- New feature flag, NL80211_EXT_FEATURE_PUNCT, to advertise
driver support for preamble puncturing in AP mode.
- New attribute, NL80211_ATTR_PUNCT_BITMAP, to receive a puncturing
bitmap from the userspace during AP bring up (NL80211_CMD_START_AP)
and channel switch (NL80211_CMD_CHANNEL_SWITCH) operations. Each bit
corresponds to a 20 MHz channel in the operating bandwidth, lowest
bit for the lowest channel. Bit set to 1 indicates that the channel
is punctured. Higher 16 bits are reserved.
- New members added to structures cfg80211_ap_settings and
cfg80211_csa_settings to propagate the bitmap to the driver after
validation.
Signed-off-by: Aloka Dixit <quic_alokad@quicinc.com>
Signed-off-by: Muna Sinada <quic_msinada@quicinc.com>
Link: https://lore.kernel.org/r/20230131001227.25014-3-quic_alokad@quicinc.com
[move validation against 0xffff into policy]
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
- Move ieee80211_valid_disable_subchannel_bitmap() from mlme.c to
chan.c, rename it as cfg80211_valid_disable_subchannel_bitmap()
and export it.
- Modify the prototype to include struct cfg80211_chan_def instead
of only bandwidth to support a check which returns false if the
primary channel is punctured.
Signed-off-by: Aloka Dixit <quic_alokad@quicinc.com>
Link: https://lore.kernel.org/r/20230131001227.25014-2-quic_alokad@quicinc.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Userspace processes such as network daemons may wish to be informed when
any AP interface is brought up on the system, for example to initiate a
(re)configuration of IP settings or to start a DHCP server.
Currently nl80211 does not broadcast any such event on its multicast
groups, leaving userspace only two options:
1. the process must be the one that actually issued the
NL80211_CMD_START_AP request, so that it can react on the response to
that request;
2. the process must react to RTM_NEWLINK events indicating a change in
carrier state, and may query for further information about the AP and
react accordingly.
Option (1) is robust, but it does not cover all scenarios. It is easy to
imagine a situation where this is not the case (e.g. hostapd +
systemd-networkd).
Option (2) is not robust, because RTM_NEWLINK events may be silently
discarded by the linkwatch logic (cf. linkwatch_fire_event()).
Concretely, consider a scenario in which the carrier state flip-flops in
the following way:
^ carrier state (high/low = carrier/no carrier)
|
| _______ _______ ...
| | | |
| ______| "foo" |____| "bar" (SSID in "quotes")
|
+-------A-------B----C---------> time
If the time interval between (A) and (C) is less than 1 second, then
linkwatch may emit only a single RTM_NEWLINK event indicating carrier
gain.
This is problematic because it is possible that the network
configuration that should be applied is a function of the AP's
properties such as SSID (cf. SSID= in systemd.network(5)). As
illustrated in the above diagram, it may be that the AP with SSID "bar"
ends up being configured as though it had SSID "foo".
Address the above issue by having nl80211 emit an NL80211_CMD_START_AP
message on the MLME nl80211 multicast group. This allows for arbitrary
processes to be reliably informed.
Signed-off-by: Alvin Šipraga <alsi@bang-olufsen.dk>
Link: https://lore.kernel.org/r/20230128125844.2407135-1-alvin@pqrs.dk
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Add support to offload OWE processing to user space for MLD AP when
driver's SME in use.
Add new parameters in struct cfg80211_update_owe_info to provide below
information in cfg80211_update_owe_info_event() call:
- MLO link ID of the AP, with which station requested (re)association.
This is applicable for both MLO and non-MLO station connections when
the AP affiliated with an MLD.
- Station's MLD address if the connection is MLO capable.
Signed-off-by: Veerendranath Jakkam <quic_vjakkam@quicinc.com>
Link: https://lore.kernel.org/r/20230126143256.960563-3-quic_vjakkam@quicinc.com
[reformat the trace event macro]
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Add support for drivers to indicate STA connection(MLO/non-MLO) when
user space SME (e.g., hostapd) is not used for MLD AP.
Add new parameters in struct station_info to provide below information
in cfg80211_new_sta() call:
- MLO link ID of the AP, with which station completed (re)association.
This is applicable for both MLO and non-MLO station connections when
the AP affiliated with an MLD.
- Station's MLD address if the connection is MLO capable.
- (Re)Association Response IEs sent to the station. User space needs
this to determine rejected and accepted affiliated links information
of the connected station if the connection is MLO capable.
Signed-off-by: Veerendranath Jakkam <quic_vjakkam@quicinc.com>
Link: https://lore.kernel.org/r/20230126143256.960563-2-quic_vjakkam@quicinc.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Currently authentication request event interface doesn't have support to
indicate the user space whether it should enable MLO or not during the
authentication with the specified AP. But driver needs such capability
since the connection is MLO or not decided by the driver in case of SME
offload to the driver.
Add support for driver to indicate MLD address of the AP in
authentication offload request to inform user space to enable MLO during
authentication process. Driver shall look at NL80211_ATTR_MLO_SUPPORT
flag capability in NL80211_CMD_CONNECT to know whether the user space
supports enabling MLO during the authentication offload.
User space should enable MLO during the authentication only when it
receives the AP MLD address in authentication offload request. User
space shouldn't enable MLO if the authentication offload request doesn't
indicate the AP MLD address even if the AP is MLO capable.
When MLO is enabled, user space should use the MAC address of the
interface (on which driver sent request) as self MLD address. User space
and driver to use MLD addresses in RA, TA and BSSID fields of the frames
between them, and driver translates the MLD addresses to/from link
addresses based on the link chosen for the authentication.
Signed-off-by: Veerendranath Jakkam <quic_vjakkam@quicinc.com>
Link: https://lore.kernel.org/r/20230116125058.1604843-1-quic_vjakkam@quicinc.com
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Oz Shlomo says:
====================
add support for per action hw stats
There are currently two mechanisms for populating hardware stats:
1. Using flow_offload api to query the flow's statistics.
The api assumes that the same stats values apply to all
the flow's actions.
This assumption breaks when action drops or jumps over following
actions.
2. Using hw_action api to query specific action stats via a driver
callback method. This api assures the correct action stats for
the offloaded action, however, it does not apply to the rest of the
actions in the flow's actions array, as elaborated below.
The current hw_action api does not apply to the following use cases:
1. Actions that are implicitly created by filters (aka bind actions).
In the following example only one counter will apply to the rule:
tc filter add dev $DEV prio 2 protocol ip parent ffff: \
flower ip_proto tcp dst_ip $IP2 \
action police rate 1mbit burst 100k conform-exceed drop/pipe \
action mirred egress redirect dev $DEV2
2. Action preceding a hw action.
In the following example the same flow stats will apply to the sample and
mirred actions:
tc action add police rate 1mbit burst 100k conform-exceed drop / pipe
tc filter add dev $DEV prio 2 protocol ip parent ffff: \
flower ip_proto tcp dst_ip $IP2 \
action sample rate 1 group 10 trunc 60 pipe \
action police index 1 \
action mirred egress redirect dev $DEV2
3. Meter action using jump control.
In the following example the same flow stats will apply to both
mirred actions:
tc action add police rate 1mbit burst 100k conform-exceed jump 2 / pipe
tc filter add dev $DEV prio 2 protocol ip parent ffff: \
flower ip_proto tcp dst_ip $IP2 \
action police index 1 \
action mirred egress redirect dev $DEV2
action mirred egress redirect dev $DEV3
This series provides the platform to query per action stats for in_hw flows.
The first four patches are preparation patches with no functionality change.
The fifth patch re-uses the existing flow action stats api to query action
stats for both classifier and action dumps.
The rest of the patches add per action stats support to the Mellanox driver.
====================
Link: https://lore.kernel.org/r/20230212132520.12571-1-ozsh@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Extend the action stats callback implementation to update stats for actions
that are associated with hw counters.
Note that the callback may be called from tc action utility or from tc
flower. Both apis expect the driver to return the stats difference from
the last update. As such, query the raw counter value and maintain
the diff from the last api call in the tc layer, instead of the fs_core
layer.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Currently a hardware counter is associated with a flow cookie.
This does not apply to flows using branching action which are required to
return per action stats.
A single counter may apply to multiple actions.
Scan the flow actions in reverse (from the last to the first action) while
caching the last counter.
Associate all the flow attribute tc action cookies with the current
cached counter.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The tc parse action phase translates the tc actions to mlx5 flow
attributes data structure that is used during the flow offload phase.
Currently, the flow offload stage instantiates hw counters while
associating them to flow cookie. However, flows with branching
actions are required to associate a hardware counter with its action
cookies.
Store the parsed tc action cookies on the flow attribute.
Use the list of cookies in the next patch to associate a tc action cookie
with its allocated hw counter.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Currently a hw count action is appended to the last action of the action
list. However, a branching action may terminate the action list before
reaching the last action.
Append a count action to a branching action.
In the next patches, filters with branching actions will read this counter
when reporting stats per action.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
There are currently two mechanisms for populating hardware stats:
1. Using flow_offload api to query the flow's statistics.
The api assumes that the same stats values apply to all
the flow's actions.
This assumption breaks when action drops or jumps over following
actions.
2. Using hw_action api to query specific action stats via a driver
callback method. This api assures the correct action stats for
the offloaded action, however, it does not apply to the rest of the
actions in the flow's actions array.
Extend the flow_offload stats callback to indicate that a per action
stats update is required.
Use the existing flow_offload_action api to query the action's hw stats.
In addition, currently the tc action stats utility only updates hw actions.
Reuse the existing action stats cb infrastructure to query any action
stats.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Currently a hardware action is uniquely identified by the <id, hw_index>
tuple. However, the id is set by the flow_act_setup callback and tc core
cannot enforce this, and it is possible that a future change could break
this. In addition, <id, hw_index> are not unique across network namespaces.
Uniquely identify the action by setting an action cookie by the tc core.
Use the unique action cookie to query the action's hardware stats.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
A single tc pedit action may be translated to multiple flow_offload
actions.
Offload only actions that translate to a single pedit command value.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Currently the hw action stats update is called from tcf_exts_hw_stats_update,
when a tc filter is dumped, and from tcf_action_copy_stats, when a hw
action is dumped.
However, the tcf_action_copy_stats is also called from tcf_action_dump.
As such, the hw action stats update cb is called 3 times for every
tc flower filter dump.
Move the tc action hw stats update from tcf_action_copy_stats to
tcf_dump_walker to update the hw action stats when tc action is dumped.
Signed-off-by: Oz Shlomo <ozsh@nvidia.com>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Reviewed-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
When setting 'snps,force_thresh_dma_mode' DT property, the following
warning is always emitted, regardless the status of force_sf_dma_mode:
dwmac-starfive 10020000.ethernet: force_sf_dma_mode is ignored if force_thresh_dma_mode is set.
Do not print the rather misleading message when DMA store and forward
mode is already disabled.
Fixes: e2a240c7d3 ("driver:net:stmmac: Disable DMA store and forward mode if platform data force_thresh_dma_mode is set.")
Signed-off-by: Cristian Ciocaltea <cristian.ciocaltea@collabora.com>
Link: https://lore.kernel.org/r/20230210202126.877548-1-cristian.ciocaltea@collabora.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This reverts commit 0243d3dfe2.
pahole 1.25 is too aggressive removing functions.
With clang compiled kernel the following is seen:
WARN: resolve_btfids: unresolved symbol tcp_reno_cong_avoid
WARN: resolve_btfids: unresolved symbol dctcp_update_alpha
WARN: resolve_btfids: unresolved symbol cubictcp_cong_avoid
WARN: resolve_btfids: unresolved symbol bpf_xdp_metadata_rx_timestamp
WARN: resolve_btfids: unresolved symbol bpf_xdp_metadata_rx_hash
WARN: resolve_btfids: unresolved symbol bpf_task_kptr_get
WARN: resolve_btfids: unresolved symbol bpf_task_acquire_not_zero
WARN: resolve_btfids: unresolved symbol bpf_rdonly_cast
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test_static_unused_arg
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test_ref
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test_pass_ctx
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test_pass2
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test_pass1
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test_mem_len_pass1
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test_mem_len_fail2
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test_mem_len_fail1
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test_kptr_get
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test_fail3
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test_fail2
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test_acquire
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test2
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_test1
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_memb_release
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_memb1_release
WARN: resolve_btfids: unresolved symbol bpf_kfunc_call_int_mem_release
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Set the DMA mask before calling dma_addressing_limited, which depends on it.
Note that this stop checking the return value of dma_set_mask_and_coherent
as this function can only fail for masks < 32-bit.
Fixes: 3f30a79c2e ("nvme-pci: set constant paramters in nvme_pci_alloc_ctrl")
Reported-by: Michael Kelley <mikelley@microsoft.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Tested-by: Michael Kelley <mikelley@microsoft.com>
The tc action act_ctinfo was using shared stats, fix it to use percpu stats
since bstats_update() must be called with locks or with a percpu pointer argument.
tdc results:
1..12
ok 1 c826 - Add ctinfo action with default setting
ok 2 0286 - Add ctinfo action with dscp
ok 3 4938 - Add ctinfo action with valid cpmark and zone
ok 4 7593 - Add ctinfo action with drop control
ok 5 2961 - Replace ctinfo action zone and action control
ok 6 e567 - Delete ctinfo action with valid index
ok 7 6a91 - Delete ctinfo action with invalid index
ok 8 5232 - List ctinfo actions
ok 9 7702 - Flush ctinfo actions
ok 10 3201 - Add ctinfo action with duplicate index
ok 11 8295 - Add ctinfo action with invalid index
ok 12 3964 - Replace ctinfo action with invalid goto_chain control
Fixes: 24ec483cec ("net: sched: Introduce act_ctinfo action")
Reviewed-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Pedro Tammela <pctammela@mojatatu.com>
Reviewed-by: Larysa Zaremba <larysa.zaremba@intel.com>
Link: https://lore.kernel.org/r/20230210200824.444856-1-pctammela@mojatatu.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
So far changing the period by just setting new period values while
running did not work.
The order as indicated by the publicly available reference manual of the i.MX8MP [1]
indicates a sequence:
* initiate the programming sequence
* set the values for PPS period and start time
* start the pulse train generation.
This is currently not used in dwmac5_flex_pps_config(), which instead does:
* initiate the programming sequence and immediately start the pulse train generation
* set the values for PPS period and start time
This caused the period values written not to take effect until the FlexPPS output was
disabled and re-enabled again.
This patch fix the order and allows the period to be set immediately.
[1] https://www.nxp.com/webapp/Download?colCode=IMX8MPRM
Fixes: 9a8a02c9d4 ("net: stmmac: Add Flexible PPS support")
Signed-off-by: Johannes Zink <j.zink@pengutronix.de>
Link: https://lore.kernel.org/r/20230210143937.3427483-1-j.zink@pengutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Creates three new drop reasons:
SKB_DROP_REASON_IPV6_NDISC_FRAG: invalid frag (suppress_frag_ndisc).
SKB_DROP_REASON_IPV6_NDISC_HOP_LIMIT: invalid hop limit.
SKB_DROP_REASON_IPV6_NDISC_BAD_CODE: invalid NDISC icmp6 code.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Accurately reports what happened in icmpv6_notify() when handling
a packet.
This makes use of the new IPV6_BAD_EXTHDR drop reason.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
pskb_may_pull() can fail for two different reasons.
Provide pskb_may_pull_reason() helper to distinguish
between these reasons.
It returns:
SKB_NOT_DROPPED_YET : Success
SKB_DROP_REASON_PKT_TOO_SMALL : packet too small
SKB_DROP_REASON_NOMEM : skb->head could not be resized
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This drop reason can be used whenever an IPv6 packet
has a malformed extension header.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Ahern <dsahern@kernel.org>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Dave Marchevsky says:
====================
This series adds a rbtree datastructure following the "next-gen
datastructure" precedent set by recently-added linked-list [0]. This is
a reimplementation of previous rbtree RFC [1] to use kfunc + kptr
instead of adding a new map type. This series adds a smaller set of API
functions than that RFC - just the minimum needed to support current
cgfifo example scheduler in ongoing sched_ext effort [2], namely:
bpf_rbtree_add
bpf_rbtree_remove
bpf_rbtree_first
The meat of this series is bugfixes and verifier infra work to support
these API functions. Adding more rbtree kfuncs in future patches should
be straightforward as a result.
First, the series refactors and extends linked_list's release_on_unlock
logic. The concept of "reference to node that was added to data
structure" is formalized as "non-owning reference". From linked_list's
perspective this non-owning reference after
linked_list_push_{front,back} has same semantics as release_on_unlock,
with the addition of writes to such references being valid in the
critical section. Such references are no longer marked PTR_UNTRUSTED.
Patches 2 and 13 go into more detail.
The series then adds rbtree API kfuncs and necessary verifier support
for them - namely support for callback args to kfuncs and some
non-owning reference interactions that linked_list didn't need.
BPF rbtree uses struct rb_root_cached + existing rbtree lib under the
hood. From the BPF program writer's perspective, a BPF rbtree is very
similar to existing linked list. Consider the following example:
struct node_data {
long key;
long data;
struct bpf_rb_node node;
}
static bool less(struct bpf_rb_node *a, const struct bpf_rb_node *b)
{
struct node_data *node_a;
struct node_data *node_b;
node_a = container_of(a, struct node_data, node);
node_b = container_of(b, struct node_data, node);
return node_a->key < node_b->key;
}
private(A) struct bpf_spin_lock glock;
private(A) struct bpf_rb_root groot __contains(node_data, node);
/* ... in BPF program */
struct node_data *n, *m;
struct bpf_rb_node *res;
n = bpf_obj_new(typeof(*n));
if (!n)
/* skip */
n->key = 5;
n->data = 10;
bpf_spin_lock(&glock);
bpf_rbtree_add(&groot, &n->node, less);
bpf_spin_unlock(&glock);
bpf_spin_lock(&glock);
res = bpf_rbtree_first(&groot);
if (!res)
/* skip */
res = bpf_rbtree_remove(&groot, res);
if (!res)
/* skip */
bpf_spin_unlock(&glock);
m = container_of(res, struct node_data, node);
bpf_obj_drop(m);
Some obvious similarities:
* Special bpf_rb_root and bpf_rb_node types have same semantics
as bpf_list_head and bpf_list_node, respectively
* __contains is used to associated node type with root
* The spin_lock associated with a rbtree must be held when using
rbtree API kfuncs
* Nodes are allocated via bpf_obj_new and dropped via bpf_obj_drop
* Rbtree takes ownership of node lifetime when a node is added.
Removing a node gives ownership back to the program, requiring a
bpf_obj_drop before program exit
Some new additions as well:
* Support for callbacks in kfunc args is added to enable 'less'
callback use above
* bpf_rbtree_first is the first graph API function to return a
non-owning reference instead of convering an arg from own->non-own
* Because all references to nodes already added to the rbtree are
non-owning, bpf_rbtree_remove must accept such a reference in order
to remove it from the tree
Summary of patches:
Patches 1 - 5 implement the meat of rbtree-specific support in this
series, gradually building up to implemented kfuncs that verify as
expected.
Patch 6 adds the bpf_rbtree_{add,first,remove} to bpf_experimental.h.
Patch 7 adds tests, Patch 9 adds documentation.
[0]: lore.kernel.org/bpf/20221118015614.2013203-1-memxor@gmail.com
[1]: lore.kernel.org/bpf/20220830172759.4069786-1-davemarchevsky@fb.com
[2]: lore.kernel.org/bpf/20221130082313.3241517-1-tj@kernel.org
Changelog:
v5 -> v6: lore.kernel.org/bpf/20230212092715.1422619-1-davemarchevsky@fb.com/
Patch #'s below refer to the patch's number in v5 unless otherwise stated.
* General / Patch 1
* Rebase onto latest bpf-next: "bpf: Migrate release_on_unlock logic to non-owning ref semantics"
* This was Patch 1 of v4, was applied, not included in v6
* Patch 3 - "bpf: Add bpf_rbtree_{add,remove,first} kfuncs"
* Use bpf_callback_t instead of plain-C fn ptr for bpf_rbtree_add. This
necessitated having bpf_rbtree_add duplicate rbtree_add's functionality.
Wrapper function was used w/ internal __bpf_rbtree_add helper so that
bpf_experimental.h proto could continue to use plain-C fn ptr so BPF progs
could benefit from typechecking (Alexei)
v4 -> v5: lore.kernel.org/bpf/20230209174144.3280955-1-davemarchevsky@fb.com/
Patch #'s below refer to the patch's number in v4 unless otherwise stated.
* General
* Rebase onto latest bpf-next: "Merge branch 'bpf, mm: introduce cgroup.memory=nobpf'"
* Patches 1-3 are squashed into "bpf: Migrate release_on_unlock logic to non-owning ref semantics".
* Added type_is_non_owning_ref helper (Alexei)
* Use a NON_OWN_REF type flag instead of separate bool (Alexei)
* Patch 8 - "bpf: Special verifier handling for bpf_rbtree_{remove, first}"
* When doing btf_parse_fields, reject structs with both bpf_list_node and
bpf_rb_node fields. This is a temporary measure that can be removed after
"collection identity" followup. See comment added in btf_parse_fields for
more detail (Kumar, Alexei)
* Add linked_list BTF test exercising check added to btf_parse_fields
* Minor changes and moving around of some reg type checks due to NON_OWN_REF type flag
introduction
* Patch 10 - "selftests/bpf: Add rbtree selftests"
* Migrate failure tests to RUN_TESTS, __failure, __msg() framework (Alexei)
v3 -> v4: lore.kernel.org/bpf/20230131180016.3368305-1-davemarchevsky@fb.com/
Patch #'s below refer to the patch's number in v3 unless otherwise stated.
* General
* Don't base this series on "bpf: Refactor release_regno searching logic",
which was submitted separately as a refactor.
* Rebase onto latest bpf-next: "samples/bpf: Add openat2() enter/exit tracepoint to syscall_tp sample"
* Patch 2 - "bpf: Improve bpf_reg_state space usage for non-owning ref lock"
* print_verifier_state change was adding redundant comma after "non_own_ref",
fix it to put comma in correct place
* invalidate_non_owning_refs no longer needs to take bpf_active_lock param,
since any non-owning ref reg in env's cur_state is assumed to use that
state's active_lock (Alexei)
* invalidate_non_owning_refs' reg loop should check that the reg being
inspected is a PTR_TO_BTF_ID before checking reg->non_owning_ref_lock,
since that field is part of a union and may be filled w/ meaningless bytes
if reg != PTR_TO_BTF_ID (Alexei)
* Patch 3 - "selftests/bpf: Update linked_list tests for non-owning ref semantics"
* Change the string searched for by the following tests:
* linked_list/incorrect_node_off1
* linked_list/double_push_front
* linked_list/double_push_back
necessary due to rebase / dropping of "release_regno searching logic" patch
(see "General" changes)
* Patch 8 - "bpf: Special verifier handling for bpf_rbtree_{remove, first}"
* Just call invalidate_non_owning_refs w/ env instead of env, lock. (see
Patch 2 changes)
* Patch 11 - "bpf, documentation: Add graph documentation for non-owning refs"
* Fix documentation formatting and improve content (David)
* v3's version of patch 11 was missing some changes, v4's patch 11 is still
addressing David's feedback from v2
v2 -> v3: lore.kernel.org/bpf/20221217082506.1570898-1-davemarchevsky@fb.com/
Patch #'s below refer to the patch's number in v2 unless otherwise stated.
* Patch 1 - "bpf: Support multiple arg regs w/ ref_obj_id for kfuncs"
* No longer needed as v3 doesn't have multiple ref_obj_id arg regs
* The refactoring pieces were submitted separately
(https://lore.kernel.org/bpf/20230121002417.1684602-1-davemarchevsky@fb.com/)
* Patch 2 - "bpf: Migrate release_on_unlock logic to non-owning ref semantics"
* Remove KF_RELEASE_NON_OWN flag from list API push methods, just match
against specific kfuncs for now (Alexei, David)
* Separate "release non owning reference" logic from KF_RELEASE logic
(Alexei, David)
* reg_find_field_offset now correctly tests 'rec' instead of 'reg' after
calling reg_btf_record (Dan Carpenter)
* New patch added after Patch 2 - "bpf: Improve bpf_reg_state space usage for non-owning ref lock"
* Eliminates extra bpf_reg_state memory usage by using a bool instead of
copying lock identity
* Patch 4 - "bpf: rename list_head -> graph_root in field info types"
* v2's version was applied to bpf-next, not including in respins
* Patch 6 - "bpf: Add bpf_rbtree_{add,remove,first} kfuncs"
* Remove KF_RELEASE_NON_OWN flag from rbtree_add, just add it to specific
kfunc matching (Alexei, David)
* Patch 9 - "bpf: Special verifier handling for bpf_rbtree_{remove, first}"
* Remove KF_INVALIDATE_NON_OWN kfunc flag, just match against specific kfunc
for now (Alexei, David)
* Patch 11 - "libbpf: Make BTF mandatory if program BTF has spin_lock or alloc_obj type"
* Drop for now, will submit separately
* Patch 12 - "selftests/bpf: Add rbtree selftests"
* Some expected-failure tests have different error messages due to "release
non-owning reference logic" being separated from KF_RELEASE logic in Patch
2 changes
* Patch 13 - "bpf, documentation: Add graph documentation for non-owning refs"
* Fix documentation formatting and improve content (David)
v1 -> v2: lore.kernel.org/bpf/20221206231000.3180914-1-davemarchevsky@fb.com/
Series-wide changes:
* Rename datastructure_{head,node,api} -> graph_{root,node,api} (Alexei)
* "graph datastructure" in patch summaries to refer to linked_list + rbtree
instead of "next-gen datastructure" (Alexei)
* Move from hacky marking of non-owning references as PTR_UNTRUSTED to
cleaner implementation (Alexei)
* Add invalidation of non-owning refs to rbtree_remove (Kumar, Alexei)
Patch #'s below refer to the patch's number in v1 unless otherwise stated.
Note that in v1 most of the meaty verifier changes were in the latter half
of the series. Here, about half of that complexity has been moved to
"bpf: Migrate release_on_unlock logic to non-owning ref semantics" - was Patch
3 in v1.
* Patch 1 - "bpf: Loosen alloc obj test in verifier's reg_btf_record"
* Was applied, dropped from further iterations
* Patch 2 - "bpf: map_check_btf should fail if btf_parse_fields fails"
* Dropped in favor of verifier check-on-use: when some normal verifier
checking expects the map to have btf_fields correctly parsed, it won't
find any and verification will fail
* New patch added before Patch 3 - "bpf: Support multiple arg regs w/ ref_obj_id for kfuncs"
* Addition of KF_RELEASE_NON_OWN flag, which requires KF_RELEASE, and tagging
of bpf_list_push_{front,back} KF_RELEASE | KF_RELEASE_NON_OWN, means that
list-in-list push_{front,back} will trigger "only one ref_obj_id arg reg"
logic. This is because "head" arg to those functions can be a list-in-list,
which itself can be an owning reference with ref_obj_id. So need to
support multiple ref_obj_id for release kfuncs.
* Patch 3 - "bpf: Minor refactor of ref_set_release_on_unlock"
* Now a major refactor w/ a rename to reflect this
* "bpf: Migrate release_on_unlock logic to non-owning ref semantics"
* Replaces release_on_unlock with active_lock logic as discussed in v1
* New patch added after Patch 3 - "selftests/bpf: Update linked_list tests for non_owning_ref logic"
* Removes "write after push" linked_list failure tests - no longer failure
scenarios.
* Patch 4 - "bpf: rename list_head -> datastructure_head in field info types"
* rename to graph_root instead. Similar renamings across the series - see
series-wide changes.
* Patch 5 - "bpf: Add basic bpf_rb_{root,node} support"
* OWNER_FIELD_MASK -> GRAPH_ROOT_MASK, OWNEE_FIELD_MASK -> GRAPH_NODE_MASK,
and change of "owner"/"ownee" in big btf_check_and_fixup_fields comment to
"root"/"node" (Alexei)
* Patch 6 - "bpf: Add bpf_rbtree_{add,remove,first} kfuncs"
* bpf_rbtree_remove can no longer return NULL. v2 continues v1's "use type
system to prevent remove of node that isn't in a datastructure" approach,
so rbtree_remove should never have been able to return NULL
* Patch 7 - "bpf: Add support for bpf_rb_root and bpf_rb_node in kfunc args"
* is_bpf_datastructure_api_kfunc -> is_bpf_graph_api_kfunc (Alexei)
* Patch 8 - "bpf: Add callback validation to kfunc verifier logic"
* Explicitly disallow rbtree_remove in rbtree callback
* Explicitly disallow bpf_spin_{lock,unlock} call in rbtree callback,
preventing possibility of "unbalanced" unlock (Alexei)
* Patch 10 - "bpf, x86: BPF_PROBE_MEM handling for insn->off < 0"
* Now that non-owning refs aren't marked PTR_UNTRUSTED it's not necessary to
include this patch as part of the series
* After conversation w/ Alexei, did another pass and submitted as an
independent series (lore.kernel.org/bpf/20221213182726.325137-1-davemarchevsky@fb.com/)
* Patch 13 - "selftests/bpf: Add rbtree selftests"
* Since bpf_rbtree_remove can no longer return null, remove null checks
* Remove test confirming that rbtree_first isn't allowed in callback. We want
this to be possible
* Add failure test confirming that rbtree_remove's new non-owning reference
invalidation behavior behaves as expected
* Add SEC("license") to rbtree_btf_fail__* progs. They were previously
failing due to lack of this section. Now they're failing for correct
reasons.
* rbtree_btf_fail__add_wrong_type.c - add locking around rbtree_add, rename
the bpf prog to something reasonable
* New patch added after patch 13 - "bpf, documentation: Add graph documentation for non-owning refs"
* Summarizes details of owning and non-owning refs which we hashed out in
v1
====================
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
It is difficult to intuit the semantics of owning and non-owning
references from verifier code. In order to keep the high-level details
from being lost in the mailing list, this patch adds documentation
explaining semantics and details.
The target audience of doc added in this patch is folks working on BPF
internals, as there's focus on "what should the verifier do here". Via
reorganization or copy-and-paste, much of the content can probably be
repurposed for BPF program writer audience as well.
Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230214004017.2534011-9-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This patch adds selftests exercising the logic changed/added in the
previous patches in the series. A variety of successful and unsuccessful
rbtree usages are validated:
Success:
* Add some nodes, let map_value bpf_rbtree_root destructor clean them
up
* Add some nodes, remove one using the non-owning ref leftover by
successful rbtree_add() call
* Add some nodes, remove one using the non-owning ref returned by
rbtree_first() call
Failure:
* BTF where bpf_rb_root owns bpf_list_node should fail to load
* BTF where node of type X is added to tree containing nodes of type Y
should fail to load
* No calling rbtree api functions in 'less' callback for rbtree_add
* No releasing lock in 'less' callback for rbtree_add
* No removing a node which hasn't been added to any tree
* No adding a node which has already been added to a tree
* No escaping of non-owning references past their lock's
critical section
* No escaping of non-owning references past other invalidation points
(rbtree_remove)
These tests mostly focus on rbtree-specific additions, but some of the
failure cases revalidate scenarios common to both linked_list and rbtree
which are covered in the former's tests. Better to be a bit redundant in
case linked_list and rbtree semantics deviate over time.
Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230214004017.2534011-8-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Newly-added bpf_rbtree_{remove,first} kfuncs have some special properties
that require handling in the verifier:
* both bpf_rbtree_remove and bpf_rbtree_first return the type containing
the bpf_rb_node field, with the offset set to that field's offset,
instead of a struct bpf_rb_node *
* mark_reg_graph_node helper added in previous patch generalizes
this logic, use it
* bpf_rbtree_remove's node input is a node that's been inserted
in the tree - a non-owning reference.
* bpf_rbtree_remove must invalidate non-owning references in order to
avoid aliasing issue. Use previously-added
invalidate_non_owning_refs helper to mark this function as a
non-owning ref invalidation point.
* Unlike other functions, which convert one of their input arg regs to
non-owning reference, bpf_rbtree_first takes no arguments and just
returns a non-owning reference (possibly null)
* For now verifier logic for this is special-cased instead of
adding new kfunc flag.
This patch, along with the previous one, complete special verifier
handling for all rbtree API functions added in this series.
With functional verifier handling of rbtree_remove, under current
non-owning reference scheme, a node type with both bpf_{list,rb}_node
fields could cause the verifier to accept programs which remove such
nodes from collections they haven't been added to.
In order to prevent this, this patch adds a check to btf_parse_fields
which rejects structs with both bpf_{list,rb}_node fields. This is a
temporary measure that can be removed after "collection identity"
followup. See comment added in btf_parse_fields. A linked_list BTF test
exercising the new check is added in this patch as well.
Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230214004017.2534011-6-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Some BPF helpers take a callback function which the helper calls. For
each helper that takes such a callback, there's a special call to
__check_func_call with a callback-state-setting callback that sets up
verifier bpf_func_state for the callback's frame.
kfuncs don't have any of this infrastructure yet, so let's add it in
this patch, following existing helper pattern as much as possible. To
validate functionality of this added plumbing, this patch adds
callback handling for the bpf_rbtree_add kfunc and hopes to lay
groundwork for future graph datastructure callbacks.
In the "general plumbing" category we have:
* check_kfunc_call doing callback verification right before clearing
CALLER_SAVED_REGS, exactly like check_helper_call
* recognition of func_ptr BTF types in kfunc args as
KF_ARG_PTR_TO_CALLBACK + propagation of subprogno for this arg type
In the "rbtree_add / graph datastructure-specific plumbing" category:
* Since bpf_rbtree_add must be called while the spin_lock associated
with the tree is held, don't complain when callback's func_state
doesn't unlock it by frame exit
* Mark rbtree_add callback's args with ref_set_non_owning
to prevent rbtree api functions from being called in the callback.
Semantically this makes sense, as less() takes no ownership of its
args when determining which comes first.
Signed-off-by: Dave Marchevsky <davemarchevsky@fb.com>
Link: https://lore.kernel.org/r/20230214004017.2534011-5-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
