2 * Copyright (c) 2010 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * Implement the swapcache daemon. When enabled swap is assumed to be
37 * configured on a fast storage device such as a SSD. Swap is assigned
38 * to clean vnode-backed pages in the inactive queue, clustered by object
39 * if possible, and written out. The swap assignment sticks around even
40 * after the underlying pages have been recycled.
42 * The daemon manages write bandwidth based on sysctl settings to control
45 * The vnode strategy code will check for the swap assignments and divert
46 * reads to the swap device.
48 * This operates on both regular files and the block device vnodes used by
49 * filesystems to manage meta-data.
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/kernel.h>
57 #include <sys/kthread.h>
58 #include <sys/resourcevar.h>
59 #include <sys/signalvar.h>
60 #include <sys/vnode.h>
61 #include <sys/vmmeter.h>
62 #include <sys/sysctl.h>
65 #include <vm/vm_param.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_page.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_pageout.h>
71 #include <vm/vm_pager.h>
72 #include <vm/swap_pager.h>
73 #include <vm/vm_extern.h>
75 #include <sys/thread2.h>
76 #include <vm/vm_page2.h>
78 #define INACTIVE_LIST (&vm_page_queues[PQ_INACTIVE].pl)
80 /* the kernel process "vm_pageout"*/
81 static void vm_swapcached (void);
82 static void vm_swapcached_flush (vm_page_t m);
83 struct thread *swapcached_thread;
85 static struct kproc_desc swpc_kp = {
90 SYSINIT(swapcached, SI_SUB_KTHREAD_PAGE, SI_ORDER_SECOND, kproc_start, &swpc_kp)
92 SYSCTL_NODE(_vm, OID_AUTO, swapcache, CTLFLAG_RW, NULL, NULL);
94 int vm_swapcache_read_enable;
95 static int vm_swapcache_sleep;
96 static int vm_swapcache_maxlaunder = 256;
97 static int vm_swapcache_data_enable = 0;
98 static int vm_swapcache_meta_enable = 0;
99 static int64_t vm_swapcache_curburst = 1000000000LL;
100 static int64_t vm_swapcache_maxburst = 1000000000LL;
101 static int64_t vm_swapcache_accrate = 1000000LL;
102 static int64_t vm_swapcache_write_count;
104 SYSCTL_INT(_vm_swapcache, OID_AUTO, maxlaunder,
105 CTLFLAG_RW, &vm_swapcache_maxlaunder, 0, "");
107 SYSCTL_INT(_vm_swapcache, OID_AUTO, data_enable,
108 CTLFLAG_RW, &vm_swapcache_data_enable, 0, "");
109 SYSCTL_INT(_vm_swapcache, OID_AUTO, meta_enable,
110 CTLFLAG_RW, &vm_swapcache_meta_enable, 0, "");
111 SYSCTL_INT(_vm_swapcache, OID_AUTO, read_enable,
112 CTLFLAG_RW, &vm_swapcache_read_enable, 0, "");
114 SYSCTL_QUAD(_vm_swapcache, OID_AUTO, curburst,
115 CTLFLAG_RW, &vm_swapcache_curburst, 0, "");
116 SYSCTL_QUAD(_vm_swapcache, OID_AUTO, maxburst,
117 CTLFLAG_RW, &vm_swapcache_maxburst, 0, "");
118 SYSCTL_QUAD(_vm_swapcache, OID_AUTO, accrate,
119 CTLFLAG_RW, &vm_swapcache_accrate, 0, "");
120 SYSCTL_QUAD(_vm_swapcache, OID_AUTO, write_count,
121 CTLFLAG_RW, &vm_swapcache_write_count, 0, "");
124 * vm_swapcached is the high level pageout daemon.
129 struct vm_page marker;
138 curthread->td_flags |= TDF_SYSTHREAD;
141 * Initialize our marker
143 bzero(&marker, sizeof(marker));
144 marker.flags = PG_BUSY | PG_FICTITIOUS | PG_MARKER;
145 marker.queue = PQ_INACTIVE;
146 marker.wire_count = 1;
149 TAILQ_INSERT_HEAD(INACTIVE_LIST, &marker, pageq);
153 * Loop once a second or so looking for work when enabled.
155 if (vm_swapcache_data_enable == 0 &&
156 vm_swapcache_meta_enable == 0) {
157 tsleep(&vm_swapcache_sleep, 0, "csleep", hz * 5);
162 * Polling rate when enabled is 10 hz. Deal with write
165 * We don't want to nickle-and-dime the scan as that will
166 * create unnecessary fragmentation.
168 tsleep(&vm_swapcache_sleep, 0, "csleep", hz / 10);
169 vm_swapcache_curburst += vm_swapcache_accrate / 10;
170 if (vm_swapcache_curburst > vm_swapcache_maxburst)
171 vm_swapcache_curburst = vm_swapcache_maxburst;
172 if (vm_swapcache_curburst < vm_swapcache_accrate)
176 * Don't load any more into the cache once we have exceeded
177 * 3/4 of available swap space. XXX need to start cleaning
178 * it out, though vnode recycling will accomplish that to
181 if (vm_swap_cache_use > vm_swap_max * 3 / 4)
185 * Calculate the number of pages to test. We don't want
186 * to get into a cpu-bound loop.
188 count = vmstats.v_inactive_count;
189 if (count > vm_swapcache_maxlaunder)
190 count = vm_swapcache_maxlaunder;
193 * Scan the inactive queue from our marker to locate
194 * suitable pages to push to the swap cache.
196 * We are looking for clean vnode-backed pages.
198 * NOTE: PG_SWAPPED pages in particular are not part of
199 * our count because once the cache stabilizes we
200 * can end up with a very high datarate of VM pages
204 while ((m = TAILQ_NEXT(m, pageq)) != NULL && count--) {
205 if (m->flags & (PG_MARKER | PG_SWAPPED)) {
209 if (vm_swapcache_curburst < 0)
211 if (m->flags & (PG_BUSY | PG_UNMANAGED))
213 if (m->busy || m->hold_count || m->wire_count)
215 if (m->valid != VM_PAGE_BITS_ALL)
217 if (m->dirty & m->valid)
219 if ((object = m->object) == NULL)
221 if (object->type != OBJT_VNODE ||
222 (object->flags & OBJ_DEAD)) {
225 vm_page_test_dirty(m);
226 if (m->dirty & m->valid)
233 if (vm_swapcache_data_enable == 0)
237 if (vm_swapcache_meta_enable == 0)
245 * Ok, move the marker and soft-busy the page.
247 TAILQ_REMOVE(INACTIVE_LIST, &marker, pageq);
248 TAILQ_INSERT_AFTER(INACTIVE_LIST, m, &marker, pageq);
251 * Assign swap and initiate I/O
253 vm_swapcached_flush(m);
256 * Setup for next loop using marker.
262 * Cleanup marker position. If we hit the end of the
263 * list the marker is placed at the tail. Newly deactivated
264 * pages will be placed after it.
266 * Earlier inactive pages that were dirty and become clean
267 * are typically moved to the end of PQ_INACTIVE by virtue
268 * of vfs_vmio_release() when they become unwired from the
271 TAILQ_REMOVE(INACTIVE_LIST, &marker, pageq);
273 TAILQ_INSERT_BEFORE(m, &marker, pageq);
275 TAILQ_INSERT_TAIL(INACTIVE_LIST, &marker, pageq);
278 TAILQ_REMOVE(INACTIVE_LIST, &marker, pageq);
283 * Flush the specified page using the swap_pager.
287 vm_swapcached_flush(vm_page_t m)
293 vm_page_protect(m, VM_PROT_READ);
296 vm_object_pip_add(object, 1);
297 swap_pager_putpages(object, &m, 1, FALSE, &rtvals);
298 vm_swapcache_write_count += PAGE_SIZE;
299 vm_swapcache_curburst -= PAGE_SIZE;
301 if (rtvals != VM_PAGER_PEND) {
302 vm_object_pip_wakeup(object);
303 vm_page_io_finish(m);