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