]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/kern/kern_shutdown.c
xnu-4903.221.2.tar.gz
[apple/xnu.git] / bsd / kern / kern_shutdown.c
index 4b59526ed4471c1f7831c5048c2a61b41d3f849c..936fd61e28212a1dc46596999941ee4df3be1b2a 100644 (file)
 #include <sys/tty.h>
 #include <kern/task.h>
 #include <sys/quota.h>
-#include <ufs/ufs/inode.h>
-#if    NCPUS > 1
-#include <kern/processor.h>
-#include <kern/thread.h>
-#include <sys/lock.h>
-#endif /* NCPUS > 1 */
 #include <vm/vm_kern.h>
 #include <mach/vm_param.h>
 #include <sys/filedesc.h>
 #include <mach/host_priv.h>
 #include <mach/host_reboot.h>
 
-#include <bsm/audit_kernel.h>
+#include <security/audit/audit.h>
 
 #include <kern/sched_prim.h>           /* for thread_block() */
 #include <kern/host.h>                 /* for host_priv_self() */
 #include <mach/task.h>                 /* for task_suspend() */
 #include <sys/sysproto.h>              /* abused for sync() */
 #include <kern/clock.h>                        /* for delay_for_interval() */
+#include <libkern/OSAtomic.h>
+#include <IOKit/IOPlatformExpert.h>
+
+#include <sys/kdebug.h>
+
+uint32_t system_inshutdown = 0;
 
 /* XXX should be in a header file somewhere, but isn't */
-extern void md_prepare_for_shutdown(int, int, char *);
+extern void (*unmountroot_pre_hook)(void);
 
-int    waittime = -1;
-static int shutting_down = 0;
+unsigned int proc_shutdown_exitcount = 0;
 
+static int  sd_openlog(vfs_context_t);
+static int  sd_closelog(vfs_context_t);
+static void sd_log(vfs_context_t, const char *, ...);
 static void proc_shutdown(void);
-int in_shutdown(void);
+static void zprint_panic_info(void);
+extern void halt_log_enter(const char * what, const void * pc, uint64_t time);
 
-extern void IOSystemShutdownNotification(void);
+#if DEVELOPMENT || DEBUG
+extern boolean_t kdp_has_polled_corefile(void);
+#endif /* DEVELOPMENT || DEBUG */
 
 struct sd_filterargs{
        int delayterm;
@@ -92,69 +97,139 @@ struct sd_filterargs{
 
 
 struct sd_iterargs {
-       int signo;      /* the signal to be posted */
-       int setsdstate;  /* shutdown state to be set */
+       int signo;              /* the signal to be posted */
+       int setsdstate;         /* shutdown state to be set */
+       int countproc;          /* count processes on action */
+       int activecount;        /* number of processes on which action was done */
 };
 
+static vnode_t sd_logvp = NULLVP;
+static off_t sd_log_offset = 0;
+
+
 static int sd_filt1(proc_t, void *);
 static int sd_filt2(proc_t, void *);
-static int  sd_callback1(proc_t p, void * arg);
-static int  sd_callback2(proc_t p, void * arg);
-static int  sd_callback3(proc_t p, void * arg);
+static int sd_callback1(proc_t p, void * arg);
+static int sd_callback2(proc_t p, void * arg);
+static int sd_callback3(proc_t p, void * arg);
 
-void
-boot(int paniced, int howto, char *command)
+extern boolean_t panic_include_zprint;
+extern mach_memory_info_t *panic_kext_memory_info;
+extern vm_size_t panic_kext_memory_size;
+
+static void
+zprint_panic_info(void)
 {
-       struct proc *p = current_proc();        /* XXX */
-       int hostboot_option=0;
-       int funnel_state;
+       unsigned int  num_sites;
+       kern_return_t kr;
+
+       panic_include_zprint = TRUE;
+       panic_kext_memory_info = NULL;
+       panic_kext_memory_size = 0;
+
+       num_sites = vm_page_diagnose_estimate();
+       panic_kext_memory_size = num_sites * sizeof(panic_kext_memory_info[0]);
+
+       kr = kmem_alloc(kernel_map, (vm_offset_t *)&panic_kext_memory_info, round_page(panic_kext_memory_size), VM_KERN_MEMORY_OSFMK);
+       if (kr != KERN_SUCCESS) {
+               panic_kext_memory_info = NULL;
+               return;
+       }
+
+       vm_page_diagnose(panic_kext_memory_info, num_sites, 0);
+}
+
+int
+get_system_inshutdown()
+{
+       return (system_inshutdown);
+}
+
+static void
+panic_kernel(int howto, char *message)
+{
+       if ((howto & RB_PANIC_ZPRINT) == RB_PANIC_ZPRINT) {
+               zprint_panic_info();
+       }
+       panic("userspace panic: %s", message);
+}
 
-       funnel_state = thread_funnel_set(kernel_flock, TRUE);
+int
+reboot_kernel(int howto, char *message)
+{
+       int hostboot_option=0;
+       uint64_t startTime;
 
-       /*
-       * Temporary hack to notify the power management root domain
-       * that the system will shut down.
-       */
-       IOSystemShutdownNotification();
+       if ((howto & (RB_PANIC | RB_QUICK)) == (RB_PANIC | RB_QUICK)) {
+               panic_kernel(howto, message);
+       }
 
-       shutting_down = 1;
-           
-       md_prepare_for_shutdown(paniced, howto, command);
+       if (!OSCompareAndSwap(0, 1, &system_inshutdown)) {
+               if ( (howto&RB_QUICK) == RB_QUICK)
+                       goto force_reboot;
+               return (EBUSY);
+       }
+       /*
+        * Notify the power management root domain that the system will shut down.
+        */
+       IOSystemShutdownNotification(kIOSystemShutdownNotificationStageProcessExit);
 
-       if ((howto&RB_NOSYNC)==0 && waittime < 0) {
+       if ((howto&RB_QUICK)==RB_QUICK) {
+               printf("Quick reboot...\n");
+               if ((howto&RB_NOSYNC)==0) {
+                       sync((proc_t)NULL, (void *)NULL, (int *)NULL);
+               }
+       }
+       else if ((howto&RB_NOSYNC)==0) {
                int iter, nbusy;
 
-               waittime = 0;
-               
                printf("syncing disks... ");
 
                /*
                 * Release vnodes held by texts before sync.
                 */
 
-               /* handle live procs (deallocate their root and current directories). */                
+               /* handle live procs (deallocate their root and current directories), suspend initproc */
+
+               startTime = mach_absolute_time();
                proc_shutdown();
+               halt_log_enter("proc_shutdown", 0, mach_absolute_time() - startTime);
 
-#if AUDIT
-               audit_shutdown();
+#if CONFIG_AUDIT
+               startTime = mach_absolute_time();
+               audit_shutdown();
+               halt_log_enter("audit_shutdown", 0, mach_absolute_time() - startTime);
 #endif
 
-               sync(p, (void *)NULL, (int *)NULL);
+               if (unmountroot_pre_hook != NULL)
+                       unmountroot_pre_hook();
 
-               /*
-                * Now that all processes have been terminated and system is
-                * sync'ed up, suspend init
-                */
+               startTime = mach_absolute_time();
+               sync((proc_t)NULL, (void *)NULL, (int *)NULL);
+
+               if (kdebug_enable) {
+                       startTime = mach_absolute_time();
+                       kdbg_dump_trace_to_file("/var/log/shutdown/shutdown.trace");
+                       halt_log_enter("shutdown.trace", 0, mach_absolute_time() - startTime);
+               }
 
-               if (initproc && p != initproc)
-                       task_suspend(initproc->task);
+               IOSystemShutdownNotification(kIOSystemShutdownNotificationStageRootUnmount);
 
                /*
                 * Unmount filesystems
                 */
-               vfs_unmountall();
+
+#if DEVELOPMENT || DEBUG
+               if (!(howto & RB_PANIC) || !kdp_has_polled_corefile())
+#endif /* DEVELOPMENT || DEBUG */
+               {
+                       startTime = mach_absolute_time();
+                       vfs_unmountall();
+                       halt_log_enter("vfs_unmountall", 0, mach_absolute_time() - startTime);
+               }
 
                /* Wait for the buffer cache to clean remaining dirty buffers */
+               startTime = mach_absolute_time();
                for (iter = 0; iter < 100; iter++) {
                        nbusy = count_busy_buffers();
                        if (nbusy == 0)
@@ -166,31 +241,100 @@ boot(int paniced, int howto, char *command)
                        printf("giving up\n");
                else
                        printf("done\n");
+               halt_log_enter("bufferclean", 0, mach_absolute_time() - startTime);
        }
-
 #if NETWORKING
        /*
         * Can't just use an splnet() here to disable the network
         * because that will lock out softints which the disk
         * drivers depend on to finish DMAs.
         */
+       startTime = mach_absolute_time();
        if_down_all();
+       halt_log_enter("if_down_all", 0, mach_absolute_time() - startTime);
 #endif /* NETWORKING */
 
+force_reboot:
+
+       if (howto & RB_PANIC) {
+               panic_kernel(howto, message);
+       }
+
        if (howto & RB_POWERDOWN)
                hostboot_option = HOST_REBOOT_HALT;
        if (howto & RB_HALT)
                hostboot_option = HOST_REBOOT_HALT;
-       if (paniced == RB_PANIC)
-               hostboot_option = HOST_REBOOT_HALT;
 
-    if (howto & RB_UPSDELAY) {
-        hostboot_option = HOST_REBOOT_UPSDELAY;
-    }
+       if (howto & RB_UPSDELAY) {
+               hostboot_option = HOST_REBOOT_UPSDELAY;
+       }
 
        host_reboot(host_priv_self(), hostboot_option);
+       /*
+        * should not be reached
+        */
+       return (0);
+}
+
+static int
+sd_openlog(vfs_context_t ctx)
+{
+       int error = 0;
+       struct timeval tv;
+       
+       /* Open shutdown log */
+       if ((error = vnode_open(PROC_SHUTDOWN_LOG, (O_CREAT | FWRITE | O_NOFOLLOW), 0644, 0, &sd_logvp, ctx))) {
+               printf("Failed to open %s: error %d\n", PROC_SHUTDOWN_LOG, error);
+               sd_logvp = NULLVP;
+               return error;
+       }
+
+       vnode_setsize(sd_logvp, (off_t)0, 0, ctx);
+
+       /* Write a little header */
+       microtime(&tv);
+       sd_log(ctx, "Process shutdown log.  Current time is %lu (in seconds).\n\n", tv.tv_sec);
+
+       return 0;
+}
+
+static int
+sd_closelog(vfs_context_t ctx)
+{
+       int error = 0;
+       if (sd_logvp != NULLVP) {
+               VNOP_FSYNC(sd_logvp, MNT_WAIT, ctx);
+               error = vnode_close(sd_logvp, FWRITE, ctx);
+       }
+
+       return error;
+}
+
+static void
+sd_log(vfs_context_t ctx, const char *fmt, ...) 
+{
+       int resid, log_error, len;
+       char logbuf[100];
+       va_list arglist;
+
+       /* If the log isn't open yet, open it */
+       if (sd_logvp == NULLVP) {
+               if (sd_openlog(ctx) != 0) {
+                       /* Couldn't open, we fail out */
+                       return;
+               }
+       }
+
+       va_start(arglist, fmt);
+       len = vsnprintf(logbuf, sizeof(logbuf), fmt, arglist);
+       log_error = vn_rdwr(UIO_WRITE, sd_logvp, (caddr_t)logbuf, len, sd_log_offset,
+                       UIO_SYSSPACE, IO_UNIT | IO_NOAUTH, vfs_context_ucred(ctx), &resid, vfs_context_proc(ctx));
+       if (log_error == EIO || log_error == 0) {
+               sd_log_offset += (len - resid);
+       }
+
+       va_end(arglist);
 
-       thread_funnel_set(kernel_flock, FALSE);
 }
 
 static int
@@ -213,22 +357,33 @@ sd_filt1(proc_t p, void * args)
 }
 
 
-static int  
+static int
 sd_callback1(proc_t p, void * args)
 {
        struct sd_iterargs * sd = (struct sd_iterargs *)args;
        int signo = sd->signo;
        int setsdstate = sd->setsdstate;
+       int countproc = sd->countproc;
 
        proc_lock(p);
        p->p_shutdownstate = setsdstate;
        if (p->p_stat != SZOMB) {
                proc_unlock(p);
+               if (countproc != 0) {
+                       proc_list_lock();
+                       p->p_listflag |= P_LIST_EXITCOUNT;
+                       proc_shutdown_exitcount++;
+                       proc_list_unlock();
+               }
+
                psignal(p, signo);
-       } else
+               if (countproc !=  0)
+                       sd->activecount++;
+       } else {
                proc_unlock(p);
-       return(PROC_RETURNED);
+       }
 
+       return PROC_RETURNED;
 }
 
 static int
@@ -249,29 +404,40 @@ sd_filt2(proc_t p, void * args)
                 return(1);
 }
 
-static int  
+static int
 sd_callback2(proc_t p, void * args)
 {
        struct sd_iterargs * sd = (struct sd_iterargs *)args;
        int signo = sd->signo;
        int setsdstate = sd->setsdstate;
+       int countproc = sd->countproc;
 
        proc_lock(p);
        p->p_shutdownstate = setsdstate;
        if (p->p_stat != SZOMB) {
                proc_unlock(p);
+               if (countproc !=  0) {
+                       proc_list_lock();
+                       p->p_listflag |= P_LIST_EXITCOUNT;
+                       proc_shutdown_exitcount++;
+                       proc_list_unlock();
+               }
                psignal(p, signo);
-       } else
+               if (countproc !=  0)
+                       sd->activecount++;
+       } else {
                proc_unlock(p);
+       }
 
-       return(PROC_RETURNED);
-
+       return PROC_RETURNED;
 }
 
-static int  
+static int
 sd_callback3(proc_t p, void * args)
 {
        struct sd_iterargs * sd = (struct sd_iterargs *)args;
+       vfs_context_t ctx = vfs_context_current();
+
        int setsdstate = sd->setsdstate;
 
        proc_lock(p);
@@ -291,13 +457,20 @@ sd_callback3(proc_t p, void * args)
                } else {
                        p->exit_thread = current_thread();
                        printf(".");
+
+                       sd_log(ctx, "%s[%d] had to be forced closed with exit1().\n", p->p_comm, p->p_pid);
+
                        proc_unlock(p);
+                       KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_FRCEXIT) | DBG_FUNC_NONE,
+                                             p->p_pid, 0, 1, 0, 0);
+                       sd->activecount++;
                        exit1(p, 1, (int *)NULL);
                }
-       } else
+       } else {
                proc_unlock(p);
+       }
 
-       return(PROC_RETURNED);
+       return PROC_RETURNED;
 }
 
 
@@ -309,27 +482,29 @@ sd_callback3(proc_t p, void * args)
  *
  * POSIX modifications:
  *
- *     For POSIX fcntl() file locking call vno_lockrelease() on 
+ *     For POSIX fcntl() file locking call vno_lockrelease() on
  *     the file to release all of its record locks, if any.
  */
 
 static void
 proc_shutdown(void)
 {
-       struct proc     *p, *self;
-       int             i, TERM_catch;
+       vfs_context_t ctx = vfs_context_current();
+       struct proc *p, *self;
        int delayterm = 0;
        struct sd_filterargs sfargs;
        struct sd_iterargs sdargs;
+       int error = 0;
+       struct timespec ts;
 
        /*
         *      Kill as many procs as we can.  (Except ourself...)
         */
        self = (struct proc *)current_proc();
-       
+
        /*
         * Signal the init with SIGTERM so that he does not launch
-        * new processes 
+        * new processes
         */
        p = proc_find(1);
        if (p && p != self) {
@@ -347,55 +522,52 @@ sigterm_loop:
        sfargs.shutdownstate = 0;
        sdargs.signo = SIGTERM;
        sdargs.setsdstate = 1;
+       sdargs.countproc = 1;
+       sdargs.activecount = 0;
 
+       error = 0;
        /* post a SIGTERM to all that catch SIGTERM and not marked for delay */
        proc_rebootscan(sd_callback1, (void *)&sdargs, sd_filt1, (void *)&sfargs);
 
-       /*
-        * now wait for up to 30 seconds to allow those procs catching SIGTERM
-        * to digest it
-        * as soon as these procs have exited, we'll continue on to the next step
-        */
-       for (i = 0; i < 300; i++) {
-               /*
-                * sleep for a tenth of a second
-                * and then check to see if the tasks that were sent a
-                * SIGTERM have exited
-                */
-               delay_for_interval(100, 1000 * 1000);
-               TERM_catch = 0;
-
-
+       if (sdargs.activecount != 0 && proc_shutdown_exitcount!= 0) {
                proc_list_lock();
-
-               for (p = allproc.lh_first; p; p = p->p_list.le_next) {
-                       if (p->p_shutdownstate == 1) {
-                               TERM_catch++;
+               if (proc_shutdown_exitcount != 0) {
+                       /*
+                        * now wait for up to 3 seconds to allow those procs catching SIGTERM
+                        * to digest it
+                        * as soon as these procs have exited, we'll continue on to the next step
+                        */
+                       ts.tv_sec = 3;
+                       ts.tv_nsec = 0;
+                       error = msleep(&proc_shutdown_exitcount, proc_list_mlock, PWAIT, "shutdownwait", &ts);
+                       if (error != 0) {
+                               for (p = allproc.lh_first; p; p = p->p_list.le_next) {
+                                       if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
+                                               p->p_listflag &= ~P_LIST_EXITCOUNT;
+                               }
+                               for (p = zombproc.lh_first; p; p = p->p_list.le_next) {
+                                       if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
+                                               p->p_listflag &= ~P_LIST_EXITCOUNT;
+                               }
                        }
                }
-
                proc_list_unlock();
-
-               if (TERM_catch == 0)
-                       break;
        }
-       if (TERM_catch) {
+       if (error == ETIMEDOUT) {
                /*
                 * log the names of the unresponsive tasks
                 */
 
-
                proc_list_lock();
 
-               for (p = allproc.lh_first; p; p = p->p_list.le_next) {
+               for (p = allproc.lh_first; p; p = p->p_list.le_next) {
                        if (p->p_shutdownstate == 1) {
-                                 printf("%s[%d]: didn't act on SIGTERM\n", p->p_comm, p->p_pid);
+                               printf("%s[%d]: didn't act on SIGTERM\n", p->p_comm, p->p_pid);
+                               sd_log(ctx, "%s[%d]: didn't act on SIGTERM\n", p->p_comm, p->p_pid);
                        }
                }
 
                proc_list_unlock();
-
-               delay_for_interval(1000 * 5, 1000 * 1000);
        }
 
        /*
@@ -405,31 +577,55 @@ sigterm_loop:
        sfargs.shutdownstate = 2;
        sdargs.signo = SIGKILL;
        sdargs.setsdstate = 2;
+       sdargs.countproc = 1;
+       sdargs.activecount = 0;
 
-       /* post a SIGTERM to all that catch SIGTERM and not marked for delay */
+       /* post a SIGKILL to all that catch SIGTERM and not marked for delay */
        proc_rebootscan(sd_callback2, (void *)&sdargs, sd_filt2, (void *)&sfargs);
 
-       /*
-        * wait for up to 60 seconds to allow these procs to exit normally
-        *
-        * History:     The delay interval was changed from 100 to 200
-        *              for NFS requests in particular.
-        */
-       for (i = 0; i < 300; i++) {
-               delay_for_interval(200, 1000 * 1000);
+       error = 0;
+
+       if (sdargs.activecount != 0 && proc_shutdown_exitcount!= 0) {
+               proc_list_lock();
+               if (proc_shutdown_exitcount != 0) {
+                       /*
+                       * wait for up to 60 seconds to allow these procs to exit normally
+                       *
+                       * History:      The delay interval was changed from 100 to 200
+                       *               for NFS requests in particular.
+                       */
+                       ts.tv_sec = 10;
+                       ts.tv_nsec = 0;
+                       error = msleep(&proc_shutdown_exitcount, proc_list_mlock, PWAIT, "shutdownwait", &ts);
+                       if (error != 0) {
+                               for (p = allproc.lh_first; p; p = p->p_list.le_next) {
+                                       if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
+                                               p->p_listflag &= ~P_LIST_EXITCOUNT;
+                               }
+                               for (p = zombproc.lh_first; p; p = p->p_list.le_next) {
+                                       if ((p->p_listflag & P_LIST_EXITCOUNT) == P_LIST_EXITCOUNT)
+                                               p->p_listflag &= ~P_LIST_EXITCOUNT;
+                               }
+                       }
+               }
+               proc_list_unlock();
+       }
 
+       if (error == ETIMEDOUT) {
+               /*
+                * log the names of the unresponsive tasks
+                */
 
                proc_list_lock();
 
-               for (p = allproc.lh_first; p; p = p->p_list.le_next) {
-                               if (p->p_shutdownstate == 2)
-                               break;
+               for (p = allproc.lh_first; p; p = p->p_list.le_next) {
+                       if (p->p_shutdownstate == 2) {
+                               printf("%s[%d]: didn't act on SIGKILL\n", p->p_comm, p->p_pid);
+                               sd_log(ctx, "%s[%d]: didn't act on SIGKILL\n", p->p_comm, p->p_pid);
+                       }
                }
 
                proc_list_unlock();
-
-               if (!p)
-                       break;
        }
 
        /*
@@ -439,6 +635,10 @@ sigterm_loop:
        sfargs.shutdownstate = 3;
        sdargs.signo = 0;
        sdargs.setsdstate = 3;
+       sdargs.countproc = 0;
+       sdargs.activecount = 0;
+
+
 
        /* post a SIGTERM to all that catch SIGTERM and not marked for delay */
        proc_rebootscan(sd_callback3, (void *)&sdargs, sd_filt2, (void *)&sfargs);
@@ -449,16 +649,16 @@ sigterm_loop:
                delayterm = 1;
                goto  sigterm_loop;
        }
+
+       sd_closelog(ctx);
+
+       /*
+        * Now that all other processes have been terminated, suspend init
+        */
+       task_suspend_internal(initproc->task);
+
        /* drop the ref on initproc */
        proc_rele(initproc);
        printf("continuing\n");
 }
 
-/*
- * Check whether the system has begun its shutdown sequence. 
- */
-int
-in_shutdown(void)
-{
-       return shutting_down;
-}