1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3 * This file and its contents are supplied under the terms of the
4 * Common Development and Distribution License ("CDDL"), version 1.0.
5 * You may only use this file in accordance with the terms of version
6 * 1.0 of the CDDL.
7 *
8 * A full copy of the text of the CDDL should have accompanied this
9 * source. A copy of the CDDL is also available via the Internet at
10 * https://opensource.org/license/CDDL-1.0.
11 */
12 /*
13 * Copyright (c) 2014, 2018 by Delphix. All rights reserved.
14 */
15
16 #include <sys/bqueue.h>
17 #include <sys/zfs_context.h>
18
19 static inline bqueue_node_t *
obj2node(bqueue_t * q,void * data)20 obj2node(bqueue_t *q, void *data)
21 {
22 return ((bqueue_node_t *)((char *)data + q->bq_node_offset));
23 }
24
25 /*
26 * Initialize a blocking queue The maximum capacity of the queue is set to
27 * size. Types that are stored in a bqueue must contain a bqueue_node_t, and
28 * node_offset must be its offset from the start of the struct. fill_fraction
29 * is a performance tuning value; when the queue is full, any threads
30 * attempting to enqueue records will block. They will block until they're
31 * signaled, which will occur when the queue is at least 1/fill_fraction
32 * empty. Similar behavior occurs on dequeue; if the queue is empty, threads
33 * block. They will be signalled when the queue has 1/fill_fraction full.
34 * As a result, you must call bqueue_enqueue_flush() when you enqueue your
35 * final record on a thread, in case the dequeuing threads are currently
36 * blocked and that enqueue does not cause them to be woken. Alternatively,
37 * this behavior can be disabled (causing signaling to happen immediately) by
38 * setting fill_fraction to any value larger than size. Return 0 on success,
39 * or -1 on failure.
40 *
41 * Note: The caller must ensure that for a given bqueue_t, there's only a
42 * single call to bqueue_enqueue() running at a time (e.g. by calling only
43 * from a single thread, or with locking around the call). Similarly, the
44 * caller must ensure that there's only a single call to bqueue_dequeue()
45 * running at a time. However, the one call to bqueue_enqueue() may be
46 * invoked concurrently with the one call to bqueue_dequeue().
47 */
48 int
bqueue_init(bqueue_t * q,uint_t fill_fraction,size_t size,size_t node_offset)49 bqueue_init(bqueue_t *q, uint_t fill_fraction, size_t size, size_t node_offset)
50 {
51 if (fill_fraction == 0) {
52 return (-1);
53 }
54 list_create(&q->bq_list, node_offset + sizeof (bqueue_node_t),
55 node_offset + offsetof(bqueue_node_t, bqn_node));
56 list_create(&q->bq_dequeuing_list, node_offset + sizeof (bqueue_node_t),
57 node_offset + offsetof(bqueue_node_t, bqn_node));
58 list_create(&q->bq_enqueuing_list, node_offset + sizeof (bqueue_node_t),
59 node_offset + offsetof(bqueue_node_t, bqn_node));
60 cv_init(&q->bq_add_cv, NULL, CV_DEFAULT, NULL);
61 cv_init(&q->bq_pop_cv, NULL, CV_DEFAULT, NULL);
62 mutex_init(&q->bq_lock, NULL, MUTEX_DEFAULT, NULL);
63 q->bq_node_offset = node_offset;
64 q->bq_size = 0;
65 q->bq_dequeuing_size = 0;
66 q->bq_enqueuing_size = 0;
67 q->bq_maxsize = size;
68 q->bq_fill_fraction = fill_fraction;
69 return (0);
70 }
71
72 /*
73 * Destroy a blocking queue. This function asserts that there are no
74 * elements in the queue, and no one is blocked on the condition
75 * variables.
76 */
77 void
bqueue_destroy(bqueue_t * q)78 bqueue_destroy(bqueue_t *q)
79 {
80 mutex_enter(&q->bq_lock);
81 ASSERT0(q->bq_size);
82 ASSERT0(q->bq_dequeuing_size);
83 ASSERT0(q->bq_enqueuing_size);
84 cv_destroy(&q->bq_add_cv);
85 cv_destroy(&q->bq_pop_cv);
86 list_destroy(&q->bq_list);
87 list_destroy(&q->bq_dequeuing_list);
88 list_destroy(&q->bq_enqueuing_list);
89 mutex_exit(&q->bq_lock);
90 mutex_destroy(&q->bq_lock);
91 }
92
93 static void
bqueue_enqueue_impl(bqueue_t * q,void * data,size_t item_size,boolean_t flush)94 bqueue_enqueue_impl(bqueue_t *q, void *data, size_t item_size, boolean_t flush)
95 {
96 ASSERT3U(item_size, >, 0);
97 ASSERT3U(item_size, <=, q->bq_maxsize);
98
99 obj2node(q, data)->bqn_size = item_size;
100 q->bq_enqueuing_size += item_size;
101 list_insert_tail(&q->bq_enqueuing_list, data);
102
103 if (flush ||
104 q->bq_enqueuing_size >= q->bq_maxsize / q->bq_fill_fraction) {
105 /* Append the enquing list to the shared list. */
106 mutex_enter(&q->bq_lock);
107 while (q->bq_size > q->bq_maxsize) {
108 cv_wait_sig(&q->bq_add_cv, &q->bq_lock);
109 }
110 q->bq_size += q->bq_enqueuing_size;
111 list_move_tail(&q->bq_list, &q->bq_enqueuing_list);
112 q->bq_enqueuing_size = 0;
113 cv_broadcast(&q->bq_pop_cv);
114 mutex_exit(&q->bq_lock);
115 }
116 }
117
118 /*
119 * Add data to q, consuming size units of capacity. If there is insufficient
120 * capacity to consume size units, block until capacity exists. Asserts size is
121 * > 0.
122 */
123 void
bqueue_enqueue(bqueue_t * q,void * data,size_t item_size)124 bqueue_enqueue(bqueue_t *q, void *data, size_t item_size)
125 {
126 bqueue_enqueue_impl(q, data, item_size, B_FALSE);
127 }
128
129 /*
130 * Enqueue an entry, and then flush the queue. This forces the popping threads
131 * to wake up, even if we're below the fill fraction. We have this in a single
132 * function, rather than having a separate call, because it prevents race
133 * conditions between the enqueuing thread and the dequeuing thread, where the
134 * enqueueing thread will wake up the dequeuing thread, that thread will
135 * destroy the condvar before the enqueuing thread is done.
136 */
137 void
bqueue_enqueue_flush(bqueue_t * q,void * data,size_t item_size)138 bqueue_enqueue_flush(bqueue_t *q, void *data, size_t item_size)
139 {
140 bqueue_enqueue_impl(q, data, item_size, B_TRUE);
141 }
142
143 /*
144 * Take the first element off of q. If there are no elements on the queue, wait
145 * until one is put there. Return the removed element.
146 */
147 void *
bqueue_dequeue(bqueue_t * q)148 bqueue_dequeue(bqueue_t *q)
149 {
150 void *ret = list_remove_head(&q->bq_dequeuing_list);
151 if (ret == NULL) {
152 /*
153 * Dequeuing list is empty. Wait for there to be something on
154 * the shared list, then move the entire shared list to the
155 * dequeuing list.
156 */
157 mutex_enter(&q->bq_lock);
158 while (q->bq_size == 0) {
159 cv_wait_sig(&q->bq_pop_cv, &q->bq_lock);
160 }
161 ASSERT0(q->bq_dequeuing_size);
162 ASSERT(list_is_empty(&q->bq_dequeuing_list));
163 list_move_tail(&q->bq_dequeuing_list, &q->bq_list);
164 q->bq_dequeuing_size = q->bq_size;
165 q->bq_size = 0;
166 cv_broadcast(&q->bq_add_cv);
167 mutex_exit(&q->bq_lock);
168 ret = list_remove_head(&q->bq_dequeuing_list);
169 }
170 q->bq_dequeuing_size -= obj2node(q, ret)->bqn_size;
171 return (ret);
172 }
173