block: fix deadlock in bdrv_co_flush
[qemu/kevin.git] / block / raw-win32.c
blob56f45fea9e7c028b762e46c7068679b9404fb629
1 /*
2 * Block driver for RAW files (win32)
4 * Copyright (c) 2006 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
24 #include "qemu/osdep.h"
25 #include "qapi/error.h"
26 #include "qemu/cutils.h"
27 #include "qemu/timer.h"
28 #include "block/block_int.h"
29 #include "qemu/module.h"
30 #include "block/raw-aio.h"
31 #include "trace.h"
32 #include "block/thread-pool.h"
33 #include "qemu/iov.h"
34 #include "qapi/qmp/qstring.h"
35 #include <windows.h>
36 #include <winioctl.h>
38 #define FTYPE_FILE 0
39 #define FTYPE_CD 1
40 #define FTYPE_HARDDISK 2
42 typedef struct RawWin32AIOData {
43 BlockDriverState *bs;
44 HANDLE hfile;
45 struct iovec *aio_iov;
46 int aio_niov;
47 size_t aio_nbytes;
48 off64_t aio_offset;
49 int aio_type;
50 } RawWin32AIOData;
52 typedef struct BDRVRawState {
53 HANDLE hfile;
54 int type;
55 char drive_path[16]; /* format: "d:\" */
56 QEMUWin32AIOState *aio;
57 } BDRVRawState;
60 * Read/writes the data to/from a given linear buffer.
62 * Returns the number of bytes handles or -errno in case of an error. Short
63 * reads are only returned if the end of the file is reached.
65 static size_t handle_aiocb_rw(RawWin32AIOData *aiocb)
67 size_t offset = 0;
68 int i;
70 for (i = 0; i < aiocb->aio_niov; i++) {
71 OVERLAPPED ov;
72 DWORD ret, ret_count, len;
74 memset(&ov, 0, sizeof(ov));
75 ov.Offset = (aiocb->aio_offset + offset);
76 ov.OffsetHigh = (aiocb->aio_offset + offset) >> 32;
77 len = aiocb->aio_iov[i].iov_len;
78 if (aiocb->aio_type & QEMU_AIO_WRITE) {
79 ret = WriteFile(aiocb->hfile, aiocb->aio_iov[i].iov_base,
80 len, &ret_count, &ov);
81 } else {
82 ret = ReadFile(aiocb->hfile, aiocb->aio_iov[i].iov_base,
83 len, &ret_count, &ov);
85 if (!ret) {
86 ret_count = 0;
88 if (ret_count != len) {
89 offset += ret_count;
90 break;
92 offset += len;
95 return offset;
98 static int aio_worker(void *arg)
100 RawWin32AIOData *aiocb = arg;
101 ssize_t ret = 0;
102 size_t count;
104 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
105 case QEMU_AIO_READ:
106 count = handle_aiocb_rw(aiocb);
107 if (count < aiocb->aio_nbytes) {
108 /* A short read means that we have reached EOF. Pad the buffer
109 * with zeros for bytes after EOF. */
110 iov_memset(aiocb->aio_iov, aiocb->aio_niov, count,
111 0, aiocb->aio_nbytes - count);
113 count = aiocb->aio_nbytes;
115 if (count == aiocb->aio_nbytes) {
116 ret = 0;
117 } else {
118 ret = -EINVAL;
120 break;
121 case QEMU_AIO_WRITE:
122 count = handle_aiocb_rw(aiocb);
123 if (count == aiocb->aio_nbytes) {
124 ret = 0;
125 } else {
126 ret = -EINVAL;
128 break;
129 case QEMU_AIO_FLUSH:
130 if (!FlushFileBuffers(aiocb->hfile)) {
131 return -EIO;
133 break;
134 default:
135 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
136 ret = -EINVAL;
137 break;
140 g_free(aiocb);
141 return ret;
144 static BlockAIOCB *paio_submit(BlockDriverState *bs, HANDLE hfile,
145 int64_t offset, QEMUIOVector *qiov, int count,
146 BlockCompletionFunc *cb, void *opaque, int type)
148 RawWin32AIOData *acb = g_new(RawWin32AIOData, 1);
149 ThreadPool *pool;
151 acb->bs = bs;
152 acb->hfile = hfile;
153 acb->aio_type = type;
155 if (qiov) {
156 acb->aio_iov = qiov->iov;
157 acb->aio_niov = qiov->niov;
158 assert(qiov->size == count);
160 acb->aio_nbytes = count;
161 acb->aio_offset = offset;
163 trace_paio_submit(acb, opaque, offset, count, type);
164 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
165 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
168 int qemu_ftruncate64(int fd, int64_t length)
170 LARGE_INTEGER li;
171 DWORD dw;
172 LONG high;
173 HANDLE h;
174 BOOL res;
176 if ((GetVersion() & 0x80000000UL) && (length >> 32) != 0)
177 return -1;
179 h = (HANDLE)_get_osfhandle(fd);
181 /* get current position, ftruncate do not change position */
182 li.HighPart = 0;
183 li.LowPart = SetFilePointer (h, 0, &li.HighPart, FILE_CURRENT);
184 if (li.LowPart == INVALID_SET_FILE_POINTER && GetLastError() != NO_ERROR) {
185 return -1;
188 high = length >> 32;
189 dw = SetFilePointer(h, (DWORD) length, &high, FILE_BEGIN);
190 if (dw == INVALID_SET_FILE_POINTER && GetLastError() != NO_ERROR) {
191 return -1;
193 res = SetEndOfFile(h);
195 /* back to old position */
196 SetFilePointer(h, li.LowPart, &li.HighPart, FILE_BEGIN);
197 return res ? 0 : -1;
200 static int set_sparse(int fd)
202 DWORD returned;
203 return (int) DeviceIoControl((HANDLE)_get_osfhandle(fd), FSCTL_SET_SPARSE,
204 NULL, 0, NULL, 0, &returned, NULL);
207 static void raw_detach_aio_context(BlockDriverState *bs)
209 BDRVRawState *s = bs->opaque;
211 if (s->aio) {
212 win32_aio_detach_aio_context(s->aio, bdrv_get_aio_context(bs));
216 static void raw_attach_aio_context(BlockDriverState *bs,
217 AioContext *new_context)
219 BDRVRawState *s = bs->opaque;
221 if (s->aio) {
222 win32_aio_attach_aio_context(s->aio, new_context);
226 static void raw_probe_alignment(BlockDriverState *bs, Error **errp)
228 BDRVRawState *s = bs->opaque;
229 DWORD sectorsPerCluster, freeClusters, totalClusters, count;
230 DISK_GEOMETRY_EX dg;
231 BOOL status;
233 if (s->type == FTYPE_CD) {
234 bs->bl.request_alignment = 2048;
235 return;
237 if (s->type == FTYPE_HARDDISK) {
238 status = DeviceIoControl(s->hfile, IOCTL_DISK_GET_DRIVE_GEOMETRY_EX,
239 NULL, 0, &dg, sizeof(dg), &count, NULL);
240 if (status != 0) {
241 bs->bl.request_alignment = dg.Geometry.BytesPerSector;
242 return;
244 /* try GetDiskFreeSpace too */
247 if (s->drive_path[0]) {
248 GetDiskFreeSpace(s->drive_path, &sectorsPerCluster,
249 &dg.Geometry.BytesPerSector,
250 &freeClusters, &totalClusters);
251 bs->bl.request_alignment = dg.Geometry.BytesPerSector;
255 static void raw_parse_flags(int flags, int *access_flags, DWORD *overlapped)
257 assert(access_flags != NULL);
258 assert(overlapped != NULL);
260 if (flags & BDRV_O_RDWR) {
261 *access_flags = GENERIC_READ | GENERIC_WRITE;
262 } else {
263 *access_flags = GENERIC_READ;
266 *overlapped = FILE_ATTRIBUTE_NORMAL;
267 if (flags & BDRV_O_NATIVE_AIO) {
268 *overlapped |= FILE_FLAG_OVERLAPPED;
270 if (flags & BDRV_O_NOCACHE) {
271 *overlapped |= FILE_FLAG_NO_BUFFERING;
275 static void raw_parse_filename(const char *filename, QDict *options,
276 Error **errp)
278 /* The filename does not have to be prefixed by the protocol name, since
279 * "file" is the default protocol; therefore, the return value of this
280 * function call can be ignored. */
281 strstart(filename, "file:", &filename);
283 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
286 static QemuOptsList raw_runtime_opts = {
287 .name = "raw",
288 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
289 .desc = {
291 .name = "filename",
292 .type = QEMU_OPT_STRING,
293 .help = "File name of the image",
295 { /* end of list */ }
299 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
300 Error **errp)
302 BDRVRawState *s = bs->opaque;
303 int access_flags;
304 DWORD overlapped;
305 QemuOpts *opts;
306 Error *local_err = NULL;
307 const char *filename;
308 int ret;
310 s->type = FTYPE_FILE;
312 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
313 qemu_opts_absorb_qdict(opts, options, &local_err);
314 if (local_err) {
315 error_propagate(errp, local_err);
316 ret = -EINVAL;
317 goto fail;
320 filename = qemu_opt_get(opts, "filename");
322 raw_parse_flags(flags, &access_flags, &overlapped);
324 if (filename[0] && filename[1] == ':') {
325 snprintf(s->drive_path, sizeof(s->drive_path), "%c:\\", filename[0]);
326 } else if (filename[0] == '\\' && filename[1] == '\\') {
327 s->drive_path[0] = 0;
328 } else {
329 /* Relative path. */
330 char buf[MAX_PATH];
331 GetCurrentDirectory(MAX_PATH, buf);
332 snprintf(s->drive_path, sizeof(s->drive_path), "%c:\\", buf[0]);
335 s->hfile = CreateFile(filename, access_flags,
336 FILE_SHARE_READ, NULL,
337 OPEN_EXISTING, overlapped, NULL);
338 if (s->hfile == INVALID_HANDLE_VALUE) {
339 int err = GetLastError();
341 if (err == ERROR_ACCESS_DENIED) {
342 ret = -EACCES;
343 } else {
344 ret = -EINVAL;
346 goto fail;
349 if (flags & BDRV_O_NATIVE_AIO) {
350 s->aio = win32_aio_init();
351 if (s->aio == NULL) {
352 CloseHandle(s->hfile);
353 error_setg(errp, "Could not initialize AIO");
354 ret = -EINVAL;
355 goto fail;
358 ret = win32_aio_attach(s->aio, s->hfile);
359 if (ret < 0) {
360 win32_aio_cleanup(s->aio);
361 CloseHandle(s->hfile);
362 error_setg_errno(errp, -ret, "Could not enable AIO");
363 goto fail;
366 win32_aio_attach_aio_context(s->aio, bdrv_get_aio_context(bs));
369 ret = 0;
370 fail:
371 qemu_opts_del(opts);
372 return ret;
375 static BlockAIOCB *raw_aio_readv(BlockDriverState *bs,
376 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
377 BlockCompletionFunc *cb, void *opaque)
379 BDRVRawState *s = bs->opaque;
380 if (s->aio) {
381 return win32_aio_submit(bs, s->aio, s->hfile, sector_num, qiov,
382 nb_sectors, cb, opaque, QEMU_AIO_READ);
383 } else {
384 return paio_submit(bs, s->hfile, sector_num << BDRV_SECTOR_BITS, qiov,
385 nb_sectors << BDRV_SECTOR_BITS,
386 cb, opaque, QEMU_AIO_READ);
390 static BlockAIOCB *raw_aio_writev(BlockDriverState *bs,
391 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
392 BlockCompletionFunc *cb, void *opaque)
394 BDRVRawState *s = bs->opaque;
395 if (s->aio) {
396 return win32_aio_submit(bs, s->aio, s->hfile, sector_num, qiov,
397 nb_sectors, cb, opaque, QEMU_AIO_WRITE);
398 } else {
399 return paio_submit(bs, s->hfile, sector_num << BDRV_SECTOR_BITS, qiov,
400 nb_sectors << BDRV_SECTOR_BITS,
401 cb, opaque, QEMU_AIO_WRITE);
405 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
406 BlockCompletionFunc *cb, void *opaque)
408 BDRVRawState *s = bs->opaque;
409 return paio_submit(bs, s->hfile, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
412 static void raw_close(BlockDriverState *bs)
414 BDRVRawState *s = bs->opaque;
416 if (s->aio) {
417 win32_aio_detach_aio_context(s->aio, bdrv_get_aio_context(bs));
418 win32_aio_cleanup(s->aio);
419 s->aio = NULL;
422 CloseHandle(s->hfile);
423 if (bs->open_flags & BDRV_O_TEMPORARY) {
424 unlink(bs->filename);
428 static int raw_truncate(BlockDriverState *bs, int64_t offset)
430 BDRVRawState *s = bs->opaque;
431 LONG low, high;
432 DWORD dwPtrLow;
434 low = offset;
435 high = offset >> 32;
438 * An error has occurred if the return value is INVALID_SET_FILE_POINTER
439 * and GetLastError doesn't return NO_ERROR.
441 dwPtrLow = SetFilePointer(s->hfile, low, &high, FILE_BEGIN);
442 if (dwPtrLow == INVALID_SET_FILE_POINTER && GetLastError() != NO_ERROR) {
443 fprintf(stderr, "SetFilePointer error: %lu\n", GetLastError());
444 return -EIO;
446 if (SetEndOfFile(s->hfile) == 0) {
447 fprintf(stderr, "SetEndOfFile error: %lu\n", GetLastError());
448 return -EIO;
450 return 0;
453 static int64_t raw_getlength(BlockDriverState *bs)
455 BDRVRawState *s = bs->opaque;
456 LARGE_INTEGER l;
457 ULARGE_INTEGER available, total, total_free;
458 DISK_GEOMETRY_EX dg;
459 DWORD count;
460 BOOL status;
462 switch(s->type) {
463 case FTYPE_FILE:
464 l.LowPart = GetFileSize(s->hfile, (PDWORD)&l.HighPart);
465 if (l.LowPart == 0xffffffffUL && GetLastError() != NO_ERROR)
466 return -EIO;
467 break;
468 case FTYPE_CD:
469 if (!GetDiskFreeSpaceEx(s->drive_path, &available, &total, &total_free))
470 return -EIO;
471 l.QuadPart = total.QuadPart;
472 break;
473 case FTYPE_HARDDISK:
474 status = DeviceIoControl(s->hfile, IOCTL_DISK_GET_DRIVE_GEOMETRY_EX,
475 NULL, 0, &dg, sizeof(dg), &count, NULL);
476 if (status != 0) {
477 l = dg.DiskSize;
479 break;
480 default:
481 return -EIO;
483 return l.QuadPart;
486 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
488 typedef DWORD (WINAPI * get_compressed_t)(const char *filename,
489 DWORD * high);
490 get_compressed_t get_compressed;
491 struct _stati64 st;
492 const char *filename = bs->filename;
493 /* WinNT support GetCompressedFileSize to determine allocate size */
494 get_compressed =
495 (get_compressed_t) GetProcAddress(GetModuleHandle("kernel32"),
496 "GetCompressedFileSizeA");
497 if (get_compressed) {
498 DWORD high, low;
499 low = get_compressed(filename, &high);
500 if (low != 0xFFFFFFFFlu || GetLastError() == NO_ERROR) {
501 return (((int64_t) high) << 32) + low;
505 if (_stati64(filename, &st) < 0) {
506 return -1;
508 return st.st_size;
511 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
513 int fd;
514 int64_t total_size = 0;
516 strstart(filename, "file:", &filename);
518 /* Read out options */
519 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
520 BDRV_SECTOR_SIZE);
522 fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
523 0644);
524 if (fd < 0) {
525 error_setg_errno(errp, errno, "Could not create file");
526 return -EIO;
528 set_sparse(fd);
529 ftruncate(fd, total_size);
530 qemu_close(fd);
531 return 0;
535 static QemuOptsList raw_create_opts = {
536 .name = "raw-create-opts",
537 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
538 .desc = {
540 .name = BLOCK_OPT_SIZE,
541 .type = QEMU_OPT_SIZE,
542 .help = "Virtual disk size"
544 { /* end of list */ }
548 BlockDriver bdrv_file = {
549 .format_name = "file",
550 .protocol_name = "file",
551 .instance_size = sizeof(BDRVRawState),
552 .bdrv_needs_filename = true,
553 .bdrv_parse_filename = raw_parse_filename,
554 .bdrv_file_open = raw_open,
555 .bdrv_refresh_limits = raw_probe_alignment,
556 .bdrv_close = raw_close,
557 .bdrv_create = raw_create,
558 .bdrv_has_zero_init = bdrv_has_zero_init_1,
560 .bdrv_aio_readv = raw_aio_readv,
561 .bdrv_aio_writev = raw_aio_writev,
562 .bdrv_aio_flush = raw_aio_flush,
564 .bdrv_truncate = raw_truncate,
565 .bdrv_getlength = raw_getlength,
566 .bdrv_get_allocated_file_size
567 = raw_get_allocated_file_size,
569 .create_opts = &raw_create_opts,
572 /***********************************************/
573 /* host device */
575 static int find_cdrom(char *cdrom_name, int cdrom_name_size)
577 char drives[256], *pdrv = drives;
578 UINT type;
580 memset(drives, 0, sizeof(drives));
581 GetLogicalDriveStrings(sizeof(drives), drives);
582 while(pdrv[0] != '\0') {
583 type = GetDriveType(pdrv);
584 switch(type) {
585 case DRIVE_CDROM:
586 snprintf(cdrom_name, cdrom_name_size, "\\\\.\\%c:", pdrv[0]);
587 return 0;
588 break;
590 pdrv += lstrlen(pdrv) + 1;
592 return -1;
595 static int find_device_type(BlockDriverState *bs, const char *filename)
597 BDRVRawState *s = bs->opaque;
598 UINT type;
599 const char *p;
601 if (strstart(filename, "\\\\.\\", &p) ||
602 strstart(filename, "//./", &p)) {
603 if (stristart(p, "PhysicalDrive", NULL))
604 return FTYPE_HARDDISK;
605 snprintf(s->drive_path, sizeof(s->drive_path), "%c:\\", p[0]);
606 type = GetDriveType(s->drive_path);
607 switch (type) {
608 case DRIVE_REMOVABLE:
609 case DRIVE_FIXED:
610 return FTYPE_HARDDISK;
611 case DRIVE_CDROM:
612 return FTYPE_CD;
613 default:
614 return FTYPE_FILE;
616 } else {
617 return FTYPE_FILE;
621 static int hdev_probe_device(const char *filename)
623 if (strstart(filename, "/dev/cdrom", NULL))
624 return 100;
625 if (is_windows_drive(filename))
626 return 100;
627 return 0;
630 static void hdev_parse_filename(const char *filename, QDict *options,
631 Error **errp)
633 /* The prefix is optional, just as for "file". */
634 strstart(filename, "host_device:", &filename);
636 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
639 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
640 Error **errp)
642 BDRVRawState *s = bs->opaque;
643 int access_flags, create_flags;
644 int ret = 0;
645 DWORD overlapped;
646 char device_name[64];
648 Error *local_err = NULL;
649 const char *filename;
651 QemuOpts *opts = qemu_opts_create(&raw_runtime_opts, NULL, 0,
652 &error_abort);
653 qemu_opts_absorb_qdict(opts, options, &local_err);
654 if (local_err) {
655 error_propagate(errp, local_err);
656 ret = -EINVAL;
657 goto done;
660 filename = qemu_opt_get(opts, "filename");
662 if (strstart(filename, "/dev/cdrom", NULL)) {
663 if (find_cdrom(device_name, sizeof(device_name)) < 0) {
664 error_setg(errp, "Could not open CD-ROM drive");
665 ret = -ENOENT;
666 goto done;
668 filename = device_name;
669 } else {
670 /* transform drive letters into device name */
671 if (((filename[0] >= 'a' && filename[0] <= 'z') ||
672 (filename[0] >= 'A' && filename[0] <= 'Z')) &&
673 filename[1] == ':' && filename[2] == '\0') {
674 snprintf(device_name, sizeof(device_name), "\\\\.\\%c:", filename[0]);
675 filename = device_name;
678 s->type = find_device_type(bs, filename);
680 raw_parse_flags(flags, &access_flags, &overlapped);
682 create_flags = OPEN_EXISTING;
684 s->hfile = CreateFile(filename, access_flags,
685 FILE_SHARE_READ, NULL,
686 create_flags, overlapped, NULL);
687 if (s->hfile == INVALID_HANDLE_VALUE) {
688 int err = GetLastError();
690 if (err == ERROR_ACCESS_DENIED) {
691 ret = -EACCES;
692 } else {
693 ret = -EINVAL;
695 error_setg_errno(errp, -ret, "Could not open device");
696 goto done;
699 done:
700 qemu_opts_del(opts);
701 return ret;
704 static BlockDriver bdrv_host_device = {
705 .format_name = "host_device",
706 .protocol_name = "host_device",
707 .instance_size = sizeof(BDRVRawState),
708 .bdrv_needs_filename = true,
709 .bdrv_parse_filename = hdev_parse_filename,
710 .bdrv_probe_device = hdev_probe_device,
711 .bdrv_file_open = hdev_open,
712 .bdrv_close = raw_close,
714 .bdrv_aio_readv = raw_aio_readv,
715 .bdrv_aio_writev = raw_aio_writev,
716 .bdrv_aio_flush = raw_aio_flush,
718 .bdrv_detach_aio_context = raw_detach_aio_context,
719 .bdrv_attach_aio_context = raw_attach_aio_context,
721 .bdrv_getlength = raw_getlength,
722 .has_variable_length = true,
724 .bdrv_get_allocated_file_size
725 = raw_get_allocated_file_size,
728 static void bdrv_file_init(void)
730 bdrv_register(&bdrv_file);
731 bdrv_register(&bdrv_host_device);
734 block_init(bdrv_file_init);