blob: d8c4e38fb5f4be1d9748dc77f214c8a285374f8d [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
Naoya Horiguchi337d9ab2016-07-26 15:24:03 -0700515 if (pmd_trans_unstable(pmd))
516 return 0;
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800517retry:
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800518 pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
519 for (; addr != end; pte++, addr += PAGE_SIZE) {
Hugh Dickins91612e02005-06-21 17:15:07 -0700520 if (!pte_present(*pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800522 page = vm_normal_page(vma, addr, *pte);
523 if (!page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 continue;
Nick Piggin053837f2006-01-18 17:42:27 -0800525 /*
Hugh Dickins62b61f62009-12-14 17:59:33 -0800526 * vm_normal_page() filters out zero pages, but there might
527 * still be PageReserved pages to skip, perhaps in a VDSO.
Nick Piggin053837f2006-01-18 17:42:27 -0800528 */
Hugh Dickinsb79bc0a2013-02-22 16:35:13 -0800529 if (PageReserved(page))
Christoph Lameterf4598c82006-01-12 01:05:20 -0800530 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800531 nid = page_to_nid(page);
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800532 if (node_isset(nid, *qp->nmask) == !!(flags & MPOL_MF_INVERT))
Christoph Lameter38e35862006-01-08 01:01:01 -0800533 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700534 if (PageTransCompound(page)) {
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800535 get_page(page);
536 pte_unmap_unlock(pte, ptl);
537 lock_page(page);
538 ret = split_huge_page(page);
539 unlock_page(page);
540 put_page(page);
541 /* Failed to split -- skip. */
542 if (ret) {
543 pte = pte_offset_map_lock(walk->mm, pmd,
544 addr, &ptl);
545 continue;
546 }
547 goto retry;
548 }
Christoph Lameter38e35862006-01-08 01:01:01 -0800549
Kirill A. Shutemov77bf45e2016-02-05 15:36:33 -0800550 migrate_page_add(page, qp->pagelist, flags);
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800551 }
552 pte_unmap_unlock(pte - 1, ptl);
553 cond_resched();
554 return 0;
Hugh Dickins91612e02005-06-21 17:15:07 -0700555}
556
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800557static int queue_pages_hugetlb(pte_t *pte, unsigned long hmask,
558 unsigned long addr, unsigned long end,
559 struct mm_walk *walk)
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700560{
561#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800562 struct queue_pages *qp = walk->private;
563 unsigned long flags = qp->flags;
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700564 int nid;
565 struct page *page;
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800566 spinlock_t *ptl;
Naoya Horiguchid4c54912014-06-06 10:00:01 -0400567 pte_t entry;
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700568
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800569 ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte);
570 entry = huge_ptep_get(pte);
Naoya Horiguchid4c54912014-06-06 10:00:01 -0400571 if (!pte_present(entry))
572 goto unlock;
573 page = pte_page(entry);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700574 nid = page_to_nid(page);
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800575 if (node_isset(nid, *qp->nmask) == !!(flags & MPOL_MF_INVERT))
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700576 goto unlock;
577 /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */
578 if (flags & (MPOL_MF_MOVE_ALL) ||
579 (flags & MPOL_MF_MOVE && page_mapcount(page) == 1))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800580 isolate_huge_page(page, qp->pagelist);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700581unlock:
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800582 spin_unlock(ptl);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700583#else
584 BUG();
585#endif
Hugh Dickins91612e02005-06-21 17:15:07 -0700586 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587}
588
Aneesh Kumar K.V58772312013-12-06 00:08:22 +0530589#ifdef CONFIG_NUMA_BALANCING
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200590/*
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200591 * This is used to mark a range of virtual addresses to be inaccessible.
592 * These are later cleared by a NUMA hinting fault. Depending on these
593 * faults, pages may be migrated for better NUMA placement.
594 *
595 * This is assuming that NUMA faults are handled using PROT_NONE. If
596 * an architecture makes a different choice, it will need further
597 * changes to the core.
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200598 */
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200599unsigned long change_prot_numa(struct vm_area_struct *vma,
600 unsigned long addr, unsigned long end)
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200601{
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200602 int nr_updated;
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200603
Mel Gorman4d942462015-02-12 14:58:28 -0800604 nr_updated = change_protection(vma, addr, end, PAGE_NONE, 0, 1);
Mel Gorman03c5a6e2012-11-02 14:52:48 +0000605 if (nr_updated)
606 count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated);
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200607
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200608 return nr_updated;
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200609}
610#else
611static unsigned long change_prot_numa(struct vm_area_struct *vma,
612 unsigned long addr, unsigned long end)
613{
614 return 0;
615}
Aneesh Kumar K.V58772312013-12-06 00:08:22 +0530616#endif /* CONFIG_NUMA_BALANCING */
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200617
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800618static int queue_pages_test_walk(unsigned long start, unsigned long end,
619 struct mm_walk *walk)
620{
621 struct vm_area_struct *vma = walk->vma;
622 struct queue_pages *qp = walk->private;
623 unsigned long endvma = vma->vm_end;
624 unsigned long flags = qp->flags;
625
Kirill A. Shutemov77bf45e2016-02-05 15:36:33 -0800626 if (!vma_migratable(vma))
Naoya Horiguchi48684a62015-02-11 15:28:06 -0800627 return 1;
628
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800629 if (endvma > end)
630 endvma = end;
631 if (vma->vm_start > start)
632 start = vma->vm_start;
633
634 if (!(flags & MPOL_MF_DISCONTIG_OK)) {
635 if (!vma->vm_next && vma->vm_end < end)
636 return -EFAULT;
637 if (qp->prev && qp->prev->vm_end < vma->vm_start)
638 return -EFAULT;
639 }
640
641 qp->prev = vma;
642
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800643 if (flags & MPOL_MF_LAZY) {
644 /* Similar to task_numa_work, skip inaccessible VMAs */
Liang Chen4355c012016-03-15 14:56:42 -0700645 if (!is_vm_hugetlb_page(vma) &&
646 (vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)) &&
647 !(vma->vm_flags & VM_MIXEDMAP))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800648 change_prot_numa(vma, start, endvma);
649 return 1;
650 }
651
Kirill A. Shutemov77bf45e2016-02-05 15:36:33 -0800652 /* queue pages from current vma */
653 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800654 return 0;
655 return 1;
656}
657
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800658/*
Naoya Horiguchi98094942013-09-11 14:22:14 -0700659 * Walk through page tables and collect pages to be migrated.
660 *
661 * If pages found in a given range are on a set of nodes (determined by
662 * @nodes and @flags,) it's isolated and queued to the pagelist which is
663 * passed via @private.)
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800664 */
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -0700665static int
Naoya Horiguchi98094942013-09-11 14:22:14 -0700666queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800667 nodemask_t *nodes, unsigned long flags,
668 struct list_head *pagelist)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800670 struct queue_pages qp = {
671 .pagelist = pagelist,
672 .flags = flags,
673 .nmask = nodes,
674 .prev = NULL,
675 };
676 struct mm_walk queue_pages_walk = {
677 .hugetlb_entry = queue_pages_hugetlb,
678 .pmd_entry = queue_pages_pte_range,
679 .test_walk = queue_pages_test_walk,
680 .mm = mm,
681 .private = &qp,
682 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800684 return walk_page_range(start, end, &queue_pages_walk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685}
686
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700687/*
688 * Apply policy to a single VMA
689 * This must be called with the mmap_sem held for writing.
690 */
691static int vma_replace_policy(struct vm_area_struct *vma,
692 struct mempolicy *pol)
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700693{
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700694 int err;
695 struct mempolicy *old;
696 struct mempolicy *new;
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700697
698 pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
699 vma->vm_start, vma->vm_end, vma->vm_pgoff,
700 vma->vm_ops, vma->vm_file,
701 vma->vm_ops ? vma->vm_ops->set_policy : NULL);
702
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700703 new = mpol_dup(pol);
704 if (IS_ERR(new))
705 return PTR_ERR(new);
706
707 if (vma->vm_ops && vma->vm_ops->set_policy) {
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700708 err = vma->vm_ops->set_policy(vma, new);
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700709 if (err)
710 goto err_out;
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700711 }
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700712
713 old = vma->vm_policy;
714 vma->vm_policy = new; /* protected by mmap_sem */
715 mpol_put(old);
716
717 return 0;
718 err_out:
719 mpol_put(new);
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700720 return err;
721}
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723/* Step 2: apply policy to a range and do splits. */
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800724static int mbind_range(struct mm_struct *mm, unsigned long start,
725 unsigned long end, struct mempolicy *new_pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
727 struct vm_area_struct *next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800728 struct vm_area_struct *prev;
729 struct vm_area_struct *vma;
730 int err = 0;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800731 pgoff_t pgoff;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800732 unsigned long vmstart;
733 unsigned long vmend;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
Linus Torvalds097d5912012-03-06 18:23:36 -0800735 vma = find_vma(mm, start);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800736 if (!vma || vma->vm_start > start)
737 return -EFAULT;
738
Linus Torvalds097d5912012-03-06 18:23:36 -0800739 prev = vma->vm_prev;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800740 if (start > vma->vm_start)
741 prev = vma;
742
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800743 for (; vma && vma->vm_start < end; prev = vma, vma = next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 next = vma->vm_next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800745 vmstart = max(start, vma->vm_start);
746 vmend = min(end, vma->vm_end);
747
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800748 if (mpol_equal(vma_policy(vma), new_pol))
749 continue;
750
751 pgoff = vma->vm_pgoff +
752 ((vmstart - vma->vm_start) >> PAGE_SHIFT);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800753 prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700754 vma->anon_vma, vma->vm_file, pgoff,
755 new_pol, vma->vm_userfaultfd_ctx);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800756 if (prev) {
757 vma = prev;
758 next = vma->vm_next;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700759 if (mpol_equal(vma_policy(vma), new_pol))
760 continue;
761 /* vma_merge() joined vma && vma->next, case 8 */
762 goto replace;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800763 }
764 if (vma->vm_start != vmstart) {
765 err = split_vma(vma->vm_mm, vma, vmstart, 1);
766 if (err)
767 goto out;
768 }
769 if (vma->vm_end != vmend) {
770 err = split_vma(vma->vm_mm, vma, vmend, 0);
771 if (err)
772 goto out;
773 }
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700774 replace:
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700775 err = vma_replace_policy(vma, new_pol);
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700776 if (err)
777 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 }
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800779
780 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 return err;
782}
783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784/* Set the process memory policy */
David Rientjes028fec42008-04-28 02:12:25 -0700785static long do_set_mempolicy(unsigned short mode, unsigned short flags,
786 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787{
Miao Xie58568d22009-06-16 15:31:49 -0700788 struct mempolicy *new, *old;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700789 NODEMASK_SCRATCH(scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700790 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700792 if (!scratch)
793 return -ENOMEM;
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700794
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700795 new = mpol_new(mode, flags, nodes);
796 if (IS_ERR(new)) {
797 ret = PTR_ERR(new);
798 goto out;
799 }
Oleg Nesterov2c7c3a72014-10-09 15:27:55 -0700800
Miao Xie58568d22009-06-16 15:31:49 -0700801 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700802 ret = mpol_set_nodemask(new, nodes, scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700803 if (ret) {
804 task_unlock(current);
Miao Xie58568d22009-06-16 15:31:49 -0700805 mpol_put(new);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700806 goto out;
Miao Xie58568d22009-06-16 15:31:49 -0700807 }
808 old = current->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 current->mempolicy = new;
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700810 if (new && new->mode == MPOL_INTERLEAVE &&
David Rientjesf5b087b2008-04-28 02:12:27 -0700811 nodes_weight(new->v.nodes))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700812 current->il_next = first_node(new->v.nodes);
Miao Xie58568d22009-06-16 15:31:49 -0700813 task_unlock(current);
Miao Xie58568d22009-06-16 15:31:49 -0700814 mpol_put(old);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700815 ret = 0;
816out:
817 NODEMASK_SCRATCH_FREE(scratch);
818 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819}
820
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700821/*
822 * Return nodemask for policy for get_mempolicy() query
Miao Xie58568d22009-06-16 15:31:49 -0700823 *
824 * Called with task's alloc_lock held
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700825 */
826static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827{
Andi Kleendfcd3c02005-10-29 18:15:48 -0700828 nodes_clear(*nodes);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700829 if (p == &default_policy)
830 return;
831
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700832 switch (p->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -0700833 case MPOL_BIND:
834 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -0700836 *nodes = p->v.nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 break;
838 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700839 if (!(p->flags & MPOL_F_LOCAL))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700840 node_set(p->v.preferred_node, *nodes);
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700841 /* else return empty node mask for local allocation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 break;
843 default:
844 BUG();
845 }
846}
847
Dave Hansend4edcf02016-02-12 13:01:56 -0800848static int lookup_node(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849{
850 struct page *p;
851 int err;
852
Dave Hansend4edcf02016-02-12 13:01:56 -0800853 err = get_user_pages(addr & PAGE_MASK, 1, 0, 0, &p, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 if (err >= 0) {
855 err = page_to_nid(p);
856 put_page(p);
857 }
858 return err;
859}
860
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861/* Retrieve NUMA policy */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700862static long do_get_mempolicy(int *policy, nodemask_t *nmask,
863 unsigned long addr, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864{
Christoph Lameter8bccd852005-10-29 18:16:59 -0700865 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 struct mm_struct *mm = current->mm;
867 struct vm_area_struct *vma = NULL;
868 struct mempolicy *pol = current->mempolicy;
869
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700870 if (flags &
871 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 return -EINVAL;
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700873
874 if (flags & MPOL_F_MEMS_ALLOWED) {
875 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
876 return -EINVAL;
877 *policy = 0; /* just so it's initialized */
Miao Xie58568d22009-06-16 15:31:49 -0700878 task_lock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700879 *nmask = cpuset_current_mems_allowed;
Miao Xie58568d22009-06-16 15:31:49 -0700880 task_unlock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700881 return 0;
882 }
883
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 if (flags & MPOL_F_ADDR) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700885 /*
886 * Do NOT fall back to task policy if the
887 * vma/shared policy at addr is NULL. We
888 * want to return MPOL_DEFAULT in this case.
889 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 down_read(&mm->mmap_sem);
891 vma = find_vma_intersection(mm, addr, addr+1);
892 if (!vma) {
893 up_read(&mm->mmap_sem);
894 return -EFAULT;
895 }
896 if (vma->vm_ops && vma->vm_ops->get_policy)
897 pol = vma->vm_ops->get_policy(vma, addr);
898 else
899 pol = vma->vm_policy;
900 } else if (addr)
901 return -EINVAL;
902
903 if (!pol)
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700904 pol = &default_policy; /* indicates default behavior */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
906 if (flags & MPOL_F_NODE) {
907 if (flags & MPOL_F_ADDR) {
Dave Hansend4edcf02016-02-12 13:01:56 -0800908 err = lookup_node(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 if (err < 0)
910 goto out;
Christoph Lameter8bccd852005-10-29 18:16:59 -0700911 *policy = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 } else if (pol == current->mempolicy &&
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700913 pol->mode == MPOL_INTERLEAVE) {
Christoph Lameter8bccd852005-10-29 18:16:59 -0700914 *policy = current->il_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 } else {
916 err = -EINVAL;
917 goto out;
918 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700919 } else {
920 *policy = pol == &default_policy ? MPOL_DEFAULT :
921 pol->mode;
David Rientjesd79df632008-07-04 12:24:13 -0700922 /*
923 * Internal mempolicy flags must be masked off before exposing
924 * the policy to userspace.
925 */
926 *policy |= (pol->flags & MPOL_MODE_FLAGS);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700927 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
929 if (vma) {
930 up_read(&current->mm->mmap_sem);
931 vma = NULL;
932 }
933
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 err = 0;
Miao Xie58568d22009-06-16 15:31:49 -0700935 if (nmask) {
Lee Schermerhornc6b6ef82010-03-23 13:35:41 -0700936 if (mpol_store_user_nodemask(pol)) {
937 *nmask = pol->w.user_nodemask;
938 } else {
939 task_lock(current);
940 get_policy_nodemask(pol, nmask);
941 task_unlock(current);
942 }
Miao Xie58568d22009-06-16 15:31:49 -0700943 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
945 out:
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700946 mpol_cond_put(pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 if (vma)
948 up_read(&current->mm->mmap_sem);
949 return err;
950}
951
Christoph Lameterb20a3502006-03-22 00:09:12 -0800952#ifdef CONFIG_MIGRATION
Christoph Lameter8bccd852005-10-29 18:16:59 -0700953/*
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800954 * page migration
955 */
Christoph Lameterfc301282006-01-18 17:42:29 -0800956static void migrate_page_add(struct page *page, struct list_head *pagelist,
957 unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800958{
959 /*
Christoph Lameterfc301282006-01-18 17:42:29 -0800960 * Avoid migrating a page that is shared with others.
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800961 */
Nick Piggin62695a82008-10-18 20:26:09 -0700962 if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) {
963 if (!isolate_lru_page(page)) {
964 list_add_tail(&page->lru, pagelist);
Mel Gorman599d0c92016-07-28 15:45:31 -0700965 inc_node_page_state(page, NR_ISOLATED_ANON +
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800966 page_is_file_cache(page));
Nick Piggin62695a82008-10-18 20:26:09 -0700967 }
968 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800969}
970
Christoph Lameter742755a2006-06-23 02:03:55 -0700971static struct page *new_node_page(struct page *page, unsigned long node, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700972{
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700973 if (PageHuge(page))
974 return alloc_huge_page_node(page_hstate(compound_head(page)),
975 node);
976 else
Vlastimil Babka96db8002015-09-08 15:03:50 -0700977 return __alloc_pages_node(node, GFP_HIGHUSER_MOVABLE |
David Rientjesb360edb2015-04-14 15:46:52 -0700978 __GFP_THISNODE, 0);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700979}
980
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800981/*
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800982 * Migrate pages from one node to a target node.
983 * Returns error or the number of pages not migrated.
984 */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700985static int migrate_to_node(struct mm_struct *mm, int source, int dest,
986 int flags)
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800987{
988 nodemask_t nmask;
989 LIST_HEAD(pagelist);
990 int err = 0;
991
992 nodes_clear(nmask);
993 node_set(source, nmask);
994
Minchan Kim08270802012-10-08 16:33:38 -0700995 /*
996 * This does not "check" the range but isolates all pages that
997 * need migration. Between passing in the full user address
998 * space range and MPOL_MF_DISCONTIG_OK, this call can not fail.
999 */
1000 VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)));
Naoya Horiguchi98094942013-09-11 14:22:14 -07001001 queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask,
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001002 flags | MPOL_MF_DISCONTIG_OK, &pagelist);
1003
Minchan Kimcf608ac2010-10-26 14:21:29 -07001004 if (!list_empty(&pagelist)) {
David Rientjes68711a72014-06-04 16:08:25 -07001005 err = migrate_pages(&pagelist, new_node_page, NULL, dest,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001006 MIGRATE_SYNC, MR_SYSCALL);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001007 if (err)
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -07001008 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001009 }
Christoph Lameter95a402c2006-06-23 02:03:53 -07001010
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001011 return err;
1012}
1013
1014/*
1015 * Move pages between the two nodesets so as to preserve the physical
1016 * layout as much as possible.
Christoph Lameter39743882006-01-08 01:00:51 -08001017 *
1018 * Returns the number of page that could not be moved.
1019 */
Andrew Morton0ce72d42012-05-29 15:06:24 -07001020int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1021 const nodemask_t *to, int flags)
Christoph Lameter39743882006-01-08 01:00:51 -08001022{
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001023 int busy = 0;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001024 int err;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001025 nodemask_t tmp;
Christoph Lameter39743882006-01-08 01:00:51 -08001026
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001027 err = migrate_prep();
1028 if (err)
1029 return err;
1030
Lee Schermerhorn53f25562008-04-28 02:13:20 -07001031 down_read(&mm->mmap_sem);
Christoph Lameter39743882006-01-08 01:00:51 -08001032
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -08001033 /*
1034 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
1035 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
1036 * bit in 'tmp', and return that <source, dest> pair for migration.
1037 * The pair of nodemasks 'to' and 'from' define the map.
1038 *
1039 * If no pair of bits is found that way, fallback to picking some
1040 * pair of 'source' and 'dest' bits that are not the same. If the
1041 * 'source' and 'dest' bits are the same, this represents a node
1042 * that will be migrating to itself, so no pages need move.
1043 *
1044 * If no bits are left in 'tmp', or if all remaining bits left
1045 * in 'tmp' correspond to the same bit in 'to', return false
1046 * (nothing left to migrate).
1047 *
1048 * This lets us pick a pair of nodes to migrate between, such that
1049 * if possible the dest node is not already occupied by some other
1050 * source node, minimizing the risk of overloading the memory on a
1051 * node that would happen if we migrated incoming memory to a node
1052 * before migrating outgoing memory source that same node.
1053 *
1054 * A single scan of tmp is sufficient. As we go, we remember the
1055 * most recent <s, d> pair that moved (s != d). If we find a pair
1056 * that not only moved, but what's better, moved to an empty slot
1057 * (d is not set in tmp), then we break out then, with that pair.
Justin P. Mattockae0e47f2011-03-01 15:06:02 +01001058 * Otherwise when we finish scanning from_tmp, we at least have the
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -08001059 * most recent <s, d> pair that moved. If we get all the way through
1060 * the scan of tmp without finding any node that moved, much less
1061 * moved to an empty node, then there is nothing left worth migrating.
1062 */
Christoph Lameterd4984712006-01-08 01:00:55 -08001063
Andrew Morton0ce72d42012-05-29 15:06:24 -07001064 tmp = *from;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001065 while (!nodes_empty(tmp)) {
1066 int s,d;
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001067 int source = NUMA_NO_NODE;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001068 int dest = 0;
1069
1070 for_each_node_mask(s, tmp) {
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001071
1072 /*
1073 * do_migrate_pages() tries to maintain the relative
1074 * node relationship of the pages established between
1075 * threads and memory areas.
1076 *
1077 * However if the number of source nodes is not equal to
1078 * the number of destination nodes we can not preserve
1079 * this node relative relationship. In that case, skip
1080 * copying memory from a node that is in the destination
1081 * mask.
1082 *
1083 * Example: [2,3,4] -> [3,4,5] moves everything.
1084 * [0-7] - > [3,4,5] moves only 0,1,2,6,7.
1085 */
1086
Andrew Morton0ce72d42012-05-29 15:06:24 -07001087 if ((nodes_weight(*from) != nodes_weight(*to)) &&
1088 (node_isset(s, *to)))
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001089 continue;
1090
Andrew Morton0ce72d42012-05-29 15:06:24 -07001091 d = node_remap(s, *from, *to);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001092 if (s == d)
1093 continue;
1094
1095 source = s; /* Node moved. Memorize */
1096 dest = d;
1097
1098 /* dest not in remaining from nodes? */
1099 if (!node_isset(dest, tmp))
1100 break;
1101 }
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001102 if (source == NUMA_NO_NODE)
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001103 break;
1104
1105 node_clear(source, tmp);
1106 err = migrate_to_node(mm, source, dest, flags);
1107 if (err > 0)
1108 busy += err;
1109 if (err < 0)
1110 break;
Christoph Lameter39743882006-01-08 01:00:51 -08001111 }
1112 up_read(&mm->mmap_sem);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001113 if (err < 0)
1114 return err;
1115 return busy;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001116
Christoph Lameter39743882006-01-08 01:00:51 -08001117}
1118
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001119/*
1120 * Allocate a new page for page migration based on vma policy.
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001121 * Start by assuming the page is mapped by the same vma as contains @start.
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001122 * Search forward from there, if not. N.B., this assumes that the
1123 * list of pages handed to migrate_pages()--which is how we get here--
1124 * is in virtual address order.
1125 */
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001126static struct page *new_page(struct page *page, unsigned long start, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001127{
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001128 struct vm_area_struct *vma;
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001129 unsigned long uninitialized_var(address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001130
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001131 vma = find_vma(current->mm, start);
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001132 while (vma) {
1133 address = page_address_in_vma(page, vma);
1134 if (address != -EFAULT)
1135 break;
1136 vma = vma->vm_next;
1137 }
1138
Wanpeng Li11c731e2013-12-18 17:08:56 -08001139 if (PageHuge(page)) {
Michal Hockocc817172014-01-23 15:53:15 -08001140 BUG_ON(!vma);
1141 return alloc_huge_page_noerr(vma, address, 1);
Wanpeng Li11c731e2013-12-18 17:08:56 -08001142 }
1143 /*
1144 * if !vma, alloc_page_vma() will use task or system default policy
1145 */
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001146 return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001147}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001148#else
1149
1150static void migrate_page_add(struct page *page, struct list_head *pagelist,
1151 unsigned long flags)
1152{
1153}
1154
Andrew Morton0ce72d42012-05-29 15:06:24 -07001155int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1156 const nodemask_t *to, int flags)
Christoph Lameterb20a3502006-03-22 00:09:12 -08001157{
1158 return -ENOSYS;
1159}
Christoph Lameter95a402c2006-06-23 02:03:53 -07001160
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001161static struct page *new_page(struct page *page, unsigned long start, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001162{
1163 return NULL;
1164}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001165#endif
1166
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001167static long do_mbind(unsigned long start, unsigned long len,
David Rientjes028fec42008-04-28 02:12:25 -07001168 unsigned short mode, unsigned short mode_flags,
1169 nodemask_t *nmask, unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001170{
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001171 struct mm_struct *mm = current->mm;
1172 struct mempolicy *new;
1173 unsigned long end;
1174 int err;
1175 LIST_HEAD(pagelist);
1176
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001177 if (flags & ~(unsigned long)MPOL_MF_VALID)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001178 return -EINVAL;
Christoph Lameter74c00242006-03-14 19:50:21 -08001179 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001180 return -EPERM;
1181
1182 if (start & ~PAGE_MASK)
1183 return -EINVAL;
1184
1185 if (mode == MPOL_DEFAULT)
1186 flags &= ~MPOL_MF_STRICT;
1187
1188 len = (len + PAGE_SIZE - 1) & PAGE_MASK;
1189 end = start + len;
1190
1191 if (end < start)
1192 return -EINVAL;
1193 if (end == start)
1194 return 0;
1195
David Rientjes028fec42008-04-28 02:12:25 -07001196 new = mpol_new(mode, mode_flags, nmask);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001197 if (IS_ERR(new))
1198 return PTR_ERR(new);
1199
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001200 if (flags & MPOL_MF_LAZY)
1201 new->flags |= MPOL_F_MOF;
1202
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001203 /*
1204 * If we are using the default policy then operation
1205 * on discontinuous address spaces is okay after all
1206 */
1207 if (!new)
1208 flags |= MPOL_MF_DISCONTIG_OK;
1209
David Rientjes028fec42008-04-28 02:12:25 -07001210 pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
1211 start, start + len, mode, mode_flags,
David Rientjes00ef2d22013-02-22 16:35:36 -08001212 nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001213
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001214 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
1215
1216 err = migrate_prep();
1217 if (err)
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001218 goto mpol_out;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001219 }
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07001220 {
1221 NODEMASK_SCRATCH(scratch);
1222 if (scratch) {
1223 down_write(&mm->mmap_sem);
1224 task_lock(current);
1225 err = mpol_set_nodemask(new, nmask, scratch);
1226 task_unlock(current);
1227 if (err)
1228 up_write(&mm->mmap_sem);
1229 } else
1230 err = -ENOMEM;
1231 NODEMASK_SCRATCH_FREE(scratch);
1232 }
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001233 if (err)
1234 goto mpol_out;
1235
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001236 err = queue_pages_range(mm, start, end, nmask,
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001237 flags | MPOL_MF_INVERT, &pagelist);
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001238 if (!err)
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -08001239 err = mbind_range(mm, start, end, new);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001240
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001241 if (!err) {
1242 int nr_failed = 0;
1243
Minchan Kimcf608ac2010-10-26 14:21:29 -07001244 if (!list_empty(&pagelist)) {
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001245 WARN_ON_ONCE(flags & MPOL_MF_LAZY);
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001246 nr_failed = migrate_pages(&pagelist, new_page, NULL,
1247 start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001248 if (nr_failed)
Naoya Horiguchi74060e42013-09-11 14:22:06 -07001249 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001250 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001251
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001252 if (nr_failed && (flags & MPOL_MF_STRICT))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001253 err = -EIO;
KOSAKI Motohiroab8a3e12009-10-26 16:49:58 -07001254 } else
Joonsoo Kimb0e5fd72013-12-18 17:08:51 -08001255 putback_movable_pages(&pagelist);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001256
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001257 up_write(&mm->mmap_sem);
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001258 mpol_out:
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001259 mpol_put(new);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001260 return err;
1261}
1262
Christoph Lameter39743882006-01-08 01:00:51 -08001263/*
Christoph Lameter8bccd852005-10-29 18:16:59 -07001264 * User space interface with variable sized bitmaps for nodelists.
1265 */
1266
1267/* Copy a node mask from user space. */
Christoph Lameter39743882006-01-08 01:00:51 -08001268static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
Christoph Lameter8bccd852005-10-29 18:16:59 -07001269 unsigned long maxnode)
1270{
1271 unsigned long k;
1272 unsigned long nlongs;
1273 unsigned long endmask;
1274
1275 --maxnode;
1276 nodes_clear(*nodes);
1277 if (maxnode == 0 || !nmask)
1278 return 0;
Andi Kleena9c930b2006-02-20 18:27:59 -08001279 if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
Chris Wright636f13c2006-02-17 13:59:36 -08001280 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001281
1282 nlongs = BITS_TO_LONGS(maxnode);
1283 if ((maxnode % BITS_PER_LONG) == 0)
1284 endmask = ~0UL;
1285 else
1286 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
1287
1288 /* When the user specified more nodes than supported just check
1289 if the non supported part is all zero. */
1290 if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
1291 if (nlongs > PAGE_SIZE/sizeof(long))
1292 return -EINVAL;
1293 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
1294 unsigned long t;
1295 if (get_user(t, nmask + k))
1296 return -EFAULT;
1297 if (k == nlongs - 1) {
1298 if (t & endmask)
1299 return -EINVAL;
1300 } else if (t)
1301 return -EINVAL;
1302 }
1303 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
1304 endmask = ~0UL;
1305 }
1306
1307 if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
1308 return -EFAULT;
1309 nodes_addr(*nodes)[nlongs-1] &= endmask;
1310 return 0;
1311}
1312
1313/* Copy a kernel node mask to user space */
1314static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1315 nodemask_t *nodes)
1316{
1317 unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1318 const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long);
1319
1320 if (copy > nbytes) {
1321 if (copy > PAGE_SIZE)
1322 return -EINVAL;
1323 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1324 return -EFAULT;
1325 copy = nbytes;
1326 }
1327 return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1328}
1329
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001330SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
Rasmus Villemoesf7f28ca2014-06-04 16:07:57 -07001331 unsigned long, mode, const unsigned long __user *, nmask,
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001332 unsigned long, maxnode, unsigned, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001333{
1334 nodemask_t nodes;
1335 int err;
David Rientjes028fec42008-04-28 02:12:25 -07001336 unsigned short mode_flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001337
David Rientjes028fec42008-04-28 02:12:25 -07001338 mode_flags = mode & MPOL_MODE_FLAGS;
1339 mode &= ~MPOL_MODE_FLAGS;
David Rientjesa3b51e02008-04-28 02:12:23 -07001340 if (mode >= MPOL_MAX)
1341 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001342 if ((mode_flags & MPOL_F_STATIC_NODES) &&
1343 (mode_flags & MPOL_F_RELATIVE_NODES))
1344 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001345 err = get_nodes(&nodes, nmask, maxnode);
1346 if (err)
1347 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001348 return do_mbind(start, len, mode, mode_flags, &nodes, flags);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001349}
1350
1351/* Set the process memory policy */
Rasmus Villemoes23c89022014-06-04 16:07:58 -07001352SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask,
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001353 unsigned long, maxnode)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001354{
1355 int err;
1356 nodemask_t nodes;
David Rientjes028fec42008-04-28 02:12:25 -07001357 unsigned short flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001358
David Rientjes028fec42008-04-28 02:12:25 -07001359 flags = mode & MPOL_MODE_FLAGS;
1360 mode &= ~MPOL_MODE_FLAGS;
1361 if ((unsigned int)mode >= MPOL_MAX)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001362 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001363 if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
1364 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001365 err = get_nodes(&nodes, nmask, maxnode);
1366 if (err)
1367 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001368 return do_set_mempolicy(mode, flags, &nodes);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001369}
1370
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001371SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
1372 const unsigned long __user *, old_nodes,
1373 const unsigned long __user *, new_nodes)
Christoph Lameter39743882006-01-08 01:00:51 -08001374{
David Howellsc69e8d92008-11-14 10:39:19 +11001375 const struct cred *cred = current_cred(), *tcred;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001376 struct mm_struct *mm = NULL;
Christoph Lameter39743882006-01-08 01:00:51 -08001377 struct task_struct *task;
Christoph Lameter39743882006-01-08 01:00:51 -08001378 nodemask_t task_nodes;
1379 int err;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001380 nodemask_t *old;
1381 nodemask_t *new;
1382 NODEMASK_SCRATCH(scratch);
Christoph Lameter39743882006-01-08 01:00:51 -08001383
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001384 if (!scratch)
1385 return -ENOMEM;
Christoph Lameter39743882006-01-08 01:00:51 -08001386
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001387 old = &scratch->mask1;
1388 new = &scratch->mask2;
1389
1390 err = get_nodes(old, old_nodes, maxnode);
Christoph Lameter39743882006-01-08 01:00:51 -08001391 if (err)
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001392 goto out;
1393
1394 err = get_nodes(new, new_nodes, maxnode);
1395 if (err)
1396 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001397
1398 /* Find the mm_struct */
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001399 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001400 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter39743882006-01-08 01:00:51 -08001401 if (!task) {
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001402 rcu_read_unlock();
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001403 err = -ESRCH;
1404 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001405 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001406 get_task_struct(task);
Christoph Lameter39743882006-01-08 01:00:51 -08001407
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001408 err = -EINVAL;
Christoph Lameter39743882006-01-08 01:00:51 -08001409
1410 /*
1411 * Check if this process has the right to modify the specified
1412 * process. The right exists if the process has administrative
Alexey Dobriyan7f927fc2006-03-28 01:56:53 -08001413 * capabilities, superuser privileges or the same
Christoph Lameter39743882006-01-08 01:00:51 -08001414 * userid as the target process.
1415 */
David Howellsc69e8d92008-11-14 10:39:19 +11001416 tcred = __task_cred(task);
Eric W. Biedermanb38a86e2012-03-12 15:48:24 -07001417 if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1418 !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) &&
Christoph Lameter74c00242006-03-14 19:50:21 -08001419 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001420 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001421 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001422 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001423 }
David Howellsc69e8d92008-11-14 10:39:19 +11001424 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001425
1426 task_nodes = cpuset_mems_allowed(task);
1427 /* Is the user allowed to access the target nodes? */
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001428 if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
Christoph Lameter39743882006-01-08 01:00:51 -08001429 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001430 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001431 }
1432
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08001433 if (!nodes_subset(*new, node_states[N_MEMORY])) {
Christoph Lameter3b42d282007-08-31 00:12:08 -07001434 err = -EINVAL;
Christoph Lameter3268c632012-03-21 16:34:06 -07001435 goto out_put;
Christoph Lameter3b42d282007-08-31 00:12:08 -07001436 }
1437
David Quigley86c3a762006-06-23 02:04:02 -07001438 err = security_task_movememory(task);
1439 if (err)
Christoph Lameter3268c632012-03-21 16:34:06 -07001440 goto out_put;
David Quigley86c3a762006-06-23 02:04:02 -07001441
Christoph Lameter3268c632012-03-21 16:34:06 -07001442 mm = get_task_mm(task);
1443 put_task_struct(task);
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001444
1445 if (!mm) {
Christoph Lameter3268c632012-03-21 16:34:06 -07001446 err = -EINVAL;
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001447 goto out;
1448 }
1449
1450 err = do_migrate_pages(mm, old, new,
1451 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
Christoph Lameter3268c632012-03-21 16:34:06 -07001452
1453 mmput(mm);
1454out:
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001455 NODEMASK_SCRATCH_FREE(scratch);
1456
Christoph Lameter39743882006-01-08 01:00:51 -08001457 return err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001458
1459out_put:
1460 put_task_struct(task);
1461 goto out;
1462
Christoph Lameter39743882006-01-08 01:00:51 -08001463}
1464
1465
Christoph Lameter8bccd852005-10-29 18:16:59 -07001466/* Retrieve NUMA policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001467SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1468 unsigned long __user *, nmask, unsigned long, maxnode,
1469 unsigned long, addr, unsigned long, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001470{
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001471 int err;
1472 int uninitialized_var(pval);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001473 nodemask_t nodes;
1474
1475 if (nmask != NULL && maxnode < MAX_NUMNODES)
1476 return -EINVAL;
1477
1478 err = do_get_mempolicy(&pval, &nodes, addr, flags);
1479
1480 if (err)
1481 return err;
1482
1483 if (policy && put_user(pval, policy))
1484 return -EFAULT;
1485
1486 if (nmask)
1487 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1488
1489 return err;
1490}
1491
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492#ifdef CONFIG_COMPAT
1493
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001494COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1495 compat_ulong_t __user *, nmask,
1496 compat_ulong_t, maxnode,
1497 compat_ulong_t, addr, compat_ulong_t, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498{
1499 long err;
1500 unsigned long __user *nm = NULL;
1501 unsigned long nr_bits, alloc_size;
1502 DECLARE_BITMAP(bm, MAX_NUMNODES);
1503
1504 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1505 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1506
1507 if (nmask)
1508 nm = compat_alloc_user_space(alloc_size);
1509
1510 err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
1511
1512 if (!err && nmask) {
KAMEZAWA Hiroyuki2bbff6c2011-09-14 16:21:02 -07001513 unsigned long copy_size;
1514 copy_size = min_t(unsigned long, sizeof(bm), alloc_size);
1515 err = copy_from_user(bm, nm, copy_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 /* ensure entire bitmap is zeroed */
1517 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
1518 err |= compat_put_bitmap(nmask, bm, nr_bits);
1519 }
1520
1521 return err;
1522}
1523
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001524COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask,
1525 compat_ulong_t, maxnode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526{
1527 long err = 0;
1528 unsigned long __user *nm = NULL;
1529 unsigned long nr_bits, alloc_size;
1530 DECLARE_BITMAP(bm, MAX_NUMNODES);
1531
1532 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1533 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1534
1535 if (nmask) {
1536 err = compat_get_bitmap(bm, nmask, nr_bits);
1537 nm = compat_alloc_user_space(alloc_size);
1538 err |= copy_to_user(nm, bm, alloc_size);
1539 }
1540
1541 if (err)
1542 return -EFAULT;
1543
1544 return sys_set_mempolicy(mode, nm, nr_bits+1);
1545}
1546
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001547COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len,
1548 compat_ulong_t, mode, compat_ulong_t __user *, nmask,
1549 compat_ulong_t, maxnode, compat_ulong_t, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550{
1551 long err = 0;
1552 unsigned long __user *nm = NULL;
1553 unsigned long nr_bits, alloc_size;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001554 nodemask_t bm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
1556 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1557 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1558
1559 if (nmask) {
Andi Kleendfcd3c02005-10-29 18:15:48 -07001560 err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 nm = compat_alloc_user_space(alloc_size);
Andi Kleendfcd3c02005-10-29 18:15:48 -07001562 err |= copy_to_user(nm, nodes_addr(bm), alloc_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 }
1564
1565 if (err)
1566 return -EFAULT;
1567
1568 return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
1569}
1570
1571#endif
1572
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001573struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
1574 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575{
Oleg Nesterov8d90274b2014-10-09 15:27:45 -07001576 struct mempolicy *pol = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
1578 if (vma) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001579 if (vma->vm_ops && vma->vm_ops->get_policy) {
Oleg Nesterov8d90274b2014-10-09 15:27:45 -07001580 pol = vma->vm_ops->get_policy(vma, addr);
Mel Gorman00442ad2012-10-08 16:29:20 -07001581 } else if (vma->vm_policy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 pol = vma->vm_policy;
Mel Gorman00442ad2012-10-08 16:29:20 -07001583
1584 /*
1585 * shmem_alloc_page() passes MPOL_F_SHARED policy with
1586 * a pseudo vma whose vma->vm_ops=NULL. Take a reference
1587 * count on these policies which will be dropped by
1588 * mpol_cond_put() later
1589 */
1590 if (mpol_needs_cond_ref(pol))
1591 mpol_get(pol);
1592 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 }
Oleg Nesterovf15ca782014-10-09 15:27:43 -07001594
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001595 return pol;
1596}
1597
1598/*
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001599 * get_vma_policy(@vma, @addr)
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001600 * @vma: virtual memory area whose policy is sought
1601 * @addr: address in @vma for shared policy lookup
1602 *
1603 * Returns effective policy for a VMA at specified address.
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001604 * Falls back to current->mempolicy or system default policy, as necessary.
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001605 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1606 * count--added by the get_policy() vm_op, as appropriate--to protect against
1607 * freeing by another task. It is the caller's responsibility to free the
1608 * extra reference for shared policies.
1609 */
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001610static struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
1611 unsigned long addr)
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001612{
1613 struct mempolicy *pol = __get_vma_policy(vma, addr);
1614
Oleg Nesterov8d90274b2014-10-09 15:27:45 -07001615 if (!pol)
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001616 pol = get_task_policy(current);
Oleg Nesterov8d90274b2014-10-09 15:27:45 -07001617
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 return pol;
1619}
1620
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001621bool vma_policy_mof(struct vm_area_struct *vma)
Mel Gormanfc3147242013-10-07 11:29:09 +01001622{
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001623 struct mempolicy *pol;
Oleg Nesterovf15ca782014-10-09 15:27:43 -07001624
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001625 if (vma->vm_ops && vma->vm_ops->get_policy) {
1626 bool ret = false;
Mel Gormanfc3147242013-10-07 11:29:09 +01001627
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001628 pol = vma->vm_ops->get_policy(vma, vma->vm_start);
1629 if (pol && (pol->flags & MPOL_F_MOF))
1630 ret = true;
1631 mpol_cond_put(pol);
Mel Gormanfc3147242013-10-07 11:29:09 +01001632
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001633 return ret;
Mel Gormanfc3147242013-10-07 11:29:09 +01001634 }
1635
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001636 pol = vma->vm_policy;
Oleg Nesterov8d90274b2014-10-09 15:27:45 -07001637 if (!pol)
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001638 pol = get_task_policy(current);
Oleg Nesterov8d90274b2014-10-09 15:27:45 -07001639
Mel Gormanfc3147242013-10-07 11:29:09 +01001640 return pol->flags & MPOL_F_MOF;
1641}
1642
Lai Jiangshand3eb1572013-02-22 16:33:22 -08001643static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone)
1644{
1645 enum zone_type dynamic_policy_zone = policy_zone;
1646
1647 BUG_ON(dynamic_policy_zone == ZONE_MOVABLE);
1648
1649 /*
1650 * if policy->v.nodes has movable memory only,
1651 * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only.
1652 *
1653 * policy->v.nodes is intersect with node_states[N_MEMORY].
1654 * so if the following test faile, it implies
1655 * policy->v.nodes has movable memory only.
1656 */
1657 if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY]))
1658 dynamic_policy_zone = ZONE_MOVABLE;
1659
1660 return zone >= dynamic_policy_zone;
1661}
1662
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001663/*
1664 * Return a nodemask representing a mempolicy for filtering nodes for
1665 * page allocation
1666 */
1667static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
Mel Gorman19770b32008-04-28 02:12:18 -07001668{
1669 /* Lower zones don't get a nodemask applied for MPOL_BIND */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001670 if (unlikely(policy->mode == MPOL_BIND) &&
Lai Jiangshand3eb1572013-02-22 16:33:22 -08001671 apply_policy_zone(policy, gfp_zone(gfp)) &&
Mel Gorman19770b32008-04-28 02:12:18 -07001672 cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
1673 return &policy->v.nodes;
1674
1675 return NULL;
1676}
1677
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001678/* Return a zonelist indicated by gfp for node representing a mempolicy */
Andi Kleen2f5f9482011-03-04 17:36:29 -08001679static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy,
1680 int nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681{
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001682 switch (policy->mode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001684 if (!(policy->flags & MPOL_F_LOCAL))
1685 nd = policy->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 break;
1687 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07001688 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001689 * Normally, MPOL_BIND allocations are node-local within the
1690 * allowed nodemask. However, if __GFP_THISNODE is set and the
Bob Liu6eb27e12010-05-24 14:32:00 -07001691 * current node isn't part of the mask, we use the zonelist for
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001692 * the first node in the mask instead.
Mel Gorman19770b32008-04-28 02:12:18 -07001693 */
Mel Gorman19770b32008-04-28 02:12:18 -07001694 if (unlikely(gfp & __GFP_THISNODE) &&
1695 unlikely(!node_isset(nd, policy->v.nodes)))
1696 nd = first_node(policy->v.nodes);
1697 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 BUG();
1700 }
Mel Gorman0e884602008-04-28 02:12:14 -07001701 return node_zonelist(nd, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702}
1703
1704/* Do dynamic interleaving for a process */
1705static unsigned interleave_nodes(struct mempolicy *policy)
1706{
1707 unsigned nid, next;
1708 struct task_struct *me = current;
1709
1710 nid = me->il_next;
Andrew Morton0edaf862016-05-19 17:10:58 -07001711 next = next_node_in(nid, policy->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001712 if (next < MAX_NUMNODES)
1713 me->il_next = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 return nid;
1715}
1716
Christoph Lameterdc85da12006-01-18 17:42:36 -08001717/*
1718 * Depending on the memory policy provide a node from which to allocate the
1719 * next slab entry.
1720 */
David Rientjes2a389612014-04-07 15:37:29 -07001721unsigned int mempolicy_slab_node(void)
Christoph Lameterdc85da12006-01-18 17:42:36 -08001722{
Andi Kleene7b691b2012-06-09 02:40:03 -07001723 struct mempolicy *policy;
David Rientjes2a389612014-04-07 15:37:29 -07001724 int node = numa_mem_id();
Andi Kleene7b691b2012-06-09 02:40:03 -07001725
1726 if (in_interrupt())
David Rientjes2a389612014-04-07 15:37:29 -07001727 return node;
Andi Kleene7b691b2012-06-09 02:40:03 -07001728
1729 policy = current->mempolicy;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001730 if (!policy || policy->flags & MPOL_F_LOCAL)
David Rientjes2a389612014-04-07 15:37:29 -07001731 return node;
Christoph Lameter765c4502006-09-27 01:50:08 -07001732
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001733 switch (policy->mode) {
1734 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001735 /*
1736 * handled MPOL_F_LOCAL above
1737 */
1738 return policy->v.preferred_node;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001739
Christoph Lameterdc85da12006-01-18 17:42:36 -08001740 case MPOL_INTERLEAVE:
1741 return interleave_nodes(policy);
1742
Mel Gormandd1a2392008-04-28 02:12:17 -07001743 case MPOL_BIND: {
Mel Gormanc33d6c02016-05-19 17:14:10 -07001744 struct zoneref *z;
1745
Christoph Lameterdc85da12006-01-18 17:42:36 -08001746 /*
1747 * Follow bind policy behavior and start allocation at the
1748 * first node.
1749 */
Mel Gorman19770b32008-04-28 02:12:18 -07001750 struct zonelist *zonelist;
Mel Gorman19770b32008-04-28 02:12:18 -07001751 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
David Rientjes2a389612014-04-07 15:37:29 -07001752 zonelist = &NODE_DATA(node)->node_zonelists[0];
Mel Gormanc33d6c02016-05-19 17:14:10 -07001753 z = first_zones_zonelist(zonelist, highest_zoneidx,
1754 &policy->v.nodes);
1755 return z->zone ? z->zone->node : node;
Mel Gormandd1a2392008-04-28 02:12:17 -07001756 }
Christoph Lameterdc85da12006-01-18 17:42:36 -08001757
Christoph Lameterdc85da12006-01-18 17:42:36 -08001758 default:
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001759 BUG();
Christoph Lameterdc85da12006-01-18 17:42:36 -08001760 }
1761}
1762
Andrew Mortonfee83b32016-05-19 17:11:43 -07001763/*
1764 * Do static interleaving for a VMA with known offset @n. Returns the n'th
1765 * node in pol->v.nodes (starting from n=0), wrapping around if n exceeds the
1766 * number of present nodes.
1767 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768static unsigned offset_il_node(struct mempolicy *pol,
Andrew Mortonfee83b32016-05-19 17:11:43 -07001769 struct vm_area_struct *vma, unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770{
Andi Kleendfcd3c02005-10-29 18:15:48 -07001771 unsigned nnodes = nodes_weight(pol->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001772 unsigned target;
Andrew Mortonfee83b32016-05-19 17:11:43 -07001773 int i;
1774 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775
David Rientjesf5b087b2008-04-28 02:12:27 -07001776 if (!nnodes)
1777 return numa_node_id();
Andrew Mortonfee83b32016-05-19 17:11:43 -07001778 target = (unsigned int)n % nnodes;
1779 nid = first_node(pol->v.nodes);
1780 for (i = 0; i < target; i++)
Andi Kleendfcd3c02005-10-29 18:15:48 -07001781 nid = next_node(nid, pol->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 return nid;
1783}
1784
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001785/* Determine a node number for interleave */
1786static inline unsigned interleave_nid(struct mempolicy *pol,
1787 struct vm_area_struct *vma, unsigned long addr, int shift)
1788{
1789 if (vma) {
1790 unsigned long off;
1791
Nishanth Aravamudan3b98b082006-08-31 21:27:53 -07001792 /*
1793 * for small pages, there is no difference between
1794 * shift and PAGE_SHIFT, so the bit-shift is safe.
1795 * for huge pages, since vm_pgoff is in units of small
1796 * pages, we need to shift off the always 0 bits to get
1797 * a useful offset.
1798 */
1799 BUG_ON(shift < PAGE_SHIFT);
1800 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001801 off += (addr - vma->vm_start) >> shift;
1802 return offset_il_node(pol, vma, off);
1803 } else
1804 return interleave_nodes(pol);
1805}
1806
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001807#ifdef CONFIG_HUGETLBFS
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001808/*
1809 * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
Fabian Frederickb46e14a2014-06-04 16:08:18 -07001810 * @vma: virtual memory area whose policy is sought
1811 * @addr: address in @vma for shared policy lookup and interleave policy
1812 * @gfp_flags: for requested zone
1813 * @mpol: pointer to mempolicy pointer for reference counted mempolicy
1814 * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001815 *
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001816 * Returns a zonelist suitable for a huge page allocation and a pointer
1817 * to the struct mempolicy for conditional unref after allocation.
1818 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1819 * @nodemask for filtering the zonelist.
Miao Xiec0ff7452010-05-24 14:32:08 -07001820 *
Mel Gormand26914d2014-04-03 14:47:24 -07001821 * Must be protected by read_mems_allowed_begin()
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001822 */
Mel Gorman396faf02007-07-17 04:03:13 -07001823struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
Mel Gorman19770b32008-04-28 02:12:18 -07001824 gfp_t gfp_flags, struct mempolicy **mpol,
1825 nodemask_t **nodemask)
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001826{
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001827 struct zonelist *zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001828
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001829 *mpol = get_vma_policy(vma, addr);
Mel Gorman19770b32008-04-28 02:12:18 -07001830 *nodemask = NULL; /* assume !MPOL_BIND */
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001831
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001832 if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
1833 zl = node_zonelist(interleave_nid(*mpol, vma, addr,
Andi Kleena5516432008-07-23 21:27:41 -07001834 huge_page_shift(hstate_vma(vma))), gfp_flags);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001835 } else {
Andi Kleen2f5f9482011-03-04 17:36:29 -08001836 zl = policy_zonelist(gfp_flags, *mpol, numa_node_id());
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001837 if ((*mpol)->mode == MPOL_BIND)
1838 *nodemask = &(*mpol)->v.nodes;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001839 }
1840 return zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001841}
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001842
1843/*
1844 * init_nodemask_of_mempolicy
1845 *
1846 * If the current task's mempolicy is "default" [NULL], return 'false'
1847 * to indicate default policy. Otherwise, extract the policy nodemask
1848 * for 'bind' or 'interleave' policy into the argument nodemask, or
1849 * initialize the argument nodemask to contain the single node for
1850 * 'preferred' or 'local' policy and return 'true' to indicate presence
1851 * of non-default mempolicy.
1852 *
1853 * We don't bother with reference counting the mempolicy [mpol_get/put]
1854 * because the current task is examining it's own mempolicy and a task's
1855 * mempolicy is only ever changed by the task itself.
1856 *
1857 * N.B., it is the caller's responsibility to free a returned nodemask.
1858 */
1859bool init_nodemask_of_mempolicy(nodemask_t *mask)
1860{
1861 struct mempolicy *mempolicy;
1862 int nid;
1863
1864 if (!(mask && current->mempolicy))
1865 return false;
1866
Miao Xiec0ff7452010-05-24 14:32:08 -07001867 task_lock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001868 mempolicy = current->mempolicy;
1869 switch (mempolicy->mode) {
1870 case MPOL_PREFERRED:
1871 if (mempolicy->flags & MPOL_F_LOCAL)
1872 nid = numa_node_id();
1873 else
1874 nid = mempolicy->v.preferred_node;
1875 init_nodemask_of_node(mask, nid);
1876 break;
1877
1878 case MPOL_BIND:
1879 /* Fall through */
1880 case MPOL_INTERLEAVE:
1881 *mask = mempolicy->v.nodes;
1882 break;
1883
1884 default:
1885 BUG();
1886 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001887 task_unlock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001888
1889 return true;
1890}
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001891#endif
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001892
David Rientjes6f48d0eb2010-08-09 17:18:52 -07001893/*
1894 * mempolicy_nodemask_intersects
1895 *
1896 * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default
1897 * policy. Otherwise, check for intersection between mask and the policy
1898 * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local'
1899 * policy, always return true since it may allocate elsewhere on fallback.
1900 *
1901 * Takes task_lock(tsk) to prevent freeing of its mempolicy.
1902 */
1903bool mempolicy_nodemask_intersects(struct task_struct *tsk,
1904 const nodemask_t *mask)
1905{
1906 struct mempolicy *mempolicy;
1907 bool ret = true;
1908
1909 if (!mask)
1910 return ret;
1911 task_lock(tsk);
1912 mempolicy = tsk->mempolicy;
1913 if (!mempolicy)
1914 goto out;
1915
1916 switch (mempolicy->mode) {
1917 case MPOL_PREFERRED:
1918 /*
1919 * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to
1920 * allocate from, they may fallback to other nodes when oom.
1921 * Thus, it's possible for tsk to have allocated memory from
1922 * nodes in mask.
1923 */
1924 break;
1925 case MPOL_BIND:
1926 case MPOL_INTERLEAVE:
1927 ret = nodes_intersects(mempolicy->v.nodes, *mask);
1928 break;
1929 default:
1930 BUG();
1931 }
1932out:
1933 task_unlock(tsk);
1934 return ret;
1935}
1936
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937/* Allocate a page in interleaved policy.
1938 Own path because it needs to do special accounting. */
Andi Kleen662f3a02005-10-29 18:15:49 -07001939static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
1940 unsigned nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941{
1942 struct zonelist *zl;
1943 struct page *page;
1944
Mel Gorman0e884602008-04-28 02:12:14 -07001945 zl = node_zonelist(nid, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 page = __alloc_pages(gfp, order, zl);
Mel Gormandd1a2392008-04-28 02:12:17 -07001947 if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
Christoph Lameterca889e62006-06-30 01:55:44 -07001948 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 return page;
1950}
1951
1952/**
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001953 * alloc_pages_vma - Allocate a page for a VMA.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 *
1955 * @gfp:
1956 * %GFP_USER user allocation.
1957 * %GFP_KERNEL kernel allocations,
1958 * %GFP_HIGHMEM highmem/user allocations,
1959 * %GFP_FS allocation should not call back into a file system.
1960 * %GFP_ATOMIC don't sleep.
1961 *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001962 * @order:Order of the GFP allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 * @vma: Pointer to VMA or NULL if not available.
1964 * @addr: Virtual Address of the allocation. Must be inside the VMA.
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001965 * @node: Which node to prefer for allocation (modulo policy).
1966 * @hugepage: for hugepages try only the preferred node if possible
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 *
1968 * This function allocates a page from the kernel page pool and applies
1969 * a NUMA policy associated with the VMA or the current process.
1970 * When VMA is not NULL caller must hold down_read on the mmap_sem of the
1971 * mm_struct of the VMA to prevent it from going away. Should be used for
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001972 * all allocations for pages that will be mapped into user space. Returns
1973 * NULL when no page can be allocated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 */
1975struct page *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001976alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001977 unsigned long addr, int node, bool hugepage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978{
Mel Gormancc9a6c82012-03-21 16:34:11 -07001979 struct mempolicy *pol;
Miao Xiec0ff7452010-05-24 14:32:08 -07001980 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001981 unsigned int cpuset_mems_cookie;
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001982 struct zonelist *zl;
1983 nodemask_t *nmask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
Mel Gormancc9a6c82012-03-21 16:34:11 -07001985retry_cpuset:
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001986 pol = get_vma_policy(vma, addr);
Mel Gormand26914d2014-04-03 14:47:24 -07001987 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07001988
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001989 if (pol->mode == MPOL_INTERLEAVE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 unsigned nid;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001991
Andi Kleen8eac5632011-02-25 14:44:28 -08001992 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001993 mpol_cond_put(pol);
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001994 page = alloc_page_interleave(gfp, order, nid);
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001995 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 }
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001997
Vlastimil Babka0867a572015-06-24 16:58:48 -07001998 if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) {
1999 int hpage_node = node;
2000
2001 /*
2002 * For hugepage allocation and non-interleave policy which
2003 * allows the current node (or other explicitly preferred
2004 * node) we only try to allocate from the current/preferred
2005 * node and don't fall back to other nodes, as the cost of
2006 * remote accesses would likely offset THP benefits.
2007 *
2008 * If the policy is interleave, or does not allow the current
2009 * node in its nodemask, we allocate the standard way.
2010 */
2011 if (pol->mode == MPOL_PREFERRED &&
2012 !(pol->flags & MPOL_F_LOCAL))
2013 hpage_node = pol->v.preferred_node;
2014
2015 nmask = policy_nodemask(gfp, pol);
2016 if (!nmask || node_isset(hpage_node, *nmask)) {
2017 mpol_cond_put(pol);
Vlastimil Babka96db8002015-09-08 15:03:50 -07002018 page = __alloc_pages_node(hpage_node,
Vlastimil Babka0867a572015-06-24 16:58:48 -07002019 gfp | __GFP_THISNODE, order);
2020 goto out;
2021 }
2022 }
2023
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002024 nmask = policy_nodemask(gfp, pol);
2025 zl = policy_zonelist(gfp, pol, node);
Oleg Nesterov23867402014-10-09 15:27:41 -07002026 mpol_cond_put(pol);
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002027 page = __alloc_pages_nodemask(gfp, order, zl, nmask);
2028out:
Mel Gormand26914d2014-04-03 14:47:24 -07002029 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002030 goto retry_cpuset;
Miao Xiec0ff7452010-05-24 14:32:08 -07002031 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032}
2033
2034/**
2035 * alloc_pages_current - Allocate pages.
2036 *
2037 * @gfp:
2038 * %GFP_USER user allocation,
2039 * %GFP_KERNEL kernel allocation,
2040 * %GFP_HIGHMEM highmem allocation,
2041 * %GFP_FS don't call back into a file system.
2042 * %GFP_ATOMIC don't sleep.
2043 * @order: Power of two of allocation size in pages. 0 is a single page.
2044 *
2045 * Allocate a page from the kernel page pool. When not in
2046 * interrupt context and apply the current process NUMA policy.
2047 * Returns NULL when no page can be allocated.
2048 *
Paul Jacksoncf2a473c2006-01-08 01:01:54 -08002049 * Don't call cpuset_update_task_memory_state() unless
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 * 1) it's ok to take cpuset_sem (can WAIT), and
2051 * 2) allocating for current task (not interrupt).
2052 */
Al Virodd0fc662005-10-07 07:46:04 +01002053struct page *alloc_pages_current(gfp_t gfp, unsigned order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054{
Oleg Nesterov8d90274b2014-10-09 15:27:45 -07002055 struct mempolicy *pol = &default_policy;
Miao Xiec0ff7452010-05-24 14:32:08 -07002056 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002057 unsigned int cpuset_mems_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
Oleg Nesterov8d90274b2014-10-09 15:27:45 -07002059 if (!in_interrupt() && !(gfp & __GFP_THISNODE))
2060 pol = get_task_policy(current);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002061
Mel Gormancc9a6c82012-03-21 16:34:11 -07002062retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07002063 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002064
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002065 /*
2066 * No reference counting needed for current->mempolicy
2067 * nor system default_policy
2068 */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002069 if (pol->mode == MPOL_INTERLEAVE)
Miao Xiec0ff7452010-05-24 14:32:08 -07002070 page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
2071 else
2072 page = __alloc_pages_nodemask(gfp, order,
Andi Kleen5c4b4be2011-03-04 17:36:32 -08002073 policy_zonelist(gfp, pol, numa_node_id()),
2074 policy_nodemask(gfp, pol));
Mel Gormancc9a6c82012-03-21 16:34:11 -07002075
Mel Gormand26914d2014-04-03 14:47:24 -07002076 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002077 goto retry_cpuset;
2078
Miao Xiec0ff7452010-05-24 14:32:08 -07002079 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080}
2081EXPORT_SYMBOL(alloc_pages_current);
2082
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002083int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
2084{
2085 struct mempolicy *pol = mpol_dup(vma_policy(src));
2086
2087 if (IS_ERR(pol))
2088 return PTR_ERR(pol);
2089 dst->vm_policy = pol;
2090 return 0;
2091}
2092
Paul Jackson42253992006-01-08 01:01:59 -08002093/*
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002094 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
Paul Jackson42253992006-01-08 01:01:59 -08002095 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
2096 * with the mems_allowed returned by cpuset_mems_allowed(). This
2097 * keeps mempolicies cpuset relative after its cpuset moves. See
2098 * further kernel/cpuset.c update_nodemask().
Miao Xie708c1bb2010-05-24 14:32:07 -07002099 *
2100 * current's mempolicy may be rebinded by the other task(the task that changes
2101 * cpuset's mems), so we needn't do rebind work for current task.
Paul Jackson42253992006-01-08 01:01:59 -08002102 */
Paul Jackson42253992006-01-08 01:01:59 -08002103
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002104/* Slow path of a mempolicy duplicate */
2105struct mempolicy *__mpol_dup(struct mempolicy *old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106{
2107 struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2108
2109 if (!new)
2110 return ERR_PTR(-ENOMEM);
Miao Xie708c1bb2010-05-24 14:32:07 -07002111
2112 /* task's mempolicy is protected by alloc_lock */
2113 if (old == current->mempolicy) {
2114 task_lock(current);
2115 *new = *old;
2116 task_unlock(current);
2117 } else
2118 *new = *old;
2119
Paul Jackson42253992006-01-08 01:01:59 -08002120 if (current_cpuset_is_being_rebound()) {
2121 nodemask_t mems = cpuset_mems_allowed(current);
Miao Xie708c1bb2010-05-24 14:32:07 -07002122 if (new->flags & MPOL_F_REBINDING)
2123 mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2);
2124 else
2125 mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE);
Paul Jackson42253992006-01-08 01:01:59 -08002126 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 atomic_set(&new->refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 return new;
2129}
2130
2131/* Slow path of a mempolicy comparison */
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002132bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133{
2134 if (!a || !b)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002135 return false;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002136 if (a->mode != b->mode)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002137 return false;
Bob Liu19800502010-05-24 14:32:01 -07002138 if (a->flags != b->flags)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002139 return false;
Bob Liu19800502010-05-24 14:32:01 -07002140 if (mpol_store_user_nodemask(a))
2141 if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask))
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002142 return false;
Bob Liu19800502010-05-24 14:32:01 -07002143
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002144 switch (a->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -07002145 case MPOL_BIND:
2146 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 case MPOL_INTERLEAVE:
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002148 return !!nodes_equal(a->v.nodes, b->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 case MPOL_PREFERRED:
Namhyung Kim75719662011-03-22 16:33:02 -07002150 return a->v.preferred_node == b->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 default:
2152 BUG();
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002153 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 }
2155}
2156
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 * Shared memory backing store policy support.
2159 *
2160 * Remember policies even when nobody has shared memory mapped.
2161 * The policies are kept in Red-Black tree linked from the inode.
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002162 * They are protected by the sp->lock rwlock, which should be held
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 * for any accesses to the tree.
2164 */
2165
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002166/*
2167 * lookup first element intersecting start-end. Caller holds sp->lock for
2168 * reading or for writing
2169 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170static struct sp_node *
2171sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
2172{
2173 struct rb_node *n = sp->root.rb_node;
2174
2175 while (n) {
2176 struct sp_node *p = rb_entry(n, struct sp_node, nd);
2177
2178 if (start >= p->end)
2179 n = n->rb_right;
2180 else if (end <= p->start)
2181 n = n->rb_left;
2182 else
2183 break;
2184 }
2185 if (!n)
2186 return NULL;
2187 for (;;) {
2188 struct sp_node *w = NULL;
2189 struct rb_node *prev = rb_prev(n);
2190 if (!prev)
2191 break;
2192 w = rb_entry(prev, struct sp_node, nd);
2193 if (w->end <= start)
2194 break;
2195 n = prev;
2196 }
2197 return rb_entry(n, struct sp_node, nd);
2198}
2199
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002200/*
2201 * Insert a new shared policy into the list. Caller holds sp->lock for
2202 * writing.
2203 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204static void sp_insert(struct shared_policy *sp, struct sp_node *new)
2205{
2206 struct rb_node **p = &sp->root.rb_node;
2207 struct rb_node *parent = NULL;
2208 struct sp_node *nd;
2209
2210 while (*p) {
2211 parent = *p;
2212 nd = rb_entry(parent, struct sp_node, nd);
2213 if (new->start < nd->start)
2214 p = &(*p)->rb_left;
2215 else if (new->end > nd->end)
2216 p = &(*p)->rb_right;
2217 else
2218 BUG();
2219 }
2220 rb_link_node(&new->nd, parent, p);
2221 rb_insert_color(&new->nd, &sp->root);
Paul Mundt140d5a42007-07-15 23:38:16 -07002222 pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002223 new->policy ? new->policy->mode : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224}
2225
2226/* Find shared policy intersecting idx */
2227struct mempolicy *
2228mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
2229{
2230 struct mempolicy *pol = NULL;
2231 struct sp_node *sn;
2232
2233 if (!sp->root.rb_node)
2234 return NULL;
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002235 read_lock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 sn = sp_lookup(sp, idx, idx+1);
2237 if (sn) {
2238 mpol_get(sn->policy);
2239 pol = sn->policy;
2240 }
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002241 read_unlock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 return pol;
2243}
2244
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002245static void sp_free(struct sp_node *n)
2246{
2247 mpol_put(n->policy);
2248 kmem_cache_free(sn_cache, n);
2249}
2250
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002251/**
2252 * mpol_misplaced - check whether current page node is valid in policy
2253 *
Fabian Frederickb46e14a2014-06-04 16:08:18 -07002254 * @page: page to be checked
2255 * @vma: vm area where page mapped
2256 * @addr: virtual address where page mapped
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002257 *
2258 * Lookup current policy node id for vma,addr and "compare to" page's
2259 * node id.
2260 *
2261 * Returns:
2262 * -1 - not misplaced, page is in the right node
2263 * node - node id where the page should be
2264 *
2265 * Policy determination "mimics" alloc_page_vma().
2266 * Called from fault path where we know the vma and faulting address.
2267 */
2268int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr)
2269{
2270 struct mempolicy *pol;
Mel Gormanc33d6c02016-05-19 17:14:10 -07002271 struct zoneref *z;
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002272 int curnid = page_to_nid(page);
2273 unsigned long pgoff;
Peter Zijlstra90572892013-10-07 11:29:20 +01002274 int thiscpu = raw_smp_processor_id();
2275 int thisnid = cpu_to_node(thiscpu);
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002276 int polnid = -1;
2277 int ret = -1;
2278
2279 BUG_ON(!vma);
2280
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07002281 pol = get_vma_policy(vma, addr);
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002282 if (!(pol->flags & MPOL_F_MOF))
2283 goto out;
2284
2285 switch (pol->mode) {
2286 case MPOL_INTERLEAVE:
2287 BUG_ON(addr >= vma->vm_end);
2288 BUG_ON(addr < vma->vm_start);
2289
2290 pgoff = vma->vm_pgoff;
2291 pgoff += (addr - vma->vm_start) >> PAGE_SHIFT;
2292 polnid = offset_il_node(pol, vma, pgoff);
2293 break;
2294
2295 case MPOL_PREFERRED:
2296 if (pol->flags & MPOL_F_LOCAL)
2297 polnid = numa_node_id();
2298 else
2299 polnid = pol->v.preferred_node;
2300 break;
2301
2302 case MPOL_BIND:
Mel Gormanc33d6c02016-05-19 17:14:10 -07002303
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002304 /*
2305 * allows binding to multiple nodes.
2306 * use current page if in policy nodemask,
2307 * else select nearest allowed node, if any.
2308 * If no allowed nodes, use current [!misplaced].
2309 */
2310 if (node_isset(curnid, pol->v.nodes))
2311 goto out;
Mel Gormanc33d6c02016-05-19 17:14:10 -07002312 z = first_zones_zonelist(
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002313 node_zonelist(numa_node_id(), GFP_HIGHUSER),
2314 gfp_zone(GFP_HIGHUSER),
Mel Gormanc33d6c02016-05-19 17:14:10 -07002315 &pol->v.nodes);
2316 polnid = z->zone->node;
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002317 break;
2318
2319 default:
2320 BUG();
2321 }
Mel Gorman5606e382012-11-02 18:19:13 +00002322
2323 /* Migrate the page towards the node whose CPU is referencing it */
Mel Gormane42c8ff2012-11-12 09:17:07 +00002324 if (pol->flags & MPOL_F_MORON) {
Peter Zijlstra90572892013-10-07 11:29:20 +01002325 polnid = thisnid;
Mel Gorman5606e382012-11-02 18:19:13 +00002326
Rik van Riel10f39042014-01-27 17:03:44 -05002327 if (!should_numa_migrate_memory(current, page, curnid, thiscpu))
Rik van Rielde1c9ce2013-10-07 11:29:39 +01002328 goto out;
Mel Gormane42c8ff2012-11-12 09:17:07 +00002329 }
2330
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002331 if (curnid != polnid)
2332 ret = polnid;
2333out:
2334 mpol_cond_put(pol);
2335
2336 return ret;
2337}
2338
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339static void sp_delete(struct shared_policy *sp, struct sp_node *n)
2340{
Paul Mundt140d5a42007-07-15 23:38:16 -07002341 pr_debug("deleting %lx-l%lx\n", n->start, n->end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 rb_erase(&n->nd, &sp->root);
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002343 sp_free(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344}
2345
Mel Gorman42288fe2012-12-21 23:10:25 +00002346static void sp_node_init(struct sp_node *node, unsigned long start,
2347 unsigned long end, struct mempolicy *pol)
2348{
2349 node->start = start;
2350 node->end = end;
2351 node->policy = pol;
2352}
2353
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07002354static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
2355 struct mempolicy *pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356{
KOSAKI Motohiro869833f2012-10-08 16:29:16 -07002357 struct sp_node *n;
2358 struct mempolicy *newpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359
KOSAKI Motohiro869833f2012-10-08 16:29:16 -07002360 n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 if (!n)
2362 return NULL;
KOSAKI Motohiro869833f2012-10-08 16:29:16 -07002363
2364 newpol = mpol_dup(pol);
2365 if (IS_ERR(newpol)) {
2366 kmem_cache_free(sn_cache, n);
2367 return NULL;
2368 }
2369 newpol->flags |= MPOL_F_SHARED;
Mel Gorman42288fe2012-12-21 23:10:25 +00002370 sp_node_init(n, start, end, newpol);
KOSAKI Motohiro869833f2012-10-08 16:29:16 -07002371
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 return n;
2373}
2374
2375/* Replace a policy range. */
2376static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
2377 unsigned long end, struct sp_node *new)
2378{
Mel Gormanb22d1272012-10-08 16:29:17 -07002379 struct sp_node *n;
Mel Gorman42288fe2012-12-21 23:10:25 +00002380 struct sp_node *n_new = NULL;
2381 struct mempolicy *mpol_new = NULL;
Mel Gormanb22d1272012-10-08 16:29:17 -07002382 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383
Mel Gorman42288fe2012-12-21 23:10:25 +00002384restart:
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002385 write_lock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 n = sp_lookup(sp, start, end);
2387 /* Take care of old policies in the same range. */
2388 while (n && n->start < end) {
2389 struct rb_node *next = rb_next(&n->nd);
2390 if (n->start >= start) {
2391 if (n->end <= end)
2392 sp_delete(sp, n);
2393 else
2394 n->start = end;
2395 } else {
2396 /* Old policy spanning whole new range. */
2397 if (n->end > end) {
Mel Gorman42288fe2012-12-21 23:10:25 +00002398 if (!n_new)
2399 goto alloc_new;
2400
2401 *mpol_new = *n->policy;
2402 atomic_set(&mpol_new->refcnt, 1);
KOSAKI Motohiro78806392013-03-08 12:43:29 -08002403 sp_node_init(n_new, end, n->end, mpol_new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 n->end = start;
Hillf Danton5ca39572013-03-08 12:43:28 -08002405 sp_insert(sp, n_new);
Mel Gorman42288fe2012-12-21 23:10:25 +00002406 n_new = NULL;
2407 mpol_new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 break;
2409 } else
2410 n->end = start;
2411 }
2412 if (!next)
2413 break;
2414 n = rb_entry(next, struct sp_node, nd);
2415 }
2416 if (new)
2417 sp_insert(sp, new);
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002418 write_unlock(&sp->lock);
Mel Gorman42288fe2012-12-21 23:10:25 +00002419 ret = 0;
2420
2421err_out:
2422 if (mpol_new)
2423 mpol_put(mpol_new);
2424 if (n_new)
2425 kmem_cache_free(sn_cache, n_new);
2426
Mel Gormanb22d1272012-10-08 16:29:17 -07002427 return ret;
Mel Gorman42288fe2012-12-21 23:10:25 +00002428
2429alloc_new:
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002430 write_unlock(&sp->lock);
Mel Gorman42288fe2012-12-21 23:10:25 +00002431 ret = -ENOMEM;
2432 n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL);
2433 if (!n_new)
2434 goto err_out;
2435 mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2436 if (!mpol_new)
2437 goto err_out;
2438 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439}
2440
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002441/**
2442 * mpol_shared_policy_init - initialize shared policy for inode
2443 * @sp: pointer to inode shared policy
2444 * @mpol: struct mempolicy to install
2445 *
2446 * Install non-NULL @mpol in inode's shared policy rb-tree.
2447 * On entry, the current task has a reference on a non-NULL @mpol.
2448 * This must be released on exit.
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002449 * This is called at get_inode() calls and we can use GFP_KERNEL.
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002450 */
2451void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
Robin Holt7339ff82006-01-14 13:20:48 -08002452{
Miao Xie58568d22009-06-16 15:31:49 -07002453 int ret;
2454
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002455 sp->root = RB_ROOT; /* empty tree == default mempolicy */
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002456 rwlock_init(&sp->lock);
Robin Holt7339ff82006-01-14 13:20:48 -08002457
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002458 if (mpol) {
2459 struct vm_area_struct pvma;
2460 struct mempolicy *new;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002461 NODEMASK_SCRATCH(scratch);
Robin Holt7339ff82006-01-14 13:20:48 -08002462
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002463 if (!scratch)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002464 goto put_mpol;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002465 /* contextualize the tmpfs mount point mempolicy */
2466 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002467 if (IS_ERR(new))
Dan Carpenter0cae3452010-05-25 23:42:58 -07002468 goto free_scratch; /* no valid nodemask intersection */
Miao Xie58568d22009-06-16 15:31:49 -07002469
2470 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002471 ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
Miao Xie58568d22009-06-16 15:31:49 -07002472 task_unlock(current);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002473 if (ret)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002474 goto put_new;
Robin Holt7339ff82006-01-14 13:20:48 -08002475
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002476 /* Create pseudo-vma that contains just the policy */
2477 memset(&pvma, 0, sizeof(struct vm_area_struct));
2478 pvma.vm_end = TASK_SIZE; /* policy covers entire file */
2479 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002480
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002481put_new:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002482 mpol_put(new); /* drop initial ref */
Dan Carpenter0cae3452010-05-25 23:42:58 -07002483free_scratch:
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002484 NODEMASK_SCRATCH_FREE(scratch);
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002485put_mpol:
2486 mpol_put(mpol); /* drop our incoming ref on sb mpol */
Robin Holt7339ff82006-01-14 13:20:48 -08002487 }
2488}
2489
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490int mpol_set_shared_policy(struct shared_policy *info,
2491 struct vm_area_struct *vma, struct mempolicy *npol)
2492{
2493 int err;
2494 struct sp_node *new = NULL;
2495 unsigned long sz = vma_pages(vma);
2496
David Rientjes028fec42008-04-28 02:12:25 -07002497 pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 vma->vm_pgoff,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002499 sz, npol ? npol->mode : -1,
David Rientjes028fec42008-04-28 02:12:25 -07002500 npol ? npol->flags : -1,
David Rientjes00ef2d22013-02-22 16:35:36 -08002501 npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502
2503 if (npol) {
2504 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
2505 if (!new)
2506 return -ENOMEM;
2507 }
2508 err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
2509 if (err && new)
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002510 sp_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 return err;
2512}
2513
2514/* Free a backing policy store on inode delete. */
2515void mpol_free_shared_policy(struct shared_policy *p)
2516{
2517 struct sp_node *n;
2518 struct rb_node *next;
2519
2520 if (!p->root.rb_node)
2521 return;
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002522 write_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 next = rb_first(&p->root);
2524 while (next) {
2525 n = rb_entry(next, struct sp_node, nd);
2526 next = rb_next(&n->nd);
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002527 sp_delete(p, n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 }
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002529 write_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530}
2531
Mel Gorman1a687c22012-11-22 11:16:36 +00002532#ifdef CONFIG_NUMA_BALANCING
Mel Gormanc2976632014-01-29 14:05:42 -08002533static int __initdata numabalancing_override;
Mel Gorman1a687c22012-11-22 11:16:36 +00002534
2535static void __init check_numabalancing_enable(void)
2536{
2537 bool numabalancing_default = false;
2538
2539 if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED))
2540 numabalancing_default = true;
2541
Mel Gormanc2976632014-01-29 14:05:42 -08002542 /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */
2543 if (numabalancing_override)
2544 set_numabalancing_state(numabalancing_override == 1);
2545
Mel Gormanb0dc2b92015-05-14 15:17:09 -07002546 if (num_online_nodes() > 1 && !numabalancing_override) {
Joe Perches756a0252016-03-17 14:19:47 -07002547 pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n",
Mel Gormanc2976632014-01-29 14:05:42 -08002548 numabalancing_default ? "Enabling" : "Disabling");
Mel Gorman1a687c22012-11-22 11:16:36 +00002549 set_numabalancing_state(numabalancing_default);
2550 }
2551}
2552
2553static int __init setup_numabalancing(char *str)
2554{
2555 int ret = 0;
2556 if (!str)
2557 goto out;
Mel Gorman1a687c22012-11-22 11:16:36 +00002558
2559 if (!strcmp(str, "enable")) {
Mel Gormanc2976632014-01-29 14:05:42 -08002560 numabalancing_override = 1;
Mel Gorman1a687c22012-11-22 11:16:36 +00002561 ret = 1;
2562 } else if (!strcmp(str, "disable")) {
Mel Gormanc2976632014-01-29 14:05:42 -08002563 numabalancing_override = -1;
Mel Gorman1a687c22012-11-22 11:16:36 +00002564 ret = 1;
2565 }
2566out:
2567 if (!ret)
Andrew Morton4a404be2014-01-29 14:05:43 -08002568 pr_warn("Unable to parse numa_balancing=\n");
Mel Gorman1a687c22012-11-22 11:16:36 +00002569
2570 return ret;
2571}
2572__setup("numa_balancing=", setup_numabalancing);
2573#else
2574static inline void __init check_numabalancing_enable(void)
2575{
2576}
2577#endif /* CONFIG_NUMA_BALANCING */
2578
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579/* assumes fs == KERNEL_DS */
2580void __init numa_policy_init(void)
2581{
Paul Mundtb71636e2007-07-15 23:38:15 -07002582 nodemask_t interleave_nodes;
2583 unsigned long largest = 0;
2584 int nid, prefer = 0;
2585
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 policy_cache = kmem_cache_create("numa_policy",
2587 sizeof(struct mempolicy),
Paul Mundt20c2df82007-07-20 10:11:58 +09002588 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589
2590 sn_cache = kmem_cache_create("shared_policy_node",
2591 sizeof(struct sp_node),
Paul Mundt20c2df82007-07-20 10:11:58 +09002592 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593
Mel Gorman5606e382012-11-02 18:19:13 +00002594 for_each_node(nid) {
2595 preferred_node_policy[nid] = (struct mempolicy) {
2596 .refcnt = ATOMIC_INIT(1),
2597 .mode = MPOL_PREFERRED,
2598 .flags = MPOL_F_MOF | MPOL_F_MORON,
2599 .v = { .preferred_node = nid, },
2600 };
2601 }
2602
Paul Mundtb71636e2007-07-15 23:38:15 -07002603 /*
2604 * Set interleaving policy for system init. Interleaving is only
2605 * enabled across suitably sized nodes (default is >= 16MB), or
2606 * fall back to the largest node if they're all smaller.
2607 */
2608 nodes_clear(interleave_nodes);
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002609 for_each_node_state(nid, N_MEMORY) {
Paul Mundtb71636e2007-07-15 23:38:15 -07002610 unsigned long total_pages = node_present_pages(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611
Paul Mundtb71636e2007-07-15 23:38:15 -07002612 /* Preserve the largest node */
2613 if (largest < total_pages) {
2614 largest = total_pages;
2615 prefer = nid;
2616 }
2617
2618 /* Interleave this node? */
2619 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
2620 node_set(nid, interleave_nodes);
2621 }
2622
2623 /* All too small, use the largest */
2624 if (unlikely(nodes_empty(interleave_nodes)))
2625 node_set(prefer, interleave_nodes);
2626
David Rientjes028fec42008-04-28 02:12:25 -07002627 if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07002628 pr_err("%s: interleaving failed\n", __func__);
Mel Gorman1a687c22012-11-22 11:16:36 +00002629
2630 check_numabalancing_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631}
2632
Christoph Lameter8bccd852005-10-29 18:16:59 -07002633/* Reset policy of current process to default */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634void numa_default_policy(void)
2635{
David Rientjes028fec42008-04-28 02:12:25 -07002636 do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637}
Paul Jackson68860ec2005-10-30 15:02:36 -08002638
Paul Jackson42253992006-01-08 01:01:59 -08002639/*
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002640 * Parse and format mempolicy from/to strings
2641 */
2642
2643/*
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002644 * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag.
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002645 */
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002646static const char * const policy_modes[] =
2647{
2648 [MPOL_DEFAULT] = "default",
2649 [MPOL_PREFERRED] = "prefer",
2650 [MPOL_BIND] = "bind",
2651 [MPOL_INTERLEAVE] = "interleave",
Lee Schermerhornd3a71032012-10-25 14:16:29 +02002652 [MPOL_LOCAL] = "local",
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002653};
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002654
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002655
2656#ifdef CONFIG_TMPFS
2657/**
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002658 * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option.
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002659 * @str: string containing mempolicy to parse
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002660 * @mpol: pointer to struct mempolicy pointer, returned on success.
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002661 *
2662 * Format of input:
2663 * <mode>[=<flags>][:<nodelist>]
2664 *
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002665 * On success, returns 0, else 1
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002666 */
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08002667int mpol_parse_str(char *str, struct mempolicy **mpol)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002668{
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002669 struct mempolicy *new = NULL;
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002670 unsigned short mode;
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002671 unsigned short mode_flags;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002672 nodemask_t nodes;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002673 char *nodelist = strchr(str, ':');
2674 char *flags = strchr(str, '=');
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002675 int err = 1;
2676
2677 if (nodelist) {
2678 /* NUL-terminate mode or flags string */
2679 *nodelist++ = '\0';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002680 if (nodelist_parse(nodelist, nodes))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002681 goto out;
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002682 if (!nodes_subset(nodes, node_states[N_MEMORY]))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002683 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002684 } else
2685 nodes_clear(nodes);
2686
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002687 if (flags)
2688 *flags++ = '\0'; /* terminate mode string */
2689
Peter Zijlstra479e2802012-10-25 14:16:28 +02002690 for (mode = 0; mode < MPOL_MAX; mode++) {
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002691 if (!strcmp(str, policy_modes[mode])) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002692 break;
2693 }
2694 }
Mel Gormana7200942012-11-16 09:37:58 +00002695 if (mode >= MPOL_MAX)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002696 goto out;
2697
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002698 switch (mode) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002699 case MPOL_PREFERRED:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002700 /*
2701 * Insist on a nodelist of one node only
2702 */
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002703 if (nodelist) {
2704 char *rest = nodelist;
2705 while (isdigit(*rest))
2706 rest++;
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002707 if (*rest)
2708 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002709 }
2710 break;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002711 case MPOL_INTERLEAVE:
2712 /*
2713 * Default to online nodes with memory if no nodelist
2714 */
2715 if (!nodelist)
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002716 nodes = node_states[N_MEMORY];
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002717 break;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002718 case MPOL_LOCAL:
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002719 /*
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002720 * Don't allow a nodelist; mpol_new() checks flags
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002721 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002722 if (nodelist)
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002723 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002724 mode = MPOL_PREFERRED;
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002725 break;
Ravikiran G Thirumalai413b43d2010-03-23 13:35:28 -07002726 case MPOL_DEFAULT:
2727 /*
2728 * Insist on a empty nodelist
2729 */
2730 if (!nodelist)
2731 err = 0;
2732 goto out;
KOSAKI Motohirod69b2e62010-03-23 13:35:30 -07002733 case MPOL_BIND:
2734 /*
2735 * Insist on a nodelist
2736 */
2737 if (!nodelist)
2738 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002739 }
2740
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002741 mode_flags = 0;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002742 if (flags) {
2743 /*
2744 * Currently, we only support two mutually exclusive
2745 * mode flags.
2746 */
2747 if (!strcmp(flags, "static"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002748 mode_flags |= MPOL_F_STATIC_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002749 else if (!strcmp(flags, "relative"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002750 mode_flags |= MPOL_F_RELATIVE_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002751 else
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002752 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002753 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002754
2755 new = mpol_new(mode, mode_flags, &nodes);
2756 if (IS_ERR(new))
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002757 goto out;
2758
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002759 /*
2760 * Save nodes for mpol_to_str() to show the tmpfs mount options
2761 * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo.
2762 */
2763 if (mode != MPOL_PREFERRED)
2764 new->v.nodes = nodes;
2765 else if (nodelist)
2766 new->v.preferred_node = first_node(nodes);
2767 else
2768 new->flags |= MPOL_F_LOCAL;
2769
2770 /*
2771 * Save nodes for contextualization: this will be used to "clone"
2772 * the mempolicy in a specific context [cpuset] at a later time.
2773 */
2774 new->w.user_nodemask = nodes;
2775
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002776 err = 0;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002777
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002778out:
2779 /* Restore string for error message */
2780 if (nodelist)
2781 *--nodelist = ':';
2782 if (flags)
2783 *--flags = '=';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002784 if (!err)
2785 *mpol = new;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002786 return err;
2787}
2788#endif /* CONFIG_TMPFS */
2789
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002790/**
2791 * mpol_to_str - format a mempolicy structure for printing
2792 * @buffer: to contain formatted mempolicy string
2793 * @maxlen: length of @buffer
2794 * @pol: pointer to mempolicy to be formatted
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002795 *
David Rientjes948927e2013-11-12 15:07:28 -08002796 * Convert @pol into a string. If @buffer is too short, truncate the string.
2797 * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the
2798 * longest flag, "relative", and to display at least a few node ids.
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002799 */
David Rientjes948927e2013-11-12 15:07:28 -08002800void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002801{
2802 char *p = buffer;
David Rientjes948927e2013-11-12 15:07:28 -08002803 nodemask_t nodes = NODE_MASK_NONE;
2804 unsigned short mode = MPOL_DEFAULT;
2805 unsigned short flags = 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002806
David Rientjes8790c712014-01-30 15:46:08 -08002807 if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002808 mode = pol->mode;
David Rientjes948927e2013-11-12 15:07:28 -08002809 flags = pol->flags;
2810 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002811
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002812 switch (mode) {
2813 case MPOL_DEFAULT:
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002814 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002815 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002816 if (flags & MPOL_F_LOCAL)
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002817 mode = MPOL_LOCAL;
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002818 else
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002819 node_set(pol->v.preferred_node, nodes);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002820 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002821 case MPOL_BIND:
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002822 case MPOL_INTERLEAVE:
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002823 nodes = pol->v.nodes;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002824 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002825 default:
David Rientjes948927e2013-11-12 15:07:28 -08002826 WARN_ON_ONCE(1);
2827 snprintf(p, maxlen, "unknown");
2828 return;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002829 }
2830
David Rientjesb7a9f422013-11-21 14:32:06 -08002831 p += snprintf(p, maxlen, "%s", policy_modes[mode]);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002832
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002833 if (flags & MPOL_MODE_FLAGS) {
David Rientjes948927e2013-11-12 15:07:28 -08002834 p += snprintf(p, buffer + maxlen - p, "=");
David Rientjesf5b087b2008-04-28 02:12:27 -07002835
Lee Schermerhorn22919902008-04-28 02:13:22 -07002836 /*
2837 * Currently, the only defined flags are mutually exclusive
2838 */
David Rientjesf5b087b2008-04-28 02:12:27 -07002839 if (flags & MPOL_F_STATIC_NODES)
Lee Schermerhorn22919902008-04-28 02:13:22 -07002840 p += snprintf(p, buffer + maxlen - p, "static");
2841 else if (flags & MPOL_F_RELATIVE_NODES)
2842 p += snprintf(p, buffer + maxlen - p, "relative");
David Rientjesf5b087b2008-04-28 02:12:27 -07002843 }
2844
Tejun Heo9e763e02015-02-13 14:38:02 -08002845 if (!nodes_empty(nodes))
2846 p += scnprintf(p, buffer + maxlen - p, ":%*pbl",
2847 nodemask_pr_args(&nodes));
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002848}