1 |
/* |
2 |
* Written by Doug Lea with assistance from members of JCP JSR-166 |
3 |
* Expert Group and released to the public domain. Use, modify, and |
4 |
* redistribute this code in any way without acknowledgement. |
5 |
*/ |
6 |
|
7 |
package java.util.concurrent; |
8 |
import java.util.*; |
9 |
import java.util.concurrent.atomic.*; |
10 |
|
11 |
|
12 |
/** |
13 |
* An unbounded thread-safe queue based on linked nodes. LinkedQueues |
14 |
* are an especially good choice when many threads will share access |
15 |
* to a common queue. |
16 |
* |
17 |
* <p> This implementation employs an efficient "wait-free" algorithm |
18 |
* based on one described in <a |
19 |
* href="http://www.cs.rochester.edu/u/michael/PODC96.html"> Simple, |
20 |
* Fast, and Practical Non-Blocking and Blocking Concurrent Queue |
21 |
* Algorithms</a> by Maged M. Michael and Michael L. Scott.) |
22 |
* |
23 |
* Beware that, unlike in most collections, the <tt>size</tt> method |
24 |
* is <em>NOT</em> a constant-time operation. Because of the |
25 |
* asynchronous nature of these queues, determining the current number |
26 |
* of elements requires an O(n) traversal. |
27 |
* |
28 |
**/ |
29 |
public class LinkedQueue<E> extends AbstractQueue<E> |
30 |
implements Queue<E>, java.io.Serializable { |
31 |
|
32 |
/* |
33 |
* This is a straight adaptation of Michael & Scott algorithm. |
34 |
* For explanation, read the paper. The only (minor) algorithmic |
35 |
* difference is that this version supports lazy deletion of |
36 |
* internal nodes (method remove(Object)) -- remove CAS'es item |
37 |
* fields to null. The normal queue operations unlink but then |
38 |
* pass over nodes with null item fields. Similarly, iteration |
39 |
* methods ignore those with nulls. |
40 |
*/ |
41 |
|
42 |
// Atomics support |
43 |
|
44 |
private final static AtomicReferenceFieldUpdater<LinkedQueue, AtomicLinkedNode> tailUpdater = new AtomicReferenceFieldUpdater<LinkedQueue, AtomicLinkedNode>(new LinkedQueue[0], new AtomicLinkedNode[0], "tail"); |
45 |
private final static AtomicReferenceFieldUpdater<LinkedQueue, AtomicLinkedNode> headUpdater = new AtomicReferenceFieldUpdater<LinkedQueue, AtomicLinkedNode>(new LinkedQueue[0], new AtomicLinkedNode[0], "head"); |
46 |
|
47 |
private boolean casTail(AtomicLinkedNode cmp, AtomicLinkedNode val) { |
48 |
return tailUpdater.compareAndSet(this, cmp, val); |
49 |
} |
50 |
|
51 |
private boolean casHead(AtomicLinkedNode cmp, AtomicLinkedNode val) { |
52 |
return headUpdater.compareAndSet(this, cmp, val); |
53 |
} |
54 |
|
55 |
|
56 |
/** |
57 |
* Pointer to header node, initialized to a dummy node. The first |
58 |
* actual node is at head.getNext(). |
59 |
*/ |
60 |
private transient volatile AtomicLinkedNode head = new AtomicLinkedNode(null, null); |
61 |
|
62 |
/** Pointer to last node on list **/ |
63 |
private transient volatile AtomicLinkedNode tail = head; |
64 |
|
65 |
|
66 |
/** |
67 |
* Creates an initially empty LinkedQueue. |
68 |
*/ |
69 |
public LinkedQueue() {} |
70 |
|
71 |
/** |
72 |
* Creates a LinkedQueue initially holding the elements |
73 |
* of the given collection. The elements are added in |
74 |
* iterator traversal order. |
75 |
* |
76 |
* @param initialElements the collections whose elements are to be added. |
77 |
*/ |
78 |
public LinkedQueue(Collection<E> initialElements) { |
79 |
for (Iterator<E> it = initialElements.iterator(); it.hasNext();) |
80 |
add(it.next()); |
81 |
} |
82 |
|
83 |
public boolean offer(E x) { |
84 |
if (x == null) throw new IllegalArgumentException(); |
85 |
AtomicLinkedNode n = new AtomicLinkedNode(x, null); |
86 |
for(;;) { |
87 |
AtomicLinkedNode t = tail; |
88 |
AtomicLinkedNode s = t.getNext(); |
89 |
if (t == tail) { |
90 |
if (s == null) { |
91 |
if (t.casNext(s, n)) { |
92 |
casTail(t, n); |
93 |
return true; |
94 |
} |
95 |
} |
96 |
else { |
97 |
casTail(t, s); |
98 |
} |
99 |
} |
100 |
} |
101 |
} |
102 |
|
103 |
public E poll() { |
104 |
for (;;) { |
105 |
AtomicLinkedNode h = head; |
106 |
AtomicLinkedNode t = tail; |
107 |
AtomicLinkedNode first = h.getNext(); |
108 |
if (h == head) { |
109 |
if (h == t) { |
110 |
if (first == null) |
111 |
return null; |
112 |
else |
113 |
casTail(t, first); |
114 |
} |
115 |
else if (casHead(h, first)) { |
116 |
E item = (E)first.getItem(); |
117 |
if (item != null) { |
118 |
first.setItem(null); |
119 |
return item; |
120 |
} |
121 |
// else skip over deleted item, continue loop, |
122 |
} |
123 |
} |
124 |
} |
125 |
} |
126 |
|
127 |
public E peek() { // same as poll except don't remove item |
128 |
for (;;) { |
129 |
AtomicLinkedNode h = head; |
130 |
AtomicLinkedNode t = tail; |
131 |
AtomicLinkedNode first = h.getNext(); |
132 |
if (h == head) { |
133 |
if (h == t) { |
134 |
if (first == null) |
135 |
return null; |
136 |
else |
137 |
casTail(t, first); |
138 |
} |
139 |
else { |
140 |
E item = (E)first.getItem(); |
141 |
if (item != null) |
142 |
return item; |
143 |
else // remove deleted node and continue |
144 |
casHead(h, first); |
145 |
} |
146 |
} |
147 |
} |
148 |
} |
149 |
|
150 |
/** |
151 |
* Return the first actual (non-header) node on list. This is yet |
152 |
* another variant of poll/peek; here returning out the first |
153 |
* node, not element (so we cannot collapse with peek() without |
154 |
* introducing race.) |
155 |
*/ |
156 |
AtomicLinkedNode first() { |
157 |
for (;;) { |
158 |
AtomicLinkedNode h = head; |
159 |
AtomicLinkedNode t = tail; |
160 |
AtomicLinkedNode first = h.getNext(); |
161 |
if (h == head) { |
162 |
if (h == t) { |
163 |
if (first == null) |
164 |
return null; |
165 |
else |
166 |
casTail(t, first); |
167 |
} |
168 |
else { |
169 |
if (first.getItem() != null) |
170 |
return first; |
171 |
else // remove deleted node and continue |
172 |
casHead(h, first); |
173 |
} |
174 |
} |
175 |
} |
176 |
} |
177 |
|
178 |
|
179 |
public boolean isEmpty() { |
180 |
return first() == null; |
181 |
} |
182 |
|
183 |
/** |
184 |
* Returns the number of elements in this collection. |
185 |
* |
186 |
* Beware that, unlike in most collection, this method> is |
187 |
* <em>NOT</em> a constant-time operation. Because of the |
188 |
* asynchronous nature of these queues, determining the current |
189 |
* number of elements requires an O(n) traversal. |
190 |
* @return the number of elements in this collection |
191 |
*/ |
192 |
public int size() { |
193 |
int count = 0; |
194 |
for (AtomicLinkedNode p = first(); p != null; p = p.getNext()) { |
195 |
if (p.getItem() != null) |
196 |
++count; |
197 |
} |
198 |
return count; |
199 |
} |
200 |
|
201 |
public boolean contains(Object x) { |
202 |
if (x == null) return false; |
203 |
for (AtomicLinkedNode p = first(); p != null; p = p.getNext()) { |
204 |
Object item = p.getItem(); |
205 |
if (item != null && |
206 |
x.equals(item)) |
207 |
return true; |
208 |
} |
209 |
return false; |
210 |
} |
211 |
|
212 |
public boolean remove(Object x) { |
213 |
if (x == null) return false; |
214 |
for (AtomicLinkedNode p = first(); p != null; p = p.getNext()) { |
215 |
Object item = p.getItem(); |
216 |
if (item != null && |
217 |
x.equals(item) && |
218 |
p.casItem(item, null)) |
219 |
return true; |
220 |
} |
221 |
return false; |
222 |
} |
223 |
|
224 |
public Object[] toArray() { |
225 |
// Use ArrayList to deal with resizing. |
226 |
ArrayList al = new ArrayList(); |
227 |
for (AtomicLinkedNode p = first(); p != null; p = p.getNext()) { |
228 |
Object item = p.getItem(); |
229 |
if (item != null) |
230 |
al.add(item); |
231 |
} |
232 |
return al.toArray(); |
233 |
} |
234 |
|
235 |
public <T> T[] toArray(T[] a) { |
236 |
// try to use sent-in array |
237 |
int k = 0; |
238 |
AtomicLinkedNode p; |
239 |
for (p = first(); p != null && k < a.length; p = p.getNext()) { |
240 |
Object item = p.getItem(); |
241 |
if (item != null) |
242 |
a[k++] = (T)item; |
243 |
} |
244 |
if (p == null) { |
245 |
if (k < a.length) |
246 |
a[k] = null; |
247 |
return a; |
248 |
} |
249 |
|
250 |
// If won't fit, use ArrayList version |
251 |
ArrayList al = new ArrayList(); |
252 |
for (AtomicLinkedNode q = first(); q != null; q = q.getNext()) { |
253 |
Object item = q.getItem(); |
254 |
if (item != null) |
255 |
al.add(item); |
256 |
} |
257 |
return (T[])al.toArray(a); |
258 |
} |
259 |
|
260 |
public Iterator<E> iterator() { |
261 |
return new Itr(); |
262 |
} |
263 |
|
264 |
private class Itr implements Iterator<E> { |
265 |
/** |
266 |
* Next node to return item for. |
267 |
*/ |
268 |
private AtomicLinkedNode nextNode; |
269 |
|
270 |
/** |
271 |
* nextItem holds on to item fields because once we claim |
272 |
* that an element exists in hasNext(), we must return it in |
273 |
* the following next() call even if it was in the process of |
274 |
* being removed when hasNext() was called. |
275 |
**/ |
276 |
private E nextItem; |
277 |
|
278 |
/** |
279 |
* Node of the last returned item, to support remove. |
280 |
*/ |
281 |
private AtomicLinkedNode lastRet; |
282 |
|
283 |
Itr() { |
284 |
advance(); |
285 |
} |
286 |
|
287 |
/** |
288 |
* Move to next valid node. |
289 |
* Return item to return for next(), or null if no such. |
290 |
*/ |
291 |
private E advance() { |
292 |
lastRet = nextNode; |
293 |
E x = (E)nextItem; |
294 |
|
295 |
AtomicLinkedNode p = (nextNode == null)? first() : nextNode.getNext(); |
296 |
for (;;) { |
297 |
if (p == null) { |
298 |
nextNode = null; |
299 |
nextItem = null; |
300 |
return x; |
301 |
} |
302 |
E item = (E)p.getItem(); |
303 |
if (item != null) { |
304 |
nextNode = p; |
305 |
nextItem = item; |
306 |
return x; |
307 |
} |
308 |
else // skip over nulls |
309 |
p = p.getNext(); |
310 |
} |
311 |
} |
312 |
|
313 |
public boolean hasNext() { |
314 |
return nextNode != null; |
315 |
} |
316 |
|
317 |
public E next() { |
318 |
if (nextNode == null) throw new NoSuchElementException(); |
319 |
return advance(); |
320 |
} |
321 |
|
322 |
public void remove() { |
323 |
AtomicLinkedNode l = lastRet; |
324 |
if (l == null) throw new IllegalStateException(); |
325 |
// rely on a future traversal to relink. |
326 |
l.setItem(null); |
327 |
lastRet = null; |
328 |
} |
329 |
} |
330 |
|
331 |
/** |
332 |
* Save the state to a stream (that is, serialize it). |
333 |
* |
334 |
* @serialData All of the elements (each an <tt>E</tt>) in |
335 |
* the proper order, followed by a null |
336 |
*/ |
337 |
private void writeObject(java.io.ObjectOutputStream s) |
338 |
throws java.io.IOException { |
339 |
|
340 |
// Write out any hidden stuff |
341 |
s.defaultWriteObject(); |
342 |
|
343 |
// Write out all elements in the proper order. |
344 |
for (AtomicLinkedNode p = first(); p != null; p = p.getNext()) { |
345 |
Object item = p.getItem(); |
346 |
if (item != null) |
347 |
s.writeObject(item); |
348 |
} |
349 |
|
350 |
// Use trailing null as sentinel |
351 |
s.writeObject(null); |
352 |
} |
353 |
|
354 |
/** |
355 |
* Reconstitute the Queue instance from a stream (that is, |
356 |
* deserialize it). |
357 |
*/ |
358 |
private void readObject(java.io.ObjectInputStream s) |
359 |
throws java.io.IOException, ClassNotFoundException { |
360 |
// Read in capacity, and any hidden stuff |
361 |
s.defaultReadObject(); |
362 |
|
363 |
// Read in all elements and place in queue |
364 |
for (;;) { |
365 |
E item = (E)s.readObject(); |
366 |
if (item == null) |
367 |
break; |
368 |
add(item); |
369 |
} |
370 |
} |
371 |
|
372 |
} |