~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/mempolicy.c

Version: ~ [ linux-5.8 ] ~ [ linux-5.7.14 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.57 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.138 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.193 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.232 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.232 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * Simple NUMA memory policy for the Linux kernel.
  3  *
  4  * Copyright 2003,2004 Andi Kleen, SuSE Labs.
  5  * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc.
  6  * Subject to the GNU Public License, version 2.
  7  *
  8  * NUMA policy allows the user to give hints in which node(s) memory should
  9  * be allocated.
 10  *
 11  * Support four policies per VMA and per process:
 12  *
 13  * The VMA policy has priority over the process policy for a page fault.
 14  *
 15  * interleave     Allocate memory interleaved over a set of nodes,
 16  *                with normal fallback if it fails.
 17  *                For VMA based allocations this interleaves based on the
 18  *                offset into the backing object or offset into the mapping
 19  *                for anonymous memory. For process policy an process counter
 20  *                is used.
 21  *
 22  * bind           Only allocate memory on a specific set of nodes,
 23  *                no fallback.
 24  *                FIXME: memory is allocated starting with the first node
 25  *                to the last. It would be better if bind would truly restrict
 26  *                the allocation to memory nodes instead
 27  *
 28  * preferred       Try a specific node first before normal fallback.
 29  *                As a special case NUMA_NO_NODE here means do the allocation
 30  *                on the local CPU. This is normally identical to default,
 31  *                but useful to set in a VMA when you have a non default
 32  *                process policy.
 33  *
 34  * default        Allocate on the local node first, or when on a VMA
 35  *                use the process policy. This is what Linux always did
 36  *                in a NUMA aware kernel and still does by, ahem, default.
 37  *
 38  * The process policy is applied for most non interrupt memory allocations
 39  * in that process' context. Interrupts ignore the policies and always
 40  * try to allocate on the local CPU. The VMA policy is only applied for memory
 41  * allocations for a VMA in the VM.
 42  *
 43  * Currently there are a few corner cases in swapping where the policy
 44  * is not applied, but the majority should be handled. When process policy
 45  * is used it is not remembered over swap outs/swap ins.
 46  *
 47  * Only the highest zone in the zone hierarchy gets policied. Allocations
 48  * requesting a lower zone just use default policy. This implies that
 49  * on systems with highmem kernel lowmem allocation don't get policied.
 50  * Same with GFP_DMA allocations.
 51  *
 52  * For shmfs/tmpfs/hugetlbfs shared memory the policy is shared between
 53  * all users and remembered even when nobody has memory mapped.
 54  */
 55 
 56 /* Notebook:
 57    fix mmap readahead to honour policy and enable policy for any page cache
 58    object
 59    statistics for bigpages
 60    global policy for page cache? currently it uses process policy. Requires
 61    first item above.
 62    handle mremap for shared memory (currently ignored for the policy)
 63    grows down?
 64    make bind policy root only? It can trigger oom much faster and the
 65    kernel is not always grateful with that.
 66 */
 67 
 68 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 69 
 70 #include <linux/mempolicy.h>
 71 #include <linux/mm.h>
 72 #include <linux/highmem.h>
 73 #include <linux/hugetlb.h>
 74 #include <linux/kernel.h>
 75 #include <linux/sched.h>
 76 #include <linux/nodemask.h>
 77 #include <linux/cpuset.h>
 78 #include <linux/slab.h>
 79 #include <linux/string.h>
 80 #include <linux/export.h>
 81 #include <linux/nsproxy.h>
 82 #include <linux/interrupt.h>
 83 #include <linux/init.h>
 84 #include <linux/compat.h>
 85 #include <linux/swap.h>
 86 #include <linux/seq_file.h>
 87 #include <linux/proc_fs.h>
 88 #include <linux/migrate.h>
 89 #include <linux/ksm.h>
 90 #include <linux/rmap.h>
 91 #include <linux/security.h>
 92 #include <linux/syscalls.h>
 93 #include <linux/ctype.h>
 94 #include <linux/mm_inline.h>
 95 #include <linux/mmu_notifier.h>
 96 #include <linux/printk.h>
 97 
 98 #include <asm/tlbflush.h>
 99 #include <asm/uaccess.h>
100 #include <linux/random.h>
101 
102 #include "internal.h"
103 
104 /* Internal flags */
105 #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0)    /* Skip checks for continuous vmas */
106 #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1)          /* Invert check for nodemask */
107 
108 static struct kmem_cache *policy_cache;
109 static struct kmem_cache *sn_cache;
110 
111 /* Highest zone. An specific allocation for a zone below that is not
112    policied. */
113 enum zone_type policy_zone = 0;
114 
115 /*
116  * run-time system-wide default policy => local allocation
117  */
118 static struct mempolicy default_policy = {
119         .refcnt = ATOMIC_INIT(1), /* never free it */
120         .mode = MPOL_PREFERRED,
121         .flags = MPOL_F_LOCAL,
122 };
123 
124 static struct mempolicy preferred_node_policy[MAX_NUMNODES];
125 
126 struct mempolicy *get_task_policy(struct task_struct *p)
127 {
128         struct mempolicy *pol = p->mempolicy;
129         int node;
130 
131         if (pol)
132                 return pol;
133 
134         node = numa_node_id();
135         if (node != NUMA_NO_NODE) {
136                 pol = &preferred_node_policy[node];
137                 /* preferred_node_policy is not initialised early in boot */
138                 if (pol->mode)
139                         return pol;
140         }
141 
142         return &default_policy;
143 }
144 
145 static const struct mempolicy_operations {
146         int (*create)(struct mempolicy *pol, const nodemask_t *nodes);
147         /*
148          * If read-side task has no lock to protect task->mempolicy, write-side
149          * task will rebind the task->mempolicy by two step. The first step is
150          * setting all the newly nodes, and the second step is cleaning all the
151          * disallowed nodes. In this way, we can avoid finding no node to alloc
152          * page.
153          * If we have a lock to protect task->mempolicy in read-side, we do
154          * rebind directly.
155          *
156          * step:
157          *      MPOL_REBIND_ONCE - do rebind work at once
158          *      MPOL_REBIND_STEP1 - set all the newly nodes
159          *      MPOL_REBIND_STEP2 - clean all the disallowed nodes
160          */
161         void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes,
162                         enum mpol_rebind_step step);
163 } mpol_ops[MPOL_MAX];
164 
165 static inline int mpol_store_user_nodemask(const struct mempolicy *pol)
166 {
167         return pol->flags & MPOL_MODE_FLAGS;
168 }
169 
170 static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig,
171                                    const nodemask_t *rel)
172 {
173         nodemask_t tmp;
174         nodes_fold(tmp, *orig, nodes_weight(*rel));
175         nodes_onto(*ret, tmp, *rel);
176 }
177 
178 static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes)
179 {
180         if (nodes_empty(*nodes))
181                 return -EINVAL;
182         pol->v.nodes = *nodes;
183         return 0;
184 }
185 
186 static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes)
187 {
188         if (!nodes)
189                 pol->flags |= MPOL_F_LOCAL;     /* local allocation */
190         else if (nodes_empty(*nodes))
191                 return -EINVAL;                 /*  no allowed nodes */
192         else
193                 pol->v.preferred_node = first_node(*nodes);
194         return 0;
195 }
196 
197 static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes)
198 {
199         if (nodes_empty(*nodes))
200                 return -EINVAL;
201         pol->v.nodes = *nodes;
202         return 0;
203 }
204 
205 /*
206  * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if
207  * any, for the new policy.  mpol_new() has already validated the nodes
208  * parameter with respect to the policy mode and flags.  But, we need to
209  * handle an empty nodemask with MPOL_PREFERRED here.
210  *
211  * Must be called holding task's alloc_lock to protect task's mems_allowed
212  * and mempolicy.  May also be called holding the mmap_semaphore for write.
213  */
214 static int mpol_set_nodemask(struct mempolicy *pol,
215                      const nodemask_t *nodes, struct nodemask_scratch *nsc)
216 {
217         int ret;
218 
219         /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */
220         if (pol == NULL)
221                 return 0;
222         /* Check N_MEMORY */
223         nodes_and(nsc->mask1,
224                   cpuset_current_mems_allowed, node_states[N_MEMORY]);
225 
226         VM_BUG_ON(!nodes);
227         if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes))
228                 nodes = NULL;   /* explicit local allocation */
229         else {
230                 if (pol->flags & MPOL_F_RELATIVE_NODES)
231                         mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1);
232                 else
233                         nodes_and(nsc->mask2, *nodes, nsc->mask1);
234 
235                 if (mpol_store_user_nodemask(pol))
236                         pol->w.user_nodemask = *nodes;
237                 else
238                         pol->w.cpuset_mems_allowed =
239                                                 cpuset_current_mems_allowed;
240         }
241 
242         if (nodes)
243                 ret = mpol_ops[pol->mode].create(pol, &nsc->mask2);
244         else
245                 ret = mpol_ops[pol->mode].create(pol, NULL);
246         return ret;
247 }
248 
249 /*
250  * This function just creates a new policy, does some check and simple
251  * initialization. You must invoke mpol_set_nodemask() to set nodes.
252  */
253 static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags,
254                                   nodemask_t *nodes)
255 {
256         struct mempolicy *policy;
257 
258         pr_debug("setting mode %d flags %d nodes[0] %lx\n",
259                  mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE);
260 
261         if (mode == MPOL_DEFAULT) {
262                 if (nodes && !nodes_empty(*nodes))
263                         return ERR_PTR(-EINVAL);
264                 return NULL;
265         }
266         VM_BUG_ON(!nodes);
267 
268         /*
269          * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or
270          * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation).
271          * All other modes require a valid pointer to a non-empty nodemask.
272          */
273         if (mode == MPOL_PREFERRED) {
274                 if (nodes_empty(*nodes)) {
275                         if (((flags & MPOL_F_STATIC_NODES) ||
276                              (flags & MPOL_F_RELATIVE_NODES)))
277                                 return ERR_PTR(-EINVAL);
278                 }
279         } else if (mode == MPOL_LOCAL) {
280                 if (!nodes_empty(*nodes))
281                         return ERR_PTR(-EINVAL);
282                 mode = MPOL_PREFERRED;
283         } else if (nodes_empty(*nodes))
284                 return ERR_PTR(-EINVAL);
285         policy = kmem_cache_alloc(policy_cache, GFP_KERNEL);
286         if (!policy)
287                 return ERR_PTR(-ENOMEM);
288         atomic_set(&policy->refcnt, 1);
289         policy->mode = mode;
290         policy->flags = flags;
291 
292         return policy;
293 }
294 
295 /* Slow path of a mpol destructor. */
296 void __mpol_put(struct mempolicy *p)
297 {
298         if (!atomic_dec_and_test(&p->refcnt))
299                 return;
300         kmem_cache_free(policy_cache, p);
301 }
302 
303 static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes,
304                                 enum mpol_rebind_step step)
305 {
306 }
307 
308 /*
309  * step:
310  *      MPOL_REBIND_ONCE  - do rebind work at once
311  *      MPOL_REBIND_STEP1 - set all the newly nodes
312  *      MPOL_REBIND_STEP2 - clean all the disallowed nodes
313  */
314 static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes,
315                                  enum mpol_rebind_step step)
316 {
317         nodemask_t tmp;
318 
319         if (pol->flags & MPOL_F_STATIC_NODES)
320                 nodes_and(tmp, pol->w.user_nodemask, *nodes);
321         else if (pol->flags & MPOL_F_RELATIVE_NODES)
322                 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
323         else {
324                 /*
325                  * if step == 1, we use ->w.cpuset_mems_allowed to cache the
326                  * result
327                  */
328                 if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) {
329                         nodes_remap(tmp, pol->v.nodes,
330                                         pol->w.cpuset_mems_allowed, *nodes);
331                         pol->w.cpuset_mems_allowed = step ? tmp : *nodes;
332                 } else if (step == MPOL_REBIND_STEP2) {
333                         tmp = pol->w.cpuset_mems_allowed;
334                         pol->w.cpuset_mems_allowed = *nodes;
335                 } else
336                         BUG();
337         }
338 
339         if (nodes_empty(tmp))
340                 tmp = *nodes;
341 
342         if (step == MPOL_REBIND_STEP1)
343                 nodes_or(pol->v.nodes, pol->v.nodes, tmp);
344         else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2)
345                 pol->v.nodes = tmp;
346         else
347                 BUG();
348 
349         if (!node_isset(current->il_next, tmp)) {
350                 current->il_next = next_node(current->il_next, tmp);
351                 if (current->il_next >= MAX_NUMNODES)
352                         current->il_next = first_node(tmp);
353                 if (current->il_next >= MAX_NUMNODES)
354                         current->il_next = numa_node_id();
355         }
356 }
357 
358 static void mpol_rebind_preferred(struct mempolicy *pol,
359                                   const nodemask_t *nodes,
360                                   enum mpol_rebind_step step)
361 {
362         nodemask_t tmp;
363 
364         if (pol->flags & MPOL_F_STATIC_NODES) {
365                 int node = first_node(pol->w.user_nodemask);
366 
367                 if (node_isset(node, *nodes)) {
368                         pol->v.preferred_node = node;
369                         pol->flags &= ~MPOL_F_LOCAL;
370                 } else
371                         pol->flags |= MPOL_F_LOCAL;
372         } else if (pol->flags & MPOL_F_RELATIVE_NODES) {
373                 mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes);
374                 pol->v.preferred_node = first_node(tmp);
375         } else if (!(pol->flags & MPOL_F_LOCAL)) {
376                 pol->v.preferred_node = node_remap(pol->v.preferred_node,
377                                                    pol->w.cpuset_mems_allowed,
378                                                    *nodes);
379                 pol->w.cpuset_mems_allowed = *nodes;
380         }
381 }
382 
383 /*
384  * mpol_rebind_policy - Migrate a policy to a different set of nodes
385  *
386  * If read-side task has no lock to protect task->mempolicy, write-side
387  * task will rebind the task->mempolicy by two step. The first step is
388  * setting all the newly nodes, and the second step is cleaning all the
389  * disallowed nodes. In this way, we can avoid finding no node to alloc
390  * page.
391  * If we have a lock to protect task->mempolicy in read-side, we do
392  * rebind directly.
393  *
394  * step:
395  *      MPOL_REBIND_ONCE  - do rebind work at once
396  *      MPOL_REBIND_STEP1 - set all the newly nodes
397  *      MPOL_REBIND_STEP2 - clean all the disallowed nodes
398  */
399 static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask,
400                                 enum mpol_rebind_step step)
401 {
402         if (!pol)
403                 return;
404         if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE &&
405             nodes_equal(pol->w.cpuset_mems_allowed, *newmask))
406                 return;
407 
408         if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING))
409                 return;
410 
411         if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING))
412                 BUG();
413 
414         if (step == MPOL_REBIND_STEP1)
415                 pol->flags |= MPOL_F_REBINDING;
416         else if (step == MPOL_REBIND_STEP2)
417                 pol->flags &= ~MPOL_F_REBINDING;
418         else if (step >= MPOL_REBIND_NSTEP)
419                 BUG();
420 
421         mpol_ops[pol->mode].rebind(pol, newmask, step);
422 }
423 
424 /*
425  * Wrapper for mpol_rebind_policy() that just requires task
426  * pointer, and updates task mempolicy.
427  *
428  * Called with task's alloc_lock held.
429  */
430 
431 void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new,
432                         enum mpol_rebind_step step)
433 {
434         mpol_rebind_policy(tsk->mempolicy, new, step);
435 }
436 
437 /*
438  * Rebind each vma in mm to new nodemask.
439  *
440  * Call holding a reference to mm.  Takes mm->mmap_sem during call.
441  */
442 
443 void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
444 {
445         struct vm_area_struct *vma;
446 
447         down_write(&mm->mmap_sem);
448         for (vma = mm->mmap; vma; vma = vma->vm_next)
449                 mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE);
450         up_write(&mm->mmap_sem);
451 }
452 
453 static const struct mempolicy_operations mpol_ops[MPOL_MAX] = {
454         [MPOL_DEFAULT] = {
455                 .rebind = mpol_rebind_default,
456         },
457         [MPOL_INTERLEAVE] = {
458                 .create = mpol_new_interleave,
459                 .rebind = mpol_rebind_nodemask,
460         },
461         [MPOL_PREFERRED] = {
462                 .create = mpol_new_preferred,
463                 .rebind = mpol_rebind_preferred,
464         },
465         [MPOL_BIND] = {
466                 .create = mpol_new_bind,
467                 .rebind = mpol_rebind_nodemask,
468         },
469 };
470 
471 static void migrate_page_add(struct page *page, struct list_head *pagelist,
472                                 unsigned long flags);
473 
474 struct queue_pages {
475         struct list_head *pagelist;
476         unsigned long flags;
477         nodemask_t *nmask;
478         struct vm_area_struct *prev;
479 };
480 
481 /*
482  * Scan through pages checking if pages follow certain conditions,
483  * and move them to the pagelist if they do.
484  */
485 static int queue_pages_pte_range(pmd_t *pmd, unsigned long addr,
486                         unsigned long end, struct mm_walk *walk)
487 {
488         struct vm_area_struct *vma = walk->vma;
489         struct page *page;
490         struct queue_pages *qp = walk->private;
491         unsigned long flags = qp->flags;
492         int nid;
493         pte_t *pte;
494         spinlock_t *ptl;
495 
496         split_huge_page_pmd(vma, addr, pmd);
497         if (pmd_trans_unstable(pmd))
498                 return 0;
499 
500         pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
501         for (; addr != end; pte++, addr += PAGE_SIZE) {
502                 if (!pte_present(*pte))
503                         continue;
504                 page = vm_normal_page(vma, addr, *pte);
505                 if (!page)
506                         continue;
507                 /*
508                  * vm_normal_page() filters out zero pages, but there might
509                  * still be PageReserved pages to skip, perhaps in a VDSO.
510                  */
511                 if (PageReserved(page))
512                         continue;
513                 nid = page_to_nid(page);
514                 if (node_isset(nid, *qp->nmask) == !!(flags & MPOL_MF_INVERT))
515                         continue;
516 
517                 if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
518                         migrate_page_add(page, qp->pagelist, flags);
519         }
520         pte_unmap_unlock(pte - 1, ptl);
521         cond_resched();
522         return 0;
523 }
524 
525 static int queue_pages_hugetlb(pte_t *pte, unsigned long hmask,
526                                unsigned long addr, unsigned long end,
527                                struct mm_walk *walk)
528 {
529 #ifdef CONFIG_HUGETLB_PAGE
530         struct queue_pages *qp = walk->private;
531         unsigned long flags = qp->flags;
532         int nid;
533         struct page *page;
534         spinlock_t *ptl;
535         pte_t entry;
536 
537         ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte);
538         entry = huge_ptep_get(pte);
539         if (!pte_present(entry))
540                 goto unlock;
541         page = pte_page(entry);
542         nid = page_to_nid(page);
543         if (node_isset(nid, *qp->nmask) == !!(flags & MPOL_MF_INVERT))
544                 goto unlock;
545         /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */
546         if (flags & (MPOL_MF_MOVE_ALL) ||
547             (flags & MPOL_MF_MOVE && page_mapcount(page) == 1))
548                 isolate_huge_page(page, qp->pagelist);
549 unlock:
550         spin_unlock(ptl);
551 #else
552         BUG();
553 #endif
554         return 0;
555 }
556 
557 #ifdef CONFIG_NUMA_BALANCING
558 /*
559  * This is used to mark a range of virtual addresses to be inaccessible.
560  * These are later cleared by a NUMA hinting fault. Depending on these
561  * faults, pages may be migrated for better NUMA placement.
562  *
563  * This is assuming that NUMA faults are handled using PROT_NONE. If
564  * an architecture makes a different choice, it will need further
565  * changes to the core.
566  */
567 unsigned long change_prot_numa(struct vm_area_struct *vma,
568                         unsigned long addr, unsigned long end)
569 {
570         int nr_updated;
571 
572         nr_updated = change_protection(vma, addr, end, PAGE_NONE, 0, 1);
573         if (nr_updated)
574                 count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated);
575 
576         return nr_updated;
577 }
578 #else
579 static unsigned long change_prot_numa(struct vm_area_struct *vma,
580                         unsigned long addr, unsigned long end)
581 {
582         return 0;
583 }
584 #endif /* CONFIG_NUMA_BALANCING */
585 
586 static int queue_pages_test_walk(unsigned long start, unsigned long end,
587                                 struct mm_walk *walk)
588 {
589         struct vm_area_struct *vma = walk->vma;
590         struct queue_pages *qp = walk->private;
591         unsigned long endvma = vma->vm_end;
592         unsigned long flags = qp->flags;
593 
594         if (vma->vm_flags & VM_PFNMAP)
595                 return 1;
596 
597         if (endvma > end)
598                 endvma = end;
599         if (vma->vm_start > start)
600                 start = vma->vm_start;
601 
602         if (!(flags & MPOL_MF_DISCONTIG_OK)) {
603                 if (!vma->vm_next && vma->vm_end < end)
604                         return -EFAULT;
605                 if (qp->prev && qp->prev->vm_end < vma->vm_start)
606                         return -EFAULT;
607         }
608 
609         qp->prev = vma;
610 
611         if (vma->vm_flags & VM_PFNMAP)
612                 return 1;
613 
614         if (flags & MPOL_MF_LAZY) {
615                 /* Similar to task_numa_work, skip inaccessible VMAs */
616                 if (vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))
617                         change_prot_numa(vma, start, endvma);
618                 return 1;
619         }
620 
621         if ((flags & MPOL_MF_STRICT) ||
622             ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) &&
623              vma_migratable(vma)))
624                 /* queue pages from current vma */
625                 return 0;
626         return 1;
627 }
628 
629 /*
630  * Walk through page tables and collect pages to be migrated.
631  *
632  * If pages found in a given range are on a set of nodes (determined by
633  * @nodes and @flags,) it's isolated and queued to the pagelist which is
634  * passed via @private.)
635  */
636 static int
637 queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end,
638                 nodemask_t *nodes, unsigned long flags,
639                 struct list_head *pagelist)
640 {
641         struct queue_pages qp = {
642                 .pagelist = pagelist,
643                 .flags = flags,
644                 .nmask = nodes,
645                 .prev = NULL,
646         };
647         struct mm_walk queue_pages_walk = {
648                 .hugetlb_entry = queue_pages_hugetlb,
649                 .pmd_entry = queue_pages_pte_range,
650                 .test_walk = queue_pages_test_walk,
651                 .mm = mm,
652                 .private = &qp,
653         };
654 
655         return walk_page_range(start, end, &queue_pages_walk);
656 }
657 
658 /*
659  * Apply policy to a single VMA
660  * This must be called with the mmap_sem held for writing.
661  */
662 static int vma_replace_policy(struct vm_area_struct *vma,
663                                                 struct mempolicy *pol)
664 {
665         int err;
666         struct mempolicy *old;
667         struct mempolicy *new;
668 
669         pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n",
670                  vma->vm_start, vma->vm_end, vma->vm_pgoff,
671                  vma->vm_ops, vma->vm_file,
672                  vma->vm_ops ? vma->vm_ops->set_policy : NULL);
673 
674         new = mpol_dup(pol);
675         if (IS_ERR(new))
676                 return PTR_ERR(new);
677 
678         if (vma->vm_ops && vma->vm_ops->set_policy) {
679                 err = vma->vm_ops->set_policy(vma, new);
680                 if (err)
681                         goto err_out;
682         }
683 
684         old = vma->vm_policy;
685         vma->vm_policy = new; /* protected by mmap_sem */
686         mpol_put(old);
687 
688         return 0;
689  err_out:
690         mpol_put(new);
691         return err;
692 }
693 
694 /* Step 2: apply policy to a range and do splits. */
695 static int mbind_range(struct mm_struct *mm, unsigned long start,
696                        unsigned long end, struct mempolicy *new_pol)
697 {
698         struct vm_area_struct *next;
699         struct vm_area_struct *prev;
700         struct vm_area_struct *vma;
701         int err = 0;
702         pgoff_t pgoff;
703         unsigned long vmstart;
704         unsigned long vmend;
705 
706         vma = find_vma(mm, start);
707         if (!vma || vma->vm_start > start)
708                 return -EFAULT;
709 
710         prev = vma->vm_prev;
711         if (start > vma->vm_start)
712                 prev = vma;
713 
714         for (; vma && vma->vm_start < end; prev = vma, vma = next) {
715                 next = vma->vm_next;
716                 vmstart = max(start, vma->vm_start);
717                 vmend   = min(end, vma->vm_end);
718 
719                 if (mpol_equal(vma_policy(vma), new_pol))
720                         continue;
721 
722                 pgoff = vma->vm_pgoff +
723                         ((vmstart - vma->vm_start) >> PAGE_SHIFT);
724                 prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags,
725                                   vma->anon_vma, vma->vm_file, pgoff,
726                                   new_pol);
727                 if (prev) {
728                         vma = prev;
729                         next = vma->vm_next;
730                         if (mpol_equal(vma_policy(vma), new_pol))
731                                 continue;
732                         /* vma_merge() joined vma && vma->next, case 8 */
733                         goto replace;
734                 }
735                 if (vma->vm_start != vmstart) {
736                         err = split_vma(vma->vm_mm, vma, vmstart, 1);
737                         if (err)
738                                 goto out;
739                 }
740                 if (vma->vm_end != vmend) {
741                         err = split_vma(vma->vm_mm, vma, vmend, 0);
742                         if (err)
743                                 goto out;
744                 }
745  replace:
746                 err = vma_replace_policy(vma, new_pol);
747                 if (err)
748                         goto out;
749         }
750 
751  out:
752         return err;
753 }
754 
755 /* Set the process memory policy */
756 static long do_set_mempolicy(unsigned short mode, unsigned short flags,
757                              nodemask_t *nodes)
758 {
759         struct mempolicy *new, *old;
760         NODEMASK_SCRATCH(scratch);
761         int ret;
762 
763         if (!scratch)
764                 return -ENOMEM;
765 
766         new = mpol_new(mode, flags, nodes);
767         if (IS_ERR(new)) {
768                 ret = PTR_ERR(new);
769                 goto out;
770         }
771 
772         task_lock(current);
773         ret = mpol_set_nodemask(new, nodes, scratch);
774         if (ret) {
775                 task_unlock(current);
776                 mpol_put(new);
777                 goto out;
778         }
779         old = current->mempolicy;
780         current->mempolicy = new;
781         if (new && new->mode == MPOL_INTERLEAVE &&
782             nodes_weight(new->v.nodes))
783                 current->il_next = first_node(new->v.nodes);
784         task_unlock(current);
785         mpol_put(old);
786         ret = 0;
787 out:
788         NODEMASK_SCRATCH_FREE(scratch);
789         return ret;
790 }
791 
792 /*
793  * Return nodemask for policy for get_mempolicy() query
794  *
795  * Called with task's alloc_lock held
796  */
797 static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes)
798 {
799         nodes_clear(*nodes);
800         if (p == &default_policy)
801                 return;
802 
803         switch (p->mode) {
804         case MPOL_BIND:
805                 /* Fall through */
806         case MPOL_INTERLEAVE:
807                 *nodes = p->v.nodes;
808                 break;
809         case MPOL_PREFERRED:
810                 if (!(p->flags & MPOL_F_LOCAL))
811                         node_set(p->v.preferred_node, *nodes);
812                 /* else return empty node mask for local allocation */
813                 break;
814         default:
815                 BUG();
816         }
817 }
818 
819 static int lookup_node(struct mm_struct *mm, unsigned long addr)
820 {
821         struct page *p;
822         int err;
823 
824         err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL);
825         if (err >= 0) {
826                 err = page_to_nid(p);
827                 put_page(p);
828         }
829         return err;
830 }
831 
832 /* Retrieve NUMA policy */
833 static long do_get_mempolicy(int *policy, nodemask_t *nmask,
834                              unsigned long addr, unsigned long flags)
835 {
836         int err;
837         struct mm_struct *mm = current->mm;
838         struct vm_area_struct *vma = NULL;
839         struct mempolicy *pol = current->mempolicy;
840 
841         if (flags &
842                 ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED))
843                 return -EINVAL;
844 
845         if (flags & MPOL_F_MEMS_ALLOWED) {
846                 if (flags & (MPOL_F_NODE|MPOL_F_ADDR))
847                         return -EINVAL;
848                 *policy = 0;    /* just so it's initialized */
849                 task_lock(current);
850                 *nmask  = cpuset_current_mems_allowed;
851                 task_unlock(current);
852                 return 0;
853         }
854 
855         if (flags & MPOL_F_ADDR) {
856                 /*
857                  * Do NOT fall back to task policy if the
858                  * vma/shared policy at addr is NULL.  We
859                  * want to return MPOL_DEFAULT in this case.
860                  */
861                 down_read(&mm->mmap_sem);
862                 vma = find_vma_intersection(mm, addr, addr+1);
863                 if (!vma) {
864                         up_read(&mm->mmap_sem);
865                         return -EFAULT;
866                 }
867                 if (vma->vm_ops && vma->vm_ops->get_policy)
868                         pol = vma->vm_ops->get_policy(vma, addr);
869                 else
870                         pol = vma->vm_policy;
871         } else if (addr)
872                 return -EINVAL;
873 
874         if (!pol)
875                 pol = &default_policy;  /* indicates default behavior */
876 
877         if (flags & MPOL_F_NODE) {
878                 if (flags & MPOL_F_ADDR) {
879                         err = lookup_node(mm, addr);
880                         if (err < 0)
881                                 goto out;
882                         *policy = err;
883                 } else if (pol == current->mempolicy &&
884                                 pol->mode == MPOL_INTERLEAVE) {
885                         *policy = current->il_next;
886                 } else {
887                         err = -EINVAL;
888                         goto out;
889                 }
890         } else {
891                 *policy = pol == &default_policy ? MPOL_DEFAULT :
892                                                 pol->mode;
893                 /*
894                  * Internal mempolicy flags must be masked off before exposing
895                  * the policy to userspace.
896                  */
897                 *policy |= (pol->flags & MPOL_MODE_FLAGS);
898         }
899 
900         if (vma) {
901                 up_read(&current->mm->mmap_sem);
902                 vma = NULL;
903         }
904 
905         err = 0;
906         if (nmask) {
907                 if (mpol_store_user_nodemask(pol)) {
908                         *nmask = pol->w.user_nodemask;
909                 } else {
910                         task_lock(current);
911                         get_policy_nodemask(pol, nmask);
912                         task_unlock(current);
913                 }
914         }
915 
916  out:
917         mpol_cond_put(pol);
918         if (vma)
919                 up_read(&current->mm->mmap_sem);
920         return err;
921 }
922 
923 #ifdef CONFIG_MIGRATION
924 /*
925  * page migration
926  */
927 static void migrate_page_add(struct page *page, struct list_head *pagelist,
928                                 unsigned long flags)
929 {
930         /*
931          * Avoid migrating a page that is shared with others.
932          */
933         if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) {
934                 if (!isolate_lru_page(page)) {
935                         list_add_tail(&page->lru, pagelist);
936                         inc_zone_page_state(page, NR_ISOLATED_ANON +
937                                             page_is_file_cache(page));
938                 }
939         }
940 }
941 
942 static struct page *new_node_page(struct page *page, unsigned long node, int **x)
943 {
944         if (PageHuge(page))
945                 return alloc_huge_page_node(page_hstate(compound_head(page)),
946                                         node);
947         else
948                 return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0);
949 }
950 
951 /*
952  * Migrate pages from one node to a target node.
953  * Returns error or the number of pages not migrated.
954  */
955 static int migrate_to_node(struct mm_struct *mm, int source, int dest,
956                            int flags)
957 {
958         nodemask_t nmask;
959         LIST_HEAD(pagelist);
960         int err = 0;
961 
962         nodes_clear(nmask);
963         node_set(source, nmask);
964 
965         /*
966          * This does not "check" the range but isolates all pages that
967          * need migration.  Between passing in the full user address
968          * space range and MPOL_MF_DISCONTIG_OK, this call can not fail.
969          */
970         VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)));
971         queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask,
972                         flags | MPOL_MF_DISCONTIG_OK, &pagelist);
973 
974         if (!list_empty(&pagelist)) {
975                 err = migrate_pages(&pagelist, new_node_page, NULL, dest,
976                                         MIGRATE_SYNC, MR_SYSCALL);
977                 if (err)
978                         putback_movable_pages(&pagelist);
979         }
980 
981         return err;
982 }
983 
984 /*
985  * Move pages between the two nodesets so as to preserve the physical
986  * layout as much as possible.
987  *
988  * Returns the number of page that could not be moved.
989  */
990 int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
991                      const nodemask_t *to, int flags)
992 {
993         int busy = 0;
994         int err;
995         nodemask_t tmp;
996 
997         err = migrate_prep();
998         if (err)
999                 return err;
1000 
1001         down_read(&mm->mmap_sem);
1002 
1003         /*
1004          * Find a 'source' bit set in 'tmp' whose corresponding 'dest'
1005          * bit in 'to' is not also set in 'tmp'.  Clear the found 'source'
1006          * bit in 'tmp', and return that <source, dest> pair for migration.
1007          * The pair of nodemasks 'to' and 'from' define the map.
1008          *
1009          * If no pair of bits is found that way, fallback to picking some
1010          * pair of 'source' and 'dest' bits that are not the same.  If the
1011          * 'source' and 'dest' bits are the same, this represents a node
1012          * that will be migrating to itself, so no pages need move.
1013          *
1014          * If no bits are left in 'tmp', or if all remaining bits left
1015          * in 'tmp' correspond to the same bit in 'to', return false
1016          * (nothing left to migrate).
1017          *
1018          * This lets us pick a pair of nodes to migrate between, such that
1019          * if possible the dest node is not already occupied by some other
1020          * source node, minimizing the risk of overloading the memory on a
1021          * node that would happen if we migrated incoming memory to a node
1022          * before migrating outgoing memory source that same node.
1023          *
1024          * A single scan of tmp is sufficient.  As we go, we remember the
1025          * most recent <s, d> pair that moved (s != d).  If we find a pair
1026          * that not only moved, but what's better, moved to an empty slot
1027          * (d is not set in tmp), then we break out then, with that pair.
1028          * Otherwise when we finish scanning from_tmp, we at least have the
1029          * most recent <s, d> pair that moved.  If we get all the way through
1030          * the scan of tmp without finding any node that moved, much less
1031          * moved to an empty node, then there is nothing left worth migrating.
1032          */
1033 
1034         tmp = *from;
1035         while (!nodes_empty(tmp)) {
1036                 int s,d;
1037                 int source = NUMA_NO_NODE;
1038                 int dest = 0;
1039 
1040                 for_each_node_mask(s, tmp) {
1041 
1042                         /*
1043                          * do_migrate_pages() tries to maintain the relative
1044                          * node relationship of the pages established between
1045                          * threads and memory areas.
1046                          *
1047                          * However if the number of source nodes is not equal to
1048                          * the number of destination nodes we can not preserve
1049                          * this node relative relationship.  In that case, skip
1050                          * copying memory from a node that is in the destination
1051                          * mask.
1052                          *
1053                          * Example: [2,3,4] -> [3,4,5] moves everything.
1054                          *          [0-7] - > [3,4,5] moves only 0,1,2,6,7.
1055                          */
1056 
1057                         if ((nodes_weight(*from) != nodes_weight(*to)) &&
1058                                                 (node_isset(s, *to)))
1059                                 continue;
1060 
1061                         d = node_remap(s, *from, *to);
1062                         if (s == d)
1063                                 continue;
1064 
1065                         source = s;     /* Node moved. Memorize */
1066                         dest = d;
1067 
1068                         /* dest not in remaining from nodes? */
1069                         if (!node_isset(dest, tmp))
1070                                 break;
1071                 }
1072                 if (source == NUMA_NO_NODE)
1073                         break;
1074 
1075                 node_clear(source, tmp);
1076                 err = migrate_to_node(mm, source, dest, flags);
1077                 if (err > 0)
1078                         busy += err;
1079                 if (err < 0)
1080                         break;
1081         }
1082         up_read(&mm->mmap_sem);
1083         if (err < 0)
1084                 return err;
1085         return busy;
1086 
1087 }
1088 
1089 /*
1090  * Allocate a new page for page migration based on vma policy.
1091  * Start by assuming the page is mapped by the same vma as contains @start.
1092  * Search forward from there, if not.  N.B., this assumes that the
1093  * list of pages handed to migrate_pages()--which is how we get here--
1094  * is in virtual address order.
1095  */
1096 static struct page *new_page(struct page *page, unsigned long start, int **x)
1097 {
1098         struct vm_area_struct *vma;
1099         unsigned long uninitialized_var(address);
1100 
1101         vma = find_vma(current->mm, start);
1102         while (vma) {
1103                 address = page_address_in_vma(page, vma);
1104                 if (address != -EFAULT)
1105                         break;
1106                 vma = vma->vm_next;
1107         }
1108 
1109         if (PageHuge(page)) {
1110                 BUG_ON(!vma);
1111                 return alloc_huge_page_noerr(vma, address, 1);
1112         }
1113         /*
1114          * if !vma, alloc_page_vma() will use task or system default policy
1115          */
1116         return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
1117 }
1118 #else
1119 
1120 static void migrate_page_add(struct page *page, struct list_head *pagelist,
1121                                 unsigned long flags)
1122 {
1123 }
1124 
1125 int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
1126                      const nodemask_t *to, int flags)
1127 {
1128         return -ENOSYS;
1129 }
1130 
1131 static struct page *new_page(struct page *page, unsigned long start, int **x)
1132 {
1133         return NULL;
1134 }
1135 #endif
1136 
1137 static long do_mbind(unsigned long start, unsigned long len,
1138                      unsigned short mode, unsigned short mode_flags,
1139                      nodemask_t *nmask, unsigned long flags)
1140 {
1141         struct mm_struct *mm = current->mm;
1142         struct mempolicy *new;
1143         unsigned long end;
1144         int err;
1145         LIST_HEAD(pagelist);
1146 
1147         if (flags & ~(unsigned long)MPOL_MF_VALID)
1148                 return -EINVAL;
1149         if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1150                 return -EPERM;
1151 
1152         if (start & ~PAGE_MASK)
1153                 return -EINVAL;
1154 
1155         if (mode == MPOL_DEFAULT)
1156                 flags &= ~MPOL_MF_STRICT;
1157 
1158         len = (len + PAGE_SIZE - 1) & PAGE_MASK;
1159         end = start + len;
1160 
1161         if (end < start)
1162                 return -EINVAL;
1163         if (end == start)
1164                 return 0;
1165 
1166         new = mpol_new(mode, mode_flags, nmask);
1167         if (IS_ERR(new))
1168                 return PTR_ERR(new);
1169 
1170         if (flags & MPOL_MF_LAZY)
1171                 new->flags |= MPOL_F_MOF;
1172 
1173         /*
1174          * If we are using the default policy then operation
1175          * on discontinuous address spaces is okay after all
1176          */
1177         if (!new)
1178                 flags |= MPOL_MF_DISCONTIG_OK;
1179 
1180         pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n",
1181                  start, start + len, mode, mode_flags,
1182                  nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE);
1183 
1184         if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
1185 
1186                 err = migrate_prep();
1187                 if (err)
1188                         goto mpol_out;
1189         }
1190         {
1191                 NODEMASK_SCRATCH(scratch);
1192                 if (scratch) {
1193                         down_write(&mm->mmap_sem);
1194                         task_lock(current);
1195                         err = mpol_set_nodemask(new, nmask, scratch);
1196                         task_unlock(current);
1197                         if (err)
1198                                 up_write(&mm->mmap_sem);
1199                 } else
1200                         err = -ENOMEM;
1201                 NODEMASK_SCRATCH_FREE(scratch);
1202         }
1203         if (err)
1204                 goto mpol_out;
1205 
1206         err = queue_pages_range(mm, start, end, nmask,
1207                           flags | MPOL_MF_INVERT, &pagelist);
1208         if (!err)
1209                 err = mbind_range(mm, start, end, new);
1210 
1211         if (!err) {
1212                 int nr_failed = 0;
1213 
1214                 if (!list_empty(&pagelist)) {
1215                         WARN_ON_ONCE(flags & MPOL_MF_LAZY);
1216                         nr_failed = migrate_pages(&pagelist, new_page, NULL,
1217                                 start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND);
1218                         if (nr_failed)
1219                                 putback_movable_pages(&pagelist);
1220                 }
1221 
1222                 if (nr_failed && (flags & MPOL_MF_STRICT))
1223                         err = -EIO;
1224         } else
1225                 putback_movable_pages(&pagelist);
1226 
1227         up_write(&mm->mmap_sem);
1228  mpol_out:
1229         mpol_put(new);
1230         return err;
1231 }
1232 
1233 /*
1234  * User space interface with variable sized bitmaps for nodelists.
1235  */
1236 
1237 /* Copy a node mask from user space. */
1238 static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask,
1239                      unsigned long maxnode)
1240 {
1241         unsigned long k;
1242         unsigned long nlongs;
1243         unsigned long endmask;
1244 
1245         --maxnode;
1246         nodes_clear(*nodes);
1247         if (maxnode == 0 || !nmask)
1248                 return 0;
1249         if (maxnode > PAGE_SIZE*BITS_PER_BYTE)
1250                 return -EINVAL;
1251 
1252         nlongs = BITS_TO_LONGS(maxnode);
1253         if ((maxnode % BITS_PER_LONG) == 0)
1254                 endmask = ~0UL;
1255         else
1256                 endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1;
1257 
1258         /* When the user specified more nodes than supported just check
1259            if the non supported part is all zero. */
1260         if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) {
1261                 if (nlongs > PAGE_SIZE/sizeof(long))
1262                         return -EINVAL;
1263                 for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) {
1264                         unsigned long t;
1265                         if (get_user(t, nmask + k))
1266                                 return -EFAULT;
1267                         if (k == nlongs - 1) {
1268                                 if (t & endmask)
1269                                         return -EINVAL;
1270                         } else if (t)
1271                                 return -EINVAL;
1272                 }
1273                 nlongs = BITS_TO_LONGS(MAX_NUMNODES);
1274                 endmask = ~0UL;
1275         }
1276 
1277         if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long)))
1278                 return -EFAULT;
1279         nodes_addr(*nodes)[nlongs-1] &= endmask;
1280         return 0;
1281 }
1282 
1283 /* Copy a kernel node mask to user space */
1284 static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode,
1285                               nodemask_t *nodes)
1286 {
1287         unsigned long copy = ALIGN(maxnode-1, 64) / 8;
1288         const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long);
1289 
1290         if (copy > nbytes) {
1291                 if (copy > PAGE_SIZE)
1292                         return -EINVAL;
1293                 if (clear_user((char __user *)mask + nbytes, copy - nbytes))
1294                         return -EFAULT;
1295                 copy = nbytes;
1296         }
1297         return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0;
1298 }
1299 
1300 SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len,
1301                 unsigned long, mode, const unsigned long __user *, nmask,
1302                 unsigned long, maxnode, unsigned, flags)
1303 {
1304         nodemask_t nodes;
1305         int err;
1306         unsigned short mode_flags;
1307 
1308         mode_flags = mode & MPOL_MODE_FLAGS;
1309         mode &= ~MPOL_MODE_FLAGS;
1310         if (mode >= MPOL_MAX)
1311                 return -EINVAL;
1312         if ((mode_flags & MPOL_F_STATIC_NODES) &&
1313             (mode_flags & MPOL_F_RELATIVE_NODES))
1314                 return -EINVAL;
1315         err = get_nodes(&nodes, nmask, maxnode);
1316         if (err)
1317                 return err;
1318         return do_mbind(start, len, mode, mode_flags, &nodes, flags);
1319 }
1320 
1321 /* Set the process memory policy */
1322 SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask,
1323                 unsigned long, maxnode)
1324 {
1325         int err;
1326         nodemask_t nodes;
1327         unsigned short flags;
1328 
1329         flags = mode & MPOL_MODE_FLAGS;
1330         mode &= ~MPOL_MODE_FLAGS;
1331         if ((unsigned int)mode >= MPOL_MAX)
1332                 return -EINVAL;
1333         if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES))
1334                 return -EINVAL;
1335         err = get_nodes(&nodes, nmask, maxnode);
1336         if (err)
1337                 return err;
1338         return do_set_mempolicy(mode, flags, &nodes);
1339 }
1340 
1341 SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
1342                 const unsigned long __user *, old_nodes,
1343                 const unsigned long __user *, new_nodes)
1344 {
1345         const struct cred *cred = current_cred(), *tcred;
1346         struct mm_struct *mm = NULL;
1347         struct task_struct *task;
1348         nodemask_t task_nodes;
1349         int err;
1350         nodemask_t *old;
1351         nodemask_t *new;
1352         NODEMASK_SCRATCH(scratch);
1353 
1354         if (!scratch)
1355                 return -ENOMEM;
1356 
1357         old = &scratch->mask1;
1358         new = &scratch->mask2;
1359 
1360         err = get_nodes(old, old_nodes, maxnode);
1361         if (err)
1362                 goto out;
1363 
1364         err = get_nodes(new, new_nodes, maxnode);
1365         if (err)
1366                 goto out;
1367 
1368         /* Find the mm_struct */
1369         rcu_read_lock();
1370         task = pid ? find_task_by_vpid(pid) : current;
1371         if (!task) {
1372                 rcu_read_unlock();
1373                 err = -ESRCH;
1374                 goto out;
1375         }
1376         get_task_struct(task);
1377 
1378         err = -EINVAL;
1379 
1380         /*
1381          * Check if this process has the right to modify the specified
1382          * process. The right exists if the process has administrative
1383          * capabilities, superuser privileges or the same
1384          * userid as the target process.
1385          */
1386         tcred = __task_cred(task);
1387         if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1388             !uid_eq(cred->uid,  tcred->suid) && !uid_eq(cred->uid,  tcred->uid) &&
1389             !capable(CAP_SYS_NICE)) {
1390                 rcu_read_unlock();
1391                 err = -EPERM;
1392                 goto out_put;
1393         }
1394         rcu_read_unlock();
1395 
1396         task_nodes = cpuset_mems_allowed(task);
1397         /* Is the user allowed to access the target nodes? */
1398         if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
1399                 err = -EPERM;
1400                 goto out_put;
1401         }
1402 
1403         if (!nodes_subset(*new, node_states[N_MEMORY])) {
1404                 err = -EINVAL;
1405                 goto out_put;
1406         }
1407 
1408         err = security_task_movememory(task);
1409         if (err)
1410                 goto out_put;
1411 
1412         mm = get_task_mm(task);
1413         put_task_struct(task);
1414 
1415         if (!mm) {
1416                 err = -EINVAL;
1417                 goto out;
1418         }
1419 
1420         err = do_migrate_pages(mm, old, new,
1421                 capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
1422 
1423         mmput(mm);
1424 out:
1425         NODEMASK_SCRATCH_FREE(scratch);
1426 
1427         return err;
1428 
1429 out_put:
1430         put_task_struct(task);
1431         goto out;
1432 
1433 }
1434 
1435 
1436 /* Retrieve NUMA policy */
1437 SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1438                 unsigned long __user *, nmask, unsigned long, maxnode,
1439                 unsigned long, addr, unsigned long, flags)
1440 {
1441         int err;
1442         int uninitialized_var(pval);
1443         nodemask_t nodes;
1444 
1445         if (nmask != NULL && maxnode < MAX_NUMNODES)
1446                 return -EINVAL;
1447 
1448         err = do_get_mempolicy(&pval, &nodes, addr, flags);
1449 
1450         if (err)
1451                 return err;
1452 
1453         if (policy && put_user(pval, policy))
1454                 return -EFAULT;
1455 
1456         if (nmask)
1457                 err = copy_nodes_to_user(nmask, maxnode, &nodes);
1458 
1459         return err;
1460 }
1461 
1462 #ifdef CONFIG_COMPAT
1463 
1464 COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy,
1465                        compat_ulong_t __user *, nmask,
1466                        compat_ulong_t, maxnode,
1467                        compat_ulong_t, addr, compat_ulong_t, flags)
1468 {
1469         long err;
1470         unsigned long __user *nm = NULL;
1471         unsigned long nr_bits, alloc_size;
1472         DECLARE_BITMAP(bm, MAX_NUMNODES);
1473 
1474         nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1475         alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1476 
1477         if (nmask)
1478                 nm = compat_alloc_user_space(alloc_size);
1479 
1480         err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags);
1481 
1482         if (!err && nmask) {
1483                 unsigned long copy_size;
1484                 copy_size = min_t(unsigned long, sizeof(bm), alloc_size);
1485                 err = copy_from_user(bm, nm, copy_size);
1486                 /* ensure entire bitmap is zeroed */
1487                 err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8);
1488                 err |= compat_put_bitmap(nmask, bm, nr_bits);
1489         }
1490 
1491         return err;
1492 }
1493 
1494 COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask,
1495                        compat_ulong_t, maxnode)
1496 {
1497         long err = 0;
1498         unsigned long __user *nm = NULL;
1499         unsigned long nr_bits, alloc_size;
1500         DECLARE_BITMAP(bm, MAX_NUMNODES);
1501 
1502         nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1503         alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1504 
1505         if (nmask) {
1506                 err = compat_get_bitmap(bm, nmask, nr_bits);
1507                 nm = compat_alloc_user_space(alloc_size);
1508                 err |= copy_to_user(nm, bm, alloc_size);
1509         }
1510 
1511         if (err)
1512                 return -EFAULT;
1513 
1514         return sys_set_mempolicy(mode, nm, nr_bits+1);
1515 }
1516 
1517 COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len,
1518                        compat_ulong_t, mode, compat_ulong_t __user *, nmask,
1519                        compat_ulong_t, maxnode, compat_ulong_t, flags)
1520 {
1521         long err = 0;
1522         unsigned long __user *nm = NULL;
1523         unsigned long nr_bits, alloc_size;
1524         nodemask_t bm;
1525 
1526         nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES);
1527         alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8;
1528 
1529         if (nmask) {
1530                 err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits);
1531                 nm = compat_alloc_user_space(alloc_size);
1532                 err |= copy_to_user(nm, nodes_addr(bm), alloc_size);
1533         }
1534 
1535         if (err)
1536                 return -EFAULT;
1537 
1538         return sys_mbind(start, len, mode, nm, nr_bits+1, flags);
1539 }
1540 
1541 #endif
1542 
1543 struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
1544                                                 unsigned long addr)
1545 {
1546         struct mempolicy *pol = NULL;
1547 
1548         if (vma) {
1549                 if (vma->vm_ops && vma->vm_ops->get_policy) {
1550                         pol = vma->vm_ops->get_policy(vma, addr);
1551                 } else if (vma->vm_policy) {
1552                         pol = vma->vm_policy;
1553 
1554                         /*
1555                          * shmem_alloc_page() passes MPOL_F_SHARED policy with
1556                          * a pseudo vma whose vma->vm_ops=NULL. Take a reference
1557                          * count on these policies which will be dropped by
1558                          * mpol_cond_put() later
1559                          */
1560                         if (mpol_needs_cond_ref(pol))
1561                                 mpol_get(pol);
1562                 }
1563         }
1564 
1565         return pol;
1566 }
1567 
1568 /*
1569  * get_vma_policy(@vma, @addr)
1570  * @vma: virtual memory area whose policy is sought
1571  * @addr: address in @vma for shared policy lookup
1572  *
1573  * Returns effective policy for a VMA at specified address.
1574  * Falls back to current->mempolicy or system default policy, as necessary.
1575  * Shared policies [those marked as MPOL_F_SHARED] require an extra reference
1576  * count--added by the get_policy() vm_op, as appropriate--to protect against
1577  * freeing by another task.  It is the caller's responsibility to free the
1578  * extra reference for shared policies.
1579  */
1580 static struct mempolicy *get_vma_policy(struct vm_area_struct *vma,
1581                                                 unsigned long addr)
1582 {
1583         struct mempolicy *pol = __get_vma_policy(vma, addr);
1584 
1585         if (!pol)
1586                 pol = get_task_policy(current);
1587 
1588         return pol;
1589 }
1590 
1591 bool vma_policy_mof(struct vm_area_struct *vma)
1592 {
1593         struct mempolicy *pol;
1594 
1595         if (vma->vm_ops && vma->vm_ops->get_policy) {
1596                 bool ret = false;
1597 
1598                 pol = vma->vm_ops->get_policy(vma, vma->vm_start);
1599                 if (pol && (pol->flags & MPOL_F_MOF))
1600                         ret = true;
1601                 mpol_cond_put(pol);
1602 
1603                 return ret;
1604         }
1605 
1606         pol = vma->vm_policy;
1607         if (!pol)
1608                 pol = get_task_policy(current);
1609 
1610         return pol->flags & MPOL_F_MOF;
1611 }
1612 
1613 static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone)
1614 {
1615         enum zone_type dynamic_policy_zone = policy_zone;
1616 
1617         BUG_ON(dynamic_policy_zone == ZONE_MOVABLE);
1618 
1619         /*
1620          * if policy->v.nodes has movable memory only,
1621          * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only.
1622          *
1623          * policy->v.nodes is intersect with node_states[N_MEMORY].
1624          * so if the following test faile, it implies
1625          * policy->v.nodes has movable memory only.
1626          */
1627         if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY]))
1628                 dynamic_policy_zone = ZONE_MOVABLE;
1629 
1630         return zone >= dynamic_policy_zone;
1631 }
1632 
1633 /*
1634  * Return a nodemask representing a mempolicy for filtering nodes for
1635  * page allocation
1636  */
1637 static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy)
1638 {
1639         /* Lower zones don't get a nodemask applied for MPOL_BIND */
1640         if (unlikely(policy->mode == MPOL_BIND) &&
1641                         apply_policy_zone(policy, gfp_zone(gfp)) &&
1642                         cpuset_nodemask_valid_mems_allowed(&policy->v.nodes))
1643                 return &policy->v.nodes;
1644 
1645         return NULL;
1646 }
1647 
1648 /* Return a zonelist indicated by gfp for node representing a mempolicy */
1649 static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy,
1650         int nd)
1651 {
1652         switch (policy->mode) {
1653         case MPOL_PREFERRED:
1654                 if (!(policy->flags & MPOL_F_LOCAL))
1655                         nd = policy->v.preferred_node;
1656                 break;
1657         case MPOL_BIND:
1658                 /*
1659                  * Normally, MPOL_BIND allocations are node-local within the
1660                  * allowed nodemask.  However, if __GFP_THISNODE is set and the
1661                  * current node isn't part of the mask, we use the zonelist for
1662                  * the first node in the mask instead.
1663                  */
1664                 if (unlikely(gfp & __GFP_THISNODE) &&
1665                                 unlikely(!node_isset(nd, policy->v.nodes)))
1666                         nd = first_node(policy->v.nodes);
1667                 break;
1668         default:
1669                 BUG();
1670         }
1671         return node_zonelist(nd, gfp);
1672 }
1673 
1674 /* Do dynamic interleaving for a process */
1675 static unsigned interleave_nodes(struct mempolicy *policy)
1676 {
1677         unsigned nid, next;
1678         struct task_struct *me = current;
1679 
1680         nid = me->il_next;
1681         next = next_node(nid, policy->v.nodes);
1682         if (next >= MAX_NUMNODES)
1683                 next = first_node(policy->v.nodes);
1684         if (next < MAX_NUMNODES)
1685                 me->il_next = next;
1686         return nid;
1687 }
1688 
1689 /*
1690  * Depending on the memory policy provide a node from which to allocate the
1691  * next slab entry.
1692  */
1693 unsigned int mempolicy_slab_node(void)
1694 {
1695         struct mempolicy *policy;
1696         int node = numa_mem_id();
1697 
1698         if (in_interrupt())
1699                 return node;
1700 
1701         policy = current->mempolicy;
1702         if (!policy || policy->flags & MPOL_F_LOCAL)
1703                 return node;
1704 
1705         switch (policy->mode) {
1706         case MPOL_PREFERRED:
1707                 /*
1708                  * handled MPOL_F_LOCAL above
1709                  */
1710                 return policy->v.preferred_node;
1711 
1712         case MPOL_INTERLEAVE:
1713                 return interleave_nodes(policy);
1714 
1715         case MPOL_BIND: {
1716                 /*
1717                  * Follow bind policy behavior and start allocation at the
1718                  * first node.
1719                  */
1720                 struct zonelist *zonelist;
1721                 struct zone *zone;
1722                 enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL);
1723                 zonelist = &NODE_DATA(node)->node_zonelists[0];
1724                 (void)first_zones_zonelist(zonelist, highest_zoneidx,
1725                                                         &policy->v.nodes,
1726                                                         &zone);
1727                 return zone ? zone->node : node;
1728         }
1729 
1730         default:
1731                 BUG();
1732         }
1733 }
1734 
1735 /* Do static interleaving for a VMA with known offset. */
1736 static unsigned offset_il_node(struct mempolicy *pol,
1737                 struct vm_area_struct *vma, unsigned long off)
1738 {
1739         unsigned nnodes = nodes_weight(pol->v.nodes);
1740         unsigned target;
1741         int c;
1742         int nid = NUMA_NO_NODE;
1743 
1744         if (!nnodes)
1745                 return numa_node_id();
1746         target = (unsigned int)off % nnodes;
1747         c = 0;
1748         do {
1749                 nid = next_node(nid, pol->v.nodes);
1750                 c++;
1751         } while (c <= target);
1752         return nid;
1753 }
1754 
1755 /* Determine a node number for interleave */
1756 static inline unsigned interleave_nid(struct mempolicy *pol,
1757                  struct vm_area_struct *vma, unsigned long addr, int shift)
1758 {
1759         if (vma) {
1760                 unsigned long off;
1761 
1762                 /*
1763                  * for small pages, there is no difference between
1764                  * shift and PAGE_SHIFT, so the bit-shift is safe.
1765                  * for huge pages, since vm_pgoff is in units of small
1766                  * pages, we need to shift off the always 0 bits to get
1767                  * a useful offset.
1768                  */
1769                 BUG_ON(shift < PAGE_SHIFT);
1770                 off = vma->vm_pgoff >> (shift - PAGE_SHIFT);
1771                 off += (addr - vma->vm_start) >> shift;
1772                 return offset_il_node(pol, vma, off);
1773         } else
1774                 return interleave_nodes(pol);
1775 }
1776 
1777 /*
1778  * Return the bit number of a random bit set in the nodemask.
1779  * (returns NUMA_NO_NODE if nodemask is empty)
1780  */
1781 int node_random(const nodemask_t *maskp)
1782 {
1783         int w, bit = NUMA_NO_NODE;
1784 
1785         w = nodes_weight(*maskp);
1786         if (w)
1787                 bit = bitmap_ord_to_pos(maskp->bits,
1788                         get_random_int() % w, MAX_NUMNODES);
1789         return bit;
1790 }
1791 
1792 #ifdef CONFIG_HUGETLBFS
1793 /*
1794  * huge_zonelist(@vma, @addr, @gfp_flags, @mpol)
1795  * @vma: virtual memory area whose policy is sought
1796  * @addr: address in @vma for shared policy lookup and interleave policy
1797  * @gfp_flags: for requested zone
1798  * @mpol: pointer to mempolicy pointer for reference counted mempolicy
1799  * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask
1800  *
1801  * Returns a zonelist suitable for a huge page allocation and a pointer
1802  * to the struct mempolicy for conditional unref after allocation.
1803  * If the effective policy is 'BIND, returns a pointer to the mempolicy's
1804  * @nodemask for filtering the zonelist.
1805  *
1806  * Must be protected by read_mems_allowed_begin()
1807  */
1808 struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr,
1809                                 gfp_t gfp_flags, struct mempolicy **mpol,
1810                                 nodemask_t **nodemask)
1811 {
1812         struct zonelist *zl;
1813 
1814         *mpol = get_vma_policy(vma, addr);
1815         *nodemask = NULL;       /* assume !MPOL_BIND */
1816 
1817         if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) {
1818                 zl = node_zonelist(interleave_nid(*mpol, vma, addr,
1819                                 huge_page_shift(hstate_vma(vma))), gfp_flags);
1820         } else {
1821                 zl = policy_zonelist(gfp_flags, *mpol, numa_node_id());
1822                 if ((*mpol)->mode == MPOL_BIND)
1823                         *nodemask = &(*mpol)->v.nodes;
1824         }
1825         return zl;
1826 }
1827 
1828 /*
1829  * init_nodemask_of_mempolicy
1830  *
1831  * If the current task's mempolicy is "default" [NULL], return 'false'
1832  * to indicate default policy.  Otherwise, extract the policy nodemask
1833  * for 'bind' or 'interleave' policy into the argument nodemask, or
1834  * initialize the argument nodemask to contain the single node for
1835  * 'preferred' or 'local' policy and return 'true' to indicate presence
1836  * of non-default mempolicy.
1837  *
1838  * We don't bother with reference counting the mempolicy [mpol_get/put]
1839  * because the current task is examining it's own mempolicy and a task's
1840  * mempolicy is only ever changed by the task itself.
1841  *
1842  * N.B., it is the caller's responsibility to free a returned nodemask.
1843  */
1844 bool init_nodemask_of_mempolicy(nodemask_t *mask)
1845 {
1846         struct mempolicy *mempolicy;
1847         int nid;
1848 
1849         if (!(mask && current->mempolicy))
1850                 return false;
1851 
1852         task_lock(current);
1853         mempolicy = current->mempolicy;
1854         switch (mempolicy->mode) {
1855         case MPOL_PREFERRED:
1856                 if (mempolicy->flags & MPOL_F_LOCAL)
1857                         nid = numa_node_id();
1858                 else
1859                         nid = mempolicy->v.preferred_node;
1860                 init_nodemask_of_node(mask, nid);
1861                 break;
1862 
1863         case MPOL_BIND:
1864                 /* Fall through */
1865         case MPOL_INTERLEAVE:
1866                 *mask =  mempolicy->v.nodes;
1867                 break;
1868 
1869         default:
1870                 BUG();
1871         }
1872         task_unlock(current);
1873 
1874         return true;
1875 }
1876 #endif
1877 
1878 /*
1879  * mempolicy_nodemask_intersects
1880  *
1881  * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default
1882  * policy.  Otherwise, check for intersection between mask and the policy
1883  * nodemask for 'bind' or 'interleave' policy.  For 'perferred' or 'local'
1884  * policy, always return true since it may allocate elsewhere on fallback.
1885  *
1886  * Takes task_lock(tsk) to prevent freeing of its mempolicy.
1887  */
1888 bool mempolicy_nodemask_intersects(struct task_struct *tsk,
1889                                         const nodemask_t *mask)
1890 {
1891         struct mempolicy *mempolicy;
1892         bool ret = true;
1893 
1894         if (!mask)
1895                 return ret;
1896         task_lock(tsk);
1897         mempolicy = tsk->mempolicy;
1898         if (!mempolicy)
1899                 goto out;
1900 
1901         switch (mempolicy->mode) {
1902         case MPOL_PREFERRED:
1903                 /*
1904                  * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to
1905                  * allocate from, they may fallback to other nodes when oom.
1906                  * Thus, it's possible for tsk to have allocated memory from
1907                  * nodes in mask.
1908                  */
1909                 break;
1910         case MPOL_BIND:
1911         case MPOL_INTERLEAVE:
1912                 ret = nodes_intersects(mempolicy->v.nodes, *mask);
1913                 break;
1914         default:
1915                 BUG();
1916         }
1917 out:
1918         task_unlock(tsk);
1919         return ret;
1920 }
1921 
1922 /* Allocate a page in interleaved policy.
1923    Own path because it needs to do special accounting. */
1924 static struct page *alloc_page_interleave(gfp_t gfp, unsigned order,
1925                                         unsigned nid)
1926 {
1927         struct zonelist *zl;
1928         struct page *page;
1929 
1930         zl = node_zonelist(nid, gfp);
1931         page = __alloc_pages(gfp, order, zl);
1932         if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0]))
1933                 inc_zone_page_state(page, NUMA_INTERLEAVE_HIT);
1934         return page;
1935 }
1936 
1937 /**
1938  *      alloc_pages_vma - Allocate a page for a VMA.
1939  *
1940  *      @gfp:
1941  *      %GFP_USER    user allocation.
1942  *      %GFP_KERNEL  kernel allocations,
1943  *      %GFP_HIGHMEM highmem/user allocations,
1944  *      %GFP_FS      allocation should not call back into a file system.
1945  *      %GFP_ATOMIC  don't sleep.
1946  *
1947  *      @order:Order of the GFP allocation.
1948  *      @vma:  Pointer to VMA or NULL if not available.
1949  *      @addr: Virtual Address of the allocation. Must be inside the VMA.
1950  *      @node: Which node to prefer for allocation (modulo policy).
1951  *      @hugepage: for hugepages try only the preferred node if possible
1952  *
1953  *      This function allocates a page from the kernel page pool and applies
1954  *      a NUMA policy associated with the VMA or the current process.
1955  *      When VMA is not NULL caller must hold down_read on the mmap_sem of the
1956  *      mm_struct of the VMA to prevent it from going away. Should be used for
1957  *      all allocations for pages that will be mapped into user space. Returns
1958  *      NULL when no page can be allocated.
1959  */
1960 struct page *
1961 alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma,
1962                 unsigned long addr, int node, bool hugepage)
1963 {
1964         struct mempolicy *pol;
1965         struct page *page;
1966         unsigned int cpuset_mems_cookie;
1967         struct zonelist *zl;
1968         nodemask_t *nmask;
1969 
1970 retry_cpuset:
1971         pol = get_vma_policy(vma, addr);
1972         cpuset_mems_cookie = read_mems_allowed_begin();
1973 
1974         if (pol->mode == MPOL_INTERLEAVE) {
1975                 unsigned nid;
1976 
1977                 nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order);
1978                 mpol_cond_put(pol);
1979                 page = alloc_page_interleave(gfp, order, nid);
1980                 goto out;
1981         }
1982 
1983         if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) {
1984                 int hpage_node = node;
1985 
1986                 /*
1987                  * For hugepage allocation and non-interleave policy which
1988                  * allows the current node (or other explicitly preferred
1989                  * node) we only try to allocate from the current/preferred
1990                  * node and don't fall back to other nodes, as the cost of
1991                  * remote accesses would likely offset THP benefits.
1992                  *
1993                  * If the policy is interleave, or does not allow the current
1994                  * node in its nodemask, we allocate the standard way.
1995                  */
1996                 if (pol->mode == MPOL_PREFERRED &&
1997                                                 !(pol->flags & MPOL_F_LOCAL))
1998                         hpage_node = pol->v.preferred_node;
1999 
2000                 nmask = policy_nodemask(gfp, pol);
2001                 if (!nmask || node_isset(hpage_node, *nmask)) {
2002                         mpol_cond_put(pol);
2003                         page = alloc_pages_exact_node(hpage_node,
2004                                                 gfp | __GFP_THISNODE, order);
2005                         goto out;
2006                 }
2007         }
2008 
2009         nmask = policy_nodemask(gfp, pol);
2010         zl = policy_zonelist(gfp, pol, node);
2011         mpol_cond_put(pol);
2012         page = __alloc_pages_nodemask(gfp, order, zl, nmask);
2013 out:
2014         if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
2015                 goto retry_cpuset;
2016         return page;
2017 }
2018 
2019 /**
2020  *      alloc_pages_current - Allocate pages.
2021  *
2022  *      @gfp:
2023  *              %GFP_USER   user allocation,
2024  *              %GFP_KERNEL kernel allocation,
2025  *              %GFP_HIGHMEM highmem allocation,
2026  *              %GFP_FS     don't call back into a file system.
2027  *              %GFP_ATOMIC don't sleep.
2028  *      @order: Power of two of allocation size in pages. 0 is a single page.
2029  *
2030  *      Allocate a page from the kernel page pool.  When not in
2031  *      interrupt context and apply the current process NUMA policy.
2032  *      Returns NULL when no page can be allocated.
2033  *
2034  *      Don't call cpuset_update_task_memory_state() unless
2035  *      1) it's ok to take cpuset_sem (can WAIT), and
2036  *      2) allocating for current task (not interrupt).
2037  */
2038 struct page *alloc_pages_current(gfp_t gfp, unsigned order)
2039 {
2040         struct mempolicy *pol = &default_policy;
2041         struct page *page;
2042         unsigned int cpuset_mems_cookie;
2043 
2044         if (!in_interrupt() && !(gfp & __GFP_THISNODE))
2045                 pol = get_task_policy(current);
2046 
2047 retry_cpuset:
2048         cpuset_mems_cookie = read_mems_allowed_begin();
2049 
2050         /*
2051          * No reference counting needed for current->mempolicy
2052          * nor system default_policy
2053          */
2054         if (pol->mode == MPOL_INTERLEAVE)
2055                 page = alloc_page_interleave(gfp, order, interleave_nodes(pol));
2056         else
2057                 page = __alloc_pages_nodemask(gfp, order,
2058                                 policy_zonelist(gfp, pol, numa_node_id()),
2059                                 policy_nodemask(gfp, pol));
2060 
2061         if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
2062                 goto retry_cpuset;
2063 
2064         return page;
2065 }
2066 EXPORT_SYMBOL(alloc_pages_current);
2067 
2068 int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst)
2069 {
2070         struct mempolicy *pol = mpol_dup(vma_policy(src));
2071 
2072         if (IS_ERR(pol))
2073                 return PTR_ERR(pol);
2074         dst->vm_policy = pol;
2075         return 0;
2076 }
2077 
2078 /*
2079  * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it
2080  * rebinds the mempolicy its copying by calling mpol_rebind_policy()
2081  * with the mems_allowed returned by cpuset_mems_allowed().  This
2082  * keeps mempolicies cpuset relative after its cpuset moves.  See
2083  * further kernel/cpuset.c update_nodemask().
2084  *
2085  * current's mempolicy may be rebinded by the other task(the task that changes
2086  * cpuset's mems), so we needn't do rebind work for current task.
2087  */
2088 
2089 /* Slow path of a mempolicy duplicate */
2090 struct mempolicy *__mpol_dup(struct mempolicy *old)
2091 {
2092         struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2093 
2094         if (!new)
2095                 return ERR_PTR(-ENOMEM);
2096 
2097         /* task's mempolicy is protected by alloc_lock */
2098         if (old == current->mempolicy) {
2099                 task_lock(current);
2100                 *new = *old;
2101                 task_unlock(current);
2102         } else
2103                 *new = *old;
2104 
2105         if (current_cpuset_is_being_rebound()) {
2106                 nodemask_t mems = cpuset_mems_allowed(current);
2107                 if (new->flags & MPOL_F_REBINDING)
2108                         mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2);
2109                 else
2110                         mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE);
2111         }
2112         atomic_set(&new->refcnt, 1);
2113         return new;
2114 }
2115 
2116 /* Slow path of a mempolicy comparison */
2117 bool __mpol_equal(struct mempolicy *a, struct mempolicy *b)
2118 {
2119         if (!a || !b)
2120                 return false;
2121         if (a->mode != b->mode)
2122                 return false;
2123         if (a->flags != b->flags)
2124                 return false;
2125         if (mpol_store_user_nodemask(a))
2126                 if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask))
2127                         return false;
2128 
2129         switch (a->mode) {
2130         case MPOL_BIND:
2131                 /* Fall through */
2132         case MPOL_INTERLEAVE:
2133                 return !!nodes_equal(a->v.nodes, b->v.nodes);
2134         case MPOL_PREFERRED:
2135                 return a->v.preferred_node == b->v.preferred_node;
2136         default:
2137                 BUG();
2138                 return false;
2139         }
2140 }
2141 
2142 /*
2143  * Shared memory backing store policy support.
2144  *
2145  * Remember policies even when nobody has shared memory mapped.
2146  * The policies are kept in Red-Black tree linked from the inode.
2147  * They are protected by the sp->lock spinlock, which should be held
2148  * for any accesses to the tree.
2149  */
2150 
2151 /* lookup first element intersecting start-end */
2152 /* Caller holds sp->lock */
2153 static struct sp_node *
2154 sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end)
2155 {
2156         struct rb_node *n = sp->root.rb_node;
2157 
2158         while (n) {
2159                 struct sp_node *p = rb_entry(n, struct sp_node, nd);
2160 
2161                 if (start >= p->end)
2162                         n = n->rb_right;
2163                 else if (end <= p->start)
2164                         n = n->rb_left;
2165                 else
2166                         break;
2167         }
2168         if (!n)
2169                 return NULL;
2170         for (;;) {
2171                 struct sp_node *w = NULL;
2172                 struct rb_node *prev = rb_prev(n);
2173                 if (!prev)
2174                         break;
2175                 w = rb_entry(prev, struct sp_node, nd);
2176                 if (w->end <= start)
2177                         break;
2178                 n = prev;
2179         }
2180         return rb_entry(n, struct sp_node, nd);
2181 }
2182 
2183 /* Insert a new shared policy into the list. */
2184 /* Caller holds sp->lock */
2185 static void sp_insert(struct shared_policy *sp, struct sp_node *new)
2186 {
2187         struct rb_node **p = &sp->root.rb_node;
2188         struct rb_node *parent = NULL;
2189         struct sp_node *nd;
2190 
2191         while (*p) {
2192                 parent = *p;
2193                 nd = rb_entry(parent, struct sp_node, nd);
2194                 if (new->start < nd->start)
2195                         p = &(*p)->rb_left;
2196                 else if (new->end > nd->end)
2197                         p = &(*p)->rb_right;
2198                 else
2199                         BUG();
2200         }
2201         rb_link_node(&new->nd, parent, p);
2202         rb_insert_color(&new->nd, &sp->root);
2203         pr_debug("inserting %lx-%lx: %d\n", new->start, new->end,
2204                  new->policy ? new->policy->mode : 0);
2205 }
2206 
2207 /* Find shared policy intersecting idx */
2208 struct mempolicy *
2209 mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx)
2210 {
2211         struct mempolicy *pol = NULL;
2212         struct sp_node *sn;
2213 
2214         if (!sp->root.rb_node)
2215                 return NULL;
2216         spin_lock(&sp->lock);
2217         sn = sp_lookup(sp, idx, idx+1);
2218         if (sn) {
2219                 mpol_get(sn->policy);
2220                 pol = sn->policy;
2221         }
2222         spin_unlock(&sp->lock);
2223         return pol;
2224 }
2225 
2226 static void sp_free(struct sp_node *n)
2227 {
2228         mpol_put(n->policy);
2229         kmem_cache_free(sn_cache, n);
2230 }
2231 
2232 /**
2233  * mpol_misplaced - check whether current page node is valid in policy
2234  *
2235  * @page: page to be checked
2236  * @vma: vm area where page mapped
2237  * @addr: virtual address where page mapped
2238  *
2239  * Lookup current policy node id for vma,addr and "compare to" page's
2240  * node id.
2241  *
2242  * Returns:
2243  *      -1      - not misplaced, page is in the right node
2244  *      node    - node id where the page should be
2245  *
2246  * Policy determination "mimics" alloc_page_vma().
2247  * Called from fault path where we know the vma and faulting address.
2248  */
2249 int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr)
2250 {
2251         struct mempolicy *pol;
2252         struct zone *zone;
2253         int curnid = page_to_nid(page);
2254         unsigned long pgoff;
2255         int thiscpu = raw_smp_processor_id();
2256         int thisnid = cpu_to_node(thiscpu);
2257         int polnid = -1;
2258         int ret = -1;
2259 
2260         BUG_ON(!vma);
2261 
2262         pol = get_vma_policy(vma, addr);
2263         if (!(pol->flags & MPOL_F_MOF))
2264                 goto out;
2265 
2266         switch (pol->mode) {
2267         case MPOL_INTERLEAVE:
2268                 BUG_ON(addr >= vma->vm_end);
2269                 BUG_ON(addr < vma->vm_start);
2270 
2271                 pgoff = vma->vm_pgoff;
2272                 pgoff += (addr - vma->vm_start) >> PAGE_SHIFT;
2273                 polnid = offset_il_node(pol, vma, pgoff);
2274                 break;
2275 
2276         case MPOL_PREFERRED:
2277                 if (pol->flags & MPOL_F_LOCAL)
2278                         polnid = numa_node_id();
2279                 else
2280                         polnid = pol->v.preferred_node;
2281                 break;
2282 
2283         case MPOL_BIND:
2284                 /*
2285                  * allows binding to multiple nodes.
2286                  * use current page if in policy nodemask,
2287                  * else select nearest allowed node, if any.
2288                  * If no allowed nodes, use current [!misplaced].
2289                  */
2290                 if (node_isset(curnid, pol->v.nodes))
2291                         goto out;
2292                 (void)first_zones_zonelist(
2293                                 node_zonelist(numa_node_id(), GFP_HIGHUSER),
2294                                 gfp_zone(GFP_HIGHUSER),
2295                                 &pol->v.nodes, &zone);
2296                 polnid = zone->node;
2297                 break;
2298 
2299         default:
2300                 BUG();
2301         }
2302 
2303         /* Migrate the page towards the node whose CPU is referencing it */
2304         if (pol->flags & MPOL_F_MORON) {
2305                 polnid = thisnid;
2306 
2307                 if (!should_numa_migrate_memory(current, page, curnid, thiscpu))
2308                         goto out;
2309         }
2310 
2311         if (curnid != polnid)
2312                 ret = polnid;
2313 out:
2314         mpol_cond_put(pol);
2315 
2316         return ret;
2317 }
2318 
2319 static void sp_delete(struct shared_policy *sp, struct sp_node *n)
2320 {
2321         pr_debug("deleting %lx-l%lx\n", n->start, n->end);
2322         rb_erase(&n->nd, &sp->root);
2323         sp_free(n);
2324 }
2325 
2326 static void sp_node_init(struct sp_node *node, unsigned long start,
2327                         unsigned long end, struct mempolicy *pol)
2328 {
2329         node->start = start;
2330         node->end = end;
2331         node->policy = pol;
2332 }
2333 
2334 static struct sp_node *sp_alloc(unsigned long start, unsigned long end,
2335                                 struct mempolicy *pol)
2336 {
2337         struct sp_node *n;
2338         struct mempolicy *newpol;
2339 
2340         n = kmem_cache_alloc(sn_cache, GFP_KERNEL);
2341         if (!n)
2342                 return NULL;
2343 
2344         newpol = mpol_dup(pol);
2345         if (IS_ERR(newpol)) {
2346                 kmem_cache_free(sn_cache, n);
2347                 return NULL;
2348         }
2349         newpol->flags |= MPOL_F_SHARED;
2350         sp_node_init(n, start, end, newpol);
2351 
2352         return n;
2353 }
2354 
2355 /* Replace a policy range. */
2356 static int shared_policy_replace(struct shared_policy *sp, unsigned long start,
2357                                  unsigned long end, struct sp_node *new)
2358 {
2359         struct sp_node *n;
2360         struct sp_node *n_new = NULL;
2361         struct mempolicy *mpol_new = NULL;
2362         int ret = 0;
2363 
2364 restart:
2365         spin_lock(&sp->lock);
2366         n = sp_lookup(sp, start, end);
2367         /* Take care of old policies in the same range. */
2368         while (n && n->start < end) {
2369                 struct rb_node *next = rb_next(&n->nd);
2370                 if (n->start >= start) {
2371                         if (n->end <= end)
2372                                 sp_delete(sp, n);
2373                         else
2374                                 n->start = end;
2375                 } else {
2376                         /* Old policy spanning whole new range. */
2377                         if (n->end > end) {
2378                                 if (!n_new)
2379                                         goto alloc_new;
2380 
2381                                 *mpol_new = *n->policy;
2382                                 atomic_set(&mpol_new->refcnt, 1);
2383                                 sp_node_init(n_new, end, n->end, mpol_new);
2384                                 n->end = start;
2385                                 sp_insert(sp, n_new);
2386                                 n_new = NULL;
2387                                 mpol_new = NULL;
2388                                 break;
2389                         } else
2390                                 n->end = start;
2391                 }
2392                 if (!next)
2393                         break;
2394                 n = rb_entry(next, struct sp_node, nd);
2395         }
2396         if (new)
2397                 sp_insert(sp, new);
2398         spin_unlock(&sp->lock);
2399         ret = 0;
2400 
2401 err_out:
2402         if (mpol_new)
2403                 mpol_put(mpol_new);
2404         if (n_new)
2405                 kmem_cache_free(sn_cache, n_new);
2406 
2407         return ret;
2408 
2409 alloc_new:
2410         spin_unlock(&sp->lock);
2411         ret = -ENOMEM;
2412         n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL);
2413         if (!n_new)
2414                 goto err_out;
2415         mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL);
2416         if (!mpol_new)
2417                 goto err_out;
2418         goto restart;
2419 }
2420 
2421 /**
2422  * mpol_shared_policy_init - initialize shared policy for inode
2423  * @sp: pointer to inode shared policy
2424  * @mpol:  struct mempolicy to install
2425  *
2426  * Install non-NULL @mpol in inode's shared policy rb-tree.
2427  * On entry, the current task has a reference on a non-NULL @mpol.
2428  * This must be released on exit.
2429  * This is called at get_inode() calls and we can use GFP_KERNEL.
2430  */
2431 void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
2432 {
2433         int ret;
2434 
2435         sp->root = RB_ROOT;             /* empty tree == default mempolicy */
2436         spin_lock_init(&sp->lock);
2437 
2438         if (mpol) {
2439                 struct vm_area_struct pvma;
2440                 struct mempolicy *new;
2441                 NODEMASK_SCRATCH(scratch);
2442 
2443                 if (!scratch)
2444                         goto put_mpol;
2445                 /* contextualize the tmpfs mount point mempolicy */
2446                 new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
2447                 if (IS_ERR(new))
2448                         goto free_scratch; /* no valid nodemask intersection */
2449 
2450                 task_lock(current);
2451                 ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch);
2452                 task_unlock(current);
2453                 if (ret)
2454                         goto put_new;
2455 
2456                 /* Create pseudo-vma that contains just the policy */
2457                 memset(&pvma, 0, sizeof(struct vm_area_struct));
2458                 pvma.vm_end = TASK_SIZE;        /* policy covers entire file */
2459                 mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
2460 
2461 put_new:
2462                 mpol_put(new);                  /* drop initial ref */
2463 free_scratch:
2464                 NODEMASK_SCRATCH_FREE(scratch);
2465 put_mpol:
2466                 mpol_put(mpol); /* drop our incoming ref on sb mpol */
2467         }
2468 }
2469 
2470 int mpol_set_shared_policy(struct shared_policy *info,
2471                         struct vm_area_struct *vma, struct mempolicy *npol)
2472 {
2473         int err;
2474         struct sp_node *new = NULL;
2475         unsigned long sz = vma_pages(vma);
2476 
2477         pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n",
2478                  vma->vm_pgoff,
2479                  sz, npol ? npol->mode : -1,
2480                  npol ? npol->flags : -1,
2481                  npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE);
2482 
2483         if (npol) {
2484                 new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol);
2485                 if (!new)
2486                         return -ENOMEM;
2487         }
2488         err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new);
2489         if (err && new)
2490                 sp_free(new);
2491         return err;
2492 }
2493 
2494 /* Free a backing policy store on inode delete. */
2495 void mpol_free_shared_policy(struct shared_policy *p)
2496 {
2497         struct sp_node *n;
2498         struct rb_node *next;
2499 
2500         if (!p->root.rb_node)
2501                 return;
2502         spin_lock(&p->lock);
2503         next = rb_first(&p->root);
2504         while (next) {
2505                 n = rb_entry(next, struct sp_node, nd);
2506                 next = rb_next(&n->nd);
2507                 sp_delete(p, n);
2508         }
2509         spin_unlock(&p->lock);
2510 }
2511 
2512 #ifdef CONFIG_NUMA_BALANCING
2513 static int __initdata numabalancing_override;
2514 
2515 static void __init check_numabalancing_enable(void)
2516 {
2517         bool numabalancing_default = false;
2518 
2519         if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED))
2520                 numabalancing_default = true;
2521 
2522         /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */
2523         if (numabalancing_override)
2524                 set_numabalancing_state(numabalancing_override == 1);
2525 
2526         if (num_online_nodes() > 1 && !numabalancing_override) {
2527                 pr_info("%s automatic NUMA balancing. "
2528                         "Configure with numa_balancing= or the "
2529                         "kernel.numa_balancing sysctl",
2530                         numabalancing_default ? "Enabling" : "Disabling");
2531                 set_numabalancing_state(numabalancing_default);
2532         }
2533 }
2534 
2535 static int __init setup_numabalancing(char *str)
2536 {
2537         int ret = 0;
2538         if (!str)
2539                 goto out;
2540 
2541         if (!strcmp(str, "enable")) {
2542                 numabalancing_override = 1;
2543                 ret = 1;
2544         } else if (!strcmp(str, "disable")) {
2545                 numabalancing_override = -1;
2546                 ret = 1;
2547         }
2548 out:
2549         if (!ret)
2550                 pr_warn("Unable to parse numa_balancing=\n");
2551 
2552         return ret;
2553 }
2554 __setup("numa_balancing=", setup_numabalancing);
2555 #else
2556 static inline void __init check_numabalancing_enable(void)
2557 {
2558 }
2559 #endif /* CONFIG_NUMA_BALANCING */
2560 
2561 /* assumes fs == KERNEL_DS */
2562 void __init numa_policy_init(void)
2563 {
2564         nodemask_t interleave_nodes;
2565         unsigned long largest = 0;
2566         int nid, prefer = 0;
2567 
2568         policy_cache = kmem_cache_create("numa_policy",
2569                                          sizeof(struct mempolicy),
2570                                          0, SLAB_PANIC, NULL);
2571 
2572         sn_cache = kmem_cache_create("shared_policy_node",
2573                                      sizeof(struct sp_node),
2574                                      0, SLAB_PANIC, NULL);
2575 
2576         for_each_node(nid) {
2577                 preferred_node_policy[nid] = (struct mempolicy) {
2578                         .refcnt = ATOMIC_INIT(1),
2579                         .mode = MPOL_PREFERRED,
2580                         .flags = MPOL_F_MOF | MPOL_F_MORON,
2581                         .v = { .preferred_node = nid, },
2582                 };
2583         }
2584 
2585         /*
2586          * Set interleaving policy for system init. Interleaving is only
2587          * enabled across suitably sized nodes (default is >= 16MB), or
2588          * fall back to the largest node if they're all smaller.
2589          */
2590         nodes_clear(interleave_nodes);
2591         for_each_node_state(nid, N_MEMORY) {
2592                 unsigned long total_pages = node_present_pages(nid);
2593 
2594                 /* Preserve the largest node */
2595                 if (largest < total_pages) {
2596                         largest = total_pages;
2597                         prefer = nid;
2598                 }
2599 
2600                 /* Interleave this node? */
2601                 if ((total_pages << PAGE_SHIFT) >= (16 << 20))
2602                         node_set(nid, interleave_nodes);
2603         }
2604 
2605         /* All too small, use the largest */
2606         if (unlikely(nodes_empty(interleave_nodes)))
2607                 node_set(prefer, interleave_nodes);
2608 
2609         if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes))
2610                 pr_err("%s: interleaving failed\n", __func__);
2611 
2612         check_numabalancing_enable();
2613 }
2614 
2615 /* Reset policy of current process to default */
2616 void numa_default_policy(void)
2617 {
2618         do_set_mempolicy(MPOL_DEFAULT, 0, NULL);
2619 }
2620 
2621 /*
2622  * Parse and format mempolicy from/to strings
2623  */
2624 
2625 /*
2626  * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag.
2627  */
2628 static const char * const policy_modes[] =
2629 {
2630         [MPOL_DEFAULT]    = "default",
2631         [MPOL_PREFERRED]  = "prefer",
2632         [MPOL_BIND]       = "bind",
2633         [MPOL_INTERLEAVE] = "interleave",
2634         [MPOL_LOCAL]      = "local",
2635 };
2636 
2637 
2638 #ifdef CONFIG_TMPFS
2639 /**
2640  * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option.
2641  * @str:  string containing mempolicy to parse
2642  * @mpol:  pointer to struct mempolicy pointer, returned on success.
2643  *
2644  * Format of input:
2645  *      <mode>[=<flags>][:<nodelist>]
2646  *
2647  * On success, returns 0, else 1
2648  */
2649 int mpol_parse_str(char *str, struct mempolicy **mpol)
2650 {
2651         struct mempolicy *new = NULL;
2652         unsigned short mode;
2653         unsigned short mode_flags;
2654         nodemask_t nodes;
2655         char *nodelist = strchr(str, ':');
2656         char *flags = strchr(str, '=');
2657         int err = 1;
2658 
2659         if (nodelist) {
2660                 /* NUL-terminate mode or flags string */
2661                 *nodelist++ = '\0';
2662                 if (nodelist_parse(nodelist, nodes))
2663                         goto out;
2664                 if (!nodes_subset(nodes, node_states[N_MEMORY]))
2665                         goto out;
2666         } else
2667                 nodes_clear(nodes);
2668 
2669         if (flags)
2670                 *flags++ = '\0';        /* terminate mode string */
2671 
2672         for (mode = 0; mode < MPOL_MAX; mode++) {
2673                 if (!strcmp(str, policy_modes[mode])) {
2674                         break;
2675                 }
2676         }
2677         if (mode >= MPOL_MAX)
2678                 goto out;
2679 
2680         switch (mode) {
2681         case MPOL_PREFERRED:
2682                 /*
2683                  * Insist on a nodelist of one node only
2684                  */
2685                 if (nodelist) {
2686                         char *rest = nodelist;
2687                         while (isdigit(*rest))
2688                                 rest++;
2689                         if (*rest)
2690                                 goto out;
2691                 }
2692                 break;
2693         case MPOL_INTERLEAVE:
2694                 /*
2695                  * Default to online nodes with memory if no nodelist
2696                  */
2697                 if (!nodelist)
2698                         nodes = node_states[N_MEMORY];
2699                 break;
2700         case MPOL_LOCAL:
2701                 /*
2702                  * Don't allow a nodelist;  mpol_new() checks flags
2703                  */
2704                 if (nodelist)
2705                         goto out;
2706                 mode = MPOL_PREFERRED;
2707                 break;
2708         case MPOL_DEFAULT:
2709                 /*
2710                  * Insist on a empty nodelist
2711                  */
2712                 if (!nodelist)
2713                         err = 0;
2714                 goto out;
2715         case MPOL_BIND:
2716                 /*
2717                  * Insist on a nodelist
2718                  */
2719                 if (!nodelist)
2720                         goto out;
2721         }
2722 
2723         mode_flags = 0;
2724         if (flags) {
2725                 /*
2726                  * Currently, we only support two mutually exclusive
2727                  * mode flags.
2728                  */
2729                 if (!strcmp(flags, "static"))
2730                         mode_flags |= MPOL_F_STATIC_NODES;
2731                 else if (!strcmp(flags, "relative"))
2732                         mode_flags |= MPOL_F_RELATIVE_NODES;
2733                 else
2734                         goto out;
2735         }
2736 
2737         new = mpol_new(mode, mode_flags, &nodes);
2738         if (IS_ERR(new))
2739                 goto out;
2740 
2741         /*
2742          * Save nodes for mpol_to_str() to show the tmpfs mount options
2743          * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo.
2744          */
2745         if (mode != MPOL_PREFERRED)
2746                 new->v.nodes = nodes;
2747         else if (nodelist)
2748                 new->v.preferred_node = first_node(nodes);
2749         else
2750                 new->flags |= MPOL_F_LOCAL;
2751 
2752         /*
2753          * Save nodes for contextualization: this will be used to "clone"
2754          * the mempolicy in a specific context [cpuset] at a later time.
2755          */
2756         new->w.user_nodemask = nodes;
2757 
2758         err = 0;
2759 
2760 out:
2761         /* Restore string for error message */
2762         if (nodelist)
2763                 *--nodelist = ':';
2764         if (flags)
2765                 *--flags = '=';
2766         if (!err)
2767                 *mpol = new;
2768         return err;
2769 }
2770 #endif /* CONFIG_TMPFS */
2771 
2772 /**
2773  * mpol_to_str - format a mempolicy structure for printing
2774  * @buffer:  to contain formatted mempolicy string
2775  * @maxlen:  length of @buffer
2776  * @pol:  pointer to mempolicy to be formatted
2777  *
2778  * Convert @pol into a string.  If @buffer is too short, truncate the string.
2779  * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the
2780  * longest flag, "relative", and to display at least a few node ids.
2781  */
2782 void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
2783 {
2784         char *p = buffer;
2785         nodemask_t nodes = NODE_MASK_NONE;
2786         unsigned short mode = MPOL_DEFAULT;
2787         unsigned short flags = 0;
2788 
2789         if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) {
2790                 mode = pol->mode;
2791                 flags = pol->flags;
2792         }
2793 
2794         switch (mode) {
2795         case MPOL_DEFAULT:
2796                 break;
2797         case MPOL_PREFERRED:
2798                 if (flags & MPOL_F_LOCAL)
2799                         mode = MPOL_LOCAL;
2800                 else
2801                         node_set(pol->v.preferred_node, nodes);
2802                 break;
2803         case MPOL_BIND:
2804         case MPOL_INTERLEAVE:
2805                 nodes = pol->v.nodes;
2806                 break;
2807         default:
2808                 WARN_ON_ONCE(1);
2809                 snprintf(p, maxlen, "unknown");
2810                 return;
2811         }
2812 
2813         p += snprintf(p, maxlen, "%s", policy_modes[mode]);
2814 
2815         if (flags & MPOL_MODE_FLAGS) {
2816                 p += snprintf(p, buffer + maxlen - p, "=");
2817 
2818                 /*
2819                  * Currently, the only defined flags are mutually exclusive
2820                  */
2821                 if (flags & MPOL_F_STATIC_NODES)
2822                         p += snprintf(p, buffer + maxlen - p, "static");
2823                 else if (flags & MPOL_F_RELATIVE_NODES)
2824                         p += snprintf(p, buffer + maxlen - p, "relative");
2825         }
2826 
2827         if (!nodes_empty(nodes))
2828                 p += scnprintf(p, buffer + maxlen - p, ":%*pbl",
2829                                nodemask_pr_args(&nodes));
2830 }
2831 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp