summaryrefslogtreecommitdiff
path: root/debian/patches/0002-fatfs-use-a-seperate-lock-to-protect-nodehash.patch
blob: 6c3f2b495596ec1dfc546d2c818e3b264d11aa43 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
From ecca4f4c8e32ff1774892718446566c2c5c36979 Mon Sep 17 00:00:00 2001
From: Justus Winter <4winter@informatik.uni-hamburg.de>
Date: Tue, 13 May 2014 15:14:53 +0200
Subject: [PATCH 2/4] fatfs: use a seperate lock to protect nodehash

Previously, fatfs used diskfs_node_refcnt_lock to serialize access to
the nodehash.

Use a separate lock to protect nodehash.  Adjust the reference
counting accordingly.  Every node in the nodehash carries a light
reference.  When we are asked to give up that light reference, we
reacquire our lock momentarily to check whether someone else
reacquired a reference through the nodehash.

* fatfs/inode.c (nodecache_lock): New lock.
(diskfs_cached_lookup): Use a separate lock to protect nodehash.
Adjust the reference counting accordingly.
(ifind): Likewise.
(diskfs_node_iterate): Likewise.
(diskfs_node_norefs): Move the code removing the node from nodehash...
(diskfs_try_dropping_softrefs): ... here, where we check whether
someone reacquired a reference, and if so hold on to our light
reference.
---
 fatfs/inode.c | 146 ++++++++++++++++++++++++++++++++++++++++------------------
 1 file changed, 101 insertions(+), 45 deletions(-)

diff --git a/fatfs/inode.c b/fatfs/inode.c
index ed6f3f0..c3997d0 100644
--- a/fatfs/inode.c
+++ b/fatfs/inode.c
@@ -44,8 +44,19 @@
 #define INOHASH(ino)    (((unsigned)(ino))%INOHSZ)
 #endif
 
+/* The nodehash is a cache of nodes.
+
+   Access to nodehash and nodehash_nr_items is protected by
+   nodecache_lock.
+
+   Every node in the nodehash carries a light reference.  When we are
+   asked to give up that light reference, we reacquire our lock
+   momentarily to check whether someone else reacquired a reference
+   through the nodehash.  */
 static struct node *nodehash[INOHSZ];
 static size_t nodehash_nr_items;
+/* nodecache_lock must be acquired before diskfs_node_refcnt_lock.  */
+static pthread_rwlock_t nodecache_lock = PTHREAD_RWLOCK_INITIALIZER;
 
 static error_t read_node (struct node *np, vm_address_t buf);
 
@@ -58,33 +69,38 @@ inode_init ()
     nodehash[n] = 0;
 }
 
+/* Lookup node with inode number INUM.  Returns NULL if the node is
+   not found in the node cache.  */
+static struct node *
+lookup (ino_t inum)
+{
+  struct node *np;
+  for (np = nodehash[INOHASH(inum)]; np; np = np->dn->hnext)
+    if (np->cache_id == inum)
+      return np;
+  return NULL;
+}
+
 /* Fetch inode INUM, set *NPP to the node structure; gain one user
    reference and lock the node.  */
 error_t
 diskfs_cached_lookup (ino64_t inum, struct node **npp)
 {
   error_t err;
-  struct node *np;
+  struct node *np, *tmp;
   struct disknode *dn;
 
-  pthread_spin_lock (&diskfs_node_refcnt_lock);
-  for (np = nodehash[INOHASH(inum)]; np; np = np->dn->hnext)
-    if (np->cache_id == inum)
-      {
-        np->references++;
-        pthread_spin_unlock (&diskfs_node_refcnt_lock);
-        pthread_mutex_lock (&np->lock);
-        *npp = np;
-        return 0;
-      }
+  pthread_rwlock_rdlock (&nodecache_lock);
+  np = lookup (inum);
+  if (np)
+    goto gotit;
+  pthread_rwlock_unlock (&nodecache_lock);
 
   /* Format specific data for the new node.  */
   dn = malloc (sizeof (struct disknode));
   if (! dn)
-    {
-      pthread_spin_unlock (&diskfs_node_refcnt_lock);
-      return ENOMEM;
-    }
+    return ENOMEM;
+
   dn->pager = 0;
   dn->first = 0;
   dn->last = 0;
@@ -102,15 +118,25 @@ diskfs_cached_lookup (ino64_t inum, struct node **npp)
   pthread_mutex_lock (&np->lock);
   
   /* Put NP in NODEHASH.  */
+  pthread_rwlock_wrlock (&nodecache_lock);
+  tmp = lookup (inum);
+  if (tmp)
+    {
+      /* We lost a race.  */
+      diskfs_nput (np);
+      np = tmp;
+      goto gotit;
+    }
+
   dn->hnext = nodehash[INOHASH(inum)];
   if (dn->hnext)
     dn->hnext->dn->hprevp = &dn->hnext;
   dn->hprevp = &nodehash[INOHASH(inum)];
   nodehash[INOHASH(inum)] = np;
+  diskfs_nref_light (np);
   nodehash_nr_items += 1;
+  pthread_rwlock_unlock (&nodecache_lock);
 
-  pthread_spin_unlock (&diskfs_node_refcnt_lock);
-  
   /* Get the contents of NP off disk.  */
   err = read_node (np, 0);
 
@@ -121,6 +147,13 @@ diskfs_cached_lookup (ino64_t inum, struct node **npp)
       *npp = np;
       return 0;
     }
+
+ gotit:
+  diskfs_nref (np);
+  pthread_rwlock_unlock (&nodecache_lock);
+  pthread_mutex_lock (&np->lock);
+  *npp = np;
+  return 0;
 }
 
 /* Fetch inode INUM, set *NPP to the node structure;
@@ -133,24 +166,23 @@ diskfs_cached_lookup_in_dirbuf (int inum, struct node **npp, vm_address_t buf)
   struct node *np;
   struct disknode *dn;
 
-  pthread_spin_lock (&diskfs_node_refcnt_lock);
+  pthread_rwlock_rdlock (&nodecache_lock);
   for (np = nodehash[INOHASH(inum)]; np; np = np->dn->hnext)
     if (np->cache_id == inum)
       {
-        np->references++;
-        pthread_spin_unlock (&diskfs_node_refcnt_lock);
+        diskfs_nref (np);
+        pthread_rwlock_unlock (&nodecache_lock);
         pthread_mutex_lock (&np->lock);
         *npp = np;
         return 0;
       }
+  pthread_rwlock_unlock (&nodecache_lock);
 
   /* Format specific data for the new node.  */
   dn = malloc (sizeof (struct disknode));
   if (! dn)
-    {
-      pthread_spin_unlock (&diskfs_node_refcnt_lock);
-      return ENOMEM;
-    }
+    return ENOMEM;
+
   dn->pager = 0;
   dn->first = 0;
   dn->last = 0;
@@ -168,15 +200,16 @@ diskfs_cached_lookup_in_dirbuf (int inum, struct node **npp, vm_address_t buf)
   pthread_mutex_lock (&np->lock);
   
   /* Put NP in NODEHASH.  */
+  pthread_rwlock_wrlock (&nodecache_lock);
   dn->hnext = nodehash[INOHASH(inum)];
   if (dn->hnext)
     dn->hnext->dn->hprevp = &dn->hnext;
   dn->hprevp = &nodehash[INOHASH(inum)];
   nodehash[INOHASH(inum)] = np;
+  diskfs_nref_light (np);
   nodehash_nr_items += 1;
+  pthread_rwlock_unlock (&nodecache_lock);
 
-  pthread_spin_unlock (&diskfs_node_refcnt_lock);
-  
   /* Get the contents of NP off disk.  */
   err = read_node (np, buf);
 
@@ -196,17 +229,12 @@ ifind (ino_t inum)
 {
   struct node *np;
 
-  pthread_spin_lock (&diskfs_node_refcnt_lock);
-  for (np = nodehash[INOHASH(inum)]; np; np = np->dn->hnext)
-    {
-      if (np->cache_id != inum)
-        continue;
+  pthread_rwlock_rdlock (&nodecache_lock);
+  np = lookup (inum);
+  pthread_rwlock_unlock (&nodecache_lock);
 
-      assert (np->references);
-      pthread_spin_unlock (&diskfs_node_refcnt_lock);
-      return np;
-    }
-  assert (0);
+  assert (np);
+  return np;
 }
 
 /* The last reference to a node has gone away; drop it from the hash
@@ -216,11 +244,6 @@ diskfs_node_norefs (struct node *np)
 {
   struct cluster_chain *last = np->dn->first;
 
-  *np->dn->hprevp = np->dn->hnext;
-  if (np->dn->hnext)
-    np->dn->hnext->dn->hprevp = np->dn->hprevp;
-  nodehash_nr_items -= 1;
-
   while (last)
     {
       struct cluster_chain *next = last->next;
@@ -251,6 +274,35 @@ diskfs_node_norefs (struct node *np)
 void
 diskfs_try_dropping_softrefs (struct node *np)
 {
+  pthread_rwlock_wrlock (&nodecache_lock);
+  if (np->dn->hnext != NULL)
+    {
+      /* Check if someone reacquired a reference through the
+	 nodehash.  */
+      unsigned int references;
+      pthread_spin_lock (&diskfs_node_refcnt_lock);
+      references = np->references;
+      pthread_spin_unlock (&diskfs_node_refcnt_lock);
+
+      /* An additional reference is acquired by libdiskfs across calls
+	 to diskfs_try_dropping_softrefs.  */
+      if (references > 1)
+	{
+	  /* A reference was reacquired through a hash table lookup.
+	     It's fine, we didn't touch anything yet. */
+	  pthread_rwlock_unlock (&nodecache_lock);
+	  return;
+	}
+
+      *np->dn->hprevp = np->dn->hnext;
+      if (np->dn->hnext)
+	np->dn->hnext->dn->hprevp = np->dn->hprevp;
+      np->dn->hnext = NULL;
+      nodehash_nr_items -= 1;
+      diskfs_nrele_light (np);
+    }
+  pthread_rwlock_unlock (&nodecache_lock);
+
   drop_pager_softrefs (np);
 }
 
@@ -554,12 +606,12 @@ diskfs_node_iterate (error_t (*fun)(struct node *))
   size_t num_nodes;
   struct node *node, **node_list, **p;
 
-  pthread_spin_lock (&diskfs_node_refcnt_lock);
+  pthread_rwlock_rdlock (&nodecache_lock);
 
   /* We must copy everything from the hash table into another data structure
      to avoid running into any problems with the hash-table being modified
      during processing (normally we delegate access to hash-table with
-     diskfs_node_refcnt_lock, but we can't hold this while locking the
+     nodecache_lock, but we can't hold this while locking the
      individual node locks).  */
 
   num_nodes = nodehash_nr_items;
@@ -570,10 +622,14 @@ diskfs_node_iterate (error_t (*fun)(struct node *))
     for (node = nodehash[n]; node; node = node->dn->hnext)
       {
         *p++ = node;
-        node->references++;
+
+	/* We acquire a hard reference for node, but without using
+	   diskfs_nref.	 We do this so that diskfs_new_hardrefs will not
+	   get called.	*/
+	node->references++;
       }
 
-  pthread_spin_unlock (&diskfs_node_refcnt_lock);
+  pthread_rwlock_unlock (&nodecache_lock);
 
   p = node_list;
   while (num_nodes-- > 0)
-- 
2.0.0.rc2