4 * Copyright (c) 2003-2008 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
24 #include "qemu/osdep.h"
26 #include "qemu/madvise.h"
27 #include "qemu/error-report.h"
29 #include "migration.h"
30 #include "migration-stats.h"
31 #include "qemu-file.h"
34 #include "qapi/error.h"
36 #define IO_BUF_SIZE 32768
37 #define MAX_IOV_SIZE MIN_CONST(IOV_MAX, 64)
40 const QEMUFileHooks
*hooks
;
44 /* The sum of bytes transferred on the wire */
45 uint64_t total_transferred
;
48 int buf_size
; /* 0 when writing */
49 uint8_t buf
[IO_BUF_SIZE
];
51 DECLARE_BITMAP(may_free
, MAX_IOV_SIZE
);
52 struct iovec iov
[MAX_IOV_SIZE
];
56 Error
*last_error_obj
;
60 * Stop a file from being read/written - not all backing files can do this
61 * typically only sockets can.
63 * TODO: convert to propagate Error objects instead of squashing
64 * to a fixed errno value
66 int qemu_file_shutdown(QEMUFile
*f
)
69 * We must set qemufile error before the real shutdown(), otherwise
70 * there can be a race window where we thought IO all went though
71 * (because last_error==NULL) but actually IO has already stopped.
73 * If without correct ordering, the race can happen like this:
75 * page receiver other thread
76 * ------------- ------------
79 * returns 0 (buffer all zero)
80 * (we didn't check this retcode)
81 * try to detect IO error
82 * last_error==NULL, IO okay
83 * install ALL-ZERO page
88 qemu_file_set_error(f
, -EIO
);
91 if (!qio_channel_has_feature(f
->ioc
,
92 QIO_CHANNEL_FEATURE_SHUTDOWN
)) {
96 if (qio_channel_shutdown(f
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
, NULL
) < 0) {
103 static QEMUFile
*qemu_file_new_impl(QIOChannel
*ioc
, bool is_writable
)
107 f
= g_new0(QEMUFile
, 1);
111 f
->is_writable
= is_writable
;
117 * Result: QEMUFile* for a 'return path' for comms in the opposite direction
118 * NULL if not available
120 QEMUFile
*qemu_file_get_return_path(QEMUFile
*f
)
122 return qemu_file_new_impl(f
->ioc
, !f
->is_writable
);
125 QEMUFile
*qemu_file_new_output(QIOChannel
*ioc
)
127 return qemu_file_new_impl(ioc
, true);
130 QEMUFile
*qemu_file_new_input(QIOChannel
*ioc
)
132 return qemu_file_new_impl(ioc
, false);
135 void qemu_file_set_hooks(QEMUFile
*f
, const QEMUFileHooks
*hooks
)
141 * Get last error for stream f with optional Error*
143 * Return negative error value if there has been an error on previous
144 * operations, return 0 if no error happened.
145 * Optional, it returns Error* in errp, but it may be NULL even if return value
149 static int qemu_file_get_error_obj(QEMUFile
*f
, Error
**errp
)
152 *errp
= f
->last_error_obj
? error_copy(f
->last_error_obj
) : NULL
;
154 return f
->last_error
;
158 * Get last error for either stream f1 or f2 with optional Error*.
159 * The error returned (non-zero) can be either from f1 or f2.
161 * If any of the qemufile* is NULL, then skip the check on that file.
163 * When there is no error on both qemufile, zero is returned.
165 int qemu_file_get_error_obj_any(QEMUFile
*f1
, QEMUFile
*f2
, Error
**errp
)
170 ret
= qemu_file_get_error_obj(f1
, errp
);
171 /* If there's already error detected, return */
178 ret
= qemu_file_get_error_obj(f2
, errp
);
185 * Set the last error for stream f with optional Error*
187 void qemu_file_set_error_obj(QEMUFile
*f
, int ret
, Error
*err
)
189 if (f
->last_error
== 0 && ret
) {
191 error_propagate(&f
->last_error_obj
, err
);
193 error_report_err(err
);
198 * Get last error for stream f
200 * Return negative error value if there has been an error on previous
201 * operations, return 0 if no error happened.
204 int qemu_file_get_error(QEMUFile
*f
)
206 return qemu_file_get_error_obj(f
, NULL
);
210 * Set the last error for stream f
212 void qemu_file_set_error(QEMUFile
*f
, int ret
)
214 qemu_file_set_error_obj(f
, ret
, NULL
);
217 static bool qemu_file_is_writable(QEMUFile
*f
)
219 return f
->is_writable
;
222 static void qemu_iovec_release_ram(QEMUFile
*f
)
227 /* Find and release all the contiguous memory ranges marked as may_free. */
228 idx
= find_next_bit(f
->may_free
, f
->iovcnt
, 0);
229 if (idx
>= f
->iovcnt
) {
234 /* The madvise() in the loop is called for iov within a continuous range and
235 * then reinitialize the iov. And in the end, madvise() is called for the
238 while ((idx
= find_next_bit(f
->may_free
, f
->iovcnt
, idx
+ 1)) < f
->iovcnt
) {
239 /* check for adjacent buffer and coalesce them */
240 if (iov
.iov_base
+ iov
.iov_len
== f
->iov
[idx
].iov_base
) {
241 iov
.iov_len
+= f
->iov
[idx
].iov_len
;
244 if (qemu_madvise(iov
.iov_base
, iov
.iov_len
, QEMU_MADV_DONTNEED
) < 0) {
245 error_report("migrate: madvise DONTNEED failed %p %zd: %s",
246 iov
.iov_base
, iov
.iov_len
, strerror(errno
));
250 if (qemu_madvise(iov
.iov_base
, iov
.iov_len
, QEMU_MADV_DONTNEED
) < 0) {
251 error_report("migrate: madvise DONTNEED failed %p %zd: %s",
252 iov
.iov_base
, iov
.iov_len
, strerror(errno
));
254 memset(f
->may_free
, 0, sizeof(f
->may_free
));
259 * Flushes QEMUFile buffer
261 * This will flush all pending data. If data was only partially flushed, it
262 * will set an error state.
264 void qemu_fflush(QEMUFile
*f
)
266 if (!qemu_file_is_writable(f
)) {
270 if (qemu_file_get_error(f
)) {
274 Error
*local_error
= NULL
;
275 if (qio_channel_writev_all(f
->ioc
,
278 qemu_file_set_error_obj(f
, -EIO
, local_error
);
280 uint64_t size
= iov_size(f
->iov
, f
->iovcnt
);
281 f
->total_transferred
+= size
;
284 qemu_iovec_release_ram(f
);
291 void ram_control_before_iterate(QEMUFile
*f
, uint64_t flags
)
295 if (f
->hooks
&& f
->hooks
->before_ram_iterate
) {
296 ret
= f
->hooks
->before_ram_iterate(f
, flags
, NULL
);
298 qemu_file_set_error(f
, ret
);
303 void ram_control_after_iterate(QEMUFile
*f
, uint64_t flags
)
307 if (f
->hooks
&& f
->hooks
->after_ram_iterate
) {
308 ret
= f
->hooks
->after_ram_iterate(f
, flags
, NULL
);
310 qemu_file_set_error(f
, ret
);
315 void ram_control_load_hook(QEMUFile
*f
, uint64_t flags
, void *data
)
317 if (f
->hooks
&& f
->hooks
->hook_ram_load
) {
318 int ret
= f
->hooks
->hook_ram_load(f
, flags
, data
);
320 qemu_file_set_error(f
, ret
);
325 int ram_control_save_page(QEMUFile
*f
, ram_addr_t block_offset
,
326 ram_addr_t offset
, size_t size
)
328 if (f
->hooks
&& f
->hooks
->save_page
) {
329 int ret
= f
->hooks
->save_page(f
, block_offset
, offset
, size
);
331 * RAM_SAVE_CONTROL_* are negative values
333 if (ret
!= RAM_SAVE_CONTROL_DELAYED
&&
334 ret
!= RAM_SAVE_CONTROL_NOT_SUPP
) {
336 qemu_file_set_error(f
, ret
);
342 return RAM_SAVE_CONTROL_NOT_SUPP
;
346 * Attempt to fill the buffer from the underlying file
347 * Returns the number of bytes read, or negative value for an error.
349 * Note that it can return a partially full buffer even in a not error/not EOF
350 * case if the underlying file descriptor gives a short read, and that can
351 * happen even on a blocking fd.
353 static ssize_t coroutine_mixed_fn
qemu_fill_buffer(QEMUFile
*f
)
357 Error
*local_error
= NULL
;
359 assert(!qemu_file_is_writable(f
));
361 pending
= f
->buf_size
- f
->buf_index
;
363 memmove(f
->buf
, f
->buf
+ f
->buf_index
, pending
);
366 f
->buf_size
= pending
;
368 if (qemu_file_get_error(f
)) {
373 len
= qio_channel_read(f
->ioc
,
374 (char *)f
->buf
+ pending
,
375 IO_BUF_SIZE
- pending
,
377 if (len
== QIO_CHANNEL_ERR_BLOCK
) {
378 if (qemu_in_coroutine()) {
379 qio_channel_yield(f
->ioc
, G_IO_IN
);
381 qio_channel_wait(f
->ioc
, G_IO_IN
);
383 } else if (len
< 0) {
386 } while (len
== QIO_CHANNEL_ERR_BLOCK
);
390 f
->total_transferred
+= len
;
391 } else if (len
== 0) {
392 qemu_file_set_error_obj(f
, -EIO
, local_error
);
394 qemu_file_set_error_obj(f
, len
, local_error
);
402 * Returns negative error value if any error happened on previous operations or
403 * while closing the file. Returns 0 or positive number on success.
405 * The meaning of return value on success depends on the specific backend
408 int qemu_fclose(QEMUFile
*f
)
412 ret
= qemu_file_get_error(f
);
414 ret2
= qio_channel_close(f
->ioc
, NULL
);
418 g_clear_pointer(&f
->ioc
, object_unref
);
420 /* If any error was spotted before closing, we should report it
421 * instead of the close() return value.
426 error_free(f
->last_error_obj
);
428 trace_qemu_file_fclose();
433 * Add buf to iovec. Do flush if iovec is full.
436 * 1 iovec is full and flushed
437 * 0 iovec is not flushed
440 static int add_to_iovec(QEMUFile
*f
, const uint8_t *buf
, size_t size
,
443 /* check for adjacent buffer and coalesce them */
444 if (f
->iovcnt
> 0 && buf
== f
->iov
[f
->iovcnt
- 1].iov_base
+
445 f
->iov
[f
->iovcnt
- 1].iov_len
&&
446 may_free
== test_bit(f
->iovcnt
- 1, f
->may_free
))
448 f
->iov
[f
->iovcnt
- 1].iov_len
+= size
;
450 if (f
->iovcnt
>= MAX_IOV_SIZE
) {
451 /* Should only happen if a previous fflush failed */
452 assert(qemu_file_get_error(f
) || !qemu_file_is_writable(f
));
456 set_bit(f
->iovcnt
, f
->may_free
);
458 f
->iov
[f
->iovcnt
].iov_base
= (uint8_t *)buf
;
459 f
->iov
[f
->iovcnt
++].iov_len
= size
;
462 if (f
->iovcnt
>= MAX_IOV_SIZE
) {
470 static void add_buf_to_iovec(QEMUFile
*f
, size_t len
)
472 if (!add_to_iovec(f
, f
->buf
+ f
->buf_index
, len
, false)) {
474 if (f
->buf_index
== IO_BUF_SIZE
) {
480 void qemu_put_buffer_async(QEMUFile
*f
, const uint8_t *buf
, size_t size
,
487 add_to_iovec(f
, buf
, size
, may_free
);
490 void qemu_put_buffer(QEMUFile
*f
, const uint8_t *buf
, size_t size
)
499 l
= IO_BUF_SIZE
- f
->buf_index
;
503 memcpy(f
->buf
+ f
->buf_index
, buf
, l
);
504 add_buf_to_iovec(f
, l
);
505 if (qemu_file_get_error(f
)) {
513 void qemu_put_byte(QEMUFile
*f
, int v
)
519 f
->buf
[f
->buf_index
] = v
;
520 add_buf_to_iovec(f
, 1);
523 void qemu_file_skip(QEMUFile
*f
, int size
)
525 if (f
->buf_index
+ size
<= f
->buf_size
) {
526 f
->buf_index
+= size
;
531 * Read 'size' bytes from file (at 'offset') without moving the
532 * pointer and set 'buf' to point to that data.
534 * It will return size bytes unless there was an error, in which case it will
535 * return as many as it managed to read (assuming blocking fd's which
536 * all current QEMUFile are)
538 size_t coroutine_mixed_fn
qemu_peek_buffer(QEMUFile
*f
, uint8_t **buf
, size_t size
, size_t offset
)
543 assert(!qemu_file_is_writable(f
));
544 assert(offset
< IO_BUF_SIZE
);
545 assert(size
<= IO_BUF_SIZE
- offset
);
547 /* The 1st byte to read from */
548 index
= f
->buf_index
+ offset
;
549 /* The number of available bytes starting at index */
550 pending
= f
->buf_size
- index
;
553 * qemu_fill_buffer might return just a few bytes, even when there isn't
554 * an error, so loop collecting them until we get enough.
556 while (pending
< size
) {
557 int received
= qemu_fill_buffer(f
);
563 index
= f
->buf_index
+ offset
;
564 pending
= f
->buf_size
- index
;
570 if (size
> pending
) {
574 *buf
= f
->buf
+ index
;
579 * Read 'size' bytes of data from the file into buf.
580 * 'size' can be larger than the internal buffer.
582 * It will return size bytes unless there was an error, in which case it will
583 * return as many as it managed to read (assuming blocking fd's which
584 * all current QEMUFile are)
586 size_t coroutine_mixed_fn
qemu_get_buffer(QEMUFile
*f
, uint8_t *buf
, size_t size
)
588 size_t pending
= size
;
591 while (pending
> 0) {
595 res
= qemu_peek_buffer(f
, &src
, MIN(pending
, IO_BUF_SIZE
), 0);
599 memcpy(buf
, src
, res
);
600 qemu_file_skip(f
, res
);
609 * Read 'size' bytes of data from the file.
610 * 'size' can be larger than the internal buffer.
613 * may be held on an internal buffer (in which case *buf is updated
614 * to point to it) that is valid until the next qemu_file operation.
616 * will be copied to the *buf that was passed in.
618 * The code tries to avoid the copy if possible.
620 * It will return size bytes unless there was an error, in which case it will
621 * return as many as it managed to read (assuming blocking fd's which
622 * all current QEMUFile are)
624 * Note: Since **buf may get changed, the caller should take care to
625 * keep a pointer to the original buffer if it needs to deallocate it.
627 size_t coroutine_mixed_fn
qemu_get_buffer_in_place(QEMUFile
*f
, uint8_t **buf
, size_t size
)
629 if (size
< IO_BUF_SIZE
) {
633 res
= qemu_peek_buffer(f
, &src
, size
, 0);
636 qemu_file_skip(f
, res
);
642 return qemu_get_buffer(f
, *buf
, size
);
646 * Peeks a single byte from the buffer; this isn't guaranteed to work if
647 * offset leaves a gap after the previous read/peeked data.
649 int coroutine_mixed_fn
qemu_peek_byte(QEMUFile
*f
, int offset
)
651 int index
= f
->buf_index
+ offset
;
653 assert(!qemu_file_is_writable(f
));
654 assert(offset
< IO_BUF_SIZE
);
656 if (index
>= f
->buf_size
) {
658 index
= f
->buf_index
+ offset
;
659 if (index
>= f
->buf_size
) {
663 return f
->buf
[index
];
666 int coroutine_mixed_fn
qemu_get_byte(QEMUFile
*f
)
670 result
= qemu_peek_byte(f
, 0);
671 qemu_file_skip(f
, 1);
675 uint64_t qemu_file_transferred_noflush(QEMUFile
*f
)
677 uint64_t ret
= f
->total_transferred
;
680 for (i
= 0; i
< f
->iovcnt
; i
++) {
681 ret
+= f
->iov
[i
].iov_len
;
687 uint64_t qemu_file_transferred(QEMUFile
*f
)
690 return f
->total_transferred
;
693 void qemu_put_be16(QEMUFile
*f
, unsigned int v
)
695 qemu_put_byte(f
, v
>> 8);
699 void qemu_put_be32(QEMUFile
*f
, unsigned int v
)
701 qemu_put_byte(f
, v
>> 24);
702 qemu_put_byte(f
, v
>> 16);
703 qemu_put_byte(f
, v
>> 8);
707 void qemu_put_be64(QEMUFile
*f
, uint64_t v
)
709 qemu_put_be32(f
, v
>> 32);
713 unsigned int qemu_get_be16(QEMUFile
*f
)
716 v
= qemu_get_byte(f
) << 8;
717 v
|= qemu_get_byte(f
);
721 unsigned int qemu_get_be32(QEMUFile
*f
)
724 v
= (unsigned int)qemu_get_byte(f
) << 24;
725 v
|= qemu_get_byte(f
) << 16;
726 v
|= qemu_get_byte(f
) << 8;
727 v
|= qemu_get_byte(f
);
731 uint64_t qemu_get_be64(QEMUFile
*f
)
734 v
= (uint64_t)qemu_get_be32(f
) << 32;
735 v
|= qemu_get_be32(f
);
739 /* return the size after compression, or negative value on error */
740 static int qemu_compress_data(z_stream
*stream
, uint8_t *dest
, size_t dest_len
,
741 const uint8_t *source
, size_t source_len
)
745 err
= deflateReset(stream
);
750 stream
->avail_in
= source_len
;
751 stream
->next_in
= (uint8_t *)source
;
752 stream
->avail_out
= dest_len
;
753 stream
->next_out
= dest
;
755 err
= deflate(stream
, Z_FINISH
);
756 if (err
!= Z_STREAM_END
) {
760 return stream
->next_out
- dest
;
763 /* Compress size bytes of data start at p and store the compressed
764 * data to the buffer of f.
766 * Since the file is dummy file with empty_ops, return -1 if f has no space to
767 * save the compressed data.
769 ssize_t
qemu_put_compression_data(QEMUFile
*f
, z_stream
*stream
,
770 const uint8_t *p
, size_t size
)
772 ssize_t blen
= IO_BUF_SIZE
- f
->buf_index
- sizeof(int32_t);
774 if (blen
< compressBound(size
)) {
778 blen
= qemu_compress_data(stream
, f
->buf
+ f
->buf_index
+ sizeof(int32_t),
784 qemu_put_be32(f
, blen
);
785 add_buf_to_iovec(f
, blen
);
786 return blen
+ sizeof(int32_t);
789 /* Put the data in the buffer of f_src to the buffer of f_des, and
790 * then reset the buf_index of f_src to 0.
793 int qemu_put_qemu_file(QEMUFile
*f_des
, QEMUFile
*f_src
)
797 if (f_src
->buf_index
> 0) {
798 len
= f_src
->buf_index
;
799 qemu_put_buffer(f_des
, f_src
->buf
, f_src
->buf_index
);
800 f_src
->buf_index
= 0;
807 * Check if the writable buffer is empty
810 bool qemu_file_buffer_empty(QEMUFile
*file
)
812 assert(qemu_file_is_writable(file
));
814 return !file
->iovcnt
;
818 * Get a string whose length is determined by a single preceding byte
819 * A preallocated 256 byte buffer must be passed in.
820 * Returns: len on success and a 0 terminated string in the buffer
822 * (Note a 0 length string will return 0 either way)
824 size_t coroutine_fn
qemu_get_counted_string(QEMUFile
*f
, char buf
[256])
826 size_t len
= qemu_get_byte(f
);
827 size_t res
= qemu_get_buffer(f
, (uint8_t *)buf
, len
);
831 return res
== len
? res
: 0;
835 * Put a string with one preceding byte containing its length. The length of
836 * the string should be less than 256.
838 void qemu_put_counted_string(QEMUFile
*f
, const char *str
)
840 size_t len
= strlen(str
);
843 qemu_put_byte(f
, len
);
844 qemu_put_buffer(f
, (const uint8_t *)str
, len
);
848 * Set the blocking state of the QEMUFile.
849 * Note: On some transports the OS only keeps a single blocking state for
850 * both directions, and thus changing the blocking on the main
851 * QEMUFile can also affect the return path.
853 void qemu_file_set_blocking(QEMUFile
*f
, bool block
)
855 qio_channel_set_blocking(f
->ioc
, block
, NULL
);
861 * Get the ioc object for the file, without incrementing
862 * the reference count.
864 * Returns: the ioc object
866 QIOChannel
*qemu_file_get_ioc(QEMUFile
*file
)
872 * Read size bytes from QEMUFile f and write them to fd.
874 int qemu_file_get_to_fd(QEMUFile
*f
, int fd
, size_t size
)
877 size_t pending
= f
->buf_size
- f
->buf_index
;
881 rc
= qemu_fill_buffer(f
);
891 rc
= write(fd
, f
->buf
+ f
->buf_index
, MIN(pending
, size
));