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"
37 #define IO_BUF_SIZE 32768
38 #define MAX_IOV_SIZE MIN_CONST(IOV_MAX, 64)
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);
136 * Get last error for stream f with optional Error*
138 * Return negative error value if there has been an error on previous
139 * operations, return 0 if no error happened.
141 * If errp is specified, a verbose error message will be copied over.
143 static int qemu_file_get_error_obj(QEMUFile
*f
, Error
**errp
)
145 if (!f
->last_error
) {
149 /* There is an error */
151 if (f
->last_error_obj
) {
152 *errp
= error_copy(f
->last_error_obj
);
154 error_setg_errno(errp
, -f
->last_error
, "Channel error");
158 return f
->last_error
;
162 * Get last error for either stream f1 or f2 with optional Error*.
163 * The error returned (non-zero) can be either from f1 or f2.
165 * If any of the qemufile* is NULL, then skip the check on that file.
167 * When there is no error on both qemufile, zero is returned.
169 int qemu_file_get_error_obj_any(QEMUFile
*f1
, QEMUFile
*f2
, Error
**errp
)
174 ret
= qemu_file_get_error_obj(f1
, errp
);
175 /* If there's already error detected, return */
182 ret
= qemu_file_get_error_obj(f2
, errp
);
189 * Set the last error for stream f with optional Error*
191 void qemu_file_set_error_obj(QEMUFile
*f
, int ret
, Error
*err
)
193 if (f
->last_error
== 0 && ret
) {
195 error_propagate(&f
->last_error_obj
, err
);
197 error_report_err(err
);
202 * Get last error for stream f
204 * Return negative error value if there has been an error on previous
205 * operations, return 0 if no error happened.
208 int qemu_file_get_error(QEMUFile
*f
)
210 return qemu_file_get_error_obj(f
, NULL
);
214 * Set the last error for stream f
216 void qemu_file_set_error(QEMUFile
*f
, int ret
)
218 qemu_file_set_error_obj(f
, ret
, NULL
);
221 static bool qemu_file_is_writable(QEMUFile
*f
)
223 return f
->is_writable
;
226 static void qemu_iovec_release_ram(QEMUFile
*f
)
231 /* Find and release all the contiguous memory ranges marked as may_free. */
232 idx
= find_next_bit(f
->may_free
, f
->iovcnt
, 0);
233 if (idx
>= f
->iovcnt
) {
238 /* The madvise() in the loop is called for iov within a continuous range and
239 * then reinitialize the iov. And in the end, madvise() is called for the
242 while ((idx
= find_next_bit(f
->may_free
, f
->iovcnt
, idx
+ 1)) < f
->iovcnt
) {
243 /* check for adjacent buffer and coalesce them */
244 if (iov
.iov_base
+ iov
.iov_len
== f
->iov
[idx
].iov_base
) {
245 iov
.iov_len
+= f
->iov
[idx
].iov_len
;
248 if (qemu_madvise(iov
.iov_base
, iov
.iov_len
, QEMU_MADV_DONTNEED
) < 0) {
249 error_report("migrate: madvise DONTNEED failed %p %zd: %s",
250 iov
.iov_base
, iov
.iov_len
, strerror(errno
));
254 if (qemu_madvise(iov
.iov_base
, iov
.iov_len
, QEMU_MADV_DONTNEED
) < 0) {
255 error_report("migrate: madvise DONTNEED failed %p %zd: %s",
256 iov
.iov_base
, iov
.iov_len
, strerror(errno
));
258 memset(f
->may_free
, 0, sizeof(f
->may_free
));
263 * Flushes QEMUFile buffer
265 * This will flush all pending data. If data was only partially flushed, it
266 * will set an error state.
268 void qemu_fflush(QEMUFile
*f
)
270 if (!qemu_file_is_writable(f
)) {
274 if (qemu_file_get_error(f
)) {
278 Error
*local_error
= NULL
;
279 if (qio_channel_writev_all(f
->ioc
,
282 qemu_file_set_error_obj(f
, -EIO
, local_error
);
284 uint64_t size
= iov_size(f
->iov
, f
->iovcnt
);
285 f
->total_transferred
+= size
;
288 qemu_iovec_release_ram(f
);
296 * Attempt to fill the buffer from the underlying file
297 * Returns the number of bytes read, or negative value for an error.
299 * Note that it can return a partially full buffer even in a not error/not EOF
300 * case if the underlying file descriptor gives a short read, and that can
301 * happen even on a blocking fd.
303 static ssize_t coroutine_mixed_fn
qemu_fill_buffer(QEMUFile
*f
)
307 Error
*local_error
= NULL
;
309 assert(!qemu_file_is_writable(f
));
311 pending
= f
->buf_size
- f
->buf_index
;
313 memmove(f
->buf
, f
->buf
+ f
->buf_index
, pending
);
316 f
->buf_size
= pending
;
318 if (qemu_file_get_error(f
)) {
323 len
= qio_channel_read(f
->ioc
,
324 (char *)f
->buf
+ pending
,
325 IO_BUF_SIZE
- pending
,
327 if (len
== QIO_CHANNEL_ERR_BLOCK
) {
328 if (qemu_in_coroutine()) {
329 qio_channel_yield(f
->ioc
, G_IO_IN
);
331 qio_channel_wait(f
->ioc
, G_IO_IN
);
333 } else if (len
< 0) {
336 } while (len
== QIO_CHANNEL_ERR_BLOCK
);
340 f
->total_transferred
+= len
;
341 } else if (len
== 0) {
342 qemu_file_set_error_obj(f
, -EIO
, local_error
);
344 qemu_file_set_error_obj(f
, len
, local_error
);
352 * Returns negative error value if any error happened on previous operations or
353 * while closing the file. Returns 0 or positive number on success.
355 * The meaning of return value on success depends on the specific backend
358 int qemu_fclose(QEMUFile
*f
)
362 ret
= qemu_file_get_error(f
);
364 ret2
= qio_channel_close(f
->ioc
, NULL
);
368 g_clear_pointer(&f
->ioc
, object_unref
);
370 /* If any error was spotted before closing, we should report it
371 * instead of the close() return value.
376 error_free(f
->last_error_obj
);
378 trace_qemu_file_fclose();
383 * Add buf to iovec. Do flush if iovec is full.
386 * 1 iovec is full and flushed
387 * 0 iovec is not flushed
390 static int add_to_iovec(QEMUFile
*f
, const uint8_t *buf
, size_t size
,
393 /* check for adjacent buffer and coalesce them */
394 if (f
->iovcnt
> 0 && buf
== f
->iov
[f
->iovcnt
- 1].iov_base
+
395 f
->iov
[f
->iovcnt
- 1].iov_len
&&
396 may_free
== test_bit(f
->iovcnt
- 1, f
->may_free
))
398 f
->iov
[f
->iovcnt
- 1].iov_len
+= size
;
400 if (f
->iovcnt
>= MAX_IOV_SIZE
) {
401 /* Should only happen if a previous fflush failed */
402 assert(qemu_file_get_error(f
) || !qemu_file_is_writable(f
));
406 set_bit(f
->iovcnt
, f
->may_free
);
408 f
->iov
[f
->iovcnt
].iov_base
= (uint8_t *)buf
;
409 f
->iov
[f
->iovcnt
++].iov_len
= size
;
412 if (f
->iovcnt
>= MAX_IOV_SIZE
) {
420 static void add_buf_to_iovec(QEMUFile
*f
, size_t len
)
422 if (!add_to_iovec(f
, f
->buf
+ f
->buf_index
, len
, false)) {
424 if (f
->buf_index
== IO_BUF_SIZE
) {
430 void qemu_put_buffer_async(QEMUFile
*f
, const uint8_t *buf
, size_t size
,
437 add_to_iovec(f
, buf
, size
, may_free
);
440 void qemu_put_buffer(QEMUFile
*f
, const uint8_t *buf
, size_t size
)
449 l
= IO_BUF_SIZE
- f
->buf_index
;
453 memcpy(f
->buf
+ f
->buf_index
, buf
, l
);
454 add_buf_to_iovec(f
, l
);
455 if (qemu_file_get_error(f
)) {
463 void qemu_put_byte(QEMUFile
*f
, int v
)
469 f
->buf
[f
->buf_index
] = v
;
470 add_buf_to_iovec(f
, 1);
473 void qemu_file_skip(QEMUFile
*f
, int size
)
475 if (f
->buf_index
+ size
<= f
->buf_size
) {
476 f
->buf_index
+= size
;
481 * Read 'size' bytes from file (at 'offset') without moving the
482 * pointer and set 'buf' to point to that data.
484 * It will return size bytes unless there was an error, in which case it will
485 * return as many as it managed to read (assuming blocking fd's which
486 * all current QEMUFile are)
488 size_t coroutine_mixed_fn
qemu_peek_buffer(QEMUFile
*f
, uint8_t **buf
, size_t size
, size_t offset
)
493 assert(!qemu_file_is_writable(f
));
494 assert(offset
< IO_BUF_SIZE
);
495 assert(size
<= IO_BUF_SIZE
- offset
);
497 /* The 1st byte to read from */
498 index
= f
->buf_index
+ offset
;
499 /* The number of available bytes starting at index */
500 pending
= f
->buf_size
- index
;
503 * qemu_fill_buffer might return just a few bytes, even when there isn't
504 * an error, so loop collecting them until we get enough.
506 while (pending
< size
) {
507 int received
= qemu_fill_buffer(f
);
513 index
= f
->buf_index
+ offset
;
514 pending
= f
->buf_size
- index
;
520 if (size
> pending
) {
524 *buf
= f
->buf
+ index
;
529 * Read 'size' bytes of data from the file into buf.
530 * 'size' can be larger than the internal buffer.
532 * It will return size bytes unless there was an error, in which case it will
533 * return as many as it managed to read (assuming blocking fd's which
534 * all current QEMUFile are)
536 size_t coroutine_mixed_fn
qemu_get_buffer(QEMUFile
*f
, uint8_t *buf
, size_t size
)
538 size_t pending
= size
;
541 while (pending
> 0) {
545 res
= qemu_peek_buffer(f
, &src
, MIN(pending
, IO_BUF_SIZE
), 0);
549 memcpy(buf
, src
, res
);
550 qemu_file_skip(f
, res
);
559 * Read 'size' bytes of data from the file.
560 * 'size' can be larger than the internal buffer.
563 * may be held on an internal buffer (in which case *buf is updated
564 * to point to it) that is valid until the next qemu_file operation.
566 * will be copied to the *buf that was passed in.
568 * The code tries to avoid the copy if possible.
570 * It will return size bytes unless there was an error, in which case it will
571 * return as many as it managed to read (assuming blocking fd's which
572 * all current QEMUFile are)
574 * Note: Since **buf may get changed, the caller should take care to
575 * keep a pointer to the original buffer if it needs to deallocate it.
577 size_t coroutine_mixed_fn
qemu_get_buffer_in_place(QEMUFile
*f
, uint8_t **buf
, size_t size
)
579 if (size
< IO_BUF_SIZE
) {
583 res
= qemu_peek_buffer(f
, &src
, size
, 0);
586 qemu_file_skip(f
, res
);
592 return qemu_get_buffer(f
, *buf
, size
);
596 * Peeks a single byte from the buffer; this isn't guaranteed to work if
597 * offset leaves a gap after the previous read/peeked data.
599 int coroutine_mixed_fn
qemu_peek_byte(QEMUFile
*f
, int offset
)
601 int index
= f
->buf_index
+ offset
;
603 assert(!qemu_file_is_writable(f
));
604 assert(offset
< IO_BUF_SIZE
);
606 if (index
>= f
->buf_size
) {
608 index
= f
->buf_index
+ offset
;
609 if (index
>= f
->buf_size
) {
613 return f
->buf
[index
];
616 int coroutine_mixed_fn
qemu_get_byte(QEMUFile
*f
)
620 result
= qemu_peek_byte(f
, 0);
621 qemu_file_skip(f
, 1);
625 uint64_t qemu_file_transferred_noflush(QEMUFile
*f
)
627 uint64_t ret
= f
->total_transferred
;
630 for (i
= 0; i
< f
->iovcnt
; i
++) {
631 ret
+= f
->iov
[i
].iov_len
;
637 uint64_t qemu_file_transferred(QEMUFile
*f
)
640 return f
->total_transferred
;
643 void qemu_put_be16(QEMUFile
*f
, unsigned int v
)
645 qemu_put_byte(f
, v
>> 8);
649 void qemu_put_be32(QEMUFile
*f
, unsigned int v
)
651 qemu_put_byte(f
, v
>> 24);
652 qemu_put_byte(f
, v
>> 16);
653 qemu_put_byte(f
, v
>> 8);
657 void qemu_put_be64(QEMUFile
*f
, uint64_t v
)
659 qemu_put_be32(f
, v
>> 32);
663 unsigned int qemu_get_be16(QEMUFile
*f
)
666 v
= qemu_get_byte(f
) << 8;
667 v
|= qemu_get_byte(f
);
671 unsigned int qemu_get_be32(QEMUFile
*f
)
674 v
= (unsigned int)qemu_get_byte(f
) << 24;
675 v
|= qemu_get_byte(f
) << 16;
676 v
|= qemu_get_byte(f
) << 8;
677 v
|= qemu_get_byte(f
);
681 uint64_t qemu_get_be64(QEMUFile
*f
)
684 v
= (uint64_t)qemu_get_be32(f
) << 32;
685 v
|= qemu_get_be32(f
);
689 /* return the size after compression, or negative value on error */
690 static int qemu_compress_data(z_stream
*stream
, uint8_t *dest
, size_t dest_len
,
691 const uint8_t *source
, size_t source_len
)
695 err
= deflateReset(stream
);
700 stream
->avail_in
= source_len
;
701 stream
->next_in
= (uint8_t *)source
;
702 stream
->avail_out
= dest_len
;
703 stream
->next_out
= dest
;
705 err
= deflate(stream
, Z_FINISH
);
706 if (err
!= Z_STREAM_END
) {
710 return stream
->next_out
- dest
;
713 /* Compress size bytes of data start at p and store the compressed
714 * data to the buffer of f.
716 * Since the file is dummy file with empty_ops, return -1 if f has no space to
717 * save the compressed data.
719 ssize_t
qemu_put_compression_data(QEMUFile
*f
, z_stream
*stream
,
720 const uint8_t *p
, size_t size
)
722 ssize_t blen
= IO_BUF_SIZE
- f
->buf_index
- sizeof(int32_t);
724 if (blen
< compressBound(size
)) {
728 blen
= qemu_compress_data(stream
, f
->buf
+ f
->buf_index
+ sizeof(int32_t),
734 qemu_put_be32(f
, blen
);
735 add_buf_to_iovec(f
, blen
);
736 return blen
+ sizeof(int32_t);
739 /* Put the data in the buffer of f_src to the buffer of f_des, and
740 * then reset the buf_index of f_src to 0.
743 int qemu_put_qemu_file(QEMUFile
*f_des
, QEMUFile
*f_src
)
747 if (f_src
->buf_index
> 0) {
748 len
= f_src
->buf_index
;
749 qemu_put_buffer(f_des
, f_src
->buf
, f_src
->buf_index
);
750 f_src
->buf_index
= 0;
757 * Check if the writable buffer is empty
760 bool qemu_file_buffer_empty(QEMUFile
*file
)
762 assert(qemu_file_is_writable(file
));
764 return !file
->iovcnt
;
768 * Get a string whose length is determined by a single preceding byte
769 * A preallocated 256 byte buffer must be passed in.
770 * Returns: len on success and a 0 terminated string in the buffer
772 * (Note a 0 length string will return 0 either way)
774 size_t coroutine_fn
qemu_get_counted_string(QEMUFile
*f
, char buf
[256])
776 size_t len
= qemu_get_byte(f
);
777 size_t res
= qemu_get_buffer(f
, (uint8_t *)buf
, len
);
781 return res
== len
? res
: 0;
785 * Put a string with one preceding byte containing its length. The length of
786 * the string should be less than 256.
788 void qemu_put_counted_string(QEMUFile
*f
, const char *str
)
790 size_t len
= strlen(str
);
793 qemu_put_byte(f
, len
);
794 qemu_put_buffer(f
, (const uint8_t *)str
, len
);
798 * Set the blocking state of the QEMUFile.
799 * Note: On some transports the OS only keeps a single blocking state for
800 * both directions, and thus changing the blocking on the main
801 * QEMUFile can also affect the return path.
803 void qemu_file_set_blocking(QEMUFile
*f
, bool block
)
805 qio_channel_set_blocking(f
->ioc
, block
, NULL
);
811 * Get the ioc object for the file, without incrementing
812 * the reference count.
814 * Returns: the ioc object
816 QIOChannel
*qemu_file_get_ioc(QEMUFile
*file
)
822 * Read size bytes from QEMUFile f and write them to fd.
824 int qemu_file_get_to_fd(QEMUFile
*f
, int fd
, size_t size
)
827 size_t pending
= f
->buf_size
- f
->buf_index
;
831 rc
= qemu_fill_buffer(f
);
841 rc
= write(fd
, f
->buf
+ f
->buf_index
, MIN(pending
, size
));