mempolicy.c 76.1 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0-only
L
Linus Torvalds 已提交
2 3 4 5
/*
 * Simple NUMA memory policy for the Linux kernel.
 *
 * Copyright 2003,2004 Andi Kleen, SuSE Labs.
6
 * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc.
L
Linus Torvalds 已提交
7 8 9 10 11 12 13 14 15 16 17 18 19 20
 *
 * NUMA policy allows the user to give hints in which node(s) memory should
 * be allocated.
 *
 * Support four policies per VMA and per process:
 *
 * The VMA policy has priority over the process policy for a page fault.
 *
 * interleave     Allocate memory interleaved over a set of nodes,
 *                with normal fallback if it fails.
 *                For VMA based allocations this interleaves based on the
 *                offset into the backing object or offset into the mapping
 *                for anonymous memory. For process policy an process counter
 *                is used.
21
 *
L
Linus Torvalds 已提交
22 23
 * bind           Only allocate memory on a specific set of nodes,
 *                no fallback.
24 25 26 27
 *                FIXME: memory is allocated starting with the first node
 *                to the last. It would be better if bind would truly restrict
 *                the allocation to memory nodes instead
 *
L
Linus Torvalds 已提交
28
 * preferred       Try a specific node first before normal fallback.
D
David Rientjes 已提交
29
 *                As a special case NUMA_NO_NODE here means do the allocation
L
Linus Torvalds 已提交
30 31 32
 *                on the local CPU. This is normally identical to default,
 *                but useful to set in a VMA when you have a non default
 *                process policy.
33
 *
L
Linus Torvalds 已提交
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67
 * default        Allocate on the local node first, or when on a VMA
 *                use the process policy. This is what Linux always did
 *		  in a NUMA aware kernel and still does by, ahem, default.
 *
 * The process policy is applied for most non interrupt memory allocations
 * in that process' context. Interrupts ignore the policies and always
 * try to allocate on the local CPU. The VMA policy is only applied for memory
 * allocations for a VMA in the VM.
 *
 * Currently there are a few corner cases in swapping where the policy
 * is not applied, but the majority should be handled. When process policy
 * is used it is not remembered over swap outs/swap ins.
 *
 * Only the highest zone in the zone hierarchy gets policied. Allocations
 * requesting a lower zone just use default policy. This implies that
 * on systems with highmem kernel lowmem allocation don't get policied.
 * Same with GFP_DMA allocations.
 *
 * For shmfs/tmpfs/hugetlbfs shared memory the policy is shared between
 * all users and remembered even when nobody has memory mapped.
 */

/* Notebook:
   fix mmap readahead to honour policy and enable policy for any page cache
   object
   statistics for bigpages
   global policy for page cache? currently it uses process policy. Requires
   first item above.
   handle mremap for shared memory (currently ignored for the policy)
   grows down?
   make bind policy root only? It can trigger oom much faster and the
   kernel is not always grateful with that.
*/

68 69
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

L
Linus Torvalds 已提交
70
#include <linux/mempolicy.h>
71
#include <linux/pagewalk.h>
L
Linus Torvalds 已提交
72 73 74 75
#include <linux/highmem.h>
#include <linux/hugetlb.h>
#include <linux/kernel.h>
#include <linux/sched.h>
76
#include <linux/sched/mm.h>
77
#include <linux/sched/numa_balancing.h>
78
#include <linux/sched/task.h>
L
Linus Torvalds 已提交
79 80 81 82
#include <linux/nodemask.h>
#include <linux/cpuset.h>
#include <linux/slab.h>
#include <linux/string.h>
83
#include <linux/export.h>
84
#include <linux/nsproxy.h>
L
Linus Torvalds 已提交
85 86 87
#include <linux/interrupt.h>
#include <linux/init.h>
#include <linux/compat.h>
88
#include <linux/ptrace.h>
89
#include <linux/swap.h>
90 91
#include <linux/seq_file.h>
#include <linux/proc_fs.h>
C
Christoph Lameter 已提交
92
#include <linux/migrate.h>
93
#include <linux/ksm.h>
94
#include <linux/rmap.h>
95
#include <linux/security.h>
A
Adrian Bunk 已提交
96
#include <linux/syscalls.h>
97
#include <linux/ctype.h>
98
#include <linux/mm_inline.h>
L
Lee Schermerhorn 已提交
99
#include <linux/mmu_notifier.h>
100
#include <linux/printk.h>
101
#include <linux/swapops.h>
102

L
Linus Torvalds 已提交
103
#include <asm/tlbflush.h>
104
#include <linux/uaccess.h>
L
Linus Torvalds 已提交
105

106 107
#include "internal.h"

108
/* Internal flags */
109
#define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0)	/* Skip checks for continuous vmas */
110
#define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1)		/* Invert check for nodemask */
111

112 113
static struct kmem_cache *policy_cache;
static struct kmem_cache *sn_cache;
L
Linus Torvalds 已提交
114 115 116

/* Highest zone. An specific allocation for a zone below that is not
   policied. */
117
enum zone_type policy_zone = 0;
L
Linus Torvalds 已提交
118

119 120 121
/*
 * run-time system-wide default policy => local allocation
 */
122
static struct mempolicy default_policy = {
L
Linus Torvalds 已提交
123
	.refcnt = ATOMIC_INIT(1), /* never free it */
124
	.mode = MPOL_PREFERRED,
125
	.flags = MPOL_F_LOCAL,
L
Linus Torvalds 已提交
126 127
};

M
Mel Gorman 已提交
128 129
static struct mempolicy preferred_node_policy[MAX_NUMNODES];

130 131
/**
 * numa_map_to_online_node - Find closest online node
132
 * @node: Node id to start the search
133 134 135 136 137
 *
 * Lookup the next closest node by distance if @nid is not online.
 */
int numa_map_to_online_node(int node)
{
138
	int min_dist = INT_MAX, dist, n, min_node;
139

140 141
	if (node == NUMA_NO_NODE || node_online(node))
		return node;
142 143

	min_node = node;
144 145 146 147 148
	for_each_online_node(n) {
		dist = node_distance(node, n);
		if (dist < min_dist) {
			min_dist = dist;
			min_node = n;
149 150 151 152 153 154 155
		}
	}

	return min_node;
}
EXPORT_SYMBOL_GPL(numa_map_to_online_node);

156
struct mempolicy *get_task_policy(struct task_struct *p)
M
Mel Gorman 已提交
157 158
{
	struct mempolicy *pol = p->mempolicy;
159
	int node;
M
Mel Gorman 已提交
160

161 162
	if (pol)
		return pol;
M
Mel Gorman 已提交
163

164 165 166 167 168 169
	node = numa_node_id();
	if (node != NUMA_NO_NODE) {
		pol = &preferred_node_policy[node];
		/* preferred_node_policy is not initialised early in boot */
		if (pol->mode)
			return pol;
M
Mel Gorman 已提交
170 171
	}

172
	return &default_policy;
M
Mel Gorman 已提交
173 174
}

175 176
static const struct mempolicy_operations {
	int (*create)(struct mempolicy *pol, const nodemask_t *nodes);
177
	void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes);
178 179
} mpol_ops[MPOL_MAX];

180 181
static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
{
B
Bob Liu 已提交
182
	return pol->flags & MPOL_MODE_FLAGS;
183 184 185 186 187 188 189 190
}

static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig,
				   const nodemask_t *rel)
{
	nodemask_t tmp;
	nodes_fold(tmp, *orig, nodes_weight(*rel));
	nodes_onto(*ret, tmp, *rel);
191 192
}

193 194 195 196 197 198 199 200 201 202 203
static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes)
{
	if (nodes_empty(*nodes))
		return -EINVAL;
	pol->v.nodes = *nodes;
	return 0;
}

static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes)
{
	if (!nodes)
204
		pol->flags |= MPOL_F_LOCAL;	/* local allocation */
205 206 207 208 209 210 211 212 213
	else if (nodes_empty(*nodes))
		return -EINVAL;			/*  no allowed nodes */
	else
		pol->v.preferred_node = first_node(*nodes);
	return 0;
}

static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes)
{
214
	if (nodes_empty(*nodes))
215 216 217 218 219
		return -EINVAL;
	pol->v.nodes = *nodes;
	return 0;
}

220 221 222 223 224 225 226
/*
 * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if
 * any, for the new policy.  mpol_new() has already validated the nodes
 * parameter with respect to the policy mode and flags.  But, we need to
 * handle an empty nodemask with MPOL_PREFERRED here.
 *
 * Must be called holding task's alloc_lock to protect task's mems_allowed
227
 * and mempolicy.  May also be called holding the mmap_lock for write.
228
 */
229 230
static int mpol_set_nodemask(struct mempolicy *pol,
		     const nodemask_t *nodes, struct nodemask_scratch *nsc)
231 232 233 234 235 236
{
	int ret;

	/* if mode is MPOL_DEFAULT, pol is NULL. This is right. */
	if (pol == NULL)
		return 0;
237
	/* Check N_MEMORY */
238
	nodes_and(nsc->mask1,
239
		  cpuset_current_mems_allowed, node_states[N_MEMORY]);
240 241 242 243 244 245

	VM_BUG_ON(!nodes);
	if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes))
		nodes = NULL;	/* explicit local allocation */
	else {
		if (pol->flags & MPOL_F_RELATIVE_NODES)
246
			mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1);
247
		else
248 249
			nodes_and(nsc->mask2, *nodes, nsc->mask1);

250 251 252 253 254 255 256
		if (mpol_store_user_nodemask(pol))
			pol->w.user_nodemask = *nodes;
		else
			pol->w.cpuset_mems_allowed =
						cpuset_current_mems_allowed;
	}

257 258 259 260
	if (nodes)
		ret = mpol_ops[pol->mode].create(pol, &nsc->mask2);
	else
		ret = mpol_ops[pol->mode].create(pol, NULL);
261 262 263 264 265 266 267
	return ret;
}

/*
 * This function just creates a new policy, does some check and simple
 * initialization. You must invoke mpol_set_nodemask() to set nodes.
 */
268 269
static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
				  nodemask_t *nodes)
L
Linus Torvalds 已提交
270 271 272
{
	struct mempolicy *policy;

273
	pr_debug("setting mode %d flags %d nodes[0] %lx\n",
D
David Rientjes 已提交
274
		 mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE);
275

276 277
	if (mode == MPOL_DEFAULT) {
		if (nodes && !nodes_empty(*nodes))
278
			return ERR_PTR(-EINVAL);
L
Lee Schermerhorn 已提交
279
		return NULL;
280
	}
281 282 283 284 285 286 287 288 289 290 291 292 293
	VM_BUG_ON(!nodes);

	/*
	 * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
	 * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
	 * All other modes require a valid pointer to a non-empty nodemask.
	 */
	if (mode == MPOL_PREFERRED) {
		if (nodes_empty(*nodes)) {
			if (((flags & MPOL_F_STATIC_NODES) ||
			     (flags & MPOL_F_RELATIVE_NODES)))
				return ERR_PTR(-EINVAL);
		}
294
	} else if (mode == MPOL_LOCAL) {
295 296 297
		if (!nodes_empty(*nodes) ||
		    (flags & MPOL_F_STATIC_NODES) ||
		    (flags & MPOL_F_RELATIVE_NODES))
298 299
			return ERR_PTR(-EINVAL);
		mode = MPOL_PREFERRED;
300 301
	} else if (nodes_empty(*nodes))
		return ERR_PTR(-EINVAL);
L
Linus Torvalds 已提交
302 303 304 305
	policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
	if (!policy)
		return ERR_PTR(-ENOMEM);
	atomic_set(&policy->refcnt, 1);
306
	policy->mode = mode;
307
	policy->flags = flags;
308

L
Linus Torvalds 已提交
309
	return policy;
310 311
}

312 313 314 315 316 317 318 319
/* Slow path of a mpol destructor. */
void __mpol_put(struct mempolicy *p)
{
	if (!atomic_dec_and_test(&p->refcnt))
		return;
	kmem_cache_free(policy_cache, p);
}

320
static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes)
321 322 323
{
}

324
static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes)
325 326 327 328 329 330 331 332
{
	nodemask_t tmp;

	if (pol->flags & MPOL_F_STATIC_NODES)
		nodes_and(tmp, pol->w.user_nodemask, *nodes);
	else if (pol->flags & MPOL_F_RELATIVE_NODES)
		mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
	else {
333 334
		nodes_remap(tmp, pol->v.nodes,pol->w.cpuset_mems_allowed,
								*nodes);
335
		pol->w.cpuset_mems_allowed = *nodes;
336
	}
337

338 339 340
	if (nodes_empty(tmp))
		tmp = *nodes;

341
	pol->v.nodes = tmp;
342 343 344
}

static void mpol_rebind_preferred(struct mempolicy *pol,
345
						const nodemask_t *nodes)
346 347 348 349 350 351
{
	nodemask_t tmp;

	if (pol->flags & MPOL_F_STATIC_NODES) {
		int node = first_node(pol->w.user_nodemask);

352
		if (node_isset(node, *nodes)) {
353
			pol->v.preferred_node = node;
354 355 356
			pol->flags &= ~MPOL_F_LOCAL;
		} else
			pol->flags |= MPOL_F_LOCAL;
357 358 359
	} else if (pol->flags & MPOL_F_RELATIVE_NODES) {
		mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
		pol->v.preferred_node = first_node(tmp);
360
	} else if (!(pol->flags & MPOL_F_LOCAL)) {
361 362 363 364 365
		pol->v.preferred_node = node_remap(pol->v.preferred_node,
						   pol->w.cpuset_mems_allowed,
						   *nodes);
		pol->w.cpuset_mems_allowed = *nodes;
	}
L
Linus Torvalds 已提交
366 367
}

368 369 370
/*
 * mpol_rebind_policy - Migrate a policy to a different set of nodes
 *
371
 * Per-vma policies are protected by mmap_lock. Allocations using per-task
372 373
 * policies are protected by task->mems_allowed_seq to prevent a premature
 * OOM/allocation failure due to parallel nodemask modification.
374
 */
375
static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask)
376 377 378
{
	if (!pol)
		return;
379
	if (!mpol_store_user_nodemask(pol) && !(pol->flags & MPOL_F_LOCAL) &&
380 381
	    nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
		return;
382

383
	mpol_ops[pol->mode].rebind(pol, newmask);
384 385 386 387 388
}

/*
 * Wrapper for mpol_rebind_policy() that just requires task
 * pointer, and updates task mempolicy.
389 390
 *
 * Called with task's alloc_lock held.
391 392
 */

393
void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new)
394
{
395
	mpol_rebind_policy(tsk->mempolicy, new);
396 397 398 399 400
}

/*
 * Rebind each vma in mm to new nodemask.
 *
401
 * Call holding a reference to mm.  Takes mm->mmap_lock during call.
402 403 404 405 406 407
 */

void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
{
	struct vm_area_struct *vma;

408
	mmap_write_lock(mm);
409
	for (vma = mm->mmap; vma; vma = vma->vm_next)
410
		mpol_rebind_policy(vma->vm_policy, new);
411
	mmap_write_unlock(mm);
412 413
}

414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431
static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
	[MPOL_DEFAULT] = {
		.rebind = mpol_rebind_default,
	},
	[MPOL_INTERLEAVE] = {
		.create = mpol_new_interleave,
		.rebind = mpol_rebind_nodemask,
	},
	[MPOL_PREFERRED] = {
		.create = mpol_new_preferred,
		.rebind = mpol_rebind_preferred,
	},
	[MPOL_BIND] = {
		.create = mpol_new_bind,
		.rebind = mpol_rebind_nodemask,
	},
};

432
static int migrate_page_add(struct page *page, struct list_head *pagelist,
433
				unsigned long flags);
434

435 436 437 438
struct queue_pages {
	struct list_head *pagelist;
	unsigned long flags;
	nodemask_t *nmask;
439 440 441
	unsigned long start;
	unsigned long end;
	struct vm_area_struct *first;
442 443
};

444 445 446 447 448 449 450 451 452 453 454 455 456 457 458
/*
 * Check if the page's nid is in qp->nmask.
 *
 * If MPOL_MF_INVERT is set in qp->flags, check if the nid is
 * in the invert of qp->nmask.
 */
static inline bool queue_pages_required(struct page *page,
					struct queue_pages *qp)
{
	int nid = page_to_nid(page);
	unsigned long flags = qp->flags;

	return node_isset(nid, *qp->nmask) == !(flags & MPOL_MF_INVERT);
}

459
/*
460 461 462 463 464 465 466 467
 * queue_pages_pmd() has four possible return values:
 * 0 - pages are placed on the right node or queued successfully.
 * 1 - there is unmovable page, and MPOL_MF_MOVE* & MPOL_MF_STRICT were
 *     specified.
 * 2 - THP was split.
 * -EIO - is migration entry or only MPOL_MF_STRICT was specified and an
 *        existing page was already on a node that does not follow the
 *        policy.
468
 */
469 470
static int queue_pages_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr,
				unsigned long end, struct mm_walk *walk)
471
	__releases(ptl)
472 473 474 475 476 477 478
{
	int ret = 0;
	struct page *page;
	struct queue_pages *qp = walk->private;
	unsigned long flags;

	if (unlikely(is_pmd_migration_entry(*pmd))) {
479
		ret = -EIO;
480 481 482 483 484 485
		goto unlock;
	}
	page = pmd_page(*pmd);
	if (is_huge_zero_page(page)) {
		spin_unlock(ptl);
		__split_huge_pmd(walk->vma, pmd, addr, false, NULL);
486
		ret = 2;
487 488
		goto out;
	}
489
	if (!queue_pages_required(page, qp))
490 491 492 493
		goto unlock;

	flags = qp->flags;
	/* go to thp migration */
494
	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
495 496
		if (!vma_migratable(walk->vma) ||
		    migrate_page_add(page, qp->pagelist, flags)) {
497
			ret = 1;
498 499 500 501
			goto unlock;
		}
	} else
		ret = -EIO;
502 503 504 505 506 507
unlock:
	spin_unlock(ptl);
out:
	return ret;
}

508 509 510
/*
 * Scan through pages checking if pages follow certain conditions,
 * and move them to the pagelist if they do.
511 512 513 514 515 516 517
 *
 * queue_pages_pte_range() has three possible return values:
 * 0 - pages are placed on the right node or queued successfully.
 * 1 - there is unmovable page, and MPOL_MF_MOVE* & MPOL_MF_STRICT were
 *     specified.
 * -EIO - only MPOL_MF_STRICT was specified and an existing page was already
 *        on a node that does not follow the policy.
518
 */
519 520
static int queue_pages_pte_range(pmd_t *pmd, unsigned long addr,
			unsigned long end, struct mm_walk *walk)
L
Linus Torvalds 已提交
521
{
522 523 524 525
	struct vm_area_struct *vma = walk->vma;
	struct page *page;
	struct queue_pages *qp = walk->private;
	unsigned long flags = qp->flags;
526
	int ret;
527
	bool has_unmovable = false;
528
	pte_t *pte, *mapped_pte;
529
	spinlock_t *ptl;
530

531 532 533
	ptl = pmd_trans_huge_lock(pmd, vma);
	if (ptl) {
		ret = queue_pages_pmd(pmd, ptl, addr, end, walk);
534
		if (ret != 2)
535
			return ret;
536
	}
537
	/* THP was split, fall through to pte walk */
538

539 540
	if (pmd_trans_unstable(pmd))
		return 0;
M
Michal Hocko 已提交
541

542
	mapped_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
543
	for (; addr != end; pte++, addr += PAGE_SIZE) {
544
		if (!pte_present(*pte))
L
Linus Torvalds 已提交
545
			continue;
546 547
		page = vm_normal_page(vma, addr, *pte);
		if (!page)
L
Linus Torvalds 已提交
548
			continue;
549
		/*
550 551
		 * vm_normal_page() filters out zero pages, but there might
		 * still be PageReserved pages to skip, perhaps in a VDSO.
552
		 */
H
Hugh Dickins 已提交
553
		if (PageReserved(page))
554
			continue;
555
		if (!queue_pages_required(page, qp))
556
			continue;
557
		if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
558 559 560
			/* MPOL_MF_STRICT must be specified if we get here */
			if (!vma_migratable(vma)) {
				has_unmovable = true;
561
				break;
562
			}
563 564 565 566 567 568 569 570

			/*
			 * Do not abort immediately since there may be
			 * temporary off LRU pages in the range.  Still
			 * need migrate other LRU pages.
			 */
			if (migrate_page_add(page, qp->pagelist, flags))
				has_unmovable = true;
571 572
		} else
			break;
573
	}
574
	pte_unmap_unlock(mapped_pte, ptl);
575
	cond_resched();
576 577 578 579

	if (has_unmovable)
		return 1;

580
	return addr != end ? -EIO : 0;
581 582
}

583 584 585
static int queue_pages_hugetlb(pte_t *pte, unsigned long hmask,
			       unsigned long addr, unsigned long end,
			       struct mm_walk *walk)
586
{
587
	int ret = 0;
588
#ifdef CONFIG_HUGETLB_PAGE
589
	struct queue_pages *qp = walk->private;
590
	unsigned long flags = (qp->flags & MPOL_MF_VALID);
591
	struct page *page;
592
	spinlock_t *ptl;
593
	pte_t entry;
594

595 596
	ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte);
	entry = huge_ptep_get(pte);
597 598 599
	if (!pte_present(entry))
		goto unlock;
	page = pte_page(entry);
600
	if (!queue_pages_required(page, qp))
601
		goto unlock;
602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622

	if (flags == MPOL_MF_STRICT) {
		/*
		 * STRICT alone means only detecting misplaced page and no
		 * need to further check other vma.
		 */
		ret = -EIO;
		goto unlock;
	}

	if (!vma_migratable(walk->vma)) {
		/*
		 * Must be STRICT with MOVE*, otherwise .test_walk() have
		 * stopped walking current vma.
		 * Detecting misplaced page but allow migrating pages which
		 * have been queued.
		 */
		ret = 1;
		goto unlock;
	}

623 624
	/* With MPOL_MF_MOVE, we migrate only unshared hugepage. */
	if (flags & (MPOL_MF_MOVE_ALL) ||
625 626 627 628 629 630 631 632 633
	    (flags & MPOL_MF_MOVE && page_mapcount(page) == 1)) {
		if (!isolate_huge_page(page, qp->pagelist) &&
			(flags & MPOL_MF_STRICT))
			/*
			 * Failed to isolate page but allow migrating pages
			 * which have been queued.
			 */
			ret = 1;
	}
634
unlock:
635
	spin_unlock(ptl);
636 637 638
#else
	BUG();
#endif
639
	return ret;
L
Linus Torvalds 已提交
640 641
}

642
#ifdef CONFIG_NUMA_BALANCING
L
Lee Schermerhorn 已提交
643
/*
644 645 646 647 648 649 650
 * This is used to mark a range of virtual addresses to be inaccessible.
 * These are later cleared by a NUMA hinting fault. Depending on these
 * faults, pages may be migrated for better NUMA placement.
 *
 * This is assuming that NUMA faults are handled using PROT_NONE. If
 * an architecture makes a different choice, it will need further
 * changes to the core.
L
Lee Schermerhorn 已提交
651
 */
652 653
unsigned long change_prot_numa(struct vm_area_struct *vma,
			unsigned long addr, unsigned long end)
L
Lee Schermerhorn 已提交
654
{
655
	int nr_updated;
L
Lee Schermerhorn 已提交
656

657
	nr_updated = change_protection(vma, addr, end, PAGE_NONE, MM_CP_PROT_NUMA);
658 659
	if (nr_updated)
		count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated);
L
Lee Schermerhorn 已提交
660

661
	return nr_updated;
L
Lee Schermerhorn 已提交
662 663 664 665 666 667 668
}
#else
static unsigned long change_prot_numa(struct vm_area_struct *vma,
			unsigned long addr, unsigned long end)
{
	return 0;
}
669
#endif /* CONFIG_NUMA_BALANCING */
L
Lee Schermerhorn 已提交
670

671 672 673 674 675 676 677 678
static int queue_pages_test_walk(unsigned long start, unsigned long end,
				struct mm_walk *walk)
{
	struct vm_area_struct *vma = walk->vma;
	struct queue_pages *qp = walk->private;
	unsigned long endvma = vma->vm_end;
	unsigned long flags = qp->flags;

679
	/* range check first */
680
	VM_BUG_ON_VMA((vma->vm_start > start) || (vma->vm_end < end), vma);
681 682 683 684 685 686

	if (!qp->first) {
		qp->first = vma;
		if (!(flags & MPOL_MF_DISCONTIG_OK) &&
			(qp->start < vma->vm_start))
			/* hole at head side of range */
687 688
			return -EFAULT;
	}
689 690 691 692 693
	if (!(flags & MPOL_MF_DISCONTIG_OK) &&
		((vma->vm_end < qp->end) &&
		(!vma->vm_next || vma->vm_end < vma->vm_next->vm_start)))
		/* hole at middle or tail of range */
		return -EFAULT;
694

695 696 697 698 699 700
	/*
	 * Need check MPOL_MF_STRICT to return -EIO if possible
	 * regardless of vma_migratable
	 */
	if (!vma_migratable(vma) &&
	    !(flags & MPOL_MF_STRICT))
701 702
		return 1;

703 704 705 706 707
	if (endvma > end)
		endvma = end;

	if (flags & MPOL_MF_LAZY) {
		/* Similar to task_numa_work, skip inaccessible VMAs */
708
		if (!is_vm_hugetlb_page(vma) && vma_is_accessible(vma) &&
709
			!(vma->vm_flags & VM_MIXEDMAP))
710 711 712 713
			change_prot_numa(vma, start, endvma);
		return 1;
	}

714
	/* queue pages from current vma */
715
	if (flags & MPOL_MF_VALID)
716 717 718 719
		return 0;
	return 1;
}

720 721 722 723 724 725
static const struct mm_walk_ops queue_pages_walk_ops = {
	.hugetlb_entry		= queue_pages_hugetlb,
	.pmd_entry		= queue_pages_pte_range,
	.test_walk		= queue_pages_test_walk,
};

726
/*
727 728 729 730
 * Walk through page tables and collect pages to be migrated.
 *
 * If pages found in a given range are on a set of nodes (determined by
 * @nodes and @flags,) it's isolated and queued to the pagelist which is
731 732 733 734 735 736
 * passed via @private.
 *
 * queue_pages_range() has three possible return values:
 * 1 - there is unmovable page, but MPOL_MF_MOVE* & MPOL_MF_STRICT were
 *     specified.
 * 0 - queue pages successfully or no misplaced page.
737 738 739
 * errno - i.e. misplaced pages with MPOL_MF_STRICT specified (-EIO) or
 *         memory range specified by nodemask and maxnode points outside
 *         your accessible address space (-EFAULT)
740
 */
741
static int
742
queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
743 744
		nodemask_t *nodes, unsigned long flags,
		struct list_head *pagelist)
L
Linus Torvalds 已提交
745
{
746
	int err;
747 748 749 750
	struct queue_pages qp = {
		.pagelist = pagelist,
		.flags = flags,
		.nmask = nodes,
751 752 753
		.start = start,
		.end = end,
		.first = NULL,
754 755
	};

756 757 758 759 760 761 762
	err = walk_page_range(mm, start, end, &queue_pages_walk_ops, &qp);

	if (!qp.first)
		/* whole range in hole */
		err = -EFAULT;

	return err;
L
Linus Torvalds 已提交
763 764
}

765 766
/*
 * Apply policy to a single VMA
767
 * This must be called with the mmap_lock held for writing.
768 769 770
 */
static int vma_replace_policy(struct vm_area_struct *vma,
						struct mempolicy *pol)
771
{
772 773 774
	int err;
	struct mempolicy *old;
	struct mempolicy *new;
775 776 777 778 779 780

	pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
		 vma->vm_start, vma->vm_end, vma->vm_pgoff,
		 vma->vm_ops, vma->vm_file,
		 vma->vm_ops ? vma->vm_ops->set_policy : NULL);

781 782 783 784 785
	new = mpol_dup(pol);
	if (IS_ERR(new))
		return PTR_ERR(new);

	if (vma->vm_ops && vma->vm_ops->set_policy) {
786
		err = vma->vm_ops->set_policy(vma, new);
787 788
		if (err)
			goto err_out;
789
	}
790 791

	old = vma->vm_policy;
792
	vma->vm_policy = new; /* protected by mmap_lock */
793 794 795 796 797
	mpol_put(old);

	return 0;
 err_out:
	mpol_put(new);
798 799 800
	return err;
}

L
Linus Torvalds 已提交
801
/* Step 2: apply policy to a range and do splits. */
802 803
static int mbind_range(struct mm_struct *mm, unsigned long start,
		       unsigned long end, struct mempolicy *new_pol)
L
Linus Torvalds 已提交
804 805
{
	struct vm_area_struct *next;
806 807 808
	struct vm_area_struct *prev;
	struct vm_area_struct *vma;
	int err = 0;
809
	pgoff_t pgoff;
810 811
	unsigned long vmstart;
	unsigned long vmend;
L
Linus Torvalds 已提交
812

813
	vma = find_vma(mm, start);
814
	VM_BUG_ON(!vma);
815

816
	prev = vma->vm_prev;
817 818 819
	if (start > vma->vm_start)
		prev = vma;

820
	for (; vma && vma->vm_start < end; prev = vma, vma = next) {
L
Linus Torvalds 已提交
821
		next = vma->vm_next;
822 823 824
		vmstart = max(start, vma->vm_start);
		vmend   = min(end, vma->vm_end);

825 826 827 828 829
		if (mpol_equal(vma_policy(vma), new_pol))
			continue;

		pgoff = vma->vm_pgoff +
			((vmstart - vma->vm_start) >> PAGE_SHIFT);
830
		prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
831 832
				 vma->anon_vma, vma->vm_file, pgoff,
				 new_pol, vma->vm_userfaultfd_ctx);
833 834 835
		if (prev) {
			vma = prev;
			next = vma->vm_next;
836 837 838 839
			if (mpol_equal(vma_policy(vma), new_pol))
				continue;
			/* vma_merge() joined vma && vma->next, case 8 */
			goto replace;
840 841 842 843 844 845 846 847 848 849 850
		}
		if (vma->vm_start != vmstart) {
			err = split_vma(vma->vm_mm, vma, vmstart, 1);
			if (err)
				goto out;
		}
		if (vma->vm_end != vmend) {
			err = split_vma(vma->vm_mm, vma, vmend, 0);
			if (err)
				goto out;
		}
851
 replace:
852
		err = vma_replace_policy(vma, new_pol);
853 854
		if (err)
			goto out;
L
Linus Torvalds 已提交
855
	}
856 857

 out:
L
Linus Torvalds 已提交
858 859 860 861
	return err;
}

/* Set the process memory policy */
862 863
static long do_set_mempolicy(unsigned short mode, unsigned short flags,
			     nodemask_t *nodes)
L
Linus Torvalds 已提交
864
{
865
	struct mempolicy *new, *old;
866
	NODEMASK_SCRATCH(scratch);
867
	int ret;
L
Linus Torvalds 已提交
868

869 870
	if (!scratch)
		return -ENOMEM;
871

872 873 874 875 876
	new = mpol_new(mode, flags, nodes);
	if (IS_ERR(new)) {
		ret = PTR_ERR(new);
		goto out;
	}
877

878
	ret = mpol_set_nodemask(new, nodes, scratch);
879 880
	if (ret) {
		mpol_put(new);
881
		goto out;
882
	}
883
	task_lock(current);
884
	old = current->mempolicy;
L
Linus Torvalds 已提交
885
	current->mempolicy = new;
886 887
	if (new && new->mode == MPOL_INTERLEAVE)
		current->il_prev = MAX_NUMNODES-1;
888 889
	task_unlock(current);
	mpol_put(old);
890 891 892 893
	ret = 0;
out:
	NODEMASK_SCRATCH_FREE(scratch);
	return ret;
L
Linus Torvalds 已提交
894 895
}

896 897
/*
 * Return nodemask for policy for get_mempolicy() query
898 899
 *
 * Called with task's alloc_lock held
900 901
 */
static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
L
Linus Torvalds 已提交
902
{
903
	nodes_clear(*nodes);
904 905 906
	if (p == &default_policy)
		return;

907
	switch (p->mode) {
908
	case MPOL_BIND:
L
Linus Torvalds 已提交
909
	case MPOL_INTERLEAVE:
910
		*nodes = p->v.nodes;
L
Linus Torvalds 已提交
911 912
		break;
	case MPOL_PREFERRED:
913
		if (!(p->flags & MPOL_F_LOCAL))
914
			node_set(p->v.preferred_node, *nodes);
915
		/* else return empty node mask for local allocation */
L
Linus Torvalds 已提交
916 917 918 919 920 921
		break;
	default:
		BUG();
	}
}

922
static int lookup_node(struct mm_struct *mm, unsigned long addr)
L
Linus Torvalds 已提交
923
{
924
	struct page *p = NULL;
L
Linus Torvalds 已提交
925 926
	int err;

927 928
	int locked = 1;
	err = get_user_pages_locked(addr & PAGE_MASK, 1, 0, &p, &locked);
929
	if (err > 0) {
L
Linus Torvalds 已提交
930 931 932
		err = page_to_nid(p);
		put_page(p);
	}
933
	if (locked)
934
		mmap_read_unlock(mm);
L
Linus Torvalds 已提交
935 936 937 938
	return err;
}

/* Retrieve NUMA policy */
A
Adrian Bunk 已提交
939 940
static long do_get_mempolicy(int *policy, nodemask_t *nmask,
			     unsigned long addr, unsigned long flags)
L
Linus Torvalds 已提交
941
{
942
	int err;
L
Linus Torvalds 已提交
943 944
	struct mm_struct *mm = current->mm;
	struct vm_area_struct *vma = NULL;
945
	struct mempolicy *pol = current->mempolicy, *pol_refcount = NULL;
L
Linus Torvalds 已提交
946

947 948
	if (flags &
		~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
L
Linus Torvalds 已提交
949
		return -EINVAL;
950 951 952 953 954

	if (flags & MPOL_F_MEMS_ALLOWED) {
		if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
			return -EINVAL;
		*policy = 0;	/* just so it's initialized */
955
		task_lock(current);
956
		*nmask  = cpuset_current_mems_allowed;
957
		task_unlock(current);
958 959 960
		return 0;
	}

L
Linus Torvalds 已提交
961
	if (flags & MPOL_F_ADDR) {
962 963 964 965 966
		/*
		 * Do NOT fall back to task policy if the
		 * vma/shared policy at addr is NULL.  We
		 * want to return MPOL_DEFAULT in this case.
		 */
967
		mmap_read_lock(mm);
L
Linus Torvalds 已提交
968 969
		vma = find_vma_intersection(mm, addr, addr+1);
		if (!vma) {
970
			mmap_read_unlock(mm);
L
Linus Torvalds 已提交
971 972 973 974 975 976 977 978 979 980
			return -EFAULT;
		}
		if (vma->vm_ops && vma->vm_ops->get_policy)
			pol = vma->vm_ops->get_policy(vma, addr);
		else
			pol = vma->vm_policy;
	} else if (addr)
		return -EINVAL;

	if (!pol)
981
		pol = &default_policy;	/* indicates default behavior */
L
Linus Torvalds 已提交
982 983 984

	if (flags & MPOL_F_NODE) {
		if (flags & MPOL_F_ADDR) {
985 986
			/*
			 * Take a refcount on the mpol, lookup_node()
987
			 * wil drop the mmap_lock, so after calling
988 989 990 991 992 993 994
			 * lookup_node() only "pol" remains valid, "vma"
			 * is stale.
			 */
			pol_refcount = pol;
			vma = NULL;
			mpol_get(pol);
			err = lookup_node(mm, addr);
L
Linus Torvalds 已提交
995 996
			if (err < 0)
				goto out;
997
			*policy = err;
L
Linus Torvalds 已提交
998
		} else if (pol == current->mempolicy &&
999
				pol->mode == MPOL_INTERLEAVE) {
1000
			*policy = next_node_in(current->il_prev, pol->v.nodes);
L
Linus Torvalds 已提交
1001 1002 1003 1004
		} else {
			err = -EINVAL;
			goto out;
		}
1005 1006 1007
	} else {
		*policy = pol == &default_policy ? MPOL_DEFAULT :
						pol->mode;
1008 1009 1010 1011 1012
		/*
		 * Internal mempolicy flags must be masked off before exposing
		 * the policy to userspace.
		 */
		*policy |= (pol->flags & MPOL_MODE_FLAGS);
1013
	}
L
Linus Torvalds 已提交
1014 1015

	err = 0;
1016
	if (nmask) {
1017 1018 1019 1020 1021 1022 1023
		if (mpol_store_user_nodemask(pol)) {
			*nmask = pol->w.user_nodemask;
		} else {
			task_lock(current);
			get_policy_nodemask(pol, nmask);
			task_unlock(current);
		}
1024
	}
L
Linus Torvalds 已提交
1025 1026

 out:
1027
	mpol_cond_put(pol);
L
Linus Torvalds 已提交
1028
	if (vma)
1029
		mmap_read_unlock(mm);
1030 1031
	if (pol_refcount)
		mpol_put(pol_refcount);
L
Linus Torvalds 已提交
1032 1033 1034
	return err;
}

C
Christoph Lameter 已提交
1035
#ifdef CONFIG_MIGRATION
1036
/*
1037
 * page migration, thp tail pages can be passed.
1038
 */
1039
static int migrate_page_add(struct page *page, struct list_head *pagelist,
1040
				unsigned long flags)
1041
{
1042
	struct page *head = compound_head(page);
1043
	/*
1044
	 * Avoid migrating a page that is shared with others.
1045
	 */
1046 1047 1048 1049
	if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(head) == 1) {
		if (!isolate_lru_page(head)) {
			list_add_tail(&head->lru, pagelist);
			mod_node_page_state(page_pgdat(head),
H
Huang Ying 已提交
1050
				NR_ISOLATED_ANON + page_is_file_lru(head),
1051
				thp_nr_pages(head));
1052 1053 1054 1055 1056 1057 1058 1059 1060
		} else if (flags & MPOL_MF_STRICT) {
			/*
			 * Non-movable page may reach here.  And, there may be
			 * temporary off LRU pages or non-LRU movable pages.
			 * Treat them as unmovable pages since they can't be
			 * isolated, so they can't be moved at the moment.  It
			 * should return -EIO for this case too.
			 */
			return -EIO;
1061 1062
		}
	}
1063 1064

	return 0;
1065
}
1066

1067 1068 1069 1070
/*
 * Migrate pages from one node to a target node.
 * Returns error or the number of pages not migrated.
 */
A
Adrian Bunk 已提交
1071 1072
static int migrate_to_node(struct mm_struct *mm, int source, int dest,
			   int flags)
1073 1074 1075 1076
{
	nodemask_t nmask;
	LIST_HEAD(pagelist);
	int err = 0;
1077 1078 1079 1080
	struct migration_target_control mtc = {
		.nid = dest,
		.gfp_mask = GFP_HIGHUSER_MOVABLE | __GFP_THISNODE,
	};
1081 1082 1083

	nodes_clear(nmask);
	node_set(source, nmask);
1084

1085 1086 1087 1088 1089 1090
	/*
	 * This does not "check" the range but isolates all pages that
	 * need migration.  Between passing in the full user address
	 * space range and MPOL_MF_DISCONTIG_OK, this call can not fail.
	 */
	VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)));
1091
	queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask,
1092 1093
			flags | MPOL_MF_DISCONTIG_OK, &pagelist);

1094
	if (!list_empty(&pagelist)) {
1095 1096
		err = migrate_pages(&pagelist, alloc_migration_target, NULL,
				(unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL);
1097
		if (err)
1098
			putback_movable_pages(&pagelist);
1099
	}
1100

1101
	return err;
1102 1103
}

1104
/*
1105 1106
 * Move pages between the two nodesets so as to preserve the physical
 * layout as much as possible.
1107 1108 1109
 *
 * Returns the number of page that could not be moved.
 */
1110 1111
int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
		     const nodemask_t *to, int flags)
1112
{
1113
	int busy = 0;
1114
	int err;
1115
	nodemask_t tmp;
1116

1117 1118 1119 1120
	err = migrate_prep();
	if (err)
		return err;

1121
	mmap_read_lock(mm);
1122

1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147
	/*
	 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
	 * bit in 'to' is not also set in 'tmp'.  Clear the found 'source'
	 * bit in 'tmp', and return that <source, dest> pair for migration.
	 * The pair of nodemasks 'to' and 'from' define the map.
	 *
	 * If no pair of bits is found that way, fallback to picking some
	 * pair of 'source' and 'dest' bits that are not the same.  If the
	 * 'source' and 'dest' bits are the same, this represents a node
	 * that will be migrating to itself, so no pages need move.
	 *
	 * If no bits are left in 'tmp', or if all remaining bits left
	 * in 'tmp' correspond to the same bit in 'to', return false
	 * (nothing left to migrate).
	 *
	 * This lets us pick a pair of nodes to migrate between, such that
	 * if possible the dest node is not already occupied by some other
	 * source node, minimizing the risk of overloading the memory on a
	 * node that would happen if we migrated incoming memory to a node
	 * before migrating outgoing memory source that same node.
	 *
	 * A single scan of tmp is sufficient.  As we go, we remember the
	 * most recent <s, d> pair that moved (s != d).  If we find a pair
	 * that not only moved, but what's better, moved to an empty slot
	 * (d is not set in tmp), then we break out then, with that pair.
1148
	 * Otherwise when we finish scanning from_tmp, we at least have the
1149 1150 1151 1152
	 * most recent <s, d> pair that moved.  If we get all the way through
	 * the scan of tmp without finding any node that moved, much less
	 * moved to an empty node, then there is nothing left worth migrating.
	 */
1153

1154
	tmp = *from;
1155 1156
	while (!nodes_empty(tmp)) {
		int s,d;
J
Jianguo Wu 已提交
1157
		int source = NUMA_NO_NODE;
1158 1159 1160
		int dest = 0;

		for_each_node_mask(s, tmp) {
1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176

			/*
			 * do_migrate_pages() tries to maintain the relative
			 * node relationship of the pages established between
			 * threads and memory areas.
                         *
			 * However if the number of source nodes is not equal to
			 * the number of destination nodes we can not preserve
			 * this node relative relationship.  In that case, skip
			 * copying memory from a node that is in the destination
			 * mask.
			 *
			 * Example: [2,3,4] -> [3,4,5] moves everything.
			 *          [0-7] - > [3,4,5] moves only 0,1,2,6,7.
			 */

1177 1178
			if ((nodes_weight(*from) != nodes_weight(*to)) &&
						(node_isset(s, *to)))
1179 1180
				continue;

1181
			d = node_remap(s, *from, *to);
1182 1183 1184 1185 1186 1187 1188 1189 1190 1191
			if (s == d)
				continue;

			source = s;	/* Node moved. Memorize */
			dest = d;

			/* dest not in remaining from nodes? */
			if (!node_isset(dest, tmp))
				break;
		}
J
Jianguo Wu 已提交
1192
		if (source == NUMA_NO_NODE)
1193 1194 1195 1196 1197 1198 1199 1200
			break;

		node_clear(source, tmp);
		err = migrate_to_node(mm, source, dest, flags);
		if (err > 0)
			busy += err;
		if (err < 0)
			break;
1201
	}
1202
	mmap_read_unlock(mm);
1203 1204 1205
	if (err < 0)
		return err;
	return busy;
C
Christoph Lameter 已提交
1206 1207 1208

}

1209 1210
/*
 * Allocate a new page for page migration based on vma policy.
1211
 * Start by assuming the page is mapped by the same vma as contains @start.
1212 1213 1214 1215
 * Search forward from there, if not.  N.B., this assumes that the
 * list of pages handed to migrate_pages()--which is how we get here--
 * is in virtual address order.
 */
1216
static struct page *new_page(struct page *page, unsigned long start)
1217
{
1218
	struct vm_area_struct *vma;
1219
	unsigned long address;
1220

1221
	vma = find_vma(current->mm, start);
1222 1223 1224 1225 1226 1227
	while (vma) {
		address = page_address_in_vma(page, vma);
		if (address != -EFAULT)
			break;
		vma = vma->vm_next;
	}
1228 1229

	if (PageHuge(page)) {
1230 1231
		return alloc_huge_page_vma(page_hstate(compound_head(page)),
				vma, address);
M
Michal Hocko 已提交
1232
	} else if (PageTransHuge(page)) {
1233 1234
		struct page *thp;

1235 1236
		thp = alloc_hugepage_vma(GFP_TRANSHUGE, vma, address,
					 HPAGE_PMD_ORDER);
1237 1238 1239 1240
		if (!thp)
			return NULL;
		prep_transhuge_page(thp);
		return thp;
1241
	}
1242
	/*
1243
	 * if !vma, alloc_page_vma() will use task or system default policy
1244
	 */
1245 1246
	return alloc_page_vma(GFP_HIGHUSER_MOVABLE | __GFP_RETRY_MAYFAIL,
			vma, address);
1247
}
C
Christoph Lameter 已提交
1248 1249
#else

1250
static int migrate_page_add(struct page *page, struct list_head *pagelist,
C
Christoph Lameter 已提交
1251 1252
				unsigned long flags)
{
1253
	return -EIO;
1254 1255
}

1256 1257
int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
		     const nodemask_t *to, int flags)
C
Christoph Lameter 已提交
1258 1259 1260
{
	return -ENOSYS;
}
1261

1262
static struct page *new_page(struct page *page, unsigned long start)
1263 1264 1265
{
	return NULL;
}
C
Christoph Lameter 已提交
1266 1267
#endif

A
Adrian Bunk 已提交
1268
static long do_mbind(unsigned long start, unsigned long len,
1269 1270
		     unsigned short mode, unsigned short mode_flags,
		     nodemask_t *nmask, unsigned long flags)
1271 1272 1273 1274 1275
{
	struct mm_struct *mm = current->mm;
	struct mempolicy *new;
	unsigned long end;
	int err;
1276
	int ret;
1277 1278
	LIST_HEAD(pagelist);

L
Lee Schermerhorn 已提交
1279
	if (flags & ~(unsigned long)MPOL_MF_VALID)
1280
		return -EINVAL;
1281
	if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297
		return -EPERM;

	if (start & ~PAGE_MASK)
		return -EINVAL;

	if (mode == MPOL_DEFAULT)
		flags &= ~MPOL_MF_STRICT;

	len = (len + PAGE_SIZE - 1) & PAGE_MASK;
	end = start + len;

	if (end < start)
		return -EINVAL;
	if (end == start)
		return 0;

1298
	new = mpol_new(mode, mode_flags, nmask);
1299 1300 1301
	if (IS_ERR(new))
		return PTR_ERR(new);

L
Lee Schermerhorn 已提交
1302 1303 1304
	if (flags & MPOL_MF_LAZY)
		new->flags |= MPOL_F_MOF;

1305 1306 1307 1308 1309 1310 1311
	/*
	 * If we are using the default policy then operation
	 * on discontinuous address spaces is okay after all
	 */
	if (!new)
		flags |= MPOL_MF_DISCONTIG_OK;

1312 1313
	pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
		 start, start + len, mode, mode_flags,
D
David Rientjes 已提交
1314
		 nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE);
1315

1316 1317 1318 1319
	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {

		err = migrate_prep();
		if (err)
K
KOSAKI Motohiro 已提交
1320
			goto mpol_out;
1321
	}
1322 1323 1324
	{
		NODEMASK_SCRATCH(scratch);
		if (scratch) {
1325
			mmap_write_lock(mm);
1326 1327
			err = mpol_set_nodemask(new, nmask, scratch);
			if (err)
1328
				mmap_write_unlock(mm);
1329 1330 1331 1332
		} else
			err = -ENOMEM;
		NODEMASK_SCRATCH_FREE(scratch);
	}
K
KOSAKI Motohiro 已提交
1333 1334 1335
	if (err)
		goto mpol_out;

1336
	ret = queue_pages_range(mm, start, end, nmask,
1337
			  flags | MPOL_MF_INVERT, &pagelist);
1338 1339

	if (ret < 0) {
1340
		err = ret;
1341 1342 1343 1344
		goto up_out;
	}

	err = mbind_range(mm, start, end, new);
1345

L
Lee Schermerhorn 已提交
1346 1347 1348
	if (!err) {
		int nr_failed = 0;

1349
		if (!list_empty(&pagelist)) {
L
Lee Schermerhorn 已提交
1350
			WARN_ON_ONCE(flags & MPOL_MF_LAZY);
1351 1352
			nr_failed = migrate_pages(&pagelist, new_page, NULL,
				start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND);
1353
			if (nr_failed)
1354
				putback_movable_pages(&pagelist);
1355
		}
1356

1357
		if ((ret > 0) || (nr_failed && (flags & MPOL_MF_STRICT)))
1358
			err = -EIO;
1359
	} else {
1360
up_out:
1361 1362 1363 1364
		if (!list_empty(&pagelist))
			putback_movable_pages(&pagelist);
	}

1365
	mmap_write_unlock(mm);
1366
mpol_out:
1367
	mpol_put(new);
1368 1369 1370
	return err;
}

1371 1372 1373 1374 1375
/*
 * User space interface with variable sized bitmaps for nodelists.
 */

/* Copy a node mask from user space. */
1376
static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
1377 1378 1379
		     unsigned long maxnode)
{
	unsigned long k;
1380
	unsigned long t;
1381 1382 1383 1384 1385 1386 1387
	unsigned long nlongs;
	unsigned long endmask;

	--maxnode;
	nodes_clear(*nodes);
	if (maxnode == 0 || !nmask)
		return 0;
A
Andi Kleen 已提交
1388
	if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
C
Chris Wright 已提交
1389
		return -EINVAL;
1390 1391 1392 1393 1394 1395 1396

	nlongs = BITS_TO_LONGS(maxnode);
	if ((maxnode % BITS_PER_LONG) == 0)
		endmask = ~0UL;
	else
		endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;

1397 1398 1399 1400 1401 1402 1403 1404 1405
	/*
	 * When the user specified more nodes than supported just check
	 * if the non supported part is all zero.
	 *
	 * If maxnode have more longs than MAX_NUMNODES, check
	 * the bits in that area first. And then go through to
	 * check the rest bits which equal or bigger than MAX_NUMNODES.
	 * Otherwise, just check bits [MAX_NUMNODES, maxnode).
	 */
1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419
	if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
		for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
			if (get_user(t, nmask + k))
				return -EFAULT;
			if (k == nlongs - 1) {
				if (t & endmask)
					return -EINVAL;
			} else if (t)
				return -EINVAL;
		}
		nlongs = BITS_TO_LONGS(MAX_NUMNODES);
		endmask = ~0UL;
	}

1420 1421 1422 1423 1424 1425 1426 1427 1428 1429
	if (maxnode > MAX_NUMNODES && MAX_NUMNODES % BITS_PER_LONG != 0) {
		unsigned long valid_mask = endmask;

		valid_mask &= ~((1UL << (MAX_NUMNODES % BITS_PER_LONG)) - 1);
		if (get_user(t, nmask + nlongs - 1))
			return -EFAULT;
		if (t & valid_mask)
			return -EINVAL;
	}

1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440
	if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
		return -EFAULT;
	nodes_addr(*nodes)[nlongs-1] &= endmask;
	return 0;
}

/* Copy a kernel node mask to user space */
static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
			      nodemask_t *nodes)
{
	unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1441
	unsigned int nbytes = BITS_TO_LONGS(nr_node_ids) * sizeof(long);
1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452

	if (copy > nbytes) {
		if (copy > PAGE_SIZE)
			return -EINVAL;
		if (clear_user((char __user *)mask + nbytes, copy - nbytes))
			return -EFAULT;
		copy = nbytes;
	}
	return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
}

1453 1454 1455
static long kernel_mbind(unsigned long start, unsigned long len,
			 unsigned long mode, const unsigned long __user *nmask,
			 unsigned long maxnode, unsigned int flags)
1456 1457 1458
{
	nodemask_t nodes;
	int err;
1459
	unsigned short mode_flags;
1460

1461
	start = untagged_addr(start);
1462 1463
	mode_flags = mode & MPOL_MODE_FLAGS;
	mode &= ~MPOL_MODE_FLAGS;
1464 1465
	if (mode >= MPOL_MAX)
		return -EINVAL;
1466 1467 1468
	if ((mode_flags & MPOL_F_STATIC_NODES) &&
	    (mode_flags & MPOL_F_RELATIVE_NODES))
		return -EINVAL;
1469 1470 1471
	err = get_nodes(&nodes, nmask, maxnode);
	if (err)
		return err;
1472
	return do_mbind(start, len, mode, mode_flags, &nodes, flags);
1473 1474
}

1475 1476 1477 1478 1479 1480 1481
SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
		unsigned long, mode, const unsigned long __user *, nmask,
		unsigned long, maxnode, unsigned int, flags)
{
	return kernel_mbind(start, len, mode, nmask, maxnode, flags);
}

1482
/* Set the process memory policy */
1483 1484
static long kernel_set_mempolicy(int mode, const unsigned long __user *nmask,
				 unsigned long maxnode)
1485 1486 1487
{
	int err;
	nodemask_t nodes;
1488
	unsigned short flags;
1489

1490 1491 1492
	flags = mode & MPOL_MODE_FLAGS;
	mode &= ~MPOL_MODE_FLAGS;
	if ((unsigned int)mode >= MPOL_MAX)
1493
		return -EINVAL;
1494 1495
	if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
		return -EINVAL;
1496 1497 1498
	err = get_nodes(&nodes, nmask, maxnode);
	if (err)
		return err;
1499
	return do_set_mempolicy(mode, flags, &nodes);
1500 1501
}

1502 1503 1504 1505 1506 1507
SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask,
		unsigned long, maxnode)
{
	return kernel_set_mempolicy(mode, nmask, maxnode);
}

1508 1509 1510
static int kernel_migrate_pages(pid_t pid, unsigned long maxnode,
				const unsigned long __user *old_nodes,
				const unsigned long __user *new_nodes)
1511
{
1512
	struct mm_struct *mm = NULL;
1513 1514 1515
	struct task_struct *task;
	nodemask_t task_nodes;
	int err;
1516 1517 1518 1519 1520 1521
	nodemask_t *old;
	nodemask_t *new;
	NODEMASK_SCRATCH(scratch);

	if (!scratch)
		return -ENOMEM;
1522

1523 1524 1525 1526
	old = &scratch->mask1;
	new = &scratch->mask2;

	err = get_nodes(old, old_nodes, maxnode);
1527
	if (err)
1528
		goto out;
1529

1530
	err = get_nodes(new, new_nodes, maxnode);
1531
	if (err)
1532
		goto out;
1533 1534

	/* Find the mm_struct */
1535
	rcu_read_lock();
1536
	task = pid ? find_task_by_vpid(pid) : current;
1537
	if (!task) {
1538
		rcu_read_unlock();
1539 1540
		err = -ESRCH;
		goto out;
1541
	}
1542
	get_task_struct(task);
1543

1544
	err = -EINVAL;
1545 1546

	/*
1547 1548
	 * Check if this process has the right to modify the specified process.
	 * Use the regular "ptrace_may_access()" checks.
1549
	 */
1550
	if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) {
1551
		rcu_read_unlock();
1552
		err = -EPERM;
1553
		goto out_put;
1554
	}
1555
	rcu_read_unlock();
1556 1557 1558

	task_nodes = cpuset_mems_allowed(task);
	/* Is the user allowed to access the target nodes? */
1559
	if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
1560
		err = -EPERM;
1561
		goto out_put;
1562 1563
	}

1564 1565 1566 1567 1568
	task_nodes = cpuset_mems_allowed(current);
	nodes_and(*new, *new, task_nodes);
	if (nodes_empty(*new))
		goto out_put;

1569 1570
	err = security_task_movememory(task);
	if (err)
1571
		goto out_put;
1572

1573 1574
	mm = get_task_mm(task);
	put_task_struct(task);
1575 1576

	if (!mm) {
1577
		err = -EINVAL;
1578 1579 1580 1581 1582
		goto out;
	}

	err = do_migrate_pages(mm, old, new,
		capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
1583 1584 1585

	mmput(mm);
out:
1586 1587
	NODEMASK_SCRATCH_FREE(scratch);

1588
	return err;
1589 1590 1591 1592 1593

out_put:
	put_task_struct(task);
	goto out;

1594 1595
}

1596 1597 1598 1599 1600 1601 1602
SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
		const unsigned long __user *, old_nodes,
		const unsigned long __user *, new_nodes)
{
	return kernel_migrate_pages(pid, maxnode, old_nodes, new_nodes);
}

1603

1604
/* Retrieve NUMA policy */
1605 1606 1607 1608 1609
static int kernel_get_mempolicy(int __user *policy,
				unsigned long __user *nmask,
				unsigned long maxnode,
				unsigned long addr,
				unsigned long flags)
1610
{
A
Adrian Bunk 已提交
1611
	int err;
1612
	int pval;
1613 1614
	nodemask_t nodes;

1615
	if (nmask != NULL && maxnode < nr_node_ids)
1616 1617
		return -EINVAL;

1618 1619
	addr = untagged_addr(addr);

1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633
	err = do_get_mempolicy(&pval, &nodes, addr, flags);

	if (err)
		return err;

	if (policy && put_user(pval, policy))
		return -EFAULT;

	if (nmask)
		err = copy_nodes_to_user(nmask, maxnode, &nodes);

	return err;
}

1634 1635 1636 1637 1638 1639 1640
SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
		unsigned long __user *, nmask, unsigned long, maxnode,
		unsigned long, addr, unsigned long, flags)
{
	return kernel_get_mempolicy(policy, nmask, maxnode, addr, flags);
}

L
Linus Torvalds 已提交
1641 1642
#ifdef CONFIG_COMPAT

1643 1644 1645 1646
COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
		       compat_ulong_t __user *, nmask,
		       compat_ulong_t, maxnode,
		       compat_ulong_t, addr, compat_ulong_t, flags)
L
Linus Torvalds 已提交
1647 1648 1649 1650 1651 1652
{
	long err;
	unsigned long __user *nm = NULL;
	unsigned long nr_bits, alloc_size;
	DECLARE_BITMAP(bm, MAX_NUMNODES);

1653
	nr_bits = min_t(unsigned long, maxnode-1, nr_node_ids);
L
Linus Torvalds 已提交
1654 1655 1656 1657 1658
	alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;

	if (nmask)
		nm = compat_alloc_user_space(alloc_size);

1659
	err = kernel_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
L
Linus Torvalds 已提交
1660 1661

	if (!err && nmask) {
1662 1663 1664
		unsigned long copy_size;
		copy_size = min_t(unsigned long, sizeof(bm), alloc_size);
		err = copy_from_user(bm, nm, copy_size);
L
Linus Torvalds 已提交
1665 1666 1667 1668 1669 1670 1671 1672
		/* ensure entire bitmap is zeroed */
		err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
		err |= compat_put_bitmap(nmask, bm, nr_bits);
	}

	return err;
}

1673 1674
COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask,
		       compat_ulong_t, maxnode)
L
Linus Torvalds 已提交
1675 1676 1677 1678 1679 1680 1681 1682 1683
{
	unsigned long __user *nm = NULL;
	unsigned long nr_bits, alloc_size;
	DECLARE_BITMAP(bm, MAX_NUMNODES);

	nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
	alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;

	if (nmask) {
1684 1685
		if (compat_get_bitmap(bm, nmask, nr_bits))
			return -EFAULT;
L
Linus Torvalds 已提交
1686
		nm = compat_alloc_user_space(alloc_size);
1687 1688
		if (copy_to_user(nm, bm, alloc_size))
			return -EFAULT;
L
Linus Torvalds 已提交
1689 1690
	}

1691
	return kernel_set_mempolicy(mode, nm, nr_bits+1);
L
Linus Torvalds 已提交
1692 1693
}

1694 1695 1696
COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len,
		       compat_ulong_t, mode, compat_ulong_t __user *, nmask,
		       compat_ulong_t, maxnode, compat_ulong_t, flags)
L
Linus Torvalds 已提交
1697 1698 1699
{
	unsigned long __user *nm = NULL;
	unsigned long nr_bits, alloc_size;
1700
	nodemask_t bm;
L
Linus Torvalds 已提交
1701 1702 1703 1704 1705

	nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
	alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;

	if (nmask) {
1706 1707
		if (compat_get_bitmap(nodes_addr(bm), nmask, nr_bits))
			return -EFAULT;
L
Linus Torvalds 已提交
1708
		nm = compat_alloc_user_space(alloc_size);
1709 1710
		if (copy_to_user(nm, nodes_addr(bm), alloc_size))
			return -EFAULT;
L
Linus Torvalds 已提交
1711 1712
	}

1713
	return kernel_mbind(start, len, mode, nm, nr_bits+1, flags);
L
Linus Torvalds 已提交
1714 1715
}

1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749
COMPAT_SYSCALL_DEFINE4(migrate_pages, compat_pid_t, pid,
		       compat_ulong_t, maxnode,
		       const compat_ulong_t __user *, old_nodes,
		       const compat_ulong_t __user *, new_nodes)
{
	unsigned long __user *old = NULL;
	unsigned long __user *new = NULL;
	nodemask_t tmp_mask;
	unsigned long nr_bits;
	unsigned long size;

	nr_bits = min_t(unsigned long, maxnode - 1, MAX_NUMNODES);
	size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
	if (old_nodes) {
		if (compat_get_bitmap(nodes_addr(tmp_mask), old_nodes, nr_bits))
			return -EFAULT;
		old = compat_alloc_user_space(new_nodes ? size * 2 : size);
		if (new_nodes)
			new = old + size / sizeof(unsigned long);
		if (copy_to_user(old, nodes_addr(tmp_mask), size))
			return -EFAULT;
	}
	if (new_nodes) {
		if (compat_get_bitmap(nodes_addr(tmp_mask), new_nodes, nr_bits))
			return -EFAULT;
		if (new == NULL)
			new = compat_alloc_user_space(size);
		if (copy_to_user(new, nodes_addr(tmp_mask), size))
			return -EFAULT;
	}
	return kernel_migrate_pages(pid, nr_bits + 1, old, new);
}

#endif /* CONFIG_COMPAT */
L
Linus Torvalds 已提交
1750

1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778
bool vma_migratable(struct vm_area_struct *vma)
{
	if (vma->vm_flags & (VM_IO | VM_PFNMAP))
		return false;

	/*
	 * DAX device mappings require predictable access latency, so avoid
	 * incurring periodic faults.
	 */
	if (vma_is_dax(vma))
		return false;

	if (is_vm_hugetlb_page(vma) &&
		!hugepage_migration_supported(hstate_vma(vma)))
		return false;

	/*
	 * Migration allocates pages in the highest zone. If we cannot
	 * do so then migration (at least from node to node) is not
	 * possible.
	 */
	if (vma->vm_file &&
		gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping))
			< policy_zone)
		return false;
	return true;
}

1779 1780
struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
						unsigned long addr)
L
Linus Torvalds 已提交
1781
{
1782
	struct mempolicy *pol = NULL;
L
Linus Torvalds 已提交
1783 1784

	if (vma) {
1785
		if (vma->vm_ops && vma->vm_ops->get_policy) {
1786
			pol = vma->vm_ops->get_policy(vma, addr);
1787
		} else if (vma->vm_policy) {
L
Linus Torvalds 已提交
1788
			pol = vma->vm_policy;
1789 1790 1791 1792 1793 1794 1795 1796 1797 1798

			/*
			 * shmem_alloc_page() passes MPOL_F_SHARED policy with
			 * a pseudo vma whose vma->vm_ops=NULL. Take a reference
			 * count on these policies which will be dropped by
			 * mpol_cond_put() later
			 */
			if (mpol_needs_cond_ref(pol))
				mpol_get(pol);
		}
L
Linus Torvalds 已提交
1799
	}
1800

1801 1802 1803 1804
	return pol;
}

/*
1805
 * get_vma_policy(@vma, @addr)
1806 1807 1808 1809
 * @vma: virtual memory area whose policy is sought
 * @addr: address in @vma for shared policy lookup
 *
 * Returns effective policy for a VMA at specified address.
1810
 * Falls back to current->mempolicy or system default policy, as necessary.
1811 1812 1813 1814 1815
 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
 * count--added by the get_policy() vm_op, as appropriate--to protect against
 * freeing by another task.  It is the caller's responsibility to free the
 * extra reference for shared policies.
 */
1816
static struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
1817
						unsigned long addr)
1818 1819 1820
{
	struct mempolicy *pol = __get_vma_policy(vma, addr);

1821
	if (!pol)
1822
		pol = get_task_policy(current);
1823

L
Linus Torvalds 已提交
1824 1825 1826
	return pol;
}

1827
bool vma_policy_mof(struct vm_area_struct *vma)
1828
{
1829
	struct mempolicy *pol;
1830

1831 1832
	if (vma->vm_ops && vma->vm_ops->get_policy) {
		bool ret = false;
1833

1834 1835 1836 1837
		pol = vma->vm_ops->get_policy(vma, vma->vm_start);
		if (pol && (pol->flags & MPOL_F_MOF))
			ret = true;
		mpol_cond_put(pol);
1838

1839
		return ret;
1840 1841
	}

1842
	pol = vma->vm_policy;
1843
	if (!pol)
1844
		pol = get_task_policy(current);
1845

1846 1847 1848
	return pol->flags & MPOL_F_MOF;
}

1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868
static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone)
{
	enum zone_type dynamic_policy_zone = policy_zone;

	BUG_ON(dynamic_policy_zone == ZONE_MOVABLE);

	/*
	 * if policy->v.nodes has movable memory only,
	 * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only.
	 *
	 * policy->v.nodes is intersect with node_states[N_MEMORY].
	 * so if the following test faile, it implies
	 * policy->v.nodes has movable memory only.
	 */
	if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY]))
		dynamic_policy_zone = ZONE_MOVABLE;

	return zone >= dynamic_policy_zone;
}

1869 1870 1871 1872
/*
 * Return a nodemask representing a mempolicy for filtering nodes for
 * page allocation
 */
1873
nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
1874 1875
{
	/* Lower zones don't get a nodemask applied for MPOL_BIND */
1876
	if (unlikely(policy->mode == MPOL_BIND) &&
1877
			apply_policy_zone(policy, gfp_zone(gfp)) &&
1878 1879 1880 1881 1882 1883
			cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
		return &policy->v.nodes;

	return NULL;
}

1884
/* Return the node id preferred by the given mempolicy, or the given id */
W
Wei Yang 已提交
1885
static int policy_node(gfp_t gfp, struct mempolicy *policy, int nd)
L
Linus Torvalds 已提交
1886
{
1887 1888 1889
	if (policy->mode == MPOL_PREFERRED && !(policy->flags & MPOL_F_LOCAL))
		nd = policy->v.preferred_node;
	else {
1890
		/*
1891 1892 1893
		 * __GFP_THISNODE shouldn't even be used with the bind policy
		 * because we might easily break the expectation to stay on the
		 * requested node and not break the policy.
1894
		 */
1895
		WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE));
L
Linus Torvalds 已提交
1896
	}
1897

1898
	return nd;
L
Linus Torvalds 已提交
1899 1900 1901 1902 1903
}

/* Do dynamic interleaving for a process */
static unsigned interleave_nodes(struct mempolicy *policy)
{
1904
	unsigned next;
L
Linus Torvalds 已提交
1905 1906
	struct task_struct *me = current;

1907
	next = next_node_in(me->il_prev, policy->v.nodes);
1908
	if (next < MAX_NUMNODES)
1909 1910
		me->il_prev = next;
	return next;
L
Linus Torvalds 已提交
1911 1912
}

1913 1914 1915 1916
/*
 * Depending on the memory policy provide a node from which to allocate the
 * next slab entry.
 */
1917
unsigned int mempolicy_slab_node(void)
1918
{
1919
	struct mempolicy *policy;
1920
	int node = numa_mem_id();
1921 1922

	if (in_interrupt())
1923
		return node;
1924 1925

	policy = current->mempolicy;
1926
	if (!policy || policy->flags & MPOL_F_LOCAL)
1927
		return node;
1928 1929 1930

	switch (policy->mode) {
	case MPOL_PREFERRED:
1931 1932 1933 1934
		/*
		 * handled MPOL_F_LOCAL above
		 */
		return policy->v.preferred_node;
1935

1936 1937 1938
	case MPOL_INTERLEAVE:
		return interleave_nodes(policy);

1939
	case MPOL_BIND: {
1940 1941
		struct zoneref *z;

1942 1943 1944 1945
		/*
		 * Follow bind policy behavior and start allocation at the
		 * first node.
		 */
1946 1947
		struct zonelist *zonelist;
		enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
1948
		zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK];
1949 1950
		z = first_zones_zonelist(zonelist, highest_zoneidx,
							&policy->v.nodes);
1951
		return z->zone ? zone_to_nid(z->zone) : node;
1952
	}
1953 1954

	default:
1955
		BUG();
1956 1957 1958
	}
}

1959 1960 1961 1962 1963
/*
 * Do static interleaving for a VMA with known offset @n.  Returns the n'th
 * node in pol->v.nodes (starting from n=0), wrapping around if n exceeds the
 * number of present nodes.
 */
1964
static unsigned offset_il_node(struct mempolicy *pol, unsigned long n)
L
Linus Torvalds 已提交
1965
{
1966
	unsigned nnodes = nodes_weight(pol->v.nodes);
1967
	unsigned target;
1968 1969
	int i;
	int nid;
L
Linus Torvalds 已提交
1970

1971 1972
	if (!nnodes)
		return numa_node_id();
1973 1974 1975
	target = (unsigned int)n % nnodes;
	nid = first_node(pol->v.nodes);
	for (i = 0; i < target; i++)
1976
		nid = next_node(nid, pol->v.nodes);
L
Linus Torvalds 已提交
1977 1978 1979
	return nid;
}

1980 1981 1982 1983 1984 1985 1986
/* Determine a node number for interleave */
static inline unsigned interleave_nid(struct mempolicy *pol,
		 struct vm_area_struct *vma, unsigned long addr, int shift)
{
	if (vma) {
		unsigned long off;

1987 1988 1989 1990 1991 1992 1993 1994 1995
		/*
		 * for small pages, there is no difference between
		 * shift and PAGE_SHIFT, so the bit-shift is safe.
		 * for huge pages, since vm_pgoff is in units of small
		 * pages, we need to shift off the always 0 bits to get
		 * a useful offset.
		 */
		BUG_ON(shift < PAGE_SHIFT);
		off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
1996
		off += (addr - vma->vm_start) >> shift;
1997
		return offset_il_node(pol, off);
1998 1999 2000 2001
	} else
		return interleave_nodes(pol);
}

2002
#ifdef CONFIG_HUGETLBFS
2003
/*
2004
 * huge_node(@vma, @addr, @gfp_flags, @mpol)
2005 2006 2007 2008 2009
 * @vma: virtual memory area whose policy is sought
 * @addr: address in @vma for shared policy lookup and interleave policy
 * @gfp_flags: for requested zone
 * @mpol: pointer to mempolicy pointer for reference counted mempolicy
 * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask
2010
 *
2011
 * Returns a nid suitable for a huge page allocation and a pointer
2012 2013 2014
 * to the struct mempolicy for conditional unref after allocation.
 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
 * @nodemask for filtering the zonelist.
2015
 *
2016
 * Must be protected by read_mems_allowed_begin()
2017
 */
2018 2019
int huge_node(struct vm_area_struct *vma, unsigned long addr, gfp_t gfp_flags,
				struct mempolicy **mpol, nodemask_t **nodemask)
2020
{
2021
	int nid;
2022

2023
	*mpol = get_vma_policy(vma, addr);
2024
	*nodemask = NULL;	/* assume !MPOL_BIND */
2025

2026
	if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
2027 2028
		nid = interleave_nid(*mpol, vma, addr,
					huge_page_shift(hstate_vma(vma)));
2029
	} else {
2030
		nid = policy_node(gfp_flags, *mpol, numa_node_id());
2031 2032
		if ((*mpol)->mode == MPOL_BIND)
			*nodemask = &(*mpol)->v.nodes;
2033
	}
2034
	return nid;
2035
}
2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060

/*
 * init_nodemask_of_mempolicy
 *
 * If the current task's mempolicy is "default" [NULL], return 'false'
 * to indicate default policy.  Otherwise, extract the policy nodemask
 * for 'bind' or 'interleave' policy into the argument nodemask, or
 * initialize the argument nodemask to contain the single node for
 * 'preferred' or 'local' policy and return 'true' to indicate presence
 * of non-default mempolicy.
 *
 * We don't bother with reference counting the mempolicy [mpol_get/put]
 * because the current task is examining it's own mempolicy and a task's
 * mempolicy is only ever changed by the task itself.
 *
 * N.B., it is the caller's responsibility to free a returned nodemask.
 */
bool init_nodemask_of_mempolicy(nodemask_t *mask)
{
	struct mempolicy *mempolicy;
	int nid;

	if (!(mask && current->mempolicy))
		return false;

2061
	task_lock(current);
2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079
	mempolicy = current->mempolicy;
	switch (mempolicy->mode) {
	case MPOL_PREFERRED:
		if (mempolicy->flags & MPOL_F_LOCAL)
			nid = numa_node_id();
		else
			nid = mempolicy->v.preferred_node;
		init_nodemask_of_node(mask, nid);
		break;

	case MPOL_BIND:
	case MPOL_INTERLEAVE:
		*mask =  mempolicy->v.nodes;
		break;

	default:
		BUG();
	}
2080
	task_unlock(current);
2081 2082 2083

	return true;
}
2084
#endif
2085

2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129
/*
 * mempolicy_nodemask_intersects
 *
 * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default
 * policy.  Otherwise, check for intersection between mask and the policy
 * nodemask for 'bind' or 'interleave' policy.  For 'perferred' or 'local'
 * policy, always return true since it may allocate elsewhere on fallback.
 *
 * Takes task_lock(tsk) to prevent freeing of its mempolicy.
 */
bool mempolicy_nodemask_intersects(struct task_struct *tsk,
					const nodemask_t *mask)
{
	struct mempolicy *mempolicy;
	bool ret = true;

	if (!mask)
		return ret;
	task_lock(tsk);
	mempolicy = tsk->mempolicy;
	if (!mempolicy)
		goto out;

	switch (mempolicy->mode) {
	case MPOL_PREFERRED:
		/*
		 * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to
		 * allocate from, they may fallback to other nodes when oom.
		 * Thus, it's possible for tsk to have allocated memory from
		 * nodes in mask.
		 */
		break;
	case MPOL_BIND:
	case MPOL_INTERLEAVE:
		ret = nodes_intersects(mempolicy->v.nodes, *mask);
		break;
	default:
		BUG();
	}
out:
	task_unlock(tsk);
	return ret;
}

L
Linus Torvalds 已提交
2130 2131
/* Allocate a page in interleaved policy.
   Own path because it needs to do special accounting. */
2132 2133
static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
					unsigned nid)
L
Linus Torvalds 已提交
2134 2135 2136
{
	struct page *page;

2137
	page = __alloc_pages(gfp, order, nid, NULL);
2138 2139 2140
	/* skip NUMA_INTERLEAVE_HIT counter update if numa stats is disabled */
	if (!static_branch_likely(&vm_numa_stat_key))
		return page;
2141 2142 2143 2144 2145
	if (page && page_to_nid(page) == nid) {
		preempt_disable();
		__inc_numa_state(page_zone(page), NUMA_INTERLEAVE_HIT);
		preempt_enable();
	}
L
Linus Torvalds 已提交
2146 2147 2148 2149
	return page;
}

/**
2150
 * 	alloc_pages_vma	- Allocate a page for a VMA.
L
Linus Torvalds 已提交
2151 2152 2153 2154 2155 2156 2157 2158
 *
 * 	@gfp:
 *      %GFP_USER    user allocation.
 *      %GFP_KERNEL  kernel allocations,
 *      %GFP_HIGHMEM highmem/user allocations,
 *      %GFP_FS      allocation should not call back into a file system.
 *      %GFP_ATOMIC  don't sleep.
 *
2159
 *	@order:Order of the GFP allocation.
L
Linus Torvalds 已提交
2160 2161
 * 	@vma:  Pointer to VMA or NULL if not available.
 *	@addr: Virtual Address of the allocation. Must be inside the VMA.
2162
 *	@node: Which node to prefer for allocation (modulo policy).
2163
 *	@hugepage: for hugepages try only the preferred node if possible
L
Linus Torvalds 已提交
2164 2165 2166
 *
 * 	This function allocates a page from the kernel page pool and applies
 *	a NUMA policy associated with the VMA or the current process.
2167
 *	When VMA is not NULL caller must read-lock the mmap_lock of the
L
Linus Torvalds 已提交
2168
 *	mm_struct of the VMA to prevent it from going away. Should be used for
2169 2170
 *	all allocations for pages that will be mapped into user space. Returns
 *	NULL when no page can be allocated.
L
Linus Torvalds 已提交
2171 2172
 */
struct page *
2173
alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
2174
		unsigned long addr, int node, bool hugepage)
L
Linus Torvalds 已提交
2175
{
2176
	struct mempolicy *pol;
2177
	struct page *page;
2178
	int preferred_nid;
2179
	nodemask_t *nmask;
2180

2181
	pol = get_vma_policy(vma, addr);
L
Linus Torvalds 已提交
2182

2183 2184 2185 2186 2187 2188 2189
	if (pol->mode == MPOL_INTERLEAVE) {
		unsigned nid;

		nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order);
		mpol_cond_put(pol);
		page = alloc_page_interleave(gfp, order, nid);
		goto out;
2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210
	}

	if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) {
		int hpage_node = node;

		/*
		 * For hugepage allocation and non-interleave policy which
		 * allows the current node (or other explicitly preferred
		 * node) we only try to allocate from the current/preferred
		 * node and don't fall back to other nodes, as the cost of
		 * remote accesses would likely offset THP benefits.
		 *
		 * If the policy is interleave, or does not allow the current
		 * node in its nodemask, we allocate the standard way.
		 */
		if (pol->mode == MPOL_PREFERRED && !(pol->flags & MPOL_F_LOCAL))
			hpage_node = pol->v.preferred_node;

		nmask = policy_nodemask(gfp, pol);
		if (!nmask || node_isset(hpage_node, *nmask)) {
			mpol_cond_put(pol);
2211 2212 2213 2214
			/*
			 * First, try to allocate THP only on local node, but
			 * don't reclaim unnecessarily, just compact.
			 */
2215
			page = __alloc_pages_node(hpage_node,
2216
				gfp | __GFP_THISNODE | __GFP_NORETRY, order);
2217 2218 2219 2220 2221

			/*
			 * If hugepage allocations are configured to always
			 * synchronous compact or the vma has been madvised
			 * to prefer hugepage backing, retry allowing remote
2222
			 * memory with both reclaim and compact as well.
2223 2224 2225
			 */
			if (!page && (gfp & __GFP_DIRECT_RECLAIM))
				page = __alloc_pages_node(hpage_node,
2226
								gfp, order);
2227

2228 2229
			goto out;
		}
2230 2231
	}

2232
	nmask = policy_nodemask(gfp, pol);
2233
	preferred_nid = policy_node(gfp, pol, node);
2234
	page = __alloc_pages(gfp, order, preferred_nid, nmask);
2235
	mpol_cond_put(pol);
2236
out:
2237
	return page;
L
Linus Torvalds 已提交
2238
}
C
Christoph Hellwig 已提交
2239
EXPORT_SYMBOL(alloc_pages_vma);
L
Linus Torvalds 已提交
2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255

/**
 * 	alloc_pages_current - Allocate pages.
 *
 *	@gfp:
 *		%GFP_USER   user allocation,
 *      	%GFP_KERNEL kernel allocation,
 *      	%GFP_HIGHMEM highmem allocation,
 *      	%GFP_FS     don't call back into a file system.
 *      	%GFP_ATOMIC don't sleep.
 *	@order: Power of two of allocation size in pages. 0 is a single page.
 *
 *	Allocate a page from the kernel page pool.  When not in
 *	interrupt context and apply the current process NUMA policy.
 *	Returns NULL when no page can be allocated.
 */
A
Al Viro 已提交
2256
struct page *alloc_pages_current(gfp_t gfp, unsigned order)
L
Linus Torvalds 已提交
2257
{
2258
	struct mempolicy *pol = &default_policy;
2259
	struct page *page;
L
Linus Torvalds 已提交
2260

2261 2262
	if (!in_interrupt() && !(gfp & __GFP_THISNODE))
		pol = get_task_policy(current);
2263 2264 2265 2266 2267

	/*
	 * No reference counting needed for current->mempolicy
	 * nor system default_policy
	 */
2268
	if (pol->mode == MPOL_INTERLEAVE)
2269 2270
		page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
	else
2271
		page = __alloc_pages(gfp, order,
2272
				policy_node(gfp, pol, numa_node_id()),
2273
				policy_nodemask(gfp, pol));
2274

2275
	return page;
L
Linus Torvalds 已提交
2276 2277 2278
}
EXPORT_SYMBOL(alloc_pages_current);

2279 2280 2281 2282 2283 2284 2285 2286 2287 2288
int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
{
	struct mempolicy *pol = mpol_dup(vma_policy(src));

	if (IS_ERR(pol))
		return PTR_ERR(pol);
	dst->vm_policy = pol;
	return 0;
}

2289
/*
2290
 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
2291 2292 2293 2294
 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
 * with the mems_allowed returned by cpuset_mems_allowed().  This
 * keeps mempolicies cpuset relative after its cpuset moves.  See
 * further kernel/cpuset.c update_nodemask().
2295 2296 2297
 *
 * current's mempolicy may be rebinded by the other task(the task that changes
 * cpuset's mems), so we needn't do rebind work for current task.
2298 2299
 */

2300 2301
/* Slow path of a mempolicy duplicate */
struct mempolicy *__mpol_dup(struct mempolicy *old)
L
Linus Torvalds 已提交
2302 2303 2304 2305 2306
{
	struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);

	if (!new)
		return ERR_PTR(-ENOMEM);
2307 2308 2309 2310 2311 2312 2313 2314 2315

	/* task's mempolicy is protected by alloc_lock */
	if (old == current->mempolicy) {
		task_lock(current);
		*new = *old;
		task_unlock(current);
	} else
		*new = *old;

2316 2317
	if (current_cpuset_is_being_rebound()) {
		nodemask_t mems = cpuset_mems_allowed(current);
2318
		mpol_rebind_policy(new, &mems);
2319
	}
L
Linus Torvalds 已提交
2320 2321 2322 2323 2324
	atomic_set(&new->refcnt, 1);
	return new;
}

/* Slow path of a mempolicy comparison */
2325
bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
L
Linus Torvalds 已提交
2326 2327
{
	if (!a || !b)
2328
		return false;
2329
	if (a->mode != b->mode)
2330
		return false;
B
Bob Liu 已提交
2331
	if (a->flags != b->flags)
2332
		return false;
B
Bob Liu 已提交
2333 2334
	if (mpol_store_user_nodemask(a))
		if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask))
2335
			return false;
B
Bob Liu 已提交
2336

2337
	switch (a->mode) {
2338
	case MPOL_BIND:
L
Linus Torvalds 已提交
2339
	case MPOL_INTERLEAVE:
2340
		return !!nodes_equal(a->v.nodes, b->v.nodes);
L
Linus Torvalds 已提交
2341
	case MPOL_PREFERRED:
2342 2343 2344
		/* a's ->flags is the same as b's */
		if (a->flags & MPOL_F_LOCAL)
			return true;
2345
		return a->v.preferred_node == b->v.preferred_node;
L
Linus Torvalds 已提交
2346 2347
	default:
		BUG();
2348
		return false;
L
Linus Torvalds 已提交
2349 2350 2351 2352 2353 2354 2355 2356
	}
}

/*
 * Shared memory backing store policy support.
 *
 * Remember policies even when nobody has shared memory mapped.
 * The policies are kept in Red-Black tree linked from the inode.
2357
 * They are protected by the sp->lock rwlock, which should be held
L
Linus Torvalds 已提交
2358 2359 2360
 * for any accesses to the tree.
 */

2361 2362 2363 2364
/*
 * lookup first element intersecting start-end.  Caller holds sp->lock for
 * reading or for writing
 */
L
Linus Torvalds 已提交
2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394
static struct sp_node *
sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
{
	struct rb_node *n = sp->root.rb_node;

	while (n) {
		struct sp_node *p = rb_entry(n, struct sp_node, nd);

		if (start >= p->end)
			n = n->rb_right;
		else if (end <= p->start)
			n = n->rb_left;
		else
			break;
	}
	if (!n)
		return NULL;
	for (;;) {
		struct sp_node *w = NULL;
		struct rb_node *prev = rb_prev(n);
		if (!prev)
			break;
		w = rb_entry(prev, struct sp_node, nd);
		if (w->end <= start)
			break;
		n = prev;
	}
	return rb_entry(n, struct sp_node, nd);
}

2395 2396 2397 2398
/*
 * Insert a new shared policy into the list.  Caller holds sp->lock for
 * writing.
 */
L
Linus Torvalds 已提交
2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416
static void sp_insert(struct shared_policy *sp, struct sp_node *new)
{
	struct rb_node **p = &sp->root.rb_node;
	struct rb_node *parent = NULL;
	struct sp_node *nd;

	while (*p) {
		parent = *p;
		nd = rb_entry(parent, struct sp_node, nd);
		if (new->start < nd->start)
			p = &(*p)->rb_left;
		else if (new->end > nd->end)
			p = &(*p)->rb_right;
		else
			BUG();
	}
	rb_link_node(&new->nd, parent, p);
	rb_insert_color(&new->nd, &sp->root);
2417
	pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
2418
		 new->policy ? new->policy->mode : 0);
L
Linus Torvalds 已提交
2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429
}

/* Find shared policy intersecting idx */
struct mempolicy *
mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
{
	struct mempolicy *pol = NULL;
	struct sp_node *sn;

	if (!sp->root.rb_node)
		return NULL;
2430
	read_lock(&sp->lock);
L
Linus Torvalds 已提交
2431 2432 2433 2434 2435
	sn = sp_lookup(sp, idx, idx+1);
	if (sn) {
		mpol_get(sn->policy);
		pol = sn->policy;
	}
2436
	read_unlock(&sp->lock);
L
Linus Torvalds 已提交
2437 2438 2439
	return pol;
}

2440 2441 2442 2443 2444 2445
static void sp_free(struct sp_node *n)
{
	mpol_put(n->policy);
	kmem_cache_free(sn_cache, n);
}

2446 2447 2448
/**
 * mpol_misplaced - check whether current page node is valid in policy
 *
2449 2450 2451
 * @page: page to be checked
 * @vma: vm area where page mapped
 * @addr: virtual address where page mapped
2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465
 *
 * Lookup current policy node id for vma,addr and "compare to" page's
 * node id.
 *
 * Returns:
 *	-1	- not misplaced, page is in the right node
 *	node	- node id where the page should be
 *
 * Policy determination "mimics" alloc_page_vma().
 * Called from fault path where we know the vma and faulting address.
 */
int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr)
{
	struct mempolicy *pol;
2466
	struct zoneref *z;
2467 2468
	int curnid = page_to_nid(page);
	unsigned long pgoff;
2469 2470
	int thiscpu = raw_smp_processor_id();
	int thisnid = cpu_to_node(thiscpu);
2471
	int polnid = NUMA_NO_NODE;
2472 2473
	int ret = -1;

2474
	pol = get_vma_policy(vma, addr);
2475 2476 2477 2478 2479 2480 2481
	if (!(pol->flags & MPOL_F_MOF))
		goto out;

	switch (pol->mode) {
	case MPOL_INTERLEAVE:
		pgoff = vma->vm_pgoff;
		pgoff += (addr - vma->vm_start) >> PAGE_SHIFT;
2482
		polnid = offset_il_node(pol, pgoff);
2483 2484 2485 2486 2487 2488 2489 2490 2491 2492
		break;

	case MPOL_PREFERRED:
		if (pol->flags & MPOL_F_LOCAL)
			polnid = numa_node_id();
		else
			polnid = pol->v.preferred_node;
		break;

	case MPOL_BIND:
2493

2494 2495 2496 2497 2498 2499 2500 2501
		/*
		 * allows binding to multiple nodes.
		 * use current page if in policy nodemask,
		 * else select nearest allowed node, if any.
		 * If no allowed nodes, use current [!misplaced].
		 */
		if (node_isset(curnid, pol->v.nodes))
			goto out;
2502
		z = first_zones_zonelist(
2503 2504
				node_zonelist(numa_node_id(), GFP_HIGHUSER),
				gfp_zone(GFP_HIGHUSER),
2505
				&pol->v.nodes);
2506
		polnid = zone_to_nid(z->zone);
2507 2508 2509 2510 2511
		break;

	default:
		BUG();
	}
M
Mel Gorman 已提交
2512 2513

	/* Migrate the page towards the node whose CPU is referencing it */
2514
	if (pol->flags & MPOL_F_MORON) {
2515
		polnid = thisnid;
M
Mel Gorman 已提交
2516

2517
		if (!should_numa_migrate_memory(current, page, curnid, thiscpu))
2518
			goto out;
2519 2520
	}

2521 2522 2523 2524 2525 2526 2527 2528
	if (curnid != polnid)
		ret = polnid;
out:
	mpol_cond_put(pol);

	return ret;
}

2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545
/*
 * Drop the (possibly final) reference to task->mempolicy.  It needs to be
 * dropped after task->mempolicy is set to NULL so that any allocation done as
 * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed
 * policy.
 */
void mpol_put_task_policy(struct task_struct *task)
{
	struct mempolicy *pol;

	task_lock(task);
	pol = task->mempolicy;
	task->mempolicy = NULL;
	task_unlock(task);
	mpol_put(pol);
}

L
Linus Torvalds 已提交
2546 2547
static void sp_delete(struct shared_policy *sp, struct sp_node *n)
{
2548
	pr_debug("deleting %lx-l%lx\n", n->start, n->end);
L
Linus Torvalds 已提交
2549
	rb_erase(&n->nd, &sp->root);
2550
	sp_free(n);
L
Linus Torvalds 已提交
2551 2552
}

2553 2554 2555 2556 2557 2558 2559 2560
static void sp_node_init(struct sp_node *node, unsigned long start,
			unsigned long end, struct mempolicy *pol)
{
	node->start = start;
	node->end = end;
	node->policy = pol;
}

A
Adrian Bunk 已提交
2561 2562
static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
				struct mempolicy *pol)
L
Linus Torvalds 已提交
2563
{
2564 2565
	struct sp_node *n;
	struct mempolicy *newpol;
L
Linus Torvalds 已提交
2566

2567
	n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
L
Linus Torvalds 已提交
2568 2569
	if (!n)
		return NULL;
2570 2571 2572 2573 2574 2575 2576

	newpol = mpol_dup(pol);
	if (IS_ERR(newpol)) {
		kmem_cache_free(sn_cache, n);
		return NULL;
	}
	newpol->flags |= MPOL_F_SHARED;
2577
	sp_node_init(n, start, end, newpol);
2578

L
Linus Torvalds 已提交
2579 2580 2581 2582 2583 2584 2585
	return n;
}

/* Replace a policy range. */
static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
				 unsigned long end, struct sp_node *new)
{
2586
	struct sp_node *n;
2587 2588
	struct sp_node *n_new = NULL;
	struct mempolicy *mpol_new = NULL;
2589
	int ret = 0;
L
Linus Torvalds 已提交
2590

2591
restart:
2592
	write_lock(&sp->lock);
L
Linus Torvalds 已提交
2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604
	n = sp_lookup(sp, start, end);
	/* Take care of old policies in the same range. */
	while (n && n->start < end) {
		struct rb_node *next = rb_next(&n->nd);
		if (n->start >= start) {
			if (n->end <= end)
				sp_delete(sp, n);
			else
				n->start = end;
		} else {
			/* Old policy spanning whole new range. */
			if (n->end > end) {
2605 2606 2607 2608 2609
				if (!n_new)
					goto alloc_new;

				*mpol_new = *n->policy;
				atomic_set(&mpol_new->refcnt, 1);
2610
				sp_node_init(n_new, end, n->end, mpol_new);
L
Linus Torvalds 已提交
2611
				n->end = start;
2612
				sp_insert(sp, n_new);
2613 2614
				n_new = NULL;
				mpol_new = NULL;
L
Linus Torvalds 已提交
2615 2616 2617 2618 2619 2620 2621 2622 2623 2624
				break;
			} else
				n->end = start;
		}
		if (!next)
			break;
		n = rb_entry(next, struct sp_node, nd);
	}
	if (new)
		sp_insert(sp, new);
2625
	write_unlock(&sp->lock);
2626 2627 2628 2629 2630 2631 2632 2633
	ret = 0;

err_out:
	if (mpol_new)
		mpol_put(mpol_new);
	if (n_new)
		kmem_cache_free(sn_cache, n_new);

2634
	return ret;
2635 2636

alloc_new:
2637
	write_unlock(&sp->lock);
2638 2639 2640 2641 2642 2643 2644 2645
	ret = -ENOMEM;
	n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL);
	if (!n_new)
		goto err_out;
	mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
	if (!mpol_new)
		goto err_out;
	goto restart;
L
Linus Torvalds 已提交
2646 2647
}

2648 2649 2650 2651 2652 2653 2654 2655
/**
 * mpol_shared_policy_init - initialize shared policy for inode
 * @sp: pointer to inode shared policy
 * @mpol:  struct mempolicy to install
 *
 * Install non-NULL @mpol in inode's shared policy rb-tree.
 * On entry, the current task has a reference on a non-NULL @mpol.
 * This must be released on exit.
2656
 * This is called at get_inode() calls and we can use GFP_KERNEL.
2657 2658 2659
 */
void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
{
2660 2661
	int ret;

2662
	sp->root = RB_ROOT;		/* empty tree == default mempolicy */
2663
	rwlock_init(&sp->lock);
2664 2665 2666 2667

	if (mpol) {
		struct vm_area_struct pvma;
		struct mempolicy *new;
2668
		NODEMASK_SCRATCH(scratch);
2669

2670
		if (!scratch)
2671
			goto put_mpol;
2672 2673
		/* contextualize the tmpfs mount point mempolicy */
		new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
2674
		if (IS_ERR(new))
2675
			goto free_scratch; /* no valid nodemask intersection */
2676 2677

		task_lock(current);
2678
		ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
2679
		task_unlock(current);
2680
		if (ret)
2681
			goto put_new;
2682 2683

		/* Create pseudo-vma that contains just the policy */
2684
		vma_init(&pvma, NULL);
2685 2686
		pvma.vm_end = TASK_SIZE;	/* policy covers entire file */
		mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
2687

2688
put_new:
2689
		mpol_put(new);			/* drop initial ref */
2690
free_scratch:
2691
		NODEMASK_SCRATCH_FREE(scratch);
2692 2693
put_mpol:
		mpol_put(mpol);	/* drop our incoming ref on sb mpol */
2694 2695 2696
	}
}

L
Linus Torvalds 已提交
2697 2698 2699 2700 2701 2702 2703
int mpol_set_shared_policy(struct shared_policy *info,
			struct vm_area_struct *vma, struct mempolicy *npol)
{
	int err;
	struct sp_node *new = NULL;
	unsigned long sz = vma_pages(vma);

2704
	pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
L
Linus Torvalds 已提交
2705
		 vma->vm_pgoff,
2706
		 sz, npol ? npol->mode : -1,
2707
		 npol ? npol->flags : -1,
D
David Rientjes 已提交
2708
		 npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE);
L
Linus Torvalds 已提交
2709 2710 2711 2712 2713 2714 2715 2716

	if (npol) {
		new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
		if (!new)
			return -ENOMEM;
	}
	err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
	if (err && new)
2717
		sp_free(new);
L
Linus Torvalds 已提交
2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728
	return err;
}

/* Free a backing policy store on inode delete. */
void mpol_free_shared_policy(struct shared_policy *p)
{
	struct sp_node *n;
	struct rb_node *next;

	if (!p->root.rb_node)
		return;
2729
	write_lock(&p->lock);
L
Linus Torvalds 已提交
2730 2731 2732 2733
	next = rb_first(&p->root);
	while (next) {
		n = rb_entry(next, struct sp_node, nd);
		next = rb_next(&n->nd);
2734
		sp_delete(p, n);
L
Linus Torvalds 已提交
2735
	}
2736
	write_unlock(&p->lock);
L
Linus Torvalds 已提交
2737 2738
}

2739
#ifdef CONFIG_NUMA_BALANCING
2740
static int __initdata numabalancing_override;
2741 2742 2743 2744 2745 2746 2747 2748

static void __init check_numabalancing_enable(void)
{
	bool numabalancing_default = false;

	if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED))
		numabalancing_default = true;

2749 2750 2751 2752
	/* Parsed by setup_numabalancing. override == 1 enables, -1 disables */
	if (numabalancing_override)
		set_numabalancing_state(numabalancing_override == 1);

2753
	if (num_online_nodes() > 1 && !numabalancing_override) {
J
Joe Perches 已提交
2754
		pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n",
2755
			numabalancing_default ? "Enabling" : "Disabling");
2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766
		set_numabalancing_state(numabalancing_default);
	}
}

static int __init setup_numabalancing(char *str)
{
	int ret = 0;
	if (!str)
		goto out;

	if (!strcmp(str, "enable")) {
2767
		numabalancing_override = 1;
2768 2769
		ret = 1;
	} else if (!strcmp(str, "disable")) {
2770
		numabalancing_override = -1;
2771 2772 2773 2774
		ret = 1;
	}
out:
	if (!ret)
2775
		pr_warn("Unable to parse numa_balancing=\n");
2776 2777 2778 2779 2780 2781 2782 2783 2784 2785

	return ret;
}
__setup("numa_balancing=", setup_numabalancing);
#else
static inline void __init check_numabalancing_enable(void)
{
}
#endif /* CONFIG_NUMA_BALANCING */

L
Linus Torvalds 已提交
2786 2787 2788
/* assumes fs == KERNEL_DS */
void __init numa_policy_init(void)
{
2789 2790 2791 2792
	nodemask_t interleave_nodes;
	unsigned long largest = 0;
	int nid, prefer = 0;

L
Linus Torvalds 已提交
2793 2794
	policy_cache = kmem_cache_create("numa_policy",
					 sizeof(struct mempolicy),
2795
					 0, SLAB_PANIC, NULL);
L
Linus Torvalds 已提交
2796 2797 2798

	sn_cache = kmem_cache_create("shared_policy_node",
				     sizeof(struct sp_node),
2799
				     0, SLAB_PANIC, NULL);
L
Linus Torvalds 已提交
2800

M
Mel Gorman 已提交
2801 2802 2803 2804 2805 2806 2807 2808 2809
	for_each_node(nid) {
		preferred_node_policy[nid] = (struct mempolicy) {
			.refcnt = ATOMIC_INIT(1),
			.mode = MPOL_PREFERRED,
			.flags = MPOL_F_MOF | MPOL_F_MORON,
			.v = { .preferred_node = nid, },
		};
	}

2810 2811 2812 2813 2814 2815
	/*
	 * Set interleaving policy for system init. Interleaving is only
	 * enabled across suitably sized nodes (default is >= 16MB), or
	 * fall back to the largest node if they're all smaller.
	 */
	nodes_clear(interleave_nodes);
2816
	for_each_node_state(nid, N_MEMORY) {
2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832
		unsigned long total_pages = node_present_pages(nid);

		/* Preserve the largest node */
		if (largest < total_pages) {
			largest = total_pages;
			prefer = nid;
		}

		/* Interleave this node? */
		if ((total_pages << PAGE_SHIFT) >= (16 << 20))
			node_set(nid, interleave_nodes);
	}

	/* All too small, use the largest */
	if (unlikely(nodes_empty(interleave_nodes)))
		node_set(prefer, interleave_nodes);
L
Linus Torvalds 已提交
2833

2834
	if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
2835
		pr_err("%s: interleaving failed\n", __func__);
2836 2837

	check_numabalancing_enable();
L
Linus Torvalds 已提交
2838 2839
}

2840
/* Reset policy of current process to default */
L
Linus Torvalds 已提交
2841 2842
void numa_default_policy(void)
{
2843
	do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
L
Linus Torvalds 已提交
2844
}
2845

2846 2847 2848 2849
/*
 * Parse and format mempolicy from/to strings
 */

2850
/*
2851
 * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag.
2852
 */
2853 2854 2855 2856 2857 2858
static const char * const policy_modes[] =
{
	[MPOL_DEFAULT]    = "default",
	[MPOL_PREFERRED]  = "prefer",
	[MPOL_BIND]       = "bind",
	[MPOL_INTERLEAVE] = "interleave",
L
Lee Schermerhorn 已提交
2859
	[MPOL_LOCAL]      = "local",
2860
};
2861

2862 2863 2864

#ifdef CONFIG_TMPFS
/**
2865
 * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option.
2866
 * @str:  string containing mempolicy to parse
2867
 * @mpol:  pointer to struct mempolicy pointer, returned on success.
2868 2869 2870 2871
 *
 * Format of input:
 *	<mode>[=<flags>][:<nodelist>]
 *
2872
 * On success, returns 0, else 1
2873
 */
2874
int mpol_parse_str(char *str, struct mempolicy **mpol)
2875
{
2876
	struct mempolicy *new = NULL;
2877
	unsigned short mode_flags;
2878
	nodemask_t nodes;
2879 2880
	char *nodelist = strchr(str, ':');
	char *flags = strchr(str, '=');
2881
	int err = 1, mode;
2882

2883 2884 2885
	if (flags)
		*flags++ = '\0';	/* terminate mode string */

2886 2887 2888
	if (nodelist) {
		/* NUL-terminate mode or flags string */
		*nodelist++ = '\0';
2889
		if (nodelist_parse(nodelist, nodes))
2890
			goto out;
2891
		if (!nodes_subset(nodes, node_states[N_MEMORY]))
2892
			goto out;
2893 2894 2895
	} else
		nodes_clear(nodes);

2896 2897
	mode = match_string(policy_modes, MPOL_MAX, str);
	if (mode < 0)
2898 2899
		goto out;

2900
	switch (mode) {
2901
	case MPOL_PREFERRED:
2902
		/*
2903 2904 2905
		 * Insist on a nodelist of one node only, although later
		 * we use first_node(nodes) to grab a single node, so here
		 * nodelist (or nodes) cannot be empty.
2906
		 */
2907 2908 2909 2910
		if (nodelist) {
			char *rest = nodelist;
			while (isdigit(*rest))
				rest++;
2911 2912
			if (*rest)
				goto out;
2913 2914
			if (nodes_empty(nodes))
				goto out;
2915 2916 2917 2918 2919 2920 2921
		}
		break;
	case MPOL_INTERLEAVE:
		/*
		 * Default to online nodes with memory if no nodelist
		 */
		if (!nodelist)
2922
			nodes = node_states[N_MEMORY];
2923
		break;
2924
	case MPOL_LOCAL:
2925
		/*
2926
		 * Don't allow a nodelist;  mpol_new() checks flags
2927
		 */
2928
		if (nodelist)
2929
			goto out;
2930
		mode = MPOL_PREFERRED;
2931
		break;
2932 2933 2934 2935 2936 2937 2938
	case MPOL_DEFAULT:
		/*
		 * Insist on a empty nodelist
		 */
		if (!nodelist)
			err = 0;
		goto out;
2939 2940 2941 2942 2943 2944
	case MPOL_BIND:
		/*
		 * Insist on a nodelist
		 */
		if (!nodelist)
			goto out;
2945 2946
	}

2947
	mode_flags = 0;
2948 2949 2950 2951 2952 2953
	if (flags) {
		/*
		 * Currently, we only support two mutually exclusive
		 * mode flags.
		 */
		if (!strcmp(flags, "static"))
2954
			mode_flags |= MPOL_F_STATIC_NODES;
2955
		else if (!strcmp(flags, "relative"))
2956
			mode_flags |= MPOL_F_RELATIVE_NODES;
2957
		else
2958
			goto out;
2959
	}
2960 2961 2962

	new = mpol_new(mode, mode_flags, &nodes);
	if (IS_ERR(new))
2963 2964
		goto out;

2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981
	/*
	 * Save nodes for mpol_to_str() to show the tmpfs mount options
	 * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo.
	 */
	if (mode != MPOL_PREFERRED)
		new->v.nodes = nodes;
	else if (nodelist)
		new->v.preferred_node = first_node(nodes);
	else
		new->flags |= MPOL_F_LOCAL;

	/*
	 * Save nodes for contextualization: this will be used to "clone"
	 * the mempolicy in a specific context [cpuset] at a later time.
	 */
	new->w.user_nodemask = nodes;

2982
	err = 0;
2983

2984 2985 2986 2987 2988 2989
out:
	/* Restore string for error message */
	if (nodelist)
		*--nodelist = ':';
	if (flags)
		*--flags = '=';
2990 2991
	if (!err)
		*mpol = new;
2992 2993 2994 2995
	return err;
}
#endif /* CONFIG_TMPFS */

2996 2997 2998 2999 3000 3001
/**
 * mpol_to_str - format a mempolicy structure for printing
 * @buffer:  to contain formatted mempolicy string
 * @maxlen:  length of @buffer
 * @pol:  pointer to mempolicy to be formatted
 *
3002 3003 3004
 * Convert @pol into a string.  If @buffer is too short, truncate the string.
 * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the
 * longest flag, "relative", and to display at least a few node ids.
3005
 */
3006
void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
3007 3008
{
	char *p = buffer;
3009 3010 3011
	nodemask_t nodes = NODE_MASK_NONE;
	unsigned short mode = MPOL_DEFAULT;
	unsigned short flags = 0;
3012

3013
	if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) {
3014
		mode = pol->mode;
3015 3016
		flags = pol->flags;
	}
3017

3018 3019 3020 3021
	switch (mode) {
	case MPOL_DEFAULT:
		break;
	case MPOL_PREFERRED:
3022
		if (flags & MPOL_F_LOCAL)
3023
			mode = MPOL_LOCAL;
3024
		else
3025
			node_set(pol->v.preferred_node, nodes);
3026 3027 3028
		break;
	case MPOL_BIND:
	case MPOL_INTERLEAVE:
3029
		nodes = pol->v.nodes;
3030 3031
		break;
	default:
3032 3033 3034
		WARN_ON_ONCE(1);
		snprintf(p, maxlen, "unknown");
		return;
3035 3036
	}

3037
	p += snprintf(p, maxlen, "%s", policy_modes[mode]);
3038

3039
	if (flags & MPOL_MODE_FLAGS) {
3040
		p += snprintf(p, buffer + maxlen - p, "=");
3041

3042 3043 3044
		/*
		 * Currently, the only defined flags are mutually exclusive
		 */
3045
		if (flags & MPOL_F_STATIC_NODES)
3046 3047 3048
			p += snprintf(p, buffer + maxlen - p, "static");
		else if (flags & MPOL_F_RELATIVE_NODES)
			p += snprintf(p, buffer + maxlen - p, "relative");
3049 3050
	}

3051 3052 3053
	if (!nodes_empty(nodes))
		p += scnprintf(p, buffer + maxlen - p, ":%*pbl",
			       nodemask_pr_args(&nodes));
3054
}