Move headers under include/
[lttng-modules.git] / lttng-statedump-impl.c
index 6d91a2c1d28dea1374a427b465e25d0567e294b8..3d010ca62679fecaa018472a8e211f54a096ff88 100644 (file)
@@ -1,4 +1,4 @@
-/* SPDX-License-Identifier: (GPL-2.0 or LGPL-2.1)
+/* SPDX-License-Identifier: (GPL-2.0-only or LGPL-2.1-only)
  *
  * lttng-statedump.c
  *
 #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/irq.h>
-#include <wrapper/tracepoint.h>
-#include <wrapper/genhd.h>
-#include <wrapper/file.h>
-#include <wrapper/time.h>
-
-#ifdef CONFIG_LTTNG_HAS_LIST_IRQ
+#include <linux/fdtable.h>
 #include <linux/irq.h>
-#endif
+#include <linux/genhd.h>
+
+#include <lttng/lttng-events.h>
+#include <lttng/lttng-tracer.h>
 
 /* Define the tracepoints, but do not build the probes */
 #define CREATE_TRACE_POINTS
@@ -62,6 +52,15 @@ 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);
+DEFINE_TRACE(lttng_statedump_process_cgroup_ns);
+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);
@@ -70,7 +69,6 @@ DEFINE_TRACE(lttng_statedump_cpu_topology);
 struct lttng_fd_ctx {
        char *page;
        struct lttng_session *session;
-       struct task_struct *p;
        struct files_struct *files;
 };
 
@@ -114,19 +112,11 @@ enum lttng_process_status {
 static
 int lttng_enumerate_block_devices(struct lttng_session *session)
 {
-       struct class *ptr_block_class;
-       struct device_type *ptr_disk_type;
        struct class_dev_iter iter;
        struct device *dev;
 
-       ptr_block_class = wrapper_get_block_class();
-       if (!ptr_block_class)
-               return -ENOSYS;
-       ptr_disk_type = wrapper_get_disk_type();
-       if (!ptr_disk_type) {
-               return -ENOSYS;
-       }
-       class_dev_iter_init(&iter, ptr_block_class, NULL, ptr_disk_type);
+       class_dev_iter_init(&iter, gendisk_block_class(), NULL,
+                           gendisk_device_type());
        while ((dev = class_dev_iter_next(&iter))) {
                struct disk_part_iter piter;
                struct gendisk *disk = dev_to_disk(dev);
@@ -145,7 +135,7 @@ int lttng_enumerate_block_devices(struct lttng_session *session)
                        char name_buf[BDEVNAME_SIZE];
                        char *p;
 
-                       p = wrapper_disk_name(disk, part->partno, name_buf);
+                       p = gendisk_name(disk, part->partno, name_buf);
                        if (!p) {
                                disk_part_iter_exit(&piter);
                                class_dev_iter_exit(&iter);
@@ -228,58 +218,34 @@ 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;
 
                /* 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);
-}
-
-static
-int lttng_enumerate_file_descriptors(struct lttng_session *session)
-{
-       struct task_struct *p;
-       char *tmp;
-
-       tmp = (char *) __get_free_page(GFP_KERNEL);
-       if (!tmp)
-               return -ENOMEM;
-
-       /* 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;
+       iterate_fd(files, 0, lttng_dump_one_fd, &ctx);
 }
 
 #ifdef LTTNG_HAVE_STATEDUMP_CPU_TOPOLOGY
@@ -328,7 +294,7 @@ 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->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);
@@ -352,12 +318,6 @@ int lttng_enumerate_vm_maps(struct lttng_session *session)
 }
 #endif
 
-#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)
 {
@@ -365,7 +325,6 @@ int lttng_list_interrupts(struct lttng_session *session)
        unsigned long flags = 0;
        struct irq_desc *desc;
 
-#define irq_to_desc    wrapper_irq_to_desc
        /* needs irq_desc */
        for_each_irq_desc(irq, desc) {
                struct irqaction *action;
@@ -373,26 +332,22 @@ 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;
-#undef irq_to_desc
-}
-#else
-static inline
-int lttng_list_interrupts(struct lttng_session *session)
-{
-       return 0;
 }
-#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
@@ -403,23 +358,69 @@ void lttng_statedump_process_ns(struct lttng_session *session,
                enum lttng_execution_submode submode,
                enum lttng_process_status status)
 {
+       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_state(session,
-                       p, type, mode, submode, status, pid_ns);
-               pid_ns = pid_ns->parent;
+               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->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.
+        */
+       proxy = p->nsproxy;
+       if (proxy) {
+               trace_lttng_statedump_process_cgroup_ns(session, p, proxy->cgroup_ns);
+               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);
+       }
 }
 
 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 =
@@ -428,6 +429,7 @@ 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)
@@ -462,13 +464,31 @@ 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;
 }
 
@@ -487,9 +507,6 @@ 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;
        /*
@@ -554,13 +571,6 @@ 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 0.026922 seconds and 4 git commands to generate.