summaryrefslogtreecommitdiff
AgeCommit message (Collapse)Author
2026-10-20ntfs: validate the MFT I/O geometryHyunchul Lee
MFT writes must consider both the device logical block size and the NTFS record geometry. Validate those values at mount time, then cache the MFT I/O unit used by all MFT write paths. Use the larger of the device logical block and MFT record sizes as the MFT I/O unit. Keep valid geometries that the current write paths cannot safely handle read-only until their mapping support is generalized. Fixes: 115380f9a2f9 ("ntfs: update mft operations") Cc: stable@vger.kernel.org Signed-off-by: Hyunchul Lee <hyc.lee@gmail.com> Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
2026-10-20ntfs: read MFT writeback bounds under size_lockHyunchul Lee
Read the MFT inode size and allocated size under the size_lock critical section before deriving the folio writeback limits. This prevents writeback from combining an old allocation boundary with a concurrently updated file size while the MFT is extended. Fixes: 115380f9a2f9 ("ntfs: update mft operations") Cc: stable@vger.kernel.org Signed-off-by: Hyunchul Lee <hyc.lee@gmail.com> Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
2026-10-20l2tp: snapshot sequencing state when building transmit headersChengfeng Ye
Updates to send_seq from netlink, PPP setsockopt and receive-side negotiation race with packet transmission. The header length is updated separately, while transmit reads both fields repeatedly without synchronization. For example, transmit can push a ten-byte L2TPv2 header, then a concurrent writer clears send_seq before the header builder reads it. The builder writes only six bytes, leaving four bytes of uninitialized skb headroom in the outgoing packet. The reverse mismatch overwrites the beginning of the payload. A later hdr_len read can also make the UDP length disagree with the skb length. KCSAN reported: BUG: KCSAN: data-race in l2tp_session_set_header_len / l2tp_xmit_skb write to 0xffff888100d0944a of 2 bytes by task 116 on cpu 1: l2tp_session_set_header_len+0x81/0xc0 l2tp_nl_cmd_session_modify+0x122/0x2c0 genl_family_rcv_msg_doit+0x170/0x1d0 read to 0xffff888100d0944a of 2 bytes by task 115 on cpu 0: l2tp_xmit_skb+0x9c/0xc80 pppol2tp_sendmsg+0x21c/0x2a0 __sys_sendto+0x28c/0x2a0 The bug can lead to leak kernel base address for KASLR bypass, the corresponding reproducer can be provided per request. Initialize hdr_len to the maximum header size and leave it unchanged when sequencing is toggled. Snapshot send_seq once per packet and use that snapshot for the actual header length and header construction, including the sequence flag. Use the same length for headroom, the header push and the UDP length. PPP callers can safely reserve the maximum size; L2TPv3 header sizes do not depend on sequencing. Make send_seq a separate bool so READ_ONCE() and WRITE_ONCE() can be used without sharing a bitfield storage unit with other configuration flags. Preserve the netlink attribute's existing low-bit truncation. Fixes: 3557baabf280 ("[L2TP]: PPP over L2TP driver core") Cc: stable@vger.kernel.org Assisted-by: GPT-6 Astra Signed-off-by: Chengfeng Ye <nicoyip.dev@gmail.com> Link: https://patch.msgid.link/20261013090500.1180376-1-nicoyip.dev@gmail.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20selftests: tc-testing: use $IP in hfsc test 8c39 setupIvan Vecera
Test 8c39 brings up lo with a bare "ip" command. nsPlugin runs it as "$IP netns exec <ns> ip ...", and as tdc runs commands with an empty environment, "ip netns exec" searches only the default execvp() path (/bin:/usr/bin). On systems that install ip only in /sbin, like Fedora, the setup fails: # ./tdc.py -e 8c39 Test 8c39: HFSC classify walk still reaches leaf after lateral drift -----> prepare stage *** Could not execute: "ip link set lo up" -----> prepare stage *** Error message: "exec of "ip" failed: No such file or directory Use $IP like all other tdc tests do. Fixes: 1e24c4f2ee44 ("selftests: tc-testing: add a lateral-drift hfsc classify-walk test") Signed-off-by: Ivan Vecera <ivecera@redhat.com> Reviewed-by: Victor Nogueira <victor@mojatatu.com> Reviewed-by: Hangbin Liu <liuhangbin@kylinos.cn> Link: https://patch.msgid.link/20261014121414.2373985-1-ivecera@redhat.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20Merge branch 'mptcp-misc-improvements-for-v7-4'Paolo Abeni
Matthieu Baerts says: ==================== mptcp: misc improvements for v7.4 Here are some unrelated improvements for MPTCP and its selftests: - Patches 1-2: avoid printing (pr_debug) uninitialised data. These are non-urgent fixes that can be backported to silence errors or if it helps with other backports. - Patches 3-4: convert IPTables to NFTables in MPTCP selftests. Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org> ==================== Link: https://patch.msgid.link/20261013-net-next-mptcp-misc-feat-7-4-part2-v1-0-1181e5d2fda4@kernel.org Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20selftests: mptcp: convert iptables to nftables for mptcp_join.shHangbin Liu
During conversion, we retain the same table and chain names used by the original iptables/ip6tables setup, so rule output is identical to the former iptables/ip6tables output. Add new function init_nftables so we only do nft table setup when test need set nft rules. Unlike iptables, nftables cannot match rules based on their full specification. Since endpoint_tests is the only test that delete rules, which inserts and deletes rules one by one. Flushing the table directly is safe and easier than using nft_handle. The cBPF bytecode matching MPTCP add‑addr is replaced with native nft matching using "tcp option mptcp subtype add-addr". But the RM_ADDR suboption is added after the MPTCP DSS option, and nft is unable to handle this because it stops parsing further TCP options once it matches the first one. We therefore use hard-coded offsets to match the RM_ADDR suboption. The config file adds CONFIG_NFT_NUMGEN (replaces iptables statistic nth), CONFIG_NFT_REJECT and CONFIG_NFT_REJECT_INET for reject‑related rules. Remove other unneeded configs. Remove the iptables/ip6tables check in mptcp_lib.sh since no script use it now. Signed-off-by: Hangbin Liu <liuhangbin@kylinos.cn> Reviewed-by: Matthieu Baerts (NGI0) <matttbe@kernel.org> Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org> Link: https://patch.msgid.link/20261013-net-next-mptcp-misc-feat-7-4-part2-v1-4-1181e5d2fda4@kernel.org Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20selftests: mptcp: convert iptables to nftables for mptcp_sockopt.shHangbin Liu
During the conversion, we retain the same filter and chain names previously used by iptables/ip6tables. Counters are not added to accept rules because the test does not inspect them. The syn mark rule is removed as it is only for debugging, which could be added easily when needed. After conversion, the generated output matches the original iptables/ip6tables behavior. Signed-off-by: Hangbin Liu <liuhangbin@kylinos.cn> Reviewed-by: Matthieu Baerts (NGI0) <matttbe@kernel.org> Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org> Link: https://patch.msgid.link/20261013-net-next-mptcp-misc-feat-7-4-part2-v1-3-1181e5d2fda4@kernel.org Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20mptcp: options: dss: avoid printing uninit csumMatthieu Baerts (NGI0)
When parsing the DSS suboption, this pr_debug() might read and print uninitialised checksum value. That's mostly harmless, but that's also confusing. Fixing that also avoids some tools to complain about accessing uninit data. So read this csum field only if it has been set. Note: this was initially reported by Clashiko as an existing issue. The code has been analysed by a human and confirmed. This is a non-urgent fix. Fixes: 390b95a5fb84 ("mptcp: receive checksum for DSS") Reviewed-by: Mat Martineau <martineau@kernel.org> Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org> Link: https://patch.msgid.link/20261013-net-next-mptcp-misc-feat-7-4-part2-v1-2-1181e5d2fda4@kernel.org Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20mptcp: options: mpc: avoid printing uninit dataMatthieu Baerts (NGI0)
When parsing the MP_CAPABLE suboption, this pr_debug() might read and print uninitialised values. That's mostly harmless, but that's also confusing. Fixing that also avoids some tools to complain about accessing uninit data. So read these fields only if they have been set. Note: this was initially reported by Clashiko as an existing issue. The code has been analysed by a human and confirmed. This is a non-urgent fix. Fixes: cc7972ea1932 ("mptcp: parse and emit MP_CAPABLE option according to v1 spec") Reviewed-by: Mat Martineau <martineau@kernel.org> Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org> Link: https://patch.msgid.link/20261013-net-next-mptcp-misc-feat-7-4-part2-v1-1-1181e5d2fda4@kernel.org Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20ALSA: compress: reject restarting an ACTIVE task via origin_seqnoPeter Ujfalusi
snd_compr_task_start_prepare() only rejects tasks with state >= SND_COMPRESS_TASK_STATE_FINISHED. Since the enum is IDLE(0) < ACTIVE(1) < FINISHED(2), this accepts ACTIVE tasks and rejects FINISHED ones, the opposite of what the origin_seqno mechanism needs: origin_seqno is meant to let a caller reuse a FINISHED task's buffers for a new job, not to restart a task that is still queued with the driver. The seqno lookup path in snd_compr_task_start() has its own explicit "task->state != IDLE" guard, but the origin_seqno lookup path relies solely on this function, so it can restart an ACTIVE task, causing ops->task_start() to be called twice for the same task and runtime->active_tasks to be incremented twice without a matching finish/stop. Accept IDLE and FINISHED, reject ACTIVE, on both entry paths. Fixes: 04177158cf98 ("ALSA: compress_offload: introduce accel operation mode") Cc: stable@vger.kernel.org Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com> Acked-by: Vinod Koul <vkoul@kernel.org> Link: https://patch.msgid.link/20261019144533.30665-5-peter.ujfalusi@linux.intel.com Signed-off-by: Takashi Iwai <tiwai@suse.de>
2026-10-20ALSA: compress: add missing OPEN-state gate to SET_METADATAPeter Ujfalusi
snd_compr_set_metadata() calls stream->ops->set_metadata() straight after copy_from_user(), without ever checking runtime->state, even though the function's own comment says a parameter change should only be allowed once the stream has left the OPEN state. Every other metadata/param entry point in this file (SET_PARAMS, WRITE, ...) enforces that gate. A codec driver's set_metadata callback can assume SET_PARAMS has already run and initialised its private state; calling it while the stream is still OPEN can operate on that uninitialised state. Add the OPEN-state check the existing comment already documents. Fixes: 9727b490e543 ("ALSA: compress: add support for gapless playback") Cc: stable@vger.kernel.org Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com> Acked-by: Vinod Koul <vkoul@kernel.org> Link: https://patch.msgid.link/20261019144533.30665-4-peter.ujfalusi@linux.intel.com Signed-off-by: Takashi Iwai <tiwai@suse.de>
2026-10-20ALSA: compress: reject buffer geometry change on gapless next_trackPeter Ujfalusi
snd_compr_set_params() is invoked from the gapless next_track re-entry path, which calls snd_compr_allocate_buffer() again and lets buffer_size change while total_bytes_available and total_bytes_transferred still hold values computed for the previous buffer. snd_compr_calc_avail() then derives an avail/count value from those stale counters against the new buffer_size, and snd_compr_write_data() copy_from_user()s that many bytes into the new, potentially smaller buffer, overflowing it with user-controlled length and content. The buffer is set up once, with the initial SET_PARAMS, and neither the core nor the DSP expects it to change for a gapless track transition. Do not call snd_compr_allocate_buffer() on the next_track path and reject a request that changes the buffer geometry, so that the codec parameters that the driver receives match the buffer in use. Fixes: 9727b490e543 ("ALSA: compress: add support for gapless playback") Cc: stable@vger.kernel.org Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com> Link: https://patch.msgid.link/20261019144533.30665-3-peter.ujfalusi@linux.intel.com Signed-off-by: Takashi Iwai <tiwai@suse.de>
2026-10-20ALSA: compress: fix buffer leak on failed SET_PARAMSPeter Ujfalusi
When the driver has neither ops->copy nor a preallocated dma_buffer_p, snd_compr_allocate_buffer() kmallocs runtime->buffer. If ops->set_params() then rejects the parameters, SET_PARAMS fails, the stream stays in the OPEN state and keeps the buffer. The next SET_PARAMS allocates a new buffer and overwrites the pointer, so repeated SET_PARAMS calls with parameters that the driver rejects leak one buffer per call. Release the framework-owned buffer when ops->set_params() fails on an OPEN stream, so that a failed SET_PARAMS leaves no buffer behind and the buffer is allocated only once per successful setup. Use the same ownership check as snd_compr_free(). Fixes: b21c60a4edd2 ("ALSA: core: add support for compress_offload") Cc: stable@vger.kernel.org Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com> Link: https://patch.msgid.link/20261019144533.30665-2-peter.ujfalusi@linux.intel.com Signed-off-by: Takashi Iwai <tiwai@suse.de>
2026-10-20ALSA: hda/realtek: Enable jack detection on Minisforum AI X1Nika Sommer
The firmware of the Minisforum AI X1 leaves the headphone jack detector reset bit asserted on its ALC245 codec. As a result, pin sense on NID 0x21 always reports the jack as absent. Enable the existing fixup ALC245_FIXUP_MINISFORUM_JACK_DETECT from commit 6bd8a57c044f ("ALSA: hda/realtek: Enable jack detection on Minisforum AI X1 Pro") to the Minisforum AI X1, which requires the same workaround. Tested on a Minisforum AI X1 (AMD Ryzen 7 255), using NixOS 26.11.20261014.55bfb4d and kernel 7.2.10. Signed-off-by: Nika Sommer <nika.sommer@mailbox.org> Link: https://patch.msgid.link/20261019183226.64985-1-nika.sommer@mailbox.org Signed-off-by: Takashi Iwai <tiwai@suse.de>
2026-10-20tun: use pskb_may_pull_reason() and remove SKB_DROP_REASON_HDR_TRUNCAnton Danilov
tun_get_user() is the only user of SKB_DROP_REASON_HDR_TRUNC. It sets it when pskb_may_pull() fails to pull the Ethernet header of a frame written to a TAP device. pskb_may_pull_reason() reports why the pull failed: PKT_TOO_SMALL if the packet is shorter than requested, NOMEM if pulling the data into the linear area fails. Use it for both header pulls in tun_get_user(), on TAP and on TUN devices, and remove the now unused SKB_DROP_REASON_HDR_TRUNC. The pull of the IP version byte on a TUN device with IFF_NO_PI was reported as NOT_SPECIFIED and now reports PKT_TOO_SMALL, for example when only the virtio-net header is written. After a successful pull drop_reason is SKB_NOT_DROPPED_YET. The later error paths without a reason of their own, the unknown IP version on TUN, the virtio-net header conversion and the path that finds the queue detached, relied on the SKB_DROP_REASON_NOT_SPECIFIED initializer. Fall back to SKB_DROP_REASON_NOT_SPECIFIED at the free_skb label, as vxlan_rcv() does, so that these paths, and any later one without a reason of its own, keep reporting it. The errors returned to user space do not change. Suggested-by: Ido Schimmel <idosch@nvidia.com> Link: https://lore.kernel.org/netdev/20260923145931.GA2859441@shredder/ Assisted-by: LLM Signed-off-by: Anton Danilov <littlesmilingcloud@gmail.com> Reviewed-by: Willem de Bruijn <willemb@google.com> Reviewed-by: Ido Schimmel <idosch@nvidia.com> Link: https://patch.msgid.link/20261014054615.1656074-1-littlesmilingcloud@gmail.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20ALSA: usb-audio: Skip sample-rate read-back on PreSonus AudioBox USBThomas Nemer
The AudioBox USB stalls the UAC1 GET_CUR SAMPLING_FREQ request on both streaming endpoints. After the sample rate is set, the read-back verification in set_sample_rate_v1() fails and logs usb 1-3.1.2: 2:1: cannot get freq at ep 0x1 usb 1-3.1.2: 3:1: cannot get freq at ep 0x82 on each endpoint configuration until the sample_rate_read_error counter gives up. Set QUIRK_FLAG_GET_SAMPLE_RATE to skip the read-back on this device. Fixes: 1869c47e5d59 ("ALSA: usb-audio: Fix missing sample-rate control for PreSonus AudioBox USB") Cc: stable@vger.kernel.org Signed-off-by: Thomas Nemer <thomas.nemer@fortytwo.fr> Link: https://patch.msgid.link/20261019204944.28302-1-thomas.nemer@fortytwo.fr Signed-off-by: Takashi Iwai <tiwai@suse.de>
2026-10-20net: gro: don't chain linear parts larger than gso_size into frag_listSorah Fukumori
skb_gro_receive() chains an skb into the frag_list of the held packet when its payload cannot be attached as page frags: when its head is not a page fragment and still holds payload, or when its head and frags would overflow the frags[] array of the last packet. Since commit 5eddb24901ee ("gro: add support of (hw)gro packets to gro stack"), that skb may itself be a GSO packet whose linear part holds several gso_size segments. skb_segment() cannot split the linear part of a frag_list member across segments, so segmenting the aggregate in software, e.g. when forwarding it to a device without NETIF_F_FRAGLIST, hits BUG_ON(skb_headlen(list_skb) > len); Fully linear GSO packets come from generic XDP. When the program does not support frags, netif_skb_check_for_xdp() linearizes any nonlinear skb with pskb_expand_head() and skb_linearize() and keeps its GSO metadata. GRO on the device runs before generic XDP, so each aggregate passed by the program carries all of its payload in a kmalloc'ed head. A tunnel that decapsulates these packets and runs GRO again through gro_cells then builds the bad frag_list. This was first seen on a router that forwards TCP received over VXLAN, with a non-frags XDP program attached in SKB mode to the underlying bond because one slave lacks native XDP: kernel BUG at net/core/skbuff.c:4917! RIP: 0010:skb_segment+0xbb6/0xf40 Call Trace: tcp_gso_segment ipv6_gso_segment skb_mac_gso_segment __skb_gso_segment validate_xmit_skb __dev_queue_xmit ip6_output ip6_sublist_rcv_finish ipv6_list_rcv netif_receive_skb_list_internal gro_complete dev_gro_receive gro_receive_skb gro_cell_poll __napi_poll net_rx_action It reproduces with three network namespaces: TCP sent over VXLAN on an IPv6 veth link to a router namespace that has an XDP_PASS program without frags support attached to the underlay veth in generic mode and GRO enabled on its VXLAN device, and forwards to a veth with tx-scatter-gather-fraglist disabled. A program with frags support in generic mode gives similar packets. skb_pp_cow_data() rebuilds the skb with up to about a page of linear data in a page_pool head_frag and keeps the GSO metadata, so once frags[] is full the next such packet is chained through frag_list with more than gso_size bytes in its head as well. Refuse the merge when the linear part to be chained exceeds the held packet's gso_size. tcp_gro_receive() then flushes the held packet and the skb starts a new GRO packet as its head, whose linear length skb_segment() handles. Fixes: 5eddb24901ee ("gro: add support of (hw)gro packets to gro stack") Cc: stable@vger.kernel.org Assisted-by: claude-opus-5-5 Signed-off-by: Sorah Fukumori <her@sorah.jp> Tested-by: Chase Hiltz <me@chasehil.tz> Reviewed-by: Eric Dumazet <edumazet@kernel.org> Reviewed-by: Jason Xing <kerneljasonxing@gmail.com> Link: https://patch.msgid.link/20261013130250.3643916-1-her@sorah.jp Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20net: stmmac: xgmac: do not treat TBU as a TX eventPengfei Zhang
Commit 8d07a7930434 ("net: stmmac: xgmac: Do not enable TBU interrupt") removed TBUE from the default DMA interrupt mask, but did not remove the TBU check from dwxgmac2_dma_interrupt(). TBU is set whenever a TX channel runs out of descriptors. Since TBUE is not enabled and only enabled bits are cleared, the TBU status bit is never cleared. As a result, TBU stays set on every idle TX channel, regardless of whether there is any TX traffic. With a shared interrupt, stmmac_dma_interrupt() checks all channels, so every interrupt from any source schedules TX NAPI on every idle TX channel. Only use TI to trigger TX processing. Fixes: 8d07a7930434 ("net: stmmac: xgmac: Do not enable TBU interrupt") Signed-off-by: Pengfei Zhang <zhangfeionline@gmail.com> Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com> Link: https://patch.msgid.link/20261013024747.2688984-1-zhangfeionline@gmail.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20ALSA: hda/hdmi: Enable HBR capabilityAaron Kling
Tegra HDA does not set the HBR capability bit, however every SoC since Tegra114 does support HBR. Override the hbr setup function for these SoCs to unconditionally enable the feature. Since Tegra30 does not support HBR, this override will not be applied. Signed-off-by: Aaron Kling <webgeek1234@gmail.com> Link: https://patch.msgid.link/20261020-tegra-hda-hbr-v2-1-6968017f585b@gmail.com Signed-off-by: Takashi Iwai <tiwai@suse.de>
2026-10-20Merge branches 'apple/dart', 'arm/smmu/updates', 'arm/smmu/bindings', ↵Joerg Roedel
'hyper-v', 'samsung/exynos', 'riscv', 'broadcom', 'intel/vt-d', 'amd/amd-vi' and 'core' into next
2026-10-20MAINTAINERS: Make Vasant Hegde AMD IOMMU maintainerJoerg Roedel
Vasant reviews and tests most of the AMD IOMMU patches and is critical for maintaining the driver. Make him an official maintainer. Cc: Vasant Hegde <vasant.hegde@amd.com> Cc: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com> Cc: Will Deacon <will@kernel.org> Acked-by: Vasant Hegde <vasant.hegde@amd.com> Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
2026-10-20RDMA/mlx5: Validate priority in STEERING_ANCHOR_CREATE handlerSerhat Kumral
In MLX5_IB_METHOD_STEERING_ANCHOR_CREATE, the user-supplied u16 priority is passed directly to _get_flow_table() without an upper bound check. This can lead to out-of-bounds array accesses on dev->flow_db priority tables. Check that priority does not exceed MLX5_IB_FLOW_LAST_PRIO before calling _get_flow_table(), matching the existing validation in raw_fs_rule_add() and mlx5_ib_create_flow(). Fixes: 0c6ab0ca9a66 ("RDMA/mlx5: Expose steering anchor to userspace") Signed-off-by: Serhat Kumral <serhatkumral1@gmail.com> Link: https://patch.msgid.link/20261018144907.20123-1-serhatkumral1@gmail.com Signed-off-by: Leon Romanovsky <leon@kernel.org>
2026-10-20Merge branch 'net-macb-implement-context-swapping'Paolo Abeni
says: ==================== net: macb: implement context swapping MACB has a pretty primitive approach to buffer management. They are all stored in `struct macb *bp`. On operations that require buffer realloc (set_ringparam & change_mtu at the moment), the only option is to close the interface, change our global state and re-open the interface. Two issues: - It doesn't fly on memory pressured systems; we free our precious buffers and don't manage to reallocate fully, meaning our machine just lost its network access. - Anecdotally, it is pretty slow because it implies a full PHY reinit. Instead, we shall: - allocate a new context (including buffers) first - if it fails, early return without any impact to the interface - stop interface - update global state (bp, netdev, etc) - pass newly allocated buffer pointers to the hardware - start interface - free old context This is what we implement here. Both .set_ringparam() and .ndo_change_mtu() are covered by this series. In the future, at least .set_channels() [0], XDP [1] and XSK [2] would benefit. Tested-by: Nicolai Buchwitz <nb@tipi-net.de> # RPi CM5 with RP1 GEM --- I wonder if we can get those changes in before the next merge window? The idea being that we can restart iterating on XDP [1] & XSK [2] during the next kernel cycle, bringing those long awaited features into MACB. [0]: https://lore.kernel.org/netdev/20260317-macb-set-channels-v4-0-1bd4f4ffcfca@bootlin.com/ [1]: https://lore.kernel.org/netdev/20260323221047.2749577-1-pvalerio@redhat.com/ [2]: https://lore.kernel.org/netdev/20260304-macb-xsk-v1-0-ba2ebe2bdaa3@bootlin.com/ [3]: https://lore.kernel.org/all/DJXGIM9EGPT8.4UHNYP2Y0GOP@bootlin.com/ [4]: https://lore.kernel.org/netdev/20260930-macb-irq-v1-1-8994a4c8f3bb@bootlin.com/ [5]: https://lore.kernel.org/netdev/20260918-macb-ncr-rmw-v1-1-30c3494f93cd@bootlin.com/ - v10: https://patch.msgid.link/20261001-macb-context-v10-0-beb541bbb7df@bootlin.com - v9: https://patch.msgid.link/20260812-macb-context-v9-0-7ddbf5f715e0@bootlin.com - v8: https://patch.msgid.link/20260805-macb-context-v8-0-bc302ffd1174@bootlin.com - v7: https://patch.msgid.link/20260803-macb-context-v7-0-4d7d4af04849@bootlin.com - v6: https://patch.msgid.link/20260731-macb-context-v6-0-49d5a1439d48@bootlin.com - v5: https://patch.msgid.link/20260724-macb-context-v5-0-569b1852bc7f@bootlin.com - v4: https://patch.msgid.link/20260717-macb-context-v4-0-0acbe7f10cdb@bootlin.com - v3: https://patch.msgid.link/20260701-macb-context-v3-0-00268d5b1502@bootlin.com - v2: https://patch.msgid.link/20260410-macb-context-v2-0-af39f71d40b6@bootlin.com - v1: https://patch.msgid.link/20260401-macb-context-v1-0-9590c5ab7272@bootlin.com ==================== Link: https://patch.msgid.link/20261013-macb-context-v11-0-b27785167566@bootlin.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20net: macb: use context swapping in .ndo_change_mtu()Théo Lebrun
Use newly introduced context buffer management to implement .ndo_change_mtu() as a context swap: allocate new context -> reconfigure HW -> free old context. This resists memory pressure well by failing without closing the interface and it is much faster by avoiding PHY reinit. AT91 EMAC is handled differently as their buffer management is separate and they don't do NAPI. They cannot land in macb_change_mtu as they use a different ndo struct. We still defensively protect against it in code. Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de> Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com> Link: https://patch.msgid.link/20261013-macb-context-v11-9-b27785167566@bootlin.com Tested-by: Nicolai Buchwitz <nb@tipi-net.de> # RPi CM5 with RP1 GEM Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20net: macb: use context swapping in .set_ringparam()Théo Lebrun
ethtool_ops.set_ringparam() is implemented using the primitive close / update ring size / reopen sequence. Under memory pressure this does not fly: we free our buffers at close and cannot reallocate new ones at open. Also, it triggers a slow PHY reinit. Instead, exploit the new context mechanism and improve our sequence to: - allocate a new context (including buffers) first - if it fails, early return without any impact to the interface - stop interface - update global state (bp, netdev, etc) - pass buffer pointers to the hardware - start interface - free old context. The HW disable sequence is inspired by macb_reset_hw() but avoids (1) setting NCR bit CLRSTAT and (2) clearing register PBUFRXCUT. The HW re-enable sequence is inspired by macb_mac_link_up(), skipping over register writes which would be redundant (because values have not changed). The generic context swapping parts are isolated into helper functions macb_context_swap_start|end(), reusable by other operations (change_mtu, set_channels, etc). Introduce a new locking primitive (mac_cfg_lock mutex) to serialise swap with phylink MAC callbacks. Avoid stopping phylink to avoid a slow PHY retrain. We cannot sync to phylink ops using phydev->lock because it is not available in the SFP case. We cannot check link state using netif_carrier_ok() because we could race with its changes; so we use a redundant bp->link_up boolean that is mac_cfg_lock protected. AT91 EMAC is handled differently as their buffer management is separate and they don't do NAPI. They must never call swap_start/end(). Anyway they do not implement set_ringparam (-EOPNOTSUPP) so we are safe. Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de> Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com> Link: https://patch.msgid.link/20261013-macb-context-v11-8-b27785167566@bootlin.com Tested-by: Nicolai Buchwitz <nb@tipi-net.de> # RPi CM5 with RP1 GEM Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20net: macb: introduce macb_context_alloc() helperThéo Lebrun
Move the context allocation sequence from inline macb_open() to its own helper function called macb_context_alloc(). All ops doing context swapping (set_ringparam, change_mtu, etc) will use this helper. Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de> Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com> Link: https://patch.msgid.link/20261013-macb-context-v11-7-b27785167566@bootlin.com Tested-by: Nicolai Buchwitz <nb@tipi-net.de> # RPi CM5 with RP1 GEM Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20net: macb: change function signatures to take contextsThéo Lebrun
For parallel MACB context to start become a reality, many functions need to stop operating on bp->ctx (the currently active context) and instead work on a context they get passed. That context might be (1) the new one that is getting allocated and initialised, or, (2) the old one to be freed. To reduce bug surface area, taint those functions to *only* take a context `struct macb_context *ctx` and no `struct macb *bp`. That way, no bug of using `bp->ctx` instead of `ctx` will ever occur. We also convert functions that take a `struct macb_queue *queue` to instead take `struct macb_context *ctx, unsigned int q`, with q indexing ctx->txq[] and ctx->rxq[]. Full list: macb_adj_dma_desc_idx() macb_tx_ring_wrap() macb_tx_desc() macb_rx_ring_wrap() macb_rx_desc() macb_get_addr() gem_rx_refill() macb_init_rx_ring() gem_free_rx_buffers() macb_free_rx_buffers() macb_tx_ring_size_per_queue() macb_rx_ring_size_per_queue() macb_free() gem_alloc_rx_buffers() macb_alloc_rx_buffers() macb_alloc() gem_init_rx_ring() gem_init_rings() macb_init_rings() Note about gem_rx_refill(): it ends with a netdev_vdbg() that prints the queue pointer. Change to print the queue index because we do not have access to the queue anymore. Note: functions that operate on the interface during its normal lifetime do not need to be converted and have no risk of using the wrong context. We do not allow concurrent swaps. When interface is alive, there can only be one context alive. Acked-by: Conor Dooley <conor.dooley@microchip.com> Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de> Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com> Link: https://patch.msgid.link/20261013-macb-context-v11-6-b27785167566@bootlin.com Tested-by: Nicolai Buchwitz <nb@tipi-net.de> # RPi CM5 with RP1 GEM Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20net: macb: change caps helpers signaturesThéo Lebrun
For parallel MACB context to start becoming a reality, many functions will soon not have access to `struct macb *bp`. Those will still have access to caps through ctx->info->caps. Change all caps helpers signatures, from taking `struct macb *bp` to taking `struct macb_info *info`. Info is accessible in both bp and ctx and it ensures type safety (versus passing raw `u32 caps`). Function list: macb_is_gem() gem_has_ptp() macb_dma64() macb_dma_ptp() macb_dma_desc_get_size() macb_set_addr() macb_get_addr() macb_64b_desc() macb_tx_unmap() Note: drop macb_64b_desc(bp, ...) parameter; it is unused and it must be dropped as macb_{set,get}_addr() call macb_64b_desc(). Acked-by: Conor Dooley <conor.dooley@microchip.com> Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de> Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com> Link: https://patch.msgid.link/20261013-macb-context-v11-5-b27785167566@bootlin.com Tested-by: Nicolai Buchwitz <nb@tipi-net.de> # RPi CM5 with RP1 GEM Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20net: macb: make `struct macb` subset reachable from macb_context structThéo Lebrun
For parallel MACB context to start become a reality, many functions need to stop operating on bp->ctx (the currently active context) and instead work on a context they get passed. That context might be (1) the new one that is getting allocated and initialised, or, (2) the old one to be freed. To reduce bug surface area, we will taint those functions to *only* take a context and no `struct macb *bp`. That way, no bug of using `bp->ctx` instead of `ctx` will ever occur. For that, we need to embed a subset of `struct macb` information into each context so that all helpers can still do their jobs. That subset must be constant once probe is completed. Do this by taking a pointer to a subset of macb called `struct macb_info`. That subset is accessible from context (ctx->info->caps) or from bp (bp->caps) using `-fms-extensions` option, thanks to commit c4781dc3d1cf ("Kbuild: enable -fms-extensions"). https://gcc.gnu.org/onlinedocs/gcc/Unnamed-Fields.html Add the structure and assign ctx->info at alloc, but nothing uses it yet. Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de> Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com> Link: https://patch.msgid.link/20261013-macb-context-v11-4-b27785167566@bootlin.com Tested-by: Nicolai Buchwitz <nb@tipi-net.de> # RPi CM5 with RP1 GEM Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20net: macb: avoid macb_init_rx_buffer_size() modifying stateThéo Lebrun
macb_init_rx_buffer_size() takes the macb private data struct and overrides its bp->ctx->rx_buffer_size. To make it usable with multiple contexts, make it return its value. Also, move the `bufsz` computation into it. The value is only used if GEM, and for historical reason it currently lives in macb_open(). Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de> Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com> Link: https://patch.msgid.link/20261013-macb-context-v11-3-b27785167566@bootlin.com Tested-by: Nicolai Buchwitz <nb@tipi-net.de> # RPi CM5 with RP1 GEM Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20net: macb: introduce macb_context struct for buffer managementThéo Lebrun
Whenever an operation requires buffer realloc, we close the interface, update parameters and reopen. To improve reliability under memory pressure, we should rather alloc new buffers, reconfigure HW and free old buffers. This requires MACB to support having multiple "contexts" in parallel. Introduce this concept by adding the macb_context struct, which owns all queue buffers and the parameters associated. We do not yet support multiple contexts in parallel, because all functions access bp->ctx (the currently active context) directly. Steps: - Introduce `struct macb_context` and its children `struct macb_rxq` and `struct macb_txq`. Context fields are stolen from `struct macb` and rxq/txq fields are from `struct macb_queue`. Making it two separate structs per queue simplifies accesses: we grab a txq/rxq local variable and access fields like txq->head instead of queue->tx_head. It also anecdotally improves data locality. - macb_init_dflt() / macb_get_ringparam() do not access bp->ctx->{rx,tx}_ring_size as they will/might run while interface is offline and ctx is NULL. Instead, introduce bp->configured_{rx,tx}_ring_size which get updated on user requests. - macb_open() starts by allocating bp->ctx. It gets freed in the open error codepath or by macb_close(). - Guided by compile errors, update all codepaths. Most diff is changing `queue->tx_*` to `txq->*` and `queue->rx_*` to `rxq->*`, with a new local variable. Also rx_buffer_size / rx_ring_size / tx_ring_size move from bp to bp->ctx. Introduce two helpers macb_txq|rxq() functions to convert macb_queue pointers. - macb_get_regs() is tweaked to support being ran while interface is offline (and context is NULL). Use default values at zero and override them only if context is present. - Protect our ctx freeing in at91ether code with a synchronize_irq() call. Note: bp->configured_{rx,tx}_ring_size contain proper values for EMAC, meaning 9/1 on ethtool -G on RM9200 whereas it reported 0/0 before. Acked-by: Conor Dooley <conor.dooley@microchip.com> Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de> Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com> Link: https://patch.msgid.link/20261013-macb-context-v11-2-b27785167566@bootlin.com Tested-by: Nicolai Buchwitz <nb@tipi-net.de> # RPi CM5 with RP1 GEM Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20net: macb: close race between NAPI polls & ndo_stopThéo Lebrun
In ndo_close, we call napi_disable() then free the ring buffers queue->rx_ring|tx_ring. In the NAPI poll functions, after napi_complete_done() we might enable interrupts and might re-check the next ring buffer slot. This is a use-after-free race on our ring buffers because napi_disable() only waits for napi_complete_done() to return, it does not wait for NAPI poll functions to finish. The poll function sequence must stay as-is, see commit 0bf476fc3624 ("net: macb: Fix lost RX packet wakeup race in NAPI receive") which explains why the sequence must be napi_complete_done() -> arm IRQs -> re-check ring slot. Fix by waiting for full end of NAPI poll functions using synchronize_net() in our ndo_close implementation. This bug has never been reproduced and was reported by Sashiko. The chances of facing it in practice sound tiny: it needs a system without any phylink attached, running PREEMPT_RT and with a kernel preemption on the NAPI poll at the right instant. We fix it here before introducing context swapping, which is another place where the race could appear (and Sashiko complains on this series). Fixes: 138badbc21a0 ("net: macb: use NAPI for TX completion path") Fixes: 1900e30d0ef7 ("net: macb: simplify/cleanup NAPI reschedule checking") Cc: stable+noautosel@kernel.org # too unlikely and never reported by users Reported-by: Sashiko <sashiko-bot@kernel.org> Link: https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261001-macb-context-v10-0-beb541bbb7df%40bootlin.com?part=7 Link: https://lore.kernel.org/all/179115680572.434549.1016291990685937691@kernel.org/ Signed-off-by: Théo Lebrun <theo.lebrun@bootlin.com> Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de> Link: https://patch.msgid.link/20261013-macb-context-v11-1-b27785167566@bootlin.com Tested-by: Nicolai Buchwitz <nb@tipi-net.de> # RPi CM5 with RP1 GEM Signed-off-by: Paolo Abeni <pabeni@redhat.com>
2026-10-20RDMA/mlx5: Fix CQ resize destination index maskLi RongQing
In copy_resize_cqes, the destination CQE index is computed as (i + 1) & cq->resize_buf->nent, but resize_buf->nent is produced by roundup_pow_of_two(), so it is a power of two with exactly one bit set. Masking with a single-bit value can only yield 0 or nent itself. The buffer has nent entries, so valid indices are 0 .. nent - 1 and nent is one past the last slot; whenever the masked result is nent the following memcpy writes one entry out of bounds. Use (nent - 1) as the wrap mask instead. For a power-of-two nent, (i + 1) & (nent - 1) is the correct modulo-nent reduction, keeping the index within [0, nent - 1]. This matches get_sw_cqe(), which addresses the CQ buffer via n & cq->ibcq.cqe, where cq->ibcq.cqe == entries - 1 is exactly the nent - 1 wrap mask. Fixes: bde51583f49b ("IB/mlx5: Add support for resize CQ") Signed-off-by: Li RongQing <lirongqing@baidu.com> Link: https://patch.msgid.link/20261019050503.2313-1-lirongqing@baidu.com Signed-off-by: Leon Romanovsky <leon@kernel.org>
2026-10-19Merge tag 'vfs-7.4-rc1.sync_close' of ↵Linus Torvalds
git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs Pull file closing updates from Christian Brauner: "This makes closing files synchronous for close_range(), exec and exit. Since fput() was switched to task work every final __fput() of an exiting task is queued on task work and run from exit_task_work(). exec and close_range() do the same. For a task with a large descriptor table that is a cmpxchg() and a TIF_NOTIFY_RESUME test-and-set per file plus a list entry for any later task_work_cancel() to search under pi_lock, and then a second pass over every struct file that isn't cache hot anymore by then. Oleg and Neil have pointed this out over the years. close_files(), close_range() and close-on-exec now flush and put each file inline the way close(2) does. Kernel threads aren't affected. They share init_files which init_task pins forever so their exit_files() never drops the last reference, and their fput() keeps punting to a workqueue. The one place that could put a table from a kthread context, a failing copy_process(), can't hold the last reference while the source table is alive. On a 64 vCPU KVM guest tearing down 4096, 65536 and 1M descriptors scattered at random positions in the table: 4096 fds 65536 fds 1M fds exit 1045 -> 895 us 18.9 -> 14.9 ms 429 -> 306 ms execve, close-on-exec 1393 -> 1190 us 21.4 -> 16.9 ms 471 -> 325 ms close_range() 1179 -> 988 us 21.9 -> 17.4 ms 484 -> 337 ms So that is 16-21% or 40-150 ns per closed file, largest at 1M descriptors. With eight processes tearing down 65536 descriptors each at once it stays at 16-21% because the second pass over 16 MB of struct file is what gets expensive when CPUs compete for the cache. With 32 or 64 processes both kernels are bound by pushing millions of files through slab and rcu and the difference shrinks to 4-8%. Files whose own release dominates gain less, 13% for eventfds, 11% for pipes, 6% for AF_UNIX sockets. close(2) is untouched and will-it-scale is within 2%. The spufs coredump support is removed. It was the only reason do_coredump() unshared the descriptor table. gdb dropped Cell debugging in 9.1 and binutils in 2.34, the Cell blades are gone and nothing on the PlayStation 3 produces or reads these notes. With that unshare_files() goes away, exec installs its private table through the new switch_files_struct() and the coredump startup completion becomes a plain thread count waited on with the new wait_var_event_state(). Three fixes for the synchronous close are included. A failed fork released its files under scx_fork_rwsem and deadlocked against the bpf link's release. A session leader's exit released its tty before disassociate_ctty() ran and the foreground job lost its SIGHUP, so the tty is hung up before the files are closed now. And descriptor tables are walked from the highest descriptor down which restores the order the deferred puts had and keeps a few long-standing release ordering bugs from deadlocking exit" * tag 'vfs-7.4-rc1.sync_close' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs: fs: close files from the highest descriptor down coredump: factor out coredump_wait_inactive() exit: hang up the tty before closing the files fork: release the files of a failed fork after sched_cancel_fork() coredump: replace the startup completion with a thread count sched: add wait_var_event_state() coredump: drop core_state->dumper fs: make close_cloexec_files() synchronous fs: rename do_close_on_exec() to close_cloexec_files() fs: make close_range() synchronous fs: make close_files() synchronous fs: add filp_close_sync() fs: remove unshare_files() fs: move unshare_fd() to fs/file.c fs: add switch_files_struct() coredump: stop unsharing the file descriptor table fs: don't open-code file_close_fd() in close_fd() powerpc: remove coredump support
2026-10-19Merge tag 'vfs-7.4-rc1.shared.super' of ↵Linus Torvalds
git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs Pull superblock updates from Christian Brauner: "This renames vfs_get_super() to get_tree_super() and exports it so a filesystem that needs its own superblock matching policy can reuse it instead of open-coding sget_fc() plus fill_super(). erofs uses it to share a superblock between file-backed mounts" * tag 'vfs-7.4-rc1.shared.super' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs: fs: rename vfs_get_super() to get_tree_super() and export it
2026-10-19Merge tag 'vfs-7.4-rc1.shared.net-next.pidfd' of ↵Linus Torvalds
git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs Pull pidfd updates from Christian Brauner: "This adds thread specific pidfds to af_unix sockets. SO_PASSPIDFD and SO_PEERPIDFD hand out a pidfd for the thread-group leader. A coredump server on the coredump socket can't get a handle on the thread that took the signal and is writing the dump, and anything doing per-thread authentication has the same problem. The sending and the connecting thread are recorded in addition to the thread-group leader and sk_peer_pid becomes an array indexed by pid type. SO_PASSPIDFD_THREAD works like SO_PASSPIDFD but delivers an SCM_PIDFD_THREAD cmsg for the sending thread. Both options are independent so a receiver can ask for both. On SOCK_STREAM sockets recvmsg() only stops merging data at a thread boundary when the receiver asked for a thread pidfd. SO_PEERPIDFD_THREAD returns a pidfd for the thread that called connect(), listen() or socketpair(). pidfs records the coredump on the dumping thread's struct pid too so its pidfd reports it like the leader's does. Selftests for af_unix and the coredump socket are included" * tag 'vfs-7.4-rc1.shared.net-next.pidfd' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs: selftests/coredump: check the dumping thread's pidfd pidfs: record the coredump on the dumping thread's pid too selftests/net: SO_PEERPIDFD_THREAD net: add SO_PEERPIDFD_THREAD to get a thread-specific pidfd af_unix: record the pid of the connecting thread net: turn sk_peer_pid into an array indexed by pid type selftests/net: SO_PASSPIDFD_THREAD net: add SO_PASSPIDFD_THREAD to get a thread-specific SCM_PIDFD af_unix: record the pid of the sending thread pid: add helpers to operate on a struct pid array
2026-10-19Merge tag 'vfs-7.4-rc1.netfs' of ↵Linus Torvalds
git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs Pull netfs updates from Christian Brauner: "This contains the netfs and cachefiles changes for this cycle from David. Cachefiles learns to cache the most common class of files properly, files that are created or truncated, written once sequentially and closed. So far it relied on the backing filesystem's idea of what is stored and asked with SEEK_DATA and SEEK_HOLE before every read it prepared, two seeks per read, and the backing filesystem is free to insert or remove runs of zeros which produces false positives and false negatives. The object type in the state xattr is now actually set to CACHEFILES_CONTENT_{SINGLE,ALL,BACKFS_MAP} instead of zero. For files written contiguously the type is ALL, the backing file's i_size tracks how much is cached and the object size lives in the xattr, and reads skip the seeks entirely. Netfslib queries the cache occupancy through a new ->query_occupancy() method that returns up to two extents at once, ->prepare_read() only limits the size of a cache read anymore and a new ->collect_write() lets the cache update its metadata once a contiguous run of writes completes. The state xattr is preset when the tmpfile is created so ENOSPC from setxattr can't scrap a fully written cache file. Two fixes for this followed, one for resize handling where the cache's idea of the size can be stale and one for an unset error that hung a request on a failed cache write. Preparatory work converts file positions that can't be negative to uoff_t, removes the writethrough code whose locking looked like it could deadlock with ceph snapshots, makes the deprecated PG_private_2 support opt-in so mm can get its page bit back, adds helpers for the NETFS_RREQ_ALL_QUEUED barriers and adds the cache object id to the read and write tracepoints. Plus an unused header removal and two documentation fixes" * tag 'vfs-7.4-rc1.netfs' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs: fscache, cachefiles: Fix resize handling cachefiles: Fix unset error when calling netfs_prepare_write_failed() netfs: remove unused fscache_internal.h header cachefiles: Preset the state xattr when creating a new file cachefiles: Don't rely on backing fs storage map for most use cases netfs: Document the remote size member and its accessors cachefiles: Add a tracepoint to log insufficient space errors netfs, cachefiles: Add a couple of traces for write failure cachefiles: Clean up cachefiles_do_prepare_read() netfs: Set subrequest->source at alloc before trace emission netfs: Add some functions to wrap the all-queued handling netfs: Make deprecated PG_private_2 support opt-in netfs: Add the cache object ID to netfs_read/write tracepoints netfs: trace: Rejig a couple of the tracepoints cachefiles: Fix path of the "debug" module parameter in help paragraph netfs: trace: Change the "clear" folio traces to "endwb" netfs: Remove the writethrough code netfs: Use uoff_t instead of unsigned long long and loff_t
2026-10-19Merge tag 'vfs-7.4-rc1.mount' of ↵Linus Torvalds
git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs Pull mount updates from Christian Brauner: "This adds OPEN_TREE_DROP_MNTNS_MOUNTS. crun started using OPEN_TREE_NAMESPACE to set up the container's mount namespace and open_tree(OPEN_TREE_CLONE) for the sources of its bind mounts. That fails on any host with a mount namespace pinned below such a source. The recursive clone picks the pinned namespace up and move_mount() refuses to attach the tree with ELOOP. Only a younger mount namespace may ever be bound into an older one or mount namespace references could form a loop. A namespace created with OPEN_TREE_NAMESPACE is younger than everything else on the system so from inside it every pinned namespace is older. By the time this fails setns() has run and the host tree is out of reach, so the runtime has to predict the situation beforehand and lose the optimization. snapd pins mount namespaces under /run/snapd/ns so on Ubuntu that is any container bind mounting the host root. Every other path crossing a mount namespace boundary already leaves these mounts behind, copy_mnt_ns() and create_new_namespace() both do. Only get_detached_copy() asks for them. The new flag lets a caller get a clone without them. It stays opt-in because open_tree(OPEN_TREE_CLONE) plus move_mount() is mount --rbind through a file descriptor and dropping them by default would silently lose /run/snapd/ns/*.mnt on a live system. It requires AT_RECURSIVE, is a no-op with OPEN_TREE_NAMESPACE and skips locked nsfs mounts the way copy_mnt_ns() does, so nothing new is exposed. Selftests are included" * tag 'vfs-7.4-rc1.mount' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs: selftests/filesystems: test OPEN_TREE_DROP_MNTNS_MOUNTS mount: add OPEN_TREE_DROP_MNTNS_MOUNTS
2026-10-20Merge branch 'xfrm: state: exact mark/mask match for control-plane SA lookups'Steffen Klassert
Antony Antony says: ==================== While looking into a XFRM_MSG_MIGRATE_STATE issue reported by Sashiko, we found the underlying problem generalizes: xfrm allows multiple SAs to coexist for the same (SPI, daddr, proto) differing only in mark, and every netlink method that resolves "which SA" - xfrm get_sa(), del_sa(), update, get_ae, new_ae, expire, migrate - uses the same wildcard mark match the data path needs. A broader-mask SA can silently shadow a more specific one: # ip xfrm state add ... spi 0x1000 mark 1 mask 1 (SA_target) # ip xfrm state add ... spi 0x1000 mark 0 mask 0 (SA_decoy, catch-all, added after -> bucket head) # ip xfrm state delete dst ... proto esp spi 0x1000 mark 1 mask 1 -> deletes SA_decoy; SA_target survives, untouched xfrm policy had the same bug, fixed in commit 4f47e8ab6ab7 ("xfrm: policy: match with both mark and mask on user interfaces"). Netlink state lookups using spi use an exact mark/mask match except for UPDSA; the wildcard match stays for the data path and state_add only. This series applies that fix across every affected method, not just XFRM_MSG_MIGRATE_STATE. Also reject marks with value bits outside the mask (state ADD, ALLOCSPI, policy ADD, MIGRATE_STATE new mark). These are misconfigurations anyway, since the extra bits can never match, and break exact match added here. This series does not touch PF_KEY, which no longer receives non-critical fixes. state_lookup_byaddr is out of scope. This is only fixing spi based lookups. ==================== Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
2026-10-19Merge tag 'vfs-7.4-rc1.misc' of ↵Linus Torvalds
git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs Pull misc vfs updates from Christian Brauner: "Fixes: - aio: a keyed poll completion on an epoll instance could recurse into the same epoll waitqueue lock when the result eventfd is watched by that very instance. An unprivileged process could make a CPU spin on the recursive lock. eventfd-backed poll completions are always deferred now. - fsnotify_inoderemove() is only called once the inode is really dead, no nlink and no aliases, and open_by_handle_at() can't add an alias to a dead inode anymore. The race could crash fsnotify or lose inode mark notifications. - devtmpfs: the conversion of the public instance to ->get_tree() broke reconfiguring the shared superblock's options on later mounts which made the boot time defaults permanent. A reconfigure context is kept in fc->fs_private so the parse ops can forward options to the underlying filesystem. - ufs truncated the 64-bit directory size to an unsigned int in ufs_last_byte(). A crafted image with a 4 GiB directory produced an out-of-folio read in ufs_find_entry() and, with CONFIG_UFS_FS_WRITE, a write into an adjacent page. - devpts rejects max=0 instead of silently lifting the per-instance pty limit. The value wrapped into ida_alloc_max() and became unlimited. - fs_pin publishes pins with the rcu list helpers so the kill paths reading the lists under rcu see the initialization. - dentry_kill() reports a Tasks-RCU quiescent state. Its cond_resched() is the only reschedule point on the dentry teardown path and cond_resched() isn't a quiescent state for Tasks-RCU, so Meta kept seeing rcu_tasks stalls in btrfs subvolume deletion. - writeback sizes foreign flushes by the target wb's dirty pages instead of the memcg-wide count which over or undersized a single bdi flush. - nfsd READ emitted two fsnotify access events on the iterator path since vfs_iocb_iter_read() already sends one. A READ on a krb5i or krb5p mount showed up twice in inotify. - pipe and splice pass a poll key when waking readers and writers so epoll, poll() and select() waiters only wake for the event they asked for. - the iso8859-14 Unicode to charset tables had page01 duplicated as page1e so 29 code points mapped to the wrong bytes and 22 that aren't in the charset were accepted. - orangefs doesn't wait for an unmount service operation that never returns on system shutdown and uses ssize_t for strscpy() errors. - __register_chrdev() checks the kobject_set_name() return value. - selftest fixes: the openat2 tests hardcoded O_LARGEFILE to 0x8000 which is O_NOFOLLOW on powerpc, resolve_test makes /tmp a mountpoint so RESOLVE_NO_XDEV has a boundary to trip over, file_stressor moves into its own directory with a 30 minute timeout, the kernfs listxattr test skips when SELinux exposes security.selinux, the namespaces active reference tests wait for the thread via a PIDFD_THREAD pidfd since pthread_join() returns before exit_nsproxy_namespaces(), the idmapped_tmpfile test chowns its directory into the mapped range, the clone3 CLONE_CLEAR_SIGHAND test adds CLONE_VM so it fails for the intended reason, the pidfd PPID error message prints the values it compares and fscontext_ns gets a .gitignore. Cleanups: - vmcore prints the release of the crashed kernel from the VMCOREINFO note so a kdump that dies before the dump lands still says which kernel it was for. - initramfs allocates hardlink hash entries with the validated cpio name length instead of PATH_MAX, which put every entry in kmalloc-8k. - the unused dax_layout_busy_page() is removed, dax_layout_busy_page_range() becomes static and dax_folio_init() drops a redundant list initialization. - vfs_empty_path() is removed, it has had no callers since getname_maybe_null(). - comment and documentation fixes in aio, bpf_fs_kfuncs, fs-writeback, inode, read_write, squashfs and the vfs.rst copy of struct file_system_type" * tag 'vfs-7.4-rc1.misc' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs: (38 commits) selftests/filesystems: Fix idmapped_tmpfile layer dir ownership selftests/namespaces: Wait for thread exit via pidfd in active-ref tests writeback: size foreign flushes by target wb dirty pages initramfs: Reduce hardlink hash allocation sizes fs: remove unused vfs_empty_path() dcache: report a Tasks-RCU quiescent state in dentry_kill() vmcore: report the release of the crashed kernel splice: pass a poll key when waking pipe readers and writers pipe: pass a poll key when waking writers after a resize orangefs: use ssize_t instead of size_t for error codes... orangefs_super: don't wait on service operation on system shutdown. docs/vfs: Fix the struct file_system_type copy nls: fix the Unicode to charset tables of iso8859-14 selftests/filesystems/openat2: make /tmp a mountpoint inside private namespace selftests/filesystems/openat2: fix O_LARGEFILE definitions for non-arm64 architectures fs: fix misleading comments in read_write.c squashfs: fix inode number description in documentation fs: fix typos in comments fs/dax: Make dax_layout_busy_page_range() static fs/dax: Remove unused dax_layout_busy_page() ...
2026-10-19ubi: use put_user() when filling the erase counter arrayLinus Torvalds
ubi_get_ec_info() checks the user's erase_counters array with access_ok() and then fills it one entry at a time with __put_user(), taking the wear-leveling lock for each entry in between. Each of those accesses enables and disables user access on its own, so the up-front check saves nothing. Use put_user(). The access_ok() stays, so a bad array still fails before anything is written. This is part of getting rid of __get_user() and __put_user() entirely. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2026-10-19PNP: isapnp: use put_user() in the proc read loopLinus Torvalds
isapnp_proc_bus_read() checks the user buffer with access_ok() and then copies the configuration registers one byte at a time with __put_user(), reading each register in between. Each of those accesses enables and disables user access on its own, so the up-front check saves nothing. Use put_user(). The ignored return value and the access_ok() stay as they were. This is part of getting rid of __get_user() and __put_user() entirely. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2026-10-19PCI/proc: Use get_user()/put_user() for config space reads and writesLinus Torvalds
proc_bus_pci_read() and proc_bus_pci_write() check the user buffer with access_ok() and then copy config space one byte, word or dword at a time with __put_user() and __get_user(). Each of those accesses enables and disables user access on its own, so the up-front check saves nothing. Use put_user() and get_user(). This deliberately changes nothing else: the return values were ignored before and still are, and the access_ok() stays, since it is the only error these functions report for a bad buffer. This is part of getting rid of __get_user() and __put_user() entirely. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2026-10-19drivers/char/mem: use get_user()/put_user() for /dev/port byte copiesLinus Torvalds
read_port() and write_port() check the whole user buffer with access_ok() and then move it one byte at a time with __put_user() and __get_user(), doing an inb() or outb() for each. Each of those accesses enables and disables user access on its own, so the up-front check saves nothing. Use put_user() and get_user(). The access_ok() stays, because it also makes a partly invalid buffer fail up front, rather than after some ports have already been read or written. This is part of getting rid of __get_user() and __put_user() entirely. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2026-10-19x86/xen: Fix a stale __get_user() reference in a commentLinus Torvalds
The comment in set_aliased_prot() says the target address is faulted in with __get_user(), but the code has used copy_from_kernel_nofault() for a long time. Make the comment say so. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2026-10-19x86/fpu: Use get_user()/put_user() for the signal frame fieldsLinus Torvalds
The FPU signal frame code reads and writes a handful of individual fields (the xstate magic words, the legacy status word, the xfeatures header and PKRU) with __get_user() and __put_user(). These rely on an access_ok() of the whole frame done earlier by the caller. Each of those accesses enables and disables user access on its own, so the up-front check saves nothing. On x86, __get_user() is even more expensive than get_user(): it does an lfence, where get_user() checks the address with a couple of plain instructions. Use get_user() and put_user(). The accesses that run with page faults disabled stay that way; get_user() and put_user() are fine there. This is part of getting rid of __get_user() and __put_user() entirely. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2026-10-19x86/ptrace: Use get_user() in the general register set writersLinus Torvalds
genregs_set() and genregs32_set() read the user buffer one word at a time with __get_user(), calling putreg() for each. The access_ok() that makes this safe is in the generic copy_regset_from_user(), on the other side of an indirect call through the regset. Each of those accesses enables and disables user access on its own, so the up-front check saves nothing. On x86, __get_user() is even more expensive than get_user(): it does an lfence, where get_user() checks the address with a couple of plain instructions. Use get_user(), so each read checks its own address. This is part of getting rid of __get_user() and __put_user() entirely. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2026-10-19perf/x86: Use scoped user access for the user callchain framesLinus Torvalds
The user callchain walkers read each two-word stack frame with __get_user(), after checking the frame with valid_user_frame(), which is just __access_ok(). They run with page faults disabled, since they can be called from NMI context. Read each frame inside a scoped_user_read_access() instead, which does the range check itself and needs only a single stac/clac pair for both loads. That leaves valid_user_frame() unused, so remove it. This is part of getting rid of __get_user() and __put_user() entirely. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2026-10-19x86/stacktrace: Use scoped user access in copy_stack_frame()Linus Torvalds
copy_stack_frame() reads the two words of a user stack frame with __get_user(), after an __access_ok() on the frame and inside pagefault_disable(), since the user stack walk can run in atomic context. Do the two reads in a scoped_user_read_access() instead. It nests inside the pagefault_disable() section, does the range check itself, and needs only a single stac/clac pair for both loads. This is part of getting rid of __get_user() and __put_user() entirely. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2026-10-19x86/signal: Use scoped user access in the sigreturn system callsLinus Torvalds
All four sigreturn variants (32-bit sigreturn and rt_sigreturn, native 64-bit and x32 rt_sigreturn) do an access_ok() on the whole signal frame only to read the signal mask with __get_user(). The 64-bit and x32 variants also read uc_flags that way. Everything else in the frame is read by helpers that do their own checking. Read those fields inside a scoped_user_read_access() instead, which does the range check once and puts a single stac/clac pair around plain loads. The scoped accessor shadows 'frame' with a checked copy only inside its block, so the badframe path still reports the original pointer. This is part of getting rid of __get_user() and __put_user() entirely. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>