1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Simple NUMA memory policy for the Linux kernel.
4 *
5 * Copyright 2003,2004 Andi Kleen, SuSE Labs.
6 * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc.
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.
21 *
22 * weighted interleave
23 * Allocate memory interleaved over a set of nodes based on
24 * a set of weights (per-node), with normal fallback if it
25 * fails. Otherwise operates the same as interleave.
26 * Example: nodeset(0,1) & weights (2,1) - 2 pages allocated
27 * on node 0 for every 1 page allocated on node 1.
28 *
29 * bind Only allocate memory on a specific set of nodes,
30 * no fallback.
31 * FIXME: memory is allocated starting with the first node
32 * to the last. It would be better if bind would truly restrict
33 * the allocation to memory nodes instead
34 *
35 * preferred Try a specific node first before normal fallback.
36 * As a special case NUMA_NO_NODE here means do the allocation
37 * on the local CPU. This is normally identical to default,
38 * but useful to set in a VMA when you have a non default
39 * process policy.
40 *
41 * preferred many Try a set of nodes first before normal fallback. This is
42 * similar to preferred without the special case.
43 *
44 * default Allocate on the local node first, or when on a VMA
45 * use the process policy. This is what Linux always did
46 * in a NUMA aware kernel and still does by, ahem, default.
47 *
48 * The process policy is applied for most non interrupt memory allocations
49 * in that process' context. Interrupts ignore the policies and always
50 * try to allocate on the local CPU. The VMA policy is only applied for memory
51 * allocations for a VMA in the VM.
52 *
53 * Currently there are a few corner cases in swapping where the policy
54 * is not applied, but the majority should be handled. When process policy
55 * is used it is not remembered over swap outs/swap ins.
56 *
57 * Only the highest zone in the zone hierarchy gets policied. Allocations
58 * requesting a lower zone just use default policy. This implies that
59 * on systems with highmem kernel lowmem allocation don't get policied.
60 * Same with GFP_DMA allocations.
61 *
62 * For shmem/tmpfs shared memory the policy is shared between
63 * all users and remembered even when nobody has memory mapped.
64 */
65
66 /* Notebook:
67 fix mmap readahead to honour policy and enable policy for any page cache
68 object
69 statistics for bigpages
70 global policy for page cache? currently it uses process policy. Requires
71 first item above.
72 handle mremap for shared memory (currently ignored for the policy)
73 grows down?
74 make bind policy root only? It can trigger oom much faster and the
75 kernel is not always grateful with that.
76 */
77
78 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
79
80 #include <linux/mempolicy.h>
81 #include <linux/pagewalk.h>
82 #include <linux/highmem.h>
83 #include <linux/hugetlb.h>
84 #include <linux/kernel.h>
85 #include <linux/sched.h>
86 #include <linux/sched/mm.h>
87 #include <linux/sched/numa_balancing.h>
88 #include <linux/sched/task.h>
89 #include <linux/nodemask.h>
90 #include <linux/cpuset.h>
91 #include <linux/slab.h>
92 #include <linux/string.h>
93 #include <linux/export.h>
94 #include <linux/nsproxy.h>
95 #include <linux/interrupt.h>
96 #include <linux/init.h>
97 #include <linux/compat.h>
98 #include <linux/ptrace.h>
99 #include <linux/swap.h>
100 #include <linux/seq_file.h>
101 #include <linux/proc_fs.h>
102 #include <linux/migrate.h>
103 #include <linux/ksm.h>
104 #include <linux/rmap.h>
105 #include <linux/security.h>
106 #include <linux/syscalls.h>
107 #include <linux/ctype.h>
108 #include <linux/mm_inline.h>
109 #include <linux/mmu_notifier.h>
110 #include <linux/printk.h>
111 #include <linux/swapops.h>
112
113 #include <asm/tlbflush.h>
114 #include <asm/tlb.h>
115 #include <linux/uaccess.h>
116
117 #include "internal.h"
118
119 /* Internal flags */
120 #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */
121 #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */
122 #define MPOL_MF_WRLOCK (MPOL_MF_INTERNAL << 2) /* Write-lock walked vmas */
123
124 static struct kmem_cache *policy_cache;
125 static struct kmem_cache *sn_cache;
126
127 /* Highest zone. An specific allocation for a zone below that is not
128 policied. */
129 enum zone_type policy_zone = 0;
130
131 /*
132 * run-time system-wide default policy => local allocation
133 */
134 static struct mempolicy default_policy = {
135 .refcnt = ATOMIC_INIT(1), /* never free it */
136 .mode = MPOL_LOCAL,
137 };
138
139 static struct mempolicy preferred_node_policy[MAX_NUMNODES];
140
141 /*
142 * iw_table is the sysfs-set interleave weight table, a value of 0 denotes
143 * system-default value should be used. A NULL iw_table also denotes that
144 * system-default values should be used. Until the system-default table
145 * is implemented, the system-default is always 1.
146 *
147 * iw_table is RCU protected
148 */
149 static u8 __rcu *iw_table;
150 static DEFINE_MUTEX(iw_table_lock);
151
get_il_weight(int node)152 static u8 get_il_weight(int node)
153 {
154 u8 *table;
155 u8 weight;
156
157 rcu_read_lock();
158 table = rcu_dereference(iw_table);
159 /* if no iw_table, use system default */
160 weight = table ? table[node] : 1;
161 /* if value in iw_table is 0, use system default */
162 weight = weight ? weight : 1;
163 rcu_read_unlock();
164 return weight;
165 }
166
167 /**
168 * numa_nearest_node - Find nearest node by state
169 * @node: Node id to start the search
170 * @state: State to filter the search
171 *
172 * Lookup the closest node by distance if @nid is not in state.
173 *
174 * Return: this @node if it is in state, otherwise the closest node by distance
175 */
numa_nearest_node(int node,unsigned int state)176 int numa_nearest_node(int node, unsigned int state)
177 {
178 int min_dist = INT_MAX, dist, n, min_node;
179
180 if (state >= NR_NODE_STATES)
181 return -EINVAL;
182
183 if (node == NUMA_NO_NODE || node_state(node, state))
184 return node;
185
186 min_node = node;
187 for_each_node_state(n, state) {
188 dist = node_distance(node, n);
189 if (dist < min_dist) {
190 min_dist = dist;
191 min_node = n;
192 }
193 }
194
195 return min_node;
196 }
197 EXPORT_SYMBOL_GPL(numa_nearest_node);
198
get_task_policy(struct task_struct * p)199 struct mempolicy *get_task_policy(struct task_struct *p)
200 {
201 struct mempolicy *pol = p->mempolicy;
202 int node;
203
204 if (pol)
205 return pol;
206
207 node = numa_node_id();
208 if (node != NUMA_NO_NODE) {
209 pol = &preferred_node_policy[node];
210 /* preferred_node_policy is not initialised early in boot */
211 if (pol->mode)
212 return pol;
213 }
214
215 return &default_policy;
216 }
217
218 static const struct mempolicy_operations {
219 int (*create)(struct mempolicy *pol, const nodemask_t *nodes);
220 void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes);
221 } mpol_ops[MPOL_MAX];
222
mpol_store_user_nodemask(const struct mempolicy * pol)223 static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
224 {
225 return pol->flags & MPOL_MODE_FLAGS;
226 }
227
mpol_relative_nodemask(nodemask_t * ret,const nodemask_t * orig,const nodemask_t * rel)228 static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig,
229 const nodemask_t *rel)
230 {
231 nodemask_t tmp;
232 nodes_fold(tmp, *orig, nodes_weight(*rel));
233 nodes_onto(*ret, tmp, *rel);
234 }
235
mpol_new_nodemask(struct mempolicy * pol,const nodemask_t * nodes)236 static int mpol_new_nodemask(struct mempolicy *pol, const nodemask_t *nodes)
237 {
238 if (nodes_empty(*nodes))
239 return -EINVAL;
240 pol->nodes = *nodes;
241 return 0;
242 }
243
mpol_new_preferred(struct mempolicy * pol,const nodemask_t * nodes)244 static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes)
245 {
246 if (nodes_empty(*nodes))
247 return -EINVAL;
248
249 nodes_clear(pol->nodes);
250 node_set(first_node(*nodes), pol->nodes);
251 return 0;
252 }
253
254 /*
255 * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if
256 * any, for the new policy. mpol_new() has already validated the nodes
257 * parameter with respect to the policy mode and flags.
258 *
259 * Must be called holding task's alloc_lock to protect task's mems_allowed
260 * and mempolicy. May also be called holding the mmap_lock for write.
261 */
mpol_set_nodemask(struct mempolicy * pol,const nodemask_t * nodes,struct nodemask_scratch * nsc)262 static int mpol_set_nodemask(struct mempolicy *pol,
263 const nodemask_t *nodes, struct nodemask_scratch *nsc)
264 {
265 int ret;
266
267 /*
268 * Default (pol==NULL) resp. local memory policies are not a
269 * subject of any remapping. They also do not need any special
270 * constructor.
271 */
272 if (!pol || pol->mode == MPOL_LOCAL)
273 return 0;
274
275 /* Check N_MEMORY */
276 nodes_and(nsc->mask1,
277 cpuset_current_mems_allowed, node_states[N_MEMORY]);
278
279 VM_BUG_ON(!nodes);
280
281 if (pol->flags & MPOL_F_RELATIVE_NODES)
282 mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1);
283 else
284 nodes_and(nsc->mask2, *nodes, nsc->mask1);
285
286 if (mpol_store_user_nodemask(pol))
287 pol->w.user_nodemask = *nodes;
288 else
289 pol->w.cpuset_mems_allowed = cpuset_current_mems_allowed;
290
291 ret = mpol_ops[pol->mode].create(pol, &nsc->mask2);
292 return ret;
293 }
294
295 /*
296 * This function just creates a new policy, does some check and simple
297 * initialization. You must invoke mpol_set_nodemask() to set nodes.
298 */
mpol_new(unsigned short mode,unsigned short flags,nodemask_t * nodes)299 static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
300 nodemask_t *nodes)
301 {
302 struct mempolicy *policy;
303
304 if (mode == MPOL_DEFAULT) {
305 if (nodes && !nodes_empty(*nodes))
306 return ERR_PTR(-EINVAL);
307 return NULL;
308 }
309 VM_BUG_ON(!nodes);
310
311 /*
312 * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
313 * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
314 * All other modes require a valid pointer to a non-empty nodemask.
315 */
316 if (mode == MPOL_PREFERRED) {
317 if (nodes_empty(*nodes)) {
318 if (((flags & MPOL_F_STATIC_NODES) ||
319 (flags & MPOL_F_RELATIVE_NODES)))
320 return ERR_PTR(-EINVAL);
321
322 mode = MPOL_LOCAL;
323 }
324 } else if (mode == MPOL_LOCAL) {
325 if (!nodes_empty(*nodes) ||
326 (flags & MPOL_F_STATIC_NODES) ||
327 (flags & MPOL_F_RELATIVE_NODES))
328 return ERR_PTR(-EINVAL);
329 } else if (nodes_empty(*nodes))
330 return ERR_PTR(-EINVAL);
331
332 policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
333 if (!policy)
334 return ERR_PTR(-ENOMEM);
335 atomic_set(&policy->refcnt, 1);
336 policy->mode = mode;
337 policy->flags = flags;
338 policy->home_node = NUMA_NO_NODE;
339
340 return policy;
341 }
342
343 /* Slow path of a mpol destructor. */
__mpol_put(struct mempolicy * pol)344 void __mpol_put(struct mempolicy *pol)
345 {
346 if (!atomic_dec_and_test(&pol->refcnt))
347 return;
348 kmem_cache_free(policy_cache, pol);
349 }
350
mpol_rebind_default(struct mempolicy * pol,const nodemask_t * nodes)351 static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes)
352 {
353 }
354
mpol_rebind_nodemask(struct mempolicy * pol,const nodemask_t * nodes)355 static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes)
356 {
357 nodemask_t tmp;
358
359 if (pol->flags & MPOL_F_STATIC_NODES)
360 nodes_and(tmp, pol->w.user_nodemask, *nodes);
361 else if (pol->flags & MPOL_F_RELATIVE_NODES)
362 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
363 else {
364 nodes_remap(tmp, pol->nodes, pol->w.cpuset_mems_allowed,
365 *nodes);
366 pol->w.cpuset_mems_allowed = *nodes;
367 }
368
369 if (nodes_empty(tmp))
370 tmp = *nodes;
371
372 pol->nodes = tmp;
373 }
374
mpol_rebind_preferred(struct mempolicy * pol,const nodemask_t * nodes)375 static void mpol_rebind_preferred(struct mempolicy *pol,
376 const nodemask_t *nodes)
377 {
378 pol->w.cpuset_mems_allowed = *nodes;
379 }
380
381 /*
382 * mpol_rebind_policy - Migrate a policy to a different set of nodes
383 *
384 * Per-vma policies are protected by mmap_lock. Allocations using per-task
385 * policies are protected by task->mems_allowed_seq to prevent a premature
386 * OOM/allocation failure due to parallel nodemask modification.
387 */
mpol_rebind_policy(struct mempolicy * pol,const nodemask_t * newmask)388 static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask)
389 {
390 if (!pol || pol->mode == MPOL_LOCAL)
391 return;
392 if (!mpol_store_user_nodemask(pol) &&
393 nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
394 return;
395
396 mpol_ops[pol->mode].rebind(pol, newmask);
397 }
398
399 /*
400 * Wrapper for mpol_rebind_policy() that just requires task
401 * pointer, and updates task mempolicy.
402 *
403 * Called with task's alloc_lock held.
404 */
mpol_rebind_task(struct task_struct * tsk,const nodemask_t * new)405 void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new)
406 {
407 mpol_rebind_policy(tsk->mempolicy, new);
408 }
409
410 /*
411 * Rebind each vma in mm to new nodemask.
412 *
413 * Call holding a reference to mm. Takes mm->mmap_lock during call.
414 */
mpol_rebind_mm(struct mm_struct * mm,nodemask_t * new)415 void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
416 {
417 struct vm_area_struct *vma;
418 VMA_ITERATOR(vmi, mm, 0);
419
420 mmap_write_lock(mm);
421 for_each_vma(vmi, vma) {
422 vma_start_write(vma);
423 mpol_rebind_policy(vma->vm_policy, new);
424 }
425 mmap_write_unlock(mm);
426 }
427
428 static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
429 [MPOL_DEFAULT] = {
430 .rebind = mpol_rebind_default,
431 },
432 [MPOL_INTERLEAVE] = {
433 .create = mpol_new_nodemask,
434 .rebind = mpol_rebind_nodemask,
435 },
436 [MPOL_PREFERRED] = {
437 .create = mpol_new_preferred,
438 .rebind = mpol_rebind_preferred,
439 },
440 [MPOL_BIND] = {
441 .create = mpol_new_nodemask,
442 .rebind = mpol_rebind_nodemask,
443 },
444 [MPOL_LOCAL] = {
445 .rebind = mpol_rebind_default,
446 },
447 [MPOL_PREFERRED_MANY] = {
448 .create = mpol_new_nodemask,
449 .rebind = mpol_rebind_preferred,
450 },
451 [MPOL_WEIGHTED_INTERLEAVE] = {
452 .create = mpol_new_nodemask,
453 .rebind = mpol_rebind_nodemask,
454 },
455 };
456
457 static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist,
458 unsigned long flags);
459 static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *pol,
460 pgoff_t ilx, int *nid);
461
strictly_unmovable(unsigned long flags)462 static bool strictly_unmovable(unsigned long flags)
463 {
464 /*
465 * STRICT without MOVE flags lets do_mbind() fail immediately with -EIO
466 * if any misplaced page is found.
467 */
468 return (flags & (MPOL_MF_STRICT | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ==
469 MPOL_MF_STRICT;
470 }
471
472 struct migration_mpol { /* for alloc_migration_target_by_mpol() */
473 struct mempolicy *pol;
474 pgoff_t ilx;
475 };
476
477 struct queue_pages {
478 struct list_head *pagelist;
479 unsigned long flags;
480 nodemask_t *nmask;
481 unsigned long start;
482 unsigned long end;
483 struct vm_area_struct *first;
484 struct folio *large; /* note last large folio encountered */
485 long nr_failed; /* could not be isolated at this time */
486 };
487
488 /*
489 * Check if the folio's nid is in qp->nmask.
490 *
491 * If MPOL_MF_INVERT is set in qp->flags, check if the nid is
492 * in the invert of qp->nmask.
493 */
queue_folio_required(struct folio * folio,struct queue_pages * qp)494 static inline bool queue_folio_required(struct folio *folio,
495 struct queue_pages *qp)
496 {
497 int nid = folio_nid(folio);
498 unsigned long flags = qp->flags;
499
500 return node_isset(nid, *qp->nmask) == !(flags & MPOL_MF_INVERT);
501 }
502
queue_folios_pmd(pmd_t * pmd,struct mm_walk * walk)503 static void queue_folios_pmd(pmd_t *pmd, struct mm_walk *walk)
504 {
505 struct folio *folio;
506 struct queue_pages *qp = walk->private;
507
508 if (unlikely(is_pmd_migration_entry(*pmd))) {
509 qp->nr_failed++;
510 return;
511 }
512 folio = pmd_folio(*pmd);
513 if (is_huge_zero_folio(folio)) {
514 walk->action = ACTION_CONTINUE;
515 return;
516 }
517 if (!queue_folio_required(folio, qp))
518 return;
519 if (!(qp->flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ||
520 !vma_migratable(walk->vma) ||
521 !migrate_folio_add(folio, qp->pagelist, qp->flags))
522 qp->nr_failed++;
523 }
524
525 /*
526 * Scan through folios, checking if they satisfy the required conditions,
527 * moving them from LRU to local pagelist for migration if they do (or not).
528 *
529 * queue_folios_pte_range() has two possible return values:
530 * 0 - continue walking to scan for more, even if an existing folio on the
531 * wrong node could not be isolated and queued for migration.
532 * -EIO - only MPOL_MF_STRICT was specified, without MPOL_MF_MOVE or ..._ALL,
533 * and an existing folio was on a node that does not follow the policy.
534 */
queue_folios_pte_range(pmd_t * pmd,unsigned long addr,unsigned long end,struct mm_walk * walk)535 static int queue_folios_pte_range(pmd_t *pmd, unsigned long addr,
536 unsigned long end, struct mm_walk *walk)
537 {
538 struct vm_area_struct *vma = walk->vma;
539 struct folio *folio;
540 struct queue_pages *qp = walk->private;
541 unsigned long flags = qp->flags;
542 pte_t *pte, *mapped_pte;
543 pte_t ptent;
544 spinlock_t *ptl;
545
546 ptl = pmd_trans_huge_lock(pmd, vma);
547 if (ptl) {
548 queue_folios_pmd(pmd, walk);
549 spin_unlock(ptl);
550 goto out;
551 }
552
553 mapped_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
554 if (!pte) {
555 walk->action = ACTION_AGAIN;
556 return 0;
557 }
558 for (; addr != end; pte++, addr += PAGE_SIZE) {
559 ptent = ptep_get(pte);
560 if (pte_none(ptent))
561 continue;
562 if (!pte_present(ptent)) {
563 if (is_migration_entry(pte_to_swp_entry(ptent)))
564 qp->nr_failed++;
565 continue;
566 }
567 folio = vm_normal_folio(vma, addr, ptent);
568 if (!folio || folio_is_zone_device(folio))
569 continue;
570 /*
571 * vm_normal_folio() filters out zero pages, but there might
572 * still be reserved folios to skip, perhaps in a VDSO.
573 */
574 if (folio_test_reserved(folio))
575 continue;
576 if (!queue_folio_required(folio, qp))
577 continue;
578 if (folio_test_large(folio)) {
579 /*
580 * A large folio can only be isolated from LRU once,
581 * but may be mapped by many PTEs (and Copy-On-Write may
582 * intersperse PTEs of other, order 0, folios). This is
583 * a common case, so don't mistake it for failure (but
584 * there can be other cases of multi-mapped pages which
585 * this quick check does not help to filter out - and a
586 * search of the pagelist might grow to be prohibitive).
587 *
588 * migrate_pages(&pagelist) returns nr_failed folios, so
589 * check "large" now so that queue_pages_range() returns
590 * a comparable nr_failed folios. This does imply that
591 * if folio could not be isolated for some racy reason
592 * at its first PTE, later PTEs will not give it another
593 * chance of isolation; but keeps the accounting simple.
594 */
595 if (folio == qp->large)
596 continue;
597 qp->large = folio;
598 }
599 if (!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ||
600 !vma_migratable(vma) ||
601 !migrate_folio_add(folio, qp->pagelist, flags)) {
602 qp->nr_failed++;
603 if (strictly_unmovable(flags))
604 break;
605 }
606 }
607 pte_unmap_unlock(mapped_pte, ptl);
608 cond_resched();
609 out:
610 if (qp->nr_failed && strictly_unmovable(flags))
611 return -EIO;
612 return 0;
613 }
614
queue_folios_hugetlb(pte_t * pte,unsigned long hmask,unsigned long addr,unsigned long end,struct mm_walk * walk)615 static int queue_folios_hugetlb(pte_t *pte, unsigned long hmask,
616 unsigned long addr, unsigned long end,
617 struct mm_walk *walk)
618 {
619 #ifdef CONFIG_HUGETLB_PAGE
620 struct queue_pages *qp = walk->private;
621 unsigned long flags = qp->flags;
622 struct folio *folio;
623 spinlock_t *ptl;
624 pte_t entry;
625
626 ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte);
627 entry = huge_ptep_get(walk->mm, addr, pte);
628 if (!pte_present(entry)) {
629 if (unlikely(is_hugetlb_entry_migration(entry)))
630 qp->nr_failed++;
631 goto unlock;
632 }
633 folio = pfn_folio(pte_pfn(entry));
634 if (!queue_folio_required(folio, qp))
635 goto unlock;
636 if (!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) ||
637 !vma_migratable(walk->vma)) {
638 qp->nr_failed++;
639 goto unlock;
640 }
641 /*
642 * Unless MPOL_MF_MOVE_ALL, we try to avoid migrating a shared folio.
643 * Choosing not to migrate a shared folio is not counted as a failure.
644 *
645 * See folio_likely_mapped_shared() on possible imprecision when we
646 * cannot easily detect if a folio is shared.
647 */
648 if ((flags & MPOL_MF_MOVE_ALL) ||
649 (!folio_likely_mapped_shared(folio) && !hugetlb_pmd_shared(pte)))
650 if (!isolate_hugetlb(folio, qp->pagelist))
651 qp->nr_failed++;
652 unlock:
653 spin_unlock(ptl);
654 if (qp->nr_failed && strictly_unmovable(flags))
655 return -EIO;
656 #endif
657 return 0;
658 }
659
660 #ifdef CONFIG_NUMA_BALANCING
661 /*
662 * This is used to mark a range of virtual addresses to be inaccessible.
663 * These are later cleared by a NUMA hinting fault. Depending on these
664 * faults, pages may be migrated for better NUMA placement.
665 *
666 * This is assuming that NUMA faults are handled using PROT_NONE. If
667 * an architecture makes a different choice, it will need further
668 * changes to the core.
669 */
change_prot_numa(struct vm_area_struct * vma,unsigned long addr,unsigned long end)670 unsigned long change_prot_numa(struct vm_area_struct *vma,
671 unsigned long addr, unsigned long end)
672 {
673 struct mmu_gather tlb;
674 long nr_updated;
675
676 tlb_gather_mmu(&tlb, vma->vm_mm);
677
678 nr_updated = change_protection(&tlb, vma, addr, end, MM_CP_PROT_NUMA);
679 if (nr_updated > 0) {
680 count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated);
681 count_memcg_events_mm(vma->vm_mm, NUMA_PTE_UPDATES, nr_updated);
682 }
683
684 tlb_finish_mmu(&tlb);
685
686 return nr_updated;
687 }
688 #endif /* CONFIG_NUMA_BALANCING */
689
queue_pages_test_walk(unsigned long start,unsigned long end,struct mm_walk * walk)690 static int queue_pages_test_walk(unsigned long start, unsigned long end,
691 struct mm_walk *walk)
692 {
693 struct vm_area_struct *next, *vma = walk->vma;
694 struct queue_pages *qp = walk->private;
695 unsigned long flags = qp->flags;
696
697 /* range check first */
698 VM_BUG_ON_VMA(!range_in_vma(vma, start, end), vma);
699
700 if (!qp->first) {
701 qp->first = vma;
702 if (!(flags & MPOL_MF_DISCONTIG_OK) &&
703 (qp->start < vma->vm_start))
704 /* hole at head side of range */
705 return -EFAULT;
706 }
707 next = find_vma(vma->vm_mm, vma->vm_end);
708 if (!(flags & MPOL_MF_DISCONTIG_OK) &&
709 ((vma->vm_end < qp->end) &&
710 (!next || vma->vm_end < next->vm_start)))
711 /* hole at middle or tail of range */
712 return -EFAULT;
713
714 /*
715 * Need check MPOL_MF_STRICT to return -EIO if possible
716 * regardless of vma_migratable
717 */
718 if (!vma_migratable(vma) &&
719 !(flags & MPOL_MF_STRICT))
720 return 1;
721
722 /*
723 * Check page nodes, and queue pages to move, in the current vma.
724 * But if no moving, and no strict checking, the scan can be skipped.
725 */
726 if (flags & (MPOL_MF_STRICT | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
727 return 0;
728 return 1;
729 }
730
731 static const struct mm_walk_ops queue_pages_walk_ops = {
732 .hugetlb_entry = queue_folios_hugetlb,
733 .pmd_entry = queue_folios_pte_range,
734 .test_walk = queue_pages_test_walk,
735 .walk_lock = PGWALK_RDLOCK,
736 };
737
738 static const struct mm_walk_ops queue_pages_lock_vma_walk_ops = {
739 .hugetlb_entry = queue_folios_hugetlb,
740 .pmd_entry = queue_folios_pte_range,
741 .test_walk = queue_pages_test_walk,
742 .walk_lock = PGWALK_WRLOCK,
743 };
744
745 /*
746 * Walk through page tables and collect pages to be migrated.
747 *
748 * If pages found in a given range are not on the required set of @nodes,
749 * and migration is allowed, they are isolated and queued to @pagelist.
750 *
751 * queue_pages_range() may return:
752 * 0 - all pages already on the right node, or successfully queued for moving
753 * (or neither strict checking nor moving requested: only range checking).
754 * >0 - this number of misplaced folios could not be queued for moving
755 * (a hugetlbfs page or a transparent huge page being counted as 1).
756 * -EIO - a misplaced page found, when MPOL_MF_STRICT specified without MOVEs.
757 * -EFAULT - a hole in the memory range, when MPOL_MF_DISCONTIG_OK unspecified.
758 */
759 static long
queue_pages_range(struct mm_struct * mm,unsigned long start,unsigned long end,nodemask_t * nodes,unsigned long flags,struct list_head * pagelist)760 queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
761 nodemask_t *nodes, unsigned long flags,
762 struct list_head *pagelist)
763 {
764 int err;
765 struct queue_pages qp = {
766 .pagelist = pagelist,
767 .flags = flags,
768 .nmask = nodes,
769 .start = start,
770 .end = end,
771 .first = NULL,
772 };
773 const struct mm_walk_ops *ops = (flags & MPOL_MF_WRLOCK) ?
774 &queue_pages_lock_vma_walk_ops : &queue_pages_walk_ops;
775
776 err = walk_page_range(mm, start, end, ops, &qp);
777
778 if (!qp.first)
779 /* whole range in hole */
780 err = -EFAULT;
781
782 return err ? : qp.nr_failed;
783 }
784
785 /*
786 * Apply policy to a single VMA
787 * This must be called with the mmap_lock held for writing.
788 */
vma_replace_policy(struct vm_area_struct * vma,struct mempolicy * pol)789 static int vma_replace_policy(struct vm_area_struct *vma,
790 struct mempolicy *pol)
791 {
792 int err;
793 struct mempolicy *old;
794 struct mempolicy *new;
795
796 vma_assert_write_locked(vma);
797
798 new = mpol_dup(pol);
799 if (IS_ERR(new))
800 return PTR_ERR(new);
801
802 if (vma->vm_ops && vma->vm_ops->set_policy) {
803 err = vma->vm_ops->set_policy(vma, new);
804 if (err)
805 goto err_out;
806 }
807
808 old = vma->vm_policy;
809 vma->vm_policy = new; /* protected by mmap_lock */
810 mpol_put(old);
811
812 return 0;
813 err_out:
814 mpol_put(new);
815 return err;
816 }
817
818 /* Split or merge the VMA (if required) and apply the new policy */
mbind_range(struct vma_iterator * vmi,struct vm_area_struct * vma,struct vm_area_struct ** prev,unsigned long start,unsigned long end,struct mempolicy * new_pol)819 static int mbind_range(struct vma_iterator *vmi, struct vm_area_struct *vma,
820 struct vm_area_struct **prev, unsigned long start,
821 unsigned long end, struct mempolicy *new_pol)
822 {
823 unsigned long vmstart, vmend;
824
825 vmend = min(end, vma->vm_end);
826 if (start > vma->vm_start) {
827 *prev = vma;
828 vmstart = start;
829 } else {
830 vmstart = vma->vm_start;
831 }
832
833 if (mpol_equal(vma->vm_policy, new_pol)) {
834 *prev = vma;
835 return 0;
836 }
837
838 vma = vma_modify_policy(vmi, *prev, vma, vmstart, vmend, new_pol);
839 if (IS_ERR(vma))
840 return PTR_ERR(vma);
841
842 *prev = vma;
843 return vma_replace_policy(vma, new_pol);
844 }
845
846 /* Set the process memory policy */
do_set_mempolicy(unsigned short mode,unsigned short flags,nodemask_t * nodes)847 static long do_set_mempolicy(unsigned short mode, unsigned short flags,
848 nodemask_t *nodes)
849 {
850 struct mempolicy *new, *old;
851 NODEMASK_SCRATCH(scratch);
852 int ret;
853
854 if (!scratch)
855 return -ENOMEM;
856
857 new = mpol_new(mode, flags, nodes);
858 if (IS_ERR(new)) {
859 ret = PTR_ERR(new);
860 goto out;
861 }
862
863 task_lock(current);
864 ret = mpol_set_nodemask(new, nodes, scratch);
865 if (ret) {
866 task_unlock(current);
867 mpol_put(new);
868 goto out;
869 }
870
871 old = current->mempolicy;
872 current->mempolicy = new;
873 if (new && (new->mode == MPOL_INTERLEAVE ||
874 new->mode == MPOL_WEIGHTED_INTERLEAVE)) {
875 current->il_prev = MAX_NUMNODES-1;
876 current->il_weight = 0;
877 }
878 task_unlock(current);
879 mpol_put(old);
880 ret = 0;
881 out:
882 NODEMASK_SCRATCH_FREE(scratch);
883 return ret;
884 }
885
886 /*
887 * Return nodemask for policy for get_mempolicy() query
888 *
889 * Called with task's alloc_lock held
890 */
get_policy_nodemask(struct mempolicy * pol,nodemask_t * nodes)891 static void get_policy_nodemask(struct mempolicy *pol, nodemask_t *nodes)
892 {
893 nodes_clear(*nodes);
894 if (pol == &default_policy)
895 return;
896
897 switch (pol->mode) {
898 case MPOL_BIND:
899 case MPOL_INTERLEAVE:
900 case MPOL_PREFERRED:
901 case MPOL_PREFERRED_MANY:
902 case MPOL_WEIGHTED_INTERLEAVE:
903 *nodes = pol->nodes;
904 break;
905 case MPOL_LOCAL:
906 /* return empty node mask for local allocation */
907 break;
908 default:
909 BUG();
910 }
911 }
912
lookup_node(struct mm_struct * mm,unsigned long addr)913 static int lookup_node(struct mm_struct *mm, unsigned long addr)
914 {
915 struct page *p = NULL;
916 int ret;
917
918 ret = get_user_pages_fast(addr & PAGE_MASK, 1, 0, &p);
919 if (ret > 0) {
920 ret = page_to_nid(p);
921 put_page(p);
922 }
923 return ret;
924 }
925
926 /* Retrieve NUMA policy */
do_get_mempolicy(int * policy,nodemask_t * nmask,unsigned long addr,unsigned long flags)927 static long do_get_mempolicy(int *policy, nodemask_t *nmask,
928 unsigned long addr, unsigned long flags)
929 {
930 int err;
931 struct mm_struct *mm = current->mm;
932 struct vm_area_struct *vma = NULL;
933 struct mempolicy *pol = current->mempolicy, *pol_refcount = NULL;
934
935 if (flags &
936 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
937 return -EINVAL;
938
939 if (flags & MPOL_F_MEMS_ALLOWED) {
940 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
941 return -EINVAL;
942 *policy = 0; /* just so it's initialized */
943 task_lock(current);
944 *nmask = cpuset_current_mems_allowed;
945 task_unlock(current);
946 return 0;
947 }
948
949 if (flags & MPOL_F_ADDR) {
950 pgoff_t ilx; /* ignored here */
951 /*
952 * Do NOT fall back to task policy if the
953 * vma/shared policy at addr is NULL. We
954 * want to return MPOL_DEFAULT in this case.
955 */
956 mmap_read_lock(mm);
957 vma = vma_lookup(mm, addr);
958 if (!vma) {
959 mmap_read_unlock(mm);
960 return -EFAULT;
961 }
962 pol = __get_vma_policy(vma, addr, &ilx);
963 } else if (addr)
964 return -EINVAL;
965
966 if (!pol)
967 pol = &default_policy; /* indicates default behavior */
968
969 if (flags & MPOL_F_NODE) {
970 if (flags & MPOL_F_ADDR) {
971 /*
972 * Take a refcount on the mpol, because we are about to
973 * drop the mmap_lock, after which only "pol" remains
974 * valid, "vma" is stale.
975 */
976 pol_refcount = pol;
977 vma = NULL;
978 mpol_get(pol);
979 mmap_read_unlock(mm);
980 err = lookup_node(mm, addr);
981 if (err < 0)
982 goto out;
983 *policy = err;
984 } else if (pol == current->mempolicy &&
985 pol->mode == MPOL_INTERLEAVE) {
986 *policy = next_node_in(current->il_prev, pol->nodes);
987 } else if (pol == current->mempolicy &&
988 pol->mode == MPOL_WEIGHTED_INTERLEAVE) {
989 if (current->il_weight)
990 *policy = current->il_prev;
991 else
992 *policy = next_node_in(current->il_prev,
993 pol->nodes);
994 } else {
995 err = -EINVAL;
996 goto out;
997 }
998 } else {
999 *policy = pol == &default_policy ? MPOL_DEFAULT :
1000 pol->mode;
1001 /*
1002 * Internal mempolicy flags must be masked off before exposing
1003 * the policy to userspace.
1004 */
1005 *policy |= (pol->flags & MPOL_MODE_FLAGS);
1006 }
1007
1008 err = 0;
1009 if (nmask) {
1010 if (mpol_store_user_nodemask(pol)) {
1011 *nmask = pol->w.user_nodemask;
1012 } else {
1013 task_lock(current);
1014 get_policy_nodemask(pol, nmask);
1015 task_unlock(current);
1016 }
1017 }
1018
1019 out:
1020 mpol_cond_put(pol);
1021 if (vma)
1022 mmap_read_unlock(mm);
1023 if (pol_refcount)
1024 mpol_put(pol_refcount);
1025 return err;
1026 }
1027
1028 #ifdef CONFIG_MIGRATION
migrate_folio_add(struct folio * folio,struct list_head * foliolist,unsigned long flags)1029 static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist,
1030 unsigned long flags)
1031 {
1032 /*
1033 * Unless MPOL_MF_MOVE_ALL, we try to avoid migrating a shared folio.
1034 * Choosing not to migrate a shared folio is not counted as a failure.
1035 *
1036 * See folio_likely_mapped_shared() on possible imprecision when we
1037 * cannot easily detect if a folio is shared.
1038 */
1039 if ((flags & MPOL_MF_MOVE_ALL) || !folio_likely_mapped_shared(folio)) {
1040 if (folio_isolate_lru(folio)) {
1041 list_add_tail(&folio->lru, foliolist);
1042 node_stat_mod_folio(folio,
1043 NR_ISOLATED_ANON + folio_is_file_lru(folio),
1044 folio_nr_pages(folio));
1045 } else {
1046 /*
1047 * Non-movable folio may reach here. And, there may be
1048 * temporary off LRU folios or non-LRU movable folios.
1049 * Treat them as unmovable folios since they can't be
1050 * isolated, so they can't be moved at the moment.
1051 */
1052 return false;
1053 }
1054 }
1055 return true;
1056 }
1057
1058 /*
1059 * Migrate pages from one node to a target node.
1060 * Returns error or the number of pages not migrated.
1061 */
migrate_to_node(struct mm_struct * mm,int source,int dest,int flags)1062 static long migrate_to_node(struct mm_struct *mm, int source, int dest,
1063 int flags)
1064 {
1065 nodemask_t nmask;
1066 struct vm_area_struct *vma;
1067 LIST_HEAD(pagelist);
1068 long nr_failed;
1069 long err = 0;
1070 struct migration_target_control mtc = {
1071 .nid = dest,
1072 .gfp_mask = GFP_HIGHUSER_MOVABLE | __GFP_THISNODE,
1073 .reason = MR_SYSCALL,
1074 };
1075
1076 nodes_clear(nmask);
1077 node_set(source, nmask);
1078
1079 VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)));
1080
1081 mmap_read_lock(mm);
1082 vma = find_vma(mm, 0);
1083
1084 /*
1085 * This does not migrate the range, but isolates all pages that
1086 * need migration. Between passing in the full user address
1087 * space range and MPOL_MF_DISCONTIG_OK, this call cannot fail,
1088 * but passes back the count of pages which could not be isolated.
1089 */
1090 nr_failed = queue_pages_range(mm, vma->vm_start, mm->task_size, &nmask,
1091 flags | MPOL_MF_DISCONTIG_OK, &pagelist);
1092 mmap_read_unlock(mm);
1093
1094 if (!list_empty(&pagelist)) {
1095 err = migrate_pages(&pagelist, alloc_migration_target, NULL,
1096 (unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL, NULL);
1097 if (err)
1098 putback_movable_pages(&pagelist);
1099 }
1100
1101 if (err >= 0)
1102 err += nr_failed;
1103 return err;
1104 }
1105
1106 /*
1107 * Move pages between the two nodesets so as to preserve the physical
1108 * layout as much as possible.
1109 *
1110 * Returns the number of page that could not be moved.
1111 */
do_migrate_pages(struct mm_struct * mm,const nodemask_t * from,const nodemask_t * to,int flags)1112 int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1113 const nodemask_t *to, int flags)
1114 {
1115 long nr_failed = 0;
1116 long err = 0;
1117 nodemask_t tmp;
1118
1119 lru_cache_disable();
1120
1121 /*
1122 * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
1123 * bit in 'to' is not also set in 'tmp'. Clear the found 'source'
1124 * bit in 'tmp', and return that <source, dest> pair for migration.
1125 * The pair of nodemasks 'to' and 'from' define the map.
1126 *
1127 * If no pair of bits is found that way, fallback to picking some
1128 * pair of 'source' and 'dest' bits that are not the same. If the
1129 * 'source' and 'dest' bits are the same, this represents a node
1130 * that will be migrating to itself, so no pages need move.
1131 *
1132 * If no bits are left in 'tmp', or if all remaining bits left
1133 * in 'tmp' correspond to the same bit in 'to', return false
1134 * (nothing left to migrate).
1135 *
1136 * This lets us pick a pair of nodes to migrate between, such that
1137 * if possible the dest node is not already occupied by some other
1138 * source node, minimizing the risk of overloading the memory on a
1139 * node that would happen if we migrated incoming memory to a node
1140 * before migrating outgoing memory source that same node.
1141 *
1142 * A single scan of tmp is sufficient. As we go, we remember the
1143 * most recent <s, d> pair that moved (s != d). If we find a pair
1144 * that not only moved, but what's better, moved to an empty slot
1145 * (d is not set in tmp), then we break out then, with that pair.
1146 * Otherwise when we finish scanning from_tmp, we at least have the
1147 * most recent <s, d> pair that moved. If we get all the way through
1148 * the scan of tmp without finding any node that moved, much less
1149 * moved to an empty node, then there is nothing left worth migrating.
1150 */
1151
1152 tmp = *from;
1153 while (!nodes_empty(tmp)) {
1154 int s, d;
1155 int source = NUMA_NO_NODE;
1156 int dest = 0;
1157
1158 for_each_node_mask(s, tmp) {
1159
1160 /*
1161 * do_migrate_pages() tries to maintain the relative
1162 * node relationship of the pages established between
1163 * threads and memory areas.
1164 *
1165 * However if the number of source nodes is not equal to
1166 * the number of destination nodes we can not preserve
1167 * this node relative relationship. In that case, skip
1168 * copying memory from a node that is in the destination
1169 * mask.
1170 *
1171 * Example: [2,3,4] -> [3,4,5] moves everything.
1172 * [0-7] - > [3,4,5] moves only 0,1,2,6,7.
1173 */
1174
1175 if ((nodes_weight(*from) != nodes_weight(*to)) &&
1176 (node_isset(s, *to)))
1177 continue;
1178
1179 d = node_remap(s, *from, *to);
1180 if (s == d)
1181 continue;
1182
1183 source = s; /* Node moved. Memorize */
1184 dest = d;
1185
1186 /* dest not in remaining from nodes? */
1187 if (!node_isset(dest, tmp))
1188 break;
1189 }
1190 if (source == NUMA_NO_NODE)
1191 break;
1192
1193 node_clear(source, tmp);
1194 err = migrate_to_node(mm, source, dest, flags);
1195 if (err > 0)
1196 nr_failed += err;
1197 if (err < 0)
1198 break;
1199 }
1200
1201 lru_cache_enable();
1202 if (err < 0)
1203 return err;
1204 return (nr_failed < INT_MAX) ? nr_failed : INT_MAX;
1205 }
1206
1207 /*
1208 * Allocate a new folio for page migration, according to NUMA mempolicy.
1209 */
alloc_migration_target_by_mpol(struct folio * src,unsigned long private)1210 static struct folio *alloc_migration_target_by_mpol(struct folio *src,
1211 unsigned long private)
1212 {
1213 struct migration_mpol *mmpol = (struct migration_mpol *)private;
1214 struct mempolicy *pol = mmpol->pol;
1215 pgoff_t ilx = mmpol->ilx;
1216 unsigned int order;
1217 int nid = numa_node_id();
1218 gfp_t gfp;
1219
1220 order = folio_order(src);
1221 ilx += src->index >> order;
1222
1223 if (folio_test_hugetlb(src)) {
1224 nodemask_t *nodemask;
1225 struct hstate *h;
1226
1227 h = folio_hstate(src);
1228 gfp = htlb_alloc_mask(h);
1229 nodemask = policy_nodemask(gfp, pol, ilx, &nid);
1230 return alloc_hugetlb_folio_nodemask(h, nid, nodemask, gfp,
1231 htlb_allow_alloc_fallback(MR_MEMPOLICY_MBIND));
1232 }
1233
1234 if (folio_test_large(src))
1235 gfp = GFP_TRANSHUGE;
1236 else
1237 gfp = GFP_HIGHUSER_MOVABLE | __GFP_RETRY_MAYFAIL | __GFP_COMP;
1238
1239 return folio_alloc_mpol(gfp, order, pol, ilx, nid);
1240 }
1241 #else
1242
migrate_folio_add(struct folio * folio,struct list_head * foliolist,unsigned long flags)1243 static bool migrate_folio_add(struct folio *folio, struct list_head *foliolist,
1244 unsigned long flags)
1245 {
1246 return false;
1247 }
1248
do_migrate_pages(struct mm_struct * mm,const nodemask_t * from,const nodemask_t * to,int flags)1249 int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1250 const nodemask_t *to, int flags)
1251 {
1252 return -ENOSYS;
1253 }
1254
alloc_migration_target_by_mpol(struct folio * src,unsigned long private)1255 static struct folio *alloc_migration_target_by_mpol(struct folio *src,
1256 unsigned long private)
1257 {
1258 return NULL;
1259 }
1260 #endif
1261
do_mbind(unsigned long start,unsigned long len,unsigned short mode,unsigned short mode_flags,nodemask_t * nmask,unsigned long flags)1262 static long do_mbind(unsigned long start, unsigned long len,
1263 unsigned short mode, unsigned short mode_flags,
1264 nodemask_t *nmask, unsigned long flags)
1265 {
1266 struct mm_struct *mm = current->mm;
1267 struct vm_area_struct *vma, *prev;
1268 struct vma_iterator vmi;
1269 struct migration_mpol mmpol;
1270 struct mempolicy *new;
1271 unsigned long end;
1272 long err;
1273 long nr_failed;
1274 LIST_HEAD(pagelist);
1275
1276 if (flags & ~(unsigned long)MPOL_MF_VALID)
1277 return -EINVAL;
1278 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1279 return -EPERM;
1280
1281 if (start & ~PAGE_MASK)
1282 return -EINVAL;
1283
1284 if (mode == MPOL_DEFAULT)
1285 flags &= ~MPOL_MF_STRICT;
1286
1287 len = PAGE_ALIGN(len);
1288 end = start + len;
1289
1290 if (end < start)
1291 return -EINVAL;
1292 if (end == start)
1293 return 0;
1294
1295 new = mpol_new(mode, mode_flags, nmask);
1296 if (IS_ERR(new))
1297 return PTR_ERR(new);
1298
1299 /*
1300 * If we are using the default policy then operation
1301 * on discontinuous address spaces is okay after all
1302 */
1303 if (!new)
1304 flags |= MPOL_MF_DISCONTIG_OK;
1305
1306 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
1307 lru_cache_disable();
1308 {
1309 NODEMASK_SCRATCH(scratch);
1310 if (scratch) {
1311 mmap_write_lock(mm);
1312 err = mpol_set_nodemask(new, nmask, scratch);
1313 if (err)
1314 mmap_write_unlock(mm);
1315 } else
1316 err = -ENOMEM;
1317 NODEMASK_SCRATCH_FREE(scratch);
1318 }
1319 if (err)
1320 goto mpol_out;
1321
1322 /*
1323 * Lock the VMAs before scanning for pages to migrate,
1324 * to ensure we don't miss a concurrently inserted page.
1325 */
1326 nr_failed = queue_pages_range(mm, start, end, nmask,
1327 flags | MPOL_MF_INVERT | MPOL_MF_WRLOCK, &pagelist);
1328
1329 if (nr_failed < 0) {
1330 err = nr_failed;
1331 nr_failed = 0;
1332 } else {
1333 vma_iter_init(&vmi, mm, start);
1334 prev = vma_prev(&vmi);
1335 for_each_vma_range(vmi, vma, end) {
1336 err = mbind_range(&vmi, vma, &prev, start, end, new);
1337 if (err)
1338 break;
1339 }
1340 }
1341
1342 if (!err && !list_empty(&pagelist)) {
1343 /* Convert MPOL_DEFAULT's NULL to task or default policy */
1344 if (!new) {
1345 new = get_task_policy(current);
1346 mpol_get(new);
1347 }
1348 mmpol.pol = new;
1349 mmpol.ilx = 0;
1350
1351 /*
1352 * In the interleaved case, attempt to allocate on exactly the
1353 * targeted nodes, for the first VMA to be migrated; for later
1354 * VMAs, the nodes will still be interleaved from the targeted
1355 * nodemask, but one by one may be selected differently.
1356 */
1357 if (new->mode == MPOL_INTERLEAVE ||
1358 new->mode == MPOL_WEIGHTED_INTERLEAVE) {
1359 struct folio *folio;
1360 unsigned int order;
1361 unsigned long addr = -EFAULT;
1362
1363 list_for_each_entry(folio, &pagelist, lru) {
1364 if (!folio_test_ksm(folio))
1365 break;
1366 }
1367 if (!list_entry_is_head(folio, &pagelist, lru)) {
1368 vma_iter_init(&vmi, mm, start);
1369 for_each_vma_range(vmi, vma, end) {
1370 addr = page_address_in_vma(
1371 folio_page(folio, 0), vma);
1372 if (addr != -EFAULT)
1373 break;
1374 }
1375 }
1376 if (addr != -EFAULT) {
1377 order = folio_order(folio);
1378 /* We already know the pol, but not the ilx */
1379 mpol_cond_put(get_vma_policy(vma, addr, order,
1380 &mmpol.ilx));
1381 /* Set base from which to increment by index */
1382 mmpol.ilx -= folio->index >> order;
1383 }
1384 }
1385 }
1386
1387 mmap_write_unlock(mm);
1388
1389 if (!err && !list_empty(&pagelist)) {
1390 nr_failed |= migrate_pages(&pagelist,
1391 alloc_migration_target_by_mpol, NULL,
1392 (unsigned long)&mmpol, MIGRATE_SYNC,
1393 MR_MEMPOLICY_MBIND, NULL);
1394 }
1395
1396 if (nr_failed && (flags & MPOL_MF_STRICT))
1397 err = -EIO;
1398 if (!list_empty(&pagelist))
1399 putback_movable_pages(&pagelist);
1400 mpol_out:
1401 mpol_put(new);
1402 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
1403 lru_cache_enable();
1404 return err;
1405 }
1406
1407 /*
1408 * User space interface with variable sized bitmaps for nodelists.
1409 */
get_bitmap(unsigned long * mask,const unsigned long __user * nmask,unsigned long maxnode)1410 static int get_bitmap(unsigned long *mask, const unsigned long __user *nmask,
1411 unsigned long maxnode)
1412 {
1413 unsigned long nlongs = BITS_TO_LONGS(maxnode);
1414 int ret;
1415
1416 if (in_compat_syscall())
1417 ret = compat_get_bitmap(mask,
1418 (const compat_ulong_t __user *)nmask,
1419 maxnode);
1420 else
1421 ret = copy_from_user(mask, nmask,
1422 nlongs * sizeof(unsigned long));
1423
1424 if (ret)
1425 return -EFAULT;
1426
1427 if (maxnode % BITS_PER_LONG)
1428 mask[nlongs - 1] &= (1UL << (maxnode % BITS_PER_LONG)) - 1;
1429
1430 return 0;
1431 }
1432
1433 /* Copy a node mask from user space. */
get_nodes(nodemask_t * nodes,const unsigned long __user * nmask,unsigned long maxnode)1434 static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
1435 unsigned long maxnode)
1436 {
1437 --maxnode;
1438 nodes_clear(*nodes);
1439 if (maxnode == 0 || !nmask)
1440 return 0;
1441 if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
1442 return -EINVAL;
1443
1444 /*
1445 * When the user specified more nodes than supported just check
1446 * if the non supported part is all zero, one word at a time,
1447 * starting at the end.
1448 */
1449 while (maxnode > MAX_NUMNODES) {
1450 unsigned long bits = min_t(unsigned long, maxnode, BITS_PER_LONG);
1451 unsigned long t;
1452
1453 if (get_bitmap(&t, &nmask[(maxnode - 1) / BITS_PER_LONG], bits))
1454 return -EFAULT;
1455
1456 if (maxnode - bits >= MAX_NUMNODES) {
1457 maxnode -= bits;
1458 } else {
1459 maxnode = MAX_NUMNODES;
1460 t &= ~((1UL << (MAX_NUMNODES % BITS_PER_LONG)) - 1);
1461 }
1462 if (t)
1463 return -EINVAL;
1464 }
1465
1466 return get_bitmap(nodes_addr(*nodes), nmask, maxnode);
1467 }
1468
1469 /* Copy a kernel node mask to user space */
copy_nodes_to_user(unsigned long __user * mask,unsigned long maxnode,nodemask_t * nodes)1470 static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1471 nodemask_t *nodes)
1472 {
1473 unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1474 unsigned int nbytes = BITS_TO_LONGS(nr_node_ids) * sizeof(long);
1475 bool compat = in_compat_syscall();
1476
1477 if (compat)
1478 nbytes = BITS_TO_COMPAT_LONGS(nr_node_ids) * sizeof(compat_long_t);
1479
1480 if (copy > nbytes) {
1481 if (copy > PAGE_SIZE)
1482 return -EINVAL;
1483 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1484 return -EFAULT;
1485 copy = nbytes;
1486 maxnode = nr_node_ids;
1487 }
1488
1489 if (compat)
1490 return compat_put_bitmap((compat_ulong_t __user *)mask,
1491 nodes_addr(*nodes), maxnode);
1492
1493 return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1494 }
1495
1496 /* Basic parameter sanity check used by both mbind() and set_mempolicy() */
sanitize_mpol_flags(int * mode,unsigned short * flags)1497 static inline int sanitize_mpol_flags(int *mode, unsigned short *flags)
1498 {
1499 *flags = *mode & MPOL_MODE_FLAGS;
1500 *mode &= ~MPOL_MODE_FLAGS;
1501
1502 if ((unsigned int)(*mode) >= MPOL_MAX)
1503 return -EINVAL;
1504 if ((*flags & MPOL_F_STATIC_NODES) && (*flags & MPOL_F_RELATIVE_NODES))
1505 return -EINVAL;
1506 if (*flags & MPOL_F_NUMA_BALANCING) {
1507 if (*mode == MPOL_BIND || *mode == MPOL_PREFERRED_MANY)
1508 *flags |= (MPOL_F_MOF | MPOL_F_MORON);
1509 else
1510 return -EINVAL;
1511 }
1512 return 0;
1513 }
1514
kernel_mbind(unsigned long start,unsigned long len,unsigned long mode,const unsigned long __user * nmask,unsigned long maxnode,unsigned int flags)1515 static long kernel_mbind(unsigned long start, unsigned long len,
1516 unsigned long mode, const unsigned long __user *nmask,
1517 unsigned long maxnode, unsigned int flags)
1518 {
1519 unsigned short mode_flags;
1520 nodemask_t nodes;
1521 int lmode = mode;
1522 int err;
1523
1524 start = untagged_addr(start);
1525 err = sanitize_mpol_flags(&lmode, &mode_flags);
1526 if (err)
1527 return err;
1528
1529 err = get_nodes(&nodes, nmask, maxnode);
1530 if (err)
1531 return err;
1532
1533 return do_mbind(start, len, lmode, mode_flags, &nodes, flags);
1534 }
1535
SYSCALL_DEFINE4(set_mempolicy_home_node,unsigned long,start,unsigned long,len,unsigned long,home_node,unsigned long,flags)1536 SYSCALL_DEFINE4(set_mempolicy_home_node, unsigned long, start, unsigned long, len,
1537 unsigned long, home_node, unsigned long, flags)
1538 {
1539 struct mm_struct *mm = current->mm;
1540 struct vm_area_struct *vma, *prev;
1541 struct mempolicy *new, *old;
1542 unsigned long end;
1543 int err = -ENOENT;
1544 VMA_ITERATOR(vmi, mm, start);
1545
1546 start = untagged_addr(start);
1547 if (start & ~PAGE_MASK)
1548 return -EINVAL;
1549 /*
1550 * flags is used for future extension if any.
1551 */
1552 if (flags != 0)
1553 return -EINVAL;
1554
1555 /*
1556 * Check home_node is online to avoid accessing uninitialized
1557 * NODE_DATA.
1558 */
1559 if (home_node >= MAX_NUMNODES || !node_online(home_node))
1560 return -EINVAL;
1561
1562 len = PAGE_ALIGN(len);
1563 end = start + len;
1564
1565 if (end < start)
1566 return -EINVAL;
1567 if (end == start)
1568 return 0;
1569 mmap_write_lock(mm);
1570 prev = vma_prev(&vmi);
1571 for_each_vma_range(vmi, vma, end) {
1572 /*
1573 * If any vma in the range got policy other than MPOL_BIND
1574 * or MPOL_PREFERRED_MANY we return error. We don't reset
1575 * the home node for vmas we already updated before.
1576 */
1577 old = vma_policy(vma);
1578 if (!old) {
1579 prev = vma;
1580 continue;
1581 }
1582 if (old->mode != MPOL_BIND && old->mode != MPOL_PREFERRED_MANY) {
1583 err = -EOPNOTSUPP;
1584 break;
1585 }
1586 new = mpol_dup(old);
1587 if (IS_ERR(new)) {
1588 err = PTR_ERR(new);
1589 break;
1590 }
1591
1592 vma_start_write(vma);
1593 new->home_node = home_node;
1594 err = mbind_range(&vmi, vma, &prev, start, end, new);
1595 mpol_put(new);
1596 if (err)
1597 break;
1598 }
1599 mmap_write_unlock(mm);
1600 return err;
1601 }
1602
SYSCALL_DEFINE6(mbind,unsigned long,start,unsigned long,len,unsigned long,mode,const unsigned long __user *,nmask,unsigned long,maxnode,unsigned int,flags)1603 SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
1604 unsigned long, mode, const unsigned long __user *, nmask,
1605 unsigned long, maxnode, unsigned int, flags)
1606 {
1607 return kernel_mbind(start, len, mode, nmask, maxnode, flags);
1608 }
1609
1610 /* Set the process memory policy */
kernel_set_mempolicy(int mode,const unsigned long __user * nmask,unsigned long maxnode)1611 static long kernel_set_mempolicy(int mode, const unsigned long __user *nmask,
1612 unsigned long maxnode)
1613 {
1614 unsigned short mode_flags;
1615 nodemask_t nodes;
1616 int lmode = mode;
1617 int err;
1618
1619 err = sanitize_mpol_flags(&lmode, &mode_flags);
1620 if (err)
1621 return err;
1622
1623 err = get_nodes(&nodes, nmask, maxnode);
1624 if (err)
1625 return err;
1626
1627 return do_set_mempolicy(lmode, mode_flags, &nodes);
1628 }
1629
SYSCALL_DEFINE3(set_mempolicy,int,mode,const unsigned long __user *,nmask,unsigned long,maxnode)1630 SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask,
1631 unsigned long, maxnode)
1632 {
1633 return kernel_set_mempolicy(mode, nmask, maxnode);
1634 }
1635
kernel_migrate_pages(pid_t pid,unsigned long maxnode,const unsigned long __user * old_nodes,const unsigned long __user * new_nodes)1636 static int kernel_migrate_pages(pid_t pid, unsigned long maxnode,
1637 const unsigned long __user *old_nodes,
1638 const unsigned long __user *new_nodes)
1639 {
1640 struct mm_struct *mm = NULL;
1641 struct task_struct *task;
1642 nodemask_t task_nodes;
1643 int err;
1644 nodemask_t *old;
1645 nodemask_t *new;
1646 NODEMASK_SCRATCH(scratch);
1647
1648 if (!scratch)
1649 return -ENOMEM;
1650
1651 old = &scratch->mask1;
1652 new = &scratch->mask2;
1653
1654 err = get_nodes(old, old_nodes, maxnode);
1655 if (err)
1656 goto out;
1657
1658 err = get_nodes(new, new_nodes, maxnode);
1659 if (err)
1660 goto out;
1661
1662 /* Find the mm_struct */
1663 rcu_read_lock();
1664 task = pid ? find_task_by_vpid(pid) : current;
1665 if (!task) {
1666 rcu_read_unlock();
1667 err = -ESRCH;
1668 goto out;
1669 }
1670 get_task_struct(task);
1671
1672 err = -EINVAL;
1673
1674 /*
1675 * Check if this process has the right to modify the specified process.
1676 * Use the regular "ptrace_may_access()" checks.
1677 */
1678 if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) {
1679 rcu_read_unlock();
1680 err = -EPERM;
1681 goto out_put;
1682 }
1683 rcu_read_unlock();
1684
1685 task_nodes = cpuset_mems_allowed(task);
1686 /* Is the user allowed to access the target nodes? */
1687 if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
1688 err = -EPERM;
1689 goto out_put;
1690 }
1691
1692 task_nodes = cpuset_mems_allowed(current);
1693 nodes_and(*new, *new, task_nodes);
1694 if (nodes_empty(*new))
1695 goto out_put;
1696
1697 err = security_task_movememory(task);
1698 if (err)
1699 goto out_put;
1700
1701 mm = get_task_mm(task);
1702 put_task_struct(task);
1703
1704 if (!mm) {
1705 err = -EINVAL;
1706 goto out;
1707 }
1708
1709 err = do_migrate_pages(mm, old, new,
1710 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
1711
1712 mmput(mm);
1713 out:
1714 NODEMASK_SCRATCH_FREE(scratch);
1715
1716 return err;
1717
1718 out_put:
1719 put_task_struct(task);
1720 goto out;
1721 }
1722
SYSCALL_DEFINE4(migrate_pages,pid_t,pid,unsigned long,maxnode,const unsigned long __user *,old_nodes,const unsigned long __user *,new_nodes)1723 SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
1724 const unsigned long __user *, old_nodes,
1725 const unsigned long __user *, new_nodes)
1726 {
1727 return kernel_migrate_pages(pid, maxnode, old_nodes, new_nodes);
1728 }
1729
1730 /* Retrieve NUMA policy */
kernel_get_mempolicy(int __user * policy,unsigned long __user * nmask,unsigned long maxnode,unsigned long addr,unsigned long flags)1731 static int kernel_get_mempolicy(int __user *policy,
1732 unsigned long __user *nmask,
1733 unsigned long maxnode,
1734 unsigned long addr,
1735 unsigned long flags)
1736 {
1737 int err;
1738 int pval;
1739 nodemask_t nodes;
1740
1741 if (nmask != NULL && maxnode < nr_node_ids)
1742 return -EINVAL;
1743
1744 addr = untagged_addr(addr);
1745
1746 err = do_get_mempolicy(&pval, &nodes, addr, flags);
1747
1748 if (err)
1749 return err;
1750
1751 if (policy && put_user(pval, policy))
1752 return -EFAULT;
1753
1754 if (nmask)
1755 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1756
1757 return err;
1758 }
1759
SYSCALL_DEFINE5(get_mempolicy,int __user *,policy,unsigned long __user *,nmask,unsigned long,maxnode,unsigned long,addr,unsigned long,flags)1760 SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1761 unsigned long __user *, nmask, unsigned long, maxnode,
1762 unsigned long, addr, unsigned long, flags)
1763 {
1764 return kernel_get_mempolicy(policy, nmask, maxnode, addr, flags);
1765 }
1766
vma_migratable(struct vm_area_struct * vma)1767 bool vma_migratable(struct vm_area_struct *vma)
1768 {
1769 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
1770 return false;
1771
1772 /*
1773 * DAX device mappings require predictable access latency, so avoid
1774 * incurring periodic faults.
1775 */
1776 if (vma_is_dax(vma))
1777 return false;
1778
1779 if (is_vm_hugetlb_page(vma) &&
1780 !hugepage_migration_supported(hstate_vma(vma)))
1781 return false;
1782
1783 /*
1784 * Migration allocates pages in the highest zone. If we cannot
1785 * do so then migration (at least from node to node) is not
1786 * possible.
1787 */
1788 if (vma->vm_file &&
1789 gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping))
1790 < policy_zone)
1791 return false;
1792 return true;
1793 }
1794
__get_vma_policy(struct vm_area_struct * vma,unsigned long addr,pgoff_t * ilx)1795 struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
1796 unsigned long addr, pgoff_t *ilx)
1797 {
1798 *ilx = 0;
1799 return (vma->vm_ops && vma->vm_ops->get_policy) ?
1800 vma->vm_ops->get_policy(vma, addr, ilx) : vma->vm_policy;
1801 }
1802
1803 /*
1804 * get_vma_policy(@vma, @addr, @order, @ilx)
1805 * @vma: virtual memory area whose policy is sought
1806 * @addr: address in @vma for shared policy lookup
1807 * @order: 0, or appropriate huge_page_order for interleaving
1808 * @ilx: interleave index (output), for use only when MPOL_INTERLEAVE or
1809 * MPOL_WEIGHTED_INTERLEAVE
1810 *
1811 * Returns effective policy for a VMA at specified address.
1812 * Falls back to current->mempolicy or system default policy, as necessary.
1813 * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1814 * count--added by the get_policy() vm_op, as appropriate--to protect against
1815 * freeing by another task. It is the caller's responsibility to free the
1816 * extra reference for shared policies.
1817 */
get_vma_policy(struct vm_area_struct * vma,unsigned long addr,int order,pgoff_t * ilx)1818 struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
1819 unsigned long addr, int order, pgoff_t *ilx)
1820 {
1821 struct mempolicy *pol;
1822
1823 pol = __get_vma_policy(vma, addr, ilx);
1824 if (!pol)
1825 pol = get_task_policy(current);
1826 if (pol->mode == MPOL_INTERLEAVE ||
1827 pol->mode == MPOL_WEIGHTED_INTERLEAVE) {
1828 *ilx += vma->vm_pgoff >> order;
1829 *ilx += (addr - vma->vm_start) >> (PAGE_SHIFT + order);
1830 }
1831 return pol;
1832 }
1833
vma_policy_mof(struct vm_area_struct * vma)1834 bool vma_policy_mof(struct vm_area_struct *vma)
1835 {
1836 struct mempolicy *pol;
1837
1838 if (vma->vm_ops && vma->vm_ops->get_policy) {
1839 bool ret = false;
1840 pgoff_t ilx; /* ignored here */
1841
1842 pol = vma->vm_ops->get_policy(vma, vma->vm_start, &ilx);
1843 if (pol && (pol->flags & MPOL_F_MOF))
1844 ret = true;
1845 mpol_cond_put(pol);
1846
1847 return ret;
1848 }
1849
1850 pol = vma->vm_policy;
1851 if (!pol)
1852 pol = get_task_policy(current);
1853
1854 return pol->flags & MPOL_F_MOF;
1855 }
1856
apply_policy_zone(struct mempolicy * policy,enum zone_type zone)1857 bool apply_policy_zone(struct mempolicy *policy, enum zone_type zone)
1858 {
1859 enum zone_type dynamic_policy_zone = policy_zone;
1860
1861 BUG_ON(dynamic_policy_zone == ZONE_MOVABLE);
1862
1863 /*
1864 * if policy->nodes has movable memory only,
1865 * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only.
1866 *
1867 * policy->nodes is intersect with node_states[N_MEMORY].
1868 * so if the following test fails, it implies
1869 * policy->nodes has movable memory only.
1870 */
1871 if (!nodes_intersects(policy->nodes, node_states[N_HIGH_MEMORY]))
1872 dynamic_policy_zone = ZONE_MOVABLE;
1873
1874 return zone >= dynamic_policy_zone;
1875 }
1876
weighted_interleave_nodes(struct mempolicy * policy)1877 static unsigned int weighted_interleave_nodes(struct mempolicy *policy)
1878 {
1879 unsigned int node;
1880 unsigned int cpuset_mems_cookie;
1881
1882 retry:
1883 /* to prevent miscount use tsk->mems_allowed_seq to detect rebind */
1884 cpuset_mems_cookie = read_mems_allowed_begin();
1885 node = current->il_prev;
1886 if (!current->il_weight || !node_isset(node, policy->nodes)) {
1887 node = next_node_in(node, policy->nodes);
1888 if (read_mems_allowed_retry(cpuset_mems_cookie))
1889 goto retry;
1890 if (node == MAX_NUMNODES)
1891 return node;
1892 current->il_prev = node;
1893 current->il_weight = get_il_weight(node);
1894 }
1895 current->il_weight--;
1896 return node;
1897 }
1898
1899 /* Do dynamic interleaving for a process */
interleave_nodes(struct mempolicy * policy)1900 static unsigned int interleave_nodes(struct mempolicy *policy)
1901 {
1902 unsigned int nid;
1903 unsigned int cpuset_mems_cookie;
1904
1905 /* to prevent miscount, use tsk->mems_allowed_seq to detect rebind */
1906 do {
1907 cpuset_mems_cookie = read_mems_allowed_begin();
1908 nid = next_node_in(current->il_prev, policy->nodes);
1909 } while (read_mems_allowed_retry(cpuset_mems_cookie));
1910
1911 if (nid < MAX_NUMNODES)
1912 current->il_prev = nid;
1913 return nid;
1914 }
1915
1916 /*
1917 * Depending on the memory policy provide a node from which to allocate the
1918 * next slab entry.
1919 */
mempolicy_slab_node(void)1920 unsigned int mempolicy_slab_node(void)
1921 {
1922 struct mempolicy *policy;
1923 int node = numa_mem_id();
1924
1925 if (!in_task())
1926 return node;
1927
1928 policy = current->mempolicy;
1929 if (!policy)
1930 return node;
1931
1932 switch (policy->mode) {
1933 case MPOL_PREFERRED:
1934 return first_node(policy->nodes);
1935
1936 case MPOL_INTERLEAVE:
1937 return interleave_nodes(policy);
1938
1939 case MPOL_WEIGHTED_INTERLEAVE:
1940 return weighted_interleave_nodes(policy);
1941
1942 case MPOL_BIND:
1943 case MPOL_PREFERRED_MANY:
1944 {
1945 struct zoneref *z;
1946
1947 /*
1948 * Follow bind policy behavior and start allocation at the
1949 * first node.
1950 */
1951 struct zonelist *zonelist;
1952 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
1953 zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK];
1954 z = first_zones_zonelist(zonelist, highest_zoneidx,
1955 &policy->nodes);
1956 return zonelist_zone(z) ? zonelist_node_idx(z) : node;
1957 }
1958 case MPOL_LOCAL:
1959 return node;
1960
1961 default:
1962 BUG();
1963 }
1964 }
1965
read_once_policy_nodemask(struct mempolicy * pol,nodemask_t * mask)1966 static unsigned int read_once_policy_nodemask(struct mempolicy *pol,
1967 nodemask_t *mask)
1968 {
1969 /*
1970 * barrier stabilizes the nodemask locally so that it can be iterated
1971 * over safely without concern for changes. Allocators validate node
1972 * selection does not violate mems_allowed, so this is safe.
1973 */
1974 barrier();
1975 memcpy(mask, &pol->nodes, sizeof(nodemask_t));
1976 barrier();
1977 return nodes_weight(*mask);
1978 }
1979
weighted_interleave_nid(struct mempolicy * pol,pgoff_t ilx)1980 static unsigned int weighted_interleave_nid(struct mempolicy *pol, pgoff_t ilx)
1981 {
1982 nodemask_t nodemask;
1983 unsigned int target, nr_nodes;
1984 u8 *table;
1985 unsigned int weight_total = 0;
1986 u8 weight;
1987 int nid;
1988
1989 nr_nodes = read_once_policy_nodemask(pol, &nodemask);
1990 if (!nr_nodes)
1991 return numa_node_id();
1992
1993 rcu_read_lock();
1994 table = rcu_dereference(iw_table);
1995 /* calculate the total weight */
1996 for_each_node_mask(nid, nodemask) {
1997 /* detect system default usage */
1998 weight = table ? table[nid] : 1;
1999 weight = weight ? weight : 1;
2000 weight_total += weight;
2001 }
2002
2003 /* Calculate the node offset based on totals */
2004 target = ilx % weight_total;
2005 nid = first_node(nodemask);
2006 while (target) {
2007 /* detect system default usage */
2008 weight = table ? table[nid] : 1;
2009 weight = weight ? weight : 1;
2010 if (target < weight)
2011 break;
2012 target -= weight;
2013 nid = next_node_in(nid, nodemask);
2014 }
2015 rcu_read_unlock();
2016 return nid;
2017 }
2018
2019 /*
2020 * Do static interleaving for interleave index @ilx. Returns the ilx'th
2021 * node in pol->nodes (starting from ilx=0), wrapping around if ilx
2022 * exceeds the number of present nodes.
2023 */
interleave_nid(struct mempolicy * pol,pgoff_t ilx)2024 static unsigned int interleave_nid(struct mempolicy *pol, pgoff_t ilx)
2025 {
2026 nodemask_t nodemask;
2027 unsigned int target, nnodes;
2028 int i;
2029 int nid;
2030
2031 nnodes = read_once_policy_nodemask(pol, &nodemask);
2032 if (!nnodes)
2033 return numa_node_id();
2034 target = ilx % nnodes;
2035 nid = first_node(nodemask);
2036 for (i = 0; i < target; i++)
2037 nid = next_node(nid, nodemask);
2038 return nid;
2039 }
2040
2041 /*
2042 * Return a nodemask representing a mempolicy for filtering nodes for
2043 * page allocation, together with preferred node id (or the input node id).
2044 */
policy_nodemask(gfp_t gfp,struct mempolicy * pol,pgoff_t ilx,int * nid)2045 static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *pol,
2046 pgoff_t ilx, int *nid)
2047 {
2048 nodemask_t *nodemask = NULL;
2049
2050 switch (pol->mode) {
2051 case MPOL_PREFERRED:
2052 /* Override input node id */
2053 *nid = first_node(pol->nodes);
2054 break;
2055 case MPOL_PREFERRED_MANY:
2056 nodemask = &pol->nodes;
2057 if (pol->home_node != NUMA_NO_NODE)
2058 *nid = pol->home_node;
2059 break;
2060 case MPOL_BIND:
2061 /* Restrict to nodemask (but not on lower zones) */
2062 if (apply_policy_zone(pol, gfp_zone(gfp)) &&
2063 cpuset_nodemask_valid_mems_allowed(&pol->nodes))
2064 nodemask = &pol->nodes;
2065 if (pol->home_node != NUMA_NO_NODE)
2066 *nid = pol->home_node;
2067 /*
2068 * __GFP_THISNODE shouldn't even be used with the bind policy
2069 * because we might easily break the expectation to stay on the
2070 * requested node and not break the policy.
2071 */
2072 WARN_ON_ONCE(gfp & __GFP_THISNODE);
2073 break;
2074 case MPOL_INTERLEAVE:
2075 /* Override input node id */
2076 *nid = (ilx == NO_INTERLEAVE_INDEX) ?
2077 interleave_nodes(pol) : interleave_nid(pol, ilx);
2078 break;
2079 case MPOL_WEIGHTED_INTERLEAVE:
2080 *nid = (ilx == NO_INTERLEAVE_INDEX) ?
2081 weighted_interleave_nodes(pol) :
2082 weighted_interleave_nid(pol, ilx);
2083 break;
2084 }
2085
2086 return nodemask;
2087 }
2088
2089 #ifdef CONFIG_HUGETLBFS
2090 /*
2091 * huge_node(@vma, @addr, @gfp_flags, @mpol)
2092 * @vma: virtual memory area whose policy is sought
2093 * @addr: address in @vma for shared policy lookup and interleave policy
2094 * @gfp_flags: for requested zone
2095 * @mpol: pointer to mempolicy pointer for reference counted mempolicy
2096 * @nodemask: pointer to nodemask pointer for 'bind' and 'prefer-many' policy
2097 *
2098 * Returns a nid suitable for a huge page allocation and a pointer
2099 * to the struct mempolicy for conditional unref after allocation.
2100 * If the effective policy is 'bind' or 'prefer-many', returns a pointer
2101 * to the mempolicy's @nodemask for filtering the zonelist.
2102 */
huge_node(struct vm_area_struct * vma,unsigned long addr,gfp_t gfp_flags,struct mempolicy ** mpol,nodemask_t ** nodemask)2103 int huge_node(struct vm_area_struct *vma, unsigned long addr, gfp_t gfp_flags,
2104 struct mempolicy **mpol, nodemask_t **nodemask)
2105 {
2106 pgoff_t ilx;
2107 int nid;
2108
2109 nid = numa_node_id();
2110 *mpol = get_vma_policy(vma, addr, hstate_vma(vma)->order, &ilx);
2111 *nodemask = policy_nodemask(gfp_flags, *mpol, ilx, &nid);
2112 return nid;
2113 }
2114
2115 /*
2116 * init_nodemask_of_mempolicy
2117 *
2118 * If the current task's mempolicy is "default" [NULL], return 'false'
2119 * to indicate default policy. Otherwise, extract the policy nodemask
2120 * for 'bind' or 'interleave' policy into the argument nodemask, or
2121 * initialize the argument nodemask to contain the single node for
2122 * 'preferred' or 'local' policy and return 'true' to indicate presence
2123 * of non-default mempolicy.
2124 *
2125 * We don't bother with reference counting the mempolicy [mpol_get/put]
2126 * because the current task is examining it's own mempolicy and a task's
2127 * mempolicy is only ever changed by the task itself.
2128 *
2129 * N.B., it is the caller's responsibility to free a returned nodemask.
2130 */
init_nodemask_of_mempolicy(nodemask_t * mask)2131 bool init_nodemask_of_mempolicy(nodemask_t *mask)
2132 {
2133 struct mempolicy *mempolicy;
2134
2135 if (!(mask && current->mempolicy))
2136 return false;
2137
2138 task_lock(current);
2139 mempolicy = current->mempolicy;
2140 switch (mempolicy->mode) {
2141 case MPOL_PREFERRED:
2142 case MPOL_PREFERRED_MANY:
2143 case MPOL_BIND:
2144 case MPOL_INTERLEAVE:
2145 case MPOL_WEIGHTED_INTERLEAVE:
2146 *mask = mempolicy->nodes;
2147 break;
2148
2149 case MPOL_LOCAL:
2150 init_nodemask_of_node(mask, numa_node_id());
2151 break;
2152
2153 default:
2154 BUG();
2155 }
2156 task_unlock(current);
2157
2158 return true;
2159 }
2160 #endif
2161
2162 /*
2163 * mempolicy_in_oom_domain
2164 *
2165 * If tsk's mempolicy is "bind", check for intersection between mask and
2166 * the policy nodemask. Otherwise, return true for all other policies
2167 * including "interleave", as a tsk with "interleave" policy may have
2168 * memory allocated from all nodes in system.
2169 *
2170 * Takes task_lock(tsk) to prevent freeing of its mempolicy.
2171 */
mempolicy_in_oom_domain(struct task_struct * tsk,const nodemask_t * mask)2172 bool mempolicy_in_oom_domain(struct task_struct *tsk,
2173 const nodemask_t *mask)
2174 {
2175 struct mempolicy *mempolicy;
2176 bool ret = true;
2177
2178 if (!mask)
2179 return ret;
2180
2181 task_lock(tsk);
2182 mempolicy = tsk->mempolicy;
2183 if (mempolicy && mempolicy->mode == MPOL_BIND)
2184 ret = nodes_intersects(mempolicy->nodes, *mask);
2185 task_unlock(tsk);
2186
2187 return ret;
2188 }
2189
alloc_pages_preferred_many(gfp_t gfp,unsigned int order,int nid,nodemask_t * nodemask)2190 static struct page *alloc_pages_preferred_many(gfp_t gfp, unsigned int order,
2191 int nid, nodemask_t *nodemask)
2192 {
2193 struct page *page;
2194 gfp_t preferred_gfp;
2195
2196 /*
2197 * This is a two pass approach. The first pass will only try the
2198 * preferred nodes but skip the direct reclaim and allow the
2199 * allocation to fail, while the second pass will try all the
2200 * nodes in system.
2201 */
2202 preferred_gfp = gfp | __GFP_NOWARN;
2203 preferred_gfp &= ~(__GFP_DIRECT_RECLAIM | __GFP_NOFAIL);
2204 page = __alloc_pages_noprof(preferred_gfp, order, nid, nodemask);
2205 if (!page)
2206 page = __alloc_pages_noprof(gfp, order, nid, NULL);
2207
2208 return page;
2209 }
2210
2211 /**
2212 * alloc_pages_mpol - Allocate pages according to NUMA mempolicy.
2213 * @gfp: GFP flags.
2214 * @order: Order of the page allocation.
2215 * @pol: Pointer to the NUMA mempolicy.
2216 * @ilx: Index for interleave mempolicy (also distinguishes alloc_pages()).
2217 * @nid: Preferred node (usually numa_node_id() but @mpol may override it).
2218 *
2219 * Return: The page on success or NULL if allocation fails.
2220 */
alloc_pages_mpol_noprof(gfp_t gfp,unsigned int order,struct mempolicy * pol,pgoff_t ilx,int nid)2221 struct page *alloc_pages_mpol_noprof(gfp_t gfp, unsigned int order,
2222 struct mempolicy *pol, pgoff_t ilx, int nid)
2223 {
2224 nodemask_t *nodemask;
2225 struct page *page;
2226
2227 nodemask = policy_nodemask(gfp, pol, ilx, &nid);
2228
2229 if (pol->mode == MPOL_PREFERRED_MANY)
2230 return alloc_pages_preferred_many(gfp, order, nid, nodemask);
2231
2232 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2233 /* filter "hugepage" allocation, unless from alloc_pages() */
2234 order == HPAGE_PMD_ORDER && ilx != NO_INTERLEAVE_INDEX) {
2235 /*
2236 * For hugepage allocation and non-interleave policy which
2237 * allows the current node (or other explicitly preferred
2238 * node) we only try to allocate from the current/preferred
2239 * node and don't fall back to other nodes, as the cost of
2240 * remote accesses would likely offset THP benefits.
2241 *
2242 * If the policy is interleave or does not allow the current
2243 * node in its nodemask, we allocate the standard way.
2244 */
2245 if (pol->mode != MPOL_INTERLEAVE &&
2246 pol->mode != MPOL_WEIGHTED_INTERLEAVE &&
2247 (!nodemask || node_isset(nid, *nodemask))) {
2248 /*
2249 * First, try to allocate THP only on local node, but
2250 * don't reclaim unnecessarily, just compact.
2251 */
2252 page = __alloc_pages_node_noprof(nid,
2253 gfp | __GFP_THISNODE | __GFP_NORETRY, order);
2254 if (page || !(gfp & __GFP_DIRECT_RECLAIM))
2255 return page;
2256 /*
2257 * If hugepage allocations are configured to always
2258 * synchronous compact or the vma has been madvised
2259 * to prefer hugepage backing, retry allowing remote
2260 * memory with both reclaim and compact as well.
2261 */
2262 }
2263 }
2264
2265 page = __alloc_pages_noprof(gfp, order, nid, nodemask);
2266
2267 if (unlikely(pol->mode == MPOL_INTERLEAVE) && page) {
2268 /* skip NUMA_INTERLEAVE_HIT update if numa stats is disabled */
2269 if (static_branch_likely(&vm_numa_stat_key) &&
2270 page_to_nid(page) == nid) {
2271 preempt_disable();
2272 __count_numa_event(page_zone(page), NUMA_INTERLEAVE_HIT);
2273 preempt_enable();
2274 }
2275 }
2276
2277 return page;
2278 }
2279
folio_alloc_mpol_noprof(gfp_t gfp,unsigned int order,struct mempolicy * pol,pgoff_t ilx,int nid)2280 struct folio *folio_alloc_mpol_noprof(gfp_t gfp, unsigned int order,
2281 struct mempolicy *pol, pgoff_t ilx, int nid)
2282 {
2283 return page_rmappable_folio(alloc_pages_mpol_noprof(gfp | __GFP_COMP,
2284 order, pol, ilx, nid));
2285 }
2286
2287 /**
2288 * vma_alloc_folio - Allocate a folio for a VMA.
2289 * @gfp: GFP flags.
2290 * @order: Order of the folio.
2291 * @vma: Pointer to VMA.
2292 * @addr: Virtual address of the allocation. Must be inside @vma.
2293 * @hugepage: Unused (was: For hugepages try only preferred node if possible).
2294 *
2295 * Allocate a folio for a specific address in @vma, using the appropriate
2296 * NUMA policy. The caller must hold the mmap_lock of the mm_struct of the
2297 * VMA to prevent it from going away. Should be used for all allocations
2298 * for folios that will be mapped into user space, excepting hugetlbfs, and
2299 * excepting where direct use of alloc_pages_mpol() is more appropriate.
2300 *
2301 * Return: The folio on success or NULL if allocation fails.
2302 */
vma_alloc_folio_noprof(gfp_t gfp,int order,struct vm_area_struct * vma,unsigned long addr,bool hugepage)2303 struct folio *vma_alloc_folio_noprof(gfp_t gfp, int order, struct vm_area_struct *vma,
2304 unsigned long addr, bool hugepage)
2305 {
2306 struct mempolicy *pol;
2307 pgoff_t ilx;
2308 struct folio *folio;
2309
2310 if (vma->vm_flags & VM_DROPPABLE)
2311 gfp |= __GFP_NOWARN;
2312
2313 pol = get_vma_policy(vma, addr, order, &ilx);
2314 folio = folio_alloc_mpol_noprof(gfp, order, pol, ilx, numa_node_id());
2315 mpol_cond_put(pol);
2316 return folio;
2317 }
2318 EXPORT_SYMBOL(vma_alloc_folio_noprof);
2319
2320 /**
2321 * alloc_pages - Allocate pages.
2322 * @gfp: GFP flags.
2323 * @order: Power of two of number of pages to allocate.
2324 *
2325 * Allocate 1 << @order contiguous pages. The physical address of the
2326 * first page is naturally aligned (eg an order-3 allocation will be aligned
2327 * to a multiple of 8 * PAGE_SIZE bytes). The NUMA policy of the current
2328 * process is honoured when in process context.
2329 *
2330 * Context: Can be called from any context, providing the appropriate GFP
2331 * flags are used.
2332 * Return: The page on success or NULL if allocation fails.
2333 */
alloc_pages_noprof(gfp_t gfp,unsigned int order)2334 struct page *alloc_pages_noprof(gfp_t gfp, unsigned int order)
2335 {
2336 struct mempolicy *pol = &default_policy;
2337
2338 /*
2339 * No reference counting needed for current->mempolicy
2340 * nor system default_policy
2341 */
2342 if (!in_interrupt() && !(gfp & __GFP_THISNODE))
2343 pol = get_task_policy(current);
2344
2345 return alloc_pages_mpol_noprof(gfp, order, pol, NO_INTERLEAVE_INDEX,
2346 numa_node_id());
2347 }
2348 EXPORT_SYMBOL(alloc_pages_noprof);
2349
folio_alloc_noprof(gfp_t gfp,unsigned int order)2350 struct folio *folio_alloc_noprof(gfp_t gfp, unsigned int order)
2351 {
2352 return page_rmappable_folio(alloc_pages_noprof(gfp | __GFP_COMP, order));
2353 }
2354 EXPORT_SYMBOL(folio_alloc_noprof);
2355
alloc_pages_bulk_array_interleave(gfp_t gfp,struct mempolicy * pol,unsigned long nr_pages,struct page ** page_array)2356 static unsigned long alloc_pages_bulk_array_interleave(gfp_t gfp,
2357 struct mempolicy *pol, unsigned long nr_pages,
2358 struct page **page_array)
2359 {
2360 int nodes;
2361 unsigned long nr_pages_per_node;
2362 int delta;
2363 int i;
2364 unsigned long nr_allocated;
2365 unsigned long total_allocated = 0;
2366
2367 nodes = nodes_weight(pol->nodes);
2368 nr_pages_per_node = nr_pages / nodes;
2369 delta = nr_pages - nodes * nr_pages_per_node;
2370
2371 for (i = 0; i < nodes; i++) {
2372 if (delta) {
2373 nr_allocated = alloc_pages_bulk_noprof(gfp,
2374 interleave_nodes(pol), NULL,
2375 nr_pages_per_node + 1, NULL,
2376 page_array);
2377 delta--;
2378 } else {
2379 nr_allocated = alloc_pages_bulk_noprof(gfp,
2380 interleave_nodes(pol), NULL,
2381 nr_pages_per_node, NULL, page_array);
2382 }
2383
2384 page_array += nr_allocated;
2385 total_allocated += nr_allocated;
2386 }
2387
2388 return total_allocated;
2389 }
2390
alloc_pages_bulk_array_weighted_interleave(gfp_t gfp,struct mempolicy * pol,unsigned long nr_pages,struct page ** page_array)2391 static unsigned long alloc_pages_bulk_array_weighted_interleave(gfp_t gfp,
2392 struct mempolicy *pol, unsigned long nr_pages,
2393 struct page **page_array)
2394 {
2395 struct task_struct *me = current;
2396 unsigned int cpuset_mems_cookie;
2397 unsigned long total_allocated = 0;
2398 unsigned long nr_allocated = 0;
2399 unsigned long rounds;
2400 unsigned long node_pages, delta;
2401 u8 *table, *weights, weight;
2402 unsigned int weight_total = 0;
2403 unsigned long rem_pages = nr_pages;
2404 nodemask_t nodes;
2405 int nnodes, node;
2406 int resume_node = MAX_NUMNODES - 1;
2407 u8 resume_weight = 0;
2408 int prev_node;
2409 int i;
2410
2411 if (!nr_pages)
2412 return 0;
2413
2414 /* read the nodes onto the stack, retry if done during rebind */
2415 do {
2416 cpuset_mems_cookie = read_mems_allowed_begin();
2417 nnodes = read_once_policy_nodemask(pol, &nodes);
2418 } while (read_mems_allowed_retry(cpuset_mems_cookie));
2419
2420 /* if the nodemask has become invalid, we cannot do anything */
2421 if (!nnodes)
2422 return 0;
2423
2424 /* Continue allocating from most recent node and adjust the nr_pages */
2425 node = me->il_prev;
2426 weight = me->il_weight;
2427 if (weight && node_isset(node, nodes)) {
2428 node_pages = min(rem_pages, weight);
2429 nr_allocated = __alloc_pages_bulk(gfp, node, NULL, node_pages,
2430 NULL, page_array);
2431 page_array += nr_allocated;
2432 total_allocated += nr_allocated;
2433 /* if that's all the pages, no need to interleave */
2434 if (rem_pages <= weight) {
2435 me->il_weight -= rem_pages;
2436 return total_allocated;
2437 }
2438 /* Otherwise we adjust remaining pages, continue from there */
2439 rem_pages -= weight;
2440 }
2441 /* clear active weight in case of an allocation failure */
2442 me->il_weight = 0;
2443 prev_node = node;
2444
2445 /* create a local copy of node weights to operate on outside rcu */
2446 weights = kzalloc(nr_node_ids, GFP_KERNEL);
2447 if (!weights)
2448 return total_allocated;
2449
2450 rcu_read_lock();
2451 table = rcu_dereference(iw_table);
2452 if (table)
2453 memcpy(weights, table, nr_node_ids);
2454 rcu_read_unlock();
2455
2456 /* calculate total, detect system default usage */
2457 for_each_node_mask(node, nodes) {
2458 if (!weights[node])
2459 weights[node] = 1;
2460 weight_total += weights[node];
2461 }
2462
2463 /*
2464 * Calculate rounds/partial rounds to minimize __alloc_pages_bulk calls.
2465 * Track which node weighted interleave should resume from.
2466 *
2467 * if (rounds > 0) and (delta == 0), resume_node will always be
2468 * the node following prev_node and its weight.
2469 */
2470 rounds = rem_pages / weight_total;
2471 delta = rem_pages % weight_total;
2472 resume_node = next_node_in(prev_node, nodes);
2473 resume_weight = weights[resume_node];
2474 for (i = 0; i < nnodes; i++) {
2475 node = next_node_in(prev_node, nodes);
2476 weight = weights[node];
2477 node_pages = weight * rounds;
2478 /* If a delta exists, add this node's portion of the delta */
2479 if (delta > weight) {
2480 node_pages += weight;
2481 delta -= weight;
2482 } else if (delta) {
2483 /* when delta is depleted, resume from that node */
2484 node_pages += delta;
2485 resume_node = node;
2486 resume_weight = weight - delta;
2487 delta = 0;
2488 }
2489 /* node_pages can be 0 if an allocation fails and rounds == 0 */
2490 if (!node_pages)
2491 break;
2492 nr_allocated = __alloc_pages_bulk(gfp, node, NULL, node_pages,
2493 NULL, page_array);
2494 page_array += nr_allocated;
2495 total_allocated += nr_allocated;
2496 if (total_allocated == nr_pages)
2497 break;
2498 prev_node = node;
2499 }
2500 me->il_prev = resume_node;
2501 me->il_weight = resume_weight;
2502 kfree(weights);
2503 return total_allocated;
2504 }
2505
alloc_pages_bulk_array_preferred_many(gfp_t gfp,int nid,struct mempolicy * pol,unsigned long nr_pages,struct page ** page_array)2506 static unsigned long alloc_pages_bulk_array_preferred_many(gfp_t gfp, int nid,
2507 struct mempolicy *pol, unsigned long nr_pages,
2508 struct page **page_array)
2509 {
2510 gfp_t preferred_gfp;
2511 unsigned long nr_allocated = 0;
2512
2513 preferred_gfp = gfp | __GFP_NOWARN;
2514 preferred_gfp &= ~(__GFP_DIRECT_RECLAIM | __GFP_NOFAIL);
2515
2516 nr_allocated = alloc_pages_bulk_noprof(preferred_gfp, nid, &pol->nodes,
2517 nr_pages, NULL, page_array);
2518
2519 if (nr_allocated < nr_pages)
2520 nr_allocated += alloc_pages_bulk_noprof(gfp, numa_node_id(), NULL,
2521 nr_pages - nr_allocated, NULL,
2522 page_array + nr_allocated);
2523 return nr_allocated;
2524 }
2525
2526 /* alloc pages bulk and mempolicy should be considered at the
2527 * same time in some situation such as vmalloc.
2528 *
2529 * It can accelerate memory allocation especially interleaving
2530 * allocate memory.
2531 */
alloc_pages_bulk_array_mempolicy_noprof(gfp_t gfp,unsigned long nr_pages,struct page ** page_array)2532 unsigned long alloc_pages_bulk_array_mempolicy_noprof(gfp_t gfp,
2533 unsigned long nr_pages, struct page **page_array)
2534 {
2535 struct mempolicy *pol = &default_policy;
2536 nodemask_t *nodemask;
2537 int nid;
2538
2539 if (!in_interrupt() && !(gfp & __GFP_THISNODE))
2540 pol = get_task_policy(current);
2541
2542 if (pol->mode == MPOL_INTERLEAVE)
2543 return alloc_pages_bulk_array_interleave(gfp, pol,
2544 nr_pages, page_array);
2545
2546 if (pol->mode == MPOL_WEIGHTED_INTERLEAVE)
2547 return alloc_pages_bulk_array_weighted_interleave(
2548 gfp, pol, nr_pages, page_array);
2549
2550 if (pol->mode == MPOL_PREFERRED_MANY)
2551 return alloc_pages_bulk_array_preferred_many(gfp,
2552 numa_node_id(), pol, nr_pages, page_array);
2553
2554 nid = numa_node_id();
2555 nodemask = policy_nodemask(gfp, pol, NO_INTERLEAVE_INDEX, &nid);
2556 return alloc_pages_bulk_noprof(gfp, nid, nodemask,
2557 nr_pages, NULL, page_array);
2558 }
2559
vma_dup_policy(struct vm_area_struct * src,struct vm_area_struct * dst)2560 int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
2561 {
2562 struct mempolicy *pol = mpol_dup(src->vm_policy);
2563
2564 if (IS_ERR(pol))
2565 return PTR_ERR(pol);
2566 dst->vm_policy = pol;
2567 return 0;
2568 }
2569
2570 /*
2571 * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
2572 * rebinds the mempolicy its copying by calling mpol_rebind_policy()
2573 * with the mems_allowed returned by cpuset_mems_allowed(). This
2574 * keeps mempolicies cpuset relative after its cpuset moves. See
2575 * further kernel/cpuset.c update_nodemask().
2576 *
2577 * current's mempolicy may be rebinded by the other task(the task that changes
2578 * cpuset's mems), so we needn't do rebind work for current task.
2579 */
2580
2581 /* Slow path of a mempolicy duplicate */
__mpol_dup(struct mempolicy * old)2582 struct mempolicy *__mpol_dup(struct mempolicy *old)
2583 {
2584 struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2585
2586 if (!new)
2587 return ERR_PTR(-ENOMEM);
2588
2589 /* task's mempolicy is protected by alloc_lock */
2590 if (old == current->mempolicy) {
2591 task_lock(current);
2592 *new = *old;
2593 task_unlock(current);
2594 } else
2595 *new = *old;
2596
2597 if (current_cpuset_is_being_rebound()) {
2598 nodemask_t mems = cpuset_mems_allowed(current);
2599 mpol_rebind_policy(new, &mems);
2600 }
2601 atomic_set(&new->refcnt, 1);
2602 return new;
2603 }
2604
2605 /* Slow path of a mempolicy comparison */
__mpol_equal(struct mempolicy * a,struct mempolicy * b)2606 bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
2607 {
2608 if (!a || !b)
2609 return false;
2610 if (a->mode != b->mode)
2611 return false;
2612 if (a->flags != b->flags)
2613 return false;
2614 if (a->home_node != b->home_node)
2615 return false;
2616 if (mpol_store_user_nodemask(a))
2617 if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask))
2618 return false;
2619
2620 switch (a->mode) {
2621 case MPOL_BIND:
2622 case MPOL_INTERLEAVE:
2623 case MPOL_PREFERRED:
2624 case MPOL_PREFERRED_MANY:
2625 case MPOL_WEIGHTED_INTERLEAVE:
2626 return !!nodes_equal(a->nodes, b->nodes);
2627 case MPOL_LOCAL:
2628 return true;
2629 default:
2630 BUG();
2631 return false;
2632 }
2633 }
2634
2635 /*
2636 * Shared memory backing store policy support.
2637 *
2638 * Remember policies even when nobody has shared memory mapped.
2639 * The policies are kept in Red-Black tree linked from the inode.
2640 * They are protected by the sp->lock rwlock, which should be held
2641 * for any accesses to the tree.
2642 */
2643
2644 /*
2645 * lookup first element intersecting start-end. Caller holds sp->lock for
2646 * reading or for writing
2647 */
sp_lookup(struct shared_policy * sp,pgoff_t start,pgoff_t end)2648 static struct sp_node *sp_lookup(struct shared_policy *sp,
2649 pgoff_t start, pgoff_t end)
2650 {
2651 struct rb_node *n = sp->root.rb_node;
2652
2653 while (n) {
2654 struct sp_node *p = rb_entry(n, struct sp_node, nd);
2655
2656 if (start >= p->end)
2657 n = n->rb_right;
2658 else if (end <= p->start)
2659 n = n->rb_left;
2660 else
2661 break;
2662 }
2663 if (!n)
2664 return NULL;
2665 for (;;) {
2666 struct sp_node *w = NULL;
2667 struct rb_node *prev = rb_prev(n);
2668 if (!prev)
2669 break;
2670 w = rb_entry(prev, struct sp_node, nd);
2671 if (w->end <= start)
2672 break;
2673 n = prev;
2674 }
2675 return rb_entry(n, struct sp_node, nd);
2676 }
2677
2678 /*
2679 * Insert a new shared policy into the list. Caller holds sp->lock for
2680 * writing.
2681 */
sp_insert(struct shared_policy * sp,struct sp_node * new)2682 static void sp_insert(struct shared_policy *sp, struct sp_node *new)
2683 {
2684 struct rb_node **p = &sp->root.rb_node;
2685 struct rb_node *parent = NULL;
2686 struct sp_node *nd;
2687
2688 while (*p) {
2689 parent = *p;
2690 nd = rb_entry(parent, struct sp_node, nd);
2691 if (new->start < nd->start)
2692 p = &(*p)->rb_left;
2693 else if (new->end > nd->end)
2694 p = &(*p)->rb_right;
2695 else
2696 BUG();
2697 }
2698 rb_link_node(&new->nd, parent, p);
2699 rb_insert_color(&new->nd, &sp->root);
2700 }
2701
2702 /* Find shared policy intersecting idx */
mpol_shared_policy_lookup(struct shared_policy * sp,pgoff_t idx)2703 struct mempolicy *mpol_shared_policy_lookup(struct shared_policy *sp,
2704 pgoff_t idx)
2705 {
2706 struct mempolicy *pol = NULL;
2707 struct sp_node *sn;
2708
2709 if (!sp->root.rb_node)
2710 return NULL;
2711 read_lock(&sp->lock);
2712 sn = sp_lookup(sp, idx, idx+1);
2713 if (sn) {
2714 mpol_get(sn->policy);
2715 pol = sn->policy;
2716 }
2717 read_unlock(&sp->lock);
2718 return pol;
2719 }
2720
sp_free(struct sp_node * n)2721 static void sp_free(struct sp_node *n)
2722 {
2723 mpol_put(n->policy);
2724 kmem_cache_free(sn_cache, n);
2725 }
2726
2727 /**
2728 * mpol_misplaced - check whether current folio node is valid in policy
2729 *
2730 * @folio: folio to be checked
2731 * @vmf: structure describing the fault
2732 * @addr: virtual address in @vma for shared policy lookup and interleave policy
2733 *
2734 * Lookup current policy node id for vma,addr and "compare to" folio's
2735 * node id. Policy determination "mimics" alloc_page_vma().
2736 * Called from fault path where we know the vma and faulting address.
2737 *
2738 * Return: NUMA_NO_NODE if the page is in a node that is valid for this
2739 * policy, or a suitable node ID to allocate a replacement folio from.
2740 */
mpol_misplaced(struct folio * folio,struct vm_fault * vmf,unsigned long addr)2741 int mpol_misplaced(struct folio *folio, struct vm_fault *vmf,
2742 unsigned long addr)
2743 {
2744 struct mempolicy *pol;
2745 pgoff_t ilx;
2746 struct zoneref *z;
2747 int curnid = folio_nid(folio);
2748 struct vm_area_struct *vma = vmf->vma;
2749 int thiscpu = raw_smp_processor_id();
2750 int thisnid = numa_node_id();
2751 int polnid = NUMA_NO_NODE;
2752 int ret = NUMA_NO_NODE;
2753
2754 /*
2755 * Make sure ptl is held so that we don't preempt and we
2756 * have a stable smp processor id
2757 */
2758 lockdep_assert_held(vmf->ptl);
2759 pol = get_vma_policy(vma, addr, folio_order(folio), &ilx);
2760 if (!(pol->flags & MPOL_F_MOF))
2761 goto out;
2762
2763 switch (pol->mode) {
2764 case MPOL_INTERLEAVE:
2765 polnid = interleave_nid(pol, ilx);
2766 break;
2767
2768 case MPOL_WEIGHTED_INTERLEAVE:
2769 polnid = weighted_interleave_nid(pol, ilx);
2770 break;
2771
2772 case MPOL_PREFERRED:
2773 if (node_isset(curnid, pol->nodes))
2774 goto out;
2775 polnid = first_node(pol->nodes);
2776 break;
2777
2778 case MPOL_LOCAL:
2779 polnid = numa_node_id();
2780 break;
2781
2782 case MPOL_BIND:
2783 case MPOL_PREFERRED_MANY:
2784 /*
2785 * Even though MPOL_PREFERRED_MANY can allocate pages outside
2786 * policy nodemask we don't allow numa migration to nodes
2787 * outside policy nodemask for now. This is done so that if we
2788 * want demotion to slow memory to happen, before allocating
2789 * from some DRAM node say 'x', we will end up using a
2790 * MPOL_PREFERRED_MANY mask excluding node 'x'. In such scenario
2791 * we should not promote to node 'x' from slow memory node.
2792 */
2793 if (pol->flags & MPOL_F_MORON) {
2794 /*
2795 * Optimize placement among multiple nodes
2796 * via NUMA balancing
2797 */
2798 if (node_isset(thisnid, pol->nodes))
2799 break;
2800 goto out;
2801 }
2802
2803 /*
2804 * use current page if in policy nodemask,
2805 * else select nearest allowed node, if any.
2806 * If no allowed nodes, use current [!misplaced].
2807 */
2808 if (node_isset(curnid, pol->nodes))
2809 goto out;
2810 z = first_zones_zonelist(
2811 node_zonelist(thisnid, GFP_HIGHUSER),
2812 gfp_zone(GFP_HIGHUSER),
2813 &pol->nodes);
2814 polnid = zonelist_node_idx(z);
2815 break;
2816
2817 default:
2818 BUG();
2819 }
2820
2821 /* Migrate the folio towards the node whose CPU is referencing it */
2822 if (pol->flags & MPOL_F_MORON) {
2823 polnid = thisnid;
2824
2825 if (!should_numa_migrate_memory(current, folio, curnid,
2826 thiscpu))
2827 goto out;
2828 }
2829
2830 if (curnid != polnid)
2831 ret = polnid;
2832 out:
2833 mpol_cond_put(pol);
2834
2835 return ret;
2836 }
2837
2838 /*
2839 * Drop the (possibly final) reference to task->mempolicy. It needs to be
2840 * dropped after task->mempolicy is set to NULL so that any allocation done as
2841 * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed
2842 * policy.
2843 */
mpol_put_task_policy(struct task_struct * task)2844 void mpol_put_task_policy(struct task_struct *task)
2845 {
2846 struct mempolicy *pol;
2847
2848 task_lock(task);
2849 pol = task->mempolicy;
2850 task->mempolicy = NULL;
2851 task_unlock(task);
2852 mpol_put(pol);
2853 }
2854
sp_delete(struct shared_policy * sp,struct sp_node * n)2855 static void sp_delete(struct shared_policy *sp, struct sp_node *n)
2856 {
2857 rb_erase(&n->nd, &sp->root);
2858 sp_free(n);
2859 }
2860
sp_node_init(struct sp_node * node,unsigned long start,unsigned long end,struct mempolicy * pol)2861 static void sp_node_init(struct sp_node *node, unsigned long start,
2862 unsigned long end, struct mempolicy *pol)
2863 {
2864 node->start = start;
2865 node->end = end;
2866 node->policy = pol;
2867 }
2868
sp_alloc(unsigned long start,unsigned long end,struct mempolicy * pol)2869 static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
2870 struct mempolicy *pol)
2871 {
2872 struct sp_node *n;
2873 struct mempolicy *newpol;
2874
2875 n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
2876 if (!n)
2877 return NULL;
2878
2879 newpol = mpol_dup(pol);
2880 if (IS_ERR(newpol)) {
2881 kmem_cache_free(sn_cache, n);
2882 return NULL;
2883 }
2884 newpol->flags |= MPOL_F_SHARED;
2885 sp_node_init(n, start, end, newpol);
2886
2887 return n;
2888 }
2889
2890 /* Replace a policy range. */
shared_policy_replace(struct shared_policy * sp,pgoff_t start,pgoff_t end,struct sp_node * new)2891 static int shared_policy_replace(struct shared_policy *sp, pgoff_t start,
2892 pgoff_t end, struct sp_node *new)
2893 {
2894 struct sp_node *n;
2895 struct sp_node *n_new = NULL;
2896 struct mempolicy *mpol_new = NULL;
2897 int ret = 0;
2898
2899 restart:
2900 write_lock(&sp->lock);
2901 n = sp_lookup(sp, start, end);
2902 /* Take care of old policies in the same range. */
2903 while (n && n->start < end) {
2904 struct rb_node *next = rb_next(&n->nd);
2905 if (n->start >= start) {
2906 if (n->end <= end)
2907 sp_delete(sp, n);
2908 else
2909 n->start = end;
2910 } else {
2911 /* Old policy spanning whole new range. */
2912 if (n->end > end) {
2913 if (!n_new)
2914 goto alloc_new;
2915
2916 *mpol_new = *n->policy;
2917 atomic_set(&mpol_new->refcnt, 1);
2918 sp_node_init(n_new, end, n->end, mpol_new);
2919 n->end = start;
2920 sp_insert(sp, n_new);
2921 n_new = NULL;
2922 mpol_new = NULL;
2923 break;
2924 } else
2925 n->end = start;
2926 }
2927 if (!next)
2928 break;
2929 n = rb_entry(next, struct sp_node, nd);
2930 }
2931 if (new)
2932 sp_insert(sp, new);
2933 write_unlock(&sp->lock);
2934 ret = 0;
2935
2936 err_out:
2937 if (mpol_new)
2938 mpol_put(mpol_new);
2939 if (n_new)
2940 kmem_cache_free(sn_cache, n_new);
2941
2942 return ret;
2943
2944 alloc_new:
2945 write_unlock(&sp->lock);
2946 ret = -ENOMEM;
2947 n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL);
2948 if (!n_new)
2949 goto err_out;
2950 mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2951 if (!mpol_new)
2952 goto err_out;
2953 atomic_set(&mpol_new->refcnt, 1);
2954 goto restart;
2955 }
2956
2957 /**
2958 * mpol_shared_policy_init - initialize shared policy for inode
2959 * @sp: pointer to inode shared policy
2960 * @mpol: struct mempolicy to install
2961 *
2962 * Install non-NULL @mpol in inode's shared policy rb-tree.
2963 * On entry, the current task has a reference on a non-NULL @mpol.
2964 * This must be released on exit.
2965 * This is called at get_inode() calls and we can use GFP_KERNEL.
2966 */
mpol_shared_policy_init(struct shared_policy * sp,struct mempolicy * mpol)2967 void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
2968 {
2969 int ret;
2970
2971 sp->root = RB_ROOT; /* empty tree == default mempolicy */
2972 rwlock_init(&sp->lock);
2973
2974 if (mpol) {
2975 struct sp_node *sn;
2976 struct mempolicy *npol;
2977 NODEMASK_SCRATCH(scratch);
2978
2979 if (!scratch)
2980 goto put_mpol;
2981
2982 /* contextualize the tmpfs mount point mempolicy to this file */
2983 npol = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
2984 if (IS_ERR(npol))
2985 goto free_scratch; /* no valid nodemask intersection */
2986
2987 task_lock(current);
2988 ret = mpol_set_nodemask(npol, &mpol->w.user_nodemask, scratch);
2989 task_unlock(current);
2990 if (ret)
2991 goto put_npol;
2992
2993 /* alloc node covering entire file; adds ref to file's npol */
2994 sn = sp_alloc(0, MAX_LFS_FILESIZE >> PAGE_SHIFT, npol);
2995 if (sn)
2996 sp_insert(sp, sn);
2997 put_npol:
2998 mpol_put(npol); /* drop initial ref on file's npol */
2999 free_scratch:
3000 NODEMASK_SCRATCH_FREE(scratch);
3001 put_mpol:
3002 mpol_put(mpol); /* drop our incoming ref on sb mpol */
3003 }
3004 }
3005
mpol_set_shared_policy(struct shared_policy * sp,struct vm_area_struct * vma,struct mempolicy * pol)3006 int mpol_set_shared_policy(struct shared_policy *sp,
3007 struct vm_area_struct *vma, struct mempolicy *pol)
3008 {
3009 int err;
3010 struct sp_node *new = NULL;
3011 unsigned long sz = vma_pages(vma);
3012
3013 if (pol) {
3014 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, pol);
3015 if (!new)
3016 return -ENOMEM;
3017 }
3018 err = shared_policy_replace(sp, vma->vm_pgoff, vma->vm_pgoff + sz, new);
3019 if (err && new)
3020 sp_free(new);
3021 return err;
3022 }
3023
3024 /* Free a backing policy store on inode delete. */
mpol_free_shared_policy(struct shared_policy * sp)3025 void mpol_free_shared_policy(struct shared_policy *sp)
3026 {
3027 struct sp_node *n;
3028 struct rb_node *next;
3029
3030 if (!sp->root.rb_node)
3031 return;
3032 write_lock(&sp->lock);
3033 next = rb_first(&sp->root);
3034 while (next) {
3035 n = rb_entry(next, struct sp_node, nd);
3036 next = rb_next(&n->nd);
3037 sp_delete(sp, n);
3038 }
3039 write_unlock(&sp->lock);
3040 }
3041
3042 #ifdef CONFIG_NUMA_BALANCING
3043 static int __initdata numabalancing_override;
3044
check_numabalancing_enable(void)3045 static void __init check_numabalancing_enable(void)
3046 {
3047 bool numabalancing_default = false;
3048
3049 if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED))
3050 numabalancing_default = true;
3051
3052 /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */
3053 if (numabalancing_override)
3054 set_numabalancing_state(numabalancing_override == 1);
3055
3056 if (num_online_nodes() > 1 && !numabalancing_override) {
3057 pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n",
3058 numabalancing_default ? "Enabling" : "Disabling");
3059 set_numabalancing_state(numabalancing_default);
3060 }
3061 }
3062
setup_numabalancing(char * str)3063 static int __init setup_numabalancing(char *str)
3064 {
3065 int ret = 0;
3066 if (!str)
3067 goto out;
3068
3069 if (!strcmp(str, "enable")) {
3070 numabalancing_override = 1;
3071 ret = 1;
3072 } else if (!strcmp(str, "disable")) {
3073 numabalancing_override = -1;
3074 ret = 1;
3075 }
3076 out:
3077 if (!ret)
3078 pr_warn("Unable to parse numa_balancing=\n");
3079
3080 return ret;
3081 }
3082 __setup("numa_balancing=", setup_numabalancing);
3083 #else
check_numabalancing_enable(void)3084 static inline void __init check_numabalancing_enable(void)
3085 {
3086 }
3087 #endif /* CONFIG_NUMA_BALANCING */
3088
numa_policy_init(void)3089 void __init numa_policy_init(void)
3090 {
3091 nodemask_t interleave_nodes;
3092 unsigned long largest = 0;
3093 int nid, prefer = 0;
3094
3095 policy_cache = kmem_cache_create("numa_policy",
3096 sizeof(struct mempolicy),
3097 0, SLAB_PANIC, NULL);
3098
3099 sn_cache = kmem_cache_create("shared_policy_node",
3100 sizeof(struct sp_node),
3101 0, SLAB_PANIC, NULL);
3102
3103 for_each_node(nid) {
3104 preferred_node_policy[nid] = (struct mempolicy) {
3105 .refcnt = ATOMIC_INIT(1),
3106 .mode = MPOL_PREFERRED,
3107 .flags = MPOL_F_MOF | MPOL_F_MORON,
3108 .nodes = nodemask_of_node(nid),
3109 };
3110 }
3111
3112 /*
3113 * Set interleaving policy for system init. Interleaving is only
3114 * enabled across suitably sized nodes (default is >= 16MB), or
3115 * fall back to the largest node if they're all smaller.
3116 */
3117 nodes_clear(interleave_nodes);
3118 for_each_node_state(nid, N_MEMORY) {
3119 unsigned long total_pages = node_present_pages(nid);
3120
3121 /* Preserve the largest node */
3122 if (largest < total_pages) {
3123 largest = total_pages;
3124 prefer = nid;
3125 }
3126
3127 /* Interleave this node? */
3128 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
3129 node_set(nid, interleave_nodes);
3130 }
3131
3132 /* All too small, use the largest */
3133 if (unlikely(nodes_empty(interleave_nodes)))
3134 node_set(prefer, interleave_nodes);
3135
3136 if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
3137 pr_err("%s: interleaving failed\n", __func__);
3138
3139 check_numabalancing_enable();
3140 }
3141
3142 /* Reset policy of current process to default */
numa_default_policy(void)3143 void numa_default_policy(void)
3144 {
3145 do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
3146 }
3147
3148 /*
3149 * Parse and format mempolicy from/to strings
3150 */
3151 static const char * const policy_modes[] =
3152 {
3153 [MPOL_DEFAULT] = "default",
3154 [MPOL_PREFERRED] = "prefer",
3155 [MPOL_BIND] = "bind",
3156 [MPOL_INTERLEAVE] = "interleave",
3157 [MPOL_WEIGHTED_INTERLEAVE] = "weighted interleave",
3158 [MPOL_LOCAL] = "local",
3159 [MPOL_PREFERRED_MANY] = "prefer (many)",
3160 };
3161
3162 #ifdef CONFIG_TMPFS
3163 /**
3164 * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option.
3165 * @str: string containing mempolicy to parse
3166 * @mpol: pointer to struct mempolicy pointer, returned on success.
3167 *
3168 * Format of input:
3169 * <mode>[=<flags>][:<nodelist>]
3170 *
3171 * Return: %0 on success, else %1
3172 */
mpol_parse_str(char * str,struct mempolicy ** mpol)3173 int mpol_parse_str(char *str, struct mempolicy **mpol)
3174 {
3175 struct mempolicy *new = NULL;
3176 unsigned short mode_flags;
3177 nodemask_t nodes;
3178 char *nodelist = strchr(str, ':');
3179 char *flags = strchr(str, '=');
3180 int err = 1, mode;
3181
3182 if (flags)
3183 *flags++ = '\0'; /* terminate mode string */
3184
3185 if (nodelist) {
3186 /* NUL-terminate mode or flags string */
3187 *nodelist++ = '\0';
3188 if (nodelist_parse(nodelist, nodes))
3189 goto out;
3190 if (!nodes_subset(nodes, node_states[N_MEMORY]))
3191 goto out;
3192 } else
3193 nodes_clear(nodes);
3194
3195 mode = match_string(policy_modes, MPOL_MAX, str);
3196 if (mode < 0)
3197 goto out;
3198
3199 switch (mode) {
3200 case MPOL_PREFERRED:
3201 /*
3202 * Insist on a nodelist of one node only, although later
3203 * we use first_node(nodes) to grab a single node, so here
3204 * nodelist (or nodes) cannot be empty.
3205 */
3206 if (nodelist) {
3207 char *rest = nodelist;
3208 while (isdigit(*rest))
3209 rest++;
3210 if (*rest)
3211 goto out;
3212 if (nodes_empty(nodes))
3213 goto out;
3214 }
3215 break;
3216 case MPOL_INTERLEAVE:
3217 case MPOL_WEIGHTED_INTERLEAVE:
3218 /*
3219 * Default to online nodes with memory if no nodelist
3220 */
3221 if (!nodelist)
3222 nodes = node_states[N_MEMORY];
3223 break;
3224 case MPOL_LOCAL:
3225 /*
3226 * Don't allow a nodelist; mpol_new() checks flags
3227 */
3228 if (nodelist)
3229 goto out;
3230 break;
3231 case MPOL_DEFAULT:
3232 /*
3233 * Insist on a empty nodelist
3234 */
3235 if (!nodelist)
3236 err = 0;
3237 goto out;
3238 case MPOL_PREFERRED_MANY:
3239 case MPOL_BIND:
3240 /*
3241 * Insist on a nodelist
3242 */
3243 if (!nodelist)
3244 goto out;
3245 }
3246
3247 mode_flags = 0;
3248 if (flags) {
3249 /*
3250 * Currently, we only support two mutually exclusive
3251 * mode flags.
3252 */
3253 if (!strcmp(flags, "static"))
3254 mode_flags |= MPOL_F_STATIC_NODES;
3255 else if (!strcmp(flags, "relative"))
3256 mode_flags |= MPOL_F_RELATIVE_NODES;
3257 else
3258 goto out;
3259 }
3260
3261 new = mpol_new(mode, mode_flags, &nodes);
3262 if (IS_ERR(new))
3263 goto out;
3264
3265 /*
3266 * Save nodes for mpol_to_str() to show the tmpfs mount options
3267 * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo.
3268 */
3269 if (mode != MPOL_PREFERRED) {
3270 new->nodes = nodes;
3271 } else if (nodelist) {
3272 nodes_clear(new->nodes);
3273 node_set(first_node(nodes), new->nodes);
3274 } else {
3275 new->mode = MPOL_LOCAL;
3276 }
3277
3278 /*
3279 * Save nodes for contextualization: this will be used to "clone"
3280 * the mempolicy in a specific context [cpuset] at a later time.
3281 */
3282 new->w.user_nodemask = nodes;
3283
3284 err = 0;
3285
3286 out:
3287 /* Restore string for error message */
3288 if (nodelist)
3289 *--nodelist = ':';
3290 if (flags)
3291 *--flags = '=';
3292 if (!err)
3293 *mpol = new;
3294 return err;
3295 }
3296 #endif /* CONFIG_TMPFS */
3297
3298 /**
3299 * mpol_to_str - format a mempolicy structure for printing
3300 * @buffer: to contain formatted mempolicy string
3301 * @maxlen: length of @buffer
3302 * @pol: pointer to mempolicy to be formatted
3303 *
3304 * Convert @pol into a string. If @buffer is too short, truncate the string.
3305 * Recommend a @maxlen of at least 51 for the longest mode, "weighted
3306 * interleave", plus the longest flag flags, "relative|balancing", and to
3307 * display at least a few node ids.
3308 */
mpol_to_str(char * buffer,int maxlen,struct mempolicy * pol)3309 void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
3310 {
3311 char *p = buffer;
3312 nodemask_t nodes = NODE_MASK_NONE;
3313 unsigned short mode = MPOL_DEFAULT;
3314 unsigned short flags = 0;
3315
3316 if (pol &&
3317 pol != &default_policy &&
3318 !(pol >= &preferred_node_policy[0] &&
3319 pol <= &preferred_node_policy[ARRAY_SIZE(preferred_node_policy) - 1])) {
3320 mode = pol->mode;
3321 flags = pol->flags;
3322 }
3323
3324 switch (mode) {
3325 case MPOL_DEFAULT:
3326 case MPOL_LOCAL:
3327 break;
3328 case MPOL_PREFERRED:
3329 case MPOL_PREFERRED_MANY:
3330 case MPOL_BIND:
3331 case MPOL_INTERLEAVE:
3332 case MPOL_WEIGHTED_INTERLEAVE:
3333 nodes = pol->nodes;
3334 break;
3335 default:
3336 WARN_ON_ONCE(1);
3337 snprintf(p, maxlen, "unknown");
3338 return;
3339 }
3340
3341 p += snprintf(p, maxlen, "%s", policy_modes[mode]);
3342
3343 if (flags & MPOL_MODE_FLAGS) {
3344 p += snprintf(p, buffer + maxlen - p, "=");
3345
3346 /*
3347 * Static and relative are mutually exclusive.
3348 */
3349 if (flags & MPOL_F_STATIC_NODES)
3350 p += snprintf(p, buffer + maxlen - p, "static");
3351 else if (flags & MPOL_F_RELATIVE_NODES)
3352 p += snprintf(p, buffer + maxlen - p, "relative");
3353
3354 if (flags & MPOL_F_NUMA_BALANCING) {
3355 if (!is_power_of_2(flags & MPOL_MODE_FLAGS))
3356 p += snprintf(p, buffer + maxlen - p, "|");
3357 p += snprintf(p, buffer + maxlen - p, "balancing");
3358 }
3359 }
3360
3361 if (!nodes_empty(nodes))
3362 p += scnprintf(p, buffer + maxlen - p, ":%*pbl",
3363 nodemask_pr_args(&nodes));
3364 }
3365
3366 #ifdef CONFIG_SYSFS
3367 struct iw_node_attr {
3368 struct kobj_attribute kobj_attr;
3369 int nid;
3370 };
3371
node_show(struct kobject * kobj,struct kobj_attribute * attr,char * buf)3372 static ssize_t node_show(struct kobject *kobj, struct kobj_attribute *attr,
3373 char *buf)
3374 {
3375 struct iw_node_attr *node_attr;
3376 u8 weight;
3377
3378 node_attr = container_of(attr, struct iw_node_attr, kobj_attr);
3379 weight = get_il_weight(node_attr->nid);
3380 return sysfs_emit(buf, "%d\n", weight);
3381 }
3382
node_store(struct kobject * kobj,struct kobj_attribute * attr,const char * buf,size_t count)3383 static ssize_t node_store(struct kobject *kobj, struct kobj_attribute *attr,
3384 const char *buf, size_t count)
3385 {
3386 struct iw_node_attr *node_attr;
3387 u8 *new;
3388 u8 *old;
3389 u8 weight = 0;
3390
3391 node_attr = container_of(attr, struct iw_node_attr, kobj_attr);
3392 if (count == 0 || sysfs_streq(buf, ""))
3393 weight = 0;
3394 else if (kstrtou8(buf, 0, &weight))
3395 return -EINVAL;
3396
3397 new = kzalloc(nr_node_ids, GFP_KERNEL);
3398 if (!new)
3399 return -ENOMEM;
3400
3401 mutex_lock(&iw_table_lock);
3402 old = rcu_dereference_protected(iw_table,
3403 lockdep_is_held(&iw_table_lock));
3404 if (old)
3405 memcpy(new, old, nr_node_ids);
3406 new[node_attr->nid] = weight;
3407 rcu_assign_pointer(iw_table, new);
3408 mutex_unlock(&iw_table_lock);
3409 synchronize_rcu();
3410 kfree(old);
3411 return count;
3412 }
3413
3414 static struct iw_node_attr **node_attrs;
3415
sysfs_wi_node_release(struct iw_node_attr * node_attr,struct kobject * parent)3416 static void sysfs_wi_node_release(struct iw_node_attr *node_attr,
3417 struct kobject *parent)
3418 {
3419 if (!node_attr)
3420 return;
3421 sysfs_remove_file(parent, &node_attr->kobj_attr.attr);
3422 kfree(node_attr->kobj_attr.attr.name);
3423 kfree(node_attr);
3424 }
3425
sysfs_wi_release(struct kobject * wi_kobj)3426 static void sysfs_wi_release(struct kobject *wi_kobj)
3427 {
3428 int i;
3429
3430 for (i = 0; i < nr_node_ids; i++)
3431 sysfs_wi_node_release(node_attrs[i], wi_kobj);
3432 kobject_put(wi_kobj);
3433 }
3434
3435 static const struct kobj_type wi_ktype = {
3436 .sysfs_ops = &kobj_sysfs_ops,
3437 .release = sysfs_wi_release,
3438 };
3439
add_weight_node(int nid,struct kobject * wi_kobj)3440 static int add_weight_node(int nid, struct kobject *wi_kobj)
3441 {
3442 struct iw_node_attr *node_attr;
3443 char *name;
3444
3445 node_attr = kzalloc(sizeof(*node_attr), GFP_KERNEL);
3446 if (!node_attr)
3447 return -ENOMEM;
3448
3449 name = kasprintf(GFP_KERNEL, "node%d", nid);
3450 if (!name) {
3451 kfree(node_attr);
3452 return -ENOMEM;
3453 }
3454
3455 sysfs_attr_init(&node_attr->kobj_attr.attr);
3456 node_attr->kobj_attr.attr.name = name;
3457 node_attr->kobj_attr.attr.mode = 0644;
3458 node_attr->kobj_attr.show = node_show;
3459 node_attr->kobj_attr.store = node_store;
3460 node_attr->nid = nid;
3461
3462 if (sysfs_create_file(wi_kobj, &node_attr->kobj_attr.attr)) {
3463 kfree(node_attr->kobj_attr.attr.name);
3464 kfree(node_attr);
3465 pr_err("failed to add attribute to weighted_interleave\n");
3466 return -ENOMEM;
3467 }
3468
3469 node_attrs[nid] = node_attr;
3470 return 0;
3471 }
3472
add_weighted_interleave_group(struct kobject * root_kobj)3473 static int add_weighted_interleave_group(struct kobject *root_kobj)
3474 {
3475 struct kobject *wi_kobj;
3476 int nid, err;
3477
3478 wi_kobj = kzalloc(sizeof(struct kobject), GFP_KERNEL);
3479 if (!wi_kobj)
3480 return -ENOMEM;
3481
3482 err = kobject_init_and_add(wi_kobj, &wi_ktype, root_kobj,
3483 "weighted_interleave");
3484 if (err) {
3485 kfree(wi_kobj);
3486 return err;
3487 }
3488
3489 for_each_node_state(nid, N_POSSIBLE) {
3490 err = add_weight_node(nid, wi_kobj);
3491 if (err) {
3492 pr_err("failed to add sysfs [node%d]\n", nid);
3493 break;
3494 }
3495 }
3496 if (err)
3497 kobject_put(wi_kobj);
3498 return 0;
3499 }
3500
mempolicy_kobj_release(struct kobject * kobj)3501 static void mempolicy_kobj_release(struct kobject *kobj)
3502 {
3503 u8 *old;
3504
3505 mutex_lock(&iw_table_lock);
3506 old = rcu_dereference_protected(iw_table,
3507 lockdep_is_held(&iw_table_lock));
3508 rcu_assign_pointer(iw_table, NULL);
3509 mutex_unlock(&iw_table_lock);
3510 synchronize_rcu();
3511 kfree(old);
3512 kfree(node_attrs);
3513 kfree(kobj);
3514 }
3515
3516 static const struct kobj_type mempolicy_ktype = {
3517 .release = mempolicy_kobj_release
3518 };
3519
mempolicy_sysfs_init(void)3520 static int __init mempolicy_sysfs_init(void)
3521 {
3522 int err;
3523 static struct kobject *mempolicy_kobj;
3524
3525 mempolicy_kobj = kzalloc(sizeof(*mempolicy_kobj), GFP_KERNEL);
3526 if (!mempolicy_kobj) {
3527 err = -ENOMEM;
3528 goto err_out;
3529 }
3530
3531 node_attrs = kcalloc(nr_node_ids, sizeof(struct iw_node_attr *),
3532 GFP_KERNEL);
3533 if (!node_attrs) {
3534 err = -ENOMEM;
3535 goto mempol_out;
3536 }
3537
3538 err = kobject_init_and_add(mempolicy_kobj, &mempolicy_ktype, mm_kobj,
3539 "mempolicy");
3540 if (err)
3541 goto node_out;
3542
3543 err = add_weighted_interleave_group(mempolicy_kobj);
3544 if (err) {
3545 pr_err("mempolicy sysfs structure failed to initialize\n");
3546 kobject_put(mempolicy_kobj);
3547 return err;
3548 }
3549
3550 return err;
3551 node_out:
3552 kfree(node_attrs);
3553 mempol_out:
3554 kfree(mempolicy_kobj);
3555 err_out:
3556 pr_err("failed to add mempolicy kobject to the system\n");
3557 return err;
3558 }
3559
3560 late_initcall(mempolicy_sysfs_init);
3561 #endif /* CONFIG_SYSFS */
3562