kernel/fs/swapfs/swap_subr.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  23  * Use is subject to license terms.
  24  */
  25
  26 #include <sys/types.h>
  27 #include <sys/param.h>
  28 #include <sys/systm.h>
  29 #include <sys/errno.h>
  30 #include <sys/kmem.h>
  31 #include <sys/vnode.h>
  32 #include <sys/vfs.h>
  33 #include <sys/swap.h>
  34 #include <sys/sysmacros.h>
  35 #include <sys/buf.h>
  36 #include <sys/callb.h>
  37 #include <sys/debug.h>
  38 #include <vm/seg.h>
  39 #include <sys/fs/swapnode.h>
  40 #include <sys/fs_subr.h>
  41 #include <sys/cmn_err.h>
  42 #include <sys/mem_config.h>
  43 #include <sys/atomic.h>
  44
  45 /*
  46  * swapfs_minfree is the amount of physical memory (actually remaining
  47  * availrmem) that we want to keep free for the rest of the system.  This
  48  * means that swapfs can only grow to availrmem - swapfs_minfree.  This
  49  * can be set as just constant value or a certain percentage of installed
  50  * physical memory. It is set in swapinit().
  51  *
  52  * Users who want to change the amount of memory that can be used as swap
  53  * space should do so by setting swapfs_desfree at boot time,
  54  * not swapfs_minfree.
  55  */
  56
  57 pgcnt_t swapfs_desfree = 0;
  58 pgcnt_t swapfs_minfree = 0;
  59 pgcnt_t swapfs_reserve = 0;
  60
  61 #ifdef SWAPFS_DEBUG
  62 int swapfs_debug;
  63 #endif /* SWAPFS_DEBUG */
  64
  65
  66 static int swapfs_vpcount;
  67 static kmutex_t swapfs_lock;
  68 static struct async_reqs *sw_ar, *sw_pendlist, *sw_freelist;
  69
  70 static struct vnode **swap_vnodes;      /* ptr's to swap vnodes */
  71
  72 static void swap_init_mem_config(void);
  73
  74 static pgcnt_t initial_swapfs_desfree;
  75 static pgcnt_t initial_swapfs_minfree;
  76 static pgcnt_t initial_swapfs_reserve;
  77
  78 static int swap_sync(struct vfs *vfsp, short flag, struct cred *cr);
  79
  80 static void
  81 swapfs_recalc_save_initial(void)
  82 {
  83         initial_swapfs_desfree = swapfs_desfree;
  84         initial_swapfs_minfree = swapfs_minfree;
  85         initial_swapfs_reserve = swapfs_reserve;
  86 }
  87
  88 static int
  89 swapfs_recalc(pgcnt_t pgs)
  90 {
  91         pgcnt_t new_swapfs_desfree;
  92         pgcnt_t new_swapfs_minfree;
  93         pgcnt_t new_swapfs_reserve;
  94
  95         new_swapfs_desfree = initial_swapfs_desfree;
  96         new_swapfs_minfree = initial_swapfs_minfree;
  97         new_swapfs_reserve = initial_swapfs_reserve;
  98
  99         if (new_swapfs_desfree == 0)
 100                 new_swapfs_desfree = btopr(7 * 512 * 1024); /* 3-1/2Mb */;
 101
 102         if (new_swapfs_minfree == 0) {
 103                 /*
 104                  * We set this lower than we'd like here, 2Mb, because we
 105                  * always boot on swapfs. It's up to a safer value,
 106                  * swapfs_desfree, when/if we add physical swap devices
 107                  * in swapadd(). Users who want to change the amount of
 108                  * memory that can be used as swap space should do so by
 109                  * setting swapfs_desfree at boot time, not swapfs_minfree.
 110                  * However, swapfs_minfree is tunable by install as a
 111                  * workaround for bugid 1147463.
 112                  */
 113                 new_swapfs_minfree = MAX(btopr(2 * 1024 * 1024), pgs >> 3);
 114         }
 115
 116         /*
 117          * priv processes can reserve memory as swap as long as availrmem
 118          * remains greater than swapfs_minfree; in the case of non-priv
 119          * processes, memory can be reserved as swap only if availrmem
 120          * doesn't fall below (swapfs_minfree + swapfs_reserve). Thus,
 121          * swapfs_reserve amount of memswap is not available to non-priv
 122          * processes. This protects daemons such as automounter dying
 123          * as a result of application processes eating away almost entire
 124          * membased swap. This safeguard becomes useless if apps are run
 125          * with root access.
 126          *
 127          * set swapfs_reserve to a minimum of 4Mb or 1/128 of physmem whichever
 128          * is greater up to the limit of 128 MB.
 129          */
 130         if (new_swapfs_reserve == 0)
 131                 new_swapfs_reserve = MIN(btopr(128 * 1024 * 1024),
 132                     MAX(btopr(4 * 1024 * 1024), pgs >> 7));
 133
 134         /* Test basic numeric viability. */
 135         if (new_swapfs_minfree > pgs)
 136                 return (0);
 137
 138         /* Equivalent test to anon_resvmem() check. */
 139         if (availrmem < new_swapfs_minfree) {
 140                 /*
 141                  * If ism pages are being used, then there must be agreement
 142                  * between these two policies.
 143                  */
 144                 if ((availrmem > segspt_minfree) && (segspt_minfree > 0)) {
 145                         new_swapfs_minfree = segspt_minfree;
 146                 } else {
 147                         return (0);
 148                 }
 149         }
 150
 151         swapfs_desfree = new_swapfs_desfree;
 152         swapfs_minfree = new_swapfs_minfree;
 153         swapfs_reserve = new_swapfs_reserve;
 154
 155         return (1);
 156 }
 157
 158 static const struct vfsops swap_vfsops = {
 159         .vfs_sync = swap_sync,
 160 };
 161
 162 /*ARGSUSED1*/
 163 int
 164 swapinit(int fstype, char *name)
 165 {                                                       /* reserve for mp */
 166         ssize_t sw_freelist_size = klustsize / PAGESIZE * 2;
 167         int i, error;
 168
 169         SWAPFS_PRINT(SWAP_SUBR, "swapinit\n", 0, 0, 0, 0, 0);
 170         mutex_init(&swapfs_lock, NULL, MUTEX_DEFAULT, NULL);
 171
 172         swap_vnodes = kmem_zalloc(MAX_SWAP_VNODES * sizeof (struct vnode *),
 173             KM_SLEEP);
 174
 175         swapfs_recalc_save_initial();
 176         if (!swapfs_recalc(physmem))
 177                 cmn_err(CE_PANIC, "swapfs_minfree(%lu) > physmem(%lu)",
 178                     swapfs_minfree, physmem);
 179
 180         /*
 181          * Arrange for a callback on memory size change.
 182          */
 183         swap_init_mem_config();
 184
 185         sw_ar = (struct async_reqs *)
 186             kmem_zalloc(sw_freelist_size*sizeof (struct async_reqs), KM_SLEEP);
 187
 188         error = vfs_setfsops(fstype, &swap_vfsops);
 189         if (error != 0) {
 190                 cmn_err(CE_WARN, "swapinit: bad vfs ops template");
 191                 return (error);
 192         }
 193
 194         sw_freelist = sw_ar;
 195         for (i = 0; i < sw_freelist_size - 1; i++)
 196                 sw_ar[i].a_next = &sw_ar[i + 1];
 197
 198         return (0);
 199 }
 200
 201 /*
 202  * Get a swapfs vnode corresponding to the specified identifier.
 203  */
 204 struct vnode *
 205 swapfs_getvp(ulong_t vidx)
 206 {
 207         struct vnode *vp;
 208
 209         vp = swap_vnodes[vidx];
 210         if (vp) {
 211                 return (vp);
 212         }
 213
 214         mutex_enter(&swapfs_lock);
 215         vp = swap_vnodes[vidx];
 216         if (vp == NULL) {
 217                 vp = vn_alloc(KM_SLEEP);
 218                 vn_setops(vp, &swap_vnodeops);
 219                 vp->v_type = VREG;
 220                 vp->v_flag |= (VISSWAP|VISSWAPFS);
 221                 swap_vnodes[vidx] = vp;
 222                 swapfs_vpcount++;
 223         }
 224         mutex_exit(&swapfs_lock);
 225         return (vp);
 226 }
 227
 228 int swap_lo;
 229
 230 /*ARGSUSED*/
 231 static int
 232 swap_sync(struct vfs *vfsp, short flag, struct cred *cr)
 233 {
 234         struct vnode *vp;
 235         int i;
 236
 237         if (!(flag & SYNC_ALL))
 238                 return (1);
 239
 240         /*
 241          * assumes that we are the only one left to access this so that
 242          * no need to use swapfs_lock (since it's staticly defined)
 243          */
 244         for (i = 0; i < MAX_SWAP_VNODES; i++) {
 245                 vp = swap_vnodes[i];
 246                 if (vp) {
 247                         VN_HOLD(vp);
 248                         (void) fop_putpage(vp, 0, 0,
 249                             (B_ASYNC | B_FREE), kcred, NULL);
 250                         VN_RELE(vp);
 251                 }
 252         }
 253         return (0);
 254 }
 255
 256 extern int sw_pending_size;
 257
 258 /*
 259  * Take an async request off the pending queue
 260  */
 261 struct async_reqs *
 262 sw_getreq()
 263 {
 264         struct async_reqs *arg;
 265
 266         mutex_enter(&swapfs_lock);
 267         arg = sw_pendlist;
 268         if (arg) {
 269                 sw_pendlist = arg->a_next;
 270                 arg->a_next = NULL;
 271                 sw_pending_size -= PAGESIZE;
 272         }
 273         ASSERT(sw_pending_size >= 0);
 274         mutex_exit(&swapfs_lock);
 275         return (arg);
 276 }
 277
 278 /*
 279  * Put an async request on the pending queue
 280  */
 281 void
 282 sw_putreq(struct async_reqs *arg)
 283 {
 284         /* Hold onto it */
 285         VN_HOLD(arg->a_vp);
 286
 287         mutex_enter(&swapfs_lock);
 288         arg->a_next = sw_pendlist;
 289         sw_pendlist = arg;
 290         sw_pending_size += PAGESIZE;
 291         mutex_exit(&swapfs_lock);
 292 }
 293
 294 /*
 295  * Put an async request back on the pending queue
 296  */
 297 void
 298 sw_putbackreq(struct async_reqs *arg)
 299 {
 300         mutex_enter(&swapfs_lock);
 301         arg->a_next = sw_pendlist;
 302         sw_pendlist = arg;
 303         sw_pending_size += PAGESIZE;
 304         mutex_exit(&swapfs_lock);
 305 }
 306
 307 /*
 308  * Take an async request structure off the free list
 309  */
 310 struct async_reqs *
 311 sw_getfree()
 312 {
 313         struct async_reqs *arg;
 314
 315         mutex_enter(&swapfs_lock);
 316         arg = sw_freelist;
 317         if (arg) {
 318                 sw_freelist = arg->a_next;
 319                 arg->a_next = NULL;
 320         }
 321         mutex_exit(&swapfs_lock);
 322         return (arg);
 323 }
 324
 325 /*
 326  * Put an async request structure on the free list
 327  */
 328 void
 329 sw_putfree(struct async_reqs *arg)
 330 {
 331         /* Release our hold - should have locked the page by now */
 332         VN_RELE(arg->a_vp);
 333
 334         mutex_enter(&swapfs_lock);
 335         arg->a_next = sw_freelist;
 336         sw_freelist = arg;
 337         mutex_exit(&swapfs_lock);
 338 }
 339
 340 static pgcnt_t swapfs_pending_delete;
 341
 342 /*ARGSUSED*/
 343 static void
 344 swap_mem_config_post_add(
 345         void *arg,
 346         pgcnt_t delta_swaps)
 347 {
 348         (void) swapfs_recalc(physmem - swapfs_pending_delete);
 349 }
 350
 351 /*ARGSUSED*/
 352 static int
 353 swap_mem_config_pre_del(
 354         void *arg,
 355         pgcnt_t delta_swaps)
 356 {
 357         pgcnt_t nv;
 358
 359         nv = atomic_add_long_nv(&swapfs_pending_delete, (spgcnt_t)delta_swaps);
 360         if (!swapfs_recalc(physmem - nv)) {
 361                 /*
 362                  * Tidy-up is done by the call to post_del which
 363                  * is always made.
 364                  */
 365                 cmn_err(CE_NOTE, "Memory operation refused to ensure system "
 366                     "doesn't deadlock due to excessive consumption by swapfs.");
 367                 return (EBUSY);
 368         }
 369         return (0);
 370 }
 371
 372 /*ARGSUSED*/
 373 static void
 374 swap_mem_config_post_del(
 375         void *arg,
 376         pgcnt_t delta_swaps,
 377         int cancelled)
 378 {
 379         pgcnt_t nv;
 380
 381         nv = atomic_add_long_nv(&swapfs_pending_delete, -(spgcnt_t)delta_swaps);
 382         (void) swapfs_recalc(physmem - nv);
 383 }
 384
 385 static kphysm_setup_vector_t swap_mem_config_vec = {
 386         KPHYSM_SETUP_VECTOR_VERSION,
 387         swap_mem_config_post_add,
 388         swap_mem_config_pre_del,
 389         swap_mem_config_post_del,
 390 };
 391
 392 static void
 393 swap_init_mem_config(void)
 394 {
 395         int ret;
 396
 397         ret = kphysm_setup_func_register(&swap_mem_config_vec, NULL);
 398         ASSERT(ret == 0);
 399 }