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 "qemu-file.h"
32 #include "qapi/error.h"
34 #define IO_BUF_SIZE 32768
35 #define MAX_IOV_SIZE MIN_CONST(IOV_MAX, 64)
38 const QEMUFileHooks *hooks;
43 * Maximum amount of data in bytes to transfer during one
44 * rate limiting time window
46 int64_t rate_limit_max;
48 * Total amount of data in bytes queued for transfer
49 * during this rate limiting time window
51 int64_t rate_limit_used;
53 /* The sum of bytes transferred on the wire */
54 uint64_t total_transferred;
57 int buf_size; /* 0 when writing */
58 uint8_t buf[IO_BUF_SIZE];
60 DECLARE_BITMAP(may_free, MAX_IOV_SIZE);
61 struct iovec iov[MAX_IOV_SIZE];
65 Error *last_error_obj;
69 * Stop a file from being read/written - not all backing files can do this
70 * typically only sockets can.
72 * TODO: convert to propagate Error objects instead of squashing
73 * to a fixed errno value
75 int qemu_file_shutdown(QEMUFile *f)
80 * We must set qemufile error before the real shutdown(), otherwise
81 * there can be a race window where we thought IO all went though
82 * (because last_error==NULL) but actually IO has already stopped.
84 * If without correct ordering, the race can happen like this:
86 * page receiver other thread
87 * ------------- ------------
90 * returns 0 (buffer all zero)
91 * (we didn't check this retcode)
92 * try to detect IO error
93 * last_error==NULL, IO okay
94 * install ALL-ZERO page
99 qemu_file_set_error(f, -EIO);
102 if (!qio_channel_has_feature(f->ioc,
103 QIO_CHANNEL_FEATURE_SHUTDOWN)) {
107 if (qio_channel_shutdown(f->ioc, QIO_CHANNEL_SHUTDOWN_BOTH, NULL) < 0) {
114 bool qemu_file_mode_is_not_valid(const char *mode)
117 (mode[0] != 'r' && mode[0] != 'w') ||
118 mode[1] != 'b' || mode[2] != 0) {
119 fprintf(stderr, "qemu_fopen: Argument validity check failed\n");
126 static QEMUFile *qemu_file_new_impl(QIOChannel *ioc, bool is_writable)
130 f = g_new0(QEMUFile, 1);
134 f->is_writable = is_writable;
140 * Result: QEMUFile* for a 'return path' for comms in the opposite direction
141 * NULL if not available
143 QEMUFile *qemu_file_get_return_path(QEMUFile *f)
145 return qemu_file_new_impl(f->ioc, !f->is_writable);
148 QEMUFile *qemu_file_new_output(QIOChannel *ioc)
150 return qemu_file_new_impl(ioc, true);
153 QEMUFile *qemu_file_new_input(QIOChannel *ioc)
155 return qemu_file_new_impl(ioc, false);
158 void qemu_file_set_hooks(QEMUFile *f, const QEMUFileHooks *hooks)
164 * Get last error for stream f with optional Error*
166 * Return negative error value if there has been an error on previous
167 * operations, return 0 if no error happened.
168 * Optional, it returns Error* in errp, but it may be NULL even if return value
172 int qemu_file_get_error_obj(QEMUFile *f, Error **errp)
175 *errp = f->last_error_obj ? error_copy(f->last_error_obj) : NULL;
177 return f->last_error;
181 * Get last error for either stream f1 or f2 with optional Error*.
182 * The error returned (non-zero) can be either from f1 or f2.
184 * If any of the qemufile* is NULL, then skip the check on that file.
186 * When there is no error on both qemufile, zero is returned.
188 int qemu_file_get_error_obj_any(QEMUFile *f1, QEMUFile *f2, Error **errp)
193 ret = qemu_file_get_error_obj(f1, errp);
194 /* If there's already error detected, return */
201 ret = qemu_file_get_error_obj(f2, errp);
208 * Set the last error for stream f with optional Error*
210 void qemu_file_set_error_obj(QEMUFile *f, int ret, Error *err)
212 if (f->last_error == 0 && ret) {
214 error_propagate(&f->last_error_obj, err);
216 error_report_err(err);
221 * Get last error for stream f
223 * Return negative error value if there has been an error on previous
224 * operations, return 0 if no error happened.
227 int qemu_file_get_error(QEMUFile *f)
229 return qemu_file_get_error_obj(f, NULL);
233 * Set the last error for stream f
235 void qemu_file_set_error(QEMUFile *f, int ret)
237 qemu_file_set_error_obj(f, ret, NULL);
240 bool qemu_file_is_writable(QEMUFile *f)
242 return f->is_writable;
245 static void qemu_iovec_release_ram(QEMUFile *f)
250 /* Find and release all the contiguous memory ranges marked as may_free. */
251 idx = find_next_bit(f->may_free, f->iovcnt, 0);
252 if (idx >= f->iovcnt) {
257 /* The madvise() in the loop is called for iov within a continuous range and
258 * then reinitialize the iov. And in the end, madvise() is called for the
261 while ((idx = find_next_bit(f->may_free, f->iovcnt, idx + 1)) < f->iovcnt) {
262 /* check for adjacent buffer and coalesce them */
263 if (iov.iov_base + iov.iov_len == f->iov[idx].iov_base) {
264 iov.iov_len += f->iov[idx].iov_len;
267 if (qemu_madvise(iov.iov_base, iov.iov_len, QEMU_MADV_DONTNEED) < 0) {
268 error_report("migrate: madvise DONTNEED failed %p %zd: %s",
269 iov.iov_base, iov.iov_len, strerror(errno));
273 if (qemu_madvise(iov.iov_base, iov.iov_len, QEMU_MADV_DONTNEED) < 0) {
274 error_report("migrate: madvise DONTNEED failed %p %zd: %s",
275 iov.iov_base, iov.iov_len, strerror(errno));
277 memset(f->may_free, 0, sizeof(f->may_free));
282 * Flushes QEMUFile buffer
284 * This will flush all pending data. If data was only partially flushed, it
285 * will set an error state.
287 void qemu_fflush(QEMUFile *f)
289 if (!qemu_file_is_writable(f)) {
293 if (qemu_file_get_error(f)) {
297 Error *local_error = NULL;
298 if (qio_channel_writev_all(f->ioc,
301 qemu_file_set_error_obj(f, -EIO, local_error);
303 f->total_transferred += iov_size(f->iov, f->iovcnt);
306 qemu_iovec_release_ram(f);
313 void ram_control_before_iterate(QEMUFile *f, uint64_t flags)
317 if (f->hooks && f->hooks->before_ram_iterate) {
318 ret = f->hooks->before_ram_iterate(f, flags, NULL);
320 qemu_file_set_error(f, ret);
325 void ram_control_after_iterate(QEMUFile *f, uint64_t flags)
329 if (f->hooks && f->hooks->after_ram_iterate) {
330 ret = f->hooks->after_ram_iterate(f, flags, NULL);
332 qemu_file_set_error(f, ret);
337 void ram_control_load_hook(QEMUFile *f, uint64_t flags, void *data)
339 if (f->hooks && f->hooks->hook_ram_load) {
340 int ret = f->hooks->hook_ram_load(f, flags, data);
342 qemu_file_set_error(f, ret);
347 size_t ram_control_save_page(QEMUFile *f, ram_addr_t block_offset,
348 ram_addr_t offset, size_t size,
349 uint64_t *bytes_sent)
351 if (f->hooks && f->hooks->save_page) {
352 int ret = f->hooks->save_page(f, block_offset,
353 offset, size, bytes_sent);
354 if (ret != RAM_SAVE_CONTROL_NOT_SUPP) {
355 f->rate_limit_used += size;
358 if (ret != RAM_SAVE_CONTROL_DELAYED &&
359 ret != RAM_SAVE_CONTROL_NOT_SUPP) {
360 if (bytes_sent && *bytes_sent > 0) {
361 qemu_file_credit_transfer(f, *bytes_sent);
362 } else if (ret < 0) {
363 qemu_file_set_error(f, ret);
370 return RAM_SAVE_CONTROL_NOT_SUPP;
374 * Attempt to fill the buffer from the underlying file
375 * Returns the number of bytes read, or negative value for an error.
377 * Note that it can return a partially full buffer even in a not error/not EOF
378 * case if the underlying file descriptor gives a short read, and that can
379 * happen even on a blocking fd.
381 static ssize_t coroutine_mixed_fn qemu_fill_buffer(QEMUFile *f)
385 Error *local_error = NULL;
387 assert(!qemu_file_is_writable(f));
389 pending = f->buf_size - f->buf_index;
391 memmove(f->buf, f->buf + f->buf_index, pending);
394 f->buf_size = pending;
396 if (qemu_file_get_error(f)) {
401 len = qio_channel_read(f->ioc,
402 (char *)f->buf + pending,
403 IO_BUF_SIZE - pending,
405 if (len == QIO_CHANNEL_ERR_BLOCK) {
406 if (qemu_in_coroutine()) {
407 qio_channel_yield(f->ioc, G_IO_IN);
409 qio_channel_wait(f->ioc, G_IO_IN);
411 } else if (len < 0) {
414 } while (len == QIO_CHANNEL_ERR_BLOCK);
418 f->total_transferred += len;
419 } else if (len == 0) {
420 qemu_file_set_error_obj(f, -EIO, local_error);
422 qemu_file_set_error_obj(f, len, local_error);
428 void qemu_file_credit_transfer(QEMUFile *f, size_t size)
430 f->total_transferred += size;
435 * Returns negative error value if any error happened on previous operations or
436 * while closing the file. Returns 0 or positive number on success.
438 * The meaning of return value on success depends on the specific backend
441 int qemu_fclose(QEMUFile *f)
445 ret = qemu_file_get_error(f);
447 ret2 = qio_channel_close(f->ioc, NULL);
451 g_clear_pointer(&f->ioc, object_unref);
453 /* If any error was spotted before closing, we should report it
454 * instead of the close() return value.
459 error_free(f->last_error_obj);
461 trace_qemu_file_fclose();
466 * Add buf to iovec. Do flush if iovec is full.
469 * 1 iovec is full and flushed
470 * 0 iovec is not flushed
473 static int add_to_iovec(QEMUFile *f, const uint8_t *buf, size_t size,
476 /* check for adjacent buffer and coalesce them */
477 if (f->iovcnt > 0 && buf == f->iov[f->iovcnt - 1].iov_base +
478 f->iov[f->iovcnt - 1].iov_len &&
479 may_free == test_bit(f->iovcnt - 1, f->may_free))
481 f->iov[f->iovcnt - 1].iov_len += size;
483 if (f->iovcnt >= MAX_IOV_SIZE) {
484 /* Should only happen if a previous fflush failed */
485 assert(qemu_file_get_error(f) || !qemu_file_is_writable(f));
489 set_bit(f->iovcnt, f->may_free);
491 f->iov[f->iovcnt].iov_base = (uint8_t *)buf;
492 f->iov[f->iovcnt++].iov_len = size;
495 if (f->iovcnt >= MAX_IOV_SIZE) {
503 static void add_buf_to_iovec(QEMUFile *f, size_t len)
505 if (!add_to_iovec(f, f->buf + f->buf_index, len, false)) {
507 if (f->buf_index == IO_BUF_SIZE) {
513 void qemu_put_buffer_async(QEMUFile *f, const uint8_t *buf, size_t size,
520 f->rate_limit_used += size;
521 add_to_iovec(f, buf, size, may_free);
524 void qemu_put_buffer(QEMUFile *f, const uint8_t *buf, size_t size)
533 l = IO_BUF_SIZE - f->buf_index;
537 memcpy(f->buf + f->buf_index, buf, l);
538 f->rate_limit_used += l;
539 add_buf_to_iovec(f, l);
540 if (qemu_file_get_error(f)) {
548 void qemu_put_byte(QEMUFile *f, int v)
554 f->buf[f->buf_index] = v;
555 f->rate_limit_used++;
556 add_buf_to_iovec(f, 1);
559 void qemu_file_skip(QEMUFile *f, int size)
561 if (f->buf_index + size <= f->buf_size) {
562 f->buf_index += size;
567 * Read 'size' bytes from file (at 'offset') without moving the
568 * pointer and set 'buf' to point to that data.
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 size_t coroutine_mixed_fn qemu_peek_buffer(QEMUFile *f, uint8_t **buf, size_t size, size_t offset)
579 assert(!qemu_file_is_writable(f));
580 assert(offset < IO_BUF_SIZE);
581 assert(size <= IO_BUF_SIZE - offset);
583 /* The 1st byte to read from */
584 index = f->buf_index + offset;
585 /* The number of available bytes starting at index */
586 pending = f->buf_size - index;
589 * qemu_fill_buffer might return just a few bytes, even when there isn't
590 * an error, so loop collecting them until we get enough.
592 while (pending < size) {
593 int received = qemu_fill_buffer(f);
599 index = f->buf_index + offset;
600 pending = f->buf_size - index;
606 if (size > pending) {
610 *buf = f->buf + index;
615 * Read 'size' bytes of data from the file into buf.
616 * 'size' can be larger than the internal buffer.
618 * It will return size bytes unless there was an error, in which case it will
619 * return as many as it managed to read (assuming blocking fd's which
620 * all current QEMUFile are)
622 size_t coroutine_mixed_fn qemu_get_buffer(QEMUFile *f, uint8_t *buf, size_t size)
624 size_t pending = size;
627 while (pending > 0) {
631 res = qemu_peek_buffer(f, &src, MIN(pending, IO_BUF_SIZE), 0);
635 memcpy(buf, src, res);
636 qemu_file_skip(f, res);
645 * Read 'size' bytes of data from the file.
646 * 'size' can be larger than the internal buffer.
649 * may be held on an internal buffer (in which case *buf is updated
650 * to point to it) that is valid until the next qemu_file operation.
652 * will be copied to the *buf that was passed in.
654 * The code tries to avoid the copy if possible.
656 * It will return size bytes unless there was an error, in which case it will
657 * return as many as it managed to read (assuming blocking fd's which
658 * all current QEMUFile are)
660 * Note: Since **buf may get changed, the caller should take care to
661 * keep a pointer to the original buffer if it needs to deallocate it.
663 size_t coroutine_mixed_fn qemu_get_buffer_in_place(QEMUFile *f, uint8_t **buf, size_t size)
665 if (size < IO_BUF_SIZE) {
669 res = qemu_peek_buffer(f, &src, size, 0);
672 qemu_file_skip(f, res);
678 return qemu_get_buffer(f, *buf, size);
682 * Peeks a single byte from the buffer; this isn't guaranteed to work if
683 * offset leaves a gap after the previous read/peeked data.
685 int coroutine_mixed_fn qemu_peek_byte(QEMUFile *f, int offset)
687 int index = f->buf_index + offset;
689 assert(!qemu_file_is_writable(f));
690 assert(offset < IO_BUF_SIZE);
692 if (index >= f->buf_size) {
694 index = f->buf_index + offset;
695 if (index >= f->buf_size) {
699 return f->buf[index];
702 int coroutine_mixed_fn qemu_get_byte(QEMUFile *f)
706 result = qemu_peek_byte(f, 0);
707 qemu_file_skip(f, 1);
711 uint64_t qemu_file_total_transferred_fast(QEMUFile *f)
713 uint64_t ret = f->total_transferred;
716 for (i = 0; i < f->iovcnt; i++) {
717 ret += f->iov[i].iov_len;
723 uint64_t qemu_file_total_transferred(QEMUFile *f)
726 return f->total_transferred;
729 int qemu_file_rate_limit(QEMUFile *f)
731 if (qemu_file_get_error(f)) {
734 if (f->rate_limit_max > 0 && f->rate_limit_used > f->rate_limit_max) {
740 int64_t qemu_file_get_rate_limit(QEMUFile *f)
742 return f->rate_limit_max;
745 void qemu_file_set_rate_limit(QEMUFile *f, int64_t limit)
747 f->rate_limit_max = limit;
750 void qemu_file_reset_rate_limit(QEMUFile *f)
752 f->rate_limit_used = 0;
755 void qemu_file_acct_rate_limit(QEMUFile *f, int64_t len)
757 f->rate_limit_used += len;
760 void qemu_put_be16(QEMUFile *f, unsigned int v)
762 qemu_put_byte(f, v >> 8);
766 void qemu_put_be32(QEMUFile *f, unsigned int v)
768 qemu_put_byte(f, v >> 24);
769 qemu_put_byte(f, v >> 16);
770 qemu_put_byte(f, v >> 8);
774 void qemu_put_be64(QEMUFile *f, uint64_t v)
776 qemu_put_be32(f, v >> 32);
780 unsigned int qemu_get_be16(QEMUFile *f)
783 v = qemu_get_byte(f) << 8;
784 v |= qemu_get_byte(f);
788 unsigned int qemu_get_be32(QEMUFile *f)
791 v = (unsigned int)qemu_get_byte(f) << 24;
792 v |= qemu_get_byte(f) << 16;
793 v |= qemu_get_byte(f) << 8;
794 v |= qemu_get_byte(f);
798 uint64_t qemu_get_be64(QEMUFile *f)
801 v = (uint64_t)qemu_get_be32(f) << 32;
802 v |= qemu_get_be32(f);
806 /* return the size after compression, or negative value on error */
807 static int qemu_compress_data(z_stream *stream, uint8_t *dest, size_t dest_len,
808 const uint8_t *source, size_t source_len)
812 err = deflateReset(stream);
817 stream->avail_in = source_len;
818 stream->next_in = (uint8_t *)source;
819 stream->avail_out = dest_len;
820 stream->next_out = dest;
822 err = deflate(stream, Z_FINISH);
823 if (err != Z_STREAM_END) {
827 return stream->next_out - dest;
830 /* Compress size bytes of data start at p and store the compressed
831 * data to the buffer of f.
833 * Since the file is dummy file with empty_ops, return -1 if f has no space to
834 * save the compressed data.
836 ssize_t qemu_put_compression_data(QEMUFile *f, z_stream *stream,
837 const uint8_t *p, size_t size)
839 ssize_t blen = IO_BUF_SIZE - f->buf_index - sizeof(int32_t);
841 if (blen < compressBound(size)) {
845 blen = qemu_compress_data(stream, f->buf + f->buf_index + sizeof(int32_t),
851 qemu_put_be32(f, blen);
852 add_buf_to_iovec(f, blen);
853 return blen + sizeof(int32_t);
856 /* Put the data in the buffer of f_src to the buffer of f_des, and
857 * then reset the buf_index of f_src to 0.
860 int qemu_put_qemu_file(QEMUFile *f_des, QEMUFile *f_src)
864 if (f_src->buf_index > 0) {
865 len = f_src->buf_index;
866 qemu_put_buffer(f_des, f_src->buf, f_src->buf_index);
867 f_src->buf_index = 0;
874 * Get a string whose length is determined by a single preceding byte
875 * A preallocated 256 byte buffer must be passed in.
876 * Returns: len on success and a 0 terminated string in the buffer
878 * (Note a 0 length string will return 0 either way)
880 size_t coroutine_fn qemu_get_counted_string(QEMUFile *f, char buf[256])
882 size_t len = qemu_get_byte(f);
883 size_t res = qemu_get_buffer(f, (uint8_t *)buf, len);
887 return res == len ? res : 0;
891 * Put a string with one preceding byte containing its length. The length of
892 * the string should be less than 256.
894 void qemu_put_counted_string(QEMUFile *f, const char *str)
896 size_t len = strlen(str);
899 qemu_put_byte(f, len);
900 qemu_put_buffer(f, (const uint8_t *)str, len);
904 * Set the blocking state of the QEMUFile.
905 * Note: On some transports the OS only keeps a single blocking state for
906 * both directions, and thus changing the blocking on the main
907 * QEMUFile can also affect the return path.
909 void qemu_file_set_blocking(QEMUFile *f, bool block)
911 qio_channel_set_blocking(f->ioc, block, NULL);
917 * Get the ioc object for the file, without incrementing
918 * the reference count.
920 * Returns: the ioc object
922 QIOChannel *qemu_file_get_ioc(QEMUFile *file)
928 * Read size bytes from QEMUFile f and write them to fd.
930 int qemu_file_get_to_fd(QEMUFile *f, int fd, size_t size)
933 size_t pending = f->buf_size - f->buf_index;
937 rc = qemu_fill_buffer(f);
947 rc = write(fd, f->buf + f->buf_index, MIN(pending, size));