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 | |
f6e92f40 | 132 | * @node: Node id to start the search |
b2ca916c DW |
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 | |
c1e8d7c6 | 227 | * and mempolicy. May also be called holding the mmap_lock for write. |
58568d2a | 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 | * | |
c1e8d7c6 | 371 | * Per-vma policies are protected by mmap_lock. Allocations using per-task |
213980c0 VB |
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 | * | |
c1e8d7c6 | 401 | * Call holding a reference to mm. Takes mm->mmap_lock during call. |
1d0d2680 DR |
402 | */ |
403 | ||
404 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
405 | { | |
406 | struct vm_area_struct *vma; | |
407 | ||
d8ed45c5 | 408 | mmap_write_lock(mm); |
1d0d2680 | 409 | for (vma = mm->mmap; vma; vma = vma->vm_next) |
213980c0 | 410 | mpol_rebind_policy(vma->vm_policy, new); |
d8ed45c5 | 411 | mmap_write_unlock(mm); |
1d0d2680 DR |
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; |
3f088420 | 528 | pte_t *pte, *mapped_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 | |
3f088420 | 542 | mapped_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); |
6f4576e3 | 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 | 573 | } |
3f088420 | 574 | pte_unmap_unlock(mapped_pte, ptl); |
6f4576e3 | 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 | |
c1e8d7c6 | 767 | * This must be called with the mmap_lock held for writing. |
869833f2 KM |
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; | |
c1e8d7c6 | 792 | vma->vm_policy = new; /* protected by mmap_lock */ |
869833f2 KM |
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 | |
4bfc4495 | 878 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a | 879 | if (ret) { |
58568d2a | 880 | mpol_put(new); |
4bfc4495 | 881 | goto out; |
58568d2a | 882 | } |
78b132e9 | 883 | task_lock(current); |
58568d2a | 884 | old = current->mempolicy; |
1da177e4 | 885 | current->mempolicy = new; |
45816682 VB |
886 | if (new && new->mode == MPOL_INTERLEAVE) |
887 | current->il_prev = MAX_NUMNODES-1; | |
58568d2a | 888 | task_unlock(current); |
58568d2a | 889 | mpol_put(old); |
4bfc4495 KH |
890 | ret = 0; |
891 | out: | |
892 | NODEMASK_SCRATCH_FREE(scratch); | |
893 | return ret; | |
1da177e4 LT |
894 | } |
895 | ||
bea904d5 LS |
896 | /* |
897 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
898 | * |
899 | * Called with task's alloc_lock held | |
bea904d5 LS |
900 | */ |
901 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 902 | { |
dfcd3c0d | 903 | nodes_clear(*nodes); |
bea904d5 LS |
904 | if (p == &default_policy) |
905 | return; | |
906 | ||
45c4745a | 907 | switch (p->mode) { |
19770b32 | 908 | case MPOL_BIND: |
1da177e4 | 909 | case MPOL_INTERLEAVE: |
dfcd3c0d | 910 | *nodes = p->v.nodes; |
1da177e4 LT |
911 | break; |
912 | case MPOL_PREFERRED: | |
fc36b8d3 | 913 | if (!(p->flags & MPOL_F_LOCAL)) |
dfcd3c0d | 914 | node_set(p->v.preferred_node, *nodes); |
53f2556b | 915 | /* else return empty node mask for local allocation */ |
1da177e4 LT |
916 | break; |
917 | default: | |
918 | BUG(); | |
919 | } | |
920 | } | |
921 | ||
3b9aadf7 | 922 | static int lookup_node(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 923 | { |
ba841078 | 924 | struct page *p = NULL; |
1da177e4 LT |
925 | int err; |
926 | ||
3b9aadf7 AA |
927 | int locked = 1; |
928 | err = get_user_pages_locked(addr & PAGE_MASK, 1, 0, &p, &locked); | |
2d3a36a4 | 929 | if (err > 0) { |
1da177e4 LT |
930 | err = page_to_nid(p); |
931 | put_page(p); | |
932 | } | |
3b9aadf7 | 933 | if (locked) |
d8ed45c5 | 934 | mmap_read_unlock(mm); |
1da177e4 LT |
935 | return err; |
936 | } | |
937 | ||
1da177e4 | 938 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
939 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
940 | unsigned long addr, unsigned long flags) | |
1da177e4 | 941 | { |
8bccd85f | 942 | int err; |
1da177e4 LT |
943 | struct mm_struct *mm = current->mm; |
944 | struct vm_area_struct *vma = NULL; | |
3b9aadf7 | 945 | struct mempolicy *pol = current->mempolicy, *pol_refcount = NULL; |
1da177e4 | 946 | |
754af6f5 LS |
947 | if (flags & |
948 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 949 | return -EINVAL; |
754af6f5 LS |
950 | |
951 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
952 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
953 | return -EINVAL; | |
954 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 955 | task_lock(current); |
754af6f5 | 956 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 957 | task_unlock(current); |
754af6f5 LS |
958 | return 0; |
959 | } | |
960 | ||
1da177e4 | 961 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
962 | /* |
963 | * Do NOT fall back to task policy if the | |
964 | * vma/shared policy at addr is NULL. We | |
965 | * want to return MPOL_DEFAULT in this case. | |
966 | */ | |
d8ed45c5 | 967 | mmap_read_lock(mm); |
1da177e4 LT |
968 | vma = find_vma_intersection(mm, addr, addr+1); |
969 | if (!vma) { | |
d8ed45c5 | 970 | mmap_read_unlock(mm); |
1da177e4 LT |
971 | return -EFAULT; |
972 | } | |
973 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
974 | pol = vma->vm_ops->get_policy(vma, addr); | |
975 | else | |
976 | pol = vma->vm_policy; | |
977 | } else if (addr) | |
978 | return -EINVAL; | |
979 | ||
980 | if (!pol) | |
bea904d5 | 981 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
982 | |
983 | if (flags & MPOL_F_NODE) { | |
984 | if (flags & MPOL_F_ADDR) { | |
3b9aadf7 AA |
985 | /* |
986 | * Take a refcount on the mpol, lookup_node() | |
c1e8d7c6 | 987 | * wil drop the mmap_lock, so after calling |
3b9aadf7 AA |
988 | * lookup_node() only "pol" remains valid, "vma" |
989 | * is stale. | |
990 | */ | |
991 | pol_refcount = pol; | |
992 | vma = NULL; | |
993 | mpol_get(pol); | |
994 | err = lookup_node(mm, addr); | |
1da177e4 LT |
995 | if (err < 0) |
996 | goto out; | |
8bccd85f | 997 | *policy = err; |
1da177e4 | 998 | } else if (pol == current->mempolicy && |
45c4745a | 999 | pol->mode == MPOL_INTERLEAVE) { |
45816682 | 1000 | *policy = next_node_in(current->il_prev, pol->v.nodes); |
1da177e4 LT |
1001 | } else { |
1002 | err = -EINVAL; | |
1003 | goto out; | |
1004 | } | |
bea904d5 LS |
1005 | } else { |
1006 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
1007 | pol->mode; | |
d79df630 DR |
1008 | /* |
1009 | * Internal mempolicy flags must be masked off before exposing | |
1010 | * the policy to userspace. | |
1011 | */ | |
1012 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 1013 | } |
1da177e4 | 1014 | |
1da177e4 | 1015 | err = 0; |
58568d2a | 1016 | if (nmask) { |
c6b6ef8b LS |
1017 | if (mpol_store_user_nodemask(pol)) { |
1018 | *nmask = pol->w.user_nodemask; | |
1019 | } else { | |
1020 | task_lock(current); | |
1021 | get_policy_nodemask(pol, nmask); | |
1022 | task_unlock(current); | |
1023 | } | |
58568d2a | 1024 | } |
1da177e4 LT |
1025 | |
1026 | out: | |
52cd3b07 | 1027 | mpol_cond_put(pol); |
1da177e4 | 1028 | if (vma) |
d8ed45c5 | 1029 | mmap_read_unlock(mm); |
3b9aadf7 AA |
1030 | if (pol_refcount) |
1031 | mpol_put(pol_refcount); | |
1da177e4 LT |
1032 | return err; |
1033 | } | |
1034 | ||
b20a3503 | 1035 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 | 1036 | /* |
c8633798 | 1037 | * page migration, thp tail pages can be passed. |
6ce3c4c0 | 1038 | */ |
a53190a4 | 1039 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
fc301289 | 1040 | unsigned long flags) |
6ce3c4c0 | 1041 | { |
c8633798 | 1042 | struct page *head = compound_head(page); |
6ce3c4c0 | 1043 | /* |
fc301289 | 1044 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 1045 | */ |
c8633798 NH |
1046 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(head) == 1) { |
1047 | if (!isolate_lru_page(head)) { | |
1048 | list_add_tail(&head->lru, pagelist); | |
1049 | mod_node_page_state(page_pgdat(head), | |
9de4f22a | 1050 | NR_ISOLATED_ANON + page_is_file_lru(head), |
6c357848 | 1051 | thp_nr_pages(head)); |
a53190a4 YS |
1052 | } else if (flags & MPOL_MF_STRICT) { |
1053 | /* | |
1054 | * Non-movable page may reach here. And, there may be | |
1055 | * temporary off LRU pages or non-LRU movable pages. | |
1056 | * Treat them as unmovable pages since they can't be | |
1057 | * isolated, so they can't be moved at the moment. It | |
1058 | * should return -EIO for this case too. | |
1059 | */ | |
1060 | return -EIO; | |
62695a84 NP |
1061 | } |
1062 | } | |
a53190a4 YS |
1063 | |
1064 | return 0; | |
7e2ab150 | 1065 | } |
6ce3c4c0 | 1066 | |
7e2ab150 CL |
1067 | /* |
1068 | * Migrate pages from one node to a target node. | |
1069 | * Returns error or the number of pages not migrated. | |
1070 | */ | |
dbcb0f19 AB |
1071 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
1072 | int flags) | |
7e2ab150 CL |
1073 | { |
1074 | nodemask_t nmask; | |
1075 | LIST_HEAD(pagelist); | |
1076 | int err = 0; | |
a0976311 JK |
1077 | struct migration_target_control mtc = { |
1078 | .nid = dest, | |
1079 | .gfp_mask = GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, | |
1080 | }; | |
7e2ab150 CL |
1081 | |
1082 | nodes_clear(nmask); | |
1083 | node_set(source, nmask); | |
6ce3c4c0 | 1084 | |
08270807 MK |
1085 | /* |
1086 | * This does not "check" the range but isolates all pages that | |
1087 | * need migration. Between passing in the full user address | |
1088 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. | |
1089 | */ | |
1090 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); | |
98094945 | 1091 | queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask, |
7e2ab150 CL |
1092 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
1093 | ||
cf608ac1 | 1094 | if (!list_empty(&pagelist)) { |
a0976311 JK |
1095 | err = migrate_pages(&pagelist, alloc_migration_target, NULL, |
1096 | (unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL); | |
cf608ac1 | 1097 | if (err) |
e2d8cf40 | 1098 | putback_movable_pages(&pagelist); |
cf608ac1 | 1099 | } |
95a402c3 | 1100 | |
7e2ab150 | 1101 | return err; |
6ce3c4c0 CL |
1102 | } |
1103 | ||
39743889 | 1104 | /* |
7e2ab150 CL |
1105 | * Move pages between the two nodesets so as to preserve the physical |
1106 | * layout as much as possible. | |
39743889 CL |
1107 | * |
1108 | * Returns the number of page that could not be moved. | |
1109 | */ | |
0ce72d4f AM |
1110 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1111 | const nodemask_t *to, int flags) | |
39743889 | 1112 | { |
7e2ab150 | 1113 | int busy = 0; |
0aedadf9 | 1114 | int err; |
7e2ab150 | 1115 | nodemask_t tmp; |
39743889 | 1116 | |
0aedadf9 CL |
1117 | err = migrate_prep(); |
1118 | if (err) | |
1119 | return err; | |
1120 | ||
d8ed45c5 | 1121 | mmap_read_lock(mm); |
39743889 | 1122 | |
da0aa138 KM |
1123 | /* |
1124 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1125 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1126 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1127 | * The pair of nodemasks 'to' and 'from' define the map. | |
1128 | * | |
1129 | * If no pair of bits is found that way, fallback to picking some | |
1130 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1131 | * 'source' and 'dest' bits are the same, this represents a node | |
1132 | * that will be migrating to itself, so no pages need move. | |
1133 | * | |
1134 | * If no bits are left in 'tmp', or if all remaining bits left | |
1135 | * in 'tmp' correspond to the same bit in 'to', return false | |
1136 | * (nothing left to migrate). | |
1137 | * | |
1138 | * This lets us pick a pair of nodes to migrate between, such that | |
1139 | * if possible the dest node is not already occupied by some other | |
1140 | * source node, minimizing the risk of overloading the memory on a | |
1141 | * node that would happen if we migrated incoming memory to a node | |
1142 | * before migrating outgoing memory source that same node. | |
1143 | * | |
1144 | * A single scan of tmp is sufficient. As we go, we remember the | |
1145 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1146 | * that not only moved, but what's better, moved to an empty slot | |
1147 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1148 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1149 | * most recent <s, d> pair that moved. If we get all the way through |
1150 | * the scan of tmp without finding any node that moved, much less | |
1151 | * moved to an empty node, then there is nothing left worth migrating. | |
1152 | */ | |
d4984711 | 1153 | |
0ce72d4f | 1154 | tmp = *from; |
7e2ab150 CL |
1155 | while (!nodes_empty(tmp)) { |
1156 | int s,d; | |
b76ac7e7 | 1157 | int source = NUMA_NO_NODE; |
7e2ab150 CL |
1158 | int dest = 0; |
1159 | ||
1160 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1161 | |
1162 | /* | |
1163 | * do_migrate_pages() tries to maintain the relative | |
1164 | * node relationship of the pages established between | |
1165 | * threads and memory areas. | |
1166 | * | |
1167 | * However if the number of source nodes is not equal to | |
1168 | * the number of destination nodes we can not preserve | |
1169 | * this node relative relationship. In that case, skip | |
1170 | * copying memory from a node that is in the destination | |
1171 | * mask. | |
1172 | * | |
1173 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1174 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1175 | */ | |
1176 | ||
0ce72d4f AM |
1177 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1178 | (node_isset(s, *to))) | |
4a5b18cc LW |
1179 | continue; |
1180 | ||
0ce72d4f | 1181 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1182 | if (s == d) |
1183 | continue; | |
1184 | ||
1185 | source = s; /* Node moved. Memorize */ | |
1186 | dest = d; | |
1187 | ||
1188 | /* dest not in remaining from nodes? */ | |
1189 | if (!node_isset(dest, tmp)) | |
1190 | break; | |
1191 | } | |
b76ac7e7 | 1192 | if (source == NUMA_NO_NODE) |
7e2ab150 CL |
1193 | break; |
1194 | ||
1195 | node_clear(source, tmp); | |
1196 | err = migrate_to_node(mm, source, dest, flags); | |
1197 | if (err > 0) | |
1198 | busy += err; | |
1199 | if (err < 0) | |
1200 | break; | |
39743889 | 1201 | } |
d8ed45c5 | 1202 | mmap_read_unlock(mm); |
7e2ab150 CL |
1203 | if (err < 0) |
1204 | return err; | |
1205 | return busy; | |
b20a3503 CL |
1206 | |
1207 | } | |
1208 | ||
3ad33b24 LS |
1209 | /* |
1210 | * Allocate a new page for page migration based on vma policy. | |
d05f0cdc | 1211 | * Start by assuming the page is mapped by the same vma as contains @start. |
3ad33b24 LS |
1212 | * Search forward from there, if not. N.B., this assumes that the |
1213 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1214 | * is in virtual address order. | |
1215 | */ | |
666feb21 | 1216 | static struct page *new_page(struct page *page, unsigned long start) |
95a402c3 | 1217 | { |
d05f0cdc | 1218 | struct vm_area_struct *vma; |
3f649ab7 | 1219 | unsigned long address; |
95a402c3 | 1220 | |
d05f0cdc | 1221 | vma = find_vma(current->mm, start); |
3ad33b24 LS |
1222 | while (vma) { |
1223 | address = page_address_in_vma(page, vma); | |
1224 | if (address != -EFAULT) | |
1225 | break; | |
1226 | vma = vma->vm_next; | |
1227 | } | |
11c731e8 WL |
1228 | |
1229 | if (PageHuge(page)) { | |
389c8178 MH |
1230 | return alloc_huge_page_vma(page_hstate(compound_head(page)), |
1231 | vma, address); | |
94723aaf | 1232 | } else if (PageTransHuge(page)) { |
c8633798 NH |
1233 | struct page *thp; |
1234 | ||
19deb769 DR |
1235 | thp = alloc_hugepage_vma(GFP_TRANSHUGE, vma, address, |
1236 | HPAGE_PMD_ORDER); | |
c8633798 NH |
1237 | if (!thp) |
1238 | return NULL; | |
1239 | prep_transhuge_page(thp); | |
1240 | return thp; | |
11c731e8 | 1241 | } |
0bf598d8 | 1242 | /* |
11c731e8 | 1243 | * if !vma, alloc_page_vma() will use task or system default policy |
0bf598d8 | 1244 | */ |
0f556856 MH |
1245 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE | __GFP_RETRY_MAYFAIL, |
1246 | vma, address); | |
95a402c3 | 1247 | } |
b20a3503 CL |
1248 | #else |
1249 | ||
a53190a4 | 1250 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
b20a3503 CL |
1251 | unsigned long flags) |
1252 | { | |
a53190a4 | 1253 | return -EIO; |
39743889 CL |
1254 | } |
1255 | ||
0ce72d4f AM |
1256 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1257 | const nodemask_t *to, int flags) | |
b20a3503 CL |
1258 | { |
1259 | return -ENOSYS; | |
1260 | } | |
95a402c3 | 1261 | |
666feb21 | 1262 | static struct page *new_page(struct page *page, unsigned long start) |
95a402c3 CL |
1263 | { |
1264 | return NULL; | |
1265 | } | |
b20a3503 CL |
1266 | #endif |
1267 | ||
dbcb0f19 | 1268 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1269 | unsigned short mode, unsigned short mode_flags, |
1270 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 | 1271 | { |
6ce3c4c0 CL |
1272 | struct mm_struct *mm = current->mm; |
1273 | struct mempolicy *new; | |
1274 | unsigned long end; | |
1275 | int err; | |
d8835445 | 1276 | int ret; |
6ce3c4c0 CL |
1277 | LIST_HEAD(pagelist); |
1278 | ||
b24f53a0 | 1279 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
6ce3c4c0 | 1280 | return -EINVAL; |
74c00241 | 1281 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1282 | return -EPERM; |
1283 | ||
1284 | if (start & ~PAGE_MASK) | |
1285 | return -EINVAL; | |
1286 | ||
1287 | if (mode == MPOL_DEFAULT) | |
1288 | flags &= ~MPOL_MF_STRICT; | |
1289 | ||
1290 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
1291 | end = start + len; | |
1292 | ||
1293 | if (end < start) | |
1294 | return -EINVAL; | |
1295 | if (end == start) | |
1296 | return 0; | |
1297 | ||
028fec41 | 1298 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1299 | if (IS_ERR(new)) |
1300 | return PTR_ERR(new); | |
1301 | ||
b24f53a0 LS |
1302 | if (flags & MPOL_MF_LAZY) |
1303 | new->flags |= MPOL_F_MOF; | |
1304 | ||
6ce3c4c0 CL |
1305 | /* |
1306 | * If we are using the default policy then operation | |
1307 | * on discontinuous address spaces is okay after all | |
1308 | */ | |
1309 | if (!new) | |
1310 | flags |= MPOL_MF_DISCONTIG_OK; | |
1311 | ||
028fec41 DR |
1312 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1313 | start, start + len, mode, mode_flags, | |
00ef2d2f | 1314 | nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE); |
6ce3c4c0 | 1315 | |
0aedadf9 CL |
1316 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1317 | ||
1318 | err = migrate_prep(); | |
1319 | if (err) | |
b05ca738 | 1320 | goto mpol_out; |
0aedadf9 | 1321 | } |
4bfc4495 KH |
1322 | { |
1323 | NODEMASK_SCRATCH(scratch); | |
1324 | if (scratch) { | |
d8ed45c5 | 1325 | mmap_write_lock(mm); |
4bfc4495 | 1326 | err = mpol_set_nodemask(new, nmask, scratch); |
4bfc4495 | 1327 | if (err) |
d8ed45c5 | 1328 | mmap_write_unlock(mm); |
4bfc4495 KH |
1329 | } else |
1330 | err = -ENOMEM; | |
1331 | NODEMASK_SCRATCH_FREE(scratch); | |
1332 | } | |
b05ca738 KM |
1333 | if (err) |
1334 | goto mpol_out; | |
1335 | ||
d8835445 | 1336 | ret = queue_pages_range(mm, start, end, nmask, |
6ce3c4c0 | 1337 | flags | MPOL_MF_INVERT, &pagelist); |
d8835445 YS |
1338 | |
1339 | if (ret < 0) { | |
a85dfc30 | 1340 | err = ret; |
d8835445 YS |
1341 | goto up_out; |
1342 | } | |
1343 | ||
1344 | err = mbind_range(mm, start, end, new); | |
7e2ab150 | 1345 | |
b24f53a0 LS |
1346 | if (!err) { |
1347 | int nr_failed = 0; | |
1348 | ||
cf608ac1 | 1349 | if (!list_empty(&pagelist)) { |
b24f53a0 | 1350 | WARN_ON_ONCE(flags & MPOL_MF_LAZY); |
d05f0cdc HD |
1351 | nr_failed = migrate_pages(&pagelist, new_page, NULL, |
1352 | start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND); | |
cf608ac1 | 1353 | if (nr_failed) |
74060e4d | 1354 | putback_movable_pages(&pagelist); |
cf608ac1 | 1355 | } |
6ce3c4c0 | 1356 | |
d8835445 | 1357 | if ((ret > 0) || (nr_failed && (flags & MPOL_MF_STRICT))) |
6ce3c4c0 | 1358 | err = -EIO; |
a85dfc30 | 1359 | } else { |
d8835445 | 1360 | up_out: |
a85dfc30 YS |
1361 | if (!list_empty(&pagelist)) |
1362 | putback_movable_pages(&pagelist); | |
1363 | } | |
1364 | ||
d8ed45c5 | 1365 | mmap_write_unlock(mm); |
d8835445 | 1366 | mpol_out: |
f0be3d32 | 1367 | mpol_put(new); |
6ce3c4c0 CL |
1368 | return err; |
1369 | } | |
1370 | ||
8bccd85f CL |
1371 | /* |
1372 | * User space interface with variable sized bitmaps for nodelists. | |
1373 | */ | |
1374 | ||
1375 | /* Copy a node mask from user space. */ | |
39743889 | 1376 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1377 | unsigned long maxnode) |
1378 | { | |
1379 | unsigned long k; | |
56521e7a | 1380 | unsigned long t; |
8bccd85f CL |
1381 | unsigned long nlongs; |
1382 | unsigned long endmask; | |
1383 | ||
1384 | --maxnode; | |
1385 | nodes_clear(*nodes); | |
1386 | if (maxnode == 0 || !nmask) | |
1387 | return 0; | |
a9c930ba | 1388 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1389 | return -EINVAL; |
8bccd85f CL |
1390 | |
1391 | nlongs = BITS_TO_LONGS(maxnode); | |
1392 | if ((maxnode % BITS_PER_LONG) == 0) | |
1393 | endmask = ~0UL; | |
1394 | else | |
1395 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1396 | ||
56521e7a YX |
1397 | /* |
1398 | * When the user specified more nodes than supported just check | |
1399 | * if the non supported part is all zero. | |
1400 | * | |
1401 | * If maxnode have more longs than MAX_NUMNODES, check | |
1402 | * the bits in that area first. And then go through to | |
1403 | * check the rest bits which equal or bigger than MAX_NUMNODES. | |
1404 | * Otherwise, just check bits [MAX_NUMNODES, maxnode). | |
1405 | */ | |
8bccd85f | 1406 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { |
8bccd85f | 1407 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { |
8bccd85f CL |
1408 | if (get_user(t, nmask + k)) |
1409 | return -EFAULT; | |
1410 | if (k == nlongs - 1) { | |
1411 | if (t & endmask) | |
1412 | return -EINVAL; | |
1413 | } else if (t) | |
1414 | return -EINVAL; | |
1415 | } | |
1416 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1417 | endmask = ~0UL; | |
1418 | } | |
1419 | ||
56521e7a YX |
1420 | if (maxnode > MAX_NUMNODES && MAX_NUMNODES % BITS_PER_LONG != 0) { |
1421 | unsigned long valid_mask = endmask; | |
1422 | ||
1423 | valid_mask &= ~((1UL << (MAX_NUMNODES % BITS_PER_LONG)) - 1); | |
1424 | if (get_user(t, nmask + nlongs - 1)) | |
1425 | return -EFAULT; | |
1426 | if (t & valid_mask) | |
1427 | return -EINVAL; | |
1428 | } | |
1429 | ||
8bccd85f CL |
1430 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) |
1431 | return -EFAULT; | |
1432 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1433 | return 0; | |
1434 | } | |
1435 | ||
1436 | /* Copy a kernel node mask to user space */ | |
1437 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1438 | nodemask_t *nodes) | |
1439 | { | |
1440 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
050c17f2 | 1441 | unsigned int nbytes = BITS_TO_LONGS(nr_node_ids) * sizeof(long); |
8bccd85f CL |
1442 | |
1443 | if (copy > nbytes) { | |
1444 | if (copy > PAGE_SIZE) | |
1445 | return -EINVAL; | |
1446 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1447 | return -EFAULT; | |
1448 | copy = nbytes; | |
1449 | } | |
1450 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1451 | } | |
1452 | ||
e7dc9ad6 DB |
1453 | static long kernel_mbind(unsigned long start, unsigned long len, |
1454 | unsigned long mode, const unsigned long __user *nmask, | |
1455 | unsigned long maxnode, unsigned int flags) | |
8bccd85f CL |
1456 | { |
1457 | nodemask_t nodes; | |
1458 | int err; | |
028fec41 | 1459 | unsigned short mode_flags; |
8bccd85f | 1460 | |
057d3389 | 1461 | start = untagged_addr(start); |
028fec41 DR |
1462 | mode_flags = mode & MPOL_MODE_FLAGS; |
1463 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
1464 | if (mode >= MPOL_MAX) |
1465 | return -EINVAL; | |
4c50bc01 DR |
1466 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
1467 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
1468 | return -EINVAL; | |
8bccd85f CL |
1469 | err = get_nodes(&nodes, nmask, maxnode); |
1470 | if (err) | |
1471 | return err; | |
028fec41 | 1472 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
1473 | } |
1474 | ||
e7dc9ad6 DB |
1475 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
1476 | unsigned long, mode, const unsigned long __user *, nmask, | |
1477 | unsigned long, maxnode, unsigned int, flags) | |
1478 | { | |
1479 | return kernel_mbind(start, len, mode, nmask, maxnode, flags); | |
1480 | } | |
1481 | ||
8bccd85f | 1482 | /* Set the process memory policy */ |
af03c4ac DB |
1483 | static long kernel_set_mempolicy(int mode, const unsigned long __user *nmask, |
1484 | unsigned long maxnode) | |
8bccd85f CL |
1485 | { |
1486 | int err; | |
1487 | nodemask_t nodes; | |
028fec41 | 1488 | unsigned short flags; |
8bccd85f | 1489 | |
028fec41 DR |
1490 | flags = mode & MPOL_MODE_FLAGS; |
1491 | mode &= ~MPOL_MODE_FLAGS; | |
1492 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 1493 | return -EINVAL; |
4c50bc01 DR |
1494 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
1495 | return -EINVAL; | |
8bccd85f CL |
1496 | err = get_nodes(&nodes, nmask, maxnode); |
1497 | if (err) | |
1498 | return err; | |
028fec41 | 1499 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1500 | } |
1501 | ||
af03c4ac DB |
1502 | SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask, |
1503 | unsigned long, maxnode) | |
1504 | { | |
1505 | return kernel_set_mempolicy(mode, nmask, maxnode); | |
1506 | } | |
1507 | ||
b6e9b0ba DB |
1508 | static int kernel_migrate_pages(pid_t pid, unsigned long maxnode, |
1509 | const unsigned long __user *old_nodes, | |
1510 | const unsigned long __user *new_nodes) | |
39743889 | 1511 | { |
596d7cfa | 1512 | struct mm_struct *mm = NULL; |
39743889 | 1513 | struct task_struct *task; |
39743889 CL |
1514 | nodemask_t task_nodes; |
1515 | int err; | |
596d7cfa KM |
1516 | nodemask_t *old; |
1517 | nodemask_t *new; | |
1518 | NODEMASK_SCRATCH(scratch); | |
1519 | ||
1520 | if (!scratch) | |
1521 | return -ENOMEM; | |
39743889 | 1522 | |
596d7cfa KM |
1523 | old = &scratch->mask1; |
1524 | new = &scratch->mask2; | |
1525 | ||
1526 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1527 | if (err) |
596d7cfa | 1528 | goto out; |
39743889 | 1529 | |
596d7cfa | 1530 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1531 | if (err) |
596d7cfa | 1532 | goto out; |
39743889 CL |
1533 | |
1534 | /* Find the mm_struct */ | |
55cfaa3c | 1535 | rcu_read_lock(); |
228ebcbe | 1536 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1537 | if (!task) { |
55cfaa3c | 1538 | rcu_read_unlock(); |
596d7cfa KM |
1539 | err = -ESRCH; |
1540 | goto out; | |
39743889 | 1541 | } |
3268c63e | 1542 | get_task_struct(task); |
39743889 | 1543 | |
596d7cfa | 1544 | err = -EINVAL; |
39743889 CL |
1545 | |
1546 | /* | |
31367466 OE |
1547 | * Check if this process has the right to modify the specified process. |
1548 | * Use the regular "ptrace_may_access()" checks. | |
39743889 | 1549 | */ |
31367466 | 1550 | if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) { |
c69e8d9c | 1551 | rcu_read_unlock(); |
39743889 | 1552 | err = -EPERM; |
3268c63e | 1553 | goto out_put; |
39743889 | 1554 | } |
c69e8d9c | 1555 | rcu_read_unlock(); |
39743889 CL |
1556 | |
1557 | task_nodes = cpuset_mems_allowed(task); | |
1558 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1559 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1560 | err = -EPERM; |
3268c63e | 1561 | goto out_put; |
39743889 CL |
1562 | } |
1563 | ||
0486a38b YX |
1564 | task_nodes = cpuset_mems_allowed(current); |
1565 | nodes_and(*new, *new, task_nodes); | |
1566 | if (nodes_empty(*new)) | |
1567 | goto out_put; | |
1568 | ||
86c3a764 DQ |
1569 | err = security_task_movememory(task); |
1570 | if (err) | |
3268c63e | 1571 | goto out_put; |
86c3a764 | 1572 | |
3268c63e CL |
1573 | mm = get_task_mm(task); |
1574 | put_task_struct(task); | |
f2a9ef88 SL |
1575 | |
1576 | if (!mm) { | |
3268c63e | 1577 | err = -EINVAL; |
f2a9ef88 SL |
1578 | goto out; |
1579 | } | |
1580 | ||
1581 | err = do_migrate_pages(mm, old, new, | |
1582 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1583 | |
1584 | mmput(mm); | |
1585 | out: | |
596d7cfa KM |
1586 | NODEMASK_SCRATCH_FREE(scratch); |
1587 | ||
39743889 | 1588 | return err; |
3268c63e CL |
1589 | |
1590 | out_put: | |
1591 | put_task_struct(task); | |
1592 | goto out; | |
1593 | ||
39743889 CL |
1594 | } |
1595 | ||
b6e9b0ba DB |
1596 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1597 | const unsigned long __user *, old_nodes, | |
1598 | const unsigned long __user *, new_nodes) | |
1599 | { | |
1600 | return kernel_migrate_pages(pid, maxnode, old_nodes, new_nodes); | |
1601 | } | |
1602 | ||
39743889 | 1603 | |
8bccd85f | 1604 | /* Retrieve NUMA policy */ |
af03c4ac DB |
1605 | static int kernel_get_mempolicy(int __user *policy, |
1606 | unsigned long __user *nmask, | |
1607 | unsigned long maxnode, | |
1608 | unsigned long addr, | |
1609 | unsigned long flags) | |
8bccd85f | 1610 | { |
dbcb0f19 | 1611 | int err; |
3f649ab7 | 1612 | int pval; |
8bccd85f CL |
1613 | nodemask_t nodes; |
1614 | ||
050c17f2 | 1615 | if (nmask != NULL && maxnode < nr_node_ids) |
8bccd85f CL |
1616 | return -EINVAL; |
1617 | ||
4605f057 WH |
1618 | addr = untagged_addr(addr); |
1619 | ||
8bccd85f CL |
1620 | err = do_get_mempolicy(&pval, &nodes, addr, flags); |
1621 | ||
1622 | if (err) | |
1623 | return err; | |
1624 | ||
1625 | if (policy && put_user(pval, policy)) | |
1626 | return -EFAULT; | |
1627 | ||
1628 | if (nmask) | |
1629 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1630 | ||
1631 | return err; | |
1632 | } | |
1633 | ||
af03c4ac DB |
1634 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1635 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1636 | unsigned long, addr, unsigned long, flags) | |
1637 | { | |
1638 | return kernel_get_mempolicy(policy, nmask, maxnode, addr, flags); | |
1639 | } | |
1640 | ||
1da177e4 LT |
1641 | #ifdef CONFIG_COMPAT |
1642 | ||
c93e0f6c HC |
1643 | COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1644 | compat_ulong_t __user *, nmask, | |
1645 | compat_ulong_t, maxnode, | |
1646 | compat_ulong_t, addr, compat_ulong_t, flags) | |
1da177e4 LT |
1647 | { |
1648 | long err; | |
1649 | unsigned long __user *nm = NULL; | |
1650 | unsigned long nr_bits, alloc_size; | |
1651 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1652 | ||
050c17f2 | 1653 | nr_bits = min_t(unsigned long, maxnode-1, nr_node_ids); |
1da177e4 LT |
1654 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; |
1655 | ||
1656 | if (nmask) | |
1657 | nm = compat_alloc_user_space(alloc_size); | |
1658 | ||
af03c4ac | 1659 | err = kernel_get_mempolicy(policy, nm, nr_bits+1, addr, flags); |
1da177e4 LT |
1660 | |
1661 | if (!err && nmask) { | |
2bbff6c7 KH |
1662 | unsigned long copy_size; |
1663 | copy_size = min_t(unsigned long, sizeof(bm), alloc_size); | |
1664 | err = copy_from_user(bm, nm, copy_size); | |
1da177e4 LT |
1665 | /* ensure entire bitmap is zeroed */ |
1666 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1667 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1668 | } | |
1669 | ||
1670 | return err; | |
1671 | } | |
1672 | ||
c93e0f6c HC |
1673 | COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask, |
1674 | compat_ulong_t, maxnode) | |
1da177e4 | 1675 | { |
1da177e4 LT |
1676 | unsigned long __user *nm = NULL; |
1677 | unsigned long nr_bits, alloc_size; | |
1678 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1679 | ||
1680 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1681 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1682 | ||
1683 | if (nmask) { | |
cf01fb99 CS |
1684 | if (compat_get_bitmap(bm, nmask, nr_bits)) |
1685 | return -EFAULT; | |
1da177e4 | 1686 | nm = compat_alloc_user_space(alloc_size); |
cf01fb99 CS |
1687 | if (copy_to_user(nm, bm, alloc_size)) |
1688 | return -EFAULT; | |
1da177e4 LT |
1689 | } |
1690 | ||
af03c4ac | 1691 | return kernel_set_mempolicy(mode, nm, nr_bits+1); |
1da177e4 LT |
1692 | } |
1693 | ||
c93e0f6c HC |
1694 | COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len, |
1695 | compat_ulong_t, mode, compat_ulong_t __user *, nmask, | |
1696 | compat_ulong_t, maxnode, compat_ulong_t, flags) | |
1da177e4 | 1697 | { |
1da177e4 LT |
1698 | unsigned long __user *nm = NULL; |
1699 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1700 | nodemask_t bm; |
1da177e4 LT |
1701 | |
1702 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1703 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1704 | ||
1705 | if (nmask) { | |
cf01fb99 CS |
1706 | if (compat_get_bitmap(nodes_addr(bm), nmask, nr_bits)) |
1707 | return -EFAULT; | |
1da177e4 | 1708 | nm = compat_alloc_user_space(alloc_size); |
cf01fb99 CS |
1709 | if (copy_to_user(nm, nodes_addr(bm), alloc_size)) |
1710 | return -EFAULT; | |
1da177e4 LT |
1711 | } |
1712 | ||
e7dc9ad6 | 1713 | return kernel_mbind(start, len, mode, nm, nr_bits+1, flags); |
1da177e4 LT |
1714 | } |
1715 | ||
b6e9b0ba DB |
1716 | COMPAT_SYSCALL_DEFINE4(migrate_pages, compat_pid_t, pid, |
1717 | compat_ulong_t, maxnode, | |
1718 | const compat_ulong_t __user *, old_nodes, | |
1719 | const compat_ulong_t __user *, new_nodes) | |
1720 | { | |
1721 | unsigned long __user *old = NULL; | |
1722 | unsigned long __user *new = NULL; | |
1723 | nodemask_t tmp_mask; | |
1724 | unsigned long nr_bits; | |
1725 | unsigned long size; | |
1726 | ||
1727 | nr_bits = min_t(unsigned long, maxnode - 1, MAX_NUMNODES); | |
1728 | size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1729 | if (old_nodes) { | |
1730 | if (compat_get_bitmap(nodes_addr(tmp_mask), old_nodes, nr_bits)) | |
1731 | return -EFAULT; | |
1732 | old = compat_alloc_user_space(new_nodes ? size * 2 : size); | |
1733 | if (new_nodes) | |
1734 | new = old + size / sizeof(unsigned long); | |
1735 | if (copy_to_user(old, nodes_addr(tmp_mask), size)) | |
1736 | return -EFAULT; | |
1737 | } | |
1738 | if (new_nodes) { | |
1739 | if (compat_get_bitmap(nodes_addr(tmp_mask), new_nodes, nr_bits)) | |
1740 | return -EFAULT; | |
1741 | if (new == NULL) | |
1742 | new = compat_alloc_user_space(size); | |
1743 | if (copy_to_user(new, nodes_addr(tmp_mask), size)) | |
1744 | return -EFAULT; | |
1745 | } | |
1746 | return kernel_migrate_pages(pid, nr_bits + 1, old, new); | |
1747 | } | |
1748 | ||
1749 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 1750 | |
20ca87f2 LX |
1751 | bool vma_migratable(struct vm_area_struct *vma) |
1752 | { | |
1753 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1754 | return false; | |
1755 | ||
1756 | /* | |
1757 | * DAX device mappings require predictable access latency, so avoid | |
1758 | * incurring periodic faults. | |
1759 | */ | |
1760 | if (vma_is_dax(vma)) | |
1761 | return false; | |
1762 | ||
1763 | if (is_vm_hugetlb_page(vma) && | |
1764 | !hugepage_migration_supported(hstate_vma(vma))) | |
1765 | return false; | |
1766 | ||
1767 | /* | |
1768 | * Migration allocates pages in the highest zone. If we cannot | |
1769 | * do so then migration (at least from node to node) is not | |
1770 | * possible. | |
1771 | */ | |
1772 | if (vma->vm_file && | |
1773 | gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping)) | |
1774 | < policy_zone) | |
1775 | return false; | |
1776 | return true; | |
1777 | } | |
1778 | ||
74d2c3a0 ON |
1779 | struct mempolicy *__get_vma_policy(struct vm_area_struct *vma, |
1780 | unsigned long addr) | |
1da177e4 | 1781 | { |
8d90274b | 1782 | struct mempolicy *pol = NULL; |
1da177e4 LT |
1783 | |
1784 | if (vma) { | |
480eccf9 | 1785 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
8d90274b | 1786 | pol = vma->vm_ops->get_policy(vma, addr); |
00442ad0 | 1787 | } else if (vma->vm_policy) { |
1da177e4 | 1788 | pol = vma->vm_policy; |
00442ad0 MG |
1789 | |
1790 | /* | |
1791 | * shmem_alloc_page() passes MPOL_F_SHARED policy with | |
1792 | * a pseudo vma whose vma->vm_ops=NULL. Take a reference | |
1793 | * count on these policies which will be dropped by | |
1794 | * mpol_cond_put() later | |
1795 | */ | |
1796 | if (mpol_needs_cond_ref(pol)) | |
1797 | mpol_get(pol); | |
1798 | } | |
1da177e4 | 1799 | } |
f15ca78e | 1800 | |
74d2c3a0 ON |
1801 | return pol; |
1802 | } | |
1803 | ||
1804 | /* | |
dd6eecb9 | 1805 | * get_vma_policy(@vma, @addr) |
74d2c3a0 ON |
1806 | * @vma: virtual memory area whose policy is sought |
1807 | * @addr: address in @vma for shared policy lookup | |
1808 | * | |
1809 | * Returns effective policy for a VMA at specified address. | |
dd6eecb9 | 1810 | * Falls back to current->mempolicy or system default policy, as necessary. |
74d2c3a0 ON |
1811 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1812 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1813 | * freeing by another task. It is the caller's responsibility to free the | |
1814 | * extra reference for shared policies. | |
1815 | */ | |
ac79f78d | 1816 | static struct mempolicy *get_vma_policy(struct vm_area_struct *vma, |
dd6eecb9 | 1817 | unsigned long addr) |
74d2c3a0 ON |
1818 | { |
1819 | struct mempolicy *pol = __get_vma_policy(vma, addr); | |
1820 | ||
8d90274b | 1821 | if (!pol) |
dd6eecb9 | 1822 | pol = get_task_policy(current); |
8d90274b | 1823 | |
1da177e4 LT |
1824 | return pol; |
1825 | } | |
1826 | ||
6b6482bb | 1827 | bool vma_policy_mof(struct vm_area_struct *vma) |
fc314724 | 1828 | { |
6b6482bb | 1829 | struct mempolicy *pol; |
fc314724 | 1830 | |
6b6482bb ON |
1831 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
1832 | bool ret = false; | |
fc314724 | 1833 | |
6b6482bb ON |
1834 | pol = vma->vm_ops->get_policy(vma, vma->vm_start); |
1835 | if (pol && (pol->flags & MPOL_F_MOF)) | |
1836 | ret = true; | |
1837 | mpol_cond_put(pol); | |
8d90274b | 1838 | |
6b6482bb | 1839 | return ret; |
fc314724 MG |
1840 | } |
1841 | ||
6b6482bb | 1842 | pol = vma->vm_policy; |
8d90274b | 1843 | if (!pol) |
6b6482bb | 1844 | pol = get_task_policy(current); |
8d90274b | 1845 | |
fc314724 MG |
1846 | return pol->flags & MPOL_F_MOF; |
1847 | } | |
1848 | ||
d3eb1570 LJ |
1849 | static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone) |
1850 | { | |
1851 | enum zone_type dynamic_policy_zone = policy_zone; | |
1852 | ||
1853 | BUG_ON(dynamic_policy_zone == ZONE_MOVABLE); | |
1854 | ||
1855 | /* | |
1856 | * if policy->v.nodes has movable memory only, | |
1857 | * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only. | |
1858 | * | |
1859 | * policy->v.nodes is intersect with node_states[N_MEMORY]. | |
1860 | * so if the following test faile, it implies | |
1861 | * policy->v.nodes has movable memory only. | |
1862 | */ | |
1863 | if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY])) | |
1864 | dynamic_policy_zone = ZONE_MOVABLE; | |
1865 | ||
1866 | return zone >= dynamic_policy_zone; | |
1867 | } | |
1868 | ||
52cd3b07 LS |
1869 | /* |
1870 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1871 | * page allocation | |
1872 | */ | |
8ca39e68 | 1873 | nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) |
19770b32 MG |
1874 | { |
1875 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1876 | if (unlikely(policy->mode == MPOL_BIND) && |
d3eb1570 | 1877 | apply_policy_zone(policy, gfp_zone(gfp)) && |
19770b32 MG |
1878 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) |
1879 | return &policy->v.nodes; | |
1880 | ||
1881 | return NULL; | |
1882 | } | |
1883 | ||
04ec6264 | 1884 | /* Return the node id preferred by the given mempolicy, or the given id */ |
f8fd5253 | 1885 | static int policy_node(gfp_t gfp, struct mempolicy *policy, int nd) |
1da177e4 | 1886 | { |
6d840958 MH |
1887 | if (policy->mode == MPOL_PREFERRED && !(policy->flags & MPOL_F_LOCAL)) |
1888 | nd = policy->v.preferred_node; | |
1889 | else { | |
19770b32 | 1890 | /* |
6d840958 MH |
1891 | * __GFP_THISNODE shouldn't even be used with the bind policy |
1892 | * because we might easily break the expectation to stay on the | |
1893 | * requested node and not break the policy. | |
19770b32 | 1894 | */ |
6d840958 | 1895 | WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE)); |
1da177e4 | 1896 | } |
6d840958 | 1897 | |
04ec6264 | 1898 | return nd; |
1da177e4 LT |
1899 | } |
1900 | ||
1901 | /* Do dynamic interleaving for a process */ | |
1902 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1903 | { | |
45816682 | 1904 | unsigned next; |
1da177e4 LT |
1905 | struct task_struct *me = current; |
1906 | ||
45816682 | 1907 | next = next_node_in(me->il_prev, policy->v.nodes); |
f5b087b5 | 1908 | if (next < MAX_NUMNODES) |
45816682 VB |
1909 | me->il_prev = next; |
1910 | return next; | |
1da177e4 LT |
1911 | } |
1912 | ||
dc85da15 CL |
1913 | /* |
1914 | * Depending on the memory policy provide a node from which to allocate the | |
1915 | * next slab entry. | |
1916 | */ | |
2a389610 | 1917 | unsigned int mempolicy_slab_node(void) |
dc85da15 | 1918 | { |
e7b691b0 | 1919 | struct mempolicy *policy; |
2a389610 | 1920 | int node = numa_mem_id(); |
e7b691b0 AK |
1921 | |
1922 | if (in_interrupt()) | |
2a389610 | 1923 | return node; |
e7b691b0 AK |
1924 | |
1925 | policy = current->mempolicy; | |
fc36b8d3 | 1926 | if (!policy || policy->flags & MPOL_F_LOCAL) |
2a389610 | 1927 | return node; |
bea904d5 LS |
1928 | |
1929 | switch (policy->mode) { | |
1930 | case MPOL_PREFERRED: | |
fc36b8d3 LS |
1931 | /* |
1932 | * handled MPOL_F_LOCAL above | |
1933 | */ | |
1934 | return policy->v.preferred_node; | |
765c4507 | 1935 | |
dc85da15 CL |
1936 | case MPOL_INTERLEAVE: |
1937 | return interleave_nodes(policy); | |
1938 | ||
dd1a239f | 1939 | case MPOL_BIND: { |
c33d6c06 MG |
1940 | struct zoneref *z; |
1941 | ||
dc85da15 CL |
1942 | /* |
1943 | * Follow bind policy behavior and start allocation at the | |
1944 | * first node. | |
1945 | */ | |
19770b32 | 1946 | struct zonelist *zonelist; |
19770b32 | 1947 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); |
c9634cf0 | 1948 | zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK]; |
c33d6c06 MG |
1949 | z = first_zones_zonelist(zonelist, highest_zoneidx, |
1950 | &policy->v.nodes); | |
c1093b74 | 1951 | return z->zone ? zone_to_nid(z->zone) : node; |
dd1a239f | 1952 | } |
dc85da15 | 1953 | |
dc85da15 | 1954 | default: |
bea904d5 | 1955 | BUG(); |
dc85da15 CL |
1956 | } |
1957 | } | |
1958 | ||
fee83b3a AM |
1959 | /* |
1960 | * Do static interleaving for a VMA with known offset @n. Returns the n'th | |
1961 | * node in pol->v.nodes (starting from n=0), wrapping around if n exceeds the | |
1962 | * number of present nodes. | |
1963 | */ | |
98c70baa | 1964 | static unsigned offset_il_node(struct mempolicy *pol, unsigned long n) |
1da177e4 | 1965 | { |
dfcd3c0d | 1966 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1967 | unsigned target; |
fee83b3a AM |
1968 | int i; |
1969 | int nid; | |
1da177e4 | 1970 | |
f5b087b5 DR |
1971 | if (!nnodes) |
1972 | return numa_node_id(); | |
fee83b3a AM |
1973 | target = (unsigned int)n % nnodes; |
1974 | nid = first_node(pol->v.nodes); | |
1975 | for (i = 0; i < target; i++) | |
dfcd3c0d | 1976 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1977 | return nid; |
1978 | } | |
1979 | ||
5da7ca86 CL |
1980 | /* Determine a node number for interleave */ |
1981 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1982 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1983 | { | |
1984 | if (vma) { | |
1985 | unsigned long off; | |
1986 | ||
3b98b087 NA |
1987 | /* |
1988 | * for small pages, there is no difference between | |
1989 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1990 | * for huge pages, since vm_pgoff is in units of small | |
1991 | * pages, we need to shift off the always 0 bits to get | |
1992 | * a useful offset. | |
1993 | */ | |
1994 | BUG_ON(shift < PAGE_SHIFT); | |
1995 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 | 1996 | off += (addr - vma->vm_start) >> shift; |
98c70baa | 1997 | return offset_il_node(pol, off); |
5da7ca86 CL |
1998 | } else |
1999 | return interleave_nodes(pol); | |
2000 | } | |
2001 | ||
00ac59ad | 2002 | #ifdef CONFIG_HUGETLBFS |
480eccf9 | 2003 | /* |
04ec6264 | 2004 | * huge_node(@vma, @addr, @gfp_flags, @mpol) |
b46e14ac FF |
2005 | * @vma: virtual memory area whose policy is sought |
2006 | * @addr: address in @vma for shared policy lookup and interleave policy | |
2007 | * @gfp_flags: for requested zone | |
2008 | * @mpol: pointer to mempolicy pointer for reference counted mempolicy | |
2009 | * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 | 2010 | * |
04ec6264 | 2011 | * Returns a nid suitable for a huge page allocation and a pointer |
52cd3b07 LS |
2012 | * to the struct mempolicy for conditional unref after allocation. |
2013 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's | |
2014 | * @nodemask for filtering the zonelist. | |
c0ff7453 | 2015 | * |
d26914d1 | 2016 | * Must be protected by read_mems_allowed_begin() |
480eccf9 | 2017 | */ |
04ec6264 VB |
2018 | int huge_node(struct vm_area_struct *vma, unsigned long addr, gfp_t gfp_flags, |
2019 | struct mempolicy **mpol, nodemask_t **nodemask) | |
5da7ca86 | 2020 | { |
04ec6264 | 2021 | int nid; |
5da7ca86 | 2022 | |
dd6eecb9 | 2023 | *mpol = get_vma_policy(vma, addr); |
19770b32 | 2024 | *nodemask = NULL; /* assume !MPOL_BIND */ |
5da7ca86 | 2025 | |
52cd3b07 | 2026 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
04ec6264 VB |
2027 | nid = interleave_nid(*mpol, vma, addr, |
2028 | huge_page_shift(hstate_vma(vma))); | |
52cd3b07 | 2029 | } else { |
04ec6264 | 2030 | nid = policy_node(gfp_flags, *mpol, numa_node_id()); |
52cd3b07 LS |
2031 | if ((*mpol)->mode == MPOL_BIND) |
2032 | *nodemask = &(*mpol)->v.nodes; | |
480eccf9 | 2033 | } |
04ec6264 | 2034 | return nid; |
5da7ca86 | 2035 | } |
06808b08 LS |
2036 | |
2037 | /* | |
2038 | * init_nodemask_of_mempolicy | |
2039 | * | |
2040 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
2041 | * to indicate default policy. Otherwise, extract the policy nodemask | |
2042 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
2043 | * initialize the argument nodemask to contain the single node for | |
2044 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
2045 | * of non-default mempolicy. | |
2046 | * | |
2047 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
2048 | * because the current task is examining it's own mempolicy and a task's | |
2049 | * mempolicy is only ever changed by the task itself. | |
2050 | * | |
2051 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
2052 | */ | |
2053 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
2054 | { | |
2055 | struct mempolicy *mempolicy; | |
2056 | int nid; | |
2057 | ||
2058 | if (!(mask && current->mempolicy)) | |
2059 | return false; | |
2060 | ||
c0ff7453 | 2061 | task_lock(current); |
06808b08 LS |
2062 | mempolicy = current->mempolicy; |
2063 | switch (mempolicy->mode) { | |
2064 | case MPOL_PREFERRED: | |
2065 | if (mempolicy->flags & MPOL_F_LOCAL) | |
2066 | nid = numa_node_id(); | |
2067 | else | |
2068 | nid = mempolicy->v.preferred_node; | |
2069 | init_nodemask_of_node(mask, nid); | |
2070 | break; | |
2071 | ||
2072 | case MPOL_BIND: | |
06808b08 LS |
2073 | case MPOL_INTERLEAVE: |
2074 | *mask = mempolicy->v.nodes; | |
2075 | break; | |
2076 | ||
2077 | default: | |
2078 | BUG(); | |
2079 | } | |
c0ff7453 | 2080 | task_unlock(current); |
06808b08 LS |
2081 | |
2082 | return true; | |
2083 | } | |
00ac59ad | 2084 | #endif |
5da7ca86 | 2085 | |
6f48d0eb DR |
2086 | /* |
2087 | * mempolicy_nodemask_intersects | |
2088 | * | |
2089 | * If tsk's mempolicy is "default" [NULL], return 'true' to indicate default | |
2090 | * policy. Otherwise, check for intersection between mask and the policy | |
2091 | * nodemask for 'bind' or 'interleave' policy. For 'perferred' or 'local' | |
2092 | * policy, always return true since it may allocate elsewhere on fallback. | |
2093 | * | |
2094 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
2095 | */ | |
2096 | bool mempolicy_nodemask_intersects(struct task_struct *tsk, | |
2097 | const nodemask_t *mask) | |
2098 | { | |
2099 | struct mempolicy *mempolicy; | |
2100 | bool ret = true; | |
2101 | ||
2102 | if (!mask) | |
2103 | return ret; | |
2104 | task_lock(tsk); | |
2105 | mempolicy = tsk->mempolicy; | |
2106 | if (!mempolicy) | |
2107 | goto out; | |
2108 | ||
2109 | switch (mempolicy->mode) { | |
2110 | case MPOL_PREFERRED: | |
2111 | /* | |
2112 | * MPOL_PREFERRED and MPOL_F_LOCAL are only preferred nodes to | |
2113 | * allocate from, they may fallback to other nodes when oom. | |
2114 | * Thus, it's possible for tsk to have allocated memory from | |
2115 | * nodes in mask. | |
2116 | */ | |
2117 | break; | |
2118 | case MPOL_BIND: | |
2119 | case MPOL_INTERLEAVE: | |
2120 | ret = nodes_intersects(mempolicy->v.nodes, *mask); | |
2121 | break; | |
2122 | default: | |
2123 | BUG(); | |
2124 | } | |
2125 | out: | |
2126 | task_unlock(tsk); | |
2127 | return ret; | |
2128 | } | |
2129 | ||
1da177e4 LT |
2130 | /* Allocate a page in interleaved policy. |
2131 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
2132 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
2133 | unsigned nid) | |
1da177e4 | 2134 | { |
1da177e4 LT |
2135 | struct page *page; |
2136 | ||
04ec6264 | 2137 | page = __alloc_pages(gfp, order, nid); |
4518085e KW |
2138 | /* skip NUMA_INTERLEAVE_HIT counter update if numa stats is disabled */ |
2139 | if (!static_branch_likely(&vm_numa_stat_key)) | |
2140 | return page; | |
de55c8b2 AR |
2141 | if (page && page_to_nid(page) == nid) { |
2142 | preempt_disable(); | |
2143 | __inc_numa_state(page_zone(page), NUMA_INTERLEAVE_HIT); | |
2144 | preempt_enable(); | |
2145 | } | |
1da177e4 LT |
2146 | return page; |
2147 | } | |
2148 | ||
2149 | /** | |
0bbbc0b3 | 2150 | * alloc_pages_vma - Allocate a page for a VMA. |
1da177e4 LT |
2151 | * |
2152 | * @gfp: | |
2153 | * %GFP_USER user allocation. | |
2154 | * %GFP_KERNEL kernel allocations, | |
2155 | * %GFP_HIGHMEM highmem/user allocations, | |
2156 | * %GFP_FS allocation should not call back into a file system. | |
2157 | * %GFP_ATOMIC don't sleep. | |
2158 | * | |
0bbbc0b3 | 2159 | * @order:Order of the GFP allocation. |
1da177e4 LT |
2160 | * @vma: Pointer to VMA or NULL if not available. |
2161 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
be97a41b | 2162 | * @node: Which node to prefer for allocation (modulo policy). |
19deb769 | 2163 | * @hugepage: for hugepages try only the preferred node if possible |
1da177e4 LT |
2164 | * |
2165 | * This function allocates a page from the kernel page pool and applies | |
2166 | * a NUMA policy associated with the VMA or the current process. | |
3e4e28c5 | 2167 | * When VMA is not NULL caller must read-lock the mmap_lock of the |
1da177e4 | 2168 | * mm_struct of the VMA to prevent it from going away. Should be used for |
be97a41b VB |
2169 | * all allocations for pages that will be mapped into user space. Returns |
2170 | * NULL when no page can be allocated. | |
1da177e4 LT |
2171 | */ |
2172 | struct page * | |
0bbbc0b3 | 2173 | alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma, |
19deb769 | 2174 | unsigned long addr, int node, bool hugepage) |
1da177e4 | 2175 | { |
cc9a6c87 | 2176 | struct mempolicy *pol; |
c0ff7453 | 2177 | struct page *page; |
04ec6264 | 2178 | int preferred_nid; |
be97a41b | 2179 | nodemask_t *nmask; |
cc9a6c87 | 2180 | |
dd6eecb9 | 2181 | pol = get_vma_policy(vma, addr); |
1da177e4 | 2182 | |
0867a57c VB |
2183 | if (pol->mode == MPOL_INTERLEAVE) { |
2184 | unsigned nid; | |
2185 | ||
2186 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); | |
2187 | mpol_cond_put(pol); | |
2188 | page = alloc_page_interleave(gfp, order, nid); | |
2189 | goto out; | |
19deb769 DR |
2190 | } |
2191 | ||
2192 | if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) { | |
2193 | int hpage_node = node; | |
2194 | ||
2195 | /* | |
2196 | * For hugepage allocation and non-interleave policy which | |
2197 | * allows the current node (or other explicitly preferred | |
2198 | * node) we only try to allocate from the current/preferred | |
2199 | * node and don't fall back to other nodes, as the cost of | |
2200 | * remote accesses would likely offset THP benefits. | |
2201 | * | |
2202 | * If the policy is interleave, or does not allow the current | |
2203 | * node in its nodemask, we allocate the standard way. | |
2204 | */ | |
2205 | if (pol->mode == MPOL_PREFERRED && !(pol->flags & MPOL_F_LOCAL)) | |
2206 | hpage_node = pol->v.preferred_node; | |
2207 | ||
2208 | nmask = policy_nodemask(gfp, pol); | |
2209 | if (!nmask || node_isset(hpage_node, *nmask)) { | |
2210 | mpol_cond_put(pol); | |
cc638f32 VB |
2211 | /* |
2212 | * First, try to allocate THP only on local node, but | |
2213 | * don't reclaim unnecessarily, just compact. | |
2214 | */ | |
19deb769 | 2215 | page = __alloc_pages_node(hpage_node, |
cc638f32 | 2216 | gfp | __GFP_THISNODE | __GFP_NORETRY, order); |
76e654cc DR |
2217 | |
2218 | /* | |
2219 | * If hugepage allocations are configured to always | |
2220 | * synchronous compact or the vma has been madvised | |
2221 | * to prefer hugepage backing, retry allowing remote | |
cc638f32 | 2222 | * memory with both reclaim and compact as well. |
76e654cc DR |
2223 | */ |
2224 | if (!page && (gfp & __GFP_DIRECT_RECLAIM)) | |
2225 | page = __alloc_pages_node(hpage_node, | |
cc638f32 | 2226 | gfp, order); |
76e654cc | 2227 | |
19deb769 DR |
2228 | goto out; |
2229 | } | |
356ff8a9 DR |
2230 | } |
2231 | ||
be97a41b | 2232 | nmask = policy_nodemask(gfp, pol); |
04ec6264 VB |
2233 | preferred_nid = policy_node(gfp, pol, node); |
2234 | page = __alloc_pages_nodemask(gfp, order, preferred_nid, nmask); | |
d51e9894 | 2235 | mpol_cond_put(pol); |
be97a41b | 2236 | out: |
c0ff7453 | 2237 | return page; |
1da177e4 | 2238 | } |
69262215 | 2239 | EXPORT_SYMBOL(alloc_pages_vma); |
1da177e4 LT |
2240 | |
2241 | /** | |
2242 | * alloc_pages_current - Allocate pages. | |
2243 | * | |
2244 | * @gfp: | |
2245 | * %GFP_USER user allocation, | |
2246 | * %GFP_KERNEL kernel allocation, | |
2247 | * %GFP_HIGHMEM highmem allocation, | |
2248 | * %GFP_FS don't call back into a file system. | |
2249 | * %GFP_ATOMIC don't sleep. | |
2250 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
2251 | * | |
2252 | * Allocate a page from the kernel page pool. When not in | |
2253 | * interrupt context and apply the current process NUMA policy. | |
2254 | * Returns NULL when no page can be allocated. | |
1da177e4 | 2255 | */ |
dd0fc66f | 2256 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 | 2257 | { |
8d90274b | 2258 | struct mempolicy *pol = &default_policy; |
c0ff7453 | 2259 | struct page *page; |
1da177e4 | 2260 | |
8d90274b ON |
2261 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) |
2262 | pol = get_task_policy(current); | |
52cd3b07 LS |
2263 | |
2264 | /* | |
2265 | * No reference counting needed for current->mempolicy | |
2266 | * nor system default_policy | |
2267 | */ | |
45c4745a | 2268 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 MX |
2269 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
2270 | else | |
2271 | page = __alloc_pages_nodemask(gfp, order, | |
04ec6264 | 2272 | policy_node(gfp, pol, numa_node_id()), |
5c4b4be3 | 2273 | policy_nodemask(gfp, pol)); |
cc9a6c87 | 2274 | |
c0ff7453 | 2275 | return page; |
1da177e4 LT |
2276 | } |
2277 | EXPORT_SYMBOL(alloc_pages_current); | |
2278 | ||
ef0855d3 ON |
2279 | int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst) |
2280 | { | |
2281 | struct mempolicy *pol = mpol_dup(vma_policy(src)); | |
2282 | ||
2283 | if (IS_ERR(pol)) | |
2284 | return PTR_ERR(pol); | |
2285 | dst->vm_policy = pol; | |
2286 | return 0; | |
2287 | } | |
2288 | ||
4225399a | 2289 | /* |
846a16bf | 2290 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2291 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2292 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2293 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2294 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2295 | * |
2296 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2297 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2298 | */ |
4225399a | 2299 | |
846a16bf LS |
2300 | /* Slow path of a mempolicy duplicate */ |
2301 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2302 | { |
2303 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2304 | ||
2305 | if (!new) | |
2306 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2307 | |
2308 | /* task's mempolicy is protected by alloc_lock */ | |
2309 | if (old == current->mempolicy) { | |
2310 | task_lock(current); | |
2311 | *new = *old; | |
2312 | task_unlock(current); | |
2313 | } else | |
2314 | *new = *old; | |
2315 | ||
4225399a PJ |
2316 | if (current_cpuset_is_being_rebound()) { |
2317 | nodemask_t mems = cpuset_mems_allowed(current); | |
213980c0 | 2318 | mpol_rebind_policy(new, &mems); |
4225399a | 2319 | } |
1da177e4 | 2320 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2321 | return new; |
2322 | } | |
2323 | ||
2324 | /* Slow path of a mempolicy comparison */ | |
fcfb4dcc | 2325 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2326 | { |
2327 | if (!a || !b) | |
fcfb4dcc | 2328 | return false; |
45c4745a | 2329 | if (a->mode != b->mode) |
fcfb4dcc | 2330 | return false; |
19800502 | 2331 | if (a->flags != b->flags) |
fcfb4dcc | 2332 | return false; |
19800502 BL |
2333 | if (mpol_store_user_nodemask(a)) |
2334 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2335 | return false; |
19800502 | 2336 | |
45c4745a | 2337 | switch (a->mode) { |
19770b32 | 2338 | case MPOL_BIND: |
1da177e4 | 2339 | case MPOL_INTERLEAVE: |
fcfb4dcc | 2340 | return !!nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 | 2341 | case MPOL_PREFERRED: |
8970a63e YX |
2342 | /* a's ->flags is the same as b's */ |
2343 | if (a->flags & MPOL_F_LOCAL) | |
2344 | return true; | |
75719661 | 2345 | return a->v.preferred_node == b->v.preferred_node; |
1da177e4 LT |
2346 | default: |
2347 | BUG(); | |
fcfb4dcc | 2348 | return false; |
1da177e4 LT |
2349 | } |
2350 | } | |
2351 | ||
1da177e4 LT |
2352 | /* |
2353 | * Shared memory backing store policy support. | |
2354 | * | |
2355 | * Remember policies even when nobody has shared memory mapped. | |
2356 | * The policies are kept in Red-Black tree linked from the inode. | |
4a8c7bb5 | 2357 | * They are protected by the sp->lock rwlock, which should be held |
1da177e4 LT |
2358 | * for any accesses to the tree. |
2359 | */ | |
2360 | ||
4a8c7bb5 NZ |
2361 | /* |
2362 | * lookup first element intersecting start-end. Caller holds sp->lock for | |
2363 | * reading or for writing | |
2364 | */ | |
1da177e4 LT |
2365 | static struct sp_node * |
2366 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2367 | { | |
2368 | struct rb_node *n = sp->root.rb_node; | |
2369 | ||
2370 | while (n) { | |
2371 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2372 | ||
2373 | if (start >= p->end) | |
2374 | n = n->rb_right; | |
2375 | else if (end <= p->start) | |
2376 | n = n->rb_left; | |
2377 | else | |
2378 | break; | |
2379 | } | |
2380 | if (!n) | |
2381 | return NULL; | |
2382 | for (;;) { | |
2383 | struct sp_node *w = NULL; | |
2384 | struct rb_node *prev = rb_prev(n); | |
2385 | if (!prev) | |
2386 | break; | |
2387 | w = rb_entry(prev, struct sp_node, nd); | |
2388 | if (w->end <= start) | |
2389 | break; | |
2390 | n = prev; | |
2391 | } | |
2392 | return rb_entry(n, struct sp_node, nd); | |
2393 | } | |
2394 | ||
4a8c7bb5 NZ |
2395 | /* |
2396 | * Insert a new shared policy into the list. Caller holds sp->lock for | |
2397 | * writing. | |
2398 | */ | |
1da177e4 LT |
2399 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) |
2400 | { | |
2401 | struct rb_node **p = &sp->root.rb_node; | |
2402 | struct rb_node *parent = NULL; | |
2403 | struct sp_node *nd; | |
2404 | ||
2405 | while (*p) { | |
2406 | parent = *p; | |
2407 | nd = rb_entry(parent, struct sp_node, nd); | |
2408 | if (new->start < nd->start) | |
2409 | p = &(*p)->rb_left; | |
2410 | else if (new->end > nd->end) | |
2411 | p = &(*p)->rb_right; | |
2412 | else | |
2413 | BUG(); | |
2414 | } | |
2415 | rb_link_node(&new->nd, parent, p); | |
2416 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2417 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2418 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2419 | } |
2420 | ||
2421 | /* Find shared policy intersecting idx */ | |
2422 | struct mempolicy * | |
2423 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2424 | { | |
2425 | struct mempolicy *pol = NULL; | |
2426 | struct sp_node *sn; | |
2427 | ||
2428 | if (!sp->root.rb_node) | |
2429 | return NULL; | |
4a8c7bb5 | 2430 | read_lock(&sp->lock); |
1da177e4 LT |
2431 | sn = sp_lookup(sp, idx, idx+1); |
2432 | if (sn) { | |
2433 | mpol_get(sn->policy); | |
2434 | pol = sn->policy; | |
2435 | } | |
4a8c7bb5 | 2436 | read_unlock(&sp->lock); |
1da177e4 LT |
2437 | return pol; |
2438 | } | |
2439 | ||
63f74ca2 KM |
2440 | static void sp_free(struct sp_node *n) |
2441 | { | |
2442 | mpol_put(n->policy); | |
2443 | kmem_cache_free(sn_cache, n); | |
2444 | } | |
2445 | ||
771fb4d8 LS |
2446 | /** |
2447 | * mpol_misplaced - check whether current page node is valid in policy | |
2448 | * | |
b46e14ac FF |
2449 | * @page: page to be checked |
2450 | * @vma: vm area where page mapped | |
2451 | * @addr: virtual address where page mapped | |
771fb4d8 LS |
2452 | * |
2453 | * Lookup current policy node id for vma,addr and "compare to" page's | |
2454 | * node id. | |
2455 | * | |
2456 | * Returns: | |
2457 | * -1 - not misplaced, page is in the right node | |
2458 | * node - node id where the page should be | |
2459 | * | |
2460 | * Policy determination "mimics" alloc_page_vma(). | |
2461 | * Called from fault path where we know the vma and faulting address. | |
2462 | */ | |
2463 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) | |
2464 | { | |
2465 | struct mempolicy *pol; | |
c33d6c06 | 2466 | struct zoneref *z; |
771fb4d8 LS |
2467 | int curnid = page_to_nid(page); |
2468 | unsigned long pgoff; | |
90572890 PZ |
2469 | int thiscpu = raw_smp_processor_id(); |
2470 | int thisnid = cpu_to_node(thiscpu); | |
98fa15f3 | 2471 | int polnid = NUMA_NO_NODE; |
771fb4d8 LS |
2472 | int ret = -1; |
2473 | ||
dd6eecb9 | 2474 | pol = get_vma_policy(vma, addr); |
771fb4d8 LS |
2475 | if (!(pol->flags & MPOL_F_MOF)) |
2476 | goto out; | |
2477 | ||
2478 | switch (pol->mode) { | |
2479 | case MPOL_INTERLEAVE: | |
771fb4d8 LS |
2480 | pgoff = vma->vm_pgoff; |
2481 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; | |
98c70baa | 2482 | polnid = offset_il_node(pol, pgoff); |
771fb4d8 LS |
2483 | break; |
2484 | ||
2485 | case MPOL_PREFERRED: | |
2486 | if (pol->flags & MPOL_F_LOCAL) | |
2487 | polnid = numa_node_id(); | |
2488 | else | |
2489 | polnid = pol->v.preferred_node; | |
2490 | break; | |
2491 | ||
2492 | case MPOL_BIND: | |
c33d6c06 | 2493 | |
771fb4d8 LS |
2494 | /* |
2495 | * allows binding to multiple nodes. | |
2496 | * use current page if in policy nodemask, | |
2497 | * else select nearest allowed node, if any. | |
2498 | * If no allowed nodes, use current [!misplaced]. | |
2499 | */ | |
2500 | if (node_isset(curnid, pol->v.nodes)) | |
2501 | goto out; | |
c33d6c06 | 2502 | z = first_zones_zonelist( |
771fb4d8 LS |
2503 | node_zonelist(numa_node_id(), GFP_HIGHUSER), |
2504 | gfp_zone(GFP_HIGHUSER), | |
c33d6c06 | 2505 | &pol->v.nodes); |
c1093b74 | 2506 | polnid = zone_to_nid(z->zone); |
771fb4d8 LS |
2507 | break; |
2508 | ||
2509 | default: | |
2510 | BUG(); | |
2511 | } | |
5606e387 MG |
2512 | |
2513 | /* Migrate the page towards the node whose CPU is referencing it */ | |
e42c8ff2 | 2514 | if (pol->flags & MPOL_F_MORON) { |
90572890 | 2515 | polnid = thisnid; |
5606e387 | 2516 | |
10f39042 | 2517 | if (!should_numa_migrate_memory(current, page, curnid, thiscpu)) |
de1c9ce6 | 2518 | goto out; |
e42c8ff2 MG |
2519 | } |
2520 | ||
771fb4d8 LS |
2521 | if (curnid != polnid) |
2522 | ret = polnid; | |
2523 | out: | |
2524 | mpol_cond_put(pol); | |
2525 | ||
2526 | return ret; | |
2527 | } | |
2528 | ||
c11600e4 DR |
2529 | /* |
2530 | * Drop the (possibly final) reference to task->mempolicy. It needs to be | |
2531 | * dropped after task->mempolicy is set to NULL so that any allocation done as | |
2532 | * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed | |
2533 | * policy. | |
2534 | */ | |
2535 | void mpol_put_task_policy(struct task_struct *task) | |
2536 | { | |
2537 | struct mempolicy *pol; | |
2538 | ||
2539 | task_lock(task); | |
2540 | pol = task->mempolicy; | |
2541 | task->mempolicy = NULL; | |
2542 | task_unlock(task); | |
2543 | mpol_put(pol); | |
2544 | } | |
2545 | ||
1da177e4 LT |
2546 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2547 | { | |
140d5a49 | 2548 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2549 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2550 | sp_free(n); |
1da177e4 LT |
2551 | } |
2552 | ||
42288fe3 MG |
2553 | static void sp_node_init(struct sp_node *node, unsigned long start, |
2554 | unsigned long end, struct mempolicy *pol) | |
2555 | { | |
2556 | node->start = start; | |
2557 | node->end = end; | |
2558 | node->policy = pol; | |
2559 | } | |
2560 | ||
dbcb0f19 AB |
2561 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2562 | struct mempolicy *pol) | |
1da177e4 | 2563 | { |
869833f2 KM |
2564 | struct sp_node *n; |
2565 | struct mempolicy *newpol; | |
1da177e4 | 2566 | |
869833f2 | 2567 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2568 | if (!n) |
2569 | return NULL; | |
869833f2 KM |
2570 | |
2571 | newpol = mpol_dup(pol); | |
2572 | if (IS_ERR(newpol)) { | |
2573 | kmem_cache_free(sn_cache, n); | |
2574 | return NULL; | |
2575 | } | |
2576 | newpol->flags |= MPOL_F_SHARED; | |
42288fe3 | 2577 | sp_node_init(n, start, end, newpol); |
869833f2 | 2578 | |
1da177e4 LT |
2579 | return n; |
2580 | } | |
2581 | ||
2582 | /* Replace a policy range. */ | |
2583 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2584 | unsigned long end, struct sp_node *new) | |
2585 | { | |
b22d127a | 2586 | struct sp_node *n; |
42288fe3 MG |
2587 | struct sp_node *n_new = NULL; |
2588 | struct mempolicy *mpol_new = NULL; | |
b22d127a | 2589 | int ret = 0; |
1da177e4 | 2590 | |
42288fe3 | 2591 | restart: |
4a8c7bb5 | 2592 | write_lock(&sp->lock); |
1da177e4 LT |
2593 | n = sp_lookup(sp, start, end); |
2594 | /* Take care of old policies in the same range. */ | |
2595 | while (n && n->start < end) { | |
2596 | struct rb_node *next = rb_next(&n->nd); | |
2597 | if (n->start >= start) { | |
2598 | if (n->end <= end) | |
2599 | sp_delete(sp, n); | |
2600 | else | |
2601 | n->start = end; | |
2602 | } else { | |
2603 | /* Old policy spanning whole new range. */ | |
2604 | if (n->end > end) { | |
42288fe3 MG |
2605 | if (!n_new) |
2606 | goto alloc_new; | |
2607 | ||
2608 | *mpol_new = *n->policy; | |
2609 | atomic_set(&mpol_new->refcnt, 1); | |
7880639c | 2610 | sp_node_init(n_new, end, n->end, mpol_new); |
1da177e4 | 2611 | n->end = start; |
5ca39575 | 2612 | sp_insert(sp, n_new); |
42288fe3 MG |
2613 | n_new = NULL; |
2614 | mpol_new = NULL; | |
1da177e4 LT |
2615 | break; |
2616 | } else | |
2617 | n->end = start; | |
2618 | } | |
2619 | if (!next) | |
2620 | break; | |
2621 | n = rb_entry(next, struct sp_node, nd); | |
2622 | } | |
2623 | if (new) | |
2624 | sp_insert(sp, new); | |
4a8c7bb5 | 2625 | write_unlock(&sp->lock); |
42288fe3 MG |
2626 | ret = 0; |
2627 | ||
2628 | err_out: | |
2629 | if (mpol_new) | |
2630 | mpol_put(mpol_new); | |
2631 | if (n_new) | |
2632 | kmem_cache_free(sn_cache, n_new); | |
2633 | ||
b22d127a | 2634 | return ret; |
42288fe3 MG |
2635 | |
2636 | alloc_new: | |
4a8c7bb5 | 2637 | write_unlock(&sp->lock); |
42288fe3 MG |
2638 | ret = -ENOMEM; |
2639 | n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
2640 | if (!n_new) | |
2641 | goto err_out; | |
2642 | mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2643 | if (!mpol_new) | |
2644 | goto err_out; | |
2645 | goto restart; | |
1da177e4 LT |
2646 | } |
2647 | ||
71fe804b LS |
2648 | /** |
2649 | * mpol_shared_policy_init - initialize shared policy for inode | |
2650 | * @sp: pointer to inode shared policy | |
2651 | * @mpol: struct mempolicy to install | |
2652 | * | |
2653 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2654 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2655 | * This must be released on exit. | |
4bfc4495 | 2656 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2657 | */ |
2658 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2659 | { | |
58568d2a MX |
2660 | int ret; |
2661 | ||
71fe804b | 2662 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
4a8c7bb5 | 2663 | rwlock_init(&sp->lock); |
71fe804b LS |
2664 | |
2665 | if (mpol) { | |
2666 | struct vm_area_struct pvma; | |
2667 | struct mempolicy *new; | |
4bfc4495 | 2668 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2669 | |
4bfc4495 | 2670 | if (!scratch) |
5c0c1654 | 2671 | goto put_mpol; |
71fe804b LS |
2672 | /* contextualize the tmpfs mount point mempolicy */ |
2673 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2674 | if (IS_ERR(new)) |
0cae3457 | 2675 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2676 | |
2677 | task_lock(current); | |
4bfc4495 | 2678 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2679 | task_unlock(current); |
15d77835 | 2680 | if (ret) |
5c0c1654 | 2681 | goto put_new; |
71fe804b LS |
2682 | |
2683 | /* Create pseudo-vma that contains just the policy */ | |
2c4541e2 | 2684 | vma_init(&pvma, NULL); |
71fe804b LS |
2685 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ |
2686 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2687 | |
5c0c1654 | 2688 | put_new: |
71fe804b | 2689 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2690 | free_scratch: |
4bfc4495 | 2691 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2692 | put_mpol: |
2693 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2694 | } |
2695 | } | |
2696 | ||
1da177e4 LT |
2697 | int mpol_set_shared_policy(struct shared_policy *info, |
2698 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2699 | { | |
2700 | int err; | |
2701 | struct sp_node *new = NULL; | |
2702 | unsigned long sz = vma_pages(vma); | |
2703 | ||
028fec41 | 2704 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2705 | vma->vm_pgoff, |
45c4745a | 2706 | sz, npol ? npol->mode : -1, |
028fec41 | 2707 | npol ? npol->flags : -1, |
00ef2d2f | 2708 | npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE); |
1da177e4 LT |
2709 | |
2710 | if (npol) { | |
2711 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2712 | if (!new) | |
2713 | return -ENOMEM; | |
2714 | } | |
2715 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2716 | if (err && new) | |
63f74ca2 | 2717 | sp_free(new); |
1da177e4 LT |
2718 | return err; |
2719 | } | |
2720 | ||
2721 | /* Free a backing policy store on inode delete. */ | |
2722 | void mpol_free_shared_policy(struct shared_policy *p) | |
2723 | { | |
2724 | struct sp_node *n; | |
2725 | struct rb_node *next; | |
2726 | ||
2727 | if (!p->root.rb_node) | |
2728 | return; | |
4a8c7bb5 | 2729 | write_lock(&p->lock); |
1da177e4 LT |
2730 | next = rb_first(&p->root); |
2731 | while (next) { | |
2732 | n = rb_entry(next, struct sp_node, nd); | |
2733 | next = rb_next(&n->nd); | |
63f74ca2 | 2734 | sp_delete(p, n); |
1da177e4 | 2735 | } |
4a8c7bb5 | 2736 | write_unlock(&p->lock); |
1da177e4 LT |
2737 | } |
2738 | ||
1a687c2e | 2739 | #ifdef CONFIG_NUMA_BALANCING |
c297663c | 2740 | static int __initdata numabalancing_override; |
1a687c2e MG |
2741 | |
2742 | static void __init check_numabalancing_enable(void) | |
2743 | { | |
2744 | bool numabalancing_default = false; | |
2745 | ||
2746 | if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED)) | |
2747 | numabalancing_default = true; | |
2748 | ||
c297663c MG |
2749 | /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */ |
2750 | if (numabalancing_override) | |
2751 | set_numabalancing_state(numabalancing_override == 1); | |
2752 | ||
b0dc2b9b | 2753 | if (num_online_nodes() > 1 && !numabalancing_override) { |
756a025f | 2754 | pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n", |
c297663c | 2755 | numabalancing_default ? "Enabling" : "Disabling"); |
1a687c2e MG |
2756 | set_numabalancing_state(numabalancing_default); |
2757 | } | |
2758 | } | |
2759 | ||
2760 | static int __init setup_numabalancing(char *str) | |
2761 | { | |
2762 | int ret = 0; | |
2763 | if (!str) | |
2764 | goto out; | |
1a687c2e MG |
2765 | |
2766 | if (!strcmp(str, "enable")) { | |
c297663c | 2767 | numabalancing_override = 1; |
1a687c2e MG |
2768 | ret = 1; |
2769 | } else if (!strcmp(str, "disable")) { | |
c297663c | 2770 | numabalancing_override = -1; |
1a687c2e MG |
2771 | ret = 1; |
2772 | } | |
2773 | out: | |
2774 | if (!ret) | |
4a404bea | 2775 | pr_warn("Unable to parse numa_balancing=\n"); |
1a687c2e MG |
2776 | |
2777 | return ret; | |
2778 | } | |
2779 | __setup("numa_balancing=", setup_numabalancing); | |
2780 | #else | |
2781 | static inline void __init check_numabalancing_enable(void) | |
2782 | { | |
2783 | } | |
2784 | #endif /* CONFIG_NUMA_BALANCING */ | |
2785 | ||
1da177e4 LT |
2786 | /* assumes fs == KERNEL_DS */ |
2787 | void __init numa_policy_init(void) | |
2788 | { | |
b71636e2 PM |
2789 | nodemask_t interleave_nodes; |
2790 | unsigned long largest = 0; | |
2791 | int nid, prefer = 0; | |
2792 | ||
1da177e4 LT |
2793 | policy_cache = kmem_cache_create("numa_policy", |
2794 | sizeof(struct mempolicy), | |
20c2df83 | 2795 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2796 | |
2797 | sn_cache = kmem_cache_create("shared_policy_node", | |
2798 | sizeof(struct sp_node), | |
20c2df83 | 2799 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2800 | |
5606e387 MG |
2801 | for_each_node(nid) { |
2802 | preferred_node_policy[nid] = (struct mempolicy) { | |
2803 | .refcnt = ATOMIC_INIT(1), | |
2804 | .mode = MPOL_PREFERRED, | |
2805 | .flags = MPOL_F_MOF | MPOL_F_MORON, | |
2806 | .v = { .preferred_node = nid, }, | |
2807 | }; | |
2808 | } | |
2809 | ||
b71636e2 PM |
2810 | /* |
2811 | * Set interleaving policy for system init. Interleaving is only | |
2812 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2813 | * fall back to the largest node if they're all smaller. | |
2814 | */ | |
2815 | nodes_clear(interleave_nodes); | |
01f13bd6 | 2816 | for_each_node_state(nid, N_MEMORY) { |
b71636e2 PM |
2817 | unsigned long total_pages = node_present_pages(nid); |
2818 | ||
2819 | /* Preserve the largest node */ | |
2820 | if (largest < total_pages) { | |
2821 | largest = total_pages; | |
2822 | prefer = nid; | |
2823 | } | |
2824 | ||
2825 | /* Interleave this node? */ | |
2826 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2827 | node_set(nid, interleave_nodes); | |
2828 | } | |
2829 | ||
2830 | /* All too small, use the largest */ | |
2831 | if (unlikely(nodes_empty(interleave_nodes))) | |
2832 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2833 | |
028fec41 | 2834 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
b1de0d13 | 2835 | pr_err("%s: interleaving failed\n", __func__); |
1a687c2e MG |
2836 | |
2837 | check_numabalancing_enable(); | |
1da177e4 LT |
2838 | } |
2839 | ||
8bccd85f | 2840 | /* Reset policy of current process to default */ |
1da177e4 LT |
2841 | void numa_default_policy(void) |
2842 | { | |
028fec41 | 2843 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2844 | } |
68860ec1 | 2845 | |
095f1fc4 LS |
2846 | /* |
2847 | * Parse and format mempolicy from/to strings | |
2848 | */ | |
2849 | ||
1a75a6c8 | 2850 | /* |
f2a07f40 | 2851 | * "local" is implemented internally by MPOL_PREFERRED with MPOL_F_LOCAL flag. |
1a75a6c8 | 2852 | */ |
345ace9c LS |
2853 | static const char * const policy_modes[] = |
2854 | { | |
2855 | [MPOL_DEFAULT] = "default", | |
2856 | [MPOL_PREFERRED] = "prefer", | |
2857 | [MPOL_BIND] = "bind", | |
2858 | [MPOL_INTERLEAVE] = "interleave", | |
d3a71033 | 2859 | [MPOL_LOCAL] = "local", |
345ace9c | 2860 | }; |
1a75a6c8 | 2861 | |
095f1fc4 LS |
2862 | |
2863 | #ifdef CONFIG_TMPFS | |
2864 | /** | |
f2a07f40 | 2865 | * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option. |
095f1fc4 | 2866 | * @str: string containing mempolicy to parse |
71fe804b | 2867 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
095f1fc4 LS |
2868 | * |
2869 | * Format of input: | |
2870 | * <mode>[=<flags>][:<nodelist>] | |
2871 | * | |
71fe804b | 2872 | * On success, returns 0, else 1 |
095f1fc4 | 2873 | */ |
a7a88b23 | 2874 | int mpol_parse_str(char *str, struct mempolicy **mpol) |
095f1fc4 | 2875 | { |
71fe804b | 2876 | struct mempolicy *new = NULL; |
f2a07f40 | 2877 | unsigned short mode_flags; |
71fe804b | 2878 | nodemask_t nodes; |
095f1fc4 LS |
2879 | char *nodelist = strchr(str, ':'); |
2880 | char *flags = strchr(str, '='); | |
dedf2c73 | 2881 | int err = 1, mode; |
095f1fc4 | 2882 | |
c7a91bc7 DC |
2883 | if (flags) |
2884 | *flags++ = '\0'; /* terminate mode string */ | |
2885 | ||
095f1fc4 LS |
2886 | if (nodelist) { |
2887 | /* NUL-terminate mode or flags string */ | |
2888 | *nodelist++ = '\0'; | |
71fe804b | 2889 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2890 | goto out; |
01f13bd6 | 2891 | if (!nodes_subset(nodes, node_states[N_MEMORY])) |
095f1fc4 | 2892 | goto out; |
71fe804b LS |
2893 | } else |
2894 | nodes_clear(nodes); | |
2895 | ||
dedf2c73 | 2896 | mode = match_string(policy_modes, MPOL_MAX, str); |
2897 | if (mode < 0) | |
095f1fc4 LS |
2898 | goto out; |
2899 | ||
71fe804b | 2900 | switch (mode) { |
095f1fc4 | 2901 | case MPOL_PREFERRED: |
71fe804b | 2902 | /* |
aa9f7d51 RD |
2903 | * Insist on a nodelist of one node only, although later |
2904 | * we use first_node(nodes) to grab a single node, so here | |
2905 | * nodelist (or nodes) cannot be empty. | |
71fe804b | 2906 | */ |
095f1fc4 LS |
2907 | if (nodelist) { |
2908 | char *rest = nodelist; | |
2909 | while (isdigit(*rest)) | |
2910 | rest++; | |
926f2ae0 KM |
2911 | if (*rest) |
2912 | goto out; | |
aa9f7d51 RD |
2913 | if (nodes_empty(nodes)) |
2914 | goto out; | |
095f1fc4 LS |
2915 | } |
2916 | break; | |
095f1fc4 LS |
2917 | case MPOL_INTERLEAVE: |
2918 | /* | |
2919 | * Default to online nodes with memory if no nodelist | |
2920 | */ | |
2921 | if (!nodelist) | |
01f13bd6 | 2922 | nodes = node_states[N_MEMORY]; |
3f226aa1 | 2923 | break; |
71fe804b | 2924 | case MPOL_LOCAL: |
3f226aa1 | 2925 | /* |
71fe804b | 2926 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 2927 | */ |
71fe804b | 2928 | if (nodelist) |
3f226aa1 | 2929 | goto out; |
71fe804b | 2930 | mode = MPOL_PREFERRED; |
3f226aa1 | 2931 | break; |
413b43de RT |
2932 | case MPOL_DEFAULT: |
2933 | /* | |
2934 | * Insist on a empty nodelist | |
2935 | */ | |
2936 | if (!nodelist) | |
2937 | err = 0; | |
2938 | goto out; | |
d69b2e63 KM |
2939 | case MPOL_BIND: |
2940 | /* | |
2941 | * Insist on a nodelist | |
2942 | */ | |
2943 | if (!nodelist) | |
2944 | goto out; | |
095f1fc4 LS |
2945 | } |
2946 | ||
71fe804b | 2947 | mode_flags = 0; |
095f1fc4 LS |
2948 | if (flags) { |
2949 | /* | |
2950 | * Currently, we only support two mutually exclusive | |
2951 | * mode flags. | |
2952 | */ | |
2953 | if (!strcmp(flags, "static")) | |
71fe804b | 2954 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 2955 | else if (!strcmp(flags, "relative")) |
71fe804b | 2956 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 2957 | else |
926f2ae0 | 2958 | goto out; |
095f1fc4 | 2959 | } |
71fe804b LS |
2960 | |
2961 | new = mpol_new(mode, mode_flags, &nodes); | |
2962 | if (IS_ERR(new)) | |
926f2ae0 KM |
2963 | goto out; |
2964 | ||
f2a07f40 HD |
2965 | /* |
2966 | * Save nodes for mpol_to_str() to show the tmpfs mount options | |
2967 | * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo. | |
2968 | */ | |
2969 | if (mode != MPOL_PREFERRED) | |
2970 | new->v.nodes = nodes; | |
2971 | else if (nodelist) | |
2972 | new->v.preferred_node = first_node(nodes); | |
2973 | else | |
2974 | new->flags |= MPOL_F_LOCAL; | |
2975 | ||
2976 | /* | |
2977 | * Save nodes for contextualization: this will be used to "clone" | |
2978 | * the mempolicy in a specific context [cpuset] at a later time. | |
2979 | */ | |
2980 | new->w.user_nodemask = nodes; | |
2981 | ||
926f2ae0 | 2982 | err = 0; |
71fe804b | 2983 | |
095f1fc4 LS |
2984 | out: |
2985 | /* Restore string for error message */ | |
2986 | if (nodelist) | |
2987 | *--nodelist = ':'; | |
2988 | if (flags) | |
2989 | *--flags = '='; | |
71fe804b LS |
2990 | if (!err) |
2991 | *mpol = new; | |
095f1fc4 LS |
2992 | return err; |
2993 | } | |
2994 | #endif /* CONFIG_TMPFS */ | |
2995 | ||
71fe804b LS |
2996 | /** |
2997 | * mpol_to_str - format a mempolicy structure for printing | |
2998 | * @buffer: to contain formatted mempolicy string | |
2999 | * @maxlen: length of @buffer | |
3000 | * @pol: pointer to mempolicy to be formatted | |
71fe804b | 3001 | * |
948927ee DR |
3002 | * Convert @pol into a string. If @buffer is too short, truncate the string. |
3003 | * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the | |
3004 | * longest flag, "relative", and to display at least a few node ids. | |
1a75a6c8 | 3005 | */ |
948927ee | 3006 | void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
1a75a6c8 CL |
3007 | { |
3008 | char *p = buffer; | |
948927ee DR |
3009 | nodemask_t nodes = NODE_MASK_NONE; |
3010 | unsigned short mode = MPOL_DEFAULT; | |
3011 | unsigned short flags = 0; | |
2291990a | 3012 | |
8790c71a | 3013 | if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) { |
bea904d5 | 3014 | mode = pol->mode; |
948927ee DR |
3015 | flags = pol->flags; |
3016 | } | |
bea904d5 | 3017 | |
1a75a6c8 CL |
3018 | switch (mode) { |
3019 | case MPOL_DEFAULT: | |
1a75a6c8 | 3020 | break; |
1a75a6c8 | 3021 | case MPOL_PREFERRED: |
fc36b8d3 | 3022 | if (flags & MPOL_F_LOCAL) |
f2a07f40 | 3023 | mode = MPOL_LOCAL; |
53f2556b | 3024 | else |
fc36b8d3 | 3025 | node_set(pol->v.preferred_node, nodes); |
1a75a6c8 | 3026 | break; |
1a75a6c8 | 3027 | case MPOL_BIND: |
1a75a6c8 | 3028 | case MPOL_INTERLEAVE: |
f2a07f40 | 3029 | nodes = pol->v.nodes; |
1a75a6c8 | 3030 | break; |
1a75a6c8 | 3031 | default: |
948927ee DR |
3032 | WARN_ON_ONCE(1); |
3033 | snprintf(p, maxlen, "unknown"); | |
3034 | return; | |
1a75a6c8 CL |
3035 | } |
3036 | ||
b7a9f420 | 3037 | p += snprintf(p, maxlen, "%s", policy_modes[mode]); |
1a75a6c8 | 3038 | |
fc36b8d3 | 3039 | if (flags & MPOL_MODE_FLAGS) { |
948927ee | 3040 | p += snprintf(p, buffer + maxlen - p, "="); |
f5b087b5 | 3041 | |
2291990a LS |
3042 | /* |
3043 | * Currently, the only defined flags are mutually exclusive | |
3044 | */ | |
f5b087b5 | 3045 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
3046 | p += snprintf(p, buffer + maxlen - p, "static"); |
3047 | else if (flags & MPOL_F_RELATIVE_NODES) | |
3048 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
3049 | } |
3050 | ||
9e763e0f TH |
3051 | if (!nodes_empty(nodes)) |
3052 | p += scnprintf(p, buffer + maxlen - p, ":%*pbl", | |
3053 | nodemask_pr_args(&nodes)); | |
1a75a6c8 | 3054 | } |