blob: 4a57f135b76e74741e7c7d285642067a3e18efe6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Simple NUMA memory policy for the Linux kernel.
3 *
4 * Copyright 2003,2004 Andi Kleen, SuSE Labs.
Christoph Lameter8bccd852005-10-29 18:16:59 -07005 * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006 * Subject to the GNU Public License, version 2.
7 *
8 * NUMA policy allows the user to give hints in which node(s) memory should
9 * be allocated.
10 *
11 * Support four policies per VMA and per process:
12 *
13 * The VMA policy has priority over the process policy for a page fault.
14 *
15 * interleave Allocate memory interleaved over a set of nodes,
16 * with normal fallback if it fails.
17 * For VMA based allocations this interleaves based on the
18 * offset into the backing object or offset into the mapping
19 * for anonymous memory. For process policy an process counter
20 * is used.
Christoph Lameter8bccd852005-10-29 18:16:59 -070021 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070022 * bind Only allocate memory on a specific set of nodes,
23 * no fallback.
Christoph Lameter8bccd852005-10-29 18:16:59 -070024 * FIXME: memory is allocated starting with the first node
25 * to the last. It would be better if bind would truly restrict
26 * the allocation to memory nodes instead
27 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070028 * preferred Try a specific node first before normal fallback.
29 * As a special case node -1 here means do the allocation
30 * on the local CPU. This is normally identical to default,
31 * but useful to set in a VMA when you have a non default
32 * process policy.
Christoph Lameter8bccd852005-10-29 18:16:59 -070033 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 * default Allocate on the local node first, or when on a VMA
35 * use the process policy. This is what Linux always did
36 * in a NUMA aware kernel and still does by, ahem, default.
37 *
38 * The process policy is applied for most non interrupt memory allocations
39 * in that process' context. Interrupts ignore the policies and always
40 * try to allocate on the local CPU. The VMA policy is only applied for memory
41 * allocations for a VMA in the VM.
42 *
43 * Currently there are a few corner cases in swapping where the policy
44 * is not applied, but the majority should be handled. When process policy
45 * is used it is not remembered over swap outs/swap ins.
46 *
47 * Only the highest zone in the zone hierarchy gets policied. Allocations
48 * requesting a lower zone just use default policy. This implies that
49 * on systems with highmem kernel lowmem allocation don't get policied.
50 * Same with GFP_DMA allocations.
51 *
52 * For shmfs/tmpfs/hugetlbfs shared memory the policy is shared between
53 * all users and remembered even when nobody has memory mapped.
54 */
55
56/* Notebook:
57 fix mmap readahead to honour policy and enable policy for any page cache
58 object
59 statistics for bigpages
60 global policy for page cache? currently it uses process policy. Requires
61 first item above.
62 handle mremap for shared memory (currently ignored for the policy)
63 grows down?
64 make bind policy root only? It can trigger oom much faster and the
65 kernel is not always grateful with that.
Linus Torvalds1da177e2005-04-16 15:20:36 -070066*/
67
68#include <linux/mempolicy.h>
69#include <linux/mm.h>
70#include <linux/highmem.h>
71#include <linux/hugetlb.h>
72#include <linux/kernel.h>
73#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#include <linux/nodemask.h>
75#include <linux/cpuset.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#include <linux/slab.h>
77#include <linux/string.h>
78#include <linux/module.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070079#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#include <linux/interrupt.h>
81#include <linux/init.h>
82#include <linux/compat.h>
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -080083#include <linux/swap.h>
Christoph Lameter1a75a6c2006-01-08 01:01:02 -080084#include <linux/seq_file.h>
85#include <linux/proc_fs.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080086#include <linux/migrate.h>
Hugh Dickins62b61f62009-12-14 17:59:33 -080087#include <linux/ksm.h>
Christoph Lameter95a402c2006-06-23 02:03:53 -070088#include <linux/rmap.h>
David Quigley86c3a762006-06-23 02:04:02 -070089#include <linux/security.h>
Adrian Bunkdbcb0f12007-10-16 01:26:26 -070090#include <linux/syscalls.h>
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -070091#include <linux/ctype.h>
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -080092#include <linux/mm_inline.h>
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -080093
Linus Torvalds1da177e2005-04-16 15:20:36 -070094#include <asm/tlbflush.h>
95#include <asm/uaccess.h>
96
Nick Piggin62695a82008-10-18 20:26:09 -070097#include "internal.h"
98
Christoph Lameter38e35862006-01-08 01:01:01 -080099/* Internal flags */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800100#define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */
Christoph Lameter38e35862006-01-08 01:01:01 -0800101#define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800102#define MPOL_MF_STATS (MPOL_MF_INTERNAL << 2) /* Gather statistics */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800103
Pekka Enbergfcc234f2006-03-22 00:08:13 -0800104static struct kmem_cache *policy_cache;
105static struct kmem_cache *sn_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107/* Highest zone. An specific allocation for a zone below that is not
108 policied. */
Christoph Lameter62672762007-02-10 01:43:07 -0800109enum zone_type policy_zone = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700111/*
112 * run-time system-wide default policy => local allocation
113 */
Andi Kleend42c6992005-07-06 19:56:03 +0200114struct mempolicy default_policy = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 .refcnt = ATOMIC_INIT(1), /* never free it */
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700116 .mode = MPOL_PREFERRED,
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700117 .flags = MPOL_F_LOCAL,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118};
119
David Rientjes37012942008-04-28 02:12:33 -0700120static const struct mempolicy_operations {
121 int (*create)(struct mempolicy *pol, const nodemask_t *nodes);
Miao Xie708c1bb2010-05-24 14:32:07 -0700122 /*
123 * If read-side task has no lock to protect task->mempolicy, write-side
124 * task will rebind the task->mempolicy by two step. The first step is
125 * setting all the newly nodes, and the second step is cleaning all the
126 * disallowed nodes. In this way, we can avoid finding no node to alloc
127 * page.
128 * If we have a lock to protect task->mempolicy in read-side, we do
129 * rebind directly.
130 *
131 * step:
132 * MPOL_REBIND_ONCE - do rebind work at once
133 * MPOL_REBIND_STEP1 - set all the newly nodes
134 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
135 */
136 void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes,
137 enum mpol_rebind_step step);
David Rientjes37012942008-04-28 02:12:33 -0700138} mpol_ops[MPOL_MAX];
139
Mel Gorman19770b32008-04-28 02:12:18 -0700140/* Check that the nodemask contains at least one populated zone */
David Rientjes37012942008-04-28 02:12:33 -0700141static int is_valid_nodemask(const nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142{
Mel Gorman19770b32008-04-28 02:12:18 -0700143 int nd, k;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Mel Gorman19770b32008-04-28 02:12:18 -0700145 for_each_node_mask(nd, *nodemask) {
146 struct zone *z;
147
148 for (k = 0; k <= policy_zone; k++) {
149 z = &NODE_DATA(nd)->node_zones[k];
150 if (z->present_pages > 0)
151 return 1;
Andi Kleendd942ae2006-02-17 01:39:16 +0100152 }
153 }
Mel Gorman19770b32008-04-28 02:12:18 -0700154
155 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156}
157
David Rientjesf5b087b2008-04-28 02:12:27 -0700158static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
159{
Bob Liu6d556292010-05-24 14:31:59 -0700160 return pol->flags & MPOL_MODE_FLAGS;
David Rientjes4c50bc02008-04-28 02:12:30 -0700161}
162
163static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig,
164 const nodemask_t *rel)
165{
166 nodemask_t tmp;
167 nodes_fold(tmp, *orig, nodes_weight(*rel));
168 nodes_onto(*ret, tmp, *rel);
David Rientjesf5b087b2008-04-28 02:12:27 -0700169}
170
David Rientjes37012942008-04-28 02:12:33 -0700171static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes)
172{
173 if (nodes_empty(*nodes))
174 return -EINVAL;
175 pol->v.nodes = *nodes;
176 return 0;
177}
178
179static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes)
180{
181 if (!nodes)
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700182 pol->flags |= MPOL_F_LOCAL; /* local allocation */
David Rientjes37012942008-04-28 02:12:33 -0700183 else if (nodes_empty(*nodes))
184 return -EINVAL; /* no allowed nodes */
185 else
186 pol->v.preferred_node = first_node(*nodes);
187 return 0;
188}
189
190static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes)
191{
192 if (!is_valid_nodemask(nodes))
193 return -EINVAL;
194 pol->v.nodes = *nodes;
195 return 0;
196}
197
Miao Xie58568d22009-06-16 15:31:49 -0700198/*
199 * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if
200 * any, for the new policy. mpol_new() has already validated the nodes
201 * parameter with respect to the policy mode and flags. But, we need to
202 * handle an empty nodemask with MPOL_PREFERRED here.
203 *
204 * Must be called holding task's alloc_lock to protect task's mems_allowed
205 * and mempolicy. May also be called holding the mmap_semaphore for write.
206 */
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700207static int mpol_set_nodemask(struct mempolicy *pol,
208 const nodemask_t *nodes, struct nodemask_scratch *nsc)
Miao Xie58568d22009-06-16 15:31:49 -0700209{
Miao Xie58568d22009-06-16 15:31:49 -0700210 int ret;
211
212 /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */
213 if (pol == NULL)
214 return 0;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700215 /* Check N_HIGH_MEMORY */
216 nodes_and(nsc->mask1,
217 cpuset_current_mems_allowed, node_states[N_HIGH_MEMORY]);
Miao Xie58568d22009-06-16 15:31:49 -0700218
219 VM_BUG_ON(!nodes);
220 if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes))
221 nodes = NULL; /* explicit local allocation */
222 else {
223 if (pol->flags & MPOL_F_RELATIVE_NODES)
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700224 mpol_relative_nodemask(&nsc->mask2, nodes,&nsc->mask1);
Miao Xie58568d22009-06-16 15:31:49 -0700225 else
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700226 nodes_and(nsc->mask2, *nodes, nsc->mask1);
227
Miao Xie58568d22009-06-16 15:31:49 -0700228 if (mpol_store_user_nodemask(pol))
229 pol->w.user_nodemask = *nodes;
230 else
231 pol->w.cpuset_mems_allowed =
232 cpuset_current_mems_allowed;
233 }
234
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700235 if (nodes)
236 ret = mpol_ops[pol->mode].create(pol, &nsc->mask2);
237 else
238 ret = mpol_ops[pol->mode].create(pol, NULL);
Miao Xie58568d22009-06-16 15:31:49 -0700239 return ret;
240}
241
242/*
243 * This function just creates a new policy, does some check and simple
244 * initialization. You must invoke mpol_set_nodemask() to set nodes.
245 */
David Rientjes028fec42008-04-28 02:12:25 -0700246static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
247 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248{
249 struct mempolicy *policy;
250
David Rientjes028fec42008-04-28 02:12:25 -0700251 pr_debug("setting mode %d flags %d nodes[0] %lx\n",
252 mode, flags, nodes ? nodes_addr(*nodes)[0] : -1);
Paul Mundt140d5a42007-07-15 23:38:16 -0700253
David Rientjes3e1f0642008-04-28 02:12:34 -0700254 if (mode == MPOL_DEFAULT) {
255 if (nodes && !nodes_empty(*nodes))
David Rientjes37012942008-04-28 02:12:33 -0700256 return ERR_PTR(-EINVAL);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700257 return NULL; /* simply delete any existing policy */
David Rientjes37012942008-04-28 02:12:33 -0700258 }
David Rientjes3e1f0642008-04-28 02:12:34 -0700259 VM_BUG_ON(!nodes);
260
261 /*
262 * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
263 * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
264 * All other modes require a valid pointer to a non-empty nodemask.
265 */
266 if (mode == MPOL_PREFERRED) {
267 if (nodes_empty(*nodes)) {
268 if (((flags & MPOL_F_STATIC_NODES) ||
269 (flags & MPOL_F_RELATIVE_NODES)))
270 return ERR_PTR(-EINVAL);
David Rientjes3e1f0642008-04-28 02:12:34 -0700271 }
272 } else if (nodes_empty(*nodes))
273 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
275 if (!policy)
276 return ERR_PTR(-ENOMEM);
277 atomic_set(&policy->refcnt, 1);
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700278 policy->mode = mode;
David Rientjes3e1f0642008-04-28 02:12:34 -0700279 policy->flags = flags;
David Rientjesf5b087b2008-04-28 02:12:27 -0700280
David Rientjes37012942008-04-28 02:12:33 -0700281 return policy;
282}
283
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700284/* Slow path of a mpol destructor. */
285void __mpol_put(struct mempolicy *p)
286{
287 if (!atomic_dec_and_test(&p->refcnt))
288 return;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700289 kmem_cache_free(policy_cache, p);
290}
291
Miao Xie708c1bb2010-05-24 14:32:07 -0700292static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes,
293 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700294{
295}
296
Miao Xie708c1bb2010-05-24 14:32:07 -0700297/*
298 * step:
299 * MPOL_REBIND_ONCE - do rebind work at once
300 * MPOL_REBIND_STEP1 - set all the newly nodes
301 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
302 */
303static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes,
304 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700305{
306 nodemask_t tmp;
307
308 if (pol->flags & MPOL_F_STATIC_NODES)
309 nodes_and(tmp, pol->w.user_nodemask, *nodes);
310 else if (pol->flags & MPOL_F_RELATIVE_NODES)
311 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
312 else {
Miao Xie708c1bb2010-05-24 14:32:07 -0700313 /*
314 * if step == 1, we use ->w.cpuset_mems_allowed to cache the
315 * result
316 */
317 if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) {
318 nodes_remap(tmp, pol->v.nodes,
319 pol->w.cpuset_mems_allowed, *nodes);
320 pol->w.cpuset_mems_allowed = step ? tmp : *nodes;
321 } else if (step == MPOL_REBIND_STEP2) {
322 tmp = pol->w.cpuset_mems_allowed;
323 pol->w.cpuset_mems_allowed = *nodes;
324 } else
325 BUG();
David Rientjes37012942008-04-28 02:12:33 -0700326 }
327
Miao Xie708c1bb2010-05-24 14:32:07 -0700328 if (nodes_empty(tmp))
329 tmp = *nodes;
330
331 if (step == MPOL_REBIND_STEP1)
332 nodes_or(pol->v.nodes, pol->v.nodes, tmp);
333 else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2)
334 pol->v.nodes = tmp;
335 else
336 BUG();
337
David Rientjes37012942008-04-28 02:12:33 -0700338 if (!node_isset(current->il_next, tmp)) {
339 current->il_next = next_node(current->il_next, tmp);
340 if (current->il_next >= MAX_NUMNODES)
341 current->il_next = first_node(tmp);
342 if (current->il_next >= MAX_NUMNODES)
343 current->il_next = numa_node_id();
344 }
345}
346
347static void mpol_rebind_preferred(struct mempolicy *pol,
Miao Xie708c1bb2010-05-24 14:32:07 -0700348 const nodemask_t *nodes,
349 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700350{
351 nodemask_t tmp;
352
David Rientjes37012942008-04-28 02:12:33 -0700353 if (pol->flags & MPOL_F_STATIC_NODES) {
354 int node = first_node(pol->w.user_nodemask);
355
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700356 if (node_isset(node, *nodes)) {
David Rientjes37012942008-04-28 02:12:33 -0700357 pol->v.preferred_node = node;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700358 pol->flags &= ~MPOL_F_LOCAL;
359 } else
360 pol->flags |= MPOL_F_LOCAL;
David Rientjes37012942008-04-28 02:12:33 -0700361 } else if (pol->flags & MPOL_F_RELATIVE_NODES) {
362 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
363 pol->v.preferred_node = first_node(tmp);
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700364 } else if (!(pol->flags & MPOL_F_LOCAL)) {
David Rientjes37012942008-04-28 02:12:33 -0700365 pol->v.preferred_node = node_remap(pol->v.preferred_node,
366 pol->w.cpuset_mems_allowed,
367 *nodes);
368 pol->w.cpuset_mems_allowed = *nodes;
369 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370}
371
Miao Xie708c1bb2010-05-24 14:32:07 -0700372/*
373 * mpol_rebind_policy - Migrate a policy to a different set of nodes
374 *
375 * If read-side task has no lock to protect task->mempolicy, write-side
376 * task will rebind the task->mempolicy by two step. The first step is
377 * setting all the newly nodes, and the second step is cleaning all the
378 * disallowed nodes. In this way, we can avoid finding no node to alloc
379 * page.
380 * If we have a lock to protect task->mempolicy in read-side, we do
381 * rebind directly.
382 *
383 * step:
384 * MPOL_REBIND_ONCE - do rebind work at once
385 * MPOL_REBIND_STEP1 - set all the newly nodes
386 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
387 */
388static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask,
389 enum mpol_rebind_step step)
David Rientjes1d0d2682008-04-28 02:12:32 -0700390{
David Rientjes1d0d2682008-04-28 02:12:32 -0700391 if (!pol)
392 return;
Miao Xie708c1bb2010-05-24 14:32:07 -0700393 if (!mpol_store_user_nodemask(pol) && step == 0 &&
David Rientjes1d0d2682008-04-28 02:12:32 -0700394 nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
395 return;
Miao Xie708c1bb2010-05-24 14:32:07 -0700396
397 if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING))
398 return;
399
400 if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING))
401 BUG();
402
403 if (step == MPOL_REBIND_STEP1)
404 pol->flags |= MPOL_F_REBINDING;
405 else if (step == MPOL_REBIND_STEP2)
406 pol->flags &= ~MPOL_F_REBINDING;
407 else if (step >= MPOL_REBIND_NSTEP)
408 BUG();
409
410 mpol_ops[pol->mode].rebind(pol, newmask, step);
David Rientjes1d0d2682008-04-28 02:12:32 -0700411}
412
413/*
414 * Wrapper for mpol_rebind_policy() that just requires task
415 * pointer, and updates task mempolicy.
Miao Xie58568d22009-06-16 15:31:49 -0700416 *
417 * Called with task's alloc_lock held.
David Rientjes1d0d2682008-04-28 02:12:32 -0700418 */
419
Miao Xie708c1bb2010-05-24 14:32:07 -0700420void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
421 enum mpol_rebind_step step)
David Rientjes1d0d2682008-04-28 02:12:32 -0700422{
Miao Xie708c1bb2010-05-24 14:32:07 -0700423 mpol_rebind_policy(tsk->mempolicy, new, step);
David Rientjes1d0d2682008-04-28 02:12:32 -0700424}
425
426/*
427 * Rebind each vma in mm to new nodemask.
428 *
429 * Call holding a reference to mm. Takes mm->mmap_sem during call.
430 */
431
432void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
433{
434 struct vm_area_struct *vma;
435
436 down_write(&mm->mmap_sem);
437 for (vma = mm->mmap; vma; vma = vma->vm_next)
Miao Xie708c1bb2010-05-24 14:32:07 -0700438 mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE);
David Rientjes1d0d2682008-04-28 02:12:32 -0700439 up_write(&mm->mmap_sem);
440}
441
David Rientjes37012942008-04-28 02:12:33 -0700442static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
443 [MPOL_DEFAULT] = {
444 .rebind = mpol_rebind_default,
445 },
446 [MPOL_INTERLEAVE] = {
447 .create = mpol_new_interleave,
448 .rebind = mpol_rebind_nodemask,
449 },
450 [MPOL_PREFERRED] = {
451 .create = mpol_new_preferred,
452 .rebind = mpol_rebind_preferred,
453 },
454 [MPOL_BIND] = {
455 .create = mpol_new_bind,
456 .rebind = mpol_rebind_nodemask,
457 },
458};
459
Christoph Lameter397874d2006-03-06 15:42:53 -0800460static void gather_stats(struct page *, void *, int pte_dirty);
Christoph Lameterfc301282006-01-18 17:42:29 -0800461static void migrate_page_add(struct page *page, struct list_head *pagelist,
462 unsigned long flags);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800463
Christoph Lameter38e35862006-01-08 01:01:01 -0800464/* Scan through pages checking if pages follow certain conditions. */
Nick Pigginb5810032005-10-29 18:16:12 -0700465static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800466 unsigned long addr, unsigned long end,
467 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800468 void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
Hugh Dickins91612e02005-06-21 17:15:07 -0700470 pte_t *orig_pte;
471 pte_t *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700472 spinlock_t *ptl;
Hugh Dickins941150a2005-06-21 17:15:06 -0700473
Hugh Dickins705e87c2005-10-29 18:16:27 -0700474 orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins91612e02005-06-21 17:15:07 -0700475 do {
Linus Torvalds6aab3412005-11-28 14:34:23 -0800476 struct page *page;
Andy Whitcroft25ba77c2006-12-06 20:33:03 -0800477 int nid;
Hugh Dickins91612e02005-06-21 17:15:07 -0700478
479 if (!pte_present(*pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800481 page = vm_normal_page(vma, addr, *pte);
482 if (!page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 continue;
Nick Piggin053837f2006-01-18 17:42:27 -0800484 /*
Hugh Dickins62b61f62009-12-14 17:59:33 -0800485 * vm_normal_page() filters out zero pages, but there might
486 * still be PageReserved pages to skip, perhaps in a VDSO.
487 * And we cannot move PageKsm pages sensibly or safely yet.
Nick Piggin053837f2006-01-18 17:42:27 -0800488 */
Hugh Dickins62b61f62009-12-14 17:59:33 -0800489 if (PageReserved(page) || PageKsm(page))
Christoph Lameterf4598c82006-01-12 01:05:20 -0800490 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800491 nid = page_to_nid(page);
Christoph Lameter38e35862006-01-08 01:01:01 -0800492 if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT))
493 continue;
494
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800495 if (flags & MPOL_MF_STATS)
Christoph Lameter397874d2006-03-06 15:42:53 -0800496 gather_stats(page, private, pte_dirty(*pte));
Nick Piggin053837f2006-01-18 17:42:27 -0800497 else if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Christoph Lameterfc301282006-01-18 17:42:29 -0800498 migrate_page_add(page, private, flags);
Christoph Lameter38e35862006-01-08 01:01:01 -0800499 else
500 break;
Hugh Dickins91612e02005-06-21 17:15:07 -0700501 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700502 pte_unmap_unlock(orig_pte, ptl);
Hugh Dickins91612e02005-06-21 17:15:07 -0700503 return addr != end;
504}
505
Nick Pigginb5810032005-10-29 18:16:12 -0700506static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800507 unsigned long addr, unsigned long end,
508 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800509 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700510{
511 pmd_t *pmd;
512 unsigned long next;
513
514 pmd = pmd_offset(pud, addr);
515 do {
516 next = pmd_addr_end(addr, end);
517 if (pmd_none_or_clear_bad(pmd))
518 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800519 if (check_pte_range(vma, pmd, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800520 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700521 return -EIO;
522 } while (pmd++, addr = next, addr != end);
523 return 0;
524}
525
Nick Pigginb5810032005-10-29 18:16:12 -0700526static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800527 unsigned long addr, unsigned long end,
528 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800529 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700530{
531 pud_t *pud;
532 unsigned long next;
533
534 pud = pud_offset(pgd, addr);
535 do {
536 next = pud_addr_end(addr, end);
537 if (pud_none_or_clear_bad(pud))
538 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800539 if (check_pmd_range(vma, pud, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800540 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700541 return -EIO;
542 } while (pud++, addr = next, addr != end);
543 return 0;
544}
545
Nick Pigginb5810032005-10-29 18:16:12 -0700546static inline int check_pgd_range(struct vm_area_struct *vma,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800547 unsigned long addr, unsigned long end,
548 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800549 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700550{
551 pgd_t *pgd;
552 unsigned long next;
553
Nick Pigginb5810032005-10-29 18:16:12 -0700554 pgd = pgd_offset(vma->vm_mm, addr);
Hugh Dickins91612e02005-06-21 17:15:07 -0700555 do {
556 next = pgd_addr_end(addr, end);
557 if (pgd_none_or_clear_bad(pgd))
558 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800559 if (check_pud_range(vma, pgd, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800560 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700561 return -EIO;
562 } while (pgd++, addr = next, addr != end);
563 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564}
565
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800566/*
567 * Check if all pages in a range are on a set of nodes.
568 * If pagelist != NULL then isolate pages from the LRU and
569 * put them on the pagelist.
570 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571static struct vm_area_struct *
572check_range(struct mm_struct *mm, unsigned long start, unsigned long end,
Christoph Lameter38e35862006-01-08 01:01:01 -0800573 const nodemask_t *nodes, unsigned long flags, void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
575 int err;
576 struct vm_area_struct *first, *vma, *prev;
577
Nick Piggin053837f2006-01-18 17:42:27 -0800578
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 first = find_vma(mm, start);
580 if (!first)
581 return ERR_PTR(-EFAULT);
582 prev = NULL;
583 for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) {
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800584 if (!(flags & MPOL_MF_DISCONTIG_OK)) {
585 if (!vma->vm_next && vma->vm_end < end)
586 return ERR_PTR(-EFAULT);
587 if (prev && prev->vm_end < vma->vm_start)
588 return ERR_PTR(-EFAULT);
589 }
590 if (!is_vm_hugetlb_page(vma) &&
591 ((flags & MPOL_MF_STRICT) ||
592 ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) &&
593 vma_migratable(vma)))) {
Andi Kleen5b952b32005-09-13 01:25:08 -0700594 unsigned long endvma = vma->vm_end;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800595
Andi Kleen5b952b32005-09-13 01:25:08 -0700596 if (endvma > end)
597 endvma = end;
598 if (vma->vm_start > start)
599 start = vma->vm_start;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800600 err = check_pgd_range(vma, start, endvma, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800601 flags, private);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 if (err) {
603 first = ERR_PTR(err);
604 break;
605 }
606 }
607 prev = vma;
608 }
609 return first;
610}
611
612/* Apply policy to a single VMA */
613static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new)
614{
615 int err = 0;
616 struct mempolicy *old = vma->vm_policy;
617
Paul Mundt140d5a42007-07-15 23:38:16 -0700618 pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 vma->vm_start, vma->vm_end, vma->vm_pgoff,
620 vma->vm_ops, vma->vm_file,
621 vma->vm_ops ? vma->vm_ops->set_policy : NULL);
622
623 if (vma->vm_ops && vma->vm_ops->set_policy)
624 err = vma->vm_ops->set_policy(vma, new);
625 if (!err) {
626 mpol_get(new);
627 vma->vm_policy = new;
Lee Schermerhornf0be3d32008-04-28 02:13:08 -0700628 mpol_put(old);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 }
630 return err;
631}
632
633/* Step 2: apply policy to a range and do splits. */
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800634static int mbind_range(struct mm_struct *mm, unsigned long start,
635 unsigned long end, struct mempolicy *new_pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636{
637 struct vm_area_struct *next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800638 struct vm_area_struct *prev;
639 struct vm_area_struct *vma;
640 int err = 0;
641 pgoff_t pgoff;
642 unsigned long vmstart;
643 unsigned long vmend;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800645 vma = find_vma_prev(mm, start, &prev);
646 if (!vma || vma->vm_start > start)
647 return -EFAULT;
648
649 for (; vma && vma->vm_start < end; prev = vma, vma = next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 next = vma->vm_next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800651 vmstart = max(start, vma->vm_start);
652 vmend = min(end, vma->vm_end);
653
654 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
655 prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
656 vma->anon_vma, vma->vm_file, pgoff, new_pol);
657 if (prev) {
658 vma = prev;
659 next = vma->vm_next;
660 continue;
661 }
662 if (vma->vm_start != vmstart) {
663 err = split_vma(vma->vm_mm, vma, vmstart, 1);
664 if (err)
665 goto out;
666 }
667 if (vma->vm_end != vmend) {
668 err = split_vma(vma->vm_mm, vma, vmend, 0);
669 if (err)
670 goto out;
671 }
672 err = policy_vma(vma, new_pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 if (err)
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800674 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 }
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800676
677 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 return err;
679}
680
Paul Jacksonc61afb12006-03-24 03:16:08 -0800681/*
682 * Update task->flags PF_MEMPOLICY bit: set iff non-default
683 * mempolicy. Allows more rapid checking of this (combined perhaps
684 * with other PF_* flag bits) on memory allocation hot code paths.
685 *
686 * If called from outside this file, the task 'p' should -only- be
687 * a newly forked child not yet visible on the task list, because
688 * manipulating the task flags of a visible task is not safe.
689 *
690 * The above limitation is why this routine has the funny name
691 * mpol_fix_fork_child_flag().
692 *
693 * It is also safe to call this with a task pointer of current,
694 * which the static wrapper mpol_set_task_struct_flag() does,
695 * for use within this file.
696 */
697
698void mpol_fix_fork_child_flag(struct task_struct *p)
699{
700 if (p->mempolicy)
701 p->flags |= PF_MEMPOLICY;
702 else
703 p->flags &= ~PF_MEMPOLICY;
704}
705
706static void mpol_set_task_struct_flag(void)
707{
708 mpol_fix_fork_child_flag(current);
709}
710
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711/* Set the process memory policy */
David Rientjes028fec42008-04-28 02:12:25 -0700712static long do_set_mempolicy(unsigned short mode, unsigned short flags,
713 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714{
Miao Xie58568d22009-06-16 15:31:49 -0700715 struct mempolicy *new, *old;
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700716 struct mm_struct *mm = current->mm;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700717 NODEMASK_SCRATCH(scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700718 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700720 if (!scratch)
721 return -ENOMEM;
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700722
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700723 new = mpol_new(mode, flags, nodes);
724 if (IS_ERR(new)) {
725 ret = PTR_ERR(new);
726 goto out;
727 }
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700728 /*
729 * prevent changing our mempolicy while show_numa_maps()
730 * is using it.
731 * Note: do_set_mempolicy() can be called at init time
732 * with no 'mm'.
733 */
734 if (mm)
735 down_write(&mm->mmap_sem);
Miao Xie58568d22009-06-16 15:31:49 -0700736 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700737 ret = mpol_set_nodemask(new, nodes, scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700738 if (ret) {
739 task_unlock(current);
740 if (mm)
741 up_write(&mm->mmap_sem);
742 mpol_put(new);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700743 goto out;
Miao Xie58568d22009-06-16 15:31:49 -0700744 }
745 old = current->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 current->mempolicy = new;
Paul Jacksonc61afb12006-03-24 03:16:08 -0800747 mpol_set_task_struct_flag();
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700748 if (new && new->mode == MPOL_INTERLEAVE &&
David Rientjesf5b087b2008-04-28 02:12:27 -0700749 nodes_weight(new->v.nodes))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700750 current->il_next = first_node(new->v.nodes);
Miao Xie58568d22009-06-16 15:31:49 -0700751 task_unlock(current);
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700752 if (mm)
753 up_write(&mm->mmap_sem);
754
Miao Xie58568d22009-06-16 15:31:49 -0700755 mpol_put(old);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700756 ret = 0;
757out:
758 NODEMASK_SCRATCH_FREE(scratch);
759 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760}
761
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700762/*
763 * Return nodemask for policy for get_mempolicy() query
Miao Xie58568d22009-06-16 15:31:49 -0700764 *
765 * Called with task's alloc_lock held
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700766 */
767static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768{
Andi Kleendfcd3c02005-10-29 18:15:48 -0700769 nodes_clear(*nodes);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700770 if (p == &default_policy)
771 return;
772
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700773 switch (p->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -0700774 case MPOL_BIND:
775 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -0700777 *nodes = p->v.nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 break;
779 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700780 if (!(p->flags & MPOL_F_LOCAL))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700781 node_set(p->v.preferred_node, *nodes);
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700782 /* else return empty node mask for local allocation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 break;
784 default:
785 BUG();
786 }
787}
788
789static int lookup_node(struct mm_struct *mm, unsigned long addr)
790{
791 struct page *p;
792 int err;
793
794 err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL);
795 if (err >= 0) {
796 err = page_to_nid(p);
797 put_page(p);
798 }
799 return err;
800}
801
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802/* Retrieve NUMA policy */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700803static long do_get_mempolicy(int *policy, nodemask_t *nmask,
804 unsigned long addr, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805{
Christoph Lameter8bccd852005-10-29 18:16:59 -0700806 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 struct mm_struct *mm = current->mm;
808 struct vm_area_struct *vma = NULL;
809 struct mempolicy *pol = current->mempolicy;
810
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700811 if (flags &
812 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 return -EINVAL;
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700814
815 if (flags & MPOL_F_MEMS_ALLOWED) {
816 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
817 return -EINVAL;
818 *policy = 0; /* just so it's initialized */
Miao Xie58568d22009-06-16 15:31:49 -0700819 task_lock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700820 *nmask = cpuset_current_mems_allowed;
Miao Xie58568d22009-06-16 15:31:49 -0700821 task_unlock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700822 return 0;
823 }
824
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 if (flags & MPOL_F_ADDR) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700826 /*
827 * Do NOT fall back to task policy if the
828 * vma/shared policy at addr is NULL. We
829 * want to return MPOL_DEFAULT in this case.
830 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 down_read(&mm->mmap_sem);
832 vma = find_vma_intersection(mm, addr, addr+1);
833 if (!vma) {
834 up_read(&mm->mmap_sem);
835 return -EFAULT;
836 }
837 if (vma->vm_ops && vma->vm_ops->get_policy)
838 pol = vma->vm_ops->get_policy(vma, addr);
839 else
840 pol = vma->vm_policy;
841 } else if (addr)
842 return -EINVAL;
843
844 if (!pol)
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700845 pol = &default_policy; /* indicates default behavior */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
847 if (flags & MPOL_F_NODE) {
848 if (flags & MPOL_F_ADDR) {
849 err = lookup_node(mm, addr);
850 if (err < 0)
851 goto out;
Christoph Lameter8bccd852005-10-29 18:16:59 -0700852 *policy = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 } else if (pol == current->mempolicy &&
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700854 pol->mode == MPOL_INTERLEAVE) {
Christoph Lameter8bccd852005-10-29 18:16:59 -0700855 *policy = current->il_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 } else {
857 err = -EINVAL;
858 goto out;
859 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700860 } else {
861 *policy = pol == &default_policy ? MPOL_DEFAULT :
862 pol->mode;
David Rientjesd79df632008-07-04 12:24:13 -0700863 /*
864 * Internal mempolicy flags must be masked off before exposing
865 * the policy to userspace.
866 */
867 *policy |= (pol->flags & MPOL_MODE_FLAGS);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700868 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869
870 if (vma) {
871 up_read(&current->mm->mmap_sem);
872 vma = NULL;
873 }
874
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 err = 0;
Miao Xie58568d22009-06-16 15:31:49 -0700876 if (nmask) {
Lee Schermerhornc6b6ef82010-03-23 13:35:41 -0700877 if (mpol_store_user_nodemask(pol)) {
878 *nmask = pol->w.user_nodemask;
879 } else {
880 task_lock(current);
881 get_policy_nodemask(pol, nmask);
882 task_unlock(current);
883 }
Miao Xie58568d22009-06-16 15:31:49 -0700884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885
886 out:
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700887 mpol_cond_put(pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 if (vma)
889 up_read(&current->mm->mmap_sem);
890 return err;
891}
892
Christoph Lameterb20a3502006-03-22 00:09:12 -0800893#ifdef CONFIG_MIGRATION
Christoph Lameter8bccd852005-10-29 18:16:59 -0700894/*
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800895 * page migration
896 */
Christoph Lameterfc301282006-01-18 17:42:29 -0800897static void migrate_page_add(struct page *page, struct list_head *pagelist,
898 unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800899{
900 /*
Christoph Lameterfc301282006-01-18 17:42:29 -0800901 * Avoid migrating a page that is shared with others.
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800902 */
Nick Piggin62695a82008-10-18 20:26:09 -0700903 if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) {
904 if (!isolate_lru_page(page)) {
905 list_add_tail(&page->lru, pagelist);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800906 inc_zone_page_state(page, NR_ISOLATED_ANON +
907 page_is_file_cache(page));
Nick Piggin62695a82008-10-18 20:26:09 -0700908 }
909 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800910}
911
Christoph Lameter742755a2006-06-23 02:03:55 -0700912static struct page *new_node_page(struct page *page, unsigned long node, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700913{
Mel Gorman6484eb32009-06-16 15:31:54 -0700914 return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700915}
916
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800917/*
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800918 * Migrate pages from one node to a target node.
919 * Returns error or the number of pages not migrated.
920 */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700921static int migrate_to_node(struct mm_struct *mm, int source, int dest,
922 int flags)
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800923{
924 nodemask_t nmask;
925 LIST_HEAD(pagelist);
926 int err = 0;
Vasiliy Kulikov0def08e2010-10-26 14:21:32 -0700927 struct vm_area_struct *vma;
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800928
929 nodes_clear(nmask);
930 node_set(source, nmask);
931
Vasiliy Kulikov0def08e2010-10-26 14:21:32 -0700932 vma = check_range(mm, mm->mmap->vm_start, mm->task_size, &nmask,
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800933 flags | MPOL_MF_DISCONTIG_OK, &pagelist);
Vasiliy Kulikov0def08e2010-10-26 14:21:32 -0700934 if (IS_ERR(vma))
935 return PTR_ERR(vma);
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800936
Minchan Kimcf608ac2010-10-26 14:21:29 -0700937 if (!list_empty(&pagelist)) {
Hugh Dickins62b61f62009-12-14 17:59:33 -0800938 err = migrate_pages(&pagelist, new_node_page, dest, 0);
Minchan Kimcf608ac2010-10-26 14:21:29 -0700939 if (err)
940 putback_lru_pages(&pagelist);
941 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700942
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800943 return err;
944}
945
946/*
947 * Move pages between the two nodesets so as to preserve the physical
948 * layout as much as possible.
Christoph Lameter39743882006-01-08 01:00:51 -0800949 *
950 * Returns the number of page that could not be moved.
951 */
952int do_migrate_pages(struct mm_struct *mm,
953 const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags)
954{
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800955 int busy = 0;
Christoph Lameter0aedadf2008-11-06 12:53:30 -0800956 int err;
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800957 nodemask_t tmp;
Christoph Lameter39743882006-01-08 01:00:51 -0800958
Christoph Lameter0aedadf2008-11-06 12:53:30 -0800959 err = migrate_prep();
960 if (err)
961 return err;
962
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700963 down_read(&mm->mmap_sem);
Christoph Lameter39743882006-01-08 01:00:51 -0800964
Christoph Lameter7b2259b2006-06-25 05:46:48 -0700965 err = migrate_vmas(mm, from_nodes, to_nodes, flags);
966 if (err)
967 goto out;
968
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -0800969 /*
970 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
971 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
972 * bit in 'tmp', and return that <source, dest> pair for migration.
973 * The pair of nodemasks 'to' and 'from' define the map.
974 *
975 * If no pair of bits is found that way, fallback to picking some
976 * pair of 'source' and 'dest' bits that are not the same. If the
977 * 'source' and 'dest' bits are the same, this represents a node
978 * that will be migrating to itself, so no pages need move.
979 *
980 * If no bits are left in 'tmp', or if all remaining bits left
981 * in 'tmp' correspond to the same bit in 'to', return false
982 * (nothing left to migrate).
983 *
984 * This lets us pick a pair of nodes to migrate between, such that
985 * if possible the dest node is not already occupied by some other
986 * source node, minimizing the risk of overloading the memory on a
987 * node that would happen if we migrated incoming memory to a node
988 * before migrating outgoing memory source that same node.
989 *
990 * A single scan of tmp is sufficient. As we go, we remember the
991 * most recent <s, d> pair that moved (s != d). If we find a pair
992 * that not only moved, but what's better, moved to an empty slot
993 * (d is not set in tmp), then we break out then, with that pair.
994 * Otherwise when we finish scannng from_tmp, we at least have the
995 * most recent <s, d> pair that moved. If we get all the way through
996 * the scan of tmp without finding any node that moved, much less
997 * moved to an empty node, then there is nothing left worth migrating.
998 */
Christoph Lameterd4984712006-01-08 01:00:55 -0800999
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001000 tmp = *from_nodes;
1001 while (!nodes_empty(tmp)) {
1002 int s,d;
1003 int source = -1;
1004 int dest = 0;
1005
1006 for_each_node_mask(s, tmp) {
1007 d = node_remap(s, *from_nodes, *to_nodes);
1008 if (s == d)
1009 continue;
1010
1011 source = s; /* Node moved. Memorize */
1012 dest = d;
1013
1014 /* dest not in remaining from nodes? */
1015 if (!node_isset(dest, tmp))
1016 break;
1017 }
1018 if (source == -1)
1019 break;
1020
1021 node_clear(source, tmp);
1022 err = migrate_to_node(mm, source, dest, flags);
1023 if (err > 0)
1024 busy += err;
1025 if (err < 0)
1026 break;
Christoph Lameter39743882006-01-08 01:00:51 -08001027 }
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001028out:
Christoph Lameter39743882006-01-08 01:00:51 -08001029 up_read(&mm->mmap_sem);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001030 if (err < 0)
1031 return err;
1032 return busy;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001033
Christoph Lameter39743882006-01-08 01:00:51 -08001034}
1035
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001036/*
1037 * Allocate a new page for page migration based on vma policy.
1038 * Start assuming that page is mapped by vma pointed to by @private.
1039 * Search forward from there, if not. N.B., this assumes that the
1040 * list of pages handed to migrate_pages()--which is how we get here--
1041 * is in virtual address order.
1042 */
Christoph Lameter742755a2006-06-23 02:03:55 -07001043static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001044{
1045 struct vm_area_struct *vma = (struct vm_area_struct *)private;
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001046 unsigned long uninitialized_var(address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001047
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001048 while (vma) {
1049 address = page_address_in_vma(page, vma);
1050 if (address != -EFAULT)
1051 break;
1052 vma = vma->vm_next;
1053 }
1054
1055 /*
1056 * if !vma, alloc_page_vma() will use task or system default policy
1057 */
1058 return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001059}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001060#else
1061
1062static void migrate_page_add(struct page *page, struct list_head *pagelist,
1063 unsigned long flags)
1064{
1065}
1066
1067int do_migrate_pages(struct mm_struct *mm,
1068 const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags)
1069{
1070 return -ENOSYS;
1071}
Christoph Lameter95a402c2006-06-23 02:03:53 -07001072
Keith Owens69939742006-10-11 01:21:28 -07001073static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001074{
1075 return NULL;
1076}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001077#endif
1078
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001079static long do_mbind(unsigned long start, unsigned long len,
David Rientjes028fec42008-04-28 02:12:25 -07001080 unsigned short mode, unsigned short mode_flags,
1081 nodemask_t *nmask, unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001082{
1083 struct vm_area_struct *vma;
1084 struct mm_struct *mm = current->mm;
1085 struct mempolicy *new;
1086 unsigned long end;
1087 int err;
1088 LIST_HEAD(pagelist);
1089
David Rientjesa3b51e02008-04-28 02:12:23 -07001090 if (flags & ~(unsigned long)(MPOL_MF_STRICT |
1091 MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001092 return -EINVAL;
Christoph Lameter74c00242006-03-14 19:50:21 -08001093 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001094 return -EPERM;
1095
1096 if (start & ~PAGE_MASK)
1097 return -EINVAL;
1098
1099 if (mode == MPOL_DEFAULT)
1100 flags &= ~MPOL_MF_STRICT;
1101
1102 len = (len + PAGE_SIZE - 1) & PAGE_MASK;
1103 end = start + len;
1104
1105 if (end < start)
1106 return -EINVAL;
1107 if (end == start)
1108 return 0;
1109
David Rientjes028fec42008-04-28 02:12:25 -07001110 new = mpol_new(mode, mode_flags, nmask);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001111 if (IS_ERR(new))
1112 return PTR_ERR(new);
1113
1114 /*
1115 * If we are using the default policy then operation
1116 * on discontinuous address spaces is okay after all
1117 */
1118 if (!new)
1119 flags |= MPOL_MF_DISCONTIG_OK;
1120
David Rientjes028fec42008-04-28 02:12:25 -07001121 pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
1122 start, start + len, mode, mode_flags,
1123 nmask ? nodes_addr(*nmask)[0] : -1);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001124
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001125 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
1126
1127 err = migrate_prep();
1128 if (err)
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001129 goto mpol_out;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001130 }
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07001131 {
1132 NODEMASK_SCRATCH(scratch);
1133 if (scratch) {
1134 down_write(&mm->mmap_sem);
1135 task_lock(current);
1136 err = mpol_set_nodemask(new, nmask, scratch);
1137 task_unlock(current);
1138 if (err)
1139 up_write(&mm->mmap_sem);
1140 } else
1141 err = -ENOMEM;
1142 NODEMASK_SCRATCH_FREE(scratch);
1143 }
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001144 if (err)
1145 goto mpol_out;
1146
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001147 vma = check_range(mm, start, end, nmask,
1148 flags | MPOL_MF_INVERT, &pagelist);
1149
1150 err = PTR_ERR(vma);
1151 if (!IS_ERR(vma)) {
1152 int nr_failed = 0;
1153
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -08001154 err = mbind_range(mm, start, end, new);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001155
Minchan Kimcf608ac2010-10-26 14:21:29 -07001156 if (!list_empty(&pagelist)) {
Christoph Lameter95a402c2006-06-23 02:03:53 -07001157 nr_failed = migrate_pages(&pagelist, new_vma_page,
Hugh Dickins62b61f62009-12-14 17:59:33 -08001158 (unsigned long)vma, 0);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001159 if (nr_failed)
1160 putback_lru_pages(&pagelist);
1161 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001162
1163 if (!err && nr_failed && (flags & MPOL_MF_STRICT))
1164 err = -EIO;
KOSAKI Motohiroab8a3e12009-10-26 16:49:58 -07001165 } else
1166 putback_lru_pages(&pagelist);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001167
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001168 up_write(&mm->mmap_sem);
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001169 mpol_out:
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001170 mpol_put(new);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001171 return err;
1172}
1173
Christoph Lameter39743882006-01-08 01:00:51 -08001174/*
Christoph Lameter8bccd852005-10-29 18:16:59 -07001175 * User space interface with variable sized bitmaps for nodelists.
1176 */
1177
1178/* Copy a node mask from user space. */
Christoph Lameter39743882006-01-08 01:00:51 -08001179static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
Christoph Lameter8bccd852005-10-29 18:16:59 -07001180 unsigned long maxnode)
1181{
1182 unsigned long k;
1183 unsigned long nlongs;
1184 unsigned long endmask;
1185
1186 --maxnode;
1187 nodes_clear(*nodes);
1188 if (maxnode == 0 || !nmask)
1189 return 0;
Andi Kleena9c930b2006-02-20 18:27:59 -08001190 if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
Chris Wright636f13c2006-02-17 13:59:36 -08001191 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001192
1193 nlongs = BITS_TO_LONGS(maxnode);
1194 if ((maxnode % BITS_PER_LONG) == 0)
1195 endmask = ~0UL;
1196 else
1197 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
1198
1199 /* When the user specified more nodes than supported just check
1200 if the non supported part is all zero. */
1201 if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
1202 if (nlongs > PAGE_SIZE/sizeof(long))
1203 return -EINVAL;
1204 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
1205 unsigned long t;
1206 if (get_user(t, nmask + k))
1207 return -EFAULT;
1208 if (k == nlongs - 1) {
1209 if (t & endmask)
1210 return -EINVAL;
1211 } else if (t)
1212 return -EINVAL;
1213 }
1214 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
1215 endmask = ~0UL;
1216 }
1217
1218 if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
1219 return -EFAULT;
1220 nodes_addr(*nodes)[nlongs-1] &= endmask;
1221 return 0;
1222}
1223
1224/* Copy a kernel node mask to user space */
1225static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1226 nodemask_t *nodes)
1227{
1228 unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1229 const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long);
1230
1231 if (copy > nbytes) {
1232 if (copy > PAGE_SIZE)
1233 return -EINVAL;
1234 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1235 return -EFAULT;
1236 copy = nbytes;
1237 }
1238 return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1239}
1240
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001241SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
1242 unsigned long, mode, unsigned long __user *, nmask,
1243 unsigned long, maxnode, unsigned, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001244{
1245 nodemask_t nodes;
1246 int err;
David Rientjes028fec42008-04-28 02:12:25 -07001247 unsigned short mode_flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001248
David Rientjes028fec42008-04-28 02:12:25 -07001249 mode_flags = mode & MPOL_MODE_FLAGS;
1250 mode &= ~MPOL_MODE_FLAGS;
David Rientjesa3b51e02008-04-28 02:12:23 -07001251 if (mode >= MPOL_MAX)
1252 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001253 if ((mode_flags & MPOL_F_STATIC_NODES) &&
1254 (mode_flags & MPOL_F_RELATIVE_NODES))
1255 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001256 err = get_nodes(&nodes, nmask, maxnode);
1257 if (err)
1258 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001259 return do_mbind(start, len, mode, mode_flags, &nodes, flags);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001260}
1261
1262/* Set the process memory policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001263SYSCALL_DEFINE3(set_mempolicy, int, mode, unsigned long __user *, nmask,
1264 unsigned long, maxnode)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001265{
1266 int err;
1267 nodemask_t nodes;
David Rientjes028fec42008-04-28 02:12:25 -07001268 unsigned short flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001269
David Rientjes028fec42008-04-28 02:12:25 -07001270 flags = mode & MPOL_MODE_FLAGS;
1271 mode &= ~MPOL_MODE_FLAGS;
1272 if ((unsigned int)mode >= MPOL_MAX)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001273 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001274 if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
1275 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001276 err = get_nodes(&nodes, nmask, maxnode);
1277 if (err)
1278 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001279 return do_set_mempolicy(mode, flags, &nodes);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001280}
1281
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001282SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
1283 const unsigned long __user *, old_nodes,
1284 const unsigned long __user *, new_nodes)
Christoph Lameter39743882006-01-08 01:00:51 -08001285{
David Howellsc69e8d92008-11-14 10:39:19 +11001286 const struct cred *cred = current_cred(), *tcred;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001287 struct mm_struct *mm = NULL;
Christoph Lameter39743882006-01-08 01:00:51 -08001288 struct task_struct *task;
Christoph Lameter39743882006-01-08 01:00:51 -08001289 nodemask_t task_nodes;
1290 int err;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001291 nodemask_t *old;
1292 nodemask_t *new;
1293 NODEMASK_SCRATCH(scratch);
Christoph Lameter39743882006-01-08 01:00:51 -08001294
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001295 if (!scratch)
1296 return -ENOMEM;
Christoph Lameter39743882006-01-08 01:00:51 -08001297
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001298 old = &scratch->mask1;
1299 new = &scratch->mask2;
1300
1301 err = get_nodes(old, old_nodes, maxnode);
Christoph Lameter39743882006-01-08 01:00:51 -08001302 if (err)
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001303 goto out;
1304
1305 err = get_nodes(new, new_nodes, maxnode);
1306 if (err)
1307 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001308
1309 /* Find the mm_struct */
1310 read_lock(&tasklist_lock);
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001311 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter39743882006-01-08 01:00:51 -08001312 if (!task) {
1313 read_unlock(&tasklist_lock);
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001314 err = -ESRCH;
1315 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001316 }
1317 mm = get_task_mm(task);
1318 read_unlock(&tasklist_lock);
1319
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001320 err = -EINVAL;
Christoph Lameter39743882006-01-08 01:00:51 -08001321 if (!mm)
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001322 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001323
1324 /*
1325 * Check if this process has the right to modify the specified
1326 * process. The right exists if the process has administrative
Alexey Dobriyan7f927fc2006-03-28 01:56:53 -08001327 * capabilities, superuser privileges or the same
Christoph Lameter39743882006-01-08 01:00:51 -08001328 * userid as the target process.
1329 */
David Howellsc69e8d92008-11-14 10:39:19 +11001330 rcu_read_lock();
1331 tcred = __task_cred(task);
David Howellsb6dff3e2008-11-14 10:39:16 +11001332 if (cred->euid != tcred->suid && cred->euid != tcred->uid &&
1333 cred->uid != tcred->suid && cred->uid != tcred->uid &&
Christoph Lameter74c00242006-03-14 19:50:21 -08001334 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001335 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001336 err = -EPERM;
1337 goto out;
1338 }
David Howellsc69e8d92008-11-14 10:39:19 +11001339 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001340
1341 task_nodes = cpuset_mems_allowed(task);
1342 /* Is the user allowed to access the target nodes? */
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001343 if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
Christoph Lameter39743882006-01-08 01:00:51 -08001344 err = -EPERM;
1345 goto out;
1346 }
1347
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001348 if (!nodes_subset(*new, node_states[N_HIGH_MEMORY])) {
Christoph Lameter3b42d282007-08-31 00:12:08 -07001349 err = -EINVAL;
1350 goto out;
1351 }
1352
David Quigley86c3a762006-06-23 02:04:02 -07001353 err = security_task_movememory(task);
1354 if (err)
1355 goto out;
1356
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001357 err = do_migrate_pages(mm, old, new,
Christoph Lameter74c00242006-03-14 19:50:21 -08001358 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
Christoph Lameter39743882006-01-08 01:00:51 -08001359out:
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001360 if (mm)
1361 mmput(mm);
1362 NODEMASK_SCRATCH_FREE(scratch);
1363
Christoph Lameter39743882006-01-08 01:00:51 -08001364 return err;
1365}
1366
1367
Christoph Lameter8bccd852005-10-29 18:16:59 -07001368/* Retrieve NUMA policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001369SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1370 unsigned long __user *, nmask, unsigned long, maxnode,
1371 unsigned long, addr, unsigned long, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001372{
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001373 int err;
1374 int uninitialized_var(pval);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001375 nodemask_t nodes;
1376
1377 if (nmask != NULL && maxnode < MAX_NUMNODES)
1378 return -EINVAL;
1379
1380 err = do_get_mempolicy(&pval, &nodes, addr, flags);
1381
1382 if (err)
1383 return err;
1384
1385 if (policy && put_user(pval, policy))
1386 return -EFAULT;
1387
1388 if (nmask)
1389 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1390
1391 return err;
1392}
1393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394#ifdef CONFIG_COMPAT
1395
1396asmlinkage long compat_sys_get_mempolicy(int __user *policy,
1397 compat_ulong_t __user *nmask,
1398 compat_ulong_t maxnode,
1399 compat_ulong_t addr, compat_ulong_t flags)
1400{
1401 long err;
1402 unsigned long __user *nm = NULL;
1403 unsigned long nr_bits, alloc_size;
1404 DECLARE_BITMAP(bm, MAX_NUMNODES);
1405
1406 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1407 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1408
1409 if (nmask)
1410 nm = compat_alloc_user_space(alloc_size);
1411
1412 err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
1413
1414 if (!err && nmask) {
1415 err = copy_from_user(bm, nm, alloc_size);
1416 /* ensure entire bitmap is zeroed */
1417 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
1418 err |= compat_put_bitmap(nmask, bm, nr_bits);
1419 }
1420
1421 return err;
1422}
1423
1424asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask,
1425 compat_ulong_t maxnode)
1426{
1427 long err = 0;
1428 unsigned long __user *nm = NULL;
1429 unsigned long nr_bits, alloc_size;
1430 DECLARE_BITMAP(bm, MAX_NUMNODES);
1431
1432 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1433 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1434
1435 if (nmask) {
1436 err = compat_get_bitmap(bm, nmask, nr_bits);
1437 nm = compat_alloc_user_space(alloc_size);
1438 err |= copy_to_user(nm, bm, alloc_size);
1439 }
1440
1441 if (err)
1442 return -EFAULT;
1443
1444 return sys_set_mempolicy(mode, nm, nr_bits+1);
1445}
1446
1447asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len,
1448 compat_ulong_t mode, compat_ulong_t __user *nmask,
1449 compat_ulong_t maxnode, compat_ulong_t flags)
1450{
1451 long err = 0;
1452 unsigned long __user *nm = NULL;
1453 unsigned long nr_bits, alloc_size;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001454 nodemask_t bm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
1456 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1457 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1458
1459 if (nmask) {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001460 err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 nm = compat_alloc_user_space(alloc_size);
Andi Kleendfcd3c02005-10-29 18:15:48 -07001462 err |= copy_to_user(nm, nodes_addr(bm), alloc_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 }
1464
1465 if (err)
1466 return -EFAULT;
1467
1468 return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
1469}
1470
1471#endif
1472
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001473/*
1474 * get_vma_policy(@task, @vma, @addr)
1475 * @task - task for fallback if vma policy == default
1476 * @vma - virtual memory area whose policy is sought
1477 * @addr - address in @vma for shared policy lookup
1478 *
1479 * Returns effective policy for a VMA at specified address.
1480 * Falls back to @task or system default policy, as necessary.
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001481 * Current or other task's task mempolicy and non-shared vma policies
1482 * are protected by the task's mmap_sem, which must be held for read by
1483 * the caller.
1484 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1485 * count--added by the get_policy() vm_op, as appropriate--to protect against
1486 * freeing by another task. It is the caller's responsibility to free the
1487 * extra reference for shared policies.
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001488 */
Lee Schermerhornae4d8c12008-04-28 02:13:11 -07001489static struct mempolicy *get_vma_policy(struct task_struct *task,
Christoph Lameter48fce342006-01-08 01:01:03 -08001490 struct vm_area_struct *vma, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491{
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07001492 struct mempolicy *pol = task->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493
1494 if (vma) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001495 if (vma->vm_ops && vma->vm_ops->get_policy) {
Lee Schermerhornae4d8c12008-04-28 02:13:11 -07001496 struct mempolicy *vpol = vma->vm_ops->get_policy(vma,
1497 addr);
1498 if (vpol)
1499 pol = vpol;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001500 } else if (vma->vm_policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 pol = vma->vm_policy;
1502 }
1503 if (!pol)
1504 pol = &default_policy;
1505 return pol;
1506}
1507
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001508/*
1509 * Return a nodemask representing a mempolicy for filtering nodes for
1510 * page allocation
1511 */
1512static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
Mel Gorman19770b32008-04-28 02:12:18 -07001513{
1514 /* Lower zones don't get a nodemask applied for MPOL_BIND */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001515 if (unlikely(policy->mode == MPOL_BIND) &&
Mel Gorman19770b32008-04-28 02:12:18 -07001516 gfp_zone(gfp) >= policy_zone &&
1517 cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
1518 return &policy->v.nodes;
1519
1520 return NULL;
1521}
1522
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001523/* Return a zonelist indicated by gfp for node representing a mempolicy */
1524static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525{
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001526 int nd = numa_node_id();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001528 switch (policy->mode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001530 if (!(policy->flags & MPOL_F_LOCAL))
1531 nd = policy->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 break;
1533 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07001534 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001535 * Normally, MPOL_BIND allocations are node-local within the
1536 * allowed nodemask. However, if __GFP_THISNODE is set and the
Bob Liu6eb27e12010-05-24 14:32:00 -07001537 * current node isn't part of the mask, we use the zonelist for
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001538 * the first node in the mask instead.
Mel Gorman19770b32008-04-28 02:12:18 -07001539 */
Mel Gorman19770b32008-04-28 02:12:18 -07001540 if (unlikely(gfp & __GFP_THISNODE) &&
1541 unlikely(!node_isset(nd, policy->v.nodes)))
1542 nd = first_node(policy->v.nodes);
1543 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 BUG();
1546 }
Mel Gorman0e884602008-04-28 02:12:14 -07001547 return node_zonelist(nd, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548}
1549
1550/* Do dynamic interleaving for a process */
1551static unsigned interleave_nodes(struct mempolicy *policy)
1552{
1553 unsigned nid, next;
1554 struct task_struct *me = current;
1555
1556 nid = me->il_next;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001557 next = next_node(nid, policy->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 if (next >= MAX_NUMNODES)
Andi Kleendfcd3c02005-10-29 18:15:48 -07001559 next = first_node(policy->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001560 if (next < MAX_NUMNODES)
1561 me->il_next = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 return nid;
1563}
1564
Christoph Lameterdc85da12006-01-18 17:42:36 -08001565/*
1566 * Depending on the memory policy provide a node from which to allocate the
1567 * next slab entry.
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001568 * @policy must be protected by freeing by the caller. If @policy is
1569 * the current task's mempolicy, this protection is implicit, as only the
1570 * task can change it's policy. The system default policy requires no
1571 * such protection.
Christoph Lameterdc85da12006-01-18 17:42:36 -08001572 */
1573unsigned slab_node(struct mempolicy *policy)
1574{
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001575 if (!policy || policy->flags & MPOL_F_LOCAL)
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001576 return numa_node_id();
Christoph Lameter765c4502006-09-27 01:50:08 -07001577
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001578 switch (policy->mode) {
1579 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001580 /*
1581 * handled MPOL_F_LOCAL above
1582 */
1583 return policy->v.preferred_node;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001584
Christoph Lameterdc85da12006-01-18 17:42:36 -08001585 case MPOL_INTERLEAVE:
1586 return interleave_nodes(policy);
1587
Mel Gormandd1a2392008-04-28 02:12:17 -07001588 case MPOL_BIND: {
Christoph Lameterdc85da12006-01-18 17:42:36 -08001589 /*
1590 * Follow bind policy behavior and start allocation at the
1591 * first node.
1592 */
Mel Gorman19770b32008-04-28 02:12:18 -07001593 struct zonelist *zonelist;
1594 struct zone *zone;
1595 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
1596 zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0];
1597 (void)first_zones_zonelist(zonelist, highest_zoneidx,
1598 &policy->v.nodes,
1599 &zone);
Eric Dumazet800416f2010-10-27 19:33:43 +02001600 return zone ? zone->node : numa_node_id();
Mel Gormandd1a2392008-04-28 02:12:17 -07001601 }
Christoph Lameterdc85da12006-01-18 17:42:36 -08001602
Christoph Lameterdc85da12006-01-18 17:42:36 -08001603 default:
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001604 BUG();
Christoph Lameterdc85da12006-01-18 17:42:36 -08001605 }
1606}
1607
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608/* Do static interleaving for a VMA with known offset. */
1609static unsigned offset_il_node(struct mempolicy *pol,
1610 struct vm_area_struct *vma, unsigned long off)
1611{
Andi Kleendfcd3c02005-10-29 18:15:48 -07001612 unsigned nnodes = nodes_weight(pol->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001613 unsigned target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 int c;
1615 int nid = -1;
1616
David Rientjesf5b087b2008-04-28 02:12:27 -07001617 if (!nnodes)
1618 return numa_node_id();
1619 target = (unsigned int)off % nnodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 c = 0;
1621 do {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001622 nid = next_node(nid, pol->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 c++;
1624 } while (c <= target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 return nid;
1626}
1627
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001628/* Determine a node number for interleave */
1629static inline unsigned interleave_nid(struct mempolicy *pol,
1630 struct vm_area_struct *vma, unsigned long addr, int shift)
1631{
1632 if (vma) {
1633 unsigned long off;
1634
Nishanth Aravamudan3b98b082006-08-31 21:27:53 -07001635 /*
1636 * for small pages, there is no difference between
1637 * shift and PAGE_SHIFT, so the bit-shift is safe.
1638 * for huge pages, since vm_pgoff is in units of small
1639 * pages, we need to shift off the always 0 bits to get
1640 * a useful offset.
1641 */
1642 BUG_ON(shift < PAGE_SHIFT);
1643 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001644 off += (addr - vma->vm_start) >> shift;
1645 return offset_il_node(pol, vma, off);
1646 } else
1647 return interleave_nodes(pol);
1648}
1649
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001650#ifdef CONFIG_HUGETLBFS
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001651/*
1652 * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
1653 * @vma = virtual memory area whose policy is sought
1654 * @addr = address in @vma for shared policy lookup and interleave policy
1655 * @gfp_flags = for requested zone
Mel Gorman19770b32008-04-28 02:12:18 -07001656 * @mpol = pointer to mempolicy pointer for reference counted mempolicy
1657 * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001658 *
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001659 * Returns a zonelist suitable for a huge page allocation and a pointer
1660 * to the struct mempolicy for conditional unref after allocation.
1661 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1662 * @nodemask for filtering the zonelist.
Miao Xiec0ff7452010-05-24 14:32:08 -07001663 *
1664 * Must be protected by get_mems_allowed()
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001665 */
Mel Gorman396faf02007-07-17 04:03:13 -07001666struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
Mel Gorman19770b32008-04-28 02:12:18 -07001667 gfp_t gfp_flags, struct mempolicy **mpol,
1668 nodemask_t **nodemask)
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001669{
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001670 struct zonelist *zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001671
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001672 *mpol = get_vma_policy(current, vma, addr);
Mel Gorman19770b32008-04-28 02:12:18 -07001673 *nodemask = NULL; /* assume !MPOL_BIND */
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001674
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001675 if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
1676 zl = node_zonelist(interleave_nid(*mpol, vma, addr,
Andi Kleena5516432008-07-23 21:27:41 -07001677 huge_page_shift(hstate_vma(vma))), gfp_flags);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001678 } else {
1679 zl = policy_zonelist(gfp_flags, *mpol);
1680 if ((*mpol)->mode == MPOL_BIND)
1681 *nodemask = &(*mpol)->v.nodes;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001682 }
1683 return zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001684}
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001685
1686/*
1687 * init_nodemask_of_mempolicy
1688 *
1689 * If the current task's mempolicy is "default" [NULL], return 'false'
1690 * to indicate default policy. Otherwise, extract the policy nodemask
1691 * for 'bind' or 'interleave' policy into the argument nodemask, or
1692 * initialize the argument nodemask to contain the single node for
1693 * 'preferred' or 'local' policy and return 'true' to indicate presence
1694 * of non-default mempolicy.
1695 *
1696 * We don't bother with reference counting the mempolicy [mpol_get/put]
1697 * because the current task is examining it's own mempolicy and a task's
1698 * mempolicy is only ever changed by the task itself.
1699 *
1700 * N.B., it is the caller's responsibility to free a returned nodemask.
1701 */
1702bool init_nodemask_of_mempolicy(nodemask_t *mask)
1703{
1704 struct mempolicy *mempolicy;
1705 int nid;
1706
1707 if (!(mask && current->mempolicy))
1708 return false;
1709
Miao Xiec0ff7452010-05-24 14:32:08 -07001710 task_lock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001711 mempolicy = current->mempolicy;
1712 switch (mempolicy->mode) {
1713 case MPOL_PREFERRED:
1714 if (mempolicy->flags & MPOL_F_LOCAL)
1715 nid = numa_node_id();
1716 else
1717 nid = mempolicy->v.preferred_node;
1718 init_nodemask_of_node(mask, nid);
1719 break;
1720
1721 case MPOL_BIND:
1722 /* Fall through */
1723 case MPOL_INTERLEAVE:
1724 *mask = mempolicy->v.nodes;
1725 break;
1726
1727 default:
1728 BUG();
1729 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001730 task_unlock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001731
1732 return true;
1733}
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001734#endif
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001735
David Rientjes6f48d0eb2010-08-09 17:18:52 -07001736/*
1737 * mempolicy_nodemask_intersects
1738 *
1739 * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default
1740 * policy. Otherwise, check for intersection between mask and the policy
1741 * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local'
1742 * policy, always return true since it may allocate elsewhere on fallback.
1743 *
1744 * Takes task_lock(tsk) to prevent freeing of its mempolicy.
1745 */
1746bool mempolicy_nodemask_intersects(struct task_struct *tsk,
1747 const nodemask_t *mask)
1748{
1749 struct mempolicy *mempolicy;
1750 bool ret = true;
1751
1752 if (!mask)
1753 return ret;
1754 task_lock(tsk);
1755 mempolicy = tsk->mempolicy;
1756 if (!mempolicy)
1757 goto out;
1758
1759 switch (mempolicy->mode) {
1760 case MPOL_PREFERRED:
1761 /*
1762 * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to
1763 * allocate from, they may fallback to other nodes when oom.
1764 * Thus, it's possible for tsk to have allocated memory from
1765 * nodes in mask.
1766 */
1767 break;
1768 case MPOL_BIND:
1769 case MPOL_INTERLEAVE:
1770 ret = nodes_intersects(mempolicy->v.nodes, *mask);
1771 break;
1772 default:
1773 BUG();
1774 }
1775out:
1776 task_unlock(tsk);
1777 return ret;
1778}
1779
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780/* Allocate a page in interleaved policy.
1781 Own path because it needs to do special accounting. */
Andi Kleen662f3a02005-10-29 18:15:49 -07001782static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
1783 unsigned nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784{
1785 struct zonelist *zl;
1786 struct page *page;
1787
Mel Gorman0e884602008-04-28 02:12:14 -07001788 zl = node_zonelist(nid, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 page = __alloc_pages(gfp, order, zl);
Mel Gormandd1a2392008-04-28 02:12:17 -07001790 if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
Christoph Lameterca889e62006-06-30 01:55:44 -07001791 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 return page;
1793}
1794
1795/**
1796 * alloc_page_vma - Allocate a page for a VMA.
1797 *
1798 * @gfp:
1799 * %GFP_USER user allocation.
1800 * %GFP_KERNEL kernel allocations,
1801 * %GFP_HIGHMEM highmem/user allocations,
1802 * %GFP_FS allocation should not call back into a file system.
1803 * %GFP_ATOMIC don't sleep.
1804 *
1805 * @vma: Pointer to VMA or NULL if not available.
1806 * @addr: Virtual Address of the allocation. Must be inside the VMA.
1807 *
1808 * This function allocates a page from the kernel page pool and applies
1809 * a NUMA policy associated with the VMA or the current process.
1810 * When VMA is not NULL caller must hold down_read on the mmap_sem of the
1811 * mm_struct of the VMA to prevent it from going away. Should be used for
1812 * all allocations for pages that will be mapped into
1813 * user space. Returns NULL when no page can be allocated.
1814 *
1815 * Should be called with the mm_sem of the vma hold.
1816 */
1817struct page *
Al Virodd0fc662005-10-07 07:46:04 +01001818alloc_page_vma(gfp_t gfp, struct vm_area_struct *vma, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819{
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07001820 struct mempolicy *pol = get_vma_policy(current, vma, addr);
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001821 struct zonelist *zl;
Miao Xiec0ff7452010-05-24 14:32:08 -07001822 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
Miao Xiec0ff7452010-05-24 14:32:08 -07001824 get_mems_allowed();
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001825 if (unlikely(pol->mode == MPOL_INTERLEAVE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 unsigned nid;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001827
1828 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001829 mpol_cond_put(pol);
Miao Xiec0ff7452010-05-24 14:32:08 -07001830 page = alloc_page_interleave(gfp, 0, nid);
1831 put_mems_allowed();
1832 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 }
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001834 zl = policy_zonelist(gfp, pol);
1835 if (unlikely(mpol_needs_cond_ref(pol))) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001836 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001837 * slow path: ref counted shared policy
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001838 */
Mel Gorman19770b32008-04-28 02:12:18 -07001839 struct page *page = __alloc_pages_nodemask(gfp, 0,
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001840 zl, policy_nodemask(gfp, pol));
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001841 __mpol_put(pol);
Miao Xiec0ff7452010-05-24 14:32:08 -07001842 put_mems_allowed();
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001843 return page;
1844 }
1845 /*
1846 * fast path: default or task policy
1847 */
Miao Xiec0ff7452010-05-24 14:32:08 -07001848 page = __alloc_pages_nodemask(gfp, 0, zl, policy_nodemask(gfp, pol));
1849 put_mems_allowed();
1850 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851}
1852
1853/**
1854 * alloc_pages_current - Allocate pages.
1855 *
1856 * @gfp:
1857 * %GFP_USER user allocation,
1858 * %GFP_KERNEL kernel allocation,
1859 * %GFP_HIGHMEM highmem allocation,
1860 * %GFP_FS don't call back into a file system.
1861 * %GFP_ATOMIC don't sleep.
1862 * @order: Power of two of allocation size in pages. 0 is a single page.
1863 *
1864 * Allocate a page from the kernel page pool. When not in
1865 * interrupt context and apply the current process NUMA policy.
1866 * Returns NULL when no page can be allocated.
1867 *
Paul Jacksoncf2a473c2006-01-08 01:01:54 -08001868 * Don't call cpuset_update_task_memory_state() unless
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 * 1) it's ok to take cpuset_sem (can WAIT), and
1870 * 2) allocating for current task (not interrupt).
1871 */
Al Virodd0fc662005-10-07 07:46:04 +01001872struct page *alloc_pages_current(gfp_t gfp, unsigned order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873{
1874 struct mempolicy *pol = current->mempolicy;
Miao Xiec0ff7452010-05-24 14:32:08 -07001875 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876
Christoph Lameter9b819d22006-09-25 23:31:40 -07001877 if (!pol || in_interrupt() || (gfp & __GFP_THISNODE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 pol = &default_policy;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001879
Miao Xiec0ff7452010-05-24 14:32:08 -07001880 get_mems_allowed();
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001881 /*
1882 * No reference counting needed for current->mempolicy
1883 * nor system default_policy
1884 */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001885 if (pol->mode == MPOL_INTERLEAVE)
Miao Xiec0ff7452010-05-24 14:32:08 -07001886 page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
1887 else
1888 page = __alloc_pages_nodemask(gfp, order,
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001889 policy_zonelist(gfp, pol), policy_nodemask(gfp, pol));
Miao Xiec0ff7452010-05-24 14:32:08 -07001890 put_mems_allowed();
1891 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892}
1893EXPORT_SYMBOL(alloc_pages_current);
1894
Paul Jackson42253992006-01-08 01:01:59 -08001895/*
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001896 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
Paul Jackson42253992006-01-08 01:01:59 -08001897 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
1898 * with the mems_allowed returned by cpuset_mems_allowed(). This
1899 * keeps mempolicies cpuset relative after its cpuset moves. See
1900 * further kernel/cpuset.c update_nodemask().
Miao Xie708c1bb2010-05-24 14:32:07 -07001901 *
1902 * current's mempolicy may be rebinded by the other task(the task that changes
1903 * cpuset's mems), so we needn't do rebind work for current task.
Paul Jackson42253992006-01-08 01:01:59 -08001904 */
Paul Jackson42253992006-01-08 01:01:59 -08001905
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001906/* Slow path of a mempolicy duplicate */
1907struct mempolicy *__mpol_dup(struct mempolicy *old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908{
1909 struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
1910
1911 if (!new)
1912 return ERR_PTR(-ENOMEM);
Miao Xie708c1bb2010-05-24 14:32:07 -07001913
1914 /* task's mempolicy is protected by alloc_lock */
1915 if (old == current->mempolicy) {
1916 task_lock(current);
1917 *new = *old;
1918 task_unlock(current);
1919 } else
1920 *new = *old;
1921
Paul E. McKenney99ee4ca2010-03-03 17:50:17 -08001922 rcu_read_lock();
Paul Jackson42253992006-01-08 01:01:59 -08001923 if (current_cpuset_is_being_rebound()) {
1924 nodemask_t mems = cpuset_mems_allowed(current);
Miao Xie708c1bb2010-05-24 14:32:07 -07001925 if (new->flags & MPOL_F_REBINDING)
1926 mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2);
1927 else
1928 mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE);
Paul Jackson42253992006-01-08 01:01:59 -08001929 }
Paul E. McKenney99ee4ca2010-03-03 17:50:17 -08001930 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 atomic_set(&new->refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 return new;
1933}
1934
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001935/*
1936 * If *frompol needs [has] an extra ref, copy *frompol to *tompol ,
1937 * eliminate the * MPOL_F_* flags that require conditional ref and
1938 * [NOTE!!!] drop the extra ref. Not safe to reference *frompol directly
1939 * after return. Use the returned value.
1940 *
1941 * Allows use of a mempolicy for, e.g., multiple allocations with a single
1942 * policy lookup, even if the policy needs/has extra ref on lookup.
1943 * shmem_readahead needs this.
1944 */
1945struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol,
1946 struct mempolicy *frompol)
1947{
1948 if (!mpol_needs_cond_ref(frompol))
1949 return frompol;
1950
1951 *tompol = *frompol;
1952 tompol->flags &= ~MPOL_F_SHARED; /* copy doesn't need unref */
1953 __mpol_put(frompol);
1954 return tompol;
1955}
1956
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957/* Slow path of a mempolicy comparison */
1958int __mpol_equal(struct mempolicy *a, struct mempolicy *b)
1959{
1960 if (!a || !b)
1961 return 0;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001962 if (a->mode != b->mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 return 0;
Bob Liu19800502010-05-24 14:32:01 -07001964 if (a->flags != b->flags)
David Rientjesf5b087b2008-04-28 02:12:27 -07001965 return 0;
Bob Liu19800502010-05-24 14:32:01 -07001966 if (mpol_store_user_nodemask(a))
1967 if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask))
1968 return 0;
1969
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001970 switch (a->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -07001971 case MPOL_BIND:
1972 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -07001974 return nodes_equal(a->v.nodes, b->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001976 return a->v.preferred_node == b->v.preferred_node &&
1977 a->flags == b->flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 default:
1979 BUG();
1980 return 0;
1981 }
1982}
1983
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 * Shared memory backing store policy support.
1986 *
1987 * Remember policies even when nobody has shared memory mapped.
1988 * The policies are kept in Red-Black tree linked from the inode.
1989 * They are protected by the sp->lock spinlock, which should be held
1990 * for any accesses to the tree.
1991 */
1992
1993/* lookup first element intersecting start-end */
1994/* Caller holds sp->lock */
1995static struct sp_node *
1996sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
1997{
1998 struct rb_node *n = sp->root.rb_node;
1999
2000 while (n) {
2001 struct sp_node *p = rb_entry(n, struct sp_node, nd);
2002
2003 if (start >= p->end)
2004 n = n->rb_right;
2005 else if (end <= p->start)
2006 n = n->rb_left;
2007 else
2008 break;
2009 }
2010 if (!n)
2011 return NULL;
2012 for (;;) {
2013 struct sp_node *w = NULL;
2014 struct rb_node *prev = rb_prev(n);
2015 if (!prev)
2016 break;
2017 w = rb_entry(prev, struct sp_node, nd);
2018 if (w->end <= start)
2019 break;
2020 n = prev;
2021 }
2022 return rb_entry(n, struct sp_node, nd);
2023}
2024
2025/* Insert a new shared policy into the list. */
2026/* Caller holds sp->lock */
2027static void sp_insert(struct shared_policy *sp, struct sp_node *new)
2028{
2029 struct rb_node **p = &sp->root.rb_node;
2030 struct rb_node *parent = NULL;
2031 struct sp_node *nd;
2032
2033 while (*p) {
2034 parent = *p;
2035 nd = rb_entry(parent, struct sp_node, nd);
2036 if (new->start < nd->start)
2037 p = &(*p)->rb_left;
2038 else if (new->end > nd->end)
2039 p = &(*p)->rb_right;
2040 else
2041 BUG();
2042 }
2043 rb_link_node(&new->nd, parent, p);
2044 rb_insert_color(&new->nd, &sp->root);
Paul Mundt140d5a42007-07-15 23:38:16 -07002045 pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002046 new->policy ? new->policy->mode : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047}
2048
2049/* Find shared policy intersecting idx */
2050struct mempolicy *
2051mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
2052{
2053 struct mempolicy *pol = NULL;
2054 struct sp_node *sn;
2055
2056 if (!sp->root.rb_node)
2057 return NULL;
2058 spin_lock(&sp->lock);
2059 sn = sp_lookup(sp, idx, idx+1);
2060 if (sn) {
2061 mpol_get(sn->policy);
2062 pol = sn->policy;
2063 }
2064 spin_unlock(&sp->lock);
2065 return pol;
2066}
2067
2068static void sp_delete(struct shared_policy *sp, struct sp_node *n)
2069{
Paul Mundt140d5a42007-07-15 23:38:16 -07002070 pr_debug("deleting %lx-l%lx\n", n->start, n->end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 rb_erase(&n->nd, &sp->root);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07002072 mpol_put(n->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 kmem_cache_free(sn_cache, n);
2074}
2075
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07002076static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
2077 struct mempolicy *pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078{
2079 struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
2080
2081 if (!n)
2082 return NULL;
2083 n->start = start;
2084 n->end = end;
2085 mpol_get(pol);
Lee Schermerhornaab0b102008-04-28 02:13:13 -07002086 pol->flags |= MPOL_F_SHARED; /* for unref */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 n->policy = pol;
2088 return n;
2089}
2090
2091/* Replace a policy range. */
2092static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
2093 unsigned long end, struct sp_node *new)
2094{
2095 struct sp_node *n, *new2 = NULL;
2096
2097restart:
2098 spin_lock(&sp->lock);
2099 n = sp_lookup(sp, start, end);
2100 /* Take care of old policies in the same range. */
2101 while (n && n->start < end) {
2102 struct rb_node *next = rb_next(&n->nd);
2103 if (n->start >= start) {
2104 if (n->end <= end)
2105 sp_delete(sp, n);
2106 else
2107 n->start = end;
2108 } else {
2109 /* Old policy spanning whole new range. */
2110 if (n->end > end) {
2111 if (!new2) {
2112 spin_unlock(&sp->lock);
2113 new2 = sp_alloc(end, n->end, n->policy);
2114 if (!new2)
2115 return -ENOMEM;
2116 goto restart;
2117 }
2118 n->end = start;
2119 sp_insert(sp, new2);
2120 new2 = NULL;
2121 break;
2122 } else
2123 n->end = start;
2124 }
2125 if (!next)
2126 break;
2127 n = rb_entry(next, struct sp_node, nd);
2128 }
2129 if (new)
2130 sp_insert(sp, new);
2131 spin_unlock(&sp->lock);
2132 if (new2) {
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07002133 mpol_put(new2->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 kmem_cache_free(sn_cache, new2);
2135 }
2136 return 0;
2137}
2138
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002139/**
2140 * mpol_shared_policy_init - initialize shared policy for inode
2141 * @sp: pointer to inode shared policy
2142 * @mpol: struct mempolicy to install
2143 *
2144 * Install non-NULL @mpol in inode's shared policy rb-tree.
2145 * On entry, the current task has a reference on a non-NULL @mpol.
2146 * This must be released on exit.
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002147 * This is called at get_inode() calls and we can use GFP_KERNEL.
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002148 */
2149void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
Robin Holt7339ff82006-01-14 13:20:48 -08002150{
Miao Xie58568d22009-06-16 15:31:49 -07002151 int ret;
2152
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002153 sp->root = RB_ROOT; /* empty tree == default mempolicy */
2154 spin_lock_init(&sp->lock);
Robin Holt7339ff82006-01-14 13:20:48 -08002155
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002156 if (mpol) {
2157 struct vm_area_struct pvma;
2158 struct mempolicy *new;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002159 NODEMASK_SCRATCH(scratch);
Robin Holt7339ff82006-01-14 13:20:48 -08002160
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002161 if (!scratch)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002162 goto put_mpol;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002163 /* contextualize the tmpfs mount point mempolicy */
2164 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002165 if (IS_ERR(new))
Dan Carpenter0cae3452010-05-25 23:42:58 -07002166 goto free_scratch; /* no valid nodemask intersection */
Miao Xie58568d22009-06-16 15:31:49 -07002167
2168 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002169 ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
Miao Xie58568d22009-06-16 15:31:49 -07002170 task_unlock(current);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002171 if (ret)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002172 goto put_new;
Robin Holt7339ff82006-01-14 13:20:48 -08002173
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002174 /* Create pseudo-vma that contains just the policy */
2175 memset(&pvma, 0, sizeof(struct vm_area_struct));
2176 pvma.vm_end = TASK_SIZE; /* policy covers entire file */
2177 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002178
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002179put_new:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002180 mpol_put(new); /* drop initial ref */
Dan Carpenter0cae3452010-05-25 23:42:58 -07002181free_scratch:
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002182 NODEMASK_SCRATCH_FREE(scratch);
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002183put_mpol:
2184 mpol_put(mpol); /* drop our incoming ref on sb mpol */
Robin Holt7339ff82006-01-14 13:20:48 -08002185 }
2186}
2187
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188int mpol_set_shared_policy(struct shared_policy *info,
2189 struct vm_area_struct *vma, struct mempolicy *npol)
2190{
2191 int err;
2192 struct sp_node *new = NULL;
2193 unsigned long sz = vma_pages(vma);
2194
David Rientjes028fec42008-04-28 02:12:25 -07002195 pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 vma->vm_pgoff,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002197 sz, npol ? npol->mode : -1,
David Rientjes028fec42008-04-28 02:12:25 -07002198 npol ? npol->flags : -1,
Paul Mundt140d5a42007-07-15 23:38:16 -07002199 npol ? nodes_addr(npol->v.nodes)[0] : -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200
2201 if (npol) {
2202 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
2203 if (!new)
2204 return -ENOMEM;
2205 }
2206 err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
2207 if (err && new)
2208 kmem_cache_free(sn_cache, new);
2209 return err;
2210}
2211
2212/* Free a backing policy store on inode delete. */
2213void mpol_free_shared_policy(struct shared_policy *p)
2214{
2215 struct sp_node *n;
2216 struct rb_node *next;
2217
2218 if (!p->root.rb_node)
2219 return;
2220 spin_lock(&p->lock);
2221 next = rb_first(&p->root);
2222 while (next) {
2223 n = rb_entry(next, struct sp_node, nd);
2224 next = rb_next(&n->nd);
Andi Kleen90c50292005-07-27 11:43:50 -07002225 rb_erase(&n->nd, &p->root);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07002226 mpol_put(n->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 kmem_cache_free(sn_cache, n);
2228 }
2229 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230}
2231
2232/* assumes fs == KERNEL_DS */
2233void __init numa_policy_init(void)
2234{
Paul Mundtb71636e2007-07-15 23:38:15 -07002235 nodemask_t interleave_nodes;
2236 unsigned long largest = 0;
2237 int nid, prefer = 0;
2238
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 policy_cache = kmem_cache_create("numa_policy",
2240 sizeof(struct mempolicy),
Paul Mundt20c2df82007-07-20 10:11:58 +09002241 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242
2243 sn_cache = kmem_cache_create("shared_policy_node",
2244 sizeof(struct sp_node),
Paul Mundt20c2df82007-07-20 10:11:58 +09002245 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246
Paul Mundtb71636e2007-07-15 23:38:15 -07002247 /*
2248 * Set interleaving policy for system init. Interleaving is only
2249 * enabled across suitably sized nodes (default is >= 16MB), or
2250 * fall back to the largest node if they're all smaller.
2251 */
2252 nodes_clear(interleave_nodes);
Christoph Lameter56bbd652007-10-16 01:25:35 -07002253 for_each_node_state(nid, N_HIGH_MEMORY) {
Paul Mundtb71636e2007-07-15 23:38:15 -07002254 unsigned long total_pages = node_present_pages(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255
Paul Mundtb71636e2007-07-15 23:38:15 -07002256 /* Preserve the largest node */
2257 if (largest < total_pages) {
2258 largest = total_pages;
2259 prefer = nid;
2260 }
2261
2262 /* Interleave this node? */
2263 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
2264 node_set(nid, interleave_nodes);
2265 }
2266
2267 /* All too small, use the largest */
2268 if (unlikely(nodes_empty(interleave_nodes)))
2269 node_set(prefer, interleave_nodes);
2270
David Rientjes028fec42008-04-28 02:12:25 -07002271 if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 printk("numa_policy_init: interleaving failed\n");
2273}
2274
Christoph Lameter8bccd852005-10-29 18:16:59 -07002275/* Reset policy of current process to default */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276void numa_default_policy(void)
2277{
David Rientjes028fec42008-04-28 02:12:25 -07002278 do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279}
Paul Jackson68860ec2005-10-30 15:02:36 -08002280
Paul Jackson42253992006-01-08 01:01:59 -08002281/*
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002282 * Parse and format mempolicy from/to strings
2283 */
2284
2285/*
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002286 * "local" is pseudo-policy: MPOL_PREFERRED with MPOL_F_LOCAL flag
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002287 * Used only for mpol_parse_str() and mpol_to_str()
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002288 */
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002289#define MPOL_LOCAL MPOL_MAX
2290static const char * const policy_modes[] =
2291{
2292 [MPOL_DEFAULT] = "default",
2293 [MPOL_PREFERRED] = "prefer",
2294 [MPOL_BIND] = "bind",
2295 [MPOL_INTERLEAVE] = "interleave",
2296 [MPOL_LOCAL] = "local"
2297};
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002298
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002299
2300#ifdef CONFIG_TMPFS
2301/**
2302 * mpol_parse_str - parse string to mempolicy
2303 * @str: string containing mempolicy to parse
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002304 * @mpol: pointer to struct mempolicy pointer, returned on success.
2305 * @no_context: flag whether to "contextualize" the mempolicy
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002306 *
2307 * Format of input:
2308 * <mode>[=<flags>][:<nodelist>]
2309 *
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002310 * if @no_context is true, save the input nodemask in w.user_nodemask in
2311 * the returned mempolicy. This will be used to "clone" the mempolicy in
2312 * a specific context [cpuset] at a later time. Used to parse tmpfs mpol
2313 * mount option. Note that if 'static' or 'relative' mode flags were
2314 * specified, the input nodemask will already have been saved. Saving
2315 * it again is redundant, but safe.
2316 *
2317 * On success, returns 0, else 1
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002318 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002319int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002320{
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002321 struct mempolicy *new = NULL;
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002322 unsigned short mode;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002323 unsigned short uninitialized_var(mode_flags);
2324 nodemask_t nodes;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002325 char *nodelist = strchr(str, ':');
2326 char *flags = strchr(str, '=');
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002327 int err = 1;
2328
2329 if (nodelist) {
2330 /* NUL-terminate mode or flags string */
2331 *nodelist++ = '\0';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002332 if (nodelist_parse(nodelist, nodes))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002333 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002334 if (!nodes_subset(nodes, node_states[N_HIGH_MEMORY]))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002335 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002336 } else
2337 nodes_clear(nodes);
2338
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002339 if (flags)
2340 *flags++ = '\0'; /* terminate mode string */
2341
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002342 for (mode = 0; mode <= MPOL_LOCAL; mode++) {
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002343 if (!strcmp(str, policy_modes[mode])) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002344 break;
2345 }
2346 }
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002347 if (mode > MPOL_LOCAL)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002348 goto out;
2349
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002350 switch (mode) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002351 case MPOL_PREFERRED:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002352 /*
2353 * Insist on a nodelist of one node only
2354 */
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002355 if (nodelist) {
2356 char *rest = nodelist;
2357 while (isdigit(*rest))
2358 rest++;
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002359 if (*rest)
2360 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002361 }
2362 break;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002363 case MPOL_INTERLEAVE:
2364 /*
2365 * Default to online nodes with memory if no nodelist
2366 */
2367 if (!nodelist)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002368 nodes = node_states[N_HIGH_MEMORY];
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002369 break;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002370 case MPOL_LOCAL:
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002371 /*
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002372 * Don't allow a nodelist; mpol_new() checks flags
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002373 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002374 if (nodelist)
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002375 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002376 mode = MPOL_PREFERRED;
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002377 break;
Ravikiran G Thirumalai413b43d2010-03-23 13:35:28 -07002378 case MPOL_DEFAULT:
2379 /*
2380 * Insist on a empty nodelist
2381 */
2382 if (!nodelist)
2383 err = 0;
2384 goto out;
KOSAKI Motohirod69b2e62010-03-23 13:35:30 -07002385 case MPOL_BIND:
2386 /*
2387 * Insist on a nodelist
2388 */
2389 if (!nodelist)
2390 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002391 }
2392
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002393 mode_flags = 0;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002394 if (flags) {
2395 /*
2396 * Currently, we only support two mutually exclusive
2397 * mode flags.
2398 */
2399 if (!strcmp(flags, "static"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002400 mode_flags |= MPOL_F_STATIC_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002401 else if (!strcmp(flags, "relative"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002402 mode_flags |= MPOL_F_RELATIVE_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002403 else
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002404 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002405 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002406
2407 new = mpol_new(mode, mode_flags, &nodes);
2408 if (IS_ERR(new))
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002409 goto out;
2410
Lee Schermerhorne17f74a2010-05-24 14:32:02 -07002411 if (no_context) {
2412 /* save for contextualization */
2413 new->w.user_nodemask = nodes;
2414 } else {
Miao Xie58568d22009-06-16 15:31:49 -07002415 int ret;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002416 NODEMASK_SCRATCH(scratch);
2417 if (scratch) {
2418 task_lock(current);
2419 ret = mpol_set_nodemask(new, &nodes, scratch);
2420 task_unlock(current);
2421 } else
2422 ret = -ENOMEM;
2423 NODEMASK_SCRATCH_FREE(scratch);
2424 if (ret) {
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002425 mpol_put(new);
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002426 goto out;
Miao Xie58568d22009-06-16 15:31:49 -07002427 }
2428 }
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002429 err = 0;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002430
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002431out:
2432 /* Restore string for error message */
2433 if (nodelist)
2434 *--nodelist = ':';
2435 if (flags)
2436 *--flags = '=';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002437 if (!err)
2438 *mpol = new;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002439 return err;
2440}
2441#endif /* CONFIG_TMPFS */
2442
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002443/**
2444 * mpol_to_str - format a mempolicy structure for printing
2445 * @buffer: to contain formatted mempolicy string
2446 * @maxlen: length of @buffer
2447 * @pol: pointer to mempolicy to be formatted
2448 * @no_context: "context free" mempolicy - use nodemask in w.user_nodemask
2449 *
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002450 * Convert a mempolicy into a string.
2451 * Returns the number of characters in buffer (if positive)
2452 * or an error (negative)
2453 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002454int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol, int no_context)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002455{
2456 char *p = buffer;
2457 int l;
2458 nodemask_t nodes;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002459 unsigned short mode;
David Rientjesf5b087b2008-04-28 02:12:27 -07002460 unsigned short flags = pol ? pol->flags : 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002461
Lee Schermerhorn22919902008-04-28 02:13:22 -07002462 /*
2463 * Sanity check: room for longest mode, flag and some nodes
2464 */
2465 VM_BUG_ON(maxlen < strlen("interleave") + strlen("relative") + 16);
2466
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002467 if (!pol || pol == &default_policy)
2468 mode = MPOL_DEFAULT;
2469 else
2470 mode = pol->mode;
2471
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002472 switch (mode) {
2473 case MPOL_DEFAULT:
2474 nodes_clear(nodes);
2475 break;
2476
2477 case MPOL_PREFERRED:
2478 nodes_clear(nodes);
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002479 if (flags & MPOL_F_LOCAL)
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002480 mode = MPOL_LOCAL; /* pseudo-policy */
2481 else
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002482 node_set(pol->v.preferred_node, nodes);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002483 break;
2484
2485 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07002486 /* Fall through */
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002487 case MPOL_INTERLEAVE:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002488 if (no_context)
2489 nodes = pol->w.user_nodemask;
2490 else
2491 nodes = pol->v.nodes;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002492 break;
2493
2494 default:
2495 BUG();
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002496 }
2497
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002498 l = strlen(policy_modes[mode]);
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002499 if (buffer + maxlen < p + l + 1)
2500 return -ENOSPC;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002501
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002502 strcpy(p, policy_modes[mode]);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002503 p += l;
2504
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002505 if (flags & MPOL_MODE_FLAGS) {
David Rientjesf5b087b2008-04-28 02:12:27 -07002506 if (buffer + maxlen < p + 2)
2507 return -ENOSPC;
2508 *p++ = '=';
2509
Lee Schermerhorn22919902008-04-28 02:13:22 -07002510 /*
2511 * Currently, the only defined flags are mutually exclusive
2512 */
David Rientjesf5b087b2008-04-28 02:12:27 -07002513 if (flags & MPOL_F_STATIC_NODES)
Lee Schermerhorn22919902008-04-28 02:13:22 -07002514 p += snprintf(p, buffer + maxlen - p, "static");
2515 else if (flags & MPOL_F_RELATIVE_NODES)
2516 p += snprintf(p, buffer + maxlen - p, "relative");
David Rientjesf5b087b2008-04-28 02:12:27 -07002517 }
2518
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002519 if (!nodes_empty(nodes)) {
2520 if (buffer + maxlen < p + 2)
2521 return -ENOSPC;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002522 *p++ = ':';
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002523 p += nodelist_scnprintf(p, buffer + maxlen - p, nodes);
2524 }
2525 return p - buffer;
2526}
2527
2528struct numa_maps {
2529 unsigned long pages;
2530 unsigned long anon;
Christoph Lameter397874d2006-03-06 15:42:53 -08002531 unsigned long active;
2532 unsigned long writeback;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002533 unsigned long mapcount_max;
Christoph Lameter397874d2006-03-06 15:42:53 -08002534 unsigned long dirty;
2535 unsigned long swapcache;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002536 unsigned long node[MAX_NUMNODES];
2537};
2538
Christoph Lameter397874d2006-03-06 15:42:53 -08002539static void gather_stats(struct page *page, void *private, int pte_dirty)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002540{
2541 struct numa_maps *md = private;
2542 int count = page_mapcount(page);
2543
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002544 md->pages++;
Christoph Lameter397874d2006-03-06 15:42:53 -08002545 if (pte_dirty || PageDirty(page))
2546 md->dirty++;
2547
2548 if (PageSwapCache(page))
2549 md->swapcache++;
2550
Lee Schermerhorn894bc312008-10-18 20:26:39 -07002551 if (PageActive(page) || PageUnevictable(page))
Christoph Lameter397874d2006-03-06 15:42:53 -08002552 md->active++;
2553
2554 if (PageWriteback(page))
2555 md->writeback++;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002556
2557 if (PageAnon(page))
2558 md->anon++;
2559
Christoph Lameter397874d2006-03-06 15:42:53 -08002560 if (count > md->mapcount_max)
2561 md->mapcount_max = count;
2562
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002563 md->node[page_to_nid(page)]++;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002564}
2565
Andrew Morton7f709ed2006-03-07 21:55:22 -08002566#ifdef CONFIG_HUGETLB_PAGE
Christoph Lameter397874d2006-03-06 15:42:53 -08002567static void check_huge_range(struct vm_area_struct *vma,
2568 unsigned long start, unsigned long end,
2569 struct numa_maps *md)
2570{
2571 unsigned long addr;
2572 struct page *page;
Andi Kleena5516432008-07-23 21:27:41 -07002573 struct hstate *h = hstate_vma(vma);
2574 unsigned long sz = huge_page_size(h);
Christoph Lameter397874d2006-03-06 15:42:53 -08002575
Andi Kleena5516432008-07-23 21:27:41 -07002576 for (addr = start; addr < end; addr += sz) {
2577 pte_t *ptep = huge_pte_offset(vma->vm_mm,
2578 addr & huge_page_mask(h));
Christoph Lameter397874d2006-03-06 15:42:53 -08002579 pte_t pte;
2580
2581 if (!ptep)
2582 continue;
2583
2584 pte = *ptep;
2585 if (pte_none(pte))
2586 continue;
2587
2588 page = pte_page(pte);
2589 if (!page)
2590 continue;
2591
2592 gather_stats(page, md, pte_dirty(*ptep));
2593 }
2594}
Andrew Morton7f709ed2006-03-07 21:55:22 -08002595#else
2596static inline void check_huge_range(struct vm_area_struct *vma,
2597 unsigned long start, unsigned long end,
2598 struct numa_maps *md)
2599{
2600}
2601#endif
Christoph Lameter397874d2006-03-06 15:42:53 -08002602
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002603/*
2604 * Display pages allocated per node and memory policy via /proc.
2605 */
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002606int show_numa_map(struct seq_file *m, void *v)
2607{
Eric W. Biederman99f89552006-06-26 00:25:55 -07002608 struct proc_maps_private *priv = m->private;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002609 struct vm_area_struct *vma = v;
2610 struct numa_maps *md;
Christoph Lameter397874d2006-03-06 15:42:53 -08002611 struct file *file = vma->vm_file;
2612 struct mm_struct *mm = vma->vm_mm;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07002613 struct mempolicy *pol;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002614 int n;
2615 char buffer[50];
2616
Christoph Lameter397874d2006-03-06 15:42:53 -08002617 if (!mm)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002618 return 0;
2619
2620 md = kzalloc(sizeof(struct numa_maps), GFP_KERNEL);
2621 if (!md)
2622 return 0;
2623
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07002624 pol = get_vma_policy(priv->task, vma, vma->vm_start);
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002625 mpol_to_str(buffer, sizeof(buffer), pol, 0);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002626 mpol_cond_put(pol);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002627
Christoph Lameter397874d2006-03-06 15:42:53 -08002628 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002629
Christoph Lameter397874d2006-03-06 15:42:53 -08002630 if (file) {
2631 seq_printf(m, " file=");
Jan Blunckc32c2f62008-02-14 19:38:43 -08002632 seq_path(m, &file->f_path, "\n\t= ");
Christoph Lameter397874d2006-03-06 15:42:53 -08002633 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
2634 seq_printf(m, " heap");
2635 } else if (vma->vm_start <= mm->start_stack &&
2636 vma->vm_end >= mm->start_stack) {
2637 seq_printf(m, " stack");
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002638 }
Christoph Lameter397874d2006-03-06 15:42:53 -08002639
2640 if (is_vm_hugetlb_page(vma)) {
2641 check_huge_range(vma, vma->vm_start, vma->vm_end, md);
2642 seq_printf(m, " huge");
2643 } else {
2644 check_pgd_range(vma, vma->vm_start, vma->vm_end,
Christoph Lameter56bbd652007-10-16 01:25:35 -07002645 &node_states[N_HIGH_MEMORY], MPOL_MF_STATS, md);
Christoph Lameter397874d2006-03-06 15:42:53 -08002646 }
2647
2648 if (!md->pages)
2649 goto out;
2650
2651 if (md->anon)
2652 seq_printf(m," anon=%lu",md->anon);
2653
2654 if (md->dirty)
2655 seq_printf(m," dirty=%lu",md->dirty);
2656
2657 if (md->pages != md->anon && md->pages != md->dirty)
2658 seq_printf(m, " mapped=%lu", md->pages);
2659
2660 if (md->mapcount_max > 1)
2661 seq_printf(m, " mapmax=%lu", md->mapcount_max);
2662
2663 if (md->swapcache)
2664 seq_printf(m," swapcache=%lu", md->swapcache);
2665
2666 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2667 seq_printf(m," active=%lu", md->active);
2668
2669 if (md->writeback)
2670 seq_printf(m," writeback=%lu", md->writeback);
2671
Christoph Lameter56bbd652007-10-16 01:25:35 -07002672 for_each_node_state(n, N_HIGH_MEMORY)
Christoph Lameter397874d2006-03-06 15:42:53 -08002673 if (md->node[n])
2674 seq_printf(m, " N%d=%lu", n, md->node[n]);
2675out:
2676 seq_putc(m, '\n');
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002677 kfree(md);
2678
2679 if (m->count < m->size)
Eric W. Biederman99f89552006-06-26 00:25:55 -07002680 m->version = (vma != priv->tail_vma) ? vma->vm_start : 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002681 return 0;
2682}