Add a new type PROC_CN_MCAST_NOTIFY to proc connector API, which allows a thread to notify the kernel that is going to exit with a non-zero exit code and specify the exit code in it. When thread exits in the kernel, it will send this exit code as a proc filter notification to any listening process. Exiting thread can call this either when it wants to call pthread_exit() with non-zero value or from signal handler.

Add a new file cn_hash.c which implements a hash table storing the exit codes of abnormally exiting threads, received by the system call above. The key used for the hash table is the pid of the thread, so when the thread actually exits, we lookup it's pid in the hash table and retrieve the exit code sent by user. If the exit code in struct task is 0, we then replace it with the user supplied non-zero exit code.

cn_hash.c implements the hash table add, delete, lookup operations. mutex_lock() and mutex_unlock() operations are used to safeguard the integrity of the hash table while adding or deleting elements. connector.c has the API calls, called from cn_proc.c, as well as calls to allocate, initialize and free the hash table.

Add a new flag in PF_* flags of task_struct - EXIT_NOTIFY. This flag is set when user sends the exit code via PROC_CN_MCAST_NOTIFY. While exiting, this flag is checked and the hash table add or delete calls are only made if this flag is set.

A refcount field hrefcnt is added in struct cn_hash_dev, to keep track of number of threads which have added an entry in hash table. Before freeing the struct cn_hash_dev, this value must be 0. This refcnt check is added in case CONFIG_CONNECTOR is compiled as a module. In that case, when unloading the module, we need to make sure no hash entries are still present in the hdev table.

Copy the task's name (task->comm) into the exit event notification. This will allow applications to filter on the name further using userspace filtering like ebpf.

Signed-off-by: Anjali Kulkarni anjali.k.kulkarni@oracle.com --- drivers/connector/Makefile | 2 +- drivers/connector/cn_hash.c | 181 ++++++++++++++++++++++++++++++++++ drivers/connector/cn_proc.c | 62 +++++++++++- drivers/connector/connector.c | 63 +++++++++++- include/linux/connector.h | 31 ++++++ include/linux/sched.h | 2 +- include/uapi/linux/cn_proc.h | 5 +- 7 files changed, 338 insertions(+), 8 deletions(-) create mode 100644 drivers/connector/cn_hash.c

diff --git a/drivers/connector/Makefile b/drivers/connector/Makefile index 1bf67d3df97d..cb1dcdf067ad 100644 --- a/drivers/connector/Makefile +++ b/drivers/connector/Makefile @@ -2,4 +2,4 @@ obj-$(CONFIG_CONNECTOR) += cn.o obj-$(CONFIG_PROC_EVENTS) += cn_proc.o

-cn-y += cn_queue.o connector.o +cn-y += cn_hash.o cn_queue.o connector.o diff --git a/drivers/connector/cn_hash.c b/drivers/connector/cn_hash.c new file mode 100644 index 000000000000..a079e9bcea6d --- /dev/null +++ b/drivers/connector/cn_hash.c @@ -0,0 +1,181 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Author: Anjali Kulkarni anjali.k.kulkarni@oracle.com + * + * Copyright (c) 2024 Oracle and/or its affiliates. + */ + +#include <linux/kernel.h> +#include <linux/init.h> +#include <linux/connector.h> +#include <linux/mutex.h> +#include <linux/pid_namespace.h> + +#include <linux/cn_proc.h> + +struct cn_hash_dev *cn_hash_alloc_dev(const char *name) +{ + struct cn_hash_dev *hdev; + + hdev = kzalloc(sizeof(*hdev), GFP_KERNEL); + if (!hdev) + return NULL; + + snprintf(hdev->name, sizeof(hdev->name), "%s", name); + atomic_set(&hdev->hrefcnt, 0); + mutex_init(&hdev->uexit_hash_lock); + hash_init(hdev->uexit_pid_htable); + return hdev; +} + +void cn_hash_free_dev(struct cn_hash_dev *hdev) +{ + struct uexit_pid_hnode *hnode; + struct hlist_node *tmp; + int bucket; + + pr_debug("%s: Freeing entire hdev %p\n", __func__, hdev); + + mutex_lock(&hdev->uexit_hash_lock); + hash_for_each_safe(hdev->uexit_pid_htable, bucket, tmp, + hnode, uexit_pid_hlist) { + hash_del(&hnode->uexit_pid_hlist); + pr_debug("%s: Freeing node for pid %d\n", + __func__, hnode->pid); + kfree(hnode); + } + + mutex_unlock(&hdev->uexit_hash_lock); + mutex_destroy(&hdev->uexit_hash_lock); + + /* + * This refcnt check is added in case CONFIG_CONNECTOR is + * compiled with =m as a module. In that case, when unloading + * the module, we need to make sure no hash entries are still + * present in the hdev table. + */ + while (atomic_read(&hdev->hrefcnt)) { + pr_info("Waiting for %s to become free: refcnt=%d\n", + hdev->name, atomic_read(&hdev->hrefcnt)); + msleep(1000); + } + + kfree(hdev); + hdev = NULL; +} + +static struct uexit_pid_hnode *cn_hash_alloc_elem(__u32 uexit_code, pid_t pid) +{ + struct uexit_pid_hnode *elem; + + elem = kzalloc(sizeof(*elem), GFP_KERNEL); + if (!elem) + return NULL; + + INIT_HLIST_NODE(&elem->uexit_pid_hlist); + elem->uexit_code = uexit_code; + elem->pid = pid; + return elem; +} + +static inline void cn_hash_free_elem(struct uexit_pid_hnode *elem) +{ + kfree(elem); +} + +int cn_hash_add_elem(struct cn_hash_dev *hdev, __u32 uexit_code, pid_t pid) +{ + struct uexit_pid_hnode *elem, *hnode; + + elem = cn_hash_alloc_elem(uexit_code, pid); + if (!elem) { + pr_err("%s: cn_hash_alloc_elem() returned NULL pid %d\n", + __func__, pid); + return -ENOMEM; + } + + mutex_lock(&hdev->uexit_hash_lock); + /* + * Check if an entry for the same pid already exists + */ + hash_for_each_possible(hdev->uexit_pid_htable, + hnode, uexit_pid_hlist, pid) { + if (hnode->pid == pid) { + mutex_unlock(&hdev->uexit_hash_lock); + cn_hash_free_elem(elem); + pr_debug("%s: pid %d already exists in hash table\n", + __func__, pid); + return -EEXIST; + } + } + + hash_add(hdev->uexit_pid_htable, &elem->uexit_pid_hlist, pid); + mutex_unlock(&hdev->uexit_hash_lock); + + atomic_inc(&hdev->hrefcnt); + + pr_debug("%s: After hash_add of pid %d elem %p hrefcnt %d\n", + __func__, pid, elem, atomic_read(&hdev->hrefcnt)); + return 0; +} + +int cn_hash_del_get_exval(struct cn_hash_dev *hdev, pid_t pid) +{ + struct uexit_pid_hnode *hnode; + struct hlist_node *tmp; + int excde; + + mutex_lock(&hdev->uexit_hash_lock); + hash_for_each_possible_safe(hdev->uexit_pid_htable, + hnode, tmp, uexit_pid_hlist, pid) { + if (hnode->pid == pid) { + excde = hnode->uexit_code; + hash_del(&hnode->uexit_pid_hlist); + mutex_unlock(&hdev->uexit_hash_lock); + kfree(hnode); + atomic_dec(&hdev->hrefcnt); + pr_debug("%s: After hash_del of pid %d, found exit code %u hrefcnt %d\n", + __func__, pid, excde, + atomic_read(&hdev->hrefcnt)); + return excde; + } + } + + mutex_unlock(&hdev->uexit_hash_lock); + pr_err("%s: pid %d not found in hash table\n", + __func__, pid); + return -EINVAL; +} + +int cn_hash_get_exval(struct cn_hash_dev *hdev, pid_t pid) +{ + struct uexit_pid_hnode *hnode; + __u32 excde; + + mutex_lock(&hdev->uexit_hash_lock); + hash_for_each_possible(hdev->uexit_pid_htable, + hnode, uexit_pid_hlist, pid) { + if (hnode->pid == pid) { + excde = hnode->uexit_code; + mutex_unlock(&hdev->uexit_hash_lock); + pr_debug("%s: Found exit code %u for pid %d\n", + __func__, excde, pid); + return excde; + } + } + + mutex_unlock(&hdev->uexit_hash_lock); + pr_debug("%s: pid %d not found in hash table\n", + __func__, pid); + return -EINVAL; +} + +bool cn_hash_table_empty(struct cn_hash_dev *hdev) +{ + bool is_empty; + + is_empty = hash_empty(hdev->uexit_pid_htable); + pr_debug("Hash table is %s\n", (is_empty ? "empty" : "not empty")); + + return is_empty; +} diff --git a/drivers/connector/cn_proc.c b/drivers/connector/cn_proc.c index 44b19e696176..0632a70a89a0 100644 --- a/drivers/connector/cn_proc.c +++ b/drivers/connector/cn_proc.c @@ -69,6 +69,8 @@ static int cn_filter(struct sock *dsk, struct sk_buff *skb, void *data) if ((__u32)val == PROC_EVENT_ALL) return 0;

+ pr_debug("%s: val %lx, what %x\n", __func__, val, what); + /* * Drop packet if we have to report only non-zero exit status * (PROC_EVENT_NONZERO_EXIT) and exit status is 0 @@ -326,9 +328,15 @@ void proc_exit_connector(struct task_struct *task) struct proc_event *ev; struct task_struct *parent; __u8 buffer[CN_PROC_MSG_SIZE] __aligned(8); + int uexit_code;

- if (atomic_read(&proc_event_num_listeners) < 1) + if (atomic_read(&proc_event_num_listeners) < 1) { + if (likely(!(task->flags & PF_EXIT_NOTIFY))) + return; + + cn_del_get_exval(task->pid); return; + }

msg = buffer_to_cn_msg(buffer); ev = (struct proc_event *)msg->data; @@ -337,7 +345,26 @@ void proc_exit_connector(struct task_struct *task) ev->what = PROC_EVENT_EXIT; ev->event_data.exit.process_pid = task->pid; ev->event_data.exit.process_tgid = task->tgid; - ev->event_data.exit.exit_code = task->exit_code; + if (unlikely(task->flags & PF_EXIT_NOTIFY)) { + task->flags &= ~PF_EXIT_NOTIFY; + + uexit_code = cn_del_get_exval(task->pid); + if (uexit_code <= 0) { + pr_debug("%s: err %d returning task's exit code %u\n", + uexit_code, __func__, + task->exit_code); + ev->event_data.exit.exit_code = task->exit_code; + } else { + ev->event_data.exit.exit_code = uexit_code; + pr_debug("%s: Reset PF_EXIT_NOTIFY & retrieved exit code %u from hash table, pid %d\n", + __func__, + ev->event_data.exit.exit_code, + task->pid); + } + } else { + ev->event_data.exit.exit_code = task->exit_code; + } + ev->event_data.exit.exit_signal = task->exit_signal;

rcu_read_lock(); @@ -348,6 +375,13 @@ void proc_exit_connector(struct task_struct *task) } rcu_read_unlock();

+ /* + * Copy task name in the packet. This will allow applications + * to filter on the name further using userspace filtering like + * ebpf + */ + get_task_comm(ev->event_data.exit.comm, task); + memcpy(&msg->id, &cn_proc_event_id, sizeof(msg->id)); msg->ack = 0; /* not used */ msg->len = sizeof(*ev); @@ -413,6 +447,13 @@ static void cn_proc_mcast_ctl(struct cn_msg *msg, if (msg->len == sizeof(*pinput)) { pinput = (struct proc_input *)msg->data; mc_op = pinput->mcast_op; + if (mc_op == PROC_CN_MCAST_NOTIFY) { + pr_debug("%s: Received PROC_CN_MCAST_NOTIFY, pid %d\n", + __func__, current->pid); + current->flags |= PF_EXIT_NOTIFY; + err = cn_add_elem(pinput->uexit_code, current->pid); + return; + } ev_type = pinput->event_type; } else if (msg->len == sizeof(mc_op)) { mc_op = *((enum proc_cn_mcast_op *)msg->data); @@ -432,6 +473,8 @@ static void cn_proc_mcast_ctl(struct cn_msg *msg, sk->sk_user_data = kzalloc(sizeof(struct proc_input), GFP_KERNEL); if (sk->sk_user_data == NULL) { + pr_err("%s: ENOMEM for sk_user_data, pid %d\n", + __func__, current->pid); err = ENOMEM; goto out; } @@ -442,21 +485,32 @@ static void cn_proc_mcast_ctl(struct cn_msg *msg, } ((struct proc_input *)(sk->sk_user_data))->event_type = ev_type; + pr_debug("%s: sk: %p pid: %d event_type: %x\n", + __func__, sk, current->pid, ev_type); ((struct proc_input *)(sk->sk_user_data))->mcast_op = mc_op; }

switch (mc_op) { case PROC_CN_MCAST_LISTEN: - if (initial || (prev_mc_op != PROC_CN_MCAST_LISTEN)) + if (initial || (prev_mc_op != PROC_CN_MCAST_LISTEN)) { atomic_inc(&proc_event_num_listeners); + pr_debug("%s: PROC_CN_MCAST_LISTEN pid %d: Incremented listeners to %d\n", + __func__, current->pid, + atomic_read(&proc_event_num_listeners)); + } break; case PROC_CN_MCAST_IGNORE: - if (!initial && (prev_mc_op != PROC_CN_MCAST_IGNORE)) + if (!initial && (prev_mc_op != PROC_CN_MCAST_IGNORE)) { atomic_dec(&proc_event_num_listeners); + pr_debug("%s: PROC_CN_MCAST_IGNORE pid %d: Decremented listeners to %d\n", + __func__, current->pid, + atomic_read(&proc_event_num_listeners)); + } ((struct proc_input *)(sk->sk_user_data))->event_type = PROC_EVENT_NONE; break; default: + pr_warn("%s: Invalid value for mc_op %d\n", __func__, mc_op); err = EINVAL; break; } diff --git a/drivers/connector/connector.c b/drivers/connector/connector.c index 4028e8eeba82..c1c0dcec53c0 100644 --- a/drivers/connector/connector.c +++ b/drivers/connector/connector.c @@ -271,6 +271,50 @@ static int __maybe_unused cn_proc_show(struct seq_file *m, void *v) return 0; }

+int cn_del_get_exval(pid_t pid) +{ + struct cn_dev *dev = &cdev; + + if (!cn_already_initialized) + return 0; + + return cn_hash_del_get_exval(dev->hdev, pid); +} +EXPORT_SYMBOL_GPL(cn_del_get_exval); + +int cn_add_elem(__u32 uexit_code, pid_t pid) +{ + struct cn_dev *dev = &cdev; + + if (!cn_already_initialized) + return 0; + + return cn_hash_add_elem(dev->hdev, uexit_code, pid); +} +EXPORT_SYMBOL_GPL(cn_add_elem); + +int cn_get_exval(pid_t pid) +{ + struct cn_dev *dev = &cdev; + + if (!cn_already_initialized) + return 0; + + return cn_hash_get_exval(dev->hdev, pid); +} +EXPORT_SYMBOL_GPL(cn_get_exval); + +bool cn_table_empty(void) +{ + struct cn_dev *dev = &cdev; + + if (!cn_already_initialized) + return 0; + + return cn_hash_table_empty(dev->hdev); +} +EXPORT_SYMBOL_GPL(cn_table_empty); + static int cn_init(void) { struct cn_dev *dev = &cdev; @@ -283,15 +327,31 @@ static int cn_init(void) };

dev->nls = netlink_kernel_create(&init_net, NETLINK_CONNECTOR, &cfg); - if (!dev->nls) + if (!dev->nls) { + pr_err("%s: netlink_kernel_create failed, connector not initialized\n", + __func__); return -EIO; + }

dev->cbdev = cn_queue_alloc_dev("cqueue", dev->nls); if (!dev->cbdev) { + pr_err("%s: Allocation of dev->cbdev failed, connector not initialized\n", + __func__); netlink_kernel_release(dev->nls); return -EINVAL; }

+ dev->hdev = cn_hash_alloc_dev("pid hash table"); + if (!dev->hdev) { + pr_err("%s: Allocation of dev->hdev failed, connector not initialized\n", + __func__); + netlink_kernel_release(dev->nls); + cn_queue_free_dev(dev->cbdev); + return -ENOMEM; + } + + pr_debug("Connector initialized, allocated hdev %p\n", dev->hdev); + cn_already_initialized = 1;

proc_create_single("connector", S_IRUGO, init_net.proc_net, cn_proc_show); @@ -308,6 +368,7 @@ static void cn_fini(void) remove_proc_entry("connector", init_net.proc_net);

cn_queue_free_dev(dev->cbdev); + cn_hash_free_dev(dev->hdev); netlink_kernel_release(dev->nls); }

diff --git a/include/linux/connector.h b/include/linux/connector.h index 70bc1160f3d8..5384e4bb98e8 100644 --- a/include/linux/connector.h +++ b/include/linux/connector.h @@ -18,6 +18,8 @@ #include <uapi/linux/connector.h>

#define CN_CBQ_NAMELEN 32 +#define HASHT_NAMELEN 32 +#define PID_HASH_TABLE_BITS 10

struct cn_queue_dev { atomic_t refcnt; @@ -45,6 +47,19 @@ struct cn_callback_entry { u32 seq, group; };

+struct uexit_pid_hnode { + __u32 uexit_code; + pid_t pid; + struct hlist_node uexit_pid_hlist; +}; + +struct cn_hash_dev { + atomic_t hrefcnt; + unsigned char name[HASHT_NAMELEN]; + struct mutex uexit_hash_lock; + DECLARE_HASHTABLE(uexit_pid_htable, PID_HASH_TABLE_BITS); +}; + struct cn_dev { struct cb_id id;

@@ -52,6 +67,7 @@ struct cn_dev { struct sock *nls;

struct cn_queue_dev *cbdev; + struct cn_hash_dev *hdev; };

/** @@ -137,4 +153,19 @@ void cn_queue_free_dev(struct cn_queue_dev *dev);

int cn_cb_equal(const struct cb_id *, const struct cb_id *);

+struct cn_hash_dev *cn_hash_alloc_dev(const char *name); +void cn_hash_free_dev(struct cn_hash_dev *hdev); +struct uexit_pid_hnode *cn_hash_find_pid_node(struct cn_hash_dev *hdev, + pid_t pid); +int cn_hash_add_elem(struct cn_hash_dev *hdev, __u32 uexit_code, pid_t pid); +int cn_hash_del_get_exval(struct cn_hash_dev *hdev, pid_t pid); +int cn_hash_get_exval(struct cn_hash_dev *hdev, pid_t pid); + +int cn_add_elem(__u32 uexit_code, pid_t pid); +int cn_del_get_exval(pid_t pid); +int cn_get_exval(pid_t pid); + +bool cn_table_empty(void); +bool cn_hash_table_empty(struct cn_hash_dev *hdev); + #endif /* __CONNECTOR_H */ diff --git a/include/linux/sched.h b/include/linux/sched.h index e6ee4258169a..a2339ae6208b 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1673,7 +1673,7 @@ extern struct pid *cad_pid; #define PF_USED_MATH 0x00002000 /* If unset the fpu must be initialized before use */ #define PF_USER_WORKER 0x00004000 /* Kernel thread cloned from userspace thread */ #define PF_NOFREEZE 0x00008000 /* This thread should not be frozen */ -#define PF__HOLE__00010000 0x00010000 +#define PF_EXIT_NOTIFY 0x00010000 /* This thread has sent an exit value to be sent as a notification to listening processes */ #define PF_KSWAPD 0x00020000 /* I am kswapd */ #define PF_MEMALLOC_NOFS 0x00040000 /* All allocations inherit GFP_NOFS. See memalloc_nfs_save() */ #define PF_MEMALLOC_NOIO 0x00080000 /* All allocations inherit GFP_NOIO. See memalloc_noio_save() */ diff --git a/include/uapi/linux/cn_proc.h b/include/uapi/linux/cn_proc.h index 18e3745b86cd..e0ddb2e2c069 100644 --- a/include/uapi/linux/cn_proc.h +++ b/include/uapi/linux/cn_proc.h @@ -27,7 +27,8 @@ */ enum proc_cn_mcast_op { PROC_CN_MCAST_LISTEN = 1, - PROC_CN_MCAST_IGNORE = 2 + PROC_CN_MCAST_IGNORE = 2, + PROC_CN_MCAST_NOTIFY = 3 };

#define PROC_EVENT_ALL (PROC_EVENT_FORK | PROC_EVENT_EXEC | PROC_EVENT_UID | \ @@ -65,6 +66,7 @@ enum proc_cn_event { struct proc_input { enum proc_cn_mcast_op mcast_op; enum proc_cn_event event_type; + __u32 uexit_code; };

static inline enum proc_cn_event valid_event(enum proc_cn_event ev_type) @@ -151,6 +153,7 @@ struct proc_event { __u32 exit_code, exit_signal; __kernel_pid_t parent_pid; __kernel_pid_t parent_tgid; + char comm[16]; } exit;

} event_data;

Test to check if setting PROC_CN_MCAST_NOTIFY in proc connector API, allows a thread's non-zero exit status to be returned to proc_filter.

The threads.c program creates 2 child threads. 1st thread handles signal SIGSEGV, and 2nd thread needs to indicate some error condition (value 1) to the kernel, instead of using pthread_exit() with 1.

In both cases, child sends notify_netlink_thread_exit(exit_code) to kernel, to let kernel know it has exited abnormally with exit_code.

Compile: make thread make proc_filter To see non-zero exit notifications, run: ./proc_filter -f Run threads code in another window: ./threads The above will also send SIGSEGV to one of the threads created. After you see the proc_filter display both tids, kill it: [Ctrl-C] in the window where "./proc_filter -f" is being run. The results will be printed in the "./threads" window.

Signed-off-by: Anjali Kulkarni anjali.k.kulkarni@oracle.com --- tools/testing/selftests/connector/Makefile | 23 +- .../testing/selftests/connector/proc_filter.c | 34 ++- tools/testing/selftests/connector/thread.c | 202 ++++++++++++++++++ .../selftests/connector/thread_filter.c | 96 +++++++++ 4 files changed, 348 insertions(+), 7 deletions(-) create mode 100644 tools/testing/selftests/connector/thread.c create mode 100644 tools/testing/selftests/connector/thread_filter.c

diff --git a/tools/testing/selftests/connector/Makefile b/tools/testing/selftests/connector/Makefile index 92188b9bac5c..bf335826bc3b 100644 --- a/tools/testing/selftests/connector/Makefile +++ b/tools/testing/selftests/connector/Makefile @@ -1,5 +1,26 @@ # SPDX-License-Identifier: GPL-2.0 -CFLAGS += -Wall $(KHDR_INCLUDES) +KERNEL="../../../.." + +CFLAGS += -Wall $(KHDR_INCLUDES) -I $(KERNEL)/include/uapi -I $(KERNEL)/include + +proc_filter: proc_filter.o + cc proc_filter.o -o proc_filter + +proc_filter.o: proc_filter.c + cc -c proc_filter.c -o proc_filter.o $(CFLAGS) + +thread: thread.o thread_filter.o + cc thread.o thread_filter.o -o thread + +thread.o: thread.c $(DEPS) + cc -c thread.c -o thread.o $(CFLAGS) + +thread_filter.o: thread_filter.c + cc -c thread_filter.c -o thread_filter.o $(CFLAGS) + +define EXTRA_CLEAN + rm *.o thread +endef

TEST_GEN_PROGS = proc_filter

diff --git a/tools/testing/selftests/connector/proc_filter.c b/tools/testing/selftests/connector/proc_filter.c index 4a825b997666..2374c7d9c7b7 100644 --- a/tools/testing/selftests/connector/proc_filter.c +++ b/tools/testing/selftests/connector/proc_filter.c @@ -1,4 +1,9 @@ // SPDX-License-Identifier: GPL-2.0-only +/* + * Author: Anjali Kulkarni anjali.k.kulkarni@oracle.com + * + * Copyright (c) 2024 Oracle and/or its affiliates. + */

#include <sys/types.h> #include <sys/epoll.h> @@ -28,6 +33,7 @@ volatile static int interrupted; static int nl_sock, ret_errno, tcount; static struct epoll_event evn; +FILE *file = NULL;

static int filter;

@@ -37,6 +43,8 @@ static int filter; #define Printf ksft_print_msg #endif

+#define EXIT_LOG + int send_message(void *pinp) { char buff[NL_MESSAGE_SIZE]; @@ -146,6 +154,12 @@ int handle_packet(char *buff, int fd, struct proc_event *event) tcount++; switch (event->what) { case PROC_EVENT_EXIT: +#ifdef EXIT_LOG + fprintf(file, "pid %d tgid %d code %d\n", + event->event_data.exit.process_pid, + event->event_data.exit.process_tgid, + event->event_data.exit.exit_code); +#endif Printf("Exit process %d (tgid %d) with code %d, signal %d\n", event->event_data.exit.process_pid, event->event_data.exit.process_tgid, @@ -279,17 +293,24 @@ int main(int argc, char *argv[]) exit(1); }

+#ifdef EXIT_LOG + file = fopen("exit.log", "w"); + if (file == NULL) { + perror("Error opening file exit.log"); + close(nl_sock); + close(epoll_fd); + exit(1); + } +#endif + while (!interrupted) { err = handle_events(epoll_fd, &proc_ev); if (err < 0) { if (ret_errno == EINTR) continue; - if (err == -2) - close(nl_sock); - if (err == -3) { - close(nl_sock); - close(epoll_fd); - } + close(nl_sock); + close(epoll_fd); + fclose(file); exit(1); } } @@ -304,6 +325,7 @@ int main(int argc, char *argv[])

close(epoll_fd); close(nl_sock); + fclose(file);

printf("Done total count: %d\n", tcount); exit(0); diff --git a/tools/testing/selftests/connector/thread.c b/tools/testing/selftests/connector/thread.c new file mode 100644 index 000000000000..afab96b9fed9 --- /dev/null +++ b/tools/testing/selftests/connector/thread.c @@ -0,0 +1,202 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Author: Anjali Kulkarni anjali.k.kulkarni@oracle.com + * + * Copyright (c) 2024 Oracle and/or its affiliates. + */ + +#include <pthread.h> +#include <stdio.h> +#include <unistd.h> +#include <stdlib.h> +#include <signal.h> + +/* + * This code tests a thread exit notification when thread exits abnormally. + * Normally, when a thread exits abnormally, the kernel is not aware of the + * exit code. This is usually only conveyed from child to parent via the + * pthread_exit() and pthread_join() calls. Sometimes, however, a parent + * process cannot monitor all child processes via pthread_join(), particularly + * when there is a huge amount of child processes. In this case, the parent + * has created the child with PTHREAD_CREATE_DETACHED attribute. + * To fix this problem, either when child wants to convey non-zero exit via + * pthread_exit() or in a signal handler, the child can notify the kernel's + * connector module it's exit status via a netlink call with new type + * PROC_CN_MCAST_NOTIFY. (Implemented in the thread_filter.c file). + * This will send the exit code from the child to the kernel, which the kernel + * can later return to proc_filter program when the child actually exits. + * To test this usecase: + * Compile: + * make thread + * make proc_filter + * To see non-zero exit notifications, run: + * ./proc_filter -f + * Start the threads code, creating 2 threads, in another window: + * ./threads + * The above will also send SIGSEGV to one of the threads created. + * After you see the proc_filter display both tids, kill it: + * [Ctrl-C] in the window where "./proc_filter -f" is being run. + * The results will be printed in the "./threads" window. + */ + +extern int notify_netlink_thread_exit(unsigned int exit_code); + +static void sigsegvh(int sig) +{ + unsigned int exit_code = (unsigned int) sig; + /* + * Send any non-zero value to get a notification. Here we are + * sending the signal number for SIGSEGV which is 11 + */ + notify_netlink_thread_exit(exit_code); +} + +void *threadc1(void *ptr) +{ + signal(SIGSEGV, sigsegvh); + + *(int *)ptr = gettid(); + + printf("Child 1 thread id %d, handling SIGSEGV\n", gettid()); + sleep(10); + pthread_exit(NULL); +} + +void *threadc2(void *ptr) +{ + int exit_val = 1; + + *(int *)ptr = gettid(); + + printf("Child 2 thread id %d, wants to exit with value %d\n", + gettid(), exit_val); + sleep(2); + notify_netlink_thread_exit(exit_val); + pthread_exit(NULL); +} + +static void verify_exit_status(int tid1, int tid2) +{ + int found1 = 0, found2 = 0; + int pid, tgid, exit_code; + size_t size = 1024; + FILE *file; + char *data; + + data = malloc(size * sizeof(char)); + if (data == NULL) { + perror("malloc for data failed"); + exit(1); + } + + file = fopen("exit.log", "r"); + if (file == NULL) { + perror("fopen of exit.log failed"); + free(data); + exit(1); + } + + while (getline(&data, &size, file) != -1) { + sscanf(data, "pid %d tgid %d code %d", + &pid, &tgid, &exit_code); + + if (tgid != getpid()) { + continue; + } + + if (pid == tid1) { + if (exit_code == 11) { + printf("Successful notification of SIGSEGV, tid %d\n", + pid); + } else { + printf("Failure SIGSEGV tid %d, exit code %d\n", + pid, exit_code); + } + found1 = 1; + } else if (pid == tid2) { + if (exit_code == 1) { + printf("Successful notification of thread exit tid %d\n", + pid); + } else { + printf("Failure thread exit tid %d, exit code %d\n", + pid, exit_code); + } + found2 = 1; + } + } + + if (!found1) + printf("tid %d not present in exit.log file\n", tid1); + + if (!found2) + printf("tid %d not present in exit.log file\n", tid2); + + fclose(file); + free(data); +} + +static inline void init_threads(pthread_attr_t *attr) +{ + int ret; + + ret = pthread_attr_init(attr); + if (ret != 0) { + perror("pthread_attr_init failed"); + exit(ret); + } + + ret = pthread_attr_setdetachstate(attr, PTHREAD_CREATE_DETACHED); + if (ret != 0) { + perror("pthread_attr_setdetachstate failed"); + exit(ret); + } +} + +static inline void destroy_thread_attr(pthread_attr_t *attr) +{ + int ret; + + ret = pthread_attr_destroy(attr); + if (ret != 0) { + perror("pthread_attr_destroy failed"); + exit(ret); + } +} + +int main(int argc, char **argv) +{ + pthread_t thread1, thread2; + pthread_attr_t attr1, attr2; + int tid1, tid2, ret; + + init_threads(&attr1); + ret = pthread_create(&thread1, &attr1, *threadc1, &tid1); + if (ret != 0) { + perror("pthread_create failed"); + exit(ret); + } + + init_threads(&attr2); + ret = pthread_create(&thread2, &attr2, *threadc2, &tid2); + if (ret != 0) { + perror("pthread_create failed"); + exit(ret); + } + + sleep(1); + + /* Send SIGSEGV to tid1 */ + kill(tid1, SIGSEGV); + + /* Wait for children to exit or be killed and for exit.log to + * be generated by ./proc_filter + */ + sleep(10); + + verify_exit_status(tid1, tid2); + + destroy_thread_attr(&attr1); + destroy_thread_attr(&attr2); + + exit(0); +} diff --git a/tools/testing/selftests/connector/thread_filter.c b/tools/testing/selftests/connector/thread_filter.c new file mode 100644 index 000000000000..3da740aa7537 --- /dev/null +++ b/tools/testing/selftests/connector/thread_filter.c @@ -0,0 +1,96 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Author: Anjali Kulkarni anjali.k.kulkarni@oracle.com + * + * Copyright (c) 2024 Oracle and/or its affiliates. + */ + +#include <sys/types.h> +#include <sys/epoll.h> +#include <sys/socket.h> +#include <linux/netlink.h> +#include <linux/connector.h> +#include <linux/cn_proc.h> + +#include <stddef.h> +#include <stdio.h> +#include <stdlib.h> +#include <unistd.h> +#include <strings.h> +#include <errno.h> +#include <signal.h> +#include <string.h> + +#define NL_MESSAGE_SIZE (sizeof(struct nlmsghdr) + sizeof(struct cn_msg) + \ + sizeof(struct proc_input)) + +/* + * Send PROC_CN_MCAST_NOTIFY type notification to the connector code in kernel. + * This will send the exit_code specified by user to the connector layer, so + * it can send a notification for that event to any listening process + */ +int send_message(int nl_sock, unsigned int exit_code) +{ + char buff[NL_MESSAGE_SIZE]; + struct nlmsghdr *hdr; + struct cn_msg *msg; + + hdr = (struct nlmsghdr *)buff; + hdr->nlmsg_len = NL_MESSAGE_SIZE; + hdr->nlmsg_type = NLMSG_DONE; + hdr->nlmsg_flags = 0; + hdr->nlmsg_seq = 0; + hdr->nlmsg_pid = getpid(); + + msg = (struct cn_msg *)NLMSG_DATA(hdr); + msg->id.idx = CN_IDX_PROC; + msg->id.val = CN_VAL_PROC; + msg->seq = 0; + msg->ack = 0; + msg->flags = 0; + + msg->len = sizeof(struct proc_input); + ((struct proc_input *)msg->data)->mcast_op = + PROC_CN_MCAST_NOTIFY; + ((struct proc_input *)msg->data)->uexit_code = exit_code; + + if (send(nl_sock, hdr, hdr->nlmsg_len, 0) == -1) { + perror("send failed"); + return -errno; + } + return 0; +} + +int notify_netlink_thread_exit(unsigned int exit_code) +{ + struct sockaddr_nl sa_nl; + int err = 0; + int nl_sock; + + nl_sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_CONNECTOR); + + if (nl_sock == -1) { + perror("socket failed"); + return -errno; + } + + bzero(&sa_nl, sizeof(sa_nl)); + sa_nl.nl_family = AF_NETLINK; + sa_nl.nl_groups = CN_IDX_PROC; + sa_nl.nl_pid = gettid(); + + if (bind(nl_sock, (struct sockaddr *)&sa_nl, sizeof(sa_nl)) == -1) { + perror("bind failed"); + close(nl_sock); + return -errno; + } + + err = send_message(nl_sock, exit_code); + + close(nl_sock); + + if (err < 0) + return err; + + return 0; +}