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 #include "io/channel-file.h"
38 #define IO_BUF_SIZE 32768
39 #define MAX_IOV_SIZE MIN_CONST(IOV_MAX, 64)
46 int buf_size
; /* 0 when writing */
47 uint8_t buf
[IO_BUF_SIZE
];
49 DECLARE_BITMAP(may_free
, MAX_IOV_SIZE
);
50 struct iovec iov
[MAX_IOV_SIZE
];
54 Error
*last_error_obj
;
58 * Stop a file from being read/written - not all backing files can do this
59 * typically only sockets can.
61 * TODO: convert to propagate Error objects instead of squashing
62 * to a fixed errno value
64 int qemu_file_shutdown(QEMUFile
*f
)
67 * We must set qemufile error before the real shutdown(), otherwise
68 * there can be a race window where we thought IO all went though
69 * (because last_error==NULL) but actually IO has already stopped.
71 * If without correct ordering, the race can happen like this:
73 * page receiver other thread
74 * ------------- ------------
77 * returns 0 (buffer all zero)
78 * (we didn't check this retcode)
79 * try to detect IO error
80 * last_error==NULL, IO okay
81 * install ALL-ZERO page
86 qemu_file_set_error(f
, -EIO
);
89 if (!qio_channel_has_feature(f
->ioc
,
90 QIO_CHANNEL_FEATURE_SHUTDOWN
)) {
94 if (qio_channel_shutdown(f
->ioc
, QIO_CHANNEL_SHUTDOWN_BOTH
, NULL
) < 0) {
101 static QEMUFile
*qemu_file_new_impl(QIOChannel
*ioc
, bool is_writable
)
105 f
= g_new0(QEMUFile
, 1);
109 f
->is_writable
= is_writable
;
115 * Result: QEMUFile* for a 'return path' for comms in the opposite direction
116 * NULL if not available
118 QEMUFile
*qemu_file_get_return_path(QEMUFile
*f
)
120 return qemu_file_new_impl(f
->ioc
, !f
->is_writable
);
123 QEMUFile
*qemu_file_new_output(QIOChannel
*ioc
)
125 return qemu_file_new_impl(ioc
, true);
128 QEMUFile
*qemu_file_new_input(QIOChannel
*ioc
)
130 return qemu_file_new_impl(ioc
, false);
134 * Get last error for stream f with optional Error*
136 * Return negative error value if there has been an error on previous
137 * operations, return 0 if no error happened.
139 * If errp is specified, a verbose error message will be copied over.
141 int qemu_file_get_error_obj(QEMUFile
*f
, Error
**errp
)
143 if (!f
->last_error
) {
147 /* There is an error */
149 if (f
->last_error_obj
) {
150 *errp
= error_copy(f
->last_error_obj
);
152 error_setg_errno(errp
, -f
->last_error
, "Channel error");
156 return f
->last_error
;
160 * Get last error for either stream f1 or f2 with optional Error*.
161 * The error returned (non-zero) can be either from f1 or f2.
163 * If any of the qemufile* is NULL, then skip the check on that file.
165 * When there is no error on both qemufile, zero is returned.
167 int qemu_file_get_error_obj_any(QEMUFile
*f1
, QEMUFile
*f2
, Error
**errp
)
172 ret
= qemu_file_get_error_obj(f1
, errp
);
173 /* If there's already error detected, return */
180 ret
= qemu_file_get_error_obj(f2
, errp
);
187 * Set the last error for stream f with optional Error*
189 void qemu_file_set_error_obj(QEMUFile
*f
, int ret
, Error
*err
)
191 if (f
->last_error
== 0 && ret
) {
193 error_propagate(&f
->last_error_obj
, err
);
195 error_report_err(err
);
200 * Get last error for stream f
202 * Return negative error value if there has been an error on previous
203 * operations, return 0 if no error happened.
206 int qemu_file_get_error(QEMUFile
*f
)
208 return f
->last_error
;
212 * Set the last error for stream f
214 void qemu_file_set_error(QEMUFile
*f
, int ret
)
216 qemu_file_set_error_obj(f
, ret
, NULL
);
219 static bool qemu_file_is_writable(QEMUFile
*f
)
221 return f
->is_writable
;
224 static void qemu_iovec_release_ram(QEMUFile
*f
)
229 /* Find and release all the contiguous memory ranges marked as may_free. */
230 idx
= find_next_bit(f
->may_free
, f
->iovcnt
, 0);
231 if (idx
>= f
->iovcnt
) {
236 /* The madvise() in the loop is called for iov within a continuous range and
237 * then reinitialize the iov. And in the end, madvise() is called for the
240 while ((idx
= find_next_bit(f
->may_free
, f
->iovcnt
, idx
+ 1)) < f
->iovcnt
) {
241 /* check for adjacent buffer and coalesce them */
242 if (iov
.iov_base
+ iov
.iov_len
== f
->iov
[idx
].iov_base
) {
243 iov
.iov_len
+= f
->iov
[idx
].iov_len
;
246 if (qemu_madvise(iov
.iov_base
, iov
.iov_len
, QEMU_MADV_DONTNEED
) < 0) {
247 error_report("migrate: madvise DONTNEED failed %p %zd: %s",
248 iov
.iov_base
, iov
.iov_len
, strerror(errno
));
252 if (qemu_madvise(iov
.iov_base
, iov
.iov_len
, QEMU_MADV_DONTNEED
) < 0) {
253 error_report("migrate: madvise DONTNEED failed %p %zd: %s",
254 iov
.iov_base
, iov
.iov_len
, strerror(errno
));
256 memset(f
->may_free
, 0, sizeof(f
->may_free
));
259 bool qemu_file_is_seekable(QEMUFile
*f
)
261 return qio_channel_has_feature(f
->ioc
, QIO_CHANNEL_FEATURE_SEEKABLE
);
265 * Flushes QEMUFile buffer
267 * This will flush all pending data. If data was only partially flushed, it
268 * will set an error state.
270 int qemu_fflush(QEMUFile
*f
)
272 if (!qemu_file_is_writable(f
)) {
273 return f
->last_error
;
277 return f
->last_error
;
280 Error
*local_error
= NULL
;
281 if (qio_channel_writev_all(f
->ioc
,
284 qemu_file_set_error_obj(f
, -EIO
, local_error
);
286 uint64_t size
= iov_size(f
->iov
, f
->iovcnt
);
287 stat64_add(&mig_stats
.qemu_file_transferred
, size
);
290 qemu_iovec_release_ram(f
);
295 return f
->last_error
;
299 * Attempt to fill the buffer from the underlying file
300 * Returns the number of bytes read, or negative value for an error.
302 * Note that it can return a partially full buffer even in a not error/not EOF
303 * case if the underlying file descriptor gives a short read, and that can
304 * happen even on a blocking fd.
306 static ssize_t coroutine_mixed_fn
qemu_fill_buffer(QEMUFile
*f
)
310 Error
*local_error
= NULL
;
312 assert(!qemu_file_is_writable(f
));
314 pending
= f
->buf_size
- f
->buf_index
;
316 memmove(f
->buf
, f
->buf
+ f
->buf_index
, pending
);
319 f
->buf_size
= pending
;
321 if (qemu_file_get_error(f
)) {
326 len
= qio_channel_read(f
->ioc
,
327 (char *)f
->buf
+ pending
,
328 IO_BUF_SIZE
- pending
,
330 if (len
== QIO_CHANNEL_ERR_BLOCK
) {
331 if (qemu_in_coroutine()) {
332 qio_channel_yield(f
->ioc
, G_IO_IN
);
334 qio_channel_wait(f
->ioc
, G_IO_IN
);
336 } else if (len
< 0) {
339 } while (len
== QIO_CHANNEL_ERR_BLOCK
);
343 } else if (len
== 0) {
344 qemu_file_set_error_obj(f
, -EIO
, local_error
);
346 qemu_file_set_error_obj(f
, len
, local_error
);
354 * Returns negative error value if any error happened on previous operations or
355 * while closing the file. Returns 0 or positive number on success.
357 * The meaning of return value on success depends on the specific backend
360 int qemu_fclose(QEMUFile
*f
)
362 int ret
= qemu_fflush(f
);
363 int ret2
= qio_channel_close(f
->ioc
, NULL
);
367 g_clear_pointer(&f
->ioc
, object_unref
);
368 error_free(f
->last_error_obj
);
370 trace_qemu_file_fclose();
375 * Add buf to iovec. Do flush if iovec is full.
378 * 1 iovec is full and flushed
379 * 0 iovec is not flushed
382 static int add_to_iovec(QEMUFile
*f
, const uint8_t *buf
, size_t size
,
385 /* check for adjacent buffer and coalesce them */
386 if (f
->iovcnt
> 0 && buf
== f
->iov
[f
->iovcnt
- 1].iov_base
+
387 f
->iov
[f
->iovcnt
- 1].iov_len
&&
388 may_free
== test_bit(f
->iovcnt
- 1, f
->may_free
))
390 f
->iov
[f
->iovcnt
- 1].iov_len
+= size
;
392 if (f
->iovcnt
>= MAX_IOV_SIZE
) {
393 /* Should only happen if a previous fflush failed */
394 assert(qemu_file_get_error(f
) || !qemu_file_is_writable(f
));
398 set_bit(f
->iovcnt
, f
->may_free
);
400 f
->iov
[f
->iovcnt
].iov_base
= (uint8_t *)buf
;
401 f
->iov
[f
->iovcnt
++].iov_len
= size
;
404 if (f
->iovcnt
>= MAX_IOV_SIZE
) {
412 static void add_buf_to_iovec(QEMUFile
*f
, size_t len
)
414 if (!add_to_iovec(f
, f
->buf
+ f
->buf_index
, len
, false)) {
416 if (f
->buf_index
== IO_BUF_SIZE
) {
422 void qemu_put_buffer_async(QEMUFile
*f
, const uint8_t *buf
, size_t size
,
429 add_to_iovec(f
, buf
, size
, may_free
);
432 void qemu_put_buffer(QEMUFile
*f
, const uint8_t *buf
, size_t size
)
441 l
= IO_BUF_SIZE
- f
->buf_index
;
445 memcpy(f
->buf
+ f
->buf_index
, buf
, l
);
446 add_buf_to_iovec(f
, l
);
447 if (qemu_file_get_error(f
)) {
455 void qemu_put_buffer_at(QEMUFile
*f
, const uint8_t *buf
, size_t buflen
,
466 ret
= qio_channel_pwrite(f
->ioc
, (char *)buf
, buflen
, pos
, &err
);
469 qemu_file_set_error_obj(f
, -EIO
, err
);
473 if ((ssize_t
)ret
== QIO_CHANNEL_ERR_BLOCK
) {
474 qemu_file_set_error_obj(f
, -EAGAIN
, NULL
);
479 error_setg(&err
, "Partial write of size %zu, expected %zu", ret
,
481 qemu_file_set_error_obj(f
, -EIO
, err
);
485 stat64_add(&mig_stats
.qemu_file_transferred
, buflen
);
491 size_t qemu_get_buffer_at(QEMUFile
*f
, const uint8_t *buf
, size_t buflen
,
501 ret
= qio_channel_pread(f
->ioc
, (char *)buf
, buflen
, pos
, &err
);
503 if ((ssize_t
)ret
== -1 || err
) {
504 qemu_file_set_error_obj(f
, -EIO
, err
);
508 if ((ssize_t
)ret
== QIO_CHANNEL_ERR_BLOCK
) {
509 qemu_file_set_error_obj(f
, -EAGAIN
, NULL
);
514 error_setg(&err
, "Partial read of size %zu, expected %zu", ret
, buflen
);
515 qemu_file_set_error_obj(f
, -EIO
, err
);
522 void qemu_set_offset(QEMUFile
*f
, off_t off
, int whence
)
527 if (qemu_file_is_writable(f
)) {
530 /* Drop all cached buffers if existed; will trigger a re-fill later */
535 ret
= qio_channel_io_seek(f
->ioc
, off
, whence
, &err
);
536 if (ret
== (off_t
)-1) {
537 qemu_file_set_error_obj(f
, -EIO
, err
);
541 off_t
qemu_get_offset(QEMUFile
*f
)
548 ret
= qio_channel_io_seek(f
->ioc
, 0, SEEK_CUR
, &err
);
549 if (ret
== (off_t
)-1) {
550 qemu_file_set_error_obj(f
, -EIO
, err
);
556 void qemu_put_byte(QEMUFile
*f
, int v
)
562 f
->buf
[f
->buf_index
] = v
;
563 add_buf_to_iovec(f
, 1);
566 void qemu_file_skip(QEMUFile
*f
, int size
)
568 if (f
->buf_index
+ size
<= f
->buf_size
) {
569 f
->buf_index
+= size
;
574 * Read 'size' bytes from file (at 'offset') without moving the
575 * pointer and set 'buf' to point to that data.
577 * It will return size bytes unless there was an error, in which case it will
578 * return as many as it managed to read (assuming blocking fd's which
579 * all current QEMUFile are)
581 size_t coroutine_mixed_fn
qemu_peek_buffer(QEMUFile
*f
, uint8_t **buf
, size_t size
, size_t offset
)
586 assert(!qemu_file_is_writable(f
));
587 assert(offset
< IO_BUF_SIZE
);
588 assert(size
<= IO_BUF_SIZE
- offset
);
590 /* The 1st byte to read from */
591 index
= f
->buf_index
+ offset
;
592 /* The number of available bytes starting at index */
593 pending
= f
->buf_size
- index
;
596 * qemu_fill_buffer might return just a few bytes, even when there isn't
597 * an error, so loop collecting them until we get enough.
599 while (pending
< size
) {
600 int received
= qemu_fill_buffer(f
);
606 index
= f
->buf_index
+ offset
;
607 pending
= f
->buf_size
- index
;
613 if (size
> pending
) {
617 *buf
= f
->buf
+ index
;
622 * Read 'size' bytes of data from the file into buf.
623 * 'size' can be larger than the internal buffer.
625 * It will return size bytes unless there was an error, in which case it will
626 * return as many as it managed to read (assuming blocking fd's which
627 * all current QEMUFile are)
629 size_t coroutine_mixed_fn
qemu_get_buffer(QEMUFile
*f
, uint8_t *buf
, size_t size
)
631 size_t pending
= size
;
634 while (pending
> 0) {
638 res
= qemu_peek_buffer(f
, &src
, MIN(pending
, IO_BUF_SIZE
), 0);
642 memcpy(buf
, src
, res
);
643 qemu_file_skip(f
, res
);
652 * Read 'size' bytes of data from the file.
653 * 'size' can be larger than the internal buffer.
656 * may be held on an internal buffer (in which case *buf is updated
657 * to point to it) that is valid until the next qemu_file operation.
659 * will be copied to the *buf that was passed in.
661 * The code tries to avoid the copy if possible.
663 * It will return size bytes unless there was an error, in which case it will
664 * return as many as it managed to read (assuming blocking fd's which
665 * all current QEMUFile are)
667 * Note: Since **buf may get changed, the caller should take care to
668 * keep a pointer to the original buffer if it needs to deallocate it.
670 size_t coroutine_mixed_fn
qemu_get_buffer_in_place(QEMUFile
*f
, uint8_t **buf
, size_t size
)
672 if (size
< IO_BUF_SIZE
) {
676 res
= qemu_peek_buffer(f
, &src
, size
, 0);
679 qemu_file_skip(f
, res
);
685 return qemu_get_buffer(f
, *buf
, size
);
689 * Peeks a single byte from the buffer; this isn't guaranteed to work if
690 * offset leaves a gap after the previous read/peeked data.
692 int coroutine_mixed_fn
qemu_peek_byte(QEMUFile
*f
, int offset
)
694 int index
= f
->buf_index
+ offset
;
696 assert(!qemu_file_is_writable(f
));
697 assert(offset
< IO_BUF_SIZE
);
699 if (index
>= f
->buf_size
) {
701 index
= f
->buf_index
+ offset
;
702 if (index
>= f
->buf_size
) {
706 return f
->buf
[index
];
709 int coroutine_mixed_fn
qemu_get_byte(QEMUFile
*f
)
713 result
= qemu_peek_byte(f
, 0);
714 qemu_file_skip(f
, 1);
718 uint64_t qemu_file_transferred(QEMUFile
*f
)
720 uint64_t ret
= stat64_get(&mig_stats
.qemu_file_transferred
);
723 g_assert(qemu_file_is_writable(f
));
725 for (i
= 0; i
< f
->iovcnt
; i
++) {
726 ret
+= f
->iov
[i
].iov_len
;
732 void qemu_put_be16(QEMUFile
*f
, unsigned int v
)
734 qemu_put_byte(f
, v
>> 8);
738 void qemu_put_be32(QEMUFile
*f
, unsigned int v
)
740 qemu_put_byte(f
, v
>> 24);
741 qemu_put_byte(f
, v
>> 16);
742 qemu_put_byte(f
, v
>> 8);
746 void qemu_put_be64(QEMUFile
*f
, uint64_t v
)
748 qemu_put_be32(f
, v
>> 32);
752 unsigned int qemu_get_be16(QEMUFile
*f
)
755 v
= qemu_get_byte(f
) << 8;
756 v
|= qemu_get_byte(f
);
760 unsigned int qemu_get_be32(QEMUFile
*f
)
763 v
= (unsigned int)qemu_get_byte(f
) << 24;
764 v
|= qemu_get_byte(f
) << 16;
765 v
|= qemu_get_byte(f
) << 8;
766 v
|= qemu_get_byte(f
);
770 uint64_t qemu_get_be64(QEMUFile
*f
)
773 v
= (uint64_t)qemu_get_be32(f
) << 32;
774 v
|= qemu_get_be32(f
);
778 /* return the size after compression, or negative value on error */
779 static int qemu_compress_data(z_stream
*stream
, uint8_t *dest
, size_t dest_len
,
780 const uint8_t *source
, size_t source_len
)
784 err
= deflateReset(stream
);
789 stream
->avail_in
= source_len
;
790 stream
->next_in
= (uint8_t *)source
;
791 stream
->avail_out
= dest_len
;
792 stream
->next_out
= dest
;
794 err
= deflate(stream
, Z_FINISH
);
795 if (err
!= Z_STREAM_END
) {
799 return stream
->next_out
- dest
;
802 /* Compress size bytes of data start at p and store the compressed
803 * data to the buffer of f.
805 * Since the file is dummy file with empty_ops, return -1 if f has no space to
806 * save the compressed data.
808 ssize_t
qemu_put_compression_data(QEMUFile
*f
, z_stream
*stream
,
809 const uint8_t *p
, size_t size
)
811 ssize_t blen
= IO_BUF_SIZE
- f
->buf_index
- sizeof(int32_t);
813 if (blen
< compressBound(size
)) {
817 blen
= qemu_compress_data(stream
, f
->buf
+ f
->buf_index
+ sizeof(int32_t),
823 qemu_put_be32(f
, blen
);
824 add_buf_to_iovec(f
, blen
);
825 return blen
+ sizeof(int32_t);
828 /* Put the data in the buffer of f_src to the buffer of f_des, and
829 * then reset the buf_index of f_src to 0.
832 int qemu_put_qemu_file(QEMUFile
*f_des
, QEMUFile
*f_src
)
836 if (f_src
->buf_index
> 0) {
837 len
= f_src
->buf_index
;
838 qemu_put_buffer(f_des
, f_src
->buf
, f_src
->buf_index
);
839 f_src
->buf_index
= 0;
846 * Check if the writable buffer is empty
849 bool qemu_file_buffer_empty(QEMUFile
*file
)
851 assert(qemu_file_is_writable(file
));
853 return !file
->iovcnt
;
857 * Get a string whose length is determined by a single preceding byte
858 * A preallocated 256 byte buffer must be passed in.
859 * Returns: len on success and a 0 terminated string in the buffer
861 * (Note a 0 length string will return 0 either way)
863 size_t coroutine_fn
qemu_get_counted_string(QEMUFile
*f
, char buf
[256])
865 size_t len
= qemu_get_byte(f
);
866 size_t res
= qemu_get_buffer(f
, (uint8_t *)buf
, len
);
870 return res
== len
? res
: 0;
874 * Put a string with one preceding byte containing its length. The length of
875 * the string should be less than 256.
877 void qemu_put_counted_string(QEMUFile
*f
, const char *str
)
879 size_t len
= strlen(str
);
882 qemu_put_byte(f
, len
);
883 qemu_put_buffer(f
, (const uint8_t *)str
, len
);
887 * Set the blocking state of the QEMUFile.
888 * Note: On some transports the OS only keeps a single blocking state for
889 * both directions, and thus changing the blocking on the main
890 * QEMUFile can also affect the return path.
892 void qemu_file_set_blocking(QEMUFile
*f
, bool block
)
894 qio_channel_set_blocking(f
->ioc
, block
, NULL
);
900 * Get the ioc object for the file, without incrementing
901 * the reference count.
903 * Returns: the ioc object
905 QIOChannel
*qemu_file_get_ioc(QEMUFile
*file
)
911 * Read size bytes from QEMUFile f and write them to fd.
913 int qemu_file_get_to_fd(QEMUFile
*f
, int fd
, size_t size
)
916 size_t pending
= f
->buf_size
- f
->buf_index
;
920 rc
= qemu_fill_buffer(f
);
930 rc
= write(fd
, f
->buf
+ f
->buf_index
, MIN(pending
, size
));