| @@ -1,1277 +1,1281 @@ | | | @@ -1,1277 +1,1281 @@ |
1 | /* $NetBSD: uvm_swap.c,v 1.139 2008/05/29 14:51:27 mrg Exp $ */ | | 1 | /* $NetBSD: uvm_swap.c,v 1.140 2008/09/23 08:56:15 ad Exp $ */ |
2 | | | 2 | |
3 | /* | | 3 | /* |
4 | * Copyright (c) 1995, 1996, 1997 Matthew R. Green | | 4 | * Copyright (c) 1995, 1996, 1997 Matthew R. Green |
5 | * All rights reserved. | | 5 | * All rights reserved. |
6 | * | | 6 | * |
7 | * Redistribution and use in source and binary forms, with or without | | 7 | * Redistribution and use in source and binary forms, with or without |
8 | * modification, are permitted provided that the following conditions | | 8 | * modification, are permitted provided that the following conditions |
9 | * are met: | | 9 | * are met: |
10 | * 1. Redistributions of source code must retain the above copyright | | 10 | * 1. Redistributions of source code must retain the above copyright |
11 | * notice, this list of conditions and the following disclaimer. | | 11 | * notice, this list of conditions and the following disclaimer. |
12 | * 2. Redistributions in binary form must reproduce the above copyright | | 12 | * 2. Redistributions in binary form must reproduce the above copyright |
13 | * notice, this list of conditions and the following disclaimer in the | | 13 | * notice, this list of conditions and the following disclaimer in the |
14 | * documentation and/or other materials provided with the distribution. | | 14 | * documentation and/or other materials provided with the distribution. |
15 | * | | 15 | * |
16 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR | | 16 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR |
17 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES | | 17 | * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES |
18 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. | | 18 | * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. |
19 | * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, | | 19 | * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, |
20 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, | | 20 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, |
21 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; | | 21 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
22 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED | | 22 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED |
23 | * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, | | 23 | * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
24 | * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY | | 24 | * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
25 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF | | 25 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
26 | * SUCH DAMAGE. | | 26 | * SUCH DAMAGE. |
27 | * | | 27 | * |
28 | * from: NetBSD: vm_swap.c,v 1.52 1997/12/02 13:47:37 pk Exp | | 28 | * from: NetBSD: vm_swap.c,v 1.52 1997/12/02 13:47:37 pk Exp |
29 | * from: Id: uvm_swap.c,v 1.1.2.42 1998/02/02 20:38:06 chuck Exp | | 29 | * from: Id: uvm_swap.c,v 1.1.2.42 1998/02/02 20:38:06 chuck Exp |
30 | */ | | 30 | */ |
31 | | | 31 | |
32 | #include <sys/cdefs.h> | | 32 | #include <sys/cdefs.h> |
33 | __KERNEL_RCSID(0, "$NetBSD: uvm_swap.c,v 1.139 2008/05/29 14:51:27 mrg Exp $"); | | 33 | __KERNEL_RCSID(0, "$NetBSD: uvm_swap.c,v 1.140 2008/09/23 08:56:15 ad Exp $"); |
34 | | | 34 | |
35 | #include "fs_nfs.h" | | 35 | #include "fs_nfs.h" |
36 | #include "opt_uvmhist.h" | | 36 | #include "opt_uvmhist.h" |
37 | #include "opt_compat_netbsd.h" | | 37 | #include "opt_compat_netbsd.h" |
38 | #include "opt_ddb.h" | | 38 | #include "opt_ddb.h" |
39 | | | 39 | |
40 | #include <sys/param.h> | | 40 | #include <sys/param.h> |
41 | #include <sys/systm.h> | | 41 | #include <sys/systm.h> |
42 | #include <sys/buf.h> | | 42 | #include <sys/buf.h> |
43 | #include <sys/bufq.h> | | 43 | #include <sys/bufq.h> |
44 | #include <sys/conf.h> | | 44 | #include <sys/conf.h> |
45 | #include <sys/proc.h> | | 45 | #include <sys/proc.h> |
46 | #include <sys/namei.h> | | 46 | #include <sys/namei.h> |
47 | #include <sys/disklabel.h> | | 47 | #include <sys/disklabel.h> |
48 | #include <sys/errno.h> | | 48 | #include <sys/errno.h> |
49 | #include <sys/kernel.h> | | 49 | #include <sys/kernel.h> |
50 | #include <sys/malloc.h> | | 50 | #include <sys/malloc.h> |
51 | #include <sys/vnode.h> | | 51 | #include <sys/vnode.h> |
52 | #include <sys/file.h> | | 52 | #include <sys/file.h> |
53 | #include <sys/vmem.h> | | 53 | #include <sys/vmem.h> |
54 | #include <sys/blist.h> | | 54 | #include <sys/blist.h> |
55 | #include <sys/mount.h> | | 55 | #include <sys/mount.h> |
56 | #include <sys/pool.h> | | 56 | #include <sys/pool.h> |
57 | #include <sys/syscallargs.h> | | 57 | #include <sys/syscallargs.h> |
58 | #include <sys/swap.h> | | 58 | #include <sys/swap.h> |
59 | #include <sys/kauth.h> | | 59 | #include <sys/kauth.h> |
60 | #include <sys/sysctl.h> | | 60 | #include <sys/sysctl.h> |
61 | #include <sys/workqueue.h> | | 61 | #include <sys/workqueue.h> |
62 | | | 62 | |
63 | #include <uvm/uvm.h> | | 63 | #include <uvm/uvm.h> |
64 | | | 64 | |
65 | #include <miscfs/specfs/specdev.h> | | 65 | #include <miscfs/specfs/specdev.h> |
66 | | | 66 | |
67 | /* | | 67 | /* |
68 | * uvm_swap.c: manage configuration and i/o to swap space. | | 68 | * uvm_swap.c: manage configuration and i/o to swap space. |
69 | */ | | 69 | */ |
70 | | | 70 | |
71 | /* | | 71 | /* |
72 | * swap space is managed in the following way: | | 72 | * swap space is managed in the following way: |
73 | * | | 73 | * |
74 | * each swap partition or file is described by a "swapdev" structure. | | 74 | * each swap partition or file is described by a "swapdev" structure. |
75 | * each "swapdev" structure contains a "swapent" structure which contains | | 75 | * each "swapdev" structure contains a "swapent" structure which contains |
76 | * information that is passed up to the user (via system calls). | | 76 | * information that is passed up to the user (via system calls). |
77 | * | | 77 | * |
78 | * each swap partition is assigned a "priority" (int) which controls | | 78 | * each swap partition is assigned a "priority" (int) which controls |
79 | * swap parition usage. | | 79 | * swap parition usage. |
80 | * | | 80 | * |
81 | * the system maintains a global data structure describing all swap | | 81 | * the system maintains a global data structure describing all swap |
82 | * partitions/files. there is a sorted LIST of "swappri" structures | | 82 | * partitions/files. there is a sorted LIST of "swappri" structures |
83 | * which describe "swapdev"'s at that priority. this LIST is headed | | 83 | * which describe "swapdev"'s at that priority. this LIST is headed |
84 | * by the "swap_priority" global var. each "swappri" contains a | | 84 | * by the "swap_priority" global var. each "swappri" contains a |
85 | * CIRCLEQ of "swapdev" structures at that priority. | | 85 | * CIRCLEQ of "swapdev" structures at that priority. |
86 | * | | 86 | * |
87 | * locking: | | 87 | * locking: |
88 | * - swap_syscall_lock (krwlock_t): this lock serializes the swapctl | | 88 | * - swap_syscall_lock (krwlock_t): this lock serializes the swapctl |
89 | * system call and prevents the swap priority list from changing | | 89 | * system call and prevents the swap priority list from changing |
90 | * while we are in the middle of a system call (e.g. SWAP_STATS). | | 90 | * while we are in the middle of a system call (e.g. SWAP_STATS). |
91 | * - uvm_swap_data_lock (kmutex_t): this lock protects all swap data | | 91 | * - uvm_swap_data_lock (kmutex_t): this lock protects all swap data |
92 | * structures including the priority list, the swapdev structures, | | 92 | * structures including the priority list, the swapdev structures, |
93 | * and the swapmap arena. | | 93 | * and the swapmap arena. |
94 | * | | 94 | * |
95 | * each swap device has the following info: | | 95 | * each swap device has the following info: |
96 | * - swap device in use (could be disabled, preventing future use) | | 96 | * - swap device in use (could be disabled, preventing future use) |
97 | * - swap enabled (allows new allocations on swap) | | 97 | * - swap enabled (allows new allocations on swap) |
98 | * - map info in /dev/drum | | 98 | * - map info in /dev/drum |
99 | * - vnode pointer | | 99 | * - vnode pointer |
100 | * for swap files only: | | 100 | * for swap files only: |
101 | * - block size | | 101 | * - block size |
102 | * - max byte count in buffer | | 102 | * - max byte count in buffer |
103 | * - buffer | | 103 | * - buffer |
104 | * | | 104 | * |
105 | * userland controls and configures swap with the swapctl(2) system call. | | 105 | * userland controls and configures swap with the swapctl(2) system call. |
106 | * the sys_swapctl performs the following operations: | | 106 | * the sys_swapctl performs the following operations: |
107 | * [1] SWAP_NSWAP: returns the number of swap devices currently configured | | 107 | * [1] SWAP_NSWAP: returns the number of swap devices currently configured |
108 | * [2] SWAP_STATS: given a pointer to an array of swapent structures | | 108 | * [2] SWAP_STATS: given a pointer to an array of swapent structures |
109 | * (passed in via "arg") of a size passed in via "misc" ... we load | | 109 | * (passed in via "arg") of a size passed in via "misc" ... we load |
110 | * the current swap config into the array. The actual work is done | | 110 | * the current swap config into the array. The actual work is done |
111 | * in the uvm_swap_stats(9) function. | | 111 | * in the uvm_swap_stats(9) function. |
112 | * [3] SWAP_ON: given a pathname in arg (could be device or file) and a | | 112 | * [3] SWAP_ON: given a pathname in arg (could be device or file) and a |
113 | * priority in "misc", start swapping on it. | | 113 | * priority in "misc", start swapping on it. |
114 | * [4] SWAP_OFF: as SWAP_ON, but stops swapping to a device | | 114 | * [4] SWAP_OFF: as SWAP_ON, but stops swapping to a device |
115 | * [5] SWAP_CTL: changes the priority of a swap device (new priority in | | 115 | * [5] SWAP_CTL: changes the priority of a swap device (new priority in |
116 | * "misc") | | 116 | * "misc") |
117 | */ | | 117 | */ |
118 | | | 118 | |
119 | /* | | 119 | /* |
120 | * swapdev: describes a single swap partition/file | | 120 | * swapdev: describes a single swap partition/file |
121 | * | | 121 | * |
122 | * note the following should be true: | | 122 | * note the following should be true: |
123 | * swd_inuse <= swd_nblks [number of blocks in use is <= total blocks] | | 123 | * swd_inuse <= swd_nblks [number of blocks in use is <= total blocks] |
124 | * swd_nblks <= swd_mapsize [because mapsize includes miniroot+disklabel] | | 124 | * swd_nblks <= swd_mapsize [because mapsize includes miniroot+disklabel] |
125 | */ | | 125 | */ |
126 | struct swapdev { | | 126 | struct swapdev { |
127 | struct oswapent swd_ose; | | 127 | struct oswapent swd_ose; |
128 | #define swd_dev swd_ose.ose_dev /* device id */ | | 128 | #define swd_dev swd_ose.ose_dev /* device id */ |
129 | #define swd_flags swd_ose.ose_flags /* flags:inuse/enable/fake */ | | 129 | #define swd_flags swd_ose.ose_flags /* flags:inuse/enable/fake */ |
130 | #define swd_priority swd_ose.ose_priority /* our priority */ | | 130 | #define swd_priority swd_ose.ose_priority /* our priority */ |
131 | /* also: swd_ose.ose_nblks, swd_ose.ose_inuse */ | | 131 | /* also: swd_ose.ose_nblks, swd_ose.ose_inuse */ |
132 | char *swd_path; /* saved pathname of device */ | | 132 | char *swd_path; /* saved pathname of device */ |
133 | int swd_pathlen; /* length of pathname */ | | 133 | int swd_pathlen; /* length of pathname */ |
134 | int swd_npages; /* #pages we can use */ | | 134 | int swd_npages; /* #pages we can use */ |
135 | int swd_npginuse; /* #pages in use */ | | 135 | int swd_npginuse; /* #pages in use */ |
136 | int swd_npgbad; /* #pages bad */ | | 136 | int swd_npgbad; /* #pages bad */ |
137 | int swd_drumoffset; /* page0 offset in drum */ | | 137 | int swd_drumoffset; /* page0 offset in drum */ |
138 | int swd_drumsize; /* #pages in drum */ | | 138 | int swd_drumsize; /* #pages in drum */ |
139 | blist_t swd_blist; /* blist for this swapdev */ | | 139 | blist_t swd_blist; /* blist for this swapdev */ |
140 | struct vnode *swd_vp; /* backing vnode */ | | 140 | struct vnode *swd_vp; /* backing vnode */ |
141 | CIRCLEQ_ENTRY(swapdev) swd_next; /* priority circleq */ | | 141 | CIRCLEQ_ENTRY(swapdev) swd_next; /* priority circleq */ |
142 | | | 142 | |
143 | int swd_bsize; /* blocksize (bytes) */ | | 143 | int swd_bsize; /* blocksize (bytes) */ |
144 | int swd_maxactive; /* max active i/o reqs */ | | 144 | int swd_maxactive; /* max active i/o reqs */ |
145 | struct bufq_state *swd_tab; /* buffer list */ | | 145 | struct bufq_state *swd_tab; /* buffer list */ |
146 | int swd_active; /* number of active buffers */ | | 146 | int swd_active; /* number of active buffers */ |
147 | }; | | 147 | }; |
148 | | | 148 | |
149 | /* | | 149 | /* |
150 | * swap device priority entry; the list is kept sorted on `spi_priority'. | | 150 | * swap device priority entry; the list is kept sorted on `spi_priority'. |
151 | */ | | 151 | */ |
152 | struct swappri { | | 152 | struct swappri { |
153 | int spi_priority; /* priority */ | | 153 | int spi_priority; /* priority */ |
154 | CIRCLEQ_HEAD(spi_swapdev, swapdev) spi_swapdev; | | 154 | CIRCLEQ_HEAD(spi_swapdev, swapdev) spi_swapdev; |
155 | /* circleq of swapdevs at this priority */ | | 155 | /* circleq of swapdevs at this priority */ |
156 | LIST_ENTRY(swappri) spi_swappri; /* global list of pri's */ | | 156 | LIST_ENTRY(swappri) spi_swappri; /* global list of pri's */ |
157 | }; | | 157 | }; |
158 | | | 158 | |
159 | /* | | 159 | /* |
160 | * The following two structures are used to keep track of data transfers | | 160 | * The following two structures are used to keep track of data transfers |
161 | * on swap devices associated with regular files. | | 161 | * on swap devices associated with regular files. |
162 | * NOTE: this code is more or less a copy of vnd.c; we use the same | | 162 | * NOTE: this code is more or less a copy of vnd.c; we use the same |
163 | * structure names here to ease porting.. | | 163 | * structure names here to ease porting.. |
164 | */ | | 164 | */ |
165 | struct vndxfer { | | 165 | struct vndxfer { |
166 | struct buf *vx_bp; /* Pointer to parent buffer */ | | 166 | struct buf *vx_bp; /* Pointer to parent buffer */ |
167 | struct swapdev *vx_sdp; | | 167 | struct swapdev *vx_sdp; |
168 | int vx_error; | | 168 | int vx_error; |
169 | int vx_pending; /* # of pending aux buffers */ | | 169 | int vx_pending; /* # of pending aux buffers */ |
170 | int vx_flags; | | 170 | int vx_flags; |
171 | #define VX_BUSY 1 | | 171 | #define VX_BUSY 1 |
172 | #define VX_DEAD 2 | | 172 | #define VX_DEAD 2 |
173 | }; | | 173 | }; |
174 | | | 174 | |
175 | struct vndbuf { | | 175 | struct vndbuf { |
176 | struct buf vb_buf; | | 176 | struct buf vb_buf; |
177 | struct vndxfer *vb_xfer; | | 177 | struct vndxfer *vb_xfer; |
178 | }; | | 178 | }; |
179 | | | 179 | |
180 | | | 180 | |
181 | /* | | 181 | /* |
182 | * We keep a of pool vndbuf's and vndxfer structures. | | 182 | * We keep a of pool vndbuf's and vndxfer structures. |
183 | */ | | 183 | */ |
184 | POOL_INIT(vndxfer_pool, sizeof(struct vndxfer), 0, 0, 0, "swp vnx", NULL, | | 184 | POOL_INIT(vndxfer_pool, sizeof(struct vndxfer), 0, 0, 0, "swp vnx", NULL, |
185 | IPL_BIO); | | 185 | IPL_BIO); |
186 | POOL_INIT(vndbuf_pool, sizeof(struct vndbuf), 0, 0, 0, "swp vnd", NULL, | | 186 | POOL_INIT(vndbuf_pool, sizeof(struct vndbuf), 0, 0, 0, "swp vnd", NULL, |
187 | IPL_BIO); | | 187 | IPL_BIO); |
188 | | | 188 | |
189 | /* | | 189 | /* |
190 | * local variables | | 190 | * local variables |
191 | */ | | 191 | */ |
192 | MALLOC_DEFINE(M_VMSWAP, "VM swap", "VM swap structures"); | | 192 | MALLOC_DEFINE(M_VMSWAP, "VM swap", "VM swap structures"); |
193 | static vmem_t *swapmap; /* controls the mapping of /dev/drum */ | | 193 | static vmem_t *swapmap; /* controls the mapping of /dev/drum */ |
194 | | | 194 | |
195 | /* list of all active swap devices [by priority] */ | | 195 | /* list of all active swap devices [by priority] */ |
196 | LIST_HEAD(swap_priority, swappri); | | 196 | LIST_HEAD(swap_priority, swappri); |
197 | static struct swap_priority swap_priority; | | 197 | static struct swap_priority swap_priority; |
198 | | | 198 | |
199 | /* locks */ | | 199 | /* locks */ |
200 | static krwlock_t swap_syscall_lock; | | 200 | static krwlock_t swap_syscall_lock; |
201 | | | 201 | |
202 | /* workqueue and use counter for swap to regular files */ | | 202 | /* workqueue and use counter for swap to regular files */ |
203 | static int sw_reg_count = 0; | | 203 | static int sw_reg_count = 0; |
204 | static struct workqueue *sw_reg_workqueue; | | 204 | static struct workqueue *sw_reg_workqueue; |
205 | | | 205 | |
206 | /* | | 206 | /* |
207 | * prototypes | | 207 | * prototypes |
208 | */ | | 208 | */ |
209 | static struct swapdev *swapdrum_getsdp(int); | | 209 | static struct swapdev *swapdrum_getsdp(int); |
210 | | | 210 | |
211 | static struct swapdev *swaplist_find(struct vnode *, bool); | | 211 | static struct swapdev *swaplist_find(struct vnode *, bool); |
212 | static void swaplist_insert(struct swapdev *, | | 212 | static void swaplist_insert(struct swapdev *, |
213 | struct swappri *, int); | | 213 | struct swappri *, int); |
214 | static void swaplist_trim(void); | | 214 | static void swaplist_trim(void); |
215 | | | 215 | |
216 | static int swap_on(struct lwp *, struct swapdev *); | | 216 | static int swap_on(struct lwp *, struct swapdev *); |
217 | static int swap_off(struct lwp *, struct swapdev *); | | 217 | static int swap_off(struct lwp *, struct swapdev *); |
218 | | | 218 | |
219 | static void uvm_swap_stats_locked(int, struct swapent *, int, register_t *); | | 219 | static void uvm_swap_stats_locked(int, struct swapent *, int, register_t *); |
220 | | | 220 | |
221 | static void sw_reg_strategy(struct swapdev *, struct buf *, int); | | 221 | static void sw_reg_strategy(struct swapdev *, struct buf *, int); |
222 | static void sw_reg_biodone(struct buf *); | | 222 | static void sw_reg_biodone(struct buf *); |
223 | static void sw_reg_iodone(struct work *wk, void *dummy); | | 223 | static void sw_reg_iodone(struct work *wk, void *dummy); |
224 | static void sw_reg_start(struct swapdev *); | | 224 | static void sw_reg_start(struct swapdev *); |
225 | | | 225 | |
226 | static int uvm_swap_io(struct vm_page **, int, int, int); | | 226 | static int uvm_swap_io(struct vm_page **, int, int, int); |
227 | | | 227 | |
228 | /* | | 228 | /* |
229 | * uvm_swap_init: init the swap system data structures and locks | | 229 | * uvm_swap_init: init the swap system data structures and locks |
230 | * | | 230 | * |
231 | * => called at boot time from init_main.c after the filesystems | | 231 | * => called at boot time from init_main.c after the filesystems |
232 | * are brought up (which happens after uvm_init()) | | 232 | * are brought up (which happens after uvm_init()) |
233 | */ | | 233 | */ |
234 | void | | 234 | void |
235 | uvm_swap_init(void) | | 235 | uvm_swap_init(void) |
236 | { | | 236 | { |
237 | UVMHIST_FUNC("uvm_swap_init"); | | 237 | UVMHIST_FUNC("uvm_swap_init"); |
238 | | | 238 | |
239 | UVMHIST_CALLED(pdhist); | | 239 | UVMHIST_CALLED(pdhist); |
240 | /* | | 240 | /* |
241 | * first, init the swap list, its counter, and its lock. | | 241 | * first, init the swap list, its counter, and its lock. |
242 | * then get a handle on the vnode for /dev/drum by using | | 242 | * then get a handle on the vnode for /dev/drum by using |
243 | * the its dev_t number ("swapdev", from MD conf.c). | | 243 | * the its dev_t number ("swapdev", from MD conf.c). |
244 | */ | | 244 | */ |
245 | | | 245 | |
246 | LIST_INIT(&swap_priority); | | 246 | LIST_INIT(&swap_priority); |
247 | uvmexp.nswapdev = 0; | | 247 | uvmexp.nswapdev = 0; |
248 | rw_init(&swap_syscall_lock); | | 248 | rw_init(&swap_syscall_lock); |
249 | cv_init(&uvm.scheduler_cv, "schedule"); | | 249 | cv_init(&uvm.scheduler_cv, "schedule"); |
250 | mutex_init(&uvm_swap_data_lock, MUTEX_DEFAULT, IPL_NONE); | | 250 | mutex_init(&uvm_swap_data_lock, MUTEX_DEFAULT, IPL_NONE); |
251 | | | 251 | |
252 | /* XXXSMP should be at IPL_VM, but for audio interrupt handlers. */ | | 252 | /* XXXSMP should be at IPL_VM, but for audio interrupt handlers. */ |
253 | mutex_init(&uvm_scheduler_mutex, MUTEX_SPIN, IPL_SCHED); | | 253 | mutex_init(&uvm_scheduler_mutex, MUTEX_SPIN, IPL_SCHED); |
254 | | | 254 | |
255 | if (bdevvp(swapdev, &swapdev_vp)) | | 255 | if (bdevvp(swapdev, &swapdev_vp)) |
256 | panic("uvm_swap_init: can't get vnode for swap device"); | | 256 | panic("uvm_swap_init: can't get vnode for swap device"); |
257 | if (vn_lock(swapdev_vp, LK_EXCLUSIVE | LK_RETRY)) | | 257 | if (vn_lock(swapdev_vp, LK_EXCLUSIVE | LK_RETRY)) |
258 | panic("uvm_swap_init: can't lock swap device"); | | 258 | panic("uvm_swap_init: can't lock swap device"); |
259 | if (VOP_OPEN(swapdev_vp, FREAD | FWRITE, NOCRED)) | | 259 | if (VOP_OPEN(swapdev_vp, FREAD | FWRITE, NOCRED)) |
260 | panic("uvm_swap_init: can't open swap device"); | | 260 | panic("uvm_swap_init: can't open swap device"); |
261 | VOP_UNLOCK(swapdev_vp, 0); | | 261 | VOP_UNLOCK(swapdev_vp, 0); |
262 | | | 262 | |
263 | /* | | 263 | /* |
264 | * create swap block resource map to map /dev/drum. the range | | 264 | * create swap block resource map to map /dev/drum. the range |
265 | * from 1 to INT_MAX allows 2 gigablocks of swap space. note | | 265 | * from 1 to INT_MAX allows 2 gigablocks of swap space. note |
266 | * that block 0 is reserved (used to indicate an allocation | | 266 | * that block 0 is reserved (used to indicate an allocation |
267 | * failure, or no allocation). | | 267 | * failure, or no allocation). |
268 | */ | | 268 | */ |
269 | swapmap = vmem_create("swapmap", 1, INT_MAX - 1, 1, NULL, NULL, NULL, 0, | | 269 | swapmap = vmem_create("swapmap", 1, INT_MAX - 1, 1, NULL, NULL, NULL, 0, |
270 | VM_NOSLEEP, IPL_NONE); | | 270 | VM_NOSLEEP, IPL_NONE); |
271 | if (swapmap == 0) | | 271 | if (swapmap == 0) |
272 | panic("uvm_swap_init: extent_create failed"); | | 272 | panic("uvm_swap_init: extent_create failed"); |
273 | | | 273 | |
274 | /* | | 274 | /* |
275 | * done! | | 275 | * done! |
276 | */ | | 276 | */ |
277 | uvm.swap_running = true; | | 277 | uvm.swap_running = true; |
| | | 278 | #ifdef __SWAP_BROKEN |
| | | 279 | uvm.swapout_enabled = 0; |
| | | 280 | #else |
278 | uvm.swapout_enabled = 1; | | 281 | uvm.swapout_enabled = 1; |
| | | 282 | #endif |
279 | UVMHIST_LOG(pdhist, "<- done", 0, 0, 0, 0); | | 283 | UVMHIST_LOG(pdhist, "<- done", 0, 0, 0, 0); |
280 | | | 284 | |
281 | sysctl_createv(NULL, 0, NULL, NULL, | | 285 | sysctl_createv(NULL, 0, NULL, NULL, |
282 | CTLFLAG_READWRITE, | | 286 | CTLFLAG_READWRITE, |
283 | CTLTYPE_INT, "swapout", | | 287 | CTLTYPE_INT, "swapout", |
284 | SYSCTL_DESCR("Set 0 to disable swapout of kernel stacks"), | | 288 | SYSCTL_DESCR("Set 0 to disable swapout of kernel stacks"), |
285 | NULL, 0, &uvm.swapout_enabled, 0, CTL_VM, CTL_CREATE, CTL_EOL); | | 289 | NULL, 0, &uvm.swapout_enabled, 0, CTL_VM, CTL_CREATE, CTL_EOL); |
286 | } | | 290 | } |
287 | | | 291 | |
288 | /* | | 292 | /* |
289 | * swaplist functions: functions that operate on the list of swap | | 293 | * swaplist functions: functions that operate on the list of swap |
290 | * devices on the system. | | 294 | * devices on the system. |
291 | */ | | 295 | */ |
292 | | | 296 | |
293 | /* | | 297 | /* |
294 | * swaplist_insert: insert swap device "sdp" into the global list | | 298 | * swaplist_insert: insert swap device "sdp" into the global list |
295 | * | | 299 | * |
296 | * => caller must hold both swap_syscall_lock and uvm_swap_data_lock | | 300 | * => caller must hold both swap_syscall_lock and uvm_swap_data_lock |
297 | * => caller must provide a newly malloc'd swappri structure (we will | | 301 | * => caller must provide a newly malloc'd swappri structure (we will |
298 | * FREE it if we don't need it... this it to prevent malloc blocking | | 302 | * FREE it if we don't need it... this it to prevent malloc blocking |
299 | * here while adding swap) | | 303 | * here while adding swap) |
300 | */ | | 304 | */ |
301 | static void | | 305 | static void |
302 | swaplist_insert(struct swapdev *sdp, struct swappri *newspp, int priority) | | 306 | swaplist_insert(struct swapdev *sdp, struct swappri *newspp, int priority) |
303 | { | | 307 | { |
304 | struct swappri *spp, *pspp; | | 308 | struct swappri *spp, *pspp; |
305 | UVMHIST_FUNC("swaplist_insert"); UVMHIST_CALLED(pdhist); | | 309 | UVMHIST_FUNC("swaplist_insert"); UVMHIST_CALLED(pdhist); |
306 | | | 310 | |
307 | /* | | 311 | /* |
308 | * find entry at or after which to insert the new device. | | 312 | * find entry at or after which to insert the new device. |
309 | */ | | 313 | */ |
310 | pspp = NULL; | | 314 | pspp = NULL; |
311 | LIST_FOREACH(spp, &swap_priority, spi_swappri) { | | 315 | LIST_FOREACH(spp, &swap_priority, spi_swappri) { |
312 | if (priority <= spp->spi_priority) | | 316 | if (priority <= spp->spi_priority) |
313 | break; | | 317 | break; |
314 | pspp = spp; | | 318 | pspp = spp; |
315 | } | | 319 | } |
316 | | | 320 | |
317 | /* | | 321 | /* |
318 | * new priority? | | 322 | * new priority? |
319 | */ | | 323 | */ |
320 | if (spp == NULL || spp->spi_priority != priority) { | | 324 | if (spp == NULL || spp->spi_priority != priority) { |
321 | spp = newspp; /* use newspp! */ | | 325 | spp = newspp; /* use newspp! */ |
322 | UVMHIST_LOG(pdhist, "created new swappri = %d", | | 326 | UVMHIST_LOG(pdhist, "created new swappri = %d", |
323 | priority, 0, 0, 0); | | 327 | priority, 0, 0, 0); |
324 | | | 328 | |
325 | spp->spi_priority = priority; | | 329 | spp->spi_priority = priority; |
326 | CIRCLEQ_INIT(&spp->spi_swapdev); | | 330 | CIRCLEQ_INIT(&spp->spi_swapdev); |
327 | | | 331 | |
328 | if (pspp) | | 332 | if (pspp) |
329 | LIST_INSERT_AFTER(pspp, spp, spi_swappri); | | 333 | LIST_INSERT_AFTER(pspp, spp, spi_swappri); |
330 | else | | 334 | else |
331 | LIST_INSERT_HEAD(&swap_priority, spp, spi_swappri); | | 335 | LIST_INSERT_HEAD(&swap_priority, spp, spi_swappri); |
332 | } else { | | 336 | } else { |
333 | /* we don't need a new priority structure, free it */ | | 337 | /* we don't need a new priority structure, free it */ |
334 | FREE(newspp, M_VMSWAP); | | 338 | FREE(newspp, M_VMSWAP); |
335 | } | | 339 | } |
336 | | | 340 | |
337 | /* | | 341 | /* |
338 | * priority found (or created). now insert on the priority's | | 342 | * priority found (or created). now insert on the priority's |
339 | * circleq list and bump the total number of swapdevs. | | 343 | * circleq list and bump the total number of swapdevs. |
340 | */ | | 344 | */ |
341 | sdp->swd_priority = priority; | | 345 | sdp->swd_priority = priority; |
342 | CIRCLEQ_INSERT_TAIL(&spp->spi_swapdev, sdp, swd_next); | | 346 | CIRCLEQ_INSERT_TAIL(&spp->spi_swapdev, sdp, swd_next); |
343 | uvmexp.nswapdev++; | | 347 | uvmexp.nswapdev++; |
344 | } | | 348 | } |
345 | | | 349 | |
346 | /* | | 350 | /* |
347 | * swaplist_find: find and optionally remove a swap device from the | | 351 | * swaplist_find: find and optionally remove a swap device from the |
348 | * global list. | | 352 | * global list. |
349 | * | | 353 | * |
350 | * => caller must hold both swap_syscall_lock and uvm_swap_data_lock | | 354 | * => caller must hold both swap_syscall_lock and uvm_swap_data_lock |
351 | * => we return the swapdev we found (and removed) | | 355 | * => we return the swapdev we found (and removed) |
352 | */ | | 356 | */ |
353 | static struct swapdev * | | 357 | static struct swapdev * |
354 | swaplist_find(struct vnode *vp, bool remove) | | 358 | swaplist_find(struct vnode *vp, bool remove) |
355 | { | | 359 | { |
356 | struct swapdev *sdp; | | 360 | struct swapdev *sdp; |
357 | struct swappri *spp; | | 361 | struct swappri *spp; |
358 | | | 362 | |
359 | /* | | 363 | /* |
360 | * search the lists for the requested vp | | 364 | * search the lists for the requested vp |
361 | */ | | 365 | */ |
362 | | | 366 | |
363 | LIST_FOREACH(spp, &swap_priority, spi_swappri) { | | 367 | LIST_FOREACH(spp, &swap_priority, spi_swappri) { |
364 | CIRCLEQ_FOREACH(sdp, &spp->spi_swapdev, swd_next) { | | 368 | CIRCLEQ_FOREACH(sdp, &spp->spi_swapdev, swd_next) { |
365 | if (sdp->swd_vp == vp) { | | 369 | if (sdp->swd_vp == vp) { |
366 | if (remove) { | | 370 | if (remove) { |
367 | CIRCLEQ_REMOVE(&spp->spi_swapdev, | | 371 | CIRCLEQ_REMOVE(&spp->spi_swapdev, |
368 | sdp, swd_next); | | 372 | sdp, swd_next); |
369 | uvmexp.nswapdev--; | | 373 | uvmexp.nswapdev--; |
370 | } | | 374 | } |
371 | return(sdp); | | 375 | return(sdp); |
372 | } | | 376 | } |
373 | } | | 377 | } |
374 | } | | 378 | } |
375 | return (NULL); | | 379 | return (NULL); |
376 | } | | 380 | } |
377 | | | 381 | |
378 | /* | | 382 | /* |
379 | * swaplist_trim: scan priority list for empty priority entries and kill | | 383 | * swaplist_trim: scan priority list for empty priority entries and kill |
380 | * them. | | 384 | * them. |
381 | * | | 385 | * |
382 | * => caller must hold both swap_syscall_lock and uvm_swap_data_lock | | 386 | * => caller must hold both swap_syscall_lock and uvm_swap_data_lock |
383 | */ | | 387 | */ |
384 | static void | | 388 | static void |
385 | swaplist_trim(void) | | 389 | swaplist_trim(void) |
386 | { | | 390 | { |
387 | struct swappri *spp, *nextspp; | | 391 | struct swappri *spp, *nextspp; |
388 | | | 392 | |
389 | for (spp = LIST_FIRST(&swap_priority); spp != NULL; spp = nextspp) { | | 393 | for (spp = LIST_FIRST(&swap_priority); spp != NULL; spp = nextspp) { |
390 | nextspp = LIST_NEXT(spp, spi_swappri); | | 394 | nextspp = LIST_NEXT(spp, spi_swappri); |
391 | if (CIRCLEQ_FIRST(&spp->spi_swapdev) != | | 395 | if (CIRCLEQ_FIRST(&spp->spi_swapdev) != |
392 | (void *)&spp->spi_swapdev) | | 396 | (void *)&spp->spi_swapdev) |
393 | continue; | | 397 | continue; |
394 | LIST_REMOVE(spp, spi_swappri); | | 398 | LIST_REMOVE(spp, spi_swappri); |
395 | free(spp, M_VMSWAP); | | 399 | free(spp, M_VMSWAP); |
396 | } | | 400 | } |
397 | } | | 401 | } |
398 | | | 402 | |
399 | /* | | 403 | /* |
400 | * swapdrum_getsdp: given a page offset in /dev/drum, convert it back | | 404 | * swapdrum_getsdp: given a page offset in /dev/drum, convert it back |
401 | * to the "swapdev" that maps that section of the drum. | | 405 | * to the "swapdev" that maps that section of the drum. |
402 | * | | 406 | * |
403 | * => each swapdev takes one big contig chunk of the drum | | 407 | * => each swapdev takes one big contig chunk of the drum |
404 | * => caller must hold uvm_swap_data_lock | | 408 | * => caller must hold uvm_swap_data_lock |
405 | */ | | 409 | */ |
406 | static struct swapdev * | | 410 | static struct swapdev * |
407 | swapdrum_getsdp(int pgno) | | 411 | swapdrum_getsdp(int pgno) |
408 | { | | 412 | { |
409 | struct swapdev *sdp; | | 413 | struct swapdev *sdp; |
410 | struct swappri *spp; | | 414 | struct swappri *spp; |
411 | | | 415 | |
412 | LIST_FOREACH(spp, &swap_priority, spi_swappri) { | | 416 | LIST_FOREACH(spp, &swap_priority, spi_swappri) { |
413 | CIRCLEQ_FOREACH(sdp, &spp->spi_swapdev, swd_next) { | | 417 | CIRCLEQ_FOREACH(sdp, &spp->spi_swapdev, swd_next) { |
414 | if (sdp->swd_flags & SWF_FAKE) | | 418 | if (sdp->swd_flags & SWF_FAKE) |
415 | continue; | | 419 | continue; |
416 | if (pgno >= sdp->swd_drumoffset && | | 420 | if (pgno >= sdp->swd_drumoffset && |
417 | pgno < (sdp->swd_drumoffset + sdp->swd_drumsize)) { | | 421 | pgno < (sdp->swd_drumoffset + sdp->swd_drumsize)) { |
418 | return sdp; | | 422 | return sdp; |
419 | } | | 423 | } |
420 | } | | 424 | } |
421 | } | | 425 | } |
422 | return NULL; | | 426 | return NULL; |
423 | } | | 427 | } |
424 | | | 428 | |
425 | | | 429 | |
426 | /* | | 430 | /* |
427 | * sys_swapctl: main entry point for swapctl(2) system call | | 431 | * sys_swapctl: main entry point for swapctl(2) system call |
428 | * [with two helper functions: swap_on and swap_off] | | 432 | * [with two helper functions: swap_on and swap_off] |
429 | */ | | 433 | */ |
430 | int | | 434 | int |
431 | sys_swapctl(struct lwp *l, const struct sys_swapctl_args *uap, register_t *retval) | | 435 | sys_swapctl(struct lwp *l, const struct sys_swapctl_args *uap, register_t *retval) |
432 | { | | 436 | { |
433 | /* { | | 437 | /* { |
434 | syscallarg(int) cmd; | | 438 | syscallarg(int) cmd; |
435 | syscallarg(void *) arg; | | 439 | syscallarg(void *) arg; |
436 | syscallarg(int) misc; | | 440 | syscallarg(int) misc; |
437 | } */ | | 441 | } */ |
438 | struct vnode *vp; | | 442 | struct vnode *vp; |
439 | struct nameidata nd; | | 443 | struct nameidata nd; |
440 | struct swappri *spp; | | 444 | struct swappri *spp; |
441 | struct swapdev *sdp; | | 445 | struct swapdev *sdp; |
442 | struct swapent *sep; | | 446 | struct swapent *sep; |
443 | #define SWAP_PATH_MAX (PATH_MAX + 1) | | 447 | #define SWAP_PATH_MAX (PATH_MAX + 1) |
444 | char *userpath; | | 448 | char *userpath; |
445 | size_t len; | | 449 | size_t len; |
446 | int error, misc; | | 450 | int error, misc; |
447 | int priority; | | 451 | int priority; |
448 | UVMHIST_FUNC("sys_swapctl"); UVMHIST_CALLED(pdhist); | | 452 | UVMHIST_FUNC("sys_swapctl"); UVMHIST_CALLED(pdhist); |
449 | | | 453 | |
450 | misc = SCARG(uap, misc); | | 454 | misc = SCARG(uap, misc); |
451 | | | 455 | |
452 | /* | | 456 | /* |
453 | * ensure serialized syscall access by grabbing the swap_syscall_lock | | 457 | * ensure serialized syscall access by grabbing the swap_syscall_lock |
454 | */ | | 458 | */ |
455 | rw_enter(&swap_syscall_lock, RW_WRITER); | | 459 | rw_enter(&swap_syscall_lock, RW_WRITER); |
456 | | | 460 | |
457 | userpath = malloc(SWAP_PATH_MAX, M_TEMP, M_WAITOK); | | 461 | userpath = malloc(SWAP_PATH_MAX, M_TEMP, M_WAITOK); |
458 | /* | | 462 | /* |
459 | * we handle the non-priv NSWAP and STATS request first. | | 463 | * we handle the non-priv NSWAP and STATS request first. |
460 | * | | 464 | * |
461 | * SWAP_NSWAP: return number of config'd swap devices | | 465 | * SWAP_NSWAP: return number of config'd swap devices |
462 | * [can also be obtained with uvmexp sysctl] | | 466 | * [can also be obtained with uvmexp sysctl] |
463 | */ | | 467 | */ |
464 | if (SCARG(uap, cmd) == SWAP_NSWAP) { | | 468 | if (SCARG(uap, cmd) == SWAP_NSWAP) { |
465 | UVMHIST_LOG(pdhist, "<- done SWAP_NSWAP=%d", uvmexp.nswapdev, | | 469 | UVMHIST_LOG(pdhist, "<- done SWAP_NSWAP=%d", uvmexp.nswapdev, |
466 | 0, 0, 0); | | 470 | 0, 0, 0); |
467 | *retval = uvmexp.nswapdev; | | 471 | *retval = uvmexp.nswapdev; |
468 | error = 0; | | 472 | error = 0; |
469 | goto out; | | 473 | goto out; |
470 | } | | 474 | } |
471 | | | 475 | |
472 | /* | | 476 | /* |
473 | * SWAP_STATS: get stats on current # of configured swap devs | | 477 | * SWAP_STATS: get stats on current # of configured swap devs |
474 | * | | 478 | * |
475 | * note that the swap_priority list can't change as long | | 479 | * note that the swap_priority list can't change as long |
476 | * as we are holding the swap_syscall_lock. we don't want | | 480 | * as we are holding the swap_syscall_lock. we don't want |
477 | * to grab the uvm_swap_data_lock because we may fault&sleep during | | 481 | * to grab the uvm_swap_data_lock because we may fault&sleep during |
478 | * copyout() and we don't want to be holding that lock then! | | 482 | * copyout() and we don't want to be holding that lock then! |
479 | */ | | 483 | */ |
480 | if (SCARG(uap, cmd) == SWAP_STATS | | 484 | if (SCARG(uap, cmd) == SWAP_STATS |
481 | #if defined(COMPAT_13) | | 485 | #if defined(COMPAT_13) |
482 | || SCARG(uap, cmd) == SWAP_OSTATS | | 486 | || SCARG(uap, cmd) == SWAP_OSTATS |
483 | #endif | | 487 | #endif |
484 | ) { | | 488 | ) { |
485 | if ((size_t)misc > (size_t)uvmexp.nswapdev) | | 489 | if ((size_t)misc > (size_t)uvmexp.nswapdev) |
486 | misc = uvmexp.nswapdev; | | 490 | misc = uvmexp.nswapdev; |
487 | #if defined(COMPAT_13) | | 491 | #if defined(COMPAT_13) |
488 | if (SCARG(uap, cmd) == SWAP_OSTATS) | | 492 | if (SCARG(uap, cmd) == SWAP_OSTATS) |
489 | len = sizeof(struct oswapent) * misc; | | 493 | len = sizeof(struct oswapent) * misc; |
490 | else | | 494 | else |
491 | #endif | | 495 | #endif |
492 | len = sizeof(struct swapent) * misc; | | 496 | len = sizeof(struct swapent) * misc; |
493 | sep = (struct swapent *)malloc(len, M_TEMP, M_WAITOK); | | 497 | sep = (struct swapent *)malloc(len, M_TEMP, M_WAITOK); |
494 | | | 498 | |
495 | uvm_swap_stats_locked(SCARG(uap, cmd), sep, misc, retval); | | 499 | uvm_swap_stats_locked(SCARG(uap, cmd), sep, misc, retval); |
496 | error = copyout(sep, SCARG(uap, arg), len); | | 500 | error = copyout(sep, SCARG(uap, arg), len); |
497 | | | 501 | |
498 | free(sep, M_TEMP); | | 502 | free(sep, M_TEMP); |
499 | UVMHIST_LOG(pdhist, "<- done SWAP_STATS", 0, 0, 0, 0); | | 503 | UVMHIST_LOG(pdhist, "<- done SWAP_STATS", 0, 0, 0, 0); |
500 | goto out; | | 504 | goto out; |
501 | } | | 505 | } |
502 | if (SCARG(uap, cmd) == SWAP_GETDUMPDEV) { | | 506 | if (SCARG(uap, cmd) == SWAP_GETDUMPDEV) { |
503 | dev_t *devp = (dev_t *)SCARG(uap, arg); | | 507 | dev_t *devp = (dev_t *)SCARG(uap, arg); |
504 | | | 508 | |
505 | error = copyout(&dumpdev, devp, sizeof(dumpdev)); | | 509 | error = copyout(&dumpdev, devp, sizeof(dumpdev)); |
506 | goto out; | | 510 | goto out; |
507 | } | | 511 | } |
508 | | | 512 | |
509 | /* | | 513 | /* |
510 | * all other requests require superuser privs. verify. | | 514 | * all other requests require superuser privs. verify. |
511 | */ | | 515 | */ |
512 | if ((error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_SWAPCTL, | | 516 | if ((error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_SWAPCTL, |
513 | 0, NULL, NULL, NULL))) | | 517 | 0, NULL, NULL, NULL))) |
514 | goto out; | | 518 | goto out; |
515 | | | 519 | |
516 | if (SCARG(uap, cmd) == SWAP_DUMPOFF) { | | 520 | if (SCARG(uap, cmd) == SWAP_DUMPOFF) { |
517 | /* drop the current dump device */ | | 521 | /* drop the current dump device */ |
518 | dumpdev = NODEV; | | 522 | dumpdev = NODEV; |
519 | dumpcdev = NODEV; | | 523 | dumpcdev = NODEV; |
520 | cpu_dumpconf(); | | 524 | cpu_dumpconf(); |
521 | goto out; | | 525 | goto out; |
522 | } | | 526 | } |
523 | | | 527 | |
524 | /* | | 528 | /* |
525 | * at this point we expect a path name in arg. we will | | 529 | * at this point we expect a path name in arg. we will |
526 | * use namei() to gain a vnode reference (vref), and lock | | 530 | * use namei() to gain a vnode reference (vref), and lock |
527 | * the vnode (VOP_LOCK). | | 531 | * the vnode (VOP_LOCK). |
528 | * | | 532 | * |
529 | * XXX: a NULL arg means use the root vnode pointer (e.g. for | | 533 | * XXX: a NULL arg means use the root vnode pointer (e.g. for |
530 | * miniroot) | | 534 | * miniroot) |
531 | */ | | 535 | */ |
532 | if (SCARG(uap, arg) == NULL) { | | 536 | if (SCARG(uap, arg) == NULL) { |
533 | vp = rootvp; /* miniroot */ | | 537 | vp = rootvp; /* miniroot */ |
534 | if (vget(vp, LK_EXCLUSIVE)) { | | 538 | if (vget(vp, LK_EXCLUSIVE)) { |
535 | error = EBUSY; | | 539 | error = EBUSY; |
536 | goto out; | | 540 | goto out; |
537 | } | | 541 | } |
538 | if (SCARG(uap, cmd) == SWAP_ON && | | 542 | if (SCARG(uap, cmd) == SWAP_ON && |
539 | copystr("miniroot", userpath, SWAP_PATH_MAX, &len)) | | 543 | copystr("miniroot", userpath, SWAP_PATH_MAX, &len)) |
540 | panic("swapctl: miniroot copy failed"); | | 544 | panic("swapctl: miniroot copy failed"); |
541 | } else { | | 545 | } else { |
542 | int space; | | 546 | int space; |
543 | char *where; | | 547 | char *where; |
544 | | | 548 | |
545 | if (SCARG(uap, cmd) == SWAP_ON) { | | 549 | if (SCARG(uap, cmd) == SWAP_ON) { |
546 | if ((error = copyinstr(SCARG(uap, arg), userpath, | | 550 | if ((error = copyinstr(SCARG(uap, arg), userpath, |
547 | SWAP_PATH_MAX, &len))) | | 551 | SWAP_PATH_MAX, &len))) |
548 | goto out; | | 552 | goto out; |
549 | space = UIO_SYSSPACE; | | 553 | space = UIO_SYSSPACE; |
550 | where = userpath; | | 554 | where = userpath; |
551 | } else { | | 555 | } else { |
552 | space = UIO_USERSPACE; | | 556 | space = UIO_USERSPACE; |
553 | where = (char *)SCARG(uap, arg); | | 557 | where = (char *)SCARG(uap, arg); |
554 | } | | 558 | } |
555 | NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, | | 559 | NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, |
556 | space, where); | | 560 | space, where); |
557 | if ((error = namei(&nd))) | | 561 | if ((error = namei(&nd))) |
558 | goto out; | | 562 | goto out; |
559 | vp = nd.ni_vp; | | 563 | vp = nd.ni_vp; |
560 | } | | 564 | } |
561 | /* note: "vp" is referenced and locked */ | | 565 | /* note: "vp" is referenced and locked */ |
562 | | | 566 | |
563 | error = 0; /* assume no error */ | | 567 | error = 0; /* assume no error */ |
564 | switch(SCARG(uap, cmd)) { | | 568 | switch(SCARG(uap, cmd)) { |
565 | | | 569 | |
566 | case SWAP_DUMPDEV: | | 570 | case SWAP_DUMPDEV: |
567 | if (vp->v_type != VBLK) { | | 571 | if (vp->v_type != VBLK) { |
568 | error = ENOTBLK; | | 572 | error = ENOTBLK; |
569 | break; | | 573 | break; |
570 | } | | 574 | } |
571 | if (bdevsw_lookup(vp->v_rdev)) { | | 575 | if (bdevsw_lookup(vp->v_rdev)) { |
572 | dumpdev = vp->v_rdev; | | 576 | dumpdev = vp->v_rdev; |
573 | dumpcdev = devsw_blk2chr(dumpdev); | | 577 | dumpcdev = devsw_blk2chr(dumpdev); |
574 | } else | | 578 | } else |
575 | dumpdev = NODEV; | | 579 | dumpdev = NODEV; |
576 | cpu_dumpconf(); | | 580 | cpu_dumpconf(); |
577 | break; | | 581 | break; |
578 | | | 582 | |
579 | case SWAP_CTL: | | 583 | case SWAP_CTL: |
580 | /* | | 584 | /* |
581 | * get new priority, remove old entry (if any) and then | | 585 | * get new priority, remove old entry (if any) and then |
582 | * reinsert it in the correct place. finally, prune out | | 586 | * reinsert it in the correct place. finally, prune out |
583 | * any empty priority structures. | | 587 | * any empty priority structures. |
584 | */ | | 588 | */ |
585 | priority = SCARG(uap, misc); | | 589 | priority = SCARG(uap, misc); |
586 | spp = malloc(sizeof *spp, M_VMSWAP, M_WAITOK); | | 590 | spp = malloc(sizeof *spp, M_VMSWAP, M_WAITOK); |
587 | mutex_enter(&uvm_swap_data_lock); | | 591 | mutex_enter(&uvm_swap_data_lock); |
588 | if ((sdp = swaplist_find(vp, true)) == NULL) { | | 592 | if ((sdp = swaplist_find(vp, true)) == NULL) { |
589 | error = ENOENT; | | 593 | error = ENOENT; |
590 | } else { | | 594 | } else { |
591 | swaplist_insert(sdp, spp, priority); | | 595 | swaplist_insert(sdp, spp, priority); |
592 | swaplist_trim(); | | 596 | swaplist_trim(); |
593 | } | | 597 | } |
594 | mutex_exit(&uvm_swap_data_lock); | | 598 | mutex_exit(&uvm_swap_data_lock); |
595 | if (error) | | 599 | if (error) |
596 | free(spp, M_VMSWAP); | | 600 | free(spp, M_VMSWAP); |
597 | break; | | 601 | break; |
598 | | | 602 | |
599 | case SWAP_ON: | | 603 | case SWAP_ON: |
600 | | | 604 | |
601 | /* | | 605 | /* |
602 | * check for duplicates. if none found, then insert a | | 606 | * check for duplicates. if none found, then insert a |
603 | * dummy entry on the list to prevent someone else from | | 607 | * dummy entry on the list to prevent someone else from |
604 | * trying to enable this device while we are working on | | 608 | * trying to enable this device while we are working on |
605 | * it. | | 609 | * it. |
606 | */ | | 610 | */ |
607 | | | 611 | |
608 | priority = SCARG(uap, misc); | | 612 | priority = SCARG(uap, misc); |
609 | sdp = malloc(sizeof *sdp, M_VMSWAP, M_WAITOK); | | 613 | sdp = malloc(sizeof *sdp, M_VMSWAP, M_WAITOK); |
610 | spp = malloc(sizeof *spp, M_VMSWAP, M_WAITOK); | | 614 | spp = malloc(sizeof *spp, M_VMSWAP, M_WAITOK); |
611 | memset(sdp, 0, sizeof(*sdp)); | | 615 | memset(sdp, 0, sizeof(*sdp)); |
612 | sdp->swd_flags = SWF_FAKE; | | 616 | sdp->swd_flags = SWF_FAKE; |
613 | sdp->swd_vp = vp; | | 617 | sdp->swd_vp = vp; |
614 | sdp->swd_dev = (vp->v_type == VBLK) ? vp->v_rdev : NODEV; | | 618 | sdp->swd_dev = (vp->v_type == VBLK) ? vp->v_rdev : NODEV; |
615 | bufq_alloc(&sdp->swd_tab, "disksort", BUFQ_SORT_RAWBLOCK); | | 619 | bufq_alloc(&sdp->swd_tab, "disksort", BUFQ_SORT_RAWBLOCK); |
616 | mutex_enter(&uvm_swap_data_lock); | | 620 | mutex_enter(&uvm_swap_data_lock); |
617 | if (swaplist_find(vp, false) != NULL) { | | 621 | if (swaplist_find(vp, false) != NULL) { |
618 | error = EBUSY; | | 622 | error = EBUSY; |
619 | mutex_exit(&uvm_swap_data_lock); | | 623 | mutex_exit(&uvm_swap_data_lock); |
620 | bufq_free(sdp->swd_tab); | | 624 | bufq_free(sdp->swd_tab); |
621 | free(sdp, M_VMSWAP); | | 625 | free(sdp, M_VMSWAP); |
622 | free(spp, M_VMSWAP); | | 626 | free(spp, M_VMSWAP); |
623 | break; | | 627 | break; |
624 | } | | 628 | } |
625 | swaplist_insert(sdp, spp, priority); | | 629 | swaplist_insert(sdp, spp, priority); |
626 | mutex_exit(&uvm_swap_data_lock); | | 630 | mutex_exit(&uvm_swap_data_lock); |
627 | | | 631 | |
628 | sdp->swd_pathlen = len; | | 632 | sdp->swd_pathlen = len; |
629 | sdp->swd_path = malloc(sdp->swd_pathlen, M_VMSWAP, M_WAITOK); | | 633 | sdp->swd_path = malloc(sdp->swd_pathlen, M_VMSWAP, M_WAITOK); |
630 | if (copystr(userpath, sdp->swd_path, sdp->swd_pathlen, 0) != 0) | | 634 | if (copystr(userpath, sdp->swd_path, sdp->swd_pathlen, 0) != 0) |
631 | panic("swapctl: copystr"); | | 635 | panic("swapctl: copystr"); |
632 | | | 636 | |
633 | /* | | 637 | /* |
634 | * we've now got a FAKE placeholder in the swap list. | | 638 | * we've now got a FAKE placeholder in the swap list. |
635 | * now attempt to enable swap on it. if we fail, undo | | 639 | * now attempt to enable swap on it. if we fail, undo |
636 | * what we've done and kill the fake entry we just inserted. | | 640 | * what we've done and kill the fake entry we just inserted. |
637 | * if swap_on is a success, it will clear the SWF_FAKE flag | | 641 | * if swap_on is a success, it will clear the SWF_FAKE flag |
638 | */ | | 642 | */ |
639 | | | 643 | |
640 | if ((error = swap_on(l, sdp)) != 0) { | | 644 | if ((error = swap_on(l, sdp)) != 0) { |
641 | mutex_enter(&uvm_swap_data_lock); | | 645 | mutex_enter(&uvm_swap_data_lock); |
642 | (void) swaplist_find(vp, true); /* kill fake entry */ | | 646 | (void) swaplist_find(vp, true); /* kill fake entry */ |
643 | swaplist_trim(); | | 647 | swaplist_trim(); |
644 | mutex_exit(&uvm_swap_data_lock); | | 648 | mutex_exit(&uvm_swap_data_lock); |
645 | bufq_free(sdp->swd_tab); | | 649 | bufq_free(sdp->swd_tab); |
646 | free(sdp->swd_path, M_VMSWAP); | | 650 | free(sdp->swd_path, M_VMSWAP); |
647 | free(sdp, M_VMSWAP); | | 651 | free(sdp, M_VMSWAP); |
648 | break; | | 652 | break; |
649 | } | | 653 | } |
650 | break; | | 654 | break; |
651 | | | 655 | |
652 | case SWAP_OFF: | | 656 | case SWAP_OFF: |
653 | mutex_enter(&uvm_swap_data_lock); | | 657 | mutex_enter(&uvm_swap_data_lock); |
654 | if ((sdp = swaplist_find(vp, false)) == NULL) { | | 658 | if ((sdp = swaplist_find(vp, false)) == NULL) { |
655 | mutex_exit(&uvm_swap_data_lock); | | 659 | mutex_exit(&uvm_swap_data_lock); |
656 | error = ENXIO; | | 660 | error = ENXIO; |
657 | break; | | 661 | break; |
658 | } | | 662 | } |
659 | | | 663 | |
660 | /* | | 664 | /* |
661 | * If a device isn't in use or enabled, we | | 665 | * If a device isn't in use or enabled, we |
662 | * can't stop swapping from it (again). | | 666 | * can't stop swapping from it (again). |
663 | */ | | 667 | */ |
664 | if ((sdp->swd_flags & (SWF_INUSE|SWF_ENABLE)) == 0) { | | 668 | if ((sdp->swd_flags & (SWF_INUSE|SWF_ENABLE)) == 0) { |
665 | mutex_exit(&uvm_swap_data_lock); | | 669 | mutex_exit(&uvm_swap_data_lock); |
666 | error = EBUSY; | | 670 | error = EBUSY; |
667 | break; | | 671 | break; |
668 | } | | 672 | } |
669 | | | 673 | |
670 | /* | | 674 | /* |
671 | * do the real work. | | 675 | * do the real work. |
672 | */ | | 676 | */ |
673 | error = swap_off(l, sdp); | | 677 | error = swap_off(l, sdp); |
674 | break; | | 678 | break; |
675 | | | 679 | |
676 | default: | | 680 | default: |
677 | error = EINVAL; | | 681 | error = EINVAL; |
678 | } | | 682 | } |
679 | | | 683 | |
680 | /* | | 684 | /* |
681 | * done! release the ref gained by namei() and unlock. | | 685 | * done! release the ref gained by namei() and unlock. |
682 | */ | | 686 | */ |
683 | vput(vp); | | 687 | vput(vp); |
684 | | | 688 | |
685 | out: | | 689 | out: |
686 | free(userpath, M_TEMP); | | 690 | free(userpath, M_TEMP); |
687 | rw_exit(&swap_syscall_lock); | | 691 | rw_exit(&swap_syscall_lock); |
688 | | | 692 | |
689 | UVMHIST_LOG(pdhist, "<- done! error=%d", error, 0, 0, 0); | | 693 | UVMHIST_LOG(pdhist, "<- done! error=%d", error, 0, 0, 0); |
690 | return (error); | | 694 | return (error); |
691 | } | | 695 | } |
692 | | | 696 | |
693 | /* | | 697 | /* |
694 | * swap_stats: implements swapctl(SWAP_STATS). The function is kept | | 698 | * swap_stats: implements swapctl(SWAP_STATS). The function is kept |
695 | * away from sys_swapctl() in order to allow COMPAT_* swapctl() | | 699 | * away from sys_swapctl() in order to allow COMPAT_* swapctl() |
696 | * emulation to use it directly without going through sys_swapctl(). | | 700 | * emulation to use it directly without going through sys_swapctl(). |
697 | * The problem with using sys_swapctl() there is that it involves | | 701 | * The problem with using sys_swapctl() there is that it involves |
698 | * copying the swapent array to the stackgap, and this array's size | | 702 | * copying the swapent array to the stackgap, and this array's size |
699 | * is not known at build time. Hence it would not be possible to | | 703 | * is not known at build time. Hence it would not be possible to |
700 | * ensure it would fit in the stackgap in any case. | | 704 | * ensure it would fit in the stackgap in any case. |
701 | */ | | 705 | */ |
702 | void | | 706 | void |
703 | uvm_swap_stats(int cmd, struct swapent *sep, int sec, register_t *retval) | | 707 | uvm_swap_stats(int cmd, struct swapent *sep, int sec, register_t *retval) |
704 | { | | 708 | { |
705 | | | 709 | |
706 | rw_enter(&swap_syscall_lock, RW_READER); | | 710 | rw_enter(&swap_syscall_lock, RW_READER); |
707 | uvm_swap_stats_locked(cmd, sep, sec, retval); | | 711 | uvm_swap_stats_locked(cmd, sep, sec, retval); |
708 | rw_exit(&swap_syscall_lock); | | 712 | rw_exit(&swap_syscall_lock); |
709 | } | | 713 | } |
710 | | | 714 | |
711 | static void | | 715 | static void |
712 | uvm_swap_stats_locked(int cmd, struct swapent *sep, int sec, register_t *retval) | | 716 | uvm_swap_stats_locked(int cmd, struct swapent *sep, int sec, register_t *retval) |
713 | { | | 717 | { |
714 | struct swappri *spp; | | 718 | struct swappri *spp; |
715 | struct swapdev *sdp; | | 719 | struct swapdev *sdp; |
716 | int count = 0; | | 720 | int count = 0; |
717 | | | 721 | |
718 | LIST_FOREACH(spp, &swap_priority, spi_swappri) { | | 722 | LIST_FOREACH(spp, &swap_priority, spi_swappri) { |
719 | for (sdp = CIRCLEQ_FIRST(&spp->spi_swapdev); | | 723 | for (sdp = CIRCLEQ_FIRST(&spp->spi_swapdev); |
720 | sdp != (void *)&spp->spi_swapdev && sec-- > 0; | | 724 | sdp != (void *)&spp->spi_swapdev && sec-- > 0; |
721 | sdp = CIRCLEQ_NEXT(sdp, swd_next)) { | | 725 | sdp = CIRCLEQ_NEXT(sdp, swd_next)) { |
722 | /* | | 726 | /* |
723 | * backwards compatibility for system call. | | 727 | * backwards compatibility for system call. |
724 | * note that we use 'struct oswapent' as an | | 728 | * note that we use 'struct oswapent' as an |
725 | * overlay into both 'struct swapdev' and | | 729 | * overlay into both 'struct swapdev' and |
726 | * the userland 'struct swapent', as we | | 730 | * the userland 'struct swapent', as we |
727 | * want to retain backwards compatibility | | 731 | * want to retain backwards compatibility |
728 | * with NetBSD 1.3. | | 732 | * with NetBSD 1.3. |
729 | */ | | 733 | */ |
730 | sdp->swd_ose.ose_inuse = | | 734 | sdp->swd_ose.ose_inuse = |
731 | btodb((uint64_t)sdp->swd_npginuse << | | 735 | btodb((uint64_t)sdp->swd_npginuse << |
732 | PAGE_SHIFT); | | 736 | PAGE_SHIFT); |
733 | (void)memcpy(sep, &sdp->swd_ose, | | 737 | (void)memcpy(sep, &sdp->swd_ose, |
734 | sizeof(struct oswapent)); | | 738 | sizeof(struct oswapent)); |
735 | | | 739 | |
736 | /* now copy out the path if necessary */ | | 740 | /* now copy out the path if necessary */ |
737 | #if !defined(COMPAT_13) | | 741 | #if !defined(COMPAT_13) |
738 | (void) cmd; | | 742 | (void) cmd; |
739 | #endif | | 743 | #endif |
740 | #if defined(COMPAT_13) | | 744 | #if defined(COMPAT_13) |
741 | if (cmd == SWAP_STATS) | | 745 | if (cmd == SWAP_STATS) |
742 | #endif | | 746 | #endif |
743 | (void)memcpy(&sep->se_path, sdp->swd_path, | | 747 | (void)memcpy(&sep->se_path, sdp->swd_path, |
744 | sdp->swd_pathlen); | | 748 | sdp->swd_pathlen); |
745 | | | 749 | |
746 | count++; | | 750 | count++; |
747 | #if defined(COMPAT_13) | | 751 | #if defined(COMPAT_13) |
748 | if (cmd == SWAP_OSTATS) | | 752 | if (cmd == SWAP_OSTATS) |
749 | sep = (struct swapent *) | | 753 | sep = (struct swapent *) |
750 | ((struct oswapent *)sep + 1); | | 754 | ((struct oswapent *)sep + 1); |
751 | else | | 755 | else |
752 | #endif | | 756 | #endif |
753 | sep++; | | 757 | sep++; |
754 | } | | 758 | } |
755 | } | | 759 | } |
756 | | | 760 | |
757 | *retval = count; | | 761 | *retval = count; |
758 | return; | | 762 | return; |
759 | } | | 763 | } |
760 | | | 764 | |
761 | /* | | 765 | /* |
762 | * swap_on: attempt to enable a swapdev for swapping. note that the | | 766 | * swap_on: attempt to enable a swapdev for swapping. note that the |
763 | * swapdev is already on the global list, but disabled (marked | | 767 | * swapdev is already on the global list, but disabled (marked |
764 | * SWF_FAKE). | | 768 | * SWF_FAKE). |
765 | * | | 769 | * |
766 | * => we avoid the start of the disk (to protect disk labels) | | 770 | * => we avoid the start of the disk (to protect disk labels) |
767 | * => we also avoid the miniroot, if we are swapping to root. | | 771 | * => we also avoid the miniroot, if we are swapping to root. |
768 | * => caller should leave uvm_swap_data_lock unlocked, we may lock it | | 772 | * => caller should leave uvm_swap_data_lock unlocked, we may lock it |
769 | * if needed. | | 773 | * if needed. |
770 | */ | | 774 | */ |
771 | static int | | 775 | static int |
772 | swap_on(struct lwp *l, struct swapdev *sdp) | | 776 | swap_on(struct lwp *l, struct swapdev *sdp) |
773 | { | | 777 | { |
774 | struct vnode *vp; | | 778 | struct vnode *vp; |
775 | int error, npages, nblocks, size; | | 779 | int error, npages, nblocks, size; |
776 | long addr; | | 780 | long addr; |
777 | u_long result; | | 781 | u_long result; |
778 | struct vattr va; | | 782 | struct vattr va; |
779 | #ifdef NFS | | 783 | #ifdef NFS |
780 | extern int (**nfsv2_vnodeop_p)(void *); | | 784 | extern int (**nfsv2_vnodeop_p)(void *); |
781 | #endif /* NFS */ | | 785 | #endif /* NFS */ |
782 | const struct bdevsw *bdev; | | 786 | const struct bdevsw *bdev; |
783 | dev_t dev; | | 787 | dev_t dev; |
784 | UVMHIST_FUNC("swap_on"); UVMHIST_CALLED(pdhist); | | 788 | UVMHIST_FUNC("swap_on"); UVMHIST_CALLED(pdhist); |
785 | | | 789 | |
786 | /* | | 790 | /* |
787 | * we want to enable swapping on sdp. the swd_vp contains | | 791 | * we want to enable swapping on sdp. the swd_vp contains |
788 | * the vnode we want (locked and ref'd), and the swd_dev | | 792 | * the vnode we want (locked and ref'd), and the swd_dev |
789 | * contains the dev_t of the file, if it a block device. | | 793 | * contains the dev_t of the file, if it a block device. |
790 | */ | | 794 | */ |
791 | | | 795 | |
792 | vp = sdp->swd_vp; | | 796 | vp = sdp->swd_vp; |
793 | dev = sdp->swd_dev; | | 797 | dev = sdp->swd_dev; |
794 | | | 798 | |
795 | /* | | 799 | /* |
796 | * open the swap file (mostly useful for block device files to | | 800 | * open the swap file (mostly useful for block device files to |
797 | * let device driver know what is up). | | 801 | * let device driver know what is up). |
798 | * | | 802 | * |
799 | * we skip the open/close for root on swap because the root | | 803 | * we skip the open/close for root on swap because the root |
800 | * has already been opened when root was mounted (mountroot). | | 804 | * has already been opened when root was mounted (mountroot). |
801 | */ | | 805 | */ |
802 | if (vp != rootvp) { | | 806 | if (vp != rootvp) { |
803 | if ((error = VOP_OPEN(vp, FREAD|FWRITE, l->l_cred))) | | 807 | if ((error = VOP_OPEN(vp, FREAD|FWRITE, l->l_cred))) |
804 | return (error); | | 808 | return (error); |
805 | } | | 809 | } |
806 | | | 810 | |
807 | /* XXX this only works for block devices */ | | 811 | /* XXX this only works for block devices */ |
808 | UVMHIST_LOG(pdhist, " dev=%d, major(dev)=%d", dev, major(dev), 0,0); | | 812 | UVMHIST_LOG(pdhist, " dev=%d, major(dev)=%d", dev, major(dev), 0,0); |
809 | | | 813 | |
810 | /* | | 814 | /* |
811 | * we now need to determine the size of the swap area. for | | 815 | * we now need to determine the size of the swap area. for |
812 | * block specials we can call the d_psize function. | | 816 | * block specials we can call the d_psize function. |
813 | * for normal files, we must stat [get attrs]. | | 817 | * for normal files, we must stat [get attrs]. |
814 | * | | 818 | * |
815 | * we put the result in nblks. | | 819 | * we put the result in nblks. |
816 | * for normal files, we also want the filesystem block size | | 820 | * for normal files, we also want the filesystem block size |
817 | * (which we get with statfs). | | 821 | * (which we get with statfs). |
818 | */ | | 822 | */ |
819 | switch (vp->v_type) { | | 823 | switch (vp->v_type) { |
820 | case VBLK: | | 824 | case VBLK: |
821 | bdev = bdevsw_lookup(dev); | | 825 | bdev = bdevsw_lookup(dev); |
822 | if (bdev == NULL || bdev->d_psize == NULL || | | 826 | if (bdev == NULL || bdev->d_psize == NULL || |
823 | (nblocks = (*bdev->d_psize)(dev)) == -1) { | | 827 | (nblocks = (*bdev->d_psize)(dev)) == -1) { |
824 | error = ENXIO; | | 828 | error = ENXIO; |
825 | goto bad; | | 829 | goto bad; |
826 | } | | 830 | } |
827 | break; | | 831 | break; |
828 | | | 832 | |
829 | case VREG: | | 833 | case VREG: |
830 | if ((error = VOP_GETATTR(vp, &va, l->l_cred))) | | 834 | if ((error = VOP_GETATTR(vp, &va, l->l_cred))) |
831 | goto bad; | | 835 | goto bad; |
832 | nblocks = (int)btodb(va.va_size); | | 836 | nblocks = (int)btodb(va.va_size); |
833 | if ((error = | | 837 | if ((error = |
834 | VFS_STATVFS(vp->v_mount, &vp->v_mount->mnt_stat)) != 0) | | 838 | VFS_STATVFS(vp->v_mount, &vp->v_mount->mnt_stat)) != 0) |
835 | goto bad; | | 839 | goto bad; |
836 | | | 840 | |
837 | sdp->swd_bsize = vp->v_mount->mnt_stat.f_iosize; | | 841 | sdp->swd_bsize = vp->v_mount->mnt_stat.f_iosize; |
838 | /* | | 842 | /* |
839 | * limit the max # of outstanding I/O requests we issue | | 843 | * limit the max # of outstanding I/O requests we issue |
840 | * at any one time. take it easy on NFS servers. | | 844 | * at any one time. take it easy on NFS servers. |
841 | */ | | 845 | */ |
842 | #ifdef NFS | | 846 | #ifdef NFS |
843 | if (vp->v_op == nfsv2_vnodeop_p) | | 847 | if (vp->v_op == nfsv2_vnodeop_p) |
844 | sdp->swd_maxactive = 2; /* XXX */ | | 848 | sdp->swd_maxactive = 2; /* XXX */ |
845 | else | | 849 | else |
846 | #endif /* NFS */ | | 850 | #endif /* NFS */ |
847 | sdp->swd_maxactive = 8; /* XXX */ | | 851 | sdp->swd_maxactive = 8; /* XXX */ |
848 | break; | | 852 | break; |
849 | | | 853 | |
850 | default: | | 854 | default: |
851 | error = ENXIO; | | 855 | error = ENXIO; |
852 | goto bad; | | 856 | goto bad; |
853 | } | | 857 | } |
854 | | | 858 | |
855 | /* | | 859 | /* |
856 | * save nblocks in a safe place and convert to pages. | | 860 | * save nblocks in a safe place and convert to pages. |
857 | */ | | 861 | */ |
858 | | | 862 | |
859 | sdp->swd_ose.ose_nblks = nblocks; | | 863 | sdp->swd_ose.ose_nblks = nblocks; |
860 | npages = dbtob((uint64_t)nblocks) >> PAGE_SHIFT; | | 864 | npages = dbtob((uint64_t)nblocks) >> PAGE_SHIFT; |
861 | | | 865 | |
862 | /* | | 866 | /* |
863 | * for block special files, we want to make sure that leave | | 867 | * for block special files, we want to make sure that leave |
864 | * the disklabel and bootblocks alone, so we arrange to skip | | 868 | * the disklabel and bootblocks alone, so we arrange to skip |
865 | * over them (arbitrarily choosing to skip PAGE_SIZE bytes). | | 869 | * over them (arbitrarily choosing to skip PAGE_SIZE bytes). |
866 | * note that because of this the "size" can be less than the | | 870 | * note that because of this the "size" can be less than the |
867 | * actual number of blocks on the device. | | 871 | * actual number of blocks on the device. |
868 | */ | | 872 | */ |
869 | if (vp->v_type == VBLK) { | | 873 | if (vp->v_type == VBLK) { |
870 | /* we use pages 1 to (size - 1) [inclusive] */ | | 874 | /* we use pages 1 to (size - 1) [inclusive] */ |
871 | size = npages - 1; | | 875 | size = npages - 1; |
872 | addr = 1; | | 876 | addr = 1; |
873 | } else { | | 877 | } else { |
874 | /* we use pages 0 to (size - 1) [inclusive] */ | | 878 | /* we use pages 0 to (size - 1) [inclusive] */ |
875 | size = npages; | | 879 | size = npages; |
876 | addr = 0; | | 880 | addr = 0; |
877 | } | | 881 | } |
878 | | | 882 | |
879 | /* | | 883 | /* |
880 | * make sure we have enough blocks for a reasonable sized swap | | 884 | * make sure we have enough blocks for a reasonable sized swap |
881 | * area. we want at least one page. | | 885 | * area. we want at least one page. |
882 | */ | | 886 | */ |
883 | | | 887 | |
884 | if (size < 1) { | | 888 | if (size < 1) { |
885 | UVMHIST_LOG(pdhist, " size <= 1!!", 0, 0, 0, 0); | | 889 | UVMHIST_LOG(pdhist, " size <= 1!!", 0, 0, 0, 0); |
886 | error = EINVAL; | | 890 | error = EINVAL; |
887 | goto bad; | | 891 | goto bad; |
888 | } | | 892 | } |
889 | | | 893 | |
890 | UVMHIST_LOG(pdhist, " dev=%x: size=%d addr=%ld\n", dev, size, addr, 0); | | 894 | UVMHIST_LOG(pdhist, " dev=%x: size=%d addr=%ld\n", dev, size, addr, 0); |
891 | | | 895 | |
892 | /* | | 896 | /* |
893 | * now we need to allocate an extent to manage this swap device | | 897 | * now we need to allocate an extent to manage this swap device |
894 | */ | | 898 | */ |
895 | | | 899 | |
896 | sdp->swd_blist = blist_create(npages); | | 900 | sdp->swd_blist = blist_create(npages); |
897 | /* mark all expect the `saved' region free. */ | | 901 | /* mark all expect the `saved' region free. */ |
898 | blist_free(sdp->swd_blist, addr, size); | | 902 | blist_free(sdp->swd_blist, addr, size); |
899 | | | 903 | |
900 | /* | | 904 | /* |
901 | * if the vnode we are swapping to is the root vnode | | 905 | * if the vnode we are swapping to is the root vnode |
902 | * (i.e. we are swapping to the miniroot) then we want | | 906 | * (i.e. we are swapping to the miniroot) then we want |
903 | * to make sure we don't overwrite it. do a statfs to | | 907 | * to make sure we don't overwrite it. do a statfs to |
904 | * find its size and skip over it. | | 908 | * find its size and skip over it. |
905 | */ | | 909 | */ |
906 | if (vp == rootvp) { | | 910 | if (vp == rootvp) { |
907 | struct mount *mp; | | 911 | struct mount *mp; |
908 | struct statvfs *sp; | | 912 | struct statvfs *sp; |
909 | int rootblocks, rootpages; | | 913 | int rootblocks, rootpages; |
910 | | | 914 | |
911 | mp = rootvnode->v_mount; | | 915 | mp = rootvnode->v_mount; |
912 | sp = &mp->mnt_stat; | | 916 | sp = &mp->mnt_stat; |
913 | rootblocks = sp->f_blocks * btodb(sp->f_frsize); | | 917 | rootblocks = sp->f_blocks * btodb(sp->f_frsize); |
914 | /* | | 918 | /* |
915 | * XXX: sp->f_blocks isn't the total number of | | 919 | * XXX: sp->f_blocks isn't the total number of |
916 | * blocks in the filesystem, it's the number of | | 920 | * blocks in the filesystem, it's the number of |
917 | * data blocks. so, our rootblocks almost | | 921 | * data blocks. so, our rootblocks almost |
918 | * definitely underestimates the total size | | 922 | * definitely underestimates the total size |
919 | * of the filesystem - how badly depends on the | | 923 | * of the filesystem - how badly depends on the |
920 | * details of the filesystem type. there isn't | | 924 | * details of the filesystem type. there isn't |
921 | * an obvious way to deal with this cleanly | | 925 | * an obvious way to deal with this cleanly |
922 | * and perfectly, so for now we just pad our | | 926 | * and perfectly, so for now we just pad our |
923 | * rootblocks estimate with an extra 5 percent. | | 927 | * rootblocks estimate with an extra 5 percent. |
924 | */ | | 928 | */ |
925 | rootblocks += (rootblocks >> 5) + | | 929 | rootblocks += (rootblocks >> 5) + |
926 | (rootblocks >> 6) + | | 930 | (rootblocks >> 6) + |
927 | (rootblocks >> 7); | | 931 | (rootblocks >> 7); |
928 | rootpages = round_page(dbtob(rootblocks)) >> PAGE_SHIFT; | | 932 | rootpages = round_page(dbtob(rootblocks)) >> PAGE_SHIFT; |
929 | if (rootpages > size) | | 933 | if (rootpages > size) |
930 | panic("swap_on: miniroot larger than swap?"); | | 934 | panic("swap_on: miniroot larger than swap?"); |
931 | | | 935 | |
932 | if (rootpages != blist_fill(sdp->swd_blist, addr, rootpages)) { | | 936 | if (rootpages != blist_fill(sdp->swd_blist, addr, rootpages)) { |
933 | panic("swap_on: unable to preserve miniroot"); | | 937 | panic("swap_on: unable to preserve miniroot"); |
934 | } | | 938 | } |
935 | | | 939 | |
936 | size -= rootpages; | | 940 | size -= rootpages; |
937 | printf("Preserved %d pages of miniroot ", rootpages); | | 941 | printf("Preserved %d pages of miniroot ", rootpages); |
938 | printf("leaving %d pages of swap\n", size); | | 942 | printf("leaving %d pages of swap\n", size); |
939 | } | | 943 | } |
940 | | | 944 | |
941 | /* | | 945 | /* |
942 | * add a ref to vp to reflect usage as a swap device. | | 946 | * add a ref to vp to reflect usage as a swap device. |
943 | */ | | 947 | */ |
944 | vref(vp); | | 948 | vref(vp); |
945 | | | 949 | |
946 | /* | | 950 | /* |
947 | * now add the new swapdev to the drum and enable. | | 951 | * now add the new swapdev to the drum and enable. |
948 | */ | | 952 | */ |
949 | result = vmem_alloc(swapmap, npages, VM_BESTFIT | VM_SLEEP); | | 953 | result = vmem_alloc(swapmap, npages, VM_BESTFIT | VM_SLEEP); |
950 | if (result == 0) | | 954 | if (result == 0) |
951 | panic("swapdrum_add"); | | 955 | panic("swapdrum_add"); |
952 | /* | | 956 | /* |
953 | * If this is the first regular swap create the workqueue. | | 957 | * If this is the first regular swap create the workqueue. |
954 | * => Protected by swap_syscall_lock. | | 958 | * => Protected by swap_syscall_lock. |
955 | */ | | 959 | */ |
956 | if (vp->v_type != VBLK) { | | 960 | if (vp->v_type != VBLK) { |
957 | if (sw_reg_count++ == 0) { | | 961 | if (sw_reg_count++ == 0) { |
958 | KASSERT(sw_reg_workqueue == NULL); | | 962 | KASSERT(sw_reg_workqueue == NULL); |
959 | if (workqueue_create(&sw_reg_workqueue, "swapiod", | | 963 | if (workqueue_create(&sw_reg_workqueue, "swapiod", |
960 | sw_reg_iodone, NULL, PRIBIO, IPL_BIO, 0) != 0) | | 964 | sw_reg_iodone, NULL, PRIBIO, IPL_BIO, 0) != 0) |
961 | panic("swap_add: workqueue_create failed"); | | 965 | panic("swap_add: workqueue_create failed"); |
962 | } | | 966 | } |
963 | } | | 967 | } |
964 | | | 968 | |
965 | sdp->swd_drumoffset = (int)result; | | 969 | sdp->swd_drumoffset = (int)result; |
966 | sdp->swd_drumsize = npages; | | 970 | sdp->swd_drumsize = npages; |
967 | sdp->swd_npages = size; | | 971 | sdp->swd_npages = size; |
968 | mutex_enter(&uvm_swap_data_lock); | | 972 | mutex_enter(&uvm_swap_data_lock); |
969 | sdp->swd_flags &= ~SWF_FAKE; /* going live */ | | 973 | sdp->swd_flags &= ~SWF_FAKE; /* going live */ |
970 | sdp->swd_flags |= (SWF_INUSE|SWF_ENABLE); | | 974 | sdp->swd_flags |= (SWF_INUSE|SWF_ENABLE); |
971 | uvmexp.swpages += size; | | 975 | uvmexp.swpages += size; |
972 | uvmexp.swpgavail += size; | | 976 | uvmexp.swpgavail += size; |
973 | mutex_exit(&uvm_swap_data_lock); | | 977 | mutex_exit(&uvm_swap_data_lock); |
974 | return (0); | | 978 | return (0); |
975 | | | 979 | |
976 | /* | | 980 | /* |
977 | * failure: clean up and return error. | | 981 | * failure: clean up and return error. |
978 | */ | | 982 | */ |
979 | | | 983 | |
980 | bad: | | 984 | bad: |
981 | if (sdp->swd_blist) { | | 985 | if (sdp->swd_blist) { |
982 | blist_destroy(sdp->swd_blist); | | 986 | blist_destroy(sdp->swd_blist); |
983 | } | | 987 | } |
984 | if (vp != rootvp) { | | 988 | if (vp != rootvp) { |
985 | (void)VOP_CLOSE(vp, FREAD|FWRITE, l->l_cred); | | 989 | (void)VOP_CLOSE(vp, FREAD|FWRITE, l->l_cred); |
986 | } | | 990 | } |
987 | return (error); | | 991 | return (error); |
988 | } | | 992 | } |
989 | | | 993 | |
990 | /* | | 994 | /* |
991 | * swap_off: stop swapping on swapdev | | 995 | * swap_off: stop swapping on swapdev |
992 | * | | 996 | * |
993 | * => swap data should be locked, we will unlock. | | 997 | * => swap data should be locked, we will unlock. |
994 | */ | | 998 | */ |
995 | static int | | 999 | static int |
996 | swap_off(struct lwp *l, struct swapdev *sdp) | | 1000 | swap_off(struct lwp *l, struct swapdev *sdp) |
997 | { | | 1001 | { |
998 | int npages = sdp->swd_npages; | | 1002 | int npages = sdp->swd_npages; |
999 | int error = 0; | | 1003 | int error = 0; |
1000 | | | 1004 | |
1001 | UVMHIST_FUNC("swap_off"); UVMHIST_CALLED(pdhist); | | 1005 | UVMHIST_FUNC("swap_off"); UVMHIST_CALLED(pdhist); |
1002 | UVMHIST_LOG(pdhist, " dev=%x, npages=%d", sdp->swd_dev,npages,0,0); | | 1006 | UVMHIST_LOG(pdhist, " dev=%x, npages=%d", sdp->swd_dev,npages,0,0); |
1003 | | | 1007 | |
1004 | /* disable the swap area being removed */ | | 1008 | /* disable the swap area being removed */ |
1005 | sdp->swd_flags &= ~SWF_ENABLE; | | 1009 | sdp->swd_flags &= ~SWF_ENABLE; |
1006 | uvmexp.swpgavail -= npages; | | 1010 | uvmexp.swpgavail -= npages; |
1007 | mutex_exit(&uvm_swap_data_lock); | | 1011 | mutex_exit(&uvm_swap_data_lock); |
1008 | | | 1012 | |
1009 | /* | | 1013 | /* |
1010 | * the idea is to find all the pages that are paged out to this | | 1014 | * the idea is to find all the pages that are paged out to this |
1011 | * device, and page them all in. in uvm, swap-backed pageable | | 1015 | * device, and page them all in. in uvm, swap-backed pageable |
1012 | * memory can take two forms: aobjs and anons. call the | | 1016 | * memory can take two forms: aobjs and anons. call the |
1013 | * swapoff hook for each subsystem to bring in pages. | | 1017 | * swapoff hook for each subsystem to bring in pages. |
1014 | */ | | 1018 | */ |
1015 | | | 1019 | |
1016 | if (uao_swap_off(sdp->swd_drumoffset, | | 1020 | if (uao_swap_off(sdp->swd_drumoffset, |
1017 | sdp->swd_drumoffset + sdp->swd_drumsize) || | | 1021 | sdp->swd_drumoffset + sdp->swd_drumsize) || |
1018 | amap_swap_off(sdp->swd_drumoffset, | | 1022 | amap_swap_off(sdp->swd_drumoffset, |
1019 | sdp->swd_drumoffset + sdp->swd_drumsize)) { | | 1023 | sdp->swd_drumoffset + sdp->swd_drumsize)) { |
1020 | error = ENOMEM; | | 1024 | error = ENOMEM; |
1021 | } else if (sdp->swd_npginuse > sdp->swd_npgbad) { | | 1025 | } else if (sdp->swd_npginuse > sdp->swd_npgbad) { |
1022 | error = EBUSY; | | 1026 | error = EBUSY; |
1023 | } | | 1027 | } |
1024 | | | 1028 | |
1025 | if (error) { | | 1029 | if (error) { |
1026 | mutex_enter(&uvm_swap_data_lock); | | 1030 | mutex_enter(&uvm_swap_data_lock); |
1027 | sdp->swd_flags |= SWF_ENABLE; | | 1031 | sdp->swd_flags |= SWF_ENABLE; |
1028 | uvmexp.swpgavail += npages; | | 1032 | uvmexp.swpgavail += npages; |
1029 | mutex_exit(&uvm_swap_data_lock); | | 1033 | mutex_exit(&uvm_swap_data_lock); |
1030 | | | 1034 | |
1031 | return error; | | 1035 | return error; |
1032 | } | | 1036 | } |
1033 | | | 1037 | |
1034 | /* | | 1038 | /* |
1035 | * If this is the last regular swap destroy the workqueue. | | 1039 | * If this is the last regular swap destroy the workqueue. |
1036 | * => Protected by swap_syscall_lock. | | 1040 | * => Protected by swap_syscall_lock. |
1037 | */ | | 1041 | */ |
1038 | if (sdp->swd_vp->v_type != VBLK) { | | 1042 | if (sdp->swd_vp->v_type != VBLK) { |
1039 | KASSERT(sw_reg_count > 0); | | 1043 | KASSERT(sw_reg_count > 0); |
1040 | KASSERT(sw_reg_workqueue != NULL); | | 1044 | KASSERT(sw_reg_workqueue != NULL); |
1041 | if (--sw_reg_count == 0) { | | 1045 | if (--sw_reg_count == 0) { |
1042 | workqueue_destroy(sw_reg_workqueue); | | 1046 | workqueue_destroy(sw_reg_workqueue); |
1043 | sw_reg_workqueue = NULL; | | 1047 | sw_reg_workqueue = NULL; |
1044 | } | | 1048 | } |
1045 | } | | 1049 | } |
1046 | | | 1050 | |
1047 | /* | | 1051 | /* |
1048 | * done with the vnode. | | 1052 | * done with the vnode. |
1049 | * drop our ref on the vnode before calling VOP_CLOSE() | | 1053 | * drop our ref on the vnode before calling VOP_CLOSE() |
1050 | * so that spec_close() can tell if this is the last close. | | 1054 | * so that spec_close() can tell if this is the last close. |
1051 | */ | | 1055 | */ |
1052 | vrele(sdp->swd_vp); | | 1056 | vrele(sdp->swd_vp); |
1053 | if (sdp->swd_vp != rootvp) { | | 1057 | if (sdp->swd_vp != rootvp) { |
1054 | (void) VOP_CLOSE(sdp->swd_vp, FREAD|FWRITE, l->l_cred); | | 1058 | (void) VOP_CLOSE(sdp->swd_vp, FREAD|FWRITE, l->l_cred); |
1055 | } | | 1059 | } |
1056 | | | 1060 | |
1057 | mutex_enter(&uvm_swap_data_lock); | | 1061 | mutex_enter(&uvm_swap_data_lock); |
1058 | uvmexp.swpages -= npages; | | 1062 | uvmexp.swpages -= npages; |
1059 | uvmexp.swpginuse -= sdp->swd_npgbad; | | 1063 | uvmexp.swpginuse -= sdp->swd_npgbad; |
1060 | | | 1064 | |
1061 | if (swaplist_find(sdp->swd_vp, true) == NULL) | | 1065 | if (swaplist_find(sdp->swd_vp, true) == NULL) |
1062 | panic("swap_off: swapdev not in list"); | | 1066 | panic("swap_off: swapdev not in list"); |
1063 | swaplist_trim(); | | 1067 | swaplist_trim(); |
1064 | mutex_exit(&uvm_swap_data_lock); | | 1068 | mutex_exit(&uvm_swap_data_lock); |
1065 | | | 1069 | |
1066 | /* | | 1070 | /* |
1067 | * free all resources! | | 1071 | * free all resources! |
1068 | */ | | 1072 | */ |
1069 | vmem_free(swapmap, sdp->swd_drumoffset, sdp->swd_drumsize); | | 1073 | vmem_free(swapmap, sdp->swd_drumoffset, sdp->swd_drumsize); |
1070 | blist_destroy(sdp->swd_blist); | | 1074 | blist_destroy(sdp->swd_blist); |
1071 | bufq_free(sdp->swd_tab); | | 1075 | bufq_free(sdp->swd_tab); |
1072 | free(sdp, M_VMSWAP); | | 1076 | free(sdp, M_VMSWAP); |
1073 | return (0); | | 1077 | return (0); |
1074 | } | | 1078 | } |
1075 | | | 1079 | |
1076 | /* | | 1080 | /* |
1077 | * /dev/drum interface and i/o functions | | 1081 | * /dev/drum interface and i/o functions |
1078 | */ | | 1082 | */ |
1079 | | | 1083 | |
1080 | /* | | 1084 | /* |
1081 | * swstrategy: perform I/O on the drum | | 1085 | * swstrategy: perform I/O on the drum |
1082 | * | | 1086 | * |
1083 | * => we must map the i/o request from the drum to the correct swapdev. | | 1087 | * => we must map the i/o request from the drum to the correct swapdev. |
1084 | */ | | 1088 | */ |
1085 | static void | | 1089 | static void |
1086 | swstrategy(struct buf *bp) | | 1090 | swstrategy(struct buf *bp) |
1087 | { | | 1091 | { |
1088 | struct swapdev *sdp; | | 1092 | struct swapdev *sdp; |
1089 | struct vnode *vp; | | 1093 | struct vnode *vp; |
1090 | int pageno, bn; | | 1094 | int pageno, bn; |
1091 | UVMHIST_FUNC("swstrategy"); UVMHIST_CALLED(pdhist); | | 1095 | UVMHIST_FUNC("swstrategy"); UVMHIST_CALLED(pdhist); |
1092 | | | 1096 | |
1093 | /* | | 1097 | /* |
1094 | * convert block number to swapdev. note that swapdev can't | | 1098 | * convert block number to swapdev. note that swapdev can't |
1095 | * be yanked out from under us because we are holding resources | | 1099 | * be yanked out from under us because we are holding resources |
1096 | * in it (i.e. the blocks we are doing I/O on). | | 1100 | * in it (i.e. the blocks we are doing I/O on). |
1097 | */ | | 1101 | */ |
1098 | pageno = dbtob((int64_t)bp->b_blkno) >> PAGE_SHIFT; | | 1102 | pageno = dbtob((int64_t)bp->b_blkno) >> PAGE_SHIFT; |
1099 | mutex_enter(&uvm_swap_data_lock); | | 1103 | mutex_enter(&uvm_swap_data_lock); |
1100 | sdp = swapdrum_getsdp(pageno); | | 1104 | sdp = swapdrum_getsdp(pageno); |
1101 | mutex_exit(&uvm_swap_data_lock); | | 1105 | mutex_exit(&uvm_swap_data_lock); |
1102 | if (sdp == NULL) { | | 1106 | if (sdp == NULL) { |
1103 | bp->b_error = EINVAL; | | 1107 | bp->b_error = EINVAL; |
1104 | biodone(bp); | | 1108 | biodone(bp); |
1105 | UVMHIST_LOG(pdhist, " failed to get swap device", 0, 0, 0, 0); | | 1109 | UVMHIST_LOG(pdhist, " failed to get swap device", 0, 0, 0, 0); |
1106 | return; | | 1110 | return; |
1107 | } | | 1111 | } |
1108 | | | 1112 | |
1109 | /* | | 1113 | /* |
1110 | * convert drum page number to block number on this swapdev. | | 1114 | * convert drum page number to block number on this swapdev. |
1111 | */ | | 1115 | */ |
1112 | | | 1116 | |
1113 | pageno -= sdp->swd_drumoffset; /* page # on swapdev */ | | 1117 | pageno -= sdp->swd_drumoffset; /* page # on swapdev */ |
1114 | bn = btodb((uint64_t)pageno << PAGE_SHIFT); /* convert to diskblock */ | | 1118 | bn = btodb((uint64_t)pageno << PAGE_SHIFT); /* convert to diskblock */ |
1115 | | | 1119 | |
1116 | UVMHIST_LOG(pdhist, " %s: mapoff=%x bn=%x bcount=%ld", | | 1120 | UVMHIST_LOG(pdhist, " %s: mapoff=%x bn=%x bcount=%ld", |
1117 | ((bp->b_flags & B_READ) == 0) ? "write" : "read", | | 1121 | ((bp->b_flags & B_READ) == 0) ? "write" : "read", |
1118 | sdp->swd_drumoffset, bn, bp->b_bcount); | | 1122 | sdp->swd_drumoffset, bn, bp->b_bcount); |
1119 | | | 1123 | |
1120 | /* | | 1124 | /* |
1121 | * for block devices we finish up here. | | 1125 | * for block devices we finish up here. |
1122 | * for regular files we have to do more work which we delegate | | 1126 | * for regular files we have to do more work which we delegate |
1123 | * to sw_reg_strategy(). | | 1127 | * to sw_reg_strategy(). |
1124 | */ | | 1128 | */ |
1125 | | | 1129 | |
1126 | vp = sdp->swd_vp; /* swapdev vnode pointer */ | | 1130 | vp = sdp->swd_vp; /* swapdev vnode pointer */ |
1127 | switch (vp->v_type) { | | 1131 | switch (vp->v_type) { |
1128 | default: | | 1132 | default: |
1129 | panic("swstrategy: vnode type 0x%x", vp->v_type); | | 1133 | panic("swstrategy: vnode type 0x%x", vp->v_type); |
1130 | | | 1134 | |
1131 | case VBLK: | | 1135 | case VBLK: |
1132 | | | 1136 | |
1133 | /* | | 1137 | /* |
1134 | * must convert "bp" from an I/O on /dev/drum to an I/O | | 1138 | * must convert "bp" from an I/O on /dev/drum to an I/O |
1135 | * on the swapdev (sdp). | | 1139 | * on the swapdev (sdp). |
1136 | */ | | 1140 | */ |
1137 | bp->b_blkno = bn; /* swapdev block number */ | | 1141 | bp->b_blkno = bn; /* swapdev block number */ |
1138 | bp->b_dev = sdp->swd_dev; /* swapdev dev_t */ | | 1142 | bp->b_dev = sdp->swd_dev; /* swapdev dev_t */ |
1139 | | | 1143 | |
1140 | /* | | 1144 | /* |
1141 | * if we are doing a write, we have to redirect the i/o on | | 1145 | * if we are doing a write, we have to redirect the i/o on |
1142 | * drum's v_numoutput counter to the swapdevs. | | 1146 | * drum's v_numoutput counter to the swapdevs. |
1143 | */ | | 1147 | */ |
1144 | if ((bp->b_flags & B_READ) == 0) { | | 1148 | if ((bp->b_flags & B_READ) == 0) { |
1145 | mutex_enter(bp->b_objlock); | | 1149 | mutex_enter(bp->b_objlock); |
1146 | vwakeup(bp); /* kills one 'v_numoutput' on drum */ | | 1150 | vwakeup(bp); /* kills one 'v_numoutput' on drum */ |
1147 | mutex_exit(bp->b_objlock); | | 1151 | mutex_exit(bp->b_objlock); |
1148 | mutex_enter(&vp->v_interlock); | | 1152 | mutex_enter(&vp->v_interlock); |
1149 | vp->v_numoutput++; /* put it on swapdev */ | | 1153 | vp->v_numoutput++; /* put it on swapdev */ |
1150 | mutex_exit(&vp->v_interlock); | | 1154 | mutex_exit(&vp->v_interlock); |
1151 | } | | 1155 | } |
1152 | | | 1156 | |
1153 | /* | | 1157 | /* |
1154 | * finally plug in swapdev vnode and start I/O | | 1158 | * finally plug in swapdev vnode and start I/O |
1155 | */ | | 1159 | */ |
1156 | bp->b_vp = vp; | | 1160 | bp->b_vp = vp; |
1157 | bp->b_objlock = &vp->v_interlock; | | 1161 | bp->b_objlock = &vp->v_interlock; |
1158 | VOP_STRATEGY(vp, bp); | | 1162 | VOP_STRATEGY(vp, bp); |
1159 | return; | | 1163 | return; |
1160 | | | 1164 | |
1161 | case VREG: | | 1165 | case VREG: |
1162 | /* | | 1166 | /* |
1163 | * delegate to sw_reg_strategy function. | | 1167 | * delegate to sw_reg_strategy function. |
1164 | */ | | 1168 | */ |
1165 | sw_reg_strategy(sdp, bp, bn); | | 1169 | sw_reg_strategy(sdp, bp, bn); |
1166 | return; | | 1170 | return; |
1167 | } | | 1171 | } |
1168 | /* NOTREACHED */ | | 1172 | /* NOTREACHED */ |
1169 | } | | 1173 | } |
1170 | | | 1174 | |
1171 | /* | | 1175 | /* |
1172 | * swread: the read function for the drum (just a call to physio) | | 1176 | * swread: the read function for the drum (just a call to physio) |
1173 | */ | | 1177 | */ |
1174 | /*ARGSUSED*/ | | 1178 | /*ARGSUSED*/ |
1175 | static int | | 1179 | static int |
1176 | swread(dev_t dev, struct uio *uio, int ioflag) | | 1180 | swread(dev_t dev, struct uio *uio, int ioflag) |
1177 | { | | 1181 | { |
1178 | UVMHIST_FUNC("swread"); UVMHIST_CALLED(pdhist); | | 1182 | UVMHIST_FUNC("swread"); UVMHIST_CALLED(pdhist); |
1179 | | | 1183 | |
1180 | UVMHIST_LOG(pdhist, " dev=%x offset=%qx", dev, uio->uio_offset, 0, 0); | | 1184 | UVMHIST_LOG(pdhist, " dev=%x offset=%qx", dev, uio->uio_offset, 0, 0); |
1181 | return (physio(swstrategy, NULL, dev, B_READ, minphys, uio)); | | 1185 | return (physio(swstrategy, NULL, dev, B_READ, minphys, uio)); |
1182 | } | | 1186 | } |
1183 | | | 1187 | |
1184 | /* | | 1188 | /* |
1185 | * swwrite: the write function for the drum (just a call to physio) | | 1189 | * swwrite: the write function for the drum (just a call to physio) |
1186 | */ | | 1190 | */ |
1187 | /*ARGSUSED*/ | | 1191 | /*ARGSUSED*/ |
1188 | static int | | 1192 | static int |
1189 | swwrite(dev_t dev, struct uio *uio, int ioflag) | | 1193 | swwrite(dev_t dev, struct uio *uio, int ioflag) |
1190 | { | | 1194 | { |
1191 | UVMHIST_FUNC("swwrite"); UVMHIST_CALLED(pdhist); | | 1195 | UVMHIST_FUNC("swwrite"); UVMHIST_CALLED(pdhist); |
1192 | | | 1196 | |
1193 | UVMHIST_LOG(pdhist, " dev=%x offset=%qx", dev, uio->uio_offset, 0, 0); | | 1197 | UVMHIST_LOG(pdhist, " dev=%x offset=%qx", dev, uio->uio_offset, 0, 0); |
1194 | return (physio(swstrategy, NULL, dev, B_WRITE, minphys, uio)); | | 1198 | return (physio(swstrategy, NULL, dev, B_WRITE, minphys, uio)); |
1195 | } | | 1199 | } |
1196 | | | 1200 | |
1197 | const struct bdevsw swap_bdevsw = { | | 1201 | const struct bdevsw swap_bdevsw = { |
1198 | nullopen, nullclose, swstrategy, noioctl, nodump, nosize, D_OTHER, | | 1202 | nullopen, nullclose, swstrategy, noioctl, nodump, nosize, D_OTHER, |
1199 | }; | | 1203 | }; |
1200 | | | 1204 | |
1201 | const struct cdevsw swap_cdevsw = { | | 1205 | const struct cdevsw swap_cdevsw = { |
1202 | nullopen, nullclose, swread, swwrite, noioctl, | | 1206 | nullopen, nullclose, swread, swwrite, noioctl, |
1203 | nostop, notty, nopoll, nommap, nokqfilter, D_OTHER, | | 1207 | nostop, notty, nopoll, nommap, nokqfilter, D_OTHER, |
1204 | }; | | 1208 | }; |
1205 | | | 1209 | |
1206 | /* | | 1210 | /* |
1207 | * sw_reg_strategy: handle swap i/o to regular files | | 1211 | * sw_reg_strategy: handle swap i/o to regular files |
1208 | */ | | 1212 | */ |
1209 | static void | | 1213 | static void |
1210 | sw_reg_strategy(struct swapdev *sdp, struct buf *bp, int bn) | | 1214 | sw_reg_strategy(struct swapdev *sdp, struct buf *bp, int bn) |
1211 | { | | 1215 | { |
1212 | struct vnode *vp; | | 1216 | struct vnode *vp; |
1213 | struct vndxfer *vnx; | | 1217 | struct vndxfer *vnx; |
1214 | daddr_t nbn; | | 1218 | daddr_t nbn; |
1215 | char *addr; | | 1219 | char *addr; |
1216 | off_t byteoff; | | 1220 | off_t byteoff; |
1217 | int s, off, nra, error, sz, resid; | | 1221 | int s, off, nra, error, sz, resid; |
1218 | UVMHIST_FUNC("sw_reg_strategy"); UVMHIST_CALLED(pdhist); | | 1222 | UVMHIST_FUNC("sw_reg_strategy"); UVMHIST_CALLED(pdhist); |
1219 | | | 1223 | |
1220 | /* | | 1224 | /* |
1221 | * allocate a vndxfer head for this transfer and point it to | | 1225 | * allocate a vndxfer head for this transfer and point it to |
1222 | * our buffer. | | 1226 | * our buffer. |
1223 | */ | | 1227 | */ |
1224 | vnx = pool_get(&vndxfer_pool, PR_WAITOK); | | 1228 | vnx = pool_get(&vndxfer_pool, PR_WAITOK); |
1225 | vnx->vx_flags = VX_BUSY; | | 1229 | vnx->vx_flags = VX_BUSY; |
1226 | vnx->vx_error = 0; | | 1230 | vnx->vx_error = 0; |
1227 | vnx->vx_pending = 0; | | 1231 | vnx->vx_pending = 0; |
1228 | vnx->vx_bp = bp; | | 1232 | vnx->vx_bp = bp; |
1229 | vnx->vx_sdp = sdp; | | 1233 | vnx->vx_sdp = sdp; |
1230 | | | 1234 | |
1231 | /* | | 1235 | /* |
1232 | * setup for main loop where we read filesystem blocks into | | 1236 | * setup for main loop where we read filesystem blocks into |
1233 | * our buffer. | | 1237 | * our buffer. |
1234 | */ | | 1238 | */ |
1235 | error = 0; | | 1239 | error = 0; |
1236 | bp->b_resid = bp->b_bcount; /* nothing transfered yet! */ | | 1240 | bp->b_resid = bp->b_bcount; /* nothing transfered yet! */ |
1237 | addr = bp->b_data; /* current position in buffer */ | | 1241 | addr = bp->b_data; /* current position in buffer */ |
1238 | byteoff = dbtob((uint64_t)bn); | | 1242 | byteoff = dbtob((uint64_t)bn); |
1239 | | | 1243 | |
1240 | for (resid = bp->b_resid; resid; resid -= sz) { | | 1244 | for (resid = bp->b_resid; resid; resid -= sz) { |
1241 | struct vndbuf *nbp; | | 1245 | struct vndbuf *nbp; |
1242 | | | 1246 | |
1243 | /* | | 1247 | /* |
1244 | * translate byteoffset into block number. return values: | | 1248 | * translate byteoffset into block number. return values: |
1245 | * vp = vnode of underlying device | | 1249 | * vp = vnode of underlying device |
1246 | * nbn = new block number (on underlying vnode dev) | | 1250 | * nbn = new block number (on underlying vnode dev) |
1247 | * nra = num blocks we can read-ahead (excludes requested | | 1251 | * nra = num blocks we can read-ahead (excludes requested |
1248 | * block) | | 1252 | * block) |
1249 | */ | | 1253 | */ |
1250 | nra = 0; | | 1254 | nra = 0; |
1251 | error = VOP_BMAP(sdp->swd_vp, byteoff / sdp->swd_bsize, | | 1255 | error = VOP_BMAP(sdp->swd_vp, byteoff / sdp->swd_bsize, |
1252 | &vp, &nbn, &nra); | | 1256 | &vp, &nbn, &nra); |
1253 | | | 1257 | |
1254 | if (error == 0 && nbn == (daddr_t)-1) { | | 1258 | if (error == 0 && nbn == (daddr_t)-1) { |
1255 | /* | | 1259 | /* |
1256 | * this used to just set error, but that doesn't | | 1260 | * this used to just set error, but that doesn't |
1257 | * do the right thing. Instead, it causes random | | 1261 | * do the right thing. Instead, it causes random |
1258 | * memory errors. The panic() should remain until | | 1262 | * memory errors. The panic() should remain until |
1259 | * this condition doesn't destabilize the system. | | 1263 | * this condition doesn't destabilize the system. |
1260 | */ | | 1264 | */ |
1261 | #if 1 | | 1265 | #if 1 |
1262 | panic("sw_reg_strategy: swap to sparse file"); | | 1266 | panic("sw_reg_strategy: swap to sparse file"); |
1263 | #else | | 1267 | #else |
1264 | error = EIO; /* failure */ | | 1268 | error = EIO; /* failure */ |
1265 | #endif | | 1269 | #endif |
1266 | } | | 1270 | } |
1267 | | | 1271 | |
1268 | /* | | 1272 | /* |
1269 | * punt if there was an error or a hole in the file. | | 1273 | * punt if there was an error or a hole in the file. |
1270 | * we must wait for any i/o ops we have already started | | 1274 | * we must wait for any i/o ops we have already started |
1271 | * to finish before returning. | | 1275 | * to finish before returning. |
1272 | * | | 1276 | * |
1273 | * XXX we could deal with holes here but it would be | | 1277 | * XXX we could deal with holes here but it would be |
1274 | * a hassle (in the write case). | | 1278 | * a hassle (in the write case). |
1275 | */ | | 1279 | */ |
1276 | if (error) { | | 1280 | if (error) { |
1277 | s = splbio(); | | 1281 | s = splbio(); |