mm/pdflush.c

   1 /*
   2  * mm/pdflush.c - worker threads for writing back filesystem data
   3  *
   4  * Copyright (C) 2002, Linus Torvalds.
   5  *
   6  * 09Apr2002    Andrew Morton
   7  *              Initial version
   8  * 29Feb2004    kaos@sgi.com
   9  *              Move worker thread creation to kthread to avoid chewing
  10  *              up stack space with nested calls to kernel_thread.
  11  */
  12
  13 #include <linux/sched.h>
  14 #include <linux/list.h>
  15 #include <linux/signal.h>
  16 #include <linux/spinlock.h>
  17 #include <linux/gfp.h>
  18 #include <linux/init.h>
  19 #include <linux/module.h>
  20 #include <linux/fs.h>           /* Needed by writeback.h          */
  21 #include <linux/writeback.h>    /* Prototypes pdflush_operation() */
  22 #include <linux/kthread.h>
  23 #include <linux/cpuset.h>
  24 #include <linux/freezer.h>
  25
  26
  27 /*
  28  * Minimum and maximum number of pdflush instances
  29  */
  30 #define MIN_PDFLUSH_THREADS     2
  31 #define MAX_PDFLUSH_THREADS     8
  32
  33 static void start_one_pdflush_thread(void);
  34
  35
  36 /*
  37  * The pdflush threads are worker threads for writing back dirty data.
  38  * Ideally, we'd like one thread per active disk spindle.  But the disk
  39  * topology is very hard to divine at this level.   Instead, we take
  40  * care in various places to prevent more than one pdflush thread from
  41  * performing writeback against a single filesystem.  pdflush threads
  42  * have the PF_FLUSHER flag set in current->flags to aid in this.
  43  */
  44
  45 /*
  46  * All the pdflush threads.  Protected by pdflush_lock
  47  */
  48 static LIST_HEAD(pdflush_list);
  49 static DEFINE_SPINLOCK(pdflush_lock);
  50
  51 /*
  52  * The count of currently-running pdflush threads.  Protected
  53  * by pdflush_lock.
  54  *
  55  * Readable by sysctl, but not writable.  Published to userspace at
  56  * /proc/sys/vm/nr_pdflush_threads.
  57  */
  58 int nr_pdflush_threads = 0;
  59
  60 /*
  61  * The time at which the pdflush thread pool last went empty
  62  */
  63 static unsigned long last_empty_jifs;
  64
  65 /*
  66  * The pdflush thread.
  67  *
  68  * Thread pool management algorithm:
  69  *
  70  * - The minimum and maximum number of pdflush instances are bound
  71  *   by MIN_PDFLUSH_THREADS and MAX_PDFLUSH_THREADS.
  72  *
  73  * - If there have been no idle pdflush instances for 1 second, create
  74  *   a new one.
  75  *
  76  * - If the least-recently-went-to-sleep pdflush thread has been asleep
  77  *   for more than one second, terminate a thread.
  78  */
  79
  80 /*
  81  * A structure for passing work to a pdflush thread.  Also for passing
  82  * state information between pdflush threads.  Protected by pdflush_lock.
  83  */
  84 struct pdflush_work {
  85         struct task_struct *who;        /* The thread */
  86         void (*fn)(unsigned long);      /* A callback function */
  87         unsigned long arg0;             /* An argument to the callback */
  88         struct list_head list;          /* On pdflush_list, when idle */
  89         unsigned long when_i_went_to_sleep;
  90 };
  91
  92 static int __pdflush(struct pdflush_work *my_work)
  93 {
  94         current->flags |= PF_FLUSHER | PF_SWAPWRITE;
  95         set_freezable();
  96         my_work->fn = NULL;
  97         my_work->who = current;
  98         INIT_LIST_HEAD(&my_work->list);
  99
 100         spin_lock_irq(&pdflush_lock);
 101         nr_pdflush_threads++;
 102         for ( ; ; ) {
 103                 struct pdflush_work *pdf;
 104
 105                 set_current_state(TASK_INTERRUPTIBLE);
 106                 list_move(&my_work->list, &pdflush_list);
 107                 my_work->when_i_went_to_sleep = jiffies;
 108                 spin_unlock_irq(&pdflush_lock);
 109                 schedule();
 110                 try_to_freeze();
 111                 spin_lock_irq(&pdflush_lock);
 112                 if (!list_empty(&my_work->list)) {
 113                         /*
 114                          * Someone woke us up, but without removing our control
 115                          * structure from the global list.  swsusp will do this
 116                          * in try_to_freeze()->refrigerator().  Handle it.
 117                          */
 118                         my_work->fn = NULL;
 119                         continue;
 120                 }
 121                 if (my_work->fn == NULL) {
 122                         printk("pdflush: bogus wakeup\n");
 123                         continue;
 124                 }
 125                 spin_unlock_irq(&pdflush_lock);
 126
 127                 (*my_work->fn)(my_work->arg0);
 128
 129                 /*
 130                  * Thread creation: For how long have there been zero
 131                  * available threads?
 132                  */
 133                 if (time_after(jiffies, last_empty_jifs + 1 * HZ)) {
 134                         /* unlocked list_empty() test is OK here */
 135                         if (list_empty(&pdflush_list)) {
 136                                 /* unlocked test is OK here */
 137                                 if (nr_pdflush_threads < MAX_PDFLUSH_THREADS)
 138                                         start_one_pdflush_thread();
 139                         }
 140                 }
 141
 142                 spin_lock_irq(&pdflush_lock);
 143                 my_work->fn = NULL;
 144
 145                 /*
 146                  * Thread destruction: For how long has the sleepiest
 147                  * thread slept?
 148                  */
 149                 if (list_empty(&pdflush_list))
 150                         continue;
 151                 if (nr_pdflush_threads <= MIN_PDFLUSH_THREADS)
 152                         continue;
 153                 pdf = list_entry(pdflush_list.prev, struct pdflush_work, list);
 154                 if (time_after(jiffies, pdf->when_i_went_to_sleep + 1 * HZ)) {
 155                         /* Limit exit rate */
 156                         pdf->when_i_went_to_sleep = jiffies;
 157                         break;                                  /* exeunt */
 158                 }
 159         }
 160         nr_pdflush_threads--;
 161         spin_unlock_irq(&pdflush_lock);
 162         return 0;
 163 }
 164
 165 /*
 166  * Of course, my_work wants to be just a local in __pdflush().  It is
 167  * separated out in this manner to hopefully prevent the compiler from
 168  * performing unfortunate optimisations against the auto variables.  Because
 169  * these are visible to other tasks and CPUs.  (No problem has actually
 170  * been observed.  This is just paranoia).
 171  */
 172 static int pdflush(void *dummy)
 173 {
 174         struct pdflush_work my_work;
 175         cpumask_var_t cpus_allowed;
 176
 177         /*
 178          * Since the caller doesn't even check kthread_run() worked, let's not
 179          * freak out too much if this fails.
 180          */
 181         if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) {
 182                 printk(KERN_WARNING "pdflush failed to allocate cpumask\n");
 183                 return 0;
 184         }
 185
 186         /*
 187          * pdflush can spend a lot of time doing encryption via dm-crypt.  We
 188          * don't want to do that at keventd's priority.
 189          */
 190         set_user_nice(current, 0);
 191
 192         /*
 193          * Some configs put our parent kthread in a limited cpuset,
 194          * which kthread() overrides, forcing cpus_allowed == CPU_MASK_ALL.
 195          * Our needs are more modest - cut back to our cpusets cpus_allowed.
 196          * This is needed as pdflush's are dynamically created and destroyed.
 197          * The boottime pdflush's are easily placed w/o these 2 lines.
 198          */
 199         cpuset_cpus_allowed(current, cpus_allowed);
 200         set_cpus_allowed_ptr(current, cpus_allowed);
 201         free_cpumask_var(cpus_allowed);
 202
 203         return __pdflush(&my_work);
 204 }
 205
 206 /*
 207  * Attempt to wake up a pdflush thread, and get it to do some work for you.
 208  * Returns zero if it indeed managed to find a worker thread, and passed your
 209  * payload to it.
 210  */
 211 int pdflush_operation(void (*fn)(unsigned long), unsigned long arg0)
 212 {
 213         unsigned long flags;
 214         int ret = 0;
 215
 216         BUG_ON(fn == NULL);     /* Hard to diagnose if it's deferred */
 217
 218         spin_lock_irqsave(&pdflush_lock, flags);
 219         if (list_empty(&pdflush_list)) {
 220                 ret = -1;
 221         } else {
 222                 struct pdflush_work *pdf;
 223
 224                 pdf = list_entry(pdflush_list.next, struct pdflush_work, list);
 225                 list_del_init(&pdf->list);
 226                 if (list_empty(&pdflush_list))
 227                         last_empty_jifs = jiffies;
 228                 pdf->fn = fn;
 229                 pdf->arg0 = arg0;
 230                 wake_up_process(pdf->who);
 231         }
 232         spin_unlock_irqrestore(&pdflush_lock, flags);
 233
 234         return ret;
 235 }
 236
 237 static void start_one_pdflush_thread(void)
 238 {
 239         kthread_run(pdflush, NULL, "pdflush");
 240 }
 241
 242 static int __init pdflush_init(void)
 243 {
 244         int i;
 245
 246         for (i = 0; i < MIN_PDFLUSH_THREADS; i++)
 247                 start_one_pdflush_thread();
 248         return 0;
 249 }
 250
 251 module_init(pdflush_init);