Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Simple NUMA memory policy for the Linux kernel. | |
3 | * | |
4 | * Copyright 2003,2004 Andi Kleen, SuSE Labs. | |
8bccd85f | 5 | * (C) Copyright 2005 Christoph Lameter, Silicon Graphics, Inc. |
1da177e4 LT |
6 | * Subject to the GNU Public License, version 2. |
7 | * | |
8 | * NUMA policy allows the user to give hints in which node(s) memory should | |
9 | * be allocated. | |
10 | * | |
11 | * Support four policies per VMA and per process: | |
12 | * | |
13 | * The VMA policy has priority over the process policy for a page fault. | |
14 | * | |
15 | * interleave Allocate memory interleaved over a set of nodes, | |
16 | * with normal fallback if it fails. | |
17 | * For VMA based allocations this interleaves based on the | |
18 | * offset into the backing object or offset into the mapping | |
19 | * for anonymous memory. For process policy an process counter | |
20 | * is used. | |
8bccd85f | 21 | * |
1da177e4 LT |
22 | * bind Only allocate memory on a specific set of nodes, |
23 | * no fallback. | |
8bccd85f CL |
24 | * FIXME: memory is allocated starting with the first node |
25 | * to the last. It would be better if bind would truly restrict | |
26 | * the allocation to memory nodes instead | |
27 | * | |
1da177e4 LT |
28 | * preferred Try a specific node first before normal fallback. |
29 | * As a special case node -1 here means do the allocation | |
30 | * on the local CPU. This is normally identical to default, | |
31 | * but useful to set in a VMA when you have a non default | |
32 | * process policy. | |
8bccd85f | 33 | * |
1da177e4 LT |
34 | * default Allocate on the local node first, or when on a VMA |
35 | * use the process policy. This is what Linux always did | |
36 | * in a NUMA aware kernel and still does by, ahem, default. | |
37 | * | |
38 | * The process policy is applied for most non interrupt memory allocations | |
39 | * in that process' context. Interrupts ignore the policies and always | |
40 | * try to allocate on the local CPU. The VMA policy is only applied for memory | |
41 | * allocations for a VMA in the VM. | |
42 | * | |
43 | * Currently there are a few corner cases in swapping where the policy | |
44 | * is not applied, but the majority should be handled. When process policy | |
45 | * is used it is not remembered over swap outs/swap ins. | |
46 | * | |
47 | * Only the highest zone in the zone hierarchy gets policied. Allocations | |
48 | * requesting a lower zone just use default policy. This implies that | |
49 | * on systems with highmem kernel lowmem allocation don't get policied. | |
50 | * Same with GFP_DMA allocations. | |
51 | * | |
52 | * For shmfs/tmpfs/hugetlbfs shared memory the policy is shared between | |
53 | * all users and remembered even when nobody has memory mapped. | |
54 | */ | |
55 | ||
56 | /* Notebook: | |
57 | fix mmap readahead to honour policy and enable policy for any page cache | |
58 | object | |
59 | statistics for bigpages | |
60 | global policy for page cache? currently it uses process policy. Requires | |
61 | first item above. | |
62 | handle mremap for shared memory (currently ignored for the policy) | |
63 | grows down? | |
64 | make bind policy root only? It can trigger oom much faster and the | |
65 | kernel is not always grateful with that. | |
66 | could replace all the switch()es with a mempolicy_ops structure. | |
67 | */ | |
68 | ||
69 | #include <linux/mempolicy.h> | |
70 | #include <linux/mm.h> | |
71 | #include <linux/highmem.h> | |
72 | #include <linux/hugetlb.h> | |
73 | #include <linux/kernel.h> | |
74 | #include <linux/sched.h> | |
1da177e4 LT |
75 | #include <linux/nodemask.h> |
76 | #include <linux/cpuset.h> | |
77 | #include <linux/gfp.h> | |
78 | #include <linux/slab.h> | |
79 | #include <linux/string.h> | |
80 | #include <linux/module.h> | |
b488893a | 81 | #include <linux/nsproxy.h> |
1da177e4 LT |
82 | #include <linux/interrupt.h> |
83 | #include <linux/init.h> | |
84 | #include <linux/compat.h> | |
dc9aa5b9 | 85 | #include <linux/swap.h> |
1a75a6c8 CL |
86 | #include <linux/seq_file.h> |
87 | #include <linux/proc_fs.h> | |
b20a3503 | 88 | #include <linux/migrate.h> |
95a402c3 | 89 | #include <linux/rmap.h> |
86c3a764 | 90 | #include <linux/security.h> |
dbcb0f19 | 91 | #include <linux/syscalls.h> |
dc9aa5b9 | 92 | |
1da177e4 LT |
93 | #include <asm/tlbflush.h> |
94 | #include <asm/uaccess.h> | |
95 | ||
38e35860 | 96 | /* Internal flags */ |
dc9aa5b9 | 97 | #define MPOL_MF_DISCONTIG_OK (MPOL_MF_INTERNAL << 0) /* Skip checks for continuous vmas */ |
38e35860 | 98 | #define MPOL_MF_INVERT (MPOL_MF_INTERNAL << 1) /* Invert check for nodemask */ |
1a75a6c8 | 99 | #define MPOL_MF_STATS (MPOL_MF_INTERNAL << 2) /* Gather statistics */ |
dc9aa5b9 | 100 | |
fcc234f8 PE |
101 | static struct kmem_cache *policy_cache; |
102 | static struct kmem_cache *sn_cache; | |
1da177e4 | 103 | |
1da177e4 LT |
104 | /* Highest zone. An specific allocation for a zone below that is not |
105 | policied. */ | |
6267276f | 106 | enum zone_type policy_zone = 0; |
1da177e4 | 107 | |
d42c6997 | 108 | struct mempolicy default_policy = { |
1da177e4 LT |
109 | .refcnt = ATOMIC_INIT(1), /* never free it */ |
110 | .policy = MPOL_DEFAULT, | |
111 | }; | |
112 | ||
19770b32 MG |
113 | /* Check that the nodemask contains at least one populated zone */ |
114 | static int is_valid_nodemask(nodemask_t *nodemask) | |
1da177e4 | 115 | { |
19770b32 | 116 | int nd, k; |
1da177e4 | 117 | |
19770b32 MG |
118 | /* Check that there is something useful in this mask */ |
119 | k = policy_zone; | |
120 | ||
121 | for_each_node_mask(nd, *nodemask) { | |
122 | struct zone *z; | |
123 | ||
124 | for (k = 0; k <= policy_zone; k++) { | |
125 | z = &NODE_DATA(nd)->node_zones[k]; | |
126 | if (z->present_pages > 0) | |
127 | return 1; | |
dd942ae3 | 128 | } |
8af5e2eb | 129 | } |
19770b32 MG |
130 | |
131 | return 0; | |
1da177e4 LT |
132 | } |
133 | ||
f5b087b5 DR |
134 | static inline int mpol_store_user_nodemask(const struct mempolicy *pol) |
135 | { | |
4c50bc01 DR |
136 | return pol->flags & (MPOL_F_STATIC_NODES | MPOL_F_RELATIVE_NODES); |
137 | } | |
138 | ||
139 | static void mpol_relative_nodemask(nodemask_t *ret, const nodemask_t *orig, | |
140 | const nodemask_t *rel) | |
141 | { | |
142 | nodemask_t tmp; | |
143 | nodes_fold(tmp, *orig, nodes_weight(*rel)); | |
144 | nodes_onto(*ret, tmp, *rel); | |
f5b087b5 DR |
145 | } |
146 | ||
1da177e4 | 147 | /* Create a new policy */ |
028fec41 DR |
148 | static struct mempolicy *mpol_new(unsigned short mode, unsigned short flags, |
149 | nodemask_t *nodes) | |
1da177e4 LT |
150 | { |
151 | struct mempolicy *policy; | |
f5b087b5 | 152 | nodemask_t cpuset_context_nmask; |
1da177e4 | 153 | |
028fec41 DR |
154 | pr_debug("setting mode %d flags %d nodes[0] %lx\n", |
155 | mode, flags, nodes ? nodes_addr(*nodes)[0] : -1); | |
140d5a49 | 156 | |
1da177e4 | 157 | if (mode == MPOL_DEFAULT) |
f5b087b5 DR |
158 | return (nodes && nodes_weight(*nodes)) ? ERR_PTR(-EINVAL) : |
159 | NULL; | |
1da177e4 LT |
160 | policy = kmem_cache_alloc(policy_cache, GFP_KERNEL); |
161 | if (!policy) | |
162 | return ERR_PTR(-ENOMEM); | |
163 | atomic_set(&policy->refcnt, 1); | |
f5b087b5 | 164 | cpuset_update_task_memory_state(); |
4c50bc01 DR |
165 | if (flags & MPOL_F_RELATIVE_NODES) |
166 | mpol_relative_nodemask(&cpuset_context_nmask, nodes, | |
167 | &cpuset_current_mems_allowed); | |
168 | else | |
169 | nodes_and(cpuset_context_nmask, *nodes, | |
170 | cpuset_current_mems_allowed); | |
1da177e4 LT |
171 | switch (mode) { |
172 | case MPOL_INTERLEAVE: | |
f5b087b5 DR |
173 | if (nodes_empty(*nodes) || nodes_empty(cpuset_context_nmask)) |
174 | goto free; | |
175 | policy->v.nodes = cpuset_context_nmask; | |
1da177e4 LT |
176 | break; |
177 | case MPOL_PREFERRED: | |
f5b087b5 | 178 | policy->v.preferred_node = first_node(cpuset_context_nmask); |
1da177e4 | 179 | if (policy->v.preferred_node >= MAX_NUMNODES) |
f5b087b5 | 180 | goto free; |
1da177e4 LT |
181 | break; |
182 | case MPOL_BIND: | |
f5b087b5 DR |
183 | if (!is_valid_nodemask(&cpuset_context_nmask)) |
184 | goto free; | |
185 | policy->v.nodes = cpuset_context_nmask; | |
1da177e4 | 186 | break; |
a3b51e01 DR |
187 | default: |
188 | BUG(); | |
1da177e4 LT |
189 | } |
190 | policy->policy = mode; | |
028fec41 | 191 | policy->flags = flags; |
f5b087b5 DR |
192 | if (mpol_store_user_nodemask(policy)) |
193 | policy->w.user_nodemask = *nodes; | |
194 | else | |
195 | policy->w.cpuset_mems_allowed = cpuset_mems_allowed(current); | |
1da177e4 | 196 | return policy; |
f5b087b5 DR |
197 | |
198 | free: | |
199 | kmem_cache_free(policy_cache, policy); | |
200 | return ERR_PTR(-EINVAL); | |
1da177e4 LT |
201 | } |
202 | ||
1d0d2680 DR |
203 | /* Migrate a policy to a different set of nodes */ |
204 | static void mpol_rebind_policy(struct mempolicy *pol, | |
205 | const nodemask_t *newmask) | |
206 | { | |
207 | nodemask_t tmp; | |
208 | int static_nodes; | |
209 | int relative_nodes; | |
210 | ||
211 | if (!pol) | |
212 | return; | |
213 | static_nodes = pol->flags & MPOL_F_STATIC_NODES; | |
214 | relative_nodes = pol->flags & MPOL_F_RELATIVE_NODES; | |
215 | if (!mpol_store_user_nodemask(pol) && | |
216 | nodes_equal(pol->w.cpuset_mems_allowed, *newmask)) | |
217 | return; | |
218 | ||
219 | switch (pol->policy) { | |
220 | case MPOL_DEFAULT: | |
221 | break; | |
222 | case MPOL_BIND: | |
223 | /* Fall through */ | |
224 | case MPOL_INTERLEAVE: | |
225 | if (static_nodes) | |
226 | nodes_and(tmp, pol->w.user_nodemask, *newmask); | |
227 | else if (relative_nodes) | |
228 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, | |
229 | newmask); | |
230 | else { | |
231 | nodes_remap(tmp, pol->v.nodes, | |
232 | pol->w.cpuset_mems_allowed, *newmask); | |
233 | pol->w.cpuset_mems_allowed = *newmask; | |
234 | } | |
235 | pol->v.nodes = tmp; | |
236 | if (!node_isset(current->il_next, tmp)) { | |
237 | current->il_next = next_node(current->il_next, tmp); | |
238 | if (current->il_next >= MAX_NUMNODES) | |
239 | current->il_next = first_node(tmp); | |
240 | if (current->il_next >= MAX_NUMNODES) | |
241 | current->il_next = numa_node_id(); | |
242 | } | |
243 | break; | |
244 | case MPOL_PREFERRED: | |
245 | if (static_nodes) { | |
246 | int node = first_node(pol->w.user_nodemask); | |
247 | ||
248 | if (node_isset(node, *newmask)) | |
249 | pol->v.preferred_node = node; | |
250 | else | |
251 | pol->v.preferred_node = -1; | |
252 | } else if (relative_nodes) { | |
253 | mpol_relative_nodemask(&tmp, &pol->w.user_nodemask, | |
254 | newmask); | |
255 | pol->v.preferred_node = first_node(tmp); | |
256 | } else { | |
257 | pol->v.preferred_node = node_remap(pol->v.preferred_node, | |
258 | pol->w.cpuset_mems_allowed, *newmask); | |
259 | pol->w.cpuset_mems_allowed = *newmask; | |
260 | } | |
261 | break; | |
262 | default: | |
263 | BUG(); | |
264 | break; | |
265 | } | |
266 | } | |
267 | ||
268 | /* | |
269 | * Wrapper for mpol_rebind_policy() that just requires task | |
270 | * pointer, and updates task mempolicy. | |
271 | */ | |
272 | ||
273 | void mpol_rebind_task(struct task_struct *tsk, const nodemask_t *new) | |
274 | { | |
275 | mpol_rebind_policy(tsk->mempolicy, new); | |
276 | } | |
277 | ||
278 | /* | |
279 | * Rebind each vma in mm to new nodemask. | |
280 | * | |
281 | * Call holding a reference to mm. Takes mm->mmap_sem during call. | |
282 | */ | |
283 | ||
284 | void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new) | |
285 | { | |
286 | struct vm_area_struct *vma; | |
287 | ||
288 | down_write(&mm->mmap_sem); | |
289 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
290 | mpol_rebind_policy(vma->vm_policy, new); | |
291 | up_write(&mm->mmap_sem); | |
292 | } | |
293 | ||
397874df | 294 | static void gather_stats(struct page *, void *, int pte_dirty); |
fc301289 CL |
295 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
296 | unsigned long flags); | |
1a75a6c8 | 297 | |
38e35860 | 298 | /* Scan through pages checking if pages follow certain conditions. */ |
b5810039 | 299 | static int check_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
dc9aa5b9 CL |
300 | unsigned long addr, unsigned long end, |
301 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 302 | void *private) |
1da177e4 | 303 | { |
91612e0d HD |
304 | pte_t *orig_pte; |
305 | pte_t *pte; | |
705e87c0 | 306 | spinlock_t *ptl; |
941150a3 | 307 | |
705e87c0 | 308 | orig_pte = pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
91612e0d | 309 | do { |
6aab341e | 310 | struct page *page; |
25ba77c1 | 311 | int nid; |
91612e0d HD |
312 | |
313 | if (!pte_present(*pte)) | |
1da177e4 | 314 | continue; |
6aab341e LT |
315 | page = vm_normal_page(vma, addr, *pte); |
316 | if (!page) | |
1da177e4 | 317 | continue; |
053837fc NP |
318 | /* |
319 | * The check for PageReserved here is important to avoid | |
320 | * handling zero pages and other pages that may have been | |
321 | * marked special by the system. | |
322 | * | |
323 | * If the PageReserved would not be checked here then f.e. | |
324 | * the location of the zero page could have an influence | |
325 | * on MPOL_MF_STRICT, zero pages would be counted for | |
326 | * the per node stats, and there would be useless attempts | |
327 | * to put zero pages on the migration list. | |
328 | */ | |
f4598c8b CL |
329 | if (PageReserved(page)) |
330 | continue; | |
6aab341e | 331 | nid = page_to_nid(page); |
38e35860 CL |
332 | if (node_isset(nid, *nodes) == !!(flags & MPOL_MF_INVERT)) |
333 | continue; | |
334 | ||
1a75a6c8 | 335 | if (flags & MPOL_MF_STATS) |
397874df | 336 | gather_stats(page, private, pte_dirty(*pte)); |
053837fc | 337 | else if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) |
fc301289 | 338 | migrate_page_add(page, private, flags); |
38e35860 CL |
339 | else |
340 | break; | |
91612e0d | 341 | } while (pte++, addr += PAGE_SIZE, addr != end); |
705e87c0 | 342 | pte_unmap_unlock(orig_pte, ptl); |
91612e0d HD |
343 | return addr != end; |
344 | } | |
345 | ||
b5810039 | 346 | static inline int check_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
dc9aa5b9 CL |
347 | unsigned long addr, unsigned long end, |
348 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 349 | void *private) |
91612e0d HD |
350 | { |
351 | pmd_t *pmd; | |
352 | unsigned long next; | |
353 | ||
354 | pmd = pmd_offset(pud, addr); | |
355 | do { | |
356 | next = pmd_addr_end(addr, end); | |
357 | if (pmd_none_or_clear_bad(pmd)) | |
358 | continue; | |
dc9aa5b9 | 359 | if (check_pte_range(vma, pmd, addr, next, nodes, |
38e35860 | 360 | flags, private)) |
91612e0d HD |
361 | return -EIO; |
362 | } while (pmd++, addr = next, addr != end); | |
363 | return 0; | |
364 | } | |
365 | ||
b5810039 | 366 | static inline int check_pud_range(struct vm_area_struct *vma, pgd_t *pgd, |
dc9aa5b9 CL |
367 | unsigned long addr, unsigned long end, |
368 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 369 | void *private) |
91612e0d HD |
370 | { |
371 | pud_t *pud; | |
372 | unsigned long next; | |
373 | ||
374 | pud = pud_offset(pgd, addr); | |
375 | do { | |
376 | next = pud_addr_end(addr, end); | |
377 | if (pud_none_or_clear_bad(pud)) | |
378 | continue; | |
dc9aa5b9 | 379 | if (check_pmd_range(vma, pud, addr, next, nodes, |
38e35860 | 380 | flags, private)) |
91612e0d HD |
381 | return -EIO; |
382 | } while (pud++, addr = next, addr != end); | |
383 | return 0; | |
384 | } | |
385 | ||
b5810039 | 386 | static inline int check_pgd_range(struct vm_area_struct *vma, |
dc9aa5b9 CL |
387 | unsigned long addr, unsigned long end, |
388 | const nodemask_t *nodes, unsigned long flags, | |
38e35860 | 389 | void *private) |
91612e0d HD |
390 | { |
391 | pgd_t *pgd; | |
392 | unsigned long next; | |
393 | ||
b5810039 | 394 | pgd = pgd_offset(vma->vm_mm, addr); |
91612e0d HD |
395 | do { |
396 | next = pgd_addr_end(addr, end); | |
397 | if (pgd_none_or_clear_bad(pgd)) | |
398 | continue; | |
dc9aa5b9 | 399 | if (check_pud_range(vma, pgd, addr, next, nodes, |
38e35860 | 400 | flags, private)) |
91612e0d HD |
401 | return -EIO; |
402 | } while (pgd++, addr = next, addr != end); | |
403 | return 0; | |
1da177e4 LT |
404 | } |
405 | ||
dc9aa5b9 CL |
406 | /* |
407 | * Check if all pages in a range are on a set of nodes. | |
408 | * If pagelist != NULL then isolate pages from the LRU and | |
409 | * put them on the pagelist. | |
410 | */ | |
1da177e4 LT |
411 | static struct vm_area_struct * |
412 | check_range(struct mm_struct *mm, unsigned long start, unsigned long end, | |
38e35860 | 413 | const nodemask_t *nodes, unsigned long flags, void *private) |
1da177e4 LT |
414 | { |
415 | int err; | |
416 | struct vm_area_struct *first, *vma, *prev; | |
417 | ||
90036ee5 | 418 | if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) { |
90036ee5 | 419 | |
b20a3503 CL |
420 | err = migrate_prep(); |
421 | if (err) | |
422 | return ERR_PTR(err); | |
90036ee5 | 423 | } |
053837fc | 424 | |
1da177e4 LT |
425 | first = find_vma(mm, start); |
426 | if (!first) | |
427 | return ERR_PTR(-EFAULT); | |
428 | prev = NULL; | |
429 | for (vma = first; vma && vma->vm_start < end; vma = vma->vm_next) { | |
dc9aa5b9 CL |
430 | if (!(flags & MPOL_MF_DISCONTIG_OK)) { |
431 | if (!vma->vm_next && vma->vm_end < end) | |
432 | return ERR_PTR(-EFAULT); | |
433 | if (prev && prev->vm_end < vma->vm_start) | |
434 | return ERR_PTR(-EFAULT); | |
435 | } | |
436 | if (!is_vm_hugetlb_page(vma) && | |
437 | ((flags & MPOL_MF_STRICT) || | |
438 | ((flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) && | |
439 | vma_migratable(vma)))) { | |
5b952b3c | 440 | unsigned long endvma = vma->vm_end; |
dc9aa5b9 | 441 | |
5b952b3c AK |
442 | if (endvma > end) |
443 | endvma = end; | |
444 | if (vma->vm_start > start) | |
445 | start = vma->vm_start; | |
dc9aa5b9 | 446 | err = check_pgd_range(vma, start, endvma, nodes, |
38e35860 | 447 | flags, private); |
1da177e4 LT |
448 | if (err) { |
449 | first = ERR_PTR(err); | |
450 | break; | |
451 | } | |
452 | } | |
453 | prev = vma; | |
454 | } | |
455 | return first; | |
456 | } | |
457 | ||
458 | /* Apply policy to a single VMA */ | |
459 | static int policy_vma(struct vm_area_struct *vma, struct mempolicy *new) | |
460 | { | |
461 | int err = 0; | |
462 | struct mempolicy *old = vma->vm_policy; | |
463 | ||
140d5a49 | 464 | pr_debug("vma %lx-%lx/%lx vm_ops %p vm_file %p set_policy %p\n", |
1da177e4 LT |
465 | vma->vm_start, vma->vm_end, vma->vm_pgoff, |
466 | vma->vm_ops, vma->vm_file, | |
467 | vma->vm_ops ? vma->vm_ops->set_policy : NULL); | |
468 | ||
469 | if (vma->vm_ops && vma->vm_ops->set_policy) | |
470 | err = vma->vm_ops->set_policy(vma, new); | |
471 | if (!err) { | |
472 | mpol_get(new); | |
473 | vma->vm_policy = new; | |
474 | mpol_free(old); | |
475 | } | |
476 | return err; | |
477 | } | |
478 | ||
479 | /* Step 2: apply policy to a range and do splits. */ | |
480 | static int mbind_range(struct vm_area_struct *vma, unsigned long start, | |
481 | unsigned long end, struct mempolicy *new) | |
482 | { | |
483 | struct vm_area_struct *next; | |
484 | int err; | |
485 | ||
486 | err = 0; | |
487 | for (; vma && vma->vm_start < end; vma = next) { | |
488 | next = vma->vm_next; | |
489 | if (vma->vm_start < start) | |
490 | err = split_vma(vma->vm_mm, vma, start, 1); | |
491 | if (!err && vma->vm_end > end) | |
492 | err = split_vma(vma->vm_mm, vma, end, 0); | |
493 | if (!err) | |
494 | err = policy_vma(vma, new); | |
495 | if (err) | |
496 | break; | |
497 | } | |
498 | return err; | |
499 | } | |
500 | ||
c61afb18 PJ |
501 | /* |
502 | * Update task->flags PF_MEMPOLICY bit: set iff non-default | |
503 | * mempolicy. Allows more rapid checking of this (combined perhaps | |
504 | * with other PF_* flag bits) on memory allocation hot code paths. | |
505 | * | |
506 | * If called from outside this file, the task 'p' should -only- be | |
507 | * a newly forked child not yet visible on the task list, because | |
508 | * manipulating the task flags of a visible task is not safe. | |
509 | * | |
510 | * The above limitation is why this routine has the funny name | |
511 | * mpol_fix_fork_child_flag(). | |
512 | * | |
513 | * It is also safe to call this with a task pointer of current, | |
514 | * which the static wrapper mpol_set_task_struct_flag() does, | |
515 | * for use within this file. | |
516 | */ | |
517 | ||
518 | void mpol_fix_fork_child_flag(struct task_struct *p) | |
519 | { | |
520 | if (p->mempolicy) | |
521 | p->flags |= PF_MEMPOLICY; | |
522 | else | |
523 | p->flags &= ~PF_MEMPOLICY; | |
524 | } | |
525 | ||
526 | static void mpol_set_task_struct_flag(void) | |
527 | { | |
528 | mpol_fix_fork_child_flag(current); | |
529 | } | |
530 | ||
1da177e4 | 531 | /* Set the process memory policy */ |
028fec41 DR |
532 | static long do_set_mempolicy(unsigned short mode, unsigned short flags, |
533 | nodemask_t *nodes) | |
1da177e4 | 534 | { |
1da177e4 | 535 | struct mempolicy *new; |
1da177e4 | 536 | |
028fec41 | 537 | new = mpol_new(mode, flags, nodes); |
1da177e4 LT |
538 | if (IS_ERR(new)) |
539 | return PTR_ERR(new); | |
540 | mpol_free(current->mempolicy); | |
541 | current->mempolicy = new; | |
c61afb18 | 542 | mpol_set_task_struct_flag(); |
f5b087b5 DR |
543 | if (new && new->policy == MPOL_INTERLEAVE && |
544 | nodes_weight(new->v.nodes)) | |
dfcd3c0d | 545 | current->il_next = first_node(new->v.nodes); |
1da177e4 LT |
546 | return 0; |
547 | } | |
548 | ||
549 | /* Fill a zone bitmap for a policy */ | |
dfcd3c0d | 550 | static void get_zonemask(struct mempolicy *p, nodemask_t *nodes) |
1da177e4 | 551 | { |
dfcd3c0d | 552 | nodes_clear(*nodes); |
1da177e4 | 553 | switch (p->policy) { |
1da177e4 LT |
554 | case MPOL_DEFAULT: |
555 | break; | |
19770b32 MG |
556 | case MPOL_BIND: |
557 | /* Fall through */ | |
1da177e4 | 558 | case MPOL_INTERLEAVE: |
dfcd3c0d | 559 | *nodes = p->v.nodes; |
1da177e4 LT |
560 | break; |
561 | case MPOL_PREFERRED: | |
56bbd65d | 562 | /* or use current node instead of memory_map? */ |
1da177e4 | 563 | if (p->v.preferred_node < 0) |
56bbd65d | 564 | *nodes = node_states[N_HIGH_MEMORY]; |
1da177e4 | 565 | else |
dfcd3c0d | 566 | node_set(p->v.preferred_node, *nodes); |
1da177e4 LT |
567 | break; |
568 | default: | |
569 | BUG(); | |
570 | } | |
571 | } | |
572 | ||
573 | static int lookup_node(struct mm_struct *mm, unsigned long addr) | |
574 | { | |
575 | struct page *p; | |
576 | int err; | |
577 | ||
578 | err = get_user_pages(current, mm, addr & PAGE_MASK, 1, 0, 0, &p, NULL); | |
579 | if (err >= 0) { | |
580 | err = page_to_nid(p); | |
581 | put_page(p); | |
582 | } | |
583 | return err; | |
584 | } | |
585 | ||
1da177e4 | 586 | /* Retrieve NUMA policy */ |
dbcb0f19 AB |
587 | static long do_get_mempolicy(int *policy, nodemask_t *nmask, |
588 | unsigned long addr, unsigned long flags) | |
1da177e4 | 589 | { |
8bccd85f | 590 | int err; |
1da177e4 LT |
591 | struct mm_struct *mm = current->mm; |
592 | struct vm_area_struct *vma = NULL; | |
593 | struct mempolicy *pol = current->mempolicy; | |
594 | ||
cf2a473c | 595 | cpuset_update_task_memory_state(); |
754af6f5 LS |
596 | if (flags & |
597 | ~(unsigned long)(MPOL_F_NODE|MPOL_F_ADDR|MPOL_F_MEMS_ALLOWED)) | |
1da177e4 | 598 | return -EINVAL; |
754af6f5 LS |
599 | |
600 | if (flags & MPOL_F_MEMS_ALLOWED) { | |
601 | if (flags & (MPOL_F_NODE|MPOL_F_ADDR)) | |
602 | return -EINVAL; | |
603 | *policy = 0; /* just so it's initialized */ | |
604 | *nmask = cpuset_current_mems_allowed; | |
605 | return 0; | |
606 | } | |
607 | ||
1da177e4 LT |
608 | if (flags & MPOL_F_ADDR) { |
609 | down_read(&mm->mmap_sem); | |
610 | vma = find_vma_intersection(mm, addr, addr+1); | |
611 | if (!vma) { | |
612 | up_read(&mm->mmap_sem); | |
613 | return -EFAULT; | |
614 | } | |
615 | if (vma->vm_ops && vma->vm_ops->get_policy) | |
616 | pol = vma->vm_ops->get_policy(vma, addr); | |
617 | else | |
618 | pol = vma->vm_policy; | |
619 | } else if (addr) | |
620 | return -EINVAL; | |
621 | ||
622 | if (!pol) | |
623 | pol = &default_policy; | |
624 | ||
625 | if (flags & MPOL_F_NODE) { | |
626 | if (flags & MPOL_F_ADDR) { | |
627 | err = lookup_node(mm, addr); | |
628 | if (err < 0) | |
629 | goto out; | |
8bccd85f | 630 | *policy = err; |
1da177e4 LT |
631 | } else if (pol == current->mempolicy && |
632 | pol->policy == MPOL_INTERLEAVE) { | |
8bccd85f | 633 | *policy = current->il_next; |
1da177e4 LT |
634 | } else { |
635 | err = -EINVAL; | |
636 | goto out; | |
637 | } | |
638 | } else | |
028fec41 | 639 | *policy = pol->policy | pol->flags; |
1da177e4 LT |
640 | |
641 | if (vma) { | |
642 | up_read(¤t->mm->mmap_sem); | |
643 | vma = NULL; | |
644 | } | |
645 | ||
1da177e4 | 646 | err = 0; |
8bccd85f CL |
647 | if (nmask) |
648 | get_zonemask(pol, nmask); | |
1da177e4 LT |
649 | |
650 | out: | |
651 | if (vma) | |
652 | up_read(¤t->mm->mmap_sem); | |
653 | return err; | |
654 | } | |
655 | ||
b20a3503 | 656 | #ifdef CONFIG_MIGRATION |
6ce3c4c0 CL |
657 | /* |
658 | * page migration | |
659 | */ | |
fc301289 CL |
660 | static void migrate_page_add(struct page *page, struct list_head *pagelist, |
661 | unsigned long flags) | |
6ce3c4c0 CL |
662 | { |
663 | /* | |
fc301289 | 664 | * Avoid migrating a page that is shared with others. |
6ce3c4c0 | 665 | */ |
b20a3503 CL |
666 | if ((flags & MPOL_MF_MOVE_ALL) || page_mapcount(page) == 1) |
667 | isolate_lru_page(page, pagelist); | |
7e2ab150 | 668 | } |
6ce3c4c0 | 669 | |
742755a1 | 670 | static struct page *new_node_page(struct page *page, unsigned long node, int **x) |
95a402c3 | 671 | { |
769848c0 | 672 | return alloc_pages_node(node, GFP_HIGHUSER_MOVABLE, 0); |
95a402c3 CL |
673 | } |
674 | ||
7e2ab150 CL |
675 | /* |
676 | * Migrate pages from one node to a target node. | |
677 | * Returns error or the number of pages not migrated. | |
678 | */ | |
dbcb0f19 AB |
679 | static int migrate_to_node(struct mm_struct *mm, int source, int dest, |
680 | int flags) | |
7e2ab150 CL |
681 | { |
682 | nodemask_t nmask; | |
683 | LIST_HEAD(pagelist); | |
684 | int err = 0; | |
685 | ||
686 | nodes_clear(nmask); | |
687 | node_set(source, nmask); | |
6ce3c4c0 | 688 | |
7e2ab150 CL |
689 | check_range(mm, mm->mmap->vm_start, TASK_SIZE, &nmask, |
690 | flags | MPOL_MF_DISCONTIG_OK, &pagelist); | |
691 | ||
aaa994b3 | 692 | if (!list_empty(&pagelist)) |
95a402c3 CL |
693 | err = migrate_pages(&pagelist, new_node_page, dest); |
694 | ||
7e2ab150 | 695 | return err; |
6ce3c4c0 CL |
696 | } |
697 | ||
39743889 | 698 | /* |
7e2ab150 CL |
699 | * Move pages between the two nodesets so as to preserve the physical |
700 | * layout as much as possible. | |
39743889 CL |
701 | * |
702 | * Returns the number of page that could not be moved. | |
703 | */ | |
704 | int do_migrate_pages(struct mm_struct *mm, | |
705 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
706 | { | |
707 | LIST_HEAD(pagelist); | |
7e2ab150 CL |
708 | int busy = 0; |
709 | int err = 0; | |
710 | nodemask_t tmp; | |
39743889 | 711 | |
7e2ab150 | 712 | down_read(&mm->mmap_sem); |
39743889 | 713 | |
7b2259b3 CL |
714 | err = migrate_vmas(mm, from_nodes, to_nodes, flags); |
715 | if (err) | |
716 | goto out; | |
717 | ||
7e2ab150 CL |
718 | /* |
719 | * Find a 'source' bit set in 'tmp' whose corresponding 'dest' | |
720 | * bit in 'to' is not also set in 'tmp'. Clear the found 'source' | |
721 | * bit in 'tmp', and return that <source, dest> pair for migration. | |
722 | * The pair of nodemasks 'to' and 'from' define the map. | |
723 | * | |
724 | * If no pair of bits is found that way, fallback to picking some | |
725 | * pair of 'source' and 'dest' bits that are not the same. If the | |
726 | * 'source' and 'dest' bits are the same, this represents a node | |
727 | * that will be migrating to itself, so no pages need move. | |
728 | * | |
729 | * If no bits are left in 'tmp', or if all remaining bits left | |
730 | * in 'tmp' correspond to the same bit in 'to', return false | |
731 | * (nothing left to migrate). | |
732 | * | |
733 | * This lets us pick a pair of nodes to migrate between, such that | |
734 | * if possible the dest node is not already occupied by some other | |
735 | * source node, minimizing the risk of overloading the memory on a | |
736 | * node that would happen if we migrated incoming memory to a node | |
737 | * before migrating outgoing memory source that same node. | |
738 | * | |
739 | * A single scan of tmp is sufficient. As we go, we remember the | |
740 | * most recent <s, d> pair that moved (s != d). If we find a pair | |
741 | * that not only moved, but what's better, moved to an empty slot | |
742 | * (d is not set in tmp), then we break out then, with that pair. | |
743 | * Otherwise when we finish scannng from_tmp, we at least have the | |
744 | * most recent <s, d> pair that moved. If we get all the way through | |
745 | * the scan of tmp without finding any node that moved, much less | |
746 | * moved to an empty node, then there is nothing left worth migrating. | |
747 | */ | |
d4984711 | 748 | |
7e2ab150 CL |
749 | tmp = *from_nodes; |
750 | while (!nodes_empty(tmp)) { | |
751 | int s,d; | |
752 | int source = -1; | |
753 | int dest = 0; | |
754 | ||
755 | for_each_node_mask(s, tmp) { | |
756 | d = node_remap(s, *from_nodes, *to_nodes); | |
757 | if (s == d) | |
758 | continue; | |
759 | ||
760 | source = s; /* Node moved. Memorize */ | |
761 | dest = d; | |
762 | ||
763 | /* dest not in remaining from nodes? */ | |
764 | if (!node_isset(dest, tmp)) | |
765 | break; | |
766 | } | |
767 | if (source == -1) | |
768 | break; | |
769 | ||
770 | node_clear(source, tmp); | |
771 | err = migrate_to_node(mm, source, dest, flags); | |
772 | if (err > 0) | |
773 | busy += err; | |
774 | if (err < 0) | |
775 | break; | |
39743889 | 776 | } |
7b2259b3 | 777 | out: |
39743889 | 778 | up_read(&mm->mmap_sem); |
7e2ab150 CL |
779 | if (err < 0) |
780 | return err; | |
781 | return busy; | |
b20a3503 CL |
782 | |
783 | } | |
784 | ||
3ad33b24 LS |
785 | /* |
786 | * Allocate a new page for page migration based on vma policy. | |
787 | * Start assuming that page is mapped by vma pointed to by @private. | |
788 | * Search forward from there, if not. N.B., this assumes that the | |
789 | * list of pages handed to migrate_pages()--which is how we get here-- | |
790 | * is in virtual address order. | |
791 | */ | |
742755a1 | 792 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
793 | { |
794 | struct vm_area_struct *vma = (struct vm_area_struct *)private; | |
3ad33b24 | 795 | unsigned long uninitialized_var(address); |
95a402c3 | 796 | |
3ad33b24 LS |
797 | while (vma) { |
798 | address = page_address_in_vma(page, vma); | |
799 | if (address != -EFAULT) | |
800 | break; | |
801 | vma = vma->vm_next; | |
802 | } | |
803 | ||
804 | /* | |
805 | * if !vma, alloc_page_vma() will use task or system default policy | |
806 | */ | |
807 | return alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); | |
95a402c3 | 808 | } |
b20a3503 CL |
809 | #else |
810 | ||
811 | static void migrate_page_add(struct page *page, struct list_head *pagelist, | |
812 | unsigned long flags) | |
813 | { | |
39743889 CL |
814 | } |
815 | ||
b20a3503 CL |
816 | int do_migrate_pages(struct mm_struct *mm, |
817 | const nodemask_t *from_nodes, const nodemask_t *to_nodes, int flags) | |
818 | { | |
819 | return -ENOSYS; | |
820 | } | |
95a402c3 | 821 | |
69939749 | 822 | static struct page *new_vma_page(struct page *page, unsigned long private, int **x) |
95a402c3 CL |
823 | { |
824 | return NULL; | |
825 | } | |
b20a3503 CL |
826 | #endif |
827 | ||
dbcb0f19 | 828 | static long do_mbind(unsigned long start, unsigned long len, |
028fec41 DR |
829 | unsigned short mode, unsigned short mode_flags, |
830 | nodemask_t *nmask, unsigned long flags) | |
6ce3c4c0 CL |
831 | { |
832 | struct vm_area_struct *vma; | |
833 | struct mm_struct *mm = current->mm; | |
834 | struct mempolicy *new; | |
835 | unsigned long end; | |
836 | int err; | |
837 | LIST_HEAD(pagelist); | |
838 | ||
a3b51e01 DR |
839 | if (flags & ~(unsigned long)(MPOL_MF_STRICT | |
840 | MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) | |
6ce3c4c0 | 841 | return -EINVAL; |
74c00241 | 842 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
6ce3c4c0 CL |
843 | return -EPERM; |
844 | ||
845 | if (start & ~PAGE_MASK) | |
846 | return -EINVAL; | |
847 | ||
848 | if (mode == MPOL_DEFAULT) | |
849 | flags &= ~MPOL_MF_STRICT; | |
850 | ||
851 | len = (len + PAGE_SIZE - 1) & PAGE_MASK; | |
852 | end = start + len; | |
853 | ||
854 | if (end < start) | |
855 | return -EINVAL; | |
856 | if (end == start) | |
857 | return 0; | |
858 | ||
028fec41 | 859 | new = mpol_new(mode, mode_flags, nmask); |
6ce3c4c0 CL |
860 | if (IS_ERR(new)) |
861 | return PTR_ERR(new); | |
862 | ||
863 | /* | |
864 | * If we are using the default policy then operation | |
865 | * on discontinuous address spaces is okay after all | |
866 | */ | |
867 | if (!new) | |
868 | flags |= MPOL_MF_DISCONTIG_OK; | |
869 | ||
028fec41 DR |
870 | pr_debug("mbind %lx-%lx mode:%d flags:%d nodes:%lx\n", |
871 | start, start + len, mode, mode_flags, | |
872 | nmask ? nodes_addr(*nmask)[0] : -1); | |
6ce3c4c0 CL |
873 | |
874 | down_write(&mm->mmap_sem); | |
875 | vma = check_range(mm, start, end, nmask, | |
876 | flags | MPOL_MF_INVERT, &pagelist); | |
877 | ||
878 | err = PTR_ERR(vma); | |
879 | if (!IS_ERR(vma)) { | |
880 | int nr_failed = 0; | |
881 | ||
882 | err = mbind_range(vma, start, end, new); | |
7e2ab150 | 883 | |
6ce3c4c0 | 884 | if (!list_empty(&pagelist)) |
95a402c3 CL |
885 | nr_failed = migrate_pages(&pagelist, new_vma_page, |
886 | (unsigned long)vma); | |
6ce3c4c0 CL |
887 | |
888 | if (!err && nr_failed && (flags & MPOL_MF_STRICT)) | |
889 | err = -EIO; | |
890 | } | |
b20a3503 | 891 | |
6ce3c4c0 CL |
892 | up_write(&mm->mmap_sem); |
893 | mpol_free(new); | |
894 | return err; | |
895 | } | |
896 | ||
8bccd85f CL |
897 | /* |
898 | * User space interface with variable sized bitmaps for nodelists. | |
899 | */ | |
900 | ||
901 | /* Copy a node mask from user space. */ | |
39743889 | 902 | static int get_nodes(nodemask_t *nodes, const unsigned long __user *nmask, |
8bccd85f CL |
903 | unsigned long maxnode) |
904 | { | |
905 | unsigned long k; | |
906 | unsigned long nlongs; | |
907 | unsigned long endmask; | |
908 | ||
909 | --maxnode; | |
910 | nodes_clear(*nodes); | |
911 | if (maxnode == 0 || !nmask) | |
912 | return 0; | |
a9c930ba | 913 | if (maxnode > PAGE_SIZE*BITS_PER_BYTE) |
636f13c1 | 914 | return -EINVAL; |
8bccd85f CL |
915 | |
916 | nlongs = BITS_TO_LONGS(maxnode); | |
917 | if ((maxnode % BITS_PER_LONG) == 0) | |
918 | endmask = ~0UL; | |
919 | else | |
920 | endmask = (1UL << (maxnode % BITS_PER_LONG)) - 1; | |
921 | ||
922 | /* When the user specified more nodes than supported just check | |
923 | if the non supported part is all zero. */ | |
924 | if (nlongs > BITS_TO_LONGS(MAX_NUMNODES)) { | |
925 | if (nlongs > PAGE_SIZE/sizeof(long)) | |
926 | return -EINVAL; | |
927 | for (k = BITS_TO_LONGS(MAX_NUMNODES); k < nlongs; k++) { | |
928 | unsigned long t; | |
929 | if (get_user(t, nmask + k)) | |
930 | return -EFAULT; | |
931 | if (k == nlongs - 1) { | |
932 | if (t & endmask) | |
933 | return -EINVAL; | |
934 | } else if (t) | |
935 | return -EINVAL; | |
936 | } | |
937 | nlongs = BITS_TO_LONGS(MAX_NUMNODES); | |
938 | endmask = ~0UL; | |
939 | } | |
940 | ||
941 | if (copy_from_user(nodes_addr(*nodes), nmask, nlongs*sizeof(unsigned long))) | |
942 | return -EFAULT; | |
943 | nodes_addr(*nodes)[nlongs-1] &= endmask; | |
944 | return 0; | |
945 | } | |
946 | ||
947 | /* Copy a kernel node mask to user space */ | |
948 | static int copy_nodes_to_user(unsigned long __user *mask, unsigned long maxnode, | |
949 | nodemask_t *nodes) | |
950 | { | |
951 | unsigned long copy = ALIGN(maxnode-1, 64) / 8; | |
952 | const int nbytes = BITS_TO_LONGS(MAX_NUMNODES) * sizeof(long); | |
953 | ||
954 | if (copy > nbytes) { | |
955 | if (copy > PAGE_SIZE) | |
956 | return -EINVAL; | |
957 | if (clear_user((char __user *)mask + nbytes, copy - nbytes)) | |
958 | return -EFAULT; | |
959 | copy = nbytes; | |
960 | } | |
961 | return copy_to_user(mask, nodes_addr(*nodes), copy) ? -EFAULT : 0; | |
962 | } | |
963 | ||
964 | asmlinkage long sys_mbind(unsigned long start, unsigned long len, | |
965 | unsigned long mode, | |
966 | unsigned long __user *nmask, unsigned long maxnode, | |
967 | unsigned flags) | |
968 | { | |
969 | nodemask_t nodes; | |
970 | int err; | |
028fec41 | 971 | unsigned short mode_flags; |
8bccd85f | 972 | |
028fec41 DR |
973 | mode_flags = mode & MPOL_MODE_FLAGS; |
974 | mode &= ~MPOL_MODE_FLAGS; | |
a3b51e01 DR |
975 | if (mode >= MPOL_MAX) |
976 | return -EINVAL; | |
4c50bc01 DR |
977 | if ((mode_flags & MPOL_F_STATIC_NODES) && |
978 | (mode_flags & MPOL_F_RELATIVE_NODES)) | |
979 | return -EINVAL; | |
8bccd85f CL |
980 | err = get_nodes(&nodes, nmask, maxnode); |
981 | if (err) | |
982 | return err; | |
028fec41 | 983 | return do_mbind(start, len, mode, mode_flags, &nodes, flags); |
8bccd85f CL |
984 | } |
985 | ||
986 | /* Set the process memory policy */ | |
987 | asmlinkage long sys_set_mempolicy(int mode, unsigned long __user *nmask, | |
988 | unsigned long maxnode) | |
989 | { | |
990 | int err; | |
991 | nodemask_t nodes; | |
028fec41 | 992 | unsigned short flags; |
8bccd85f | 993 | |
028fec41 DR |
994 | flags = mode & MPOL_MODE_FLAGS; |
995 | mode &= ~MPOL_MODE_FLAGS; | |
996 | if ((unsigned int)mode >= MPOL_MAX) | |
8bccd85f | 997 | return -EINVAL; |
4c50bc01 DR |
998 | if ((flags & MPOL_F_STATIC_NODES) && (flags & MPOL_F_RELATIVE_NODES)) |
999 | return -EINVAL; | |
8bccd85f CL |
1000 | err = get_nodes(&nodes, nmask, maxnode); |
1001 | if (err) | |
1002 | return err; | |
028fec41 | 1003 | return do_set_mempolicy(mode, flags, &nodes); |
8bccd85f CL |
1004 | } |
1005 | ||
39743889 CL |
1006 | asmlinkage long sys_migrate_pages(pid_t pid, unsigned long maxnode, |
1007 | const unsigned long __user *old_nodes, | |
1008 | const unsigned long __user *new_nodes) | |
1009 | { | |
1010 | struct mm_struct *mm; | |
1011 | struct task_struct *task; | |
1012 | nodemask_t old; | |
1013 | nodemask_t new; | |
1014 | nodemask_t task_nodes; | |
1015 | int err; | |
1016 | ||
1017 | err = get_nodes(&old, old_nodes, maxnode); | |
1018 | if (err) | |
1019 | return err; | |
1020 | ||
1021 | err = get_nodes(&new, new_nodes, maxnode); | |
1022 | if (err) | |
1023 | return err; | |
1024 | ||
1025 | /* Find the mm_struct */ | |
1026 | read_lock(&tasklist_lock); | |
228ebcbe | 1027 | task = pid ? find_task_by_vpid(pid) : current; |
39743889 CL |
1028 | if (!task) { |
1029 | read_unlock(&tasklist_lock); | |
1030 | return -ESRCH; | |
1031 | } | |
1032 | mm = get_task_mm(task); | |
1033 | read_unlock(&tasklist_lock); | |
1034 | ||
1035 | if (!mm) | |
1036 | return -EINVAL; | |
1037 | ||
1038 | /* | |
1039 | * Check if this process has the right to modify the specified | |
1040 | * process. The right exists if the process has administrative | |
7f927fcc | 1041 | * capabilities, superuser privileges or the same |
39743889 CL |
1042 | * userid as the target process. |
1043 | */ | |
1044 | if ((current->euid != task->suid) && (current->euid != task->uid) && | |
1045 | (current->uid != task->suid) && (current->uid != task->uid) && | |
74c00241 | 1046 | !capable(CAP_SYS_NICE)) { |
39743889 CL |
1047 | err = -EPERM; |
1048 | goto out; | |
1049 | } | |
1050 | ||
1051 | task_nodes = cpuset_mems_allowed(task); | |
1052 | /* Is the user allowed to access the target nodes? */ | |
74c00241 | 1053 | if (!nodes_subset(new, task_nodes) && !capable(CAP_SYS_NICE)) { |
39743889 CL |
1054 | err = -EPERM; |
1055 | goto out; | |
1056 | } | |
1057 | ||
37b07e41 | 1058 | if (!nodes_subset(new, node_states[N_HIGH_MEMORY])) { |
3b42d28b CL |
1059 | err = -EINVAL; |
1060 | goto out; | |
1061 | } | |
1062 | ||
86c3a764 DQ |
1063 | err = security_task_movememory(task); |
1064 | if (err) | |
1065 | goto out; | |
1066 | ||
511030bc | 1067 | err = do_migrate_pages(mm, &old, &new, |
74c00241 | 1068 | capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE); |
39743889 CL |
1069 | out: |
1070 | mmput(mm); | |
1071 | return err; | |
1072 | } | |
1073 | ||
1074 | ||
8bccd85f CL |
1075 | /* Retrieve NUMA policy */ |
1076 | asmlinkage long sys_get_mempolicy(int __user *policy, | |
1077 | unsigned long __user *nmask, | |
1078 | unsigned long maxnode, | |
1079 | unsigned long addr, unsigned long flags) | |
1080 | { | |
dbcb0f19 AB |
1081 | int err; |
1082 | int uninitialized_var(pval); | |
8bccd85f CL |
1083 | nodemask_t nodes; |
1084 | ||
1085 | if (nmask != NULL && maxnode < MAX_NUMNODES) | |
1086 | return -EINVAL; | |
1087 | ||
1088 | err = do_get_mempolicy(&pval, &nodes, addr, flags); | |
1089 | ||
1090 | if (err) | |
1091 | return err; | |
1092 | ||
1093 | if (policy && put_user(pval, policy)) | |
1094 | return -EFAULT; | |
1095 | ||
1096 | if (nmask) | |
1097 | err = copy_nodes_to_user(nmask, maxnode, &nodes); | |
1098 | ||
1099 | return err; | |
1100 | } | |
1101 | ||
1da177e4 LT |
1102 | #ifdef CONFIG_COMPAT |
1103 | ||
1104 | asmlinkage long compat_sys_get_mempolicy(int __user *policy, | |
1105 | compat_ulong_t __user *nmask, | |
1106 | compat_ulong_t maxnode, | |
1107 | compat_ulong_t addr, compat_ulong_t flags) | |
1108 | { | |
1109 | long err; | |
1110 | unsigned long __user *nm = NULL; | |
1111 | unsigned long nr_bits, alloc_size; | |
1112 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1113 | ||
1114 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1115 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1116 | ||
1117 | if (nmask) | |
1118 | nm = compat_alloc_user_space(alloc_size); | |
1119 | ||
1120 | err = sys_get_mempolicy(policy, nm, nr_bits+1, addr, flags); | |
1121 | ||
1122 | if (!err && nmask) { | |
1123 | err = copy_from_user(bm, nm, alloc_size); | |
1124 | /* ensure entire bitmap is zeroed */ | |
1125 | err |= clear_user(nmask, ALIGN(maxnode-1, 8) / 8); | |
1126 | err |= compat_put_bitmap(nmask, bm, nr_bits); | |
1127 | } | |
1128 | ||
1129 | return err; | |
1130 | } | |
1131 | ||
1132 | asmlinkage long compat_sys_set_mempolicy(int mode, compat_ulong_t __user *nmask, | |
1133 | compat_ulong_t maxnode) | |
1134 | { | |
1135 | long err = 0; | |
1136 | unsigned long __user *nm = NULL; | |
1137 | unsigned long nr_bits, alloc_size; | |
1138 | DECLARE_BITMAP(bm, MAX_NUMNODES); | |
1139 | ||
1140 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1141 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1142 | ||
1143 | if (nmask) { | |
1144 | err = compat_get_bitmap(bm, nmask, nr_bits); | |
1145 | nm = compat_alloc_user_space(alloc_size); | |
1146 | err |= copy_to_user(nm, bm, alloc_size); | |
1147 | } | |
1148 | ||
1149 | if (err) | |
1150 | return -EFAULT; | |
1151 | ||
1152 | return sys_set_mempolicy(mode, nm, nr_bits+1); | |
1153 | } | |
1154 | ||
1155 | asmlinkage long compat_sys_mbind(compat_ulong_t start, compat_ulong_t len, | |
1156 | compat_ulong_t mode, compat_ulong_t __user *nmask, | |
1157 | compat_ulong_t maxnode, compat_ulong_t flags) | |
1158 | { | |
1159 | long err = 0; | |
1160 | unsigned long __user *nm = NULL; | |
1161 | unsigned long nr_bits, alloc_size; | |
dfcd3c0d | 1162 | nodemask_t bm; |
1da177e4 LT |
1163 | |
1164 | nr_bits = min_t(unsigned long, maxnode-1, MAX_NUMNODES); | |
1165 | alloc_size = ALIGN(nr_bits, BITS_PER_LONG) / 8; | |
1166 | ||
1167 | if (nmask) { | |
dfcd3c0d | 1168 | err = compat_get_bitmap(nodes_addr(bm), nmask, nr_bits); |
1da177e4 | 1169 | nm = compat_alloc_user_space(alloc_size); |
dfcd3c0d | 1170 | err |= copy_to_user(nm, nodes_addr(bm), alloc_size); |
1da177e4 LT |
1171 | } |
1172 | ||
1173 | if (err) | |
1174 | return -EFAULT; | |
1175 | ||
1176 | return sys_mbind(start, len, mode, nm, nr_bits+1, flags); | |
1177 | } | |
1178 | ||
1179 | #endif | |
1180 | ||
480eccf9 LS |
1181 | /* |
1182 | * get_vma_policy(@task, @vma, @addr) | |
1183 | * @task - task for fallback if vma policy == default | |
1184 | * @vma - virtual memory area whose policy is sought | |
1185 | * @addr - address in @vma for shared policy lookup | |
1186 | * | |
1187 | * Returns effective policy for a VMA at specified address. | |
1188 | * Falls back to @task or system default policy, as necessary. | |
1189 | * Returned policy has extra reference count if shared, vma, | |
1190 | * or some other task's policy [show_numa_maps() can pass | |
1191 | * @task != current]. It is the caller's responsibility to | |
1192 | * free the reference in these cases. | |
1193 | */ | |
48fce342 CL |
1194 | static struct mempolicy * get_vma_policy(struct task_struct *task, |
1195 | struct vm_area_struct *vma, unsigned long addr) | |
1da177e4 | 1196 | { |
6e21c8f1 | 1197 | struct mempolicy *pol = task->mempolicy; |
480eccf9 | 1198 | int shared_pol = 0; |
1da177e4 LT |
1199 | |
1200 | if (vma) { | |
480eccf9 | 1201 | if (vma->vm_ops && vma->vm_ops->get_policy) { |
8bccd85f | 1202 | pol = vma->vm_ops->get_policy(vma, addr); |
480eccf9 LS |
1203 | shared_pol = 1; /* if pol non-NULL, add ref below */ |
1204 | } else if (vma->vm_policy && | |
1da177e4 LT |
1205 | vma->vm_policy->policy != MPOL_DEFAULT) |
1206 | pol = vma->vm_policy; | |
1207 | } | |
1208 | if (!pol) | |
1209 | pol = &default_policy; | |
480eccf9 LS |
1210 | else if (!shared_pol && pol != current->mempolicy) |
1211 | mpol_get(pol); /* vma or other task's policy */ | |
1da177e4 LT |
1212 | return pol; |
1213 | } | |
1214 | ||
19770b32 MG |
1215 | /* Return a nodemask representing a mempolicy */ |
1216 | static nodemask_t *nodemask_policy(gfp_t gfp, struct mempolicy *policy) | |
1217 | { | |
1218 | /* Lower zones don't get a nodemask applied for MPOL_BIND */ | |
1219 | if (unlikely(policy->policy == MPOL_BIND) && | |
1220 | gfp_zone(gfp) >= policy_zone && | |
1221 | cpuset_nodemask_valid_mems_allowed(&policy->v.nodes)) | |
1222 | return &policy->v.nodes; | |
1223 | ||
1224 | return NULL; | |
1225 | } | |
1226 | ||
1da177e4 | 1227 | /* Return a zonelist representing a mempolicy */ |
dd0fc66f | 1228 | static struct zonelist *zonelist_policy(gfp_t gfp, struct mempolicy *policy) |
1da177e4 LT |
1229 | { |
1230 | int nd; | |
1231 | ||
1232 | switch (policy->policy) { | |
1233 | case MPOL_PREFERRED: | |
1234 | nd = policy->v.preferred_node; | |
1235 | if (nd < 0) | |
1236 | nd = numa_node_id(); | |
1237 | break; | |
1238 | case MPOL_BIND: | |
19770b32 MG |
1239 | /* |
1240 | * Normally, MPOL_BIND allocations node-local are node-local | |
1241 | * within the allowed nodemask. However, if __GFP_THISNODE is | |
1242 | * set and the current node is part of the mask, we use the | |
1243 | * the zonelist for the first node in the mask instead. | |
1244 | */ | |
1245 | nd = numa_node_id(); | |
1246 | if (unlikely(gfp & __GFP_THISNODE) && | |
1247 | unlikely(!node_isset(nd, policy->v.nodes))) | |
1248 | nd = first_node(policy->v.nodes); | |
1249 | break; | |
1da177e4 LT |
1250 | case MPOL_INTERLEAVE: /* should not happen */ |
1251 | case MPOL_DEFAULT: | |
1252 | nd = numa_node_id(); | |
1253 | break; | |
1254 | default: | |
1255 | nd = 0; | |
1256 | BUG(); | |
1257 | } | |
0e88460d | 1258 | return node_zonelist(nd, gfp); |
1da177e4 LT |
1259 | } |
1260 | ||
1261 | /* Do dynamic interleaving for a process */ | |
1262 | static unsigned interleave_nodes(struct mempolicy *policy) | |
1263 | { | |
1264 | unsigned nid, next; | |
1265 | struct task_struct *me = current; | |
1266 | ||
1267 | nid = me->il_next; | |
dfcd3c0d | 1268 | next = next_node(nid, policy->v.nodes); |
1da177e4 | 1269 | if (next >= MAX_NUMNODES) |
dfcd3c0d | 1270 | next = first_node(policy->v.nodes); |
f5b087b5 DR |
1271 | if (next < MAX_NUMNODES) |
1272 | me->il_next = next; | |
1da177e4 LT |
1273 | return nid; |
1274 | } | |
1275 | ||
dc85da15 CL |
1276 | /* |
1277 | * Depending on the memory policy provide a node from which to allocate the | |
1278 | * next slab entry. | |
1279 | */ | |
1280 | unsigned slab_node(struct mempolicy *policy) | |
1281 | { | |
a3b51e01 | 1282 | unsigned short pol = policy ? policy->policy : MPOL_DEFAULT; |
765c4507 CL |
1283 | |
1284 | switch (pol) { | |
dc85da15 CL |
1285 | case MPOL_INTERLEAVE: |
1286 | return interleave_nodes(policy); | |
1287 | ||
dd1a239f | 1288 | case MPOL_BIND: { |
dc85da15 CL |
1289 | /* |
1290 | * Follow bind policy behavior and start allocation at the | |
1291 | * first node. | |
1292 | */ | |
19770b32 MG |
1293 | struct zonelist *zonelist; |
1294 | struct zone *zone; | |
1295 | enum zone_type highest_zoneidx = gfp_zone(GFP_KERNEL); | |
1296 | zonelist = &NODE_DATA(numa_node_id())->node_zonelists[0]; | |
1297 | (void)first_zones_zonelist(zonelist, highest_zoneidx, | |
1298 | &policy->v.nodes, | |
1299 | &zone); | |
1300 | return zone->node; | |
dd1a239f | 1301 | } |
dc85da15 CL |
1302 | |
1303 | case MPOL_PREFERRED: | |
1304 | if (policy->v.preferred_node >= 0) | |
1305 | return policy->v.preferred_node; | |
1306 | /* Fall through */ | |
1307 | ||
1308 | default: | |
1309 | return numa_node_id(); | |
1310 | } | |
1311 | } | |
1312 | ||
1da177e4 LT |
1313 | /* Do static interleaving for a VMA with known offset. */ |
1314 | static unsigned offset_il_node(struct mempolicy *pol, | |
1315 | struct vm_area_struct *vma, unsigned long off) | |
1316 | { | |
dfcd3c0d | 1317 | unsigned nnodes = nodes_weight(pol->v.nodes); |
f5b087b5 | 1318 | unsigned target; |
1da177e4 LT |
1319 | int c; |
1320 | int nid = -1; | |
1321 | ||
f5b087b5 DR |
1322 | if (!nnodes) |
1323 | return numa_node_id(); | |
1324 | target = (unsigned int)off % nnodes; | |
1da177e4 LT |
1325 | c = 0; |
1326 | do { | |
dfcd3c0d | 1327 | nid = next_node(nid, pol->v.nodes); |
1da177e4 LT |
1328 | c++; |
1329 | } while (c <= target); | |
1da177e4 LT |
1330 | return nid; |
1331 | } | |
1332 | ||
5da7ca86 CL |
1333 | /* Determine a node number for interleave */ |
1334 | static inline unsigned interleave_nid(struct mempolicy *pol, | |
1335 | struct vm_area_struct *vma, unsigned long addr, int shift) | |
1336 | { | |
1337 | if (vma) { | |
1338 | unsigned long off; | |
1339 | ||
3b98b087 NA |
1340 | /* |
1341 | * for small pages, there is no difference between | |
1342 | * shift and PAGE_SHIFT, so the bit-shift is safe. | |
1343 | * for huge pages, since vm_pgoff is in units of small | |
1344 | * pages, we need to shift off the always 0 bits to get | |
1345 | * a useful offset. | |
1346 | */ | |
1347 | BUG_ON(shift < PAGE_SHIFT); | |
1348 | off = vma->vm_pgoff >> (shift - PAGE_SHIFT); | |
5da7ca86 CL |
1349 | off += (addr - vma->vm_start) >> shift; |
1350 | return offset_il_node(pol, vma, off); | |
1351 | } else | |
1352 | return interleave_nodes(pol); | |
1353 | } | |
1354 | ||
00ac59ad | 1355 | #ifdef CONFIG_HUGETLBFS |
480eccf9 LS |
1356 | /* |
1357 | * huge_zonelist(@vma, @addr, @gfp_flags, @mpol) | |
1358 | * @vma = virtual memory area whose policy is sought | |
1359 | * @addr = address in @vma for shared policy lookup and interleave policy | |
1360 | * @gfp_flags = for requested zone | |
19770b32 MG |
1361 | * @mpol = pointer to mempolicy pointer for reference counted mempolicy |
1362 | * @nodemask = pointer to nodemask pointer for MPOL_BIND nodemask | |
480eccf9 LS |
1363 | * |
1364 | * Returns a zonelist suitable for a huge page allocation. | |
19770b32 MG |
1365 | * If the effective policy is 'BIND, returns pointer to local node's zonelist, |
1366 | * and a pointer to the mempolicy's @nodemask for filtering the zonelist. | |
480eccf9 | 1367 | * If it is also a policy for which get_vma_policy() returns an extra |
19770b32 | 1368 | * reference, we must hold that reference until after the allocation. |
480eccf9 | 1369 | * In that case, return policy via @mpol so hugetlb allocation can drop |
19770b32 | 1370 | * the reference. For non-'BIND referenced policies, we can/do drop the |
480eccf9 LS |
1371 | * reference here, so the caller doesn't need to know about the special case |
1372 | * for default and current task policy. | |
1373 | */ | |
396faf03 | 1374 | struct zonelist *huge_zonelist(struct vm_area_struct *vma, unsigned long addr, |
19770b32 MG |
1375 | gfp_t gfp_flags, struct mempolicy **mpol, |
1376 | nodemask_t **nodemask) | |
5da7ca86 CL |
1377 | { |
1378 | struct mempolicy *pol = get_vma_policy(current, vma, addr); | |
480eccf9 | 1379 | struct zonelist *zl; |
5da7ca86 | 1380 | |
480eccf9 | 1381 | *mpol = NULL; /* probably no unref needed */ |
19770b32 MG |
1382 | *nodemask = NULL; /* assume !MPOL_BIND */ |
1383 | if (pol->policy == MPOL_BIND) { | |
1384 | *nodemask = &pol->v.nodes; | |
1385 | } else if (pol->policy == MPOL_INTERLEAVE) { | |
5da7ca86 CL |
1386 | unsigned nid; |
1387 | ||
1388 | nid = interleave_nid(pol, vma, addr, HPAGE_SHIFT); | |
69682d85 LS |
1389 | if (unlikely(pol != &default_policy && |
1390 | pol != current->mempolicy)) | |
1391 | __mpol_free(pol); /* finished with pol */ | |
0e88460d | 1392 | return node_zonelist(nid, gfp_flags); |
5da7ca86 | 1393 | } |
480eccf9 LS |
1394 | |
1395 | zl = zonelist_policy(GFP_HIGHUSER, pol); | |
1396 | if (unlikely(pol != &default_policy && pol != current->mempolicy)) { | |
1397 | if (pol->policy != MPOL_BIND) | |
1398 | __mpol_free(pol); /* finished with pol */ | |
1399 | else | |
1400 | *mpol = pol; /* unref needed after allocation */ | |
1401 | } | |
1402 | return zl; | |
5da7ca86 | 1403 | } |
00ac59ad | 1404 | #endif |
5da7ca86 | 1405 | |
1da177e4 LT |
1406 | /* Allocate a page in interleaved policy. |
1407 | Own path because it needs to do special accounting. */ | |
662f3a0b AK |
1408 | static struct page *alloc_page_interleave(gfp_t gfp, unsigned order, |
1409 | unsigned nid) | |
1da177e4 LT |
1410 | { |
1411 | struct zonelist *zl; | |
1412 | struct page *page; | |
1413 | ||
0e88460d | 1414 | zl = node_zonelist(nid, gfp); |
1da177e4 | 1415 | page = __alloc_pages(gfp, order, zl); |
dd1a239f | 1416 | if (page && page_zone(page) == zonelist_zone(&zl->_zonerefs[0])) |
ca889e6c | 1417 | inc_zone_page_state(page, NUMA_INTERLEAVE_HIT); |
1da177e4 LT |
1418 | return page; |
1419 | } | |
1420 | ||
1421 | /** | |
1422 | * alloc_page_vma - Allocate a page for a VMA. | |
1423 | * | |
1424 | * @gfp: | |
1425 | * %GFP_USER user allocation. | |
1426 | * %GFP_KERNEL kernel allocations, | |
1427 | * %GFP_HIGHMEM highmem/user allocations, | |
1428 | * %GFP_FS allocation should not call back into a file system. | |
1429 | * %GFP_ATOMIC don't sleep. | |
1430 | * | |
1431 | * @vma: Pointer to VMA or NULL if not available. | |
1432 | * @addr: Virtual Address of the allocation. Must be inside the VMA. | |
1433 | * | |
1434 | * This function allocates a page from the kernel page pool and applies | |
1435 | * a NUMA policy associated with the VMA or the current process. | |
1436 | * When VMA is not NULL caller must hold down_read on the mmap_sem of the | |
1437 | * mm_struct of the VMA to prevent it from going away. Should be used for | |
1438 | * all allocations for pages that will be mapped into | |
1439 | * user space. Returns NULL when no page can be allocated. | |
1440 | * | |
1441 | * Should be called with the mm_sem of the vma hold. | |
1442 | */ | |
1443 | struct page * | |
dd0fc66f | 1444 | alloc_page_vma(gfp_t gfp, struct vm_area_struct *vma, unsigned long addr) |
1da177e4 | 1445 | { |
6e21c8f1 | 1446 | struct mempolicy *pol = get_vma_policy(current, vma, addr); |
480eccf9 | 1447 | struct zonelist *zl; |
1da177e4 | 1448 | |
cf2a473c | 1449 | cpuset_update_task_memory_state(); |
1da177e4 LT |
1450 | |
1451 | if (unlikely(pol->policy == MPOL_INTERLEAVE)) { | |
1452 | unsigned nid; | |
5da7ca86 CL |
1453 | |
1454 | nid = interleave_nid(pol, vma, addr, PAGE_SHIFT); | |
69682d85 LS |
1455 | if (unlikely(pol != &default_policy && |
1456 | pol != current->mempolicy)) | |
1457 | __mpol_free(pol); /* finished with pol */ | |
1da177e4 LT |
1458 | return alloc_page_interleave(gfp, 0, nid); |
1459 | } | |
480eccf9 LS |
1460 | zl = zonelist_policy(gfp, pol); |
1461 | if (pol != &default_policy && pol != current->mempolicy) { | |
1462 | /* | |
1463 | * slow path: ref counted policy -- shared or vma | |
1464 | */ | |
19770b32 MG |
1465 | struct page *page = __alloc_pages_nodemask(gfp, 0, |
1466 | zl, nodemask_policy(gfp, pol)); | |
480eccf9 LS |
1467 | __mpol_free(pol); |
1468 | return page; | |
1469 | } | |
1470 | /* | |
1471 | * fast path: default or task policy | |
1472 | */ | |
19770b32 | 1473 | return __alloc_pages_nodemask(gfp, 0, zl, nodemask_policy(gfp, pol)); |
1da177e4 LT |
1474 | } |
1475 | ||
1476 | /** | |
1477 | * alloc_pages_current - Allocate pages. | |
1478 | * | |
1479 | * @gfp: | |
1480 | * %GFP_USER user allocation, | |
1481 | * %GFP_KERNEL kernel allocation, | |
1482 | * %GFP_HIGHMEM highmem allocation, | |
1483 | * %GFP_FS don't call back into a file system. | |
1484 | * %GFP_ATOMIC don't sleep. | |
1485 | * @order: Power of two of allocation size in pages. 0 is a single page. | |
1486 | * | |
1487 | * Allocate a page from the kernel page pool. When not in | |
1488 | * interrupt context and apply the current process NUMA policy. | |
1489 | * Returns NULL when no page can be allocated. | |
1490 | * | |
cf2a473c | 1491 | * Don't call cpuset_update_task_memory_state() unless |
1da177e4 LT |
1492 | * 1) it's ok to take cpuset_sem (can WAIT), and |
1493 | * 2) allocating for current task (not interrupt). | |
1494 | */ | |
dd0fc66f | 1495 | struct page *alloc_pages_current(gfp_t gfp, unsigned order) |
1da177e4 LT |
1496 | { |
1497 | struct mempolicy *pol = current->mempolicy; | |
1498 | ||
1499 | if ((gfp & __GFP_WAIT) && !in_interrupt()) | |
cf2a473c | 1500 | cpuset_update_task_memory_state(); |
9b819d20 | 1501 | if (!pol || in_interrupt() || (gfp & __GFP_THISNODE)) |
1da177e4 LT |
1502 | pol = &default_policy; |
1503 | if (pol->policy == MPOL_INTERLEAVE) | |
1504 | return alloc_page_interleave(gfp, order, interleave_nodes(pol)); | |
19770b32 MG |
1505 | return __alloc_pages_nodemask(gfp, order, |
1506 | zonelist_policy(gfp, pol), nodemask_policy(gfp, pol)); | |
1da177e4 LT |
1507 | } |
1508 | EXPORT_SYMBOL(alloc_pages_current); | |
1509 | ||
4225399a PJ |
1510 | /* |
1511 | * If mpol_copy() sees current->cpuset == cpuset_being_rebound, then it | |
1512 | * rebinds the mempolicy its copying by calling mpol_rebind_policy() | |
1513 | * with the mems_allowed returned by cpuset_mems_allowed(). This | |
1514 | * keeps mempolicies cpuset relative after its cpuset moves. See | |
1515 | * further kernel/cpuset.c update_nodemask(). | |
1516 | */ | |
4225399a | 1517 | |
1da177e4 LT |
1518 | /* Slow path of a mempolicy copy */ |
1519 | struct mempolicy *__mpol_copy(struct mempolicy *old) | |
1520 | { | |
1521 | struct mempolicy *new = kmem_cache_alloc(policy_cache, GFP_KERNEL); | |
1522 | ||
1523 | if (!new) | |
1524 | return ERR_PTR(-ENOMEM); | |
4225399a PJ |
1525 | if (current_cpuset_is_being_rebound()) { |
1526 | nodemask_t mems = cpuset_mems_allowed(current); | |
1527 | mpol_rebind_policy(old, &mems); | |
1528 | } | |
1da177e4 LT |
1529 | *new = *old; |
1530 | atomic_set(&new->refcnt, 1); | |
1da177e4 LT |
1531 | return new; |
1532 | } | |
1533 | ||
f5b087b5 DR |
1534 | static int mpol_match_intent(const struct mempolicy *a, |
1535 | const struct mempolicy *b) | |
1536 | { | |
1537 | if (a->flags != b->flags) | |
1538 | return 0; | |
1539 | if (!mpol_store_user_nodemask(a)) | |
1540 | return 1; | |
1541 | return nodes_equal(a->w.user_nodemask, b->w.user_nodemask); | |
1542 | } | |
1543 | ||
1da177e4 LT |
1544 | /* Slow path of a mempolicy comparison */ |
1545 | int __mpol_equal(struct mempolicy *a, struct mempolicy *b) | |
1546 | { | |
1547 | if (!a || !b) | |
1548 | return 0; | |
1549 | if (a->policy != b->policy) | |
1550 | return 0; | |
f5b087b5 DR |
1551 | if (a->policy != MPOL_DEFAULT && !mpol_match_intent(a, b)) |
1552 | return 0; | |
1da177e4 LT |
1553 | switch (a->policy) { |
1554 | case MPOL_DEFAULT: | |
1555 | return 1; | |
19770b32 MG |
1556 | case MPOL_BIND: |
1557 | /* Fall through */ | |
1da177e4 | 1558 | case MPOL_INTERLEAVE: |
dfcd3c0d | 1559 | return nodes_equal(a->v.nodes, b->v.nodes); |
1da177e4 LT |
1560 | case MPOL_PREFERRED: |
1561 | return a->v.preferred_node == b->v.preferred_node; | |
1da177e4 LT |
1562 | default: |
1563 | BUG(); | |
1564 | return 0; | |
1565 | } | |
1566 | } | |
1567 | ||
1568 | /* Slow path of a mpol destructor. */ | |
1569 | void __mpol_free(struct mempolicy *p) | |
1570 | { | |
1571 | if (!atomic_dec_and_test(&p->refcnt)) | |
1572 | return; | |
1da177e4 LT |
1573 | p->policy = MPOL_DEFAULT; |
1574 | kmem_cache_free(policy_cache, p); | |
1575 | } | |
1576 | ||
1da177e4 LT |
1577 | /* |
1578 | * Shared memory backing store policy support. | |
1579 | * | |
1580 | * Remember policies even when nobody has shared memory mapped. | |
1581 | * The policies are kept in Red-Black tree linked from the inode. | |
1582 | * They are protected by the sp->lock spinlock, which should be held | |
1583 | * for any accesses to the tree. | |
1584 | */ | |
1585 | ||
1586 | /* lookup first element intersecting start-end */ | |
1587 | /* Caller holds sp->lock */ | |
1588 | static struct sp_node * | |
1589 | sp_lookup(struct shared_policy *sp, unsigned long start, unsigned long end) | |
1590 | { | |
1591 | struct rb_node *n = sp->root.rb_node; | |
1592 | ||
1593 | while (n) { | |
1594 | struct sp_node *p = rb_entry(n, struct sp_node, nd); | |
1595 | ||
1596 | if (start >= p->end) | |
1597 | n = n->rb_right; | |
1598 | else if (end <= p->start) | |
1599 | n = n->rb_left; | |
1600 | else | |
1601 | break; | |
1602 | } | |
1603 | if (!n) | |
1604 | return NULL; | |
1605 | for (;;) { | |
1606 | struct sp_node *w = NULL; | |
1607 | struct rb_node *prev = rb_prev(n); | |
1608 | if (!prev) | |
1609 | break; | |
1610 | w = rb_entry(prev, struct sp_node, nd); | |
1611 | if (w->end <= start) | |
1612 | break; | |
1613 | n = prev; | |
1614 | } | |
1615 | return rb_entry(n, struct sp_node, nd); | |
1616 | } | |
1617 | ||
1618 | /* Insert a new shared policy into the list. */ | |
1619 | /* Caller holds sp->lock */ | |
1620 | static void sp_insert(struct shared_policy *sp, struct sp_node *new) | |
1621 | { | |
1622 | struct rb_node **p = &sp->root.rb_node; | |
1623 | struct rb_node *parent = NULL; | |
1624 | struct sp_node *nd; | |
1625 | ||
1626 | while (*p) { | |
1627 | parent = *p; | |
1628 | nd = rb_entry(parent, struct sp_node, nd); | |
1629 | if (new->start < nd->start) | |
1630 | p = &(*p)->rb_left; | |
1631 | else if (new->end > nd->end) | |
1632 | p = &(*p)->rb_right; | |
1633 | else | |
1634 | BUG(); | |
1635 | } | |
1636 | rb_link_node(&new->nd, parent, p); | |
1637 | rb_insert_color(&new->nd, &sp->root); | |
140d5a49 | 1638 | pr_debug("inserting %lx-%lx: %d\n", new->start, new->end, |
1da177e4 LT |
1639 | new->policy ? new->policy->policy : 0); |
1640 | } | |
1641 | ||
1642 | /* Find shared policy intersecting idx */ | |
1643 | struct mempolicy * | |
1644 | mpol_shared_policy_lookup(struct shared_policy *sp, unsigned long idx) | |
1645 | { | |
1646 | struct mempolicy *pol = NULL; | |
1647 | struct sp_node *sn; | |
1648 | ||
1649 | if (!sp->root.rb_node) | |
1650 | return NULL; | |
1651 | spin_lock(&sp->lock); | |
1652 | sn = sp_lookup(sp, idx, idx+1); | |
1653 | if (sn) { | |
1654 | mpol_get(sn->policy); | |
1655 | pol = sn->policy; | |
1656 | } | |
1657 | spin_unlock(&sp->lock); | |
1658 | return pol; | |
1659 | } | |
1660 | ||
1661 | static void sp_delete(struct shared_policy *sp, struct sp_node *n) | |
1662 | { | |
140d5a49 | 1663 | pr_debug("deleting %lx-l%lx\n", n->start, n->end); |
1da177e4 LT |
1664 | rb_erase(&n->nd, &sp->root); |
1665 | mpol_free(n->policy); | |
1666 | kmem_cache_free(sn_cache, n); | |
1667 | } | |
1668 | ||
dbcb0f19 AB |
1669 | static struct sp_node *sp_alloc(unsigned long start, unsigned long end, |
1670 | struct mempolicy *pol) | |
1da177e4 LT |
1671 | { |
1672 | struct sp_node *n = kmem_cache_alloc(sn_cache, GFP_KERNEL); | |
1673 | ||
1674 | if (!n) | |
1675 | return NULL; | |
1676 | n->start = start; | |
1677 | n->end = end; | |
1678 | mpol_get(pol); | |
1679 | n->policy = pol; | |
1680 | return n; | |
1681 | } | |
1682 | ||
1683 | /* Replace a policy range. */ | |
1684 | static int shared_policy_replace(struct shared_policy *sp, unsigned long start, | |
1685 | unsigned long end, struct sp_node *new) | |
1686 | { | |
1687 | struct sp_node *n, *new2 = NULL; | |
1688 | ||
1689 | restart: | |
1690 | spin_lock(&sp->lock); | |
1691 | n = sp_lookup(sp, start, end); | |
1692 | /* Take care of old policies in the same range. */ | |
1693 | while (n && n->start < end) { | |
1694 | struct rb_node *next = rb_next(&n->nd); | |
1695 | if (n->start >= start) { | |
1696 | if (n->end <= end) | |
1697 | sp_delete(sp, n); | |
1698 | else | |
1699 | n->start = end; | |
1700 | } else { | |
1701 | /* Old policy spanning whole new range. */ | |
1702 | if (n->end > end) { | |
1703 | if (!new2) { | |
1704 | spin_unlock(&sp->lock); | |
1705 | new2 = sp_alloc(end, n->end, n->policy); | |
1706 | if (!new2) | |
1707 | return -ENOMEM; | |
1708 | goto restart; | |
1709 | } | |
1710 | n->end = start; | |
1711 | sp_insert(sp, new2); | |
1712 | new2 = NULL; | |
1713 | break; | |
1714 | } else | |
1715 | n->end = start; | |
1716 | } | |
1717 | if (!next) | |
1718 | break; | |
1719 | n = rb_entry(next, struct sp_node, nd); | |
1720 | } | |
1721 | if (new) | |
1722 | sp_insert(sp, new); | |
1723 | spin_unlock(&sp->lock); | |
1724 | if (new2) { | |
1725 | mpol_free(new2->policy); | |
1726 | kmem_cache_free(sn_cache, new2); | |
1727 | } | |
1728 | return 0; | |
1729 | } | |
1730 | ||
a3b51e01 | 1731 | void mpol_shared_policy_init(struct shared_policy *info, unsigned short policy, |
028fec41 | 1732 | unsigned short flags, nodemask_t *policy_nodes) |
7339ff83 RH |
1733 | { |
1734 | info->root = RB_ROOT; | |
1735 | spin_lock_init(&info->lock); | |
1736 | ||
1737 | if (policy != MPOL_DEFAULT) { | |
1738 | struct mempolicy *newpol; | |
1739 | ||
1740 | /* Falls back to MPOL_DEFAULT on any error */ | |
028fec41 | 1741 | newpol = mpol_new(policy, flags, policy_nodes); |
7339ff83 RH |
1742 | if (!IS_ERR(newpol)) { |
1743 | /* Create pseudo-vma that contains just the policy */ | |
1744 | struct vm_area_struct pvma; | |
1745 | ||
1746 | memset(&pvma, 0, sizeof(struct vm_area_struct)); | |
1747 | /* Policy covers entire file */ | |
1748 | pvma.vm_end = TASK_SIZE; | |
1749 | mpol_set_shared_policy(info, &pvma, newpol); | |
1750 | mpol_free(newpol); | |
1751 | } | |
1752 | } | |
1753 | } | |
1754 | ||
1da177e4 LT |
1755 | int mpol_set_shared_policy(struct shared_policy *info, |
1756 | struct vm_area_struct *vma, struct mempolicy *npol) | |
1757 | { | |
1758 | int err; | |
1759 | struct sp_node *new = NULL; | |
1760 | unsigned long sz = vma_pages(vma); | |
1761 | ||
028fec41 | 1762 | pr_debug("set_shared_policy %lx sz %lu %d %d %lx\n", |
1da177e4 | 1763 | vma->vm_pgoff, |
028fec41 DR |
1764 | sz, npol ? npol->policy : -1, |
1765 | npol ? npol->flags : -1, | |
140d5a49 | 1766 | npol ? nodes_addr(npol->v.nodes)[0] : -1); |
1da177e4 LT |
1767 | |
1768 | if (npol) { | |
1769 | new = sp_alloc(vma->vm_pgoff, vma->vm_pgoff + sz, npol); | |
1770 | if (!new) | |
1771 | return -ENOMEM; | |
1772 | } | |
1773 | err = shared_policy_replace(info, vma->vm_pgoff, vma->vm_pgoff+sz, new); | |
1774 | if (err && new) | |
1775 | kmem_cache_free(sn_cache, new); | |
1776 | return err; | |
1777 | } | |
1778 | ||
1779 | /* Free a backing policy store on inode delete. */ | |
1780 | void mpol_free_shared_policy(struct shared_policy *p) | |
1781 | { | |
1782 | struct sp_node *n; | |
1783 | struct rb_node *next; | |
1784 | ||
1785 | if (!p->root.rb_node) | |
1786 | return; | |
1787 | spin_lock(&p->lock); | |
1788 | next = rb_first(&p->root); | |
1789 | while (next) { | |
1790 | n = rb_entry(next, struct sp_node, nd); | |
1791 | next = rb_next(&n->nd); | |
90c5029e | 1792 | rb_erase(&n->nd, &p->root); |
1da177e4 LT |
1793 | mpol_free(n->policy); |
1794 | kmem_cache_free(sn_cache, n); | |
1795 | } | |
1796 | spin_unlock(&p->lock); | |
1da177e4 LT |
1797 | } |
1798 | ||
1799 | /* assumes fs == KERNEL_DS */ | |
1800 | void __init numa_policy_init(void) | |
1801 | { | |
b71636e2 PM |
1802 | nodemask_t interleave_nodes; |
1803 | unsigned long largest = 0; | |
1804 | int nid, prefer = 0; | |
1805 | ||
1da177e4 LT |
1806 | policy_cache = kmem_cache_create("numa_policy", |
1807 | sizeof(struct mempolicy), | |
20c2df83 | 1808 | 0, SLAB_PANIC, NULL); |
1da177e4 LT |
1809 | |
1810 | sn_cache = kmem_cache_create("shared_policy_node", | |
1811 | sizeof(struct sp_node), | |
20c2df83 | 1812 | 0, SLAB_PANIC, NULL); |
1da177e4 | 1813 | |
b71636e2 PM |
1814 | /* |
1815 | * Set interleaving policy for system init. Interleaving is only | |
1816 | * enabled across suitably sized nodes (default is >= 16MB), or | |
1817 | * fall back to the largest node if they're all smaller. | |
1818 | */ | |
1819 | nodes_clear(interleave_nodes); | |
56bbd65d | 1820 | for_each_node_state(nid, N_HIGH_MEMORY) { |
b71636e2 PM |
1821 | unsigned long total_pages = node_present_pages(nid); |
1822 | ||
1823 | /* Preserve the largest node */ | |
1824 | if (largest < total_pages) { | |
1825 | largest = total_pages; | |
1826 | prefer = nid; | |
1827 | } | |
1828 | ||
1829 | /* Interleave this node? */ | |
1830 | if ((total_pages << PAGE_SHIFT) >= (16 << 20)) | |
1831 | node_set(nid, interleave_nodes); | |
1832 | } | |
1833 | ||
1834 | /* All too small, use the largest */ | |
1835 | if (unlikely(nodes_empty(interleave_nodes))) | |
1836 | node_set(prefer, interleave_nodes); | |
1da177e4 | 1837 | |
028fec41 | 1838 | if (do_set_mempolicy(MPOL_INTERLEAVE, 0, &interleave_nodes)) |
1da177e4 LT |
1839 | printk("numa_policy_init: interleaving failed\n"); |
1840 | } | |
1841 | ||
8bccd85f | 1842 | /* Reset policy of current process to default */ |
1da177e4 LT |
1843 | void numa_default_policy(void) |
1844 | { | |
028fec41 | 1845 | do_set_mempolicy(MPOL_DEFAULT, 0, NULL); |
1da177e4 | 1846 | } |
68860ec1 | 1847 | |
1a75a6c8 CL |
1848 | /* |
1849 | * Display pages allocated per node and memory policy via /proc. | |
1850 | */ | |
1851 | ||
15ad7cdc HD |
1852 | static const char * const policy_types[] = |
1853 | { "default", "prefer", "bind", "interleave" }; | |
1a75a6c8 CL |
1854 | |
1855 | /* | |
1856 | * Convert a mempolicy into a string. | |
1857 | * Returns the number of characters in buffer (if positive) | |
1858 | * or an error (negative) | |
1859 | */ | |
1860 | static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol) | |
1861 | { | |
1862 | char *p = buffer; | |
1863 | int l; | |
1864 | nodemask_t nodes; | |
a3b51e01 | 1865 | unsigned short mode = pol ? pol->policy : MPOL_DEFAULT; |
f5b087b5 | 1866 | unsigned short flags = pol ? pol->flags : 0; |
1a75a6c8 CL |
1867 | |
1868 | switch (mode) { | |
1869 | case MPOL_DEFAULT: | |
1870 | nodes_clear(nodes); | |
1871 | break; | |
1872 | ||
1873 | case MPOL_PREFERRED: | |
1874 | nodes_clear(nodes); | |
1875 | node_set(pol->v.preferred_node, nodes); | |
1876 | break; | |
1877 | ||
1878 | case MPOL_BIND: | |
19770b32 | 1879 | /* Fall through */ |
1a75a6c8 CL |
1880 | case MPOL_INTERLEAVE: |
1881 | nodes = pol->v.nodes; | |
1882 | break; | |
1883 | ||
1884 | default: | |
1885 | BUG(); | |
1886 | return -EFAULT; | |
1887 | } | |
1888 | ||
1889 | l = strlen(policy_types[mode]); | |
1890 | if (buffer + maxlen < p + l + 1) | |
1891 | return -ENOSPC; | |
1892 | ||
1893 | strcpy(p, policy_types[mode]); | |
1894 | p += l; | |
1895 | ||
f5b087b5 DR |
1896 | if (flags) { |
1897 | int need_bar = 0; | |
1898 | ||
1899 | if (buffer + maxlen < p + 2) | |
1900 | return -ENOSPC; | |
1901 | *p++ = '='; | |
1902 | ||
1903 | if (flags & MPOL_F_STATIC_NODES) | |
1904 | p += sprintf(p, "%sstatic", need_bar++ ? "|" : ""); | |
4c50bc01 DR |
1905 | if (flags & MPOL_F_RELATIVE_NODES) |
1906 | p += sprintf(p, "%srelative", need_bar++ ? "|" : ""); | |
f5b087b5 DR |
1907 | } |
1908 | ||
1a75a6c8 CL |
1909 | if (!nodes_empty(nodes)) { |
1910 | if (buffer + maxlen < p + 2) | |
1911 | return -ENOSPC; | |
1912 | *p++ = '='; | |
1913 | p += nodelist_scnprintf(p, buffer + maxlen - p, nodes); | |
1914 | } | |
1915 | return p - buffer; | |
1916 | } | |
1917 | ||
1918 | struct numa_maps { | |
1919 | unsigned long pages; | |
1920 | unsigned long anon; | |
397874df CL |
1921 | unsigned long active; |
1922 | unsigned long writeback; | |
1a75a6c8 | 1923 | unsigned long mapcount_max; |
397874df CL |
1924 | unsigned long dirty; |
1925 | unsigned long swapcache; | |
1a75a6c8 CL |
1926 | unsigned long node[MAX_NUMNODES]; |
1927 | }; | |
1928 | ||
397874df | 1929 | static void gather_stats(struct page *page, void *private, int pte_dirty) |
1a75a6c8 CL |
1930 | { |
1931 | struct numa_maps *md = private; | |
1932 | int count = page_mapcount(page); | |
1933 | ||
397874df CL |
1934 | md->pages++; |
1935 | if (pte_dirty || PageDirty(page)) | |
1936 | md->dirty++; | |
1a75a6c8 | 1937 | |
397874df CL |
1938 | if (PageSwapCache(page)) |
1939 | md->swapcache++; | |
1a75a6c8 | 1940 | |
397874df CL |
1941 | if (PageActive(page)) |
1942 | md->active++; | |
1943 | ||
1944 | if (PageWriteback(page)) | |
1945 | md->writeback++; | |
1a75a6c8 CL |
1946 | |
1947 | if (PageAnon(page)) | |
1948 | md->anon++; | |
1949 | ||
397874df CL |
1950 | if (count > md->mapcount_max) |
1951 | md->mapcount_max = count; | |
1952 | ||
1a75a6c8 | 1953 | md->node[page_to_nid(page)]++; |
1a75a6c8 CL |
1954 | } |
1955 | ||
7f709ed0 | 1956 | #ifdef CONFIG_HUGETLB_PAGE |
397874df CL |
1957 | static void check_huge_range(struct vm_area_struct *vma, |
1958 | unsigned long start, unsigned long end, | |
1959 | struct numa_maps *md) | |
1960 | { | |
1961 | unsigned long addr; | |
1962 | struct page *page; | |
1963 | ||
1964 | for (addr = start; addr < end; addr += HPAGE_SIZE) { | |
1965 | pte_t *ptep = huge_pte_offset(vma->vm_mm, addr & HPAGE_MASK); | |
1966 | pte_t pte; | |
1967 | ||
1968 | if (!ptep) | |
1969 | continue; | |
1970 | ||
1971 | pte = *ptep; | |
1972 | if (pte_none(pte)) | |
1973 | continue; | |
1974 | ||
1975 | page = pte_page(pte); | |
1976 | if (!page) | |
1977 | continue; | |
1978 | ||
1979 | gather_stats(page, md, pte_dirty(*ptep)); | |
1980 | } | |
1981 | } | |
7f709ed0 AM |
1982 | #else |
1983 | static inline void check_huge_range(struct vm_area_struct *vma, | |
1984 | unsigned long start, unsigned long end, | |
1985 | struct numa_maps *md) | |
1986 | { | |
1987 | } | |
1988 | #endif | |
397874df | 1989 | |
1a75a6c8 CL |
1990 | int show_numa_map(struct seq_file *m, void *v) |
1991 | { | |
99f89551 | 1992 | struct proc_maps_private *priv = m->private; |
1a75a6c8 CL |
1993 | struct vm_area_struct *vma = v; |
1994 | struct numa_maps *md; | |
397874df CL |
1995 | struct file *file = vma->vm_file; |
1996 | struct mm_struct *mm = vma->vm_mm; | |
480eccf9 | 1997 | struct mempolicy *pol; |
1a75a6c8 CL |
1998 | int n; |
1999 | char buffer[50]; | |
2000 | ||
397874df | 2001 | if (!mm) |
1a75a6c8 CL |
2002 | return 0; |
2003 | ||
2004 | md = kzalloc(sizeof(struct numa_maps), GFP_KERNEL); | |
2005 | if (!md) | |
2006 | return 0; | |
2007 | ||
480eccf9 LS |
2008 | pol = get_vma_policy(priv->task, vma, vma->vm_start); |
2009 | mpol_to_str(buffer, sizeof(buffer), pol); | |
2010 | /* | |
2011 | * unref shared or other task's mempolicy | |
2012 | */ | |
2013 | if (pol != &default_policy && pol != current->mempolicy) | |
2014 | __mpol_free(pol); | |
397874df CL |
2015 | |
2016 | seq_printf(m, "%08lx %s", vma->vm_start, buffer); | |
2017 | ||
2018 | if (file) { | |
2019 | seq_printf(m, " file="); | |
c32c2f63 | 2020 | seq_path(m, &file->f_path, "\n\t= "); |
397874df CL |
2021 | } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { |
2022 | seq_printf(m, " heap"); | |
2023 | } else if (vma->vm_start <= mm->start_stack && | |
2024 | vma->vm_end >= mm->start_stack) { | |
2025 | seq_printf(m, " stack"); | |
2026 | } | |
2027 | ||
2028 | if (is_vm_hugetlb_page(vma)) { | |
2029 | check_huge_range(vma, vma->vm_start, vma->vm_end, md); | |
2030 | seq_printf(m, " huge"); | |
2031 | } else { | |
a57ebfdb | 2032 | check_pgd_range(vma, vma->vm_start, vma->vm_end, |
56bbd65d | 2033 | &node_states[N_HIGH_MEMORY], MPOL_MF_STATS, md); |
397874df CL |
2034 | } |
2035 | ||
2036 | if (!md->pages) | |
2037 | goto out; | |
1a75a6c8 | 2038 | |
397874df CL |
2039 | if (md->anon) |
2040 | seq_printf(m," anon=%lu",md->anon); | |
1a75a6c8 | 2041 | |
397874df CL |
2042 | if (md->dirty) |
2043 | seq_printf(m," dirty=%lu",md->dirty); | |
1a75a6c8 | 2044 | |
397874df CL |
2045 | if (md->pages != md->anon && md->pages != md->dirty) |
2046 | seq_printf(m, " mapped=%lu", md->pages); | |
1a75a6c8 | 2047 | |
397874df CL |
2048 | if (md->mapcount_max > 1) |
2049 | seq_printf(m, " mapmax=%lu", md->mapcount_max); | |
1a75a6c8 | 2050 | |
397874df CL |
2051 | if (md->swapcache) |
2052 | seq_printf(m," swapcache=%lu", md->swapcache); | |
2053 | ||
2054 | if (md->active < md->pages && !is_vm_hugetlb_page(vma)) | |
2055 | seq_printf(m," active=%lu", md->active); | |
2056 | ||
2057 | if (md->writeback) | |
2058 | seq_printf(m," writeback=%lu", md->writeback); | |
2059 | ||
56bbd65d | 2060 | for_each_node_state(n, N_HIGH_MEMORY) |
397874df CL |
2061 | if (md->node[n]) |
2062 | seq_printf(m, " N%d=%lu", n, md->node[n]); | |
2063 | out: | |
2064 | seq_putc(m, '\n'); | |
1a75a6c8 CL |
2065 | kfree(md); |
2066 | ||
2067 | if (m->count < m->size) | |
99f89551 | 2068 | m->version = (vma != priv->tail_vma) ? vma->vm_start : 0; |
1a75a6c8 CL |
2069 | return 0; |
2070 | } |