blob: c60807625fd5162f8f86086a8e7e3e98319ef2e2 [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>
Ingo Molnar6e84f312017-02-08 18:51:29 +010076#include <linux/sched/mm.h>
Ingo Molnar6a3827d2017-02-08 18:51:31 +010077#include <linux/sched/numa_balancing.h>
Ingo Molnarf719ff9b2017-02-06 10:57:33 +010078#include <linux/sched/task.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070079#include <linux/nodemask.h>
80#include <linux/cpuset.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070081#include <linux/slab.h>
82#include <linux/string.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040083#include <linux/export.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070084#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070085#include <linux/interrupt.h>
86#include <linux/init.h>
87#include <linux/compat.h>
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -080088#include <linux/swap.h>
Christoph Lameter1a75a6c2006-01-08 01:01:02 -080089#include <linux/seq_file.h>
90#include <linux/proc_fs.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080091#include <linux/migrate.h>
Hugh Dickins62b61f62009-12-14 17:59:33 -080092#include <linux/ksm.h>
Christoph Lameter95a402c2006-06-23 02:03:53 -070093#include <linux/rmap.h>
David Quigley86c3a762006-06-23 02:04:02 -070094#include <linux/security.h>
Adrian Bunkdbcb0f12007-10-16 01:26:26 -070095#include <linux/syscalls.h>
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -070096#include <linux/ctype.h>
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -080097#include <linux/mm_inline.h>
Lee Schermerhornb24f53a2012-10-25 14:16:32 +020098#include <linux/mmu_notifier.h>
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070099#include <linux/printk.h>
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101#include <asm/tlbflush.h>
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -0800102#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
Nick Piggin62695a82008-10-18 20:26:09 -0700104#include "internal.h"
105
Christoph Lameter38e35862006-01-08 01:01:01 -0800106/* Internal flags */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800107#define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */
Christoph Lameter38e35862006-01-08 01:01:01 -0800108#define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800109
Pekka Enbergfcc234f2006-03-22 00:08:13 -0800110static struct kmem_cache *policy_cache;
111static struct kmem_cache *sn_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113/* Highest zone. An specific allocation for a zone below that is not
114 policied. */
Christoph Lameter62672762007-02-10 01:43:07 -0800115enum zone_type policy_zone = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700117/*
118 * run-time system-wide default policy => local allocation
119 */
H Hartley Sweetene754d792011-10-31 17:09:23 -0700120static struct mempolicy default_policy = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 .refcnt = ATOMIC_INIT(1), /* never free it */
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700122 .mode = MPOL_PREFERRED,
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700123 .flags = MPOL_F_LOCAL,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124};
125
Mel Gorman5606e382012-11-02 18:19:13 +0000126static struct mempolicy preferred_node_policy[MAX_NUMNODES];
127
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -0700128struct mempolicy *get_task_policy(struct task_struct *p)
Mel Gorman5606e382012-11-02 18:19:13 +0000129{
130 struct mempolicy *pol = p->mempolicy;
Oleg Nesterovf15ca782014-10-09 15:27:43 -0700131 int node;
Mel Gorman5606e382012-11-02 18:19:13 +0000132
Oleg Nesterovf15ca782014-10-09 15:27:43 -0700133 if (pol)
134 return pol;
Mel Gorman5606e382012-11-02 18:19:13 +0000135
Oleg Nesterovf15ca782014-10-09 15:27:43 -0700136 node = numa_node_id();
137 if (node != NUMA_NO_NODE) {
138 pol = &preferred_node_policy[node];
139 /* preferred_node_policy is not initialised early in boot */
140 if (pol->mode)
141 return pol;
Mel Gorman5606e382012-11-02 18:19:13 +0000142 }
143
Oleg Nesterovf15ca782014-10-09 15:27:43 -0700144 return &default_policy;
Mel Gorman5606e382012-11-02 18:19:13 +0000145}
146
David Rientjes37012942008-04-28 02:12:33 -0700147static const struct mempolicy_operations {
148 int (*create)(struct mempolicy *pol, const nodemask_t *nodes);
Miao Xie708c1bb2010-05-24 14:32:07 -0700149 /*
150 * If read-side task has no lock to protect task->mempolicy, write-side
151 * task will rebind the task->mempolicy by two step. The first step is
152 * setting all the newly nodes, and the second step is cleaning all the
153 * disallowed nodes. In this way, we can avoid finding no node to alloc
154 * page.
155 * If we have a lock to protect task->mempolicy in read-side, we do
156 * rebind directly.
157 *
158 * step:
159 * MPOL_REBIND_ONCE - do rebind work at once
160 * MPOL_REBIND_STEP1 - set all the newly nodes
161 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
162 */
163 void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes,
164 enum mpol_rebind_step step);
David Rientjes37012942008-04-28 02:12:33 -0700165} mpol_ops[MPOL_MAX];
166
David Rientjesf5b087b2008-04-28 02:12:27 -0700167static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
168{
Bob Liu6d556292010-05-24 14:31:59 -0700169 return pol->flags & MPOL_MODE_FLAGS;
David Rientjes4c50bc02008-04-28 02:12:30 -0700170}
171
172static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig,
173 const nodemask_t *rel)
174{
175 nodemask_t tmp;
176 nodes_fold(tmp, *orig, nodes_weight(*rel));
177 nodes_onto(*ret, tmp, *rel);
David Rientjesf5b087b2008-04-28 02:12:27 -0700178}
179
David Rientjes37012942008-04-28 02:12:33 -0700180static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes)
181{
182 if (nodes_empty(*nodes))
183 return -EINVAL;
184 pol->v.nodes = *nodes;
185 return 0;
186}
187
188static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes)
189{
190 if (!nodes)
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700191 pol->flags |= MPOL_F_LOCAL; /* local allocation */
David Rientjes37012942008-04-28 02:12:33 -0700192 else if (nodes_empty(*nodes))
193 return -EINVAL; /* no allowed nodes */
194 else
195 pol->v.preferred_node = first_node(*nodes);
196 return 0;
197}
198
199static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes)
200{
Zhihui Zhang859f7ef2014-12-18 16:17:09 -0800201 if (nodes_empty(*nodes))
David Rientjes37012942008-04-28 02:12:33 -0700202 return -EINVAL;
203 pol->v.nodes = *nodes;
204 return 0;
205}
206
Miao Xie58568d22009-06-16 15:31:49 -0700207/*
208 * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if
209 * any, for the new policy. mpol_new() has already validated the nodes
210 * parameter with respect to the policy mode and flags. But, we need to
211 * handle an empty nodemask with MPOL_PREFERRED here.
212 *
213 * Must be called holding task's alloc_lock to protect task's mems_allowed
214 * and mempolicy. May also be called holding the mmap_semaphore for write.
215 */
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700216static int mpol_set_nodemask(struct mempolicy *pol,
217 const nodemask_t *nodes, struct nodemask_scratch *nsc)
Miao Xie58568d22009-06-16 15:31:49 -0700218{
Miao Xie58568d22009-06-16 15:31:49 -0700219 int ret;
220
221 /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */
222 if (pol == NULL)
223 return 0;
Lai Jiangshan01f13bd2012-12-12 13:51:33 -0800224 /* Check N_MEMORY */
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700225 nodes_and(nsc->mask1,
Lai Jiangshan01f13bd2012-12-12 13:51:33 -0800226 cpuset_current_mems_allowed, node_states[N_MEMORY]);
Miao Xie58568d22009-06-16 15:31:49 -0700227
228 VM_BUG_ON(!nodes);
229 if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes))
230 nodes = NULL; /* explicit local allocation */
231 else {
232 if (pol->flags & MPOL_F_RELATIVE_NODES)
Zhihui Zhang859f7ef2014-12-18 16:17:09 -0800233 mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1);
Miao Xie58568d22009-06-16 15:31:49 -0700234 else
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700235 nodes_and(nsc->mask2, *nodes, nsc->mask1);
236
Miao Xie58568d22009-06-16 15:31:49 -0700237 if (mpol_store_user_nodemask(pol))
238 pol->w.user_nodemask = *nodes;
239 else
240 pol->w.cpuset_mems_allowed =
241 cpuset_current_mems_allowed;
242 }
243
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700244 if (nodes)
245 ret = mpol_ops[pol->mode].create(pol, &nsc->mask2);
246 else
247 ret = mpol_ops[pol->mode].create(pol, NULL);
Miao Xie58568d22009-06-16 15:31:49 -0700248 return ret;
249}
250
251/*
252 * This function just creates a new policy, does some check and simple
253 * initialization. You must invoke mpol_set_nodemask() to set nodes.
254 */
David Rientjes028fec42008-04-28 02:12:25 -0700255static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
256 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257{
258 struct mempolicy *policy;
259
David Rientjes028fec42008-04-28 02:12:25 -0700260 pr_debug("setting mode %d flags %d nodes[0] %lx\n",
David Rientjes00ef2d22013-02-22 16:35:36 -0800261 mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE);
Paul Mundt140d5a42007-07-15 23:38:16 -0700262
David Rientjes3e1f06452008-04-28 02:12:34 -0700263 if (mode == MPOL_DEFAULT) {
264 if (nodes && !nodes_empty(*nodes))
David Rientjes37012942008-04-28 02:12:33 -0700265 return ERR_PTR(-EINVAL);
Lee Schermerhornd3a71032012-10-25 14:16:29 +0200266 return NULL;
David Rientjes37012942008-04-28 02:12:33 -0700267 }
David Rientjes3e1f06452008-04-28 02:12:34 -0700268 VM_BUG_ON(!nodes);
269
270 /*
271 * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
272 * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
273 * All other modes require a valid pointer to a non-empty nodemask.
274 */
275 if (mode == MPOL_PREFERRED) {
276 if (nodes_empty(*nodes)) {
277 if (((flags & MPOL_F_STATIC_NODES) ||
278 (flags & MPOL_F_RELATIVE_NODES)))
279 return ERR_PTR(-EINVAL);
David Rientjes3e1f06452008-04-28 02:12:34 -0700280 }
Peter Zijlstra479e2802012-10-25 14:16:28 +0200281 } else if (mode == MPOL_LOCAL) {
Piotr Kwapulinski8d303e42016-12-12 16:42:49 -0800282 if (!nodes_empty(*nodes) ||
283 (flags & MPOL_F_STATIC_NODES) ||
284 (flags & MPOL_F_RELATIVE_NODES))
Peter Zijlstra479e2802012-10-25 14:16:28 +0200285 return ERR_PTR(-EINVAL);
286 mode = MPOL_PREFERRED;
David Rientjes3e1f06452008-04-28 02:12:34 -0700287 } else if (nodes_empty(*nodes))
288 return ERR_PTR(-EINVAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
290 if (!policy)
291 return ERR_PTR(-ENOMEM);
292 atomic_set(&policy->refcnt, 1);
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700293 policy->mode = mode;
David Rientjes3e1f06452008-04-28 02:12:34 -0700294 policy->flags = flags;
David Rientjesf5b087b2008-04-28 02:12:27 -0700295
David Rientjes37012942008-04-28 02:12:33 -0700296 return policy;
297}
298
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700299/* Slow path of a mpol destructor. */
300void __mpol_put(struct mempolicy *p)
301{
302 if (!atomic_dec_and_test(&p->refcnt))
303 return;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700304 kmem_cache_free(policy_cache, p);
305}
306
Miao Xie708c1bb2010-05-24 14:32:07 -0700307static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes,
308 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700309{
310}
311
Miao Xie708c1bb2010-05-24 14:32:07 -0700312/*
313 * step:
314 * MPOL_REBIND_ONCE - do rebind work at once
315 * MPOL_REBIND_STEP1 - set all the newly nodes
316 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
317 */
318static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes,
319 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700320{
321 nodemask_t tmp;
322
323 if (pol->flags & MPOL_F_STATIC_NODES)
324 nodes_and(tmp, pol->w.user_nodemask, *nodes);
325 else if (pol->flags & MPOL_F_RELATIVE_NODES)
326 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
327 else {
Miao Xie708c1bb2010-05-24 14:32:07 -0700328 /*
329 * if step == 1, we use ->w.cpuset_mems_allowed to cache the
330 * result
331 */
332 if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) {
333 nodes_remap(tmp, pol->v.nodes,
334 pol->w.cpuset_mems_allowed, *nodes);
335 pol->w.cpuset_mems_allowed = step ? tmp : *nodes;
336 } else if (step == MPOL_REBIND_STEP2) {
337 tmp = pol->w.cpuset_mems_allowed;
338 pol->w.cpuset_mems_allowed = *nodes;
339 } else
340 BUG();
David Rientjes37012942008-04-28 02:12:33 -0700341 }
342
Miao Xie708c1bb2010-05-24 14:32:07 -0700343 if (nodes_empty(tmp))
344 tmp = *nodes;
345
346 if (step == MPOL_REBIND_STEP1)
347 nodes_or(pol->v.nodes, pol->v.nodes, tmp);
348 else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2)
349 pol->v.nodes = tmp;
350 else
351 BUG();
David Rientjes37012942008-04-28 02:12:33 -0700352}
353
354static void mpol_rebind_preferred(struct mempolicy *pol,
Miao Xie708c1bb2010-05-24 14:32:07 -0700355 const nodemask_t *nodes,
356 enum mpol_rebind_step step)
David Rientjes37012942008-04-28 02:12:33 -0700357{
358 nodemask_t tmp;
359
David Rientjes37012942008-04-28 02:12:33 -0700360 if (pol->flags & MPOL_F_STATIC_NODES) {
361 int node = first_node(pol->w.user_nodemask);
362
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700363 if (node_isset(node, *nodes)) {
David Rientjes37012942008-04-28 02:12:33 -0700364 pol->v.preferred_node = node;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700365 pol->flags &= ~MPOL_F_LOCAL;
366 } else
367 pol->flags |= MPOL_F_LOCAL;
David Rientjes37012942008-04-28 02:12:33 -0700368 } else if (pol->flags & MPOL_F_RELATIVE_NODES) {
369 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
370 pol->v.preferred_node = first_node(tmp);
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700371 } else if (!(pol->flags & MPOL_F_LOCAL)) {
David Rientjes37012942008-04-28 02:12:33 -0700372 pol->v.preferred_node = node_remap(pol->v.preferred_node,
373 pol->w.cpuset_mems_allowed,
374 *nodes);
375 pol->w.cpuset_mems_allowed = *nodes;
376 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377}
378
Miao Xie708c1bb2010-05-24 14:32:07 -0700379/*
380 * mpol_rebind_policy - Migrate a policy to a different set of nodes
381 *
382 * If read-side task has no lock to protect task->mempolicy, write-side
383 * task will rebind the task->mempolicy by two step. The first step is
384 * setting all the newly nodes, and the second step is cleaning all the
385 * disallowed nodes. In this way, we can avoid finding no node to alloc
386 * page.
387 * If we have a lock to protect task->mempolicy in read-side, we do
388 * rebind directly.
389 *
390 * step:
391 * MPOL_REBIND_ONCE - do rebind work at once
392 * MPOL_REBIND_STEP1 - set all the newly nodes
393 * MPOL_REBIND_STEP2 - clean all the disallowed nodes
394 */
395static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask,
396 enum mpol_rebind_step step)
David Rientjes1d0d2682008-04-28 02:12:32 -0700397{
David Rientjes1d0d2682008-04-28 02:12:32 -0700398 if (!pol)
399 return;
Wang Sheng-Hui89c522c2012-05-29 15:06:16 -0700400 if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE &&
David Rientjes1d0d2682008-04-28 02:12:32 -0700401 nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
402 return;
Miao Xie708c1bb2010-05-24 14:32:07 -0700403
404 if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING))
405 return;
406
407 if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING))
408 BUG();
409
410 if (step == MPOL_REBIND_STEP1)
411 pol->flags |= MPOL_F_REBINDING;
412 else if (step == MPOL_REBIND_STEP2)
413 pol->flags &= ~MPOL_F_REBINDING;
414 else if (step >= MPOL_REBIND_NSTEP)
415 BUG();
416
417 mpol_ops[pol->mode].rebind(pol, newmask, step);
David Rientjes1d0d2682008-04-28 02:12:32 -0700418}
419
420/*
421 * Wrapper for mpol_rebind_policy() that just requires task
422 * pointer, and updates task mempolicy.
Miao Xie58568d22009-06-16 15:31:49 -0700423 *
424 * Called with task's alloc_lock held.
David Rientjes1d0d2682008-04-28 02:12:32 -0700425 */
426
Miao Xie708c1bb2010-05-24 14:32:07 -0700427void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
428 enum mpol_rebind_step step)
David Rientjes1d0d2682008-04-28 02:12:32 -0700429{
Miao Xie708c1bb2010-05-24 14:32:07 -0700430 mpol_rebind_policy(tsk->mempolicy, new, step);
David Rientjes1d0d2682008-04-28 02:12:32 -0700431}
432
433/*
434 * Rebind each vma in mm to new nodemask.
435 *
436 * Call holding a reference to mm. Takes mm->mmap_sem during call.
437 */
438
439void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
440{
441 struct vm_area_struct *vma;
442
443 down_write(&mm->mmap_sem);
444 for (vma = mm->mmap; vma; vma = vma->vm_next)
Miao Xie708c1bb2010-05-24 14:32:07 -0700445 mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE);
David Rientjes1d0d2682008-04-28 02:12:32 -0700446 up_write(&mm->mmap_sem);
447}
448
David Rientjes37012942008-04-28 02:12:33 -0700449static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
450 [MPOL_DEFAULT] = {
451 .rebind = mpol_rebind_default,
452 },
453 [MPOL_INTERLEAVE] = {
454 .create = mpol_new_interleave,
455 .rebind = mpol_rebind_nodemask,
456 },
457 [MPOL_PREFERRED] = {
458 .create = mpol_new_preferred,
459 .rebind = mpol_rebind_preferred,
460 },
461 [MPOL_BIND] = {
462 .create = mpol_new_bind,
463 .rebind = mpol_rebind_nodemask,
464 },
465};
466
Christoph Lameterfc301282006-01-18 17:42:29 -0800467static void migrate_page_add(struct page *page, struct list_head *pagelist,
468 unsigned long flags);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800469
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800470struct queue_pages {
471 struct list_head *pagelist;
472 unsigned long flags;
473 nodemask_t *nmask;
474 struct vm_area_struct *prev;
475};
476
Naoya Horiguchi98094942013-09-11 14:22:14 -0700477/*
478 * Scan through pages checking if pages follow certain conditions,
479 * and move them to the pagelist if they do.
480 */
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800481static int queue_pages_pte_range(pmd_t *pmd, unsigned long addr,
482 unsigned long end, struct mm_walk *walk)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800484 struct vm_area_struct *vma = walk->vma;
485 struct page *page;
486 struct queue_pages *qp = walk->private;
487 unsigned long flags = qp->flags;
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800488 int nid, ret;
Hugh Dickins91612e02005-06-21 17:15:07 -0700489 pte_t *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700490 spinlock_t *ptl;
Hugh Dickins941150a2005-06-21 17:15:06 -0700491
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800492 if (pmd_trans_huge(*pmd)) {
493 ptl = pmd_lock(walk->mm, pmd);
494 if (pmd_trans_huge(*pmd)) {
495 page = pmd_page(*pmd);
496 if (is_huge_zero_page(page)) {
497 spin_unlock(ptl);
David Rientjesfd607752016-12-12 16:42:20 -0800498 __split_huge_pmd(vma, pmd, addr, false, NULL);
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800499 } else {
500 get_page(page);
501 spin_unlock(ptl);
502 lock_page(page);
503 ret = split_huge_page(page);
504 unlock_page(page);
505 put_page(page);
506 if (ret)
507 return 0;
508 }
509 } else {
510 spin_unlock(ptl);
511 }
512 }
Hugh Dickins91612e02005-06-21 17:15:07 -0700513
Naoya Horiguchi337d9ab2016-07-26 15:24:03 -0700514 if (pmd_trans_unstable(pmd))
515 return 0;
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800516retry:
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800517 pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
518 for (; addr != end; pte++, addr += PAGE_SIZE) {
Hugh Dickins91612e02005-06-21 17:15:07 -0700519 if (!pte_present(*pte))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800521 page = vm_normal_page(vma, addr, *pte);
522 if (!page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 continue;
Nick Piggin053837f2006-01-18 17:42:27 -0800524 /*
Hugh Dickins62b61f62009-12-14 17:59:33 -0800525 * vm_normal_page() filters out zero pages, but there might
526 * still be PageReserved pages to skip, perhaps in a VDSO.
Nick Piggin053837f2006-01-18 17:42:27 -0800527 */
Hugh Dickinsb79bc0a2013-02-22 16:35:13 -0800528 if (PageReserved(page))
Christoph Lameterf4598c82006-01-12 01:05:20 -0800529 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800530 nid = page_to_nid(page);
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800531 if (node_isset(nid, *qp->nmask) == !!(flags & MPOL_MF_INVERT))
Christoph Lameter38e35862006-01-08 01:01:01 -0800532 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700533 if (PageTransCompound(page)) {
Kirill A. Shutemov248db922016-01-15 16:54:14 -0800534 get_page(page);
535 pte_unmap_unlock(pte, ptl);
536 lock_page(page);
537 ret = split_huge_page(page);
538 unlock_page(page);
539 put_page(page);
540 /* Failed to split -- skip. */
541 if (ret) {
542 pte = pte_offset_map_lock(walk->mm, pmd,
543 addr, &ptl);
544 continue;
545 }
546 goto retry;
547 }
Christoph Lameter38e35862006-01-08 01:01:01 -0800548
Kirill A. Shutemov77bf45e2016-02-05 15:36:33 -0800549 migrate_page_add(page, qp->pagelist, flags);
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800550 }
551 pte_unmap_unlock(pte - 1, ptl);
552 cond_resched();
553 return 0;
Hugh Dickins91612e02005-06-21 17:15:07 -0700554}
555
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800556static int queue_pages_hugetlb(pte_t *pte, unsigned long hmask,
557 unsigned long addr, unsigned long end,
558 struct mm_walk *walk)
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700559{
560#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800561 struct queue_pages *qp = walk->private;
562 unsigned long flags = qp->flags;
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700563 int nid;
564 struct page *page;
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800565 spinlock_t *ptl;
Naoya Horiguchid4c54912014-06-06 10:00:01 -0400566 pte_t entry;
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700567
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800568 ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte);
569 entry = huge_ptep_get(pte);
Naoya Horiguchid4c54912014-06-06 10:00:01 -0400570 if (!pte_present(entry))
571 goto unlock;
572 page = pte_page(entry);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700573 nid = page_to_nid(page);
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800574 if (node_isset(nid, *qp->nmask) == !!(flags & MPOL_MF_INVERT))
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700575 goto unlock;
576 /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */
577 if (flags & (MPOL_MF_MOVE_ALL) ||
578 (flags & MPOL_MF_MOVE && page_mapcount(page) == 1))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800579 isolate_huge_page(page, qp->pagelist);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700580unlock:
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800581 spin_unlock(ptl);
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700582#else
583 BUG();
584#endif
Hugh Dickins91612e02005-06-21 17:15:07 -0700585 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586}
587
Aneesh Kumar K.V58772312013-12-06 00:08:22 +0530588#ifdef CONFIG_NUMA_BALANCING
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200589/*
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200590 * This is used to mark a range of virtual addresses to be inaccessible.
591 * These are later cleared by a NUMA hinting fault. Depending on these
592 * faults, pages may be migrated for better NUMA placement.
593 *
594 * This is assuming that NUMA faults are handled using PROT_NONE. If
595 * an architecture makes a different choice, it will need further
596 * changes to the core.
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200597 */
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200598unsigned long change_prot_numa(struct vm_area_struct *vma,
599 unsigned long addr, unsigned long end)
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200600{
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200601 int nr_updated;
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200602
Mel Gorman4d942462015-02-12 14:58:28 -0800603 nr_updated = change_protection(vma, addr, end, PAGE_NONE, 0, 1);
Mel Gorman03c5a6e2012-11-02 14:52:48 +0000604 if (nr_updated)
605 count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated);
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200606
Mel Gorman4b10e7d2012-10-25 14:16:32 +0200607 return nr_updated;
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200608}
609#else
610static unsigned long change_prot_numa(struct vm_area_struct *vma,
611 unsigned long addr, unsigned long end)
612{
613 return 0;
614}
Aneesh Kumar K.V58772312013-12-06 00:08:22 +0530615#endif /* CONFIG_NUMA_BALANCING */
Lee Schermerhornb24f53a2012-10-25 14:16:32 +0200616
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800617static int queue_pages_test_walk(unsigned long start, unsigned long end,
618 struct mm_walk *walk)
619{
620 struct vm_area_struct *vma = walk->vma;
621 struct queue_pages *qp = walk->private;
622 unsigned long endvma = vma->vm_end;
623 unsigned long flags = qp->flags;
624
Kirill A. Shutemov77bf45e2016-02-05 15:36:33 -0800625 if (!vma_migratable(vma))
Naoya Horiguchi48684a62015-02-11 15:28:06 -0800626 return 1;
627
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800628 if (endvma > end)
629 endvma = end;
630 if (vma->vm_start > start)
631 start = vma->vm_start;
632
633 if (!(flags & MPOL_MF_DISCONTIG_OK)) {
634 if (!vma->vm_next && vma->vm_end < end)
635 return -EFAULT;
636 if (qp->prev && qp->prev->vm_end < vma->vm_start)
637 return -EFAULT;
638 }
639
640 qp->prev = vma;
641
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800642 if (flags & MPOL_MF_LAZY) {
643 /* Similar to task_numa_work, skip inaccessible VMAs */
Liang Chen4355c012016-03-15 14:56:42 -0700644 if (!is_vm_hugetlb_page(vma) &&
645 (vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)) &&
646 !(vma->vm_flags & VM_MIXEDMAP))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800647 change_prot_numa(vma, start, endvma);
648 return 1;
649 }
650
Kirill A. Shutemov77bf45e2016-02-05 15:36:33 -0800651 /* queue pages from current vma */
652 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800653 return 0;
654 return 1;
655}
656
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800657/*
Naoya Horiguchi98094942013-09-11 14:22:14 -0700658 * Walk through page tables and collect pages to be migrated.
659 *
660 * If pages found in a given range are on a set of nodes (determined by
661 * @nodes and @flags,) it's isolated and queued to the pagelist which is
662 * passed via @private.)
Christoph Lameterdc9aa5b2006-01-08 01:00:50 -0800663 */
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -0700664static int
Naoya Horiguchi98094942013-09-11 14:22:14 -0700665queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800666 nodemask_t *nodes, unsigned long flags,
667 struct list_head *pagelist)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668{
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800669 struct queue_pages qp = {
670 .pagelist = pagelist,
671 .flags = flags,
672 .nmask = nodes,
673 .prev = NULL,
674 };
675 struct mm_walk queue_pages_walk = {
676 .hugetlb_entry = queue_pages_hugetlb,
677 .pmd_entry = queue_pages_pte_range,
678 .test_walk = queue_pages_test_walk,
679 .mm = mm,
680 .private = &qp,
681 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682
Naoya Horiguchi6f4576e2015-02-11 15:28:03 -0800683 return walk_page_range(start, end, &queue_pages_walk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684}
685
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700686/*
687 * Apply policy to a single VMA
688 * This must be called with the mmap_sem held for writing.
689 */
690static int vma_replace_policy(struct vm_area_struct *vma,
691 struct mempolicy *pol)
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700692{
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700693 int err;
694 struct mempolicy *old;
695 struct mempolicy *new;
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700696
697 pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
698 vma->vm_start, vma->vm_end, vma->vm_pgoff,
699 vma->vm_ops, vma->vm_file,
700 vma->vm_ops ? vma->vm_ops->set_policy : NULL);
701
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700702 new = mpol_dup(pol);
703 if (IS_ERR(new))
704 return PTR_ERR(new);
705
706 if (vma->vm_ops && vma->vm_ops->set_policy) {
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700707 err = vma->vm_ops->set_policy(vma, new);
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700708 if (err)
709 goto err_out;
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700710 }
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700711
712 old = vma->vm_policy;
713 vma->vm_policy = new; /* protected by mmap_sem */
714 mpol_put(old);
715
716 return 0;
717 err_out:
718 mpol_put(new);
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700719 return err;
720}
721
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722/* Step 2: apply policy to a range and do splits. */
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800723static int mbind_range(struct mm_struct *mm, unsigned long start,
724 unsigned long end, struct mempolicy *new_pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
726 struct vm_area_struct *next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800727 struct vm_area_struct *prev;
728 struct vm_area_struct *vma;
729 int err = 0;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800730 pgoff_t pgoff;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800731 unsigned long vmstart;
732 unsigned long vmend;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
Linus Torvalds097d5912012-03-06 18:23:36 -0800734 vma = find_vma(mm, start);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800735 if (!vma || vma->vm_start > start)
736 return -EFAULT;
737
Linus Torvalds097d5912012-03-06 18:23:36 -0800738 prev = vma->vm_prev;
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800739 if (start > vma->vm_start)
740 prev = vma;
741
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800742 for (; vma && vma->vm_start < end; prev = vma, vma = next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 next = vma->vm_next;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800744 vmstart = max(start, vma->vm_start);
745 vmend = min(end, vma->vm_end);
746
KOSAKI Motohiroe26a5112011-12-28 15:57:11 -0800747 if (mpol_equal(vma_policy(vma), new_pol))
748 continue;
749
750 pgoff = vma->vm_pgoff +
751 ((vmstart - vma->vm_start) >> PAGE_SHIFT);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800752 prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
Andrea Arcangeli19a809a2015-09-04 15:46:24 -0700753 vma->anon_vma, vma->vm_file, pgoff,
754 new_pol, vma->vm_userfaultfd_ctx);
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800755 if (prev) {
756 vma = prev;
757 next = vma->vm_next;
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700758 if (mpol_equal(vma_policy(vma), new_pol))
759 continue;
760 /* vma_merge() joined vma && vma->next, case 8 */
761 goto replace;
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800762 }
763 if (vma->vm_start != vmstart) {
764 err = split_vma(vma->vm_mm, vma, vmstart, 1);
765 if (err)
766 goto out;
767 }
768 if (vma->vm_end != vmend) {
769 err = split_vma(vma->vm_mm, vma, vmend, 0);
770 if (err)
771 goto out;
772 }
Oleg Nesterov3964acd2013-07-31 13:53:28 -0700773 replace:
KOSAKI Motohiro869833f2012-10-08 16:29:16 -0700774 err = vma_replace_policy(vma, new_pol);
KOSAKI Motohiro8d34694c12012-10-08 16:29:14 -0700775 if (err)
776 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 }
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -0800778
779 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 return err;
781}
782
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783/* Set the process memory policy */
David Rientjes028fec42008-04-28 02:12:25 -0700784static long do_set_mempolicy(unsigned short mode, unsigned short flags,
785 nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786{
Miao Xie58568d22009-06-16 15:31:49 -0700787 struct mempolicy *new, *old;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700788 NODEMASK_SCRATCH(scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700789 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700791 if (!scratch)
792 return -ENOMEM;
Lee Schermerhornf4e53d92008-04-28 02:13:10 -0700793
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700794 new = mpol_new(mode, flags, nodes);
795 if (IS_ERR(new)) {
796 ret = PTR_ERR(new);
797 goto out;
798 }
Oleg Nesterov2c7c3a72014-10-09 15:27:55 -0700799
Miao Xie58568d22009-06-16 15:31:49 -0700800 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700801 ret = mpol_set_nodemask(new, nodes, scratch);
Miao Xie58568d22009-06-16 15:31:49 -0700802 if (ret) {
803 task_unlock(current);
Miao Xie58568d22009-06-16 15:31:49 -0700804 mpol_put(new);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700805 goto out;
Miao Xie58568d22009-06-16 15:31:49 -0700806 }
807 old = current->mempolicy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 current->mempolicy = new;
Vlastimil Babka45816682017-07-06 15:39:59 -0700809 if (new && new->mode == MPOL_INTERLEAVE)
810 current->il_prev = MAX_NUMNODES-1;
Miao Xie58568d22009-06-16 15:31:49 -0700811 task_unlock(current);
Miao Xie58568d22009-06-16 15:31:49 -0700812 mpol_put(old);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -0700813 ret = 0;
814out:
815 NODEMASK_SCRATCH_FREE(scratch);
816 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817}
818
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700819/*
820 * Return nodemask for policy for get_mempolicy() query
Miao Xie58568d22009-06-16 15:31:49 -0700821 *
822 * Called with task's alloc_lock held
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700823 */
824static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825{
Andi Kleendfcd3c02005-10-29 18:15:48 -0700826 nodes_clear(*nodes);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700827 if (p == &default_policy)
828 return;
829
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700830 switch (p->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -0700831 case MPOL_BIND:
832 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 case MPOL_INTERLEAVE:
Andi Kleendfcd3c02005-10-29 18:15:48 -0700834 *nodes = p->v.nodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 break;
836 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -0700837 if (!(p->flags & MPOL_F_LOCAL))
Andi Kleendfcd3c02005-10-29 18:15:48 -0700838 node_set(p->v.preferred_node, *nodes);
Lee Schermerhorn53f25562008-04-28 02:13:20 -0700839 /* else return empty node mask for local allocation */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 break;
841 default:
842 BUG();
843 }
844}
845
Dave Hansend4edcf02016-02-12 13:01:56 -0800846static int lookup_node(unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847{
848 struct page *p;
849 int err;
850
Lorenzo Stoakes768ae302016-10-13 01:20:16 +0100851 err = get_user_pages(addr & PAGE_MASK, 1, 0, &p, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 if (err >= 0) {
853 err = page_to_nid(p);
854 put_page(p);
855 }
856 return err;
857}
858
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859/* Retrieve NUMA policy */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700860static long do_get_mempolicy(int *policy, nodemask_t *nmask,
861 unsigned long addr, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862{
Christoph Lameter8bccd852005-10-29 18:16:59 -0700863 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 struct mm_struct *mm = current->mm;
865 struct vm_area_struct *vma = NULL;
866 struct mempolicy *pol = current->mempolicy;
867
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700868 if (flags &
869 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 return -EINVAL;
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700871
872 if (flags & MPOL_F_MEMS_ALLOWED) {
873 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
874 return -EINVAL;
875 *policy = 0; /* just so it's initialized */
Miao Xie58568d22009-06-16 15:31:49 -0700876 task_lock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700877 *nmask = cpuset_current_mems_allowed;
Miao Xie58568d22009-06-16 15:31:49 -0700878 task_unlock(current);
Lee Schermerhorn754af6f2007-10-16 01:24:51 -0700879 return 0;
880 }
881
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 if (flags & MPOL_F_ADDR) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700883 /*
884 * Do NOT fall back to task policy if the
885 * vma/shared policy at addr is NULL. We
886 * want to return MPOL_DEFAULT in this case.
887 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 down_read(&mm->mmap_sem);
889 vma = find_vma_intersection(mm, addr, addr+1);
890 if (!vma) {
891 up_read(&mm->mmap_sem);
892 return -EFAULT;
893 }
894 if (vma->vm_ops && vma->vm_ops->get_policy)
895 pol = vma->vm_ops->get_policy(vma, addr);
896 else
897 pol = vma->vm_policy;
898 } else if (addr)
899 return -EINVAL;
900
901 if (!pol)
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700902 pol = &default_policy; /* indicates default behavior */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
904 if (flags & MPOL_F_NODE) {
905 if (flags & MPOL_F_ADDR) {
Dave Hansend4edcf02016-02-12 13:01:56 -0800906 err = lookup_node(addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 if (err < 0)
908 goto out;
Christoph Lameter8bccd852005-10-29 18:16:59 -0700909 *policy = err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 } else if (pol == current->mempolicy &&
Lee Schermerhorn45c47452008-04-28 02:13:12 -0700911 pol->mode == MPOL_INTERLEAVE) {
Vlastimil Babka45816682017-07-06 15:39:59 -0700912 *policy = next_node_in(current->il_prev, pol->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 } else {
914 err = -EINVAL;
915 goto out;
916 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700917 } else {
918 *policy = pol == &default_policy ? MPOL_DEFAULT :
919 pol->mode;
David Rientjesd79df632008-07-04 12:24:13 -0700920 /*
921 * Internal mempolicy flags must be masked off before exposing
922 * the policy to userspace.
923 */
924 *policy |= (pol->flags & MPOL_MODE_FLAGS);
Lee Schermerhornbea904d2008-04-28 02:13:18 -0700925 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
927 if (vma) {
928 up_read(&current->mm->mmap_sem);
929 vma = NULL;
930 }
931
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 err = 0;
Miao Xie58568d22009-06-16 15:31:49 -0700933 if (nmask) {
Lee Schermerhornc6b6ef82010-03-23 13:35:41 -0700934 if (mpol_store_user_nodemask(pol)) {
935 *nmask = pol->w.user_nodemask;
936 } else {
937 task_lock(current);
938 get_policy_nodemask(pol, nmask);
939 task_unlock(current);
940 }
Miao Xie58568d22009-06-16 15:31:49 -0700941 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
943 out:
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700944 mpol_cond_put(pol);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 if (vma)
946 up_read(&current->mm->mmap_sem);
947 return err;
948}
949
Christoph Lameterb20a3502006-03-22 00:09:12 -0800950#ifdef CONFIG_MIGRATION
Christoph Lameter8bccd852005-10-29 18:16:59 -0700951/*
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800952 * page migration
953 */
Christoph Lameterfc301282006-01-18 17:42:29 -0800954static void migrate_page_add(struct page *page, struct list_head *pagelist,
955 unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800956{
957 /*
Christoph Lameterfc301282006-01-18 17:42:29 -0800958 * Avoid migrating a page that is shared with others.
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800959 */
Nick Piggin62695a82008-10-18 20:26:09 -0700960 if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) {
961 if (!isolate_lru_page(page)) {
962 list_add_tail(&page->lru, pagelist);
Mel Gorman599d0c92016-07-28 15:45:31 -0700963 inc_node_page_state(page, NR_ISOLATED_ANON +
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -0800964 page_is_file_cache(page));
Nick Piggin62695a82008-10-18 20:26:09 -0700965 }
966 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800967}
968
Christoph Lameter742755a2006-06-23 02:03:55 -0700969static struct page *new_node_page(struct page *page, unsigned long node, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -0700970{
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -0700971 if (PageHuge(page))
972 return alloc_huge_page_node(page_hstate(compound_head(page)),
973 node);
974 else
Vlastimil Babka96db8002015-09-08 15:03:50 -0700975 return __alloc_pages_node(node, GFP_HIGHUSER_MOVABLE |
David Rientjesb360edb2015-04-14 15:46:52 -0700976 __GFP_THISNODE, 0);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700977}
978
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -0800979/*
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800980 * Migrate pages from one node to a target node.
981 * Returns error or the number of pages not migrated.
982 */
Adrian Bunkdbcb0f12007-10-16 01:26:26 -0700983static int migrate_to_node(struct mm_struct *mm, int source, int dest,
984 int flags)
Christoph Lameter7e2ab152006-02-01 03:05:40 -0800985{
986 nodemask_t nmask;
987 LIST_HEAD(pagelist);
988 int err = 0;
989
990 nodes_clear(nmask);
991 node_set(source, nmask);
992
Minchan Kim08270802012-10-08 16:33:38 -0700993 /*
994 * This does not "check" the range but isolates all pages that
995 * need migration. Between passing in the full user address
996 * space range and MPOL_MF_DISCONTIG_OK, this call can not fail.
997 */
998 VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)));
Naoya Horiguchi98094942013-09-11 14:22:14 -0700999 queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask,
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001000 flags | MPOL_MF_DISCONTIG_OK, &pagelist);
1001
Minchan Kimcf608ac2010-10-26 14:21:29 -07001002 if (!list_empty(&pagelist)) {
David Rientjes68711a72014-06-04 16:08:25 -07001003 err = migrate_pages(&pagelist, new_node_page, NULL, dest,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001004 MIGRATE_SYNC, MR_SYSCALL);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001005 if (err)
Naoya Horiguchie2d8cf42013-09-11 14:22:03 -07001006 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001007 }
Christoph Lameter95a402c2006-06-23 02:03:53 -07001008
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001009 return err;
1010}
1011
1012/*
1013 * Move pages between the two nodesets so as to preserve the physical
1014 * layout as much as possible.
Christoph Lameter39743882006-01-08 01:00:51 -08001015 *
1016 * Returns the number of page that could not be moved.
1017 */
Andrew Morton0ce72d42012-05-29 15:06:24 -07001018int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1019 const nodemask_t *to, int flags)
Christoph Lameter39743882006-01-08 01:00:51 -08001020{
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001021 int busy = 0;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001022 int err;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001023 nodemask_t tmp;
Christoph Lameter39743882006-01-08 01:00:51 -08001024
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001025 err = migrate_prep();
1026 if (err)
1027 return err;
1028
Lee Schermerhorn53f25562008-04-28 02:13:20 -07001029 down_read(&mm->mmap_sem);
Christoph Lameter39743882006-01-08 01:00:51 -08001030
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -08001031 /*
1032 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
1033 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
1034 * bit in 'tmp', and return that <source, dest> pair for migration.
1035 * The pair of nodemasks 'to' and 'from' define the map.
1036 *
1037 * If no pair of bits is found that way, fallback to picking some
1038 * pair of 'source' and 'dest' bits that are not the same. If the
1039 * 'source' and 'dest' bits are the same, this represents a node
1040 * that will be migrating to itself, so no pages need move.
1041 *
1042 * If no bits are left in 'tmp', or if all remaining bits left
1043 * in 'tmp' correspond to the same bit in 'to', return false
1044 * (nothing left to migrate).
1045 *
1046 * This lets us pick a pair of nodes to migrate between, such that
1047 * if possible the dest node is not already occupied by some other
1048 * source node, minimizing the risk of overloading the memory on a
1049 * node that would happen if we migrated incoming memory to a node
1050 * before migrating outgoing memory source that same node.
1051 *
1052 * A single scan of tmp is sufficient. As we go, we remember the
1053 * most recent <s, d> pair that moved (s != d). If we find a pair
1054 * that not only moved, but what's better, moved to an empty slot
1055 * (d is not set in tmp), then we break out then, with that pair.
Justin P. Mattockae0e47f2011-03-01 15:06:02 +01001056 * Otherwise when we finish scanning from_tmp, we at least have the
KOSAKI Motohiroda0aa132010-03-05 13:41:59 -08001057 * most recent <s, d> pair that moved. If we get all the way through
1058 * the scan of tmp without finding any node that moved, much less
1059 * moved to an empty node, then there is nothing left worth migrating.
1060 */
Christoph Lameterd4984712006-01-08 01:00:55 -08001061
Andrew Morton0ce72d42012-05-29 15:06:24 -07001062 tmp = *from;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001063 while (!nodes_empty(tmp)) {
1064 int s,d;
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001065 int source = NUMA_NO_NODE;
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001066 int dest = 0;
1067
1068 for_each_node_mask(s, tmp) {
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001069
1070 /*
1071 * do_migrate_pages() tries to maintain the relative
1072 * node relationship of the pages established between
1073 * threads and memory areas.
1074 *
1075 * However if the number of source nodes is not equal to
1076 * the number of destination nodes we can not preserve
1077 * this node relative relationship. In that case, skip
1078 * copying memory from a node that is in the destination
1079 * mask.
1080 *
1081 * Example: [2,3,4] -> [3,4,5] moves everything.
1082 * [0-7] - > [3,4,5] moves only 0,1,2,6,7.
1083 */
1084
Andrew Morton0ce72d42012-05-29 15:06:24 -07001085 if ((nodes_weight(*from) != nodes_weight(*to)) &&
1086 (node_isset(s, *to)))
Larry Woodman4a5b18c2012-05-29 15:06:24 -07001087 continue;
1088
Andrew Morton0ce72d42012-05-29 15:06:24 -07001089 d = node_remap(s, *from, *to);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001090 if (s == d)
1091 continue;
1092
1093 source = s; /* Node moved. Memorize */
1094 dest = d;
1095
1096 /* dest not in remaining from nodes? */
1097 if (!node_isset(dest, tmp))
1098 break;
1099 }
Jianguo Wub76ac7e2013-11-12 15:07:39 -08001100 if (source == NUMA_NO_NODE)
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001101 break;
1102
1103 node_clear(source, tmp);
1104 err = migrate_to_node(mm, source, dest, flags);
1105 if (err > 0)
1106 busy += err;
1107 if (err < 0)
1108 break;
Christoph Lameter39743882006-01-08 01:00:51 -08001109 }
1110 up_read(&mm->mmap_sem);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001111 if (err < 0)
1112 return err;
1113 return busy;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001114
Christoph Lameter39743882006-01-08 01:00:51 -08001115}
1116
Lee Schermerhorn3ad33b242007-11-14 16:59:10 -08001117/*
1118 * Allocate a new page for page migration based on vma policy.
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001119 * Start by assuming the page is mapped by the same vma as contains @start.
Lee Schermerhorn3ad33b242007-11-14 16:59:10 -08001120 * Search forward from there, if not. N.B., this assumes that the
1121 * list of pages handed to migrate_pages()--which is how we get here--
1122 * is in virtual address order.
1123 */
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001124static struct page *new_page(struct page *page, unsigned long start, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001125{
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001126 struct vm_area_struct *vma;
Lee Schermerhorn3ad33b242007-11-14 16:59:10 -08001127 unsigned long uninitialized_var(address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001128
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001129 vma = find_vma(current->mm, start);
Lee Schermerhorn3ad33b242007-11-14 16:59:10 -08001130 while (vma) {
1131 address = page_address_in_vma(page, vma);
1132 if (address != -EFAULT)
1133 break;
1134 vma = vma->vm_next;
1135 }
1136
Wanpeng Li11c731e2013-12-18 17:08:56 -08001137 if (PageHuge(page)) {
Michal Hockocc817172014-01-23 15:53:15 -08001138 BUG_ON(!vma);
1139 return alloc_huge_page_noerr(vma, address, 1);
Wanpeng Li11c731e2013-12-18 17:08:56 -08001140 }
1141 /*
1142 * if !vma, alloc_page_vma() will use task or system default policy
1143 */
Lee Schermerhorn3ad33b242007-11-14 16:59:10 -08001144 return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
Christoph Lameter95a402c2006-06-23 02:03:53 -07001145}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001146#else
1147
1148static void migrate_page_add(struct page *page, struct list_head *pagelist,
1149 unsigned long flags)
1150{
1151}
1152
Andrew Morton0ce72d42012-05-29 15:06:24 -07001153int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1154 const nodemask_t *to, int flags)
Christoph Lameterb20a3502006-03-22 00:09:12 -08001155{
1156 return -ENOSYS;
1157}
Christoph Lameter95a402c2006-06-23 02:03:53 -07001158
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001159static struct page *new_page(struct page *page, unsigned long start, int **x)
Christoph Lameter95a402c2006-06-23 02:03:53 -07001160{
1161 return NULL;
1162}
Christoph Lameterb20a3502006-03-22 00:09:12 -08001163#endif
1164
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001165static long do_mbind(unsigned long start, unsigned long len,
David Rientjes028fec42008-04-28 02:12:25 -07001166 unsigned short mode, unsigned short mode_flags,
1167 nodemask_t *nmask, unsigned long flags)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001168{
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001169 struct mm_struct *mm = current->mm;
1170 struct mempolicy *new;
1171 unsigned long end;
1172 int err;
1173 LIST_HEAD(pagelist);
1174
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001175 if (flags & ~(unsigned long)MPOL_MF_VALID)
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001176 return -EINVAL;
Christoph Lameter74c00242006-03-14 19:50:21 -08001177 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001178 return -EPERM;
1179
1180 if (start & ~PAGE_MASK)
1181 return -EINVAL;
1182
1183 if (mode == MPOL_DEFAULT)
1184 flags &= ~MPOL_MF_STRICT;
1185
1186 len = (len + PAGE_SIZE - 1) & PAGE_MASK;
1187 end = start + len;
1188
1189 if (end < start)
1190 return -EINVAL;
1191 if (end == start)
1192 return 0;
1193
David Rientjes028fec42008-04-28 02:12:25 -07001194 new = mpol_new(mode, mode_flags, nmask);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001195 if (IS_ERR(new))
1196 return PTR_ERR(new);
1197
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001198 if (flags & MPOL_MF_LAZY)
1199 new->flags |= MPOL_F_MOF;
1200
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001201 /*
1202 * If we are using the default policy then operation
1203 * on discontinuous address spaces is okay after all
1204 */
1205 if (!new)
1206 flags |= MPOL_MF_DISCONTIG_OK;
1207
David Rientjes028fec42008-04-28 02:12:25 -07001208 pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
1209 start, start + len, mode, mode_flags,
David Rientjes00ef2d22013-02-22 16:35:36 -08001210 nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001211
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001212 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
1213
1214 err = migrate_prep();
1215 if (err)
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001216 goto mpol_out;
Christoph Lameter0aedadf2008-11-06 12:53:30 -08001217 }
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07001218 {
1219 NODEMASK_SCRATCH(scratch);
1220 if (scratch) {
1221 down_write(&mm->mmap_sem);
1222 task_lock(current);
1223 err = mpol_set_nodemask(new, nmask, scratch);
1224 task_unlock(current);
1225 if (err)
1226 up_write(&mm->mmap_sem);
1227 } else
1228 err = -ENOMEM;
1229 NODEMASK_SCRATCH_FREE(scratch);
1230 }
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001231 if (err)
1232 goto mpol_out;
1233
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001234 err = queue_pages_range(mm, start, end, nmask,
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001235 flags | MPOL_MF_INVERT, &pagelist);
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001236 if (!err)
KOSAKI Motohiro9d8cebd2010-03-05 13:41:57 -08001237 err = mbind_range(mm, start, end, new);
Christoph Lameter7e2ab152006-02-01 03:05:40 -08001238
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001239 if (!err) {
1240 int nr_failed = 0;
1241
Minchan Kimcf608ac2010-10-26 14:21:29 -07001242 if (!list_empty(&pagelist)) {
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001243 WARN_ON_ONCE(flags & MPOL_MF_LAZY);
Hugh Dickinsd05f0cdcb2014-06-23 13:22:07 -07001244 nr_failed = migrate_pages(&pagelist, new_page, NULL,
1245 start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001246 if (nr_failed)
Naoya Horiguchi74060e42013-09-11 14:22:06 -07001247 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001248 }
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001249
Lee Schermerhornb24f53a2012-10-25 14:16:32 +02001250 if (nr_failed && (flags & MPOL_MF_STRICT))
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001251 err = -EIO;
KOSAKI Motohiroab8a3e12009-10-26 16:49:58 -07001252 } else
Joonsoo Kimb0e5fd72013-12-18 17:08:51 -08001253 putback_movable_pages(&pagelist);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001254
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001255 up_write(&mm->mmap_sem);
KOSAKI Motohirob05ca732009-10-26 16:49:59 -07001256 mpol_out:
Lee Schermerhornf0be3d32008-04-28 02:13:08 -07001257 mpol_put(new);
Christoph Lameter6ce3c4c2006-01-08 01:01:04 -08001258 return err;
1259}
1260
Christoph Lameter39743882006-01-08 01:00:51 -08001261/*
Christoph Lameter8bccd852005-10-29 18:16:59 -07001262 * User space interface with variable sized bitmaps for nodelists.
1263 */
1264
1265/* Copy a node mask from user space. */
Christoph Lameter39743882006-01-08 01:00:51 -08001266static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
Christoph Lameter8bccd852005-10-29 18:16:59 -07001267 unsigned long maxnode)
1268{
1269 unsigned long k;
1270 unsigned long nlongs;
1271 unsigned long endmask;
1272
1273 --maxnode;
1274 nodes_clear(*nodes);
1275 if (maxnode == 0 || !nmask)
1276 return 0;
Andi Kleena9c930b2006-02-20 18:27:59 -08001277 if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
Chris Wright636f13c2006-02-17 13:59:36 -08001278 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001279
1280 nlongs = BITS_TO_LONGS(maxnode);
1281 if ((maxnode % BITS_PER_LONG) == 0)
1282 endmask = ~0UL;
1283 else
1284 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
1285
1286 /* When the user specified more nodes than supported just check
1287 if the non supported part is all zero. */
1288 if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
1289 if (nlongs > PAGE_SIZE/sizeof(long))
1290 return -EINVAL;
1291 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
1292 unsigned long t;
1293 if (get_user(t, nmask + k))
1294 return -EFAULT;
1295 if (k == nlongs - 1) {
1296 if (t & endmask)
1297 return -EINVAL;
1298 } else if (t)
1299 return -EINVAL;
1300 }
1301 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
1302 endmask = ~0UL;
1303 }
1304
1305 if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
1306 return -EFAULT;
1307 nodes_addr(*nodes)[nlongs-1] &= endmask;
1308 return 0;
1309}
1310
1311/* Copy a kernel node mask to user space */
1312static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1313 nodemask_t *nodes)
1314{
1315 unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1316 const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long);
1317
1318 if (copy > nbytes) {
1319 if (copy > PAGE_SIZE)
1320 return -EINVAL;
1321 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1322 return -EFAULT;
1323 copy = nbytes;
1324 }
1325 return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1326}
1327
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001328SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
Rasmus Villemoesf7f28ca2014-06-04 16:07:57 -07001329 unsigned long, mode, const unsigned long __user *, nmask,
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001330 unsigned long, maxnode, unsigned, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001331{
1332 nodemask_t nodes;
1333 int err;
David Rientjes028fec42008-04-28 02:12:25 -07001334 unsigned short mode_flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001335
David Rientjes028fec42008-04-28 02:12:25 -07001336 mode_flags = mode & MPOL_MODE_FLAGS;
1337 mode &= ~MPOL_MODE_FLAGS;
David Rientjesa3b51e02008-04-28 02:12:23 -07001338 if (mode >= MPOL_MAX)
1339 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001340 if ((mode_flags & MPOL_F_STATIC_NODES) &&
1341 (mode_flags & MPOL_F_RELATIVE_NODES))
1342 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001343 err = get_nodes(&nodes, nmask, maxnode);
1344 if (err)
1345 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001346 return do_mbind(start, len, mode, mode_flags, &nodes, flags);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001347}
1348
1349/* Set the process memory policy */
Rasmus Villemoes23c89022014-06-04 16:07:58 -07001350SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask,
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001351 unsigned long, maxnode)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001352{
1353 int err;
1354 nodemask_t nodes;
David Rientjes028fec42008-04-28 02:12:25 -07001355 unsigned short flags;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001356
David Rientjes028fec42008-04-28 02:12:25 -07001357 flags = mode & MPOL_MODE_FLAGS;
1358 mode &= ~MPOL_MODE_FLAGS;
1359 if ((unsigned int)mode >= MPOL_MAX)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001360 return -EINVAL;
David Rientjes4c50bc02008-04-28 02:12:30 -07001361 if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
1362 return -EINVAL;
Christoph Lameter8bccd852005-10-29 18:16:59 -07001363 err = get_nodes(&nodes, nmask, maxnode);
1364 if (err)
1365 return err;
David Rientjes028fec42008-04-28 02:12:25 -07001366 return do_set_mempolicy(mode, flags, &nodes);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001367}
1368
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001369SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
1370 const unsigned long __user *, old_nodes,
1371 const unsigned long __user *, new_nodes)
Christoph Lameter39743882006-01-08 01:00:51 -08001372{
David Howellsc69e8d92008-11-14 10:39:19 +11001373 const struct cred *cred = current_cred(), *tcred;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001374 struct mm_struct *mm = NULL;
Christoph Lameter39743882006-01-08 01:00:51 -08001375 struct task_struct *task;
Christoph Lameter39743882006-01-08 01:00:51 -08001376 nodemask_t task_nodes;
1377 int err;
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001378 nodemask_t *old;
1379 nodemask_t *new;
1380 NODEMASK_SCRATCH(scratch);
Christoph Lameter39743882006-01-08 01:00:51 -08001381
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001382 if (!scratch)
1383 return -ENOMEM;
Christoph Lameter39743882006-01-08 01:00:51 -08001384
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001385 old = &scratch->mask1;
1386 new = &scratch->mask2;
1387
1388 err = get_nodes(old, old_nodes, maxnode);
Christoph Lameter39743882006-01-08 01:00:51 -08001389 if (err)
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001390 goto out;
1391
1392 err = get_nodes(new, new_nodes, maxnode);
1393 if (err)
1394 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001395
1396 /* Find the mm_struct */
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001397 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001398 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter39743882006-01-08 01:00:51 -08001399 if (!task) {
Zeng Zhaoming55cfaa32010-12-02 14:31:13 -08001400 rcu_read_unlock();
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001401 err = -ESRCH;
1402 goto out;
Christoph Lameter39743882006-01-08 01:00:51 -08001403 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001404 get_task_struct(task);
Christoph Lameter39743882006-01-08 01:00:51 -08001405
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001406 err = -EINVAL;
Christoph Lameter39743882006-01-08 01:00:51 -08001407
1408 /*
1409 * Check if this process has the right to modify the specified
1410 * process. The right exists if the process has administrative
Alexey Dobriyan7f927fc2006-03-28 01:56:53 -08001411 * capabilities, superuser privileges or the same
Christoph Lameter39743882006-01-08 01:00:51 -08001412 * userid as the target process.
1413 */
David Howellsc69e8d92008-11-14 10:39:19 +11001414 tcred = __task_cred(task);
Eric W. Biedermanb38a86e2012-03-12 15:48:24 -07001415 if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1416 !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) &&
Christoph Lameter74c00242006-03-14 19:50:21 -08001417 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001418 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001419 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001420 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001421 }
David Howellsc69e8d92008-11-14 10:39:19 +11001422 rcu_read_unlock();
Christoph Lameter39743882006-01-08 01:00:51 -08001423
1424 task_nodes = cpuset_mems_allowed(task);
1425 /* Is the user allowed to access the target nodes? */
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001426 if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
Christoph Lameter39743882006-01-08 01:00:51 -08001427 err = -EPERM;
Christoph Lameter3268c632012-03-21 16:34:06 -07001428 goto out_put;
Christoph Lameter39743882006-01-08 01:00:51 -08001429 }
1430
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08001431 if (!nodes_subset(*new, node_states[N_MEMORY])) {
Christoph Lameter3b42d282007-08-31 00:12:08 -07001432 err = -EINVAL;
Christoph Lameter3268c632012-03-21 16:34:06 -07001433 goto out_put;
Christoph Lameter3b42d282007-08-31 00:12:08 -07001434 }
1435
David Quigley86c3a762006-06-23 02:04:02 -07001436 err = security_task_movememory(task);
1437 if (err)
Christoph Lameter3268c632012-03-21 16:34:06 -07001438 goto out_put;
David Quigley86c3a762006-06-23 02:04:02 -07001439
Christoph Lameter3268c632012-03-21 16:34:06 -07001440 mm = get_task_mm(task);
1441 put_task_struct(task);
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001442
1443 if (!mm) {
Christoph Lameter3268c632012-03-21 16:34:06 -07001444 err = -EINVAL;
Sasha Levinf2a9ef82012-04-25 16:01:52 -07001445 goto out;
1446 }
1447
1448 err = do_migrate_pages(mm, old, new,
1449 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
Christoph Lameter3268c632012-03-21 16:34:06 -07001450
1451 mmput(mm);
1452out:
KOSAKI Motohiro596d7cf2010-08-09 17:19:01 -07001453 NODEMASK_SCRATCH_FREE(scratch);
1454
Christoph Lameter39743882006-01-08 01:00:51 -08001455 return err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001456
1457out_put:
1458 put_task_struct(task);
1459 goto out;
1460
Christoph Lameter39743882006-01-08 01:00:51 -08001461}
1462
1463
Christoph Lameter8bccd852005-10-29 18:16:59 -07001464/* Retrieve NUMA policy */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001465SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1466 unsigned long __user *, nmask, unsigned long, maxnode,
1467 unsigned long, addr, unsigned long, flags)
Christoph Lameter8bccd852005-10-29 18:16:59 -07001468{
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07001469 int err;
1470 int uninitialized_var(pval);
Christoph Lameter8bccd852005-10-29 18:16:59 -07001471 nodemask_t nodes;
1472
1473 if (nmask != NULL && maxnode < MAX_NUMNODES)
1474 return -EINVAL;
1475
1476 err = do_get_mempolicy(&pval, &nodes, addr, flags);
1477
1478 if (err)
1479 return err;
1480
1481 if (policy && put_user(pval, policy))
1482 return -EFAULT;
1483
1484 if (nmask)
1485 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1486
1487 return err;
1488}
1489
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490#ifdef CONFIG_COMPAT
1491
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001492COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1493 compat_ulong_t __user *, nmask,
1494 compat_ulong_t, maxnode,
1495 compat_ulong_t, addr, compat_ulong_t, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496{
1497 long err;
1498 unsigned long __user *nm = NULL;
1499 unsigned long nr_bits, alloc_size;
1500 DECLARE_BITMAP(bm, MAX_NUMNODES);
1501
1502 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1503 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1504
1505 if (nmask)
1506 nm = compat_alloc_user_space(alloc_size);
1507
1508 err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
1509
1510 if (!err && nmask) {
KAMEZAWA Hiroyuki2bbff6c2011-09-14 16:21:02 -07001511 unsigned long copy_size;
1512 copy_size = min_t(unsigned long, sizeof(bm), alloc_size);
1513 err = copy_from_user(bm, nm, copy_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 /* ensure entire bitmap is zeroed */
1515 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
1516 err |= compat_put_bitmap(nmask, bm, nr_bits);
1517 }
1518
1519 return err;
1520}
1521
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001522COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask,
1523 compat_ulong_t, maxnode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 unsigned long __user *nm = NULL;
1526 unsigned long nr_bits, alloc_size;
1527 DECLARE_BITMAP(bm, MAX_NUMNODES);
1528
1529 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1530 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1531
1532 if (nmask) {
Chris Sallscf01fb92017-04-07 23:48:11 -07001533 if (compat_get_bitmap(bm, nmask, nr_bits))
1534 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 nm = compat_alloc_user_space(alloc_size);
Chris Sallscf01fb92017-04-07 23:48:11 -07001536 if (copy_to_user(nm, bm, alloc_size))
1537 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 }
1539
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 return sys_set_mempolicy(mode, nm, nr_bits+1);
1541}
1542
Heiko Carstensc93e0f62014-03-03 16:32:26 +01001543COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len,
1544 compat_ulong_t, mode, compat_ulong_t __user *, nmask,
1545 compat_ulong_t, maxnode, compat_ulong_t, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 unsigned long __user *nm = NULL;
1548 unsigned long nr_bits, alloc_size;
Andi Kleendfcd3c02005-10-29 18:15:48 -07001549 nodemask_t bm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
1551 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1552 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1553
1554 if (nmask) {
Chris Sallscf01fb92017-04-07 23:48:11 -07001555 if (compat_get_bitmap(nodes_addr(bm), nmask, nr_bits))
1556 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 nm = compat_alloc_user_space(alloc_size);
Chris Sallscf01fb92017-04-07 23:48:11 -07001558 if (copy_to_user(nm, nodes_addr(bm), alloc_size))
1559 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 }
1561
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
1563}
1564
1565#endif
1566
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001567struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
1568 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569{
Oleg Nesterov8d902742014-10-09 15:27:45 -07001570 struct mempolicy *pol = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
1572 if (vma) {
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001573 if (vma->vm_ops && vma->vm_ops->get_policy) {
Oleg Nesterov8d902742014-10-09 15:27:45 -07001574 pol = vma->vm_ops->get_policy(vma, addr);
Mel Gorman00442ad2012-10-08 16:29:20 -07001575 } else if (vma->vm_policy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 pol = vma->vm_policy;
Mel Gorman00442ad2012-10-08 16:29:20 -07001577
1578 /*
1579 * shmem_alloc_page() passes MPOL_F_SHARED policy with
1580 * a pseudo vma whose vma->vm_ops=NULL. Take a reference
1581 * count on these policies which will be dropped by
1582 * mpol_cond_put() later
1583 */
1584 if (mpol_needs_cond_ref(pol))
1585 mpol_get(pol);
1586 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 }
Oleg Nesterovf15ca782014-10-09 15:27:43 -07001588
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001589 return pol;
1590}
1591
1592/*
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001593 * get_vma_policy(@vma, @addr)
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001594 * @vma: virtual memory area whose policy is sought
1595 * @addr: address in @vma for shared policy lookup
1596 *
1597 * Returns effective policy for a VMA at specified address.
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001598 * Falls back to current->mempolicy or system default policy, as necessary.
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001599 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1600 * count--added by the get_policy() vm_op, as appropriate--to protect against
1601 * freeing by another task. It is the caller's responsibility to free the
1602 * extra reference for shared policies.
1603 */
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001604static struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
1605 unsigned long addr)
Oleg Nesterov74d2c3a2014-10-09 15:27:50 -07001606{
1607 struct mempolicy *pol = __get_vma_policy(vma, addr);
1608
Oleg Nesterov8d902742014-10-09 15:27:45 -07001609 if (!pol)
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001610 pol = get_task_policy(current);
Oleg Nesterov8d902742014-10-09 15:27:45 -07001611
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 return pol;
1613}
1614
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001615bool vma_policy_mof(struct vm_area_struct *vma)
Mel Gormanfc3147242013-10-07 11:29:09 +01001616{
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001617 struct mempolicy *pol;
Oleg Nesterovf15ca782014-10-09 15:27:43 -07001618
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001619 if (vma->vm_ops && vma->vm_ops->get_policy) {
1620 bool ret = false;
Mel Gormanfc3147242013-10-07 11:29:09 +01001621
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001622 pol = vma->vm_ops->get_policy(vma, vma->vm_start);
1623 if (pol && (pol->flags & MPOL_F_MOF))
1624 ret = true;
1625 mpol_cond_put(pol);
Mel Gormanfc3147242013-10-07 11:29:09 +01001626
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001627 return ret;
Mel Gormanfc3147242013-10-07 11:29:09 +01001628 }
1629
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001630 pol = vma->vm_policy;
Oleg Nesterov8d902742014-10-09 15:27:45 -07001631 if (!pol)
Oleg Nesterov6b6482b2014-10-09 15:27:48 -07001632 pol = get_task_policy(current);
Oleg Nesterov8d902742014-10-09 15:27:45 -07001633
Mel Gormanfc3147242013-10-07 11:29:09 +01001634 return pol->flags & MPOL_F_MOF;
1635}
1636
Lai Jiangshand3eb1572013-02-22 16:33:22 -08001637static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone)
1638{
1639 enum zone_type dynamic_policy_zone = policy_zone;
1640
1641 BUG_ON(dynamic_policy_zone == ZONE_MOVABLE);
1642
1643 /*
1644 * if policy->v.nodes has movable memory only,
1645 * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only.
1646 *
1647 * policy->v.nodes is intersect with node_states[N_MEMORY].
1648 * so if the following test faile, it implies
1649 * policy->v.nodes has movable memory only.
1650 */
1651 if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY]))
1652 dynamic_policy_zone = ZONE_MOVABLE;
1653
1654 return zone >= dynamic_policy_zone;
1655}
1656
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001657/*
1658 * Return a nodemask representing a mempolicy for filtering nodes for
1659 * page allocation
1660 */
1661static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
Mel Gorman19770b32008-04-28 02:12:18 -07001662{
1663 /* Lower zones don't get a nodemask applied for MPOL_BIND */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07001664 if (unlikely(policy->mode == MPOL_BIND) &&
Lai Jiangshand3eb1572013-02-22 16:33:22 -08001665 apply_policy_zone(policy, gfp_zone(gfp)) &&
Mel Gorman19770b32008-04-28 02:12:18 -07001666 cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
1667 return &policy->v.nodes;
1668
1669 return NULL;
1670}
1671
Vlastimil Babka04ec6262017-07-06 15:40:03 -07001672/* Return the node id preferred by the given mempolicy, or the given id */
1673static int policy_node(gfp_t gfp, struct mempolicy *policy,
1674 int nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675{
Michal Hocko6d840952016-12-12 16:42:23 -08001676 if (policy->mode == MPOL_PREFERRED && !(policy->flags & MPOL_F_LOCAL))
1677 nd = policy->v.preferred_node;
1678 else {
Mel Gorman19770b32008-04-28 02:12:18 -07001679 /*
Michal Hocko6d840952016-12-12 16:42:23 -08001680 * __GFP_THISNODE shouldn't even be used with the bind policy
1681 * because we might easily break the expectation to stay on the
1682 * requested node and not break the policy.
Mel Gorman19770b32008-04-28 02:12:18 -07001683 */
Michal Hocko6d840952016-12-12 16:42:23 -08001684 WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 }
Michal Hocko6d840952016-12-12 16:42:23 -08001686
Vlastimil Babka04ec6262017-07-06 15:40:03 -07001687 return nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688}
1689
1690/* Do dynamic interleaving for a process */
1691static unsigned interleave_nodes(struct mempolicy *policy)
1692{
Vlastimil Babka45816682017-07-06 15:39:59 -07001693 unsigned next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 struct task_struct *me = current;
1695
Vlastimil Babka45816682017-07-06 15:39:59 -07001696 next = next_node_in(me->il_prev, policy->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001697 if (next < MAX_NUMNODES)
Vlastimil Babka45816682017-07-06 15:39:59 -07001698 me->il_prev = next;
1699 return next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700}
1701
Christoph Lameterdc85da12006-01-18 17:42:36 -08001702/*
1703 * Depending on the memory policy provide a node from which to allocate the
1704 * next slab entry.
1705 */
David Rientjes2a389612014-04-07 15:37:29 -07001706unsigned int mempolicy_slab_node(void)
Christoph Lameterdc85da12006-01-18 17:42:36 -08001707{
Andi Kleene7b691b2012-06-09 02:40:03 -07001708 struct mempolicy *policy;
David Rientjes2a389612014-04-07 15:37:29 -07001709 int node = numa_mem_id();
Andi Kleene7b691b2012-06-09 02:40:03 -07001710
1711 if (in_interrupt())
David Rientjes2a389612014-04-07 15:37:29 -07001712 return node;
Andi Kleene7b691b2012-06-09 02:40:03 -07001713
1714 policy = current->mempolicy;
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001715 if (!policy || policy->flags & MPOL_F_LOCAL)
David Rientjes2a389612014-04-07 15:37:29 -07001716 return node;
Christoph Lameter765c4502006-09-27 01:50:08 -07001717
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001718 switch (policy->mode) {
1719 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07001720 /*
1721 * handled MPOL_F_LOCAL above
1722 */
1723 return policy->v.preferred_node;
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001724
Christoph Lameterdc85da12006-01-18 17:42:36 -08001725 case MPOL_INTERLEAVE:
1726 return interleave_nodes(policy);
1727
Mel Gormandd1a2392008-04-28 02:12:17 -07001728 case MPOL_BIND: {
Mel Gormanc33d6c02016-05-19 17:14:10 -07001729 struct zoneref *z;
1730
Christoph Lameterdc85da12006-01-18 17:42:36 -08001731 /*
1732 * Follow bind policy behavior and start allocation at the
1733 * first node.
1734 */
Mel Gorman19770b32008-04-28 02:12:18 -07001735 struct zonelist *zonelist;
Mel Gorman19770b32008-04-28 02:12:18 -07001736 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
Aneesh Kumar K.Vc9634cf2016-10-07 16:59:12 -07001737 zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK];
Mel Gormanc33d6c02016-05-19 17:14:10 -07001738 z = first_zones_zonelist(zonelist, highest_zoneidx,
1739 &policy->v.nodes);
1740 return z->zone ? z->zone->node : node;
Mel Gormandd1a2392008-04-28 02:12:17 -07001741 }
Christoph Lameterdc85da12006-01-18 17:42:36 -08001742
Christoph Lameterdc85da12006-01-18 17:42:36 -08001743 default:
Lee Schermerhornbea904d2008-04-28 02:13:18 -07001744 BUG();
Christoph Lameterdc85da12006-01-18 17:42:36 -08001745 }
1746}
1747
Andrew Mortonfee83b32016-05-19 17:11:43 -07001748/*
1749 * Do static interleaving for a VMA with known offset @n. Returns the n'th
1750 * node in pol->v.nodes (starting from n=0), wrapping around if n exceeds the
1751 * number of present nodes.
1752 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753static unsigned offset_il_node(struct mempolicy *pol,
Andrew Mortonfee83b32016-05-19 17:11:43 -07001754 struct vm_area_struct *vma, unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755{
Andi Kleendfcd3c02005-10-29 18:15:48 -07001756 unsigned nnodes = nodes_weight(pol->v.nodes);
David Rientjesf5b087b2008-04-28 02:12:27 -07001757 unsigned target;
Andrew Mortonfee83b32016-05-19 17:11:43 -07001758 int i;
1759 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
David Rientjesf5b087b2008-04-28 02:12:27 -07001761 if (!nnodes)
1762 return numa_node_id();
Andrew Mortonfee83b32016-05-19 17:11:43 -07001763 target = (unsigned int)n % nnodes;
1764 nid = first_node(pol->v.nodes);
1765 for (i = 0; i < target; i++)
Andi Kleendfcd3c02005-10-29 18:15:48 -07001766 nid = next_node(nid, pol->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 return nid;
1768}
1769
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001770/* Determine a node number for interleave */
1771static inline unsigned interleave_nid(struct mempolicy *pol,
1772 struct vm_area_struct *vma, unsigned long addr, int shift)
1773{
1774 if (vma) {
1775 unsigned long off;
1776
Nishanth Aravamudan3b98b082006-08-31 21:27:53 -07001777 /*
1778 * for small pages, there is no difference between
1779 * shift and PAGE_SHIFT, so the bit-shift is safe.
1780 * for huge pages, since vm_pgoff is in units of small
1781 * pages, we need to shift off the always 0 bits to get
1782 * a useful offset.
1783 */
1784 BUG_ON(shift < PAGE_SHIFT);
1785 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001786 off += (addr - vma->vm_start) >> shift;
1787 return offset_il_node(pol, vma, off);
1788 } else
1789 return interleave_nodes(pol);
1790}
1791
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001792#ifdef CONFIG_HUGETLBFS
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001793/*
Vlastimil Babka04ec6262017-07-06 15:40:03 -07001794 * huge_node(@vma, @addr, @gfp_flags, @mpol)
Fabian Frederickb46e14a2014-06-04 16:08:18 -07001795 * @vma: virtual memory area whose policy is sought
1796 * @addr: address in @vma for shared policy lookup and interleave policy
1797 * @gfp_flags: for requested zone
1798 * @mpol: pointer to mempolicy pointer for reference counted mempolicy
1799 * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001800 *
Vlastimil Babka04ec6262017-07-06 15:40:03 -07001801 * Returns a nid suitable for a huge page allocation and a pointer
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001802 * to the struct mempolicy for conditional unref after allocation.
1803 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1804 * @nodemask for filtering the zonelist.
Miao Xiec0ff7452010-05-24 14:32:08 -07001805 *
Mel Gormand26914d2014-04-03 14:47:24 -07001806 * Must be protected by read_mems_allowed_begin()
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001807 */
Vlastimil Babka04ec6262017-07-06 15:40:03 -07001808int huge_node(struct vm_area_struct *vma, unsigned long addr, gfp_t gfp_flags,
1809 struct mempolicy **mpol, nodemask_t **nodemask)
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001810{
Vlastimil Babka04ec6262017-07-06 15:40:03 -07001811 int nid;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001812
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001813 *mpol = get_vma_policy(vma, addr);
Mel Gorman19770b32008-04-28 02:12:18 -07001814 *nodemask = NULL; /* assume !MPOL_BIND */
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001815
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001816 if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
Vlastimil Babka04ec6262017-07-06 15:40:03 -07001817 nid = interleave_nid(*mpol, vma, addr,
1818 huge_page_shift(hstate_vma(vma)));
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001819 } else {
Vlastimil Babka04ec6262017-07-06 15:40:03 -07001820 nid = policy_node(gfp_flags, *mpol, numa_node_id());
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001821 if ((*mpol)->mode == MPOL_BIND)
1822 *nodemask = &(*mpol)->v.nodes;
Lee Schermerhorn480eccf2007-09-18 22:46:47 -07001823 }
Vlastimil Babka04ec6262017-07-06 15:40:03 -07001824 return nid;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001825}
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001826
1827/*
1828 * init_nodemask_of_mempolicy
1829 *
1830 * If the current task's mempolicy is "default" [NULL], return 'false'
1831 * to indicate default policy. Otherwise, extract the policy nodemask
1832 * for 'bind' or 'interleave' policy into the argument nodemask, or
1833 * initialize the argument nodemask to contain the single node for
1834 * 'preferred' or 'local' policy and return 'true' to indicate presence
1835 * of non-default mempolicy.
1836 *
1837 * We don't bother with reference counting the mempolicy [mpol_get/put]
1838 * because the current task is examining it's own mempolicy and a task's
1839 * mempolicy is only ever changed by the task itself.
1840 *
1841 * N.B., it is the caller's responsibility to free a returned nodemask.
1842 */
1843bool init_nodemask_of_mempolicy(nodemask_t *mask)
1844{
1845 struct mempolicy *mempolicy;
1846 int nid;
1847
1848 if (!(mask && current->mempolicy))
1849 return false;
1850
Miao Xiec0ff7452010-05-24 14:32:08 -07001851 task_lock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001852 mempolicy = current->mempolicy;
1853 switch (mempolicy->mode) {
1854 case MPOL_PREFERRED:
1855 if (mempolicy->flags & MPOL_F_LOCAL)
1856 nid = numa_node_id();
1857 else
1858 nid = mempolicy->v.preferred_node;
1859 init_nodemask_of_node(mask, nid);
1860 break;
1861
1862 case MPOL_BIND:
1863 /* Fall through */
1864 case MPOL_INTERLEAVE:
1865 *mask = mempolicy->v.nodes;
1866 break;
1867
1868 default:
1869 BUG();
1870 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001871 task_unlock(current);
Lee Schermerhorn06808b02009-12-14 17:58:21 -08001872
1873 return true;
1874}
Chen, Kenneth W00ac59a2006-02-03 21:51:14 +01001875#endif
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001876
David Rientjes6f48d0eb2010-08-09 17:18:52 -07001877/*
1878 * mempolicy_nodemask_intersects
1879 *
1880 * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default
1881 * policy. Otherwise, check for intersection between mask and the policy
1882 * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local'
1883 * policy, always return true since it may allocate elsewhere on fallback.
1884 *
1885 * Takes task_lock(tsk) to prevent freeing of its mempolicy.
1886 */
1887bool mempolicy_nodemask_intersects(struct task_struct *tsk,
1888 const nodemask_t *mask)
1889{
1890 struct mempolicy *mempolicy;
1891 bool ret = true;
1892
1893 if (!mask)
1894 return ret;
1895 task_lock(tsk);
1896 mempolicy = tsk->mempolicy;
1897 if (!mempolicy)
1898 goto out;
1899
1900 switch (mempolicy->mode) {
1901 case MPOL_PREFERRED:
1902 /*
1903 * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to
1904 * allocate from, they may fallback to other nodes when oom.
1905 * Thus, it's possible for tsk to have allocated memory from
1906 * nodes in mask.
1907 */
1908 break;
1909 case MPOL_BIND:
1910 case MPOL_INTERLEAVE:
1911 ret = nodes_intersects(mempolicy->v.nodes, *mask);
1912 break;
1913 default:
1914 BUG();
1915 }
1916out:
1917 task_unlock(tsk);
1918 return ret;
1919}
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921/* Allocate a page in interleaved policy.
1922 Own path because it needs to do special accounting. */
Andi Kleen662f3a02005-10-29 18:15:49 -07001923static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
1924 unsigned nid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 struct page *page;
1927
Vlastimil Babka04ec6262017-07-06 15:40:03 -07001928 page = __alloc_pages(gfp, order, nid);
1929 if (page && page_to_nid(page) == nid)
Christoph Lameterca889e62006-06-30 01:55:44 -07001930 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 return page;
1932}
1933
1934/**
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001935 * alloc_pages_vma - Allocate a page for a VMA.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 *
1937 * @gfp:
1938 * %GFP_USER user allocation.
1939 * %GFP_KERNEL kernel allocations,
1940 * %GFP_HIGHMEM highmem/user allocations,
1941 * %GFP_FS allocation should not call back into a file system.
1942 * %GFP_ATOMIC don't sleep.
1943 *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001944 * @order:Order of the GFP allocation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 * @vma: Pointer to VMA or NULL if not available.
1946 * @addr: Virtual Address of the allocation. Must be inside the VMA.
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001947 * @node: Which node to prefer for allocation (modulo policy).
1948 * @hugepage: for hugepages try only the preferred node if possible
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 *
1950 * This function allocates a page from the kernel page pool and applies
1951 * a NUMA policy associated with the VMA or the current process.
1952 * When VMA is not NULL caller must hold down_read on the mmap_sem of the
1953 * mm_struct of the VMA to prevent it from going away. Should be used for
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001954 * all allocations for pages that will be mapped into user space. Returns
1955 * NULL when no page can be allocated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 */
1957struct page *
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001958alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001959 unsigned long addr, int node, bool hugepage)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960{
Mel Gormancc9a6c82012-03-21 16:34:11 -07001961 struct mempolicy *pol;
Miao Xiec0ff7452010-05-24 14:32:08 -07001962 struct page *page;
Vlastimil Babka04ec6262017-07-06 15:40:03 -07001963 int preferred_nid;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001964 unsigned int cpuset_mems_cookie;
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001965 nodemask_t *nmask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966
Mel Gormancc9a6c82012-03-21 16:34:11 -07001967retry_cpuset:
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07001968 pol = get_vma_policy(vma, addr);
Mel Gormand26914d2014-04-03 14:47:24 -07001969 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07001970
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001971 if (pol->mode == MPOL_INTERLEAVE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 unsigned nid;
Christoph Lameter5da7ca82006-01-06 00:10:46 -08001973
Andi Kleen8eac5632011-02-25 14:44:28 -08001974 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07001975 mpol_cond_put(pol);
Andrea Arcangeli0bbbc0b2011-01-13 15:47:05 -08001976 page = alloc_page_interleave(gfp, order, nid);
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001977 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 }
Vlastimil Babkabe97a412015-02-11 15:27:15 -08001979
Vlastimil Babka0867a572015-06-24 16:58:48 -07001980 if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) {
1981 int hpage_node = node;
1982
1983 /*
1984 * For hugepage allocation and non-interleave policy which
1985 * allows the current node (or other explicitly preferred
1986 * node) we only try to allocate from the current/preferred
1987 * node and don't fall back to other nodes, as the cost of
1988 * remote accesses would likely offset THP benefits.
1989 *
1990 * If the policy is interleave, or does not allow the current
1991 * node in its nodemask, we allocate the standard way.
1992 */
1993 if (pol->mode == MPOL_PREFERRED &&
1994 !(pol->flags & MPOL_F_LOCAL))
1995 hpage_node = pol->v.preferred_node;
1996
1997 nmask = policy_nodemask(gfp, pol);
1998 if (!nmask || node_isset(hpage_node, *nmask)) {
1999 mpol_cond_put(pol);
Vlastimil Babka96db8002015-09-08 15:03:50 -07002000 page = __alloc_pages_node(hpage_node,
Vlastimil Babka0867a572015-06-24 16:58:48 -07002001 gfp | __GFP_THISNODE, order);
2002 goto out;
2003 }
2004 }
2005
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002006 nmask = policy_nodemask(gfp, pol);
Vlastimil Babka04ec6262017-07-06 15:40:03 -07002007 preferred_nid = policy_node(gfp, pol, node);
2008 page = __alloc_pages_nodemask(gfp, order, preferred_nid, nmask);
Vlastimil Babkad51e9892017-01-24 15:18:18 -08002009 mpol_cond_put(pol);
Vlastimil Babkabe97a412015-02-11 15:27:15 -08002010out:
Mel Gormand26914d2014-04-03 14:47:24 -07002011 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002012 goto retry_cpuset;
Miao Xiec0ff7452010-05-24 14:32:08 -07002013 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014}
2015
2016/**
2017 * alloc_pages_current - Allocate pages.
2018 *
2019 * @gfp:
2020 * %GFP_USER user allocation,
2021 * %GFP_KERNEL kernel allocation,
2022 * %GFP_HIGHMEM highmem allocation,
2023 * %GFP_FS don't call back into a file system.
2024 * %GFP_ATOMIC don't sleep.
2025 * @order: Power of two of allocation size in pages. 0 is a single page.
2026 *
2027 * Allocate a page from the kernel page pool. When not in
2028 * interrupt context and apply the current process NUMA policy.
2029 * Returns NULL when no page can be allocated.
2030 *
Paul Jacksoncf2a473c2006-01-08 01:01:54 -08002031 * Don't call cpuset_update_task_memory_state() unless
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 * 1) it's ok to take cpuset_sem (can WAIT), and
2033 * 2) allocating for current task (not interrupt).
2034 */
Al Virodd0fc662005-10-07 07:46:04 +01002035struct page *alloc_pages_current(gfp_t gfp, unsigned order)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036{
Oleg Nesterov8d902742014-10-09 15:27:45 -07002037 struct mempolicy *pol = &default_policy;
Miao Xiec0ff7452010-05-24 14:32:08 -07002038 struct page *page;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002039 unsigned int cpuset_mems_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
Oleg Nesterov8d902742014-10-09 15:27:45 -07002041 if (!in_interrupt() && !(gfp & __GFP_THISNODE))
2042 pol = get_task_policy(current);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002043
Mel Gormancc9a6c82012-03-21 16:34:11 -07002044retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07002045 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -07002046
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -07002047 /*
2048 * No reference counting needed for current->mempolicy
2049 * nor system default_policy
2050 */
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002051 if (pol->mode == MPOL_INTERLEAVE)
Miao Xiec0ff7452010-05-24 14:32:08 -07002052 page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
2053 else
2054 page = __alloc_pages_nodemask(gfp, order,
Vlastimil Babka04ec6262017-07-06 15:40:03 -07002055 policy_node(gfp, pol, numa_node_id()),
Andi Kleen5c4b4be2011-03-04 17:36:32 -08002056 policy_nodemask(gfp, pol));
Mel Gormancc9a6c82012-03-21 16:34:11 -07002057
Mel Gormand26914d2014-04-03 14:47:24 -07002058 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07002059 goto retry_cpuset;
2060
Miao Xiec0ff7452010-05-24 14:32:08 -07002061 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062}
2063EXPORT_SYMBOL(alloc_pages_current);
2064
Oleg Nesterovef0855d2013-09-11 14:20:14 -07002065int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
2066{
2067 struct mempolicy *pol = mpol_dup(vma_policy(src));
2068
2069 if (IS_ERR(pol))
2070 return PTR_ERR(pol);
2071 dst->vm_policy = pol;
2072 return 0;
2073}
2074
Paul Jackson42253992006-01-08 01:01:59 -08002075/*
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002076 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
Paul Jackson42253992006-01-08 01:01:59 -08002077 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
2078 * with the mems_allowed returned by cpuset_mems_allowed(). This
2079 * keeps mempolicies cpuset relative after its cpuset moves. See
2080 * further kernel/cpuset.c update_nodemask().
Miao Xie708c1bb2010-05-24 14:32:07 -07002081 *
2082 * current's mempolicy may be rebinded by the other task(the task that changes
2083 * cpuset's mems), so we needn't do rebind work for current task.
Paul Jackson42253992006-01-08 01:01:59 -08002084 */
Paul Jackson42253992006-01-08 01:01:59 -08002085
Lee Schermerhorn846a16b2008-04-28 02:13:09 -07002086/* Slow path of a mempolicy duplicate */
2087struct mempolicy *__mpol_dup(struct mempolicy *old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088{
2089 struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2090
2091 if (!new)
2092 return ERR_PTR(-ENOMEM);
Miao Xie708c1bb2010-05-24 14:32:07 -07002093
2094 /* task's mempolicy is protected by alloc_lock */
2095 if (old == current->mempolicy) {
2096 task_lock(current);
2097 *new = *old;
2098 task_unlock(current);
2099 } else
2100 *new = *old;
2101
Paul Jackson42253992006-01-08 01:01:59 -08002102 if (current_cpuset_is_being_rebound()) {
2103 nodemask_t mems = cpuset_mems_allowed(current);
Miao Xie708c1bb2010-05-24 14:32:07 -07002104 if (new->flags & MPOL_F_REBINDING)
2105 mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2);
2106 else
2107 mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE);
Paul Jackson42253992006-01-08 01:01:59 -08002108 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 atomic_set(&new->refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 return new;
2111}
2112
2113/* Slow path of a mempolicy comparison */
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002114bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115{
2116 if (!a || !b)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002117 return false;
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002118 if (a->mode != b->mode)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002119 return false;
Bob Liu19800502010-05-24 14:32:01 -07002120 if (a->flags != b->flags)
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002121 return false;
Bob Liu19800502010-05-24 14:32:01 -07002122 if (mpol_store_user_nodemask(a))
2123 if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask))
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002124 return false;
Bob Liu19800502010-05-24 14:32:01 -07002125
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002126 switch (a->mode) {
Mel Gorman19770b32008-04-28 02:12:18 -07002127 case MPOL_BIND:
2128 /* Fall through */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 case MPOL_INTERLEAVE:
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002130 return !!nodes_equal(a->v.nodes, b->v.nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 case MPOL_PREFERRED:
Namhyung Kim75719662011-03-22 16:33:02 -07002132 return a->v.preferred_node == b->v.preferred_node;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 default:
2134 BUG();
KOSAKI Motohirofcfb4dc2012-01-10 15:08:21 -08002135 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 }
2137}
2138
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 * Shared memory backing store policy support.
2141 *
2142 * Remember policies even when nobody has shared memory mapped.
2143 * The policies are kept in Red-Black tree linked from the inode.
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002144 * They are protected by the sp->lock rwlock, which should be held
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 * for any accesses to the tree.
2146 */
2147
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002148/*
2149 * lookup first element intersecting start-end. Caller holds sp->lock for
2150 * reading or for writing
2151 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152static struct sp_node *
2153sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
2154{
2155 struct rb_node *n = sp->root.rb_node;
2156
2157 while (n) {
2158 struct sp_node *p = rb_entry(n, struct sp_node, nd);
2159
2160 if (start >= p->end)
2161 n = n->rb_right;
2162 else if (end <= p->start)
2163 n = n->rb_left;
2164 else
2165 break;
2166 }
2167 if (!n)
2168 return NULL;
2169 for (;;) {
2170 struct sp_node *w = NULL;
2171 struct rb_node *prev = rb_prev(n);
2172 if (!prev)
2173 break;
2174 w = rb_entry(prev, struct sp_node, nd);
2175 if (w->end <= start)
2176 break;
2177 n = prev;
2178 }
2179 return rb_entry(n, struct sp_node, nd);
2180}
2181
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002182/*
2183 * Insert a new shared policy into the list. Caller holds sp->lock for
2184 * writing.
2185 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186static void sp_insert(struct shared_policy *sp, struct sp_node *new)
2187{
2188 struct rb_node **p = &sp->root.rb_node;
2189 struct rb_node *parent = NULL;
2190 struct sp_node *nd;
2191
2192 while (*p) {
2193 parent = *p;
2194 nd = rb_entry(parent, struct sp_node, nd);
2195 if (new->start < nd->start)
2196 p = &(*p)->rb_left;
2197 else if (new->end > nd->end)
2198 p = &(*p)->rb_right;
2199 else
2200 BUG();
2201 }
2202 rb_link_node(&new->nd, parent, p);
2203 rb_insert_color(&new->nd, &sp->root);
Paul Mundt140d5a42007-07-15 23:38:16 -07002204 pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002205 new->policy ? new->policy->mode : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206}
2207
2208/* Find shared policy intersecting idx */
2209struct mempolicy *
2210mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
2211{
2212 struct mempolicy *pol = NULL;
2213 struct sp_node *sn;
2214
2215 if (!sp->root.rb_node)
2216 return NULL;
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002217 read_lock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 sn = sp_lookup(sp, idx, idx+1);
2219 if (sn) {
2220 mpol_get(sn->policy);
2221 pol = sn->policy;
2222 }
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002223 read_unlock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 return pol;
2225}
2226
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002227static void sp_free(struct sp_node *n)
2228{
2229 mpol_put(n->policy);
2230 kmem_cache_free(sn_cache, n);
2231}
2232
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002233/**
2234 * mpol_misplaced - check whether current page node is valid in policy
2235 *
Fabian Frederickb46e14a2014-06-04 16:08:18 -07002236 * @page: page to be checked
2237 * @vma: vm area where page mapped
2238 * @addr: virtual address where page mapped
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002239 *
2240 * Lookup current policy node id for vma,addr and "compare to" page's
2241 * node id.
2242 *
2243 * Returns:
2244 * -1 - not misplaced, page is in the right node
2245 * node - node id where the page should be
2246 *
2247 * Policy determination "mimics" alloc_page_vma().
2248 * Called from fault path where we know the vma and faulting address.
2249 */
2250int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr)
2251{
2252 struct mempolicy *pol;
Mel Gormanc33d6c02016-05-19 17:14:10 -07002253 struct zoneref *z;
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002254 int curnid = page_to_nid(page);
2255 unsigned long pgoff;
Peter Zijlstra90572892013-10-07 11:29:20 +01002256 int thiscpu = raw_smp_processor_id();
2257 int thisnid = cpu_to_node(thiscpu);
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002258 int polnid = -1;
2259 int ret = -1;
2260
2261 BUG_ON(!vma);
2262
Oleg Nesterovdd6eecb2014-10-09 15:27:57 -07002263 pol = get_vma_policy(vma, addr);
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002264 if (!(pol->flags & MPOL_F_MOF))
2265 goto out;
2266
2267 switch (pol->mode) {
2268 case MPOL_INTERLEAVE:
2269 BUG_ON(addr >= vma->vm_end);
2270 BUG_ON(addr < vma->vm_start);
2271
2272 pgoff = vma->vm_pgoff;
2273 pgoff += (addr - vma->vm_start) >> PAGE_SHIFT;
2274 polnid = offset_il_node(pol, vma, pgoff);
2275 break;
2276
2277 case MPOL_PREFERRED:
2278 if (pol->flags & MPOL_F_LOCAL)
2279 polnid = numa_node_id();
2280 else
2281 polnid = pol->v.preferred_node;
2282 break;
2283
2284 case MPOL_BIND:
Mel Gormanc33d6c02016-05-19 17:14:10 -07002285
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002286 /*
2287 * allows binding to multiple nodes.
2288 * use current page if in policy nodemask,
2289 * else select nearest allowed node, if any.
2290 * If no allowed nodes, use current [!misplaced].
2291 */
2292 if (node_isset(curnid, pol->v.nodes))
2293 goto out;
Mel Gormanc33d6c02016-05-19 17:14:10 -07002294 z = first_zones_zonelist(
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002295 node_zonelist(numa_node_id(), GFP_HIGHUSER),
2296 gfp_zone(GFP_HIGHUSER),
Mel Gormanc33d6c02016-05-19 17:14:10 -07002297 &pol->v.nodes);
2298 polnid = z->zone->node;
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002299 break;
2300
2301 default:
2302 BUG();
2303 }
Mel Gorman5606e382012-11-02 18:19:13 +00002304
2305 /* Migrate the page towards the node whose CPU is referencing it */
Mel Gormane42c8ff2012-11-12 09:17:07 +00002306 if (pol->flags & MPOL_F_MORON) {
Peter Zijlstra90572892013-10-07 11:29:20 +01002307 polnid = thisnid;
Mel Gorman5606e382012-11-02 18:19:13 +00002308
Rik van Riel10f39042014-01-27 17:03:44 -05002309 if (!should_numa_migrate_memory(current, page, curnid, thiscpu))
Rik van Rielde1c9ce2013-10-07 11:29:39 +01002310 goto out;
Mel Gormane42c8ff2012-11-12 09:17:07 +00002311 }
2312
Lee Schermerhorn771fb4d2012-10-25 14:16:30 +02002313 if (curnid != polnid)
2314 ret = polnid;
2315out:
2316 mpol_cond_put(pol);
2317
2318 return ret;
2319}
2320
David Rientjesc11600e2016-09-01 16:15:07 -07002321/*
2322 * Drop the (possibly final) reference to task->mempolicy. It needs to be
2323 * dropped after task->mempolicy is set to NULL so that any allocation done as
2324 * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed
2325 * policy.
2326 */
2327void mpol_put_task_policy(struct task_struct *task)
2328{
2329 struct mempolicy *pol;
2330
2331 task_lock(task);
2332 pol = task->mempolicy;
2333 task->mempolicy = NULL;
2334 task_unlock(task);
2335 mpol_put(pol);
2336}
2337
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338static void sp_delete(struct shared_policy *sp, struct sp_node *n)
2339{
Paul Mundt140d5a42007-07-15 23:38:16 -07002340 pr_debug("deleting %lx-l%lx\n", n->start, n->end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 rb_erase(&n->nd, &sp->root);
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002342 sp_free(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343}
2344
Mel Gorman42288fe2012-12-21 23:10:25 +00002345static void sp_node_init(struct sp_node *node, unsigned long start,
2346 unsigned long end, struct mempolicy *pol)
2347{
2348 node->start = start;
2349 node->end = end;
2350 node->policy = pol;
2351}
2352
Adrian Bunkdbcb0f12007-10-16 01:26:26 -07002353static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
2354 struct mempolicy *pol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355{
KOSAKI Motohiro869833f2012-10-08 16:29:16 -07002356 struct sp_node *n;
2357 struct mempolicy *newpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358
KOSAKI Motohiro869833f2012-10-08 16:29:16 -07002359 n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 if (!n)
2361 return NULL;
KOSAKI Motohiro869833f2012-10-08 16:29:16 -07002362
2363 newpol = mpol_dup(pol);
2364 if (IS_ERR(newpol)) {
2365 kmem_cache_free(sn_cache, n);
2366 return NULL;
2367 }
2368 newpol->flags |= MPOL_F_SHARED;
Mel Gorman42288fe2012-12-21 23:10:25 +00002369 sp_node_init(n, start, end, newpol);
KOSAKI Motohiro869833f2012-10-08 16:29:16 -07002370
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 return n;
2372}
2373
2374/* Replace a policy range. */
2375static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
2376 unsigned long end, struct sp_node *new)
2377{
Mel Gormanb22d1272012-10-08 16:29:17 -07002378 struct sp_node *n;
Mel Gorman42288fe2012-12-21 23:10:25 +00002379 struct sp_node *n_new = NULL;
2380 struct mempolicy *mpol_new = NULL;
Mel Gormanb22d1272012-10-08 16:29:17 -07002381 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382
Mel Gorman42288fe2012-12-21 23:10:25 +00002383restart:
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002384 write_lock(&sp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 n = sp_lookup(sp, start, end);
2386 /* Take care of old policies in the same range. */
2387 while (n && n->start < end) {
2388 struct rb_node *next = rb_next(&n->nd);
2389 if (n->start >= start) {
2390 if (n->end <= end)
2391 sp_delete(sp, n);
2392 else
2393 n->start = end;
2394 } else {
2395 /* Old policy spanning whole new range. */
2396 if (n->end > end) {
Mel Gorman42288fe2012-12-21 23:10:25 +00002397 if (!n_new)
2398 goto alloc_new;
2399
2400 *mpol_new = *n->policy;
2401 atomic_set(&mpol_new->refcnt, 1);
KOSAKI Motohiro78806392013-03-08 12:43:29 -08002402 sp_node_init(n_new, end, n->end, mpol_new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 n->end = start;
Hillf Danton5ca39572013-03-08 12:43:28 -08002404 sp_insert(sp, n_new);
Mel Gorman42288fe2012-12-21 23:10:25 +00002405 n_new = NULL;
2406 mpol_new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 break;
2408 } else
2409 n->end = start;
2410 }
2411 if (!next)
2412 break;
2413 n = rb_entry(next, struct sp_node, nd);
2414 }
2415 if (new)
2416 sp_insert(sp, new);
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002417 write_unlock(&sp->lock);
Mel Gorman42288fe2012-12-21 23:10:25 +00002418 ret = 0;
2419
2420err_out:
2421 if (mpol_new)
2422 mpol_put(mpol_new);
2423 if (n_new)
2424 kmem_cache_free(sn_cache, n_new);
2425
Mel Gormanb22d1272012-10-08 16:29:17 -07002426 return ret;
Mel Gorman42288fe2012-12-21 23:10:25 +00002427
2428alloc_new:
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002429 write_unlock(&sp->lock);
Mel Gorman42288fe2012-12-21 23:10:25 +00002430 ret = -ENOMEM;
2431 n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL);
2432 if (!n_new)
2433 goto err_out;
2434 mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2435 if (!mpol_new)
2436 goto err_out;
2437 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438}
2439
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002440/**
2441 * mpol_shared_policy_init - initialize shared policy for inode
2442 * @sp: pointer to inode shared policy
2443 * @mpol: struct mempolicy to install
2444 *
2445 * Install non-NULL @mpol in inode's shared policy rb-tree.
2446 * On entry, the current task has a reference on a non-NULL @mpol.
2447 * This must be released on exit.
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002448 * This is called at get_inode() calls and we can use GFP_KERNEL.
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002449 */
2450void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
Robin Holt7339ff82006-01-14 13:20:48 -08002451{
Miao Xie58568d22009-06-16 15:31:49 -07002452 int ret;
2453
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002454 sp->root = RB_ROOT; /* empty tree == default mempolicy */
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002455 rwlock_init(&sp->lock);
Robin Holt7339ff82006-01-14 13:20:48 -08002456
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002457 if (mpol) {
2458 struct vm_area_struct pvma;
2459 struct mempolicy *new;
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002460 NODEMASK_SCRATCH(scratch);
Robin Holt7339ff82006-01-14 13:20:48 -08002461
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002462 if (!scratch)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002463 goto put_mpol;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002464 /* contextualize the tmpfs mount point mempolicy */
2465 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002466 if (IS_ERR(new))
Dan Carpenter0cae3452010-05-25 23:42:58 -07002467 goto free_scratch; /* no valid nodemask intersection */
Miao Xie58568d22009-06-16 15:31:49 -07002468
2469 task_lock(current);
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002470 ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
Miao Xie58568d22009-06-16 15:31:49 -07002471 task_unlock(current);
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002472 if (ret)
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002473 goto put_new;
Robin Holt7339ff82006-01-14 13:20:48 -08002474
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002475 /* Create pseudo-vma that contains just the policy */
2476 memset(&pvma, 0, sizeof(struct vm_area_struct));
2477 pvma.vm_end = TASK_SIZE; /* policy covers entire file */
2478 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
Lee Schermerhorn15d77832010-05-24 14:32:04 -07002479
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002480put_new:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002481 mpol_put(new); /* drop initial ref */
Dan Carpenter0cae3452010-05-25 23:42:58 -07002482free_scratch:
KAMEZAWA Hiroyuki4bfc4492009-08-06 15:07:33 -07002483 NODEMASK_SCRATCH_FREE(scratch);
Lee Schermerhorn5c0c1652010-06-29 15:05:30 -07002484put_mpol:
2485 mpol_put(mpol); /* drop our incoming ref on sb mpol */
Robin Holt7339ff82006-01-14 13:20:48 -08002486 }
2487}
2488
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489int mpol_set_shared_policy(struct shared_policy *info,
2490 struct vm_area_struct *vma, struct mempolicy *npol)
2491{
2492 int err;
2493 struct sp_node *new = NULL;
2494 unsigned long sz = vma_pages(vma);
2495
David Rientjes028fec42008-04-28 02:12:25 -07002496 pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 vma->vm_pgoff,
Lee Schermerhorn45c47452008-04-28 02:13:12 -07002498 sz, npol ? npol->mode : -1,
David Rientjes028fec42008-04-28 02:12:25 -07002499 npol ? npol->flags : -1,
David Rientjes00ef2d22013-02-22 16:35:36 -08002500 npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501
2502 if (npol) {
2503 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
2504 if (!new)
2505 return -ENOMEM;
2506 }
2507 err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
2508 if (err && new)
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002509 sp_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 return err;
2511}
2512
2513/* Free a backing policy store on inode delete. */
2514void mpol_free_shared_policy(struct shared_policy *p)
2515{
2516 struct sp_node *n;
2517 struct rb_node *next;
2518
2519 if (!p->root.rb_node)
2520 return;
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002521 write_lock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 next = rb_first(&p->root);
2523 while (next) {
2524 n = rb_entry(next, struct sp_node, nd);
2525 next = rb_next(&n->nd);
KOSAKI Motohiro63f74ca2012-10-08 16:29:19 -07002526 sp_delete(p, n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 }
Nathan Zimmer4a8c7bb2016-01-14 15:18:36 -08002528 write_unlock(&p->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529}
2530
Mel Gorman1a687c22012-11-22 11:16:36 +00002531#ifdef CONFIG_NUMA_BALANCING
Mel Gormanc2976632014-01-29 14:05:42 -08002532static int __initdata numabalancing_override;
Mel Gorman1a687c22012-11-22 11:16:36 +00002533
2534static void __init check_numabalancing_enable(void)
2535{
2536 bool numabalancing_default = false;
2537
2538 if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED))
2539 numabalancing_default = true;
2540
Mel Gormanc2976632014-01-29 14:05:42 -08002541 /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */
2542 if (numabalancing_override)
2543 set_numabalancing_state(numabalancing_override == 1);
2544
Mel Gormanb0dc2b92015-05-14 15:17:09 -07002545 if (num_online_nodes() > 1 && !numabalancing_override) {
Joe Perches756a0252016-03-17 14:19:47 -07002546 pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n",
Mel Gormanc2976632014-01-29 14:05:42 -08002547 numabalancing_default ? "Enabling" : "Disabling");
Mel Gorman1a687c22012-11-22 11:16:36 +00002548 set_numabalancing_state(numabalancing_default);
2549 }
2550}
2551
2552static int __init setup_numabalancing(char *str)
2553{
2554 int ret = 0;
2555 if (!str)
2556 goto out;
Mel Gorman1a687c22012-11-22 11:16:36 +00002557
2558 if (!strcmp(str, "enable")) {
Mel Gormanc2976632014-01-29 14:05:42 -08002559 numabalancing_override = 1;
Mel Gorman1a687c22012-11-22 11:16:36 +00002560 ret = 1;
2561 } else if (!strcmp(str, "disable")) {
Mel Gormanc2976632014-01-29 14:05:42 -08002562 numabalancing_override = -1;
Mel Gorman1a687c22012-11-22 11:16:36 +00002563 ret = 1;
2564 }
2565out:
2566 if (!ret)
Andrew Morton4a404be2014-01-29 14:05:43 -08002567 pr_warn("Unable to parse numa_balancing=\n");
Mel Gorman1a687c22012-11-22 11:16:36 +00002568
2569 return ret;
2570}
2571__setup("numa_balancing=", setup_numabalancing);
2572#else
2573static inline void __init check_numabalancing_enable(void)
2574{
2575}
2576#endif /* CONFIG_NUMA_BALANCING */
2577
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578/* assumes fs == KERNEL_DS */
2579void __init numa_policy_init(void)
2580{
Paul Mundtb71636e2007-07-15 23:38:15 -07002581 nodemask_t interleave_nodes;
2582 unsigned long largest = 0;
2583 int nid, prefer = 0;
2584
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 policy_cache = kmem_cache_create("numa_policy",
2586 sizeof(struct mempolicy),
Paul Mundt20c2df82007-07-20 10:11:58 +09002587 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588
2589 sn_cache = kmem_cache_create("shared_policy_node",
2590 sizeof(struct sp_node),
Paul Mundt20c2df82007-07-20 10:11:58 +09002591 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592
Mel Gorman5606e382012-11-02 18:19:13 +00002593 for_each_node(nid) {
2594 preferred_node_policy[nid] = (struct mempolicy) {
2595 .refcnt = ATOMIC_INIT(1),
2596 .mode = MPOL_PREFERRED,
2597 .flags = MPOL_F_MOF | MPOL_F_MORON,
2598 .v = { .preferred_node = nid, },
2599 };
2600 }
2601
Paul Mundtb71636e2007-07-15 23:38:15 -07002602 /*
2603 * Set interleaving policy for system init. Interleaving is only
2604 * enabled across suitably sized nodes (default is >= 16MB), or
2605 * fall back to the largest node if they're all smaller.
2606 */
2607 nodes_clear(interleave_nodes);
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002608 for_each_node_state(nid, N_MEMORY) {
Paul Mundtb71636e2007-07-15 23:38:15 -07002609 unsigned long total_pages = node_present_pages(nid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610
Paul Mundtb71636e2007-07-15 23:38:15 -07002611 /* Preserve the largest node */
2612 if (largest < total_pages) {
2613 largest = total_pages;
2614 prefer = nid;
2615 }
2616
2617 /* Interleave this node? */
2618 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
2619 node_set(nid, interleave_nodes);
2620 }
2621
2622 /* All too small, use the largest */
2623 if (unlikely(nodes_empty(interleave_nodes)))
2624 node_set(prefer, interleave_nodes);
2625
David Rientjes028fec42008-04-28 02:12:25 -07002626 if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07002627 pr_err("%s: interleaving failed\n", __func__);
Mel Gorman1a687c22012-11-22 11:16:36 +00002628
2629 check_numabalancing_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630}
2631
Christoph Lameter8bccd852005-10-29 18:16:59 -07002632/* Reset policy of current process to default */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633void numa_default_policy(void)
2634{
David Rientjes028fec42008-04-28 02:12:25 -07002635 do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636}
Paul Jackson68860ec2005-10-30 15:02:36 -08002637
Paul Jackson42253992006-01-08 01:01:59 -08002638/*
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002639 * Parse and format mempolicy from/to strings
2640 */
2641
2642/*
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002643 * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag.
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002644 */
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002645static const char * const policy_modes[] =
2646{
2647 [MPOL_DEFAULT] = "default",
2648 [MPOL_PREFERRED] = "prefer",
2649 [MPOL_BIND] = "bind",
2650 [MPOL_INTERLEAVE] = "interleave",
Lee Schermerhornd3a71032012-10-25 14:16:29 +02002651 [MPOL_LOCAL] = "local",
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002652};
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002653
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002654
2655#ifdef CONFIG_TMPFS
2656/**
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002657 * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option.
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002658 * @str: string containing mempolicy to parse
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002659 * @mpol: pointer to struct mempolicy pointer, returned on success.
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002660 *
2661 * Format of input:
2662 * <mode>[=<flags>][:<nodelist>]
2663 *
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002664 * On success, returns 0, else 1
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002665 */
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08002666int mpol_parse_str(char *str, struct mempolicy **mpol)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002667{
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002668 struct mempolicy *new = NULL;
Lee Schermerhornb4652e82010-05-24 14:32:03 -07002669 unsigned short mode;
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002670 unsigned short mode_flags;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002671 nodemask_t nodes;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002672 char *nodelist = strchr(str, ':');
2673 char *flags = strchr(str, '=');
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002674 int err = 1;
2675
2676 if (nodelist) {
2677 /* NUL-terminate mode or flags string */
2678 *nodelist++ = '\0';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002679 if (nodelist_parse(nodelist, nodes))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002680 goto out;
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002681 if (!nodes_subset(nodes, node_states[N_MEMORY]))
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002682 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002683 } else
2684 nodes_clear(nodes);
2685
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002686 if (flags)
2687 *flags++ = '\0'; /* terminate mode string */
2688
Peter Zijlstra479e2802012-10-25 14:16:28 +02002689 for (mode = 0; mode < MPOL_MAX; mode++) {
Lee Schermerhorn345ace92010-05-24 14:32:04 -07002690 if (!strcmp(str, policy_modes[mode])) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002691 break;
2692 }
2693 }
Mel Gormana7200942012-11-16 09:37:58 +00002694 if (mode >= MPOL_MAX)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002695 goto out;
2696
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002697 switch (mode) {
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002698 case MPOL_PREFERRED:
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002699 /*
2700 * Insist on a nodelist of one node only
2701 */
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002702 if (nodelist) {
2703 char *rest = nodelist;
2704 while (isdigit(*rest))
2705 rest++;
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002706 if (*rest)
2707 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002708 }
2709 break;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002710 case MPOL_INTERLEAVE:
2711 /*
2712 * Default to online nodes with memory if no nodelist
2713 */
2714 if (!nodelist)
Lai Jiangshan01f13bd2012-12-12 13:51:33 -08002715 nodes = node_states[N_MEMORY];
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002716 break;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002717 case MPOL_LOCAL:
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002718 /*
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002719 * Don't allow a nodelist; mpol_new() checks flags
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002720 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002721 if (nodelist)
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002722 goto out;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002723 mode = MPOL_PREFERRED;
Lee Schermerhorn3f226aa2008-04-28 02:13:24 -07002724 break;
Ravikiran G Thirumalai413b43d2010-03-23 13:35:28 -07002725 case MPOL_DEFAULT:
2726 /*
2727 * Insist on a empty nodelist
2728 */
2729 if (!nodelist)
2730 err = 0;
2731 goto out;
KOSAKI Motohirod69b2e62010-03-23 13:35:30 -07002732 case MPOL_BIND:
2733 /*
2734 * Insist on a nodelist
2735 */
2736 if (!nodelist)
2737 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002738 }
2739
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002740 mode_flags = 0;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002741 if (flags) {
2742 /*
2743 * Currently, we only support two mutually exclusive
2744 * mode flags.
2745 */
2746 if (!strcmp(flags, "static"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002747 mode_flags |= MPOL_F_STATIC_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002748 else if (!strcmp(flags, "relative"))
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002749 mode_flags |= MPOL_F_RELATIVE_NODES;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002750 else
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002751 goto out;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002752 }
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002753
2754 new = mpol_new(mode, mode_flags, &nodes);
2755 if (IS_ERR(new))
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002756 goto out;
2757
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002758 /*
2759 * Save nodes for mpol_to_str() to show the tmpfs mount options
2760 * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo.
2761 */
2762 if (mode != MPOL_PREFERRED)
2763 new->v.nodes = nodes;
2764 else if (nodelist)
2765 new->v.preferred_node = first_node(nodes);
2766 else
2767 new->flags |= MPOL_F_LOCAL;
2768
2769 /*
2770 * Save nodes for contextualization: this will be used to "clone"
2771 * the mempolicy in a specific context [cpuset] at a later time.
2772 */
2773 new->w.user_nodemask = nodes;
2774
KOSAKI Motohiro926f2ae2010-03-23 13:35:32 -07002775 err = 0;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002776
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002777out:
2778 /* Restore string for error message */
2779 if (nodelist)
2780 *--nodelist = ':';
2781 if (flags)
2782 *--flags = '=';
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002783 if (!err)
2784 *mpol = new;
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07002785 return err;
2786}
2787#endif /* CONFIG_TMPFS */
2788
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002789/**
2790 * mpol_to_str - format a mempolicy structure for printing
2791 * @buffer: to contain formatted mempolicy string
2792 * @maxlen: length of @buffer
2793 * @pol: pointer to mempolicy to be formatted
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002794 *
David Rientjes948927e2013-11-12 15:07:28 -08002795 * Convert @pol into a string. If @buffer is too short, truncate the string.
2796 * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the
2797 * longest flag, "relative", and to display at least a few node ids.
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002798 */
David Rientjes948927e2013-11-12 15:07:28 -08002799void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002800{
2801 char *p = buffer;
David Rientjes948927e2013-11-12 15:07:28 -08002802 nodemask_t nodes = NODE_MASK_NONE;
2803 unsigned short mode = MPOL_DEFAULT;
2804 unsigned short flags = 0;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002805
David Rientjes8790c71a2014-01-30 15:46:08 -08002806 if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) {
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002807 mode = pol->mode;
David Rientjes948927e2013-11-12 15:07:28 -08002808 flags = pol->flags;
2809 }
Lee Schermerhornbea904d2008-04-28 02:13:18 -07002810
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002811 switch (mode) {
2812 case MPOL_DEFAULT:
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002813 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002814 case MPOL_PREFERRED:
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002815 if (flags & MPOL_F_LOCAL)
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002816 mode = MPOL_LOCAL;
Lee Schermerhorn53f25562008-04-28 02:13:20 -07002817 else
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002818 node_set(pol->v.preferred_node, nodes);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002819 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002820 case MPOL_BIND:
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002821 case MPOL_INTERLEAVE:
Hugh Dickinsf2a07f42013-01-02 02:01:33 -08002822 nodes = pol->v.nodes;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002823 break;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002824 default:
David Rientjes948927e2013-11-12 15:07:28 -08002825 WARN_ON_ONCE(1);
2826 snprintf(p, maxlen, "unknown");
2827 return;
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002828 }
2829
David Rientjesb7a9f422013-11-21 14:32:06 -08002830 p += snprintf(p, maxlen, "%s", policy_modes[mode]);
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002831
Lee Schermerhornfc36b8d2008-04-28 02:13:21 -07002832 if (flags & MPOL_MODE_FLAGS) {
David Rientjes948927e2013-11-12 15:07:28 -08002833 p += snprintf(p, buffer + maxlen - p, "=");
David Rientjesf5b087b2008-04-28 02:12:27 -07002834
Lee Schermerhorn22919902008-04-28 02:13:22 -07002835 /*
2836 * Currently, the only defined flags are mutually exclusive
2837 */
David Rientjesf5b087b2008-04-28 02:12:27 -07002838 if (flags & MPOL_F_STATIC_NODES)
Lee Schermerhorn22919902008-04-28 02:13:22 -07002839 p += snprintf(p, buffer + maxlen - p, "static");
2840 else if (flags & MPOL_F_RELATIVE_NODES)
2841 p += snprintf(p, buffer + maxlen - p, "relative");
David Rientjesf5b087b2008-04-28 02:12:27 -07002842 }
2843
Tejun Heo9e763e02015-02-13 14:38:02 -08002844 if (!nodes_empty(nodes))
2845 p += scnprintf(p, buffer + maxlen - p, ":%*pbl",
2846 nodemask_pr_args(&nodes));
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08002847}