]> bbs.cooldavid.org Git - net-next-2.6.git/blame - mm/mempolicy.c
tmpfs: fix oops on mounts with mpol=default
[net-next-2.6.git] / mm / mempolicy.c
CommitLineData
1da177e4
LT
1/*
2 * Simple NUMA memory policy for the Linux kernel.
3 *
4 * Copyright 2003,2004 Andi Kleen, SuSE Labs.
8bccd85f 5 * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc.
1da177e4
LT
6 * Subject to the GNU Public License, version 2.
7 *
8 * NUMA policy allows the user to give hints in which node(s) memory should
9 * be allocated.
10 *
11 * Support four policies per VMA and per process:
12 *
13 * The VMA policy has priority over the process policy for a page fault.
14 *
15 * interleave Allocate memory interleaved over a set of nodes,
16 * with normal fallback if it fails.
17 * For VMA based allocations this interleaves based on the
18 * offset into the backing object or offset into the mapping
19 * for anonymous memory. For process policy an process counter
20 * is used.
8bccd85f 21 *
1da177e4
LT
22 * bind Only allocate memory on a specific set of nodes,
23 * no fallback.
8bccd85f
CL
24 * FIXME: memory is allocated starting with the first node
25 * to the last. It would be better if bind would truly restrict
26 * the allocation to memory nodes instead
27 *
1da177e4
LT
28 * preferred Try a specific node first before normal fallback.
29 * As a special case node -1 here means do the allocation
30 * on the local CPU. This is normally identical to default,
31 * but useful to set in a VMA when you have a non default
32 * process policy.
8bccd85f 33 *
1da177e4
LT
34 * default Allocate on the local node first, or when on a VMA
35 * use the process policy. This is what Linux always did
36 * in a NUMA aware kernel and still does by, ahem, default.
37 *
38 * The process policy is applied for most non interrupt memory allocations
39 * in that process' context. Interrupts ignore the policies and always
40 * try to allocate on the local CPU. The VMA policy is only applied for memory
41 * allocations for a VMA in the VM.
42 *
43 * Currently there are a few corner cases in swapping where the policy
44 * is not applied, but the majority should be handled. When process policy
45 * is used it is not remembered over swap outs/swap ins.
46 *
47 * Only the highest zone in the zone hierarchy gets policied. Allocations
48 * requesting a lower zone just use default policy. This implies that
49 * on systems with highmem kernel lowmem allocation don't get policied.
50 * Same with GFP_DMA allocations.
51 *
52 * For shmfs/tmpfs/hugetlbfs shared memory the policy is shared between
53 * all users and remembered even when nobody has memory mapped.
54 */
55
56/* Notebook:
57 fix mmap readahead to honour policy and enable policy for any page cache
58 object
59 statistics for bigpages
60 global policy for page cache? currently it uses process policy. Requires
61 first item above.
62 handle mremap for shared memory (currently ignored for the policy)
63 grows down?
64 make bind policy root only? It can trigger oom much faster and the
65 kernel is not always grateful with that.
1da177e4
LT
66*/
67
68#include <linux/mempolicy.h>
69#include <linux/mm.h>
70#include <linux/highmem.h>
71#include <linux/hugetlb.h>
72#include <linux/kernel.h>
73#include <linux/sched.h>
1da177e4
LT
74#include <linux/nodemask.h>
75#include <linux/cpuset.h>
76#include <linux/gfp.h>
77#include <linux/slab.h>
78#include <linux/string.h>
79#include <linux/module.h>
b488893a 80#include <linux/nsproxy.h>
1da177e4
LT
81#include <linux/interrupt.h>
82#include <linux/init.h>
83#include <linux/compat.h>
dc9aa5b9 84#include <linux/swap.h>
1a75a6c8
CL
85#include <linux/seq_file.h>
86#include <linux/proc_fs.h>
b20a3503 87#include <linux/migrate.h>
62b61f61 88#include <linux/ksm.h>
95a402c3 89#include <linux/rmap.h>
86c3a764 90#include <linux/security.h>
dbcb0f19 91#include <linux/syscalls.h>
095f1fc4 92#include <linux/ctype.h>
6d9c285a 93#include <linux/mm_inline.h>
dc9aa5b9 94
1da177e4
LT
95#include <asm/tlbflush.h>
96#include <asm/uaccess.h>
97
62695a84
NP
98#include "internal.h"
99
38e35860 100/* Internal flags */
dc9aa5b9 101#define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */
38e35860 102#define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */
1a75a6c8 103#define MPOL_MF_STATS (MPOL_MF_INTERNAL << 2) /* Gather statistics */
dc9aa5b9 104
fcc234f8
PE
105static struct kmem_cache *policy_cache;
106static struct kmem_cache *sn_cache;
1da177e4 107
1da177e4
LT
108/* Highest zone. An specific allocation for a zone below that is not
109 policied. */
6267276f 110enum zone_type policy_zone = 0;
1da177e4 111
bea904d5
LS
112/*
113 * run-time system-wide default policy => local allocation
114 */
d42c6997 115struct mempolicy default_policy = {
1da177e4 116 .refcnt = ATOMIC_INIT(1), /* never free it */
bea904d5 117 .mode = MPOL_PREFERRED,
fc36b8d3 118 .flags = MPOL_F_LOCAL,
1da177e4
LT
119};
120
37012946
DR
121static const struct mempolicy_operations {
122 int (*create)(struct mempolicy *pol, const nodemask_t *nodes);
123 void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes);
124} mpol_ops[MPOL_MAX];
125
19770b32 126/* Check that the nodemask contains at least one populated zone */
37012946 127static int is_valid_nodemask(const nodemask_t *nodemask)
1da177e4 128{
19770b32 129 int nd, k;
1da177e4 130
19770b32
MG
131 /* Check that there is something useful in this mask */
132 k = policy_zone;
133
134 for_each_node_mask(nd, *nodemask) {
135 struct zone *z;
136
137 for (k = 0; k <= policy_zone; k++) {
138 z = &NODE_DATA(nd)->node_zones[k];
139 if (z->present_pages > 0)
140 return 1;
dd942ae3 141 }
8af5e2eb 142 }
19770b32
MG
143
144 return 0;
1da177e4
LT
145}
146
f5b087b5
DR
147static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
148{
4c50bc01
DR
149 return pol->flags & (MPOL_F_STATIC_NODES | MPOL_F_RELATIVE_NODES);
150}
151
152static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig,
153 const nodemask_t *rel)
154{
155 nodemask_t tmp;
156 nodes_fold(tmp, *orig, nodes_weight(*rel));
157 nodes_onto(*ret, tmp, *rel);
f5b087b5
DR
158}
159
37012946
DR
160static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes)
161{
162 if (nodes_empty(*nodes))
163 return -EINVAL;
164 pol->v.nodes = *nodes;
165 return 0;
166}
167
168static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes)
169{
170 if (!nodes)
fc36b8d3 171 pol->flags |= MPOL_F_LOCAL; /* local allocation */
37012946
DR
172 else if (nodes_empty(*nodes))
173 return -EINVAL; /* no allowed nodes */
174 else
175 pol->v.preferred_node = first_node(*nodes);
176 return 0;
177}
178
179static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes)
180{
181 if (!is_valid_nodemask(nodes))
182 return -EINVAL;
183 pol->v.nodes = *nodes;
184 return 0;
185}
186
58568d2a
MX
187/*
188 * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if
189 * any, for the new policy. mpol_new() has already validated the nodes
190 * parameter with respect to the policy mode and flags. But, we need to
191 * handle an empty nodemask with MPOL_PREFERRED here.
192 *
193 * Must be called holding task's alloc_lock to protect task's mems_allowed
194 * and mempolicy. May also be called holding the mmap_semaphore for write.
195 */
4bfc4495
KH
196static int mpol_set_nodemask(struct mempolicy *pol,
197 const nodemask_t *nodes, struct nodemask_scratch *nsc)
58568d2a 198{
58568d2a
MX
199 int ret;
200
201 /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */
202 if (pol == NULL)
203 return 0;
4bfc4495
KH
204 /* Check N_HIGH_MEMORY */
205 nodes_and(nsc->mask1,
206 cpuset_current_mems_allowed, node_states[N_HIGH_MEMORY]);
58568d2a
MX
207
208 VM_BUG_ON(!nodes);
209 if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes))
210 nodes = NULL; /* explicit local allocation */
211 else {
212 if (pol->flags & MPOL_F_RELATIVE_NODES)
4bfc4495 213 mpol_relative_nodemask(&nsc->mask2, nodes,&nsc->mask1);
58568d2a 214 else
4bfc4495
KH
215 nodes_and(nsc->mask2, *nodes, nsc->mask1);
216
58568d2a
MX
217 if (mpol_store_user_nodemask(pol))
218 pol->w.user_nodemask = *nodes;
219 else
220 pol->w.cpuset_mems_allowed =
221 cpuset_current_mems_allowed;
222 }
223
4bfc4495
KH
224 if (nodes)
225 ret = mpol_ops[pol->mode].create(pol, &nsc->mask2);
226 else
227 ret = mpol_ops[pol->mode].create(pol, NULL);
58568d2a
MX
228 return ret;
229}
230
231/*
232 * This function just creates a new policy, does some check and simple
233 * initialization. You must invoke mpol_set_nodemask() to set nodes.
234 */
028fec41
DR
235static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
236 nodemask_t *nodes)
1da177e4
LT
237{
238 struct mempolicy *policy;
239
028fec41
DR
240 pr_debug("setting mode %d flags %d nodes[0] %lx\n",
241 mode, flags, nodes ? nodes_addr(*nodes)[0] : -1);
140d5a49 242
3e1f0645
DR
243 if (mode == MPOL_DEFAULT) {
244 if (nodes && !nodes_empty(*nodes))
37012946 245 return ERR_PTR(-EINVAL);
bea904d5 246 return NULL; /* simply delete any existing policy */
37012946 247 }
3e1f0645
DR
248 VM_BUG_ON(!nodes);
249
250 /*
251 * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
252 * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
253 * All other modes require a valid pointer to a non-empty nodemask.
254 */
255 if (mode == MPOL_PREFERRED) {
256 if (nodes_empty(*nodes)) {
257 if (((flags & MPOL_F_STATIC_NODES) ||
258 (flags & MPOL_F_RELATIVE_NODES)))
259 return ERR_PTR(-EINVAL);
3e1f0645
DR
260 }
261 } else if (nodes_empty(*nodes))
262 return ERR_PTR(-EINVAL);
1da177e4
LT
263 policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
264 if (!policy)
265 return ERR_PTR(-ENOMEM);
266 atomic_set(&policy->refcnt, 1);
45c4745a 267 policy->mode = mode;
3e1f0645 268 policy->flags = flags;
37012946 269
1da177e4 270 return policy;
37012946
DR
271}
272
52cd3b07
LS
273/* Slow path of a mpol destructor. */
274void __mpol_put(struct mempolicy *p)
275{
276 if (!atomic_dec_and_test(&p->refcnt))
277 return;
52cd3b07
LS
278 kmem_cache_free(policy_cache, p);
279}
280
37012946
DR
281static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes)
282{
283}
284
285static void mpol_rebind_nodemask(struct mempolicy *pol,
286 const nodemask_t *nodes)
287{
288 nodemask_t tmp;
289
290 if (pol->flags & MPOL_F_STATIC_NODES)
291 nodes_and(tmp, pol->w.user_nodemask, *nodes);
292 else if (pol->flags & MPOL_F_RELATIVE_NODES)
293 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
294 else {
295 nodes_remap(tmp, pol->v.nodes, pol->w.cpuset_mems_allowed,
296 *nodes);
297 pol->w.cpuset_mems_allowed = *nodes;
298 }
f5b087b5 299
37012946
DR
300 pol->v.nodes = tmp;
301 if (!node_isset(current->il_next, tmp)) {
302 current->il_next = next_node(current->il_next, tmp);
303 if (current->il_next >= MAX_NUMNODES)
304 current->il_next = first_node(tmp);
305 if (current->il_next >= MAX_NUMNODES)
306 current->il_next = numa_node_id();
307 }
308}
309
310static void mpol_rebind_preferred(struct mempolicy *pol,
311 const nodemask_t *nodes)
312{
313 nodemask_t tmp;
314
37012946
DR
315 if (pol->flags & MPOL_F_STATIC_NODES) {
316 int node = first_node(pol->w.user_nodemask);
317
fc36b8d3 318 if (node_isset(node, *nodes)) {
37012946 319 pol->v.preferred_node = node;
fc36b8d3
LS
320 pol->flags &= ~MPOL_F_LOCAL;
321 } else
322 pol->flags |= MPOL_F_LOCAL;
37012946
DR
323 } else if (pol->flags & MPOL_F_RELATIVE_NODES) {
324 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
325 pol->v.preferred_node = first_node(tmp);
fc36b8d3 326 } else if (!(pol->flags & MPOL_F_LOCAL)) {
37012946
DR
327 pol->v.preferred_node = node_remap(pol->v.preferred_node,
328 pol->w.cpuset_mems_allowed,
329 *nodes);
330 pol->w.cpuset_mems_allowed = *nodes;
331 }
1da177e4
LT
332}
333
1d0d2680
DR
334/* Migrate a policy to a different set of nodes */
335static void mpol_rebind_policy(struct mempolicy *pol,
336 const nodemask_t *newmask)
337{
1d0d2680
DR
338 if (!pol)
339 return;
1d0d2680
DR
340 if (!mpol_store_user_nodemask(pol) &&
341 nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
342 return;
45c4745a 343 mpol_ops[pol->mode].rebind(pol, newmask);
1d0d2680
DR
344}
345
346/*
347 * Wrapper for mpol_rebind_policy() that just requires task
348 * pointer, and updates task mempolicy.
58568d2a
MX
349 *
350 * Called with task's alloc_lock held.
1d0d2680
DR
351 */
352
353void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new)
354{
355 mpol_rebind_policy(tsk->mempolicy, new);
356}
357
358/*
359 * Rebind each vma in mm to new nodemask.
360 *
361 * Call holding a reference to mm. Takes mm->mmap_sem during call.
362 */
363
364void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
365{
366 struct vm_area_struct *vma;
367
368 down_write(&mm->mmap_sem);
369 for (vma = mm->mmap; vma; vma = vma->vm_next)
370 mpol_rebind_policy(vma->vm_policy, new);
371 up_write(&mm->mmap_sem);
372}
373
37012946
DR
374static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
375 [MPOL_DEFAULT] = {
376 .rebind = mpol_rebind_default,
377 },
378 [MPOL_INTERLEAVE] = {
379 .create = mpol_new_interleave,
380 .rebind = mpol_rebind_nodemask,
381 },
382 [MPOL_PREFERRED] = {
383 .create = mpol_new_preferred,
384 .rebind = mpol_rebind_preferred,
385 },
386 [MPOL_BIND] = {
387 .create = mpol_new_bind,
388 .rebind = mpol_rebind_nodemask,
389 },
390};
391
397874df 392static void gather_stats(struct page *, void *, int pte_dirty);
fc301289
CL
393static void migrate_page_add(struct page *page, struct list_head *pagelist,
394 unsigned long flags);
1a75a6c8 395
38e35860 396/* Scan through pages checking if pages follow certain conditions. */
b5810039 397static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
dc9aa5b9
CL
398 unsigned long addr, unsigned long end,
399 const nodemask_t *nodes, unsigned long flags,
38e35860 400 void *private)
1da177e4 401{
91612e0d
HD
402 pte_t *orig_pte;
403 pte_t *pte;
705e87c0 404 spinlock_t *ptl;
941150a3 405
705e87c0 406 orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
91612e0d 407 do {
6aab341e 408 struct page *page;
25ba77c1 409 int nid;
91612e0d
HD
410
411 if (!pte_present(*pte))
1da177e4 412 continue;
6aab341e
LT
413 page = vm_normal_page(vma, addr, *pte);
414 if (!page)
1da177e4 415 continue;
053837fc 416 /*
62b61f61
HD
417 * vm_normal_page() filters out zero pages, but there might
418 * still be PageReserved pages to skip, perhaps in a VDSO.
419 * And we cannot move PageKsm pages sensibly or safely yet.
053837fc 420 */
62b61f61 421 if (PageReserved(page) || PageKsm(page))
f4598c8b 422 continue;
6aab341e 423 nid = page_to_nid(page);
38e35860
CL
424 if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT))
425 continue;
426
1a75a6c8 427 if (flags & MPOL_MF_STATS)
397874df 428 gather_stats(page, private, pte_dirty(*pte));
053837fc 429 else if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
fc301289 430 migrate_page_add(page, private, flags);
38e35860
CL
431 else
432 break;
91612e0d 433 } while (pte++, addr += PAGE_SIZE, addr != end);
705e87c0 434 pte_unmap_unlock(orig_pte, ptl);
91612e0d
HD
435 return addr != end;
436}
437
b5810039 438static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud,
dc9aa5b9
CL
439 unsigned long addr, unsigned long end,
440 const nodemask_t *nodes, unsigned long flags,
38e35860 441 void *private)
91612e0d
HD
442{
443 pmd_t *pmd;
444 unsigned long next;
445
446 pmd = pmd_offset(pud, addr);
447 do {
448 next = pmd_addr_end(addr, end);
449 if (pmd_none_or_clear_bad(pmd))
450 continue;
dc9aa5b9 451 if (check_pte_range(vma, pmd, addr, next, nodes,
38e35860 452 flags, private))
91612e0d
HD
453 return -EIO;
454 } while (pmd++, addr = next, addr != end);
455 return 0;
456}
457
b5810039 458static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
dc9aa5b9
CL
459 unsigned long addr, unsigned long end,
460 const nodemask_t *nodes, unsigned long flags,
38e35860 461 void *private)
91612e0d
HD
462{
463 pud_t *pud;
464 unsigned long next;
465
466 pud = pud_offset(pgd, addr);
467 do {
468 next = pud_addr_end(addr, end);
469 if (pud_none_or_clear_bad(pud))
470 continue;
dc9aa5b9 471 if (check_pmd_range(vma, pud, addr, next, nodes,
38e35860 472 flags, private))
91612e0d
HD
473 return -EIO;
474 } while (pud++, addr = next, addr != end);
475 return 0;
476}
477
b5810039 478static inline int check_pgd_range(struct vm_area_struct *vma,
dc9aa5b9
CL
479 unsigned long addr, unsigned long end,
480 const nodemask_t *nodes, unsigned long flags,
38e35860 481 void *private)
91612e0d
HD
482{
483 pgd_t *pgd;
484 unsigned long next;
485
b5810039 486 pgd = pgd_offset(vma->vm_mm, addr);
91612e0d
HD
487 do {
488 next = pgd_addr_end(addr, end);
489 if (pgd_none_or_clear_bad(pgd))
490 continue;
dc9aa5b9 491 if (check_pud_range(vma, pgd, addr, next, nodes,
38e35860 492 flags, private))
91612e0d
HD
493 return -EIO;
494 } while (pgd++, addr = next, addr != end);
495 return 0;
1da177e4
LT
496}
497
dc9aa5b9
CL
498/*
499 * Check if all pages in a range are on a set of nodes.
500 * If pagelist != NULL then isolate pages from the LRU and
501 * put them on the pagelist.
502 */
1da177e4
LT
503static struct vm_area_struct *
504check_range(struct mm_struct *mm, unsigned long start, unsigned long end,
38e35860 505 const nodemask_t *nodes, unsigned long flags, void *private)
1da177e4
LT
506{
507 int err;
508 struct vm_area_struct *first, *vma, *prev;
509
053837fc 510
1da177e4
LT
511 first = find_vma(mm, start);
512 if (!first)
513 return ERR_PTR(-EFAULT);
514 prev = NULL;
515 for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) {
dc9aa5b9
CL
516 if (!(flags & MPOL_MF_DISCONTIG_OK)) {
517 if (!vma->vm_next && vma->vm_end < end)
518 return ERR_PTR(-EFAULT);
519 if (prev && prev->vm_end < vma->vm_start)
520 return ERR_PTR(-EFAULT);
521 }
522 if (!is_vm_hugetlb_page(vma) &&
523 ((flags & MPOL_MF_STRICT) ||
524 ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) &&
525 vma_migratable(vma)))) {
5b952b3c 526 unsigned long endvma = vma->vm_end;
dc9aa5b9 527
5b952b3c
AK
528 if (endvma > end)
529 endvma = end;
530 if (vma->vm_start > start)
531 start = vma->vm_start;
dc9aa5b9 532 err = check_pgd_range(vma, start, endvma, nodes,
38e35860 533 flags, private);
1da177e4
LT
534 if (err) {
535 first = ERR_PTR(err);
536 break;
537 }
538 }
539 prev = vma;
540 }
541 return first;
542}
543
544/* Apply policy to a single VMA */
545static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new)
546{
547 int err = 0;
548 struct mempolicy *old = vma->vm_policy;
549
140d5a49 550 pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
1da177e4
LT
551 vma->vm_start, vma->vm_end, vma->vm_pgoff,
552 vma->vm_ops, vma->vm_file,
553 vma->vm_ops ? vma->vm_ops->set_policy : NULL);
554
555 if (vma->vm_ops && vma->vm_ops->set_policy)
556 err = vma->vm_ops->set_policy(vma, new);
557 if (!err) {
558 mpol_get(new);
559 vma->vm_policy = new;
f0be3d32 560 mpol_put(old);
1da177e4
LT
561 }
562 return err;
563}
564
565/* Step 2: apply policy to a range and do splits. */
9d8cebd4
KM
566static int mbind_range(struct mm_struct *mm, unsigned long start,
567 unsigned long end, struct mempolicy *new_pol)
1da177e4
LT
568{
569 struct vm_area_struct *next;
9d8cebd4
KM
570 struct vm_area_struct *prev;
571 struct vm_area_struct *vma;
572 int err = 0;
573 pgoff_t pgoff;
574 unsigned long vmstart;
575 unsigned long vmend;
1da177e4 576
9d8cebd4
KM
577 vma = find_vma_prev(mm, start, &prev);
578 if (!vma || vma->vm_start > start)
579 return -EFAULT;
580
581 for (; vma && vma->vm_start < end; prev = vma, vma = next) {
1da177e4 582 next = vma->vm_next;
9d8cebd4
KM
583 vmstart = max(start, vma->vm_start);
584 vmend = min(end, vma->vm_end);
585
586 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
587 prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
588 vma->anon_vma, vma->vm_file, pgoff, new_pol);
589 if (prev) {
590 vma = prev;
591 next = vma->vm_next;
592 continue;
593 }
594 if (vma->vm_start != vmstart) {
595 err = split_vma(vma->vm_mm, vma, vmstart, 1);
596 if (err)
597 goto out;
598 }
599 if (vma->vm_end != vmend) {
600 err = split_vma(vma->vm_mm, vma, vmend, 0);
601 if (err)
602 goto out;
603 }
604 err = policy_vma(vma, new_pol);
1da177e4 605 if (err)
9d8cebd4 606 goto out;
1da177e4 607 }
9d8cebd4
KM
608
609 out:
1da177e4
LT
610 return err;
611}
612
c61afb18
PJ
613/*
614 * Update task->flags PF_MEMPOLICY bit: set iff non-default
615 * mempolicy. Allows more rapid checking of this (combined perhaps
616 * with other PF_* flag bits) on memory allocation hot code paths.
617 *
618 * If called from outside this file, the task 'p' should -only- be
619 * a newly forked child not yet visible on the task list, because
620 * manipulating the task flags of a visible task is not safe.
621 *
622 * The above limitation is why this routine has the funny name
623 * mpol_fix_fork_child_flag().
624 *
625 * It is also safe to call this with a task pointer of current,
626 * which the static wrapper mpol_set_task_struct_flag() does,
627 * for use within this file.
628 */
629
630void mpol_fix_fork_child_flag(struct task_struct *p)
631{
632 if (p->mempolicy)
633 p->flags |= PF_MEMPOLICY;
634 else
635 p->flags &= ~PF_MEMPOLICY;
636}
637
638static void mpol_set_task_struct_flag(void)
639{
640 mpol_fix_fork_child_flag(current);
641}
642
1da177e4 643/* Set the process memory policy */
028fec41
DR
644static long do_set_mempolicy(unsigned short mode, unsigned short flags,
645 nodemask_t *nodes)
1da177e4 646{
58568d2a 647 struct mempolicy *new, *old;
f4e53d91 648 struct mm_struct *mm = current->mm;
4bfc4495 649 NODEMASK_SCRATCH(scratch);
58568d2a 650 int ret;
1da177e4 651
4bfc4495
KH
652 if (!scratch)
653 return -ENOMEM;
f4e53d91 654
4bfc4495
KH
655 new = mpol_new(mode, flags, nodes);
656 if (IS_ERR(new)) {
657 ret = PTR_ERR(new);
658 goto out;
659 }
f4e53d91
LS
660 /*
661 * prevent changing our mempolicy while show_numa_maps()
662 * is using it.
663 * Note: do_set_mempolicy() can be called at init time
664 * with no 'mm'.
665 */
666 if (mm)
667 down_write(&mm->mmap_sem);
58568d2a 668 task_lock(current);
4bfc4495 669 ret = mpol_set_nodemask(new, nodes, scratch);
58568d2a
MX
670 if (ret) {
671 task_unlock(current);
672 if (mm)
673 up_write(&mm->mmap_sem);
674 mpol_put(new);
4bfc4495 675 goto out;
58568d2a
MX
676 }
677 old = current->mempolicy;
1da177e4 678 current->mempolicy = new;
c61afb18 679 mpol_set_task_struct_flag();
45c4745a 680 if (new && new->mode == MPOL_INTERLEAVE &&
f5b087b5 681 nodes_weight(new->v.nodes))
dfcd3c0d 682 current->il_next = first_node(new->v.nodes);
58568d2a 683 task_unlock(current);
f4e53d91
LS
684 if (mm)
685 up_write(&mm->mmap_sem);
686
58568d2a 687 mpol_put(old);
4bfc4495
KH
688 ret = 0;
689out:
690 NODEMASK_SCRATCH_FREE(scratch);
691 return ret;
1da177e4
LT
692}
693
bea904d5
LS
694/*
695 * Return nodemask for policy for get_mempolicy() query
58568d2a
MX
696 *
697 * Called with task's alloc_lock held
bea904d5
LS
698 */
699static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
1da177e4 700{
dfcd3c0d 701 nodes_clear(*nodes);
bea904d5
LS
702 if (p == &default_policy)
703 return;
704
45c4745a 705 switch (p->mode) {
19770b32
MG
706 case MPOL_BIND:
707 /* Fall through */
1da177e4 708 case MPOL_INTERLEAVE:
dfcd3c0d 709 *nodes = p->v.nodes;
1da177e4
LT
710 break;
711 case MPOL_PREFERRED:
fc36b8d3 712 if (!(p->flags & MPOL_F_LOCAL))
dfcd3c0d 713 node_set(p->v.preferred_node, *nodes);
53f2556b 714 /* else return empty node mask for local allocation */
1da177e4
LT
715 break;
716 default:
717 BUG();
718 }
719}
720
721static int lookup_node(struct mm_struct *mm, unsigned long addr)
722{
723 struct page *p;
724 int err;
725
726 err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL);
727 if (err >= 0) {
728 err = page_to_nid(p);
729 put_page(p);
730 }
731 return err;
732}
733
1da177e4 734/* Retrieve NUMA policy */
dbcb0f19
AB
735static long do_get_mempolicy(int *policy, nodemask_t *nmask,
736 unsigned long addr, unsigned long flags)
1da177e4 737{
8bccd85f 738 int err;
1da177e4
LT
739 struct mm_struct *mm = current->mm;
740 struct vm_area_struct *vma = NULL;
741 struct mempolicy *pol = current->mempolicy;
742
754af6f5
LS
743 if (flags &
744 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
1da177e4 745 return -EINVAL;
754af6f5
LS
746
747 if (flags & MPOL_F_MEMS_ALLOWED) {
748 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
749 return -EINVAL;
750 *policy = 0; /* just so it's initialized */
58568d2a 751 task_lock(current);
754af6f5 752 *nmask = cpuset_current_mems_allowed;
58568d2a 753 task_unlock(current);
754af6f5
LS
754 return 0;
755 }
756
1da177e4 757 if (flags & MPOL_F_ADDR) {
bea904d5
LS
758 /*
759 * Do NOT fall back to task policy if the
760 * vma/shared policy at addr is NULL. We
761 * want to return MPOL_DEFAULT in this case.
762 */
1da177e4
LT
763 down_read(&mm->mmap_sem);
764 vma = find_vma_intersection(mm, addr, addr+1);
765 if (!vma) {
766 up_read(&mm->mmap_sem);
767 return -EFAULT;
768 }
769 if (vma->vm_ops && vma->vm_ops->get_policy)
770 pol = vma->vm_ops->get_policy(vma, addr);
771 else
772 pol = vma->vm_policy;
773 } else if (addr)
774 return -EINVAL;
775
776 if (!pol)
bea904d5 777 pol = &default_policy; /* indicates default behavior */
1da177e4
LT
778
779 if (flags & MPOL_F_NODE) {
780 if (flags & MPOL_F_ADDR) {
781 err = lookup_node(mm, addr);
782 if (err < 0)
783 goto out;
8bccd85f 784 *policy = err;
1da177e4 785 } else if (pol == current->mempolicy &&
45c4745a 786 pol->mode == MPOL_INTERLEAVE) {
8bccd85f 787 *policy = current->il_next;
1da177e4
LT
788 } else {
789 err = -EINVAL;
790 goto out;
791 }
bea904d5
LS
792 } else {
793 *policy = pol == &default_policy ? MPOL_DEFAULT :
794 pol->mode;
d79df630
DR
795 /*
796 * Internal mempolicy flags must be masked off before exposing
797 * the policy to userspace.
798 */
799 *policy |= (pol->flags & MPOL_MODE_FLAGS);
bea904d5 800 }
1da177e4
LT
801
802 if (vma) {
803 up_read(&current->mm->mmap_sem);
804 vma = NULL;
805 }
806
1da177e4 807 err = 0;
58568d2a
MX
808 if (nmask) {
809 task_lock(current);
bea904d5 810 get_policy_nodemask(pol, nmask);
58568d2a
MX
811 task_unlock(current);
812 }
1da177e4
LT
813
814 out:
52cd3b07 815 mpol_cond_put(pol);
1da177e4
LT
816 if (vma)
817 up_read(&current->mm->mmap_sem);
818 return err;
819}
820
b20a3503 821#ifdef CONFIG_MIGRATION
6ce3c4c0
CL
822/*
823 * page migration
824 */
fc301289
CL
825static void migrate_page_add(struct page *page, struct list_head *pagelist,
826 unsigned long flags)
6ce3c4c0
CL
827{
828 /*
fc301289 829 * Avoid migrating a page that is shared with others.
6ce3c4c0 830 */
62695a84
NP
831 if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) {
832 if (!isolate_lru_page(page)) {
833 list_add_tail(&page->lru, pagelist);
6d9c285a
KM
834 inc_zone_page_state(page, NR_ISOLATED_ANON +
835 page_is_file_cache(page));
62695a84
NP
836 }
837 }
7e2ab150 838}
6ce3c4c0 839
742755a1 840static struct page *new_node_page(struct page *page, unsigned long node, int **x)
95a402c3 841{
6484eb3e 842 return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0);
95a402c3
CL
843}
844
7e2ab150
CL
845/*
846 * Migrate pages from one node to a target node.
847 * Returns error or the number of pages not migrated.
848 */
dbcb0f19
AB
849static int migrate_to_node(struct mm_struct *mm, int source, int dest,
850 int flags)
7e2ab150
CL
851{
852 nodemask_t nmask;
853 LIST_HEAD(pagelist);
854 int err = 0;
855
856 nodes_clear(nmask);
857 node_set(source, nmask);
6ce3c4c0 858
7e2ab150
CL
859 check_range(mm, mm->mmap->vm_start, TASK_SIZE, &nmask,
860 flags | MPOL_MF_DISCONTIG_OK, &pagelist);
861
aaa994b3 862 if (!list_empty(&pagelist))
62b61f61 863 err = migrate_pages(&pagelist, new_node_page, dest, 0);
95a402c3 864
7e2ab150 865 return err;
6ce3c4c0
CL
866}
867
39743889 868/*
7e2ab150
CL
869 * Move pages between the two nodesets so as to preserve the physical
870 * layout as much as possible.
39743889
CL
871 *
872 * Returns the number of page that could not be moved.
873 */
874int do_migrate_pages(struct mm_struct *mm,
875 const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags)
876{
7e2ab150 877 int busy = 0;
0aedadf9 878 int err;
7e2ab150 879 nodemask_t tmp;
39743889 880
0aedadf9
CL
881 err = migrate_prep();
882 if (err)
883 return err;
884
53f2556b 885 down_read(&mm->mmap_sem);
39743889 886
7b2259b3
CL
887 err = migrate_vmas(mm, from_nodes, to_nodes, flags);
888 if (err)
889 goto out;
890
da0aa138
KM
891 /*
892 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
893 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
894 * bit in 'tmp', and return that <source, dest> pair for migration.
895 * The pair of nodemasks 'to' and 'from' define the map.
896 *
897 * If no pair of bits is found that way, fallback to picking some
898 * pair of 'source' and 'dest' bits that are not the same. If the
899 * 'source' and 'dest' bits are the same, this represents a node
900 * that will be migrating to itself, so no pages need move.
901 *
902 * If no bits are left in 'tmp', or if all remaining bits left
903 * in 'tmp' correspond to the same bit in 'to', return false
904 * (nothing left to migrate).
905 *
906 * This lets us pick a pair of nodes to migrate between, such that
907 * if possible the dest node is not already occupied by some other
908 * source node, minimizing the risk of overloading the memory on a
909 * node that would happen if we migrated incoming memory to a node
910 * before migrating outgoing memory source that same node.
911 *
912 * A single scan of tmp is sufficient. As we go, we remember the
913 * most recent <s, d> pair that moved (s != d). If we find a pair
914 * that not only moved, but what's better, moved to an empty slot
915 * (d is not set in tmp), then we break out then, with that pair.
916 * Otherwise when we finish scannng from_tmp, we at least have the
917 * most recent <s, d> pair that moved. If we get all the way through
918 * the scan of tmp without finding any node that moved, much less
919 * moved to an empty node, then there is nothing left worth migrating.
920 */
d4984711 921
7e2ab150
CL
922 tmp = *from_nodes;
923 while (!nodes_empty(tmp)) {
924 int s,d;
925 int source = -1;
926 int dest = 0;
927
928 for_each_node_mask(s, tmp) {
929 d = node_remap(s, *from_nodes, *to_nodes);
930 if (s == d)
931 continue;
932
933 source = s; /* Node moved. Memorize */
934 dest = d;
935
936 /* dest not in remaining from nodes? */
937 if (!node_isset(dest, tmp))
938 break;
939 }
940 if (source == -1)
941 break;
942
943 node_clear(source, tmp);
944 err = migrate_to_node(mm, source, dest, flags);
945 if (err > 0)
946 busy += err;
947 if (err < 0)
948 break;
39743889 949 }
7b2259b3 950out:
39743889 951 up_read(&mm->mmap_sem);
7e2ab150
CL
952 if (err < 0)
953 return err;
954 return busy;
b20a3503
CL
955
956}
957
3ad33b24
LS
958/*
959 * Allocate a new page for page migration based on vma policy.
960 * Start assuming that page is mapped by vma pointed to by @private.
961 * Search forward from there, if not. N.B., this assumes that the
962 * list of pages handed to migrate_pages()--which is how we get here--
963 * is in virtual address order.
964 */
742755a1 965static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
95a402c3
CL
966{
967 struct vm_area_struct *vma = (struct vm_area_struct *)private;
3ad33b24 968 unsigned long uninitialized_var(address);
95a402c3 969
3ad33b24
LS
970 while (vma) {
971 address = page_address_in_vma(page, vma);
972 if (address != -EFAULT)
973 break;
974 vma = vma->vm_next;
975 }
976
977 /*
978 * if !vma, alloc_page_vma() will use task or system default policy
979 */
980 return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
95a402c3 981}
b20a3503
CL
982#else
983
984static void migrate_page_add(struct page *page, struct list_head *pagelist,
985 unsigned long flags)
986{
39743889
CL
987}
988
b20a3503
CL
989int do_migrate_pages(struct mm_struct *mm,
990 const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags)
991{
992 return -ENOSYS;
993}
95a402c3 994
69939749 995static struct page *new_vma_page(struct page *page, unsigned long private, int **x)
95a402c3
CL
996{
997 return NULL;
998}
b20a3503
CL
999#endif
1000
dbcb0f19 1001static long do_mbind(unsigned long start, unsigned long len,
028fec41
DR
1002 unsigned short mode, unsigned short mode_flags,
1003 nodemask_t *nmask, unsigned long flags)
6ce3c4c0
CL
1004{
1005 struct vm_area_struct *vma;
1006 struct mm_struct *mm = current->mm;
1007 struct mempolicy *new;
1008 unsigned long end;
1009 int err;
1010 LIST_HEAD(pagelist);
1011
a3b51e01
DR
1012 if (flags & ~(unsigned long)(MPOL_MF_STRICT |
1013 MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
6ce3c4c0 1014 return -EINVAL;
74c00241 1015 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
6ce3c4c0
CL
1016 return -EPERM;
1017
1018 if (start & ~PAGE_MASK)
1019 return -EINVAL;
1020
1021 if (mode == MPOL_DEFAULT)
1022 flags &= ~MPOL_MF_STRICT;
1023
1024 len = (len + PAGE_SIZE - 1) & PAGE_MASK;
1025 end = start + len;
1026
1027 if (end < start)
1028 return -EINVAL;
1029 if (end == start)
1030 return 0;
1031
028fec41 1032 new = mpol_new(mode, mode_flags, nmask);
6ce3c4c0
CL
1033 if (IS_ERR(new))
1034 return PTR_ERR(new);
1035
1036 /*
1037 * If we are using the default policy then operation
1038 * on discontinuous address spaces is okay after all
1039 */
1040 if (!new)
1041 flags |= MPOL_MF_DISCONTIG_OK;
1042
028fec41
DR
1043 pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
1044 start, start + len, mode, mode_flags,
1045 nmask ? nodes_addr(*nmask)[0] : -1);
6ce3c4c0 1046
0aedadf9
CL
1047 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
1048
1049 err = migrate_prep();
1050 if (err)
b05ca738 1051 goto mpol_out;
0aedadf9 1052 }
4bfc4495
KH
1053 {
1054 NODEMASK_SCRATCH(scratch);
1055 if (scratch) {
1056 down_write(&mm->mmap_sem);
1057 task_lock(current);
1058 err = mpol_set_nodemask(new, nmask, scratch);
1059 task_unlock(current);
1060 if (err)
1061 up_write(&mm->mmap_sem);
1062 } else
1063 err = -ENOMEM;
1064 NODEMASK_SCRATCH_FREE(scratch);
1065 }
b05ca738
KM
1066 if (err)
1067 goto mpol_out;
1068
6ce3c4c0
CL
1069 vma = check_range(mm, start, end, nmask,
1070 flags | MPOL_MF_INVERT, &pagelist);
1071
1072 err = PTR_ERR(vma);
1073 if (!IS_ERR(vma)) {
1074 int nr_failed = 0;
1075
9d8cebd4 1076 err = mbind_range(mm, start, end, new);
7e2ab150 1077
6ce3c4c0 1078 if (!list_empty(&pagelist))
95a402c3 1079 nr_failed = migrate_pages(&pagelist, new_vma_page,
62b61f61 1080 (unsigned long)vma, 0);
6ce3c4c0
CL
1081
1082 if (!err && nr_failed && (flags & MPOL_MF_STRICT))
1083 err = -EIO;
ab8a3e14
KM
1084 } else
1085 putback_lru_pages(&pagelist);
b20a3503 1086
6ce3c4c0 1087 up_write(&mm->mmap_sem);
b05ca738 1088 mpol_out:
f0be3d32 1089 mpol_put(new);
6ce3c4c0
CL
1090 return err;
1091}
1092
8bccd85f
CL
1093/*
1094 * User space interface with variable sized bitmaps for nodelists.
1095 */
1096
1097/* Copy a node mask from user space. */
39743889 1098static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
8bccd85f
CL
1099 unsigned long maxnode)
1100{
1101 unsigned long k;
1102 unsigned long nlongs;
1103 unsigned long endmask;
1104
1105 --maxnode;
1106 nodes_clear(*nodes);
1107 if (maxnode == 0 || !nmask)
1108 return 0;
a9c930ba 1109 if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
636f13c1 1110 return -EINVAL;
8bccd85f
CL
1111
1112 nlongs = BITS_TO_LONGS(maxnode);
1113 if ((maxnode % BITS_PER_LONG) == 0)
1114 endmask = ~0UL;
1115 else
1116 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
1117
1118 /* When the user specified more nodes than supported just check
1119 if the non supported part is all zero. */
1120 if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
1121 if (nlongs > PAGE_SIZE/sizeof(long))
1122 return -EINVAL;
1123 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
1124 unsigned long t;
1125 if (get_user(t, nmask + k))
1126 return -EFAULT;
1127 if (k == nlongs - 1) {
1128 if (t & endmask)
1129 return -EINVAL;
1130 } else if (t)
1131 return -EINVAL;
1132 }
1133 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
1134 endmask = ~0UL;
1135 }
1136
1137 if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
1138 return -EFAULT;
1139 nodes_addr(*nodes)[nlongs-1] &= endmask;
1140 return 0;
1141}
1142
1143/* Copy a kernel node mask to user space */
1144static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1145 nodemask_t *nodes)
1146{
1147 unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1148 const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long);
1149
1150 if (copy > nbytes) {
1151 if (copy > PAGE_SIZE)
1152 return -EINVAL;
1153 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1154 return -EFAULT;
1155 copy = nbytes;
1156 }
1157 return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1158}
1159
938bb9f5
HC
1160SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
1161 unsigned long, mode, unsigned long __user *, nmask,
1162 unsigned long, maxnode, unsigned, flags)
8bccd85f
CL
1163{
1164 nodemask_t nodes;
1165 int err;
028fec41 1166 unsigned short mode_flags;
8bccd85f 1167
028fec41
DR
1168 mode_flags = mode & MPOL_MODE_FLAGS;
1169 mode &= ~MPOL_MODE_FLAGS;
a3b51e01
DR
1170 if (mode >= MPOL_MAX)
1171 return -EINVAL;
4c50bc01
DR
1172 if ((mode_flags & MPOL_F_STATIC_NODES) &&
1173 (mode_flags & MPOL_F_RELATIVE_NODES))
1174 return -EINVAL;
8bccd85f
CL
1175 err = get_nodes(&nodes, nmask, maxnode);
1176 if (err)
1177 return err;
028fec41 1178 return do_mbind(start, len, mode, mode_flags, &nodes, flags);
8bccd85f
CL
1179}
1180
1181/* Set the process memory policy */
938bb9f5
HC
1182SYSCALL_DEFINE3(set_mempolicy, int, mode, unsigned long __user *, nmask,
1183 unsigned long, maxnode)
8bccd85f
CL
1184{
1185 int err;
1186 nodemask_t nodes;
028fec41 1187 unsigned short flags;
8bccd85f 1188
028fec41
DR
1189 flags = mode & MPOL_MODE_FLAGS;
1190 mode &= ~MPOL_MODE_FLAGS;
1191 if ((unsigned int)mode >= MPOL_MAX)
8bccd85f 1192 return -EINVAL;
4c50bc01
DR
1193 if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
1194 return -EINVAL;
8bccd85f
CL
1195 err = get_nodes(&nodes, nmask, maxnode);
1196 if (err)
1197 return err;
028fec41 1198 return do_set_mempolicy(mode, flags, &nodes);
8bccd85f
CL
1199}
1200
938bb9f5
HC
1201SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
1202 const unsigned long __user *, old_nodes,
1203 const unsigned long __user *, new_nodes)
39743889 1204{
c69e8d9c 1205 const struct cred *cred = current_cred(), *tcred;
39743889
CL
1206 struct mm_struct *mm;
1207 struct task_struct *task;
1208 nodemask_t old;
1209 nodemask_t new;
1210 nodemask_t task_nodes;
1211 int err;
1212
1213 err = get_nodes(&old, old_nodes, maxnode);
1214 if (err)
1215 return err;
1216
1217 err = get_nodes(&new, new_nodes, maxnode);
1218 if (err)
1219 return err;
1220
1221 /* Find the mm_struct */
1222 read_lock(&tasklist_lock);
228ebcbe 1223 task = pid ? find_task_by_vpid(pid) : current;
39743889
CL
1224 if (!task) {
1225 read_unlock(&tasklist_lock);
1226 return -ESRCH;
1227 }
1228 mm = get_task_mm(task);
1229 read_unlock(&tasklist_lock);
1230
1231 if (!mm)
1232 return -EINVAL;
1233
1234 /*
1235 * Check if this process has the right to modify the specified
1236 * process. The right exists if the process has administrative
7f927fcc 1237 * capabilities, superuser privileges or the same
39743889
CL
1238 * userid as the target process.
1239 */
c69e8d9c
DH
1240 rcu_read_lock();
1241 tcred = __task_cred(task);
b6dff3ec
DH
1242 if (cred->euid != tcred->suid && cred->euid != tcred->uid &&
1243 cred->uid != tcred->suid && cred->uid != tcred->uid &&
74c00241 1244 !capable(CAP_SYS_NICE)) {
c69e8d9c 1245 rcu_read_unlock();
39743889
CL
1246 err = -EPERM;
1247 goto out;
1248 }
c69e8d9c 1249 rcu_read_unlock();
39743889
CL
1250
1251 task_nodes = cpuset_mems_allowed(task);
1252 /* Is the user allowed to access the target nodes? */
74c00241 1253 if (!nodes_subset(new, task_nodes) && !capable(CAP_SYS_NICE)) {
39743889
CL
1254 err = -EPERM;
1255 goto out;
1256 }
1257
37b07e41 1258 if (!nodes_subset(new, node_states[N_HIGH_MEMORY])) {
3b42d28b
CL
1259 err = -EINVAL;
1260 goto out;
1261 }
1262
86c3a764
DQ
1263 err = security_task_movememory(task);
1264 if (err)
1265 goto out;
1266
511030bc 1267 err = do_migrate_pages(mm, &old, &new,
74c00241 1268 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
39743889
CL
1269out:
1270 mmput(mm);
1271 return err;
1272}
1273
1274
8bccd85f 1275/* Retrieve NUMA policy */
938bb9f5
HC
1276SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1277 unsigned long __user *, nmask, unsigned long, maxnode,
1278 unsigned long, addr, unsigned long, flags)
8bccd85f 1279{
dbcb0f19
AB
1280 int err;
1281 int uninitialized_var(pval);
8bccd85f
CL
1282 nodemask_t nodes;
1283
1284 if (nmask != NULL && maxnode < MAX_NUMNODES)
1285 return -EINVAL;
1286
1287 err = do_get_mempolicy(&pval, &nodes, addr, flags);
1288
1289 if (err)
1290 return err;
1291
1292 if (policy && put_user(pval, policy))
1293 return -EFAULT;
1294
1295 if (nmask)
1296 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1297
1298 return err;
1299}
1300
1da177e4
LT
1301#ifdef CONFIG_COMPAT
1302
1303asmlinkage long compat_sys_get_mempolicy(int __user *policy,
1304 compat_ulong_t __user *nmask,
1305 compat_ulong_t maxnode,
1306 compat_ulong_t addr, compat_ulong_t flags)
1307{
1308 long err;
1309 unsigned long __user *nm = NULL;
1310 unsigned long nr_bits, alloc_size;
1311 DECLARE_BITMAP(bm, MAX_NUMNODES);
1312
1313 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1314 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1315
1316 if (nmask)
1317 nm = compat_alloc_user_space(alloc_size);
1318
1319 err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
1320
1321 if (!err && nmask) {
1322 err = copy_from_user(bm, nm, alloc_size);
1323 /* ensure entire bitmap is zeroed */
1324 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
1325 err |= compat_put_bitmap(nmask, bm, nr_bits);
1326 }
1327
1328 return err;
1329}
1330
1331asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask,
1332 compat_ulong_t maxnode)
1333{
1334 long err = 0;
1335 unsigned long __user *nm = NULL;
1336 unsigned long nr_bits, alloc_size;
1337 DECLARE_BITMAP(bm, MAX_NUMNODES);
1338
1339 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1340 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1341
1342 if (nmask) {
1343 err = compat_get_bitmap(bm, nmask, nr_bits);
1344 nm = compat_alloc_user_space(alloc_size);
1345 err |= copy_to_user(nm, bm, alloc_size);
1346 }
1347
1348 if (err)
1349 return -EFAULT;
1350
1351 return sys_set_mempolicy(mode, nm, nr_bits+1);
1352}
1353
1354asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len,
1355 compat_ulong_t mode, compat_ulong_t __user *nmask,
1356 compat_ulong_t maxnode, compat_ulong_t flags)
1357{
1358 long err = 0;
1359 unsigned long __user *nm = NULL;
1360 unsigned long nr_bits, alloc_size;
dfcd3c0d 1361 nodemask_t bm;
1da177e4
LT
1362
1363 nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1364 alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1365
1366 if (nmask) {
dfcd3c0d 1367 err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits);
1da177e4 1368 nm = compat_alloc_user_space(alloc_size);
dfcd3c0d 1369 err |= copy_to_user(nm, nodes_addr(bm), alloc_size);
1da177e4
LT
1370 }
1371
1372 if (err)
1373 return -EFAULT;
1374
1375 return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
1376}
1377
1378#endif
1379
480eccf9
LS
1380/*
1381 * get_vma_policy(@task, @vma, @addr)
1382 * @task - task for fallback if vma policy == default
1383 * @vma - virtual memory area whose policy is sought
1384 * @addr - address in @vma for shared policy lookup
1385 *
1386 * Returns effective policy for a VMA at specified address.
1387 * Falls back to @task or system default policy, as necessary.
52cd3b07
LS
1388 * Current or other task's task mempolicy and non-shared vma policies
1389 * are protected by the task's mmap_sem, which must be held for read by
1390 * the caller.
1391 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1392 * count--added by the get_policy() vm_op, as appropriate--to protect against
1393 * freeing by another task. It is the caller's responsibility to free the
1394 * extra reference for shared policies.
480eccf9 1395 */
ae4d8c16 1396static struct mempolicy *get_vma_policy(struct task_struct *task,
48fce342 1397 struct vm_area_struct *vma, unsigned long addr)
1da177e4 1398{
6e21c8f1 1399 struct mempolicy *pol = task->mempolicy;
1da177e4
LT
1400
1401 if (vma) {
480eccf9 1402 if (vma->vm_ops && vma->vm_ops->get_policy) {
ae4d8c16
LS
1403 struct mempolicy *vpol = vma->vm_ops->get_policy(vma,
1404 addr);
1405 if (vpol)
1406 pol = vpol;
bea904d5 1407 } else if (vma->vm_policy)
1da177e4
LT
1408 pol = vma->vm_policy;
1409 }
1410 if (!pol)
1411 pol = &default_policy;
1412 return pol;
1413}
1414
52cd3b07
LS
1415/*
1416 * Return a nodemask representing a mempolicy for filtering nodes for
1417 * page allocation
1418 */
1419static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
19770b32
MG
1420{
1421 /* Lower zones don't get a nodemask applied for MPOL_BIND */
45c4745a 1422 if (unlikely(policy->mode == MPOL_BIND) &&
19770b32
MG
1423 gfp_zone(gfp) >= policy_zone &&
1424 cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
1425 return &policy->v.nodes;
1426
1427 return NULL;
1428}
1429
52cd3b07
LS
1430/* Return a zonelist indicated by gfp for node representing a mempolicy */
1431static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy)
1da177e4 1432{
fc36b8d3 1433 int nd = numa_node_id();
1da177e4 1434
45c4745a 1435 switch (policy->mode) {
1da177e4 1436 case MPOL_PREFERRED:
fc36b8d3
LS
1437 if (!(policy->flags & MPOL_F_LOCAL))
1438 nd = policy->v.preferred_node;
1da177e4
LT
1439 break;
1440 case MPOL_BIND:
19770b32 1441 /*
52cd3b07
LS
1442 * Normally, MPOL_BIND allocations are node-local within the
1443 * allowed nodemask. However, if __GFP_THISNODE is set and the
1444 * current node is part of the mask, we use the zonelist for
1445 * the first node in the mask instead.
19770b32 1446 */
19770b32
MG
1447 if (unlikely(gfp & __GFP_THISNODE) &&
1448 unlikely(!node_isset(nd, policy->v.nodes)))
1449 nd = first_node(policy->v.nodes);
1450 break;
1da177e4 1451 case MPOL_INTERLEAVE: /* should not happen */
1da177e4
LT
1452 break;
1453 default:
1da177e4
LT
1454 BUG();
1455 }
0e88460d 1456 return node_zonelist(nd, gfp);
1da177e4
LT
1457}
1458
1459/* Do dynamic interleaving for a process */
1460static unsigned interleave_nodes(struct mempolicy *policy)
1461{
1462 unsigned nid, next;
1463 struct task_struct *me = current;
1464
1465 nid = me->il_next;
dfcd3c0d 1466 next = next_node(nid, policy->v.nodes);
1da177e4 1467 if (next >= MAX_NUMNODES)
dfcd3c0d 1468 next = first_node(policy->v.nodes);
f5b087b5
DR
1469 if (next < MAX_NUMNODES)
1470 me->il_next = next;
1da177e4
LT
1471 return nid;
1472}
1473
dc85da15
CL
1474/*
1475 * Depending on the memory policy provide a node from which to allocate the
1476 * next slab entry.
52cd3b07
LS
1477 * @policy must be protected by freeing by the caller. If @policy is
1478 * the current task's mempolicy, this protection is implicit, as only the
1479 * task can change it's policy. The system default policy requires no
1480 * such protection.
dc85da15
CL
1481 */
1482unsigned slab_node(struct mempolicy *policy)
1483{
fc36b8d3 1484 if (!policy || policy->flags & MPOL_F_LOCAL)
bea904d5
LS
1485 return numa_node_id();
1486
1487 switch (policy->mode) {
1488 case MPOL_PREFERRED:
fc36b8d3
LS
1489 /*
1490 * handled MPOL_F_LOCAL above
1491 */
1492 return policy->v.preferred_node;
765c4507 1493
dc85da15
CL
1494 case MPOL_INTERLEAVE:
1495 return interleave_nodes(policy);
1496
dd1a239f 1497 case MPOL_BIND: {
dc85da15
CL
1498 /*
1499 * Follow bind policy behavior and start allocation at the
1500 * first node.
1501 */
19770b32
MG
1502 struct zonelist *zonelist;
1503 struct zone *zone;
1504 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
1505 zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0];
1506 (void)first_zones_zonelist(zonelist, highest_zoneidx,
1507 &policy->v.nodes,
1508 &zone);
1509 return zone->node;
dd1a239f 1510 }
dc85da15 1511
dc85da15 1512 default:
bea904d5 1513 BUG();
dc85da15
CL
1514 }
1515}
1516
1da177e4
LT
1517/* Do static interleaving for a VMA with known offset. */
1518static unsigned offset_il_node(struct mempolicy *pol,
1519 struct vm_area_struct *vma, unsigned long off)
1520{
dfcd3c0d 1521 unsigned nnodes = nodes_weight(pol->v.nodes);
f5b087b5 1522 unsigned target;
1da177e4
LT
1523 int c;
1524 int nid = -1;
1525
f5b087b5
DR
1526 if (!nnodes)
1527 return numa_node_id();
1528 target = (unsigned int)off % nnodes;
1da177e4
LT
1529 c = 0;
1530 do {
dfcd3c0d 1531 nid = next_node(nid, pol->v.nodes);
1da177e4
LT
1532 c++;
1533 } while (c <= target);
1da177e4
LT
1534 return nid;
1535}
1536
5da7ca86
CL
1537/* Determine a node number for interleave */
1538static inline unsigned interleave_nid(struct mempolicy *pol,
1539 struct vm_area_struct *vma, unsigned long addr, int shift)
1540{
1541 if (vma) {
1542 unsigned long off;
1543
3b98b087
NA
1544 /*
1545 * for small pages, there is no difference between
1546 * shift and PAGE_SHIFT, so the bit-shift is safe.
1547 * for huge pages, since vm_pgoff is in units of small
1548 * pages, we need to shift off the always 0 bits to get
1549 * a useful offset.
1550 */
1551 BUG_ON(shift < PAGE_SHIFT);
1552 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
5da7ca86
CL
1553 off += (addr - vma->vm_start) >> shift;
1554 return offset_il_node(pol, vma, off);
1555 } else
1556 return interleave_nodes(pol);
1557}
1558
00ac59ad 1559#ifdef CONFIG_HUGETLBFS
480eccf9
LS
1560/*
1561 * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
1562 * @vma = virtual memory area whose policy is sought
1563 * @addr = address in @vma for shared policy lookup and interleave policy
1564 * @gfp_flags = for requested zone
19770b32
MG
1565 * @mpol = pointer to mempolicy pointer for reference counted mempolicy
1566 * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask
480eccf9 1567 *
52cd3b07
LS
1568 * Returns a zonelist suitable for a huge page allocation and a pointer
1569 * to the struct mempolicy for conditional unref after allocation.
1570 * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1571 * @nodemask for filtering the zonelist.
480eccf9 1572 */
396faf03 1573struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
19770b32
MG
1574 gfp_t gfp_flags, struct mempolicy **mpol,
1575 nodemask_t **nodemask)
5da7ca86 1576{
480eccf9 1577 struct zonelist *zl;
5da7ca86 1578
52cd3b07 1579 *mpol = get_vma_policy(current, vma, addr);
19770b32 1580 *nodemask = NULL; /* assume !MPOL_BIND */
5da7ca86 1581
52cd3b07
LS
1582 if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
1583 zl = node_zonelist(interleave_nid(*mpol, vma, addr,
a5516438 1584 huge_page_shift(hstate_vma(vma))), gfp_flags);
52cd3b07
LS
1585 } else {
1586 zl = policy_zonelist(gfp_flags, *mpol);
1587 if ((*mpol)->mode == MPOL_BIND)
1588 *nodemask = &(*mpol)->v.nodes;
480eccf9
LS
1589 }
1590 return zl;
5da7ca86 1591}
06808b08
LS
1592
1593/*
1594 * init_nodemask_of_mempolicy
1595 *
1596 * If the current task's mempolicy is "default" [NULL], return 'false'
1597 * to indicate default policy. Otherwise, extract the policy nodemask
1598 * for 'bind' or 'interleave' policy into the argument nodemask, or
1599 * initialize the argument nodemask to contain the single node for
1600 * 'preferred' or 'local' policy and return 'true' to indicate presence
1601 * of non-default mempolicy.
1602 *
1603 * We don't bother with reference counting the mempolicy [mpol_get/put]
1604 * because the current task is examining it's own mempolicy and a task's
1605 * mempolicy is only ever changed by the task itself.
1606 *
1607 * N.B., it is the caller's responsibility to free a returned nodemask.
1608 */
1609bool init_nodemask_of_mempolicy(nodemask_t *mask)
1610{
1611 struct mempolicy *mempolicy;
1612 int nid;
1613
1614 if (!(mask && current->mempolicy))
1615 return false;
1616
1617 mempolicy = current->mempolicy;
1618 switch (mempolicy->mode) {
1619 case MPOL_PREFERRED:
1620 if (mempolicy->flags & MPOL_F_LOCAL)
1621 nid = numa_node_id();
1622 else
1623 nid = mempolicy->v.preferred_node;
1624 init_nodemask_of_node(mask, nid);
1625 break;
1626
1627 case MPOL_BIND:
1628 /* Fall through */
1629 case MPOL_INTERLEAVE:
1630 *mask = mempolicy->v.nodes;
1631 break;
1632
1633 default:
1634 BUG();
1635 }
1636
1637 return true;
1638}
00ac59ad 1639#endif
5da7ca86 1640
1da177e4
LT
1641/* Allocate a page in interleaved policy.
1642 Own path because it needs to do special accounting. */
662f3a0b
AK
1643static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
1644 unsigned nid)
1da177e4
LT
1645{
1646 struct zonelist *zl;
1647 struct page *page;
1648
0e88460d 1649 zl = node_zonelist(nid, gfp);
1da177e4 1650 page = __alloc_pages(gfp, order, zl);
dd1a239f 1651 if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
ca889e6c 1652 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
1da177e4
LT
1653 return page;
1654}
1655
1656/**
1657 * alloc_page_vma - Allocate a page for a VMA.
1658 *
1659 * @gfp:
1660 * %GFP_USER user allocation.
1661 * %GFP_KERNEL kernel allocations,
1662 * %GFP_HIGHMEM highmem/user allocations,
1663 * %GFP_FS allocation should not call back into a file system.
1664 * %GFP_ATOMIC don't sleep.
1665 *
1666 * @vma: Pointer to VMA or NULL if not available.
1667 * @addr: Virtual Address of the allocation. Must be inside the VMA.
1668 *
1669 * This function allocates a page from the kernel page pool and applies
1670 * a NUMA policy associated with the VMA or the current process.
1671 * When VMA is not NULL caller must hold down_read on the mmap_sem of the
1672 * mm_struct of the VMA to prevent it from going away. Should be used for
1673 * all allocations for pages that will be mapped into
1674 * user space. Returns NULL when no page can be allocated.
1675 *
1676 * Should be called with the mm_sem of the vma hold.
1677 */
1678struct page *
dd0fc66f 1679alloc_page_vma(gfp_t gfp, struct vm_area_struct *vma, unsigned long addr)
1da177e4 1680{
6e21c8f1 1681 struct mempolicy *pol = get_vma_policy(current, vma, addr);
480eccf9 1682 struct zonelist *zl;
1da177e4 1683
45c4745a 1684 if (unlikely(pol->mode == MPOL_INTERLEAVE)) {
1da177e4 1685 unsigned nid;
5da7ca86
CL
1686
1687 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT);
52cd3b07 1688 mpol_cond_put(pol);
1da177e4
LT
1689 return alloc_page_interleave(gfp, 0, nid);
1690 }
52cd3b07
LS
1691 zl = policy_zonelist(gfp, pol);
1692 if (unlikely(mpol_needs_cond_ref(pol))) {
480eccf9 1693 /*
52cd3b07 1694 * slow path: ref counted shared policy
480eccf9 1695 */
19770b32 1696 struct page *page = __alloc_pages_nodemask(gfp, 0,
52cd3b07 1697 zl, policy_nodemask(gfp, pol));
f0be3d32 1698 __mpol_put(pol);
480eccf9
LS
1699 return page;
1700 }
1701 /*
1702 * fast path: default or task policy
1703 */
52cd3b07 1704 return __alloc_pages_nodemask(gfp, 0, zl, policy_nodemask(gfp, pol));
1da177e4
LT
1705}
1706
1707/**
1708 * alloc_pages_current - Allocate pages.
1709 *
1710 * @gfp:
1711 * %GFP_USER user allocation,
1712 * %GFP_KERNEL kernel allocation,
1713 * %GFP_HIGHMEM highmem allocation,
1714 * %GFP_FS don't call back into a file system.
1715 * %GFP_ATOMIC don't sleep.
1716 * @order: Power of two of allocation size in pages. 0 is a single page.
1717 *
1718 * Allocate a page from the kernel page pool. When not in
1719 * interrupt context and apply the current process NUMA policy.
1720 * Returns NULL when no page can be allocated.
1721 *
cf2a473c 1722 * Don't call cpuset_update_task_memory_state() unless
1da177e4
LT
1723 * 1) it's ok to take cpuset_sem (can WAIT), and
1724 * 2) allocating for current task (not interrupt).
1725 */
dd0fc66f 1726struct page *alloc_pages_current(gfp_t gfp, unsigned order)
1da177e4
LT
1727{
1728 struct mempolicy *pol = current->mempolicy;
1729
9b819d20 1730 if (!pol || in_interrupt() || (gfp & __GFP_THISNODE))
1da177e4 1731 pol = &default_policy;
52cd3b07
LS
1732
1733 /*
1734 * No reference counting needed for current->mempolicy
1735 * nor system default_policy
1736 */
45c4745a 1737 if (pol->mode == MPOL_INTERLEAVE)
1da177e4 1738 return alloc_page_interleave(gfp, order, interleave_nodes(pol));
19770b32 1739 return __alloc_pages_nodemask(gfp, order,
52cd3b07 1740 policy_zonelist(gfp, pol), policy_nodemask(gfp, pol));
1da177e4
LT
1741}
1742EXPORT_SYMBOL(alloc_pages_current);
1743
4225399a 1744/*
846a16bf 1745 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
4225399a
PJ
1746 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
1747 * with the mems_allowed returned by cpuset_mems_allowed(). This
1748 * keeps mempolicies cpuset relative after its cpuset moves. See
1749 * further kernel/cpuset.c update_nodemask().
1750 */
4225399a 1751
846a16bf
LS
1752/* Slow path of a mempolicy duplicate */
1753struct mempolicy *__mpol_dup(struct mempolicy *old)
1da177e4
LT
1754{
1755 struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
1756
1757 if (!new)
1758 return ERR_PTR(-ENOMEM);
99ee4ca7 1759 rcu_read_lock();
4225399a
PJ
1760 if (current_cpuset_is_being_rebound()) {
1761 nodemask_t mems = cpuset_mems_allowed(current);
1762 mpol_rebind_policy(old, &mems);
1763 }
99ee4ca7 1764 rcu_read_unlock();
1da177e4
LT
1765 *new = *old;
1766 atomic_set(&new->refcnt, 1);
1da177e4
LT
1767 return new;
1768}
1769
52cd3b07
LS
1770/*
1771 * If *frompol needs [has] an extra ref, copy *frompol to *tompol ,
1772 * eliminate the * MPOL_F_* flags that require conditional ref and
1773 * [NOTE!!!] drop the extra ref. Not safe to reference *frompol directly
1774 * after return. Use the returned value.
1775 *
1776 * Allows use of a mempolicy for, e.g., multiple allocations with a single
1777 * policy lookup, even if the policy needs/has extra ref on lookup.
1778 * shmem_readahead needs this.
1779 */
1780struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol,
1781 struct mempolicy *frompol)
1782{
1783 if (!mpol_needs_cond_ref(frompol))
1784 return frompol;
1785
1786 *tompol = *frompol;
1787 tompol->flags &= ~MPOL_F_SHARED; /* copy doesn't need unref */
1788 __mpol_put(frompol);
1789 return tompol;
1790}
1791
f5b087b5
DR
1792static int mpol_match_intent(const struct mempolicy *a,
1793 const struct mempolicy *b)
1794{
1795 if (a->flags != b->flags)
1796 return 0;
1797 if (!mpol_store_user_nodemask(a))
1798 return 1;
1799 return nodes_equal(a->w.user_nodemask, b->w.user_nodemask);
1800}
1801
1da177e4
LT
1802/* Slow path of a mempolicy comparison */
1803int __mpol_equal(struct mempolicy *a, struct mempolicy *b)
1804{
1805 if (!a || !b)
1806 return 0;
45c4745a 1807 if (a->mode != b->mode)
1da177e4 1808 return 0;
45c4745a 1809 if (a->mode != MPOL_DEFAULT && !mpol_match_intent(a, b))
f5b087b5 1810 return 0;
45c4745a 1811 switch (a->mode) {
19770b32
MG
1812 case MPOL_BIND:
1813 /* Fall through */
1da177e4 1814 case MPOL_INTERLEAVE:
dfcd3c0d 1815 return nodes_equal(a->v.nodes, b->v.nodes);
1da177e4 1816 case MPOL_PREFERRED:
fc36b8d3
LS
1817 return a->v.preferred_node == b->v.preferred_node &&
1818 a->flags == b->flags;
1da177e4
LT
1819 default:
1820 BUG();
1821 return 0;
1822 }
1823}
1824
1da177e4
LT
1825/*
1826 * Shared memory backing store policy support.
1827 *
1828 * Remember policies even when nobody has shared memory mapped.
1829 * The policies are kept in Red-Black tree linked from the inode.
1830 * They are protected by the sp->lock spinlock, which should be held
1831 * for any accesses to the tree.
1832 */
1833
1834/* lookup first element intersecting start-end */
1835/* Caller holds sp->lock */
1836static struct sp_node *
1837sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
1838{
1839 struct rb_node *n = sp->root.rb_node;
1840
1841 while (n) {
1842 struct sp_node *p = rb_entry(n, struct sp_node, nd);
1843
1844 if (start >= p->end)
1845 n = n->rb_right;
1846 else if (end <= p->start)
1847 n = n->rb_left;
1848 else
1849 break;
1850 }
1851 if (!n)
1852 return NULL;
1853 for (;;) {
1854 struct sp_node *w = NULL;
1855 struct rb_node *prev = rb_prev(n);
1856 if (!prev)
1857 break;
1858 w = rb_entry(prev, struct sp_node, nd);
1859 if (w->end <= start)
1860 break;
1861 n = prev;
1862 }
1863 return rb_entry(n, struct sp_node, nd);
1864}
1865
1866/* Insert a new shared policy into the list. */
1867/* Caller holds sp->lock */
1868static void sp_insert(struct shared_policy *sp, struct sp_node *new)
1869{
1870 struct rb_node **p = &sp->root.rb_node;
1871 struct rb_node *parent = NULL;
1872 struct sp_node *nd;
1873
1874 while (*p) {
1875 parent = *p;
1876 nd = rb_entry(parent, struct sp_node, nd);
1877 if (new->start < nd->start)
1878 p = &(*p)->rb_left;
1879 else if (new->end > nd->end)
1880 p = &(*p)->rb_right;
1881 else
1882 BUG();
1883 }
1884 rb_link_node(&new->nd, parent, p);
1885 rb_insert_color(&new->nd, &sp->root);
140d5a49 1886 pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
45c4745a 1887 new->policy ? new->policy->mode : 0);
1da177e4
LT
1888}
1889
1890/* Find shared policy intersecting idx */
1891struct mempolicy *
1892mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
1893{
1894 struct mempolicy *pol = NULL;
1895 struct sp_node *sn;
1896
1897 if (!sp->root.rb_node)
1898 return NULL;
1899 spin_lock(&sp->lock);
1900 sn = sp_lookup(sp, idx, idx+1);
1901 if (sn) {
1902 mpol_get(sn->policy);
1903 pol = sn->policy;
1904 }
1905 spin_unlock(&sp->lock);
1906 return pol;
1907}
1908
1909static void sp_delete(struct shared_policy *sp, struct sp_node *n)
1910{
140d5a49 1911 pr_debug("deleting %lx-l%lx\n", n->start, n->end);
1da177e4 1912 rb_erase(&n->nd, &sp->root);
f0be3d32 1913 mpol_put(n->policy);
1da177e4
LT
1914 kmem_cache_free(sn_cache, n);
1915}
1916
dbcb0f19
AB
1917static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
1918 struct mempolicy *pol)
1da177e4
LT
1919{
1920 struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
1921
1922 if (!n)
1923 return NULL;
1924 n->start = start;
1925 n->end = end;
1926 mpol_get(pol);
aab0b102 1927 pol->flags |= MPOL_F_SHARED; /* for unref */
1da177e4
LT
1928 n->policy = pol;
1929 return n;
1930}
1931
1932/* Replace a policy range. */
1933static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
1934 unsigned long end, struct sp_node *new)
1935{
1936 struct sp_node *n, *new2 = NULL;
1937
1938restart:
1939 spin_lock(&sp->lock);
1940 n = sp_lookup(sp, start, end);
1941 /* Take care of old policies in the same range. */
1942 while (n && n->start < end) {
1943 struct rb_node *next = rb_next(&n->nd);
1944 if (n->start >= start) {
1945 if (n->end <= end)
1946 sp_delete(sp, n);
1947 else
1948 n->start = end;
1949 } else {
1950 /* Old policy spanning whole new range. */
1951 if (n->end > end) {
1952 if (!new2) {
1953 spin_unlock(&sp->lock);
1954 new2 = sp_alloc(end, n->end, n->policy);
1955 if (!new2)
1956 return -ENOMEM;
1957 goto restart;
1958 }
1959 n->end = start;
1960 sp_insert(sp, new2);
1961 new2 = NULL;
1962 break;
1963 } else
1964 n->end = start;
1965 }
1966 if (!next)
1967 break;
1968 n = rb_entry(next, struct sp_node, nd);
1969 }
1970 if (new)
1971 sp_insert(sp, new);
1972 spin_unlock(&sp->lock);
1973 if (new2) {
f0be3d32 1974 mpol_put(new2->policy);
1da177e4
LT
1975 kmem_cache_free(sn_cache, new2);
1976 }
1977 return 0;
1978}
1979
71fe804b
LS
1980/**
1981 * mpol_shared_policy_init - initialize shared policy for inode
1982 * @sp: pointer to inode shared policy
1983 * @mpol: struct mempolicy to install
1984 *
1985 * Install non-NULL @mpol in inode's shared policy rb-tree.
1986 * On entry, the current task has a reference on a non-NULL @mpol.
1987 * This must be released on exit.
4bfc4495 1988 * This is called at get_inode() calls and we can use GFP_KERNEL.
71fe804b
LS
1989 */
1990void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
1991{
58568d2a
MX
1992 int ret;
1993
71fe804b
LS
1994 sp->root = RB_ROOT; /* empty tree == default mempolicy */
1995 spin_lock_init(&sp->lock);
1996
1997 if (mpol) {
1998 struct vm_area_struct pvma;
1999 struct mempolicy *new;
4bfc4495 2000 NODEMASK_SCRATCH(scratch);
71fe804b 2001
4bfc4495
KH
2002 if (!scratch)
2003 return;
71fe804b
LS
2004 /* contextualize the tmpfs mount point mempolicy */
2005 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
58568d2a
MX
2006 if (IS_ERR(new)) {
2007 mpol_put(mpol); /* drop our ref on sb mpol */
4bfc4495 2008 NODEMASK_SCRATCH_FREE(scratch);
71fe804b 2009 return; /* no valid nodemask intersection */
58568d2a
MX
2010 }
2011
2012 task_lock(current);
4bfc4495 2013 ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
58568d2a
MX
2014 task_unlock(current);
2015 mpol_put(mpol); /* drop our ref on sb mpol */
2016 if (ret) {
4bfc4495 2017 NODEMASK_SCRATCH_FREE(scratch);
58568d2a
MX
2018 mpol_put(new);
2019 return;
2020 }
71fe804b
LS
2021
2022 /* Create pseudo-vma that contains just the policy */
2023 memset(&pvma, 0, sizeof(struct vm_area_struct));
2024 pvma.vm_end = TASK_SIZE; /* policy covers entire file */
2025 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
2026 mpol_put(new); /* drop initial ref */
4bfc4495 2027 NODEMASK_SCRATCH_FREE(scratch);
7339ff83
RH
2028 }
2029}
2030
1da177e4
LT
2031int mpol_set_shared_policy(struct shared_policy *info,
2032 struct vm_area_struct *vma, struct mempolicy *npol)
2033{
2034 int err;
2035 struct sp_node *new = NULL;
2036 unsigned long sz = vma_pages(vma);
2037
028fec41 2038 pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
1da177e4 2039 vma->vm_pgoff,
45c4745a 2040 sz, npol ? npol->mode : -1,
028fec41 2041 npol ? npol->flags : -1,
140d5a49 2042 npol ? nodes_addr(npol->v.nodes)[0] : -1);
1da177e4
LT
2043
2044 if (npol) {
2045 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
2046 if (!new)
2047 return -ENOMEM;
2048 }
2049 err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
2050 if (err && new)
2051 kmem_cache_free(sn_cache, new);
2052 return err;
2053}
2054
2055/* Free a backing policy store on inode delete. */
2056void mpol_free_shared_policy(struct shared_policy *p)
2057{
2058 struct sp_node *n;
2059 struct rb_node *next;
2060
2061 if (!p->root.rb_node)
2062 return;
2063 spin_lock(&p->lock);
2064 next = rb_first(&p->root);
2065 while (next) {
2066 n = rb_entry(next, struct sp_node, nd);
2067 next = rb_next(&n->nd);
90c5029e 2068 rb_erase(&n->nd, &p->root);
f0be3d32 2069 mpol_put(n->policy);
1da177e4
LT
2070 kmem_cache_free(sn_cache, n);
2071 }
2072 spin_unlock(&p->lock);
1da177e4
LT
2073}
2074
2075/* assumes fs == KERNEL_DS */
2076void __init numa_policy_init(void)
2077{
b71636e2
PM
2078 nodemask_t interleave_nodes;
2079 unsigned long largest = 0;
2080 int nid, prefer = 0;
2081
1da177e4
LT
2082 policy_cache = kmem_cache_create("numa_policy",
2083 sizeof(struct mempolicy),
20c2df83 2084 0, SLAB_PANIC, NULL);
1da177e4
LT
2085
2086 sn_cache = kmem_cache_create("shared_policy_node",
2087 sizeof(struct sp_node),
20c2df83 2088 0, SLAB_PANIC, NULL);
1da177e4 2089
b71636e2
PM
2090 /*
2091 * Set interleaving policy for system init. Interleaving is only
2092 * enabled across suitably sized nodes (default is >= 16MB), or
2093 * fall back to the largest node if they're all smaller.
2094 */
2095 nodes_clear(interleave_nodes);
56bbd65d 2096 for_each_node_state(nid, N_HIGH_MEMORY) {
b71636e2
PM
2097 unsigned long total_pages = node_present_pages(nid);
2098
2099 /* Preserve the largest node */
2100 if (largest < total_pages) {
2101 largest = total_pages;
2102 prefer = nid;
2103 }
2104
2105 /* Interleave this node? */
2106 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
2107 node_set(nid, interleave_nodes);
2108 }
2109
2110 /* All too small, use the largest */
2111 if (unlikely(nodes_empty(interleave_nodes)))
2112 node_set(prefer, interleave_nodes);
1da177e4 2113
028fec41 2114 if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
1da177e4
LT
2115 printk("numa_policy_init: interleaving failed\n");
2116}
2117
8bccd85f 2118/* Reset policy of current process to default */
1da177e4
LT
2119void numa_default_policy(void)
2120{
028fec41 2121 do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
1da177e4 2122}
68860ec1 2123
095f1fc4
LS
2124/*
2125 * Parse and format mempolicy from/to strings
2126 */
2127
1a75a6c8 2128/*
fc36b8d3 2129 * "local" is pseudo-policy: MPOL_PREFERRED with MPOL_F_LOCAL flag
3f226aa1 2130 * Used only for mpol_parse_str() and mpol_to_str()
1a75a6c8 2131 */
53f2556b 2132#define MPOL_LOCAL (MPOL_INTERLEAVE + 1)
15ad7cdc 2133static const char * const policy_types[] =
53f2556b 2134 { "default", "prefer", "bind", "interleave", "local" };
1a75a6c8 2135
095f1fc4
LS
2136
2137#ifdef CONFIG_TMPFS
2138/**
2139 * mpol_parse_str - parse string to mempolicy
2140 * @str: string containing mempolicy to parse
71fe804b
LS
2141 * @mpol: pointer to struct mempolicy pointer, returned on success.
2142 * @no_context: flag whether to "contextualize" the mempolicy
095f1fc4
LS
2143 *
2144 * Format of input:
2145 * <mode>[=<flags>][:<nodelist>]
2146 *
71fe804b
LS
2147 * if @no_context is true, save the input nodemask in w.user_nodemask in
2148 * the returned mempolicy. This will be used to "clone" the mempolicy in
2149 * a specific context [cpuset] at a later time. Used to parse tmpfs mpol
2150 * mount option. Note that if 'static' or 'relative' mode flags were
2151 * specified, the input nodemask will already have been saved. Saving
2152 * it again is redundant, but safe.
2153 *
2154 * On success, returns 0, else 1
095f1fc4 2155 */
71fe804b 2156int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context)
095f1fc4 2157{
71fe804b
LS
2158 struct mempolicy *new = NULL;
2159 unsigned short uninitialized_var(mode);
2160 unsigned short uninitialized_var(mode_flags);
2161 nodemask_t nodes;
095f1fc4
LS
2162 char *nodelist = strchr(str, ':');
2163 char *flags = strchr(str, '=');
2164 int i;
2165 int err = 1;
2166
2167 if (nodelist) {
2168 /* NUL-terminate mode or flags string */
2169 *nodelist++ = '\0';
71fe804b 2170 if (nodelist_parse(nodelist, nodes))
095f1fc4 2171 goto out;
71fe804b 2172 if (!nodes_subset(nodes, node_states[N_HIGH_MEMORY]))
095f1fc4 2173 goto out;
71fe804b
LS
2174 } else
2175 nodes_clear(nodes);
2176
095f1fc4
LS
2177 if (flags)
2178 *flags++ = '\0'; /* terminate mode string */
2179
3f226aa1 2180 for (i = 0; i <= MPOL_LOCAL; i++) {
095f1fc4 2181 if (!strcmp(str, policy_types[i])) {
71fe804b 2182 mode = i;
095f1fc4
LS
2183 break;
2184 }
2185 }
3f226aa1 2186 if (i > MPOL_LOCAL)
095f1fc4
LS
2187 goto out;
2188
71fe804b 2189 switch (mode) {
095f1fc4 2190 case MPOL_PREFERRED:
71fe804b
LS
2191 /*
2192 * Insist on a nodelist of one node only
2193 */
095f1fc4
LS
2194 if (nodelist) {
2195 char *rest = nodelist;
2196 while (isdigit(*rest))
2197 rest++;
2198 if (!*rest)
2199 err = 0;
2200 }
2201 break;
095f1fc4
LS
2202 case MPOL_INTERLEAVE:
2203 /*
2204 * Default to online nodes with memory if no nodelist
2205 */
2206 if (!nodelist)
71fe804b 2207 nodes = node_states[N_HIGH_MEMORY];
095f1fc4 2208 err = 0;
3f226aa1 2209 break;
71fe804b 2210 case MPOL_LOCAL:
3f226aa1 2211 /*
71fe804b 2212 * Don't allow a nodelist; mpol_new() checks flags
3f226aa1 2213 */
71fe804b 2214 if (nodelist)
3f226aa1 2215 goto out;
71fe804b 2216 mode = MPOL_PREFERRED;
3f226aa1 2217 break;
413b43de
RT
2218 case MPOL_DEFAULT:
2219 /*
2220 * Insist on a empty nodelist
2221 */
2222 if (!nodelist)
2223 err = 0;
2224 goto out;
71fe804b
LS
2225 /*
2226 * case MPOL_BIND: mpol_new() enforces non-empty nodemask.
71fe804b 2227 */
095f1fc4
LS
2228 }
2229
71fe804b 2230 mode_flags = 0;
095f1fc4
LS
2231 if (flags) {
2232 /*
2233 * Currently, we only support two mutually exclusive
2234 * mode flags.
2235 */
2236 if (!strcmp(flags, "static"))
71fe804b 2237 mode_flags |= MPOL_F_STATIC_NODES;
095f1fc4 2238 else if (!strcmp(flags, "relative"))
71fe804b 2239 mode_flags |= MPOL_F_RELATIVE_NODES;
095f1fc4
LS
2240 else
2241 err = 1;
2242 }
71fe804b
LS
2243
2244 new = mpol_new(mode, mode_flags, &nodes);
2245 if (IS_ERR(new))
2246 err = 1;
58568d2a
MX
2247 else {
2248 int ret;
4bfc4495
KH
2249 NODEMASK_SCRATCH(scratch);
2250 if (scratch) {
2251 task_lock(current);
2252 ret = mpol_set_nodemask(new, &nodes, scratch);
2253 task_unlock(current);
2254 } else
2255 ret = -ENOMEM;
2256 NODEMASK_SCRATCH_FREE(scratch);
2257 if (ret) {
58568d2a 2258 err = 1;
4bfc4495
KH
2259 mpol_put(new);
2260 } else if (no_context) {
58568d2a
MX
2261 /* save for contextualization */
2262 new->w.user_nodemask = nodes;
2263 }
2264 }
71fe804b 2265
095f1fc4
LS
2266out:
2267 /* Restore string for error message */
2268 if (nodelist)
2269 *--nodelist = ':';
2270 if (flags)
2271 *--flags = '=';
71fe804b
LS
2272 if (!err)
2273 *mpol = new;
095f1fc4
LS
2274 return err;
2275}
2276#endif /* CONFIG_TMPFS */
2277
71fe804b
LS
2278/**
2279 * mpol_to_str - format a mempolicy structure for printing
2280 * @buffer: to contain formatted mempolicy string
2281 * @maxlen: length of @buffer
2282 * @pol: pointer to mempolicy to be formatted
2283 * @no_context: "context free" mempolicy - use nodemask in w.user_nodemask
2284 *
1a75a6c8
CL
2285 * Convert a mempolicy into a string.
2286 * Returns the number of characters in buffer (if positive)
2287 * or an error (negative)
2288 */
71fe804b 2289int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol, int no_context)
1a75a6c8
CL
2290{
2291 char *p = buffer;
2292 int l;
2293 nodemask_t nodes;
bea904d5 2294 unsigned short mode;
f5b087b5 2295 unsigned short flags = pol ? pol->flags : 0;
1a75a6c8 2296
2291990a
LS
2297 /*
2298 * Sanity check: room for longest mode, flag and some nodes
2299 */
2300 VM_BUG_ON(maxlen < strlen("interleave") + strlen("relative") + 16);
2301
bea904d5
LS
2302 if (!pol || pol == &default_policy)
2303 mode = MPOL_DEFAULT;
2304 else
2305 mode = pol->mode;
2306
1a75a6c8
CL
2307 switch (mode) {
2308 case MPOL_DEFAULT:
2309 nodes_clear(nodes);
2310 break;
2311
2312 case MPOL_PREFERRED:
2313 nodes_clear(nodes);
fc36b8d3 2314 if (flags & MPOL_F_LOCAL)
53f2556b
LS
2315 mode = MPOL_LOCAL; /* pseudo-policy */
2316 else
fc36b8d3 2317 node_set(pol->v.preferred_node, nodes);
1a75a6c8
CL
2318 break;
2319
2320 case MPOL_BIND:
19770b32 2321 /* Fall through */
1a75a6c8 2322 case MPOL_INTERLEAVE:
71fe804b
LS
2323 if (no_context)
2324 nodes = pol->w.user_nodemask;
2325 else
2326 nodes = pol->v.nodes;
1a75a6c8
CL
2327 break;
2328
2329 default:
2330 BUG();
1a75a6c8
CL
2331 }
2332
2333 l = strlen(policy_types[mode]);
53f2556b
LS
2334 if (buffer + maxlen < p + l + 1)
2335 return -ENOSPC;
1a75a6c8
CL
2336
2337 strcpy(p, policy_types[mode]);
2338 p += l;
2339
fc36b8d3 2340 if (flags & MPOL_MODE_FLAGS) {
f5b087b5
DR
2341 if (buffer + maxlen < p + 2)
2342 return -ENOSPC;
2343 *p++ = '=';
2344
2291990a
LS
2345 /*
2346 * Currently, the only defined flags are mutually exclusive
2347 */
f5b087b5 2348 if (flags & MPOL_F_STATIC_NODES)
2291990a
LS
2349 p += snprintf(p, buffer + maxlen - p, "static");
2350 else if (flags & MPOL_F_RELATIVE_NODES)
2351 p += snprintf(p, buffer + maxlen - p, "relative");
f5b087b5
DR
2352 }
2353
1a75a6c8
CL
2354 if (!nodes_empty(nodes)) {
2355 if (buffer + maxlen < p + 2)
2356 return -ENOSPC;
095f1fc4 2357 *p++ = ':';
1a75a6c8
CL
2358 p += nodelist_scnprintf(p, buffer + maxlen - p, nodes);
2359 }
2360 return p - buffer;
2361}
2362
2363struct numa_maps {
2364 unsigned long pages;
2365 unsigned long anon;
397874df
CL
2366 unsigned long active;
2367 unsigned long writeback;
1a75a6c8 2368 unsigned long mapcount_max;
397874df
CL
2369 unsigned long dirty;
2370 unsigned long swapcache;
1a75a6c8
CL
2371 unsigned long node[MAX_NUMNODES];
2372};
2373
397874df 2374static void gather_stats(struct page *page, void *private, int pte_dirty)
1a75a6c8
CL
2375{
2376 struct numa_maps *md = private;
2377 int count = page_mapcount(page);
2378
397874df
CL
2379 md->pages++;
2380 if (pte_dirty || PageDirty(page))
2381 md->dirty++;
1a75a6c8 2382
397874df
CL
2383 if (PageSwapCache(page))
2384 md->swapcache++;
1a75a6c8 2385
894bc310 2386 if (PageActive(page) || PageUnevictable(page))
397874df
CL
2387 md->active++;
2388
2389 if (PageWriteback(page))
2390 md->writeback++;
1a75a6c8
CL
2391
2392 if (PageAnon(page))
2393 md->anon++;
2394
397874df
CL
2395 if (count > md->mapcount_max)
2396 md->mapcount_max = count;
2397
1a75a6c8 2398 md->node[page_to_nid(page)]++;
1a75a6c8
CL
2399}
2400
7f709ed0 2401#ifdef CONFIG_HUGETLB_PAGE
397874df
CL
2402static void check_huge_range(struct vm_area_struct *vma,
2403 unsigned long start, unsigned long end,
2404 struct numa_maps *md)
2405{
2406 unsigned long addr;
2407 struct page *page;
a5516438
AK
2408 struct hstate *h = hstate_vma(vma);
2409 unsigned long sz = huge_page_size(h);
397874df 2410
a5516438
AK
2411 for (addr = start; addr < end; addr += sz) {
2412 pte_t *ptep = huge_pte_offset(vma->vm_mm,
2413 addr & huge_page_mask(h));
397874df
CL
2414 pte_t pte;
2415
2416 if (!ptep)
2417 continue;
2418
2419 pte = *ptep;
2420 if (pte_none(pte))
2421 continue;
2422
2423 page = pte_page(pte);
2424 if (!page)
2425 continue;
2426
2427 gather_stats(page, md, pte_dirty(*ptep));
2428 }
2429}
7f709ed0
AM
2430#else
2431static inline void check_huge_range(struct vm_area_struct *vma,
2432 unsigned long start, unsigned long end,
2433 struct numa_maps *md)
2434{
2435}
2436#endif
397874df 2437
53f2556b
LS
2438/*
2439 * Display pages allocated per node and memory policy via /proc.
2440 */
1a75a6c8
CL
2441int show_numa_map(struct seq_file *m, void *v)
2442{
99f89551 2443 struct proc_maps_private *priv = m->private;
1a75a6c8
CL
2444 struct vm_area_struct *vma = v;
2445 struct numa_maps *md;
397874df
CL
2446 struct file *file = vma->vm_file;
2447 struct mm_struct *mm = vma->vm_mm;
480eccf9 2448 struct mempolicy *pol;
1a75a6c8
CL
2449 int n;
2450 char buffer[50];
2451
397874df 2452 if (!mm)
1a75a6c8
CL
2453 return 0;
2454
2455 md = kzalloc(sizeof(struct numa_maps), GFP_KERNEL);
2456 if (!md)
2457 return 0;
2458
480eccf9 2459 pol = get_vma_policy(priv->task, vma, vma->vm_start);
71fe804b 2460 mpol_to_str(buffer, sizeof(buffer), pol, 0);
52cd3b07 2461 mpol_cond_put(pol);
397874df
CL
2462
2463 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
2464
2465 if (file) {
2466 seq_printf(m, " file=");
c32c2f63 2467 seq_path(m, &file->f_path, "\n\t= ");
397874df
CL
2468 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
2469 seq_printf(m, " heap");
2470 } else if (vma->vm_start <= mm->start_stack &&
2471 vma->vm_end >= mm->start_stack) {
2472 seq_printf(m, " stack");
2473 }
2474
2475 if (is_vm_hugetlb_page(vma)) {
2476 check_huge_range(vma, vma->vm_start, vma->vm_end, md);
2477 seq_printf(m, " huge");
2478 } else {
a57ebfdb 2479 check_pgd_range(vma, vma->vm_start, vma->vm_end,
56bbd65d 2480 &node_states[N_HIGH_MEMORY], MPOL_MF_STATS, md);
397874df
CL
2481 }
2482
2483 if (!md->pages)
2484 goto out;
1a75a6c8 2485
397874df
CL
2486 if (md->anon)
2487 seq_printf(m," anon=%lu",md->anon);
1a75a6c8 2488
397874df
CL
2489 if (md->dirty)
2490 seq_printf(m," dirty=%lu",md->dirty);
1a75a6c8 2491
397874df
CL
2492 if (md->pages != md->anon && md->pages != md->dirty)
2493 seq_printf(m, " mapped=%lu", md->pages);
1a75a6c8 2494
397874df
CL
2495 if (md->mapcount_max > 1)
2496 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1a75a6c8 2497
397874df
CL
2498 if (md->swapcache)
2499 seq_printf(m," swapcache=%lu", md->swapcache);
2500
2501 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
2502 seq_printf(m," active=%lu", md->active);
2503
2504 if (md->writeback)
2505 seq_printf(m," writeback=%lu", md->writeback);
2506
56bbd65d 2507 for_each_node_state(n, N_HIGH_MEMORY)
397874df
CL
2508 if (md->node[n])
2509 seq_printf(m, " N%d=%lu", n, md->node[n]);
2510out:
2511 seq_putc(m, '\n');
1a75a6c8
CL
2512 kfree(md);
2513
2514 if (m->count < m->size)
99f89551 2515 m->version = (vma != priv->tail_vma) ? vma->vm_start : 0;
1a75a6c8
CL
2516 return 0;
2517}