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> | |
1da177e4 LT |
76 | #include <linux/slab.h> |
77 | #include <linux/string.h> | |
b95f1b31 | 78 | #include <linux/export.h> |
b488893a | 79 | #include <linux/nsproxy.h> |
1da177e4 LT |
80 | #include <linux/interrupt.h> |
81 | #include <linux/init.h> | |
82 | #include <linux/compat.h> | |
dc9aa5b9 | 83 | #include <linux/swap.h> |
1a75a6c8 CL |
84 | #include <linux/seq_file.h> |
85 | #include <linux/proc_fs.h> | |
b20a3503 | 86 | #include <linux/migrate.h> |
62b61f61 | 87 | #include <linux/ksm.h> |
95a402c3 | 88 | #include <linux/rmap.h> |
86c3a764 | 89 | #include <linux/security.h> |
dbcb0f19 | 90 | #include <linux/syscalls.h> |
095f1fc4 | 91 | #include <linux/ctype.h> |
6d9c285a | 92 | #include <linux/mm_inline.h> |
dc9aa5b9 | 93 | |
1da177e4 LT |
94 | #include <asm/tlbflush.h> |
95 | #include <asm/uaccess.h> | |
778d3b0f | 96 | #include <linux/random.h> |
1da177e4 | 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 */ |
dc9aa5b9 | 103 | |
fcc234f8 PE |
104 | static struct kmem_cache *policy_cache; |
105 | static struct kmem_cache *sn_cache; | |
1da177e4 | 106 | |
1da177e4 LT |
107 | /* Highest zone. An specific allocation for a zone below that is not |
108 | policied. */ | |
6267276f | 109 | enum zone_type policy_zone = 0; |
1da177e4 | 110 | |
bea904d5 LS |
111 | /* |
112 | * run-time system-wide default policy => local allocation | |
113 | */ | |
e754d79d | 114 | static struct mempolicy default_policy = { |
1da177e4 | 115 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
bea904d5 | 116 | .mode = MPOL_PREFERRED, |
fc36b8d3 | 117 | .flags = MPOL_F_LOCAL, |
1da177e4 LT |
118 | }; |
119 | ||
37012946 DR |
120 | static const struct mempolicy_operations { |
121 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
708c1bbc MX |
122 | /* |
123 | * If read-side task has no lock to protect task->mempolicy, write-side | |
124 | * task will rebind the task->mempolicy by two step. The first step is | |
125 | * setting all the newly nodes, and the second step is cleaning all the | |
126 | * disallowed nodes. In this way, we can avoid finding no node to alloc | |
127 | * page. | |
128 | * If we have a lock to protect task->mempolicy in read-side, we do | |
129 | * rebind directly. | |
130 | * | |
131 | * step: | |
132 | * MPOL_REBIND_ONCE - do rebind work at once | |
133 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
134 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
135 | */ | |
136 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes, | |
137 | enum mpol_rebind_step step); | |
37012946 DR |
138 | } mpol_ops[MPOL_MAX]; |
139 | ||
19770b32 | 140 | /* Check that the nodemask contains at least one populated zone */ |
37012946 | 141 | static int is_valid_nodemask(const nodemask_t *nodemask) |
1da177e4 | 142 | { |
19770b32 | 143 | int nd, k; |
1da177e4 | 144 | |
19770b32 MG |
145 | for_each_node_mask(nd, *nodemask) { |
146 | struct zone *z; | |
147 | ||
148 | for (k = 0; k <= policy_zone; k++) { | |
149 | z = &NODE_DATA(nd)->node_zones[k]; | |
150 | if (z->present_pages > 0) | |
151 | return 1; | |
dd942ae3 | 152 | } |
8af5e2eb | 153 | } |
19770b32 MG |
154 | |
155 | return 0; | |
1da177e4 LT |
156 | } |
157 | ||
f5b087b5 DR |
158 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
159 | { | |
6d556294 | 160 | return pol->flags & MPOL_MODE_FLAGS; |
4c50bc01 DR |
161 | } |
162 | ||
163 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
164 | const nodemask_t *rel) | |
165 | { | |
166 | nodemask_t tmp; | |
167 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
168 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
169 | } |
170 | ||
37012946 DR |
171 | static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes) |
172 | { | |
173 | if (nodes_empty(*nodes)) | |
174 | return -EINVAL; | |
175 | pol->v.nodes = *nodes; | |
176 | return 0; | |
177 | } | |
178 | ||
179 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
180 | { | |
181 | if (!nodes) | |
fc36b8d3 | 182 | pol->flags |= MPOL_F_LOCAL; /* local allocation */ |
37012946 DR |
183 | else if (nodes_empty(*nodes)) |
184 | return -EINVAL; /* no allowed nodes */ | |
185 | else | |
186 | pol->v.preferred_node = first_node(*nodes); | |
187 | return 0; | |
188 | } | |
189 | ||
190 | static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes) | |
191 | { | |
192 | if (!is_valid_nodemask(nodes)) | |
193 | return -EINVAL; | |
194 | pol->v.nodes = *nodes; | |
195 | return 0; | |
196 | } | |
197 | ||
58568d2a MX |
198 | /* |
199 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if | |
200 | * any, for the new policy. mpol_new() has already validated the nodes | |
201 | * parameter with respect to the policy mode and flags. But, we need to | |
202 | * handle an empty nodemask with MPOL_PREFERRED here. | |
203 | * | |
204 | * Must be called holding task's alloc_lock to protect task's mems_allowed | |
205 | * and mempolicy. May also be called holding the mmap_semaphore for write. | |
206 | */ | |
4bfc4495 KH |
207 | static int mpol_set_nodemask(struct mempolicy *pol, |
208 | const nodemask_t *nodes, struct nodemask_scratch *nsc) | |
58568d2a | 209 | { |
58568d2a MX |
210 | int ret; |
211 | ||
212 | /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */ | |
213 | if (pol == NULL) | |
214 | return 0; | |
4bfc4495 KH |
215 | /* Check N_HIGH_MEMORY */ |
216 | nodes_and(nsc->mask1, | |
217 | cpuset_current_mems_allowed, node_states[N_HIGH_MEMORY]); | |
58568d2a MX |
218 | |
219 | VM_BUG_ON(!nodes); | |
220 | if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes)) | |
221 | nodes = NULL; /* explicit local allocation */ | |
222 | else { | |
223 | if (pol->flags & MPOL_F_RELATIVE_NODES) | |
4bfc4495 | 224 | mpol_relative_nodemask(&nsc->mask2, nodes,&nsc->mask1); |
58568d2a | 225 | else |
4bfc4495 KH |
226 | nodes_and(nsc->mask2, *nodes, nsc->mask1); |
227 | ||
58568d2a MX |
228 | if (mpol_store_user_nodemask(pol)) |
229 | pol->w.user_nodemask = *nodes; | |
230 | else | |
231 | pol->w.cpuset_mems_allowed = | |
232 | cpuset_current_mems_allowed; | |
233 | } | |
234 | ||
4bfc4495 KH |
235 | if (nodes) |
236 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); | |
237 | else | |
238 | ret = mpol_ops[pol->mode].create(pol, NULL); | |
58568d2a MX |
239 | return ret; |
240 | } | |
241 | ||
242 | /* | |
243 | * This function just creates a new policy, does some check and simple | |
244 | * initialization. You must invoke mpol_set_nodemask() to set nodes. | |
245 | */ | |
028fec41 DR |
246 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
247 | nodemask_t *nodes) | |
1da177e4 LT |
248 | { |
249 | struct mempolicy *policy; | |
250 | ||
028fec41 DR |
251 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
252 | mode, flags, nodes ? nodes_addr(*nodes)[0] : -1); | |
140d5a49 | 253 | |
d3a71033 | 254 | if (mode == MPOL_DEFAULT || mode == MPOL_NOOP) { |
3e1f0645 | 255 | if (nodes && !nodes_empty(*nodes)) |
37012946 | 256 | return ERR_PTR(-EINVAL); |
d3a71033 | 257 | return NULL; |
37012946 | 258 | } |
3e1f0645 DR |
259 | VM_BUG_ON(!nodes); |
260 | ||
261 | /* | |
262 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
263 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
264 | * All other modes require a valid pointer to a non-empty nodemask. | |
265 | */ | |
266 | if (mode == MPOL_PREFERRED) { | |
267 | if (nodes_empty(*nodes)) { | |
268 | if (((flags & MPOL_F_STATIC_NODES) || | |
269 | (flags & MPOL_F_RELATIVE_NODES))) | |
270 | return ERR_PTR(-EINVAL); | |
3e1f0645 | 271 | } |
479e2802 PZ |
272 | } else if (mode == MPOL_LOCAL) { |
273 | if (!nodes_empty(*nodes)) | |
274 | return ERR_PTR(-EINVAL); | |
275 | mode = MPOL_PREFERRED; | |
3e1f0645 DR |
276 | } else if (nodes_empty(*nodes)) |
277 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
278 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
279 | if (!policy) | |
280 | return ERR_PTR(-ENOMEM); | |
281 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 282 | policy->mode = mode; |
3e1f0645 | 283 | policy->flags = flags; |
37012946 | 284 | |
1da177e4 | 285 | return policy; |
37012946 DR |
286 | } |
287 | ||
52cd3b07 LS |
288 | /* Slow path of a mpol destructor. */ |
289 | void __mpol_put(struct mempolicy *p) | |
290 | { | |
291 | if (!atomic_dec_and_test(&p->refcnt)) | |
292 | return; | |
52cd3b07 LS |
293 | kmem_cache_free(policy_cache, p); |
294 | } | |
295 | ||
708c1bbc MX |
296 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes, |
297 | enum mpol_rebind_step step) | |
37012946 DR |
298 | { |
299 | } | |
300 | ||
708c1bbc MX |
301 | /* |
302 | * step: | |
303 | * MPOL_REBIND_ONCE - do rebind work at once | |
304 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
305 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
306 | */ | |
307 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes, | |
308 | enum mpol_rebind_step step) | |
37012946 DR |
309 | { |
310 | nodemask_t tmp; | |
311 | ||
312 | if (pol->flags & MPOL_F_STATIC_NODES) | |
313 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
314 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
315 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
316 | else { | |
708c1bbc MX |
317 | /* |
318 | * if step == 1, we use ->w.cpuset_mems_allowed to cache the | |
319 | * result | |
320 | */ | |
321 | if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) { | |
322 | nodes_remap(tmp, pol->v.nodes, | |
323 | pol->w.cpuset_mems_allowed, *nodes); | |
324 | pol->w.cpuset_mems_allowed = step ? tmp : *nodes; | |
325 | } else if (step == MPOL_REBIND_STEP2) { | |
326 | tmp = pol->w.cpuset_mems_allowed; | |
327 | pol->w.cpuset_mems_allowed = *nodes; | |
328 | } else | |
329 | BUG(); | |
37012946 | 330 | } |
f5b087b5 | 331 | |
708c1bbc MX |
332 | if (nodes_empty(tmp)) |
333 | tmp = *nodes; | |
334 | ||
335 | if (step == MPOL_REBIND_STEP1) | |
336 | nodes_or(pol->v.nodes, pol->v.nodes, tmp); | |
337 | else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2) | |
338 | pol->v.nodes = tmp; | |
339 | else | |
340 | BUG(); | |
341 | ||
37012946 DR |
342 | if (!node_isset(current->il_next, tmp)) { |
343 | current->il_next = next_node(current->il_next, tmp); | |
344 | if (current->il_next >= MAX_NUMNODES) | |
345 | current->il_next = first_node(tmp); | |
346 | if (current->il_next >= MAX_NUMNODES) | |
347 | current->il_next = numa_node_id(); | |
348 | } | |
349 | } | |
350 | ||
351 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
708c1bbc MX |
352 | const nodemask_t *nodes, |
353 | enum mpol_rebind_step step) | |
37012946 DR |
354 | { |
355 | nodemask_t tmp; | |
356 | ||
37012946 DR |
357 | if (pol->flags & MPOL_F_STATIC_NODES) { |
358 | int node = first_node(pol->w.user_nodemask); | |
359 | ||
fc36b8d3 | 360 | if (node_isset(node, *nodes)) { |
37012946 | 361 | pol->v.preferred_node = node; |
fc36b8d3 LS |
362 | pol->flags &= ~MPOL_F_LOCAL; |
363 | } else | |
364 | pol->flags |= MPOL_F_LOCAL; | |
37012946 DR |
365 | } else if (pol->flags & MPOL_F_RELATIVE_NODES) { |
366 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
367 | pol->v.preferred_node = first_node(tmp); | |
fc36b8d3 | 368 | } else if (!(pol->flags & MPOL_F_LOCAL)) { |
37012946 DR |
369 | pol->v.preferred_node = node_remap(pol->v.preferred_node, |
370 | pol->w.cpuset_mems_allowed, | |
371 | *nodes); | |
372 | pol->w.cpuset_mems_allowed = *nodes; | |
373 | } | |
1da177e4 LT |
374 | } |
375 | ||
708c1bbc MX |
376 | /* |
377 | * mpol_rebind_policy - Migrate a policy to a different set of nodes | |
378 | * | |
379 | * If read-side task has no lock to protect task->mempolicy, write-side | |
380 | * task will rebind the task->mempolicy by two step. The first step is | |
381 | * setting all the newly nodes, and the second step is cleaning all the | |
382 | * disallowed nodes. In this way, we can avoid finding no node to alloc | |
383 | * page. | |
384 | * If we have a lock to protect task->mempolicy in read-side, we do | |
385 | * rebind directly. | |
386 | * | |
387 | * step: | |
388 | * MPOL_REBIND_ONCE - do rebind work at once | |
389 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
390 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
391 | */ | |
392 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask, | |
393 | enum mpol_rebind_step step) | |
1d0d2680 | 394 | { |
1d0d2680 DR |
395 | if (!pol) |
396 | return; | |
89c522c7 | 397 | if (!mpol_store_user_nodemask(pol) && step == MPOL_REBIND_ONCE && |
1d0d2680 DR |
398 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
399 | return; | |
708c1bbc MX |
400 | |
401 | if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING)) | |
402 | return; | |
403 | ||
404 | if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING)) | |
405 | BUG(); | |
406 | ||
407 | if (step == MPOL_REBIND_STEP1) | |
408 | pol->flags |= MPOL_F_REBINDING; | |
409 | else if (step == MPOL_REBIND_STEP2) | |
410 | pol->flags &= ~MPOL_F_REBINDING; | |
411 | else if (step >= MPOL_REBIND_NSTEP) | |
412 | BUG(); | |
413 | ||
414 | mpol_ops[pol->mode].rebind(pol, newmask, step); | |
1d0d2680 DR |
415 | } |
416 | ||
417 | /* | |
418 | * Wrapper for mpol_rebind_policy() that just requires task | |
419 | * pointer, and updates task mempolicy. | |
58568d2a MX |
420 | * |
421 | * Called with task's alloc_lock held. | |
1d0d2680 DR |
422 | */ |
423 | ||
708c1bbc MX |
424 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new, |
425 | enum mpol_rebind_step step) | |
1d0d2680 | 426 | { |
708c1bbc | 427 | mpol_rebind_policy(tsk->mempolicy, new, step); |
1d0d2680 DR |
428 | } |
429 | ||
430 | /* | |
431 | * Rebind each vma in mm to new nodemask. | |
432 | * | |
433 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
434 | */ | |
435 | ||
436 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
437 | { | |
438 | struct vm_area_struct *vma; | |
439 | ||
440 | down_write(&mm->mmap_sem); | |
441 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
708c1bbc | 442 | mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE); |
1d0d2680 DR |
443 | up_write(&mm->mmap_sem); |
444 | } | |
445 | ||
37012946 DR |
446 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
447 | [MPOL_DEFAULT] = { | |
448 | .rebind = mpol_rebind_default, | |
449 | }, | |
450 | [MPOL_INTERLEAVE] = { | |
451 | .create = mpol_new_interleave, | |
452 | .rebind = mpol_rebind_nodemask, | |
453 | }, | |
454 | [MPOL_PREFERRED] = { | |
455 | .create = mpol_new_preferred, | |
456 | .rebind = mpol_rebind_preferred, | |
457 | }, | |
458 | [MPOL_BIND] = { | |
459 | .create = mpol_new_bind, | |
460 | .rebind = mpol_rebind_nodemask, | |
461 | }, | |
462 | }; | |
463 | ||
fc301289 CL |
464 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
465 | unsigned long flags); | |
1a75a6c8 | 466 | |
38e35860 | 467 | /* Scan through pages checking if pages follow certain conditions. */ |
b5810039 | 468 | static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
dc9aa5b9 CL |
469 | unsigned long addr, unsigned long end, |
470 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 471 | void *private) |
1da177e4 | 472 | { |
91612e0d HD |
473 | pte_t *orig_pte; |
474 | pte_t *pte; | |
705e87c0 | 475 | spinlock_t *ptl; |
941150a3 | 476 | |
705e87c0 | 477 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
91612e0d | 478 | do { |
6aab341e | 479 | struct page *page; |
25ba77c1 | 480 | int nid; |
91612e0d HD |
481 | |
482 | if (!pte_present(*pte)) | |
1da177e4 | 483 | continue; |
6aab341e LT |
484 | page = vm_normal_page(vma, addr, *pte); |
485 | if (!page) | |
1da177e4 | 486 | continue; |
053837fc | 487 | /* |
62b61f61 HD |
488 | * vm_normal_page() filters out zero pages, but there might |
489 | * still be PageReserved pages to skip, perhaps in a VDSO. | |
490 | * And we cannot move PageKsm pages sensibly or safely yet. | |
053837fc | 491 | */ |
62b61f61 | 492 | if (PageReserved(page) || PageKsm(page)) |
f4598c8b | 493 | continue; |
6aab341e | 494 | nid = page_to_nid(page); |
38e35860 CL |
495 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
496 | continue; | |
497 | ||
b1f72d18 | 498 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
fc301289 | 499 | migrate_page_add(page, private, flags); |
38e35860 CL |
500 | else |
501 | break; | |
91612e0d | 502 | } while (pte++, addr += PAGE_SIZE, addr != end); |
705e87c0 | 503 | pte_unmap_unlock(orig_pte, ptl); |
91612e0d HD |
504 | return addr != end; |
505 | } | |
506 | ||
b5810039 | 507 | static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
dc9aa5b9 CL |
508 | unsigned long addr, unsigned long end, |
509 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 510 | void *private) |
91612e0d HD |
511 | { |
512 | pmd_t *pmd; | |
513 | unsigned long next; | |
514 | ||
515 | pmd = pmd_offset(pud, addr); | |
516 | do { | |
517 | next = pmd_addr_end(addr, end); | |
bae9c19b | 518 | split_huge_page_pmd(vma->vm_mm, pmd); |
1a5a9906 | 519 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
91612e0d | 520 | continue; |
dc9aa5b9 | 521 | if (check_pte_range(vma, pmd, addr, next, nodes, |
38e35860 | 522 | flags, private)) |
91612e0d HD |
523 | return -EIO; |
524 | } while (pmd++, addr = next, addr != end); | |
525 | return 0; | |
526 | } | |
527 | ||
b5810039 | 528 | static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
dc9aa5b9 CL |
529 | unsigned long addr, unsigned long end, |
530 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 531 | void *private) |
91612e0d HD |
532 | { |
533 | pud_t *pud; | |
534 | unsigned long next; | |
535 | ||
536 | pud = pud_offset(pgd, addr); | |
537 | do { | |
538 | next = pud_addr_end(addr, end); | |
539 | if (pud_none_or_clear_bad(pud)) | |
540 | continue; | |
dc9aa5b9 | 541 | if (check_pmd_range(vma, pud, addr, next, nodes, |
38e35860 | 542 | flags, private)) |
91612e0d HD |
543 | return -EIO; |
544 | } while (pud++, addr = next, addr != end); | |
545 | return 0; | |
546 | } | |
547 | ||
b5810039 | 548 | static inline int check_pgd_range(struct vm_area_struct *vma, |
dc9aa5b9 CL |
549 | unsigned long addr, unsigned long end, |
550 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 551 | void *private) |
91612e0d HD |
552 | { |
553 | pgd_t *pgd; | |
554 | unsigned long next; | |
555 | ||
b5810039 | 556 | pgd = pgd_offset(vma->vm_mm, addr); |
91612e0d HD |
557 | do { |
558 | next = pgd_addr_end(addr, end); | |
559 | if (pgd_none_or_clear_bad(pgd)) | |
560 | continue; | |
dc9aa5b9 | 561 | if (check_pud_range(vma, pgd, addr, next, nodes, |
38e35860 | 562 | flags, private)) |
91612e0d HD |
563 | return -EIO; |
564 | } while (pgd++, addr = next, addr != end); | |
565 | return 0; | |
1da177e4 LT |
566 | } |
567 | ||
dc9aa5b9 CL |
568 | /* |
569 | * Check if all pages in a range are on a set of nodes. | |
570 | * If pagelist != NULL then isolate pages from the LRU and | |
571 | * put them on the pagelist. | |
572 | */ | |
1da177e4 LT |
573 | static struct vm_area_struct * |
574 | check_range(struct mm_struct *mm, unsigned long start, unsigned long end, | |
38e35860 | 575 | const nodemask_t *nodes, unsigned long flags, void *private) |
1da177e4 LT |
576 | { |
577 | int err; | |
578 | struct vm_area_struct *first, *vma, *prev; | |
579 | ||
053837fc | 580 | |
1da177e4 LT |
581 | first = find_vma(mm, start); |
582 | if (!first) | |
583 | return ERR_PTR(-EFAULT); | |
584 | prev = NULL; | |
585 | for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) { | |
dc9aa5b9 CL |
586 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
587 | if (!vma->vm_next && vma->vm_end < end) | |
588 | return ERR_PTR(-EFAULT); | |
589 | if (prev && prev->vm_end < vma->vm_start) | |
590 | return ERR_PTR(-EFAULT); | |
591 | } | |
592 | if (!is_vm_hugetlb_page(vma) && | |
593 | ((flags & MPOL_MF_STRICT) || | |
594 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && | |
595 | vma_migratable(vma)))) { | |
5b952b3c | 596 | unsigned long endvma = vma->vm_end; |
dc9aa5b9 | 597 | |
5b952b3c AK |
598 | if (endvma > end) |
599 | endvma = end; | |
600 | if (vma->vm_start > start) | |
601 | start = vma->vm_start; | |
dc9aa5b9 | 602 | err = check_pgd_range(vma, start, endvma, nodes, |
38e35860 | 603 | flags, private); |
1da177e4 LT |
604 | if (err) { |
605 | first = ERR_PTR(err); | |
606 | break; | |
607 | } | |
608 | } | |
609 | prev = vma; | |
610 | } | |
611 | return first; | |
612 | } | |
613 | ||
869833f2 KM |
614 | /* |
615 | * Apply policy to a single VMA | |
616 | * This must be called with the mmap_sem held for writing. | |
617 | */ | |
618 | static int vma_replace_policy(struct vm_area_struct *vma, | |
619 | struct mempolicy *pol) | |
8d34694c | 620 | { |
869833f2 KM |
621 | int err; |
622 | struct mempolicy *old; | |
623 | struct mempolicy *new; | |
8d34694c KM |
624 | |
625 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", | |
626 | vma->vm_start, vma->vm_end, vma->vm_pgoff, | |
627 | vma->vm_ops, vma->vm_file, | |
628 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
629 | ||
869833f2 KM |
630 | new = mpol_dup(pol); |
631 | if (IS_ERR(new)) | |
632 | return PTR_ERR(new); | |
633 | ||
634 | if (vma->vm_ops && vma->vm_ops->set_policy) { | |
8d34694c | 635 | err = vma->vm_ops->set_policy(vma, new); |
869833f2 KM |
636 | if (err) |
637 | goto err_out; | |
8d34694c | 638 | } |
869833f2 KM |
639 | |
640 | old = vma->vm_policy; | |
641 | vma->vm_policy = new; /* protected by mmap_sem */ | |
642 | mpol_put(old); | |
643 | ||
644 | return 0; | |
645 | err_out: | |
646 | mpol_put(new); | |
8d34694c KM |
647 | return err; |
648 | } | |
649 | ||
1da177e4 | 650 | /* Step 2: apply policy to a range and do splits. */ |
9d8cebd4 KM |
651 | static int mbind_range(struct mm_struct *mm, unsigned long start, |
652 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 LT |
653 | { |
654 | struct vm_area_struct *next; | |
9d8cebd4 KM |
655 | struct vm_area_struct *prev; |
656 | struct vm_area_struct *vma; | |
657 | int err = 0; | |
e26a5114 | 658 | pgoff_t pgoff; |
9d8cebd4 KM |
659 | unsigned long vmstart; |
660 | unsigned long vmend; | |
1da177e4 | 661 | |
097d5910 | 662 | vma = find_vma(mm, start); |
9d8cebd4 KM |
663 | if (!vma || vma->vm_start > start) |
664 | return -EFAULT; | |
665 | ||
097d5910 | 666 | prev = vma->vm_prev; |
e26a5114 KM |
667 | if (start > vma->vm_start) |
668 | prev = vma; | |
669 | ||
9d8cebd4 | 670 | for (; vma && vma->vm_start < end; prev = vma, vma = next) { |
1da177e4 | 671 | next = vma->vm_next; |
9d8cebd4 KM |
672 | vmstart = max(start, vma->vm_start); |
673 | vmend = min(end, vma->vm_end); | |
674 | ||
e26a5114 KM |
675 | if (mpol_equal(vma_policy(vma), new_pol)) |
676 | continue; | |
677 | ||
678 | pgoff = vma->vm_pgoff + | |
679 | ((vmstart - vma->vm_start) >> PAGE_SHIFT); | |
9d8cebd4 | 680 | prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags, |
e26a5114 | 681 | vma->anon_vma, vma->vm_file, pgoff, |
8aacc9f5 | 682 | new_pol); |
9d8cebd4 KM |
683 | if (prev) { |
684 | vma = prev; | |
685 | next = vma->vm_next; | |
686 | continue; | |
687 | } | |
688 | if (vma->vm_start != vmstart) { | |
689 | err = split_vma(vma->vm_mm, vma, vmstart, 1); | |
690 | if (err) | |
691 | goto out; | |
692 | } | |
693 | if (vma->vm_end != vmend) { | |
694 | err = split_vma(vma->vm_mm, vma, vmend, 0); | |
695 | if (err) | |
696 | goto out; | |
697 | } | |
869833f2 | 698 | err = vma_replace_policy(vma, new_pol); |
8d34694c KM |
699 | if (err) |
700 | goto out; | |
1da177e4 | 701 | } |
9d8cebd4 KM |
702 | |
703 | out: | |
1da177e4 LT |
704 | return err; |
705 | } | |
706 | ||
c61afb18 PJ |
707 | /* |
708 | * Update task->flags PF_MEMPOLICY bit: set iff non-default | |
709 | * mempolicy. Allows more rapid checking of this (combined perhaps | |
710 | * with other PF_* flag bits) on memory allocation hot code paths. | |
711 | * | |
712 | * If called from outside this file, the task 'p' should -only- be | |
713 | * a newly forked child not yet visible on the task list, because | |
714 | * manipulating the task flags of a visible task is not safe. | |
715 | * | |
716 | * The above limitation is why this routine has the funny name | |
717 | * mpol_fix_fork_child_flag(). | |
718 | * | |
719 | * It is also safe to call this with a task pointer of current, | |
720 | * which the static wrapper mpol_set_task_struct_flag() does, | |
721 | * for use within this file. | |
722 | */ | |
723 | ||
724 | void mpol_fix_fork_child_flag(struct task_struct *p) | |
725 | { | |
726 | if (p->mempolicy) | |
727 | p->flags |= PF_MEMPOLICY; | |
728 | else | |
729 | p->flags &= ~PF_MEMPOLICY; | |
730 | } | |
731 | ||
732 | static void mpol_set_task_struct_flag(void) | |
733 | { | |
734 | mpol_fix_fork_child_flag(current); | |
735 | } | |
736 | ||
1da177e4 | 737 | /* Set the process memory policy */ |
028fec41 DR |
738 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
739 | nodemask_t *nodes) | |
1da177e4 | 740 | { |
58568d2a | 741 | struct mempolicy *new, *old; |
f4e53d91 | 742 | struct mm_struct *mm = current->mm; |
4bfc4495 | 743 | NODEMASK_SCRATCH(scratch); |
58568d2a | 744 | int ret; |
1da177e4 | 745 | |
4bfc4495 KH |
746 | if (!scratch) |
747 | return -ENOMEM; | |
f4e53d91 | 748 | |
4bfc4495 KH |
749 | new = mpol_new(mode, flags, nodes); |
750 | if (IS_ERR(new)) { | |
751 | ret = PTR_ERR(new); | |
752 | goto out; | |
753 | } | |
f4e53d91 LS |
754 | /* |
755 | * prevent changing our mempolicy while show_numa_maps() | |
756 | * is using it. | |
757 | * Note: do_set_mempolicy() can be called at init time | |
758 | * with no 'mm'. | |
759 | */ | |
760 | if (mm) | |
761 | down_write(&mm->mmap_sem); | |
58568d2a | 762 | task_lock(current); |
4bfc4495 | 763 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a MX |
764 | if (ret) { |
765 | task_unlock(current); | |
766 | if (mm) | |
767 | up_write(&mm->mmap_sem); | |
768 | mpol_put(new); | |
4bfc4495 | 769 | goto out; |
58568d2a MX |
770 | } |
771 | old = current->mempolicy; | |
1da177e4 | 772 | current->mempolicy = new; |
c61afb18 | 773 | mpol_set_task_struct_flag(); |
45c4745a | 774 | if (new && new->mode == MPOL_INTERLEAVE && |
f5b087b5 | 775 | nodes_weight(new->v.nodes)) |
dfcd3c0d | 776 | current->il_next = first_node(new->v.nodes); |
58568d2a | 777 | task_unlock(current); |
f4e53d91 LS |
778 | if (mm) |
779 | up_write(&mm->mmap_sem); | |
780 | ||
58568d2a | 781 | mpol_put(old); |
4bfc4495 KH |
782 | ret = 0; |
783 | out: | |
784 | NODEMASK_SCRATCH_FREE(scratch); | |
785 | return ret; | |
1da177e4 LT |
786 | } |
787 | ||
bea904d5 LS |
788 | /* |
789 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
790 | * |
791 | * Called with task's alloc_lock held | |
bea904d5 LS |
792 | */ |
793 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 794 | { |
dfcd3c0d | 795 | nodes_clear(*nodes); |
bea904d5 LS |
796 | if (p == &default_policy) |
797 | return; | |
798 | ||
45c4745a | 799 | switch (p->mode) { |
19770b32 MG |
800 | case MPOL_BIND: |
801 | /* Fall through */ | |
1da177e4 | 802 | case MPOL_INTERLEAVE: |
dfcd3c0d | 803 | *nodes = p->v.nodes; |
1da177e4 LT |
804 | break; |
805 | case MPOL_PREFERRED: | |
fc36b8d3 | 806 | if (!(p->flags & MPOL_F_LOCAL)) |
dfcd3c0d | 807 | node_set(p->v.preferred_node, *nodes); |
53f2556b | 808 | /* else return empty node mask for local allocation */ |
1da177e4 LT |
809 | break; |
810 | default: | |
811 | BUG(); | |
812 | } | |
813 | } | |
814 | ||
815 | static int lookup_node(struct mm_struct *mm, unsigned long addr) | |
816 | { | |
817 | struct page *p; | |
818 | int err; | |
819 | ||
820 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); | |
821 | if (err >= 0) { | |
822 | err = page_to_nid(p); | |
823 | put_page(p); | |
824 | } | |
825 | return err; | |
826 | } | |
827 | ||
1da177e4 | 828 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
829 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
830 | unsigned long addr, unsigned long flags) | |
1da177e4 | 831 | { |
8bccd85f | 832 | int err; |
1da177e4 LT |
833 | struct mm_struct *mm = current->mm; |
834 | struct vm_area_struct *vma = NULL; | |
835 | struct mempolicy *pol = current->mempolicy; | |
836 | ||
754af6f5 LS |
837 | if (flags & |
838 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 839 | return -EINVAL; |
754af6f5 LS |
840 | |
841 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
842 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
843 | return -EINVAL; | |
844 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 845 | task_lock(current); |
754af6f5 | 846 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 847 | task_unlock(current); |
754af6f5 LS |
848 | return 0; |
849 | } | |
850 | ||
1da177e4 | 851 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
852 | /* |
853 | * Do NOT fall back to task policy if the | |
854 | * vma/shared policy at addr is NULL. We | |
855 | * want to return MPOL_DEFAULT in this case. | |
856 | */ | |
1da177e4 LT |
857 | down_read(&mm->mmap_sem); |
858 | vma = find_vma_intersection(mm, addr, addr+1); | |
859 | if (!vma) { | |
860 | up_read(&mm->mmap_sem); | |
861 | return -EFAULT; | |
862 | } | |
863 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
864 | pol = vma->vm_ops->get_policy(vma, addr); | |
865 | else | |
866 | pol = vma->vm_policy; | |
867 | } else if (addr) | |
868 | return -EINVAL; | |
869 | ||
870 | if (!pol) | |
bea904d5 | 871 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
872 | |
873 | if (flags & MPOL_F_NODE) { | |
874 | if (flags & MPOL_F_ADDR) { | |
875 | err = lookup_node(mm, addr); | |
876 | if (err < 0) | |
877 | goto out; | |
8bccd85f | 878 | *policy = err; |
1da177e4 | 879 | } else if (pol == current->mempolicy && |
45c4745a | 880 | pol->mode == MPOL_INTERLEAVE) { |
8bccd85f | 881 | *policy = current->il_next; |
1da177e4 LT |
882 | } else { |
883 | err = -EINVAL; | |
884 | goto out; | |
885 | } | |
bea904d5 LS |
886 | } else { |
887 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
888 | pol->mode; | |
d79df630 DR |
889 | /* |
890 | * Internal mempolicy flags must be masked off before exposing | |
891 | * the policy to userspace. | |
892 | */ | |
893 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 894 | } |
1da177e4 LT |
895 | |
896 | if (vma) { | |
897 | up_read(¤t->mm->mmap_sem); | |
898 | vma = NULL; | |
899 | } | |
900 | ||
1da177e4 | 901 | err = 0; |
58568d2a | 902 | if (nmask) { |
c6b6ef8b LS |
903 | if (mpol_store_user_nodemask(pol)) { |
904 | *nmask = pol->w.user_nodemask; | |
905 | } else { | |
906 | task_lock(current); | |
907 | get_policy_nodemask(pol, nmask); | |
908 | task_unlock(current); | |
909 | } | |
58568d2a | 910 | } |
1da177e4 LT |
911 | |
912 | out: | |
52cd3b07 | 913 | mpol_cond_put(pol); |
1da177e4 LT |
914 | if (vma) |
915 | up_read(¤t->mm->mmap_sem); | |
916 | return err; | |
917 | } | |
918 | ||
b20a3503 | 919 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 CL |
920 | /* |
921 | * page migration | |
922 | */ | |
fc301289 CL |
923 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
924 | unsigned long flags) | |
6ce3c4c0 CL |
925 | { |
926 | /* | |
fc301289 | 927 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 928 | */ |
62695a84 NP |
929 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) { |
930 | if (!isolate_lru_page(page)) { | |
931 | list_add_tail(&page->lru, pagelist); | |
6d9c285a KM |
932 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
933 | page_is_file_cache(page)); | |
62695a84 NP |
934 | } |
935 | } | |
7e2ab150 | 936 | } |
6ce3c4c0 | 937 | |
742755a1 | 938 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
95a402c3 | 939 | { |
6484eb3e | 940 | return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0); |
95a402c3 CL |
941 | } |
942 | ||
7e2ab150 CL |
943 | /* |
944 | * Migrate pages from one node to a target node. | |
945 | * Returns error or the number of pages not migrated. | |
946 | */ | |
dbcb0f19 AB |
947 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
948 | int flags) | |
7e2ab150 CL |
949 | { |
950 | nodemask_t nmask; | |
951 | LIST_HEAD(pagelist); | |
952 | int err = 0; | |
953 | ||
954 | nodes_clear(nmask); | |
955 | node_set(source, nmask); | |
6ce3c4c0 | 956 | |
08270807 MK |
957 | /* |
958 | * This does not "check" the range but isolates all pages that | |
959 | * need migration. Between passing in the full user address | |
960 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. | |
961 | */ | |
962 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); | |
963 | check_range(mm, mm->mmap->vm_start, mm->task_size, &nmask, | |
7e2ab150 CL |
964 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
965 | ||
cf608ac1 | 966 | if (!list_empty(&pagelist)) { |
7f0f2496 | 967 | err = migrate_pages(&pagelist, new_node_page, dest, |
7b2a2d4a MG |
968 | false, MIGRATE_SYNC, |
969 | MR_SYSCALL); | |
cf608ac1 MK |
970 | if (err) |
971 | putback_lru_pages(&pagelist); | |
972 | } | |
95a402c3 | 973 | |
7e2ab150 | 974 | return err; |
6ce3c4c0 CL |
975 | } |
976 | ||
39743889 | 977 | /* |
7e2ab150 CL |
978 | * Move pages between the two nodesets so as to preserve the physical |
979 | * layout as much as possible. | |
39743889 CL |
980 | * |
981 | * Returns the number of page that could not be moved. | |
982 | */ | |
0ce72d4f AM |
983 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
984 | const nodemask_t *to, int flags) | |
39743889 | 985 | { |
7e2ab150 | 986 | int busy = 0; |
0aedadf9 | 987 | int err; |
7e2ab150 | 988 | nodemask_t tmp; |
39743889 | 989 | |
0aedadf9 CL |
990 | err = migrate_prep(); |
991 | if (err) | |
992 | return err; | |
993 | ||
53f2556b | 994 | down_read(&mm->mmap_sem); |
39743889 | 995 | |
0ce72d4f | 996 | err = migrate_vmas(mm, from, to, flags); |
7b2259b3 CL |
997 | if (err) |
998 | goto out; | |
999 | ||
da0aa138 KM |
1000 | /* |
1001 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1002 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1003 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1004 | * The pair of nodemasks 'to' and 'from' define the map. | |
1005 | * | |
1006 | * If no pair of bits is found that way, fallback to picking some | |
1007 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1008 | * 'source' and 'dest' bits are the same, this represents a node | |
1009 | * that will be migrating to itself, so no pages need move. | |
1010 | * | |
1011 | * If no bits are left in 'tmp', or if all remaining bits left | |
1012 | * in 'tmp' correspond to the same bit in 'to', return false | |
1013 | * (nothing left to migrate). | |
1014 | * | |
1015 | * This lets us pick a pair of nodes to migrate between, such that | |
1016 | * if possible the dest node is not already occupied by some other | |
1017 | * source node, minimizing the risk of overloading the memory on a | |
1018 | * node that would happen if we migrated incoming memory to a node | |
1019 | * before migrating outgoing memory source that same node. | |
1020 | * | |
1021 | * A single scan of tmp is sufficient. As we go, we remember the | |
1022 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1023 | * that not only moved, but what's better, moved to an empty slot | |
1024 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1025 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1026 | * most recent <s, d> pair that moved. If we get all the way through |
1027 | * the scan of tmp without finding any node that moved, much less | |
1028 | * moved to an empty node, then there is nothing left worth migrating. | |
1029 | */ | |
d4984711 | 1030 | |
0ce72d4f | 1031 | tmp = *from; |
7e2ab150 CL |
1032 | while (!nodes_empty(tmp)) { |
1033 | int s,d; | |
1034 | int source = -1; | |
1035 | int dest = 0; | |
1036 | ||
1037 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1038 | |
1039 | /* | |
1040 | * do_migrate_pages() tries to maintain the relative | |
1041 | * node relationship of the pages established between | |
1042 | * threads and memory areas. | |
1043 | * | |
1044 | * However if the number of source nodes is not equal to | |
1045 | * the number of destination nodes we can not preserve | |
1046 | * this node relative relationship. In that case, skip | |
1047 | * copying memory from a node that is in the destination | |
1048 | * mask. | |
1049 | * | |
1050 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1051 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1052 | */ | |
1053 | ||
0ce72d4f AM |
1054 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1055 | (node_isset(s, *to))) | |
4a5b18cc LW |
1056 | continue; |
1057 | ||
0ce72d4f | 1058 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1059 | if (s == d) |
1060 | continue; | |
1061 | ||
1062 | source = s; /* Node moved. Memorize */ | |
1063 | dest = d; | |
1064 | ||
1065 | /* dest not in remaining from nodes? */ | |
1066 | if (!node_isset(dest, tmp)) | |
1067 | break; | |
1068 | } | |
1069 | if (source == -1) | |
1070 | break; | |
1071 | ||
1072 | node_clear(source, tmp); | |
1073 | err = migrate_to_node(mm, source, dest, flags); | |
1074 | if (err > 0) | |
1075 | busy += err; | |
1076 | if (err < 0) | |
1077 | break; | |
39743889 | 1078 | } |
7b2259b3 | 1079 | out: |
39743889 | 1080 | up_read(&mm->mmap_sem); |
7e2ab150 CL |
1081 | if (err < 0) |
1082 | return err; | |
1083 | return busy; | |
b20a3503 CL |
1084 | |
1085 | } | |
1086 | ||
3ad33b24 LS |
1087 | /* |
1088 | * Allocate a new page for page migration based on vma policy. | |
1089 | * Start assuming that page is mapped by vma pointed to by @private. | |
1090 | * Search forward from there, if not. N.B., this assumes that the | |
1091 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1092 | * is in virtual address order. | |
1093 | */ | |
742755a1 | 1094 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
1095 | { |
1096 | struct vm_area_struct *vma = (struct vm_area_struct *)private; | |
3ad33b24 | 1097 | unsigned long uninitialized_var(address); |
95a402c3 | 1098 | |
3ad33b24 LS |
1099 | while (vma) { |
1100 | address = page_address_in_vma(page, vma); | |
1101 | if (address != -EFAULT) | |
1102 | break; | |
1103 | vma = vma->vm_next; | |
1104 | } | |
1105 | ||
1106 | /* | |
1107 | * if !vma, alloc_page_vma() will use task or system default policy | |
1108 | */ | |
1109 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
95a402c3 | 1110 | } |
b20a3503 CL |
1111 | #else |
1112 | ||
1113 | static void migrate_page_add(struct page *page, struct list_head *pagelist, | |
1114 | unsigned long flags) | |
1115 | { | |
39743889 CL |
1116 | } |
1117 | ||
0ce72d4f AM |
1118 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1119 | const nodemask_t *to, int flags) | |
b20a3503 CL |
1120 | { |
1121 | return -ENOSYS; | |
1122 | } | |
95a402c3 | 1123 | |
69939749 | 1124 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
1125 | { |
1126 | return NULL; | |
1127 | } | |
b20a3503 CL |
1128 | #endif |
1129 | ||
dbcb0f19 | 1130 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1131 | unsigned short mode, unsigned short mode_flags, |
1132 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 CL |
1133 | { |
1134 | struct vm_area_struct *vma; | |
1135 | struct mm_struct *mm = current->mm; | |
1136 | struct mempolicy *new; | |
1137 | unsigned long end; | |
1138 | int err; | |
1139 | LIST_HEAD(pagelist); | |
1140 | ||
a3b51e01 DR |
1141 | if (flags & ~(unsigned long)(MPOL_MF_STRICT | |
1142 | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) | |
6ce3c4c0 | 1143 | return -EINVAL; |
74c00241 | 1144 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1145 | return -EPERM; |
1146 | ||
1147 | if (start & ~PAGE_MASK) | |
1148 | return -EINVAL; | |
1149 | ||
d3a71033 | 1150 | if (mode == MPOL_DEFAULT || mode == MPOL_NOOP) |
6ce3c4c0 CL |
1151 | flags &= ~MPOL_MF_STRICT; |
1152 | ||
1153 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
1154 | end = start + len; | |
1155 | ||
1156 | if (end < start) | |
1157 | return -EINVAL; | |
1158 | if (end == start) | |
1159 | return 0; | |
1160 | ||
028fec41 | 1161 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1162 | if (IS_ERR(new)) |
1163 | return PTR_ERR(new); | |
1164 | ||
1165 | /* | |
1166 | * If we are using the default policy then operation | |
1167 | * on discontinuous address spaces is okay after all | |
1168 | */ | |
1169 | if (!new) | |
1170 | flags |= MPOL_MF_DISCONTIG_OK; | |
1171 | ||
028fec41 DR |
1172 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1173 | start, start + len, mode, mode_flags, | |
1174 | nmask ? nodes_addr(*nmask)[0] : -1); | |
6ce3c4c0 | 1175 | |
0aedadf9 CL |
1176 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1177 | ||
1178 | err = migrate_prep(); | |
1179 | if (err) | |
b05ca738 | 1180 | goto mpol_out; |
0aedadf9 | 1181 | } |
4bfc4495 KH |
1182 | { |
1183 | NODEMASK_SCRATCH(scratch); | |
1184 | if (scratch) { | |
1185 | down_write(&mm->mmap_sem); | |
1186 | task_lock(current); | |
1187 | err = mpol_set_nodemask(new, nmask, scratch); | |
1188 | task_unlock(current); | |
1189 | if (err) | |
1190 | up_write(&mm->mmap_sem); | |
1191 | } else | |
1192 | err = -ENOMEM; | |
1193 | NODEMASK_SCRATCH_FREE(scratch); | |
1194 | } | |
b05ca738 KM |
1195 | if (err) |
1196 | goto mpol_out; | |
1197 | ||
6ce3c4c0 CL |
1198 | vma = check_range(mm, start, end, nmask, |
1199 | flags | MPOL_MF_INVERT, &pagelist); | |
1200 | ||
1201 | err = PTR_ERR(vma); | |
1202 | if (!IS_ERR(vma)) { | |
1203 | int nr_failed = 0; | |
1204 | ||
9d8cebd4 | 1205 | err = mbind_range(mm, start, end, new); |
7e2ab150 | 1206 | |
cf608ac1 | 1207 | if (!list_empty(&pagelist)) { |
95a402c3 | 1208 | nr_failed = migrate_pages(&pagelist, new_vma_page, |
7f0f2496 | 1209 | (unsigned long)vma, |
7b2a2d4a MG |
1210 | false, MIGRATE_SYNC, |
1211 | MR_MEMPOLICY_MBIND); | |
cf608ac1 MK |
1212 | if (nr_failed) |
1213 | putback_lru_pages(&pagelist); | |
1214 | } | |
6ce3c4c0 CL |
1215 | |
1216 | if (!err && nr_failed && (flags & MPOL_MF_STRICT)) | |
1217 | err = -EIO; | |
ab8a3e14 KM |
1218 | } else |
1219 | putback_lru_pages(&pagelist); | |
b20a3503 | 1220 | |
6ce3c4c0 | 1221 | up_write(&mm->mmap_sem); |
b05ca738 | 1222 | mpol_out: |
f0be3d32 | 1223 | mpol_put(new); |
6ce3c4c0 CL |
1224 | return err; |
1225 | } | |
1226 | ||
8bccd85f CL |
1227 | /* |
1228 | * User space interface with variable sized bitmaps for nodelists. | |
1229 | */ | |
1230 | ||
1231 | /* Copy a node mask from user space. */ | |
39743889 | 1232 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1233 | unsigned long maxnode) |
1234 | { | |
1235 | unsigned long k; | |
1236 | unsigned long nlongs; | |
1237 | unsigned long endmask; | |
1238 | ||
1239 | --maxnode; | |
1240 | nodes_clear(*nodes); | |
1241 | if (maxnode == 0 || !nmask) | |
1242 | return 0; | |
a9c930ba | 1243 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1244 | return -EINVAL; |
8bccd85f CL |
1245 | |
1246 | nlongs = BITS_TO_LONGS(maxnode); | |
1247 | if ((maxnode % BITS_PER_LONG) == 0) | |
1248 | endmask = ~0UL; | |
1249 | else | |
1250 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1251 | ||
1252 | /* When the user specified more nodes than supported just check | |
1253 | if the non supported part is all zero. */ | |
1254 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
1255 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
1256 | return -EINVAL; | |
1257 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
1258 | unsigned long t; | |
1259 | if (get_user(t, nmask + k)) | |
1260 | return -EFAULT; | |
1261 | if (k == nlongs - 1) { | |
1262 | if (t & endmask) | |
1263 | return -EINVAL; | |
1264 | } else if (t) | |
1265 | return -EINVAL; | |
1266 | } | |
1267 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1268 | endmask = ~0UL; | |
1269 | } | |
1270 | ||
1271 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
1272 | return -EFAULT; | |
1273 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1274 | return 0; | |
1275 | } | |
1276 | ||
1277 | /* Copy a kernel node mask to user space */ | |
1278 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1279 | nodemask_t *nodes) | |
1280 | { | |
1281 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
1282 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
1283 | ||
1284 | if (copy > nbytes) { | |
1285 | if (copy > PAGE_SIZE) | |
1286 | return -EINVAL; | |
1287 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1288 | return -EFAULT; | |
1289 | copy = nbytes; | |
1290 | } | |
1291 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1292 | } | |
1293 | ||
938bb9f5 HC |
1294 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
1295 | unsigned long, mode, unsigned long __user *, nmask, | |
1296 | unsigned long, maxnode, unsigned, flags) | |
8bccd85f CL |
1297 | { |
1298 | nodemask_t nodes; | |
1299 | int err; | |
028fec41 | 1300 | unsigned short mode_flags; |
8bccd85f | 1301 | |
028fec41 DR |
1302 | mode_flags = mode & MPOL_MODE_FLAGS; |
1303 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
1304 | if (mode >= MPOL_MAX) |
1305 | return -EINVAL; | |
4c50bc01 DR |
1306 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
1307 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
1308 | return -EINVAL; | |
8bccd85f CL |
1309 | err = get_nodes(&nodes, nmask, maxnode); |
1310 | if (err) | |
1311 | return err; | |
028fec41 | 1312 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
1313 | } |
1314 | ||
1315 | /* Set the process memory policy */ | |
938bb9f5 HC |
1316 | SYSCALL_DEFINE3(set_mempolicy, int, mode, unsigned long __user *, nmask, |
1317 | unsigned long, maxnode) | |
8bccd85f CL |
1318 | { |
1319 | int err; | |
1320 | nodemask_t nodes; | |
028fec41 | 1321 | unsigned short flags; |
8bccd85f | 1322 | |
028fec41 DR |
1323 | flags = mode & MPOL_MODE_FLAGS; |
1324 | mode &= ~MPOL_MODE_FLAGS; | |
1325 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 1326 | return -EINVAL; |
4c50bc01 DR |
1327 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
1328 | return -EINVAL; | |
8bccd85f CL |
1329 | err = get_nodes(&nodes, nmask, maxnode); |
1330 | if (err) | |
1331 | return err; | |
028fec41 | 1332 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1333 | } |
1334 | ||
938bb9f5 HC |
1335 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1336 | const unsigned long __user *, old_nodes, | |
1337 | const unsigned long __user *, new_nodes) | |
39743889 | 1338 | { |
c69e8d9c | 1339 | const struct cred *cred = current_cred(), *tcred; |
596d7cfa | 1340 | struct mm_struct *mm = NULL; |
39743889 | 1341 | struct task_struct *task; |
39743889 CL |
1342 | nodemask_t task_nodes; |
1343 | int err; | |
596d7cfa KM |
1344 | nodemask_t *old; |
1345 | nodemask_t *new; | |
1346 | NODEMASK_SCRATCH(scratch); | |
1347 | ||
1348 | if (!scratch) | |
1349 | return -ENOMEM; | |
39743889 | 1350 | |
596d7cfa KM |
1351 | old = &scratch->mask1; |
1352 | new = &scratch->mask2; | |
1353 | ||
1354 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1355 | if (err) |
596d7cfa | 1356 | goto out; |
39743889 | 1357 | |
596d7cfa | 1358 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1359 | if (err) |
596d7cfa | 1360 | goto out; |
39743889 CL |
1361 | |
1362 | /* Find the mm_struct */ | |
55cfaa3c | 1363 | rcu_read_lock(); |
228ebcbe | 1364 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1365 | if (!task) { |
55cfaa3c | 1366 | rcu_read_unlock(); |
596d7cfa KM |
1367 | err = -ESRCH; |
1368 | goto out; | |
39743889 | 1369 | } |
3268c63e | 1370 | get_task_struct(task); |
39743889 | 1371 | |
596d7cfa | 1372 | err = -EINVAL; |
39743889 CL |
1373 | |
1374 | /* | |
1375 | * Check if this process has the right to modify the specified | |
1376 | * process. The right exists if the process has administrative | |
7f927fcc | 1377 | * capabilities, superuser privileges or the same |
39743889 CL |
1378 | * userid as the target process. |
1379 | */ | |
c69e8d9c | 1380 | tcred = __task_cred(task); |
b38a86eb EB |
1381 | if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) && |
1382 | !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) && | |
74c00241 | 1383 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1384 | rcu_read_unlock(); |
39743889 | 1385 | err = -EPERM; |
3268c63e | 1386 | goto out_put; |
39743889 | 1387 | } |
c69e8d9c | 1388 | rcu_read_unlock(); |
39743889 CL |
1389 | |
1390 | task_nodes = cpuset_mems_allowed(task); | |
1391 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1392 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1393 | err = -EPERM; |
3268c63e | 1394 | goto out_put; |
39743889 CL |
1395 | } |
1396 | ||
596d7cfa | 1397 | if (!nodes_subset(*new, node_states[N_HIGH_MEMORY])) { |
3b42d28b | 1398 | err = -EINVAL; |
3268c63e | 1399 | goto out_put; |
3b42d28b CL |
1400 | } |
1401 | ||
86c3a764 DQ |
1402 | err = security_task_movememory(task); |
1403 | if (err) | |
3268c63e | 1404 | goto out_put; |
86c3a764 | 1405 | |
3268c63e CL |
1406 | mm = get_task_mm(task); |
1407 | put_task_struct(task); | |
f2a9ef88 SL |
1408 | |
1409 | if (!mm) { | |
3268c63e | 1410 | err = -EINVAL; |
f2a9ef88 SL |
1411 | goto out; |
1412 | } | |
1413 | ||
1414 | err = do_migrate_pages(mm, old, new, | |
1415 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1416 | |
1417 | mmput(mm); | |
1418 | out: | |
596d7cfa KM |
1419 | NODEMASK_SCRATCH_FREE(scratch); |
1420 | ||
39743889 | 1421 | return err; |
3268c63e CL |
1422 | |
1423 | out_put: | |
1424 | put_task_struct(task); | |
1425 | goto out; | |
1426 | ||
39743889 CL |
1427 | } |
1428 | ||
1429 | ||
8bccd85f | 1430 | /* Retrieve NUMA policy */ |
938bb9f5 HC |
1431 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1432 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1433 | unsigned long, addr, unsigned long, flags) | |
8bccd85f | 1434 | { |
dbcb0f19 AB |
1435 | int err; |
1436 | int uninitialized_var(pval); | |
8bccd85f CL |
1437 | nodemask_t nodes; |
1438 | ||
1439 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
1440 | return -EINVAL; | |
1441 | ||
1442 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1443 | ||
1444 | if (err) | |
1445 | return err; | |
1446 | ||
1447 | if (policy && put_user(pval, policy)) | |
1448 | return -EFAULT; | |
1449 | ||
1450 | if (nmask) | |
1451 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1452 | ||
1453 | return err; | |
1454 | } | |
1455 | ||
1da177e4 LT |
1456 | #ifdef CONFIG_COMPAT |
1457 | ||
1458 | asmlinkage long compat_sys_get_mempolicy(int __user *policy, | |
1459 | compat_ulong_t __user *nmask, | |
1460 | compat_ulong_t maxnode, | |
1461 | compat_ulong_t addr, compat_ulong_t flags) | |
1462 | { | |
1463 | long err; | |
1464 | unsigned long __user *nm = NULL; | |
1465 | unsigned long nr_bits, alloc_size; | |
1466 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1467 | ||
1468 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1469 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1470 | ||
1471 | if (nmask) | |
1472 | nm = compat_alloc_user_space(alloc_size); | |
1473 | ||
1474 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
1475 | ||
1476 | if (!err && nmask) { | |
2bbff6c7 KH |
1477 | unsigned long copy_size; |
1478 | copy_size = min_t(unsigned long, sizeof(bm), alloc_size); | |
1479 | err = copy_from_user(bm, nm, copy_size); | |
1da177e4 LT |
1480 | /* ensure entire bitmap is zeroed */ |
1481 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1482 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1483 | } | |
1484 | ||
1485 | return err; | |
1486 | } | |
1487 | ||
1488 | asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask, | |
1489 | compat_ulong_t maxnode) | |
1490 | { | |
1491 | long err = 0; | |
1492 | unsigned long __user *nm = NULL; | |
1493 | unsigned long nr_bits, alloc_size; | |
1494 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1495 | ||
1496 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1497 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1498 | ||
1499 | if (nmask) { | |
1500 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
1501 | nm = compat_alloc_user_space(alloc_size); | |
1502 | err |= copy_to_user(nm, bm, alloc_size); | |
1503 | } | |
1504 | ||
1505 | if (err) | |
1506 | return -EFAULT; | |
1507 | ||
1508 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
1509 | } | |
1510 | ||
1511 | asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len, | |
1512 | compat_ulong_t mode, compat_ulong_t __user *nmask, | |
1513 | compat_ulong_t maxnode, compat_ulong_t flags) | |
1514 | { | |
1515 | long err = 0; | |
1516 | unsigned long __user *nm = NULL; | |
1517 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1518 | nodemask_t bm; |
1da177e4 LT |
1519 | |
1520 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1521 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1522 | ||
1523 | if (nmask) { | |
dfcd3c0d | 1524 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 1525 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 1526 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
1527 | } |
1528 | ||
1529 | if (err) | |
1530 | return -EFAULT; | |
1531 | ||
1532 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
1533 | } | |
1534 | ||
1535 | #endif | |
1536 | ||
480eccf9 LS |
1537 | /* |
1538 | * get_vma_policy(@task, @vma, @addr) | |
1539 | * @task - task for fallback if vma policy == default | |
1540 | * @vma - virtual memory area whose policy is sought | |
1541 | * @addr - address in @vma for shared policy lookup | |
1542 | * | |
1543 | * Returns effective policy for a VMA at specified address. | |
1544 | * Falls back to @task or system default policy, as necessary. | |
32f8516a DR |
1545 | * Current or other task's task mempolicy and non-shared vma policies must be |
1546 | * protected by task_lock(task) by the caller. | |
52cd3b07 LS |
1547 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1548 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1549 | * freeing by another task. It is the caller's responsibility to free the | |
1550 | * extra reference for shared policies. | |
480eccf9 | 1551 | */ |
d98f6cb6 | 1552 | struct mempolicy *get_vma_policy(struct task_struct *task, |
48fce342 | 1553 | struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1554 | { |
6e21c8f1 | 1555 | struct mempolicy *pol = task->mempolicy; |
1da177e4 LT |
1556 | |
1557 | if (vma) { | |
480eccf9 | 1558 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
ae4d8c16 LS |
1559 | struct mempolicy *vpol = vma->vm_ops->get_policy(vma, |
1560 | addr); | |
1561 | if (vpol) | |
1562 | pol = vpol; | |
00442ad0 | 1563 | } else if (vma->vm_policy) { |
1da177e4 | 1564 | pol = vma->vm_policy; |
00442ad0 MG |
1565 | |
1566 | /* | |
1567 | * shmem_alloc_page() passes MPOL_F_SHARED policy with | |
1568 | * a pseudo vma whose vma->vm_ops=NULL. Take a reference | |
1569 | * count on these policies which will be dropped by | |
1570 | * mpol_cond_put() later | |
1571 | */ | |
1572 | if (mpol_needs_cond_ref(pol)) | |
1573 | mpol_get(pol); | |
1574 | } | |
1da177e4 LT |
1575 | } |
1576 | if (!pol) | |
1577 | pol = &default_policy; | |
1578 | return pol; | |
1579 | } | |
1580 | ||
52cd3b07 LS |
1581 | /* |
1582 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1583 | * page allocation | |
1584 | */ | |
1585 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) | |
19770b32 MG |
1586 | { |
1587 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1588 | if (unlikely(policy->mode == MPOL_BIND) && |
19770b32 MG |
1589 | gfp_zone(gfp) >= policy_zone && |
1590 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) | |
1591 | return &policy->v.nodes; | |
1592 | ||
1593 | return NULL; | |
1594 | } | |
1595 | ||
52cd3b07 | 1596 | /* Return a zonelist indicated by gfp for node representing a mempolicy */ |
2f5f9486 AK |
1597 | static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy, |
1598 | int nd) | |
1da177e4 | 1599 | { |
45c4745a | 1600 | switch (policy->mode) { |
1da177e4 | 1601 | case MPOL_PREFERRED: |
fc36b8d3 LS |
1602 | if (!(policy->flags & MPOL_F_LOCAL)) |
1603 | nd = policy->v.preferred_node; | |
1da177e4 LT |
1604 | break; |
1605 | case MPOL_BIND: | |
19770b32 | 1606 | /* |
52cd3b07 LS |
1607 | * Normally, MPOL_BIND allocations are node-local within the |
1608 | * allowed nodemask. However, if __GFP_THISNODE is set and the | |
6eb27e1f | 1609 | * current node isn't part of the mask, we use the zonelist for |
52cd3b07 | 1610 | * the first node in the mask instead. |
19770b32 | 1611 | */ |
19770b32 MG |
1612 | if (unlikely(gfp & __GFP_THISNODE) && |
1613 | unlikely(!node_isset(nd, policy->v.nodes))) | |
1614 | nd = first_node(policy->v.nodes); | |
1615 | break; | |
1da177e4 | 1616 | default: |
1da177e4 LT |
1617 | BUG(); |
1618 | } | |
0e88460d | 1619 | return node_zonelist(nd, gfp); |
1da177e4 LT |
1620 | } |
1621 | ||
1622 | /* Do dynamic interleaving for a process */ | |
1623 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1624 | { | |
1625 | unsigned nid, next; | |
1626 | struct task_struct *me = current; | |
1627 | ||
1628 | nid = me->il_next; | |
dfcd3c0d | 1629 | next = next_node(nid, policy->v.nodes); |
1da177e4 | 1630 | if (next >= MAX_NUMNODES) |
dfcd3c0d | 1631 | next = first_node(policy->v.nodes); |
f5b087b5 DR |
1632 | if (next < MAX_NUMNODES) |
1633 | me->il_next = next; | |
1da177e4 LT |
1634 | return nid; |
1635 | } | |
1636 | ||
dc85da15 CL |
1637 | /* |
1638 | * Depending on the memory policy provide a node from which to allocate the | |
1639 | * next slab entry. | |
52cd3b07 LS |
1640 | * @policy must be protected by freeing by the caller. If @policy is |
1641 | * the current task's mempolicy, this protection is implicit, as only the | |
1642 | * task can change it's policy. The system default policy requires no | |
1643 | * such protection. | |
dc85da15 | 1644 | */ |
e7b691b0 | 1645 | unsigned slab_node(void) |
dc85da15 | 1646 | { |
e7b691b0 AK |
1647 | struct mempolicy *policy; |
1648 | ||
1649 | if (in_interrupt()) | |
1650 | return numa_node_id(); | |
1651 | ||
1652 | policy = current->mempolicy; | |
fc36b8d3 | 1653 | if (!policy || policy->flags & MPOL_F_LOCAL) |
bea904d5 LS |
1654 | return numa_node_id(); |
1655 | ||
1656 | switch (policy->mode) { | |
1657 | case MPOL_PREFERRED: | |
fc36b8d3 LS |
1658 | /* |
1659 | * handled MPOL_F_LOCAL above | |
1660 | */ | |
1661 | return policy->v.preferred_node; | |
765c4507 | 1662 | |
dc85da15 CL |
1663 | case MPOL_INTERLEAVE: |
1664 | return interleave_nodes(policy); | |
1665 | ||
dd1a239f | 1666 | case MPOL_BIND: { |
dc85da15 CL |
1667 | /* |
1668 | * Follow bind policy behavior and start allocation at the | |
1669 | * first node. | |
1670 | */ | |
19770b32 MG |
1671 | struct zonelist *zonelist; |
1672 | struct zone *zone; | |
1673 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); | |
1674 | zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0]; | |
1675 | (void)first_zones_zonelist(zonelist, highest_zoneidx, | |
1676 | &policy->v.nodes, | |
1677 | &zone); | |
800416f7 | 1678 | return zone ? zone->node : numa_node_id(); |
dd1a239f | 1679 | } |
dc85da15 | 1680 | |
dc85da15 | 1681 | default: |
bea904d5 | 1682 | BUG(); |
dc85da15 CL |
1683 | } |
1684 | } | |
1685 | ||
1da177e4 LT |
1686 | /* Do static interleaving for a VMA with known offset. */ |
1687 | static unsigned offset_il_node(struct mempolicy *pol, | |
1688 | struct vm_area_struct *vma, unsigned long off) | |
1689 | { | |
dfcd3c0d | 1690 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1691 | unsigned target; |
1da177e4 LT |
1692 | int c; |
1693 | int nid = -1; | |
1694 | ||
f5b087b5 DR |
1695 | if (!nnodes) |
1696 | return numa_node_id(); | |
1697 | target = (unsigned int)off % nnodes; | |
1da177e4 LT |
1698 | c = 0; |
1699 | do { | |
dfcd3c0d | 1700 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1701 | c++; |
1702 | } while (c <= target); | |
1da177e4 LT |
1703 | return nid; |
1704 | } | |
1705 | ||
5da7ca86 CL |
1706 | /* Determine a node number for interleave */ |
1707 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1708 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1709 | { | |
1710 | if (vma) { | |
1711 | unsigned long off; | |
1712 | ||
3b98b087 NA |
1713 | /* |
1714 | * for small pages, there is no difference between | |
1715 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1716 | * for huge pages, since vm_pgoff is in units of small | |
1717 | * pages, we need to shift off the always 0 bits to get | |
1718 | * a useful offset. | |
1719 | */ | |
1720 | BUG_ON(shift < PAGE_SHIFT); | |
1721 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 CL |
1722 | off += (addr - vma->vm_start) >> shift; |
1723 | return offset_il_node(pol, vma, off); | |
1724 | } else | |
1725 | return interleave_nodes(pol); | |
1726 | } | |
1727 | ||
778d3b0f MH |
1728 | /* |
1729 | * Return the bit number of a random bit set in the nodemask. | |
1730 | * (returns -1 if nodemask is empty) | |
1731 | */ | |
1732 | int node_random(const nodemask_t *maskp) | |
1733 | { | |
1734 | int w, bit = -1; | |
1735 | ||
1736 | w = nodes_weight(*maskp); | |
1737 | if (w) | |
1738 | bit = bitmap_ord_to_pos(maskp->bits, | |
1739 | get_random_int() % w, MAX_NUMNODES); | |
1740 | return bit; | |
1741 | } | |
1742 | ||
00ac59ad | 1743 | #ifdef CONFIG_HUGETLBFS |
480eccf9 LS |
1744 | /* |
1745 | * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) | |
1746 | * @vma = virtual memory area whose policy is sought | |
1747 | * @addr = address in @vma for shared policy lookup and interleave policy | |
1748 | * @gfp_flags = for requested zone | |
19770b32 MG |
1749 | * @mpol = pointer to mempolicy pointer for reference counted mempolicy |
1750 | * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 | 1751 | * |
52cd3b07 LS |
1752 | * Returns a zonelist suitable for a huge page allocation and a pointer |
1753 | * to the struct mempolicy for conditional unref after allocation. | |
1754 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's | |
1755 | * @nodemask for filtering the zonelist. | |
c0ff7453 MX |
1756 | * |
1757 | * Must be protected by get_mems_allowed() | |
480eccf9 | 1758 | */ |
396faf03 | 1759 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
19770b32 MG |
1760 | gfp_t gfp_flags, struct mempolicy **mpol, |
1761 | nodemask_t **nodemask) | |
5da7ca86 | 1762 | { |
480eccf9 | 1763 | struct zonelist *zl; |
5da7ca86 | 1764 | |
52cd3b07 | 1765 | *mpol = get_vma_policy(current, vma, addr); |
19770b32 | 1766 | *nodemask = NULL; /* assume !MPOL_BIND */ |
5da7ca86 | 1767 | |
52cd3b07 LS |
1768 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
1769 | zl = node_zonelist(interleave_nid(*mpol, vma, addr, | |
a5516438 | 1770 | huge_page_shift(hstate_vma(vma))), gfp_flags); |
52cd3b07 | 1771 | } else { |
2f5f9486 | 1772 | zl = policy_zonelist(gfp_flags, *mpol, numa_node_id()); |
52cd3b07 LS |
1773 | if ((*mpol)->mode == MPOL_BIND) |
1774 | *nodemask = &(*mpol)->v.nodes; | |
480eccf9 LS |
1775 | } |
1776 | return zl; | |
5da7ca86 | 1777 | } |
06808b08 LS |
1778 | |
1779 | /* | |
1780 | * init_nodemask_of_mempolicy | |
1781 | * | |
1782 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
1783 | * to indicate default policy. Otherwise, extract the policy nodemask | |
1784 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
1785 | * initialize the argument nodemask to contain the single node for | |
1786 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
1787 | * of non-default mempolicy. | |
1788 | * | |
1789 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
1790 | * because the current task is examining it's own mempolicy and a task's | |
1791 | * mempolicy is only ever changed by the task itself. | |
1792 | * | |
1793 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
1794 | */ | |
1795 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
1796 | { | |
1797 | struct mempolicy *mempolicy; | |
1798 | int nid; | |
1799 | ||
1800 | if (!(mask && current->mempolicy)) | |
1801 | return false; | |
1802 | ||
c0ff7453 | 1803 | task_lock(current); |
06808b08 LS |
1804 | mempolicy = current->mempolicy; |
1805 | switch (mempolicy->mode) { | |
1806 | case MPOL_PREFERRED: | |
1807 | if (mempolicy->flags & MPOL_F_LOCAL) | |
1808 | nid = numa_node_id(); | |
1809 | else | |
1810 | nid = mempolicy->v.preferred_node; | |
1811 | init_nodemask_of_node(mask, nid); | |
1812 | break; | |
1813 | ||
1814 | case MPOL_BIND: | |
1815 | /* Fall through */ | |
1816 | case MPOL_INTERLEAVE: | |
1817 | *mask = mempolicy->v.nodes; | |
1818 | break; | |
1819 | ||
1820 | default: | |
1821 | BUG(); | |
1822 | } | |
c0ff7453 | 1823 | task_unlock(current); |
06808b08 LS |
1824 | |
1825 | return true; | |
1826 | } | |
00ac59ad | 1827 | #endif |
5da7ca86 | 1828 | |
6f48d0eb DR |
1829 | /* |
1830 | * mempolicy_nodemask_intersects | |
1831 | * | |
1832 | * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default | |
1833 | * policy. Otherwise, check for intersection between mask and the policy | |
1834 | * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local' | |
1835 | * policy, always return true since it may allocate elsewhere on fallback. | |
1836 | * | |
1837 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
1838 | */ | |
1839 | bool mempolicy_nodemask_intersects(struct task_struct *tsk, | |
1840 | const nodemask_t *mask) | |
1841 | { | |
1842 | struct mempolicy *mempolicy; | |
1843 | bool ret = true; | |
1844 | ||
1845 | if (!mask) | |
1846 | return ret; | |
1847 | task_lock(tsk); | |
1848 | mempolicy = tsk->mempolicy; | |
1849 | if (!mempolicy) | |
1850 | goto out; | |
1851 | ||
1852 | switch (mempolicy->mode) { | |
1853 | case MPOL_PREFERRED: | |
1854 | /* | |
1855 | * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to | |
1856 | * allocate from, they may fallback to other nodes when oom. | |
1857 | * Thus, it's possible for tsk to have allocated memory from | |
1858 | * nodes in mask. | |
1859 | */ | |
1860 | break; | |
1861 | case MPOL_BIND: | |
1862 | case MPOL_INTERLEAVE: | |
1863 | ret = nodes_intersects(mempolicy->v.nodes, *mask); | |
1864 | break; | |
1865 | default: | |
1866 | BUG(); | |
1867 | } | |
1868 | out: | |
1869 | task_unlock(tsk); | |
1870 | return ret; | |
1871 | } | |
1872 | ||
1da177e4 LT |
1873 | /* Allocate a page in interleaved policy. |
1874 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1875 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1876 | unsigned nid) | |
1da177e4 LT |
1877 | { |
1878 | struct zonelist *zl; | |
1879 | struct page *page; | |
1880 | ||
0e88460d | 1881 | zl = node_zonelist(nid, gfp); |
1da177e4 | 1882 | page = __alloc_pages(gfp, order, zl); |
dd1a239f | 1883 | if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0])) |
ca889e6c | 1884 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); |
1da177e4 LT |
1885 | return page; |
1886 | } | |
1887 | ||
1888 | /** | |
0bbbc0b3 | 1889 | * alloc_pages_vma - Allocate a page for a VMA. |
1da177e4 LT |
1890 | * |
1891 | * @gfp: | |
1892 | * %GFP_USER user allocation. | |
1893 | * %GFP_KERNEL kernel allocations, | |
1894 | * %GFP_HIGHMEM highmem/user allocations, | |
1895 | * %GFP_FS allocation should not call back into a file system. | |
1896 | * %GFP_ATOMIC don't sleep. | |
1897 | * | |
0bbbc0b3 | 1898 | * @order:Order of the GFP allocation. |
1da177e4 LT |
1899 | * @vma: Pointer to VMA or NULL if not available. |
1900 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
1901 | * | |
1902 | * This function allocates a page from the kernel page pool and applies | |
1903 | * a NUMA policy associated with the VMA or the current process. | |
1904 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
1905 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
1906 | * all allocations for pages that will be mapped into | |
1907 | * user space. Returns NULL when no page can be allocated. | |
1908 | * | |
1909 | * Should be called with the mm_sem of the vma hold. | |
1910 | */ | |
1911 | struct page * | |
0bbbc0b3 | 1912 | alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma, |
2f5f9486 | 1913 | unsigned long addr, int node) |
1da177e4 | 1914 | { |
cc9a6c87 | 1915 | struct mempolicy *pol; |
480eccf9 | 1916 | struct zonelist *zl; |
c0ff7453 | 1917 | struct page *page; |
cc9a6c87 MG |
1918 | unsigned int cpuset_mems_cookie; |
1919 | ||
1920 | retry_cpuset: | |
1921 | pol = get_vma_policy(current, vma, addr); | |
1922 | cpuset_mems_cookie = get_mems_allowed(); | |
1da177e4 | 1923 | |
45c4745a | 1924 | if (unlikely(pol->mode == MPOL_INTERLEAVE)) { |
1da177e4 | 1925 | unsigned nid; |
5da7ca86 | 1926 | |
8eac563c | 1927 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); |
52cd3b07 | 1928 | mpol_cond_put(pol); |
0bbbc0b3 | 1929 | page = alloc_page_interleave(gfp, order, nid); |
cc9a6c87 MG |
1930 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
1931 | goto retry_cpuset; | |
1932 | ||
c0ff7453 | 1933 | return page; |
1da177e4 | 1934 | } |
2f5f9486 | 1935 | zl = policy_zonelist(gfp, pol, node); |
52cd3b07 | 1936 | if (unlikely(mpol_needs_cond_ref(pol))) { |
480eccf9 | 1937 | /* |
52cd3b07 | 1938 | * slow path: ref counted shared policy |
480eccf9 | 1939 | */ |
0bbbc0b3 | 1940 | struct page *page = __alloc_pages_nodemask(gfp, order, |
52cd3b07 | 1941 | zl, policy_nodemask(gfp, pol)); |
f0be3d32 | 1942 | __mpol_put(pol); |
cc9a6c87 MG |
1943 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
1944 | goto retry_cpuset; | |
480eccf9 LS |
1945 | return page; |
1946 | } | |
1947 | /* | |
1948 | * fast path: default or task policy | |
1949 | */ | |
0bbbc0b3 AA |
1950 | page = __alloc_pages_nodemask(gfp, order, zl, |
1951 | policy_nodemask(gfp, pol)); | |
cc9a6c87 MG |
1952 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) |
1953 | goto retry_cpuset; | |
c0ff7453 | 1954 | return page; |
1da177e4 LT |
1955 | } |
1956 | ||
1957 | /** | |
1958 | * alloc_pages_current - Allocate pages. | |
1959 | * | |
1960 | * @gfp: | |
1961 | * %GFP_USER user allocation, | |
1962 | * %GFP_KERNEL kernel allocation, | |
1963 | * %GFP_HIGHMEM highmem allocation, | |
1964 | * %GFP_FS don't call back into a file system. | |
1965 | * %GFP_ATOMIC don't sleep. | |
1966 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
1967 | * | |
1968 | * Allocate a page from the kernel page pool. When not in | |
1969 | * interrupt context and apply the current process NUMA policy. | |
1970 | * Returns NULL when no page can be allocated. | |
1971 | * | |
cf2a473c | 1972 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
1973 | * 1) it's ok to take cpuset_sem (can WAIT), and |
1974 | * 2) allocating for current task (not interrupt). | |
1975 | */ | |
dd0fc66f | 1976 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 LT |
1977 | { |
1978 | struct mempolicy *pol = current->mempolicy; | |
c0ff7453 | 1979 | struct page *page; |
cc9a6c87 | 1980 | unsigned int cpuset_mems_cookie; |
1da177e4 | 1981 | |
9b819d20 | 1982 | if (!pol || in_interrupt() || (gfp & __GFP_THISNODE)) |
1da177e4 | 1983 | pol = &default_policy; |
52cd3b07 | 1984 | |
cc9a6c87 MG |
1985 | retry_cpuset: |
1986 | cpuset_mems_cookie = get_mems_allowed(); | |
1987 | ||
52cd3b07 LS |
1988 | /* |
1989 | * No reference counting needed for current->mempolicy | |
1990 | * nor system default_policy | |
1991 | */ | |
45c4745a | 1992 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 MX |
1993 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
1994 | else | |
1995 | page = __alloc_pages_nodemask(gfp, order, | |
5c4b4be3 AK |
1996 | policy_zonelist(gfp, pol, numa_node_id()), |
1997 | policy_nodemask(gfp, pol)); | |
cc9a6c87 MG |
1998 | |
1999 | if (unlikely(!put_mems_allowed(cpuset_mems_cookie) && !page)) | |
2000 | goto retry_cpuset; | |
2001 | ||
c0ff7453 | 2002 | return page; |
1da177e4 LT |
2003 | } |
2004 | EXPORT_SYMBOL(alloc_pages_current); | |
2005 | ||
4225399a | 2006 | /* |
846a16bf | 2007 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2008 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2009 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2010 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2011 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2012 | * |
2013 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2014 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2015 | */ |
4225399a | 2016 | |
846a16bf LS |
2017 | /* Slow path of a mempolicy duplicate */ |
2018 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2019 | { |
2020 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2021 | ||
2022 | if (!new) | |
2023 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2024 | |
2025 | /* task's mempolicy is protected by alloc_lock */ | |
2026 | if (old == current->mempolicy) { | |
2027 | task_lock(current); | |
2028 | *new = *old; | |
2029 | task_unlock(current); | |
2030 | } else | |
2031 | *new = *old; | |
2032 | ||
99ee4ca7 | 2033 | rcu_read_lock(); |
4225399a PJ |
2034 | if (current_cpuset_is_being_rebound()) { |
2035 | nodemask_t mems = cpuset_mems_allowed(current); | |
708c1bbc MX |
2036 | if (new->flags & MPOL_F_REBINDING) |
2037 | mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2); | |
2038 | else | |
2039 | mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE); | |
4225399a | 2040 | } |
99ee4ca7 | 2041 | rcu_read_unlock(); |
1da177e4 | 2042 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2043 | return new; |
2044 | } | |
2045 | ||
52cd3b07 LS |
2046 | /* |
2047 | * If *frompol needs [has] an extra ref, copy *frompol to *tompol , | |
2048 | * eliminate the * MPOL_F_* flags that require conditional ref and | |
2049 | * [NOTE!!!] drop the extra ref. Not safe to reference *frompol directly | |
2050 | * after return. Use the returned value. | |
2051 | * | |
2052 | * Allows use of a mempolicy for, e.g., multiple allocations with a single | |
2053 | * policy lookup, even if the policy needs/has extra ref on lookup. | |
2054 | * shmem_readahead needs this. | |
2055 | */ | |
2056 | struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol, | |
2057 | struct mempolicy *frompol) | |
2058 | { | |
2059 | if (!mpol_needs_cond_ref(frompol)) | |
2060 | return frompol; | |
2061 | ||
2062 | *tompol = *frompol; | |
2063 | tompol->flags &= ~MPOL_F_SHARED; /* copy doesn't need unref */ | |
2064 | __mpol_put(frompol); | |
2065 | return tompol; | |
2066 | } | |
2067 | ||
1da177e4 | 2068 | /* Slow path of a mempolicy comparison */ |
fcfb4dcc | 2069 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2070 | { |
2071 | if (!a || !b) | |
fcfb4dcc | 2072 | return false; |
45c4745a | 2073 | if (a->mode != b->mode) |
fcfb4dcc | 2074 | return false; |
19800502 | 2075 | if (a->flags != b->flags) |
fcfb4dcc | 2076 | return false; |
19800502 BL |
2077 | if (mpol_store_user_nodemask(a)) |
2078 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2079 | return false; |
19800502 | 2080 | |
45c4745a | 2081 | switch (a->mode) { |
19770b32 MG |
2082 | case MPOL_BIND: |
2083 | /* Fall through */ | |
1da177e4 | 2084 | case MPOL_INTERLEAVE: |
fcfb4dcc | 2085 | return !!nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 | 2086 | case MPOL_PREFERRED: |
75719661 | 2087 | return a->v.preferred_node == b->v.preferred_node; |
1da177e4 LT |
2088 | default: |
2089 | BUG(); | |
fcfb4dcc | 2090 | return false; |
1da177e4 LT |
2091 | } |
2092 | } | |
2093 | ||
1da177e4 LT |
2094 | /* |
2095 | * Shared memory backing store policy support. | |
2096 | * | |
2097 | * Remember policies even when nobody has shared memory mapped. | |
2098 | * The policies are kept in Red-Black tree linked from the inode. | |
2099 | * They are protected by the sp->lock spinlock, which should be held | |
2100 | * for any accesses to the tree. | |
2101 | */ | |
2102 | ||
2103 | /* lookup first element intersecting start-end */ | |
b22d127a | 2104 | /* Caller holds sp->mutex */ |
1da177e4 LT |
2105 | static struct sp_node * |
2106 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2107 | { | |
2108 | struct rb_node *n = sp->root.rb_node; | |
2109 | ||
2110 | while (n) { | |
2111 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2112 | ||
2113 | if (start >= p->end) | |
2114 | n = n->rb_right; | |
2115 | else if (end <= p->start) | |
2116 | n = n->rb_left; | |
2117 | else | |
2118 | break; | |
2119 | } | |
2120 | if (!n) | |
2121 | return NULL; | |
2122 | for (;;) { | |
2123 | struct sp_node *w = NULL; | |
2124 | struct rb_node *prev = rb_prev(n); | |
2125 | if (!prev) | |
2126 | break; | |
2127 | w = rb_entry(prev, struct sp_node, nd); | |
2128 | if (w->end <= start) | |
2129 | break; | |
2130 | n = prev; | |
2131 | } | |
2132 | return rb_entry(n, struct sp_node, nd); | |
2133 | } | |
2134 | ||
2135 | /* Insert a new shared policy into the list. */ | |
2136 | /* Caller holds sp->lock */ | |
2137 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) | |
2138 | { | |
2139 | struct rb_node **p = &sp->root.rb_node; | |
2140 | struct rb_node *parent = NULL; | |
2141 | struct sp_node *nd; | |
2142 | ||
2143 | while (*p) { | |
2144 | parent = *p; | |
2145 | nd = rb_entry(parent, struct sp_node, nd); | |
2146 | if (new->start < nd->start) | |
2147 | p = &(*p)->rb_left; | |
2148 | else if (new->end > nd->end) | |
2149 | p = &(*p)->rb_right; | |
2150 | else | |
2151 | BUG(); | |
2152 | } | |
2153 | rb_link_node(&new->nd, parent, p); | |
2154 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2155 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2156 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2157 | } |
2158 | ||
2159 | /* Find shared policy intersecting idx */ | |
2160 | struct mempolicy * | |
2161 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2162 | { | |
2163 | struct mempolicy *pol = NULL; | |
2164 | struct sp_node *sn; | |
2165 | ||
2166 | if (!sp->root.rb_node) | |
2167 | return NULL; | |
b22d127a | 2168 | mutex_lock(&sp->mutex); |
1da177e4 LT |
2169 | sn = sp_lookup(sp, idx, idx+1); |
2170 | if (sn) { | |
2171 | mpol_get(sn->policy); | |
2172 | pol = sn->policy; | |
2173 | } | |
b22d127a | 2174 | mutex_unlock(&sp->mutex); |
1da177e4 LT |
2175 | return pol; |
2176 | } | |
2177 | ||
63f74ca2 KM |
2178 | static void sp_free(struct sp_node *n) |
2179 | { | |
2180 | mpol_put(n->policy); | |
2181 | kmem_cache_free(sn_cache, n); | |
2182 | } | |
2183 | ||
771fb4d8 LS |
2184 | /** |
2185 | * mpol_misplaced - check whether current page node is valid in policy | |
2186 | * | |
2187 | * @page - page to be checked | |
2188 | * @vma - vm area where page mapped | |
2189 | * @addr - virtual address where page mapped | |
2190 | * | |
2191 | * Lookup current policy node id for vma,addr and "compare to" page's | |
2192 | * node id. | |
2193 | * | |
2194 | * Returns: | |
2195 | * -1 - not misplaced, page is in the right node | |
2196 | * node - node id where the page should be | |
2197 | * | |
2198 | * Policy determination "mimics" alloc_page_vma(). | |
2199 | * Called from fault path where we know the vma and faulting address. | |
2200 | */ | |
2201 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) | |
2202 | { | |
2203 | struct mempolicy *pol; | |
2204 | struct zone *zone; | |
2205 | int curnid = page_to_nid(page); | |
2206 | unsigned long pgoff; | |
2207 | int polnid = -1; | |
2208 | int ret = -1; | |
2209 | ||
2210 | BUG_ON(!vma); | |
2211 | ||
2212 | pol = get_vma_policy(current, vma, addr); | |
2213 | if (!(pol->flags & MPOL_F_MOF)) | |
2214 | goto out; | |
2215 | ||
2216 | switch (pol->mode) { | |
2217 | case MPOL_INTERLEAVE: | |
2218 | BUG_ON(addr >= vma->vm_end); | |
2219 | BUG_ON(addr < vma->vm_start); | |
2220 | ||
2221 | pgoff = vma->vm_pgoff; | |
2222 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; | |
2223 | polnid = offset_il_node(pol, vma, pgoff); | |
2224 | break; | |
2225 | ||
2226 | case MPOL_PREFERRED: | |
2227 | if (pol->flags & MPOL_F_LOCAL) | |
2228 | polnid = numa_node_id(); | |
2229 | else | |
2230 | polnid = pol->v.preferred_node; | |
2231 | break; | |
2232 | ||
2233 | case MPOL_BIND: | |
2234 | /* | |
2235 | * allows binding to multiple nodes. | |
2236 | * use current page if in policy nodemask, | |
2237 | * else select nearest allowed node, if any. | |
2238 | * If no allowed nodes, use current [!misplaced]. | |
2239 | */ | |
2240 | if (node_isset(curnid, pol->v.nodes)) | |
2241 | goto out; | |
2242 | (void)first_zones_zonelist( | |
2243 | node_zonelist(numa_node_id(), GFP_HIGHUSER), | |
2244 | gfp_zone(GFP_HIGHUSER), | |
2245 | &pol->v.nodes, &zone); | |
2246 | polnid = zone->node; | |
2247 | break; | |
2248 | ||
2249 | default: | |
2250 | BUG(); | |
2251 | } | |
2252 | if (curnid != polnid) | |
2253 | ret = polnid; | |
2254 | out: | |
2255 | mpol_cond_put(pol); | |
2256 | ||
2257 | return ret; | |
2258 | } | |
2259 | ||
1da177e4 LT |
2260 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2261 | { | |
140d5a49 | 2262 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2263 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2264 | sp_free(n); |
1da177e4 LT |
2265 | } |
2266 | ||
dbcb0f19 AB |
2267 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2268 | struct mempolicy *pol) | |
1da177e4 | 2269 | { |
869833f2 KM |
2270 | struct sp_node *n; |
2271 | struct mempolicy *newpol; | |
1da177e4 | 2272 | |
869833f2 | 2273 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2274 | if (!n) |
2275 | return NULL; | |
869833f2 KM |
2276 | |
2277 | newpol = mpol_dup(pol); | |
2278 | if (IS_ERR(newpol)) { | |
2279 | kmem_cache_free(sn_cache, n); | |
2280 | return NULL; | |
2281 | } | |
2282 | newpol->flags |= MPOL_F_SHARED; | |
2283 | ||
1da177e4 LT |
2284 | n->start = start; |
2285 | n->end = end; | |
869833f2 KM |
2286 | n->policy = newpol; |
2287 | ||
1da177e4 LT |
2288 | return n; |
2289 | } | |
2290 | ||
2291 | /* Replace a policy range. */ | |
2292 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2293 | unsigned long end, struct sp_node *new) | |
2294 | { | |
b22d127a MG |
2295 | struct sp_node *n; |
2296 | int ret = 0; | |
1da177e4 | 2297 | |
b22d127a | 2298 | mutex_lock(&sp->mutex); |
1da177e4 LT |
2299 | n = sp_lookup(sp, start, end); |
2300 | /* Take care of old policies in the same range. */ | |
2301 | while (n && n->start < end) { | |
2302 | struct rb_node *next = rb_next(&n->nd); | |
2303 | if (n->start >= start) { | |
2304 | if (n->end <= end) | |
2305 | sp_delete(sp, n); | |
2306 | else | |
2307 | n->start = end; | |
2308 | } else { | |
2309 | /* Old policy spanning whole new range. */ | |
2310 | if (n->end > end) { | |
b22d127a MG |
2311 | struct sp_node *new2; |
2312 | new2 = sp_alloc(end, n->end, n->policy); | |
1da177e4 | 2313 | if (!new2) { |
b22d127a MG |
2314 | ret = -ENOMEM; |
2315 | goto out; | |
1da177e4 LT |
2316 | } |
2317 | n->end = start; | |
2318 | sp_insert(sp, new2); | |
1da177e4 LT |
2319 | break; |
2320 | } else | |
2321 | n->end = start; | |
2322 | } | |
2323 | if (!next) | |
2324 | break; | |
2325 | n = rb_entry(next, struct sp_node, nd); | |
2326 | } | |
2327 | if (new) | |
2328 | sp_insert(sp, new); | |
b22d127a MG |
2329 | out: |
2330 | mutex_unlock(&sp->mutex); | |
2331 | return ret; | |
1da177e4 LT |
2332 | } |
2333 | ||
71fe804b LS |
2334 | /** |
2335 | * mpol_shared_policy_init - initialize shared policy for inode | |
2336 | * @sp: pointer to inode shared policy | |
2337 | * @mpol: struct mempolicy to install | |
2338 | * | |
2339 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2340 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2341 | * This must be released on exit. | |
4bfc4495 | 2342 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2343 | */ |
2344 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2345 | { | |
58568d2a MX |
2346 | int ret; |
2347 | ||
71fe804b | 2348 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
b22d127a | 2349 | mutex_init(&sp->mutex); |
71fe804b LS |
2350 | |
2351 | if (mpol) { | |
2352 | struct vm_area_struct pvma; | |
2353 | struct mempolicy *new; | |
4bfc4495 | 2354 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2355 | |
4bfc4495 | 2356 | if (!scratch) |
5c0c1654 | 2357 | goto put_mpol; |
71fe804b LS |
2358 | /* contextualize the tmpfs mount point mempolicy */ |
2359 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2360 | if (IS_ERR(new)) |
0cae3457 | 2361 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2362 | |
2363 | task_lock(current); | |
4bfc4495 | 2364 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2365 | task_unlock(current); |
15d77835 | 2366 | if (ret) |
5c0c1654 | 2367 | goto put_new; |
71fe804b LS |
2368 | |
2369 | /* Create pseudo-vma that contains just the policy */ | |
2370 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
2371 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ | |
2372 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2373 | |
5c0c1654 | 2374 | put_new: |
71fe804b | 2375 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2376 | free_scratch: |
4bfc4495 | 2377 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2378 | put_mpol: |
2379 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2380 | } |
2381 | } | |
2382 | ||
1da177e4 LT |
2383 | int mpol_set_shared_policy(struct shared_policy *info, |
2384 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2385 | { | |
2386 | int err; | |
2387 | struct sp_node *new = NULL; | |
2388 | unsigned long sz = vma_pages(vma); | |
2389 | ||
028fec41 | 2390 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2391 | vma->vm_pgoff, |
45c4745a | 2392 | sz, npol ? npol->mode : -1, |
028fec41 | 2393 | npol ? npol->flags : -1, |
140d5a49 | 2394 | npol ? nodes_addr(npol->v.nodes)[0] : -1); |
1da177e4 LT |
2395 | |
2396 | if (npol) { | |
2397 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2398 | if (!new) | |
2399 | return -ENOMEM; | |
2400 | } | |
2401 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2402 | if (err && new) | |
63f74ca2 | 2403 | sp_free(new); |
1da177e4 LT |
2404 | return err; |
2405 | } | |
2406 | ||
2407 | /* Free a backing policy store on inode delete. */ | |
2408 | void mpol_free_shared_policy(struct shared_policy *p) | |
2409 | { | |
2410 | struct sp_node *n; | |
2411 | struct rb_node *next; | |
2412 | ||
2413 | if (!p->root.rb_node) | |
2414 | return; | |
b22d127a | 2415 | mutex_lock(&p->mutex); |
1da177e4 LT |
2416 | next = rb_first(&p->root); |
2417 | while (next) { | |
2418 | n = rb_entry(next, struct sp_node, nd); | |
2419 | next = rb_next(&n->nd); | |
63f74ca2 | 2420 | sp_delete(p, n); |
1da177e4 | 2421 | } |
b22d127a | 2422 | mutex_unlock(&p->mutex); |
1da177e4 LT |
2423 | } |
2424 | ||
2425 | /* assumes fs == KERNEL_DS */ | |
2426 | void __init numa_policy_init(void) | |
2427 | { | |
b71636e2 PM |
2428 | nodemask_t interleave_nodes; |
2429 | unsigned long largest = 0; | |
2430 | int nid, prefer = 0; | |
2431 | ||
1da177e4 LT |
2432 | policy_cache = kmem_cache_create("numa_policy", |
2433 | sizeof(struct mempolicy), | |
20c2df83 | 2434 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2435 | |
2436 | sn_cache = kmem_cache_create("shared_policy_node", | |
2437 | sizeof(struct sp_node), | |
20c2df83 | 2438 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2439 | |
b71636e2 PM |
2440 | /* |
2441 | * Set interleaving policy for system init. Interleaving is only | |
2442 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2443 | * fall back to the largest node if they're all smaller. | |
2444 | */ | |
2445 | nodes_clear(interleave_nodes); | |
56bbd65d | 2446 | for_each_node_state(nid, N_HIGH_MEMORY) { |
b71636e2 PM |
2447 | unsigned long total_pages = node_present_pages(nid); |
2448 | ||
2449 | /* Preserve the largest node */ | |
2450 | if (largest < total_pages) { | |
2451 | largest = total_pages; | |
2452 | prefer = nid; | |
2453 | } | |
2454 | ||
2455 | /* Interleave this node? */ | |
2456 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2457 | node_set(nid, interleave_nodes); | |
2458 | } | |
2459 | ||
2460 | /* All too small, use the largest */ | |
2461 | if (unlikely(nodes_empty(interleave_nodes))) | |
2462 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2463 | |
028fec41 | 2464 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
1da177e4 LT |
2465 | printk("numa_policy_init: interleaving failed\n"); |
2466 | } | |
2467 | ||
8bccd85f | 2468 | /* Reset policy of current process to default */ |
1da177e4 LT |
2469 | void numa_default_policy(void) |
2470 | { | |
028fec41 | 2471 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2472 | } |
68860ec1 | 2473 | |
095f1fc4 LS |
2474 | /* |
2475 | * Parse and format mempolicy from/to strings | |
2476 | */ | |
2477 | ||
1a75a6c8 | 2478 | /* |
fc36b8d3 | 2479 | * "local" is pseudo-policy: MPOL_PREFERRED with MPOL_F_LOCAL flag |
3f226aa1 | 2480 | * Used only for mpol_parse_str() and mpol_to_str() |
1a75a6c8 | 2481 | */ |
345ace9c LS |
2482 | static const char * const policy_modes[] = |
2483 | { | |
2484 | [MPOL_DEFAULT] = "default", | |
2485 | [MPOL_PREFERRED] = "prefer", | |
2486 | [MPOL_BIND] = "bind", | |
2487 | [MPOL_INTERLEAVE] = "interleave", | |
d3a71033 LS |
2488 | [MPOL_LOCAL] = "local", |
2489 | [MPOL_NOOP] = "noop", /* should not actually be used */ | |
345ace9c | 2490 | }; |
1a75a6c8 | 2491 | |
095f1fc4 LS |
2492 | |
2493 | #ifdef CONFIG_TMPFS | |
2494 | /** | |
2495 | * mpol_parse_str - parse string to mempolicy | |
2496 | * @str: string containing mempolicy to parse | |
71fe804b LS |
2497 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
2498 | * @no_context: flag whether to "contextualize" the mempolicy | |
095f1fc4 LS |
2499 | * |
2500 | * Format of input: | |
2501 | * <mode>[=<flags>][:<nodelist>] | |
2502 | * | |
71fe804b LS |
2503 | * if @no_context is true, save the input nodemask in w.user_nodemask in |
2504 | * the returned mempolicy. This will be used to "clone" the mempolicy in | |
2505 | * a specific context [cpuset] at a later time. Used to parse tmpfs mpol | |
2506 | * mount option. Note that if 'static' or 'relative' mode flags were | |
2507 | * specified, the input nodemask will already have been saved. Saving | |
2508 | * it again is redundant, but safe. | |
2509 | * | |
2510 | * On success, returns 0, else 1 | |
095f1fc4 | 2511 | */ |
71fe804b | 2512 | int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context) |
095f1fc4 | 2513 | { |
71fe804b | 2514 | struct mempolicy *new = NULL; |
b4652e84 | 2515 | unsigned short mode; |
71fe804b LS |
2516 | unsigned short uninitialized_var(mode_flags); |
2517 | nodemask_t nodes; | |
095f1fc4 LS |
2518 | char *nodelist = strchr(str, ':'); |
2519 | char *flags = strchr(str, '='); | |
095f1fc4 LS |
2520 | int err = 1; |
2521 | ||
2522 | if (nodelist) { | |
2523 | /* NUL-terminate mode or flags string */ | |
2524 | *nodelist++ = '\0'; | |
71fe804b | 2525 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2526 | goto out; |
71fe804b | 2527 | if (!nodes_subset(nodes, node_states[N_HIGH_MEMORY])) |
095f1fc4 | 2528 | goto out; |
71fe804b LS |
2529 | } else |
2530 | nodes_clear(nodes); | |
2531 | ||
095f1fc4 LS |
2532 | if (flags) |
2533 | *flags++ = '\0'; /* terminate mode string */ | |
2534 | ||
479e2802 | 2535 | for (mode = 0; mode < MPOL_MAX; mode++) { |
345ace9c | 2536 | if (!strcmp(str, policy_modes[mode])) { |
095f1fc4 LS |
2537 | break; |
2538 | } | |
2539 | } | |
d3a71033 | 2540 | if (mode >= MPOL_MAX || mode == MPOL_NOOP) |
095f1fc4 LS |
2541 | goto out; |
2542 | ||
71fe804b | 2543 | switch (mode) { |
095f1fc4 | 2544 | case MPOL_PREFERRED: |
71fe804b LS |
2545 | /* |
2546 | * Insist on a nodelist of one node only | |
2547 | */ | |
095f1fc4 LS |
2548 | if (nodelist) { |
2549 | char *rest = nodelist; | |
2550 | while (isdigit(*rest)) | |
2551 | rest++; | |
926f2ae0 KM |
2552 | if (*rest) |
2553 | goto out; | |
095f1fc4 LS |
2554 | } |
2555 | break; | |
095f1fc4 LS |
2556 | case MPOL_INTERLEAVE: |
2557 | /* | |
2558 | * Default to online nodes with memory if no nodelist | |
2559 | */ | |
2560 | if (!nodelist) | |
71fe804b | 2561 | nodes = node_states[N_HIGH_MEMORY]; |
3f226aa1 | 2562 | break; |
71fe804b | 2563 | case MPOL_LOCAL: |
3f226aa1 | 2564 | /* |
71fe804b | 2565 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 2566 | */ |
71fe804b | 2567 | if (nodelist) |
3f226aa1 | 2568 | goto out; |
71fe804b | 2569 | mode = MPOL_PREFERRED; |
3f226aa1 | 2570 | break; |
413b43de RT |
2571 | case MPOL_DEFAULT: |
2572 | /* | |
2573 | * Insist on a empty nodelist | |
2574 | */ | |
2575 | if (!nodelist) | |
2576 | err = 0; | |
2577 | goto out; | |
d69b2e63 KM |
2578 | case MPOL_BIND: |
2579 | /* | |
2580 | * Insist on a nodelist | |
2581 | */ | |
2582 | if (!nodelist) | |
2583 | goto out; | |
095f1fc4 LS |
2584 | } |
2585 | ||
71fe804b | 2586 | mode_flags = 0; |
095f1fc4 LS |
2587 | if (flags) { |
2588 | /* | |
2589 | * Currently, we only support two mutually exclusive | |
2590 | * mode flags. | |
2591 | */ | |
2592 | if (!strcmp(flags, "static")) | |
71fe804b | 2593 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 2594 | else if (!strcmp(flags, "relative")) |
71fe804b | 2595 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 2596 | else |
926f2ae0 | 2597 | goto out; |
095f1fc4 | 2598 | } |
71fe804b LS |
2599 | |
2600 | new = mpol_new(mode, mode_flags, &nodes); | |
2601 | if (IS_ERR(new)) | |
926f2ae0 KM |
2602 | goto out; |
2603 | ||
e17f74af LS |
2604 | if (no_context) { |
2605 | /* save for contextualization */ | |
2606 | new->w.user_nodemask = nodes; | |
2607 | } else { | |
58568d2a | 2608 | int ret; |
4bfc4495 KH |
2609 | NODEMASK_SCRATCH(scratch); |
2610 | if (scratch) { | |
2611 | task_lock(current); | |
2612 | ret = mpol_set_nodemask(new, &nodes, scratch); | |
2613 | task_unlock(current); | |
2614 | } else | |
2615 | ret = -ENOMEM; | |
2616 | NODEMASK_SCRATCH_FREE(scratch); | |
2617 | if (ret) { | |
4bfc4495 | 2618 | mpol_put(new); |
926f2ae0 | 2619 | goto out; |
58568d2a MX |
2620 | } |
2621 | } | |
926f2ae0 | 2622 | err = 0; |
71fe804b | 2623 | |
095f1fc4 LS |
2624 | out: |
2625 | /* Restore string for error message */ | |
2626 | if (nodelist) | |
2627 | *--nodelist = ':'; | |
2628 | if (flags) | |
2629 | *--flags = '='; | |
71fe804b LS |
2630 | if (!err) |
2631 | *mpol = new; | |
095f1fc4 LS |
2632 | return err; |
2633 | } | |
2634 | #endif /* CONFIG_TMPFS */ | |
2635 | ||
71fe804b LS |
2636 | /** |
2637 | * mpol_to_str - format a mempolicy structure for printing | |
2638 | * @buffer: to contain formatted mempolicy string | |
2639 | * @maxlen: length of @buffer | |
2640 | * @pol: pointer to mempolicy to be formatted | |
2641 | * @no_context: "context free" mempolicy - use nodemask in w.user_nodemask | |
2642 | * | |
1a75a6c8 CL |
2643 | * Convert a mempolicy into a string. |
2644 | * Returns the number of characters in buffer (if positive) | |
2645 | * or an error (negative) | |
2646 | */ | |
71fe804b | 2647 | int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol, int no_context) |
1a75a6c8 CL |
2648 | { |
2649 | char *p = buffer; | |
2650 | int l; | |
2651 | nodemask_t nodes; | |
bea904d5 | 2652 | unsigned short mode; |
f5b087b5 | 2653 | unsigned short flags = pol ? pol->flags : 0; |
1a75a6c8 | 2654 | |
2291990a LS |
2655 | /* |
2656 | * Sanity check: room for longest mode, flag and some nodes | |
2657 | */ | |
2658 | VM_BUG_ON(maxlen < strlen("interleave") + strlen("relative") + 16); | |
2659 | ||
bea904d5 LS |
2660 | if (!pol || pol == &default_policy) |
2661 | mode = MPOL_DEFAULT; | |
2662 | else | |
2663 | mode = pol->mode; | |
2664 | ||
1a75a6c8 CL |
2665 | switch (mode) { |
2666 | case MPOL_DEFAULT: | |
2667 | nodes_clear(nodes); | |
2668 | break; | |
2669 | ||
2670 | case MPOL_PREFERRED: | |
2671 | nodes_clear(nodes); | |
fc36b8d3 | 2672 | if (flags & MPOL_F_LOCAL) |
53f2556b LS |
2673 | mode = MPOL_LOCAL; /* pseudo-policy */ |
2674 | else | |
fc36b8d3 | 2675 | node_set(pol->v.preferred_node, nodes); |
1a75a6c8 CL |
2676 | break; |
2677 | ||
2678 | case MPOL_BIND: | |
19770b32 | 2679 | /* Fall through */ |
1a75a6c8 | 2680 | case MPOL_INTERLEAVE: |
71fe804b LS |
2681 | if (no_context) |
2682 | nodes = pol->w.user_nodemask; | |
2683 | else | |
2684 | nodes = pol->v.nodes; | |
1a75a6c8 CL |
2685 | break; |
2686 | ||
2687 | default: | |
80de7c31 | 2688 | return -EINVAL; |
1a75a6c8 CL |
2689 | } |
2690 | ||
345ace9c | 2691 | l = strlen(policy_modes[mode]); |
53f2556b LS |
2692 | if (buffer + maxlen < p + l + 1) |
2693 | return -ENOSPC; | |
1a75a6c8 | 2694 | |
345ace9c | 2695 | strcpy(p, policy_modes[mode]); |
1a75a6c8 CL |
2696 | p += l; |
2697 | ||
fc36b8d3 | 2698 | if (flags & MPOL_MODE_FLAGS) { |
f5b087b5 DR |
2699 | if (buffer + maxlen < p + 2) |
2700 | return -ENOSPC; | |
2701 | *p++ = '='; | |
2702 | ||
2291990a LS |
2703 | /* |
2704 | * Currently, the only defined flags are mutually exclusive | |
2705 | */ | |
f5b087b5 | 2706 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
2707 | p += snprintf(p, buffer + maxlen - p, "static"); |
2708 | else if (flags & MPOL_F_RELATIVE_NODES) | |
2709 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
2710 | } |
2711 | ||
1a75a6c8 CL |
2712 | if (!nodes_empty(nodes)) { |
2713 | if (buffer + maxlen < p + 2) | |
2714 | return -ENOSPC; | |
095f1fc4 | 2715 | *p++ = ':'; |
1a75a6c8 CL |
2716 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); |
2717 | } | |
2718 | return p - buffer; | |
2719 | } |