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