fs/btrfs/locking.c

   1 /*
   2  * Copyright (C) 2008 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/page-flags.h>
  22 #include <asm/bug.h>
  23 #include "ctree.h"
  24 #include "extent_io.h"
  25 #include "locking.h"
  26
  27 void btrfs_assert_tree_read_locked(struct extent_buffer *eb);
  28
  29 /*
  30  * if we currently have a spinning reader or writer lock
  31  * (indicated by the rw flag) this will bump the count
  32  * of blocking holders and drop the spinlock.
  33  */
  34 void btrfs_set_lock_blocking_rw(struct extent_buffer *eb, int rw)
  35 {
  36         if (eb->lock_nested) {
  37                 read_lock(&eb->lock);
  38                 if (eb->lock_nested && current->pid == eb->lock_owner) {
  39                         read_unlock(&eb->lock);
  40                         return;
  41                 }
  42                 read_unlock(&eb->lock);
  43         }
  44         if (rw == BTRFS_WRITE_LOCK) {
  45                 if (atomic_read(&eb->blocking_writers) == 0) {
  46                         WARN_ON(atomic_read(&eb->spinning_writers) != 1);
  47                         atomic_dec(&eb->spinning_writers);
  48                         btrfs_assert_tree_locked(eb);
  49                         atomic_inc(&eb->blocking_writers);
  50                         write_unlock(&eb->lock);
  51                 }
  52         } else if (rw == BTRFS_READ_LOCK) {
  53                 btrfs_assert_tree_read_locked(eb);
  54                 atomic_inc(&eb->blocking_readers);
  55                 WARN_ON(atomic_read(&eb->spinning_readers) == 0);
  56                 atomic_dec(&eb->spinning_readers);
  57                 read_unlock(&eb->lock);
  58         }
  59         return;
  60 }
  61
  62 /*
  63  * if we currently have a blocking lock, take the spinlock
  64  * and drop our blocking count
  65  */
  66 void btrfs_clear_lock_blocking_rw(struct extent_buffer *eb, int rw)
  67 {
  68         if (eb->lock_nested) {
  69                 read_lock(&eb->lock);
  70                 if (&eb->lock_nested && current->pid == eb->lock_owner) {
  71                         read_unlock(&eb->lock);
  72                         return;
  73                 }
  74                 read_unlock(&eb->lock);
  75         }
  76         if (rw == BTRFS_WRITE_LOCK_BLOCKING) {
  77                 BUG_ON(atomic_read(&eb->blocking_writers) != 1);
  78                 write_lock(&eb->lock);
  79                 WARN_ON(atomic_read(&eb->spinning_writers));
  80                 atomic_inc(&eb->spinning_writers);
  81                 if (atomic_dec_and_test(&eb->blocking_writers))
  82                         wake_up(&eb->write_lock_wq);
  83         } else if (rw == BTRFS_READ_LOCK_BLOCKING) {
  84                 BUG_ON(atomic_read(&eb->blocking_readers) == 0);
  85                 read_lock(&eb->lock);
  86                 atomic_inc(&eb->spinning_readers);
  87                 if (atomic_dec_and_test(&eb->blocking_readers))
  88                         wake_up(&eb->read_lock_wq);
  89         }
  90         return;
  91 }
  92
  93 /*
  94  * take a spinning read lock.  This will wait for any blocking
  95  * writers
  96  */
  97 void btrfs_tree_read_lock(struct extent_buffer *eb)
  98 {
  99 again:
 100         read_lock(&eb->lock);
 101         if (atomic_read(&eb->blocking_writers) &&
 102             current->pid == eb->lock_owner) {
 103                 /*
 104                  * This extent is already write-locked by our thread. We allow
 105                  * an additional read lock to be added because it's for the same
 106                  * thread. btrfs_find_all_roots() depends on this as it may be
 107                  * called on a partly (write-)locked tree.
 108                  */
 109                 BUG_ON(eb->lock_nested);
 110                 eb->lock_nested = 1;
 111                 read_unlock(&eb->lock);
 112                 return;
 113         }
 114         read_unlock(&eb->lock);
 115         wait_event(eb->write_lock_wq, atomic_read(&eb->blocking_writers) == 0);
 116         read_lock(&eb->lock);
 117         if (atomic_read(&eb->blocking_writers)) {
 118                 read_unlock(&eb->lock);
 119                 goto again;
 120         }
 121         atomic_inc(&eb->read_locks);
 122         atomic_inc(&eb->spinning_readers);
 123 }
 124
 125 /*
 126  * returns 1 if we get the read lock and 0 if we don't
 127  * this won't wait for blocking writers
 128  */
 129 int btrfs_try_tree_read_lock(struct extent_buffer *eb)
 130 {
 131         if (atomic_read(&eb->blocking_writers))
 132                 return 0;
 133
 134         read_lock(&eb->lock);
 135         if (atomic_read(&eb->blocking_writers)) {
 136                 read_unlock(&eb->lock);
 137                 return 0;
 138         }
 139         atomic_inc(&eb->read_locks);
 140         atomic_inc(&eb->spinning_readers);
 141         return 1;
 142 }
 143
 144 /*
 145  * returns 1 if we get the read lock and 0 if we don't
 146  * this won't wait for blocking writers or readers
 147  */
 148 int btrfs_try_tree_write_lock(struct extent_buffer *eb)
 149 {
 150         if (atomic_read(&eb->blocking_writers) ||
 151             atomic_read(&eb->blocking_readers))
 152                 return 0;
 153         write_lock(&eb->lock);
 154         if (atomic_read(&eb->blocking_writers) ||
 155             atomic_read(&eb->blocking_readers)) {
 156                 write_unlock(&eb->lock);
 157                 return 0;
 158         }
 159         atomic_inc(&eb->write_locks);
 160         atomic_inc(&eb->spinning_writers);
 161         eb->lock_owner = current->pid;
 162         return 1;
 163 }
 164
 165 /*
 166  * drop a spinning read lock
 167  */
 168 void btrfs_tree_read_unlock(struct extent_buffer *eb)
 169 {
 170         if (eb->lock_nested) {
 171                 read_lock(&eb->lock);
 172                 if (eb->lock_nested && current->pid == eb->lock_owner) {
 173                         eb->lock_nested = 0;
 174                         read_unlock(&eb->lock);
 175                         return;
 176                 }
 177                 read_unlock(&eb->lock);
 178         }
 179         btrfs_assert_tree_read_locked(eb);
 180         WARN_ON(atomic_read(&eb->spinning_readers) == 0);
 181         atomic_dec(&eb->spinning_readers);
 182         atomic_dec(&eb->read_locks);
 183         read_unlock(&eb->lock);
 184 }
 185
 186 /*
 187  * drop a blocking read lock
 188  */
 189 void btrfs_tree_read_unlock_blocking(struct extent_buffer *eb)
 190 {
 191         if (eb->lock_nested) {
 192                 read_lock(&eb->lock);
 193                 if (eb->lock_nested && current->pid == eb->lock_owner) {
 194                         eb->lock_nested = 0;
 195                         read_unlock(&eb->lock);
 196                         return;
 197                 }
 198                 read_unlock(&eb->lock);
 199         }
 200         btrfs_assert_tree_read_locked(eb);
 201         WARN_ON(atomic_read(&eb->blocking_readers) == 0);
 202         if (atomic_dec_and_test(&eb->blocking_readers))
 203                 wake_up(&eb->read_lock_wq);
 204         atomic_dec(&eb->read_locks);
 205 }
 206
 207 /*
 208  * take a spinning write lock.  This will wait for both
 209  * blocking readers or writers
 210  */
 211 int btrfs_tree_lock(struct extent_buffer *eb)
 212 {
 213 again:
 214         wait_event(eb->read_lock_wq, atomic_read(&eb->blocking_readers) == 0);
 215         wait_event(eb->write_lock_wq, atomic_read(&eb->blocking_writers) == 0);
 216         write_lock(&eb->lock);
 217         if (atomic_read(&eb->blocking_readers)) {
 218                 write_unlock(&eb->lock);
 219                 wait_event(eb->read_lock_wq,
 220                            atomic_read(&eb->blocking_readers) == 0);
 221                 goto again;
 222         }
 223         if (atomic_read(&eb->blocking_writers)) {
 224                 write_unlock(&eb->lock);
 225                 wait_event(eb->write_lock_wq,
 226                            atomic_read(&eb->blocking_writers) == 0);
 227                 goto again;
 228         }
 229         WARN_ON(atomic_read(&eb->spinning_writers));
 230         atomic_inc(&eb->spinning_writers);
 231         atomic_inc(&eb->write_locks);
 232         eb->lock_owner = current->pid;
 233         return 0;
 234 }
 235
 236 /*
 237  * drop a spinning or a blocking write lock.
 238  */
 239 int btrfs_tree_unlock(struct extent_buffer *eb)
 240 {
 241         int blockers = atomic_read(&eb->blocking_writers);
 242
 243         BUG_ON(blockers > 1);
 244
 245         btrfs_assert_tree_locked(eb);
 246         atomic_dec(&eb->write_locks);
 247
 248         if (blockers) {
 249                 WARN_ON(atomic_read(&eb->spinning_writers));
 250                 atomic_dec(&eb->blocking_writers);
 251                 smp_wmb();
 252                 wake_up(&eb->write_lock_wq);
 253         } else {
 254                 WARN_ON(atomic_read(&eb->spinning_writers) != 1);
 255                 atomic_dec(&eb->spinning_writers);
 256                 write_unlock(&eb->lock);
 257         }
 258         return 0;
 259 }
 260
 261 void btrfs_assert_tree_locked(struct extent_buffer *eb)
 262 {
 263         BUG_ON(!atomic_read(&eb->write_locks));
 264 }
 265
 266 void btrfs_assert_tree_read_locked(struct extent_buffer *eb)
 267 {
 268         BUG_ON(!atomic_read(&eb->read_locks));
 269 }