2 * Back-end of the driver for virtual network devices. This portion of the
3 * driver exports a 'unified' network-device interface that can be accessed
4 * by any operating system that implements a compatible front end. A
5 * reference front-end implementation can be found in:
6 * drivers/net/xen-netfront.c
8 * Copyright (c) 2002-2005, K A Fraser
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License version 2
12 * as published by the Free Software Foundation; or, when distributed
13 * separately from the Linux kernel or incorporated into other
14 * software packages, subject to the following license:
16 * Permission is hereby granted, free of charge, to any person obtaining a copy
17 * of this source file (the "Software"), to deal in the Software without
18 * restriction, including without limitation the rights to use, copy, modify,
19 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
20 * and to permit persons to whom the Software is furnished to do so, subject to
21 * the following conditions:
23 * The above copyright notice and this permission notice shall be included in
24 * all copies or substantial portions of the Software.
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
27 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
28 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
29 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
30 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
31 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
37 #include <linux/kthread.h>
38 #include <linux/if_vlan.h>
39 #include <linux/udp.h>
44 #include <xen/events.h>
45 #include <xen/interface/memory.h>
47 #include <asm/xen/hypercall.h>
48 #include <asm/xen/page.h>
50 /* Provide an option to disable split event channels at load time as
51 * event channels are limited resource. Split event channels are
54 bool separate_tx_rx_irq = 1;
55 module_param(separate_tx_rx_irq, bool, 0644);
58 * This is the maximum slots a skb can have. If a guest sends a skb
59 * which exceeds this limit it is considered malicious.
61 #define FATAL_SKB_SLOTS_DEFAULT 20
62 static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT;
63 module_param(fatal_skb_slots, uint, 0444);
66 * To avoid confusion, we define XEN_NETBK_LEGACY_SLOTS_MAX indicating
67 * the maximum slots a valid packet can use. Now this value is defined
68 * to be XEN_NETIF_NR_SLOTS_MIN, which is supposed to be supported by
71 #define XEN_NETBK_LEGACY_SLOTS_MAX XEN_NETIF_NR_SLOTS_MIN
74 * If head != INVALID_PENDING_RING_IDX, it means this tx request is head of
75 * one or more merged tx requests, otherwise it is the continuation of
76 * previous tx request.
78 static inline int pending_tx_is_head(struct xenvif *vif, RING_IDX idx)
80 return vif->pending_tx_info[idx].head != INVALID_PENDING_RING_IDX;
83 static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx,
86 static void make_tx_response(struct xenvif *vif,
87 struct xen_netif_tx_request *txp,
90 static inline int tx_work_todo(struct xenvif *vif);
91 static inline int rx_work_todo(struct xenvif *vif);
93 static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
100 static inline unsigned long idx_to_pfn(struct xenvif *vif,
103 return page_to_pfn(vif->mmap_pages[idx]);
106 static inline unsigned long idx_to_kaddr(struct xenvif *vif,
109 return (unsigned long)pfn_to_kaddr(idx_to_pfn(vif, idx));
112 /* This is a miniumum size for the linear area to avoid lots of
113 * calls to __pskb_pull_tail() as we set up checksum offsets. The
114 * value 128 was chosen as it covers all IPv4 and most likely
117 #define PKT_PROT_LEN 128
119 static u16 frag_get_pending_idx(skb_frag_t *frag)
121 return (u16)frag->page_offset;
124 static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx)
126 frag->page_offset = pending_idx;
129 static inline pending_ring_idx_t pending_index(unsigned i)
131 return i & (MAX_PENDING_REQS-1);
134 static inline pending_ring_idx_t nr_pending_reqs(struct xenvif *vif)
136 return MAX_PENDING_REQS -
137 vif->pending_prod + vif->pending_cons;
140 bool xenvif_rx_ring_slots_available(struct xenvif *vif, int needed)
145 prod = vif->rx.sring->req_prod;
146 cons = vif->rx.req_cons;
148 if (prod - cons >= needed)
151 vif->rx.sring->req_event = prod + 1;
153 /* Make sure event is visible before we check prod
157 } while (vif->rx.sring->req_prod != prod);
163 * Returns true if we should start a new receive buffer instead of
164 * adding 'size' bytes to a buffer which currently contains 'offset'
167 static bool start_new_rx_buffer(int offset, unsigned long size, int head)
169 /* simple case: we have completely filled the current buffer. */
170 if (offset == MAX_BUFFER_OFFSET)
174 * complex case: start a fresh buffer if the current frag
175 * would overflow the current buffer but only if:
176 * (i) this frag would fit completely in the next buffer
177 * and (ii) there is already some data in the current buffer
178 * and (iii) this is not the head buffer.
181 * - (i) stops us splitting a frag into two copies
182 * unless the frag is too large for a single buffer.
183 * - (ii) stops us from leaving a buffer pointlessly empty.
184 * - (iii) stops us leaving the first buffer
185 * empty. Strictly speaking this is already covered
186 * by (ii) but is explicitly checked because
187 * netfront relies on the first buffer being
188 * non-empty and can crash otherwise.
190 * This means we will effectively linearise small
191 * frags but do not needlessly split large buffers
192 * into multiple copies tend to give large frags their
193 * own buffers as before.
195 BUG_ON(size > MAX_BUFFER_OFFSET);
196 if ((offset + size > MAX_BUFFER_OFFSET) && offset && !head)
202 struct netrx_pending_operations {
203 unsigned copy_prod, copy_cons;
204 unsigned meta_prod, meta_cons;
205 struct gnttab_copy *copy;
206 struct xenvif_rx_meta *meta;
208 grant_ref_t copy_gref;
211 static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif *vif,
212 struct netrx_pending_operations *npo)
214 struct xenvif_rx_meta *meta;
215 struct xen_netif_rx_request *req;
217 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
219 meta = npo->meta + npo->meta_prod++;
220 meta->gso_type = XEN_NETIF_GSO_TYPE_NONE;
226 npo->copy_gref = req->gref;
232 * Set up the grant operations for this fragment. If it's a flipping
233 * interface, we also set up the unmap request from here.
235 static void xenvif_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb,
236 struct netrx_pending_operations *npo,
237 struct page *page, unsigned long size,
238 unsigned long offset, int *head)
240 struct gnttab_copy *copy_gop;
241 struct xenvif_rx_meta *meta;
243 int gso_type = XEN_NETIF_GSO_TYPE_NONE;
245 /* Data must not cross a page boundary. */
246 BUG_ON(size + offset > PAGE_SIZE<<compound_order(page));
248 meta = npo->meta + npo->meta_prod - 1;
250 /* Skip unused frames from start of page */
251 page += offset >> PAGE_SHIFT;
252 offset &= ~PAGE_MASK;
255 BUG_ON(offset >= PAGE_SIZE);
256 BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET);
258 bytes = PAGE_SIZE - offset;
263 if (start_new_rx_buffer(npo->copy_off, bytes, *head)) {
265 * Netfront requires there to be some data in the head
270 meta = get_next_rx_buffer(vif, npo);
273 if (npo->copy_off + bytes > MAX_BUFFER_OFFSET)
274 bytes = MAX_BUFFER_OFFSET - npo->copy_off;
276 copy_gop = npo->copy + npo->copy_prod++;
277 copy_gop->flags = GNTCOPY_dest_gref;
278 copy_gop->len = bytes;
280 copy_gop->source.domid = DOMID_SELF;
281 copy_gop->source.u.gmfn = virt_to_mfn(page_address(page));
282 copy_gop->source.offset = offset;
284 copy_gop->dest.domid = vif->domid;
285 copy_gop->dest.offset = npo->copy_off;
286 copy_gop->dest.u.ref = npo->copy_gref;
288 npo->copy_off += bytes;
295 if (offset == PAGE_SIZE && size) {
296 BUG_ON(!PageCompound(page));
301 /* Leave a gap for the GSO descriptor. */
302 if (skb_is_gso(skb)) {
303 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
304 gso_type = XEN_NETIF_GSO_TYPE_TCPV4;
305 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
306 gso_type = XEN_NETIF_GSO_TYPE_TCPV6;
309 if (*head && ((1 << gso_type) & vif->gso_mask))
312 *head = 0; /* There must be something in this buffer now. */
318 * Prepare an SKB to be transmitted to the frontend.
320 * This function is responsible for allocating grant operations, meta
323 * It returns the number of meta structures consumed. The number of
324 * ring slots used is always equal to the number of meta slots used
325 * plus the number of GSO descriptors used. Currently, we use either
326 * zero GSO descriptors (for non-GSO packets) or one descriptor (for
327 * frontend-side LRO).
329 static int xenvif_gop_skb(struct sk_buff *skb,
330 struct netrx_pending_operations *npo)
332 struct xenvif *vif = netdev_priv(skb->dev);
333 int nr_frags = skb_shinfo(skb)->nr_frags;
335 struct xen_netif_rx_request *req;
336 struct xenvif_rx_meta *meta;
342 old_meta_prod = npo->meta_prod;
344 gso_type = XEN_NETIF_GSO_TYPE_NONE;
345 if (skb_is_gso(skb)) {
346 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
347 gso_type = XEN_NETIF_GSO_TYPE_TCPV4;
348 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
349 gso_type = XEN_NETIF_GSO_TYPE_TCPV6;
352 /* Set up a GSO prefix descriptor, if necessary */
353 if ((1 << gso_type) & vif->gso_prefix_mask) {
354 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
355 meta = npo->meta + npo->meta_prod++;
356 meta->gso_type = gso_type;
357 meta->gso_size = skb_shinfo(skb)->gso_size;
362 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
363 meta = npo->meta + npo->meta_prod++;
365 if ((1 << gso_type) & vif->gso_mask) {
366 meta->gso_type = gso_type;
367 meta->gso_size = skb_shinfo(skb)->gso_size;
369 meta->gso_type = XEN_NETIF_GSO_TYPE_NONE;
376 npo->copy_gref = req->gref;
379 while (data < skb_tail_pointer(skb)) {
380 unsigned int offset = offset_in_page(data);
381 unsigned int len = PAGE_SIZE - offset;
383 if (data + len > skb_tail_pointer(skb))
384 len = skb_tail_pointer(skb) - data;
386 xenvif_gop_frag_copy(vif, skb, npo,
387 virt_to_page(data), len, offset, &head);
391 for (i = 0; i < nr_frags; i++) {
392 xenvif_gop_frag_copy(vif, skb, npo,
393 skb_frag_page(&skb_shinfo(skb)->frags[i]),
394 skb_frag_size(&skb_shinfo(skb)->frags[i]),
395 skb_shinfo(skb)->frags[i].page_offset,
399 return npo->meta_prod - old_meta_prod;
403 * This is a twin to xenvif_gop_skb. Assume that xenvif_gop_skb was
404 * used to set up the operations on the top of
405 * netrx_pending_operations, which have since been done. Check that
406 * they didn't give any errors and advance over them.
408 static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots,
409 struct netrx_pending_operations *npo)
411 struct gnttab_copy *copy_op;
412 int status = XEN_NETIF_RSP_OKAY;
415 for (i = 0; i < nr_meta_slots; i++) {
416 copy_op = npo->copy + npo->copy_cons++;
417 if (copy_op->status != GNTST_okay) {
419 "Bad status %d from copy to DOM%d.\n",
420 copy_op->status, vif->domid);
421 status = XEN_NETIF_RSP_ERROR;
428 static void xenvif_add_frag_responses(struct xenvif *vif, int status,
429 struct xenvif_rx_meta *meta,
433 unsigned long offset;
435 /* No fragments used */
436 if (nr_meta_slots <= 1)
441 for (i = 0; i < nr_meta_slots; i++) {
443 if (i == nr_meta_slots - 1)
446 flags = XEN_NETRXF_more_data;
449 make_rx_response(vif, meta[i].id, status, offset,
450 meta[i].size, flags);
454 struct skb_cb_overlay {
458 void xenvif_kick_thread(struct xenvif *vif)
463 static void xenvif_rx_action(struct xenvif *vif)
467 struct xen_netif_rx_response *resp;
468 struct sk_buff_head rxq;
472 unsigned long offset;
473 struct skb_cb_overlay *sco;
474 bool need_to_notify = false;
476 struct netrx_pending_operations npo = {
477 .copy = vif->grant_copy_op,
481 skb_queue_head_init(&rxq);
483 while ((skb = skb_dequeue(&vif->rx_queue)) != NULL) {
484 RING_IDX max_slots_needed;
487 /* We need a cheap worse case estimate for the number of
491 max_slots_needed = DIV_ROUND_UP(offset_in_page(skb->data) +
494 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
498 size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
499 offset = skb_shinfo(skb)->frags[i].page_offset;
501 /* For a worse-case estimate we need to factor in
502 * the fragment page offset as this will affect the
503 * number of times xenvif_gop_frag_copy() will
504 * call start_new_rx_buffer().
506 max_slots_needed += DIV_ROUND_UP(offset + size,
510 /* To avoid the estimate becoming too pessimal for some
511 * frontends that limit posted rx requests, cap the estimate
514 if (max_slots_needed > MAX_SKB_FRAGS)
515 max_slots_needed = MAX_SKB_FRAGS;
517 /* We may need one more slot for GSO metadata */
518 if (skb_is_gso(skb) &&
519 (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4 ||
520 skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6))
523 /* If the skb may not fit then bail out now */
524 if (!xenvif_rx_ring_slots_available(vif, max_slots_needed)) {
525 skb_queue_head(&vif->rx_queue, skb);
526 need_to_notify = true;
527 vif->rx_last_skb_slots = max_slots_needed;
530 vif->rx_last_skb_slots = 0;
532 sco = (struct skb_cb_overlay *)skb->cb;
533 sco->meta_slots_used = xenvif_gop_skb(skb, &npo);
534 BUG_ON(sco->meta_slots_used > max_slots_needed);
536 __skb_queue_tail(&rxq, skb);
539 BUG_ON(npo.meta_prod > ARRAY_SIZE(vif->meta));
544 BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS);
545 gnttab_batch_copy(vif->grant_copy_op, npo.copy_prod);
547 while ((skb = __skb_dequeue(&rxq)) != NULL) {
548 sco = (struct skb_cb_overlay *)skb->cb;
550 if ((1 << vif->meta[npo.meta_cons].gso_type) &
551 vif->gso_prefix_mask) {
552 resp = RING_GET_RESPONSE(&vif->rx,
553 vif->rx.rsp_prod_pvt++);
555 resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data;
557 resp->offset = vif->meta[npo.meta_cons].gso_size;
558 resp->id = vif->meta[npo.meta_cons].id;
559 resp->status = sco->meta_slots_used;
562 sco->meta_slots_used--;
566 vif->dev->stats.tx_bytes += skb->len;
567 vif->dev->stats.tx_packets++;
569 status = xenvif_check_gop(vif, sco->meta_slots_used, &npo);
571 if (sco->meta_slots_used == 1)
574 flags = XEN_NETRXF_more_data;
576 if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */
577 flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated;
578 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
579 /* remote but checksummed. */
580 flags |= XEN_NETRXF_data_validated;
583 resp = make_rx_response(vif, vif->meta[npo.meta_cons].id,
585 vif->meta[npo.meta_cons].size,
588 if ((1 << vif->meta[npo.meta_cons].gso_type) &
590 struct xen_netif_extra_info *gso =
591 (struct xen_netif_extra_info *)
592 RING_GET_RESPONSE(&vif->rx,
593 vif->rx.rsp_prod_pvt++);
595 resp->flags |= XEN_NETRXF_extra_info;
597 gso->u.gso.type = vif->meta[npo.meta_cons].gso_type;
598 gso->u.gso.size = vif->meta[npo.meta_cons].gso_size;
600 gso->u.gso.features = 0;
602 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
606 xenvif_add_frag_responses(vif, status,
607 vif->meta + npo.meta_cons + 1,
608 sco->meta_slots_used);
610 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret);
612 need_to_notify |= !!ret;
614 npo.meta_cons += sco->meta_slots_used;
620 notify_remote_via_irq(vif->rx_irq);
623 void xenvif_check_rx_xenvif(struct xenvif *vif)
627 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do);
630 napi_schedule(&vif->napi);
633 static void tx_add_credit(struct xenvif *vif)
635 unsigned long max_burst, max_credit;
638 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
639 * Otherwise the interface can seize up due to insufficient credit.
641 max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size;
642 max_burst = min(max_burst, 131072UL);
643 max_burst = max(max_burst, vif->credit_bytes);
645 /* Take care that adding a new chunk of credit doesn't wrap to zero. */
646 max_credit = vif->remaining_credit + vif->credit_bytes;
647 if (max_credit < vif->remaining_credit)
648 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
650 vif->remaining_credit = min(max_credit, max_burst);
653 static void tx_credit_callback(unsigned long data)
655 struct xenvif *vif = (struct xenvif *)data;
657 xenvif_check_rx_xenvif(vif);
660 static void xenvif_tx_err(struct xenvif *vif,
661 struct xen_netif_tx_request *txp, RING_IDX end)
663 RING_IDX cons = vif->tx.req_cons;
666 make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR);
669 txp = RING_GET_REQUEST(&vif->tx, cons++);
671 vif->tx.req_cons = cons;
674 static void xenvif_fatal_tx_err(struct xenvif *vif)
676 netdev_err(vif->dev, "fatal error; disabling device\n");
677 xenvif_carrier_off(vif);
680 static int xenvif_count_requests(struct xenvif *vif,
681 struct xen_netif_tx_request *first,
682 struct xen_netif_tx_request *txp,
685 RING_IDX cons = vif->tx.req_cons;
690 if (!(first->flags & XEN_NETTXF_more_data))
694 struct xen_netif_tx_request dropped_tx = { 0 };
696 if (slots >= work_to_do) {
698 "Asked for %d slots but exceeds this limit\n",
700 xenvif_fatal_tx_err(vif);
704 /* This guest is really using too many slots and
705 * considered malicious.
707 if (unlikely(slots >= fatal_skb_slots)) {
709 "Malicious frontend using %d slots, threshold %u\n",
710 slots, fatal_skb_slots);
711 xenvif_fatal_tx_err(vif);
715 /* Xen network protocol had implicit dependency on
716 * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to
717 * the historical MAX_SKB_FRAGS value 18 to honor the
718 * same behavior as before. Any packet using more than
719 * 18 slots but less than fatal_skb_slots slots is
722 if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) {
725 "Too many slots (%d) exceeding limit (%d), dropping packet\n",
726 slots, XEN_NETBK_LEGACY_SLOTS_MAX);
733 memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + slots),
736 /* If the guest submitted a frame >= 64 KiB then
737 * first->size overflowed and following slots will
738 * appear to be larger than the frame.
740 * This cannot be fatal error as there are buggy
741 * frontends that do this.
743 * Consume all slots and drop the packet.
745 if (!drop_err && txp->size > first->size) {
748 "Invalid tx request, slot size %u > remaining size %u\n",
749 txp->size, first->size);
753 first->size -= txp->size;
756 if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) {
757 netdev_err(vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n",
758 txp->offset, txp->size);
759 xenvif_fatal_tx_err(vif);
763 more_data = txp->flags & XEN_NETTXF_more_data;
771 xenvif_tx_err(vif, first, cons + slots);
778 static struct page *xenvif_alloc_page(struct xenvif *vif,
783 page = alloc_page(GFP_ATOMIC|__GFP_COLD);
786 vif->mmap_pages[pending_idx] = page;
791 static struct gnttab_copy *xenvif_get_requests(struct xenvif *vif,
793 struct xen_netif_tx_request *txp,
794 struct gnttab_copy *gop)
796 struct skb_shared_info *shinfo = skb_shinfo(skb);
797 skb_frag_t *frags = shinfo->frags;
798 u16 pending_idx = *((u16 *)skb->data);
802 pending_ring_idx_t index, start_idx = 0;
804 unsigned int nr_slots;
805 struct pending_tx_info *first = NULL;
807 /* At this point shinfo->nr_frags is in fact the number of
808 * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX.
810 nr_slots = shinfo->nr_frags;
812 /* Skip first skb fragment if it is on same page as header fragment. */
813 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
815 /* Coalesce tx requests, at this point the packet passed in
816 * should be <= 64K. Any packets larger than 64K have been
817 * handled in xenvif_count_requests().
819 for (shinfo->nr_frags = slot = start; slot < nr_slots;
820 shinfo->nr_frags++) {
821 struct pending_tx_info *pending_tx_info =
822 vif->pending_tx_info;
824 page = alloc_page(GFP_ATOMIC|__GFP_COLD);
830 while (dst_offset < PAGE_SIZE && slot < nr_slots) {
831 gop->flags = GNTCOPY_source_gref;
833 gop->source.u.ref = txp->gref;
834 gop->source.domid = vif->domid;
835 gop->source.offset = txp->offset;
837 gop->dest.domid = DOMID_SELF;
839 gop->dest.offset = dst_offset;
840 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
842 if (dst_offset + txp->size > PAGE_SIZE) {
843 /* This page can only merge a portion
844 * of tx request. Do not increment any
845 * pointer / counter here. The txp
846 * will be dealt with in future
847 * rounds, eventually hitting the
850 gop->len = PAGE_SIZE - dst_offset;
851 txp->offset += gop->len;
852 txp->size -= gop->len;
853 dst_offset += gop->len; /* quit loop */
855 /* This tx request can be merged in the page */
856 gop->len = txp->size;
857 dst_offset += gop->len;
859 index = pending_index(vif->pending_cons++);
861 pending_idx = vif->pending_ring[index];
863 memcpy(&pending_tx_info[pending_idx].req, txp,
866 /* Poison these fields, corresponding
867 * fields for head tx req will be set
868 * to correct values after the loop.
870 vif->mmap_pages[pending_idx] = (void *)(~0UL);
871 pending_tx_info[pending_idx].head =
872 INVALID_PENDING_RING_IDX;
875 first = &pending_tx_info[pending_idx];
877 head_idx = pending_idx;
887 first->req.offset = 0;
888 first->req.size = dst_offset;
889 first->head = start_idx;
890 vif->mmap_pages[head_idx] = page;
891 frag_set_pending_idx(&frags[shinfo->nr_frags], head_idx);
894 BUG_ON(shinfo->nr_frags > MAX_SKB_FRAGS);
898 /* Unwind, freeing all pages and sending error responses. */
899 while (shinfo->nr_frags-- > start) {
900 xenvif_idx_release(vif,
901 frag_get_pending_idx(&frags[shinfo->nr_frags]),
902 XEN_NETIF_RSP_ERROR);
904 /* The head too, if necessary. */
906 xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR);
911 static int xenvif_tx_check_gop(struct xenvif *vif,
913 struct gnttab_copy **gopp)
915 struct gnttab_copy *gop = *gopp;
916 u16 pending_idx = *((u16 *)skb->data);
917 struct skb_shared_info *shinfo = skb_shinfo(skb);
918 struct pending_tx_info *tx_info;
919 int nr_frags = shinfo->nr_frags;
921 u16 peek; /* peek into next tx request */
923 /* Check status of header. */
926 xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR);
928 /* Skip first skb fragment if it is on same page as header fragment. */
929 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
931 for (i = start; i < nr_frags; i++) {
933 pending_ring_idx_t head;
935 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
936 tx_info = &vif->pending_tx_info[pending_idx];
937 head = tx_info->head;
939 /* Check error status: if okay then remember grant handle. */
941 newerr = (++gop)->status;
944 peek = vif->pending_ring[pending_index(++head)];
945 } while (!pending_tx_is_head(vif, peek));
947 if (likely(!newerr)) {
948 /* Had a previous error? Invalidate this fragment. */
950 xenvif_idx_release(vif, pending_idx,
955 /* Error on this fragment: respond to client with an error. */
956 xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_ERROR);
958 /* Not the first error? Preceding frags already invalidated. */
962 /* First error: invalidate header and preceding fragments. */
963 pending_idx = *((u16 *)skb->data);
964 xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY);
965 for (j = start; j < i; j++) {
966 pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
967 xenvif_idx_release(vif, pending_idx,
971 /* Remember the error: invalidate all subsequent fragments. */
979 static void xenvif_fill_frags(struct xenvif *vif, struct sk_buff *skb)
981 struct skb_shared_info *shinfo = skb_shinfo(skb);
982 int nr_frags = shinfo->nr_frags;
985 for (i = 0; i < nr_frags; i++) {
986 skb_frag_t *frag = shinfo->frags + i;
987 struct xen_netif_tx_request *txp;
991 pending_idx = frag_get_pending_idx(frag);
993 txp = &vif->pending_tx_info[pending_idx].req;
994 page = virt_to_page(idx_to_kaddr(vif, pending_idx));
995 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
996 skb->len += txp->size;
997 skb->data_len += txp->size;
998 skb->truesize += txp->size;
1000 /* Take an extra reference to offset xenvif_idx_release */
1001 get_page(vif->mmap_pages[pending_idx]);
1002 xenvif_idx_release(vif, pending_idx, XEN_NETIF_RSP_OKAY);
1006 static int xenvif_get_extras(struct xenvif *vif,
1007 struct xen_netif_extra_info *extras,
1010 struct xen_netif_extra_info extra;
1011 RING_IDX cons = vif->tx.req_cons;
1014 if (unlikely(work_to_do-- <= 0)) {
1015 netdev_err(vif->dev, "Missing extra info\n");
1016 xenvif_fatal_tx_err(vif);
1020 memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons),
1022 if (unlikely(!extra.type ||
1023 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1024 vif->tx.req_cons = ++cons;
1025 netdev_err(vif->dev,
1026 "Invalid extra type: %d\n", extra.type);
1027 xenvif_fatal_tx_err(vif);
1031 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
1032 vif->tx.req_cons = ++cons;
1033 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
1038 static int xenvif_set_skb_gso(struct xenvif *vif,
1039 struct sk_buff *skb,
1040 struct xen_netif_extra_info *gso)
1042 if (!gso->u.gso.size) {
1043 netdev_err(vif->dev, "GSO size must not be zero.\n");
1044 xenvif_fatal_tx_err(vif);
1048 switch (gso->u.gso.type) {
1049 case XEN_NETIF_GSO_TYPE_TCPV4:
1050 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1052 case XEN_NETIF_GSO_TYPE_TCPV6:
1053 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
1056 netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
1057 xenvif_fatal_tx_err(vif);
1061 skb_shinfo(skb)->gso_size = gso->u.gso.size;
1062 /* gso_segs will be calculated later */
1067 static int checksum_setup(struct xenvif *vif, struct sk_buff *skb)
1069 bool recalculate_partial_csum = false;
1071 /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1072 * peers can fail to set NETRXF_csum_blank when sending a GSO
1073 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1074 * recalculate the partial checksum.
1076 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1077 vif->rx_gso_checksum_fixup++;
1078 skb->ip_summed = CHECKSUM_PARTIAL;
1079 recalculate_partial_csum = true;
1082 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1083 if (skb->ip_summed != CHECKSUM_PARTIAL)
1086 return skb_checksum_setup(skb, recalculate_partial_csum);
1089 static bool tx_credit_exceeded(struct xenvif *vif, unsigned size)
1091 u64 now = get_jiffies_64();
1092 u64 next_credit = vif->credit_window_start +
1093 msecs_to_jiffies(vif->credit_usec / 1000);
1095 /* Timer could already be pending in rare cases. */
1096 if (timer_pending(&vif->credit_timeout))
1099 /* Passed the point where we can replenish credit? */
1100 if (time_after_eq64(now, next_credit)) {
1101 vif->credit_window_start = now;
1105 /* Still too big to send right now? Set a callback. */
1106 if (size > vif->remaining_credit) {
1107 vif->credit_timeout.data =
1109 vif->credit_timeout.function =
1111 mod_timer(&vif->credit_timeout,
1113 vif->credit_window_start = next_credit;
1121 static unsigned xenvif_tx_build_gops(struct xenvif *vif, int budget)
1123 struct gnttab_copy *gop = vif->tx_copy_ops, *request_gop;
1124 struct sk_buff *skb;
1127 while ((nr_pending_reqs(vif) + XEN_NETBK_LEGACY_SLOTS_MAX
1128 < MAX_PENDING_REQS) &&
1129 (skb_queue_len(&vif->tx_queue) < budget)) {
1130 struct xen_netif_tx_request txreq;
1131 struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX];
1133 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
1137 unsigned int data_len;
1138 pending_ring_idx_t index;
1140 if (vif->tx.sring->req_prod - vif->tx.req_cons >
1141 XEN_NETIF_TX_RING_SIZE) {
1142 netdev_err(vif->dev,
1143 "Impossible number of requests. "
1144 "req_prod %d, req_cons %d, size %ld\n",
1145 vif->tx.sring->req_prod, vif->tx.req_cons,
1146 XEN_NETIF_TX_RING_SIZE);
1147 xenvif_fatal_tx_err(vif);
1151 work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&vif->tx);
1155 idx = vif->tx.req_cons;
1156 rmb(); /* Ensure that we see the request before we copy it. */
1157 memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq));
1159 /* Credit-based scheduling. */
1160 if (txreq.size > vif->remaining_credit &&
1161 tx_credit_exceeded(vif, txreq.size))
1164 vif->remaining_credit -= txreq.size;
1167 vif->tx.req_cons = ++idx;
1169 memset(extras, 0, sizeof(extras));
1170 if (txreq.flags & XEN_NETTXF_extra_info) {
1171 work_to_do = xenvif_get_extras(vif, extras,
1173 idx = vif->tx.req_cons;
1174 if (unlikely(work_to_do < 0))
1178 ret = xenvif_count_requests(vif, &txreq, txfrags, work_to_do);
1179 if (unlikely(ret < 0))
1184 if (unlikely(txreq.size < ETH_HLEN)) {
1185 netdev_dbg(vif->dev,
1186 "Bad packet size: %d\n", txreq.size);
1187 xenvif_tx_err(vif, &txreq, idx);
1191 /* No crossing a page as the payload mustn't fragment. */
1192 if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) {
1193 netdev_err(vif->dev,
1194 "txreq.offset: %x, size: %u, end: %lu\n",
1195 txreq.offset, txreq.size,
1196 (txreq.offset&~PAGE_MASK) + txreq.size);
1197 xenvif_fatal_tx_err(vif);
1201 index = pending_index(vif->pending_cons);
1202 pending_idx = vif->pending_ring[index];
1204 data_len = (txreq.size > PKT_PROT_LEN &&
1205 ret < XEN_NETBK_LEGACY_SLOTS_MAX) ?
1206 PKT_PROT_LEN : txreq.size;
1208 skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN,
1209 GFP_ATOMIC | __GFP_NOWARN);
1210 if (unlikely(skb == NULL)) {
1211 netdev_dbg(vif->dev,
1212 "Can't allocate a skb in start_xmit.\n");
1213 xenvif_tx_err(vif, &txreq, idx);
1217 /* Packets passed to netif_rx() must have some headroom. */
1218 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1220 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1221 struct xen_netif_extra_info *gso;
1222 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1224 if (xenvif_set_skb_gso(vif, skb, gso)) {
1225 /* Failure in xenvif_set_skb_gso is fatal. */
1231 /* XXX could copy straight to head */
1232 page = xenvif_alloc_page(vif, pending_idx);
1235 xenvif_tx_err(vif, &txreq, idx);
1239 gop->source.u.ref = txreq.gref;
1240 gop->source.domid = vif->domid;
1241 gop->source.offset = txreq.offset;
1243 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
1244 gop->dest.domid = DOMID_SELF;
1245 gop->dest.offset = txreq.offset;
1247 gop->len = txreq.size;
1248 gop->flags = GNTCOPY_source_gref;
1252 memcpy(&vif->pending_tx_info[pending_idx].req,
1253 &txreq, sizeof(txreq));
1254 vif->pending_tx_info[pending_idx].head = index;
1255 *((u16 *)skb->data) = pending_idx;
1257 __skb_put(skb, data_len);
1259 skb_shinfo(skb)->nr_frags = ret;
1260 if (data_len < txreq.size) {
1261 skb_shinfo(skb)->nr_frags++;
1262 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1265 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1266 INVALID_PENDING_IDX);
1269 vif->pending_cons++;
1271 request_gop = xenvif_get_requests(vif, skb, txfrags, gop);
1272 if (request_gop == NULL) {
1274 xenvif_tx_err(vif, &txreq, idx);
1279 __skb_queue_tail(&vif->tx_queue, skb);
1281 vif->tx.req_cons = idx;
1283 if ((gop-vif->tx_copy_ops) >= ARRAY_SIZE(vif->tx_copy_ops))
1287 return gop - vif->tx_copy_ops;
1291 static int xenvif_tx_submit(struct xenvif *vif)
1293 struct gnttab_copy *gop = vif->tx_copy_ops;
1294 struct sk_buff *skb;
1297 while ((skb = __skb_dequeue(&vif->tx_queue)) != NULL) {
1298 struct xen_netif_tx_request *txp;
1302 pending_idx = *((u16 *)skb->data);
1303 txp = &vif->pending_tx_info[pending_idx].req;
1305 /* Check the remap error code. */
1306 if (unlikely(xenvif_tx_check_gop(vif, skb, &gop))) {
1307 netdev_dbg(vif->dev, "netback grant failed.\n");
1308 skb_shinfo(skb)->nr_frags = 0;
1313 data_len = skb->len;
1315 (void *)(idx_to_kaddr(vif, pending_idx)|txp->offset),
1317 if (data_len < txp->size) {
1318 /* Append the packet payload as a fragment. */
1319 txp->offset += data_len;
1320 txp->size -= data_len;
1322 /* Schedule a response immediately. */
1323 xenvif_idx_release(vif, pending_idx,
1324 XEN_NETIF_RSP_OKAY);
1327 if (txp->flags & XEN_NETTXF_csum_blank)
1328 skb->ip_summed = CHECKSUM_PARTIAL;
1329 else if (txp->flags & XEN_NETTXF_data_validated)
1330 skb->ip_summed = CHECKSUM_UNNECESSARY;
1332 xenvif_fill_frags(vif, skb);
1334 if (skb_is_nonlinear(skb) && skb_headlen(skb) < PKT_PROT_LEN) {
1335 int target = min_t(int, skb->len, PKT_PROT_LEN);
1336 __pskb_pull_tail(skb, target - skb_headlen(skb));
1339 skb->dev = vif->dev;
1340 skb->protocol = eth_type_trans(skb, skb->dev);
1341 skb_reset_network_header(skb);
1343 if (checksum_setup(vif, skb)) {
1344 netdev_dbg(vif->dev,
1345 "Can't setup checksum in net_tx_action\n");
1350 skb_probe_transport_header(skb, 0);
1352 /* If the packet is GSO then we will have just set up the
1353 * transport header offset in checksum_setup so it's now
1354 * straightforward to calculate gso_segs.
1356 if (skb_is_gso(skb)) {
1357 int mss = skb_shinfo(skb)->gso_size;
1358 int hdrlen = skb_transport_header(skb) -
1359 skb_mac_header(skb) +
1362 skb_shinfo(skb)->gso_segs =
1363 DIV_ROUND_UP(skb->len - hdrlen, mss);
1366 vif->dev->stats.rx_bytes += skb->len;
1367 vif->dev->stats.rx_packets++;
1371 netif_receive_skb(skb);
1377 /* Called after netfront has transmitted */
1378 int xenvif_tx_action(struct xenvif *vif, int budget)
1383 if (unlikely(!tx_work_todo(vif)))
1386 nr_gops = xenvif_tx_build_gops(vif, budget);
1391 gnttab_batch_copy(vif->tx_copy_ops, nr_gops);
1393 work_done = xenvif_tx_submit(vif);
1398 static void xenvif_idx_release(struct xenvif *vif, u16 pending_idx,
1401 struct pending_tx_info *pending_tx_info;
1402 pending_ring_idx_t head;
1403 u16 peek; /* peek into next tx request */
1405 BUG_ON(vif->mmap_pages[pending_idx] == (void *)(~0UL));
1407 /* Already complete? */
1408 if (vif->mmap_pages[pending_idx] == NULL)
1411 pending_tx_info = &vif->pending_tx_info[pending_idx];
1413 head = pending_tx_info->head;
1415 BUG_ON(!pending_tx_is_head(vif, head));
1416 BUG_ON(vif->pending_ring[pending_index(head)] != pending_idx);
1419 pending_ring_idx_t index;
1420 pending_ring_idx_t idx = pending_index(head);
1421 u16 info_idx = vif->pending_ring[idx];
1423 pending_tx_info = &vif->pending_tx_info[info_idx];
1424 make_tx_response(vif, &pending_tx_info->req, status);
1426 /* Setting any number other than
1427 * INVALID_PENDING_RING_IDX indicates this slot is
1428 * starting a new packet / ending a previous packet.
1430 pending_tx_info->head = 0;
1432 index = pending_index(vif->pending_prod++);
1433 vif->pending_ring[index] = vif->pending_ring[info_idx];
1435 peek = vif->pending_ring[pending_index(++head)];
1437 } while (!pending_tx_is_head(vif, peek));
1439 put_page(vif->mmap_pages[pending_idx]);
1440 vif->mmap_pages[pending_idx] = NULL;
1444 static void make_tx_response(struct xenvif *vif,
1445 struct xen_netif_tx_request *txp,
1448 RING_IDX i = vif->tx.rsp_prod_pvt;
1449 struct xen_netif_tx_response *resp;
1452 resp = RING_GET_RESPONSE(&vif->tx, i);
1456 if (txp->flags & XEN_NETTXF_extra_info)
1457 RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1459 vif->tx.rsp_prod_pvt = ++i;
1460 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify);
1462 notify_remote_via_irq(vif->tx_irq);
1465 static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
1472 RING_IDX i = vif->rx.rsp_prod_pvt;
1473 struct xen_netif_rx_response *resp;
1475 resp = RING_GET_RESPONSE(&vif->rx, i);
1476 resp->offset = offset;
1477 resp->flags = flags;
1479 resp->status = (s16)size;
1481 resp->status = (s16)st;
1483 vif->rx.rsp_prod_pvt = ++i;
1488 static inline int rx_work_todo(struct xenvif *vif)
1490 return !skb_queue_empty(&vif->rx_queue) &&
1491 xenvif_rx_ring_slots_available(vif, vif->rx_last_skb_slots);
1494 static inline int tx_work_todo(struct xenvif *vif)
1497 if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->tx)) &&
1498 (nr_pending_reqs(vif) + XEN_NETBK_LEGACY_SLOTS_MAX
1499 < MAX_PENDING_REQS))
1505 void xenvif_unmap_frontend_rings(struct xenvif *vif)
1508 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1511 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1515 int xenvif_map_frontend_rings(struct xenvif *vif,
1516 grant_ref_t tx_ring_ref,
1517 grant_ref_t rx_ring_ref)
1520 struct xen_netif_tx_sring *txs;
1521 struct xen_netif_rx_sring *rxs;
1525 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1526 tx_ring_ref, &addr);
1530 txs = (struct xen_netif_tx_sring *)addr;
1531 BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE);
1533 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1534 rx_ring_ref, &addr);
1538 rxs = (struct xen_netif_rx_sring *)addr;
1539 BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE);
1544 xenvif_unmap_frontend_rings(vif);
1548 void xenvif_stop_queue(struct xenvif *vif)
1550 if (!vif->can_queue)
1553 netif_stop_queue(vif->dev);
1556 static void xenvif_start_queue(struct xenvif *vif)
1558 if (xenvif_schedulable(vif))
1559 netif_wake_queue(vif->dev);
1562 int xenvif_kthread(void *data)
1564 struct xenvif *vif = data;
1565 struct sk_buff *skb;
1567 while (!kthread_should_stop()) {
1568 wait_event_interruptible(vif->wq,
1569 rx_work_todo(vif) ||
1570 kthread_should_stop());
1571 if (kthread_should_stop())
1574 if (!skb_queue_empty(&vif->rx_queue))
1575 xenvif_rx_action(vif);
1577 if (skb_queue_empty(&vif->rx_queue) &&
1578 netif_queue_stopped(vif->dev))
1579 xenvif_start_queue(vif);
1584 /* Bin any remaining skbs */
1585 while ((skb = skb_dequeue(&vif->rx_queue)) != NULL)
1591 static int __init netback_init(void)
1598 if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) {
1599 pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n",
1600 fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX);
1601 fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX;
1604 rc = xenvif_xenbus_init();
1614 module_init(netback_init);
1616 static void __exit netback_fini(void)
1618 xenvif_xenbus_fini();
1620 module_exit(netback_fini);
1622 MODULE_LICENSE("Dual BSD/GPL");
1623 MODULE_ALIAS("xen-backend:vif");