Generic zcopy_* functions
Subsystems (7)
v2 cover Generic zcopy_* functions
From: Jonathan Lemon <[email protected]>
This is set of cleanup patches for zerocopy which are intended to allow a introduction of a different zerocopy implementation.
The top level API will use the skb_zcopy_*() functions, while the current TCP specific zerocopy ends up using msg_zerocopy_*() calls.
There should be no functional changes from these patches.
v1->v2:
Break changes to skb_zcopy_put into 3 patches, in order to make it easier to follow the changes. Add Willem's suggestion about renaming sock_zerocopy_
Patch 1:
Move zerocopy bits from tx_flags into zc_flags for clarity. These bits will be used in the RX path in the future.
Patch 2: remove dead function
Patch 3: simplify sock_zerocopy_put
Patch 4: push status/refcounts into sock_zerocopy_callback Patch 5: replace sock_zerocopy_put with skb_zcopy_put Patch 6: rename sock_zerocopy_get
Patch 7:
Add an optional skb parameter to callback, allowing access to the attached skb from the callback.
Patch 8:
Add skb_zcopy_put_abort, and move zerocopy logic into the callback function. There unfortunately is still a check against the callback type here.
Patch 9:
Set the skb zc_flags from the ubuf being attached, instead of a fixed value, allowing different initialization types.
Patch 10: Replace open-coded assignments
Patch 11: Relocate skb_zcopy_clear() in skb_release_data() Patch 12: rename sock_zerocopy_ to msg_zerocpy_
Jonathan Lemon (12):
net: group skb_shinfo zerocopy related bits together.
skbuff: remove unused skb_zcopy_abort function
skbuff: simplify sock_zerocopy_put
skbuff: Push status and refcounts into sock_zerocopy_callback
skbuff: replace sock_zerocopy_put() with skb_zcopy_put()
skbuff: replace sock_zerocopy_get with skb_zcopy_get
skbuff: Add skb parameter to the ubuf zerocopy callback
skbuff: Call sock_zerocopy_put_abort from skb_zcopy_put_abort
skbuff: add zc_flags to ubuf_info for ubuf setup tap/tun: use skb_zcopy_set() instead of open coded assignment
skbuff: Call skb_zcopy_clear() before unref'ing fragments
skbuff: rename sock_zerocopy_* to msg_zerocopy_*
drivers/net/tap.c | 6 +- drivers/net/tun.c | 6 +- drivers/net/xen-netback/common.h | 3 +- drivers/net/xen-netback/interface.c | 4 +- drivers/net/xen-netback/netback.c | 7 ++- drivers/vhost/net.c | 4 +- include/linux/skbuff.h | 95 +++++++++++++++-------------- net/core/skbuff.c | 66 ++++++++++---------- net/ipv4/ip_output.c | 5 +- net/ipv4/tcp.c | 8 +-- net/ipv6/ip6_output.c | 5 +- net/kcm/kcmsock.c | 4 +- 12 files changed, 106 insertions(+), 107 deletions(-)
Signature · 1 line
2.24.1
RFC PATCH v2 1/12 net: group skb_shinfo zerocopy related bits together.
From: Jonathan Lemon <[email protected]>
In preparation for expanded zerocopy (TX and RX), move the ZC related bits out of tx_flags into their own flag word.
--- drivers/net/tap.c | 3 +-- drivers/net/tun.c | 3 +-- drivers/net/xen-netback/interface.c | 4 ++-- include/linux/skbuff.h | 33 ++++++++++++++++------------- net/core/skbuff.c | 10 ++++----- net/ipv4/tcp.c | 2 +- net/kcm/kcmsock.c | 4 ++-- 7 files changed, 30 insertions(+), 29 deletions(-)
drivers/net/tap.c +1 -2
@@ -723,8 +723,7 @@ static ssize_t tap_get_user(struct tap_queue *q, void *msg_control,/* copy skb_ubuf_info for callback when skb has no error */ if (zerocopy) {skb_shinfo (skb)->destructor_arg = msg_control; -skb_shinfo (skb)->tx_flags |=SKBTX_DEV_ZEROCOPY ; -skb_shinfo (skb)->tx_flags |=SKBTX_SHARED_FRAG ; +skb_shinfo (skb)->zc_flags |=SKBZC_FRAGMENTS ; }else if (msg_control) {struct ubuf_info *uarg = msg_control; uarg->callback (uarg, false);
drivers/net/tun.c +1 -2
@@ -1815,8 +1815,7 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,/* copy skb_ubuf_info for callback when skb has no error */ if (zerocopy) {skb_shinfo (skb)->destructor_arg = msg_control; -skb_shinfo (skb)->tx_flags |=SKBTX_DEV_ZEROCOPY ; -skb_shinfo (skb)->tx_flags |=SKBTX_SHARED_FRAG ; +skb_shinfo (skb)->zc_flags |=SKBZC_FRAGMENTS ; }else if (msg_control) {struct ubuf_info *uarg = msg_control; uarg->callback (uarg, false);
drivers/net/xen-netback/interface.c +2 -2
@@ -47,7 +47,7 @@/* Number of bytes allowed on the internal guest Rx queue. */ #define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE) -/* This function is used to set SKBTX_DEV_ZEROCOPY as well as +/* This function is used to set SKBZC_ENABLE as well as * increasing the inflight counter. We need to increase the inflight * counter because core driver calls into xenvif_zerocopy_callback * which calls xenvif_skb_zerocopy_complete.
@@ -55,7 +55,7 @@void xenvif_skb_zerocopy_prepare (struct xenvif_queue *queue,struct sk_buff *skb) { -skb_shinfo (skb)->tx_flags |=SKBTX_DEV_ZEROCOPY ; +skb_shinfo (skb)->zc_flags |=SKBZC_ENABLE ;atomic_inc (&queue->inflight_packets); }
include/linux/skbuff.h +18 -15
@@ -430,24 +430,27 @@ enum {/* device driver is going to provide hardware time stamp */ SKBTX_IN_PROGRESS =1 <<2 , -/* device driver supports TX zero-copy buffers */ -SKBTX_DEV_ZEROCOPY =1 <<3 , -/* generate wifi status information (where possible) */ SKBTX_WIFI_STATUS =1 <<4 , +/* generate software time stamp when entering packet scheduling */ + SKBTX_SCHED_TSTAMP = 1 << 6, +}; + +/* Definitions for zc_flags in struct skb_shared_info */ +enum { +/* use zcopy routines */ +SKBZC_ENABLE =BIT (0 ), +/* This indicates at least one fragment might be overwritten * (as in vmsplice(), sendfile() ...) * If we need to compute a TX checksum, we'll need to copy * all frags to avoid possible bad checksum */ -SKBTX_SHARED_FRAG =1 <<5 , - -/* generate software time stamp when entering packet scheduling */ - SKBTX_SCHED_TSTAMP = 1 << 6, +SKBZC_SHARED_FRAG =BIT (1 ), }; -#define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG) +#define SKBZC_FRAGMENTS (SKBZC_ENABLE | SKBZC_SHARED_FRAG)#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ SKBTX_SCHED_TSTAMP)#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
@@ -510,7 +513,7 @@ int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,* the end of the header data, ie. at skb->end. */ struct skb_shared_info { -__u8 __unused; +__u8 zc_flags;__u8 meta_len;__u8 nr_frags;__u8 tx_flags;
@@ -1437,7 +1440,7 @@ static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)static inline struct ubuf_info *skb_zcopy (struct sk_buff *skb) { -bool is_zcopy = skb &&skb_shinfo (skb)->tx_flags &SKBTX_DEV_ZEROCOPY ; +bool is_zcopy = skb &&skb_shinfo (skb)->zc_flags &SKBZC_ENABLE ;return is_zcopy ?skb_uarg (skb) :NULL ; }
@@ -1451,14 +1454,14 @@ static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,else sock_zerocopy_get (uarg);skb_shinfo (skb)->destructor_arg = uarg; -skb_shinfo (skb)->tx_flags |=SKBTX_ZEROCOPY_FRAG ; +skb_shinfo (skb)->zc_flags |=SKBZC_FRAGMENTS ; } }static inline void skb_zcopy_set_nouarg (struct sk_buff *skb,void *val) {skb_shinfo (skb)->destructor_arg = (void *)((uintptr_t ) val |0x1UL ); -skb_shinfo (skb)->tx_flags |=SKBTX_ZEROCOPY_FRAG ; +skb_shinfo (skb)->zc_flags |=SKBZC_FRAGMENTS ; }static inline bool skb_zcopy_is_nouarg (struct sk_buff *skb)
@@ -1486,7 +1489,7 @@ static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy) uarg->callback (uarg, zerocopy); } -skb_shinfo (skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG ; +skb_shinfo (skb)->zc_flags &= ~SKBZC_FRAGMENTS ; } }
@@ -1497,7 +1500,7 @@ static inline void skb_zcopy_abort(struct sk_buff *skb)if (uarg) {sock_zerocopy_put_abort (uarg, false); -skb_shinfo (skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG ; +skb_shinfo (skb)->zc_flags &= ~SKBZC_FRAGMENTS ; } }
@@ -3323,7 +3326,7 @@ static inline int skb_linearize(struct sk_buff *skb)static inline bool skb_has_shared_frag (const struct sk_buff *skb) {return skb_is_nonlinear (skb) && -skb_shinfo (skb)->tx_flags &SKBTX_SHARED_FRAG ; +skb_shinfo (skb)->zc_flags &SKBZC_SHARED_FRAG ; }/**
net/core/skbuff.c +5 -5
@@ -1330,7 +1330,7 @@ static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,* @skb: the skb to modify * @gfp_mask: allocation priority * -* This must be called on SKBTX_DEV_ZEROCOPY skb. +* This must be called on SKBZC_ENABLE skb. * It will copy all frags into kernel and drop the reference * to userspace pages. *
@@ -3267,8 +3267,8 @@ void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) {int pos =skb_headlen (skb); -skb_shinfo (skb1)->tx_flags |=skb_shinfo (skb)->tx_flags & -SKBTX_SHARED_FRAG ; +skb_shinfo (skb1)->zc_flags |=skb_shinfo (skb)->zc_flags & +SKBZC_SHARED_FRAG ;skb_zerocopy_clone (skb1, skb,0 );if (len < pos)/* Split line is inside header. */ skb_split_inside_header (skb, skb1, len, pos);
@@ -3957,8 +3957,8 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,skb_copy_from_linear_data_offset (head_skb, offset,skb_put (nskb, hsize), hsize); -skb_shinfo (nskb)->tx_flags |=skb_shinfo (head_skb)->tx_flags & -SKBTX_SHARED_FRAG ; +skb_shinfo (nskb)->zc_flags |=skb_shinfo (head_skb)->zc_flags & +SKBZC_SHARED_FRAG ;if (skb_orphan_frags (frag_skb,GFP_ATOMIC ) ||skb_zerocopy_clone (nskb, frag_skb,GFP_ATOMIC ))
net/ipv4/tcp.c +1 -1
@@ -1010,7 +1010,7 @@ struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags, }if (!(flags &MSG_NO_SHARED_FRAGS )) -skb_shinfo (skb)->tx_flags |=SKBTX_SHARED_FRAG ; +skb_shinfo (skb)->zc_flags |=SKBZC_SHARED_FRAG ; skb->len += copy; skb->data_len += copy;
net/kcm/kcmsock.c +2 -2
@@ -786,7 +786,7 @@ static ssize_t kcm_sendpage(struct socket *sock, struct page *page,if (skb_can_coalesce (skb, i, page, offset)) {skb_frag_size_add (&skb_shinfo (skb)->frags[i -1 ], size); -skb_shinfo (skb)->tx_flags |=SKBTX_SHARED_FRAG ; +skb_shinfo (skb)->zc_flags |=SKBZC_SHARED_FRAG ; goto coalesced; }
@@ -834,7 +834,7 @@ static ssize_t kcm_sendpage(struct socket *sock, struct page *page,get_page (page);skb_fill_page_desc (skb, i, page, offset, size); -skb_shinfo (skb)->tx_flags |=SKBTX_SHARED_FRAG ; +skb_shinfo (skb)->zc_flags |=SKBZC_SHARED_FRAG ; coalesced: skb->len += size;
Signature · 1 line
2.24.1
Willem de Bruijn
maintainerRe: [PATCH 01/12 v2 RFC] net: group skb_shinfo zerocopy related bits together.
Quoted 8 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
In preparation for expanded zerocopy (TX and RX), move the ZC related bits out of
tx_flagsinto their own flag word.Signed-off-by Jonathan Lemon <[email protected]>
I think it's better to expand tx_flags to a u16 and add the two new flags that you need.
That allows the additional 7 bits to be used for arbitrary flags, not stranding 8 bits exactly only for zerocopy features.
Moving around a few u8's in the same cacheline won't be a problem.
I also prefer not to rename flags that some of us are familiar with, if it's not needed.
Re: [PATCH 01/12 v2 RFC] net: group skb_shinfo zerocopy related bits together.
Quoted 4 lines · Willem de Bruijn
Quoted 3 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
Quoted 8 lines
Quoted 5 lines
In preparation for expanded zerocopy (TX and RX), move the ZC related bits out of
tx_flagsinto their own flag word.Signed-off-by Jonathan Lemon <[email protected]>I think it's better to expand
tx_flagsto a u16 and add the two new flags that you need.
Okay, but in that case, tx_flags is now wrong, since some of these flags will also be checked on the rx path.
Quoted 4 lines
That allows the additional 7 bits to be used for arbitrary flags, not stranding 8 bits exactly only for zerocopy features.
Moving around a few u8's in the same cacheline won't be a problem.
How about rearranging them to form 16 bits, and just calling it 'flags'?
Quoted 2 lines
I also prefer not to rename flags that some of us are familiar with, if it's not needed.
Ack.
--
Jonathan
Willem de Bruijn
maintainerRe: [PATCH 01/12 v2 RFC] net: group skb_shinfo zerocopy related bits together.
Quoted 19 lines · Jonathan Lemon
Quoted 4 lines · Willem de Bruijn
Quoted 3 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
Quoted 8 lines
Quoted 5 lines
In preparation for expanded zerocopy (TX and RX), move the ZC related bits out of
tx_flagsinto their own flag word.Signed-off-by Jonathan Lemon <[email protected]>I think it's better to expand
tx_flagsto a u16 and add the two new flags that you need.Okay, but in that case,
tx_flagsis now wrong, since some of these flags will also be checked on the rx path.
Quoted 6 lines
Quoted 4 lines
That allows the additional 7 bits to be used for arbitrary flags, not stranding 8 bits exactly only for zerocopy features.
Moving around a few u8's in the same cacheline won't be a problem.
How about rearranging them to form 16 bits, and just calling it 'flags'?
I thought that would be a good idea, but tx_flags is used in a lot more places than I expected. That will be a lot of renaming. Let's just either keep the name or add a new field.
Re: [PATCH 01/12 v2 RFC] net: group skb_shinfo zerocopy related bits together.
Quoted 24 lines · Willem de Bruijn
Quoted 22 lines · Jonathan Lemon
Quoted 4 lines · Willem de Bruijn
Quoted 3 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
Quoted 8 lines
Quoted 5 lines
In preparation for expanded zerocopy (TX and RX), move the ZC related bits out of
tx_flagsinto their own flag word.Signed-off-by Jonathan Lemon <[email protected]>I think it's better to expand
tx_flagsto a u16 and add the two new flags that you need.Okay, but in that case,
tx_flagsis now wrong, since some of these flags will also be checked on the rx path.Quoted 3 lines
That allows the additional 7 bits to be used for arbitrary flags, not stranding 8 bits exactly only for zerocopy features.
Quoted 7 lines
Quoted 3 lines
Quoted 1 line
Moving around a few u8's in the same cacheline won't be a problem.
How about rearranging them to form 16 bits, and just calling it 'flags'?
I thought that would be a good idea, but
tx_flagsis used in a lot more places than I expected. That will be a lot of renaming. Let's just either keep the name or add a new field.
Hmm, which? I went with "add new field" = zc_flags. I can rename this to something else if that's too specific. Suggestions?
Signature · 1 line
Jonathan "naming is hard!"
Willem de Bruijn
maintainerRe: [PATCH 01/12 v2 RFC] net: group skb_shinfo zerocopy related bits together.
Quoted 28 lines · Jonathan Lemon
Quoted 26 lines · Willem de Bruijn
Quoted 24 lines · Jonathan Lemon
Quoted 4 lines · Willem de Bruijn
Quoted 3 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
Quoted 8 lines
Quoted 5 lines
In preparation for expanded zerocopy (TX and RX), move the ZC related bits out of
tx_flagsinto their own flag word.Signed-off-by Jonathan Lemon <[email protected]>I think it's better to expand
tx_flagsto a u16 and add the two new flags that you need.Okay, but in that case,
tx_flagsis now wrong, since some of these flags will also be checked on the rx path.Quoted 4 lines
That allows the additional 7 bits to be used for arbitrary flags, not stranding 8 bits exactly only for zerocopy features.
Moving around a few u8's in the same cacheline won't be a problem.
Quoted 8 lines
Quoted 5 lines
Quoted 1 line
How about rearranging them to form 16 bits, and just calling it 'flags'?
I thought that would be a good idea, but
tx_flagsis used in a lot more places than I expected. That will be a lot of renaming. Let's just either keep the name or add a new field.Hmm, which? I went with "add new field" =
zc_flags. I can rename this to something else if that's too specific. Suggestions?
Just flags, for now? High order bit is to not move and rename all the existing flags for the sake of it. Name itself is less important.
RFC PATCH v2 2/12 skbuff: remove unused skb_zcopy_abort function
In reply to Jonathan LemonFrom: Jonathan Lemon <[email protected]>
skb_zcopy_abort() has no in-tree consumers, remove it.
--- include/linux/skbuff.h | 11 ----------- 1 file changed, 11 deletions(-)
include/linux/skbuff.h +0 -11
@@ -1493,17 +1493,6 @@ static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy) } } -/* Abort a zerocopy operation and revert zckey on error in send syscall */ -static inline void skb_zcopy_abort (struct sk_buff *skb) -{ -struct ubuf_info *uarg =skb_zcopy (skb); - -if (uarg) { -sock_zerocopy_put_abort (uarg, false); -skb_shinfo (skb)->zc_flags &= ~SKBZC_FRAGMENTS ; - } -} -static inline void skb_mark_not_on_list (struct sk_buff *skb) { skb->next =NULL ;
Signature · 1 line
2.24.1
Re: [PATCH 02/12 v2 RFC] skbuff: remove unused skb_zcopy_abort function
Quoted 6 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
skb_zcopy_abort()has no in-tree consumers, remove it.Signed-off-by Jonathan Lemon <[email protected]>
Indeed unused. This was used in the packet and raw zerocopy patches in the original patchset. But we never merged those, for lack of copy avoidance benefit at small packet sizes.
RFC PATCH v2 3/12 skbuff: simplify sock_zerocopy_put
In reply to Jonathan LemonFrom: Jonathan Lemon <[email protected]>
All 'struct ubuf_info' users should have a callback defined. Remove the dead code path to consume_skb(), which makes unwarranted assumptions about how the structure was allocated.
--- net/core/skbuff.c | 8 ++------ 1 file changed, 2 insertions(+), 6 deletions(-)
net/core/skbuff.c +2 -6
@@ -1245,12 +1245,8 @@ EXPORT_SYMBOL_GPL(sock_zerocopy_callback);void sock_zerocopy_put (struct ubuf_info *uarg) { -if (uarg &&refcount_dec_and_test (&uarg->refcnt)) { -if (uarg->callback) - uarg->callback (uarg, uarg->zerocopy); -else -consume_skb (skb_from_uarg (uarg)); - } +if (uarg &&refcount_dec_and_test (&uarg->refcnt)) + uarg->callback (uarg, uarg->zerocopy); }EXPORT_SYMBOL_GPL (sock_zerocopy_put);
Signature · 1 line
2.24.1
Re: [PATCH 03/12 v2 RFC] skbuff: simplify sock_zerocopy_put
Quoted 8 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
All 'struct
ubuf_info' users should have a callback defined. Remove the dead code path toconsume_skb(), which makes unwarranted assumptions about how the structure was allocated.Signed-off-by Jonathan Lemon <[email protected]>
Please link to the commit I shared that made the consume_skb path obsolete. Before that this would have been unsafe.
should have a callback defined -> have a callback defined as of commit 0a4a060bb204 ("sock: fixzerocopy_success regression with msg_zerocopy").
With that explanation why this is correct
Re: [PATCH 03/12 v2 RFC] skbuff: simplify sock_zerocopy_put
In reply to Jonathan LemonQuoted 10 lines · Jonathan Lemon
net/core/skbuff.c +0
-1@@ -1245,12 +1245,8 @@ EXPORT_SYMBOL_GPL(sock_zerocopy_callback);void sock_zerocopy_put (struct ubuf_info *uarg) { -if (uarg &&refcount_dec_and_test (&uarg->refcnt)) {
net/core/skbuff.c EXPORT_SYMBOL_GPL(sock_zerocopy_callback); · line 1252
Quoted 10 lines
net/core/skbuff.c +2
-5-if (uarg->callback) - uarg->callback (uarg, uarg->zerocopy); -else -consume_skb (skb_from_uarg (uarg)); - } +if (uarg &&refcount_dec_and_test (&uarg->refcnt)) + uarg->callback (uarg, uarg->zerocopy); }EXPORT_SYMBOL_GPL (sock_zerocopy_put);
since it is down to 2 lines, move to skbuff.h as an inline?
Re: [PATCH 03/12 v2 RFC] skbuff: simplify sock_zerocopy_put
Quoted 7 lines · David Ahern
Quoted 6 lines · Jonathan Lemon
net/core/skbuff.c +0
-0@@ -1245,12 +1245,8 @@ EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
Quoted 17 lines
Quoted 4 lines
net/core/skbuff.c +0
-1void sock_zerocopy_put (struct ubuf_info *uarg) { -if (uarg &&refcount_dec_and_test (&uarg->refcnt)) {net/core/skbuff.c EXPORT_SYMBOL_GPL(sock_zerocopy_callback); · line 1252
Quoted 10 lines
net/core/skbuff.c +2
-5-if (uarg->callback) - uarg->callback (uarg, uarg->zerocopy); -else -consume_skb (skb_from_uarg (uarg)); - } +if (uarg &&refcount_dec_and_test (&uarg->refcnt)) + uarg->callback (uarg, uarg->zerocopy); }EXPORT_SYMBOL_GPL (sock_zerocopy_put);since it is down to 2 lines, move to
skbuff.has an inline?
nm. that is done in patch 5.
RFC PATCH v2 4/12 skbuff: Push status and refcounts into sock_zerocopy_callback
In reply to Jonathan LemonFrom: Jonathan Lemon <[email protected]>
Before this change, the caller of sock_zerocopy_callback would need to save the zerocopy status, decrement and check the refcount, and then call the callback function - the callback was only invoked when the refcount reached zero.
Now, the caller just passes the status into the callback function, which saves the status and handles its own refcounts.
This makes the behavior of the sock_zerocopy_callback identical to the tpacket and vhost callbacks.
--- include/linux/skbuff.h | 3 --- net/core/skbuff.c | 14 +++++++++++--- 2 files changed, 11 insertions(+), 6 deletions(-)
include/linux/skbuff.h +0 -3
@@ -1482,9 +1482,6 @@ static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)if (uarg) {if (skb_zcopy_is_nouarg (skb)) {/* no notification callback */ - }else if (uarg->callback == sock_zerocopy_callback) { - uarg->zerocopy = uarg->zerocopy && zerocopy; -sock_zerocopy_put (uarg); }else { uarg->callback (uarg, zerocopy); }
net/core/skbuff.c +11 -3
@@ -1194,7 +1194,7 @@ static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)return true; } -void sock_zerocopy_callback (struct ubuf_info *uarg,bool success) +static void __sock_zerocopy_callback (struct ubuf_info *uarg) {struct sk_buff *tail, *skb =skb_from_uarg (uarg);struct sock_exterr_skb *serr;
@@ -1222,7 +1222,7 @@ void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) serr->ee.ee_origin =SO_EE_ORIGIN_ZEROCOPY ; serr->ee.ee_data = hi; serr->ee.ee_info = lo; -if (!success) +if (!uarg->zerocopy) serr->ee.ee_code |=SO_EE_CODE_ZEROCOPY_COPIED ; q = &sk->sk_error_queue;
@@ -1241,11 +1241,19 @@ void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)consume_skb (skb);sock_put (sk); } + +void sock_zerocopy_callback (struct ubuf_info *uarg,bool success) +{ + uarg->zerocopy = uarg->zerocopy & success; + +if (refcount_dec_and_test (&uarg->refcnt)) +__sock_zerocopy_callback (uarg); +}EXPORT_SYMBOL_GPL (sock_zerocopy_callback);void sock_zerocopy_put (struct ubuf_info *uarg) { -if (uarg &&refcount_dec_and_test (&uarg->refcnt)) +if (uarg) uarg->callback (uarg, uarg->zerocopy); }EXPORT_SYMBOL_GPL (sock_zerocopy_put);
Signature · 1 line
2.24.1
Willem de Bruijn
maintainerRe: [PATCH 04/12 v2 RFC] skbuff: Push status and refcounts into sock_zerocopy_callback
Quoted 59 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
Before this change, the caller of
sock_zerocopy_callbackwould need to save the zerocopy status, decrement and check the refcount, and then call the callback function - the callback was only invoked when the refcount reached zero.Now, the caller just passes the status into the callback function, which saves the status and handles its own refcounts.
This makes the behavior of the
sock_zerocopy_callbackidentical to the tpacket and vhost callbacks.Signed-off-by Jonathan Lemon <[email protected]>--- include/linux/skbuff.h | 3 --- net/core/skbuff.c | 14 +++++++++++--- 2 files changed, 11 insertions(+), 6 deletions(-)include/linux/skbuff.h +0
-3@@ -1482,9 +1482,6 @@ static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)if (uarg) {if (skb_zcopy_is_nouarg (skb)) {/* no notification callback */ - }else if (uarg->callback == sock_zerocopy_callback) { - uarg->zerocopy = uarg->zerocopy && zerocopy; -sock_zerocopy_put (uarg); }else { uarg->callback (uarg, zerocopy); }net/core/skbuff.c +2
-2@@ -1194,7 +1194,7 @@ static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)return true; } -void sock_zerocopy_callback (struct ubuf_info *uarg,bool success) +static void __sock_zerocopy_callback (struct ubuf_info *uarg) {struct sk_buff *tail, *skb =skb_from_uarg (uarg);struct sock_exterr_skb *serr;@@ -1222,7 +1222,7 @@ void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) serr->ee.ee_origin =SO_EE_ORIGIN_ZEROCOPY ; serr->ee.ee_data = hi; serr->ee.ee_info = lo; -if (!success) +if (!uarg->zerocopy) serr->ee.ee_code |=SO_EE_CODE_ZEROCOPY_COPIED ; q = &sk->sk_error_queue;@@ -1241,11 +1241,19 @@ void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)consume_skb (skb);sock_put (sk);
net/core/skbuff.c void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) · line 1252
Quoted 10 lines
net/core/skbuff.c +8
-0} + +void sock_zerocopy_callback (struct ubuf_info *uarg,bool success) +{ + uarg->zerocopy = uarg->zerocopy & success; + +if (refcount_dec_and_test (&uarg->refcnt)) +__sock_zerocopy_callback (uarg); +}EXPORT_SYMBOL_GPL (sock_zerocopy_callback);
I still think this helper is unnecessary. Just return immediately in existing sock_zerocopy_callback if refcount is not zero.
net/core/skbuff.c void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) · line 1259
Quoted 7 lines
net/core/skbuff.c +1
-1void sock_zerocopy_put (struct ubuf_info *uarg) { -if (uarg &&refcount_dec_and_test (&uarg->refcnt)) +if (uarg) uarg->callback (uarg, uarg->zerocopy); }EXPORT_SYMBOL_GPL (sock_zerocopy_put);
This does increase the number of indirect function calls. Which are not cheap post spectre.
In the common case for msg_zerocopy we only have two clones, one sent out and one on the retransmit queue. So I guess the cost will be acceptable.
Re: [PATCH 04/12 v2 RFC] skbuff: Push status and refcounts into sock_zerocopy_callback
Quoted 9 lines · Willem de Bruijn
net/core/skbuff.c void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) · line 1259
Quoted 8 lines
void
sock_zerocopy_put(struct ubuf_info *uarg)
{net/core/skbuff.c +1
-1-if (uarg &&refcount_dec_and_test (&uarg->refcnt)) +if (uarg) uarg->callback (uarg, uarg->zerocopy); }EXPORT_SYMBOL_GPL (sock_zerocopy_put);
Quoted 7 lines
This does increase the number of indirect function calls. Which are not cheap post spectre.
In the common case for
msg_zerocopywe only have two clones, one sent out and one on the retransmit queue. So I guess the cost will be acceptable.
sock_zerocopy_callback seems to be the only one at the moment - or the most dominant if I goofed my search. Could use the INDIRECT_CALL macros for it.
Re: [PATCH 04/12 v2 RFC] skbuff: Push status and refcounts into sock_zerocopy_callback
In reply to Willem de BruijnQuoted 65 lines · Willem de Bruijn
Quoted 64 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
Before this change, the caller of
sock_zerocopy_callbackwould need to save the zerocopy status, decrement and check the refcount, and then call the callback function - the callback was only invoked when the refcount reached zero.Now, the caller just passes the status into the callback function, which saves the status and handles its own refcounts.
This makes the behavior of the
sock_zerocopy_callbackidentical to the tpacket and vhost callbacks.Signed-off-by Jonathan Lemon <[email protected]>--- include/linux/skbuff.h | 3 --- net/core/skbuff.c | 14 +++++++++++--- 2 files changed, 11 insertions(+), 6 deletions(-)include/linux/skbuff.h +0
-3@@ -1482,9 +1482,6 @@ static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)if (uarg) {if (skb_zcopy_is_nouarg (skb)) {/* no notification callback */ - }else if (uarg->callback == sock_zerocopy_callback) { - uarg->zerocopy = uarg->zerocopy && zerocopy; -sock_zerocopy_put (uarg); }else { uarg->callback (uarg, zerocopy); }net/core/skbuff.c +6
-2@@ -1194,7 +1194,7 @@ static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)return true; } -void sock_zerocopy_callback (struct ubuf_info *uarg,bool success) +static void __sock_zerocopy_callback (struct ubuf_info *uarg) {struct sk_buff *tail, *skb =skb_from_uarg (uarg);struct sock_exterr_skb *serr;@@ -1222,7 +1222,7 @@ void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) serr->ee.ee_origin =SO_EE_ORIGIN_ZEROCOPY ; serr->ee.ee_data = hi; serr->ee.ee_info = lo; -if (!success) +if (!uarg->zerocopy) serr->ee.ee_code |=SO_EE_CODE_ZEROCOPY_COPIED ; q = &sk->sk_error_queue;@@ -1241,11 +1241,19 @@ void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)consume_skb (skb);sock_put (sk); } + +void sock_zerocopy_callback (struct ubuf_info *uarg,bool success) +{ + uarg->zerocopy = uarg->zerocopy & success;
Quoted 8 lines
net/core/skbuff.c void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) · line 1252
Quoted 5 lines
net/core/skbuff.c +4
-0+ +if (refcount_dec_and_test (&uarg->refcnt)) +__sock_zerocopy_callback (uarg); +}EXPORT_SYMBOL_GPL (sock_zerocopy_callback);I still think this helper is unnecessary. Just return immediately in existing
sock_zerocopy_callbackif refcount is not zero.
I think the helper makes the logic clearer, and prevents misuse of the success variable in the main function (use of last value vs the latched value). If you really feel that strongly about it, I'll fold it into one function.
Quoted 8 lines
net/core/skbuff.c void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) · line 1259
Quoted 7 lines
net/core/skbuff.c +1
-1void sock_zerocopy_put (struct ubuf_info *uarg) { -if (uarg &&refcount_dec_and_test (&uarg->refcnt)) +if (uarg) uarg->callback (uarg, uarg->zerocopy); }EXPORT_SYMBOL_GPL (sock_zerocopy_put);
Quoted 6 lines
This does increase the number of indirect function calls. Which are not cheap post spectre.
In the common case for
msg_zerocopywe only have two clones, one sent out and one on the retransmit queue. So I guess the cost will be acceptable.
Yes, this was the source of my original comment about this being a slight pessimization - moving the refcount into the function.
I briefly considered adding a flag like 'use_refcnt', so the logic becomes:
if (uarg && (!uarg->use_refcnt || refcount_dec_and_test(&uarg->refcnt)))
But thought this might be too much micro-optimization. But if the call overhead is significant, I can put this back in.
Signature · 1 line
Jonathan
Willem de Bruijn
maintainerRe: [PATCH 04/12 v2 RFC] skbuff: Push status and refcounts into sock_zerocopy_callback
Quoted 76 lines · Jonathan Lemon
Quoted 65 lines · Willem de Bruijn
Quoted 64 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
Before this change, the caller of
sock_zerocopy_callbackwould need to save the zerocopy status, decrement and check the refcount, and then call the callback function - the callback was only invoked when the refcount reached zero.Now, the caller just passes the status into the callback function, which saves the status and handles its own refcounts.
This makes the behavior of the
sock_zerocopy_callbackidentical to the tpacket and vhost callbacks.Signed-off-by Jonathan Lemon <[email protected]>--- include/linux/skbuff.h | 3 --- net/core/skbuff.c | 14 +++++++++++--- 2 files changed, 11 insertions(+), 6 deletions(-)include/linux/skbuff.h +0
-3@@ -1482,9 +1482,6 @@ static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)if (uarg) {if (skb_zcopy_is_nouarg (skb)) {/* no notification callback */ - }else if (uarg->callback == sock_zerocopy_callback) { - uarg->zerocopy = uarg->zerocopy && zerocopy; -sock_zerocopy_put (uarg); }else { uarg->callback (uarg, zerocopy); }net/core/skbuff.c +6
-2@@ -1194,7 +1194,7 @@ static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)return true; } -void sock_zerocopy_callback (struct ubuf_info *uarg,bool success) +static void __sock_zerocopy_callback (struct ubuf_info *uarg) {struct sk_buff *tail, *skb =skb_from_uarg (uarg);struct sock_exterr_skb *serr;@@ -1222,7 +1222,7 @@ void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) serr->ee.ee_origin =SO_EE_ORIGIN_ZEROCOPY ; serr->ee.ee_data = hi; serr->ee.ee_info = lo; -if (!success) +if (!uarg->zerocopy) serr->ee.ee_code |=SO_EE_CODE_ZEROCOPY_COPIED ; q = &sk->sk_error_queue;@@ -1241,11 +1241,19 @@ void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)consume_skb (skb);sock_put (sk); } + +void sock_zerocopy_callback (struct ubuf_info *uarg,bool success) +{ + uarg->zerocopy = uarg->zerocopy & success;Quoted 8 lines
net/core/skbuff.c void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) · line 1252
Quoted 5 lines
net/core/skbuff.c +4
-0+ +if (refcount_dec_and_test (&uarg->refcnt)) +__sock_zerocopy_callback (uarg); +}EXPORT_SYMBOL_GPL (sock_zerocopy_callback);I still think this helper is unnecessary. Just return immediately in existing
sock_zerocopy_callbackif refcount is not zero.
Quoted 4 lines
I think the helper makes the logic clearer, and prevents misuse of the success variable in the main function (use of last value vs the latched value). If you really feel that strongly about it, I'll fold it into one function.
Ok. Thanks, no, it's fine.
Quoted 19 lines
Quoted 9 lines
net/core/skbuff.c void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) · line 1259
Quoted 8 lines
net/core/skbuff.c +1
-1void sock_zerocopy_put (struct ubuf_info *uarg) { -if (uarg &&refcount_dec_and_test (&uarg->refcnt)) +if (uarg) uarg->callback (uarg, uarg->zerocopy); }EXPORT_SYMBOL_GPL (sock_zerocopy_put);Quoted 6 lines
This does increase the number of indirect function calls. Which are not cheap post spectre.
In the common case for
msg_zerocopywe only have two clones, one sent out and one on the retransmit queue. So I guess the cost will be acceptable.Yes, this was the source of my original comment about this being a slight pessimization - moving the refcount into the function.
Quoted 7 lines
I briefly considered adding a flag like '
use_refcnt', so the logic becomes:if (uarg && (!uarg->use_refcnt || refcount_dec_and_test(&uarg->refcnt)))But thought this might be too much micro-optimization. But if the call overhead is significant, I can put this back in.
Agreed on the premature optimization. Let's find out :)
RFC PATCH v2 5/12 skbuff: replace sock_zerocopy_put() with skb_zcopy_put()
In reply to Jonathan LemonFrom: Jonathan Lemon <[email protected]>
Replace sock_zerocopy_put with the generic skb_zcopy_put() function. Pass 'true' as the success argument, as this is identical to no change.
--- include/linux/skbuff.h | 7 ++++++- net/core/skbuff.c | 9 +-------- net/ipv4/tcp.c | 2 +- 3 files changed, 8 insertions(+), 10 deletions(-)
include/linux/skbuff.h +6 -1
@@ -499,7 +499,6 @@ static inline void sock_zerocopy_get(struct ubuf_info *uarg)refcount_inc (&uarg->refcnt); } -void sock_zerocopy_put (struct ubuf_info *uarg);void sock_zerocopy_put_abort (struct ubuf_info *uarg,bool have_uref);void sock_zerocopy_callback (struct ubuf_info *uarg,bool success);
@@ -1474,6 +1473,12 @@ static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)return (void *)((uintptr_t )skb_shinfo (skb)->destructor_arg & ~0x1UL ); } +static inline void skb_zcopy_put (struct ubuf_info *uarg) +{ +if (uarg) + uarg->callback (uarg, true); +} +/* Release a reference on a zerocopy structure */ static inline void skb_zcopy_clear (struct sk_buff *skb,bool zerocopy) {
net/core/skbuff.c +1 -8
@@ -1251,13 +1251,6 @@ void sock_zerocopy_callback(struct ubuf_info *uarg, bool success) }EXPORT_SYMBOL_GPL (sock_zerocopy_callback); -void sock_zerocopy_put (struct ubuf_info *uarg) -{ -if (uarg) - uarg->callback (uarg, uarg->zerocopy); -} -EXPORT_SYMBOL_GPL (sock_zerocopy_put); -void sock_zerocopy_put_abort (struct ubuf_info *uarg,bool have_uref) {if (uarg) {
@@ -1267,7 +1260,7 @@ void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) uarg->len--;if (have_uref) -sock_zerocopy_put (uarg); +skb_zcopy_put (uarg); } }EXPORT_SYMBOL_GPL (sock_zerocopy_put_abort);
net/ipv4/tcp.c +1 -1
@@ -1429,7 +1429,7 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)tcp_push (sk, flags, mss_now, tp->nonagle, size_goal); } out_nopush: -sock_zerocopy_put (uarg); +skb_zcopy_put (uarg);return copied + copied_syn; do_error:
Signature · 1 line
2.24.1
Willem de Bruijn
maintainerRe: [PATCH 05/12 v2 RFC] skbuff: replace sock_zerocopy_put() with skb_zcopy_put()
Quoted 8 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
Replace
sock_zerocopy_putwith the genericskb_zcopy_put()function. Pass 'true' as the success argument, as this is identical to no change.Signed-off-by Jonathan Lemon <[email protected]>
uarg->zerocopy may be false if sock_zerocopy_put_abort is called from tcp_sendmsg_locked
Re: [PATCH 05/12 v2 RFC] skbuff: replace sock_zerocopy_put() with skb_zcopy_put()
Quoted 4 lines · Willem de Bruijn
Quoted 3 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
Quoted 8 lines
Quoted 5 lines
Replace
sock_zerocopy_putwith the genericskb_zcopy_put()function. Pass 'true' as the success argument, as this is identical to no change.Signed-off-by Jonathan Lemon <[email protected]>
uarg->zerocopymay be false ifsock_zerocopy_put_abortis called fromtcp_sendmsg_locked
Yes, it may well be false. The original logic goes:
sock_zerocopy_put_abort() sock_zerocopy_put() sock_zerocopy_callback(..., success = uarg->zerocopy)
if (success)
The new logic is now:
sock_zerocopy_put_abort()
sock_zerocopy_callback(..., success = true)
uarg->zerocopy = uarg->zerocopy & success
if (uarg->zerocopy)The success value ls latched into uarg->zerocopy, and any failure is persistent. Hence my comment about passing 'true' not changing the logic.
Signature · 1 line
Jonathan
Willem de Bruijn
maintainerRe: [PATCH 05/12 v2 RFC] skbuff: replace sock_zerocopy_put() with skb_zcopy_put()
Quoted 24 lines · Jonathan Lemon
Quoted 4 lines · Willem de Bruijn
Quoted 3 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
Quoted 8 lines
Quoted 5 lines
Replace
sock_zerocopy_putwith the genericskb_zcopy_put()function. Pass 'true' as the success argument, as this is identical to no change.Signed-off-by Jonathan Lemon <[email protected]>
uarg->zerocopymay be false ifsock_zerocopy_put_abortis called fromtcp_sendmsg_lockedYes, it may well be false. The original logic goes:
sock_zerocopy_put_abort() sock_zerocopy_put() sock_zerocopy_callback(..., success = uarg->zerocopy) if (success)The new logic is now:
Quoted 7 lines
sock_zerocopy_put_abort() sock_zerocopy_callback(..., success = true) uarg->zerocopy = uarg->zerocopy & success if (uarg->zerocopy)The success value ls latched into
uarg->zerocopy, and any failure is persistent.
Good point.
It's not entirely obvious, and a bit unintuitive to pass success in an abort statement. But it works.
Abort does not schedule a notification if no skb associated with the uarg was sent. And if it was, the zerocopy state will reflect the & of all those packets. On which note, if renaming the currently inconsistent name, maybe renaming to zerocopy_success is the more descriptive one, then.
RFC PATCH v2 6/12 skbuff: replace sock_zerocopy_get with skb_zcopy_get
In reply to Jonathan LemonFrom: Jonathan Lemon <[email protected]>
Rename the get routines for consistency.
--- include/linux/skbuff.h | 12 ++++++------ net/core/skbuff.c | 2 +- 2 files changed, 7 insertions(+), 7 deletions(-)
include/linux/skbuff.h +6 -6
@@ -494,11 +494,6 @@ struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);struct ubuf_info *sock_zerocopy_realloc (struct sock *sk,size_t size,struct ubuf_info *uarg); -static inline void sock_zerocopy_get (struct ubuf_info *uarg) -{ -refcount_inc (&uarg->refcnt); -} -void sock_zerocopy_put_abort (struct ubuf_info *uarg,bool have_uref);void sock_zerocopy_callback (struct ubuf_info *uarg,bool success);
@@ -1444,6 +1439,11 @@ static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)return is_zcopy ?skb_uarg (skb) :NULL ; } +static inline void skb_zcopy_get (struct ubuf_info *uarg) +{ +refcount_inc (&uarg->refcnt); +} +static inline void skb_zcopy_set (struct sk_buff *skb,struct ubuf_info *uarg,bool *have_ref) {
@@ -1451,7 +1451,7 @@ static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,if (unlikely (have_ref && *have_ref)) *have_ref = false;else -sock_zerocopy_get (uarg); +skb_zcopy_get (uarg);skb_shinfo (skb)->destructor_arg = uarg;skb_shinfo (skb)->zc_flags |=SKBZC_FRAGMENTS ; }
net/core/skbuff.c +1 -1
@@ -1163,7 +1163,7 @@ struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,/* no extra ref when appending to datagram (MSG_MORE) */ if (sk->sk_type ==SOCK_STREAM ) -sock_zerocopy_get (uarg); +skb_zcopy_get (uarg);return uarg; }
Signature · 1 line
2.24.1
RFC PATCH v2 7/12 skbuff: Add skb parameter to the ubuf zerocopy callback
In reply to Jonathan LemonFrom: Jonathan Lemon <[email protected]>
Add an optional skb parameter to the zerocopy callback parameter, which is passed down from skb_zcopy_clear(). This gives access to the original skb, which is needed for upcoming RX zero-copy error handling.
Make the name of the the zerocopy success pararmeter consistent for all callbacks.
--- drivers/net/tap.c | 2 +- drivers/net/tun.c | 2 +- drivers/net/xen-netback/common.h | 3 ++- drivers/net/xen-netback/netback.c | 7 ++++--- drivers/vhost/net.c | 3 ++- include/linux/skbuff.h | 20 +++++++++----------- net/core/skbuff.c | 3 ++- 7 files changed, 21 insertions(+), 19 deletions(-)
drivers/net/tap.c +1 -1
@@ -726,7 +726,7 @@ static ssize_t tap_get_user(struct tap_queue *q, void *msg_control,skb_shinfo (skb)->zc_flags |=SKBZC_FRAGMENTS ; }else if (msg_control) {struct ubuf_info *uarg = msg_control; - uarg->callback (uarg, false); + uarg->callback (NULL , uarg, false); }if (tap) {
drivers/net/tun.c +1 -1
@@ -1818,7 +1818,7 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,skb_shinfo (skb)->zc_flags |=SKBZC_FRAGMENTS ; }else if (msg_control) {struct ubuf_info *uarg = msg_control; - uarg->callback (uarg, false); + uarg->callback (NULL , uarg, false); }skb_reset_network_header (skb);
drivers/net/xen-netback/common.h +2 -1
@@ -399,7 +399,8 @@ void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb);void xenvif_carrier_on (struct xenvif *vif);/* Callback from stack when TX packet can be released */ -void xenvif_zerocopy_callback (struct ubuf_info *ubuf,bool zerocopy_success); +void xenvif_zerocopy_callback (struct sk_buff *skb,struct ubuf_info *ubuf, +bool success);/* Unmap a pending page and release it back to the guest */ void xenvif_idx_unmap (struct xenvif_queue *queue,u16 pending_idx);
drivers/net/xen-netback/netback.c +4 -3
@@ -1091,7 +1091,7 @@ static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *s uarg =skb_shinfo (skb)->destructor_arg;/* increase inflight counter to offset decrement in callback */ atomic_inc (&queue->inflight_packets); - uarg->callback (uarg, true); + uarg->callback (NULL , uarg, true);skb_shinfo (skb)->destructor_arg =NULL ;/* Fill the skb with the new (local) frags. */
@@ -1228,7 +1228,8 @@ static int xenvif_tx_submit(struct xenvif_queue *queue)return work_done; } -void xenvif_zerocopy_callback (struct ubuf_info *ubuf,bool zerocopy_success) +void xenvif_zerocopy_callback (struct sk_buff *skb,struct ubuf_info *ubuf, +bool success) {unsigned long flags;pending_ring_idx_t index;
@@ -1253,7 +1254,7 @@ void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success) }while (ubuf);spin_unlock_irqrestore (&queue->callback_lock, flags); -if (likely (zerocopy_success)) +if (likely (success)) queue->stats.tx_zerocopy_success++;else queue->stats.tx_zerocopy_fail++;
drivers/vhost/net.c +2 -1
@@ -381,7 +381,8 @@ static void vhost_zerocopy_signal_used(struct vhost_net *net, } } -static void vhost_zerocopy_callback (struct ubuf_info *ubuf,bool success) +static void vhost_zerocopy_callback (struct sk_buff *skb, +struct ubuf_info *ubuf,bool success) {struct vhost_net_ubuf_ref *ubufs = ubuf->ctx;struct vhost_virtqueue *vq = ubufs->vq;
include/linux/skbuff.h +9 -11
@@ -458,13 +458,13 @@ enum {/* * The callback notifies userspace to release buffers when skb DMA is done in * lower device, the skb last reference should be 0 when calling this. -* The zerocopy_success argument is true if zero copy transmit occurred, -* false on data copy or out of memory error caused by data copy attempt. +* The success argument is true if zero copy transmit occurred, false on +* data copy or out of memory error caused by data copy attempt. * The ctx field is used to track device context. * The desc field is used to track userspace buffer index. */ struct ubuf_info { - void (*callback)(struct ubuf_info *, bool zerocopy_success); + void (*callback)(struct sk_buff *, struct ubuf_info *, bool success); union { struct { unsigned long desc;
@@ -496,7 +496,8 @@ struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,void sock_zerocopy_put_abort (struct ubuf_info *uarg,bool have_uref); -void sock_zerocopy_callback (struct ubuf_info *uarg,bool success); +void sock_zerocopy_callback (struct sk_buff *skb,struct ubuf_info *uarg, +bool success);int skb_zerocopy_iter_dgram (struct sk_buff *skb,struct msghdr *msg,int len); int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
@@ -1476,20 +1477,17 @@ static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)static inline void skb_zcopy_put (struct ubuf_info *uarg) {if (uarg) - uarg->callback (uarg, true); + uarg->callback (NULL , uarg, true); }/* Release a reference on a zerocopy structure */ -static inline void skb_zcopy_clear (struct sk_buff *skb,bool zerocopy) +static inline void skb_zcopy_clear (struct sk_buff *skb,bool success) {struct ubuf_info *uarg =skb_zcopy (skb);if (uarg) { -if (skb_zcopy_is_nouarg (skb)) { -/* no notification callback */ - }else { - uarg->callback (uarg, zerocopy); - } +if (!skb_zcopy_is_nouarg (skb)) + uarg->callback (skb, uarg, success);skb_shinfo (skb)->zc_flags &= ~SKBZC_FRAGMENTS ; }
net/core/skbuff.c +2 -1
@@ -1242,7 +1242,8 @@ static void __sock_zerocopy_callback(struct ubuf_info *uarg)sock_put (sk); } -void sock_zerocopy_callback (struct ubuf_info *uarg,bool success) +void sock_zerocopy_callback (struct sk_buff *skb,struct ubuf_info *uarg, +bool success) { uarg->zerocopy = uarg->zerocopy & success;
Signature · 1 line
2.24.1
RFC PATCH v2 8/12 skbuff: Call sock_zerocopy_put_abort from skb_zcopy_put_abort
In reply to Jonathan LemonFrom: Jonathan Lemon <[email protected]>
The sock_zerocopy_put_abort function contains logic which is specific to the current zerocopy implementation. Add a wrapper which checks the callback and dispatches apppropriately.
--- include/linux/skbuff.h | 10 ++++++++++ net/core/skbuff.c | 12 +++++------- net/ipv4/ip_output.c | 3 +-- net/ipv4/tcp.c | 2 +- net/ipv6/ip6_output.c | 3 +-- 5 files changed, 18 insertions(+), 12 deletions(-)
include/linux/skbuff.h +10 -0
@@ -1480,6 +1480,16 @@ static inline void skb_zcopy_put(struct ubuf_info *uarg) uarg->callback (NULL , uarg, true); } +static inline void skb_zcopy_put_abort (struct ubuf_info *uarg,bool have_uref) +{ +if (uarg) { +if (uarg->callback == sock_zerocopy_callback) +sock_zerocopy_put_abort (uarg, have_uref); +else if (have_uref) +skb_zcopy_put (uarg); + } +} +/* Release a reference on a zerocopy structure */ static inline void skb_zcopy_clear (struct sk_buff *skb,bool success) {
net/core/skbuff.c +5 -7
@@ -1254,15 +1254,13 @@ EXPORT_SYMBOL_GPL(sock_zerocopy_callback);void sock_zerocopy_put_abort (struct ubuf_info *uarg,bool have_uref) { -if (uarg) { -struct sock *sk =skb_from_uarg (uarg)->sk; +struct sock *sk =skb_from_uarg (uarg)->sk; -atomic_dec (&sk->sk_zckey); - uarg->len--; +atomic_dec (&sk->sk_zckey); + uarg->len--; -if (have_uref) -skb_zcopy_put (uarg); - } +if (have_uref) +sock_zerocopy_callback (NULL , uarg, true); }EXPORT_SYMBOL_GPL (sock_zerocopy_put_abort);
net/ipv4/ip_output.c +1 -2
@@ -1230,8 +1230,7 @@ static int __ip_append_data(struct sock *sk, error_efault: err = -EFAULT ; error: -if (uarg) -sock_zerocopy_put_abort (uarg, extra_uref); +skb_zcopy_put_abort (uarg, extra_uref); cork->length -= length;IP_INC_STATS (sock_net (sk),IPSTATS_MIB_OUTDISCARDS );refcount_add (wmem_alloc_delta, &sk->sk_wmem_alloc);
net/ipv4/tcp.c +1 -1
@@ -1440,7 +1440,7 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)if (copied + copied_syn) goto out; out_err: -sock_zerocopy_put_abort (uarg, true); +skb_zcopy_put_abort (uarg, true); err =sk_stream_error (sk, flags, err);/* make sure we wake any epoll edge trigger waiter */ if (unlikely (tcp_rtx_and_write_queues_empty (sk) && err == -EAGAIN )) {
net/ipv6/ip6_output.c +1 -2
@@ -1715,8 +1715,7 @@ static int __ip6_append_data(struct sock *sk, error_efault: err = -EFAULT ; error: -if (uarg) -sock_zerocopy_put_abort (uarg, extra_uref); +skb_zcopy_put_abort (uarg, extra_uref); cork->length -= length;IP6_INC_STATS (sock_net (sk), rt->rt6i_idev,IPSTATS_MIB_OUTDISCARDS );refcount_add (wmem_alloc_delta, &sk->sk_wmem_alloc);
Signature · 1 line
2.24.1
RFC PATCH v2 9/12 skbuff: add zc_flags to ubuf_info for ubuf setup
In reply to Jonathan LemonFrom: Jonathan Lemon <[email protected]>
Currently, an ubuf is attached to a new skb, the skb zc_flags is initialized to a fixed value. Instead of doing this, set the default zc_flags in the ubuf, and have new skb's inherit from this default.
This is needed when setting up different zerocopy types.
--- include/linux/skbuff.h | 3 ++- net/core/skbuff.c | 1 + 2 files changed, 3 insertions(+), 1 deletion(-)
include/linux/skbuff.h +2 -1
@@ -478,6 +478,7 @@ struct ubuf_info { }; };refcount_t refcnt; +u8 zc_flags;struct mmpin {struct user_struct *user;
@@ -1454,7 +1455,7 @@ static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,else skb_zcopy_get (uarg);skb_shinfo (skb)->destructor_arg = uarg; -skb_shinfo (skb)->zc_flags |=SKBZC_FRAGMENTS ; +skb_shinfo (skb)->zc_flags |= uarg->zc_flags; } }
net/core/skbuff.c +1 -0
@@ -1118,6 +1118,7 @@ struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size) uarg->len =1 ; uarg->bytelen = size; uarg->zerocopy =1 ; + uarg->zc_flags =SKBZC_FRAGMENTS ;refcount_set (&uarg->refcnt,1 );sock_hold (sk);
Signature · 1 line
2.24.1
Willem de Bruijn
maintainerRe: [PATCH 09/12 v2 RFC] skbuff: add zc_flags to ubuf_info for ubuf setup
Quoted 21 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
Currently, an ubuf is attached to a new skb, the skb
zc_flagsis initialized to a fixed value. Instead of doing this, set the defaultzc_flagsin the ubuf, and have new skb's inherit from this default.This is needed when setting up different zerocopy types.
Signed-off-by Jonathan Lemon <[email protected]>--- include/linux/skbuff.h | 3 ++- net/core/skbuff.c | 1 + 2 files changed, 3 insertions(+), 1 deletion(-)include/linux/skbuff.h +0
-0
include/linux/skbuff.h struct ubuf_info { · line 484
Quoted 7 lines
include/linux/skbuff.h +1
-0@@ -478,6 +478,7 @@ struct ubuf_info { }; };refcount_t refcnt; +u8 zc_flags;struct mmpin {struct user_struct *user;
When allocating ubuf_info for msg_zerocopy, we actually allocate the notification skb, to be sure that notifications won't be dropped due to memory pressure at notification time. We actually allocate the skb and place ubuf_info in skb->cb[].
The struct is exactly 48 bytes on 64-bit platforms, filling all of cb. This new field fills a 4B hole, so it should still be fine.
Just being very explicit, as this is a fragile bit of code. Come to think of it, this probably deserves a BUILD_BUG_ON.
Re: [PATCH 09/12 v2 RFC] skbuff: add zc_flags to ubuf_info for ubuf setup
Quoted 35 lines · Willem de Bruijn
Quoted 21 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
Currently, an ubuf is attached to a new skb, the skb
zc_flagsis initialized to a fixed value. Instead of doing this, set the defaultzc_flagsin the ubuf, and have new skb's inherit from this default.This is needed when setting up different zerocopy types.
Signed-off-by Jonathan Lemon <[email protected]>--- include/linux/skbuff.h | 3 ++- net/core/skbuff.c | 1 + 2 files changed, 3 insertions(+), 1 deletion(-)include/linux/skbuff.h +0
-0include/linux/skbuff.h struct ubuf_info { · line 484
Quoted 7 lines
include/linux/skbuff.h +1
-0@@ -478,6 +478,7 @@ struct ubuf_info { }; };refcount_t refcnt; +u8 zc_flags;struct mmpin {struct user_struct *user;When allocating
ubuf_infoformsg_zerocopy, we actually allocate the notification skb, to be sure that notifications won't be dropped due to memory pressure at notification time. We actually allocate the skb and placeubuf_infoin skb->cb[].
Quoted 2 lines
The struct is exactly 48 bytes on 64-bit platforms, filling all of cb. This new field fills a 4B hole, so it should still be fine.
Yes, I was careful not to increase the size. I have future changes
for this structure, moving 'struct mmpin' into a union. Making the
flags 16 bits shouldn't be a problem either.
Quoted 2 lines
Just being very explicit, as this is a fragile bit of code. Come to think of it, this probably deserves a
BUILD_BUG_ON.
You mean like the one which exists in sock_zerocopy_alloc()?
BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
Signature · 1 line
Jonathan
Willem de Bruijn
maintainerRe: [PATCH 09/12 v2 RFC] skbuff: add zc_flags to ubuf_info for ubuf setup
Quoted 45 lines · Jonathan Lemon
Quoted 35 lines · Willem de Bruijn
Quoted 21 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
Currently, an ubuf is attached to a new skb, the skb
zc_flagsis initialized to a fixed value. Instead of doing this, set the defaultzc_flagsin the ubuf, and have new skb's inherit from this default.This is needed when setting up different zerocopy types.
Signed-off-by Jonathan Lemon <[email protected]>--- include/linux/skbuff.h | 3 ++- net/core/skbuff.c | 1 + 2 files changed, 3 insertions(+), 1 deletion(-)include/linux/skbuff.h +0
-0include/linux/skbuff.h struct ubuf_info { · line 484
Quoted 7 lines
include/linux/skbuff.h +1
-0@@ -478,6 +478,7 @@ struct ubuf_info { }; };refcount_t refcnt; +u8 zc_flags;struct mmpin {struct user_struct *user;When allocating
ubuf_infoformsg_zerocopy, we actually allocate the notification skb, to be sure that notifications won't be dropped due to memory pressure at notification time. We actually allocate the skb and placeubuf_infoin skb->cb[].Quoted 2 lines
The struct is exactly 48 bytes on 64-bit platforms, filling all of cb. This new field fills a 4B hole, so it should still be fine.
Yes, I was careful not to increase the size. I have future changes
for this structure, moving 'struct mmpin' into a union. Making the
flags 16 bits shouldn't be a problem either.
Quoted 6 lines
Quoted 2 lines
Just being very explicit, as this is a fragile bit of code. Come to think of it, this probably deserves a
BUILD_BUG_ON.You mean like the one which exists in
sock_zerocopy_alloc()?BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
Ah good. Yes. I should have remembered that was there.
RFC PATCH v2 10/12 tap/tun: use skb_zcopy_set() instead of open coded assignment
In reply to Jonathan LemonFrom: Jonathan Lemon <[email protected]>
Replace direct assignments with skb_zcopy_set() for clarity.
--- drivers/net/tap.c | 3 +-- drivers/net/tun.c | 3 +-- drivers/vhost/net.c | 1 + 3 files changed, 3 insertions(+), 4 deletions(-)
drivers/net/tap.c +1 -2
@@ -722,8 +722,7 @@ static ssize_t tap_get_user(struct tap_queue *q, void *msg_control, tap =rcu_dereference (q->tap);/* copy skb_ubuf_info for callback when skb has no error */ if (zerocopy) { -skb_shinfo (skb)->destructor_arg = msg_control; -skb_shinfo (skb)->zc_flags |=SKBZC_FRAGMENTS ; +skb_zcopy_set (skb, msg_control,NULL ); }else if (msg_control) {struct ubuf_info *uarg = msg_control; uarg->callback (NULL , uarg, false);
drivers/net/tun.c +1 -2
@@ -1814,8 +1814,7 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,/* copy skb_ubuf_info for callback when skb has no error */ if (zerocopy) { -skb_shinfo (skb)->destructor_arg = msg_control; -skb_shinfo (skb)->zc_flags |=SKBZC_FRAGMENTS ; +skb_zcopy_set (skb, msg_control,NULL ); }else if (msg_control) {struct ubuf_info *uarg = msg_control; uarg->callback (NULL , uarg, false);
drivers/vhost/net.c +1 -0
@@ -904,6 +904,7 @@ static void handle_tx_zerocopy(struct vhost_net *net, struct socket *sock) ubuf->callback = vhost_zerocopy_callback; ubuf->ctx = nvq->ubufs; ubuf->desc = nvq->upend_idx; + ubuf->zc_flags =SKBZC_FRAGMENTS ;refcount_set (&ubuf->refcnt,1 ); msg.msg_control = &ctl; ctl.type =TUN_MSG_UBUF ;
Signature · 1 line
2.24.1
RFC PATCH v2 11/12 skbuff: Call skb_zcopy_clear() before unref'ing fragments
In reply to Jonathan LemonFrom: Jonathan Lemon <[email protected]>
RX zerocopy fragment pages which are not allocated from the system page pool require special handling. Give the callback in skb_zcopy_clear() a chance to process them first.
--- net/core/skbuff.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
net/core/skbuff.c +2 -1
@@ -605,13 +605,14 @@ static void skb_release_data(struct sk_buff *skb) &shinfo->dataref))return ; +skb_zcopy_clear (skb, true); +for (i =0 ; i < shinfo->nr_frags; i++)__skb_frag_unref (&shinfo->frags[i]);if (shinfo->frag_list)kfree_skb_list (shinfo->frag_list); -skb_zcopy_clear (skb, true);skb_free_head (skb); }
Signature · 1 line
2.24.1
RFC PATCH v2 12/12 skbuff: rename sock_zerocopy_* to msg_zerocopy_*
In reply to Jonathan LemonFrom: Jonathan Lemon <[email protected]>
At Willem's suggestion, rename the sock_zerocopy_* functions so that they match the MSG_ZEROCOPY flag, which makes it clear they are specific to this zerocopy implementation.
--- include/linux/skbuff.h | 16 ++++++++-------- net/core/skbuff.c | 28 ++++++++++++++-------------- net/ipv4/ip_output.c | 2 +- net/ipv4/tcp.c | 2 +- net/ipv6/ip6_output.c | 2 +- 5 files changed, 25 insertions(+), 25 deletions(-)
include/linux/skbuff.h +8 -8
@@ -491,13 +491,13 @@ struct ubuf_info {int mm_account_pinned_pages (struct mmpin *mmp,size_t size);void mm_unaccount_pinned_pages (struct mmpin *mmp); -struct ubuf_info *sock_zerocopy_alloc (struct sock *sk,size_t size); -struct ubuf_info *sock_zerocopy_realloc (struct sock *sk,size_t size, -struct ubuf_info *uarg); +struct ubuf_info *msg_zerocopy_alloc (struct sock *sk,size_t size); +struct ubuf_info *msg_zerocopy_realloc (struct sock *sk,size_t size, +struct ubuf_info *uarg); -void sock_zerocopy_put_abort (struct ubuf_info *uarg,bool have_uref); +void msg_zerocopy_put_abort (struct ubuf_info *uarg,bool have_uref); -void sock_zerocopy_callback (struct sk_buff *skb,struct ubuf_info *uarg, +void msg_zerocopy_callback (struct sk_buff *skb,struct ubuf_info *uarg,bool success);int skb_zerocopy_iter_dgram (struct sk_buff *skb,struct msghdr *msg,int len);
@@ -1484,8 +1484,8 @@ static inline void skb_zcopy_put(struct ubuf_info *uarg)static inline void skb_zcopy_put_abort (struct ubuf_info *uarg,bool have_uref) {if (uarg) { -if (uarg->callback == sock_zerocopy_callback) -sock_zerocopy_put_abort (uarg, have_uref); +if (uarg->callback == msg_zerocopy_callback) +msg_zerocopy_put_abort (uarg, have_uref);else if (have_uref)skb_zcopy_put (uarg); }
@@ -2779,7 +2779,7 @@ static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)if (likely (!skb_zcopy (skb)))return 0 ;if (!skb_zcopy_is_nouarg (skb) && -skb_uarg (skb)->callback == sock_zerocopy_callback) +skb_uarg (skb)->callback == msg_zerocopy_callback)return 0 ;return skb_copy_ubufs (skb, gfp_mask); }
net/core/skbuff.c +14 -14
@@ -1094,7 +1094,7 @@ void mm_unaccount_pinned_pages(struct mmpin *mmp) }EXPORT_SYMBOL_GPL (mm_unaccount_pinned_pages); -struct ubuf_info *sock_zerocopy_alloc (struct sock *sk,size_t size) +struct ubuf_info *msg_zerocopy_alloc (struct sock *sk,size_t size) {struct ubuf_info *uarg;struct sk_buff *skb;
@@ -1114,7 +1114,7 @@ struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)return NULL ; } - uarg->callback = sock_zerocopy_callback; + uarg->callback = msg_zerocopy_callback; uarg->id = ((u32 )atomic_inc_return (&sk->sk_zckey)) -1 ; uarg->len =1 ; uarg->bytelen = size;
@@ -1125,15 +1125,15 @@ struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)return uarg; } -EXPORT_SYMBOL_GPL (sock_zerocopy_alloc); +EXPORT_SYMBOL_GPL (msg_zerocopy_alloc);static inline struct sk_buff *skb_from_uarg (struct ubuf_info *uarg) { return container_of((void *)uarg, struct sk_buff, cb); } -struct ubuf_info *sock_zerocopy_realloc (struct sock *sk,size_t size, -struct ubuf_info *uarg) +struct ubuf_info *msg_zerocopy_realloc (struct sock *sk,size_t size, +struct ubuf_info *uarg) {if (uarg) {const u32 byte_limit =1 <<19 ;/* limit to a few TSO */
@@ -1172,9 +1172,9 @@ struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size, } new_alloc: -return sock_zerocopy_alloc (sk, size); +return msg_zerocopy_alloc (sk, size); } -EXPORT_SYMBOL_GPL (sock_zerocopy_realloc); +EXPORT_SYMBOL_GPL (msg_zerocopy_realloc);static bool skb_zerocopy_notify_extend (struct sk_buff *skb,u32 lo,u16 len) {
@@ -1196,7 +1196,7 @@ static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)return true; } -static void __sock_zerocopy_callback (struct ubuf_info *uarg) +static void __msg_zerocopy_callback (struct ubuf_info *uarg) {struct sk_buff *tail, *skb =skb_from_uarg (uarg);struct sock_exterr_skb *serr;
@@ -1244,17 +1244,17 @@ static void __sock_zerocopy_callback(struct ubuf_info *uarg)sock_put (sk); } -void sock_zerocopy_callback (struct sk_buff *skb,struct ubuf_info *uarg, +void msg_zerocopy_callback (struct sk_buff *skb,struct ubuf_info *uarg,bool success) { uarg->zerocopy = uarg->zerocopy & success;if (refcount_dec_and_test (&uarg->refcnt)) -__sock_zerocopy_callback (uarg); +__msg_zerocopy_callback (uarg); } -EXPORT_SYMBOL_GPL (sock_zerocopy_callback); +EXPORT_SYMBOL_GPL (msg_zerocopy_callback); -void sock_zerocopy_put_abort (struct ubuf_info *uarg,bool have_uref) +void msg_zerocopy_put_abort (struct ubuf_info *uarg,bool have_uref) {struct sock *sk =skb_from_uarg (uarg)->sk;
@@ -1262,9 +1262,9 @@ void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref) uarg->len--;if (have_uref) -sock_zerocopy_callback (NULL , uarg, true); +msg_zerocopy_callback (NULL , uarg, true); } -EXPORT_SYMBOL_GPL (sock_zerocopy_put_abort); +EXPORT_SYMBOL_GPL (msg_zerocopy_put_abort);int skb_zerocopy_iter_dgram (struct sk_buff *skb,struct msghdr *msg,int len) {
net/ipv4/ip_output.c +1 -1
@@ -1018,7 +1018,7 @@ static int __ip_append_data(struct sock *sk, csummode =CHECKSUM_PARTIAL ;if (flags &MSG_ZEROCOPY && length &&sock_flag (sk,SOCK_ZEROCOPY )) { - uarg =sock_zerocopy_realloc (sk, length,skb_zcopy (skb)); + uarg =msg_zerocopy_realloc (sk, length,skb_zcopy (skb));if (!uarg)return -ENOBUFS ; extra_uref = !skb_zcopy (skb);/* only ref on new uarg */
net/ipv4/tcp.c +1 -1
@@ -1217,7 +1217,7 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)if (flags &MSG_ZEROCOPY && size &&sock_flag (sk,SOCK_ZEROCOPY )) { skb =tcp_write_queue_tail (sk); - uarg =sock_zerocopy_realloc (sk, size,skb_zcopy (skb)); + uarg =msg_zerocopy_realloc (sk, size,skb_zcopy (skb));if (!uarg) { err = -ENOBUFS ; goto out_err;
net/ipv6/ip6_output.c +1 -1
@@ -1471,7 +1471,7 @@ static int __ip6_append_data(struct sock *sk, csummode =CHECKSUM_PARTIAL ;if (flags &MSG_ZEROCOPY && length &&sock_flag (sk,SOCK_ZEROCOPY )) { - uarg =sock_zerocopy_realloc (sk, length,skb_zcopy (skb)); + uarg =msg_zerocopy_realloc (sk, length,skb_zcopy (skb));if (!uarg)return -ENOBUFS ; extra_uref = !skb_zcopy (skb);/* only ref on new uarg */
Signature · 1 line
2.24.1
Re: [PATCH 12/12 v2 RFC] skbuff: rename sock_zerocopy_* to msg_zerocopy_*
Quoted 8 lines · Jonathan Lemon
From: Jonathan Lemon <[email protected]>
At Willem's suggestion, rename the sock_zerocopy_* functions so that they match the
MSG_ZEROCOPYflag, which makes it clear they are specific to this zerocopy implementation.Signed-off-by Jonathan Lemon <[email protected]>