes -= ret; 17521efad060SFam Zheng } 17531efad060SFam Zheng return 0; 17541efad060SFam Zheng } 17551efad060SFam Zheng 175646ee0f46SKevin Wolf static int handle_aiocb_discard(void *opaque) 1757c1bb86cdSEric Blake { 175846ee0f46SKevin Wolf RawPosixAIOData *aiocb = opaque; 1759c1bb86cdSEric Blake int ret = -EOPNOTSUPP; 1760c1bb86cdSEric Blake BDRVRawState *s = aiocb->bs->opaque; 1761c1bb86cdSEric Blake 1762c1bb86cdSEric Blake if (!s->has_discard) { 1763c1bb86cdSEric Blake return -ENOTSUP; 1764c1bb86cdSEric Blake } 1765c1bb86cdSEric Blake 1766c1bb86cdSEric Blake if (aiocb->aio_type & QEMU_AIO_BLKDEV) { 1767c1bb86cdSEric Blake #ifdef BLKDISCARD 1768c1bb86cdSEric Blake do { 1769c1bb86cdSEric Blake uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; 1770c1bb86cdSEric Blake if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { 1771c1bb86cdSEric Blake return 0; 1772c1bb86cdSEric Blake } 1773c1bb86cdSEric Blake } while (errno == EINTR); 1774c1bb86cdSEric Blake 1775c1bb86cdSEric Blake ret = -errno; 1776c1bb86cdSEric Blake #endif 1777c1bb86cdSEric Blake } else { 1778c1bb86cdSEric Blake #ifdef CONFIG_FALLOCATE_PUNCH_HOLE 1779c1bb86cdSEric Blake ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 1780c1bb86cdSEric Blake aiocb->aio_offset, aiocb->aio_nbytes); 1781c1bb86cdSEric Blake #endif 1782c1bb86cdSEric Blake } 1783c1bb86cdSEric Blake 1784c1bb86cdSEric Blake ret = translate_err(ret); 1785c1bb86cdSEric Blake if (ret == -ENOTSUP) { 1786c1bb86cdSEric Blake s->has_discard = false; 1787c1bb86cdSEric Blake } 1788c1bb86cdSEric Blake return ret; 1789c1bb86cdSEric Blake } 1790c1bb86cdSEric Blake 17913a20013fSNir Soffer /* 17923a20013fSNir Soffer * Help alignment probing by allocating the first block. 17933a20013fSNir Soffer * 17943a20013fSNir Soffer * When reading with direct I/O from unallocated area on Gluster backed by XFS, 17953a20013fSNir Soffer * reading succeeds regardless of request length. In this case we fallback to 17963a20013fSNir Soffer * safe alignment which is not optimal. Allocating the first block avoids this 17973a20013fSNir Soffer * fallback. 17983a20013fSNir Soffer * 17993a20013fSNir Soffer * fd may be opened with O_DIRECT, but we don't know the buffer alignment or 18003a20013fSNir Soffer * request alignment, so we use safe values. 18013a20013fSNir Soffer * 18023a20013fSNir Soffer * Returns: 0 on success, -errno on failure. Since this is an optimization, 18033a20013fSNir Soffer * caller may ignore failures. 18043a20013fSNir Soffer */ 18053a20013fSNir Soffer static int allocate_first_block(int fd, size_t max_size) 18063a20013fSNir Soffer { 18073a20013fSNir Soffer size_t write_size = (max_size < MAX_BLOCKSIZE) 18083a20013fSNir Soffer ? BDRV_SECTOR_SIZE 18093a20013fSNir Soffer : MAX_BLOCKSIZE; 1810038adc2fSWei Yang size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); 18113a20013fSNir Soffer void *buf; 18123a20013fSNir Soffer ssize_t n; 18133a20013fSNir Soffer int ret; 18143a20013fSNir Soffer 18153a20013fSNir Soffer buf = qemu_memalign(max_align, write_size); 18163a20013fSNir Soffer memset(buf, 0, write_size); 18173a20013fSNir Soffer 18183a20013fSNir Soffer do { 18193a20013fSNir Soffer n = pwrite(fd, buf, write_size, 0); 18203a20013fSNir Soffer } while (n == -1 && errno == EINTR); 18213a20013fSNir Soffer 18223a20013fSNir Soffer ret = (n == -1) ? -errno : 0; 18233a20013fSNir Soffer 18243a20013fSNir Soffer qemu_vfree(buf); 18253a20013fSNir Soffer return ret; 18263a20013fSNir Soffer } 18273a20013fSNir Soffer 182829cb4c01SKevin Wolf static int handle_aiocb_truncate(void *opaque) 182993f4e2ffSKevin Wolf { 183029cb4c01SKevin Wolf RawPosixAIOData *aiocb = opaque; 183193f4e2ffSKevin Wolf int result = 0; 183293f4e2ffSKevin Wolf int64_t current_length = 0; 183393f4e2ffSKevin Wolf char *buf = NULL; 183493f4e2ffSKevin Wolf struct stat st; 183593f4e2ffSKevin Wolf int fd = aiocb->aio_fildes; 183693f4e2ffSKevin Wolf int64_t offset = aiocb->aio_offset; 1837d57c44d0SKevin Wolf PreallocMode prealloc = aiocb->truncate.prealloc; 1838d57c44d0SKevin Wolf Error **errp = aiocb->truncate.errp; 183993f4e2ffSKevin Wolf 184093f4e2ffSKevin Wolf if (fstat(fd, &st) < 0) { 184193f4e2ffSKevin Wolf result = -errno; 184293f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not stat file"); 184393f4e2ffSKevin Wolf return result; 184493f4e2ffSKevin Wolf } 184593f4e2ffSKevin Wolf 184693f4e2ffSKevin Wolf current_length = st.st_size; 1847d57c44d0SKevin Wolf if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { 184893f4e2ffSKevin Wolf error_setg(errp, "Cannot use preallocation for shrinking files"); 184993f4e2ffSKevin Wolf return -ENOTSUP; 185093f4e2ffSKevin Wolf } 185193f4e2ffSKevin Wolf 1852d57c44d0SKevin Wolf switch (prealloc) { 185393f4e2ffSKevin Wolf #ifdef CONFIG_POSIX_FALLOCATE 185493f4e2ffSKevin Wolf case PREALLOC_MODE_FALLOC: 185593f4e2ffSKevin Wolf /* 185693f4e2ffSKevin Wolf * Truncating before posix_fallocate() makes it about twice slower on 185793f4e2ffSKevin Wolf * file systems that do not support fallocate(), trying to check if a 185893f4e2ffSKevin Wolf * block is allocated before allocating it, so don't do that here. 185993f4e2ffSKevin Wolf */ 186093f4e2ffSKevin Wolf if (offset != current_length) { 186193f4e2ffSKevin Wolf result = -posix_fallocate(fd, current_length, 186293f4e2ffSKevin Wolf offset - current_length); 186393f4e2ffSKevin Wolf if (result != 0) { 186493f4e2ffSKevin Wolf /* posix_fallocate() doesn't set errno. */ 186593f4e2ffSKevin Wolf error_setg_errno(errp, -result, 186693f4e2ffSKevin Wolf "Could not preallocate new data"); 18673a20013fSNir Soffer } else if (current_length == 0) { 18683a20013fSNir Soffer /* 18693a20013fSNir Soffer * posix_fallocate() uses fallocate() if the filesystem 18703a20013fSNir Soffer * supports it, or fallback to manually writing zeroes. If 18713a20013fSNir Soffer * fallocate() was used, unaligned reads from the fallocated 18723a20013fSNir Soffer * area in raw_probe_alignment() will succeed, hence we need to 18733a20013fSNir Soffer * allocate the first block. 18743a20013fSNir Soffer * 18753a20013fSNir Soffer * Optimize future alignment probing; ignore failures. 18763a20013fSNir Soffer */ 18773a20013fSNir Soffer allocate_first_block(fd, offset); 187893f4e2ffSKevin Wolf } 187993f4e2ffSKevin Wolf } else { 188093f4e2ffSKevin Wolf result = 0; 188193f4e2ffSKevin Wolf } 188293f4e2ffSKevin Wolf goto out; 188393f4e2ffSKevin Wolf #endif 188493f4e2ffSKevin Wolf case PREALLOC_MODE_FULL: 188593f4e2ffSKevin Wolf { 188693f4e2ffSKevin Wolf int64_t num = 0, left = offset - current_length; 188793f4e2ffSKevin Wolf off_t seek_result; 188893f4e2ffSKevin Wolf 188993f4e2ffSKevin Wolf /* 189093f4e2ffSKevin Wolf * Knowing the final size from the beginning could allow the file 189193f4e2ffSKevin Wolf * system driver to do less allocations and possibly avoid 189293f4e2ffSKevin Wolf * fragmentation of the file. 189393f4e2ffSKevin Wolf */ 189493f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 189593f4e2ffSKevin Wolf result = -errno; 189693f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 189793f4e2ffSKevin Wolf goto out; 189893f4e2ffSKevin Wolf } 189993f4e2ffSKevin Wolf 190093f4e2ffSKevin Wolf buf = g_malloc0(65536); 190193f4e2ffSKevin Wolf 190293f4e2ffSKevin Wolf seek_result = lseek(fd, current_length, SEEK_SET); 190393f4e2ffSKevin Wolf if (seek_result < 0) { 190493f4e2ffSKevin Wolf result = -errno; 190593f4e2ffSKevin Wolf error_setg_errno(errp, -result, 190693f4e2ffSKevin Wolf "Failed to seek to the old end of file"); 190793f4e2ffSKevin Wolf goto out; 190893f4e2ffSKevin Wolf } 190993f4e2ffSKevin Wolf 191093f4e2ffSKevin Wolf while (left > 0) { 191193f4e2ffSKevin Wolf num = MIN(left, 65536); 191293f4e2ffSKevin Wolf result = write(fd, buf, num); 191393f4e2ffSKevin Wolf if (result < 0) { 1914a1c81f4fSFam Zheng if (errno == EINTR) { 1915a1c81f4fSFam Zheng continue; 1916a1c81f4fSFam Zheng } 191793f4e2ffSKevin Wolf result = -errno; 191893f4e2ffSKevin Wolf error_setg_errno(errp, -result, 191993f4e2ffSKevin Wolf "Could not write zeros for preallocation"); 192093f4e2ffSKevin Wolf goto out; 192193f4e2ffSKevin Wolf } 192293f4e2ffSKevin Wolf left -= result; 192393f4e2ffSKevin Wolf } 192493f4e2ffSKevin Wolf if (result >= 0) { 192593f4e2ffSKevin Wolf result = fsync(fd); 192693f4e2ffSKevin Wolf if (result < 0) { 192793f4e2ffSKevin Wolf result = -errno; 192893f4e2ffSKevin Wolf error_setg_errno(errp, -result, 192993f4e2ffSKevin Wolf "Could not flush file to disk"); 193093f4e2ffSKevin Wolf goto out; 193193f4e2ffSKevin Wolf } 193293f4e2ffSKevin Wolf } 193393f4e2ffSKevin Wolf goto out; 193493f4e2ffSKevin Wolf } 193593f4e2ffSKevin Wolf case PREALLOC_MODE_OFF: 193693f4e2ffSKevin Wolf if (ftruncate(fd, offset) != 0) { 193793f4e2ffSKevin Wolf result = -errno; 193893f4e2ffSKevin Wolf error_setg_errno(errp, -result, "Could not resize file"); 19393a20013fSNir Soffer } else if (current_length == 0 && offset > current_length) { 19403a20013fSNir Soffer /* Optimize future alignment probing; ignore failures. */ 19413a20013fSNir Soffer allocate_first_block(fd, offset); 194293f4e2ffSKevin Wolf } 194393f4e2ffSKevin Wolf return result; 194493f4e2ffSKevin Wolf default: 194593f4e2ffSKevin Wolf result = -ENOTSUP; 194693f4e2ffSKevin Wolf error_setg(errp, "Unsupported preallocation mode: %s", 1947d57c44d0SKevin Wolf PreallocMode_str(prealloc)); 194893f4e2ffSKevin Wolf return result; 194993f4e2ffSKevin Wolf } 195093f4e2ffSKevin Wolf 195193f4e2ffSKevin Wolf out: 195293f4e2ffSKevin Wolf if (result < 0) { 195393f4e2ffSKevin Wolf if (ftruncate(fd, current_length) < 0) { 195493f4e2ffSKevin Wolf error_report("Failed to restore old file length: %s", 195593f4e2ffSKevin Wolf strerror(errno)); 195693f4e2ffSKevin Wolf } 195793f4e2ffSKevin Wolf } 195893f4e2ffSKevin Wolf 195993f4e2ffSKevin Wolf g_free(buf); 196093f4e2ffSKevin Wolf return result; 196193f4e2ffSKevin Wolf } 196293f4e2ffSKevin Wolf 19635d5de250SKevin Wolf static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs, 19645d5de250SKevin Wolf ThreadPoolFunc func, void *arg) 19655d5de250SKevin Wolf { 19665d5de250SKevin Wolf /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ 19675d5de250SKevin Wolf ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); 19685d5de250SKevin Wolf return thread_pool_submit_co(pool, func, arg); 19695d5de250SKevin Wolf } 19705d5de250SKevin Wolf 1971c1bb86cdSEric Blake static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, 1972c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, int type) 1973c1bb86cdSEric Blake { 1974c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 1975999e6b69SKevin Wolf RawPosixAIOData acb; 1976c1bb86cdSEric Blake 1977c1bb86cdSEric Blake if (fd_open(bs) < 0) 1978c1bb86cdSEric Blake return -EIO; 1979c1bb86cdSEric Blake 1980c1bb86cdSEric Blake /* 1981c6447510SAarushi Mehta * When using O_DIRECT, the request must be aligned to be able to use 1982c6447510SAarushi Mehta * either libaio or io_uring interface. If not fail back to regular thread 1983c6447510SAarushi Mehta * pool read/write code which emulates this for us if we 1984c6447510SAarushi Mehta * set QEMU_AIO_MISALIGNED. 1985c1bb86cdSEric Blake */ 1986c6447510SAarushi Mehta if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) { 1987c1bb86cdSEric Blake type |= QEMU_AIO_MISALIGNED; 1988c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 1989c6447510SAarushi Mehta } else if (s->use_linux_io_uring) { 1990c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 1991c6447510SAarushi Mehta assert(qiov->size == bytes); 1992c6447510SAarushi Mehta return luring_co_submit(bs, aio, s->fd, offset, qiov, type); 1993c6447510SAarushi Mehta #endif 1994c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 1995c1bb86cdSEric Blake } else if (s->use_linux_aio) { 1996c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 1997c1bb86cdSEric Blake assert(qiov->size == bytes); 1998c1bb86cdSEric Blake return laio_co_submit(bs, aio, s->fd, offset, qiov, type); 1999c1bb86cdSEric Blake #endif 2000c1bb86cdSEric Blake } 2001c1bb86cdSEric Blake 2002999e6b69SKevin Wolf acb = (RawPosixAIOData) { 2003999e6b69SKevin Wolf .bs = bs, 2004999e6b69SKevin Wolf .aio_fildes = s->fd, 2005999e6b69SKevin Wolf .aio_type = type, 2006999e6b69SKevin Wolf .aio_offset = offset, 2007999e6b69SKevin Wolf .aio_nbytes = bytes, 2008999e6b69SKevin Wolf .io = { 2009999e6b69SKevin Wolf .iov = qiov->iov, 2010999e6b69SKevin Wolf .niov = qiov->niov, 2011999e6b69SKevin Wolf }, 2012999e6b69SKevin Wolf }; 2013999e6b69SKevin Wolf 2014999e6b69SKevin Wolf assert(qiov->size == bytes); 2015999e6b69SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb); 2016c1bb86cdSEric Blake } 2017c1bb86cdSEric Blake 2018c1bb86cdSEric Blake static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset, 2019c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 2020c1bb86cdSEric Blake int flags) 2021c1bb86cdSEric Blake { 2022c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); 2023c1bb86cdSEric Blake } 2024c1bb86cdSEric Blake 2025c1bb86cdSEric Blake static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset, 2026c1bb86cdSEric Blake uint64_t bytes, QEMUIOVector *qiov, 2027c1bb86cdSEric Blake int flags) 2028c1bb86cdSEric Blake { 2029c1bb86cdSEric Blake assert(flags == 0); 2030c1bb86cdSEric Blake return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); 2031c1bb86cdSEric Blake } 2032c1bb86cdSEric Blake 2033c1bb86cdSEric Blake static void raw_aio_plug(BlockDriverState *bs) 2034c1bb86cdSEric Blake { 2035c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 2036c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 2037c1bb86cdSEric Blake if (s->use_linux_aio) { 2038c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 2039c1bb86cdSEric Blake laio_io_plug(bs, aio); 2040c1bb86cdSEric Blake } 2041c1bb86cdSEric Blake #endif 2042c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2043c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2044c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2045c6447510SAarushi Mehta luring_io_plug(bs, aio); 2046c6447510SAarushi Mehta } 2047c6447510SAarushi Mehta #endif 2048c1bb86cdSEric Blake } 2049c1bb86cdSEric Blake 2050c1bb86cdSEric Blake static void raw_aio_unplug(BlockDriverState *bs) 2051c1bb86cdSEric Blake { 2052c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 2053c1bb86cdSEric Blake #ifdef CONFIG_LINUX_AIO 2054c1bb86cdSEric Blake if (s->use_linux_aio) { 2055c1bb86cdSEric Blake LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); 2056c1bb86cdSEric Blake laio_io_unplug(bs, aio); 2057c1bb86cdSEric Blake } 2058c1bb86cdSEric Blake #endif 2059c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2060c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2061c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2062c6447510SAarushi Mehta luring_io_unplug(bs, aio); 2063c6447510SAarushi Mehta } 2064c6447510SAarushi Mehta #endif 2065c1bb86cdSEric Blake } 2066c1bb86cdSEric Blake 206733d70fb6SKevin Wolf static int raw_co_flush_to_disk(BlockDriverState *bs) 2068c1bb86cdSEric Blake { 2069c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 207006dc9bd5SKevin Wolf RawPosixAIOData acb; 207133d70fb6SKevin Wolf int ret; 2072c1bb86cdSEric Blake 207333d70fb6SKevin Wolf ret = fd_open(bs); 207433d70fb6SKevin Wolf if (ret < 0) { 207533d70fb6SKevin Wolf return ret; 207633d70fb6SKevin Wolf } 2077c1bb86cdSEric Blake 207806dc9bd5SKevin Wolf acb = (RawPosixAIOData) { 207906dc9bd5SKevin Wolf .bs = bs, 208006dc9bd5SKevin Wolf .aio_fildes = s->fd, 208106dc9bd5SKevin Wolf .aio_type = QEMU_AIO_FLUSH, 208206dc9bd5SKevin Wolf }; 208306dc9bd5SKevin Wolf 2084c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2085c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2086c6447510SAarushi Mehta LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); 2087c6447510SAarushi Mehta return luring_co_submit(bs, aio, s->fd, 0, NULL, QEMU_AIO_FLUSH); 2088c6447510SAarushi Mehta } 2089c6447510SAarushi Mehta #endif 209006dc9bd5SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb); 2091c1bb86cdSEric Blake } 2092c1bb86cdSEric Blake 2093ed6e2161SNishanth Aravamudan static void raw_aio_attach_aio_context(BlockDriverState *bs, 2094ed6e2161SNishanth Aravamudan AioContext *new_context) 2095ed6e2161SNishanth Aravamudan { 2096c6447510SAarushi Mehta BDRVRawState __attribute__((unused)) *s = bs->opaque; 2097ed6e2161SNishanth Aravamudan #ifdef CONFIG_LINUX_AIO 2098ed6e2161SNishanth Aravamudan if (s->use_linux_aio) { 2099cb09104eSMarkus Armbruster Error *local_err = NULL; 2100ed6e2161SNishanth Aravamudan if (!aio_setup_linux_aio(new_context, &local_err)) { 2101ed6e2161SNishanth Aravamudan error_reportf_err(local_err, "Unable to use native AIO, " 2102ed6e2161SNishanth Aravamudan "falling back to thread pool: "); 2103ed6e2161SNishanth Aravamudan s->use_linux_aio = false; 2104ed6e2161SNishanth Aravamudan } 2105ed6e2161SNishanth Aravamudan } 2106ed6e2161SNishanth Aravamudan #endif 2107c6447510SAarushi Mehta #ifdef CONFIG_LINUX_IO_URING 2108c6447510SAarushi Mehta if (s->use_linux_io_uring) { 2109c6447510SAarushi Mehta Error *local_err; 2110c6447510SAarushi Mehta if (!aio_setup_linux_io_uring(new_context, &local_err)) { 2111c6447510SAarushi Mehta error_reportf_err(local_err, "Unable to use linux io_uring, " 2112c6447510SAarushi Mehta "falling back to thread pool: "); 2113c6447510SAarushi Mehta s->use_linux_io_uring = false; 2114c6447510SAarushi Mehta } 2115c6447510SAarushi Mehta } 2116c6447510SAarushi Mehta #endif 2117ed6e2161SNishanth Aravamudan } 2118ed6e2161SNishanth Aravamudan 2119c1bb86cdSEric Blake static void raw_close(BlockDriverState *bs) 2120c1bb86cdSEric Blake { 2121c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2122c1bb86cdSEric Blake 2123c1bb86cdSEric Blake if (s->fd >= 0) { 2124c1bb86cdSEric Blake qemu_close(s->fd); 2125c1bb86cdSEric Blake s->fd = -1; 2126c1bb86cdSEric Blake } 2127c1bb86cdSEric Blake } 2128c1bb86cdSEric Blake 2129d0bc9e5dSMax Reitz /** 2130d0bc9e5dSMax Reitz * Truncates the given regular file @fd to @offset and, when growing, fills the 2131d0bc9e5dSMax Reitz * new space according to @prealloc. 2132d0bc9e5dSMax Reitz * 2133d0bc9e5dSMax Reitz * Returns: 0 on success, -errno on failure. 2134d0bc9e5dSMax Reitz */ 213593f4e2ffSKevin Wolf static int coroutine_fn 213693f4e2ffSKevin Wolf raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, 213793f4e2ffSKevin Wolf PreallocMode prealloc, Error **errp) 21389f63b07eSMax Reitz { 213929cb4c01SKevin Wolf RawPosixAIOData acb; 2140d0bc9e5dSMax Reitz 214129cb4c01SKevin Wolf acb = (RawPosixAIOData) { 214293f4e2ffSKevin Wolf .bs = bs, 214393f4e2ffSKevin Wolf .aio_fildes = fd, 214493f4e2ffSKevin Wolf .aio_type = QEMU_AIO_TRUNCATE, 214593f4e2ffSKevin Wolf .aio_offset = offset, 2146d57c44d0SKevin Wolf .truncate = { 214793f4e2ffSKevin Wolf .prealloc = prealloc, 214893f4e2ffSKevin Wolf .errp = errp, 2149d57c44d0SKevin Wolf }, 215093f4e2ffSKevin Wolf }; 2151d0bc9e5dSMax Reitz 215229cb4c01SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb); 21539f63b07eSMax Reitz } 21549f63b07eSMax Reitz 2155061ca8a3SKevin Wolf static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, 2156c80d8b06SMax Reitz bool exact, PreallocMode prealloc, 215792b92799SKevin Wolf BdrvRequestFlags flags, Error **errp) 2158c1bb86cdSEric Blake { 2159c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2160c1bb86cdSEric Blake struct stat st; 2161f59adb32SMax Reitz int ret; 2162c1bb86cdSEric Blake 2163c1bb86cdSEric Blake if (fstat(s->fd, &st)) { 2164f59adb32SMax Reitz ret = -errno; 2165f59adb32SMax Reitz error_setg_errno(errp, -ret, "Failed to fstat() the file"); 2166f59adb32SMax Reitz return ret; 2167c1bb86cdSEric Blake } 2168c1bb86cdSEric Blake 2169c1bb86cdSEric Blake if (S_ISREG(st.st_mode)) { 217082325ae5SMax Reitz /* Always resizes to the exact @offset */ 217193f4e2ffSKevin Wolf return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); 2172c1bb86cdSEric Blake } 217335d72602SMax Reitz 217435d72602SMax Reitz if (prealloc != PREALLOC_MODE_OFF) { 217535d72602SMax Reitz error_setg(errp, "Preallocation mode '%s' unsupported for this " 2176977c736fSMarkus Armbruster "non-regular file", PreallocMode_str(prealloc)); 217735d72602SMax Reitz return -ENOTSUP; 217835d72602SMax Reitz } 217935d72602SMax Reitz 218035d72602SMax Reitz if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 218182325ae5SMax Reitz int64_t cur_length = raw_getlength(bs); 218282325ae5SMax Reitz 218382325ae5SMax Reitz if (offset != cur_length && exact) { 218482325ae5SMax Reitz error_setg(errp, "Cannot resize device files"); 218582325ae5SMax Reitz return -ENOTSUP; 218682325ae5SMax Reitz } else if (offset > cur_length) { 2187f59adb32SMax Reitz error_setg(errp, "Cannot grow device files"); 2188c1bb86cdSEric Blake return -EINVAL; 2189c1bb86cdSEric Blake } 2190c1bb86cdSEric Blake } else { 2191f59adb32SMax Reitz error_setg(errp, "Resizing this file is not supported"); 2192c1bb86cdSEric Blake return -ENOTSUP; 2193c1bb86cdSEric Blake } 2194c1bb86cdSEric Blake 2195c1bb86cdSEric Blake return 0; 2196c1bb86cdSEric Blake } 2197c1bb86cdSEric Blake 2198c1bb86cdSEric Blake #ifdef __OpenBSD__ 2199c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2200c1bb86cdSEric Blake { 2201c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2202c1bb86cdSEric Blake int fd = s->fd; 2203c1bb86cdSEric Blake struct stat st; 2204c1bb86cdSEric Blake 2205c1bb86cdSEric Blake if (fstat(fd, &st)) 2206c1bb86cdSEric Blake return -errno; 2207c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2208c1bb86cdSEric Blake struct disklabel dl; 2209c1bb86cdSEric Blake 2210c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2211c1bb86cdSEric Blake return -errno; 2212c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2213c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2214c1bb86cdSEric Blake } else 2215c1bb86cdSEric Blake return st.st_size; 2216c1bb86cdSEric Blake } 2217c1bb86cdSEric Blake #elif defined(__NetBSD__) 2218c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2219c1bb86cdSEric Blake { 2220c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2221c1bb86cdSEric Blake int fd = s->fd; 2222c1bb86cdSEric Blake struct stat st; 2223c1bb86cdSEric Blake 2224c1bb86cdSEric Blake if (fstat(fd, &st)) 2225c1bb86cdSEric Blake return -errno; 2226c1bb86cdSEric Blake if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { 2227c1bb86cdSEric Blake struct dkwedge_info dkw; 2228c1bb86cdSEric Blake 2229c1bb86cdSEric Blake if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { 2230c1bb86cdSEric Blake return dkw.dkw_size * 512; 2231c1bb86cdSEric Blake } else { 2232c1bb86cdSEric Blake struct disklabel dl; 2233c1bb86cdSEric Blake 2234c1bb86cdSEric Blake if (ioctl(fd, DIOCGDINFO, &dl)) 2235c1bb86cdSEric Blake return -errno; 2236c1bb86cdSEric Blake return (uint64_t)dl.d_secsize * 2237c1bb86cdSEric Blake dl.d_partitions[DISKPART(st.st_rdev)].p_size; 2238c1bb86cdSEric Blake } 2239c1bb86cdSEric Blake } else 2240c1bb86cdSEric Blake return st.st_size; 2241c1bb86cdSEric Blake } 2242c1bb86cdSEric Blake #elif defined(__sun__) 2243c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2244c1bb86cdSEric Blake { 2245c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2246c1bb86cdSEric Blake struct dk_minfo minfo; 2247c1bb86cdSEric Blake int ret; 2248c1bb86cdSEric Blake int64_t size; 2249c1bb86cdSEric Blake 2250c1bb86cdSEric Blake ret = fd_open(bs); 2251c1bb86cdSEric Blake if (ret < 0) { 2252c1bb86cdSEric Blake return ret; 2253c1bb86cdSEric Blake } 2254c1bb86cdSEric Blake 2255c1bb86cdSEric Blake /* 2256c1bb86cdSEric Blake * Use the DKIOCGMEDIAINFO ioctl to read the size. 2257c1bb86cdSEric Blake */ 2258c1bb86cdSEric Blake ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); 2259c1bb86cdSEric Blake if (ret != -1) { 2260c1bb86cdSEric Blake return minfo.dki_lbsize * minfo.dki_capacity; 2261c1bb86cdSEric Blake } 2262c1bb86cdSEric Blake 2263c1bb86cdSEric Blake /* 2264c1bb86cdSEric Blake * There are reports that lseek on some devices fails, but 2265c1bb86cdSEric Blake * irc discussion said that contingency on contingency was overkill. 2266c1bb86cdSEric Blake */ 2267c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2268c1bb86cdSEric Blake if (size < 0) { 2269c1bb86cdSEric Blake return -errno; 2270c1bb86cdSEric Blake } 2271c1bb86cdSEric Blake return size; 2272c1bb86cdSEric Blake } 2273c1bb86cdSEric Blake #elif defined(CONFIG_BSD) 2274c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2275c1bb86cdSEric Blake { 2276c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2277c1bb86cdSEric Blake int fd = s->fd; 2278c1bb86cdSEric Blake int64_t size; 2279c1bb86cdSEric Blake struct stat sb; 2280c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2281c1bb86cdSEric Blake int reopened = 0; 2282c1bb86cdSEric Blake #endif 2283c1bb86cdSEric Blake int ret; 2284c1bb86cdSEric Blake 2285c1bb86cdSEric Blake ret = fd_open(bs); 2286c1bb86cdSEric Blake if (ret < 0) 2287c1bb86cdSEric Blake return ret; 2288c1bb86cdSEric Blake 2289c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 2290c1bb86cdSEric Blake again: 2291c1bb86cdSEric Blake #endif 2292c1bb86cdSEric Blake if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { 2293c1bb86cdSEric Blake #ifdef DIOCGMEDIASIZE 2294c1bb86cdSEric Blake if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) 2295c1bb86cdSEric Blake #elif defined(DIOCGPART) 2296c1bb86cdSEric Blake { 2297c1bb86cdSEric Blake struct partinfo pi; 2298c1bb86cdSEric Blake if (ioctl(fd, DIOCGPART, &pi) == 0) 2299c1bb86cdSEric Blake size = pi.media_size; 2300c1bb86cdSEric Blake else 2301c1bb86cdSEric Blake size = 0; 2302c1bb86cdSEric Blake } 2303c1bb86cdSEric Blake if (size == 0) 2304c1bb86cdSEric Blake #endif 2305c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 2306c1bb86cdSEric Blake { 2307c1bb86cdSEric Blake uint64_t sectors = 0; 2308c1bb86cdSEric Blake uint32_t sector_size = 0; 2309c1bb86cdSEric Blake 2310c1bb86cdSEric Blake if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 2311c1bb86cdSEric Blake && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { 2312c1bb86cdSEric Blake size = sectors * sector_size; 2313c1bb86cdSEric Blake } else { 2314c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2315c1bb86cdSEric Blake if (size < 0) { 2316c1bb86cdSEric Blake return -errno; 2317c1bb86cdSEric Blake } 2318c1bb86cdSEric Blake } 2319c1bb86cdSEric Blake } 2320c1bb86cdSEric Blake #else 2321c1bb86cdSEric Blake size = lseek(fd, 0LL, SEEK_END); 2322c1bb86cdSEric Blake if (size < 0) { 2323c1bb86cdSEric Blake return -errno; 2324c1bb86cdSEric Blake } 2325c1bb86cdSEric Blake #endif 2326c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 2327c1bb86cdSEric Blake switch(s->type) { 2328c1bb86cdSEric Blake case FTYPE_CD: 2329c1bb86cdSEric Blake /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ 2330c1bb86cdSEric Blake if (size == 2048LL * (unsigned)-1) 2331c1bb86cdSEric Blake size = 0; 2332c1bb86cdSEric Blake /* XXX no disc? maybe we need to reopen... */ 2333c1bb86cdSEric Blake if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { 2334c1bb86cdSEric Blake reopened = 1; 2335c1bb86cdSEric Blake goto again; 2336c1bb86cdSEric Blake } 2337c1bb86cdSEric Blake } 2338c1bb86cdSEric Blake #endif 2339c1bb86cdSEric Blake } else { 2340c1bb86cdSEric Blake size = lseek(fd, 0, SEEK_END); 2341c1bb86cdSEric Blake if (size < 0) { 2342c1bb86cdSEric Blake return -errno; 2343c1bb86cdSEric Blake } 2344c1bb86cdSEric Blake } 2345c1bb86cdSEric Blake return size; 2346c1bb86cdSEric Blake } 2347c1bb86cdSEric Blake #else 2348c1bb86cdSEric Blake static int64_t raw_getlength(BlockDriverState *bs) 2349c1bb86cdSEric Blake { 2350c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2351c1bb86cdSEric Blake int ret; 2352c1bb86cdSEric Blake int64_t size; 2353c1bb86cdSEric Blake 2354c1bb86cdSEric Blake ret = fd_open(bs); 2355c1bb86cdSEric Blake if (ret < 0) { 2356c1bb86cdSEric Blake return ret; 2357c1bb86cdSEric Blake } 2358c1bb86cdSEric Blake 2359c1bb86cdSEric Blake size = lseek(s->fd, 0, SEEK_END); 2360c1bb86cdSEric Blake if (size < 0) { 2361c1bb86cdSEric Blake return -errno; 2362c1bb86cdSEric Blake } 2363c1bb86cdSEric Blake return size; 2364c1bb86cdSEric Blake } 2365c1bb86cdSEric Blake #endif 2366c1bb86cdSEric Blake 2367c1bb86cdSEric Blake static int64_t raw_get_allocated_file_size(BlockDriverState *bs) 2368c1bb86cdSEric Blake { 2369c1bb86cdSEric Blake struct stat st; 2370c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2371c1bb86cdSEric Blake 2372c1bb86cdSEric Blake if (fstat(s->fd, &st) < 0) { 2373c1bb86cdSEric Blake return -errno; 2374c1bb86cdSEric Blake } 2375c1bb86cdSEric Blake return (int64_t)st.st_blocks * 512; 2376c1bb86cdSEric Blake } 2377c1bb86cdSEric Blake 237893f4e2ffSKevin Wolf static int coroutine_fn 237993f4e2ffSKevin Wolf raw_co_create(BlockdevCreateOptions *options, Error **errp) 2380c1bb86cdSEric Blake { 2381927f11e1SKevin Wolf BlockdevCreateOptionsFile *file_opts; 23827c20c808SMax Reitz Error *local_err = NULL; 2383c1bb86cdSEric Blake int fd; 2384d815efcaSMax Reitz uint64_t perm, shared; 2385c1bb86cdSEric Blake int result = 0; 2386c1bb86cdSEric Blake 2387927f11e1SKevin Wolf /* Validate options and set default values */ 2388927f11e1SKevin Wolf assert(options->driver == BLOCKDEV_DRIVER_FILE); 2389927f11e1SKevin Wolf file_opts = &options->u.file; 2390c1bb86cdSEric Blake 2391927f11e1SKevin Wolf if (!file_opts->has_nocow) { 2392927f11e1SKevin Wolf file_opts->nocow = false; 2393927f11e1SKevin Wolf } 2394927f11e1SKevin Wolf if (!file_opts->has_preallocation) { 2395927f11e1SKevin Wolf file_opts->preallocation = PREALLOC_MODE_OFF; 2396c1bb86cdSEric Blake } 2397ffa244c8SKevin Wolf if (!file_opts->has_extent_size_hint) { 2398ffa244c8SKevin Wolf file_opts->extent_size_hint = 1 * MiB; 2399ffa244c8SKevin Wolf } 2400ffa244c8SKevin Wolf if (file_opts->extent_size_hint > UINT32_MAX) { 2401ffa244c8SKevin Wolf result = -EINVAL; 2402ffa244c8SKevin Wolf error_setg(errp, "Extent size hint is too large"); 2403ffa244c8SKevin Wolf goto out; 2404ffa244c8SKevin Wolf } 2405c1bb86cdSEric Blake 2406927f11e1SKevin Wolf /* Create file */ 2407b8cf1913SMax Reitz fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_BINARY, 0644); 2408c1bb86cdSEric Blake if (fd < 0) { 2409c1bb86cdSEric Blake result = -errno; 2410c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not create file"); 2411c1bb86cdSEric Blake goto out; 2412c1bb86cdSEric Blake } 2413c1bb86cdSEric Blake 2414b8cf1913SMax Reitz /* Take permissions: We want to discard everything, so we need 2415b8cf1913SMax Reitz * BLK_PERM_WRITE; and truncation to the desired size requires 2416b8cf1913SMax Reitz * BLK_PERM_RESIZE. 2417b8cf1913SMax Reitz * On the other hand, we cannot share the RESIZE permission 2418b8cf1913SMax Reitz * because we promise that after this function, the file has the 2419b8cf1913SMax Reitz * size given in the options. If someone else were to resize it 2420b8cf1913SMax Reitz * concurrently, we could not guarantee that. 2421b8cf1913SMax Reitz * Note that after this function, we can no longer guarantee that 2422b8cf1913SMax Reitz * the file is not touched by a third party, so it may be resized 2423b8cf1913SMax Reitz * then. */ 2424b8cf1913SMax Reitz perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; 2425b8cf1913SMax Reitz shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; 2426b8cf1913SMax Reitz 2427b8cf1913SMax Reitz /* Step one: Take locks */ 24282996ffadSFam Zheng result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); 2429b8cf1913SMax Reitz if (result < 0) { 2430b8cf1913SMax Reitz goto out_close; 2431b8cf1913SMax Reitz } 2432b8cf1913SMax Reitz 2433b8cf1913SMax Reitz /* Step two: Check that nobody else has taken conflicting locks */ 2434b8cf1913SMax Reitz result = raw_check_lock_bytes(fd, perm, shared, errp); 2435b8cf1913SMax Reitz if (result < 0) { 2436b857431dSFam Zheng error_append_hint(errp, 2437b857431dSFam Zheng "Is another process using the image [%s]?\n", 2438b857431dSFam Zheng file_opts->filename); 24397c20c808SMax Reitz goto out_unlock; 2440b8cf1913SMax Reitz } 2441b8cf1913SMax Reitz 2442b8cf1913SMax Reitz /* Clear the file by truncating it to 0 */ 244393f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); 2444b8cf1913SMax Reitz if (result < 0) { 24457c20c808SMax Reitz goto out_unlock; 2446b8cf1913SMax Reitz } 2447b8cf1913SMax Reitz 2448927f11e1SKevin Wolf if (file_opts->nocow) { 2449c1bb86cdSEric Blake #ifdef __linux__ 2450c1bb86cdSEric Blake /* Set NOCOW flag to solve performance issue on fs like btrfs. 2451c1bb86cdSEric Blake * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value 2452c1bb86cdSEric Blake * will be ignored since any failure of this operation should not 2453c1bb86cdSEric Blake * block the left work. 2454c1bb86cdSEric Blake */ 2455c1bb86cdSEric Blake int attr; 2456c1bb86cdSEric Blake if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { 2457c1bb86cdSEric Blake attr |= FS_NOCOW_FL; 2458c1bb86cdSEric Blake ioctl(fd, FS_IOC_SETFLAGS, &attr); 2459c1bb86cdSEric Blake } 2460c1bb86cdSEric Blake #endif 2461c1bb86cdSEric Blake } 2462ffa244c8SKevin Wolf #ifdef FS_IOC_FSSETXATTR 2463ffa244c8SKevin Wolf /* 2464ffa244c8SKevin Wolf * Try to set the extent size hint. Failure is not fatal, and a warning is 2465ffa244c8SKevin Wolf * only printed if the option was explicitly specified. 2466ffa244c8SKevin Wolf */ 2467ffa244c8SKevin Wolf { 2468ffa244c8SKevin Wolf struct fsxattr attr; 2469ffa244c8SKevin Wolf result = ioctl(fd, FS_IOC_FSGETXATTR, &attr); 2470ffa244c8SKevin Wolf if (result == 0) { 2471ffa244c8SKevin Wolf attr.fsx_xflags |= FS_XFLAG_EXTSIZE; 2472ffa244c8SKevin Wolf attr.fsx_extsize = file_opts->extent_size_hint; 2473ffa244c8SKevin Wolf result = ioctl(fd, FS_IOC_FSSETXATTR, &attr); 2474ffa244c8SKevin Wolf } 2475ffa244c8SKevin Wolf if (result < 0 && file_opts->has_extent_size_hint && 2476ffa244c8SKevin Wolf file_opts->extent_size_hint) 2477ffa244c8SKevin Wolf { 2478ffa244c8SKevin Wolf warn_report("Failed to set extent size hint: %s", 2479ffa244c8SKevin Wolf strerror(errno)); 2480ffa244c8SKevin Wolf } 2481ffa244c8SKevin Wolf } 2482ffa244c8SKevin Wolf #endif 2483c1bb86cdSEric Blake 2484b8cf1913SMax Reitz /* Resize and potentially preallocate the file to the desired 2485b8cf1913SMax Reitz * final size */ 248693f4e2ffSKevin Wolf result = raw_regular_truncate(NULL, fd, file_opts->size, 248793f4e2ffSKevin Wolf file_opts->preallocation, errp); 24889f63b07eSMax Reitz if (result < 0) { 24897c20c808SMax Reitz goto out_unlock; 24907c20c808SMax Reitz } 24917c20c808SMax Reitz 24927c20c808SMax Reitz out_unlock: 24932996ffadSFam Zheng raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); 24947c20c808SMax Reitz if (local_err) { 24957c20c808SMax Reitz /* The above call should not fail, and if it does, that does 24967c20c808SMax Reitz * not mean the whole creation operation has failed. So 24977c20c808SMax Reitz * report it the user for their convenience, but do not report 24987c20c808SMax Reitz * it to the caller. */ 2499db0754dfSFam Zheng warn_report_err(local_err); 25005a1dad9dSNir Soffer } 25015a1dad9dSNir Soffer 25025a1dad9dSNir Soffer out_close: 2503c1bb86cdSEric Blake if (qemu_close(fd) != 0 && result == 0) { 2504c1bb86cdSEric Blake result = -errno; 2505c1bb86cdSEric Blake error_setg_errno(errp, -result, "Could not close the new file"); 2506c1bb86cdSEric Blake } 2507c1bb86cdSEric Blake out: 2508c1bb86cdSEric Blake return result; 2509c1bb86cdSEric Blake } 2510c1bb86cdSEric Blake 2511b92902dfSMaxim Levitsky static int coroutine_fn raw_co_create_opts(BlockDriver *drv, 2512b92902dfSMaxim Levitsky const char *filename, 2513b92902dfSMaxim Levitsky QemuOpts *opts, 2514927f11e1SKevin Wolf Error **errp) 2515927f11e1SKevin Wolf { 2516927f11e1SKevin Wolf BlockdevCreateOptions options; 2517927f11e1SKevin Wolf int64_t total_size = 0; 2518ffa244c8SKevin Wolf int64_t extent_size_hint = 0; 2519ffa244c8SKevin Wolf bool has_extent_size_hint = false; 2520927f11e1SKevin Wolf bool nocow = false; 2521927f11e1SKevin Wolf PreallocMode prealloc; 2522927f11e1SKevin Wolf char *buf = NULL; 2523927f11e1SKevin Wolf Error *local_err = NULL; 2524927f11e1SKevin Wolf 2525927f11e1SKevin Wolf /* Skip file: protocol prefix */ 2526927f11e1SKevin Wolf strstart(filename, "file:", &filename); 2527927f11e1SKevin Wolf 2528927f11e1SKevin Wolf /* Read out options */ 2529927f11e1SKevin Wolf total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), 2530927f11e1SKevin Wolf BDRV_SECTOR_SIZE); 2531ffa244c8SKevin Wolf if (qemu_opt_get(opts, BLOCK_OPT_EXTENT_SIZE_HINT)) { 2532ffa244c8SKevin Wolf has_extent_size_hint = true; 2533ffa244c8SKevin Wolf extent_size_hint = 2534ffa244c8SKevin Wolf qemu_opt_get_size_del(opts, BLOCK_OPT_EXTENT_SIZE_HINT, -1); 2535ffa244c8SKevin Wolf } 2536927f11e1SKevin Wolf nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); 2537927f11e1SKevin Wolf buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); 2538927f11e1SKevin Wolf prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, 2539927f11e1SKevin Wolf PREALLOC_MODE_OFF, &local_err); 2540927f11e1SKevin Wolf g_free(buf); 2541927f11e1SKevin Wolf if (local_err) { 2542927f11e1SKevin Wolf error_propagate(errp, local_err); 2543927f11e1SKevin Wolf return -EINVAL; 2544927f11e1SKevin Wolf } 2545927f11e1SKevin Wolf 2546927f11e1SKevin Wolf options = (BlockdevCreateOptions) { 2547927f11e1SKevin Wolf .driver = BLOCKDEV_DRIVER_FILE, 2548927f11e1SKevin Wolf .u.file = { 2549927f11e1SKevin Wolf .filename = (char *) filename, 2550927f11e1SKevin Wolf .size = total_size, 2551927f11e1SKevin Wolf .has_preallocation = true, 2552927f11e1SKevin Wolf .preallocation = prealloc, 2553927f11e1SKevin Wolf .has_nocow = true, 2554927f11e1SKevin Wolf .nocow = nocow, 2555ffa244c8SKevin Wolf .has_extent_size_hint = has_extent_size_hint, 2556ffa244c8SKevin Wolf .extent_size_hint = extent_size_hint, 2557927f11e1SKevin Wolf }, 2558927f11e1SKevin Wolf }; 2559927f11e1SKevin Wolf return raw_co_create(&options, errp); 2560927f11e1SKevin Wolf } 2561927f11e1SKevin Wolf 25629bffae14SDaniel Henrique Barboza static int coroutine_fn raw_co_delete_file(BlockDriverState *bs, 25639bffae14SDaniel Henrique Barboza Error **errp) 25649bffae14SDaniel Henrique Barboza { 25659bffae14SDaniel Henrique Barboza struct stat st; 25669bffae14SDaniel Henrique Barboza int ret; 25679bffae14SDaniel Henrique Barboza 25689bffae14SDaniel Henrique Barboza if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) { 25699bffae14SDaniel Henrique Barboza error_setg_errno(errp, ENOENT, "%s is not a regular file", 25709bffae14SDaniel Henrique Barboza bs->filename); 25719bffae14SDaniel Henrique Barboza return -ENOENT; 25729bffae14SDaniel Henrique Barboza } 25739bffae14SDaniel Henrique Barboza 25749bffae14SDaniel Henrique Barboza ret = unlink(bs->filename); 25759bffae14SDaniel Henrique Barboza if (ret < 0) { 25769bffae14SDaniel Henrique Barboza ret = -errno; 25779bffae14SDaniel Henrique Barboza error_setg_errno(errp, -ret, "Error when deleting file %s", 25789bffae14SDaniel Henrique Barboza bs->filename); 25799bffae14SDaniel Henrique Barboza } 25809bffae14SDaniel Henrique Barboza 25819bffae14SDaniel Henrique Barboza return ret; 25829bffae14SDaniel Henrique Barboza } 25839bffae14SDaniel Henrique Barboza 2584c1bb86cdSEric Blake /* 2585c1bb86cdSEric Blake * Find allocation range in @bs around offset @start. 2586c1bb86cdSEric Blake * May change underlying file descriptor's file offset. 2587c1bb86cdSEric Blake * If @start is not in a hole, store @start in @data, and the 2588c1bb86cdSEric Blake * beginning of the next hole in @hole, and return 0. 2589c1bb86cdSEric Blake * If @start is in a non-trailing hole, store @start in @hole and the 2590c1bb86cdSEric Blake * beginning of the next non-hole in @data, and return 0. 2591c1bb86cdSEric Blake * If @start is in a trailing hole or beyond EOF, return -ENXIO. 2592c1bb86cdSEric Blake * If we can't find out, return a negative errno other than -ENXIO. 2593c1bb86cdSEric Blake */ 2594c1bb86cdSEric Blake static int find_allocation(BlockDriverState *bs, off_t start, 2595c1bb86cdSEric Blake off_t *data, off_t *hole) 2596c1bb86cdSEric Blake { 2597c1bb86cdSEric Blake #if defined SEEK_HOLE && defined SEEK_DATA 2598c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 2599c1bb86cdSEric Blake off_t offs; 2600c1bb86cdSEric Blake 2601c1bb86cdSEric Blake /* 2602c1bb86cdSEric Blake * SEEK_DATA cases: 2603c1bb86cdSEric Blake * D1. offs == start: start is in data 2604c1bb86cdSEric Blake * D2. offs > start: start is in a hole, next data at offs 2605c1bb86cdSEric Blake * D3. offs < 0, errno = ENXIO: either start is in a trailing hole 2606c1bb86cdSEric Blake * or start is beyond EOF 2607c1bb86cdSEric Blake * If the latter happens, the file has been truncated behind 2608c1bb86cdSEric Blake * our back since we opened it. All bets are off then. 2609c1bb86cdSEric Blake * Treating like a trailing hole is simplest. 2610c1bb86cdSEric Blake * D4. offs < 0, errno != ENXIO: we learned nothing 2611c1bb86cdSEric Blake */ 2612c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_DATA); 2613c1bb86cdSEric Blake if (offs < 0) { 2614c1bb86cdSEric Blake return -errno; /* D3 or D4 */ 2615c1bb86cdSEric Blake } 2616a03083a0SJeff Cody 2617a03083a0SJeff Cody if (offs < start) { 2618a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2619a03083a0SJeff Cody * -EIO in this case, and we'll treat it like D4. */ 2620a03083a0SJeff Cody return -EIO; 2621a03083a0SJeff Cody } 2622c1bb86cdSEric Blake 2623c1bb86cdSEric Blake if (offs > start) { 2624c1bb86cdSEric Blake /* D2: in hole, next data at offs */ 2625c1bb86cdSEric Blake *hole = start; 2626c1bb86cdSEric Blake *data = offs; 2627c1bb86cdSEric Blake return 0; 2628c1bb86cdSEric Blake } 2629c1bb86cdSEric Blake 2630c1bb86cdSEric Blake /* D1: in data, end not yet known */ 2631c1bb86cdSEric Blake 2632c1bb86cdSEric Blake /* 2633c1bb86cdSEric Blake * SEEK_HOLE cases: 2634c1bb86cdSEric Blake * H1. offs == start: start is in a hole 2635c1bb86cdSEric Blake * If this happens here, a hole has been dug behind our back 2636c1bb86cdSEric Blake * since the previous lseek(). 2637c1bb86cdSEric Blake * H2. offs > start: either start is in data, next hole at offs, 2638c1bb86cdSEric Blake * or start is in trailing hole, EOF at offs 2639c1bb86cdSEric Blake * Linux treats trailing holes like any other hole: offs == 2640c1bb86cdSEric Blake * start. Solaris seeks to EOF instead: offs > start (blech). 2641c1bb86cdSEric Blake * If that happens here, a hole has been dug behind our back 2642c1bb86cdSEric Blake * since the previous lseek(). 2643c1bb86cdSEric Blake * H3. offs < 0, errno = ENXIO: start is beyond EOF 2644c1bb86cdSEric Blake * If this happens, the file has been truncated behind our 2645c1bb86cdSEric Blake * back since we opened it. Treat it like a trailing hole. 2646c1bb86cdSEric Blake * H4. offs < 0, errno != ENXIO: we learned nothing 2647c1bb86cdSEric Blake * Pretend we know nothing at all, i.e. "forget" about D1. 2648c1bb86cdSEric Blake */ 2649c1bb86cdSEric Blake offs = lseek(s->fd, start, SEEK_HOLE); 2650c1bb86cdSEric Blake if (offs < 0) { 2651c1bb86cdSEric Blake return -errno; /* D1 and (H3 or H4) */ 2652c1bb86cdSEric Blake } 2653a03083a0SJeff Cody 2654a03083a0SJeff Cody if (offs < start) { 2655a03083a0SJeff Cody /* This is not a valid return by lseek(). We are safe to just return 2656a03083a0SJeff Cody * -EIO in this case, and we'll treat it like H4. */ 2657a03083a0SJeff Cody return -EIO; 2658a03083a0SJeff Cody } 2659c1bb86cdSEric Blake 2660c1bb86cdSEric Blake if (offs > start) { 2661c1bb86cdSEric Blake /* 2662c1bb86cdSEric Blake * D1 and H2: either in data, next hole at offs, or it was in 2663c1bb86cdSEric Blake * data but is now in a trailing hole. In the latter case, 2664c1bb86cdSEric Blake * all bets are off. Treating it as if it there was data all 2665c1bb86cdSEric Blake * the way to EOF is safe, so simply do that. 2666c1bb86cdSEric Blake */ 2667c1bb86cdSEric Blake *data = start; 2668c1bb86cdSEric Blake *hole = offs; 2669c1bb86cdSEric Blake return 0; 2670c1bb86cdSEric Blake } 2671c1bb86cdSEric Blake 2672c1bb86cdSEric Blake /* D1 and H1 */ 2673c1bb86cdSEric Blake return -EBUSY; 2674c1bb86cdSEric Blake #else 2675c1bb86cdSEric Blake return -ENOTSUP; 2676c1bb86cdSEric Blake #endif 2677c1bb86cdSEric Blake } 2678c1bb86cdSEric Blake 2679c1bb86cdSEric Blake /* 2680a290f085SEric Blake * Returns the allocation status of the specified offset. 2681c1bb86cdSEric Blake * 2682a290f085SEric Blake * The block layer guarantees 'offset' and 'bytes' are within bounds. 2683c1bb86cdSEric Blake * 2684a290f085SEric Blake * 'pnum' is set to the number of bytes (including and immediately following 2685a290f085SEric Blake * the specified offset) that are known to be in the same 2686c1bb86cdSEric Blake * allocated/unallocated state. 2687c1bb86cdSEric Blake * 2688a290f085SEric Blake * 'bytes' is the max value 'pnum' should be set to. 2689c1bb86cdSEric Blake */ 2690a290f085SEric Blake static int coroutine_fn raw_co_block_status(BlockDriverState *bs, 2691a290f085SEric Blake bool want_zero, 2692a290f085SEric Blake int64_t offset, 2693a290f085SEric Blake int64_t bytes, int64_t *pnum, 2694a290f085SEric Blake int64_t *map, 2695c1bb86cdSEric Blake BlockDriverState **file) 2696c1bb86cdSEric Blake { 2697a290f085SEric Blake off_t data = 0, hole = 0; 2698c1bb86cdSEric Blake int ret; 2699c1bb86cdSEric Blake 27009c3db310SMax Reitz assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); 27019c3db310SMax Reitz 2702c1bb86cdSEric Blake ret = fd_open(bs); 2703c1bb86cdSEric Blake if (ret < 0) { 2704c1bb86cdSEric Blake return ret; 2705c1bb86cdSEric Blake } 2706c1bb86cdSEric Blake 2707a290f085SEric Blake if (!want_zero) { 2708a290f085SEric Blake *pnum = bytes; 2709a290f085SEric Blake *map = offset; 2710a290f085SEric Blake *file = bs; 2711a290f085SEric Blake return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; 2712c1bb86cdSEric Blake } 2713c1bb86cdSEric Blake 2714a290f085SEric Blake ret = find_allocation(bs, offset, &data, &hole); 2715c1bb86cdSEric Blake if (ret == -ENXIO) { 2716c1bb86cdSEric Blake /* Trailing hole */ 2717a290f085SEric Blake *pnum = bytes; 2718c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2719c1bb86cdSEric Blake } else if (ret < 0) { 2720c1bb86cdSEric Blake /* No info available, so pretend there are no holes */ 2721a290f085SEric Blake *pnum = bytes; 2722c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2723a290f085SEric Blake } else if (data == offset) { 2724a290f085SEric Blake /* On a data extent, compute bytes to the end of the extent, 2725c1bb86cdSEric Blake * possibly including a partial sector at EOF. */ 2726a290f085SEric Blake *pnum = MIN(bytes, hole - offset); 27279c3db310SMax Reitz 27289c3db310SMax Reitz /* 27299c3db310SMax Reitz * We are not allowed to return partial sectors, though, so 27309c3db310SMax Reitz * round up if necessary. 27319c3db310SMax Reitz */ 27329c3db310SMax Reitz if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { 27339c3db310SMax Reitz int64_t file_length = raw_getlength(bs); 27349c3db310SMax Reitz if (file_length > 0) { 27359c3db310SMax Reitz /* Ignore errors, this is just a safeguard */ 27369c3db310SMax Reitz assert(hole == file_length); 27379c3db310SMax Reitz } 27389c3db310SMax Reitz *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); 27399c3db310SMax Reitz } 27409c3db310SMax Reitz 2741c1bb86cdSEric Blake ret = BDRV_BLOCK_DATA; 2742c1bb86cdSEric Blake } else { 2743a290f085SEric Blake /* On a hole, compute bytes to the beginning of the next extent. */ 2744a290f085SEric Blake assert(hole == offset); 2745a290f085SEric Blake *pnum = MIN(bytes, data - offset); 2746c1bb86cdSEric Blake ret = BDRV_BLOCK_ZERO; 2747c1bb86cdSEric Blake } 2748a290f085SEric Blake *map = offset; 2749c1bb86cdSEric Blake *file = bs; 2750a290f085SEric Blake return ret | BDRV_BLOCK_OFFSET_VALID; 2751c1bb86cdSEric Blake } 2752c1bb86cdSEric Blake 275331be8a2aSStefan Hajnoczi #if defined(__linux__) 275431be8a2aSStefan Hajnoczi /* Verify that the file is not in the page cache */ 275531be8a2aSStefan Hajnoczi static void check_cache_dropped(BlockDriverState *bs, Error **errp) 275631be8a2aSStefan Hajnoczi { 275731be8a2aSStefan Hajnoczi const size_t window_size = 128 * 1024 * 1024; 275831be8a2aSStefan Hajnoczi BDRVRawState *s = bs->opaque; 275931be8a2aSStefan Hajnoczi void *window = NULL; 276031be8a2aSStefan Hajnoczi size_t length = 0; 276131be8a2aSStefan Hajnoczi unsigned char *vec; 276231be8a2aSStefan Hajnoczi size_t page_size; 276331be8a2aSStefan Hajnoczi off_t offset; 276431be8a2aSStefan Hajnoczi off_t end; 276531be8a2aSStefan Hajnoczi 276631be8a2aSStefan Hajnoczi /* mincore(2) page status information requires 1 byte per page */ 276731be8a2aSStefan Hajnoczi page_size = sysconf(_SC_PAGESIZE); 276831be8a2aSStefan Hajnoczi vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); 276931be8a2aSStefan Hajnoczi 277031be8a2aSStefan Hajnoczi end = raw_getlength(bs); 277131be8a2aSStefan Hajnoczi 277231be8a2aSStefan Hajnoczi for (offset = 0; offset < end; offset += window_size) { 277331be8a2aSStefan Hajnoczi void *new_window; 277431be8a2aSStefan Hajnoczi size_t new_length; 277531be8a2aSStefan Hajnoczi size_t vec_end; 277631be8a2aSStefan Hajnoczi size_t i; 277731be8a2aSStefan Hajnoczi int ret; 277831be8a2aSStefan Hajnoczi 277931be8a2aSStefan Hajnoczi /* Unmap previous window if size has changed */ 278031be8a2aSStefan Hajnoczi new_length = MIN(end - offset, window_size); 278131be8a2aSStefan Hajnoczi if (new_length != length) { 278231be8a2aSStefan Hajnoczi munmap(window, length); 278331be8a2aSStefan Hajnoczi window = NULL; 278431be8a2aSStefan Hajnoczi length = 0; 278531be8a2aSStefan Hajnoczi } 278631be8a2aSStefan Hajnoczi 278731be8a2aSStefan Hajnoczi new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, 278831be8a2aSStefan Hajnoczi s->fd, offset); 278931be8a2aSStefan Hajnoczi if (new_window == MAP_FAILED) { 279031be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mmap failed"); 279131be8a2aSStefan Hajnoczi break; 279231be8a2aSStefan Hajnoczi } 279331be8a2aSStefan Hajnoczi 279431be8a2aSStefan Hajnoczi window = new_window; 279531be8a2aSStefan Hajnoczi length = new_length; 279631be8a2aSStefan Hajnoczi 279731be8a2aSStefan Hajnoczi ret = mincore(window, length, vec); 279831be8a2aSStefan Hajnoczi if (ret < 0) { 279931be8a2aSStefan Hajnoczi error_setg_errno(errp, errno, "mincore failed"); 280031be8a2aSStefan Hajnoczi break; 280131be8a2aSStefan Hajnoczi } 280231be8a2aSStefan Hajnoczi 280331be8a2aSStefan Hajnoczi vec_end = DIV_ROUND_UP(length, page_size); 280431be8a2aSStefan Hajnoczi for (i = 0; i < vec_end; i++) { 280531be8a2aSStefan Hajnoczi if (vec[i] & 0x1) { 280631be8a2aSStefan Hajnoczi break; 280731be8a2aSStefan Hajnoczi } 280831be8a2aSStefan Hajnoczi } 280977ed971bSMarkus Armbruster if (i < vec_end) { 281077ed971bSMarkus Armbruster error_setg(errp, "page cache still in use!"); 281177ed971bSMarkus Armbruster break; 281277ed971bSMarkus Armbruster } 281331be8a2aSStefan Hajnoczi } 281431be8a2aSStefan Hajnoczi 281531be8a2aSStefan Hajnoczi if (window) { 281631be8a2aSStefan Hajnoczi munmap(window, length); 281731be8a2aSStefan Hajnoczi } 281831be8a2aSStefan Hajnoczi 281931be8a2aSStefan Hajnoczi g_free(vec); 282031be8a2aSStefan Hajnoczi } 282131be8a2aSStefan Hajnoczi #endif /* __linux__ */ 282231be8a2aSStefan Hajnoczi 2823dd577a26SStefan Hajnoczi static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, 2824dd577a26SStefan Hajnoczi Error **errp) 2825dd577a26SStefan Hajnoczi { 2826dd577a26SStefan Hajnoczi BDRVRawState *s = bs->opaque; 2827dd577a26SStefan Hajnoczi int ret; 2828dd577a26SStefan Hajnoczi 2829dd577a26SStefan Hajnoczi ret = fd_open(bs); 2830dd577a26SStefan Hajnoczi if (ret < 0) { 2831dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "The file descriptor is not open"); 2832dd577a26SStefan Hajnoczi return; 2833dd577a26SStefan Hajnoczi } 2834dd577a26SStefan Hajnoczi 2835f357fcd8SStefan Hajnoczi if (!s->drop_cache) { 2836f357fcd8SStefan Hajnoczi return; 2837f357fcd8SStefan Hajnoczi } 2838f357fcd8SStefan Hajnoczi 2839dd577a26SStefan Hajnoczi if (s->open_flags & O_DIRECT) { 2840dd577a26SStefan Hajnoczi return; /* No host kernel page cache */ 2841dd577a26SStefan Hajnoczi } 2842dd577a26SStefan Hajnoczi 2843dd577a26SStefan Hajnoczi #if defined(__linux__) 2844dd577a26SStefan Hajnoczi /* This sets the scene for the next syscall... */ 2845dd577a26SStefan Hajnoczi ret = bdrv_co_flush(bs); 2846dd577a26SStefan Hajnoczi if (ret < 0) { 2847dd577a26SStefan Hajnoczi error_setg_errno(errp, -ret, "flush failed"); 2848dd577a26SStefan Hajnoczi return; 2849dd577a26SStefan Hajnoczi } 2850dd577a26SStefan Hajnoczi 2851dd577a26SStefan Hajnoczi /* Linux does not invalidate pages that are dirty, locked, or mmapped by a 2852dd577a26SStefan Hajnoczi * process. These limitations are okay because we just fsynced the file, 2853dd577a26SStefan Hajnoczi * we don't use mmap, and the file should not be in use by other processes. 2854dd577a26SStefan Hajnoczi */ 2855dd577a26SStefan Hajnoczi ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); 2856dd577a26SStefan Hajnoczi if (ret != 0) { /* the return value is a positive errno */ 2857dd577a26SStefan Hajnoczi error_setg_errno(errp, ret, "fadvise failed"); 2858dd577a26SStefan Hajnoczi return; 2859dd577a26SStefan Hajnoczi } 286031be8a2aSStefan Hajnoczi 286131be8a2aSStefan Hajnoczi if (s->check_cache_dropped) { 286231be8a2aSStefan Hajnoczi check_cache_dropped(bs, errp); 286331be8a2aSStefan Hajnoczi } 2864dd577a26SStefan Hajnoczi #else /* __linux__ */ 2865dd577a26SStefan Hajnoczi /* Do nothing. Live migration to a remote host with cache.direct=off is 2866dd577a26SStefan Hajnoczi * unsupported on other host operating systems. Cache consistency issues 2867dd577a26SStefan Hajnoczi * may occur but no error is reported here, partly because that's the 2868dd577a26SStefan Hajnoczi * historical behavior and partly because it's hard to differentiate valid 2869dd577a26SStefan Hajnoczi * configurations that should not cause errors. 2870dd577a26SStefan Hajnoczi */ 2871dd577a26SStefan Hajnoczi #endif /* !__linux__ */ 2872dd577a26SStefan Hajnoczi } 2873dd577a26SStefan Hajnoczi 28741c450366SAnton Nefedov static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret) 28751c450366SAnton Nefedov { 28761c450366SAnton Nefedov if (ret) { 28771c450366SAnton Nefedov s->stats.discard_nb_failed++; 28781c450366SAnton Nefedov } else { 28791c450366SAnton Nefedov s->stats.discard_nb_ok++; 28801c450366SAnton Nefedov s->stats.discard_bytes_ok += nbytes; 28811c450366SAnton Nefedov } 28821c450366SAnton Nefedov } 28831c450366SAnton Nefedov 288433d70fb6SKevin Wolf static coroutine_fn int 288546ee0f46SKevin Wolf raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev) 2886c1bb86cdSEric Blake { 2887c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 288846ee0f46SKevin Wolf RawPosixAIOData acb; 28891c450366SAnton Nefedov int ret; 2890c1bb86cdSEric Blake 289146ee0f46SKevin Wolf acb = (RawPosixAIOData) { 289246ee0f46SKevin Wolf .bs = bs, 289346ee0f46SKevin Wolf .aio_fildes = s->fd, 289446ee0f46SKevin Wolf .aio_type = QEMU_AIO_DISCARD, 289546ee0f46SKevin Wolf .aio_offset = offset, 289646ee0f46SKevin Wolf .aio_nbytes = bytes, 289746ee0f46SKevin Wolf }; 289846ee0f46SKevin Wolf 289946ee0f46SKevin Wolf if (blkdev) { 290046ee0f46SKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 290146ee0f46SKevin Wolf } 290246ee0f46SKevin Wolf 29031c450366SAnton Nefedov ret = raw_thread_pool_submit(bs, handle_aiocb_discard, &acb); 29041c450366SAnton Nefedov raw_account_discard(s, bytes, ret); 29051c450366SAnton Nefedov return ret; 290646ee0f46SKevin Wolf } 290746ee0f46SKevin Wolf 290846ee0f46SKevin Wolf static coroutine_fn int 290946ee0f46SKevin Wolf raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 291046ee0f46SKevin Wolf { 291146ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, false); 2912c1bb86cdSEric Blake } 2913c1bb86cdSEric Blake 29147154d8aeSKevin Wolf static int coroutine_fn 29157154d8aeSKevin Wolf raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, 29167154d8aeSKevin Wolf BdrvRequestFlags flags, bool blkdev) 29177154d8aeSKevin Wolf { 29187154d8aeSKevin Wolf BDRVRawState *s = bs->opaque; 29197154d8aeSKevin Wolf RawPosixAIOData acb; 29207154d8aeSKevin Wolf ThreadPoolFunc *handler; 29217154d8aeSKevin Wolf 2922292d06b9SMax Reitz #ifdef CONFIG_FALLOCATE 2923292d06b9SMax Reitz if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { 2924292d06b9SMax Reitz BdrvTrackedRequest *req; 2925292d06b9SMax Reitz uint64_t end; 2926292d06b9SMax Reitz 2927292d06b9SMax Reitz /* 2928292d06b9SMax Reitz * This is a workaround for a bug in the Linux XFS driver, 2929292d06b9SMax Reitz * where writes submitted through the AIO interface will be 2930292d06b9SMax Reitz * discarded if they happen beyond a concurrently running 2931292d06b9SMax Reitz * fallocate() that increases the file length (i.e., both the 2932292d06b9SMax Reitz * write and the fallocate() happen beyond the EOF). 2933292d06b9SMax Reitz * 2934292d06b9SMax Reitz * To work around it, we extend the tracked request for this 2935292d06b9SMax Reitz * zero write until INT64_MAX (effectively infinity), and mark 2936292d06b9SMax Reitz * it as serializing. 2937292d06b9SMax Reitz * 2938292d06b9SMax Reitz * We have to enable this workaround for all filesystems and 2939292d06b9SMax Reitz * AIO modes (not just XFS with aio=native), because for 2940292d06b9SMax Reitz * remote filesystems we do not know the host configuration. 2941292d06b9SMax Reitz */ 2942292d06b9SMax Reitz 2943292d06b9SMax Reitz req = bdrv_co_get_self_request(bs); 2944292d06b9SMax Reitz assert(req); 2945292d06b9SMax Reitz assert(req->type == BDRV_TRACKED_WRITE); 2946292d06b9SMax Reitz assert(req->offset <= offset); 2947292d06b9SMax Reitz assert(req->offset + req->bytes >= offset + bytes); 2948292d06b9SMax Reitz 2949292d06b9SMax Reitz end = INT64_MAX & -(uint64_t)bs->bl.request_alignment; 2950292d06b9SMax Reitz req->bytes = end - req->offset; 2951292d06b9SMax Reitz req->overlap_bytes = req->bytes; 2952292d06b9SMax Reitz 2953292d06b9SMax Reitz bdrv_mark_request_serialising(req, bs->bl.request_alignment); 2954292d06b9SMax Reitz } 2955292d06b9SMax Reitz #endif 2956292d06b9SMax Reitz 29577154d8aeSKevin Wolf acb = (RawPosixAIOData) { 29587154d8aeSKevin Wolf .bs = bs, 29597154d8aeSKevin Wolf .aio_fildes = s->fd, 29607154d8aeSKevin Wolf .aio_type = QEMU_AIO_WRITE_ZEROES, 29617154d8aeSKevin Wolf .aio_offset = offset, 29627154d8aeSKevin Wolf .aio_nbytes = bytes, 29637154d8aeSKevin Wolf }; 29647154d8aeSKevin Wolf 29657154d8aeSKevin Wolf if (blkdev) { 29667154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_BLKDEV; 29677154d8aeSKevin Wolf } 2968738301e1SKevin Wolf if (flags & BDRV_REQ_NO_FALLBACK) { 2969738301e1SKevin Wolf acb.aio_type |= QEMU_AIO_NO_FALLBACK; 2970738301e1SKevin Wolf } 29717154d8aeSKevin Wolf 29727154d8aeSKevin Wolf if (flags & BDRV_REQ_MAY_UNMAP) { 29737154d8aeSKevin Wolf acb.aio_type |= QEMU_AIO_DISCARD; 29747154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes_unmap; 29757154d8aeSKevin Wolf } else { 29767154d8aeSKevin Wolf handler = handle_aiocb_write_zeroes; 29777154d8aeSKevin Wolf } 29787154d8aeSKevin Wolf 29797154d8aeSKevin Wolf return raw_thread_pool_submit(bs, handler, &acb); 29807154d8aeSKevin Wolf } 29817154d8aeSKevin Wolf 2982c1bb86cdSEric Blake static int coroutine_fn raw_co_pwrite_zeroes( 2983c1bb86cdSEric Blake BlockDriverState *bs, int64_t offset, 2984f5a5ca79SManos Pitsidianakis int bytes, BdrvRequestFlags flags) 2985c1bb86cdSEric Blake { 29867154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); 2987c1bb86cdSEric Blake } 2988c1bb86cdSEric Blake 2989c1bb86cdSEric Blake static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) 2990c1bb86cdSEric Blake { 2991c1bb86cdSEric Blake return 0; 2992c1bb86cdSEric Blake } 2993c1bb86cdSEric Blake 2994d9245599SAnton Nefedov static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs) 2995d9245599SAnton Nefedov { 2996d9245599SAnton Nefedov BDRVRawState *s = bs->opaque; 2997d9245599SAnton Nefedov return (BlockStatsSpecificFile) { 2998d9245599SAnton Nefedov .discard_nb_ok = s->stats.discard_nb_ok, 2999d9245599SAnton Nefedov .discard_nb_failed = s->stats.discard_nb_failed, 3000d9245599SAnton Nefedov .discard_bytes_ok = s->stats.discard_bytes_ok, 3001d9245599SAnton Nefedov }; 3002d9245599SAnton Nefedov } 3003d9245599SAnton Nefedov 3004d9245599SAnton Nefedov static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs) 3005d9245599SAnton Nefedov { 3006d9245599SAnton Nefedov BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 3007d9245599SAnton Nefedov 3008d9245599SAnton Nefedov stats->driver = BLOCKDEV_DRIVER_FILE; 3009d9245599SAnton Nefedov stats->u.file = get_blockstats_specific_file(bs); 3010d9245599SAnton Nefedov 3011d9245599SAnton Nefedov return stats; 3012d9245599SAnton Nefedov } 3013d9245599SAnton Nefedov 3014d9245599SAnton Nefedov static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs) 3015d9245599SAnton Nefedov { 3016d9245599SAnton Nefedov BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); 3017d9245599SAnton Nefedov 3018d9245599SAnton Nefedov stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE; 3019d9245599SAnton Nefedov stats->u.host_device = get_blockstats_specific_file(bs); 3020d9245599SAnton Nefedov 3021d9245599SAnton Nefedov return stats; 3022d9245599SAnton Nefedov } 3023d9245599SAnton Nefedov 3024c1bb86cdSEric Blake static QemuOptsList raw_create_opts = { 3025c1bb86cdSEric Blake .name = "raw-create-opts", 3026c1bb86cdSEric Blake .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), 3027c1bb86cdSEric Blake .desc = { 3028c1bb86cdSEric Blake { 3029c1bb86cdSEric Blake .name = BLOCK_OPT_SIZE, 3030c1bb86cdSEric Blake .type = QEMU_OPT_SIZE, 3031c1bb86cdSEric Blake .help = "Virtual disk size" 3032c1bb86cdSEric Blake }, 3033c1bb86cdSEric Blake { 3034c1bb86cdSEric Blake .name = BLOCK_OPT_NOCOW, 3035c1bb86cdSEric Blake .type = QEMU_OPT_BOOL, 3036c1bb86cdSEric Blake .help = "Turn off copy-on-write (valid only on btrfs)" 3037c1bb86cdSEric Blake }, 3038c1bb86cdSEric Blake { 3039c1bb86cdSEric Blake .name = BLOCK_OPT_PREALLOC, 3040c1bb86cdSEric Blake .type = QEMU_OPT_STRING, 3041abea0053SStefano Garzarella .help = "Preallocation mode (allowed values: off" 3042abea0053SStefano Garzarella #ifdef CONFIG_POSIX_FALLOCATE 3043abea0053SStefano Garzarella ", falloc" 3044abea0053SStefano Garzarella #endif 3045abea0053SStefano Garzarella ", full)" 3046c1bb86cdSEric Blake }, 3047ffa244c8SKevin Wolf { 3048ffa244c8SKevin Wolf .name = BLOCK_OPT_EXTENT_SIZE_HINT, 3049ffa244c8SKevin Wolf .type = QEMU_OPT_SIZE, 3050ffa244c8SKevin Wolf .help = "Extent size hint for the image file, 0 to disable" 3051ffa244c8SKevin Wolf }, 3052c1bb86cdSEric Blake { /* end of list */ } 3053c1bb86cdSEric Blake } 3054c1bb86cdSEric Blake }; 3055c1bb86cdSEric Blake 3056244a5668SFam Zheng static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, 3057244a5668SFam Zheng Error **errp) 3058244a5668SFam Zheng { 30596ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 30606ceabe6fSKevin Wolf BDRVRawReopenState *rs = NULL; 30616ceabe6fSKevin Wolf int open_flags; 30626ceabe6fSKevin Wolf int ret; 30636ceabe6fSKevin Wolf 30646ceabe6fSKevin Wolf if (s->perm_change_fd) { 30656ceabe6fSKevin Wolf /* 30666ceabe6fSKevin Wolf * In the context of reopen, this function may be called several times 30676ceabe6fSKevin Wolf * (directly and recursively while change permissions of the parent). 30686ceabe6fSKevin Wolf * This is even true for children that don't inherit from the original 30696ceabe6fSKevin Wolf * reopen node, so s->reopen_state is not set. 30706ceabe6fSKevin Wolf * 30716ceabe6fSKevin Wolf * Ignore all but the first call. 30726ceabe6fSKevin Wolf */ 30736ceabe6fSKevin Wolf return 0; 30746ceabe6fSKevin Wolf } 30756ceabe6fSKevin Wolf 30766ceabe6fSKevin Wolf if (s->reopen_state) { 30776ceabe6fSKevin Wolf /* We already have a new file descriptor to set permissions for */ 30786ceabe6fSKevin Wolf assert(s->reopen_state->perm == perm); 30796ceabe6fSKevin Wolf assert(s->reopen_state->shared_perm == shared); 30806ceabe6fSKevin Wolf rs = s->reopen_state->opaque; 30816ceabe6fSKevin Wolf s->perm_change_fd = rs->fd; 3082094e3639SMax Reitz s->perm_change_flags = rs->open_flags; 30836ceabe6fSKevin Wolf } else { 30846ceabe6fSKevin Wolf /* We may need a new fd if auto-read-only switches the mode */ 308523dece19SKevin Wolf ret = raw_reconfigure_getfd(bs, bs->open_flags, &open_flags, perm, 30866ceabe6fSKevin Wolf false, errp); 30876ceabe6fSKevin Wolf if (ret < 0) { 30886ceabe6fSKevin Wolf return ret; 30896ceabe6fSKevin Wolf } else if (ret != s->fd) { 30906ceabe6fSKevin Wolf s->perm_change_fd = ret; 3091094e3639SMax Reitz s->perm_change_flags = open_flags; 30926ceabe6fSKevin Wolf } 30936ceabe6fSKevin Wolf } 30946ceabe6fSKevin Wolf 30956ceabe6fSKevin Wolf /* Prepare permissions on old fd to avoid conflicts between old and new, 30966ceabe6fSKevin Wolf * but keep everything locked that new will need. */ 30976ceabe6fSKevin Wolf ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); 30986ceabe6fSKevin Wolf if (ret < 0) { 30996ceabe6fSKevin Wolf goto fail; 31006ceabe6fSKevin Wolf } 31016ceabe6fSKevin Wolf 31026ceabe6fSKevin Wolf /* Copy locks to the new fd */ 31036ceabe6fSKevin Wolf if (s->perm_change_fd) { 31046ceabe6fSKevin Wolf ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, 31056ceabe6fSKevin Wolf false, errp); 31066ceabe6fSKevin Wolf if (ret < 0) { 31076ceabe6fSKevin Wolf raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 31086ceabe6fSKevin Wolf goto fail; 31096ceabe6fSKevin Wolf } 31106ceabe6fSKevin Wolf } 31116ceabe6fSKevin Wolf return 0; 31126ceabe6fSKevin Wolf 31136ceabe6fSKevin Wolf fail: 31146ceabe6fSKevin Wolf if (s->perm_change_fd && !s->reopen_state) { 31156ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 31166ceabe6fSKevin Wolf } 31176ceabe6fSKevin Wolf s->perm_change_fd = 0; 31186ceabe6fSKevin Wolf return ret; 3119244a5668SFam Zheng } 3120244a5668SFam Zheng 3121244a5668SFam Zheng static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) 3122244a5668SFam Zheng { 3123244a5668SFam Zheng BDRVRawState *s = bs->opaque; 31246ceabe6fSKevin Wolf 31256ceabe6fSKevin Wolf /* For reopen, we have already switched to the new fd (.bdrv_set_perm is 31266ceabe6fSKevin Wolf * called after .bdrv_reopen_commit) */ 31276ceabe6fSKevin Wolf if (s->perm_change_fd && s->fd != s->perm_change_fd) { 31286ceabe6fSKevin Wolf qemu_close(s->fd); 31296ceabe6fSKevin Wolf s->fd = s->perm_change_fd; 3130094e3639SMax Reitz s->open_flags = s->perm_change_flags; 31316ceabe6fSKevin Wolf } 31326ceabe6fSKevin Wolf s->perm_change_fd = 0; 31336ceabe6fSKevin Wolf 3134244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); 3135244a5668SFam Zheng s->perm = perm; 3136244a5668SFam Zheng s->shared_perm = shared; 3137244a5668SFam Zheng } 3138244a5668SFam Zheng 3139244a5668SFam Zheng static void raw_abort_perm_update(BlockDriverState *bs) 3140244a5668SFam Zheng { 31416ceabe6fSKevin Wolf BDRVRawState *s = bs->opaque; 31426ceabe6fSKevin Wolf 31436ceabe6fSKevin Wolf /* For reopen, .bdrv_reopen_abort is called afterwards and will close 31446ceabe6fSKevin Wolf * the file descriptor. */ 31456ceabe6fSKevin Wolf if (s->perm_change_fd && !s->reopen_state) { 31466ceabe6fSKevin Wolf qemu_close(s->perm_change_fd); 31476ceabe6fSKevin Wolf } 31486ceabe6fSKevin Wolf s->perm_change_fd = 0; 31496ceabe6fSKevin Wolf 3150244a5668SFam Zheng raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); 3151244a5668SFam Zheng } 3152244a5668SFam Zheng 315367b51fb9SVladimir Sementsov-Ogievskiy static int coroutine_fn raw_co_copy_range_from( 315467b51fb9SVladimir Sementsov-Ogievskiy BlockDriverState *bs, BdrvChild *src, uint64_t src_offset, 315567b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, uint64_t dst_offset, uint64_t bytes, 315667b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) 31571efad060SFam Zheng { 315867b51fb9SVladimir Sementsov-Ogievskiy return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, 315967b51fb9SVladimir Sementsov-Ogievskiy read_flags, write_flags); 31601efad060SFam Zheng } 31611efad060SFam Zheng 31621efad060SFam Zheng static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, 316367b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *src, 316467b51fb9SVladimir Sementsov-Ogievskiy uint64_t src_offset, 316567b51fb9SVladimir Sementsov-Ogievskiy BdrvChild *dst, 316667b51fb9SVladimir Sementsov-Ogievskiy uint64_t dst_offset, 316767b51fb9SVladimir Sementsov-Ogievskiy uint64_t bytes, 316867b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags read_flags, 316967b51fb9SVladimir Sementsov-Ogievskiy BdrvRequestFlags write_flags) 31701efad060SFam Zheng { 317158a209c4SKevin Wolf RawPosixAIOData acb; 31721efad060SFam Zheng BDRVRawState *s = bs->opaque; 31731efad060SFam Zheng BDRVRawState *src_s; 31741efad060SFam Zheng 31751efad060SFam Zheng assert(dst->bs == bs); 31761efad060SFam Zheng if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { 31771efad060SFam Zheng return -ENOTSUP; 31781efad060SFam Zheng } 31791efad060SFam Zheng 31801efad060SFam Zheng src_s = src->bs->opaque; 31819f850f67SFam Zheng if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { 31821efad060SFam Zheng return -EIO; 31831efad060SFam Zheng } 318458a209c4SKevin Wolf 318558a209c4SKevin Wolf acb = (RawPosixAIOData) { 318658a209c4SKevin Wolf .bs = bs, 318758a209c4SKevin Wolf .aio_type = QEMU_AIO_COPY_RANGE, 318858a209c4SKevin Wolf .aio_fildes = src_s->fd, 318958a209c4SKevin Wolf .aio_offset = src_offset, 319058a209c4SKevin Wolf .aio_nbytes = bytes, 319158a209c4SKevin Wolf .copy_range = { 319258a209c4SKevin Wolf .aio_fd2 = s->fd, 319358a209c4SKevin Wolf .aio_offset2 = dst_offset, 319458a209c4SKevin Wolf }, 319558a209c4SKevin Wolf }; 319658a209c4SKevin Wolf 319758a209c4SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); 31981efad060SFam Zheng } 31991efad060SFam Zheng 3200c1bb86cdSEric Blake BlockDriver bdrv_file = { 3201c1bb86cdSEric Blake .format_name = "file", 3202c1bb86cdSEric Blake .protocol_name = "file", 3203c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3204c1bb86cdSEric Blake .bdrv_needs_filename = true, 3205c1bb86cdSEric Blake .bdrv_probe = NULL, /* no probe for protocols */ 3206c1bb86cdSEric Blake .bdrv_parse_filename = raw_parse_filename, 3207c1bb86cdSEric Blake .bdrv_file_open = raw_open, 3208c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3209c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3210c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 3211c1bb86cdSEric Blake .bdrv_close = raw_close, 3212927f11e1SKevin Wolf .bdrv_co_create = raw_co_create, 3213efc75e2aSStefan Hajnoczi .bdrv_co_create_opts = raw_co_create_opts, 3214c1bb86cdSEric Blake .bdrv_has_zero_init = bdrv_has_zero_init_1, 3215a290f085SEric Blake .bdrv_co_block_status = raw_co_block_status, 3216dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3217c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, 32189bffae14SDaniel Henrique Barboza .bdrv_co_delete_file = raw_co_delete_file, 3219c1bb86cdSEric Blake 3220c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3221c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 322233d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 322333d70fb6SKevin Wolf .bdrv_co_pdiscard = raw_co_pdiscard, 32241efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 32251efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 3226c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3227c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3228c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3229ed6e2161SNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3230c1bb86cdSEric Blake 3231061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3232c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3233c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 3234c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3235c1bb86cdSEric Blake = raw_get_allocated_file_size, 3236d9245599SAnton Nefedov .bdrv_get_specific_stats = raw_get_specific_stats, 3237244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 3238244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 3239244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 3240c1bb86cdSEric Blake .create_opts = &raw_create_opts, 32418a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3242c1bb86cdSEric Blake }; 3243c1bb86cdSEric Blake 3244c1bb86cdSEric Blake /***********************************************/ 3245c1bb86cdSEric Blake /* host device */ 3246c1bb86cdSEric Blake 3247c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3248c1bb86cdSEric Blake static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3249c1bb86cdSEric Blake CFIndex maxPathSize, int flags); 3250c1bb86cdSEric Blake static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) 3251c1bb86cdSEric Blake { 3252c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 3253c1bb86cdSEric Blake mach_port_t masterPort; 3254c1bb86cdSEric Blake CFMutableDictionaryRef classesToMatch; 3255c1bb86cdSEric Blake const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; 3256c1bb86cdSEric Blake char *mediaType = NULL; 3257c1bb86cdSEric Blake 3258c1bb86cdSEric Blake kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); 3259c1bb86cdSEric Blake if ( KERN_SUCCESS != kernResult ) { 3260c1bb86cdSEric Blake printf( "IOMasterPort returned %d\n", kernResult ); 3261c1bb86cdSEric Blake } 3262c1bb86cdSEric Blake 3263c1bb86cdSEric Blake int index; 3264c1bb86cdSEric Blake for (index = 0; index < ARRAY_SIZE(matching_array); index++) { 3265c1bb86cdSEric Blake classesToMatch = IOServiceMatching(matching_array[index]); 3266c1bb86cdSEric Blake if (classesToMatch == NULL) { 3267c1bb86cdSEric Blake error_report("IOServiceMatching returned NULL for %s", 3268c1bb86cdSEric Blake matching_array[index]); 3269c1bb86cdSEric Blake continue; 3270c1bb86cdSEric Blake } 3271c1bb86cdSEric Blake CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), 3272c1bb86cdSEric Blake kCFBooleanTrue); 3273c1bb86cdSEric Blake kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, 3274c1bb86cdSEric Blake mediaIterator); 3275c1bb86cdSEric Blake if (kernResult != KERN_SUCCESS) { 3276c1bb86cdSEric Blake error_report("Note: IOServiceGetMatchingServices returned %d", 3277c1bb86cdSEric Blake kernResult); 3278c1bb86cdSEric Blake continue; 3279c1bb86cdSEric Blake } 3280c1bb86cdSEric Blake 3281c1bb86cdSEric Blake /* If a match was found, leave the loop */ 3282c1bb86cdSEric Blake if (*mediaIterator != 0) { 32834f7d28d7SLaurent Vivier trace_file_FindEjectableOpticalMedia(matching_array[index]); 3284c1bb86cdSEric Blake mediaType = g_strdup(matching_array[index]); 3285c1bb86cdSEric Blake break; 3286c1bb86cdSEric Blake } 3287c1bb86cdSEric Blake } 3288c1bb86cdSEric Blake return mediaType; 3289c1bb86cdSEric Blake } 3290c1bb86cdSEric Blake 3291c1bb86cdSEric Blake kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, 3292c1bb86cdSEric Blake CFIndex maxPathSize, int flags) 3293c1bb86cdSEric Blake { 3294c1bb86cdSEric Blake io_object_t nextMedia; 3295c1bb86cdSEric Blake kern_return_t kernResult = KERN_FAILURE; 3296c1bb86cdSEric Blake *bsdPath = '\0'; 3297c1bb86cdSEric Blake nextMedia = IOIteratorNext( mediaIterator ); 3298c1bb86cdSEric Blake if ( nextMedia ) 3299c1bb86cdSEric Blake { 3300c1bb86cdSEric Blake CFTypeRef bsdPathAsCFString; 3301c1bb86cdSEric Blake bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); 3302c1bb86cdSEric Blake if ( bsdPathAsCFString ) { 3303c1bb86cdSEric Blake size_t devPathLength; 3304c1bb86cdSEric Blake strcpy( bsdPath, _PATH_DEV ); 3305c1bb86cdSEric Blake if (flags & BDRV_O_NOCACHE) { 3306c1bb86cdSEric Blake strcat(bsdPath, "r"); 3307c1bb86cdSEric Blake } 3308c1bb86cdSEric Blake devPathLength = strlen( bsdPath ); 3309c1bb86cdSEric Blake if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { 3310c1bb86cdSEric Blake kernResult = KERN_SUCCESS; 3311c1bb86cdSEric Blake } 3312c1bb86cdSEric Blake CFRelease( bsdPathAsCFString ); 3313c1bb86cdSEric Blake } 3314c1bb86cdSEric Blake IOObjectRelease( nextMedia ); 3315c1bb86cdSEric Blake } 3316c1bb86cdSEric Blake 3317c1bb86cdSEric Blake return kernResult; 3318c1bb86cdSEric Blake } 3319c1bb86cdSEric Blake 3320c1bb86cdSEric Blake /* Sets up a real cdrom for use in QEMU */ 3321c1bb86cdSEric Blake static bool setup_cdrom(char *bsd_path, Error **errp) 3322c1bb86cdSEric Blake { 3323c1bb86cdSEric Blake int index, num_of_test_partitions = 2, fd; 3324c1bb86cdSEric Blake char test_partition[MAXPATHLEN]; 3325c1bb86cdSEric Blake bool partition_found = false; 3326c1bb86cdSEric Blake 3327c1bb86cdSEric Blake /* look for a working partition */ 3328c1bb86cdSEric Blake for (index = 0; index < num_of_test_partitions; index++) { 3329c1bb86cdSEric Blake snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, 3330c1bb86cdSEric Blake index); 3331c1bb86cdSEric Blake fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE); 3332c1bb86cdSEric Blake if (fd >= 0) { 3333c1bb86cdSEric Blake partition_found = true; 3334c1bb86cdSEric Blake qemu_close(fd); 3335c1bb86cdSEric Blake break; 3336c1bb86cdSEric Blake } 3337c1bb86cdSEric Blake } 3338c1bb86cdSEric Blake 3339c1bb86cdSEric Blake /* if a working partition on the device was not found */ 3340c1bb86cdSEric Blake if (partition_found == false) { 3341c1bb86cdSEric Blake error_setg(errp, "Failed to find a working partition on disc"); 3342c1bb86cdSEric Blake } else { 33434f7d28d7SLaurent Vivier trace_file_setup_cdrom(test_partition); 3344c1bb86cdSEric Blake pstrcpy(bsd_path, MAXPATHLEN, test_partition); 3345c1bb86cdSEric Blake } 3346c1bb86cdSEric Blake return partition_found; 3347c1bb86cdSEric Blake } 3348c1bb86cdSEric Blake 3349c1bb86cdSEric Blake /* Prints directions on mounting and unmounting a device */ 3350c1bb86cdSEric Blake static void print_unmounting_directions(const char *file_name) 3351c1bb86cdSEric Blake { 3352c1bb86cdSEric Blake error_report("If device %s is mounted on the desktop, unmount" 3353c1bb86cdSEric Blake " it first before using it in QEMU", file_name); 3354c1bb86cdSEric Blake error_report("Command to unmount device: diskutil unmountDisk %s", 3355c1bb86cdSEric Blake file_name); 3356c1bb86cdSEric Blake error_report("Command to mount device: diskutil mountDisk %s", file_name); 3357c1bb86cdSEric Blake } 3358c1bb86cdSEric Blake 3359c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3360c1bb86cdSEric Blake 3361c1bb86cdSEric Blake static int hdev_probe_device(const char *filename) 3362c1bb86cdSEric Blake { 3363c1bb86cdSEric Blake struct stat st; 3364c1bb86cdSEric Blake 3365c1bb86cdSEric Blake /* allow a dedicated CD-ROM driver to match with a higher priority */ 3366c1bb86cdSEric Blake if (strstart(filename, "/dev/cdrom", NULL)) 3367c1bb86cdSEric Blake return 50; 3368c1bb86cdSEric Blake 3369c1bb86cdSEric Blake if (stat(filename, &st) >= 0 && 3370c1bb86cdSEric Blake (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { 3371c1bb86cdSEric Blake return 100; 3372c1bb86cdSEric Blake } 3373c1bb86cdSEric Blake 3374c1bb86cdSEric Blake return 0; 3375c1bb86cdSEric Blake } 3376c1bb86cdSEric Blake 3377c1bb86cdSEric Blake static void hdev_parse_filename(const char *filename, QDict *options, 3378c1bb86cdSEric Blake Error **errp) 3379c1bb86cdSEric Blake { 338003c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_device:", options); 3381c1bb86cdSEric Blake } 3382c1bb86cdSEric Blake 3383c1bb86cdSEric Blake static bool hdev_is_sg(BlockDriverState *bs) 3384c1bb86cdSEric Blake { 3385c1bb86cdSEric Blake 3386c1bb86cdSEric Blake #if defined(__linux__) 3387c1bb86cdSEric Blake 3388c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3389c1bb86cdSEric Blake struct stat st; 3390c1bb86cdSEric Blake struct sg_scsi_id scsiid; 3391c1bb86cdSEric Blake int sg_version; 3392c1bb86cdSEric Blake int ret; 3393c1bb86cdSEric Blake 3394c1bb86cdSEric Blake if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { 3395c1bb86cdSEric Blake return false; 3396c1bb86cdSEric Blake } 3397c1bb86cdSEric Blake 3398c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); 3399c1bb86cdSEric Blake if (ret < 0) { 3400c1bb86cdSEric Blake return false; 3401c1bb86cdSEric Blake } 3402c1bb86cdSEric Blake 3403c1bb86cdSEric Blake ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); 3404c1bb86cdSEric Blake if (ret >= 0) { 34054f7d28d7SLaurent Vivier trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); 3406c1bb86cdSEric Blake return true; 3407c1bb86cdSEric Blake } 3408c1bb86cdSEric Blake 3409c1bb86cdSEric Blake #endif 3410c1bb86cdSEric Blake 3411c1bb86cdSEric Blake return false; 3412c1bb86cdSEric Blake } 3413c1bb86cdSEric Blake 3414c1bb86cdSEric Blake static int hdev_open(BlockDriverState *bs, QDict *options, int flags, 3415c1bb86cdSEric Blake Error **errp) 3416c1bb86cdSEric Blake { 3417c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3418c1bb86cdSEric Blake int ret; 3419c1bb86cdSEric Blake 3420c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3421129c7d1cSMarkus Armbruster /* 3422129c7d1cSMarkus Armbruster * Caution: while qdict_get_str() is fine, getting non-string types 3423129c7d1cSMarkus Armbruster * would require more care. When @options come from -blockdev or 3424129c7d1cSMarkus Armbruster * blockdev_add, its members are typed according to the QAPI 3425129c7d1cSMarkus Armbruster * schema, but when they come from -drive, they're all QString. 3426129c7d1cSMarkus Armbruster */ 3427c1bb86cdSEric Blake const char *filename = qdict_get_str(options, "filename"); 3428c1bb86cdSEric Blake char bsd_path[MAXPATHLEN] = ""; 3429c1bb86cdSEric Blake bool error_occurred = false; 3430c1bb86cdSEric Blake 3431c1bb86cdSEric Blake /* If using a real cdrom */ 3432c1bb86cdSEric Blake if (strcmp(filename, "/dev/cdrom") == 0) { 3433c1bb86cdSEric Blake char *mediaType = NULL; 3434c1bb86cdSEric Blake kern_return_t ret_val; 3435c1bb86cdSEric Blake io_iterator_t mediaIterator = 0; 3436c1bb86cdSEric Blake 3437c1bb86cdSEric Blake mediaType = FindEjectableOpticalMedia(&mediaIterator); 3438c1bb86cdSEric Blake if (mediaType == NULL) { 3439c1bb86cdSEric Blake error_setg(errp, "Please make sure your CD/DVD is in the optical" 3440c1bb86cdSEric Blake " drive"); 3441c1bb86cdSEric Blake error_occurred = true; 3442c1bb86cdSEric Blake goto hdev_open_Mac_error; 3443c1bb86cdSEric Blake } 3444c1bb86cdSEric Blake 3445c1bb86cdSEric Blake ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); 3446c1bb86cdSEric Blake if (ret_val != KERN_SUCCESS) { 3447c1bb86cdSEric Blake error_setg(errp, "Could not get BSD path for optical drive"); 3448c1bb86cdSEric Blake error_occurred = true; 3449c1bb86cdSEric Blake goto hdev_open_Mac_error; 3450c1bb86cdSEric Blake } 3451c1bb86cdSEric Blake 3452c1bb86cdSEric Blake /* If a real optical drive was not found */ 3453c1bb86cdSEric Blake if (bsd_path[0] == '\0') { 3454c1bb86cdSEric Blake error_setg(errp, "Failed to obtain bsd path for optical drive"); 3455c1bb86cdSEric Blake error_occurred = true; 3456c1bb86cdSEric Blake goto hdev_open_Mac_error; 3457c1bb86cdSEric Blake } 3458c1bb86cdSEric Blake 3459c1bb86cdSEric Blake /* If using a cdrom disc and finding a partition on the disc failed */ 3460c1bb86cdSEric Blake if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && 3461c1bb86cdSEric Blake setup_cdrom(bsd_path, errp) == false) { 3462c1bb86cdSEric Blake print_unmounting_directions(bsd_path); 3463c1bb86cdSEric Blake error_occurred = true; 3464c1bb86cdSEric Blake goto hdev_open_Mac_error; 3465c1bb86cdSEric Blake } 3466c1bb86cdSEric Blake 346746f5ac20SEric Blake qdict_put_str(options, "filename", bsd_path); 3468c1bb86cdSEric Blake 3469c1bb86cdSEric Blake hdev_open_Mac_error: 3470c1bb86cdSEric Blake g_free(mediaType); 3471c1bb86cdSEric Blake if (mediaIterator) { 3472c1bb86cdSEric Blake IOObjectRelease(mediaIterator); 3473c1bb86cdSEric Blake } 3474c1bb86cdSEric Blake if (error_occurred) { 3475c1bb86cdSEric Blake return -ENOENT; 3476c1bb86cdSEric Blake } 3477c1bb86cdSEric Blake } 3478c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3479c1bb86cdSEric Blake 3480c1bb86cdSEric Blake s->type = FTYPE_FILE; 3481c1bb86cdSEric Blake 3482668f62ecSMarkus Armbruster ret = raw_open_common(bs, options, flags, 0, true, errp); 3483c1bb86cdSEric Blake if (ret < 0) { 3484c1bb86cdSEric Blake #if defined(__APPLE__) && defined(__MACH__) 3485c1bb86cdSEric Blake if (*bsd_path) { 3486c1bb86cdSEric Blake filename = bsd_path; 3487c1bb86cdSEric Blake } 3488c1bb86cdSEric Blake /* if a physical device experienced an error while being opened */ 3489c1bb86cdSEric Blake if (strncmp(filename, "/dev/", 5) == 0) { 3490c1bb86cdSEric Blake print_unmounting_directions(filename); 3491c1bb86cdSEric Blake } 3492c1bb86cdSEric Blake #endif /* defined(__APPLE__) && defined(__MACH__) */ 3493c1bb86cdSEric Blake return ret; 3494c1bb86cdSEric Blake } 3495c1bb86cdSEric Blake 3496c1bb86cdSEric Blake /* Since this does ioctl the device must be already opened */ 3497c1bb86cdSEric Blake bs->sg = hdev_is_sg(bs); 3498c1bb86cdSEric Blake 3499c1bb86cdSEric Blake return ret; 3500c1bb86cdSEric Blake } 3501c1bb86cdSEric Blake 3502c1bb86cdSEric Blake #if defined(__linux__) 35032f3a7ab3SKevin Wolf static int coroutine_fn 35042f3a7ab3SKevin Wolf hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) 3505c1bb86cdSEric Blake { 3506c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 350703425671SKevin Wolf RawPosixAIOData acb; 35082f3a7ab3SKevin Wolf int ret; 3509c1bb86cdSEric Blake 35102f3a7ab3SKevin Wolf ret = fd_open(bs); 35112f3a7ab3SKevin Wolf if (ret < 0) { 35122f3a7ab3SKevin Wolf return ret; 35132f3a7ab3SKevin Wolf } 3514c1bb86cdSEric Blake 35157c9e5276SPaolo Bonzini if (req == SG_IO && s->pr_mgr) { 35167c9e5276SPaolo Bonzini struct sg_io_hdr *io_hdr = buf; 35177c9e5276SPaolo Bonzini if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || 35187c9e5276SPaolo Bonzini io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { 35197c9e5276SPaolo Bonzini return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), 35202f3a7ab3SKevin Wolf s->fd, io_hdr); 35217c9e5276SPaolo Bonzini } 35227c9e5276SPaolo Bonzini } 35237c9e5276SPaolo Bonzini 352403425671SKevin Wolf acb = (RawPosixAIOData) { 352503425671SKevin Wolf .bs = bs, 352603425671SKevin Wolf .aio_type = QEMU_AIO_IOCTL, 352703425671SKevin Wolf .aio_fildes = s->fd, 352803425671SKevin Wolf .aio_offset = 0, 352903425671SKevin Wolf .ioctl = { 353003425671SKevin Wolf .buf = buf, 353103425671SKevin Wolf .cmd = req, 353203425671SKevin Wolf }, 353303425671SKevin Wolf }; 353403425671SKevin Wolf 353503425671SKevin Wolf return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); 3536c1bb86cdSEric Blake } 3537c1bb86cdSEric Blake #endif /* linux */ 3538c1bb86cdSEric Blake 3539c1bb86cdSEric Blake static int fd_open(BlockDriverState *bs) 3540c1bb86cdSEric Blake { 3541c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3542c1bb86cdSEric Blake 3543c1bb86cdSEric Blake /* this is just to ensure s->fd is sane (its called by io ops) */ 3544c1bb86cdSEric Blake if (s->fd >= 0) 3545c1bb86cdSEric Blake return 0; 3546c1bb86cdSEric Blake return -EIO; 3547c1bb86cdSEric Blake } 3548c1bb86cdSEric Blake 354933d70fb6SKevin Wolf static coroutine_fn int 355033d70fb6SKevin Wolf hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) 3551c1bb86cdSEric Blake { 35521c450366SAnton Nefedov BDRVRawState *s = bs->opaque; 355333d70fb6SKevin Wolf int ret; 3554c1bb86cdSEric Blake 355533d70fb6SKevin Wolf ret = fd_open(bs); 355633d70fb6SKevin Wolf if (ret < 0) { 35571c450366SAnton Nefedov raw_account_discard(s, bytes, ret); 355833d70fb6SKevin Wolf return ret; 3559c1bb86cdSEric Blake } 356046ee0f46SKevin Wolf return raw_do_pdiscard(bs, offset, bytes, true); 3561c1bb86cdSEric Blake } 3562c1bb86cdSEric Blake 3563c1bb86cdSEric Blake static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, 3564f5a5ca79SManos Pitsidianakis int64_t offset, int bytes, BdrvRequestFlags flags) 3565c1bb86cdSEric Blake { 3566c1bb86cdSEric Blake int rc; 3567c1bb86cdSEric Blake 3568c1bb86cdSEric Blake rc = fd_open(bs); 3569c1bb86cdSEric Blake if (rc < 0) { 3570c1bb86cdSEric Blake return rc; 3571c1bb86cdSEric Blake } 357234fa110eSKevin Wolf 35737154d8aeSKevin Wolf return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); 3574c1bb86cdSEric Blake } 3575c1bb86cdSEric Blake 3576c1bb86cdSEric Blake static BlockDriver bdrv_host_device = { 3577c1bb86cdSEric Blake .format_name = "host_device", 3578c1bb86cdSEric Blake .protocol_name = "host_device", 3579c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3580c1bb86cdSEric Blake .bdrv_needs_filename = true, 3581c1bb86cdSEric Blake .bdrv_probe_device = hdev_probe_device, 3582c1bb86cdSEric Blake .bdrv_parse_filename = hdev_parse_filename, 3583c1bb86cdSEric Blake .bdrv_file_open = hdev_open, 3584c1bb86cdSEric Blake .bdrv_close = raw_close, 3585c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3586c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3587c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 35885a5e7f8cSMaxim Levitsky .bdrv_co_create_opts = bdrv_co_create_opts_simple, 35895a5e7f8cSMaxim Levitsky .create_opts = &bdrv_create_opts_simple, 35908a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3591dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3592c1bb86cdSEric Blake .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, 3593c1bb86cdSEric Blake 3594c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3595c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 359633d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 359733d70fb6SKevin Wolf .bdrv_co_pdiscard = hdev_co_pdiscard, 35981efad060SFam Zheng .bdrv_co_copy_range_from = raw_co_copy_range_from, 35991efad060SFam Zheng .bdrv_co_copy_range_to = raw_co_copy_range_to, 3600c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3601c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3602c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3603042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3604c1bb86cdSEric Blake 3605061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3606c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3607c1bb86cdSEric Blake .bdrv_get_info = raw_get_info, 3608c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3609c1bb86cdSEric Blake = raw_get_allocated_file_size, 3610d9245599SAnton Nefedov .bdrv_get_specific_stats = hdev_get_specific_stats, 3611244a5668SFam Zheng .bdrv_check_perm = raw_check_perm, 3612244a5668SFam Zheng .bdrv_set_perm = raw_set_perm, 3613244a5668SFam Zheng .bdrv_abort_perm_update = raw_abort_perm_update, 3614c1bb86cdSEric Blake .bdrv_probe_blocksizes = hdev_probe_blocksizes, 3615c1bb86cdSEric Blake .bdrv_probe_geometry = hdev_probe_geometry, 3616c1bb86cdSEric Blake 3617c1bb86cdSEric Blake /* generic scsi device */ 3618c1bb86cdSEric Blake #ifdef __linux__ 36192f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3620c1bb86cdSEric Blake #endif 3621c1bb86cdSEric Blake }; 3622c1bb86cdSEric Blake 3623c1bb86cdSEric Blake #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3624c1bb86cdSEric Blake static void cdrom_parse_filename(const char *filename, QDict *options, 3625c1bb86cdSEric Blake Error **errp) 3626c1bb86cdSEric Blake { 362703c320d8SMax Reitz bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); 3628c1bb86cdSEric Blake } 3629c1bb86cdSEric Blake #endif 3630c1bb86cdSEric Blake 3631c1bb86cdSEric Blake #ifdef __linux__ 3632c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3633c1bb86cdSEric Blake Error **errp) 3634c1bb86cdSEric Blake { 3635c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3636c1bb86cdSEric Blake 3637c1bb86cdSEric Blake s->type = FTYPE_CD; 3638c1bb86cdSEric Blake 3639c1bb86cdSEric Blake /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ 3640230ff739SJohn Snow return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); 3641c1bb86cdSEric Blake } 3642c1bb86cdSEric Blake 3643c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3644c1bb86cdSEric Blake { 3645c1bb86cdSEric Blake int fd, ret; 3646c1bb86cdSEric Blake int prio = 0; 3647c1bb86cdSEric Blake struct stat st; 3648c1bb86cdSEric Blake 3649c1bb86cdSEric Blake fd = qemu_open(filename, O_RDONLY | O_NONBLOCK); 3650c1bb86cdSEric Blake if (fd < 0) { 3651c1bb86cdSEric Blake goto out; 3652c1bb86cdSEric Blake } 3653c1bb86cdSEric Blake ret = fstat(fd, &st); 3654c1bb86cdSEric Blake if (ret == -1 || !S_ISBLK(st.st_mode)) { 3655c1bb86cdSEric Blake goto outc; 3656c1bb86cdSEric Blake } 3657c1bb86cdSEric Blake 3658c1bb86cdSEric Blake /* Attempt to detect via a CDROM specific ioctl */ 3659c1bb86cdSEric Blake ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3660c1bb86cdSEric Blake if (ret >= 0) 3661c1bb86cdSEric Blake prio = 100; 3662c1bb86cdSEric Blake 3663c1bb86cdSEric Blake outc: 3664c1bb86cdSEric Blake qemu_close(fd); 3665c1bb86cdSEric Blake out: 3666c1bb86cdSEric Blake return prio; 3667c1bb86cdSEric Blake } 3668c1bb86cdSEric Blake 3669c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3670c1bb86cdSEric Blake { 3671c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3672c1bb86cdSEric Blake int ret; 3673c1bb86cdSEric Blake 3674c1bb86cdSEric Blake ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); 3675c1bb86cdSEric Blake return ret == CDS_DISC_OK; 3676c1bb86cdSEric Blake } 3677c1bb86cdSEric Blake 3678c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3679c1bb86cdSEric Blake { 3680c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3681c1bb86cdSEric Blake 3682c1bb86cdSEric Blake if (eject_flag) { 3683c1bb86cdSEric Blake if (ioctl(s->fd, CDROMEJECT, NULL) < 0) 3684c1bb86cdSEric Blake perror("CDROMEJECT"); 3685c1bb86cdSEric Blake } else { 3686c1bb86cdSEric Blake if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) 3687c1bb86cdSEric Blake perror("CDROMEJECT"); 3688c1bb86cdSEric Blake } 3689c1bb86cdSEric Blake } 3690c1bb86cdSEric Blake 3691c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3692c1bb86cdSEric Blake { 3693c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3694c1bb86cdSEric Blake 3695c1bb86cdSEric Blake if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { 3696c1bb86cdSEric Blake /* 3697c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3698c1bb86cdSEric Blake * mounts the CD-ROM 3699c1bb86cdSEric Blake */ 3700c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3701c1bb86cdSEric Blake } 3702c1bb86cdSEric Blake } 3703c1bb86cdSEric Blake 3704c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3705c1bb86cdSEric Blake .format_name = "host_cdrom", 3706c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3707c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3708c1bb86cdSEric Blake .bdrv_needs_filename = true, 3709c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3710c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3711c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3712c1bb86cdSEric Blake .bdrv_close = raw_close, 3713c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3714c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3715c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 37165a5e7f8cSMaxim Levitsky .bdrv_co_create_opts = bdrv_co_create_opts_simple, 37175a5e7f8cSMaxim Levitsky .create_opts = &bdrv_create_opts_simple, 37188a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3719dd577a26SStefan Hajnoczi .bdrv_co_invalidate_cache = raw_co_invalidate_cache, 3720c1bb86cdSEric Blake 3721c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3722c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 372333d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3724c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3725c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3726c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3727042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3728c1bb86cdSEric Blake 3729061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3730c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3731c1bb86cdSEric Blake .has_variable_length = true, 3732c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3733c1bb86cdSEric Blake = raw_get_allocated_file_size, 3734c1bb86cdSEric Blake 3735c1bb86cdSEric Blake /* removable device support */ 3736c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3737c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3738c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3739c1bb86cdSEric Blake 3740c1bb86cdSEric Blake /* generic scsi device */ 37412f3a7ab3SKevin Wolf .bdrv_co_ioctl = hdev_co_ioctl, 3742c1bb86cdSEric Blake }; 3743c1bb86cdSEric Blake #endif /* __linux__ */ 3744c1bb86cdSEric Blake 3745c1bb86cdSEric Blake #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) 3746c1bb86cdSEric Blake static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, 3747c1bb86cdSEric Blake Error **errp) 3748c1bb86cdSEric Blake { 3749c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3750c1bb86cdSEric Blake int ret; 3751c1bb86cdSEric Blake 3752c1bb86cdSEric Blake s->type = FTYPE_CD; 3753c1bb86cdSEric Blake 3754668f62ecSMarkus Armbruster ret = raw_open_common(bs, options, flags, 0, true, errp); 3755c1bb86cdSEric Blake if (ret) { 3756c1bb86cdSEric Blake return ret; 3757c1bb86cdSEric Blake } 3758c1bb86cdSEric Blake 3759c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3760c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3761c1bb86cdSEric Blake return 0; 3762c1bb86cdSEric Blake } 3763c1bb86cdSEric Blake 3764c1bb86cdSEric Blake static int cdrom_probe_device(const char *filename) 3765c1bb86cdSEric Blake { 3766c1bb86cdSEric Blake if (strstart(filename, "/dev/cd", NULL) || 3767c1bb86cdSEric Blake strstart(filename, "/dev/acd", NULL)) 3768c1bb86cdSEric Blake return 100; 3769c1bb86cdSEric Blake return 0; 3770c1bb86cdSEric Blake } 3771c1bb86cdSEric Blake 3772c1bb86cdSEric Blake static int cdrom_reopen(BlockDriverState *bs) 3773c1bb86cdSEric Blake { 3774c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3775c1bb86cdSEric Blake int fd; 3776c1bb86cdSEric Blake 3777c1bb86cdSEric Blake /* 3778c1bb86cdSEric Blake * Force reread of possibly changed/newly loaded disc, 3779c1bb86cdSEric Blake * FreeBSD seems to not notice sometimes... 3780c1bb86cdSEric Blake */ 3781c1bb86cdSEric Blake if (s->fd >= 0) 3782c1bb86cdSEric Blake qemu_close(s->fd); 3783c1bb86cdSEric Blake fd = qemu_open(bs->filename, s->open_flags, 0644); 3784c1bb86cdSEric Blake if (fd < 0) { 3785c1bb86cdSEric Blake s->fd = -1; 3786c1bb86cdSEric Blake return -EIO; 3787c1bb86cdSEric Blake } 3788c1bb86cdSEric Blake s->fd = fd; 3789c1bb86cdSEric Blake 3790c1bb86cdSEric Blake /* make sure the door isn't locked at this time */ 3791c1bb86cdSEric Blake ioctl(s->fd, CDIOCALLOW); 3792c1bb86cdSEric Blake return 0; 3793c1bb86cdSEric Blake } 3794c1bb86cdSEric Blake 3795c1bb86cdSEric Blake static bool cdrom_is_inserted(BlockDriverState *bs) 3796c1bb86cdSEric Blake { 3797c1bb86cdSEric Blake return raw_getlength(bs) > 0; 3798c1bb86cdSEric Blake } 3799c1bb86cdSEric Blake 3800c1bb86cdSEric Blake static void cdrom_eject(BlockDriverState *bs, bool eject_flag) 3801c1bb86cdSEric Blake { 3802c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3803c1bb86cdSEric Blake 3804c1bb86cdSEric Blake if (s->fd < 0) 3805c1bb86cdSEric Blake return; 3806c1bb86cdSEric Blake 3807c1bb86cdSEric Blake (void) ioctl(s->fd, CDIOCALLOW); 3808c1bb86cdSEric Blake 3809c1bb86cdSEric Blake if (eject_flag) { 3810c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCEJECT) < 0) 3811c1bb86cdSEric Blake perror("CDIOCEJECT"); 3812c1bb86cdSEric Blake } else { 3813c1bb86cdSEric Blake if (ioctl(s->fd, CDIOCCLOSE) < 0) 3814c1bb86cdSEric Blake perror("CDIOCCLOSE"); 3815c1bb86cdSEric Blake } 3816c1bb86cdSEric Blake 3817c1bb86cdSEric Blake cdrom_reopen(bs); 3818c1bb86cdSEric Blake } 3819c1bb86cdSEric Blake 3820c1bb86cdSEric Blake static void cdrom_lock_medium(BlockDriverState *bs, bool locked) 3821c1bb86cdSEric Blake { 3822c1bb86cdSEric Blake BDRVRawState *s = bs->opaque; 3823c1bb86cdSEric Blake 3824c1bb86cdSEric Blake if (s->fd < 0) 3825c1bb86cdSEric Blake return; 3826c1bb86cdSEric Blake if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { 3827c1bb86cdSEric Blake /* 3828c1bb86cdSEric Blake * Note: an error can happen if the distribution automatically 3829c1bb86cdSEric Blake * mounts the CD-ROM 3830c1bb86cdSEric Blake */ 3831c1bb86cdSEric Blake /* perror("CDROM_LOCKDOOR"); */ 3832c1bb86cdSEric Blake } 3833c1bb86cdSEric Blake } 3834c1bb86cdSEric Blake 3835c1bb86cdSEric Blake static BlockDriver bdrv_host_cdrom = { 3836c1bb86cdSEric Blake .format_name = "host_cdrom", 3837c1bb86cdSEric Blake .protocol_name = "host_cdrom", 3838c1bb86cdSEric Blake .instance_size = sizeof(BDRVRawState), 3839c1bb86cdSEric Blake .bdrv_needs_filename = true, 3840c1bb86cdSEric Blake .bdrv_probe_device = cdrom_probe_device, 3841c1bb86cdSEric Blake .bdrv_parse_filename = cdrom_parse_filename, 3842c1bb86cdSEric Blake .bdrv_file_open = cdrom_open, 3843c1bb86cdSEric Blake .bdrv_close = raw_close, 3844c1bb86cdSEric Blake .bdrv_reopen_prepare = raw_reopen_prepare, 3845c1bb86cdSEric Blake .bdrv_reopen_commit = raw_reopen_commit, 3846c1bb86cdSEric Blake .bdrv_reopen_abort = raw_reopen_abort, 38475a5e7f8cSMaxim Levitsky .bdrv_co_create_opts = bdrv_co_create_opts_simple, 38485a5e7f8cSMaxim Levitsky .create_opts = &bdrv_create_opts_simple, 38498a2ce0bcSAlberto Garcia .mutable_opts = mutable_opts, 3850c1bb86cdSEric Blake 3851c1bb86cdSEric Blake .bdrv_co_preadv = raw_co_preadv, 3852c1bb86cdSEric Blake .bdrv_co_pwritev = raw_co_pwritev, 385333d70fb6SKevin Wolf .bdrv_co_flush_to_disk = raw_co_flush_to_disk, 3854c1bb86cdSEric Blake .bdrv_refresh_limits = raw_refresh_limits, 3855c1bb86cdSEric Blake .bdrv_io_plug = raw_aio_plug, 3856c1bb86cdSEric Blake .bdrv_io_unplug = raw_aio_unplug, 3857042b757cSNishanth Aravamudan .bdrv_attach_aio_context = raw_aio_attach_aio_context, 3858c1bb86cdSEric Blake 3859061ca8a3SKevin Wolf .bdrv_co_truncate = raw_co_truncate, 3860c1bb86cdSEric Blake .bdrv_getlength = raw_getlength, 3861c1bb86cdSEric Blake .has_variable_length = true, 3862c1bb86cdSEric Blake .bdrv_get_allocated_file_size 3863c1bb86cdSEric Blake = raw_get_allocated_file_size, 3864c1bb86cdSEric Blake 3865c1bb86cdSEric Blake /* removable device support */ 3866c1bb86cdSEric Blake .bdrv_is_inserted = cdrom_is_inserted, 3867c1bb86cdSEric Blake .bdrv_eject = cdrom_eject, 3868c1bb86cdSEric Blake .bdrv_lock_medium = cdrom_lock_medium, 3869c1bb86cdSEric Blake }; 3870c1bb86cdSEric Blake #endif /* __FreeBSD__ */ 3871c1bb86cdSEric Blake 3872c1bb86cdSEric Blake static void bdrv_file_init(void) 3873c1bb86cdSEric Blake { 3874c1bb86cdSEric Blake /* 3875c1bb86cdSEric Blake * Register all the drivers. Note that order is important, the driver 3876c1bb86cdSEric Blake * registered last will get probed first. 3877c1bb86cdSEric Blake */ 3878c1bb86cdSEric Blake bdrv_register(&bdrv_file); 3879c1bb86cdSEric Blake bdrv_register(&bdrv_host_device); 3880c1bb86cdSEric Blake #ifdef __linux__ 3881c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3882c1bb86cdSEric Blake #endif 3883c1bb86cdSEric Blake #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 3884c1bb86cdSEric Blake bdrv_register(&bdrv_host_cdrom); 3885c1bb86cdSEric Blake #endif 3886c1bb86cdSEric Blake } 3887c1bb86cdSEric Blake 3888c1bb86cdSEric Blake block_init(bdrv_file_init); 3889