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
73 typedef unsigned int pending_ring_idx_t;
74 #define INVALID_PENDING_RING_IDX (~0U)
76 struct pending_tx_info {
77 struct xen_netif_tx_request req; /* coalesced tx request */
79 pending_ring_idx_t head; /* head != INVALID_PENDING_RING_IDX
80 * if it is head of one or more tx
85 struct netbk_rx_meta {
91 #define MAX_PENDING_REQS 256
93 /* Discriminate from any valid pending_idx value. */
94 #define INVALID_PENDING_IDX 0xFFFF
96 #define MAX_BUFFER_OFFSET PAGE_SIZE
98 /* extra field used in struct page */
101 #if BITS_PER_LONG < 64
103 #define GROUP_WIDTH (BITS_PER_LONG - IDX_WIDTH)
104 unsigned int group:GROUP_WIDTH;
105 unsigned int idx:IDX_WIDTH;
107 unsigned int group, idx;
114 wait_queue_head_t wq;
115 struct task_struct *task;
117 struct sk_buff_head rx_queue;
118 struct sk_buff_head tx_queue;
120 struct timer_list net_timer;
122 struct page *mmap_pages[MAX_PENDING_REQS];
124 pending_ring_idx_t pending_prod;
125 pending_ring_idx_t pending_cons;
126 struct list_head net_schedule_list;
128 /* Protect the net_schedule_list in netif. */
129 spinlock_t net_schedule_list_lock;
131 atomic_t netfront_count;
133 struct pending_tx_info pending_tx_info[MAX_PENDING_REQS];
134 /* Coalescing tx requests before copying makes number of grant
135 * copy ops greater or equal to number of slots required. In
136 * worst case a tx request consumes 2 gnttab_copy.
138 struct gnttab_copy tx_copy_ops[2*MAX_PENDING_REQS];
140 u16 pending_ring[MAX_PENDING_REQS];
143 * Given MAX_BUFFER_OFFSET of 4096 the worst case is that each
144 * head/fragment page uses 2 copy operations because it
145 * straddles two buffers in the frontend.
147 struct gnttab_copy grant_copy_op[2*XEN_NETIF_RX_RING_SIZE];
148 struct netbk_rx_meta meta[2*XEN_NETIF_RX_RING_SIZE];
151 static struct xen_netbk *xen_netbk;
152 static int xen_netbk_group_nr;
155 * If head != INVALID_PENDING_RING_IDX, it means this tx request is head of
156 * one or more merged tx requests, otherwise it is the continuation of
157 * previous tx request.
159 static inline int pending_tx_is_head(struct xen_netbk *netbk, RING_IDX idx)
161 return netbk->pending_tx_info[idx].head != INVALID_PENDING_RING_IDX;
164 void xen_netbk_add_xenvif(struct xenvif *vif)
167 int min_netfront_count;
169 struct xen_netbk *netbk;
171 min_netfront_count = atomic_read(&xen_netbk[0].netfront_count);
172 for (i = 0; i < xen_netbk_group_nr; i++) {
173 int netfront_count = atomic_read(&xen_netbk[i].netfront_count);
174 if (netfront_count < min_netfront_count) {
176 min_netfront_count = netfront_count;
180 netbk = &xen_netbk[min_group];
183 atomic_inc(&netbk->netfront_count);
186 void xen_netbk_remove_xenvif(struct xenvif *vif)
188 struct xen_netbk *netbk = vif->netbk;
190 atomic_dec(&netbk->netfront_count);
193 static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx,
195 static void make_tx_response(struct xenvif *vif,
196 struct xen_netif_tx_request *txp,
198 static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
205 static inline unsigned long idx_to_pfn(struct xen_netbk *netbk,
208 return page_to_pfn(netbk->mmap_pages[idx]);
211 static inline unsigned long idx_to_kaddr(struct xen_netbk *netbk,
214 return (unsigned long)pfn_to_kaddr(idx_to_pfn(netbk, idx));
217 /* extra field used in struct page */
218 static inline void set_page_ext(struct page *pg, struct xen_netbk *netbk,
221 unsigned int group = netbk - xen_netbk;
222 union page_ext ext = { .e = { .group = group + 1, .idx = idx } };
224 BUILD_BUG_ON(sizeof(ext) > sizeof(ext.mapping));
225 pg->mapping = ext.mapping;
228 static int get_page_ext(struct page *pg,
229 unsigned int *pgroup, unsigned int *pidx)
231 union page_ext ext = { .mapping = pg->mapping };
232 struct xen_netbk *netbk;
233 unsigned int group, idx;
235 group = ext.e.group - 1;
237 if (group < 0 || group >= xen_netbk_group_nr)
240 netbk = &xen_netbk[group];
244 if ((idx < 0) || (idx >= MAX_PENDING_REQS))
247 if (netbk->mmap_pages[idx] != pg)
257 * This is the amount of packet we copy rather than map, so that the
258 * guest can't fiddle with the contents of the headers while we do
259 * packet processing on them (netfilter, routing, etc).
261 #define PKT_PROT_LEN (ETH_HLEN + \
263 sizeof(struct iphdr) + MAX_IPOPTLEN + \
264 sizeof(struct tcphdr) + MAX_TCP_OPTION_SPACE)
266 static u16 frag_get_pending_idx(skb_frag_t *frag)
268 return (u16)frag->page_offset;
271 static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx)
273 frag->page_offset = pending_idx;
276 static inline pending_ring_idx_t pending_index(unsigned i)
278 return i & (MAX_PENDING_REQS-1);
281 static inline pending_ring_idx_t nr_pending_reqs(struct xen_netbk *netbk)
283 return MAX_PENDING_REQS -
284 netbk->pending_prod + netbk->pending_cons;
287 static void xen_netbk_kick_thread(struct xen_netbk *netbk)
292 static int max_required_rx_slots(struct xenvif *vif)
294 int max = DIV_ROUND_UP(vif->dev->mtu, PAGE_SIZE);
296 /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */
297 if (vif->can_sg || vif->gso || vif->gso_prefix)
298 max += MAX_SKB_FRAGS + 1; /* extra_info + frags */
303 int xen_netbk_rx_ring_full(struct xenvif *vif)
305 RING_IDX peek = vif->rx_req_cons_peek;
306 RING_IDX needed = max_required_rx_slots(vif);
308 return ((vif->rx.sring->req_prod - peek) < needed) ||
309 ((vif->rx.rsp_prod_pvt + XEN_NETIF_RX_RING_SIZE - peek) < needed);
312 int xen_netbk_must_stop_queue(struct xenvif *vif)
314 if (!xen_netbk_rx_ring_full(vif))
317 vif->rx.sring->req_event = vif->rx_req_cons_peek +
318 max_required_rx_slots(vif);
319 mb(); /* request notification /then/ check the queue */
321 return xen_netbk_rx_ring_full(vif);
325 * Returns true if we should start a new receive buffer instead of
326 * adding 'size' bytes to a buffer which currently contains 'offset'
329 static bool start_new_rx_buffer(int offset, unsigned long size, int head)
331 /* simple case: we have completely filled the current buffer. */
332 if (offset == MAX_BUFFER_OFFSET)
336 * complex case: start a fresh buffer if the current frag
337 * would overflow the current buffer but only if:
338 * (i) this frag would fit completely in the next buffer
339 * and (ii) there is already some data in the current buffer
340 * and (iii) this is not the head buffer.
343 * - (i) stops us splitting a frag into two copies
344 * unless the frag is too large for a single buffer.
345 * - (ii) stops us from leaving a buffer pointlessly empty.
346 * - (iii) stops us leaving the first buffer
347 * empty. Strictly speaking this is already covered
348 * by (ii) but is explicitly checked because
349 * netfront relies on the first buffer being
350 * non-empty and can crash otherwise.
352 * This means we will effectively linearise small
353 * frags but do not needlessly split large buffers
354 * into multiple copies tend to give large frags their
355 * own buffers as before.
357 if ((offset + size > MAX_BUFFER_OFFSET) &&
358 (size <= MAX_BUFFER_OFFSET) && offset && !head)
365 * Figure out how many ring slots we're going to need to send @skb to
366 * the guest. This function is essentially a dry run of
367 * netbk_gop_frag_copy.
369 unsigned int xen_netbk_count_skb_slots(struct xenvif *vif, struct sk_buff *skb)
374 count = DIV_ROUND_UP(skb_headlen(skb), PAGE_SIZE);
376 copy_off = skb_headlen(skb) % PAGE_SIZE;
378 if (skb_shinfo(skb)->gso_size)
381 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
382 unsigned long size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
383 unsigned long offset = skb_shinfo(skb)->frags[i].page_offset;
386 offset &= ~PAGE_MASK;
389 BUG_ON(offset >= PAGE_SIZE);
390 BUG_ON(copy_off > MAX_BUFFER_OFFSET);
392 bytes = PAGE_SIZE - offset;
397 if (start_new_rx_buffer(copy_off, bytes, 0)) {
402 if (copy_off + bytes > MAX_BUFFER_OFFSET)
403 bytes = MAX_BUFFER_OFFSET - copy_off;
410 if (offset == PAGE_SIZE)
417 struct netrx_pending_operations {
418 unsigned copy_prod, copy_cons;
419 unsigned meta_prod, meta_cons;
420 struct gnttab_copy *copy;
421 struct netbk_rx_meta *meta;
423 grant_ref_t copy_gref;
426 static struct netbk_rx_meta *get_next_rx_buffer(struct xenvif *vif,
427 struct netrx_pending_operations *npo)
429 struct netbk_rx_meta *meta;
430 struct xen_netif_rx_request *req;
432 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
434 meta = npo->meta + npo->meta_prod++;
440 npo->copy_gref = req->gref;
446 * Set up the grant operations for this fragment. If it's a flipping
447 * interface, we also set up the unmap request from here.
449 static void netbk_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb,
450 struct netrx_pending_operations *npo,
451 struct page *page, unsigned long size,
452 unsigned long offset, int *head)
454 struct gnttab_copy *copy_gop;
455 struct netbk_rx_meta *meta;
457 * These variables are used iff get_page_ext returns true,
458 * in which case they are guaranteed to be initialized.
460 unsigned int uninitialized_var(group), uninitialized_var(idx);
461 int foreign = get_page_ext(page, &group, &idx);
464 /* Data must not cross a page boundary. */
465 BUG_ON(size + offset > PAGE_SIZE<<compound_order(page));
467 meta = npo->meta + npo->meta_prod - 1;
469 /* Skip unused frames from start of page */
470 page += offset >> PAGE_SHIFT;
471 offset &= ~PAGE_MASK;
474 BUG_ON(offset >= PAGE_SIZE);
475 BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET);
477 bytes = PAGE_SIZE - offset;
482 if (start_new_rx_buffer(npo->copy_off, bytes, *head)) {
484 * Netfront requires there to be some data in the head
489 meta = get_next_rx_buffer(vif, npo);
492 if (npo->copy_off + bytes > MAX_BUFFER_OFFSET)
493 bytes = MAX_BUFFER_OFFSET - npo->copy_off;
495 copy_gop = npo->copy + npo->copy_prod++;
496 copy_gop->flags = GNTCOPY_dest_gref;
498 struct xen_netbk *netbk = &xen_netbk[group];
499 struct pending_tx_info *src_pend;
501 src_pend = &netbk->pending_tx_info[idx];
503 copy_gop->source.domid = src_pend->vif->domid;
504 copy_gop->source.u.ref = src_pend->req.gref;
505 copy_gop->flags |= GNTCOPY_source_gref;
507 void *vaddr = page_address(page);
508 copy_gop->source.domid = DOMID_SELF;
509 copy_gop->source.u.gmfn = virt_to_mfn(vaddr);
511 copy_gop->source.offset = offset;
512 copy_gop->dest.domid = vif->domid;
514 copy_gop->dest.offset = npo->copy_off;
515 copy_gop->dest.u.ref = npo->copy_gref;
516 copy_gop->len = bytes;
518 npo->copy_off += bytes;
525 if (offset == PAGE_SIZE && size) {
526 BUG_ON(!PageCompound(page));
531 /* Leave a gap for the GSO descriptor. */
532 if (*head && skb_shinfo(skb)->gso_size && !vif->gso_prefix)
535 *head = 0; /* There must be something in this buffer now. */
541 * Prepare an SKB to be transmitted to the frontend.
543 * This function is responsible for allocating grant operations, meta
546 * It returns the number of meta structures consumed. The number of
547 * ring slots used is always equal to the number of meta slots used
548 * plus the number of GSO descriptors used. Currently, we use either
549 * zero GSO descriptors (for non-GSO packets) or one descriptor (for
550 * frontend-side LRO).
552 static int netbk_gop_skb(struct sk_buff *skb,
553 struct netrx_pending_operations *npo)
555 struct xenvif *vif = netdev_priv(skb->dev);
556 int nr_frags = skb_shinfo(skb)->nr_frags;
558 struct xen_netif_rx_request *req;
559 struct netbk_rx_meta *meta;
564 old_meta_prod = npo->meta_prod;
566 /* Set up a GSO prefix descriptor, if necessary */
567 if (skb_shinfo(skb)->gso_size && vif->gso_prefix) {
568 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
569 meta = npo->meta + npo->meta_prod++;
570 meta->gso_size = skb_shinfo(skb)->gso_size;
575 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
576 meta = npo->meta + npo->meta_prod++;
578 if (!vif->gso_prefix)
579 meta->gso_size = skb_shinfo(skb)->gso_size;
586 npo->copy_gref = req->gref;
589 while (data < skb_tail_pointer(skb)) {
590 unsigned int offset = offset_in_page(data);
591 unsigned int len = PAGE_SIZE - offset;
593 if (data + len > skb_tail_pointer(skb))
594 len = skb_tail_pointer(skb) - data;
596 netbk_gop_frag_copy(vif, skb, npo,
597 virt_to_page(data), len, offset, &head);
601 for (i = 0; i < nr_frags; i++) {
602 netbk_gop_frag_copy(vif, skb, npo,
603 skb_frag_page(&skb_shinfo(skb)->frags[i]),
604 skb_frag_size(&skb_shinfo(skb)->frags[i]),
605 skb_shinfo(skb)->frags[i].page_offset,
609 return npo->meta_prod - old_meta_prod;
613 * This is a twin to netbk_gop_skb. Assume that netbk_gop_skb was
614 * used to set up the operations on the top of
615 * netrx_pending_operations, which have since been done. Check that
616 * they didn't give any errors and advance over them.
618 static int netbk_check_gop(struct xenvif *vif, int nr_meta_slots,
619 struct netrx_pending_operations *npo)
621 struct gnttab_copy *copy_op;
622 int status = XEN_NETIF_RSP_OKAY;
625 for (i = 0; i < nr_meta_slots; i++) {
626 copy_op = npo->copy + npo->copy_cons++;
627 if (copy_op->status != GNTST_okay) {
629 "Bad status %d from copy to DOM%d.\n",
630 copy_op->status, vif->domid);
631 status = XEN_NETIF_RSP_ERROR;
638 static void netbk_add_frag_responses(struct xenvif *vif, int status,
639 struct netbk_rx_meta *meta,
643 unsigned long offset;
645 /* No fragments used */
646 if (nr_meta_slots <= 1)
651 for (i = 0; i < nr_meta_slots; i++) {
653 if (i == nr_meta_slots - 1)
656 flags = XEN_NETRXF_more_data;
659 make_rx_response(vif, meta[i].id, status, offset,
660 meta[i].size, flags);
664 struct skb_cb_overlay {
668 static void xen_netbk_rx_action(struct xen_netbk *netbk)
670 struct xenvif *vif = NULL, *tmp;
673 struct xen_netif_rx_response *resp;
674 struct sk_buff_head rxq;
680 unsigned long offset;
681 struct skb_cb_overlay *sco;
683 struct netrx_pending_operations npo = {
684 .copy = netbk->grant_copy_op,
688 skb_queue_head_init(&rxq);
692 while ((skb = skb_dequeue(&netbk->rx_queue)) != NULL) {
693 vif = netdev_priv(skb->dev);
694 nr_frags = skb_shinfo(skb)->nr_frags;
696 sco = (struct skb_cb_overlay *)skb->cb;
697 sco->meta_slots_used = netbk_gop_skb(skb, &npo);
699 count += nr_frags + 1;
701 __skb_queue_tail(&rxq, skb);
703 /* Filled the batch queue? */
704 /* XXX FIXME: RX path dependent on MAX_SKB_FRAGS */
705 if (count + MAX_SKB_FRAGS >= XEN_NETIF_RX_RING_SIZE)
709 BUG_ON(npo.meta_prod > ARRAY_SIZE(netbk->meta));
714 BUG_ON(npo.copy_prod > ARRAY_SIZE(netbk->grant_copy_op));
715 gnttab_batch_copy(netbk->grant_copy_op, npo.copy_prod);
717 while ((skb = __skb_dequeue(&rxq)) != NULL) {
718 sco = (struct skb_cb_overlay *)skb->cb;
720 vif = netdev_priv(skb->dev);
722 if (netbk->meta[npo.meta_cons].gso_size && vif->gso_prefix) {
723 resp = RING_GET_RESPONSE(&vif->rx,
724 vif->rx.rsp_prod_pvt++);
726 resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data;
728 resp->offset = netbk->meta[npo.meta_cons].gso_size;
729 resp->id = netbk->meta[npo.meta_cons].id;
730 resp->status = sco->meta_slots_used;
733 sco->meta_slots_used--;
737 vif->dev->stats.tx_bytes += skb->len;
738 vif->dev->stats.tx_packets++;
740 status = netbk_check_gop(vif, sco->meta_slots_used, &npo);
742 if (sco->meta_slots_used == 1)
745 flags = XEN_NETRXF_more_data;
747 if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */
748 flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated;
749 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
750 /* remote but checksummed. */
751 flags |= XEN_NETRXF_data_validated;
754 resp = make_rx_response(vif, netbk->meta[npo.meta_cons].id,
756 netbk->meta[npo.meta_cons].size,
759 if (netbk->meta[npo.meta_cons].gso_size && !vif->gso_prefix) {
760 struct xen_netif_extra_info *gso =
761 (struct xen_netif_extra_info *)
762 RING_GET_RESPONSE(&vif->rx,
763 vif->rx.rsp_prod_pvt++);
765 resp->flags |= XEN_NETRXF_extra_info;
767 gso->u.gso.size = netbk->meta[npo.meta_cons].gso_size;
768 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
770 gso->u.gso.features = 0;
772 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
776 netbk_add_frag_responses(vif, status,
777 netbk->meta + npo.meta_cons + 1,
778 sco->meta_slots_used);
780 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret);
781 if (ret && list_empty(&vif->notify_list))
782 list_add_tail(&vif->notify_list, ¬ify);
784 xenvif_notify_tx_completion(vif);
787 npo.meta_cons += sco->meta_slots_used;
791 list_for_each_entry_safe(vif, tmp, ¬ify, notify_list) {
792 notify_remote_via_irq(vif->rx_irq);
793 list_del_init(&vif->notify_list);
796 /* More work to do? */
797 if (!skb_queue_empty(&netbk->rx_queue) &&
798 !timer_pending(&netbk->net_timer))
799 xen_netbk_kick_thread(netbk);
802 void xen_netbk_queue_tx_skb(struct xenvif *vif, struct sk_buff *skb)
804 struct xen_netbk *netbk = vif->netbk;
806 skb_queue_tail(&netbk->rx_queue, skb);
808 xen_netbk_kick_thread(netbk);
811 static void xen_netbk_alarm(unsigned long data)
813 struct xen_netbk *netbk = (struct xen_netbk *)data;
814 xen_netbk_kick_thread(netbk);
817 static int __on_net_schedule_list(struct xenvif *vif)
819 return !list_empty(&vif->schedule_list);
822 /* Must be called with net_schedule_list_lock held */
823 static void remove_from_net_schedule_list(struct xenvif *vif)
825 if (likely(__on_net_schedule_list(vif))) {
826 list_del_init(&vif->schedule_list);
831 static struct xenvif *poll_net_schedule_list(struct xen_netbk *netbk)
833 struct xenvif *vif = NULL;
835 spin_lock_irq(&netbk->net_schedule_list_lock);
836 if (list_empty(&netbk->net_schedule_list))
839 vif = list_first_entry(&netbk->net_schedule_list,
840 struct xenvif, schedule_list);
846 remove_from_net_schedule_list(vif);
848 spin_unlock_irq(&netbk->net_schedule_list_lock);
852 void xen_netbk_schedule_xenvif(struct xenvif *vif)
855 struct xen_netbk *netbk = vif->netbk;
857 if (__on_net_schedule_list(vif))
860 spin_lock_irqsave(&netbk->net_schedule_list_lock, flags);
861 if (!__on_net_schedule_list(vif) &&
862 likely(xenvif_schedulable(vif))) {
863 list_add_tail(&vif->schedule_list, &netbk->net_schedule_list);
866 spin_unlock_irqrestore(&netbk->net_schedule_list_lock, flags);
870 if ((nr_pending_reqs(netbk) < (MAX_PENDING_REQS/2)) &&
871 !list_empty(&netbk->net_schedule_list))
872 xen_netbk_kick_thread(netbk);
875 void xen_netbk_deschedule_xenvif(struct xenvif *vif)
877 struct xen_netbk *netbk = vif->netbk;
878 spin_lock_irq(&netbk->net_schedule_list_lock);
879 remove_from_net_schedule_list(vif);
880 spin_unlock_irq(&netbk->net_schedule_list_lock);
883 void xen_netbk_check_rx_xenvif(struct xenvif *vif)
887 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do);
890 xen_netbk_schedule_xenvif(vif);
893 static void tx_add_credit(struct xenvif *vif)
895 unsigned long max_burst, max_credit;
898 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
899 * Otherwise the interface can seize up due to insufficient credit.
901 max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size;
902 max_burst = min(max_burst, 131072UL);
903 max_burst = max(max_burst, vif->credit_bytes);
905 /* Take care that adding a new chunk of credit doesn't wrap to zero. */
906 max_credit = vif->remaining_credit + vif->credit_bytes;
907 if (max_credit < vif->remaining_credit)
908 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
910 vif->remaining_credit = min(max_credit, max_burst);
913 static void tx_credit_callback(unsigned long data)
915 struct xenvif *vif = (struct xenvif *)data;
917 xen_netbk_check_rx_xenvif(vif);
920 static void netbk_tx_err(struct xenvif *vif,
921 struct xen_netif_tx_request *txp, RING_IDX end)
923 RING_IDX cons = vif->tx.req_cons;
926 make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR);
929 txp = RING_GET_REQUEST(&vif->tx, cons++);
931 vif->tx.req_cons = cons;
932 xen_netbk_check_rx_xenvif(vif);
936 static void netbk_fatal_tx_err(struct xenvif *vif)
938 netdev_err(vif->dev, "fatal error; disabling device\n");
939 xenvif_carrier_off(vif);
943 static int netbk_count_requests(struct xenvif *vif,
944 struct xen_netif_tx_request *first,
945 struct xen_netif_tx_request *txp,
948 RING_IDX cons = vif->tx.req_cons;
953 if (!(first->flags & XEN_NETTXF_more_data))
957 struct xen_netif_tx_request dropped_tx = { 0 };
959 if (slots >= work_to_do) {
961 "Asked for %d slots but exceeds this limit\n",
963 netbk_fatal_tx_err(vif);
967 /* This guest is really using too many slots and
968 * considered malicious.
970 if (unlikely(slots >= fatal_skb_slots)) {
972 "Malicious frontend using %d slots, threshold %u\n",
973 slots, fatal_skb_slots);
974 netbk_fatal_tx_err(vif);
978 /* Xen network protocol had implicit dependency on
979 * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to
980 * the historical MAX_SKB_FRAGS value 18 to honor the
981 * same behavior as before. Any packet using more than
982 * 18 slots but less than fatal_skb_slots slots is
985 if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) {
988 "Too many slots (%d) exceeding limit (%d), dropping packet\n",
989 slots, XEN_NETBK_LEGACY_SLOTS_MAX);
996 memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + slots),
999 /* If the guest submitted a frame >= 64 KiB then
1000 * first->size overflowed and following slots will
1001 * appear to be larger than the frame.
1003 * This cannot be fatal error as there are buggy
1004 * frontends that do this.
1006 * Consume all slots and drop the packet.
1008 if (!drop_err && txp->size > first->size) {
1009 if (net_ratelimit())
1010 netdev_dbg(vif->dev,
1011 "Invalid tx request, slot size %u > remaining size %u\n",
1012 txp->size, first->size);
1016 first->size -= txp->size;
1019 if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) {
1020 netdev_err(vif->dev, "Cross page boundary, txp->offset: %x, size: %u\n",
1021 txp->offset, txp->size);
1022 netbk_fatal_tx_err(vif);
1026 more_data = txp->flags & XEN_NETTXF_more_data;
1031 } while (more_data);
1034 netbk_tx_err(vif, first, cons + slots);
1041 static struct page *xen_netbk_alloc_page(struct xen_netbk *netbk,
1045 page = alloc_page(GFP_KERNEL|__GFP_COLD);
1048 set_page_ext(page, netbk, pending_idx);
1049 netbk->mmap_pages[pending_idx] = page;
1053 static struct gnttab_copy *xen_netbk_get_requests(struct xen_netbk *netbk,
1055 struct sk_buff *skb,
1056 struct xen_netif_tx_request *txp,
1057 struct gnttab_copy *gop)
1059 struct skb_shared_info *shinfo = skb_shinfo(skb);
1060 skb_frag_t *frags = shinfo->frags;
1061 u16 pending_idx = *((u16 *)skb->data);
1065 pending_ring_idx_t index, start_idx = 0;
1066 uint16_t dst_offset;
1067 unsigned int nr_slots;
1068 struct pending_tx_info *first = NULL;
1070 /* At this point shinfo->nr_frags is in fact the number of
1071 * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX.
1073 nr_slots = shinfo->nr_frags;
1075 /* Skip first skb fragment if it is on same page as header fragment. */
1076 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
1078 /* Coalesce tx requests, at this point the packet passed in
1079 * should be <= 64K. Any packets larger than 64K have been
1080 * handled in netbk_count_requests().
1082 for (shinfo->nr_frags = slot = start; slot < nr_slots;
1083 shinfo->nr_frags++) {
1084 struct pending_tx_info *pending_tx_info =
1085 netbk->pending_tx_info;
1087 page = alloc_page(GFP_KERNEL|__GFP_COLD);
1093 while (dst_offset < PAGE_SIZE && slot < nr_slots) {
1094 gop->flags = GNTCOPY_source_gref;
1096 gop->source.u.ref = txp->gref;
1097 gop->source.domid = vif->domid;
1098 gop->source.offset = txp->offset;
1100 gop->dest.domid = DOMID_SELF;
1102 gop->dest.offset = dst_offset;
1103 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
1105 if (dst_offset + txp->size > PAGE_SIZE) {
1106 /* This page can only merge a portion
1107 * of tx request. Do not increment any
1108 * pointer / counter here. The txp
1109 * will be dealt with in future
1110 * rounds, eventually hitting the
1113 gop->len = PAGE_SIZE - dst_offset;
1114 txp->offset += gop->len;
1115 txp->size -= gop->len;
1116 dst_offset += gop->len; /* quit loop */
1118 /* This tx request can be merged in the page */
1119 gop->len = txp->size;
1120 dst_offset += gop->len;
1122 index = pending_index(netbk->pending_cons++);
1124 pending_idx = netbk->pending_ring[index];
1126 memcpy(&pending_tx_info[pending_idx].req, txp,
1130 pending_tx_info[pending_idx].vif = vif;
1132 /* Poison these fields, corresponding
1133 * fields for head tx req will be set
1134 * to correct values after the loop.
1136 netbk->mmap_pages[pending_idx] = (void *)(~0UL);
1137 pending_tx_info[pending_idx].head =
1138 INVALID_PENDING_RING_IDX;
1141 first = &pending_tx_info[pending_idx];
1143 head_idx = pending_idx;
1153 first->req.offset = 0;
1154 first->req.size = dst_offset;
1155 first->head = start_idx;
1156 set_page_ext(page, netbk, head_idx);
1157 netbk->mmap_pages[head_idx] = page;
1158 frag_set_pending_idx(&frags[shinfo->nr_frags], head_idx);
1161 BUG_ON(shinfo->nr_frags > MAX_SKB_FRAGS);
1165 /* Unwind, freeing all pages and sending error responses. */
1166 while (shinfo->nr_frags-- > start) {
1167 xen_netbk_idx_release(netbk,
1168 frag_get_pending_idx(&frags[shinfo->nr_frags]),
1169 XEN_NETIF_RSP_ERROR);
1171 /* The head too, if necessary. */
1173 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
1178 static int xen_netbk_tx_check_gop(struct xen_netbk *netbk,
1179 struct sk_buff *skb,
1180 struct gnttab_copy **gopp)
1182 struct gnttab_copy *gop = *gopp;
1183 u16 pending_idx = *((u16 *)skb->data);
1184 struct skb_shared_info *shinfo = skb_shinfo(skb);
1185 struct pending_tx_info *tx_info;
1186 int nr_frags = shinfo->nr_frags;
1188 u16 peek; /* peek into next tx request */
1190 /* Check status of header. */
1193 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
1195 /* Skip first skb fragment if it is on same page as header fragment. */
1196 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
1198 for (i = start; i < nr_frags; i++) {
1200 pending_ring_idx_t head;
1202 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
1203 tx_info = &netbk->pending_tx_info[pending_idx];
1204 head = tx_info->head;
1206 /* Check error status: if okay then remember grant handle. */
1208 newerr = (++gop)->status;
1211 peek = netbk->pending_ring[pending_index(++head)];
1212 } while (!pending_tx_is_head(netbk, peek));
1214 if (likely(!newerr)) {
1215 /* Had a previous error? Invalidate this fragment. */
1217 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1221 /* Error on this fragment: respond to client with an error. */
1222 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
1224 /* Not the first error? Preceding frags already invalidated. */
1228 /* First error: invalidate header and preceding fragments. */
1229 pending_idx = *((u16 *)skb->data);
1230 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1231 for (j = start; j < i; j++) {
1232 pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
1233 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1236 /* Remember the error: invalidate all subsequent fragments. */
1244 static void xen_netbk_fill_frags(struct xen_netbk *netbk, struct sk_buff *skb)
1246 struct skb_shared_info *shinfo = skb_shinfo(skb);
1247 int nr_frags = shinfo->nr_frags;
1250 for (i = 0; i < nr_frags; i++) {
1251 skb_frag_t *frag = shinfo->frags + i;
1252 struct xen_netif_tx_request *txp;
1256 pending_idx = frag_get_pending_idx(frag);
1258 txp = &netbk->pending_tx_info[pending_idx].req;
1259 page = virt_to_page(idx_to_kaddr(netbk, pending_idx));
1260 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
1261 skb->len += txp->size;
1262 skb->data_len += txp->size;
1263 skb->truesize += txp->size;
1265 /* Take an extra reference to offset xen_netbk_idx_release */
1266 get_page(netbk->mmap_pages[pending_idx]);
1267 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1271 static int xen_netbk_get_extras(struct xenvif *vif,
1272 struct xen_netif_extra_info *extras,
1275 struct xen_netif_extra_info extra;
1276 RING_IDX cons = vif->tx.req_cons;
1279 if (unlikely(work_to_do-- <= 0)) {
1280 netdev_err(vif->dev, "Missing extra info\n");
1281 netbk_fatal_tx_err(vif);
1285 memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons),
1287 if (unlikely(!extra.type ||
1288 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1289 vif->tx.req_cons = ++cons;
1290 netdev_err(vif->dev,
1291 "Invalid extra type: %d\n", extra.type);
1292 netbk_fatal_tx_err(vif);
1296 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
1297 vif->tx.req_cons = ++cons;
1298 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
1303 static int netbk_set_skb_gso(struct xenvif *vif,
1304 struct sk_buff *skb,
1305 struct xen_netif_extra_info *gso)
1307 if (!gso->u.gso.size) {
1308 netdev_err(vif->dev, "GSO size must not be zero.\n");
1309 netbk_fatal_tx_err(vif);
1313 /* Currently only TCPv4 S.O. is supported. */
1314 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
1315 netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
1316 netbk_fatal_tx_err(vif);
1320 skb_shinfo(skb)->gso_size = gso->u.gso.size;
1321 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1323 /* Header must be checked, and gso_segs computed. */
1324 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1325 skb_shinfo(skb)->gso_segs = 0;
1330 static int checksum_setup(struct xenvif *vif, struct sk_buff *skb)
1334 int recalculate_partial_csum = 0;
1337 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1338 * peers can fail to set NETRXF_csum_blank when sending a GSO
1339 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1340 * recalculate the partial checksum.
1342 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1343 vif->rx_gso_checksum_fixup++;
1344 skb->ip_summed = CHECKSUM_PARTIAL;
1345 recalculate_partial_csum = 1;
1348 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1349 if (skb->ip_summed != CHECKSUM_PARTIAL)
1352 if (skb->protocol != htons(ETH_P_IP))
1355 iph = (void *)skb->data;
1356 switch (iph->protocol) {
1358 if (!skb_partial_csum_set(skb, 4 * iph->ihl,
1359 offsetof(struct tcphdr, check)))
1362 if (recalculate_partial_csum) {
1363 struct tcphdr *tcph = tcp_hdr(skb);
1364 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1365 skb->len - iph->ihl*4,
1370 if (!skb_partial_csum_set(skb, 4 * iph->ihl,
1371 offsetof(struct udphdr, check)))
1374 if (recalculate_partial_csum) {
1375 struct udphdr *udph = udp_hdr(skb);
1376 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1377 skb->len - iph->ihl*4,
1382 if (net_ratelimit())
1383 netdev_err(vif->dev,
1384 "Attempting to checksum a non-TCP/UDP packet, dropping a protocol %d packet\n",
1395 static bool tx_credit_exceeded(struct xenvif *vif, unsigned size)
1397 unsigned long now = jiffies;
1398 unsigned long next_credit =
1399 vif->credit_timeout.expires +
1400 msecs_to_jiffies(vif->credit_usec / 1000);
1402 /* Timer could already be pending in rare cases. */
1403 if (timer_pending(&vif->credit_timeout))
1406 /* Passed the point where we can replenish credit? */
1407 if (time_after_eq(now, next_credit)) {
1408 vif->credit_timeout.expires = now;
1412 /* Still too big to send right now? Set a callback. */
1413 if (size > vif->remaining_credit) {
1414 vif->credit_timeout.data =
1416 vif->credit_timeout.function =
1418 mod_timer(&vif->credit_timeout,
1427 static unsigned xen_netbk_tx_build_gops(struct xen_netbk *netbk)
1429 struct gnttab_copy *gop = netbk->tx_copy_ops, *request_gop;
1430 struct sk_buff *skb;
1433 while ((nr_pending_reqs(netbk) + XEN_NETBK_LEGACY_SLOTS_MAX
1434 < MAX_PENDING_REQS) &&
1435 !list_empty(&netbk->net_schedule_list)) {
1437 struct xen_netif_tx_request txreq;
1438 struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX];
1440 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
1444 unsigned int data_len;
1445 pending_ring_idx_t index;
1447 /* Get a netif from the list with work to do. */
1448 vif = poll_net_schedule_list(netbk);
1449 /* This can sometimes happen because the test of
1450 * list_empty(net_schedule_list) at the top of the
1451 * loop is unlocked. Just go back and have another
1457 if (vif->tx.sring->req_prod - vif->tx.req_cons >
1458 XEN_NETIF_TX_RING_SIZE) {
1459 netdev_err(vif->dev,
1460 "Impossible number of requests. "
1461 "req_prod %d, req_cons %d, size %ld\n",
1462 vif->tx.sring->req_prod, vif->tx.req_cons,
1463 XEN_NETIF_TX_RING_SIZE);
1464 netbk_fatal_tx_err(vif);
1468 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, work_to_do);
1474 idx = vif->tx.req_cons;
1475 rmb(); /* Ensure that we see the request before we copy it. */
1476 memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq));
1478 /* Credit-based scheduling. */
1479 if (txreq.size > vif->remaining_credit &&
1480 tx_credit_exceeded(vif, txreq.size)) {
1485 vif->remaining_credit -= txreq.size;
1488 vif->tx.req_cons = ++idx;
1490 memset(extras, 0, sizeof(extras));
1491 if (txreq.flags & XEN_NETTXF_extra_info) {
1492 work_to_do = xen_netbk_get_extras(vif, extras,
1494 idx = vif->tx.req_cons;
1495 if (unlikely(work_to_do < 0))
1499 ret = netbk_count_requests(vif, &txreq, txfrags, work_to_do);
1500 if (unlikely(ret < 0))
1505 if (unlikely(txreq.size < ETH_HLEN)) {
1506 netdev_dbg(vif->dev,
1507 "Bad packet size: %d\n", txreq.size);
1508 netbk_tx_err(vif, &txreq, idx);
1512 /* No crossing a page as the payload mustn't fragment. */
1513 if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) {
1514 netdev_err(vif->dev,
1515 "txreq.offset: %x, size: %u, end: %lu\n",
1516 txreq.offset, txreq.size,
1517 (txreq.offset&~PAGE_MASK) + txreq.size);
1518 netbk_fatal_tx_err(vif);
1522 index = pending_index(netbk->pending_cons);
1523 pending_idx = netbk->pending_ring[index];
1525 data_len = (txreq.size > PKT_PROT_LEN &&
1526 ret < XEN_NETBK_LEGACY_SLOTS_MAX) ?
1527 PKT_PROT_LEN : txreq.size;
1529 skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN,
1530 GFP_ATOMIC | __GFP_NOWARN);
1531 if (unlikely(skb == NULL)) {
1532 netdev_dbg(vif->dev,
1533 "Can't allocate a skb in start_xmit.\n");
1534 netbk_tx_err(vif, &txreq, idx);
1538 /* Packets passed to netif_rx() must have some headroom. */
1539 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1541 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1542 struct xen_netif_extra_info *gso;
1543 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1545 if (netbk_set_skb_gso(vif, skb, gso)) {
1546 /* Failure in netbk_set_skb_gso is fatal. */
1552 /* XXX could copy straight to head */
1553 page = xen_netbk_alloc_page(netbk, pending_idx);
1556 netbk_tx_err(vif, &txreq, idx);
1560 gop->source.u.ref = txreq.gref;
1561 gop->source.domid = vif->domid;
1562 gop->source.offset = txreq.offset;
1564 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
1565 gop->dest.domid = DOMID_SELF;
1566 gop->dest.offset = txreq.offset;
1568 gop->len = txreq.size;
1569 gop->flags = GNTCOPY_source_gref;
1573 memcpy(&netbk->pending_tx_info[pending_idx].req,
1574 &txreq, sizeof(txreq));
1575 netbk->pending_tx_info[pending_idx].vif = vif;
1576 netbk->pending_tx_info[pending_idx].head = index;
1577 *((u16 *)skb->data) = pending_idx;
1579 __skb_put(skb, data_len);
1581 skb_shinfo(skb)->nr_frags = ret;
1582 if (data_len < txreq.size) {
1583 skb_shinfo(skb)->nr_frags++;
1584 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1587 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1588 INVALID_PENDING_IDX);
1591 netbk->pending_cons++;
1593 request_gop = xen_netbk_get_requests(netbk, vif,
1595 if (request_gop == NULL) {
1597 netbk_tx_err(vif, &txreq, idx);
1602 __skb_queue_tail(&netbk->tx_queue, skb);
1604 vif->tx.req_cons = idx;
1605 xen_netbk_check_rx_xenvif(vif);
1607 if ((gop-netbk->tx_copy_ops) >= ARRAY_SIZE(netbk->tx_copy_ops))
1611 return gop - netbk->tx_copy_ops;
1614 static void xen_netbk_tx_submit(struct xen_netbk *netbk)
1616 struct gnttab_copy *gop = netbk->tx_copy_ops;
1617 struct sk_buff *skb;
1619 while ((skb = __skb_dequeue(&netbk->tx_queue)) != NULL) {
1620 struct xen_netif_tx_request *txp;
1625 pending_idx = *((u16 *)skb->data);
1626 vif = netbk->pending_tx_info[pending_idx].vif;
1627 txp = &netbk->pending_tx_info[pending_idx].req;
1629 /* Check the remap error code. */
1630 if (unlikely(xen_netbk_tx_check_gop(netbk, skb, &gop))) {
1631 netdev_dbg(vif->dev, "netback grant failed.\n");
1632 skb_shinfo(skb)->nr_frags = 0;
1637 data_len = skb->len;
1639 (void *)(idx_to_kaddr(netbk, pending_idx)|txp->offset),
1641 if (data_len < txp->size) {
1642 /* Append the packet payload as a fragment. */
1643 txp->offset += data_len;
1644 txp->size -= data_len;
1646 /* Schedule a response immediately. */
1647 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1650 if (txp->flags & XEN_NETTXF_csum_blank)
1651 skb->ip_summed = CHECKSUM_PARTIAL;
1652 else if (txp->flags & XEN_NETTXF_data_validated)
1653 skb->ip_summed = CHECKSUM_UNNECESSARY;
1655 xen_netbk_fill_frags(netbk, skb);
1658 * If the initial fragment was < PKT_PROT_LEN then
1659 * pull through some bytes from the other fragments to
1660 * increase the linear region to PKT_PROT_LEN bytes.
1662 if (skb_headlen(skb) < PKT_PROT_LEN && skb_is_nonlinear(skb)) {
1663 int target = min_t(int, skb->len, PKT_PROT_LEN);
1664 __pskb_pull_tail(skb, target - skb_headlen(skb));
1667 skb->dev = vif->dev;
1668 skb->protocol = eth_type_trans(skb, skb->dev);
1669 skb_reset_network_header(skb);
1671 if (checksum_setup(vif, skb)) {
1672 netdev_dbg(vif->dev,
1673 "Can't setup checksum in net_tx_action\n");
1678 skb_probe_transport_header(skb, 0);
1680 vif->dev->stats.rx_bytes += skb->len;
1681 vif->dev->stats.rx_packets++;
1683 xenvif_receive_skb(vif, skb);
1687 /* Called after netfront has transmitted */
1688 static void xen_netbk_tx_action(struct xen_netbk *netbk)
1692 nr_gops = xen_netbk_tx_build_gops(netbk);
1697 gnttab_batch_copy(netbk->tx_copy_ops, nr_gops);
1699 xen_netbk_tx_submit(netbk);
1702 static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx,
1706 struct pending_tx_info *pending_tx_info;
1707 pending_ring_idx_t head;
1708 u16 peek; /* peek into next tx request */
1710 BUG_ON(netbk->mmap_pages[pending_idx] == (void *)(~0UL));
1712 /* Already complete? */
1713 if (netbk->mmap_pages[pending_idx] == NULL)
1716 pending_tx_info = &netbk->pending_tx_info[pending_idx];
1718 vif = pending_tx_info->vif;
1719 head = pending_tx_info->head;
1721 BUG_ON(!pending_tx_is_head(netbk, head));
1722 BUG_ON(netbk->pending_ring[pending_index(head)] != pending_idx);
1725 pending_ring_idx_t index;
1726 pending_ring_idx_t idx = pending_index(head);
1727 u16 info_idx = netbk->pending_ring[idx];
1729 pending_tx_info = &netbk->pending_tx_info[info_idx];
1730 make_tx_response(vif, &pending_tx_info->req, status);
1732 /* Setting any number other than
1733 * INVALID_PENDING_RING_IDX indicates this slot is
1734 * starting a new packet / ending a previous packet.
1736 pending_tx_info->head = 0;
1738 index = pending_index(netbk->pending_prod++);
1739 netbk->pending_ring[index] = netbk->pending_ring[info_idx];
1743 peek = netbk->pending_ring[pending_index(++head)];
1745 } while (!pending_tx_is_head(netbk, peek));
1747 netbk->mmap_pages[pending_idx]->mapping = 0;
1748 put_page(netbk->mmap_pages[pending_idx]);
1749 netbk->mmap_pages[pending_idx] = NULL;
1753 static void make_tx_response(struct xenvif *vif,
1754 struct xen_netif_tx_request *txp,
1757 RING_IDX i = vif->tx.rsp_prod_pvt;
1758 struct xen_netif_tx_response *resp;
1761 resp = RING_GET_RESPONSE(&vif->tx, i);
1765 if (txp->flags & XEN_NETTXF_extra_info)
1766 RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1768 vif->tx.rsp_prod_pvt = ++i;
1769 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify);
1771 notify_remote_via_irq(vif->tx_irq);
1774 static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
1781 RING_IDX i = vif->rx.rsp_prod_pvt;
1782 struct xen_netif_rx_response *resp;
1784 resp = RING_GET_RESPONSE(&vif->rx, i);
1785 resp->offset = offset;
1786 resp->flags = flags;
1788 resp->status = (s16)size;
1790 resp->status = (s16)st;
1792 vif->rx.rsp_prod_pvt = ++i;
1797 static inline int rx_work_todo(struct xen_netbk *netbk)
1799 return !skb_queue_empty(&netbk->rx_queue);
1802 static inline int tx_work_todo(struct xen_netbk *netbk)
1805 if ((nr_pending_reqs(netbk) + XEN_NETBK_LEGACY_SLOTS_MAX
1806 < MAX_PENDING_REQS) &&
1807 !list_empty(&netbk->net_schedule_list))
1813 static int xen_netbk_kthread(void *data)
1815 struct xen_netbk *netbk = data;
1816 while (!kthread_should_stop()) {
1817 wait_event_interruptible(netbk->wq,
1818 rx_work_todo(netbk) ||
1819 tx_work_todo(netbk) ||
1820 kthread_should_stop());
1823 if (kthread_should_stop())
1826 if (rx_work_todo(netbk))
1827 xen_netbk_rx_action(netbk);
1829 if (tx_work_todo(netbk))
1830 xen_netbk_tx_action(netbk);
1836 void xen_netbk_unmap_frontend_rings(struct xenvif *vif)
1839 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1842 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1846 int xen_netbk_map_frontend_rings(struct xenvif *vif,
1847 grant_ref_t tx_ring_ref,
1848 grant_ref_t rx_ring_ref)
1851 struct xen_netif_tx_sring *txs;
1852 struct xen_netif_rx_sring *rxs;
1856 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1857 tx_ring_ref, &addr);
1861 txs = (struct xen_netif_tx_sring *)addr;
1862 BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE);
1864 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1865 rx_ring_ref, &addr);
1869 rxs = (struct xen_netif_rx_sring *)addr;
1870 BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE);
1872 vif->rx_req_cons_peek = 0;
1877 xen_netbk_unmap_frontend_rings(vif);
1881 static int __init netback_init(void)
1890 if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) {
1892 "xen-netback: fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n",
1893 fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX);
1894 fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX;
1897 xen_netbk_group_nr = num_online_cpus();
1898 xen_netbk = vzalloc(sizeof(struct xen_netbk) * xen_netbk_group_nr);
1902 for (group = 0; group < xen_netbk_group_nr; group++) {
1903 struct xen_netbk *netbk = &xen_netbk[group];
1904 skb_queue_head_init(&netbk->rx_queue);
1905 skb_queue_head_init(&netbk->tx_queue);
1907 init_timer(&netbk->net_timer);
1908 netbk->net_timer.data = (unsigned long)netbk;
1909 netbk->net_timer.function = xen_netbk_alarm;
1911 netbk->pending_cons = 0;
1912 netbk->pending_prod = MAX_PENDING_REQS;
1913 for (i = 0; i < MAX_PENDING_REQS; i++)
1914 netbk->pending_ring[i] = i;
1916 init_waitqueue_head(&netbk->wq);
1917 netbk->task = kthread_create(xen_netbk_kthread,
1919 "netback/%u", group);
1921 if (IS_ERR(netbk->task)) {
1922 printk(KERN_ALERT "kthread_create() fails at netback\n");
1923 del_timer(&netbk->net_timer);
1924 rc = PTR_ERR(netbk->task);
1928 kthread_bind(netbk->task, group);
1930 INIT_LIST_HEAD(&netbk->net_schedule_list);
1932 spin_lock_init(&netbk->net_schedule_list_lock);
1934 atomic_set(&netbk->netfront_count, 0);
1936 wake_up_process(netbk->task);
1939 rc = xenvif_xenbus_init();
1946 while (--group >= 0) {
1947 struct xen_netbk *netbk = &xen_netbk[group];
1948 del_timer(&netbk->net_timer);
1949 kthread_stop(netbk->task);
1956 module_init(netback_init);
1958 static void __exit netback_fini(void)
1962 xenvif_xenbus_fini();
1964 for (i = 0; i < xen_netbk_group_nr; i++) {
1965 struct xen_netbk *netbk = &xen_netbk[i];
1966 del_timer_sync(&netbk->net_timer);
1967 kthread_stop(netbk->task);
1968 for (j = 0; j < MAX_PENDING_REQS; j++) {
1969 if (netbk->mmap_pages[i])
1970 __free_page(netbk->mmap_pages[i]);
1976 module_exit(netback_fini);
1978 MODULE_LICENSE("Dual BSD/GPL");
1979 MODULE_ALIAS("xen-backend:vif");