Commit | Line | Data |
---|---|---|
46aeb7e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * Simple NUMA memory policy for the Linux kernel. | |
4 | * | |
5 | * Copyright 2003,2004 Andi Kleen, SuSE Labs. | |
8bccd85f | 6 | * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc. |
1da177e4 LT |
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 | 28 | * preferred Try a specific node first before normal fallback. |
00ef2d2f | 29 | * As a special case NUMA_NO_NODE here means do the allocation |
1da177e4 LT |
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 | ||
b1de0d13 MH |
68 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
69 | ||
1da177e4 | 70 | #include <linux/mempolicy.h> |
a520110e | 71 | #include <linux/pagewalk.h> |
1da177e4 LT |
72 | #include <linux/highmem.h> |
73 | #include <linux/hugetlb.h> | |
74 | #include <linux/kernel.h> | |
75 | #include <linux/sched.h> | |
6e84f315 | 76 | #include <linux/sched/mm.h> |
6a3827d7 | 77 | #include <linux/sched/numa_balancing.h> |
f719ff9b | 78 | #include <linux/sched/task.h> |
1da177e4 LT |
79 | #include <linux/nodemask.h> |
80 | #include <linux/cpuset.h> | |
1da177e4 LT |
81 | #include <linux/slab.h> |
82 | #include <linux/string.h> | |
b95f1b31 | 83 | #include <linux/export.h> |
b488893a | 84 | #include <linux/nsproxy.h> |
1da177e4 LT |
85 | #include <linux/interrupt.h> |
86 | #include <linux/init.h> | |
87 | #include <linux/compat.h> | |
31367466 | 88 | #include <linux/ptrace.h> |
dc9aa5b9 | 89 | #include <linux/swap.h> |
1a75a6c8 CL |
90 | #include <linux/seq_file.h> |
91 | #include <linux/proc_fs.h> | |
b20a3503 | 92 | #include <linux/migrate.h> |
62b61f61 | 93 | #include <linux/ksm.h> |
95a402c3 | 94 | #include <linux/rmap.h> |
86c3a764 | 95 | #include <linux/security.h> |
dbcb0f19 | 96 | #include <linux/syscalls.h> |
095f1fc4 | 97 | #include <linux/ctype.h> |
6d9c285a | 98 | #include <linux/mm_inline.h> |
b24f53a0 | 99 | #include <linux/mmu_notifier.h> |
b1de0d13 | 100 | #include <linux/printk.h> |
c8633798 | 101 | #include <linux/swapops.h> |
dc9aa5b9 | 102 | |
1da177e4 | 103 | #include <asm/tlbflush.h> |
7c0f6ba6 | 104 | #include <linux/uaccess.h> |
1da177e4 | 105 | |
62695a84 NP |
106 | #include "internal.h" |
107 | ||
38e35860 | 108 | /* Internal flags */ |
dc9aa5b9 | 109 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
38e35860 | 110 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
dc9aa5b9 | 111 | |
fcc234f8 PE |
112 | static struct kmem_cache *policy_cache; |
113 | static struct kmem_cache *sn_cache; | |
1da177e4 | 114 | |
1da177e4 LT |
115 | /* Highest zone. An specific allocation for a zone below that is not |
116 | policied. */ | |
6267276f | 117 | enum zone_type policy_zone = 0; |
1da177e4 | 118 | |
bea904d5 LS |
119 | /* |
120 | * run-time system-wide default policy => local allocation | |
121 | */ | |
e754d79d | 122 | static struct mempolicy default_policy = { |
1da177e4 | 123 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
bea904d5 | 124 | .mode = MPOL_PREFERRED, |
fc36b8d3 | 125 | .flags = MPOL_F_LOCAL, |
1da177e4 LT |
126 | }; |
127 | ||
5606e387 MG |
128 | static struct mempolicy preferred_node_policy[MAX_NUMNODES]; |
129 | ||
b2ca916c DW |
130 | /** |
131 | * numa_map_to_online_node - Find closest online node | |
132 | * @nid: Node id to start the search | |
133 | * | |
134 | * Lookup the next closest node by distance if @nid is not online. | |
135 | */ | |
136 | int numa_map_to_online_node(int node) | |
137 | { | |
4fcbe96e | 138 | int min_dist = INT_MAX, dist, n, min_node; |
b2ca916c | 139 | |
4fcbe96e DW |
140 | if (node == NUMA_NO_NODE || node_online(node)) |
141 | return node; | |
b2ca916c DW |
142 | |
143 | min_node = node; | |
4fcbe96e DW |
144 | for_each_online_node(n) { |
145 | dist = node_distance(node, n); | |
146 | if (dist < min_dist) { | |
147 | min_dist = dist; | |
148 | min_node = n; | |
b2ca916c DW |
149 | } |
150 | } | |
151 | ||
152 | return min_node; | |
153 | } | |
154 | EXPORT_SYMBOL_GPL(numa_map_to_online_node); | |
155 | ||
74d2c3a0 | 156 | struct mempolicy *get_task_policy(struct task_struct *p) |
5606e387 MG |
157 | { |
158 | struct mempolicy *pol = p->mempolicy; | |
f15ca78e | 159 | int node; |
5606e387 | 160 | |
f15ca78e ON |
161 | if (pol) |
162 | return pol; | |
5606e387 | 163 | |
f15ca78e ON |
164 | node = numa_node_id(); |
165 | if (node != NUMA_NO_NODE) { | |
166 | pol = &preferred_node_policy[node]; | |
167 | /* preferred_node_policy is not initialised early in boot */ | |
168 | if (pol->mode) | |
169 | return pol; | |
5606e387 MG |
170 | } |
171 | ||
f15ca78e | 172 | return &default_policy; |
5606e387 MG |
173 | } |
174 | ||
37012946 DR |
175 | static const struct mempolicy_operations { |
176 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
213980c0 | 177 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes); |
37012946 DR |
178 | } mpol_ops[MPOL_MAX]; |
179 | ||
f5b087b5 DR |
180 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
181 | { | |
6d556294 | 182 | return pol->flags & MPOL_MODE_FLAGS; |
4c50bc01 DR |
183 | } |
184 | ||
185 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
186 | const nodemask_t *rel) | |
187 | { | |
188 | nodemask_t tmp; | |
189 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
190 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
191 | } |
192 | ||
37012946 DR |
193 | static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes) |
194 | { | |
195 | if (nodes_empty(*nodes)) | |
196 | return -EINVAL; | |
197 | pol->v.nodes = *nodes; | |
198 | return 0; | |
199 | } | |
200 | ||
201 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
202 | { | |
203 | if (!nodes) | |
fc36b8d3 | 204 | pol->flags |= MPOL_F_LOCAL; /* local allocation */ |
37012946 DR |
205 | else if (nodes_empty(*nodes)) |
206 | return -EINVAL; /* no allowed nodes */ | |
207 | else | |
208 | pol->v.preferred_node = first_node(*nodes); | |
209 | return 0; | |
210 | } | |
211 | ||
212 | static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes) | |
213 | { | |
859f7ef1 | 214 | if (nodes_empty(*nodes)) |
37012946 DR |
215 | return -EINVAL; |
216 | pol->v.nodes = *nodes; | |
217 | return 0; | |
218 | } | |
219 | ||
58568d2a MX |
220 | /* |
221 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if | |
222 | * any, for the new policy. mpol_new() has already validated the nodes | |
223 | * parameter with respect to the policy mode and flags. But, we need to | |
224 | * handle an empty nodemask with MPOL_PREFERRED here. | |
225 | * | |
226 | * Must be called holding task's alloc_lock to protect task's mems_allowed | |
227 | * and mempolicy. May also be called holding the mmap_semaphore for write. | |
228 | */ | |
4bfc4495 KH |
229 | static int mpol_set_nodemask(struct mempolicy *pol, |
230 | const nodemask_t *nodes, struct nodemask_scratch *nsc) | |
58568d2a | 231 | { |
58568d2a MX |
232 | int ret; |
233 | ||
234 | /* if mode is MPOL_DEFAULT, pol is NULL. This is right. */ | |
235 | if (pol == NULL) | |
236 | return 0; | |
01f13bd6 | 237 | /* Check N_MEMORY */ |
4bfc4495 | 238 | nodes_and(nsc->mask1, |
01f13bd6 | 239 | cpuset_current_mems_allowed, node_states[N_MEMORY]); |
58568d2a MX |
240 | |
241 | VM_BUG_ON(!nodes); | |
242 | if (pol->mode == MPOL_PREFERRED && nodes_empty(*nodes)) | |
243 | nodes = NULL; /* explicit local allocation */ | |
244 | else { | |
245 | if (pol->flags & MPOL_F_RELATIVE_NODES) | |
859f7ef1 | 246 | mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1); |
58568d2a | 247 | else |
4bfc4495 KH |
248 | nodes_and(nsc->mask2, *nodes, nsc->mask1); |
249 | ||
58568d2a MX |
250 | if (mpol_store_user_nodemask(pol)) |
251 | pol->w.user_nodemask = *nodes; | |
252 | else | |
253 | pol->w.cpuset_mems_allowed = | |
254 | cpuset_current_mems_allowed; | |
255 | } | |
256 | ||
4bfc4495 KH |
257 | if (nodes) |
258 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); | |
259 | else | |
260 | ret = mpol_ops[pol->mode].create(pol, NULL); | |
58568d2a MX |
261 | return ret; |
262 | } | |
263 | ||
264 | /* | |
265 | * This function just creates a new policy, does some check and simple | |
266 | * initialization. You must invoke mpol_set_nodemask() to set nodes. | |
267 | */ | |
028fec41 DR |
268 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
269 | nodemask_t *nodes) | |
1da177e4 LT |
270 | { |
271 | struct mempolicy *policy; | |
272 | ||
028fec41 | 273 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
00ef2d2f | 274 | mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE); |
140d5a49 | 275 | |
3e1f0645 DR |
276 | if (mode == MPOL_DEFAULT) { |
277 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 278 | return ERR_PTR(-EINVAL); |
d3a71033 | 279 | return NULL; |
37012946 | 280 | } |
3e1f0645 DR |
281 | VM_BUG_ON(!nodes); |
282 | ||
283 | /* | |
284 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
285 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
286 | * All other modes require a valid pointer to a non-empty nodemask. | |
287 | */ | |
288 | if (mode == MPOL_PREFERRED) { | |
289 | if (nodes_empty(*nodes)) { | |
290 | if (((flags & MPOL_F_STATIC_NODES) || | |
291 | (flags & MPOL_F_RELATIVE_NODES))) | |
292 | return ERR_PTR(-EINVAL); | |
3e1f0645 | 293 | } |
479e2802 | 294 | } else if (mode == MPOL_LOCAL) { |
8d303e44 PK |
295 | if (!nodes_empty(*nodes) || |
296 | (flags & MPOL_F_STATIC_NODES) || | |
297 | (flags & MPOL_F_RELATIVE_NODES)) | |
479e2802 PZ |
298 | return ERR_PTR(-EINVAL); |
299 | mode = MPOL_PREFERRED; | |
3e1f0645 DR |
300 | } else if (nodes_empty(*nodes)) |
301 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
302 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
303 | if (!policy) | |
304 | return ERR_PTR(-ENOMEM); | |
305 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 306 | policy->mode = mode; |
3e1f0645 | 307 | policy->flags = flags; |
37012946 | 308 | |
1da177e4 | 309 | return policy; |
37012946 DR |
310 | } |
311 | ||
52cd3b07 LS |
312 | /* Slow path of a mpol destructor. */ |
313 | void __mpol_put(struct mempolicy *p) | |
314 | { | |
315 | if (!atomic_dec_and_test(&p->refcnt)) | |
316 | return; | |
52cd3b07 LS |
317 | kmem_cache_free(policy_cache, p); |
318 | } | |
319 | ||
213980c0 | 320 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
321 | { |
322 | } | |
323 | ||
213980c0 | 324 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
325 | { |
326 | nodemask_t tmp; | |
327 | ||
328 | if (pol->flags & MPOL_F_STATIC_NODES) | |
329 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
330 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
331 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
332 | else { | |
213980c0 VB |
333 | nodes_remap(tmp, pol->v.nodes,pol->w.cpuset_mems_allowed, |
334 | *nodes); | |
29b190fa | 335 | pol->w.cpuset_mems_allowed = *nodes; |
37012946 | 336 | } |
f5b087b5 | 337 | |
708c1bbc MX |
338 | if (nodes_empty(tmp)) |
339 | tmp = *nodes; | |
340 | ||
213980c0 | 341 | pol->v.nodes = tmp; |
37012946 DR |
342 | } |
343 | ||
344 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
213980c0 | 345 | const nodemask_t *nodes) |
37012946 DR |
346 | { |
347 | nodemask_t tmp; | |
348 | ||
37012946 DR |
349 | if (pol->flags & MPOL_F_STATIC_NODES) { |
350 | int node = first_node(pol->w.user_nodemask); | |
351 | ||
fc36b8d3 | 352 | if (node_isset(node, *nodes)) { |
37012946 | 353 | pol->v.preferred_node = node; |
fc36b8d3 LS |
354 | pol->flags &= ~MPOL_F_LOCAL; |
355 | } else | |
356 | pol->flags |= MPOL_F_LOCAL; | |
37012946 DR |
357 | } else if (pol->flags & MPOL_F_RELATIVE_NODES) { |
358 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
359 | pol->v.preferred_node = first_node(tmp); | |
fc36b8d3 | 360 | } else if (!(pol->flags & MPOL_F_LOCAL)) { |
37012946 DR |
361 | pol->v.preferred_node = node_remap(pol->v.preferred_node, |
362 | pol->w.cpuset_mems_allowed, | |
363 | *nodes); | |
364 | pol->w.cpuset_mems_allowed = *nodes; | |
365 | } | |
1da177e4 LT |
366 | } |
367 | ||
708c1bbc MX |
368 | /* |
369 | * mpol_rebind_policy - Migrate a policy to a different set of nodes | |
370 | * | |
213980c0 VB |
371 | * Per-vma policies are protected by mmap_sem. Allocations using per-task |
372 | * policies are protected by task->mems_allowed_seq to prevent a premature | |
373 | * OOM/allocation failure due to parallel nodemask modification. | |
708c1bbc | 374 | */ |
213980c0 | 375 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask) |
1d0d2680 | 376 | { |
1d0d2680 DR |
377 | if (!pol) |
378 | return; | |
2e25644e | 379 | if (!mpol_store_user_nodemask(pol) && !(pol->flags & MPOL_F_LOCAL) && |
1d0d2680 DR |
380 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
381 | return; | |
708c1bbc | 382 | |
213980c0 | 383 | mpol_ops[pol->mode].rebind(pol, newmask); |
1d0d2680 DR |
384 | } |
385 | ||
386 | /* | |
387 | * Wrapper for mpol_rebind_policy() that just requires task | |
388 | * pointer, and updates task mempolicy. | |
58568d2a MX |
389 | * |
390 | * Called with task's alloc_lock held. | |
1d0d2680 DR |
391 | */ |
392 | ||
213980c0 | 393 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) |
1d0d2680 | 394 | { |
213980c0 | 395 | mpol_rebind_policy(tsk->mempolicy, new); |
1d0d2680 DR |
396 | } |
397 | ||
398 | /* | |
399 | * Rebind each vma in mm to new nodemask. | |
400 | * | |
401 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
402 | */ | |
403 | ||
404 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
405 | { | |
406 | struct vm_area_struct *vma; | |
407 | ||
408 | down_write(&mm->mmap_sem); | |
409 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
213980c0 | 410 | mpol_rebind_policy(vma->vm_policy, new); |
1d0d2680 DR |
411 | up_write(&mm->mmap_sem); |
412 | } | |
413 | ||
37012946 DR |
414 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
415 | [MPOL_DEFAULT] = { | |
416 | .rebind = mpol_rebind_default, | |
417 | }, | |
418 | [MPOL_INTERLEAVE] = { | |
419 | .create = mpol_new_interleave, | |
420 | .rebind = mpol_rebind_nodemask, | |
421 | }, | |
422 | [MPOL_PREFERRED] = { | |
423 | .create = mpol_new_preferred, | |
424 | .rebind = mpol_rebind_preferred, | |
425 | }, | |
426 | [MPOL_BIND] = { | |
427 | .create = mpol_new_bind, | |
428 | .rebind = mpol_rebind_nodemask, | |
429 | }, | |
430 | }; | |
431 | ||
a53190a4 | 432 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
fc301289 | 433 | unsigned long flags); |
1a75a6c8 | 434 | |
6f4576e3 NH |
435 | struct queue_pages { |
436 | struct list_head *pagelist; | |
437 | unsigned long flags; | |
438 | nodemask_t *nmask; | |
f18da660 LX |
439 | unsigned long start; |
440 | unsigned long end; | |
441 | struct vm_area_struct *first; | |
6f4576e3 NH |
442 | }; |
443 | ||
88aaa2a1 NH |
444 | /* |
445 | * Check if the page's nid is in qp->nmask. | |
446 | * | |
447 | * If MPOL_MF_INVERT is set in qp->flags, check if the nid is | |
448 | * in the invert of qp->nmask. | |
449 | */ | |
450 | static inline bool queue_pages_required(struct page *page, | |
451 | struct queue_pages *qp) | |
452 | { | |
453 | int nid = page_to_nid(page); | |
454 | unsigned long flags = qp->flags; | |
455 | ||
456 | return node_isset(nid, *qp->nmask) == !(flags & MPOL_MF_INVERT); | |
457 | } | |
458 | ||
a7f40cfe | 459 | /* |
d8835445 YS |
460 | * queue_pages_pmd() has four possible return values: |
461 | * 0 - pages are placed on the right node or queued successfully. | |
462 | * 1 - there is unmovable page, and MPOL_MF_MOVE* & MPOL_MF_STRICT were | |
463 | * specified. | |
464 | * 2 - THP was split. | |
465 | * -EIO - is migration entry or only MPOL_MF_STRICT was specified and an | |
466 | * existing page was already on a node that does not follow the | |
467 | * policy. | |
a7f40cfe | 468 | */ |
c8633798 NH |
469 | static int queue_pages_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr, |
470 | unsigned long end, struct mm_walk *walk) | |
959a7e13 | 471 | __releases(ptl) |
c8633798 NH |
472 | { |
473 | int ret = 0; | |
474 | struct page *page; | |
475 | struct queue_pages *qp = walk->private; | |
476 | unsigned long flags; | |
477 | ||
478 | if (unlikely(is_pmd_migration_entry(*pmd))) { | |
a7f40cfe | 479 | ret = -EIO; |
c8633798 NH |
480 | goto unlock; |
481 | } | |
482 | page = pmd_page(*pmd); | |
483 | if (is_huge_zero_page(page)) { | |
484 | spin_unlock(ptl); | |
485 | __split_huge_pmd(walk->vma, pmd, addr, false, NULL); | |
d8835445 | 486 | ret = 2; |
c8633798 NH |
487 | goto out; |
488 | } | |
d8835445 | 489 | if (!queue_pages_required(page, qp)) |
c8633798 | 490 | goto unlock; |
c8633798 | 491 | |
c8633798 NH |
492 | flags = qp->flags; |
493 | /* go to thp migration */ | |
a7f40cfe | 494 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
a53190a4 YS |
495 | if (!vma_migratable(walk->vma) || |
496 | migrate_page_add(page, qp->pagelist, flags)) { | |
d8835445 | 497 | ret = 1; |
a7f40cfe YS |
498 | goto unlock; |
499 | } | |
a7f40cfe YS |
500 | } else |
501 | ret = -EIO; | |
c8633798 NH |
502 | unlock: |
503 | spin_unlock(ptl); | |
504 | out: | |
505 | return ret; | |
506 | } | |
507 | ||
98094945 NH |
508 | /* |
509 | * Scan through pages checking if pages follow certain conditions, | |
510 | * and move them to the pagelist if they do. | |
d8835445 YS |
511 | * |
512 | * queue_pages_pte_range() has three possible return values: | |
513 | * 0 - pages are placed on the right node or queued successfully. | |
514 | * 1 - there is unmovable page, and MPOL_MF_MOVE* & MPOL_MF_STRICT were | |
515 | * specified. | |
516 | * -EIO - only MPOL_MF_STRICT was specified and an existing page was already | |
517 | * on a node that does not follow the policy. | |
98094945 | 518 | */ |
6f4576e3 NH |
519 | static int queue_pages_pte_range(pmd_t *pmd, unsigned long addr, |
520 | unsigned long end, struct mm_walk *walk) | |
1da177e4 | 521 | { |
6f4576e3 NH |
522 | struct vm_area_struct *vma = walk->vma; |
523 | struct page *page; | |
524 | struct queue_pages *qp = walk->private; | |
525 | unsigned long flags = qp->flags; | |
c8633798 | 526 | int ret; |
d8835445 | 527 | bool has_unmovable = false; |
91612e0d | 528 | pte_t *pte; |
705e87c0 | 529 | spinlock_t *ptl; |
941150a3 | 530 | |
c8633798 NH |
531 | ptl = pmd_trans_huge_lock(pmd, vma); |
532 | if (ptl) { | |
533 | ret = queue_pages_pmd(pmd, ptl, addr, end, walk); | |
d8835445 | 534 | if (ret != 2) |
a7f40cfe | 535 | return ret; |
248db92d | 536 | } |
d8835445 | 537 | /* THP was split, fall through to pte walk */ |
91612e0d | 538 | |
337d9abf NH |
539 | if (pmd_trans_unstable(pmd)) |
540 | return 0; | |
94723aaf | 541 | |
6f4576e3 NH |
542 | pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); |
543 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
91612e0d | 544 | if (!pte_present(*pte)) |
1da177e4 | 545 | continue; |
6aab341e LT |
546 | page = vm_normal_page(vma, addr, *pte); |
547 | if (!page) | |
1da177e4 | 548 | continue; |
053837fc | 549 | /* |
62b61f61 HD |
550 | * vm_normal_page() filters out zero pages, but there might |
551 | * still be PageReserved pages to skip, perhaps in a VDSO. | |
053837fc | 552 | */ |
b79bc0a0 | 553 | if (PageReserved(page)) |
f4598c8b | 554 | continue; |
88aaa2a1 | 555 | if (!queue_pages_required(page, qp)) |
38e35860 | 556 | continue; |
a7f40cfe | 557 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
d8835445 YS |
558 | /* MPOL_MF_STRICT must be specified if we get here */ |
559 | if (!vma_migratable(vma)) { | |
560 | has_unmovable = true; | |
a7f40cfe | 561 | break; |
d8835445 | 562 | } |
a53190a4 YS |
563 | |
564 | /* | |
565 | * Do not abort immediately since there may be | |
566 | * temporary off LRU pages in the range. Still | |
567 | * need migrate other LRU pages. | |
568 | */ | |
569 | if (migrate_page_add(page, qp->pagelist, flags)) | |
570 | has_unmovable = true; | |
a7f40cfe YS |
571 | } else |
572 | break; | |
6f4576e3 NH |
573 | } |
574 | pte_unmap_unlock(pte - 1, ptl); | |
575 | cond_resched(); | |
d8835445 YS |
576 | |
577 | if (has_unmovable) | |
578 | return 1; | |
579 | ||
a7f40cfe | 580 | return addr != end ? -EIO : 0; |
91612e0d HD |
581 | } |
582 | ||
6f4576e3 NH |
583 | static int queue_pages_hugetlb(pte_t *pte, unsigned long hmask, |
584 | unsigned long addr, unsigned long end, | |
585 | struct mm_walk *walk) | |
e2d8cf40 | 586 | { |
dcf17635 | 587 | int ret = 0; |
e2d8cf40 | 588 | #ifdef CONFIG_HUGETLB_PAGE |
6f4576e3 | 589 | struct queue_pages *qp = walk->private; |
dcf17635 | 590 | unsigned long flags = (qp->flags & MPOL_MF_VALID); |
e2d8cf40 | 591 | struct page *page; |
cb900f41 | 592 | spinlock_t *ptl; |
d4c54919 | 593 | pte_t entry; |
e2d8cf40 | 594 | |
6f4576e3 NH |
595 | ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte); |
596 | entry = huge_ptep_get(pte); | |
d4c54919 NH |
597 | if (!pte_present(entry)) |
598 | goto unlock; | |
599 | page = pte_page(entry); | |
88aaa2a1 | 600 | if (!queue_pages_required(page, qp)) |
e2d8cf40 | 601 | goto unlock; |
dcf17635 LX |
602 | |
603 | if (flags == MPOL_MF_STRICT) { | |
604 | /* | |
605 | * STRICT alone means only detecting misplaced page and no | |
606 | * need to further check other vma. | |
607 | */ | |
608 | ret = -EIO; | |
609 | goto unlock; | |
610 | } | |
611 | ||
612 | if (!vma_migratable(walk->vma)) { | |
613 | /* | |
614 | * Must be STRICT with MOVE*, otherwise .test_walk() have | |
615 | * stopped walking current vma. | |
616 | * Detecting misplaced page but allow migrating pages which | |
617 | * have been queued. | |
618 | */ | |
619 | ret = 1; | |
620 | goto unlock; | |
621 | } | |
622 | ||
e2d8cf40 NH |
623 | /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */ |
624 | if (flags & (MPOL_MF_MOVE_ALL) || | |
dcf17635 LX |
625 | (flags & MPOL_MF_MOVE && page_mapcount(page) == 1)) { |
626 | if (!isolate_huge_page(page, qp->pagelist) && | |
627 | (flags & MPOL_MF_STRICT)) | |
628 | /* | |
629 | * Failed to isolate page but allow migrating pages | |
630 | * which have been queued. | |
631 | */ | |
632 | ret = 1; | |
633 | } | |
e2d8cf40 | 634 | unlock: |
cb900f41 | 635 | spin_unlock(ptl); |
e2d8cf40 NH |
636 | #else |
637 | BUG(); | |
638 | #endif | |
dcf17635 | 639 | return ret; |
1da177e4 LT |
640 | } |
641 | ||
5877231f | 642 | #ifdef CONFIG_NUMA_BALANCING |
b24f53a0 | 643 | /* |
4b10e7d5 MG |
644 | * This is used to mark a range of virtual addresses to be inaccessible. |
645 | * These are later cleared by a NUMA hinting fault. Depending on these | |
646 | * faults, pages may be migrated for better NUMA placement. | |
647 | * | |
648 | * This is assuming that NUMA faults are handled using PROT_NONE. If | |
649 | * an architecture makes a different choice, it will need further | |
650 | * changes to the core. | |
b24f53a0 | 651 | */ |
4b10e7d5 MG |
652 | unsigned long change_prot_numa(struct vm_area_struct *vma, |
653 | unsigned long addr, unsigned long end) | |
b24f53a0 | 654 | { |
4b10e7d5 | 655 | int nr_updated; |
b24f53a0 | 656 | |
58705444 | 657 | nr_updated = change_protection(vma, addr, end, PAGE_NONE, MM_CP_PROT_NUMA); |
03c5a6e1 MG |
658 | if (nr_updated) |
659 | count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated); | |
b24f53a0 | 660 | |
4b10e7d5 | 661 | return nr_updated; |
b24f53a0 LS |
662 | } |
663 | #else | |
664 | static unsigned long change_prot_numa(struct vm_area_struct *vma, | |
665 | unsigned long addr, unsigned long end) | |
666 | { | |
667 | return 0; | |
668 | } | |
5877231f | 669 | #endif /* CONFIG_NUMA_BALANCING */ |
b24f53a0 | 670 | |
6f4576e3 NH |
671 | static int queue_pages_test_walk(unsigned long start, unsigned long end, |
672 | struct mm_walk *walk) | |
673 | { | |
674 | struct vm_area_struct *vma = walk->vma; | |
675 | struct queue_pages *qp = walk->private; | |
676 | unsigned long endvma = vma->vm_end; | |
677 | unsigned long flags = qp->flags; | |
678 | ||
a18b3ac2 | 679 | /* range check first */ |
d888fb2b | 680 | VM_BUG_ON_VMA((vma->vm_start > start) || (vma->vm_end < end), vma); |
f18da660 LX |
681 | |
682 | if (!qp->first) { | |
683 | qp->first = vma; | |
684 | if (!(flags & MPOL_MF_DISCONTIG_OK) && | |
685 | (qp->start < vma->vm_start)) | |
686 | /* hole at head side of range */ | |
a18b3ac2 LX |
687 | return -EFAULT; |
688 | } | |
f18da660 LX |
689 | if (!(flags & MPOL_MF_DISCONTIG_OK) && |
690 | ((vma->vm_end < qp->end) && | |
691 | (!vma->vm_next || vma->vm_end < vma->vm_next->vm_start))) | |
692 | /* hole at middle or tail of range */ | |
693 | return -EFAULT; | |
a18b3ac2 | 694 | |
a7f40cfe YS |
695 | /* |
696 | * Need check MPOL_MF_STRICT to return -EIO if possible | |
697 | * regardless of vma_migratable | |
698 | */ | |
699 | if (!vma_migratable(vma) && | |
700 | !(flags & MPOL_MF_STRICT)) | |
48684a65 NH |
701 | return 1; |
702 | ||
6f4576e3 NH |
703 | if (endvma > end) |
704 | endvma = end; | |
6f4576e3 | 705 | |
6f4576e3 NH |
706 | if (flags & MPOL_MF_LAZY) { |
707 | /* Similar to task_numa_work, skip inaccessible VMAs */ | |
3122e80e | 708 | if (!is_vm_hugetlb_page(vma) && vma_is_accessible(vma) && |
4355c018 | 709 | !(vma->vm_flags & VM_MIXEDMAP)) |
6f4576e3 NH |
710 | change_prot_numa(vma, start, endvma); |
711 | return 1; | |
712 | } | |
713 | ||
77bf45e7 | 714 | /* queue pages from current vma */ |
a7f40cfe | 715 | if (flags & MPOL_MF_VALID) |
6f4576e3 NH |
716 | return 0; |
717 | return 1; | |
718 | } | |
719 | ||
7b86ac33 CH |
720 | static const struct mm_walk_ops queue_pages_walk_ops = { |
721 | .hugetlb_entry = queue_pages_hugetlb, | |
722 | .pmd_entry = queue_pages_pte_range, | |
723 | .test_walk = queue_pages_test_walk, | |
724 | }; | |
725 | ||
dc9aa5b9 | 726 | /* |
98094945 NH |
727 | * Walk through page tables and collect pages to be migrated. |
728 | * | |
729 | * If pages found in a given range are on a set of nodes (determined by | |
730 | * @nodes and @flags,) it's isolated and queued to the pagelist which is | |
d8835445 YS |
731 | * passed via @private. |
732 | * | |
733 | * queue_pages_range() has three possible return values: | |
734 | * 1 - there is unmovable page, but MPOL_MF_MOVE* & MPOL_MF_STRICT were | |
735 | * specified. | |
736 | * 0 - queue pages successfully or no misplaced page. | |
a85dfc30 YS |
737 | * errno - i.e. misplaced pages with MPOL_MF_STRICT specified (-EIO) or |
738 | * memory range specified by nodemask and maxnode points outside | |
739 | * your accessible address space (-EFAULT) | |
dc9aa5b9 | 740 | */ |
d05f0cdc | 741 | static int |
98094945 | 742 | queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end, |
6f4576e3 NH |
743 | nodemask_t *nodes, unsigned long flags, |
744 | struct list_head *pagelist) | |
1da177e4 | 745 | { |
f18da660 | 746 | int err; |
6f4576e3 NH |
747 | struct queue_pages qp = { |
748 | .pagelist = pagelist, | |
749 | .flags = flags, | |
750 | .nmask = nodes, | |
f18da660 LX |
751 | .start = start, |
752 | .end = end, | |
753 | .first = NULL, | |
6f4576e3 | 754 | }; |
6f4576e3 | 755 | |
f18da660 LX |
756 | err = walk_page_range(mm, start, end, &queue_pages_walk_ops, &qp); |
757 | ||
758 | if (!qp.first) | |
759 | /* whole range in hole */ | |
760 | err = -EFAULT; | |
761 | ||
762 | return err; | |
1da177e4 LT |
763 | } |
764 | ||
869833f2 KM |
765 | /* |
766 | * Apply policy to a single VMA | |
767 | * This must be called with the mmap_sem held for writing. | |
768 | */ | |
769 | static int vma_replace_policy(struct vm_area_struct *vma, | |
770 | struct mempolicy *pol) | |
8d34694c | 771 | { |
869833f2 KM |
772 | int err; |
773 | struct mempolicy *old; | |
774 | struct mempolicy *new; | |
8d34694c KM |
775 | |
776 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", | |
777 | vma->vm_start, vma->vm_end, vma->vm_pgoff, | |
778 | vma->vm_ops, vma->vm_file, | |
779 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
780 | ||
869833f2 KM |
781 | new = mpol_dup(pol); |
782 | if (IS_ERR(new)) | |
783 | return PTR_ERR(new); | |
784 | ||
785 | if (vma->vm_ops && vma->vm_ops->set_policy) { | |
8d34694c | 786 | err = vma->vm_ops->set_policy(vma, new); |
869833f2 KM |
787 | if (err) |
788 | goto err_out; | |
8d34694c | 789 | } |
869833f2 KM |
790 | |
791 | old = vma->vm_policy; | |
792 | vma->vm_policy = new; /* protected by mmap_sem */ | |
793 | mpol_put(old); | |
794 | ||
795 | return 0; | |
796 | err_out: | |
797 | mpol_put(new); | |
8d34694c KM |
798 | return err; |
799 | } | |
800 | ||
1da177e4 | 801 | /* Step 2: apply policy to a range and do splits. */ |
9d8cebd4 KM |
802 | static int mbind_range(struct mm_struct *mm, unsigned long start, |
803 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 LT |
804 | { |
805 | struct vm_area_struct *next; | |
9d8cebd4 KM |
806 | struct vm_area_struct *prev; |
807 | struct vm_area_struct *vma; | |
808 | int err = 0; | |
e26a5114 | 809 | pgoff_t pgoff; |
9d8cebd4 KM |
810 | unsigned long vmstart; |
811 | unsigned long vmend; | |
1da177e4 | 812 | |
097d5910 | 813 | vma = find_vma(mm, start); |
f18da660 | 814 | VM_BUG_ON(!vma); |
9d8cebd4 | 815 | |
097d5910 | 816 | prev = vma->vm_prev; |
e26a5114 KM |
817 | if (start > vma->vm_start) |
818 | prev = vma; | |
819 | ||
9d8cebd4 | 820 | for (; vma && vma->vm_start < end; prev = vma, vma = next) { |
1da177e4 | 821 | next = vma->vm_next; |
9d8cebd4 KM |
822 | vmstart = max(start, vma->vm_start); |
823 | vmend = min(end, vma->vm_end); | |
824 | ||
e26a5114 KM |
825 | if (mpol_equal(vma_policy(vma), new_pol)) |
826 | continue; | |
827 | ||
828 | pgoff = vma->vm_pgoff + | |
829 | ((vmstart - vma->vm_start) >> PAGE_SHIFT); | |
9d8cebd4 | 830 | prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags, |
19a809af AA |
831 | vma->anon_vma, vma->vm_file, pgoff, |
832 | new_pol, vma->vm_userfaultfd_ctx); | |
9d8cebd4 KM |
833 | if (prev) { |
834 | vma = prev; | |
835 | next = vma->vm_next; | |
3964acd0 ON |
836 | if (mpol_equal(vma_policy(vma), new_pol)) |
837 | continue; | |
838 | /* vma_merge() joined vma && vma->next, case 8 */ | |
839 | goto replace; | |
9d8cebd4 KM |
840 | } |
841 | if (vma->vm_start != vmstart) { | |
842 | err = split_vma(vma->vm_mm, vma, vmstart, 1); | |
843 | if (err) | |
844 | goto out; | |
845 | } | |
846 | if (vma->vm_end != vmend) { | |
847 | err = split_vma(vma->vm_mm, vma, vmend, 0); | |
848 | if (err) | |
849 | goto out; | |
850 | } | |
3964acd0 | 851 | replace: |
869833f2 | 852 | err = vma_replace_policy(vma, new_pol); |
8d34694c KM |
853 | if (err) |
854 | goto out; | |
1da177e4 | 855 | } |
9d8cebd4 KM |
856 | |
857 | out: | |
1da177e4 LT |
858 | return err; |
859 | } | |
860 | ||
1da177e4 | 861 | /* Set the process memory policy */ |
028fec41 DR |
862 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
863 | nodemask_t *nodes) | |
1da177e4 | 864 | { |
58568d2a | 865 | struct mempolicy *new, *old; |
4bfc4495 | 866 | NODEMASK_SCRATCH(scratch); |
58568d2a | 867 | int ret; |
1da177e4 | 868 | |
4bfc4495 KH |
869 | if (!scratch) |
870 | return -ENOMEM; | |
f4e53d91 | 871 | |
4bfc4495 KH |
872 | new = mpol_new(mode, flags, nodes); |
873 | if (IS_ERR(new)) { | |
874 | ret = PTR_ERR(new); | |
875 | goto out; | |
876 | } | |
2c7c3a7d | 877 | |
58568d2a | 878 | task_lock(current); |
4bfc4495 | 879 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a MX |
880 | if (ret) { |
881 | task_unlock(current); | |
58568d2a | 882 | mpol_put(new); |
4bfc4495 | 883 | goto out; |
58568d2a MX |
884 | } |
885 | old = current->mempolicy; | |
1da177e4 | 886 | current->mempolicy = new; |
45816682 VB |
887 | if (new && new->mode == MPOL_INTERLEAVE) |
888 | current->il_prev = MAX_NUMNODES-1; | |
58568d2a | 889 | task_unlock(current); |
58568d2a | 890 | mpol_put(old); |
4bfc4495 KH |
891 | ret = 0; |
892 | out: | |
893 | NODEMASK_SCRATCH_FREE(scratch); | |
894 | return ret; | |
1da177e4 LT |
895 | } |
896 | ||
bea904d5 LS |
897 | /* |
898 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
899 | * |
900 | * Called with task's alloc_lock held | |
bea904d5 LS |
901 | */ |
902 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 903 | { |
dfcd3c0d | 904 | nodes_clear(*nodes); |
bea904d5 LS |
905 | if (p == &default_policy) |
906 | return; | |
907 | ||
45c4745a | 908 | switch (p->mode) { |
19770b32 | 909 | case MPOL_BIND: |
1da177e4 | 910 | case MPOL_INTERLEAVE: |
dfcd3c0d | 911 | *nodes = p->v.nodes; |
1da177e4 LT |
912 | break; |
913 | case MPOL_PREFERRED: | |
fc36b8d3 | 914 | if (!(p->flags & MPOL_F_LOCAL)) |
dfcd3c0d | 915 | node_set(p->v.preferred_node, *nodes); |
53f2556b | 916 | /* else return empty node mask for local allocation */ |
1da177e4 LT |
917 | break; |
918 | default: | |
919 | BUG(); | |
920 | } | |
921 | } | |
922 | ||
3b9aadf7 | 923 | static int lookup_node(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 924 | { |
ba841078 | 925 | struct page *p = NULL; |
1da177e4 LT |
926 | int err; |
927 | ||
3b9aadf7 AA |
928 | int locked = 1; |
929 | err = get_user_pages_locked(addr & PAGE_MASK, 1, 0, &p, &locked); | |
ba841078 PX |
930 | if (err == 0) { |
931 | /* E.g. GUP interrupted by fatal signal */ | |
932 | err = -EFAULT; | |
933 | } else if (err > 0) { | |
1da177e4 LT |
934 | err = page_to_nid(p); |
935 | put_page(p); | |
936 | } | |
3b9aadf7 AA |
937 | if (locked) |
938 | up_read(&mm->mmap_sem); | |
1da177e4 LT |
939 | return err; |
940 | } | |
941 | ||
1da177e4 | 942 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
943 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
944 | unsigned long addr, unsigned long flags) | |
1da177e4 | 945 | { |
8bccd85f | 946 | int err; |
1da177e4 LT |
947 | struct mm_struct *mm = current->mm; |
948 | struct vm_area_struct *vma = NULL; | |
3b9aadf7 | 949 | struct mempolicy *pol = current->mempolicy, *pol_refcount = NULL; |
1da177e4 | 950 | |
754af6f5 LS |
951 | if (flags & |
952 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 953 | return -EINVAL; |
754af6f5 LS |
954 | |
955 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
956 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
957 | return -EINVAL; | |
958 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 959 | task_lock(current); |
754af6f5 | 960 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 961 | task_unlock(current); |
754af6f5 LS |
962 | return 0; |
963 | } | |
964 | ||
1da177e4 | 965 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
966 | /* |
967 | * Do NOT fall back to task policy if the | |
968 | * vma/shared policy at addr is NULL. We | |
969 | * want to return MPOL_DEFAULT in this case. | |
970 | */ | |
1da177e4 LT |
971 | down_read(&mm->mmap_sem); |
972 | vma = find_vma_intersection(mm, addr, addr+1); | |
973 | if (!vma) { | |
974 | up_read(&mm->mmap_sem); | |
975 | return -EFAULT; | |
976 | } | |
977 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
978 | pol = vma->vm_ops->get_policy(vma, addr); | |
979 | else | |
980 | pol = vma->vm_policy; | |
981 | } else if (addr) | |
982 | return -EINVAL; | |
983 | ||
984 | if (!pol) | |
bea904d5 | 985 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
986 | |
987 | if (flags & MPOL_F_NODE) { | |
988 | if (flags & MPOL_F_ADDR) { | |
3b9aadf7 AA |
989 | /* |
990 | * Take a refcount on the mpol, lookup_node() | |
991 | * wil drop the mmap_sem, so after calling | |
992 | * lookup_node() only "pol" remains valid, "vma" | |
993 | * is stale. | |
994 | */ | |
995 | pol_refcount = pol; | |
996 | vma = NULL; | |
997 | mpol_get(pol); | |
998 | err = lookup_node(mm, addr); | |
1da177e4 LT |
999 | if (err < 0) |
1000 | goto out; | |
8bccd85f | 1001 | *policy = err; |
1da177e4 | 1002 | } else if (pol == current->mempolicy && |
45c4745a | 1003 | pol->mode == MPOL_INTERLEAVE) { |
45816682 | 1004 | *policy = next_node_in(current->il_prev, pol->v.nodes); |
1da177e4 LT |
1005 | } else { |
1006 | err = -EINVAL; | |
1007 | goto out; | |
1008 | } | |
bea904d5 LS |
1009 | } else { |
1010 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
1011 | pol->mode; | |
d79df630 DR |
1012 | /* |
1013 | * Internal mempolicy flags must be masked off before exposing | |
1014 | * the policy to userspace. | |
1015 | */ | |
1016 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 1017 | } |
1da177e4 | 1018 | |
1da177e4 | 1019 | err = 0; |
58568d2a | 1020 | if (nmask) { |
c6b6ef8b LS |
1021 | if (mpol_store_user_nodemask(pol)) { |
1022 | *nmask = pol->w.user_nodemask; | |
1023 | } else { | |
1024 | task_lock(current); | |
1025 | get_policy_nodemask(pol, nmask); | |
1026 | task_unlock(current); | |
1027 | } | |
58568d2a | 1028 | } |
1da177e4 LT |
1029 | |
1030 | out: | |
52cd3b07 | 1031 | mpol_cond_put(pol); |
1da177e4 | 1032 | if (vma) |
3b9aadf7 AA |
1033 | up_read(&mm->mmap_sem); |
1034 | if (pol_refcount) | |
1035 | mpol_put(pol_refcount); | |
1da177e4 LT |
1036 | return err; |
1037 | } | |
1038 | ||
b20a3503 | 1039 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 | 1040 | /* |
c8633798 | 1041 | * page migration, thp tail pages can be passed. |
6ce3c4c0 | 1042 | */ |
a53190a4 | 1043 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
fc301289 | 1044 | unsigned long flags) |
6ce3c4c0 | 1045 | { |
c8633798 | 1046 | struct page *head = compound_head(page); |
6ce3c4c0 | 1047 | /* |
fc301289 | 1048 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 1049 | */ |
c8633798 NH |
1050 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(head) == 1) { |
1051 | if (!isolate_lru_page(head)) { | |
1052 | list_add_tail(&head->lru, pagelist); | |
1053 | mod_node_page_state(page_pgdat(head), | |
9de4f22a | 1054 | NR_ISOLATED_ANON + page_is_file_lru(head), |
c8633798 | 1055 | hpage_nr_pages(head)); |
a53190a4 YS |
1056 | } else if (flags & MPOL_MF_STRICT) { |
1057 | /* | |
1058 | * Non-movable page may reach here. And, there may be | |
1059 | * temporary off LRU pages or non-LRU movable pages. | |
1060 | * Treat them as unmovable pages since they can't be | |
1061 | * isolated, so they can't be moved at the moment. It | |
1062 | * should return -EIO for this case too. | |
1063 | */ | |
1064 | return -EIO; | |
62695a84 NP |
1065 | } |
1066 | } | |
a53190a4 YS |
1067 | |
1068 | return 0; | |
7e2ab150 | 1069 | } |
6ce3c4c0 | 1070 | |
a49bd4d7 | 1071 | /* page allocation callback for NUMA node migration */ |
666feb21 | 1072 | struct page *alloc_new_node_page(struct page *page, unsigned long node) |
95a402c3 | 1073 | { |
e2d8cf40 NH |
1074 | if (PageHuge(page)) |
1075 | return alloc_huge_page_node(page_hstate(compound_head(page)), | |
1076 | node); | |
94723aaf | 1077 | else if (PageTransHuge(page)) { |
c8633798 NH |
1078 | struct page *thp; |
1079 | ||
1080 | thp = alloc_pages_node(node, | |
1081 | (GFP_TRANSHUGE | __GFP_THISNODE), | |
1082 | HPAGE_PMD_ORDER); | |
1083 | if (!thp) | |
1084 | return NULL; | |
1085 | prep_transhuge_page(thp); | |
1086 | return thp; | |
1087 | } else | |
96db800f | 1088 | return __alloc_pages_node(node, GFP_HIGHUSER_MOVABLE | |
b360edb4 | 1089 | __GFP_THISNODE, 0); |
95a402c3 CL |
1090 | } |
1091 | ||
7e2ab150 CL |
1092 | /* |
1093 | * Migrate pages from one node to a target node. | |
1094 | * Returns error or the number of pages not migrated. | |
1095 | */ | |
dbcb0f19 AB |
1096 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
1097 | int flags) | |
7e2ab150 CL |
1098 | { |
1099 | nodemask_t nmask; | |
1100 | LIST_HEAD(pagelist); | |
1101 | int err = 0; | |
1102 | ||
1103 | nodes_clear(nmask); | |
1104 | node_set(source, nmask); | |
6ce3c4c0 | 1105 | |
08270807 MK |
1106 | /* |
1107 | * This does not "check" the range but isolates all pages that | |
1108 | * need migration. Between passing in the full user address | |
1109 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. | |
1110 | */ | |
1111 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); | |
98094945 | 1112 | queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask, |
7e2ab150 CL |
1113 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
1114 | ||
cf608ac1 | 1115 | if (!list_empty(&pagelist)) { |
a49bd4d7 | 1116 | err = migrate_pages(&pagelist, alloc_new_node_page, NULL, dest, |
9c620e2b | 1117 | MIGRATE_SYNC, MR_SYSCALL); |
cf608ac1 | 1118 | if (err) |
e2d8cf40 | 1119 | putback_movable_pages(&pagelist); |
cf608ac1 | 1120 | } |
95a402c3 | 1121 | |
7e2ab150 | 1122 | return err; |
6ce3c4c0 CL |
1123 | } |
1124 | ||
39743889 | 1125 | /* |
7e2ab150 CL |
1126 | * Move pages between the two nodesets so as to preserve the physical |
1127 | * layout as much as possible. | |
39743889 CL |
1128 | * |
1129 | * Returns the number of page that could not be moved. | |
1130 | */ | |
0ce72d4f AM |
1131 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1132 | const nodemask_t *to, int flags) | |
39743889 | 1133 | { |
7e2ab150 | 1134 | int busy = 0; |
0aedadf9 | 1135 | int err; |
7e2ab150 | 1136 | nodemask_t tmp; |
39743889 | 1137 | |
0aedadf9 CL |
1138 | err = migrate_prep(); |
1139 | if (err) | |
1140 | return err; | |
1141 | ||
53f2556b | 1142 | down_read(&mm->mmap_sem); |
39743889 | 1143 | |
da0aa138 KM |
1144 | /* |
1145 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1146 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1147 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1148 | * The pair of nodemasks 'to' and 'from' define the map. | |
1149 | * | |
1150 | * If no pair of bits is found that way, fallback to picking some | |
1151 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1152 | * 'source' and 'dest' bits are the same, this represents a node | |
1153 | * that will be migrating to itself, so no pages need move. | |
1154 | * | |
1155 | * If no bits are left in 'tmp', or if all remaining bits left | |
1156 | * in 'tmp' correspond to the same bit in 'to', return false | |
1157 | * (nothing left to migrate). | |
1158 | * | |
1159 | * This lets us pick a pair of nodes to migrate between, such that | |
1160 | * if possible the dest node is not already occupied by some other | |
1161 | * source node, minimizing the risk of overloading the memory on a | |
1162 | * node that would happen if we migrated incoming memory to a node | |
1163 | * before migrating outgoing memory source that same node. | |
1164 | * | |
1165 | * A single scan of tmp is sufficient. As we go, we remember the | |
1166 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1167 | * that not only moved, but what's better, moved to an empty slot | |
1168 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1169 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1170 | * most recent <s, d> pair that moved. If we get all the way through |
1171 | * the scan of tmp without finding any node that moved, much less | |
1172 | * moved to an empty node, then there is nothing left worth migrating. | |
1173 | */ | |
d4984711 | 1174 | |
0ce72d4f | 1175 | tmp = *from; |
7e2ab150 CL |
1176 | while (!nodes_empty(tmp)) { |
1177 | int s,d; | |
b76ac7e7 | 1178 | int source = NUMA_NO_NODE; |
7e2ab150 CL |
1179 | int dest = 0; |
1180 | ||
1181 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1182 | |
1183 | /* | |
1184 | * do_migrate_pages() tries to maintain the relative | |
1185 | * node relationship of the pages established between | |
1186 | * threads and memory areas. | |
1187 | * | |
1188 | * However if the number of source nodes is not equal to | |
1189 | * the number of destination nodes we can not preserve | |
1190 | * this node relative relationship. In that case, skip | |
1191 | * copying memory from a node that is in the destination | |
1192 | * mask. | |
1193 | * | |
1194 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1195 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1196 | */ | |
1197 | ||
0ce72d4f AM |
1198 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1199 | (node_isset(s, *to))) | |
4a5b18cc LW |
1200 | continue; |
1201 | ||
0ce72d4f | 1202 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1203 | if (s == d) |
1204 | continue; | |
1205 | ||
1206 | source = s; /* Node moved. Memorize */ | |
1207 | dest = d; | |
1208 | ||
1209 | /* dest not in remaining from nodes? */ | |
1210 | if (!node_isset(dest, tmp)) | |
1211 | break; | |
1212 | } | |
b76ac7e7 | 1213 | if (source == NUMA_NO_NODE) |
7e2ab150 CL |
1214 | break; |
1215 | ||
1216 | node_clear(source, tmp); | |
1217 | err = migrate_to_node(mm, source, dest, flags); | |
1218 | if (err > 0) | |
1219 | busy += err; | |
1220 | if (err < 0) | |
1221 | break; | |
39743889 CL |
1222 | } |
1223 | up_read(&mm->mmap_sem); | |
7e2ab150 CL |
1224 | if (err < 0) |
1225 | return err; | |
1226 | return busy; | |
b20a3503 CL |
1227 | |
1228 | } | |
1229 | ||
3ad33b24 LS |
1230 | /* |
1231 | * Allocate a new page for page migration based on vma policy. | |
d05f0cdc | 1232 | * Start by assuming the page is mapped by the same vma as contains @start. |
3ad33b24 LS |
1233 | * Search forward from there, if not. N.B., this assumes that the |
1234 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1235 | * is in virtual address order. | |
1236 | */ | |
666feb21 | 1237 | static struct page *new_page(struct page *page, unsigned long start) |
95a402c3 | 1238 | { |
d05f0cdc | 1239 | struct vm_area_struct *vma; |
3ad33b24 | 1240 | unsigned long uninitialized_var(address); |
95a402c3 | 1241 | |
d05f0cdc | 1242 | vma = find_vma(current->mm, start); |
3ad33b24 LS |
1243 | while (vma) { |
1244 | address = page_address_in_vma(page, vma); | |
1245 | if (address != -EFAULT) | |
1246 | break; | |
1247 | vma = vma->vm_next; | |
1248 | } | |
11c731e8 WL |
1249 | |
1250 | if (PageHuge(page)) { | |
389c8178 MH |
1251 | return alloc_huge_page_vma(page_hstate(compound_head(page)), |
1252 | vma, address); | |
94723aaf | 1253 | } else if (PageTransHuge(page)) { |
c8633798 NH |
1254 | struct page *thp; |
1255 | ||
19deb769 DR |
1256 | thp = alloc_hugepage_vma(GFP_TRANSHUGE, vma, address, |
1257 | HPAGE_PMD_ORDER); | |
c8633798 NH |
1258 | if (!thp) |
1259 | return NULL; | |
1260 | prep_transhuge_page(thp); | |
1261 | return thp; | |
11c731e8 | 1262 | } |
0bf598d8 | 1263 | /* |
11c731e8 | 1264 | * if !vma, alloc_page_vma() will use task or system default policy |
0bf598d8 | 1265 | */ |
0f556856 MH |
1266 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE | __GFP_RETRY_MAYFAIL, |
1267 | vma, address); | |
95a402c3 | 1268 | } |
b20a3503 CL |
1269 | #else |
1270 | ||
a53190a4 | 1271 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
b20a3503 CL |
1272 | unsigned long flags) |
1273 | { | |
a53190a4 | 1274 | return -EIO; |
39743889 CL |
1275 | } |
1276 | ||
0ce72d4f AM |
1277 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1278 | const nodemask_t *to, int flags) | |
b20a3503 CL |
1279 | { |
1280 | return -ENOSYS; | |
1281 | } | |
95a402c3 | 1282 | |
666feb21 | 1283 | static struct page *new_page(struct page *page, unsigned long start) |
95a402c3 CL |
1284 | { |
1285 | return NULL; | |
1286 | } | |
b20a3503 CL |
1287 | #endif |
1288 | ||
dbcb0f19 | 1289 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1290 | unsigned short mode, unsigned short mode_flags, |
1291 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 | 1292 | { |
6ce3c4c0 CL |
1293 | struct mm_struct *mm = current->mm; |
1294 | struct mempolicy *new; | |
1295 | unsigned long end; | |
1296 | int err; | |
d8835445 | 1297 | int ret; |
6ce3c4c0 CL |
1298 | LIST_HEAD(pagelist); |
1299 | ||
b24f53a0 | 1300 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
6ce3c4c0 | 1301 | return -EINVAL; |
74c00241 | 1302 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1303 | return -EPERM; |
1304 | ||
1305 | if (start & ~PAGE_MASK) | |
1306 | return -EINVAL; | |
1307 | ||
1308 | if (mode == MPOL_DEFAULT) | |
1309 | flags &= ~MPOL_MF_STRICT; | |
1310 | ||
1311 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
1312 | end = start + len; | |
1313 | ||
1314 | if (end < start) | |
1315 | return -EINVAL; | |
1316 | if (end == start) | |
1317 | return 0; | |
1318 | ||
028fec41 | 1319 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1320 | if (IS_ERR(new)) |
1321 | return PTR_ERR(new); | |
1322 | ||
b24f53a0 LS |
1323 | if (flags & MPOL_MF_LAZY) |
1324 | new->flags |= MPOL_F_MOF; | |
1325 | ||
6ce3c4c0 CL |
1326 | /* |
1327 | * If we are using the default policy then operation | |
1328 | * on discontinuous address spaces is okay after all | |
1329 | */ | |
1330 | if (!new) | |
1331 | flags |= MPOL_MF_DISCONTIG_OK; | |
1332 | ||
028fec41 DR |
1333 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1334 | start, start + len, mode, mode_flags, | |
00ef2d2f | 1335 | nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE); |
6ce3c4c0 | 1336 | |
0aedadf9 CL |
1337 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1338 | ||
1339 | err = migrate_prep(); | |
1340 | if (err) | |
b05ca738 | 1341 | goto mpol_out; |
0aedadf9 | 1342 | } |
4bfc4495 KH |
1343 | { |
1344 | NODEMASK_SCRATCH(scratch); | |
1345 | if (scratch) { | |
1346 | down_write(&mm->mmap_sem); | |
1347 | task_lock(current); | |
1348 | err = mpol_set_nodemask(new, nmask, scratch); | |
1349 | task_unlock(current); | |
1350 | if (err) | |
1351 | up_write(&mm->mmap_sem); | |
1352 | } else | |
1353 | err = -ENOMEM; | |
1354 | NODEMASK_SCRATCH_FREE(scratch); | |
1355 | } | |
b05ca738 KM |
1356 | if (err) |
1357 | goto mpol_out; | |
1358 | ||
d8835445 | 1359 | ret = queue_pages_range(mm, start, end, nmask, |
6ce3c4c0 | 1360 | flags | MPOL_MF_INVERT, &pagelist); |
d8835445 YS |
1361 | |
1362 | if (ret < 0) { | |
a85dfc30 | 1363 | err = ret; |
d8835445 YS |
1364 | goto up_out; |
1365 | } | |
1366 | ||
1367 | err = mbind_range(mm, start, end, new); | |
7e2ab150 | 1368 | |
b24f53a0 LS |
1369 | if (!err) { |
1370 | int nr_failed = 0; | |
1371 | ||
cf608ac1 | 1372 | if (!list_empty(&pagelist)) { |
b24f53a0 | 1373 | WARN_ON_ONCE(flags & MPOL_MF_LAZY); |
d05f0cdc HD |
1374 | nr_failed = migrate_pages(&pagelist, new_page, NULL, |
1375 | start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND); | |
cf608ac1 | 1376 | if (nr_failed) |
74060e4d | 1377 | putback_movable_pages(&pagelist); |
cf608ac1 | 1378 | } |
6ce3c4c0 | 1379 | |
d8835445 | 1380 | if ((ret > 0) || (nr_failed && (flags & MPOL_MF_STRICT))) |
6ce3c4c0 | 1381 | err = -EIO; |
a85dfc30 | 1382 | } else { |
d8835445 | 1383 | up_out: |
a85dfc30 YS |
1384 | if (!list_empty(&pagelist)) |
1385 | putback_movable_pages(&pagelist); | |
1386 | } | |
1387 | ||
6ce3c4c0 | 1388 | up_write(&mm->mmap_sem); |
d8835445 | 1389 | mpol_out: |
f0be3d32 | 1390 | mpol_put(new); |
6ce3c4c0 CL |
1391 | return err; |
1392 | } | |
1393 | ||
8bccd85f CL |
1394 | /* |
1395 | * User space interface with variable sized bitmaps for nodelists. | |
1396 | */ | |
1397 | ||
1398 | /* Copy a node mask from user space. */ | |
39743889 | 1399 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1400 | unsigned long maxnode) |
1401 | { | |
1402 | unsigned long k; | |
56521e7a | 1403 | unsigned long t; |
8bccd85f CL |
1404 | unsigned long nlongs; |
1405 | unsigned long endmask; | |
1406 | ||
1407 | --maxnode; | |
1408 | nodes_clear(*nodes); | |
1409 | if (maxnode == 0 || !nmask) | |
1410 | return 0; | |
a9c930ba | 1411 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1412 | return -EINVAL; |
8bccd85f CL |
1413 | |
1414 | nlongs = BITS_TO_LONGS(maxnode); | |
1415 | if ((maxnode % BITS_PER_LONG) == 0) | |
1416 | endmask = ~0UL; | |
1417 | else | |
1418 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1419 | ||
56521e7a YX |
1420 | /* |
1421 | * When the user specified more nodes than supported just check | |
1422 | * if the non supported part is all zero. | |
1423 | * | |
1424 | * If maxnode have more longs than MAX_NUMNODES, check | |
1425 | * the bits in that area first. And then go through to | |
1426 | * check the rest bits which equal or bigger than MAX_NUMNODES. | |
1427 | * Otherwise, just check bits [MAX_NUMNODES, maxnode). | |
1428 | */ | |
8bccd85f | 1429 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { |
8bccd85f | 1430 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { |
8bccd85f CL |
1431 | if (get_user(t, nmask + k)) |
1432 | return -EFAULT; | |
1433 | if (k == nlongs - 1) { | |
1434 | if (t & endmask) | |
1435 | return -EINVAL; | |
1436 | } else if (t) | |
1437 | return -EINVAL; | |
1438 | } | |
1439 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1440 | endmask = ~0UL; | |
1441 | } | |
1442 | ||
56521e7a YX |
1443 | if (maxnode > MAX_NUMNODES && MAX_NUMNODES % BITS_PER_LONG != 0) { |
1444 | unsigned long valid_mask = endmask; | |
1445 | ||
1446 | valid_mask &= ~((1UL << (MAX_NUMNODES % BITS_PER_LONG)) - 1); | |
1447 | if (get_user(t, nmask + nlongs - 1)) | |
1448 | return -EFAULT; | |
1449 | if (t & valid_mask) | |
1450 | return -EINVAL; | |
1451 | } | |
1452 | ||
8bccd85f CL |
1453 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) |
1454 | return -EFAULT; | |
1455 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1456 | return 0; | |
1457 | } | |
1458 | ||
1459 | /* Copy a kernel node mask to user space */ | |
1460 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1461 | nodemask_t *nodes) | |
1462 | { | |
1463 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
050c17f2 | 1464 | unsigned int nbytes = BITS_TO_LONGS(nr_node_ids) * sizeof(long); |
8bccd85f CL |
1465 | |
1466 | if (copy > nbytes) { | |
1467 | if (copy > PAGE_SIZE) | |
1468 | return -EINVAL; | |
1469 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1470 | return -EFAULT; | |
1471 | copy = nbytes; | |
1472 | } | |
1473 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1474 | } | |
1475 | ||
e7dc9ad6 DB |
1476 | static long kernel_mbind(unsigned long start, unsigned long len, |
1477 | unsigned long mode, const unsigned long __user *nmask, | |
1478 | unsigned long maxnode, unsigned int flags) | |
8bccd85f CL |
1479 | { |
1480 | nodemask_t nodes; | |
1481 | int err; | |
028fec41 | 1482 | unsigned short mode_flags; |
8bccd85f | 1483 | |
057d3389 | 1484 | start = untagged_addr(start); |
028fec41 DR |
1485 | mode_flags = mode & MPOL_MODE_FLAGS; |
1486 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
1487 | if (mode >= MPOL_MAX) |
1488 | return -EINVAL; | |
4c50bc01 DR |
1489 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
1490 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
1491 | return -EINVAL; | |
8bccd85f CL |
1492 | err = get_nodes(&nodes, nmask, maxnode); |
1493 | if (err) | |
1494 | return err; | |
028fec41 | 1495 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
1496 | } |
1497 | ||
e7dc9ad6 DB |
1498 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
1499 | unsigned long, mode, const unsigned long __user *, nmask, | |
1500 | unsigned long, maxnode, unsigned int, flags) | |
1501 | { | |
1502 | return kernel_mbind(start, len, mode, nmask, maxnode, flags); | |
1503 | } | |
1504 | ||
8bccd85f | 1505 | /* Set the process memory policy */ |
af03c4ac DB |
1506 | static long kernel_set_mempolicy(int mode, const unsigned long __user *nmask, |
1507 | unsigned long maxnode) | |
8bccd85f CL |
1508 | { |
1509 | int err; | |
1510 | nodemask_t nodes; | |
028fec41 | 1511 | unsigned short flags; |
8bccd85f | 1512 | |
028fec41 DR |
1513 | flags = mode & MPOL_MODE_FLAGS; |
1514 | mode &= ~MPOL_MODE_FLAGS; | |
1515 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 1516 | return -EINVAL; |
4c50bc01 DR |
1517 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
1518 | return -EINVAL; | |
8bccd85f CL |
1519 | err = get_nodes(&nodes, nmask, maxnode); |
1520 | if (err) | |
1521 | return err; | |
028fec41 | 1522 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1523 | } |
1524 | ||
af03c4ac DB |
1525 | SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask, |
1526 | unsigned long, maxnode) | |
1527 | { | |
1528 | return kernel_set_mempolicy(mode, nmask, maxnode); | |
1529 | } | |
1530 | ||
b6e9b0ba DB |
1531 | static int kernel_migrate_pages(pid_t pid, unsigned long maxnode, |
1532 | const unsigned long __user *old_nodes, | |
1533 | const unsigned long __user *new_nodes) | |
39743889 | 1534 | { |
596d7cfa | 1535 | struct mm_struct *mm = NULL; |
39743889 | 1536 | struct task_struct *task; |
39743889 CL |
1537 | nodemask_t task_nodes; |
1538 | int err; | |
596d7cfa KM |
1539 | nodemask_t *old; |
1540 | nodemask_t *new; | |
1541 | NODEMASK_SCRATCH(scratch); | |
1542 | ||
1543 | if (!scratch) | |
1544 | return -ENOMEM; | |
39743889 | 1545 | |
596d7cfa KM |
1546 | old = &scratch->mask1; |
1547 | new = &scratch->mask2; | |
1548 | ||
1549 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1550 | if (err) |
596d7cfa | 1551 | goto out; |
39743889 | 1552 | |
596d7cfa | 1553 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1554 | if (err) |
596d7cfa | 1555 | goto out; |
39743889 CL |
1556 | |
1557 | /* Find the mm_struct */ | |
55cfaa3c | 1558 | rcu_read_lock(); |
228ebcbe | 1559 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1560 | if (!task) { |
55cfaa3c | 1561 | rcu_read_unlock(); |
596d7cfa KM |
1562 | err = -ESRCH; |
1563 | goto out; | |
39743889 | 1564 | } |
3268c63e | 1565 | get_task_struct(task); |
39743889 | 1566 | |
596d7cfa | 1567 | err = -EINVAL; |
39743889 CL |
1568 | |
1569 | /* | |
31367466 OE |
1570 | * Check if this process has the right to modify the specified process. |
1571 | * Use the regular "ptrace_may_access()" checks. | |
39743889 | 1572 | */ |
31367466 | 1573 | if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) { |
c69e8d9c | 1574 | rcu_read_unlock(); |
39743889 | 1575 | err = -EPERM; |
3268c63e | 1576 | goto out_put; |
39743889 | 1577 | } |
c69e8d9c | 1578 | rcu_read_unlock(); |
39743889 CL |
1579 | |
1580 | task_nodes = cpuset_mems_allowed(task); | |
1581 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1582 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1583 | err = -EPERM; |
3268c63e | 1584 | goto out_put; |
39743889 CL |
1585 | } |
1586 | ||
0486a38b YX |
1587 | task_nodes = cpuset_mems_allowed(current); |
1588 | nodes_and(*new, *new, task_nodes); | |
1589 | if (nodes_empty(*new)) | |
1590 | goto out_put; | |
1591 | ||
86c3a764 DQ |
1592 | err = security_task_movememory(task); |
1593 | if (err) | |
3268c63e | 1594 | goto out_put; |
86c3a764 | 1595 | |
3268c63e CL |
1596 | mm = get_task_mm(task); |
1597 | put_task_struct(task); | |
f2a9ef88 SL |
1598 | |
1599 | if (!mm) { | |
3268c63e | 1600 | err = -EINVAL; |
f2a9ef88 SL |
1601 | goto out; |
1602 | } | |
1603 | ||
1604 | err = do_migrate_pages(mm, old, new, | |
1605 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1606 | |
1607 | mmput(mm); | |
1608 | out: | |
596d7cfa KM |
1609 | NODEMASK_SCRATCH_FREE(scratch); |
1610 | ||
39743889 | 1611 | return err; |
3268c63e CL |
1612 | |
1613 | out_put: | |
1614 | put_task_struct(task); | |
1615 | goto out; | |
1616 | ||
39743889 CL |
1617 | } |
1618 | ||
b6e9b0ba DB |
1619 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1620 | const unsigned long __user *, old_nodes, | |
1621 | const unsigned long __user *, new_nodes) | |
1622 | { | |
1623 | return kernel_migrate_pages(pid, maxnode, old_nodes, new_nodes); | |
1624 | } | |
1625 | ||
39743889 | 1626 | |
8bccd85f | 1627 | /* Retrieve NUMA policy */ |
af03c4ac DB |
1628 | static int kernel_get_mempolicy(int __user *policy, |
1629 | unsigned long __user *nmask, | |
1630 | unsigned long maxnode, | |
1631 | unsigned long addr, | |
1632 | unsigned long flags) | |
8bccd85f | 1633 | { |
dbcb0f19 AB |
1634 | int err; |
1635 | int uninitialized_var(pval); | |
8bccd85f CL |
1636 | nodemask_t nodes; |
1637 | ||
057d3389 AK |
1638 | addr = untagged_addr(addr); |
1639 | ||
050c17f2 | 1640 | if (nmask != NULL && maxnode < nr_node_ids) |
8bccd85f CL |
1641 | return -EINVAL; |
1642 | ||
1643 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1644 | ||
1645 | if (err) | |
1646 | return err; | |
1647 | ||
1648 | if (policy && put_user(pval, policy)) | |
1649 | return -EFAULT; | |
1650 | ||
1651 | if (nmask) | |
1652 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1653 | ||
1654 | return err; | |
1655 | } | |
1656 | ||
af03c4ac DB |
1657 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1658 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1659 | unsigned long, addr, unsigned long, flags) | |
1660 | { | |
1661 | return kernel_get_mempolicy(policy, nmask, maxnode, addr, flags); | |
1662 | } | |
1663 | ||
1da177e4 LT |
1664 | #ifdef CONFIG_COMPAT |
1665 | ||
c93e0f6c HC |
1666 | COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1667 | compat_ulong_t __user *, nmask, | |
1668 | compat_ulong_t, maxnode, | |
1669 | compat_ulong_t, addr, compat_ulong_t, flags) | |
1da177e4 LT |
1670 | { |
1671 | long err; | |
1672 | unsigned long __user *nm = NULL; | |
1673 | unsigned long nr_bits, alloc_size; | |
1674 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1675 | ||
050c17f2 | 1676 | nr_bits = min_t(unsigned long, maxnode-1, nr_node_ids); |
1da177e4 LT |
1677 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; |
1678 | ||
1679 | if (nmask) | |
1680 | nm = compat_alloc_user_space(alloc_size); | |
1681 | ||
af03c4ac | 1682 | err = kernel_get_mempolicy(policy, nm, nr_bits+1, addr, flags); |
1da177e4 LT |
1683 | |
1684 | if (!err && nmask) { | |
2bbff6c7 KH |
1685 | unsigned long copy_size; |
1686 | copy_size = min_t(unsigned long, sizeof(bm), alloc_size); | |
1687 | err = copy_from_user(bm, nm, copy_size); | |
1da177e4 LT |
1688 | /* ensure entire bitmap is zeroed */ |
1689 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1690 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1691 | } | |
1692 | ||
1693 | return err; | |
1694 | } | |
1695 | ||
c93e0f6c HC |
1696 | COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask, |
1697 | compat_ulong_t, maxnode) | |
1da177e4 | 1698 | { |
1da177e4 LT |
1699 | unsigned long __user *nm = NULL; |
1700 | unsigned long nr_bits, alloc_size; | |
1701 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1702 | ||
1703 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1704 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1705 | ||
1706 | if (nmask) { | |
cf01fb99 CS |
1707 | if (compat_get_bitmap(bm, nmask, nr_bits)) |
1708 | return -EFAULT; | |
1da177e4 | 1709 | nm = compat_alloc_user_space(alloc_size); |
cf01fb99 CS |
1710 | if (copy_to_user(nm, bm, alloc_size)) |
1711 | return -EFAULT; | |
1da177e4 LT |
1712 | } |
1713 | ||
af03c4ac | 1714 | return kernel_set_mempolicy(mode, nm, nr_bits+1); |
1da177e4 LT |
1715 | } |
1716 | ||
c93e0f6c HC |
1717 | COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len, |
1718 | compat_ulong_t, mode, compat_ulong_t __user *, nmask, | |
1719 | compat_ulong_t, maxnode, compat_ulong_t, flags) | |
1da177e4 | 1720 | { |
1da177e4 LT |
1721 | unsigned long __user *nm = NULL; |
1722 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1723 | nodemask_t bm; |
1da177e4 LT |
1724 | |
1725 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1726 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1727 | ||
1728 | if (nmask) { | |
cf01fb99 CS |
1729 | if (compat_get_bitmap(nodes_addr(bm), nmask, nr_bits)) |
1730 | return -EFAULT; | |
1da177e4 | 1731 | nm = compat_alloc_user_space(alloc_size); |
cf01fb99 CS |
1732 | if (copy_to_user(nm, nodes_addr(bm), alloc_size)) |
1733 | return -EFAULT; | |
1da177e4 LT |
1734 | } |
1735 | ||
e7dc9ad6 | 1736 | return kernel_mbind(start, len, mode, nm, nr_bits+1, flags); |
1da177e4 LT |
1737 | } |
1738 | ||
b6e9b0ba DB |
1739 | COMPAT_SYSCALL_DEFINE4(migrate_pages, compat_pid_t, pid, |
1740 | compat_ulong_t, maxnode, | |
1741 | const compat_ulong_t __user *, old_nodes, | |
1742 | const compat_ulong_t __user *, new_nodes) | |
1743 | { | |
1744 | unsigned long __user *old = NULL; | |
1745 | unsigned long __user *new = NULL; | |
1746 | nodemask_t tmp_mask; | |
1747 | unsigned long nr_bits; | |
1748 | unsigned long size; | |
1749 | ||
1750 | nr_bits = min_t(unsigned long, maxnode - 1, MAX_NUMNODES); | |
1751 | size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1752 | if (old_nodes) { | |
1753 | if (compat_get_bitmap(nodes_addr(tmp_mask), old_nodes, nr_bits)) | |
1754 | return -EFAULT; | |
1755 | old = compat_alloc_user_space(new_nodes ? size * 2 : size); | |
1756 | if (new_nodes) | |
1757 | new = old + size / sizeof(unsigned long); | |
1758 | if (copy_to_user(old, nodes_addr(tmp_mask), size)) | |
1759 | return -EFAULT; | |
1760 | } | |
1761 | if (new_nodes) { | |
1762 | if (compat_get_bitmap(nodes_addr(tmp_mask), new_nodes, nr_bits)) | |
1763 | return -EFAULT; | |
1764 | if (new == NULL) | |
1765 | new = compat_alloc_user_space(size); | |
1766 | if (copy_to_user(new, nodes_addr(tmp_mask), size)) | |
1767 | return -EFAULT; | |
1768 | } | |
1769 | return kernel_migrate_pages(pid, nr_bits + 1, old, new); | |
1770 | } | |
1771 | ||
1772 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 1773 | |
20ca87f2 LX |
1774 | bool vma_migratable(struct vm_area_struct *vma) |
1775 | { | |
1776 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1777 | return false; | |
1778 | ||
1779 | /* | |
1780 | * DAX device mappings require predictable access latency, so avoid | |
1781 | * incurring periodic faults. | |
1782 | */ | |
1783 | if (vma_is_dax(vma)) | |
1784 | return false; | |
1785 | ||
1786 | if (is_vm_hugetlb_page(vma) && | |
1787 | !hugepage_migration_supported(hstate_vma(vma))) | |
1788 | return false; | |
1789 | ||
1790 | /* | |
1791 | * Migration allocates pages in the highest zone. If we cannot | |
1792 | * do so then migration (at least from node to node) is not | |
1793 | * possible. | |
1794 | */ | |
1795 | if (vma->vm_file && | |
1796 | gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping)) | |
1797 | < policy_zone) | |
1798 | return false; | |
1799 | return true; | |
1800 | } | |
1801 | ||
74d2c3a0 ON |
1802 | struct mempolicy *__get_vma_policy(struct vm_area_struct *vma, |
1803 | unsigned long addr) | |
1da177e4 | 1804 | { |
8d90274b | 1805 | struct mempolicy *pol = NULL; |
1da177e4 LT |
1806 | |
1807 | if (vma) { | |
480eccf9 | 1808 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
8d90274b | 1809 | pol = vma->vm_ops->get_policy(vma, addr); |
00442ad0 | 1810 | } else if (vma->vm_policy) { |
1da177e4 | 1811 | pol = vma->vm_policy; |
00442ad0 MG |
1812 | |
1813 | /* | |
1814 | * shmem_alloc_page() passes MPOL_F_SHARED policy with | |
1815 | * a pseudo vma whose vma->vm_ops=NULL. Take a reference | |
1816 | * count on these policies which will be dropped by | |
1817 | * mpol_cond_put() later | |
1818 | */ | |
1819 | if (mpol_needs_cond_ref(pol)) | |
1820 | mpol_get(pol); | |
1821 | } | |
1da177e4 | 1822 | } |
f15ca78e | 1823 | |
74d2c3a0 ON |
1824 | return pol; |
1825 | } | |
1826 | ||
1827 | /* | |
dd6eecb9 | 1828 | * get_vma_policy(@vma, @addr) |
74d2c3a0 ON |
1829 | * @vma: virtual memory area whose policy is sought |
1830 | * @addr: address in @vma for shared policy lookup | |
1831 | * | |
1832 | * Returns effective policy for a VMA at specified address. | |
dd6eecb9 | 1833 | * Falls back to current->mempolicy or system default policy, as necessary. |
74d2c3a0 ON |
1834 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1835 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1836 | * freeing by another task. It is the caller's responsibility to free the | |
1837 | * extra reference for shared policies. | |
1838 | */ | |
ac79f78d | 1839 | static struct mempolicy *get_vma_policy(struct vm_area_struct *vma, |
dd6eecb9 | 1840 | unsigned long addr) |
74d2c3a0 ON |
1841 | { |
1842 | struct mempolicy *pol = __get_vma_policy(vma, addr); | |
1843 | ||
8d90274b | 1844 | if (!pol) |
dd6eecb9 | 1845 | pol = get_task_policy(current); |
8d90274b | 1846 | |
1da177e4 LT |
1847 | return pol; |
1848 | } | |
1849 | ||
6b6482bb | 1850 | bool vma_policy_mof(struct vm_area_struct *vma) |
fc314724 | 1851 | { |
6b6482bb | 1852 | struct mempolicy *pol; |
fc314724 | 1853 | |
6b6482bb ON |
1854 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
1855 | bool ret = false; | |
fc314724 | 1856 | |
6b6482bb ON |
1857 | pol = vma->vm_ops->get_policy(vma, vma->vm_start); |
1858 | if (pol && (pol->flags & MPOL_F_MOF)) | |
1859 | ret = true; | |
1860 | mpol_cond_put(pol); | |
8d90274b | 1861 | |
6b6482bb | 1862 | return ret; |
fc314724 MG |
1863 | } |
1864 | ||
6b6482bb | 1865 | pol = vma->vm_policy; |
8d90274b | 1866 | if (!pol) |
6b6482bb | 1867 | pol = get_task_policy(current); |
8d90274b | 1868 | |
fc314724 MG |
1869 | return pol->flags & MPOL_F_MOF; |
1870 | } | |
1871 | ||
d3eb1570 LJ |
1872 | static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone) |
1873 | { | |
1874 | enum zone_type dynamic_policy_zone = policy_zone; | |
1875 | ||
1876 | BUG_ON(dynamic_policy_zone == ZONE_MOVABLE); | |
1877 | ||
1878 | /* | |
1879 | * if policy->v.nodes has movable memory only, | |
1880 | * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only. | |
1881 | * | |
1882 | * policy->v.nodes is intersect with node_states[N_MEMORY]. | |
1883 | * so if the following test faile, it implies | |
1884 | * policy->v.nodes has movable memory only. | |
1885 | */ | |
1886 | if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY])) | |
1887 | dynamic_policy_zone = ZONE_MOVABLE; | |
1888 | ||
1889 | return zone >= dynamic_policy_zone; | |
1890 | } | |
1891 | ||
52cd3b07 LS |
1892 | /* |
1893 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1894 | * page allocation | |
1895 | */ | |
1896 | static nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) | |
19770b32 MG |
1897 | { |
1898 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1899 | if (unlikely(policy->mode == MPOL_BIND) && |
d3eb1570 | 1900 | apply_policy_zone(policy, gfp_zone(gfp)) && |
19770b32 MG |
1901 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) |
1902 | return &policy->v.nodes; | |
1903 | ||
1904 | return NULL; | |
1905 | } | |
1906 | ||
04ec6264 VB |
1907 | /* Return the node id preferred by the given mempolicy, or the given id */ |
1908 | static int policy_node(gfp_t gfp, struct mempolicy *policy, | |
1909 | int nd) | |
1da177e4 | 1910 | { |
6d840958 MH |
1911 | if (policy->mode == MPOL_PREFERRED && !(policy->flags & MPOL_F_LOCAL)) |
1912 | nd = policy->v.preferred_node; | |
1913 | else { | |
19770b32 | 1914 | /* |
6d840958 MH |
1915 | * __GFP_THISNODE shouldn't even be used with the bind policy |
1916 | * because we might easily break the expectation to stay on the | |
1917 | * requested node and not break the policy. | |
19770b32 | 1918 | */ |
6d840958 | 1919 | WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE)); |
1da177e4 | 1920 | } |
6d840958 | 1921 | |
04ec6264 | 1922 | return nd; |
1da177e4 LT |
1923 | } |
1924 | ||
1925 | /* Do dynamic interleaving for a process */ | |
1926 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1927 | { | |
45816682 | 1928 | unsigned next; |
1da177e4 LT |
1929 | struct task_struct *me = current; |
1930 | ||
45816682 | 1931 | next = next_node_in(me->il_prev, policy->v.nodes); |
f5b087b5 | 1932 | if (next < MAX_NUMNODES) |
45816682 VB |
1933 | me->il_prev = next; |
1934 | return next; | |
1da177e4 LT |
1935 | } |
1936 | ||
dc85da15 CL |
1937 | /* |
1938 | * Depending on the memory policy provide a node from which to allocate the | |
1939 | * next slab entry. | |
1940 | */ | |
2a389610 | 1941 | unsigned int mempolicy_slab_node(void) |
dc85da15 | 1942 | { |
e7b691b0 | 1943 | struct mempolicy *policy; |
2a389610 | 1944 | int node = numa_mem_id(); |
e7b691b0 AK |
1945 | |
1946 | if (in_interrupt()) | |
2a389610 | 1947 | return node; |
e7b691b0 AK |
1948 | |
1949 | policy = current->mempolicy; | |
fc36b8d3 | 1950 | if (!policy || policy->flags & MPOL_F_LOCAL) |
2a389610 | 1951 | return node; |
bea904d5 LS |
1952 | |
1953 | switch (policy->mode) { | |
1954 | case MPOL_PREFERRED: | |
fc36b8d3 LS |
1955 | /* |
1956 | * handled MPOL_F_LOCAL above | |
1957 | */ | |
1958 | return policy->v.preferred_node; | |
765c4507 | 1959 | |
dc85da15 CL |
1960 | case MPOL_INTERLEAVE: |
1961 | return interleave_nodes(policy); | |
1962 | ||
dd1a239f | 1963 | case MPOL_BIND: { |
c33d6c06 MG |
1964 | struct zoneref *z; |
1965 | ||
dc85da15 CL |
1966 | /* |
1967 | * Follow bind policy behavior and start allocation at the | |
1968 | * first node. | |
1969 | */ | |
19770b32 | 1970 | struct zonelist *zonelist; |
19770b32 | 1971 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); |
c9634cf0 | 1972 | zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK]; |
c33d6c06 MG |
1973 | z = first_zones_zonelist(zonelist, highest_zoneidx, |
1974 | &policy->v.nodes); | |
c1093b74 | 1975 | return z->zone ? zone_to_nid(z->zone) : node; |
dd1a239f | 1976 | } |
dc85da15 | 1977 | |
dc85da15 | 1978 | default: |
bea904d5 | 1979 | BUG(); |
dc85da15 CL |
1980 | } |
1981 | } | |
1982 | ||
fee83b3a AM |
1983 | /* |
1984 | * Do static interleaving for a VMA with known offset @n. Returns the n'th | |
1985 | * node in pol->v.nodes (starting from n=0), wrapping around if n exceeds the | |
1986 | * number of present nodes. | |
1987 | */ | |
98c70baa | 1988 | static unsigned offset_il_node(struct mempolicy *pol, unsigned long n) |
1da177e4 | 1989 | { |
dfcd3c0d | 1990 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1991 | unsigned target; |
fee83b3a AM |
1992 | int i; |
1993 | int nid; | |
1da177e4 | 1994 | |
f5b087b5 DR |
1995 | if (!nnodes) |
1996 | return numa_node_id(); | |
fee83b3a AM |
1997 | target = (unsigned int)n % nnodes; |
1998 | nid = first_node(pol->v.nodes); | |
1999 | for (i = 0; i < target; i++) | |
dfcd3c0d | 2000 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
2001 | return nid; |
2002 | } | |
2003 | ||
5da7ca86 CL |
2004 | /* Determine a node number for interleave */ |
2005 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
2006 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
2007 | { | |
2008 | if (vma) { | |
2009 | unsigned long off; | |
2010 | ||
3b98b087 NA |
2011 | /* |
2012 | * for small pages, there is no difference between | |
2013 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
2014 | * for huge pages, since vm_pgoff is in units of small | |
2015 | * pages, we need to shift off the always 0 bits to get | |
2016 | * a useful offset. | |
2017 | */ | |
2018 | BUG_ON(shift < PAGE_SHIFT); | |
2019 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 | 2020 | off += (addr - vma->vm_start) >> shift; |
98c70baa | 2021 | return offset_il_node(pol, off); |
5da7ca86 CL |
2022 | } else |
2023 | return interleave_nodes(pol); | |
2024 | } | |
2025 | ||
00ac59ad | 2026 | #ifdef CONFIG_HUGETLBFS |
480eccf9 | 2027 | /* |
04ec6264 | 2028 | * huge_node(@vma, @addr, @gfp_flags, @mpol) |
b46e14ac FF |
2029 | * @vma: virtual memory area whose policy is sought |
2030 | * @addr: address in @vma for shared policy lookup and interleave policy | |
2031 | * @gfp_flags: for requested zone | |
2032 | * @mpol: pointer to mempolicy pointer for reference counted mempolicy | |
2033 | * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 | 2034 | * |
04ec6264 | 2035 | * Returns a nid suitable for a huge page allocation and a pointer |
52cd3b07 LS |
2036 | * to the struct mempolicy for conditional unref after allocation. |
2037 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's | |
2038 | * @nodemask for filtering the zonelist. | |
c0ff7453 | 2039 | * |
d26914d1 | 2040 | * Must be protected by read_mems_allowed_begin() |
480eccf9 | 2041 | */ |
04ec6264 VB |
2042 | int huge_node(struct vm_area_struct *vma, unsigned long addr, gfp_t gfp_flags, |
2043 | struct mempolicy **mpol, nodemask_t **nodemask) | |
5da7ca86 | 2044 | { |
04ec6264 | 2045 | int nid; |
5da7ca86 | 2046 | |
dd6eecb9 | 2047 | *mpol = get_vma_policy(vma, addr); |
19770b32 | 2048 | *nodemask = NULL; /* assume !MPOL_BIND */ |
5da7ca86 | 2049 | |
52cd3b07 | 2050 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
04ec6264 VB |
2051 | nid = interleave_nid(*mpol, vma, addr, |
2052 | huge_page_shift(hstate_vma(vma))); | |
52cd3b07 | 2053 | } else { |
04ec6264 | 2054 | nid = policy_node(gfp_flags, *mpol, numa_node_id()); |
52cd3b07 LS |
2055 | if ((*mpol)->mode == MPOL_BIND) |
2056 | *nodemask = &(*mpol)->v.nodes; | |
480eccf9 | 2057 | } |
04ec6264 | 2058 | return nid; |
5da7ca86 | 2059 | } |
06808b08 LS |
2060 | |
2061 | /* | |
2062 | * init_nodemask_of_mempolicy | |
2063 | * | |
2064 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
2065 | * to indicate default policy. Otherwise, extract the policy nodemask | |
2066 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
2067 | * initialize the argument nodemask to contain the single node for | |
2068 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
2069 | * of non-default mempolicy. | |
2070 | * | |
2071 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
2072 | * because the current task is examining it's own mempolicy and a task's | |
2073 | * mempolicy is only ever changed by the task itself. | |
2074 | * | |
2075 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
2076 | */ | |
2077 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
2078 | { | |
2079 | struct mempolicy *mempolicy; | |
2080 | int nid; | |
2081 | ||
2082 | if (!(mask && current->mempolicy)) | |
2083 | return false; | |
2084 | ||
c0ff7453 | 2085 | task_lock(current); |
06808b08 LS |
2086 | mempolicy = current->mempolicy; |
2087 | switch (mempolicy->mode) { | |
2088 | case MPOL_PREFERRED: | |
2089 | if (mempolicy->flags & MPOL_F_LOCAL) | |
2090 | nid = numa_node_id(); | |
2091 | else | |
2092 | nid = mempolicy->v.preferred_node; | |
2093 | init_nodemask_of_node(mask, nid); | |
2094 | break; | |
2095 | ||
2096 | case MPOL_BIND: | |
06808b08 LS |
2097 | case MPOL_INTERLEAVE: |
2098 | *mask = mempolicy->v.nodes; | |
2099 | break; | |
2100 | ||
2101 | default: | |
2102 | BUG(); | |
2103 | } | |
c0ff7453 | 2104 | task_unlock(current); |
06808b08 LS |
2105 | |
2106 | return true; | |
2107 | } | |
00ac59ad | 2108 | #endif |
5da7ca86 | 2109 | |
6f48d0eb DR |
2110 | /* |
2111 | * mempolicy_nodemask_intersects | |
2112 | * | |
2113 | * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default | |
2114 | * policy. Otherwise, check for intersection between mask and the policy | |
2115 | * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local' | |
2116 | * policy, always return true since it may allocate elsewhere on fallback. | |
2117 | * | |
2118 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
2119 | */ | |
2120 | bool mempolicy_nodemask_intersects(struct task_struct *tsk, | |
2121 | const nodemask_t *mask) | |
2122 | { | |
2123 | struct mempolicy *mempolicy; | |
2124 | bool ret = true; | |
2125 | ||
2126 | if (!mask) | |
2127 | return ret; | |
2128 | task_lock(tsk); | |
2129 | mempolicy = tsk->mempolicy; | |
2130 | if (!mempolicy) | |
2131 | goto out; | |
2132 | ||
2133 | switch (mempolicy->mode) { | |
2134 | case MPOL_PREFERRED: | |
2135 | /* | |
2136 | * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to | |
2137 | * allocate from, they may fallback to other nodes when oom. | |
2138 | * Thus, it's possible for tsk to have allocated memory from | |
2139 | * nodes in mask. | |
2140 | */ | |
2141 | break; | |
2142 | case MPOL_BIND: | |
2143 | case MPOL_INTERLEAVE: | |
2144 | ret = nodes_intersects(mempolicy->v.nodes, *mask); | |
2145 | break; | |
2146 | default: | |
2147 | BUG(); | |
2148 | } | |
2149 | out: | |
2150 | task_unlock(tsk); | |
2151 | return ret; | |
2152 | } | |
2153 | ||
1da177e4 LT |
2154 | /* Allocate a page in interleaved policy. |
2155 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
2156 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
2157 | unsigned nid) | |
1da177e4 | 2158 | { |
1da177e4 LT |
2159 | struct page *page; |
2160 | ||
04ec6264 | 2161 | page = __alloc_pages(gfp, order, nid); |
4518085e KW |
2162 | /* skip NUMA_INTERLEAVE_HIT counter update if numa stats is disabled */ |
2163 | if (!static_branch_likely(&vm_numa_stat_key)) | |
2164 | return page; | |
de55c8b2 AR |
2165 | if (page && page_to_nid(page) == nid) { |
2166 | preempt_disable(); | |
2167 | __inc_numa_state(page_zone(page), NUMA_INTERLEAVE_HIT); | |
2168 | preempt_enable(); | |
2169 | } | |
1da177e4 LT |
2170 | return page; |
2171 | } | |
2172 | ||
2173 | /** | |
0bbbc0b3 | 2174 | * alloc_pages_vma - Allocate a page for a VMA. |
1da177e4 LT |
2175 | * |
2176 | * @gfp: | |
2177 | * %GFP_USER user allocation. | |
2178 | * %GFP_KERNEL kernel allocations, | |
2179 | * %GFP_HIGHMEM highmem/user allocations, | |
2180 | * %GFP_FS allocation should not call back into a file system. | |
2181 | * %GFP_ATOMIC don't sleep. | |
2182 | * | |
0bbbc0b3 | 2183 | * @order:Order of the GFP allocation. |
1da177e4 LT |
2184 | * @vma: Pointer to VMA or NULL if not available. |
2185 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
be97a41b | 2186 | * @node: Which node to prefer for allocation (modulo policy). |
19deb769 | 2187 | * @hugepage: for hugepages try only the preferred node if possible |
1da177e4 LT |
2188 | * |
2189 | * This function allocates a page from the kernel page pool and applies | |
2190 | * a NUMA policy associated with the VMA or the current process. | |
2191 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
2192 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
be97a41b VB |
2193 | * all allocations for pages that will be mapped into user space. Returns |
2194 | * NULL when no page can be allocated. | |
1da177e4 LT |
2195 | */ |
2196 | struct page * | |
0bbbc0b3 | 2197 | alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma, |
19deb769 | 2198 | unsigned long addr, int node, bool hugepage) |
1da177e4 | 2199 | { |
cc9a6c87 | 2200 | struct mempolicy *pol; |
c0ff7453 | 2201 | struct page *page; |
04ec6264 | 2202 | int preferred_nid; |
be97a41b | 2203 | nodemask_t *nmask; |
cc9a6c87 | 2204 | |
dd6eecb9 | 2205 | pol = get_vma_policy(vma, addr); |
1da177e4 | 2206 | |
0867a57c VB |
2207 | if (pol->mode == MPOL_INTERLEAVE) { |
2208 | unsigned nid; | |
2209 | ||
2210 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); | |
2211 | mpol_cond_put(pol); | |
2212 | page = alloc_page_interleave(gfp, order, nid); | |
2213 | goto out; | |
19deb769 DR |
2214 | } |
2215 | ||
2216 | if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) { | |
2217 | int hpage_node = node; | |
2218 | ||
2219 | /* | |
2220 | * For hugepage allocation and non-interleave policy which | |
2221 | * allows the current node (or other explicitly preferred | |
2222 | * node) we only try to allocate from the current/preferred | |
2223 | * node and don't fall back to other nodes, as the cost of | |
2224 | * remote accesses would likely offset THP benefits. | |
2225 | * | |
2226 | * If the policy is interleave, or does not allow the current | |
2227 | * node in its nodemask, we allocate the standard way. | |
2228 | */ | |
2229 | if (pol->mode == MPOL_PREFERRED && !(pol->flags & MPOL_F_LOCAL)) | |
2230 | hpage_node = pol->v.preferred_node; | |
2231 | ||
2232 | nmask = policy_nodemask(gfp, pol); | |
2233 | if (!nmask || node_isset(hpage_node, *nmask)) { | |
2234 | mpol_cond_put(pol); | |
cc638f32 VB |
2235 | /* |
2236 | * First, try to allocate THP only on local node, but | |
2237 | * don't reclaim unnecessarily, just compact. | |
2238 | */ | |
19deb769 | 2239 | page = __alloc_pages_node(hpage_node, |
cc638f32 | 2240 | gfp | __GFP_THISNODE | __GFP_NORETRY, order); |
76e654cc DR |
2241 | |
2242 | /* | |
2243 | * If hugepage allocations are configured to always | |
2244 | * synchronous compact or the vma has been madvised | |
2245 | * to prefer hugepage backing, retry allowing remote | |
cc638f32 | 2246 | * memory with both reclaim and compact as well. |
76e654cc DR |
2247 | */ |
2248 | if (!page && (gfp & __GFP_DIRECT_RECLAIM)) | |
2249 | page = __alloc_pages_node(hpage_node, | |
cc638f32 | 2250 | gfp, order); |
76e654cc | 2251 | |
19deb769 DR |
2252 | goto out; |
2253 | } | |
356ff8a9 DR |
2254 | } |
2255 | ||
be97a41b | 2256 | nmask = policy_nodemask(gfp, pol); |
04ec6264 VB |
2257 | preferred_nid = policy_node(gfp, pol, node); |
2258 | page = __alloc_pages_nodemask(gfp, order, preferred_nid, nmask); | |
d51e9894 | 2259 | mpol_cond_put(pol); |
be97a41b | 2260 | out: |
c0ff7453 | 2261 | return page; |
1da177e4 | 2262 | } |
69262215 | 2263 | EXPORT_SYMBOL(alloc_pages_vma); |
1da177e4 LT |
2264 | |
2265 | /** | |
2266 | * alloc_pages_current - Allocate pages. | |
2267 | * | |
2268 | * @gfp: | |
2269 | * %GFP_USER user allocation, | |
2270 | * %GFP_KERNEL kernel allocation, | |
2271 | * %GFP_HIGHMEM highmem allocation, | |
2272 | * %GFP_FS don't call back into a file system. | |
2273 | * %GFP_ATOMIC don't sleep. | |
2274 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
2275 | * | |
2276 | * Allocate a page from the kernel page pool. When not in | |
2277 | * interrupt context and apply the current process NUMA policy. | |
2278 | * Returns NULL when no page can be allocated. | |
1da177e4 | 2279 | */ |
dd0fc66f | 2280 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 | 2281 | { |
8d90274b | 2282 | struct mempolicy *pol = &default_policy; |
c0ff7453 | 2283 | struct page *page; |
1da177e4 | 2284 | |
8d90274b ON |
2285 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) |
2286 | pol = get_task_policy(current); | |
52cd3b07 LS |
2287 | |
2288 | /* | |
2289 | * No reference counting needed for current->mempolicy | |
2290 | * nor system default_policy | |
2291 | */ | |
45c4745a | 2292 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 MX |
2293 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
2294 | else | |
2295 | page = __alloc_pages_nodemask(gfp, order, | |
04ec6264 | 2296 | policy_node(gfp, pol, numa_node_id()), |
5c4b4be3 | 2297 | policy_nodemask(gfp, pol)); |
cc9a6c87 | 2298 | |
c0ff7453 | 2299 | return page; |
1da177e4 LT |
2300 | } |
2301 | EXPORT_SYMBOL(alloc_pages_current); | |
2302 | ||
ef0855d3 ON |
2303 | int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst) |
2304 | { | |
2305 | struct mempolicy *pol = mpol_dup(vma_policy(src)); | |
2306 | ||
2307 | if (IS_ERR(pol)) | |
2308 | return PTR_ERR(pol); | |
2309 | dst->vm_policy = pol; | |
2310 | return 0; | |
2311 | } | |
2312 | ||
4225399a | 2313 | /* |
846a16bf | 2314 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2315 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2316 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2317 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2318 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2319 | * |
2320 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2321 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2322 | */ |
4225399a | 2323 | |
846a16bf LS |
2324 | /* Slow path of a mempolicy duplicate */ |
2325 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2326 | { |
2327 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2328 | ||
2329 | if (!new) | |
2330 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2331 | |
2332 | /* task's mempolicy is protected by alloc_lock */ | |
2333 | if (old == current->mempolicy) { | |
2334 | task_lock(current); | |
2335 | *new = *old; | |
2336 | task_unlock(current); | |
2337 | } else | |
2338 | *new = *old; | |
2339 | ||
4225399a PJ |
2340 | if (current_cpuset_is_being_rebound()) { |
2341 | nodemask_t mems = cpuset_mems_allowed(current); | |
213980c0 | 2342 | mpol_rebind_policy(new, &mems); |
4225399a | 2343 | } |
1da177e4 | 2344 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2345 | return new; |
2346 | } | |
2347 | ||
2348 | /* Slow path of a mempolicy comparison */ | |
fcfb4dcc | 2349 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2350 | { |
2351 | if (!a || !b) | |
fcfb4dcc | 2352 | return false; |
45c4745a | 2353 | if (a->mode != b->mode) |
fcfb4dcc | 2354 | return false; |
19800502 | 2355 | if (a->flags != b->flags) |
fcfb4dcc | 2356 | return false; |
19800502 BL |
2357 | if (mpol_store_user_nodemask(a)) |
2358 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2359 | return false; |
19800502 | 2360 | |
45c4745a | 2361 | switch (a->mode) { |
19770b32 | 2362 | case MPOL_BIND: |
1da177e4 | 2363 | case MPOL_INTERLEAVE: |
fcfb4dcc | 2364 | return !!nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 | 2365 | case MPOL_PREFERRED: |
8970a63e YX |
2366 | /* a's ->flags is the same as b's */ |
2367 | if (a->flags & MPOL_F_LOCAL) | |
2368 | return true; | |
75719661 | 2369 | return a->v.preferred_node == b->v.preferred_node; |
1da177e4 LT |
2370 | default: |
2371 | BUG(); | |
fcfb4dcc | 2372 | return false; |
1da177e4 LT |
2373 | } |
2374 | } | |
2375 | ||
1da177e4 LT |
2376 | /* |
2377 | * Shared memory backing store policy support. | |
2378 | * | |
2379 | * Remember policies even when nobody has shared memory mapped. | |
2380 | * The policies are kept in Red-Black tree linked from the inode. | |
4a8c7bb5 | 2381 | * They are protected by the sp->lock rwlock, which should be held |
1da177e4 LT |
2382 | * for any accesses to the tree. |
2383 | */ | |
2384 | ||
4a8c7bb5 NZ |
2385 | /* |
2386 | * lookup first element intersecting start-end. Caller holds sp->lock for | |
2387 | * reading or for writing | |
2388 | */ | |
1da177e4 LT |
2389 | static struct sp_node * |
2390 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2391 | { | |
2392 | struct rb_node *n = sp->root.rb_node; | |
2393 | ||
2394 | while (n) { | |
2395 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2396 | ||
2397 | if (start >= p->end) | |
2398 | n = n->rb_right; | |
2399 | else if (end <= p->start) | |
2400 | n = n->rb_left; | |
2401 | else | |
2402 | break; | |
2403 | } | |
2404 | if (!n) | |
2405 | return NULL; | |
2406 | for (;;) { | |
2407 | struct sp_node *w = NULL; | |
2408 | struct rb_node *prev = rb_prev(n); | |
2409 | if (!prev) | |
2410 | break; | |
2411 | w = rb_entry(prev, struct sp_node, nd); | |
2412 | if (w->end <= start) | |
2413 | break; | |
2414 | n = prev; | |
2415 | } | |
2416 | return rb_entry(n, struct sp_node, nd); | |
2417 | } | |
2418 | ||
4a8c7bb5 NZ |
2419 | /* |
2420 | * Insert a new shared policy into the list. Caller holds sp->lock for | |
2421 | * writing. | |
2422 | */ | |
1da177e4 LT |
2423 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) |
2424 | { | |
2425 | struct rb_node **p = &sp->root.rb_node; | |
2426 | struct rb_node *parent = NULL; | |
2427 | struct sp_node *nd; | |
2428 | ||
2429 | while (*p) { | |
2430 | parent = *p; | |
2431 | nd = rb_entry(parent, struct sp_node, nd); | |
2432 | if (new->start < nd->start) | |
2433 | p = &(*p)->rb_left; | |
2434 | else if (new->end > nd->end) | |
2435 | p = &(*p)->rb_right; | |
2436 | else | |
2437 | BUG(); | |
2438 | } | |
2439 | rb_link_node(&new->nd, parent, p); | |
2440 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2441 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2442 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2443 | } |
2444 | ||
2445 | /* Find shared policy intersecting idx */ | |
2446 | struct mempolicy * | |
2447 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2448 | { | |
2449 | struct mempolicy *pol = NULL; | |
2450 | struct sp_node *sn; | |
2451 | ||
2452 | if (!sp->root.rb_node) | |
2453 | return NULL; | |
4a8c7bb5 | 2454 | read_lock(&sp->lock); |
1da177e4 LT |
2455 | sn = sp_lookup(sp, idx, idx+1); |
2456 | if (sn) { | |
2457 | mpol_get(sn->policy); | |
2458 | pol = sn->policy; | |
2459 | } | |
4a8c7bb5 | 2460 | read_unlock(&sp->lock); |
1da177e4 LT |
2461 | return pol; |
2462 | } | |
2463 | ||
63f74ca2 KM |
2464 | static void sp_free(struct sp_node *n) |
2465 | { | |
2466 | mpol_put(n->policy); | |
2467 | kmem_cache_free(sn_cache, n); | |
2468 | } | |
2469 | ||
771fb4d8 LS |
2470 | /** |
2471 | * mpol_misplaced - check whether current page node is valid in policy | |
2472 | * | |
b46e14ac FF |
2473 | * @page: page to be checked |
2474 | * @vma: vm area where page mapped | |
2475 | * @addr: virtual address where page mapped | |
771fb4d8 LS |
2476 | * |
2477 | * Lookup current policy node id for vma,addr and "compare to" page's | |
2478 | * node id. | |
2479 | * | |
2480 | * Returns: | |
2481 | * -1 - not misplaced, page is in the right node | |
2482 | * node - node id where the page should be | |
2483 | * | |
2484 | * Policy determination "mimics" alloc_page_vma(). | |
2485 | * Called from fault path where we know the vma and faulting address. | |
2486 | */ | |
2487 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) | |
2488 | { | |
2489 | struct mempolicy *pol; | |
c33d6c06 | 2490 | struct zoneref *z; |
771fb4d8 LS |
2491 | int curnid = page_to_nid(page); |
2492 | unsigned long pgoff; | |
90572890 PZ |
2493 | int thiscpu = raw_smp_processor_id(); |
2494 | int thisnid = cpu_to_node(thiscpu); | |
98fa15f3 | 2495 | int polnid = NUMA_NO_NODE; |
771fb4d8 LS |
2496 | int ret = -1; |
2497 | ||
dd6eecb9 | 2498 | pol = get_vma_policy(vma, addr); |
771fb4d8 LS |
2499 | if (!(pol->flags & MPOL_F_MOF)) |
2500 | goto out; | |
2501 | ||
2502 | switch (pol->mode) { | |
2503 | case MPOL_INTERLEAVE: | |
771fb4d8 LS |
2504 | pgoff = vma->vm_pgoff; |
2505 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; | |
98c70baa | 2506 | polnid = offset_il_node(pol, pgoff); |
771fb4d8 LS |
2507 | break; |
2508 | ||
2509 | case MPOL_PREFERRED: | |
2510 | if (pol->flags & MPOL_F_LOCAL) | |
2511 | polnid = numa_node_id(); | |
2512 | else | |
2513 | polnid = pol->v.preferred_node; | |
2514 | break; | |
2515 | ||
2516 | case MPOL_BIND: | |
c33d6c06 | 2517 | |
771fb4d8 LS |
2518 | /* |
2519 | * allows binding to multiple nodes. | |
2520 | * use current page if in policy nodemask, | |
2521 | * else select nearest allowed node, if any. | |
2522 | * If no allowed nodes, use current [!misplaced]. | |
2523 | */ | |
2524 | if (node_isset(curnid, pol->v.nodes)) | |
2525 | goto out; | |
c33d6c06 | 2526 | z = first_zones_zonelist( |
771fb4d8 LS |
2527 | node_zonelist(numa_node_id(), GFP_HIGHUSER), |
2528 | gfp_zone(GFP_HIGHUSER), | |
c33d6c06 | 2529 | &pol->v.nodes); |
c1093b74 | 2530 | polnid = zone_to_nid(z->zone); |
771fb4d8 LS |
2531 | break; |
2532 | ||
2533 | default: | |
2534 | BUG(); | |
2535 | } | |
5606e387 MG |
2536 | |
2537 | /* Migrate the page towards the node whose CPU is referencing it */ | |
e42c8ff2 | 2538 | if (pol->flags & MPOL_F_MORON) { |
90572890 | 2539 | polnid = thisnid; |
5606e387 | 2540 | |
10f39042 | 2541 | if (!should_numa_migrate_memory(current, page, curnid, thiscpu)) |
de1c9ce6 | 2542 | goto out; |
e42c8ff2 MG |
2543 | } |
2544 | ||
771fb4d8 LS |
2545 | if (curnid != polnid) |
2546 | ret = polnid; | |
2547 | out: | |
2548 | mpol_cond_put(pol); | |
2549 | ||
2550 | return ret; | |
2551 | } | |
2552 | ||
c11600e4 DR |
2553 | /* |
2554 | * Drop the (possibly final) reference to task->mempolicy. It needs to be | |
2555 | * dropped after task->mempolicy is set to NULL so that any allocation done as | |
2556 | * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed | |
2557 | * policy. | |
2558 | */ | |
2559 | void mpol_put_task_policy(struct task_struct *task) | |
2560 | { | |
2561 | struct mempolicy *pol; | |
2562 | ||
2563 | task_lock(task); | |
2564 | pol = task->mempolicy; | |
2565 | task->mempolicy = NULL; | |
2566 | task_unlock(task); | |
2567 | mpol_put(pol); | |
2568 | } | |
2569 | ||
1da177e4 LT |
2570 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2571 | { | |
140d5a49 | 2572 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2573 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2574 | sp_free(n); |
1da177e4 LT |
2575 | } |
2576 | ||
42288fe3 MG |
2577 | static void sp_node_init(struct sp_node *node, unsigned long start, |
2578 | unsigned long end, struct mempolicy *pol) | |
2579 | { | |
2580 | node->start = start; | |
2581 | node->end = end; | |
2582 | node->policy = pol; | |
2583 | } | |
2584 | ||
dbcb0f19 AB |
2585 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2586 | struct mempolicy *pol) | |
1da177e4 | 2587 | { |
869833f2 KM |
2588 | struct sp_node *n; |
2589 | struct mempolicy *newpol; | |
1da177e4 | 2590 | |
869833f2 | 2591 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2592 | if (!n) |
2593 | return NULL; | |
869833f2 KM |
2594 | |
2595 | newpol = mpol_dup(pol); | |
2596 | if (IS_ERR(newpol)) { | |
2597 | kmem_cache_free(sn_cache, n); | |
2598 | return NULL; | |
2599 | } | |
2600 | newpol->flags |= MPOL_F_SHARED; | |
42288fe3 | 2601 | sp_node_init(n, start, end, newpol); |
869833f2 | 2602 | |
1da177e4 LT |
2603 | return n; |
2604 | } | |
2605 | ||
2606 | /* Replace a policy range. */ | |
2607 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2608 | unsigned long end, struct sp_node *new) | |
2609 | { | |
b22d127a | 2610 | struct sp_node *n; |
42288fe3 MG |
2611 | struct sp_node *n_new = NULL; |
2612 | struct mempolicy *mpol_new = NULL; | |
b22d127a | 2613 | int ret = 0; |
1da177e4 | 2614 | |
42288fe3 | 2615 | restart: |
4a8c7bb5 | 2616 | write_lock(&sp->lock); |
1da177e4 LT |
2617 | n = sp_lookup(sp, start, end); |
2618 | /* Take care of old policies in the same range. */ | |
2619 | while (n && n->start < end) { | |
2620 | struct rb_node *next = rb_next(&n->nd); | |
2621 | if (n->start >= start) { | |
2622 | if (n->end <= end) | |
2623 | sp_delete(sp, n); | |
2624 | else | |
2625 | n->start = end; | |
2626 | } else { | |
2627 | /* Old policy spanning whole new range. */ | |
2628 | if (n->end > end) { | |
42288fe3 MG |
2629 | if (!n_new) |
2630 | goto alloc_new; | |
2631 | ||
2632 | *mpol_new = *n->policy; | |
2633 | atomic_set(&mpol_new->refcnt, 1); | |
7880639c | 2634 | sp_node_init(n_new, end, n->end, mpol_new); |
1da177e4 | 2635 | n->end = start; |
5ca39575 | 2636 | sp_insert(sp, n_new); |
42288fe3 MG |
2637 | n_new = NULL; |
2638 | mpol_new = NULL; | |
1da177e4 LT |
2639 | break; |
2640 | } else | |
2641 | n->end = start; | |
2642 | } | |
2643 | if (!next) | |
2644 | break; | |
2645 | n = rb_entry(next, struct sp_node, nd); | |
2646 | } | |
2647 | if (new) | |
2648 | sp_insert(sp, new); | |
4a8c7bb5 | 2649 | write_unlock(&sp->lock); |
42288fe3 MG |
2650 | ret = 0; |
2651 | ||
2652 | err_out: | |
2653 | if (mpol_new) | |
2654 | mpol_put(mpol_new); | |
2655 | if (n_new) | |
2656 | kmem_cache_free(sn_cache, n_new); | |
2657 | ||
b22d127a | 2658 | return ret; |
42288fe3 MG |
2659 | |
2660 | alloc_new: | |
4a8c7bb5 | 2661 | write_unlock(&sp->lock); |
42288fe3 MG |
2662 | ret = -ENOMEM; |
2663 | n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
2664 | if (!n_new) | |
2665 | goto err_out; | |
2666 | mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2667 | if (!mpol_new) | |
2668 | goto err_out; | |
2669 | goto restart; | |
1da177e4 LT |
2670 | } |
2671 | ||
71fe804b LS |
2672 | /** |
2673 | * mpol_shared_policy_init - initialize shared policy for inode | |
2674 | * @sp: pointer to inode shared policy | |
2675 | * @mpol: struct mempolicy to install | |
2676 | * | |
2677 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2678 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2679 | * This must be released on exit. | |
4bfc4495 | 2680 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2681 | */ |
2682 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2683 | { | |
58568d2a MX |
2684 | int ret; |
2685 | ||
71fe804b | 2686 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
4a8c7bb5 | 2687 | rwlock_init(&sp->lock); |
71fe804b LS |
2688 | |
2689 | if (mpol) { | |
2690 | struct vm_area_struct pvma; | |
2691 | struct mempolicy *new; | |
4bfc4495 | 2692 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2693 | |
4bfc4495 | 2694 | if (!scratch) |
5c0c1654 | 2695 | goto put_mpol; |
71fe804b LS |
2696 | /* contextualize the tmpfs mount point mempolicy */ |
2697 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2698 | if (IS_ERR(new)) |
0cae3457 | 2699 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2700 | |
2701 | task_lock(current); | |
4bfc4495 | 2702 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2703 | task_unlock(current); |
15d77835 | 2704 | if (ret) |
5c0c1654 | 2705 | goto put_new; |
71fe804b LS |
2706 | |
2707 | /* Create pseudo-vma that contains just the policy */ | |
2c4541e2 | 2708 | vma_init(&pvma, NULL); |
71fe804b LS |
2709 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ |
2710 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2711 | |
5c0c1654 | 2712 | put_new: |
71fe804b | 2713 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2714 | free_scratch: |
4bfc4495 | 2715 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2716 | put_mpol: |
2717 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2718 | } |
2719 | } | |
2720 | ||
1da177e4 LT |
2721 | int mpol_set_shared_policy(struct shared_policy *info, |
2722 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2723 | { | |
2724 | int err; | |
2725 | struct sp_node *new = NULL; | |
2726 | unsigned long sz = vma_pages(vma); | |
2727 | ||
028fec41 | 2728 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2729 | vma->vm_pgoff, |
45c4745a | 2730 | sz, npol ? npol->mode : -1, |
028fec41 | 2731 | npol ? npol->flags : -1, |
00ef2d2f | 2732 | npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE); |
1da177e4 LT |
2733 | |
2734 | if (npol) { | |
2735 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2736 | if (!new) | |
2737 | return -ENOMEM; | |
2738 | } | |
2739 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2740 | if (err && new) | |
63f74ca2 | 2741 | sp_free(new); |
1da177e4 LT |
2742 | return err; |
2743 | } | |
2744 | ||
2745 | /* Free a backing policy store on inode delete. */ | |
2746 | void mpol_free_shared_policy(struct shared_policy *p) | |
2747 | { | |
2748 | struct sp_node *n; | |
2749 | struct rb_node *next; | |
2750 | ||
2751 | if (!p->root.rb_node) | |
2752 | return; | |
4a8c7bb5 | 2753 | write_lock(&p->lock); |
1da177e4 LT |
2754 | next = rb_first(&p->root); |
2755 | while (next) { | |
2756 | n = rb_entry(next, struct sp_node, nd); | |
2757 | next = rb_next(&n->nd); | |
63f74ca2 | 2758 | sp_delete(p, n); |
1da177e4 | 2759 | } |
4a8c7bb5 | 2760 | write_unlock(&p->lock); |
1da177e4 LT |
2761 | } |
2762 | ||
1a687c2e | 2763 | #ifdef CONFIG_NUMA_BALANCING |
c297663c | 2764 | static int __initdata numabalancing_override; |
1a687c2e MG |
2765 | |
2766 | static void __init check_numabalancing_enable(void) | |
2767 | { | |
2768 | bool numabalancing_default = false; | |
2769 | ||
2770 | if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED)) | |
2771 | numabalancing_default = true; | |
2772 | ||
c297663c MG |
2773 | /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */ |
2774 | if (numabalancing_override) | |
2775 | set_numabalancing_state(numabalancing_override == 1); | |
2776 | ||
b0dc2b9b | 2777 | if (num_online_nodes() > 1 && !numabalancing_override) { |
756a025f | 2778 | pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n", |
c297663c | 2779 | numabalancing_default ? "Enabling" : "Disabling"); |
1a687c2e MG |
2780 | set_numabalancing_state(numabalancing_default); |
2781 | } | |
2782 | } | |
2783 | ||
2784 | static int __init setup_numabalancing(char *str) | |
2785 | { | |
2786 | int ret = 0; | |
2787 | if (!str) | |
2788 | goto out; | |
1a687c2e MG |
2789 | |
2790 | if (!strcmp(str, "enable")) { | |
c297663c | 2791 | numabalancing_override = 1; |
1a687c2e MG |
2792 | ret = 1; |
2793 | } else if (!strcmp(str, "disable")) { | |
c297663c | 2794 | numabalancing_override = -1; |
1a687c2e MG |
2795 | ret = 1; |
2796 | } | |
2797 | out: | |
2798 | if (!ret) | |
4a404bea | 2799 | pr_warn("Unable to parse numa_balancing=\n"); |
1a687c2e MG |
2800 | |
2801 | return ret; | |
2802 | } | |
2803 | __setup("numa_balancing=", setup_numabalancing); | |
2804 | #else | |
2805 | static inline void __init check_numabalancing_enable(void) | |
2806 | { | |
2807 | } | |
2808 | #endif /* CONFIG_NUMA_BALANCING */ | |
2809 | ||
1da177e4 LT |
2810 | /* assumes fs == KERNEL_DS */ |
2811 | void __init numa_policy_init(void) | |
2812 | { | |
b71636e2 PM |
2813 | nodemask_t interleave_nodes; |
2814 | unsigned long largest = 0; | |
2815 | int nid, prefer = 0; | |
2816 | ||
1da177e4 LT |
2817 | policy_cache = kmem_cache_create("numa_policy", |
2818 | sizeof(struct mempolicy), | |
20c2df83 | 2819 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2820 | |
2821 | sn_cache = kmem_cache_create("shared_policy_node", | |
2822 | sizeof(struct sp_node), | |
20c2df83 | 2823 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2824 | |
5606e387 MG |
2825 | for_each_node(nid) { |
2826 | preferred_node_policy[nid] = (struct mempolicy) { | |
2827 | .refcnt = ATOMIC_INIT(1), | |
2828 | .mode = MPOL_PREFERRED, | |
2829 | .flags = MPOL_F_MOF | MPOL_F_MORON, | |
2830 | .v = { .preferred_node = nid, }, | |
2831 | }; | |
2832 | } | |
2833 | ||
b71636e2 PM |
2834 | /* |
2835 | * Set interleaving policy for system init. Interleaving is only | |
2836 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2837 | * fall back to the largest node if they're all smaller. | |
2838 | */ | |
2839 | nodes_clear(interleave_nodes); | |
01f13bd6 | 2840 | for_each_node_state(nid, N_MEMORY) { |
b71636e2 PM |
2841 | unsigned long total_pages = node_present_pages(nid); |
2842 | ||
2843 | /* Preserve the largest node */ | |
2844 | if (largest < total_pages) { | |
2845 | largest = total_pages; | |
2846 | prefer = nid; | |
2847 | } | |
2848 | ||
2849 | /* Interleave this node? */ | |
2850 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2851 | node_set(nid, interleave_nodes); | |
2852 | } | |
2853 | ||
2854 | /* All too small, use the largest */ | |
2855 | if (unlikely(nodes_empty(interleave_nodes))) | |
2856 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2857 | |
028fec41 | 2858 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
b1de0d13 | 2859 | pr_err("%s: interleaving failed\n", __func__); |
1a687c2e MG |
2860 | |
2861 | check_numabalancing_enable(); | |
1da177e4 LT |
2862 | } |
2863 | ||
8bccd85f | 2864 | /* Reset policy of current process to default */ |
1da177e4 LT |
2865 | void numa_default_policy(void) |
2866 | { | |
028fec41 | 2867 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2868 | } |
68860ec1 | 2869 | |
095f1fc4 LS |
2870 | /* |
2871 | * Parse and format mempolicy from/to strings | |
2872 | */ | |
2873 | ||
1a75a6c8 | 2874 | /* |
f2a07f40 | 2875 | * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag. |
1a75a6c8 | 2876 | */ |
345ace9c LS |
2877 | static const char * const policy_modes[] = |
2878 | { | |
2879 | [MPOL_DEFAULT] = "default", | |
2880 | [MPOL_PREFERRED] = "prefer", | |
2881 | [MPOL_BIND] = "bind", | |
2882 | [MPOL_INTERLEAVE] = "interleave", | |
d3a71033 | 2883 | [MPOL_LOCAL] = "local", |
345ace9c | 2884 | }; |
1a75a6c8 | 2885 | |
095f1fc4 LS |
2886 | |
2887 | #ifdef CONFIG_TMPFS | |
2888 | /** | |
f2a07f40 | 2889 | * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option. |
095f1fc4 | 2890 | * @str: string containing mempolicy to parse |
71fe804b | 2891 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
095f1fc4 LS |
2892 | * |
2893 | * Format of input: | |
2894 | * <mode>[=<flags>][:<nodelist>] | |
2895 | * | |
71fe804b | 2896 | * On success, returns 0, else 1 |
095f1fc4 | 2897 | */ |
a7a88b23 | 2898 | int mpol_parse_str(char *str, struct mempolicy **mpol) |
095f1fc4 | 2899 | { |
71fe804b | 2900 | struct mempolicy *new = NULL; |
f2a07f40 | 2901 | unsigned short mode_flags; |
71fe804b | 2902 | nodemask_t nodes; |
095f1fc4 LS |
2903 | char *nodelist = strchr(str, ':'); |
2904 | char *flags = strchr(str, '='); | |
dedf2c73 | 2905 | int err = 1, mode; |
095f1fc4 | 2906 | |
c7a91bc7 DC |
2907 | if (flags) |
2908 | *flags++ = '\0'; /* terminate mode string */ | |
2909 | ||
095f1fc4 LS |
2910 | if (nodelist) { |
2911 | /* NUL-terminate mode or flags string */ | |
2912 | *nodelist++ = '\0'; | |
71fe804b | 2913 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2914 | goto out; |
01f13bd6 | 2915 | if (!nodes_subset(nodes, node_states[N_MEMORY])) |
095f1fc4 | 2916 | goto out; |
71fe804b LS |
2917 | } else |
2918 | nodes_clear(nodes); | |
2919 | ||
dedf2c73 | 2920 | mode = match_string(policy_modes, MPOL_MAX, str); |
2921 | if (mode < 0) | |
095f1fc4 LS |
2922 | goto out; |
2923 | ||
71fe804b | 2924 | switch (mode) { |
095f1fc4 | 2925 | case MPOL_PREFERRED: |
71fe804b | 2926 | /* |
aa9f7d51 RD |
2927 | * Insist on a nodelist of one node only, although later |
2928 | * we use first_node(nodes) to grab a single node, so here | |
2929 | * nodelist (or nodes) cannot be empty. | |
71fe804b | 2930 | */ |
095f1fc4 LS |
2931 | if (nodelist) { |
2932 | char *rest = nodelist; | |
2933 | while (isdigit(*rest)) | |
2934 | rest++; | |
926f2ae0 KM |
2935 | if (*rest) |
2936 | goto out; | |
aa9f7d51 RD |
2937 | if (nodes_empty(nodes)) |
2938 | goto out; | |
095f1fc4 LS |
2939 | } |
2940 | break; | |
095f1fc4 LS |
2941 | case MPOL_INTERLEAVE: |
2942 | /* | |
2943 | * Default to online nodes with memory if no nodelist | |
2944 | */ | |
2945 | if (!nodelist) | |
01f13bd6 | 2946 | nodes = node_states[N_MEMORY]; |
3f226aa1 | 2947 | break; |
71fe804b | 2948 | case MPOL_LOCAL: |
3f226aa1 | 2949 | /* |
71fe804b | 2950 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 2951 | */ |
71fe804b | 2952 | if (nodelist) |
3f226aa1 | 2953 | goto out; |
71fe804b | 2954 | mode = MPOL_PREFERRED; |
3f226aa1 | 2955 | break; |
413b43de RT |
2956 | case MPOL_DEFAULT: |
2957 | /* | |
2958 | * Insist on a empty nodelist | |
2959 | */ | |
2960 | if (!nodelist) | |
2961 | err = 0; | |
2962 | goto out; | |
d69b2e63 KM |
2963 | case MPOL_BIND: |
2964 | /* | |
2965 | * Insist on a nodelist | |
2966 | */ | |
2967 | if (!nodelist) | |
2968 | goto out; | |
095f1fc4 LS |
2969 | } |
2970 | ||
71fe804b | 2971 | mode_flags = 0; |
095f1fc4 LS |
2972 | if (flags) { |
2973 | /* | |
2974 | * Currently, we only support two mutually exclusive | |
2975 | * mode flags. | |
2976 | */ | |
2977 | if (!strcmp(flags, "static")) | |
71fe804b | 2978 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 2979 | else if (!strcmp(flags, "relative")) |
71fe804b | 2980 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 2981 | else |
926f2ae0 | 2982 | goto out; |
095f1fc4 | 2983 | } |
71fe804b LS |
2984 | |
2985 | new = mpol_new(mode, mode_flags, &nodes); | |
2986 | if (IS_ERR(new)) | |
926f2ae0 KM |
2987 | goto out; |
2988 | ||
f2a07f40 HD |
2989 | /* |
2990 | * Save nodes for mpol_to_str() to show the tmpfs mount options | |
2991 | * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo. | |
2992 | */ | |
2993 | if (mode != MPOL_PREFERRED) | |
2994 | new->v.nodes = nodes; | |
2995 | else if (nodelist) | |
2996 | new->v.preferred_node = first_node(nodes); | |
2997 | else | |
2998 | new->flags |= MPOL_F_LOCAL; | |
2999 | ||
3000 | /* | |
3001 | * Save nodes for contextualization: this will be used to "clone" | |
3002 | * the mempolicy in a specific context [cpuset] at a later time. | |
3003 | */ | |
3004 | new->w.user_nodemask = nodes; | |
3005 | ||
926f2ae0 | 3006 | err = 0; |
71fe804b | 3007 | |
095f1fc4 LS |
3008 | out: |
3009 | /* Restore string for error message */ | |
3010 | if (nodelist) | |
3011 | *--nodelist = ':'; | |
3012 | if (flags) | |
3013 | *--flags = '='; | |
71fe804b LS |
3014 | if (!err) |
3015 | *mpol = new; | |
095f1fc4 LS |
3016 | return err; |
3017 | } | |
3018 | #endif /* CONFIG_TMPFS */ | |
3019 | ||
71fe804b LS |
3020 | /** |
3021 | * mpol_to_str - format a mempolicy structure for printing | |
3022 | * @buffer: to contain formatted mempolicy string | |
3023 | * @maxlen: length of @buffer | |
3024 | * @pol: pointer to mempolicy to be formatted | |
71fe804b | 3025 | * |
948927ee DR |
3026 | * Convert @pol into a string. If @buffer is too short, truncate the string. |
3027 | * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the | |
3028 | * longest flag, "relative", and to display at least a few node ids. | |
1a75a6c8 | 3029 | */ |
948927ee | 3030 | void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
1a75a6c8 CL |
3031 | { |
3032 | char *p = buffer; | |
948927ee DR |
3033 | nodemask_t nodes = NODE_MASK_NONE; |
3034 | unsigned short mode = MPOL_DEFAULT; | |
3035 | unsigned short flags = 0; | |
2291990a | 3036 | |
8790c71a | 3037 | if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) { |
bea904d5 | 3038 | mode = pol->mode; |
948927ee DR |
3039 | flags = pol->flags; |
3040 | } | |
bea904d5 | 3041 | |
1a75a6c8 CL |
3042 | switch (mode) { |
3043 | case MPOL_DEFAULT: | |
1a75a6c8 | 3044 | break; |
1a75a6c8 | 3045 | case MPOL_PREFERRED: |
fc36b8d3 | 3046 | if (flags & MPOL_F_LOCAL) |
f2a07f40 | 3047 | mode = MPOL_LOCAL; |
53f2556b | 3048 | else |
fc36b8d3 | 3049 | node_set(pol->v.preferred_node, nodes); |
1a75a6c8 | 3050 | break; |
1a75a6c8 | 3051 | case MPOL_BIND: |
1a75a6c8 | 3052 | case MPOL_INTERLEAVE: |
f2a07f40 | 3053 | nodes = pol->v.nodes; |
1a75a6c8 | 3054 | break; |
1a75a6c8 | 3055 | default: |
948927ee DR |
3056 | WARN_ON_ONCE(1); |
3057 | snprintf(p, maxlen, "unknown"); | |
3058 | return; | |
1a75a6c8 CL |
3059 | } |
3060 | ||
b7a9f420 | 3061 | p += snprintf(p, maxlen, "%s", policy_modes[mode]); |
1a75a6c8 | 3062 | |
fc36b8d3 | 3063 | if (flags & MPOL_MODE_FLAGS) { |
948927ee | 3064 | p += snprintf(p, buffer + maxlen - p, "="); |
f5b087b5 | 3065 | |
2291990a LS |
3066 | /* |
3067 | * Currently, the only defined flags are mutually exclusive | |
3068 | */ | |
f5b087b5 | 3069 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
3070 | p += snprintf(p, buffer + maxlen - p, "static"); |
3071 | else if (flags & MPOL_F_RELATIVE_NODES) | |
3072 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
3073 | } |
3074 | ||
9e763e0f TH |
3075 | if (!nodes_empty(nodes)) |
3076 | p += scnprintf(p, buffer + maxlen - p, ":%*pbl", | |
3077 | nodemask_pr_args(&nodes)); | |
1a75a6c8 | 3078 | } |