blob: 1d771e4200d222eea46458bdd6089279a78ffeb5 [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>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040078#include <linux/export.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>
Michal Hocko778d3b02011-07-26 16:08:30 -070096#include <linux/random.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Nick Piggin62695a82008-10-18 20:26:09 -070098#include "internal.h"
99
Christoph Lameter38e35862006-01-08 01:01:01 -0800100/* Internal flags */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800101#define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */
Christoph Lameter38e35862006-01-08 01:01:01 -0800102#define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */
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 */
H Hartley Sweetene754d792011-10-31 17:09:23 -0700114static struct 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;
Wang Sheng-Hui89c522c2012-05-29 15:06:16 -0700393 if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE &&
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 Lameterfc301282006-01-18 17:42:29 -0800460static void migrate_page_add(struct page *page, struct list_head *pagelist,
461 unsigned long flags);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800462
Christoph Lameter38e35862006-01-08 01:01:01 -0800463/* Scan through pages checking if pages follow certain conditions. */
Nick Pigginb5810032005-10-29 18:16:12 -0700464static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800465 unsigned long addr, unsigned long end,
466 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800467 void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468{
Hugh Dickins91612e02005-06-21 17:15:07 -0700469 pte_t *orig_pte;
470 pte_t *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700471 spinlock_t *ptl;
Hugh Dickins941150a2005-06-21 17:15:06 -0700472
Hugh Dickins705e87c2005-10-29 18:16:27 -0700473 orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Hugh Dickins91612e02005-06-21 17:15:07 -0700474 do {
Linus Torvalds6aab3412005-11-28 14:34:23 -0800475 struct page *page;
Andy Whitcroft25ba77c2006-12-06 20:33:03 -0800476 int nid;
Hugh Dickins91612e02005-06-21 17:15:07 -0700477
478 if (!pte_present(*pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800480 page = vm_normal_page(vma, addr, *pte);
481 if (!page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 continue;
Nick Piggin053837f2006-01-18 17:42:27 -0800483 /*
Hugh Dickins62b61f62009-12-14 17:59:33 -0800484 * vm_normal_page() filters out zero pages, but there might
485 * still be PageReserved pages to skip, perhaps in a VDSO.
486 * And we cannot move PageKsm pages sensibly or safely yet.
Nick Piggin053837f2006-01-18 17:42:27 -0800487 */
Hugh Dickins62b61f62009-12-14 17:59:33 -0800488 if (PageReserved(page) || PageKsm(page))
Christoph Lameterf4598c82006-01-12 01:05:20 -0800489 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800490 nid = page_to_nid(page);
Christoph Lameter38e35862006-01-08 01:01:01 -0800491 if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT))
492 continue;
493
Stephen Wilsonb1f72d12011-05-24 17:12:43 -0700494 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Christoph Lameterfc301282006-01-18 17:42:29 -0800495 migrate_page_add(page, private, flags);
Christoph Lameter38e35862006-01-08 01:01:01 -0800496 else
497 break;
Hugh Dickins91612e02005-06-21 17:15:07 -0700498 } while (pte++, addr += PAGE_SIZE, addr != end);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700499 pte_unmap_unlock(orig_pte, ptl);
Hugh Dickins91612e02005-06-21 17:15:07 -0700500 return addr != end;
501}
502
Nick Pigginb5810032005-10-29 18:16:12 -0700503static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800504 unsigned long addr, unsigned long end,
505 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800506 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700507{
508 pmd_t *pmd;
509 unsigned long next;
510
511 pmd = pmd_offset(pud, addr);
512 do {
513 next = pmd_addr_end(addr, end);
Andrea Arcangelibae9c192011-01-13 15:46:46 -0800514 split_huge_page_pmd(vma->vm_mm, pmd);
Andrea Arcangeli1a5a9902012-03-21 16:33:42 -0700515 if (pmd_none_or_trans_huge_or_clear_bad(pmd))
Hugh Dickins91612e02005-06-21 17:15:07 -0700516 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800517 if (check_pte_range(vma, pmd, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800518 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700519 return -EIO;
520 } while (pmd++, addr = next, addr != end);
521 return 0;
522}
523
Nick Pigginb5810032005-10-29 18:16:12 -0700524static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800525 unsigned long addr, unsigned long end,
526 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800527 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700528{
529 pud_t *pud;
530 unsigned long next;
531
532 pud = pud_offset(pgd, addr);
533 do {
534 next = pud_addr_end(addr, end);
535 if (pud_none_or_clear_bad(pud))
536 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800537 if (check_pmd_range(vma, pud, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800538 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700539 return -EIO;
540 } while (pud++, addr = next, addr != end);
541 return 0;
542}
543
Nick Pigginb5810032005-10-29 18:16:12 -0700544static inline int check_pgd_range(struct vm_area_struct *vma,
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800545 unsigned long addr, unsigned long end,
546 const nodemask_t *nodes, unsigned long flags,
Christoph Lameter38e35862006-01-08 01:01:01 -0800547 void *private)
Hugh Dickins91612e02005-06-21 17:15:07 -0700548{
549 pgd_t *pgd;
550 unsigned long next;
551
Nick Pigginb5810032005-10-29 18:16:12 -0700552 pgd = pgd_offset(vma->vm_mm, addr);
Hugh Dickins91612e02005-06-21 17:15:07 -0700553 do {
554 next = pgd_addr_end(addr, end);
555 if (pgd_none_or_clear_bad(pgd))
556 continue;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800557 if (check_pud_range(vma, pgd, addr, next, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800558 flags, private))
Hugh Dickins91612e02005-06-21 17:15:07 -0700559 return -EIO;
560 } while (pgd++, addr = next, addr != end);
561 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562}
563
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800564/*
565 * Check if all pages in a range are on a set of nodes.
566 * If pagelist != NULL then isolate pages from the LRU and
567 * put them on the pagelist.
568 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569static struct vm_area_struct *
570check_range(struct mm_struct *mm, unsigned long start, unsigned long end,
Christoph Lameter38e35862006-01-08 01:01:01 -0800571 const nodemask_t *nodes, unsigned long flags, void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572{
573 int err;
574 struct vm_area_struct *first, *vma, *prev;
575
Nick Piggin053837f2006-01-18 17:42:27 -0800576
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 first = find_vma(mm, start);
578 if (!first)
579 return ERR_PTR(-EFAULT);
580 prev = NULL;
581 for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) {
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800582 if (!(flags & MPOL_MF_DISCONTIG_OK)) {
583 if (!vma->vm_next && vma->vm_end < end)
584 return ERR_PTR(-EFAULT);
585 if (prev && prev->vm_end < vma->vm_start)
586 return ERR_PTR(-EFAULT);
587 }
588 if (!is_vm_hugetlb_page(vma) &&
589 ((flags & MPOL_MF_STRICT) ||
590 ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) &&
591 vma_migratable(vma)))) {
Andi Kleen5b952b32005-09-13 01:25:08 -0700592 unsigned long endvma = vma->vm_end;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800593
Andi Kleen5b952b32005-09-13 01:25:08 -0700594 if (endvma > end)
595 endvma = end;
596 if (vma->vm_start > start)
597 start = vma->vm_start;
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800598 err = check_pgd_range(vma, start, endvma, nodes,
Christoph Lameter38e35862006-01-08 01:01:01 -0800599 flags, private);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 if (err) {
601 first = ERR_PTR(err);
602 break;
603 }
604 }
605 prev = vma;
606 }
607 return first;
608}
609
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610/* Step 2: apply policy to a range and do splits. */
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800611static int mbind_range(struct mm_struct *mm, unsigned long start,
612 unsigned long end, struct mempolicy *new_pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613{
614 struct vm_area_struct *next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800615 struct vm_area_struct *prev;
616 struct vm_area_struct *vma;
617 int err = 0;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800618 pgoff_t pgoff;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800619 unsigned long vmstart;
620 unsigned long vmend;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
Linus Torvalds097d5912012-03-06 18:23:36 -0800622 vma = find_vma(mm, start);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800623 if (!vma || vma->vm_start > start)
624 return -EFAULT;
625
Linus Torvalds097d5912012-03-06 18:23:36 -0800626 prev = vma->vm_prev;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800627 if (start > vma->vm_start)
628 prev = vma;
629
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800630 for (; vma && vma->vm_start < end; prev = vma, vma = next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 next = vma->vm_next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800632 vmstart = max(start, vma->vm_start);
633 vmend = min(end, vma->vm_end);
634
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800635 if (mpol_equal(vma_policy(vma), new_pol))
636 continue;
637
638 pgoff = vma->vm_pgoff +
639 ((vmstart - vma->vm_start) >> PAGE_SHIFT);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800640 prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800641 vma->anon_vma, vma->vm_file, pgoff,
Caspar Zhang8aacc9f2011-09-14 16:20:58 -0700642 new_pol);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800643 if (prev) {
644 vma = prev;
645 next = vma->vm_next;
646 continue;
647 }
648 if (vma->vm_start != vmstart) {
649 err = split_vma(vma->vm_mm, vma, vmstart, 1);
650 if (err)
651 goto out;
652 }
653 if (vma->vm_end != vmend) {
654 err = split_vma(vma->vm_mm, vma, vmend, 0);
655 if (err)
656 goto out;
657 }
Mel Gorman05f144a2012-05-23 12:48:13 +0100658
659 /*
660 * Apply policy to a single VMA. The reference counting of
661 * policy for vma_policy linkages has already been handled by
662 * vma_merge and split_vma as necessary. If this is a shared
663 * policy then ->set_policy will increment the reference count
664 * for an sp node.
665 */
666 pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
667 vma->vm_start, vma->vm_end, vma->vm_pgoff,
668 vma->vm_ops, vma->vm_file,
669 vma->vm_ops ? vma->vm_ops->set_policy : NULL);
670 if (vma->vm_ops && vma->vm_ops->set_policy) {
671 err = vma->vm_ops->set_policy(vma, new_pol);
672 if (err)
673 goto out;
674 }
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)) {
Mel Gorman7f0f2492011-01-13 15:45:58 -0800938 err = migrate_pages(&pagelist, new_node_page, dest,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800939 false, MIGRATE_SYNC);
Minchan Kimcf608ac2010-10-26 14:21:29 -0700940 if (err)
941 putback_lru_pages(&pagelist);
942 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700943
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800944 return err;
945}
946
947/*
948 * Move pages between the two nodesets so as to preserve the physical
949 * layout as much as possible.
Christoph Lameter39743882006-01-08 01:00:51 -0800950 *
951 * Returns the number of page that could not be moved.
952 */
Andrew Morton0ce72d42012-05-29 15:06:24 -0700953int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
954 const nodemask_t *to, int flags)
Christoph Lameter39743882006-01-08 01:00:51 -0800955{
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800956 int busy = 0;
Christoph Lameter0aedadf2008-11-06 12:53:30 -0800957 int err;
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800958 nodemask_t tmp;
Christoph Lameter39743882006-01-08 01:00:51 -0800959
Christoph Lameter0aedadf2008-11-06 12:53:30 -0800960 err = migrate_prep();
961 if (err)
962 return err;
963
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700964 down_read(&mm->mmap_sem);
Christoph Lameter39743882006-01-08 01:00:51 -0800965
Andrew Morton0ce72d42012-05-29 15:06:24 -0700966 err = migrate_vmas(mm, from, to, flags);
Christoph Lameter7b2259b2006-06-25 05:46:48 -0700967 if (err)
968 goto out;
969
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -0800970 /*
971 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
972 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
973 * bit in 'tmp', and return that <source, dest> pair for migration.
974 * The pair of nodemasks 'to' and 'from' define the map.
975 *
976 * If no pair of bits is found that way, fallback to picking some
977 * pair of 'source' and 'dest' bits that are not the same. If the
978 * 'source' and 'dest' bits are the same, this represents a node
979 * that will be migrating to itself, so no pages need move.
980 *
981 * If no bits are left in 'tmp', or if all remaining bits left
982 * in 'tmp' correspond to the same bit in 'to', return false
983 * (nothing left to migrate).
984 *
985 * This lets us pick a pair of nodes to migrate between, such that
986 * if possible the dest node is not already occupied by some other
987 * source node, minimizing the risk of overloading the memory on a
988 * node that would happen if we migrated incoming memory to a node
989 * before migrating outgoing memory source that same node.
990 *
991 * A single scan of tmp is sufficient. As we go, we remember the
992 * most recent <s, d> pair that moved (s != d). If we find a pair
993 * that not only moved, but what's better, moved to an empty slot
994 * (d is not set in tmp), then we break out then, with that pair.
Justin P. Mattockae0e47f2011-03-01 15:06:02 +0100995 * Otherwise when we finish scanning from_tmp, we at least have the
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -0800996 * most recent <s, d> pair that moved. If we get all the way through
997 * the scan of tmp without finding any node that moved, much less
998 * moved to an empty node, then there is nothing left worth migrating.
999 */
Christoph Lameterd4984712006-01-08 01:00:55 -08001000
Andrew Morton0ce72d42012-05-29 15:06:24 -07001001 tmp = *from;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001002 while (!nodes_empty(tmp)) {
1003 int s,d;
1004 int source = -1;
1005 int dest = 0;
1006
1007 for_each_node_mask(s, tmp) {
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001008
1009 /*
1010 * do_migrate_pages() tries to maintain the relative
1011 * node relationship of the pages established between
1012 * threads and memory areas.
1013 *
1014 * However if the number of source nodes is not equal to
1015 * the number of destination nodes we can not preserve
1016 * this node relative relationship. In that case, skip
1017 * copying memory from a node that is in the destination
1018 * mask.
1019 *
1020 * Example: [2,3,4] -> [3,4,5] moves everything.
1021 * [0-7] - > [3,4,5] moves only 0,1,2,6,7.
1022 */
1023
Andrew Morton0ce72d42012-05-29 15:06:24 -07001024 if ((nodes_weight(*from) != nodes_weight(*to)) &&
1025 (node_isset(s, *to)))
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001026 continue;
1027
Andrew Morton0ce72d42012-05-29 15:06:24 -07001028 d = node_remap(s, *from, *to);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001029 if (s == d)
1030 continue;
1031
1032 source = s; /* Node moved. Memorize */
1033 dest = d;
1034
1035 /* dest not in remaining from nodes? */
1036 if (!node_isset(dest, tmp))
1037 break;
1038 }
1039 if (source == -1)
1040 break;
1041
1042 node_clear(source, tmp);
1043 err = migrate_to_node(mm, source, dest, flags);
1044 if (err > 0)
1045 busy += err;
1046 if (err < 0)
1047 break;
Christoph Lameter39743882006-01-08 01:00:51 -08001048 }
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001049out:
Christoph Lameter39743882006-01-08 01:00:51 -08001050 up_read(&mm->mmap_sem);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001051 if (err < 0)
1052 return err;
1053 return busy;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001054
Christoph Lameter39743882006-01-08 01:00:51 -08001055}
1056
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001057/*
1058 * Allocate a new page for page migration based on vma policy.
1059 * Start assuming that page is mapped by vma pointed to by @private.
1060 * Search forward from there, if not. N.B., this assumes that the
1061 * list of pages handed to migrate_pages()--which is how we get here--
1062 * is in virtual address order.
1063 */
Christoph Lameter742755a2006-06-23 02:03:55 -07001064static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001065{
1066 struct vm_area_struct *vma = (struct vm_area_struct *)private;
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001067 unsigned long uninitialized_var(address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001068
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001069 while (vma) {
1070 address = page_address_in_vma(page, vma);
1071 if (address != -EFAULT)
1072 break;
1073 vma = vma->vm_next;
1074 }
1075
1076 /*
1077 * if !vma, alloc_page_vma() will use task or system default policy
1078 */
1079 return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001080}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001081#else
1082
1083static void migrate_page_add(struct page *page, struct list_head *pagelist,
1084 unsigned long flags)
1085{
1086}
1087
Andrew Morton0ce72d42012-05-29 15:06:24 -07001088int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1089 const nodemask_t *to, int flags)
Christoph Lameterb20a3502006-03-22 00:09:12 -08001090{
1091 return -ENOSYS;
1092}
Christoph Lameter95a402c2006-06-23 02:03:53 -07001093
Keith Owens69939742006-10-11 01:21:28 -07001094static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001095{
1096 return NULL;
1097}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001098#endif
1099
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001100static long do_mbind(unsigned long start, unsigned long len,
David Rientjes028fec42008-04-28 02:12:25 -07001101 unsigned short mode, unsigned short mode_flags,
1102 nodemask_t *nmask, unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001103{
1104 struct vm_area_struct *vma;
1105 struct mm_struct *mm = current->mm;
1106 struct mempolicy *new;
1107 unsigned long end;
1108 int err;
1109 LIST_HEAD(pagelist);
1110
David Rientjesa3b51e02008-04-28 02:12:23 -07001111 if (flags & ~(unsigned long)(MPOL_MF_STRICT |
1112 MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001113 return -EINVAL;
Christoph Lameter74c00242006-03-14 19:50:21 -08001114 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001115 return -EPERM;
1116
1117 if (start & ~PAGE_MASK)
1118 return -EINVAL;
1119
1120 if (mode == MPOL_DEFAULT)
1121 flags &= ~MPOL_MF_STRICT;
1122
1123 len = (len + PAGE_SIZE - 1) & PAGE_MASK;
1124 end = start + len;
1125
1126 if (end < start)
1127 return -EINVAL;
1128 if (end == start)
1129 return 0;
1130
David Rientjes028fec42008-04-28 02:12:25 -07001131 new = mpol_new(mode, mode_flags, nmask);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001132 if (IS_ERR(new))
1133 return PTR_ERR(new);
1134
1135 /*
1136 * If we are using the default policy then operation
1137 * on discontinuous address spaces is okay after all
1138 */
1139 if (!new)
1140 flags |= MPOL_MF_DISCONTIG_OK;
1141
David Rientjes028fec42008-04-28 02:12:25 -07001142 pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
1143 start, start + len, mode, mode_flags,
1144 nmask ? nodes_addr(*nmask)[0] : -1);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001145
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001146 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
1147
1148 err = migrate_prep();
1149 if (err)
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001150 goto mpol_out;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001151 }
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07001152 {
1153 NODEMASK_SCRATCH(scratch);
1154 if (scratch) {
1155 down_write(&mm->mmap_sem);
1156 task_lock(current);
1157 err = mpol_set_nodemask(new, nmask, scratch);
1158 task_unlock(current);
1159 if (err)
1160 up_write(&mm->mmap_sem);
1161 } else
1162 err = -ENOMEM;
1163 NODEMASK_SCRATCH_FREE(scratch);
1164 }
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001165 if (err)
1166 goto mpol_out;
1167
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001168 vma = check_range(mm, start, end, nmask,
1169 flags | MPOL_MF_INVERT, &pagelist);
1170
1171 err = PTR_ERR(vma);
1172 if (!IS_ERR(vma)) {
1173 int nr_failed = 0;
1174
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -08001175 err = mbind_range(mm, start, end, new);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001176
Minchan Kimcf608ac2010-10-26 14:21:29 -07001177 if (!list_empty(&pagelist)) {
Christoph Lameter95a402c2006-06-23 02:03:53 -07001178 nr_failed = migrate_pages(&pagelist, new_vma_page,
Mel Gorman7f0f2492011-01-13 15:45:58 -08001179 (unsigned long)vma,
David Rientjesc4c0e9e2012-06-20 18:00:12 -07001180 false, MIGRATE_SYNC);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001181 if (nr_failed)
1182 putback_lru_pages(&pagelist);
1183 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001184
1185 if (!err && nr_failed && (flags & MPOL_MF_STRICT))
1186 err = -EIO;
KOSAKI Motohiroab8a3e12009-10-26 16:49:58 -07001187 } else
1188 putback_lru_pages(&pagelist);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001189
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001190 up_write(&mm->mmap_sem);
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001191 mpol_out:
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001192 mpol_put(new);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001193 return err;
1194}
1195
Christoph Lameter39743882006-01-08 01:00:51 -08001196/*
Christoph Lameter8bccd852005-10-29 18:16:59 -07001197 * User space interface with variable sized bitmaps for nodelists.
1198 */
1199
1200/* Copy a node mask from user space. */
Christoph Lameter39743882006-01-08 01:00:51 -08001201static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
Christoph Lameter8bccd852005-10-29 18:16:59 -07001202 unsigned long maxnode)
1203{
1204 unsigned long k;
1205 unsigned long nlongs;
1206 unsigned long endmask;
1207
1208 --maxnode;
1209 nodes_clear(*nodes);
1210 if (maxnode == 0 || !nmask)
1211 return 0;
Andi Kleena9c930b2006-02-20 18:27:59 -08001212 if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
Chris Wright636f13c2006-02-17 13:59:36 -08001213 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001214
1215 nlongs = BITS_TO_LONGS(maxnode);
1216 if ((maxnode % BITS_PER_LONG) == 0)
1217 endmask = ~0UL;
1218 else
1219 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
1220
1221 /* When the user specified more nodes than supported just check
1222 if the non supported part is all zero. */
1223 if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
1224 if (nlongs > PAGE_SIZE/sizeof(long))
1225 return -EINVAL;
1226 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
1227 unsigned long t;
1228 if (get_user(t, nmask + k))
1229 return -EFAULT;
1230 if (k == nlongs - 1) {
1231 if (t & endmask)
1232 return -EINVAL;
1233 } else if (t)
1234 return -EINVAL;
1235 }
1236 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
1237 endmask = ~0UL;
1238 }
1239
1240 if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
1241 return -EFAULT;
1242 nodes_addr(*nodes)[nlongs-1] &= endmask;
1243 return 0;
1244}
1245
1246/* Copy a kernel node mask to user space */
1247static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1248 nodemask_t *nodes)
1249{
1250 unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1251 const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long);
1252
1253 if (copy > nbytes) {
1254 if (copy > PAGE_SIZE)
1255 return -EINVAL;
1256 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1257 return -EFAULT;
1258 copy = nbytes;
1259 }
1260 return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1261}
1262
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001263SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
1264 unsigned long, mode, unsigned long __user *, nmask,
1265 unsigned long, maxnode, unsigned, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001266{
1267 nodemask_t nodes;
1268 int err;
David Rientjes028fec42008-04-28 02:12:25 -07001269 unsigned short mode_flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001270
David Rientjes028fec42008-04-28 02:12:25 -07001271 mode_flags = mode & MPOL_MODE_FLAGS;
1272 mode &= ~MPOL_MODE_FLAGS;
David Rientjesa3b51e02008-04-28 02:12:23 -07001273 if (mode >= MPOL_MAX)
1274 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001275 if ((mode_flags & MPOL_F_STATIC_NODES) &&
1276 (mode_flags & MPOL_F_RELATIVE_NODES))
1277 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001278 err = get_nodes(&nodes, nmask, maxnode);
1279 if (err)
1280 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001281 return do_mbind(start, len, mode, mode_flags, &nodes, flags);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001282}
1283
1284/* Set the process memory policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001285SYSCALL_DEFINE3(set_mempolicy, int, mode, unsigned long __user *, nmask,
1286 unsigned long, maxnode)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001287{
1288 int err;
1289 nodemask_t nodes;
David Rientjes028fec42008-04-28 02:12:25 -07001290 unsigned short flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001291
David Rientjes028fec42008-04-28 02:12:25 -07001292 flags = mode & MPOL_MODE_FLAGS;
1293 mode &= ~MPOL_MODE_FLAGS;
1294 if ((unsigned int)mode >= MPOL_MAX)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001295 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001296 if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
1297 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001298 err = get_nodes(&nodes, nmask, maxnode);
1299 if (err)
1300 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001301 return do_set_mempolicy(mode, flags, &nodes);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001302}
1303
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001304SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
1305 const unsigned long __user *, old_nodes,
1306 const unsigned long __user *, new_nodes)
Christoph Lameter39743882006-01-08 01:00:51 -08001307{
David Howellsc69e8d92008-11-14 10:39:19 +11001308 const struct cred *cred = current_cred(), *tcred;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001309 struct mm_struct *mm = NULL;
Christoph Lameter39743882006-01-08 01:00:51 -08001310 struct task_struct *task;
Christoph Lameter39743882006-01-08 01:00:51 -08001311 nodemask_t task_nodes;
1312 int err;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001313 nodemask_t *old;
1314 nodemask_t *new;
1315 NODEMASK_SCRATCH(scratch);
Christoph Lameter39743882006-01-08 01:00:51 -08001316
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001317 if (!scratch)
1318 return -ENOMEM;
Christoph Lameter39743882006-01-08 01:00:51 -08001319
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001320 old = &scratch->mask1;
1321 new = &scratch->mask2;
1322
1323 err = get_nodes(old, old_nodes, maxnode);
Christoph Lameter39743882006-01-08 01:00:51 -08001324 if (err)
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001325 goto out;
1326
1327 err = get_nodes(new, new_nodes, maxnode);
1328 if (err)
1329 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001330
1331 /* Find the mm_struct */
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001332 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001333 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter39743882006-01-08 01:00:51 -08001334 if (!task) {
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001335 rcu_read_unlock();
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001336 err = -ESRCH;
1337 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001338 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001339 get_task_struct(task);
Christoph Lameter39743882006-01-08 01:00:51 -08001340
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001341 err = -EINVAL;
Christoph Lameter39743882006-01-08 01:00:51 -08001342
1343 /*
1344 * Check if this process has the right to modify the specified
1345 * process. The right exists if the process has administrative
Alexey Dobriyan7f927fc2006-03-28 01:56:53 -08001346 * capabilities, superuser privileges or the same
Christoph Lameter39743882006-01-08 01:00:51 -08001347 * userid as the target process.
1348 */
David Howellsc69e8d92008-11-14 10:39:19 +11001349 tcred = __task_cred(task);
Eric W. Biedermanb38a86e2012-03-12 15:48:24 -07001350 if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1351 !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) &&
Christoph Lameter74c00242006-03-14 19:50:21 -08001352 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001353 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001354 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001355 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001356 }
David Howellsc69e8d92008-11-14 10:39:19 +11001357 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001358
1359 task_nodes = cpuset_mems_allowed(task);
1360 /* Is the user allowed to access the target nodes? */
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001361 if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
Christoph Lameter39743882006-01-08 01:00:51 -08001362 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001363 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001364 }
1365
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001366 if (!nodes_subset(*new, node_states[N_HIGH_MEMORY])) {
Christoph Lameter3b42d282007-08-31 00:12:08 -07001367 err = -EINVAL;
Christoph Lameter3268c632012-03-21 16:34:06 -07001368 goto out_put;
Christoph Lameter3b42d282007-08-31 00:12:08 -07001369 }
1370
David Quigley86c3a762006-06-23 02:04:02 -07001371 err = security_task_movememory(task);
1372 if (err)
Christoph Lameter3268c632012-03-21 16:34:06 -07001373 goto out_put;
David Quigley86c3a762006-06-23 02:04:02 -07001374
Christoph Lameter3268c632012-03-21 16:34:06 -07001375 mm = get_task_mm(task);
1376 put_task_struct(task);
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001377
1378 if (!mm) {
Christoph Lameter3268c632012-03-21 16:34:06 -07001379 err = -EINVAL;
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001380 goto out;
1381 }
1382
1383 err = do_migrate_pages(mm, old, new,
1384 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
Christoph Lameter3268c632012-03-21 16:34:06 -07001385
1386 mmput(mm);
1387out:
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001388 NODEMASK_SCRATCH_FREE(scratch);
1389
Christoph Lameter39743882006-01-08 01:00:51 -08001390 return err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001391
1392out_put:
1393 put_task_struct(task);
1394 goto out;
1395
Christoph Lameter39743882006-01-08 01:00:51 -08001396}
1397
1398
Christoph Lameter8bccd852005-10-29 18:16:59 -07001399/* Retrieve NUMA policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001400SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1401 unsigned long __user *, nmask, unsigned long, maxnode,
1402 unsigned long, addr, unsigned long, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001403{
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001404 int err;
1405 int uninitialized_var(pval);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001406 nodemask_t nodes;
1407
1408 if (nmask != NULL && maxnode < MAX_NUMNODES)
1409 return -EINVAL;
1410
1411 err = do_get_mempolicy(&pval, &nodes, addr, flags);
1412
1413 if (err)
1414 return err;
1415
1416 if (policy && put_user(pval, policy))
1417 return -EFAULT;
1418
1419 if (nmask)
1420 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1421
1422 return err;
1423}
1424
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425#ifdef CONFIG_COMPAT
1426
1427asmlinkage long compat_sys_get_mempolicy(int __user *policy,
1428 compat_ulong_t __user *nmask,
1429 compat_ulong_t maxnode,
1430 compat_ulong_t addr, compat_ulong_t flags)
1431{
1432 long err;
1433 unsigned long __user *nm = NULL;
1434 unsigned long nr_bits, alloc_size;
1435 DECLARE_BITMAP(bm, MAX_NUMNODES);
1436
1437 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1438 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1439
1440 if (nmask)
1441 nm = compat_alloc_user_space(alloc_size);
1442
1443 err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
1444
1445 if (!err && nmask) {
KAMEZAWA Hiroyuki2bbff6c2011-09-14 16:21:02 -07001446 unsigned long copy_size;
1447 copy_size = min_t(unsigned long, sizeof(bm), alloc_size);
1448 err = copy_from_user(bm, nm, copy_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 /* ensure entire bitmap is zeroed */
1450 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
1451 err |= compat_put_bitmap(nmask, bm, nr_bits);
1452 }
1453
1454 return err;
1455}
1456
1457asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask,
1458 compat_ulong_t maxnode)
1459{
1460 long err = 0;
1461 unsigned long __user *nm = NULL;
1462 unsigned long nr_bits, alloc_size;
1463 DECLARE_BITMAP(bm, MAX_NUMNODES);
1464
1465 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1466 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1467
1468 if (nmask) {
1469 err = compat_get_bitmap(bm, nmask, nr_bits);
1470 nm = compat_alloc_user_space(alloc_size);
1471 err |= copy_to_user(nm, bm, alloc_size);
1472 }
1473
1474 if (err)
1475 return -EFAULT;
1476
1477 return sys_set_mempolicy(mode, nm, nr_bits+1);
1478}
1479
1480asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len,
1481 compat_ulong_t mode, compat_ulong_t __user *nmask,
1482 compat_ulong_t maxnode, compat_ulong_t flags)
1483{
1484 long err = 0;
1485 unsigned long __user *nm = NULL;
1486 unsigned long nr_bits, alloc_size;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001487 nodemask_t bm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
1489 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1490 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1491
1492 if (nmask) {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001493 err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 nm = compat_alloc_user_space(alloc_size);
Andi Kleendfcd3c02005-10-29 18:15:48 -07001495 err |= copy_to_user(nm, nodes_addr(bm), alloc_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 }
1497
1498 if (err)
1499 return -EFAULT;
1500
1501 return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
1502}
1503
1504#endif
1505
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001506/*
1507 * get_vma_policy(@task, @vma, @addr)
1508 * @task - task for fallback if vma policy == default
1509 * @vma - virtual memory area whose policy is sought
1510 * @addr - address in @vma for shared policy lookup
1511 *
1512 * Returns effective policy for a VMA at specified address.
1513 * Falls back to @task or system default policy, as necessary.
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001514 * Current or other task's task mempolicy and non-shared vma policies
1515 * are protected by the task's mmap_sem, which must be held for read by
1516 * the caller.
1517 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1518 * count--added by the get_policy() vm_op, as appropriate--to protect against
1519 * freeing by another task. It is the caller's responsibility to free the
1520 * extra reference for shared policies.
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001521 */
Stephen Wilsond98f6cb2011-05-24 17:12:41 -07001522struct mempolicy *get_vma_policy(struct task_struct *task,
Christoph Lameter48fce342006-01-08 01:01:03 -08001523 struct vm_area_struct *vma, unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07001525 struct mempolicy *pol = task->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526
1527 if (vma) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001528 if (vma->vm_ops && vma->vm_ops->get_policy) {
Lee Schermerhornae4d8c12008-04-28 02:13:11 -07001529 struct mempolicy *vpol = vma->vm_ops->get_policy(vma,
1530 addr);
1531 if (vpol)
1532 pol = vpol;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001533 } else if (vma->vm_policy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 pol = vma->vm_policy;
1535 }
1536 if (!pol)
1537 pol = &default_policy;
1538 return pol;
1539}
1540
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001541/*
1542 * Return a nodemask representing a mempolicy for filtering nodes for
1543 * page allocation
1544 */
1545static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
Mel Gorman19770b32008-04-28 02:12:18 -07001546{
1547 /* Lower zones don't get a nodemask applied for MPOL_BIND */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001548 if (unlikely(policy->mode == MPOL_BIND) &&
Mel Gorman19770b32008-04-28 02:12:18 -07001549 gfp_zone(gfp) >= policy_zone &&
1550 cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
1551 return &policy->v.nodes;
1552
1553 return NULL;
1554}
1555
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001556/* Return a zonelist indicated by gfp for node representing a mempolicy */
Andi Kleen2f5f9482011-03-04 17:36:29 -08001557static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy,
1558 int nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559{
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001560 switch (policy->mode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001562 if (!(policy->flags & MPOL_F_LOCAL))
1563 nd = policy->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 break;
1565 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07001566 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001567 * Normally, MPOL_BIND allocations are node-local within the
1568 * allowed nodemask. However, if __GFP_THISNODE is set and the
Bob Liu6eb27e12010-05-24 14:32:00 -07001569 * current node isn't part of the mask, we use the zonelist for
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001570 * the first node in the mask instead.
Mel Gorman19770b32008-04-28 02:12:18 -07001571 */
Mel Gorman19770b32008-04-28 02:12:18 -07001572 if (unlikely(gfp & __GFP_THISNODE) &&
1573 unlikely(!node_isset(nd, policy->v.nodes)))
1574 nd = first_node(policy->v.nodes);
1575 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 BUG();
1578 }
Mel Gorman0e884602008-04-28 02:12:14 -07001579 return node_zonelist(nd, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580}
1581
1582/* Do dynamic interleaving for a process */
1583static unsigned interleave_nodes(struct mempolicy *policy)
1584{
1585 unsigned nid, next;
1586 struct task_struct *me = current;
1587
1588 nid = me->il_next;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001589 next = next_node(nid, policy->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 if (next >= MAX_NUMNODES)
Andi Kleendfcd3c02005-10-29 18:15:48 -07001591 next = first_node(policy->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001592 if (next < MAX_NUMNODES)
1593 me->il_next = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 return nid;
1595}
1596
Christoph Lameterdc85da12006-01-18 17:42:36 -08001597/*
1598 * Depending on the memory policy provide a node from which to allocate the
1599 * next slab entry.
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001600 * @policy must be protected by freeing by the caller. If @policy is
1601 * the current task's mempolicy, this protection is implicit, as only the
1602 * task can change it's policy. The system default policy requires no
1603 * such protection.
Christoph Lameterdc85da12006-01-18 17:42:36 -08001604 */
1605unsigned slab_node(struct mempolicy *policy)
1606{
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001607 if (!policy || policy->flags & MPOL_F_LOCAL)
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001608 return numa_node_id();
Christoph Lameter765c4502006-09-27 01:50:08 -07001609
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001610 switch (policy->mode) {
1611 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001612 /*
1613 * handled MPOL_F_LOCAL above
1614 */
1615 return policy->v.preferred_node;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001616
Christoph Lameterdc85da12006-01-18 17:42:36 -08001617 case MPOL_INTERLEAVE:
1618 return interleave_nodes(policy);
1619
Mel Gormandd1a2392008-04-28 02:12:17 -07001620 case MPOL_BIND: {
Christoph Lameterdc85da12006-01-18 17:42:36 -08001621 /*
1622 * Follow bind policy behavior and start allocation at the
1623 * first node.
1624 */
Mel Gorman19770b32008-04-28 02:12:18 -07001625 struct zonelist *zonelist;
1626 struct zone *zone;
1627 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
1628 zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0];
1629 (void)first_zones_zonelist(zonelist, highest_zoneidx,
1630 &policy->v.nodes,
1631 &zone);
Eric Dumazet800416f2010-10-27 19:33:43 +02001632 return zone ? zone->node : numa_node_id();
Mel Gormandd1a2392008-04-28 02:12:17 -07001633 }
Christoph Lameterdc85da12006-01-18 17:42:36 -08001634
Christoph Lameterdc85da12006-01-18 17:42:36 -08001635 default:
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001636 BUG();
Christoph Lameterdc85da12006-01-18 17:42:36 -08001637 }
1638}
1639
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640/* Do static interleaving for a VMA with known offset. */
1641static unsigned offset_il_node(struct mempolicy *pol,
1642 struct vm_area_struct *vma, unsigned long off)
1643{
Andi Kleendfcd3c02005-10-29 18:15:48 -07001644 unsigned nnodes = nodes_weight(pol->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001645 unsigned target;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 int c;
1647 int nid = -1;
1648
David Rientjesf5b087b2008-04-28 02:12:27 -07001649 if (!nnodes)
1650 return numa_node_id();
1651 target = (unsigned int)off % nnodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 c = 0;
1653 do {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001654 nid = next_node(nid, pol->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 c++;
1656 } while (c <= target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 return nid;
1658}
1659
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001660/* Determine a node number for interleave */
1661static inline unsigned interleave_nid(struct mempolicy *pol,
1662 struct vm_area_struct *vma, unsigned long addr, int shift)
1663{
1664 if (vma) {
1665 unsigned long off;
1666
Nishanth Aravamudan3b98b082006-08-31 21:27:53 -07001667 /*
1668 * for small pages, there is no difference between
1669 * shift and PAGE_SHIFT, so the bit-shift is safe.
1670 * for huge pages, since vm_pgoff is in units of small
1671 * pages, we need to shift off the always 0 bits to get
1672 * a useful offset.
1673 */
1674 BUG_ON(shift < PAGE_SHIFT);
1675 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001676 off += (addr - vma->vm_start) >> shift;
1677 return offset_il_node(pol, vma, off);
1678 } else
1679 return interleave_nodes(pol);
1680}
1681
Michal Hocko778d3b02011-07-26 16:08:30 -07001682/*
1683 * Return the bit number of a random bit set in the nodemask.
1684 * (returns -1 if nodemask is empty)
1685 */
1686int node_random(const nodemask_t *maskp)
1687{
1688 int w, bit = -1;
1689
1690 w = nodes_weight(*maskp);
1691 if (w)
1692 bit = bitmap_ord_to_pos(maskp->bits,
1693 get_random_int() % w, MAX_NUMNODES);
1694 return bit;
1695}
1696
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001697#ifdef CONFIG_HUGETLBFS
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001698/*
1699 * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
1700 * @vma = virtual memory area whose policy is sought
1701 * @addr = address in @vma for shared policy lookup and interleave policy
1702 * @gfp_flags = for requested zone
Mel Gorman19770b32008-04-28 02:12:18 -07001703 * @mpol = pointer to mempolicy pointer for reference counted mempolicy
1704 * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001705 *
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001706 * Returns a zonelist suitable for a huge page allocation and a pointer
1707 * to the struct mempolicy for conditional unref after allocation.
1708 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1709 * @nodemask for filtering the zonelist.
Miao Xiec0ff7452010-05-24 14:32:08 -07001710 *
1711 * Must be protected by get_mems_allowed()
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001712 */
Mel Gorman396faf02007-07-17 04:03:13 -07001713struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
Mel Gorman19770b32008-04-28 02:12:18 -07001714 gfp_t gfp_flags, struct mempolicy **mpol,
1715 nodemask_t **nodemask)
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001716{
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001717 struct zonelist *zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001718
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001719 *mpol = get_vma_policy(current, vma, addr);
Mel Gorman19770b32008-04-28 02:12:18 -07001720 *nodemask = NULL; /* assume !MPOL_BIND */
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001721
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001722 if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
1723 zl = node_zonelist(interleave_nid(*mpol, vma, addr,
Andi Kleena5516432008-07-23 21:27:41 -07001724 huge_page_shift(hstate_vma(vma))), gfp_flags);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001725 } else {
Andi Kleen2f5f9482011-03-04 17:36:29 -08001726 zl = policy_zonelist(gfp_flags, *mpol, numa_node_id());
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001727 if ((*mpol)->mode == MPOL_BIND)
1728 *nodemask = &(*mpol)->v.nodes;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001729 }
1730 return zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001731}
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001732
1733/*
1734 * init_nodemask_of_mempolicy
1735 *
1736 * If the current task's mempolicy is "default" [NULL], return 'false'
1737 * to indicate default policy. Otherwise, extract the policy nodemask
1738 * for 'bind' or 'interleave' policy into the argument nodemask, or
1739 * initialize the argument nodemask to contain the single node for
1740 * 'preferred' or 'local' policy and return 'true' to indicate presence
1741 * of non-default mempolicy.
1742 *
1743 * We don't bother with reference counting the mempolicy [mpol_get/put]
1744 * because the current task is examining it's own mempolicy and a task's
1745 * mempolicy is only ever changed by the task itself.
1746 *
1747 * N.B., it is the caller's responsibility to free a returned nodemask.
1748 */
1749bool init_nodemask_of_mempolicy(nodemask_t *mask)
1750{
1751 struct mempolicy *mempolicy;
1752 int nid;
1753
1754 if (!(mask && current->mempolicy))
1755 return false;
1756
Miao Xiec0ff7452010-05-24 14:32:08 -07001757 task_lock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001758 mempolicy = current->mempolicy;
1759 switch (mempolicy->mode) {
1760 case MPOL_PREFERRED:
1761 if (mempolicy->flags & MPOL_F_LOCAL)
1762 nid = numa_node_id();
1763 else
1764 nid = mempolicy->v.preferred_node;
1765 init_nodemask_of_node(mask, nid);
1766 break;
1767
1768 case MPOL_BIND:
1769 /* Fall through */
1770 case MPOL_INTERLEAVE:
1771 *mask = mempolicy->v.nodes;
1772 break;
1773
1774 default:
1775 BUG();
1776 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001777 task_unlock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001778
1779 return true;
1780}
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001781#endif
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001782
David Rientjes6f48d0eb2010-08-09 17:18:52 -07001783/*
1784 * mempolicy_nodemask_intersects
1785 *
1786 * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default
1787 * policy. Otherwise, check for intersection between mask and the policy
1788 * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local'
1789 * policy, always return true since it may allocate elsewhere on fallback.
1790 *
1791 * Takes task_lock(tsk) to prevent freeing of its mempolicy.
1792 */
1793bool mempolicy_nodemask_intersects(struct task_struct *tsk,
1794 const nodemask_t *mask)
1795{
1796 struct mempolicy *mempolicy;
1797 bool ret = true;
1798
1799 if (!mask)
1800 return ret;
1801 task_lock(tsk);
1802 mempolicy = tsk->mempolicy;
1803 if (!mempolicy)
1804 goto out;
1805
1806 switch (mempolicy->mode) {
1807 case MPOL_PREFERRED:
1808 /*
1809 * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to
1810 * allocate from, they may fallback to other nodes when oom.
1811 * Thus, it's possible for tsk to have allocated memory from
1812 * nodes in mask.
1813 */
1814 break;
1815 case MPOL_BIND:
1816 case MPOL_INTERLEAVE:
1817 ret = nodes_intersects(mempolicy->v.nodes, *mask);
1818 break;
1819 default:
1820 BUG();
1821 }
1822out:
1823 task_unlock(tsk);
1824 return ret;
1825}
1826
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827/* Allocate a page in interleaved policy.
1828 Own path because it needs to do special accounting. */
Andi Kleen662f3a02005-10-29 18:15:49 -07001829static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
1830 unsigned nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831{
1832 struct zonelist *zl;
1833 struct page *page;
1834
Mel Gorman0e884602008-04-28 02:12:14 -07001835 zl = node_zonelist(nid, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 page = __alloc_pages(gfp, order, zl);
Mel Gormandd1a2392008-04-28 02:12:17 -07001837 if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
Christoph Lameterca889e62006-06-30 01:55:44 -07001838 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 return page;
1840}
1841
1842/**
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001843 * alloc_pages_vma - Allocate a page for a VMA.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 *
1845 * @gfp:
1846 * %GFP_USER user allocation.
1847 * %GFP_KERNEL kernel allocations,
1848 * %GFP_HIGHMEM highmem/user allocations,
1849 * %GFP_FS allocation should not call back into a file system.
1850 * %GFP_ATOMIC don't sleep.
1851 *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001852 * @order:Order of the GFP allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 * @vma: Pointer to VMA or NULL if not available.
1854 * @addr: Virtual Address of the allocation. Must be inside the VMA.
1855 *
1856 * This function allocates a page from the kernel page pool and applies
1857 * a NUMA policy associated with the VMA or the current process.
1858 * When VMA is not NULL caller must hold down_read on the mmap_sem of the
1859 * mm_struct of the VMA to prevent it from going away. Should be used for
1860 * all allocations for pages that will be mapped into
1861 * user space. Returns NULL when no page can be allocated.
1862 *
1863 * Should be called with the mm_sem of the vma hold.
1864 */
1865struct page *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001866alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
Andi Kleen2f5f9482011-03-04 17:36:29 -08001867 unsigned long addr, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868{
Mel Gormancc9a6c82012-03-21 16:34:11 -07001869 struct mempolicy *pol;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001870 struct zonelist *zl;
Miao Xiec0ff7452010-05-24 14:32:08 -07001871 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001872 unsigned int cpuset_mems_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
Mel Gormancc9a6c82012-03-21 16:34:11 -07001874retry_cpuset:
1875 pol = get_vma_policy(current, vma, addr);
1876 cpuset_mems_cookie = get_mems_allowed();
1877
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001878 if (unlikely(pol->mode == MPOL_INTERLEAVE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 unsigned nid;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001880
Andi Kleen8eac5632011-02-25 14:44:28 -08001881 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001882 mpol_cond_put(pol);
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001883 page = alloc_page_interleave(gfp, order, nid);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001884 if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
1885 goto retry_cpuset;
1886
Miao Xiec0ff7452010-05-24 14:32:08 -07001887 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 }
Andi Kleen2f5f9482011-03-04 17:36:29 -08001889 zl = policy_zonelist(gfp, pol, node);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001890 if (unlikely(mpol_needs_cond_ref(pol))) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001891 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001892 * slow path: ref counted shared policy
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001893 */
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001894 struct page *page = __alloc_pages_nodemask(gfp, order,
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001895 zl, policy_nodemask(gfp, pol));
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001896 __mpol_put(pol);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001897 if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
1898 goto retry_cpuset;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001899 return page;
1900 }
1901 /*
1902 * fast path: default or task policy
1903 */
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001904 page = __alloc_pages_nodemask(gfp, order, zl,
1905 policy_nodemask(gfp, pol));
Mel Gormancc9a6c82012-03-21 16:34:11 -07001906 if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
1907 goto retry_cpuset;
Miao Xiec0ff7452010-05-24 14:32:08 -07001908 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909}
1910
1911/**
1912 * alloc_pages_current - Allocate pages.
1913 *
1914 * @gfp:
1915 * %GFP_USER user allocation,
1916 * %GFP_KERNEL kernel allocation,
1917 * %GFP_HIGHMEM highmem allocation,
1918 * %GFP_FS don't call back into a file system.
1919 * %GFP_ATOMIC don't sleep.
1920 * @order: Power of two of allocation size in pages. 0 is a single page.
1921 *
1922 * Allocate a page from the kernel page pool. When not in
1923 * interrupt context and apply the current process NUMA policy.
1924 * Returns NULL when no page can be allocated.
1925 *
Paul Jacksoncf2a473c2006-01-08 01:01:54 -08001926 * Don't call cpuset_update_task_memory_state() unless
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 * 1) it's ok to take cpuset_sem (can WAIT), and
1928 * 2) allocating for current task (not interrupt).
1929 */
Al Virodd0fc662005-10-07 07:46:04 +01001930struct page *alloc_pages_current(gfp_t gfp, unsigned order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931{
1932 struct mempolicy *pol = current->mempolicy;
Miao Xiec0ff7452010-05-24 14:32:08 -07001933 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001934 unsigned int cpuset_mems_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
Christoph Lameter9b819d22006-09-25 23:31:40 -07001936 if (!pol || in_interrupt() || (gfp & __GFP_THISNODE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 pol = &default_policy;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001938
Mel Gormancc9a6c82012-03-21 16:34:11 -07001939retry_cpuset:
1940 cpuset_mems_cookie = get_mems_allowed();
1941
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001942 /*
1943 * No reference counting needed for current->mempolicy
1944 * nor system default_policy
1945 */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001946 if (pol->mode == MPOL_INTERLEAVE)
Miao Xiec0ff7452010-05-24 14:32:08 -07001947 page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
1948 else
1949 page = __alloc_pages_nodemask(gfp, order,
Andi Kleen5c4b4be2011-03-04 17:36:32 -08001950 policy_zonelist(gfp, pol, numa_node_id()),
1951 policy_nodemask(gfp, pol));
Mel Gormancc9a6c82012-03-21 16:34:11 -07001952
1953 if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page))
1954 goto retry_cpuset;
1955
Miao Xiec0ff7452010-05-24 14:32:08 -07001956 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957}
1958EXPORT_SYMBOL(alloc_pages_current);
1959
Paul Jackson42253992006-01-08 01:01:59 -08001960/*
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001961 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
Paul Jackson42253992006-01-08 01:01:59 -08001962 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
1963 * with the mems_allowed returned by cpuset_mems_allowed(). This
1964 * keeps mempolicies cpuset relative after its cpuset moves. See
1965 * further kernel/cpuset.c update_nodemask().
Miao Xie708c1bb2010-05-24 14:32:07 -07001966 *
1967 * current's mempolicy may be rebinded by the other task(the task that changes
1968 * cpuset's mems), so we needn't do rebind work for current task.
Paul Jackson42253992006-01-08 01:01:59 -08001969 */
Paul Jackson42253992006-01-08 01:01:59 -08001970
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07001971/* Slow path of a mempolicy duplicate */
1972struct mempolicy *__mpol_dup(struct mempolicy *old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973{
1974 struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
1975
1976 if (!new)
1977 return ERR_PTR(-ENOMEM);
Miao Xie708c1bb2010-05-24 14:32:07 -07001978
1979 /* task's mempolicy is protected by alloc_lock */
1980 if (old == current->mempolicy) {
1981 task_lock(current);
1982 *new = *old;
1983 task_unlock(current);
1984 } else
1985 *new = *old;
1986
Paul E. McKenney99ee4ca2010-03-03 17:50:17 -08001987 rcu_read_lock();
Paul Jackson42253992006-01-08 01:01:59 -08001988 if (current_cpuset_is_being_rebound()) {
1989 nodemask_t mems = cpuset_mems_allowed(current);
Miao Xie708c1bb2010-05-24 14:32:07 -07001990 if (new->flags & MPOL_F_REBINDING)
1991 mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2);
1992 else
1993 mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE);
Paul Jackson42253992006-01-08 01:01:59 -08001994 }
Paul E. McKenney99ee4ca2010-03-03 17:50:17 -08001995 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 atomic_set(&new->refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 return new;
1998}
1999
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002000/*
2001 * If *frompol needs [has] an extra ref, copy *frompol to *tompol ,
2002 * eliminate the * MPOL_F_* flags that require conditional ref and
2003 * [NOTE!!!] drop the extra ref. Not safe to reference *frompol directly
2004 * after return. Use the returned value.
2005 *
2006 * Allows use of a mempolicy for, e.g., multiple allocations with a single
2007 * policy lookup, even if the policy needs/has extra ref on lookup.
2008 * shmem_readahead needs this.
2009 */
2010struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol,
2011 struct mempolicy *frompol)
2012{
2013 if (!mpol_needs_cond_ref(frompol))
2014 return frompol;
2015
2016 *tompol = *frompol;
2017 tompol->flags &= ~MPOL_F_SHARED; /* copy doesn't need unref */
2018 __mpol_put(frompol);
2019 return tompol;
2020}
2021
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022/* Slow path of a mempolicy comparison */
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002023bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024{
2025 if (!a || !b)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002026 return false;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002027 if (a->mode != b->mode)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002028 return false;
Bob Liu19800502010-05-24 14:32:01 -07002029 if (a->flags != b->flags)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002030 return false;
Bob Liu19800502010-05-24 14:32:01 -07002031 if (mpol_store_user_nodemask(a))
2032 if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask))
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002033 return false;
Bob Liu19800502010-05-24 14:32:01 -07002034
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002035 switch (a->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -07002036 case MPOL_BIND:
2037 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 case MPOL_INTERLEAVE:
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002039 return !!nodes_equal(a->v.nodes, b->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 case MPOL_PREFERRED:
Namhyung Kim75719662011-03-22 16:33:02 -07002041 return a->v.preferred_node == b->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 default:
2043 BUG();
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002044 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 }
2046}
2047
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 * Shared memory backing store policy support.
2050 *
2051 * Remember policies even when nobody has shared memory mapped.
2052 * The policies are kept in Red-Black tree linked from the inode.
2053 * They are protected by the sp->lock spinlock, which should be held
2054 * for any accesses to the tree.
2055 */
2056
2057/* lookup first element intersecting start-end */
2058/* Caller holds sp->lock */
2059static struct sp_node *
2060sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
2061{
2062 struct rb_node *n = sp->root.rb_node;
2063
2064 while (n) {
2065 struct sp_node *p = rb_entry(n, struct sp_node, nd);
2066
2067 if (start >= p->end)
2068 n = n->rb_right;
2069 else if (end <= p->start)
2070 n = n->rb_left;
2071 else
2072 break;
2073 }
2074 if (!n)
2075 return NULL;
2076 for (;;) {
2077 struct sp_node *w = NULL;
2078 struct rb_node *prev = rb_prev(n);
2079 if (!prev)
2080 break;
2081 w = rb_entry(prev, struct sp_node, nd);
2082 if (w->end <= start)
2083 break;
2084 n = prev;
2085 }
2086 return rb_entry(n, struct sp_node, nd);
2087}
2088
2089/* Insert a new shared policy into the list. */
2090/* Caller holds sp->lock */
2091static void sp_insert(struct shared_policy *sp, struct sp_node *new)
2092{
2093 struct rb_node **p = &sp->root.rb_node;
2094 struct rb_node *parent = NULL;
2095 struct sp_node *nd;
2096
2097 while (*p) {
2098 parent = *p;
2099 nd = rb_entry(parent, struct sp_node, nd);
2100 if (new->start < nd->start)
2101 p = &(*p)->rb_left;
2102 else if (new->end > nd->end)
2103 p = &(*p)->rb_right;
2104 else
2105 BUG();
2106 }
2107 rb_link_node(&new->nd, parent, p);
2108 rb_insert_color(&new->nd, &sp->root);
Paul Mundt140d5a42007-07-15 23:38:16 -07002109 pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002110 new->policy ? new->policy->mode : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111}
2112
2113/* Find shared policy intersecting idx */
2114struct mempolicy *
2115mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
2116{
2117 struct mempolicy *pol = NULL;
2118 struct sp_node *sn;
2119
2120 if (!sp->root.rb_node)
2121 return NULL;
2122 spin_lock(&sp->lock);
2123 sn = sp_lookup(sp, idx, idx+1);
2124 if (sn) {
2125 mpol_get(sn->policy);
2126 pol = sn->policy;
2127 }
2128 spin_unlock(&sp->lock);
2129 return pol;
2130}
2131
2132static void sp_delete(struct shared_policy *sp, struct sp_node *n)
2133{
Paul Mundt140d5a42007-07-15 23:38:16 -07002134 pr_debug("deleting %lx-l%lx\n", n->start, n->end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 rb_erase(&n->nd, &sp->root);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07002136 mpol_put(n->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 kmem_cache_free(sn_cache, n);
2138}
2139
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07002140static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
2141 struct mempolicy *pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142{
2143 struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
2144
2145 if (!n)
2146 return NULL;
2147 n->start = start;
2148 n->end = end;
2149 mpol_get(pol);
Lee Schermerhornaab0b102008-04-28 02:13:13 -07002150 pol->flags |= MPOL_F_SHARED; /* for unref */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 n->policy = pol;
2152 return n;
2153}
2154
2155/* Replace a policy range. */
2156static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
2157 unsigned long end, struct sp_node *new)
2158{
2159 struct sp_node *n, *new2 = NULL;
2160
2161restart:
2162 spin_lock(&sp->lock);
2163 n = sp_lookup(sp, start, end);
2164 /* Take care of old policies in the same range. */
2165 while (n && n->start < end) {
2166 struct rb_node *next = rb_next(&n->nd);
2167 if (n->start >= start) {
2168 if (n->end <= end)
2169 sp_delete(sp, n);
2170 else
2171 n->start = end;
2172 } else {
2173 /* Old policy spanning whole new range. */
2174 if (n->end > end) {
2175 if (!new2) {
2176 spin_unlock(&sp->lock);
2177 new2 = sp_alloc(end, n->end, n->policy);
2178 if (!new2)
2179 return -ENOMEM;
2180 goto restart;
2181 }
2182 n->end = start;
2183 sp_insert(sp, new2);
2184 new2 = NULL;
2185 break;
2186 } else
2187 n->end = start;
2188 }
2189 if (!next)
2190 break;
2191 n = rb_entry(next, struct sp_node, nd);
2192 }
2193 if (new)
2194 sp_insert(sp, new);
2195 spin_unlock(&sp->lock);
2196 if (new2) {
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07002197 mpol_put(new2->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 kmem_cache_free(sn_cache, new2);
2199 }
2200 return 0;
2201}
2202
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002203/**
2204 * mpol_shared_policy_init - initialize shared policy for inode
2205 * @sp: pointer to inode shared policy
2206 * @mpol: struct mempolicy to install
2207 *
2208 * Install non-NULL @mpol in inode's shared policy rb-tree.
2209 * On entry, the current task has a reference on a non-NULL @mpol.
2210 * This must be released on exit.
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002211 * This is called at get_inode() calls and we can use GFP_KERNEL.
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002212 */
2213void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
Robin Holt7339ff82006-01-14 13:20:48 -08002214{
Miao Xie58568d22009-06-16 15:31:49 -07002215 int ret;
2216
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002217 sp->root = RB_ROOT; /* empty tree == default mempolicy */
2218 spin_lock_init(&sp->lock);
Robin Holt7339ff82006-01-14 13:20:48 -08002219
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002220 if (mpol) {
2221 struct vm_area_struct pvma;
2222 struct mempolicy *new;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002223 NODEMASK_SCRATCH(scratch);
Robin Holt7339ff82006-01-14 13:20:48 -08002224
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002225 if (!scratch)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002226 goto put_mpol;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002227 /* contextualize the tmpfs mount point mempolicy */
2228 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002229 if (IS_ERR(new))
Dan Carpenter0cae3452010-05-25 23:42:58 -07002230 goto free_scratch; /* no valid nodemask intersection */
Miao Xie58568d22009-06-16 15:31:49 -07002231
2232 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002233 ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
Miao Xie58568d22009-06-16 15:31:49 -07002234 task_unlock(current);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002235 if (ret)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002236 goto put_new;
Robin Holt7339ff82006-01-14 13:20:48 -08002237
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002238 /* Create pseudo-vma that contains just the policy */
2239 memset(&pvma, 0, sizeof(struct vm_area_struct));
2240 pvma.vm_end = TASK_SIZE; /* policy covers entire file */
2241 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002242
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002243put_new:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002244 mpol_put(new); /* drop initial ref */
Dan Carpenter0cae3452010-05-25 23:42:58 -07002245free_scratch:
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002246 NODEMASK_SCRATCH_FREE(scratch);
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002247put_mpol:
2248 mpol_put(mpol); /* drop our incoming ref on sb mpol */
Robin Holt7339ff82006-01-14 13:20:48 -08002249 }
2250}
2251
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252int mpol_set_shared_policy(struct shared_policy *info,
2253 struct vm_area_struct *vma, struct mempolicy *npol)
2254{
2255 int err;
2256 struct sp_node *new = NULL;
2257 unsigned long sz = vma_pages(vma);
2258
David Rientjes028fec42008-04-28 02:12:25 -07002259 pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 vma->vm_pgoff,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002261 sz, npol ? npol->mode : -1,
David Rientjes028fec42008-04-28 02:12:25 -07002262 npol ? npol->flags : -1,
Paul Mundt140d5a42007-07-15 23:38:16 -07002263 npol ? nodes_addr(npol->v.nodes)[0] : -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
2265 if (npol) {
2266 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
2267 if (!new)
2268 return -ENOMEM;
2269 }
2270 err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
2271 if (err && new)
2272 kmem_cache_free(sn_cache, new);
2273 return err;
2274}
2275
2276/* Free a backing policy store on inode delete. */
2277void mpol_free_shared_policy(struct shared_policy *p)
2278{
2279 struct sp_node *n;
2280 struct rb_node *next;
2281
2282 if (!p->root.rb_node)
2283 return;
2284 spin_lock(&p->lock);
2285 next = rb_first(&p->root);
2286 while (next) {
2287 n = rb_entry(next, struct sp_node, nd);
2288 next = rb_next(&n->nd);
Andi Kleen90c50292005-07-27 11:43:50 -07002289 rb_erase(&n->nd, &p->root);
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07002290 mpol_put(n->policy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 kmem_cache_free(sn_cache, n);
2292 }
2293 spin_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294}
2295
2296/* assumes fs == KERNEL_DS */
2297void __init numa_policy_init(void)
2298{
Paul Mundtb71636e2007-07-15 23:38:15 -07002299 nodemask_t interleave_nodes;
2300 unsigned long largest = 0;
2301 int nid, prefer = 0;
2302
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 policy_cache = kmem_cache_create("numa_policy",
2304 sizeof(struct mempolicy),
Paul Mundt20c2df82007-07-20 10:11:58 +09002305 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306
2307 sn_cache = kmem_cache_create("shared_policy_node",
2308 sizeof(struct sp_node),
Paul Mundt20c2df82007-07-20 10:11:58 +09002309 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310
Paul Mundtb71636e2007-07-15 23:38:15 -07002311 /*
2312 * Set interleaving policy for system init. Interleaving is only
2313 * enabled across suitably sized nodes (default is >= 16MB), or
2314 * fall back to the largest node if they're all smaller.
2315 */
2316 nodes_clear(interleave_nodes);
Christoph Lameter56bbd652007-10-16 01:25:35 -07002317 for_each_node_state(nid, N_HIGH_MEMORY) {
Paul Mundtb71636e2007-07-15 23:38:15 -07002318 unsigned long total_pages = node_present_pages(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319
Paul Mundtb71636e2007-07-15 23:38:15 -07002320 /* Preserve the largest node */
2321 if (largest < total_pages) {
2322 largest = total_pages;
2323 prefer = nid;
2324 }
2325
2326 /* Interleave this node? */
2327 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
2328 node_set(nid, interleave_nodes);
2329 }
2330
2331 /* All too small, use the largest */
2332 if (unlikely(nodes_empty(interleave_nodes)))
2333 node_set(prefer, interleave_nodes);
2334
David Rientjes028fec42008-04-28 02:12:25 -07002335 if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 printk("numa_policy_init: interleaving failed\n");
2337}
2338
Christoph Lameter8bccd852005-10-29 18:16:59 -07002339/* Reset policy of current process to default */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340void numa_default_policy(void)
2341{
David Rientjes028fec42008-04-28 02:12:25 -07002342 do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343}
Paul Jackson68860ec2005-10-30 15:02:36 -08002344
Paul Jackson42253992006-01-08 01:01:59 -08002345/*
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002346 * Parse and format mempolicy from/to strings
2347 */
2348
2349/*
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002350 * "local" is pseudo-policy: MPOL_PREFERRED with MPOL_F_LOCAL flag
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002351 * Used only for mpol_parse_str() and mpol_to_str()
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002352 */
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002353#define MPOL_LOCAL MPOL_MAX
2354static const char * const policy_modes[] =
2355{
2356 [MPOL_DEFAULT] = "default",
2357 [MPOL_PREFERRED] = "prefer",
2358 [MPOL_BIND] = "bind",
2359 [MPOL_INTERLEAVE] = "interleave",
2360 [MPOL_LOCAL] = "local"
2361};
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002362
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002363
2364#ifdef CONFIG_TMPFS
2365/**
2366 * mpol_parse_str - parse string to mempolicy
2367 * @str: string containing mempolicy to parse
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002368 * @mpol: pointer to struct mempolicy pointer, returned on success.
2369 * @no_context: flag whether to "contextualize" the mempolicy
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002370 *
2371 * Format of input:
2372 * <mode>[=<flags>][:<nodelist>]
2373 *
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002374 * if @no_context is true, save the input nodemask in w.user_nodemask in
2375 * the returned mempolicy. This will be used to "clone" the mempolicy in
2376 * a specific context [cpuset] at a later time. Used to parse tmpfs mpol
2377 * mount option. Note that if 'static' or 'relative' mode flags were
2378 * specified, the input nodemask will already have been saved. Saving
2379 * it again is redundant, but safe.
2380 *
2381 * On success, returns 0, else 1
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002382 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002383int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002384{
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002385 struct mempolicy *new = NULL;
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002386 unsigned short mode;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002387 unsigned short uninitialized_var(mode_flags);
2388 nodemask_t nodes;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002389 char *nodelist = strchr(str, ':');
2390 char *flags = strchr(str, '=');
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002391 int err = 1;
2392
2393 if (nodelist) {
2394 /* NUL-terminate mode or flags string */
2395 *nodelist++ = '\0';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002396 if (nodelist_parse(nodelist, nodes))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002397 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002398 if (!nodes_subset(nodes, node_states[N_HIGH_MEMORY]))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002399 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002400 } else
2401 nodes_clear(nodes);
2402
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002403 if (flags)
2404 *flags++ = '\0'; /* terminate mode string */
2405
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002406 for (mode = 0; mode <= MPOL_LOCAL; mode++) {
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002407 if (!strcmp(str, policy_modes[mode])) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002408 break;
2409 }
2410 }
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002411 if (mode > MPOL_LOCAL)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002412 goto out;
2413
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002414 switch (mode) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002415 case MPOL_PREFERRED:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002416 /*
2417 * Insist on a nodelist of one node only
2418 */
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002419 if (nodelist) {
2420 char *rest = nodelist;
2421 while (isdigit(*rest))
2422 rest++;
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002423 if (*rest)
2424 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002425 }
2426 break;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002427 case MPOL_INTERLEAVE:
2428 /*
2429 * Default to online nodes with memory if no nodelist
2430 */
2431 if (!nodelist)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002432 nodes = node_states[N_HIGH_MEMORY];
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002433 break;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002434 case MPOL_LOCAL:
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002435 /*
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002436 * Don't allow a nodelist; mpol_new() checks flags
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002437 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002438 if (nodelist)
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002439 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002440 mode = MPOL_PREFERRED;
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002441 break;
Ravikiran G Thirumalai413b43d2010-03-23 13:35:28 -07002442 case MPOL_DEFAULT:
2443 /*
2444 * Insist on a empty nodelist
2445 */
2446 if (!nodelist)
2447 err = 0;
2448 goto out;
KOSAKI Motohirod69b2e62010-03-23 13:35:30 -07002449 case MPOL_BIND:
2450 /*
2451 * Insist on a nodelist
2452 */
2453 if (!nodelist)
2454 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002455 }
2456
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002457 mode_flags = 0;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002458 if (flags) {
2459 /*
2460 * Currently, we only support two mutually exclusive
2461 * mode flags.
2462 */
2463 if (!strcmp(flags, "static"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002464 mode_flags |= MPOL_F_STATIC_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002465 else if (!strcmp(flags, "relative"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002466 mode_flags |= MPOL_F_RELATIVE_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002467 else
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002468 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002469 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002470
2471 new = mpol_new(mode, mode_flags, &nodes);
2472 if (IS_ERR(new))
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002473 goto out;
2474
Lee Schermerhorne17f74a2010-05-24 14:32:02 -07002475 if (no_context) {
2476 /* save for contextualization */
2477 new->w.user_nodemask = nodes;
2478 } else {
Miao Xie58568d22009-06-16 15:31:49 -07002479 int ret;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002480 NODEMASK_SCRATCH(scratch);
2481 if (scratch) {
2482 task_lock(current);
2483 ret = mpol_set_nodemask(new, &nodes, scratch);
2484 task_unlock(current);
2485 } else
2486 ret = -ENOMEM;
2487 NODEMASK_SCRATCH_FREE(scratch);
2488 if (ret) {
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002489 mpol_put(new);
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002490 goto out;
Miao Xie58568d22009-06-16 15:31:49 -07002491 }
2492 }
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002493 err = 0;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002494
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002495out:
2496 /* Restore string for error message */
2497 if (nodelist)
2498 *--nodelist = ':';
2499 if (flags)
2500 *--flags = '=';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002501 if (!err)
2502 *mpol = new;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002503 return err;
2504}
2505#endif /* CONFIG_TMPFS */
2506
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002507/**
2508 * mpol_to_str - format a mempolicy structure for printing
2509 * @buffer: to contain formatted mempolicy string
2510 * @maxlen: length of @buffer
2511 * @pol: pointer to mempolicy to be formatted
2512 * @no_context: "context free" mempolicy - use nodemask in w.user_nodemask
2513 *
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002514 * Convert a mempolicy into a string.
2515 * Returns the number of characters in buffer (if positive)
2516 * or an error (negative)
2517 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002518int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol, int no_context)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002519{
2520 char *p = buffer;
2521 int l;
2522 nodemask_t nodes;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002523 unsigned short mode;
David Rientjesf5b087b2008-04-28 02:12:27 -07002524 unsigned short flags = pol ? pol->flags : 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002525
Lee Schermerhorn22919902008-04-28 02:13:22 -07002526 /*
2527 * Sanity check: room for longest mode, flag and some nodes
2528 */
2529 VM_BUG_ON(maxlen < strlen("interleave") + strlen("relative") + 16);
2530
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002531 if (!pol || pol == &default_policy)
2532 mode = MPOL_DEFAULT;
2533 else
2534 mode = pol->mode;
2535
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002536 switch (mode) {
2537 case MPOL_DEFAULT:
2538 nodes_clear(nodes);
2539 break;
2540
2541 case MPOL_PREFERRED:
2542 nodes_clear(nodes);
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002543 if (flags & MPOL_F_LOCAL)
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002544 mode = MPOL_LOCAL; /* pseudo-policy */
2545 else
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002546 node_set(pol->v.preferred_node, nodes);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002547 break;
2548
2549 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07002550 /* Fall through */
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002551 case MPOL_INTERLEAVE:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002552 if (no_context)
2553 nodes = pol->w.user_nodemask;
2554 else
2555 nodes = pol->v.nodes;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002556 break;
2557
2558 default:
2559 BUG();
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002560 }
2561
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002562 l = strlen(policy_modes[mode]);
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002563 if (buffer + maxlen < p + l + 1)
2564 return -ENOSPC;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002565
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002566 strcpy(p, policy_modes[mode]);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002567 p += l;
2568
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002569 if (flags & MPOL_MODE_FLAGS) {
David Rientjesf5b087b2008-04-28 02:12:27 -07002570 if (buffer + maxlen < p + 2)
2571 return -ENOSPC;
2572 *p++ = '=';
2573
Lee Schermerhorn22919902008-04-28 02:13:22 -07002574 /*
2575 * Currently, the only defined flags are mutually exclusive
2576 */
David Rientjesf5b087b2008-04-28 02:12:27 -07002577 if (flags & MPOL_F_STATIC_NODES)
Lee Schermerhorn22919902008-04-28 02:13:22 -07002578 p += snprintf(p, buffer + maxlen - p, "static");
2579 else if (flags & MPOL_F_RELATIVE_NODES)
2580 p += snprintf(p, buffer + maxlen - p, "relative");
David Rientjesf5b087b2008-04-28 02:12:27 -07002581 }
2582
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002583 if (!nodes_empty(nodes)) {
2584 if (buffer + maxlen < p + 2)
2585 return -ENOSPC;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002586 *p++ = ':';
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002587 p += nodelist_scnprintf(p, buffer + maxlen - p, nodes);
2588 }
2589 return p - buffer;
2590}