]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
81819f0f CL |
2 | /* |
3 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
4 | * objects in per cpu and per node lists. | |
5 | * | |
dc84207d | 6 | * The allocator synchronizes using per slab locks or atomic operations |
881db7fb | 7 | * and only uses a centralized lock to manage a pool of partial slabs. |
81819f0f | 8 | * |
cde53535 | 9 | * (C) 2007 SGI, Christoph Lameter |
881db7fb | 10 | * (C) 2011 Linux Foundation, Christoph Lameter |
81819f0f CL |
11 | */ |
12 | ||
13 | #include <linux/mm.h> | |
1eb5ac64 | 14 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
15 | #include <linux/module.h> |
16 | #include <linux/bit_spinlock.h> | |
17 | #include <linux/interrupt.h> | |
1b3865d0 | 18 | #include <linux/swab.h> |
81819f0f CL |
19 | #include <linux/bitops.h> |
20 | #include <linux/slab.h> | |
97d06609 | 21 | #include "slab.h" |
7b3c3a50 | 22 | #include <linux/proc_fs.h> |
81819f0f | 23 | #include <linux/seq_file.h> |
a79316c6 | 24 | #include <linux/kasan.h> |
81819f0f CL |
25 | #include <linux/cpu.h> |
26 | #include <linux/cpuset.h> | |
27 | #include <linux/mempolicy.h> | |
28 | #include <linux/ctype.h> | |
3ac7fe5a | 29 | #include <linux/debugobjects.h> |
81819f0f | 30 | #include <linux/kallsyms.h> |
b89fb5ef | 31 | #include <linux/kfence.h> |
b9049e23 | 32 | #include <linux/memory.h> |
f8bd2258 | 33 | #include <linux/math64.h> |
773ff60e | 34 | #include <linux/fault-inject.h> |
bfa71457 | 35 | #include <linux/stacktrace.h> |
4de900b4 | 36 | #include <linux/prefetch.h> |
2633d7a0 | 37 | #include <linux/memcontrol.h> |
2482ddec | 38 | #include <linux/random.h> |
1f9f78b1 | 39 | #include <kunit/test.h> |
81819f0f | 40 | |
64dd6849 | 41 | #include <linux/debugfs.h> |
4a92379b RK |
42 | #include <trace/events/kmem.h> |
43 | ||
072bb0aa MG |
44 | #include "internal.h" |
45 | ||
81819f0f CL |
46 | /* |
47 | * Lock order: | |
18004c5d | 48 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
49 | * 2. node->list_lock |
50 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 51 | * |
18004c5d | 52 | * slab_mutex |
881db7fb | 53 | * |
18004c5d | 54 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
55 | * and to synchronize major metadata changes to slab cache structures. |
56 | * | |
57 | * The slab_lock is only used for debugging and on arches that do not | |
b7ccc7f8 | 58 | * have the ability to do a cmpxchg_double. It only protects: |
881db7fb | 59 | * A. page->freelist -> List of object free in a page |
b7ccc7f8 MW |
60 | * B. page->inuse -> Number of objects in use |
61 | * C. page->objects -> Number of objects in page | |
62 | * D. page->frozen -> frozen state | |
881db7fb CL |
63 | * |
64 | * If a slab is frozen then it is exempt from list management. It is not | |
632b2ef0 LX |
65 | * on any list except per cpu partial list. The processor that froze the |
66 | * slab is the one who can perform list operations on the page. Other | |
67 | * processors may put objects onto the freelist but the processor that | |
68 | * froze the slab is the only one that can retrieve the objects from the | |
69 | * page's freelist. | |
81819f0f CL |
70 | * |
71 | * The list_lock protects the partial and full list on each node and | |
72 | * the partial slab counter. If taken then no new slabs may be added or | |
73 | * removed from the lists nor make the number of partial slabs be modified. | |
74 | * (Note that the total number of slabs is an atomic value that may be | |
75 | * modified without taking the list lock). | |
76 | * | |
77 | * The list_lock is a centralized lock and thus we avoid taking it as | |
78 | * much as possible. As long as SLUB does not have to handle partial | |
79 | * slabs, operations can continue without any centralized lock. F.e. | |
80 | * allocating a long series of objects that fill up slabs does not require | |
81 | * the list lock. | |
81819f0f CL |
82 | * Interrupts are disabled during allocation and deallocation in order to |
83 | * make the slab allocator safe to use in the context of an irq. In addition | |
84 | * interrupts are disabled to ensure that the processor does not change | |
85 | * while handling per_cpu slabs, due to kernel preemption. | |
86 | * | |
87 | * SLUB assigns one slab for allocation to each processor. | |
88 | * Allocations only occur from these slabs called cpu slabs. | |
89 | * | |
672bba3a CL |
90 | * Slabs with free elements are kept on a partial list and during regular |
91 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 92 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
93 | * We track full slabs for debugging purposes though because otherwise we |
94 | * cannot scan all objects. | |
81819f0f CL |
95 | * |
96 | * Slabs are freed when they become empty. Teardown and setup is | |
97 | * minimal so we rely on the page allocators per cpu caches for | |
98 | * fast frees and allocs. | |
99 | * | |
aed68148 | 100 | * page->frozen The slab is frozen and exempt from list processing. |
4b6f0750 CL |
101 | * This means that the slab is dedicated to a purpose |
102 | * such as satisfying allocations for a specific | |
103 | * processor. Objects may be freed in the slab while | |
104 | * it is frozen but slab_free will then skip the usual | |
105 | * list operations. It is up to the processor holding | |
106 | * the slab to integrate the slab into the slab lists | |
107 | * when the slab is no longer needed. | |
108 | * | |
109 | * One use of this flag is to mark slabs that are | |
110 | * used for allocations. Then such a slab becomes a cpu | |
111 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 112 | * freelist that allows lockless access to |
894b8788 CL |
113 | * free objects in addition to the regular freelist |
114 | * that requires the slab lock. | |
81819f0f | 115 | * |
aed68148 | 116 | * SLAB_DEBUG_FLAGS Slab requires special handling due to debug |
81819f0f | 117 | * options set. This moves slab handling out of |
894b8788 | 118 | * the fast path and disables lockless freelists. |
81819f0f CL |
119 | */ |
120 | ||
ca0cab65 | 121 | #ifdef CONFIG_SLUB_DEBUG |
79270291 | 122 | |
ca0cab65 VB |
123 | #ifdef CONFIG_SLUB_DEBUG_ON |
124 | DEFINE_STATIC_KEY_TRUE(slub_debug_enabled); | |
125 | #else | |
126 | DEFINE_STATIC_KEY_FALSE(slub_debug_enabled); | |
127 | #endif | |
79270291 SB |
128 | |
129 | static inline bool __slub_debug_enabled(void) | |
130 | { | |
131 | return static_branch_unlikely(&slub_debug_enabled); | |
132 | } | |
133 | ||
134 | #else /* CONFIG_SLUB_DEBUG */ | |
135 | ||
136 | static inline bool __slub_debug_enabled(void) | |
137 | { | |
138 | return false; | |
139 | } | |
140 | ||
141 | #endif /* CONFIG_SLUB_DEBUG */ | |
ca0cab65 | 142 | |
59052e89 VB |
143 | static inline bool kmem_cache_debug(struct kmem_cache *s) |
144 | { | |
145 | return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS); | |
af537b0a | 146 | } |
5577bd8a | 147 | |
117d54df | 148 | void *fixup_red_left(struct kmem_cache *s, void *p) |
d86bd1be | 149 | { |
59052e89 | 150 | if (kmem_cache_debug_flags(s, SLAB_RED_ZONE)) |
d86bd1be JK |
151 | p += s->red_left_pad; |
152 | ||
153 | return p; | |
154 | } | |
155 | ||
345c905d JK |
156 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
157 | { | |
158 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
159 | return !kmem_cache_debug(s); | |
160 | #else | |
161 | return false; | |
162 | #endif | |
163 | } | |
164 | ||
81819f0f CL |
165 | /* |
166 | * Issues still to be resolved: | |
167 | * | |
81819f0f CL |
168 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
169 | * | |
81819f0f CL |
170 | * - Variable sizing of the per node arrays |
171 | */ | |
172 | ||
b789ef51 CL |
173 | /* Enable to log cmpxchg failures */ |
174 | #undef SLUB_DEBUG_CMPXCHG | |
175 | ||
2086d26a | 176 | /* |
dc84207d | 177 | * Minimum number of partial slabs. These will be left on the partial |
2086d26a CL |
178 | * lists even if they are empty. kmem_cache_shrink may reclaim them. |
179 | */ | |
76be8950 | 180 | #define MIN_PARTIAL 5 |
e95eed57 | 181 | |
2086d26a CL |
182 | /* |
183 | * Maximum number of desirable partial slabs. | |
184 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 185 | * sort the partial list by the number of objects in use. |
2086d26a CL |
186 | */ |
187 | #define MAX_PARTIAL 10 | |
188 | ||
becfda68 | 189 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 190 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 191 | |
149daaf3 LA |
192 | /* |
193 | * These debug flags cannot use CMPXCHG because there might be consistency | |
194 | * issues when checking or reading debug information | |
195 | */ | |
196 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
197 | SLAB_TRACE) | |
198 | ||
199 | ||
fa5ec8a1 | 200 | /* |
3de47213 DR |
201 | * Debugging flags that require metadata to be stored in the slab. These get |
202 | * disabled when slub_debug=O is used and a cache's min order increases with | |
203 | * metadata. | |
fa5ec8a1 | 204 | */ |
3de47213 | 205 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 206 | |
210b5c06 CG |
207 | #define OO_SHIFT 16 |
208 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 209 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 210 | |
81819f0f | 211 | /* Internal SLUB flags */ |
d50112ed | 212 | /* Poison object */ |
4fd0b46e | 213 | #define __OBJECT_POISON ((slab_flags_t __force)0x80000000U) |
d50112ed | 214 | /* Use cmpxchg_double */ |
4fd0b46e | 215 | #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U) |
81819f0f | 216 | |
02cbc874 CL |
217 | /* |
218 | * Tracking user of a slab. | |
219 | */ | |
d6543e39 | 220 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 221 | struct track { |
ce71e27c | 222 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
223 | #ifdef CONFIG_STACKTRACE |
224 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
225 | #endif | |
02cbc874 CL |
226 | int cpu; /* Was running on cpu */ |
227 | int pid; /* Pid context */ | |
228 | unsigned long when; /* When did the operation occur */ | |
229 | }; | |
230 | ||
231 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
232 | ||
ab4d5ed5 | 233 | #ifdef CONFIG_SYSFS |
81819f0f CL |
234 | static int sysfs_slab_add(struct kmem_cache *); |
235 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
81819f0f | 236 | #else |
0c710013 CL |
237 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
238 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
239 | { return 0; } | |
81819f0f CL |
240 | #endif |
241 | ||
64dd6849 FM |
242 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG) |
243 | static void debugfs_slab_add(struct kmem_cache *); | |
244 | #else | |
245 | static inline void debugfs_slab_add(struct kmem_cache *s) { } | |
246 | #endif | |
247 | ||
4fdccdfb | 248 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
249 | { |
250 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
251 | /* |
252 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
253 | * avoid this_cpu_add()'s irq-disable overhead. | |
254 | */ | |
255 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
256 | #endif |
257 | } | |
258 | ||
7e1fa93d VB |
259 | /* |
260 | * Tracks for which NUMA nodes we have kmem_cache_nodes allocated. | |
261 | * Corresponds to node_state[N_NORMAL_MEMORY], but can temporarily | |
262 | * differ during memory hotplug/hotremove operations. | |
263 | * Protected by slab_mutex. | |
264 | */ | |
265 | static nodemask_t slab_nodes; | |
266 | ||
81819f0f CL |
267 | /******************************************************************** |
268 | * Core slab cache functions | |
269 | *******************************************************************/ | |
270 | ||
2482ddec KC |
271 | /* |
272 | * Returns freelist pointer (ptr). With hardening, this is obfuscated | |
273 | * with an XOR of the address where the pointer is held and a per-cache | |
274 | * random number. | |
275 | */ | |
276 | static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr, | |
277 | unsigned long ptr_addr) | |
278 | { | |
279 | #ifdef CONFIG_SLAB_FREELIST_HARDENED | |
d36a63a9 | 280 | /* |
aa1ef4d7 | 281 | * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged. |
d36a63a9 AK |
282 | * Normally, this doesn't cause any issues, as both set_freepointer() |
283 | * and get_freepointer() are called with a pointer with the same tag. | |
284 | * However, there are some issues with CONFIG_SLUB_DEBUG code. For | |
285 | * example, when __free_slub() iterates over objects in a cache, it | |
286 | * passes untagged pointers to check_object(). check_object() in turns | |
287 | * calls get_freepointer() with an untagged pointer, which causes the | |
288 | * freepointer to be restored incorrectly. | |
289 | */ | |
290 | return (void *)((unsigned long)ptr ^ s->random ^ | |
1ad53d9f | 291 | swab((unsigned long)kasan_reset_tag((void *)ptr_addr))); |
2482ddec KC |
292 | #else |
293 | return ptr; | |
294 | #endif | |
295 | } | |
296 | ||
297 | /* Returns the freelist pointer recorded at location ptr_addr. */ | |
298 | static inline void *freelist_dereference(const struct kmem_cache *s, | |
299 | void *ptr_addr) | |
300 | { | |
301 | return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr), | |
302 | (unsigned long)ptr_addr); | |
303 | } | |
304 | ||
7656c72b CL |
305 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
306 | { | |
aa1ef4d7 | 307 | object = kasan_reset_tag(object); |
2482ddec | 308 | return freelist_dereference(s, object + s->offset); |
7656c72b CL |
309 | } |
310 | ||
0ad9500e ED |
311 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
312 | { | |
0882ff91 | 313 | prefetch(object + s->offset); |
0ad9500e ED |
314 | } |
315 | ||
1393d9a1 CL |
316 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
317 | { | |
2482ddec | 318 | unsigned long freepointer_addr; |
1393d9a1 CL |
319 | void *p; |
320 | ||
8e57f8ac | 321 | if (!debug_pagealloc_enabled_static()) |
922d566c JK |
322 | return get_freepointer(s, object); |
323 | ||
f70b0049 | 324 | object = kasan_reset_tag(object); |
2482ddec | 325 | freepointer_addr = (unsigned long)object + s->offset; |
fe557319 | 326 | copy_from_kernel_nofault(&p, (void **)freepointer_addr, sizeof(p)); |
2482ddec | 327 | return freelist_ptr(s, p, freepointer_addr); |
1393d9a1 CL |
328 | } |
329 | ||
7656c72b CL |
330 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
331 | { | |
2482ddec KC |
332 | unsigned long freeptr_addr = (unsigned long)object + s->offset; |
333 | ||
ce6fa91b AP |
334 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
335 | BUG_ON(object == fp); /* naive detection of double free or corruption */ | |
336 | #endif | |
337 | ||
aa1ef4d7 | 338 | freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr); |
2482ddec | 339 | *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr); |
7656c72b CL |
340 | } |
341 | ||
342 | /* Loop over all objects in a slab */ | |
224a88be | 343 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
344 | for (__p = fixup_red_left(__s, __addr); \ |
345 | __p < (__addr) + (__objects) * (__s)->size; \ | |
346 | __p += (__s)->size) | |
7656c72b | 347 | |
9736d2a9 | 348 | static inline unsigned int order_objects(unsigned int order, unsigned int size) |
ab9a0f19 | 349 | { |
9736d2a9 | 350 | return ((unsigned int)PAGE_SIZE << order) / size; |
ab9a0f19 LJ |
351 | } |
352 | ||
19af27af | 353 | static inline struct kmem_cache_order_objects oo_make(unsigned int order, |
9736d2a9 | 354 | unsigned int size) |
834f3d11 CL |
355 | { |
356 | struct kmem_cache_order_objects x = { | |
9736d2a9 | 357 | (order << OO_SHIFT) + order_objects(order, size) |
834f3d11 CL |
358 | }; |
359 | ||
360 | return x; | |
361 | } | |
362 | ||
19af27af | 363 | static inline unsigned int oo_order(struct kmem_cache_order_objects x) |
834f3d11 | 364 | { |
210b5c06 | 365 | return x.x >> OO_SHIFT; |
834f3d11 CL |
366 | } |
367 | ||
19af27af | 368 | static inline unsigned int oo_objects(struct kmem_cache_order_objects x) |
834f3d11 | 369 | { |
210b5c06 | 370 | return x.x & OO_MASK; |
834f3d11 CL |
371 | } |
372 | ||
881db7fb CL |
373 | /* |
374 | * Per slab locking using the pagelock | |
375 | */ | |
376 | static __always_inline void slab_lock(struct page *page) | |
377 | { | |
48c935ad | 378 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
379 | bit_spin_lock(PG_locked, &page->flags); |
380 | } | |
381 | ||
382 | static __always_inline void slab_unlock(struct page *page) | |
383 | { | |
48c935ad | 384 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
385 | __bit_spin_unlock(PG_locked, &page->flags); |
386 | } | |
387 | ||
1d07171c CL |
388 | /* Interrupts must be disabled (for the fallback code to work right) */ |
389 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
390 | void *freelist_old, unsigned long counters_old, | |
391 | void *freelist_new, unsigned long counters_new, | |
392 | const char *n) | |
393 | { | |
394 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
395 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
396 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 397 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 398 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
399 | freelist_old, counters_old, |
400 | freelist_new, counters_new)) | |
6f6528a1 | 401 | return true; |
1d07171c CL |
402 | } else |
403 | #endif | |
404 | { | |
405 | slab_lock(page); | |
d0e0ac97 CG |
406 | if (page->freelist == freelist_old && |
407 | page->counters == counters_old) { | |
1d07171c | 408 | page->freelist = freelist_new; |
7d27a04b | 409 | page->counters = counters_new; |
1d07171c | 410 | slab_unlock(page); |
6f6528a1 | 411 | return true; |
1d07171c CL |
412 | } |
413 | slab_unlock(page); | |
414 | } | |
415 | ||
416 | cpu_relax(); | |
417 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
418 | ||
419 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 420 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
421 | #endif |
422 | ||
6f6528a1 | 423 | return false; |
1d07171c CL |
424 | } |
425 | ||
b789ef51 CL |
426 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
427 | void *freelist_old, unsigned long counters_old, | |
428 | void *freelist_new, unsigned long counters_new, | |
429 | const char *n) | |
430 | { | |
2565409f HC |
431 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
432 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 433 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 434 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
435 | freelist_old, counters_old, |
436 | freelist_new, counters_new)) | |
6f6528a1 | 437 | return true; |
b789ef51 CL |
438 | } else |
439 | #endif | |
440 | { | |
1d07171c CL |
441 | unsigned long flags; |
442 | ||
443 | local_irq_save(flags); | |
881db7fb | 444 | slab_lock(page); |
d0e0ac97 CG |
445 | if (page->freelist == freelist_old && |
446 | page->counters == counters_old) { | |
b789ef51 | 447 | page->freelist = freelist_new; |
7d27a04b | 448 | page->counters = counters_new; |
881db7fb | 449 | slab_unlock(page); |
1d07171c | 450 | local_irq_restore(flags); |
6f6528a1 | 451 | return true; |
b789ef51 | 452 | } |
881db7fb | 453 | slab_unlock(page); |
1d07171c | 454 | local_irq_restore(flags); |
b789ef51 CL |
455 | } |
456 | ||
457 | cpu_relax(); | |
458 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
459 | ||
460 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 461 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
462 | #endif |
463 | ||
6f6528a1 | 464 | return false; |
b789ef51 CL |
465 | } |
466 | ||
41ecc55b | 467 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 YZ |
468 | static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)]; |
469 | static DEFINE_SPINLOCK(object_map_lock); | |
470 | ||
1f9f78b1 OG |
471 | #if IS_ENABLED(CONFIG_KUNIT) |
472 | static bool slab_add_kunit_errors(void) | |
473 | { | |
474 | struct kunit_resource *resource; | |
475 | ||
476 | if (likely(!current->kunit_test)) | |
477 | return false; | |
478 | ||
479 | resource = kunit_find_named_resource(current->kunit_test, "slab_errors"); | |
480 | if (!resource) | |
481 | return false; | |
482 | ||
483 | (*(int *)resource->data)++; | |
484 | kunit_put_resource(resource); | |
485 | return true; | |
486 | } | |
487 | #else | |
488 | static inline bool slab_add_kunit_errors(void) { return false; } | |
489 | #endif | |
490 | ||
5f80b13a CL |
491 | /* |
492 | * Determine a map of object in use on a page. | |
493 | * | |
881db7fb | 494 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
495 | * not vanish from under us. |
496 | */ | |
90e9f6a6 | 497 | static unsigned long *get_map(struct kmem_cache *s, struct page *page) |
31364c2e | 498 | __acquires(&object_map_lock) |
5f80b13a CL |
499 | { |
500 | void *p; | |
501 | void *addr = page_address(page); | |
502 | ||
90e9f6a6 YZ |
503 | VM_BUG_ON(!irqs_disabled()); |
504 | ||
505 | spin_lock(&object_map_lock); | |
506 | ||
507 | bitmap_zero(object_map, page->objects); | |
508 | ||
5f80b13a | 509 | for (p = page->freelist; p; p = get_freepointer(s, p)) |
4138fdfc | 510 | set_bit(__obj_to_index(s, addr, p), object_map); |
90e9f6a6 YZ |
511 | |
512 | return object_map; | |
513 | } | |
514 | ||
81aba9e0 | 515 | static void put_map(unsigned long *map) __releases(&object_map_lock) |
90e9f6a6 YZ |
516 | { |
517 | VM_BUG_ON(map != object_map); | |
90e9f6a6 | 518 | spin_unlock(&object_map_lock); |
5f80b13a CL |
519 | } |
520 | ||
870b1fbb | 521 | static inline unsigned int size_from_object(struct kmem_cache *s) |
d86bd1be JK |
522 | { |
523 | if (s->flags & SLAB_RED_ZONE) | |
524 | return s->size - s->red_left_pad; | |
525 | ||
526 | return s->size; | |
527 | } | |
528 | ||
529 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
530 | { | |
531 | if (s->flags & SLAB_RED_ZONE) | |
532 | p -= s->red_left_pad; | |
533 | ||
534 | return p; | |
535 | } | |
536 | ||
41ecc55b CL |
537 | /* |
538 | * Debug settings: | |
539 | */ | |
89d3c87e | 540 | #if defined(CONFIG_SLUB_DEBUG_ON) |
d50112ed | 541 | static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS; |
f0630fff | 542 | #else |
d50112ed | 543 | static slab_flags_t slub_debug; |
f0630fff | 544 | #endif |
41ecc55b | 545 | |
e17f1dfb | 546 | static char *slub_debug_string; |
fa5ec8a1 | 547 | static int disable_higher_order_debug; |
41ecc55b | 548 | |
a79316c6 AR |
549 | /* |
550 | * slub is about to manipulate internal object metadata. This memory lies | |
551 | * outside the range of the allocated object, so accessing it would normally | |
552 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
553 | * to tell kasan that these accesses are OK. | |
554 | */ | |
555 | static inline void metadata_access_enable(void) | |
556 | { | |
557 | kasan_disable_current(); | |
558 | } | |
559 | ||
560 | static inline void metadata_access_disable(void) | |
561 | { | |
562 | kasan_enable_current(); | |
563 | } | |
564 | ||
81819f0f CL |
565 | /* |
566 | * Object debugging | |
567 | */ | |
d86bd1be JK |
568 | |
569 | /* Verify that a pointer has an address that is valid within a slab page */ | |
570 | static inline int check_valid_pointer(struct kmem_cache *s, | |
571 | struct page *page, void *object) | |
572 | { | |
573 | void *base; | |
574 | ||
575 | if (!object) | |
576 | return 1; | |
577 | ||
578 | base = page_address(page); | |
338cfaad | 579 | object = kasan_reset_tag(object); |
d86bd1be JK |
580 | object = restore_red_left(s, object); |
581 | if (object < base || object >= base + page->objects * s->size || | |
582 | (object - base) % s->size) { | |
583 | return 0; | |
584 | } | |
585 | ||
586 | return 1; | |
587 | } | |
588 | ||
aa2efd5e DT |
589 | static void print_section(char *level, char *text, u8 *addr, |
590 | unsigned int length) | |
81819f0f | 591 | { |
a79316c6 | 592 | metadata_access_enable(); |
aa1ef4d7 AK |
593 | print_hex_dump(level, kasan_reset_tag(text), DUMP_PREFIX_ADDRESS, |
594 | 16, 1, addr, length, 1); | |
a79316c6 | 595 | metadata_access_disable(); |
81819f0f CL |
596 | } |
597 | ||
cbfc35a4 WL |
598 | /* |
599 | * See comment in calculate_sizes(). | |
600 | */ | |
601 | static inline bool freeptr_outside_object(struct kmem_cache *s) | |
602 | { | |
603 | return s->offset >= s->inuse; | |
604 | } | |
605 | ||
606 | /* | |
607 | * Return offset of the end of info block which is inuse + free pointer if | |
608 | * not overlapping with object. | |
609 | */ | |
610 | static inline unsigned int get_info_end(struct kmem_cache *s) | |
611 | { | |
612 | if (freeptr_outside_object(s)) | |
613 | return s->inuse + sizeof(void *); | |
614 | else | |
615 | return s->inuse; | |
616 | } | |
617 | ||
81819f0f CL |
618 | static struct track *get_track(struct kmem_cache *s, void *object, |
619 | enum track_item alloc) | |
620 | { | |
621 | struct track *p; | |
622 | ||
cbfc35a4 | 623 | p = object + get_info_end(s); |
81819f0f | 624 | |
aa1ef4d7 | 625 | return kasan_reset_tag(p + alloc); |
81819f0f CL |
626 | } |
627 | ||
628 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 629 | enum track_item alloc, unsigned long addr) |
81819f0f | 630 | { |
1a00df4a | 631 | struct track *p = get_track(s, object, alloc); |
81819f0f | 632 | |
81819f0f | 633 | if (addr) { |
d6543e39 | 634 | #ifdef CONFIG_STACKTRACE |
79716799 | 635 | unsigned int nr_entries; |
d6543e39 | 636 | |
a79316c6 | 637 | metadata_access_enable(); |
aa1ef4d7 AK |
638 | nr_entries = stack_trace_save(kasan_reset_tag(p->addrs), |
639 | TRACK_ADDRS_COUNT, 3); | |
a79316c6 | 640 | metadata_access_disable(); |
d6543e39 | 641 | |
79716799 TG |
642 | if (nr_entries < TRACK_ADDRS_COUNT) |
643 | p->addrs[nr_entries] = 0; | |
d6543e39 | 644 | #endif |
81819f0f CL |
645 | p->addr = addr; |
646 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 647 | p->pid = current->pid; |
81819f0f | 648 | p->when = jiffies; |
b8ca7ff7 | 649 | } else { |
81819f0f | 650 | memset(p, 0, sizeof(struct track)); |
b8ca7ff7 | 651 | } |
81819f0f CL |
652 | } |
653 | ||
81819f0f CL |
654 | static void init_tracking(struct kmem_cache *s, void *object) |
655 | { | |
24922684 CL |
656 | if (!(s->flags & SLAB_STORE_USER)) |
657 | return; | |
658 | ||
ce71e27c EGM |
659 | set_track(s, object, TRACK_FREE, 0UL); |
660 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
661 | } |
662 | ||
86609d33 | 663 | static void print_track(const char *s, struct track *t, unsigned long pr_time) |
81819f0f CL |
664 | { |
665 | if (!t->addr) | |
666 | return; | |
667 | ||
96b94abc | 668 | pr_err("%s in %pS age=%lu cpu=%u pid=%d\n", |
86609d33 | 669 | s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid); |
d6543e39 BG |
670 | #ifdef CONFIG_STACKTRACE |
671 | { | |
672 | int i; | |
673 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
674 | if (t->addrs[i]) | |
f9f58285 | 675 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
676 | else |
677 | break; | |
678 | } | |
679 | #endif | |
24922684 CL |
680 | } |
681 | ||
e42f174e | 682 | void print_tracking(struct kmem_cache *s, void *object) |
24922684 | 683 | { |
86609d33 | 684 | unsigned long pr_time = jiffies; |
24922684 CL |
685 | if (!(s->flags & SLAB_STORE_USER)) |
686 | return; | |
687 | ||
86609d33 CP |
688 | print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time); |
689 | print_track("Freed", get_track(s, object, TRACK_FREE), pr_time); | |
24922684 CL |
690 | } |
691 | ||
692 | static void print_page_info(struct page *page) | |
693 | { | |
96b94abc | 694 | pr_err("Slab 0x%p objects=%u used=%u fp=0x%p flags=%#lx(%pGp)\n", |
4a8ef190 YS |
695 | page, page->objects, page->inuse, page->freelist, |
696 | page->flags, &page->flags); | |
24922684 CL |
697 | |
698 | } | |
699 | ||
700 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
701 | { | |
ecc42fbe | 702 | struct va_format vaf; |
24922684 | 703 | va_list args; |
24922684 CL |
704 | |
705 | va_start(args, fmt); | |
ecc42fbe FF |
706 | vaf.fmt = fmt; |
707 | vaf.va = &args; | |
f9f58285 | 708 | pr_err("=============================================================================\n"); |
ecc42fbe | 709 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 710 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 711 | |
373d4d09 | 712 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 713 | va_end(args); |
81819f0f CL |
714 | } |
715 | ||
582d1212 | 716 | __printf(2, 3) |
24922684 CL |
717 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
718 | { | |
ecc42fbe | 719 | struct va_format vaf; |
24922684 | 720 | va_list args; |
24922684 | 721 | |
1f9f78b1 OG |
722 | if (slab_add_kunit_errors()) |
723 | return; | |
724 | ||
24922684 | 725 | va_start(args, fmt); |
ecc42fbe FF |
726 | vaf.fmt = fmt; |
727 | vaf.va = &args; | |
728 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 729 | va_end(args); |
24922684 CL |
730 | } |
731 | ||
52f23478 | 732 | static bool freelist_corrupted(struct kmem_cache *s, struct page *page, |
dc07a728 | 733 | void **freelist, void *nextfree) |
52f23478 DZ |
734 | { |
735 | if ((s->flags & SLAB_CONSISTENCY_CHECKS) && | |
dc07a728 ER |
736 | !check_valid_pointer(s, page, nextfree) && freelist) { |
737 | object_err(s, page, *freelist, "Freechain corrupt"); | |
738 | *freelist = NULL; | |
52f23478 DZ |
739 | slab_fix(s, "Isolate corrupted freechain"); |
740 | return true; | |
741 | } | |
742 | ||
743 | return false; | |
744 | } | |
745 | ||
24922684 | 746 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) |
81819f0f CL |
747 | { |
748 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 749 | u8 *addr = page_address(page); |
24922684 CL |
750 | |
751 | print_tracking(s, p); | |
752 | ||
753 | print_page_info(page); | |
754 | ||
96b94abc | 755 | pr_err("Object 0x%p @offset=%tu fp=0x%p\n\n", |
f9f58285 | 756 | p, p - addr, get_freepointer(s, p)); |
24922684 | 757 | |
d86bd1be | 758 | if (s->flags & SLAB_RED_ZONE) |
8669dbab | 759 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
aa2efd5e | 760 | s->red_left_pad); |
d86bd1be | 761 | else if (p > addr + 16) |
aa2efd5e | 762 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 763 | |
8669dbab | 764 | print_section(KERN_ERR, "Object ", p, |
1b473f29 | 765 | min_t(unsigned int, s->object_size, PAGE_SIZE)); |
81819f0f | 766 | if (s->flags & SLAB_RED_ZONE) |
8669dbab | 767 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 768 | s->inuse - s->object_size); |
81819f0f | 769 | |
cbfc35a4 | 770 | off = get_info_end(s); |
81819f0f | 771 | |
24922684 | 772 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 773 | off += 2 * sizeof(struct track); |
81819f0f | 774 | |
80a9201a AP |
775 | off += kasan_metadata_size(s); |
776 | ||
d86bd1be | 777 | if (off != size_from_object(s)) |
81819f0f | 778 | /* Beginning of the filler is the free pointer */ |
8669dbab | 779 | print_section(KERN_ERR, "Padding ", p + off, |
aa2efd5e | 780 | size_from_object(s) - off); |
24922684 CL |
781 | |
782 | dump_stack(); | |
81819f0f CL |
783 | } |
784 | ||
75c66def | 785 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
786 | u8 *object, char *reason) |
787 | { | |
1f9f78b1 OG |
788 | if (slab_add_kunit_errors()) |
789 | return; | |
790 | ||
3dc50637 | 791 | slab_bug(s, "%s", reason); |
24922684 | 792 | print_trailer(s, page, object); |
81819f0f CL |
793 | } |
794 | ||
a38965bf | 795 | static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page, |
d0e0ac97 | 796 | const char *fmt, ...) |
81819f0f CL |
797 | { |
798 | va_list args; | |
799 | char buf[100]; | |
800 | ||
1f9f78b1 OG |
801 | if (slab_add_kunit_errors()) |
802 | return; | |
803 | ||
24922684 CL |
804 | va_start(args, fmt); |
805 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 806 | va_end(args); |
3dc50637 | 807 | slab_bug(s, "%s", buf); |
24922684 | 808 | print_page_info(page); |
81819f0f CL |
809 | dump_stack(); |
810 | } | |
811 | ||
f7cb1933 | 812 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f | 813 | { |
aa1ef4d7 | 814 | u8 *p = kasan_reset_tag(object); |
81819f0f | 815 | |
d86bd1be JK |
816 | if (s->flags & SLAB_RED_ZONE) |
817 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
818 | ||
81819f0f | 819 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
820 | memset(p, POISON_FREE, s->object_size - 1); |
821 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
822 | } |
823 | ||
824 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 825 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
826 | } |
827 | ||
24922684 CL |
828 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
829 | void *from, void *to) | |
830 | { | |
582d1212 | 831 | slab_fix(s, "Restoring %s 0x%p-0x%p=0x%x", message, from, to - 1, data); |
24922684 CL |
832 | memset(from, data, to - from); |
833 | } | |
834 | ||
835 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
836 | u8 *object, char *what, | |
06428780 | 837 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
838 | { |
839 | u8 *fault; | |
840 | u8 *end; | |
e1b70dd1 | 841 | u8 *addr = page_address(page); |
24922684 | 842 | |
a79316c6 | 843 | metadata_access_enable(); |
aa1ef4d7 | 844 | fault = memchr_inv(kasan_reset_tag(start), value, bytes); |
a79316c6 | 845 | metadata_access_disable(); |
24922684 CL |
846 | if (!fault) |
847 | return 1; | |
848 | ||
849 | end = start + bytes; | |
850 | while (end > fault && end[-1] == value) | |
851 | end--; | |
852 | ||
1f9f78b1 OG |
853 | if (slab_add_kunit_errors()) |
854 | goto skip_bug_print; | |
855 | ||
24922684 | 856 | slab_bug(s, "%s overwritten", what); |
96b94abc | 857 | pr_err("0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n", |
e1b70dd1 MC |
858 | fault, end - 1, fault - addr, |
859 | fault[0], value); | |
24922684 CL |
860 | print_trailer(s, page, object); |
861 | ||
1f9f78b1 | 862 | skip_bug_print: |
24922684 CL |
863 | restore_bytes(s, what, value, fault, end); |
864 | return 0; | |
81819f0f CL |
865 | } |
866 | ||
81819f0f CL |
867 | /* |
868 | * Object layout: | |
869 | * | |
870 | * object address | |
871 | * Bytes of the object to be managed. | |
872 | * If the freepointer may overlay the object then the free | |
cbfc35a4 | 873 | * pointer is at the middle of the object. |
672bba3a | 874 | * |
81819f0f CL |
875 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
876 | * 0xa5 (POISON_END) | |
877 | * | |
3b0efdfa | 878 | * object + s->object_size |
81819f0f | 879 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 880 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 881 | * object_size == inuse. |
672bba3a | 882 | * |
81819f0f CL |
883 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
884 | * 0xcc (RED_ACTIVE) for objects in use. | |
885 | * | |
886 | * object + s->inuse | |
672bba3a CL |
887 | * Meta data starts here. |
888 | * | |
81819f0f CL |
889 | * A. Free pointer (if we cannot overwrite object on free) |
890 | * B. Tracking data for SLAB_STORE_USER | |
dc84207d | 891 | * C. Padding to reach required alignment boundary or at minimum |
6446faa2 | 892 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
893 | * before the word boundary. |
894 | * | |
895 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
896 | * |
897 | * object + s->size | |
672bba3a | 898 | * Nothing is used beyond s->size. |
81819f0f | 899 | * |
3b0efdfa | 900 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 901 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
902 | * may be used with merged slabcaches. |
903 | */ | |
904 | ||
81819f0f CL |
905 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
906 | { | |
cbfc35a4 | 907 | unsigned long off = get_info_end(s); /* The end of info */ |
81819f0f CL |
908 | |
909 | if (s->flags & SLAB_STORE_USER) | |
910 | /* We also have user information there */ | |
911 | off += 2 * sizeof(struct track); | |
912 | ||
80a9201a AP |
913 | off += kasan_metadata_size(s); |
914 | ||
d86bd1be | 915 | if (size_from_object(s) == off) |
81819f0f CL |
916 | return 1; |
917 | ||
24922684 | 918 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 919 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
920 | } |
921 | ||
39b26464 | 922 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
923 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
924 | { | |
24922684 CL |
925 | u8 *start; |
926 | u8 *fault; | |
927 | u8 *end; | |
5d682681 | 928 | u8 *pad; |
24922684 CL |
929 | int length; |
930 | int remainder; | |
81819f0f CL |
931 | |
932 | if (!(s->flags & SLAB_POISON)) | |
933 | return 1; | |
934 | ||
a973e9dd | 935 | start = page_address(page); |
a50b854e | 936 | length = page_size(page); |
39b26464 CL |
937 | end = start + length; |
938 | remainder = length % s->size; | |
81819f0f CL |
939 | if (!remainder) |
940 | return 1; | |
941 | ||
5d682681 | 942 | pad = end - remainder; |
a79316c6 | 943 | metadata_access_enable(); |
aa1ef4d7 | 944 | fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder); |
a79316c6 | 945 | metadata_access_disable(); |
24922684 CL |
946 | if (!fault) |
947 | return 1; | |
948 | while (end > fault && end[-1] == POISON_INUSE) | |
949 | end--; | |
950 | ||
e1b70dd1 MC |
951 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p @offset=%tu", |
952 | fault, end - 1, fault - start); | |
5d682681 | 953 | print_section(KERN_ERR, "Padding ", pad, remainder); |
24922684 | 954 | |
5d682681 | 955 | restore_bytes(s, "slab padding", POISON_INUSE, fault, end); |
24922684 | 956 | return 0; |
81819f0f CL |
957 | } |
958 | ||
959 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 960 | void *object, u8 val) |
81819f0f CL |
961 | { |
962 | u8 *p = object; | |
3b0efdfa | 963 | u8 *endobject = object + s->object_size; |
81819f0f CL |
964 | |
965 | if (s->flags & SLAB_RED_ZONE) { | |
8669dbab | 966 | if (!check_bytes_and_report(s, page, object, "Left Redzone", |
d86bd1be JK |
967 | object - s->red_left_pad, val, s->red_left_pad)) |
968 | return 0; | |
969 | ||
8669dbab | 970 | if (!check_bytes_and_report(s, page, object, "Right Redzone", |
3b0efdfa | 971 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 972 | return 0; |
81819f0f | 973 | } else { |
3b0efdfa | 974 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 975 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
976 | endobject, POISON_INUSE, |
977 | s->inuse - s->object_size); | |
3adbefee | 978 | } |
81819f0f CL |
979 | } |
980 | ||
981 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 982 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 983 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 984 | POISON_FREE, s->object_size - 1) || |
8669dbab | 985 | !check_bytes_and_report(s, page, p, "End Poison", |
3b0efdfa | 986 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 987 | return 0; |
81819f0f CL |
988 | /* |
989 | * check_pad_bytes cleans up on its own. | |
990 | */ | |
991 | check_pad_bytes(s, page, p); | |
992 | } | |
993 | ||
cbfc35a4 | 994 | if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE) |
81819f0f CL |
995 | /* |
996 | * Object and freepointer overlap. Cannot check | |
997 | * freepointer while object is allocated. | |
998 | */ | |
999 | return 1; | |
1000 | ||
1001 | /* Check free pointer validity */ | |
1002 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
1003 | object_err(s, page, p, "Freepointer corrupt"); | |
1004 | /* | |
9f6c708e | 1005 | * No choice but to zap it and thus lose the remainder |
81819f0f | 1006 | * of the free objects in this slab. May cause |
672bba3a | 1007 | * another error because the object count is now wrong. |
81819f0f | 1008 | */ |
a973e9dd | 1009 | set_freepointer(s, p, NULL); |
81819f0f CL |
1010 | return 0; |
1011 | } | |
1012 | return 1; | |
1013 | } | |
1014 | ||
1015 | static int check_slab(struct kmem_cache *s, struct page *page) | |
1016 | { | |
39b26464 CL |
1017 | int maxobj; |
1018 | ||
81819f0f CL |
1019 | VM_BUG_ON(!irqs_disabled()); |
1020 | ||
1021 | if (!PageSlab(page)) { | |
24922684 | 1022 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
1023 | return 0; |
1024 | } | |
39b26464 | 1025 | |
9736d2a9 | 1026 | maxobj = order_objects(compound_order(page), s->size); |
39b26464 CL |
1027 | if (page->objects > maxobj) { |
1028 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 1029 | page->objects, maxobj); |
39b26464 CL |
1030 | return 0; |
1031 | } | |
1032 | if (page->inuse > page->objects) { | |
24922684 | 1033 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 1034 | page->inuse, page->objects); |
81819f0f CL |
1035 | return 0; |
1036 | } | |
1037 | /* Slab_pad_check fixes things up after itself */ | |
1038 | slab_pad_check(s, page); | |
1039 | return 1; | |
1040 | } | |
1041 | ||
1042 | /* | |
672bba3a CL |
1043 | * Determine if a certain object on a page is on the freelist. Must hold the |
1044 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
1045 | */ |
1046 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
1047 | { | |
1048 | int nr = 0; | |
881db7fb | 1049 | void *fp; |
81819f0f | 1050 | void *object = NULL; |
f6edde9c | 1051 | int max_objects; |
81819f0f | 1052 | |
881db7fb | 1053 | fp = page->freelist; |
39b26464 | 1054 | while (fp && nr <= page->objects) { |
81819f0f CL |
1055 | if (fp == search) |
1056 | return 1; | |
1057 | if (!check_valid_pointer(s, page, fp)) { | |
1058 | if (object) { | |
1059 | object_err(s, page, object, | |
1060 | "Freechain corrupt"); | |
a973e9dd | 1061 | set_freepointer(s, object, NULL); |
81819f0f | 1062 | } else { |
24922684 | 1063 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 1064 | page->freelist = NULL; |
39b26464 | 1065 | page->inuse = page->objects; |
24922684 | 1066 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
1067 | return 0; |
1068 | } | |
1069 | break; | |
1070 | } | |
1071 | object = fp; | |
1072 | fp = get_freepointer(s, object); | |
1073 | nr++; | |
1074 | } | |
1075 | ||
9736d2a9 | 1076 | max_objects = order_objects(compound_order(page), s->size); |
210b5c06 CG |
1077 | if (max_objects > MAX_OBJS_PER_PAGE) |
1078 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
1079 | |
1080 | if (page->objects != max_objects) { | |
756a025f JP |
1081 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
1082 | page->objects, max_objects); | |
224a88be | 1083 | page->objects = max_objects; |
582d1212 | 1084 | slab_fix(s, "Number of objects adjusted"); |
224a88be | 1085 | } |
39b26464 | 1086 | if (page->inuse != page->objects - nr) { |
756a025f JP |
1087 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
1088 | page->inuse, page->objects - nr); | |
39b26464 | 1089 | page->inuse = page->objects - nr; |
582d1212 | 1090 | slab_fix(s, "Object count adjusted"); |
81819f0f CL |
1091 | } |
1092 | return search == NULL; | |
1093 | } | |
1094 | ||
0121c619 CL |
1095 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
1096 | int alloc) | |
3ec09742 CL |
1097 | { |
1098 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 1099 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
1100 | s->name, |
1101 | alloc ? "alloc" : "free", | |
1102 | object, page->inuse, | |
1103 | page->freelist); | |
1104 | ||
1105 | if (!alloc) | |
aa2efd5e | 1106 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 1107 | s->object_size); |
3ec09742 CL |
1108 | |
1109 | dump_stack(); | |
1110 | } | |
1111 | } | |
1112 | ||
643b1138 | 1113 | /* |
672bba3a | 1114 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1115 | */ |
5cc6eee8 CL |
1116 | static void add_full(struct kmem_cache *s, |
1117 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1118 | { |
5cc6eee8 CL |
1119 | if (!(s->flags & SLAB_STORE_USER)) |
1120 | return; | |
1121 | ||
255d0884 | 1122 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1123 | list_add(&page->slab_list, &n->full); |
643b1138 CL |
1124 | } |
1125 | ||
c65c1877 | 1126 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1127 | { |
643b1138 CL |
1128 | if (!(s->flags & SLAB_STORE_USER)) |
1129 | return; | |
1130 | ||
255d0884 | 1131 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1132 | list_del(&page->slab_list); |
643b1138 CL |
1133 | } |
1134 | ||
0f389ec6 CL |
1135 | /* Tracking of the number of slabs for debugging purposes */ |
1136 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1137 | { | |
1138 | struct kmem_cache_node *n = get_node(s, node); | |
1139 | ||
1140 | return atomic_long_read(&n->nr_slabs); | |
1141 | } | |
1142 | ||
26c02cf0 AB |
1143 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1144 | { | |
1145 | return atomic_long_read(&n->nr_slabs); | |
1146 | } | |
1147 | ||
205ab99d | 1148 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1149 | { |
1150 | struct kmem_cache_node *n = get_node(s, node); | |
1151 | ||
1152 | /* | |
1153 | * May be called early in order to allocate a slab for the | |
1154 | * kmem_cache_node structure. Solve the chicken-egg | |
1155 | * dilemma by deferring the increment of the count during | |
1156 | * bootstrap (see early_kmem_cache_node_alloc). | |
1157 | */ | |
338b2642 | 1158 | if (likely(n)) { |
0f389ec6 | 1159 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1160 | atomic_long_add(objects, &n->total_objects); |
1161 | } | |
0f389ec6 | 1162 | } |
205ab99d | 1163 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1164 | { |
1165 | struct kmem_cache_node *n = get_node(s, node); | |
1166 | ||
1167 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1168 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1169 | } |
1170 | ||
1171 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1172 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1173 | void *object) | |
1174 | { | |
8fc8d666 | 1175 | if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)) |
3ec09742 CL |
1176 | return; |
1177 | ||
f7cb1933 | 1178 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1179 | init_tracking(s, object); |
1180 | } | |
1181 | ||
a50b854e MWO |
1182 | static |
1183 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) | |
a7101224 | 1184 | { |
8fc8d666 | 1185 | if (!kmem_cache_debug_flags(s, SLAB_POISON)) |
a7101224 AK |
1186 | return; |
1187 | ||
1188 | metadata_access_enable(); | |
aa1ef4d7 | 1189 | memset(kasan_reset_tag(addr), POISON_INUSE, page_size(page)); |
a7101224 AK |
1190 | metadata_access_disable(); |
1191 | } | |
1192 | ||
becfda68 | 1193 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
278d7756 | 1194 | struct page *page, void *object) |
81819f0f CL |
1195 | { |
1196 | if (!check_slab(s, page)) | |
becfda68 | 1197 | return 0; |
81819f0f | 1198 | |
81819f0f CL |
1199 | if (!check_valid_pointer(s, page, object)) { |
1200 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1201 | return 0; |
81819f0f CL |
1202 | } |
1203 | ||
f7cb1933 | 1204 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1205 | return 0; |
1206 | ||
1207 | return 1; | |
1208 | } | |
1209 | ||
1210 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1211 | struct page *page, | |
1212 | void *object, unsigned long addr) | |
1213 | { | |
1214 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
278d7756 | 1215 | if (!alloc_consistency_checks(s, page, object)) |
becfda68 LA |
1216 | goto bad; |
1217 | } | |
81819f0f | 1218 | |
3ec09742 CL |
1219 | /* Success perform special debug activities for allocs */ |
1220 | if (s->flags & SLAB_STORE_USER) | |
1221 | set_track(s, object, TRACK_ALLOC, addr); | |
1222 | trace(s, page, object, 1); | |
f7cb1933 | 1223 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1224 | return 1; |
3ec09742 | 1225 | |
81819f0f CL |
1226 | bad: |
1227 | if (PageSlab(page)) { | |
1228 | /* | |
1229 | * If this is a slab page then lets do the best we can | |
1230 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1231 | * as used avoids touching the remaining objects. |
81819f0f | 1232 | */ |
24922684 | 1233 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1234 | page->inuse = page->objects; |
a973e9dd | 1235 | page->freelist = NULL; |
81819f0f CL |
1236 | } |
1237 | return 0; | |
1238 | } | |
1239 | ||
becfda68 LA |
1240 | static inline int free_consistency_checks(struct kmem_cache *s, |
1241 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1242 | { |
81819f0f | 1243 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1244 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1245 | return 0; |
81819f0f CL |
1246 | } |
1247 | ||
1248 | if (on_freelist(s, page, object)) { | |
24922684 | 1249 | object_err(s, page, object, "Object already free"); |
becfda68 | 1250 | return 0; |
81819f0f CL |
1251 | } |
1252 | ||
f7cb1933 | 1253 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1254 | return 0; |
81819f0f | 1255 | |
1b4f59e3 | 1256 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1257 | if (!PageSlab(page)) { |
756a025f JP |
1258 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1259 | object); | |
1b4f59e3 | 1260 | } else if (!page->slab_cache) { |
f9f58285 FF |
1261 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1262 | object); | |
70d71228 | 1263 | dump_stack(); |
06428780 | 1264 | } else |
24922684 CL |
1265 | object_err(s, page, object, |
1266 | "page slab pointer corrupt."); | |
becfda68 LA |
1267 | return 0; |
1268 | } | |
1269 | return 1; | |
1270 | } | |
1271 | ||
1272 | /* Supports checking bulk free of a constructed freelist */ | |
1273 | static noinline int free_debug_processing( | |
1274 | struct kmem_cache *s, struct page *page, | |
1275 | void *head, void *tail, int bulk_cnt, | |
1276 | unsigned long addr) | |
1277 | { | |
1278 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1279 | void *object = head; | |
1280 | int cnt = 0; | |
3f649ab7 | 1281 | unsigned long flags; |
becfda68 LA |
1282 | int ret = 0; |
1283 | ||
1284 | spin_lock_irqsave(&n->list_lock, flags); | |
1285 | slab_lock(page); | |
1286 | ||
1287 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1288 | if (!check_slab(s, page)) | |
1289 | goto out; | |
1290 | } | |
1291 | ||
1292 | next_object: | |
1293 | cnt++; | |
1294 | ||
1295 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1296 | if (!free_consistency_checks(s, page, object, addr)) | |
1297 | goto out; | |
81819f0f | 1298 | } |
3ec09742 | 1299 | |
3ec09742 CL |
1300 | if (s->flags & SLAB_STORE_USER) |
1301 | set_track(s, object, TRACK_FREE, addr); | |
1302 | trace(s, page, object, 0); | |
81084651 | 1303 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1304 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1305 | |
1306 | /* Reached end of constructed freelist yet? */ | |
1307 | if (object != tail) { | |
1308 | object = get_freepointer(s, object); | |
1309 | goto next_object; | |
1310 | } | |
804aa132 LA |
1311 | ret = 1; |
1312 | ||
5c2e4bbb | 1313 | out: |
81084651 JDB |
1314 | if (cnt != bulk_cnt) |
1315 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1316 | bulk_cnt, cnt); | |
1317 | ||
881db7fb | 1318 | slab_unlock(page); |
282acb43 | 1319 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1320 | if (!ret) |
1321 | slab_fix(s, "Object at 0x%p not freed", object); | |
1322 | return ret; | |
81819f0f CL |
1323 | } |
1324 | ||
e17f1dfb VB |
1325 | /* |
1326 | * Parse a block of slub_debug options. Blocks are delimited by ';' | |
1327 | * | |
1328 | * @str: start of block | |
1329 | * @flags: returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified | |
1330 | * @slabs: return start of list of slabs, or NULL when there's no list | |
1331 | * @init: assume this is initial parsing and not per-kmem-create parsing | |
1332 | * | |
1333 | * returns the start of next block if there's any, or NULL | |
1334 | */ | |
1335 | static char * | |
1336 | parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init) | |
41ecc55b | 1337 | { |
e17f1dfb | 1338 | bool higher_order_disable = false; |
f0630fff | 1339 | |
e17f1dfb VB |
1340 | /* Skip any completely empty blocks */ |
1341 | while (*str && *str == ';') | |
1342 | str++; | |
1343 | ||
1344 | if (*str == ',') { | |
f0630fff CL |
1345 | /* |
1346 | * No options but restriction on slabs. This means full | |
1347 | * debugging for slabs matching a pattern. | |
1348 | */ | |
e17f1dfb | 1349 | *flags = DEBUG_DEFAULT_FLAGS; |
f0630fff | 1350 | goto check_slabs; |
e17f1dfb VB |
1351 | } |
1352 | *flags = 0; | |
f0630fff | 1353 | |
e17f1dfb VB |
1354 | /* Determine which debug features should be switched on */ |
1355 | for (; *str && *str != ',' && *str != ';'; str++) { | |
f0630fff | 1356 | switch (tolower(*str)) { |
e17f1dfb VB |
1357 | case '-': |
1358 | *flags = 0; | |
1359 | break; | |
f0630fff | 1360 | case 'f': |
e17f1dfb | 1361 | *flags |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1362 | break; |
1363 | case 'z': | |
e17f1dfb | 1364 | *flags |= SLAB_RED_ZONE; |
f0630fff CL |
1365 | break; |
1366 | case 'p': | |
e17f1dfb | 1367 | *flags |= SLAB_POISON; |
f0630fff CL |
1368 | break; |
1369 | case 'u': | |
e17f1dfb | 1370 | *flags |= SLAB_STORE_USER; |
f0630fff CL |
1371 | break; |
1372 | case 't': | |
e17f1dfb | 1373 | *flags |= SLAB_TRACE; |
f0630fff | 1374 | break; |
4c13dd3b | 1375 | case 'a': |
e17f1dfb | 1376 | *flags |= SLAB_FAILSLAB; |
4c13dd3b | 1377 | break; |
08303a73 CA |
1378 | case 'o': |
1379 | /* | |
1380 | * Avoid enabling debugging on caches if its minimum | |
1381 | * order would increase as a result. | |
1382 | */ | |
e17f1dfb | 1383 | higher_order_disable = true; |
08303a73 | 1384 | break; |
f0630fff | 1385 | default: |
e17f1dfb VB |
1386 | if (init) |
1387 | pr_err("slub_debug option '%c' unknown. skipped\n", *str); | |
f0630fff | 1388 | } |
41ecc55b | 1389 | } |
f0630fff | 1390 | check_slabs: |
41ecc55b | 1391 | if (*str == ',') |
e17f1dfb VB |
1392 | *slabs = ++str; |
1393 | else | |
1394 | *slabs = NULL; | |
1395 | ||
1396 | /* Skip over the slab list */ | |
1397 | while (*str && *str != ';') | |
1398 | str++; | |
1399 | ||
1400 | /* Skip any completely empty blocks */ | |
1401 | while (*str && *str == ';') | |
1402 | str++; | |
1403 | ||
1404 | if (init && higher_order_disable) | |
1405 | disable_higher_order_debug = 1; | |
1406 | ||
1407 | if (*str) | |
1408 | return str; | |
1409 | else | |
1410 | return NULL; | |
1411 | } | |
1412 | ||
1413 | static int __init setup_slub_debug(char *str) | |
1414 | { | |
1415 | slab_flags_t flags; | |
1416 | char *saved_str; | |
1417 | char *slab_list; | |
1418 | bool global_slub_debug_changed = false; | |
1419 | bool slab_list_specified = false; | |
1420 | ||
1421 | slub_debug = DEBUG_DEFAULT_FLAGS; | |
1422 | if (*str++ != '=' || !*str) | |
1423 | /* | |
1424 | * No options specified. Switch on full debugging. | |
1425 | */ | |
1426 | goto out; | |
1427 | ||
1428 | saved_str = str; | |
1429 | while (str) { | |
1430 | str = parse_slub_debug_flags(str, &flags, &slab_list, true); | |
1431 | ||
1432 | if (!slab_list) { | |
1433 | slub_debug = flags; | |
1434 | global_slub_debug_changed = true; | |
1435 | } else { | |
1436 | slab_list_specified = true; | |
1437 | } | |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * For backwards compatibility, a single list of flags with list of | |
1442 | * slabs means debugging is only enabled for those slabs, so the global | |
1443 | * slub_debug should be 0. We can extended that to multiple lists as | |
1444 | * long as there is no option specifying flags without a slab list. | |
1445 | */ | |
1446 | if (slab_list_specified) { | |
1447 | if (!global_slub_debug_changed) | |
1448 | slub_debug = 0; | |
1449 | slub_debug_string = saved_str; | |
1450 | } | |
f0630fff | 1451 | out: |
ca0cab65 VB |
1452 | if (slub_debug != 0 || slub_debug_string) |
1453 | static_branch_enable(&slub_debug_enabled); | |
02ac47d0 SB |
1454 | else |
1455 | static_branch_disable(&slub_debug_enabled); | |
6471384a AP |
1456 | if ((static_branch_unlikely(&init_on_alloc) || |
1457 | static_branch_unlikely(&init_on_free)) && | |
1458 | (slub_debug & SLAB_POISON)) | |
1459 | pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n"); | |
41ecc55b CL |
1460 | return 1; |
1461 | } | |
1462 | ||
1463 | __setup("slub_debug", setup_slub_debug); | |
1464 | ||
c5fd3ca0 AT |
1465 | /* |
1466 | * kmem_cache_flags - apply debugging options to the cache | |
1467 | * @object_size: the size of an object without meta data | |
1468 | * @flags: flags to set | |
1469 | * @name: name of the cache | |
c5fd3ca0 AT |
1470 | * |
1471 | * Debug option(s) are applied to @flags. In addition to the debug | |
1472 | * option(s), if a slab name (or multiple) is specified i.e. | |
1473 | * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ... | |
1474 | * then only the select slabs will receive the debug option(s). | |
1475 | */ | |
0293d1fd | 1476 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 1477 | slab_flags_t flags, const char *name) |
41ecc55b | 1478 | { |
c5fd3ca0 AT |
1479 | char *iter; |
1480 | size_t len; | |
e17f1dfb VB |
1481 | char *next_block; |
1482 | slab_flags_t block_flags; | |
ca220593 JB |
1483 | slab_flags_t slub_debug_local = slub_debug; |
1484 | ||
1485 | /* | |
1486 | * If the slab cache is for debugging (e.g. kmemleak) then | |
1487 | * don't store user (stack trace) information by default, | |
1488 | * but let the user enable it via the command line below. | |
1489 | */ | |
1490 | if (flags & SLAB_NOLEAKTRACE) | |
1491 | slub_debug_local &= ~SLAB_STORE_USER; | |
c5fd3ca0 | 1492 | |
c5fd3ca0 | 1493 | len = strlen(name); |
e17f1dfb VB |
1494 | next_block = slub_debug_string; |
1495 | /* Go through all blocks of debug options, see if any matches our slab's name */ | |
1496 | while (next_block) { | |
1497 | next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false); | |
1498 | if (!iter) | |
1499 | continue; | |
1500 | /* Found a block that has a slab list, search it */ | |
1501 | while (*iter) { | |
1502 | char *end, *glob; | |
1503 | size_t cmplen; | |
1504 | ||
1505 | end = strchrnul(iter, ','); | |
1506 | if (next_block && next_block < end) | |
1507 | end = next_block - 1; | |
1508 | ||
1509 | glob = strnchr(iter, end - iter, '*'); | |
1510 | if (glob) | |
1511 | cmplen = glob - iter; | |
1512 | else | |
1513 | cmplen = max_t(size_t, len, (end - iter)); | |
c5fd3ca0 | 1514 | |
e17f1dfb VB |
1515 | if (!strncmp(name, iter, cmplen)) { |
1516 | flags |= block_flags; | |
1517 | return flags; | |
1518 | } | |
c5fd3ca0 | 1519 | |
e17f1dfb VB |
1520 | if (!*end || *end == ';') |
1521 | break; | |
1522 | iter = end + 1; | |
c5fd3ca0 | 1523 | } |
c5fd3ca0 | 1524 | } |
ba0268a8 | 1525 | |
ca220593 | 1526 | return flags | slub_debug_local; |
41ecc55b | 1527 | } |
b4a64718 | 1528 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1529 | static inline void setup_object_debug(struct kmem_cache *s, |
1530 | struct page *page, void *object) {} | |
a50b854e MWO |
1531 | static inline |
1532 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) {} | |
41ecc55b | 1533 | |
3ec09742 | 1534 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1535 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1536 | |
282acb43 | 1537 | static inline int free_debug_processing( |
81084651 JDB |
1538 | struct kmem_cache *s, struct page *page, |
1539 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1540 | unsigned long addr) { return 0; } |
41ecc55b | 1541 | |
41ecc55b CL |
1542 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1543 | { return 1; } | |
1544 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1545 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1546 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1547 | struct page *page) {} | |
c65c1877 PZ |
1548 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1549 | struct page *page) {} | |
0293d1fd | 1550 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 1551 | slab_flags_t flags, const char *name) |
ba0268a8 CL |
1552 | { |
1553 | return flags; | |
1554 | } | |
41ecc55b | 1555 | #define slub_debug 0 |
0f389ec6 | 1556 | |
fdaa45e9 IM |
1557 | #define disable_higher_order_debug 0 |
1558 | ||
0f389ec6 CL |
1559 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1560 | { return 0; } | |
26c02cf0 AB |
1561 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1562 | { return 0; } | |
205ab99d CL |
1563 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1564 | int objects) {} | |
1565 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1566 | int objects) {} | |
7d550c56 | 1567 | |
52f23478 | 1568 | static bool freelist_corrupted(struct kmem_cache *s, struct page *page, |
dc07a728 | 1569 | void **freelist, void *nextfree) |
52f23478 DZ |
1570 | { |
1571 | return false; | |
1572 | } | |
02e72cc6 AR |
1573 | #endif /* CONFIG_SLUB_DEBUG */ |
1574 | ||
1575 | /* | |
1576 | * Hooks for other subsystems that check memory allocations. In a typical | |
1577 | * production configuration these hooks all should produce no code at all. | |
1578 | */ | |
0116523c | 1579 | static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
d56791b3 | 1580 | { |
53128245 | 1581 | ptr = kasan_kmalloc_large(ptr, size, flags); |
a2f77575 | 1582 | /* As ptr might get tagged, call kmemleak hook after KASAN. */ |
d56791b3 | 1583 | kmemleak_alloc(ptr, size, 1, flags); |
53128245 | 1584 | return ptr; |
d56791b3 RB |
1585 | } |
1586 | ||
ee3ce779 | 1587 | static __always_inline void kfree_hook(void *x) |
d56791b3 RB |
1588 | { |
1589 | kmemleak_free(x); | |
027b37b5 | 1590 | kasan_kfree_large(x); |
d56791b3 RB |
1591 | } |
1592 | ||
d57a964e AK |
1593 | static __always_inline bool slab_free_hook(struct kmem_cache *s, |
1594 | void *x, bool init) | |
d56791b3 RB |
1595 | { |
1596 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1597 | |
02e72cc6 AR |
1598 | /* |
1599 | * Trouble is that we may no longer disable interrupts in the fast path | |
1600 | * So in order to make the debug calls that expect irqs to be | |
1601 | * disabled we need to disable interrupts temporarily. | |
1602 | */ | |
4675ff05 | 1603 | #ifdef CONFIG_LOCKDEP |
02e72cc6 AR |
1604 | { |
1605 | unsigned long flags; | |
1606 | ||
1607 | local_irq_save(flags); | |
02e72cc6 AR |
1608 | debug_check_no_locks_freed(x, s->object_size); |
1609 | local_irq_restore(flags); | |
1610 | } | |
1611 | #endif | |
1612 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1613 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1614 | |
cfbe1636 ME |
1615 | /* Use KCSAN to help debug racy use-after-free. */ |
1616 | if (!(s->flags & SLAB_TYPESAFE_BY_RCU)) | |
1617 | __kcsan_check_access(x, s->object_size, | |
1618 | KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT); | |
1619 | ||
d57a964e AK |
1620 | /* |
1621 | * As memory initialization might be integrated into KASAN, | |
1622 | * kasan_slab_free and initialization memset's must be | |
1623 | * kept together to avoid discrepancies in behavior. | |
1624 | * | |
1625 | * The initialization memset's clear the object and the metadata, | |
1626 | * but don't touch the SLAB redzone. | |
1627 | */ | |
1628 | if (init) { | |
1629 | int rsize; | |
1630 | ||
1631 | if (!kasan_has_integrated_init()) | |
1632 | memset(kasan_reset_tag(x), 0, s->object_size); | |
1633 | rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0; | |
1634 | memset((char *)kasan_reset_tag(x) + s->inuse, 0, | |
1635 | s->size - s->inuse - rsize); | |
1636 | } | |
1637 | /* KASAN might put x into memory quarantine, delaying its reuse. */ | |
1638 | return kasan_slab_free(s, x, init); | |
02e72cc6 | 1639 | } |
205ab99d | 1640 | |
c3895391 AK |
1641 | static inline bool slab_free_freelist_hook(struct kmem_cache *s, |
1642 | void **head, void **tail) | |
81084651 | 1643 | { |
6471384a AP |
1644 | |
1645 | void *object; | |
1646 | void *next = *head; | |
1647 | void *old_tail = *tail ? *tail : *head; | |
6471384a | 1648 | |
b89fb5ef | 1649 | if (is_kfence_address(next)) { |
d57a964e | 1650 | slab_free_hook(s, next, false); |
b89fb5ef AP |
1651 | return true; |
1652 | } | |
1653 | ||
aea4df4c LA |
1654 | /* Head and tail of the reconstructed freelist */ |
1655 | *head = NULL; | |
1656 | *tail = NULL; | |
1b7e816f | 1657 | |
aea4df4c LA |
1658 | do { |
1659 | object = next; | |
1660 | next = get_freepointer(s, object); | |
1661 | ||
c3895391 | 1662 | /* If object's reuse doesn't have to be delayed */ |
d57a964e | 1663 | if (!slab_free_hook(s, object, slab_want_init_on_free(s))) { |
c3895391 AK |
1664 | /* Move object to the new freelist */ |
1665 | set_freepointer(s, object, *head); | |
1666 | *head = object; | |
1667 | if (!*tail) | |
1668 | *tail = object; | |
1669 | } | |
1670 | } while (object != old_tail); | |
1671 | ||
1672 | if (*head == *tail) | |
1673 | *tail = NULL; | |
1674 | ||
1675 | return *head != NULL; | |
81084651 JDB |
1676 | } |
1677 | ||
4d176711 | 1678 | static void *setup_object(struct kmem_cache *s, struct page *page, |
588f8ba9 TG |
1679 | void *object) |
1680 | { | |
1681 | setup_object_debug(s, page, object); | |
4d176711 | 1682 | object = kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1683 | if (unlikely(s->ctor)) { |
1684 | kasan_unpoison_object_data(s, object); | |
1685 | s->ctor(object); | |
1686 | kasan_poison_object_data(s, object); | |
1687 | } | |
4d176711 | 1688 | return object; |
588f8ba9 TG |
1689 | } |
1690 | ||
81819f0f CL |
1691 | /* |
1692 | * Slab allocation and freeing | |
1693 | */ | |
5dfb4175 VD |
1694 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1695 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1696 | { |
5dfb4175 | 1697 | struct page *page; |
19af27af | 1698 | unsigned int order = oo_order(oo); |
65c3376a | 1699 | |
2154a336 | 1700 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1701 | page = alloc_pages(flags, order); |
65c3376a | 1702 | else |
96db800f | 1703 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1704 | |
5dfb4175 | 1705 | return page; |
65c3376a CL |
1706 | } |
1707 | ||
210e7a43 TG |
1708 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1709 | /* Pre-initialize the random sequence cache */ | |
1710 | static int init_cache_random_seq(struct kmem_cache *s) | |
1711 | { | |
19af27af | 1712 | unsigned int count = oo_objects(s->oo); |
210e7a43 | 1713 | int err; |
210e7a43 | 1714 | |
a810007a SR |
1715 | /* Bailout if already initialised */ |
1716 | if (s->random_seq) | |
1717 | return 0; | |
1718 | ||
210e7a43 TG |
1719 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1720 | if (err) { | |
1721 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1722 | s->name); | |
1723 | return err; | |
1724 | } | |
1725 | ||
1726 | /* Transform to an offset on the set of pages */ | |
1727 | if (s->random_seq) { | |
19af27af AD |
1728 | unsigned int i; |
1729 | ||
210e7a43 TG |
1730 | for (i = 0; i < count; i++) |
1731 | s->random_seq[i] *= s->size; | |
1732 | } | |
1733 | return 0; | |
1734 | } | |
1735 | ||
1736 | /* Initialize each random sequence freelist per cache */ | |
1737 | static void __init init_freelist_randomization(void) | |
1738 | { | |
1739 | struct kmem_cache *s; | |
1740 | ||
1741 | mutex_lock(&slab_mutex); | |
1742 | ||
1743 | list_for_each_entry(s, &slab_caches, list) | |
1744 | init_cache_random_seq(s); | |
1745 | ||
1746 | mutex_unlock(&slab_mutex); | |
1747 | } | |
1748 | ||
1749 | /* Get the next entry on the pre-computed freelist randomized */ | |
1750 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1751 | unsigned long *pos, void *start, | |
1752 | unsigned long page_limit, | |
1753 | unsigned long freelist_count) | |
1754 | { | |
1755 | unsigned int idx; | |
1756 | ||
1757 | /* | |
1758 | * If the target page allocation failed, the number of objects on the | |
1759 | * page might be smaller than the usual size defined by the cache. | |
1760 | */ | |
1761 | do { | |
1762 | idx = s->random_seq[*pos]; | |
1763 | *pos += 1; | |
1764 | if (*pos >= freelist_count) | |
1765 | *pos = 0; | |
1766 | } while (unlikely(idx >= page_limit)); | |
1767 | ||
1768 | return (char *)start + idx; | |
1769 | } | |
1770 | ||
1771 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1772 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1773 | { | |
1774 | void *start; | |
1775 | void *cur; | |
1776 | void *next; | |
1777 | unsigned long idx, pos, page_limit, freelist_count; | |
1778 | ||
1779 | if (page->objects < 2 || !s->random_seq) | |
1780 | return false; | |
1781 | ||
1782 | freelist_count = oo_objects(s->oo); | |
1783 | pos = get_random_int() % freelist_count; | |
1784 | ||
1785 | page_limit = page->objects * s->size; | |
1786 | start = fixup_red_left(s, page_address(page)); | |
1787 | ||
1788 | /* First entry is used as the base of the freelist */ | |
1789 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1790 | freelist_count); | |
4d176711 | 1791 | cur = setup_object(s, page, cur); |
210e7a43 TG |
1792 | page->freelist = cur; |
1793 | ||
1794 | for (idx = 1; idx < page->objects; idx++) { | |
210e7a43 TG |
1795 | next = next_freelist_entry(s, page, &pos, start, page_limit, |
1796 | freelist_count); | |
4d176711 | 1797 | next = setup_object(s, page, next); |
210e7a43 TG |
1798 | set_freepointer(s, cur, next); |
1799 | cur = next; | |
1800 | } | |
210e7a43 TG |
1801 | set_freepointer(s, cur, NULL); |
1802 | ||
1803 | return true; | |
1804 | } | |
1805 | #else | |
1806 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1807 | { | |
1808 | return 0; | |
1809 | } | |
1810 | static inline void init_freelist_randomization(void) { } | |
1811 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1812 | { | |
1813 | return false; | |
1814 | } | |
1815 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1816 | ||
81819f0f CL |
1817 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1818 | { | |
06428780 | 1819 | struct page *page; |
834f3d11 | 1820 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1821 | gfp_t alloc_gfp; |
4d176711 | 1822 | void *start, *p, *next; |
a50b854e | 1823 | int idx; |
210e7a43 | 1824 | bool shuffle; |
81819f0f | 1825 | |
7e0528da CL |
1826 | flags &= gfp_allowed_mask; |
1827 | ||
d0164adc | 1828 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1829 | local_irq_enable(); |
1830 | ||
b7a49f0d | 1831 | flags |= s->allocflags; |
e12ba74d | 1832 | |
ba52270d PE |
1833 | /* |
1834 | * Let the initial higher-order allocation fail under memory pressure | |
1835 | * so we fall-back to the minimum order allocation. | |
1836 | */ | |
1837 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1838 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1839 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1840 | |
5dfb4175 | 1841 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1842 | if (unlikely(!page)) { |
1843 | oo = s->min; | |
80c3a998 | 1844 | alloc_gfp = flags; |
65c3376a CL |
1845 | /* |
1846 | * Allocation may have failed due to fragmentation. | |
1847 | * Try a lower order alloc if possible | |
1848 | */ | |
5dfb4175 | 1849 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1850 | if (unlikely(!page)) |
1851 | goto out; | |
1852 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1853 | } |
5a896d9e | 1854 | |
834f3d11 | 1855 | page->objects = oo_objects(oo); |
81819f0f | 1856 | |
2e9bd483 | 1857 | account_slab_page(page, oo_order(oo), s, flags); |
1f3147b4 | 1858 | |
1b4f59e3 | 1859 | page->slab_cache = s; |
c03f94cc | 1860 | __SetPageSlab(page); |
2f064f34 | 1861 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1862 | SetPageSlabPfmemalloc(page); |
81819f0f | 1863 | |
a7101224 | 1864 | kasan_poison_slab(page); |
81819f0f | 1865 | |
a7101224 | 1866 | start = page_address(page); |
81819f0f | 1867 | |
a50b854e | 1868 | setup_page_debug(s, page, start); |
0316bec2 | 1869 | |
210e7a43 TG |
1870 | shuffle = shuffle_freelist(s, page); |
1871 | ||
1872 | if (!shuffle) { | |
4d176711 AK |
1873 | start = fixup_red_left(s, start); |
1874 | start = setup_object(s, page, start); | |
1875 | page->freelist = start; | |
18e50661 AK |
1876 | for (idx = 0, p = start; idx < page->objects - 1; idx++) { |
1877 | next = p + s->size; | |
1878 | next = setup_object(s, page, next); | |
1879 | set_freepointer(s, p, next); | |
1880 | p = next; | |
1881 | } | |
1882 | set_freepointer(s, p, NULL); | |
81819f0f | 1883 | } |
81819f0f | 1884 | |
e6e82ea1 | 1885 | page->inuse = page->objects; |
8cb0a506 | 1886 | page->frozen = 1; |
588f8ba9 | 1887 | |
81819f0f | 1888 | out: |
d0164adc | 1889 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1890 | local_irq_disable(); |
1891 | if (!page) | |
1892 | return NULL; | |
1893 | ||
588f8ba9 TG |
1894 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1895 | ||
81819f0f CL |
1896 | return page; |
1897 | } | |
1898 | ||
588f8ba9 TG |
1899 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1900 | { | |
44405099 LL |
1901 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) |
1902 | flags = kmalloc_fix_flags(flags); | |
588f8ba9 TG |
1903 | |
1904 | return allocate_slab(s, | |
1905 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1906 | } | |
1907 | ||
81819f0f CL |
1908 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1909 | { | |
834f3d11 CL |
1910 | int order = compound_order(page); |
1911 | int pages = 1 << order; | |
81819f0f | 1912 | |
8fc8d666 | 1913 | if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) { |
81819f0f CL |
1914 | void *p; |
1915 | ||
1916 | slab_pad_check(s, page); | |
224a88be CL |
1917 | for_each_object(p, s, page_address(page), |
1918 | page->objects) | |
f7cb1933 | 1919 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1920 | } |
1921 | ||
072bb0aa | 1922 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1923 | __ClearPageSlab(page); |
0c06dd75 VB |
1924 | /* In union with page->mapping where page allocator expects NULL */ |
1925 | page->slab_cache = NULL; | |
1eb5ac64 NP |
1926 | if (current->reclaim_state) |
1927 | current->reclaim_state->reclaimed_slab += pages; | |
74d555be | 1928 | unaccount_slab_page(page, order, s); |
27ee57c9 | 1929 | __free_pages(page, order); |
81819f0f CL |
1930 | } |
1931 | ||
1932 | static void rcu_free_slab(struct rcu_head *h) | |
1933 | { | |
bf68c214 | 1934 | struct page *page = container_of(h, struct page, rcu_head); |
da9a638c | 1935 | |
1b4f59e3 | 1936 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1937 | } |
1938 | ||
1939 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1940 | { | |
5f0d5a3a | 1941 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
bf68c214 | 1942 | call_rcu(&page->rcu_head, rcu_free_slab); |
81819f0f CL |
1943 | } else |
1944 | __free_slab(s, page); | |
1945 | } | |
1946 | ||
1947 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1948 | { | |
205ab99d | 1949 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1950 | free_slab(s, page); |
1951 | } | |
1952 | ||
1953 | /* | |
5cc6eee8 | 1954 | * Management of partially allocated slabs. |
81819f0f | 1955 | */ |
1e4dd946 SR |
1956 | static inline void |
1957 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1958 | { |
e95eed57 | 1959 | n->nr_partial++; |
136333d1 | 1960 | if (tail == DEACTIVATE_TO_TAIL) |
916ac052 | 1961 | list_add_tail(&page->slab_list, &n->partial); |
7c2e132c | 1962 | else |
916ac052 | 1963 | list_add(&page->slab_list, &n->partial); |
81819f0f CL |
1964 | } |
1965 | ||
1e4dd946 SR |
1966 | static inline void add_partial(struct kmem_cache_node *n, |
1967 | struct page *page, int tail) | |
62e346a8 | 1968 | { |
c65c1877 | 1969 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1970 | __add_partial(n, page, tail); |
1971 | } | |
c65c1877 | 1972 | |
1e4dd946 SR |
1973 | static inline void remove_partial(struct kmem_cache_node *n, |
1974 | struct page *page) | |
1975 | { | |
1976 | lockdep_assert_held(&n->list_lock); | |
916ac052 | 1977 | list_del(&page->slab_list); |
52b4b950 | 1978 | n->nr_partial--; |
1e4dd946 SR |
1979 | } |
1980 | ||
81819f0f | 1981 | /* |
7ced3719 CL |
1982 | * Remove slab from the partial list, freeze it and |
1983 | * return the pointer to the freelist. | |
81819f0f | 1984 | * |
497b66f2 | 1985 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1986 | */ |
497b66f2 | 1987 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1988 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1989 | int mode, int *objects) |
81819f0f | 1990 | { |
2cfb7455 CL |
1991 | void *freelist; |
1992 | unsigned long counters; | |
1993 | struct page new; | |
1994 | ||
c65c1877 PZ |
1995 | lockdep_assert_held(&n->list_lock); |
1996 | ||
2cfb7455 CL |
1997 | /* |
1998 | * Zap the freelist and set the frozen bit. | |
1999 | * The old freelist is the list of objects for the | |
2000 | * per cpu allocation list. | |
2001 | */ | |
7ced3719 CL |
2002 | freelist = page->freelist; |
2003 | counters = page->counters; | |
2004 | new.counters = counters; | |
633b0764 | 2005 | *objects = new.objects - new.inuse; |
23910c50 | 2006 | if (mode) { |
7ced3719 | 2007 | new.inuse = page->objects; |
23910c50 PE |
2008 | new.freelist = NULL; |
2009 | } else { | |
2010 | new.freelist = freelist; | |
2011 | } | |
2cfb7455 | 2012 | |
a0132ac0 | 2013 | VM_BUG_ON(new.frozen); |
7ced3719 | 2014 | new.frozen = 1; |
2cfb7455 | 2015 | |
7ced3719 | 2016 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 2017 | freelist, counters, |
02d7633f | 2018 | new.freelist, new.counters, |
7ced3719 | 2019 | "acquire_slab")) |
7ced3719 | 2020 | return NULL; |
2cfb7455 CL |
2021 | |
2022 | remove_partial(n, page); | |
7ced3719 | 2023 | WARN_ON(!freelist); |
49e22585 | 2024 | return freelist; |
81819f0f CL |
2025 | } |
2026 | ||
633b0764 | 2027 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 2028 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 2029 | |
81819f0f | 2030 | /* |
672bba3a | 2031 | * Try to allocate a partial slab from a specific node. |
81819f0f | 2032 | */ |
8ba00bb6 JK |
2033 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
2034 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 2035 | { |
49e22585 CL |
2036 | struct page *page, *page2; |
2037 | void *object = NULL; | |
e5d9998f | 2038 | unsigned int available = 0; |
633b0764 | 2039 | int objects; |
81819f0f CL |
2040 | |
2041 | /* | |
2042 | * Racy check. If we mistakenly see no partial slabs then we | |
2043 | * just allocate an empty slab. If we mistakenly try to get a | |
70b6d25e | 2044 | * partial slab and there is none available then get_partial() |
672bba3a | 2045 | * will return NULL. |
81819f0f CL |
2046 | */ |
2047 | if (!n || !n->nr_partial) | |
2048 | return NULL; | |
2049 | ||
2050 | spin_lock(&n->list_lock); | |
916ac052 | 2051 | list_for_each_entry_safe(page, page2, &n->partial, slab_list) { |
8ba00bb6 | 2052 | void *t; |
49e22585 | 2053 | |
8ba00bb6 JK |
2054 | if (!pfmemalloc_match(page, flags)) |
2055 | continue; | |
2056 | ||
633b0764 | 2057 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 | 2058 | if (!t) |
9b1ea29b | 2059 | break; |
49e22585 | 2060 | |
633b0764 | 2061 | available += objects; |
12d79634 | 2062 | if (!object) { |
49e22585 | 2063 | c->page = page; |
49e22585 | 2064 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 2065 | object = t; |
49e22585 | 2066 | } else { |
633b0764 | 2067 | put_cpu_partial(s, page, 0); |
8028dcea | 2068 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 2069 | } |
345c905d | 2070 | if (!kmem_cache_has_cpu_partial(s) |
e6d0e1dc | 2071 | || available > slub_cpu_partial(s) / 2) |
49e22585 CL |
2072 | break; |
2073 | ||
497b66f2 | 2074 | } |
81819f0f | 2075 | spin_unlock(&n->list_lock); |
497b66f2 | 2076 | return object; |
81819f0f CL |
2077 | } |
2078 | ||
2079 | /* | |
672bba3a | 2080 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 2081 | */ |
de3ec035 | 2082 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 2083 | struct kmem_cache_cpu *c) |
81819f0f CL |
2084 | { |
2085 | #ifdef CONFIG_NUMA | |
2086 | struct zonelist *zonelist; | |
dd1a239f | 2087 | struct zoneref *z; |
54a6eb5c | 2088 | struct zone *zone; |
97a225e6 | 2089 | enum zone_type highest_zoneidx = gfp_zone(flags); |
497b66f2 | 2090 | void *object; |
cc9a6c87 | 2091 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
2092 | |
2093 | /* | |
672bba3a CL |
2094 | * The defrag ratio allows a configuration of the tradeoffs between |
2095 | * inter node defragmentation and node local allocations. A lower | |
2096 | * defrag_ratio increases the tendency to do local allocations | |
2097 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 2098 | * |
672bba3a CL |
2099 | * If the defrag_ratio is set to 0 then kmalloc() always |
2100 | * returns node local objects. If the ratio is higher then kmalloc() | |
2101 | * may return off node objects because partial slabs are obtained | |
2102 | * from other nodes and filled up. | |
81819f0f | 2103 | * |
43efd3ea LP |
2104 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
2105 | * (which makes defrag_ratio = 1000) then every (well almost) | |
2106 | * allocation will first attempt to defrag slab caches on other nodes. | |
2107 | * This means scanning over all nodes to look for partial slabs which | |
2108 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 2109 | * with available objects. |
81819f0f | 2110 | */ |
9824601e CL |
2111 | if (!s->remote_node_defrag_ratio || |
2112 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
2113 | return NULL; |
2114 | ||
cc9a6c87 | 2115 | do { |
d26914d1 | 2116 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 2117 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
97a225e6 | 2118 | for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) { |
cc9a6c87 MG |
2119 | struct kmem_cache_node *n; |
2120 | ||
2121 | n = get_node(s, zone_to_nid(zone)); | |
2122 | ||
dee2f8aa | 2123 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 2124 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 2125 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
2126 | if (object) { |
2127 | /* | |
d26914d1 MG |
2128 | * Don't check read_mems_allowed_retry() |
2129 | * here - if mems_allowed was updated in | |
2130 | * parallel, that was a harmless race | |
2131 | * between allocation and the cpuset | |
2132 | * update | |
cc9a6c87 | 2133 | */ |
cc9a6c87 MG |
2134 | return object; |
2135 | } | |
c0ff7453 | 2136 | } |
81819f0f | 2137 | } |
d26914d1 | 2138 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
6dfd1b65 | 2139 | #endif /* CONFIG_NUMA */ |
81819f0f CL |
2140 | return NULL; |
2141 | } | |
2142 | ||
2143 | /* | |
2144 | * Get a partial page, lock it and return it. | |
2145 | */ | |
497b66f2 | 2146 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 2147 | struct kmem_cache_cpu *c) |
81819f0f | 2148 | { |
497b66f2 | 2149 | void *object; |
a561ce00 JK |
2150 | int searchnode = node; |
2151 | ||
2152 | if (node == NUMA_NO_NODE) | |
2153 | searchnode = numa_mem_id(); | |
81819f0f | 2154 | |
8ba00bb6 | 2155 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
2156 | if (object || node != NUMA_NO_NODE) |
2157 | return object; | |
81819f0f | 2158 | |
acd19fd1 | 2159 | return get_any_partial(s, flags, c); |
81819f0f CL |
2160 | } |
2161 | ||
923717cb | 2162 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2163 | /* |
0d645ed1 | 2164 | * Calculate the next globally unique transaction for disambiguation |
8a5ec0ba CL |
2165 | * during cmpxchg. The transactions start with the cpu number and are then |
2166 | * incremented by CONFIG_NR_CPUS. | |
2167 | */ | |
2168 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
2169 | #else | |
2170 | /* | |
2171 | * No preemption supported therefore also no need to check for | |
2172 | * different cpus. | |
2173 | */ | |
2174 | #define TID_STEP 1 | |
2175 | #endif | |
2176 | ||
2177 | static inline unsigned long next_tid(unsigned long tid) | |
2178 | { | |
2179 | return tid + TID_STEP; | |
2180 | } | |
2181 | ||
9d5f0be0 | 2182 | #ifdef SLUB_DEBUG_CMPXCHG |
8a5ec0ba CL |
2183 | static inline unsigned int tid_to_cpu(unsigned long tid) |
2184 | { | |
2185 | return tid % TID_STEP; | |
2186 | } | |
2187 | ||
2188 | static inline unsigned long tid_to_event(unsigned long tid) | |
2189 | { | |
2190 | return tid / TID_STEP; | |
2191 | } | |
9d5f0be0 | 2192 | #endif |
8a5ec0ba CL |
2193 | |
2194 | static inline unsigned int init_tid(int cpu) | |
2195 | { | |
2196 | return cpu; | |
2197 | } | |
2198 | ||
2199 | static inline void note_cmpxchg_failure(const char *n, | |
2200 | const struct kmem_cache *s, unsigned long tid) | |
2201 | { | |
2202 | #ifdef SLUB_DEBUG_CMPXCHG | |
2203 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2204 | ||
f9f58285 | 2205 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba | 2206 | |
923717cb | 2207 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2208 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) |
f9f58285 | 2209 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
2210 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
2211 | else | |
2212 | #endif | |
2213 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 2214 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
2215 | tid_to_event(tid), tid_to_event(actual_tid)); |
2216 | else | |
f9f58285 | 2217 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
2218 | actual_tid, tid, next_tid(tid)); |
2219 | #endif | |
4fdccdfb | 2220 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
2221 | } |
2222 | ||
788e1aad | 2223 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 2224 | { |
8a5ec0ba CL |
2225 | int cpu; |
2226 | ||
2227 | for_each_possible_cpu(cpu) | |
2228 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 2229 | } |
2cfb7455 | 2230 | |
81819f0f CL |
2231 | /* |
2232 | * Remove the cpu slab | |
2233 | */ | |
d0e0ac97 | 2234 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
d4ff6d35 | 2235 | void *freelist, struct kmem_cache_cpu *c) |
81819f0f | 2236 | { |
2cfb7455 | 2237 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 | 2238 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
d930ff03 | 2239 | int lock = 0, free_delta = 0; |
2cfb7455 | 2240 | enum slab_modes l = M_NONE, m = M_NONE; |
d930ff03 | 2241 | void *nextfree, *freelist_iter, *freelist_tail; |
136333d1 | 2242 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
2243 | struct page new; |
2244 | struct page old; | |
2245 | ||
2246 | if (page->freelist) { | |
84e554e6 | 2247 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2248 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2249 | } |
2250 | ||
894b8788 | 2251 | /* |
d930ff03 VB |
2252 | * Stage one: Count the objects on cpu's freelist as free_delta and |
2253 | * remember the last object in freelist_tail for later splicing. | |
2cfb7455 | 2254 | */ |
d930ff03 VB |
2255 | freelist_tail = NULL; |
2256 | freelist_iter = freelist; | |
2257 | while (freelist_iter) { | |
2258 | nextfree = get_freepointer(s, freelist_iter); | |
2cfb7455 | 2259 | |
52f23478 DZ |
2260 | /* |
2261 | * If 'nextfree' is invalid, it is possible that the object at | |
d930ff03 VB |
2262 | * 'freelist_iter' is already corrupted. So isolate all objects |
2263 | * starting at 'freelist_iter' by skipping them. | |
52f23478 | 2264 | */ |
d930ff03 | 2265 | if (freelist_corrupted(s, page, &freelist_iter, nextfree)) |
52f23478 DZ |
2266 | break; |
2267 | ||
d930ff03 VB |
2268 | freelist_tail = freelist_iter; |
2269 | free_delta++; | |
2cfb7455 | 2270 | |
d930ff03 | 2271 | freelist_iter = nextfree; |
2cfb7455 CL |
2272 | } |
2273 | ||
894b8788 | 2274 | /* |
d930ff03 VB |
2275 | * Stage two: Unfreeze the page while splicing the per-cpu |
2276 | * freelist to the head of page's freelist. | |
2277 | * | |
2278 | * Ensure that the page is unfrozen while the list presence | |
2279 | * reflects the actual number of objects during unfreeze. | |
2cfb7455 CL |
2280 | * |
2281 | * We setup the list membership and then perform a cmpxchg | |
2282 | * with the count. If there is a mismatch then the page | |
2283 | * is not unfrozen but the page is on the wrong list. | |
2284 | * | |
2285 | * Then we restart the process which may have to remove | |
2286 | * the page from the list that we just put it on again | |
2287 | * because the number of objects in the slab may have | |
2288 | * changed. | |
894b8788 | 2289 | */ |
2cfb7455 | 2290 | redo: |
894b8788 | 2291 | |
d930ff03 VB |
2292 | old.freelist = READ_ONCE(page->freelist); |
2293 | old.counters = READ_ONCE(page->counters); | |
a0132ac0 | 2294 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2295 | |
2cfb7455 CL |
2296 | /* Determine target state of the slab */ |
2297 | new.counters = old.counters; | |
d930ff03 VB |
2298 | if (freelist_tail) { |
2299 | new.inuse -= free_delta; | |
2300 | set_freepointer(s, freelist_tail, old.freelist); | |
2cfb7455 CL |
2301 | new.freelist = freelist; |
2302 | } else | |
2303 | new.freelist = old.freelist; | |
2304 | ||
2305 | new.frozen = 0; | |
2306 | ||
8a5b20ae | 2307 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2308 | m = M_FREE; |
2309 | else if (new.freelist) { | |
2310 | m = M_PARTIAL; | |
2311 | if (!lock) { | |
2312 | lock = 1; | |
2313 | /* | |
8bb4e7a2 | 2314 | * Taking the spinlock removes the possibility |
2cfb7455 CL |
2315 | * that acquire_slab() will see a slab page that |
2316 | * is frozen | |
2317 | */ | |
2318 | spin_lock(&n->list_lock); | |
2319 | } | |
2320 | } else { | |
2321 | m = M_FULL; | |
965c4848 | 2322 | if (kmem_cache_debug_flags(s, SLAB_STORE_USER) && !lock) { |
2cfb7455 CL |
2323 | lock = 1; |
2324 | /* | |
2325 | * This also ensures that the scanning of full | |
2326 | * slabs from diagnostic functions will not see | |
2327 | * any frozen slabs. | |
2328 | */ | |
2329 | spin_lock(&n->list_lock); | |
2330 | } | |
2331 | } | |
2332 | ||
2333 | if (l != m) { | |
2cfb7455 | 2334 | if (l == M_PARTIAL) |
2cfb7455 | 2335 | remove_partial(n, page); |
2cfb7455 | 2336 | else if (l == M_FULL) |
c65c1877 | 2337 | remove_full(s, n, page); |
2cfb7455 | 2338 | |
88349a28 | 2339 | if (m == M_PARTIAL) |
2cfb7455 | 2340 | add_partial(n, page, tail); |
88349a28 | 2341 | else if (m == M_FULL) |
2cfb7455 | 2342 | add_full(s, n, page); |
2cfb7455 CL |
2343 | } |
2344 | ||
2345 | l = m; | |
1d07171c | 2346 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2347 | old.freelist, old.counters, |
2348 | new.freelist, new.counters, | |
2349 | "unfreezing slab")) | |
2350 | goto redo; | |
2351 | ||
2cfb7455 CL |
2352 | if (lock) |
2353 | spin_unlock(&n->list_lock); | |
2354 | ||
88349a28 WY |
2355 | if (m == M_PARTIAL) |
2356 | stat(s, tail); | |
2357 | else if (m == M_FULL) | |
2358 | stat(s, DEACTIVATE_FULL); | |
2359 | else if (m == M_FREE) { | |
2cfb7455 CL |
2360 | stat(s, DEACTIVATE_EMPTY); |
2361 | discard_slab(s, page); | |
2362 | stat(s, FREE_SLAB); | |
894b8788 | 2363 | } |
d4ff6d35 WY |
2364 | |
2365 | c->page = NULL; | |
2366 | c->freelist = NULL; | |
81819f0f CL |
2367 | } |
2368 | ||
d24ac77f JK |
2369 | /* |
2370 | * Unfreeze all the cpu partial slabs. | |
2371 | * | |
59a09917 CL |
2372 | * This function must be called with interrupts disabled |
2373 | * for the cpu using c (or some other guarantee must be there | |
2374 | * to guarantee no concurrent accesses). | |
d24ac77f | 2375 | */ |
59a09917 CL |
2376 | static void unfreeze_partials(struct kmem_cache *s, |
2377 | struct kmem_cache_cpu *c) | |
49e22585 | 2378 | { |
345c905d | 2379 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2380 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2381 | struct page *page, *discard_page = NULL; |
49e22585 | 2382 | |
4c7ba22e | 2383 | while ((page = slub_percpu_partial(c))) { |
49e22585 CL |
2384 | struct page new; |
2385 | struct page old; | |
2386 | ||
4c7ba22e | 2387 | slub_set_percpu_partial(c, page); |
43d77867 JK |
2388 | |
2389 | n2 = get_node(s, page_to_nid(page)); | |
2390 | if (n != n2) { | |
2391 | if (n) | |
2392 | spin_unlock(&n->list_lock); | |
2393 | ||
2394 | n = n2; | |
2395 | spin_lock(&n->list_lock); | |
2396 | } | |
49e22585 CL |
2397 | |
2398 | do { | |
2399 | ||
2400 | old.freelist = page->freelist; | |
2401 | old.counters = page->counters; | |
a0132ac0 | 2402 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2403 | |
2404 | new.counters = old.counters; | |
2405 | new.freelist = old.freelist; | |
2406 | ||
2407 | new.frozen = 0; | |
2408 | ||
d24ac77f | 2409 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2410 | old.freelist, old.counters, |
2411 | new.freelist, new.counters, | |
2412 | "unfreezing slab")); | |
2413 | ||
8a5b20ae | 2414 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2415 | page->next = discard_page; |
2416 | discard_page = page; | |
43d77867 JK |
2417 | } else { |
2418 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2419 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2420 | } |
2421 | } | |
2422 | ||
2423 | if (n) | |
2424 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2425 | |
2426 | while (discard_page) { | |
2427 | page = discard_page; | |
2428 | discard_page = discard_page->next; | |
2429 | ||
2430 | stat(s, DEACTIVATE_EMPTY); | |
2431 | discard_slab(s, page); | |
2432 | stat(s, FREE_SLAB); | |
2433 | } | |
6dfd1b65 | 2434 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2435 | } |
2436 | ||
2437 | /* | |
9234bae9 WY |
2438 | * Put a page that was just frozen (in __slab_free|get_partial_node) into a |
2439 | * partial page slot if available. | |
49e22585 CL |
2440 | * |
2441 | * If we did not find a slot then simply move all the partials to the | |
2442 | * per node partial list. | |
2443 | */ | |
633b0764 | 2444 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2445 | { |
345c905d | 2446 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2447 | struct page *oldpage; |
2448 | int pages; | |
2449 | int pobjects; | |
2450 | ||
d6e0b7fa | 2451 | preempt_disable(); |
49e22585 CL |
2452 | do { |
2453 | pages = 0; | |
2454 | pobjects = 0; | |
2455 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2456 | ||
2457 | if (oldpage) { | |
2458 | pobjects = oldpage->pobjects; | |
2459 | pages = oldpage->pages; | |
bbd4e305 | 2460 | if (drain && pobjects > slub_cpu_partial(s)) { |
49e22585 CL |
2461 | unsigned long flags; |
2462 | /* | |
2463 | * partial array is full. Move the existing | |
2464 | * set to the per node partial list. | |
2465 | */ | |
2466 | local_irq_save(flags); | |
59a09917 | 2467 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2468 | local_irq_restore(flags); |
e24fc410 | 2469 | oldpage = NULL; |
49e22585 CL |
2470 | pobjects = 0; |
2471 | pages = 0; | |
8028dcea | 2472 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2473 | } |
2474 | } | |
2475 | ||
2476 | pages++; | |
2477 | pobjects += page->objects - page->inuse; | |
2478 | ||
2479 | page->pages = pages; | |
2480 | page->pobjects = pobjects; | |
2481 | page->next = oldpage; | |
2482 | ||
d0e0ac97 CG |
2483 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2484 | != oldpage); | |
bbd4e305 | 2485 | if (unlikely(!slub_cpu_partial(s))) { |
d6e0b7fa VD |
2486 | unsigned long flags; |
2487 | ||
2488 | local_irq_save(flags); | |
2489 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2490 | local_irq_restore(flags); | |
2491 | } | |
2492 | preempt_enable(); | |
6dfd1b65 | 2493 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2494 | } |
2495 | ||
dfb4f096 | 2496 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2497 | { |
84e554e6 | 2498 | stat(s, CPUSLAB_FLUSH); |
d4ff6d35 | 2499 | deactivate_slab(s, c->page, c->freelist, c); |
c17dda40 CL |
2500 | |
2501 | c->tid = next_tid(c->tid); | |
81819f0f CL |
2502 | } |
2503 | ||
2504 | /* | |
2505 | * Flush cpu slab. | |
6446faa2 | 2506 | * |
81819f0f CL |
2507 | * Called from IPI handler with interrupts disabled. |
2508 | */ | |
0c710013 | 2509 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2510 | { |
9dfc6e68 | 2511 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2512 | |
1265ef2d WY |
2513 | if (c->page) |
2514 | flush_slab(s, c); | |
49e22585 | 2515 | |
1265ef2d | 2516 | unfreeze_partials(s, c); |
81819f0f CL |
2517 | } |
2518 | ||
2519 | static void flush_cpu_slab(void *d) | |
2520 | { | |
2521 | struct kmem_cache *s = d; | |
81819f0f | 2522 | |
dfb4f096 | 2523 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2524 | } |
2525 | ||
a8364d55 GBY |
2526 | static bool has_cpu_slab(int cpu, void *info) |
2527 | { | |
2528 | struct kmem_cache *s = info; | |
2529 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2530 | ||
a93cf07b | 2531 | return c->page || slub_percpu_partial(c); |
a8364d55 GBY |
2532 | } |
2533 | ||
81819f0f CL |
2534 | static void flush_all(struct kmem_cache *s) |
2535 | { | |
cb923159 | 2536 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1); |
81819f0f CL |
2537 | } |
2538 | ||
a96a87bf SAS |
2539 | /* |
2540 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2541 | * necessary. | |
2542 | */ | |
2543 | static int slub_cpu_dead(unsigned int cpu) | |
2544 | { | |
2545 | struct kmem_cache *s; | |
2546 | unsigned long flags; | |
2547 | ||
2548 | mutex_lock(&slab_mutex); | |
2549 | list_for_each_entry(s, &slab_caches, list) { | |
2550 | local_irq_save(flags); | |
2551 | __flush_cpu_slab(s, cpu); | |
2552 | local_irq_restore(flags); | |
2553 | } | |
2554 | mutex_unlock(&slab_mutex); | |
2555 | return 0; | |
2556 | } | |
2557 | ||
dfb4f096 CL |
2558 | /* |
2559 | * Check if the objects in a per cpu structure fit numa | |
2560 | * locality expectations. | |
2561 | */ | |
57d437d2 | 2562 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2563 | { |
2564 | #ifdef CONFIG_NUMA | |
6159d0f5 | 2565 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2566 | return 0; |
2567 | #endif | |
2568 | return 1; | |
2569 | } | |
2570 | ||
9a02d699 | 2571 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2572 | static int count_free(struct page *page) |
2573 | { | |
2574 | return page->objects - page->inuse; | |
2575 | } | |
2576 | ||
9a02d699 DR |
2577 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2578 | { | |
2579 | return atomic_long_read(&n->total_objects); | |
2580 | } | |
2581 | #endif /* CONFIG_SLUB_DEBUG */ | |
2582 | ||
2583 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2584 | static unsigned long count_partial(struct kmem_cache_node *n, |
2585 | int (*get_count)(struct page *)) | |
2586 | { | |
2587 | unsigned long flags; | |
2588 | unsigned long x = 0; | |
2589 | struct page *page; | |
2590 | ||
2591 | spin_lock_irqsave(&n->list_lock, flags); | |
916ac052 | 2592 | list_for_each_entry(page, &n->partial, slab_list) |
781b2ba6 PE |
2593 | x += get_count(page); |
2594 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2595 | return x; | |
2596 | } | |
9a02d699 | 2597 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2598 | |
781b2ba6 PE |
2599 | static noinline void |
2600 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2601 | { | |
9a02d699 DR |
2602 | #ifdef CONFIG_SLUB_DEBUG |
2603 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2604 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2605 | int node; |
fa45dc25 | 2606 | struct kmem_cache_node *n; |
781b2ba6 | 2607 | |
9a02d699 DR |
2608 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2609 | return; | |
2610 | ||
5b3810e5 VB |
2611 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2612 | nid, gfpflags, &gfpflags); | |
19af27af | 2613 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", |
f9f58285 FF |
2614 | s->name, s->object_size, s->size, oo_order(s->oo), |
2615 | oo_order(s->min)); | |
781b2ba6 | 2616 | |
3b0efdfa | 2617 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2618 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2619 | s->name); | |
fa5ec8a1 | 2620 | |
fa45dc25 | 2621 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2622 | unsigned long nr_slabs; |
2623 | unsigned long nr_objs; | |
2624 | unsigned long nr_free; | |
2625 | ||
26c02cf0 AB |
2626 | nr_free = count_partial(n, count_free); |
2627 | nr_slabs = node_nr_slabs(n); | |
2628 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2629 | |
f9f58285 | 2630 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2631 | node, nr_slabs, nr_objs, nr_free); |
2632 | } | |
9a02d699 | 2633 | #endif |
781b2ba6 PE |
2634 | } |
2635 | ||
497b66f2 CL |
2636 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2637 | int node, struct kmem_cache_cpu **pc) | |
2638 | { | |
6faa6833 | 2639 | void *freelist; |
188fd063 CL |
2640 | struct kmem_cache_cpu *c = *pc; |
2641 | struct page *page; | |
497b66f2 | 2642 | |
128227e7 MW |
2643 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); |
2644 | ||
188fd063 | 2645 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2646 | |
188fd063 CL |
2647 | if (freelist) |
2648 | return freelist; | |
2649 | ||
2650 | page = new_slab(s, flags, node); | |
497b66f2 | 2651 | if (page) { |
7c8e0181 | 2652 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2653 | if (c->page) |
2654 | flush_slab(s, c); | |
2655 | ||
2656 | /* | |
2657 | * No other reference to the page yet so we can | |
2658 | * muck around with it freely without cmpxchg | |
2659 | */ | |
6faa6833 | 2660 | freelist = page->freelist; |
497b66f2 CL |
2661 | page->freelist = NULL; |
2662 | ||
2663 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2664 | c->page = page; |
2665 | *pc = c; | |
edde82b6 | 2666 | } |
497b66f2 | 2667 | |
6faa6833 | 2668 | return freelist; |
497b66f2 CL |
2669 | } |
2670 | ||
072bb0aa MG |
2671 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2672 | { | |
2673 | if (unlikely(PageSlabPfmemalloc(page))) | |
2674 | return gfp_pfmemalloc_allowed(gfpflags); | |
2675 | ||
2676 | return true; | |
2677 | } | |
2678 | ||
213eeb9f | 2679 | /* |
d0e0ac97 CG |
2680 | * Check the page->freelist of a page and either transfer the freelist to the |
2681 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2682 | * |
2683 | * The page is still frozen if the return value is not NULL. | |
2684 | * | |
2685 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2686 | * |
2687 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2688 | */ |
2689 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2690 | { | |
2691 | struct page new; | |
2692 | unsigned long counters; | |
2693 | void *freelist; | |
2694 | ||
2695 | do { | |
2696 | freelist = page->freelist; | |
2697 | counters = page->counters; | |
6faa6833 | 2698 | |
213eeb9f | 2699 | new.counters = counters; |
a0132ac0 | 2700 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2701 | |
2702 | new.inuse = page->objects; | |
2703 | new.frozen = freelist != NULL; | |
2704 | ||
d24ac77f | 2705 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2706 | freelist, counters, |
2707 | NULL, new.counters, | |
2708 | "get_freelist")); | |
2709 | ||
2710 | return freelist; | |
2711 | } | |
2712 | ||
81819f0f | 2713 | /* |
894b8788 CL |
2714 | * Slow path. The lockless freelist is empty or we need to perform |
2715 | * debugging duties. | |
2716 | * | |
894b8788 CL |
2717 | * Processing is still very fast if new objects have been freed to the |
2718 | * regular freelist. In that case we simply take over the regular freelist | |
2719 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2720 | * |
894b8788 CL |
2721 | * If that is not working then we fall back to the partial lists. We take the |
2722 | * first element of the freelist as the object to allocate now and move the | |
2723 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2724 | * |
894b8788 | 2725 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2726 | * we need to allocate a new slab. This is the slowest path since it involves |
2727 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2728 | * |
2729 | * Version of __slab_alloc to use when we know that interrupts are | |
2730 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2731 | */ |
a380a3c7 | 2732 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2733 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2734 | { |
6faa6833 | 2735 | void *freelist; |
f6e7def7 | 2736 | struct page *page; |
81819f0f | 2737 | |
9f986d99 AW |
2738 | stat(s, ALLOC_SLOWPATH); |
2739 | ||
f6e7def7 | 2740 | page = c->page; |
0715e6c5 VB |
2741 | if (!page) { |
2742 | /* | |
2743 | * if the node is not online or has no normal memory, just | |
2744 | * ignore the node constraint | |
2745 | */ | |
2746 | if (unlikely(node != NUMA_NO_NODE && | |
7e1fa93d | 2747 | !node_isset(node, slab_nodes))) |
0715e6c5 | 2748 | node = NUMA_NO_NODE; |
81819f0f | 2749 | goto new_slab; |
0715e6c5 | 2750 | } |
49e22585 | 2751 | redo: |
6faa6833 | 2752 | |
57d437d2 | 2753 | if (unlikely(!node_match(page, node))) { |
0715e6c5 VB |
2754 | /* |
2755 | * same as above but node_match() being false already | |
2756 | * implies node != NUMA_NO_NODE | |
2757 | */ | |
7e1fa93d | 2758 | if (!node_isset(node, slab_nodes)) { |
0715e6c5 VB |
2759 | node = NUMA_NO_NODE; |
2760 | goto redo; | |
2761 | } else { | |
a561ce00 | 2762 | stat(s, ALLOC_NODE_MISMATCH); |
d4ff6d35 | 2763 | deactivate_slab(s, page, c->freelist, c); |
a561ce00 JK |
2764 | goto new_slab; |
2765 | } | |
fc59c053 | 2766 | } |
6446faa2 | 2767 | |
072bb0aa MG |
2768 | /* |
2769 | * By rights, we should be searching for a slab page that was | |
2770 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2771 | * information when the page leaves the per-cpu allocator | |
2772 | */ | |
2773 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
d4ff6d35 | 2774 | deactivate_slab(s, page, c->freelist, c); |
072bb0aa MG |
2775 | goto new_slab; |
2776 | } | |
2777 | ||
73736e03 | 2778 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2779 | freelist = c->freelist; |
2780 | if (freelist) | |
73736e03 | 2781 | goto load_freelist; |
03e404af | 2782 | |
f6e7def7 | 2783 | freelist = get_freelist(s, page); |
6446faa2 | 2784 | |
6faa6833 | 2785 | if (!freelist) { |
03e404af CL |
2786 | c->page = NULL; |
2787 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2788 | goto new_slab; |
03e404af | 2789 | } |
6446faa2 | 2790 | |
84e554e6 | 2791 | stat(s, ALLOC_REFILL); |
6446faa2 | 2792 | |
894b8788 | 2793 | load_freelist: |
507effea CL |
2794 | /* |
2795 | * freelist is pointing to the list of objects to be used. | |
2796 | * page is pointing to the page from which the objects are obtained. | |
2797 | * That page must be frozen for per cpu allocations to work. | |
2798 | */ | |
a0132ac0 | 2799 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2800 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2801 | c->tid = next_tid(c->tid); |
6faa6833 | 2802 | return freelist; |
81819f0f | 2803 | |
81819f0f | 2804 | new_slab: |
2cfb7455 | 2805 | |
a93cf07b WY |
2806 | if (slub_percpu_partial(c)) { |
2807 | page = c->page = slub_percpu_partial(c); | |
2808 | slub_set_percpu_partial(c, page); | |
49e22585 | 2809 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 2810 | goto redo; |
81819f0f CL |
2811 | } |
2812 | ||
188fd063 | 2813 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2814 | |
f4697436 | 2815 | if (unlikely(!freelist)) { |
9a02d699 | 2816 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2817 | return NULL; |
81819f0f | 2818 | } |
2cfb7455 | 2819 | |
f6e7def7 | 2820 | page = c->page; |
5091b74a | 2821 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2822 | goto load_freelist; |
2cfb7455 | 2823 | |
497b66f2 | 2824 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2825 | if (kmem_cache_debug(s) && |
2826 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2827 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2828 | |
d4ff6d35 | 2829 | deactivate_slab(s, page, get_freepointer(s, freelist), c); |
6faa6833 | 2830 | return freelist; |
894b8788 CL |
2831 | } |
2832 | ||
a380a3c7 CL |
2833 | /* |
2834 | * Another one that disabled interrupt and compensates for possible | |
2835 | * cpu changes by refetching the per cpu area pointer. | |
2836 | */ | |
2837 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2838 | unsigned long addr, struct kmem_cache_cpu *c) | |
2839 | { | |
2840 | void *p; | |
2841 | unsigned long flags; | |
2842 | ||
2843 | local_irq_save(flags); | |
923717cb | 2844 | #ifdef CONFIG_PREEMPTION |
a380a3c7 CL |
2845 | /* |
2846 | * We may have been preempted and rescheduled on a different | |
2847 | * cpu before disabling interrupts. Need to reload cpu area | |
2848 | * pointer. | |
2849 | */ | |
2850 | c = this_cpu_ptr(s->cpu_slab); | |
2851 | #endif | |
2852 | ||
2853 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2854 | local_irq_restore(flags); | |
2855 | return p; | |
2856 | } | |
2857 | ||
0f181f9f AP |
2858 | /* |
2859 | * If the object has been wiped upon free, make sure it's fully initialized by | |
2860 | * zeroing out freelist pointer. | |
2861 | */ | |
2862 | static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s, | |
2863 | void *obj) | |
2864 | { | |
2865 | if (unlikely(slab_want_init_on_free(s)) && obj) | |
ce5716c6 AK |
2866 | memset((void *)((char *)kasan_reset_tag(obj) + s->offset), |
2867 | 0, sizeof(void *)); | |
0f181f9f AP |
2868 | } |
2869 | ||
894b8788 CL |
2870 | /* |
2871 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2872 | * have the fastpath folded into their functions. So no function call | |
2873 | * overhead for requests that can be satisfied on the fastpath. | |
2874 | * | |
2875 | * The fastpath works by first checking if the lockless freelist can be used. | |
2876 | * If not then __slab_alloc is called for slow processing. | |
2877 | * | |
2878 | * Otherwise we can simply pick the next object from the lockless free list. | |
2879 | */ | |
2b847c3c | 2880 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
b89fb5ef | 2881 | gfp_t gfpflags, int node, unsigned long addr, size_t orig_size) |
894b8788 | 2882 | { |
03ec0ed5 | 2883 | void *object; |
dfb4f096 | 2884 | struct kmem_cache_cpu *c; |
57d437d2 | 2885 | struct page *page; |
8a5ec0ba | 2886 | unsigned long tid; |
964d4bd3 | 2887 | struct obj_cgroup *objcg = NULL; |
da844b78 | 2888 | bool init = false; |
1f84260c | 2889 | |
964d4bd3 | 2890 | s = slab_pre_alloc_hook(s, &objcg, 1, gfpflags); |
8135be5a | 2891 | if (!s) |
773ff60e | 2892 | return NULL; |
b89fb5ef AP |
2893 | |
2894 | object = kfence_alloc(s, orig_size, gfpflags); | |
2895 | if (unlikely(object)) | |
2896 | goto out; | |
2897 | ||
8a5ec0ba | 2898 | redo: |
8a5ec0ba CL |
2899 | /* |
2900 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2901 | * enabled. We may switch back and forth between cpus while | |
2902 | * reading from one cpu area. That does not matter as long | |
2903 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2904 | * |
9aabf810 | 2905 | * We should guarantee that tid and kmem_cache are retrieved on |
923717cb | 2906 | * the same cpu. It could be different if CONFIG_PREEMPTION so we need |
9aabf810 | 2907 | * to check if it is matched or not. |
8a5ec0ba | 2908 | */ |
9aabf810 JK |
2909 | do { |
2910 | tid = this_cpu_read(s->cpu_slab->tid); | |
2911 | c = raw_cpu_ptr(s->cpu_slab); | |
923717cb | 2912 | } while (IS_ENABLED(CONFIG_PREEMPTION) && |
859b7a0e | 2913 | unlikely(tid != READ_ONCE(c->tid))); |
9aabf810 JK |
2914 | |
2915 | /* | |
2916 | * Irqless object alloc/free algorithm used here depends on sequence | |
2917 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2918 | * on c to guarantee that object and page associated with previous tid | |
2919 | * won't be used with current tid. If we fetch tid first, object and | |
2920 | * page could be one associated with next tid and our alloc/free | |
2921 | * request will be failed. In this case, we will retry. So, no problem. | |
2922 | */ | |
2923 | barrier(); | |
8a5ec0ba | 2924 | |
8a5ec0ba CL |
2925 | /* |
2926 | * The transaction ids are globally unique per cpu and per operation on | |
2927 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2928 | * occurs on the right processor and that there was no operation on the | |
2929 | * linked list in between. | |
2930 | */ | |
8a5ec0ba | 2931 | |
9dfc6e68 | 2932 | object = c->freelist; |
57d437d2 | 2933 | page = c->page; |
22e4663e | 2934 | if (unlikely(!object || !page || !node_match(page, node))) { |
dfb4f096 | 2935 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 | 2936 | } else { |
0ad9500e ED |
2937 | void *next_object = get_freepointer_safe(s, object); |
2938 | ||
8a5ec0ba | 2939 | /* |
25985edc | 2940 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2941 | * operation and if we are on the right processor. |
2942 | * | |
d0e0ac97 CG |
2943 | * The cmpxchg does the following atomically (without lock |
2944 | * semantics!) | |
8a5ec0ba CL |
2945 | * 1. Relocate first pointer to the current per cpu area. |
2946 | * 2. Verify that tid and freelist have not been changed | |
2947 | * 3. If they were not changed replace tid and freelist | |
2948 | * | |
d0e0ac97 CG |
2949 | * Since this is without lock semantics the protection is only |
2950 | * against code executing on this cpu *not* from access by | |
2951 | * other cpus. | |
8a5ec0ba | 2952 | */ |
933393f5 | 2953 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2954 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2955 | object, tid, | |
0ad9500e | 2956 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2957 | |
2958 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2959 | goto redo; | |
2960 | } | |
0ad9500e | 2961 | prefetch_freepointer(s, next_object); |
84e554e6 | 2962 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2963 | } |
0f181f9f | 2964 | |
ce5716c6 | 2965 | maybe_wipe_obj_freeptr(s, object); |
da844b78 | 2966 | init = slab_want_init_on_alloc(gfpflags, s); |
d07dbea4 | 2967 | |
b89fb5ef | 2968 | out: |
da844b78 | 2969 | slab_post_alloc_hook(s, objcg, gfpflags, 1, &object, init); |
5a896d9e | 2970 | |
894b8788 | 2971 | return object; |
81819f0f CL |
2972 | } |
2973 | ||
2b847c3c | 2974 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
b89fb5ef | 2975 | gfp_t gfpflags, unsigned long addr, size_t orig_size) |
2b847c3c | 2976 | { |
b89fb5ef | 2977 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr, orig_size); |
2b847c3c EG |
2978 | } |
2979 | ||
81819f0f CL |
2980 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2981 | { | |
b89fb5ef | 2982 | void *ret = slab_alloc(s, gfpflags, _RET_IP_, s->object_size); |
5b882be4 | 2983 | |
d0e0ac97 CG |
2984 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2985 | s->size, gfpflags); | |
5b882be4 EGM |
2986 | |
2987 | return ret; | |
81819f0f CL |
2988 | } |
2989 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2990 | ||
0f24f128 | 2991 | #ifdef CONFIG_TRACING |
4a92379b RK |
2992 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2993 | { | |
b89fb5ef | 2994 | void *ret = slab_alloc(s, gfpflags, _RET_IP_, size); |
4a92379b | 2995 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0116523c | 2996 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2997 | return ret; |
2998 | } | |
2999 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
3000 | #endif |
3001 | ||
81819f0f CL |
3002 | #ifdef CONFIG_NUMA |
3003 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
3004 | { | |
b89fb5ef | 3005 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_, s->object_size); |
5b882be4 | 3006 | |
ca2b84cb | 3007 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 3008 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
3009 | |
3010 | return ret; | |
81819f0f CL |
3011 | } |
3012 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 3013 | |
0f24f128 | 3014 | #ifdef CONFIG_TRACING |
4a92379b | 3015 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 3016 | gfp_t gfpflags, |
4a92379b | 3017 | int node, size_t size) |
5b882be4 | 3018 | { |
b89fb5ef | 3019 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_, size); |
4a92379b RK |
3020 | |
3021 | trace_kmalloc_node(_RET_IP_, ret, | |
3022 | size, s->size, gfpflags, node); | |
0316bec2 | 3023 | |
0116523c | 3024 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 3025 | return ret; |
5b882be4 | 3026 | } |
4a92379b | 3027 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 3028 | #endif |
6dfd1b65 | 3029 | #endif /* CONFIG_NUMA */ |
5b882be4 | 3030 | |
81819f0f | 3031 | /* |
94e4d712 | 3032 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 3033 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 3034 | * |
894b8788 CL |
3035 | * So we still attempt to reduce cache line usage. Just take the slab |
3036 | * lock and free the item. If there is no additional partial page | |
3037 | * handling required then we can return immediately. | |
81819f0f | 3038 | */ |
894b8788 | 3039 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
3040 | void *head, void *tail, int cnt, |
3041 | unsigned long addr) | |
3042 | ||
81819f0f CL |
3043 | { |
3044 | void *prior; | |
2cfb7455 | 3045 | int was_frozen; |
2cfb7455 CL |
3046 | struct page new; |
3047 | unsigned long counters; | |
3048 | struct kmem_cache_node *n = NULL; | |
3f649ab7 | 3049 | unsigned long flags; |
81819f0f | 3050 | |
8a5ec0ba | 3051 | stat(s, FREE_SLOWPATH); |
81819f0f | 3052 | |
b89fb5ef AP |
3053 | if (kfence_free(head)) |
3054 | return; | |
3055 | ||
19c7ff9e | 3056 | if (kmem_cache_debug(s) && |
282acb43 | 3057 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 3058 | return; |
6446faa2 | 3059 | |
2cfb7455 | 3060 | do { |
837d678d JK |
3061 | if (unlikely(n)) { |
3062 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3063 | n = NULL; | |
3064 | } | |
2cfb7455 CL |
3065 | prior = page->freelist; |
3066 | counters = page->counters; | |
81084651 | 3067 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
3068 | new.counters = counters; |
3069 | was_frozen = new.frozen; | |
81084651 | 3070 | new.inuse -= cnt; |
837d678d | 3071 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 3072 | |
c65c1877 | 3073 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
3074 | |
3075 | /* | |
d0e0ac97 CG |
3076 | * Slab was on no list before and will be |
3077 | * partially empty | |
3078 | * We can defer the list move and instead | |
3079 | * freeze it. | |
49e22585 CL |
3080 | */ |
3081 | new.frozen = 1; | |
3082 | ||
c65c1877 | 3083 | } else { /* Needs to be taken off a list */ |
49e22585 | 3084 | |
b455def2 | 3085 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
3086 | /* |
3087 | * Speculatively acquire the list_lock. | |
3088 | * If the cmpxchg does not succeed then we may | |
3089 | * drop the list_lock without any processing. | |
3090 | * | |
3091 | * Otherwise the list_lock will synchronize with | |
3092 | * other processors updating the list of slabs. | |
3093 | */ | |
3094 | spin_lock_irqsave(&n->list_lock, flags); | |
3095 | ||
3096 | } | |
2cfb7455 | 3097 | } |
81819f0f | 3098 | |
2cfb7455 CL |
3099 | } while (!cmpxchg_double_slab(s, page, |
3100 | prior, counters, | |
81084651 | 3101 | head, new.counters, |
2cfb7455 | 3102 | "__slab_free")); |
81819f0f | 3103 | |
2cfb7455 | 3104 | if (likely(!n)) { |
49e22585 | 3105 | |
c270cf30 AW |
3106 | if (likely(was_frozen)) { |
3107 | /* | |
3108 | * The list lock was not taken therefore no list | |
3109 | * activity can be necessary. | |
3110 | */ | |
3111 | stat(s, FREE_FROZEN); | |
3112 | } else if (new.frozen) { | |
3113 | /* | |
3114 | * If we just froze the page then put it onto the | |
3115 | * per cpu partial list. | |
3116 | */ | |
49e22585 | 3117 | put_cpu_partial(s, page, 1); |
8028dcea AS |
3118 | stat(s, CPU_PARTIAL_FREE); |
3119 | } | |
c270cf30 | 3120 | |
b455def2 L |
3121 | return; |
3122 | } | |
81819f0f | 3123 | |
8a5b20ae | 3124 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
3125 | goto slab_empty; |
3126 | ||
81819f0f | 3127 | /* |
837d678d JK |
3128 | * Objects left in the slab. If it was not on the partial list before |
3129 | * then add it. | |
81819f0f | 3130 | */ |
345c905d | 3131 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
a4d3f891 | 3132 | remove_full(s, n, page); |
837d678d JK |
3133 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
3134 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 3135 | } |
80f08c19 | 3136 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
3137 | return; |
3138 | ||
3139 | slab_empty: | |
a973e9dd | 3140 | if (prior) { |
81819f0f | 3141 | /* |
6fbabb20 | 3142 | * Slab on the partial list. |
81819f0f | 3143 | */ |
5cc6eee8 | 3144 | remove_partial(n, page); |
84e554e6 | 3145 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 3146 | } else { |
6fbabb20 | 3147 | /* Slab must be on the full list */ |
c65c1877 PZ |
3148 | remove_full(s, n, page); |
3149 | } | |
2cfb7455 | 3150 | |
80f08c19 | 3151 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 3152 | stat(s, FREE_SLAB); |
81819f0f | 3153 | discard_slab(s, page); |
81819f0f CL |
3154 | } |
3155 | ||
894b8788 CL |
3156 | /* |
3157 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
3158 | * can perform fastpath freeing without additional function calls. | |
3159 | * | |
3160 | * The fastpath is only possible if we are freeing to the current cpu slab | |
3161 | * of this processor. This typically the case if we have just allocated | |
3162 | * the item before. | |
3163 | * | |
3164 | * If fastpath is not possible then fall back to __slab_free where we deal | |
3165 | * with all sorts of special processing. | |
81084651 JDB |
3166 | * |
3167 | * Bulk free of a freelist with several objects (all pointing to the | |
3168 | * same page) possible by specifying head and tail ptr, plus objects | |
3169 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 3170 | */ |
80a9201a AP |
3171 | static __always_inline void do_slab_free(struct kmem_cache *s, |
3172 | struct page *page, void *head, void *tail, | |
3173 | int cnt, unsigned long addr) | |
894b8788 | 3174 | { |
81084651 | 3175 | void *tail_obj = tail ? : head; |
dfb4f096 | 3176 | struct kmem_cache_cpu *c; |
8a5ec0ba | 3177 | unsigned long tid; |
964d4bd3 | 3178 | |
d1b2cf6c | 3179 | memcg_slab_free_hook(s, &head, 1); |
8a5ec0ba CL |
3180 | redo: |
3181 | /* | |
3182 | * Determine the currently cpus per cpu slab. | |
3183 | * The cpu may change afterward. However that does not matter since | |
3184 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 3185 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 3186 | */ |
9aabf810 JK |
3187 | do { |
3188 | tid = this_cpu_read(s->cpu_slab->tid); | |
3189 | c = raw_cpu_ptr(s->cpu_slab); | |
923717cb | 3190 | } while (IS_ENABLED(CONFIG_PREEMPTION) && |
859b7a0e | 3191 | unlikely(tid != READ_ONCE(c->tid))); |
c016b0bd | 3192 | |
9aabf810 JK |
3193 | /* Same with comment on barrier() in slab_alloc_node() */ |
3194 | barrier(); | |
c016b0bd | 3195 | |
442b06bc | 3196 | if (likely(page == c->page)) { |
5076190d LT |
3197 | void **freelist = READ_ONCE(c->freelist); |
3198 | ||
3199 | set_freepointer(s, tail_obj, freelist); | |
8a5ec0ba | 3200 | |
933393f5 | 3201 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba | 3202 | s->cpu_slab->freelist, s->cpu_slab->tid, |
5076190d | 3203 | freelist, tid, |
81084651 | 3204 | head, next_tid(tid)))) { |
8a5ec0ba CL |
3205 | |
3206 | note_cmpxchg_failure("slab_free", s, tid); | |
3207 | goto redo; | |
3208 | } | |
84e554e6 | 3209 | stat(s, FREE_FASTPATH); |
894b8788 | 3210 | } else |
81084651 | 3211 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 3212 | |
894b8788 CL |
3213 | } |
3214 | ||
80a9201a AP |
3215 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
3216 | void *head, void *tail, int cnt, | |
3217 | unsigned long addr) | |
3218 | { | |
80a9201a | 3219 | /* |
c3895391 AK |
3220 | * With KASAN enabled slab_free_freelist_hook modifies the freelist |
3221 | * to remove objects, whose reuse must be delayed. | |
80a9201a | 3222 | */ |
c3895391 AK |
3223 | if (slab_free_freelist_hook(s, &head, &tail)) |
3224 | do_slab_free(s, page, head, tail, cnt, addr); | |
80a9201a AP |
3225 | } |
3226 | ||
2bd926b4 | 3227 | #ifdef CONFIG_KASAN_GENERIC |
80a9201a AP |
3228 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) |
3229 | { | |
3230 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
3231 | } | |
3232 | #endif | |
3233 | ||
81819f0f CL |
3234 | void kmem_cache_free(struct kmem_cache *s, void *x) |
3235 | { | |
b9ce5ef4 GC |
3236 | s = cache_from_obj(s, x); |
3237 | if (!s) | |
79576102 | 3238 | return; |
81084651 | 3239 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
3544de8e | 3240 | trace_kmem_cache_free(_RET_IP_, x, s->name); |
81819f0f CL |
3241 | } |
3242 | EXPORT_SYMBOL(kmem_cache_free); | |
3243 | ||
d0ecd894 | 3244 | struct detached_freelist { |
fbd02630 | 3245 | struct page *page; |
d0ecd894 JDB |
3246 | void *tail; |
3247 | void *freelist; | |
3248 | int cnt; | |
376bf125 | 3249 | struct kmem_cache *s; |
d0ecd894 | 3250 | }; |
fbd02630 | 3251 | |
d0ecd894 JDB |
3252 | /* |
3253 | * This function progressively scans the array with free objects (with | |
3254 | * a limited look ahead) and extract objects belonging to the same | |
3255 | * page. It builds a detached freelist directly within the given | |
3256 | * page/objects. This can happen without any need for | |
3257 | * synchronization, because the objects are owned by running process. | |
3258 | * The freelist is build up as a single linked list in the objects. | |
3259 | * The idea is, that this detached freelist can then be bulk | |
3260 | * transferred to the real freelist(s), but only requiring a single | |
3261 | * synchronization primitive. Look ahead in the array is limited due | |
3262 | * to performance reasons. | |
3263 | */ | |
376bf125 JDB |
3264 | static inline |
3265 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3266 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
3267 | { |
3268 | size_t first_skipped_index = 0; | |
3269 | int lookahead = 3; | |
3270 | void *object; | |
ca257195 | 3271 | struct page *page; |
fbd02630 | 3272 | |
d0ecd894 JDB |
3273 | /* Always re-init detached_freelist */ |
3274 | df->page = NULL; | |
fbd02630 | 3275 | |
d0ecd894 JDB |
3276 | do { |
3277 | object = p[--size]; | |
ca257195 | 3278 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3279 | } while (!object && size); |
3eed034d | 3280 | |
d0ecd894 JDB |
3281 | if (!object) |
3282 | return 0; | |
fbd02630 | 3283 | |
ca257195 JDB |
3284 | page = virt_to_head_page(object); |
3285 | if (!s) { | |
3286 | /* Handle kalloc'ed objects */ | |
3287 | if (unlikely(!PageSlab(page))) { | |
3288 | BUG_ON(!PageCompound(page)); | |
3289 | kfree_hook(object); | |
4949148a | 3290 | __free_pages(page, compound_order(page)); |
ca257195 JDB |
3291 | p[size] = NULL; /* mark object processed */ |
3292 | return size; | |
3293 | } | |
3294 | /* Derive kmem_cache from object */ | |
3295 | df->s = page->slab_cache; | |
3296 | } else { | |
3297 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3298 | } | |
376bf125 | 3299 | |
b89fb5ef | 3300 | if (is_kfence_address(object)) { |
d57a964e | 3301 | slab_free_hook(df->s, object, false); |
b89fb5ef AP |
3302 | __kfence_free(object); |
3303 | p[size] = NULL; /* mark object processed */ | |
3304 | return size; | |
3305 | } | |
3306 | ||
d0ecd894 | 3307 | /* Start new detached freelist */ |
ca257195 | 3308 | df->page = page; |
376bf125 | 3309 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3310 | df->tail = object; |
3311 | df->freelist = object; | |
3312 | p[size] = NULL; /* mark object processed */ | |
3313 | df->cnt = 1; | |
3314 | ||
3315 | while (size) { | |
3316 | object = p[--size]; | |
3317 | if (!object) | |
3318 | continue; /* Skip processed objects */ | |
3319 | ||
3320 | /* df->page is always set at this point */ | |
3321 | if (df->page == virt_to_head_page(object)) { | |
3322 | /* Opportunity build freelist */ | |
376bf125 | 3323 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3324 | df->freelist = object; |
3325 | df->cnt++; | |
3326 | p[size] = NULL; /* mark object processed */ | |
3327 | ||
3328 | continue; | |
fbd02630 | 3329 | } |
d0ecd894 JDB |
3330 | |
3331 | /* Limit look ahead search */ | |
3332 | if (!--lookahead) | |
3333 | break; | |
3334 | ||
3335 | if (!first_skipped_index) | |
3336 | first_skipped_index = size + 1; | |
fbd02630 | 3337 | } |
d0ecd894 JDB |
3338 | |
3339 | return first_skipped_index; | |
3340 | } | |
3341 | ||
d0ecd894 | 3342 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3343 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3344 | { |
3345 | if (WARN_ON(!size)) | |
3346 | return; | |
3347 | ||
d1b2cf6c | 3348 | memcg_slab_free_hook(s, p, size); |
d0ecd894 JDB |
3349 | do { |
3350 | struct detached_freelist df; | |
3351 | ||
3352 | size = build_detached_freelist(s, size, p, &df); | |
84582c8a | 3353 | if (!df.page) |
d0ecd894 JDB |
3354 | continue; |
3355 | ||
457c82c3 | 3356 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt, _RET_IP_); |
d0ecd894 | 3357 | } while (likely(size)); |
484748f0 CL |
3358 | } |
3359 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3360 | ||
994eb764 | 3361 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3362 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3363 | void **p) | |
484748f0 | 3364 | { |
994eb764 JDB |
3365 | struct kmem_cache_cpu *c; |
3366 | int i; | |
964d4bd3 | 3367 | struct obj_cgroup *objcg = NULL; |
994eb764 | 3368 | |
03ec0ed5 | 3369 | /* memcg and kmem_cache debug support */ |
964d4bd3 | 3370 | s = slab_pre_alloc_hook(s, &objcg, size, flags); |
03ec0ed5 JDB |
3371 | if (unlikely(!s)) |
3372 | return false; | |
994eb764 JDB |
3373 | /* |
3374 | * Drain objects in the per cpu slab, while disabling local | |
3375 | * IRQs, which protects against PREEMPT and interrupts | |
3376 | * handlers invoking normal fastpath. | |
3377 | */ | |
3378 | local_irq_disable(); | |
3379 | c = this_cpu_ptr(s->cpu_slab); | |
3380 | ||
3381 | for (i = 0; i < size; i++) { | |
b89fb5ef | 3382 | void *object = kfence_alloc(s, s->object_size, flags); |
994eb764 | 3383 | |
b89fb5ef AP |
3384 | if (unlikely(object)) { |
3385 | p[i] = object; | |
3386 | continue; | |
3387 | } | |
3388 | ||
3389 | object = c->freelist; | |
ebe909e0 | 3390 | if (unlikely(!object)) { |
fd4d9c7d JH |
3391 | /* |
3392 | * We may have removed an object from c->freelist using | |
3393 | * the fastpath in the previous iteration; in that case, | |
3394 | * c->tid has not been bumped yet. | |
3395 | * Since ___slab_alloc() may reenable interrupts while | |
3396 | * allocating memory, we should bump c->tid now. | |
3397 | */ | |
3398 | c->tid = next_tid(c->tid); | |
3399 | ||
ebe909e0 JDB |
3400 | /* |
3401 | * Invoking slow path likely have side-effect | |
3402 | * of re-populating per CPU c->freelist | |
3403 | */ | |
87098373 | 3404 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3405 | _RET_IP_, c); |
87098373 CL |
3406 | if (unlikely(!p[i])) |
3407 | goto error; | |
3408 | ||
ebe909e0 | 3409 | c = this_cpu_ptr(s->cpu_slab); |
0f181f9f AP |
3410 | maybe_wipe_obj_freeptr(s, p[i]); |
3411 | ||
ebe909e0 JDB |
3412 | continue; /* goto for-loop */ |
3413 | } | |
994eb764 JDB |
3414 | c->freelist = get_freepointer(s, object); |
3415 | p[i] = object; | |
0f181f9f | 3416 | maybe_wipe_obj_freeptr(s, p[i]); |
994eb764 JDB |
3417 | } |
3418 | c->tid = next_tid(c->tid); | |
3419 | local_irq_enable(); | |
3420 | ||
da844b78 AK |
3421 | /* |
3422 | * memcg and kmem_cache debug support and memory initialization. | |
3423 | * Done outside of the IRQ disabled fastpath loop. | |
3424 | */ | |
3425 | slab_post_alloc_hook(s, objcg, flags, size, p, | |
3426 | slab_want_init_on_alloc(flags, s)); | |
865762a8 | 3427 | return i; |
87098373 | 3428 | error: |
87098373 | 3429 | local_irq_enable(); |
da844b78 | 3430 | slab_post_alloc_hook(s, objcg, flags, i, p, false); |
03ec0ed5 | 3431 | __kmem_cache_free_bulk(s, i, p); |
865762a8 | 3432 | return 0; |
484748f0 CL |
3433 | } |
3434 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3435 | ||
3436 | ||
81819f0f | 3437 | /* |
672bba3a CL |
3438 | * Object placement in a slab is made very easy because we always start at |
3439 | * offset 0. If we tune the size of the object to the alignment then we can | |
3440 | * get the required alignment by putting one properly sized object after | |
3441 | * another. | |
81819f0f CL |
3442 | * |
3443 | * Notice that the allocation order determines the sizes of the per cpu | |
3444 | * caches. Each processor has always one slab available for allocations. | |
3445 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3446 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3447 | * locking overhead. |
81819f0f CL |
3448 | */ |
3449 | ||
3450 | /* | |
f0953a1b | 3451 | * Minimum / Maximum order of slab pages. This influences locking overhead |
81819f0f CL |
3452 | * and slab fragmentation. A higher order reduces the number of partial slabs |
3453 | * and increases the number of allocations possible without having to | |
3454 | * take the list_lock. | |
3455 | */ | |
19af27af AD |
3456 | static unsigned int slub_min_order; |
3457 | static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; | |
3458 | static unsigned int slub_min_objects; | |
81819f0f | 3459 | |
81819f0f CL |
3460 | /* |
3461 | * Calculate the order of allocation given an slab object size. | |
3462 | * | |
672bba3a CL |
3463 | * The order of allocation has significant impact on performance and other |
3464 | * system components. Generally order 0 allocations should be preferred since | |
3465 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3466 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3467 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3468 | * would be wasted. |
3469 | * | |
3470 | * In order to reach satisfactory performance we must ensure that a minimum | |
3471 | * number of objects is in one slab. Otherwise we may generate too much | |
3472 | * activity on the partial lists which requires taking the list_lock. This is | |
3473 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3474 | * |
672bba3a CL |
3475 | * slub_max_order specifies the order where we begin to stop considering the |
3476 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3477 | * we try to keep the page order as low as possible. So we accept more waste | |
3478 | * of space in favor of a small page order. | |
81819f0f | 3479 | * |
672bba3a CL |
3480 | * Higher order allocations also allow the placement of more objects in a |
3481 | * slab and thereby reduce object handling overhead. If the user has | |
dc84207d | 3482 | * requested a higher minimum order then we start with that one instead of |
672bba3a | 3483 | * the smallest order which will fit the object. |
81819f0f | 3484 | */ |
19af27af AD |
3485 | static inline unsigned int slab_order(unsigned int size, |
3486 | unsigned int min_objects, unsigned int max_order, | |
9736d2a9 | 3487 | unsigned int fract_leftover) |
81819f0f | 3488 | { |
19af27af AD |
3489 | unsigned int min_order = slub_min_order; |
3490 | unsigned int order; | |
81819f0f | 3491 | |
9736d2a9 | 3492 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3493 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3494 | |
9736d2a9 | 3495 | for (order = max(min_order, (unsigned int)get_order(min_objects * size)); |
5e6d444e | 3496 | order <= max_order; order++) { |
81819f0f | 3497 | |
19af27af AD |
3498 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; |
3499 | unsigned int rem; | |
81819f0f | 3500 | |
9736d2a9 | 3501 | rem = slab_size % size; |
81819f0f | 3502 | |
5e6d444e | 3503 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3504 | break; |
81819f0f | 3505 | } |
672bba3a | 3506 | |
81819f0f CL |
3507 | return order; |
3508 | } | |
3509 | ||
9736d2a9 | 3510 | static inline int calculate_order(unsigned int size) |
5e6d444e | 3511 | { |
19af27af AD |
3512 | unsigned int order; |
3513 | unsigned int min_objects; | |
3514 | unsigned int max_objects; | |
3286222f | 3515 | unsigned int nr_cpus; |
5e6d444e CL |
3516 | |
3517 | /* | |
3518 | * Attempt to find best configuration for a slab. This | |
3519 | * works by first attempting to generate a layout with | |
3520 | * the best configuration and backing off gradually. | |
3521 | * | |
422ff4d7 | 3522 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3523 | * we reduce the minimum objects required in a slab. |
3524 | */ | |
3525 | min_objects = slub_min_objects; | |
3286222f VB |
3526 | if (!min_objects) { |
3527 | /* | |
3528 | * Some architectures will only update present cpus when | |
3529 | * onlining them, so don't trust the number if it's just 1. But | |
3530 | * we also don't want to use nr_cpu_ids always, as on some other | |
3531 | * architectures, there can be many possible cpus, but never | |
3532 | * onlined. Here we compromise between trying to avoid too high | |
3533 | * order on systems that appear larger than they are, and too | |
3534 | * low order on systems that appear smaller than they are. | |
3535 | */ | |
3536 | nr_cpus = num_present_cpus(); | |
3537 | if (nr_cpus <= 1) | |
3538 | nr_cpus = nr_cpu_ids; | |
3539 | min_objects = 4 * (fls(nr_cpus) + 1); | |
3540 | } | |
9736d2a9 | 3541 | max_objects = order_objects(slub_max_order, size); |
e8120ff1 ZY |
3542 | min_objects = min(min_objects, max_objects); |
3543 | ||
5e6d444e | 3544 | while (min_objects > 1) { |
19af27af AD |
3545 | unsigned int fraction; |
3546 | ||
c124f5b5 | 3547 | fraction = 16; |
5e6d444e CL |
3548 | while (fraction >= 4) { |
3549 | order = slab_order(size, min_objects, | |
9736d2a9 | 3550 | slub_max_order, fraction); |
5e6d444e CL |
3551 | if (order <= slub_max_order) |
3552 | return order; | |
3553 | fraction /= 2; | |
3554 | } | |
5086c389 | 3555 | min_objects--; |
5e6d444e CL |
3556 | } |
3557 | ||
3558 | /* | |
3559 | * We were unable to place multiple objects in a slab. Now | |
3560 | * lets see if we can place a single object there. | |
3561 | */ | |
9736d2a9 | 3562 | order = slab_order(size, 1, slub_max_order, 1); |
5e6d444e CL |
3563 | if (order <= slub_max_order) |
3564 | return order; | |
3565 | ||
3566 | /* | |
3567 | * Doh this slab cannot be placed using slub_max_order. | |
3568 | */ | |
9736d2a9 | 3569 | order = slab_order(size, 1, MAX_ORDER, 1); |
818cf590 | 3570 | if (order < MAX_ORDER) |
5e6d444e CL |
3571 | return order; |
3572 | return -ENOSYS; | |
3573 | } | |
3574 | ||
5595cffc | 3575 | static void |
4053497d | 3576 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3577 | { |
3578 | n->nr_partial = 0; | |
81819f0f CL |
3579 | spin_lock_init(&n->list_lock); |
3580 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3581 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3582 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3583 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3584 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3585 | #endif |
81819f0f CL |
3586 | } |
3587 | ||
55136592 | 3588 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3589 | { |
6c182dc0 | 3590 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3591 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3592 | |
8a5ec0ba | 3593 | /* |
d4d84fef CM |
3594 | * Must align to double word boundary for the double cmpxchg |
3595 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3596 | */ |
d4d84fef CM |
3597 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3598 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3599 | |
3600 | if (!s->cpu_slab) | |
3601 | return 0; | |
3602 | ||
3603 | init_kmem_cache_cpus(s); | |
4c93c355 | 3604 | |
8a5ec0ba | 3605 | return 1; |
4c93c355 | 3606 | } |
4c93c355 | 3607 | |
51df1142 CL |
3608 | static struct kmem_cache *kmem_cache_node; |
3609 | ||
81819f0f CL |
3610 | /* |
3611 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3612 | * slab on the node for this slabcache. There are no concurrent accesses | |
3613 | * possible. | |
3614 | * | |
721ae22a ZYW |
3615 | * Note that this function only works on the kmem_cache_node |
3616 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3617 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3618 | */ |
55136592 | 3619 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3620 | { |
3621 | struct page *page; | |
3622 | struct kmem_cache_node *n; | |
3623 | ||
51df1142 | 3624 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3625 | |
51df1142 | 3626 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3627 | |
3628 | BUG_ON(!page); | |
a2f92ee7 | 3629 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3630 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3631 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3632 | } |
3633 | ||
81819f0f CL |
3634 | n = page->freelist; |
3635 | BUG_ON(!n); | |
8ab1372f | 3636 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3637 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3638 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3639 | #endif |
da844b78 | 3640 | n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false); |
12b22386 AK |
3641 | page->freelist = get_freepointer(kmem_cache_node, n); |
3642 | page->inuse = 1; | |
3643 | page->frozen = 0; | |
3644 | kmem_cache_node->node[node] = n; | |
4053497d | 3645 | init_kmem_cache_node(n); |
51df1142 | 3646 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3647 | |
67b6c900 | 3648 | /* |
1e4dd946 SR |
3649 | * No locks need to be taken here as it has just been |
3650 | * initialized and there is no concurrent access. | |
67b6c900 | 3651 | */ |
1e4dd946 | 3652 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3653 | } |
3654 | ||
3655 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3656 | { | |
3657 | int node; | |
fa45dc25 | 3658 | struct kmem_cache_node *n; |
81819f0f | 3659 | |
fa45dc25 | 3660 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 3661 | s->node[node] = NULL; |
ea37df54 | 3662 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
3663 | } |
3664 | } | |
3665 | ||
52b4b950 DS |
3666 | void __kmem_cache_release(struct kmem_cache *s) |
3667 | { | |
210e7a43 | 3668 | cache_random_seq_destroy(s); |
52b4b950 DS |
3669 | free_percpu(s->cpu_slab); |
3670 | free_kmem_cache_nodes(s); | |
3671 | } | |
3672 | ||
55136592 | 3673 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3674 | { |
3675 | int node; | |
81819f0f | 3676 | |
7e1fa93d | 3677 | for_each_node_mask(node, slab_nodes) { |
81819f0f CL |
3678 | struct kmem_cache_node *n; |
3679 | ||
73367bd8 | 3680 | if (slab_state == DOWN) { |
55136592 | 3681 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3682 | continue; |
3683 | } | |
51df1142 | 3684 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3685 | GFP_KERNEL, node); |
81819f0f | 3686 | |
73367bd8 AD |
3687 | if (!n) { |
3688 | free_kmem_cache_nodes(s); | |
3689 | return 0; | |
81819f0f | 3690 | } |
73367bd8 | 3691 | |
4053497d | 3692 | init_kmem_cache_node(n); |
ea37df54 | 3693 | s->node[node] = n; |
81819f0f CL |
3694 | } |
3695 | return 1; | |
3696 | } | |
81819f0f | 3697 | |
c0bdb232 | 3698 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3699 | { |
3700 | if (min < MIN_PARTIAL) | |
3701 | min = MIN_PARTIAL; | |
3702 | else if (min > MAX_PARTIAL) | |
3703 | min = MAX_PARTIAL; | |
3704 | s->min_partial = min; | |
3705 | } | |
3706 | ||
e6d0e1dc WY |
3707 | static void set_cpu_partial(struct kmem_cache *s) |
3708 | { | |
3709 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
3710 | /* | |
3711 | * cpu_partial determined the maximum number of objects kept in the | |
3712 | * per cpu partial lists of a processor. | |
3713 | * | |
3714 | * Per cpu partial lists mainly contain slabs that just have one | |
3715 | * object freed. If they are used for allocation then they can be | |
3716 | * filled up again with minimal effort. The slab will never hit the | |
3717 | * per node partial lists and therefore no locking will be required. | |
3718 | * | |
3719 | * This setting also determines | |
3720 | * | |
3721 | * A) The number of objects from per cpu partial slabs dumped to the | |
3722 | * per node list when we reach the limit. | |
3723 | * B) The number of objects in cpu partial slabs to extract from the | |
3724 | * per node list when we run out of per cpu objects. We only fetch | |
3725 | * 50% to keep some capacity around for frees. | |
3726 | */ | |
3727 | if (!kmem_cache_has_cpu_partial(s)) | |
bbd4e305 | 3728 | slub_set_cpu_partial(s, 0); |
e6d0e1dc | 3729 | else if (s->size >= PAGE_SIZE) |
bbd4e305 | 3730 | slub_set_cpu_partial(s, 2); |
e6d0e1dc | 3731 | else if (s->size >= 1024) |
bbd4e305 | 3732 | slub_set_cpu_partial(s, 6); |
e6d0e1dc | 3733 | else if (s->size >= 256) |
bbd4e305 | 3734 | slub_set_cpu_partial(s, 13); |
e6d0e1dc | 3735 | else |
bbd4e305 | 3736 | slub_set_cpu_partial(s, 30); |
e6d0e1dc WY |
3737 | #endif |
3738 | } | |
3739 | ||
81819f0f CL |
3740 | /* |
3741 | * calculate_sizes() determines the order and the distribution of data within | |
3742 | * a slab object. | |
3743 | */ | |
06b285dc | 3744 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f | 3745 | { |
d50112ed | 3746 | slab_flags_t flags = s->flags; |
be4a7988 | 3747 | unsigned int size = s->object_size; |
19af27af | 3748 | unsigned int order; |
81819f0f | 3749 | |
d8b42bf5 CL |
3750 | /* |
3751 | * Round up object size to the next word boundary. We can only | |
3752 | * place the free pointer at word boundaries and this determines | |
3753 | * the possible location of the free pointer. | |
3754 | */ | |
3755 | size = ALIGN(size, sizeof(void *)); | |
3756 | ||
3757 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3758 | /* |
3759 | * Determine if we can poison the object itself. If the user of | |
3760 | * the slab may touch the object after free or before allocation | |
3761 | * then we should never poison the object itself. | |
3762 | */ | |
5f0d5a3a | 3763 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 3764 | !s->ctor) |
81819f0f CL |
3765 | s->flags |= __OBJECT_POISON; |
3766 | else | |
3767 | s->flags &= ~__OBJECT_POISON; | |
3768 | ||
81819f0f CL |
3769 | |
3770 | /* | |
672bba3a | 3771 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3772 | * end of the object and the free pointer. If not then add an |
672bba3a | 3773 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3774 | */ |
3b0efdfa | 3775 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3776 | size += sizeof(void *); |
41ecc55b | 3777 | #endif |
81819f0f CL |
3778 | |
3779 | /* | |
672bba3a | 3780 | * With that we have determined the number of bytes in actual use |
e41a49fa | 3781 | * by the object and redzoning. |
81819f0f CL |
3782 | */ |
3783 | s->inuse = size; | |
3784 | ||
74c1d3e0 KC |
3785 | if ((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
3786 | ((flags & SLAB_RED_ZONE) && s->object_size < sizeof(void *)) || | |
3787 | s->ctor) { | |
81819f0f CL |
3788 | /* |
3789 | * Relocate free pointer after the object if it is not | |
3790 | * permitted to overwrite the first word of the object on | |
3791 | * kmem_cache_free. | |
3792 | * | |
3793 | * This is the case if we do RCU, have a constructor or | |
74c1d3e0 KC |
3794 | * destructor, are poisoning the objects, or are |
3795 | * redzoning an object smaller than sizeof(void *). | |
cbfc35a4 WL |
3796 | * |
3797 | * The assumption that s->offset >= s->inuse means free | |
3798 | * pointer is outside of the object is used in the | |
3799 | * freeptr_outside_object() function. If that is no | |
3800 | * longer true, the function needs to be modified. | |
81819f0f CL |
3801 | */ |
3802 | s->offset = size; | |
3803 | size += sizeof(void *); | |
e41a49fa | 3804 | } else { |
3202fa62 KC |
3805 | /* |
3806 | * Store freelist pointer near middle of object to keep | |
3807 | * it away from the edges of the object to avoid small | |
3808 | * sized over/underflows from neighboring allocations. | |
3809 | */ | |
e41a49fa | 3810 | s->offset = ALIGN_DOWN(s->object_size / 2, sizeof(void *)); |
81819f0f CL |
3811 | } |
3812 | ||
c12b3c62 | 3813 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3814 | if (flags & SLAB_STORE_USER) |
3815 | /* | |
3816 | * Need to store information about allocs and frees after | |
3817 | * the object. | |
3818 | */ | |
3819 | size += 2 * sizeof(struct track); | |
80a9201a | 3820 | #endif |
81819f0f | 3821 | |
80a9201a AP |
3822 | kasan_cache_create(s, &size, &s->flags); |
3823 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 3824 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3825 | /* |
3826 | * Add some empty padding so that we can catch | |
3827 | * overwrites from earlier objects rather than let | |
3828 | * tracking information or the free pointer be | |
0211a9c8 | 3829 | * corrupted if a user writes before the start |
81819f0f CL |
3830 | * of the object. |
3831 | */ | |
3832 | size += sizeof(void *); | |
d86bd1be JK |
3833 | |
3834 | s->red_left_pad = sizeof(void *); | |
3835 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3836 | size += s->red_left_pad; | |
3837 | } | |
41ecc55b | 3838 | #endif |
672bba3a | 3839 | |
81819f0f CL |
3840 | /* |
3841 | * SLUB stores one object immediately after another beginning from | |
3842 | * offset 0. In order to align the objects we have to simply size | |
3843 | * each object to conform to the alignment. | |
3844 | */ | |
45906855 | 3845 | size = ALIGN(size, s->align); |
81819f0f | 3846 | s->size = size; |
4138fdfc | 3847 | s->reciprocal_size = reciprocal_value(size); |
06b285dc CL |
3848 | if (forced_order >= 0) |
3849 | order = forced_order; | |
3850 | else | |
9736d2a9 | 3851 | order = calculate_order(size); |
81819f0f | 3852 | |
19af27af | 3853 | if ((int)order < 0) |
81819f0f CL |
3854 | return 0; |
3855 | ||
b7a49f0d | 3856 | s->allocflags = 0; |
834f3d11 | 3857 | if (order) |
b7a49f0d CL |
3858 | s->allocflags |= __GFP_COMP; |
3859 | ||
3860 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3861 | s->allocflags |= GFP_DMA; |
b7a49f0d | 3862 | |
6d6ea1e9 NB |
3863 | if (s->flags & SLAB_CACHE_DMA32) |
3864 | s->allocflags |= GFP_DMA32; | |
3865 | ||
b7a49f0d CL |
3866 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
3867 | s->allocflags |= __GFP_RECLAIMABLE; | |
3868 | ||
81819f0f CL |
3869 | /* |
3870 | * Determine the number of objects per slab | |
3871 | */ | |
9736d2a9 MW |
3872 | s->oo = oo_make(order, size); |
3873 | s->min = oo_make(get_order(size), size); | |
205ab99d CL |
3874 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3875 | s->max = s->oo; | |
81819f0f | 3876 | |
834f3d11 | 3877 | return !!oo_objects(s->oo); |
81819f0f CL |
3878 | } |
3879 | ||
d50112ed | 3880 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 3881 | { |
37540008 | 3882 | s->flags = kmem_cache_flags(s->size, flags, s->name); |
2482ddec KC |
3883 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
3884 | s->random = get_random_long(); | |
3885 | #endif | |
81819f0f | 3886 | |
06b285dc | 3887 | if (!calculate_sizes(s, -1)) |
81819f0f | 3888 | goto error; |
3de47213 DR |
3889 | if (disable_higher_order_debug) { |
3890 | /* | |
3891 | * Disable debugging flags that store metadata if the min slab | |
3892 | * order increased. | |
3893 | */ | |
3b0efdfa | 3894 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3895 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3896 | s->offset = 0; | |
3897 | if (!calculate_sizes(s, -1)) | |
3898 | goto error; | |
3899 | } | |
3900 | } | |
81819f0f | 3901 | |
2565409f HC |
3902 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3903 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3904 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3905 | /* Enable fast mode */ |
3906 | s->flags |= __CMPXCHG_DOUBLE; | |
3907 | #endif | |
3908 | ||
3b89d7d8 DR |
3909 | /* |
3910 | * The larger the object size is, the more pages we want on the partial | |
3911 | * list to avoid pounding the page allocator excessively. | |
3912 | */ | |
49e22585 CL |
3913 | set_min_partial(s, ilog2(s->size) / 2); |
3914 | ||
e6d0e1dc | 3915 | set_cpu_partial(s); |
49e22585 | 3916 | |
81819f0f | 3917 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3918 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3919 | #endif |
210e7a43 TG |
3920 | |
3921 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3922 | if (slab_state >= UP) { | |
3923 | if (init_cache_random_seq(s)) | |
3924 | goto error; | |
3925 | } | |
3926 | ||
55136592 | 3927 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3928 | goto error; |
81819f0f | 3929 | |
55136592 | 3930 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3931 | return 0; |
ff12059e | 3932 | |
4c93c355 | 3933 | free_kmem_cache_nodes(s); |
81819f0f | 3934 | error: |
278b1bb1 | 3935 | return -EINVAL; |
81819f0f | 3936 | } |
81819f0f | 3937 | |
33b12c38 | 3938 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
55860d96 | 3939 | const char *text) |
33b12c38 CL |
3940 | { |
3941 | #ifdef CONFIG_SLUB_DEBUG | |
3942 | void *addr = page_address(page); | |
55860d96 | 3943 | unsigned long *map; |
33b12c38 | 3944 | void *p; |
aa456c7a | 3945 | |
945cf2b6 | 3946 | slab_err(s, page, text, s->name); |
33b12c38 | 3947 | slab_lock(page); |
33b12c38 | 3948 | |
90e9f6a6 | 3949 | map = get_map(s, page); |
33b12c38 CL |
3950 | for_each_object(p, s, addr, page->objects) { |
3951 | ||
4138fdfc | 3952 | if (!test_bit(__obj_to_index(s, addr, p), map)) { |
96b94abc | 3953 | pr_err("Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3954 | print_tracking(s, p); |
3955 | } | |
3956 | } | |
55860d96 | 3957 | put_map(map); |
33b12c38 CL |
3958 | slab_unlock(page); |
3959 | #endif | |
3960 | } | |
3961 | ||
81819f0f | 3962 | /* |
599870b1 | 3963 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3964 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3965 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3966 | */ |
599870b1 | 3967 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3968 | { |
60398923 | 3969 | LIST_HEAD(discard); |
81819f0f CL |
3970 | struct page *page, *h; |
3971 | ||
52b4b950 DS |
3972 | BUG_ON(irqs_disabled()); |
3973 | spin_lock_irq(&n->list_lock); | |
916ac052 | 3974 | list_for_each_entry_safe(page, h, &n->partial, slab_list) { |
81819f0f | 3975 | if (!page->inuse) { |
52b4b950 | 3976 | remove_partial(n, page); |
916ac052 | 3977 | list_add(&page->slab_list, &discard); |
33b12c38 CL |
3978 | } else { |
3979 | list_slab_objects(s, page, | |
55860d96 | 3980 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3981 | } |
33b12c38 | 3982 | } |
52b4b950 | 3983 | spin_unlock_irq(&n->list_lock); |
60398923 | 3984 | |
916ac052 | 3985 | list_for_each_entry_safe(page, h, &discard, slab_list) |
60398923 | 3986 | discard_slab(s, page); |
81819f0f CL |
3987 | } |
3988 | ||
f9e13c0a SB |
3989 | bool __kmem_cache_empty(struct kmem_cache *s) |
3990 | { | |
3991 | int node; | |
3992 | struct kmem_cache_node *n; | |
3993 | ||
3994 | for_each_kmem_cache_node(s, node, n) | |
3995 | if (n->nr_partial || slabs_node(s, node)) | |
3996 | return false; | |
3997 | return true; | |
3998 | } | |
3999 | ||
81819f0f | 4000 | /* |
672bba3a | 4001 | * Release all resources used by a slab cache. |
81819f0f | 4002 | */ |
52b4b950 | 4003 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
4004 | { |
4005 | int node; | |
fa45dc25 | 4006 | struct kmem_cache_node *n; |
81819f0f CL |
4007 | |
4008 | flush_all(s); | |
81819f0f | 4009 | /* Attempt to free all objects */ |
fa45dc25 | 4010 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
4011 | free_partial(s, n); |
4012 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
4013 | return 1; |
4014 | } | |
81819f0f CL |
4015 | return 0; |
4016 | } | |
4017 | ||
5bb1bb35 | 4018 | #ifdef CONFIG_PRINTK |
8e7f37f2 PM |
4019 | void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct page *page) |
4020 | { | |
4021 | void *base; | |
4022 | int __maybe_unused i; | |
4023 | unsigned int objnr; | |
4024 | void *objp; | |
4025 | void *objp0; | |
4026 | struct kmem_cache *s = page->slab_cache; | |
4027 | struct track __maybe_unused *trackp; | |
4028 | ||
4029 | kpp->kp_ptr = object; | |
4030 | kpp->kp_page = page; | |
4031 | kpp->kp_slab_cache = s; | |
4032 | base = page_address(page); | |
4033 | objp0 = kasan_reset_tag(object); | |
4034 | #ifdef CONFIG_SLUB_DEBUG | |
4035 | objp = restore_red_left(s, objp0); | |
4036 | #else | |
4037 | objp = objp0; | |
4038 | #endif | |
4039 | objnr = obj_to_index(s, page, objp); | |
4040 | kpp->kp_data_offset = (unsigned long)((char *)objp0 - (char *)objp); | |
4041 | objp = base + s->size * objnr; | |
4042 | kpp->kp_objp = objp; | |
4043 | if (WARN_ON_ONCE(objp < base || objp >= base + page->objects * s->size || (objp - base) % s->size) || | |
4044 | !(s->flags & SLAB_STORE_USER)) | |
4045 | return; | |
4046 | #ifdef CONFIG_SLUB_DEBUG | |
4047 | trackp = get_track(s, objp, TRACK_ALLOC); | |
4048 | kpp->kp_ret = (void *)trackp->addr; | |
4049 | #ifdef CONFIG_STACKTRACE | |
4050 | for (i = 0; i < KS_ADDRS_COUNT && i < TRACK_ADDRS_COUNT; i++) { | |
4051 | kpp->kp_stack[i] = (void *)trackp->addrs[i]; | |
4052 | if (!kpp->kp_stack[i]) | |
4053 | break; | |
4054 | } | |
4055 | #endif | |
4056 | #endif | |
4057 | } | |
5bb1bb35 | 4058 | #endif |
8e7f37f2 | 4059 | |
81819f0f CL |
4060 | /******************************************************************** |
4061 | * Kmalloc subsystem | |
4062 | *******************************************************************/ | |
4063 | ||
81819f0f CL |
4064 | static int __init setup_slub_min_order(char *str) |
4065 | { | |
19af27af | 4066 | get_option(&str, (int *)&slub_min_order); |
81819f0f CL |
4067 | |
4068 | return 1; | |
4069 | } | |
4070 | ||
4071 | __setup("slub_min_order=", setup_slub_min_order); | |
4072 | ||
4073 | static int __init setup_slub_max_order(char *str) | |
4074 | { | |
19af27af AD |
4075 | get_option(&str, (int *)&slub_max_order); |
4076 | slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1); | |
81819f0f CL |
4077 | |
4078 | return 1; | |
4079 | } | |
4080 | ||
4081 | __setup("slub_max_order=", setup_slub_max_order); | |
4082 | ||
4083 | static int __init setup_slub_min_objects(char *str) | |
4084 | { | |
19af27af | 4085 | get_option(&str, (int *)&slub_min_objects); |
81819f0f CL |
4086 | |
4087 | return 1; | |
4088 | } | |
4089 | ||
4090 | __setup("slub_min_objects=", setup_slub_min_objects); | |
4091 | ||
81819f0f CL |
4092 | void *__kmalloc(size_t size, gfp_t flags) |
4093 | { | |
aadb4bc4 | 4094 | struct kmem_cache *s; |
5b882be4 | 4095 | void *ret; |
81819f0f | 4096 | |
95a05b42 | 4097 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 4098 | return kmalloc_large(size, flags); |
aadb4bc4 | 4099 | |
2c59dd65 | 4100 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
4101 | |
4102 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
4103 | return s; |
4104 | ||
b89fb5ef | 4105 | ret = slab_alloc(s, flags, _RET_IP_, size); |
5b882be4 | 4106 | |
ca2b84cb | 4107 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 4108 | |
0116523c | 4109 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 4110 | |
5b882be4 | 4111 | return ret; |
81819f0f CL |
4112 | } |
4113 | EXPORT_SYMBOL(__kmalloc); | |
4114 | ||
5d1f57e4 | 4115 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
4116 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
4117 | { | |
b1eeab67 | 4118 | struct page *page; |
e4f7c0b4 | 4119 | void *ptr = NULL; |
6a486c0a | 4120 | unsigned int order = get_order(size); |
f619cfe1 | 4121 | |
75f296d9 | 4122 | flags |= __GFP_COMP; |
6a486c0a VB |
4123 | page = alloc_pages_node(node, flags, order); |
4124 | if (page) { | |
e4f7c0b4 | 4125 | ptr = page_address(page); |
96403bfe MS |
4126 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, |
4127 | PAGE_SIZE << order); | |
6a486c0a | 4128 | } |
e4f7c0b4 | 4129 | |
0116523c | 4130 | return kmalloc_large_node_hook(ptr, size, flags); |
f619cfe1 CL |
4131 | } |
4132 | ||
81819f0f CL |
4133 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
4134 | { | |
aadb4bc4 | 4135 | struct kmem_cache *s; |
5b882be4 | 4136 | void *ret; |
81819f0f | 4137 | |
95a05b42 | 4138 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
4139 | ret = kmalloc_large_node(size, flags, node); |
4140 | ||
ca2b84cb EGM |
4141 | trace_kmalloc_node(_RET_IP_, ret, |
4142 | size, PAGE_SIZE << get_order(size), | |
4143 | flags, node); | |
5b882be4 EGM |
4144 | |
4145 | return ret; | |
4146 | } | |
aadb4bc4 | 4147 | |
2c59dd65 | 4148 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
4149 | |
4150 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
4151 | return s; |
4152 | ||
b89fb5ef | 4153 | ret = slab_alloc_node(s, flags, node, _RET_IP_, size); |
5b882be4 | 4154 | |
ca2b84cb | 4155 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 4156 | |
0116523c | 4157 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 4158 | |
5b882be4 | 4159 | return ret; |
81819f0f CL |
4160 | } |
4161 | EXPORT_SYMBOL(__kmalloc_node); | |
6dfd1b65 | 4162 | #endif /* CONFIG_NUMA */ |
81819f0f | 4163 | |
ed18adc1 KC |
4164 | #ifdef CONFIG_HARDENED_USERCOPY |
4165 | /* | |
afcc90f8 KC |
4166 | * Rejects incorrectly sized objects and objects that are to be copied |
4167 | * to/from userspace but do not fall entirely within the containing slab | |
4168 | * cache's usercopy region. | |
ed18adc1 KC |
4169 | * |
4170 | * Returns NULL if check passes, otherwise const char * to name of cache | |
4171 | * to indicate an error. | |
4172 | */ | |
f4e6e289 KC |
4173 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, |
4174 | bool to_user) | |
ed18adc1 KC |
4175 | { |
4176 | struct kmem_cache *s; | |
44065b2e | 4177 | unsigned int offset; |
ed18adc1 | 4178 | size_t object_size; |
b89fb5ef | 4179 | bool is_kfence = is_kfence_address(ptr); |
ed18adc1 | 4180 | |
96fedce2 AK |
4181 | ptr = kasan_reset_tag(ptr); |
4182 | ||
ed18adc1 KC |
4183 | /* Find object and usable object size. */ |
4184 | s = page->slab_cache; | |
ed18adc1 KC |
4185 | |
4186 | /* Reject impossible pointers. */ | |
4187 | if (ptr < page_address(page)) | |
f4e6e289 KC |
4188 | usercopy_abort("SLUB object not in SLUB page?!", NULL, |
4189 | to_user, 0, n); | |
ed18adc1 KC |
4190 | |
4191 | /* Find offset within object. */ | |
b89fb5ef AP |
4192 | if (is_kfence) |
4193 | offset = ptr - kfence_object_start(ptr); | |
4194 | else | |
4195 | offset = (ptr - page_address(page)) % s->size; | |
ed18adc1 KC |
4196 | |
4197 | /* Adjust for redzone and reject if within the redzone. */ | |
b89fb5ef | 4198 | if (!is_kfence && kmem_cache_debug_flags(s, SLAB_RED_ZONE)) { |
ed18adc1 | 4199 | if (offset < s->red_left_pad) |
f4e6e289 KC |
4200 | usercopy_abort("SLUB object in left red zone", |
4201 | s->name, to_user, offset, n); | |
ed18adc1 KC |
4202 | offset -= s->red_left_pad; |
4203 | } | |
4204 | ||
afcc90f8 KC |
4205 | /* Allow address range falling entirely within usercopy region. */ |
4206 | if (offset >= s->useroffset && | |
4207 | offset - s->useroffset <= s->usersize && | |
4208 | n <= s->useroffset - offset + s->usersize) | |
f4e6e289 | 4209 | return; |
ed18adc1 | 4210 | |
afcc90f8 KC |
4211 | /* |
4212 | * If the copy is still within the allocated object, produce | |
4213 | * a warning instead of rejecting the copy. This is intended | |
4214 | * to be a temporary method to find any missing usercopy | |
4215 | * whitelists. | |
4216 | */ | |
4217 | object_size = slab_ksize(s); | |
2d891fbc KC |
4218 | if (usercopy_fallback && |
4219 | offset <= object_size && n <= object_size - offset) { | |
afcc90f8 KC |
4220 | usercopy_warn("SLUB object", s->name, to_user, offset, n); |
4221 | return; | |
4222 | } | |
ed18adc1 | 4223 | |
f4e6e289 | 4224 | usercopy_abort("SLUB object", s->name, to_user, offset, n); |
ed18adc1 KC |
4225 | } |
4226 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
4227 | ||
10d1f8cb | 4228 | size_t __ksize(const void *object) |
81819f0f | 4229 | { |
272c1d21 | 4230 | struct page *page; |
81819f0f | 4231 | |
ef8b4520 | 4232 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
4233 | return 0; |
4234 | ||
294a80a8 | 4235 | page = virt_to_head_page(object); |
294a80a8 | 4236 | |
76994412 PE |
4237 | if (unlikely(!PageSlab(page))) { |
4238 | WARN_ON(!PageCompound(page)); | |
a50b854e | 4239 | return page_size(page); |
76994412 | 4240 | } |
81819f0f | 4241 | |
1b4f59e3 | 4242 | return slab_ksize(page->slab_cache); |
81819f0f | 4243 | } |
10d1f8cb | 4244 | EXPORT_SYMBOL(__ksize); |
81819f0f CL |
4245 | |
4246 | void kfree(const void *x) | |
4247 | { | |
81819f0f | 4248 | struct page *page; |
5bb983b0 | 4249 | void *object = (void *)x; |
81819f0f | 4250 | |
2121db74 PE |
4251 | trace_kfree(_RET_IP_, x); |
4252 | ||
2408c550 | 4253 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
4254 | return; |
4255 | ||
b49af68f | 4256 | page = virt_to_head_page(x); |
aadb4bc4 | 4257 | if (unlikely(!PageSlab(page))) { |
6a486c0a VB |
4258 | unsigned int order = compound_order(page); |
4259 | ||
0937502a | 4260 | BUG_ON(!PageCompound(page)); |
47adccce | 4261 | kfree_hook(object); |
96403bfe MS |
4262 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, |
4263 | -(PAGE_SIZE << order)); | |
6a486c0a | 4264 | __free_pages(page, order); |
aadb4bc4 CL |
4265 | return; |
4266 | } | |
81084651 | 4267 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
4268 | } |
4269 | EXPORT_SYMBOL(kfree); | |
4270 | ||
832f37f5 VD |
4271 | #define SHRINK_PROMOTE_MAX 32 |
4272 | ||
2086d26a | 4273 | /* |
832f37f5 VD |
4274 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
4275 | * up most to the head of the partial lists. New allocations will then | |
4276 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
4277 | * |
4278 | * The slabs with the least items are placed last. This results in them | |
4279 | * being allocated from last increasing the chance that the last objects | |
4280 | * are freed in them. | |
2086d26a | 4281 | */ |
c9fc5864 | 4282 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
4283 | { |
4284 | int node; | |
4285 | int i; | |
4286 | struct kmem_cache_node *n; | |
4287 | struct page *page; | |
4288 | struct page *t; | |
832f37f5 VD |
4289 | struct list_head discard; |
4290 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 4291 | unsigned long flags; |
ce3712d7 | 4292 | int ret = 0; |
2086d26a | 4293 | |
2086d26a | 4294 | flush_all(s); |
fa45dc25 | 4295 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
4296 | INIT_LIST_HEAD(&discard); |
4297 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
4298 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
4299 | |
4300 | spin_lock_irqsave(&n->list_lock, flags); | |
4301 | ||
4302 | /* | |
832f37f5 | 4303 | * Build lists of slabs to discard or promote. |
2086d26a | 4304 | * |
672bba3a CL |
4305 | * Note that concurrent frees may occur while we hold the |
4306 | * list_lock. page->inuse here is the upper limit. | |
2086d26a | 4307 | */ |
916ac052 | 4308 | list_for_each_entry_safe(page, t, &n->partial, slab_list) { |
832f37f5 VD |
4309 | int free = page->objects - page->inuse; |
4310 | ||
4311 | /* Do not reread page->inuse */ | |
4312 | barrier(); | |
4313 | ||
4314 | /* We do not keep full slabs on the list */ | |
4315 | BUG_ON(free <= 0); | |
4316 | ||
4317 | if (free == page->objects) { | |
916ac052 | 4318 | list_move(&page->slab_list, &discard); |
69cb8e6b | 4319 | n->nr_partial--; |
832f37f5 | 4320 | } else if (free <= SHRINK_PROMOTE_MAX) |
916ac052 | 4321 | list_move(&page->slab_list, promote + free - 1); |
2086d26a CL |
4322 | } |
4323 | ||
2086d26a | 4324 | /* |
832f37f5 VD |
4325 | * Promote the slabs filled up most to the head of the |
4326 | * partial list. | |
2086d26a | 4327 | */ |
832f37f5 VD |
4328 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
4329 | list_splice(promote + i, &n->partial); | |
2086d26a | 4330 | |
2086d26a | 4331 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
4332 | |
4333 | /* Release empty slabs */ | |
916ac052 | 4334 | list_for_each_entry_safe(page, t, &discard, slab_list) |
69cb8e6b | 4335 | discard_slab(s, page); |
ce3712d7 VD |
4336 | |
4337 | if (slabs_node(s, node)) | |
4338 | ret = 1; | |
2086d26a CL |
4339 | } |
4340 | ||
ce3712d7 | 4341 | return ret; |
2086d26a | 4342 | } |
2086d26a | 4343 | |
b9049e23 YG |
4344 | static int slab_mem_going_offline_callback(void *arg) |
4345 | { | |
4346 | struct kmem_cache *s; | |
4347 | ||
18004c5d | 4348 | mutex_lock(&slab_mutex); |
b9049e23 | 4349 | list_for_each_entry(s, &slab_caches, list) |
c9fc5864 | 4350 | __kmem_cache_shrink(s); |
18004c5d | 4351 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4352 | |
4353 | return 0; | |
4354 | } | |
4355 | ||
4356 | static void slab_mem_offline_callback(void *arg) | |
4357 | { | |
b9049e23 YG |
4358 | struct memory_notify *marg = arg; |
4359 | int offline_node; | |
4360 | ||
b9d5ab25 | 4361 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4362 | |
4363 | /* | |
4364 | * If the node still has available memory. we need kmem_cache_node | |
4365 | * for it yet. | |
4366 | */ | |
4367 | if (offline_node < 0) | |
4368 | return; | |
4369 | ||
18004c5d | 4370 | mutex_lock(&slab_mutex); |
7e1fa93d | 4371 | node_clear(offline_node, slab_nodes); |
666716fd VB |
4372 | /* |
4373 | * We no longer free kmem_cache_node structures here, as it would be | |
4374 | * racy with all get_node() users, and infeasible to protect them with | |
4375 | * slab_mutex. | |
4376 | */ | |
18004c5d | 4377 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4378 | } |
4379 | ||
4380 | static int slab_mem_going_online_callback(void *arg) | |
4381 | { | |
4382 | struct kmem_cache_node *n; | |
4383 | struct kmem_cache *s; | |
4384 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4385 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4386 | int ret = 0; |
4387 | ||
4388 | /* | |
4389 | * If the node's memory is already available, then kmem_cache_node is | |
4390 | * already created. Nothing to do. | |
4391 | */ | |
4392 | if (nid < 0) | |
4393 | return 0; | |
4394 | ||
4395 | /* | |
0121c619 | 4396 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4397 | * allocate a kmem_cache_node structure in order to bring the node |
4398 | * online. | |
4399 | */ | |
18004c5d | 4400 | mutex_lock(&slab_mutex); |
b9049e23 | 4401 | list_for_each_entry(s, &slab_caches, list) { |
666716fd VB |
4402 | /* |
4403 | * The structure may already exist if the node was previously | |
4404 | * onlined and offlined. | |
4405 | */ | |
4406 | if (get_node(s, nid)) | |
4407 | continue; | |
b9049e23 YG |
4408 | /* |
4409 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4410 | * since memory is not yet available from the node that | |
4411 | * is brought up. | |
4412 | */ | |
8de66a0c | 4413 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4414 | if (!n) { |
4415 | ret = -ENOMEM; | |
4416 | goto out; | |
4417 | } | |
4053497d | 4418 | init_kmem_cache_node(n); |
b9049e23 YG |
4419 | s->node[nid] = n; |
4420 | } | |
7e1fa93d VB |
4421 | /* |
4422 | * Any cache created after this point will also have kmem_cache_node | |
4423 | * initialized for the new node. | |
4424 | */ | |
4425 | node_set(nid, slab_nodes); | |
b9049e23 | 4426 | out: |
18004c5d | 4427 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4428 | return ret; |
4429 | } | |
4430 | ||
4431 | static int slab_memory_callback(struct notifier_block *self, | |
4432 | unsigned long action, void *arg) | |
4433 | { | |
4434 | int ret = 0; | |
4435 | ||
4436 | switch (action) { | |
4437 | case MEM_GOING_ONLINE: | |
4438 | ret = slab_mem_going_online_callback(arg); | |
4439 | break; | |
4440 | case MEM_GOING_OFFLINE: | |
4441 | ret = slab_mem_going_offline_callback(arg); | |
4442 | break; | |
4443 | case MEM_OFFLINE: | |
4444 | case MEM_CANCEL_ONLINE: | |
4445 | slab_mem_offline_callback(arg); | |
4446 | break; | |
4447 | case MEM_ONLINE: | |
4448 | case MEM_CANCEL_OFFLINE: | |
4449 | break; | |
4450 | } | |
dc19f9db KH |
4451 | if (ret) |
4452 | ret = notifier_from_errno(ret); | |
4453 | else | |
4454 | ret = NOTIFY_OK; | |
b9049e23 YG |
4455 | return ret; |
4456 | } | |
4457 | ||
3ac38faa AM |
4458 | static struct notifier_block slab_memory_callback_nb = { |
4459 | .notifier_call = slab_memory_callback, | |
4460 | .priority = SLAB_CALLBACK_PRI, | |
4461 | }; | |
b9049e23 | 4462 | |
81819f0f CL |
4463 | /******************************************************************** |
4464 | * Basic setup of slabs | |
4465 | *******************************************************************/ | |
4466 | ||
51df1142 CL |
4467 | /* |
4468 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4469 | * the page allocator. Allocate them properly then fix up the pointers |
4470 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4471 | */ |
4472 | ||
dffb4d60 | 4473 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4474 | { |
4475 | int node; | |
dffb4d60 | 4476 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4477 | struct kmem_cache_node *n; |
51df1142 | 4478 | |
dffb4d60 | 4479 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4480 | |
7d557b3c GC |
4481 | /* |
4482 | * This runs very early, and only the boot processor is supposed to be | |
4483 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4484 | * IPIs around. | |
4485 | */ | |
4486 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4487 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4488 | struct page *p; |
4489 | ||
916ac052 | 4490 | list_for_each_entry(p, &n->partial, slab_list) |
fa45dc25 | 4491 | p->slab_cache = s; |
51df1142 | 4492 | |
607bf324 | 4493 | #ifdef CONFIG_SLUB_DEBUG |
916ac052 | 4494 | list_for_each_entry(p, &n->full, slab_list) |
fa45dc25 | 4495 | p->slab_cache = s; |
51df1142 | 4496 | #endif |
51df1142 | 4497 | } |
dffb4d60 CL |
4498 | list_add(&s->list, &slab_caches); |
4499 | return s; | |
51df1142 CL |
4500 | } |
4501 | ||
81819f0f CL |
4502 | void __init kmem_cache_init(void) |
4503 | { | |
dffb4d60 CL |
4504 | static __initdata struct kmem_cache boot_kmem_cache, |
4505 | boot_kmem_cache_node; | |
7e1fa93d | 4506 | int node; |
51df1142 | 4507 | |
fc8d8620 SG |
4508 | if (debug_guardpage_minorder()) |
4509 | slub_max_order = 0; | |
4510 | ||
79270291 SB |
4511 | /* Print slub debugging pointers without hashing */ |
4512 | if (__slub_debug_enabled()) | |
4513 | no_hash_pointers_enable(NULL); | |
4514 | ||
dffb4d60 CL |
4515 | kmem_cache_node = &boot_kmem_cache_node; |
4516 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4517 | |
7e1fa93d VB |
4518 | /* |
4519 | * Initialize the nodemask for which we will allocate per node | |
4520 | * structures. Here we don't need taking slab_mutex yet. | |
4521 | */ | |
4522 | for_each_node_state(node, N_NORMAL_MEMORY) | |
4523 | node_set(node, slab_nodes); | |
4524 | ||
dffb4d60 | 4525 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
8eb8284b | 4526 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0); |
b9049e23 | 4527 | |
3ac38faa | 4528 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4529 | |
4530 | /* Able to allocate the per node structures */ | |
4531 | slab_state = PARTIAL; | |
4532 | ||
dffb4d60 CL |
4533 | create_boot_cache(kmem_cache, "kmem_cache", |
4534 | offsetof(struct kmem_cache, node) + | |
4535 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
8eb8284b | 4536 | SLAB_HWCACHE_ALIGN, 0, 0); |
8a13a4cc | 4537 | |
dffb4d60 | 4538 | kmem_cache = bootstrap(&boot_kmem_cache); |
dffb4d60 | 4539 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4540 | |
4541 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4542 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4543 | create_kmalloc_caches(0); |
81819f0f | 4544 | |
210e7a43 TG |
4545 | /* Setup random freelists for each cache */ |
4546 | init_freelist_randomization(); | |
4547 | ||
a96a87bf SAS |
4548 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4549 | slub_cpu_dead); | |
81819f0f | 4550 | |
b9726c26 | 4551 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n", |
f97d5f63 | 4552 | cache_line_size(), |
81819f0f CL |
4553 | slub_min_order, slub_max_order, slub_min_objects, |
4554 | nr_cpu_ids, nr_node_ids); | |
4555 | } | |
4556 | ||
7e85ee0c PE |
4557 | void __init kmem_cache_init_late(void) |
4558 | { | |
7e85ee0c PE |
4559 | } |
4560 | ||
2633d7a0 | 4561 | struct kmem_cache * |
f4957d5b | 4562 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 4563 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 4564 | { |
10befea9 | 4565 | struct kmem_cache *s; |
81819f0f | 4566 | |
a44cb944 | 4567 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4568 | if (s) { |
4569 | s->refcount++; | |
84d0ddd6 | 4570 | |
81819f0f CL |
4571 | /* |
4572 | * Adjust the object sizes so that we clear | |
4573 | * the complete object on kzalloc. | |
4574 | */ | |
1b473f29 | 4575 | s->object_size = max(s->object_size, size); |
52ee6d74 | 4576 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4577 | |
7b8f3b66 | 4578 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4579 | s->refcount--; |
cbb79694 | 4580 | s = NULL; |
7b8f3b66 | 4581 | } |
a0e1d1be | 4582 | } |
6446faa2 | 4583 | |
cbb79694 CL |
4584 | return s; |
4585 | } | |
84c1cf62 | 4586 | |
d50112ed | 4587 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 4588 | { |
aac3a166 PE |
4589 | int err; |
4590 | ||
4591 | err = kmem_cache_open(s, flags); | |
4592 | if (err) | |
4593 | return err; | |
20cea968 | 4594 | |
45530c44 CL |
4595 | /* Mutex is not taken during early boot */ |
4596 | if (slab_state <= UP) | |
4597 | return 0; | |
4598 | ||
aac3a166 | 4599 | err = sysfs_slab_add(s); |
aac3a166 | 4600 | if (err) |
52b4b950 | 4601 | __kmem_cache_release(s); |
20cea968 | 4602 | |
64dd6849 FM |
4603 | if (s->flags & SLAB_STORE_USER) |
4604 | debugfs_slab_add(s); | |
4605 | ||
aac3a166 | 4606 | return err; |
81819f0f | 4607 | } |
81819f0f | 4608 | |
ce71e27c | 4609 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4610 | { |
aadb4bc4 | 4611 | struct kmem_cache *s; |
94b528d0 | 4612 | void *ret; |
aadb4bc4 | 4613 | |
95a05b42 | 4614 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4615 | return kmalloc_large(size, gfpflags); |
4616 | ||
2c59dd65 | 4617 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4618 | |
2408c550 | 4619 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4620 | return s; |
81819f0f | 4621 | |
b89fb5ef | 4622 | ret = slab_alloc(s, gfpflags, caller, size); |
94b528d0 | 4623 | |
25985edc | 4624 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4625 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4626 | |
4627 | return ret; | |
81819f0f | 4628 | } |
fd7cb575 | 4629 | EXPORT_SYMBOL(__kmalloc_track_caller); |
81819f0f | 4630 | |
5d1f57e4 | 4631 | #ifdef CONFIG_NUMA |
81819f0f | 4632 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4633 | int node, unsigned long caller) |
81819f0f | 4634 | { |
aadb4bc4 | 4635 | struct kmem_cache *s; |
94b528d0 | 4636 | void *ret; |
aadb4bc4 | 4637 | |
95a05b42 | 4638 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4639 | ret = kmalloc_large_node(size, gfpflags, node); |
4640 | ||
4641 | trace_kmalloc_node(caller, ret, | |
4642 | size, PAGE_SIZE << get_order(size), | |
4643 | gfpflags, node); | |
4644 | ||
4645 | return ret; | |
4646 | } | |
eada35ef | 4647 | |
2c59dd65 | 4648 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4649 | |
2408c550 | 4650 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4651 | return s; |
81819f0f | 4652 | |
b89fb5ef | 4653 | ret = slab_alloc_node(s, gfpflags, node, caller, size); |
94b528d0 | 4654 | |
25985edc | 4655 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4656 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4657 | |
4658 | return ret; | |
81819f0f | 4659 | } |
fd7cb575 | 4660 | EXPORT_SYMBOL(__kmalloc_node_track_caller); |
5d1f57e4 | 4661 | #endif |
81819f0f | 4662 | |
ab4d5ed5 | 4663 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4664 | static int count_inuse(struct page *page) |
4665 | { | |
4666 | return page->inuse; | |
4667 | } | |
4668 | ||
4669 | static int count_total(struct page *page) | |
4670 | { | |
4671 | return page->objects; | |
4672 | } | |
ab4d5ed5 | 4673 | #endif |
205ab99d | 4674 | |
ab4d5ed5 | 4675 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 | 4676 | static void validate_slab(struct kmem_cache *s, struct page *page) |
53e15af0 CL |
4677 | { |
4678 | void *p; | |
a973e9dd | 4679 | void *addr = page_address(page); |
90e9f6a6 YZ |
4680 | unsigned long *map; |
4681 | ||
4682 | slab_lock(page); | |
53e15af0 | 4683 | |
dd98afd4 | 4684 | if (!check_slab(s, page) || !on_freelist(s, page, NULL)) |
90e9f6a6 | 4685 | goto unlock; |
53e15af0 CL |
4686 | |
4687 | /* Now we know that a valid freelist exists */ | |
90e9f6a6 | 4688 | map = get_map(s, page); |
5f80b13a | 4689 | for_each_object(p, s, addr, page->objects) { |
4138fdfc | 4690 | u8 val = test_bit(__obj_to_index(s, addr, p), map) ? |
dd98afd4 | 4691 | SLUB_RED_INACTIVE : SLUB_RED_ACTIVE; |
53e15af0 | 4692 | |
dd98afd4 YZ |
4693 | if (!check_object(s, page, p, val)) |
4694 | break; | |
4695 | } | |
90e9f6a6 YZ |
4696 | put_map(map); |
4697 | unlock: | |
881db7fb | 4698 | slab_unlock(page); |
53e15af0 CL |
4699 | } |
4700 | ||
434e245d | 4701 | static int validate_slab_node(struct kmem_cache *s, |
90e9f6a6 | 4702 | struct kmem_cache_node *n) |
53e15af0 CL |
4703 | { |
4704 | unsigned long count = 0; | |
4705 | struct page *page; | |
4706 | unsigned long flags; | |
4707 | ||
4708 | spin_lock_irqsave(&n->list_lock, flags); | |
4709 | ||
916ac052 | 4710 | list_for_each_entry(page, &n->partial, slab_list) { |
90e9f6a6 | 4711 | validate_slab(s, page); |
53e15af0 CL |
4712 | count++; |
4713 | } | |
1f9f78b1 | 4714 | if (count != n->nr_partial) { |
f9f58285 FF |
4715 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4716 | s->name, count, n->nr_partial); | |
1f9f78b1 OG |
4717 | slab_add_kunit_errors(); |
4718 | } | |
53e15af0 CL |
4719 | |
4720 | if (!(s->flags & SLAB_STORE_USER)) | |
4721 | goto out; | |
4722 | ||
916ac052 | 4723 | list_for_each_entry(page, &n->full, slab_list) { |
90e9f6a6 | 4724 | validate_slab(s, page); |
53e15af0 CL |
4725 | count++; |
4726 | } | |
1f9f78b1 | 4727 | if (count != atomic_long_read(&n->nr_slabs)) { |
f9f58285 FF |
4728 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4729 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
1f9f78b1 OG |
4730 | slab_add_kunit_errors(); |
4731 | } | |
53e15af0 CL |
4732 | |
4733 | out: | |
4734 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4735 | return count; | |
4736 | } | |
4737 | ||
1f9f78b1 | 4738 | long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4739 | { |
4740 | int node; | |
4741 | unsigned long count = 0; | |
fa45dc25 | 4742 | struct kmem_cache_node *n; |
53e15af0 CL |
4743 | |
4744 | flush_all(s); | |
fa45dc25 | 4745 | for_each_kmem_cache_node(s, node, n) |
90e9f6a6 YZ |
4746 | count += validate_slab_node(s, n); |
4747 | ||
53e15af0 CL |
4748 | return count; |
4749 | } | |
1f9f78b1 OG |
4750 | EXPORT_SYMBOL(validate_slab_cache); |
4751 | ||
64dd6849 | 4752 | #ifdef CONFIG_DEBUG_FS |
88a420e4 | 4753 | /* |
672bba3a | 4754 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4755 | * and freed. |
4756 | */ | |
4757 | ||
4758 | struct location { | |
4759 | unsigned long count; | |
ce71e27c | 4760 | unsigned long addr; |
45edfa58 CL |
4761 | long long sum_time; |
4762 | long min_time; | |
4763 | long max_time; | |
4764 | long min_pid; | |
4765 | long max_pid; | |
174596a0 | 4766 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4767 | nodemask_t nodes; |
88a420e4 CL |
4768 | }; |
4769 | ||
4770 | struct loc_track { | |
4771 | unsigned long max; | |
4772 | unsigned long count; | |
4773 | struct location *loc; | |
4774 | }; | |
4775 | ||
64dd6849 FM |
4776 | static struct dentry *slab_debugfs_root; |
4777 | ||
88a420e4 CL |
4778 | static void free_loc_track(struct loc_track *t) |
4779 | { | |
4780 | if (t->max) | |
4781 | free_pages((unsigned long)t->loc, | |
4782 | get_order(sizeof(struct location) * t->max)); | |
4783 | } | |
4784 | ||
68dff6a9 | 4785 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4786 | { |
4787 | struct location *l; | |
4788 | int order; | |
4789 | ||
88a420e4 CL |
4790 | order = get_order(sizeof(struct location) * max); |
4791 | ||
68dff6a9 | 4792 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4793 | if (!l) |
4794 | return 0; | |
4795 | ||
4796 | if (t->count) { | |
4797 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4798 | free_loc_track(t); | |
4799 | } | |
4800 | t->max = max; | |
4801 | t->loc = l; | |
4802 | return 1; | |
4803 | } | |
4804 | ||
4805 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4806 | const struct track *track) |
88a420e4 CL |
4807 | { |
4808 | long start, end, pos; | |
4809 | struct location *l; | |
ce71e27c | 4810 | unsigned long caddr; |
45edfa58 | 4811 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4812 | |
4813 | start = -1; | |
4814 | end = t->count; | |
4815 | ||
4816 | for ( ; ; ) { | |
4817 | pos = start + (end - start + 1) / 2; | |
4818 | ||
4819 | /* | |
4820 | * There is nothing at "end". If we end up there | |
4821 | * we need to add something to before end. | |
4822 | */ | |
4823 | if (pos == end) | |
4824 | break; | |
4825 | ||
4826 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4827 | if (track->addr == caddr) { |
4828 | ||
4829 | l = &t->loc[pos]; | |
4830 | l->count++; | |
4831 | if (track->when) { | |
4832 | l->sum_time += age; | |
4833 | if (age < l->min_time) | |
4834 | l->min_time = age; | |
4835 | if (age > l->max_time) | |
4836 | l->max_time = age; | |
4837 | ||
4838 | if (track->pid < l->min_pid) | |
4839 | l->min_pid = track->pid; | |
4840 | if (track->pid > l->max_pid) | |
4841 | l->max_pid = track->pid; | |
4842 | ||
174596a0 RR |
4843 | cpumask_set_cpu(track->cpu, |
4844 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4845 | } |
4846 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4847 | return 1; |
4848 | } | |
4849 | ||
45edfa58 | 4850 | if (track->addr < caddr) |
88a420e4 CL |
4851 | end = pos; |
4852 | else | |
4853 | start = pos; | |
4854 | } | |
4855 | ||
4856 | /* | |
672bba3a | 4857 | * Not found. Insert new tracking element. |
88a420e4 | 4858 | */ |
68dff6a9 | 4859 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4860 | return 0; |
4861 | ||
4862 | l = t->loc + pos; | |
4863 | if (pos < t->count) | |
4864 | memmove(l + 1, l, | |
4865 | (t->count - pos) * sizeof(struct location)); | |
4866 | t->count++; | |
4867 | l->count = 1; | |
45edfa58 CL |
4868 | l->addr = track->addr; |
4869 | l->sum_time = age; | |
4870 | l->min_time = age; | |
4871 | l->max_time = age; | |
4872 | l->min_pid = track->pid; | |
4873 | l->max_pid = track->pid; | |
174596a0 RR |
4874 | cpumask_clear(to_cpumask(l->cpus)); |
4875 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4876 | nodes_clear(l->nodes); |
4877 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4878 | return 1; |
4879 | } | |
4880 | ||
4881 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
90e9f6a6 | 4882 | struct page *page, enum track_item alloc) |
88a420e4 | 4883 | { |
a973e9dd | 4884 | void *addr = page_address(page); |
88a420e4 | 4885 | void *p; |
90e9f6a6 | 4886 | unsigned long *map; |
88a420e4 | 4887 | |
90e9f6a6 | 4888 | map = get_map(s, page); |
224a88be | 4889 | for_each_object(p, s, addr, page->objects) |
4138fdfc | 4890 | if (!test_bit(__obj_to_index(s, addr, p), map)) |
45edfa58 | 4891 | add_location(t, s, get_track(s, p, alloc)); |
90e9f6a6 | 4892 | put_map(map); |
88a420e4 | 4893 | } |
64dd6849 | 4894 | #endif /* CONFIG_DEBUG_FS */ |
6dfd1b65 | 4895 | #endif /* CONFIG_SLUB_DEBUG */ |
88a420e4 | 4896 | |
ab4d5ed5 | 4897 | #ifdef CONFIG_SYSFS |
81819f0f | 4898 | enum slab_stat_type { |
205ab99d CL |
4899 | SL_ALL, /* All slabs */ |
4900 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4901 | SL_CPU, /* Only slabs used for cpu caches */ | |
4902 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4903 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4904 | }; |
4905 | ||
205ab99d | 4906 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4907 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4908 | #define SO_CPU (1 << SL_CPU) | |
4909 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4910 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4911 | |
62e5c4b4 | 4912 | static ssize_t show_slab_objects(struct kmem_cache *s, |
bf16d19a | 4913 | char *buf, unsigned long flags) |
81819f0f CL |
4914 | { |
4915 | unsigned long total = 0; | |
81819f0f CL |
4916 | int node; |
4917 | int x; | |
4918 | unsigned long *nodes; | |
bf16d19a | 4919 | int len = 0; |
81819f0f | 4920 | |
6396bb22 | 4921 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); |
62e5c4b4 CG |
4922 | if (!nodes) |
4923 | return -ENOMEM; | |
81819f0f | 4924 | |
205ab99d CL |
4925 | if (flags & SO_CPU) { |
4926 | int cpu; | |
81819f0f | 4927 | |
205ab99d | 4928 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4929 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4930 | cpu); | |
ec3ab083 | 4931 | int node; |
49e22585 | 4932 | struct page *page; |
dfb4f096 | 4933 | |
4db0c3c2 | 4934 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4935 | if (!page) |
4936 | continue; | |
205ab99d | 4937 | |
ec3ab083 CL |
4938 | node = page_to_nid(page); |
4939 | if (flags & SO_TOTAL) | |
4940 | x = page->objects; | |
4941 | else if (flags & SO_OBJECTS) | |
4942 | x = page->inuse; | |
4943 | else | |
4944 | x = 1; | |
49e22585 | 4945 | |
ec3ab083 CL |
4946 | total += x; |
4947 | nodes[node] += x; | |
4948 | ||
a93cf07b | 4949 | page = slub_percpu_partial_read_once(c); |
49e22585 | 4950 | if (page) { |
8afb1474 LZ |
4951 | node = page_to_nid(page); |
4952 | if (flags & SO_TOTAL) | |
4953 | WARN_ON_ONCE(1); | |
4954 | else if (flags & SO_OBJECTS) | |
4955 | WARN_ON_ONCE(1); | |
4956 | else | |
4957 | x = page->pages; | |
bc6697d8 ED |
4958 | total += x; |
4959 | nodes[node] += x; | |
49e22585 | 4960 | } |
81819f0f CL |
4961 | } |
4962 | } | |
4963 | ||
e4f8e513 QC |
4964 | /* |
4965 | * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex" | |
4966 | * already held which will conflict with an existing lock order: | |
4967 | * | |
4968 | * mem_hotplug_lock->slab_mutex->kernfs_mutex | |
4969 | * | |
4970 | * We don't really need mem_hotplug_lock (to hold off | |
4971 | * slab_mem_going_offline_callback) here because slab's memory hot | |
4972 | * unplug code doesn't destroy the kmem_cache->node[] data. | |
4973 | */ | |
4974 | ||
ab4d5ed5 | 4975 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4976 | if (flags & SO_ALL) { |
fa45dc25 CL |
4977 | struct kmem_cache_node *n; |
4978 | ||
4979 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4980 | |
d0e0ac97 CG |
4981 | if (flags & SO_TOTAL) |
4982 | x = atomic_long_read(&n->total_objects); | |
4983 | else if (flags & SO_OBJECTS) | |
4984 | x = atomic_long_read(&n->total_objects) - | |
4985 | count_partial(n, count_free); | |
81819f0f | 4986 | else |
205ab99d | 4987 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4988 | total += x; |
4989 | nodes[node] += x; | |
4990 | } | |
4991 | ||
ab4d5ed5 CL |
4992 | } else |
4993 | #endif | |
4994 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4995 | struct kmem_cache_node *n; |
81819f0f | 4996 | |
fa45dc25 | 4997 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4998 | if (flags & SO_TOTAL) |
4999 | x = count_partial(n, count_total); | |
5000 | else if (flags & SO_OBJECTS) | |
5001 | x = count_partial(n, count_inuse); | |
81819f0f | 5002 | else |
205ab99d | 5003 | x = n->nr_partial; |
81819f0f CL |
5004 | total += x; |
5005 | nodes[node] += x; | |
5006 | } | |
5007 | } | |
bf16d19a JP |
5008 | |
5009 | len += sysfs_emit_at(buf, len, "%lu", total); | |
81819f0f | 5010 | #ifdef CONFIG_NUMA |
bf16d19a | 5011 | for (node = 0; node < nr_node_ids; node++) { |
81819f0f | 5012 | if (nodes[node]) |
bf16d19a JP |
5013 | len += sysfs_emit_at(buf, len, " N%d=%lu", |
5014 | node, nodes[node]); | |
5015 | } | |
81819f0f | 5016 | #endif |
bf16d19a | 5017 | len += sysfs_emit_at(buf, len, "\n"); |
81819f0f | 5018 | kfree(nodes); |
bf16d19a JP |
5019 | |
5020 | return len; | |
81819f0f CL |
5021 | } |
5022 | ||
81819f0f | 5023 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) |
497888cf | 5024 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
5025 | |
5026 | struct slab_attribute { | |
5027 | struct attribute attr; | |
5028 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
5029 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
5030 | }; | |
5031 | ||
5032 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
5033 | static struct slab_attribute _name##_attr = \ |
5034 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
5035 | |
5036 | #define SLAB_ATTR(_name) \ | |
5037 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 5038 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 5039 | |
81819f0f CL |
5040 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
5041 | { | |
bf16d19a | 5042 | return sysfs_emit(buf, "%u\n", s->size); |
81819f0f CL |
5043 | } |
5044 | SLAB_ATTR_RO(slab_size); | |
5045 | ||
5046 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
5047 | { | |
bf16d19a | 5048 | return sysfs_emit(buf, "%u\n", s->align); |
81819f0f CL |
5049 | } |
5050 | SLAB_ATTR_RO(align); | |
5051 | ||
5052 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
5053 | { | |
bf16d19a | 5054 | return sysfs_emit(buf, "%u\n", s->object_size); |
81819f0f CL |
5055 | } |
5056 | SLAB_ATTR_RO(object_size); | |
5057 | ||
5058 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
5059 | { | |
bf16d19a | 5060 | return sysfs_emit(buf, "%u\n", oo_objects(s->oo)); |
81819f0f CL |
5061 | } |
5062 | SLAB_ATTR_RO(objs_per_slab); | |
5063 | ||
5064 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
5065 | { | |
bf16d19a | 5066 | return sysfs_emit(buf, "%u\n", oo_order(s->oo)); |
81819f0f | 5067 | } |
32a6f409 | 5068 | SLAB_ATTR_RO(order); |
81819f0f | 5069 | |
73d342b1 DR |
5070 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
5071 | { | |
bf16d19a | 5072 | return sysfs_emit(buf, "%lu\n", s->min_partial); |
73d342b1 DR |
5073 | } |
5074 | ||
5075 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
5076 | size_t length) | |
5077 | { | |
5078 | unsigned long min; | |
5079 | int err; | |
5080 | ||
3dbb95f7 | 5081 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
5082 | if (err) |
5083 | return err; | |
5084 | ||
c0bdb232 | 5085 | set_min_partial(s, min); |
73d342b1 DR |
5086 | return length; |
5087 | } | |
5088 | SLAB_ATTR(min_partial); | |
5089 | ||
49e22585 CL |
5090 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
5091 | { | |
bf16d19a | 5092 | return sysfs_emit(buf, "%u\n", slub_cpu_partial(s)); |
49e22585 CL |
5093 | } |
5094 | ||
5095 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
5096 | size_t length) | |
5097 | { | |
e5d9998f | 5098 | unsigned int objects; |
49e22585 CL |
5099 | int err; |
5100 | ||
e5d9998f | 5101 | err = kstrtouint(buf, 10, &objects); |
49e22585 CL |
5102 | if (err) |
5103 | return err; | |
345c905d | 5104 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 5105 | return -EINVAL; |
49e22585 | 5106 | |
e6d0e1dc | 5107 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
5108 | flush_all(s); |
5109 | return length; | |
5110 | } | |
5111 | SLAB_ATTR(cpu_partial); | |
5112 | ||
81819f0f CL |
5113 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
5114 | { | |
62c70bce JP |
5115 | if (!s->ctor) |
5116 | return 0; | |
bf16d19a | 5117 | return sysfs_emit(buf, "%pS\n", s->ctor); |
81819f0f CL |
5118 | } |
5119 | SLAB_ATTR_RO(ctor); | |
5120 | ||
81819f0f CL |
5121 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
5122 | { | |
bf16d19a | 5123 | return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
5124 | } |
5125 | SLAB_ATTR_RO(aliases); | |
5126 | ||
81819f0f CL |
5127 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
5128 | { | |
d9acf4b7 | 5129 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
5130 | } |
5131 | SLAB_ATTR_RO(partial); | |
5132 | ||
5133 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
5134 | { | |
d9acf4b7 | 5135 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
5136 | } |
5137 | SLAB_ATTR_RO(cpu_slabs); | |
5138 | ||
5139 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
5140 | { | |
205ab99d | 5141 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
5142 | } |
5143 | SLAB_ATTR_RO(objects); | |
5144 | ||
205ab99d CL |
5145 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
5146 | { | |
5147 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5148 | } | |
5149 | SLAB_ATTR_RO(objects_partial); | |
5150 | ||
49e22585 CL |
5151 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
5152 | { | |
5153 | int objects = 0; | |
5154 | int pages = 0; | |
5155 | int cpu; | |
bf16d19a | 5156 | int len = 0; |
49e22585 CL |
5157 | |
5158 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5159 | struct page *page; |
5160 | ||
5161 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5162 | |
5163 | if (page) { | |
5164 | pages += page->pages; | |
5165 | objects += page->pobjects; | |
5166 | } | |
5167 | } | |
5168 | ||
bf16d19a | 5169 | len += sysfs_emit_at(buf, len, "%d(%d)", objects, pages); |
49e22585 CL |
5170 | |
5171 | #ifdef CONFIG_SMP | |
5172 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5173 | struct page *page; |
5174 | ||
5175 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
bf16d19a JP |
5176 | if (page) |
5177 | len += sysfs_emit_at(buf, len, " C%d=%d(%d)", | |
5178 | cpu, page->pobjects, page->pages); | |
49e22585 CL |
5179 | } |
5180 | #endif | |
bf16d19a JP |
5181 | len += sysfs_emit_at(buf, len, "\n"); |
5182 | ||
5183 | return len; | |
49e22585 CL |
5184 | } |
5185 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5186 | ||
a5a84755 CL |
5187 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5188 | { | |
bf16d19a | 5189 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); |
a5a84755 | 5190 | } |
8f58119a | 5191 | SLAB_ATTR_RO(reclaim_account); |
a5a84755 CL |
5192 | |
5193 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5194 | { | |
bf16d19a | 5195 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
a5a84755 CL |
5196 | } |
5197 | SLAB_ATTR_RO(hwcache_align); | |
5198 | ||
5199 | #ifdef CONFIG_ZONE_DMA | |
5200 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5201 | { | |
bf16d19a | 5202 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); |
a5a84755 CL |
5203 | } |
5204 | SLAB_ATTR_RO(cache_dma); | |
5205 | #endif | |
5206 | ||
8eb8284b DW |
5207 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) |
5208 | { | |
bf16d19a | 5209 | return sysfs_emit(buf, "%u\n", s->usersize); |
8eb8284b DW |
5210 | } |
5211 | SLAB_ATTR_RO(usersize); | |
5212 | ||
a5a84755 CL |
5213 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) |
5214 | { | |
bf16d19a | 5215 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5216 | } |
5217 | SLAB_ATTR_RO(destroy_by_rcu); | |
5218 | ||
ab4d5ed5 | 5219 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5220 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5221 | { | |
5222 | return show_slab_objects(s, buf, SO_ALL); | |
5223 | } | |
5224 | SLAB_ATTR_RO(slabs); | |
5225 | ||
205ab99d CL |
5226 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5227 | { | |
5228 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5229 | } | |
5230 | SLAB_ATTR_RO(total_objects); | |
5231 | ||
81819f0f CL |
5232 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5233 | { | |
bf16d19a | 5234 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f | 5235 | } |
060807f8 | 5236 | SLAB_ATTR_RO(sanity_checks); |
81819f0f CL |
5237 | |
5238 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5239 | { | |
bf16d19a | 5240 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE)); |
81819f0f | 5241 | } |
060807f8 | 5242 | SLAB_ATTR_RO(trace); |
81819f0f | 5243 | |
81819f0f CL |
5244 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5245 | { | |
bf16d19a | 5246 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); |
81819f0f CL |
5247 | } |
5248 | ||
ad38b5b1 | 5249 | SLAB_ATTR_RO(red_zone); |
81819f0f CL |
5250 | |
5251 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5252 | { | |
bf16d19a | 5253 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON)); |
81819f0f CL |
5254 | } |
5255 | ||
ad38b5b1 | 5256 | SLAB_ATTR_RO(poison); |
81819f0f CL |
5257 | |
5258 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5259 | { | |
bf16d19a | 5260 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); |
81819f0f CL |
5261 | } |
5262 | ||
ad38b5b1 | 5263 | SLAB_ATTR_RO(store_user); |
81819f0f | 5264 | |
53e15af0 CL |
5265 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5266 | { | |
5267 | return 0; | |
5268 | } | |
5269 | ||
5270 | static ssize_t validate_store(struct kmem_cache *s, | |
5271 | const char *buf, size_t length) | |
5272 | { | |
434e245d CL |
5273 | int ret = -EINVAL; |
5274 | ||
5275 | if (buf[0] == '1') { | |
5276 | ret = validate_slab_cache(s); | |
5277 | if (ret >= 0) | |
5278 | ret = length; | |
5279 | } | |
5280 | return ret; | |
53e15af0 CL |
5281 | } |
5282 | SLAB_ATTR(validate); | |
a5a84755 | 5283 | |
a5a84755 CL |
5284 | #endif /* CONFIG_SLUB_DEBUG */ |
5285 | ||
5286 | #ifdef CONFIG_FAILSLAB | |
5287 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5288 | { | |
bf16d19a | 5289 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); |
a5a84755 | 5290 | } |
060807f8 | 5291 | SLAB_ATTR_RO(failslab); |
ab4d5ed5 | 5292 | #endif |
53e15af0 | 5293 | |
2086d26a CL |
5294 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5295 | { | |
5296 | return 0; | |
5297 | } | |
5298 | ||
5299 | static ssize_t shrink_store(struct kmem_cache *s, | |
5300 | const char *buf, size_t length) | |
5301 | { | |
832f37f5 | 5302 | if (buf[0] == '1') |
10befea9 | 5303 | kmem_cache_shrink(s); |
832f37f5 | 5304 | else |
2086d26a CL |
5305 | return -EINVAL; |
5306 | return length; | |
5307 | } | |
5308 | SLAB_ATTR(shrink); | |
5309 | ||
81819f0f | 5310 | #ifdef CONFIG_NUMA |
9824601e | 5311 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5312 | { |
bf16d19a | 5313 | return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5314 | } |
5315 | ||
9824601e | 5316 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5317 | const char *buf, size_t length) |
5318 | { | |
eb7235eb | 5319 | unsigned int ratio; |
0121c619 CL |
5320 | int err; |
5321 | ||
eb7235eb | 5322 | err = kstrtouint(buf, 10, &ratio); |
0121c619 CL |
5323 | if (err) |
5324 | return err; | |
eb7235eb AD |
5325 | if (ratio > 100) |
5326 | return -ERANGE; | |
0121c619 | 5327 | |
eb7235eb | 5328 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5329 | |
81819f0f CL |
5330 | return length; |
5331 | } | |
9824601e | 5332 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5333 | #endif |
5334 | ||
8ff12cfc | 5335 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5336 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5337 | { | |
5338 | unsigned long sum = 0; | |
5339 | int cpu; | |
bf16d19a | 5340 | int len = 0; |
6da2ec56 | 5341 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); |
8ff12cfc CL |
5342 | |
5343 | if (!data) | |
5344 | return -ENOMEM; | |
5345 | ||
5346 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5347 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5348 | |
5349 | data[cpu] = x; | |
5350 | sum += x; | |
5351 | } | |
5352 | ||
bf16d19a | 5353 | len += sysfs_emit_at(buf, len, "%lu", sum); |
8ff12cfc | 5354 | |
50ef37b9 | 5355 | #ifdef CONFIG_SMP |
8ff12cfc | 5356 | for_each_online_cpu(cpu) { |
bf16d19a JP |
5357 | if (data[cpu]) |
5358 | len += sysfs_emit_at(buf, len, " C%d=%u", | |
5359 | cpu, data[cpu]); | |
8ff12cfc | 5360 | } |
50ef37b9 | 5361 | #endif |
8ff12cfc | 5362 | kfree(data); |
bf16d19a JP |
5363 | len += sysfs_emit_at(buf, len, "\n"); |
5364 | ||
5365 | return len; | |
8ff12cfc CL |
5366 | } |
5367 | ||
78eb00cc DR |
5368 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5369 | { | |
5370 | int cpu; | |
5371 | ||
5372 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5373 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5374 | } |
5375 | ||
8ff12cfc CL |
5376 | #define STAT_ATTR(si, text) \ |
5377 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5378 | { \ | |
5379 | return show_stat(s, buf, si); \ | |
5380 | } \ | |
78eb00cc DR |
5381 | static ssize_t text##_store(struct kmem_cache *s, \ |
5382 | const char *buf, size_t length) \ | |
5383 | { \ | |
5384 | if (buf[0] != '0') \ | |
5385 | return -EINVAL; \ | |
5386 | clear_stat(s, si); \ | |
5387 | return length; \ | |
5388 | } \ | |
5389 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5390 | |
5391 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5392 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5393 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5394 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5395 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5396 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5397 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5398 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5399 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5400 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5401 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5402 | STAT_ATTR(FREE_SLAB, free_slab); |
5403 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5404 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5405 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5406 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5407 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5408 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5409 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5410 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5411 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5412 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5413 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5414 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5415 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5416 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
6dfd1b65 | 5417 | #endif /* CONFIG_SLUB_STATS */ |
8ff12cfc | 5418 | |
06428780 | 5419 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5420 | &slab_size_attr.attr, |
5421 | &object_size_attr.attr, | |
5422 | &objs_per_slab_attr.attr, | |
5423 | &order_attr.attr, | |
73d342b1 | 5424 | &min_partial_attr.attr, |
49e22585 | 5425 | &cpu_partial_attr.attr, |
81819f0f | 5426 | &objects_attr.attr, |
205ab99d | 5427 | &objects_partial_attr.attr, |
81819f0f CL |
5428 | &partial_attr.attr, |
5429 | &cpu_slabs_attr.attr, | |
5430 | &ctor_attr.attr, | |
81819f0f CL |
5431 | &aliases_attr.attr, |
5432 | &align_attr.attr, | |
81819f0f CL |
5433 | &hwcache_align_attr.attr, |
5434 | &reclaim_account_attr.attr, | |
5435 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5436 | &shrink_attr.attr, |
49e22585 | 5437 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5438 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5439 | &total_objects_attr.attr, |
5440 | &slabs_attr.attr, | |
5441 | &sanity_checks_attr.attr, | |
5442 | &trace_attr.attr, | |
81819f0f CL |
5443 | &red_zone_attr.attr, |
5444 | &poison_attr.attr, | |
5445 | &store_user_attr.attr, | |
53e15af0 | 5446 | &validate_attr.attr, |
ab4d5ed5 | 5447 | #endif |
81819f0f CL |
5448 | #ifdef CONFIG_ZONE_DMA |
5449 | &cache_dma_attr.attr, | |
5450 | #endif | |
5451 | #ifdef CONFIG_NUMA | |
9824601e | 5452 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5453 | #endif |
5454 | #ifdef CONFIG_SLUB_STATS | |
5455 | &alloc_fastpath_attr.attr, | |
5456 | &alloc_slowpath_attr.attr, | |
5457 | &free_fastpath_attr.attr, | |
5458 | &free_slowpath_attr.attr, | |
5459 | &free_frozen_attr.attr, | |
5460 | &free_add_partial_attr.attr, | |
5461 | &free_remove_partial_attr.attr, | |
5462 | &alloc_from_partial_attr.attr, | |
5463 | &alloc_slab_attr.attr, | |
5464 | &alloc_refill_attr.attr, | |
e36a2652 | 5465 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5466 | &free_slab_attr.attr, |
5467 | &cpuslab_flush_attr.attr, | |
5468 | &deactivate_full_attr.attr, | |
5469 | &deactivate_empty_attr.attr, | |
5470 | &deactivate_to_head_attr.attr, | |
5471 | &deactivate_to_tail_attr.attr, | |
5472 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5473 | &deactivate_bypass_attr.attr, |
65c3376a | 5474 | &order_fallback_attr.attr, |
b789ef51 CL |
5475 | &cmpxchg_double_fail_attr.attr, |
5476 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5477 | &cpu_partial_alloc_attr.attr, |
5478 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5479 | &cpu_partial_node_attr.attr, |
5480 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5481 | #endif |
4c13dd3b DM |
5482 | #ifdef CONFIG_FAILSLAB |
5483 | &failslab_attr.attr, | |
5484 | #endif | |
8eb8284b | 5485 | &usersize_attr.attr, |
4c13dd3b | 5486 | |
81819f0f CL |
5487 | NULL |
5488 | }; | |
5489 | ||
1fdaaa23 | 5490 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
5491 | .attrs = slab_attrs, |
5492 | }; | |
5493 | ||
5494 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5495 | struct attribute *attr, | |
5496 | char *buf) | |
5497 | { | |
5498 | struct slab_attribute *attribute; | |
5499 | struct kmem_cache *s; | |
5500 | int err; | |
5501 | ||
5502 | attribute = to_slab_attr(attr); | |
5503 | s = to_slab(kobj); | |
5504 | ||
5505 | if (!attribute->show) | |
5506 | return -EIO; | |
5507 | ||
5508 | err = attribute->show(s, buf); | |
5509 | ||
5510 | return err; | |
5511 | } | |
5512 | ||
5513 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5514 | struct attribute *attr, | |
5515 | const char *buf, size_t len) | |
5516 | { | |
5517 | struct slab_attribute *attribute; | |
5518 | struct kmem_cache *s; | |
5519 | int err; | |
5520 | ||
5521 | attribute = to_slab_attr(attr); | |
5522 | s = to_slab(kobj); | |
5523 | ||
5524 | if (!attribute->store) | |
5525 | return -EIO; | |
5526 | ||
5527 | err = attribute->store(s, buf, len); | |
81819f0f CL |
5528 | return err; |
5529 | } | |
5530 | ||
41a21285 CL |
5531 | static void kmem_cache_release(struct kobject *k) |
5532 | { | |
5533 | slab_kmem_cache_release(to_slab(k)); | |
5534 | } | |
5535 | ||
52cf25d0 | 5536 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5537 | .show = slab_attr_show, |
5538 | .store = slab_attr_store, | |
5539 | }; | |
5540 | ||
5541 | static struct kobj_type slab_ktype = { | |
5542 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5543 | .release = kmem_cache_release, |
81819f0f CL |
5544 | }; |
5545 | ||
27c3a314 | 5546 | static struct kset *slab_kset; |
81819f0f | 5547 | |
9a41707b VD |
5548 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5549 | { | |
9a41707b VD |
5550 | return slab_kset; |
5551 | } | |
5552 | ||
81819f0f CL |
5553 | #define ID_STR_LENGTH 64 |
5554 | ||
5555 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5556 | * |
5557 | * Format :[flags-]size | |
81819f0f CL |
5558 | */ |
5559 | static char *create_unique_id(struct kmem_cache *s) | |
5560 | { | |
5561 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5562 | char *p = name; | |
5563 | ||
5564 | BUG_ON(!name); | |
5565 | ||
5566 | *p++ = ':'; | |
5567 | /* | |
5568 | * First flags affecting slabcache operations. We will only | |
5569 | * get here for aliasable slabs so we do not need to support | |
5570 | * too many flags. The flags here must cover all flags that | |
5571 | * are matched during merging to guarantee that the id is | |
5572 | * unique. | |
5573 | */ | |
5574 | if (s->flags & SLAB_CACHE_DMA) | |
5575 | *p++ = 'd'; | |
6d6ea1e9 NB |
5576 | if (s->flags & SLAB_CACHE_DMA32) |
5577 | *p++ = 'D'; | |
81819f0f CL |
5578 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
5579 | *p++ = 'a'; | |
becfda68 | 5580 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5581 | *p++ = 'F'; |
230e9fc2 VD |
5582 | if (s->flags & SLAB_ACCOUNT) |
5583 | *p++ = 'A'; | |
81819f0f CL |
5584 | if (p != name + 1) |
5585 | *p++ = '-'; | |
44065b2e | 5586 | p += sprintf(p, "%07u", s->size); |
2633d7a0 | 5587 | |
81819f0f CL |
5588 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5589 | return name; | |
5590 | } | |
5591 | ||
5592 | static int sysfs_slab_add(struct kmem_cache *s) | |
5593 | { | |
5594 | int err; | |
5595 | const char *name; | |
1663f26d | 5596 | struct kset *kset = cache_kset(s); |
45530c44 | 5597 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5598 | |
1663f26d TH |
5599 | if (!kset) { |
5600 | kobject_init(&s->kobj, &slab_ktype); | |
5601 | return 0; | |
5602 | } | |
5603 | ||
11066386 MC |
5604 | if (!unmergeable && disable_higher_order_debug && |
5605 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
5606 | unmergeable = 1; | |
5607 | ||
81819f0f CL |
5608 | if (unmergeable) { |
5609 | /* | |
5610 | * Slabcache can never be merged so we can use the name proper. | |
5611 | * This is typically the case for debug situations. In that | |
5612 | * case we can catch duplicate names easily. | |
5613 | */ | |
27c3a314 | 5614 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5615 | name = s->name; |
5616 | } else { | |
5617 | /* | |
5618 | * Create a unique name for the slab as a target | |
5619 | * for the symlinks. | |
5620 | */ | |
5621 | name = create_unique_id(s); | |
5622 | } | |
5623 | ||
1663f26d | 5624 | s->kobj.kset = kset; |
26e4f205 | 5625 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
757fed1d | 5626 | if (err) |
80da026a | 5627 | goto out; |
81819f0f CL |
5628 | |
5629 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5630 | if (err) |
5631 | goto out_del_kobj; | |
9a41707b | 5632 | |
81819f0f CL |
5633 | if (!unmergeable) { |
5634 | /* Setup first alias */ | |
5635 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5636 | } |
54b6a731 DJ |
5637 | out: |
5638 | if (!unmergeable) | |
5639 | kfree(name); | |
5640 | return err; | |
5641 | out_del_kobj: | |
5642 | kobject_del(&s->kobj); | |
54b6a731 | 5643 | goto out; |
81819f0f CL |
5644 | } |
5645 | ||
d50d82fa MP |
5646 | void sysfs_slab_unlink(struct kmem_cache *s) |
5647 | { | |
5648 | if (slab_state >= FULL) | |
5649 | kobject_del(&s->kobj); | |
5650 | } | |
5651 | ||
bf5eb3de TH |
5652 | void sysfs_slab_release(struct kmem_cache *s) |
5653 | { | |
5654 | if (slab_state >= FULL) | |
5655 | kobject_put(&s->kobj); | |
81819f0f CL |
5656 | } |
5657 | ||
5658 | /* | |
5659 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5660 | * available lest we lose that information. |
81819f0f CL |
5661 | */ |
5662 | struct saved_alias { | |
5663 | struct kmem_cache *s; | |
5664 | const char *name; | |
5665 | struct saved_alias *next; | |
5666 | }; | |
5667 | ||
5af328a5 | 5668 | static struct saved_alias *alias_list; |
81819f0f CL |
5669 | |
5670 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5671 | { | |
5672 | struct saved_alias *al; | |
5673 | ||
97d06609 | 5674 | if (slab_state == FULL) { |
81819f0f CL |
5675 | /* |
5676 | * If we have a leftover link then remove it. | |
5677 | */ | |
27c3a314 GKH |
5678 | sysfs_remove_link(&slab_kset->kobj, name); |
5679 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5680 | } |
5681 | ||
5682 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5683 | if (!al) | |
5684 | return -ENOMEM; | |
5685 | ||
5686 | al->s = s; | |
5687 | al->name = name; | |
5688 | al->next = alias_list; | |
5689 | alias_list = al; | |
5690 | return 0; | |
5691 | } | |
5692 | ||
5693 | static int __init slab_sysfs_init(void) | |
5694 | { | |
5b95a4ac | 5695 | struct kmem_cache *s; |
81819f0f CL |
5696 | int err; |
5697 | ||
18004c5d | 5698 | mutex_lock(&slab_mutex); |
2bce6485 | 5699 | |
d7660ce5 | 5700 | slab_kset = kset_create_and_add("slab", NULL, kernel_kobj); |
27c3a314 | 5701 | if (!slab_kset) { |
18004c5d | 5702 | mutex_unlock(&slab_mutex); |
f9f58285 | 5703 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5704 | return -ENOSYS; |
5705 | } | |
5706 | ||
97d06609 | 5707 | slab_state = FULL; |
26a7bd03 | 5708 | |
5b95a4ac | 5709 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5710 | err = sysfs_slab_add(s); |
5d540fb7 | 5711 | if (err) |
f9f58285 FF |
5712 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5713 | s->name); | |
26a7bd03 | 5714 | } |
81819f0f CL |
5715 | |
5716 | while (alias_list) { | |
5717 | struct saved_alias *al = alias_list; | |
5718 | ||
5719 | alias_list = alias_list->next; | |
5720 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5721 | if (err) |
f9f58285 FF |
5722 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5723 | al->name); | |
81819f0f CL |
5724 | kfree(al); |
5725 | } | |
5726 | ||
18004c5d | 5727 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5728 | return 0; |
5729 | } | |
5730 | ||
5731 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5732 | #endif /* CONFIG_SYSFS */ |
57ed3eda | 5733 | |
64dd6849 FM |
5734 | #if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS) |
5735 | static int slab_debugfs_show(struct seq_file *seq, void *v) | |
5736 | { | |
5737 | ||
5738 | struct location *l; | |
5739 | unsigned int idx = *(unsigned int *)v; | |
5740 | struct loc_track *t = seq->private; | |
5741 | ||
5742 | if (idx < t->count) { | |
5743 | l = &t->loc[idx]; | |
5744 | ||
5745 | seq_printf(seq, "%7ld ", l->count); | |
5746 | ||
5747 | if (l->addr) | |
5748 | seq_printf(seq, "%pS", (void *)l->addr); | |
5749 | else | |
5750 | seq_puts(seq, "<not-available>"); | |
5751 | ||
5752 | if (l->sum_time != l->min_time) { | |
5753 | seq_printf(seq, " age=%ld/%llu/%ld", | |
5754 | l->min_time, div_u64(l->sum_time, l->count), | |
5755 | l->max_time); | |
5756 | } else | |
5757 | seq_printf(seq, " age=%ld", l->min_time); | |
5758 | ||
5759 | if (l->min_pid != l->max_pid) | |
5760 | seq_printf(seq, " pid=%ld-%ld", l->min_pid, l->max_pid); | |
5761 | else | |
5762 | seq_printf(seq, " pid=%ld", | |
5763 | l->min_pid); | |
5764 | ||
5765 | if (num_online_cpus() > 1 && !cpumask_empty(to_cpumask(l->cpus))) | |
5766 | seq_printf(seq, " cpus=%*pbl", | |
5767 | cpumask_pr_args(to_cpumask(l->cpus))); | |
5768 | ||
5769 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes)) | |
5770 | seq_printf(seq, " nodes=%*pbl", | |
5771 | nodemask_pr_args(&l->nodes)); | |
5772 | ||
5773 | seq_puts(seq, "\n"); | |
5774 | } | |
5775 | ||
5776 | if (!idx && !t->count) | |
5777 | seq_puts(seq, "No data\n"); | |
5778 | ||
5779 | return 0; | |
5780 | } | |
5781 | ||
5782 | static void slab_debugfs_stop(struct seq_file *seq, void *v) | |
5783 | { | |
5784 | } | |
5785 | ||
5786 | static void *slab_debugfs_next(struct seq_file *seq, void *v, loff_t *ppos) | |
5787 | { | |
5788 | struct loc_track *t = seq->private; | |
5789 | ||
5790 | v = ppos; | |
5791 | ++*ppos; | |
5792 | if (*ppos <= t->count) | |
5793 | return v; | |
5794 | ||
5795 | return NULL; | |
5796 | } | |
5797 | ||
5798 | static void *slab_debugfs_start(struct seq_file *seq, loff_t *ppos) | |
5799 | { | |
5800 | return ppos; | |
5801 | } | |
5802 | ||
5803 | static const struct seq_operations slab_debugfs_sops = { | |
5804 | .start = slab_debugfs_start, | |
5805 | .next = slab_debugfs_next, | |
5806 | .stop = slab_debugfs_stop, | |
5807 | .show = slab_debugfs_show, | |
5808 | }; | |
5809 | ||
5810 | static int slab_debug_trace_open(struct inode *inode, struct file *filep) | |
5811 | { | |
5812 | ||
5813 | struct kmem_cache_node *n; | |
5814 | enum track_item alloc; | |
5815 | int node; | |
5816 | struct loc_track *t = __seq_open_private(filep, &slab_debugfs_sops, | |
5817 | sizeof(struct loc_track)); | |
5818 | struct kmem_cache *s = file_inode(filep)->i_private; | |
5819 | ||
5820 | if (strcmp(filep->f_path.dentry->d_name.name, "alloc_traces") == 0) | |
5821 | alloc = TRACK_ALLOC; | |
5822 | else | |
5823 | alloc = TRACK_FREE; | |
5824 | ||
5825 | if (!alloc_loc_track(t, PAGE_SIZE / sizeof(struct location), GFP_KERNEL)) | |
5826 | return -ENOMEM; | |
5827 | ||
5828 | /* Push back cpu slabs */ | |
5829 | flush_all(s); | |
5830 | ||
5831 | for_each_kmem_cache_node(s, node, n) { | |
5832 | unsigned long flags; | |
5833 | struct page *page; | |
5834 | ||
5835 | if (!atomic_long_read(&n->nr_slabs)) | |
5836 | continue; | |
5837 | ||
5838 | spin_lock_irqsave(&n->list_lock, flags); | |
5839 | list_for_each_entry(page, &n->partial, slab_list) | |
5840 | process_slab(t, s, page, alloc); | |
5841 | list_for_each_entry(page, &n->full, slab_list) | |
5842 | process_slab(t, s, page, alloc); | |
5843 | spin_unlock_irqrestore(&n->list_lock, flags); | |
5844 | } | |
5845 | ||
5846 | return 0; | |
5847 | } | |
5848 | ||
5849 | static int slab_debug_trace_release(struct inode *inode, struct file *file) | |
5850 | { | |
5851 | struct seq_file *seq = file->private_data; | |
5852 | struct loc_track *t = seq->private; | |
5853 | ||
5854 | free_loc_track(t); | |
5855 | return seq_release_private(inode, file); | |
5856 | } | |
5857 | ||
5858 | static const struct file_operations slab_debugfs_fops = { | |
5859 | .open = slab_debug_trace_open, | |
5860 | .read = seq_read, | |
5861 | .llseek = seq_lseek, | |
5862 | .release = slab_debug_trace_release, | |
5863 | }; | |
5864 | ||
5865 | static void debugfs_slab_add(struct kmem_cache *s) | |
5866 | { | |
5867 | struct dentry *slab_cache_dir; | |
5868 | ||
5869 | if (unlikely(!slab_debugfs_root)) | |
5870 | return; | |
5871 | ||
5872 | slab_cache_dir = debugfs_create_dir(s->name, slab_debugfs_root); | |
5873 | ||
5874 | debugfs_create_file("alloc_traces", 0400, | |
5875 | slab_cache_dir, s, &slab_debugfs_fops); | |
5876 | ||
5877 | debugfs_create_file("free_traces", 0400, | |
5878 | slab_cache_dir, s, &slab_debugfs_fops); | |
5879 | } | |
5880 | ||
5881 | void debugfs_slab_release(struct kmem_cache *s) | |
5882 | { | |
5883 | debugfs_remove_recursive(debugfs_lookup(s->name, slab_debugfs_root)); | |
5884 | } | |
5885 | ||
5886 | static int __init slab_debugfs_init(void) | |
5887 | { | |
5888 | struct kmem_cache *s; | |
5889 | ||
5890 | slab_debugfs_root = debugfs_create_dir("slab", NULL); | |
5891 | ||
5892 | list_for_each_entry(s, &slab_caches, list) | |
5893 | if (s->flags & SLAB_STORE_USER) | |
5894 | debugfs_slab_add(s); | |
5895 | ||
5896 | return 0; | |
5897 | ||
5898 | } | |
5899 | __initcall(slab_debugfs_init); | |
5900 | #endif | |
57ed3eda PE |
5901 | /* |
5902 | * The /proc/slabinfo ABI | |
5903 | */ | |
5b365771 | 5904 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 5905 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5906 | { |
57ed3eda | 5907 | unsigned long nr_slabs = 0; |
205ab99d CL |
5908 | unsigned long nr_objs = 0; |
5909 | unsigned long nr_free = 0; | |
57ed3eda | 5910 | int node; |
fa45dc25 | 5911 | struct kmem_cache_node *n; |
57ed3eda | 5912 | |
fa45dc25 | 5913 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5914 | nr_slabs += node_nr_slabs(n); |
5915 | nr_objs += node_nr_objs(n); | |
205ab99d | 5916 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5917 | } |
5918 | ||
0d7561c6 GC |
5919 | sinfo->active_objs = nr_objs - nr_free; |
5920 | sinfo->num_objs = nr_objs; | |
5921 | sinfo->active_slabs = nr_slabs; | |
5922 | sinfo->num_slabs = nr_slabs; | |
5923 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5924 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5925 | } |
5926 | ||
0d7561c6 | 5927 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5928 | { |
7b3c3a50 AD |
5929 | } |
5930 | ||
b7454ad3 GC |
5931 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5932 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5933 | { |
b7454ad3 | 5934 | return -EIO; |
7b3c3a50 | 5935 | } |
5b365771 | 5936 | #endif /* CONFIG_SLUB_DEBUG */ |