Rename "tsc" to "timestamp"
[lttng-modules.git] / src / lttng-statedump-impl.c
index 394f661861a88a8f00a52002b2440112b9df2b68..4a228ce41a577422b6ac4c4c50abc0620d808c8c 100644 (file)
 #include <linux/proc_fs.h>
 #include <linux/file.h>
 #include <linux/interrupt.h>
+#include <linux/irq.h>
 #include <linux/irqnr.h>
-#include <linux/cpu.h>
 #include <linux/netdevice.h>
 #include <linux/inetdevice.h>
-#include <linux/sched.h>
 #include <linux/mm.h>
 #include <linux/swap.h>
 #include <linux/wait.h>
 #include <linux/mutex.h>
 #include <linux/device.h>
 
+#include <linux/blkdev.h>
+
 #include <lttng/events.h>
 #include <lttng/tracer.h>
+#include <wrapper/cpu.h>
 #include <wrapper/irqdesc.h>
 #include <wrapper/fdtable.h>
-#include <wrapper/namespace.h>
-#include <wrapper/irq.h>
 #include <wrapper/tracepoint.h>
-#include <wrapper/genhd.h>
-#include <wrapper/file.h>
+#include <wrapper/blkdev.h>
 #include <wrapper/fdtable.h>
-
-#ifdef CONFIG_LTTNG_HAS_LIST_IRQ
-#include <linux/irq.h>
-#endif
+#include <wrapper/sched.h>
 
 /* Define the tracepoints, but do not build the probes */
 #define CREATE_TRACE_POINTS
@@ -195,7 +191,30 @@ enum lttng_process_status {
 };
 
 
-#if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0))
+#if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,0,0) || \
+    LTTNG_RHEL_KERNEL_RANGE(5,14,0,163,0,0, 5,15,0,0,0,0))
+
+#define LTTNG_PART_STRUCT_TYPE struct block_device
+
+static
+int lttng_get_part_name(struct gendisk *disk, struct block_device *part, char *name_buf)
+{
+       int ret;
+
+       ret = snprintf(name_buf, BDEVNAME_SIZE, "%pg", part);
+       if (ret < 0 || ret >= BDEVNAME_SIZE)
+               return -ENOSYS;
+
+       return 0;
+}
+
+static
+dev_t lttng_get_part_devt(struct block_device *part)
+{
+       return part->bd_dev;
+}
+
+#elif (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,11,0))
 
 #define LTTNG_PART_STRUCT_TYPE struct block_device
 
@@ -249,6 +268,61 @@ dev_t lttng_get_part_devt(struct hd_struct *part)
 }
 #endif
 
+#if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,12,0))
+static
+int lttng_statedump_each_block_device(struct lttng_kernel_session *session, struct gendisk *disk)
+{
+       struct block_device *part;
+       unsigned long idx;
+       int ret = 0;
+
+       /* Include partition 0 */
+       idx = 0;
+
+       rcu_read_lock();
+       xa_for_each(&disk->part_tbl, idx, part) {
+               char name_buf[BDEVNAME_SIZE];
+
+               /* Exclude non-partitions bdev and empty partitions. */
+               if (bdev_is_partition(part) && !bdev_nr_sectors(part))
+                       continue;
+
+               if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) {
+                       ret = -ENOSYS;
+                       goto end;
+               }
+               trace_lttng_statedump_block_device(session, lttng_get_part_devt(part),
+                               name_buf);
+       }
+end:
+       rcu_read_unlock();
+       return ret;
+}
+#else
+static
+int lttng_statedump_each_block_device(struct lttng_kernel_session *session, struct gendisk *disk)
+{
+       struct disk_part_iter piter;
+       LTTNG_PART_STRUCT_TYPE *part;
+
+       disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
+
+       while ((part = disk_part_iter_next(&piter))) {
+               char name_buf[BDEVNAME_SIZE];
+
+               if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) {
+                       disk_part_iter_exit(&piter);
+                       return -ENOSYS;
+               }
+               trace_lttng_statedump_block_device(session, lttng_get_part_devt(part),
+                               name_buf);
+       }
+       disk_part_iter_exit(&piter);
+
+       return 0;
+}
+#endif
+
 static
 int lttng_enumerate_block_devices(struct lttng_kernel_session *session)
 {
@@ -270,41 +344,18 @@ int lttng_enumerate_block_devices(struct lttng_kernel_session *session)
        }
        class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
        while ((dev = class_dev_iter_next(&iter))) {
-               struct disk_part_iter piter;
                struct gendisk *disk = dev_to_disk(dev);
-               LTTNG_PART_STRUCT_TYPE *part;
 
                /*
                 * Don't show empty devices or things that have been
                 * suppressed
                 */
                if (get_capacity(disk) == 0 ||
-                   (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO))
+                   (disk->flags & LTTNG_GENHD_FL_HIDDEN))
                        continue;
 
-               /*
-                * The original 'disk_part_iter_init' returns void, but our
-                * wrapper can fail to lookup the original symbol.
-                */
-               if (wrapper_disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0) < 0) {
-                       ret = -ENOSYS;
-                       goto iter_exit;
-               }
-
-               while ((part = wrapper_disk_part_iter_next(&piter))) {
-                       char name_buf[BDEVNAME_SIZE];
-
-                       if (lttng_get_part_name(disk, part, name_buf) == -ENOSYS) {
-                               wrapper_disk_part_iter_exit(&piter);
-                               ret = -ENOSYS;
-                               goto iter_exit;
-                       }
-                       trace_lttng_statedump_block_device(session,
-                                       lttng_get_part_devt(part), name_buf);
-               }
-               wrapper_disk_part_iter_exit(&piter);
+               ret = lttng_statedump_each_block_device(session, disk);
        }
-iter_exit:
        class_dev_iter_exit(&iter);
 end:
        return ret;
@@ -335,6 +386,20 @@ void lttng_enumerate_device(struct lttng_kernel_session *session,
        }
 }
 
+#if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,9,0))
+static
+int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session)
+{
+       struct net_device *dev;
+
+       rtnl_lock();
+       for_each_netdev(&init_net, dev)
+               lttng_enumerate_device(session, dev);
+       rtnl_unlock();
+
+       return 0;
+}
+#else
 static
 int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session)
 {
@@ -347,6 +412,7 @@ int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session)
 
        return 0;
 }
+#endif /* (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(6,9,0)) */
 #else /* CONFIG_INET */
 static inline
 int lttng_enumerate_network_ip_interface(struct lttng_kernel_session *session)
@@ -378,7 +444,7 @@ int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
         * the lock is taken, but we are not aware whether this is
         * guaranteed or not, so play safe.
         */
-       if (fd < fdt->max_fds && lttng_close_on_exec(fd, fdt))
+       if (fd < fdt->max_fds && close_on_exec(fd, fdt))
                flags |= O_CLOEXEC;
        if (IS_ERR(s)) {
                struct dentry *dentry = file->f_path.dentry;
@@ -405,7 +471,7 @@ void lttng_enumerate_files(struct lttng_kernel_session *session,
 {
        struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .files = files, };
 
-       lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
+       iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
 }
 
 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
@@ -454,7 +520,7 @@ void lttng_enumerate_task_vm_maps(struct lttng_kernel_session *session,
                down_read(&mm->mmap_sem);
                while (map) {
                        if (map->vm_file)
-                               ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
+                               ino = map->vm_file->f_path.dentry->d_inode->i_ino;
                        else
                                ino = 0;
                        trace_lttng_statedump_vm_map(session, p, map, ino);
@@ -478,8 +544,6 @@ int lttng_enumerate_vm_maps(struct lttng_kernel_session *session)
 }
 #endif
 
-#ifdef CONFIG_LTTNG_HAS_LIST_IRQ
-
 static
 int lttng_list_interrupts(struct lttng_kernel_session *session)
 {
@@ -506,13 +570,6 @@ int lttng_list_interrupts(struct lttng_kernel_session *session)
        return 0;
 #undef irq_to_desc
 }
-#else
-static inline
-int lttng_list_interrupts(struct lttng_kernel_session *session)
-{
-       return 0;
-}
-#endif
 
 /*
  * Statedump the task's namespaces using the proc filesystem inode number as
@@ -557,24 +614,11 @@ void lttng_statedump_process_ns(struct lttng_kernel_session *session,
                 * paranoid behavior of
                 * trace_lttng_statedump_process_user_ns().
                 */
-               user_ns = user_ns ? user_ns->lttng_user_ns_parent : NULL;
+               user_ns = user_ns ? user_ns->parent : NULL;
        } while (user_ns);
 
-       /*
-        * Back and forth on locking strategy within Linux upstream for nsproxy.
-        * See Linux upstream commit 728dba3a39c66b3d8ac889ddbe38b5b1c264aec3
-        * "namespaces: Use task_lock and not rcu to protect nsproxy"
-        * for details.
-        */
-#if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \
-               LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
-               LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
-               LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
        proxy = p->nsproxy;
-#else
-       rcu_read_lock();
-       proxy = task_nsproxy(p);
-#endif
+
        if (proxy) {
 #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,6,0))
                trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns);
@@ -585,18 +629,11 @@ void lttng_statedump_process_ns(struct lttng_kernel_session *session,
 #endif
                trace_lttng_statedump_process_net_ns(session, p, proxy->net_ns);
                trace_lttng_statedump_process_uts_ns(session, p, proxy->uts_ns);
-#if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,6,0))
+#if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,6,0) || \
+       LTTNG_RHEL_KERNEL_RANGE(4,18,0,305,0,0, 4,19,0,0,0,0))
                trace_lttng_statedump_process_time_ns(session, p, proxy->time_ns);
 #endif
        }
-#if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(3,17,0) || \
-               LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0) || \
-               LTTNG_UBUNTU_KERNEL_RANGE(3,16,1,11, 3,17,0,0) || \
-               LTTNG_RHEL_KERNEL_RANGE(3,10,0,229,13,0, 3,11,0,0,0,0))
-       /* (nothing) */
-#else
-       rcu_read_unlock();
-#endif
 }
 
 static
@@ -628,7 +665,7 @@ int lttng_enumerate_process_states(struct lttng_kernel_session *session)
                                status = LTTNG_ZOMBIE;
                        else if (p->exit_state == EXIT_DEAD)
                                status = LTTNG_DEAD;
-                       else if (p->state == TASK_RUNNING) {
+                       else if (lttng_task_is_running(p)) {
                                /* Is this a forked child that has not run yet? */
                                if (list_empty(&p->rt.run_list))
                                        status = LTTNG_WAIT_FORK;
@@ -639,7 +676,7 @@ int lttng_enumerate_process_states(struct lttng_kernel_session *session)
                                         * was really running at this time.
                                         */
                                        status = LTTNG_WAIT_CPU;
-                       } else if (p->state &
+                       } else if (lttng_get_task_state(p) &
                                (TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE)) {
                                /* Task is waiting for something to complete */
                                status = LTTNG_WAIT;
@@ -737,7 +774,7 @@ int do_lttng_statedump(struct lttng_kernel_session *session)
         * is to guarantee that each CPU has been in a state where is was in
         * syscall mode (i.e. not in a trap, an IRQ or a soft IRQ).
         */
-       get_online_cpus();
+       lttng_cpus_read_lock();
        atomic_set(&kernel_threads_to_run, num_online_cpus());
        for_each_online_cpu(cpu) {
                INIT_DELAYED_WORK(&cpu_work[cpu], lttng_statedump_work_func);
@@ -745,7 +782,7 @@ int do_lttng_statedump(struct lttng_kernel_session *session)
        }
        /* Wait for all threads to run */
        __wait_event(statedump_wq, (atomic_read(&kernel_threads_to_run) == 0));
-       put_online_cpus();
+       lttng_cpus_read_unlock();
        /* Our work is done */
        trace_lttng_statedump_end(session);
        return 0;
@@ -763,13 +800,6 @@ EXPORT_SYMBOL_GPL(lttng_statedump_start);
 static
 int __init lttng_statedump_init(void)
 {
-       /*
-        * Allow module to load even if the fixup cannot be done. This
-        * will allow seemless transition when the underlying issue fix
-        * is merged into the Linux kernel, and when tracepoint.c
-        * "tracepoint_module_notify" is turned into a static function.
-        */
-       (void) wrapper_lttng_fixup_sig(THIS_MODULE);
        return 0;
 }
 
This page took 0.030536 seconds and 4 git commands to generate.