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 | |
3e1f0645 DR |
254 | if (mode == MPOL_DEFAULT) { |
255 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 256 | return ERR_PTR(-EINVAL); |
bea904d5 | 257 | return NULL; /* simply delete any existing policy */ |
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 DR |
271 | } |
272 | } else if (nodes_empty(*nodes)) | |
273 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
274 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
275 | if (!policy) | |
276 | return ERR_PTR(-ENOMEM); | |
277 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 278 | policy->mode = mode; |
3e1f0645 | 279 | policy->flags = flags; |
37012946 | 280 | |
1da177e4 | 281 | return policy; |
37012946 DR |
282 | } |
283 | ||
52cd3b07 LS |
284 | /* Slow path of a mpol destructor. */ |
285 | void __mpol_put(struct mempolicy *p) | |
286 | { | |
287 | if (!atomic_dec_and_test(&p->refcnt)) | |
288 | return; | |
52cd3b07 LS |
289 | kmem_cache_free(policy_cache, p); |
290 | } | |
291 | ||
708c1bbc MX |
292 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes, |
293 | enum mpol_rebind_step step) | |
37012946 DR |
294 | { |
295 | } | |
296 | ||
708c1bbc MX |
297 | /* |
298 | * step: | |
299 | * MPOL_REBIND_ONCE - do rebind work at once | |
300 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
301 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
302 | */ | |
303 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes, | |
304 | enum mpol_rebind_step step) | |
37012946 DR |
305 | { |
306 | nodemask_t tmp; | |
307 | ||
308 | if (pol->flags & MPOL_F_STATIC_NODES) | |
309 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
310 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
311 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
312 | else { | |
708c1bbc MX |
313 | /* |
314 | * if step == 1, we use ->w.cpuset_mems_allowed to cache the | |
315 | * result | |
316 | */ | |
317 | if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP1) { | |
318 | nodes_remap(tmp, pol->v.nodes, | |
319 | pol->w.cpuset_mems_allowed, *nodes); | |
320 | pol->w.cpuset_mems_allowed = step ? tmp : *nodes; | |
321 | } else if (step == MPOL_REBIND_STEP2) { | |
322 | tmp = pol->w.cpuset_mems_allowed; | |
323 | pol->w.cpuset_mems_allowed = *nodes; | |
324 | } else | |
325 | BUG(); | |
37012946 | 326 | } |
f5b087b5 | 327 | |
708c1bbc MX |
328 | if (nodes_empty(tmp)) |
329 | tmp = *nodes; | |
330 | ||
331 | if (step == MPOL_REBIND_STEP1) | |
332 | nodes_or(pol->v.nodes, pol->v.nodes, tmp); | |
333 | else if (step == MPOL_REBIND_ONCE || step == MPOL_REBIND_STEP2) | |
334 | pol->v.nodes = tmp; | |
335 | else | |
336 | BUG(); | |
337 | ||
37012946 DR |
338 | if (!node_isset(current->il_next, tmp)) { |
339 | current->il_next = next_node(current->il_next, tmp); | |
340 | if (current->il_next >= MAX_NUMNODES) | |
341 | current->il_next = first_node(tmp); | |
342 | if (current->il_next >= MAX_NUMNODES) | |
343 | current->il_next = numa_node_id(); | |
344 | } | |
345 | } | |
346 | ||
347 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
708c1bbc MX |
348 | const nodemask_t *nodes, |
349 | enum mpol_rebind_step step) | |
37012946 DR |
350 | { |
351 | nodemask_t tmp; | |
352 | ||
37012946 DR |
353 | if (pol->flags & MPOL_F_STATIC_NODES) { |
354 | int node = first_node(pol->w.user_nodemask); | |
355 | ||
fc36b8d3 | 356 | if (node_isset(node, *nodes)) { |
37012946 | 357 | pol->v.preferred_node = node; |
fc36b8d3 LS |
358 | pol->flags &= ~MPOL_F_LOCAL; |
359 | } else | |
360 | pol->flags |= MPOL_F_LOCAL; | |
37012946 DR |
361 | } else if (pol->flags & MPOL_F_RELATIVE_NODES) { |
362 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
363 | pol->v.preferred_node = first_node(tmp); | |
fc36b8d3 | 364 | } else if (!(pol->flags & MPOL_F_LOCAL)) { |
37012946 DR |
365 | pol->v.preferred_node = node_remap(pol->v.preferred_node, |
366 | pol->w.cpuset_mems_allowed, | |
367 | *nodes); | |
368 | pol->w.cpuset_mems_allowed = *nodes; | |
369 | } | |
1da177e4 LT |
370 | } |
371 | ||
708c1bbc MX |
372 | /* |
373 | * mpol_rebind_policy - Migrate a policy to a different set of nodes | |
374 | * | |
375 | * If read-side task has no lock to protect task->mempolicy, write-side | |
376 | * task will rebind the task->mempolicy by two step. The first step is | |
377 | * setting all the newly nodes, and the second step is cleaning all the | |
378 | * disallowed nodes. In this way, we can avoid finding no node to alloc | |
379 | * page. | |
380 | * If we have a lock to protect task->mempolicy in read-side, we do | |
381 | * rebind directly. | |
382 | * | |
383 | * step: | |
384 | * MPOL_REBIND_ONCE - do rebind work at once | |
385 | * MPOL_REBIND_STEP1 - set all the newly nodes | |
386 | * MPOL_REBIND_STEP2 - clean all the disallowed nodes | |
387 | */ | |
388 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask, | |
389 | enum mpol_rebind_step step) | |
1d0d2680 | 390 | { |
1d0d2680 DR |
391 | if (!pol) |
392 | return; | |
708c1bbc | 393 | if (!mpol_store_user_nodemask(pol) && step == 0 && |
1d0d2680 DR |
394 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
395 | return; | |
708c1bbc MX |
396 | |
397 | if (step == MPOL_REBIND_STEP1 && (pol->flags & MPOL_F_REBINDING)) | |
398 | return; | |
399 | ||
400 | if (step == MPOL_REBIND_STEP2 && !(pol->flags & MPOL_F_REBINDING)) | |
401 | BUG(); | |
402 | ||
403 | if (step == MPOL_REBIND_STEP1) | |
404 | pol->flags |= MPOL_F_REBINDING; | |
405 | else if (step == MPOL_REBIND_STEP2) | |
406 | pol->flags &= ~MPOL_F_REBINDING; | |
407 | else if (step >= MPOL_REBIND_NSTEP) | |
408 | BUG(); | |
409 | ||
410 | mpol_ops[pol->mode].rebind(pol, newmask, step); | |
1d0d2680 DR |
411 | } |
412 | ||
413 | /* | |
414 | * Wrapper for mpol_rebind_policy() that just requires task | |
415 | * pointer, and updates task mempolicy. | |
58568d2a MX |
416 | * |
417 | * Called with task's alloc_lock held. | |
1d0d2680 DR |
418 | */ |
419 | ||
708c1bbc MX |
420 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new, |
421 | enum mpol_rebind_step step) | |
1d0d2680 | 422 | { |
708c1bbc | 423 | mpol_rebind_policy(tsk->mempolicy, new, step); |
1d0d2680 DR |
424 | } |
425 | ||
426 | /* | |
427 | * Rebind each vma in mm to new nodemask. | |
428 | * | |
429 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
430 | */ | |
431 | ||
432 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
433 | { | |
434 | struct vm_area_struct *vma; | |
435 | ||
436 | down_write(&mm->mmap_sem); | |
437 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
708c1bbc | 438 | mpol_rebind_policy(vma->vm_policy, new, MPOL_REBIND_ONCE); |
1d0d2680 DR |
439 | up_write(&mm->mmap_sem); |
440 | } | |
441 | ||
37012946 DR |
442 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
443 | [MPOL_DEFAULT] = { | |
444 | .rebind = mpol_rebind_default, | |
445 | }, | |
446 | [MPOL_INTERLEAVE] = { | |
447 | .create = mpol_new_interleave, | |
448 | .rebind = mpol_rebind_nodemask, | |
449 | }, | |
450 | [MPOL_PREFERRED] = { | |
451 | .create = mpol_new_preferred, | |
452 | .rebind = mpol_rebind_preferred, | |
453 | }, | |
454 | [MPOL_BIND] = { | |
455 | .create = mpol_new_bind, | |
456 | .rebind = mpol_rebind_nodemask, | |
457 | }, | |
458 | }; | |
459 | ||
fc301289 CL |
460 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
461 | unsigned long flags); | |
1a75a6c8 | 462 | |
38e35860 | 463 | /* Scan through pages checking if pages follow certain conditions. */ |
b5810039 | 464 | static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
dc9aa5b9 CL |
465 | unsigned long addr, unsigned long end, |
466 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 467 | void *private) |
1da177e4 | 468 | { |
91612e0d HD |
469 | pte_t *orig_pte; |
470 | pte_t *pte; | |
705e87c0 | 471 | spinlock_t *ptl; |
941150a3 | 472 | |
705e87c0 | 473 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
91612e0d | 474 | do { |
6aab341e | 475 | struct page *page; |
25ba77c1 | 476 | int nid; |
91612e0d HD |
477 | |
478 | if (!pte_present(*pte)) | |
1da177e4 | 479 | continue; |
6aab341e LT |
480 | page = vm_normal_page(vma, addr, *pte); |
481 | if (!page) | |
1da177e4 | 482 | continue; |
053837fc | 483 | /* |
62b61f61 HD |
484 | * vm_normal_page() filters out zero pages, but there might |
485 | * still be PageReserved pages to skip, perhaps in a VDSO. | |
486 | * And we cannot move PageKsm pages sensibly or safely yet. | |
053837fc | 487 | */ |
62b61f61 | 488 | if (PageReserved(page) || PageKsm(page)) |
f4598c8b | 489 | continue; |
6aab341e | 490 | nid = page_to_nid(page); |
38e35860 CL |
491 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
492 | continue; | |
493 | ||
b1f72d18 | 494 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
fc301289 | 495 | migrate_page_add(page, private, flags); |
38e35860 CL |
496 | else |
497 | break; | |
91612e0d | 498 | } while (pte++, addr += PAGE_SIZE, addr != end); |
705e87c0 | 499 | pte_unmap_unlock(orig_pte, ptl); |
91612e0d HD |
500 | return addr != end; |
501 | } | |
502 | ||
b5810039 | 503 | static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
dc9aa5b9 CL |
504 | unsigned long addr, unsigned long end, |
505 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 506 | void *private) |
91612e0d HD |
507 | { |
508 | pmd_t *pmd; | |
509 | unsigned long next; | |
510 | ||
511 | pmd = pmd_offset(pud, addr); | |
512 | do { | |
513 | next = pmd_addr_end(addr, end); | |
bae9c19b | 514 | split_huge_page_pmd(vma->vm_mm, pmd); |
91612e0d HD |
515 | if (pmd_none_or_clear_bad(pmd)) |
516 | continue; | |
dc9aa5b9 | 517 | if (check_pte_range(vma, pmd, addr, next, nodes, |
38e35860 | 518 | flags, private)) |
91612e0d HD |
519 | return -EIO; |
520 | } while (pmd++, addr = next, addr != end); | |
521 | return 0; | |
522 | } | |
523 | ||
b5810039 | 524 | static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
dc9aa5b9 CL |
525 | unsigned long addr, unsigned long end, |
526 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 527 | void *private) |
91612e0d HD |
528 | { |
529 | pud_t *pud; | |
530 | unsigned long next; | |
531 | ||
532 | pud = pud_offset(pgd, addr); | |
533 | do { | |
534 | next = pud_addr_end(addr, end); | |
535 | if (pud_none_or_clear_bad(pud)) | |
536 | continue; | |
dc9aa5b9 | 537 | if (check_pmd_range(vma, pud, addr, next, nodes, |
38e35860 | 538 | flags, private)) |
91612e0d HD |
539 | return -EIO; |
540 | } while (pud++, addr = next, addr != end); | |
541 | return 0; | |
542 | } | |
543 | ||
b5810039 | 544 | static inline int check_pgd_range(struct vm_area_struct *vma, |
dc9aa5b9 CL |
545 | unsigned long addr, unsigned long end, |
546 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 547 | void *private) |
91612e0d HD |
548 | { |
549 | pgd_t *pgd; | |
550 | unsigned long next; | |
551 | ||
b5810039 | 552 | pgd = pgd_offset(vma->vm_mm, addr); |
91612e0d HD |
553 | do { |
554 | next = pgd_addr_end(addr, end); | |
555 | if (pgd_none_or_clear_bad(pgd)) | |
556 | continue; | |
dc9aa5b9 | 557 | if (check_pud_range(vma, pgd, addr, next, nodes, |
38e35860 | 558 | flags, private)) |
91612e0d HD |
559 | return -EIO; |
560 | } while (pgd++, addr = next, addr != end); | |
561 | return 0; | |
1da177e4 LT |
562 | } |
563 | ||
dc9aa5b9 CL |
564 | /* |
565 | * Check if all pages in a range are on a set of nodes. | |
566 | * If pagelist != NULL then isolate pages from the LRU and | |
567 | * put them on the pagelist. | |
568 | */ | |
1da177e4 LT |
569 | static struct vm_area_struct * |
570 | check_range(struct mm_struct *mm, unsigned long start, unsigned long end, | |
38e35860 | 571 | const nodemask_t *nodes, unsigned long flags, void *private) |
1da177e4 LT |
572 | { |
573 | int err; | |
574 | struct vm_area_struct *first, *vma, *prev; | |
575 | ||
053837fc | 576 | |
1da177e4 LT |
577 | first = find_vma(mm, start); |
578 | if (!first) | |
579 | return ERR_PTR(-EFAULT); | |
580 | prev = NULL; | |
581 | for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) { | |
dc9aa5b9 CL |
582 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
583 | if (!vma->vm_next && vma->vm_end < end) | |
584 | return ERR_PTR(-EFAULT); | |
585 | if (prev && prev->vm_end < vma->vm_start) | |
586 | return ERR_PTR(-EFAULT); | |
587 | } | |
588 | if (!is_vm_hugetlb_page(vma) && | |
589 | ((flags & MPOL_MF_STRICT) || | |
590 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && | |
591 | vma_migratable(vma)))) { | |
5b952b3c | 592 | unsigned long endvma = vma->vm_end; |
dc9aa5b9 | 593 | |
5b952b3c AK |
594 | if (endvma > end) |
595 | endvma = end; | |
596 | if (vma->vm_start > start) | |
597 | start = vma->vm_start; | |
dc9aa5b9 | 598 | err = check_pgd_range(vma, start, endvma, nodes, |
38e35860 | 599 | flags, private); |
1da177e4 LT |
600 | if (err) { |
601 | first = ERR_PTR(err); | |
602 | break; | |
603 | } | |
604 | } | |
605 | prev = vma; | |
606 | } | |
607 | return first; | |
608 | } | |
609 | ||
610 | /* Apply policy to a single VMA */ | |
611 | static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new) | |
612 | { | |
613 | int err = 0; | |
614 | struct mempolicy *old = vma->vm_policy; | |
615 | ||
140d5a49 | 616 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", |
1da177e4 LT |
617 | vma->vm_start, vma->vm_end, vma->vm_pgoff, |
618 | vma->vm_ops, vma->vm_file, | |
619 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
620 | ||
621 | if (vma->vm_ops && vma->vm_ops->set_policy) | |
622 | err = vma->vm_ops->set_policy(vma, new); | |
623 | if (!err) { | |
624 | mpol_get(new); | |
625 | vma->vm_policy = new; | |
f0be3d32 | 626 | mpol_put(old); |
1da177e4 LT |
627 | } |
628 | return err; | |
629 | } | |
630 | ||
631 | /* Step 2: apply policy to a range and do splits. */ | |
9d8cebd4 KM |
632 | static int mbind_range(struct mm_struct *mm, unsigned long start, |
633 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 LT |
634 | { |
635 | struct vm_area_struct *next; | |
9d8cebd4 KM |
636 | struct vm_area_struct *prev; |
637 | struct vm_area_struct *vma; | |
638 | int err = 0; | |
e26a5114 | 639 | pgoff_t pgoff; |
9d8cebd4 KM |
640 | unsigned long vmstart; |
641 | unsigned long vmend; | |
1da177e4 | 642 | |
9d8cebd4 KM |
643 | vma = find_vma_prev(mm, start, &prev); |
644 | if (!vma || vma->vm_start > start) | |
645 | return -EFAULT; | |
646 | ||
e26a5114 KM |
647 | if (start > vma->vm_start) |
648 | prev = vma; | |
649 | ||
9d8cebd4 | 650 | for (; vma && vma->vm_start < end; prev = vma, vma = next) { |
1da177e4 | 651 | next = vma->vm_next; |
9d8cebd4 KM |
652 | vmstart = max(start, vma->vm_start); |
653 | vmend = min(end, vma->vm_end); | |
654 | ||
e26a5114 KM |
655 | if (mpol_equal(vma_policy(vma), new_pol)) |
656 | continue; | |
657 | ||
658 | pgoff = vma->vm_pgoff + | |
659 | ((vmstart - vma->vm_start) >> PAGE_SHIFT); | |
9d8cebd4 | 660 | prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags, |
e26a5114 | 661 | vma->anon_vma, vma->vm_file, pgoff, |
8aacc9f5 | 662 | new_pol); |
9d8cebd4 KM |
663 | if (prev) { |
664 | vma = prev; | |
665 | next = vma->vm_next; | |
666 | continue; | |
667 | } | |
668 | if (vma->vm_start != vmstart) { | |
669 | err = split_vma(vma->vm_mm, vma, vmstart, 1); | |
670 | if (err) | |
671 | goto out; | |
672 | } | |
673 | if (vma->vm_end != vmend) { | |
674 | err = split_vma(vma->vm_mm, vma, vmend, 0); | |
675 | if (err) | |
676 | goto out; | |
677 | } | |
678 | err = policy_vma(vma, new_pol); | |
1da177e4 | 679 | if (err) |
9d8cebd4 | 680 | goto out; |
1da177e4 | 681 | } |
9d8cebd4 KM |
682 | |
683 | out: | |
1da177e4 LT |
684 | return err; |
685 | } | |
686 | ||
c61afb18 PJ |
687 | /* |
688 | * Update task->flags PF_MEMPOLICY bit: set iff non-default | |
689 | * mempolicy. Allows more rapid checking of this (combined perhaps | |
690 | * with other PF_* flag bits) on memory allocation hot code paths. | |
691 | * | |
692 | * If called from outside this file, the task 'p' should -only- be | |
693 | * a newly forked child not yet visible on the task list, because | |
694 | * manipulating the task flags of a visible task is not safe. | |
695 | * | |
696 | * The above limitation is why this routine has the funny name | |
697 | * mpol_fix_fork_child_flag(). | |
698 | * | |
699 | * It is also safe to call this with a task pointer of current, | |
700 | * which the static wrapper mpol_set_task_struct_flag() does, | |
701 | * for use within this file. | |
702 | */ | |
703 | ||
704 | void mpol_fix_fork_child_flag(struct task_struct *p) | |
705 | { | |
706 | if (p->mempolicy) | |
707 | p->flags |= PF_MEMPOLICY; | |
708 | else | |
709 | p->flags &= ~PF_MEMPOLICY; | |
710 | } | |
711 | ||
712 | static void mpol_set_task_struct_flag(void) | |
713 | { | |
714 | mpol_fix_fork_child_flag(current); | |
715 | } | |
716 | ||
1da177e4 | 717 | /* Set the process memory policy */ |
028fec41 DR |
718 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
719 | nodemask_t *nodes) | |
1da177e4 | 720 | { |
58568d2a | 721 | struct mempolicy *new, *old; |
f4e53d91 | 722 | struct mm_struct *mm = current->mm; |
4bfc4495 | 723 | NODEMASK_SCRATCH(scratch); |
58568d2a | 724 | int ret; |
1da177e4 | 725 | |
4bfc4495 KH |
726 | if (!scratch) |
727 | return -ENOMEM; | |
f4e53d91 | 728 | |
4bfc4495 KH |
729 | new = mpol_new(mode, flags, nodes); |
730 | if (IS_ERR(new)) { | |
731 | ret = PTR_ERR(new); | |
732 | goto out; | |
733 | } | |
f4e53d91 LS |
734 | /* |
735 | * prevent changing our mempolicy while show_numa_maps() | |
736 | * is using it. | |
737 | * Note: do_set_mempolicy() can be called at init time | |
738 | * with no 'mm'. | |
739 | */ | |
740 | if (mm) | |
741 | down_write(&mm->mmap_sem); | |
58568d2a | 742 | task_lock(current); |
4bfc4495 | 743 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a MX |
744 | if (ret) { |
745 | task_unlock(current); | |
746 | if (mm) | |
747 | up_write(&mm->mmap_sem); | |
748 | mpol_put(new); | |
4bfc4495 | 749 | goto out; |
58568d2a MX |
750 | } |
751 | old = current->mempolicy; | |
1da177e4 | 752 | current->mempolicy = new; |
c61afb18 | 753 | mpol_set_task_struct_flag(); |
45c4745a | 754 | if (new && new->mode == MPOL_INTERLEAVE && |
f5b087b5 | 755 | nodes_weight(new->v.nodes)) |
dfcd3c0d | 756 | current->il_next = first_node(new->v.nodes); |
58568d2a | 757 | task_unlock(current); |
f4e53d91 LS |
758 | if (mm) |
759 | up_write(&mm->mmap_sem); | |
760 | ||
58568d2a | 761 | mpol_put(old); |
4bfc4495 KH |
762 | ret = 0; |
763 | out: | |
764 | NODEMASK_SCRATCH_FREE(scratch); | |
765 | return ret; | |
1da177e4 LT |
766 | } |
767 | ||
bea904d5 LS |
768 | /* |
769 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
770 | * |
771 | * Called with task's alloc_lock held | |
bea904d5 LS |
772 | */ |
773 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 774 | { |
dfcd3c0d | 775 | nodes_clear(*nodes); |
bea904d5 LS |
776 | if (p == &default_policy) |
777 | return; | |
778 | ||
45c4745a | 779 | switch (p->mode) { |
19770b32 MG |
780 | case MPOL_BIND: |
781 | /* Fall through */ | |
1da177e4 | 782 | case MPOL_INTERLEAVE: |
dfcd3c0d | 783 | *nodes = p->v.nodes; |
1da177e4 LT |
784 | break; |
785 | case MPOL_PREFERRED: | |
fc36b8d3 | 786 | if (!(p->flags & MPOL_F_LOCAL)) |
dfcd3c0d | 787 | node_set(p->v.preferred_node, *nodes); |
53f2556b | 788 | /* else return empty node mask for local allocation */ |
1da177e4 LT |
789 | break; |
790 | default: | |
791 | BUG(); | |
792 | } | |
793 | } | |
794 | ||
795 | static int lookup_node(struct mm_struct *mm, unsigned long addr) | |
796 | { | |
797 | struct page *p; | |
798 | int err; | |
799 | ||
800 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); | |
801 | if (err >= 0) { | |
802 | err = page_to_nid(p); | |
803 | put_page(p); | |
804 | } | |
805 | return err; | |
806 | } | |
807 | ||
1da177e4 | 808 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
809 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
810 | unsigned long addr, unsigned long flags) | |
1da177e4 | 811 | { |
8bccd85f | 812 | int err; |
1da177e4 LT |
813 | struct mm_struct *mm = current->mm; |
814 | struct vm_area_struct *vma = NULL; | |
815 | struct mempolicy *pol = current->mempolicy; | |
816 | ||
754af6f5 LS |
817 | if (flags & |
818 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 819 | return -EINVAL; |
754af6f5 LS |
820 | |
821 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
822 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
823 | return -EINVAL; | |
824 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 825 | task_lock(current); |
754af6f5 | 826 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 827 | task_unlock(current); |
754af6f5 LS |
828 | return 0; |
829 | } | |
830 | ||
1da177e4 | 831 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
832 | /* |
833 | * Do NOT fall back to task policy if the | |
834 | * vma/shared policy at addr is NULL. We | |
835 | * want to return MPOL_DEFAULT in this case. | |
836 | */ | |
1da177e4 LT |
837 | down_read(&mm->mmap_sem); |
838 | vma = find_vma_intersection(mm, addr, addr+1); | |
839 | if (!vma) { | |
840 | up_read(&mm->mmap_sem); | |
841 | return -EFAULT; | |
842 | } | |
843 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
844 | pol = vma->vm_ops->get_policy(vma, addr); | |
845 | else | |
846 | pol = vma->vm_policy; | |
847 | } else if (addr) | |
848 | return -EINVAL; | |
849 | ||
850 | if (!pol) | |
bea904d5 | 851 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
852 | |
853 | if (flags & MPOL_F_NODE) { | |
854 | if (flags & MPOL_F_ADDR) { | |
855 | err = lookup_node(mm, addr); | |
856 | if (err < 0) | |
857 | goto out; | |
8bccd85f | 858 | *policy = err; |
1da177e4 | 859 | } else if (pol == current->mempolicy && |
45c4745a | 860 | pol->mode == MPOL_INTERLEAVE) { |
8bccd85f | 861 | *policy = current->il_next; |
1da177e4 LT |
862 | } else { |
863 | err = -EINVAL; | |
864 | goto out; | |
865 | } | |
bea904d5 LS |
866 | } else { |
867 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
868 | pol->mode; | |
d79df630 DR |
869 | /* |
870 | * Internal mempolicy flags must be masked off before exposing | |
871 | * the policy to userspace. | |
872 | */ | |
873 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 874 | } |
1da177e4 LT |
875 | |
876 | if (vma) { | |
877 | up_read(¤t->mm->mmap_sem); | |
878 | vma = NULL; | |
879 | } | |
880 | ||
1da177e4 | 881 | err = 0; |
58568d2a | 882 | if (nmask) { |
c6b6ef8b LS |
883 | if (mpol_store_user_nodemask(pol)) { |
884 | *nmask = pol->w.user_nodemask; | |
885 | } else { | |
886 | task_lock(current); | |
887 | get_policy_nodemask(pol, nmask); | |
888 | task_unlock(current); | |
889 | } | |
58568d2a | 890 | } |
1da177e4 LT |
891 | |
892 | out: | |
52cd3b07 | 893 | mpol_cond_put(pol); |
1da177e4 LT |
894 | if (vma) |
895 | up_read(¤t->mm->mmap_sem); | |
896 | return err; | |
897 | } | |
898 | ||
b20a3503 | 899 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 CL |
900 | /* |
901 | * page migration | |
902 | */ | |
fc301289 CL |
903 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
904 | unsigned long flags) | |
6ce3c4c0 CL |
905 | { |
906 | /* | |
fc301289 | 907 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 908 | */ |
62695a84 NP |
909 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) { |
910 | if (!isolate_lru_page(page)) { | |
911 | list_add_tail(&page->lru, pagelist); | |
6d9c285a KM |
912 | inc_zone_page_state(page, NR_ISOLATED_ANON + |
913 | page_is_file_cache(page)); | |
62695a84 NP |
914 | } |
915 | } | |
7e2ab150 | 916 | } |
6ce3c4c0 | 917 | |
742755a1 | 918 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
95a402c3 | 919 | { |
6484eb3e | 920 | return alloc_pages_exact_node(node, GFP_HIGHUSER_MOVABLE, 0); |
95a402c3 CL |
921 | } |
922 | ||
7e2ab150 CL |
923 | /* |
924 | * Migrate pages from one node to a target node. | |
925 | * Returns error or the number of pages not migrated. | |
926 | */ | |
dbcb0f19 AB |
927 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
928 | int flags) | |
7e2ab150 CL |
929 | { |
930 | nodemask_t nmask; | |
931 | LIST_HEAD(pagelist); | |
932 | int err = 0; | |
0def08e3 | 933 | struct vm_area_struct *vma; |
7e2ab150 CL |
934 | |
935 | nodes_clear(nmask); | |
936 | node_set(source, nmask); | |
6ce3c4c0 | 937 | |
0def08e3 | 938 | vma = check_range(mm, mm->mmap->vm_start, mm->task_size, &nmask, |
7e2ab150 | 939 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
0def08e3 VK |
940 | if (IS_ERR(vma)) |
941 | return PTR_ERR(vma); | |
7e2ab150 | 942 | |
cf608ac1 | 943 | if (!list_empty(&pagelist)) { |
7f0f2496 | 944 | err = migrate_pages(&pagelist, new_node_page, dest, |
a6bc32b8 | 945 | false, MIGRATE_SYNC); |
cf608ac1 MK |
946 | if (err) |
947 | putback_lru_pages(&pagelist); | |
948 | } | |
95a402c3 | 949 | |
7e2ab150 | 950 | return err; |
6ce3c4c0 CL |
951 | } |
952 | ||
39743889 | 953 | /* |
7e2ab150 CL |
954 | * Move pages between the two nodesets so as to preserve the physical |
955 | * layout as much as possible. | |
39743889 CL |
956 | * |
957 | * Returns the number of page that could not be moved. | |
958 | */ | |
959 | int do_migrate_pages(struct mm_struct *mm, | |
960 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
961 | { | |
7e2ab150 | 962 | int busy = 0; |
0aedadf9 | 963 | int err; |
7e2ab150 | 964 | nodemask_t tmp; |
39743889 | 965 | |
0aedadf9 CL |
966 | err = migrate_prep(); |
967 | if (err) | |
968 | return err; | |
969 | ||
53f2556b | 970 | down_read(&mm->mmap_sem); |
39743889 | 971 | |
7b2259b3 CL |
972 | err = migrate_vmas(mm, from_nodes, to_nodes, flags); |
973 | if (err) | |
974 | goto out; | |
975 | ||
da0aa138 KM |
976 | /* |
977 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
978 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
979 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
980 | * The pair of nodemasks 'to' and 'from' define the map. | |
981 | * | |
982 | * If no pair of bits is found that way, fallback to picking some | |
983 | * pair of 'source' and 'dest' bits that are not the same. If the | |
984 | * 'source' and 'dest' bits are the same, this represents a node | |
985 | * that will be migrating to itself, so no pages need move. | |
986 | * | |
987 | * If no bits are left in 'tmp', or if all remaining bits left | |
988 | * in 'tmp' correspond to the same bit in 'to', return false | |
989 | * (nothing left to migrate). | |
990 | * | |
991 | * This lets us pick a pair of nodes to migrate between, such that | |
992 | * if possible the dest node is not already occupied by some other | |
993 | * source node, minimizing the risk of overloading the memory on a | |
994 | * node that would happen if we migrated incoming memory to a node | |
995 | * before migrating outgoing memory source that same node. | |
996 | * | |
997 | * A single scan of tmp is sufficient. As we go, we remember the | |
998 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
999 | * that not only moved, but what's better, moved to an empty slot | |
1000 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1001 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1002 | * most recent <s, d> pair that moved. If we get all the way through |
1003 | * the scan of tmp without finding any node that moved, much less | |
1004 | * moved to an empty node, then there is nothing left worth migrating. | |
1005 | */ | |
d4984711 | 1006 | |
7e2ab150 CL |
1007 | tmp = *from_nodes; |
1008 | while (!nodes_empty(tmp)) { | |
1009 | int s,d; | |
1010 | int source = -1; | |
1011 | int dest = 0; | |
1012 | ||
1013 | for_each_node_mask(s, tmp) { | |
1014 | d = node_remap(s, *from_nodes, *to_nodes); | |
1015 | if (s == d) | |
1016 | continue; | |
1017 | ||
1018 | source = s; /* Node moved. Memorize */ | |
1019 | dest = d; | |
1020 | ||
1021 | /* dest not in remaining from nodes? */ | |
1022 | if (!node_isset(dest, tmp)) | |
1023 | break; | |
1024 | } | |
1025 | if (source == -1) | |
1026 | break; | |
1027 | ||
1028 | node_clear(source, tmp); | |
1029 | err = migrate_to_node(mm, source, dest, flags); | |
1030 | if (err > 0) | |
1031 | busy += err; | |
1032 | if (err < 0) | |
1033 | break; | |
39743889 | 1034 | } |
7b2259b3 | 1035 | out: |
39743889 | 1036 | up_read(&mm->mmap_sem); |
7e2ab150 CL |
1037 | if (err < 0) |
1038 | return err; | |
1039 | return busy; | |
b20a3503 CL |
1040 | |
1041 | } | |
1042 | ||
3ad33b24 LS |
1043 | /* |
1044 | * Allocate a new page for page migration based on vma policy. | |
1045 | * Start assuming that page is mapped by vma pointed to by @private. | |
1046 | * Search forward from there, if not. N.B., this assumes that the | |
1047 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1048 | * is in virtual address order. | |
1049 | */ | |
742755a1 | 1050 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
1051 | { |
1052 | struct vm_area_struct *vma = (struct vm_area_struct *)private; | |
3ad33b24 | 1053 | unsigned long uninitialized_var(address); |
95a402c3 | 1054 | |
3ad33b24 LS |
1055 | while (vma) { |
1056 | address = page_address_in_vma(page, vma); | |
1057 | if (address != -EFAULT) | |
1058 | break; | |
1059 | vma = vma->vm_next; | |
1060 | } | |
1061 | ||
1062 | /* | |
1063 | * if !vma, alloc_page_vma() will use task or system default policy | |
1064 | */ | |
1065 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
95a402c3 | 1066 | } |
b20a3503 CL |
1067 | #else |
1068 | ||
1069 | static void migrate_page_add(struct page *page, struct list_head *pagelist, | |
1070 | unsigned long flags) | |
1071 | { | |
39743889 CL |
1072 | } |
1073 | ||
b20a3503 CL |
1074 | int do_migrate_pages(struct mm_struct *mm, |
1075 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
1076 | { | |
1077 | return -ENOSYS; | |
1078 | } | |
95a402c3 | 1079 | |
69939749 | 1080 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
1081 | { |
1082 | return NULL; | |
1083 | } | |
b20a3503 CL |
1084 | #endif |
1085 | ||
dbcb0f19 | 1086 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1087 | unsigned short mode, unsigned short mode_flags, |
1088 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 CL |
1089 | { |
1090 | struct vm_area_struct *vma; | |
1091 | struct mm_struct *mm = current->mm; | |
1092 | struct mempolicy *new; | |
1093 | unsigned long end; | |
1094 | int err; | |
1095 | LIST_HEAD(pagelist); | |
1096 | ||
a3b51e01 DR |
1097 | if (flags & ~(unsigned long)(MPOL_MF_STRICT | |
1098 | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) | |
6ce3c4c0 | 1099 | return -EINVAL; |
74c00241 | 1100 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1101 | return -EPERM; |
1102 | ||
1103 | if (start & ~PAGE_MASK) | |
1104 | return -EINVAL; | |
1105 | ||
1106 | if (mode == MPOL_DEFAULT) | |
1107 | flags &= ~MPOL_MF_STRICT; | |
1108 | ||
1109 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
1110 | end = start + len; | |
1111 | ||
1112 | if (end < start) | |
1113 | return -EINVAL; | |
1114 | if (end == start) | |
1115 | return 0; | |
1116 | ||
028fec41 | 1117 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1118 | if (IS_ERR(new)) |
1119 | return PTR_ERR(new); | |
1120 | ||
1121 | /* | |
1122 | * If we are using the default policy then operation | |
1123 | * on discontinuous address spaces is okay after all | |
1124 | */ | |
1125 | if (!new) | |
1126 | flags |= MPOL_MF_DISCONTIG_OK; | |
1127 | ||
028fec41 DR |
1128 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1129 | start, start + len, mode, mode_flags, | |
1130 | nmask ? nodes_addr(*nmask)[0] : -1); | |
6ce3c4c0 | 1131 | |
0aedadf9 CL |
1132 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1133 | ||
1134 | err = migrate_prep(); | |
1135 | if (err) | |
b05ca738 | 1136 | goto mpol_out; |
0aedadf9 | 1137 | } |
4bfc4495 KH |
1138 | { |
1139 | NODEMASK_SCRATCH(scratch); | |
1140 | if (scratch) { | |
1141 | down_write(&mm->mmap_sem); | |
1142 | task_lock(current); | |
1143 | err = mpol_set_nodemask(new, nmask, scratch); | |
1144 | task_unlock(current); | |
1145 | if (err) | |
1146 | up_write(&mm->mmap_sem); | |
1147 | } else | |
1148 | err = -ENOMEM; | |
1149 | NODEMASK_SCRATCH_FREE(scratch); | |
1150 | } | |
b05ca738 KM |
1151 | if (err) |
1152 | goto mpol_out; | |
1153 | ||
6ce3c4c0 CL |
1154 | vma = check_range(mm, start, end, nmask, |
1155 | flags | MPOL_MF_INVERT, &pagelist); | |
1156 | ||
1157 | err = PTR_ERR(vma); | |
1158 | if (!IS_ERR(vma)) { | |
1159 | int nr_failed = 0; | |
1160 | ||
9d8cebd4 | 1161 | err = mbind_range(mm, start, end, new); |
7e2ab150 | 1162 | |
cf608ac1 | 1163 | if (!list_empty(&pagelist)) { |
95a402c3 | 1164 | nr_failed = migrate_pages(&pagelist, new_vma_page, |
7f0f2496 MG |
1165 | (unsigned long)vma, |
1166 | false, true); | |
cf608ac1 MK |
1167 | if (nr_failed) |
1168 | putback_lru_pages(&pagelist); | |
1169 | } | |
6ce3c4c0 CL |
1170 | |
1171 | if (!err && nr_failed && (flags & MPOL_MF_STRICT)) | |
1172 | err = -EIO; | |
ab8a3e14 KM |
1173 | } else |
1174 | putback_lru_pages(&pagelist); | |
b20a3503 | 1175 | |
6ce3c4c0 | 1176 | up_write(&mm->mmap_sem); |
b05ca738 | 1177 | mpol_out: |
f0be3d32 | 1178 | mpol_put(new); |
6ce3c4c0 CL |
1179 | return err; |
1180 | } | |
1181 | ||
8bccd85f CL |
1182 | /* |
1183 | * User space interface with variable sized bitmaps for nodelists. | |
1184 | */ | |
1185 | ||
1186 | /* Copy a node mask from user space. */ | |
39743889 | 1187 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1188 | unsigned long maxnode) |
1189 | { | |
1190 | unsigned long k; | |
1191 | unsigned long nlongs; | |
1192 | unsigned long endmask; | |
1193 | ||
1194 | --maxnode; | |
1195 | nodes_clear(*nodes); | |
1196 | if (maxnode == 0 || !nmask) | |
1197 | return 0; | |
a9c930ba | 1198 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1199 | return -EINVAL; |
8bccd85f CL |
1200 | |
1201 | nlongs = BITS_TO_LONGS(maxnode); | |
1202 | if ((maxnode % BITS_PER_LONG) == 0) | |
1203 | endmask = ~0UL; | |
1204 | else | |
1205 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1206 | ||
1207 | /* When the user specified more nodes than supported just check | |
1208 | if the non supported part is all zero. */ | |
1209 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
1210 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
1211 | return -EINVAL; | |
1212 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
1213 | unsigned long t; | |
1214 | if (get_user(t, nmask + k)) | |
1215 | return -EFAULT; | |
1216 | if (k == nlongs - 1) { | |
1217 | if (t & endmask) | |
1218 | return -EINVAL; | |
1219 | } else if (t) | |
1220 | return -EINVAL; | |
1221 | } | |
1222 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1223 | endmask = ~0UL; | |
1224 | } | |
1225 | ||
1226 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
1227 | return -EFAULT; | |
1228 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1229 | return 0; | |
1230 | } | |
1231 | ||
1232 | /* Copy a kernel node mask to user space */ | |
1233 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1234 | nodemask_t *nodes) | |
1235 | { | |
1236 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
1237 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
1238 | ||
1239 | if (copy > nbytes) { | |
1240 | if (copy > PAGE_SIZE) | |
1241 | return -EINVAL; | |
1242 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1243 | return -EFAULT; | |
1244 | copy = nbytes; | |
1245 | } | |
1246 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1247 | } | |
1248 | ||
938bb9f5 HC |
1249 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
1250 | unsigned long, mode, unsigned long __user *, nmask, | |
1251 | unsigned long, maxnode, unsigned, flags) | |
8bccd85f CL |
1252 | { |
1253 | nodemask_t nodes; | |
1254 | int err; | |
028fec41 | 1255 | unsigned short mode_flags; |
8bccd85f | 1256 | |
028fec41 DR |
1257 | mode_flags = mode & MPOL_MODE_FLAGS; |
1258 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
1259 | if (mode >= MPOL_MAX) |
1260 | return -EINVAL; | |
4c50bc01 DR |
1261 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
1262 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
1263 | return -EINVAL; | |
8bccd85f CL |
1264 | err = get_nodes(&nodes, nmask, maxnode); |
1265 | if (err) | |
1266 | return err; | |
028fec41 | 1267 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
1268 | } |
1269 | ||
1270 | /* Set the process memory policy */ | |
938bb9f5 HC |
1271 | SYSCALL_DEFINE3(set_mempolicy, int, mode, unsigned long __user *, nmask, |
1272 | unsigned long, maxnode) | |
8bccd85f CL |
1273 | { |
1274 | int err; | |
1275 | nodemask_t nodes; | |
028fec41 | 1276 | unsigned short flags; |
8bccd85f | 1277 | |
028fec41 DR |
1278 | flags = mode & MPOL_MODE_FLAGS; |
1279 | mode &= ~MPOL_MODE_FLAGS; | |
1280 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 1281 | return -EINVAL; |
4c50bc01 DR |
1282 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
1283 | return -EINVAL; | |
8bccd85f CL |
1284 | err = get_nodes(&nodes, nmask, maxnode); |
1285 | if (err) | |
1286 | return err; | |
028fec41 | 1287 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1288 | } |
1289 | ||
938bb9f5 HC |
1290 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1291 | const unsigned long __user *, old_nodes, | |
1292 | const unsigned long __user *, new_nodes) | |
39743889 | 1293 | { |
c69e8d9c | 1294 | const struct cred *cred = current_cred(), *tcred; |
596d7cfa | 1295 | struct mm_struct *mm = NULL; |
39743889 | 1296 | struct task_struct *task; |
39743889 CL |
1297 | nodemask_t task_nodes; |
1298 | int err; | |
596d7cfa KM |
1299 | nodemask_t *old; |
1300 | nodemask_t *new; | |
1301 | NODEMASK_SCRATCH(scratch); | |
1302 | ||
1303 | if (!scratch) | |
1304 | return -ENOMEM; | |
39743889 | 1305 | |
596d7cfa KM |
1306 | old = &scratch->mask1; |
1307 | new = &scratch->mask2; | |
1308 | ||
1309 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1310 | if (err) |
596d7cfa | 1311 | goto out; |
39743889 | 1312 | |
596d7cfa | 1313 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1314 | if (err) |
596d7cfa | 1315 | goto out; |
39743889 CL |
1316 | |
1317 | /* Find the mm_struct */ | |
55cfaa3c | 1318 | rcu_read_lock(); |
228ebcbe | 1319 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1320 | if (!task) { |
55cfaa3c | 1321 | rcu_read_unlock(); |
596d7cfa KM |
1322 | err = -ESRCH; |
1323 | goto out; | |
39743889 CL |
1324 | } |
1325 | mm = get_task_mm(task); | |
55cfaa3c | 1326 | rcu_read_unlock(); |
39743889 | 1327 | |
596d7cfa | 1328 | err = -EINVAL; |
39743889 | 1329 | if (!mm) |
596d7cfa | 1330 | goto out; |
39743889 CL |
1331 | |
1332 | /* | |
1333 | * Check if this process has the right to modify the specified | |
1334 | * process. The right exists if the process has administrative | |
7f927fcc | 1335 | * capabilities, superuser privileges or the same |
39743889 CL |
1336 | * userid as the target process. |
1337 | */ | |
c69e8d9c DH |
1338 | rcu_read_lock(); |
1339 | tcred = __task_cred(task); | |
b6dff3ec DH |
1340 | if (cred->euid != tcred->suid && cred->euid != tcred->uid && |
1341 | cred->uid != tcred->suid && cred->uid != tcred->uid && | |
74c00241 | 1342 | !capable(CAP_SYS_NICE)) { |
c69e8d9c | 1343 | rcu_read_unlock(); |
39743889 CL |
1344 | err = -EPERM; |
1345 | goto out; | |
1346 | } | |
c69e8d9c | 1347 | rcu_read_unlock(); |
39743889 CL |
1348 | |
1349 | task_nodes = cpuset_mems_allowed(task); | |
1350 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1351 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 CL |
1352 | err = -EPERM; |
1353 | goto out; | |
1354 | } | |
1355 | ||
596d7cfa | 1356 | if (!nodes_subset(*new, node_states[N_HIGH_MEMORY])) { |
3b42d28b CL |
1357 | err = -EINVAL; |
1358 | goto out; | |
1359 | } | |
1360 | ||
86c3a764 DQ |
1361 | err = security_task_movememory(task); |
1362 | if (err) | |
1363 | goto out; | |
1364 | ||
596d7cfa | 1365 | err = do_migrate_pages(mm, old, new, |
74c00241 | 1366 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); |
39743889 | 1367 | out: |
596d7cfa KM |
1368 | if (mm) |
1369 | mmput(mm); | |
1370 | NODEMASK_SCRATCH_FREE(scratch); | |
1371 | ||
39743889 CL |
1372 | return err; |
1373 | } | |
1374 | ||
1375 | ||
8bccd85f | 1376 | /* Retrieve NUMA policy */ |
938bb9f5 HC |
1377 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1378 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1379 | unsigned long, addr, unsigned long, flags) | |
8bccd85f | 1380 | { |
dbcb0f19 AB |
1381 | int err; |
1382 | int uninitialized_var(pval); | |
8bccd85f CL |
1383 | nodemask_t nodes; |
1384 | ||
1385 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
1386 | return -EINVAL; | |
1387 | ||
1388 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1389 | ||
1390 | if (err) | |
1391 | return err; | |
1392 | ||
1393 | if (policy && put_user(pval, policy)) | |
1394 | return -EFAULT; | |
1395 | ||
1396 | if (nmask) | |
1397 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1398 | ||
1399 | return err; | |
1400 | } | |
1401 | ||
1da177e4 LT |
1402 | #ifdef CONFIG_COMPAT |
1403 | ||
1404 | asmlinkage long compat_sys_get_mempolicy(int __user *policy, | |
1405 | compat_ulong_t __user *nmask, | |
1406 | compat_ulong_t maxnode, | |
1407 | compat_ulong_t addr, compat_ulong_t flags) | |
1408 | { | |
1409 | long err; | |
1410 | unsigned long __user *nm = NULL; | |
1411 | unsigned long nr_bits, alloc_size; | |
1412 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1413 | ||
1414 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1415 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1416 | ||
1417 | if (nmask) | |
1418 | nm = compat_alloc_user_space(alloc_size); | |
1419 | ||
1420 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
1421 | ||
1422 | if (!err && nmask) { | |
2bbff6c7 KH |
1423 | unsigned long copy_size; |
1424 | copy_size = min_t(unsigned long, sizeof(bm), alloc_size); | |
1425 | err = copy_from_user(bm, nm, copy_size); | |
1da177e4 LT |
1426 | /* ensure entire bitmap is zeroed */ |
1427 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1428 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1429 | } | |
1430 | ||
1431 | return err; | |
1432 | } | |
1433 | ||
1434 | asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask, | |
1435 | compat_ulong_t maxnode) | |
1436 | { | |
1437 | long err = 0; | |
1438 | unsigned long __user *nm = NULL; | |
1439 | unsigned long nr_bits, alloc_size; | |
1440 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1441 | ||
1442 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1443 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1444 | ||
1445 | if (nmask) { | |
1446 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
1447 | nm = compat_alloc_user_space(alloc_size); | |
1448 | err |= copy_to_user(nm, bm, alloc_size); | |
1449 | } | |
1450 | ||
1451 | if (err) | |
1452 | return -EFAULT; | |
1453 | ||
1454 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
1455 | } | |
1456 | ||
1457 | asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len, | |
1458 | compat_ulong_t mode, compat_ulong_t __user *nmask, | |
1459 | compat_ulong_t maxnode, compat_ulong_t flags) | |
1460 | { | |
1461 | long err = 0; | |
1462 | unsigned long __user *nm = NULL; | |
1463 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1464 | nodemask_t bm; |
1da177e4 LT |
1465 | |
1466 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1467 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1468 | ||
1469 | if (nmask) { | |
dfcd3c0d | 1470 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 1471 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 1472 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
1473 | } |
1474 | ||
1475 | if (err) | |
1476 | return -EFAULT; | |
1477 | ||
1478 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
1479 | } | |
1480 | ||
1481 | #endif | |
1482 | ||
480eccf9 LS |
1483 | /* |
1484 | * get_vma_policy(@task, @vma, @addr) | |
1485 | * @task - task for fallback if vma policy == default | |
1486 | * @vma - virtual memory area whose policy is sought | |
1487 | * @addr - address in @vma for shared policy lookup | |
1488 | * | |
1489 | * Returns effective policy for a VMA at specified address. | |
1490 | * Falls back to @task or system default policy, as necessary. | |
52cd3b07 LS |
1491 | * Current or other task's task mempolicy and non-shared vma policies |
1492 | * are protected by the task's mmap_sem, which must be held for read by | |
1493 | * the caller. | |
1494 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference | |
1495 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1496 | * freeing by another task. It is the caller's responsibility to free the | |
1497 | * extra reference for shared policies. | |
480eccf9 | 1498 | */ |
d98f6cb6 | 1499 | struct mempolicy *get_vma_policy(struct task_struct *task, |
48fce342 | 1500 | struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1501 | { |
6e21c8f1 | 1502 | struct mempolicy *pol = task->mempolicy; |
1da177e4 LT |
1503 | |
1504 | if (vma) { | |
480eccf9 | 1505 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
ae4d8c16 LS |
1506 | struct mempolicy *vpol = vma->vm_ops->get_policy(vma, |
1507 | addr); | |
1508 | if (vpol) | |
1509 | pol = vpol; | |
bea904d5 | 1510 | } else if (vma->vm_policy) |
1da177e4 LT |
1511 | pol = vma->vm_policy; |
1512 | } | |
1513 | if (!pol) | |
1514 | pol = &default_policy; | |
1515 | return pol; | |
1516 | } | |
1517 | ||
52cd3b07 LS |
1518 | /* |
1519 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1520 | * page allocation | |
1521 | */ | |
1522 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) | |
19770b32 MG |
1523 | { |
1524 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1525 | if (unlikely(policy->mode == MPOL_BIND) && |
19770b32 MG |
1526 | gfp_zone(gfp) >= policy_zone && |
1527 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) | |
1528 | return &policy->v.nodes; | |
1529 | ||
1530 | return NULL; | |
1531 | } | |
1532 | ||
52cd3b07 | 1533 | /* Return a zonelist indicated by gfp for node representing a mempolicy */ |
2f5f9486 AK |
1534 | static struct zonelist *policy_zonelist(gfp_t gfp, struct mempolicy *policy, |
1535 | int nd) | |
1da177e4 | 1536 | { |
45c4745a | 1537 | switch (policy->mode) { |
1da177e4 | 1538 | case MPOL_PREFERRED: |
fc36b8d3 LS |
1539 | if (!(policy->flags & MPOL_F_LOCAL)) |
1540 | nd = policy->v.preferred_node; | |
1da177e4 LT |
1541 | break; |
1542 | case MPOL_BIND: | |
19770b32 | 1543 | /* |
52cd3b07 LS |
1544 | * Normally, MPOL_BIND allocations are node-local within the |
1545 | * allowed nodemask. However, if __GFP_THISNODE is set and the | |
6eb27e1f | 1546 | * current node isn't part of the mask, we use the zonelist for |
52cd3b07 | 1547 | * the first node in the mask instead. |
19770b32 | 1548 | */ |
19770b32 MG |
1549 | if (unlikely(gfp & __GFP_THISNODE) && |
1550 | unlikely(!node_isset(nd, policy->v.nodes))) | |
1551 | nd = first_node(policy->v.nodes); | |
1552 | break; | |
1da177e4 | 1553 | default: |
1da177e4 LT |
1554 | BUG(); |
1555 | } | |
0e88460d | 1556 | return node_zonelist(nd, gfp); |
1da177e4 LT |
1557 | } |
1558 | ||
1559 | /* Do dynamic interleaving for a process */ | |
1560 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1561 | { | |
1562 | unsigned nid, next; | |
1563 | struct task_struct *me = current; | |
1564 | ||
1565 | nid = me->il_next; | |
dfcd3c0d | 1566 | next = next_node(nid, policy->v.nodes); |
1da177e4 | 1567 | if (next >= MAX_NUMNODES) |
dfcd3c0d | 1568 | next = first_node(policy->v.nodes); |
f5b087b5 DR |
1569 | if (next < MAX_NUMNODES) |
1570 | me->il_next = next; | |
1da177e4 LT |
1571 | return nid; |
1572 | } | |
1573 | ||
dc85da15 CL |
1574 | /* |
1575 | * Depending on the memory policy provide a node from which to allocate the | |
1576 | * next slab entry. | |
52cd3b07 LS |
1577 | * @policy must be protected by freeing by the caller. If @policy is |
1578 | * the current task's mempolicy, this protection is implicit, as only the | |
1579 | * task can change it's policy. The system default policy requires no | |
1580 | * such protection. | |
dc85da15 CL |
1581 | */ |
1582 | unsigned slab_node(struct mempolicy *policy) | |
1583 | { | |
fc36b8d3 | 1584 | if (!policy || policy->flags & MPOL_F_LOCAL) |
bea904d5 LS |
1585 | return numa_node_id(); |
1586 | ||
1587 | switch (policy->mode) { | |
1588 | case MPOL_PREFERRED: | |
fc36b8d3 LS |
1589 | /* |
1590 | * handled MPOL_F_LOCAL above | |
1591 | */ | |
1592 | return policy->v.preferred_node; | |
765c4507 | 1593 | |
dc85da15 CL |
1594 | case MPOL_INTERLEAVE: |
1595 | return interleave_nodes(policy); | |
1596 | ||
dd1a239f | 1597 | case MPOL_BIND: { |
dc85da15 CL |
1598 | /* |
1599 | * Follow bind policy behavior and start allocation at the | |
1600 | * first node. | |
1601 | */ | |
19770b32 MG |
1602 | struct zonelist *zonelist; |
1603 | struct zone *zone; | |
1604 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); | |
1605 | zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0]; | |
1606 | (void)first_zones_zonelist(zonelist, highest_zoneidx, | |
1607 | &policy->v.nodes, | |
1608 | &zone); | |
800416f7 | 1609 | return zone ? zone->node : numa_node_id(); |
dd1a239f | 1610 | } |
dc85da15 | 1611 | |
dc85da15 | 1612 | default: |
bea904d5 | 1613 | BUG(); |
dc85da15 CL |
1614 | } |
1615 | } | |
1616 | ||
1da177e4 LT |
1617 | /* Do static interleaving for a VMA with known offset. */ |
1618 | static unsigned offset_il_node(struct mempolicy *pol, | |
1619 | struct vm_area_struct *vma, unsigned long off) | |
1620 | { | |
dfcd3c0d | 1621 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1622 | unsigned target; |
1da177e4 LT |
1623 | int c; |
1624 | int nid = -1; | |
1625 | ||
f5b087b5 DR |
1626 | if (!nnodes) |
1627 | return numa_node_id(); | |
1628 | target = (unsigned int)off % nnodes; | |
1da177e4 LT |
1629 | c = 0; |
1630 | do { | |
dfcd3c0d | 1631 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1632 | c++; |
1633 | } while (c <= target); | |
1da177e4 LT |
1634 | return nid; |
1635 | } | |
1636 | ||
5da7ca86 CL |
1637 | /* Determine a node number for interleave */ |
1638 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1639 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1640 | { | |
1641 | if (vma) { | |
1642 | unsigned long off; | |
1643 | ||
3b98b087 NA |
1644 | /* |
1645 | * for small pages, there is no difference between | |
1646 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1647 | * for huge pages, since vm_pgoff is in units of small | |
1648 | * pages, we need to shift off the always 0 bits to get | |
1649 | * a useful offset. | |
1650 | */ | |
1651 | BUG_ON(shift < PAGE_SHIFT); | |
1652 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 CL |
1653 | off += (addr - vma->vm_start) >> shift; |
1654 | return offset_il_node(pol, vma, off); | |
1655 | } else | |
1656 | return interleave_nodes(pol); | |
1657 | } | |
1658 | ||
778d3b0f MH |
1659 | /* |
1660 | * Return the bit number of a random bit set in the nodemask. | |
1661 | * (returns -1 if nodemask is empty) | |
1662 | */ | |
1663 | int node_random(const nodemask_t *maskp) | |
1664 | { | |
1665 | int w, bit = -1; | |
1666 | ||
1667 | w = nodes_weight(*maskp); | |
1668 | if (w) | |
1669 | bit = bitmap_ord_to_pos(maskp->bits, | |
1670 | get_random_int() % w, MAX_NUMNODES); | |
1671 | return bit; | |
1672 | } | |
1673 | ||
00ac59ad | 1674 | #ifdef CONFIG_HUGETLBFS |
480eccf9 LS |
1675 | /* |
1676 | * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) | |
1677 | * @vma = virtual memory area whose policy is sought | |
1678 | * @addr = address in @vma for shared policy lookup and interleave policy | |
1679 | * @gfp_flags = for requested zone | |
19770b32 MG |
1680 | * @mpol = pointer to mempolicy pointer for reference counted mempolicy |
1681 | * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 | 1682 | * |
52cd3b07 LS |
1683 | * Returns a zonelist suitable for a huge page allocation and a pointer |
1684 | * to the struct mempolicy for conditional unref after allocation. | |
1685 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's | |
1686 | * @nodemask for filtering the zonelist. | |
c0ff7453 MX |
1687 | * |
1688 | * Must be protected by get_mems_allowed() | |
480eccf9 | 1689 | */ |
396faf03 | 1690 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
19770b32 MG |
1691 | gfp_t gfp_flags, struct mempolicy **mpol, |
1692 | nodemask_t **nodemask) | |
5da7ca86 | 1693 | { |
480eccf9 | 1694 | struct zonelist *zl; |
5da7ca86 | 1695 | |
52cd3b07 | 1696 | *mpol = get_vma_policy(current, vma, addr); |
19770b32 | 1697 | *nodemask = NULL; /* assume !MPOL_BIND */ |
5da7ca86 | 1698 | |
52cd3b07 LS |
1699 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
1700 | zl = node_zonelist(interleave_nid(*mpol, vma, addr, | |
a5516438 | 1701 | huge_page_shift(hstate_vma(vma))), gfp_flags); |
52cd3b07 | 1702 | } else { |
2f5f9486 | 1703 | zl = policy_zonelist(gfp_flags, *mpol, numa_node_id()); |
52cd3b07 LS |
1704 | if ((*mpol)->mode == MPOL_BIND) |
1705 | *nodemask = &(*mpol)->v.nodes; | |
480eccf9 LS |
1706 | } |
1707 | return zl; | |
5da7ca86 | 1708 | } |
06808b08 LS |
1709 | |
1710 | /* | |
1711 | * init_nodemask_of_mempolicy | |
1712 | * | |
1713 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
1714 | * to indicate default policy. Otherwise, extract the policy nodemask | |
1715 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
1716 | * initialize the argument nodemask to contain the single node for | |
1717 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
1718 | * of non-default mempolicy. | |
1719 | * | |
1720 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
1721 | * because the current task is examining it's own mempolicy and a task's | |
1722 | * mempolicy is only ever changed by the task itself. | |
1723 | * | |
1724 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
1725 | */ | |
1726 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
1727 | { | |
1728 | struct mempolicy *mempolicy; | |
1729 | int nid; | |
1730 | ||
1731 | if (!(mask && current->mempolicy)) | |
1732 | return false; | |
1733 | ||
c0ff7453 | 1734 | task_lock(current); |
06808b08 LS |
1735 | mempolicy = current->mempolicy; |
1736 | switch (mempolicy->mode) { | |
1737 | case MPOL_PREFERRED: | |
1738 | if (mempolicy->flags & MPOL_F_LOCAL) | |
1739 | nid = numa_node_id(); | |
1740 | else | |
1741 | nid = mempolicy->v.preferred_node; | |
1742 | init_nodemask_of_node(mask, nid); | |
1743 | break; | |
1744 | ||
1745 | case MPOL_BIND: | |
1746 | /* Fall through */ | |
1747 | case MPOL_INTERLEAVE: | |
1748 | *mask = mempolicy->v.nodes; | |
1749 | break; | |
1750 | ||
1751 | default: | |
1752 | BUG(); | |
1753 | } | |
c0ff7453 | 1754 | task_unlock(current); |
06808b08 LS |
1755 | |
1756 | return true; | |
1757 | } | |
00ac59ad | 1758 | #endif |
5da7ca86 | 1759 | |
6f48d0eb DR |
1760 | /* |
1761 | * mempolicy_nodemask_intersects | |
1762 | * | |
1763 | * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default | |
1764 | * policy. Otherwise, check for intersection between mask and the policy | |
1765 | * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local' | |
1766 | * policy, always return true since it may allocate elsewhere on fallback. | |
1767 | * | |
1768 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
1769 | */ | |
1770 | bool mempolicy_nodemask_intersects(struct task_struct *tsk, | |
1771 | const nodemask_t *mask) | |
1772 | { | |
1773 | struct mempolicy *mempolicy; | |
1774 | bool ret = true; | |
1775 | ||
1776 | if (!mask) | |
1777 | return ret; | |
1778 | task_lock(tsk); | |
1779 | mempolicy = tsk->mempolicy; | |
1780 | if (!mempolicy) | |
1781 | goto out; | |
1782 | ||
1783 | switch (mempolicy->mode) { | |
1784 | case MPOL_PREFERRED: | |
1785 | /* | |
1786 | * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to | |
1787 | * allocate from, they may fallback to other nodes when oom. | |
1788 | * Thus, it's possible for tsk to have allocated memory from | |
1789 | * nodes in mask. | |
1790 | */ | |
1791 | break; | |
1792 | case MPOL_BIND: | |
1793 | case MPOL_INTERLEAVE: | |
1794 | ret = nodes_intersects(mempolicy->v.nodes, *mask); | |
1795 | break; | |
1796 | default: | |
1797 | BUG(); | |
1798 | } | |
1799 | out: | |
1800 | task_unlock(tsk); | |
1801 | return ret; | |
1802 | } | |
1803 | ||
1da177e4 LT |
1804 | /* Allocate a page in interleaved policy. |
1805 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1806 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1807 | unsigned nid) | |
1da177e4 LT |
1808 | { |
1809 | struct zonelist *zl; | |
1810 | struct page *page; | |
1811 | ||
0e88460d | 1812 | zl = node_zonelist(nid, gfp); |
1da177e4 | 1813 | page = __alloc_pages(gfp, order, zl); |
dd1a239f | 1814 | if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0])) |
ca889e6c | 1815 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); |
1da177e4 LT |
1816 | return page; |
1817 | } | |
1818 | ||
1819 | /** | |
0bbbc0b3 | 1820 | * alloc_pages_vma - Allocate a page for a VMA. |
1da177e4 LT |
1821 | * |
1822 | * @gfp: | |
1823 | * %GFP_USER user allocation. | |
1824 | * %GFP_KERNEL kernel allocations, | |
1825 | * %GFP_HIGHMEM highmem/user allocations, | |
1826 | * %GFP_FS allocation should not call back into a file system. | |
1827 | * %GFP_ATOMIC don't sleep. | |
1828 | * | |
0bbbc0b3 | 1829 | * @order:Order of the GFP allocation. |
1da177e4 LT |
1830 | * @vma: Pointer to VMA or NULL if not available. |
1831 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
1832 | * | |
1833 | * This function allocates a page from the kernel page pool and applies | |
1834 | * a NUMA policy associated with the VMA or the current process. | |
1835 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
1836 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
1837 | * all allocations for pages that will be mapped into | |
1838 | * user space. Returns NULL when no page can be allocated. | |
1839 | * | |
1840 | * Should be called with the mm_sem of the vma hold. | |
1841 | */ | |
1842 | struct page * | |
0bbbc0b3 | 1843 | alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma, |
2f5f9486 | 1844 | unsigned long addr, int node) |
1da177e4 | 1845 | { |
6e21c8f1 | 1846 | struct mempolicy *pol = get_vma_policy(current, vma, addr); |
480eccf9 | 1847 | struct zonelist *zl; |
c0ff7453 | 1848 | struct page *page; |
1da177e4 | 1849 | |
c0ff7453 | 1850 | get_mems_allowed(); |
45c4745a | 1851 | if (unlikely(pol->mode == MPOL_INTERLEAVE)) { |
1da177e4 | 1852 | unsigned nid; |
5da7ca86 | 1853 | |
8eac563c | 1854 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); |
52cd3b07 | 1855 | mpol_cond_put(pol); |
0bbbc0b3 | 1856 | page = alloc_page_interleave(gfp, order, nid); |
c0ff7453 MX |
1857 | put_mems_allowed(); |
1858 | return page; | |
1da177e4 | 1859 | } |
2f5f9486 | 1860 | zl = policy_zonelist(gfp, pol, node); |
52cd3b07 | 1861 | if (unlikely(mpol_needs_cond_ref(pol))) { |
480eccf9 | 1862 | /* |
52cd3b07 | 1863 | * slow path: ref counted shared policy |
480eccf9 | 1864 | */ |
0bbbc0b3 | 1865 | struct page *page = __alloc_pages_nodemask(gfp, order, |
52cd3b07 | 1866 | zl, policy_nodemask(gfp, pol)); |
f0be3d32 | 1867 | __mpol_put(pol); |
c0ff7453 | 1868 | put_mems_allowed(); |
480eccf9 LS |
1869 | return page; |
1870 | } | |
1871 | /* | |
1872 | * fast path: default or task policy | |
1873 | */ | |
0bbbc0b3 AA |
1874 | page = __alloc_pages_nodemask(gfp, order, zl, |
1875 | policy_nodemask(gfp, pol)); | |
c0ff7453 MX |
1876 | put_mems_allowed(); |
1877 | return page; | |
1da177e4 LT |
1878 | } |
1879 | ||
1880 | /** | |
1881 | * alloc_pages_current - Allocate pages. | |
1882 | * | |
1883 | * @gfp: | |
1884 | * %GFP_USER user allocation, | |
1885 | * %GFP_KERNEL kernel allocation, | |
1886 | * %GFP_HIGHMEM highmem allocation, | |
1887 | * %GFP_FS don't call back into a file system. | |
1888 | * %GFP_ATOMIC don't sleep. | |
1889 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
1890 | * | |
1891 | * Allocate a page from the kernel page pool. When not in | |
1892 | * interrupt context and apply the current process NUMA policy. | |
1893 | * Returns NULL when no page can be allocated. | |
1894 | * | |
cf2a473c | 1895 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
1896 | * 1) it's ok to take cpuset_sem (can WAIT), and |
1897 | * 2) allocating for current task (not interrupt). | |
1898 | */ | |
dd0fc66f | 1899 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 LT |
1900 | { |
1901 | struct mempolicy *pol = current->mempolicy; | |
c0ff7453 | 1902 | struct page *page; |
1da177e4 | 1903 | |
9b819d20 | 1904 | if (!pol || in_interrupt() || (gfp & __GFP_THISNODE)) |
1da177e4 | 1905 | pol = &default_policy; |
52cd3b07 | 1906 | |
c0ff7453 | 1907 | get_mems_allowed(); |
52cd3b07 LS |
1908 | /* |
1909 | * No reference counting needed for current->mempolicy | |
1910 | * nor system default_policy | |
1911 | */ | |
45c4745a | 1912 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 MX |
1913 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
1914 | else | |
1915 | page = __alloc_pages_nodemask(gfp, order, | |
5c4b4be3 AK |
1916 | policy_zonelist(gfp, pol, numa_node_id()), |
1917 | policy_nodemask(gfp, pol)); | |
c0ff7453 MX |
1918 | put_mems_allowed(); |
1919 | return page; | |
1da177e4 LT |
1920 | } |
1921 | EXPORT_SYMBOL(alloc_pages_current); | |
1922 | ||
4225399a | 1923 | /* |
846a16bf | 1924 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
1925 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
1926 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
1927 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
1928 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
1929 | * |
1930 | * current's mempolicy may be rebinded by the other task(the task that changes | |
1931 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 1932 | */ |
4225399a | 1933 | |
846a16bf LS |
1934 | /* Slow path of a mempolicy duplicate */ |
1935 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
1936 | { |
1937 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
1938 | ||
1939 | if (!new) | |
1940 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
1941 | |
1942 | /* task's mempolicy is protected by alloc_lock */ | |
1943 | if (old == current->mempolicy) { | |
1944 | task_lock(current); | |
1945 | *new = *old; | |
1946 | task_unlock(current); | |
1947 | } else | |
1948 | *new = *old; | |
1949 | ||
99ee4ca7 | 1950 | rcu_read_lock(); |
4225399a PJ |
1951 | if (current_cpuset_is_being_rebound()) { |
1952 | nodemask_t mems = cpuset_mems_allowed(current); | |
708c1bbc MX |
1953 | if (new->flags & MPOL_F_REBINDING) |
1954 | mpol_rebind_policy(new, &mems, MPOL_REBIND_STEP2); | |
1955 | else | |
1956 | mpol_rebind_policy(new, &mems, MPOL_REBIND_ONCE); | |
4225399a | 1957 | } |
99ee4ca7 | 1958 | rcu_read_unlock(); |
1da177e4 | 1959 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
1960 | return new; |
1961 | } | |
1962 | ||
52cd3b07 LS |
1963 | /* |
1964 | * If *frompol needs [has] an extra ref, copy *frompol to *tompol , | |
1965 | * eliminate the * MPOL_F_* flags that require conditional ref and | |
1966 | * [NOTE!!!] drop the extra ref. Not safe to reference *frompol directly | |
1967 | * after return. Use the returned value. | |
1968 | * | |
1969 | * Allows use of a mempolicy for, e.g., multiple allocations with a single | |
1970 | * policy lookup, even if the policy needs/has extra ref on lookup. | |
1971 | * shmem_readahead needs this. | |
1972 | */ | |
1973 | struct mempolicy *__mpol_cond_copy(struct mempolicy *tompol, | |
1974 | struct mempolicy *frompol) | |
1975 | { | |
1976 | if (!mpol_needs_cond_ref(frompol)) | |
1977 | return frompol; | |
1978 | ||
1979 | *tompol = *frompol; | |
1980 | tompol->flags &= ~MPOL_F_SHARED; /* copy doesn't need unref */ | |
1981 | __mpol_put(frompol); | |
1982 | return tompol; | |
1983 | } | |
1984 | ||
1da177e4 | 1985 | /* Slow path of a mempolicy comparison */ |
fcfb4dcc | 1986 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
1987 | { |
1988 | if (!a || !b) | |
fcfb4dcc | 1989 | return false; |
45c4745a | 1990 | if (a->mode != b->mode) |
fcfb4dcc | 1991 | return false; |
19800502 | 1992 | if (a->flags != b->flags) |
fcfb4dcc | 1993 | return false; |
19800502 BL |
1994 | if (mpol_store_user_nodemask(a)) |
1995 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 1996 | return false; |
19800502 | 1997 | |
45c4745a | 1998 | switch (a->mode) { |
19770b32 MG |
1999 | case MPOL_BIND: |
2000 | /* Fall through */ | |
1da177e4 | 2001 | case MPOL_INTERLEAVE: |
fcfb4dcc | 2002 | return !!nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 | 2003 | case MPOL_PREFERRED: |
75719661 | 2004 | return a->v.preferred_node == b->v.preferred_node; |
1da177e4 LT |
2005 | default: |
2006 | BUG(); | |
fcfb4dcc | 2007 | return false; |
1da177e4 LT |
2008 | } |
2009 | } | |
2010 | ||
1da177e4 LT |
2011 | /* |
2012 | * Shared memory backing store policy support. | |
2013 | * | |
2014 | * Remember policies even when nobody has shared memory mapped. | |
2015 | * The policies are kept in Red-Black tree linked from the inode. | |
2016 | * They are protected by the sp->lock spinlock, which should be held | |
2017 | * for any accesses to the tree. | |
2018 | */ | |
2019 | ||
2020 | /* lookup first element intersecting start-end */ | |
2021 | /* Caller holds sp->lock */ | |
2022 | static struct sp_node * | |
2023 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2024 | { | |
2025 | struct rb_node *n = sp->root.rb_node; | |
2026 | ||
2027 | while (n) { | |
2028 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2029 | ||
2030 | if (start >= p->end) | |
2031 | n = n->rb_right; | |
2032 | else if (end <= p->start) | |
2033 | n = n->rb_left; | |
2034 | else | |
2035 | break; | |
2036 | } | |
2037 | if (!n) | |
2038 | return NULL; | |
2039 | for (;;) { | |
2040 | struct sp_node *w = NULL; | |
2041 | struct rb_node *prev = rb_prev(n); | |
2042 | if (!prev) | |
2043 | break; | |
2044 | w = rb_entry(prev, struct sp_node, nd); | |
2045 | if (w->end <= start) | |
2046 | break; | |
2047 | n = prev; | |
2048 | } | |
2049 | return rb_entry(n, struct sp_node, nd); | |
2050 | } | |
2051 | ||
2052 | /* Insert a new shared policy into the list. */ | |
2053 | /* Caller holds sp->lock */ | |
2054 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) | |
2055 | { | |
2056 | struct rb_node **p = &sp->root.rb_node; | |
2057 | struct rb_node *parent = NULL; | |
2058 | struct sp_node *nd; | |
2059 | ||
2060 | while (*p) { | |
2061 | parent = *p; | |
2062 | nd = rb_entry(parent, struct sp_node, nd); | |
2063 | if (new->start < nd->start) | |
2064 | p = &(*p)->rb_left; | |
2065 | else if (new->end > nd->end) | |
2066 | p = &(*p)->rb_right; | |
2067 | else | |
2068 | BUG(); | |
2069 | } | |
2070 | rb_link_node(&new->nd, parent, p); | |
2071 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2072 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2073 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2074 | } |
2075 | ||
2076 | /* Find shared policy intersecting idx */ | |
2077 | struct mempolicy * | |
2078 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2079 | { | |
2080 | struct mempolicy *pol = NULL; | |
2081 | struct sp_node *sn; | |
2082 | ||
2083 | if (!sp->root.rb_node) | |
2084 | return NULL; | |
2085 | spin_lock(&sp->lock); | |
2086 | sn = sp_lookup(sp, idx, idx+1); | |
2087 | if (sn) { | |
2088 | mpol_get(sn->policy); | |
2089 | pol = sn->policy; | |
2090 | } | |
2091 | spin_unlock(&sp->lock); | |
2092 | return pol; | |
2093 | } | |
2094 | ||
2095 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) | |
2096 | { | |
140d5a49 | 2097 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2098 | rb_erase(&n->nd, &sp->root); |
f0be3d32 | 2099 | mpol_put(n->policy); |
1da177e4 LT |
2100 | kmem_cache_free(sn_cache, n); |
2101 | } | |
2102 | ||
dbcb0f19 AB |
2103 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2104 | struct mempolicy *pol) | |
1da177e4 LT |
2105 | { |
2106 | struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
2107 | ||
2108 | if (!n) | |
2109 | return NULL; | |
2110 | n->start = start; | |
2111 | n->end = end; | |
2112 | mpol_get(pol); | |
aab0b102 | 2113 | pol->flags |= MPOL_F_SHARED; /* for unref */ |
1da177e4 LT |
2114 | n->policy = pol; |
2115 | return n; | |
2116 | } | |
2117 | ||
2118 | /* Replace a policy range. */ | |
2119 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2120 | unsigned long end, struct sp_node *new) | |
2121 | { | |
2122 | struct sp_node *n, *new2 = NULL; | |
2123 | ||
2124 | restart: | |
2125 | spin_lock(&sp->lock); | |
2126 | n = sp_lookup(sp, start, end); | |
2127 | /* Take care of old policies in the same range. */ | |
2128 | while (n && n->start < end) { | |
2129 | struct rb_node *next = rb_next(&n->nd); | |
2130 | if (n->start >= start) { | |
2131 | if (n->end <= end) | |
2132 | sp_delete(sp, n); | |
2133 | else | |
2134 | n->start = end; | |
2135 | } else { | |
2136 | /* Old policy spanning whole new range. */ | |
2137 | if (n->end > end) { | |
2138 | if (!new2) { | |
2139 | spin_unlock(&sp->lock); | |
2140 | new2 = sp_alloc(end, n->end, n->policy); | |
2141 | if (!new2) | |
2142 | return -ENOMEM; | |
2143 | goto restart; | |
2144 | } | |
2145 | n->end = start; | |
2146 | sp_insert(sp, new2); | |
2147 | new2 = NULL; | |
2148 | break; | |
2149 | } else | |
2150 | n->end = start; | |
2151 | } | |
2152 | if (!next) | |
2153 | break; | |
2154 | n = rb_entry(next, struct sp_node, nd); | |
2155 | } | |
2156 | if (new) | |
2157 | sp_insert(sp, new); | |
2158 | spin_unlock(&sp->lock); | |
2159 | if (new2) { | |
f0be3d32 | 2160 | mpol_put(new2->policy); |
1da177e4 LT |
2161 | kmem_cache_free(sn_cache, new2); |
2162 | } | |
2163 | return 0; | |
2164 | } | |
2165 | ||
71fe804b LS |
2166 | /** |
2167 | * mpol_shared_policy_init - initialize shared policy for inode | |
2168 | * @sp: pointer to inode shared policy | |
2169 | * @mpol: struct mempolicy to install | |
2170 | * | |
2171 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2172 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2173 | * This must be released on exit. | |
4bfc4495 | 2174 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2175 | */ |
2176 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2177 | { | |
58568d2a MX |
2178 | int ret; |
2179 | ||
71fe804b LS |
2180 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
2181 | spin_lock_init(&sp->lock); | |
2182 | ||
2183 | if (mpol) { | |
2184 | struct vm_area_struct pvma; | |
2185 | struct mempolicy *new; | |
4bfc4495 | 2186 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2187 | |
4bfc4495 | 2188 | if (!scratch) |
5c0c1654 | 2189 | goto put_mpol; |
71fe804b LS |
2190 | /* contextualize the tmpfs mount point mempolicy */ |
2191 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2192 | if (IS_ERR(new)) |
0cae3457 | 2193 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2194 | |
2195 | task_lock(current); | |
4bfc4495 | 2196 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2197 | task_unlock(current); |
15d77835 | 2198 | if (ret) |
5c0c1654 | 2199 | goto put_new; |
71fe804b LS |
2200 | |
2201 | /* Create pseudo-vma that contains just the policy */ | |
2202 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
2203 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ | |
2204 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2205 | |
5c0c1654 | 2206 | put_new: |
71fe804b | 2207 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2208 | free_scratch: |
4bfc4495 | 2209 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2210 | put_mpol: |
2211 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2212 | } |
2213 | } | |
2214 | ||
1da177e4 LT |
2215 | int mpol_set_shared_policy(struct shared_policy *info, |
2216 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2217 | { | |
2218 | int err; | |
2219 | struct sp_node *new = NULL; | |
2220 | unsigned long sz = vma_pages(vma); | |
2221 | ||
028fec41 | 2222 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2223 | vma->vm_pgoff, |
45c4745a | 2224 | sz, npol ? npol->mode : -1, |
028fec41 | 2225 | npol ? npol->flags : -1, |
140d5a49 | 2226 | npol ? nodes_addr(npol->v.nodes)[0] : -1); |
1da177e4 LT |
2227 | |
2228 | if (npol) { | |
2229 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2230 | if (!new) | |
2231 | return -ENOMEM; | |
2232 | } | |
2233 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2234 | if (err && new) | |
2235 | kmem_cache_free(sn_cache, new); | |
2236 | return err; | |
2237 | } | |
2238 | ||
2239 | /* Free a backing policy store on inode delete. */ | |
2240 | void mpol_free_shared_policy(struct shared_policy *p) | |
2241 | { | |
2242 | struct sp_node *n; | |
2243 | struct rb_node *next; | |
2244 | ||
2245 | if (!p->root.rb_node) | |
2246 | return; | |
2247 | spin_lock(&p->lock); | |
2248 | next = rb_first(&p->root); | |
2249 | while (next) { | |
2250 | n = rb_entry(next, struct sp_node, nd); | |
2251 | next = rb_next(&n->nd); | |
90c5029e | 2252 | rb_erase(&n->nd, &p->root); |
f0be3d32 | 2253 | mpol_put(n->policy); |
1da177e4 LT |
2254 | kmem_cache_free(sn_cache, n); |
2255 | } | |
2256 | spin_unlock(&p->lock); | |
1da177e4 LT |
2257 | } |
2258 | ||
2259 | /* assumes fs == KERNEL_DS */ | |
2260 | void __init numa_policy_init(void) | |
2261 | { | |
b71636e2 PM |
2262 | nodemask_t interleave_nodes; |
2263 | unsigned long largest = 0; | |
2264 | int nid, prefer = 0; | |
2265 | ||
1da177e4 LT |
2266 | policy_cache = kmem_cache_create("numa_policy", |
2267 | sizeof(struct mempolicy), | |
20c2df83 | 2268 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2269 | |
2270 | sn_cache = kmem_cache_create("shared_policy_node", | |
2271 | sizeof(struct sp_node), | |
20c2df83 | 2272 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2273 | |
b71636e2 PM |
2274 | /* |
2275 | * Set interleaving policy for system init. Interleaving is only | |
2276 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2277 | * fall back to the largest node if they're all smaller. | |
2278 | */ | |
2279 | nodes_clear(interleave_nodes); | |
56bbd65d | 2280 | for_each_node_state(nid, N_HIGH_MEMORY) { |
b71636e2 PM |
2281 | unsigned long total_pages = node_present_pages(nid); |
2282 | ||
2283 | /* Preserve the largest node */ | |
2284 | if (largest < total_pages) { | |
2285 | largest = total_pages; | |
2286 | prefer = nid; | |
2287 | } | |
2288 | ||
2289 | /* Interleave this node? */ | |
2290 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2291 | node_set(nid, interleave_nodes); | |
2292 | } | |
2293 | ||
2294 | /* All too small, use the largest */ | |
2295 | if (unlikely(nodes_empty(interleave_nodes))) | |
2296 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2297 | |
028fec41 | 2298 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
1da177e4 LT |
2299 | printk("numa_policy_init: interleaving failed\n"); |
2300 | } | |
2301 | ||
8bccd85f | 2302 | /* Reset policy of current process to default */ |
1da177e4 LT |
2303 | void numa_default_policy(void) |
2304 | { | |
028fec41 | 2305 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2306 | } |
68860ec1 | 2307 | |
095f1fc4 LS |
2308 | /* |
2309 | * Parse and format mempolicy from/to strings | |
2310 | */ | |
2311 | ||
1a75a6c8 | 2312 | /* |
fc36b8d3 | 2313 | * "local" is pseudo-policy: MPOL_PREFERRED with MPOL_F_LOCAL flag |
3f226aa1 | 2314 | * Used only for mpol_parse_str() and mpol_to_str() |
1a75a6c8 | 2315 | */ |
345ace9c LS |
2316 | #define MPOL_LOCAL MPOL_MAX |
2317 | static const char * const policy_modes[] = | |
2318 | { | |
2319 | [MPOL_DEFAULT] = "default", | |
2320 | [MPOL_PREFERRED] = "prefer", | |
2321 | [MPOL_BIND] = "bind", | |
2322 | [MPOL_INTERLEAVE] = "interleave", | |
2323 | [MPOL_LOCAL] = "local" | |
2324 | }; | |
1a75a6c8 | 2325 | |
095f1fc4 LS |
2326 | |
2327 | #ifdef CONFIG_TMPFS | |
2328 | /** | |
2329 | * mpol_parse_str - parse string to mempolicy | |
2330 | * @str: string containing mempolicy to parse | |
71fe804b LS |
2331 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
2332 | * @no_context: flag whether to "contextualize" the mempolicy | |
095f1fc4 LS |
2333 | * |
2334 | * Format of input: | |
2335 | * <mode>[=<flags>][:<nodelist>] | |
2336 | * | |
71fe804b LS |
2337 | * if @no_context is true, save the input nodemask in w.user_nodemask in |
2338 | * the returned mempolicy. This will be used to "clone" the mempolicy in | |
2339 | * a specific context [cpuset] at a later time. Used to parse tmpfs mpol | |
2340 | * mount option. Note that if 'static' or 'relative' mode flags were | |
2341 | * specified, the input nodemask will already have been saved. Saving | |
2342 | * it again is redundant, but safe. | |
2343 | * | |
2344 | * On success, returns 0, else 1 | |
095f1fc4 | 2345 | */ |
71fe804b | 2346 | int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context) |
095f1fc4 | 2347 | { |
71fe804b | 2348 | struct mempolicy *new = NULL; |
b4652e84 | 2349 | unsigned short mode; |
71fe804b LS |
2350 | unsigned short uninitialized_var(mode_flags); |
2351 | nodemask_t nodes; | |
095f1fc4 LS |
2352 | char *nodelist = strchr(str, ':'); |
2353 | char *flags = strchr(str, '='); | |
095f1fc4 LS |
2354 | int err = 1; |
2355 | ||
2356 | if (nodelist) { | |
2357 | /* NUL-terminate mode or flags string */ | |
2358 | *nodelist++ = '\0'; | |
71fe804b | 2359 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2360 | goto out; |
71fe804b | 2361 | if (!nodes_subset(nodes, node_states[N_HIGH_MEMORY])) |
095f1fc4 | 2362 | goto out; |
71fe804b LS |
2363 | } else |
2364 | nodes_clear(nodes); | |
2365 | ||
095f1fc4 LS |
2366 | if (flags) |
2367 | *flags++ = '\0'; /* terminate mode string */ | |
2368 | ||
b4652e84 | 2369 | for (mode = 0; mode <= MPOL_LOCAL; mode++) { |
345ace9c | 2370 | if (!strcmp(str, policy_modes[mode])) { |
095f1fc4 LS |
2371 | break; |
2372 | } | |
2373 | } | |
b4652e84 | 2374 | if (mode > MPOL_LOCAL) |
095f1fc4 LS |
2375 | goto out; |
2376 | ||
71fe804b | 2377 | switch (mode) { |
095f1fc4 | 2378 | case MPOL_PREFERRED: |
71fe804b LS |
2379 | /* |
2380 | * Insist on a nodelist of one node only | |
2381 | */ | |
095f1fc4 LS |
2382 | if (nodelist) { |
2383 | char *rest = nodelist; | |
2384 | while (isdigit(*rest)) | |
2385 | rest++; | |
926f2ae0 KM |
2386 | if (*rest) |
2387 | goto out; | |
095f1fc4 LS |
2388 | } |
2389 | break; | |
095f1fc4 LS |
2390 | case MPOL_INTERLEAVE: |
2391 | /* | |
2392 | * Default to online nodes with memory if no nodelist | |
2393 | */ | |
2394 | if (!nodelist) | |
71fe804b | 2395 | nodes = node_states[N_HIGH_MEMORY]; |
3f226aa1 | 2396 | break; |
71fe804b | 2397 | case MPOL_LOCAL: |
3f226aa1 | 2398 | /* |
71fe804b | 2399 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 2400 | */ |
71fe804b | 2401 | if (nodelist) |
3f226aa1 | 2402 | goto out; |
71fe804b | 2403 | mode = MPOL_PREFERRED; |
3f226aa1 | 2404 | break; |
413b43de RT |
2405 | case MPOL_DEFAULT: |
2406 | /* | |
2407 | * Insist on a empty nodelist | |
2408 | */ | |
2409 | if (!nodelist) | |
2410 | err = 0; | |
2411 | goto out; | |
d69b2e63 KM |
2412 | case MPOL_BIND: |
2413 | /* | |
2414 | * Insist on a nodelist | |
2415 | */ | |
2416 | if (!nodelist) | |
2417 | goto out; | |
095f1fc4 LS |
2418 | } |
2419 | ||
71fe804b | 2420 | mode_flags = 0; |
095f1fc4 LS |
2421 | if (flags) { |
2422 | /* | |
2423 | * Currently, we only support two mutually exclusive | |
2424 | * mode flags. | |
2425 | */ | |
2426 | if (!strcmp(flags, "static")) | |
71fe804b | 2427 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 2428 | else if (!strcmp(flags, "relative")) |
71fe804b | 2429 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 2430 | else |
926f2ae0 | 2431 | goto out; |
095f1fc4 | 2432 | } |
71fe804b LS |
2433 | |
2434 | new = mpol_new(mode, mode_flags, &nodes); | |
2435 | if (IS_ERR(new)) | |
926f2ae0 KM |
2436 | goto out; |
2437 | ||
e17f74af LS |
2438 | if (no_context) { |
2439 | /* save for contextualization */ | |
2440 | new->w.user_nodemask = nodes; | |
2441 | } else { | |
58568d2a | 2442 | int ret; |
4bfc4495 KH |
2443 | NODEMASK_SCRATCH(scratch); |
2444 | if (scratch) { | |
2445 | task_lock(current); | |
2446 | ret = mpol_set_nodemask(new, &nodes, scratch); | |
2447 | task_unlock(current); | |
2448 | } else | |
2449 | ret = -ENOMEM; | |
2450 | NODEMASK_SCRATCH_FREE(scratch); | |
2451 | if (ret) { | |
4bfc4495 | 2452 | mpol_put(new); |
926f2ae0 | 2453 | goto out; |
58568d2a MX |
2454 | } |
2455 | } | |
926f2ae0 | 2456 | err = 0; |
71fe804b | 2457 | |
095f1fc4 LS |
2458 | out: |
2459 | /* Restore string for error message */ | |
2460 | if (nodelist) | |
2461 | *--nodelist = ':'; | |
2462 | if (flags) | |
2463 | *--flags = '='; | |
71fe804b LS |
2464 | if (!err) |
2465 | *mpol = new; | |
095f1fc4 LS |
2466 | return err; |
2467 | } | |
2468 | #endif /* CONFIG_TMPFS */ | |
2469 | ||
71fe804b LS |
2470 | /** |
2471 | * mpol_to_str - format a mempolicy structure for printing | |
2472 | * @buffer: to contain formatted mempolicy string | |
2473 | * @maxlen: length of @buffer | |
2474 | * @pol: pointer to mempolicy to be formatted | |
2475 | * @no_context: "context free" mempolicy - use nodemask in w.user_nodemask | |
2476 | * | |
1a75a6c8 CL |
2477 | * Convert a mempolicy into a string. |
2478 | * Returns the number of characters in buffer (if positive) | |
2479 | * or an error (negative) | |
2480 | */ | |
71fe804b | 2481 | int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol, int no_context) |
1a75a6c8 CL |
2482 | { |
2483 | char *p = buffer; | |
2484 | int l; | |
2485 | nodemask_t nodes; | |
bea904d5 | 2486 | unsigned short mode; |
f5b087b5 | 2487 | unsigned short flags = pol ? pol->flags : 0; |
1a75a6c8 | 2488 | |
2291990a LS |
2489 | /* |
2490 | * Sanity check: room for longest mode, flag and some nodes | |
2491 | */ | |
2492 | VM_BUG_ON(maxlen < strlen("interleave") + strlen("relative") + 16); | |
2493 | ||
bea904d5 LS |
2494 | if (!pol || pol == &default_policy) |
2495 | mode = MPOL_DEFAULT; | |
2496 | else | |
2497 | mode = pol->mode; | |
2498 | ||
1a75a6c8 CL |
2499 | switch (mode) { |
2500 | case MPOL_DEFAULT: | |
2501 | nodes_clear(nodes); | |
2502 | break; | |
2503 | ||
2504 | case MPOL_PREFERRED: | |
2505 | nodes_clear(nodes); | |
fc36b8d3 | 2506 | if (flags & MPOL_F_LOCAL) |
53f2556b LS |
2507 | mode = MPOL_LOCAL; /* pseudo-policy */ |
2508 | else | |
fc36b8d3 | 2509 | node_set(pol->v.preferred_node, nodes); |
1a75a6c8 CL |
2510 | break; |
2511 | ||
2512 | case MPOL_BIND: | |
19770b32 | 2513 | /* Fall through */ |
1a75a6c8 | 2514 | case MPOL_INTERLEAVE: |
71fe804b LS |
2515 | if (no_context) |
2516 | nodes = pol->w.user_nodemask; | |
2517 | else | |
2518 | nodes = pol->v.nodes; | |
1a75a6c8 CL |
2519 | break; |
2520 | ||
2521 | default: | |
2522 | BUG(); | |
1a75a6c8 CL |
2523 | } |
2524 | ||
345ace9c | 2525 | l = strlen(policy_modes[mode]); |
53f2556b LS |
2526 | if (buffer + maxlen < p + l + 1) |
2527 | return -ENOSPC; | |
1a75a6c8 | 2528 | |
345ace9c | 2529 | strcpy(p, policy_modes[mode]); |
1a75a6c8 CL |
2530 | p += l; |
2531 | ||
fc36b8d3 | 2532 | if (flags & MPOL_MODE_FLAGS) { |
f5b087b5 DR |
2533 | if (buffer + maxlen < p + 2) |
2534 | return -ENOSPC; | |
2535 | *p++ = '='; | |
2536 | ||
2291990a LS |
2537 | /* |
2538 | * Currently, the only defined flags are mutually exclusive | |
2539 | */ | |
f5b087b5 | 2540 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
2541 | p += snprintf(p, buffer + maxlen - p, "static"); |
2542 | else if (flags & MPOL_F_RELATIVE_NODES) | |
2543 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
2544 | } |
2545 | ||
1a75a6c8 CL |
2546 | if (!nodes_empty(nodes)) { |
2547 | if (buffer + maxlen < p + 2) | |
2548 | return -ENOSPC; | |
095f1fc4 | 2549 | *p++ = ':'; |
1a75a6c8 CL |
2550 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); |
2551 | } | |
2552 | return p - buffer; | |
2553 | } |