2 * linux/net/sunrpc/xdr.c
6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
9 #include <linux/types.h>
10 #include <linux/socket.h>
11 #include <linux/string.h>
12 #include <linux/kernel.h>
13 #include <linux/pagemap.h>
14 #include <linux/errno.h>
16 #include <linux/net.h>
18 #include <linux/sunrpc/xdr.h>
19 #include <linux/sunrpc/msg_prot.h>
22 * XDR functions for basic NFS types
25 xdr_encode_netobj(u32 *p, const struct xdr_netobj *obj)
27 unsigned int quadlen = XDR_QUADLEN(obj->len);
29 p[quadlen] = 0; /* zero trailing bytes */
30 *p++ = htonl(obj->len);
31 memcpy(p, obj->data, obj->len);
32 return p + XDR_QUADLEN(obj->len);
36 xdr_decode_netobj(u32 *p, struct xdr_netobj *obj)
40 if ((len = ntohl(*p++)) > XDR_MAX_NETOBJ)
44 return p + XDR_QUADLEN(len);
48 * xdr_encode_opaque_fixed - Encode fixed length opaque data
49 * @p: pointer to current position in XDR buffer.
50 * @ptr: pointer to data to encode (or NULL)
51 * @nbytes: size of data.
53 * Copy the array of data of length nbytes at ptr to the XDR buffer
54 * at position p, then align to the next 32-bit boundary by padding
55 * with zero bytes (see RFC1832).
56 * Note: if ptr is NULL, only the padding is performed.
58 * Returns the updated current XDR buffer position
61 u32 *xdr_encode_opaque_fixed(u32 *p, const void *ptr, unsigned int nbytes)
63 if (likely(nbytes != 0)) {
64 unsigned int quadlen = XDR_QUADLEN(nbytes);
65 unsigned int padding = (quadlen << 2) - nbytes;
68 memcpy(p, ptr, nbytes);
70 memset((char *)p + nbytes, 0, padding);
75 EXPORT_SYMBOL(xdr_encode_opaque_fixed);
78 * xdr_encode_opaque - Encode variable length opaque data
79 * @p: pointer to current position in XDR buffer.
80 * @ptr: pointer to data to encode (or NULL)
81 * @nbytes: size of data.
83 * Returns the updated current XDR buffer position
85 u32 *xdr_encode_opaque(u32 *p, const void *ptr, unsigned int nbytes)
88 return xdr_encode_opaque_fixed(p, ptr, nbytes);
90 EXPORT_SYMBOL(xdr_encode_opaque);
93 xdr_encode_string(u32 *p, const char *string)
95 return xdr_encode_array(p, string, strlen(string));
99 xdr_decode_string(u32 *p, char **sp, int *lenp, int maxlen)
104 if ((len = ntohl(*p++)) > maxlen)
108 if ((len % 4) != 0) {
111 string = (char *) (p - 1);
112 memmove(string, p, len);
116 return p + XDR_QUADLEN(len);
120 xdr_decode_string_inplace(u32 *p, char **sp, int *lenp, int maxlen)
124 if ((len = ntohl(*p++)) > maxlen)
128 return p + XDR_QUADLEN(len);
132 xdr_encode_pages(struct xdr_buf *xdr, struct page **pages, unsigned int base,
135 struct kvec *tail = xdr->tail;
139 xdr->page_base = base;
142 p = (u32 *)xdr->head[0].iov_base + XDR_QUADLEN(xdr->head[0].iov_len);
147 unsigned int pad = 4 - (len & 3);
150 tail->iov_base = (char *)p + (len & 3);
159 xdr_inline_pages(struct xdr_buf *xdr, unsigned int offset,
160 struct page **pages, unsigned int base, unsigned int len)
162 struct kvec *head = xdr->head;
163 struct kvec *tail = xdr->tail;
164 char *buf = (char *)head->iov_base;
165 unsigned int buflen = head->iov_len;
167 head->iov_len = offset;
170 xdr->page_base = base;
173 tail->iov_base = buf + offset;
174 tail->iov_len = buflen - offset;
180 xdr_partial_copy_from_skb(struct xdr_buf *xdr, unsigned int base,
182 skb_read_actor_t copy_actor)
184 struct page **ppage = xdr->pages;
185 unsigned int len, pglen = xdr->page_len;
188 len = xdr->head[0].iov_len;
191 ret = copy_actor(desc, (char *)xdr->head[0].iov_base + base, len);
192 if (ret != len || !desc->count)
204 if (base || xdr->page_base) {
206 base += xdr->page_base;
207 ppage += base >> PAGE_CACHE_SHIFT;
208 base &= ~PAGE_CACHE_MASK;
213 len = PAGE_CACHE_SIZE;
214 kaddr = kmap_atomic(*ppage, KM_SKB_SUNRPC_DATA);
219 ret = copy_actor(desc, kaddr + base, len);
224 ret = copy_actor(desc, kaddr, len);
226 flush_dcache_page(*ppage);
227 kunmap_atomic(kaddr, KM_SKB_SUNRPC_DATA);
228 if (ret != len || !desc->count)
231 } while ((pglen -= len) != 0);
233 len = xdr->tail[0].iov_len;
235 copy_actor(desc, (char *)xdr->tail[0].iov_base + base, len - base);
240 xdr_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen,
241 struct xdr_buf *xdr, unsigned int base, int msgflags)
243 struct page **ppage = xdr->pages;
244 unsigned int len, pglen = xdr->page_len;
246 ssize_t (*sendpage)(struct socket *, struct page *, int, size_t, int);
248 len = xdr->head[0].iov_len;
249 if (base < len || (addr != NULL && base == 0)) {
251 .iov_base = xdr->head[0].iov_base + base,
252 .iov_len = len - base,
254 struct msghdr msg = {
256 .msg_namelen = addrlen,
257 .msg_flags = msgflags,
260 msg.msg_flags |= MSG_MORE;
262 if (iov.iov_len != 0)
263 err = kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
265 err = kernel_sendmsg(sock, &msg, NULL, 0, 0);
270 if (err != iov.iov_len)
282 if (base || xdr->page_base) {
284 base += xdr->page_base;
285 ppage += base >> PAGE_CACHE_SHIFT;
286 base &= ~PAGE_CACHE_MASK;
289 sendpage = sock->ops->sendpage ? : sock_no_sendpage;
291 int flags = msgflags;
293 len = PAGE_CACHE_SIZE;
299 if (pglen != len || xdr->tail[0].iov_len != 0)
302 /* Hmm... We might be dealing with highmem pages */
303 if (PageHighMem(*ppage))
304 sendpage = sock_no_sendpage;
305 err = sendpage(sock, *ppage, base, len, flags);
314 } while ((pglen -= len) != 0);
316 len = xdr->tail[0].iov_len;
319 .iov_base = xdr->tail[0].iov_base + base,
320 .iov_len = len - base,
322 struct msghdr msg = {
323 .msg_flags = msgflags,
325 err = kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
337 * Helper routines for doing 'memmove' like operations on a struct xdr_buf
339 * _shift_data_right_pages
340 * @pages: vector of pages containing both the source and dest memory area.
341 * @pgto_base: page vector address of destination
342 * @pgfrom_base: page vector address of source
343 * @len: number of bytes to copy
345 * Note: the addresses pgto_base and pgfrom_base are both calculated in
347 * if a memory area starts at byte 'base' in page 'pages[i]',
348 * then its address is given as (i << PAGE_CACHE_SHIFT) + base
349 * Also note: pgfrom_base must be < pgto_base, but the memory areas
350 * they point to may overlap.
353 _shift_data_right_pages(struct page **pages, size_t pgto_base,
354 size_t pgfrom_base, size_t len)
356 struct page **pgfrom, **pgto;
360 BUG_ON(pgto_base <= pgfrom_base);
365 pgto = pages + (pgto_base >> PAGE_CACHE_SHIFT);
366 pgfrom = pages + (pgfrom_base >> PAGE_CACHE_SHIFT);
368 pgto_base &= ~PAGE_CACHE_MASK;
369 pgfrom_base &= ~PAGE_CACHE_MASK;
372 /* Are any pointers crossing a page boundary? */
373 if (pgto_base == 0) {
374 flush_dcache_page(*pgto);
375 pgto_base = PAGE_CACHE_SIZE;
378 if (pgfrom_base == 0) {
379 pgfrom_base = PAGE_CACHE_SIZE;
384 if (copy > pgto_base)
386 if (copy > pgfrom_base)
391 vto = kmap_atomic(*pgto, KM_USER0);
392 vfrom = kmap_atomic(*pgfrom, KM_USER1);
393 memmove(vto + pgto_base, vfrom + pgfrom_base, copy);
394 kunmap_atomic(vfrom, KM_USER1);
395 kunmap_atomic(vto, KM_USER0);
397 } while ((len -= copy) != 0);
398 flush_dcache_page(*pgto);
403 * @pages: array of pages
404 * @pgbase: page vector address of destination
405 * @p: pointer to source data
408 * Copies data from an arbitrary memory location into an array of pages
409 * The copy is assumed to be non-overlapping.
412 _copy_to_pages(struct page **pages, size_t pgbase, const char *p, size_t len)
418 pgto = pages + (pgbase >> PAGE_CACHE_SHIFT);
419 pgbase &= ~PAGE_CACHE_MASK;
422 copy = PAGE_CACHE_SIZE - pgbase;
426 vto = kmap_atomic(*pgto, KM_USER0);
427 memcpy(vto + pgbase, p, copy);
428 kunmap_atomic(vto, KM_USER0);
431 if (pgbase == PAGE_CACHE_SIZE) {
432 flush_dcache_page(*pgto);
438 } while ((len -= copy) != 0);
439 flush_dcache_page(*pgto);
444 * @p: pointer to destination
445 * @pages: array of pages
446 * @pgbase: offset of source data
449 * Copies data into an arbitrary memory location from an array of pages
450 * The copy is assumed to be non-overlapping.
453 _copy_from_pages(char *p, struct page **pages, size_t pgbase, size_t len)
455 struct page **pgfrom;
459 pgfrom = pages + (pgbase >> PAGE_CACHE_SHIFT);
460 pgbase &= ~PAGE_CACHE_MASK;
463 copy = PAGE_CACHE_SIZE - pgbase;
467 vfrom = kmap_atomic(*pgfrom, KM_USER0);
468 memcpy(p, vfrom + pgbase, copy);
469 kunmap_atomic(vfrom, KM_USER0);
472 if (pgbase == PAGE_CACHE_SIZE) {
478 } while ((len -= copy) != 0);
484 * @len: bytes to remove from buf->head[0]
486 * Shrinks XDR buffer's header kvec buf->head[0] by
487 * 'len' bytes. The extra data is not lost, but is instead
488 * moved into the inlined pages and/or the tail.
491 xdr_shrink_bufhead(struct xdr_buf *buf, size_t len)
493 struct kvec *head, *tail;
495 unsigned int pglen = buf->page_len;
499 BUG_ON (len > head->iov_len);
501 /* Shift the tail first */
502 if (tail->iov_len != 0) {
503 if (tail->iov_len > len) {
504 copy = tail->iov_len - len;
505 memmove((char *)tail->iov_base + len,
506 tail->iov_base, copy);
508 /* Copy from the inlined pages into the tail */
513 if (offs >= tail->iov_len)
515 else if (copy > tail->iov_len - offs)
516 copy = tail->iov_len - offs;
518 _copy_from_pages((char *)tail->iov_base + offs,
520 buf->page_base + pglen + offs - len,
522 /* Do we also need to copy data from the head into the tail ? */
524 offs = copy = len - pglen;
525 if (copy > tail->iov_len)
526 copy = tail->iov_len;
527 memcpy(tail->iov_base,
528 (char *)head->iov_base +
529 head->iov_len - offs,
533 /* Now handle pages */
536 _shift_data_right_pages(buf->pages,
537 buf->page_base + len,
543 _copy_to_pages(buf->pages, buf->page_base,
544 (char *)head->iov_base + head->iov_len - len,
547 head->iov_len -= len;
549 /* Have we truncated the message? */
550 if (buf->len > buf->buflen)
551 buf->len = buf->buflen;
557 * @len: bytes to remove from buf->pages
559 * Shrinks XDR buffer's page array buf->pages by
560 * 'len' bytes. The extra data is not lost, but is instead
561 * moved into the tail.
564 xdr_shrink_pagelen(struct xdr_buf *buf, size_t len)
569 unsigned int pglen = buf->page_len;
572 BUG_ON (len > pglen);
574 /* Shift the tail first */
575 if (tail->iov_len != 0) {
576 p = (char *)tail->iov_base + len;
577 if (tail->iov_len > len) {
578 copy = tail->iov_len - len;
579 memmove(p, tail->iov_base, copy);
582 /* Copy from the inlined pages into the tail */
584 if (copy > tail->iov_len)
585 copy = tail->iov_len;
586 _copy_from_pages((char *)tail->iov_base,
587 buf->pages, buf->page_base + pglen - len,
590 buf->page_len -= len;
592 /* Have we truncated the message? */
593 if (buf->len > buf->buflen)
594 buf->len = buf->buflen;
598 xdr_shift_buf(struct xdr_buf *buf, size_t len)
600 xdr_shrink_bufhead(buf, len);
604 * xdr_init_encode - Initialize a struct xdr_stream for sending data.
605 * @xdr: pointer to xdr_stream struct
606 * @buf: pointer to XDR buffer in which to encode data
607 * @p: current pointer inside XDR buffer
609 * Note: at the moment the RPC client only passes the length of our
610 * scratch buffer in the xdr_buf's header kvec. Previously this
611 * meant we needed to call xdr_adjust_iovec() after encoding the
612 * data. With the new scheme, the xdr_stream manages the details
613 * of the buffer length, and takes care of adjusting the kvec
616 void xdr_init_encode(struct xdr_stream *xdr, struct xdr_buf *buf, uint32_t *p)
618 struct kvec *iov = buf->head;
622 xdr->end = (uint32_t *)((char *)iov->iov_base + iov->iov_len);
623 buf->len = iov->iov_len = (char *)p - (char *)iov->iov_base;
626 EXPORT_SYMBOL(xdr_init_encode);
629 * xdr_reserve_space - Reserve buffer space for sending
630 * @xdr: pointer to xdr_stream
631 * @nbytes: number of bytes to reserve
633 * Checks that we have enough buffer space to encode 'nbytes' more
634 * bytes of data. If so, update the total xdr_buf length, and
635 * adjust the length of the current kvec.
637 uint32_t * xdr_reserve_space(struct xdr_stream *xdr, size_t nbytes)
639 uint32_t *p = xdr->p;
642 /* align nbytes on the next 32-bit boundary */
645 q = p + (nbytes >> 2);
646 if (unlikely(q > xdr->end || q < p))
649 xdr->iov->iov_len += nbytes;
650 xdr->buf->len += nbytes;
653 EXPORT_SYMBOL(xdr_reserve_space);
656 * xdr_write_pages - Insert a list of pages into an XDR buffer for sending
657 * @xdr: pointer to xdr_stream
658 * @pages: list of pages
659 * @base: offset of first byte
660 * @len: length of data in bytes
663 void xdr_write_pages(struct xdr_stream *xdr, struct page **pages, unsigned int base,
666 struct xdr_buf *buf = xdr->buf;
667 struct kvec *iov = buf->tail;
669 buf->page_base = base;
672 iov->iov_base = (char *)xdr->p;
677 unsigned int pad = 4 - (len & 3);
679 BUG_ON(xdr->p >= xdr->end);
680 iov->iov_base = (char *)xdr->p + (len & 3);
688 EXPORT_SYMBOL(xdr_write_pages);
691 * xdr_init_decode - Initialize an xdr_stream for decoding data.
692 * @xdr: pointer to xdr_stream struct
693 * @buf: pointer to XDR buffer from which to decode data
694 * @p: current pointer inside XDR buffer
696 void xdr_init_decode(struct xdr_stream *xdr, struct xdr_buf *buf, uint32_t *p)
698 struct kvec *iov = buf->head;
699 unsigned int len = iov->iov_len;
706 xdr->end = (uint32_t *)((char *)iov->iov_base + len);
708 EXPORT_SYMBOL(xdr_init_decode);
711 * xdr_inline_decode - Retrieve non-page XDR data to decode
712 * @xdr: pointer to xdr_stream struct
713 * @nbytes: number of bytes of data to decode
715 * Check if the input buffer is long enough to enable us to decode
716 * 'nbytes' more bytes of data starting at the current position.
717 * If so return the current pointer, then update the current
720 uint32_t * xdr_inline_decode(struct xdr_stream *xdr, size_t nbytes)
722 uint32_t *p = xdr->p;
723 uint32_t *q = p + XDR_QUADLEN(nbytes);
725 if (unlikely(q > xdr->end || q < p))
730 EXPORT_SYMBOL(xdr_inline_decode);
733 * xdr_read_pages - Ensure page-based XDR data to decode is aligned at current pointer position
734 * @xdr: pointer to xdr_stream struct
735 * @len: number of bytes of page data
737 * Moves data beyond the current pointer position from the XDR head[] buffer
738 * into the page list. Any data that lies beyond current position + "len"
739 * bytes is moved into the XDR tail[]. The current pointer is then
740 * repositioned at the beginning of the XDR tail.
742 void xdr_read_pages(struct xdr_stream *xdr, unsigned int len)
744 struct xdr_buf *buf = xdr->buf;
750 /* Realign pages to current pointer position */
752 shift = iov->iov_len + (char *)iov->iov_base - (char *)xdr->p;
754 xdr_shrink_bufhead(buf, shift);
756 /* Truncate page data and move it into the tail */
757 if (buf->page_len > len)
758 xdr_shrink_pagelen(buf, buf->page_len - len);
759 padding = (XDR_QUADLEN(len) << 2) - len;
760 xdr->iov = iov = buf->tail;
761 /* Compute remaining message length. */
763 shift = buf->buflen - buf->len;
769 * Position current pointer at beginning of tail, and
770 * set remaining message length.
772 xdr->p = (uint32_t *)((char *)iov->iov_base + padding);
773 xdr->end = (uint32_t *)((char *)iov->iov_base + end);
775 EXPORT_SYMBOL(xdr_read_pages);
777 static struct kvec empty_iov = {.iov_base = NULL, .iov_len = 0};
780 xdr_buf_from_iov(struct kvec *iov, struct xdr_buf *buf)
783 buf->tail[0] = empty_iov;
785 buf->buflen = buf->len = iov->iov_len;
788 /* Sets subiov to the intersection of iov with the buffer of length len
789 * starting base bytes after iov. Indicates empty intersection by setting
790 * length of subiov to zero. Decrements len by length of subiov, sets base
791 * to zero (or decrements it by length of iov if subiov is empty). */
793 iov_subsegment(struct kvec *iov, struct kvec *subiov, int *base, int *len)
795 if (*base > iov->iov_len) {
796 subiov->iov_base = NULL;
798 *base -= iov->iov_len;
800 subiov->iov_base = iov->iov_base + *base;
801 subiov->iov_len = min(*len, (int)iov->iov_len - *base);
804 *len -= subiov->iov_len;
807 /* Sets subbuf to the portion of buf of length len beginning base bytes
808 * from the start of buf. Returns -1 if base of length are out of bounds. */
810 xdr_buf_subsegment(struct xdr_buf *buf, struct xdr_buf *subbuf,
815 subbuf->buflen = subbuf->len = len;
816 iov_subsegment(buf->head, subbuf->head, &base, &len);
818 if (base < buf->page_len) {
819 i = (base + buf->page_base) >> PAGE_CACHE_SHIFT;
820 subbuf->pages = &buf->pages[i];
821 subbuf->page_base = (base + buf->page_base) & ~PAGE_CACHE_MASK;
822 subbuf->page_len = min((int)buf->page_len - base, len);
823 len -= subbuf->page_len;
826 base -= buf->page_len;
827 subbuf->page_len = 0;
830 iov_subsegment(buf->tail, subbuf->tail, &base, &len);
836 /* obj is assumed to point to allocated memory of size at least len: */
838 read_bytes_from_xdr_buf(struct xdr_buf *buf, int base, void *obj, int len)
840 struct xdr_buf subbuf;
844 status = xdr_buf_subsegment(buf, &subbuf, base, len);
847 this_len = min(len, (int)subbuf.head[0].iov_len);
848 memcpy(obj, subbuf.head[0].iov_base, this_len);
851 this_len = min(len, (int)subbuf.page_len);
853 _copy_from_pages(obj, subbuf.pages, subbuf.page_base, this_len);
856 this_len = min(len, (int)subbuf.tail[0].iov_len);
857 memcpy(obj, subbuf.tail[0].iov_base, this_len);
863 read_u32_from_xdr_buf(struct xdr_buf *buf, int base, u32 *obj)
868 status = read_bytes_from_xdr_buf(buf, base, &raw, sizeof(*obj));
875 /* If the netobj starting offset bytes from the start of xdr_buf is contained
876 * entirely in the head or the tail, set object to point to it; otherwise
877 * try to find space for it at the end of the tail, copy it there, and
878 * set obj to point to it. */
880 xdr_buf_read_netobj(struct xdr_buf *buf, struct xdr_netobj *obj, int offset)
882 u32 tail_offset = buf->head[0].iov_len + buf->page_len;
885 if (read_u32_from_xdr_buf(buf, offset, &obj->len))
887 obj_end_offset = offset + 4 + obj->len;
889 if (obj_end_offset <= buf->head[0].iov_len) {
890 /* The obj is contained entirely in the head: */
891 obj->data = buf->head[0].iov_base + offset + 4;
892 } else if (offset + 4 >= tail_offset) {
893 if (obj_end_offset - tail_offset
894 > buf->tail[0].iov_len)
896 /* The obj is contained entirely in the tail: */
897 obj->data = buf->tail[0].iov_base
898 + offset - tail_offset + 4;
900 /* use end of tail as storage for obj:
901 * (We don't copy to the beginning because then we'd have
902 * to worry about doing a potentially overlapping copy.
903 * This assumes the object is at most half the length of the
905 if (obj->len > buf->tail[0].iov_len)
907 obj->data = buf->tail[0].iov_base + buf->tail[0].iov_len -
909 if (read_bytes_from_xdr_buf(buf, offset + 4,
910 obj->data, obj->len))