blob: 297d6854f84920f4420d5929fd3bc1084dd8a15e [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.
David Rientjes00ef2d22013-02-22 16:35:36 -080029 * As a special case NUMA_NO_NODE here means do the allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 * 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
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070068#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
69
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <linux/mempolicy.h>
71#include <linux/mm.h>
72#include <linux/highmem.h>
73#include <linux/hugetlb.h>
74#include <linux/kernel.h>
75#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#include <linux/nodemask.h>
77#include <linux/cpuset.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070078#include <linux/slab.h>
79#include <linux/string.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040080#include <linux/export.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070081#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070082#include <linux/interrupt.h>
83#include <linux/init.h>
84#include <linux/compat.h>
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -080085#include <linux/swap.h>
Christoph Lameter1a75a6c2006-01-08 01:01:02 -080086#include <linux/seq_file.h>
87#include <linux/proc_fs.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080088#include <linux/migrate.h>
Hugh Dickins62b61f62009-12-14 17:59:33 -080089#include <linux/ksm.h>
Christoph Lameter95a402c2006-06-23 02:03:53 -070090#include <linux/rmap.h>
David Quigley86c3a762006-06-23 02:04:02 -070091#include <linux/security.h>
Adrian Bunkdbcb0f12007-10-16 01:26:26 -070092#include <linux/syscalls.h>
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -070093#include <linux/ctype.h>
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -080094#include <linux/mm_inline.h>
Lee Schermerhornb24f53a2012-10-25 14:16:32 +020095#include <linux/mmu_notifier.h>
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070096#include <linux/printk.h>
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -080097
Linus Torvalds1da177e2005-04-16 15:20:36 -070098#include <asm/tlbflush.h>
99#include <asm/uaccess.h>
100
Nick Piggin62695a82008-10-18 20:26:09 -0700101#include "internal.h"
102
Christoph Lameter38e35862006-01-08 01:01:01 -0800103/* Internal flags */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800104#define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */
Christoph Lameter38e35862006-01-08 01:01:01 -0800105#define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800106
Pekka Enbergfcc234f2006-03-22 00:08:13 -0800107static struct kmem_cache *policy_cache;
108static struct kmem_cache *sn_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110/* Highest zone. An specific allocation for a zone below that is not
111 policied. */
Christoph Lameter62672762007-02-10 01:43:07 -0800112enum zone_type policy_zone = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700114/*
115 * run-time system-wide default policy => local allocation
116 */
H Hartley Sweetene754d792011-10-31 17:09:23 -0700117static struct mempolicy default_policy = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 .refcnt = ATOMIC_INIT(1), /* never free it */
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700119 .mode = MPOL_PREFERRED,
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700120 .flags = MPOL_F_LOCAL,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121};
122
Mel Gorman5606e382012-11-02 18:19:13 +0000123static struct mempolicy preferred_node_policy[MAX_NUMNODES];
124
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -0700125struct mempolicy *get_task_policy(struct task_struct *p)
Mel Gorman5606e382012-11-02 18:19:13 +0000126{
127 struct mempolicy *pol = p->mempolicy;
Oleg Nesterovf15ca782014-10-09 15:27:43 -0700128 int node;
Mel Gorman5606e382012-11-02 18:19:13 +0000129
Oleg Nesterovf15ca782014-10-09 15:27:43 -0700130 if (pol)
131 return pol;
Mel Gorman5606e382012-11-02 18:19:13 +0000132
Oleg Nesterovf15ca782014-10-09 15:27:43 -0700133 node = numa_node_id();
134 if (node != NUMA_NO_NODE) {
135 pol = &preferred_node_policy[node];
136 /* preferred_node_policy is not initialised early in boot */
137 if (pol->mode)
138 return pol;
Mel Gorman5606e382012-11-02 18:19:13 +0000139 }
140
Oleg Nesterovf15ca782014-10-09 15:27:43 -0700141 return &default_policy;
Mel Gorman5606e382012-11-02 18:19:13 +0000142}
143
David Rientjes37012942008-04-28 02:12:33 -0700144static const struct mempolicy_operations {
145 int (*create)(struct mempolicy *pol, const nodemask_t *nodes);
Miao Xie708c1bb2010-05-24 14:32:07 -0700146 /*
147 * If read-side task has no lock to protect task->mempolicy, write-side
148 * task will rebind the task->mempolicy by two step. The first step is
149 * setting all the newly nodes, and the second step is cleaning all the
150 * disallowed nodes. In this way, we can avoid finding no node to alloc
151 * page.
152 * If we have a lock to protect task->mempolicy in read-side, we do
153 * rebind directly.
154 *
155 * step:
156 * MPOL_REBIND_ONCE - do rebind work at once
157 * MPOL_REBIND_STEP1 - set all the newly nodes
158 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
159 */
160 void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes,
161 enum mpol_rebind_step step);
David Rientjes37012942008-04-28 02:12:33 -0700162} mpol_ops[MPOL_MAX];
163
David Rientjesf5b087b2008-04-28 02:12:27 -0700164static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
165{
Bob Liu6d556292010-05-24 14:31:59 -0700166 return pol->flags & MPOL_MODE_FLAGS;
David Rientjes4c50bc02008-04-28 02:12:30 -0700167}
168
169static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig,
170 const nodemask_t *rel)
171{
172 nodemask_t tmp;
173 nodes_fold(tmp, *orig, nodes_weight(*rel));
174 nodes_onto(*ret, tmp, *rel);
David Rientjesf5b087b2008-04-28 02:12:27 -0700175}
176
David Rientjes37012942008-04-28 02:12:33 -0700177static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes)
178{
179 if (nodes_empty(*nodes))
180 return -EINVAL;
181 pol->v.nodes = *nodes;
182 return 0;
183}
184
185static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes)
186{
187 if (!nodes)
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700188 pol->flags |= MPOL_F_LOCAL; /* local allocation */
David Rientjes37012942008-04-28 02:12:33 -0700189 else if (nodes_empty(*nodes))
190 return -EINVAL; /* no allowed nodes */
191 else
192 pol->v.preferred_node = first_node(*nodes);
193 return 0;
194}
195
196static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes)
197{
Zhihui Zhang859f7ef2014-12-18 16:17:09 -0800198 if (nodes_empty(*nodes))
David Rientjes37012942008-04-28 02:12:33 -0700199 return -EINVAL;
200 pol->v.nodes = *nodes;
201 return 0;
202}
203
Miao Xie58568d22009-06-16 15:31:49 -0700204/*
205 * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if
206 * any, for the new policy. mpol_new() has already validated the nodes
207 * parameter with respect to the policy mode and flags. But, we need to
208 * handle an empty nodemask with MPOL_PREFERRED here.
209 *
210 * Must be called holding task's alloc_lock to protect task's mems_allowed
211 * and mempolicy. May also be called holding the mmap_semaphore for write.
212 */
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700213static int mpol_set_nodemask(struct mempolicy *pol,
214 const nodemask_t *nodes, struct nodemask_scratch *nsc)
Miao Xie58568d22009-06-16 15:31:49 -0700215{
Miao Xie58568d22009-06-16 15:31:49 -0700216 int ret;
217
218 /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */
219 if (pol == NULL)
220 return 0;
Lai Jiangshan01f13bd2012-12-12 13:51:33 -0800221 /* Check N_MEMORY */
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700222 nodes_and(nsc->mask1,
Lai Jiangshan01f13bd2012-12-12 13:51:33 -0800223 cpuset_current_mems_allowed, node_states[N_MEMORY]);
Miao Xie58568d22009-06-16 15:31:49 -0700224
225 VM_BUG_ON(!nodes);
226 if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes))
227 nodes = NULL; /* explicit local allocation */
228 else {
229 if (pol->flags & MPOL_F_RELATIVE_NODES)
Zhihui Zhang859f7ef2014-12-18 16:17:09 -0800230 mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1);
Miao Xie58568d22009-06-16 15:31:49 -0700231 else
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700232 nodes_and(nsc->mask2, *nodes, nsc->mask1);
233
Miao Xie58568d22009-06-16 15:31:49 -0700234 if (mpol_store_user_nodemask(pol))
235 pol->w.user_nodemask = *nodes;
236 else
237 pol->w.cpuset_mems_allowed =
238 cpuset_current_mems_allowed;
239 }
240
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700241 if (nodes)
242 ret = mpol_ops[pol->mode].create(pol, &nsc->mask2);
243 else
244 ret = mpol_ops[pol->mode].create(pol, NULL);
Miao Xie58568d22009-06-16 15:31:49 -0700245 return ret;
246}
247
248/*
249 * This function just creates a new policy, does some check and simple
250 * initialization. You must invoke mpol_set_nodemask() to set nodes.
251 */
David Rientjes028fec42008-04-28 02:12:25 -0700252static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
253 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254{
255 struct mempolicy *policy;
256
David Rientjes028fec42008-04-28 02:12:25 -0700257 pr_debug("setting mode %d flags %d nodes[0] %lx\n",
David Rientjes00ef2d22013-02-22 16:35:36 -0800258 mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE);
Paul Mundt140d5a42007-07-15 23:38:16 -0700259
David Rientjes3e1f0642008-04-28 02:12:34 -0700260 if (mode == MPOL_DEFAULT) {
261 if (nodes && !nodes_empty(*nodes))
David Rientjes37012942008-04-28 02:12:33 -0700262 return ERR_PTR(-EINVAL);
Lee Schermerhornd3a71032012-10-25 14:16:29 +0200263 return NULL;
David Rientjes37012942008-04-28 02:12:33 -0700264 }
David Rientjes3e1f0642008-04-28 02:12:34 -0700265 VM_BUG_ON(!nodes);
266
267 /*
268 * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
269 * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
270 * All other modes require a valid pointer to a non-empty nodemask.
271 */
272 if (mode == MPOL_PREFERRED) {
273 if (nodes_empty(*nodes)) {
274 if (((flags & MPOL_F_STATIC_NODES) ||
275 (flags & MPOL_F_RELATIVE_NODES)))
276 return ERR_PTR(-EINVAL);
David Rientjes3e1f0642008-04-28 02:12:34 -0700277 }
Peter Zijlstra479e2802012-10-25 14:16:28 +0200278 } else if (mode == MPOL_LOCAL) {
279 if (!nodes_empty(*nodes))
280 return ERR_PTR(-EINVAL);
281 mode = MPOL_PREFERRED;
David Rientjes3e1f0642008-04-28 02:12:34 -0700282 } else if (nodes_empty(*nodes))
283 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
285 if (!policy)
286 return ERR_PTR(-ENOMEM);
287 atomic_set(&policy->refcnt, 1);
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700288 policy->mode = mode;
David Rientjes3e1f0642008-04-28 02:12:34 -0700289 policy->flags = flags;
David Rientjesf5b087b2008-04-28 02:12:27 -0700290
David Rientjes37012942008-04-28 02:12:33 -0700291 return policy;
292}
293
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700294/* Slow path of a mpol destructor. */
295void __mpol_put(struct mempolicy *p)
296{
297 if (!atomic_dec_and_test(&p->refcnt))
298 return;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700299 kmem_cache_free(policy_cache, p);
300}
301
Miao Xie708c1bb2010-05-24 14:32:07 -0700302static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes,
303 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700304{
305}
306
Miao Xie708c1bb2010-05-24 14:32:07 -0700307/*
308 * step:
309 * MPOL_REBIND_ONCE - do rebind work at once
310 * MPOL_REBIND_STEP1 - set all the newly nodes
311 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
312 */
313static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes,
314 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700315{
316 nodemask_t tmp;
317
318 if (pol->flags & MPOL_F_STATIC_NODES)
319 nodes_and(tmp, pol->w.user_nodemask, *nodes);
320 else if (pol->flags & MPOL_F_RELATIVE_NODES)
321 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
322 else {
Miao Xie708c1bb2010-05-24 14:32:07 -0700323 /*
324 * if step == 1, we use ->w.cpuset_mems_allowed to cache the
325 * result
326 */
327 if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) {
328 nodes_remap(tmp, pol->v.nodes,
329 pol->w.cpuset_mems_allowed, *nodes);
330 pol->w.cpuset_mems_allowed = step ? tmp : *nodes;
331 } else if (step == MPOL_REBIND_STEP2) {
332 tmp = pol->w.cpuset_mems_allowed;
333 pol->w.cpuset_mems_allowed = *nodes;
334 } else
335 BUG();
David Rientjes37012942008-04-28 02:12:33 -0700336 }
337
Miao Xie708c1bb2010-05-24 14:32:07 -0700338 if (nodes_empty(tmp))
339 tmp = *nodes;
340
341 if (step == MPOL_REBIND_STEP1)
342 nodes_or(pol->v.nodes, pol->v.nodes, tmp);
343 else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2)
344 pol->v.nodes = tmp;
345 else
346 BUG();
347
David Rientjes37012942008-04-28 02:12:33 -0700348 if (!node_isset(current->il_next, tmp)) {
Andrew Morton0edaf862016-05-19 17:10:58 -0700349 current->il_next = next_node_in(current->il_next, tmp);
David Rientjes37012942008-04-28 02:12:33 -0700350 if (current->il_next >= MAX_NUMNODES)
351 current->il_next = numa_node_id();
352 }
353}
354
355static void mpol_rebind_preferred(struct mempolicy *pol,
Miao Xie708c1bb2010-05-24 14:32:07 -0700356 const nodemask_t *nodes,
357 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700358{
359 nodemask_t tmp;
360
David Rientjes37012942008-04-28 02:12:33 -0700361 if (pol->flags & MPOL_F_STATIC_NODES) {
362 int node = first_node(pol->w.user_nodemask);
363
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700364 if (node_isset(node, *nodes)) {
David Rientjes37012942008-04-28 02:12:33 -0700365 pol->v.preferred_node = node;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700366 pol->flags &= ~MPOL_F_LOCAL;
367 } else
368 pol->flags |= MPOL_F_LOCAL;
David Rientjes37012942008-04-28 02:12:33 -0700369 } else if (pol->flags & MPOL_F_RELATIVE_NODES) {
370 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
371 pol->v.preferred_node = first_node(tmp);
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700372 } else if (!(pol->flags & MPOL_F_LOCAL)) {
David Rientjes37012942008-04-28 02:12:33 -0700373 pol->v.preferred_node = node_remap(pol->v.preferred_node,
374 pol->w.cpuset_mems_allowed,
375 *nodes);
376 pol->w.cpuset_mems_allowed = *nodes;
377 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378}
379
Miao Xie708c1bb2010-05-24 14:32:07 -0700380/*
381 * mpol_rebind_policy - Migrate a policy to a different set of nodes
382 *
383 * If read-side task has no lock to protect task->mempolicy, write-side
384 * task will rebind the task->mempolicy by two step. The first step is
385 * setting all the newly nodes, and the second step is cleaning all the
386 * disallowed nodes. In this way, we can avoid finding no node to alloc
387 * page.
388 * If we have a lock to protect task->mempolicy in read-side, we do
389 * rebind directly.
390 *
391 * step:
392 * MPOL_REBIND_ONCE - do rebind work at once
393 * MPOL_REBIND_STEP1 - set all the newly nodes
394 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
395 */
396static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask,
397 enum mpol_rebind_step step)
David Rientjes1d0d2682008-04-28 02:12:32 -0700398{
David Rientjes1d0d2682008-04-28 02:12:32 -0700399 if (!pol)
400 return;
Wang Sheng-Hui89c522c2012-05-29 15:06:16 -0700401 if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE &&
David Rientjes1d0d2682008-04-28 02:12:32 -0700402 nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
403 return;
Miao Xie708c1bb2010-05-24 14:32:07 -0700404
405 if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING))
406 return;
407
408 if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING))
409 BUG();
410
411 if (step == MPOL_REBIND_STEP1)
412 pol->flags |= MPOL_F_REBINDING;
413 else if (step == MPOL_REBIND_STEP2)
414 pol->flags &= ~MPOL_F_REBINDING;
415 else if (step >= MPOL_REBIND_NSTEP)
416 BUG();
417
418 mpol_ops[pol->mode].rebind(pol, newmask, step);
David Rientjes1d0d2682008-04-28 02:12:32 -0700419}
420
421/*
422 * Wrapper for mpol_rebind_policy() that just requires task
423 * pointer, and updates task mempolicy.
Miao Xie58568d22009-06-16 15:31:49 -0700424 *
425 * Called with task's alloc_lock held.
David Rientjes1d0d2682008-04-28 02:12:32 -0700426 */
427
Miao Xie708c1bb2010-05-24 14:32:07 -0700428void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
429 enum mpol_rebind_step step)
David Rientjes1d0d2682008-04-28 02:12:32 -0700430{
Miao Xie708c1bb2010-05-24 14:32:07 -0700431 mpol_rebind_policy(tsk->mempolicy, new, step);
David Rientjes1d0d2682008-04-28 02:12:32 -0700432}
433
434/*
435 * Rebind each vma in mm to new nodemask.
436 *
437 * Call holding a reference to mm. Takes mm->mmap_sem during call.
438 */
439
440void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
441{
442 struct vm_area_struct *vma;
443
444 down_write(&mm->mmap_sem);
445 for (vma = mm->mmap; vma; vma = vma->vm_next)
Miao Xie708c1bb2010-05-24 14:32:07 -0700446 mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE);
David Rientjes1d0d2682008-04-28 02:12:32 -0700447 up_write(&mm->mmap_sem);
448}
449
David Rientjes37012942008-04-28 02:12:33 -0700450static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
451 [MPOL_DEFAULT] = {
452 .rebind = mpol_rebind_default,
453 },
454 [MPOL_INTERLEAVE] = {
455 .create = mpol_new_interleave,
456 .rebind = mpol_rebind_nodemask,
457 },
458 [MPOL_PREFERRED] = {
459 .create = mpol_new_preferred,
460 .rebind = mpol_rebind_preferred,
461 },
462 [MPOL_BIND] = {
463 .create = mpol_new_bind,
464 .rebind = mpol_rebind_nodemask,
465 },
466};
467
Christoph Lameterfc301282006-01-18 17:42:29 -0800468static void migrate_page_add(struct page *page, struct list_head *pagelist,
469 unsigned long flags);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800470
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800471struct queue_pages {
472 struct list_head *pagelist;
473 unsigned long flags;
474 nodemask_t *nmask;
475 struct vm_area_struct *prev;
476};
477
Naoya Horiguchi98094942013-09-11 14:22:14 -0700478/*
479 * Scan through pages checking if pages follow certain conditions,
480 * and move them to the pagelist if they do.
481 */
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800482static int queue_pages_pte_range(pmd_t *pmd, unsigned long addr,
483 unsigned long end, struct mm_walk *walk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484{
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800485 struct vm_area_struct *vma = walk->vma;
486 struct page *page;
487 struct queue_pages *qp = walk->private;
488 unsigned long flags = qp->flags;
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800489 int nid, ret;
Hugh Dickins91612e02005-06-21 17:15:07 -0700490 pte_t *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700491 spinlock_t *ptl;
Hugh Dickins941150a2005-06-21 17:15:06 -0700492
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800493 if (pmd_trans_huge(*pmd)) {
494 ptl = pmd_lock(walk->mm, pmd);
495 if (pmd_trans_huge(*pmd)) {
496 page = pmd_page(*pmd);
497 if (is_huge_zero_page(page)) {
498 spin_unlock(ptl);
499 split_huge_pmd(vma, pmd, addr);
500 } else {
501 get_page(page);
502 spin_unlock(ptl);
503 lock_page(page);
504 ret = split_huge_page(page);
505 unlock_page(page);
506 put_page(page);
507 if (ret)
508 return 0;
509 }
510 } else {
511 spin_unlock(ptl);
512 }
513 }
Hugh Dickins91612e02005-06-21 17:15:07 -0700514
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800515retry:
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800516 pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
517 for (; addr != end; pte++, addr += PAGE_SIZE) {
Hugh Dickins91612e02005-06-21 17:15:07 -0700518 if (!pte_present(*pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800520 page = vm_normal_page(vma, addr, *pte);
521 if (!page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 continue;
Nick Piggin053837f2006-01-18 17:42:27 -0800523 /*
Hugh Dickins62b61f62009-12-14 17:59:33 -0800524 * vm_normal_page() filters out zero pages, but there might
525 * still be PageReserved pages to skip, perhaps in a VDSO.
Nick Piggin053837f2006-01-18 17:42:27 -0800526 */
Hugh Dickinsb79bc0a2013-02-22 16:35:13 -0800527 if (PageReserved(page))
Christoph Lameterf4598c82006-01-12 01:05:20 -0800528 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800529 nid = page_to_nid(page);
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800530 if (node_isset(nid, *qp->nmask) == !!(flags & MPOL_MF_INVERT))
Christoph Lameter38e35862006-01-08 01:01:01 -0800531 continue;
Kirill A. Shutemov0a2e2802016-03-09 14:08:30 -0800532 if (PageTransCompound(page) && PageAnon(page)) {
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800533 get_page(page);
534 pte_unmap_unlock(pte, ptl);
535 lock_page(page);
536 ret = split_huge_page(page);
537 unlock_page(page);
538 put_page(page);
539 /* Failed to split -- skip. */
540 if (ret) {
541 pte = pte_offset_map_lock(walk->mm, pmd,
542 addr, &ptl);
543 continue;
544 }
545 goto retry;
546 }
Christoph Lameter38e35862006-01-08 01:01:01 -0800547
Kirill A. Shutemov77bf45e2016-02-05 15:36:33 -0800548 migrate_page_add(page, qp->pagelist, flags);
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800549 }
550 pte_unmap_unlock(pte - 1, ptl);
551 cond_resched();
552 return 0;
Hugh Dickins91612e02005-06-21 17:15:07 -0700553}
554
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800555static int queue_pages_hugetlb(pte_t *pte, unsigned long hmask,
556 unsigned long addr, unsigned long end,
557 struct mm_walk *walk)
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700558{
559#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800560 struct queue_pages *qp = walk->private;
561 unsigned long flags = qp->flags;
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700562 int nid;
563 struct page *page;
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800564 spinlock_t *ptl;
Naoya Horiguchid4c54912014-06-06 10:00:01 -0400565 pte_t entry;
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700566
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800567 ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte);
568 entry = huge_ptep_get(pte);
Naoya Horiguchid4c54912014-06-06 10:00:01 -0400569 if (!pte_present(entry))
570 goto unlock;
571 page = pte_page(entry);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700572 nid = page_to_nid(page);
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800573 if (node_isset(nid, *qp->nmask) == !!(flags & MPOL_MF_INVERT))
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700574 goto unlock;
575 /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */
576 if (flags & (MPOL_MF_MOVE_ALL) ||
577 (flags & MPOL_MF_MOVE && page_mapcount(page) == 1))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800578 isolate_huge_page(page, qp->pagelist);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700579unlock:
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800580 spin_unlock(ptl);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700581#else
582 BUG();
583#endif
Hugh Dickins91612e02005-06-21 17:15:07 -0700584 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585}
586
Aneesh Kumar K.V58772312013-12-06 00:08:22 +0530587#ifdef CONFIG_NUMA_BALANCING
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200588/*
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200589 * This is used to mark a range of virtual addresses to be inaccessible.
590 * These are later cleared by a NUMA hinting fault. Depending on these
591 * faults, pages may be migrated for better NUMA placement.
592 *
593 * This is assuming that NUMA faults are handled using PROT_NONE. If
594 * an architecture makes a different choice, it will need further
595 * changes to the core.
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200596 */
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200597unsigned long change_prot_numa(struct vm_area_struct *vma,
598 unsigned long addr, unsigned long end)
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200599{
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200600 int nr_updated;
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200601
Mel Gorman4d942462015-02-12 14:58:28 -0800602 nr_updated = change_protection(vma, addr, end, PAGE_NONE, 0, 1);
Mel Gorman03c5a6e2012-11-02 14:52:48 +0000603 if (nr_updated)
604 count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated);
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200605
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200606 return nr_updated;
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200607}
608#else
609static unsigned long change_prot_numa(struct vm_area_struct *vma,
610 unsigned long addr, unsigned long end)
611{
612 return 0;
613}
Aneesh Kumar K.V58772312013-12-06 00:08:22 +0530614#endif /* CONFIG_NUMA_BALANCING */
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200615
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800616static int queue_pages_test_walk(unsigned long start, unsigned long end,
617 struct mm_walk *walk)
618{
619 struct vm_area_struct *vma = walk->vma;
620 struct queue_pages *qp = walk->private;
621 unsigned long endvma = vma->vm_end;
622 unsigned long flags = qp->flags;
623
Kirill A. Shutemov77bf45e2016-02-05 15:36:33 -0800624 if (!vma_migratable(vma))
Naoya Horiguchi48684a62015-02-11 15:28:06 -0800625 return 1;
626
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800627 if (endvma > end)
628 endvma = end;
629 if (vma->vm_start > start)
630 start = vma->vm_start;
631
632 if (!(flags & MPOL_MF_DISCONTIG_OK)) {
633 if (!vma->vm_next && vma->vm_end < end)
634 return -EFAULT;
635 if (qp->prev && qp->prev->vm_end < vma->vm_start)
636 return -EFAULT;
637 }
638
639 qp->prev = vma;
640
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800641 if (flags & MPOL_MF_LAZY) {
642 /* Similar to task_numa_work, skip inaccessible VMAs */
Liang Chen4355c012016-03-15 14:56:42 -0700643 if (!is_vm_hugetlb_page(vma) &&
644 (vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)) &&
645 !(vma->vm_flags & VM_MIXEDMAP))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800646 change_prot_numa(vma, start, endvma);
647 return 1;
648 }
649
Kirill A. Shutemov77bf45e2016-02-05 15:36:33 -0800650 /* queue pages from current vma */
651 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800652 return 0;
653 return 1;
654}
655
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800656/*
Naoya Horiguchi98094942013-09-11 14:22:14 -0700657 * Walk through page tables and collect pages to be migrated.
658 *
659 * If pages found in a given range are on a set of nodes (determined by
660 * @nodes and @flags,) it's isolated and queued to the pagelist which is
661 * passed via @private.)
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800662 */
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -0700663static int
Naoya Horiguchi98094942013-09-11 14:22:14 -0700664queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800665 nodemask_t *nodes, unsigned long flags,
666 struct list_head *pagelist)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800668 struct queue_pages qp = {
669 .pagelist = pagelist,
670 .flags = flags,
671 .nmask = nodes,
672 .prev = NULL,
673 };
674 struct mm_walk queue_pages_walk = {
675 .hugetlb_entry = queue_pages_hugetlb,
676 .pmd_entry = queue_pages_pte_range,
677 .test_walk = queue_pages_test_walk,
678 .mm = mm,
679 .private = &qp,
680 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800682 return walk_page_range(start, end, &queue_pages_walk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683}
684
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700685/*
686 * Apply policy to a single VMA
687 * This must be called with the mmap_sem held for writing.
688 */
689static int vma_replace_policy(struct vm_area_struct *vma,
690 struct mempolicy *pol)
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700691{
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700692 int err;
693 struct mempolicy *old;
694 struct mempolicy *new;
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700695
696 pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
697 vma->vm_start, vma->vm_end, vma->vm_pgoff,
698 vma->vm_ops, vma->vm_file,
699 vma->vm_ops ? vma->vm_ops->set_policy : NULL);
700
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700701 new = mpol_dup(pol);
702 if (IS_ERR(new))
703 return PTR_ERR(new);
704
705 if (vma->vm_ops && vma->vm_ops->set_policy) {
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700706 err = vma->vm_ops->set_policy(vma, new);
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700707 if (err)
708 goto err_out;
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700709 }
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700710
711 old = vma->vm_policy;
712 vma->vm_policy = new; /* protected by mmap_sem */
713 mpol_put(old);
714
715 return 0;
716 err_out:
717 mpol_put(new);
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700718 return err;
719}
720
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721/* Step 2: apply policy to a range and do splits. */
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800722static int mbind_range(struct mm_struct *mm, unsigned long start,
723 unsigned long end, struct mempolicy *new_pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724{
725 struct vm_area_struct *next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800726 struct vm_area_struct *prev;
727 struct vm_area_struct *vma;
728 int err = 0;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800729 pgoff_t pgoff;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800730 unsigned long vmstart;
731 unsigned long vmend;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Linus Torvalds097d5912012-03-06 18:23:36 -0800733 vma = find_vma(mm, start);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800734 if (!vma || vma->vm_start > start)
735 return -EFAULT;
736
Linus Torvalds097d5912012-03-06 18:23:36 -0800737 prev = vma->vm_prev;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800738 if (start > vma->vm_start)
739 prev = vma;
740
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800741 for (; vma && vma->vm_start < end; prev = vma, vma = next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 next = vma->vm_next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800743 vmstart = max(start, vma->vm_start);
744 vmend = min(end, vma->vm_end);
745
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800746 if (mpol_equal(vma_policy(vma), new_pol))
747 continue;
748
749 pgoff = vma->vm_pgoff +
750 ((vmstart - vma->vm_start) >> PAGE_SHIFT);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800751 prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700752 vma->anon_vma, vma->vm_file, pgoff,
753 new_pol, vma->vm_userfaultfd_ctx);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800754 if (prev) {
755 vma = prev;
756 next = vma->vm_next;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700757 if (mpol_equal(vma_policy(vma), new_pol))
758 continue;
759 /* vma_merge() joined vma && vma->next, case 8 */
760 goto replace;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800761 }
762 if (vma->vm_start != vmstart) {
763 err = split_vma(vma->vm_mm, vma, vmstart, 1);
764 if (err)
765 goto out;
766 }
767 if (vma->vm_end != vmend) {
768 err = split_vma(vma->vm_mm, vma, vmend, 0);
769 if (err)
770 goto out;
771 }
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700772 replace:
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700773 err = vma_replace_policy(vma, new_pol);
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700774 if (err)
775 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 }
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800777
778 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 return err;
780}
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782/* Set the process memory policy */
David Rientjes028fec42008-04-28 02:12:25 -0700783static long do_set_mempolicy(unsigned short mode, unsigned short flags,
784 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
Miao Xie58568d22009-06-16 15:31:49 -0700786 struct mempolicy *new, *old;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700787 NODEMASK_SCRATCH(scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700788 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700790 if (!scratch)
791 return -ENOMEM;
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700792
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700793 new = mpol_new(mode, flags, nodes);
794 if (IS_ERR(new)) {
795 ret = PTR_ERR(new);
796 goto out;
797 }
Oleg Nesterov2c7c3a72014-10-09 15:27:55 -0700798
Miao Xie58568d22009-06-16 15:31:49 -0700799 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700800 ret = mpol_set_nodemask(new, nodes, scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700801 if (ret) {
802 task_unlock(current);
Miao Xie58568d22009-06-16 15:31:49 -0700803 mpol_put(new);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700804 goto out;
Miao Xie58568d22009-06-16 15:31:49 -0700805 }
806 old = current->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 current->mempolicy = new;
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700808 if (new && new->mode == MPOL_INTERLEAVE &&
David Rientjesf5b087b2008-04-28 02:12:27 -0700809 nodes_weight(new->v.nodes))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700810 current->il_next = first_node(new->v.nodes);
Miao Xie58568d22009-06-16 15:31:49 -0700811 task_unlock(current);
Miao Xie58568d22009-06-16 15:31:49 -0700812 mpol_put(old);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700813 ret = 0;
814out:
815 NODEMASK_SCRATCH_FREE(scratch);
816 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817}
818
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700819/*
820 * Return nodemask for policy for get_mempolicy() query
Miao Xie58568d22009-06-16 15:31:49 -0700821 *
822 * Called with task's alloc_lock held
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700823 */
824static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825{
Andi Kleendfcd3c02005-10-29 18:15:48 -0700826 nodes_clear(*nodes);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700827 if (p == &default_policy)
828 return;
829
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700830 switch (p->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -0700831 case MPOL_BIND:
832 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -0700834 *nodes = p->v.nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 break;
836 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700837 if (!(p->flags & MPOL_F_LOCAL))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700838 node_set(p->v.preferred_node, *nodes);
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700839 /* else return empty node mask for local allocation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 break;
841 default:
842 BUG();
843 }
844}
845
Dave Hansend4edcf02016-02-12 13:01:56 -0800846static int lookup_node(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847{
848 struct page *p;
849 int err;
850
Dave Hansend4edcf02016-02-12 13:01:56 -0800851 err = get_user_pages(addr & PAGE_MASK, 1, 0, 0, &p, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 if (err >= 0) {
853 err = page_to_nid(p);
854 put_page(p);
855 }
856 return err;
857}
858
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859/* Retrieve NUMA policy */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700860static long do_get_mempolicy(int *policy, nodemask_t *nmask,
861 unsigned long addr, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862{
Christoph Lameter8bccd852005-10-29 18:16:59 -0700863 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 struct mm_struct *mm = current->mm;
865 struct vm_area_struct *vma = NULL;
866 struct mempolicy *pol = current->mempolicy;
867
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700868 if (flags &
869 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 return -EINVAL;
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700871
872 if (flags & MPOL_F_MEMS_ALLOWED) {
873 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
874 return -EINVAL;
875 *policy = 0; /* just so it's initialized */
Miao Xie58568d22009-06-16 15:31:49 -0700876 task_lock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700877 *nmask = cpuset_current_mems_allowed;
Miao Xie58568d22009-06-16 15:31:49 -0700878 task_unlock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700879 return 0;
880 }
881
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 if (flags & MPOL_F_ADDR) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700883 /*
884 * Do NOT fall back to task policy if the
885 * vma/shared policy at addr is NULL. We
886 * want to return MPOL_DEFAULT in this case.
887 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 down_read(&mm->mmap_sem);
889 vma = find_vma_intersection(mm, addr, addr+1);
890 if (!vma) {
891 up_read(&mm->mmap_sem);
892 return -EFAULT;
893 }
894 if (vma->vm_ops && vma->vm_ops->get_policy)
895 pol = vma->vm_ops->get_policy(vma, addr);
896 else
897 pol = vma->vm_policy;
898 } else if (addr)
899 return -EINVAL;
900
901 if (!pol)
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700902 pol = &default_policy; /* indicates default behavior */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
904 if (flags & MPOL_F_NODE) {
905 if (flags & MPOL_F_ADDR) {
Dave Hansend4edcf02016-02-12 13:01:56 -0800906 err = lookup_node(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 if (err < 0)
908 goto out;
Christoph Lameter8bccd852005-10-29 18:16:59 -0700909 *policy = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 } else if (pol == current->mempolicy &&
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700911 pol->mode == MPOL_INTERLEAVE) {
Christoph Lameter8bccd852005-10-29 18:16:59 -0700912 *policy = current->il_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 } else {
914 err = -EINVAL;
915 goto out;
916 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700917 } else {
918 *policy = pol == &default_policy ? MPOL_DEFAULT :
919 pol->mode;
David Rientjesd79df632008-07-04 12:24:13 -0700920 /*
921 * Internal mempolicy flags must be masked off before exposing
922 * the policy to userspace.
923 */
924 *policy |= (pol->flags & MPOL_MODE_FLAGS);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700925 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
927 if (vma) {
928 up_read(&current->mm->mmap_sem);
929 vma = NULL;
930 }
931
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 err = 0;
Miao Xie58568d22009-06-16 15:31:49 -0700933 if (nmask) {
Lee Schermerhornc6b6ef82010-03-23 13:35:41 -0700934 if (mpol_store_user_nodemask(pol)) {
935 *nmask = pol->w.user_nodemask;
936 } else {
937 task_lock(current);
938 get_policy_nodemask(pol, nmask);
939 task_unlock(current);
940 }
Miao Xie58568d22009-06-16 15:31:49 -0700941 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
943 out:
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700944 mpol_cond_put(pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 if (vma)
946 up_read(&current->mm->mmap_sem);
947 return err;
948}
949
Christoph Lameterb20a3502006-03-22 00:09:12 -0800950#ifdef CONFIG_MIGRATION
Christoph Lameter8bccd852005-10-29 18:16:59 -0700951/*
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800952 * page migration
953 */
Christoph Lameterfc301282006-01-18 17:42:29 -0800954static void migrate_page_add(struct page *page, struct list_head *pagelist,
955 unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800956{
957 /*
Christoph Lameterfc301282006-01-18 17:42:29 -0800958 * Avoid migrating a page that is shared with others.
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800959 */
Nick Piggin62695a82008-10-18 20:26:09 -0700960 if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) {
961 if (!isolate_lru_page(page)) {
962 list_add_tail(&page->lru, pagelist);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800963 inc_zone_page_state(page, NR_ISOLATED_ANON +
964 page_is_file_cache(page));
Nick Piggin62695a82008-10-18 20:26:09 -0700965 }
966 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800967}
968
Christoph Lameter742755a2006-06-23 02:03:55 -0700969static struct page *new_node_page(struct page *page, unsigned long node, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700970{
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700971 if (PageHuge(page))
972 return alloc_huge_page_node(page_hstate(compound_head(page)),
973 node);
974 else
Vlastimil Babka96db8002015-09-08 15:03:50 -0700975 return __alloc_pages_node(node, GFP_HIGHUSER_MOVABLE |
David Rientjesb360edb2015-04-14 15:46:52 -0700976 __GFP_THISNODE, 0);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700977}
978
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800979/*
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800980 * Migrate pages from one node to a target node.
981 * Returns error or the number of pages not migrated.
982 */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700983static int migrate_to_node(struct mm_struct *mm, int source, int dest,
984 int flags)
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800985{
986 nodemask_t nmask;
987 LIST_HEAD(pagelist);
988 int err = 0;
989
990 nodes_clear(nmask);
991 node_set(source, nmask);
992
Minchan Kim08270802012-10-08 16:33:38 -0700993 /*
994 * This does not "check" the range but isolates all pages that
995 * need migration. Between passing in the full user address
996 * space range and MPOL_MF_DISCONTIG_OK, this call can not fail.
997 */
998 VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)));
Naoya Horiguchi98094942013-09-11 14:22:14 -0700999 queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask,
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001000 flags | MPOL_MF_DISCONTIG_OK, &pagelist);
1001
Minchan Kimcf608ac2010-10-26 14:21:29 -07001002 if (!list_empty(&pagelist)) {
David Rientjes68711a72014-06-04 16:08:25 -07001003 err = migrate_pages(&pagelist, new_node_page, NULL, dest,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001004 MIGRATE_SYNC, MR_SYSCALL);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001005 if (err)
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -07001006 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001007 }
Christoph Lameter95a402c2006-06-23 02:03:53 -07001008
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001009 return err;
1010}
1011
1012/*
1013 * Move pages between the two nodesets so as to preserve the physical
1014 * layout as much as possible.
Christoph Lameter39743882006-01-08 01:00:51 -08001015 *
1016 * Returns the number of page that could not be moved.
1017 */
Andrew Morton0ce72d42012-05-29 15:06:24 -07001018int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1019 const nodemask_t *to, int flags)
Christoph Lameter39743882006-01-08 01:00:51 -08001020{
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001021 int busy = 0;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001022 int err;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001023 nodemask_t tmp;
Christoph Lameter39743882006-01-08 01:00:51 -08001024
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001025 err = migrate_prep();
1026 if (err)
1027 return err;
1028
Lee Schermerhorn53f25562008-04-28 02:13:20 -07001029 down_read(&mm->mmap_sem);
Christoph Lameter39743882006-01-08 01:00:51 -08001030
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -08001031 /*
1032 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
1033 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
1034 * bit in 'tmp', and return that <source, dest> pair for migration.
1035 * The pair of nodemasks 'to' and 'from' define the map.
1036 *
1037 * If no pair of bits is found that way, fallback to picking some
1038 * pair of 'source' and 'dest' bits that are not the same. If the
1039 * 'source' and 'dest' bits are the same, this represents a node
1040 * that will be migrating to itself, so no pages need move.
1041 *
1042 * If no bits are left in 'tmp', or if all remaining bits left
1043 * in 'tmp' correspond to the same bit in 'to', return false
1044 * (nothing left to migrate).
1045 *
1046 * This lets us pick a pair of nodes to migrate between, such that
1047 * if possible the dest node is not already occupied by some other
1048 * source node, minimizing the risk of overloading the memory on a
1049 * node that would happen if we migrated incoming memory to a node
1050 * before migrating outgoing memory source that same node.
1051 *
1052 * A single scan of tmp is sufficient. As we go, we remember the
1053 * most recent <s, d> pair that moved (s != d). If we find a pair
1054 * that not only moved, but what's better, moved to an empty slot
1055 * (d is not set in tmp), then we break out then, with that pair.
Justin P. Mattockae0e47f2011-03-01 15:06:02 +01001056 * Otherwise when we finish scanning from_tmp, we at least have the
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -08001057 * most recent <s, d> pair that moved. If we get all the way through
1058 * the scan of tmp without finding any node that moved, much less
1059 * moved to an empty node, then there is nothing left worth migrating.
1060 */
Christoph Lameterd4984712006-01-08 01:00:55 -08001061
Andrew Morton0ce72d42012-05-29 15:06:24 -07001062 tmp = *from;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001063 while (!nodes_empty(tmp)) {
1064 int s,d;
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001065 int source = NUMA_NO_NODE;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001066 int dest = 0;
1067
1068 for_each_node_mask(s, tmp) {
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001069
1070 /*
1071 * do_migrate_pages() tries to maintain the relative
1072 * node relationship of the pages established between
1073 * threads and memory areas.
1074 *
1075 * However if the number of source nodes is not equal to
1076 * the number of destination nodes we can not preserve
1077 * this node relative relationship. In that case, skip
1078 * copying memory from a node that is in the destination
1079 * mask.
1080 *
1081 * Example: [2,3,4] -> [3,4,5] moves everything.
1082 * [0-7] - > [3,4,5] moves only 0,1,2,6,7.
1083 */
1084
Andrew Morton0ce72d42012-05-29 15:06:24 -07001085 if ((nodes_weight(*from) != nodes_weight(*to)) &&
1086 (node_isset(s, *to)))
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001087 continue;
1088
Andrew Morton0ce72d42012-05-29 15:06:24 -07001089 d = node_remap(s, *from, *to);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001090 if (s == d)
1091 continue;
1092
1093 source = s; /* Node moved. Memorize */
1094 dest = d;
1095
1096 /* dest not in remaining from nodes? */
1097 if (!node_isset(dest, tmp))
1098 break;
1099 }
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001100 if (source == NUMA_NO_NODE)
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001101 break;
1102
1103 node_clear(source, tmp);
1104 err = migrate_to_node(mm, source, dest, flags);
1105 if (err > 0)
1106 busy += err;
1107 if (err < 0)
1108 break;
Christoph Lameter39743882006-01-08 01:00:51 -08001109 }
1110 up_read(&mm->mmap_sem);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001111 if (err < 0)
1112 return err;
1113 return busy;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001114
Christoph Lameter39743882006-01-08 01:00:51 -08001115}
1116
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001117/*
1118 * Allocate a new page for page migration based on vma policy.
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001119 * Start by assuming the page is mapped by the same vma as contains @start.
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001120 * Search forward from there, if not. N.B., this assumes that the
1121 * list of pages handed to migrate_pages()--which is how we get here--
1122 * is in virtual address order.
1123 */
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001124static struct page *new_page(struct page *page, unsigned long start, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001125{
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001126 struct vm_area_struct *vma;
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001127 unsigned long uninitialized_var(address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001128
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001129 vma = find_vma(current->mm, start);
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001130 while (vma) {
1131 address = page_address_in_vma(page, vma);
1132 if (address != -EFAULT)
1133 break;
1134 vma = vma->vm_next;
1135 }
1136
Wanpeng Li11c731e2013-12-18 17:08:56 -08001137 if (PageHuge(page)) {
Michal Hockocc817172014-01-23 15:53:15 -08001138 BUG_ON(!vma);
1139 return alloc_huge_page_noerr(vma, address, 1);
Wanpeng Li11c731e2013-12-18 17:08:56 -08001140 }
1141 /*
1142 * if !vma, alloc_page_vma() will use task or system default policy
1143 */
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001144 return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001145}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001146#else
1147
1148static void migrate_page_add(struct page *page, struct list_head *pagelist,
1149 unsigned long flags)
1150{
1151}
1152
Andrew Morton0ce72d42012-05-29 15:06:24 -07001153int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1154 const nodemask_t *to, int flags)
Christoph Lameterb20a3502006-03-22 00:09:12 -08001155{
1156 return -ENOSYS;
1157}
Christoph Lameter95a402c2006-06-23 02:03:53 -07001158
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001159static struct page *new_page(struct page *page, unsigned long start, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001160{
1161 return NULL;
1162}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001163#endif
1164
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001165static long do_mbind(unsigned long start, unsigned long len,
David Rientjes028fec42008-04-28 02:12:25 -07001166 unsigned short mode, unsigned short mode_flags,
1167 nodemask_t *nmask, unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001168{
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001169 struct mm_struct *mm = current->mm;
1170 struct mempolicy *new;
1171 unsigned long end;
1172 int err;
1173 LIST_HEAD(pagelist);
1174
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001175 if (flags & ~(unsigned long)MPOL_MF_VALID)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001176 return -EINVAL;
Christoph Lameter74c00242006-03-14 19:50:21 -08001177 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001178 return -EPERM;
1179
1180 if (start & ~PAGE_MASK)
1181 return -EINVAL;
1182
1183 if (mode == MPOL_DEFAULT)
1184 flags &= ~MPOL_MF_STRICT;
1185
1186 len = (len + PAGE_SIZE - 1) & PAGE_MASK;
1187 end = start + len;
1188
1189 if (end < start)
1190 return -EINVAL;
1191 if (end == start)
1192 return 0;
1193
David Rientjes028fec42008-04-28 02:12:25 -07001194 new = mpol_new(mode, mode_flags, nmask);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001195 if (IS_ERR(new))
1196 return PTR_ERR(new);
1197
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001198 if (flags & MPOL_MF_LAZY)
1199 new->flags |= MPOL_F_MOF;
1200
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001201 /*
1202 * If we are using the default policy then operation
1203 * on discontinuous address spaces is okay after all
1204 */
1205 if (!new)
1206 flags |= MPOL_MF_DISCONTIG_OK;
1207
David Rientjes028fec42008-04-28 02:12:25 -07001208 pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
1209 start, start + len, mode, mode_flags,
David Rientjes00ef2d22013-02-22 16:35:36 -08001210 nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001211
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001212 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
1213
1214 err = migrate_prep();
1215 if (err)
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001216 goto mpol_out;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001217 }
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07001218 {
1219 NODEMASK_SCRATCH(scratch);
1220 if (scratch) {
1221 down_write(&mm->mmap_sem);
1222 task_lock(current);
1223 err = mpol_set_nodemask(new, nmask, scratch);
1224 task_unlock(current);
1225 if (err)
1226 up_write(&mm->mmap_sem);
1227 } else
1228 err = -ENOMEM;
1229 NODEMASK_SCRATCH_FREE(scratch);
1230 }
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001231 if (err)
1232 goto mpol_out;
1233
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001234 err = queue_pages_range(mm, start, end, nmask,
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001235 flags | MPOL_MF_INVERT, &pagelist);
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001236 if (!err)
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -08001237 err = mbind_range(mm, start, end, new);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001238
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001239 if (!err) {
1240 int nr_failed = 0;
1241
Minchan Kimcf608ac2010-10-26 14:21:29 -07001242 if (!list_empty(&pagelist)) {
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001243 WARN_ON_ONCE(flags & MPOL_MF_LAZY);
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001244 nr_failed = migrate_pages(&pagelist, new_page, NULL,
1245 start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001246 if (nr_failed)
Naoya Horiguchi74060e42013-09-11 14:22:06 -07001247 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001248 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001249
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001250 if (nr_failed && (flags & MPOL_MF_STRICT))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001251 err = -EIO;
KOSAKI Motohiroab8a3e12009-10-26 16:49:58 -07001252 } else
Joonsoo Kimb0e5fd72013-12-18 17:08:51 -08001253 putback_movable_pages(&pagelist);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001254
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001255 up_write(&mm->mmap_sem);
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001256 mpol_out:
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001257 mpol_put(new);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001258 return err;
1259}
1260
Christoph Lameter39743882006-01-08 01:00:51 -08001261/*
Christoph Lameter8bccd852005-10-29 18:16:59 -07001262 * User space interface with variable sized bitmaps for nodelists.
1263 */
1264
1265/* Copy a node mask from user space. */
Christoph Lameter39743882006-01-08 01:00:51 -08001266static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
Christoph Lameter8bccd852005-10-29 18:16:59 -07001267 unsigned long maxnode)
1268{
1269 unsigned long k;
1270 unsigned long nlongs;
1271 unsigned long endmask;
1272
1273 --maxnode;
1274 nodes_clear(*nodes);
1275 if (maxnode == 0 || !nmask)
1276 return 0;
Andi Kleena9c930b2006-02-20 18:27:59 -08001277 if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
Chris Wright636f13c2006-02-17 13:59:36 -08001278 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001279
1280 nlongs = BITS_TO_LONGS(maxnode);
1281 if ((maxnode % BITS_PER_LONG) == 0)
1282 endmask = ~0UL;
1283 else
1284 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
1285
1286 /* When the user specified more nodes than supported just check
1287 if the non supported part is all zero. */
1288 if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
1289 if (nlongs > PAGE_SIZE/sizeof(long))
1290 return -EINVAL;
1291 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
1292 unsigned long t;
1293 if (get_user(t, nmask + k))
1294 return -EFAULT;
1295 if (k == nlongs - 1) {
1296 if (t & endmask)
1297 return -EINVAL;
1298 } else if (t)
1299 return -EINVAL;
1300 }
1301 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
1302 endmask = ~0UL;
1303 }
1304
1305 if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
1306 return -EFAULT;
1307 nodes_addr(*nodes)[nlongs-1] &= endmask;
1308 return 0;
1309}
1310
1311/* Copy a kernel node mask to user space */
1312static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1313 nodemask_t *nodes)
1314{
1315 unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1316 const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long);
1317
1318 if (copy > nbytes) {
1319 if (copy > PAGE_SIZE)
1320 return -EINVAL;
1321 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1322 return -EFAULT;
1323 copy = nbytes;
1324 }
1325 return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1326}
1327
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001328SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
Rasmus Villemoesf7f28ca2014-06-04 16:07:57 -07001329 unsigned long, mode, const unsigned long __user *, nmask,
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001330 unsigned long, maxnode, unsigned, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001331{
1332 nodemask_t nodes;
1333 int err;
David Rientjes028fec42008-04-28 02:12:25 -07001334 unsigned short mode_flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001335
David Rientjes028fec42008-04-28 02:12:25 -07001336 mode_flags = mode & MPOL_MODE_FLAGS;
1337 mode &= ~MPOL_MODE_FLAGS;
David Rientjesa3b51e02008-04-28 02:12:23 -07001338 if (mode >= MPOL_MAX)
1339 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001340 if ((mode_flags & MPOL_F_STATIC_NODES) &&
1341 (mode_flags & MPOL_F_RELATIVE_NODES))
1342 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001343 err = get_nodes(&nodes, nmask, maxnode);
1344 if (err)
1345 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001346 return do_mbind(start, len, mode, mode_flags, &nodes, flags);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001347}
1348
1349/* Set the process memory policy */
Rasmus Villemoes23c89022014-06-04 16:07:58 -07001350SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask,
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001351 unsigned long, maxnode)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001352{
1353 int err;
1354 nodemask_t nodes;
David Rientjes028fec42008-04-28 02:12:25 -07001355 unsigned short flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001356
David Rientjes028fec42008-04-28 02:12:25 -07001357 flags = mode & MPOL_MODE_FLAGS;
1358 mode &= ~MPOL_MODE_FLAGS;
1359 if ((unsigned int)mode >= MPOL_MAX)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001360 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001361 if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
1362 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001363 err = get_nodes(&nodes, nmask, maxnode);
1364 if (err)
1365 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001366 return do_set_mempolicy(mode, flags, &nodes);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001367}
1368
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001369SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
1370 const unsigned long __user *, old_nodes,
1371 const unsigned long __user *, new_nodes)
Christoph Lameter39743882006-01-08 01:00:51 -08001372{
David Howellsc69e8d92008-11-14 10:39:19 +11001373 const struct cred *cred = current_cred(), *tcred;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001374 struct mm_struct *mm = NULL;
Christoph Lameter39743882006-01-08 01:00:51 -08001375 struct task_struct *task;
Christoph Lameter39743882006-01-08 01:00:51 -08001376 nodemask_t task_nodes;
1377 int err;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001378 nodemask_t *old;
1379 nodemask_t *new;
1380 NODEMASK_SCRATCH(scratch);
Christoph Lameter39743882006-01-08 01:00:51 -08001381
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001382 if (!scratch)
1383 return -ENOMEM;
Christoph Lameter39743882006-01-08 01:00:51 -08001384
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001385 old = &scratch->mask1;
1386 new = &scratch->mask2;
1387
1388 err = get_nodes(old, old_nodes, maxnode);
Christoph Lameter39743882006-01-08 01:00:51 -08001389 if (err)
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001390 goto out;
1391
1392 err = get_nodes(new, new_nodes, maxnode);
1393 if (err)
1394 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001395
1396 /* Find the mm_struct */
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001397 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001398 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter39743882006-01-08 01:00:51 -08001399 if (!task) {
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001400 rcu_read_unlock();
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001401 err = -ESRCH;
1402 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001403 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001404 get_task_struct(task);
Christoph Lameter39743882006-01-08 01:00:51 -08001405
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001406 err = -EINVAL;
Christoph Lameter39743882006-01-08 01:00:51 -08001407
1408 /*
1409 * Check if this process has the right to modify the specified
1410 * process. The right exists if the process has administrative
Alexey Dobriyan7f927fc2006-03-28 01:56:53 -08001411 * capabilities, superuser privileges or the same
Christoph Lameter39743882006-01-08 01:00:51 -08001412 * userid as the target process.
1413 */
David Howellsc69e8d92008-11-14 10:39:19 +11001414 tcred = __task_cred(task);
Eric W. Biedermanb38a86e2012-03-12 15:48:24 -07001415 if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1416 !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) &&
Christoph Lameter74c00242006-03-14 19:50:21 -08001417 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001418 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001419 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001420 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001421 }
David Howellsc69e8d92008-11-14 10:39:19 +11001422 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001423
1424 task_nodes = cpuset_mems_allowed(task);
1425 /* Is the user allowed to access the target nodes? */
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001426 if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
Christoph Lameter39743882006-01-08 01:00:51 -08001427 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001428 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001429 }
1430
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08001431 if (!nodes_subset(*new, node_states[N_MEMORY])) {
Christoph Lameter3b42d282007-08-31 00:12:08 -07001432 err = -EINVAL;
Christoph Lameter3268c632012-03-21 16:34:06 -07001433 goto out_put;
Christoph Lameter3b42d282007-08-31 00:12:08 -07001434 }
1435
David Quigley86c3a762006-06-23 02:04:02 -07001436 err = security_task_movememory(task);
1437 if (err)
Christoph Lameter3268c632012-03-21 16:34:06 -07001438 goto out_put;
David Quigley86c3a762006-06-23 02:04:02 -07001439
Christoph Lameter3268c632012-03-21 16:34:06 -07001440 mm = get_task_mm(task);
1441 put_task_struct(task);
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001442
1443 if (!mm) {
Christoph Lameter3268c632012-03-21 16:34:06 -07001444 err = -EINVAL;
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001445 goto out;
1446 }
1447
1448 err = do_migrate_pages(mm, old, new,
1449 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
Christoph Lameter3268c632012-03-21 16:34:06 -07001450
1451 mmput(mm);
1452out:
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001453 NODEMASK_SCRATCH_FREE(scratch);
1454
Christoph Lameter39743882006-01-08 01:00:51 -08001455 return err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001456
1457out_put:
1458 put_task_struct(task);
1459 goto out;
1460
Christoph Lameter39743882006-01-08 01:00:51 -08001461}
1462
1463
Christoph Lameter8bccd852005-10-29 18:16:59 -07001464/* Retrieve NUMA policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001465SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1466 unsigned long __user *, nmask, unsigned long, maxnode,
1467 unsigned long, addr, unsigned long, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001468{
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001469 int err;
1470 int uninitialized_var(pval);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001471 nodemask_t nodes;
1472
1473 if (nmask != NULL && maxnode < MAX_NUMNODES)
1474 return -EINVAL;
1475
1476 err = do_get_mempolicy(&pval, &nodes, addr, flags);
1477
1478 if (err)
1479 return err;
1480
1481 if (policy && put_user(pval, policy))
1482 return -EFAULT;
1483
1484 if (nmask)
1485 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1486
1487 return err;
1488}
1489
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490#ifdef CONFIG_COMPAT
1491
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001492COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1493 compat_ulong_t __user *, nmask,
1494 compat_ulong_t, maxnode,
1495 compat_ulong_t, addr, compat_ulong_t, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496{
1497 long err;
1498 unsigned long __user *nm = NULL;
1499 unsigned long nr_bits, alloc_size;
1500 DECLARE_BITMAP(bm, MAX_NUMNODES);
1501
1502 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1503 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1504
1505 if (nmask)
1506 nm = compat_alloc_user_space(alloc_size);
1507
1508 err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
1509
1510 if (!err && nmask) {
KAMEZAWA Hiroyuki2bbff6c2011-09-14 16:21:02 -07001511 unsigned long copy_size;
1512 copy_size = min_t(unsigned long, sizeof(bm), alloc_size);
1513 err = copy_from_user(bm, nm, copy_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 /* ensure entire bitmap is zeroed */
1515 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
1516 err |= compat_put_bitmap(nmask, bm, nr_bits);
1517 }
1518
1519 return err;
1520}
1521
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001522COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask,
1523 compat_ulong_t, maxnode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
1525 long err = 0;
1526 unsigned long __user *nm = NULL;
1527 unsigned long nr_bits, alloc_size;
1528 DECLARE_BITMAP(bm, MAX_NUMNODES);
1529
1530 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1531 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1532
1533 if (nmask) {
1534 err = compat_get_bitmap(bm, nmask, nr_bits);
1535 nm = compat_alloc_user_space(alloc_size);
1536 err |= copy_to_user(nm, bm, alloc_size);
1537 }
1538
1539 if (err)
1540 return -EFAULT;
1541
1542 return sys_set_mempolicy(mode, nm, nr_bits+1);
1543}
1544
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001545COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len,
1546 compat_ulong_t, mode, compat_ulong_t __user *, nmask,
1547 compat_ulong_t, maxnode, compat_ulong_t, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548{
1549 long err = 0;
1550 unsigned long __user *nm = NULL;
1551 unsigned long nr_bits, alloc_size;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001552 nodemask_t bm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
1554 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1555 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1556
1557 if (nmask) {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001558 err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 nm = compat_alloc_user_space(alloc_size);
Andi Kleendfcd3c02005-10-29 18:15:48 -07001560 err |= copy_to_user(nm, nodes_addr(bm), alloc_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 }
1562
1563 if (err)
1564 return -EFAULT;
1565
1566 return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
1567}
1568
1569#endif
1570
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001571struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
1572 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573{
Oleg Nesterov8d902742014-10-09 15:27:45 -07001574 struct mempolicy *pol = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
1576 if (vma) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001577 if (vma->vm_ops && vma->vm_ops->get_policy) {
Oleg Nesterov8d902742014-10-09 15:27:45 -07001578 pol = vma->vm_ops->get_policy(vma, addr);
Mel Gorman00442ad2012-10-08 16:29:20 -07001579 } else if (vma->vm_policy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 pol = vma->vm_policy;
Mel Gorman00442ad2012-10-08 16:29:20 -07001581
1582 /*
1583 * shmem_alloc_page() passes MPOL_F_SHARED policy with
1584 * a pseudo vma whose vma->vm_ops=NULL. Take a reference
1585 * count on these policies which will be dropped by
1586 * mpol_cond_put() later
1587 */
1588 if (mpol_needs_cond_ref(pol))
1589 mpol_get(pol);
1590 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 }
Oleg Nesterovf15ca782014-10-09 15:27:43 -07001592
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001593 return pol;
1594}
1595
1596/*
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001597 * get_vma_policy(@vma, @addr)
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001598 * @vma: virtual memory area whose policy is sought
1599 * @addr: address in @vma for shared policy lookup
1600 *
1601 * Returns effective policy for a VMA at specified address.
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001602 * Falls back to current->mempolicy or system default policy, as necessary.
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001603 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1604 * count--added by the get_policy() vm_op, as appropriate--to protect against
1605 * freeing by another task. It is the caller's responsibility to free the
1606 * extra reference for shared policies.
1607 */
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001608static struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
1609 unsigned long addr)
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001610{
1611 struct mempolicy *pol = __get_vma_policy(vma, addr);
1612
Oleg Nesterov8d902742014-10-09 15:27:45 -07001613 if (!pol)
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001614 pol = get_task_policy(current);
Oleg Nesterov8d902742014-10-09 15:27:45 -07001615
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 return pol;
1617}
1618
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001619bool vma_policy_mof(struct vm_area_struct *vma)
Mel Gormanfc3147242013-10-07 11:29:09 +01001620{
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001621 struct mempolicy *pol;
Oleg Nesterovf15ca782014-10-09 15:27:43 -07001622
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001623 if (vma->vm_ops && vma->vm_ops->get_policy) {
1624 bool ret = false;
Mel Gormanfc3147242013-10-07 11:29:09 +01001625
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001626 pol = vma->vm_ops->get_policy(vma, vma->vm_start);
1627 if (pol && (pol->flags & MPOL_F_MOF))
1628 ret = true;
1629 mpol_cond_put(pol);
Mel Gormanfc3147242013-10-07 11:29:09 +01001630
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001631 return ret;
Mel Gormanfc3147242013-10-07 11:29:09 +01001632 }
1633
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001634 pol = vma->vm_policy;
Oleg Nesterov8d902742014-10-09 15:27:45 -07001635 if (!pol)
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001636 pol = get_task_policy(current);
Oleg Nesterov8d902742014-10-09 15:27:45 -07001637
Mel Gormanfc3147242013-10-07 11:29:09 +01001638 return pol->flags & MPOL_F_MOF;
1639}
1640
Lai Jiangshand3eb1572013-02-22 16:33:22 -08001641static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone)
1642{
1643 enum zone_type dynamic_policy_zone = policy_zone;
1644
1645 BUG_ON(dynamic_policy_zone == ZONE_MOVABLE);
1646
1647 /*
1648 * if policy->v.nodes has movable memory only,
1649 * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only.
1650 *
1651 * policy->v.nodes is intersect with node_states[N_MEMORY].
1652 * so if the following test faile, it implies
1653 * policy->v.nodes has movable memory only.
1654 */
1655 if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY]))
1656 dynamic_policy_zone = ZONE_MOVABLE;
1657
1658 return zone >= dynamic_policy_zone;
1659}
1660
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001661/*
1662 * Return a nodemask representing a mempolicy for filtering nodes for
1663 * page allocation
1664 */
1665static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
Mel Gorman19770b32008-04-28 02:12:18 -07001666{
1667 /* Lower zones don't get a nodemask applied for MPOL_BIND */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001668 if (unlikely(policy->mode == MPOL_BIND) &&
Lai Jiangshand3eb1572013-02-22 16:33:22 -08001669 apply_policy_zone(policy, gfp_zone(gfp)) &&
Mel Gorman19770b32008-04-28 02:12:18 -07001670 cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
1671 return &policy->v.nodes;
1672
1673 return NULL;
1674}
1675
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001676/* Return a zonelist indicated by gfp for node representing a mempolicy */
Andi Kleen2f5f9482011-03-04 17:36:29 -08001677static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy,
1678 int nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679{
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001680 switch (policy->mode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001682 if (!(policy->flags & MPOL_F_LOCAL))
1683 nd = policy->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 break;
1685 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07001686 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001687 * Normally, MPOL_BIND allocations are node-local within the
1688 * allowed nodemask. However, if __GFP_THISNODE is set and the
Bob Liu6eb27e12010-05-24 14:32:00 -07001689 * current node isn't part of the mask, we use the zonelist for
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001690 * the first node in the mask instead.
Mel Gorman19770b32008-04-28 02:12:18 -07001691 */
Mel Gorman19770b32008-04-28 02:12:18 -07001692 if (unlikely(gfp & __GFP_THISNODE) &&
1693 unlikely(!node_isset(nd, policy->v.nodes)))
1694 nd = first_node(policy->v.nodes);
1695 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 BUG();
1698 }
Mel Gorman0e884602008-04-28 02:12:14 -07001699 return node_zonelist(nd, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700}
1701
1702/* Do dynamic interleaving for a process */
1703static unsigned interleave_nodes(struct mempolicy *policy)
1704{
1705 unsigned nid, next;
1706 struct task_struct *me = current;
1707
1708 nid = me->il_next;
Andrew Morton0edaf862016-05-19 17:10:58 -07001709 next = next_node_in(nid, policy->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001710 if (next < MAX_NUMNODES)
1711 me->il_next = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 return nid;
1713}
1714
Christoph Lameterdc85da12006-01-18 17:42:36 -08001715/*
1716 * Depending on the memory policy provide a node from which to allocate the
1717 * next slab entry.
1718 */
David Rientjes2a389612014-04-07 15:37:29 -07001719unsigned int mempolicy_slab_node(void)
Christoph Lameterdc85da12006-01-18 17:42:36 -08001720{
Andi Kleene7b691b2012-06-09 02:40:03 -07001721 struct mempolicy *policy;
David Rientjes2a389612014-04-07 15:37:29 -07001722 int node = numa_mem_id();
Andi Kleene7b691b2012-06-09 02:40:03 -07001723
1724 if (in_interrupt())
David Rientjes2a389612014-04-07 15:37:29 -07001725 return node;
Andi Kleene7b691b2012-06-09 02:40:03 -07001726
1727 policy = current->mempolicy;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001728 if (!policy || policy->flags & MPOL_F_LOCAL)
David Rientjes2a389612014-04-07 15:37:29 -07001729 return node;
Christoph Lameter765c4502006-09-27 01:50:08 -07001730
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001731 switch (policy->mode) {
1732 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001733 /*
1734 * handled MPOL_F_LOCAL above
1735 */
1736 return policy->v.preferred_node;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001737
Christoph Lameterdc85da12006-01-18 17:42:36 -08001738 case MPOL_INTERLEAVE:
1739 return interleave_nodes(policy);
1740
Mel Gormandd1a2392008-04-28 02:12:17 -07001741 case MPOL_BIND: {
Mel Gormanc33d6c02016-05-19 17:14:10 -07001742 struct zoneref *z;
1743
Christoph Lameterdc85da12006-01-18 17:42:36 -08001744 /*
1745 * Follow bind policy behavior and start allocation at the
1746 * first node.
1747 */
Mel Gorman19770b32008-04-28 02:12:18 -07001748 struct zonelist *zonelist;
Mel Gorman19770b32008-04-28 02:12:18 -07001749 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
David Rientjes2a389612014-04-07 15:37:29 -07001750 zonelist = &NODE_DATA(node)->node_zonelists[0];
Mel Gormanc33d6c02016-05-19 17:14:10 -07001751 z = first_zones_zonelist(zonelist, highest_zoneidx,
1752 &policy->v.nodes);
1753 return z->zone ? z->zone->node : node;
Mel Gormandd1a2392008-04-28 02:12:17 -07001754 }
Christoph Lameterdc85da12006-01-18 17:42:36 -08001755
Christoph Lameterdc85da12006-01-18 17:42:36 -08001756 default:
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001757 BUG();
Christoph Lameterdc85da12006-01-18 17:42:36 -08001758 }
1759}
1760
Andrew Mortonfee83b32016-05-19 17:11:43 -07001761/*
1762 * Do static interleaving for a VMA with known offset @n. Returns the n'th
1763 * node in pol->v.nodes (starting from n=0), wrapping around if n exceeds the
1764 * number of present nodes.
1765 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766static unsigned offset_il_node(struct mempolicy *pol,
Andrew Mortonfee83b32016-05-19 17:11:43 -07001767 struct vm_area_struct *vma, unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768{
Andi Kleendfcd3c02005-10-29 18:15:48 -07001769 unsigned nnodes = nodes_weight(pol->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001770 unsigned target;
Andrew Mortonfee83b32016-05-19 17:11:43 -07001771 int i;
1772 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773
David Rientjesf5b087b2008-04-28 02:12:27 -07001774 if (!nnodes)
1775 return numa_node_id();
Andrew Mortonfee83b32016-05-19 17:11:43 -07001776 target = (unsigned int)n % nnodes;
1777 nid = first_node(pol->v.nodes);
1778 for (i = 0; i < target; i++)
Andi Kleendfcd3c02005-10-29 18:15:48 -07001779 nid = next_node(nid, pol->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 return nid;
1781}
1782
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001783/* Determine a node number for interleave */
1784static inline unsigned interleave_nid(struct mempolicy *pol,
1785 struct vm_area_struct *vma, unsigned long addr, int shift)
1786{
1787 if (vma) {
1788 unsigned long off;
1789
Nishanth Aravamudan3b98b082006-08-31 21:27:53 -07001790 /*
1791 * for small pages, there is no difference between
1792 * shift and PAGE_SHIFT, so the bit-shift is safe.
1793 * for huge pages, since vm_pgoff is in units of small
1794 * pages, we need to shift off the always 0 bits to get
1795 * a useful offset.
1796 */
1797 BUG_ON(shift < PAGE_SHIFT);
1798 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001799 off += (addr - vma->vm_start) >> shift;
1800 return offset_il_node(pol, vma, off);
1801 } else
1802 return interleave_nodes(pol);
1803}
1804
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001805#ifdef CONFIG_HUGETLBFS
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001806/*
1807 * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
Fabian Frederickb46e14a2014-06-04 16:08:18 -07001808 * @vma: virtual memory area whose policy is sought
1809 * @addr: address in @vma for shared policy lookup and interleave policy
1810 * @gfp_flags: for requested zone
1811 * @mpol: pointer to mempolicy pointer for reference counted mempolicy
1812 * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001813 *
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001814 * Returns a zonelist suitable for a huge page allocation and a pointer
1815 * to the struct mempolicy for conditional unref after allocation.
1816 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1817 * @nodemask for filtering the zonelist.
Miao Xiec0ff7452010-05-24 14:32:08 -07001818 *
Mel Gormand26914d2014-04-03 14:47:24 -07001819 * Must be protected by read_mems_allowed_begin()
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001820 */
Mel Gorman396faf02007-07-17 04:03:13 -07001821struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
Mel Gorman19770b32008-04-28 02:12:18 -07001822 gfp_t gfp_flags, struct mempolicy **mpol,
1823 nodemask_t **nodemask)
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001824{
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001825 struct zonelist *zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001826
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001827 *mpol = get_vma_policy(vma, addr);
Mel Gorman19770b32008-04-28 02:12:18 -07001828 *nodemask = NULL; /* assume !MPOL_BIND */
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001829
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001830 if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
1831 zl = node_zonelist(interleave_nid(*mpol, vma, addr,
Andi Kleena5516432008-07-23 21:27:41 -07001832 huge_page_shift(hstate_vma(vma))), gfp_flags);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001833 } else {
Andi Kleen2f5f9482011-03-04 17:36:29 -08001834 zl = policy_zonelist(gfp_flags, *mpol, numa_node_id());
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001835 if ((*mpol)->mode == MPOL_BIND)
1836 *nodemask = &(*mpol)->v.nodes;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001837 }
1838 return zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001839}
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001840
1841/*
1842 * init_nodemask_of_mempolicy
1843 *
1844 * If the current task's mempolicy is "default" [NULL], return 'false'
1845 * to indicate default policy. Otherwise, extract the policy nodemask
1846 * for 'bind' or 'interleave' policy into the argument nodemask, or
1847 * initialize the argument nodemask to contain the single node for
1848 * 'preferred' or 'local' policy and return 'true' to indicate presence
1849 * of non-default mempolicy.
1850 *
1851 * We don't bother with reference counting the mempolicy [mpol_get/put]
1852 * because the current task is examining it's own mempolicy and a task's
1853 * mempolicy is only ever changed by the task itself.
1854 *
1855 * N.B., it is the caller's responsibility to free a returned nodemask.
1856 */
1857bool init_nodemask_of_mempolicy(nodemask_t *mask)
1858{
1859 struct mempolicy *mempolicy;
1860 int nid;
1861
1862 if (!(mask && current->mempolicy))
1863 return false;
1864
Miao Xiec0ff7452010-05-24 14:32:08 -07001865 task_lock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001866 mempolicy = current->mempolicy;
1867 switch (mempolicy->mode) {
1868 case MPOL_PREFERRED:
1869 if (mempolicy->flags & MPOL_F_LOCAL)
1870 nid = numa_node_id();
1871 else
1872 nid = mempolicy->v.preferred_node;
1873 init_nodemask_of_node(mask, nid);
1874 break;
1875
1876 case MPOL_BIND:
1877 /* Fall through */
1878 case MPOL_INTERLEAVE:
1879 *mask = mempolicy->v.nodes;
1880 break;
1881
1882 default:
1883 BUG();
1884 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001885 task_unlock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001886
1887 return true;
1888}
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001889#endif
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001890
David Rientjes6f48d0eb2010-08-09 17:18:52 -07001891/*
1892 * mempolicy_nodemask_intersects
1893 *
1894 * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default
1895 * policy. Otherwise, check for intersection between mask and the policy
1896 * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local'
1897 * policy, always return true since it may allocate elsewhere on fallback.
1898 *
1899 * Takes task_lock(tsk) to prevent freeing of its mempolicy.
1900 */
1901bool mempolicy_nodemask_intersects(struct task_struct *tsk,
1902 const nodemask_t *mask)
1903{
1904 struct mempolicy *mempolicy;
1905 bool ret = true;
1906
1907 if (!mask)
1908 return ret;
1909 task_lock(tsk);
1910 mempolicy = tsk->mempolicy;
1911 if (!mempolicy)
1912 goto out;
1913
1914 switch (mempolicy->mode) {
1915 case MPOL_PREFERRED:
1916 /*
1917 * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to
1918 * allocate from, they may fallback to other nodes when oom.
1919 * Thus, it's possible for tsk to have allocated memory from
1920 * nodes in mask.
1921 */
1922 break;
1923 case MPOL_BIND:
1924 case MPOL_INTERLEAVE:
1925 ret = nodes_intersects(mempolicy->v.nodes, *mask);
1926 break;
1927 default:
1928 BUG();
1929 }
1930out:
1931 task_unlock(tsk);
1932 return ret;
1933}
1934
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935/* Allocate a page in interleaved policy.
1936 Own path because it needs to do special accounting. */
Andi Kleen662f3a02005-10-29 18:15:49 -07001937static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
1938 unsigned nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939{
1940 struct zonelist *zl;
1941 struct page *page;
1942
Mel Gorman0e884602008-04-28 02:12:14 -07001943 zl = node_zonelist(nid, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 page = __alloc_pages(gfp, order, zl);
Mel Gormandd1a2392008-04-28 02:12:17 -07001945 if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
Christoph Lameterca889e62006-06-30 01:55:44 -07001946 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 return page;
1948}
1949
1950/**
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001951 * alloc_pages_vma - Allocate a page for a VMA.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 *
1953 * @gfp:
1954 * %GFP_USER user allocation.
1955 * %GFP_KERNEL kernel allocations,
1956 * %GFP_HIGHMEM highmem/user allocations,
1957 * %GFP_FS allocation should not call back into a file system.
1958 * %GFP_ATOMIC don't sleep.
1959 *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001960 * @order:Order of the GFP allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 * @vma: Pointer to VMA or NULL if not available.
1962 * @addr: Virtual Address of the allocation. Must be inside the VMA.
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001963 * @node: Which node to prefer for allocation (modulo policy).
1964 * @hugepage: for hugepages try only the preferred node if possible
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 *
1966 * This function allocates a page from the kernel page pool and applies
1967 * a NUMA policy associated with the VMA or the current process.
1968 * When VMA is not NULL caller must hold down_read on the mmap_sem of the
1969 * mm_struct of the VMA to prevent it from going away. Should be used for
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001970 * all allocations for pages that will be mapped into user space. Returns
1971 * NULL when no page can be allocated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 */
1973struct page *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001974alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001975 unsigned long addr, int node, bool hugepage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976{
Mel Gormancc9a6c82012-03-21 16:34:11 -07001977 struct mempolicy *pol;
Miao Xiec0ff7452010-05-24 14:32:08 -07001978 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001979 unsigned int cpuset_mems_cookie;
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001980 struct zonelist *zl;
1981 nodemask_t *nmask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
Mel Gormancc9a6c82012-03-21 16:34:11 -07001983retry_cpuset:
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001984 pol = get_vma_policy(vma, addr);
Mel Gormand26914d2014-04-03 14:47:24 -07001985 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07001986
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001987 if (pol->mode == MPOL_INTERLEAVE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 unsigned nid;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001989
Andi Kleen8eac5632011-02-25 14:44:28 -08001990 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001991 mpol_cond_put(pol);
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001992 page = alloc_page_interleave(gfp, order, nid);
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001993 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 }
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001995
Vlastimil Babka0867a572015-06-24 16:58:48 -07001996 if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) {
1997 int hpage_node = node;
1998
1999 /*
2000 * For hugepage allocation and non-interleave policy which
2001 * allows the current node (or other explicitly preferred
2002 * node) we only try to allocate from the current/preferred
2003 * node and don't fall back to other nodes, as the cost of
2004 * remote accesses would likely offset THP benefits.
2005 *
2006 * If the policy is interleave, or does not allow the current
2007 * node in its nodemask, we allocate the standard way.
2008 */
2009 if (pol->mode == MPOL_PREFERRED &&
2010 !(pol->flags & MPOL_F_LOCAL))
2011 hpage_node = pol->v.preferred_node;
2012
2013 nmask = policy_nodemask(gfp, pol);
2014 if (!nmask || node_isset(hpage_node, *nmask)) {
2015 mpol_cond_put(pol);
Vlastimil Babka96db8002015-09-08 15:03:50 -07002016 page = __alloc_pages_node(hpage_node,
Vlastimil Babka0867a572015-06-24 16:58:48 -07002017 gfp | __GFP_THISNODE, order);
2018 goto out;
2019 }
2020 }
2021
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002022 nmask = policy_nodemask(gfp, pol);
2023 zl = policy_zonelist(gfp, pol, node);
Oleg Nesterov23867402014-10-09 15:27:41 -07002024 mpol_cond_put(pol);
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002025 page = __alloc_pages_nodemask(gfp, order, zl, nmask);
2026out:
Mel Gormand26914d2014-04-03 14:47:24 -07002027 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002028 goto retry_cpuset;
Miao Xiec0ff7452010-05-24 14:32:08 -07002029 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030}
2031
2032/**
2033 * alloc_pages_current - Allocate pages.
2034 *
2035 * @gfp:
2036 * %GFP_USER user allocation,
2037 * %GFP_KERNEL kernel allocation,
2038 * %GFP_HIGHMEM highmem allocation,
2039 * %GFP_FS don't call back into a file system.
2040 * %GFP_ATOMIC don't sleep.
2041 * @order: Power of two of allocation size in pages. 0 is a single page.
2042 *
2043 * Allocate a page from the kernel page pool. When not in
2044 * interrupt context and apply the current process NUMA policy.
2045 * Returns NULL when no page can be allocated.
2046 *
Paul Jacksoncf2a473c2006-01-08 01:01:54 -08002047 * Don't call cpuset_update_task_memory_state() unless
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 * 1) it's ok to take cpuset_sem (can WAIT), and
2049 * 2) allocating for current task (not interrupt).
2050 */
Al Virodd0fc662005-10-07 07:46:04 +01002051struct page *alloc_pages_current(gfp_t gfp, unsigned order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052{
Oleg Nesterov8d902742014-10-09 15:27:45 -07002053 struct mempolicy *pol = &default_policy;
Miao Xiec0ff7452010-05-24 14:32:08 -07002054 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002055 unsigned int cpuset_mems_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056
Oleg Nesterov8d902742014-10-09 15:27:45 -07002057 if (!in_interrupt() && !(gfp & __GFP_THISNODE))
2058 pol = get_task_policy(current);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002059
Mel Gormancc9a6c82012-03-21 16:34:11 -07002060retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07002061 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002062
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002063 /*
2064 * No reference counting needed for current->mempolicy
2065 * nor system default_policy
2066 */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002067 if (pol->mode == MPOL_INTERLEAVE)
Miao Xiec0ff7452010-05-24 14:32:08 -07002068 page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
2069 else
2070 page = __alloc_pages_nodemask(gfp, order,
Andi Kleen5c4b4be2011-03-04 17:36:32 -08002071 policy_zonelist(gfp, pol, numa_node_id()),
2072 policy_nodemask(gfp, pol));
Mel Gormancc9a6c82012-03-21 16:34:11 -07002073
Mel Gormand26914d2014-04-03 14:47:24 -07002074 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002075 goto retry_cpuset;
2076
Miao Xiec0ff7452010-05-24 14:32:08 -07002077 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078}
2079EXPORT_SYMBOL(alloc_pages_current);
2080
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002081int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
2082{
2083 struct mempolicy *pol = mpol_dup(vma_policy(src));
2084
2085 if (IS_ERR(pol))
2086 return PTR_ERR(pol);
2087 dst->vm_policy = pol;
2088 return 0;
2089}
2090
Paul Jackson42253992006-01-08 01:01:59 -08002091/*
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002092 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
Paul Jackson42253992006-01-08 01:01:59 -08002093 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
2094 * with the mems_allowed returned by cpuset_mems_allowed(). This
2095 * keeps mempolicies cpuset relative after its cpuset moves. See
2096 * further kernel/cpuset.c update_nodemask().
Miao Xie708c1bb2010-05-24 14:32:07 -07002097 *
2098 * current's mempolicy may be rebinded by the other task(the task that changes
2099 * cpuset's mems), so we needn't do rebind work for current task.
Paul Jackson42253992006-01-08 01:01:59 -08002100 */
Paul Jackson42253992006-01-08 01:01:59 -08002101
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002102/* Slow path of a mempolicy duplicate */
2103struct mempolicy *__mpol_dup(struct mempolicy *old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104{
2105 struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2106
2107 if (!new)
2108 return ERR_PTR(-ENOMEM);
Miao Xie708c1bb2010-05-24 14:32:07 -07002109
2110 /* task's mempolicy is protected by alloc_lock */
2111 if (old == current->mempolicy) {
2112 task_lock(current);
2113 *new = *old;
2114 task_unlock(current);
2115 } else
2116 *new = *old;
2117
Paul Jackson42253992006-01-08 01:01:59 -08002118 if (current_cpuset_is_being_rebound()) {
2119 nodemask_t mems = cpuset_mems_allowed(current);
Miao Xie708c1bb2010-05-24 14:32:07 -07002120 if (new->flags & MPOL_F_REBINDING)
2121 mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2);
2122 else
2123 mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE);
Paul Jackson42253992006-01-08 01:01:59 -08002124 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 atomic_set(&new->refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 return new;
2127}
2128
2129/* Slow path of a mempolicy comparison */
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002130bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131{
2132 if (!a || !b)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002133 return false;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002134 if (a->mode != b->mode)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002135 return false;
Bob Liu19800502010-05-24 14:32:01 -07002136 if (a->flags != b->flags)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002137 return false;
Bob Liu19800502010-05-24 14:32:01 -07002138 if (mpol_store_user_nodemask(a))
2139 if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask))
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002140 return false;
Bob Liu19800502010-05-24 14:32:01 -07002141
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002142 switch (a->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -07002143 case MPOL_BIND:
2144 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 case MPOL_INTERLEAVE:
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002146 return !!nodes_equal(a->v.nodes, b->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 case MPOL_PREFERRED:
Namhyung Kim75719662011-03-22 16:33:02 -07002148 return a->v.preferred_node == b->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 default:
2150 BUG();
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002151 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 }
2153}
2154
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 * Shared memory backing store policy support.
2157 *
2158 * Remember policies even when nobody has shared memory mapped.
2159 * The policies are kept in Red-Black tree linked from the inode.
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002160 * They are protected by the sp->lock rwlock, which should be held
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 * for any accesses to the tree.
2162 */
2163
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002164/*
2165 * lookup first element intersecting start-end. Caller holds sp->lock for
2166 * reading or for writing
2167 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168static struct sp_node *
2169sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
2170{
2171 struct rb_node *n = sp->root.rb_node;
2172
2173 while (n) {
2174 struct sp_node *p = rb_entry(n, struct sp_node, nd);
2175
2176 if (start >= p->end)
2177 n = n->rb_right;
2178 else if (end <= p->start)
2179 n = n->rb_left;
2180 else
2181 break;
2182 }
2183 if (!n)
2184 return NULL;
2185 for (;;) {
2186 struct sp_node *w = NULL;
2187 struct rb_node *prev = rb_prev(n);
2188 if (!prev)
2189 break;
2190 w = rb_entry(prev, struct sp_node, nd);
2191 if (w->end <= start)
2192 break;
2193 n = prev;
2194 }
2195 return rb_entry(n, struct sp_node, nd);
2196}
2197
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002198/*
2199 * Insert a new shared policy into the list. Caller holds sp->lock for
2200 * writing.
2201 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202static void sp_insert(struct shared_policy *sp, struct sp_node *new)
2203{
2204 struct rb_node **p = &sp->root.rb_node;
2205 struct rb_node *parent = NULL;
2206 struct sp_node *nd;
2207
2208 while (*p) {
2209 parent = *p;
2210 nd = rb_entry(parent, struct sp_node, nd);
2211 if (new->start < nd->start)
2212 p = &(*p)->rb_left;
2213 else if (new->end > nd->end)
2214 p = &(*p)->rb_right;
2215 else
2216 BUG();
2217 }
2218 rb_link_node(&new->nd, parent, p);
2219 rb_insert_color(&new->nd, &sp->root);
Paul Mundt140d5a42007-07-15 23:38:16 -07002220 pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002221 new->policy ? new->policy->mode : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222}
2223
2224/* Find shared policy intersecting idx */
2225struct mempolicy *
2226mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
2227{
2228 struct mempolicy *pol = NULL;
2229 struct sp_node *sn;
2230
2231 if (!sp->root.rb_node)
2232 return NULL;
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002233 read_lock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 sn = sp_lookup(sp, idx, idx+1);
2235 if (sn) {
2236 mpol_get(sn->policy);
2237 pol = sn->policy;
2238 }
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002239 read_unlock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 return pol;
2241}
2242
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002243static void sp_free(struct sp_node *n)
2244{
2245 mpol_put(n->policy);
2246 kmem_cache_free(sn_cache, n);
2247}
2248
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002249/**
2250 * mpol_misplaced - check whether current page node is valid in policy
2251 *
Fabian Frederickb46e14a2014-06-04 16:08:18 -07002252 * @page: page to be checked
2253 * @vma: vm area where page mapped
2254 * @addr: virtual address where page mapped
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002255 *
2256 * Lookup current policy node id for vma,addr and "compare to" page's
2257 * node id.
2258 *
2259 * Returns:
2260 * -1 - not misplaced, page is in the right node
2261 * node - node id where the page should be
2262 *
2263 * Policy determination "mimics" alloc_page_vma().
2264 * Called from fault path where we know the vma and faulting address.
2265 */
2266int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr)
2267{
2268 struct mempolicy *pol;
Mel Gormanc33d6c02016-05-19 17:14:10 -07002269 struct zoneref *z;
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002270 int curnid = page_to_nid(page);
2271 unsigned long pgoff;
Peter Zijlstra90572892013-10-07 11:29:20 +01002272 int thiscpu = raw_smp_processor_id();
2273 int thisnid = cpu_to_node(thiscpu);
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002274 int polnid = -1;
2275 int ret = -1;
2276
2277 BUG_ON(!vma);
2278
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07002279 pol = get_vma_policy(vma, addr);
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002280 if (!(pol->flags & MPOL_F_MOF))
2281 goto out;
2282
2283 switch (pol->mode) {
2284 case MPOL_INTERLEAVE:
2285 BUG_ON(addr >= vma->vm_end);
2286 BUG_ON(addr < vma->vm_start);
2287
2288 pgoff = vma->vm_pgoff;
2289 pgoff += (addr - vma->vm_start) >> PAGE_SHIFT;
2290 polnid = offset_il_node(pol, vma, pgoff);
2291 break;
2292
2293 case MPOL_PREFERRED:
2294 if (pol->flags & MPOL_F_LOCAL)
2295 polnid = numa_node_id();
2296 else
2297 polnid = pol->v.preferred_node;
2298 break;
2299
2300 case MPOL_BIND:
Mel Gormanc33d6c02016-05-19 17:14:10 -07002301
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002302 /*
2303 * allows binding to multiple nodes.
2304 * use current page if in policy nodemask,
2305 * else select nearest allowed node, if any.
2306 * If no allowed nodes, use current [!misplaced].
2307 */
2308 if (node_isset(curnid, pol->v.nodes))
2309 goto out;
Mel Gormanc33d6c02016-05-19 17:14:10 -07002310 z = first_zones_zonelist(
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002311 node_zonelist(numa_node_id(), GFP_HIGHUSER),
2312 gfp_zone(GFP_HIGHUSER),
Mel Gormanc33d6c02016-05-19 17:14:10 -07002313 &pol->v.nodes);
2314 polnid = z->zone->node;
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002315 break;
2316
2317 default:
2318 BUG();
2319 }
Mel Gorman5606e382012-11-02 18:19:13 +00002320
2321 /* Migrate the page towards the node whose CPU is referencing it */
Mel Gormane42c8ff2012-11-12 09:17:07 +00002322 if (pol->flags & MPOL_F_MORON) {
Peter Zijlstra90572892013-10-07 11:29:20 +01002323 polnid = thisnid;
Mel Gorman5606e382012-11-02 18:19:13 +00002324
Rik van Riel10f39042014-01-27 17:03:44 -05002325 if (!should_numa_migrate_memory(current, page, curnid, thiscpu))
Rik van Rielde1c9ce2013-10-07 11:29:39 +01002326 goto out;
Mel Gormane42c8ff2012-11-12 09:17:07 +00002327 }
2328
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002329 if (curnid != polnid)
2330 ret = polnid;
2331out:
2332 mpol_cond_put(pol);
2333
2334 return ret;
2335}
2336
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337static void sp_delete(struct shared_policy *sp, struct sp_node *n)
2338{
Paul Mundt140d5a42007-07-15 23:38:16 -07002339 pr_debug("deleting %lx-l%lx\n", n->start, n->end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 rb_erase(&n->nd, &sp->root);
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002341 sp_free(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342}
2343
Mel Gorman42288fe2012-12-21 23:10:25 +00002344static void sp_node_init(struct sp_node *node, unsigned long start,
2345 unsigned long end, struct mempolicy *pol)
2346{
2347 node->start = start;
2348 node->end = end;
2349 node->policy = pol;
2350}
2351
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07002352static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
2353 struct mempolicy *pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354{
KOSAKI Motohiro869833f22012-10-08 16:29:16 -07002355 struct sp_node *n;
2356 struct mempolicy *newpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357
KOSAKI Motohiro869833f22012-10-08 16:29:16 -07002358 n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 if (!n)
2360 return NULL;
KOSAKI Motohiro869833f22012-10-08 16:29:16 -07002361
2362 newpol = mpol_dup(pol);
2363 if (IS_ERR(newpol)) {
2364 kmem_cache_free(sn_cache, n);
2365 return NULL;
2366 }
2367 newpol->flags |= MPOL_F_SHARED;
Mel Gorman42288fe2012-12-21 23:10:25 +00002368 sp_node_init(n, start, end, newpol);
KOSAKI Motohiro869833f22012-10-08 16:29:16 -07002369
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 return n;
2371}
2372
2373/* Replace a policy range. */
2374static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
2375 unsigned long end, struct sp_node *new)
2376{
Mel Gormanb22d1272012-10-08 16:29:17 -07002377 struct sp_node *n;
Mel Gorman42288fe2012-12-21 23:10:25 +00002378 struct sp_node *n_new = NULL;
2379 struct mempolicy *mpol_new = NULL;
Mel Gormanb22d1272012-10-08 16:29:17 -07002380 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381
Mel Gorman42288fe2012-12-21 23:10:25 +00002382restart:
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002383 write_lock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 n = sp_lookup(sp, start, end);
2385 /* Take care of old policies in the same range. */
2386 while (n && n->start < end) {
2387 struct rb_node *next = rb_next(&n->nd);
2388 if (n->start >= start) {
2389 if (n->end <= end)
2390 sp_delete(sp, n);
2391 else
2392 n->start = end;
2393 } else {
2394 /* Old policy spanning whole new range. */
2395 if (n->end > end) {
Mel Gorman42288fe2012-12-21 23:10:25 +00002396 if (!n_new)
2397 goto alloc_new;
2398
2399 *mpol_new = *n->policy;
2400 atomic_set(&mpol_new->refcnt, 1);
KOSAKI Motohiro78806392013-03-08 12:43:29 -08002401 sp_node_init(n_new, end, n->end, mpol_new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 n->end = start;
Hillf Danton5ca39572013-03-08 12:43:28 -08002403 sp_insert(sp, n_new);
Mel Gorman42288fe2012-12-21 23:10:25 +00002404 n_new = NULL;
2405 mpol_new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 break;
2407 } else
2408 n->end = start;
2409 }
2410 if (!next)
2411 break;
2412 n = rb_entry(next, struct sp_node, nd);
2413 }
2414 if (new)
2415 sp_insert(sp, new);
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002416 write_unlock(&sp->lock);
Mel Gorman42288fe2012-12-21 23:10:25 +00002417 ret = 0;
2418
2419err_out:
2420 if (mpol_new)
2421 mpol_put(mpol_new);
2422 if (n_new)
2423 kmem_cache_free(sn_cache, n_new);
2424
Mel Gormanb22d1272012-10-08 16:29:17 -07002425 return ret;
Mel Gorman42288fe2012-12-21 23:10:25 +00002426
2427alloc_new:
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002428 write_unlock(&sp->lock);
Mel Gorman42288fe2012-12-21 23:10:25 +00002429 ret = -ENOMEM;
2430 n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL);
2431 if (!n_new)
2432 goto err_out;
2433 mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2434 if (!mpol_new)
2435 goto err_out;
2436 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437}
2438
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002439/**
2440 * mpol_shared_policy_init - initialize shared policy for inode
2441 * @sp: pointer to inode shared policy
2442 * @mpol: struct mempolicy to install
2443 *
2444 * Install non-NULL @mpol in inode's shared policy rb-tree.
2445 * On entry, the current task has a reference on a non-NULL @mpol.
2446 * This must be released on exit.
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002447 * This is called at get_inode() calls and we can use GFP_KERNEL.
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002448 */
2449void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
Robin Holt7339ff82006-01-14 13:20:48 -08002450{
Miao Xie58568d22009-06-16 15:31:49 -07002451 int ret;
2452
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002453 sp->root = RB_ROOT; /* empty tree == default mempolicy */
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002454 rwlock_init(&sp->lock);
Robin Holt7339ff82006-01-14 13:20:48 -08002455
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002456 if (mpol) {
2457 struct vm_area_struct pvma;
2458 struct mempolicy *new;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002459 NODEMASK_SCRATCH(scratch);
Robin Holt7339ff82006-01-14 13:20:48 -08002460
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002461 if (!scratch)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002462 goto put_mpol;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002463 /* contextualize the tmpfs mount point mempolicy */
2464 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002465 if (IS_ERR(new))
Dan Carpenter0cae3452010-05-25 23:42:58 -07002466 goto free_scratch; /* no valid nodemask intersection */
Miao Xie58568d22009-06-16 15:31:49 -07002467
2468 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002469 ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
Miao Xie58568d22009-06-16 15:31:49 -07002470 task_unlock(current);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002471 if (ret)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002472 goto put_new;
Robin Holt7339ff82006-01-14 13:20:48 -08002473
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002474 /* Create pseudo-vma that contains just the policy */
2475 memset(&pvma, 0, sizeof(struct vm_area_struct));
2476 pvma.vm_end = TASK_SIZE; /* policy covers entire file */
2477 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002478
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002479put_new:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002480 mpol_put(new); /* drop initial ref */
Dan Carpenter0cae3452010-05-25 23:42:58 -07002481free_scratch:
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002482 NODEMASK_SCRATCH_FREE(scratch);
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002483put_mpol:
2484 mpol_put(mpol); /* drop our incoming ref on sb mpol */
Robin Holt7339ff82006-01-14 13:20:48 -08002485 }
2486}
2487
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488int mpol_set_shared_policy(struct shared_policy *info,
2489 struct vm_area_struct *vma, struct mempolicy *npol)
2490{
2491 int err;
2492 struct sp_node *new = NULL;
2493 unsigned long sz = vma_pages(vma);
2494
David Rientjes028fec42008-04-28 02:12:25 -07002495 pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 vma->vm_pgoff,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002497 sz, npol ? npol->mode : -1,
David Rientjes028fec42008-04-28 02:12:25 -07002498 npol ? npol->flags : -1,
David Rientjes00ef2d22013-02-22 16:35:36 -08002499 npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500
2501 if (npol) {
2502 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
2503 if (!new)
2504 return -ENOMEM;
2505 }
2506 err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
2507 if (err && new)
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002508 sp_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 return err;
2510}
2511
2512/* Free a backing policy store on inode delete. */
2513void mpol_free_shared_policy(struct shared_policy *p)
2514{
2515 struct sp_node *n;
2516 struct rb_node *next;
2517
2518 if (!p->root.rb_node)
2519 return;
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002520 write_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521 next = rb_first(&p->root);
2522 while (next) {
2523 n = rb_entry(next, struct sp_node, nd);
2524 next = rb_next(&n->nd);
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002525 sp_delete(p, n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 }
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002527 write_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528}
2529
Mel Gorman1a687c22012-11-22 11:16:36 +00002530#ifdef CONFIG_NUMA_BALANCING
Mel Gormanc2976632014-01-29 14:05:42 -08002531static int __initdata numabalancing_override;
Mel Gorman1a687c22012-11-22 11:16:36 +00002532
2533static void __init check_numabalancing_enable(void)
2534{
2535 bool numabalancing_default = false;
2536
2537 if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED))
2538 numabalancing_default = true;
2539
Mel Gormanc2976632014-01-29 14:05:42 -08002540 /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */
2541 if (numabalancing_override)
2542 set_numabalancing_state(numabalancing_override == 1);
2543
Mel Gormanb0dc2b92015-05-14 15:17:09 -07002544 if (num_online_nodes() > 1 && !numabalancing_override) {
Joe Perches756a025f02016-03-17 14:19:47 -07002545 pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n",
Mel Gormanc2976632014-01-29 14:05:42 -08002546 numabalancing_default ? "Enabling" : "Disabling");
Mel Gorman1a687c22012-11-22 11:16:36 +00002547 set_numabalancing_state(numabalancing_default);
2548 }
2549}
2550
2551static int __init setup_numabalancing(char *str)
2552{
2553 int ret = 0;
2554 if (!str)
2555 goto out;
Mel Gorman1a687c22012-11-22 11:16:36 +00002556
2557 if (!strcmp(str, "enable")) {
Mel Gormanc2976632014-01-29 14:05:42 -08002558 numabalancing_override = 1;
Mel Gorman1a687c22012-11-22 11:16:36 +00002559 ret = 1;
2560 } else if (!strcmp(str, "disable")) {
Mel Gormanc2976632014-01-29 14:05:42 -08002561 numabalancing_override = -1;
Mel Gorman1a687c22012-11-22 11:16:36 +00002562 ret = 1;
2563 }
2564out:
2565 if (!ret)
Andrew Morton4a404be2014-01-29 14:05:43 -08002566 pr_warn("Unable to parse numa_balancing=\n");
Mel Gorman1a687c22012-11-22 11:16:36 +00002567
2568 return ret;
2569}
2570__setup("numa_balancing=", setup_numabalancing);
2571#else
2572static inline void __init check_numabalancing_enable(void)
2573{
2574}
2575#endif /* CONFIG_NUMA_BALANCING */
2576
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577/* assumes fs == KERNEL_DS */
2578void __init numa_policy_init(void)
2579{
Paul Mundtb71636e2007-07-15 23:38:15 -07002580 nodemask_t interleave_nodes;
2581 unsigned long largest = 0;
2582 int nid, prefer = 0;
2583
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 policy_cache = kmem_cache_create("numa_policy",
2585 sizeof(struct mempolicy),
Paul Mundt20c2df82007-07-20 10:11:58 +09002586 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587
2588 sn_cache = kmem_cache_create("shared_policy_node",
2589 sizeof(struct sp_node),
Paul Mundt20c2df82007-07-20 10:11:58 +09002590 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591
Mel Gorman5606e382012-11-02 18:19:13 +00002592 for_each_node(nid) {
2593 preferred_node_policy[nid] = (struct mempolicy) {
2594 .refcnt = ATOMIC_INIT(1),
2595 .mode = MPOL_PREFERRED,
2596 .flags = MPOL_F_MOF | MPOL_F_MORON,
2597 .v = { .preferred_node = nid, },
2598 };
2599 }
2600
Paul Mundtb71636e2007-07-15 23:38:15 -07002601 /*
2602 * Set interleaving policy for system init. Interleaving is only
2603 * enabled across suitably sized nodes (default is >= 16MB), or
2604 * fall back to the largest node if they're all smaller.
2605 */
2606 nodes_clear(interleave_nodes);
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002607 for_each_node_state(nid, N_MEMORY) {
Paul Mundtb71636e2007-07-15 23:38:15 -07002608 unsigned long total_pages = node_present_pages(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609
Paul Mundtb71636e2007-07-15 23:38:15 -07002610 /* Preserve the largest node */
2611 if (largest < total_pages) {
2612 largest = total_pages;
2613 prefer = nid;
2614 }
2615
2616 /* Interleave this node? */
2617 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
2618 node_set(nid, interleave_nodes);
2619 }
2620
2621 /* All too small, use the largest */
2622 if (unlikely(nodes_empty(interleave_nodes)))
2623 node_set(prefer, interleave_nodes);
2624
David Rientjes028fec42008-04-28 02:12:25 -07002625 if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07002626 pr_err("%s: interleaving failed\n", __func__);
Mel Gorman1a687c22012-11-22 11:16:36 +00002627
2628 check_numabalancing_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629}
2630
Christoph Lameter8bccd852005-10-29 18:16:59 -07002631/* Reset policy of current process to default */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632void numa_default_policy(void)
2633{
David Rientjes028fec42008-04-28 02:12:25 -07002634 do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635}
Paul Jackson68860ec2005-10-30 15:02:36 -08002636
Paul Jackson42253992006-01-08 01:01:59 -08002637/*
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002638 * Parse and format mempolicy from/to strings
2639 */
2640
2641/*
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002642 * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag.
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002643 */
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002644static const char * const policy_modes[] =
2645{
2646 [MPOL_DEFAULT] = "default",
2647 [MPOL_PREFERRED] = "prefer",
2648 [MPOL_BIND] = "bind",
2649 [MPOL_INTERLEAVE] = "interleave",
Lee Schermerhornd3a71032012-10-25 14:16:29 +02002650 [MPOL_LOCAL] = "local",
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002651};
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002652
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002653
2654#ifdef CONFIG_TMPFS
2655/**
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002656 * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option.
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002657 * @str: string containing mempolicy to parse
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002658 * @mpol: pointer to struct mempolicy pointer, returned on success.
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002659 *
2660 * Format of input:
2661 * <mode>[=<flags>][:<nodelist>]
2662 *
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002663 * On success, returns 0, else 1
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002664 */
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08002665int mpol_parse_str(char *str, struct mempolicy **mpol)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002666{
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002667 struct mempolicy *new = NULL;
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002668 unsigned short mode;
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002669 unsigned short mode_flags;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002670 nodemask_t nodes;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002671 char *nodelist = strchr(str, ':');
2672 char *flags = strchr(str, '=');
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002673 int err = 1;
2674
2675 if (nodelist) {
2676 /* NUL-terminate mode or flags string */
2677 *nodelist++ = '\0';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002678 if (nodelist_parse(nodelist, nodes))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002679 goto out;
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002680 if (!nodes_subset(nodes, node_states[N_MEMORY]))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002681 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002682 } else
2683 nodes_clear(nodes);
2684
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002685 if (flags)
2686 *flags++ = '\0'; /* terminate mode string */
2687
Peter Zijlstra479e2802012-10-25 14:16:28 +02002688 for (mode = 0; mode < MPOL_MAX; mode++) {
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002689 if (!strcmp(str, policy_modes[mode])) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002690 break;
2691 }
2692 }
Mel Gormana7200942012-11-16 09:37:58 +00002693 if (mode >= MPOL_MAX)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002694 goto out;
2695
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002696 switch (mode) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002697 case MPOL_PREFERRED:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002698 /*
2699 * Insist on a nodelist of one node only
2700 */
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002701 if (nodelist) {
2702 char *rest = nodelist;
2703 while (isdigit(*rest))
2704 rest++;
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002705 if (*rest)
2706 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002707 }
2708 break;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002709 case MPOL_INTERLEAVE:
2710 /*
2711 * Default to online nodes with memory if no nodelist
2712 */
2713 if (!nodelist)
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002714 nodes = node_states[N_MEMORY];
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002715 break;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002716 case MPOL_LOCAL:
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002717 /*
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002718 * Don't allow a nodelist; mpol_new() checks flags
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002719 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002720 if (nodelist)
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002721 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002722 mode = MPOL_PREFERRED;
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002723 break;
Ravikiran G Thirumalai413b43d2010-03-23 13:35:28 -07002724 case MPOL_DEFAULT:
2725 /*
2726 * Insist on a empty nodelist
2727 */
2728 if (!nodelist)
2729 err = 0;
2730 goto out;
KOSAKI Motohirod69b2e62010-03-23 13:35:30 -07002731 case MPOL_BIND:
2732 /*
2733 * Insist on a nodelist
2734 */
2735 if (!nodelist)
2736 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002737 }
2738
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002739 mode_flags = 0;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002740 if (flags) {
2741 /*
2742 * Currently, we only support two mutually exclusive
2743 * mode flags.
2744 */
2745 if (!strcmp(flags, "static"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002746 mode_flags |= MPOL_F_STATIC_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002747 else if (!strcmp(flags, "relative"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002748 mode_flags |= MPOL_F_RELATIVE_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002749 else
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002750 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002751 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002752
2753 new = mpol_new(mode, mode_flags, &nodes);
2754 if (IS_ERR(new))
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002755 goto out;
2756
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002757 /*
2758 * Save nodes for mpol_to_str() to show the tmpfs mount options
2759 * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo.
2760 */
2761 if (mode != MPOL_PREFERRED)
2762 new->v.nodes = nodes;
2763 else if (nodelist)
2764 new->v.preferred_node = first_node(nodes);
2765 else
2766 new->flags |= MPOL_F_LOCAL;
2767
2768 /*
2769 * Save nodes for contextualization: this will be used to "clone"
2770 * the mempolicy in a specific context [cpuset] at a later time.
2771 */
2772 new->w.user_nodemask = nodes;
2773
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002774 err = 0;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002775
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002776out:
2777 /* Restore string for error message */
2778 if (nodelist)
2779 *--nodelist = ':';
2780 if (flags)
2781 *--flags = '=';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002782 if (!err)
2783 *mpol = new;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002784 return err;
2785}
2786#endif /* CONFIG_TMPFS */
2787
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002788/**
2789 * mpol_to_str - format a mempolicy structure for printing
2790 * @buffer: to contain formatted mempolicy string
2791 * @maxlen: length of @buffer
2792 * @pol: pointer to mempolicy to be formatted
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002793 *
David Rientjes948927e2013-11-12 15:07:28 -08002794 * Convert @pol into a string. If @buffer is too short, truncate the string.
2795 * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the
2796 * longest flag, "relative", and to display at least a few node ids.
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002797 */
David Rientjes948927e2013-11-12 15:07:28 -08002798void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002799{
2800 char *p = buffer;
David Rientjes948927e2013-11-12 15:07:28 -08002801 nodemask_t nodes = NODE_MASK_NONE;
2802 unsigned short mode = MPOL_DEFAULT;
2803 unsigned short flags = 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002804
David Rientjes8790c712014-01-30 15:46:08 -08002805 if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002806 mode = pol->mode;
David Rientjes948927e2013-11-12 15:07:28 -08002807 flags = pol->flags;
2808 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002809
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002810 switch (mode) {
2811 case MPOL_DEFAULT:
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002812 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002813 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002814 if (flags & MPOL_F_LOCAL)
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002815 mode = MPOL_LOCAL;
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002816 else
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002817 node_set(pol->v.preferred_node, nodes);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002818 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002819 case MPOL_BIND:
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002820 case MPOL_INTERLEAVE:
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002821 nodes = pol->v.nodes;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002822 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002823 default:
David Rientjes948927e2013-11-12 15:07:28 -08002824 WARN_ON_ONCE(1);
2825 snprintf(p, maxlen, "unknown");
2826 return;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002827 }
2828
David Rientjesb7a9f422013-11-21 14:32:06 -08002829 p += snprintf(p, maxlen, "%s", policy_modes[mode]);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002830
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002831 if (flags & MPOL_MODE_FLAGS) {
David Rientjes948927e2013-11-12 15:07:28 -08002832 p += snprintf(p, buffer + maxlen - p, "=");
David Rientjesf5b087b2008-04-28 02:12:27 -07002833
Lee Schermerhorn22919902008-04-28 02:13:22 -07002834 /*
2835 * Currently, the only defined flags are mutually exclusive
2836 */
David Rientjesf5b087b2008-04-28 02:12:27 -07002837 if (flags & MPOL_F_STATIC_NODES)
Lee Schermerhorn22919902008-04-28 02:13:22 -07002838 p += snprintf(p, buffer + maxlen - p, "static");
2839 else if (flags & MPOL_F_RELATIVE_NODES)
2840 p += snprintf(p, buffer + maxlen - p, "relative");
David Rientjesf5b087b2008-04-28 02:12:27 -07002841 }
2842
Tejun Heo9e763e02015-02-13 14:38:02 -08002843 if (!nodes_empty(nodes))
2844 p += scnprintf(p, buffer + maxlen - p, ":%*pbl",
2845 nodemask_pr_args(&nodes));
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002846}