blob: 5cb5147235dfe9aa2c6b76b0b75c92cf934a0f03 [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
Yang Shid59794a2019-03-28 20:43:55 -0700550 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
551 if (!vma_migratable(vma))
552 break;
553 migrate_page_add(page, qp->pagelist, flags);
554 } else
555 break;
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800556 }
557 pte_unmap_unlock(pte - 1, ptl);
558 cond_resched();
Yang Shid59794a2019-03-28 20:43:55 -0700559 return addr != end ? -EIO : 0;
Hugh Dickins91612e02005-06-21 17:15:07 -0700560}
561
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800562static int queue_pages_hugetlb(pte_t *pte, unsigned long hmask,
563 unsigned long addr, unsigned long end,
564 struct mm_walk *walk)
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700565{
566#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800567 struct queue_pages *qp = walk->private;
568 unsigned long flags = qp->flags;
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700569 int nid;
570 struct page *page;
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800571 spinlock_t *ptl;
Naoya Horiguchid4c54912014-06-06 10:00:01 -0400572 pte_t entry;
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700573
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800574 ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte);
575 entry = huge_ptep_get(pte);
Naoya Horiguchid4c54912014-06-06 10:00:01 -0400576 if (!pte_present(entry))
577 goto unlock;
578 page = pte_page(entry);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700579 nid = page_to_nid(page);
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800580 if (node_isset(nid, *qp->nmask) == !!(flags & MPOL_MF_INVERT))
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700581 goto unlock;
582 /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */
583 if (flags & (MPOL_MF_MOVE_ALL) ||
584 (flags & MPOL_MF_MOVE && page_mapcount(page) == 1))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800585 isolate_huge_page(page, qp->pagelist);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700586unlock:
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800587 spin_unlock(ptl);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700588#else
589 BUG();
590#endif
Hugh Dickins91612e02005-06-21 17:15:07 -0700591 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592}
593
Aneesh Kumar K.V58772312013-12-06 00:08:22 +0530594#ifdef CONFIG_NUMA_BALANCING
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200595/*
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200596 * This is used to mark a range of virtual addresses to be inaccessible.
597 * These are later cleared by a NUMA hinting fault. Depending on these
598 * faults, pages may be migrated for better NUMA placement.
599 *
600 * This is assuming that NUMA faults are handled using PROT_NONE. If
601 * an architecture makes a different choice, it will need further
602 * changes to the core.
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200603 */
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200604unsigned long change_prot_numa(struct vm_area_struct *vma,
605 unsigned long addr, unsigned long end)
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200606{
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200607 int nr_updated;
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200608
Mel Gorman4d942462015-02-12 14:58:28 -0800609 nr_updated = change_protection(vma, addr, end, PAGE_NONE, 0, 1);
Mel Gorman03c5a6e2012-11-02 14:52:48 +0000610 if (nr_updated)
611 count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated);
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200612
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200613 return nr_updated;
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200614}
615#else
616static unsigned long change_prot_numa(struct vm_area_struct *vma,
617 unsigned long addr, unsigned long end)
618{
619 return 0;
620}
Aneesh Kumar K.V58772312013-12-06 00:08:22 +0530621#endif /* CONFIG_NUMA_BALANCING */
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200622
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800623static int queue_pages_test_walk(unsigned long start, unsigned long end,
624 struct mm_walk *walk)
625{
626 struct vm_area_struct *vma = walk->vma;
627 struct queue_pages *qp = walk->private;
628 unsigned long endvma = vma->vm_end;
629 unsigned long flags = qp->flags;
630
Yang Shid59794a2019-03-28 20:43:55 -0700631 /*
632 * Need check MPOL_MF_STRICT to return -EIO if possible
633 * regardless of vma_migratable
634 */
635 if (!vma_migratable(vma) &&
636 !(flags & MPOL_MF_STRICT))
Naoya Horiguchi48684a62015-02-11 15:28:06 -0800637 return 1;
638
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800639 if (endvma > end)
640 endvma = end;
641 if (vma->vm_start > start)
642 start = vma->vm_start;
643
644 if (!(flags & MPOL_MF_DISCONTIG_OK)) {
645 if (!vma->vm_next && vma->vm_end < end)
646 return -EFAULT;
647 if (qp->prev && qp->prev->vm_end < vma->vm_start)
648 return -EFAULT;
649 }
650
651 qp->prev = vma;
652
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800653 if (flags & MPOL_MF_LAZY) {
654 /* Similar to task_numa_work, skip inaccessible VMAs */
Liang Chen4355c012016-03-15 14:56:42 -0700655 if (!is_vm_hugetlb_page(vma) &&
656 (vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)) &&
657 !(vma->vm_flags & VM_MIXEDMAP))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800658 change_prot_numa(vma, start, endvma);
659 return 1;
660 }
661
Kirill A. Shutemov77bf45e2016-02-05 15:36:33 -0800662 /* queue pages from current vma */
Yang Shid59794a2019-03-28 20:43:55 -0700663 if (flags & MPOL_MF_VALID)
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800664 return 0;
665 return 1;
666}
667
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800668/*
Naoya Horiguchi98094942013-09-11 14:22:14 -0700669 * Walk through page tables and collect pages to be migrated.
670 *
671 * If pages found in a given range are on a set of nodes (determined by
672 * @nodes and @flags,) it's isolated and queued to the pagelist which is
673 * passed via @private.)
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800674 */
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -0700675static int
Naoya Horiguchi98094942013-09-11 14:22:14 -0700676queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800677 nodemask_t *nodes, unsigned long flags,
678 struct list_head *pagelist)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679{
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800680 struct queue_pages qp = {
681 .pagelist = pagelist,
682 .flags = flags,
683 .nmask = nodes,
684 .prev = NULL,
685 };
686 struct mm_walk queue_pages_walk = {
687 .hugetlb_entry = queue_pages_hugetlb,
688 .pmd_entry = queue_pages_pte_range,
689 .test_walk = queue_pages_test_walk,
690 .mm = mm,
691 .private = &qp,
692 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800694 return walk_page_range(start, end, &queue_pages_walk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695}
696
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700697/*
698 * Apply policy to a single VMA
699 * This must be called with the mmap_sem held for writing.
700 */
701static int vma_replace_policy(struct vm_area_struct *vma,
702 struct mempolicy *pol)
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700703{
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700704 int err;
705 struct mempolicy *old;
706 struct mempolicy *new;
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700707
708 pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
709 vma->vm_start, vma->vm_end, vma->vm_pgoff,
710 vma->vm_ops, vma->vm_file,
711 vma->vm_ops ? vma->vm_ops->set_policy : NULL);
712
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700713 new = mpol_dup(pol);
714 if (IS_ERR(new))
715 return PTR_ERR(new);
716
717 if (vma->vm_ops && vma->vm_ops->set_policy) {
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700718 err = vma->vm_ops->set_policy(vma, new);
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700719 if (err)
720 goto err_out;
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700721 }
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700722
723 old = vma->vm_policy;
724 vma->vm_policy = new; /* protected by mmap_sem */
725 mpol_put(old);
726
727 return 0;
728 err_out:
729 mpol_put(new);
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700730 return err;
731}
732
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733/* Step 2: apply policy to a range and do splits. */
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800734static int mbind_range(struct mm_struct *mm, unsigned long start,
735 unsigned long end, struct mempolicy *new_pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736{
737 struct vm_area_struct *next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800738 struct vm_area_struct *prev;
739 struct vm_area_struct *vma;
740 int err = 0;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800741 pgoff_t pgoff;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800742 unsigned long vmstart;
743 unsigned long vmend;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
Linus Torvalds097d5912012-03-06 18:23:36 -0800745 vma = find_vma(mm, start);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800746 if (!vma || vma->vm_start > start)
747 return -EFAULT;
748
Linus Torvalds097d5912012-03-06 18:23:36 -0800749 prev = vma->vm_prev;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800750 if (start > vma->vm_start)
751 prev = vma;
752
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800753 for (; vma && vma->vm_start < end; prev = vma, vma = next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 next = vma->vm_next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800755 vmstart = max(start, vma->vm_start);
756 vmend = min(end, vma->vm_end);
757
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800758 if (mpol_equal(vma_policy(vma), new_pol))
759 continue;
760
761 pgoff = vma->vm_pgoff +
762 ((vmstart - vma->vm_start) >> PAGE_SHIFT);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800763 prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700764 vma->anon_vma, vma->vm_file, pgoff,
765 new_pol, vma->vm_userfaultfd_ctx);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800766 if (prev) {
767 vma = prev;
768 next = vma->vm_next;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700769 if (mpol_equal(vma_policy(vma), new_pol))
770 continue;
771 /* vma_merge() joined vma && vma->next, case 8 */
772 goto replace;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800773 }
774 if (vma->vm_start != vmstart) {
775 err = split_vma(vma->vm_mm, vma, vmstart, 1);
776 if (err)
777 goto out;
778 }
779 if (vma->vm_end != vmend) {
780 err = split_vma(vma->vm_mm, vma, vmend, 0);
781 if (err)
782 goto out;
783 }
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700784 replace:
KOSAKI Motohiro869833f22012-10-08 16:29:16 -0700785 err = vma_replace_policy(vma, new_pol);
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700786 if (err)
787 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 }
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800789
790 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 return err;
792}
793
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794/* Set the process memory policy */
David Rientjes028fec42008-04-28 02:12:25 -0700795static long do_set_mempolicy(unsigned short mode, unsigned short flags,
796 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
Miao Xie58568d22009-06-16 15:31:49 -0700798 struct mempolicy *new, *old;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700799 NODEMASK_SCRATCH(scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700800 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700802 if (!scratch)
803 return -ENOMEM;
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700804
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700805 new = mpol_new(mode, flags, nodes);
806 if (IS_ERR(new)) {
807 ret = PTR_ERR(new);
808 goto out;
809 }
Oleg Nesterov2c7c3a72014-10-09 15:27:55 -0700810
Miao Xie58568d22009-06-16 15:31:49 -0700811 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700812 ret = mpol_set_nodemask(new, nodes, scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700813 if (ret) {
814 task_unlock(current);
Miao Xie58568d22009-06-16 15:31:49 -0700815 mpol_put(new);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700816 goto out;
Miao Xie58568d22009-06-16 15:31:49 -0700817 }
818 old = current->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 current->mempolicy = new;
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700820 if (new && new->mode == MPOL_INTERLEAVE &&
David Rientjesf5b087b2008-04-28 02:12:27 -0700821 nodes_weight(new->v.nodes))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700822 current->il_next = first_node(new->v.nodes);
Miao Xie58568d22009-06-16 15:31:49 -0700823 task_unlock(current);
Miao Xie58568d22009-06-16 15:31:49 -0700824 mpol_put(old);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700825 ret = 0;
826out:
827 NODEMASK_SCRATCH_FREE(scratch);
828 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829}
830
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700831/*
832 * Return nodemask for policy for get_mempolicy() query
Miao Xie58568d22009-06-16 15:31:49 -0700833 *
834 * Called with task's alloc_lock held
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700835 */
836static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837{
Andi Kleendfcd3c02005-10-29 18:15:48 -0700838 nodes_clear(*nodes);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700839 if (p == &default_policy)
840 return;
841
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700842 switch (p->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -0700843 case MPOL_BIND:
844 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -0700846 *nodes = p->v.nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 break;
848 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700849 if (!(p->flags & MPOL_F_LOCAL))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700850 node_set(p->v.preferred_node, *nodes);
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700851 /* else return empty node mask for local allocation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 break;
853 default:
854 BUG();
855 }
856}
857
Dave Hansend4edcf02016-02-12 13:01:56 -0800858static int lookup_node(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859{
860 struct page *p;
861 int err;
862
Lorenzo Stoakes768ae302016-10-13 01:20:16 +0100863 err = get_user_pages(addr & PAGE_MASK, 1, 0, &p, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 if (err >= 0) {
865 err = page_to_nid(p);
866 put_page(p);
867 }
868 return err;
869}
870
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871/* Retrieve NUMA policy */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700872static long do_get_mempolicy(int *policy, nodemask_t *nmask,
873 unsigned long addr, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874{
Christoph Lameter8bccd852005-10-29 18:16:59 -0700875 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 struct mm_struct *mm = current->mm;
877 struct vm_area_struct *vma = NULL;
878 struct mempolicy *pol = current->mempolicy;
879
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700880 if (flags &
881 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 return -EINVAL;
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700883
884 if (flags & MPOL_F_MEMS_ALLOWED) {
885 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
886 return -EINVAL;
887 *policy = 0; /* just so it's initialized */
Miao Xie58568d22009-06-16 15:31:49 -0700888 task_lock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700889 *nmask = cpuset_current_mems_allowed;
Miao Xie58568d22009-06-16 15:31:49 -0700890 task_unlock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700891 return 0;
892 }
893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 if (flags & MPOL_F_ADDR) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700895 /*
896 * Do NOT fall back to task policy if the
897 * vma/shared policy at addr is NULL. We
898 * want to return MPOL_DEFAULT in this case.
899 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 down_read(&mm->mmap_sem);
901 vma = find_vma_intersection(mm, addr, addr+1);
902 if (!vma) {
903 up_read(&mm->mmap_sem);
904 return -EFAULT;
905 }
906 if (vma->vm_ops && vma->vm_ops->get_policy)
907 pol = vma->vm_ops->get_policy(vma, addr);
908 else
909 pol = vma->vm_policy;
910 } else if (addr)
911 return -EINVAL;
912
913 if (!pol)
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700914 pol = &default_policy; /* indicates default behavior */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
916 if (flags & MPOL_F_NODE) {
917 if (flags & MPOL_F_ADDR) {
Dave Hansend4edcf02016-02-12 13:01:56 -0800918 err = lookup_node(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 if (err < 0)
920 goto out;
Christoph Lameter8bccd852005-10-29 18:16:59 -0700921 *policy = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 } else if (pol == current->mempolicy &&
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700923 pol->mode == MPOL_INTERLEAVE) {
Christoph Lameter8bccd852005-10-29 18:16:59 -0700924 *policy = current->il_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 } else {
926 err = -EINVAL;
927 goto out;
928 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700929 } else {
930 *policy = pol == &default_policy ? MPOL_DEFAULT :
931 pol->mode;
David Rientjesd79df632008-07-04 12:24:13 -0700932 /*
933 * Internal mempolicy flags must be masked off before exposing
934 * the policy to userspace.
935 */
936 *policy |= (pol->flags & MPOL_MODE_FLAGS);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700937 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 err = 0;
Miao Xie58568d22009-06-16 15:31:49 -0700940 if (nmask) {
Lee Schermerhornc6b6ef82010-03-23 13:35:41 -0700941 if (mpol_store_user_nodemask(pol)) {
942 *nmask = pol->w.user_nodemask;
943 } else {
944 task_lock(current);
945 get_policy_nodemask(pol, nmask);
946 task_unlock(current);
947 }
Miao Xie58568d22009-06-16 15:31:49 -0700948 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
950 out:
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700951 mpol_cond_put(pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 if (vma)
953 up_read(&current->mm->mmap_sem);
954 return err;
955}
956
Christoph Lameterb20a3502006-03-22 00:09:12 -0800957#ifdef CONFIG_MIGRATION
Christoph Lameter8bccd852005-10-29 18:16:59 -0700958/*
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800959 * page migration
960 */
Christoph Lameterfc301282006-01-18 17:42:29 -0800961static void migrate_page_add(struct page *page, struct list_head *pagelist,
962 unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800963{
964 /*
Christoph Lameterfc301282006-01-18 17:42:29 -0800965 * Avoid migrating a page that is shared with others.
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800966 */
Nick Piggin62695a82008-10-18 20:26:09 -0700967 if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) {
968 if (!isolate_lru_page(page)) {
969 list_add_tail(&page->lru, pagelist);
Mel Gorman599d0c92016-07-28 15:45:31 -0700970 inc_node_page_state(page, NR_ISOLATED_ANON +
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800971 page_is_file_cache(page));
Nick Piggin62695a82008-10-18 20:26:09 -0700972 }
973 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800974}
975
Christoph Lameter742755a2006-06-23 02:03:55 -0700976static struct page *new_node_page(struct page *page, unsigned long node, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700977{
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700978 if (PageHuge(page))
979 return alloc_huge_page_node(page_hstate(compound_head(page)),
980 node);
981 else
Vlastimil Babka96db8002015-09-08 15:03:50 -0700982 return __alloc_pages_node(node, GFP_HIGHUSER_MOVABLE |
David Rientjesb360edb2015-04-14 15:46:52 -0700983 __GFP_THISNODE, 0);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700984}
985
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800986/*
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800987 * Migrate pages from one node to a target node.
988 * Returns error or the number of pages not migrated.
989 */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700990static int migrate_to_node(struct mm_struct *mm, int source, int dest,
991 int flags)
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800992{
993 nodemask_t nmask;
994 LIST_HEAD(pagelist);
995 int err = 0;
996
997 nodes_clear(nmask);
998 node_set(source, nmask);
999
Minchan Kim08270802012-10-08 16:33:38 -07001000 /*
1001 * This does not "check" the range but isolates all pages that
1002 * need migration. Between passing in the full user address
1003 * space range and MPOL_MF_DISCONTIG_OK, this call can not fail.
1004 */
1005 VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)));
Naoya Horiguchi98094942013-09-11 14:22:14 -07001006 queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask,
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001007 flags | MPOL_MF_DISCONTIG_OK, &pagelist);
1008
Minchan Kimcf608ac2010-10-26 14:21:29 -07001009 if (!list_empty(&pagelist)) {
David Rientjes68711a72014-06-04 16:08:25 -07001010 err = migrate_pages(&pagelist, new_node_page, NULL, dest,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001011 MIGRATE_SYNC, MR_SYSCALL);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001012 if (err)
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -07001013 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001014 }
Christoph Lameter95a402c2006-06-23 02:03:53 -07001015
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001016 return err;
1017}
1018
1019/*
1020 * Move pages between the two nodesets so as to preserve the physical
1021 * layout as much as possible.
Christoph Lameter39743882006-01-08 01:00:51 -08001022 *
1023 * Returns the number of page that could not be moved.
1024 */
Andrew Morton0ce72d42012-05-29 15:06:24 -07001025int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1026 const nodemask_t *to, int flags)
Christoph Lameter39743882006-01-08 01:00:51 -08001027{
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001028 int busy = 0;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001029 int err;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001030 nodemask_t tmp;
Christoph Lameter39743882006-01-08 01:00:51 -08001031
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001032 err = migrate_prep();
1033 if (err)
1034 return err;
1035
Lee Schermerhorn53f25562008-04-28 02:13:20 -07001036 down_read(&mm->mmap_sem);
Christoph Lameter39743882006-01-08 01:00:51 -08001037
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -08001038 /*
1039 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
1040 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
1041 * bit in 'tmp', and return that <source, dest> pair for migration.
1042 * The pair of nodemasks 'to' and 'from' define the map.
1043 *
1044 * If no pair of bits is found that way, fallback to picking some
1045 * pair of 'source' and 'dest' bits that are not the same. If the
1046 * 'source' and 'dest' bits are the same, this represents a node
1047 * that will be migrating to itself, so no pages need move.
1048 *
1049 * If no bits are left in 'tmp', or if all remaining bits left
1050 * in 'tmp' correspond to the same bit in 'to', return false
1051 * (nothing left to migrate).
1052 *
1053 * This lets us pick a pair of nodes to migrate between, such that
1054 * if possible the dest node is not already occupied by some other
1055 * source node, minimizing the risk of overloading the memory on a
1056 * node that would happen if we migrated incoming memory to a node
1057 * before migrating outgoing memory source that same node.
1058 *
1059 * A single scan of tmp is sufficient. As we go, we remember the
1060 * most recent <s, d> pair that moved (s != d). If we find a pair
1061 * that not only moved, but what's better, moved to an empty slot
1062 * (d is not set in tmp), then we break out then, with that pair.
Justin P. Mattockae0e47f2011-03-01 15:06:02 +01001063 * Otherwise when we finish scanning from_tmp, we at least have the
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -08001064 * most recent <s, d> pair that moved. If we get all the way through
1065 * the scan of tmp without finding any node that moved, much less
1066 * moved to an empty node, then there is nothing left worth migrating.
1067 */
Christoph Lameterd4984712006-01-08 01:00:55 -08001068
Andrew Morton0ce72d42012-05-29 15:06:24 -07001069 tmp = *from;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001070 while (!nodes_empty(tmp)) {
1071 int s,d;
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001072 int source = NUMA_NO_NODE;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001073 int dest = 0;
1074
1075 for_each_node_mask(s, tmp) {
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001076
1077 /*
1078 * do_migrate_pages() tries to maintain the relative
1079 * node relationship of the pages established between
1080 * threads and memory areas.
1081 *
1082 * However if the number of source nodes is not equal to
1083 * the number of destination nodes we can not preserve
1084 * this node relative relationship. In that case, skip
1085 * copying memory from a node that is in the destination
1086 * mask.
1087 *
1088 * Example: [2,3,4] -> [3,4,5] moves everything.
1089 * [0-7] - > [3,4,5] moves only 0,1,2,6,7.
1090 */
1091
Andrew Morton0ce72d42012-05-29 15:06:24 -07001092 if ((nodes_weight(*from) != nodes_weight(*to)) &&
1093 (node_isset(s, *to)))
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001094 continue;
1095
Andrew Morton0ce72d42012-05-29 15:06:24 -07001096 d = node_remap(s, *from, *to);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001097 if (s == d)
1098 continue;
1099
1100 source = s; /* Node moved. Memorize */
1101 dest = d;
1102
1103 /* dest not in remaining from nodes? */
1104 if (!node_isset(dest, tmp))
1105 break;
1106 }
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001107 if (source == NUMA_NO_NODE)
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001108 break;
1109
1110 node_clear(source, tmp);
1111 err = migrate_to_node(mm, source, dest, flags);
1112 if (err > 0)
1113 busy += err;
1114 if (err < 0)
1115 break;
Christoph Lameter39743882006-01-08 01:00:51 -08001116 }
1117 up_read(&mm->mmap_sem);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001118 if (err < 0)
1119 return err;
1120 return busy;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001121
Christoph Lameter39743882006-01-08 01:00:51 -08001122}
1123
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001124/*
1125 * Allocate a new page for page migration based on vma policy.
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001126 * Start by assuming the page is mapped by the same vma as contains @start.
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001127 * Search forward from there, if not. N.B., this assumes that the
1128 * list of pages handed to migrate_pages()--which is how we get here--
1129 * is in virtual address order.
1130 */
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001131static struct page *new_page(struct page *page, unsigned long start, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001132{
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001133 struct vm_area_struct *vma;
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001134 unsigned long uninitialized_var(address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001135
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001136 vma = find_vma(current->mm, start);
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001137 while (vma) {
1138 address = page_address_in_vma(page, vma);
1139 if (address != -EFAULT)
1140 break;
1141 vma = vma->vm_next;
1142 }
1143
Wanpeng Li11c731e2013-12-18 17:08:56 -08001144 if (PageHuge(page)) {
Michal Hockocc817172014-01-23 15:53:15 -08001145 BUG_ON(!vma);
1146 return alloc_huge_page_noerr(vma, address, 1);
Wanpeng Li11c731e2013-12-18 17:08:56 -08001147 }
1148 /*
1149 * if !vma, alloc_page_vma() will use task or system default policy
1150 */
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -08001151 return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001152}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001153#else
1154
1155static void migrate_page_add(struct page *page, struct list_head *pagelist,
1156 unsigned long flags)
1157{
1158}
1159
Andrew Morton0ce72d42012-05-29 15:06:24 -07001160int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1161 const nodemask_t *to, int flags)
Christoph Lameterb20a3502006-03-22 00:09:12 -08001162{
1163 return -ENOSYS;
1164}
Christoph Lameter95a402c2006-06-23 02:03:53 -07001165
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001166static struct page *new_page(struct page *page, unsigned long start, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001167{
1168 return NULL;
1169}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001170#endif
1171
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001172static long do_mbind(unsigned long start, unsigned long len,
David Rientjes028fec42008-04-28 02:12:25 -07001173 unsigned short mode, unsigned short mode_flags,
1174 nodemask_t *nmask, unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001175{
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001176 struct mm_struct *mm = current->mm;
1177 struct mempolicy *new;
1178 unsigned long end;
1179 int err;
1180 LIST_HEAD(pagelist);
1181
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001182 if (flags & ~(unsigned long)MPOL_MF_VALID)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001183 return -EINVAL;
Christoph Lameter74c00242006-03-14 19:50:21 -08001184 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001185 return -EPERM;
1186
1187 if (start & ~PAGE_MASK)
1188 return -EINVAL;
1189
1190 if (mode == MPOL_DEFAULT)
1191 flags &= ~MPOL_MF_STRICT;
1192
1193 len = (len + PAGE_SIZE - 1) & PAGE_MASK;
1194 end = start + len;
1195
1196 if (end < start)
1197 return -EINVAL;
1198 if (end == start)
1199 return 0;
1200
David Rientjes028fec42008-04-28 02:12:25 -07001201 new = mpol_new(mode, mode_flags, nmask);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001202 if (IS_ERR(new))
1203 return PTR_ERR(new);
1204
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001205 if (flags & MPOL_MF_LAZY)
1206 new->flags |= MPOL_F_MOF;
1207
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001208 /*
1209 * If we are using the default policy then operation
1210 * on discontinuous address spaces is okay after all
1211 */
1212 if (!new)
1213 flags |= MPOL_MF_DISCONTIG_OK;
1214
David Rientjes028fec42008-04-28 02:12:25 -07001215 pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
1216 start, start + len, mode, mode_flags,
David Rientjes00ef2d22013-02-22 16:35:36 -08001217 nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001218
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001219 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
1220
1221 err = migrate_prep();
1222 if (err)
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001223 goto mpol_out;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001224 }
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07001225 {
1226 NODEMASK_SCRATCH(scratch);
1227 if (scratch) {
1228 down_write(&mm->mmap_sem);
1229 task_lock(current);
1230 err = mpol_set_nodemask(new, nmask, scratch);
1231 task_unlock(current);
1232 if (err)
1233 up_write(&mm->mmap_sem);
1234 } else
1235 err = -ENOMEM;
1236 NODEMASK_SCRATCH_FREE(scratch);
1237 }
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001238 if (err)
1239 goto mpol_out;
1240
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001241 err = queue_pages_range(mm, start, end, nmask,
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001242 flags | MPOL_MF_INVERT, &pagelist);
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001243 if (!err)
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -08001244 err = mbind_range(mm, start, end, new);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001245
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001246 if (!err) {
1247 int nr_failed = 0;
1248
Minchan Kimcf608ac2010-10-26 14:21:29 -07001249 if (!list_empty(&pagelist)) {
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001250 WARN_ON_ONCE(flags & MPOL_MF_LAZY);
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001251 nr_failed = migrate_pages(&pagelist, new_page, NULL,
1252 start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001253 if (nr_failed)
Naoya Horiguchi74060e42013-09-11 14:22:06 -07001254 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001255 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001256
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001257 if (nr_failed && (flags & MPOL_MF_STRICT))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001258 err = -EIO;
KOSAKI Motohiroab8a3e12009-10-26 16:49:58 -07001259 } else
Joonsoo Kimb0e5fd72013-12-18 17:08:51 -08001260 putback_movable_pages(&pagelist);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001261
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001262 up_write(&mm->mmap_sem);
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001263 mpol_out:
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001264 mpol_put(new);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001265 return err;
1266}
1267
Christoph Lameter39743882006-01-08 01:00:51 -08001268/*
Christoph Lameter8bccd852005-10-29 18:16:59 -07001269 * User space interface with variable sized bitmaps for nodelists.
1270 */
1271
1272/* Copy a node mask from user space. */
Christoph Lameter39743882006-01-08 01:00:51 -08001273static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
Christoph Lameter8bccd852005-10-29 18:16:59 -07001274 unsigned long maxnode)
1275{
1276 unsigned long k;
Yisheng Xie2851e3b2018-01-31 16:16:11 -08001277 unsigned long t;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001278 unsigned long nlongs;
1279 unsigned long endmask;
1280
1281 --maxnode;
1282 nodes_clear(*nodes);
1283 if (maxnode == 0 || !nmask)
1284 return 0;
Andi Kleena9c930b2006-02-20 18:27:59 -08001285 if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
Chris Wright636f13c2006-02-17 13:59:36 -08001286 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001287
1288 nlongs = BITS_TO_LONGS(maxnode);
1289 if ((maxnode % BITS_PER_LONG) == 0)
1290 endmask = ~0UL;
1291 else
1292 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
1293
Yisheng Xie2851e3b2018-01-31 16:16:11 -08001294 /*
1295 * When the user specified more nodes than supported just check
1296 * if the non supported part is all zero.
1297 *
1298 * If maxnode have more longs than MAX_NUMNODES, check
1299 * the bits in that area first. And then go through to
1300 * check the rest bits which equal or bigger than MAX_NUMNODES.
1301 * Otherwise, just check bits [MAX_NUMNODES, maxnode).
1302 */
Christoph Lameter8bccd852005-10-29 18:16:59 -07001303 if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
1304 if (nlongs > PAGE_SIZE/sizeof(long))
1305 return -EINVAL;
1306 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
Christoph Lameter8bccd852005-10-29 18:16:59 -07001307 if (get_user(t, nmask + k))
1308 return -EFAULT;
1309 if (k == nlongs - 1) {
1310 if (t & endmask)
1311 return -EINVAL;
1312 } else if (t)
1313 return -EINVAL;
1314 }
1315 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
1316 endmask = ~0UL;
1317 }
1318
Yisheng Xie2851e3b2018-01-31 16:16:11 -08001319 if (maxnode > MAX_NUMNODES && MAX_NUMNODES % BITS_PER_LONG != 0) {
1320 unsigned long valid_mask = endmask;
1321
1322 valid_mask &= ~((1UL << (MAX_NUMNODES % BITS_PER_LONG)) - 1);
1323 if (get_user(t, nmask + nlongs - 1))
1324 return -EFAULT;
1325 if (t & valid_mask)
1326 return -EINVAL;
1327 }
1328
Christoph Lameter8bccd852005-10-29 18:16:59 -07001329 if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
1330 return -EFAULT;
1331 nodes_addr(*nodes)[nlongs-1] &= endmask;
1332 return 0;
1333}
1334
1335/* Copy a kernel node mask to user space */
1336static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1337 nodemask_t *nodes)
1338{
1339 unsigned long copy = ALIGN(maxnode-1, 64) / 8;
Ralph Campbell17ef0852019-02-20 22:18:58 -08001340 unsigned int nbytes = BITS_TO_LONGS(nr_node_ids) * sizeof(long);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001341
1342 if (copy > nbytes) {
1343 if (copy > PAGE_SIZE)
1344 return -EINVAL;
1345 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1346 return -EFAULT;
1347 copy = nbytes;
1348 }
1349 return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1350}
1351
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001352SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
Rasmus Villemoesf7f28ca2014-06-04 16:07:57 -07001353 unsigned long, mode, const unsigned long __user *, nmask,
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001354 unsigned long, maxnode, unsigned, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001355{
1356 nodemask_t nodes;
1357 int err;
David Rientjes028fec42008-04-28 02:12:25 -07001358 unsigned short mode_flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001359
David Rientjes028fec42008-04-28 02:12:25 -07001360 mode_flags = mode & MPOL_MODE_FLAGS;
1361 mode &= ~MPOL_MODE_FLAGS;
David Rientjesa3b51e02008-04-28 02:12:23 -07001362 if (mode >= MPOL_MAX)
1363 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001364 if ((mode_flags & MPOL_F_STATIC_NODES) &&
1365 (mode_flags & MPOL_F_RELATIVE_NODES))
1366 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001367 err = get_nodes(&nodes, nmask, maxnode);
1368 if (err)
1369 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001370 return do_mbind(start, len, mode, mode_flags, &nodes, flags);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001371}
1372
1373/* Set the process memory policy */
Rasmus Villemoes23c89022014-06-04 16:07:58 -07001374SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask,
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001375 unsigned long, maxnode)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001376{
1377 int err;
1378 nodemask_t nodes;
David Rientjes028fec42008-04-28 02:12:25 -07001379 unsigned short flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001380
David Rientjes028fec42008-04-28 02:12:25 -07001381 flags = mode & MPOL_MODE_FLAGS;
1382 mode &= ~MPOL_MODE_FLAGS;
1383 if ((unsigned int)mode >= MPOL_MAX)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001384 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001385 if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
1386 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001387 err = get_nodes(&nodes, nmask, maxnode);
1388 if (err)
1389 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001390 return do_set_mempolicy(mode, flags, &nodes);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001391}
1392
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001393SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
1394 const unsigned long __user *, old_nodes,
1395 const unsigned long __user *, new_nodes)
Christoph Lameter39743882006-01-08 01:00:51 -08001396{
David Howellsc69e8d92008-11-14 10:39:19 +11001397 const struct cred *cred = current_cred(), *tcred;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001398 struct mm_struct *mm = NULL;
Christoph Lameter39743882006-01-08 01:00:51 -08001399 struct task_struct *task;
Christoph Lameter39743882006-01-08 01:00:51 -08001400 nodemask_t task_nodes;
1401 int err;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001402 nodemask_t *old;
1403 nodemask_t *new;
1404 NODEMASK_SCRATCH(scratch);
Christoph Lameter39743882006-01-08 01:00:51 -08001405
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001406 if (!scratch)
1407 return -ENOMEM;
Christoph Lameter39743882006-01-08 01:00:51 -08001408
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001409 old = &scratch->mask1;
1410 new = &scratch->mask2;
1411
1412 err = get_nodes(old, old_nodes, maxnode);
Christoph Lameter39743882006-01-08 01:00:51 -08001413 if (err)
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001414 goto out;
1415
1416 err = get_nodes(new, new_nodes, maxnode);
1417 if (err)
1418 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001419
1420 /* Find the mm_struct */
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001421 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001422 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter39743882006-01-08 01:00:51 -08001423 if (!task) {
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001424 rcu_read_unlock();
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001425 err = -ESRCH;
1426 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001427 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001428 get_task_struct(task);
Christoph Lameter39743882006-01-08 01:00:51 -08001429
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001430 err = -EINVAL;
Christoph Lameter39743882006-01-08 01:00:51 -08001431
1432 /*
1433 * Check if this process has the right to modify the specified
1434 * process. The right exists if the process has administrative
Alexey Dobriyan7f927fc2006-03-28 01:56:53 -08001435 * capabilities, superuser privileges or the same
Christoph Lameter39743882006-01-08 01:00:51 -08001436 * userid as the target process.
1437 */
David Howellsc69e8d92008-11-14 10:39:19 +11001438 tcred = __task_cred(task);
Eric W. Biedermanb38a86e2012-03-12 15:48:24 -07001439 if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1440 !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) &&
Christoph Lameter74c00242006-03-14 19:50:21 -08001441 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001442 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001443 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001444 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001445 }
David Howellsc69e8d92008-11-14 10:39:19 +11001446 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001447
1448 task_nodes = cpuset_mems_allowed(task);
1449 /* Is the user allowed to access the target nodes? */
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001450 if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
Christoph Lameter39743882006-01-08 01:00:51 -08001451 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001452 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001453 }
1454
Yisheng Xie4cf24632018-01-31 16:16:15 -08001455 task_nodes = cpuset_mems_allowed(current);
1456 nodes_and(*new, *new, task_nodes);
1457 if (nodes_empty(*new))
Christoph Lameter3268c632012-03-21 16:34:06 -07001458 goto out_put;
Yisheng Xie4cf24632018-01-31 16:16:15 -08001459
1460 nodes_and(*new, *new, node_states[N_MEMORY]);
1461 if (nodes_empty(*new))
1462 goto out_put;
Christoph Lameter3b42d282007-08-31 00:12:08 -07001463
David Quigley86c3a762006-06-23 02:04:02 -07001464 err = security_task_movememory(task);
1465 if (err)
Christoph Lameter3268c632012-03-21 16:34:06 -07001466 goto out_put;
David Quigley86c3a762006-06-23 02:04:02 -07001467
Christoph Lameter3268c632012-03-21 16:34:06 -07001468 mm = get_task_mm(task);
1469 put_task_struct(task);
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001470
1471 if (!mm) {
Christoph Lameter3268c632012-03-21 16:34:06 -07001472 err = -EINVAL;
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001473 goto out;
1474 }
1475
1476 err = do_migrate_pages(mm, old, new,
1477 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
Christoph Lameter3268c632012-03-21 16:34:06 -07001478
1479 mmput(mm);
1480out:
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001481 NODEMASK_SCRATCH_FREE(scratch);
1482
Christoph Lameter39743882006-01-08 01:00:51 -08001483 return err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001484
1485out_put:
1486 put_task_struct(task);
1487 goto out;
1488
Christoph Lameter39743882006-01-08 01:00:51 -08001489}
1490
1491
Christoph Lameter8bccd852005-10-29 18:16:59 -07001492/* Retrieve NUMA policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001493SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1494 unsigned long __user *, nmask, unsigned long, maxnode,
1495 unsigned long, addr, unsigned long, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001496{
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001497 int err;
1498 int uninitialized_var(pval);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001499 nodemask_t nodes;
1500
Ralph Campbell17ef0852019-02-20 22:18:58 -08001501 if (nmask != NULL && maxnode < nr_node_ids)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001502 return -EINVAL;
1503
1504 err = do_get_mempolicy(&pval, &nodes, addr, flags);
1505
1506 if (err)
1507 return err;
1508
1509 if (policy && put_user(pval, policy))
1510 return -EFAULT;
1511
1512 if (nmask)
1513 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1514
1515 return err;
1516}
1517
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518#ifdef CONFIG_COMPAT
1519
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001520COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1521 compat_ulong_t __user *, nmask,
1522 compat_ulong_t, maxnode,
1523 compat_ulong_t, addr, compat_ulong_t, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
1525 long err;
1526 unsigned long __user *nm = NULL;
1527 unsigned long nr_bits, alloc_size;
1528 DECLARE_BITMAP(bm, MAX_NUMNODES);
1529
Ralph Campbell17ef0852019-02-20 22:18:58 -08001530 nr_bits = min_t(unsigned long, maxnode-1, nr_node_ids);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1532
1533 if (nmask)
1534 nm = compat_alloc_user_space(alloc_size);
1535
1536 err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
1537
1538 if (!err && nmask) {
KAMEZAWA Hiroyuki2bbff6c2011-09-14 16:21:02 -07001539 unsigned long copy_size;
1540 copy_size = min_t(unsigned long, sizeof(bm), alloc_size);
1541 err = copy_from_user(bm, nm, copy_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 /* ensure entire bitmap is zeroed */
1543 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
1544 err |= compat_put_bitmap(nmask, bm, nr_bits);
1545 }
1546
1547 return err;
1548}
1549
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001550COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask,
1551 compat_ulong_t, maxnode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 unsigned long __user *nm = NULL;
1554 unsigned long nr_bits, alloc_size;
1555 DECLARE_BITMAP(bm, MAX_NUMNODES);
1556
1557 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1558 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1559
1560 if (nmask) {
Chris Sallscddab762017-04-07 23:48:11 -07001561 if (compat_get_bitmap(bm, nmask, nr_bits))
1562 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 nm = compat_alloc_user_space(alloc_size);
Chris Sallscddab762017-04-07 23:48:11 -07001564 if (copy_to_user(nm, bm, alloc_size))
1565 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 }
1567
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 return sys_set_mempolicy(mode, nm, nr_bits+1);
1569}
1570
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001571COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len,
1572 compat_ulong_t, mode, compat_ulong_t __user *, nmask,
1573 compat_ulong_t, maxnode, compat_ulong_t, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 unsigned long __user *nm = NULL;
1576 unsigned long nr_bits, alloc_size;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001577 nodemask_t bm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
1579 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1580 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1581
1582 if (nmask) {
Chris Sallscddab762017-04-07 23:48:11 -07001583 if (compat_get_bitmap(nodes_addr(bm), nmask, nr_bits))
1584 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 nm = compat_alloc_user_space(alloc_size);
Chris Sallscddab762017-04-07 23:48:11 -07001586 if (copy_to_user(nm, nodes_addr(bm), alloc_size))
1587 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 }
1589
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
1591}
1592
1593#endif
1594
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001595struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
1596 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597{
Oleg Nesterov8d902742014-10-09 15:27:45 -07001598 struct mempolicy *pol = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
1600 if (vma) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001601 if (vma->vm_ops && vma->vm_ops->get_policy) {
Oleg Nesterov8d902742014-10-09 15:27:45 -07001602 pol = vma->vm_ops->get_policy(vma, addr);
Mel Gorman00442ad2012-10-08 16:29:20 -07001603 } else if (vma->vm_policy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 pol = vma->vm_policy;
Mel Gorman00442ad2012-10-08 16:29:20 -07001605
1606 /*
1607 * shmem_alloc_page() passes MPOL_F_SHARED policy with
1608 * a pseudo vma whose vma->vm_ops=NULL. Take a reference
1609 * count on these policies which will be dropped by
1610 * mpol_cond_put() later
1611 */
1612 if (mpol_needs_cond_ref(pol))
1613 mpol_get(pol);
1614 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 }
Oleg Nesterovf15ca782014-10-09 15:27:43 -07001616
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001617 return pol;
1618}
1619
1620/*
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001621 * get_vma_policy(@vma, @addr)
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001622 * @vma: virtual memory area whose policy is sought
1623 * @addr: address in @vma for shared policy lookup
1624 *
1625 * Returns effective policy for a VMA at specified address.
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001626 * Falls back to current->mempolicy or system default policy, as necessary.
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001627 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1628 * count--added by the get_policy() vm_op, as appropriate--to protect against
1629 * freeing by another task. It is the caller's responsibility to free the
1630 * extra reference for shared policies.
1631 */
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001632static struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
1633 unsigned long addr)
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001634{
1635 struct mempolicy *pol = __get_vma_policy(vma, addr);
1636
Oleg Nesterov8d902742014-10-09 15:27:45 -07001637 if (!pol)
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001638 pol = get_task_policy(current);
Oleg Nesterov8d902742014-10-09 15:27:45 -07001639
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 return pol;
1641}
1642
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001643bool vma_policy_mof(struct vm_area_struct *vma)
Mel Gormanfc3147242013-10-07 11:29:09 +01001644{
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001645 struct mempolicy *pol;
Oleg Nesterovf15ca782014-10-09 15:27:43 -07001646
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001647 if (vma->vm_ops && vma->vm_ops->get_policy) {
1648 bool ret = false;
Mel Gormanfc3147242013-10-07 11:29:09 +01001649
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001650 pol = vma->vm_ops->get_policy(vma, vma->vm_start);
1651 if (pol && (pol->flags & MPOL_F_MOF))
1652 ret = true;
1653 mpol_cond_put(pol);
Mel Gormanfc3147242013-10-07 11:29:09 +01001654
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001655 return ret;
Mel Gormanfc3147242013-10-07 11:29:09 +01001656 }
1657
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001658 pol = vma->vm_policy;
Oleg Nesterov8d902742014-10-09 15:27:45 -07001659 if (!pol)
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001660 pol = get_task_policy(current);
Oleg Nesterov8d902742014-10-09 15:27:45 -07001661
Mel Gormanfc3147242013-10-07 11:29:09 +01001662 return pol->flags & MPOL_F_MOF;
1663}
1664
Lai Jiangshand3eb1572013-02-22 16:33:22 -08001665static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone)
1666{
1667 enum zone_type dynamic_policy_zone = policy_zone;
1668
1669 BUG_ON(dynamic_policy_zone == ZONE_MOVABLE);
1670
1671 /*
1672 * if policy->v.nodes has movable memory only,
1673 * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only.
1674 *
1675 * policy->v.nodes is intersect with node_states[N_MEMORY].
1676 * so if the following test faile, it implies
1677 * policy->v.nodes has movable memory only.
1678 */
1679 if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY]))
1680 dynamic_policy_zone = ZONE_MOVABLE;
1681
1682 return zone >= dynamic_policy_zone;
1683}
1684
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001685/*
1686 * Return a nodemask representing a mempolicy for filtering nodes for
1687 * page allocation
1688 */
1689static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
Mel Gorman19770b32008-04-28 02:12:18 -07001690{
1691 /* Lower zones don't get a nodemask applied for MPOL_BIND */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001692 if (unlikely(policy->mode == MPOL_BIND) &&
Lai Jiangshand3eb1572013-02-22 16:33:22 -08001693 apply_policy_zone(policy, gfp_zone(gfp)) &&
Mel Gorman19770b32008-04-28 02:12:18 -07001694 cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
1695 return &policy->v.nodes;
1696
1697 return NULL;
1698}
1699
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001700/* Return a zonelist indicated by gfp for node representing a mempolicy */
Andi Kleen2f5f9482011-03-04 17:36:29 -08001701static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy,
1702 int nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703{
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001704 switch (policy->mode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001706 if (!(policy->flags & MPOL_F_LOCAL))
1707 nd = policy->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 break;
1709 case MPOL_BIND:
Mel Gorman19770b32008-04-28 02:12:18 -07001710 /*
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001711 * Normally, MPOL_BIND allocations are node-local within the
1712 * allowed nodemask. However, if __GFP_THISNODE is set and the
Bob Liu6eb27e12010-05-24 14:32:00 -07001713 * current node isn't part of the mask, we use the zonelist for
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001714 * the first node in the mask instead.
Mel Gorman19770b32008-04-28 02:12:18 -07001715 */
Mel Gorman19770b32008-04-28 02:12:18 -07001716 if (unlikely(gfp & __GFP_THISNODE) &&
1717 unlikely(!node_isset(nd, policy->v.nodes)))
1718 nd = first_node(policy->v.nodes);
1719 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 BUG();
1722 }
Mel Gorman0e884602008-04-28 02:12:14 -07001723 return node_zonelist(nd, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724}
1725
1726/* Do dynamic interleaving for a process */
1727static unsigned interleave_nodes(struct mempolicy *policy)
1728{
1729 unsigned nid, next;
1730 struct task_struct *me = current;
1731
1732 nid = me->il_next;
Andrew Morton0edaf862016-05-19 17:10:58 -07001733 next = next_node_in(nid, policy->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001734 if (next < MAX_NUMNODES)
1735 me->il_next = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 return nid;
1737}
1738
Christoph Lameterdc85da12006-01-18 17:42:36 -08001739/*
1740 * Depending on the memory policy provide a node from which to allocate the
1741 * next slab entry.
1742 */
David Rientjes2a389612014-04-07 15:37:29 -07001743unsigned int mempolicy_slab_node(void)
Christoph Lameterdc85da12006-01-18 17:42:36 -08001744{
Andi Kleene7b691b2012-06-09 02:40:03 -07001745 struct mempolicy *policy;
David Rientjes2a389612014-04-07 15:37:29 -07001746 int node = numa_mem_id();
Andi Kleene7b691b2012-06-09 02:40:03 -07001747
1748 if (in_interrupt())
David Rientjes2a389612014-04-07 15:37:29 -07001749 return node;
Andi Kleene7b691b2012-06-09 02:40:03 -07001750
1751 policy = current->mempolicy;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001752 if (!policy || policy->flags & MPOL_F_LOCAL)
David Rientjes2a389612014-04-07 15:37:29 -07001753 return node;
Christoph Lameter765c4502006-09-27 01:50:08 -07001754
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001755 switch (policy->mode) {
1756 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001757 /*
1758 * handled MPOL_F_LOCAL above
1759 */
1760 return policy->v.preferred_node;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001761
Christoph Lameterdc85da12006-01-18 17:42:36 -08001762 case MPOL_INTERLEAVE:
1763 return interleave_nodes(policy);
1764
Mel Gormandd1a2392008-04-28 02:12:17 -07001765 case MPOL_BIND: {
Mel Gormanc33d6c02016-05-19 17:14:10 -07001766 struct zoneref *z;
1767
Christoph Lameterdc85da12006-01-18 17:42:36 -08001768 /*
1769 * Follow bind policy behavior and start allocation at the
1770 * first node.
1771 */
Mel Gorman19770b32008-04-28 02:12:18 -07001772 struct zonelist *zonelist;
Mel Gorman19770b32008-04-28 02:12:18 -07001773 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
Aneesh Kumar K.Vc9634cf2016-10-07 16:59:12 -07001774 zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK];
Mel Gormanc33d6c02016-05-19 17:14:10 -07001775 z = first_zones_zonelist(zonelist, highest_zoneidx,
1776 &policy->v.nodes);
1777 return z->zone ? z->zone->node : node;
Mel Gormandd1a2392008-04-28 02:12:17 -07001778 }
Christoph Lameterdc85da12006-01-18 17:42:36 -08001779
Christoph Lameterdc85da12006-01-18 17:42:36 -08001780 default:
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001781 BUG();
Christoph Lameterdc85da12006-01-18 17:42:36 -08001782 }
1783}
1784
Andrew Mortonfee83b32016-05-19 17:11:43 -07001785/*
1786 * Do static interleaving for a VMA with known offset @n. Returns the n'th
1787 * node in pol->v.nodes (starting from n=0), wrapping around if n exceeds the
1788 * number of present nodes.
1789 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790static unsigned offset_il_node(struct mempolicy *pol,
Andrew Mortonfee83b32016-05-19 17:11:43 -07001791 struct vm_area_struct *vma, unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792{
Andi Kleendfcd3c02005-10-29 18:15:48 -07001793 unsigned nnodes = nodes_weight(pol->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001794 unsigned target;
Andrew Mortonfee83b32016-05-19 17:11:43 -07001795 int i;
1796 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797
David Rientjesf5b087b2008-04-28 02:12:27 -07001798 if (!nnodes)
1799 return numa_node_id();
Andrew Mortonfee83b32016-05-19 17:11:43 -07001800 target = (unsigned int)n % nnodes;
1801 nid = first_node(pol->v.nodes);
1802 for (i = 0; i < target; i++)
Andi Kleendfcd3c02005-10-29 18:15:48 -07001803 nid = next_node(nid, pol->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 return nid;
1805}
1806
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001807/* Determine a node number for interleave */
1808static inline unsigned interleave_nid(struct mempolicy *pol,
1809 struct vm_area_struct *vma, unsigned long addr, int shift)
1810{
1811 if (vma) {
1812 unsigned long off;
1813
Nishanth Aravamudan3b98b082006-08-31 21:27:53 -07001814 /*
1815 * for small pages, there is no difference between
1816 * shift and PAGE_SHIFT, so the bit-shift is safe.
1817 * for huge pages, since vm_pgoff is in units of small
1818 * pages, we need to shift off the always 0 bits to get
1819 * a useful offset.
1820 */
1821 BUG_ON(shift < PAGE_SHIFT);
1822 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001823 off += (addr - vma->vm_start) >> shift;
1824 return offset_il_node(pol, vma, off);
1825 } else
1826 return interleave_nodes(pol);
1827}
1828
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001829#ifdef CONFIG_HUGETLBFS
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001830/*
1831 * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
Fabian Frederickb46e14a2014-06-04 16:08:18 -07001832 * @vma: virtual memory area whose policy is sought
1833 * @addr: address in @vma for shared policy lookup and interleave policy
1834 * @gfp_flags: for requested zone
1835 * @mpol: pointer to mempolicy pointer for reference counted mempolicy
1836 * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001837 *
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001838 * Returns a zonelist suitable for a huge page allocation and a pointer
1839 * to the struct mempolicy for conditional unref after allocation.
1840 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1841 * @nodemask for filtering the zonelist.
Miao Xiec0ff7452010-05-24 14:32:08 -07001842 *
Mel Gormand26914d2014-04-03 14:47:24 -07001843 * Must be protected by read_mems_allowed_begin()
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001844 */
Mel Gorman396faf02007-07-17 04:03:13 -07001845struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
Mel Gorman19770b32008-04-28 02:12:18 -07001846 gfp_t gfp_flags, struct mempolicy **mpol,
1847 nodemask_t **nodemask)
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001848{
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001849 struct zonelist *zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001850
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001851 *mpol = get_vma_policy(vma, addr);
Mel Gorman19770b32008-04-28 02:12:18 -07001852 *nodemask = NULL; /* assume !MPOL_BIND */
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001853
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001854 if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
1855 zl = node_zonelist(interleave_nid(*mpol, vma, addr,
Andi Kleena5516432008-07-23 21:27:41 -07001856 huge_page_shift(hstate_vma(vma))), gfp_flags);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001857 } else {
Andi Kleen2f5f9482011-03-04 17:36:29 -08001858 zl = policy_zonelist(gfp_flags, *mpol, numa_node_id());
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001859 if ((*mpol)->mode == MPOL_BIND)
1860 *nodemask = &(*mpol)->v.nodes;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001861 }
1862 return zl;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001863}
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001864
1865/*
1866 * init_nodemask_of_mempolicy
1867 *
1868 * If the current task's mempolicy is "default" [NULL], return 'false'
1869 * to indicate default policy. Otherwise, extract the policy nodemask
1870 * for 'bind' or 'interleave' policy into the argument nodemask, or
1871 * initialize the argument nodemask to contain the single node for
1872 * 'preferred' or 'local' policy and return 'true' to indicate presence
1873 * of non-default mempolicy.
1874 *
1875 * We don't bother with reference counting the mempolicy [mpol_get/put]
1876 * because the current task is examining it's own mempolicy and a task's
1877 * mempolicy is only ever changed by the task itself.
1878 *
1879 * N.B., it is the caller's responsibility to free a returned nodemask.
1880 */
1881bool init_nodemask_of_mempolicy(nodemask_t *mask)
1882{
1883 struct mempolicy *mempolicy;
1884 int nid;
1885
1886 if (!(mask && current->mempolicy))
1887 return false;
1888
Miao Xiec0ff7452010-05-24 14:32:08 -07001889 task_lock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001890 mempolicy = current->mempolicy;
1891 switch (mempolicy->mode) {
1892 case MPOL_PREFERRED:
1893 if (mempolicy->flags & MPOL_F_LOCAL)
1894 nid = numa_node_id();
1895 else
1896 nid = mempolicy->v.preferred_node;
1897 init_nodemask_of_node(mask, nid);
1898 break;
1899
1900 case MPOL_BIND:
1901 /* Fall through */
1902 case MPOL_INTERLEAVE:
1903 *mask = mempolicy->v.nodes;
1904 break;
1905
1906 default:
1907 BUG();
1908 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001909 task_unlock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001910
1911 return true;
1912}
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001913#endif
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001914
David Rientjes6f48d0eb2010-08-09 17:18:52 -07001915/*
1916 * mempolicy_nodemask_intersects
1917 *
1918 * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default
1919 * policy. Otherwise, check for intersection between mask and the policy
1920 * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local'
1921 * policy, always return true since it may allocate elsewhere on fallback.
1922 *
1923 * Takes task_lock(tsk) to prevent freeing of its mempolicy.
1924 */
1925bool mempolicy_nodemask_intersects(struct task_struct *tsk,
1926 const nodemask_t *mask)
1927{
1928 struct mempolicy *mempolicy;
1929 bool ret = true;
1930
1931 if (!mask)
1932 return ret;
1933 task_lock(tsk);
1934 mempolicy = tsk->mempolicy;
1935 if (!mempolicy)
1936 goto out;
1937
1938 switch (mempolicy->mode) {
1939 case MPOL_PREFERRED:
1940 /*
1941 * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to
1942 * allocate from, they may fallback to other nodes when oom.
1943 * Thus, it's possible for tsk to have allocated memory from
1944 * nodes in mask.
1945 */
1946 break;
1947 case MPOL_BIND:
1948 case MPOL_INTERLEAVE:
1949 ret = nodes_intersects(mempolicy->v.nodes, *mask);
1950 break;
1951 default:
1952 BUG();
1953 }
1954out:
1955 task_unlock(tsk);
1956 return ret;
1957}
1958
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959/* Allocate a page in interleaved policy.
1960 Own path because it needs to do special accounting. */
Andi Kleen662f3a02005-10-29 18:15:49 -07001961static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
1962 unsigned nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963{
1964 struct zonelist *zl;
1965 struct page *page;
1966
Mel Gorman0e884602008-04-28 02:12:14 -07001967 zl = node_zonelist(nid, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 page = __alloc_pages(gfp, order, zl);
Mel Gormandd1a2392008-04-28 02:12:17 -07001969 if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
Christoph Lameterca889e62006-06-30 01:55:44 -07001970 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 return page;
1972}
1973
1974/**
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001975 * alloc_pages_vma - Allocate a page for a VMA.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 *
1977 * @gfp:
1978 * %GFP_USER user allocation.
1979 * %GFP_KERNEL kernel allocations,
1980 * %GFP_HIGHMEM highmem/user allocations,
1981 * %GFP_FS allocation should not call back into a file system.
1982 * %GFP_ATOMIC don't sleep.
1983 *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001984 * @order:Order of the GFP allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 * @vma: Pointer to VMA or NULL if not available.
1986 * @addr: Virtual Address of the allocation. Must be inside the VMA.
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001987 * @node: Which node to prefer for allocation (modulo policy).
1988 * @hugepage: for hugepages try only the preferred node if possible
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 *
1990 * This function allocates a page from the kernel page pool and applies
1991 * a NUMA policy associated with the VMA or the current process.
1992 * When VMA is not NULL caller must hold down_read on the mmap_sem of the
1993 * mm_struct of the VMA to prevent it from going away. Should be used for
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001994 * all allocations for pages that will be mapped into user space. Returns
1995 * NULL when no page can be allocated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 */
1997struct page *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001998alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001999 unsigned long addr, int node, bool hugepage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000{
Mel Gormancc9a6c82012-03-21 16:34:11 -07002001 struct mempolicy *pol;
Miao Xiec0ff7452010-05-24 14:32:08 -07002002 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002003 unsigned int cpuset_mems_cookie;
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002004 struct zonelist *zl;
2005 nodemask_t *nmask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006
Mel Gormancc9a6c82012-03-21 16:34:11 -07002007retry_cpuset:
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07002008 pol = get_vma_policy(vma, addr);
Mel Gormand26914d2014-04-03 14:47:24 -07002009 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002010
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002011 if (pol->mode == MPOL_INTERLEAVE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 unsigned nid;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08002013
Andi Kleen8eac5632011-02-25 14:44:28 -08002014 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002015 mpol_cond_put(pol);
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08002016 page = alloc_page_interleave(gfp, order, nid);
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002017 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 }
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002019
Vlastimil Babka0867a572015-06-24 16:58:48 -07002020 if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) {
2021 int hpage_node = node;
2022
2023 /*
2024 * For hugepage allocation and non-interleave policy which
2025 * allows the current node (or other explicitly preferred
2026 * node) we only try to allocate from the current/preferred
2027 * node and don't fall back to other nodes, as the cost of
2028 * remote accesses would likely offset THP benefits.
2029 *
2030 * If the policy is interleave, or does not allow the current
2031 * node in its nodemask, we allocate the standard way.
2032 */
2033 if (pol->mode == MPOL_PREFERRED &&
2034 !(pol->flags & MPOL_F_LOCAL))
2035 hpage_node = pol->v.preferred_node;
2036
2037 nmask = policy_nodemask(gfp, pol);
2038 if (!nmask || node_isset(hpage_node, *nmask)) {
2039 mpol_cond_put(pol);
Andrea Arcangeli818e5842018-11-02 15:47:59 -07002040 /*
2041 * We cannot invoke reclaim if __GFP_THISNODE
2042 * is set. Invoking reclaim with
2043 * __GFP_THISNODE set, would cause THP
2044 * allocations to trigger heavy swapping
2045 * despite there may be tons of free memory
2046 * (including potentially plenty of THP
2047 * already available in the buddy) on all the
2048 * other NUMA nodes.
2049 *
2050 * At most we could invoke compaction when
2051 * __GFP_THISNODE is set (but we would need to
2052 * refrain from invoking reclaim even if
2053 * compaction returned COMPACT_SKIPPED because
2054 * there wasn't not enough memory to succeed
2055 * compaction). For now just avoid
2056 * __GFP_THISNODE instead of limiting the
2057 * allocation path to a strict and single
2058 * compaction invocation.
2059 *
2060 * Supposedly if direct reclaim was enabled by
2061 * the caller, the app prefers THP regardless
2062 * of the node it comes from so this would be
2063 * more desiderable behavior than only
2064 * providing THP originated from the local
2065 * node in such case.
2066 */
2067 if (!(gfp & __GFP_DIRECT_RECLAIM))
2068 gfp |= __GFP_THISNODE;
2069 page = __alloc_pages_node(hpage_node, gfp, order);
Vlastimil Babka0867a572015-06-24 16:58:48 -07002070 goto out;
2071 }
2072 }
2073
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002074 nmask = policy_nodemask(gfp, pol);
2075 zl = policy_zonelist(gfp, pol, node);
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002076 page = __alloc_pages_nodemask(gfp, order, zl, nmask);
Vlastimil Babka9b1a1ae2017-01-24 15:18:18 -08002077 mpol_cond_put(pol);
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002078out:
Mel Gormand26914d2014-04-03 14:47:24 -07002079 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002080 goto retry_cpuset;
Miao Xiec0ff7452010-05-24 14:32:08 -07002081 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082}
2083
2084/**
2085 * alloc_pages_current - Allocate pages.
2086 *
2087 * @gfp:
2088 * %GFP_USER user allocation,
2089 * %GFP_KERNEL kernel allocation,
2090 * %GFP_HIGHMEM highmem allocation,
2091 * %GFP_FS don't call back into a file system.
2092 * %GFP_ATOMIC don't sleep.
2093 * @order: Power of two of allocation size in pages. 0 is a single page.
2094 *
2095 * Allocate a page from the kernel page pool. When not in
2096 * interrupt context and apply the current process NUMA policy.
2097 * Returns NULL when no page can be allocated.
2098 *
Paul Jacksoncf2a473c2006-01-08 01:01:54 -08002099 * Don't call cpuset_update_task_memory_state() unless
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 * 1) it's ok to take cpuset_sem (can WAIT), and
2101 * 2) allocating for current task (not interrupt).
2102 */
Al Virodd0fc662005-10-07 07:46:04 +01002103struct page *alloc_pages_current(gfp_t gfp, unsigned order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104{
Oleg Nesterov8d902742014-10-09 15:27:45 -07002105 struct mempolicy *pol = &default_policy;
Miao Xiec0ff7452010-05-24 14:32:08 -07002106 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002107 unsigned int cpuset_mems_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108
Oleg Nesterov8d902742014-10-09 15:27:45 -07002109 if (!in_interrupt() && !(gfp & __GFP_THISNODE))
2110 pol = get_task_policy(current);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002111
Mel Gormancc9a6c82012-03-21 16:34:11 -07002112retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07002113 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002114
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002115 /*
2116 * No reference counting needed for current->mempolicy
2117 * nor system default_policy
2118 */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002119 if (pol->mode == MPOL_INTERLEAVE)
Miao Xiec0ff7452010-05-24 14:32:08 -07002120 page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
2121 else
2122 page = __alloc_pages_nodemask(gfp, order,
Andi Kleen5c4b4be2011-03-04 17:36:32 -08002123 policy_zonelist(gfp, pol, numa_node_id()),
2124 policy_nodemask(gfp, pol));
Mel Gormancc9a6c82012-03-21 16:34:11 -07002125
Mel Gormand26914d2014-04-03 14:47:24 -07002126 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002127 goto retry_cpuset;
2128
Miao Xiec0ff7452010-05-24 14:32:08 -07002129 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130}
2131EXPORT_SYMBOL(alloc_pages_current);
2132
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002133int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
2134{
2135 struct mempolicy *pol = mpol_dup(vma_policy(src));
2136
2137 if (IS_ERR(pol))
2138 return PTR_ERR(pol);
2139 dst->vm_policy = pol;
2140 return 0;
2141}
2142
Paul Jackson42253992006-01-08 01:01:59 -08002143/*
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002144 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
Paul Jackson42253992006-01-08 01:01:59 -08002145 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
2146 * with the mems_allowed returned by cpuset_mems_allowed(). This
2147 * keeps mempolicies cpuset relative after its cpuset moves. See
2148 * further kernel/cpuset.c update_nodemask().
Miao Xie708c1bb2010-05-24 14:32:07 -07002149 *
2150 * current's mempolicy may be rebinded by the other task(the task that changes
2151 * cpuset's mems), so we needn't do rebind work for current task.
Paul Jackson42253992006-01-08 01:01:59 -08002152 */
Paul Jackson42253992006-01-08 01:01:59 -08002153
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002154/* Slow path of a mempolicy duplicate */
2155struct mempolicy *__mpol_dup(struct mempolicy *old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156{
2157 struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2158
2159 if (!new)
2160 return ERR_PTR(-ENOMEM);
Miao Xie708c1bb2010-05-24 14:32:07 -07002161
2162 /* task's mempolicy is protected by alloc_lock */
2163 if (old == current->mempolicy) {
2164 task_lock(current);
2165 *new = *old;
2166 task_unlock(current);
2167 } else
2168 *new = *old;
2169
Paul Jackson42253992006-01-08 01:01:59 -08002170 if (current_cpuset_is_being_rebound()) {
2171 nodemask_t mems = cpuset_mems_allowed(current);
Miao Xie708c1bb2010-05-24 14:32:07 -07002172 if (new->flags & MPOL_F_REBINDING)
2173 mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2);
2174 else
2175 mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE);
Paul Jackson42253992006-01-08 01:01:59 -08002176 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 atomic_set(&new->refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 return new;
2179}
2180
2181/* Slow path of a mempolicy comparison */
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002182bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183{
2184 if (!a || !b)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002185 return false;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002186 if (a->mode != b->mode)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002187 return false;
Bob Liu19800502010-05-24 14:32:01 -07002188 if (a->flags != b->flags)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002189 return false;
Bob Liu19800502010-05-24 14:32:01 -07002190 if (mpol_store_user_nodemask(a))
2191 if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask))
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002192 return false;
Bob Liu19800502010-05-24 14:32:01 -07002193
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002194 switch (a->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -07002195 case MPOL_BIND:
2196 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 case MPOL_INTERLEAVE:
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002198 return !!nodes_equal(a->v.nodes, b->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 case MPOL_PREFERRED:
Yisheng Xiebe1a9d12018-03-22 16:17:02 -07002200 /* a's ->flags is the same as b's */
2201 if (a->flags & MPOL_F_LOCAL)
2202 return true;
Namhyung Kim75719662011-03-22 16:33:02 -07002203 return a->v.preferred_node == b->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 default:
2205 BUG();
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002206 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 }
2208}
2209
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 * Shared memory backing store policy support.
2212 *
2213 * Remember policies even when nobody has shared memory mapped.
2214 * The policies are kept in Red-Black tree linked from the inode.
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002215 * They are protected by the sp->lock rwlock, which should be held
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 * for any accesses to the tree.
2217 */
2218
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002219/*
2220 * lookup first element intersecting start-end. Caller holds sp->lock for
2221 * reading or for writing
2222 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223static struct sp_node *
2224sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
2225{
2226 struct rb_node *n = sp->root.rb_node;
2227
2228 while (n) {
2229 struct sp_node *p = rb_entry(n, struct sp_node, nd);
2230
2231 if (start >= p->end)
2232 n = n->rb_right;
2233 else if (end <= p->start)
2234 n = n->rb_left;
2235 else
2236 break;
2237 }
2238 if (!n)
2239 return NULL;
2240 for (;;) {
2241 struct sp_node *w = NULL;
2242 struct rb_node *prev = rb_prev(n);
2243 if (!prev)
2244 break;
2245 w = rb_entry(prev, struct sp_node, nd);
2246 if (w->end <= start)
2247 break;
2248 n = prev;
2249 }
2250 return rb_entry(n, struct sp_node, nd);
2251}
2252
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002253/*
2254 * Insert a new shared policy into the list. Caller holds sp->lock for
2255 * writing.
2256 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257static void sp_insert(struct shared_policy *sp, struct sp_node *new)
2258{
2259 struct rb_node **p = &sp->root.rb_node;
2260 struct rb_node *parent = NULL;
2261 struct sp_node *nd;
2262
2263 while (*p) {
2264 parent = *p;
2265 nd = rb_entry(parent, struct sp_node, nd);
2266 if (new->start < nd->start)
2267 p = &(*p)->rb_left;
2268 else if (new->end > nd->end)
2269 p = &(*p)->rb_right;
2270 else
2271 BUG();
2272 }
2273 rb_link_node(&new->nd, parent, p);
2274 rb_insert_color(&new->nd, &sp->root);
Paul Mundt140d5a42007-07-15 23:38:16 -07002275 pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002276 new->policy ? new->policy->mode : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277}
2278
2279/* Find shared policy intersecting idx */
2280struct mempolicy *
2281mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
2282{
2283 struct mempolicy *pol = NULL;
2284 struct sp_node *sn;
2285
2286 if (!sp->root.rb_node)
2287 return NULL;
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002288 read_lock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 sn = sp_lookup(sp, idx, idx+1);
2290 if (sn) {
2291 mpol_get(sn->policy);
2292 pol = sn->policy;
2293 }
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002294 read_unlock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 return pol;
2296}
2297
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002298static void sp_free(struct sp_node *n)
2299{
2300 mpol_put(n->policy);
2301 kmem_cache_free(sn_cache, n);
2302}
2303
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002304/**
2305 * mpol_misplaced - check whether current page node is valid in policy
2306 *
Fabian Frederickb46e14a2014-06-04 16:08:18 -07002307 * @page: page to be checked
2308 * @vma: vm area where page mapped
2309 * @addr: virtual address where page mapped
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002310 *
2311 * Lookup current policy node id for vma,addr and "compare to" page's
2312 * node id.
2313 *
2314 * Returns:
2315 * -1 - not misplaced, page is in the right node
2316 * node - node id where the page should be
2317 *
2318 * Policy determination "mimics" alloc_page_vma().
2319 * Called from fault path where we know the vma and faulting address.
2320 */
2321int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr)
2322{
2323 struct mempolicy *pol;
Mel Gormanc33d6c02016-05-19 17:14:10 -07002324 struct zoneref *z;
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002325 int curnid = page_to_nid(page);
2326 unsigned long pgoff;
Peter Zijlstra90572892013-10-07 11:29:20 +01002327 int thiscpu = raw_smp_processor_id();
2328 int thisnid = cpu_to_node(thiscpu);
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002329 int polnid = -1;
2330 int ret = -1;
2331
2332 BUG_ON(!vma);
2333
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07002334 pol = get_vma_policy(vma, addr);
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002335 if (!(pol->flags & MPOL_F_MOF))
2336 goto out;
2337
2338 switch (pol->mode) {
2339 case MPOL_INTERLEAVE:
2340 BUG_ON(addr >= vma->vm_end);
2341 BUG_ON(addr < vma->vm_start);
2342
2343 pgoff = vma->vm_pgoff;
2344 pgoff += (addr - vma->vm_start) >> PAGE_SHIFT;
2345 polnid = offset_il_node(pol, vma, pgoff);
2346 break;
2347
2348 case MPOL_PREFERRED:
2349 if (pol->flags & MPOL_F_LOCAL)
2350 polnid = numa_node_id();
2351 else
2352 polnid = pol->v.preferred_node;
2353 break;
2354
2355 case MPOL_BIND:
Mel Gormanc33d6c02016-05-19 17:14:10 -07002356
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002357 /*
2358 * allows binding to multiple nodes.
2359 * use current page if in policy nodemask,
2360 * else select nearest allowed node, if any.
2361 * If no allowed nodes, use current [!misplaced].
2362 */
2363 if (node_isset(curnid, pol->v.nodes))
2364 goto out;
Mel Gormanc33d6c02016-05-19 17:14:10 -07002365 z = first_zones_zonelist(
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002366 node_zonelist(numa_node_id(), GFP_HIGHUSER),
2367 gfp_zone(GFP_HIGHUSER),
Mel Gormanc33d6c02016-05-19 17:14:10 -07002368 &pol->v.nodes);
2369 polnid = z->zone->node;
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002370 break;
2371
2372 default:
2373 BUG();
2374 }
Mel Gorman5606e382012-11-02 18:19:13 +00002375
2376 /* Migrate the page towards the node whose CPU is referencing it */
Mel Gormane42c8ff2012-11-12 09:17:07 +00002377 if (pol->flags & MPOL_F_MORON) {
Peter Zijlstra90572892013-10-07 11:29:20 +01002378 polnid = thisnid;
Mel Gorman5606e382012-11-02 18:19:13 +00002379
Rik van Riel10f39042014-01-27 17:03:44 -05002380 if (!should_numa_migrate_memory(current, page, curnid, thiscpu))
Rik van Rielde1c9ce2013-10-07 11:29:39 +01002381 goto out;
Mel Gormane42c8ff2012-11-12 09:17:07 +00002382 }
2383
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002384 if (curnid != polnid)
2385 ret = polnid;
2386out:
2387 mpol_cond_put(pol);
2388
2389 return ret;
2390}
2391
David Rientjesc11600e2016-09-01 16:15:07 -07002392/*
2393 * Drop the (possibly final) reference to task->mempolicy. It needs to be
2394 * dropped after task->mempolicy is set to NULL so that any allocation done as
2395 * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed
2396 * policy.
2397 */
2398void mpol_put_task_policy(struct task_struct *task)
2399{
2400 struct mempolicy *pol;
2401
2402 task_lock(task);
2403 pol = task->mempolicy;
2404 task->mempolicy = NULL;
2405 task_unlock(task);
2406 mpol_put(pol);
2407}
2408
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409static void sp_delete(struct shared_policy *sp, struct sp_node *n)
2410{
Paul Mundt140d5a42007-07-15 23:38:16 -07002411 pr_debug("deleting %lx-l%lx\n", n->start, n->end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 rb_erase(&n->nd, &sp->root);
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002413 sp_free(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414}
2415
Mel Gorman42288fe2012-12-21 23:10:25 +00002416static void sp_node_init(struct sp_node *node, unsigned long start,
2417 unsigned long end, struct mempolicy *pol)
2418{
2419 node->start = start;
2420 node->end = end;
2421 node->policy = pol;
2422}
2423
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07002424static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
2425 struct mempolicy *pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426{
KOSAKI Motohiro869833f22012-10-08 16:29:16 -07002427 struct sp_node *n;
2428 struct mempolicy *newpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429
KOSAKI Motohiro869833f22012-10-08 16:29:16 -07002430 n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 if (!n)
2432 return NULL;
KOSAKI Motohiro869833f22012-10-08 16:29:16 -07002433
2434 newpol = mpol_dup(pol);
2435 if (IS_ERR(newpol)) {
2436 kmem_cache_free(sn_cache, n);
2437 return NULL;
2438 }
2439 newpol->flags |= MPOL_F_SHARED;
Mel Gorman42288fe2012-12-21 23:10:25 +00002440 sp_node_init(n, start, end, newpol);
KOSAKI Motohiro869833f22012-10-08 16:29:16 -07002441
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 return n;
2443}
2444
2445/* Replace a policy range. */
2446static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
2447 unsigned long end, struct sp_node *new)
2448{
Mel Gormanb22d1272012-10-08 16:29:17 -07002449 struct sp_node *n;
Mel Gorman42288fe2012-12-21 23:10:25 +00002450 struct sp_node *n_new = NULL;
2451 struct mempolicy *mpol_new = NULL;
Mel Gormanb22d1272012-10-08 16:29:17 -07002452 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453
Mel Gorman42288fe2012-12-21 23:10:25 +00002454restart:
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002455 write_lock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 n = sp_lookup(sp, start, end);
2457 /* Take care of old policies in the same range. */
2458 while (n && n->start < end) {
2459 struct rb_node *next = rb_next(&n->nd);
2460 if (n->start >= start) {
2461 if (n->end <= end)
2462 sp_delete(sp, n);
2463 else
2464 n->start = end;
2465 } else {
2466 /* Old policy spanning whole new range. */
2467 if (n->end > end) {
Mel Gorman42288fe2012-12-21 23:10:25 +00002468 if (!n_new)
2469 goto alloc_new;
2470
2471 *mpol_new = *n->policy;
2472 atomic_set(&mpol_new->refcnt, 1);
KOSAKI Motohiro78806392013-03-08 12:43:29 -08002473 sp_node_init(n_new, end, n->end, mpol_new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 n->end = start;
Hillf Danton5ca39572013-03-08 12:43:28 -08002475 sp_insert(sp, n_new);
Mel Gorman42288fe2012-12-21 23:10:25 +00002476 n_new = NULL;
2477 mpol_new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 break;
2479 } else
2480 n->end = start;
2481 }
2482 if (!next)
2483 break;
2484 n = rb_entry(next, struct sp_node, nd);
2485 }
2486 if (new)
2487 sp_insert(sp, new);
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002488 write_unlock(&sp->lock);
Mel Gorman42288fe2012-12-21 23:10:25 +00002489 ret = 0;
2490
2491err_out:
2492 if (mpol_new)
2493 mpol_put(mpol_new);
2494 if (n_new)
2495 kmem_cache_free(sn_cache, n_new);
2496
Mel Gormanb22d1272012-10-08 16:29:17 -07002497 return ret;
Mel Gorman42288fe2012-12-21 23:10:25 +00002498
2499alloc_new:
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002500 write_unlock(&sp->lock);
Mel Gorman42288fe2012-12-21 23:10:25 +00002501 ret = -ENOMEM;
2502 n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL);
2503 if (!n_new)
2504 goto err_out;
2505 mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2506 if (!mpol_new)
2507 goto err_out;
2508 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509}
2510
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002511/**
2512 * mpol_shared_policy_init - initialize shared policy for inode
2513 * @sp: pointer to inode shared policy
2514 * @mpol: struct mempolicy to install
2515 *
2516 * Install non-NULL @mpol in inode's shared policy rb-tree.
2517 * On entry, the current task has a reference on a non-NULL @mpol.
2518 * This must be released on exit.
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002519 * This is called at get_inode() calls and we can use GFP_KERNEL.
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002520 */
2521void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
Robin Holt7339ff82006-01-14 13:20:48 -08002522{
Miao Xie58568d22009-06-16 15:31:49 -07002523 int ret;
2524
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002525 sp->root = RB_ROOT; /* empty tree == default mempolicy */
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002526 rwlock_init(&sp->lock);
Robin Holt7339ff82006-01-14 13:20:48 -08002527
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002528 if (mpol) {
2529 struct vm_area_struct pvma;
2530 struct mempolicy *new;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002531 NODEMASK_SCRATCH(scratch);
Robin Holt7339ff82006-01-14 13:20:48 -08002532
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002533 if (!scratch)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002534 goto put_mpol;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002535 /* contextualize the tmpfs mount point mempolicy */
2536 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002537 if (IS_ERR(new))
Dan Carpenter0cae3452010-05-25 23:42:58 -07002538 goto free_scratch; /* no valid nodemask intersection */
Miao Xie58568d22009-06-16 15:31:49 -07002539
2540 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002541 ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
Miao Xie58568d22009-06-16 15:31:49 -07002542 task_unlock(current);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002543 if (ret)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002544 goto put_new;
Robin Holt7339ff82006-01-14 13:20:48 -08002545
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002546 /* Create pseudo-vma that contains just the policy */
2547 memset(&pvma, 0, sizeof(struct vm_area_struct));
2548 pvma.vm_end = TASK_SIZE; /* policy covers entire file */
2549 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002550
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002551put_new:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002552 mpol_put(new); /* drop initial ref */
Dan Carpenter0cae3452010-05-25 23:42:58 -07002553free_scratch:
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002554 NODEMASK_SCRATCH_FREE(scratch);
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002555put_mpol:
2556 mpol_put(mpol); /* drop our incoming ref on sb mpol */
Robin Holt7339ff82006-01-14 13:20:48 -08002557 }
2558}
2559
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560int mpol_set_shared_policy(struct shared_policy *info,
2561 struct vm_area_struct *vma, struct mempolicy *npol)
2562{
2563 int err;
2564 struct sp_node *new = NULL;
2565 unsigned long sz = vma_pages(vma);
2566
David Rientjes028fec42008-04-28 02:12:25 -07002567 pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 vma->vm_pgoff,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002569 sz, npol ? npol->mode : -1,
David Rientjes028fec42008-04-28 02:12:25 -07002570 npol ? npol->flags : -1,
David Rientjes00ef2d22013-02-22 16:35:36 -08002571 npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572
2573 if (npol) {
2574 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
2575 if (!new)
2576 return -ENOMEM;
2577 }
2578 err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
2579 if (err && new)
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002580 sp_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 return err;
2582}
2583
2584/* Free a backing policy store on inode delete. */
2585void mpol_free_shared_policy(struct shared_policy *p)
2586{
2587 struct sp_node *n;
2588 struct rb_node *next;
2589
2590 if (!p->root.rb_node)
2591 return;
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002592 write_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 next = rb_first(&p->root);
2594 while (next) {
2595 n = rb_entry(next, struct sp_node, nd);
2596 next = rb_next(&n->nd);
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002597 sp_delete(p, n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598 }
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002599 write_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600}
2601
Mel Gorman1a687c22012-11-22 11:16:36 +00002602#ifdef CONFIG_NUMA_BALANCING
Mel Gormanc2976632014-01-29 14:05:42 -08002603static int __initdata numabalancing_override;
Mel Gorman1a687c22012-11-22 11:16:36 +00002604
2605static void __init check_numabalancing_enable(void)
2606{
2607 bool numabalancing_default = false;
2608
2609 if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED))
2610 numabalancing_default = true;
2611
Mel Gormanc2976632014-01-29 14:05:42 -08002612 /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */
2613 if (numabalancing_override)
2614 set_numabalancing_state(numabalancing_override == 1);
2615
Mel Gormanb0dc2b92015-05-14 15:17:09 -07002616 if (num_online_nodes() > 1 && !numabalancing_override) {
Joe Perches756a025f02016-03-17 14:19:47 -07002617 pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n",
Mel Gormanc2976632014-01-29 14:05:42 -08002618 numabalancing_default ? "Enabling" : "Disabling");
Mel Gorman1a687c22012-11-22 11:16:36 +00002619 set_numabalancing_state(numabalancing_default);
2620 }
2621}
2622
2623static int __init setup_numabalancing(char *str)
2624{
2625 int ret = 0;
2626 if (!str)
2627 goto out;
Mel Gorman1a687c22012-11-22 11:16:36 +00002628
2629 if (!strcmp(str, "enable")) {
Mel Gormanc2976632014-01-29 14:05:42 -08002630 numabalancing_override = 1;
Mel Gorman1a687c22012-11-22 11:16:36 +00002631 ret = 1;
2632 } else if (!strcmp(str, "disable")) {
Mel Gormanc2976632014-01-29 14:05:42 -08002633 numabalancing_override = -1;
Mel Gorman1a687c22012-11-22 11:16:36 +00002634 ret = 1;
2635 }
2636out:
2637 if (!ret)
Andrew Morton4a404be2014-01-29 14:05:43 -08002638 pr_warn("Unable to parse numa_balancing=\n");
Mel Gorman1a687c22012-11-22 11:16:36 +00002639
2640 return ret;
2641}
2642__setup("numa_balancing=", setup_numabalancing);
2643#else
2644static inline void __init check_numabalancing_enable(void)
2645{
2646}
2647#endif /* CONFIG_NUMA_BALANCING */
2648
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649/* assumes fs == KERNEL_DS */
2650void __init numa_policy_init(void)
2651{
Paul Mundtb71636e2007-07-15 23:38:15 -07002652 nodemask_t interleave_nodes;
2653 unsigned long largest = 0;
2654 int nid, prefer = 0;
2655
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 policy_cache = kmem_cache_create("numa_policy",
2657 sizeof(struct mempolicy),
Paul Mundt20c2df82007-07-20 10:11:58 +09002658 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659
2660 sn_cache = kmem_cache_create("shared_policy_node",
2661 sizeof(struct sp_node),
Paul Mundt20c2df82007-07-20 10:11:58 +09002662 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663
Mel Gorman5606e382012-11-02 18:19:13 +00002664 for_each_node(nid) {
2665 preferred_node_policy[nid] = (struct mempolicy) {
2666 .refcnt = ATOMIC_INIT(1),
2667 .mode = MPOL_PREFERRED,
2668 .flags = MPOL_F_MOF | MPOL_F_MORON,
2669 .v = { .preferred_node = nid, },
2670 };
2671 }
2672
Paul Mundtb71636e2007-07-15 23:38:15 -07002673 /*
2674 * Set interleaving policy for system init. Interleaving is only
2675 * enabled across suitably sized nodes (default is >= 16MB), or
2676 * fall back to the largest node if they're all smaller.
2677 */
2678 nodes_clear(interleave_nodes);
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002679 for_each_node_state(nid, N_MEMORY) {
Paul Mundtb71636e2007-07-15 23:38:15 -07002680 unsigned long total_pages = node_present_pages(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681
Paul Mundtb71636e2007-07-15 23:38:15 -07002682 /* Preserve the largest node */
2683 if (largest < total_pages) {
2684 largest = total_pages;
2685 prefer = nid;
2686 }
2687
2688 /* Interleave this node? */
2689 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
2690 node_set(nid, interleave_nodes);
2691 }
2692
2693 /* All too small, use the largest */
2694 if (unlikely(nodes_empty(interleave_nodes)))
2695 node_set(prefer, interleave_nodes);
2696
David Rientjes028fec42008-04-28 02:12:25 -07002697 if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07002698 pr_err("%s: interleaving failed\n", __func__);
Mel Gorman1a687c22012-11-22 11:16:36 +00002699
2700 check_numabalancing_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701}
2702
Christoph Lameter8bccd852005-10-29 18:16:59 -07002703/* Reset policy of current process to default */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704void numa_default_policy(void)
2705{
David Rientjes028fec42008-04-28 02:12:25 -07002706 do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707}
Paul Jackson68860ec2005-10-30 15:02:36 -08002708
Paul Jackson42253992006-01-08 01:01:59 -08002709/*
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002710 * Parse and format mempolicy from/to strings
2711 */
2712
2713/*
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002714 * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag.
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002715 */
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002716static const char * const policy_modes[] =
2717{
2718 [MPOL_DEFAULT] = "default",
2719 [MPOL_PREFERRED] = "prefer",
2720 [MPOL_BIND] = "bind",
2721 [MPOL_INTERLEAVE] = "interleave",
Lee Schermerhornd3a71032012-10-25 14:16:29 +02002722 [MPOL_LOCAL] = "local",
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002723};
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002724
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002725
2726#ifdef CONFIG_TMPFS
2727/**
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002728 * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option.
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002729 * @str: string containing mempolicy to parse
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002730 * @mpol: pointer to struct mempolicy pointer, returned on success.
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002731 *
2732 * Format of input:
2733 * <mode>[=<flags>][:<nodelist>]
2734 *
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002735 * On success, returns 0, else 1
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002736 */
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08002737int mpol_parse_str(char *str, struct mempolicy **mpol)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002738{
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002739 struct mempolicy *new = NULL;
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002740 unsigned short mode;
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002741 unsigned short mode_flags;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002742 nodemask_t nodes;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002743 char *nodelist = strchr(str, ':');
2744 char *flags = strchr(str, '=');
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002745 int err = 1;
2746
2747 if (nodelist) {
2748 /* NUL-terminate mode or flags string */
2749 *nodelist++ = '\0';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002750 if (nodelist_parse(nodelist, nodes))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002751 goto out;
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002752 if (!nodes_subset(nodes, node_states[N_MEMORY]))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002753 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002754 } else
2755 nodes_clear(nodes);
2756
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002757 if (flags)
2758 *flags++ = '\0'; /* terminate mode string */
2759
Peter Zijlstra479e2802012-10-25 14:16:28 +02002760 for (mode = 0; mode < MPOL_MAX; mode++) {
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002761 if (!strcmp(str, policy_modes[mode])) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002762 break;
2763 }
2764 }
Mel Gormana7200942012-11-16 09:37:58 +00002765 if (mode >= MPOL_MAX)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002766 goto out;
2767
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002768 switch (mode) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002769 case MPOL_PREFERRED:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002770 /*
2771 * Insist on a nodelist of one node only
2772 */
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002773 if (nodelist) {
2774 char *rest = nodelist;
2775 while (isdigit(*rest))
2776 rest++;
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002777 if (*rest)
2778 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002779 }
2780 break;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002781 case MPOL_INTERLEAVE:
2782 /*
2783 * Default to online nodes with memory if no nodelist
2784 */
2785 if (!nodelist)
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002786 nodes = node_states[N_MEMORY];
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002787 break;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002788 case MPOL_LOCAL:
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002789 /*
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002790 * Don't allow a nodelist; mpol_new() checks flags
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002791 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002792 if (nodelist)
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002793 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002794 mode = MPOL_PREFERRED;
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002795 break;
Ravikiran G Thirumalai413b43d2010-03-23 13:35:28 -07002796 case MPOL_DEFAULT:
2797 /*
2798 * Insist on a empty nodelist
2799 */
2800 if (!nodelist)
2801 err = 0;
2802 goto out;
KOSAKI Motohirod69b2e62010-03-23 13:35:30 -07002803 case MPOL_BIND:
2804 /*
2805 * Insist on a nodelist
2806 */
2807 if (!nodelist)
2808 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002809 }
2810
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002811 mode_flags = 0;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002812 if (flags) {
2813 /*
2814 * Currently, we only support two mutually exclusive
2815 * mode flags.
2816 */
2817 if (!strcmp(flags, "static"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002818 mode_flags |= MPOL_F_STATIC_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002819 else if (!strcmp(flags, "relative"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002820 mode_flags |= MPOL_F_RELATIVE_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002821 else
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002822 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002823 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002824
2825 new = mpol_new(mode, mode_flags, &nodes);
2826 if (IS_ERR(new))
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002827 goto out;
2828
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002829 /*
2830 * Save nodes for mpol_to_str() to show the tmpfs mount options
2831 * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo.
2832 */
2833 if (mode != MPOL_PREFERRED)
2834 new->v.nodes = nodes;
2835 else if (nodelist)
2836 new->v.preferred_node = first_node(nodes);
2837 else
2838 new->flags |= MPOL_F_LOCAL;
2839
2840 /*
2841 * Save nodes for contextualization: this will be used to "clone"
2842 * the mempolicy in a specific context [cpuset] at a later time.
2843 */
2844 new->w.user_nodemask = nodes;
2845
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002846 err = 0;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002847
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002848out:
2849 /* Restore string for error message */
2850 if (nodelist)
2851 *--nodelist = ':';
2852 if (flags)
2853 *--flags = '=';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002854 if (!err)
2855 *mpol = new;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002856 return err;
2857}
2858#endif /* CONFIG_TMPFS */
2859
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002860/**
2861 * mpol_to_str - format a mempolicy structure for printing
2862 * @buffer: to contain formatted mempolicy string
2863 * @maxlen: length of @buffer
2864 * @pol: pointer to mempolicy to be formatted
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002865 *
David Rientjes948927e2013-11-12 15:07:28 -08002866 * Convert @pol into a string. If @buffer is too short, truncate the string.
2867 * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the
2868 * longest flag, "relative", and to display at least a few node ids.
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002869 */
David Rientjes948927e2013-11-12 15:07:28 -08002870void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002871{
2872 char *p = buffer;
David Rientjes948927e2013-11-12 15:07:28 -08002873 nodemask_t nodes = NODE_MASK_NONE;
2874 unsigned short mode = MPOL_DEFAULT;
2875 unsigned short flags = 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002876
David Rientjes8790c712014-01-30 15:46:08 -08002877 if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002878 mode = pol->mode;
David Rientjes948927e2013-11-12 15:07:28 -08002879 flags = pol->flags;
2880 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002881
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002882 switch (mode) {
2883 case MPOL_DEFAULT:
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002884 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002885 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002886 if (flags & MPOL_F_LOCAL)
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002887 mode = MPOL_LOCAL;
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002888 else
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002889 node_set(pol->v.preferred_node, nodes);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002890 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002891 case MPOL_BIND:
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002892 case MPOL_INTERLEAVE:
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002893 nodes = pol->v.nodes;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002894 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002895 default:
David Rientjes948927e2013-11-12 15:07:28 -08002896 WARN_ON_ONCE(1);
2897 snprintf(p, maxlen, "unknown");
2898 return;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002899 }
2900
David Rientjesb7a9f422013-11-21 14:32:06 -08002901 p += snprintf(p, maxlen, "%s", policy_modes[mode]);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002902
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002903 if (flags & MPOL_MODE_FLAGS) {
David Rientjes948927e2013-11-12 15:07:28 -08002904 p += snprintf(p, buffer + maxlen - p, "=");
David Rientjesf5b087b2008-04-28 02:12:27 -07002905
Lee Schermerhorn22919902008-04-28 02:13:22 -07002906 /*
2907 * Currently, the only defined flags are mutually exclusive
2908 */
David Rientjesf5b087b2008-04-28 02:12:27 -07002909 if (flags & MPOL_F_STATIC_NODES)
Lee Schermerhorn22919902008-04-28 02:13:22 -07002910 p += snprintf(p, buffer + maxlen - p, "static");
2911 else if (flags & MPOL_F_RELATIVE_NODES)
2912 p += snprintf(p, buffer + maxlen - p, "relative");
David Rientjesf5b087b2008-04-28 02:12:27 -07002913 }
2914
Tejun Heo9e763e02015-02-13 14:38:02 -08002915 if (!nodes_empty(nodes))
2916 p += scnprintf(p, buffer + maxlen - p, ":%*pbl",
2917 nodemask_pr_args(&nodes));
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002918}