kernel/linux.git/drivers/infiniband/sw, branch v6.18.22

RDMA/rxe: Fix race condition in QP timer handlers

2026-02-26T22:59:26+00:00

[ Upstream commit 87bf646921430e303176edc4eb07c30160361b73 ] I encontered the following warning: WARNING: drivers/infiniband/sw/rxe/rxe_task.c:249 at rxe_sched_task+0x1c8/0x238 [rdma_rxe], CPU#0: swapper/0/0 ... libsha1 [last unloaded: ip6_udp_tunnel] CPU: 0 UID: 0 PID: 0 Comm: swapper/0 Tainted: G C 6.19.0-rc5-64k-v8+ #37 PREEMPT Tainted: [C]=CRAP Hardware name: Raspberry Pi 4 Model B Rev 1.2 Call trace: rxe_sched_task+0x1c8/0x238 [rdma_rxe] (P) retransmit_timer+0x130/0x188 [rdma_rxe] call_timer_fn+0x68/0x4d0 __run_timers+0x630/0x888 ... WARNING: drivers/infiniband/sw/rxe/rxe_task.c:38 at rxe_sched_task+0x1c0/0x238 [rdma_rxe], CPU#0: swapper/0/0 ... WARNING: drivers/infiniband/sw/rxe/rxe_task.c:111 at do_work+0x488/0x5c8 [rdma_rxe], CPU#3: kworker/u17:4/93400 ... refcount_t: underflow; use-after-free. WARNING: lib/refcount.c:28 at refcount_warn_saturate+0x138/0x1a0, CPU#3: kworker/u17:4/93400 The issue is caused by a race condition between retransmit_timer() and rxe_destroy_qp, leading to the Queue Pair's (QP) reference count dropping to zero during timer handler execution. It seems this warning is harmless because rxe_qp_do_cleanup() will flush all pending timers and requests. Example of flow causing the issue: CPU0 CPU1 retransmit_timer() { spin_lock_irqsave rxe_destroy_qp() __rxe_cleanup() __rxe_put() // qp->ref_count decrease to 0 rxe_qp_do_cleanup() { if (qp->valid) { rxe_sched_task() { WARN_ON(rxe_read(task->qp) <= 0); } } spin_unlock_irqrestore } spin_lock_irqsave qp->valid = 0 spin_unlock_irqrestore } Ensure the QP's reference count is maintained and its validity is checked within the timer callbacks by adding calls to rxe_get(qp) and corresponding rxe_put(qp) after use. Signed-off-by: Li Zhijian Fixes: d94671632572 ("RDMA/rxe: Rewrite rxe_task.c") Link: https://patch.msgid.link/20260120074437.623018-1-lizhijian@fujitsu.com Reviewed-by: Zhu Yanjun Signed-off-by: Leon Romanovsky Signed-off-by: Sasha Levin

RDMA/rxe: Fix iova-to-va conversion for MR page sizes != PAGE_SIZE

2026-02-26T22:59:26+00:00

[ Upstream commit 12985e5915a0b8354796efadaaeb201eed115377 ] The current implementation incorrectly handles memory regions (MRs) with page sizes different from the system PAGE_SIZE. The core issue is that rxe_set_page() is called with mr->page_size step increments, but the page_list stores individual struct page pointers, each representing PAGE_SIZE of memory. ib_sg_to_page() has ensured that when i>=1 either a) SG[i-1].dma_end and SG[i].dma_addr are contiguous or b) SG[i-1].dma_end and SG[i].dma_addr are mr->page_size aligned. This leads to incorrect iova-to-va conversion in scenarios: 1) page_size < PAGE_SIZE (e.g., MR: 4K, system: 64K): ibmr->iova = 0x181800 sg[0]: dma_addr=0x181800, len=0x800 sg[1]: dma_addr=0x173000, len=0x1000 Access iova = 0x181800 + 0x810 = 0x182010 Expected VA: 0x173010 (second SG, offset 0x10) Before fix: - index = (0x182010 >> 12) - (0x181800 >> 12) = 1 - page_offset = 0x182010 & 0xFFF = 0x10 - xarray[1] stores system page base 0x170000 - Resulting VA: 0x170000 + 0x10 = 0x170010 (wrong) 2) page_size > PAGE_SIZE (e.g., MR: 64K, system: 4K): ibmr->iova = 0x18f800 sg[0]: dma_addr=0x18f800, len=0x800 sg[1]: dma_addr=0x170000, len=0x1000 Access iova = 0x18f800 + 0x810 = 0x190010 Expected VA: 0x170010 (second SG, offset 0x10) Before fix: - index = (0x190010 >> 16) - (0x18f800 >> 16) = 1 - page_offset = 0x190010 & 0xFFFF = 0x10 - xarray[1] stores system page for dma_addr 0x170000 - Resulting VA: system page of 0x170000 + 0x10 = 0x170010 (wrong) Yi Zhang reported a kernel panic[1] years ago related to this defect. Solution: 1. Replace xarray with pre-allocated rxe_mr_page array for sequential indexing (all MR page indices are contiguous) 2. Each rxe_mr_page stores both struct page* and offset within the system page 3. Handle MR page_size != PAGE_SIZE relationships: - page_size > PAGE_SIZE: Split MR pages into multiple system pages - page_size <= PAGE_SIZE: Store offset within system page 4. Add boundary checks and compatibility validation This ensures correct iova-to-va conversion regardless of MR page size and system PAGE_SIZE relationship, while improving performance through array-based sequential access. Tests on 4K and 64K PAGE_SIZE hosts: - rdma-core/pytests $ ./build/bin/run_tests.py --dev eth0_rxe - blktest: $ TIMEOUT=30 QUICK_RUN=1 USE_RXE=1 NVMET_TRTYPES=rdma ./check nvme srp rnbd [1] https://lore.kernel.org/all/CAHj4cs9XRqE25jyVw9rj9YugffLn5+f=1znaBEnu1usLOciD+g@mail.gmail.com/T/ Fixes: 592627ccbdff ("RDMA/rxe: Replace rxe_map and rxe_phys_buf by xarray") Signed-off-by: Li Zhijian Link: https://patch.msgid.link/20260116032753.2574363-1-lizhijian@fujitsu.com Signed-off-by: Leon Romanovsky Signed-off-by: Sasha Levin

RDMA/rxe: Fix double free in rxe_srq_from_init

2026-02-26T22:59:25+00:00

[ Upstream commit 0beefd0e15d962f497aad750b2d5e9c3570b66d1 ] In rxe_srq_from_init(), the queue pointer 'q' is assigned to 'srq->rq.queue' before copying the SRQ number to user space. If copy_to_user() fails, the function calls rxe_queue_cleanup() to free the queue, but leaves the now-invalid pointer in 'srq->rq.queue'. The caller of rxe_srq_from_init() (rxe_create_srq) eventually calls rxe_srq_cleanup() upon receiving the error, which triggers a second rxe_queue_cleanup() on the same memory, leading to a double free. The call trace looks like this: kmem_cache_free+0x.../0x... rxe_queue_cleanup+0x1a/0x30 [rdma_rxe] rxe_srq_cleanup+0x42/0x60 [rdma_rxe] rxe_elem_release+0x31/0x70 [rdma_rxe] rxe_create_srq+0x12b/0x1a0 [rdma_rxe] ib_create_srq_user+0x9a/0x150 [ib_core] Fix this by moving 'srq->rq.queue = q' after copy_to_user. Fixes: aae0484e15f0 ("IB/rxe: avoid srq memory leak") Signed-off-by: Jiasheng Jiang Link: https://patch.msgid.link/20260112015412.29458-1-jiashengjiangcool@gmail.com Reviewed-by: Zhu Yanjun Signed-off-by: Leon Romanovsky Signed-off-by: Sasha Levin

RDMA/siw: Fix potential NULL pointer dereference in header processing

2026-02-26T22:58:56+00:00

commit 14ab3da122bd18920ad57428f6cf4fade8385142 upstream. If siw_get_hdr() returns -EINVAL before set_rx_fpdu_context(), qp->rx_fpdu can be NULL. The error path in siw_tcp_rx_data() dereferences qp->rx_fpdu->more_ddp_segs without checking, which may lead to a NULL pointer deref. Only check more_ddp_segs when rx_fpdu is present. KASAN splat: [ 101.384271] KASAN: null-ptr-deref in range [0x00000000000000c0-0x00000000000000c7] [ 101.385869] RIP: 0010:siw_tcp_rx_data+0x13ad/0x1e50 Fixes: 8b6a361b8c48 ("rdma/siw: receive path") Signed-off-by: YunJe Shin Link: https://patch.msgid.link/20260204092546.489842-1-ioerts@kookmin.ac.kr Acked-by: Bernard Metzler Signed-off-by: Leon Romanovsky Signed-off-by: Greg Kroah-Hartman

IB/rxe: Fix missing umem_odp->umem_mutex unlock on error path

2026-01-08T09:16:51+00:00

[ Upstream commit 3c68cf68233e556e0102f45b69f7448908dc1f44 ] rxe_odp_map_range_and_lock() must release umem_odp->umem_mutex when an error occurs, including cases where rxe_check_pagefault() fails. Fixes: 2fae67ab63db ("RDMA/rxe: Add support for Send/Recv/Write/Read with ODP") Signed-off-by: Li Zhijian Link: https://patch.msgid.link/20251226094112.3042583-1-lizhijian@fujitsu.com Reviewed-by: Zhu Yanjun Signed-off-by: Leon Romanovsky Signed-off-by: Sasha Levin

RDMA/rxe: Fix null deref on srq->rq.queue after resize failure

2025-12-18T13:02:41+00:00

[ Upstream commit 503a5e4690ae14c18570141bc0dcf7501a8419b0 ] A NULL pointer dereference can occur in rxe_srq_chk_attr() when ibv_modify_srq() is invoked twice in succession under certain error conditions. The first call may fail in rxe_queue_resize(), which leads rxe_srq_from_attr() to set srq->rq.queue = NULL. The second call then triggers a crash (null deref) when accessing srq->rq.queue->buf->index_mask. Call Trace: rxe_modify_srq+0x170/0x480 [rdma_rxe] ? __pfx_rxe_modify_srq+0x10/0x10 [rdma_rxe] ? uverbs_try_lock_object+0x4f/0xa0 [ib_uverbs] ? rdma_lookup_get_uobject+0x1f0/0x380 [ib_uverbs] ib_uverbs_modify_srq+0x204/0x290 [ib_uverbs] ? __pfx_ib_uverbs_modify_srq+0x10/0x10 [ib_uverbs] ? tryinc_node_nr_active+0xe6/0x150 ? uverbs_fill_udata+0xed/0x4f0 [ib_uverbs] ib_uverbs_handler_UVERBS_METHOD_INVOKE_WRITE+0x2c0/0x470 [ib_uverbs] ? __pfx_ib_uverbs_handler_UVERBS_METHOD_INVOKE_WRITE+0x10/0x10 [ib_uverbs] ? uverbs_fill_udata+0xed/0x4f0 [ib_uverbs] ib_uverbs_run_method+0x55a/0x6e0 [ib_uverbs] ? __pfx_ib_uverbs_handler_UVERBS_METHOD_INVOKE_WRITE+0x10/0x10 [ib_uverbs] ib_uverbs_cmd_verbs+0x54d/0x800 [ib_uverbs] ? __pfx_ib_uverbs_cmd_verbs+0x10/0x10 [ib_uverbs] ? __pfx___raw_spin_lock_irqsave+0x10/0x10 ? __pfx_do_vfs_ioctl+0x10/0x10 ? ioctl_has_perm.constprop.0.isra.0+0x2c7/0x4c0 ? __pfx_ioctl_has_perm.constprop.0.isra.0+0x10/0x10 ib_uverbs_ioctl+0x13e/0x220 [ib_uverbs] ? __pfx_ib_uverbs_ioctl+0x10/0x10 [ib_uverbs] __x64_sys_ioctl+0x138/0x1c0 do_syscall_64+0x82/0x250 ? fdget_pos+0x58/0x4c0 ? ksys_write+0xf3/0x1c0 ? __pfx_ksys_write+0x10/0x10 ? do_syscall_64+0xc8/0x250 ? __pfx_vm_mmap_pgoff+0x10/0x10 ? fget+0x173/0x230 ? fput+0x2a/0x80 ? ksys_mmap_pgoff+0x224/0x4c0 ? do_syscall_64+0xc8/0x250 ? do_user_addr_fault+0x37b/0xfe0 ? clear_bhb_loop+0x50/0xa0 ? clear_bhb_loop+0x50/0xa0 ? clear_bhb_loop+0x50/0xa0 entry_SYSCALL_64_after_hwframe+0x76/0x7e Fixes: 8700e3e7c485 ("Soft RoCE driver") Tested-by: Liu Yi Signed-off-by: Zhu Yanjun Link: https://patch.msgid.link/20251027215203.1321-1-yanjun.zhu@linux.dev Signed-off-by: Leon Romanovsky Signed-off-by: Sasha Levin

Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma

2025-10-04T01:35:22+00:00

Pull rdma updates from Jason Gunthorpe: "A new Pensando ionic driver, a new Gen 3 HW support for Intel irdma, and lots of small bnxt_re improvements. - Small bug fixes and improves to hfi1, efa, mlx5, erdma, rdmarvt, siw - Allow userspace access to IB service records through the rdmacm - Optimize dma mapping for erdma - Fix shutdown of the GSI QP in mana - Support relaxed ordering MR and fix a corruption bug with mlx5 DMA Data Direct - Many improvement to bnxt_re: - Debugging features and counters - Improve performance of some commands - Change flow_label reporting in completions - Mirror vnic - RDMA flow support - New RDMA driver for Pensando Ethernet devices: ionic - Gen 3 hardware support for the Intel irdma driver - Fix rdma routing resolution with VRFs" * tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma: (85 commits) RDMA/ionic: Fix memory leak of admin q_wr RDMA/siw: Always report immediate post SQ errors RDMA/bnxt_re: improve clarity in ALLOC_PAGE handler RDMA/irdma: Remove unused struct irdma_cq fields RDMA/irdma: Fix positive vs negative error codes in irdma_post_send() RDMA/bnxt_re: Remove non-statistics counters from hw_counters RDMA/bnxt_re: Add debugfs info entry for device and resource information RDMA/bnxt_re: Fix incorrect errno used in function comments RDMA: Use %pe format specifier for error pointers RDMA/ionic: Use ether_addr_copy instead of memcpy RDMA/ionic: Fix build failure on SPARC due to xchg() operand size RDMA/rxe: Fix race in do_task() when draining IB/sa: Fix sa_local_svc_timeout_ms read race IB/ipoib: Ignore L3 master device RDMA/core: Use route entry flag to decide on loopback traffic RDMA/core: Resolve MAC of next-hop device without ARP support RDMA/core: Squash a single user static function RDMA/irdma: Update Kconfig RDMA/irdma: Extend CQE Error and Flush Handling for GEN3 Devices RDMA/irdma: Add Atomic Operations support ...

RDMA/siw: Always report immediate post SQ errors

2025-09-26T16:12:19+00:00

In siw_post_send(), any immediate error encountered during processing of the work request list must be reported to the caller, even if previous work requests in that list were just accepted and added to the send queue. Not reporting those errors confuses the caller, which would wait indefinitely for the failing and potentially subsequently aborted work requests completion. This fixes a case where immediate errors were overwritten by subsequent code in siw_post_send(). Fixes: 303ae1cdfdf7 ("rdma/siw: application interface") Link: https://patch.msgid.link/r/20250923144536.103825-1-bernard.metzler@linux.dev Suggested-by: Stefan Metzmacher Signed-off-by: Bernard Metzler Signed-off-by: Jason Gunthorpe

RDMA/rxe: Fix race in do_task() when draining

2025-09-21T11:22:28+00:00

When do_task() exhausts its iteration budget (!ret), it sets the state to TASK_STATE_IDLE to reschedule, without a secondary check on the current task->state. This can overwrite the TASK_STATE_DRAINING state set by a concurrent call to rxe_cleanup_task() or rxe_disable_task(). While state changes are protected by a spinlock, both rxe_cleanup_task() and rxe_disable_task() release the lock while waiting for the task to finish draining in the while(!is_done(task)) loop. The race occurs if do_task() hits its iteration limit and acquires the lock in this window. The cleanup logic may then proceed while the task incorrectly reschedules itself, leading to a potential use-after-free. This bug was introduced during the migration from tasklets to workqueues, where the special handling for the draining case was lost. Fix this by restoring the original pre-migration behavior. If the state is TASK_STATE_DRAINING when iterations are exhausted, set cont to 1 to force a new loop iteration. This allows the task to finish its work, so that a subsequent iteration can reach the switch statement and correctly transition the state to TASK_STATE_DRAINED, stopping the task as intended. Fixes: 9b4b7c1f9f54 ("RDMA/rxe: Add workqueue support for rxe tasks") Reviewed-by: Zhu Yanjun Signed-off-by: Gui-Dong Han Link: https://patch.msgid.link/20250919025212.1682087-1-hanguidong02@gmail.com Signed-off-by: Leon Romanovsky

RDMA/rdmavt: Use int type to store negative error codes

2025-09-11T06:18:35+00:00

Change 'ret' from u32 to int in alloc_qpn() to store -EINVAL, and remove the 'bail' label as it simply returns 'ret'. Storing negative error codes in an u32 causes no runtime issues, but it's ugly as pants, Change 'ret' from u32 to int type - this change has no runtime impact. Signed-off-by: Qianfeng Rong Link: https://patch.msgid.link/20250826150556.541440-1-rongqianfeng@vivo.com Signed-off-by: Leon Romanovsky