Cleanup: Move patches.i to include/generated/
[lttng-modules.git] / lttng-statedump-impl.c
index bab5d9ff4c2cf1dcdef90e03e777ac5f96a6ebe2..1a2a12ba7f650ceb9a633b2cd7f79729aeadec32 100644 (file)
@@ -1,4 +1,5 @@
-/*
+/* SPDX-License-Identifier: (GPL-2.0-only or LGPL-2.1-only)
+ *
  * lttng-statedump.c
  *
  * Linux Trace Toolkit Next Generation Kernel State Dump
@@ -6,20 +7,6 @@
  * Copyright 2005 Jean-Hugues Deschenes <jean-hugues.deschenes@polymtl.ca>
  * Copyright 2006-2012 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
  *
- * This library is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; only
- * version 2.1 of the License.
- *
- * This library is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- *
  * Changes:
  *     Eric Clement:                   Add listing of network IP interface
  *     2006, 2007 Mathieu Desnoyers    Fix kernel threads
 #include <linux/inetdevice.h>
 #include <linux/sched.h>
 #include <linux/mm.h>
-#include <linux/fdtable.h>
 #include <linux/swap.h>
 #include <linux/wait.h>
 #include <linux/mutex.h>
 #include <linux/device.h>
 
-#include "lttng-events.h"
-#include "lttng-tracer.h"
-#include "wrapper/irqdesc.h"
-#include "wrapper/spinlock.h"
-#include "wrapper/fdtable.h"
-#include "wrapper/nsproxy.h"
-#include "wrapper/irq.h"
-#include "wrapper/tracepoint.h"
-#include "wrapper/genhd.h"
+#include <lttng/events.h>
+#include <lttng/tracer.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/fdtable.h>
 
 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 #include <linux/irq.h>
 
 /* Define the tracepoints, but do not build the probes */
 #define CREATE_TRACE_POINTS
-#define TRACE_INCLUDE_PATH ../instrumentation/events/lttng-module
+#define TRACE_INCLUDE_PATH instrumentation/events
 #define TRACE_INCLUDE_FILE lttng-statedump
 #define LTTNG_INSTRUMENTATION
-#include "instrumentation/events/lttng-module/lttng-statedump.h"
+#include <instrumentation/events/lttng-statedump.h>
 
 DEFINE_TRACE(lttng_statedump_block_device);
 DEFINE_TRACE(lttng_statedump_end);
@@ -74,12 +61,25 @@ DEFINE_TRACE(lttng_statedump_interrupt);
 DEFINE_TRACE(lttng_statedump_file_descriptor);
 DEFINE_TRACE(lttng_statedump_start);
 DEFINE_TRACE(lttng_statedump_process_state);
+DEFINE_TRACE(lttng_statedump_process_pid_ns);
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
+DEFINE_TRACE(lttng_statedump_process_cgroup_ns);
+#endif
+DEFINE_TRACE(lttng_statedump_process_ipc_ns);
+#ifndef LTTNG_MNT_NS_MISSING_HEADER
+DEFINE_TRACE(lttng_statedump_process_mnt_ns);
+#endif
+DEFINE_TRACE(lttng_statedump_process_net_ns);
+DEFINE_TRACE(lttng_statedump_process_user_ns);
+DEFINE_TRACE(lttng_statedump_process_uts_ns);
 DEFINE_TRACE(lttng_statedump_network_interface);
+#ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
+DEFINE_TRACE(lttng_statedump_cpu_topology);
+#endif
 
 struct lttng_fd_ctx {
        char *page;
        struct lttng_session *session;
-       struct task_struct *p;
        struct files_struct *files;
 };
 
@@ -214,18 +214,6 @@ int lttng_enumerate_network_ip_interface(struct lttng_session *session)
 }
 #endif /* CONFIG_INET */
 
-#ifdef FD_ISSET        /* For old kernels lacking close_on_exec() */
-static inline bool lttng_close_on_exec(int fd, const struct fdtable *fdt)
-{
-       return FD_ISSET(fd, fdt->close_on_exec);
-}
-#else
-static inline bool lttng_close_on_exec(int fd, const struct fdtable *fdt)
-{
-       return close_on_exec(fd, fdt);
-}
-#endif
-
 static
 int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 {
@@ -256,52 +244,50 @@ int lttng_dump_one_fd(const void *p, struct file *file, unsigned int fd)
 
                /* Make sure we give at least some info */
                spin_lock(&dentry->d_lock);
-               trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd,
-                       dentry->d_name.name, flags, file->f_mode);
+               trace_lttng_statedump_file_descriptor(ctx->session,
+                       ctx->files, fd, dentry->d_name.name, flags,
+                       file->f_mode);
                spin_unlock(&dentry->d_lock);
                goto end;
        }
-       trace_lttng_statedump_file_descriptor(ctx->session, ctx->p, fd, s,
-               flags, file->f_mode);
+       trace_lttng_statedump_file_descriptor(ctx->session,
+               ctx->files, fd, s, flags, file->f_mode);
 end:
        return 0;
 }
 
+/* Called with task lock held. */
 static
-void lttng_enumerate_task_fd(struct lttng_session *session,
-               struct task_struct *p, char *tmp)
+void lttng_enumerate_files(struct lttng_session *session,
+               struct files_struct *files,
+               char *tmp)
 {
-       struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .p = p };
-       struct files_struct *files;
+       struct lttng_fd_ctx ctx = { .page = tmp, .session = session, .files = files, };
 
-       task_lock(p);
-       files = p->files;
-       if (!files)
-               goto end;
-       ctx.files = files;
        lttng_iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
-end:
-       task_unlock(p);
 }
 
+#ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
 static
-int lttng_enumerate_file_descriptors(struct lttng_session *session)
+int lttng_enumerate_cpu_topology(struct lttng_session *session)
 {
-       struct task_struct *p;
-       char *tmp;
+       int cpu;
+       const cpumask_t *cpumask = cpu_possible_mask;
 
-       tmp = (char *) __get_free_page(GFP_KERNEL);
-       if (!tmp)
-               return -ENOMEM;
+       for (cpu = cpumask_first(cpumask); cpu < nr_cpu_ids;
+                       cpu = cpumask_next(cpu, cpumask)) {
+               trace_lttng_statedump_cpu_topology(session, &cpu_data(cpu));
+       }
 
-       /* Enumerate active file descriptors */
-       rcu_read_lock();
-       for_each_process(p)
-               lttng_enumerate_task_fd(session, p, tmp);
-       rcu_read_unlock();
-       free_page((unsigned long) tmp);
        return 0;
 }
+#else
+static
+int lttng_enumerate_cpu_topology(struct lttng_session *session)
+{
+       return 0;
+}
+#endif
 
 #if 0
 /*
@@ -327,7 +313,7 @@ void lttng_enumerate_task_vm_maps(struct lttng_session *session,
                down_read(&mm->mmap_sem);
                while (map) {
                        if (map->vm_file)
-                               ino = map->vm_file->f_dentry->d_inode->i_ino;
+                               ino = map->vm_file->lttng_f_dentry->d_inode->i_ino;
                        else
                                ino = 0;
                        trace_lttng_statedump_vm_map(session, p, map, ino);
@@ -353,10 +339,6 @@ int lttng_enumerate_vm_maps(struct lttng_session *session)
 
 #ifdef CONFIG_LTTNG_HAS_LIST_IRQ
 
-#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39))
-#define irq_desc_get_chip(desc) get_irq_desc_chip(desc)
-#endif
-
 static
 int lttng_list_interrupts(struct lttng_session *session)
 {
@@ -372,12 +354,12 @@ int lttng_list_interrupts(struct lttng_session *session)
                        irq_desc_get_chip(desc)->name ? : "unnamed_irq_chip";
 
                local_irq_save(flags);
-               wrapper_desc_spin_lock(&desc->lock);
+               raw_spin_lock(&desc->lock);
                for (action = desc->action; action; action = action->next) {
                        trace_lttng_statedump_interrupt(session,
                                irq, irq_chip_name, action);
                }
-               wrapper_desc_spin_unlock(&desc->lock);
+               raw_spin_unlock(&desc->lock);
                local_irq_restore(flags);
        }
        return 0;
@@ -392,6 +374,10 @@ int lttng_list_interrupts(struct lttng_session *session)
 #endif
 
 /*
+ * Statedump the task's namespaces using the proc filesystem inode number as
+ * the unique identifier. The user and pid ns are nested and will be dumped
+ * recursively.
+ *
  * Called with task lock held.
  */
 static
@@ -404,6 +390,34 @@ void lttng_statedump_process_ns(struct lttng_session *session,
 {
        struct nsproxy *proxy;
        struct pid_namespace *pid_ns;
+       struct user_namespace *user_ns;
+
+       /*
+        * The pid and user namespaces are special, they are nested and
+        * accessed with specific functions instead of the nsproxy struct
+        * like the other namespaces.
+        */
+       pid_ns = task_active_pid_ns(p);
+       do {
+               trace_lttng_statedump_process_pid_ns(session, p, pid_ns);
+               pid_ns = pid_ns ? pid_ns->parent : NULL;
+       } while (pid_ns);
+
+
+       user_ns = task_cred_xxx(p, user_ns);
+       do {
+               trace_lttng_statedump_process_user_ns(session, p, user_ns);
+               /*
+                * trace_lttng_statedump_process_user_ns() internally
+                * checks whether user_ns is NULL. While this does not
+                * appear to be a possible return value for
+                * task_cred_xxx(), err on the safe side and check
+                * for NULL here as well to be consistent with the
+                * paranoid behavior of
+                * trace_lttng_statedump_process_user_ns().
+                */
+               user_ns = user_ns ? user_ns->lttng_user_ns_parent : NULL;
+       } while (user_ns);
 
        /*
         * Back and forth on locking strategy within Linux upstream for nsproxy.
@@ -412,25 +426,29 @@ void lttng_statedump_process_ns(struct lttng_session *session,
         * for details.
         */
 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
-               LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0))
+               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) {
-               pid_ns = lttng_get_proxy_pid_ns(proxy);
-               do {
-                       trace_lttng_statedump_process_state(session,
-                               p, type, mode, submode, status, pid_ns);
-                       pid_ns = pid_ns->parent;
-               } while (pid_ns);
-       } else {
-               trace_lttng_statedump_process_state(session,
-                       p, type, mode, submode, status, NULL);
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,6,0))
+               trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns);
+#endif
+               trace_lttng_statedump_process_ipc_ns(session, p, proxy->ipc_ns);
+#ifndef LTTNG_MNT_NS_MISSING_HEADER
+               trace_lttng_statedump_process_mnt_ns(session, p, proxy->mnt_ns);
+#endif
+               trace_lttng_statedump_process_net_ns(session, p, proxy->net_ns);
+               trace_lttng_statedump_process_uts_ns(session, p, proxy->uts_ns);
        }
 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,17,0) || \
-               LTTNG_UBUNTU_KERNEL_RANGE(3,13,11,36, 3,14,0,0))
+               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();
@@ -441,9 +459,16 @@ static
 int lttng_enumerate_process_states(struct lttng_session *session)
 {
        struct task_struct *g, *p;
+       char *tmp;
+
+       tmp = (char *) __get_free_page(GFP_KERNEL);
+       if (!tmp)
+               return -ENOMEM;
 
        rcu_read_lock();
        for_each_process(g) {
+               struct files_struct *prev_files = NULL;
+
                p = g;
                do {
                        enum lttng_execution_mode mode =
@@ -452,6 +477,7 @@ int lttng_enumerate_process_states(struct lttng_session *session)
                                LTTNG_UNKNOWN;
                        enum lttng_process_status status;
                        enum lttng_thread_type type;
+                       struct files_struct *files;
 
                        task_lock(p);
                        if (p->exit_state == EXIT_ZOMBIE)
@@ -486,13 +512,31 @@ int lttng_enumerate_process_states(struct lttng_session *session)
                                type = LTTNG_USER_THREAD;
                        else
                                type = LTTNG_KERNEL_THREAD;
+                       files = p->files;
+
+                       trace_lttng_statedump_process_state(session,
+                               p, type, mode, submode, status, files);
                        lttng_statedump_process_ns(session,
                                p, type, mode, submode, status);
+                       /*
+                        * As an optimisation for the common case, do not
+                        * repeat information for the same files_struct in
+                        * two consecutive threads. This is the common case
+                        * for threads sharing the same fd table. RCU guarantees
+                        * that the same files_struct pointer is not re-used
+                        * throughout processes/threads iteration.
+                        */
+                       if (files && files != prev_files) {
+                               lttng_enumerate_files(session, files, tmp);
+                               prev_files = files;
+                       }
                        task_unlock(p);
                } while_each_thread(g, p);
        }
        rcu_read_unlock();
 
+       free_page((unsigned long) tmp);
+
        return 0;
 }
 
@@ -511,9 +555,6 @@ int do_lttng_statedump(struct lttng_session *session)
 
        trace_lttng_statedump_start(session);
        ret = lttng_enumerate_process_states(session);
-       if (ret)
-               return ret;
-       ret = lttng_enumerate_file_descriptors(session);
        if (ret)
                return ret;
        /*
@@ -530,12 +571,17 @@ int do_lttng_statedump(struct lttng_session *session)
                return ret;
        ret = lttng_enumerate_block_devices(session);
        switch (ret) {
+       case 0:
+               break;
        case -ENOSYS:
                printk(KERN_WARNING "LTTng: block device enumeration is not supported by kernel\n");
                break;
        default:
                return ret;
        }
+       ret = lttng_enumerate_cpu_topology(session);
+       if (ret)
+               return ret;
 
        /* TODO lttng_dump_idt_table(session); */
        /* TODO lttng_dump_softirq_vec(session); */
@@ -594,7 +640,7 @@ module_exit(lttng_statedump_exit);
 
 MODULE_LICENSE("GPL and additional rights");
 MODULE_AUTHOR("Jean-Hugues Deschenes");
-MODULE_DESCRIPTION("Linux Trace Toolkit Next Generation Statedump");
+MODULE_DESCRIPTION("LTTng statedump provider");
 MODULE_VERSION(__stringify(LTTNG_MODULES_MAJOR_VERSION) "."
        __stringify(LTTNG_MODULES_MINOR_VERSION) "."
        __stringify(LTTNG_MODULES_PATCHLEVEL_VERSION)
This page took 0.028024 seconds and 4 git commands to generate.