"value":"In the Linux kernel, the following vulnerability has been resolved:\n\ntcp: prevent concurrent execution of tcp_sk_exit_batch\n\nIts possible that two threads call tcp_sk_exit_batch() concurrently,\nonce from the cleanup_net workqueue, once from a task that failed to clone\na new netns. In the latter case, error unwinding calls the exit handlers\nin reverse order for the 'failed' netns.\n\ntcp_sk_exit_batch() calls tcp_twsk_purge().\nProblem is that since commit b099ce2602d8 (\"net: Batch inet_twsk_purge\"),\nthis function picks up twsk in any dying netns, not just the one passed\nin via exit_batch list.\n\nThis means that the error unwind of setup_net() can \"steal\" and destroy\ntimewait sockets belonging to the exiting netns.\n\nThis allows the netns exit worker to proceed to call\n\nWARN_ON_ONCE(!refcount_dec_and_test(&net->ipv4.tcp_death_row.tw_refcount));\n\nwithout the expected 1 -> 0 transition, which then splats.\n\nAt same time, error unwind path that is also running inet_twsk_purge()\nwill splat as well:\n\nWARNING: .. at lib/refcount.c:31 refcount_warn_saturate+0x1ed/0x210\n...\n refcount_dec include/linux/refcount.h:351 [inline]\n inet_twsk_kill+0x758/0x9c0 net/ipv4/inet_timewait_sock.c:70\n inet_twsk_deschedule_put net/ipv4/inet_timewait_sock.c:221\n inet_twsk_purge+0x725/0x890 net/ipv4/inet_timewait_sock.c:304\n tcp_sk_exit_batch+0x1c/0x170 net/ipv4/tcp_ipv4.c:3522\n ops_exit_list+0x128/0x180 net/core/net_namespace.c:178\n setup_net+0x714/0xb40 net/core/net_namespace.c:375\n copy_net_ns+0x2f0/0x670 net/core/net_namespace.c:508\n create_new_namespaces+0x3ea/0xb10 kernel/nsproxy.c:110\n\n... because refcount_dec() of tw_refcount unexpectedly dropped to 0.\n\nThis doesn't seem like an actual bug (no tw sockets got lost and I don't\nsee a use-after-free) but as erroneous trigger of debug check.\n\nAdd a mutex to force strict ordering: the task that calls tcp_twsk_purge()\nblocks other task from doing final _dec_and_test before mutex-owner has\nremoved all tw sockets of dying netns."
"value":"En el kernel de Linux, se ha resuelto la siguiente vulnerabilidad: tcp: evitar la ejecuci\u00f3n concurrente de tcp_sk_exit_batch Es posible que dos subprocesos llamen a tcp_sk_exit_batch() simult\u00e1neamente, una vez desde la cola de trabajo cleanup_net, otra desde una tarea que no pudo clonar una nueva netns. En el \u00faltimo caso, el desenrollado de errores llama a los controladores de salida en orden inverso para las netns \"fallidas\". tcp_sk_exit_batch() llama a tcp_twsk_purge(). El problema es que desde el commit b099ce2602d8 (\"net: Batch inet_twsk_purge\"), esta funci\u00f3n recoge twsk en cualquier netn moribundo, no solo en el que se pasa a trav\u00e9s de la lista exit_batch. Esto significa que el desenrollado de errores de setup_net() puede \"robar\" y destruir los sockets timewait que pertenecen a las netns que salen. Esto permite que el trabajador de salida netns proceda a llamar a WARN_ON_ONCE(!refcount_dec_and_test(&net->ipv4.tcp_death_row.tw_refcount)); sin la transici\u00f3n esperada de 1 -> 0, que luego falla. Al mismo tiempo, la ruta de desenrollado de error que tambi\u00e9n est\u00e1 ejecutando inet_twsk_purge() tambi\u00e9n se mostrar\u00e1: ADVERTENCIA: .. en lib/refcount.c:31 refcount_warn_saturate+0x1ed/0x210 ... refcount_dec include/linux/refcount.h:351 [en l\u00ednea] inet_twsk_kill+0x758/0x9c0 net/ipv4/inet_timewait_sock.c:70 inet_twsk_deschedule_put net/ipv4/inet_timewait_sock.c:221 inet_twsk_purge+0x725/0x890 net/ipv4/inet_timewait_sock.c:304 tcp_sk_exit_batch+0x1c/0x170 net/ipv4/tcp_ipv4.c:3522 ops_exit_list+0x128/0x180 net/core/net_namespace.c:178 setup_net+0x714/0xb40 net/core/net_namespace.c:375 copy_net_ns+0x2f0/0x670 net/core/net_namespace.c:508 create_new_namespaces+0x3ea/0xb10 kernel/nsproxy.c:110 ... porque refcount_dec() de tw_refcount cay\u00f3 inesperadamente a 0. Esto no parece un error real (no se perdieron sockets tw y no veo un use-after-free) sino un disparador err\u00f3neo de la comprobaci\u00f3n de depuraci\u00f3n. Agregue un mutex para forzar un orden estricto: la tarea que llama a tcp_twsk_purge() impide que otra tarea realice _dec_and_test final antes de que el propietario del mutex haya eliminado todos los sockets tw de los netn moribundos."
