]>
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 */ |
7858d7bc | 124 | .mode = MPOL_LOCAL, |
1da177e4 LT |
125 | }; |
126 | ||
5606e387 MG |
127 | static struct mempolicy preferred_node_policy[MAX_NUMNODES]; |
128 | ||
b2ca916c DW |
129 | /** |
130 | * numa_map_to_online_node - Find closest online node | |
f6e92f40 | 131 | * @node: Node id to start the search |
b2ca916c DW |
132 | * |
133 | * Lookup the next closest node by distance if @nid is not online. | |
134 | */ | |
135 | int numa_map_to_online_node(int node) | |
136 | { | |
4fcbe96e | 137 | int min_dist = INT_MAX, dist, n, min_node; |
b2ca916c | 138 | |
4fcbe96e DW |
139 | if (node == NUMA_NO_NODE || node_online(node)) |
140 | return node; | |
b2ca916c DW |
141 | |
142 | min_node = node; | |
4fcbe96e DW |
143 | for_each_online_node(n) { |
144 | dist = node_distance(node, n); | |
145 | if (dist < min_dist) { | |
146 | min_dist = dist; | |
147 | min_node = n; | |
b2ca916c DW |
148 | } |
149 | } | |
150 | ||
151 | return min_node; | |
152 | } | |
153 | EXPORT_SYMBOL_GPL(numa_map_to_online_node); | |
154 | ||
74d2c3a0 | 155 | struct mempolicy *get_task_policy(struct task_struct *p) |
5606e387 MG |
156 | { |
157 | struct mempolicy *pol = p->mempolicy; | |
f15ca78e | 158 | int node; |
5606e387 | 159 | |
f15ca78e ON |
160 | if (pol) |
161 | return pol; | |
5606e387 | 162 | |
f15ca78e ON |
163 | node = numa_node_id(); |
164 | if (node != NUMA_NO_NODE) { | |
165 | pol = &preferred_node_policy[node]; | |
166 | /* preferred_node_policy is not initialised early in boot */ | |
167 | if (pol->mode) | |
168 | return pol; | |
5606e387 MG |
169 | } |
170 | ||
f15ca78e | 171 | return &default_policy; |
5606e387 MG |
172 | } |
173 | ||
37012946 DR |
174 | static const struct mempolicy_operations { |
175 | int (*create)(struct mempolicy *pol, const nodemask_t *nodes); | |
213980c0 | 176 | void (*rebind)(struct mempolicy *pol, const nodemask_t *nodes); |
37012946 DR |
177 | } mpol_ops[MPOL_MAX]; |
178 | ||
f5b087b5 DR |
179 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
180 | { | |
6d556294 | 181 | return pol->flags & MPOL_MODE_FLAGS; |
4c50bc01 DR |
182 | } |
183 | ||
184 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
185 | const nodemask_t *rel) | |
186 | { | |
187 | nodemask_t tmp; | |
188 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
189 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
190 | } |
191 | ||
37012946 DR |
192 | static int mpol_new_interleave(struct mempolicy *pol, const nodemask_t *nodes) |
193 | { | |
194 | if (nodes_empty(*nodes)) | |
195 | return -EINVAL; | |
196 | pol->v.nodes = *nodes; | |
197 | return 0; | |
198 | } | |
199 | ||
200 | static int mpol_new_preferred(struct mempolicy *pol, const nodemask_t *nodes) | |
201 | { | |
7858d7bc FT |
202 | if (nodes_empty(*nodes)) |
203 | return -EINVAL; | |
204 | pol->v.preferred_node = first_node(*nodes); | |
37012946 DR |
205 | return 0; |
206 | } | |
207 | ||
208 | static int mpol_new_bind(struct mempolicy *pol, const nodemask_t *nodes) | |
209 | { | |
859f7ef1 | 210 | if (nodes_empty(*nodes)) |
37012946 DR |
211 | return -EINVAL; |
212 | pol->v.nodes = *nodes; | |
213 | return 0; | |
214 | } | |
215 | ||
58568d2a MX |
216 | /* |
217 | * mpol_set_nodemask is called after mpol_new() to set up the nodemask, if | |
218 | * any, for the new policy. mpol_new() has already validated the nodes | |
7858d7bc | 219 | * parameter with respect to the policy mode and flags. |
58568d2a MX |
220 | * |
221 | * Must be called holding task's alloc_lock to protect task's mems_allowed | |
c1e8d7c6 | 222 | * and mempolicy. May also be called holding the mmap_lock for write. |
58568d2a | 223 | */ |
4bfc4495 KH |
224 | static int mpol_set_nodemask(struct mempolicy *pol, |
225 | const nodemask_t *nodes, struct nodemask_scratch *nsc) | |
58568d2a | 226 | { |
58568d2a MX |
227 | int ret; |
228 | ||
7858d7bc FT |
229 | /* |
230 | * Default (pol==NULL) resp. local memory policies are not a | |
231 | * subject of any remapping. They also do not need any special | |
232 | * constructor. | |
233 | */ | |
234 | if (!pol || pol->mode == MPOL_LOCAL) | |
58568d2a | 235 | return 0; |
7858d7bc | 236 | |
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); | |
4bfc4495 | 242 | |
7858d7bc FT |
243 | if (pol->flags & MPOL_F_RELATIVE_NODES) |
244 | mpol_relative_nodemask(&nsc->mask2, nodes, &nsc->mask1); | |
245 | else | |
246 | nodes_and(nsc->mask2, *nodes, nsc->mask1); | |
58568d2a | 247 | |
7858d7bc FT |
248 | if (mpol_store_user_nodemask(pol)) |
249 | pol->w.user_nodemask = *nodes; | |
4bfc4495 | 250 | else |
7858d7bc FT |
251 | pol->w.cpuset_mems_allowed = cpuset_current_mems_allowed; |
252 | ||
253 | ret = mpol_ops[pol->mode].create(pol, &nsc->mask2); | |
58568d2a MX |
254 | return ret; |
255 | } | |
256 | ||
257 | /* | |
258 | * This function just creates a new policy, does some check and simple | |
259 | * initialization. You must invoke mpol_set_nodemask() to set nodes. | |
260 | */ | |
028fec41 DR |
261 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
262 | nodemask_t *nodes) | |
1da177e4 LT |
263 | { |
264 | struct mempolicy *policy; | |
265 | ||
028fec41 | 266 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
00ef2d2f | 267 | mode, flags, nodes ? nodes_addr(*nodes)[0] : NUMA_NO_NODE); |
140d5a49 | 268 | |
3e1f0645 DR |
269 | if (mode == MPOL_DEFAULT) { |
270 | if (nodes && !nodes_empty(*nodes)) | |
37012946 | 271 | return ERR_PTR(-EINVAL); |
d3a71033 | 272 | return NULL; |
37012946 | 273 | } |
3e1f0645 DR |
274 | VM_BUG_ON(!nodes); |
275 | ||
276 | /* | |
277 | * MPOL_PREFERRED cannot be used with MPOL_F_STATIC_NODES or | |
278 | * MPOL_F_RELATIVE_NODES if the nodemask is empty (local allocation). | |
279 | * All other modes require a valid pointer to a non-empty nodemask. | |
280 | */ | |
281 | if (mode == MPOL_PREFERRED) { | |
282 | if (nodes_empty(*nodes)) { | |
283 | if (((flags & MPOL_F_STATIC_NODES) || | |
284 | (flags & MPOL_F_RELATIVE_NODES))) | |
285 | return ERR_PTR(-EINVAL); | |
7858d7bc FT |
286 | |
287 | mode = MPOL_LOCAL; | |
3e1f0645 | 288 | } |
479e2802 | 289 | } else if (mode == MPOL_LOCAL) { |
8d303e44 PK |
290 | if (!nodes_empty(*nodes) || |
291 | (flags & MPOL_F_STATIC_NODES) || | |
292 | (flags & MPOL_F_RELATIVE_NODES)) | |
479e2802 | 293 | return ERR_PTR(-EINVAL); |
3e1f0645 DR |
294 | } else if (nodes_empty(*nodes)) |
295 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
296 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
297 | if (!policy) | |
298 | return ERR_PTR(-ENOMEM); | |
299 | atomic_set(&policy->refcnt, 1); | |
45c4745a | 300 | policy->mode = mode; |
3e1f0645 | 301 | policy->flags = flags; |
37012946 | 302 | |
1da177e4 | 303 | return policy; |
37012946 DR |
304 | } |
305 | ||
52cd3b07 LS |
306 | /* Slow path of a mpol destructor. */ |
307 | void __mpol_put(struct mempolicy *p) | |
308 | { | |
309 | if (!atomic_dec_and_test(&p->refcnt)) | |
310 | return; | |
52cd3b07 LS |
311 | kmem_cache_free(policy_cache, p); |
312 | } | |
313 | ||
213980c0 | 314 | static void mpol_rebind_default(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
315 | { |
316 | } | |
317 | ||
213980c0 | 318 | static void mpol_rebind_nodemask(struct mempolicy *pol, const nodemask_t *nodes) |
37012946 DR |
319 | { |
320 | nodemask_t tmp; | |
321 | ||
322 | if (pol->flags & MPOL_F_STATIC_NODES) | |
323 | nodes_and(tmp, pol->w.user_nodemask, *nodes); | |
324 | else if (pol->flags & MPOL_F_RELATIVE_NODES) | |
325 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, nodes); | |
326 | else { | |
68d68ff6 | 327 | nodes_remap(tmp, pol->v.nodes, pol->w.cpuset_mems_allowed, |
213980c0 | 328 | *nodes); |
29b190fa | 329 | pol->w.cpuset_mems_allowed = *nodes; |
37012946 | 330 | } |
f5b087b5 | 331 | |
708c1bbc MX |
332 | if (nodes_empty(tmp)) |
333 | tmp = *nodes; | |
334 | ||
213980c0 | 335 | pol->v.nodes = tmp; |
37012946 DR |
336 | } |
337 | ||
338 | static void mpol_rebind_preferred(struct mempolicy *pol, | |
213980c0 | 339 | const nodemask_t *nodes) |
37012946 | 340 | { |
7858d7bc | 341 | pol->w.cpuset_mems_allowed = *nodes; |
1da177e4 LT |
342 | } |
343 | ||
708c1bbc MX |
344 | /* |
345 | * mpol_rebind_policy - Migrate a policy to a different set of nodes | |
346 | * | |
c1e8d7c6 | 347 | * Per-vma policies are protected by mmap_lock. Allocations using per-task |
213980c0 VB |
348 | * policies are protected by task->mems_allowed_seq to prevent a premature |
349 | * OOM/allocation failure due to parallel nodemask modification. | |
708c1bbc | 350 | */ |
213980c0 | 351 | static void mpol_rebind_policy(struct mempolicy *pol, const nodemask_t *newmask) |
1d0d2680 | 352 | { |
1d0d2680 DR |
353 | if (!pol) |
354 | return; | |
7858d7bc | 355 | if (!mpol_store_user_nodemask(pol) && |
1d0d2680 DR |
356 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) |
357 | return; | |
708c1bbc | 358 | |
213980c0 | 359 | mpol_ops[pol->mode].rebind(pol, newmask); |
1d0d2680 DR |
360 | } |
361 | ||
362 | /* | |
363 | * Wrapper for mpol_rebind_policy() that just requires task | |
364 | * pointer, and updates task mempolicy. | |
58568d2a MX |
365 | * |
366 | * Called with task's alloc_lock held. | |
1d0d2680 DR |
367 | */ |
368 | ||
213980c0 | 369 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) |
1d0d2680 | 370 | { |
213980c0 | 371 | mpol_rebind_policy(tsk->mempolicy, new); |
1d0d2680 DR |
372 | } |
373 | ||
374 | /* | |
375 | * Rebind each vma in mm to new nodemask. | |
376 | * | |
c1e8d7c6 | 377 | * Call holding a reference to mm. Takes mm->mmap_lock during call. |
1d0d2680 DR |
378 | */ |
379 | ||
380 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
381 | { | |
382 | struct vm_area_struct *vma; | |
383 | ||
d8ed45c5 | 384 | mmap_write_lock(mm); |
1d0d2680 | 385 | for (vma = mm->mmap; vma; vma = vma->vm_next) |
213980c0 | 386 | mpol_rebind_policy(vma->vm_policy, new); |
d8ed45c5 | 387 | mmap_write_unlock(mm); |
1d0d2680 DR |
388 | } |
389 | ||
37012946 DR |
390 | static const struct mempolicy_operations mpol_ops[MPOL_MAX] = { |
391 | [MPOL_DEFAULT] = { | |
392 | .rebind = mpol_rebind_default, | |
393 | }, | |
394 | [MPOL_INTERLEAVE] = { | |
395 | .create = mpol_new_interleave, | |
396 | .rebind = mpol_rebind_nodemask, | |
397 | }, | |
398 | [MPOL_PREFERRED] = { | |
399 | .create = mpol_new_preferred, | |
400 | .rebind = mpol_rebind_preferred, | |
401 | }, | |
402 | [MPOL_BIND] = { | |
403 | .create = mpol_new_bind, | |
404 | .rebind = mpol_rebind_nodemask, | |
405 | }, | |
7858d7bc FT |
406 | [MPOL_LOCAL] = { |
407 | .rebind = mpol_rebind_default, | |
408 | }, | |
37012946 DR |
409 | }; |
410 | ||
a53190a4 | 411 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
fc301289 | 412 | unsigned long flags); |
1a75a6c8 | 413 | |
6f4576e3 NH |
414 | struct queue_pages { |
415 | struct list_head *pagelist; | |
416 | unsigned long flags; | |
417 | nodemask_t *nmask; | |
f18da660 LX |
418 | unsigned long start; |
419 | unsigned long end; | |
420 | struct vm_area_struct *first; | |
6f4576e3 NH |
421 | }; |
422 | ||
88aaa2a1 NH |
423 | /* |
424 | * Check if the page's nid is in qp->nmask. | |
425 | * | |
426 | * If MPOL_MF_INVERT is set in qp->flags, check if the nid is | |
427 | * in the invert of qp->nmask. | |
428 | */ | |
429 | static inline bool queue_pages_required(struct page *page, | |
430 | struct queue_pages *qp) | |
431 | { | |
432 | int nid = page_to_nid(page); | |
433 | unsigned long flags = qp->flags; | |
434 | ||
435 | return node_isset(nid, *qp->nmask) == !(flags & MPOL_MF_INVERT); | |
436 | } | |
437 | ||
a7f40cfe | 438 | /* |
d8835445 | 439 | * queue_pages_pmd() has four possible return values: |
e5947d23 YS |
440 | * 0 - pages are placed on the right node or queued successfully, or |
441 | * special page is met, i.e. huge zero page. | |
d8835445 YS |
442 | * 1 - there is unmovable page, and MPOL_MF_MOVE* & MPOL_MF_STRICT were |
443 | * specified. | |
444 | * 2 - THP was split. | |
445 | * -EIO - is migration entry or only MPOL_MF_STRICT was specified and an | |
446 | * existing page was already on a node that does not follow the | |
447 | * policy. | |
a7f40cfe | 448 | */ |
c8633798 NH |
449 | static int queue_pages_pmd(pmd_t *pmd, spinlock_t *ptl, unsigned long addr, |
450 | unsigned long end, struct mm_walk *walk) | |
959a7e13 | 451 | __releases(ptl) |
c8633798 NH |
452 | { |
453 | int ret = 0; | |
454 | struct page *page; | |
455 | struct queue_pages *qp = walk->private; | |
456 | unsigned long flags; | |
457 | ||
458 | if (unlikely(is_pmd_migration_entry(*pmd))) { | |
a7f40cfe | 459 | ret = -EIO; |
c8633798 NH |
460 | goto unlock; |
461 | } | |
462 | page = pmd_page(*pmd); | |
463 | if (is_huge_zero_page(page)) { | |
464 | spin_unlock(ptl); | |
e5947d23 | 465 | walk->action = ACTION_CONTINUE; |
c8633798 NH |
466 | goto out; |
467 | } | |
d8835445 | 468 | if (!queue_pages_required(page, qp)) |
c8633798 | 469 | goto unlock; |
c8633798 | 470 | |
c8633798 NH |
471 | flags = qp->flags; |
472 | /* go to thp migration */ | |
a7f40cfe | 473 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
a53190a4 YS |
474 | if (!vma_migratable(walk->vma) || |
475 | migrate_page_add(page, qp->pagelist, flags)) { | |
d8835445 | 476 | ret = 1; |
a7f40cfe YS |
477 | goto unlock; |
478 | } | |
a7f40cfe YS |
479 | } else |
480 | ret = -EIO; | |
c8633798 NH |
481 | unlock: |
482 | spin_unlock(ptl); | |
483 | out: | |
484 | return ret; | |
485 | } | |
486 | ||
98094945 NH |
487 | /* |
488 | * Scan through pages checking if pages follow certain conditions, | |
489 | * and move them to the pagelist if they do. | |
d8835445 YS |
490 | * |
491 | * queue_pages_pte_range() has three possible return values: | |
e5947d23 YS |
492 | * 0 - pages are placed on the right node or queued successfully, or |
493 | * special page is met, i.e. zero page. | |
d8835445 YS |
494 | * 1 - there is unmovable page, and MPOL_MF_MOVE* & MPOL_MF_STRICT were |
495 | * specified. | |
496 | * -EIO - only MPOL_MF_STRICT was specified and an existing page was already | |
497 | * on a node that does not follow the policy. | |
98094945 | 498 | */ |
6f4576e3 NH |
499 | static int queue_pages_pte_range(pmd_t *pmd, unsigned long addr, |
500 | unsigned long end, struct mm_walk *walk) | |
1da177e4 | 501 | { |
6f4576e3 NH |
502 | struct vm_area_struct *vma = walk->vma; |
503 | struct page *page; | |
504 | struct queue_pages *qp = walk->private; | |
505 | unsigned long flags = qp->flags; | |
c8633798 | 506 | int ret; |
d8835445 | 507 | bool has_unmovable = false; |
3f088420 | 508 | pte_t *pte, *mapped_pte; |
705e87c0 | 509 | spinlock_t *ptl; |
941150a3 | 510 | |
c8633798 NH |
511 | ptl = pmd_trans_huge_lock(pmd, vma); |
512 | if (ptl) { | |
513 | ret = queue_pages_pmd(pmd, ptl, addr, end, walk); | |
d8835445 | 514 | if (ret != 2) |
a7f40cfe | 515 | return ret; |
248db92d | 516 | } |
d8835445 | 517 | /* THP was split, fall through to pte walk */ |
91612e0d | 518 | |
337d9abf NH |
519 | if (pmd_trans_unstable(pmd)) |
520 | return 0; | |
94723aaf | 521 | |
3f088420 | 522 | mapped_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); |
6f4576e3 | 523 | for (; addr != end; pte++, addr += PAGE_SIZE) { |
91612e0d | 524 | if (!pte_present(*pte)) |
1da177e4 | 525 | continue; |
6aab341e LT |
526 | page = vm_normal_page(vma, addr, *pte); |
527 | if (!page) | |
1da177e4 | 528 | continue; |
053837fc | 529 | /* |
62b61f61 HD |
530 | * vm_normal_page() filters out zero pages, but there might |
531 | * still be PageReserved pages to skip, perhaps in a VDSO. | |
053837fc | 532 | */ |
b79bc0a0 | 533 | if (PageReserved(page)) |
f4598c8b | 534 | continue; |
88aaa2a1 | 535 | if (!queue_pages_required(page, qp)) |
38e35860 | 536 | continue; |
a7f40cfe | 537 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
d8835445 YS |
538 | /* MPOL_MF_STRICT must be specified if we get here */ |
539 | if (!vma_migratable(vma)) { | |
540 | has_unmovable = true; | |
a7f40cfe | 541 | break; |
d8835445 | 542 | } |
a53190a4 YS |
543 | |
544 | /* | |
545 | * Do not abort immediately since there may be | |
546 | * temporary off LRU pages in the range. Still | |
547 | * need migrate other LRU pages. | |
548 | */ | |
549 | if (migrate_page_add(page, qp->pagelist, flags)) | |
550 | has_unmovable = true; | |
a7f40cfe YS |
551 | } else |
552 | break; | |
6f4576e3 | 553 | } |
3f088420 | 554 | pte_unmap_unlock(mapped_pte, ptl); |
6f4576e3 | 555 | cond_resched(); |
d8835445 YS |
556 | |
557 | if (has_unmovable) | |
558 | return 1; | |
559 | ||
a7f40cfe | 560 | return addr != end ? -EIO : 0; |
91612e0d HD |
561 | } |
562 | ||
6f4576e3 NH |
563 | static int queue_pages_hugetlb(pte_t *pte, unsigned long hmask, |
564 | unsigned long addr, unsigned long end, | |
565 | struct mm_walk *walk) | |
e2d8cf40 | 566 | { |
dcf17635 | 567 | int ret = 0; |
e2d8cf40 | 568 | #ifdef CONFIG_HUGETLB_PAGE |
6f4576e3 | 569 | struct queue_pages *qp = walk->private; |
dcf17635 | 570 | unsigned long flags = (qp->flags & MPOL_MF_VALID); |
e2d8cf40 | 571 | struct page *page; |
cb900f41 | 572 | spinlock_t *ptl; |
d4c54919 | 573 | pte_t entry; |
e2d8cf40 | 574 | |
6f4576e3 NH |
575 | ptl = huge_pte_lock(hstate_vma(walk->vma), walk->mm, pte); |
576 | entry = huge_ptep_get(pte); | |
d4c54919 NH |
577 | if (!pte_present(entry)) |
578 | goto unlock; | |
579 | page = pte_page(entry); | |
88aaa2a1 | 580 | if (!queue_pages_required(page, qp)) |
e2d8cf40 | 581 | goto unlock; |
dcf17635 LX |
582 | |
583 | if (flags == MPOL_MF_STRICT) { | |
584 | /* | |
585 | * STRICT alone means only detecting misplaced page and no | |
586 | * need to further check other vma. | |
587 | */ | |
588 | ret = -EIO; | |
589 | goto unlock; | |
590 | } | |
591 | ||
592 | if (!vma_migratable(walk->vma)) { | |
593 | /* | |
594 | * Must be STRICT with MOVE*, otherwise .test_walk() have | |
595 | * stopped walking current vma. | |
596 | * Detecting misplaced page but allow migrating pages which | |
597 | * have been queued. | |
598 | */ | |
599 | ret = 1; | |
600 | goto unlock; | |
601 | } | |
602 | ||
e2d8cf40 NH |
603 | /* With MPOL_MF_MOVE, we migrate only unshared hugepage. */ |
604 | if (flags & (MPOL_MF_MOVE_ALL) || | |
dcf17635 LX |
605 | (flags & MPOL_MF_MOVE && page_mapcount(page) == 1)) { |
606 | if (!isolate_huge_page(page, qp->pagelist) && | |
607 | (flags & MPOL_MF_STRICT)) | |
608 | /* | |
609 | * Failed to isolate page but allow migrating pages | |
610 | * which have been queued. | |
611 | */ | |
612 | ret = 1; | |
613 | } | |
e2d8cf40 | 614 | unlock: |
cb900f41 | 615 | spin_unlock(ptl); |
e2d8cf40 NH |
616 | #else |
617 | BUG(); | |
618 | #endif | |
dcf17635 | 619 | return ret; |
1da177e4 LT |
620 | } |
621 | ||
5877231f | 622 | #ifdef CONFIG_NUMA_BALANCING |
b24f53a0 | 623 | /* |
4b10e7d5 MG |
624 | * This is used to mark a range of virtual addresses to be inaccessible. |
625 | * These are later cleared by a NUMA hinting fault. Depending on these | |
626 | * faults, pages may be migrated for better NUMA placement. | |
627 | * | |
628 | * This is assuming that NUMA faults are handled using PROT_NONE. If | |
629 | * an architecture makes a different choice, it will need further | |
630 | * changes to the core. | |
b24f53a0 | 631 | */ |
4b10e7d5 MG |
632 | unsigned long change_prot_numa(struct vm_area_struct *vma, |
633 | unsigned long addr, unsigned long end) | |
b24f53a0 | 634 | { |
4b10e7d5 | 635 | int nr_updated; |
b24f53a0 | 636 | |
58705444 | 637 | nr_updated = change_protection(vma, addr, end, PAGE_NONE, MM_CP_PROT_NUMA); |
03c5a6e1 MG |
638 | if (nr_updated) |
639 | count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated); | |
b24f53a0 | 640 | |
4b10e7d5 | 641 | return nr_updated; |
b24f53a0 LS |
642 | } |
643 | #else | |
644 | static unsigned long change_prot_numa(struct vm_area_struct *vma, | |
645 | unsigned long addr, unsigned long end) | |
646 | { | |
647 | return 0; | |
648 | } | |
5877231f | 649 | #endif /* CONFIG_NUMA_BALANCING */ |
b24f53a0 | 650 | |
6f4576e3 NH |
651 | static int queue_pages_test_walk(unsigned long start, unsigned long end, |
652 | struct mm_walk *walk) | |
653 | { | |
654 | struct vm_area_struct *vma = walk->vma; | |
655 | struct queue_pages *qp = walk->private; | |
656 | unsigned long endvma = vma->vm_end; | |
657 | unsigned long flags = qp->flags; | |
658 | ||
a18b3ac2 | 659 | /* range check first */ |
ce33135c | 660 | VM_BUG_ON_VMA(!range_in_vma(vma, start, end), vma); |
f18da660 LX |
661 | |
662 | if (!qp->first) { | |
663 | qp->first = vma; | |
664 | if (!(flags & MPOL_MF_DISCONTIG_OK) && | |
665 | (qp->start < vma->vm_start)) | |
666 | /* hole at head side of range */ | |
a18b3ac2 LX |
667 | return -EFAULT; |
668 | } | |
f18da660 LX |
669 | if (!(flags & MPOL_MF_DISCONTIG_OK) && |
670 | ((vma->vm_end < qp->end) && | |
671 | (!vma->vm_next || vma->vm_end < vma->vm_next->vm_start))) | |
672 | /* hole at middle or tail of range */ | |
673 | return -EFAULT; | |
a18b3ac2 | 674 | |
a7f40cfe YS |
675 | /* |
676 | * Need check MPOL_MF_STRICT to return -EIO if possible | |
677 | * regardless of vma_migratable | |
678 | */ | |
679 | if (!vma_migratable(vma) && | |
680 | !(flags & MPOL_MF_STRICT)) | |
48684a65 NH |
681 | return 1; |
682 | ||
6f4576e3 NH |
683 | if (endvma > end) |
684 | endvma = end; | |
6f4576e3 | 685 | |
6f4576e3 NH |
686 | if (flags & MPOL_MF_LAZY) { |
687 | /* Similar to task_numa_work, skip inaccessible VMAs */ | |
3122e80e | 688 | if (!is_vm_hugetlb_page(vma) && vma_is_accessible(vma) && |
4355c018 | 689 | !(vma->vm_flags & VM_MIXEDMAP)) |
6f4576e3 NH |
690 | change_prot_numa(vma, start, endvma); |
691 | return 1; | |
692 | } | |
693 | ||
77bf45e7 | 694 | /* queue pages from current vma */ |
a7f40cfe | 695 | if (flags & MPOL_MF_VALID) |
6f4576e3 NH |
696 | return 0; |
697 | return 1; | |
698 | } | |
699 | ||
7b86ac33 CH |
700 | static const struct mm_walk_ops queue_pages_walk_ops = { |
701 | .hugetlb_entry = queue_pages_hugetlb, | |
702 | .pmd_entry = queue_pages_pte_range, | |
703 | .test_walk = queue_pages_test_walk, | |
704 | }; | |
705 | ||
dc9aa5b9 | 706 | /* |
98094945 NH |
707 | * Walk through page tables and collect pages to be migrated. |
708 | * | |
709 | * If pages found in a given range are on a set of nodes (determined by | |
710 | * @nodes and @flags,) it's isolated and queued to the pagelist which is | |
d8835445 YS |
711 | * passed via @private. |
712 | * | |
713 | * queue_pages_range() has three possible return values: | |
714 | * 1 - there is unmovable page, but MPOL_MF_MOVE* & MPOL_MF_STRICT were | |
715 | * specified. | |
716 | * 0 - queue pages successfully or no misplaced page. | |
a85dfc30 YS |
717 | * errno - i.e. misplaced pages with MPOL_MF_STRICT specified (-EIO) or |
718 | * memory range specified by nodemask and maxnode points outside | |
719 | * your accessible address space (-EFAULT) | |
dc9aa5b9 | 720 | */ |
d05f0cdc | 721 | static int |
98094945 | 722 | queue_pages_range(struct mm_struct *mm, unsigned long start, unsigned long end, |
6f4576e3 NH |
723 | nodemask_t *nodes, unsigned long flags, |
724 | struct list_head *pagelist) | |
1da177e4 | 725 | { |
f18da660 | 726 | int err; |
6f4576e3 NH |
727 | struct queue_pages qp = { |
728 | .pagelist = pagelist, | |
729 | .flags = flags, | |
730 | .nmask = nodes, | |
f18da660 LX |
731 | .start = start, |
732 | .end = end, | |
733 | .first = NULL, | |
6f4576e3 | 734 | }; |
6f4576e3 | 735 | |
f18da660 LX |
736 | err = walk_page_range(mm, start, end, &queue_pages_walk_ops, &qp); |
737 | ||
738 | if (!qp.first) | |
739 | /* whole range in hole */ | |
740 | err = -EFAULT; | |
741 | ||
742 | return err; | |
1da177e4 LT |
743 | } |
744 | ||
869833f2 KM |
745 | /* |
746 | * Apply policy to a single VMA | |
c1e8d7c6 | 747 | * This must be called with the mmap_lock held for writing. |
869833f2 KM |
748 | */ |
749 | static int vma_replace_policy(struct vm_area_struct *vma, | |
750 | struct mempolicy *pol) | |
8d34694c | 751 | { |
869833f2 KM |
752 | int err; |
753 | struct mempolicy *old; | |
754 | struct mempolicy *new; | |
8d34694c KM |
755 | |
756 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", | |
757 | vma->vm_start, vma->vm_end, vma->vm_pgoff, | |
758 | vma->vm_ops, vma->vm_file, | |
759 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
760 | ||
869833f2 KM |
761 | new = mpol_dup(pol); |
762 | if (IS_ERR(new)) | |
763 | return PTR_ERR(new); | |
764 | ||
765 | if (vma->vm_ops && vma->vm_ops->set_policy) { | |
8d34694c | 766 | err = vma->vm_ops->set_policy(vma, new); |
869833f2 KM |
767 | if (err) |
768 | goto err_out; | |
8d34694c | 769 | } |
869833f2 KM |
770 | |
771 | old = vma->vm_policy; | |
c1e8d7c6 | 772 | vma->vm_policy = new; /* protected by mmap_lock */ |
869833f2 KM |
773 | mpol_put(old); |
774 | ||
775 | return 0; | |
776 | err_out: | |
777 | mpol_put(new); | |
8d34694c KM |
778 | return err; |
779 | } | |
780 | ||
1da177e4 | 781 | /* Step 2: apply policy to a range and do splits. */ |
9d8cebd4 KM |
782 | static int mbind_range(struct mm_struct *mm, unsigned long start, |
783 | unsigned long end, struct mempolicy *new_pol) | |
1da177e4 LT |
784 | { |
785 | struct vm_area_struct *next; | |
9d8cebd4 KM |
786 | struct vm_area_struct *prev; |
787 | struct vm_area_struct *vma; | |
788 | int err = 0; | |
e26a5114 | 789 | pgoff_t pgoff; |
9d8cebd4 KM |
790 | unsigned long vmstart; |
791 | unsigned long vmend; | |
1da177e4 | 792 | |
097d5910 | 793 | vma = find_vma(mm, start); |
f18da660 | 794 | VM_BUG_ON(!vma); |
9d8cebd4 | 795 | |
097d5910 | 796 | prev = vma->vm_prev; |
e26a5114 KM |
797 | if (start > vma->vm_start) |
798 | prev = vma; | |
799 | ||
9d8cebd4 | 800 | for (; vma && vma->vm_start < end; prev = vma, vma = next) { |
1da177e4 | 801 | next = vma->vm_next; |
9d8cebd4 KM |
802 | vmstart = max(start, vma->vm_start); |
803 | vmend = min(end, vma->vm_end); | |
804 | ||
e26a5114 KM |
805 | if (mpol_equal(vma_policy(vma), new_pol)) |
806 | continue; | |
807 | ||
808 | pgoff = vma->vm_pgoff + | |
809 | ((vmstart - vma->vm_start) >> PAGE_SHIFT); | |
9d8cebd4 | 810 | prev = vma_merge(mm, prev, vmstart, vmend, vma->vm_flags, |
19a809af AA |
811 | vma->anon_vma, vma->vm_file, pgoff, |
812 | new_pol, vma->vm_userfaultfd_ctx); | |
9d8cebd4 KM |
813 | if (prev) { |
814 | vma = prev; | |
815 | next = vma->vm_next; | |
3964acd0 ON |
816 | if (mpol_equal(vma_policy(vma), new_pol)) |
817 | continue; | |
818 | /* vma_merge() joined vma && vma->next, case 8 */ | |
819 | goto replace; | |
9d8cebd4 KM |
820 | } |
821 | if (vma->vm_start != vmstart) { | |
822 | err = split_vma(vma->vm_mm, vma, vmstart, 1); | |
823 | if (err) | |
824 | goto out; | |
825 | } | |
826 | if (vma->vm_end != vmend) { | |
827 | err = split_vma(vma->vm_mm, vma, vmend, 0); | |
828 | if (err) | |
829 | goto out; | |
830 | } | |
3964acd0 | 831 | replace: |
869833f2 | 832 | err = vma_replace_policy(vma, new_pol); |
8d34694c KM |
833 | if (err) |
834 | goto out; | |
1da177e4 | 835 | } |
9d8cebd4 KM |
836 | |
837 | out: | |
1da177e4 LT |
838 | return err; |
839 | } | |
840 | ||
1da177e4 | 841 | /* Set the process memory policy */ |
028fec41 DR |
842 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
843 | nodemask_t *nodes) | |
1da177e4 | 844 | { |
58568d2a | 845 | struct mempolicy *new, *old; |
4bfc4495 | 846 | NODEMASK_SCRATCH(scratch); |
58568d2a | 847 | int ret; |
1da177e4 | 848 | |
4bfc4495 KH |
849 | if (!scratch) |
850 | return -ENOMEM; | |
f4e53d91 | 851 | |
4bfc4495 KH |
852 | new = mpol_new(mode, flags, nodes); |
853 | if (IS_ERR(new)) { | |
854 | ret = PTR_ERR(new); | |
855 | goto out; | |
856 | } | |
2c7c3a7d | 857 | |
bda420b9 HY |
858 | if (flags & MPOL_F_NUMA_BALANCING) { |
859 | if (new && new->mode == MPOL_BIND) { | |
860 | new->flags |= (MPOL_F_MOF | MPOL_F_MORON); | |
861 | } else { | |
862 | ret = -EINVAL; | |
863 | mpol_put(new); | |
864 | goto out; | |
865 | } | |
866 | } | |
867 | ||
4bfc4495 | 868 | ret = mpol_set_nodemask(new, nodes, scratch); |
58568d2a | 869 | if (ret) { |
58568d2a | 870 | mpol_put(new); |
4bfc4495 | 871 | goto out; |
58568d2a | 872 | } |
78b132e9 | 873 | task_lock(current); |
58568d2a | 874 | old = current->mempolicy; |
1da177e4 | 875 | current->mempolicy = new; |
45816682 VB |
876 | if (new && new->mode == MPOL_INTERLEAVE) |
877 | current->il_prev = MAX_NUMNODES-1; | |
58568d2a | 878 | task_unlock(current); |
58568d2a | 879 | mpol_put(old); |
4bfc4495 KH |
880 | ret = 0; |
881 | out: | |
882 | NODEMASK_SCRATCH_FREE(scratch); | |
883 | return ret; | |
1da177e4 LT |
884 | } |
885 | ||
bea904d5 LS |
886 | /* |
887 | * Return nodemask for policy for get_mempolicy() query | |
58568d2a MX |
888 | * |
889 | * Called with task's alloc_lock held | |
bea904d5 LS |
890 | */ |
891 | static void get_policy_nodemask(struct mempolicy *p, nodemask_t *nodes) | |
1da177e4 | 892 | { |
dfcd3c0d | 893 | nodes_clear(*nodes); |
bea904d5 LS |
894 | if (p == &default_policy) |
895 | return; | |
896 | ||
45c4745a | 897 | switch (p->mode) { |
19770b32 | 898 | case MPOL_BIND: |
1da177e4 | 899 | case MPOL_INTERLEAVE: |
dfcd3c0d | 900 | *nodes = p->v.nodes; |
1da177e4 | 901 | break; |
7858d7bc FT |
902 | case MPOL_LOCAL: |
903 | /* return empty node mask for local allocation */ | |
904 | break; | |
905 | ||
1da177e4 | 906 | case MPOL_PREFERRED: |
7858d7bc | 907 | node_set(p->v.preferred_node, *nodes); |
1da177e4 LT |
908 | break; |
909 | default: | |
910 | BUG(); | |
911 | } | |
912 | } | |
913 | ||
3b9aadf7 | 914 | static int lookup_node(struct mm_struct *mm, unsigned long addr) |
1da177e4 | 915 | { |
ba841078 | 916 | struct page *p = NULL; |
1da177e4 LT |
917 | int err; |
918 | ||
3b9aadf7 AA |
919 | int locked = 1; |
920 | err = get_user_pages_locked(addr & PAGE_MASK, 1, 0, &p, &locked); | |
2d3a36a4 | 921 | if (err > 0) { |
1da177e4 LT |
922 | err = page_to_nid(p); |
923 | put_page(p); | |
924 | } | |
3b9aadf7 | 925 | if (locked) |
d8ed45c5 | 926 | mmap_read_unlock(mm); |
1da177e4 LT |
927 | return err; |
928 | } | |
929 | ||
1da177e4 | 930 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
931 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
932 | unsigned long addr, unsigned long flags) | |
1da177e4 | 933 | { |
8bccd85f | 934 | int err; |
1da177e4 LT |
935 | struct mm_struct *mm = current->mm; |
936 | struct vm_area_struct *vma = NULL; | |
3b9aadf7 | 937 | struct mempolicy *pol = current->mempolicy, *pol_refcount = NULL; |
1da177e4 | 938 | |
754af6f5 LS |
939 | if (flags & |
940 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 941 | return -EINVAL; |
754af6f5 LS |
942 | |
943 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
944 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
945 | return -EINVAL; | |
946 | *policy = 0; /* just so it's initialized */ | |
58568d2a | 947 | task_lock(current); |
754af6f5 | 948 | *nmask = cpuset_current_mems_allowed; |
58568d2a | 949 | task_unlock(current); |
754af6f5 LS |
950 | return 0; |
951 | } | |
952 | ||
1da177e4 | 953 | if (flags & MPOL_F_ADDR) { |
bea904d5 LS |
954 | /* |
955 | * Do NOT fall back to task policy if the | |
956 | * vma/shared policy at addr is NULL. We | |
957 | * want to return MPOL_DEFAULT in this case. | |
958 | */ | |
d8ed45c5 | 959 | mmap_read_lock(mm); |
33e3575c | 960 | vma = vma_lookup(mm, addr); |
1da177e4 | 961 | if (!vma) { |
d8ed45c5 | 962 | mmap_read_unlock(mm); |
1da177e4 LT |
963 | return -EFAULT; |
964 | } | |
965 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
966 | pol = vma->vm_ops->get_policy(vma, addr); | |
967 | else | |
968 | pol = vma->vm_policy; | |
969 | } else if (addr) | |
970 | return -EINVAL; | |
971 | ||
972 | if (!pol) | |
bea904d5 | 973 | pol = &default_policy; /* indicates default behavior */ |
1da177e4 LT |
974 | |
975 | if (flags & MPOL_F_NODE) { | |
976 | if (flags & MPOL_F_ADDR) { | |
3b9aadf7 AA |
977 | /* |
978 | * Take a refcount on the mpol, lookup_node() | |
baf2f90b | 979 | * will drop the mmap_lock, so after calling |
3b9aadf7 AA |
980 | * lookup_node() only "pol" remains valid, "vma" |
981 | * is stale. | |
982 | */ | |
983 | pol_refcount = pol; | |
984 | vma = NULL; | |
985 | mpol_get(pol); | |
986 | err = lookup_node(mm, addr); | |
1da177e4 LT |
987 | if (err < 0) |
988 | goto out; | |
8bccd85f | 989 | *policy = err; |
1da177e4 | 990 | } else if (pol == current->mempolicy && |
45c4745a | 991 | pol->mode == MPOL_INTERLEAVE) { |
45816682 | 992 | *policy = next_node_in(current->il_prev, pol->v.nodes); |
1da177e4 LT |
993 | } else { |
994 | err = -EINVAL; | |
995 | goto out; | |
996 | } | |
bea904d5 LS |
997 | } else { |
998 | *policy = pol == &default_policy ? MPOL_DEFAULT : | |
999 | pol->mode; | |
d79df630 DR |
1000 | /* |
1001 | * Internal mempolicy flags must be masked off before exposing | |
1002 | * the policy to userspace. | |
1003 | */ | |
1004 | *policy |= (pol->flags & MPOL_MODE_FLAGS); | |
bea904d5 | 1005 | } |
1da177e4 | 1006 | |
1da177e4 | 1007 | err = 0; |
58568d2a | 1008 | if (nmask) { |
c6b6ef8b LS |
1009 | if (mpol_store_user_nodemask(pol)) { |
1010 | *nmask = pol->w.user_nodemask; | |
1011 | } else { | |
1012 | task_lock(current); | |
1013 | get_policy_nodemask(pol, nmask); | |
1014 | task_unlock(current); | |
1015 | } | |
58568d2a | 1016 | } |
1da177e4 LT |
1017 | |
1018 | out: | |
52cd3b07 | 1019 | mpol_cond_put(pol); |
1da177e4 | 1020 | if (vma) |
d8ed45c5 | 1021 | mmap_read_unlock(mm); |
3b9aadf7 AA |
1022 | if (pol_refcount) |
1023 | mpol_put(pol_refcount); | |
1da177e4 LT |
1024 | return err; |
1025 | } | |
1026 | ||
b20a3503 | 1027 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 | 1028 | /* |
c8633798 | 1029 | * page migration, thp tail pages can be passed. |
6ce3c4c0 | 1030 | */ |
a53190a4 | 1031 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
fc301289 | 1032 | unsigned long flags) |
6ce3c4c0 | 1033 | { |
c8633798 | 1034 | struct page *head = compound_head(page); |
6ce3c4c0 | 1035 | /* |
fc301289 | 1036 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 1037 | */ |
c8633798 NH |
1038 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(head) == 1) { |
1039 | if (!isolate_lru_page(head)) { | |
1040 | list_add_tail(&head->lru, pagelist); | |
1041 | mod_node_page_state(page_pgdat(head), | |
9de4f22a | 1042 | NR_ISOLATED_ANON + page_is_file_lru(head), |
6c357848 | 1043 | thp_nr_pages(head)); |
a53190a4 YS |
1044 | } else if (flags & MPOL_MF_STRICT) { |
1045 | /* | |
1046 | * Non-movable page may reach here. And, there may be | |
1047 | * temporary off LRU pages or non-LRU movable pages. | |
1048 | * Treat them as unmovable pages since they can't be | |
1049 | * isolated, so they can't be moved at the moment. It | |
1050 | * should return -EIO for this case too. | |
1051 | */ | |
1052 | return -EIO; | |
62695a84 NP |
1053 | } |
1054 | } | |
a53190a4 YS |
1055 | |
1056 | return 0; | |
7e2ab150 | 1057 | } |
6ce3c4c0 | 1058 | |
7e2ab150 CL |
1059 | /* |
1060 | * Migrate pages from one node to a target node. | |
1061 | * Returns error or the number of pages not migrated. | |
1062 | */ | |
dbcb0f19 AB |
1063 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
1064 | int flags) | |
7e2ab150 CL |
1065 | { |
1066 | nodemask_t nmask; | |
1067 | LIST_HEAD(pagelist); | |
1068 | int err = 0; | |
a0976311 JK |
1069 | struct migration_target_control mtc = { |
1070 | .nid = dest, | |
1071 | .gfp_mask = GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, | |
1072 | }; | |
7e2ab150 CL |
1073 | |
1074 | nodes_clear(nmask); | |
1075 | node_set(source, nmask); | |
6ce3c4c0 | 1076 | |
08270807 MK |
1077 | /* |
1078 | * This does not "check" the range but isolates all pages that | |
1079 | * need migration. Between passing in the full user address | |
1080 | * space range and MPOL_MF_DISCONTIG_OK, this call can not fail. | |
1081 | */ | |
1082 | VM_BUG_ON(!(flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))); | |
98094945 | 1083 | queue_pages_range(mm, mm->mmap->vm_start, mm->task_size, &nmask, |
7e2ab150 CL |
1084 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); |
1085 | ||
cf608ac1 | 1086 | if (!list_empty(&pagelist)) { |
a0976311 JK |
1087 | err = migrate_pages(&pagelist, alloc_migration_target, NULL, |
1088 | (unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL); | |
cf608ac1 | 1089 | if (err) |
e2d8cf40 | 1090 | putback_movable_pages(&pagelist); |
cf608ac1 | 1091 | } |
95a402c3 | 1092 | |
7e2ab150 | 1093 | return err; |
6ce3c4c0 CL |
1094 | } |
1095 | ||
39743889 | 1096 | /* |
7e2ab150 CL |
1097 | * Move pages between the two nodesets so as to preserve the physical |
1098 | * layout as much as possible. | |
39743889 CL |
1099 | * |
1100 | * Returns the number of page that could not be moved. | |
1101 | */ | |
0ce72d4f AM |
1102 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1103 | const nodemask_t *to, int flags) | |
39743889 | 1104 | { |
7e2ab150 | 1105 | int busy = 0; |
f555befd | 1106 | int err = 0; |
7e2ab150 | 1107 | nodemask_t tmp; |
39743889 | 1108 | |
361a2a22 | 1109 | lru_cache_disable(); |
0aedadf9 | 1110 | |
d8ed45c5 | 1111 | mmap_read_lock(mm); |
39743889 | 1112 | |
da0aa138 KM |
1113 | /* |
1114 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
1115 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
1116 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
1117 | * The pair of nodemasks 'to' and 'from' define the map. | |
1118 | * | |
1119 | * If no pair of bits is found that way, fallback to picking some | |
1120 | * pair of 'source' and 'dest' bits that are not the same. If the | |
1121 | * 'source' and 'dest' bits are the same, this represents a node | |
1122 | * that will be migrating to itself, so no pages need move. | |
1123 | * | |
1124 | * If no bits are left in 'tmp', or if all remaining bits left | |
1125 | * in 'tmp' correspond to the same bit in 'to', return false | |
1126 | * (nothing left to migrate). | |
1127 | * | |
1128 | * This lets us pick a pair of nodes to migrate between, such that | |
1129 | * if possible the dest node is not already occupied by some other | |
1130 | * source node, minimizing the risk of overloading the memory on a | |
1131 | * node that would happen if we migrated incoming memory to a node | |
1132 | * before migrating outgoing memory source that same node. | |
1133 | * | |
1134 | * A single scan of tmp is sufficient. As we go, we remember the | |
1135 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
1136 | * that not only moved, but what's better, moved to an empty slot | |
1137 | * (d is not set in tmp), then we break out then, with that pair. | |
ae0e47f0 | 1138 | * Otherwise when we finish scanning from_tmp, we at least have the |
da0aa138 KM |
1139 | * most recent <s, d> pair that moved. If we get all the way through |
1140 | * the scan of tmp without finding any node that moved, much less | |
1141 | * moved to an empty node, then there is nothing left worth migrating. | |
1142 | */ | |
d4984711 | 1143 | |
0ce72d4f | 1144 | tmp = *from; |
7e2ab150 | 1145 | while (!nodes_empty(tmp)) { |
68d68ff6 | 1146 | int s, d; |
b76ac7e7 | 1147 | int source = NUMA_NO_NODE; |
7e2ab150 CL |
1148 | int dest = 0; |
1149 | ||
1150 | for_each_node_mask(s, tmp) { | |
4a5b18cc LW |
1151 | |
1152 | /* | |
1153 | * do_migrate_pages() tries to maintain the relative | |
1154 | * node relationship of the pages established between | |
1155 | * threads and memory areas. | |
1156 | * | |
1157 | * However if the number of source nodes is not equal to | |
1158 | * the number of destination nodes we can not preserve | |
1159 | * this node relative relationship. In that case, skip | |
1160 | * copying memory from a node that is in the destination | |
1161 | * mask. | |
1162 | * | |
1163 | * Example: [2,3,4] -> [3,4,5] moves everything. | |
1164 | * [0-7] - > [3,4,5] moves only 0,1,2,6,7. | |
1165 | */ | |
1166 | ||
0ce72d4f AM |
1167 | if ((nodes_weight(*from) != nodes_weight(*to)) && |
1168 | (node_isset(s, *to))) | |
4a5b18cc LW |
1169 | continue; |
1170 | ||
0ce72d4f | 1171 | d = node_remap(s, *from, *to); |
7e2ab150 CL |
1172 | if (s == d) |
1173 | continue; | |
1174 | ||
1175 | source = s; /* Node moved. Memorize */ | |
1176 | dest = d; | |
1177 | ||
1178 | /* dest not in remaining from nodes? */ | |
1179 | if (!node_isset(dest, tmp)) | |
1180 | break; | |
1181 | } | |
b76ac7e7 | 1182 | if (source == NUMA_NO_NODE) |
7e2ab150 CL |
1183 | break; |
1184 | ||
1185 | node_clear(source, tmp); | |
1186 | err = migrate_to_node(mm, source, dest, flags); | |
1187 | if (err > 0) | |
1188 | busy += err; | |
1189 | if (err < 0) | |
1190 | break; | |
39743889 | 1191 | } |
d8ed45c5 | 1192 | mmap_read_unlock(mm); |
d479960e | 1193 | |
361a2a22 | 1194 | lru_cache_enable(); |
7e2ab150 CL |
1195 | if (err < 0) |
1196 | return err; | |
1197 | return busy; | |
b20a3503 CL |
1198 | |
1199 | } | |
1200 | ||
3ad33b24 LS |
1201 | /* |
1202 | * Allocate a new page for page migration based on vma policy. | |
d05f0cdc | 1203 | * Start by assuming the page is mapped by the same vma as contains @start. |
3ad33b24 LS |
1204 | * Search forward from there, if not. N.B., this assumes that the |
1205 | * list of pages handed to migrate_pages()--which is how we get here-- | |
1206 | * is in virtual address order. | |
1207 | */ | |
666feb21 | 1208 | static struct page *new_page(struct page *page, unsigned long start) |
95a402c3 | 1209 | { |
d05f0cdc | 1210 | struct vm_area_struct *vma; |
3f649ab7 | 1211 | unsigned long address; |
95a402c3 | 1212 | |
d05f0cdc | 1213 | vma = find_vma(current->mm, start); |
3ad33b24 LS |
1214 | while (vma) { |
1215 | address = page_address_in_vma(page, vma); | |
1216 | if (address != -EFAULT) | |
1217 | break; | |
1218 | vma = vma->vm_next; | |
1219 | } | |
11c731e8 WL |
1220 | |
1221 | if (PageHuge(page)) { | |
389c8178 MH |
1222 | return alloc_huge_page_vma(page_hstate(compound_head(page)), |
1223 | vma, address); | |
94723aaf | 1224 | } else if (PageTransHuge(page)) { |
c8633798 NH |
1225 | struct page *thp; |
1226 | ||
19deb769 DR |
1227 | thp = alloc_hugepage_vma(GFP_TRANSHUGE, vma, address, |
1228 | HPAGE_PMD_ORDER); | |
c8633798 NH |
1229 | if (!thp) |
1230 | return NULL; | |
1231 | prep_transhuge_page(thp); | |
1232 | return thp; | |
11c731e8 | 1233 | } |
0bf598d8 | 1234 | /* |
11c731e8 | 1235 | * if !vma, alloc_page_vma() will use task or system default policy |
0bf598d8 | 1236 | */ |
0f556856 MH |
1237 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE | __GFP_RETRY_MAYFAIL, |
1238 | vma, address); | |
95a402c3 | 1239 | } |
b20a3503 CL |
1240 | #else |
1241 | ||
a53190a4 | 1242 | static int migrate_page_add(struct page *page, struct list_head *pagelist, |
b20a3503 CL |
1243 | unsigned long flags) |
1244 | { | |
a53190a4 | 1245 | return -EIO; |
39743889 CL |
1246 | } |
1247 | ||
0ce72d4f AM |
1248 | int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from, |
1249 | const nodemask_t *to, int flags) | |
b20a3503 CL |
1250 | { |
1251 | return -ENOSYS; | |
1252 | } | |
95a402c3 | 1253 | |
666feb21 | 1254 | static struct page *new_page(struct page *page, unsigned long start) |
95a402c3 CL |
1255 | { |
1256 | return NULL; | |
1257 | } | |
b20a3503 CL |
1258 | #endif |
1259 | ||
dbcb0f19 | 1260 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
1261 | unsigned short mode, unsigned short mode_flags, |
1262 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 | 1263 | { |
6ce3c4c0 CL |
1264 | struct mm_struct *mm = current->mm; |
1265 | struct mempolicy *new; | |
1266 | unsigned long end; | |
1267 | int err; | |
d8835445 | 1268 | int ret; |
6ce3c4c0 CL |
1269 | LIST_HEAD(pagelist); |
1270 | ||
b24f53a0 | 1271 | if (flags & ~(unsigned long)MPOL_MF_VALID) |
6ce3c4c0 | 1272 | return -EINVAL; |
74c00241 | 1273 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
1274 | return -EPERM; |
1275 | ||
1276 | if (start & ~PAGE_MASK) | |
1277 | return -EINVAL; | |
1278 | ||
1279 | if (mode == MPOL_DEFAULT) | |
1280 | flags &= ~MPOL_MF_STRICT; | |
1281 | ||
1282 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
1283 | end = start + len; | |
1284 | ||
1285 | if (end < start) | |
1286 | return -EINVAL; | |
1287 | if (end == start) | |
1288 | return 0; | |
1289 | ||
028fec41 | 1290 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
1291 | if (IS_ERR(new)) |
1292 | return PTR_ERR(new); | |
1293 | ||
b24f53a0 LS |
1294 | if (flags & MPOL_MF_LAZY) |
1295 | new->flags |= MPOL_F_MOF; | |
1296 | ||
6ce3c4c0 CL |
1297 | /* |
1298 | * If we are using the default policy then operation | |
1299 | * on discontinuous address spaces is okay after all | |
1300 | */ | |
1301 | if (!new) | |
1302 | flags |= MPOL_MF_DISCONTIG_OK; | |
1303 | ||
028fec41 DR |
1304 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
1305 | start, start + len, mode, mode_flags, | |
00ef2d2f | 1306 | nmask ? nodes_addr(*nmask)[0] : NUMA_NO_NODE); |
6ce3c4c0 | 1307 | |
0aedadf9 CL |
1308 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
1309 | ||
361a2a22 | 1310 | lru_cache_disable(); |
0aedadf9 | 1311 | } |
4bfc4495 KH |
1312 | { |
1313 | NODEMASK_SCRATCH(scratch); | |
1314 | if (scratch) { | |
d8ed45c5 | 1315 | mmap_write_lock(mm); |
4bfc4495 | 1316 | err = mpol_set_nodemask(new, nmask, scratch); |
4bfc4495 | 1317 | if (err) |
d8ed45c5 | 1318 | mmap_write_unlock(mm); |
4bfc4495 KH |
1319 | } else |
1320 | err = -ENOMEM; | |
1321 | NODEMASK_SCRATCH_FREE(scratch); | |
1322 | } | |
b05ca738 KM |
1323 | if (err) |
1324 | goto mpol_out; | |
1325 | ||
d8835445 | 1326 | ret = queue_pages_range(mm, start, end, nmask, |
6ce3c4c0 | 1327 | flags | MPOL_MF_INVERT, &pagelist); |
d8835445 YS |
1328 | |
1329 | if (ret < 0) { | |
a85dfc30 | 1330 | err = ret; |
d8835445 YS |
1331 | goto up_out; |
1332 | } | |
1333 | ||
1334 | err = mbind_range(mm, start, end, new); | |
7e2ab150 | 1335 | |
b24f53a0 LS |
1336 | if (!err) { |
1337 | int nr_failed = 0; | |
1338 | ||
cf608ac1 | 1339 | if (!list_empty(&pagelist)) { |
b24f53a0 | 1340 | WARN_ON_ONCE(flags & MPOL_MF_LAZY); |
d05f0cdc HD |
1341 | nr_failed = migrate_pages(&pagelist, new_page, NULL, |
1342 | start, MIGRATE_SYNC, MR_MEMPOLICY_MBIND); | |
cf608ac1 | 1343 | if (nr_failed) |
74060e4d | 1344 | putback_movable_pages(&pagelist); |
cf608ac1 | 1345 | } |
6ce3c4c0 | 1346 | |
d8835445 | 1347 | if ((ret > 0) || (nr_failed && (flags & MPOL_MF_STRICT))) |
6ce3c4c0 | 1348 | err = -EIO; |
a85dfc30 | 1349 | } else { |
d8835445 | 1350 | up_out: |
a85dfc30 YS |
1351 | if (!list_empty(&pagelist)) |
1352 | putback_movable_pages(&pagelist); | |
1353 | } | |
1354 | ||
d8ed45c5 | 1355 | mmap_write_unlock(mm); |
d8835445 | 1356 | mpol_out: |
f0be3d32 | 1357 | mpol_put(new); |
d479960e | 1358 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
361a2a22 | 1359 | lru_cache_enable(); |
6ce3c4c0 CL |
1360 | return err; |
1361 | } | |
1362 | ||
8bccd85f CL |
1363 | /* |
1364 | * User space interface with variable sized bitmaps for nodelists. | |
1365 | */ | |
1366 | ||
1367 | /* Copy a node mask from user space. */ | |
39743889 | 1368 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
1369 | unsigned long maxnode) |
1370 | { | |
1371 | unsigned long k; | |
56521e7a | 1372 | unsigned long t; |
8bccd85f CL |
1373 | unsigned long nlongs; |
1374 | unsigned long endmask; | |
1375 | ||
1376 | --maxnode; | |
1377 | nodes_clear(*nodes); | |
1378 | if (maxnode == 0 || !nmask) | |
1379 | return 0; | |
a9c930ba | 1380 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 1381 | return -EINVAL; |
8bccd85f CL |
1382 | |
1383 | nlongs = BITS_TO_LONGS(maxnode); | |
1384 | if ((maxnode % BITS_PER_LONG) == 0) | |
1385 | endmask = ~0UL; | |
1386 | else | |
1387 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
1388 | ||
56521e7a YX |
1389 | /* |
1390 | * When the user specified more nodes than supported just check | |
1391 | * if the non supported part is all zero. | |
1392 | * | |
1393 | * If maxnode have more longs than MAX_NUMNODES, check | |
1394 | * the bits in that area first. And then go through to | |
1395 | * check the rest bits which equal or bigger than MAX_NUMNODES. | |
1396 | * Otherwise, just check bits [MAX_NUMNODES, maxnode). | |
1397 | */ | |
8bccd85f | 1398 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { |
8bccd85f | 1399 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { |
8bccd85f CL |
1400 | if (get_user(t, nmask + k)) |
1401 | return -EFAULT; | |
1402 | if (k == nlongs - 1) { | |
1403 | if (t & endmask) | |
1404 | return -EINVAL; | |
1405 | } else if (t) | |
1406 | return -EINVAL; | |
1407 | } | |
1408 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
1409 | endmask = ~0UL; | |
1410 | } | |
1411 | ||
56521e7a YX |
1412 | if (maxnode > MAX_NUMNODES && MAX_NUMNODES % BITS_PER_LONG != 0) { |
1413 | unsigned long valid_mask = endmask; | |
1414 | ||
1415 | valid_mask &= ~((1UL << (MAX_NUMNODES % BITS_PER_LONG)) - 1); | |
1416 | if (get_user(t, nmask + nlongs - 1)) | |
1417 | return -EFAULT; | |
1418 | if (t & valid_mask) | |
1419 | return -EINVAL; | |
1420 | } | |
1421 | ||
8bccd85f CL |
1422 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) |
1423 | return -EFAULT; | |
1424 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
1425 | return 0; | |
1426 | } | |
1427 | ||
1428 | /* Copy a kernel node mask to user space */ | |
1429 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
1430 | nodemask_t *nodes) | |
1431 | { | |
1432 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
050c17f2 | 1433 | unsigned int nbytes = BITS_TO_LONGS(nr_node_ids) * sizeof(long); |
8bccd85f CL |
1434 | |
1435 | if (copy > nbytes) { | |
1436 | if (copy > PAGE_SIZE) | |
1437 | return -EINVAL; | |
1438 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
1439 | return -EFAULT; | |
1440 | copy = nbytes; | |
1441 | } | |
1442 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
1443 | } | |
1444 | ||
95837924 FT |
1445 | /* Basic parameter sanity check used by both mbind() and set_mempolicy() */ |
1446 | static inline int sanitize_mpol_flags(int *mode, unsigned short *flags) | |
1447 | { | |
1448 | *flags = *mode & MPOL_MODE_FLAGS; | |
1449 | *mode &= ~MPOL_MODE_FLAGS; | |
1450 | if ((unsigned int)(*mode) >= MPOL_MAX) | |
1451 | return -EINVAL; | |
1452 | if ((*flags & MPOL_F_STATIC_NODES) && (*flags & MPOL_F_RELATIVE_NODES)) | |
1453 | return -EINVAL; | |
1454 | ||
1455 | return 0; | |
1456 | } | |
1457 | ||
e7dc9ad6 DB |
1458 | static long kernel_mbind(unsigned long start, unsigned long len, |
1459 | unsigned long mode, const unsigned long __user *nmask, | |
1460 | unsigned long maxnode, unsigned int flags) | |
8bccd85f | 1461 | { |
95837924 | 1462 | unsigned short mode_flags; |
8bccd85f | 1463 | nodemask_t nodes; |
95837924 | 1464 | int lmode = mode; |
8bccd85f CL |
1465 | int err; |
1466 | ||
057d3389 | 1467 | start = untagged_addr(start); |
95837924 FT |
1468 | err = sanitize_mpol_flags(&lmode, &mode_flags); |
1469 | if (err) | |
1470 | return err; | |
1471 | ||
8bccd85f CL |
1472 | err = get_nodes(&nodes, nmask, maxnode); |
1473 | if (err) | |
1474 | return err; | |
95837924 FT |
1475 | |
1476 | return do_mbind(start, len, lmode, mode_flags, &nodes, flags); | |
8bccd85f CL |
1477 | } |
1478 | ||
e7dc9ad6 DB |
1479 | SYSCALL_DEFINE6(mbind, unsigned long, start, unsigned long, len, |
1480 | unsigned long, mode, const unsigned long __user *, nmask, | |
1481 | unsigned long, maxnode, unsigned int, flags) | |
1482 | { | |
1483 | return kernel_mbind(start, len, mode, nmask, maxnode, flags); | |
1484 | } | |
1485 | ||
8bccd85f | 1486 | /* Set the process memory policy */ |
af03c4ac DB |
1487 | static long kernel_set_mempolicy(int mode, const unsigned long __user *nmask, |
1488 | unsigned long maxnode) | |
8bccd85f | 1489 | { |
95837924 | 1490 | unsigned short mode_flags; |
8bccd85f | 1491 | nodemask_t nodes; |
95837924 FT |
1492 | int lmode = mode; |
1493 | int err; | |
1494 | ||
1495 | err = sanitize_mpol_flags(&lmode, &mode_flags); | |
1496 | if (err) | |
1497 | return err; | |
8bccd85f | 1498 | |
8bccd85f CL |
1499 | err = get_nodes(&nodes, nmask, maxnode); |
1500 | if (err) | |
1501 | return err; | |
95837924 FT |
1502 | |
1503 | return do_set_mempolicy(lmode, mode_flags, &nodes); | |
8bccd85f CL |
1504 | } |
1505 | ||
af03c4ac DB |
1506 | SYSCALL_DEFINE3(set_mempolicy, int, mode, const unsigned long __user *, nmask, |
1507 | unsigned long, maxnode) | |
1508 | { | |
1509 | return kernel_set_mempolicy(mode, nmask, maxnode); | |
1510 | } | |
1511 | ||
b6e9b0ba DB |
1512 | static int kernel_migrate_pages(pid_t pid, unsigned long maxnode, |
1513 | const unsigned long __user *old_nodes, | |
1514 | const unsigned long __user *new_nodes) | |
39743889 | 1515 | { |
596d7cfa | 1516 | struct mm_struct *mm = NULL; |
39743889 | 1517 | struct task_struct *task; |
39743889 CL |
1518 | nodemask_t task_nodes; |
1519 | int err; | |
596d7cfa KM |
1520 | nodemask_t *old; |
1521 | nodemask_t *new; | |
1522 | NODEMASK_SCRATCH(scratch); | |
1523 | ||
1524 | if (!scratch) | |
1525 | return -ENOMEM; | |
39743889 | 1526 | |
596d7cfa KM |
1527 | old = &scratch->mask1; |
1528 | new = &scratch->mask2; | |
1529 | ||
1530 | err = get_nodes(old, old_nodes, maxnode); | |
39743889 | 1531 | if (err) |
596d7cfa | 1532 | goto out; |
39743889 | 1533 | |
596d7cfa | 1534 | err = get_nodes(new, new_nodes, maxnode); |
39743889 | 1535 | if (err) |
596d7cfa | 1536 | goto out; |
39743889 CL |
1537 | |
1538 | /* Find the mm_struct */ | |
55cfaa3c | 1539 | rcu_read_lock(); |
228ebcbe | 1540 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 | 1541 | if (!task) { |
55cfaa3c | 1542 | rcu_read_unlock(); |
596d7cfa KM |
1543 | err = -ESRCH; |
1544 | goto out; | |
39743889 | 1545 | } |
3268c63e | 1546 | get_task_struct(task); |
39743889 | 1547 | |
596d7cfa | 1548 | err = -EINVAL; |
39743889 CL |
1549 | |
1550 | /* | |
31367466 OE |
1551 | * Check if this process has the right to modify the specified process. |
1552 | * Use the regular "ptrace_may_access()" checks. | |
39743889 | 1553 | */ |
31367466 | 1554 | if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) { |
c69e8d9c | 1555 | rcu_read_unlock(); |
39743889 | 1556 | err = -EPERM; |
3268c63e | 1557 | goto out_put; |
39743889 | 1558 | } |
c69e8d9c | 1559 | rcu_read_unlock(); |
39743889 CL |
1560 | |
1561 | task_nodes = cpuset_mems_allowed(task); | |
1562 | /* Is the user allowed to access the target nodes? */ | |
596d7cfa | 1563 | if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 | 1564 | err = -EPERM; |
3268c63e | 1565 | goto out_put; |
39743889 CL |
1566 | } |
1567 | ||
0486a38b YX |
1568 | task_nodes = cpuset_mems_allowed(current); |
1569 | nodes_and(*new, *new, task_nodes); | |
1570 | if (nodes_empty(*new)) | |
1571 | goto out_put; | |
1572 | ||
86c3a764 DQ |
1573 | err = security_task_movememory(task); |
1574 | if (err) | |
3268c63e | 1575 | goto out_put; |
86c3a764 | 1576 | |
3268c63e CL |
1577 | mm = get_task_mm(task); |
1578 | put_task_struct(task); | |
f2a9ef88 SL |
1579 | |
1580 | if (!mm) { | |
3268c63e | 1581 | err = -EINVAL; |
f2a9ef88 SL |
1582 | goto out; |
1583 | } | |
1584 | ||
1585 | err = do_migrate_pages(mm, old, new, | |
1586 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); | |
3268c63e CL |
1587 | |
1588 | mmput(mm); | |
1589 | out: | |
596d7cfa KM |
1590 | NODEMASK_SCRATCH_FREE(scratch); |
1591 | ||
39743889 | 1592 | return err; |
3268c63e CL |
1593 | |
1594 | out_put: | |
1595 | put_task_struct(task); | |
1596 | goto out; | |
1597 | ||
39743889 CL |
1598 | } |
1599 | ||
b6e9b0ba DB |
1600 | SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode, |
1601 | const unsigned long __user *, old_nodes, | |
1602 | const unsigned long __user *, new_nodes) | |
1603 | { | |
1604 | return kernel_migrate_pages(pid, maxnode, old_nodes, new_nodes); | |
1605 | } | |
1606 | ||
39743889 | 1607 | |
8bccd85f | 1608 | /* Retrieve NUMA policy */ |
af03c4ac DB |
1609 | static int kernel_get_mempolicy(int __user *policy, |
1610 | unsigned long __user *nmask, | |
1611 | unsigned long maxnode, | |
1612 | unsigned long addr, | |
1613 | unsigned long flags) | |
8bccd85f | 1614 | { |
dbcb0f19 | 1615 | int err; |
3f649ab7 | 1616 | int pval; |
8bccd85f CL |
1617 | nodemask_t nodes; |
1618 | ||
050c17f2 | 1619 | if (nmask != NULL && maxnode < nr_node_ids) |
8bccd85f CL |
1620 | return -EINVAL; |
1621 | ||
4605f057 WH |
1622 | addr = untagged_addr(addr); |
1623 | ||
8bccd85f CL |
1624 | err = do_get_mempolicy(&pval, &nodes, addr, flags); |
1625 | ||
1626 | if (err) | |
1627 | return err; | |
1628 | ||
1629 | if (policy && put_user(pval, policy)) | |
1630 | return -EFAULT; | |
1631 | ||
1632 | if (nmask) | |
1633 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1634 | ||
1635 | return err; | |
1636 | } | |
1637 | ||
af03c4ac DB |
1638 | SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1639 | unsigned long __user *, nmask, unsigned long, maxnode, | |
1640 | unsigned long, addr, unsigned long, flags) | |
1641 | { | |
1642 | return kernel_get_mempolicy(policy, nmask, maxnode, addr, flags); | |
1643 | } | |
1644 | ||
1da177e4 LT |
1645 | #ifdef CONFIG_COMPAT |
1646 | ||
c93e0f6c HC |
1647 | COMPAT_SYSCALL_DEFINE5(get_mempolicy, int __user *, policy, |
1648 | compat_ulong_t __user *, nmask, | |
1649 | compat_ulong_t, maxnode, | |
1650 | compat_ulong_t, addr, compat_ulong_t, flags) | |
1da177e4 LT |
1651 | { |
1652 | long err; | |
1653 | unsigned long __user *nm = NULL; | |
1654 | unsigned long nr_bits, alloc_size; | |
1655 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1656 | ||
050c17f2 | 1657 | nr_bits = min_t(unsigned long, maxnode-1, nr_node_ids); |
1da177e4 LT |
1658 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; |
1659 | ||
1660 | if (nmask) | |
1661 | nm = compat_alloc_user_space(alloc_size); | |
1662 | ||
af03c4ac | 1663 | err = kernel_get_mempolicy(policy, nm, nr_bits+1, addr, flags); |
1da177e4 LT |
1664 | |
1665 | if (!err && nmask) { | |
2bbff6c7 KH |
1666 | unsigned long copy_size; |
1667 | copy_size = min_t(unsigned long, sizeof(bm), alloc_size); | |
1668 | err = copy_from_user(bm, nm, copy_size); | |
1da177e4 LT |
1669 | /* ensure entire bitmap is zeroed */ |
1670 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1671 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1672 | } | |
1673 | ||
1674 | return err; | |
1675 | } | |
1676 | ||
c93e0f6c HC |
1677 | COMPAT_SYSCALL_DEFINE3(set_mempolicy, int, mode, compat_ulong_t __user *, nmask, |
1678 | compat_ulong_t, maxnode) | |
1da177e4 | 1679 | { |
1da177e4 LT |
1680 | unsigned long __user *nm = NULL; |
1681 | unsigned long nr_bits, alloc_size; | |
1682 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1683 | ||
1684 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1685 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1686 | ||
1687 | if (nmask) { | |
cf01fb99 CS |
1688 | if (compat_get_bitmap(bm, nmask, nr_bits)) |
1689 | return -EFAULT; | |
1da177e4 | 1690 | nm = compat_alloc_user_space(alloc_size); |
cf01fb99 CS |
1691 | if (copy_to_user(nm, bm, alloc_size)) |
1692 | return -EFAULT; | |
1da177e4 LT |
1693 | } |
1694 | ||
af03c4ac | 1695 | return kernel_set_mempolicy(mode, nm, nr_bits+1); |
1da177e4 LT |
1696 | } |
1697 | ||
c93e0f6c HC |
1698 | COMPAT_SYSCALL_DEFINE6(mbind, compat_ulong_t, start, compat_ulong_t, len, |
1699 | compat_ulong_t, mode, compat_ulong_t __user *, nmask, | |
1700 | compat_ulong_t, maxnode, compat_ulong_t, flags) | |
1da177e4 | 1701 | { |
1da177e4 LT |
1702 | unsigned long __user *nm = NULL; |
1703 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1704 | nodemask_t bm; |
1da177e4 LT |
1705 | |
1706 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1707 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1708 | ||
1709 | if (nmask) { | |
cf01fb99 CS |
1710 | if (compat_get_bitmap(nodes_addr(bm), nmask, nr_bits)) |
1711 | return -EFAULT; | |
1da177e4 | 1712 | nm = compat_alloc_user_space(alloc_size); |
cf01fb99 CS |
1713 | if (copy_to_user(nm, nodes_addr(bm), alloc_size)) |
1714 | return -EFAULT; | |
1da177e4 LT |
1715 | } |
1716 | ||
e7dc9ad6 | 1717 | return kernel_mbind(start, len, mode, nm, nr_bits+1, flags); |
1da177e4 LT |
1718 | } |
1719 | ||
b6e9b0ba DB |
1720 | COMPAT_SYSCALL_DEFINE4(migrate_pages, compat_pid_t, pid, |
1721 | compat_ulong_t, maxnode, | |
1722 | const compat_ulong_t __user *, old_nodes, | |
1723 | const compat_ulong_t __user *, new_nodes) | |
1724 | { | |
1725 | unsigned long __user *old = NULL; | |
1726 | unsigned long __user *new = NULL; | |
1727 | nodemask_t tmp_mask; | |
1728 | unsigned long nr_bits; | |
1729 | unsigned long size; | |
1730 | ||
1731 | nr_bits = min_t(unsigned long, maxnode - 1, MAX_NUMNODES); | |
1732 | size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1733 | if (old_nodes) { | |
1734 | if (compat_get_bitmap(nodes_addr(tmp_mask), old_nodes, nr_bits)) | |
1735 | return -EFAULT; | |
1736 | old = compat_alloc_user_space(new_nodes ? size * 2 : size); | |
1737 | if (new_nodes) | |
1738 | new = old + size / sizeof(unsigned long); | |
1739 | if (copy_to_user(old, nodes_addr(tmp_mask), size)) | |
1740 | return -EFAULT; | |
1741 | } | |
1742 | if (new_nodes) { | |
1743 | if (compat_get_bitmap(nodes_addr(tmp_mask), new_nodes, nr_bits)) | |
1744 | return -EFAULT; | |
1745 | if (new == NULL) | |
1746 | new = compat_alloc_user_space(size); | |
1747 | if (copy_to_user(new, nodes_addr(tmp_mask), size)) | |
1748 | return -EFAULT; | |
1749 | } | |
1750 | return kernel_migrate_pages(pid, nr_bits + 1, old, new); | |
1751 | } | |
1752 | ||
1753 | #endif /* CONFIG_COMPAT */ | |
1da177e4 | 1754 | |
20ca87f2 LX |
1755 | bool vma_migratable(struct vm_area_struct *vma) |
1756 | { | |
1757 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1758 | return false; | |
1759 | ||
1760 | /* | |
1761 | * DAX device mappings require predictable access latency, so avoid | |
1762 | * incurring periodic faults. | |
1763 | */ | |
1764 | if (vma_is_dax(vma)) | |
1765 | return false; | |
1766 | ||
1767 | if (is_vm_hugetlb_page(vma) && | |
1768 | !hugepage_migration_supported(hstate_vma(vma))) | |
1769 | return false; | |
1770 | ||
1771 | /* | |
1772 | * Migration allocates pages in the highest zone. If we cannot | |
1773 | * do so then migration (at least from node to node) is not | |
1774 | * possible. | |
1775 | */ | |
1776 | if (vma->vm_file && | |
1777 | gfp_zone(mapping_gfp_mask(vma->vm_file->f_mapping)) | |
1778 | < policy_zone) | |
1779 | return false; | |
1780 | return true; | |
1781 | } | |
1782 | ||
74d2c3a0 ON |
1783 | struct mempolicy *__get_vma_policy(struct vm_area_struct *vma, |
1784 | unsigned long addr) | |
1da177e4 | 1785 | { |
8d90274b | 1786 | struct mempolicy *pol = NULL; |
1da177e4 LT |
1787 | |
1788 | if (vma) { | |
480eccf9 | 1789 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
8d90274b | 1790 | pol = vma->vm_ops->get_policy(vma, addr); |
00442ad0 | 1791 | } else if (vma->vm_policy) { |
1da177e4 | 1792 | pol = vma->vm_policy; |
00442ad0 MG |
1793 | |
1794 | /* | |
1795 | * shmem_alloc_page() passes MPOL_F_SHARED policy with | |
1796 | * a pseudo vma whose vma->vm_ops=NULL. Take a reference | |
1797 | * count on these policies which will be dropped by | |
1798 | * mpol_cond_put() later | |
1799 | */ | |
1800 | if (mpol_needs_cond_ref(pol)) | |
1801 | mpol_get(pol); | |
1802 | } | |
1da177e4 | 1803 | } |
f15ca78e | 1804 | |
74d2c3a0 ON |
1805 | return pol; |
1806 | } | |
1807 | ||
1808 | /* | |
dd6eecb9 | 1809 | * get_vma_policy(@vma, @addr) |
74d2c3a0 ON |
1810 | * @vma: virtual memory area whose policy is sought |
1811 | * @addr: address in @vma for shared policy lookup | |
1812 | * | |
1813 | * Returns effective policy for a VMA at specified address. | |
dd6eecb9 | 1814 | * Falls back to current->mempolicy or system default policy, as necessary. |
74d2c3a0 ON |
1815 | * Shared policies [those marked as MPOL_F_SHARED] require an extra reference |
1816 | * count--added by the get_policy() vm_op, as appropriate--to protect against | |
1817 | * freeing by another task. It is the caller's responsibility to free the | |
1818 | * extra reference for shared policies. | |
1819 | */ | |
ac79f78d | 1820 | static struct mempolicy *get_vma_policy(struct vm_area_struct *vma, |
dd6eecb9 | 1821 | unsigned long addr) |
74d2c3a0 ON |
1822 | { |
1823 | struct mempolicy *pol = __get_vma_policy(vma, addr); | |
1824 | ||
8d90274b | 1825 | if (!pol) |
dd6eecb9 | 1826 | pol = get_task_policy(current); |
8d90274b | 1827 | |
1da177e4 LT |
1828 | return pol; |
1829 | } | |
1830 | ||
6b6482bb | 1831 | bool vma_policy_mof(struct vm_area_struct *vma) |
fc314724 | 1832 | { |
6b6482bb | 1833 | struct mempolicy *pol; |
fc314724 | 1834 | |
6b6482bb ON |
1835 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
1836 | bool ret = false; | |
fc314724 | 1837 | |
6b6482bb ON |
1838 | pol = vma->vm_ops->get_policy(vma, vma->vm_start); |
1839 | if (pol && (pol->flags & MPOL_F_MOF)) | |
1840 | ret = true; | |
1841 | mpol_cond_put(pol); | |
8d90274b | 1842 | |
6b6482bb | 1843 | return ret; |
fc314724 MG |
1844 | } |
1845 | ||
6b6482bb | 1846 | pol = vma->vm_policy; |
8d90274b | 1847 | if (!pol) |
6b6482bb | 1848 | pol = get_task_policy(current); |
8d90274b | 1849 | |
fc314724 MG |
1850 | return pol->flags & MPOL_F_MOF; |
1851 | } | |
1852 | ||
d3eb1570 LJ |
1853 | static int apply_policy_zone(struct mempolicy *policy, enum zone_type zone) |
1854 | { | |
1855 | enum zone_type dynamic_policy_zone = policy_zone; | |
1856 | ||
1857 | BUG_ON(dynamic_policy_zone == ZONE_MOVABLE); | |
1858 | ||
1859 | /* | |
1860 | * if policy->v.nodes has movable memory only, | |
1861 | * we apply policy when gfp_zone(gfp) = ZONE_MOVABLE only. | |
1862 | * | |
1863 | * policy->v.nodes is intersect with node_states[N_MEMORY]. | |
f0953a1b | 1864 | * so if the following test fails, it implies |
d3eb1570 LJ |
1865 | * policy->v.nodes has movable memory only. |
1866 | */ | |
1867 | if (!nodes_intersects(policy->v.nodes, node_states[N_HIGH_MEMORY])) | |
1868 | dynamic_policy_zone = ZONE_MOVABLE; | |
1869 | ||
1870 | return zone >= dynamic_policy_zone; | |
1871 | } | |
1872 | ||
52cd3b07 LS |
1873 | /* |
1874 | * Return a nodemask representing a mempolicy for filtering nodes for | |
1875 | * page allocation | |
1876 | */ | |
8ca39e68 | 1877 | nodemask_t *policy_nodemask(gfp_t gfp, struct mempolicy *policy) |
19770b32 MG |
1878 | { |
1879 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
45c4745a | 1880 | if (unlikely(policy->mode == MPOL_BIND) && |
d3eb1570 | 1881 | apply_policy_zone(policy, gfp_zone(gfp)) && |
19770b32 MG |
1882 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) |
1883 | return &policy->v.nodes; | |
1884 | ||
1885 | return NULL; | |
1886 | } | |
1887 | ||
04ec6264 | 1888 | /* Return the node id preferred by the given mempolicy, or the given id */ |
f8fd5253 | 1889 | static int policy_node(gfp_t gfp, struct mempolicy *policy, int nd) |
1da177e4 | 1890 | { |
7858d7bc | 1891 | if (policy->mode == MPOL_PREFERRED) { |
6d840958 | 1892 | nd = policy->v.preferred_node; |
7858d7bc | 1893 | } else { |
19770b32 | 1894 | /* |
6d840958 MH |
1895 | * __GFP_THISNODE shouldn't even be used with the bind policy |
1896 | * because we might easily break the expectation to stay on the | |
1897 | * requested node and not break the policy. | |
19770b32 | 1898 | */ |
6d840958 | 1899 | WARN_ON_ONCE(policy->mode == MPOL_BIND && (gfp & __GFP_THISNODE)); |
1da177e4 | 1900 | } |
6d840958 | 1901 | |
04ec6264 | 1902 | return nd; |
1da177e4 LT |
1903 | } |
1904 | ||
1905 | /* Do dynamic interleaving for a process */ | |
1906 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1907 | { | |
45816682 | 1908 | unsigned next; |
1da177e4 LT |
1909 | struct task_struct *me = current; |
1910 | ||
45816682 | 1911 | next = next_node_in(me->il_prev, policy->v.nodes); |
f5b087b5 | 1912 | if (next < MAX_NUMNODES) |
45816682 VB |
1913 | me->il_prev = next; |
1914 | return next; | |
1da177e4 LT |
1915 | } |
1916 | ||
dc85da15 CL |
1917 | /* |
1918 | * Depending on the memory policy provide a node from which to allocate the | |
1919 | * next slab entry. | |
1920 | */ | |
2a389610 | 1921 | unsigned int mempolicy_slab_node(void) |
dc85da15 | 1922 | { |
e7b691b0 | 1923 | struct mempolicy *policy; |
2a389610 | 1924 | int node = numa_mem_id(); |
e7b691b0 AK |
1925 | |
1926 | if (in_interrupt()) | |
2a389610 | 1927 | return node; |
e7b691b0 AK |
1928 | |
1929 | policy = current->mempolicy; | |
7858d7bc | 1930 | if (!policy) |
2a389610 | 1931 | return node; |
bea904d5 LS |
1932 | |
1933 | switch (policy->mode) { | |
1934 | case MPOL_PREFERRED: | |
fc36b8d3 | 1935 | return policy->v.preferred_node; |
765c4507 | 1936 | |
dc85da15 CL |
1937 | case MPOL_INTERLEAVE: |
1938 | return interleave_nodes(policy); | |
1939 | ||
dd1a239f | 1940 | case MPOL_BIND: { |
c33d6c06 MG |
1941 | struct zoneref *z; |
1942 | ||
dc85da15 CL |
1943 | /* |
1944 | * Follow bind policy behavior and start allocation at the | |
1945 | * first node. | |
1946 | */ | |
19770b32 | 1947 | struct zonelist *zonelist; |
19770b32 | 1948 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); |
c9634cf0 | 1949 | zonelist = &NODE_DATA(node)->node_zonelists[ZONELIST_FALLBACK]; |
c33d6c06 MG |
1950 | z = first_zones_zonelist(zonelist, highest_zoneidx, |
1951 | &policy->v.nodes); | |
c1093b74 | 1952 | return z->zone ? zone_to_nid(z->zone) : node; |
dd1a239f | 1953 | } |
7858d7bc FT |
1954 | case MPOL_LOCAL: |
1955 | return node; | |
dc85da15 | 1956 | |
dc85da15 | 1957 | default: |
bea904d5 | 1958 | BUG(); |
dc85da15 CL |
1959 | } |
1960 | } | |
1961 | ||
fee83b3a AM |
1962 | /* |
1963 | * Do static interleaving for a VMA with known offset @n. Returns the n'th | |
1964 | * node in pol->v.nodes (starting from n=0), wrapping around if n exceeds the | |
1965 | * number of present nodes. | |
1966 | */ | |
98c70baa | 1967 | static unsigned offset_il_node(struct mempolicy *pol, unsigned long n) |
1da177e4 | 1968 | { |
dfcd3c0d | 1969 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1970 | unsigned target; |
fee83b3a AM |
1971 | int i; |
1972 | int nid; | |
1da177e4 | 1973 | |
f5b087b5 DR |
1974 | if (!nnodes) |
1975 | return numa_node_id(); | |
fee83b3a AM |
1976 | target = (unsigned int)n % nnodes; |
1977 | nid = first_node(pol->v.nodes); | |
1978 | for (i = 0; i < target; i++) | |
dfcd3c0d | 1979 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1980 | return nid; |
1981 | } | |
1982 | ||
5da7ca86 CL |
1983 | /* Determine a node number for interleave */ |
1984 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1985 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1986 | { | |
1987 | if (vma) { | |
1988 | unsigned long off; | |
1989 | ||
3b98b087 NA |
1990 | /* |
1991 | * for small pages, there is no difference between | |
1992 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1993 | * for huge pages, since vm_pgoff is in units of small | |
1994 | * pages, we need to shift off the always 0 bits to get | |
1995 | * a useful offset. | |
1996 | */ | |
1997 | BUG_ON(shift < PAGE_SHIFT); | |
1998 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 | 1999 | off += (addr - vma->vm_start) >> shift; |
98c70baa | 2000 | return offset_il_node(pol, off); |
5da7ca86 CL |
2001 | } else |
2002 | return interleave_nodes(pol); | |
2003 | } | |
2004 | ||
00ac59ad | 2005 | #ifdef CONFIG_HUGETLBFS |
480eccf9 | 2006 | /* |
04ec6264 | 2007 | * huge_node(@vma, @addr, @gfp_flags, @mpol) |
b46e14ac FF |
2008 | * @vma: virtual memory area whose policy is sought |
2009 | * @addr: address in @vma for shared policy lookup and interleave policy | |
2010 | * @gfp_flags: for requested zone | |
2011 | * @mpol: pointer to mempolicy pointer for reference counted mempolicy | |
2012 | * @nodemask: pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 | 2013 | * |
04ec6264 | 2014 | * Returns a nid suitable for a huge page allocation and a pointer |
52cd3b07 LS |
2015 | * to the struct mempolicy for conditional unref after allocation. |
2016 | * If the effective policy is 'BIND, returns a pointer to the mempolicy's | |
2017 | * @nodemask for filtering the zonelist. | |
c0ff7453 | 2018 | * |
d26914d1 | 2019 | * Must be protected by read_mems_allowed_begin() |
480eccf9 | 2020 | */ |
04ec6264 VB |
2021 | int huge_node(struct vm_area_struct *vma, unsigned long addr, gfp_t gfp_flags, |
2022 | struct mempolicy **mpol, nodemask_t **nodemask) | |
5da7ca86 | 2023 | { |
04ec6264 | 2024 | int nid; |
5da7ca86 | 2025 | |
dd6eecb9 | 2026 | *mpol = get_vma_policy(vma, addr); |
19770b32 | 2027 | *nodemask = NULL; /* assume !MPOL_BIND */ |
5da7ca86 | 2028 | |
52cd3b07 | 2029 | if (unlikely((*mpol)->mode == MPOL_INTERLEAVE)) { |
04ec6264 VB |
2030 | nid = interleave_nid(*mpol, vma, addr, |
2031 | huge_page_shift(hstate_vma(vma))); | |
52cd3b07 | 2032 | } else { |
04ec6264 | 2033 | nid = policy_node(gfp_flags, *mpol, numa_node_id()); |
52cd3b07 LS |
2034 | if ((*mpol)->mode == MPOL_BIND) |
2035 | *nodemask = &(*mpol)->v.nodes; | |
480eccf9 | 2036 | } |
04ec6264 | 2037 | return nid; |
5da7ca86 | 2038 | } |
06808b08 LS |
2039 | |
2040 | /* | |
2041 | * init_nodemask_of_mempolicy | |
2042 | * | |
2043 | * If the current task's mempolicy is "default" [NULL], return 'false' | |
2044 | * to indicate default policy. Otherwise, extract the policy nodemask | |
2045 | * for 'bind' or 'interleave' policy into the argument nodemask, or | |
2046 | * initialize the argument nodemask to contain the single node for | |
2047 | * 'preferred' or 'local' policy and return 'true' to indicate presence | |
2048 | * of non-default mempolicy. | |
2049 | * | |
2050 | * We don't bother with reference counting the mempolicy [mpol_get/put] | |
2051 | * because the current task is examining it's own mempolicy and a task's | |
2052 | * mempolicy is only ever changed by the task itself. | |
2053 | * | |
2054 | * N.B., it is the caller's responsibility to free a returned nodemask. | |
2055 | */ | |
2056 | bool init_nodemask_of_mempolicy(nodemask_t *mask) | |
2057 | { | |
2058 | struct mempolicy *mempolicy; | |
2059 | int nid; | |
2060 | ||
2061 | if (!(mask && current->mempolicy)) | |
2062 | return false; | |
2063 | ||
c0ff7453 | 2064 | task_lock(current); |
06808b08 LS |
2065 | mempolicy = current->mempolicy; |
2066 | switch (mempolicy->mode) { | |
2067 | case MPOL_PREFERRED: | |
7858d7bc | 2068 | nid = mempolicy->v.preferred_node; |
06808b08 LS |
2069 | init_nodemask_of_node(mask, nid); |
2070 | break; | |
2071 | ||
2072 | case MPOL_BIND: | |
06808b08 | 2073 | case MPOL_INTERLEAVE: |
7858d7bc FT |
2074 | *mask = mempolicy->v.nodes; |
2075 | break; | |
2076 | ||
2077 | case MPOL_LOCAL: | |
2078 | nid = numa_node_id(); | |
2079 | init_nodemask_of_node(mask, nid); | |
06808b08 LS |
2080 | break; |
2081 | ||
2082 | default: | |
2083 | BUG(); | |
2084 | } | |
c0ff7453 | 2085 | task_unlock(current); |
06808b08 LS |
2086 | |
2087 | return true; | |
2088 | } | |
00ac59ad | 2089 | #endif |
5da7ca86 | 2090 | |
6f48d0eb | 2091 | /* |
b26e517a | 2092 | * mempolicy_in_oom_domain |
6f48d0eb | 2093 | * |
b26e517a FT |
2094 | * If tsk's mempolicy is "bind", check for intersection between mask and |
2095 | * the policy nodemask. Otherwise, return true for all other policies | |
2096 | * including "interleave", as a tsk with "interleave" policy may have | |
2097 | * memory allocated from all nodes in system. | |
6f48d0eb DR |
2098 | * |
2099 | * Takes task_lock(tsk) to prevent freeing of its mempolicy. | |
2100 | */ | |
b26e517a | 2101 | bool mempolicy_in_oom_domain(struct task_struct *tsk, |
6f48d0eb DR |
2102 | const nodemask_t *mask) |
2103 | { | |
2104 | struct mempolicy *mempolicy; | |
2105 | bool ret = true; | |
2106 | ||
2107 | if (!mask) | |
2108 | return ret; | |
b26e517a | 2109 | |
6f48d0eb DR |
2110 | task_lock(tsk); |
2111 | mempolicy = tsk->mempolicy; | |
b26e517a | 2112 | if (mempolicy && mempolicy->mode == MPOL_BIND) |
6f48d0eb | 2113 | ret = nodes_intersects(mempolicy->v.nodes, *mask); |
6f48d0eb | 2114 | task_unlock(tsk); |
b26e517a | 2115 | |
6f48d0eb DR |
2116 | return ret; |
2117 | } | |
2118 | ||
1da177e4 LT |
2119 | /* Allocate a page in interleaved policy. |
2120 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
2121 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
2122 | unsigned nid) | |
1da177e4 | 2123 | { |
1da177e4 LT |
2124 | struct page *page; |
2125 | ||
84172f4b | 2126 | page = __alloc_pages(gfp, order, nid, NULL); |
4518085e KW |
2127 | /* skip NUMA_INTERLEAVE_HIT counter update if numa stats is disabled */ |
2128 | if (!static_branch_likely(&vm_numa_stat_key)) | |
2129 | return page; | |
de55c8b2 AR |
2130 | if (page && page_to_nid(page) == nid) { |
2131 | preempt_disable(); | |
f19298b9 | 2132 | __count_numa_event(page_zone(page), NUMA_INTERLEAVE_HIT); |
de55c8b2 AR |
2133 | preempt_enable(); |
2134 | } | |
1da177e4 LT |
2135 | return page; |
2136 | } | |
2137 | ||
2138 | /** | |
eb350739 MWO |
2139 | * alloc_pages_vma - Allocate a page for a VMA. |
2140 | * @gfp: GFP flags. | |
2141 | * @order: Order of the GFP allocation. | |
2142 | * @vma: Pointer to VMA or NULL if not available. | |
2143 | * @addr: Virtual address of the allocation. Must be inside @vma. | |
2144 | * @node: Which node to prefer for allocation (modulo policy). | |
2145 | * @hugepage: For hugepages try only the preferred node if possible. | |
1da177e4 | 2146 | * |
eb350739 MWO |
2147 | * Allocate a page for a specific address in @vma, using the appropriate |
2148 | * NUMA policy. When @vma is not NULL the caller must hold the mmap_lock | |
2149 | * of the mm_struct of the VMA to prevent it from going away. Should be | |
2150 | * used for all allocations for pages that will be mapped into user space. | |
1da177e4 | 2151 | * |
eb350739 | 2152 | * Return: The page on success or NULL if allocation fails. |
1da177e4 | 2153 | */ |
eb350739 | 2154 | struct page *alloc_pages_vma(gfp_t gfp, int order, struct vm_area_struct *vma, |
19deb769 | 2155 | unsigned long addr, int node, bool hugepage) |
1da177e4 | 2156 | { |
cc9a6c87 | 2157 | struct mempolicy *pol; |
c0ff7453 | 2158 | struct page *page; |
04ec6264 | 2159 | int preferred_nid; |
be97a41b | 2160 | nodemask_t *nmask; |
cc9a6c87 | 2161 | |
dd6eecb9 | 2162 | pol = get_vma_policy(vma, addr); |
1da177e4 | 2163 | |
0867a57c VB |
2164 | if (pol->mode == MPOL_INTERLEAVE) { |
2165 | unsigned nid; | |
2166 | ||
2167 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT + order); | |
2168 | mpol_cond_put(pol); | |
2169 | page = alloc_page_interleave(gfp, order, nid); | |
2170 | goto out; | |
19deb769 DR |
2171 | } |
2172 | ||
2173 | if (unlikely(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hugepage)) { | |
2174 | int hpage_node = node; | |
2175 | ||
2176 | /* | |
2177 | * For hugepage allocation and non-interleave policy which | |
2178 | * allows the current node (or other explicitly preferred | |
2179 | * node) we only try to allocate from the current/preferred | |
2180 | * node and don't fall back to other nodes, as the cost of | |
2181 | * remote accesses would likely offset THP benefits. | |
2182 | * | |
2183 | * If the policy is interleave, or does not allow the current | |
2184 | * node in its nodemask, we allocate the standard way. | |
2185 | */ | |
7858d7bc | 2186 | if (pol->mode == MPOL_PREFERRED) |
19deb769 DR |
2187 | hpage_node = pol->v.preferred_node; |
2188 | ||
2189 | nmask = policy_nodemask(gfp, pol); | |
2190 | if (!nmask || node_isset(hpage_node, *nmask)) { | |
2191 | mpol_cond_put(pol); | |
cc638f32 VB |
2192 | /* |
2193 | * First, try to allocate THP only on local node, but | |
2194 | * don't reclaim unnecessarily, just compact. | |
2195 | */ | |
19deb769 | 2196 | page = __alloc_pages_node(hpage_node, |
cc638f32 | 2197 | gfp | __GFP_THISNODE | __GFP_NORETRY, order); |
76e654cc DR |
2198 | |
2199 | /* | |
2200 | * If hugepage allocations are configured to always | |
2201 | * synchronous compact or the vma has been madvised | |
2202 | * to prefer hugepage backing, retry allowing remote | |
cc638f32 | 2203 | * memory with both reclaim and compact as well. |
76e654cc DR |
2204 | */ |
2205 | if (!page && (gfp & __GFP_DIRECT_RECLAIM)) | |
2206 | page = __alloc_pages_node(hpage_node, | |
cc638f32 | 2207 | gfp, order); |
76e654cc | 2208 | |
19deb769 DR |
2209 | goto out; |
2210 | } | |
356ff8a9 DR |
2211 | } |
2212 | ||
be97a41b | 2213 | nmask = policy_nodemask(gfp, pol); |
04ec6264 | 2214 | preferred_nid = policy_node(gfp, pol, node); |
84172f4b | 2215 | page = __alloc_pages(gfp, order, preferred_nid, nmask); |
d51e9894 | 2216 | mpol_cond_put(pol); |
be97a41b | 2217 | out: |
c0ff7453 | 2218 | return page; |
1da177e4 | 2219 | } |
69262215 | 2220 | EXPORT_SYMBOL(alloc_pages_vma); |
1da177e4 LT |
2221 | |
2222 | /** | |
6421ec76 MWO |
2223 | * alloc_pages - Allocate pages. |
2224 | * @gfp: GFP flags. | |
2225 | * @order: Power of two of number of pages to allocate. | |
1da177e4 | 2226 | * |
6421ec76 MWO |
2227 | * Allocate 1 << @order contiguous pages. The physical address of the |
2228 | * first page is naturally aligned (eg an order-3 allocation will be aligned | |
2229 | * to a multiple of 8 * PAGE_SIZE bytes). The NUMA policy of the current | |
2230 | * process is honoured when in process context. | |
1da177e4 | 2231 | * |
6421ec76 MWO |
2232 | * Context: Can be called from any context, providing the appropriate GFP |
2233 | * flags are used. | |
2234 | * Return: The page on success or NULL if allocation fails. | |
1da177e4 | 2235 | */ |
d7f946d0 | 2236 | struct page *alloc_pages(gfp_t gfp, unsigned order) |
1da177e4 | 2237 | { |
8d90274b | 2238 | struct mempolicy *pol = &default_policy; |
c0ff7453 | 2239 | struct page *page; |
1da177e4 | 2240 | |
8d90274b ON |
2241 | if (!in_interrupt() && !(gfp & __GFP_THISNODE)) |
2242 | pol = get_task_policy(current); | |
52cd3b07 LS |
2243 | |
2244 | /* | |
2245 | * No reference counting needed for current->mempolicy | |
2246 | * nor system default_policy | |
2247 | */ | |
45c4745a | 2248 | if (pol->mode == MPOL_INTERLEAVE) |
c0ff7453 MX |
2249 | page = alloc_page_interleave(gfp, order, interleave_nodes(pol)); |
2250 | else | |
84172f4b | 2251 | page = __alloc_pages(gfp, order, |
04ec6264 | 2252 | policy_node(gfp, pol, numa_node_id()), |
5c4b4be3 | 2253 | policy_nodemask(gfp, pol)); |
cc9a6c87 | 2254 | |
c0ff7453 | 2255 | return page; |
1da177e4 | 2256 | } |
d7f946d0 | 2257 | EXPORT_SYMBOL(alloc_pages); |
1da177e4 | 2258 | |
ef0855d3 ON |
2259 | int vma_dup_policy(struct vm_area_struct *src, struct vm_area_struct *dst) |
2260 | { | |
2261 | struct mempolicy *pol = mpol_dup(vma_policy(src)); | |
2262 | ||
2263 | if (IS_ERR(pol)) | |
2264 | return PTR_ERR(pol); | |
2265 | dst->vm_policy = pol; | |
2266 | return 0; | |
2267 | } | |
2268 | ||
4225399a | 2269 | /* |
846a16bf | 2270 | * If mpol_dup() sees current->cpuset == cpuset_being_rebound, then it |
4225399a PJ |
2271 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() |
2272 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
2273 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
2274 | * further kernel/cpuset.c update_nodemask(). | |
708c1bbc MX |
2275 | * |
2276 | * current's mempolicy may be rebinded by the other task(the task that changes | |
2277 | * cpuset's mems), so we needn't do rebind work for current task. | |
4225399a | 2278 | */ |
4225399a | 2279 | |
846a16bf LS |
2280 | /* Slow path of a mempolicy duplicate */ |
2281 | struct mempolicy *__mpol_dup(struct mempolicy *old) | |
1da177e4 LT |
2282 | { |
2283 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2284 | ||
2285 | if (!new) | |
2286 | return ERR_PTR(-ENOMEM); | |
708c1bbc MX |
2287 | |
2288 | /* task's mempolicy is protected by alloc_lock */ | |
2289 | if (old == current->mempolicy) { | |
2290 | task_lock(current); | |
2291 | *new = *old; | |
2292 | task_unlock(current); | |
2293 | } else | |
2294 | *new = *old; | |
2295 | ||
4225399a PJ |
2296 | if (current_cpuset_is_being_rebound()) { |
2297 | nodemask_t mems = cpuset_mems_allowed(current); | |
213980c0 | 2298 | mpol_rebind_policy(new, &mems); |
4225399a | 2299 | } |
1da177e4 | 2300 | atomic_set(&new->refcnt, 1); |
1da177e4 LT |
2301 | return new; |
2302 | } | |
2303 | ||
2304 | /* Slow path of a mempolicy comparison */ | |
fcfb4dcc | 2305 | bool __mpol_equal(struct mempolicy *a, struct mempolicy *b) |
1da177e4 LT |
2306 | { |
2307 | if (!a || !b) | |
fcfb4dcc | 2308 | return false; |
45c4745a | 2309 | if (a->mode != b->mode) |
fcfb4dcc | 2310 | return false; |
19800502 | 2311 | if (a->flags != b->flags) |
fcfb4dcc | 2312 | return false; |
19800502 BL |
2313 | if (mpol_store_user_nodemask(a)) |
2314 | if (!nodes_equal(a->w.user_nodemask, b->w.user_nodemask)) | |
fcfb4dcc | 2315 | return false; |
19800502 | 2316 | |
45c4745a | 2317 | switch (a->mode) { |
19770b32 | 2318 | case MPOL_BIND: |
1da177e4 | 2319 | case MPOL_INTERLEAVE: |
fcfb4dcc | 2320 | return !!nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 | 2321 | case MPOL_PREFERRED: |
75719661 | 2322 | return a->v.preferred_node == b->v.preferred_node; |
7858d7bc FT |
2323 | case MPOL_LOCAL: |
2324 | return true; | |
1da177e4 LT |
2325 | default: |
2326 | BUG(); | |
fcfb4dcc | 2327 | return false; |
1da177e4 LT |
2328 | } |
2329 | } | |
2330 | ||
1da177e4 LT |
2331 | /* |
2332 | * Shared memory backing store policy support. | |
2333 | * | |
2334 | * Remember policies even when nobody has shared memory mapped. | |
2335 | * The policies are kept in Red-Black tree linked from the inode. | |
4a8c7bb5 | 2336 | * They are protected by the sp->lock rwlock, which should be held |
1da177e4 LT |
2337 | * for any accesses to the tree. |
2338 | */ | |
2339 | ||
4a8c7bb5 NZ |
2340 | /* |
2341 | * lookup first element intersecting start-end. Caller holds sp->lock for | |
2342 | * reading or for writing | |
2343 | */ | |
1da177e4 LT |
2344 | static struct sp_node * |
2345 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
2346 | { | |
2347 | struct rb_node *n = sp->root.rb_node; | |
2348 | ||
2349 | while (n) { | |
2350 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
2351 | ||
2352 | if (start >= p->end) | |
2353 | n = n->rb_right; | |
2354 | else if (end <= p->start) | |
2355 | n = n->rb_left; | |
2356 | else | |
2357 | break; | |
2358 | } | |
2359 | if (!n) | |
2360 | return NULL; | |
2361 | for (;;) { | |
2362 | struct sp_node *w = NULL; | |
2363 | struct rb_node *prev = rb_prev(n); | |
2364 | if (!prev) | |
2365 | break; | |
2366 | w = rb_entry(prev, struct sp_node, nd); | |
2367 | if (w->end <= start) | |
2368 | break; | |
2369 | n = prev; | |
2370 | } | |
2371 | return rb_entry(n, struct sp_node, nd); | |
2372 | } | |
2373 | ||
4a8c7bb5 NZ |
2374 | /* |
2375 | * Insert a new shared policy into the list. Caller holds sp->lock for | |
2376 | * writing. | |
2377 | */ | |
1da177e4 LT |
2378 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) |
2379 | { | |
2380 | struct rb_node **p = &sp->root.rb_node; | |
2381 | struct rb_node *parent = NULL; | |
2382 | struct sp_node *nd; | |
2383 | ||
2384 | while (*p) { | |
2385 | parent = *p; | |
2386 | nd = rb_entry(parent, struct sp_node, nd); | |
2387 | if (new->start < nd->start) | |
2388 | p = &(*p)->rb_left; | |
2389 | else if (new->end > nd->end) | |
2390 | p = &(*p)->rb_right; | |
2391 | else | |
2392 | BUG(); | |
2393 | } | |
2394 | rb_link_node(&new->nd, parent, p); | |
2395 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 2396 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
45c4745a | 2397 | new->policy ? new->policy->mode : 0); |
1da177e4 LT |
2398 | } |
2399 | ||
2400 | /* Find shared policy intersecting idx */ | |
2401 | struct mempolicy * | |
2402 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
2403 | { | |
2404 | struct mempolicy *pol = NULL; | |
2405 | struct sp_node *sn; | |
2406 | ||
2407 | if (!sp->root.rb_node) | |
2408 | return NULL; | |
4a8c7bb5 | 2409 | read_lock(&sp->lock); |
1da177e4 LT |
2410 | sn = sp_lookup(sp, idx, idx+1); |
2411 | if (sn) { | |
2412 | mpol_get(sn->policy); | |
2413 | pol = sn->policy; | |
2414 | } | |
4a8c7bb5 | 2415 | read_unlock(&sp->lock); |
1da177e4 LT |
2416 | return pol; |
2417 | } | |
2418 | ||
63f74ca2 KM |
2419 | static void sp_free(struct sp_node *n) |
2420 | { | |
2421 | mpol_put(n->policy); | |
2422 | kmem_cache_free(sn_cache, n); | |
2423 | } | |
2424 | ||
771fb4d8 LS |
2425 | /** |
2426 | * mpol_misplaced - check whether current page node is valid in policy | |
2427 | * | |
b46e14ac FF |
2428 | * @page: page to be checked |
2429 | * @vma: vm area where page mapped | |
2430 | * @addr: virtual address where page mapped | |
771fb4d8 LS |
2431 | * |
2432 | * Lookup current policy node id for vma,addr and "compare to" page's | |
5f076944 | 2433 | * node id. Policy determination "mimics" alloc_page_vma(). |
771fb4d8 | 2434 | * Called from fault path where we know the vma and faulting address. |
5f076944 MWO |
2435 | * |
2436 | * Return: -1 if the page is in a node that is valid for this policy, or a | |
2437 | * suitable node ID to allocate a replacement page from. | |
771fb4d8 LS |
2438 | */ |
2439 | int mpol_misplaced(struct page *page, struct vm_area_struct *vma, unsigned long addr) | |
2440 | { | |
2441 | struct mempolicy *pol; | |
c33d6c06 | 2442 | struct zoneref *z; |
771fb4d8 LS |
2443 | int curnid = page_to_nid(page); |
2444 | unsigned long pgoff; | |
90572890 PZ |
2445 | int thiscpu = raw_smp_processor_id(); |
2446 | int thisnid = cpu_to_node(thiscpu); | |
98fa15f3 | 2447 | int polnid = NUMA_NO_NODE; |
771fb4d8 LS |
2448 | int ret = -1; |
2449 | ||
dd6eecb9 | 2450 | pol = get_vma_policy(vma, addr); |
771fb4d8 LS |
2451 | if (!(pol->flags & MPOL_F_MOF)) |
2452 | goto out; | |
2453 | ||
2454 | switch (pol->mode) { | |
2455 | case MPOL_INTERLEAVE: | |
771fb4d8 LS |
2456 | pgoff = vma->vm_pgoff; |
2457 | pgoff += (addr - vma->vm_start) >> PAGE_SHIFT; | |
98c70baa | 2458 | polnid = offset_il_node(pol, pgoff); |
771fb4d8 LS |
2459 | break; |
2460 | ||
2461 | case MPOL_PREFERRED: | |
7858d7bc FT |
2462 | polnid = pol->v.preferred_node; |
2463 | break; | |
2464 | ||
2465 | case MPOL_LOCAL: | |
2466 | polnid = numa_node_id(); | |
771fb4d8 LS |
2467 | break; |
2468 | ||
2469 | case MPOL_BIND: | |
bda420b9 HY |
2470 | /* Optimize placement among multiple nodes via NUMA balancing */ |
2471 | if (pol->flags & MPOL_F_MORON) { | |
2472 | if (node_isset(thisnid, pol->v.nodes)) | |
2473 | break; | |
2474 | goto out; | |
2475 | } | |
c33d6c06 | 2476 | |
771fb4d8 LS |
2477 | /* |
2478 | * allows binding to multiple nodes. | |
2479 | * use current page if in policy nodemask, | |
2480 | * else select nearest allowed node, if any. | |
2481 | * If no allowed nodes, use current [!misplaced]. | |
2482 | */ | |
2483 | if (node_isset(curnid, pol->v.nodes)) | |
2484 | goto out; | |
c33d6c06 | 2485 | z = first_zones_zonelist( |
771fb4d8 LS |
2486 | node_zonelist(numa_node_id(), GFP_HIGHUSER), |
2487 | gfp_zone(GFP_HIGHUSER), | |
c33d6c06 | 2488 | &pol->v.nodes); |
c1093b74 | 2489 | polnid = zone_to_nid(z->zone); |
771fb4d8 LS |
2490 | break; |
2491 | ||
2492 | default: | |
2493 | BUG(); | |
2494 | } | |
5606e387 MG |
2495 | |
2496 | /* Migrate the page towards the node whose CPU is referencing it */ | |
e42c8ff2 | 2497 | if (pol->flags & MPOL_F_MORON) { |
90572890 | 2498 | polnid = thisnid; |
5606e387 | 2499 | |
10f39042 | 2500 | if (!should_numa_migrate_memory(current, page, curnid, thiscpu)) |
de1c9ce6 | 2501 | goto out; |
e42c8ff2 MG |
2502 | } |
2503 | ||
771fb4d8 LS |
2504 | if (curnid != polnid) |
2505 | ret = polnid; | |
2506 | out: | |
2507 | mpol_cond_put(pol); | |
2508 | ||
2509 | return ret; | |
2510 | } | |
2511 | ||
c11600e4 DR |
2512 | /* |
2513 | * Drop the (possibly final) reference to task->mempolicy. It needs to be | |
2514 | * dropped after task->mempolicy is set to NULL so that any allocation done as | |
2515 | * part of its kmem_cache_free(), such as by KASAN, doesn't reference a freed | |
2516 | * policy. | |
2517 | */ | |
2518 | void mpol_put_task_policy(struct task_struct *task) | |
2519 | { | |
2520 | struct mempolicy *pol; | |
2521 | ||
2522 | task_lock(task); | |
2523 | pol = task->mempolicy; | |
2524 | task->mempolicy = NULL; | |
2525 | task_unlock(task); | |
2526 | mpol_put(pol); | |
2527 | } | |
2528 | ||
1da177e4 LT |
2529 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) |
2530 | { | |
140d5a49 | 2531 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 | 2532 | rb_erase(&n->nd, &sp->root); |
63f74ca2 | 2533 | sp_free(n); |
1da177e4 LT |
2534 | } |
2535 | ||
42288fe3 MG |
2536 | static void sp_node_init(struct sp_node *node, unsigned long start, |
2537 | unsigned long end, struct mempolicy *pol) | |
2538 | { | |
2539 | node->start = start; | |
2540 | node->end = end; | |
2541 | node->policy = pol; | |
2542 | } | |
2543 | ||
dbcb0f19 AB |
2544 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
2545 | struct mempolicy *pol) | |
1da177e4 | 2546 | { |
869833f2 KM |
2547 | struct sp_node *n; |
2548 | struct mempolicy *newpol; | |
1da177e4 | 2549 | |
869833f2 | 2550 | n = kmem_cache_alloc(sn_cache, GFP_KERNEL); |
1da177e4 LT |
2551 | if (!n) |
2552 | return NULL; | |
869833f2 KM |
2553 | |
2554 | newpol = mpol_dup(pol); | |
2555 | if (IS_ERR(newpol)) { | |
2556 | kmem_cache_free(sn_cache, n); | |
2557 | return NULL; | |
2558 | } | |
2559 | newpol->flags |= MPOL_F_SHARED; | |
42288fe3 | 2560 | sp_node_init(n, start, end, newpol); |
869833f2 | 2561 | |
1da177e4 LT |
2562 | return n; |
2563 | } | |
2564 | ||
2565 | /* Replace a policy range. */ | |
2566 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
2567 | unsigned long end, struct sp_node *new) | |
2568 | { | |
b22d127a | 2569 | struct sp_node *n; |
42288fe3 MG |
2570 | struct sp_node *n_new = NULL; |
2571 | struct mempolicy *mpol_new = NULL; | |
b22d127a | 2572 | int ret = 0; |
1da177e4 | 2573 | |
42288fe3 | 2574 | restart: |
4a8c7bb5 | 2575 | write_lock(&sp->lock); |
1da177e4 LT |
2576 | n = sp_lookup(sp, start, end); |
2577 | /* Take care of old policies in the same range. */ | |
2578 | while (n && n->start < end) { | |
2579 | struct rb_node *next = rb_next(&n->nd); | |
2580 | if (n->start >= start) { | |
2581 | if (n->end <= end) | |
2582 | sp_delete(sp, n); | |
2583 | else | |
2584 | n->start = end; | |
2585 | } else { | |
2586 | /* Old policy spanning whole new range. */ | |
2587 | if (n->end > end) { | |
42288fe3 MG |
2588 | if (!n_new) |
2589 | goto alloc_new; | |
2590 | ||
2591 | *mpol_new = *n->policy; | |
2592 | atomic_set(&mpol_new->refcnt, 1); | |
7880639c | 2593 | sp_node_init(n_new, end, n->end, mpol_new); |
1da177e4 | 2594 | n->end = start; |
5ca39575 | 2595 | sp_insert(sp, n_new); |
42288fe3 MG |
2596 | n_new = NULL; |
2597 | mpol_new = NULL; | |
1da177e4 LT |
2598 | break; |
2599 | } else | |
2600 | n->end = start; | |
2601 | } | |
2602 | if (!next) | |
2603 | break; | |
2604 | n = rb_entry(next, struct sp_node, nd); | |
2605 | } | |
2606 | if (new) | |
2607 | sp_insert(sp, new); | |
4a8c7bb5 | 2608 | write_unlock(&sp->lock); |
42288fe3 MG |
2609 | ret = 0; |
2610 | ||
2611 | err_out: | |
2612 | if (mpol_new) | |
2613 | mpol_put(mpol_new); | |
2614 | if (n_new) | |
2615 | kmem_cache_free(sn_cache, n_new); | |
2616 | ||
b22d127a | 2617 | return ret; |
42288fe3 MG |
2618 | |
2619 | alloc_new: | |
4a8c7bb5 | 2620 | write_unlock(&sp->lock); |
42288fe3 MG |
2621 | ret = -ENOMEM; |
2622 | n_new = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
2623 | if (!n_new) | |
2624 | goto err_out; | |
2625 | mpol_new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
2626 | if (!mpol_new) | |
2627 | goto err_out; | |
2628 | goto restart; | |
1da177e4 LT |
2629 | } |
2630 | ||
71fe804b LS |
2631 | /** |
2632 | * mpol_shared_policy_init - initialize shared policy for inode | |
2633 | * @sp: pointer to inode shared policy | |
2634 | * @mpol: struct mempolicy to install | |
2635 | * | |
2636 | * Install non-NULL @mpol in inode's shared policy rb-tree. | |
2637 | * On entry, the current task has a reference on a non-NULL @mpol. | |
2638 | * This must be released on exit. | |
4bfc4495 | 2639 | * This is called at get_inode() calls and we can use GFP_KERNEL. |
71fe804b LS |
2640 | */ |
2641 | void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol) | |
2642 | { | |
58568d2a MX |
2643 | int ret; |
2644 | ||
71fe804b | 2645 | sp->root = RB_ROOT; /* empty tree == default mempolicy */ |
4a8c7bb5 | 2646 | rwlock_init(&sp->lock); |
71fe804b LS |
2647 | |
2648 | if (mpol) { | |
2649 | struct vm_area_struct pvma; | |
2650 | struct mempolicy *new; | |
4bfc4495 | 2651 | NODEMASK_SCRATCH(scratch); |
71fe804b | 2652 | |
4bfc4495 | 2653 | if (!scratch) |
5c0c1654 | 2654 | goto put_mpol; |
71fe804b LS |
2655 | /* contextualize the tmpfs mount point mempolicy */ |
2656 | new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask); | |
15d77835 | 2657 | if (IS_ERR(new)) |
0cae3457 | 2658 | goto free_scratch; /* no valid nodemask intersection */ |
58568d2a MX |
2659 | |
2660 | task_lock(current); | |
4bfc4495 | 2661 | ret = mpol_set_nodemask(new, &mpol->w.user_nodemask, scratch); |
58568d2a | 2662 | task_unlock(current); |
15d77835 | 2663 | if (ret) |
5c0c1654 | 2664 | goto put_new; |
71fe804b LS |
2665 | |
2666 | /* Create pseudo-vma that contains just the policy */ | |
2c4541e2 | 2667 | vma_init(&pvma, NULL); |
71fe804b LS |
2668 | pvma.vm_end = TASK_SIZE; /* policy covers entire file */ |
2669 | mpol_set_shared_policy(sp, &pvma, new); /* adds ref */ | |
15d77835 | 2670 | |
5c0c1654 | 2671 | put_new: |
71fe804b | 2672 | mpol_put(new); /* drop initial ref */ |
0cae3457 | 2673 | free_scratch: |
4bfc4495 | 2674 | NODEMASK_SCRATCH_FREE(scratch); |
5c0c1654 LS |
2675 | put_mpol: |
2676 | mpol_put(mpol); /* drop our incoming ref on sb mpol */ | |
7339ff83 RH |
2677 | } |
2678 | } | |
2679 | ||
1da177e4 LT |
2680 | int mpol_set_shared_policy(struct shared_policy *info, |
2681 | struct vm_area_struct *vma, struct mempolicy *npol) | |
2682 | { | |
2683 | int err; | |
2684 | struct sp_node *new = NULL; | |
2685 | unsigned long sz = vma_pages(vma); | |
2686 | ||
028fec41 | 2687 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 2688 | vma->vm_pgoff, |
45c4745a | 2689 | sz, npol ? npol->mode : -1, |
028fec41 | 2690 | npol ? npol->flags : -1, |
00ef2d2f | 2691 | npol ? nodes_addr(npol->v.nodes)[0] : NUMA_NO_NODE); |
1da177e4 LT |
2692 | |
2693 | if (npol) { | |
2694 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
2695 | if (!new) | |
2696 | return -ENOMEM; | |
2697 | } | |
2698 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
2699 | if (err && new) | |
63f74ca2 | 2700 | sp_free(new); |
1da177e4 LT |
2701 | return err; |
2702 | } | |
2703 | ||
2704 | /* Free a backing policy store on inode delete. */ | |
2705 | void mpol_free_shared_policy(struct shared_policy *p) | |
2706 | { | |
2707 | struct sp_node *n; | |
2708 | struct rb_node *next; | |
2709 | ||
2710 | if (!p->root.rb_node) | |
2711 | return; | |
4a8c7bb5 | 2712 | write_lock(&p->lock); |
1da177e4 LT |
2713 | next = rb_first(&p->root); |
2714 | while (next) { | |
2715 | n = rb_entry(next, struct sp_node, nd); | |
2716 | next = rb_next(&n->nd); | |
63f74ca2 | 2717 | sp_delete(p, n); |
1da177e4 | 2718 | } |
4a8c7bb5 | 2719 | write_unlock(&p->lock); |
1da177e4 LT |
2720 | } |
2721 | ||
1a687c2e | 2722 | #ifdef CONFIG_NUMA_BALANCING |
c297663c | 2723 | static int __initdata numabalancing_override; |
1a687c2e MG |
2724 | |
2725 | static void __init check_numabalancing_enable(void) | |
2726 | { | |
2727 | bool numabalancing_default = false; | |
2728 | ||
2729 | if (IS_ENABLED(CONFIG_NUMA_BALANCING_DEFAULT_ENABLED)) | |
2730 | numabalancing_default = true; | |
2731 | ||
c297663c MG |
2732 | /* Parsed by setup_numabalancing. override == 1 enables, -1 disables */ |
2733 | if (numabalancing_override) | |
2734 | set_numabalancing_state(numabalancing_override == 1); | |
2735 | ||
b0dc2b9b | 2736 | if (num_online_nodes() > 1 && !numabalancing_override) { |
756a025f | 2737 | pr_info("%s automatic NUMA balancing. Configure with numa_balancing= or the kernel.numa_balancing sysctl\n", |
c297663c | 2738 | numabalancing_default ? "Enabling" : "Disabling"); |
1a687c2e MG |
2739 | set_numabalancing_state(numabalancing_default); |
2740 | } | |
2741 | } | |
2742 | ||
2743 | static int __init setup_numabalancing(char *str) | |
2744 | { | |
2745 | int ret = 0; | |
2746 | if (!str) | |
2747 | goto out; | |
1a687c2e MG |
2748 | |
2749 | if (!strcmp(str, "enable")) { | |
c297663c | 2750 | numabalancing_override = 1; |
1a687c2e MG |
2751 | ret = 1; |
2752 | } else if (!strcmp(str, "disable")) { | |
c297663c | 2753 | numabalancing_override = -1; |
1a687c2e MG |
2754 | ret = 1; |
2755 | } | |
2756 | out: | |
2757 | if (!ret) | |
4a404bea | 2758 | pr_warn("Unable to parse numa_balancing=\n"); |
1a687c2e MG |
2759 | |
2760 | return ret; | |
2761 | } | |
2762 | __setup("numa_balancing=", setup_numabalancing); | |
2763 | #else | |
2764 | static inline void __init check_numabalancing_enable(void) | |
2765 | { | |
2766 | } | |
2767 | #endif /* CONFIG_NUMA_BALANCING */ | |
2768 | ||
1da177e4 LT |
2769 | /* assumes fs == KERNEL_DS */ |
2770 | void __init numa_policy_init(void) | |
2771 | { | |
b71636e2 PM |
2772 | nodemask_t interleave_nodes; |
2773 | unsigned long largest = 0; | |
2774 | int nid, prefer = 0; | |
2775 | ||
1da177e4 LT |
2776 | policy_cache = kmem_cache_create("numa_policy", |
2777 | sizeof(struct mempolicy), | |
20c2df83 | 2778 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
2779 | |
2780 | sn_cache = kmem_cache_create("shared_policy_node", | |
2781 | sizeof(struct sp_node), | |
20c2df83 | 2782 | 0, SLAB_PANIC, NULL); |
1da177e4 | 2783 | |
5606e387 MG |
2784 | for_each_node(nid) { |
2785 | preferred_node_policy[nid] = (struct mempolicy) { | |
2786 | .refcnt = ATOMIC_INIT(1), | |
2787 | .mode = MPOL_PREFERRED, | |
2788 | .flags = MPOL_F_MOF | MPOL_F_MORON, | |
2789 | .v = { .preferred_node = nid, }, | |
2790 | }; | |
2791 | } | |
2792 | ||
b71636e2 PM |
2793 | /* |
2794 | * Set interleaving policy for system init. Interleaving is only | |
2795 | * enabled across suitably sized nodes (default is >= 16MB), or | |
2796 | * fall back to the largest node if they're all smaller. | |
2797 | */ | |
2798 | nodes_clear(interleave_nodes); | |
01f13bd6 | 2799 | for_each_node_state(nid, N_MEMORY) { |
b71636e2 PM |
2800 | unsigned long total_pages = node_present_pages(nid); |
2801 | ||
2802 | /* Preserve the largest node */ | |
2803 | if (largest < total_pages) { | |
2804 | largest = total_pages; | |
2805 | prefer = nid; | |
2806 | } | |
2807 | ||
2808 | /* Interleave this node? */ | |
2809 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
2810 | node_set(nid, interleave_nodes); | |
2811 | } | |
2812 | ||
2813 | /* All too small, use the largest */ | |
2814 | if (unlikely(nodes_empty(interleave_nodes))) | |
2815 | node_set(prefer, interleave_nodes); | |
1da177e4 | 2816 | |
028fec41 | 2817 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
b1de0d13 | 2818 | pr_err("%s: interleaving failed\n", __func__); |
1a687c2e MG |
2819 | |
2820 | check_numabalancing_enable(); | |
1da177e4 LT |
2821 | } |
2822 | ||
8bccd85f | 2823 | /* Reset policy of current process to default */ |
1da177e4 LT |
2824 | void numa_default_policy(void) |
2825 | { | |
028fec41 | 2826 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 2827 | } |
68860ec1 | 2828 | |
095f1fc4 LS |
2829 | /* |
2830 | * Parse and format mempolicy from/to strings | |
2831 | */ | |
2832 | ||
345ace9c LS |
2833 | static const char * const policy_modes[] = |
2834 | { | |
2835 | [MPOL_DEFAULT] = "default", | |
2836 | [MPOL_PREFERRED] = "prefer", | |
2837 | [MPOL_BIND] = "bind", | |
2838 | [MPOL_INTERLEAVE] = "interleave", | |
d3a71033 | 2839 | [MPOL_LOCAL] = "local", |
345ace9c | 2840 | }; |
1a75a6c8 | 2841 | |
095f1fc4 LS |
2842 | |
2843 | #ifdef CONFIG_TMPFS | |
2844 | /** | |
f2a07f40 | 2845 | * mpol_parse_str - parse string to mempolicy, for tmpfs mpol mount option. |
095f1fc4 | 2846 | * @str: string containing mempolicy to parse |
71fe804b | 2847 | * @mpol: pointer to struct mempolicy pointer, returned on success. |
095f1fc4 LS |
2848 | * |
2849 | * Format of input: | |
2850 | * <mode>[=<flags>][:<nodelist>] | |
2851 | * | |
71fe804b | 2852 | * On success, returns 0, else 1 |
095f1fc4 | 2853 | */ |
a7a88b23 | 2854 | int mpol_parse_str(char *str, struct mempolicy **mpol) |
095f1fc4 | 2855 | { |
71fe804b | 2856 | struct mempolicy *new = NULL; |
f2a07f40 | 2857 | unsigned short mode_flags; |
71fe804b | 2858 | nodemask_t nodes; |
095f1fc4 LS |
2859 | char *nodelist = strchr(str, ':'); |
2860 | char *flags = strchr(str, '='); | |
dedf2c73 | 2861 | int err = 1, mode; |
095f1fc4 | 2862 | |
c7a91bc7 DC |
2863 | if (flags) |
2864 | *flags++ = '\0'; /* terminate mode string */ | |
2865 | ||
095f1fc4 LS |
2866 | if (nodelist) { |
2867 | /* NUL-terminate mode or flags string */ | |
2868 | *nodelist++ = '\0'; | |
71fe804b | 2869 | if (nodelist_parse(nodelist, nodes)) |
095f1fc4 | 2870 | goto out; |
01f13bd6 | 2871 | if (!nodes_subset(nodes, node_states[N_MEMORY])) |
095f1fc4 | 2872 | goto out; |
71fe804b LS |
2873 | } else |
2874 | nodes_clear(nodes); | |
2875 | ||
dedf2c73 | 2876 | mode = match_string(policy_modes, MPOL_MAX, str); |
2877 | if (mode < 0) | |
095f1fc4 LS |
2878 | goto out; |
2879 | ||
71fe804b | 2880 | switch (mode) { |
095f1fc4 | 2881 | case MPOL_PREFERRED: |
71fe804b | 2882 | /* |
aa9f7d51 RD |
2883 | * Insist on a nodelist of one node only, although later |
2884 | * we use first_node(nodes) to grab a single node, so here | |
2885 | * nodelist (or nodes) cannot be empty. | |
71fe804b | 2886 | */ |
095f1fc4 LS |
2887 | if (nodelist) { |
2888 | char *rest = nodelist; | |
2889 | while (isdigit(*rest)) | |
2890 | rest++; | |
926f2ae0 KM |
2891 | if (*rest) |
2892 | goto out; | |
aa9f7d51 RD |
2893 | if (nodes_empty(nodes)) |
2894 | goto out; | |
095f1fc4 LS |
2895 | } |
2896 | break; | |
095f1fc4 LS |
2897 | case MPOL_INTERLEAVE: |
2898 | /* | |
2899 | * Default to online nodes with memory if no nodelist | |
2900 | */ | |
2901 | if (!nodelist) | |
01f13bd6 | 2902 | nodes = node_states[N_MEMORY]; |
3f226aa1 | 2903 | break; |
71fe804b | 2904 | case MPOL_LOCAL: |
3f226aa1 | 2905 | /* |
71fe804b | 2906 | * Don't allow a nodelist; mpol_new() checks flags |
3f226aa1 | 2907 | */ |
71fe804b | 2908 | if (nodelist) |
3f226aa1 | 2909 | goto out; |
3f226aa1 | 2910 | break; |
413b43de RT |
2911 | case MPOL_DEFAULT: |
2912 | /* | |
2913 | * Insist on a empty nodelist | |
2914 | */ | |
2915 | if (!nodelist) | |
2916 | err = 0; | |
2917 | goto out; | |
d69b2e63 KM |
2918 | case MPOL_BIND: |
2919 | /* | |
2920 | * Insist on a nodelist | |
2921 | */ | |
2922 | if (!nodelist) | |
2923 | goto out; | |
095f1fc4 LS |
2924 | } |
2925 | ||
71fe804b | 2926 | mode_flags = 0; |
095f1fc4 LS |
2927 | if (flags) { |
2928 | /* | |
2929 | * Currently, we only support two mutually exclusive | |
2930 | * mode flags. | |
2931 | */ | |
2932 | if (!strcmp(flags, "static")) | |
71fe804b | 2933 | mode_flags |= MPOL_F_STATIC_NODES; |
095f1fc4 | 2934 | else if (!strcmp(flags, "relative")) |
71fe804b | 2935 | mode_flags |= MPOL_F_RELATIVE_NODES; |
095f1fc4 | 2936 | else |
926f2ae0 | 2937 | goto out; |
095f1fc4 | 2938 | } |
71fe804b LS |
2939 | |
2940 | new = mpol_new(mode, mode_flags, &nodes); | |
2941 | if (IS_ERR(new)) | |
926f2ae0 KM |
2942 | goto out; |
2943 | ||
f2a07f40 HD |
2944 | /* |
2945 | * Save nodes for mpol_to_str() to show the tmpfs mount options | |
2946 | * for /proc/mounts, /proc/pid/mounts and /proc/pid/mountinfo. | |
2947 | */ | |
2948 | if (mode != MPOL_PREFERRED) | |
2949 | new->v.nodes = nodes; | |
2950 | else if (nodelist) | |
2951 | new->v.preferred_node = first_node(nodes); | |
2952 | else | |
7858d7bc | 2953 | new->mode = MPOL_LOCAL; |
f2a07f40 HD |
2954 | |
2955 | /* | |
2956 | * Save nodes for contextualization: this will be used to "clone" | |
2957 | * the mempolicy in a specific context [cpuset] at a later time. | |
2958 | */ | |
2959 | new->w.user_nodemask = nodes; | |
2960 | ||
926f2ae0 | 2961 | err = 0; |
71fe804b | 2962 | |
095f1fc4 LS |
2963 | out: |
2964 | /* Restore string for error message */ | |
2965 | if (nodelist) | |
2966 | *--nodelist = ':'; | |
2967 | if (flags) | |
2968 | *--flags = '='; | |
71fe804b LS |
2969 | if (!err) |
2970 | *mpol = new; | |
095f1fc4 LS |
2971 | return err; |
2972 | } | |
2973 | #endif /* CONFIG_TMPFS */ | |
2974 | ||
71fe804b LS |
2975 | /** |
2976 | * mpol_to_str - format a mempolicy structure for printing | |
2977 | * @buffer: to contain formatted mempolicy string | |
2978 | * @maxlen: length of @buffer | |
2979 | * @pol: pointer to mempolicy to be formatted | |
71fe804b | 2980 | * |
948927ee DR |
2981 | * Convert @pol into a string. If @buffer is too short, truncate the string. |
2982 | * Recommend a @maxlen of at least 32 for the longest mode, "interleave", the | |
2983 | * longest flag, "relative", and to display at least a few node ids. | |
1a75a6c8 | 2984 | */ |
948927ee | 2985 | void mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) |
1a75a6c8 CL |
2986 | { |
2987 | char *p = buffer; | |
948927ee DR |
2988 | nodemask_t nodes = NODE_MASK_NONE; |
2989 | unsigned short mode = MPOL_DEFAULT; | |
2990 | unsigned short flags = 0; | |
2291990a | 2991 | |
8790c71a | 2992 | if (pol && pol != &default_policy && !(pol->flags & MPOL_F_MORON)) { |
bea904d5 | 2993 | mode = pol->mode; |
948927ee DR |
2994 | flags = pol->flags; |
2995 | } | |
bea904d5 | 2996 | |
1a75a6c8 CL |
2997 | switch (mode) { |
2998 | case MPOL_DEFAULT: | |
7858d7bc | 2999 | case MPOL_LOCAL: |
1a75a6c8 | 3000 | break; |
1a75a6c8 | 3001 | case MPOL_PREFERRED: |
7858d7bc | 3002 | node_set(pol->v.preferred_node, nodes); |
1a75a6c8 | 3003 | break; |
1a75a6c8 | 3004 | case MPOL_BIND: |
1a75a6c8 | 3005 | case MPOL_INTERLEAVE: |
f2a07f40 | 3006 | nodes = pol->v.nodes; |
1a75a6c8 | 3007 | break; |
1a75a6c8 | 3008 | default: |
948927ee DR |
3009 | WARN_ON_ONCE(1); |
3010 | snprintf(p, maxlen, "unknown"); | |
3011 | return; | |
1a75a6c8 CL |
3012 | } |
3013 | ||
b7a9f420 | 3014 | p += snprintf(p, maxlen, "%s", policy_modes[mode]); |
1a75a6c8 | 3015 | |
fc36b8d3 | 3016 | if (flags & MPOL_MODE_FLAGS) { |
948927ee | 3017 | p += snprintf(p, buffer + maxlen - p, "="); |
f5b087b5 | 3018 | |
2291990a LS |
3019 | /* |
3020 | * Currently, the only defined flags are mutually exclusive | |
3021 | */ | |
f5b087b5 | 3022 | if (flags & MPOL_F_STATIC_NODES) |
2291990a LS |
3023 | p += snprintf(p, buffer + maxlen - p, "static"); |
3024 | else if (flags & MPOL_F_RELATIVE_NODES) | |
3025 | p += snprintf(p, buffer + maxlen - p, "relative"); | |
f5b087b5 DR |
3026 | } |
3027 | ||
9e763e0f TH |
3028 | if (!nodes_empty(nodes)) |
3029 | p += scnprintf(p, buffer + maxlen - p, ":%*pbl", | |
3030 | nodemask_pr_args(&nodes)); | |
1a75a6c8 | 3031 | } |